Datasets:

timaeus

/

pile-pubmed_central

like 0

Follow

Timaeus 10

Background ========== Psychosocial factors including chronic stress, depression, dejection, and lack of social support have been proved risk factors for cancer occurrence and progression by psychological and epidemiological studies \[[@B1]-[@B4]\]. It is well known that chronic stress impacts on immune system, neuroendocrine system, lymphatic and hematopoietic system. Stress inhibits the immune response ability in antigen-specific T-cells and natural killer cells while stimulates the secretion of proinflammatory cytokines, such as IL-1, IL-2, IL-6, IL-8, IL-11 and TNF-α, which were regarded as co-factors for modulating the growth and progression of tumor \[[@B5],[@B6]\]. Recent studies reported that chronic stress can also immediately affect the growth, development and metastasis of malignant tumors via hormone receptors on tumor cells \[[@B7]-[@B10]\]. In mammals under stress, an increased level of stress-related hormone (**SRH**) can be induced by the activation of the hypothalamic-pituitary-adrenal and the sympathetic-adrenal medullary axes. Activation by stress on sympathetic nervous system results in the release of catecholamines from the adrenal medulla and sympathetic nerve terminals \[[@B6],[@B10]\]. Catecholamines consist of several kinds of substances such as dopamine, histamine, serotonin, epinephrine and norepinephrine (**NE**). The last one is regarded as the most potential SRH related to tumors in mammals \[[@B10],[@B11]\]. As ligands, catecholamines can bind adrenergic receptors (**ARs**) coupled with G-protein which can be classified as several subtypes such as α1, α2, β1, β2 and β3 ARs. Many types of ARs locate on tumor cells, providing the theory that chronic stress impacts on the progression of cancer. Furthermore, the effect of stress could be mimicked with NE or β2-AR agonists, and abolished with surgical sympathetic denervation, β-AR antagonists or knocking down β2-AR gene by small interfering RNA \[[@B6],[@B10],[@B12]\]. It is accepted that a solid tumor can not progress without angiogenesis. VEGF, one of the most important angiogenic factors, can recruit and induce endothelial cells to proliferate and migrate, thereby starting the critical step of tumor expansion. Previous studies have demonstrated that NE upregulates VEGF, IL-8, IL-6 and MMP expression levels in some kinds of tumor cells *in vitro* such as melanoma, breast cancer, colon cancer, prostate cancer, ovary cancer, pancreatic cancer and nasopharynx cancer. Besides, migration of cancer cells can be stimulated by NE, which can be blocked by nonselective β-AR antagonist, propranolol \[[@B7]-[@B9],[@B13]-[@B18]\]. In mouse models *in vivo*, chronic stress stimulates the growth, progression and metastasis of tumors, which can also be inhibited by propranolol \[[@B13]-[@B15],[@B19]\]. The clinical research reported that propranolol lowered the rate of breast cancer-specific mortality, cancer recurrence and distant metastasis, thus improved relapse-free survival and cancer specific survival \[[@B20]-[@B22]\]. Tumor angiogenesis plays a key role in development of solid tumors. Sunitinib, one kind of anti-angiogenic drugs, is a tyrosine kinase inhibitor with the ability of blocking VEGFR1, VEGFR2, VEGFR3, PDGFRα, PDGFRβ, c-Kit and RET. It can induce tumor cell death and inhibit tumor proliferation and vascularization \[[@B23]-[@B25]\]. However, in clinic, treatment with sunitinib alone is of poor curative effect or even inefficacious for many types of solid tumors. On the contrary, sunitinib exhibits satisfactory efficacy in mouse homografts of melanoma, Lewis lung cancer, renal cancer and colon cancer, and xenografts of human colorectal cancer *in vivo*\[[@B24],[@B26]-[@B28]\]. Additionally, monotherapy with anti-angiogenic drugs including endostatin and bevacizumab also shows the discrepancy between clinical and preclinical results \[[@B29],[@B30]\]. Thus the question should be presented: Why does the difference of the curative response between the human and animal occur? Different from tumor-bearing mice, receiving a diagnosis of malignancy and battling with chronic uncertainties as regards treatment, progression, recurrence, and mortality is a major chronic stressor imaginable for patients with cancer. Given the impact of chronic stress on a cancer patient, the confluence of the psychological and physical discomfort places the patient at high risk for the occurrence of stress-induced behavioral alterations which usually presents depression, anxiety, sadness, fear and hopelessness \[[@B4],[@B11],[@B31],[@B32]\]. We reported previously that 39.5% of cancer patients were unwilling to realize the diagnosis of cancer, 63.0% were burdened with mental stress and 33.0% considered the impact of mental stress above that of somatic symptoms \[[@B33]\]. We hypothesize that the discrepancy of the efficacy of anti-angiogenic drugs between clinical and preclinical results is caused by chronic stress, which has not been yet identified. So in this research, the goal is to investigate whether NE, one of the most potent stress related hormones, can attenuate the efficacy of sunitinib in a mouse model and whether this effect can be blocked by propranolol. Materials and methods ===================== Cell culture ------------ The murine melanoma B16F1 cells and human lung adenocarcinoma A549 cells, kind gifts from State Key Laboratory of Biotherapy (Sichuan University, Chengdu), were authenticated by the supplier \[[@B29]\] and cultured in RPMI 1640 complete medium containing 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin at 37°C with 5% CO2 in humidified atmosphere. Reagents -------- NE, 3-(4,5-dimethylthiazol-2-yl)-2,5 diphenyltetrazolium bromide (**MTT**), dimethylsulfoxide (**DMSO**), isoproterenol, dobutamine and terbutaline were purchased from Sigma (St. Louis, MO, USA); propranolol and 8-CPT from Enzo (Germany); forskolin from Biovision (USA); H-89 and myristoylated PKI from Calbiochem (USA); sunitinib from Pfizer (USA); RNAiso plus and One Step SYBR® PrimeScript™ RT-PCR Kit from TaKaRa (Japan). *In vitro* cell proliferation assays for measuring the IC~50~ (half maximal inhibitory concentration) of sunitinib in B16F1 cells --------------------------------------------------------------------------------------------------------------------------------- B16F1 cells were harvested and seeded in 96-well plates (5,000 cells/200 μL complete medium/ well). After 24 hours incubation, the cells were exposed to various concentrations (0--100 μM, each concentration had six replicate wells) of sunitinib for 48 h. Following sunitinib treatment, 20 μL of 5 mg/mL MTT was added to each well and incubated at 37°C for 4 hours. The plates were centrifuged, the supernatants were carefully discarded and formazan crystals were dissolved in 150 μL DMSO. At last, the light absorbance at 490 nm was determined in a luminescence plate reader (PerkinElmer, USA) according to the manufacturer's instructions. Evaluation of the influence of NE on mRNA and protein expression *in vitro* --------------------------------------------------------------------------- B16F1 and A549 cells were dispensed in six-well culture plates (2 × 10^5^/well). After incubation overnight, 2 mL complete RPMI 1640 medium was replaced by serum-free medium for 24 hours to make the cells adapt serum-starvation. Then cells were incubated in 2 mL renewed serum-free medium containing 0, 0.1, 1, 10 μM NE or 10 μM NE +10 μM propranolol (propranolol was added 30 minutes prior to NE). Culture supernatants were gathered and cells were homogenized in RNAiso plus at different time points designed for detection by ELISA (3, 6, 12 and 24 hours) and real-time PCR (1, 2, 3 and 4 hours), respectively. In addition, we evaluated the influence of 10 μM NE in B16F1 cells treated with sunitinib at the concentration equal to **IC**~**50**~ (sunitinib was added 30 minutes following NE)~**.**~ Evaluation of β-AR (β-adrenoceptor)/cAMP/PKA signaling pathway -------------------------------------------------------------- A recent study identified that the β2-AR/cAMP/PKA signaling pathway mediated the up-regulation of VEGF by NE on human ovarian cancer cells \[[@B9]\]. Here we tested the role of this pathway on A549 cells. First, 10 μM α-AR antagonist phentolamine and 10 μM β-AR antagonist propranolol were added into the cell cultures 30 minutes before adding 10 μM NE in order to assess the role of AR subtypes (α-AR VS β-AR). Second, A549 cells were incubated in serum-free medium containing 10 μM β-AR agonist isoproterenol, 10 μM β1-AR agonist dobutamine, 10 μM β2-AR agonist terbutaline, 100 μM selective activator of the cAMP receptor 8-CPT, 10 μM adenylate cyclase agonist forskolin, 100 μM cAMP-dependent protein kinase inhibitor H-89 or 10 μM myristoylated protein kinase inhibitor PKI. Similar to propranolol, H-89 or PKI was added 30 minutes before the addition of 10 μM NE \[[@B17]\]. Culture supernatants were harvested 6 hours after treatment for ELISA and cells were homogenized in RNAiso plus 2 hours after treatment for RT-PCR. In order to evaluate the proliferation and migration of A549 cells under the inhibitors PKI and H-89, MTT assay and scratch wound healing assay were performed as previously described \[[@B34]-[@B36]\]. *In vivo* tumor model --------------------- C57BL6 female mice (4--6 weeks old) were purchased from the Laboratory Animal Center of Sichuan University. Male mice should be excluded for possible stress from mates in the cage. The animal experiments with the C57BL6 mice were consistent with protocols approved by the Institutional Animal Care and Treatment Committee of Sichuan University. The mice were maintained under pathogen-free conditions with food and water *ad libitum*, on 12 h/12 h day/night cycle, a temperature of 21--25°C, three mice per cage. B16F1 cells were trypsinized, centrifuged and then resuspended in serum-free medium. For implantation, tumors cells were subcutaneously inoculated in the right flanks of mice (5 × 10^5^ cells per mouse). Tumor measurements were made periodically with manual calipers every three days, and tumor volume was calculated applying the formula: π/6 × length × width^2^. At the end of the test, mice were sacrificed and tumors were excised, weighed and photographed. The serum from mice was harvested. Establishment of chronic stress *in vivo* and treatment with sunitinib ---------------------------------------------------------------------- Eight days after inoculation when the tumors reached an average diameter of 5 mm, mice were randomly assigned to four groups each consisting of six mice. The mice were narcotized by chloral hydrate i.p. (4%, 10 mL/kg) and then microosmotic pumps (Alzet model 1004, Durect, Cupertino, CA) were implanted subcutaneously on the left back of the mice for the establishment of chronic stress. The microosmotic pumps implanted in the body could keep functional and pump drugs contained continuously for up to 4 weeks. The pumps were filled with 100 μL normal saline containing 56 mM NE, 56 mM propranolol or both of them at a dose of 1 μmol/100 g/day \[[@B14]\]. Ascorbic acid (0.2%) was added as a preservative into every pump. The pumps full of just normal saline and ascorbic acid were used in the control group. The initiation of treatment with sunitinib by oral gavage (80 mg/kg/day \[[@B27]\]) was on the next day. The animals were sacrificed after 14 days of treatment. ELISA ----- The concentrations of VEGF, IL-8 and IL-6 proteins in culture supernatants or serum were detected using mouse or human ELISA Kits (Neobioscience, Beijing) following the manufacturer's protocol. The light absorbance at 450 nm was read in a luminescence plate reader (PerkinElmer, USA)**.** The values of concentrations were calculated by interpolation from a standard curve. Each experiment was repeated at least three times in duplicate. Immunohistochemistry for CD31, VEGF, β1-AR and β2-AR ---------------------------------------------------- Immunohistochemical studies were performed as previously described \[[@B26]\] using antibodies against CD31 (rat antimouse monoclonal antibody, diluted 1:300; BD Pharmingen, San Diego, CA, USA), VEGF (rabbit antimouse polyclonal antibody, diluted 1:200; Bioss Biotechology, Beijng), β1-AR & β2-AR (rabbit antimouse polyclonal antibody, diluted 1:300; Bioss Biotechology, Beijng). CD31 was stained on the frozen sections from B16F1 tumors for measuring microvessel density (**MVD**), VEGF on the formalin-fixed and paraffin-embedded sections from B16F1 tumors for comparing the expression levels among four groups and β1-AR and β2-AR on the slides of B16F1 cells for detecting the status of β-ARs in cells. Phosphate buffered saline was used instead of the primary antibody for negative controls. Assessment of microvessel density --------------------------------- MVD was assessed by choosing three areas with thickest microvessel distribution (hot spot) according to immunoreactivity for CD31 at low microscopic magnification (40 ×) and then counting the number of immunoreactive endothelial cells and microvessels from three 200 × high power fields in hot pot areas \[[@B37],[@B38]\]. RT-PCR analysis --------------- The assessment of VEGF, IL-8 and IL-6 gene expression was conducted using semiquantitative real-time reverse transcription-PCR (**RT-PCR**). Total RNA from A549 cells was isolated with RNAiso plus according to the RNA extraction protocols. Then the RNA was separated by 1% agarose gel electrophoresis and visualized by golden view to test the quality and integrity of RNA samples using the Gel Doc image system (Bio-Rad, Hercules, CA, USA). RT-PCR was conducted using One Step SYBR® PrimeScript™ RT-PCR Kit (Perfect Real Time) and amplified with CFX 96™ Real-Time System in C1000™ Thermal Cycler (Bio-Rad, USA). Glyceraldehyde-3-phosphate dehydrogenase (**GAPDH**) was applied as an internal positive control. The primers in this study were as follows: GAPDH: sense 5′- ACCACAGTCCATGCCATCAC -3′, antisense 5′- TCCACCACCCTGTTGCTGTA -3′; VEGF: sense 5′- TGGATCCATGAACTTTCTGCTGTC -3′, antisense 5′- TCACCGCCTTGGCTTGTCACAT -3′; IL-8: sense 5′-CTTTGTCCATTCCCACTTCTGA-3′, antisense 5′-TCCCTAACGGTTGCCTTTGTA T-3′; IL-6: sense 5′- ATGAACTCCTTCTCCACAAGCGC -3′, antisense 5′- GAAGAGCCCTCAGGCTGGACTG -3′ \[[@B12],[@B39]-[@B41]\]. The PCR cycler condition was according to the recommendations in the manufacturer's instructions. Reactions were performed in a 25-μL volume and each sample was run at least in duplicate. The levels of expression of VEGF, IL-8, and IL-6 mRNA in each sample were normalized to the GAPDH mRNA level. The relative expression of VEGF, IL-8, and IL-6 mRNA was calculated applying the comparative C~T~ method \[[@B18],[@B39]\]. Statistical analysis -------------------- The data are expressed as the mean ± SD. Changes in protein and mRNA levels of VEGF, IL-8 and IL-6, the averaged tumor volume and weight were calculated by one way analysis of variance (ANOVA) with an LSD post-hoc test and an unpaired student' t test using SPSS, version 15.0 (SPSS, Chicago, IL). A p value less than 0.05 was considered as statistically significant. Results ======= NE upregulates VEGF, IL-8, and IL-6 protein levels in culture supernatants of B16F1 (with or without sunitinib) and A549 cells, which can be blocked by propranolol ------------------------------------------------------------------------------------------------------------------------------------------------------------------- A NE dose-dependent and time-dependent increase in VEGF, IL-8 and IL-6 protein levels in culture supernatants of both B16F1 and A549 cells with a peak increase at the 6 hours time point and 10 μM concentration, which could be blocked by 10 μM propranolol (Figure  [1](#F1){ref-type="fig"}A-F). In A549 cells, treatment with 10 μM NE for 6 h caused a remarkable increase to 242.79 ± 19.86%, 331.56 ± 24.41% and 685.85 ± 34.72% (P \< 0.001) of control levels for VEGF, IL-8 and IL-6 protein levels, respectively (Figure  [1](#F1){ref-type="fig"}A-C). Likewise, in B16F1 cells, VEGF, IL-8 and IL-6 protein levels arrived at 185.15 ± 12.13%, 301.35 ± 24.98% and 294.40 ± 23.17% (P \< 0.001) of control levels in response to exposure to 10 μM NE for 6 hours (Figure  [1](#F1){ref-type="fig"}D-F). Overall, the increase could be most seen in both two cells at the NE concentration ranging from 0.1 to 10 μM since 3 hours after treatment. However, as time went on, the extent of the increase reduced 6 hours later. {#F1} In addition, the IC~50~ of sunitinib in B16F1 cells measured by cell proliferation assays was 3.35 μM. The results about B16F1 cells treated with sunitinib at the concentration equal to IC~50~ indicated that NE could also upregulate VEGF, IL-8, and IL-6 proteins with a peak increase at the 6 hours time, which could also be blocked by 10 μM propranolol (Figure  [1](#F1){ref-type="fig"}G-I). NE promotes tumor growth in the murine B16F1 model under the treatment of sunitinib and can be blocked by propranolol --------------------------------------------------------------------------------------------------------------------- Our results showed that NE speeded up the tumor growth rate in the B16F1 model treated with sunitinib. Similar with the results *in vitro* as above, the effect of NE could be blocked by propranolol (P \< 0.05) (Figure  [2](#F2){ref-type="fig"}A-E). NE increased the tumor weight by 51.65% compared with normal saline (0.99 ± 0.28 g VS 0.65 ± 0.27 g, P = 0.014) and 79.22% compared with the combination of NE and propranolol (0.99 ± 0.28 g VS 0.55 ± 0.08 g, P = 0.002) (Figure  [2](#F2){ref-type="fig"}D). {#F2} As shown in Figure  [2](#F2){ref-type="fig"}F, VEGF, IL-8 and IL-6 protein levels tested by the ELISA assay were upregulated by NE in the serum from the B16F1 model, which could be blocked by propranolol. NE increased VEGF, IL-8 and IL-6 protein levels by 155.77%, 417.77% and 586.21% compared with normal saline, respectively (P \< 0.001). NE stimulates tumor angiogenesis in the B16F1 model treated with sunitinib -------------------------------------------------------------------------- Immunohistochemical staining for VEGF on the formalin-fixed and paraffin-embedded sections showed a much stronger staining in the tumors of the group stimulated by NE than the other three groups (normal saline, propranolol and NE + propranolol) (Figure  [3](#F3){ref-type="fig"}A). There is no brown or yellow staining in negative control slides for VEGF wherein no primary antibodies were used (Figure  [3](#F3){ref-type="fig"}D). {#F3} Similar to VEGF, the significant increase in MVD, detected by immunohistochemical staining for CD31 on frozen sections, occurred in the tumors of the mice treated with sunitinib and stimulated by NE (P \< 0.05) (Figure  [3](#F3){ref-type="fig"}B-C). Beta1-AR and β2-AR are expressed in B16F1 cells ----------------------------------------------- Immunohistochemical staining for β1-AR and β2-AR on the slides of B16F1 cells was utilized to evaluate the status of β-AR via which NE affected cells. The results showed strong β1 and β2-AR immunoreactivivty located in the cytoplasma (Figure  [3](#F3){ref-type="fig"}E, F, respectively). The staining was invisible in negative control slides (not shown). NE upregulates VEGF, IL-8, and IL-6 gene expression in A549 cells ----------------------------------------------------------------- Although the up-regulation of VEGF, IL-8, and IL-6 protein levels by NE was described as above, we assessed the effect of NE on the expression of these three genes to further clarify the mechanism concerning the modulation of these three proteins in A549 cells. The results indicated that the levels of VEGF, IL-8, and IL-6 mRNA increased rapidly with a peak after 2 hours of treatment and decreased gradually thereafter in A549 cells exposed to 10 μM NE (Figure  [4](#F4){ref-type="fig"}A-C). {#F4} Beta-AR/cAMP/PKA signaling pathway contributes to the NE effect in A549 cells ----------------------------------------------------------------------------- For determining whether β-AR mediated the NE effect, phentolamine (α-AR antagonist) was used here to contrast with propranolol. We observed that, opposite to propranolol, phentolamine could not abrogate the NE-induced increase of VEGF, IL-8, and IL-6 mRNA levels in A549 cells (Figure  [4](#F4){ref-type="fig"}D). Isoproterenol (nonselective β-AR agonist), dobutamine (selective β1-AR agonist) and terbutaline (selective β2-AR agonist) upregulated VEGF, IL-8, and IL-6 mRNA levels, which indicated that both β1-AR and β2-AR mediated the NE-dependent effect (Figure  [4](#F4){ref-type="fig"}E-G). Moreover, comparing with β1-AR, β2-AR played a key role as a mediator special for the NE-induced stimulation of VEGF and IL-8 gene expression in A549 cells because terbutaline had a higher degree of up-regulation than dobutamine. Additionally, 8-CPT and forskolin (cAMP analogs) both raised VEGF, IL-8, and IL-6 mRNA levels implicating cAMP as a mediator. Lastly, H-89 (PKA inhibitor) nearly checked the effect of NE which could be just partially inhibited by PKI. To further identify the role of β-AR/cAMP/PKA signaling pathway in NE-treated A549 cells, the changes in VEGF, IL-8, and IL-6 protein levels tested by the ELISA assay related to mRNA levels as above were also analyzed. We observed similar changes in VEGF, IL-8, and IL-6 protein levels with their mRNA levels (Figure  [5](#F5){ref-type="fig"}A-D). {#F5} We also evaluated the proliferation and migration of A549 cells under the inhibitors PKI and H-89. The results showed that, different from PKI, H-89 inhibited the proliferation (Figure  [6](#F6){ref-type="fig"}A) and migration (Figure  [6](#F6){ref-type="fig"}B-C) of A549 cells. These results were consistent with the protein and gene levels of VEGF, IL-8 and IL-6 of A549 cells under PKI and H-89. {#F6} Discussion ========== In this study we showed that NE spurred tumor growth in the murine melanoma model treated with sunitinib by gavage *in vivo* and could be inhibited by propranolol. We also identified that NE upregulated VEGF, IL-8, and IL-6 protein levels in B16F1 cells in the presence or absence of the treatment with sunitinib at the concentration equal to IC~50,~ which was blocked by propranolol. In addition, NE-dependent up-regulation in both protein and gene levels of VEGF, IL-8, and IL-6 was observed in human lung adenocarcinoma cells in which β-AR/cAMP/PKA signaling pathway was proved as the important mechanism. Chronic stress has been acknowledged as an important factor affecting patients with cancer and the effect of chronic stress may be persistent during the process from diagnosis for cancer to death of cancer. The activation on sympathetic nervous system by stress gives rise to the increased level of catecholamines resulting in several biological effects via ARs such as VEGF-caused stimulation in angiogenesis, raised levels of cytokines including IL-8 and IL-6 \[[@B42]\]. These effects were also proved in our study and found as at least a part of factors attenuating the efficacy of sunitinib in preclinical models. In order to mimic chronic stress in patients, a wide variety of stress models in animals were established, e.g. addition of corticosterone to drinking water, transfer to a cold room at 4°C, subcutaneously administration with NE or β2-AR agonists, restraint procedure using open-ended Plexiglas cylindrical restrainers, social defeat, social isolation, unpredictable chronic mild stress, repeated social defeat, subcutaneous microosmotic pumps containing NE \[[@B12],[@B43]-[@B49]\]. However, some of stress models aforementioned have limitations more or less and thus induce unpredictable impacts on tests *in vivo*. For addition of corticosterone to drinking water, this test might not control the volume of water drunk by animals and thus the reliable uptake of corticosterone can not be evaluated especially when uptake of water was interrupted by the disorders in animals such as a heavy tumor burden \[[@B49]\]. For the restraint test, it was found in our laboratory that mice would adapt the open-ended Plexiglas cylindrical restrainers in the later stage. So the restraint test might not sustain enough stress if the observation in a test *in vivo* should be kept for a long time \[[@B45]\]. Seeing that microosmotic pumps (1004 type) are of the ability of pumping drugs contained incessantly for up to 4 weeks and exhibit reliable effects in mouse models, the pumps were taken into account in our research to deal with the short half life period of NE. It is well known that in clinic patients are under chronic stress after diagnosed with cancer prior to treatment. Thereby, in order to mimic patients in clinic as possible, sunitinib was administrated 30 minutes following NE in tests *in vitro*, and treatment with sunitinib was started 1 day after the implantation of pumps containing NE in tests *in vivo*. Tumor neovascularization or angiogenesis is closely related with proangiogenic factors such as VEGF, IL-8, IL-6, TGF and TNF released by tumor cells and immune cells. In analogy to tumors cells, lymphocytes and macrophages in the tumor microenviroment also express β-ARs triggered by NE with the following increased levels of VEGF, IL-8, and IL-6 \[[@B50]-[@B53]\]. The NE-induced up-regulation of VEGF, IL-8, and IL-6 protein levels was found in a number of human cancer cell lines such as colon cancer, nasopharyngeal cancer, ovarian cancer, prostate cancer and melanoma \[[@B7],[@B8],[@B13],[@B17],[@B18]\]. This effect of NE was identified in murine melanoma B16F1 cells and human lung adenocarcinoma A549 cells in our study. In addition, this phenomenon was also observed in murine colon cancer CT26 cells and some human cancer cells (e.g., nasopharyngeal cancer HNE1 & CNE2 cells, breast cancer MDA-MB-231 & MDA-MB-468 cells and colon cancer HT-29 & SW480 cells) in other studies in our laboratory (unpublished date not shown). However, to our knowledge, nothing is known of the influence of NE in cancer cells treated with sunitinib *in vitro*. Our date indicated that, in B16F1 cells treated with sunitinib at IC~50~ concentration, NE also increased VEGF, IL-8, and IL-6 protein expression in culture supernatants, which could be inhibited by propranolol. This result offered at least a mechanism for the difference in the efficacy of sunitinib between clinical and preclinical trials. It should be considered if sunitinib acts via some of its targets on B16 cells. Previous studies reported that B16 cells did not express VEGFR1, VEGFR2, VEGFR3 \[[@B54],[@B55]\], PDGFRα and PDGFRβ \[[@B56]\] but no more than 10% of B16 cells expressed c-Kit \[[@B57]\]. Whether sunitinib acts on B16 cells through the c-Kit target remains to be investigated in the further study. Chronic stress has been demonstrated to promote development and progression of tumors in several human cancer cells in xenografts including prostate cancer, ovarian cancer and breast cancer \[[@B9],[@B13],[@B15],[@B46],[@B58]\], whereas no date regarding the influence of chronic stress in cancer models under sunitinib *in vivo* has been reported so far. This study showed that consecutive NE pumped stimulated the growth of primary tumor in a mouse melanoma model and could be blocked by propranolol. This result provided a piece of evidence for the discrepancy in the efficacy of sunitinib between clinical and preclinical trials and was consistent with the results in the other studies in our laboratory (mouse colon cancer CT26 homograft and human colon cancer SW480 and HT-29 xenografts, unpublished date not shown). To further investigate stress-induced angiogenesis *in vivo*, we analysed the immunoreactivity for VEGF and CD31, counted the MVD and measured the protein levels of VEGF, IL-8 and IL-6 in mouse serums. As expected, in accordance with the results *in vivo* as mentioned in the previous paragraph, chronic stress promoted angiogenesis and neovascularization in B16F1 tumors, thus withstood the anti-angiogenic treatment of sunitinib. Interestingly, relatively low VEGF expression was found in tumor and endothelial cells while stronger VEGF expression usually found in peri-necrotic tumors cells mainly by reason of hypoxia as reported in the other study \[[@B59]\]. In clinic, the serum levels of VEGF, IL-8 and IL-6 have been suggested as potentially predictive markers for survival in cancer patients under sunitinib. Bauerschlag et al. \[[@B60]\] found that 18 cases with a decrease in VEGF serum concentration out of 29 ovarian cancer patients with sunitinib therapy had a longer progression-free survival (**PFS**) compared to 11 cases with an increase in VEGF serum concentration (10.5 VS 2.9 months). Likewise, the lower serum VEGF level was reported to be associated with longer PFS and objective response rate in patients under sunitinib with bevacizumab-refractory metastatic renal cancer \[[@B61]\]. Bellmunt et al. \[[@B62]\] announced that the low serum IL-8 level was related to long median time to progression in urothelial cancer patients receiving sunitinib as first-line treatment. Comparing with healthy donors, an increased level of IL-8 was detected in serums from medullary thyroid carcinoma patients with distant metastases \[[@B63]\]. Zhu et al. \[[@B64]\] reported that advanced hepatocellular carcinoma patients with high serum levels of IL-8 and IL-6 were of high mortality and rapid tumor progression after sunitinib. On the other hand, patients with a decrease level of IL-6 had better PFS and overall survival. Additionally, during sunitinib treatment, a more elevated IL-6 level was in correspondence with higher hazard of mortality or immediate progression. ARs are a family of G protein-coupled receptors, also called serpentine receptors whose ligands mainly include chemokines and neurotransmitters \[[@B31]\]. Since the expression of β-ARs was observed in human lung adenocarcinoma A549 cells \[[@B65],[@B66]\], only an immunohistochemical analysis for β-ARs in B16F1 cells was carried out. Hegener et al. \[[@B65]\] also found that the internalization and endocytosis of β2-AR in A549 cells were stimulated by terbutaline (selective β2-AR agonist) and forskolin (cAMP analogs), whereas blocked by propranolol. In our study, the strong expression of β-ARs located in the cytoplasma and there was no difference of staining intensity between β1-AR and β2-AR discerned with naked eyes. This finding in our study provided the basis for following research on the β-AR/cAMP/PKA pathway in B16F1 cells. Considering ARs play a key role mediating the effect on tumors induced by chronic stress and endow tumor cells the potential to respond to neurotransmitters, few scholars suggest the receptor-based interference of intracellular ARs signaling pathway as a new approach to resist this effect \[[@B9],[@B42],[@B67],[@B68]\]. Powe et al. \[[@B69]\] found, in breast cancer, β2-AR strongly immunoreactive in cases with a luminal phenotype and good clinic outcome while α1b-AR and α2c-AR over-expressed in basal-like phenotypes of poor prognosis. So ARs might be supposed to be potential predictors for survival and probable indicators for targeted therapy with AR blockers. In the present research, it was approved in A549 cells that the NE-induced up-regulation in both protein and gene levels of VEGF, IL-8 and IL-6 was chiefly mediated by β-AR/cAMP/PKA signaling pathway which had been found to play a key role in mouse xenografts of melanoma and ovarian cancer \[[@B9],[@B17]\]. The stimulation of β-ARs by neurotransmitters induces multiple signaling pathways of which the most important one approved is cAMP/PKA/**CREB** (cAMP response element binding protein). Then the activation of CREB, a transcription factor, initiates the arachidonic acid cascade, the Src/STAT and the EGFR pathways followed by a wide variety of biological effects \[[@B9],[@B70]\]. Conclusions =========== Taken together, our data support the hypothesis that exogenous norepinephrine gives rise to the attenuation in the efficacy of sunitinib in a mouse melanoma model and provide a reason for the discrepancy of the efficacy of anti-angiogenic drugs between clinical and preclinical results. The combination of anti-angiogenic agents with β-AR antagonists such as propranolol, a drug for cardio-vascular disease for decades, might eliminate the ineffectiveness of anti-angiogenic agents alone and enhance their efficacy in some types of tumors, which has yet to be approved in prospective randomized controlled trials in clinic. Competing interests =================== The authors declare no conflict of interests. Authors' contributions ====================== YJ and YQW designed the procedure of the study. GHD carried out the plan and drafted the manuscript. JL, JZ and YW participated in cell culture, animal experiments and immunohistological analysis. XCP assisted in RT-PCR and statistical analysis. YJ and YQW supervised the whole experimental work and revised the manuscript. All authors read and approved the manuscript. Acknowledgement =============== This work was supported by the National Natural Sciences Foundation of China (81172202).

Background {#Sec1} ========== Pancreatic cancer is the most lethal of the common human malignancies and the fifth leading cause of cancer-related death in Japan \[[@CR1]\]. Even after pathologically curative resection, most patients have disease recurrence \[[@CR2]\]. Multimodality therapy, such as the combination of surgery and chemotherapy, has been important for pancreatic cancer \[[@CR2]\]. Pancreatic cancer often develops both hematogenous and lymphatic metastases, with metastasis usually occurring in the liver, lung, peritoneum, or bones \[[@CR3]\]. Gastric involvement is occasionally observed in pancreatic cancer patients, although it results not from metastasis but from direct invasion \[[@CR4]\]. Gastrointestinal metastasis from pancreatic cancer is quite rare \[[@CR5]\]. In addition, metastatic tumors from pancreatic cancer are usually solid tumors. As we are not aware of a similar case in the English literature, we would like to present a case of cystic gastric metastasis from pancreatic cancer. Case presentation {#Sec2} ================= A 58-year-old male patient was found to have hyperamylasemia upon medical examination. A contrast-enhanced computed tomography (CT) scan revealed a 30-mm low-density solid tumor in the tail of the pancreas (Fig. [1a](#Fig1){ref-type="fig"}). Positron emission tomography--computed tomography (PET-CT) showed a maximum standardized uptake value (SUV max) of the tumor of 3.5 to 3.7 (Fig. [1b](#Fig1){ref-type="fig"}). There was no distant metastasis on enhanced CT or PET-CT. The serum levels of carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA 19-9) were not elevated (CEA = 1.0 ng/mL, CA 19-9 = 5.8 U/ml). We diagnosed a resectable pancreatic tail cancer. Moreover, prior to the initial surgery, there was no tumor lesion in the gastric antrum (Fig. [1c](#Fig1){ref-type="fig"}), and endoscopic ultrasonography (EUS) was not performed.Fig. 1Preoperative imaging of the primary pancreatic cancer. Enhanced computed tomography (CT) showed a 30-mm solid tumor in the tail of the pancreas (**a**, arrowheads). Positron emission tomography--CT showed abnormal accumulation in the tumor (**b**, arrowheads). There was no gastric before the initial surgery (**c**) The patient underwent distal pancreatectomy with D2 lymph node dissection and splenectomy (Fig. [2a](#Fig2){ref-type="fig"},[b](#Fig2){ref-type="fig"}). The intraoperative peritoneal lavage cytology was negative. Pathological diagnosis was tubular adenocarcinoma (moderately \> well-differentiated), Pbt, TS2, tumor diameter of 39 × 26 mm, nodular type, pT3, int, INFβ, ly0, v1, ne2, mpd0, pCH0, pDU0, pS1, pRP1, pPVsp1, pAsp1, pPL0, pOO0, pPCM0, pBCMX, pDPM0, pN0, M0, CY0, pStage IIA, D2, and pR0 according to the 7th edition of Japanese Pancreatic Society staging system and T3N0M0 Stage IIA (pathological grade G2) in the 7th edition of the AJCC/UICC staging system (Fig. [2c](#Fig2){ref-type="fig"}). The patient received adjuvant chemotherapy with 16 courses of gemcitabine.Fig. 2Pathological findings of the primary pancreatic cancer. Macroscopically, the size of the primary pancreatic cancer was 3.9 × 2.6 cm (**a**, **b**). Microscopically, pathological diagnosis was moderately \> well-differentiated tubular adenocarcinoma (**c**) Abdominal enhanced CT showed a 15-mm cystic tumor in the gastric antrum 4 years after the surgery (Fig. [3a](#Fig3){ref-type="fig"}). Tumor markers were not elevated. Because the cystic lesion had grown to 25 mm, 6 months later (Fig. [3b](#Fig3){ref-type="fig"}), upper gastrointestinal endoscopy and EUS were performed. Endoscopy revealed no epithelial lesion in the antrum (Fig. [3c](#Fig3){ref-type="fig"}); however, EUS showed a submucosal tumor (Fig. [3d](#Fig3){ref-type="fig"}). EUS-guided fine-needle aspiration biopsy (EUS-FNA) was performed, and light brown serous cystic fluid was collected. Cytological diagnosis of the cyst fluid was class III. Interestingly, CEA of the cyst fluid was 6770 ng/mL, whereas serum CEA and CA19-9 levels were 1.5 ng/mL and 6.6 U/mL, respectively. Thus, the cystic tumor could have been gastric metastasis from pancreatic cancer. The tumor existed at the anterior wall of the stomach; so, it was not considered as direct invasion from the pancreatic cancer. Because there was no other metastasis or recurrence, we planned resection of the cystic tumor. After laparotomy and adhesiolysis, the cystic tumor was detected at the anterior wall of the stomach (Fig. [4a](#Fig4){ref-type="fig"}). Partial gastrectomy was performed, and the defect of the stomach was closed in the minor axis direction (Fig. [4b](#Fig4){ref-type="fig"}). The intraoperative peritoneal lavage cytology result was negative. There were no postoperative complications, and the postoperative course was good.Fig. 3Preoperative imaging of the cystic tumor in the gastric antrum. Enhanced CT showed a 15-mm cystic lesion of the gastric antrum (**a**, arrowheads) 4 years after the initial operation. Six months later, the cystic lesion had grown up to 25 mm (**b**, arrowheads). Endoscopy revealed no epithelial lesion (**c**, arrowheads). Endoscopic ultrasonography showed a submucosal tumor (**d**)Fig. 4Operative findings of the cystic tumor in the gastric antrum. Laparotomy showed the cystic tumor at the anterior wall of the stomach (**a**). Partial gastrectomy was performed, and the defect of the stomach was closed (**b**) Macroscopically, there was a 25-mm round, cystic tumor (Fig. [5a](#Fig5){ref-type="fig"},[b](#Fig5){ref-type="fig"}). Microscopically, adenocarcinoma arising from the mucinous atypical acinar epithelium was observed from the intrinsic muscle layer of the stomach wall to the subserosal tissue (Fig. [5c](#Fig5){ref-type="fig"}). Adenocarcinoma mainly took on a cystic structure, in contrast to primary pancreatic cancer, which presented as a solid adenocarcinoma tumor with rich stromal cells. Some parts of the tumor represented a papillary structure with irregular ducts (Fig. [5](#Fig5){ref-type="fig"}d). There was no tumor proliferation in the epithelium of the gastric mucosa. Thus, the pathological diagnosis was adenocarcinoma, consistent with metastasis of pancreatic cancer.Fig. 5Pathological findings of the cystic gastric metastasis. Macroscopically, there was a 25-mm round cystic tumor (**a**, **b**). Microscopically, the cystic tumor existed in the submucosal layer (**c**). The tumor comprised tubular adenocarcinoma cells (**d**). Invasion of the adenocarcinoma cells did not reach the epithelium The patient received adjuvant chemotherapy with S-1. The patient has survived without obvious recurrence or metastasis for 13 months after the partial gastrectomy. Generally, metastasis or recurrence of pancreatic cancer is found in the liver, lymph nodes, peritoneum, and lung \[[@CR4], [@CR6]\]. Involvement of the stomach due to direct invasion by cancer of the pancreatic body and tail has occasionally been observed \[[@CR4]\], but gastric metastasis from pancreatic cancer is quite a rare event. Possible mechanisms of cancer involvement of the stomach are direct invasion, intraoperative seeding, hematogenous metastasis, lymphatic metastasis, and intraluminal or intramural dissemination \[[@CR7]\]. In our case, a growing cystic tumor in the gastric antrum wall was detected. Because the tumor was located in the submucosal layer and separated from the space where the primary pancreatic cancer existed or the pancreatic cut ended, we diagnosed it as hematogenous metastasis to the stomach. Sasajima et al. indicated that it was difficult to distinguish between gastric metastasis and direct invasion of the stomach in advanced cancer \[[@CR5]\]. Autopsy studies often lead to the diagnosis of metastasis when cancer cells are detected histologically in other organs \[[@CR8]\]. Gastric metastasis from pancreatic cancer was described as a rare condition in those studies \[[@CR4], [@CR6]\]. Campoli et al. reported 20 patients with metastatic cancer of the stomach and indicated that the primary sites were the esophagus, skin, lung, cervix, breast, sigmoid colon, and testis, but not the pancreas \[[@CR9]\]. Oda et al. reported that 347 cases of gastric metastases from solid malignant tumors were identified from a series of 6380 autopsy reports; thus, the incidence of gastric metastasis was 5.4% \[[@CR10]\]. In their study, the most common primary malignancy of gastric metastasis was lung cancer, followed by breast and esophageal cancers \[[@CR10]\]. Additionally, they performed autopsies on 209 patients with pancreatic cancer and reported only 2 cases with gastric metastasis \[[@CR10]\]. These results indicate that the pancreas is a rare primary lesion of gastric metastasis. It is also difficult to distinguish a gastric metastasis of pancreatic cancer from a heterochrony primary pancreatic cancer originated from aberrant pancreas in the stomach. Gillou et al. described that the possibility of a tumor from ectopic pancreatic tissue origin is acceptable only if the following three conditions are met: (1) the tumor must be found within or close to the ectopic pancreatic tissue, (2) direct transition between pancreatic structures and the carcinoma must be observed, and (3) the nonneoplastic pancreatic tissue must comprise, at least, fully developed acini and ductal structures \[[@CR11]\]. Because all of the above conditions are not found in the tumor microscopically in our case, we thought the tumor was not a primary pancreatic cancer originated from ectopic pancreas in the stomach. In addition, metastatic tumors from pancreatic cancer, such as liver metastasis and lymph node metastasis, are usually solid tumors. However, our case presented with a cystic metastatic tumor. English language publications were systematically searched to find similar cases in the National Library of Medicine (MEDLINE \[PubMed, available at: <http://www.pubmed.com>/\]) published between 1970 and 2017 using the medical terms "pancreatic cancer," "gastric metastasis," and "cystic tumor". However, no similar cases were found. In addition, no cases of cystic metastasis of the stomach in other primary malignancies were found. It is known that a metastatic lung tumor can present as a cystic lesion \[[@CR12]\]. The mechanism of how it became a cystic tumor has been considered such as ischemic necrosis of the center of the tumor. Several papers have reported good prognoses after resection for solitary metastasis from pancreatic cancer, in the lung or remnant pancreas, which developed long after the initial pancreatectomy for the primary lesion \[[@CR13]--[@CR15]\]. Therefore, partial gastrectomy for this case seems to be a feasible operation. Conclusions {#Sec3} =========== The gastric metastasis from pancreatic cancer is quite rare. In addition, the reported patients developed a cystic submucosal tumor. In carefully selected patients, we believe that surgical excision of a solitary metastasis like our case may prolong survival time. CA19-9 : Carbohydrate antigen 19-9 CEA : Carcinoembryonic antigen CT : Computed tomography EUS : Endoscopic ultrasonography FNA : Fine-needle aspiration biopsy PET : Positron emission tomography SUV : Standardized uptake value NU, DH, and SN prepared the manuscript. TY, MT, and YK contributed to the acquisition of data and participated in drafting the article. KA, KY, TM, and HO reviewed it. AC, FM, and HB were responsible for the manuscript. All authors read and approved the final manuscript. Consent for publication {#FPar1} ======================= Written informed consent was obtained from the patient for the publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Ethics approval and consent to participate {#FPar2} ========================================== None Competing interests {#FPar3} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar4} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Background {#s1} ========== Uncontrolled hypertension has become a major health problem worldwide. Previous study reveals that the prevalence of uncontrolled hypertension in both developed and developing countries remains high \[[@R01]\]. Many studies suggest that uncontrolled hypertension leads to a half of coronary heart disease and cerebrovascular disease incidents \[[@R02]\]. Moreover, it contributes to 7.5 million deaths per year (13% of all death causes) \[[@R03]\]. Therefore, achieving blood pressure target is essential to reduce the mortality and morbidity of hypertension related diseases \[[@R04], [@R05], [@R06]\]. Many factors may contribute to uncontrolled hypertension, such as antihypertensive drug combination, patients\' compliance, and genetic variants in renin angiotensin aldosterone system. Renin is an aspartyl protease produced and secreted by juxtaglomerular cell in kidney \[[@R07]\]. This enzyme induces enzymatic cascade that generates angiotensin II peptide as the main effector in the system of renin angiotensin aldosterone as blood pressure regulator \[[@R08]\]. Plasma renin level has been used for many years to determine the responses to antihypertensive therapy. The transcription level of renin is determined by many factors such as the presence of distal enhancer in renin gene and polymorphism in this area. Studies on human chorionic cultured cell suggest that renin distal enhancer plays a pivotal role in renin gene expression. Deletion of this area resulted in 10 fold loss of enhancer activity \[[@R09]\]. Fuchs, *et al*. suggest that SNP in C-5312T of renin enhancer increases transcription level of renin gene by 45% in -5312T compared to -5312C9 \[[@R10]\]. On the other hand, Pan L., *et al*. (2003) report that renin gene expression is regulated by Sp1/Sp3 proteins, mutations in these proteins decrease renin levels up to 40% in mice \[[@R09]\]. This phenomenon suggests that Sp1 acts as a positive regulatory protein that binds renin enhancer. Thus, this study has been designed to investigate whether renin enhancer has a Sp1-binding site, and to examine the influence of genetic variant of C-5312T on Sp1 binding pattern to renin enhancer. Methodology {#s2} =========== *Detection of polymorphism*: {#s2a} ---------------------------- SNP within renin distal enhancer of human renin gene was identified by direct sequencing of genomic DNA obtained from 34 hypertensive patients. Each sample was amplified by these primers 5\'-CGTAGTGCCATTTTTAGGAAC-3\' and 5\'-AACACCA AAGCAGGCTTAA-3\'. All samples were sequenced by automatic sequencing method (Macrogen) and the genetic variants were analyzed using Genescane software (Applied Biosystem). The existence of Sp1-binding site of these sequences was determined based on Sp1-binding site consensus from CORE Jaspar database (<http://jaspar.genereg.net/cgi-bin/jaspar_db.pl>). *DNA Modeling*: {#s2b} --------------- We generated 3D structural models of DNA from sequences of renin enhancer (5\'gtgcccccagattaagcctg3\') normal and (5\'gtgccctcagattaagcctg3\') mutant using 3D-DART provided by haddock (<http://haddock.science.uu.nl/services/3DDART/>) \[[@R11]\]. *Molecular Docking*: {#s2c} -------------------- Models of SP1 protein (1R8U) and ER-Estradiol (1A52) were retrieved from protein structure databases, PDB (<http://www.rcsb.org>). Binding interaction among molecule complex of ER-Estradiol, Sp1, and DNA was done by using Escher NG in VEGAZZ. Further for detailed analysis, we used Hex Protein Docking to predict interactions between Zinc Finger Domain 2 (Zf-2) of SP1 and DNA of renin enhancer. Hex is FFT-based approach that was first used as a rapid way to calculate shape complementarity within a 3D Cartesian grid \[[@R12]\]. Thereafter, the complex of docking result was analyzed using ligand Scout to know the amino acids and nucleotides responsible for the interaction \[[@R13]\]. *Visualization*: {#s2d} ---------------- All the visualization of the structure files was done using PyMol ([www.pymol.org](www.pymol.org)) and YASARA molecular graphics system. Results {#s3} ======= Proportions of genetic variant of the patients were CC (47.1%), CT (44.1%), and TT (8.8%). Sp1 protein is a trans-acting transcription factor-1, has 3 zinc finger domains which serves to bind DNA. Since the Sp1 bound to DNA in the GC rich motif, then we searched the motif in renin enhancer of 34 hypertensive patients that we analyzed. The data showed that renin enhancer of the hypertensive patients contained Sp1-DNA binding motif. Further, we found two genetic variations in Sp1-binding site from this study, SNPs in C-5312T ([Figure 1](#F1){ref-type="fig"}). Sp1 is activated by a complex of ER alpha receptor (ER) and 17β-Estradiol, and then binds DNA to stimulate transcription of the gene being regulated. Therefore, we used the ER-Estradiol-Sp1 complex to elucidate pattern of the regulator bind to renin enhancer by using docking methode. The docking results showed that Sp1 most preferably bind to DNA in GC rich sequence. Sp1 was bound in the center of the GC rich sequence and form bond with C base in -5312C variant. However, Sp1 was bound in downstream of the GC rich sequence in -5312T variant. Taken together the phenomenon, indicated that genetic variations C-5312T of renin enhancer causes differences of Sp1 binding to renin enhancer ([Figure 2](#F2){ref-type="fig"}). Furthermore, we analyzed more detail the differences of Sp1 binding pattern to SNPs in C-5312T of renin enhancer by docking Zinc Finger domain-2 of SP1 with 21 BP DNA containing the C-5312T SNP. The result indicated that change from T to C causing displacement of Sp1 binding site from center to dowstream of GC rich sequence, and Sp1 loss of contact with bases -5312 ([Figure 3](#F3){ref-type="fig"}). The shift in the Sp1 binding suggested that Sp1 loos contact with the center of rich GC sequence, which was very likely not be able to stimulate transcription of renin gene. This result explained why the transcription level of renin gene in -5312T was higher by 45% compared to -5312C. Discussion {#s4} ========== Renin angiotensin aldosterone system (RAAS) plays a pivotal role in blood pressure regulation \[[@R08]\]. Renin, that catalyzes angiotensinogen to angiotensin I, plays an essential role on blood pressure regulation and electrolyte homeostasis \[[@R14]\]. Plasma renin level has been used for many years to determine the responses to antihypertensive therapy. The transcription level of renin is determined by many factors such as the presence of distal enhancer in renin gene and polymorphism in this area. Several studies have been conducted to identify the single nucleotide polymorphism (SNP) in renin gene and its association with hypertension \[[@R15], [@R14], [@R16]\]. These studies suggested that polymorphism in intron 1 of renin gene is strongly correlated with hypertension in Caucasian \[[@R15]\]. Furthermore, SNP in renin distal enhancer at -5312 is associated with hypertension in whites \[[@R14], [@R16]\]. This SNP also becomes a major determinant of patients\' response to angiotensin-receptor blocker therapy in Ireland and Japan. This study shows two polymorphisms in renin distal enhancer, in -5190 site and -5312 site. Previous study has reported the functionality of polymorphism in -5312 in Japan and French \[[@R10], [@R17]\]. Sp1 binds to GC rich promoter especially in CACCC box and -5312 area might be identified by Sp1. This transcription factor will be activated by ER alpha and 17β-Estradiol (E2) \[[@R18]\]. Thus, polymorphism within C-5312T might contribute to the reduction of renin transcription. Fuchs, *et al*. stated that polymorphism in this site increases transcription level by 45% in -5312T compared to that of -5312C \[[@R10]\]. However, these studies do not explain the mechanism of transcriptional level increase. Therefore, according to all of the results of the analyses, we concluded that Genetic variant of C-5312T has changed binding pattern of Sp1 to renin enhancer, which is very likely to influence Sp1 activity to stimulate expression of renin gene. Morever, this study helped to explain why SNP -5312C reduces expression levels of renin gene. Conclusion {#s5} ========== This study shows that Sp1 most preferably binds to DNA at the GC rich region. The genetic variants C-5312T has changed binding pattern of Sp1 to renin enhancer. This is likely to influence Sp1 activity to stimulate the expression of renin gene. **Citation:**Lukitasari *et al*, Bioinformation 9(20): 1010-1012 (2013) {#F1} {#F2} {#F3}

Introduction {#Sec1} ============ The occurrence of symptomatic maternal arrhythmias during pregnancy is a cause for concern for the well-being of both the mother and the foetus. As in the non-pregnant population, sustained symptomatic arrhythmias should be treated. Treatment depends upon the diagnosed or suspected specific arrhythmia. In case of supraventricular arrhythmias, electrical cardioversion (ECV) is applied when physical treatment such as sinus carotid massage or Valsalva manoeuvres and drug therapy fail or in life-threatening situations when the patient is haemodynamically unstable \[[@CR1]--[@CR8]\]. No large-scale studies concerning the safety of ECV in pregnancy are available. This article describes two cases and gives an overview of the reported efficacy and safety of ECV during pregnancy. Cases {#Sec2} ===== A 34-year-old woman without a relevant medical or obstetric history is pregnant of her third child. The course of the pregnancy is uncomplicated until a gestational age of 21 + 1 weeks, when she experiences progressive palpitations and feelings of agitation. Physical examination reveals an irregular pulse rate of 160 beats/min. Blood pressure (BP) is 127/86 mmHg, and there is no peripheral oedema. Auscultation of heart and lungs is normal. Anaemia and hyperthyroidism are excluded. The electrocardiogram (ECG) shows atrial fibrillation with a ventricular rate of 194 beats/min, and slight repolarisation disturbance in leads II, III, aVF and V3--V6. Echocardiography demonstrates no structural or functional abnormalities of the maternal heart. Treatment with a β-blocker is started to control ventricular rate. This does not result in conversion to sinus rhythm, and ECV (50 J) is successfully applied. The foetal condition is monitored by ultrasound, and no adverse effects are seen. The patient is discharged the same day without medication. At a gestational age of 40 + 1 weeks, she gives birth to a healthy boy of 3620 g. During 24 h of postpartum observation, no complications are noted. Paediatric evaluation shows no abnormalities. The second patient is a 29-year-old primigravida with no relevant medical history and with antenatal care provided by a community midwife. At a gestational age of 34 weeks, she is referred to the hospital with palpitations. Maternal pulse rate is 160 beats/min, BP 118/79 mmHg, and SpO~2~ 98%. Anaemia and hyperthyroidism are excluded. Auscultation of heart and lungs reveals no abnormalities, and oedema of lower extremities is absent. The ECG shows a small complex tachycardia and a long r-p interval with a positive p wave in leads I and aVL and a negative p wave in leads II, III and aVF. Occasionally, 2:1 block is shown to the ventricle which excluded an atrioventricular re-entry tachycardia (AVRT); however, atrioventricular nodal re-entry tachycardia (AVNRT) and adenosine-sensitive atrial tachycardia cannot be ruled out. The diagnosis is most likely ectopic atrial tachycardia. Since adenosine and verapamil did not terminate the tachycardia, ECV is performed. ECV (100, 200 and 360 J) decreases the heart rate to some extent, but conversion to sinus rhythm does not occur. For additional rate control, digoxin and propranolol are given. Echocardiography shows normal systolic LV function and mild mitral and tricuspid insufficiency, which can be physiological in the third trimester of pregnancy. Foetal non-stress test and biophysical profile remain reactive throughout pharmacological treatment and ECV procedures. Maternal atrial tachycardia persists with variance in conduction, resulting in fluctuations of ventricular rate. Because of her declining clinical condition, echocardiography is repeated and reveals a decrease in left ventricular function probably due to tachycardiomyopathy. At a gestational age of 36 + 6 weeks, the foetal non-stress test shows deceleration of the foetal heart rate and a caesarean section is performed for foetal indications. A healthy baby girl of 3095 g is born. Postpartum drug-refractory symptomatic paroxysmal atrial tachycardia persists, and 18 months after delivery, a catheter ablation was performed. A subsequent pregnancy was uneventful. Discussion {#Sec3} ========== Cardiac arrhythmias are frequently diagnosed for the first time during pregnancy. An important risk factor for arrhythmias during pregnancy is the presence of organic heart disease. In most cases, there is no previous history of heart disease, and fortunately, arrhythmias that occur during pregnancy are almost always benign and no treatment is necessary. Sustained symptomatic arrhythmia, however, requires treatment. The incidence of maternal arrhythmias is 1.2 per 1000 pregnant women, of which 50% is asymptomatic. The incidence increases in the third trimester. During pregnancy, impressive haemodynamic, neuro-cardiogenic and autonomic changes occur, and it is supposed that the hyperdynamic circulation and the increased sensitivity of adrenergic receptors play a provocative role \[[@CR1], [@CR3], [@CR5], [@CR9]\]. Symptoms are dyspnoea, palpitations, dizziness and (pre)syncope \[[@CR3], [@CR5]\]. Most common arrhythmias in pregnancy are AVNRT, AVRT, atrial tachycardia and atrial fibrillation \[[@CR3]--[@CR6]\]. Atrial flutter and atrial fibrillation are often associated with hyperthyroidism. Diagnosis can be hampered by the fact that symptoms of shortness of breath, palpitations, dizziness and presyncope frequently occur during normal pregnancy and increase with gestational age \[[@CR3]\]. The presence of an arrhythmia is likely to cause an increase or even a new onset of these symptoms. As a result, the patient feels unwell. Clinical assessment and ECG investigation are mandatory for an accurate diagnosis of the arrhythmia \[[@CR5]\]. Echocardiography is essential to exclude structural and functional heart disease. Treatment of supraventricular tachycardias in pregnancy is the same as for the non-pregnant population. Initial treatment consists of stimulation of the vagus nerve by means of carotid massage or Valsalva manoeuvres. First-line pharmacological treatment in case of failure of physical procedures is with adenosine. Second choice is verapamil but only after the first trimester of pregnancy and only in acute circumstances \[[@CR4]--[@CR6], [@CR9]\]. A low dose of β-blockers can be effective treatment for supraventricular extrasystoles or tachycardia**.** When drugs fail or in case of life-threatening symptoms as shock of pulmonary oedema, ECV is indicated \[[@CR2]--[@CR6], [@CR9], [@CR10]\]. We searched the English literature from 1965 and found 44 case reports that describe the use of ECV during pregnancy. There is considerable variation in specific arrhythmias for which ECV is applied and required energy varied from 50 to 400 J. Successful ECV after one or more attempts is reported in 41 pregnant women (93.2%). ECV in the non-pregnant population is reported to be successful in 42--92% \[[@CR11], [@CR12]\]. ECV success depends on the type and length of the arrhythmia, cardioversion method, voltage and type of energy. In three cases, a normal sinus rhythm is acquired only after additional drug treatment \[[@CR13]--[@CR15]\]. Two maternal deaths are reported shortly after cardioversion \[[@CR16], [@CR17]\]. In both cases, maternal death is attributed to the severity of underlying heart disease and no relation with cardioversion has been assumed. There are limited data on perinatal outcome: 22 cases do not report pregnancy outcome \[[@CR18]--[@CR20]\], and 13 cases report an uncomplicated continuation of pregnancy resulting in term vaginal deliveries \[[@CR2], [@CR3], [@CR13]--[@CR15], [@CR17], [@CR21]--[@CR24]\]. Three cases of a spontaneous preterm delivery are reported \[[@CR8], [@CR21], [@CR25]\]. However, as in two cases where the delivery is 4 to 8 weeks after cardioversion, a direct relation with ECV is unlikely \[[@CR8], [@CR21]\]. In two cases, foetal distress directly after the cardioversion is noted, necessitating an immediate caesarean section at 37 and 28 weeks of gestational age, respectively \[[@CR10], [@CR26]\]. Both women are known with underlying cardiac disease. Both neonates are born healthy. The first case shows a notably hypertonic uterus, while the current used was only 50 J. It was hypothesised that the hypertonic uterus resulted in foetal bradycardia, which could be a direct consequence of the cardioversion. Caution should be applied since the hyperaemic uterine muscle as well as the amnion fluid are excellent conductors of electricity \[[@CR10], [@CR27]\]. However, the uterus is usually not involved in the ECV trajectory, and only a minimum amount of current reaches the uterus. It is supposed that there is little effect on the foetal heart because of the high fibrillation threshold of a small heart \[[@CR1], [@CR8], [@CR10], [@CR20], [@CR28], [@CR29]\]. Possible explanations are placement of the pads over the apex beneath the left breast and/or the relatively large third trimester uterus. These cases underline the importance of carrying out the procedure with facilities available for foetal monitoring and emergency caesarean section \[[@CR17]\]. Conclusion {#Sec4} ========== Provided that a multidisciplinary approach, continuous foetal heart rate monitoring and the possibility to perform a caesarean section are applied, it can be concluded that cardioversion is a safe and effective treatment for maternal tachycardia in pregnancy. **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.

1. Introduction {#sec1} =============== Oxidative stress represents an imbalance between the production of reactive oxygen species (ROS) and the detoxification of their reactive intermediates. ROS, such as hydroxyl radicals, superoxide anions, hydrogen peroxide, and singlet oxygen, are primarily generated by mitochondria in most mammalian cells and are generally regarded as the toxic side-products of cellular metabolism \[[@B1]--[@B3]\]. ROS are normally detoxified by a variety of antioxidant enzymatic scavengers, including superoxide dismutase (SOD), catalase, glutathione S-transferase (GST), and glutathione peroxidase (GPX) \[[@B4]\]. Mitochondria are a major site of ROS-induced oxidative damage \[[@B5], [@B6]\]. ROS generated by mitochondria are hypothesized to damage key mitochondrial components such as mitochondrial DNA (mtDNA), mitochondrial membranes, and respiratory chain proteins and nuclear DNA that affect mitochondrial function. mtDNA is a circular, closed, double-stranded molecule and is not protected by histones. Therefore, mtDNA is more susceptible to DNA insults in comparison with nuclear DNA. Most of mtDNA mutations are characterized by heteroplasmy, which is defined as the presence of more than one an organellar genome within a cell or tissue from a single individual. As the percentage of mutant alleles increases, the mitochondrial bioenergetic defect becomes more severe. The expression of disease depends on the percentage of mutant alleles. It has been widely considered that aging is the process of accumulated oxidative damage caused by ROS \[[@B7], [@B8]\]. This damage accumulates over time, causing mitochondrial dysfunction and an associated decrease of energy production, and results in tissue dysfunction. ROS production increases with age and it is known that oxidative stress and associated mitochondrial dysfunction play an important role in aging and age-related diseases \[[@B1], [@B2]\]. Age-related hearing loss (ARHL), which is also called presbycusis, is the progressive loss of hearing associated with aging and is the most common sensory disorder in the elderly population \[[@B9]--[@B11]\]. ARHL afflicts approximately half of the people over 65 years of age in the United States \[[@B12]\]. The prevalence of the ARHL is expected to increase as the elderly population grows \[[@B9], [@B13], [@B14]\]. It has been proposed that ARHL is associated with many factors, including environmental, medical, and hereditary factors \[[@B12], [@B15]\]. So far, no effective treatment has been found for this age-related disorder. Many studies have been conducted based on the assumption that age-related oxidative stress and mitochondrial dysfunction could be an underlying pathology of ARHL as well as other age-related diseases. In this review, we will focus on previous research concerning the role of the oxidative stress and mitochondrial dysfunction in the pathology of ARHL in both animal models and humans and introduce concepts that have recently emerged as potential mechanisms for the development of ARHL. 2. Pathological Findings in ARHL {#sec2} ================================ Sound waves travel down the external ear canal and cause the tympanic membrane to vibrate. The ossicles in the middle ear link the vibrating tympanic membrane to the cochlea, the auditory end organ of the inner ear. The cochlea is filled with fluid that vibrates in response to the movement of the ossicles. The inner and outer sensory hair cells are located within a core component of the cochlea, the organ of Corti. When a sound pressure wave travels from the basal turn to the apical turn of the cochlea, the basilar membrane vibrates \[[@B16]\]. Displacement of stereocilia, the mechanosensing organelles of the hair cell, in association with the vibration of the basilar membrane, opens transduction ion channels, allowing entry of potassium ions from the endolymph produced by the stria vascularis. This transduction current then activates voltage-dependent calcium channels along the hair cell lateral wall and base \[[@B17]\]. The inner hair cells release the neurotransmitter glutamate to encode acoustic signals for the adjacent spiral ganglion neurons (SGNs), which are the primary auditory neurons \[[@B18]\]. Based on postmortem pathological analysis, ARHL in humans is generally classified into 3 types: sensory hearing loss (loss of sensory hair cells), neuronal hearing loss (loss of SGNs), and metabolic hearing loss (atrophy of the stria vascularis) \[[@B9], [@B19]\], although it is now well established that most cases of ARHL exhibit mixed pathological changes \[[@B9]\]. This idea is supported by the observation that the progressive loss of hair cells and SGNs leads to ARHL because these two cell types do not regenerate in mammals. 3. Candidate Genes for ARHL Associated with Oxidative Stress and Mitochondrial Dysfunction {#sec3} ========================================================================================== Many genetic investigations of ARHL, such as genome-wide association studies and candidate-gene-based association studies, have been performed recently \[[@B20]\]. With regard to oxidative stress and mitochondrial function, several genes and loci have been proposed as a result of candidate-gene-based association studies, which are based on hypotheses about the relationship between specific known loci and phenotypes. The superoxide dismutases (SODs), which catalyze the dismutation of superoxide into oxygen and hydrogen peroxide, are an important part of the antioxidant defense system against ROS. Recently, evidence from the London ARHL cohort suggested an effect of common superoxide dismutase 2 (SOD2, also known as manganese SOD or mitochondrial SOD) promoter variation, −38 C \> G, on SOD2 promoter regulation and linked it to ARHL risk in men; however, this association was only suggestive due to a lack of replication \[[@B21]\]. The glutathione S-transferases (GSTs) catalyze the detoxification of electrophilic substrates by conjugation with reduced glutathione and participate in intracellular binding and transport of lipophilic substances. Decreased glutathione and GST activity levels cause an increase in susceptibility to cell damage. A previous study investigated the association between ARHL and genes related to oxidative stress using a large set of samples from two population groups, a general European group and a Finnish group \[[@B22]\]. Although an association between the polymorphisms of glutathione S-transferase, mu 1 (*GSTM1*) or glutathione S-transferase, theta 1 (*GSTT1*), and ARHL was not detected in the former population, there were significant associations between both genes and ARHL in the latter population. Mitochondrial uncoupling proteins (UCPs), which are members of the larger family of mitochondrial anion carrier proteins, facilitate the transfer of anions from the inner to the outer mitochondrial membrane and the return transfer of protons from the outer to the inner mitochondrial membrane. UCPs reduce the mitochondrial membrane potential in mammalian cells. The main function of uncoupling protein 2 (UCP2) is the control of mitochondria-derived ROS \[[@B23]\].*UCP2* Ala55Val polymorphisms exhibited a significant association with ARHL in a Japanese population \[[@B24]\]. 4. Deletions and Mutations of mtDNA in the Peripheral Auditory System of ARHL Patients {#sec4} ====================================================================================== Acquired mtDNA defects have been proposed as important factors in aging. Increases in deletions, mutations, or both, in mtDNA have been reported in human temporal bone studies from ARHL patients in comparison with normal-hearing control tissues. A 4977-base pair deletion of mtDNA from celloidin-embedded temporal bone sections was significantly more frequent in cochlear tissue from ARHL patients in comparison to those with normal hearing \[[@B25]\]. Another study reported that quantitative analysis of the mtDNA in archival cochlear tissue samples revealed a mean common deletion level of 32 ± 14% in ARHL patients, in comparison with a level of 12 ± 2% in age-matched controls with normal hearing, and showed a significant correlation between the common deletion level and the severity of hearing loss \[[@B26]\]. Cytochrome c oxidase subunit 3 (COX3) expression was significantly diminished in SGNs from ARHL patients in comparison with age-matched normal-hearing individuals. In addition to the mtDNA common deletion, other deletions involving the mtDNA major arc contributed to the observed deficit in COX3 expression \[[@B27]\]. Mutations within the cytochrome c oxidase subunit 2 (*COX2*) gene in the spiral ganglion and membranous labyrinth from archival temporal bones occur more commonly in ARHL patients relative to controls \[[@B28]\]. 5. Basic Research in Animals on the Role of Oxidative Stresses and Mitochondrial Dysfunction in ARHL {#sec5} ==================================================================================================== Although details of the aging process differ in various organisms, there is a common understanding that oxidative stress and mitochondrial dysfunction play a major part in aging. The auditory system is no exception and it is thought that oxidative damage caused by ROS and mitochondrial dysfunction plays a causal role in ARHL. The fast-aging senescence-accelerated mouse-prone 8 (SAMP8) strain that is a useful model for probing the effects of aging on biological processes displays premature hearing loss associated with strial, sensory, and neural degeneration \[[@B29]\]. The molecular mechanisms associated with premature ARHL in SAMP8 strain mice involve oxidative stress, altered levels of antioxidant enzymes, and decreased activity of complexes I, II, and IV, which lead to triggering of apoptotic cell death pathways. In the organ of Corti of CBA/J mice, glutathione-conjugated proteins, markers of H~2~O~2~-mediated oxidation, were shown to begin to increase at 12 months, and 4-hydroxynonenal and 3-nitrotyrosine, products of hydroxyl radical and peroxynitrite action, respectively, were elevated by 18 months \[[@B30]\]. On the other hand, apoptosis-inducing factor and SOD2 were decreased by 18 months in the organ of Corti and SGNs \[[@B30]\]. Mice lacking superoxide dismutase 1 (*Sod1*) showed premature ARHL \[[@B31], [@B32]\]. Age-related cochlear hair cell loss was observed in*Sod1* knockout mice \[[@B32]\] and a reduced thickness of the stria vascularis and severe degeneration of SGNs were observed at middle age \[[@B31]\]. A previous study showed that increased GPX activity was observed in the stria vascularis and spiral ligament of the cochlea in aged Fischer 344 rats \[[@B33]\]. Two-month-old knockout mice with a targeted inactivating mutation of the gene coding for glutathione peroxidase 1 (*Gpx1*) showed a significant increase in hearing thresholds at high frequency \[[@B34]\]. Mice lacking senescence marker protein 30 (SMP30)/gluconolactonase (GNL), which are not able to synthesize vitamin C, showed a reduction of vitamin C in the inner ear, an increase of hearing thresholds, and loss of spiral ganglion cells, suggesting that depletion of vitamin C accelerates ARHL \[[@B35]\]. Oxidative stress induces the expression of BCL2-antagonist/killer 1 (*Bak*); the mitochondrial proapoptotic gene, in primary cochlear cells and Bak deficiency prevents apoptotic cell death \[[@B36]\]. C57BL/6J mice with a deletion of*Bak* exhibit reduced age-related apoptotic cell death of SGNs and hair cells in the cochlea and prevention of ARHL \[[@B36]\]. A mitochondrially targeted catalase transgene suppresses*Bak* expression in the cochlea, reduces cochlear cell death, and prevents ARHL \[[@B36]\]. Collectively, these findings indicate that age-related increases in ROS levels play an important role in the development of ARHL. It has been shown that accumulation of mtDNA mutations leads to premature aging in mice expressing a proofreading-deficient version of the mtDNA polymerase g (POLG D257A mice), indicating a causal role of mtDNA mutations in mammalian aging \[[@B37], [@B38]\]. POLG D257A mice accumulate mitochondrial mutations more rapidly than wild-type mice. At 9-10 months old, POLG D257A mice showed a variety of premature aging phenotypes, including the early onset of ARHL. Histological findings in the cochlear basal turn confirmed that POLG D257A mice at the age of 9-10 months showed a severe loss of SGNs and hair cells and significant elevation in TUNEL-positive cells and cleaved caspase-3-positive cells in the cochlea \[[@B39]\]. Mitochondrial biogenesis and degradation are involved in mitochondrial turnover. In the SGNs of SAMP8 strain mice, mitochondrial biogenesis, characterized by the ratio of mtDNA/nuclear DNA and the activity of citrate synthase, was increased at younger ages and decreased in old age \[[@B29]\]. Age-related reductions of peroxisome proliferator-activated receptor c coactivator a (PGC-1a), one of the key regulators of mitochondrial biogenesis, might be an important factor for mitochondrial function in age-related diseases \[[@B40]\]. When it comes to mitochondrial function in the cochlea, the overexpression of PGC-1a with a consequent increase of nuclear respiratory factor 1 (NRF1) and mitochondrial transcription factor A (TFAM) caused a significant decrease in the accumulation of damaged mtDNA and the number of apoptotic cells in the strial marginal cells senescence model \[[@B41]\]. Autophagy is one of the major intracellular degradation pathways along with the ubiquitin-proteasome system \[[@B42]\]. Unnecessary cytoplasmic proteins and organelles are enclosed by the autophagosome and then delivered to the lysosome by autophagy. It has been reported that the SGNs of SAMP8 undergo autophagic stress with accumulation of lipofuscin inside these cells \[[@B29]\]. Downregulation of mitophagy, the selective removal of damaged and dysfunctional mitochondria by autophagosomes will cause abnormal mitochondrial morphological changes. Impairment of mitophagy might result in the formation of giant mitochondria, which have been characterized as having low ATP production, a loss of cristae structure, and a swollen morphology \[[@B43]\]. Accumulation of abnormally functioning and shaped mitochondria accelerates apoptosis \[[@B44]\], which merits further investigation in the cochlea. 6. Prevention and Retardation of ARHL by Supplementation or Caloric Restriction {#sec6} =============================================================================== Several studies have reported the effects of supplementation of antioxidants against ARHL. A cross-sectional and 5-year longitudinal study in Australia demonstrated that dietary vitamin A and vitamin E has a significant association with the prevalence of hearing loss, although dietary antioxidant intake did not increase the incidence of hearing loss \[[@B45]\]. Another cross-sectional study in Australia showed that higher carbohydrate, vitamin C, vitamin E, riboflavin, magnesium, and lycopene intakes were significantly associated with larger transiently evoked otoacoustic emission (TEOAE) amplitudes and better pure tone averages (PTAs) whereas higher cholesterol, fat, and retinol intakes were significantly associated with lower TEOAE amplitude and worse PTAs \[[@B46]\]. Another further cross-sectional study in the United States showed that higher intakes of beta-carotene, vitamin C, and magnesium were associated with better PTAs at both speech and high frequencies, and high intakes of beta-carotene or vitamin C combined with high magnesium compared with low intakes of both nutrients were significantly associated with better PTAs at high frequencies \[[@B47]\]. In animal studies, Fischer 344 rats given vitamin C, vitamin E, melatonin, or lazaroid had better auditory sensitivities and a trend for fewer mtDNA deletions in comparison with placebo subjects \[[@B48]\]. Fischer 344 rats of 18--20 months old supplemented orally for 6 months with lecithin, a polyunsaturated phosphatidylcholine (PCP) which has antioxidant effects, showed significantly better hearing sensitivities, higher mitochondrial membrane potentials, and reduced frequency of the common aging mtDNA deletion in the cochlear tissues compared with controls \[[@B49]\]. Aged dogs fed a high antioxidant diet for the last 3 years of their life showed less degeneration of the spiral ganglion cells and stria vascularis in comparison with dogs fed a control-diet \[[@B50]\]. In C57BL/6 mice, supplementation with vitamin C did not increase vitamin C levels in the cochlea or slow ARHL \[[@B35]\], but animals fed with a diet comprising 6 antioxidant agents (L-cysteine-glutathione mixed disulfide, ribose-cysteine, NW-nitro-L-arginine methyl ester, vitamin B12, folate, and ascorbic acid) showed significantly better auditory sensitivity \[[@B51]\]. When C57BL/6 mice were fed with a diet containing one of 17 antioxidant agents (acetyl-L-carnitine, alpha-lipoic acid, beta-carotene, carnosine, coenzyme Q10, curcumin, d-alpha-tocopherol, epigallocatechin gallate, gallic acid, lutein, lycopene, melatonin, N-acetyl-L-cysteine, proanthocyanidin, quercetin, resveratrol, and tannic acid), ARHL was nearly completely prevented by alpha-lipoic acid and coenzyme Q10 and partially by N-acetyl-L-cysteine, but not by other agents \[[@B36]\]. When CBA/J mice were fed with an antioxidant-enriched diet containing vitamin A, vitamin C, vitamin E, L-carnitine, and a-lipoic acid from 10 months through 24 months of age, the antioxidant capacity of the inner ear tissues was significantly increased, but the loss of hair cells and spiral ganglion cells and the magnitude of ARHL were not improved \[[@B52]\]. These studies show that the prevention and retardation of ARHL by supplementation with antioxidants can be influenced by many factors such as the type and dosage of antioxidant compounds, the timing and duration of the treatment, and the species and strains involved. Caloric restriction (CR) extends the lifespan of various organisms including yeast, worms, flies, rodents and non-human primates. It has been reported that CR plays an important role in reducing age-related diseases such as cancer \[[@B53]\], protecting age-related mitochondrial dysfunction \[[@B54]\] and reducing mtDNA damage \[[@B55]\]. It has also been reported that CR can protect neurons against degeneration in animal models of neurodegenerative diseases, as well as promote neurogenesis and enhance synaptic plasticity \[[@B56]\]. The ability of CR to prevent cochlear pathology and ARHL has been extensively studied using laboratory animals \[[@B57]\]. C57BL/6 mice with CR by 15 months of age maintained normal hearing and showed no obvious cochlear degeneration and a significant reduction in the number of TUNEL-positive and cleaved caspase-3-positive cells in the spiral ganglion cells in comparison with controls \[[@B58]\]. Fischer 344 rats with CR to 70% of the control intake beginning at one month of age and then housed for 24-25 months showed significantly better hearing thresholds, reduced hair cell loss, and decreased mtDNA common deletion in the auditory nerve and stria vascularis of the cochlea than control rats \[[@B48]\]. Beneficial effects of CR for the prevention of ARHL has been reported in the AU/Ss, CBA/J strains of mice as well as the C57BL/6 strain, but not in the DBA/2J, WB/ReJ, or BALB/cByJ strains \[[@B57]\]. The effects of CR may depend on genetic background. On the other hand, a high fat diet given to Sprague Dawley rats for 12 months resulted in elevated hearing thresholds in the high-frequency region, increased ROS generation, expression of reduced nicotinamide adenine dinucleotide phosphate (NADPH) oxidase and UCP, accumulation of mtDNA common deletion, and cleaved caspase-3 and TUNEL-positive cells in the inner ear \[[@B59]\]. A microarray analysis study of the cochlea revealed that CR down-regulated the expression of 24 apoptotic genes, including*Bak* and BCL2-like 11 (*Bim*), suggesting that CR could prevent apoptosis of cochlear cells \[[@B58]\]. It has been reported that the mitochondrial deacetylase Sirtuin 3 (*Sirt3*) mediates reduction of oxidative damage and prevention of ARHL under CR \[[@B60]\]. CR failed to reduce oxidative DNA damage or prevent ARHL in C57B/6 mice lacking*Sirt3* \[[@B60]\]. In response to CR,*Sirt3* directly deacetylated and activated mitochondrial isocitrate dehydrogenase 2 (*Idh2*), leading to increased NADPH levels and an increased ratio of reduced-to-oxidized glutathione in mitochondria \[[@B60]\]. In cultured human kidney cells (HEK293), overexpression of Sirt3 and/or Idh2 increased NADPH levels and gave protection from oxidative stress-induced cell death \[[@B60]\]. 7. Putative Role of Oxidative Stress and Mitochondrial Dysfunction in ARHL {#sec7} ========================================================================== The important role of oxidative stress and mitochondrial dysfunction in the development of ARHL has been established by reviewing previous studies. The severity of hearing loss is probably associated with cochlear degeneration. Accumulation of mtDNA damage, ROS production, and decreased antioxidant function are primarily involved in the process of cochlear senescence in response to aging stress. Mitochondria play a crucial role in the induction of intrinsic apoptosis in cochlear cells. ARHL in laboratory animals can be prevented by certain interventions, such as CR and supplementation with antioxidants. Further large clinical studies are needed to confirm whether ARHL can be prevented by the above-mentioned interventions in humans. This work was supported by grants from the Ministry of Education, Culture, Sports, Science & Technology in Japan to Chisato Fujimoto and Tatsuya Yamasoba and from the Ministry of Education, Culture, Sports, Science, and Technology to Tatsuya Yamasoba. Conflict of Interests ===================== The authors declare that there is no conflict of interests regarding the publication of this paper. [^1]: Academic Editor: Robb E. Moses

1. Introduction {#sec1-viruses-11-00460} =============== 1.1. Enteroviruses {#sec1dot1-viruses-11-00460} ------------------ According to the Centers for Disease Control and Prevention, enteroviruses cause at least 10--15 million symptomatic infections in the United States each year \[[@B1-viruses-11-00460]\]. These viruses belong to the *Picornaviridae* family and are small, non-enveloped viruses that have a single stranded positive sense RNA genome. The enterovirus genus includes poliovirus (PV), coxsackieviruses, echoviruses, enterovirus 71 (EV71), enterovirus D68 (EV-D68), and rhinoviruses. These viruses are spread primarily through the fecal--oral route, but some species can be spread through respiratory secretions (e.g., EV-D68 and rhinovirus). Nonpolio enteroviruses are typically asymptomatic or cause minor clinical symptoms which include hand-foot-and-mouth disease and respiratory illness. In some cases, enteroviruses can cause severe complications which include acute flaccid myelitis, myocarditis and encephalitis, pancreatitis, hepatitis, and even death \[[@B2-viruses-11-00460],[@B3-viruses-11-00460],[@B4-viruses-11-00460],[@B5-viruses-11-00460]\]. The pediatric and neonatal populations can develop severe symptoms and grave clinical outcomes of enterovirus infections \[[@B6-viruses-11-00460],[@B7-viruses-11-00460],[@B8-viruses-11-00460]\]. In fact, enteroviruses are one of the top viral pathogens that cause outbreaks in neonatal intensive care units (NICUs) across the United States each year \[[@B9-viruses-11-00460],[@B10-viruses-11-00460]\], and infections in infants and neonates are associated with high morbidity and mortality. Additionally, enterovirus infections particularly impact young children during outbreaks, as seen in the EV71 outbreak in China from 2008 to 2012. This outbreak was responsible for over 7 million infections with a majority of infections in children under the age of five \[[@B11-viruses-11-00460]\]. In addition to EV71 outbreaks, EV-D68 outbreaks have been intensifying throughout the world with outbreaks in 2014, 2016, and 2018 \[[@B12-viruses-11-00460]\]. EV-D68 outbreaks also typically impact neonates and children and have caused many cases of acute flaccid myelitis (AFM) in the United States, including 120 confirmed cases across 34 states in 2014 \[[@B13-viruses-11-00460],[@B14-viruses-11-00460]\]. Although these severe outcomes are the focus of many studies, most individuals infected with enteroviruses are asymptomatic. Overall, enteroviruses are a significant public health concern, particularly in the pediatric population, due to severe complications from infection in children and neonates. The immune response to enteroviruses is imperative for successful host clearance. A sufficient immune response to clear enterovirus infection includes the activation of innate immune signaling and a strong B cell response. The antibody response can be extremely important to clear an enterovirus infection. Previous studies have shown that about 50% of adults and older children have neutralizing antibodies against at least two non-polio enteroviruses and over 75% of adults and children have neutralizing antibodies to PV serotypes \[[@B15-viruses-11-00460],[@B16-viruses-11-00460]\]. Neutralizing antibodies to PV arise from vaccination-induced long-lived memory B cells and neutralizing antibodies that are protective against infection \[[@B17-viruses-11-00460],[@B18-viruses-11-00460]\]. This suggests that neutralizing antibodies are important for protection from re-exposure and may explain why children and neonates are most likely to experience severe infection since they likely lack these antibodies \[[@B15-viruses-11-00460]\]. Consistent with this, individuals with X-linked agammaglobulinemia, where the patient has little to no B cells, are highly susceptible to enterovirus infection \[[@B19-viruses-11-00460],[@B20-viruses-11-00460]\]. In addition, mice deficient in B cells have high coxsackievirus B (CVB) titers in their tissues and experience chronic infection and an inability to clear the virus \[[@B21-viruses-11-00460]\]. Thus, the concerted actions of the innate and adaptive immune response allow for the clearance of enteroviruses. 1.2. The Gastrointestinal Tract {#sec1dot2-viruses-11-00460} ------------------------------- The gastrointestinal (GI) tract is a key defensive barrier against pathogenic bacteria and viruses. The GI tract is divided into different subsections: The duodenum, jejunum, and ileum, which make up the small intestine, large intestine, and colon. The GI tract is composed of an epithelial layer that forms a physical cellular barrier as well as a lamina propria that contains immune cells \[[@B22-viruses-11-00460]\]. The lamina propria is essential to elicit an adaptive immune response to pathogens that breach the epithelium. This region contains dendritic cells and macrophages that are able to present viral antigens as well as many other immune cells that are important for initiating a cellular immune response ([Figure 1](#viruses-11-00460-f001){ref-type="fig"}A). In addition to these two compartments, specialized subsections of the epithelium and the lamina propria contain Peyer's patches. Peyer's patches contain organized lymphoid structures that sample the intestinal lumen to initiate mucosal immune responses. The formation and role of Peyer's patches in mucosal immunity have been extensively reviewed elsewhere \[[@B23-viruses-11-00460],[@B24-viruses-11-00460],[@B25-viruses-11-00460]\]. The GI tract, like many barrier surfaces, has important defense mechanisms to prevent microbial invasion. The cells that comprise the epithelium are polarized, meaning they have distinct apical and basolateral surfaces that contain distinct lipid and protein components. The apical surface of enterocytes, which make up a high proportion of the epithelium, contain microvilli that create a dense brush border. The GI epithelium forms a physical barrier due to junctional complexes composed of tight and adherens junctions as well as desmosomes ([Figure 1](#viruses-11-00460-f001){ref-type="fig"}B) \[[@B26-viruses-11-00460]\]. These junctional complexes are important for restricting the free flow of ions and solutes \[[@B27-viruses-11-00460]\]. In addition, differentiated enterocytes have a dense cortical actin network that is critical for preventing pathogens from gaining access to the subcellular domain \[[@B28-viruses-11-00460]\]. Finally, the epithelium utilizes chemical defenses and secretes type I and III interferons (IFNs) to trigger an antiviral state during viral infections \[[@B29-viruses-11-00460],[@B30-viruses-11-00460],[@B31-viruses-11-00460]\]. The concerted actions of the epithelium and the cells that comprise the lamina propria are essential in the defense against enteric pathogens. The GI epithelium has villus and crypt structures that influence the morphology of the intestine ([Figure 1](#viruses-11-00460-f001){ref-type="fig"}A). The villi project into the intestinal lumen and are mainly composed of enterocytes with other cell types scattered throughout the villi. The base of the crypts contain stem cells that are responsible for the renewal of all the cell types of the GI epithelium \[[@B32-viruses-11-00460],[@B33-viruses-11-00460]\]. The LGR5+ stem cells at the base of the crypt are lined by Paneth cells, which are critical for maintaining stem cell homeostasis (growth factor production) and the secretion of antimicrobial peptides \[[@B34-viruses-11-00460],[@B35-viruses-11-00460]\]. Other cell types are also critical for stem cell differentiation, including crypt-specific fibroblasts \[[@B36-viruses-11-00460]\]. In addition to stem cells and Paneth cells, the epithelium is composed of at least six distinct cell types that help execute the barrier's essential functions. The cell types can be broken into two main subgroups: those of absorptive and secretory lineages. The absorptive lineage is comprised mainly of enterocytes and microfold (M) cells. M cells are grouped into the absorptive lineage due to their role as conduits between luminal contents and immune cells in the lamina propria and Peyer's patches \[[@B37-viruses-11-00460]\]. The secretory lineage includes enteroendocrine cells, Paneth cells, and goblet cells \[[@B38-viruses-11-00460]\]. As the name suggests, cells comprising the secretory lineage mainly secrete proteins into the lumen of the GI tract. Goblet cells produce and secrete mucus, which covers the epithelium and has a protective function against pathogens \[[@B39-viruses-11-00460]\]. On the other hand, enteroendocrine cells produce hormones that are thought to provide signals to stem cells. Each cell type is responsible for separate functions to maintain homeostasis of the GI epithelium. Without each cell type, the delicate balancing act of protecting against pathogens, maintaining correct equilibrium with the microbiome, and the absorption of nutrients would be disrupted. 2. Enterovirus Infections in the GI Tract {#sec2-viruses-11-00460} ========================================= Enteroviruses are primarily transmitted through the fecal--oral route and target the GI epithelium. Enteroviruses are typically ingested through contact with contaminated surfaces, food, and/or water. These viruses are not thought to cause gastrointestinal illness such as severe vomiting or diarrhea, but GI-associated complications may occur \[[@B40-viruses-11-00460]\]. However, most cases of enterovirus infection are asymptomatic \[[@B41-viruses-11-00460]\]. Once they infect the GI epithelium, enteroviruses can disseminate into secondary target tissues and can cause clinical disease in some cases. Enteroviruses have specific secondary tissue tropism that vary between enterovirus species. EV71 has been shown to disseminate into the skin and brain causing hand-foot-and-mouth disease and aseptic meningitis or acute flaccid myelitis, respectively \[[@B42-viruses-11-00460]\]. On the other hand, coxsackievirus B (CVB) can disseminate to the heart and pancreas to cause myocarditis and pancreatitis \[[@B5-viruses-11-00460],[@B43-viruses-11-00460],[@B44-viruses-11-00460]\]. Additionally, echoviruses target the liver as well as the brain causing acute liver failure and aseptic meningitis \[[@B4-viruses-11-00460],[@B45-viruses-11-00460]\]. Despite these differences in secondary target tissues, most enteroviruses, with the exception of rhinoviruses and EV-D68, replicate in the GI epithelium. Enteroviruses target the epithelium for replication and; therefore, this barrier surface is an important defense mechanism for preventing the dissemination of these viruses into secondary target tissues. Enteroviruses initiate entry into a host cell by binding to cell surface receptors and undergoing receptor-mediated endocytosis. Entry receptors vary between enteroviruses and include scavenger receptor B2 (SCARB2) and P-selectin glycoprotein ligand 1 (PSGL-1) for EV71 \[[@B46-viruses-11-00460],[@B47-viruses-11-00460],[@B48-viruses-11-00460]\], the coxsackievirus and adenovirus receptor (CAR) for CVB \[[@B49-viruses-11-00460],[@B50-viruses-11-00460],[@B51-viruses-11-00460]\], the poliovirus receptor (PVR/CD155) for PV \[[@B52-viruses-11-00460]\], and the neonatal Fc receptor (FcRn) for echoviruses \[[@B53-viruses-11-00460],[@B54-viruses-11-00460]\], amongst others ([Table 1](#viruses-11-00460-t001){ref-type="table"}). In some cases, enteroviruses bind to additional attachment factors, the most common of which is decay accelerating factor (DAF)/CD55 \[[@B55-viruses-11-00460],[@B56-viruses-11-00460]\]. Despite differences in cellular receptors, enteroviruses have generally well-conserved life cycles ([Figure 2](#viruses-11-00460-f002){ref-type="fig"}). In intestinal epithelial cell lines, CVB binding to DAF has been proposed to facilitate the induction of cell signaling from the apical domain, which in turn facilitates delivery of viral particles to their primary receptors \[[@B57-viruses-11-00460]\]. Similarly, echovirus binding to DAF has also been proposed to facilitate entry into intestinal epithelial cell lines, although the role of intracellular cell signaling and the primary echovirus receptor in this process remains unclear \[[@B58-viruses-11-00460]\]. After binding and entry, enteroviruses undergo uncoating in order to release the viral genome. Uncoating occurs either after the virus binds to the cell receptor or is initiated through a pH change in the endosome. This uncoating process allows the RNA genome to be released from the protective capsid into the cytoplasm or endosome. Several studies have investigated the speed at which PV virions uncoat in nonpolarized cells using either fluorescently-labeled capsids and viral RNA (vRNA) or neutral red-incorporated vRNA. These studies reveled that vRNA is released from the capsid within 30 minutes of entry \[[@B64-viruses-11-00460],[@B65-viruses-11-00460]\]. However, other studies using polarized cells of the blood--brain barrier with apical and basolateral domains suggest that uncoating is a slower process that requires actin cytoskeleton remodeling \[[@B66-viruses-11-00460]\]. Additionally, the speed of uncoating may differ between enteroviruses based on the requirements of attachment factors, such as DAF, or other cellular proteins required for entry \[[@B57-viruses-11-00460]\]. Once viral RNA has entered the cytoplasm it is translated by host ribosomes ([Figure 2](#viruses-11-00460-f002){ref-type="fig"}). Historically, it was thought that the viral RNA was translated into a single polyprotein. However, a recent study discovered a second open reading frame (ORF) in some enterovirus genomes \[[@B67-viruses-11-00460]\]. This study found a small ORF that is located in the 5' end of the untranslated region and is suggested to be important in replication in intestinal epithelial cells \[[@B67-viruses-11-00460]\]. Regardless of whether it is a single polyprotein or two proteins, the resulting product is then proteolytically cleaved by the viral proteases 2A and 3C. The resulting proteins include 10 proteins such as capsid proteins and other replication proteins including the RNA-dependent RNA polymerase 3D (3D^pol^). 3D^pol^ initiates the synthesis of the negative-stranded copy of the genome making a dsRNA intermediate, which becomes the template to generate new positive-stranded genomes. Replication of viral RNA occurs in replication organelles that derive from host membranes that are induced upon viral infection \[[@B68-viruses-11-00460]\]. These replication organelles can protect the RNA and replication intermediates from cytosolic localized innate immune pattern recognition receptors (PRRs) that are important for the detection of foreign RNA \[[@B68-viruses-11-00460]\]. The newly synthesized positive-stranded genome is packaged into virions for release out of the cell. The virions are assembled into protomers and pentamers using the capsid proteins VP0, VP1 and VP3. After the RNA is packaged into the virion, VP0 is processed into VP2 and VP4, which results in mature enterovirus virions \[[@B41-viruses-11-00460],[@B69-viruses-11-00460]\]. Classically, enteroviruses have been thought to exit the cell through a lytic form of cell death where the cell undergoes lysis and releases the progeny virions to infect neighboring cells \[[@B41-viruses-11-00460]\] ([Figure 2](#viruses-11-00460-f002){ref-type="fig"}). Recently, new studies have suggested that enteroviruses can also exit the host cell through non-lytic pathways \[[@B70-viruses-11-00460],[@B71-viruses-11-00460],[@B72-viruses-11-00460]\]. These studies showed that during infection of PV and a related Picornavirus, hepatitis A virus, progeny virions are able to acquire host cell membranes to exit the cell in a vesicle in order to infect new cells \[[@B70-viruses-11-00460],[@B71-viruses-11-00460]\]. These studies have shifted how enterovirus release is considered, but more work is needed to establish whether all enterovirus species are able to undergo non-lytic release and whether GI-derived cells permit this form of release. 3. Detection of Enteroviruses by Pattern Recognition Receptors {#sec3-viruses-11-00460} ============================================================== Detection of pathogens by the host immune system is an important first step in the clearance of viral pathogens. Previous studies have shown that echovirus 11 (E11), EV71, and CVB induce a robust innate immune response in intestinal epithelial cell lines and primary cells \[[@B73-viruses-11-00460],[@B74-viruses-11-00460],[@B75-viruses-11-00460]\]. Viruses can be detected by the innate immune system in a variety of ways. PRRs are imperative in the detection and response to viral pathogens and are germ line encoded. PRRs detect pathogen associated molecular patterns (PAMPs) and respond by inducing an antiviral state. 3.1. Detection by TLRs {#sec3dot1-viruses-11-00460} ---------------------- One class of PRRs is Toll-like receptors (TLRs). TLRs are a class of 10 transmembrane PRRs that recognize a variety of PAMPs. Within the TLR family, two additional categories exist, which are TLRs that are localized to the cell surface as well as TLRs that are localized to the endosome \[[@B76-viruses-11-00460]\]. Generally, TLRs have a PAMP binding domain on the N terminal region of the protein that is either on the extracellular domain or in the endosomal lumen and an intracellular signaling region on the C terminal end \[[@B77-viruses-11-00460]\]. Here we will only discuss TLRs that detect RNA viruses but note other TLRs exist to sense DNA virus and bacterial derived PAMPs, such as TLR5 (flagellin), which have been extensively reviewed elsewhere \[[@B77-viruses-11-00460],[@B78-viruses-11-00460],[@B79-viruses-11-00460]\]. TLR3 is primarily expressed on the endosome and recognizes dsRNA \[[@B80-viruses-11-00460]\]. TLR3 is expressed under basal conditions in most cells and is not typically induced by interferon (IFN) or enterovirus infection \[[@B81-viruses-11-00460]\]. TLR3 mediates an IFN response through Toll/IL-1 receptor domain-containing adaptor inducing interferon-beta (TRIF) and interferon regulatory factor (IRF)-3. The antiviral response mediated through TLR3 has been shown to be important in controlling PV, coxsackievirus A16 (CVA16), and coxsackievirus B3 (CVB3) \[[@B82-viruses-11-00460],[@B83-viruses-11-00460],[@B84-viruses-11-00460]\]. In fact, the interferon stimulated genes (ISGs) that are produced by the induction of type I IFNs downstream of TLR3 are directly antiviral against CVA16. When expression of TLR3 is knocked out in mice, these animals have a more severe CVA16 infection compared to wild type control animals and develop severe paralysis and death \[[@B83-viruses-11-00460]\]. Others have shown that TLR3 is also imperative for antiviral signaling during PV infection in mice \[[@B85-viruses-11-00460]\]. Additionally, in vitro studies have shown that when cells are depleted of TRIF, a downstream adaptor molecule of TLR3, EV71 replication increases \[[@B86-viruses-11-00460]\]. Although a rare polymorphism in TLR3 was identified in a patient who developed CVB-associated myocarditis, genetic variants in TLR3 or other IFN-associated factors are not commonly found in patients with viral-associated myocarditis \[[@B87-viruses-11-00460],[@B88-viruses-11-00460]\]. Instead, these patients often express variants in genes associated with inherited cardiomyopathies, suggesting that TLR3 signaling is not the sole determinant of CVB-induced myocarditis. Nonetheless, in vitro and in vivo studies provide strong evidence that TLR3 is important for the detection and antiviral control of many enterovirus species. Several studies provide support that TLR3 is an essential TLR in enterovirus infection. However, other TLRs can also play significant roles. Although TLR4 is thought to be key during bacterial infection since it mainly senses lipopolysaccharide (LPS), a protein found on gram negative bacteria, TLR4 plays an important role in secondary target tissues (tissues other than the route of entry) of enterovirus infection. TLR4 is localized to the cell surface where it can detect extracellular bacterial pathogens and has been shown to be important in myocarditis associated with CVB3 infection \[[@B89-viruses-11-00460],[@B90-viruses-11-00460]\]. Studies have shown the TLR4 activation induces proinflammatory cytokines, which is seen in dilated cardiomyopathy, and a positive correlation between TLR4 and enterovirus RNA in endomyocardial biopsy tissues \[[@B90-viruses-11-00460]\]. Furthermore, coxsackievirus B4 (CVB4) has been shown to induce proinflammatory cytokines through TLR4 in the pancreas, which leads to the progression of type I diabetes \[[@B91-viruses-11-00460]\]. It is still unclear how TLR4 detects enterovirus infection; however, evidence points to a role of TLR4 in the induction of proinflammatory cytokines and clinical pathology during infection. Both TLR7 and TLR8 sense ssRNA and are localized to the endosome \[[@B92-viruses-11-00460]\]. Typically, these TLRs are not thought to be ISGs and; therefore, their expression is independent of IFN induction. However, a number of studies have demonstrated that TLR7 and TLR8 can be induced upon enteroviral infection. CVB3 can induce TLR7 and TLR8 expression after 48 h of infection at a low multiplicity of infection (MOI) \[[@B81-viruses-11-00460]\]. Additionally, EV71 induces expression of TLR8 in cell lines and expression of TLR7 and TLR8 are increased in lung and brain tissues from children who died from EV71 infection \[[@B73-viruses-11-00460],[@B93-viruses-11-00460]\]. Although the role of TLR7 and TLR8 have not been extensively studied during enterovirus infection, it is becoming clear that these PRRs may play key roles in the induction of proinflammatory cytokines. In fact, CVB is known to cause myocarditis due to chronic inflammation of the myocardium. This release of inflammatory cytokines has been linked to TLR8 and TLR4 \[[@B94-viruses-11-00460],[@B95-viruses-11-00460]\]. These studies suggest that TLR7 and TLR8 play a significant role in enterovirus infection. 3.2. Detection by RLRs {#sec3dot2-viruses-11-00460} ---------------------- TLRs play a key role in the detection of extracellular and endosomal localized pathogens, but RIG-I like receptors (RLRs) are arguably the crucial sensors for the detection of enteroviruses due to their localization to the cytoplasm. RLRs that are able to detect RNA virus infection are Retinoic acid-inducible gene I (RIG-I) and Melanoma differentiation-associated antigen 5 (MDA5). Both MDA5 and RIG-I have two caspase recruitment domains (CARD-like domains) at the N terminus as well as a DExD box RNA helicase, which is important for the detection of viral PAMPs \[[@B96-viruses-11-00460]\]. RIG-I is a cytosolic PRR that recognizes RNA ligands such as 5' triphosphate RNA (5' pppRNA) \[[@B97-viruses-11-00460]\]. In vitro data suggests that RIG-I is not always activated by enterovirus infection due to the VPg protein binding to the free 5' triphosphate RNA, which would normally activate RIG-I \[[@B41-viruses-11-00460]\]. However, recent studies have suggested a role of RIG-I in CVB3 infection \[[@B98-viruses-11-00460],[@B99-viruses-11-00460]\]. Feng et al, suggests that the 5' clover leaf of CVB3 is able to activate RIG-I since it contains triphosphate containing RNA \[[@B99-viruses-11-00460]\]. However, this may be specific to CVB3 in cell line models since mice that are deficient in RIG-I have no difference in susceptibility to enterovirus infection compared to WT controls \[[@B82-viruses-11-00460]\]. MDA5 detects long cytoplasmic dsRNA \[[@B100-viruses-11-00460],[@B101-viruses-11-00460],[@B102-viruses-11-00460]\]. Several studies have indicated that MDA5 specifically interacts with enterovirus dsRNA, a replication intermediate, during CVA, CVB, EV71, and other enteroviruses \[[@B103-viruses-11-00460],[@B104-viruses-11-00460],[@B105-viruses-11-00460],[@B106-viruses-11-00460]\]. Moreover, a polymorphism in MDA5 has been suggested to be a risk factor for more severe EV71 infection \[[@B107-viruses-11-00460]\]. Children with this polymorphism exhibited more severe symptoms during EV71 infection compared to children without the polymorphism, suggesting a role of MDA5 in the detection of enterovirus infection. Furthermore, mice that are deficient in MDA5 are more susceptible to enterovirus infection and succumb to disease much more rapidly \[[@B105-viruses-11-00460],[@B108-viruses-11-00460]\]. In addition to this increase in susceptibility, MDA5 deficient animals infected with CVB3 display severe hepatic necrosis of the liver \[[@B105-viruses-11-00460]\]. Collectively, these studies point to the essential role of MDA5 in the detection of enteroviruses. The adaptor protein for both RIG-I and MDA5 is the mitochondrial antiviral-signaling protein (MAVS), which is localized to the mitochondria and peroxisomes. When RIG-I or MDA5 are activated by dsRNA, the CARD domains become ubiquitinated \[[@B109-viruses-11-00460],[@B110-viruses-11-00460]\], leading to the formation of MAVS aggregates in the mitochondrial membrane \[[@B111-viruses-11-00460]\]. Aggregation of MAVS leads to the activation of NF-κB and IRF3, which then induce IFN \[[@B112-viruses-11-00460]\]. In vitro studies have concluded that overexpression of MAVS can inhibit CVB3 replication by increasing the amount of IFN induction \[[@B113-viruses-11-00460]\]. Although some studies have showed that MAVS deficient mice do not have an increased CVB3 viral load compared to WT controls, these animals succumb to infection much earlier than WT animals, suggesting that MAVS signaling and MDA5 dependent activation of IFN is imperative to host response to infection \[[@B105-viruses-11-00460]\]. Overall, these PRRs and adaptor molecules have been shown to be imperative for sensing enterovirus infections. 4. Evasion of Innate Immunity by Enteroviruses {#sec4-viruses-11-00460} ============================================== Viruses have evolved mechanisms to evade the induction of the antiviral state of the cell. Viral protease-mediated cleavage of PRRs allow enteroviruses to impact downstream signaling cascades, resulting in loss or reduced induction of IFN or interferon stimulated genes (ISGs). [Table 2](#viruses-11-00460-t002){ref-type="table"} summarizes these specific events and will be detailed below. As a result of these cleavage events, viruses are able to replicate more efficiently in the cell. Each species of enterovirus has developed its own set of mechanisms of evasion. Here, we will discuss current knowledge of evasion mechanisms of enteroviruses and how they antagonize host innate immune signaling. 4.1. Evasion of TLRs by Enteroviruses {#sec4dot1-viruses-11-00460} ------------------------------------- Enteroviruses are very efficient at disrupting downstream innate immune signaling. We have previously shown that, in human embryonic kidney cells (HEK293), CVB3 3C protease (3C^pro^) cleaves TRIF, a downstream adaptor molecule of TLR3 \[[@B114-viruses-11-00460]\]. Cleavage of TRIF results in loss of TLR3-dependent induction of IFN and NF-κB. Other groups have shown that EV-D68 and EV71 3C^pro^ can also cleave TRIF resulting in decreased signaling downstream of TLR3 \[[@B115-viruses-11-00460],[@B116-viruses-11-00460]\]. These studies, which include many different enterovirus species, show that TLR3-dependent IFN induction is antiviral against enteroviruses and is a key evasion target for these viruses ([Figure 3](#viruses-11-00460-f003){ref-type="fig"}). In addition to TLR3, TLR7 has been shown to be targeted by some enterovirus species, but the mechanisms that they use to target it are not well understood. As discussed previously, some enteroviruses, such as CVB3, can induce expression of TLR7 during infection. However, other enteroviruses seem to target TLR7. The detection of vRNA by TLR7 has been shown to increase autophagic flux \[[@B124-viruses-11-00460]\]. In fact, one study showed that, in human bronchial epithelial cells (16HBE), TLR7-dependent type I IFN induction is reduced by EV71 and CVA16 \[[@B125-viruses-11-00460]\]. This study concluded that autophagy induced by these viruses reduces endosome formation, resulting in the decreased expression of TLR7 to evade TLR7-dependent induction of autophagic flux in this cell type. This finding potentially demonstrates that some enterovirus species evade detection by TLR7, but others benefit from the induction of proinflammatory cytokines by TLR7. 4.2. Evasion of RLRs by Enteroviruses {#sec4dot2-viruses-11-00460} ------------------------------------- In addition to TLRs, enteroviruses also target members of the RLR family for cleavage to evade innate immune signaling. MDA5, which is important for the sensing of enteroviruses in the host cell, is a target of viral proteases in many different studies. CVB3 2A^pro^ has been shown to cleave MDA5 in HeLa cells \[[@B121-viruses-11-00460]\]. However, this study does not determine whether the cleavage products are still able to induce IFN signaling or whether cleavage of MDA5 hinders MDA5-depedent IFN induction. Similar studies using CVA16, CVA6, and EV-D68 have indicated that 3C^pro^ cleaves MDA5 \[[@B120-viruses-11-00460]\]. Although the authors show that IFN signaling is disrupted when cells were transfected with 3C^pro^, they do not specifically show that the cleavage products are not functional in inducing an IFN response. Additionally, MDA5 has been shown to be cleaved in PV-infected HeLa cells \[[@B122-viruses-11-00460]\]. However, unlike the prior studies, this study concluded that the cleavage was not dependent on viral proteases but was instead mediated by cellular caspases activated during infection \[[@B122-viruses-11-00460]\]. Furthermore, EV71 is able to cleave MDA5, but the mechanism is less clear \[[@B104-viruses-11-00460]\]. Apart from the different mechanisms enteroviruses use to disrupt MDA5 signaling, infected cell lines have been shown to have cleavage products resulting in the inhibition of IFN induction. RIG-I has also been shown to be cleaved in cells infected with different enteroviruses. Since RIG-I mainly detects 5' pppRNA, the reason why enteroviruses would target this RLR is not well understood but, as discussed before, new evidence suggests that RIG-I may detect enteroviruses (see [Section 3.2](#sec3dot2-viruses-11-00460){ref-type="sec"}). PV 3C^pro^ is able to cleave RIG-I in infected HeLa cells by 6 h post infection \[[@B119-viruses-11-00460]\]. In addition to viral protease-mediated cleavage of these sensors, EV71 alters IFN induction by targeting the ubiquitination of RIG-I \[[@B117-viruses-11-00460]\], which is critical for downstream signaling \[[@B117-viruses-11-00460],[@B118-viruses-11-00460]\]. Previous studies have shown that CYLD (cylindromatosis), a deubiquitinating enzyme, is a negative regulator of RIG-I \[[@B126-viruses-11-00460]\]. During viral infection, a cellular microRNA, miR-526a, is upregulated and induces the downregulation of negative regulator, CYLD, leading to enhanced signaling of RIG-I. However, EV71 is able to downregulate miR-526a resulting in normal levels of CYLD \[[@B127-viruses-11-00460]\]. As a result, RIG-I ubiquitination decreases inhibiting IFN induction. Further research to delineate the specific enterovirus PAMP that RIG-I is able to detect to induce IFN and the mechanisms enteroviruses use to target RIG-I will be needed to understand this aspect of enterovirus infection. Numerous studies have investigated the viral protease mediated cleavage of MAVS. Targeting MAVS, the adaptor protein of RIG-I and MDA5, ablates IFN induction of both RIG-I and MDA5, making this protein an essential target of many enteroviruses. CVB3 2A^pro^ and 3C^pro^ cleave MAVS in various cell lines \[[@B114-viruses-11-00460],[@B121-viruses-11-00460]\]. The resulting cleavage products are nonfunctional and are deficient in NF-κB and IFN signaling \[[@B114-viruses-11-00460]\]. EV71 2A^pro^ is able to cleave MAVS in HeLa cells \[[@B121-viruses-11-00460],[@B123-viruses-11-00460]\]. These studies showed that, similar to CVB infection, the products of MAVS in EV71-infected cells are deficient in NF-κB and IFN signaling \[[@B121-viruses-11-00460],[@B123-viruses-11-00460]\]. However, since these studies are mainly performed in cell lines, further research is needed to determine if enteroviruses behave similarly in primary cells such as those of the GI tract. 5. Models to Study Enteroviruses in the Gut {#sec5-viruses-11-00460} =========================================== 5.1. In Vitro and Ex Vivo Models to Study Enterovirus Infection {#sec5dot1-viruses-11-00460} --------------------------------------------------------------- Many different models to study enteroviruses in the GI tract exist. These include cell lines, three-dimensional cell culture-based models, mouse models, and non-human primate models. Cell lines that model the GI tract and have been applied to enterovirus research include Caco-2, HT-29, T84, MODE-K (murine), and IEC-6 (rat) cells. The main cell line that has historically been used to model enteroviral infections of the human intestinal epithelium is Caco-2 cells. Caco-2 cells have characteristics of enterocytes, which includes a brush border and tight junctions \[[@B128-viruses-11-00460]\]. In addition to standard culture systems that utilize Caco-2 cells, we have also developed a three-dimensional culture model using Caco-2 cells grown on beads in a rotating wall vessel bioreactor that exhibit the properties of the intestinal epithelium, and have applied this system to model enterovirus infections in the GI tract \[[@B129-viruses-11-00460]\]. Other 3D culture model systems include organoids. Organoids are 3D enterospheres that are derived from pluripotent stem cells or embryonic stem cells \[[@B130-viruses-11-00460]\]. Organoids are spherical structures that are hollow in the middle, have apical and basolateral polarity, and form a spherical layer of epithelium \[[@B131-viruses-11-00460],[@B132-viruses-11-00460]\]. Pluripotent stem cells are differentiated into ectoderm, then hindgut ectoderm, and finally form spheroids with the addition of correct growth factors \[[@B130-viruses-11-00460],[@B133-viruses-11-00460]\]. In addition, organoids contain a mesenchymal cell layer that develops under the organoid. Studies have shown that organoids are able to differentiate into the absorptive and secretory lineages of the GI epithelium \[[@B130-viruses-11-00460]\]. However, this culture model has not yet been applied to enterovirus research. In addition to organoids, enteroids are used as another 3D system to model the GI epithelium. Enteroids are formed through the isolation of intestinal crypts from whole human and murine intestinal tissues \[[@B134-viruses-11-00460]\], which contain LGR5+ stem cells \[[@B33-viruses-11-00460]\]. The crypts can be isolated and plated in Matrigel, where they form 3D spherical enteroids, or on transwells, where they form 2D monolayers that exhibit barrier function ([Figure 4](#viruses-11-00460-f004){ref-type="fig"}) \[[@B74-viruses-11-00460],[@B75-viruses-11-00460],[@B135-viruses-11-00460]\]. Enteroids that are plated in matrigel have an "inside-out" phenotype, where the apical surface is facing into the lumen and the basolateral surface is on the outside of the structure \[[@B75-viruses-11-00460]\]. This makes the Matrigel model a difficult model for studying viruses that use receptors that are localized to the apical surface, since the apical surface is not accessible without disruption of the 3D nature of these structures. To overcome this limitation, we and others have developed a transwell-based model system in order to gain access to both the apical and basolateral surfaces \[[@B75-viruses-11-00460],[@B136-viruses-11-00460],[@B137-viruses-11-00460]\]. These transwell models allow for infection at either the apical or basolateral surfaces and for collection of growth medium from these distinct compartments. We have applied both Matrigel- and transwell-based enteroid models to study enterovirus--GI interactions ([Figure 4](#viruses-11-00460-f004){ref-type="fig"}). Using these systems, we have identified differences in the cell-type specificity by which enterovirus target the GI epithelium. For example, whereas E11 preferentially infects enterocytes and enteroendocrine cells, EV71 replication is largely restricted to goblet cells \[[@B74-viruses-11-00460],[@B75-viruses-11-00460]\]. In addition, using a transwell-based model, we have shown that enteroviruses also exhibit differences in the polarity by which they enter into and egress from the epithelium, with E11 exhibiting a basolateral polarity of entry and a bidirectional manner of egress whereas EV71 both enter and is released preferentially from the apical domain \[[@B75-viruses-11-00460]\]. Perhaps most striking in these models is the robust antiviral response elicited in response to enterovirus infections. In contrast to most cell lines, which induce little to no IFN signaling, primary human enteroid models potently induce an antiviral response to enteroviral infections \[[@B75-viruses-11-00460]\]. Perhaps not surprisingly given their role in barrier defenses, these models almost exclusively induce antiviral type III IFNs in response to infection. Collectively, these data highlight the potential relevance of primary-based intestinal cell systems to model enterovirus infections. Although these in vitro models recapitulate the multicellular complexity of the small intestine, making them a more physiologically relevant model compared to cell lines, they do; however, lack bacterial interactions, which can impact enteroviral infection. In vitro studies have shown that PV virions can bind to bacteria and that some bacterial strains can facilitate enterovirus infection \[[@B138-viruses-11-00460]\]. In fact, bacteria can aid in co-infection of different enteroviruses, which allows for genetic recombination \[[@B138-viruses-11-00460]\]. Other studies have shown that certain species of bacteria can increase thermal stabilization of PV and CVB \[[@B139-viruses-11-00460]\]. This leads to the question of whether other enteroviruses are also impacted by bacterial co-infection and what impact this has on pathogenesis in vivo. Studies investigating the role of the microbiome on enteroviruses are imperative to understand in vivo pathogenesis. 5.2. In Vivo Models to Study Enterovirus Infection {#sec5dot2-viruses-11-00460} -------------------------------------------------- While the above-described in vitro models have provided many insights into various aspects of enterovirus infections of the GI tract, in vivo models are also needed to understand complex interactions that occur during enteroviral infections, such as the interaction of viral particles with bacteria or the complex interaction with the immune system. One of the first mouse models to study enterovirus infection was the transgenic PV receptor mouse. The authors demonstrated that mice expressing the human homologue of the poliovirus receptor (PVR) were able to be infected with PV through intracerebral injection, where they displayed signs of paralysis similar to human disease \[[@B140-viruses-11-00460]\]. Due to the route of infection, animals did not need to be immunosuppressed such as blocking and depleting type I interferon receptor (IFNAR). Since this model was established, others have developed models for other enterovirus infection using a variety of different methods. Many mouse models of enterovirus infection use "humanized" mice that express the human form of the viral receptor. Since mice are not the natural host of enteroviruses, the mouse homologs of the entry receptors are often not sufficient for infection or the affinity of viruses much less. Several groups have used this strategy, including generating "knock-in" animals expressing human SCARB2 for EV71 infection \[[@B141-viruses-11-00460]\]. Very commonly, ablation of IFNAR or oral infection at high viral doses is required to generate in vivo mouse models of enteroviruses \[[@B142-viruses-11-00460],[@B143-viruses-11-00460],[@B144-viruses-11-00460]\]. These strategies allow the infection of enteroviruses in mice, which normally do not support robust replication. However, many of these models are based on IP injection or other non-oral infection routes. However, this route of administration bypasses the primary site of infection observed in humans. Thus, models that include oral infection are imperative to understand how enteroviruses infect the GI tract and disseminate into secondary target tissues causing clinical disease. Several models of oral infection have been established for a E11, PV, CVB, and EV71, and have been shown to recapitulate human disease \[[@B53-viruses-11-00460],[@B143-viruses-11-00460],[@B145-viruses-11-00460],[@B146-viruses-11-00460],[@B147-viruses-11-00460]\]. An adult model of oral infection of CVB using IFNAR-deficient mice investigated the pathogenesis of a mutant CVB virus that emerged after passage through a mouse that exhibited a large plaque phenotype \[[@B142-viruses-11-00460]\]. In addition to adult mouse models of oral infection, several studies have established neonatal infection models for a number of enteroviruses. We recently established a neonatal model for E11 infection by the enteral route in human transgenic mice expressing the human homolog of FcRn \[[@B53-viruses-11-00460]\] and showed that only transgenic mice exhibited viral replication in the small intestine, liver, and blood seven days post oral infection \[[@B53-viruses-11-00460]\]. An oral infection model of PV was established using transgenic animals expressing the human homolog of PVR and using IFNAR-deficient mice \[[@B143-viruses-11-00460]\]. Neonatal, transgenic, IFNAR-deficient mice infected with PV exhibited viral replication in the blood and small intestine two and three days post oral infection \[[@B143-viruses-11-00460]\]. In addition to E11 and PV, multiple neonatal models of oral EV71 infection have been established \[[@B146-viruses-11-00460],[@B147-viruses-11-00460]\]. One study established an oral infection model using chimeric receptor-expressing transgenic mice, showing that oral infection of clinical isolates of EV71 leads to viral replication in the stomach, small intestine, colon, and brain seven days post infection \[[@B146-viruses-11-00460]\]. Another model of EV71 infection using outbred mice showed seven-day-old outbred mice that were orally infected with EV71 displayed skin rashes early during infection, which progressed to hind limb paralysis \[[@B147-viruses-11-00460]\]. In addition to mouse models, several studies have used non-human primate models to study EV71, CVB, and PV pathogenesis. One study showed that rhesus monkeys can be infected with EV71 through the intravenous, respiratory, and oral routes but had limited viral replication in the blood after intracerebral infection \[[@B148-viruses-11-00460]\]. This study showed that EV71 disseminated to the brain and causes neuropathological damage \[[@B148-viruses-11-00460]\]. Moreover, oral infection models of EV71 have been established in cynomolgus monkeys that showed degeneration and necrosis of neurons in the central nervous system of infected monkeys \[[@B149-viruses-11-00460]\]. Additionally, neonatal rhesus monkeys animals infected with EV71 show signs of clinical hand-foot-and-mouth disease as seen in humans \[[@B150-viruses-11-00460]\]. Furthermore, a model of CVB-induced myocarditis was established using cynomolgus monkeys which exhibited viral myocarditis similar to human disease. This study showed that, following intravenous inoculation with CVB, animals experienced myocardial injury and inflammatory cells infiltration in the heart of infected animals \[[@B151-viruses-11-00460]\]. Another study using patas monkeys showed that intravenous CVB infection caused abnormalities in blood glucose as well as impaired insulin secretion \[[@B152-viruses-11-00460]\]. In addition to EV71 and CVB, several models to study PV have been established which include oral, subcutaneous, intravenous, intraspinal, and intracerebral infection \[[@B153-viruses-11-00460],[@B154-viruses-11-00460],[@B155-viruses-11-00460],[@B156-viruses-11-00460],[@B157-viruses-11-00460]\]. These models were incredibly important for understanding the immune response to the PV vaccine \[[@B17-viruses-11-00460],[@B158-viruses-11-00460]\]. One study showed that rhesus, cynomolgus, and bonnet macaques were all susceptible to oral PV infection \[[@B156-viruses-11-00460]\]. When these macaques were fed PV, they developed paralysis \[[@B156-viruses-11-00460]\]. Other studies have shown that infant cynomolgus monkeys that were fed PV developed paralytic poliomyelitis \[[@B159-viruses-11-00460]\]. Together, the use of in vitro models and in vivo models, including mouse models and non-human primate models, will aid in our understanding of enterovirus entry, detection by the host immune response, and evasion mechanisms these viruses use to subvert the innate immune response. 6. Concluding Remarks {#sec6-viruses-11-00460} ===================== Enteroviruses remain a significant global public health concern. The field has made significant progress in determining how enteroviruses are detected by host cells and the mechanisms they use to evade this detection. With the continued development of in vitro, ex vivo, and in vivo models that fully recapitulate the GI epithelium, we will gain a better understanding of the mechanisms used by enteroviruses to breach the intestinal barrier. These models could also facilitate the development of novel therapeutic targets and/or strategies to prevent or treat enterovirus infections and ultimately alleviate morbidity and mortality caused by these infections. Our work on enteroviruses is supported by NIH R01-AI081759 (C.B.C.) and the Children's Hospital of Pittsburgh of the UPMC Health System (C.B.C). The author declares no conflict of interest. {#viruses-11-00460-f001} {#viruses-11-00460-f002} {#viruses-11-00460-f003} {#viruses-11-00460-f004} viruses-11-00460-t001_Table 1 ###### Enterovirus receptors and attachment factors. Host Protein EV Serotype Role Reference --------------------------- ----------------------------------------------------------------------------------------------------------- --------------------------- ----------------------------------------------------------------------------- PVR (CD155) PV Binding, entry, uncoating \[[@B52-viruses-11-00460]\] CAR CVB1, CVB2, CVB3, CVB4, CVB5, and CVB6 Binding, entry, uncoating \[[@B49-viruses-11-00460],[@B50-viruses-11-00460],[@B51-viruses-11-00460]\] DAF CVA21, CVB1, CVB3 (some isolates), CVB5, E3, E6, E7, E11, E12, E13, E19, E19, E20, E21, E25, E29, and E30 Attachment \[[@B55-viruses-11-00460],[@B56-viruses-11-00460]\] SCARB2 EV71, CVA7, CVA14, and CVA16 Binding, entry, uncoating \[[@B46-viruses-11-00460]\] PSGL1 EV71, CVA2, CVA7, CVA10, CVA14, and CVA16 Attachment \[[@B47-viruses-11-00460],[@B59-viruses-11-00460]\] KREMEN1 CVA10 Binding, entry \[[@B60-viruses-11-00460]\] Sialic acid EV71 Attachment \[[@B61-viruses-11-00460]\] ICAM5 EV-D68 Binding, entry \[[@B62-viruses-11-00460]\] Integrin ⍺~2~β~1~ (VLA-2) E1 Binding, entry, uncoating \[[@B63-viruses-11-00460]\] FcRn Echoviruses Binding, entry, uncoating \[[@B53-viruses-11-00460],[@B54-viruses-11-00460]\] viruses-11-00460-t002_Table 2 ###### Enterovirus targets of PRRs. Host Protein EV Serotype Mechanism of Cleavage Reference -------------- ------------- ------------------------------------------------------------------ ------------------------------------------------------- TRIF CVB3 3C^pro^ \[[@B114-viruses-11-00460]\] EV-D68 3C^pro^ \[[@B115-viruses-11-00460]\] EV71 3C^pro^ \[[@B116-viruses-11-00460]\] RIG-I EV71 Decreases ubiquitination of RIG-I inhibiting recruitment to MAVS \[[@B117-viruses-11-00460],[@B118-viruses-11-00460]\] PV 3C^pro^ \[[@B119-viruses-11-00460]\] MDA5 CVA6 3C^pro^ \[[@B120-viruses-11-00460]\] CVA16 3C^pro^ \[[@B120-viruses-11-00460]\] CVB3 2A^pro^ \[[@B121-viruses-11-00460]\] EV-D68 3C^pro^ \[[@B120-viruses-11-00460]\] EV71 Unknown \[[@B104-viruses-11-00460]\] PV Caspase Dependent \[[@B122-viruses-11-00460]\] MAVS CVB3 2A^pro,^ 3C^pro^ \[[@B114-viruses-11-00460],[@B121-viruses-11-00460]\] EV71 2A^pro^ \[[@B121-viruses-11-00460],[@B123-viruses-11-00460]\]

I[NTRODUCTION]{.smallcaps} {#sec1-1} ========================== Cannabinoid compounds are being extensively studied for their wide range of therapeutic potentials. Although there is an abundance of information on the role of cannabinoids in the central and peripheral nervous systems, their role in the cardiovascular system seems to be more complex. The aim of this review is to cover the role that cannabinoids play in the treatment or prevention of various cardiovascular diseases including blood pressure, metabolic disorders, atherosclerosis, myocardial infarction, and cardiac arrhythmia. In addition, the most recent advances in this field and the challenges and negative effects of cannabinoids on human participants will be highlighted. H[ISTORY OF]{.smallcaps} C[ANNABINOIDS AND]{.smallcaps} T[HEIR]{.smallcaps} M[ECHANISM OF]{.smallcaps} A[CTION]{.smallcaps} {#sec1-2} =========================================================================================================================== Cannabinoids have traditionally been known for acting on the central nervous system. tetrahydrocannabinol (THC), which was first extracted from Cannabis sativa by Gaoni and Mechoulam in 1964, has been extensively studied and shown to have psychoactive properties.\[[@ref1]\] Since then, it has been shown that cannabis contains over 80 active constituents.\[[@ref2]\] In addition to naturally occurring cannabinoids, several endogenous cannabinoid compounds (endocannabinoids), such as N-arachidonoyl ethanolamine (AEA)\[[@ref3]\] and 2-arachidonoyl glycerol (2-AG),\[[@ref4][@ref5]\] have been identified and examined. AEA is inactivated *in vivo* by fatty acid amidohydrolase\[[@ref6]\] whereas 2-AG is inactivated by monoglyceride lipase with other enzymes playing a more minor role in their inactivation.\[[@ref7]\] Cannabinoid receptors consist of two main forms, namely cannabinoid 1 receptor (CB~1~R) and cannabinoid 2 receptor (CB~2~R). They are both classified as G-protein-coupled receptors.\[[@ref8]\] CB~1~R is present in the central nervous system as well as other peripheral sites such as the cardiovascular, respiratory, reproductive, and digestive systems.\[[@ref9]\] CB~2~R is a peripheral receptor found in immune cells and organs.\[[@ref10]\] The behavioral properties of cannabinoids have been shown to be mediated by CB~1~R\[[@ref8][@ref11][@ref12]\] whereas CB~2~R is mainly implicated in immunomodulation.\[[@ref13]\] Several studies have explored targeting the endocannabinoid system to treat various vascular diseases and disorders such as myocardial\[[@ref14]\] and cerebral ischemia, hypertension\[[@ref15]\] circulatory shock,\[[@ref16]\] atherosclerosis, metabolic syndrome, stroke,\[[@ref17]\] arrhythmia, and myocardial infarction.\[[@ref18]\] Furthermore, studies have reported positive as well as negative effects of cannabinoids when used to combat cardiac disorders\[[@ref19]\] and there seems to be a complicated relationship between the endocannabinoid system, the cardiovascular, and the immune system. E[FFECTS OF]{.smallcaps} C[ANNABINOIDS ON THE]{.smallcaps} H[EART]{.smallcaps}, B[LOOD]{.smallcaps} P[RESSURE, AND]{.smallcaps} V[ASCULATURE]{.smallcaps} {#sec1-3} ========================================================================================================================================================= It is well known that acute exposure to cannabis leads to tachycardia, although the effect on blood pressure is less consistent.\[[@ref20][@ref21][@ref22]\] On the contrary, chronic exposure was reported to cause bradycardia and a lowering of blood pressure.\[[@ref15][@ref23]\] The endocannabinoids AEA and 2-AG are present in various parts of the vasculature including red blood cells, platelets, serum and vascular cells, and the myocardium.\[[@ref24][@ref25][@ref26]\] The activation of CB~1~R in the myocardial tissue produces a negative inotropic response on the heart.\[[@ref17][@ref27][@ref28][@ref29]\] Although the expression of CB~2~R has been demonstrated in cardiac myocytes\[[@ref30][@ref31]\] and in endothelial and smooth muscle cells of coronary arteries,\[[@ref32][@ref33]\] their role has been less well characterized and requires further investigation.\[[@ref15]\] The effects of endocannabinoids on blood pressure are not different from those of cannabis. The intravenous administration of AEA and 2-AG has been reported to cause a triphasic response that ultimately causes a lowering of blood pressure in experimental animals.\[[@ref34]\] Although it was postulated in the 1970s that modulation of cannabinoid receptors could lead to the development of blood pressure-lowering agents, this was complicated by an overlap in the cardiovascular and neurological effects.\[[@ref15]\] In recent years, however, it was shown that cannabinoids have a more profound blood pressure-lowering effect on hypertensive animals when compared to their nondiseased counterparts.\[[@ref17][@ref28][@ref35][@ref36][@ref37]\] This has led to the rebirth of the hypothesis that cannabinoid ligands may indeed be used to target hypertension. The vasodilatory response of endocannabinoids has been shown to be partially attributed to the blockage of norepinephrine release from perivascular nerves in the sympathetic nervous system,\[[@ref38]\] and partially, due to their ability to directly affect endothelial and vascular smooth muscle.\[[@ref39]\] When the effects of synthetic and endocannabinoids were studied on isolated blood vessels, there seemed to be a complex range of responses which proved to be species and tissue dependent.\[[@ref17]\] In addition to the CB~1~R, the involvement in a variety of other receptors, including the transient receptor potential vanilloid 1 receptor (TRPV1), has been shown to mediate the vascular effects of cannabinoids.\[[@ref23][@ref40][@ref41]\] The activation of perivascular TRPV1 receptors by AEA in animal studies causes a cascade of events, whereby calcitonin gene-related peptide and various neuropeptides are produced ultimately activating vascular potassium channels and producing a dilatory response.\[[@ref25][@ref42][@ref43]\] The vasodilatory effect of 2-AG has also been linked to the activation of TRPV4 in the endothelium and which ultimately activates calcium-dependent potassium channels in vascular smooth muscle cells.\[[@ref44]\] The endocannabinoids, AEA, and 2-AG were shown to activate peroxisome proliferator-activated receptors (PPAR).\[[@ref45]\] These may be exploited when using cannabinoids in treating cardiometabolic conditions since PPAR have widespread use in conditions displaying inflammation and tissue fibrosis.\[[@ref46]\] Therefore, the actions of cannabinoids on vascular tone are complex and are mediated by a vast number of receptors which activate a wide range of signaling pathways.\[[@ref22]\] Cannabinoids may also affect local blood flow since they possess autocrine functions.\[[@ref47]\] Studies have shown that endocannabinoid release in isolated arteries increases after exposure to vasoconstrictors such as angiotensin II.\[[@ref24][@ref48][@ref49]\] To further support this activity, the presence of enzymes that metabolize cannabinoids in the vasculature has also been shown in a number of studies.\[[@ref24][@ref25]\] It has been demonstrated that endocannabinoids are released in ample amounts by white blood cells and platelets in certain inflammatory diseases.\[[@ref26][@ref28][@ref50]\] These endocannabinoids subsequently exert their actions on the heart and vascular cells leading to vasodilation, lowering of blood pressure, and negative inotropy. Furthermore, various studies suggest that endocannabinoids may even mediate tissue remodeling in certain diseases.\[[@ref51][@ref52][@ref53]\] C[ANNABINOIDS IN THE]{.smallcaps} T[REATMENT OF]{.smallcaps} M[ETABOLIC]{.smallcaps} S[YNDROME AND ITS]{.smallcaps} A[SSOCIATED]{.smallcaps} C[ARDIAC]{.smallcaps} C[OMPLICATIONS]{.smallcaps} {#sec1-4} ============================================================================================================================================================================================== The role of the endocannabinoid system in metabolic syndrome has been an area of growing research interest over the past several decades. Furthermore, several animal models and clinical human trials have explored the therapeutic potential of CB~1~R antagonists, such as rimonabant, for the treatment of metabolic diseases.\[[@ref15]\] The expression of the CB~1~R has been demonstrated in adipose tissue, whereby its activation is believed to cause an augmentation in lipolysis and an attenuation of oxidation of fatty acids.\[[@ref54]\] CB~1~R is also expressed in the liver where it is block counteracts fats and stenosis.\[[@ref55]\] There have been reports of improved insulin resistance and glucose tolerance after CB~1~R blockage in diet-induced and genetic animal obesity studies.\[[@ref56]\] Another beneficial effect of CB~1~R blockage is the reversal of reduced heart contractility seen in many patients with cardiac diseases.\[[@ref30][@ref31]\] The patients suffering from obesity tend to have reduced heart contractility,\[[@ref57][@ref58]\] although this is frequently counteracted by excessive sympathetic stimulation in these patients with a net result of a slight elevation of blood pressure, as is commonly seen in metabolic disorders where norepinephrine levels are reported to increase more than two-fold.\[[@ref59]\] Rimonabant administration does not cause any change in blood pressure which further supports the idea of opposing mechanisms of action. Although cannabinoid antagonists may have beneficial effects, their use has been largely limited by their neuropsychiatric adverse effects such as depression and anxiety disorders.\[[@ref60][@ref61][@ref62][@ref63]\] Therefore, it is imperative that peripheral CB~1~R antagonists are developed which possess the beneficial effects, yet lack the unwanted adverse effects. C[ANNABINOIDS FOR THE]{.smallcaps} T[REATMENT OF]{.smallcaps} A[THEROSCLEROSIS]{.smallcaps} {#sec1-5} =========================================================================================== It is still unclear whether cannabinoids can be truly useful for the treatment of atherosclerosis. Although endocannabinoids may have beneficial effects in certain cardiovascular disorders, they actually possess a prothrombic effect.\[[@ref64]\] Studies in humans and rats have reported platelet activation by anandamide and 2-AG. Platelets have a well-recognized role in maintaining blood homeostasis and in the formation of thrombi. Furthermore, the platelets possess anti-inflammatory actions and regulate growth. These properties may collectively contribute to the development of atherosclerosis.\[[@ref65]\] Endocannabinoids may also be produced by platelets, macrophages, and endothelial cells, when an atherosclerotic plaque is formed. The increased level of endocannabinoids is counteracted by the metabolic properties of platelets, macrophages, and endothelial cells.\[[@ref64]\] CB~2~R receptors have been proposed to play a role in the development of atherosclerosis. A study in an experimental model of atherosclerosis showed that THC in low doses was beneficial for treating atherosclerotic plaques.\[[@ref66]\] The same study showed that the CB~2~R receptor blocker SR144528 reversed this effect. Both synthetic and endogenous cannabinoids are involved in the mobilization of immune cells by CB~2~R activation.\[[@ref60]\] T[HE]{.smallcaps} R[OLE OF]{.smallcaps} E[NDOCANNABINOIDS IN]{.smallcaps} M[YOCARDIAL]{.smallcaps} I[NFARCTION]{.smallcaps} {#sec1-6} =========================================================================================================================== Several studies point to the important beneficial effects of endocannabinoids in protecting against myocardial infarction. One of the earliest studies testing this phenomenon looked at the effects on endocannabinoids on ischemic-isolated hearts with induced shock.\[[@ref67]\] In that study, the cardioprotection was abolished after administering the CB~2~R antagonist SR144528. Blocking CB~1~R had no effect on the cardioprotection.\[[@ref67]\] Other studies have reported that endocannabinoids lead to decreased necrosis of the myocardial tissue as well as lowering the risk of arrhythmia in acute myocardial infarction, and they are also involved in remodeling in the chronic stage.\[[@ref68][@ref69]\] The protective effect of endocannabinoids may be attributed to an action on endothelial cells as well as other mechanisms involving CB~1~R and CB~2~R.\[[@ref70][@ref31]\] CB~2~R has been shown to play a key protective role when activated.\[[@ref9]\] This has been linked to an attenuation of the inflammatory response and endothelial activation. Studies on knock-out mice also confirm the important role of CB~2~R activation in protecting against myocardial infarction.\[[@ref9][@ref71]\] Intravenous AEA injection was shown to upregulate heat-shock protein 72 in cardiac tissues. This provided cardiac protection against ischemia/reperfusion injuries in rodents. This effect was unaffected by the administration of a CB~1~R blocker, but abolished after CB~2~R blockade further supporting the important role of CB~2~R in myocardial infarction.\[[@ref72]\] Other studies looked at the benefits of CB~2~R agonists in mice suffering from myocardial infarction using a model of ischemia/reperfusion and showed that mitogen-activated protein kinase as well as protein kinase C maybe involved in the CB~2~R-mediated protection.\[[@ref73][@ref74]\] The effects on leukocyte migration are thought to underlie the cardioprotective effects of cannabinoids, as demonstrated in a study by Di Filippo *et al*.\[[@ref75]\] They showed that the activation of CB~2~R reduced the myocardial infarction size by half and caused a decline in leukocyte migration. This effect was reversed when a cannabinoid antagonist was administered further supporting their findings.\[[@ref75]\] C[ANNABINOIDS FOR]{.smallcaps} T[REATING]{.smallcaps} C[ARDIAC]{.smallcaps} A[RRHYTHMIA]{.smallcaps} {#sec1-7} ==================================================================================================== Since most of the patients that suffer from a myocardial infarction tend to develop arrhythmia as a complication, it is also important to explore the role that the cannabinoid system plays in controlling this detrimental complication. Endocannabinoids have been shown to decrease the risk of developing cardiac arrhythmia, an effect which is mediated mainly by CB~2~R.\[[@ref70][@ref76][@ref77][@ref78]\] A study looking at the effects of AEA in an arrhythmia model induced by ischemia found that it led to a significant decrease in the arrhythmia events.\[[@ref77]\] This effect of AEA disappeared if the CB~2~R antagonist SR144528 was given, but not after administration of the CB~1~R antagonist rimonabant.\[[@ref78]\] The CB~2~R agonist, HU-210, showed nearly a complete reversal of the arrhythmia observed in both an ischemia model and epinephrine-induced and aconitine-induced model.\[[@ref79][@ref80][@ref81]\] All these studies highlighted the important role that CB~2~R receptors play over CB~1~R in the antiarrhythmia effects of cannabinoids.\[[@ref9]\] Furthermore, a study of cannabidiol in rats with cardiac injury showed that it caused a significant reduction in the ischemia-induced arrhythmias\[[@ref82]\] an effect which was later shown to be mediated by potassium channels.\[[@ref83]\] N[EGATIVE]{.smallcaps} E[FFECTS OF]{.smallcaps} C[ANNABINOIDS ON THE]{.smallcaps} C[ARDIOVASCULAR]{.smallcaps} S[YSTEM IN]{.smallcaps} H[UMANS]{.smallcaps} {#sec1-8} =========================================================================================================================================================== Although several animal studies have provided a promising role for both endocannabinoids and synthetic cannabinoids in treating and preventing certain cardiovascular disorders, this unfortunately does not always translate well in human trials. In fact, a large body of research has shown that the signaling of CB~1~R can have detrimental effects on the cardiovascular system. Drop in blood pressure,\[[@ref84]\] tachycardia,\[[@ref84]\] increased heart rate\[[@ref85]\] and higher incidence of a heart attack were all reported in otherwise healthy young cannabis users.\[[@ref86]\] Furthermore, it has been demonstrated that the synthetic cannabinoid K2 may cause healthy young children to suffer from myocardial infarction.\[[@ref87]\] Another study by Heath *et al*.\[[@ref88]\] reported that adolescents suffered from an increase in heart rate, unconsciousness, and pain. It is postulated that the mediation of CB~1~R leads to unwanted cardiac effects whereas CB~2~R has more protective and anti-inflammatory action.\[[@ref9][@ref89]\] Other than affecting the cardiovascular system, the endocannabinoid system was reported to play a key role in various body functions including the gastrointestinal tract.\[[@ref90]\] Endocannabinoids have also been reported to mediate the activities of cyclooxygenase enzyme, which is activated in certain gastric inhibitory responses to THC as well as in the hindbrain to mediate the inhibitory cardiac effects.\[[@ref91]\] C[ONCLUSION]{.smallcaps} {#sec1-9} ======================== Cannabinoids are increasingly being recognized for their wide range of therapeutic effects both in the cardiovascular system and on other systems in the human body. Both endocannabinoids and synthetic cannabinoid compounds have been widely studied and proven to be useful in treating a large number of cardiac disorders. However, there still seems to be a dimension of complexity that accompanies these compounds which exert effects on CB~1~R, CB~2~R as well as other noncannabinoid receptors. Although several rodent models have shown promising actions of cannabinoids on the cardiovascular system, indeed we are a long way from seeing these compounds on the market. It is, therefore, required that the cannabinoid compounds with promising effects in animal studies be taken to the next stage and studied in humans. These clinical studies will shed light on the true therapeutic potential of cannabinoids in the cardiovascular system. Financial support and sponsorship {#sec2-1} --------------------------------- Nil. Conflicts of interest {#sec2-2} --------------------- There are no conflicts of interest.

Polycystic ovary syndrome (PCOS) is a complex disorder affecting about 5% to 15% of women in the reproductive age group and accounting for 75% of cases of anovulatory infertility.[@b1-asm-4-296] Women with PCOS have a significant degree of insulin resistance and compensatory hyperinsulinemia[@b2-asm-4-296] and a high prevalence of type 2 diabetes has been documented in these women.[@b3-asm-4-296] Hyperinsulinaemia in PCOS has been observed to lead to hyperandrogenism, which creates a hostile hormonal milieu for ovulatory cycles and for conception to occur.[@b4-asm-4-296] Thus, targeting insulin resistance and hyperinsulinemia can favourably alter the hormonal environment allowing ovulation and conception. There is some evidence to suggest that the thiazolidinediones may alter the intra-ovarian hormonal milieu of women with PCOS. Aziz et al[@b5-asm-4-296] reported that troglitazone improves defects in insulin action, insulin secretion and ovarian steroidogenesis in women with PCOS. Rosiglitazone has been reported to improve insulin sensitivity, and improve the menstrual pattern,[@b6-asm-4-296] decrease ovarian androgen production and improve ovulatory cycles in women with PCOS.[@b7-asm-4-296] The use of rosiglitazone in a young woman with PCOS has ameliorated insulin resistance, lowered androgen levels and helped spontaneous conception.[@b8-asm-4-296] Recently Vaughan and Bell reported a 46-year-old woman with PCOS who conceived spontaneously following the use of rosiglitazone to treat supervening diabetes.[@b9-asm-4-296] We report two women with PCOS and longstanding primary infertility, where the addition of rosiglitazone to metformin to treat type 2 diabetes resulted in spontaneous pregnancy with a viable singleton baby in the first woman and twin babies in the second case. Case 1 ====== A 30-year-old Saudi female, diagnosed as suffering from PCOS since her teenage years, had been followed and managed for years in another hospital. The diagnosis was based on a clinical profile of obesity, mild hirsutism and irregular cycles as well as polycystic ovaries on pelvic ultrasound. She had been trying for pregnancy since age 20, had received several courses of clomiphene citrate, and eventually had a total of three trials of in vitro fertilization (IVF), but with no success. At age 25 she developed type 2 diabetes, following presentation with frank osmotic symptoms, hyperglycaemic malaise and a random blood glucose of 15 mmol/L. There was no ketonuria or profound weight loss. She was treated initially with metformin at a dose of 500 mg twice daily, later increased to 2 g daily. For the last 4 years she had been followed in our diabetic clinic, and continued to have significant obesity (height 160 cm, weight 87 kg, BMI 34 kg/m^2^), and erratic menstrual cycle and mild hirsutism. Despite encouragement for physical exercise and dieting, her metabolic control failed to improve with HbA1c remaining elevated \>8.5% (normal reference non-diabetic range, 4.5% to 6%). In early 2003, rosiglitazone maleate at a dose of 4 mg was added to achieve better glycemic control. The patient noticed some changes in the pattern of her menstruation, but overall they remained erratic, occurring every 2 to 3 months. She put on more weight, an average of 3--5 kg since the start of rosiglitazone, but her HbA1c dropped to 6.6% in June 2003. In July 2003, she had persistent symptoms of nausea and vomiting occurring early in the morning, and by then she was 4-months amenorrhoeic. Pregnancy was suspected and a subsequent pregnancy test was found to be positive. Ultrasonography confirmed a viable 17-week-old fetus. Metformin and rosiglitazone were both stopped and twice-daily pre-mixture insulin was started to control glycaemia. Pregnancy progressed well with no complications and the baby was delivered at 37 weeks by elective caesarean delivery due to fetal macrosomia. A healthy male baby weighing 4 kgs was born who had no neonatal complications and achieved normal developmental milestones on regular follow up. Case 2 ====== A 41-year-old British woman of Caucasian descent had PCOS with a longstanding history of primary infertility for which she underwent extensive investigations at a fertility clinic from 1984 to 1992. She received several treatment modalities for ovulation induction, including two long courses of clomiphene citrate, and laparascopic ovarian electrocautery with no successful outcome. She was finally offered the option of IVF, which she was not too keen to pursue. In 1997, by then age 35 years, she was diagnosed with type 2 diabetes, following symptoms of hyperglycaemia and a random blood glucose of 14 mmol/L and HbA1c of 8.8% (normal 4.5--5.6%). On initial assessment she was noted to be morbidly obese with a body mass index (BMI) of 46 kg/m^2^, a normal blood pressure and significant hirsutism with a Ferrimen-Galleway score of 13. She had normal renal, hepatic, calcium and lipid profiles. She was started on metformin 500 mg three times daily, given dietary advice and encouraged to adopt a more active life style with a view toward weight loss. On follow up in the subsequent year she managed to lose about 6 kg of body weight, and her menstrual cycles improved to some extent but remained erratic. Her diabetes control, however, remained poor with HbA1c running \>10%. At this stage acarbose 50mg three times daily was introduced to achieve a better diabetic control. By October 2000, her diabetes control was still unsatisfactory, so rosiglitazone maleate 4 mg was added and later increased to 8 mg by early 2001. During this time she was noted to have hypertension, so perindopril at a dose of 2.5 mg was started and later felodipine 5 mg was added. With the above regimen her glycaemic control improved with the HbA1c dropping to 6% by the end of 2001. Also she realised that her menstruation showed some improvement, despite that she had gained an average of 5 kg in weight over the preceding 12 months. In August 2002, now older than 40 years of age, she visited her general practitioner with symptoms of breast fullness and tenderness. By then, she was amenorrhoeic for seven months but was not particularly bothered about it, as this could be quite normal for her. Her primary care physician suspected that she might be pregnant but the patient denied any symptoms of pregnancy. A pregnancy test turned out to be positive, so she was referred to our joint antenatal-diabetic clinic where ultrasonographic examination confirmed pregnancy with twins with an estimated gestational age of 27 weeks. Both metformin and rosiglitazone were discontinued, and the ACE inhibitor was substituted for labetalol 200 mcg/d. The pregnancy progressed well in subsequent weeks, with diet only to control her diabetes, and her blood sugar series were running within the target range for our clinic. By 32 weeks she had premature rupture of the membranes for which she was admitted to hospital and managed conservatively with beclomethasone. Following this her glycaemic control deteriorated significantly necessitating insulin therapy. Two weeks later she delivered two healthy male babies by elective caesarean delivery. The babies spent a few weeks in a special care baby unit, were breastfed, and later discharged in good health. On follow up both babies were well and healthy and achieving normal developmental milestones. Discussion ========== Insulin resistance and compensatory hyperinsulinemia are established as the central pathophysiologic mechanisms in women with PCOS. There is a mounting body of evidence to suggest that tackling insulin resistance either via physical exercise, weight loss or by use of insulin sensitizers can improve the menstrual pattern and aid conception in these women.[@b10-asm-4-296],[@b11-asm-4-296] One might argue that the successful conception in our cases occurred by shear chance and has nothing to do with the medication, as spontaneous conception can occur in women with PCOS and longstanding infertility. However, we argue that the use of the two drugs (to treat diabetes) has resulted in ovulation induction and spontaneous conception. Metformin acts primarily by decreasing hepatic glucose output and enhancing peripheral glucose uptake and has in addition an insulin sensitizing effect. It has been shown to be helpful in restoring a normal menstrual cycle,[@b12-asm-4-296] improving ovulation rate,[@b13-asm-4-296] and enhancing the conception rate when used alone[@b14-asm-4-296] or in combination with clomiphene citrate in women with PCOS.[@b15-asm-4-296] Moghetti et al,[@b16-asm-4-296] in a randomised placebo controlled trial, showed that long-term metformin treatment in obese PCOS women reduces hyperinsulinemia and hyperandrogenaemia independently of changes in body weight and improved menstrual abnormality. The thiazolidindiones directly target insulin resistance by increasing insulin sensitivity and glucose uptake in the muscle and to some extent in the liver via their action on the peroxisome proliferator-activator receptor gamma.[@b17-asm-4-296] This direct effect on insulin resistance could influence the intra-ovarian hormonal milieu in PCOS similar to that achieved with metformin. Troglitazone, the first of the thiazolidindione group was shown in a few studies to increase the number of ovulatory cycles when used alone or in combination with clomiphene citrate.[@b5-asm-4-296],[@b18-asm-4-296] The newer members, rosiglitazone and pioglitazone can potentially produce the same beneficial effects in PCOS but there are limited data to support their use in ovulation induction. Shobokshi and Shaarawy[@b19-asm-4-296] reported an improvement of insulin sensitivity via reducing IGF-1 bioavailability following combined use of rosiglitazone with clomiphene citrate, a finding confirmed by Ghazeeri et al.[@b20-asm-4-296] Recently Yilmaz et al demonstrated that both metformin and rosiglitazone improve androgen profile, menstrual cyclicity and the hirsutism score with somewhat superior effect of rosiglitazone over metformin.[@b21-asm-4-296] The same group later reported that both agents also improve insulin sensitivity and serum androgen level in lean and obese women with PCOS.[@b22-asm-4-296] In a study involving non-obese women with PCOS, the frequency of ovulation following treatment with rosiglitazone improved significantly, though the combination of rosiglitazone with metformin was not superior over metformin alone.[@b23-asm-4-296] However the combination of pioglitazone and metformin was superior to metformin alone in ameliorating insulin resistance, hyperandrogenism and improving menstruation in women with PCOS.[@b24-asm-4-296] On the other hand, metformin alone has been shown to be less effective in obese women with PCOS.[@b13-asm-4-296] Our two patients had been on metformin for some time, with some beneficial effects upon their menstrual cycles. We hypothesise that the addition of rosiglitazone to metformin to improve diabetes control resulted in unexpected ovulation and conception. It is plausible that the two drugs in combination have corrected the metabolic defect(s) that were responsible for their longstanding infertility and paved the way for successful conception. Our report compliments the observation of Vaughan and Bell where the addition of rosiglitazone to metformin resulted in spontaneous conception in a perimenopausal woman with PCOS.[@b9-asm-4-296] Arlt et al[@b25-asm-4-296] demonstrated that the thiazolidinedione class of drugs have a direct inhibitory effect on the steroidogenic enzymes P450c17 and 3βHSD. Metformin has been recently reported to possess similar property.[@b26-asm-4-296] Those cases highlight that women with PCOS and type 2 diabetes may conceive spontaneously following the use of this combination, and this fact should be pointed out to them and they should be counselled for such an outcome. Furthermore the safety issue of the use of the thiazolidinediones, including potential teratogenicity, should not be forgotten, despite the fact that there is some anecdotal evidence of its safety in pregnancy.[@b9-asm-4-296],[@b27-asm-4-296],[@b28-asm-4-296] Therefore, these women should be warned about such a possibility and where appropriate contraceptive measures should be advised. In conclusion, these case series highlight the need for patient awareness of the possible effect of the thiazolidinediones, either alone or in combination with metformin on fertility in women with PCOS. Taking into account their direct insulin-sensitising effect, these agents may be capable in solo in initiating ovulation and aid conception in women with PCOS. This may prove more attractive as a new therapeutic armentarium to treat infertility in these women. However, proof awaits the completion of randomised controlled trials. In the meantime, women should be counselled and where appropriate contraceptive measures offered. We have no conflict of interest and no source of outside funding

Aspiration of oral and gastric secretions is a well-known complication of intubation and mechanical ventilation. Aspiration can lead to complications, such as ventilator-associated pneumonia (VAP), pneumonitis, and other aspiration syndromes ([@R1]--[@R4]). Colonization of oropharyngeal secretions that accumulate above the cuff of the endotracheal tube (ETT), and bacterial translocation from the stomach contribute to the pathogenesis of VAP and pulmonary complications associated with aspiration ([@R5], [@R6]). It is challenging to accurately detect when aspiration occurs, source of aspiration (oral or gastric), and the degree of aspiration. Detection of pepsin in tracheal or bronchoalveolar lavage (BAL) secretions is often considered the gold standard for identifying aspiration of gastric secretions ([@R7], [@R8]). Researchers have assessed levels of salivary (alpha) amylase in pulmonary secretions to detect aspiration of oral secretions. Alpha-amylase is a digestive enzyme that is present in saliva. If oral secretions are aspirated, α-amylase should be detectable in pulmonary secretions, but data related to interpretation of α-amylase values are limited. This brief article summarizes what is known about α-amylase values and discusses issues associated with its use as a biomarker for detection of aspiration of oral secretions. ALPHA-AMYLASE VALUES FOR DIFFERENT SPECIMEN SOURCES =================================================== Alpha-amylase values are highest in the mouth (normal expected finding) and decrease depending on proximity of sample collection from the mouth. **Table [1](#T1){ref-type="table"}** compares α-amylase values reported for oral, subglottic, tracheal, and BAL specimens. Because of extremes in values, median levels provide more meaningful comparative data. Median values for "oral" samples ranged between 195,029 to 307,606 U/L ([@R9], [@R10]). "Subglottic" secretions were assessed in only one study, and the median α-amylase value was 130,750 U/L ([@R9]). Median values of α-amylase in "tracheal" secretions varied widely from 0 in a small pilot study to over 22,000 U/L in patients with documented aspiration or VAP ([@R10], [@R11], [@R15]). Median values for "BAL" specimens ranged from 134 U/L to 927 U/L; a mean value of 1,722 U/L was reported in those at high risk for aspiration ([@R17]--[@R20]). ###### Levels of Alpha-Amylase Based Upon Type of Specimen  It has been suggested that the ratio of tracheal values to oral α-amylase values (T/O ratio) is an indicator of the degree of aspiration ([@R9]). T/O ratios in recent studies have ranged from 0.055 (5.5%) in a sample of 26 patients to 0.07 (7.0%) in a sample of 410 patients ([@R9], [@R16]). Although one study showed a higher rate of pulmonary infection in patients with a higher T/O ratio ([@R14]), another found no differences in clinical outcomes ([@R16]). SENSITIVITY AND SPECIFICITY OF VALUES ===================================== Sensitivity and specificity of α-amylase values are limited. Filloux et al ([@R9]) reported a tracheal α-amylase cutoff value of 1,832 U/L had an 88.0% sensitivity and 100.0% specificity. Dewavrin et al ([@R11]) reported a similar sensitivity of 87.0% using a tracheal α-amylase value of 1,685 U/L, but the specificity was only 29.0%. Methods for determining sensitivity and specificity varied between the two studies, and Dewavrin et al ([@R11]) classified aspiration of oral secretions using pepsin biomarkers, which is likely the reason for the difference in specificity between the two studies ([@R9]). With the emergence of using α-amylase as a biomarker, researchers have begun to use a cut-point of 1,685 U/L or higher to classify aspiration. APPLICATION OF ALPHA-AMYLASE RESULTS ==================================== Incidence of Aspiration ----------------------- Alpha-amylase has been used to assess frequency of aspiration of oral secretions (**Table [2](#T2){ref-type="table"}**). Sample sizes and frequency of measures of α-amylase vary widely across studies. Studies used varying cutoff values of α-amylase to report the occurrence of aspiration, and many reports did not list values at all. With the exception of a pilot study, aspiration of oral secretions exceeded 75%, validating that the majority of intubated patients aspirate secretions around the cuff of the ETT. ###### Estimates of Aspiration  Duration of Alpha-Amylase in Secretions After Aspiration -------------------------------------------------------- Once oral secretions are aspirated, it is unknown how long α-amylase is detectable in pulmonary secretions. Studies with multiple measures detected α-amylase up to 14 days after intubation ([@R11], [@R12], [@R16]), and no changes in either the α-amylase values or T/O ratio over time ([@R16]). Reported values may reflect ongoing aspiration around the ETT cuff and/or stability of α-amylase in the pulmonary system once aspiration occurs. CLINICAL SIGNIFICANCE ===================== Alpha-Amylase Levels and Pulmonary Complications ------------------------------------------------ Several researchers assessed the relationship between α-amylase and development of pulmonary complications, including VAP and ventilator-associated events (VAEs). In some studies, α-amylase in tracheal or BAL specimens was significantly higher in those with VAP ([@R15], [@R18], [@R20]). An α-amylase of 4,681 U/L had a 90% sensitivity and 79% specificity for identifying VAP ([@R15]), whereas a value of 163 U/L in mini-BAL specimens had a reported sensitivity of 73.0% and specificity of 68.6% ([@R18]). Another study reported that an α-amylase value less than 125 U/L in BAL fluid reduced the odds for developing VAP after adjusting for gender, chest radiography, and preintubation risk factors (odds ratio, 0.39; 95% CI, 0.21--0.71; *p* = 0.002) ([@R20]). In contrast, other researchers used α-amylase to evaluate outcomes of interventions and reported data related to VAE. In these studies, no relationships were observed between α-amylase levels and development of VAE ([@R12], [@R16]). Aspiration Risk Factors and Alpha-Amylase ----------------------------------------- Having risk factors for aspiration prior to and after intubation was associated with higher α-amylase levels in tracheal and/or BAL specimens ([@R15], [@R17], [@R18], [@R20]). Risk factors include altered consciousness, swallowing dysfunction, cardiac arrest, preintubation vomiting, and/or difficulty with intubation ([@R20]). ETT Type Impact on Aspiration ----------------------------- Specialized ETTs with subglottic suction ports and different cuff configurations have been developed and tested. Although some devices have been shown to reduce VAP, findings vary when using α-amylase as an outcome measure ([@R12], [@R13], [@R19]). No differences were noted in the percentage of patients with aspiration between a tapered-cuff versus a standard-cuff ETT (90.1% vs 93.3%; *p* = 0.095) ([@R12]). Additionally, no differences were found in the proportion of tracheal specimens positive for α-amylase between those with a subglottic secretion drainage tube compared with a standard tracheal tube (88.0% vs 100.0%; *p* = 0.859) ([@R13]). It is not known if the lack of differences is attributed to the devices or if α-amylase was unable to detect differences. DISCUSSION ========== Several researchers have used α-amylase as a biomarker to assess for aspiration of oral secretions. As expected, values of α-amylase decreased depending on the proximity of specimen collection from the mouth. Approximately 5--7% of oral α-amylase is detectable in tracheal samples calculating a T/O ratio ([@R9], [@R16]). The majority of subjects across studies had aspiration as indicated by positive levels of α-amylase in pulmonary specimens. Aspiration was noted with different ETT cuff configurations and with subglottic suction tubes ([@R12], [@R22]). Challenges in Interpretation of Values -------------------------------------- Several challenges in interpretation of study findings exist including lack of standardization. Studies reported different cutoff points of α-amylase to identify aspiration. Alpha-amylase values that indicate microaspiration or macroaspiration are unknown. Some researchers classified patients as having aspirated if α-amylase was detected at a minimal level detectable by laboratory methods ([@R10], [@R16]--[@R18]). Other studies used a higher value based on sensitivity and specificity analysis, requiring values of 1,685 U/L or higher to indicate aspiration ([@R11], [@R13]). Some studies classified aspiration only if a percentage of the samples (e.g., 30% or more) were higher than a predetermined cutoff point ([@R12], [@R21]). Clinical Significance --------------------- The clinical significance of aspiration of oral secretions remains unknown. Some, but not all, researchers identified higher α-amylase levels in those with pulmonary complications, such as infection or VAP ([@R14]--[@R16]). Macroaspiration of secretions is a risk factor for aspiration pneumonia, secondary to the bacterial load that is introduced from either the oral cavity or upper gastrointestinal tract ([@R4]). Therefore, colonization of the oral secretions may be a more important factor in identifying complications than α-amylase levels. Studies will likely be more meaningful if aspiration is measured by both pepsin and α-amylase biomarkers. Since aspiration of gastric contents can result in an inflammatory reaction and chemical pneumonitis ([@R2]), assessment of pepsin in pulmonary secretions may be more important in assessing risk for complications. It is unknown how stable α-amylase is in pulmonary secretions as research has found no differences in values over time after intubation ([@R16]). Therefore, detection of α-amylase only indicates that aspiration of oral secretions occurred, but not when. Despite the recent increase in measurement of α-amylase in detecting aspiration, many questions remain unanswered: 1) what is the duration of detecting α-amylase in the lungs following aspiration; 2) what values which are considered "positive for aspiration;" 3) what values indicate microaspiration versus macroaspiration; and 4) what is the clinical significance of values? CONCLUSIONS =========== Although α-amylase is detected in most intubated patients, many unknowns exist about interpretation and usefulness of values. A lack of standardization in measurement and interpretation suggest that overall usefulness of α-amylase, even as a research outcome, has many limitations. This work was performed at the University of Central Florida College of Nursing, Orlando, FL. Supported, in part, by grant from the National Institutes of Health (NIH). The NIH grant number for the study is 1R01NR014508-01A1. The authors have disclosed that they do not have any potential conflicts of interest.

Introduction {#Sec1} ============ Vasomotor symptoms (63.5%), fatigue (50.7%) and palpitations (35.1%) are common symptoms in peri- and postmenopausal Chinese women^[@CR1]^. Hot flashes or night sweats, which are vasomotor symptoms, are regarded as thermoregulatory events that might be associated with estrogen deprivation or fluctuation^[@CR2]^ and may exert a negative impact on the quality of life of menopausal women^[@CR3]^. In addition, a growing body of evidence suggests a link between vasomotor symptoms and cardiovascular risks^[@CR4]--[@CR6]^; however, not all studies in the field agree^[@CR7]--[@CR9]^, possibly due to the use of different biomarkers, sample selection bias, and use of observational versus interventional studies. The pulse wave velocity is the wave velocity between two points of an arterial system; it is positively correlated with arteriosclerosis and is useful in clinical applications to evaluate arterial stiffness^[@CR10]^. The ability to measure brachial-ankle pulse wave velocity (baPWV) is commonly available in clinical practice; it is precise, not time-consuming, simple and non-invasive^[@CR11]^. Women's risk of cardiovascular disease may increase as they progress through menopause^[@CR12],[@CR13]^. Studies of the association between hot flashes/sweats and cardiovascular disease incidence reported conflicting results. It is not clear whether an association exists between hot flashes/sweats and baPWV in middle-aged women. Therefore, we aim to evaluate the impact of hot flashes/sweats on arterial stiffness by studying the correlation between hot flashes/sweats and baPWV in an effort to guide protocols for handling hot flashes/sweats and to aid in monitoring arterial stiffness in peri- and postmenopausal women. Results {#Sec2} ======= Participants {#Sec3} ------------ A total of 589 subjects were enrolled in this study; the characteristics of the study population are listed in Table [1](#Tab1){ref-type="table"}. The mean age (SD) of participants was 50.5(5.7). The mean height, weight, BMI, systolic pressure, diastolic pressure and heart rate all followed normal distributions, the incidence of hot flashes/sweats was 41.77%.Table 1Characteristics of the study population.CharacteristicMenopausal StatusTotalPre-menopausePeri-menopauseEarly Post-menopauseLate Post-menopausen = 589n = 194n = 114n = 140n = 141Age(year)50.50 ± 5.7045.07 ± 3.2849.14 ± 3.6052.95 ± 3.4156.62 ± 3.66Height (cm)159.60 ± 5.03160.44 ± 4.35159.11 ± 4.85159.67 ± 4.75158.78 ± 6.07Weight(kg)59.56 ± 7.9959.21 ± 7.6459.46 ± 7.6060.72 ± 8.1158.97 ± 8.59BMI(kg/m^2^)23.38 ± 3.0123.00 ± 2.8323.50 ± 3.0123.81 ± 3.0223.38 ± 3.20SBP(mmHg)125.27 ± 17.87119.93 ± 16.18124.19 ± 19.11127.05 ± 18.05131.72 ± 16.65DBP(mmHg)76.38 ± 11.5473.06 ± 10.0775.64 ± 12.4178.26 ± 10.0479.69 ± 10.90Heart rate(bpm)69.81 ± 10.2870.29 ± 9.7768.33 ± 9.6969.54 ± 10.0770.60 ± 11.51Age groups, n(%) 40--44131(22.2)107(55.2)19(16.7)3(2.1)2(1.4) 45--49131(22.2)71(36.6)37(32.5)18(12.9)5(3.5) 50--54181(30.7)16(8.2)56(49.1)80(57.1)29(20.6) 55--60146(24.8)0(0.0)2(1.8)39(27.9)105(74.5)Marital status, n(%) Married555(94.2)181(93.3)110(96.5)133(95.0)131(92.9) Divorced20(3.4)8(4.1)2(1.8)6(4.3)4(2.8) Separated, Widowed10(1.7)2(1.0)2(1.8)1(0.7)5(3.5) Unmarried4(0.7)3(1.5)0(0.0)0(0.0)1(0.7)Employment status, n(%) Yes248(42.1)125(64.4)62(54.4)44(31.4)17(12.1) No341(57.9)69(35.6)52(45.6)96(68.6)124(87.9)Education, n(%) None52(8.8)12(6.2)9(7.9)11(7.9)20(14.2) Primary75(12.7)24(12.4)15(13.2)21(15.0)15(10.6) Junior high148(25.1)45(23.2)40(35.1)30(21.4)33(23.4) Senior high172(29.3)51(26.3)28(24.6)46(32.9)47(33.3) College130(22.1)58(29.9)20(17.5)29(20.7)23(16.3) Postgraduate12(2.0)4(2.1)2(1.8)3(2.1)3(2.1)Gynecological diseases, n(%) Uterine fibroid212(36.0)56(28.9)46(40.4)60(42.9)50(35.5) Endometriosis9(1.5)3(1.5)1(0.9)4(2.9)1(0.7)Chronic disease, n(%) Hypertension105(17.8)23(11.9)14(12.3)30(21.4)38(27.0) Diabetes42(7.1)9(4.6)4(3.5)17(12.1)12(8.5) Hotflashes/sweats(yes), n(%)41(21.1)57(50)61(56.4)4(48.9)246(41.4) BaPWV data {#Sec4} ---------- According to one-way ANOVA, there was a statistically significant difference in baPWV values across age groups (p \< 0.001), menopausal status (p \< 0.001), and severity of hot flashes/sweats (p \< 0.001). The baPWV of each group was significantly different. (p \< 0.001) (Table [2](#Tab2){ref-type="table"}).Table 2Comparison of baPWV among different groups by age, menopausal status and degree of severity of hot flashes/sweats.characteristicsnmean ± SD95% CIlowerupperAge 40--441131193.32 ± 169.72\*1161.691224.96Age 45--491411241.54 ± 167.50\*1213.691269.43Age 50--541831381.92 ± 236.69\*1347.401416.44Age 55--601521462.64 ± 272.34\*1419.001506.29Premenopause1941226.89 ± 185.55\*1200.611253.16Perimenopause1141292.51 ± 188.83\*1257.231327.91Early Postmenopause1401389.36 ± 233.12\*1350.401428.31Late Postmenopause1411455.62 ± 289.89\*1407.351503.89Hot flashes/sweats 0 points3431287.09 ± 227.30\*1262.951311.23Hot flashes/sweats 1 point1781376.07 ± 229.59\*1342.111410.03Hot flashes/sweats 2 points551433.86 ± 289.78\*1355.531512.20Hot flashes/sweats 3 points131526.27 ± 243.72\*1327.841724.70\*Means p \<0.001 CI: Confidence Interval. Pearson's correlation analyses revealed that systolic blood pressure, diastolic blood pressure, age, heart rate, triglyceride (TG), body mass index (BMI), and low density lipoprotein (LDL) levels were positively correlated with baPWV (p \< 0.05). Height and HDL levels were significantly negatively correlated with baPWV (Table [3](#Tab3){ref-type="table"}). Spearman's correlation analyses revealed that hot flashes/sweats (r = 0.243, p \< 0.001), menopausal status (r = 0.367, p \< 0.001), and the KMI total score (r = 0.237, p \< 0.001) were significantly positively correlated with baPWV menopausal status and hot flashes/sweats (Table [3](#Tab3){ref-type="table"}).Table 3Correlation analysis of baPWV with some characteristics.Itemsrp valueAge(year)0.439\*\*\<0.001HDL(mmol/L)−0.093\*0.024LDL(mmol/L)0.136\*\*0.001TG(mmol/L)0.200\*\*\<0.001TC(mmol/L)0.139\*\*0.001Height(cm)−0.114\*\*0.006Weight(kg)0.0810.050BMI(kg/m^2^)0.140\*\*0.001Systolic pressure(mmHg)0.594\*\*\<0.001Diastolic pressure(mmHg)0.548\*\*\<0.001Heart rate(bpm)0.302\*\*\<0.001Menopausal status0.367\*\*\<0.001Flashes/sweats(points)0.243\*\*\<0.001KMI total score(points)0.237\*\*\<0.001\*\*Means p \<0.001; \*Means p \<0.05. HDL, high-density lipoprotein; LDL, Low density lipoprotein; TC, total cholesterol; TG, Triglyceride; BMI, body mass index. Hot flash and KMI total score were analyzed by Spearman's correlation, others were computed by Pearson's correlation. As shown in Table [4](#Tab4){ref-type="table"}, we observed a significant positive association between baPWV and hot flashes/sweats (p = 0.010). BaPWV was also significantly associated with SBP, age, hypertension, heart rate, and diabetes mellitus in the full multiple linear regression model following corrections for age, BMI, HDL, LDL, TC, TG, DBP, height, and menopausal status.Table 4Linear regression analysis of baPWV with hot flashes/sweats in an adjusted model.TraitsUnstandardized coefficients95% CIp valueCorrected R²BStd ErrorSBP(mmHg)3.760.64(2.5, 5.025)\<0.0010.54Age(year)10.671.31(8.11, 12.24)\<0.001Hypertension(yes)−147.5320.21(187.21, 107.84)\<0.001Heart rate (bpm)3.680.70(2.31, 5.05)\<0.001Diabetes mellitus(yes)−98.5827.04(−151.69, −45.46)\<0.001Flashes/sweats24.549.52(5.86, 43.23)0.010Covariates: BMI, HDL, LDL, TC, TG, DBP, height, menopausal status. Discussions {#Sec5} =========== Based on 589 women's modified KMI scores, the prevalence of hot flashes/sweats was found to be 41.77%, which is similar to previous findings from other Asian countries, such as Japan, Hong Kong, Singapore^[@CR14]^ and South Korea^[@CR15]^, and was significantly lower than that found in white populations^[@CR16],[@CR17]^. The discrepancy may be explained by the racial and cultural context^[@CR18]^, demographic and socioeconomic characteristics^[@CR19]^, and methods of symptom identification^[@CR20]^. This study revealed that the pulse wave velocity is significantly associated with aging, menopausal status and duration of menopause; this finding parallels the relationship between the decrease in estrogen levels in menopausal women and vascular aging. As a result, baPWV is considered to be a possibly effective measure to evaluate arterial stiffness in middle-aged women. Our study found that the pulse wave velocity was positively correlated with the frequency of hot flashes/sweats and the severity of symptoms (r = 0.243, p \< 0.001). After adjusting for established cardiovascular risk factors, such as systolic blood pressure, diastolic blood pressure, age, menopause, heart rate, TG, BMI, TC, LDL, HDL, and hot flashes/sweats (95% CI: 5.86--43.23, p = 0.01), the results remain significant using a linear regression analysis. Therefore, we can safely infer that the assessment of baPWV is a valuable tool to evaluate symptoms of hot flashes/sweats, the method provides an objective standard by which to assess symptoms of hot flashes in menopausal women. It is known that SBP, DBP, age, menopause, BMI, TG, TC and LDL are risk factors for cardiovascular diseases^[@CR21]^. Our analyses have shown that SBP, DBP, age, heart rate, TG, BMI, TC, LDL were significantly positively correlated with baPWV, which is consistent with previous studies^[@CR22]^. After adjusting for these items in addition to a history of hypertension and diabetes mellitus, the independent risk factors for higher arterial stiffness were found to be age, systolic blood pressure, history of hypertension and diabetes mellitus, heart rate and hot flashes/sweats. We can therefore extrapolate self-reporting hot flashes/sweats has clinical implications for predicting arterial stiffness in menopause beyond other CVD risk factors. Additionally, we should pay more attention to hot flashes/sweats. We suggest that people who experience severe hot flashes/sweats should focus on atherosclerosis and baPWV, and health staff should prioritize the treatment of people with hot flashes/sweats by, for example, providing menopausal hormone therapy. To the best of our knowledge, this may be the first study to suggest that hot flashes/sweats and their severity are associated with higher baPWV. We suggest that baPWV may serve as a metric to monitor arterial stiffness for middle-aged women. Additionally, baPWV can likely be regarded as an objective index for evaluating the severity of hot flashes/sweats; furthermore, self-reporting hot flashes/sweats is of prominent value to assess arterial stiffness independent of CVD risk factors in middle-aged women. however, it is worthy of further exploration and research. Because of the limited survey samples, more community-based studies on large groups of people are important. Our team will continue to conduct future relevant research to confirm the validity of these results. We are now conducting a cohort study of measuring baPWV before and after MHT to explore changes in pulse wave velocity after MHT in middle-aged women. Limitations {#Sec6} ----------- Several limitations need to be mentioned. First, the inherent drawback of an observational survey may attenuate the causal relationship. Secondly, systemic errors due to the baPWV tool may be produced, for its calculation of path length comes from a height-based formula for Japanese population. However, the height of Chinese is similar to that of Japanese. Finally, hot flashes/sweat ascertained by questionnaire would produce memory bias. Therefore, further longitudinal study is needed to confirm these relationships. Our team is now working on the following-up investigation. Conclusions {#Sec7} =========== BaPWV may serve as a metric to monitor arterial stiffness in middle-aged womenBaPWV may provide an objective standard to evaluate the severity of vasomotor symptoms.Self-reporting hot flashes/sweats may be regarded as an easy and quick predictor of arterial stiffness independent of CVD risk factors in middle-aged women. Methods {#Sec8} ======= Study Subjects {#Sec9} -------------- Women who received a physical examination at the Center of Health Examination of Shanghai Jiao Tong University Affiliated Sixth People's Hospital from the 28st of November 2016 to the 28th of September 2017 were enrolled in this study. The inclusion criteria included the following: 1) aged between 40 and 60 years; 2) normal cognition, has the ability to complete the questionnaire by herself; and 3) volunteered to participate in this research. The exclusion criteria included the following: 1) female who received menopausal hormone treatment (MHT) or any traditional Chinese medicine indicated for menopause in the past 6 months; 2) history of mental disorder; and 3) history of serious organic diseases (e.g., coronary heart disease, stroke, systemic autoimmune disease). A total of 589 women met to the criteria. The reproductive medicine center of the Shanghai Sixth People's Hospital institutional review board approved this study, This analytic cross-sectional study enrolled 1904 participants aged 40--60 years who visited the physical examination center in the Shanghai Sixth people's Hospital, Shanghai Jiao Tong University School of Medicine, China, from January 2016 to November 2016. The study protocol was approved by the Ethics Committee of Shanghai Sixth People's Hospital, and the study was performed in accordance with the approved guidelines. All the participants provided written informed consents after full explanation of the study. Questionnaire {#Sec10} ------------- The women's health demographic questionnaire included basic information, such as age, education, marital status, occupation, history of MHT and traditional Chinese medicine, menopausal states, menstrual states, gynecological history, and past history. This questionnaire has been applied previously^[@CR23],[@CR24]^. Menopausal status group {#Sec11} ----------------------- According to the stages of reproductive aging workshop (STRAW + 10)^[@CR25]^, we divided subjects into the following groups: premenopausal (with regular menstrual cycle), perimenopausal (consecutive irregularities \>7 days from their normal cycle), early postmenopausal (absence of menstrual periods for at least 12 months and less than 5 years) and late postmenopausal (absence of menstrual periods for more than 5 years). Menopausal symptoms and hot flashes/sweats group {#Sec12} ------------------------------------------------ We used a valid modified KMI scale in Chinese to evaluate the severity of menopausal symptoms^[@CR26]^. The modified KMI scale consists of three aspects: 1) somatic symptoms, such as hot flashes/sweats, palpitation, vertigo, headache, paresthesia, formication, arthralgia and myalgia; 2) mental symptoms, such as fatigue, nervousness and melancholia; and 3) genitourinary tract symptoms, such as urinary infections and sexual complaints. The sum of the scores from 0 to 63 points were categorized into four grades:score ≤6, asymptomatic (148 subjects)6\< score ≤15, mildly symptomatic (259 subjects)15\< score ≤30, moderately symptomatic (159 subjects)score \>30, severely symptomatic (23 subjects) According to the modified KMI subproject, we divided the hot flashes/sweats group into four groups:10 point, no hot flashes/sweats (342 subjects)1 point, hot flashes and sweats ≤3 times/day (178 subjects)2 points, 3\< hot flashes and sweats ≤9 times/day (55 subjects)3 points, hot flashes and sweats \>9 times/day (14 subjects). Measurement of PWV {#Sec13} ------------------ We used an automatic waveform analyzer (BP-203RPE III, OMRON, Japan) to measure baPWV. Participants were asked to remain supine and at rest for 5 minutes before the PWV examination^[@CR27]^. We record the heart rate concurrently. Laboratory examination {#Sec14} ---------------------- All participants underwent a fasting lipid profile consisting of total cholesterol (TC), triglycerides triglyceride (TG), high-density lipoprotein (HDL)and low density lipoprotein (LDL). Height, weight and systolic blood pressure (SBP), diastolic blood pressure (DBP) were determined on the same day. Body mass index (BMI) was computed by dividing weight in kilograms by the square of their height in meters. Statistical analysis {#Sec15} -------------------- SPSS Statistics 23.0 (IBM Corporation, Armonk, NY, USA) was used for all analyses. All variables were presented as the mean ± standard deviation (SD) or number (%). One-way ANOVAs were used to analyze the variation in PWV by menopausal status and degree of hot flashes/sweats. Pearson correlation analysis was performed to evaluate the relationship between age, HDL, LDL, TC, TG, height, weight, BMI, SBP, DBP, heart rate and baPWV. Spearman correlation analysis was performed to evaluate the relationship between menopausal status, hot flashes/sweats and baPWV. Associations with the pulse wave velocity and hot flashes/sweats were computed with a linear regression analysis. R² values were derived from linear regression models. Residuals analysis was performed and diagnostic plots were made to verify model assumptions. A two-sided P-value \< 0.05 was considered statistically significant. Ruwei Yang and Yang Zhou contributed equally to this work. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors work the study participants and the research associates who made it possible to complete this research project. This study was supported by grants from the Shanghai Science and Technology Committee (154119050202) and Shanghai Health Development Planning Commission (GWIV15). Minfang Tao conceived and designed the study. Ruwei Yang wrote the first draft. Yang Zhou revised the manuscripts. Changbin Li input and managed the data. All of the authors read and approved the final manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec005} ============ Glaucoma is an optic neuropathy in which visual disturbance corresponds to optic disc cupping and optic nerve fiber degeneration \[[@pone.0190012.ref001]\]. Lowering intraocular pressure (IOP) is an effective, evidence-based treatment for open-angle glaucoma (OAG) \[[@pone.0190012.ref002],[@pone.0190012.ref003]\], but meta-analysis has shown that non-IOP risk factors also contribute to progression \[[@pone.0190012.ref004]\], and glaucoma is now regarded as multifactorial \[[@pone.0190012.ref005]\]. Therefore, ophthalmologists must consider IOP-independent factors and varying pathophysiologies in glaucoma patients, and adjust treatments strategies accordingly, to most effectively preserve quality of life. Nicolela et al. described characteristic inter-individual variations in optic disc morphology, and classified the glaucomatous disc into 4 types \[[@pone.0190012.ref006]\]. This revealed disc-dependent variations in age, rate of spasm, arteriosclerosis, and myopia in patients. Follow-up investigations showed that disc type classification is a useful addition to the management of OAG \[[@pone.0190012.ref007]--[@pone.0190012.ref010]\]. However, classification of the optic disc can sometimes be difficult because it relies on subjective assessment methods. Thus, we previously attempted to develop new, objective, and more accurate methods of classifying the optic disc, based on stereophotography \[[@pone.0190012.ref011]\] and optical coherence tomography (OCT) \[[@pone.0190012.ref012]\]. Recently, machine learning technology has seen dramatic progress, and has enabled the development of new algorithms to diagnose age-related macular disease and glaucoma \[[@pone.0190012.ref013]--[@pone.0190012.ref015]\]. Thus, in this report, we used a variety of parameters to set up a machine-learning-based system for objective optic disc classification. We then investigated the accuracy of this method in training and testing groups randomly selected from among OAG patients at our clinic. Our results suggest that our method is highly reproducible, and that it might contribute not only to daily glaucoma care, but also to ophthalmological research by enabling big-data analysis in clinical trials of new disc-type-specific treatments for glaucoma. Material and methods {#sec006} ==================== Subjects {#sec007} -------- This study included 163 eyes of 105 OAG patients with a glaucomatous visual field meeting the Anderson-Patella classification criteria \[[@pone.0190012.ref016]\]. All patients underwent testing with the Humphrey field analyzer (HFA, SITA standard 24--2, Carl Zeiss Meditec) with only reliable and repeatable results being included. Patients were excluded if they had a spherical equivalent (SE) refractive error of \< -8.00 diopters, ocular disease other than OAG, systemic disease affecting the visual field, or cataract progression. This study adhered to the tenets of the Declaration of Helsinki, and the protocols were approved by the Clinical Research Ethics Committee of the Tohoku University Graduate School of Medicine (study 2014-1-805). Participants provided their written informed consent to participate in this study. The ethics committees approved this consent procedure. In this study, minors were not included. The patients were classified into Nicolela's 4 types: focal ischemic (FI), generalized enlargement (GE), myopic (MY), and senile sclerotic (SS) with a method we have previously described \[[@pone.0190012.ref011],[@pone.0190012.ref012]\]. The distinctive characteristics of each disc type include rim notching in FI, a diffusely enlarged, rounded cup in GE, a tilted disc and temporal crescent peripapillary atrophy (PPA) in MY, and shallow cupping and haloing in SS. Three glaucoma specialists performed the classification (TN, KO, and ST). Cases were excluded when classification was not consistent between all three graders. Demographic data are listed in [Table 1](#pone.0190012.t001){ref-type="table"} and assignment data are listed in [Table 2](#pone.0190012.t002){ref-type="table"}. There were no significant differences in background between training and test data (t-test). 10.1371/journal.pone.0190012.t001 ###### Demographic data. {#pone.0190012.t001g} All FI[^a^](#t001fn001){ref-type="table-fn"} GE[^b^](#t001fn002){ref-type="table-fn"} MY[^c^](#t001fn003){ref-type="table-fn"} SS[^d^](#t001fn004){ref-type="table-fn"} ------------------------------------------------------------------------------------------------ ------------- ------------------------------------------ ------------------------------------------ ------------------------------------------ ------------------------------------------ **Male / Female** 85 / 78 9 / 17 31 / 19 27 / 28 18 / 14 **Age (years)** 62.3 ± 12.6 63.0 ± 11.9 65.7 ± 10.0 53.6 ± 11.6 71.5 ± 9.1 **MD**[^e^](#t001fn005){ref-type="table-fn"} **(dB)** -8.9 ± 7.5 -5.2 ± 5.8 -12.9 ± 7.7 -7.8 ± 7.1 -7.8 ± 6.0 **SE**[^f^](#t001fn006){ref-type="table-fn"} **(D**[^g^](#t001fn007){ref-type="table-fn"}**)** -2.5 ± 2.9 -1.2 ± 1.7 -0.5 ± 1.5 -5.6 ± 2.2 -1.5 ± 2.1 **IOP**[^h^](#t001fn008){ref-type="table-fn"} **(mmHg)** 13.3 ± 3.5 13.6 ± 3.4 13.3 ± 2.9 13.4 ± 2.9 13.3 ± 2.9 ^a^FI: focal ischemic, ^b^GE: generalized enlargement, ^c^MY: myopic, ^d^SS: senile sclerotic, ^e^MD: Humphrey-field analyzer (HFA)-measured mean deviation, ^f^SE: spherical equivalent, ^g^D: diopters, ^h^IOP: intraocular pressure. Data are presented as the mean ± standard deviation. 10.1371/journal.pone.0190012.t002 ###### Assignment data. {#pone.0190012.t002g} Training data Test data *P* value ------------------------------------------------------------------------------------------------ --------------- ------------- ----------- **Male / Female** 58 / 56 27 / 22 1.000 **Age (years)** 62.8 ± 12.5 61.1 ± 13.0 0.432 **MD**[^e^](#t002fn005){ref-type="table-fn"} **(dB)** -9.6 ± 7.6 -7.4 ± 7.0 0.085 **SE**[^f^](#t002fn006){ref-type="table-fn"} **(D**[^g^](#t002fn007){ref-type="table-fn"}**)** -2.6 ± 2.9 -2.4 ± 3.1 0.693 **IOP**[^h^](#t002fn008){ref-type="table-fn"} **(mmHg)** 13.3 ± 2.6 13.1 ± 2.4 0.646 ^a^FI: focal ischemic, ^b^GE: generalized enlargement, ^c^MY: myopic, ^d^SS: senile sclerotic, ^e^MD: Humphrey-field analyzer (HFA)-measured mean deviation, ^f^SE: spherical equivalent, ^g^D: diopters, ^h^IOP: intraocular pressure. Data are presented as the mean ± standard deviation. Differences were considered significant at p \< 0.05. Measurement of clinical variables {#sec008} --------------------------------- We obtained biographical data for the patients (including sex, age, family history, and medical history) from medical records. All 91 types of data used in this study are listed in [Fig 1](#pone.0190012.g001){ref-type="fig"}. {#pone.0190012.g001} Best-corrected visual acuity (BCVA) was measured with the 5-meter Landolt's international ring-type chart, and was calculated as the logarithmic minimum angle of resolution (logMAR) with decimal values. Following a slit lamp examination and gonioscopy, IOP was measured with Goldman applanation tonometry. Central corneal thickness (CCT) was measured with anterior-segment OCT (CASIA, Tomey Corp.). Following pupil dilation with 0.4% tropicamide (Mydrin M; Santen Pharmaceutical), OCT parameters, including 22 parameters related to disc topography and 26 parameters related to cpRNFLT, were measured with SS-OCT (DRI OCT Atlantis, Topcon Corp.). CpRNFLT was calculated in the quadrants, 6 radial sectors, and the clockwise sectors. These local measurements were used to quantify asymmetries in cpRNFLT. Mean blood pressure (MBP) and ocular perfusion pressure (OPP) were calculated as follows: MBP = diastolic BP + 1/3 (systolic BP---diastolic BP); OPP = 2/3 MBP---IOP. To assess blood flow (BF) in the ONH, the laser speckle flowgraphy (LSFG-NAVI device, Softcare Co., Ltd., Fukutsu, Japan) was used. Mean blur rate (MBR), an LSFG variable that represents a relative BF index, is expressed in arbitrary units. The accompanying analysis software then automatically divided the region of interest (ROI) into large-vessel and tissue (i.e., capillary) areas and determined specific MBR values in each area (vessel-area MBR: MV; tissue-area MBR: MT). A total of 36 BF waveform parameters were also measured. All data were obtained within a 3-month period. Machine learning {#sec009} ---------------- To build a machine learning model, we used 91 types of quantitative data from 7 aspects of patient background and 84 types of data, included 22 parameters of optic disc topography, as well as 26 measurement parameters related to cpRNFLT, and 36 LSFG BF parameters ([Fig 1](#pone.0190012.g001){ref-type="fig"}). We used a neural network (NN) as the machine-learning classifier, with a structure of 9 input layer units, 8 hidden layer units, and 4 output layer units. After standardization to the training data, we used minimum redundancy maximum relevance to quickly limit the candidate features to 15, and then selected the characteristics with the best classification performance with a genetic algorithm. We used Cohen\'s Kappa of 10-fold cross validation (CV) for an evaluation index of the hereditary classification performance. Results {#sec010} ======= Demographic data of this study was listed in the [Table 1](#pone.0190012.t001){ref-type="table"} and assigned data was listed in [Table 2](#pone.0190012.t002){ref-type="table"}. The accuracy and Cohen's Kappa 10-fold CV were 91.2% and 0.88 for the NN. The nine most important discriminative characteristics selected by the NN were spherical equivalent, age, average rim disc ratio (nasal), average cup depth, horizontal disc angle, 6-sector superior-temporal cpRNFLT, superior-quadrant cpRNFLT, maximum cup depth, and cup area ([Fig 2](#pone.0190012.g002){ref-type="fig"}). {#pone.0190012.g002} Disc-type classification by the NN matched the test data at rates of 66.7% for FI, 93.3% for GE, 83.3% for MY, and 100.0% for SS, and recall at rates of 75.0% for FI, 93.3% for GE, 93.8% for MY, and 80.0% for SS. Overall, the NN had the high validated accuracy against the test data, at 87.8% (Cohen's Kappa = 0.83). [Fig 3](#pone.0190012.g003){ref-type="fig"} shows the confidence level for the classification of each of Nicolela's types, as performed by the NN. [Fig 3a and 3b](#pone.0190012.g003){ref-type="fig"} show cases with accurate classifications and a high confidence level. [Fig 3c](#pone.0190012.g003){ref-type="fig"} shows a case where two potential disc types was assessed by our NN system. {#pone.0190012.g003} Discussion {#sec011} ========== This report describes a new machine-learning algorithm to classify optic disc topography, based on 91 quantitative data types derived from patient background, OCT-measured structure, and blood flow. The NN had the high accuracy, with a match rate to the validated data of 87.8%. Previously, we reported \[[@pone.0190012.ref012]\] an identification rate of 80.0% and a Cohen\'s Kappa of 0.73. Thus, the algorithms described in this study promise to help doctors to classify the disc type of patients into Nicolela's 4 types and to provide useful information for glaucoma care. Accurate functional classification of glaucoma is an important way of identifying risk factors for progression, which can occur even with successful IOP-lowering treatment. Reported IOP-independent risk factors for NTG progression include disc hemorrhage \[[@pone.0190012.ref004],[@pone.0190012.ref017]\], arteriosclerosis \[[@pone.0190012.ref004],[@pone.0190012.ref018]\], migraine \[[@pone.0190012.ref019]\], blood flow abnormalities in the optic nerve (derived from increased resistance in the retrobulbar vessels) \[[@pone.0190012.ref020]\], Flammer syndrome \[[@pone.0190012.ref005],[@pone.0190012.ref021]\], hypotension \[[@pone.0190012.ref022],[@pone.0190012.ref023]\], and night dip \[[@pone.0190012.ref024]\]. Additionally, Nicolela's 4 types of disc morphology have each been linked to various factors, such as spasm, myopia, IOP, and high or low retrobulbar blood flow. Previously, we also found that the lamina cribrosa was thinner in GE discs, and that lamina cribrosa thickness was independently associated with cup size and tissue blood flow in the optic nerve head \[[@pone.0190012.ref025]\]. All these findings point to the importance of Nicolela's disc types for the functional classification of glaucoma. In addition to identifying at-risk patients, classifying glaucoma into progressive subtypes is very important for analyzing large patient data sets. Our algorithm is a first step towards the establishment of machine-learning methods for the functional classification of glaucoma, and in the future, we hope that such methods will open new directions for research and improve treatment outcomes, not only for established IOP-lowering approaches, but also for newly reported IOP-independent strategies, such as those targeting blood flow and oxidative stress. It is widely believed that glaucoma diagnosis would be subject to less variation if the relationship between structure and function were better understood. Supervised learning approaches have been most frequently used to discriminate between glaucomatous and non-glaucomatous eyes, and most published studies in the field of glaucoma research have used supervised machine learning techniques to improve diagnoses \[[@pone.0190012.ref014]\]. Belghith et al. showed that the AUC to differentiate glaucoma was 0.91 for Bayes, 0.69 for an artificial NN, and 0.6 for a support vector machine (SVM) \[[@pone.0190012.ref013]\]. Mookiah et al. compared different methods of machine learning and found that a simple linear SVM was superior to decision tree, nearest neighbor, naïve Bayes, and probabilistic neural network (PNN) analyses for the diagnosis of age-related macular degeneration \[[@pone.0190012.ref015]\]. Torok et al. investigated six machine-learning algorithms, including an SVM, recursive partitioning, random forest, Naïve Bayes, logistic regression, and K-nearest neighbor (k-NN), and found that the results of recursive partitioning could most accurately screen for diabetic retinopathy \[[@pone.0190012.ref026]\]. Therefore, the accuracy of machine-learning algorithms varies with different conditions. Here, we found that 9 clinical and morphological characteristics were important components for automatically classifying optic disc type, including spherical equivalent, age, average rim disc ratio (nasal), average cup depth, horizontal disc angle, 6-sector superior-temporal cpRNFLT, superior-quadrant cpRNFLT, maximum cup depth and cup area. Our previous research \[[@pone.0190012.ref012]\] showed that six parameters were most significant for disc type discrimination: disc angle (horizontal), average cup depth, cup/disc ratio, rim-decentering ratio, average rim/disc ratio (upper and lower nasal). These previous findings are closely consistent with the present study. Generally, eyes with the MY disc type have a low spherical equivalent, and the onset of glaucoma occurs at a younger age. Moreover, MY discs are tilted temporally, resulting in a high horizontal disc angle, and have a high nasal cup to disc area ratio. On the other hand, GE discs generally have a thin nasal rim and a large average and maximum cup depth and cup area, while SS discs have shallow cupping and are associated with the onset of glaucoma at an older age. FI discs showed thickening of the cpRNFLT in the 6-sector superior-temporal and superior quadrants. These characteristics of different disc types likely underlie the similarities between the most important parameters for discriminating disc type in each machine-learning model. Furthermore, we calculated the contribution of these characteristics to Nicolela's types ([Fig 2](#pone.0190012.g002){ref-type="fig"}), and found that in general, horizontal disc angle was the most useful for automatic classification. All these characteristics will be recognizable to experienced clinicians, and are likely similar to their own sense of the most important contributors to disc classification. Interestingly, however, the NN in this study selected two quantified parameters from patient background information and seven from purely digital, OCT data, without any parameters derived from LSFG. This may be because the definitions of Nicolela's 4 disc types do not include any blood flow parameters. Thus, future research into functional OAG classification based on LSFG parameters may be necessary. Nevertheless, this study showed that it may be possible to use OCT data to classify the disc into Nicolela's 4 types \[[@pone.0190012.ref006]\], without reference to blood flow parameters. The NN model allowed calculation of the confidence level of the prediction ([Fig 3](#pone.0190012.g003){ref-type="fig"}). This showed that many individual predictions had high confidence levels ([Fig 3a and 3b](#pone.0190012.g003){ref-type="fig"}), and that the overall accuracy of our method was 87.8%. We returned to our data to reconsider why the NN misidentified certain discs, and found that in most cases, the correct disc type had been the second choice. When we recalculated the accuracy of the model to include the first and second choices, the accuracy increased to 95.9%. This result may reflect the real-life experiences of glaucoma specialists, who are sometimes unsure which disc type is correct, and classify discs as having a mixed type, such as mixed FI and MY discs ([Fig 3c](#pone.0190012.g003){ref-type="fig"}). Thus, the presence of mixed disc type influences the accuracy of disc type classification with our machine-learning system. Nevertheless, each disc type had a good confidence level, including more than 80% for the pure MY disc type, which should help future efforts to find accurate classification methods and help improve daily clinical glaucoma care. Limitations of this study included a cross-sectional design, the inclusion of only Japanese subjects, and a relatively small study population, which may have affected the statistical power of our analyses. However, the entire analysis was performed 3 times, with randomized selection for the training and validation groups, which supports the accuracy of our classification system. An additional limitation was the unavoidable inclusion of factors that might have influenced the OCT and LSFG measurements, such as myopia. In Asia, the combination of myopia and glaucoma is very common. Nevertheless, we excluded cases with high myopia and adjusted the number of cases with glaucomatous myopic disc to minimize possible bias. In conclusion, this study used various parameters derived from OCT and LSFG and from biographical and clinical data to set up an automated, objective, machine-learning based method for the classification of Nicolela's 4 disc types, with a success rate of 87.8%. This technology has the potential to standardize diagnostic criteria for optic disc classification in future clinical trials, thereby reducing possible bias. However, in daily practice, this technology is not superior to the judgements of experienced glaucoma clinicians. Thus, the techniques described here have the potential to be very useful, powerful tools for supporting clinical decision making. Our findings should help improve the classification of OAG and lead to better glaucoma management, by demonstrating the confidence level of predictions for each disc type. Supporting information {#sec012} ====================== ###### Demographic data and quantitative ocular parameters from ophthalmic examination instruments. F1 File contains 91 types of quantitative data from 7 aspects of patient background and 84 types of data, included 22 parameters of optic disc topography, as well as 26 measurement parameters related to circumpapillary retinal nerve fiber layer thickness from OCT, and 36 blood flow parameters from laser speckle flowgraphy. (CSV) ###### Click here for additional data file. The authors thank Mr. Tim Hilts for editing this manuscript and Ms. Seri Takahashi for technical assistance. [^1]: **Competing Interests:**Co-authors GA, TK, and MA are employed by Topcon Corporation, a commercial company. There are no patents, products in development or marketed products to declare. This does not alter our adherence to all the PLOS ONE policies on sharing data and materials.

The power conversion efficiency (PCE) of single-junction organic solar cells has increased significantly during the last decade to 9--10 %, now approaching the threshold considered necessary to commercialize the technology.^1^ During this period, the structural diversity of semiconducting donor polymers for solar cells has increased dramatically.^2^ These materials have helped enable an accelerated development of bulk heterojunction (BHJ) organic solar cells based on polymer donor materials and molecular fullerene derivatives. However, the development of electron-accepting materials that lead to BHJs with high PCE has been significantly slower.^3^ The most commonly used *n*-type acceptors to date remain \[6,6\]-phenyl-C~61~-butyric acid methyl ester (PC~61~BM) and its slightly larger counterpart PC~71~BM. The morphology of polymer:PCBM BHJs is crucial for optimal PCE; it is important to achieve the right trade-off between domain size and finely interconnected network throughout the bulk of the active layer that allows efficient charge generation and extraction. The morphology can be controlled during device fabrication by employing solvent mixtures to deposit the active layer, adding high boiling solvent additives or alternatively by post-deposition treatments like solvent vapor or thermal annealing.^4^ However once the active layer is deposited and the desired morphology achieved, it is essential to "lock" that morphology in order to maintain high power efficiencies and to avoid device degradation over time. The "locking" of the active layer morphology is extremely non-trivial, due in part to the fullerene derivatives miscibility in the polymer phase, where they readily diffuse throughout the polymer on short timescales.^5^ Over an extended period of time or under thermal stress, the fullerene derivatives, especially PC~61~BM, form micron-scale crystallites within the polymer phase acting as charge traps and inhibiting efficient charge transport to the electrodes.^6^ One of the most common approaches to avoid aggregation of fullerenes into large domains is to chemically crosslink either or even both the materials in the BHJ after deposition.^7^ Crosslinking, usually triggered by external stimuli (e.g. heat or light), has been successfully used in lithography for decades, but has proven rather difficult to apply to organic semiconductor thin films. In most cases crosslinking the active layer inadvertently causes a significant performance decrease. We have recently demonstrated a new approach upon which the active layer morphology can be stabilized by the photoinduced \[2+2\] cycloaddition of the PC~61~BM derivatives, which results in covalent bonding between the fullerene groups.[@b8] The strategy was found to hold generally for a variety of polymer--fullerene pairs.[@b6c], ^9^ Light soaking significantly reduced the formation of PC~61~BM crystallites under thermal stress, and thus the device lifetime subsequently improved. Mechanistic studies indicated the enhancement in device thermal stability is primarily associated with inhibition of the nucleation of nanoscale PC~61~BM aggregates.[@b6c] These observations have motivated us to consider whether enhancement of device stability could also be achieved by the direct addition of covalently linked PC~61~BM dimers in solution prior to deposition of the active layer. Such an approach could provide a more structurally controlled and robust route to inhibit PC~61~BM aggregation, as well as enabling direct quantification of the degree of PC~61~BM dimerization required to enhance device stability, thus avoiding limitations associated with thermally induced dissociation observed for photoinduced PC~61~BM oligomers.[@b6c] Our synthetic strategy was to make dumbbell-shaped dimeric fullerenes where the fullerenes are linked by an alkyl bridge between the ester functional group on PCBM. In a first step commercially available PCBM was hydrolyzed using a previously described literature procedure and the resulting PCBA was recovered as a black soot.[@b10] The PCBA was immediately used in the next reaction step without any further purification. The use of thionyl chloride to convert the PCBA into the corresponding acyl chloride was disregarded because thionyl chloride potentially p-dopes carbon-based molecules like fullerenes, leading ultimately to poor device efficiencies.[@b11] To prevent undesired doping with thionyl chloride, two equivalents of PCBA were reacted with one equivalent of ethylene glycol under mild reaction conditions using Steglich esterification. Isolating the desired (PCB)~2~C~2~ dumbbell in sufficient purity from the complex reaction mixture however proved difficult and gave unsatisfactory yields (\<10 %). As a result of the impossibility to purify the poorly soluble PCBA starting material, it is likely that the acid and diol were not present in the required stoichiometric ratio, leading to the formation of a variety of compounds, difficult to separate. To circumvent these problems, the esterification reaction was performed in two steps as shown in Scheme [1](#sch01){ref-type="fig"}. The crude PCBA was reacted with a large excess of diol (10 equivalents) to encourage the formation of PCBC~2~OH. The highly soluble PCBC~2~OH intermediate was readily purified by column chromatography and recovered in good yields. After a second esterification between PCBC~2~OH and PCBA, the (PCB)~2~C~2~ dumbbell could be successfully assembled and isolated in reasonable yield and excellent purity as evidenced by the ^1^H-MALDI and NMR spectra shown in Figures S5 and S10 in the Supporting Information. The singlet associated with the terminal methyl protons in PCBM appears at 3.67 ppm, whereas the singlet of the bridging methylene protons in (PCB)~2~C~2~ are shifted towards lower fields and appear at 4.26 ppm. All remaining signals associated with the aromatic and alkyl protons are identical in both cases and confirm that our dimerization approach has no effect on the chemical environment of the fullerenes. {#sch01} Because our goal was to blend the dimeric fullerenes with PCBM, we investigated the impact of the dimerization on their electronic structure. There have been numerous approaches to synthesize covalently bound fullerene dimers, however most of these synthetic procedures also modify the frontier energy levels of the fullerene dumbbells.^12^ From this perspective our controlled dimerization approach is advantageous, allowing the synthesis of structurally well-defined PCBM dumbbells with similar frontier energy levels to PCBM, as evidenced by density functional theory (DFT) calculations (Table S1) and cyclic voltammetry (CV) measurements (Figure S11). In both cases the HOMO was estimated at −5.94 eV and the LUMO at −3.85 eV, thus confirming the trend predicted by DFT calculations. The result might seem intuitive, but ensuring similar energy levels for both PCBM and the dimer (PCB)~2~C~2~ is crucial in order to achieve high PCE. In case the LUMO energy level of the dimer would be significantly lower than the one of PCBM, the added dimer would act as a charge trap, which would be detrimental for device performance. To investigate the effect of addition of (PCB)~2~C~2~ on the active layer morphology and device performance, a blend of PCBM with the carbazole-based PCDTBT polymer was chosen as model system. PCDTBT is a high-performing amorphous polymer with a glass-transition temperature (*T*~g~≈106 °C) in neat films below the temperatures used here for examination of thermal stability.[@b13] To determine the minimum weight percentage of (PCB)~2~C~2~ dumbbell necessary to stabilize the PCDTBT:PCBM blend under thermal stress, blends of PCDTBT:fullerene (1:2) were spin-coated on silicon substrates from chlorobenzene and thermally annealed at 140 °C for 1 hour. We have studied the effect of dimer content in the fullerene component, while maintaining the overall weight ratio of polymer:fullerene constant. From the optical micrographs presented in Figure S14, it is apparent that the addition of dimer leads to a more thermally stable blend. At low dimer loadings (\<1 % by mass) the growth of micron-sized PCBM crystallites in the blend cannot be inhibited. However, the formation of micron-sized PCBM crystallites at temperatures above the *T*~g~ of the pure polymer is completely impeded upon addition of 20 % of (PCB)~2~C~2~ (Figure [1](#fig01){ref-type="fig"}), and indeed the number density of the PCBM crystallites is suppressed by over 50 % by adding only 5 % of dimer to the blend (Figure [2 A](#fig02){ref-type="fig"}). This low threshold is consistent with our previous conclusion that the primary role of the PC~61~BM oligomers, respectively (PCB)~2~C~2~, is to frustrate and eventually inhibit PC~61~BM crystal nucleation, with such nucleation processes often being hindered by low compositions of additive materials.[@b6c], ^14^ {#fig01} {#fig02} The effect of (PCB)~2~C~2~ on the nanomorphological behavior upon thermal stress analogous to solar-cell-operating conditions was also investigated. Our previous studies have shown that thermal stress at temperatures relevant to device operation does not lead to micron-size crystallites, but instead results in the formation of nanometer-sized PCBM features in PCDTBT:PCBM blend films.[@b6c] The evolution of the blend morphology upon thermal annealing at 85 °C, customary for standard thermal stress tests, on PEDOT:PSS substrates is shown in Figure [1](#fig01){ref-type="fig"}.[@b15] As revealed by AFM, as cast samples are relatively smooth and featureless, with a similar surface roughness for both with and without (PCB)~2~C~2~. Upon thermal annealing at 85 °C for 1 h, the samples without (PCB)~2~C~2~ exhibit densely formed PCBM crystallites of the order of 150 nm in diameter and 30 nm in height. We have previously assigned these nanoscale features to PCBM aggregates, and correlated their appearance with the degradation of device efficiency under modest thermal stress.[@b6c] In clear contrast, the samples with 20 % (PCB)~2~C~2~ show drastically suppressed PCBM aggregation with no measurable change in film surface morphology after thermal treatment. The film morphology was further probed by bright-field transmission microscopy (TEM). At 80 °C, both blend films, with and without (PCB)~2~C~2~, present very comparable features. The non-(PCB)~2~C~2~ containing blend contains spherical aggregates dispersed homogenously throughout the film (Figures S15 to S17), with diameters ranging from about 150--250 nm. The (PCB)~2~C~2~ containing blend showed comparable features, but a much lower density of fullerene nuclei of approximate diameter of 100--150 nm was found. This modest difference suggests that the observed spherical aggregates in the non-(PCB)~2~C~2~ containing blend might play a key role in PCBM crystal nucleation and device degradation even though the exact mechanism is not yet elucidated. Upon thermal annealing at 140 °C, the formation of fullerene nuclei varying in size (about 50--100, 100--150, and 150--200 nm) was observed in the non-(PCB)~2~C~2~ containing blend. More importantly however, we observed the formation of needle-like structures, similar to the ones previously observed in optical microscopy image, but much smaller in size (50--100 nm in length and 10 nm in width). In the annealed blend film containing 20 % of (PCB)~2~C~2~ these sheaf-like structures were not observed, but the morphology was instead characterized by the presence of spherical aggregates, on average smaller than the ones observed prior to 140 °C annealing. These results are in further agreement with our previously made observation that a low concentration of (PCB)~2~C~2~ in the fullerene phase is stabilizing the blend nanomorphology and seems to influence the shape and number density of PCBM aggregates under modest thermal stress. Knowing that the PCBM dumbbell has a positive effect on the thermal stability of active layer blend films, we focused on its effect on the performance of PCDTBT:PCBM solar cells. A standardized thermal stress of 85 °C was applied to the devices under nitrogen atmosphere, with the current--voltage (*J*--*V*) curve recorded using repeated dark thermal annealing/room-temperature device testing cycles. For *J*--*V* measurements, devices were exposed under a solar simulator for less than 30 seconds during each measurement to minimize the effect of PCBM photo-oligomerization. Figure [2](#fig02){ref-type="fig"} demonstrates the impact of a low percentage of PCBM dimers on the efficiency and thermal stability of PCDTBT:PCBM devices. Devices containing 20 % of (PCB)~2~C~2~ show a modest improvement in efficiency compared to devices without dimer. This improvement is primarily due to an enhancement in fill factor (FF) from around 0.56 to 0.62, whereas both *V*~oc~ and *J*~sc~ are nearly identical in both sets of devices (Table S2). The cause of improved FF is not obvious from the AFM images (e) and (f) in Figure [1](#fig01){ref-type="fig"}, but is likely due to an improvement in the blend morphology induced by the dimers during film deposition. For the devices fabricated without (PCB)~2~C~2~, a rapid degradation in device performance was observed upon thermal stress, with a 20 % loss in power conversion efficiency (PCE) within the first 25 min, primarily due a loss of fill factor, and likely linked to the densely formed nanoscale PCBM aggregates. Remarkably, devices containing 20 % of the (PCB)~2~C~2~ show significantly improved thermal device stability with a reduction in the PCE by 20 % occurring only after 2000 minutes. It is thus apparent that the device thermal stability of conventional PCDTBT:PCBM devices can be enhanced by an order of magnitude by using a low percentage of (PCB)~2~C~2~, at least under the thermal stress conditions and time scale studied herein. Given that the addition of (PCB)~2~C~2~ not only enhances the long-term stability, but also the device efficiency, we evaluated the potential to further enhance the performance by systematically increasing the PCBM dimer loading from 0 % (100 % of PCBM) to 100 % (0 % of PCBM) while keeping the total concentration of fullerene in the active layer blend constant. The initial device *J*--*V* characteristics, as a function of the (PCB)~2~C~2~ loading, are shown in Figure S18. From 0 to 20 % of (PCB)~2~C~2~ loading, modest increases in PCE are observed because of enhanced FF values. However, further increase in dimer loading results in decreasing device PCE, caused by decreases in both *J*~sc~ and FF (see Table S2 in the Supporting Information for detailed device parameters). The loss in PCE results in part from a higher series resistance, indicative of a drop in electron mobility in the fullerene phase with increasing PCBM dimer loading. To confirm this assumption, organic field-effect transistors (OFET) were built to measure the electron mobility as a function of (PCB)~2~C~2~ loading. The electron mobility of a neat PCBM film was found to be around 1×10^−3^ cm^2^ V^−1^ s^−1^, which is in good agreement with previously reported literature values.^16^ As evidenced from Figure [3 A](#fig03){ref-type="fig"}, a sharp drop in electron mobility is observed at (PCB)~2~C~2~ loadings exceeding 10--20 %, which correlates well with the lower *J*~sc~ measured at higher dumbbell loadings and the resulting drop in PCE. These findings are in line with previous work by Distler et al., who demonstrated that photodimerization of PCBM causes a significant reduction of charge carrier mobility in a polymer--fullerene system, resulting in a reversible degradation of device efficiency under constant illumination.[@b17] The electron mobility measurements in conjunction with the device stability data therefore conclusively show that there is an optimal (PCB)~2~C~2~ loading around 20 % at which both the device efficiency and the device lifetime can be simultaneously enhanced. {#fig03} Based on the aforementioned results, the addition of (PCB)~2~C~2~ leads to improved power conversion efficiencies, mainly because the growth of nanometer-sized PCBM crystallites in the active layer under thermal stress is prevented as evidenced by AFM measurements. To further investigate by what mechanism this crystallite growth is suppressed, we have carried out interdiffusion measurements. We hypothesized above that the addition of (PCB)~2~C~2~ would lead to lower mass diffusion coefficient, thus slowing down and eventually preventing, the growth of PCBM nanocrystallites. The diffusion of (PCB)~2~C~2~ from a bulk heterojunction blend with PCDTBT into a neat PCDTBT thin film was thus investigated with dynamic secondary ion mass spectrometry as a function of annealing temperature (Figure [4](#fig04){ref-type="fig"}). The bilayers remained stable in composition, as prepared from solution casting and film lamination, until the annealing temperature approached the *T*~g~ of PCDTBT (beginning at 100 °C; Figure S19). At temperatures near or slightly below *T*~g~, the (PCB)~2~C~2~ infiltrated only partially into the originally neat PCDTBT film, with a decreasing dimer concentration with diffusion distance. At temperatures above *T*~g~, the (PCB)~2~C~2~ diffused entirely across the PCDTBT film with uniform concentration. These results suggest that the molecular mobility of PCDTBT is critical in determining the ability for the dimer molecule to traverse the polymer phase. While (PCB)~2~C~2~ and PCBM show qualitatively the same behavior with PCDTBT, we cannot rule out that the diffusion of the former is slowed down, as the DSIMS measurement provides only a bound for the diffusion constant. We can state however that it is not slowed so much as to retard diffusion at timescales commonly used in processing of BHJs, that is, minutes. More critically, the miscibility in PCDTBT seems unaffected by the dimerization process. These results imply that in blends with PCDTBT, dimerized and undimerized PCBM show similar diffusion kinetics and suggest that rather than slowing growth of aggregates and crystals by mass diffusion in such blends, that the stability of the microstructure is achieved through frustration of molecular packing of PCBM by (PCB)~2~C~2~. Grazing incidence wide-angle X-ray scattering (GIWAXS) measurements performed on PCDTBT:fullerene blends revealed only some weak ordering from PCDTBT, with (100) and (200) peaks that are broad and weakly observable in the out-of-plane and in-plane directions (Figure S25). The large, broad aggregate peak centered at 1.4 Å^−1^ has contributions from the weakly ordered/aggregated PCBM and possibly (PCB)~2~C~2~, as well as from the amorphous regions of PCDTBT. No diffraction peaks related to the spherical fullerene aggregates observed in TEM images could be identified, which is likely because these are only very weakly ordered. The hypothesis of frustrated molecular packing is further supported by differential scanning calorimetric (DSC) measurements as evidenced from Figures S20--S24. Pure PCBM exhibits a sharp melting peak at 280 °C with an associated enthalpy of fusion of 14.8 J g^−1^. However upon addition of 20 % of (PCB)~2~C~2~ dumbbell, it is possible to quench the PCBM:(PCB)~2~C~2~ blend in a glassy state (*T*~g~ of about 200 °C) and even after 2 h of thermal annealing at 85 °C the sample did not return to a crystalline state comparable to the neat PCBM sample (Figure [3 B](#fig03){ref-type="fig"}). ![Dynamic secondary ion mass spectrometry profiles of ^2^H and ^32^S in a PCDTBT:\[D\]PCBM bilayer (solid lines) and in a PCDTBT:(BHJ PCDTBT:\[D\](PCB)~2~C~2~) bilayer (dotted line) fabricated on a SiO~2~/Si substrate. A) Sharp changes in the ^2^H and ^32^S counts are observed in the as-cast bilayer structures. B) The homogenous distributions of the deuterated species is achieved at an annealing temperature (140 °C) exceeding the *T*~g~ of the polymers.](anie0053-12870-f4){#fig04} These results demonstrate conclusively that the addition of a low-weight percentage of modified fullerene to a polymer--fullerene active layer does not negatively affect the PCE of BHJs and can lead to the significantly enhanced morphological stability. The underlying mechanism appears to be, not a significant reduction in the mass diffusion of the fullerene, but rather the frustration of fullerene nanocrystal nucleation. We find that there is an optimal loading, of approximately 20 % fullerene dimer, that yields a device lifetime increase of around 20 % under realistic operating conditions. In contrast with our previously reported strategy of photo-induced fullerene oligomerisation in the active layer, which yields a time-dependent fullerene--oligomer population (as a function of time and illumination), this approach is robust to environmental conditions and thus potentially more attractive for practical implementation.[@b6c], [@b8] Supporting Information ====================== Supporting information for this article, including synthetic details and additional graphs (CV, UV/Vis, OPV, DSC, OFET) and images (microscopy, TEM), is available on the WWW under http://dx.doi. org/10.1002/anie.201407310. ###### miscellaneous_information

{#sp1 .205} {#sp2 .206}

Introduction {#Sec1} ============ Hydrogen is the cleanest fuel and a cost-effective energy carrier of the future \[[@CR1]--[@CR6]\], which produces three times more energy per unit mass than fossil fuels \[[@CR7]\]. Industrially, hydrogen is generated from fossil fuels releasing higher amounts of greenhouse gas, CO~2~ \[[@CR8], [@CR9]\]. A sustainable supply of hydrogen from renewable resources, such as biomass, is highly desirable \[[@CR4], [@CR10]\]. Development of homogeneous transition metal catalysts for complete decomposition of polyhydroxy biomass constituents, i.e., carbohydrates, into H~2~ and CO~2~, could revolutionize H~2~ production from renewable resources. Methanol is the simplest model for such carbohydrates, but complete dehydrogenation of this compound according to$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathrm{MeOH}+{\mathrm{H}}_2\mathrm{O}\to {3\mathrm{H}}_2+{\mathrm{CO}}_2 $$\end{document}$$would be of considerable interest in its own right. This decomposition of methanol could follow a sequence of three reactions, (i) dehydrogenation, (ii) decarbonylation, and (iii) water--gas shift reaction (WGSR, see Scheme [1](#Sch1){ref-type="fig"}), all of which are well known.Scheme 1Putative reaction sequence for complete methanol dehydrogenation Development of homogeneous transition metal catalysts for H~2~ production from methanol \[[@CR11], [@CR12]\] has attracted much attention in the past few decades. Morton and Cole-Hamilton developed a ruthenium catalyst \[Ru(H)~2~(X~2~)(PPh~3~)~3~\] (X = N, H) for partial dehydrogenation of alcohols (including methanol) with notable turnover frequencies \[[@CR13]\]. Aldehydes and ketones were the main products \[formaldehyde in case of methanol, Scheme [1](#Sch1){ref-type="fig"}, step (i)\]. During the conversion of ethanol, significant amounts of methane and a carbonyl complex, \[RuH~2~(CO)(PPh~3~)~3~\], were produced through decarbonylation \[Scheme [1](#Sch1){ref-type="fig"}, step (ii)\]. However, since no CO~2~ was noticed, apparently these Ru complexes are not active as WGSR catalysts \[[@CR14]\]. Recently, based on density functional theory (DFT) calculations, we studied the mechanisms for the dehydrogenation \[[@CR15]\] and decarbonylation \[[@CR16]\] of aliphatic alcohols catalyzed by the Morton and Cole-Hamilton system, \[RuH~2~(H~2~)(PPh~3~)~3~\] \[Scheme [1](#Sch1){ref-type="fig"}, steps (i) and (ii)\]. When exploring the feasibility of WGSR with these complexes computationally \[[@CR17]\], we noticed that the first step of WGSR, the attack of water (in the form of OH^−^) on the CO ligand (Scheme [2](#Sch2){ref-type="fig"}) is highly endergonic.Scheme 2General mechanism for transition-metal-catalyzed WGSR under basic conditions, where OH^−^ is the nucleophile Significant research has been undertaken to reveal the mechanism of WGSRs catalyzed by homogeneous transition metal complexes, particularly the metal carbonyls of Fe, Ru, and Os \[[@CR18]--[@CR30]\]. Recently, Guo et al. have studied the WGSR mechanism catalyzed by hexacarbonyl complexes of Mo and W \[[@CR31]\]. In all of these reactions, which are conducted under basic conditions, OH^−^ is the nucleophile and its uptake to form a transient metallacarboxylic acid is considered as the initial step (in the grey box in Scheme [2](#Sch2){ref-type="fig"}). Such attack of OH^−^ on carbonyl ligands is well known as Hieber base reaction \[[@CR32]\]. In all studies of metal carbonyl catalyzed WGSR, this OH^−^ uptake appeared to be highly exothermic and essentially barrierless. In contrast, in our study of WGSR in the Morton and Cole-Hamilton system \[[@CR17]\], this step is predicted to be highly endergonic. The driving force for formation of the metallacarboxylic acid depends notably on the co-ligands that are present, in particular on the number of CO ligands. While large negative enthalpies and free energies are computed for the OH^−^ uptake of Ru(CO)~5~ \[[@CR28], [@CR29]\] ΔG = 127.7 kJ/mol and 81.6 kJ/mol are predicted for \[RuH~2~(CO)(PPh~3~)~3~\] and \[RuH~2~(CO)~2~(PPh~3~)~2~\], respectively. It therefore appears that suitable ligand design, by varying the steric or electronic properties of the ligands, could make the process of OH^−^ uptake feasible. In this work, we now report DFT-computed driving forces for OH^−^ uptake in a number of metal-carbonyl complexes of Ru, Fe, and Os. Along with CO, we have made the choice of trimethylphosphine (comparable to triphenylphosphine), trifluorophosphine (a strong π-acceptor ligand, comparable to CO) \[[@CR33]\], pyridine, and bipyridine ligands. For a perfect catalytic system, the OH^−^ entry into the cycle should be facile and should not produce a very low-lying intermediate on the reaction profile that would eventually deactivate the catalytic system. This work can lead to the rational design of better catalysts for WGSR and, eventually, towards the complete decomposition of alcohols by dehydrogenation, decarbonylation, and the finally WGSR, which could facilitate entry into a hydrogen-based economy. Results and discussion {#Sec2} ====================== Ligand effects on the initial uptake of OH^−^ to form the metallacarboxylic acid {#Sec3} -------------------------------------------------------------------------------- In metal carbonyls, OH^−^ uptake is usually considered a fast process, and the resulting metallacarboxylic acid is usually too reactive to be isolated and decarboxylates under CO~2~ evolution to form a hydride (Scheme [2](#Sch2){ref-type="fig"}). Protonation of this hydride intermediate takes the system uphill on the free energy reaction profile and the H~2~ evolution tends to be associated with the highest energy transition state. Our work focuses on the ligand effects on the initial uptake of OH^−^ to metal carbonyls (Eq. [2](#Equ2){ref-type=""}, corresponding to the first step in the grey box in Scheme [2](#Sch2){ref-type="fig"}), taking M(CO)~5~ pentacarbonyls as prototypical representatives (M = Fe, Ru, Os). Step by step, we replaced each of the CO in the metal pentacarbonyl system with selected ligands, namely trimethylphosphine (PMe~3~), trifluorophosphine (PF~3~), pyridine (py), and bipyridine (bipy). We compared the free energies of the reactants (metal carbonyls) and the products (metallacarboxylic acids) to see how such a ligand change affects the driving force for the OH- uptake to the system. We have included the results for Ru carbonyls and metallacarboxylic acids in the main paper, whereas the results for the Fe and Os analogs are included within the supporting information (SI).$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\mathrm{L}}_n\mathrm{M}\left(\mathrm{CO}\right)+{\mathrm{OH}}^{-}\to {\left[{\mathrm{L}}_{\mathrm{n}}\mathrm{M}\left({\mathrm{CO}}_2\mathrm{H}\right)\right]}^{-},\mathrm{M}=\mathrm{Fe},\mathrm{Ru},\mathrm{Os},\mathrm{L}=\mathrm{CO},{\mathrm{PMe}}_3,{\mathrm{PF}}_3,\mathrm{py},\mathrm{bipy},\mathrm{Cl},\mathrm{H} $$\end{document}$$ Unlike CO, PMe~3~ is a weak π-acceptor ligand and experiences weak backbonding with the metal center. At our chosen level of theory, B97-D/ECP2//RI-BP86/ECP1, Ru(CO)~5~ has a driving force of ΔG = −93.5 kJ/mol for the initial OH^−^ uptake. This driving force decreases (i.e., ΔG increases) as we increase the number of PMe~3~ ligands that replace CO. On substituting one CO with one PMe~3~ ligand at the axial position, the free energy increases to −38.9 kJ/mol, which further increases to 49.2 kJ/mol on replacing the second CO on the axial position with another PMe~3~ ligand \[[@CR34]\]. The OH^−^ uptake to Ru(CO)~2~(PMe~3~)~3~ is unfavorable by a free energy of 87.8 kJ/mol, that of Ru(CO)(PMe~3~)~4~ by 122.2 kJ/mol (Fig. [1](#Fig1){ref-type="fig"}). Since the subsequent steps on the WGSR reaction profile (Scheme [2](#Sch2){ref-type="fig"}) add additional barriers, the latter two complexes are expected to be only weakly active or unproductive as WGSR catalysts.Fig. 1Computed free energies (B97-D level, kJ/mol) for the OH^−^ uptake of the carbonyl reactant (note that this reactant is different for each product). The number of PMe~3~ ligands increases from left to right (see Fig. [S1](#MOESM1){ref-type="media"} in the ESI for a plot showing three-dimensional representations of the complexes) Based on the results obtained for the PMe~3~ ligand exchange with CO, one can assume that the driving force for the OH^−^ uptake should be affected by replacing CO ligands with ligands of slightly greater π-acceptor strength, e.g., PF~3~. A slight increase of the driving force for OH^−^ uptake is observed on replacing one CO at the axial position with a PF~3~ ligand, from −93.5 kJ/mol in Ru(CO)~5~ (Fig. [1](#Fig1){ref-type="fig"}) to −101.8 kJ/mol for Ru(CO)~4~(PF~3~) (Fig. [2](#Fig2){ref-type="fig"}). On exchanging both axial CO ligands with PF~3~, the free energy decreases by 31.7 kJ/mol for the OH^−^ uptake as compared to the free energy of the OH^−^ uptake in Ru(CO)~5~. One would expect that exchanging three CO ligands, two at the axial and one at the equatorial position, should further favor OH^−^ uptake, but this is not the case. After the OH^−^ uptake by Ru(CO)~2~(PF~3~)~3~, the metalla-acid, \[Ru(CO)(COOH)(PF~3~)~3~\]^−^, is obtained with a free energy of −112.8 kJ/mol, which is higher by 12.5 kJ/mol than the free energy for the OH^−^ uptake in Ru(CO)~2~(PF~3~)~3~, possibly because of the *trans* influence of CO. When employing another, higher-lying isomer of Ru(CO)~2~(PF~3~)~3~, where one of the CO ligands is positioned *trans* to PF~3~, the resulting metalla-acid is obtained at a ΔG of −121.4 kJ/mol. This value is lower by 8.6 kJ/mol compared to the ΔG for the most stable isomer included in Fig. [2](#Fig2){ref-type="fig"}. There is thus a noticeable *trans* influence on the driving force under scrutiny, although it does not seem to override other electronic effects.Fig. 2Relative free energy (kJ/mol) for the OH^−^ uptake with that of the respective carbonyl reactant set to 0.0 kJ/mol in each case. The number of PF~3~ ligands increases from left to right (see Fig. [S2](#MOESM1){ref-type="media"} in the ESI for a plot showing three-dimensional representations of the complexes) Ru(CO)(PF~3~)~4~ follows the expected trend in terms of free energy for the OH^−^ uptake and has the largest predicted affinity for OH^−^ of all complexes studied here (−143.7 kJ/mol, Fig. [2](#Fig2){ref-type="fig"}). It should be noted that such a large driving force for OH^−^ uptake does not necessarily make this complex a good target for a WGSR catalyst, because a correspondingly higher energy needs to be invested to close the cycle and re-form the initial catalyst. A variety of \[Ru\]-CO~2~H complexes are known, some of which have been structurally characterized, notably with bidentate aromatic N-donor ligands \[[@CR35], [@CR36]\]. We therefore included a couple of model complexes with aromatic N-donor ligands, namely pyridine (py) and bipy. Both are coordinated through the lone pair of an electronegative N atom providing inductive donation, with the aromatic backbone allowing for significant π-backbonding interaction. As the inductive donation from the nitrogen lone pair is counteracted by the backbonding into the aromatic system, on exchanging one CO at the axial position in Ru(CO)~5~ with a pyridine (py), the OH^−^ uptake becomes unfavorable as compared to that in parent Ru(CO)~5~, but not by that much as in case of PMe~3~ ligand. On replacing two CO with a bipy bidentate ligand, the OH^−^ uptake becomes further unfavorable as compared to that in parent Ru(CO)~5~, but remains favorable by 63.4 kJ/mol as compared to the OH^−^ uptake in the Ru(CO)~3~(PMe~3~)~2~ system (Fig. [3](#Fig3){ref-type="fig"}).Fig. 3Relative free energy (kJ/mol) for the OH^−^ uptake on replacing one CO ligand with py (*left*) and two CO ligands with bipy (*right*) (see Fig. [S3](#MOESM1){ref-type="media"} in the ESI for a plot showing three-dimensional representations of the complexes) Finally, in addition to the pentacoordinate Ru(0) species, we considered the OH^−^ uptake in a few selected octahedral Ru(II) complexes including Ru(CO)(H)~2~(PMe~3~)~3~, Ru(CO)(H)~2~(PF~3~)~3~, and \[Ru(CO)~3~Cl~3~\]^−^ (Fig. [4](#Fig4){ref-type="fig"}). The driving force for OH^−^ uptake in Ru(CO)(H)~2~(PMe~3~)~3~, endergonic by 82.7 kJ/mol, is comparable to that in the Morton and Cole-Hamilton system, Ru(CO)(H)~2~(PPh~3~)~3~, where it is endergonic by 127 kJ/mol at essentially the same level (at a higher temperature, however, 150 °C) \[[@CR17], [@CR37]\]. On replacing PMe~3~ ligands with PF~3~ ligands, the product is obtained at a free energy of −61.2 kJ/mol, obviously because of strong π-backbonding interaction.Fig. 4Relative free energy (kJ/mol) for the OH^−^ uptake with that of the respective octahedral carbonyl reactant set to 0.0 kJ/mol in each case (see Fig. [S4](#MOESM1){ref-type="media"} in the ESI for a plot showing three-dimensional representations of the complexes) The results for the analogous Os complexes are very similar to those for the Ru species just discussed, with individual driving forces for OH^−^ uptake within typically 10 kJ/mol of each other (ca. 20 kJ/mol for the bipy complex, compare Tables [S1](#MOESM1){ref-type="media"} and [S3](#MOESM1){ref-type="media"} in the [SI](#MOESM1){ref-type="media"}). On going from Ru to Fe congeners, the changes in this driving force become somewhat more variable (up to ca. 30 kJ/mol, compare Tables [S1](#MOESM1){ref-type="media"} and [S2](#MOESM1){ref-type="media"} in the [SI](#MOESM1){ref-type="media"}), but overall the same trends are obtained irrespective of the group 8 metal. The metallacarboxylic acid arising from Hieber base reaction of \[Ru(CO)~3~Cl~3~\]^−^ has been implicated as a key reactive intermediate in a complex variety of reactions \[[@CR32]\]. Indeed, despite forming a dianion from two monoions, OH^−^ uptake of \[Ru(CO)~3~Cl~3~\]^−^ affording \[Ru(CO)~2~(CO~2~H)Cl~3~\]^2−^ is highly exergonic, with a free energy of −119.0 kJ/mol. This large driving force is fully consistent with the fact that this complex is a reactive intermediate that can be formed through Hieber base reaction \[[@CR32]\]. Experimentally, \[Ru(CO)~2~(CO~2~H)Cl~3~\]^2−^ appears to lose a chloride ion consistent with our calculations as at our level, as this process is computed to be slightly exergonic, by −5.9 kJ/mol. Natural population analysis {#Sec4} --------------------------- What is the origin of the huge variation in driving forces for OH^−^ uptake in these complexes? Hypothesizing that a key factor should be delocalization of the additional negative charge brought into the complex, we used natural population analysis (NPA) \[[@CR38]\] to evaluate the extent of charge transfer from OH^−^ upon attack on the carbonyl ligand. To this end, we simply calculated the natural charge on the OH^−^ fragments in the ruthenacarboxylic acid products, assessing how it changes from the value in free OH^−^, where it is −1. A substantial reduction from this absolute value is found in the complexes, indicating that most of the charge is actually delocalized into the complex, but there is still a notable variation of this charge, between −0.32 and −0.19 (see Table [S4](#MOESM1){ref-type="media"} in the [SI](#MOESM1){ref-type="media"}). The PMe~3~ ligand has σ-donating abilities, which pushes electron density to the metal center, which increases the amount of backbonding interaction between the filled metal *d*-orbital and the empty π\*- orbital of the carbon atom of CO. The overall affect makes it difficult for the OH^−^ fragment to delocalize electron density over the metal complex. As we replace more CO ligands with PMe~3~ ligands, the natural charge at the OH^−^ fragment decreases. In \[Ru(CO)~2~(CO~2~H)(PMe~3~)~2~\]^−^, presence of a CO ligand at the axial position *trans* to the --CO~2~H^−^ fragment increases its distance from the metal center, making it less available for OH^−^ fragment to accommodate the charge density. Here the *trans* influence dominates the electronic nature of the ligands and a small discrepancy in the natural charges of the OH^−^ fragment occurs when we move from \[Ru(CO)~2~(CO~2~H)(PMe~3~)~2~\]^−^ to \[Ru(CO)(CO~2~H)(PMe~3~)~3~\]^−^, similar is the case with py and bipy ligands. For the PF~3~ ligands, on replacing each with the CO ligands, the natural charge at the OH^−^ fragment gradually increases, as expected (Table [S4](#MOESM1){ref-type="media"}). A plot of the computed driving forces vs. OH^−^ charges indeed reveals an overall trend towards more favorable OH^−^ uptake with decreasing charge on this fragment (see Fig. [5](#Fig5){ref-type="fig"}; essentially the same correlation is obtained when enthalpies are used instead of free energies, see Fig. [S5](#MOESM1){ref-type="media"} in the [SI](#MOESM1){ref-type="media"}). No strict relationship is apparent, and there are a few outliers, notably the bipy complex, but overall our results are consistent with the charge delocalization being a key factor for the driving force of this reaction step. For instance, the PF~3~ ligands, which are predicted to strongly promote Hieber base reaction (Fig. [2](#Fig2){ref-type="fig"}), are indicated to do so because they are very efficient in delocalizing the incoming negative charge (see red triangles in the lower right of Fig. [5](#Fig5){ref-type="fig"}). Arguably, the extent of this charge delocalization will depend on the balance between σ-donating and π-backdonating capabilities of the co-ligands, which should allow for a rational tailoring of complexes toward Hieber base reaction and, eventually, for designing new WGSR catalysts.Fig. 5Plot of driving forces for OH^−^ uptake vs. natural charges of the OH^−^ fragments in the products (B97-D level) Conclusions {#Sec5} =========== In summary, using an appropriate DFT level, we have computed the driving forces for formation of metallacarboxylic acids from group 8 carbonyl complexes through uptake of OH^−^. This reaction (Eq. [2](#Equ2){ref-type=""}), known as Hieber base reaction, is the first step of water--gas shift reaction (WGSR) that can be catalyzed by transition metal complexes under basic conditions. According to our findings, the driving force for this step is surprisingly sensitive to the nature of the co-ligands at the metal, and can range from ΔG = −144 kJ/mol to +122 kJ/mol \[for R = F and Me, respectively, in Ru(CO)(PR~3~)~4~\]. Far from being innocent spectator ligands, these co-ligands actively take part in OH^−^ uptake through delocalization of the negative charge, as apparent in the computed atomic charges from natural population analysis. Fe and Ru pentacarbonyls are prototypical WGSR catalysts; it is remarkable how replacement of CO ligands with electron-rich phosphines (which are ubiquitous in modern transition metal chemistry) can impede the first step of this WGSR catalytic cycle. In that case, use of phosphines with electron-withdrawing substituents (where we have used PF~3~ as extreme example) or aromatic N-donor ligands can increase the driving force for Hieber base reaction. Compared to these ligand effects, the nature of the metal (Fe, Ru, or Os) or its oxidation state \[e.g., Ru(0) vs. Ru(II)\] seems to be of lesser importance for OH^−^ uptake. We are convinced that the tunability of the driving force for Hieber base reaction through appropriate choice of co-ligands can inform on the rational design of new WGSR catalysts. As this quantity, a simple reaction (free) energy, can be readily computed with modern DFT tools, large libraries of ligands can be screened computationally, opening up new avenues for applications of molecular modeling in homogeneous catalysis. Computational methodology {#Sec6} ========================= In this work, we are mainly interested in calculating the driving force related to change in Gibbs free energy for the OH^−^ uptake, which is considered as the initial step of the WGSR. Our calculations follow a computationally cost-effective protocol based on density functional theory (DFT) that had been validated \[[@CR39]\] and fruitfully applied to mechanistic DFT studies of related Ru complexes \[[@CR15], [@CR16]\]. Geometry optimizations were carried out at a lower level, RI-BP86/ECP1, whereas the energies were refined at B97-D/ECP2 \[[@CR40]\] level. All the metal complexes were fully optimized at the RI-BP86/ECP1 level, i.e., by the use of Becke \[[@CR41]\] and Perdew \[[@CR42]\] exchange and correlation functionals along with SDD \[[@CR43]\] core potential and valence basis on the metal atoms, whereas all the other atoms were treated with the standard 6-31G(d,p) basis. The nature of all the minima was verified by frequency calculations within the harmonic approximation, which were further used to obtain the enthalpic and entropic corrections under standard conditions (1 atm and 298.15 K). Thermochemical correction terms *δE*~*G*~ were obtained as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \delta {E}_G=\varDelta {G}_{RI- BP86/ ECP1}-\varDelta {E}_{RI- BP86/ ECP1} $$\end{document}$$where *ΔE*~*RI-BP86/ECP1*~ is the reaction energy and *ΔG*~*RI-BP86/ECP1*~ is the corresponding Gibbs free energy (analogously for corrections to enthaly, *δE*~*H*~ from *ΔH*~*RI-BP86/ECP1*~ - *ΔE*~*RI-BP86/ECP1*~). The energies of the optimized complexes were refined through single-point calculations at the B97-D/ECP2 level, i.e., using dispersion-corrected B97-D functional, which includes Grimme's dispersion correction \[[@CR40]\] along with 6-311+G(d,p) basis set for all the non-metal atoms (SDD on the metals). Solvent effects were included by a polarizable continuum model (PCM) \[[@CR44], [@CR45]\] using methanol as a model solvent with self-consistent reaction field (SCRF) method. The solvent energy correction, *δE*~solv~, was performed as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \delta {E}_{solv}=\varDelta {E}_{PCM}-\varDelta E $$\end{document}$$ Difference between the reaction energy in the continuum is donated as *ΔE*~*PCM*~ and *ΔE* is the difference between the reaction energy in gas phase at B97-D/ECP2 level. The final enthalpies and free energies, *ΔH* and *ΔG*, were obtained as sum of all energy correction terms:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \varDelta H=\varDelta E+\delta {E}_{solv}+\delta {E}_H $$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \varDelta G=\varDelta E+\delta {E}_{solv}+\delta {E}_G $$\end{document}$$where *ΔE* and *δE*~solv~ were calculated at the B97-D/ECP2 level, whereas *δE*~H~ and *δE*~G~ were obtained at RI-BP86/ECP1. Atomic charges from natural population analysis \[[@CR38]\] were evaluated at the B97-D/ECP2/PCM level. All calculations were performed using the Gaussian 09 suite of programs \[[@CR46]\]. In order to identify the most stable isomers and conformers of each reactant and product, an exhaustive screening of the possible stereoisomers was undertaken. Only the results for the most stable forms are reported. The conformation of the carboxylic acid group was uniformly taken as that where the hydrogen of the OH^−^ fragment is pointing towards the metal center. We investigated the stability of these metalla-acids against those in which the hydrogen of the OH^−^ fragments points away from the metal center, particularly in Fe complexes, the former complexes are appeared to be more stable (see Table [S5](#MOESM1){ref-type="media"}). Electronic supplementary material ================================= {#Sec7} ESM 1(PDF 1045 kb) This paper belongs to the Topical Collection Tim Clark 70th Birthday Festschrift **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank EaStCHEM and the School of Chemistry for support. Computations were carried out on a local Opteron PC cluster maintained by Dr. H. Früchtl.

1. Introduction =============== Liver fibrosis is a major complication of various chronic hepatic diseases, resulting from increased production and decreased degradation of the extracellular matrix. Activation and phenotypic transformation of hepatic stellate cells (HSCs) play a central role in the development and resolution of liver fibrosis and cirrhosis \[[@B1-molecules-19-19845],[@B2-molecules-19-19845]\], and in the promotion of hepatocellular carcinoma \[[@B3-molecules-19-19845]\]. Transforming growth factor-β1 (TGF-β1), as the key fibrogenic cytokine \[[@B4-molecules-19-19845],[@B5-molecules-19-19845]\], activates HSCs via the TGFβ1/activin receptor-like kinase 1 (ALK1)/Smad1 pathway and stimulates collagen transcription in HSCs \[[@B6-molecules-19-19845]\]. Glycosylation is the major posttranslational modification to the secretory and the membrane-anchored proteins in eukaryotic cells \[[@B7-molecules-19-19845]\], which plays important roles in cell-cell adhesion, bacterial infection, viral attachment, ligand-receptor binding, and other key cellular processes \[[@B8-molecules-19-19845]\]. In recent years, a growing number of studies have reported that protein glycosylations are frequently altered during physiological and pathological changes \[[@B9-molecules-19-19845],[@B10-molecules-19-19845],[@B11-molecules-19-19845]\]. Many receptor--ligand interactions lead to alteration of protein-bound glycans, and numerous receptors are regulated by glycosylation to a lesser or greater degree; for example, Notch receptors are the recipient substrates of specific glycosyltransferases \[[@B12-molecules-19-19845],[@B13-molecules-19-19845]\], which regulate the activation of Notch by its ligands and thus affect various processes during development. In HSCs, interleukin-1 and tumor necrosis factor-α have lectin-like activities \[[@B14-molecules-19-19845]\] and recognize disialylated diantennary N-glycans bearing two Neu5Aca2,3 residues \[[@B15-molecules-19-19845]\] and heparin \[[@B16-molecules-19-19845],[@B17-molecules-19-19845]\], respectively. Our previous study also indicated that concanavalin A (ConA) and six other lectins showed stronger binding to the activated HSCs than to the quiescent HSCs \[[@B18-molecules-19-19845]\]. Lectins are carbohydrate-binding proteins that discriminate terminal glycans based on subtle differences in their structures, which can be exploited in the laboratory to detect or isolate specific carbohydrate structures of glycoconjugates, including glycoproteins and glycolipids as well as glycosaminoglycans \[[@B19-molecules-19-19845],[@B20-molecules-19-19845]\]. Along with the rapid improvement of cutting-edge separation technologies coupled with mass spectrometry in the fields of glycoproteomics and glycomics, glycoproteins can be enriched for analysis by hydrazide chemistry \[[@B21-molecules-19-19845],[@B22-molecules-19-19845]\], lectin affinity \[[@B23-molecules-19-19845],[@B24-molecules-19-19845]\], immunoaffinity \[[@B25-molecules-19-19845],[@B26-molecules-19-19845]\], boric acid chemistry \[[@B27-molecules-19-19845],[@B28-molecules-19-19845]\], size exclusion chromatography \[[@B29-molecules-19-19845],[@B30-molecules-19-19845]\], hydrophilic interaction \[[@B31-molecules-19-19845],[@B32-molecules-19-19845]\], and other methods. Lectin affinity chromatography is a unique method that can isolate glycoproteins/glycopeptides containing similar terminal glycan structures from complex samples. In this study, we employed a targeted glycoproteomics approach using lectin-magnetic particle conjugate-based liquid chromatography-tandem mass spectrometry (LC-MS/MS) to compare the profiles of ConA-binding glycoproteins (CBGs) between human HSCs (LX-2 cell line) with and without activation by TGF-β1. The aim of this study was to determine whether any novel CBGs may be differentially expressed in the activated, compared with quiescent LX-2 cells, and consider the potential roles that these CBGs may play in the pathophysiology of liver fibrosis. 2. Results and Discussion ========================= 2.1. Isolation of Con A-Binding Glycoproteins (CBGs) from LX-2 Cells -------------------------------------------------------------------- ConA-magnetic particle conjugates (CMPCs) were used to isolate CBGs from total protein lysates of the activated and quiescent LX-2 cells. The total, unbound, and eluted proteins were resolved by SDS-PAGE and visualized by silver staining. The results showed that the bands of CBGs ([Figure 1](#molecules-19-19845-f001){ref-type="fig"}A, lanes 5 and 6) were complementary to those of the unbound proteins (lanes 3 and 4), and that the eluted CBGs were enriched to some degree from the total proteins (lanes 1 and 2) ([Figure 1](#molecules-19-19845-f001){ref-type="fig"}A). {#molecules-19-19845-f001} 2.2. Identification of CBGs Isolated by CMPCs by LC-MS/MS --------------------------------------------------------- CBGs isolated by CMPCs from total proteins of the activated and quiescent LX-2 cells were identified by Agilent 6530 accurate-mass Q-TOF LC-MS/MS. A total of 118 and 152 unique peptides were identified, respectively, from the quiescent and activated LX-2 cells, and these peptides represented 54 and 77 proteins, respectively ([Figure 1](#molecules-19-19845-f001){ref-type="fig"}B,C). Then, the percentage amounts by which emPAI-based protein abundance estimates deviated from the actual measured values were calculated to permit the analysis and comparison of protein expression between the activated and quiescent LX-2 cells. As a result, four proteins (calreticulin \[CALR\], histone H4, galectin \[LGALS\]1, and isoform 1 of nucleoside diphosphate kinase A) were upregulated (fold-change \> 2), and two proteins (14-3-3 protein ζ/δ and protein disulfide-isomerase) were downregulated in the activated LX-2 cells (fold-change \< 0.5). In addition, 28 proteins (including 14-3-3 protein ε, protease serine 4 isoform B, and 14-3-3 protein γ) were specifically identified in the quiescent LX-2 cells and 51 proteins (including isoform 2 of procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2, POTE ankyrin domain family member F, and vimentin) were specifically identified in the activated LX-2 cells. In total, by mapping to the Uniprot database, 13 (12.4%) of the identified proteins were classified as unreviewed proteins, 15 proteins (14.3%) were classified as glycoproteins, and 77 proteins (73.3%) were classified as unknown proteins ([Table 1](#molecules-19-19845-t001){ref-type="table"}). According to two online glycosylation site prediction servers (NetNGlyc 1.0 \[[@B33-molecules-19-19845]\] and NetOGlyc 4.0 \[[@B34-molecules-19-19845]\]), 62 of the 77 "unknown" proteins were predicted to have potential N-glycosylation sites, and 15 proteins were predicted to have potential O-glycosylation sites. There were still three proteins that were not identified as glycoproteins by the online servers ([Table 1](#molecules-19-19845-t001){ref-type="table"}). Lectins are defined as carbohydrate-binding proteins that are neither antibodies nor enzymes, which have a wide range of glycan-binding specificities, and are therefore suitable for the partial isolation and characterization of a glycome. ConA is a lectin originally extracted from the jack-bean *Canavalia ensiformis*, which binds specifically to internal and nonreducing terminal α-[d]{.smallcaps}-mannosyl (α-Man) and α-[d]{.smallcaps}-glucosyl groups (α-Glu) \[[@B35-molecules-19-19845]\]. In this study, two possible consequences of the affinity attachment of CBGs to ConA were considered. First, total proteins were extracted under non-denaturing conditions; therefore, some non-target proteins that interacted with CBGs might have been pulled down together. Second, ConA is known to be a glycoprotein containing N-glycans linked to the 152th amino acid (Asn) for each sub-unit \[[@B36-molecules-19-19845]\], and therefore, it was inevitable that cellular glycan-binding proteins with an affinity for the glycans of ConA might also have been collected by CMPCs. These two factors, together with the addition of the possible unusual O-glycosylations (e.g., O-mannosylation and O-GlcNAc) of the CBGs may explain the identification of some "non-glycoproteins" by MS ([Table 1](#molecules-19-19845-t001){ref-type="table"}). To solve those two problems as effectively as possible, denaturing rinses were conducted several times with 0.1% (v/v) Tween^®^ 20, and specific competitive elution was performed with 0.5 M methyl-α-[d]{.smallcaps}-mannose and 0.2 M methyl-α-[d]{.smallcaps}-glucose. molecules-19-19845-t001_Table 1 ###### Detailed information about the CBGs (ConA-binding glycoproteins) isolated by CMPCs (ConA-magnetic particle conjugates) from activated and quiescent LX-2 cells. No. Accession Description Gene Name Search Score Coverage % A/Q Ratio ^a^ Q or A ^b^ Known ^c^ ----- -------------------- --------------------------------------------------------------------------------------- ---------------- -------------- ------------ --------------- ------------ ----------- 1 IPI00003865 Isoform 1 of Heat shock cognate 71 kDa protein HSPA8 375 26.8 --- Q, A P^N^ 2 IPI00022434 Putative uncharacterized protein ALB ALB 260 6.5 --- Q, A Y^N^ 3 IPI00021263 14-3-3 protein zeta/delta YWHAZ 164 15.1 0.16 Q, A P^N^ 4 IPI00010796 Protein disulfide-isomerase P4HB 109 17.5 0.5 Q, A P^N^ 5 IPI00011134 Putative heat shock 70 kDa protein 7 HSPA7 96 6.5 --- Q, A P^N^ 6 IPI00220327 Keratin, type II cytoskeletal 1 KRT1 92 3.4 --- Q, A P^N^ 7 IPI00009865 Keratin, type I cytoskeletal 10 KRT10 75 14.6 --- Q, A P^N^ 8 IPI00217966 Isoform 1 of L-lactate dehydrogenase A chain LDHA 68 4.5 --- Q, A P^N^ 9 IPI00021439 Actin, cytoplasmic 1 ACTB 65 17.9 --- Q, A P^N^ 10 IPI01969230 LOC644914; H3F3B; LOC440926; H3F3A Histone H3;xx Null 60 8.9 --- Q, A Null 11 IPI00418471 Vimentin VIM 56 2.1 --- Q, A Y° 12 IPI00020599 Calreticulin CALR 53 10.8 2.44 Q, A Y^N^ 13 IPI00183968 tropomyosin alpha-3 chain isoform 1 TPM3 53 7.4 --- Q, A P^N^ 14 IPI00218820 Isoform 3 of Tropomyosin beta chain TPM2 53 8.9 --- Q, A P^N^ 15 IPI00293665 Keratin, type II cytoskeletal 6B KRT6B 49 4.4 --- Q, A P^N^ 16 IPI00453473 Histone H4 HIST1H4H 36 29.1 3.13 Q, A P° 17 IPI00003269 Beta-actin-like protein 2 ACTBL2 32 6.6 --- Q, A P^N^ 18 IPI00216783 ubiquitin carboxyl-terminal hydrolase 2 isoform b USP2 32 2.3 --- Q, A P° 19 IPI00219219 Galectin-1 LGALS1 29 8.9 2.8 Q, A P^N^ 20 IPI00009636 Membrane-spanning 4-domains subfamily A member 7 MS4A7 28 7.1 --- Q, A N 21 IPI00375617 Isoform 2 of Abhydrolase domain-containing protein 12B ABHD12B 26 2.1 --- Q, A P° 22 IPI00012048 Isoform 1 of Nucleoside diphosphate kinase A NME1-NME2 24 11.2 2.73 Q, A N 23 IPI00005685 Paraneoplastic antigen Ma1 PNMA1 22 3.7 --- Q, A P^N^ 24 IPI00022774 Transitional endoplasmic reticulum ATPase VCP 22 3 --- Q, A P^N^ 25 IPI00470859 Putative uncharacterized protein DKFZp686C04126 MAN1B1 22 0.8 --- Q, A P° 26 IPI00888712 actin, beta-like 3 POTEE 20 2.6 --- Q, A P^N^ 27 IPI00000816 14-3-3 protein epsilon YWHAE 116 14.5 0.01 Q P^N^ 28 IPI00385250 Protease serine 4 isoform B PRSS3 75 9.6 0.01 Q P^N^ 29 IPI00013508 Alpha-actinin-1 ACTN1 61 1.3 0.01 Q P^N^ 30 [IPI00451401]{.ul} Triosephosphate isomerase TPI1 17 7.4 0.01 Q P^N^ 31 IPI00220642 14-3-3 protein gamma YWHAG 58 15 0.01 Q P^N^ 32 IPI00465028 TPI1 triosephosphate isomerase isoform 2 TPI1P1 56 18.9 0.01 Q P^N^ 33 IPI00479722 Proteasome activator complex subunit 1 PSME1 56 4.4 0.01 Q P^N^ 34 IPI00453476 29 kDa protein Null 37 15.7 0.01 Q Null 35 IPI00021304 Keratin, type II cytoskeletal 2 epidermal KRT2 33 4.5 0.01 Q P^N^ 36 IPI00008527 60S acidic ribosomal protein P1 RPLP1 32 14 0.01 Q P^N^ 37 IPI00021428 Actin, alpha skeletal muscle ACTA1 26 15.6 0.01 Q P^N^ 38 IPI00479743 Isoform 1 of POTE ankyrin domain family member E POTEKP 26 5.8 0.01 Q P^N^ 39 IPI00009791 Isoform WB of Plasma membrane calcium-transporting ATPase 2 ATP2B2 25 0.7 0.01 Q P^N^ 40 IPI00014845 Isoform 1 of Dynein heavy chain 8, axonemal DNAH8 25 0.9 0.01 Q P^N^ 41 IPI00886949 LOC100129520, similar to hCG2044193 Null 25 4.6 0.01 Q Null 42 IPI00002850 Hepatic leukemia factor HLF 24 4.1 0.01 Q P^N^ 43 IPI00550020 Parathymosin PTMS 24 10.8 0.01 Q P^N^ 44 IPI00064885 Zinc finger protein 3 homolog ZFP3 23 1.8 0.01 Q P^N^ 45 IPI00925740 ECT2 Protein ECT2 20 47.8 0.01 Q P^N^ 46 IPI00002349 Nuclear fragile X mental retardation-interacting protein 2 NUFIP2 19 2.6 0.01 Q P^N^ 47 IPI00218667 Stathmin-2 STMN2 19 5 0.01 Q P^N^ 48 IPI00018146 14-3-3 protein theta YWHAQ 18 9.8 0.01 Q P^N^ 49 IPI00156689 Synaptic vesicle membrane protein VAT-1 homolog VAT1 18 6.4 0.01 Q P^N^ 50 IPI00457114 Isoform 1 of IQ motif and SEC7 domain-containing protein 1 IQSEC1 18 3.7 0.01 Q P^N^ 51 IPI00797750 11 kDa protein Null 17 16.7 0.01 Q Null 52 IPI00938247 LOC100287408,hypothetical protein XP_002344194 Null 17 8.8 0.01 Q Null 53 IPI00966637 Putative uncharacterized protein SDAD1 SDAD1 17 23.9 0.01 Q P^N^ 54 IPI00937642 LOC100290701 hypothetical protein XP_002347764 Null 15 5.6 0.01 Q Null 55 IPI00304925 Heat shock 70 kDa protein 1A/1B HSPA1A; HSPA1B 85 9.2 100 A P^N^ 56 IPI00830052 62 kDa protein Null 85 10.5 100 A Null 57 IPI00015614 Isoform A of Trypsin-3 PRSS3 78 6.6 100 A P° 58 IPI00337495 Isoform 2 of Procollagen-lysine,2-oxoglutarate 5-dioxygenase 2 PLOD2 73 1.8 100 A Y^N^ 59 IPI00739539 POTE ankyrin domain family member F POTEKP 63 3.6 100 A P^N^ 60 IPI00000230 tropomyosin alpha-1 chain isoform 2 TPM1 59 7.7 100 A P° 61 IPI00013991 Isoform 1 of Tropomyosin beta chain TPM2 59 8.8 100 A P° 62 IPI00026260 Isoform 1 of Nucleoside diphosphate kinase B NME1-NME2 59 25 100 A N 63 IPI00879437 Protein disulfide-isomerase A1 P4HB 54 16.5 100 A P° 64 IPI00937995 Actin-like protein (Fragment) ACTB 50 19.4 100 A P° 65 IPI00024320 Putative RNA-binding protein 3 RBM3 40 20.4 100 A P° 66 IPI00171611 Histone H3.2 HIST2H3A 40 11.8 100 A P^N^ 67 IPI00798360 SARNP 18 kDa protein SARNP 40 4.3 100 A P^N^ 68 IPI00878173 cDNA FLJ39583 fis, clone SKMUS2004897, highly similar to ACTIN, ALPHA SKELETAL MUSCLE ACTA1 38 15.2 100 A P° 69 IPI00045396 Calumenin isoform 4 CALU 34 2.2 100 A Y^N^ 70 IPI00016768 L-lactate dehydrogenase A-like 6B LDHAL6B 31 4.5 100 A P^N^ 71 IPI00784327 Isoform 1 of 1-phosphatidylinositol-4,5-bisphosphate phosphodiesterase beta-2 PLCB2 30 1.9 100 A P^N^ 72 IPI00000874 Peroxiredoxin-1 PRDX1 29 10.1 100 A P^N^ 73 IPI00216817 HEAT repeat containing 7B1 HEATR7B1 29 1.3 100 A P^N^ 74 IPI00293975 16 kDa protein Null 29 6.2 100 A Null 75 IPI00298622 Intestinal-type alkaline phosphatase ALPI 29 2.1 100 A Y^N^ 76 IPI00945706 BOC Protein BOC 29 14.3 100 A P^N^ 77 IPI00329801 Annexin A5 ANXA5 26 2.8 100 A P° 78 IPI00014424 Elongation factor 1-alpha 2 EEF1A2 25 3.9 100 A P^N^ 79 IPI00030282 Isoform 1 of Filensin BFSP1 25 3.9 100 A P° 80 IPI00910689 Polyserase-2 PRSS36 25 3.2 100 A Y^N^ 81 IPI00023673 Galectin-3-binding protein LGALS3BP 24 3.8 100 A Y^N^ 82 IPI00414717 Isoform 2 of Golgi apparatus protein 1 GLG1 24 2.1 100 A Y^N^ 83 IPI00007249 Ectonucleotide pyrophosphatase/phosphodiesterase family member 4 ENPP4 23 5.7 100 A Y^N^ 84 IPI00008692 Isoform 1 of Keratin, type I cuticular Ha6 KRT36 23 4.9 100 A P^N^ 85 IPI00290077 Keratin, type I cytoskeletal 15 KRT15 23 4.4 100 A P^N^ 86 IPI00384497 Protein-tyrosine phosphatase-like member B PTPLB 23 7.1 100 A Y^N^ 87 IPI00908888 cDNA FLJ57836, highly similar to Myb-binding protein 1A MYBBP1A 23 2.2 100 A P^N^ 88 IPI00024658 OTU domain-containing protein 7B OTUD7B 22 2.5 100 A P° 89 IPI00384972 Isoform 1 of MLL1/MLL complex subunit KIAA1267 KIAA1267 22 1.5 100 A P^N^ 90 IPI00394814 Serine protease 55 PRSS55 22 4.8 100 A Y^N^ 91 IPI00216704 Isoform 2 of Spectrin beta chain, erythrocyte SPTB I 21 1.7 100 A P^N^ 92 IPI00455675 Centrosomal protein of 192 kDa CEP192 20 1.1 100 A P^N^ 93 IPI00552749 DNAH8 478 kDa protein DNAH8 20 1.7 100 A P^N^ 94 IPI00748715 SEPT9 protein (Fragment) Null 19 3.5 100 A Null 95 IPI00020035 Protein NipSnap homolog 3B NIPSNAP3B 18 12.6 100 A P° 96 IPI00217899 E3 ubiquitin-protein ligase RNF168 RNF168 18 1.9 100 A P^N^ 97 IPI00010289 D(1A) dopamine receptor DRD1 17 4 100 A Y^N^ 98 IPI00478586 Isoform 2 of Vacuolar protein sorting-associated protein 13A VPS13A 17 1.9 100 A P^N^ 99 IPI00929137 Conserved hypothetical protein Null 17 9.9 100 A Null 100 IPI00011500 Isoform 1 of Testicular acid phosphatase ACPT 16 2.8 100 A Y^N^ 101 IPI00299507 Condensin complex subunit 2 NCAPH 16 0.9 100 A P^N^ 102 IPI00009101 Isoform 2 of Helicase SRCAP SRCAP 15 1 100 A P^N^ 103 IPI00442299 Isoform 1 of Neurexin-1-alpha NRXN1 15 3.3 100 A Y^N^ 104 IPI00947233 PCOLCE2 Protein PCOLCE2 14 42.6 100 A P^N^ 105 IPI00294052 Single-stranded DNA-binding protein 2 SSBP2 13 3.6 100 A P^N^ ^a^: Glycoproteins with an emPAI ratio of \>2.0 or \<0.5 are shown. "A/Q" represents the ratio of the emPAI value of a protein in the activated LX-2 cells *vs.* that in the quiescent LX-2 cells. The "A/Q" value of glycoproteins identified specifically in quiescent LX-2 was assigned as 0.01, while that of glycoproteins identified specifically in the activated LX-2 was assigned as 100; ^b^: Data in column "Q or A" represent the CBG identified in the quiescent LX-2 (Q) or the activated LX-2 (A); ^c^: "Y^N^" represents the CBGs annotated as "N-linked glycosylated" in Swiss-Prot; "Y°" represents the CBGs annotated as "O-linked glycosylated" in Swiss-Prot; "P^N^" represents potential N-linked glycoproteins predicted by the software NetNGlyc 1.0 Server; "P°" represents potential O-linked glycoproteins predicted by the software NetOGlyc 4.0 Server; "N" represents proteins with no typical glycosylation site. 2.3. GO Classification of the Identified CBGs Using Blast2GO^®^ --------------------------------------------------------------- To investigate the major biological functions of CBGs in LX-2 cells, the commercially available software program Blast2GO was used to analyze the proteins that were identified in LX-2 cells for functional enrichment according to three grouping classifications: cellular components, biological processes, and molecular functions ([Figure 2](#molecules-19-19845-f002){ref-type="fig"}A). Of 105 identified CBGs, 98 had gene ontology (GO) annotations available. In terms of cellular components, 62 proteins (30.1%) were grouped as cell proteins, while 58 proteins (28.2%) were grouped as organelle proteins. Moreover, other protein groupings were found, with groups including: cell membrane (26, 12.6%), macromolecular complex (26, 12.6%), membrane-enclosed lumen (14, 6.8%), and extracellular region (10, 4.9%). In terms of biological processes, about 48 proteins (16.3%), 44 proteins (15.0%), 31 proteins (10.5%), 29 proteins (9.9%), and 27 proteins (9.2%) were included in the groups of cellular process, single-organism process, biological regulation, metabolic process, and response to stimulus, respectively. In terms of molecular function, proteins with binding ability formed the largest group (30, 43.9%) and other smaller groups identified included catalytic activity (14, 21.9%), structural molecule activity (11, 17.2%), and receptor activity (3, 4.7%). Detailed analysis of the binding ability group revealed that proteins with the ability to bind to proteins (21, 47.7%) and ions (8, 18.2%) comprised the most abundant subgroups. To determine whether there were any differences in the three aspects of GO annotations between activated and quiescent LX-2 cells, GO enrichment analysis was performed. Greater enrichment was found in the response to stimulus, detection of stimulus, activation of immune response, and gene silencing groups, while enrichment of the translation initiation and cell cycle process groups was much lesser in activated LX-2 cells than in the quiescent control cells ([Figure 2](#molecules-19-19845-f002){ref-type="fig"}B). In our previous study, we found large differences in the distribution of glycoproteins bound among ConA and other lectins in LX-2 cells. The glycoproteins recognized by *Aleuria aurantia* lectin (AAL), *Erythrina cristagalli* lectin (ECA), and phytohemagglutinin (PHA-E) were mainly located on the cytoplasmic membrane and the perinuclear cytoplasm (*i.e.*, the endoplasmic reticulum \[ER\] and Golgi complex), while ConA showed positive binding to the glycoproteins located in the whole cytoplasm and/or cytoplasmic membrane \[[@B18-molecules-19-19845]\]. In the present study, 62 proteins (30.1%) were grouped as intracellular proteins and 58 (28.2%) as organelle proteins. Most of the proteins in the organelle proteins group were also included in the intracellular proteins group. It is known that the common protein-bound glycan adduct, N-linked core oligosaccharide, is composed of two N-acetylglucosamine, nine mannose, and three glucose residues, and this oligosaccharide molecule is synthesized and processed in the ER \[[@B37-molecules-19-19845]\]. Glycans are subjected to extensive modifications as glycoproteins mature and move through the ER via the Golgi complex to their final destinations inside and outside the cell \[[@B38-molecules-19-19845]\]. Con A can specifically recognize the trimannoside core of the N-glycan \[[@B39-molecules-19-19845]\], so N-glycosylated proteins that are still undergoing processing and sorting in the ER and Golgi can also be recognized by ConA, in addition to the mature membrane and secreted glycoproteins. For some organelle proteins, such as Golgi glycoprotein 1 (GLG1) ([Table 1](#molecules-19-19845-t001){ref-type="table"}) and lysosomal enzymes, their synthesis and positioning also depend on the glycosylation associated with ER-Golgi trafficking \[[@B40-molecules-19-19845],[@B41-molecules-19-19845]\]. In addition, O-glycosylations such as O-GlcNAc, which contribute to nutrient sensing and regulation of insulin signaling, cell cycle, calcium signaling pathway, and cellular stress response \[[@B42-molecules-19-19845],[@B43-molecules-19-19845],[@B44-molecules-19-19845]\], may contain multiple mannose residues that can also be recognized by ConA, since CBGs were distributed not only in the perinuclear cytoplasm of the LX-2 cells, but also in the whole cytoplasm and cell membrane. {#molecules-19-19845-f002} 2.4. KEGG Pathway Analysis and Functional Protein Association Networks ---------------------------------------------------------------------- In total, 91 of 105 identified CBGs had been annotated in DAVID Bioinformatics Resources version 6.7. These glycoproteins were mapped to 6 KEGG pathways with the thresholds of count of \>2 and *p* value of \<0.05 compared with the background signal of the human genome; the identified KEGG pathways included protein processing in the ER, calcium signaling pathway, cell cycle, glycolysis/gluconeogenesis, and others ([Figure 3](#molecules-19-19845-f003){ref-type="fig"}A,B, [Table S1](#app1-molecules-19-19845){ref-type="app"}). Proteins involved in protein processing in the ER (e.g., CALR, protein disulfide-isomerase A1, and heat shock 70-kDa protein 1A/1B) and calcium signaling pathway (e.g., D\[1A\] dopamine receptor \[DRD1\] and 1-phosphatidylinositol-4,5-bisphosphate phosphodiesterase β-2) were specifically identified or upregulated in the activated LX-2 cells. In contrast, 14-3-3 protein family members (e.g., 14-3-3 ζ/δ, 14-3-3 ε, and 14-3-3 γ) involved in the cell cycle and the neurotrophin signaling pathway were specifically identified in the quiescent LX-2 cells. ###### KEGG pathway analysis and functional protein association networks. (**A**,**B**) show the CBGs mapped to the KEGG pathways of protein processing in the ER and calcium signaling. The CBGs involved in these networks are labeled with a red frame; (**C**,**D**) display the potential interactions among total CBGs and show the CBGs that were determined to show significant correlations by STRING analysis.    A total of 90 matched CBGs were queried against the STRING *Homo sapiens* database to determine their functional relevance ([Figure 3](#molecules-19-19845-f003){ref-type="fig"}C). Through K-mean clustering analysis, associations among TPM1, TPM2, ACTB, DNAH8, and ACTA1; LGALS1, ANXA5, and LGALS3BP; P4HB and CALR; and HSPA1B and HSPA1A were specifically identified in the activated LX-2 cells, whereas associations among YWHAZ, YWHAE, YWHAG, and YWHAQ were specifically identified in the quiescent LX-2 cells ([Figure 3](#molecules-19-19845-f003){ref-type="fig"}D). The aims of this study were not only to find novel CBGs that differentially expressed in the activated HSCs, but also to speculate the possible pathway networks associated with fibrogenesis in HSCs. Proteins involved in protein processing in ER and calcium signaling pathway were higher expressed in the activated HSCs ([Figure 3](#molecules-19-19845-f003){ref-type="fig"}A,B, and [Table S1](#app1-molecules-19-19845){ref-type="app"}), which partially demonstrated that these pathways were activated in HSCs when stimulated by TGF-β1. Interestingly, the expression levels of galectin-1 (LGALS1) and galectin-3-binding protein (LGALS3BP) were upregulated or specifically identified in the activated LX-2 cells ([Table 1](#molecules-19-19845-t001){ref-type="table"}). The functions of galectins have been reported to be involved in physiological and pathological processes of the liver \[[@B45-molecules-19-19845],[@B46-molecules-19-19845]\]. A previous proteomics analysis of rat HSC proteins revealed that the production and secretion of LGALS1 is greatly increased in activated HSCs compared to that in quiescent HSCs \[[@B45-molecules-19-19845]\]. LGALS3 expression was found to be induced in regenerative nodules of liver cirrhosis tissues and in hepatocellular carcinomas \[[@B47-molecules-19-19845]\]. A further study demonstrated that both LGALS1 and LGALS3 activate mitogen-activated protein kinase (MAPK) pathways, presumably by forming cross-links with target molecules through their β-galactoside-containing glycoconjugates, leading to the proliferation of HSCs \[[@B48-molecules-19-19845]\]. In addition, an increased concentration of cytoplasm Ca^2+^ can also activate the Ca^2+^/calmodulin-dependent protein kinase (CaMKII)/MAPK signaling pathway \[[@B49-molecules-19-19845]\]. Intracellular free Ca^2+^ is a crucial second messenger that plays various roles in regulating a wide range of cellular processes in different cells and tissues. To maintain a low Ca^2+^ concentration of 10--100 nM in the cytoplasm, Ca^2+^ is actively pumped from the cytoplasm to the extracellular space and ER, and sometimes into the mitochondria. Specific signals can trigger an increase in the cytoplasmic Ca^2+^ level up to 500--1000 nM by opening channels in the plasma membrane or ER \[[@B50-molecules-19-19845]\]. DRD1 is a G protein-coupled receptor that stimulates adenylate cyclase and was found to inhibit Ca^2+^ from being pumped from the cytosol to the ER \[[@B51-molecules-19-19845]\]. The plasma membrane Ca^2+^ ATPase (ATP2B) is a transport protein in the plasma membrane of cells that serves to remove Ca^2+^ from the cytosol \[[@B52-molecules-19-19845]\]. 1-phosphatidylinositol-4,5-bisphosphate phosphodiesterase β-2 (PLCB2) participates in phosphatidylinositol 4,5-bisphosphate (PIP2) metabolism and indirectly promotes the release of Ca^2+^ from the ER to the cytosol \[[@B53-molecules-19-19845]\]. According to our results, DRD1 and PLCB2 were specifically identified in the activated LX-2 cells, whereas ATP2B was specifically identified in the quiescent LX-2 cells, indicating that Ca^2+^ became more inclined to accumulation in the cytoplasm, rather than the extracellular space or the ER in the LX-2 cells activated with TGF-β1 ([Figure 3](#molecules-19-19845-f003){ref-type="fig"}B). Thus, it can be conjectured that increases in galectin expression and intracellular Ca^2+^ concentration might collectively activate the MAPK signaling pathway to realize the activation of HSC induced by TGF-β1. This view is different from the traditional view that platelet derived growth factor (PDGF) induced the activation of HSCs through MAPK signaling pathway \[[@B54-molecules-19-19845]\] and TGF-β1 through TGFβ1/ALK1/Smad1 pathway, and might be therapeutic targets of liver fibrosis. Actually, a recent study suggested that the production of CTGF by TGF-β activated MAPK signaling in hepatic progenitor cells \[[@B55-molecules-19-19845]\]. However, the detailed molecular mechanisms of these cytokines still need to be further studied. 2.5. Expression of PLCB2 in Activated LX-2 Cells and Human Liver Cirrhosis Tissue --------------------------------------------------------------------------------- To further validate and investigate the expression of CBGs and their relationship to liver fibrosis and cirrhosis, the increased expression of PLCB2, which was specifically identified in the activated LX-2 cells, was measured in LX-2 cells and human liver by western blotting and immunohistochemistry, respectively. The expression of PLCB2 was found to be distinctly increased in activated, compared with quiescent LX-2 cells ([Figure 4](#molecules-19-19845-f004){ref-type="fig"}A,B) and was also increased in liver cirrhosis tissues compared with normal para-carcinoma tissues ([Figure 4](#molecules-19-19845-f004){ref-type="fig"}C). PLCB2 was found to be mainly distributed in the cytoplasm of the activated LX-2 cells ([Figure 4](#molecules-19-19845-f004){ref-type="fig"}B), as well as that of hepatocytes and sinusoidal cells in liver tissues ([Figure 4](#molecules-19-19845-f004){ref-type="fig"}C). These results were consistent with the MS analysis and provided further evidence indicating a possible vital function of PLCB2 in the development of liver fibrosis and cirrhosis. {#molecules-19-19845-f004} In this study, 62 proteins (30.1%) were intracellular proteins and 26 (12.6%) were located on the cell membrane, in accordance with the result that ConA showed strong binding to the cytoplasm and/or membrane of LX-2 cells in our previous study \[[@B18-molecules-19-19845]\]. The activation of HSCs is a complex and persistent response to chronic liver injury. In the initial stages, HSCs receive paracrine stimulation from cytokines secreted by neighboring cells such as Kupffer cells, hepatocytes, and leukocytes. In the persistent stages, autocrine cytokines (e.g., TGF-β and connective tissue growth factor) persistently stimulate HSCs so as to regulate and maintain their activation \[[@B56-molecules-19-19845],[@B57-molecules-19-19845]\]. In this study, a greater number of CBGs were identified from the activated LX-2 cells than from the quiescent LX-2 cells. In addition, glycoproteins involved in the biological processes of response to stimulus, detection of stimulus, and activation of immune response were upregulated in the activated LX-2 cells. For example, PLCB2 was validated to be upregulated in the cytoplasm of the activated LX-2 cells, as well as in the hepatocytes and sinusoidal cells of liver cirrhosis tissues. Based upon the above findings, CBGs might play important roles in signal conduction and response to stimulus, and be associated with the formation and development of liver fibrosis and cirrhosis. 3. Experimental Section ======================= 3.1. Materials -------------- Dulbecco's modified Eagle's medium (DMEM) and fetal bovine serum (FBS) were purchased from Invitrogen (Carlsbad, CA, USA). TGF-β1 was purchased from R&D systems (Minneapolis, MN, USA). Con A was purchased from Vector Laboratories (Burlingame, CA, USA). DTT and iodoacetamide were purchased from GE Healthcare (Little Chalfont, Buckinghamshire, UK). Mouse anti-PLCB2 monoclonal antibody, urea, ammonia bicarbonate (NH~4~HCO~3~), sodium periodate (NaIO~4~), HPLC-grade acetonitrile (ACN) and trifluoroacetic acid (TFA), protease inhibitor cocktail, and Bradford assay reagent were purchased from Sigma-Aldrich (St. Louis, MO, USA). Trypsin was purchased from Promega (Madison, WI, USA). Peptide-N-glycosidase F (PNGase F) was purchased from New England Biolabs (Ipswich, MA, USA). Bovine serum albumin (BSA) was purchased from Merck (Darmstadt, Germany). T-PER^®^ reagent was purchased from Pierce Biotechnology (Rockford, IL, USA). Other chemical reagents were obtained from commercial suppliers and used without further purification. All of the solutions were prepared with ultra-pure water obtained from a Milli-Q^®^ 50 SP Reagent Water System (Millipore, Bradford, MA, USA). Amicon Ultra-0.5 10-kDa devices were obtained from Millipore. C18 SepPak^®^ columns were from Waters (Milford, CT, USA). The ZHWY- 2101C Oscillator was purchased from ZhiCheng Corporation, China. The SpeedVac^TM^ vacuum concentrator was obtained from Thermo Scientific (Waltham, MA, USA). Nanospray CHIP-LC-MS/MS 6530 mass spectrometers equipped with HP 1200 solvent delivery systems were purchased from Agilent (Santa Clara, CA, USA). 3.2. Cell Culture and Stimulation --------------------------------- The LX-2 cells were a gift from Dr. Friedman of the Mount Sinai School of Medicine, New York, NY, USA. Cells were cultured and stimulated as described previously \[[@B18-molecules-19-19845],[@B58-molecules-19-19845]\]. About 2 × 10^5^ LX-2 cells were seeded and rested in high-glucose DMEM supplemented with 10% (v/v) FBS for 24 h. After 24 h of starvation, cells were stimulated by adding 2 ng/mL TGF-β1 for 24 h. The control (quiescent) groups were not treated with TGF-β1. 3.3. Non-Denaturing Extraction of Total Proteins ------------------------------------------------ Total protein was extracted from LX-2 cells using non-denaturing T-PER reagent as described previously \[[@B18-molecules-19-19845]\]. The adherent cells were washed twice with ice-cold 1× PBS (0.01 M phosphate buffer containing 0.15 M NaCl, pH 7.4) to remove culture medium. The cells were incubated for 15 min with T-PER reagent mixed with protease inhibitors (10 μL/mL T-PER reagent) and then homogenization was achieved by triturating cells with a pipette. Following centrifugation at 10,000× *g* for 10 min, the supernatants were either immediately collected for use or stored at −80 °C. The protein concentration in the supernatants was determined by Bradford assay. 3.4. Isolation of Glycoproteins from LX-2 Cells Using ConA-Magnetic Particle Conjugates (CMPCs) ----------------------------------------------------------------------------------------------- A subset of glycoproteins binding to ConA were selectively isolated by CMPCs as described previously \[[@B59-molecules-19-19845],[@B60-molecules-19-19845]\]. Briefly, 2 mg of total proteins from LX-2 cells were diluted into 600 µL of binding buffer (0.1 M Tris-HCl, 150 mM NaCl, 1 mM CaCl~2~, 1 mM MgCl~2,~ and 1 mM MnCl~2~, pH 7.4) supplemented with 1% (v/v) proteinase inhibitor. The homemade ConA-magnetic particle conjugates (CMPCs) were rinsed three times with binding buffer and then incubated with proteins at room temperature for 3 h under gentle shaking. The unbound proteins were removed by washing three times with the rinsing buffer (binding buffer with the addition of 0.1% (v/v) Tween 20) thoroughly. Then, the glycoproteins bound to CMPCs were eluted with 300 µL of the elution buffer (binding buffer supplemented with 0.5 M methyl-α-[d]{.smallcaps}-mannose, 0.2 M methyl-α-[d]{.smallcaps}-glucose, and 1% (v/v) proteinase inhibitor) at room temperature for 1 h under gentle shaking, and the supernatants were collected. 3.5. Identification of Peptides by LC-MS/MS ------------------------------------------- The glycoproteins isolated by CMPCs were concentrated and desalted by a size-exclusion spin filter (Amicon Ultra-0.5 10 K device), with a molecular mass cutoff of 10 kDa. The obtained glycoproteins were denatured in 8 M urea, then deoxidated with 10 mM DTT and carboxyamidomethylated with 20 mM iodoacetamide. Subsequently, trypsin was added at a 1:100 (w/w) ratio of enzyme to protein and the samples were incubated overnight at 37 °C. Glacial acetic acid (5 µL, pH 2.0) was added to stop the reaction, samples were centrifuged, and the supernatants were collected and lyophilized. Then, the N-linked glycans of the glycopeptides were released by incubation with PNGase F overnight at 37 °C. The peptides were desalted using C18 SepPak columns and lyophilized. Finally, the peptides were resolubilized in 0.1% (v/v) formic acid and analyzed using nanospray CHIP-LC-MS/MS 6530 mass spectrometers equipped with HP 1200 solvent delivery systems. 3.6. Label-Free Relative Quantification of Identified Glycoproteins by emPAI ---------------------------------------------------------------------------- Label-free quantification methods are popular because of their convenience and accuracy. Mascot search engine software (Matrix Science, Boston, MA, USA) includes a semi-quantification method, emPAI, which is a convenient tool for relative quantification \[[@B61-molecules-19-19845]\]. To compare the activated *versus* quiescent LX-2 glycoproteomes, we calculated normalized emPAI values for each protein (see Formula (1)), and estimated the fold-change in the activated *versus* quiescent LX-2 cells. When a protein was not identified in the quiescent LX-2 cells, we arbitrarily assigned the fold-change as 100 to avoid dividing by zero. 3.7. Data Mining and Bioinformatics ----------------------------------- Ontology analysis was performed according to the standard procedure of Blast2GO \[[@B62-molecules-19-19845]\] to gain insights into functional groupings within the set of identified glycoproteins in three aspects of GO. The differences in the GO terms of glycoproteins between the activated and quiescent LX-2 cells were also discovered by analysis using WEGO \[[@B63-molecules-19-19845]\]. Biological pathways that were enriched in the identified glycoproteins originated from the KEGG human pathway database \[[@B64-molecules-19-19845]\]. To validate the results of Blast2GO and KEGG, an independent functional enrichment analysis was performed using DAVID (Database for Annotation, Visualization and Integrated Discovery) \[[@B65-molecules-19-19845]\]. Finally, to determine the functional relevance of the identified glycoproteins, we performed an analysis of protein association networks using STRING 9.05 \[[@B66-molecules-19-19845]\]. 3.8. SDS-PAGE and Western Blotting ---------------------------------- For SDS-PAGE, protein samples were mixed with 5× loading buffer and boiled for 4 min at 100 °C, and then separated on a 10% polyacrylamide resolving gel and a 3% stacking gel. Molecular mass standards (Thermo Scientific) were run for each gel. Gels were then stained directly with alkaline silver. For western blotting, 30 μg of proteins were separated by 10% SDS-PAGE and then transferred to PVDF membranes (Millipore). After blocking for 2 h at room temperature with 5% (w/v) skimmed milk powder in TBS (100 mM Tris-HCl, 150 mM NaCl, pH 7.6), the membrane was incubated overnight at 4 °C with mouse monoclonal antibodies against PLCB2 in antibody dilution buffer (5% (w/v) skimmed milk in TBST (0.05% (v/v) Tween 20 in TBS, pH 7.6)). The membranes were washed three times with TBST, and then incubated with alkaline phosphatase-conjugated secondary antibodies (1:1000 (v/v) dilution in antibody dilution buffer) for 1 h at room temperature. The membranes were washed three times with TBST and detected using a 5-bromo-4-chloro-3-indolyl phosphate/nitroblue tetrazolium salt substrate kit. β-Actin was used as an internal control. 3.9. Immunohistochemistry ------------------------- Immunohistochemistry was performed as described previously \[[@B67-molecules-19-19845]\]. Briefly, 2 × 10^5^ LX-2 cells were seeded in 6-well culture plates containing sterile coverslips. The adherent cells were fixed with 4% (w/v) paraformaldehyde and permeabilized in ice-cold 1× PBS supplemented with 1% (v/v) Triton^TM^ X-100 at 4 °C for 10 min and rinsed twice in 1× PBS. The permeabilized cells were then treated with hydrogen peroxide (H~2~O~2~) (3% (v/v) in methanol) for 20 min to block endogenous peroxidase. Prior to incubation with anti-PLCB2, cells were blocked with blocking solution (10% (w/v) BSA in 1× PBS) for 1 h at room temperature. Then, cells were incubated with PLCB2 antibody diluted at 1:1000 in blocking solution overnight at 4 °C. Then, cells were incubated with alkaline phosphatase-conjugated secondary antibodies in blocking solution for 1 h at room temperature, and horseradish peroxidase colorimetric development solution was applied according to the manufacturer's instructions. Finally, the cells were further stained with hematoxylin and histological images were obtained. 4. Conclusions ============== The profile of CBGs in human HSCs (LX-2 cell line) activated with TGF-β1 was compared with that of quiescent HSCs by employing a glycoproteomics approach using a CMPCs coupled MS technique. Novel CBGs and the signaling pathways/networks in which they were involved were discovered and partially validated to be connected with the activation of HSCs. The results of this study may provide useful information to aid future investigation to identify new molecular mechanisms for HSC activation and antifibrotic therapeutic targets. This work was supported by National Natural Science Foundation of China (No. 81401137) and the Fundamental Research Funds for the Central Universities, China (No. XJJ2014069). *Sample Availability*: Samples of the compounds upon request are available from the authors. Supplementary materials can be accessed at: <http://www.mdpi.com/1420-3049/19/12/19845/s1>. ###### Click here for additional data file. Y.Q. carried out enrichment of CBGs by CMPCs, generated the graphs for proteomic data and wrote the manuscript; Y.Z. participated in the culture and stimulation of LX-2 cells; G.Y. carried out MS identification; T.M. performed the preparation of CMPCs; L.J. participated in data analysis; C.H. participated in the revision of the manuscript; and Z.L. participated in the design of the project, coordination and helped to draft the manuscript. All authors read and approved the final manuscript. The authors declare no conflict of interest. [^1]: These authors contributed equally to this work.

INTRODUCTION {#sec1} ============ Porcine reproductive and respiratory syndrome (PRRS) is arguably the most economically important infectious disease affecting pigs worldwide. The causative agent of PRRS is PRRS virus (PRRSV), a member of the Arteriviridae family and the order *Nidovirales*. Infected pigs of all ages may present with symptoms involving inappetence, fever, lethargy, and respiratory distress. However, the most devastating effects of PRRSV infection are observed in young piglets and pregnant sows. In pregnant sows, full abortions or death and mummification of fetuses *in utero* are observed, and live-born piglets from an antenatal infection are often weak and display severe respiratory symptoms ([@B1][@B2][@B3]). Piglets infected with PRRSV in early life can show diarrhea and, more commonly, severe respiratory distress due to active PRRSV replication in pulmonary macrophages and subsequent damage in lung tissues ([@B4]). Due to the reduction or loss of pregnancies, death in young piglets, and decreased growth rates in all PRRSV-infected pigs, it is estimated that the economic impact of PRRSV to pork producers in the United States alone is more than \$650 million annually ([@B5], [@B6]). There are two different species of PRRSV with distinct geographic distributions: PRRSV-1 is found primarily in Europe and Asia, overlapping the range of PRRSV-2, which is found in Asia and the Americas. PRRSV-1 can be further divided into at least three subtypes, currently based on open reading frame 7 (ORF7) sequences and geographical distribution, with subtype 1 being pan-European and subtypes 2 and 3 currently being limited to eastern Europe ([@B7]). PRRSV has a very narrow host cell range, infecting only specific subsets of porcine macrophages ([@B8][@B9][@B10]). Entry of PRRSV into macrophages has been shown to occur via pH-dependent, receptor-mediated endocytosis ([@B11], [@B12]). Various attachment factors and receptors have been indicated to be involved in the PRRSV entry process (reviewed in reference [@B13]). However, only the scavenger receptor CD163, also known as a hemoglobin (Hb)-haptoglobin (Hp) scavenger receptor or p155, has been confirmed to be an essential fusion receptor *in vitro* and *in vivo* ([@B14][@B15][@B16]). CD163 is expressed on specific subtypes of macrophages. The extracellular portion of CD163 forms a pearl-on-a-string structure of nine scavenger receptor cysteine-rich (SRCR) domains and is anchored by a single transmembrane segment and a short cytoplasmic domain ([@B17]). A proteolytically cleaved, soluble form of the protein ectodomain is found in the bloodstream and is involved in the inflammation and ischemic repair response ([@B18], [@B19]). Transmembrane anchoring and interaction with SRCR domain 5 (SRCR5) of CD163 were found to be essential for successful infection with PRRSV ([@B20], [@B21]). CD163 has a variety of biological functions, including mediating systemic inflammation and the removal of hemoglobin from blood plasma (reviewed in references [@B21] and [@B22]). Overexpression of CD163 renders nonsusceptible cells permissive to PRRSV infection ([@B20]), and it was found that CD163 does not mediate internalization but is crucial for fusion ([@B16]). Recent *in vivo* challenge experiments of pigs in which both copies of the CD163 gene had been knocked out using gene-editing technology confirmed that CD163 is required for infection by PRRSV-2 and highly pathogenic PRRSV-2 (HP-PRRSV) ([@B14], [@B23]). Gene-editing technology has also been used to generate pigs in which the CD163 SRCR5-encoding sequence has been replaced with a sequence encoding human CD163L1 SRCR8 ([@B24]), in effect replicating *in vivo* the previous *in vitro* domain-swapping experiment of Van Gorp and colleagues ([@B25]). This attempt to maintain CD163 function rendered pigs and macrophages resistant to PRRSV-1 but not PRRSV-2 infection ([@B24]), making this strategy ineffective in combating both PRRSV species. CD163 has important biological functions, and the complete knockout could have a negative physiological impact on the animal, particularly with respect to the inflammation response and/or infection by other pathogens. Interestingly, whereas all the other eight SRCR domains have been shown to be involved in different biological functions, no specific role has been associated with SRCR5, other than in PRRSV infection ([@B21]). Therefore, we generated pigs lacking SRCR5 by the deletion of exon 7 of *CD163* using CRISPR/Cas9 editing and showed that macrophages from these pigs were resistant to both PRRSV-1 and PRRSV-2 infection *in vitro* ([@B15]). The aim of the experiments described here was to determine whether our *in vitro* results would translate directly *in vivo* by conducting a PRRSV challenge of pigs with a CD163 SRCR5 deletion. Furthermore, we aimed to characterize the biological function of the modified CD163 protein (ΔSRCR5 CD163) as both a soluble and a cell-bound protein. RESULTS {#sec2} ======= Genome editing in zygotes for ΔSRCR5 CD163 pigs and breeding genotypically uniform F2 pigs. {#sec2-1} ------------------------------------------------------------------------------------------- Founder-generation (F0) animals carrying a deletion of exon 7 in the CD163 gene, which encodes SRCR5 of the protein, were generated by CRISPR/Cas9 gene editing as previously described (ΔSRCR5 pigs) ([@B15]). Briefly, zygotes were microinjected with a combination of Cas9 mRNA and guide RNAs targeting sites flanking exon 7, resulting in double-strand breaks (DSBs) and the deletion of the exon ([Fig. 1A](#F1){ref-type="fig"}). Cross-breeding of heterozygous founder animals and outbreeding with wild-type pigs yielded a first generation composed of both heterozygous and biallelic edited animals (F1 generation) ([@B15]). We selected for breeding heterozygous F1 animals displaying a "clean" deletion between the DSBs, i.e., no "on-target" sequence changes beyond the desired deletion region. To generate a cohort with comparable genetic backgrounds, half-sibling heterozygous animals and wild-type animals were bred to yield homozygous ΔSRCR5 animals ([Fig. 1A](#F1){ref-type="fig"}) and wild-type half- and full-sibling animals. {#F1} As described above, ΔSRCR5 animals express the ΔSRCR5 CD163 mRNA and protein at levels equivalent to CD163 transcript and protein levels in wild-type siblings. Furthermore, the native-structure ΔSRCR5 CD163 is recognized on the surface of pulmonary alveolar macrophages (PAMs) by a commercial antibody ([@B15]). We further analyzed whether template-based protein structure analysis by RaptorX predicted the folding of each subdomain compatible with a functional ΔSRCR5 CD163 protein ([@B26]). As demonstrated in [Fig. 1B](#F1){ref-type="fig"}, all subdomains in both full-length and ΔSRCR5 CD163 are predicted to adopt the globular structure and retain the pearl-on-a-string configuration of the native CD163 protein. This supports our findings indicating the proper folding and expression of the ΔSRCR5 protein. Previously, we have shown that PAMs and *in vitro*-differentiated peripheral blood monocytes (PBMCs) are resistant to infection with both PRRSV-1 and PRRSV-2. Now, we aimed to assess the resistance of these pigs to PRRSV-1 infection *in vivo*. We selected four homozygous ΔSRCR5 F2 animals and four wild-type controls that were cohoused from weaning ([Fig. 1C](#F1){ref-type="fig"}). Blood samples were taken from all eight pigs, and a full blood count was conducted by the diagnostics laboratory at the Royal (Dick) School of Veterinary Sciences, University of Edinburgh. The blood counts of all animals were within reference values, indicating good general health and the absence of infection or inflammation. Furthermore, the hemoglobin levels of all animals were within reference values, indicating normal function of the hemoglobin/haptoglobin-scavenging activity of CD163 ([Table 1](#T1){ref-type="table"}). ###### Whole-blood-count results for ΔSRCR5 (animals 4 to 7) and wild-type (animals 8 to 11) piglets at week 6[^*a*^](#T1F1){ref-type="table-fn"} Indicator Value for animal Reference value (range) ---------------------------------------------- ------------------ ------------------------- ------- ------- ------- ------- -------- ------- ------------ WBC count (10E9/liter) 22.5 24 14 15.1 12.4 19.6 26.1 14.4 11--22 Neutrophil count (segmented) (10E9/liter) 5.85 4.8 4.62 5.889 4.34 7.252 7.83 4.32 2--15 \% neutrophils (segmented) 26 20 33 39 35 37 30 30 20--70 Neutrophil count (nonsegmented) (10E9/liter) 0 0 0 0 0 0 0 0 0--0.8 \% neutrophils (nonsegmented) 0 0 0 0 0 0 0 0 0--4 Lymphocyte count (10E9/liter) 15.3 18.72 8.82 8.305 7.564 11.76 16.182 9.36 3.8--16.5 \% lymphocytes 68 78 63 55 61 60 62 65 35--75 Monocyte count (10E9/liter) 0.675 0.48 0.42 0.755 0.496 0.588 1.044 0.576 0--1 \% monocytes 3 2 3 5 4 3 4 4 0--10 Eosinophil count (10E9/liter) 0.675 0 0 0.151 0 0 1.044 0.144 0--1.5 \% eosinophils 3 0 0 1 0 0 4 1 0--15 Basophil count 0 0 0.14 0 0 0 0 0 0--0.5 \% basophils 0 0 1 0 0 0 0 0 0--3 RBC count (10E12/liter) 6.03 6.64 6.99 6.58 6.3 6.53 7.52 6.97 5--9 PCV/hematocrit 0.384 0.391 0.383 0.388 0.382 0.39 0.429 0.421 0.36--0.43 Hb level (g/dl) 11.5 11.9 10.9 11.8 11.6 12 13.8 12.3 10--16 MCV (fl) 63.7 58.9 54.8 58.9 60.7 59.8 57.1 60.5 50--62 MCHC (g/dl) 29.9 30.4 28.3 30.5 30.3 30.9 32.1 29.1 30--36 No. of platelets 219 230 605 397 483 519 219 606 120--720 RDW 20.9 23.1 28.9 20.6 21 18 17 22.6 WBC, white blood cell; RBC, red blood cell; PCV, packed cell volume; Hb, hemoglobin; MCV, mean corpuscular volume; MCHC, mean corpuscular hemoglobin concentration; RDW, red cell distribution width. At 6 weeks of age, a serum sample was collected from all animals prior to movement to a specific-pathogen-free unit (SPFU). The cohort was cohoused for the duration of the experiment and allowed to settle for 1 week prior to the initiation of the challenge. On day 0 of the challenge with PRRSV-1, a second serum sample was taken, and soluble CD163 (sCD163) serum levels were assessed using a commercially available enzyme-linked immunosorbent assay (ELISA) recognizing soluble porcine CD163. Serum CD163 levels were found to be 433.2 ± 69.57 ng/ml in wild-type pigs and 463.5 ± 68.99 ng/ml in ΔSRCR5 pigs ([Fig. 2](#F2){ref-type="fig"}). These results are not significantly different from each other (*P* = 0.7512) and are comparable to sCD163 levels in humans (for example, see reference [@B27]). {#F2} ΔSRCR5 pigs are resistant to PRRSV-1 infection. {#sec2-2} ----------------------------------------------- At 7 to 8 weeks of age, the pigs were inoculated intranasally with PRRSV-1 subtype 2 strain BOR-57 ([@B28]). Eastern European strains are often associated with higher pathogenicity than other PRRSV-1 strains. For BOR-57, we previously observed mild respiratory symptoms, elevated core temperature, extensive lung pathology, and high viral loads in serum. For this study, the strain was selected due to the high viremia and shedding levels expected to occur under study conditions. The experiment was conducted for a period of 14 days following inoculation of each pig on days 0 and 1 with 5E6 50% tissue culture infective doses (TCID~50~), as determined by assessment on PAMs, of the virus isolate used. Rectal temperature, respiration, nasal discharge, coughing, and demeanor were recorded every day, and serum samples were collected on day 0 (prior to challenge) and on days 3, 7, 10, and 14 (prior to euthanasia). Weights were recorded on days 0, 7, and 14 (prior to euthanasia). People assessing the pigs clinically, conducting the challenge, and analyzing pathology were blind to the genotype status of the animals. The rectal temperatures were significantly elevated on days 6 to 9 of the challenge in the wild-type animals, whereas no fever was observed in the ΔSRCR5 animals ([Fig. 3A](#F3){ref-type="fig"}). The average daily weight gain of the ΔSRCR5 pigs was higher than that of their wild-type counterparts over the entire challenge period and significantly higher over days 7 to 14 (*P* = 0.0465) ([Fig. 3B](#F3){ref-type="fig"}). One wild-type pig showed decreased demeanor on days 7 to 8; no respiratory symptoms or other abnormalities in behavior were observed in any of the other animals during the course of the study, as expected for a PRRSV-1 infection at this age. Viral RNA was isolated from serum, virus levels were quantified by using a DNA fragment template standard, and viral RNA was extracted from virus stocks of known infectivity. Whereas the wild-type pigs showed high viremia, no viral RNA was detected in the serum of ΔSRCR5 pigs ([Fig. 3C](#F3){ref-type="fig"}). The presence of antibodies against PRRSV was assessed using a commercial ELISA able to detect antibodies against all PRRSV-1 subtypes and PRRSV-2. PRRSV antibodies were detected in wild-type pigs from day 7 and present at significant levels (according to the manufacturer\'s indicated positive threshold of a sample-to-positive (s/p) ratio of \>0.4) on days 10 and 14 ([Fig. 3D](#F3){ref-type="fig"}) but were not detected in samples from the ΔSRCR5 pigs collected during the course of the experiment. {#F3} During necropsy, lungs were assessed initially, and detailed photographs of the dorsal and ventral sides were taken. Lungs were scored for the presence of lesions. An established scoring system, based on the approximate contribution of each lung section to the complete lung volume, was employed ([@B29]). On average, 61% of the lung surface of control pigs was found to be mottled tan with areas of consolidations, compared to 0.25% of lung surfaces in ΔSRCR5 pigs ([Fig. 3E](#F3){ref-type="fig"} and [G](#F3){ref-type="fig"}). Samples of the lungs were fixed in formalin, embedded in paraffin, cut into sections, and stained for further analysis. To assess general lung histology, samples were stained with hematoxylin and eosin. Sections from each pig were assessed for the presence of interstitial pneumonia on a scale of 0 to 6 (0, normal; 1, mild multifocal; 2, mild diffuse; 3, moderate multifocal; 4, moderate diffuse; 5, severe multifocal; 6, severe diffuse). Microscopic lung lesions characterized by multifocal-to-diffuse interstitial pneumonia with type 2 pneumocytes, hypertrophy, and hyperplasia were observed only in wild-type animals (average score, 4) and were not present in lung sections of ΔSRCR5 pigs ([Fig. 3E](#F3){ref-type="fig"} and [F](#F3){ref-type="fig"}, top). The presence of PRRSV antigens was assessed by immunohistochemistry on lung sections and lymph node sections using a mixture of two different antibodies against the PRRSV N protein as described previously ([@B30]). PRRSV antigens were detected in 3 out of 4 lung sections and 1 out of 4 lymph node sections of wild-type animals, but no PRRSV antigens were detected in sections from ΔSRCR5 pigs ([Fig. 3E](#F3){ref-type="fig"} and [F](#F3){ref-type="fig"}, bottom). Overall, no signs of infection were detected in ΔSRCR5 animals despite the high initial inoculation volume and persistent exposure to infected wild-type animals that actively shed virus (the wild-type and edited pigs were cohoused throughout the experiment). This is a clear demonstration that ΔSRCR5 animals are resistant to PRRSV-1 infection, confirming our previous *in vitro* results ([@B15]). ΔSRCR5 pigs show no cytokine response to PRRSV-1 infection and generally normal cytokine levels. {#sec2-3} ------------------------------------------------------------------------------------------------ CD163 is involved in the cytokine response to infection and immune stimuli as well as hemoglobin-haptoglobin (Hb-Hp) uptake. Hb-Hp stimulation has been shown to lead to increased levels of interleukin-6 (IL-6) and IL-10 as well as IL-1 and tumor necrosis factor alpha (TNF-α) downregulation ([@B31], [@B32]). SRCR domains 1 to 4 and 6 to 9 contain binding sites for TNF-like weak inducer of apoptosis (TWEAK), an anti-inflammatory cytokine that negatively regulates CD163 expression ([@B33], [@B34]). Stimulation with inflammatory mediators can induce the secretion of soluble CD163 and TNF-α in an ADAM17-mediated manner ([@B35]). In order to assess these biological functions in the ΔSRCR5 pigs, we assayed expression levels of key cytokines. Quantitative antibody arrays were used to assess the expression levels of 20 cytokines in serum collected from pigs on day 0 (prior to challenge) and on days 3, 7, 10, and 14 of challenge. Overall, baseline cytokine levels as determined on day 0, considered a baseline, were similar between ΔSRCR5 and wild-type pigs. However, the monokine induced by gamma interferon (IFN-γ) (MIG; also known as CXCL9) was found to show consistently higher levels in wild-type pigs until day 10, after which no significant difference was detected. MIG is a T-cell chemoattractant to inflammation sites and involved in the repair of tissue damage. In wild-type animals, MIG was strongly upregulated on days 7 and 10 of the challenge ([@B36]) ([Fig. 4H](#F4){ref-type="fig"}). The level of chemokine ligand 3-like 1 (CCL3L1) (an isoform of macrophage inflammatory protein 1α \[MIP-1α\]), involved in the inflammation response via CCR5 signaling and downregulated by IL-10 ([@B37]), was found to be higher in wild-type than in ΔSRCR5 animals ([Fig. 4J](#F4){ref-type="fig"}). As IL-10 levels were found to be comparable in both genotype groups, IL-10-mediated downregulation is unlikely to be the cause of low CCL3L1 levels. In wild-type animals, the level of CCL3L1 was elevated in the serum on days 10 and 14, whereas no significant IL-10 elevation was found to occur over the period of the challenge ([Fig. 4O](#F4){ref-type="fig"}). {#F4} We observed sequential cytokine responses to PRRSV-1 infection in wild-type animals, with an early increase in the levels of IFN-α, IL-17A, and the IL-1 receptor antagonist (IL-1ra) ([Fig. 4A](#F4){ref-type="fig"} to [C](#F4){ref-type="fig"}), followed by an increase in the levels of IL-4, IL-6, and IL-8 at the high point of viremia, from 7 days postinoculation (dpi) ([Fig. 4D](#F4){ref-type="fig"} to [F](#F4){ref-type="fig"}). Increased levels of MIG and MIP-1β (also known as CCL4) were observed only transiently at 10 dpi ([Fig. 4G](#F4){ref-type="fig"} and [H](#F4){ref-type="fig"}). In the last period of the challenge, with moderate but reducing levels of viremia, elevations in the levels of CCL3L1, granulocyte-macrophage colony-stimulating factor (GM-CSF), IL-12, and IL-1β were detected ([Fig. 4I](#F4){ref-type="fig"} to [M](#F4){ref-type="fig"}). All of these cytokine responses were restricted to wild-type animals, with no cytokine response being recorded for ΔSRCR5 pigs. IL-10, transforming growth factor β1 (TGFβ1), and IFN-γ levels showed no significant difference between wild-type and ΔSRCR5 pigs at any of the time points but were found to change significantly over time in the wild-type animals (calculated using two-way analysis of variance \[ANOVA\] and Sidak\'s multiple-comparison test) ([Fig. 4N](#F4){ref-type="fig"} to [P](#F4){ref-type="fig"}). IL-18 levels decreased significantly over time in wild-type animals but were not significantly different from those of ΔSRCR5 pigs at each time point ([Fig. 4Q](#F4){ref-type="fig"}). The level of platelet endothelial cell adhesion molecule 1 (PECAM1) was significantly elevated on day 3 of the challenge and decreased on day 10 compared to the levels in ΔSRCR5 pigs ([Fig. 4R](#F4){ref-type="fig"}). No significant difference in the levels of IL-1α and IL-13 was found between ΔSRCR5 and wild-type pigs or over time ([Fig. 4S](#F4){ref-type="fig"} and [T](#F4){ref-type="fig"}). DISCUSSION {#sec3} ========== The results of this study show that ΔSRCR5 pigs are healthy under standard husbandry conditions and maintain the biological function of the CD163 protein while being resistant to PRRSV infection. So far, we have bred three generations of edited animals with over 10 litters and have not observed any abnormalities in breeding. ΔSRCR5 pigs were generated, as previously described ([@B15]), by using two short guide RNAs (sgRNAs) flanking exon 7 of *CD163* CRISPR/Cas9. Here, we have shown that heterozygous and homozygous animals can be bred naturally and yield normal-litter-size offspring. Our previous data showed that primary PAMs and peripheral blood monocyte (PBMC)-derived macrophages from ΔSRCR5 pigs are fully resistant to PRRSV-1 subtype 1, 2, and 3 infection as well as to both typical and atypical PRRSV-2 ([@B15]). To confirm that these results translated to the *in vivo* model, we show here that ΔSRCR5 pigs are completely resistant to infection with a highly virulent PRRSV-1 subtype 2 strain. The edited animals displayed no clinical or pathological signs of infection, no viral replication was observed, and no cytokine response (indicative of low-level virus replication) was observed. This confirms that our previous *in vitro* results directly translate to the *in vivo* situation. The ΔSRCR5 CD163 protein was previously detected on the surface of CD163-expressing macrophages using a native-confirmation antibody ([@B15]). Analysis of the ΔSRCR5 CD163 amino acid sequence *in silico* using RaptorX predicts that posttranslational folding will yield a protein that closely mimics the structure of full-length CD163. The expression of ΔSRCR5 CD163 in animals has several advantages over previously described PRRSV-resistant CD163 knockout animals generated by the random introduction of a premature stop codon in exon 3 or exon 7 of the CD163 gene ([@B14], [@B23]). Free hemoglobin, often released following hemolytic events, can cause serious toxicity to a system (reviewed in reference [@B38]). CD163 is a direct mediator of Hb-Hp complex uptake into macrophages, which sequesters and degrades this potentially toxic compound ([@B39]). We have previously shown that PBMC-derived macrophages from ΔSRCR5 animals are still capable of CD163-mediated Hb-Hp uptake, as demonstrated by HO-1 upregulation and the uptake of the fluorescently labeled Hb-Hp complex (Hb~AF488~-Hp) ([@B15]). All ΔSRCR5 animals showed normal hemoglobin levels in their blood, confirming the proper clearance of Hb-Hp complexes. Yang et al. recently remade pigs with a premature stop codon in exon 7 of CD163, resulting in a functional CD163 knockout, as previously reported ([@B40]). Surprisingly, that paper claims that PBMC-derived macrophages from CD163 knockout animals are able to take up Hb~AF488~-Hp complexes *in vitro* ([@B23]), a result that directly contradicts the findings by Schaer et al. and Nielsen et al. ([@B39], [@B41]) in human macrophages, Etzerodt et al. in cells with murine CD163 ([@B42]), and Boretti et al. in canine macrophages ([@B49]), all of which highlight the essential requirement for CD163 to be present for Hb-Hp uptake in macrophages. CD163, in both its cell-bound as well as its secreted forms, has been shown to have multiple other functions in addition to Hb-Hp scavenging (reviewed in reference [@B18]). One important aspect is the regulatory function of soluble CD163 following inflammation and ischemic repair, which was found to result in enhanced regeneration activity in CD163 knockout mice, resulting in abnormal peripheral blood vessel development and systemic rather than local regeneration after injury ([@B43]). Soluble CD163 is also able to bind Staphylococcus aureus, which promotes the recognition, phagocytosis, and killing of this important livestock and human pathogen ([@B44]). Soluble CD163 is not the result of alternative splicing but results from proteolytic cleavage, likely by the metalloprotease ADAM17 in the juxtamembrane area following SRCR domain 9 of the protein ([@B45]). Proper folding of CD163 and accessibility of this area are essential for the secretion of soluble CD163. Here, we show the presence of soluble CD163 in the serum of ΔSRCR5 pigs at levels comparable to those in wild-type animals. CD163 knockout mice have been reported to be significantly more susceptible to intra-abdominal sepsis, which is linked to haptoglobin-HMGB1 signaling and the cytokine response ([@B46]). It was also found that CD163 plays a role in asthmatic human patients, and CD163 knockout mice were found to have increased airway eosinophils and mucus cell metaplasia linked to CCL24 chemokine signaling upon dust mite challenge ([@B47]). We analyzed the function of ΔSRCR5 CD163 in signaling and the cytokine response by measuring the baseline cytokine levels in ΔSRCR5 pigs compared to their wild-type counterparts. We also monitored the cytokine levels during the course of the *in vivo* PRRSV challenge to identify any changes that could result from low-level PRRSV replication. Whereas we found an orchestrated sequence of inflammation and immune response signaling in the PRRSV-infected wild-type animals, no cytokine response was observed in the ΔSRCR5 pigs. Of the panel of 20 cytokines analyzed, the levels of only one cytokine, CCL3L1, were significantly different between the two groups of animals over the entire course of the challenge. The inflammation response protein CCL3L1 is involved in the inflammation response and is downregulated by IL-10, but no significant differences in IL-10 levels between the two genotype groups could be found. We have no explanation for the higher CCL3L1 level in ΔSRCR5 pigs, but the limited time frame and number of animals in this study warrant further investigation of this cytokine in the future. Another cytokine, MIG, showed higher levels in wild-type animals up to day 10, but no significant difference was observed on day 14. It will be interesting to investigate this observation further, over a longer period and with larger numbers of animals. The creation of ΔSRCR5 pigs holds tremendous opportunity for the pork industry worldwide to improve both animal welfare and productivity at the same time. PRRSV infection has immunomodulatory outcomes and plays an important role in polymicrobial disease, such as the porcine respiratory disease complex. As such, PRRSV-resistant animals could benefit general health and decrease the need for antimicrobial use at the same time. However, for the implementation of next-generation breeding/genome-editing techniques in animal production, both consumer acceptance and the legislative framework need to be in place. MATERIALS AND METHODS {#sec4} ===================== All animal work was approved by The Roslin Institute\'s and the University of Edinburgh\'s Protocols and Ethics Committees as well as the ethics group at Moredun Scientific Ltd. The experiments were carried out under the authority of UK Home Office project licenses PPL60/4518, PPL60/4482, and PPL70/8827 under the regulations of the Animal (Scientific Procedures) Act of 1986. Humane endpoints were clearly defined. Cells and viruses. {#sec4-1} ------------------ Primary pulmonary alveolar macrophages (PAMs) for the propagation of PRRSV-1 subtype 2 strain BOR-57 (isolated from a sample from a Belarusian pig in 2009 by T. Stadejek) were harvested from wild-type research animals aged 6 to 9 weeks as previously described ([@B48]). Briefly, animals were euthanized according to a schedule I method. Lungs were removed and transferred on ice to a sterile environment. Lung lavage with warm phosphate-buffered saline (PBS) and gentle massage were used to recover PAMs. Cells were collected by centrifugation for 10 min at 400 × *g*. When necessary, red blood cells were removed using red cell lysis buffer (10 mM KHCO~3~, 155 mM NH~4~Cl, 0.1 mM EDTA \[pH 8.0\]) for 5 min before washing again with PBS. Cells were collected by centrifugation as described above and frozen in a solution containing 90% fetal bovine serum (FBS) (heat inactivated \[HI\]; GE Healthcare) and 10% dimethyl sulfoxide (DMSO) (Sigma). Cells were frozen gradually at 1°C/min in a −80°C freezer before being transferred to −150°C. PAM cells were cultivated in a solution containing RPMI 1640, GlutaMAX (Invitrogen), 10% FBS (HI; Gibco), 100 IU/ml penicillin, and 100 μg/ml streptomycin (Invitrogen) (RPMI^+/+^). Breeding and genotyping of animals. {#sec4-2} ----------------------------------- Unrelated founder animals generated by zygote injection of Cas9 mRNA and sgRNAs SL26 and SL28, as described previously ([@B15]), were bred to each other or to wild-type animals to generate heterozygous F1 offspring. F1 animals with a double-strand break and religation without insertions or deletions at the cutting sites of SL26 and SL28 were selected for breeding the F2 generation. Animals were genotyped as described previously ([@B15]); briefly, genomic DNA was extracted from ear biopsy specimens using the DNeasy blood and tissue kit (Qiagen). The region spanning intron 6 to exon 8 was amplified using primers oSL46 (ACCTTGATGATTGCGCTCTT) and oSL47 (TGTCCCAGTGAGAGTTGCAG), generating a 904-bp product from the intact allele and a 454-bp product if the complete deletion of exon 7 had occurred. The PCR products were analyzed by separation on a 1% agarose gel and by Sanger sequencing. Animal challenge with BOR-57. {#sec4-3} ----------------------------- Four days prior to transfer of the animals to the specific-pathogen-free unit (SPFU), blood and serum samples were taken from all animals by jugular venipuncture, and the blood samples were subjected to whole-blood-count analyses. Sera were screened by using the Idexx PRRSV X3 ELISA to confirm that none of the animals had previously been exposed to PRRSV. Animals were acclimated in the SPFU for 1 week prior to challenge. Infectivity of BOR-57 stocks was assessed using a TCID~50~ assay on PAMs immediately after production, prior to challenge, and on both challenge dates after administration. The BOR-57 inoculum was tested for the absence of mycoplasma and other major viral pathogens. Animals were inoculated intranasally in the left nostril with 5 ml of 1E6 TCID~50~/ml BOR-57 in RPMI^+/+^. Body weights of the animals were measured on day 0 prior to challenge and on day 7 and day 14 prior to euthanasia. Serum samples were collected on day 0 prior to challenge and on days 3, 7, 10, and 14 prior to euthanasia by jugular venipuncture into Vacutainer tubes. After clotting, samples were centrifuged at 2,000 × *g* for 10 min at 4°C, and samples were aliquoted and frozen at −80°C for further analysis. Clinical observations were recorded daily, making note of the rectal temperature, demeanor, nasal discharge, coughing, and respiration. Feeding, water consumption, and general health were observed and recorded daily. Humane endpoints were defined prior to challenge. No animal reached the criteria for premature termination during the challenge. Necropsy, histopathology, and immunohistochemistry. {#sec4-4} --------------------------------------------------- On day 14 of the challenge, animals were euthanized according to a schedule I method. During necropsy, the lungs were removed and initially assessed, and detailed photographs were taken from the dorsal and ventral sides for detailed scoring of macroscopic lung lesions. An established scoring system, based on the approximate contribution of each lung section to the complete lung volume, was employed as previously described ([@B29]). Briefly, each lung lobe is assigned a number to reflect the approximate volume percentage of the entire lung represented. The affected area of each lobe is scored relative to the assigned volume percentage. Lung, mediastinal lymph node, and PAM samples were collected and frozen, and lung and lymph node samples were fixed in 10% neutral buffered formalin. Formalin-fixed sections were embedded in paraffin and routinely processed for histological examination with hematoxylin and eosin staining. Lung sections were scored for the presence and severity of interstitial pneumonia, ranging from 0 (normal) to 6 (severe diffuse), as previously described ([@B29]). Immunohistochemical analysis of lung and lymph nodes for the detection of PRRSV antigen was performed as previously described ([@B30]), using a mixture of two monoclonal antibodies, SDOW-17 at 1:5,000 and SR-30 at 1:1,500 (both Research\|Technology\|Innovation \[RTI\]), as primary antibodies. Sections were counterstained with hematoxylin. Assessment of PRRSV and anti-PRRSV antibody levels in serum. {#sec4-5} ------------------------------------------------------------ Viral RNA was extracted from the sera collected on days 0, 3, 7, 10, and 14 using the QIAamp viral RNA minikit (Qiagen) according to the manufacturer\'s instructions. Viral RNA levels were assessed by reverse transcription-quantitative PCR (RT-qPCR) using the GoTaq 1-Step RT-qPCR system (Promega) and analyzed on a LightCycler II 480 instrument (Roche). Viral RNA extracted from the PAM cell culture supernatant with a known multiplicity of infection (MOI) and a custom-synthesized DNA fragment (Invitrogen) of a known concentration (GAGAGCGGCCGCTAATACGACTCACTATAGTCAGCTGTGTCAGCTGCTGGGAAAAATGATGAAATCCCAGCGCCAGCAACCCAGGGGAGGACAGGCCAACAAAAGGAAAAAGCCTGAGAAGCCTCATTTTCCCTTGGCTGCTGAAGATGACGTTCGGCACCATCTCACCGCAACTGAGCGTTCCCTCTGTCTGCAATCGATCCAGACAGCCTTCAATCAGGGTGCAGGAACTGCGTCGCTTTCACCCAGTGGGAAGGTCAGTTTTCAGGTAGAGTTCATGCTGCCCCTGCAGGGAGA) were used as standards. The primers used were BOR57_FWD (GAAATCCCAGCGCCAGCAAC) and BOR57_REV (TTCCCACTGGGTGAAAGCGA). Assessment of soluble CD163 serum levels. {#sec4-6} ----------------------------------------- Serum samples collected a week prior to and on day 0 of the challenge were analyzed for the presence and level of soluble CD163. A sandwich ELISA was performed using the porcine CD163 ELISA kit (Elabscience) according to the manufacturer\'s instructions and using serial dilutions of serum. Analysis of serum cytokine levels using cytokine arrays. {#sec4-7} -------------------------------------------------------- Serum samples collected on day 0 (prior to challenge) and on days 3, 7, 10, and 14 of the challenge were analyzed for the levels of 20 different cytokines. Cytokine array analyses were performed using porcine cytokine antibody array A (catalog number ab197479) and array B (catalog number ab197480) (both from Abcam) according to the manufacturer\'s instructions. We acknowledge financial support from the BBSRC Animal Health Research Club (BB/L004143/1), Genus plc, and BBSRC Institute Strategic Programme grant funding to The Roslin Institute (BBS/E/D/20241866, BBS/E/D/20002172, and BBS/E/D/20002174). We have the following competing interests: C.B., S.G.L., T.A.-A., C.B.A.W., and A.L.A. have patent claims on the targeted deletion described in the manuscript. [^1]: **Citation** Burkard C, Opriessnig T, Mileham AJ, Stadejek T, Ait-Ali T, Lillico SG, Whitelaw CBA, Archibald AL. 2018. Pigs lacking the scavenger receptor cysteine-rich domain 5 of CD163 are resistant to porcine reproductive and respiratory syndrome virus 1 infection. J Virol 92:e00415-18. <https://doi.org/10.1128/JVI.00415-18>.

Introduction {#s1} ============ Human ESCs are a powerful tool for drug screening, studying early lineage differentiation *in vitro*, and generating specific cell phenotypes for therapeutic applications. However, optimizing the efficient expansion and differentiation of these cells still requires further understanding, especially of the signaling pathways responsible for regulating hESC. Several pathways have been implicated in hESC self-renewal, including transforming growth factor-β/Activin-A/Nodal [@pone.0017538-Vallier1], sphingosine-1-phosphate/platelet-derived growth factor (S1P/PDGF) [@pone.0017538-Pebay1], insulin growth factor (IGF)/insulin [@pone.0017538-Bendall1] and fibroblast growth factor-2 (FGF-2) [@pone.0017538-Dvorak1] (reviewed in Avery *et. al*. 2006 [@pone.0017538-Avery1]). The process of self-renewal appears to be regulated synergistically through various pathways via growth factor or cytokine supplementation. Interestingly, FGF-2 signaling appears indispensible to hESC self-renewal just as leukemia inhibitory factor is to mESC [@pone.0017538-Ying1]. Therefore, FGF-2 is widely used for sustained long-term culture of human embryonic stem cells (hESC) and induced pluripotent stem (iPS) cells under both feeder and feeder--free culture conditions [@pone.0017538-Xu1]--[@pone.0017538-Ludwig1]. FGFs execute their biological actions by activating cell surface fibroblast growth factor receptors (FGFRs) [@pone.0017538-Mohammadi1], [@pone.0017538-Ibrahimi1]. The four human FGFRs, namely FGFR1, 2, 3 and 4, are members of the receptor tyrosine kinase (RTK) family. These receptors govern a wide variety of cellular processes from cell motility and differentiation to proliferation. hESC express all four FGFR [@pone.0017538-Dvorak1], [@pone.0017538-Ding1]--[@pone.0017538-Bhattacharya1] whereby blocking of FGFR signaling leads to rapid differentiation [@pone.0017538-Dvorak1], [@pone.0017538-Ding1]. This suggests that FGF-mediated signaling is important for hESC self-renewal. Following FGF-2 stimulation, activation of the FGF/FGFRs in hESC typically results in signal transduction of the FGF canonical pathways, namely the mitogen activated protein kinase (MAPK) and phosphoinositide 3- kinase (PI3-K) pathways [@pone.0017538-Dvorak1], [@pone.0017538-Ding1], [@pone.0017538-Li1]. Further downstream signaling events upon FGF-2 stimulation and its link to hESC self-renewal and the maintenance of pluripotency remain to be determined. The large scale analysis of cellular phosphorylation by liquid chromatography-mass spectrometry (LC-MS) is challenging due to the low stoichiometry of phosphorylation, causing phosphopeptides to remain largely undetected in the overwhelming background of non-modified peptides. Several of these large scale proteomic methods have been applied to profile phosphorylation in hESC [@pone.0017538-Swaney1]--[@pone.0017538-Brill1]. Although impressive in the number of reported sites, these latter studies largely failed to monitor tyrosine phosphorylation. Therefore, specific enrichment of tyrosine phosphorylated peptides by phosphotyrosine-specific antibodies may provide a more targeted approach to study tyrosine phosphorylation [@pone.0017538-Rush1]--[@pone.0017538-Boersema1]. To understand the broader implications of FGF signaling in hESC, we adopted a large-scale, targeted, phosphoproteomics approach to investigate tyrosine phosphorylation events following FGF-2 stimulation. Using a peptide-centered immuno-affinity purification strategy [@pone.0017538-Boersema1], [@pone.0017538-Boersema2] 735 unique tyrosine phosphorylation sites on 430 proteins were detected in two biological hESC replicates. Combining this enrichment technique with a stable isotope dimethyl labeling strategy [@pone.0017538-Boersema2], [@pone.0017538-Boersema3], a quantitative picture of the early signaling events in FGF-2 stimulated hESC was generated. [Fig. 1](#pone-0017538-g001){ref-type="fig"} represents an overview of the quantitative phosphotyrosine proteomics strategy. Results from our quantitative dataset suggest that all four FGFRs were activated upon FGF-2 stimulation. Induced tyrosine phosphorylation of members of both MAPK and PI3-K pathways was also observed. Additionally, diverse trans-activation of receptors in the EGF family (EGFR, ERBB2, and ERBB3), Insulin family (IGF1-R and INSR), Ephrin receptors, and Vascular endothelial growth factor receptor 2 (VEGFR2/KDR) was detected. An increase in phosphorylation of Src kinase substrates was also observed, suggesting a possible role of FGF-2 in regulating cytoskeletal and actin dependent processes. The impact of changes to the cytoskeletal processes on hESC pluripotency should be further explored. {#pone-0017538-g001} Results {#s2} ======= Effect of FGF-2 stimulation on the hESC phosphoproteome {#s2a} ------------------------------------------------------- To specifically investigate the effects of activating FGFR, exogenous FGF-2 was added to hESC that were deprived of FGF-2 for 5 days (1 day ∼ 1 population doubling). Activation of the hESC was validated by Western blotting using two antibodies specific for the phosphorylation of proteins downstream of the FGFR. After 15 min of stimulation, a rapid upregulation of both phosphorylated ERK1/2 (ph-ERK1/2) and AKT (ph-AKT) was observed ([Fig. S1](#pone.0017538.s001){ref-type="supplementary-material"}). A targeted phosphoproteomics experiment was subsequently carried out to profile the global tyrosine phosphorylation events post FGF-2 stimulation using an anti-pY specific antibody to enrich for tyrosine phosphorylated peptides [@pone.0017538-Boersema1], [@pone.0017538-Rikova1]. A total of 6 mg of cell lysate per cell state (different time points following FGF-2 stimulation) was digested with trypsin and this peptide digest was subsequently enriched for tyrosine phosphorylated peptides. The enriched peptides were then analyzed by LC-MS. A total of 153, 376, 273, 285 and 287 unique tyrosine phosphorylation sites could be identified respectively from the 0, 1, 5, 15 and 60 min post-FGF-2 stimulated cells. A cumulative total of 597 unique tyrosine phosphorylation sites were observed. In a biological replicate performed with similar amounts of sample and with similar FGF-2 stimulation conditions, a total of 574 tyrosine phosphorylation sites could be identified. Both experiments led to an overall cumulative 735 unique tyrosine phosphorylation sites in the two biological replicates ([Table S1](#pone.0017538.s004){ref-type="supplementary-material"}). The observed more than double increase in tyrosine phosphorylation immediately after FGF-2 stimulation suggests a prompt activation of numerous tyrosine phosphorylation signaling pathways. The 76% overlap we observed between the here performed biological replicates is much larger than typically observed in shotgun LC-MS approaches focusing on global serine and threonine phosphorylation [@pone.0017538-Gauci1]. This is likely caused by the significant reduction in sample complexity after enrichment of tyrosine phosphorylated peptides, thus allowing a more comprehensive and reproducible profiling of tyrosine phosphorylation [@pone.0017538-Boersema1]. A first classification using Panther [@pone.0017538-Thomas1] indicated that many of the detected pY sites were on proteins involved in (growth factor) signaling, such as the expected FGFs, but also EGF, VEGF and PDGF, and proteins involved in their downstream pathways. A comparison of our hESC dataset with known pY site locations obtained from Phospho.ELM (version 8.2) and Rikova et al. [@pone.0017538-Rikova1] yielded overlaps of only 23% and 41%, respectively ([Fig. S2](#pone.0017538.s002){ref-type="supplementary-material"}). A global motif analysis was performed on the residues adjoining the phosphorylation sites in our hESC dataset. Overrepresented sequence motifs were extracted using Motif-X [@pone.0017538-Amanchy1]. Employing the human IPI database for background sequences, 7 significant motif patterns could be identified, covering over 53% of our dataset ([Fig. 2A--B](#pone-0017538-g002){ref-type="fig"}). Comparison of the seven motif patterns identified in hESC with both the phospho.ELM database and the Rikova data revealed that motif 1 and motif 5 are unique to our hESC dataset. Amongst the proteins exhibiting such motifs were PI3-K (in motif 1), MAPK1, MAPK3, GAB1, SHB, ERBB3, TJP1, PKP2, PKP4, and CDK5 (all in motif 5) ([Table S2](#pone.0017538.s005){ref-type="supplementary-material"}). Further comparison to the Human Protein Reference Database [@pone.0017538-Amanchy1] confirmed that both motif 1 and motif 5 are indeed unique to our cumulative hESC dataset ([Fig. 2C](#pone-0017538-g002){ref-type="fig"}). ![Tyrosine phosphorylation motifs overrepresented in the hESC cumulative dataset.\ Seven motifs were found to be significantly overrepresented using the Motif-X algorithm. (A) Distribution of different motifs in the dataset. For approximately half of the phospho-peptides (53%) an overrepresented motif could be identified, with the most abundant motifs being motifs 3 (12%) and 4 (11%). (B) Sequence logos of the various motifs. (C) Comparison of motifs among the various datasets. The table shows the motifs found in the different datasets, and their respective motif score. Motifs 2, 3, 4, 6, and 7 were also found either in the Phospho.ELM (ver. 8.2) or the Rikova et.al [@pone.0017538-Rikova1] datasets. Motifs 1 and 5, however, are unique to the hESC dataset. The number of foreground matches, background matches and the level of fold increase for each motif are also shown in the table.](pone.0017538.g002){#pone-0017538-g002} Quantitative phosphoproteomics {#s2b} ------------------------------ In order to accurately quantify the tyrosine phosphorylation events, we next applied a combined approach of pY-peptide immunoprecipitation (IP) and stable isotope labeling. Using this combined approach, we have generated a more precise and quantifiable tyrosine phosphorylation profiles [@pone.0017538-Boersema1]. Our Western blot results on ERK1/2 and AKT ([Fig. S1](#pone.0017538.s001){ref-type="supplementary-material"}) indicated that most of the tyrosine phosphorylation events occur within 15 min of FGF-2 stimulation. Therefore, 0, 5 and 15 min post FGF-2 stimulation time points were chosen for this profiling study. hESC lysate for each time point (6 mg) was digested, followed by stable isotope dimethyl labeling, whereby the non-stimulated hESC (0 min) was labeled with light dimethyl labels, the 5 min time point with intermediate labels, and the 15 min time point with heavy dimethyl labels. The differentially labeled samples were mixed 1∶1∶1 and enriched simultaneously for pY peptides by IP. The enriched pY peptides were then analyzed by LC-MS, using a 3 h elution gradient. By comparing their signal intensities, peptides from the different FGF-2 stimulated time points could be relatively quantified. From our analysis, 316 unique pY-peptide triplets (light, intermediate, heavy) with 300 unique tyrosine phosphorylation sites were identified ([Table S3](#pone.0017538.s006){ref-type="supplementary-material"}). The quantitative analysis demonstrated an increase in tyrosine phosphorylation of all 4 FGFRs and some of their canonical downstream effectors (eg. PLC-, MAPK, PI3-K) following FGF-2 stimulation [Fig. 3](#pone-0017538-g003){ref-type="fig"}. Furthermore, a large number of pY-peptides not directly involved in the canonical FGF pathway were also identified with increased phosphorylation upon FGF-2 stimulation. These include Src kinase substrates (e.g. FAK, CTTN, PXN, SHANK2), additional receptor tyrosine kinases (e.g. INSR, IGF1R, KDR, ERBB2, MEGF-10), and many others (e.g. Occludin, TJP1, TJP2, Kirrel) ([Fig. 3](#pone-0017538-g003){ref-type="fig"}). To classify the global response to FGF-2, a cluster analysis was performed based on the temporal tyrosine phosphorylation profiles. Phosphopeptides that showed at least a 2-fold increase (arbitrarily chosen as a substantial increase) in tyrosine phosphorylation at either of the stimulated time points (5 min or 15 min) were clustered into 5 different groups ([Fig. 4](#pone-0017538-g004){ref-type="fig"} and [Table S4](#pone.0017538.s007){ref-type="supplementary-material"}). Most of the pY phosphopeptides (approximately 34.5%) showed sustained or transient activation (Clusters 3 and 4) upon stimulation. The pY peptides of all FGFRs and some of their downstream targets (PLC\--, MAPK1, MAPK3, PI3-K, SHC) were found within these clusters. {#pone-0017538-g003} {#pone-0017538-g004} Interestingly, various substrates of the Src kinase family were also found within these clusters, namely Shank2, F11R, PKP2, TJP2, CDK5 (all in cluster 3), and CTTN, OCLN, FAK, PKP3, PXN, and TJP1 (all in cluster 4) (see [Table S4](#pone.0017538.s007){ref-type="supplementary-material"}). The Src family kinases have been implicated in the maintenance of stem cell pluripotency[@pone.0017538-Anneren1]. Selective inhibition of these kinases by the chemical inhibitor SU6656 resulted in the downregulation of expression of such pluripotent marker as Oct3/4 and Nanog and decreased growth of hESCs[@pone.0017538-Anneren1]. Similar results were also observed in our study ([Fig. S3A--C](#pone.0017538.s003){ref-type="supplementary-material"}). Despite the continuous stimulation of hESC with FGF-2, inhibition of Src kinases with 1 µM SU6656 for 6 days led to significant downregulation of Oct3/4, Nanog and EpCAM expression, suggesting the importance of Src kinase signaling in maintaining the undifferentiated hESC phenotype. Concomitantly, decrease in cell numbers was observed following SU6656 treatment, however, this was more pronounced at 4 µM ([Fig. S3B](#pone.0017538.s003){ref-type="supplementary-material"}). To confirm the increased phosphorylation of the RTKs identified post FGF-2 stimulation, a protein array containing 42 different human RTKs (including receptors from the FGF, EGF, VEGF, and insulin receptor families) was used ([Fig. 5](#pone-0017538-g005){ref-type="fig"}). In the array experiment, we also included an additional time point (60 min) to understand the pattern of phosphorylation in stimulated hESC. A rapid increase in phosphorylation of all FGF receptors was observed ([Fig. 5A--B](#pone-0017538-g005){ref-type="fig"}). Furthermore, phosphorylation levels for INSR, IGF1-R, EphA 1 and 2, VEGFR2/KDR were also increased as compared to the basal level (no stimulation). ([Fig. 5C--F](#pone-0017538-g005){ref-type="fig"}). Results from the array confirmed an elevated level of phosphorylation in several RTK, and all four FGFRs (FGFR1--4) were activated simultaneously upon FGF-2 treatment. {#pone-0017538-g005} Discussion {#s3} ========== Many cytokines and growth factors have been shown to play a role in maintaining self-renewal of hESCs (reviewed in [@pone.0017538-Avery1], [@pone.0017538-Stewart1]). The central role of FGF-2 in maintaining hESC self-renewal remains undisputable, as almost all hESC culture platforms require FGF-2 supplementation. Withdrawal of exogenous FGF-2from hESC results in the loss of pluripotent marker expression (Oct-3/4, Tra-1-60, and Podocalyxin) [@pone.0017538-Ding1]. Furthermore, blocking FGFR signaling in hESC leads to rapid differentiation [@pone.0017538-Dvorak1], [@pone.0017538-Ding1]. Interestingly, co-localization of all four FGFRs occurs on Oct-3/4 positive cells. Taken together, these results suggest that FGF-mediated signaling is important for the maintenance of the undifferentiated hESC phenotype. To further understand FGF-2 mediated FGFR signaling in hESC, we profiled tyrosine phosphorylation events following FGF-2 stimulation. The main challenge in tyrosine phosphoproteomics is the low abundance of tyrosine phosphorylated proteins. In this study, we used a pY-specific antibody to enrich for these low abundant pY-peptides from large quantities of complex samples. Using this approach, 735 unique tyrosine phosphorylation sites in the two biological replicates were identified. To our knowledge, this is to date the largest tyrosine phosphorylation dataset reported for a hESC phosphoproteome. The overlap of our cumulative dataset with the datasets from Phospho.ELM and Rikova *et. al*. [@pone.0017538-Rikova1] is 23% and 41% respectively. Dissimilar cell constitutions (human/mouse cell lines/tissues for phosho.ELM, lung cancer cell lines/tissues from Rikova) together with different enrichment strategies could account for this difference. It has been estimated that tyrosine phosphorylation accounts for about 1% of all phosphorylated events in human cells [@pone.0017538-Olsen1], hinting at potentially over ∼75,000 phosphorylation events in our hESC. This is still an order of magnitude higher than currently achievable in large-scale phosphoproteomics screens [@pone.0017538-Swaney1], [@pone.0017538-Johnson1]. We identified, using Motif-X [@pone.0017538-Schwartz1], a total of 7 tyrosine phosphorylated sequence motif. Interestingly, motif 1 and motif 5 have not been previously reported in any pY dataset, and could be specifically enriched in hESC. The hESC unique motif 5 has an arginine residue at the P+3 position, and was detected on proteins including MAPK1, MAPK3, GAB1, SHB, ERBB3, TJP1, PKP2, PKP4, and CDK5. Motif 1 contains a glutamic acid residue at P-2, and an alanine residue at P+1 position. A phosphopeptide of the regulatory subunit of PI3-K has a peptide sequence that belongs to motif 1, which is of interest as the PI3-K signaling pathway has been reported to be important for hESC self-renewal [@pone.0017538-Ding1], [@pone.0017538-Armstrong1]. The position of the glutamic acid at P-2 was also observed in motif 4. Hence, it is possible that motif 1 represents a subset of motif 4. Src kinase has been proposed to be the upstream kinase recognizing motif 4 [@pone.0017538-Schwartz1]. The close resemblance between motifs 1 and 4 may imply that proteins/peptides belonging to motif 1 are also phosphorylated by Src kinases. Quantitative profiling of tyrosine phosphorylation {#s3a} -------------------------------------------------- In order to understand early key signaling events of FGF-2 stimulated hESC, it is important to study the temporal involvement of the FGFRs, other RTKs, and their subsequent substrates post FGF-2 stimulation. Results from Western blotting of FGF signaling downstream effectors (ph-ERK1/2 and ph-AKT) demonstrated rapid response (5--15 min) of these downstream effectors to FGF-2 stimulation, hence we selected 0, 5, and 15 min for a more in-depth quantitative study. Stable isotope dimethyl labeling method was selected to be used in conjunction with pY-peptide IP. By performing stable isotope labeling prior to IP, potential variation in the IP or LC-MS step is mitigated as sample handling and analysis is performed simultaneously for all samples. A potential downside is that up to 6 mg of material needs to be isotope labeled, necessitating a cost-effective labeling approach provided by reductive dimethyl labeling. Using the above combination, 316 unique quantifiable tyrosine phosphorylated peptides, occurring as a triplet in the MS data, could be identified and quantified. As expected, increased tyrosine phosphorylation at multiple sites of FGFRs was observed after FGF-2 stimulation, notably on all four FGFRs. The use of the RTK array confirmed the simultaneous activation of the four FGFRs. A separate study by Livia *et. al* (2004) [@pone.0017538-Livia1] using the RTK array also demonstrated activation of FGFRs in hESC by FGF-2 stimulation. Increases in tyrosine phosphorylation of downstream targets implied in canonical FGF signaling, such as PLC--, GAB1, MAPK1/3, PI3-K could also be detected. The temporal response of pY residues at the autophosphorylation sites on the FGFRs indicates propagation of the signal due to the presence of exogenous FGF-2. Apart from FGFRs, several other receptors were found to be rapidly tyrosine phosphorylated following FGF-2 stimulation. These included INSR, IGF1-R, ERBB2, ERBB3, EphA1 and 2, EphB3 and 4, VEGFR2/KDR, and MEGF10. The activation of most of these receptors was confirmed by a human phospho-RTK array. IGF1-R and IR have been shown to play a role in self-renewal of hESC [@pone.0017538-Bendall1], [@pone.0017538-Wang1]. Furthermore, tyrosine phosphorylation of ERBB2, ERBB3, VEGFR2/KDR, FGFR3, and FGFR4 has been demonstrated previously upon stimulation with conditioned medium supplemented with FGF-2 [@pone.0017538-Wang1]. MEGF10 showed a noteworthy increase in tyrosine phosphorylation at multiple sites. There are very limited reports on MEGF10 and its function in hESC remains to be further explored (see [Text S1](#pone.0017538.s009){ref-type="supplementary-material"} for further discussion). Although many of the results obtained in the pY-IP approach could be validated using the human phospho-RTK array, the sequencing approach of the phosphoproteomics method allows the determination of exactly which tyrosine sites are differentially phosphorylated and therefore provide higher resolution data. One possible explanation for the induced tyrosine phosphorylation of other RTKs following FGF-2 stimulation could be transactivation [@pone.0017538-Yokote1], [@pone.0017538-Esposito1]. For example, FGFR1 has been shown to be capable of tyrosine phosphorylating EphA4 in an adult kinase-negative mutant cell line [@pone.0017538-Yokote1]. Also, it has been demonstrated that a transactivation mechanism might even be required to establish certain physiological processes [@pone.0017538-Esposito1]. The elucidation of these RTK transactivations and cross-talk between different pathways, once again, illustrates that a linear view of signal transduction is an oversimplification of the actual signaling process [@pone.0017538-VanHoof1]. Some RTKs share several of the same downstream targets, including the above mentioned PLC--, GAB1, MAPK1/3, and PI3-K, which may complicate the determination whether phosphorylation of any of these proteins is the direct result of upstream activation of FGFR or any of the other RTKs. Therefore, a more comprehensive view based on the data provided here is required to deconvolute the complex interplay of signaling pathways underlying the effect of cellular stimuli. Src family kinases and substrates {#s3b} --------------------------------- Src family kinases (SFKs) represent a group of tyrosine kinases that are strongly activated by, amongst others, RTKs [@pone.0017538-Brunton1]. Several tyrosine phosphorylation sites were identified on members of SFKs, but the extensive homology, in particular, in the activation loop of different members of the family complicates the determination of the exact family member. Interestingly, although only relatively small increases in tyrosine phosphorylation were observed for the SFK members, larger increases in tyrosine phosphorylation were observed for Src substrates, for example, focal adhesion kinase (FAK). Src and FAK are known to form a tight complex after activation by RTKs or integrins [@pone.0017538-Parsons1], [@pone.0017538-vanNimwegen1]. Src binds to phosphorylated FAK and promotes further phosphorylation of tyrosine residues on FAK. Cortactin has long been known to be a very efficient substrate for SFKs and was found in our screen to exhibit increased tyrosine phosphorylation levels. These phosphorylation events are supposed to create docking sites for proteins containing the SH2 phosphotyrosine binding domain, such as Src family kinases themselves or adaptor proteins [@pone.0017538-Daly1], [@pone.0017538-Buday1]. These interactions have been proposed to influence actin polymerization and thus the turnover of actin networks. SHANK2 was also found to be heavily phosphorylated upon induction with FGF. It is a large scaffold protein associated with actin, and also a binding partner of Cortactin [@pone.0017538-Boeckers1]. Another protein that showed increased tyrosine phosphorylation is the scaffold protein Paxillin, which localizes at the sites of cell adhesions. Together with a large number of interaction partners, Paxillin is heavily involved in cytoskeletal reorganization and cell adhesion [@pone.0017538-Mitra1]--[@pone.0017538-Deakin1]. Plakophilin-3, Plakophilin-4, Catenin delta-1, F11R, Occludin, KIRREL/Neph1, TJP-1 and TJP-2 are found in tight junctions and an increased tyrosine phosphorylation was detected in our screen. Of these proteins Catenin delta-1, Occludin, TJP-1 and TJP-2 have been described as genuine Src kinase substrates; the phosphorylation of the aforementioned sites has been shown to modulate the regulation of cell-cell adhesions as well as the formation of tight junctions [@pone.0017538-Mitra1], [@pone.0017538-Rao1], [@pone.0017538-Saito1]. The prevalence of activity of Src family kinases can be seen from both Motif mapping (Motif 1 and 4) and the large number of Src kinase substrates identified. SFKs can also be recruited by receptor mediated phosphorylation and influence the signaling dynamics of the cell. It has been stated that Src activity controls FGFR activation and FGF-2 induced AKT activation is dependent on activation of Src [@pone.0017538-Sandilands1]. Furthermore, SFKs are known to promote cell growth and survival by activation of MAP kinase pathways (e.g. MEKK2/5, Erk5) or the signal transducers and activators of transcription (STAT) family (e.g. STAT3). In this study, we demonstrate that Src kinase activity is crucial for the maintenance of pluripotency by FGF-2. Inhibition of Src kinases resulted in decreased growth and expression of markers representing the undifferentiated hESC state. Results from this investigation and previous study by Annaren *et. al*. (2004) suggest a considerable involvement of these kinases and their substrates in hESC in maintenance of undifferentiated hESC[@pone.0017538-Anneren1] `.` Other interesting targets {#s3c} ------------------------- CDK5 was also found to be upregulated upon FGF-2 treatment. CDK5 is an atypical member of the cyclin dependent kinase family, which has rather been found to be involved in the regulation of CNS development [@pone.0017538-Dhariwala1]. Many of its substrates are also involved in cytoskeletal regulation and neuronal migration such as c-Abl, Src, PAK1 or β-catenin. Multiple EGF-like domains 10 (MEGF10) was identified in the dataset with an average of 5 fold-change on all the four pY-peptides identified (within 5 min of FGF-2 stimulation). Out of the four pY-peptides identified, two were not reported previously. Additionally, this is the first account for identification of tyrosine phosphorylation of MEGF10 in hESC, but its precise function in hESC is largely unknown. MEGF10 is believed to be the mammalian ortholog of nematode CED-1, a cell surface receptor involved in phagocytosis of dead cells [@pone.0017538-Suzuki1], [@pone.0017538-Hamon1]. It has been known to interact with clathrin assembly protein complex 2 medium chain (AP50) [@pone.0017538-Suzuki2], and ATP binding cassette transporter (ABCA1) [@pone.0017538-Hamon1]. The function of this protein is largely thought to be involved in cell engulfment due to its similarity with CED-1 [@pone.0017538-Hamon1], [@pone.0017538-Suzuki2]. It has also been reported to have a role in novel adhesion pattern, restricting cell motility, and inducing large vacuole formation [@pone.0017538-Suzuki1], [@pone.0017538-Suzuki2]. There are limited reports on the MEGF10 signaling pathway. Whether the function of the protein is supporting self-renewal of hESC remains to be elucidated. Conclusion {#s3d} ---------- In this study, we have provided a global temporal tyrosine phosphorylation profile of hESC after FGF-2 stimulation. In [Figure 6](#pone-0017538-g006){ref-type="fig"}, the events following the introduction of FGF-2 to hESC are summarized. Remarkably, we did not only observed an activation of the FGFR pathways, but also detected the tyrosine phosphorylation of several other RTKs upon FGF-2 stimulation. This activation could be either a direct effect of FGF-2 (although no specificity for FGF-2 is known for the mentioned RTKs), through transactivation, which would involved an FGFR, or via a yet unknown mechanism. All RTKs, including FGFRs, have downstream target substrates that are tyrosine phosphorylated. Not surprisingly, we found a plethora of proteins with increased levels of tyrosine phosphorylation upon FGF-2 stimulation, many of which have not been associated with FGF signaling before. Many of these proteins have been described as integral components and also key players in the aforementioned processes in adult cells with functions including modulation of cell-adhesion, cell-cell interaction, migration and the formation of tight junctions. Our preliminary results from our Src kinase inhibition experiment suggest possible involvement of the Src kinase and its substrate in the maintenance of undifferentiated hESC phenotype. Interestingly, changes to cytoskeletal and actin dependent processes are major events during hESC differentiation [@pone.0017538-Chae1]. Therefore, we hypothesize that the pluripotency maintaining effect of FGF-2 in hESC might be partially executed through these processes. {#pone-0017538-g006} Materials and Methods {#s4} ===================== Cell culture, stimulation, and cell lysis {#s4a} ----------------------------------------- Human ESC line, HES-3 (46 X,X) was obtained from ES Cell International (ESI, Singapore). Briefly, the cells were cultured on Matrigel-coated (Becton, Dickinson and Company, Franklin Lakes, NJ) tissue culture dishes and supplemented with conditioned medium from MEF. For routine culture, the medium was supplemented with 10 ng/ml of FGF-2 (Invitrogen, Carlsbad, CA, USA), and medium was changed daily. The cultures were passaged weekly following enzymatic treatment as previously described [@pone.0017538-Ding2]. For FGF-2 starved cultures, cells were maintained in the absence of FGF-2 for 5--7 population doublings (PD, 1 PD  =  ∼1 day). Cells were then stimulated with 10 ng/ml of FGF-2 at the indicated time points. Cells were lysed on ice in 7 M urea, 2 M thiourea, 4% CHAPS, 40 mM Tris, 50 µg/ml DNase, 50 µg/ml RNase, 1 mM sodium orthovanadate and 1X *PhosSTOP* (Roche Diagnostics, Switzerland, Rotkreuz, <http://www.roche.com/>) in the presence of protease inhibitors. Protein concentration was determined using Bradford Assay. Total protein lysate of 6 mg per time point were reduced with dithiothreitol (DTT) at a final concentration of 10 mM at 56°C. Subsequently, lysates were alkylated with 55 mM iodoacetamide. Lysates were diluted 6-fold with 100 mM ammonium bicarbonate and digested overnight with trypsin. Stable isotope labeling by reductive dimethylation of tryptic peptides {#s4b} ---------------------------------------------------------------------- Tryptic peptides were desalted using a Sep-Pak C18 column (Waters, USA, Massachusetts), eluted peptides were lyophilized, and re-suspended in 100 µL of triethylammonium bicarbonate (100 mM). Subsequently, stable isotope dimethyl labeling was performed on the full digested lysate (∼6 mg) as described before [@pone.0017538-Boersema1] using formaldehyde-H~2~ and cyanoborohydride, formaldehyde-D~2~ and cyanoborohydride and formaldehyde-^13^C--D~2~ and cyanoborodeuteride to generate light, intermediate and heavy dimethyl labels, respectively. The light, intermediate and heavy dimethyl labeled samples were mixed in 1∶1∶1 ratio based on total peptide amount, determined by analyzing an aliquot of the labeled samples on a regular LC-MS run and comparing overall peptide signal intensities. Immunoprecipitation (IP) of phosphopeptides {#s4c} ------------------------------------------- Prior to LC-MS analysis, the differentially labeled peptides from the three different digested lysates were mixed, desalted with Sep-Pak C18 column, and lypholized. Labeled peptide mixtures were dissolved in IP buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 1% NOG, and 1x complete mini protease inhibitor cocktail (Roche diagnostics). Agarose-conjugated anti-p-Tyr (pY99) antibodies (Santa Cruz Biotechnology Inc., USA, CA) (prewashed three times with IP buffer) were added into each peptide mixture and incubated overnight at 4°C with gentle rotation. After incubation, the beads were then washed three times with 1 ml of IP buffer followed by two times with 1 ml of water, all at 4°C. Peptides were eluted with 0.15% TFA and centrifuged at 1500 g for 1 min to separate the antibody beads from the eluate. Eluted peptides were desalted and concentrated on STAGE-tips. On-line nanoflow LC-MS {#s4d} ---------------------- Nanoflow LC-MS/MS was performed by coupling an Agilent 1100 HPLC system (Agilent Technologies, Waldbronn, Germany) to a LTQ-Orbitrap mass spectrometer (Thermo Electron, Bremen, Germany) as described previously [@pone.0017538-Raijmakers1]. Dried fractions were reconstituted in 10 µL 0.1 M acetic acid and delivered to a trap column (Aqua^tm^ C18, 5 µm, (Phenomenex, Torrance, CA, USA); 20 mm × 100 µm ID, packed in-house) at 5 µL/min in 100% solvent A (0.1 M acetic acid in water). Subsequently, peptides were transferred to an analytical column (ReproSil-Pur C18-AQ, 3 µm, Dr. Maisch GmbH, Ammerbuch, Germany; 40 cm × 50 µm ID, packed in-house) at ∼100 nL/min in a 2 hour (non-labeled) or 3 hour (stable isotope dimethyl labeled) gradient from 0 to 40% solvent B (0.1 M acetic acid in 8/2 (v/v) acetonitrile/water). The eluent was sprayed via distal coated emitter tips (New Objective), butt-connected to the analytical column. The mass spectrometer was operated in data dependent mode, automatically switching between MS and MS/MS. Full scan MS spectra (from *m/z* 300--1500) were acquired in the Orbitrap with a resolution of 60,000 at *m/z* 400 after accumulation to target value of 500,000. The three most intense ions at a threshold above 5000 were selected for collision-induced fragmentation in the linear ion trap at normalized collision energy of 35% after accumulation to a target value of 10,000. Data analysis {#s4e} ------------- All MS^2^ spectra were converted to single DTA files using Bioworks 3.3 at default settings. Runs were searched using an in-house licensed MASCOT search engine (Mascot version 2.1.0) software platform (Matrix Science, London, UK) against the Human IPI database version 3.36 (labeled sample; 63012 sequences) or version 3.37 (non-labeled sample; 69164 sequences) with carbamidomethyl cysteine as a fixed modification. Light, intermediate and heavy dimethylation of peptide N-termini and lysine residues (for labeled samples only), oxidized methionine and phosphorylation of tyrosine, serine and threonine were set as variable modifications. Trypsin was specified as the proteolytic enzyme and up to two missed cleavages were allowed. The mass tolerance of the precursor ion was set to 5 ppm and for fragment ions 0.6 Da. Peptides were assigned to the first protein hit reported by Mascot. The assignment of phosphorylation sites of identified phosphopeptides was performed by the PTM scoring algorithm implemented in MSQuant as described previously [@pone.0017538-Olsen1]. Individual MS/MS spectra from phosphopeptides were accepted for a Mascot score ≥20 [@pone.0017538-Boersema1]. The FDR at this score was estimated to be less than 3.5% (and less than 1% for tyrosine phosphorylated peptides only) by performing a concatenated decoy database search [@pone.0017538-Boersema1]. All identified phosphopeptides that were found to be differentially phosphorylated were manually validated as previously described[@pone.0017538-Nichols1]. The quality of the spectrum was judged based on a sufficient number of b- and/or y-ions --preferably those that contain the pY residue- to consolidate the peptide sequence. Furthermore, no dominant neutral loss of phosphoric acid should be observed when no additional serine or threonine phosphorylation was annotated to assure phosphorylation of a tyrosine residue. Quantification {#s4f} -------------- Quantification of peptide triplets of which at least one has obtained a Mascot peptide score of 20 was performed using an in-house dimethyl-adapted version of MSQuant [@pone.0017538-Mortensen1], as described previously [@pone.0017538-Boersema1], [@pone.0017538-Boersema2]. Briefly, peptide ratios were obtained by calculating the extracted ion chromatograms (XIC) of the "light", "intermediate" and "heavy" forms of the peptide using the monoisotopic peaks only. The total XIC for each of the peptide forms was obtained by summing the XIC in consecutive MS cycles for the duration of their respective LC-MS peaks in the total ion chromatogram using FT-MS scans. This total XIC was then used to compute the peptide ratio. Quantified proteins were normalized against the Log~2~ of the median of all peptides quantified. StatQuant, an in-house developed program [@pone.0017538-vanBreukelen1], was used for normalization, outlier detection and determination of standard deviation. Ratios of phosphotyrosine levels were normalized to the ratios of (non-specifically binding) non-phosphorylated peptides. Western Blotting {#s4g} ---------------- Cells were lysed in 1X Cell Lysis Buffer (Cell Signaling Technology, Beverly MA, USA) supplemented with 1 mM phenylmethylsulphonyl fluoride (PMSF). Protein concentrations were determined using the DC protein assay (Bio-Rad laboratories Inc., Hercules CA, USA). Twenty micrograms of each sample was mixed with Laemmli buffer and boiled for 5 min at 95°C. All samples were subjected to SDS-PAGE and electro-transferred onto PDVF membranes (0.2 µm, Bio-Rad). Membranes were probed using the corresponding primary antibodies at the indicated dilutions. After incubation with the primary antibodies, appropriate peroxidase-conjugated secondary antibodies (Dako, Denmark) or fluorescent secondary antibodies (LI-COR Biosciences) were used to detect the bound antibodies. Protein bands were visualized either using a chemiluminescence detection reagent ECL Plus (Amersham, GE Healthcare, UK) or LI-COR ODYSSY imaging system (LI-COR Biosciences, Nebraska). All antibodies used are shown in [Table S5](#pone.0017538.s008){ref-type="supplementary-material"}. Human phospho-receptor tyrosine kinase array {#s4h} -------------------------------------------- Analysis of protein expression using the human phospho-RTK antibody array (R&D Systems Inc., Minneapolis) was performed according to manufacturer\'s instructions. Briefly, capture and control antibodies were spotted in duplicates on nitrocellulose membranes and incubated overnight with 300 µg protein lysate. The membrane was then washed extensively with buffer provided and, further incubated with pan anti-phospho-tyrosine antibody conjugated to horseradish peroxidase (HRP). After incubation, arrays were washed and visualized using chemiluminescence ECL Plus (Amersham). Dataset Comparison {#s4i} ------------------ The hESC phosphoproteome was compared to other datasets by mapping known pY site locations obtained from Phospho.ELM (version 8.2) and Rikova *et. al*. [@pone.0017538-Rikova1]. Using the IPI human database, 1382 and 4117 unique pY sites were mapped out from phospho.ELM and Rikova *et. al*. datasets respectively. Overlap was determined by counting the number of identical sites between the datasets. Supporting Information {#s5} ====================== ###### **Activation of FGF-signaling in hESC.** Activation profile of HES-3 cells post-FGF-2 induction, by Western Blotting. Cells were starved of FGF-2 for 5 days and 10 ng/ml of FGF-2 was added at the indicated time. (TIF) ###### Click here for additional data file. ###### **Comparison of hESC cumulative tyrosine phosphorylated dataset.** Overlap of identified phosphotyrosine sites between our dataset (hESC), those deposited in Phospho.ELM (ver. 8.2), and those identified in Rikova et. al. [@pone.0017538-Rikova1]. Of the 735 phospho-peptides identified in FGF-2 stimulated hESCs, 374 (50.9%) peptides are unique to our dataset. (TIF) ###### Click here for additional data file. ###### **Effect of Src Kinase inhibitor on undifferentiated hESC.** Cells were treated with increase concentrations of Src Kinase inhibitor (SU6656) for 6 PD. (A) Morphology of hESC after treatment with SU6656. Cells treated with 4 µM of SU6656 showed significant reduce in cell number when compared to control (untreated hESC) Scale bar  =  500 µm and 100 µm respectively (B) Cell count of SU6656 treated hESC. (C) Pluripotent marker expression of SU6656 treated hESC using quantitative real time PCR. Data were expressed as mean + SEM and results were from triplicate runs. (TIF) ###### Click here for additional data file. ###### **Summary of antibodies used.** (PDF) ###### Click here for additional data file. ###### **List of cumulative phosphopeptides detected from FGF-2 stimulated hESC.** (PDF) ###### Click here for additional data file. ###### **Motif assignment for phosphopeptides from the cumulative dataset.** (PDF) ###### Click here for additional data file. ###### **List of phosphopeptides from the quantitated dataset.** (PDF) ###### Click here for additional data file. ###### **List of phosphopeptides classified into 5 clusters using cluster analysis.** (TIF) ###### Click here for additional data file. ###### **Supplementary Materials and Methods.** (DOC) ###### Click here for additional data file. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work is supported by the Agency for Science Technology and Research (A\*STAR), Singapore and the Netherlands Proteomics Centre (<http://www.netherlandsproteomicscentre.nl>), a program embedded in the Netherlands Genomics Initiative. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: VMYD PJB AJRH AC. Performed the experiments: VMYD PJB LYF SN SL. Analyzed the data: VMYD PJB CP GK DYL JB BS. Wrote the paper: VMYD PJB CP. [^2]: Current address: Department of Proteomics and Signal Transduction, Max Planck Institute of Biochemistry, Martinsried, Germany [^3]: Current address: Lehrstuhl für Bioanalytik, Technische Universität München, Freising, Germany

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by abnormalities in social interaction, impairment in language and communication, restrictive or repetitive interests, and stereotyped behaviours and movements[@b1]. ASD includes autistic disorder, Asperger syndrome, and pervasive developmental disorder not otherwise specified[@b1]. Over the last decade, the prevalence of ASD has been reported as 11.3 per 1,000 with a male-to-female ratio of 3--4:1[@b2]. There is growing evidence that ASD may be influenced by genetic, neurological, environmental and immunological factors[@b3][@b4]. However, the underlying mechanism of ASD has not yet been identified. Behavioural abnormalities are not evident until approximately 12--18 months of age[@b5][@b6]. Furthermore, ASD individuals vary enormously in clinical manifestation, severity, developmental trajectory, and treatment response. This complexity has propelled an intensive search to identify biomarkers to aid clinicians in achieving earlier diagnoses and in predicting treatment response[@b7]. Brain-derived neurotrophic factor (BDNF) is a small protein found throughout the central nervous system (CNS) and peripheral blood. BDNF plays an important regulatory function in cell proliferation, migration, and survival during neurodevelopment and can also modulate axonal and dendritic outgrowth, synapse formation, neurotransmitter release and other neuroplastic processes[@b8][@b9]. Converging lines of evidence implicate the role of BDNF in the pathophysiology of ASD. It is reported that the concentration of BDNF in serum and the CNS are closely correlated in rats[@b10]. However, evidence of such a correlation in humans is still lacking. Therefore, whether altered BDNF values in the periphery reflect altered BDNF levels in the human CNS requires further investigation. Scientists assume that peripheral BDNF levels mirror and indirectly reflect BDNF levels in the brain[@b11]. This indirect measure of BDNF levels is much easier to acquire than direct assessments of BDNF levels in the CNS. For this reason, the concentration of BDNF in peripheral blood may be a useful potential biological marker for ASD. The concept of the "periphery as a window to the brain" has led to an ever-increasing number of clinical studies assessing peripheral BDNF levels in ASD. However, reports of peripheral BDNF levels in ASD are inconsistent. Some studies have reported that serum BDNF is significantly reduced in ASD compared with healthy controls[@b12][@b13], while other studies have reported higher serum BDNF levels in children with ASD compared with controls[@b14][@b15][@b16]. Similarly, two studies that have examined BDNF levels in neonatal specimens from individuals later diagnosed with ASD have yielded inconsistent results[@b17][@b18]. Thus, we undertook a systematic review of studies assessing peripheral BDNF levels in ASD and controls, followed by a series of meta-analyses to provide an overall estimate of the effect size and between-study heterogeneity of the association between peripheral BDNF levels and ASD. Results ======= Literature search ----------------- The initial search yielded a total of 205 citations: 91 from PubMed, 109 from Embase, 3 from Cochrane Library and 2 from reviewing references. After excluding 69 duplicate studies, 59 with irrelevant topics, 18 reviews and 21 letters/meetings, 38 studies of peripheral BDNF levels in ASD were identified and subjected to detailed evaluation. Subsequently, 16 studies were excluded due to irrelevant outcomes. Seven reports that lacked sufficient data (raw data mean and standard deviation (SD) or median and interquartile range (IQR)) were also excluded. One report was excluded because the ASD was comorbid with attention deficit hyperactivity disorder. Finally, 14 studies, including 2,707 participants and 1,131 incident cases, fulfilled all of the inclusion criteria and were included in the meta-analysis. A detailed flow chart of the search and selection process is presented in [Fig. 1](#f1){ref-type="fig"}. Study characteristics --------------------- The characteristics of the fourteen selected studies are presented in [Table 1](#t1){ref-type="table"}. All of the studies were published between 2001 and 2016. 14 studies including 2,707 participants and 1,131 incident cases were included in this meta-analysis. Three studies were conducted in the United States[@b18][@b19][@b20], three in China[@b16][@b21][@b22], three in Japan[@b13][@b23][@b24], one in Denmark[@b17], one in Ireland[@b14], one in Saudi Arabia[@b25], one in India[@b26], and one in Italy[@b15]. The sample sizes varied widely, ranging from 18[@b13][@b24] to 359[@b17] ASD individuals and from 16[@b24] to 741[@b17] controls. Similarly, the mean age of the ASD and control individuals varied broadly and ranged from 0[@b17][@b18][@b19][@b20] to 22.2 ± 2.2[@b13] years old. The systematic review identified two different biomaterials used for BDNF assays: serum[@b13][@b15][@b16][@b17][@b18][@b19][@b20][@b21][@b22][@b23][@b24][@b25][@b26] and plasma[@b14]. Moreover, 11 studies assessed the BDNF levels using enzyme-linked immunosorbent assay (ELISA) as the analytical procedure[@b13][@b14][@b15][@b16][@b20][@b21][@b22][@b23][@b24][@b25][@b26], while 3 adopted Luminex[@b17][@b19][@b20] and 1 used recycling immunoaffinity chromatography (RIAC)[@b18]. Meta-analysis of peripheral BDNF levels in ASD ---------------------------------------------- Based on estimates pooled from 14 studies, a significantly higher level of BDNF was found in ASD compared with controls \[standardized mean difference (SMD) = 0.63, 95% CI = 0.18--1.08; *P* = 0.006)\]. However, there was significant statistical heterogeneity across studies (I^2^ = 96%, *P* \< 0.001) ([Fig. 2](#f2){ref-type="fig"}). Quality evaluation ------------------ The results of the quality assessment of the included studies are shown in [Table 2](#t2){ref-type="table"}. Fourteen studies were of high quality, with an average score of 7.4. Publication bias ---------------- The funnel plot was symmetric ([Fig. 3](#f3){ref-type="fig"}). Moreover, Begg's and Egger's tests did not reveal significant evidence of publication bias among the included studies (Begg's test, *P* = 1.000; Egger's test, *P* = 0.156). Subgroup analysis ----------------- Of the fourteen studies included in this meta-analysis, thirteen described BDNF levels assessed in serum. One study reported result assessed in plasma. Higher BDNF levels were found in ASD compared with controls in both subgroups of studies. The pooled SMD was 0.58 (95% CI = 0.11--1.04, *P* = 0.02) for BDNF levels assessed in serum and 1.27 (95% CI = 0.92--1.61, *P* \< 0.001) for BDNF levels assessed in plasma ([Table 3](#t3){ref-type="table"}). We conducted a meta-analysis of the five studies of ASD and the nine studies of autism. The pooled SMD was 0.64 (95% CI = 0.1--1.19, *P* = 0.02) for the studies of ASD and 0.55 (95% CI = −0.03--1.13, *P* = 0.06) for the studies of autism ([Table 3](#t3){ref-type="table"}). We conducted a meta-analysis of the ten studies of childhood and the four studies of adulthood. The pooled SMD was 0.78 (95% CI = 0.31--1.26, *P* = 0.001) for the studies of childhood and 0.04 (95% CI = −1.72--1.80, *P* = 0.97) for the studies of adulthood ([Table 3](#t3){ref-type="table"}). We also conducted a meta-analysis of the studies based on the analytic technology used: Luminex, ELISA or RIAC. The effect size of the difference in BDNF levels measured in ASD and controls when these different analytical technologies were applied was 0.66 (95% CI = 0.12--1.2, *P* = 0.02) for ELISA, −0.03 (95% CI = −0.14--0.08, *P* = 0.56) for Luminex, and 1.57 (95% CI = 1.16--1.98, *P* \< 0.001) for RIAC ([Table 3](#t3){ref-type="table"}). Meta-analysis of the studies based on both subject age and the analytical technology used showed that the pooled SMD was 0.93 (95% CI = 0.54--1.33, *P* \< 0.001) for the studies of childhood and 0.04 (95% CI = −1.72--1.8, *P* = 0.97) for the studies of adulthood when applying ELISA. The pooled SMD was −0.03 (95% CI = −0.14--0.08, *P* = 0.56) for the studies of childhood and used Luminex as the analytical method. The pooled SMD was 1.57 (95% CI = 1.16--1.98, *P* \< 0.001) for the study of childhood and employed RIAC as the analytical method ([Table 3](#t3){ref-type="table"}). Meta-regression analysis ------------------------ We performed meta-regression analyses in an exploratory attempt to identify the sources of heterogeneity between the studies and the effect of moderators. Using univariable meta-regression models, we found a positive relationship between gender and BDNF levels (slope = 0.06, 95% CI = 0.01--0.11; *P* = 0.024). There was no relationship between the mean age, study design or confounders adjustment and BDNF levels ([Table 4](#t4){ref-type="table"}). Sensitivity analysis -------------------- The influence of each study on the overall estimate was assessed by removing studies one by one and comparing the pooled estimate from the remaining thirteen studies to the pooled estimate from all fourteen studies. The results revealed higher peripheral BDNF levels in ASD compared with controls in all 14 analyses, indicating that the removal of any one study would not alter the overall results. Discussion ========== Over the fourteen studies, 2,707 participants and 1,131 incident cases were included in this meta-analysis. A random-effect model was used to compute the pooled estimates because there was significant between-study heterogeneity. The pooled SMD indicated that the peripheral BDNF level was higher in ASD compared to the controls. Sensitivity analysis showed that the pooled results were robust. The symmetrical funnel plot and the results of Begg's and Egger's tests also suggested the lack of significant publication bias. BDNF, which is the most abundant neurotrophin in the CNS, can cross the blood-brain barrier. Therefore, its levels in serum and plasma are highly correlated with the levels in cerebrospinal fluid (r = 0.8)[@b10][@b27]. Studies have shown that BDNF protein levels in serum and the brain are similar in developmental rats, with a positive correlation between serum and cortical BDNF levels[@b10]. In addition, Klein *et al*.[@b28] reported a significant positive correlation between whole blood and hippocampal BDNF levels in rats. Furthermore, they demonstrated that blood and plasma BDNF levels also reflected brain-tissue BDNF levels in rats and pigs. However, BDNF levels in blood are undetectable in other species, such as mice. Additionally, evidence of such a correlation in humans is still lacking. Therefore, whether altered BDNF values in the periphery reflect altered BDNF levels in the human CNS requires further investigation. Interestingly, we found that peripheral BDNF levels were higher in the ASD samples from childhood but not from adulthood. Evidence from animal models of ASD suggests that BDNF levels increase in the foetal brain[@b29][@b30]. In the ASD model of sodium valproate (VPA) exposure in utero, VPA administration has been shown to increase BDNF protein levels in the feotal mouse brain 5- to 6-fold *in vitro* and *in vivo*[@b29]. Higher BDNF expression in the foetal brain was also demonstrated in the BTBR T+tf/J ASD model[@b30]. BDNF hyperactivity during early life may play an aetiological role in ASD. Early BDNF hyperactivity could result in the overgrowth of brain tissue, which is found in many ASD children[@b31]. Increased BDNF levels in children with ASD may reflect a regional compensatory mechanism in response to late brain maturation[@b32]. Therefore, accurately detecting BDNF levels is important to analyse the brain development in children with ASD. Different analytical technologies for the assessment of BDNF were used across the studies. Four studies of adulthood and seven studies of childhood used ELISA to analyse BDNF levels. BDNF levels were found to be significantly increased in the childhood studies but not in the adulthood studies, further demonstrating the difference in peripheral BDNF levels between the age groups. Furthermore, subgroup analyses revealed that BDNF levels were significantly higher in children with ASD compared with controls when the levels were measured through ELISA and RIAC, but not when they were measured using Luminex. Therefore, the ELISA and RIAC assays may have been more sensitive than the Luminex assay in detecting BDNF levels in the included studies. Significant heterogeneity was found in this analysis. To clarity the sources of heterogeneity and make a more comprehensive analysis, we performed the subgroup analyses and meta-regression. We found that three different factors contributed to the heterogeneity. First, the positive SMD is significantly higher in plasma-based samples than that in serum-based samples ([Table 3](#t3){ref-type="table"}). This difference may due to the difference of clotting factors between plasma and serum. Second, that different analytic technologies used in this study may also contribute to heterogeneity ([Table 3](#t3){ref-type="table"}). Finally, the gender factor especially for the percentage of male was also one of the reasons for the heterogeneity ([Table 4](#t4){ref-type="table"}). Our study had some advantages. First, this is the first comprehensive meta-analysis conducted to assess the association between peripheral BDNF levels and ASD. Second, the sensitivity analysis indicated that the removal of individual studies did not alter the final results, which increased the robustness of the conclusions of this analysis. Third, no significant publication bias was detected, suggesting that the results are reliable. However, our meta-analysis also had some limitations. Firstly, detailed information regarding medication use was not provided in some studies. Medication could have influenced peripheral BDNF levels in ASD. Thus, future research should take into consideration the possible effect of medication on peripheral BDNF levels. Secondly, the potential correlation between the severity of ASD and BDNF levels was not assessed because few studies have analysed the relationship between the clinical severity of ASD and BDNF levels. The association between the severity of ASD and BDNF should be evaluated in future studies. Thirdly, the meta-analysis of BDNF levels in ASD generated a pooled result that largely originated from cross-sectional studies. Therefore, we cannot draw any conclusions about causality. We do not know if an increase in BDNF levels is a cause or consequence of ASD development. Fourthly, there was significant heterogeneity in this analysis, which may have affected the precision of the overall results. In conclusion, this meta-analysis indicated that peripheral BDNF levels are higher in ASD compared with controls, suggesting that peripheral BDNF levels may serve as a potential biomarker for the diagnosis of ASD. Future studies need to clarify the influence of medication, the causal nature of the relationship and the association between the severity of clinical ASD symptoms and BDNF levels. Methods ======= Literature search ----------------- Two authors searched the PubMed, Embase and Cochrane Library databases for relevant articles published before February 2016 using both Medical Subject Heading (MeSH) terms and the free text terms "ASD," "autism," "autism spectrum disorder," "autistic disorder," "Asperger syndrome," "pervasive developmental disorder," and "BDNF," "brain-derived neurotrophic factor," and "peripheral," "levels," "serum," "plasma," "urine," "saliva," "blood," "platelets," "cerebrospinal fluid," "red blood cells." In addition, the references of the included articles and previous meta-analyses were manually searched to identify additional studies. We restricted the search to human studies published in English. The titles and abstracts of the retrieved studies were reviewed to exclude studies that were clearly irrelevant. Then, two authors independently read the full text of the remaining studies to assess their eligibility according to the inclusion criteria. Disagreements about the inclusion/exclusion of a study were resolved by a third author, who independently examined the studies, and consensus was reached. Study selection --------------- Studies were eligible for analysis if they met all of the following criteria: (1) they were about the association between peripheral BDNF levels and ASD *in vivo*; and (2) they provided the raw data mean and SD or median and IQR. The following types of studies were excluded: (1) reviews, case reports, case-only studies, animal studies, and simple commentaries; (2) overlapping publications; (3) publications lacking measures of peripheral BDNF levels, including pharmacological, genetic, brain imaging, and post-mortem studies; (4) studies in which ASD was comorbid with other conditions; and (5) studies that showed BDNF levels in dot plot and histogram format but did not provide numerical results. Data extraction --------------- Two authors extracted data from the included articles, with particular regard to the following: first author's name, publication year, country of region, number of cases and controls, age of subjects (mean ± SD), percentage of females and males, analytical technology employed, biomaterial evaluated, BDNF level (mean ± SD), unit of measure and adjusted confounders. BDNF levels were measured from serum and plasma, and different units of measurement were used across studies (ng/ml or pg/ml). Therefore, we reported all BDNF levels in ng/ml (1 ng/ml = 1000 pg/ml). If the data were presented using the median (IQR) format, then the formula "IQR/1.35" was used to calculate SD. If participants overlapped between studies, the study with the largest sample size was included in the meta-analysis. Quality evaluation ------------------ Two authors independently assessed the quality of each included study using the Newcastle-Ottawa Quality Assessment Scale (NOS) to determine the quality of selection, comparability, exposure, and outcome of study participants, with a maximum score of 9 points. We divided the study quality into three categories: (1) high quality (scored 7--9); (2) moderate quality (scored 4--6); and (3) low quality (scored 0--3). Disagreements were resolved through mutual discussion. Statistical analysis -------------------- The SMD was used to assess the association between peripheral BDNF levels and ASD. We pooled the SMD across studies using the Mantel-Haenszel formula (fixed-effect model) or the DerSimonian-Laird formula (random-effect model). A fixed-effect model was chosen when heterogeneity was low; otherwise, a random-effect model was adopted. Across-study heterogeneity was evaluated using the I^2^ and Q statistics; these statistics provide a quantitative measure of inconsistency across studies, with suggested thresholds for low (25--50%), moderate (50--75%) and high (\>75%) heterogeneity. The Q statistic was considered significant if *P* \< 0.1, and I^2^ \> 50% indicated high heterogeneity. The results of the analyses are shown in forest plots. Potential publication bias was assessed via visual inspection of the funnel plot. Begg's and Egger's tests were used to estimate the severity of publication bias, with *P* \< 0.05 considered statistically significant. We analysed subgroups of studies to examine the source of potential heterogeneity based on the biomaterial used (serum or plasma), condition (ASD or autism) and subject age (childhood \< 18 years old or adulthood ≥18 years old). The analytical technologies employed included Luminex, ELISA and RIAC. Unrestricted maximum likelihood random effects meta-regressions of effect size were performed with mean age, gender (% male), study design and confounder adjustment as moderators to determine whether these covariates influenced the effect size. We carried out the sensitivity analysis by removing studies one by one and comparing the SMD of the remaining studies to the SMD for all studies. Statistical analysis was performed using Stata 12.0 (Stata Corp, College Station, Texas, USA) and Cochrane Collaboration Review Manager 5.1.2 (Cochrane Collaboration, Oxford, UK) software. Additional Information ====================== **How to cite this article**: Zheng, Z. *et al*. Peripheral brain-derived neurotrophic factor in autism spectrum disorder: a systematic review and meta-analysis. *Sci. Rep*. **6**, 31241; doi: 10.1038/srep31241 (2016). This work was supported by the National Science Foundation of China (No. 81330016 to Dezhi Mu and No. 81270724 to Yi Qu), a grant from the Major State Basic Research Development Program (2013CB967404), grants from the Ministry of Education of China (313037, IRT0935), a grant from the State Commission of Science Technology of China (2012BAI04B04), grants from the Science and Technology Bureau of Sichuan Province (2014SZ0149, 2016TD0002), and a grant from the clinical discipline program (neonatology) of the Ministry of Health of China (1311200003303). **Author Contributions** Z.Z. and L.Z. designed the study and wrote the manuscript. T.Z. performed the literature searches and collected the data. J.H. and Y.Q. conducted the statistical analysis. D.M. revised the manuscript. {#f1} {#f2} {#f3} ###### Characteristics of the fourteen studies included in the meta-analysis. Author Year Country Sample size ASD controls Age Mean ± SD (range) ASD controls Sex (F/M) ASD controls Analytical technology Biomaterial BDNF Mean ± SD ASD controls Unit of measure Adjusted founders ------------------- -------------- -------------------------- ------------------------------------ ------------------------ ----------------------- ------------- ------------------------------- ----------------- ------------------------------------ Abdallah[@b17] Denmark 359741 00 68/291146/595 Luminex Serum 6.77 ± 3.466.94 ± 3.51 ng/ml Age, gender Correia[@b14] Ireland 14650 7.17.5 nr ELISA Plasma 40.44 ± 13.8723.26 ± 12.34 ng/ml nr Croen[@b19] America 84159 00 nr Luminex Serum 0.0541 ± 0.0400.0537 ± 0.048 ng/ml Age, gender Halepoto[@b25] Saudi Arabia 6025 6 ± 1.727.04 ± 1.74 nr ELISA Serum 0.392 ± 0.2430.290 ± 0.162 ng/ml Age, gender Hashimoto[@b13] Japan 1818 21.2 ± 2.122.2 ± 2.2 0/180/18 ELISA Serum 25.6 ± 2.1561.6 ± 10.9 ng/ml Age Kasarpalkar[@b26] India 4829 7.47.4 nr ELISA Serum 250.56 ± 92.7225.16 ± 79.54 ng/ml Age Katoh-Semba[@b23] Japan 56218 \<60\<60 nr ELISA Serum 0.359 ± 0.2770.331 ± 0.187 ng/ml Age Miyazaki[@b24] Japan 1816 7.6 ± 6.123.3 ± 0.9 1/1711/5 ELISA Serum 25.22 ± 2.4517.5 ± 2.00 ng/ml nr Nelson[@b18] America 6954 00 8/6127/27 RIAC Serum 0.0374 ± 0.01990.0133 ± 0.005 ng/ml nr Nelson[@b20] America 2720 00 nr Luminex and ELISA Serum 3.404 ± 1.1313.299 ± 0.844 ng/ml Mean gestational age, birth weight Ricci[@b15] Italy 2929 2--212--21 2/272/27 ELISA Serum 0.728 ± 0.4550.351 ± 0.347 ng/ml Age, gender Wang[@b16] China 7575 4.0 ± 1.254.0 ± 1.25 13/6213/62 ELISA Serum 17.59 ± 5.5511.21 ± 2.79 ng/ml Age, gender Zhang[@b21] China 6060 3.78 ± 1.223.78 ± 1.22 12/4812/48 ELISA Serum 17.6 ± 5.711.5 ± 3.1 ng/ml Age, gender Meng[@b22] China 8282 4.02 ± 1.274.02 ± 1.27 17/6517/65 ELISA Serum 17.75 ± 5.4311.49 ± 2.85 ng/ml Age, gender ASD: autism spectrum disorder; nr: not reported; F/M: female/male; RIAC: recycling immunoaffinity chromatography; ELISA: enzyme-linked immunosorbent assay. ###### Quality assessment of the included studies based on the Newcastle--Ottawa Scale. Publication year Study design Selection Comparability Exposure/Outcome Total scores ------------------- --------------- ----------- --------------- ------------------ -------------- Abdallah[@b17] Cross-section ★★★★ ★★ ★★ 8 Correia[@b14] Cross-section ★★★ ★ ★★ 6 Croen[@b19] Case-control ★★★★ ★★ ★★ 8 Halepoto[@b25] Cross-section ★★★★ ★★ ★★ 8 Hashimoto[@b13] Cross-section ★★★ ★★ ★★ 7 Kasarpalkar[@b26] Cross-section ★★★★ ★★ ★★ 8 Katoh-Semba[@b23] Cross-section ★★★ ★★ ★★ 7 Miyazaki[@b24] Cross-section ★★★★ ★ ★★ 7 Nelson[@b18] Case-control ★★★★ ★ ★★ 7 Nelson[@b20] Case-control ★★★★ ★★ ★★ 8 Ricci[@b15] Cross-section ★★★ ★★ ★★ 7 Wang[@b16] Cross-section ★★★★ ★★ ★★ 8 Zhang[@b21] Cross-section ★★★★ ★★ ★★ 8 Meng[@b22] Cross-section ★★★ ★★ ★★ 7 ###### Summary results of peripheral BDNF levels in persons with ASD and healthy controls. Variables No. of comparisions No. of subjects Meta-analysis Heterogeneity Test for subgroup differences ------------------------------------- --------------------- ----------------- --------------- --------------- ------------------------------- ------ --------- ---------------- ---------------- ---------------- ---------------- Biomaterial  Serum 13 985 1526 0.58 0.11 1.04 0.02 96 \<0.001 81.6 0.02  Plasma 1 146 50 1.27 0.92 1.61 \<0.001 Not applicable Not applicable     Condition  ASD 5 538 974 0.64 0.1 1.19 0.02 92 \<0.001 0 0.82  autism 9 593 602 0.55 −0.03 1.13 0.06 94 \<0.001     Subject age  Childhood 10 1010 1295 0.78 0.31 1.26 0.001 96 \<0.001 0 0.42  Adulthood 4 121 281 0.04 −1.72 1.8 0.97 97 \<0.001     Analytical technology  ELISA 11 619 622 0.66 0.12 1.2 0.02 94 \<0.001 96.6 \<0.001  Luminex 3 470 920 −0.03 −0.14 0.08 0.56 0 0.84      RIAC 1 69 54 1.57 1.16 1.98 \<0.001 Not applicable Not applicable     Subject age & analytical technology Childhood  ELISA 7 498 341 0.93 0.54 1.33 \<0.001 85 \<0.001 97.2 \<0.001  Luminex 3 470 920 −0.03 −0.14 0.08 0.56 0 0.84      RIAC 1 69 54 1.57 1.16 1.98 \<0.001 Not applicable Not applicable     Adulthood  ELISA 4 121 281 0.04 −1.72 1.8 0.97 97 \<0.001 Not applicable Not applicable ASD: autism spectrum disorder; No.: number; CI: confidence interval; RIAC: recycling immunoaffinity chromatography; ELISA: enzyme-linked immunosorbent assay. ###### Meta-regression of peripheral BDNF levels in persons with ASD and healthy controls. Moderator No. of comparisions No. of subjects Meta-regression Proportion of variance explained ------------------------ --------------------- ----------------- ----------------- ---------------------------------- --------- -------- ------- ------- Age (mean, years) 12 1046 1329 −0.1028 −0.285 0.079 0.237 4.66 Gender (% male) 8 710 1075 0.0628 0.0117 0.114 0.024 58.18 Study design 14 1131 1576 0.248 −1.9765 2.4726 0.812 0 Confounders adjustment 14 1131 1576 −1.8329 −3.992 0.326 0.089 19.18 ASD: autism spectrum disorder; No.: number; CI: confidence interval. [^1]: These authors contributed equally to this work.

All small RNA data series were submitted to NCBI Gene Expression Omnibus (GEO) with accession number GSE29053. Introduction {#sec001} ============ Hibernation (multiday torpor) and daily torpor in heterothermic mammals are characterized by substantial, temporal reductions of energy expenditure and body temperature (T~b~). Torpor is often expressed when animals are exposed to low ambient temperature (T~a~), and the decrease in energy use is crucial for survival during adverse conditions and food shortages \[[@pone.0135064.ref001]\]. Mammalian hibernation is an extreme example of hypometabolism with metabolic rate depression often to \<5% of the euthermic rate, a minimum T~b~ of approximately 0°C, and cell preservation during prolonged dormancy periods \[[@pone.0135064.ref002], [@pone.0135064.ref003]\]. The physiological changes, such as the decrease and subsequent increase of metabolism and T~b~ during entry and arousal from torpor, are active and adaptive processes. They are often initiated in response to changes in environmental temperature and photoperiod and require the coordination of core and peripheral tissues by the brain \[[@pone.0135064.ref004], [@pone.0135064.ref005]\]. Importantly, the carbohydrate oxidation pathway shifts upon hormone activation towards lipid metabolism, and fat (mainly stored in white adipose tissue, WAT) becomes the primary energy source during the hibernation season \[[@pone.0135064.ref006]\]. Considering the brain's important role in integrating peripheral tissues and WAT to provide energy during hibernation, we hypothesized that to adequately perform these tasks, both must be activated during hibernation, in contrast to other tissues and organs, where metabolism is generally suppressed \[[@pone.0135064.ref007]\]. A detailed understanding of the molecular mechanisms of this phenomenon would improve our understanding of hibernation at the subcellular level, but may also be useful in human medicine for improving organ transplantation, preventing skeletal muscle atrophy during long-term bed rest, and suppressing carcinogenesis \[[@pone.0135064.ref007]\]. In the past ten years, molecules that regulate hibernation and the molecular mechanisms underlying hibernation have been identified by transcriptome analysis \[[@pone.0135064.ref008]--[@pone.0135064.ref012]\]. Thousands of genes and proteins are differentially expressed in many species during hibernation \[[@pone.0135064.ref012]--[@pone.0135064.ref017]\]. However, the precise and timely control of gene expression during hibernation remains unclear. Previous studies indicated that post-transcriptional regulation occurs during hibernation \[[@pone.0135064.ref018]\]. MicroRNAs (miRNAs) are short RNA molecules (18--25nt), and they are known to play key roles in directing post-transcriptional activity of genes involved in all cellular processes, including differentiation and development \[[@pone.0135064.ref019]\]. Since Morin and colleagues \[[@pone.0135064.ref020]\] first proved that miRNAs regulate ground squirrel \[*Ictidomys* (*Spermophilus) tridecemlineatus*\] hibernation, many detailed studies have analyzed miRNAs during torpor \[[@pone.0135064.ref018], [@pone.0135064.ref021]--[@pone.0135064.ref025]\] and found that miRNAs are involved in cold adaptation, epigenetic control and metabolic rate depression \[[@pone.0135064.ref026]--[@pone.0135064.ref028]\]. Differential miRNA expression, and thus their target genes, is an important component of the molecular adaptation required by organisms to respond to environmental changes, especially heterothermic mammals \[[@pone.0135064.ref018], [@pone.0135064.ref026]--[@pone.0135064.ref028]\]. There are more than 1100 species of bats (order Chiroptera), and they are among the most geographically widespread mammals \[[@pone.0135064.ref029]\]. Most bat families include species that enter some form of torpor, whereas multiday torpor (i.e., hibernation) occurs in at least five families \[[@pone.0135064.ref030]\]. The wide taxonomic diversity of heterothermy in bats suggests that bats are a good model to explore the evolutionary history of mammalian heterothermy \[[@pone.0135064.ref031]\]. Importantly, unlike ground squirrels, for which most molecular work on hibernation has been conducted, bats are not strongly seasonal in their torpor expression and many use multiday torpor throughout the year \[[@pone.0135064.ref032]\]. Therefore, a systematic study of genes or proteins involved in bat hibernation by comparing active and hibernating individuals will help us better understand the molecular mechanisms of hibernation and to explore the evolution of mammalian heterothermy \[[@pone.0135064.ref011], [@pone.0135064.ref031], [@pone.0135064.ref033], [@pone.0135064.ref034]\]. However, a systematic analysis of small RNAs, especially miRNAs, with regard to function and evolution of torpor in bats is currently lacking. The latest 7× coverage genome sequence of *Myotis lucifugus* (<http://www.ensembl.org/Myotis_lucifugus/Info/Index>) provides us with the opportunity to survey miRNA expression in hibernating bats and to conduct comparative analyses with sequences from other metazoan species. Moreover, the RNA-seq sequencing-by-synthesis technique allows to sequence millions of short cDNAs and increases sensitivity by reducing personal error in library construction \[[@pone.0135064.ref035]\]. Based on the Illumina/Solexa sequencing platform and combined with qPCR validation, we performed the first in-depth miRNA analysis on two key tissues (brain and WAT) of the hibernating bat *Myotis ricketti*, which is congeneric with *M*. *lucifugus*. Our study aimed to discover bat-specific miRNAs, illustrate their potential role during bat hibernation, and contribute to the understanding of the evolution of mammalian heterothermy. Methods {#sec002} ======= Ethics statement {#sec003} ---------------- All procedures involving animals were performed by strictly following the Guidelines and Regulations for the Administration of Laboratory Animals (Decree No. 2 of the State Science and Technology Commission of the People's Republic of China on November 14, 1988) and were approved by the Animal Ethics Committee of East China Normal University (20080729). Sample and RNA preparation {#sec004} -------------------------- A total of six *Myotis ricketti* individuals with a body mass of 18--22 g were used in this study. Three active *M*. *ricketti* bats (T~b~\~36°C) were captured using mist nets in the Fangshan area (Beijing, China; 39°48′N, 115°42′E) on October 26, 2008 (T~a~ = 21°C). On March 6, 2009, three bats in deep hibernation (T~b~\~9°C) were captured in the same location (T~a~ = 7°C). The field studies did not involve endangered or protected species, and no specific permissions were required for *M*. *ricketti* in the Fangshan area. The bats were sacrificed humanely by decapitation immediately after measuring T~b~ and body mass. Whole brain and white intra-abdominal adipose tissue (WAT) of hibernating and euthermic bats were rapidly removed, immediately frozen in liquid nitrogen, and stored at -80°C until RNA extraction. Total RNA was isolated from entire brains and adipose tissues (\~0.2 g) of six bats using the RNAiso kit (TakaRa, Japan) according to the manufacturer's instructions, and the concentration and RNA Integrity Number (RIN) was determined by an Agilent 2100 bioanalyzer. RNA samples were pooled to construct the transcriptome sequencing library as described previously \[[@pone.0135064.ref012], [@pone.0135064.ref021]\]. Briefly, total RNA from bat brains of the same stage were mixed in equal amounts (10 μg each) into two pooled samples, representing the hibernation and active state, hereafter referred to as HB (hibernating brain) and AB (active brain). Similarly, total WAT RNA from hibernating and active bats (10 μg each) were pooled and referred to as HA (hibernating WAT), or AA (active WAT). In total, four RNA pools (two tissues in two physiological states) were used to construct small RNA libraries. The RNA quality and quantity were as follows: HB: 480 ng/μl, RIN = 8.3, 28S/18S = 1.2; AB: 2210 ng/μl, RIN = 8.1, 28S/18S = 1.5; HA: 390 ng/μl, RIN = 8.4, 28S/18S = 1.4; AA: 1440 ng/μl, RIN = 7.5, 28S/18S = 1.5. Construction and high-throughput sequencing of small RNA libraries {#sec005} ------------------------------------------------------------------ Small RNA cDNA libraries were constructed as described previously \[[@pone.0135064.ref036]\]. Briefly, small RNA (18\~33 nt) was purified and enriched from each total RNA sample by polyacrylamide gel electrophoresis (PAGE), and 10 μg small RNA was ligated with the proprietary adapters used for cDNA synthesis and library construction, then Solexa sequencing-by-synthesis was performed (Illumina). All small RNA data series were submitted to NCBI Gene Expression Omnibus (GEO) with the accession number GSE29053. Filter of small RNA reads {#sec006} ------------------------- Individual sequence reads with base quality scores were produced by Illumina/Solexa. After removing contaminant reads (adapters, low quality and redundant reads), clean unique reads were mapped onto the *Myotis lucifugus* genome (<http://asia.ensembl.org/Myotis_lucifugus/Info/Index>) using the Bowtie program \[[@pone.0135064.ref037]\]. Perfectly mapped reads were scanned against the metazoan mature miRNA in the Sanger miRBase (Release 19) \[[@pone.0135064.ref038]\] to identify orthologs of known miRNAs using the Patscan program \[[@pone.0135064.ref039]\] with two mismatches allowed. Non-conserved unique reads were screened against Rfam databases (Release 10) \[[@pone.0135064.ref040]\] to filter the sequences of tRNA, rRNA, snoRNA, and other ncRNAs except miRNAs using Bowtie. Reads that matched to the genome more than 20 times were removed by miRDeep \[[@pone.0135064.ref041]\], and reads sequenced only one time were also removed. Finally, the remaining reads were considered potential miRNA reads and were used for miRNA identification. MiRNA identification {#sec007} -------------------- MiRNA and its antisense strand can both be sequenced from the same precursor. Thus, to avoid repeated predictions and to reduce calculations, candidate reads whose distance in the reference genome were \<200 nt were combined and examined as one genomic block. For each block, 150 nt of upstream and downstream extensions were extracted for secondary structure prediction. First, inverted repeats (IR) with stem-loop or hairpin structures were identified by Einverted of Emboss \[[@pone.0135064.ref042]\], with the following parameters: threshold = 30, match score = 3, mismatch score = 3, gap penalty = 6, and maximum repeat length = 240, as described previously \[[@pone.0135064.ref043]\]. The IR secondary structure was then predicted by RNAfold \[[@pone.0135064.ref044]\] with 10 nt upstream and downstream extensions and evaluated by MirCheck \[[@pone.0135064.ref043]\]. Predicted precursors that miRNA and antisense strand can be found in both arms were deemed miRNA candidates. Moreover, if the candidates have two or more unique reads resulted from the ±2 nt bias during cleavage and located at mature positions, they were deemed highly probable. When several length variants of the same miRNA were sequenced, only variants with the highest representation were considered. Finally, the miRNA candidates were submitted again to miRBase, and the miRNA precursors (hairpins) that passed MirCheck were manually inspected against the canonical miRNA structure to remove false predictions. Repeat-derived siRNA detection {#sec008} ------------------------------ To screen repeat-derived small interfering RNA (siRNAs), the repeat sequences of the *M*. *lucifugus* genome were annotated by Repeatmasker (<http://www.repeatmasker.org/>) \[[@pone.0135064.ref045]\]. Reads that perfectly matched the *M*. *lucifugus* genome were aligned to repetitive elements using Bowtie, and reads that perfectly matched the repeats were considered genomic repeats-derived siRNAs. Differential miRNA expression analysis and target prediction {#sec009} ------------------------------------------------------------ Differentially expressed miRNAs that were statistically significant in relative abundance (reflected by Transcript Per Million, TPM) between hibernating and active states were identified by the edgeR function in Bioconductor \[[@pone.0135064.ref046]\]. Empirical Bayes estimation and exact tests based on the negative binomial distribution were used, and *P*≤0.01 & │logFC│≥1 considered statistically significant. As there is no or little information about 3' untranslated region (UTR) of *M*. *lucifugus* reference genes in database of Ensembl (<http://www.ensembl.org/Myotis_lucifugus/Info/Index>), to predict miRNA target genes, we first identified the orthologs of *M*. *lucifugus* genes by searching against *Homo sapiens* mRNA (<http://www.ncbi.nlm.nih.gov/RefSeq/>) with tblastx (e value ≤1e-10 and identity ≥60%) \[[@pone.0135064.ref047]\]. Orthologs with the best hits were kept. The *H*. *sapiens* mRNAs were used to analyze the 3' UTR lengths, and 93.6% (67,594/72,204) of human mRNAs had a 3' UTR less than 3 Kb (see [S1 Fig](#pone.0135064.s001){ref-type="supplementary-material"}), excluding 32,559 mRNAs that did not have a 3' UTR. Thus, the 3 Kb region downstream of *M*. *lucifugus* reference gene coding sequences were extracted and aligned with the 3'UTR of *H*. *sapiens* mRNA orthologs by CLUSTALW \[[@pone.0135064.ref048]\] to trim the candidate sequences of 3' UTR of *M*. *lucifugus*. For *M*. *lucifugus* reference genes that did not have a human ortholog, the 3 Kb region downstream of the coding sequence was considered the 3'UTR. The miRNA target was then predicted using the PITA program based on the interaction between miRNAs and their targets with the default criterion and ΔΔG≤−10 kcal/mol \[[@pone.0135064.ref049]\]. Finally, target sequence Gene Ontology (GO) annotation was performed by Interproscan \[[@pone.0135064.ref050]\] and target sequences were compared to the Kyoto Encyclopedia of Genes and Genomes database (KEGG, release 50) by BLASTX at E values ≤1e-10 \[[@pone.0135064.ref047], [@pone.0135064.ref051]\]. A Perl script was used to retrieve the KO (KEGG Orthology) information from BLAST results and to establish pathway associations between target genes and databases. Quantitative miRNA expression by qPCR assay {#sec010} ------------------------------------------- Stem-loop reverse transcription (RT) qPCR was performed to quantify the expression level of 10 differentially expressed miRNAs obtained by Solexa sequencing in brains and WAT as described previously \[[@pone.0135064.ref052]\], and 5S rRNA was used as an endogenous control. Primers used in this study are listed in [S1 Table](#pone.0135064.s004){ref-type="supplementary-material"}. Each RT reaction contained 1 μg total RNA, 1× Reaction Buffer, dNTPs (5 mM each), 50 nM miRNA-specific stem-loop RT primer, 100 U RevertAid Premium Reverse Transcriptase and 0.5 U RiboLock RNase Inhibitor. RT reactions were incubated at 37°C for 30 min, 42°C for 60 min and 85°C for 5 min. PCR reactions were performed in a total of 20 μl, including 0.8 μl RT product, 10 μl 2× SYBR Green I Master Mix, 6 μM each of forward primer and reverse primer with the following program: 95°C for 10 min, followed by 40 cycles of 95°C for 15 s, 60°C for 30 s and 72°C for 15 s, then 95°C for 15 s, 60°C for 1 min and a final hold at 4°C. All reactions were performed on three biological replicates in each tissue and each physiological state, and each was run three times. Negative controls containing all reagents except template were included on each reaction plate. The 2^−ΔΔCT^ method was used to calculate relative expression (fold change); and data were analyzed by One-Way ANOVA using SPSS 18.0 software. Results {#sec011} ======= Small RNA sequencing and statistics {#sec012} ----------------------------------- We isolated total RNA from hibernating and active *Myotis ricketti* brain and intra-abdominal adipose tissue, and the RNA concentration and RNA Integrity Number (RIN) met the requirements for Solexa small RNA library construction and sequencing. After small RNAs were isolated and processed for deep sequencing on the Illumina/Solexa platform, we sequenced a total of 1.56×10^7^, 1.54×10^7^, 1.53×10^7^ and 1.55×10^7^ reads from the HB, AB, HA and AA libraries, respectively. We obtained a total of 4.51×10^7^ high-quality small RNA reads after removing ambiguous reads. In these high quality reads, 6.3×10^5^ (HB), 4.2×10^5^ (AB), 1.2×10^6^ (HA), 5.4×10^5^ (AA) and 2.3×10^6^ (in total) were clean unique reads, and the percentage of small RNA reads (\>18 bp) in four libraries were above 60% (see [S2](#pone.0135064.s005){ref-type="supplementary-material"} and [S3](#pone.0135064.s006){ref-type="supplementary-material"} Tables). The redundancies of the four libraries were 94.82% (HB), 95.85% (AB), 89.92% (HA) and 95.31% (AA), and among these, the redundancy of co-expressed unique reads among libraries was 11.3% and that of state/tissue-specific reads was 88.7% ([S3 Table](#pone.0135064.s006){ref-type="supplementary-material"}). Less than 5% of total small RNA reads from the bat libraries perfectly matched the *Myotis lucifugus* genome \>20 times ([S4 Table](#pone.0135064.s007){ref-type="supplementary-material"}); thus, we removed these unique reads to avoid self-contradiction of miRNA prediction. Therefore, a total of 3.16×10^7^ remained after mapping the sequences to the *M*. *lucifugus* genome ([Table 1](#pone.0135064.t001){ref-type="table"} and [S2 Fig](#pone.0135064.s002){ref-type="supplementary-material"}). The size distribution of perfectly matched small RNA reads is shown in [Fig 1A](#pone.0135064.g001){ref-type="fig"}. We identified a total of 2.47×10^7^ potential miRNA reads, including 159,807 distinct reads, by searching against miRBase and filtering the ncRNAs and low-expressing reads ([S5 Table](#pone.0135064.s008){ref-type="supplementary-material"}). Subsequently, 1.51×10^7^ (47.7% in total) miRNA reads met the fold-back structure (hairpin) and MirCheck criteria, and miRNA reads were the most abundant fraction (range from 32.7% in HA to 55.6% in AB, [Table 1](#pone.0135064.t001){ref-type="table"} and [Fig 1B](#pone.0135064.g001){ref-type="fig"}). Other small RNAs, such as ncRNAs, genomic repeats, transcript repeats, and unknown genomic regions, comprised 20.4%, 2.3%, 2.8% and 26.8%, respectively ([Table 1](#pone.0135064.t001){ref-type="table"} and [Fig 1B](#pone.0135064.g001){ref-type="fig"}). Upon further inspection of genomic repeat-derived siRNAs, we determined that Short INterspersed Element (SINE) and Long INterspersed Element (LINE) were the major contributors to the four libraries ([Table 1](#pone.0135064.t001){ref-type="table"} and [Fig 2](#pone.0135064.g002){ref-type="fig"}). {#pone.0135064.g001} {#pone.0135064.g002} 10.1371/journal.pone.0135064.t001 ###### Classification of unique small RNA reads. {#pone.0135064.t001g} Classification Total HB AB HA AA ------------------------------------------------------ --------- ------ ---------- ------ -------- ------ ---------- ------ -------- ------ ---------- ------ --------- ------ ---------- ------ -------- ------ ---------- ------ Total Unique small RNA Reads 2323170 100 45161406 100 629751 100 12160105 100 415452 100 10012888 100 1164049 100 11549959 100 536995 100 11438454 100 Total Perfect Matched small RNA reads 720862 31.0 31590053 69.9 201875 32.1 8682951 71.4 120695 29.1 7266543 72.6 373224 32.1 7502137 65.0 152674 28.4 8138422 71.1 miRNAs[^a^](#t001fn002){ref-type="table-fn"} 7530 1.0 15068478 47.7 3365 1.7 4451386 51.3 2684 2.2 4041090 55.6 3967 1.1 2454320 32.7 3164 2.1 4121682 50.6 Non-Coding RNA[^b^](#t001fn003){ref-type="table-fn"} 71131 9.9 6443708 20.4 39956 19.8 1556064 17.9 27417 22.7 709223 9.8 47132 12.6 2798792 37.3 29197 19.1 1379629 17.0 rRNA 39273 5.4 5495818 17.4 25585 12.7 1311512 15.1 18091 15.0 539528 7.4 31298 8.4 2452251 32.7 20826 13.6 1192527 14.7 tRNA 18189 2.5 244828 0.8 7063 3.5 54006 0.6 3977 3.3 36283 0.5 7954 2.1 112091 1.5 3540 2.3 42448 0.5 snoRNA 2360 0.3 36480 0.1 1030 0.5 9531 0.1 660 0.5 5882 0.1 1280 0.3 13861 0.2 729 0.5 7206 0.1 snRNA 454 0.1 7249 0.0 214 0.1 1107 0.0 130 0.1 614 0.0 309 0.1 4669 0.1 107 0.1 859 0.0 other ncRNA 10855 1.5 659333 2.1 6064 3.0 179908 2.1 4559 3.8 126916 1.7 6291 1.7 215920 2.9 3995 2.6 136589 1.7 Transcript-derived reads 289993 40.2 873840 2.8 52157 25.8 173709 2.0 31598 26.2 105720 1.5 173270 46.4 445779 5.9 52878 34.6 148632 1.8 Genomic repeat-derived reads 67509 9.4 732080 2.3 24669 12.2 66751 0.8 11222 9.3 47251 0.7 25197 6.8 91641 1.2 10348 6.8 526437 6.5 LINE 33783 50.0 124881 17.1 12312 49.9 36423 54.6 5312 47.3 26188 55.4 13129 52.1 32850 35.8 4622 44.7 29420 5.6 SINE 5681 8.4 512264 70.0 1830 7.4 8885 13.3 941 8.4 10160 21.5 2178 8.6 26692 29.1 1016 9.8 466527 88.6 LTR 10571 15.7 35643 4.9 3637 14.7 10730 16.1 1797 16.0 5286 11.2 4007 15.9 13374 14.6 1751 16.9 6253 1.2 DNA 8644 12.8 19283 2.6 2881 11.7 5062 7.6 1674 14.9 2877 6.1 3398 13.5 8261 9.0 1444 14.0 3083 0.6 SSR 2293 3.4 3002 0.4 1012 4.1 1216 1.8 372 3.3 427 0.9 658 2.6 868 0.9 316 3.1 491 0.1 Low 2524 3.7 3194 0.4 1895 7.7 2287 3.4 178 1.6 196 0.4 338 1.3 483 0.5 173 1.7 228 0.0 Unknown Repeats 4013 5.9 33813 4.6 1102 4.5 2148 3.2 948 8.4 2117 4.5 1489 5.9 9113 9.9 1026 9.9 20435 3.9 Unknown 284699 39.5 8471947 26.8 81728 40.5 2435041 28.0 47774 39.6 2363259 32.5 123658 33.1 1711605 22.8 57087 37.4 1962042 24.1 HB: hibernating state brain; AB: active state brain; HA: hibernating state adipose tissue; AA: active state adipose tissue. rRNA, ribosomal RNA; tRNA, transfer RNA; snoRNA, small nucleolar RNA; snRNA, small nuclear RNA; LINE, Long INterspersed Elements; SINE, Short INterspersed Elements; LTR, Transposable elements with Long Terminal Repeats; DNA, DNA transposons; SSR, Simple Sequence Repeats; Low, Low Complexity Sequences; unknown, derived from unannotated/intergenic regions. ^a^ miRNA reads identified by miCheck, searched against the Sanger miRBase (Release 19) and manually inspected for canonical structure from 24656421 total potential reads ([S2 Fig](#pone.0135064.s002){ref-type="supplementary-material"} and [S5 Table](#pone.0135064.s008){ref-type="supplementary-material"}). ^b^ Screening against non-coding RNA databases (Release 10) to filter ncRNAs (tRNA, rRNA, snoRNA and other ncRNA) by Bowtie. MiRNA identification {#sec013} -------------------- To better understand the length distribution of metazoan miRNA precursors, we analyzed known miRNA precursors in five out-grouped organisms (two vertebrates and three invertebrates, [S3 Fig](#pone.0135064.s003){ref-type="supplementary-material"}) in miRBase. The majority of pre-mature miRNAs from these five organisms were less than 150 nt with a mean length of approximately 90 nt ([S3 Fig](#pone.0135064.s003){ref-type="supplementary-material"}). We mapped the potential miRNA reads onto the reference genome and retrieved 509,206 block sequences with 150 nt upstream and downstream extensions; we then used these sequences for secondary structure prediction. From these, we used Einverted of Emboss \[[@pone.0135064.ref042]\] and identified 380,605 potential reads with inverted repeats. We identified 90 conserved hairpins in 28 families and 101 novel hairpins ([S6 Table](#pone.0135064.s009){ref-type="supplementary-material"}). Finally, we obtained a total of 196 mature miRNAs, including 119 conserved miRNAs and 77 novel bat-specific miRNAs, and sequence raw counts varied from one to 1,946,308 ([S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). Further analysis showed that 12.2% (23/188, brain) and 4.6% (8/173, WAT) of the miRNA was expressed in a tissue-specific manner, whereas 10.1% (19/188, brain), and 10.9% (19/173, WAT) exhibited state-specific expression ([Fig 3](#pone.0135064.g003){ref-type="fig"} and [S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). In the 31 state/tissue-specific miRNAs, six (miR-1298-5p, miR-124-5p, miR-153-5p, miR-153-3p, miR-551-3p and miR-1298-3p) had counts above 100, and they were all specifically expressed in the brain ([S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). {ref-type="supplementary-material"}. HB: hibernating state brain; AB: active state brain; HA: hibernating state adipose tissue; AA: active state adipose tissue.](pone.0135064.g003){#pone.0135064.g003} Differential miRNA expression analysis and validation {#sec014} ----------------------------------------------------- A total 49 out of 196 miRNAs were differentially expressed (*P*≤0.01 & │logFC│≥1) between the hibernation vs. the active state, including 33 miRNAs in brain, 25 miRNAs in WAT and 9 miRNAs in both tissues ([Table 2](#pone.0135064.t002){ref-type="table"} and [S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). Of 33 miRNAs previously shown to be differentially expressed during torpor ([S8 Table](#pone.0135064.s011){ref-type="supplementary-material"}), ten (mir-1, mir-24, miR-29, miR-124a, mir-142, mir-181a, mir-181b mir-206, mir-378 and mir-486) were detected in our study ([S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). Moreover, we found that mir-378 was differentially expressed, and mir-142, mir-181b and mir-486 were also somewhat differentially expressed during hibernation ([S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). 10.1371/journal.pone.0135064.t002 ###### 49 differentially expressed miRNAs identified by Solexa sequencing in the brain or adipose tissues during hibernation. {#pone.0135064.t002g} HA vs. AA HB vs. AB --------------- -------- -------------- ----------- ----------- ------ ------------ ----------- ----------- ------ miR-124b-3p 101576 2.259908382 5.6190503 0.0018537 UP -0.536674 14.312553 0.4127694 DOWN miR-139-3p 3183 0.37423932 8.5019562 0.5727299 UP -2.1306716 8.4057206 0.0019162 DOWN miR-200a-3p 334 7.578561186 3.2895048 2.45E-06 UP -0.886215 5.923925 0.1986791 DOWN miR-200b-3p 420 4.364567511 6.7026081 3.08E-08 UP -1.8932045 4.6037305 0.0113473 DOWN miR-222-3p 12491 1.429407306 10.720804 0.0321579 UP -1.8021156 10.33994 0.0075425 DOWN miR-222-5p 290 3.030812254 5.7697748 5.80E-05 UP 2.1583491 4.6655436 0.0044099 UP miR-3135-5p 6682 -2.637365825 9.938779 0.0001553 DOWN -0.1976903 8.1896251 0.7652108 DOWN miR-363-3p 590 2.133071814 5.9211148 0.0028745 UP -0.0815074 6.3009552 0.9118204 DOWN miR-378-3p 198036 -2.818466705 15.121184 5.51E-05 DOWN -1.5232279 10.232562 0.0228152 DOWN miR-448-3p 8783 -2.143218357 7.3054357 0.0022535 DOWN 0.0288125 10.656037 0.9653039 UP miR-455-3p 1010 -1.921311717 6.8853723 0.0061558 DOWN 0.1968478 6.3795769 0.7761823 UP miR-486-3p 809 -0.156052775 6.9078207 0.825439 DOWN -1.9680416 5.8518332 0.0056228 DOWN miR-574a-5p 400 -2.521905714 6.1292436 0.0006325 DOWN -2.7422469 2.7923067 0.0039648 DOWN miR-574b-5p 153 -4.250113976 4.6476538 1.32E-05 DOWN -3.8191272 1.987474 0.0030428 DOWN miR-590-3p 136 0.378106825 4.4405947 0.6392554 UP 2.6462719 3.3179244 0.0035346 UP miR-7c-5p 32436 2.712605457 8.7286431 0.0001116 UP 1.1902664 12.649161 0.0723922 UP miR-7f-5p 227183 0.390233519 13.83588 0.5510089 UP -1.8441761 15.019439 0.0062266 DOWN miR-92b-3p 9041 2.884880248 5.7195942 0.0001081 UP -0.1793825 10.807021 0.7843292 DOWN Novel-10-3p 111 3.016403333 12.848138 0.0001838 UP -4.4272759 12.275541 4.45E-06 DOWN Novel-14-3p 97 0.936226836 12.677215 0.2187093 UP -4.3606505 11.683673 3.70E-05 DOWN Novel-15-3p 1669 -2.944335536 11.581472 0.0018222 DOWN 7.2036728 15.19745 3.14E-14 UP Novel-16-3p 87 -2.344769395 12.036684 0.0062019 DOWN -1.2012847 9.4737976 0.3901332 DOWN Novel-17-3p 81 -2.304954477 12.263073 0.0052596 DOWN -0.5052158 9.9593452 0.7250474 DOWN Novel-18a-3p 1471 -0.426177519 15.116184 0.5299252 DOWN 2.3688352 16.00558 0.0006634 UP Novel-18b-3p 344 -1.02311079 13.052043 0.1656782 DOWN 2.5486332 13.71619 0.0004685 UP Novel-1b-3p 121 -2.293820099 12.911157 0.0026867 DOWN -1.1555324 10.404784 0.2531532 DOWN Novel-21-3p 1424 2.451014248 15.801168 0.0004914 UP -0.4894437 15.830676 0.4628932 DOWN Novel-22-3p 71 0 0 1 \_ 7.397751 11.114078 2.45E-06 UP Novel-25a-5p 1088 1.581470973 15.986822 0.0200239 UP -3.2647271 15.214248 8.67E-06 DOWN Novel-27-3p 52 0 0 1 \_ 3.8813759 10.646132 0.0002438 UP Novel-28-5p 50 2.607474245 11.636459 0.0034908 UP -3.7091105 11.077865 0.0008149 DOWN Novel-29-5p 50 1.097898289 11.646378 0.1811178 UP -4.2594804 10.735381 0.0002438 DOWN Novel-2a-3p 157 0.806179078 13.221511 0.2539667 UP -1.9817143 12.411679 0.0100787 DOWN Novel-2a-1-5p 38 3.353470354 10.81473 0.002812 UP -0.7705951 10.858367 0.4453967 DOWN Novel-31-3p 910 -2.058174551 12.059167 0.0145875 DOWN 5.722207 14.947979 2.83E-11 UP Novel-32-3p 649 -2.308464878 13.31712 0.0020344 DOWN 3.1963163 13.574877 3.13E-05 UP Novel-34-3p 329 -0.707605215 12.444825 0.3636482 DOWN 3.301607 13.981698 1.33E-05 UP Novel-36-3p 309 1.484842174 11.704508 0.070754 UP 2.9220594 13.854956 8.30E-05 UP Novel-37-3p 3841 -2.056735288 13.166302 0.0056955 DOWN 6.1069365 16.429287 4.25E-13 UP Novel-38-5p 238 -0.844950584 13.898719 0.2282207 DOWN -2.8099732 12.134971 0.0009641 DOWN Novel-3b-3p 153 -0.732186096 11.093626 0.4453967 DOWN 2.9094233 12.254351 0.0004231 UP Novel-3c-3p 127 -0.989181873 10.999282 0.2980811 DOWN 2.707911 11.810942 0.0012308 UP Novel-3f-3p 66 -1.301092394 10.228842 0.2836528 DOWN 2.9447601 11.027109 0.0019476 UP Novel-4a-3p 148 -1.243075915 13.337668 0.0862047 DOWN -3.8778764 11.233557 0.0003435 DOWN Novel-4b-3p 1962 -0.712484479 17.03775 0.2811898 DOWN -4.5924403 15.198042 8.55E-09 DOWN Novel-4d-3p 93 -1.074767845 12.588034 0.1491761 DOWN -3.2491131 10.65837 0.0047291 DOWN Novel-5-3p 146 0 0 1 \_ 4.8140123 11.573737 2.47E-06 UP Novel-6-3p 139 2.426333279 12.19918 0.0032194 UP 0.5738677 12.756475 0.4367159 UP Novel-9-5p 126 2.212281384 13.005661 0.0042355 UP -5.0103113 12.29918 1.27E-06 DOWN To validate miRNA expression level, we tested 10 differentially expressed miRNAs by miRNA-specific stem-loop qPCR. According to the results of Solexa sequencing, five of these miRNAs (mir-222-5p, mir-574a-5p, Novel-9-5p, Novel-10-3p and Novel-37-3p) were differentially expressed during hibernation in both the brain and WAT; another five miRNAs, including three (mir-7f-5p, mir-139-3p and mir-222-3p) differentially expressed only in the brain and two (mir-124b-3p and mir-378-3p) only in WAT, were used as negative controls. PCR confirmed the Solexa sequencing expression patterns ([Table 3](#pone.0135064.t003){ref-type="table"}). In WAT (HA/AA), six miRNAs (mir-124b-3p, mir-222-5p, mir-378-3p, mir-574a-5p, Novel-9-5p and Novel-10-3p) showed consistent expression patterns and all reached significance by both technologies ([Table 3](#pone.0135064.t003){ref-type="table"}). In the brain (HB/AB), the differentially expressed patterns of four miRNAs (mir-222-5p, Novel-9-5p, Novel-10-3p and Novel-37-3p) were identical in both technologies. Moreover, two negative controls, mir-7f-5p in WAT and mir-378-3p in brain showed no significant change detected in sequencing data, were confirmed by qPCR. 10.1371/journal.pone.0135064.t003 ###### Comparison of ten miRNA expression pattern in four libraries, as detected by Solexa sequencing and qPCR. {#pone.0135064.t003g} HA vs. AA HB vs. AB ------------- ----------- ----------- ---------------------------------------------------------------------------- ------- ---------- ---------------------------------------------------------------------------- -------- ---------- ---------------------------------------------------------------------------- -------- ---------- ---------------------------------------------------------------------------- miR-124b-3p 2.26 1.85E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 1.87 1.74E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.542 0.41 [↓](#t003fn003){ref-type="table-fn"} -0.98 0.12 [↓](#t003fn003){ref-type="table-fn"} miR-139-3p 0.37 0.57 [↑](#t003fn002){ref-type="table-fn"} -2.89 3.91E-04 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -2.13 1.92E-03 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.67 0.11 [↓](#t003fn003){ref-type="table-fn"} miR-222-3p 1.43 0.03 [↑](#t003fn002){ref-type="table-fn"} 1.07 1.36E-05 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -1.8 7.54E-03 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.34 0.15 [↓](#t003fn003){ref-type="table-fn"} miR-222-5p 3.03 5.79E-05 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 0.31 6.81E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 2.16 4.41E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 0.59 7.67E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} miR-378-3p -2.82 5.51E-05 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.82 1.57E-06 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -1.52 0.02 [↓](#t003fn003){ref-type="table-fn"} -0.1 0.04 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} miR-574a-5p -2.52 6.32E-04 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -6.23 2.08E-07 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -2.74 3.96E-03 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.24 0.83 [↓](#t003fn003){ref-type="table-fn"} miR-7f-5p 0.39 0.55 [↑](#t003fn002){ref-type="table-fn"} -0.05 0.12 [↓](#t003fn003){ref-type="table-fn"} -1.84 6.23E-03 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.796 0.24 [↓](#t003fn003){ref-type="table-fn"} Novel-10-3p 3.02 1.84E-04 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 1.05 0.01 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -4.43 4.45E-06 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.67 8.52E-07 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} Novel-37-3p -2.06 5.69E-03 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -0.04 0.25 [↓](#t003fn003){ref-type="table-fn"} 6.11 4.20E-13 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 0.36 7.27E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} Novel-9-5p 2.21 4.24E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} 35.07 1.69E-03 [↑](#t003fn002){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -5.01 1.27E-06 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} -1.86 3.82E-24 [↓](#t003fn003){ref-type="table-fn"} [\*](#t003fn004){ref-type="table-fn"} PCR was performed in three biological replicates of each state with triplicate wells for each individual sample. ↑, up-regulated ↓, down-regulated \*, *P*-value^a^\<0.01 or *P*-value^b^\<0.05. Target prediction and annotation {#sec015} -------------------------------- To better understand the potential role and mechanism of miRNAs in hibernation, we predicted the miRNA target genes. Firstly, we obtained 22,432 *M*. *lucifugus* reference genes and 104,763 *H*. *sapiens* mRNAs and searched for orthologous genes by tblastx. A total of 20,282 *M*. *lucifugus* genes were orthologous to *H*. *sapiens* mRNAs with a BLAST e value ≤1e-10 and identity ≥60%. As 93.6% of human mRNAs have a 3' UTR less than 3 Kb ([S1 Fig](#pone.0135064.s001){ref-type="supplementary-material"}), we extracted 3 Kb downstream of *M*. *lucifugus* coding sequences, including 2151 *M*. *lucifugus* genes with no human orthologs. Finally, we identified 87.8% (19,698 / 22,432) reference genes as the targets of 196 miRNAs. Both KEGG pathway and GO annotation analyses indicated that most physiological processes and cell functions are regulated by 196 miRNAs. We did not observe any significant differences in the number of target genes when we compared 119 conserved miRNAs vs. 196 miRNAs and 77 novel miRNAs vs. 196 miRNAs ([S9 Table](#pone.0135064.s012){ref-type="supplementary-material"}). In contrast, further analysis showed that many pathways in both tissues were markedly affected by the differentially expressed miRNAs, especially by the miRNAs down-regulated during hibernation ([S9 Table](#pone.0135064.s012){ref-type="supplementary-material"}). We further examined the adipocytokine signaling pathway to determine the relationship between miRNAs and target mRNAs during hibernation, as adipocytokine signaling is co-regulated by both brain and adipose tissues. Moreover, we searched for mRNAs from our on-going project of digital gene expression (DGE) sequencing to enrich for target genes. DGE sequencing has been conducted on the same animals used here (unpublished data). A total of 64 mRNAs involved in adipocytokine signaling pathway were identified by DGE sequencing ([S10 Table](#pone.0135064.s013){ref-type="supplementary-material"}, DGE unpublished data). By combining the miRNA and target mRNA expression patterns, we found that 41 of 49 differentially expressed miRNAs regulated the expression of 51 genes, and more than half of the genes in each tissue were up-regulated during hibernation. Among these, the differential expression of three mRNAs (Insulin Receptor Substrate, IRS; Adenosine 5'-monophosphate (AMP)-activated protein kinase, AMPK; Retinoid X receptor, RXR) was significant (*P*≤0.01), including IRS over-expression. The others were down-regulated during HA/AA ([Fig 4](#pone.0135064.g004){ref-type="fig"} and [S10 Table](#pone.0135064.s013){ref-type="supplementary-material"}). {#pone.0135064.g004} Discussion {#sec016} ========== Our study shows that in the hibernating bat *Myotis ricketti*, 33 brain and 25 white adipose tissue (WAT) miRNAs were differentially expressed between euthermia and hibernation. Consequently, they are likely involved in regulating functional processes and energy metabolism during hibernation. Our findings were possible with the ultrahigh-throughput sequencing technique (RNA-seq) Illumina/Solexa sequencing, which provides a great platform to discover and analyze small RNAs. It has been previously used in non-model species, such as the Arctic ground squirrel \[*Urocitellus* (*Spermophilus) parryii*\] and the sea cucumber (*Apostichopus japonicus*) during hypometabolism, but these species' genomic backgrounds and miRNA data are not available in miRbase \[[@pone.0135064.ref018], [@pone.0135064.ref021]\]. The clear advantages of Solexa sequencing are the generation of several million small RNA sequences in each small RNA library of one run, the ability to identify novel miRNAs, the potential to overcome the drawbacks of microarrays, and the generation of expression profiles in a greater and reproducible dynamic range \[[@pone.0135064.ref053]\]. However, this technology can be influenced by sequencing errors and raw data processing before miRNA identification \[[@pone.0135064.ref053]\]. In the present study, we performed Solexa high-throughput sequencing with stem-loop qPCR validation of expression levels to analyze miRNAs involved in hibernation. Moreover, we chose a widely distributed bat species in China, congeneric with *Myotis lucifugus*, as the animal model to explore miRNA expression in two vital organs (brain and WAT) during hibernation. Our data show that the sequencing information from four libraries was saturated, and 88.7% unique reads were tissue/state-specifically expressed ([S3 Table](#pone.0135064.s006){ref-type="supplementary-material"}). These data indicate a high quality of small RNA library construction and screening and that the libraries meet the needs of small RNA annotation and analysis. After mapping the unique reads to the *M*. *lucifugus* genome database and removing the reads that matched genome ≥20, the small RNA annotation indicated that miRNAs (47.7% in total) are the most abundant fraction of small RNAs (see [Table 1](#pone.0135064.t001){ref-type="table"} and [Fig 1](#pone.0135064.g001){ref-type="fig"}). These results are similar to those of previous miRNA profiling of multi-animal species \[[@pone.0135064.ref018], [@pone.0135064.ref021], [@pone.0135064.ref036]\]. Furthermore, we found that SINE and LINE, which are retroposons widely used as phylogenetic markers \[[@pone.0135064.ref054], [@pone.0135064.ref055]\], were the major source of genomic repeat-derived siRNAs in all four libraries. The detailed information of SINE and LINE provided us with the ability to further explore the phylogenetic evolution of heterothermy in mammals, especially in bats. Our data also revealed a large number of conserved and novel miRNAs. We identified a total of 196 mature miRNAs, including 119 conserved miRNAs and 77 novel miRNAs ([S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}). Moreover, we found that 31 (22 conserved and nine novel) of 196 miRNAs were expressed in a tissue/state-specific manner, suggesting a complex mechanism of miRNA regulation during hibernation. Of those, six conserved miRNAs with expression counts \>100 were specifically expressed in the brain, including miR-153, which has been shown to function in neuronal protection and tumorigenesis \[[@pone.0135064.ref056], [@pone.0135064.ref057]\], miR-124, which regulates metabolic and immune processes \[[@pone.0135064.ref058], [@pone.0135064.ref059]\], miR-551, which is implicated in the stress response \[[@pone.0135064.ref060]\], and miR-1298, which has been associated with neural and endocrinological disease \[[@pone.0135064.ref061]\]. However, in the absence of direct and detailed evidence between the molecular data and animal function, it is necessary to further investigate the roles of these tissue- and state-specific miRNAs during hibernation. Of 196 miRNAs, 49 were differentially expressed (*P*≤0.01 & │logFC│≥1) ([Table 2](#pone.0135064.t002){ref-type="table"} and [S7 Table](#pone.0135064.s010){ref-type="supplementary-material"}), including ten miRNAs that had been shown to regulate mammalian hibernation in other species ([S8 Table](#pone.0135064.s011){ref-type="supplementary-material"}) \[[@pone.0135064.ref018], [@pone.0135064.ref020], [@pone.0135064.ref022]--[@pone.0135064.ref025]\]. Both microarray and qPCR have been used to validate miRNA expression in many studies, as they are more accurate than RNA-seq sequencing \[[@pone.0135064.ref018], [@pone.0135064.ref021]\]. In our study, qPCR validation of 10 out of 49 differentially expressed miRNAs provided consistent results with those of Solexa sequencing ([Table 3](#pone.0135064.t003){ref-type="table"}). We conclude that our data are reliable. Although we validated a few differentially expressed miRNAs by qPCR in our study, the limitations of RNA-seq, and the value of sample pooling in high-throughput sequencing, we suggest that microarray hybridization should be performed in the future to further validate the significance of differentially expressed miRNAs obtained here. Microarray analysis could potentially compensate for the drawbacks of RNA-seq sequencing \[[@pone.0135064.ref053]\] and estimate the biological variability. To gain insight into the potential broader functions of the miRNAs we identified, we predicted their putative targets and classified them by KEGG pathway assay and GO annotation. Comparing the targets of 196 miRNAs, we found that miRNAs were differentially expressed in both brain and adipose tissues between euthermia and hibernation states, and this is likely a reflection of the different activities of many essential processes during these states (e.g., lipid metabolism and signal transduction, [S9 Table](#pone.0135064.s012){ref-type="supplementary-material"}). We found that only 49 of 196 miRNAs regulated most target gene expression (12,909 of 20,370), and thus most physiological processes during hibernation, contributing to reduced energy expenditure during hypometabolism. Moreover, both KEGG and GO assays identified the same enriched functional groups (*P*\<0.05), and all of them were affected by differentially expressed miRNAs, particularly by down-regulated miRNAs ([S9 Table](#pone.0135064.s012){ref-type="supplementary-material"}). By binding to target mRNAs, miRNAs can reversibly degrade their targets at the transcriptional level and/or inhibit their translation \[[@pone.0135064.ref019]\]. Thus, during torpor, brain and WAT physiological processes were globally affected by down-regulated miRNAs, suggesting that these processes were activated, supporting our hypothesis. Moreover, previous studies have shown that some specific brain regions such as the hypothalamus are involved in regulating mammalian hibernation \[[@pone.0135064.ref062]--[@pone.0135064.ref064]\]. Our study was limited to the use of entire brains, but future work on specific brain regions may identify the precise regulation of miRNA expression in important brain areas. Such studies could further expand our knowledge of hibernation regulation. The brain plays an important role in controlling and synchronizing peripheral physiology to reduce energy expenditure and minimizing tissue damage and/or disease during hibernation \[[@pone.0135064.ref003], [@pone.0135064.ref004], [@pone.0135064.ref007]\]. In contrast, adipose tissue is the primary energy source during hibernation and is especially critical during periodic rewarming \[[@pone.0135064.ref006], [@pone.0135064.ref065]\]. Thus, specific molecular changes are required to accommodate the specific roles of brain and adipose tissue during hibernation. Because the potential network of miRNAs and targets is highly complicated, we chose to examine the adipocytokine signaling pathway, which plays an important role in signaling transduction and energy metabolism \[[@pone.0135064.ref066], [@pone.0135064.ref067]\]. In both tissues, more than half mRNAs were up-regulated ([Fig 4A](#pone.0135064.g004){ref-type="fig"}) in contrast to miRNA expression, indicating that this pathway was activated to some extent. According to KEGG analysis, the adipocytokine signaling pathway can be subdivided into three interrelated parts, TNFα (Tumor Necrosis Factor α)-related genes, LEP (Leptin)-related genes and ADIPO (Adiponectin)-related genes ([Fig 4B](#pone.0135064.g004){ref-type="fig"}). Andrews et al. \[[@pone.0135064.ref013]\] previously showed that energy metabolism of hibernators is shifted at the gene level from glycometabolism in the liver and/or pancreas to lipid metabolism in adipose tissue during hibernation. We found that the significant up-regulation of IRS in WAT could be indicative of glucose uptake inhibition, whereas a decrease in AMPK in WAT could promote gluconeogenesis. Our data help to illustrate how miRNAs and mRNAs control the metabolic switch by inhibiting the glycometabolic pathway to burning fat as the main energy source, supporting previous reports \[[@pone.0135064.ref065]\]. While the regulation of fat metabolism is crucial during hibernation, the regulation of food intake is important during euthermia. Negative AMPK expression or the inhibition of AMPK and AMPK-related gene activity in the hypothalamus result in reduced food intake \[[@pone.0135064.ref068], [@pone.0135064.ref069]\]. We hypothesize that down-regulation of AMPK and RXR in WAT lead to decreased food intake and increased energy expenditure. Network construction increases our understanding of molecular mechanisms mediated by miRNAs and target genes. It also provides clues to uncover how brain and adipose tissue cooperate to regulate hibernation by decreasing food intake, inhibiting glucose metabolism and increasing fatty acid metabolism to meet the physiological needs of hibernators. Conclusions {#sec017} =========== We identified 196 miRNAs (77 novel miRNAs) in bats by in-depth analysis of the small RNAs in the hibernating species *Myotis ricketti*. Differential miRNA expression suggests that some physiological pathways in the brain and adipose tissue are activated during hibernation. In contrast to other peripheral tissues that are physiologically suppressed, our findings advance the understanding of the molecular mechanisms of hibernation. However, further study is required to validate the expression pattern during hibernation vs. euthermia, estimate biological variability, and elucidate the mechanisms of specific miRNAs and their target mRNAs in hibernation. Moreover, the Interspersed Elements (LINE and SINE) identified here may help further explore the phylogenetic evolution of heterothermy in bats. With the progress of the *M*. *lucifugus* genome sequencing project and the increased accuracy of gene annotation, more detailed information on miRNAs involved in hibernation will likely be uncovered from bat transcriptome libraries. These can be used to better understand heterothermy the physiology and evolution of heterothermy. Supporting Information {#sec018} ====================== ###### Analysis of 3' UTR length of *Homo sapiens* mRNAs. *Homo sapiens* reference sequences were downloaded from the NCBI Reference Sequence database (<http://www.ncbi.nlm.nih.gov/RefSeq/>). (JPG) ###### Click here for additional data file. ###### Flow chart of *Myotis ricketti* small RNA filtering and miRNA identification. (PNG) ###### Click here for additional data file. ###### Length distribution of known miRNA precursors of five organisms in metazoan miRBase. gga: *Gallus gallus*; has: *Homo sapiens*; cel: *Caenorhabditis elegans*; dme: *Drosophila melanogaster*; sme: *Schmidtea mediterranea*. (JPG) ###### Click here for additional data file. ###### Forward, stem-loop and universal primers used to amplify miRNAs and 5S rRNA. (DOC) ###### Click here for additional data file. ###### Summary of Solexa small RNA reads statistics. The table summarizes the total number of small RNA reads by Solexa sequencing. After removing adapter contaminants, reads less than 18 nt and containing Ns, the clean reads from each library were used for the small RNA screen. HB: hibernating state brain; AB: active state brain; HA: hibernating state adipose tissue; AA: active state adipose tissue. (DOC) ###### Click here for additional data file. ###### Summary of small RNA read expression profiles in four libraries. Redundancy among libraries was calculated to estimate the quality of the Solexa small RNA libraries. Moreover, the numbers of state/tissue-specific expressed reads were identified. HB: hibernating state brain; AB: active state brain; HA: hibernating state adipose tissue; AA: active state adipose tissue. (DOC) ###### Click here for additional data file. ###### Summary of perfectly mapped total small RNA unique reads. (DOC) ###### Click here for additional data file. ###### Summary statistics of unique miRNA read identification. (DOC) ###### Click here for additional data file. ###### Details of miRNA hairpin screening. \(A\) 90 conserved miRNA hairpins. (B) 101 novel miRNA hairpins. (XLSX) ###### Click here for additional data file. ###### List of miRNA expression and distribution. \(A\) Expression details of 196 miRNAs (119 conserved miRNAs and 77 novel miRNAs). (B) List of 33 miRNAs differentially expressed in the brain (*P*≤0.01 & │logFC ≥1). (C) List of 25 miRNAs differentially expressed in adipose tissue (*P*≤0.01 & │logFC│≥1). (D) List of 9 miRNAs differentially expressed in both brain and adipose tissue (*P*≤0.01 & │logFC│≥1). (E) List of 31 tissue/state-specifically expressed miRNAs in brain (23) and adipose tissue (8). (XLSX) ###### Click here for additional data file. ###### Previously identified miRNAs involved in hibernation. (XLSX) ###### Click here for additional data file. ###### Classification of miRNA target genes. KEGG pathway analysis (A) and GO annotation (B) of target genes regulated by 196 miRNAs, 119 conserved miRNAs, 8 conserved miRNAs differentially expressed in the brain, 13 conserved miRNAs differentially expressed in adipose tissue, 77 novel miRNAs, 25 novel miRNAs differentially expressed in the brain, and 12 novel miRNAs differentially expressed in adipose tissue. T-tests were performed to analyze the bio-activity of each physiological process in both tissues by comparing hibernation vs. active state. (XLS) ###### Click here for additional data file. ###### Summary of mRNAs identified by DGE sequencing. Genes, only identified by DGE sequencing and not the targets of differentially expressed miRNAs which obtained in this study, are in green. (XLS) ###### Click here for additional data file. The authors wish to thank Junpeng Zhang and Jinshuo Zhang for their assistance collecting samples and Bronwyn McAllan for commenting on the manuscript. [^1]: **Competing Interests:**HBS is an employee of a commercial company (MininGene Biotechnology Co. Ltd), and this does not alter the authors\' adherence to all the PLOS ONE policies on sharing data and materials. There are also no issues such as consultancy, patents, products in development, or marketed products etc. [^2]: Conceived and designed the experiments: LHY SYZ. Performed the experiments: LHY BFL. Analyzed the data: LHY HBS BFL. Contributed reagents/materials/analysis tools: LHY SYZ. Wrote the paper: LHY FG BFL JPC SYZ.

Introduction {#Sec1} ============ PAHs are a class of organic compounds that consist of two ore more fused benzene rings forming ring-shaped super molecules. PAHs are important drivers of chronic toxicity to aquatic organisms^[@CR1],[@CR2]^. They are extremely restive molecules that can persist in the environment due their hydrophobicity and low water solubility^[@CR3],[@CR4]^. As a result, they are ubiquitous in air, water and soil at varying concentrations depending on proximity to point and non-point sources^[@CR1]^. While they are generally lipophilic, some PAHs are more soluble in water and are highly toxic to aquatic biota, including fish and invertebrates^[@CR5]^. Specifically, PAHs with 3--5 aromatic rings are the most important source of chronic toxicity to aquatic organisms exposed at e.g. oil spill sites^[@CR6],[@CR7]^. Because of the wide distribution of PAHs, it is important to understand natural and anthropogenic sources of these compounds and to monitor their baseline concentrations in the aquatic environment^[@CR8]^. Discovered growing on the periphery of deep-sea hydrothermal vents, *Staphylothermus marinus* is a heterotrophic hyperthermophilic archaea that requires elemental sulfur for growth. The ability of *S. marinus* to sequester elemental sulfur from its environment relies in part on the Right Hand Coiled-Coil Nanotube (RHCC-NT), a protein fragment in the surface layer of the microorganism. The RHCC-NT encapsulates elemental cyclo-octasulfur in a hydrophobic cavity with a diameter of \~8.4 Å, where it is then available for metabolism. These cavities, adapted to capture the small, hydrophobic cyclo-octasulfur are also of a suitable size to encapsulate low molecular weight PAHs^[@CR9]^. Due to the extreme conditions of its origin the RHCC-NT can function with a wide range of temperature and chemical stability, and would easily maintain its fold and binding capacity in aquatic environments surrounding hydrocarbon extraction and transportation sites. Here we present an in-depth characterization of RHCC-NT and its unique uptake properties for PAHs (Fig. [1](#Fig1){ref-type="fig"}). Specifically, we quantified the capacity of RHCC-NT to bind 7 lower MW weight PAHs using their intrinsic fluorescence in steady-state binding experiments. We also studied the ability of the nanotube to uptake PAHs by using pyrene as a model compound. Moreover, these binding experiments were coupled with calculations of the transfer free energy based on molecular dynamics simulations and structural studies based on X-ray crystallography to determine the PAH location within the RHCC-NT. Our studies provide an insight for future binding site optimization within the nanotube, with the end goal to use it as a monitoring device in the field application.Figure 1Upon binding the RHCC-NT, the PAH displaces the waters in the cavity and moves from an aqueous environment to the hydrophobic interior of the protein. Due to the fluorescent hydrophobic effect, this change in environment results in an increase in quantum yield, a shift in the emission maximum, or both. Results {#Sec2} ======= Fluorescence monitoring studies and binding constants {#Sec3} ----------------------------------------------------- Fluorescence assays were used to examine the binding of RHCC-NT to PAHs as they allow discrimination between PAHs in solution and those bound to the RHCC-NT, as well as the ability to monitor PAH uptake as a function of time (Fig. [2](#Fig2){ref-type="fig"}). This is possible due to the polarity effect exhibited by many fluorophores, where changes in the polarity of the immediate environment of the fluorophore results in changes in the intensity or peak position in the fluorescence spectrum; the capacity of RHCC-NT to bind PAHs is examined by the changes in the PAH fluorescence spectrum as a function of increasing nanotube concentrations. Titration of RHCC-NT into a naphthalene solution results in an increase of its quantum yield consistent with the movement of naphthalene into a lower polarity environment. These fluorescence data could then be fit to Equation ([1](#Equ1){ref-type=""}), yielding a binding constant for naphthalene of *K*~*d*~ = 17 ± 1.4 µM (Fig. [2](#Fig2){ref-type="fig"}). It must be noted that the RHCC-NT also fluoresces when excited at 275 nm (Figure [S1](#MOESM1){ref-type="media"}); therefore, it was necessary to correct for this contribution in the binding analysis. In order to minimize contributions from tyrosine fluorescence, the fluorescence emission at 350 nm was used to determine the binding affinity of naphthalene to RHCC-NT (Table [1](#Tab1){ref-type="table"}). The approach described above was applied to the analysis of other low molecular weight PAH's fluorescence data (Fig. [3](#Fig3){ref-type="fig"}), however, excitation wavelengths longer than 275 nm were available for the remaining PAHs and RHCC-NT interference was minimized (for excitation wavelengths refer to Methods).Figure 2Fluorescence emission spectra (λ~exc~ = 275 nm) of 4 µM Naphthalene at different RHCC-NT concentrations. A maximum was assigned at 350 nm to eliminate any contribution from tyrosine fluorescence present in RHCC-NT. Slits were set to 1 nm. The inset shows the fraction of Naphthalene bound plotted as a function of RHCC-NT concentration.Table 1Change in measured fluorescence intensity of naphthalene at 350 nm with increasing RHCC concentration.\[RHCC\], µMFluorescence intensity at 350 nm (λ~exc~ = 275 nm)07082920.127098240.247305720.487336940.967806441.99061203.810036507.7117791015.4142026030.75184591061.51873130Figure 3Binding capacity of RHCC-NT in uptake of lower MW PAHs in aqueous solution. PAH binding constants were calculated based on the Equation ([1](#Equ1){ref-type=""}) using the same principle described in Methods section. Fluorescence emission maxima were assigned as following: acenaphthene 322 nm, acenaphthylene 323 nm, fluorene 315 nm, anthracene 421 nm, pyrene 384 nm, chrysene 400 nm. Though phenanthrene, fluoranthene, and benz(a)anthracene were also tested (Figure [S2](#MOESM1){ref-type="media"}), they did not exhibit any concentration dependent fluorescence changes. The maximum dimension of these compounds and their hydrophobicity are not significantly different from the other low MW PAHs examined, however, they do exhibit a lower degree of symmetry. In general, electron distributions in the π orbitals have a significant effect on the affinity of the RHCC-NT for structurally similar PAHs. Whereas acenaphthene yields a *K*~*d*~-value of 8.9 ± 1.8 µM, acenaphthylene is bound significantly weaker (*K*~*d*~ = 32 ± 7.0 µM) (Table [2](#Tab2){ref-type="table"}). Naphthalene, of particular interest with respect to toxicity, binds with a *K*~*d*~ of 17 ± 1.4 µM. These results suggest that RHCC will be most useful as a passive sampling media, for binding symmetrical 2--4 ring symmetrical PAHs. Introducing alkyl groups to the structure of the parent PAHs will impact symmetry and lipophilicity^[@CR10]^, which may also impact binding to the RHCC matrix.Table 2*K*~*d*~ values and molar Gibbs Free energy for individual PAHsCompound*K*~*d*~, µMR^2^-value∆G, kJ/mol^a^Fluorene5.4 ± 1.60.9570−29.9Anthracene8.7 ± 1.50.9827−28.7Acenaphthene8.9 ± 1.80.9854−28.6Naphthalene17 ± 1.40.9970−27.1Chrysene27 ± 5.60.9926−25.9Pyrene30 ± 2.00.9982−25.6Acenaphthylene32 ± 7.00.9823−25.5^a^Molar Gibbs free energy for the ligand binding is defined as ∆G = RT ln (*K*~*d*~). Time-scale study {#Sec4} ---------------- To probe the uptake kinetics of low MW PAHs, we examined the fluorescence characteristics of pyrene in solvent in comparison to RHCC-NT bound pyrene (Fig. [4](#Fig4){ref-type="fig"}). As the bulkiest of the PAHs tested, it is expected that pyrene provides an upper bound on the equilibration time of these low MW PAHs. We took advantage of the major vibronic bands (I and III) of pyrene with defined peaks at \~373 and 384 nm^[@CR11],[@CR12]^. In comparison to the emission intensity of Band I at 373 nm, the intensity of Band III at 384 nm is significantly enhanced in hydrophobic environments. In contrast, the intensity of Band I is significantly higher than that of Band III in polar environments. The ratio of the fluorescence emission intensities of Band I/III is defined as *Py*-value^[@CR13],[@CR14]^ and can be employed to detect polarity in the vicinity of the probed microenvironment. Our findings clearly demonstrate a *Py* shift from 1.7288 (RHCC-NT:Pyrene time = 0: (I~I~/I~III~ = 1067230/617334) to *Py* = 0.6395 (RHCC-NT:Pyrene time = 75 hrs: (I~I~/I~III~ = 834882/1305448). Figure [4](#Fig4){ref-type="fig"} demonstrates the change in pyrene fluorescence emission spectrum with the change of the polarity. A significant increase in the intensity at 384 nm (Band III) is observed when pyrene is incorporated into the hydrophobic microenvironment of the RHCC-NT cavities. By monitoring the change in pyrene emission at 384 nm and plotting this change as a function of time we determined that complete uptake of pyrene by RHCC-NT at 10 °C takes around 80hrs, with an average uptake rate of 1.59 nmol/hr∙mol RHCC-NT. Under these conditions, we found that naphthalene, acenaphthene and acenaphthylene all reach equilibrium in *ca*. 1 min (data not shown).Figure 4Time-scale uptake of pyrene by RHCC-NT. (**a**) Fluorescence emission spectrum (λ~exc~ = 334 nm) of 2 µM pyrene measured in the presence of 70 µM RHCC at the beginning of experiment (dashed line -Time 0) and after 75 hours (solid line) when reaction reached complete equilibrium. A maximum was assigned at 384 nm. Slits were set to 1 nm. (**b**) Fluorescence emission maxima at 384 nm measured during the course of the experiment plotted as a function of time to show time-dependent uptake of pyrene by RHCC-NT. Structural elucidation of RHCC-NT in complex with naphthalene and pyrene {#Sec5} ------------------------------------------------------------------------ To verify our previous findings and to explore the molecular mechanisms of PAH uptake we determined the X-ray crystal structures of RHCC-NT in complex with both naphthalene and pyrene (Fig. [5](#Fig5){ref-type="fig"} and Table [3](#Tab3){ref-type="table"}). Our findings reveal that the RHCC-NT binds both 2- and 4- ring PAHs in the same hydrophobic pockets, justifying the use of a common model to measure the binding coefficients through the subset of PAHs examined in this study. Comparing unliganded *vs* liganded structures, there is no major conformational rearrangement at the protein backbone nor of individual side chains detectable. The large-sized cavities act as fixed size container, which are able to uptake large hydrophobic ring-like structures. Remarkably, in the unliganded RHCC-NT structures, the cavities are filled with water clusters (pdb:1FE6)^[@CR15]^. As these cavities are hydrophobic in nature, it is expected that the presence of polar waters in the cavities is unfavorable relative to a less hydrophilic moiety. Indeed, both naphthalene and pyrene completely displace the water molecules within the binding pocket (Figure [S4](#MOESM1){ref-type="media"}). The plane of the PAHs lies perpendicular to the long axis of the RHCC-NT, even though the cavity is slightly elongated along this helical axis. This could be due to the interaction of the π- electron cloud with the dipole of the nanotube, though further experimentation will be necessary to confirm this.Figure 5X-ray Crystal structures of Naphthalene (NAPH; **A**,**B**) and Pyrene (PYR; **C**,**D**). Cross sections of cavities 2 (**A,C**) and 3 (**B,D**) are drawn with individual chains of the tetrameric RHCC-NT shown in different colors. To mark cavities 2 and 3, tyrosine moieties of Tyr^[@CR23]^ and the guanidine moieties of Arg^[@CR33]^ are highlighted in stick mode.Table 3X-ray structure determination.Data CollectionRHCC-Naphthalene (5VKF)RHCC-Pyrene (5VH0)λ (Å)1.541921.54192Space GroupP31 2 1P1 2 1**Cell dimensions***a, b, c* (Å)110.36 110.36 70.8335.02 77.00 35.10*α, β, γ* (°)90.00 90.00 120.0090.00 90.00 90.00No. reflections66293 (9151)42470 (2720)Resolution (Å)19.12--2.75 (2.90--2.75)19.25--2.06 (2.11--2.06)*R* ~*merge*~0.167 (0.689)0.157 (0.789)*I/*σI7.2 (2.4)8.3 (2.4)Completeness (%)98.9 (95.6)99.3 (93.4)Multiplicity5.1 (5.0)3.7 (3.3)**Refinement***R* ~*work*~ */R* ~*free*~ ^b^0.199/0.2390.212/0.259**No. atoms**Protein16891510Ligand/Ion4568Water126109*B-*factor (Å^2^)Protein42.3235.57Water35.5135.42Ligands74.5340.31**R.m.s. deviations**Bond lengths (Å)0.0020.003Bond angles (°)0.330.41^a^Statistics of the highest resolution shell are shown in parenthesis. ^b^The R~free~ was calculated by selecting 10% of observed reflections from refinement. Molecular Dynamics Simulations {#Sec6} ------------------------------ The computed values of the transfer free energy for the three PAH ligands are listed in Table [4](#Tab4){ref-type="table"} for both cavities 2 and 3. The calculated solvation free energies $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{G}_{solv}=-{\rm{\Delta }}{G}_{3}$$\end{document}$ in Table [4](#Tab4){ref-type="table"} are comparable in magnitude to the measured solvation free energies^[@CR16]^, Δ*G*~*solv*~(*naphthalene*) = −10.0 ± 2.5,Δ*G*~*solv*~(*phenanthrene*) = −16.2 ± 2.5, Δ*G*~*solv*~(*pyrene*) = −18.91 ± 2.5 and show the same systematic trend through the series naphthalene → phenanthrene → pyrene. The calculated transfer free energies $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{G}_{transfer}^{0}$$\end{document}$ for naphthalene and pyrene in Table [4](#Tab4){ref-type="table"} are significantly larger in magnitude than the measured values in Table [2](#Tab2){ref-type="table"}. Such systematic discrepancies between calculated and measured binding free energies have been reported elsewhere in the literature when MCTI is used to compute *absolute* binding energies in strongly hydrophobic systems. In particular, the absolute binding free energy of isobutyl-methoxy-pyrazine (IMBP) bound to porcine odorant binding proteins (OBP) is computed to be \~−67 *kJ*/*mol*^[@CR17]^ compared to the experimental value of −38.5 *kJ*/*mol* (similar to the discrepancy in the current investigation) and the absolute binding free energy of FKBP, a protein from the immunophilin group, bound to several ligands has been shown to be systematically more negative than the experimental values by \~13.4 *kJ*/*mol*^[@CR18]^. By contrast, the application of MCTI to more hydrophilic systems yields much better agreement. This behaviour appears to be independent of the choice of force field, sampling time or implicit versus explicit solvent models and has been attributed to the higher mobility of ligands in hydrophobic cavities although, thus far, no successful protocols have emerged for improving the discrepancy. Nevertheless, the calculated binding free energies do correctly reproduce the experimental systematics with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|{\rm{\Delta }}{G}_{transfer}^{0}(naphthalene)| > |{\rm{\Delta }}{G}_{transfer}^{0}(pyrene)|$$\end{document}$.Table 4Contributions to the free energy of transfer (kJ mol^−1^) from MD simulations.PAHcavity∆G~1~∆G~2~∆G~3~$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{{\bf{G}}}_{{\bf{transfer}}}^{0}$$\end{document}$naphthalene279.2 ± 1.2−1.54.9 ± 1.0−72.8 ± 1.1374.7 ± 1.4−1.54.9 ± 1.0−68.3 ± 1.2phenanthrene280.2 ± 3.8−1.510.1 ± 1.1−68.6 ± 2.8358.7 ± 3.8−1.510.1 ± 1.1−47.1 ± 2.8pyrene276.3 ± 8.8−1.512.5 ± 1.1−62.3 ± 6.3362.7 ± 2.9−1.512.5 ± 1.1−48.7 ± 2.2 The MD simulations predict that phenanthrene is also stable in both cavities, in spite of its lower degree of symmetry, with a transfer free energy intermediate between that for naphthalene and pyrene in cavity 2 and equal to that for pyrene in cavity 3. The discrepancy between MD simulations and experiment in the case of phenanthrene is most likely the result of a limitation on the fluorescence technique. Other techniques (GC-MS/MS) clearly show an uptake of all 2, 3 and 4 ring PAHs by RHCC-tetrabrachion, including phenanthrene, fluoranthrene and benz(a)anthracene. A visual examination of the fully interacting, *FBHW* − restrained, *λ* = 0 trajectories frame-by-frame showed that, on average, the plane of all three PAH ligands maintained an orientation transverse to the central cavity of the RHCC-NT throughout the entire trajectory, as observed experimentally, with angular fluctuations Δ*θ* \~± 10  relative to the initial orientation. However, all three PAH ligands exhibited rapid transitions between discrete orientations within the transverse plane, with an angular separation of Δ*ϕ*\~30 . Discussion {#Sec7} ========== Using fluorescence probe techniques, X-ray crystallography, and Molecular Dynamics Simulations we have demonstrated that the RHCC-NT can bind low MW PAHs with low equilibration times and high selectivity. To our knowledge, this is the first study showing that a proteinaceous Nano container can encapsulate PAHs. The cavities of RHCC-NT are ellipsoidal in shape with a volume of \~380 Å^3^ and allow for the uptake of PAHs up to 4-rings in size, while binding was not detected for any 5 or higher ring PAH systems. This suggests that such molecules may be too large to fit inside the cavities. Coiled coils are mostly known as strict oligomerisation motifs which function as entropic clamps to oligomerise target systems and to increase the local copy number of protein domains^[@CR19],[@CR20]^. However, most recent studies have shown that RHCC-NT and COMPcc can uptake a number of different cargos, including vitamin D3^[@CR21]^, fatty acids^[@CR22]^, S8 crowns^[@CR9]^ and metallic Hg cluster^[@CR23]^. Whereas COMPcc shows a very dynamic behavior and extensive "breathing effect" during uptake of the ligand via the open N-terminal chamber, the RHCC-NT is a highly rigid homo-tetramer, which seems to allow uptake of hydrophobic cargos just via the inter-helical space. The closed storage cavities of RHCC-NT are very rigid and do not show any sign of significant structural re-arrangements upon binding. A common feature for both systems is that strictly hydrophobic interior channels are releasing water clusters upon hydrophobic ligand binding. The genotoxoicity of PAHs is linked to metabolic activation via cytochrome P450 family enzymes into highly reactive electrophiles interacting with DNA sites^[@CR24]^. The crystal structure of human P450 1A2 reveals a compact, highly adaptive active site pocket for positioning and oxidation of the relatively large planar disk-shaped structures^[@CR25]^. In general, these metabolites are considerably more toxic than the PAH precursor molecules. There is a wealth of studies published describing the bioremediation of PAHs by pollutant-degrading microorganisms^[@CR26]^. However, many environmental factors (incl. pH, Temperature, salinity, nutrient availability and bioavailability of the contaminant) play a crucial role in these processes and are difficult to control. In conclusion, we have successfully studied the unique encapsulation of low MW PAHs in Nano containers of the archaea RHCC-NT. As shown in a most recent study, the large-sized cavities of RHCC-NT are not only able to store elemental sulfur S8 crowns^[@CR9]^, however the Nano container can also reduce and store metallic metal clusters^[@CR23]^. Moreover, RHCC-NT can be seen as a new protein-based matrix for removing PAHs and heavy metals from water. Using fluorescence probe techniques and X-ray crystallography we have demonstrated that RHCC-NT provides many advantages over traditional Passive Sample Devices for low MW PAHs. With low equilibration times and selective binding, the biodegradable RHCC-NT is ideally suited for determining changes in PAH concentrations at an oil spill site. This is critical as proportions of the toxic 3--5 ring PAHs changes rapidly at a spill site due to weathering and traditional PAH PSD devices take too long to reach equilibrium for any meaningful short-term measurements. In addition, our RHCC-NT matrix is extremely robust and because it is not susceptible to chemical changes because of temperature and/or pH it has the potential to be used in a wide range of aquatic environments. Further work is ongoing in our group to evaluate the performance of the RHCC-NT in the field. Methods {#Sec8} ======= Chemicals Reagents {#Sec9} ------------------ Individual PAHs (naphthalene, acenaphthene, anthracene, pyrene, chrysene, fluorine and acenaphthylene) were purchased from AccuStandard (New Haven, CT) as 1000 µg/µL solutions in methanol. Ethanol (99.8% HPLC grade), sodium chloride (CAS No. 7647-14-5), sodium dihydrogen phosphate monohydrate (CAS No. 10049-21-5), disodium hydrogen phosphate (CAS No. 7558-79-4), ammonium sulfate (CAS No. 7783-20-2) and Lysogeny broth (LB) (CAS No. 69-52-3) were purchased from Thermo Fisher Scientific. Antifoam 204 (A6426), IPTG (CAS No. 367-93-1), glucose (CAS No. 50-99-7), guanidine hydrochloride (CAS No. 50-01-1), and imidazole (CAS No. 288-32-4) were obtained from Sigma-Aldrich. RHCC-NT expression and purification {#Sec10} ----------------------------------- The inoculum was grown in Lysogeny broth (LB) containing 100 mg/L ampicillin as selective pressure. Two 2.5 L shake flasks each containing 500 mL of LB medium were inoculated from *E. coli* BL21 (DE3) freshly transformed with pET-15b.his6-TCS-RHCC plasmid and incubated at 37 °C, 220 rpm. Once the OD600 value of inoculum was 30 (after 10 hrs) the bioreactor was inoculated with 500 mL of inoculum (5% of the initial working volume). High density *E. coli* growth in the 15 L New Brunswick BioFlo 115 (Eppendorf) was achieved using a fed-batch fermentation method. Ten liters (working volume) of LB containing 100 mg/L ampicillin was used as the initial fermentation medium. Temperature, pH, and dissolved oxygen were kept constant at 37 °C, 7.5, and 30ppm during the experiment, respectively. Antifoam 204 was added only when needed (between 12--18 hrs of fermentation). The inducer IPTG was added to the media after 9 hrs of fermentation (OD600 = 30) at a final concentration of 1 mM. Samples were taken periodically to monitor the cell growth and glucose concentration. Glucose concentrations were measured using a glucose meter (Bioreactor sciences, BRS GM100). Feeding fermentation medium with 20% glucose solution was initiated when the glucose concentration dropped below 2 g/L, which occurred at 6 h of cultivation. Twenty hours from the start of fermentation, 3 L of medium containing cells removed from the bioreactor and replaced with the same volume (3 L) of LB containing 100 mg/L ampicillin. After 28 hrs of cultivation, cells were pelleted for protein purification. RHCC-NT was purified as described previously. Briefly, pelleted cells are suspended in buffer containing 6 M guanidine hydrochloride and lysed via sonication. After centrifugation (Beckman Coulter, Avanti J-26 XPI) to remove cellular debris, the lysate is passed through a cobalt affinity column and washed with resuspension buffer. The RHCC-NT is eluted with buffer containing 200 mM imidazole. The polyhistidine tag is then removed by thrombin cleavage. A cartoon showing the amino acid sequence is presented in Figure [S5](#MOESM1){ref-type="media"}. Fluorescence binding assay {#Sec11} -------------------------- Initial binding screens were performed by titrating known amounts of a single PAH with varying amounts of RHCC-NT. All assays were performed in 20 mM sodium phosphate solution containing 1% ethanol buffered at pH 7 and adjusted to 150 mM ionic strength. All samples were prepared in triplicate to a final volume of 1 mL and were left to incubate overnight at 25 °C to ensure complete equilibration. PAH fluorescence was then measured as a function of RHCC-NT concentration. Steady-state fluorescence spectra were collected using Fluorolog-3 Horiba Jobin Yvon spectrofluorometer (Edison, NJ). Excitation wavelength was set to 275 nm for naphthalene, 300 nm for acenaphthene, 350 nm for anthracene, 334 nm for chrysene and pyrene, 290 nm for acenaphthylene and fluorene. Excitation and emission slits were set to 1 nm band pass resolution. All samples were measured at 20 °C in a 10 × 3 mm^2^ quartz cuvette. The obtained fluorescence data was used to calculate the binding constant *K*~*d*~ assuming that each PAH binds to one of two equal bindings sites in the RHCC-NT, cavity 2 or 3. The fraction of PAH bound and the concentration of unbound RHCC-NT was determined based on the following equation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Fraction\,bound=y=\frac{F-{F}_{0}}{{F}_{\infty }-{F}_{0}}=\frac{{B}_{max}[RHC{C}_{free}]}{{K}_{d}+[RHC{C}_{free}]}$$\end{document}$$where *F* is the experimental fluorescence signal of each sample, *F*~∞~ is the fluorescence of the PAH at infinite protein concentration, *F*~0~ is the fluorescence of the free PAH in solution, *B*~*max*~ is a constant and equal to 1. The y values thus obtained were plotted as a function of free RHCC concentrations. The plotted data were then fit to Equation ([1](#Equ1){ref-type=""}) using Sigma Plot 13 (Point Richmond, CA), yielding the binding constant *K*~*d*~ for each PAH bound to RHCC-NT^[@CR27]^. Time-scale binding assay {#Sec12} ------------------------ The time-scale assay was set-up by addition of pyrene to a 35-fold molar excess of RHCC in 20 mM sodium phosphate solution buffered at pH 7 and adjusted to 150 mM ionic strength. Both solutions were equilibrated to 10 °C prior to the assay. Data was collected on a Fluorolog-3 Horiba Jobin Yvon Spectrofluorometer (Edison, NJ). Excitation wavelength was set to 334 nm; emission was monitored at 384 nm; excitation and emission slits were set to 1 nm band pass resolution. Temperature was kept at 10 °C and controlled with a water bath throughout the experiment. Data was acquired every 1 minute for the first 25 minutes, following by 30 minute intervals until fluorescence signal plateaued and equilibrium achieved. Structural elucidation of RHCC-NT:PAH complexes {#Sec13} ----------------------------------------------- RHCC-NT:Naphthalene was formed by adding a 5-fold molar excess of naphthalene from a 50% (w/v) naphthalene solution in 95% ethanol to 11.2 mg/mL RHCC in 10 mM Tris pH 7.5 I = 154 mM (NaCl). Due to the lower solubility of pyrene in water, RHCC-NT:Pyrene was formed by heating 225 mg pyrene with 1.6 mL of 5.0 mg/mL RHCC in 20 mM Tris pH 8.0 I = 154 mM (NaCl) to 70 °C and incubating for 1 week. After cooling to 20 °C, excess insoluble pyrene was removed by centrifugation and the complex concentrated to 11.0 mg/mL for crystallization. RHCC-NT:Naph crystal setups were performed by mixing 2 μL protein and 2 μL reservoir containing 1.5 M ammonium sulfate and 0.1 M Tris pH 8.5 at 277 K with a 1 mL reservoir. Crystals appeared after 1 week. RHCC-NT:Pyrene crystals were grown in a 0.7 μL protein and 0.7 μL reservoir solution drop over 50 μL reservoir composed of 25% PEG 3350 m 0.1 M Bis-Tris pH 5.5, 200 mM NaCl (Jena Bioscience Crystals JBScreen JCSG +  + 4). Crystals were soaked with 15% glycerol in reservoir solution for 1 min prior to flash freezing in LN2. Data were collected on a Rigaku rotating anode MM-007HF diffractometer with a wavelength of 1.54178 Å in 1° wedges at 100 K. Data were processed with XDS and the CCP4-package. Phases were calculated in Phaser using a polyserine RHCC model based on a previously solved RHCC structure (pdb code 1FE6)^[@CR15]^. For RHCC-NT:Pyrene the search model was truncated to two of the four chains of the RHCC tetramer. The structures were built manually and refined refined crystallographically using the Coot suite and the Phenix software package. Coordinates and chemical restraints for pyrene were generated using JLigand and eLBOW. Molecular dynamics simulations {#Sec14} ------------------------------ Molecular dynamics simulations were performed on three RHCC-PAH complexes (RHCC-naphthalene, RHCC-phenanthrene and RHCC-pyrene) using the GROMACS molecular dynamics simulation package^[@CR28]^ with the Gromos 43a2 force field and an SPC water model. Each complex is representative of a particular ring structure, one ring (naphthalene), two rings (phenanthrene) and three rings (pyrene). In each case, the simulation data were used to compute the standard free energies for transferring the ligand to cavity 2 and cavity 3 of the RHCC-NT from the solvent. For the RHCC-naphthalene complex and the RHCC-pyrene complex, the starting structures for the simulations were the measured X-ray crystal structures 5VKF and 5VH0, respectively. As no structure of phenanthrene bound to the RHCC-NT was available, it was necessary to construct the RHCC-phenanthrene complex by inserting phenanthrene manually into the RHCC-NT cavities in the same position and orientation as the naphthalene and pyrene ligands. The topologies of all three ligands were generated by PRODRG^[@CR29]^. The MD simulation protocols and free energy analysis were based closely on those employed by the authors in a previous investigation^[@CR9]^ of *RHCC*−*NT* and only the principle features are reproduced here. Each RHCC-PAH complex was solvated in a rectangular simulation box with dimensions 5.2 nm × 5.2 nm × 9.8 nm containing 7900 SPC water molecules and a set of charge neutralizing Na^+^ ions (16 for RHCC-naphthalene and 12 for RHCC-phenanthrene and RHCC-pyrene). A minimum distance of 1.2 nm was maintained between the surface of the RHCC-PAH complex and the walls of the simulation box. Particle mesh Ewald was employed to treat electrostatic interactions and the cut-off radius for non-bonded interactions was taken to be 1.0 nm. The initial X-ray diffraction structure was energy minimized using the method of steepest descent until convergence was achieved, typically around 40--50 kJ/mol. Solvation free energies were calculated by performing an independent series of simulations on a single energy-minimized PAH ligand in a periodic box with dimensions identical to those listed above. The system was subjected to three position-restrained equilibrations at 50 K, 150 K and 300 K each for a period of 20 ps. During the 2 ns production simulation, the position restraints were removed and the temperature and pressure of the system (RHCC-NT/cavity ligand/solvent bath) were held at *T* = 300*K* and *P* = 1*atm* with thermostats and barostats of the Berendsen type and time constants *τ* = 0.1 *ps*. The absolute standard state free energy for transferring each PAH ligand from the solvent into one of the cavities of RHCC-NT was computed with the method of double-decoupling^[@CR30]--[@CR32]^ using the thermodynamic paths in Figure [S3](#MOESM1){ref-type="media"}. The upper path represents the conversion of a restrained, fully-interacting ligand bound to the RHCC-PAH complex into a restrained gas-phase particle by turning off the non-bonded interactions while simultaneously restricting the translational degrees of freedom of the ligand to the cavity volume with a flat-bottom harmonic well (FBHW)^[@CR33]^ with functional form *U*~*FBHW*~(*r* \< *r*~0~) = 0; *U*~*FBHW*~(*r* \> *r*~0~) = *k*(*r* − *r*~0~)^2^ where *k* = 1000 *kJmol*^−1^*nm*^−2^, *r*~0~ = 0.55 *nm*. (No other conformational, angle or dihedral restraints were employed in the current simulations, where the *RHCC* − *NT* cavities behave simply as storage receptacles.) Δ*G*~1~ is the free energy for this conversion. Δ*G*~2~ is the free energy cost of removing the FBHW restraint, which includes a correction for the standard concentration. The lower path in Figure [S3](#MOESM1){ref-type="media"} represents the conversion of a single, fully-interacting, unrestrained ligand immersed in solvent into a free gas-phase particle and Δ*G*~3~ = −Δ*G*~*solv*~ is the associated free energy (Δ*G*~*solv*~ is the solvation free energy). The point symmetry group of each PAH ligand would, in principle, require the inclusion of another term in the free energy Δ*G*~*symm*~ = −*RT lnσ*, where *σ* is the rotational symmetry number. For phenanthrene with point group symmetry *C*~2*v*~, Δ*G*~*symm*~ = −*RT ln*4 = −3.5 *kJ mol*^−1^, while for naphthalene and pyrene with symmetry group *D*~2*h*~, Δ*G*~*symm*~ = −*RT ln*8 = −5.2 *kJ mol*^−1^. However, given the absence of body restraints that would artificially restrict the orientation of the PAH ligands, as mentioned above, this term was omitted from the current analysis. The absolute free energy for transferring a PAH ligand to the cavity from the solvent bath, corrected for the standard state, is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{G}_{transfer}^{0}=-\,{\rm{\Delta }}{G}_{1}-{\rm{\Delta }}{G}_{2}+{\rm{\Delta }}{G}_{3}$$\end{document}$$ The terms Δ*G*~1~ and Δ*G*~3~ were calculated with the method of Multi-Configurational Thermodynamic Integration (MCTI)^[@CR30]--[@CR37]^. Linear scaling of the nonbonded parameters was accomplished using a coupling parameter λ which was varied from λ = 0 (ligand interacting) to λ = 1 (ligand non-interacting) while holding the bond lengths and bond angles fixed. The restraint free energy Δ*G*~2~ was calculated analytically^[@CR9]^ using:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{G}_{2}=+RTln({V}_{FBHW}/{V}_{0})$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{FBHW}={\int }_{0}^{\infty }exp(-{U}_{FBHW}(r)/RT){d}^{3}r=(4/3)\pi {r}_{0}^{3}+4\pi {r}_{0}^{2}{(\pi RT/k)}^{1/2}+2\pi {r}_{0}(\pi RT/k)+{(\pi RT/k)}^{3/2}\cong 0.90\,n{m}^{3}$$\end{document}$ is the effective FBHW simulation volume and *V*~0~ is the standard volume *V*~0~ = 1.660 nm^3^ corresponding to a concentration of 1*M* = 1*molecule*/1.660 nm^3^. The minimization and equilibration protocols were repeated for each λ before generating the 2 ns production runs. Structures were saved every 0.5 ps and the accumulated set of 4000 structures was used to compute *dG*/*dλ* for each λ. The mean value *dG*/*dλ* was computed using the Gromacs g_analyse routine and the errors were calculated by block averaging. Soft-core interactions with a soft-core parameter *α* = 0.1 were included to obtain smooth *dG*/*dλ* curves and to eliminate discontinuities in the non-bonded parameters and ensure convergence in the limit *λ* → 1. Supplementary information ========================= {#Sec15} Supplement 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/s41598-018-37323-x. Equal contribution to the research and writing of the manuscript by M.M and O.F is noted. The Natural Science and Engineering Council of Canada (RGPIN-004954-2017 and STGP 479210-2015) funded this project. JS is supported by the CRC program as a Tier-1 Canada Research Chair in Structural Biology and Biophysics. The data collection was performed partially at the Canadian Light Source (CLS). M. Mc and O.F. characterized RHCC-NT uptake via Fluorescence spectroscopy. S.S. and F.H. performed the kinetic studies. M. Mc performed the structural elucidation of the PAH-RHCC-NT complexes. C.H.-V. and R.R. performed the MD simulations. M.K. assisted in the fluorescence data analysis. J.S. led the project, supervised M. Mc. O.F. S.S. and F.H. G.T. supervised the quantitative analysis of PAH uptakes. J.L. and V.P. carried out other experiments, supervised by G.T. and J.S. M. Mc and O.F. analyzed the data, and wrote the paper under the guidance of J.S. Competing Interests {#FPar1} =================== The authors declare no competing interests.

Introduction ============ Diffuse large B-cell lymphoma (DLBCL) represents the most common subtype of non-Hodgkin lymphoma worldwide, accounting for 30%--40% of all newly diagnosed cases.[@b1-ott-9-5349] It is an aggressive disease featuring heterogeneous genetic, phenotypic, and clinical characteristics. The most commonly used initial therapy is an anthracycline-based combinatorial chemotherapy regimen composed of cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP).[@b2-ott-9-5349] Even though complete response rates of 40%--50% with chemotherapy can be attained, most patients who do not respond to CHOP will ultimately succumb to their disease.[@b3-ott-9-5349] The major cause of treatment failure is chemotherapeutic drug resistance, which imposes major obstacles to the successful therapy of DLBCL. Although considerable efforts have been made to understand the molecular basis for development of multidrug resistance in DLBCL, current knowledge remains limited.[@b4-ott-9-5349] Aldehyde dehydrogenases (ALDHs) are a family of intracellular enzymes responsible for oxidizing aldehydes to carboxylic acids.[@b5-ott-9-5349] ALDH1A1, one of 19 ALDH isoforms, is not only crucial for protecting cells from toxic aldehydes but is also known to play important roles in retinoic acid metabolism, cancer development, and drug resistance.[@b6-ott-9-5349] Mounting evidence has shown that ALDH1A1 can offer cellular protection against many cytotoxic drugs.[@b7-ott-9-5349] This protective role was first observed over 2 decades ago when it was reported that leukemic cells with ALDH activity were highly resistant to cyclophosphamide.[@b8-ott-9-5349] Recently, we have demonstrated that knockdown or inhibition of ALDH1A1 could increase chemosensitivity in DLBCL Farage cells.[@b9-ott-9-5349] However, the role of ALDH1A1 in chemoresistance of DLBCL has not been identified completely. Further research efforts are needed to clarify the exact mechanisms of ALDH1A1 in DLBCL chemoresistance. JAK/STAT3 signaling pathway is a well-known cell survival signal that contributes to chemoresistance in a variety of cancer cells.[@b10-ott-9-5349] In lymphoid malignancies, a pathogenic role for STAT3 has been shown in multiple myeloma, anaplastic large T-cell lymphoma, Hodgkin lymphoma, and, recently, in activated B-cell-DLBCL.[@b11-ott-9-5349] Moreover, Alas et al demonstrated that chemical inhibitors of the STAT3 pathway are a novel class of compounds capable of reversing drug resistance in lymphoid-derived tumors.[@b12-ott-9-5349] We have previously demonstrated that knockdown or inhibition of ALDH1A1 could decrease STAT3/NF-κB activity.[@b9-ott-9-5349] However, the pathway through which ALDH1A1 induces chemoresistance is not known, and the potential role of pathway inhibitors in reversing chemoresistance has not been investigated. In the present study, we investigated the role of ALDH1A1 in CHOP resistance using the human DLBCL cell line Pfeiffer transduced with an ALDH1A1-expressing or ALDH1A1-shRNA (short hairpin RNA) lentivirus. Moreover, the correlation between ALDH1A1 expression and clinical drug response was analyzed in DLBCL patients. We also examined the association between ALDH1A1 and the JAK2/STAT3 pathway in relation to CHOP chemoresistance. Taken together, our results support a model in which ALDH1A1 confers resistance to CHOP through the activation of the JAK2/STAT3 pathway, indicating its potential value as a therapeutic target in DLBCL. Materials and methods ===================== Reagents and antibodies ----------------------- Antibodies against ALDH1A1 and phospho-JAK2 (tyrosine 570, tyrosine 931, and tyrosine 1007) were purchased from Abcam (Cambridge, UK). Antibodies against JAK2, STAT3, phospho-STAT3 (tyrosine705, serine 727), Bcl-2, and cyclin D1 were obtained from Cell Signaling Technology (Danvers, MA, USA). Cyclophosphamide and prednisone were purchased from Sigma-Aldrich Co. (St Louis, MO, USA) and dissolved according to the manufacturer's instructions. Doxorubicin and vincristine were obtained from KeyGen Biotechnology (Nanjing, People's Republic of China). All CHOP reagents were stored at −80° C. The JAK2/STAT3 inhibitor, WP1066, was obtained from Selleckchem (Houston, TX, USA). Patients and specimens ---------------------- Eighty-eight patients treated for DLBCL in the Department of Oncology and Hematology, Xiangya Hospital, Central South University (Changsha, People's Republic of China), between January 2012 and November 2013, were selected for this study. The criteria for study enrollment were as follows: 1) histopathologically diagnosed DLBCL, 2) newly diagnosed and without a history of other tumors, 3) availability of paraffin-embedded specimens, and 4) a treatment regime involving at least four cycles of the CHOP regimen as induction therapy. Detailed clinicopathological information of the patients is listed in [Table 1](#t1-ott-9-5349){ref-type="table"}. After the fourth cycle of CHOP therapy, the response to chemotherapy was assessed according to the International Working Group standardized response criteria for non-Hodgkin lymphoma.[@b13-ott-9-5349] We defined the high-sensitivity group as patients showing partial or complete response following CHOP and the low-sensitivity group as patients showing stable or progressive disease. Overall survival time was calculated as the period between the date of diagnosis and the date of death or the last follow-up. The study was approved by the Research Ethics Committee of Xiangya Hospital, Central South University, and written informed consent was obtained from all the patients involved in this study. Cell lines and cultures ----------------------- The human DLBCL cell lines Pfeiffer (CRL 2632) and Farage (CRL 2630) were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). The cells were cultured in Roswell Park Memorial Institute (RPMI) medium 1640 (Corning Incorporated, Corning, NY, USA), supplemented with 10% fetal bovine serum (Corning Incorporated) in a humidified atmosphere of 37°C at 5% CO~2~. Cells were passaged every 2--3 days to maintain a density of 1--2×10^6^ cells/mL. Construction of recombinant plasmids and lentivirus production -------------------------------------------------------------- ALDH1A1 coding sequence and ALDH1A1 shRNA in a recombinant lentivirus gene delivery system were constructed by GeneChem Biomedical Co. Ltd (Shanghai, People's Republic of China) for ALDH1A1 overexpression and knockdown. A blank vector lentivirus gene delivery system was used as a negative control. Lentivirus was transfected into Pfeiffer cells at an optimal multiplicity of infection of 100 supplemented with enhanced infection solution and polybrene (GeneChem Biomedical Co. Ltd) according to the manufacturer's recommended protocol. Validation of the knockdown and overexpression of ALDH1A1 was performed at the protein level by Western blotting. In vitro drug sensitivity assay ------------------------------- Cells were seeded into 96-well plates at a density of 2×10^4^ cells per well in a volume of 100 μL. CHOP were added at the standard clinically used ratio of 80/5.5/0.16/11.1[@b14-ott-9-5349] and concentrations ranging from 40 to 1,280 ng/mL. Cells were treated with various concentrations of CHOP for 48 hours. Control columns contained cells without drug and blank columns contained only the medium. Cell survival was then analyzed via the Cell Counting Kit-8 assay (CCK-8; 7SeaPharmTech, Shanghai, People's Republic of China) according to the manufacturer's instructions. After incubation with 10 μL of CCK-8 reagent for 4 hours, the absorbance was measured at a wavelength of 450 nm by an automatic plate reader. All procedures were performed in triplicate. Colony formation in soft agar ----------------------------- For colony formation assay, a bottom layer (0.6% agar) was prepared with RPMI 1640 medium containing 10% fetal bovine serum. A top layer (0.3% agar) was prepared with the same medium as described earlier but containing 1,000 indicated cells per well treated with or without 640 ng/mL CHOP. Plates were incubated for nearly 14 days at 37°C with 5% CO~2~ in a humidified incubator until the formed colonies were visible. The visible colonies were then counted and images were captured. All procedures were performed in triplicate. Apoptosis assay by flow cytometry --------------------------------- Cells were seeded into six-well plates and treated with or without 640 ng/mL CHOP for 48 hours. Cells were harvested for Annexin V-fluorescein isothiocyanate and propidium iodide staining. The cells were then analyzed by flow cytometry using fluorescence-activated cell-sorting flow cytometer (BD Biosciences, San Jose, CA, USA). All assays were performed independently in triplicate. Western blotting ---------------- Total protein was extracted from cells using radioimmunoprecipitation assay buffer and quantified using bicinchoninic acid assay. Equal amounts of protein were separated by sodium dodecyl sulfate polyacrylamide gel electrophoresis before being electrotransferred to polyvinylidene fluoride membranes. After the membranes were blocked with 5% skimmed milk for 1 hour at room temperature, they were incubated with primary antibodies overnight at 4°C. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as a protein-loading control. After being washed with a mixture of Tris-buffered saline and 0.1% Tween 20 (TBST) three times, the membranes were incubated with horseradish peroxidase-conjugated secondary antibody for 1 hour at room temperature. After washing again with TBST, the protein bands were detected by chemiluminescence. Blots shown are representative of a minimum of three separate experiments. Immunohistochemistry -------------------- The paraffin-embedded sections were cut at 4 mm thickness and then deparaffinized and rehydrated. After washing in phosphate buffered saline, the slides were blocked for endogenous peroxidase with 3% hydrogen peroxide in methanol for 10 minutes at room temperature. Sections were then blocked with 10% goat serum for 10 minutes, followed by incubation with a primary antibody overnight at 4°C. Monoclonal antibodies used were as follows: ALDH1A1, STAT3, and p-STAT3. After washing, sections were incubated with a biotin-labeled secondary antibody for 10 minutes at room temperature. Then, streptavidin peroxidase reagent was applied for 10 minutes. Coloration was developed by a 4-minute incubation with 3,3′-diaminobenzidine solution and the sections were counterstained with hematoxylin. The immunohistochemistry-stained slides were independently evaluated by two pathologists blinded to the clinical parameters. The extent of stained cells was scored as 0 (0%--5%), 1 (6%--25%), 2 (26%--50%), or 3 (51%--100%). The staining intensity was scored as 0 (negative), 1 (weak), 2 (medium), or 3 (strong). The sum of the extent and intensity scores was used as the final staining score (0--6). For the statistical analysis, we divided cases into two groups: low expression (with final scores \<3 points) and high expression (with final scores ≥3 points). Statistical analysis -------------------- All experiments were run in triplicate, and results are presented as the mean ± standard deviation. The association between protein expression levels and clinicopathological parameters was analyzed by the chi-square test. Statistical analyses were carried out using either the Student's *t*-test or an analysis of variance. Kaplan--Meier and log-rank methods were used to analyze survival curves. The Cox regression model was used for analysis of prognostic factors. Differences were considered statistically significant when the two-sided *P*-value was less than 0.05. Statistical analyses were performed with the SPSS 17.0 software (SPSS Inc., Chicago, IL, USA). Results ======= The protein expression level of ALDH1A1 is upregulated in DLBCL Pfeiffer and Farage cells ----------------------------------------------------------------------------------------- We first analyzed ALDH1A1 expression in two human DLBCL cell lines, Farage and Pfeiffer, by Western blot. The human B-cell was used as a control. Although Pfeiffer showed higher ALDH1A1 expression than Farage, the expression in both these cell lines was significantly higher than that of human B-cells (sixfold and 4.5-fold, respectively, *P*\<0.05; [Figure 1A](#f1-ott-9-5349){ref-type="fig"}). Thus, these results indicate that the protein expression level of ALDH1A1 is upregulated in DLBCL (Pfeiffer and Farage) cells. ALDH1A1 is upregulated in CHOP-resistant DLBCL patients and is associated with STAT3/p-STAT3 -------------------------------------------------------------------------------------------- We first determined the level of ALDH1A1 expression in cancer tissues of DLBCL patients by immunohistochemical analysis. ALDH1A1 showed a significantly higher expression in samples from patients in the low-sensitivity group compared with those in the high-sensitivity group ([Figure 1B](#f1-ott-9-5349){ref-type="fig"}). Next, we assayed the expression levels of STAT3 and p-STAT3 in 88 DLBCL specimens by immunohistochemistry and assessed the correlations between the expression levels of STAT3/p-STAT3 and ALDH1A1. As shown in [Table 2](#t2-ott-9-5349){ref-type="table"} and [Figure 1C](#f1-ott-9-5349){ref-type="fig"}, the expression levels of ALDH1A1, STAT3, and p-STAT3 in tissues from low-sensitivity group patients were significantly higher than those from patients in the high-sensitivity group (*P*\<0.05). The correlations between the clinicopathological features of DLBCL patients and expression levels of ALDH1A1, STAT3, and p-STAT3 are summarized in [Table 1](#t1-ott-9-5349){ref-type="table"}. There was a close association between the level of ALDH1A1 and clinical stage and International Prognostic Index. STAT3 and p-STAT3 showed association with DLBCL subtypes. The frequencies of STAT3 and p-STAT3 in the non-germinal center B-cell (GCB)-DLBCL were significantly higher than those in the GCB-DLBCL group (*P*\<0.05). [Table 3](#t3-ott-9-5349){ref-type="table"} shows that the expression of ALDH1A1 was positively associated with that of STAT3 and p-STAT3 in DLBCL tissues. High expression of ALDH1A1 is correlated with poor outcome in CHOP-treated DLBCL patients ----------------------------------------------------------------------------------------- We conducted Kaplan--Meier survival analysis to study the relationship between ALDH1A1 protein expression and survival. Overall survival was significantly shorter in patients expressing high level of ALDH1A1 (log-rank test; *P*=0.0033; [Figure 1D](#f1-ott-9-5349){ref-type="fig"}), with a median overall survival of 27 months. In contrast, the median survival could not be determined for patients with low ALDH1A1 expression because many were still alive at the conclusion of the study. We also investigated the influence of ALDH1A1 expression and other clinicopathological parameters on overall survival using multivariate Cox regression analysis. As shown in [Table 4](#t4-ott-9-5349){ref-type="table"}, ALDH1A1 expression (hazard ratio \[HR\] 3.648; 95% confidence interval \[CI\] 1.655--8.039; *P*\<0.05) and International Prognostic Index (HR 2.353; 95% CI 1.033--5.361; *P*\<0.05) were independent prognostic factors in DLBCL. ALDH1A1 promotes resistance to CHOP in Pfeiffer cells ----------------------------------------------------- To investigate the role of ALDH1A1 in CHOP resistance in DLBCL cells, we modulated ALDH1A1 expression by transfecting Pfeiffer cells with an ALDH1A1-shRNA vector ("Pfeiffer-shRNA") or an ALDH1A1 expression vector ("Pfeiffer-ALDH1A1"). Cells transfected with empty vector ("Pfeiffer-NC") were used as a control. Forty-eight hours after transfection, ALDH1A1 expression was evaluated by Western blot, which showed that Pfeiffer-ALDH1A1 cells displayed a twofold increase of ALDH1A1 protein, while Pfeiffer-shRNA cells showed a 95.6% downregulation compared with Pfeiffer-NC cells ([Figure 2A](#f2-ott-9-5349){ref-type="fig"}). The half maximal inhibitory concentration (IC50) values were determined to evaluate the sensitivity of Pfeiffer-shRNA cells or Pfeiffer-ALDH1A1 cells to CHOP. It was observed that the silencing of ALDH1A1 expression in Pfeiffer cells led to an increased sensitivity to CHOP (IC50: Pfeiffer-shRNA, 314.48 \[272.35--366.06\] ng/mL vs Pfeiffer-NC, 645.30 \[564.33--750.73\] ng/mL). In contrast, overexpression of ALDH1A1 in Pfeiffer cells showed a decreased sensitivity to CHOP (IC50: Pfeiffer-ALDH1A1, 1,122.10 \[979.68--1,324.51\] ng/mL vs Pfeiffer-NC, 648.02 \[566.30--754.66\] ng/mL) (*P*\<0.05; [Figure 2B](#f2-ott-9-5349){ref-type="fig"}). Next, to determine the long-term effects of ALDH1A1 on CHOP-inhibited cell proliferation, colony formation assay was performed. As depicted in [Figure 2C](#f2-ott-9-5349){ref-type="fig"}, when exposed to CHOP treatment (640 ng/mL), the Pfeiffer-shRNA cells showed decreased colony formation capacity compared with Pfeiffer-NC cells, while ALDH1A1 overexpression had the opposite effect (*P*\<0.05). To further investigate the effect of ALDH1A1 on cell apoptosis induced by CHOP, flow cytometry analysis was performed. [Figure 2D](#f2-ott-9-5349){ref-type="fig"} shows that the silencing of ALDH1A1 expression resulted in increased CHOP-induced apoptosis compared with the Pfeiffer-NC group, while ALDH1A1 overexpression in Pfeiffer cells prevented the drug-mediated apoptosis (*P*\<0.05). ALDH1A1 regulates the JAK2/STAT3 pathway in Pfeiffer cells ---------------------------------------------------------- The JAK2/STAT3 signaling pathway is implicated in diverse biological processes, including drug resistance.[@b10-ott-9-5349] We performed Western blot to measure whether ALDH1A1 regulated the JAK2/STAT3 pathway in Pfeiffer cells. First, we hypothesized that ALDH1A1 may regulate the expression of JAK2, an upstream activator of STAT3. Western blotting showed that ALDH1A1 overexpression and silencing could increase and decrease the expressions of p-JAK2 (Y931, Y1007), respectively. In contrast, no impact of ALDH1A1 on the expression of JAK2 and p-JAK2 (Y570) was observed ([Figure 3A](#f3-ott-9-5349){ref-type="fig"}). Then, we investigated whether ALDH1A1 could regulate STAT3 and p-STAT3 in DLBCL. As shown in [Figure 3B](#f3-ott-9-5349){ref-type="fig"}, overexpression of ALDH1A1 significantly increased the protein levels of STAT3 and p-STAT3 (Y705, S727), while ALDH1A1 silencing significantly reduced the protein levels of these targets. Overall, these results indicate that ALDH1A1 might be involved in activating the JAK2/STAT3 pathway in DLBCL cells. ALDH1A1 mediates CHOP resistance through the JAK2/STAT3 pathway --------------------------------------------------------------- In the previous experiments, we found that the expression of ALDH1A1 was positively correlated with that of STAT3 and p-STAT3 in DLBCL tissues. Moreover, ALDH1A1 could regulate the JAK2/STAT3 pathway in Pfeiffer cells. To further explore whether ALDH1A1 mediates CHOP resistance through the JAK2/STAT3 pathway, we used JAK2/STAT3 signaling-specific inhibitor WP1066. The treatment concentration of WP1066 was 2 μmol/L. Western blotting showed that the expression levels of p-JAK2, STAT3, p-STAT3, Bcl-2, and cyclin D1 were increased in ALDH1A1-overexpressing cells, while WP1066 neutralized the effect of ALDH1A1 overexpression ([Figure 4A](#f4-ott-9-5349){ref-type="fig"}). Furthermore, we evaluated IC50, colony formation capacity, and cell apoptosis in Pfeiffer-ALDH1A1 cells treated with WP1066. As shown in [Figure 4B](#f4-ott-9-5349){ref-type="fig"}, WP1066 could increase sensitivity of Pfeiffer-NC cells to CHOP (IC50: Pfeiffer-NC, 644.61 \[514.92--851.01\] ng/mL vs Pfeiffer-NC+WP1066, 339.64 \[270.22--437.68\] ng/mL). Moreover, ALDH1A1-induced resistance to CHOP, according to IC50, was markedly inhibited by WP1066 (Pfeiffer-ALDH1A1: 1,111.28 \[881.06--1,513.95\] ng/mL, Pfeiffer-ALDH1A1+WP1066: 504.894 \[439.335--589.474\] ng/mL; *P*\<0.05). Colony formation assay was also performed to determine the long-term effects of WP1066. As shown in [Figure 4C](#f4-ott-9-5349){ref-type="fig"}, WP1066 attenuated ALDH1A1-induced colony formation in Pfeiffer-ALDH1A1 cells exposed to CHOP. Upregulating ALDH1A1 expression in Pfeiffer cells prevented drug-induced apoptosis while WP1066 impaired these effects ([Figure 4D](#f4-ott-9-5349){ref-type="fig"}). Discussion ========== Despite initial excellent response rates to standard treatment, recurrent disease is still the major cause of mortality in DLBCL.[@b15-ott-9-5349],[@b16-ott-9-5349] The resistance of lymphoma cells to CHOP has been one of the major challenges in the clinical treatment of DLBCL. Therefore, it is of paramount importance to explore the molecular mechanisms involved in chemoresistance in this disease to identify novel therapeutic targets and develop effective treatment strategies. ALDH1A1 is the predominant isoform of a family of cytosolic enzymes that catalyze aldehydes to corresponding carboxylic acids, detoxifying many of the free oxygen radicals generated by chemotherapeutic agents.[@b17-ott-9-5349],[@b18-ott-9-5349] Overexpression of ALDH1A1 in many drug-resistant cancer cell lines and tissues derived from chemotherapy-resistant patients indicates its potential role in chemoresistance.[@b19-ott-9-5349] Herein, we observed that expression of ALDH1A1 was significantly higher in low-sensitivity DLBCL tissues when compared with their high-sensitivity counterparts. In line with this observation, we found that ALDH1A1 is significantly upregulated in DLBCL cell lines compared with human B-cells. In addition to conferring resistance to cyclophosphamide, accumulating evidence suggests that ALDH1A1 may also be capable of mediating resistance to many other agents such as epirubicin, paclitaxel, cisplatin, and tyrosine kinase inhibitors.[@b20-ott-9-5349]--[@b22-ott-9-5349] ALDH1A1 expression can also identify tumorigenic cells in both solid tumors and B-cell malignancies multiple myeloma and classic Hodgkin lymphoma.[@b17-ott-9-5349],[@b23-ott-9-5349] In the present study, we confirmed the function of ALDH1A1 on CHOP resistance in Pfeiffer cells by modulation of ALDH1A1 expression using lentivirus transfection. We found that overexpression of ALDH1A1 increased resistance to CHOP. In contrast, ALDH1A1 silencing with shRNA enhanced CHOP-induced cytotoxicity. Taken together, these findings indicate that ALDH1A1 plays a significant role in CHOP resistance. Importantly, disulfiram, a potent inhibitor of ALDH1A1, has entered a Phase II clinical trial for newly diagnosed glioblastoma ([ClinicalTrials.gov](http://ClinicalTrials.gov) identifier NCT01777919). Hence, investigation of the efficacy of targeting ALDH1A1 in DLBCL may be clinically feasible in the near future. It is generally accepted that JAK2/STAT3 is a critical survival signal, which is involved in cancer development and chemoresistance because of its role in enhancing the expression of survival proteins and preventing cell cycle arrest.[@b10-ott-9-5349] In view of its central role, we investigated whether JAK2/STAT3 is related to the ALDH1A1 regulation of CHOP sensitivity in DLBCL cells. In the current study, we found that expression of ALDH1A1 was positively correlated with that of STAT3 and p-STAT3 in the cancer tissues obtained from DLBCL patients. Moreover, upregulation and silencing of ALDH1A1 increased and decreased the p-JAK2 and p-STAT3 expression in Pfeiffer cells, respectively, suggesting that ALDH1A1 activates JAK2/STAT3 signaling. In addition, we found that treatment of ALDH1A1-Pfeiffer cells with the JAK2/STAT3 inhibitor WP1066 resumed the CHOP cytotoxicity suppressed by ALDH1A1 overexpression. Taken together, our findings may provide new mechanistic insights for CHOP chemoresistance in supporting a model in which ALDH1A1 enhances CHOP resistance through the JAK2/STAT3 pathway. CHOP chemotherapy induces cell death and cell cycle arrest.[@b24-ott-9-5349] Here, we found that upregulation of ALDH1A1 decreased annexin V-positive apoptotic cells induced by CHOP. Our study also showed that the antiapoptotic protein Bcl-2, which is a downstream effector of STAT3, was increased dramatically after overexpression of ALDH1A1. In a wide range of tumor cells, STAT3 has been shown to induce the expression of Bcl-2, the prosurvival protein that prevents cytochrome c release and inhibits mitochondria-dependent cell death.[@b25-ott-9-5349] In addition to inhibition of the intrinsic apoptotic pathway, STAT3 binds directly to the FAS promoter in association with JUN to suppress the transcription of the death receptor, thus leading to inhibition of the extrinsic apoptotic pathway.[@b26-ott-9-5349] Alas et al reported that inhibition of STAT3 activity by AG490 sensitizes resistant non-Hodgkin's lymphoma to chemotherapeutic drug-mediated apoptosis.[@b12-ott-9-5349] In agreement with this observation, we found that the JAK2/STAT3 inhibitor WP1066 also promoted apoptosis induced by CHOP. Moreover, we found that overexpression of ALDH1A1 increased the expression level of cyclin D1, which is another downstream target gene of STAT3. In several cell lines, STAT3 and cyclin D1 associate to form a transcriptional repressor complex that binds to the promoter of the cell cycle inhibitor p21WAF1 to downregulate its expression, thus resulting in inhibition of cell cycle arrest.[@b10-ott-9-5349] Together, these observations suggest that the role of JAK2/STAT3 in ALDH1A1-mediated chemoresistance may be dependent on the inhibition of cell cycle arrest and cell apoptosis. Conclusion ========== The effects of ALDH1A1 on CHOP chemosensitivity in vivo and the applicability of the present findings to other DLBCL cell lines remain to be determined. Notwithstanding these limitations, our study provides a better understanding on both the functional role and molecular mechanism of ALDH1A1 in CHOP chemoresistance of DLBCL cells. Our current work reveals that ALDH1A1 confers resistance to CHOP through activation of the JAK2/STAT3 pathway, indicating its potential value as a therapeutic target in reversing CHOP resistance. This study was supported by the National Natural Science Foundation of China (Grant number 81570200). **Disclosure** The authors report no conflicts of interest in this work. {#f1-ott-9-5349} {#f2-ott-9-5349} {#f3-ott-9-5349} {#f4-ott-9-5349} ###### Correlation between ALDH1A1, STAT3, or p-STAT3 expression and clinicopathological parameters in 88 DLBCL patients Parameters n ALDH1A1 expression STAT3 expression p-STAT3 expression ---------------- ---- -------------------- ------------------ -------------------- ---- ---- ------- ---- ---- ------- Age (years)  \<60 70 30 40 0.586 50 20 0.769 37 33 0.139  ≥60 18 9 9 14 4 13 5 Sex  Male 37 17 20 0.793 29 8 0.311 23 14 0.389  Female 51 22 29 35 16 27 24 B symptoms  Positive 19 12 7 0.062 11 8 0.101 13 6 0.249  Negative 69 27 42 53 16 37 32 LDH  Elevated 33 19 14 0.052 22 11 0.323 21 12 0.317  Normal 55 20 35 42 13 29 26 β2-MG  Elevated 15 10 5 0.056 9 6 0.224 6 9 0.149  Normal 73 29 44 55 18 44 29 CA125  Elevated 29 10 19 0.193 18 11 0.116 13 16 0.111  Normal 59 29 30 46 13 37 22 DLBCL subtypes  GCB 40 15 25 0.240 25 15 0.049 18 22 0.041  Non-GCB 48 24 24 39 9 32 16 Clinical stage  I +II 36 11 25 0.031 23 13 0.121 17 19 0.051  III +IV 52 28 24 41 11 33 19 IPI score  0--2 63 22 41 0.005 43 20 0.104 39 24 0.126  3--5 25 17 8 21 4 11 14 **Abbreviations:** LDH, lactate dehydrogenase; β2-MG, β2-microglobulin; CA125, cancer antigen 125; DLBCL, diffuse large B-cell lymphoma; GCB, germinal center B-cell; IPI, International Prognostic Index; ALDH1A1, aldehyde dehydrogenase 1A1; p-STAT3, phosphorylated-STAT3. ###### Immunohistochemical detection of ALDH1A1, STAT3, and p-STAT3 expression in low-sensitivity and high-sensitivity groups of diffuse large B-cell lymphoma patients Group ALDH1A1 STAT3 p-STAT3 ------------------------ ----------- ------- --------- ----------- ---- ------- ----------- ---- ------- Low-sensitivity group 20 (62.5) 12 0.009 29 (90.6) 3 0.004 28 (87.5) 4 0.000 High-sensitivity group 19 (33.9) 37 35 (62.5) 21 22 (39.2) 34 **Abbreviations:** ALDH1A1, aldehyde dehydrogenase 1A1; p-STAT3, phosphorylated-STAT3. ###### Correlation between ALDH1A1 and STAT3, p-STAT3 in DLBCL ALDH1A1 expression -------------------- -------------------- ---- ------- ------- STAT3 expression  High 33 31 0.238 0.025  Low 6 18 p-STAT3 expression  High 30 20 0.362 0.001  Low 9 29 **Abbreviations:** ALDH1A1, aldehyde dehydrogenase 1A1; DLBCL, diffuse large B-cell lymphoma; p-STAT3, phosphorylated-STAT3. ###### Multivariate analysis of factors contributing to overall survival in DLBCL patients Variables HR (95% CI) *P*-value ---------------- ---------------------- ----------- Age 0.998 (0.972--1.024) 0.852 Sex 1.095 (0.513--2.335) 0.815 B symptoms 1.790 (0.679--4.721) 0.239 LDH 1.539 (0.632--3.748) 0.343 β2-MG 1.644 (0.554--4.883) 0.370 CA125 1.476 (0.643--3.390) 0.358 DLBCL subtypes 0.933 (0.411--2.121) 0.869 Clinical stage 1.328 (0.592--2.978) 0.491 IPI 2.353 (1.033--5.361) 0.042 ALDH1A1 3.648 (1.655--8.039) 0.001 **Abbreviations:** LDH, lactate dehydrogenase; β2-MG, β2-microglobulin; CA125, cancer antigen 125; DLBCL, diffuse large B-cell lymphoma; IPI, International Prognostic Index; ALDH1A1, aldehyde dehydrogenase 1A1; HR, hazard ratio; CI, confidence interval.

1. Introduction {#sec1} =============== Proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors have demonstrated significant effects on most lipoprotein particles and their respective cholesterol content. The recently completed Fourier outcomes trial demonstrated significant reductions in low-density lipoprotein (LDL) cholesterol, nonhigh density lipoprotein (HDL) cholesterol, total cholesterol, Apolipoproteins B (Apo B) and A1, triglycerides, and lipoprotein a in a large population of patients with stable coronary disease using evolocumab \[[@B1]\]. In earlier efficacy trials, both alirocumab and evolocumab had been demonstrated to lower concentrations of intermediate density cholesterol (IDL) \[[@B2], [@B3]\]. The pharmacologic reduction of most particles that include Apo B has shown prognostic benefit \[[@B4]\]. Very low-density lipoprotein (VLDL) remnants, IDL, and chylomicron remnants contain Apo B (either Apo B100 or 48) and are often collectively referred to as remnant cholesterol or triglyceride-rich lipoproteins (TRL). In fasting specimens, directly measured remnant lipoprotein cholesterol has been demonstrated to be associated with increased risk of coronary heart disease and large artery atherosclerotic stroke \[[@B5], [@B6]\]. In nonfasting specimens, a simpler method to estimate remnant cholesterol content has also been suggested to be an independent causal risk factor for ischemic heart disease \[[@B7]\]. While all patients prescribed PCSK9 inhibitors will have unacceptably high LDL cholesterol, all patients prescribed PCSK9 inhibitors will not necessarily have elevated remnant cholesterol or triglycerides. To date, there are limited reports on the effect of PCSK9 inhibitors on remnant cholesterol \[[@B2], [@B3]\]. The purpose of this study was to evaluate the effect of these medications on nonfasting remnant cholesterol in a real world population of patients with FDA- and payer-approved clinical indications for the medication and to confirm that the effect of PCSK9 inhibition on remnant cholesterol is dependent upon the baseline triglyceride level. 2. Materials and Methods {#sec2} ======================== 2.1. Patient Population {#sec2.1} ----------------------- All patients were \>18 years of age and referred to our adult lipid clinic at the West Virginia University Heart and Vascular Institute because of an unacceptably high LDL cholesterol level. All patients had either atherosclerotic cardiovascular disease (ASCVD) or heterozygous familial hypercholesterolemia (FH) without clinical ASCVD as qualifying diagnoses for approval of PCSK9 inhibitors. The vast majority of patients had some adjustment made to their lipid-lowering therapy upon arrival to the lipid clinic. However, all patients were on a stable regimen of lipid-lowering therapy at the time of enrollment and use of the PCSK9 inhibitor. No changes in lipid-lowering therapy were made during the initiation and follow-up of the PCSK9 inhibitor. All patients received approval from their respective payers for the administration of the particular PCSK9 inhibitor that was acceptable to the payer. Criteria used were the patient\'s qualifying diagnosis, documented record of lipid-lowering therapy, and current lipid levels on that therapy. All patients had unacceptably high LDL cholesterol levels on the maximal medical therapy they were able to tolerate, meaning that their LDL cholesterol levels were \>70 mg/dl for ASCVD and \>100 mg/dl for FH. Patients could be on moderate to high-intensity statin therapy, completely statin intolerant on 0 mg of statin, tolerant of small doses of statin, on other nonstatin therapies, or some combination of these. All patients included in this analysis had standard nonfasting lipid laboratory data available just before administration and after the administration of at least 3 doses of the PCSK9 inhibitor. Dosing interval for PCSK9 inhibitors was biweekly. All patients received either evolocumab or alirocumab. Approval to use patient data was obtained from the institutional review board and a waiver of consent was granted. 2.2. Laboratory Testing {#sec2.2} ----------------------- All patients had nonfasting lipid testing. Standard lipid measurements were performed on ARCHITECT c-analyzers using ARCHITECT reagents (Abbott Diagnostics, Abbott Park, IL, USA) including measured lipid profile components (total cholesterol, HDL cholesterol, and triglycerides) and direct LDL measurements at baseline (pretreatment) and minimally after the third dose of PCSK9 inhibitor. An attempt was also made to obtain Apo B measurements before and after therapy. Total cholesterol measurements were performed using standard enzymatic assay employing coupled cholesterol esterase/cholesterol oxidase reactions and detection of quinoneimine product at 500 nm. HDL cholesterol measurements were performed employing selective dissolution of HDL with proprietary detergent and subsequent coupled cholesterol esterase/cholesterol oxidase reactions as for total cholesterol measurement. Triglycerides measurements were performed using standard enzymatic assay employing coupled lipase/glycerol kinase/glycerol phosphate oxidase reactions and detection of quinoneimine product at 500 nm. Direct LDL cholesterol measurements were performed employing selective dissolution of LDL with proprietary detergent and subsequent coupled cholesterol esterase/cholesterol oxidase reactions as for total cholesterol measurement. Apo B measurements were performed at a reference laboratory using the Tina-quant Apolipoprotein B immunoturbidimetric assay (Roche Diagnostics, Indianapolis, IN, USA). 2.3. Calculations {#sec2.3} ----------------- Remnant cholesterol was determined by calculating the difference between non-HDL cholesterol and LDL cholesterol as previously described \[[@B7]\]. Non-HDL cholesterol was determined by calculating the difference between total cholesterol and HDL cholesterol. In addition to the direct measurement, LDL cholesterol was calculated using 2 methods: Friedewald equation (LDL = total cholesterol -- HDL cholesterol -- (triglycerides/5)) \[[@B8]\] and the recently described Martin/Hopkins method \[[@B8]\]. Because the Friedewald equation and Martin/Hopkins method both fail when triglycerides are \> 400 mg/dL, the direct LDL measurement was substituted. 2.4. Statistical Analysis {#sec2.4} ------------------------- All analyses were performed with SPSS Statistics Version 24.0 (IBM Corp., Armonk, NY).*P* value \< 0.05 was considered statistically significant. Continuous variables were presented as mean ± standard deviation (SD) and categorical variables were presented as frequency (percent). Pretreatment to posttreatment lipid measures were compared using paired-samples *t* tests. Pearson correlations were conducted to compare changes in remnant cholesterol with changes in other lipid measures. To examine the impact of pretreatment triglyceride levels on change in remnant cholesterol measures, groups were created above and below the median value of pretreatment triglycerides in this sample (\>223 versus ≤223). Changes in LDL and remnant cholesterol by pretreatment triglyceride groups were examined using repeated measures ANOVA. Multivariable linear regression analyses were used to examine the impact of pretreatment triglyceride groups on change in remnant cholesterol measures after adjustment for age, gender, diabetes, ASCVD, any statin use, and type/dose of PCSK9 inhibitor (evolocumab 140 mg, alirocumab 75 mg, and alirocumab 150 mg). Beta and B coefficient values are presented from these models to show the standardized and unstandardized estimate of effect, respectively, in remnant cholesterol change for each factor in the model, although only factors significantly associated with remnant cholesterol change were interpreted. The B coefficients can be used to form the regression equation for each model and represent the estimate of effect for each variable using the original units of measurement, whereas the beta coefficient is standardized and can demonstrate the relative strength of effect amongst all factors in the model regardless of differing units of measurement. It is possible that the phenomenon of regression to the mean could play a role in the results of this study given that there was only one pretreatment measurement and one posttreatment measurement. Therefore, for the purpose of sensitivity analyses, an equation was used to estimate the regression to the mean effect \[[@B9]\] and doubly robust ANCOVA analyses of the triglyceride group effect on change in remnant cholesterol levels, adjusted again by baseline remnant cholesterol levels, were conducted. 3. Results {#sec3} ========== 3.1. Patient Population {#sec3.1} ----------------------- Between September 2015 and January 2018, 122 patients were treated with PCSK9 inhibitors. At the time of this analysis, 10 patients had not received 3 doses of drug and had not obtained follow-up lipid profiles. Three additional patients failed to get adequate follow-up lipid profiles after beginning medication. Therefore, 109 patients had lipid profile data before and after PCSK9 inhibition, including 62 patients with evolocumab and 47 patients with alirocumab ([Table 1](#tab1){ref-type="table"}). Most patients were treated because of ASCVD (mean age = 65.4 ± 10.4 years). Most with ASCVD had statin intolerance, hypercholesterolemia alone, and coronary artery disease as the principle manifestation and were not on any lipid-lowering therapy. Of the 10% of patients who had familial hypercholesterolemia without ASCVD (mean age = 52.0 ± 15.6 years), most had statin intolerance, pure hypercholesterolemia (*n* = 1 with combined hyperlipidemia) and were not on any lipid-lowering therapy. Diabetes of any type was present in 34% of patients, all with ASCVD. 3.2. Basic Lipid Data {#sec3.2} --------------------- With the exception of HDL cholesterol, all lipid measures decreased significantly after PCSK9 inhibition ([Table 2](#tab2){ref-type="table"}). Examining continuous variable change scores, greater absolute decrease in triglycerides after PCSK9 inhibition was significantly correlated with greater absolute decrease in remnant cholesterol when LDL was derived using any of the 3 methods (Friedewald equation \[*r* = 0.87,*P* \< 0.001\], Martin/Hopkins method \[*r* = 0.77,*P* \< 0.001\], and direct measurement \[*r* = 0.71,*P* \< 0.001\]). The absolute decrease in remnant cholesterol was significantly greater when LDL was derived by Martin/Hopkins method than when derived by Friedewald equation (--14.3 versus --11.7,*P* = 0.002) or measured directly (--14.3 versus --10.99,*P* = 0.022). 3.3. Impact of Pretreatment Triglyceride Levels {#sec3.3} ----------------------------------------------- Repeated measures ANOVA revealed that patients in the high pretreatment triglyceride group (*n* = 54) had greater reductions in remnant cholesterol than patients in the low pretreatment triglyceride group (*n* = 55) regardless of the way LDL was determined (Friedewald equation \[*F* = 20.0,*P* \< 0.001\], direct measurement \[*F* = 11.4,*P* = 0.001\], and Martin/Hopkins method \[*F* = 16.9,*P* \< 0.001\]; [Figure 1](#fig1){ref-type="fig"}). Repeated measures ANOVA also revealed that patients in the high pretreatment triglyceride group had similar reductions in Friedewald derived LDL (*F* = 0.3,*P* = 0.591), direct measurement LDL (*F* = 1.2,*P* = 0.275), and Martin/Hopkins derived LDL (*F* = 1.0,*P* = 0.315) to patients in the low pretreatment triglyceride group ([Figure 2](#fig2){ref-type="fig"}). In multivariable analyses, pretreatment triglycerides group (low versus high) was a significant independent factor associated with changes in remnant cholesterol after PCSK9 inhibition ([Table 3](#tab3){ref-type="table"}). Specifically, patients in the high pretreatment triglyceride group had greater reductions in remnant cholesterol regardless of the way LDL was determined (Friedewald equation \[B = −20.1,*P* \< 0.001\], Martin/Hopkins method \[B = −16.8,*P* \< 0.001\], and direct measurement \[B = −16.3,*P* = 0.001\]). Within these multivariable models, the high pretreatment triglyceride group was associated with a 20.1 point, 16.8 point, and 16.3 point greater reduction in remnant cholesterol, respectively, when LDL was determined by Friedewald equation, Martin/Hopkins method, and direct measurement. Multivariable analyses also demonstrated that the type of drug used was significantly associated with changes in remnant cholesterol when LDL was derived with the Friedewald equation and Martin/Hopkins method, but not when direct LDL measurement was utilized. Specifically, patients who received alirocumab 150 mg had greater reductions in remnant cholesterol than patients who received evolocumab 140 mg when LDL was derived by the Friedewald equation (B = −12.1,*P* = 0.036) and Martin/Hopkins method (B = −12.8,*P* = 0.015). Within these multivariable models, alirocumab 150 mg was associated with a 12.1-point and 12.8-point greater reduction in remnant cholesterol compared to evolocumab 140 mg when LDL was derived by Friedewald equation and Martin/Hopkins method, respectively. The results of sensitivity analysis estimates found that it is possible that the change in remnant cholesterol levels for the low triglyceride group was mostly due to regression to the mean, but the change in remnant cholesterol levels for the high triglyceride group had an effect over and above the effect for regression to the mean. In addition, doubly robust ANCOVA analyses found similar effects for triglyceride group as the primary univariate analyses in the study (Friedewald equation \[F = 3.8,*P* = 0.055\], direct measurement \[F = 4.9,*P* = 0.029\], and Martin/Hopkins method \[F = 5.3,*P* = 0.023\]). 4. Discussion {#sec4} ============= This study demonstrates that remnant cholesterol, like all other Apo B containing lipid fractions, is effectively lowered by PCSK9 inhibition. Importantly, these observations were made in a real world population outside of a clinical trial. In addition, as expected, our findings demonstrate that the degree of remnant cholesterol lowering by PCSK9 inhibition is associated with pretreatment triglyceride concentration, although the same is not true for LDL cholesterol lowering. This finding is consistent with the concept that remnant cholesterol containing larger TRL particles is predicted to be elevated when the serum triglycerides are elevated. While all of the study patients had approved indications for PCSK9 inhibition, most were not treated with high-intensity statins and thus were considered to be statin intolerant. As a result, pretreatment LDL cholesterol was significantly higher in our study patients than those seen in clinical trials \[[@B1]\]. When compared with the results of the Fourier study \[[@B1]\], the percent decrease in nonHDL, Apo B, and LDL cholesterol in our study was comparable. However, the percent decrease in triglycerides was greater in our study, likely reflecting the nonfasting lipid testing and higher pretreatment triglyceride levels. We used 3 different methods to generate LDL cholesterol results and then subsequent remnant cholesterol results. Each of the 3 methodologies has strengths and weaknesses. The commonly calculated and nearly universal Friedewald method should not be used when triglycerides are \> 400 mg/dL; in such instances, direct LDL measurement is preferred. The recently described Martin/Hopkins method \[[@B8]\] utilizes the same measured variables as the Friedewald method. However, when rapid ultracentrifugation is used as the standard, the Martin/Hopkins method is more accurate than the Friedewald method in estimate of LDL cholesterol especially when the LDL levels are low \[[@B10]\]. Nevertheless, the Martin/Hopkins method still has the limitation when triglycerides are \> 400 mg/dL. Direct LDL measurement is not dependent on the measurement of triglycerides for its determination and should not be effected by elevated triglycerides as are the other 2 methods. In fact in the current study, the pretreatment direct LDL was higher than the LDLs estimated by the other 2 methods perhaps reflecting the absence of the triglyceride influence. While directly measured LDL has none of the aforementioned limitations due to triglycerides, it is not as standardized as are the measurements in the standard lipid profile (total cholesterol, HDL, and triglycerides) used to calculate LDL cholesterol. The effect of PCSK9 inhibition on LDL and remnant cholesterol is qualitatively similar using each of the 3 methods. The quantitative differences we may have found in this study will need to be reevaluated using a larger sample size. Current knowledge concerning how PCSK9 inhibitors work indicates that the interaction between LDL particles and the LDL receptor is effected by PCSK9 such that the LDL receptor (as well as the LDL particle) is ultimately destroyed and never recycled back to the hepatic cellular surface. Remnant particles have a somewhat different pathway for clearance \[[@B11]\]; as the larger TRL particles such as VLDL have their triglyceride content reduced by lipase hydrolysis, they eventually become IDL or smaller VLDL particles. Given sufficient time, they can eventually become LDL particles. Most TRL particles have Apolipoprotein E (Apo E) as well as Apo B on their surfaces, acquired during the process of VLDL production in the liver. Apo E is a ligand for the LDL receptor in the liver as well as for hepatic heparan sulfate proteoglycan receptors (HSPG-R). HSPG-R contain hepatic lipase and polypeptide strands used to capture lipoproteins. Due to configurational issues, many VLDL remnants cannot bind to LDL receptors and must rely on HSPG-R for remnant clearance. Remnant particles become attached to HSPG-R via Apo E binding and undergo further triglyceride lipolysis by hepatic lipase. By this mechanism, many remnant particles diminish in size and become LDL particles suitable for clearance by the LDL receptor. Recent information \[[@B12]\] suggests that, in addition to being involved in remnant metabolism, HSPG-R are PCSK9 receptors that facilitate subsequent PCSK9 and LDL receptor complex formation. Much like the interference with LDL clearance, this PCSK9/LDL receptor complex formation would interfere with the process of remnant clearance as well, due to the reduction in available LDL receptors as VLDL and IDL become LDL particles. Therefore, it would be expected that PCSK9 inhibition by currently available monoclonal antibodies would facilitate the process of remnant as well as LDL clearance. The differences found by multivariable analysis concerning the effect of equipotent doses of 2 PCSK9 inhibitors on the reduction of remnant cholesterol may be related to sample size or chance alone. Thus, further study in larger groups of patients is needed. 4.1. Study Limitations {#sec4.1} ---------------------- This study is limited by its lack of a control group, small sample size, lack of specific remnant particle analysis, and direct measurement of remnant cholesterol content. Previously published data involved 81 healthy, normolipemic, nonobese men for evolocumab \[[@B2]\], and 18 healthy volunteers for alirocumab \[[@B3]\]. Evolocumab significantly reduced remnant pool size and increased the fractional catabolic rate of IDL and VLDL \[[@B2]\]. Alirocumab significantly reduced IDL cholesterol content but not VLDL cholesterol content \[[@B3]\]. Larger pools of more clinically relevant patients with appropriate control groups are available from the Fourier \[[@B1]\] (evolocumab,*N* = 27,564), Spire \[[@B13]\] (bococizumab,*N* = 27,438), and Odyssey \[[@B14]\] (alirocumab,*N* = 18,924) outcomes trials. It is unclear whether these large datasets have detailed remnant particle data or directly measured remnant cholesterol as used in prior studies \[[@B5], [@B6], [@B15]\]. Nevertheless, the analysis of remnant cholesterol and risk as presented by Varbo et al. \[[@B7]\] suggests that the simple calculation of remnant cholesterol from non-HDL cholesterol minus LDL cholesterol could be sufficient to assess clinical impact. These simpler remnant cholesterol data are available in the 3 large randomized trial populations and could be explored to confirm the findings of our study and determine whether these changes in remnant cholesterol have a prognostic impact independent of the effect on LDL. We limited our evaluation to nonfasting specimens principally for the sake of patient convenience. However, the use of nonfasting lipids is endorsed by numerous societies, guidelines, and statements including the American Heart Association, the European Society of Cardiology, and the American Association of Clinical Endocrinology \[[@B16]\]. In addition, Varbo et al. \[[@B7]\] have demonstrated through both observational hazard assessments and Mendelian randomization studies that nonfasting remnant cholesterol assessments using the simpler calculation have merit as a means to assess for an effect on clinical outcomes. 4.2. Summary {#sec4.2} ------------ In this real world population receiving PCSK9 inhibition for approved indications and using nonfasting lipid measurements, all Apo B containing particle subgroups had significant reductions in their respective cholesterol concentrations, including remnants. Remnant cholesterol was estimated by the simple subtraction of LDL cholesterol from non-HDL cholesterol. As anticipated, those patients with higher pretreatment triglyceride levels and therefore higher levels of pretreatment remnant cholesterol had a greater decrease in remnant cholesterol after treatment than patients with relatively normal pretreatment triglyceride and remnant levels. Further study in larger populations and with more measurement points will be needed to confirm these findings and determine if these findings will have independent prognostic significance. Data Availability ================= The data used to support the findings of this study are available from the corresponding author upon request. Conflicts of Interest ===================== The authors declare that they have no conflicts of interest. {#fig1} {#fig2} ###### Baseline characteristics of patient sample. --------------------------------------------------------------------------------------     **Total Sample** \   ***N* = 109** ----------------------------- ---------------------- -------------------- ------------ Age (years)   64.1 ± 11.7   Female   65 (60)   PCSK9 Inhibitor        Evolocumab 140mg   62 (57)    Alirocumab 75mg   23 (21)    Alirocumab 150 mg   24 (22)   ASCVD   98 (90)    Coronary   87    Peripheral   2    Cerebral   1    Polyvascular   2    Coronary calcium   6   FH without ASCVD   11 (10)   Diabetes   37 (34)   Lipid diagnosis ASCVD        Hypercholesterolemia alone   61 (56)    Combined hyperlipidemia   37 (34)   Therapy *Statin Intolerance* *Any Statin* *Any LLRx*  ASCVD 78 (80) 20 (20) 44 (45)  FH 9 (82) 2 (18) 2 (18) -------------------------------------------------------------------------------------- ASCVD, atherosclerotic cardiovascular disease; FH, familial hypercholesterolemia; LLRx, lipid lowering therapy; PCSK9, proprotein convertase subtilisin kexin 9. Data presented as frequency (%) or mean ± SD. ###### Lipid levels before and after treatment.   **Pre-treatment** **Post-treatment** \%**Change** ***P* value** ----------------------------- ------------------- -------------------- -------------- --------------- Total cholesterol 259.6 ± 70.1 151.5 ± 52.7 --41% \<0.001 HDLc 45.1 ± 11.3 46.0 ± 11.8 2% 0.228 NonHDLc 215.1 ± 68.6 106.5 ± 52.4 --50% \<0.001 Triglycerides 255.3 ± 161.7 191.9 ± 99.8 --24% \<0.001 LDLc          Friedewald equation 167.0 ± 57.2 70.11 ± 47.2 --58% \<0.001  Direct measurement 174.5 ± 57.2 76.9 ± 46.4 --56% \<0.001  Martin/Hopkins method 171.3 ± 56.4 76.9 ± 46.0 --55% \<0.001 Remnant cholesterol          Friedewald equation LDLc 48.1 ± 28.2 36.4 ± 16.5 --24% \<0.001  Direct measurement LDLc 40.6 ± 29.9 29.6 ± 15.4 --27% \<0.001  Martin/Hopkins method LDLc 43.8 ± 27.2 29.6 ± 13.1 --32% \<0.001 Apolipoprotein B (*n* = 98) 141.4 ± 40.8 74.2 ± 30.5 --48% \<0.001 HDLc = high density lipoprotein cholesterol, LDLc = low density lipoprotein cholesterol. ###### Results of the multivariable linear regression analyses for change in remnant cholesterol when LDL was derived by 3 different methodologies. **Friedewald Equation** **B** **Beta** ***P* Value** -------------------------------------- --------- ---------- --------------- Age --0.21 --0.10 0.708 Female 8.79 0.18 0.074 Diabetes --5.25 --0.10 0.278 ASCVD 0.87 0.01 0.913 Statin --2.45 --0.04 0.669 Alirocumab 75mg vs evolocumab 140mg --8.39 --0.14 0.156 Alirocumab 150mg vs evolocumab 140mg --12.09 --0.21 0.036 Triglyceride group (high vs low) --20.06 --0.41 \<0.001 **Direct Measurement**       Age --0.31 --0.14 0.193 Female 8.39 0.16 0.113 Diabetes --1.61 --0.03 0.757 ASCVD 2.90 0.04 0.734 Statin --8.60 --0.14 0.165 Alirocumab 75mg vs evolocumab 140mg --1.56 --0.03 0.806 Alirocumab 150mg vs evolocumab 140mg --7.29 --0.12 0.236 Triglyceride group (high vs low) --16.29 --0.32 0.001 **Martin/Hopkins Method**       Age --0.27 --0.14 0.185 Female 8.62 0.19 0.053 Diabetes --4.16 --0.09 0.342 ASCVD 1.15 0.02 0.872 Statin --0.87 --0.02 0.866 Alirocumab 75mg vs evolocumab 140mg --6.04 --0.11 0.259 Alirocumab 150mg vs evolocumab 140mg --12.76 --0.24 0.015 Triglyceride group (high vs low) --16.76 --0.38 \<0.001 ASCVD = atherosclerotic cardiovascular disease. [^1]: Academic Editor: Akihiro Inazu

A. V. Diez Roux, S. C. Slesinski, M. Alazraqui, W. T. Caiaffa, P. Frenz, R. J. Fuchs, J. J. Miranda, D. A. Rodriguez, O. L. S. Dueñas, J. Siri, A. V. Vergara, Global Challenges 2019, 3, 1800013 10.1002/gch2.201800013 The SALURBAL Project and LAC‐Urban Health Executive Committees on behalf of the SALURBAL investigators^\[+\]^: ^\[+\]^All SALURBAL Co‐Investigators are listed in the Acknowledgements section. 1. Introduction {#gch2201800013-sec-0010} =============== For the first time, more than half the world\'s population lives in urban areas; by 2050 this figure is expected to approach 70%.[1](#gch2201800013-bib-0001){ref-type="ref"} The dense intersection of the social, built, and natural environments in cities brings many benefits but also poses serious risks to both human health and the environment. Choices about how cities develop and grow, how they organize transportation and land use, how they manage access to quality housing and other resources, and the social and economic policies they prioritize and implement will have profound consequences for the health and environmental sustainability of cities. Cities are also home to large social inequities. Because of residential segregation by class and ethnicity, these social inequities are often manifested spatially as large differences across city neighborhoods in social and physical environments. Social inequities and differences across neighborhood environments have profound impacts on health. The sheer concentration of population in cities and the possibility to intervene proactively to promote the creation of healthy environments present an opportunity to cities not only to improve population health but also to reduce inequities in health and promote environmental sustainability. The past 15 years have witnessed increasing collaborations between urban planners, public health researchers, and other social scientists in the study of how urban neighborhood and urban design may impact health and health inequities, but this research was initially largely focused on high income countries.[2](#gch2201800013-bib-0002){ref-type="ref"}, [3](#gch2201800013-bib-0003){ref-type="ref"} More recently, researchers from the highly urban or rapidly urbanizing regions of the global south have also begun to explore how city living affects health or how urban policies could be used to improve health.[4](#gch2201800013-bib-0004){ref-type="ref"}, [5](#gch2201800013-bib-0005){ref-type="ref"}, [6](#gch2201800013-bib-0006){ref-type="ref"} Discussions of urban health in the context of the New Urban Agenda and the Sustainable Development Goals have focused primarily on access to health care and to some extent on traditional public health issues of water and sanitation and infant mortality with limited discussion of the impact of urban environments on the new epidemics of chronic disease and injuries.[7](#gch2201800013-bib-0007){ref-type="ref"}, [8](#gch2201800013-bib-0008){ref-type="ref"}, [9](#gch2201800013-bib-0009){ref-type="ref"} In addition, there has been limited research into the effects of urban policies on health or health inequities. Several features of Latin America make it uniquely suited to study how urban environments affect health and how urbanization can be managed to create a sustainable and healthy future for the planet. At 80% urban, the Latin American region is substantially more urban than other regions such as Africa (40%) and Asia (48%), although both of these regions are urbanizing rapidly.[1](#gch2201800013-bib-0001){ref-type="ref"}, [10](#gch2201800013-bib-0010){ref-type="ref"} The region now contains 20% of the largest cities in the world.[10](#gch2201800013-bib-0010){ref-type="ref"} Latin American cities encompass not only many megacities but also very large numbers of rapidly growing ("emerging") small and middle‐sized cities. Latin American cities also vary in physical, economic, and social environments. This heterogeneity in size and in other features across Latin American cities can be exploited to investigate how urban environments influence health and environmental sustainability. Another unique feature of Latin American cities is that many city governments have begun to implement policies and interventions that could impact not only health but also long‐term urban environmental sustainability (see, e.g., refs. [11](#gch2201800013-bib-0011){ref-type="ref"}, [12](#gch2201800013-bib-0012){ref-type="ref"}, [13](#gch2201800013-bib-0013){ref-type="ref"}). These include innovations in sustainable transportation (including mass transit and active transportation), urban redevelopment, food policy, and social programs, among others. By studying these policies (via case studies or "natural experiments"), including how and why they were implemented and what their health and environmental impacts might be, we can build a critical evidence base relevant to improving urban health worldwide. Last but not least, the Latin American region is also characterized by large social inequalities. These social inequalities are manifested in cities: 19 of the world\'s 30 most unequal cities are located in Latin America.[14](#gch2201800013-bib-0014){ref-type="ref"} These social inequalities have profound implications for health and health equity. Despite a long tradition of documenting health inequalities in high income countries, there is limited documentation of how health inequities vary across cities, or of what factors or policies ate associated with smaller or larger urban health inequities by class, ethnicity, gender, or place. Understanding the drivers of these inequities, and the policies that might be most beneficial in reducing them, remains a critical need. This article describes the origins and characteristics of an interdisciplinary multinational collaboration aimed at generating and disseminating actionable evidence on the drivers of health in cities in Latin America with the goals of supporting healthy urban policies in the region while at the same time producing generalizable knowledge for urban areas across the globe. The collaboration emerged both to address critical data, research, and policy gaps related to urban health in the region, and to take advantage of the Latin American context to generate knowledge relevant to other regions, especially those in the global south. We describe the processes, the challenges, as well as lessons learned to date in launching this novel initiative. Given the novel nature and the complexity of this multicountry partnership, we describe in detail the history of the collaborations as well as the steps the partnership has taken to formulate a common vision and agenda, and to create the basic administrative structures necessary to function in a way that is participatory, collaborative, and productive. We view the creation of both the vision and the structures as fundamental to the long‐term sustainability of the partnership and to capacity building in the region generally. 1.1. Origins of the Initiative: The Network for Urban Health in Latin America and the Caribbean {#gch2201800013-sec-0020} ----------------------------------------------------------------------------------------------- Building on growing interest in urban health in the Latin American region (including urban health initiatives led by the Pan‐American Health Organization and the first International Society for Urban Health conference in the region in Belo Horizonte in 2011), representatives of academic institutions and international organizations came together convened by the Dornsife School of Public Health in September of 2015. The group, which included representatives from Argentina, Brazil, Peru, and Colombia as well as the United Nations University International Institute for Global Health (UNU‐IIGH) and the Economic Commission for Latin America and the Caribbean (ECLAC), identified urban health research, capacity building, and policy translation as key needs and opportunities for the region. To address these needs, the group agreed to constitute the Urban Health Network for Latin America and the Caribbean (LAC‐Urban Health) with three key goals: (1) to stimulate policy‐relevant research on the state and drivers of health in cities of Latin America and the Caribbean (LAC); (2) to disseminate information and findings to policy‐makers and the public; and (3) to promote translation of existing knowledge into actions that improve health and reduce health inequities, and to provide insights relevant to urban areas worldwide. The group described itself as a "regional learning network" and emphasized the critical role of exchanges and collaborations focused on research, training, and translation. The Network developed a set of eight guiding principles for its work (see **Table** [**1**](#gch2201800013-tbl-0001){ref-type="table"}). These principles have proven to be important value statements that have guided the partnership as it developed. ###### Guiding principles of the Network for Urban Health in Latin America and the Caribbean (LAC‐Urban Health) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ The value of describing the state of urban health (including both physical and mental health) including comparative studies over time and across cities in the region. The importance of understanding the particularities of urbanization (and the potential health consequences of urbanization) in the historical, economic, social, and political context of LAC. Recognition of the multidimensional nature of health in cities, its roots in social, environmental, and other contextual determinants, and therefore the need to seek health solutions beyond the health sector. Understanding how "places" can affect health and the role of place‐based initiatives and policies. The critical importance of addressing large health inequities within and across cities. Promoting capacity building and exchange of trainees. The need to incorporate multiple methodologies including qualitative studies and historical analysis, observational studies, experiments and natural experiments, and systems modeling approaches. The need to understand the health consequences of a range of policies currently being implemented in cities of LAC, not only from the point of view of specific evaluation of policies but also to generate knowledge on the drivers of urban health generally. This necessitates partnerships between researchers, decision‐makers, and communities, and a new way of thinking about the links between research and action. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ John Wiley & Sons, Ltd. Over time, LAC‐Urban Health has grown to over 75 members from many disciplines from more than 10 countries in the region. Regular in‐person meetings, a monthly newsletter, a website (<http://www.lacurbanhealth.org/>), and an active social media presence keep members informed and connected. New collaborations have already emerged. Most importantly, the Network became the context that gave rise to the Wellcome Trust‐funded Salud Urbana en América Latina (SALURBAL) project, through which the Network is working toward its foundational goals. 1.2. Making the Aims of the LAC‐Urban Health Network a Reality: The Wellcome Trust‐Funded SALURBAL Project {#gch2201800013-sec-0030} ---------------------------------------------------------------------------------------------------------- The SALURBAL proposal, funded by the Wellcome Trust in 2017 as part of the "Our Planet, Our Health" initiative was crafted by multiple members of LAC‐Urban Health as a platform through which they could achieve the network\'s research and policy translation priorities while remaining faithful to the network\'s guiding principles. The sense of cohesion, mission, and focus created by the Network were fundamental to the success of the team in developing the proposal. The project aims and approach were developed through an extensive participatory process including an in‐person workshop and subsequent electronic communications that brought together experts in epidemiology, medicine, urban planning and transportation, nutrition, physical activity, environmental and occupational health, sociology, politics, and economics from across the region. A history of successful prior collaborations between subset of team members was critical to the success of this important phase of proposal development. A key element of the proposal development phase was the creation of an agreed‐upon conceptual model of the drivers of the health and environmental sustainability of cities (**Figure** [**1**](#gch2201800013-fig-0001){ref-type="fig"}). The process of developing such a model began during prior meetings of LAC‐Urban Health but it was rendered more focused by the development of the SALURBAL proposal. Key elements of the model include: (1) multilevel determinants, ranging from global trends to local neighborhood environments and individual‐level characteristics; (2) dynamic processes involving feedbacks and dependencies; (3) interrelated nature of social processes and the physical environment; (4) involvement of different stakeholders---civil society, government, nonprofits, businesses, international organizations; (5) role of policies and practices within and outside the traditional health sector; and (6) interrelated nature of human health and environmental sustainability. The model adds to previous urban health frameworks by emphasizing the critical role of urban policies/interventions and by highlighting the interrelatedness of health, equity, and environmental sustainability.[15](#gch2201800013-bib-0015){ref-type="ref"} {#gch2201800013-fig-0001} Based on this conceptual model, gaps identified in the available research, and consideration of unique opportunities in the Latin American context, including a desire for comparative and innovative approaches, the team designed four overall project "aims" under which project activities are organized. 2. SALURBAL Aims and Approach {#gch2201800013-sec-0040} ============================= The SALURBAL project leverages a large, multicountry interdisciplinary team spanning 16 institutions and 7 countries across Latin America (in addition to three U.S. academic partners and two United Nations entities), a multiplicity of data sources, heterogeneity within and among Latin American cities, and intersectoral partnerships to enhance understanding of how changes in urban environments affect health, social inequities in health and environmental sustainability across the globe. The project is structured around four aims: three empirical/analytical aims and one policy dissemination aim. The three empirical/analytical aims were designed to be complementary and reinforce each other over the course of the project (see **Figure** [**2**](#gch2201800013-fig-0002){ref-type="fig"}). {#gch2201800013-fig-0002} The fourth aim is a critical dissemination and translation aim that seeks to communicate the results of the project but also more broadly to promote a new way of thinking about urban health and its drivers among policy makers and the public. SALURBAL aims and key research questions and activities are summarized in **Table** [**2**](#gch2201800013-tbl-0002){ref-type="table"}. ###### SALURBAL aims, sample research questions, and core activities Project aim Sample questions or objectives Core activities ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Aim 1: To quantify the contributions of city and neighborhood‐level factors to differences in levels of health and health inequalities among and within cities. \(1\) What is the impact of city‐level and neighborhood‐level factors on levels of population health in cities? (2) What is the impact of city‐level and neighborhood‐level factors on the magnitude of social inequities in health within cities? (3) What city and neighborhood‐level factors are associated with beneficial health and environmental indicators? Compile link, and document a rich harmonized data resource on health, city, and neighborhood‐level factors in the Latin American region. Units and variables are being defined at each level allowing for a variety of flexible analytical approaches, including multilevel and longitudinal analyses. Aim 2: To evaluate the health and environmental impact of city and neighborhood‐level policies/interventions by capitalizing on natural experiments and by combining quantitative and qualitative approaches. \(1\) What policies or interventions are associated with better population health and lower health inequities within cities? (2) What policies or interventions are related to beneficial urban environmental conditions and lower inequities in exposures to adverse urban environments? (3) What policies or interventions result in the most beneficial health and environmental trajectories? The project team identified four thematic areas of interest based on interventions and policies currently prioritized in the region: mobility and emissions control, comprehensive urban development policies, reduction of social inequities, and promotion of healthy behaviors. Strategic opportunities for health impact evaluation using quantitative and qualitative approaches are being identified through working groups and a call for proposals. Aim 3: To employ systems thinking and formal simulation models in order to (1) better understand the dynamic relations between the urban environment, health, and environmental sustainability; and (2) identify the plausible impacts of selected policies under varying conditions and dynamic relations. \(1\) What are the dynamic relations between the urban environment, health, and environmental sustainability; and (2) what are the plausible impacts of selected policies under varying conditions and dynamic relations? The team identified two areas for systems modeling based on interest in the region, team expertise, and relevance of systems approaches: transportation and food policy. Conduct facilitated workshops\[16,17\] engaging scientists, policy‐makers, and representatives from civil society in order to promote systems thinking and generate causal loop diagrams. This stage will help identify key systems components and refine the research questions In a second stage, we will use system dynamics or agent‐based modeling\[18,19\] to address key questions identified. Aim 4: To engage with the scientific community, the public, and policy makers in order to disseminate findings and translate them into policies and interventions. Objectives include: (1) to promote new ways of thinking about drivers of urban health and the types of policies and interventions that could improve health and sustainability in cities. (2) To engage various stakeholders in research and evaluation process in order to shape questions and facilitate dissemination. (3) To disseminate our vision and our findings broadly. (4) To advocate for and support the translation of research findings into policies and interventions. Conduct stake holder mapping. Incorporate policy‐make input at multiple steps in the project through workshops and other events. Implement rapid research dissemination and policy translation activities across digital, print, and in‐person platforms. Evaluate impact. John Wiley & Sons, Ltd. **Tables** [**3**](#gch2201800013-tbl-0003){ref-type="table"} and [**4**](#gch2201800013-tbl-0004){ref-type="table"} provide examples of the data that will be compiled by SALURBAL and of the types of policies that will be studied for health and environmental impact. ###### Examples of health and environmental data to be compiled by SALURBAL Domain Sample measures -------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Health Mortality data by cause, survey data on physical and mental health and health risks factors in children and adults, objective measures and hospitalization data when available Economic and social structure Poverty and income, Gini coefficient, GDP, unemployment, education Built environment Urban footprint/land cover, compactness, density, street and intersection density, public transport infrastructure by type, active transport infrastructure, water, and sanitation Emissions and natural environment Air pollution, surface temperatures, green space Social and behavioral environment Violence, social cohesion, travel mode and motorization, housing, transit fares, gasoline cost Organizational/Institutional factors Governance, social services, health care, municipal taxation, land use plan, hazards plan, transit subsidies John Wiley & Sons, Ltd. ###### Examples of policy themes and expected impacts that could be assessed Policy themes Sample expected impacts --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Mobility and emissions control. This includes policies related to mobility and emissions control (transit/cycling infrastructure, vehicle restrictions). Proximal impacts on air quality, traffic, availability/density of transit/cycling infrastructureDistal impacts on physical activity levels (and travel mode), weight, respiratory conditions, cause‐specific mortality Comprehensive urban development. This includes policies related to land access and housing subsidies and comprehensive housing/ education/ health programs in poor areas, water/sanitation. Proximal impacts on crime, availability of sanitation and clean waterDistal impacts on child health and nutrition and growth indices, behaviors and risks factors, injuries, mortality Social inequities. This includes policies related to social inclusion and reduction of social inequities (e.g., conditional cash transfers, minimum wage). Proximal impacts on food security and dietary quality, nutrition status and growth indices, mental healthDistal impacts on child health and mortality, adult risk factors, violent deaths Promotion of healthy behaviors. This includes policies related to food, leisure physical activity, and tobacco control (e.g., taxes and regulations regarding availability and labeling). Proximal impacts on food security and nutrition; smoking prevalence, intensity, and cessationDistal impact on malnutrition (overweight and underweight), diabetes, hypertension, cancer, mortality John Wiley & Sons, Ltd. 3. Organization and Coordination {#gch2201800013-sec-0050} ================================ A critical component of a large, multicountry and interdisciplinary initiative like SALURBAL is ensuring an organizational structure that facilitates and promotes engagement of all partners while ensuring productivity and the ability of the project to meet its aims. To address this organizational challenge from the very beginning, SALURBAL developed an organizational and governance structure inspired by other multisite projects. The project is led by the Principal Investigator (PI) with assistance from an Executive Committee composed of five members representing different countries and project cores. The Executive Committee meets regularly to review study progress, troubleshoot any problems, and develop strategies and processes to operationalize the activities necessary to achieve study aims. A larger Steering Committee, including members of country hubs and cores (representing all institutions) meet via video conference call monthly to set broad strategies, establish priorities, monitor progress, and provide input to the PI and the Executive Committee. The entire research team meets in person twice a year to review study progress. A Publications and Presentations Committee (P&P) reviews proposals for presentations and publications. It ensures that priority papers identified by the Executive Committee are completed; promotes the quality of presentations and publications; tracks progress; and identifies gaps and promotes publications. The Executive Committee is serving as the P&P committee during the initial phase of the project. An Ancillary Studies Review Committee will be constituted to oversee the review and selection of proposals for ancillary studies. SALURBAL has convened science and policy advisory boards to obtain timely and expert external advice. The science advisory board provides input into the scientific priorities and approaches of the study and connects the project to other endeavors in related areas. It is integrated by scientists and researchers from a range of relevant fields. The policy advisory board provides input on project priorities with an eye toward policy relevance, provides guidance on policy translation, and connects the project to external stakeholders and policy‐makers. It is integrated by representatives from government, nonprofits, and other partners such as development banks. To achieve study aims the project team is organized into project cores (an administrative core plus three resource cores) and country hubs (see **Figure** [**3**](#gch2201800013-fig-0003){ref-type="fig"}). The *administrative core* provides administrative oversight and coordination. It facilitates interaction and collaboration across country hubs and project cores through project meetings and video conference calls. It engages the advisory boards and supports the review or paper and ancillary studies proposals. In addition, the administrative core coordinates various dissemination, policy and community outreach activities. {#gch2201800013-fig-0003} *Resource cores* (data and methods, physical, and social environment cores) have the function of integrating activities and facilitating the development of collaborative products across countries and institutions involved in the project. Cores work collaboratively and with country hubs. Country investigators are affiliated with one or more cores as appropriate to their expertise and interests. The *data and methods core* (DMC) provides overall data and analytical support to the project. The core pools health and other datasets, harmonizes variables, performs data linkages, and provides statistical and GIS expertise for all analyses. The core also provides expertise on systems approaches This core is responsible for rigorously protecting data privacy, which is especially complex due to the geographically specific nature of the data SALURBAL works with. *The physical and built environment core* provides expertise on the measurement of the physical and built environment, on the identification and characterization of interventions or policies affecting the physical and built environment, and on relevant research questions involving the physical and built environments (or policy interventions on these environments). *The social environment and inequities* core provides expertise on the measurement of the social environment and social inequities, on the identification and characterization of interventions or policies affecting the social environment or social inequities, and on relevant research questions involving the social environment and social inequities (including policies and interventions in this domain). Each *country hub* is led by one lead or two colead institutions (and a country PI or Co‐PIs) that coordinate and integrate activities at all institutions engaged with the project throughout the country. Country hubs facilitate access to appropriate health, mortality, environment, and other datasets; they actively engage in data harmonization efforts, data analyses, and in the development of manuscripts for presentation and publication as lead authors or coauthors; they contribute scientific and data expertise to study cores and they engage in the dissemination of evidence to policy‐makers and communities. In addition to the resource cores, various working groups have been organized focused on specific areas of emphasis. A systems working group has been constituted to advance Aim 3. A food environment working group is coordinating analyses related to food environments and health. Working groups focused on urban redevelopment and mass transit are exploring opportunities for health impact evaluation of urban development activities and the installation and implementation of new transit interventions. 4. Progress to Date {#gch2201800013-sec-0060} =================== Since its launch, SALURBAL has made substantial progress. A critical component has been the consolidation of the organizational and governance structure described above, with input and participation from all country teams. The DMC has worked collaboratively with researchers from the project\'s country hubs to identify two key foundational elements of the research: (1) identification of the "universe" of SALURBAL cities, including ways of operationalizing them so that they can be linked to available health data; (2) the compilation, harmonization, and geographic linkage of mortality and survey data. The DMC has also created a flexible data structure composed of cities, subcity units, and neighborhoods that can be linked to health, mortality, and other data at different levels allowing a flexible set of analyses. The physical and social environment cores have each identified a core set of domains and the levels at which they can be characterized and have initiated the process of data compilation for each domain. The project has launched a policy scoping paper aimed at characterizing the range of potentially health relevant urban policies that are currently being implemented or have recently been implemented in cities of the region. The review will identify research gaps and useful examples for future analysis, guiding policy evaluation work during the subsequent years of the project. Simultaneously, SALURBAL is working with various potential partners to identify policies and interventions that are planned for construction or initiation during the first two years of the project\'s lifespan. The team plans to take advantage of these natural experiments to understand their impact on health, equity, and environmental sustainability. Aim 3 activities to date include a training for 15 team members in the tools of participatory group model building. Following this training the team developed plans for three regional workshops with a diverse set of stakeholders (government, nonprofits, community members, international organizations, academics). Workshop goals include: (1) to engage diverse stakeholders in the SALURBAL project and vision; (2) to provide stakeholders with experience in the application of systems approaches to urban health problems, and (3) to obtain stakeholder input that will help identify and prioritize research questions to be pursued by the SALURBAL project using systems modeling in the future. The first workshop was held in Lima in November 2017. As part of the activity, stakeholders developed causal loop diagrams illustrating the relations between transport policies and health and food policies and health. Two additional workshops are planned for the first half of 2018. The results of the workshops will be summarized in a report and a manuscript for publication. To further focus systems work, the team has also launched two systematic reviews (on transport and food policy) that will inform the focus of future modeling efforts. The project has developed a research dissemination and policy translation strategy and has launched several related activities. These include the creation of formal communication channels to LAC‐Urban Health members such as a monthly newsletter and an active social media presence. A public website for LAC‐Urban Health, which will be a primary platform for research dissemination, has been under development. SALURBAL has hosted two policy‐maker‐focused events (Bogota, May 2017; Lima, November 2017) attached to regional meetings. In these events, local policy‐makers presented on challenges and opportunities they face in improving health in cities using an intersectoral perspective. These events have resulted in one policy brief on sustainable transportation and health (already produced) and a second brief on food systems and health (under development). The project has also been active presenting work at in various venues including the International Society for Urban Health and the World Urban Forum. To jumpstart the development of papers, the team engaged in a participatory process to identify promising "low hanging fruit" ideas that can be pursued as scientific papers early in the project. This had multiple objectives including team building, identifying early feasible but also impactful papers, and focusing the data compilation efforts. Researchers from all participating countries together with LAC‐Urban Health members present at the first in‐person SALURBAL meeting heard presentations on what data might be available during the first year of the project (including health survey data, mortality data, census data, physical and built environment data, and social environment data) and worked in small groups to brainstorm how they could capitalize on that data to answer more specific research questions during the first two years of the project. These small working group sessions were organized around five general themes: mortality data, urban form and physical environment data, social environment data, food environment data, and survey data. These discussions resulted in a preliminary list of research questions that was subsequently refined and prioritized. Writing group members were identified through a participatory process. Paper proposals based on these ideas (as well as new ideas emerging subsequently) have been reviewed by the Publications and Presentations Committee and data compilation efforts linked to these proposals are underway. 5. Lessons Learned {#gch2201800013-sec-0070} ================== SALURBAL is still in its early stages. However, a few key lessons that may be relevant to similar initiatives across the global south have already emerged. 5.1. Building the Team {#gch2201800013-sec-0080} ---------------------- The team is critical to the feasibility and success of a project like SALURBAL. Several sets of team members had collaborated in various capacities before SALURBAL, but others had not. Once a core set of investigators had been identified, the team worked to expand the group via networks to encompass countries that were not represented and to expand on the expertise and disciplines included. The existence of the fledgling LAC‐Urban Health as well as the goals and values of the network were fundamental to the ability of the team to come together as it created a platform within which SALURBAL could be nested. SALURBAL aims are perfectly aligned with LAC‐Urban Health goals, and thus SALURBAL became a project through which the network can pursue its agenda. An important aspect of the SALURBAL team is the openness to the engagement and inclusion of researchers, especially junior researchers, who are not formally supported through SALURBAL. Three key strategies have been used to promote the engagement of others throughout the region: (1) All LAC‐Urban Health members have been invited to SALURBAL meetings and selected members have been supported to attend SALURBAL meetings through additional funds provided by the network sponsors; SALURBAL meetings have also included a third day focused on Network activities; (2) the paper development and paper proposal process is open to outside investigators in order to encourage their engagement and participation; and (3) SALURBAL actively encourages and supports additional grant submissions that are synergistic and complementary with SALURBAL aims in order to facilitate the funding of investigators not currently supported through SALURBAL. 5.2. Organizational Structure and Processes That Promote Transparency and Engagement {#gch2201800013-sec-0090} ------------------------------------------------------------------------------------ SALURBAL has invested considerable effort in creating a structure that promotes strategic leadership, the development of various policies and guidelines, engagement, and accountability. The Executive and Steering Committee meetings are used to discuss and agree on complex issues such as how data will be obtained and shared, how publications will be coordinated, and to review goals and agendas for major meetings as well as to review work plans for various parts of the project. They also serve to share and socialize important information. The creation of policies, such as the publications policy, in ways that promote and enhance collaboration with clear guiding principles in mind is also fundamental. For example, the SALURBAL Publications Policy clearly articulates four guiding principles: (1) to support and encourage the development and publication of high‐quality papers related to the aims of the project among all team members; (2) to ensure that the project fulfills its top scientific priorities across a range of areas and its deliverables to the funder; (3) to ensure that all team members have the opportunity to participate in publications; and (4) to foster capacity building for the entire team as well as the career development of junior and early stage investigators through their appropriate involvement in the publication process. The process for jump‐starting SALURBAL papers was also designed and implemented with the goals of maximizing transparency and participation while ensuring productivity and accountability. 5.3. Balance of Practicality and Aspirations {#gch2201800013-sec-0100} -------------------------------------------- A key lesson that has emerged is the importance of balancing aspirations with practical realities. This implies being ambitious in goals but also being practical in the sense of identifying the "low‐hanging fruit" activities that can move forward at a faster pace while more sophisticated analyses and activities are being developed. The use of data is a case in point. In Latin America, as in many countries, health data present many challenges. There are challenges in access, in harmonization, in quality, and in completeness. And yet there is much that can be done even with imperfect data, and the insights provided by rigorous analyses of imperfect data will spur data improvements in the future. Moreover, pushing forward with imperfect data often reveals that there is more data or better data than originally anticipated. SALURBAL is already discovering a wealth of untapped data relevant to urban health, as the project works to provide, for the first time, information of the heterogeneity within and between cities in the countries of the Latin American region. SALURBAL has moved forward with the principle of balancing practicality and feasibility with perfection, in ways that maximize the informativeness of the work and promote better and better work as the project evolves. 5.4. Careful Consideration of Early Decisions as They May Have Long‐Term Implications {#gch2201800013-sec-0110} ------------------------------------------------------------------------------------- Early in the process of compiling SALURBAL data it became obvious that key early decisions related to definitions and structures could have long‐term implications, and that therefore careful consideration need to be given to these foundational decisions. A key example is methods used to define and operationalize the universe of cities that would be included in the project. For example, "cities" can be defined in various ways: based on country‐specific definitions of metropolitan areas, based on administrative boundaries linked to censuses, or based on urban footprints assessed through data‐driven efforts.[10](#gch2201800013-bib-0010){ref-type="ref"} Various definitions have major implications for data compilation and linkages. Another key decision is the selection of a data structure that allows the incorporation and linkage of data with varying levels of aggregation and spatial specificity and that can be shared across various cores and types of data. SALURBAL addressed these challenges through deliberative discussions and the development of processes and standards encoded in study documentation that are systematic and standardized and yet flexible enough to accommodate a variety of circumstances. 5.5. Capacity Building as Key and Broad {#gch2201800013-sec-0120} --------------------------------------- SALURBAL views capacity building as integral to all its activities. The multidisciplinary team includes and supports junior and senior researchers as well as trainees. SALURBAL provides a unique opportunity to build research capacity and regional collaboration across multiple disciplines in ten countries. The project has planned specific capacity building activities such as workshops (two training activities have already occurred). The engagement of trainees and early career investigators, critical to the long‐term success of SALURBAL, has been promoted through the attendance at meetings and through a publications process and policy that explicitly encourages and supports junior investigators. However, capacity building also extends to building experience in how to work in large multicountry teams. In this sense, the creation of the organizational structure and policies of SALURBAL is a critical capacity building activity. The team is building capacity for multisite, cross‐country work in ways that are new to many in the region. Capacity building also extends beyond the SALURBAL team to regional and country stakeholders through their engagement in data processing and data sharing activities, through the support for data linkages that will benefit local partners, and through activities like the participatory group modeling workshops. In many ways, this broad view of capacity building may be one of the most lasting legacies of the project. 5.6. Openness to Serendipity and Unanticipated Opportunities {#gch2201800013-sec-0130} ------------------------------------------------------------ A critical element in projects like SALURBAL is the need to have the flexibility to be open to new opportunities and unanticipated directions that emerge as the project develops. For this reason, these projects are better conceptualized as research networks than as strictly and very tightly predefined research studies. The opportunity to be open to new collaborations and opportunities is critical to their success. A case in point is provided by Aim 2 of SALURBAL. The specific policy evaluation opportunities were purposefully left open to maximize our ability to capitalize on the most rigorous evaluation contexts. An example that has emerged is the health impact evaluation of a new aerial tram project that will be implemented in a low‐income area in Bogotá over the next year. This could not have been anticipated when the proposal was written and the flexibility in the project has allowed us to deploy expertise and resources to capitalize on this important opportunity. This flexibility and adaptability is critical to SALURBAL\'s ability to produce the most informative and rigorous research. 5.7. Sustainability and Partnerships {#gch2201800013-sec-0140} ------------------------------------ It is critical that projects like SALURBAL be developed with an eye toward sustainability after the funding period ends. SALURBAL has approached this by nesting the project and its communications within the broader LAC‐Urban Health Network and the Urban Health Collaborative (UHC) at the Drexel University Dornsife School of Public Health. The Network will ensure continuity of the effort after the project ends. Indeed, since initiating SALURBAL, LAC‐Urban Health has seen expansive growth of its network, which can be seen in the network map below in **Figure** [**4**](#gch2201800013-fig-0004){ref-type="fig"}. We anticipate that during the five years of the project, this growth will continue, leaving behind a strong and well‐connected network primed to continue pursuing the research priorities of LAC‐Urban Health. Aligning our work with the objectives of our partners be they local academic, city governments, or international organizations is also key to creating the momentum and structures that will allow continued investment and support for this kind of work after funding for SALURBAL ends. {#gch2201800013-fig-0004} In addition, SALURBAL has been proactive in supporting new grant submissions even early in the history of the project to allow off shoots and supplementary activities that will expand and capitalize on the SALURBAL resource long into the future. Building partnerships and extending the project reach through continuous incorporation of new team members is also a way to ensure sustainability. SALURBAL has created partnerships with two United Nations agencies and is exploring partnerships with health organizations, development banks, and other regional stakeholders. 6. Challenges Ahead {#gch2201800013-sec-0150} =================== SALURBAL is groundbreaking in its vision and team but it is a complex project and several challenges lie ahead. Key challenges include: (1) keeping the full team engaged despite geographic distances and heterogeneity in expertise, experience, and primary language (the team includes Spanish, Portuguese, and English speakers); (2) facilitating data access and analyses in a context that has limited experience with data sharing and pooling while protecting confidentiality and privacy; (3) producing rigorous research in a timely manner in the context of imperfect data and supporting improvements to data collection for the future; (4) capitalizing on emerging opportunities in a way that is timely and informative; (5) engaging with policy‐makers and other nonacademic stakeholders from the very beginning of the project in a way that influences the direction of the project so that its impact is truly transformative; and (6) developing strategies to communicate key facts to policy‐makers and the public in ways that are clear and impactful. The project has aimed to position itself to address these challenges in several ways. We have a governance structure and have planned a set of systematic engagement opportunities virtually and in person that maximize engagement. We have developed a publications policy that is open and transparent and aimed at supporting and promoting the contributions of all team members. We have developed draft data use agreements that can be tailored and adapted to country needs and requirements. We have created a flexible data structure that allows us to deal with varying levels of geographic specificity and have detailed procedures for ensuring data confidentiality. Through the paper proposal development process, we have aimed to balance early productivity with more refined analyses as the project develops. Our data compilation efforts have proceeded in steps, beginning with the most easily available and proceeding to data that may be more difficult to access and process. We provide analytical and methodological support to papers in various ways, locally and centrally. Key documents are provided in English, Spanish, and Portuguese. Meetings are, at times, multilingual to accommodate all team members. We have designed the project to allow flexibility in identifying and supporting policy evaluation opportunities as they arise. This is accomplished through funds set aside for an ancillary studies mechanism to support these activities. We have also set in place activities to engage stakeholders early in the process, not waiting until the project itself has produced new results. These activities include policy briefs based on existing knowledge, engagement in systems thinking through the group modeling workshops, and the presence of SALURBAL at events like the Ninth Word Urban Forum. We have also set in place mechanisms for evaluating the effectiveness of our policy‐maker engagement activities. Prioritizing new collaborative opportunities and new partnerships so that the project remains open to new ideas and opportunities but still retains focus and does not get distracted from delivering on its key goals is also a major challenge. SALURBAL is addressing this challenge by exploring many new opportunities but also prioritizing our efforts and resources so that the activities we pursue are synergistic and do not distract from our key goals. 7. Conclusion {#gch2201800013-sec-0160} ============= The mere existence of the SALURBAL project is transformative, and we have already begun to see early fruits of this ground‐breaking collaboration. There are many challenges ahead but the structures and processes we have put in place will help address these challenges as the project advances. In a transformative project of this magnitude and ambition, it is to be expected that not all will turn out as planned and that new opportunities that we did not anticipate will arise. Progress to date shows that SALURBAL is well poised to capitalize on opportunities, to overcome barriers, and to deliver on its key aims. We envision a future where other regional networks focused on urban health and sustainability may emerge, capitalizing and building on the lessons of SALURBAL, providing more relevant evidence and engagement around urban health, and partnership with SALURBAL to promote health, equity, and sustainability of cities worldwide. Conflict of Interest {#gch2201800013-sec-0180} ==================== The authors declare no conflict of interest. Ana V. Diez Roux, Amy Auchincloss, Brent Langellier, Gina Lovasi, Leslie McClure, Yvonne Michael, Harrison Quick, Duane Alexander Quistberg, Jose Tapia Granados: Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania, USA. Marcio Alazraqui, Hugo Spinelli, Carlos Guevel, Vanessa Di Cecco, Adela Tisnés, Carlos Leveau, Adrián Santoro, Damián Herkovits: National University of Lanus, Buenos Aires, Argentina. Waleska Teixeira Caiaffa, Amélia Augusta de Lima Friche: School of Medicine, Federal University of Minas Gerais, Brazil. Nelson Gouveia: Universidad de São Paulo, São Paulo, Brazil. Mauricio Barreto: Oswaldo Cruz Foundation, Salvador Bahia, Brazil. Leticia Cardoso, Maria de Fatima de Pina: Oswaldo Cruz Foundation, Rio de Janeiro, Brazil. Patricia Frenz, Tania Alfaro, Pablo Ruiz: School of Public Health, University of Chile, Santiago, Chile. Alejandra Vives Vergara, Alejandro Salazar, Andrea Cortinez‐O\'Ryan, Cristián Schmitt, Francisca Gonzalez, Fernando Baeza, Flavia Angelini: Department of Public Health, Pontificia Universidad Católica de Chile, Santiago, Chile. Olga L. Sarmiento, Diana Higuera, Catalina González: School of Medicine, Universidad de los Andes, Bogotá, Colombia. Felipe Montes, Andres F. Useche, Oscar Guaje, Ana Maria Jaramillo, Luis Angel Guzmán: School of Engineering, Universidad de los Andes, Bogotá, Colombia. Philipp Hessel, Diego Lucumi: School of Government, Universidad de los Andes, Bogotá, Colombia. Jose David Meisel: Universidad de Ibagué, Ibagué, Colombia. Eliana Martinez: Universidad de Antioquia, Medellín, Colombia. María F. Kroker‐Lobos, Manuel Ramirez‐Zea: INCAP Research Center for the Prevention of Chronic Diseases (CIIPEC). Institute of Nutrition of Central America and Panama (INCAP), Guatemala City, Guatemala. Tonatiuh Barrientos‐Gutierrez, Carolina Perez‐Ferrer, Javier Prado‐Galbarro, Filipa de Castro, Rosalba Rojas‐Martínez: Instituto Nacional de Salud Pública, Mexico City, Mexico. J. Jaime Miranda, Francisco Diez‐Canseco: School of Medicine, Universidad Peruana Cayetano Heredia, Lima, Peru. Ross Hammond: Brookings Institute, Washington, D.C., USA. Daniel A. Rodriguez, Iryna Dronova, Melissa Burroughs‐Pena: Department of City and Regional Planning, the University of California Berkeley, USA. Brisa N. Sanchez: University of Michigan School of Public Health, Ann Arbor, Michigan, USA. Joaquin Barnoya, Peter Hovmand, Ellis Ballard: Washington University in St Louis, St. Louis, Missouri, USA. Ricardo Jordan Fuchs, Juliet Braslow: Economic Commission for Latin America and the Caribbean (ECLAC). Jose Siri: United Nations University International Institute for Global Health (UNU‐IIGH).

INTRODUCTION {#s1} ============ Legionella pneumophila is an intracellular human pathogen that multiplies within alveolar macrophages and causes a severe pneumonia known as Legionnaires' disease ([@B1][@B2][@B3]). In the environment, L. pneumophila thrives in many different protozoan cells ([@B4][@B5][@B6]), which serve as its training grounds for pathogenesis ([@B7]). Inside its eukaryotic host cells, the bacterium remodels its phagosome to generate the *Legionella*-containing vacuole (LCV) ([@B8], [@B9]). Establishment of the LCV depends on the Icm/Dot type IV secretion system, which delivers more than 300 effector proteins, which modulate numerous host-cell functions during infection ([@B10][@B11][@B14]). The enormous number of effectors that participate in LCV establishment and the various host cell pathways manipulated by L. pneumophila effectors ([@B15][@B16][@B17]) imply that a successful infection will require different levels of coordination among the effectors, including on the level of gene expression. To date, five regulatory systems have been shown to directly regulate the expression of effector-encoding genes (EEGs) in L. pneumophila: (i) the PmrAB two-component system (TCS) activates the expression of about 40 EEGs ([@B18], [@B19]); (ii) the CpxRA TCS activates or represses the expression of about 30 EEGs and also regulates the expression of several *icm/dot* genes ([@B20][@B21][@B23]); (iii) the LetAS-RsmYZ-CsrA regulatory cascade represses the expression of about 40 EEGs during exponential phase ([@B24][@B25][@B31]); (iv) two Fis regulators repress the expression of about 20 EEGs during exponential phase ([@B32]); and (v) the Fur regulator controls the expression of a single EEG (*mavN*) as well as several other proteins involved in iron acquisition ([@B33][@B34][@B37]). In addition, these regulatory systems have been shown to assemble into an interconnected regulatory network using accessory components, such as modulators (AckA-Pta and PTS^Ntr^ \[[@B22], [@B38]\]) and connectors (LetE and LerC \[[@B22], [@B39], [@B40]\]). All the direct regulators of EEGs described above (PmrA, CpxR, CsrA, Fis1, Fis3, and Fur) were found to be present in all the *Legionella* species examined ([@B41], [@B42]), and they function as global regulators that regulate the expression of a large number of genes, including EEGs, scattered throughout the L. pneumophila genome ([@B18][@B19][@B20], [@B22]). Besides global regulators, local regulators that regulate the expression of a small number of adjacent genes are also common in many bacterial systems ([@B43], [@B44]), including pathogenesis-related genes in Vibrio cholerae and Salmonella enterica ([@B45]). Local regulators were shown to function as either activators or repressors, and in many cases, they are found in genomic islands together with the genes they regulate ([@B46][@B47][@B49]). Genomic islands are genetic elements acquired via horizontal gene transfer (HGT) that include sets of genes that encode proteins that may be beneficial for the bacteria under certain conditions ([@B50], [@B51]). Many pathogenicity islands have been described in bacteria, some of which contain a large number of genes and encode complete pathogenesis systems ([@B50], [@B52]). For example, the Salmonella enterica pathogenicity island 2 (SPI-2) encodes the complex components of a type III system, effector proteins, and the SsrAB TCS, which functions as a local regulator and coordinates their expression ([@B53]). Smaller pathogenicity islands that encode a few proteins are sometimes referred to as islets ([@B50]). For example, the Streptococcus pneumoniae RlrA pathogenicity islet encodes the RlrA local transcriptional regulator, which controls the expression of genes located on the same islet that are essential for lung infection ([@B54]). Another example is the Listeria monocytogenes virulence gene cluster that encodes phospholipases, listeriolysin, metalloprotease, the ActA protein, and the PrfA transcriptional regulator, which controls their expression as well as the expression of genes outside the island ([@B55]). All known L. pneumophila TCSs that regulate the expression of EEGs are global regulators found in all the *Legionella* species sequenced. They are not part of genomic islands, and the signal sensed by their cognate sensor histidine kinases (SHKs) is unknown. Here, we describe a novel L. pneumophila TCS effector island. The TCS is composed of the LciS SHK, which specifically senses copper and activates the cognate LciR response regulator (RR). LciR functions as a local regulator, activating the expression of a single adjacent EEG (*lciE*). The LciRS-LciE genomic island undergoes HGT throughout the *Legionella* genus and represents a novel type of effector regulation in *Legionella*. RESULTS {#s2} ======= The CpxR and PmrA direct regulators of effector-encoding genes (EEGs) belong to the winged helix-turn-helix (wHTH) family of response regulators (RRs), which function as part of the CpxRA and PmrAB two-component systems (TCSs) ([@B18][@B19][@B20], [@B22], [@B23]). L. pneumophila harbors a third TCS from the same family, consisting of lpg0714, which encodes a sensor-histidine kinase (SHK), and lpg0715, which encodes a wHTH-type RR. This TCS is found in L. pneumophila, in five other characterized *Legionella* species, and in three uncharacterized *Legionella* species. In all these species, a gene, which was shown in L. pneumophila to encode an effector protein ([@B56]) (lpg0716 in L. pneumophila), is located next to it, forming a regulator-effector island ([Fig. 1A](#fig1){ref-type="fig"}; see also [Fig. S1](#figS1){ref-type="supplementary-material"} in the supplemental material). {#fig1} 10.1128/mBio.03232-19.1 Presence and absence of lpg0714-lpg0715-lpg0716 in the *Legionella* genus. A maximum likelihood tree of 41 *Legionella* species was reconstructed on the basis of concatenated amino acid alignment of 78 orthologous open reading frames ([@B41]). For each species, the presence (grey) or absence (white) of lpg0714, lpg0715, and lpg0716 genes is indicated. Download FIG S1, TIF file, 0.9 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . The lpg0714-lpg0715-lpg0716 island undergoes horizontal gene transfer in the *Legionella* genus. {#s2.1} ------------------------------------------------------------------------------------------------ Comparison of the genomic location of the regulator-effector island in the nine *Legionella* species in which it was found indicated that it is usually located in different positions ([Fig. 1A](#fig1){ref-type="fig"}). However, in genomes of the four closely related species (L. worsleiensis, L. moravica, L. quateirensis, and *Legionella* sp. strain Km535), it is positioned in a similar genomic region, which underwent additional changes in its gene content ([Fig. 1A](#fig1){ref-type="fig"}). Comparison of the genomic region of this regulator-effector island is found in L. pneumophila and in the other eight *Legionella* species revealed that it contains highly variable genes ([Fig. S2A](#figS2){ref-type="supplementary-material"}), indicating that it is prone to changes in gene content. To examine whether this regulator-effector island undergoes HGT in the *Legionella* genus, we reconstructed the phylogenetic tree of these nine species based on the protein sequences encoded by the lpg0714-lpg0715 TCS and the lpg0716 effector and compared it to the phylogenetic tree reconstructed using a TCS (PmrAB) and a core effector (LegA3) present in all the *Legionella* species. While the analysis using PmrAB-LegA3 resulted in a tree structure similar to that of the known *Legionella* phylogenetic tree (as far as the characterized *Legionella* species are concerned \[[@B41]\]), the tree based on the regulator-effector island resulted in a different topology ([Fig. 1B](#fig1){ref-type="fig"}). Moreover, while the GC content of the *pmrAB-legA3* DNA sequence was similar to the genomic GC content in all nine species ([Fig. S2B](#figS2){ref-type="supplementary-material"}), the GC content of the regulator-effector island in three of the species (L. pneumophila, L. brunensis, and L. nautarum) was considerably lower, suggesting a recent event of HGT in these species. The genomic position, the GC content, and the tree structure of L. worsleiensis, L. moravica, L. quateirensis, and *Legionella* sp. strain Km535 imply a single HGT event, which occurred before their speciation. Collectively, these results suggest that this regulator-effector island undergoes HGT in the *Legionella* genus as a unit. 10.1128/mBio.03232-19.2 Regulator-effector island undergoes HGT in the *Legionella* genus. (A) Schematic presentation of the genomic region where the lpg0714-lpg0715-lpg0716 L. pneumophila orthologs are located in the six characterized *Legionella* species, L. pneumophila (Lpn), *L. moravica* (Lmor), *L. quateirensis* (Lqua), *L. worsleiensis* (Lwor), *L. nautarum* (Lnau), and *L. brunensis* (Lbru), and the three uncharacterized *Legionella* species, *Legionella* sp. strain 13.8642 (L_13.8642), *Legionella* sp. strain FW215 (L_FW215), and *Legionella* sp. strain Km535 (L_Km535), harboring the regulator-effector island. Homologous genes are indicated by the same color, and nonhomologous genes are white. The regions between lpg0713, encoding an oligopeptide transporter, and lpg0719, encoding a valyl-tRNA synthetase in the nine species, are presented. The genes are indicated by their locus tag number. (B) Comparison of the GC content of the lpg0714-lpg0715-lpg0716 genes and the *pmrA-pmrB-legA3* genes to the genomic GC content of the nine *Legionella* species harboring the regulator-effector island. Download FIG S2, TIF file, 1.5 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . The L. pneumophila lpg0714-lpg0715 TCS is homologous to TCSs involved in copper sensing in other bacteria. {#s2.2} ---------------------------------------------------------------------------------------------------------- Examination of the proteins encoded by lpg0714 and lpg0715 indicated that homologous TCSs are present in many bacteria. In most of them, the genes located next to it encode different proteins and systems involved in copper resistance ([Fig. 2](#fig2){ref-type="fig"}); however, cases of genes encoding resistance to silver and zinc were also described ([@B57], [@B58]). A few of these systems were studied before, including the CusRS TCS of Escherichia coli, which is located next to the CusCFBA copper transport system ([@B59], [@B60]). Examination of the lpg0716 EEG and of all other genes located next to the homologous TCSs from the different bacteria indicated that a conserved sequence is found at a precise distance from the predicted or validated −10 promoter elements in one or two of the surrounding genes ([Fig. 2](#fig2){ref-type="fig"} and [Fig. S3](#figS3){ref-type="supplementary-material"}). {ref-type="supplementary-material"}). The different designations of the homologous RRs and SHKs are indicated.](mBio.03232-19-f0002){#fig2} 10.1128/mBio.03232-19.3 Consensus regulatory element of genes found next to the LciRS homologous TCSs in different bacteria. The regulatory region of genes located near the genes encoding the LciRS homologous TCS in different bacteria is shown: L. pneumophila (Lpg), *L. moravica* (Lmor), *L. quateirensis* (Lqua), *L. worsleiensis* (Lwor), *L. nautarum* (Lnau), *L. brunensis* (Lbru), *Legionella* sp. strain 13.8642 (L_13.8642), *Legionella* sp. strain FW215 (L_FW215), *Legionella* sp. strain Km535 (L_Km535), Escherichia coli (Esc), Salmonella enterica (Sal), Enterobacter cloacae (Ent), *Achromobacter* species (Ach), Pseudomonas aeruginosa (Pa), Pseudomonas fluorescens (Pf), Pseudomonas putida (Pp), Pseudomonas stutzeri (Ps), and Bordetella pertussis (Bper). In purple are the putative LciR inverted-repeat regulatory elements, in light blue are the two additional regulatory elements found between the inverted-repeat and the −10 promoter element, in dark blue are the −10 promoter elements, and the experimentally determined transcription start sites (TSSs) are underlined and marked in blue. Download FIG S3, TIF file, 1.5 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . Due to the homology of this TCS and its predicted regulatory element to systems involved in copper sensing and detoxification, we named the L. pneumophila lpg0714-lpg0716 genes *lci*, for *Legionella* copper island. The SHK lpg0714 was named LciS, the RR lpg0715 was named LciR, and the effector lpg0716 was named LciE. *lciE* is induced by copper in L. pneumophila in an LciRS-dependent manner. {#s2.3} --------------------------------------------------------------------------- To determine whether *lciE* is induced in response to copper exposure, two fusions were constructed ([Fig. 3A](#fig3){ref-type="fig"}). The first fusion contains the 300-bp regulatory region of *lciE* fused to *lacZ* (designated *lciE-lacZ*). The second construct contains the same fusion as well as the *lciRS* genes in their original genomic organization (designated *lciRS-lciE-lacZ*) ([Fig. 3A](#fig3){ref-type="fig"}). To examine the expression of these constructs in response to copper exposure, we first determined the concentrations of copper that L. pneumophila can tolerate ([Fig. S4A](#figS4){ref-type="supplementary-material"}) and examined the effect of copper on the expression of *lciE* using a range of concentrations, from 1 to 50 μM. Using the *lciE-lacZ* fusion, the level of expression of *lciE* increased gradually as the concentration of copper was elevated, reaching more than 100-fold induction at the highest copper concentration ([Fig. 3B](#fig3){ref-type="fig"}). To determine the importance of the LciRS TCS for LciE expression, four deletion mutants were constructed, *lciR*, *lciS*, and *lciE* single deletion mutants and an *lciRS-lciE* triple deletion mutant. None of these mutants had an intracellular growth phenotype when examined in the amoeba host Acanthamoeba castellanii and in HL-60-derived human macrophages, as well as in a competition assay in amoeba with and without the addition of copper ([Fig. S5](#figS5){ref-type="supplementary-material"}). The copper induction of *lciE* was found to be completely dependent on the *lciR* and *lciS* genes, and as expected, the *lciE* deletion did not affect the levels of expression of the *lciE-lacZ* fusion ([Fig. 3B](#fig3){ref-type="fig"}). To further substantiate these results, copper induction was examined in an L. pneumophila strain deleted for the entire *lciRS-lciE* region ([Fig. 3C](#fig3){ref-type="fig"}). Using the *lciE-lacZ* fusion ([Fig. 3A](#fig3){ref-type="fig"}), no induction by copper was obtained, but when the *lciRS-lciE-lacZ* fusion ([Fig. 3A](#fig3){ref-type="fig"}) was examined, high levels of expression were obtained due to exposure to copper, which increased gradually as the concentrations of copper increased ([Fig. 3C](#fig3){ref-type="fig"}). In addition, the expression of *lciE* was found to be completely dependent on the presence of a functional LciRS TCS, since no induction was obtained when the conserved histidine of the LciS SHK and conserved aspartic acid of the LciR RR were mutated ([Fig. S6A](#figS6){ref-type="supplementary-material"}). Collectively, these results indicate that the expression of the Icm/Dot effector LciE is activated by copper, and its activation is completely dependent on the presence and functionality of the LciRS TCS. ![*lciE* is induced by copper in L. pneumophila and E. coli in a TCS-dependent manner. (A) Schematic representation of the L. pneumophila *lciRS-lciE* genomic region and the two *lacZ* fusions constructed (*lciE-lacZ* and *lciRS-lciE-lacZ*) to determine the expression of *lciE*. (B) The expression of the *lciE-lacZ* fusion was examined in the wild-type strain (white bars), in the *lciE* deletion mutant (gray bars), in the *lciR* deletion mutant (dotted bars), and in the *lciS* deletion mutant (hatched bars) at the stationary phase under different copper concentrations (indicated below the bars). (C) The expression of the *lciE-lacZ* fusion (white bars) and the *lciRS-lciE-lacZ* fusion (gray bars) was examined in the L. pneumophila triple *lciRS-lciE* deletion mutant at the stationary phase under different copper concentrations (indicated below the bars). (D) The expression of the *lciE-lacZ* fusion was examined in E. coli (white bars), in the E. coli *cusC* deletion mutant (gray bars), in the E. coli *cusR* deletion mutant (dotted bars), and in the *cusS* deletion mutant (hatched bars) at the stationary phase under different copper concentrations (indicated below the bars). (E) The expression of the *lciRS-lciE-lacZ* fusion was examined in the E. coli *cusR* deletion mutant at the stationary phase under different copper concentrations (indicated below the bars). The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student\'s *t* test) between expression of the same *lacZ* fusion under the same copper concentrations between the mutants and the one in the wild-type strain (panels B and D) or between the expression of the same strain grown without copper and the one grown under different copper concentrations (panels C and E). β-Galactosidase activity was measured as described in Materials and Methods. Data (expressed in Miller units \[M.U.\]) are the averages ± standard deviations (error bars) of the results from at least three different experiments.](mBio.03232-19-f0003){#fig3} 10.1128/mBio.03232-19.4 Effect of copper on L. pneumophila and E. coli growth in media. L. pneumophila (A) and E. coli (B) were grown with different concentrations of copper (indicated on the right), and their growth (OD~600~) was examined in intervals of 1 h for the time indicated. Download FIG S4, TIF file, 1.7 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . 10.1128/mBio.03232-19.5 LciRS TCS and the effector LciE are dispensable for intracellular growth. (A and B) The intracellular growth of the *lciR*, *lciS*, *lciE*, and *lciRS-lciE* deletion mutants was examined in A. castellanii (A) and HL-60-derived human macrophages (B). Symbols: diamonds, L. pneumophila wild-type strain JR32; squares, *icmT* mutant; triangles, *lciS* deletion mutant; black circles, *lciR* deletion mutant; star, *lciE* deletion mutant; white circles, *lciRS-lciE* triple deletion mutant. The experiments were performed three times, and similar results were obtained. (C and D) Intracellular competition assay between the L. pneumophila *lciRS-lciE* triple deletion mutant (white circles) and the JR32 wild-type strain (diamonds) in A. castellanii. The experiment was performed without (C) and with (D) 200 μM copper. Download FIG S5, TIF file, 0.8 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . 10.1128/mBio.03232-19.6 Effect of mutations and copper on the expression of *lciE* and *lciR*. (A) The effect of mutations on the amino acids predicted to be required for phosphorylation of LciR (LciR^D51A^) and LciS (LciS^H148A^) on *lciE* induction by copper. The levels of expression of wild-type *lciRS-lciE-lacZ* fusion and the *lciRS-lciE-lacZ* fusion containing a mutation in the amino acids, with and without 50 μM copper, are indicated below the bars. The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student's *t* test) between mutated fusions and the wild-type fusion after exposure to copper. (B) *lciR* expression is activated neither by copper nor by the LciR regulator. The expression of the *lciR-lacZ* fusion was examined in the wild-type strain and in the *lciR* deletion mutant with (grey bars) and without (white bars) the addition of 50 μM copper. Download FIG S6, TIF file, 0.7 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . *lciE* is induced in E. coli after copper exposure in a CusRS- or LciRS-dependent manner. {#s2.4} ----------------------------------------------------------------------------------------- As indicated above, the LciRS homologous TCSs as well as the LciR predicted regulatory element are conserved in many bacteria. To examine if these TCSs function similarly, we examined the expression of the *lciE-lacZ* construct ([Fig. 3A](#fig3){ref-type="fig"}) in E. coli. To this end, we determined the maximal copper concentration that E. coli tolerates without any effect on growth to be 1 mM ([Fig. S4B](#figS4){ref-type="supplementary-material"}), which is much higher than the one found for L. pneumophila ([Fig. S4A](#figS4){ref-type="supplementary-material"}). Analysis of the *lciE-lacZ* construct in E. coli resulted in gradual induction of *lciE* expression as the concentration of copper increased ([Fig. 3D](#fig3){ref-type="fig"}). Moreover, this induction was completely dependent on the E. coli CusRS TCS ([Fig. 3D](#fig3){ref-type="fig"}). Interestingly, when *lciE-lacZ* expression was examined in the *cusC* deletion mutant (CusC together with CusFBA form a copper transporter located in the E. coli envelope \[[@B59], [@B61]\]), higher levels of expression of *lciE* were observed at all copper concentrations examined ([Fig. 3D](#fig3){ref-type="fig"}). This result was obtained probably because E. coli lacking a functional CusCFBA system is exposed to higher concentrations of copper, which led to higher expression of *lciE-lacZ*. Comparing this result to the one obtained with the L. pneumophila *lciE* deletion mutant (compare [Fig. 3B](#fig3){ref-type="fig"} and [D](#fig3){ref-type="fig"}), which was induced similarly to wild-type L. pneumophila at all copper concentrations, indicates that in L. pneumophila, LciE is not involved in copper transport when the bacteria are grown in media, as expected from an Icm/Dot effector protein that functions during infection of host cells. Furthermore, the higher concentrations of copper tolerated by E. coli led us to examine the copper induction of the L. pneumophila LciRS TCS using E. coli as an *in vivo* heterologous system. Examination of the *lciRS-lciE-lacZ* construct in an E. coli *cusR* deletion mutant indicated that the L. pneumophila LciRS functions in E. coli and can respond to higher concentrations of copper than those tolerated by L. pneumophila ([Fig. 3E](#fig3){ref-type="fig"}). Collectively, these results indicate that the CusRS and LciRS TCSs function similarly, and both activate the expression of *lciE* in response to copper exposure. LciR recognizes a conserved regulatory element located upstream of *lciE*. {#s2.5} -------------------------------------------------------------------------- We identified a conserved regulatory element located upstream of *lciE* and other genes known or expected to be regulated by LciRS homologous TCSs ([Fig. 2](#fig2){ref-type="fig"} and [Fig. S3](#figS3){ref-type="supplementary-material"}). This regulatory element constitutes a 7-bp inverted repeat (ATTACAAnnTTGTAAT) as well as two shorter (4 bp each) conserved sequences located between the inverted repeat and the *lciE* −10 promoter element ([Fig. 4A](#fig4){ref-type="fig"} and [Fig. S3](#figS3){ref-type="supplementary-material"}). Examination of the L. pneumophila genomic sequence for the presence of additional such regulatory elements revealed that this site is not present elsewhere in the genome, which is in line with our observation that the LciRS-LciE genomic island was acquired by HGT. To determine the importance of each of the putative *lciE* regulatory elements for the copper induction by LciR, they were mutagenized and examined for their levels of expression before and after copper induction. Three of these mutated *lciE-lacZ* fusions completely lost their ability to be induced by copper, and the fourth mutant retained a very limited ability to respond to copper ([Fig. 4B](#fig4){ref-type="fig"}). ![LciR directly regulates the expression of *lciE* using a conserved regulatory element. (A) The intergenic DNA sequence located between *lciR* and *lciE*. The *lciR* and *lciE* −10 promoter elements are in dark blue, and the nucleotides representing the putative LciR consensus are in purple (the inverted-repeat sequence) or light blue (the two sequences located between the inverted repeat and the *lciE* −10 promoter); the inverted repeat is also marked with arrows. The transcription start sites are boldface and underlined. The four triplets of base pairs mutated in each part of the suspected regulatory element are indicated. (B) The effect of mutations in the putative LciR regulatory element. The levels of expression of wild-type *lciE-lacZ* fusion and the *lciE-lacZ* fusions containing mutations in the nucleotides marked in panel A were examined with and without 50 μM copper. The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student\'s *t* test) between the fusions containing the wild-type regulatory region and the mutated regulatory region under the same copper concentrations. β-Galactosidase activity was measured as described in Materials and Methods. Data (expressed in Miller units \[M.U.\]) are the averages ± standard deviations (error bars) of the results from at least three different experiments. (C) L. pneumophila His~6~-LciR protein binds to the *lciE* regulatory region. Gel mobility shift assay was performed with purified His~6~-LciR protein and the DIG-labeled *lciE* regulatory region. The first lane did not contain any protein. The rest of the lanes contained increasing amounts of the His~6~-LciR protein in 2-fold increments, starting from 35 nM. Competition was performed using unlabeled probe as a specific competitor (unlabeled WT) or a probe containing a mutation in the LciR regulatory element (unlabeled mut.).](mBio.03232-19-f0004){#fig4} Since the *lciR* gene that encodes the RR and the *lciE* EEG share an intergenic region ([Fig. 4A](#fig4){ref-type="fig"}), we examined whether the *lciR* gene is activated by copper and whether it is an autoregulator. The *lciR* gene was neither induced by copper nor affected by the *lciR* deletion mutant ([Fig. S6B](#figS6){ref-type="supplementary-material"}). These results indicate that the LciRS TCS specifically activates the expression of *lciE* in response to copper and suggest that the two conserved short regulatory elements located between the inverted repeat and the −10 promoter element of *lciE* play a critical role in directing the activation by LciR to *lciE* and not to *lciR*. The L. pneumophila LciR protein directly binds to the regulatory region of *lciE*. {#s2.6} ---------------------------------------------------------------------------------- To further support the results presented, the L. pneumophila LciR protein was His tagged, overexpressed, purified, and used for gel mobility shift assays with the *lciE* regulatory region. The L. pneumophila His~6~-LciR protein was found to bind to the regulatory region of the *lciE* gene, as evidenced by a shift in the migration of the DNA probe ([Fig. 4C](#fig4){ref-type="fig"}). The band shift degree as well as the amount of the shifted probe correlated with the increasing amounts of His~6~-LciR ([Fig. 4C](#fig4){ref-type="fig"}). In addition, competition with an unlabeled probe reduced the band shift ([Fig. 4C](#fig4){ref-type="fig"}, compare lanes 5 and 6). To further validate the specificity of the binding, we performed additional competition assays with an unlabeled probe containing a mutation in the LciR binding site. When the unlabeled mutated probe was used, a dramatic decrease in competition was observed compared to that of the unlabeled wild-type probe ([Fig. 4C](#fig4){ref-type="fig"}, compare lanes 6 and 7). The mobility shift assay ([Fig. 4C](#fig4){ref-type="fig"}), together with the examination of *lciE* gene expression in the *lciR* and *lciS* deletion mutants ([Fig. 3](#fig3){ref-type="fig"}) and the analysis of the mutations in the LciR consensus sequence ([Fig. 4B](#fig4){ref-type="fig"}), establish LciR as a direct regulator of the *lciE* EEG in L. pneumophila. Identification of amino acids required for LciS copper sensing. {#s2.7} --------------------------------------------------------------- Even though the E. coli CusRS and the L. pneumophila LciRS TCSs are homologous and both respond to elevated copper concentrations ([Fig. 3](#fig3){ref-type="fig"}), the periplasmic sensing domain of CusS and LciS are considerably different in size (150 amino acids in E. coli compared to 28 amino acids in L. pneumophila) and show no sequence homology ([Fig. 5A](#fig5){ref-type="fig"}). To identify amino acids required for copper induction in the L. pneumophila LciS periplasmic sensor domain, we aligned the sequences of LciS from the nine *Legionella* species harboring it. Since the amino acids histidine, cysteine, and methionine were previously shown to be involved in copper binding ([@B62], [@B63]), we specifically looked for these residues. Several histidine residues and a single cysteine were found in the short LciS periplasmic domain, together with other conserved amino acids ([Fig. 5A](#fig5){ref-type="fig"}). To examine the importance of these conserved amino acids for copper sensing by LciS, seven amino acid residues in the periplasmic domain of LciS (H43, H44, H46, N48, E50, H51, and C60) were changed separately to alanine residues. These mutations were introduced into the *lciRS-lciE-lacZ* fusion and used to determine the levels of copper induction ([Fig. 5B](#fig5){ref-type="fig"}). The most striking result was obtained with the mutation of the cysteine at position 60, which completely abolished the induction by copper. In addition, two of the histidine residues mutated (H44 and H51) significantly reduced induction by copper. We noticed that the number of histidine residues varies between the periplasmic sensor domains of the different *Legionella* LciS proteins, but at least two histidine residues were present in each LciS periplasmic domain. Thus, we constructed a triple mutation in which the three adjacent histidine residues (H43, H44, and H46) all were changed to alanine residues. This combined mutation resulted in a complete lack of induction by copper ([Fig. 5B](#fig5){ref-type="fig"}). To determine the specificity of LciS, we also exposed the bacteria to other similar metals ([@B57], [@B64]), and as seen in [Fig. 5C](#fig5){ref-type="fig"}, the LciRS TCS was found to be specific to copper and no induction was obtained with any of the other metals examined. Collectively, these results indicate that the *Legionella* LciS small periplasmic domain contains few histidine residues and a single cysteine residue specifically required for copper sensing. ![Identification of amino acids required for copper induction located in the LciS periplasmic sensor domain. (A) Schematic illustration of the L. pneumophila *lciRS* and the E. coli *cusRS* genes. The periplasmic domain of the nine *Legionella* SHKs is much smaller than the periplasmic domain found in other bacteria, such as E. coli (28 amino acids compared to about 150 amino acids, respectively). In addition, protein sequence alignment of the LciS transmembrane regions and periplasmic domains from nine *Legionella* species harboring the LciRS TCS is shown. The amino acids that compose the SHK predicted transmembrane domains are colored yellow. Amino acids found in the periplasmic domain which might be involved in copper-binding are marked by different colors: histidine residues, green; cysteine, light blue; and asparagine and glutamic acid, red. (B) The effect of mutations in the periplasmic domain of the L. pneumophila LciS on copper induction. The levels of expression of wild-type *lciRS-lciE-lacZ* fusion and the *lciRS-lciE-lacZ* fusions containing mutations in the amino acids indicated below the bars, with and without 50 μM copper. The levels of expression of the *lacZ* fusions were found to be significantly different (\*\*, *P \< *10^−4^; \*, *P \< *10^−5^; both by paired Student\'s *t* test) between the mutated fusions and the wild-type fusion with or without copper. (C) The effect of different metals on *lciE* expression. The levels of expression of wild-type *lciRS-lciE-lacZ* fusion after exposure to the metals indicated below the bars (50 μM). The level of expression of the *lacZ* fusion was found to be significantly different (*P \< *10^−5^, paired Student\'s *t* test) between expression with metals and the one without metal exposure. β-Galactosidase activity was measured as described in Materials and Methods. Data (expressed in Miller units \[M.U.\]) are the averages ± standard deviations (error bars) of the results from at least three different experiments.](mBio.03232-19-f0005){#fig5} The LciRS-LciE genomic island is regulated by the Fis repressors. {#s2.8} ----------------------------------------------------------------- It was previously shown in S. enterica as well as in other bacteria that genomic islands that undergo HGT are silenced by nucleoid-associated proteins (NAPs), such as H-NS and Hha ([@B65][@B66][@B67]). Since the LciRS-LciE island undergoes HGT in the *Legionella* genus ([Fig. 1](#fig1){ref-type="fig"} and [Fig. S1](#figS1){ref-type="supplementary-material"} and [S2](#figS2){ref-type="supplementary-material"}), we were interested in examining whether this island is silenced by NAPs. The H-NS NAP is not present in *Legionella*, but three Fis paralogs (Fis1, Fis2, and Fis3), which are also NAPs, were previously shown to directly repress the expression of EEGs ([@B32]). Examination of the *lciRS-lciE* intergenic region led to the identification of four potential Fis regulatory elements (TG-N~13~-C), two close to or overlapping the −10 promoter element of *lciE* ([Fig. S7](#figS7){ref-type="supplementary-material"}) and two others close to or overlapping the −10 promoter element of *lciR* ([Fig. 6A](#fig6){ref-type="fig"} and [Fig. S7](#figS7){ref-type="supplementary-material"}). To determine whether the Fis repressors are involved in the regulation of the LciRS-LciE island, the expression of the *lciE-lacZ* fusion was examined in deletion mutants of each of the three *fis* genes. In the absence of copper, the expression levels of the *lciE-lacZ* fusion in the three *fis* deletion mutants were similar to those in the wild-type strain ([Fig. 6B](#fig6){ref-type="fig"}). However, in the presence of copper, the expression of the *lciE-lacZ* fusion was significantly higher in the *fis1* and *fis3* deletion mutants ([Fig. 6B](#fig6){ref-type="fig"}). In addition, when the expression of the *lciR-lacZ* fusion was examined in the same mutants under the same conditions, there was an increase in the level of expression of the *lciR-lacZ* fusion in the *fis1* deletion mutant that was independent of the presence of copper ([Fig. 6C](#fig6){ref-type="fig"}). These results indicate a direct repression of Fis on *lciE*, on *lciR*, or on both genes. ![*lciRS-lciE* island is repressed by Fis. (A) The intergenic DNA sequence located between *lciE* and *lciR*. The *lciR* and *lciE* −10 promoter elements are in dark blue, and the nucleotides representing the LciR consensus are in purple (the inverted-repeat sequence) or light blue (the two sequences located between the inverted repeat and the −10 promoter of *lciE*); the inverted repeat is also marked with arrows. The transcription start sites are boldface and underlined. The putative *lciR* Fis regulatory elements are shaded in yellow, conserved nucleotides of the Fis consensus are marked in red, and the nucleotides mutated are marked by asterisks. (B and C) The expression levels of the *lciE-lacZ* fusion (B) and the *lciR-lacZ* fusion (C) were examined in the wild-type strain and in the three *fis* deletion mutants at the stationary phase. Expression was examined with (gray bars) and without (white bars) 50 μM copper. The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student\'s *t* test) between expression of the wild-type strain and each *fis* deletion mutant under the same conditions. (D and E) The levels of expression of wild-type *lciE-lacZ* fusion and the two *lciE-lacZ* fusions containing mutations (mut-1 and mut-2) in the putative Fis regulatory elements (D) and wild-type *lciR-lacZ* fusion and the three *lciR-lacZ* fusions containing mutations (mut-1, mut-2, and mut-1 + 2) in the putative Fis regulatory elements (E) were examined with (gray bars) and without (white bars) 50 μM copper. The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student\'s *t* test) between fusions containing the wild-type regulatory region and the mutated regulatory region under the same copper concentrations. (F) The levels of expression of wild-type *lciRS-lciE-lacZ* fusion (white bars) and the same fusion containing a mutation in the downstream Fis regulatory element of *lciR* (hatched gray bars) were examined in the *lciRS-lciE* deletion mutant without copper and with 5 μM and 50 μM copper. The levels of expression of the *lacZ* fusions were found to be significantly different (\*, *P \< *10^−5^, paired Student\'s *t* test) between the wild-type fusion and the mutated fusion under the same copper concentrations. β-Galactosidase activity was measured as described in Materials and Methods. Data (expressed in Miller units \[M.U.\]) are the averages ± standard deviations (error bars) of the results from at least three different experiments.](mBio.03232-19-f0006){#fig6} 10.1128/mBio.03232-19.7 Predicted Fis regulatory elements in the regulatory regions of *lciE* and *lciR* homologs in different *Legionella* species. The promoter region of the *lciE* (A) and *lciR* (B) genes in the nine *Legionella* species harboring the island, L. pneumophila (*lciE* and *lciR*), *L. moravica* (Lmor), *L. quateirensis* (Lqua), *L. worsleiensis* (Lwor), *L. nautarum* (Lnau), *L. brunensis* (Lbru), *Legionella* sp. strain 13.8642 (L_13.8642), *Legionella* sp. strain FW215 (L_FW215), and *Legionella* sp. strain Km535 (L_Km535). In purple are the putative LciR inverted-repeat regulatory elements, in light blue are the two additional regulatory elements found between the inverted repeat and the −10 promoter element of *lciE*, in dark blue are the −10 promoter elements, and the experimentally determined transcription start sites (TSSs) are underlined and marked in blue. The predicted Fis sites in both promoter regions are marked: the site overlapping or close to the −10 promoter element is colored yellow, and a second predicted Fis site (when present) is underlined. In both predicted Fis sites, the conserved TG and CA consensus nucleotides are in red and the nucleotides mutated are marked by asterisks. Download FIG S7, TIF file, 1.8 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . Fis1 and Fis3 repress the expression of *lciR* and affect the copper induction of *lciE*. {#s2.9} ----------------------------------------------------------------------------------------- To distinguish between the three possibilities described above, we constructed site-directed mutations in the four described putative Fis regulatory elements ([Fig. 6A](#fig6){ref-type="fig"} and [Fig. S7](#figS7){ref-type="supplementary-material"}). The two mutations constructed in the putative Fis regulatory elements of *lciE* did not affect its level of expression, with or without copper ([Fig. 6D](#fig6){ref-type="fig"}). However, both mutations constructed in the putative Fis regulatory element of *lciR* ([Fig. 6A](#fig6){ref-type="fig"}) showed a relief of repression ([Fig. 6E](#fig6){ref-type="fig"}). The mutation in the Fis regulatory element located immediately downstream to the −10 promoter element of *lciR* ([Fig. 6A](#fig6){ref-type="fig"}) showed a 6-fold relief of repression, and the mutation in the upstream Fis site showed 2-fold relief of repression. A combined mutation in both Fis regulatory elements led to a level of expression similar to that of the mutation in the downstream Fis site ([Fig. 6E](#fig6){ref-type="fig"}). It is important to note that the Fis regulatory element located downstream to the *lciR* −10 promoter element is conserved in all nine *Legionella* species harboring the LciRS-LciE genomic island, while the three other putative Fis regulatory elements examined are not conserved ([Fig. S7](#figS7){ref-type="supplementary-material"}). The effect observed on the level of expression of the *lciR-lacZ* fusion containing the mutations in the Fis regulatory elements was considerably stronger than the effect observed on the wild-type *lciR-lacZ* fusion in the *fis1* deletion mutant. Since a double deletion of both *fis1* and *fis3* is not viable in L. pneumophila ([@B32]), only the mutation in the Fis regulatory element can fully expose the degree of repression mediated by both Fis1 and Fis3 on the expression of LciR. To determine the effect of the Fis repression of LciR on the expression of LciE, we used the *lciRS-lciE-lacZ* fusion to construct a single-base-pair mutation in the downstream Fis regulatory element of LciR and examined the expression of LciE from this fusion. As can be seen in [Fig. 6F](#fig6){ref-type="fig"}, the level of expression of the *lciRS-lciE-lacZ* fusion containing the mutation in the *lciR* regulatory element was much higher in the absence of copper (6-fold) as well as in the presence of copper compared to that of the wild-type fusion. Collectively, these results indicate that the repression mediated by Fis on the LciR regulator affects the level of expression of the LciE effector with and without copper, showing that Fis proteins silence this genomic island by repressing the expression of the positive regulator LciR. DISCUSSION {#s3} ========== The L. pneumophila Icm/Dot secretion system translocates into host cells the largest number of effectors known in a single bacterium, and these effectors manipulate numerous host cell processes for the benefit of the bacteria ([@B10][@B11][@B14]). One of the challenges encountered by bacteria using such a multicomponent pathogenesis system is the coordination of the expression of the pathogenesis genes encoding these components to result in a successful infection. Thus far, several regulatory systems that control the expression of EEGs were identified in L. pneumophila ([Fig. 7](#fig7){ref-type="fig"}), including a pair of TCSs (PmrAB and CpxRA) that belong to the wHTH family of transcriptional regulators ([@B68]). Here, we described a third L. pneumophila TCS (LciRS) that belongs to the wHTH family of RRs, which directly regulates the expression of a single EEG ([Fig. 7](#fig7){ref-type="fig"}). In contrast to PmrA and CpxR, LciR is present only in several *Legionella* species and undergoes HGT as part of a genomic island together with the single EEG it regulates, representing the first case of local regulation of Icm/Dot effectors in *Legionella*. {#fig7} One of the most interesting findings regarding the LciRS-LciE genomic island is that it undergoes HGT in the *Legionella* genus ([Fig. 1](#fig1){ref-type="fig"}; see also [Fig. S1](#figS1){ref-type="supplementary-material"} and [S2](#figS2){ref-type="supplementary-material"} in the supplemental material). Two aspects regarding this genomic island were left unresolved, namely, (i) the way by which it is transferred between *Legionella* species and (ii) the way by which it integrates into the bacterial genome. A hint regarding these two issues was obtained when we analyzed the LciRS-LciE genomic island found in two uncharacterized *Legionella* species (*Legionella* sp. strain 13.8642 and *Legionella* sp. strain FW215). In *Legionella* sp. strain 13.8642, a pseudogene was found to be located next to the LciRS-LciE genomic island ([Fig. 1](#fig1){ref-type="fig"}). This pseudogene contains a deletion of a single nucleotide after nucleotide 41, making the protein it used to encode nonfunctional. However, prior to its pseudogenization, this gene encoded a protein with a high degree of homology to a phage integrase. Homologous integrase-encoding genes were found intact in several *Legionella* species (but not next to the LciRS-LciE genomic island) and in *Legionella* sp. strain 13.8642 it is completely intact, excluding the single-nucleotide deletion described above. It is possible that this integrase was involved in the integration of the LciRS-LciE genomic island in this *Legionella* species. Integrase-encoding genes, as well as mutated integrase-encoding genes, were previously shown to be located near genomic islands in other bacteria ([@B69], [@B70]). Another known feature of genomic islands is that they sometimes integrate into bacterial genomes next to tRNA genes ([@B50], [@B71]). In this case too, in one of the uncharacterized *Legionella* species, *Legionella* sp. strain FW215, the LciRS-LciE genomic island was found in proximity to a Lys tRNA gene ([Fig. 1](#fig1){ref-type="fig"}). This indicates that tRNA genes are also an entry site for the LciRS-LciE genomic island in *Legionella*. Although the precise mechanism by which the LciRS-LciE genomic island undergoes HGT in bacteria is not known, the above-mentioned findings suggest that it utilizes transfer and integration mechanisms similar to the ones previously described for genomic islands in other bacteria. The LciRS TCS was found to be activated in the presence of copper ([Fig. 3](#fig3){ref-type="fig"}), and homologous TCSs present in other bacteria are also activated by copper and are usually located next to genes involved in metal resistance (mainly copper) ([@B59]). Comparison of the *Legionella* LciRS TCS to the homolog TCSs present in other bacteria resulted in the identification of a major difference: the *Legionella* SHK LciS was found to contain a small (28 amino acids long) ([Fig. 5A](#fig5){ref-type="fig"}) periplasmic sensor domain, which is completely nonhomologous to the periplasmic sensing domains found in the homologous SHKs present in other bacteria, including the ones that sense copper. This finding is intriguing, since SHKs were previously shown to change their sensing domain by recombination or mutations and, in this way, change the signal they respond to and deviate from other SHKs ([@B72]). However, in the case of LciS, the sensing domain was completely changed but the signal recognized by the SHK remained the same. One can speculate on the evolutionary driving forces that can lead a sensing domain to be replaced without changing the signal it responds to. For example, an advantage can be achieved if the new sensing domain alters the sensitivity or specificity to the signal. However, examination of these two aspects, by comparing the E. coli CusS and L. pneumophila LciS SHKs, did not result in differences in the sensitivity to copper (compare [Fig. 3B](#fig3){ref-type="fig"} and [D](#fig3){ref-type="fig"}) or in the specificity to copper ([Fig. 5C](#fig5){ref-type="fig"} and data not shown). Even though the LciS periplasmic sensor domain is not homologous to other sensing domains that sense copper, it is important to mention that we did recognize specific amino acids critical for the sensing of copper by the L. pneumophila LciS, and the same amino acids were previously shown to be involved in copper binding in different proteins ([@B62], [@B63]). Genes regulated by regulatory systems that sense a specific signal were shown to encode proteins with functions directly related to the signal sensed by their regulators. This phenomenon was demonstrated in many systems, including regulators of amino acids biosynthetic pathways and regulators of metal resistance systems ([@B59], [@B73]). In addition, the L. pneumophila *mavN* EEG, which encodes an iron transporter localized to the LCV, is regulated by the iron-specific repressor Fur ([@B33], [@B34]). However, in many cases the signal sensed by a regulator is not directly related to the function mediated by the genes it regulates. In these cases, the signal sensed indicates a change in the bacterial environment. Such cases were described in many pathogenesis systems, such as the S. enterica PhoPQ TCS, which senses Mg^2+^ and cationic antimicrobial peptides ([@B74], [@B75]), and the S. enterica PmrAB TCS, which senses acidic pH and high Fe^3+^ concentrations ([@B76], [@B77]). Both of these TCSs regulate the expression of numerous genes encoding diverse functions ([@B78], [@B79]). The function of LciE, which is activated by the LciRS TCS in response to copper, is currently unknown, and it contains no known protein domains. However, we identified two other L. pneumophila effectors with unknown functions that show a significant degree of homology to LciE ([Fig. S8](#figS8){ref-type="supplementary-material"}). The core effector CetLp1 (lpg0140), and another effector (lpg2888) that is found in most of the *Legionella* species examined, harbor a protein domain homologous to LciE ([Fig. S8](#figS8){ref-type="supplementary-material"}). In addition, these three effectors contain four predicted transmembrane domains located at the C-terminal half of the protein ([Fig. S8](#figS8){ref-type="supplementary-material"}). These similarities among the three effectors suggest that they perform a related function. Since both copper and zinc are used by eukaryotic cells to kill invading bacteria ([@B80], [@B81]), we examined the copper- or zinc-dependent induction of CetLp1 and lpg2888, or an effect of LciR on their levels of expression, but their expression was unaffected ([Fig. S9A](#figS9){ref-type="supplementary-material"} and [B](#figS9){ref-type="supplementary-material"}). Moreover, we generated a triple deletion mutant of *cetLp1*, lpg288, and *lciE* and examined this mutant using a competition assay in amoeba with and without the addition of copper, and no intracellular growth phenotype was observed ([Fig. S9C to](#figS9){ref-type="supplementary-material"} F). 10.1128/mBio.03232-19.8 L. pneumophila contains two effectors with a domain homologous to LciE. Schematic representation of the effectors LciE (lpg0716), CetLp1 (lpg0140), and lpg2888. The domain homologous among the three effectors is marked in light blue, and putative transmembrane domains are marked in black. In addition, protein sequence alignment of the homologous domain found in the three effectors is shown from the six characterized *Legionella* species in which the LciRS-LciE genomic island was found, L. pneumophila (LciE/lpg2888/CetLp1), *L. moravica* (Lmor), *L. quateirensis* (Lqua), *L. worsleiensis* (Lwor), *L. nautarum* (Lnau), and *L. brunensis* (Lbru). Download FIG S8, TIF file, 2.1 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . 10.1128/mBio.03232-19.9 Examination of the connection of CetLp1 and lpg2888 to copper and LciE. The levels of expression of *cetLp1* (A) and lpg2888 (B) *lacZ* fusions were examined without the addition of metals and after exposure to 50 μM copper (grey bars) or zinc (hatched bars). The same fusions were also examined in the *lciR* deletion mutant under the same conditions. (C and D) Intracellular competition assay between the L. pneumophila *cetLp1*-lpg2888 double deletion mutant (circles) and the JR32 wild-type strain (diamonds) in A. castellanii. (E and F) Intracellular competition assay between the L. pneumophila *cetLp1*-lpg2888-*lciE* triple deletion mutant (circles) and the JR32 wild-type strain (diamonds) in A. castellanii. The experiment was performed without (C and E) and with (D and F) 200 μM copper. Download FIG S9, TIF file, 1.5 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . Unlike all the regulatory systems described thus far for L. pneumophila effectors, the LciRS-LciE genomic island represents a new type of effector regulation, unprecedented in the *Legionella* genus. In this system, a single effector is locally regulated by a dedicated regulatory system in response to a specific signal. Moreover, both the regulatory system and the effector form a unit that undergoes HGT in the *Legionella* genus. Uncovering this novel type of regulation sheds new light on the ways the effector repertoire of *Legionella* evolves and the activating signals of effectors expand. MATERIALS AND METHODS {#s4} ===================== Bacteria strains and media. {#s4.1} --------------------------- The L. pneumophila wild-type strain used in this work was JR32, a streptomycin-resistant, restriction-negative mutant of L. pneumophila Philadelphia-1, which is a wild-type strain in terms of intracellular growth ([@B82]). In addition, mutant strains derived from JR32 that were used in this study are listed in [Data Set S1](#dataS1){ref-type="supplementary-material"} in the supplemental material. The E. coli strains used in this work are also listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}. Bacterial media, plates, and antibiotics were as previously described ([@B83]). 10.1128/mBio.03232-19.10 Strains, plasmids, and primers used in this study. Download Data Set S1, XLSX file, 0.03 MB. Copyright © 2020 Linsky et al. 2020 Linsky et al. This content is distributed under the terms of the Creative Commons Attribution 4.0 International license . Plasmid construction. {#s4.2} --------------------- To construct *lacZ* translational fusions ([Data Set S1](#dataS1){ref-type="supplementary-material"}), the 300-bp regulatory regions of the *lciE* and *lciR* genes were amplified by PCR using the primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}. The PCR products were then digested with BamHI and EcoRI, cloned into pGS-lac-02, and sequenced. The *lciE lacZ* fusion, which does not contain the *lciRS* genes, was designated *lciE-lacZ*. In addition, a second *lciE lacZ* fusion was constructed containing the *lciRS* genes in an organization similar to that in the genome. To construct this fusion, an internal EcoRI site present in the *lciR* gene was mutagenized in a way that does not change the LciR amino acid sequence. A 1,950-bp region was amplified by PCR using the four primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}, digested with BamHI and EcoRI cloned into pGS-lac-02, and sequenced. This *lciE lacZ* fusion, which contains the *lciRS* genes, was designated *lciRS-lciE-lacZ*. To construct a substitution mutation in the putative LciR binding site in the regulatory region of the *lciE* gene, a substitution mutation in the putative Fis binding site in the regulatory region of the *lciE* and *lciR* genes, and substitution mutations in the *lciR* and *lciS* coding sequence, site-directed mutagenesis was performed by regular PCR or the PCR overlap extension approach ([@B84]), as previously described ([@B19]). The primers used for the mutagenesis are listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}, and the plasmids resulting from the site-directed mutagenesis are listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}. To construct deletion substitution mutants in the L. pneumophila *lciE* and *lciS* genes, a 1-kb DNA fragment located on each side of the planned deletion was amplified by PCR using the primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}. The resulting plasmids were digested with the suitable enzymes, and the inserts were used for a four-way ligation containing the Km resistance cassette (Pharmacia). The plasmids generated, pAA-lpg0714-Km and pMLpUC18+0716Up-Km-Dw ([Data Set S1](#dataS1){ref-type="supplementary-material"}), were digested with PvuII, and the resulting fragment was cloned into the pLAW344 allelic exchange vector digested with EcoRV to generate the plasmids pAA-lpg0714-pLAW and pMLpLAW344-0716-Up-Km-Dw ([Data Set S1](#dataS1){ref-type="supplementary-material"}). In addition, the insert of pMLpUC18+0716Up-Km-Dw was also cloned into the pGY100 allelic exchange vector, digested with XmnI, to generate the plasmid pMLpGY100+lpg0716-Up-Km-Dw, which was later digested with SalI to take out the Km resistance cassette and generate the plasmid pMLpGY100+lpg0176-UP+Dw ([Data Set S1](#dataS1){ref-type="supplementary-material"}). The latter plasmid was used to generate a clean *lciE* deletion. To generate the triple *lciRS-lciE* deletion mutant, the insert containing the 1-kb upstream region of *lciS* and the insert containing the 1-kb downstream region of *lciE* were used for a four-way ligation, as described above, to generated plasmid pMLpUC18+0714Dw-Km-0716Dw ([Data Set S1](#dataS1){ref-type="supplementary-material"}). This plasmid was digested with PvuII, and the resulting fragment was cloned into the pLAW344 allelic exchange vector digested with EcoRV to generate the plasmid pMLpLAW344-0714Dw-Km-0716Dw ([Data Set S1](#dataS1){ref-type="supplementary-material"}). To generate a clean deletion mutant in lpg2888, a 1-kb DNA fragment located on each side of the planned deletion was amplified by PCR using the primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}. The resulting plasmids were digested with the suitable enzymes, and the inserts were used for a three-way ligation into pUC-18 to generate pMW-18-Δlpg2888-3W ([Data Set S1](#dataS1){ref-type="supplementary-material"}). This plasmid was digested with PvuII, and the resulting fragment was cloned into the pGY100 allelic exchange vector digested with XmnI to generate the plasmid pMW-100-Δlpg2888 ([Data Set S1](#dataS1){ref-type="supplementary-material"}). The clean and marked allelic exchange deletion mutants were constructed as previously described ([@B83], [@B85]). For the construction of the plasmid expressing the His-tagged LciR, the *lciR* gene was amplified by PCR using the primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"}, cloned into pET-15b, and sequenced to generate the plasmid pML-pET15b+lpg0715 ([Data Set S1](#dataS1){ref-type="supplementary-material"}). Bacterial growth in the presence of copper. {#s4.3} ------------------------------------------- To determine the copper concentrations to be used in the β-galactosidase assays with both L. pneumophila and E. coli, the bacteria were grown in fresh AYE lacking Fe(NO~3~)~3~ or LB, respectively, with a wide range of copper concentrations, and the optical density at 600 nm (OD~600~) was determined in intervals of 1 h until reaching stationary phase. The same analysis was also performed with the other metals examined. β-Galactosidase assay. {#s4.4} ---------------------- β-Galactosidase assays were performed as previously described ([@B19]). L. pneumophila strains were grown for 48 h on charcoal-yeast extract (CYE) plates containing chloramphenicol (Cm). The bacteria were scraped off the plate and suspended in ACES-yeast extract (AYE) broth, and the bacterial OD~600~ was calibrated to 0.1 in fresh AYE lacking Fe(NO~3~)~3~, containing different concentrations of copper (or other metals, when indicated) and Cm. When other metals were used, 2 μM bathocuproine sulfonate (BCS) was added to the medium to adsorb any copper traces present in the stocks of the other metals. This concentration of BCS was also included when copper was used, and it did not affect the induction by copper. The resulting cultures were grown on a roller drum for about 18 h, until reaching an OD~600~ of about 3.2 (early stationary phase), and used for β-galactosidase assay. β-Galactosidase assays in E. coli were performed similarly, but the E. coli strains were grown for about 6 h in LB containing different concentrations of copper, until reaching an OD~600~ of about 2.5 (early stationary phase), and used for β-galactosidase assay. The assays were done for 20, 50, or 100 μl of culture, and the substrate for β-galactosidase hydrolysis was *o*-nitrophenyl-β-[d]{.smallcaps}-galactopyranoside. Protein purification and gel mobility shift assay. {#s4.5} -------------------------------------------------- His~6~-LciR was purified from E. coli BL21(DE3) using nickel bead columns (Qiagen) according to the manufacturer's instructions. After purification, the fractions containing the protein were dialyzed overnight against a buffer containing 10 mM Tris-HCl (pH 7.5), 5 mM MgCl~2~, 50 mM KCl, 0.1 mM EDTA, and 0.1 mM dithiothreitol. Glycerol was added to a concentration of 50%, and the purified protein was then stored at --20°C. A gel mobility shift assay was performed as previously described ([@B32]), with a few modifications. The regulatory region *lciE* (176 bp) was amplified by PCR using the primers listed in [Data Set S1](#dataS1){ref-type="supplementary-material"} and 3′ end labeled with digoxigenin (DIG) by using DIG-11-ddUTP (Roche). Increasing amounts of the purified His~6~-LciR protein (between 35 and 280 nM) were mixed with 0.75 nM the DIG-labeled probe in buffer containing 10 mM Tris-HCl (pH 7.5), 5 mM MgCl~2~, 50 mM KCl, 0.1 mM EDTA, 0.1 mM dithiothreitol, 250 μg/ml bovine serum albumin, and 50 μg/ml herring sperm DNA. For the competition experiments, a 100-fold excess of the unlabeled probe or mutated unlabeled probe was allowed to bind the His~6~-LciR protein for 30 min before addition of the DIG-labeled probe. The binding reaction was carried out for 30 min at room temperature, and samples then were loaded onto a 5% polyacrylamide--0.25× Tris-acetate-EDTA gel in 0.5× Tris-acetate-EDTA running buffer. Following electrophoresis, the gel was transferred to a nylon membrane and fixed by UV cross-linking. The DIG-labeled DNA fragments were detected by following the manufacturer's instructions (Roche). Intracellular growth assays. {#s4.6} ---------------------------- Intracellular growth assays of L. pneumophila strains in A. castellanii and HL-60-derived human macrophages were performed as previously described ([@B86]). Intracellular competition assays of L. pneumophila strains in A. castellanii also were performed as previously described ([@B22]). Reconstruction of phylogenetic trees. {#s4.7} ------------------------------------- Trees were reconstructed on the basis of concatenated alignments of the three proteins indicated for each tree. The trees were reconstructed using RAxML ([@B87]) under the LG + GAMMA evolutionary model with 100 bootstrap resampling. This article is a direct contribution from Gil Segal, a Fellow of the American Academy of Microbiology, who arranged for and secured reviews by Ralph Isberg, Tufts Medical School, and Craig Roy, Yale University School of Medicine. **Citation** Linsky M, Vitkin Y, Segal G. 2020. A novel *Legionella* genomic island encodes a copper-responsive regulatory system and a single Icm/Dot effector protein transcriptionally activated by copper. mBio 11:e03232-19. <https://doi.org/10.1128/mBio.03232-19>. We thank Yaron S. Feldheim for plasmid construction and for reading the manuscript. We thank Ziv Lifshitz and Michael Wexler for plasmids and strains construction. We thank Anna Pasechnek for plasmid construction. We thank Tal Zusman for her ideas and help throughout this study and for reading the manuscript. We thank David Burstein for his help with the bioinformatic analyses and for reading the manuscript. This research was supported by Israeli Science Foundation grant 877/15 (to G.S.).

INTRODUCTION ============ The approval of imatinib (IM) as a targeted therapy for chronic myeloid leukemia (CML) by the Federal Drug Agency (FDA) in 2001 has opened a new era in the treatment of this disease and has considerably improved patients\' survival [@b1-cln_70p322][@b2-cln_70p322][@b3-cln_70p322][@b4-cln_70p322][@b5-cln_70p322]. In Brazil, the Ministry of Health initially approved IM for CML in the chronic phase as a second-line treatment and in 2008 as front-line therapy [@b6-cln_70p322][@b7-cln_70p322]. The drug is provided for free to all Brazilian residents through the public health program, Sistema Único de Saúde (SUS). Additionally, recently published national guidelines regulate the cytogenetic and molecular monitoring of patients [@b7-cln_70p322]. In the first Brazilian publication on the results of using IM, presented by a center from São Paulo participating in international cooperative studies [@b8-cln_70p322], 68.8% of the patients starting treatment in the early chronic phase achieved a complete cytogenetic response (CCyR) within 12 months. However, it might be expected that the large social and educational inequalities of the Brazilian population could influence the treatment outcome. Thus, we evaluated socioeconomic factors influencing the treatment success of patients with CML treated in the state of Piauí, located in the northeast region of Brazil. PATIENTS AND METHODS ==================== All consecutive patients diagnosed with CML in the chronic phase treated at the São Marcos Hospital in Teresina, Piauí, between February 2004 and February 2013 were included in the study. This hospital is the only reference center for CML treatment in the state of Piauí and also treats some patients from the neighboring state of Maranhão. Diagnosis was based on clinical data, peripheral blood counts and cytogenetics or on the detection of the BCR-ABL1 rearrangement by multiplex PCR. Cytogenetics and molecular tests were performed at the central laboratory of the National Institute of Cancer (INCA) in Rio de Janeiro. Patients' risk categories were assessed by the Sokal, Hasford and EUTOS scores [@b9-cln_70p322][@b10-cln_70p322][@b11-cln_70p322]. Patients diagnosed before 2008 and who were treated first with hydroxyurea or interferon but were intolerant to these medications and were thus switched to imatinib were also included in the study. The diagnosis of chronic or accelerated phase disease and blast crisis was performed according to the WHO 2008 criteria [@b12-cln_70p322]. Patients were considered to be in the early chronic phase when imatinib therapy was started before 12 months following diagnosis. For each patient, we collected the geographic distance from the patient\'s home to the hospital (in kilometers), the years spent in school, the average monthly income of each family member and the time interval between diagnosis and the start of imatinib treatment. The per capita income was also grouped into the income classes established by the Brazilian government (Secretariat of Strategic Affairs). All patients received 400 mg imatinib daily. Previous treatments included hydroxyurea or interferon. All patents provided informed consent for their participation in the study, which was approved by the local Ethics Committee (proc. no. 045 - 2010 - Federal University of Piauí). Laboratory monitoring comprised monthly peripheral blood counts performed by local laboratories. Karyotyping was scheduled to be performed at 3, 6 and 12 months; however, in some cases, karyotyping was only possible with some delay. After achieving a complete cytogenetic response, molecular monitoring (quantitative PCR) was performed every 3 months. The response criteria were evaluated according to the 2009 European LeukemiaNet (ELN) guidelines [@b13-cln_70p322]. All time-to-event analyses were made with the log-rank test or Cox regressions. All time-to-event calculations were based on the interval between the beginning of imatinib treatment and the complete cytogenetic response. Cases were censored when there was no response after the third karyotyping (approximately 12 months) or when the patients changed therapy or dropped out of the study before the end of the study. Non-normally distributed variables were normalized by logarithmic transformation. All variables presenting *p*\<0.15 were included in the multivariate analysis. A bootstrap resampling procedure was performed to test the stability of the model, as previously described [@b14-cln_70p322][@b15-cln_70p322][@b16-cln_70p322][@b17-cln_70p322]. Winstat 3.1 and SPSS 15.0 software were used. RESULTS ======= A total of 103 patients with CML in the chronic phase entered this study. Their descriptive data are summarized in [Table 1](#t1-cln_70p322){ref-type="table"}. Among the patients, 101 were Ph+ (had a Philadelphia chromosome) in the cytogenetic examination; the remaining two had normal karyotypes and were BCR-ABL1 positive. All patients had received hydroxyurea for cytoreduction until the result of karyotyping was received, and 35 had also received interferon. Regarding the delay before imatinib treatment, 69 patients were in the early chronic phase, and the remaining 34 started imatinib more than one year following diagnosis. The median time of imatinib treatment in the patients was 45 months (range 7-141). At the end of the observation period, 70 patients (68%) remained on this tyrosine kinase inhibitor (TKI). Imatinib was discontinued due to a loss of response in 28 cases and disease progression in 5 cases. One patient received an allogeneic bone marrow transplantation, and two were lost to follow-up. Among the 68 patients who had achieved a CCR within approximately one year, only 8 later became resistant or progressed. According to the income classes, 12 patients earned between 0 and 81 reais, 17 between 82-162 reais, 38 between 163-441 reais, 18 between 292-441 reais, 9 between 442-641 and 9 had an income \>642 reais. The median time of formal education of the patients was 4 years ([Table 1](#t1-cln_70p322){ref-type="table"}). Eleven patients had never gone to school or had attended school for only one year, 67 patients had spent 2-9 years in school, 17 had spent between 10-12 years in school, and only 8 had some university education (\>13 years of education). There was a positive correlation (r=0.20; *p*=0.012) between the patients\' income and the years they had spent in school ([Figure 1](#f1-cln_70p322){ref-type="fig"}). However, the distribution of the patients according to the risk groups of any of the scores assessed was not affected by their income classes or years of formal education. Most patients lived far from the healthcare facilities ([Table 1](#t1-cln_70p322){ref-type="table"}). In the Kaplan-Meier plot, the median time to remission for patients with 1 year or less spent in school was 404 days. The median time to remission was 369 days for both the groups with 2 to 9 years of education (basic level) and with 10 to 12 years of education (secondary level) and was 252 days for patients with 13 or more years of education (university level) (log-rank test *p*=0.006) ([Figure 2](#f2-cln_70p322){ref-type="fig"}). In the Cox regression analysis, the Sokal score was of no prognostic relevance, in contrast with the Hasford (*p=*0.03) and EUTOS (*p=*0.08) scores. The delay between diagnosis and start of the imatinib therapy was an unfavorable predictive factor for complete cytogenetic remission (B=-0.001; *p=*0.02), and age was a weak risk factor (B=-0.013; *p*=0.117). Concerning the demographic and socioeconomic variables, only the educational level had a significant influence on the time to acquire a complete cytogenetic response (B=0.065; *p=*0.008). Gender, income and the distance from the home to the health facility did not affect the cytogenetic response. In the multivariate Cox regression analysis, considering age, Hasford and EUTOS scores, years spent in school and the delay to begin IM treatment, only the Hasford score (B=-0.531; *p=*0.038) and years spent in school remained as independent predictive factors (B=0.374; *p=*0.019). In the bootstrap stability test, "Hasford score" was present in 71% of the new sets, "years in school" was present in 62%, "delay to begin imatinib" appeared in 32%, and "EUTOS score" appeared in 18%. DISCUSSION ========== The introduction of targeted therapy with TKIs has substantially improved the life expectancy and quality of life of patients with CML. However, it has repeatedly been observed that treatment results in daily praxis do not reproduce those of clinical trials [@b2-cln_70p322][@b3-cln_70p322]. The reasons for this observation have been a matter of debate, but more recent data have shown that the results in population-based patient cohorts may be equally good, provided that drug access is unrestricted [@b3-cln_70p322]. Since 2008, the Brazilian public healthcare system has provided IM as the first-line treatment for CML for every resident in the country. The main question of our study was to investigate whether the significant social inequalities in the population could be of major clinical importance for the success of long-term TKI treatment for CML. Notably, in our study, the socioeconomic or educational level was not related to the stage of disease at diagnosis. Among our patients, the only demographic factor influencing the optimal cytogenetic response according to the ELN criteria was the education level, whereas personal income and the distance between home and the health care facility did not. In a population with unrestricted access to medication, patients with a better education level may better understand the disease mechanism and the importance of long-term treatment, as has been previously noted [@b4-cln_70p322][@b5-cln_70p322][@b19-cln_70p322]. Social and educational inequalities have also been shown to influence breast cancer mortality [@b20-cln_70p322] as well as the normal development of low birth weight children [@b21-cln_70p322]. The fact that a family\'s earnings do not affect therapy is surprising at first glance. The literature is contradictory in this regard [@b3-cln_70p322][@b5-cln_70p322][@b22-cln_70p322][@b23-cln_70p322][@b24-cln_70p322]. On the one hand, familial resources have been reported to have a significant influence on the outcome of acute lymphoid leukemia (ALL) in both developing and wealthier countries [@b22-cln_70p322][@b24-cln_70p322]. Patient malnourishment and lack of family support have also been implicated in a patient's incapacity to maintain long-term treatment. On the other hand, family earnings did not affect the outcomes of children with ALL in Norway [@b23-cln_70p322] and the United Kingdom [@b24-cln_70p322], where the public health system provides state of the art treatment for all patients. Nevertheless, several publications have called attention to the fact that better educated patients and those with good family support have much better outcomes after cancer treatment, especially with oral anti-cancer agents, in European countries [@b5-cln_70p322][@b23-cln_70p322]. Highly educated patients read more, better understand the nature of their disease, and are more perseverant and more motivated to adhere to long-lasting treatments, regularly using the prescribed drugs and even arriving at the hospital on the appropriate day [@b4-cln_70p322][@b5-cln_70p322][@b19-cln_70p322][@b23-cln_70p322]. This was fully confirmed in the present study, in which the years spent in school was a strong predictive factor for an optimal cytogenetic response. According to the Kaplan-Meier curve, it was evident that many patients with only one year of formal education delayed the third cytogenetic control. However, even after eliminating these patients from the survival curve, the log rank test still indicated significant differences. As the Kaplan-Meier curve suggests, patients with at least some university education show the best therapy success. A similar finding was reported from Norway, where the outcomes in children with cancer were better when the mothers were highly educated [@b23-cln_70p322]. Additionally, in a former work, performed in a more affluent region of our country [@b19-cln_70p322], the time of TKI treatment, participation in clinical trial, higher quality of life scores and socioeconomic statuses, measured by the patients' income and education levels, respectively, influenced IM treatment compliance. Although we have found a significant correlation between income and education level, only the years of formal education influenced achievement of CCyR within one year of IM treatment. This finding is in agreement with the hypothesis that in a setting where all patients have equal access to oral anticancer therapy, the patients' capacity to understand the importance of their treatment is essential for good compliance. It is even more important for the long-term maintenance of therapy. Thus, to optimize the invested public resources in drugs and laboratory support, broad additional healthcare information for patients is necessary. AUTHOR CONTRIBUTIONS ==================== Rego MN was the postgraduate student who performed her work (her doctor thesis) in the context of the Interinstitutional Postgraduating program from CAPES. She was in charge of patient care, developed the Care Unit, performed the data collection and analysis and participated in the manuscript writing. Metze K participated in the study design, performed the statistics and critically reviewed the manuscript. Lorand-Metze I was the supervising doctor who participated in the study design and reviewed the data collection, data analysis and manuscript. No potential conflict of interest was reported. {#f1-cln_70p322} {#f2-cln_70p322} ###### Features of the patients at diagnosis. Variable Value ----------------------------------------------------------------------------------------------- ------------------------ Gender **(for all 103 cases)** Male 56 Female 47 Age in years (median, range) 42 (8-81) Sokal score (100 cases) Low risk 30 Intermediate risk 42 High risk 28 Hasford score (100 cases) Low risk 67 Intermediate risk 27 High risk 6 EUTOS score (99 cases) Low risk **92** Intermediate risk **7** Treatment before imatinib (no. Patients) Hydroxyurea 103 Interferon-α 34 Median interval diagnosis - start imatinib (days) 99 (27-1780) For patients in early chronic phase (days) 60 For patients in late chronic phase (days) 963 Years in school 4 (0-17) Median income per family member and month (in R\$)[\*](#tfn1-cln_70p322){ref-type="table-fn"} 226.00 (9.00-3,000.00) Distance from the home to the hospital (in km) 150 (1-794) These values correspond to a mean of 1 Real=0.5 US\$ during the study period

Introduction ============ In 1989, a new hypothalamic hormone with the significant ability to activate adenylate cyclase in rat pituitary cell cultures was discovered and aptly named pituitary adenylate cyclase-activating polypeptide or PACAP \[[@B1],[@B2]\]. As reported in those first papers, PACAP exists in two amidated forms, PACAP38 and its truncated form, PACAP27. The ovine PACAP38 is 176 amino acids long, proceeded by a putative signal peptide and a proregion (107 amino acids), and followed by a Gly-Arg-Arg sequence for proteolytic processing and C-terminal amidation; the human PACAP38 has been shown to be identical to the isolated ovine PACAP \[[@B3]\]. The sequence of PACAP38 encompasses an internal cleavage-amidation site (Gly28-Lys29-Arg30), suggesting that it can generate a 27 residue-amidated polypeptide fragment or PACAP27. Similarly, the rat (PACAP precursor that is highly similar to the ovine and human PACAP precursors) and the mouse (PACAP precursor shows 81 to 93% sequence similarity) PACAP cDNAs were cloned (reviewed in \[[@B4]\]). To briefly note some characteristics, PACAP is a member of the secretin, glucagon, and vasoactive intestinal polypeptide (VIP) family showing the most homology to VIP; all peptides are present in the tissue but PACAP38 is the most dominant form \[[@B5]\]. An important finding was on the potency of adenylate cyclase activation by PACAP 1,000 to 10,000 times greater than VIP, as demonstrated in pituitary cell, neuron, and astrocyte cultures. PACAP and its receptors (PAC1, VPAC1, and VPAC2) are widely distributed in the brain (central nervous system, CNS) and peripheral organs (notably endocrine pancreas, gonads, respiratory, and urogenital tracts). Consequently, PACAP exerts pleiotropic effects including control of neurotransmitter release, vasodilation, bronchodilation, activation of intestinal motility, increase in insulin and histamine secretion, immune modulation, and stimulation of cell proliferation and/or differentiation. These and other characteristics and functions of PACAP are comprehensively reviewed by Arimura \[[@B4]\] and Vaudry *et al*. \[[@B6]\]. PACAP is now considered a potent neurotrophic and neuroprotective peptide \[[@B7]-[@B11]\]. The specific role of PACAP in neuroprotection is of interest to our study, and forms the basis for the present experiments with an aim to unravel new potential targets of PACAP at the gene and/or protein level, and possible mechanisms involved therein. PACAP exerts potent neuroprotective effects not only *in vitro* but also in *in vivo* models of Parkinson\'s disease, Huntington\'s disease, traumatic brain injury, and stroke. The neuroprotective effects of PACAP are based on its capacity to prevent neuronal apoptosis by acting directly on neurons or indirectly through the release of neuroprotective factors by astrocytes \[[@B12],[@B13]\]. These biological activities are mainly mediated through activation of the PAC1 receptor which is currently considered a potential target for the treatment of neurodegenerative diseases. However, the use of native PACAP, the endogenous ligand of PAC1, as an efficient neuroprotective drug is actually limited by its rapid degradation. Moreover, injection of PACAP in humans induces peripheral side effects that are mainly mediated through VPAC1 and VPAC2 receptors \[[@B7]\]. The intraventricular infusion of PACAP38 at 1 pmol/h significantly delayed neuronal cell death in the hippocampal CA1 region that would normally occur a couple of days after induction of ischemia. The intravenous administration of PACAP (only 16 pmol/h) decreased delayed neuronal cell death in the hippocampus. Subsequently, another study showed that PACAP quickly crosses the blood--brain barrier at approximately 0.12% by a saturable system. PACAP prevented delayed neuronal cell death even when it was given intravenously 24 h after ischemia induction. Moreover, anti-apoptotic factors and anti-cell-death pathways were increased by PACAP and decreased by the addition of the PAC1R antagonist, PACAP 6--38, after ischemia and in other *in vitro* studies. The studies suggest that the neuroprotection by PACAP depends on the activation of the PAC1R, and exogenous PACAP might extend the therapeutic time window for treatment of ischemia-related conditions, such as stroke \[[@B10]\]. In an effort to unravel PACAP molecular targets in the brain, especially under ischemia, we further advanced our research using the established permanent middle cerebral artery occlusion (hereafter referred to as PMCAO) model mice and the optimized DNA microarray approach with the 44K mouse whole genome chip \[[@B14]\]. Our study employed the intraluminal filament technique PMCAO model because of its simplicity and noninvasive characteristics compared to the classical electrocoagulation method for PMCAO. The transient MCAO results in reperfusion injury, and to avoid this complication, the method of choice was in favor of intraluminal PMCAO. Further, we also applied the proteomics approach \[[@B15],[@B16]\] to examine the protein changes in the same sample/brain hemisphere as was used for the transcriptomics study. We used DNA microarray analysis to answer two questions. 1) What are the PACAP influenced genes? 2) Can we pinpoint the transcripts specifically altered (increased or decreased) by PACAP, that is, the genes that are not the result of ischemia itself, and are predominantly regulated by PACAP? By proteomics, we aimed to identify proteins from the two-dimensional gel profile that was differentially regulated by PACAP under the ischemic condition. Our results reveal numerous potential gene candidates, such as interleukin family members, *Gabra6*, and *Crtam* that might be involved in PACAP-mediated neuroprotective mechanisms. The identification of an important protein involved in neuronal function, the collapsin response mediator protein (CRMP2) \[[@B17]\], was another key finding of a PACAP-regulated protein in the ischemic hemisphere. Results presented show the usefulness of omics approaches in screening of potential targets of PACAP-regulated genes and proteins. Methods ======= Animals and husbandry --------------------- Animal care and experimental procedures were used as approved by the Institutional Animal Care and Use Committee of Showa University (School of Medicine), Tokyo, Japan. Thirty male mice (C57BL/6J), 9-weeks-old, body weight 25 to 35 g, were purchased from Charles River (Kanagawa, Japan). Mice were housed at the Animal Institution in Showa University in acrylic cages (eight mice/cage) maintained at 23°C with a standard 12-h light/dark cycle, optimum humidity, and temperature control. Animals were given access to tap water and laboratory chow *ad libitum*. Permanent middle cerebral artery occlusion (PMCAO), PACAP treatment, and SHAM control ------------------------------------------------------------------------------------- The experimental design is presented in Figure [1](#F1){ref-type="fig"}. The PMCAO-model mice were generated as described previously \[[@B14]\]. Briefly, the mice were anesthetized with 4% sevoflurane (induction) and 2% sevoflurane (maintenance) in a 30% O~2~ and 70% N~2~O gas mixture via a face mask. An incision was then made in the cervical skin followed by opening of the salivary gland, and the right common carotid artery was visualized. A midline cervical incision was made to expose the external carotid artery. The intraluminal filament technique was used, as reported by Hori *et al*. \[[@B14]\] to generate the PMCAO model. The PACAP38 (1 μL containing 1 pmol) or 1 μL of saline (0.9% NaCl) was injected intracerebroventrically, immediately after PMCAO. PACAP38 (Peptide Institute Inc., Osaka, Japan; supplier temperature was −20°C) was dissolved at 10^-5^M concentration by saline, and stored at −80°C. PACAP test solution (for injection) was diluted × 10 times with 0.9% NaCl just before use. In the sham control animals, the wound was sutured following exposure of the external carotid artery. The PACAP38 or saline was injected intracerebroventrically in the same concentrations as above. After injection, the animals were returned to their cages. A total of eight groups were prepared: four groups of three, seven, four and five mice in the PMCAO plus PACAP38 and PMCAO plus saline cohorts at 6 and 24 h after the operation, respectively, and five mice each in the control (sham) plus PACAP and saline groups at 6 and 24 h after operation, respectively (see Additional file [1](#S1){ref-type="supplementary-material"}: Table S1). We used three mice each in PMCAO groups that exhibited neurological grades G1 and G2 \[[@B13],[@B14]\], and three mice each at random in sham groups for the subsequent downstream analysis. Some of the mice were examined for ischemia by TTC (2, 3, 5-triphenyltetrazolium chloride) staining of brain sections (2-mm slices) at 37°C for 10 minutes \[[@B13],[@B14],[@B18]\]. {#F1} Dissection of brain, sampling and storage, extraction of total RNA, cDNA synthesis, and semiquantitative RT-PCR --------------------------------------------------------------------------------------------------------------- Six or 24 hours post-injection of PACAP38 or saline, the mice were removed from their cages, decapitated, and their brains carefully removed on ice. The left (contralateral) and right (ipsilateral; ischemic) hemispheres were dissected and placed in 2-mL Eppendorf tubes, which were then quickly immersed in liquid nitrogen before being stored at −80°C prior to further analysis (Figure [1](#F1){ref-type="fig"}). Stored tissues were ground to a very fine powder with liquid nitrogen, and used for total RNA extraction, followed by a quantity and quality check \[[@B14]-[@B16]\]. To validate the total RNA quality and subsequently synthesized cDNA, RT-PCR was carried out using two commonly used housekeeping genes *glyceraldehyde 3-phosphate dehydrogenase* (*GAPDH*) and *beta-actin* as positive controls \[[@B14],[@B19]\]. The 3'-UTR gene-specific primers were designed (see Additional file [2](#S2){ref-type="supplementary-material"}: Table S2). The cDNA synthesis and RT-PCR analysis protocol used is as follows: Total RNA samples were first DNase-treated with RNase-free DNase (Stratagene, Agilent Technologies, La Jolla, CA, USA). First-strand cDNA was then synthesized in a 20 μL reaction mixture with an AffinityScript QPCR cDNA Synthesis Kit (Stratagene) according to the protocol provided by the manufacturer, using 1 μg total RNA. The reaction conditions were: 25°C for 5 minutes, 42°C for 5 minutes, 55°C for 40 minutes and 95°C for 5 minutes. The synthesized cDNA was made up to a volume of 50 μL with sterile water supplied in the kit. The reaction mixture contained 0.6 μL of the first-strand cDNA, 7 pmol of each primer set and 6.0 μL of the Emerald Amp PCR Master Mix (2X premix) (TaKaRa Shuzo, Shiga, Japan) in a total volume of 12 μL. Thermal-cycling (Applied Biosystems, Tokyo, Japan) parameters were as follows: after an initial denaturation at 97°C for 5 minutes, samples were subjected to a cycling regime of 20 to 40 cycles at 95°C for 45 s, 55°C for 45 s, and 72°C for 1 minute. At the end of the final cycle, an additional extension step was carried out for 10 minutes at 72°C. After completion of the PCR the total reaction mixture was spun down and mixed (3 μL), before being loaded into the wells of a 1.2/1.8% agarose fine powder (catalogue number 02468--95, Nacalai Tesque, Kyoto, Japan) gel. Electrophoresis was then performed for approximately 22 minutes at 100 Volts in 1X TAE buffer using a Mupid-ex electrophoresis system (ADVANCE, Tokyo, Japan). The gels were stained (8 μL of 10 mg/mL ethidium bromide in 200 mL 1X TAE buffer) for approximately 7 minutes and the stained bands were visualized with the ChemiDoc XRS+ imaging system (Bio-Rad, 6000 Alfred Nobel Drive, Hercules, CA 94547, USA). DNA microarray analysis in the ipsilateral (right) hemisphere ------------------------------------------------------------- A mouse 4 × 44K whole genome oligo DNA microarray chip (G4122F, Agilent Technologies, Palo Alto, CA, USA) was used for global gene expression analysis using the ipsilateral (ischemic) hemisphere. Briefly, total RNA (900 ng; 300 ng for each replicate pooled) was labeled with either Cy3 or Cy5 dye using an Agilent Low RNA Input Fluorescent Linear Amplification Kit (Agilent). Fluorescent-labeled targets of control (sham) as well as treated (PMCAO) samples with PACAP38 or without PACAP38 (saline) were hybridized to the same microarray slide with 60-mer probes. As illustrated in Figure [1](#F1){ref-type="fig"}, in this experiment we compared the PMCAO plus PACAP38-injected mice to PMCAO plus saline, that is, the ipsilateral brain region of the PMCAO mice was compared with the same right hemisphere of the control mice. Similarly, sham control plus PACAP38-injected mice to sham control plus saline were analyzed to identify only the PACAP38-influenced genes. A flip labeling (dye-swap or reverse labeling with Cy3 and Cy5 dyes) procedure was followed to nullify the dye bias associated with unequal incorporation of the two Cy dyes into cDNA \[[@B14],[@B20]\]; and references therein\]. Briefly, the same total RNA (900 ng) samples were labeled twice with Cy3 or Cy5: a Cy5-labeled treatment (T^Cy5^) and a Cy3-labeled control (C^Cy3^) was hybridized on a slide and then a Cy3-labeled treatment (T^Cy3^) and a Cy5-labeled control (C^Cy5^) were reversely hybridized on another to revise the dye bias associated with unequal incorporation of two Cy dyes into cRNA. Hybridization and wash processes were performed according to the manufacturer's instructions, and hybridized microarrays were scanned using an Agilent Microarray scanner G2565BA. For the detection of significantly differentially expressed genes between control and treated samples, each slide image was processed by Agilent Feature Extraction software (version 9.5.3.1). Signal ratios of each spot on all slides were normalized by the software and finally, the log ratio of each spot for technical replicates was averaged to give a representative value for each pooled sample and to adjust remaining gene-specific dye bias. The output of microarray analysis used in this study are available under the series number GSE 37565, at the NCBI Gene Expression Omnibus (GEO) public functional genomics data repository (<http://www.ncbi.nlm.nih.gov/geo/info/linking.html>). To validate the microarray data (ipsilateral hemisphere) RT-PCR, as described above, was also performed on randomly up- and down-regulated genes using 3'-UTR specific gene primers (see Additional file [2](#S2){ref-type="supplementary-material"}: Table S2). Extraction of total soluble protein ----------------------------------- Total protein was extracted from sample powders (around 50 mg) of the contralateral and ipsilateral hemispheres (control and treatment) using a previously used lysis buffer containing thiourea and Tris (LB-TT) for extraction of brain proteins \[[@B15],[@B16]\]. The composition of slightly modified LB-TT was as follows: 7 M (w/v) urea, 42 g; 2 M (w/v) thiourea, 15.2 g; 4% (w/v) CHAPS, 4.0 g; 18 mM (w/v) Tris--HCl (pH 8.0), 1.8 mL; 14 mM (w/v) Trizma base, 169.5 mg; 0.2% (v/v) Triton X-100, 0.2 mL; 50 mM (w/v) DTT, 771.5 mg; 1% (v/v) pH 3--10 Ampholyte, 1 mL; and two EDTA-free proteinase inhibitor (Roche Diagnostics GmbH, Mannheim, Germany) tablets in a total volume of 100 mL; see Additional file [3](#S3){ref-type="supplementary-material"}: Figure S1 for preparation of LB-TT at room temperature (RT). To extract protein, 1 mL of LB-TT was quickly added to the 2-mL microfuge tube containing the sample powder (immediately after removal from the −80°C deep freezer) and immediately mixed by vortexing (at full speed using a Lab mixer, Iwaki, Tokyo, Japan) for 1 minute at RT. The protein solution in LB-TT was incubated at RT for 30 minutes with mixing by vortexing (for 30 sec) and sonication (for 30 sec in a water bath-type sonicator) for a total of five times. The insoluble protein pellet and/or debris were pelleted by centrifugation at 18,500 g for 15 minutes at 20°C in a high-speed refrigerated micro centrifuge (MX-150, TOMY, Tokyo, Japan). The clear supernatant (around 900 μL) was transferred to a new 1.5-mL microfuge tube, and stored at −80°C as the total soluble protein. Prior to storage, a 0.2 mL aliquot of total protein solution was transferred to a new 2-mL microfuge tube and processed for a second purification, clean up, and concentration step, using the ProteoExtract Protein Precipitation Kit (Calbiochem, Darmstadt, Germany; catalague number 539180). This additional step (see Additional file [4](#S4){ref-type="supplementary-material"}: Figure S2) removes any impurities while concentrating the protein. The pelleted protein, visible as a white solid precipitate at the bottom of the tube, was re-suspended in LB-TT (around 0.2 mL) as above. Protein concentration was determined with a Coomassie Plus^TM^ Protein Assay Kit (PIERCE, Rockford, IL, USA) using bovine serum albumin (BSA) as a standard and a NanoDrop 2000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA). Two-dimensional gel electrophoresis and visualization of the separated proteins ------------------------------------------------------------------------------- The first-dimension separation was carried out using pre-cast IPG strip (pH 4 to 7, 18 cm; GE Healthcare Bio-Sciences AB, Uppsala, Sweden) gels on a CoolPhoreStar IPG-IEF Type-PX unit (Anatech, Tokyo, Japan) followed by the second dimension using hand-cast polyacrylamide gels on an Anatech CoolPhoreStar SDS-PAGE Dual-200 K unit (Anatech, Tokyo, Japan). Two-dimensional gel electrophoresis was performed as essentially described by Toda and Kimura \[[@B21]\]. The detailed protocol for the first- and second dimension-run is described in Additional file [5](#S5){ref-type="supplementary-material"}: Figure S3. After two-dimensional gel electrophoresis, the separated proteins were visualized by staining with Flamingo fluorescent gel stain (catalogue number 161--0491, Bio-Rad). Briefly, the gel was fixed in 200 mL (two-dimensional gel) of fixing solution (40% methanol and 10% acetic acid in MQ water) for 30 minutes at RT, followed by staining with 1X diluted (in MQ water) Flamingo stain (stock of 10X) for 1 h in the dark. The stained proteins were visualized after washing three times for 1 minute each in MQ water with a FluoroPhoreStar 3000 image analysis system (Anatech). The gel images were saved as 8-bit TIFF files for further analysis. Protein identification by matrix-assisted laser desorption/ionization-time of flight-mass spectrometry (MALDI-TOF-MS) --------------------------------------------------------------------------------------------------------------------- Protein spot was excised from the gel using a gel picker (1.8 mm diameter) in combination with a robotic spot analyzer (FluoroPhoreStar 3000, Anatech) and placed in a sterile 0.5 mL microfuge tube (Safe-lock tubes, Eppendorf AG, Hamburg, Germany). Excised protein spots can be stored at 4°C till further analysis. For enzymatic digestion prior to mass spectrometry (MS) analysis 100 μL of 50 mM ammonium bicarbonate (NH~4~HCO~3~) (w/v) solution in 50% acetonitrile (ACN) (v/v) was added to the excised spot in the tube, and rinsed for 10 minutes under constant shaking in a tube shaker at RT. Following spin-down of the washed gel piece, the NH~4~HCO~3~-ACN solution was removed and the gel was vacuum-dried for 30 minutes (Centrifugal vacuum/SpeedVAc evaporator, Eyela, Tokyo, Japan) at RT. To the dried gel pieces, 2 μl of trypsin solution (10 mM NH~4~HCO~3~ and 10 ng/μL of trypsin) was added, and the microfuge tube was kept on ice for 30 minutes, followed by incubation at 37°C for 4 h to overnight for digestion of the protein in gel. Tryptic peptides were extracted with 10 μL of 75% ACN containing 0.025% trifluoroacetic acid (TFA) (v/v) for 15 minutes by ultrasonication (water bath-type). Following a spin-down of any gel debris, the supernatant was analyzed by MS. The supernatant can also be stored at −40°C/-80°C for long-term (1 year) storage. For MALDI-TOF-MS analysis, we first prepared the matrix solution by dissolving α-Cyano-4-hydroxycinnamic acid (4-CHCA, Shimadzu Biotech, Kyoto, Japan) in 1:1 100% ACN and 0.1% TFA (5 mg of 4-CHCA is dissolved in 500 μL of ACN-TFA solution to give 10 μg/μL 4-CHCA. We next loaded 0.5 μL of the above matrix mixture onto any numbered spot area on the stainless steel MALDI sample plate and left to dry for 5 minutes at RT followed by addition of 1 μL of the peptide sample solution to the same spot, dried for a further 10 to 15 minutes. The ready sample plate was inserted into the vacuum slot of the AXIMA Performance MALDI-TOF-MS (Shimadzu Biotech, Kyoto, Japan) and peptides were identified by Peptide Mass Fingerprint (PMF) analysis. The parent ion masses were measured in the reflectron mode with an accelerating voltage of 24.4 kV. A two-point internal standard for calibration was used with an angiotensin 2 (human, A9525, Sigma; m/z 1046.54 Da) and ACTH fragment 18--39 (human, A0673, Sigma; *m/z* 2465.20 Da). Peptides were selected in the mass range of 900 to 5,000 Da. For data processing, we used MASCOT program (<http://www.matrixscience.com>). Searching conditions/parameters were fixed (Carbamidomethyl, C) and variable modifications (methionine, M), mass values (monoisotopic), peptide mass (unrestricted), peptide mass tolerance (± 0.3 Da), peptide charge state (1+), and maximum mixed cleavage (1). NCBInr \[20120113 (16852130 sequences; 5788275850 residues) (Taxonomy, *Mus musculus* (house mouse) 144612 sequences)\] and SWISSPROT \[2012_12 (533657 sequences; 189261966 residues) (Taxonomy, *Mus musculus* (house mouse) 16415 sequences)\] databases were used. Western blot analysis --------------------- The separated proteins after SDS-PAGE (mini-gel; Additional file [6](#S6){ref-type="supplementary-material"}: Figure S4) were transferred onto a polyvinyldifluoride (PVDF) (Trans-Blot Turbo Midi PVDF, 0.2 μM, Transfer Packs kit; catalogue number 170--4157). The Trans-Blot Turbo Transfer System (Bio-Rad) was used for the electrotransfer (StandardSD protocol; 25V, 1.0 A, 30 minutes). Following the transfer of proteins on the PVDF membrane (and also confirmed by visualizing all the 10 colored molecular mass standards), it was incubated in 25 mL of 5% blocking solution (Block-Ace powder, catalogue number UK-B80, DS Pharma, Osaka, Japan; Yukizurishi, Sapporo, Hokkaido, Japan) for 1 h under constant slow shaking at RT. Blocking solution was prepared by dissolving 4 g powder in 80 mL 1X TTBS \[10X TTBS: NaCl, 80 g; 1M Tris--HCl, pH 7.5, 200 mL; Tween-20, 5 mL\]. Western blotting and detection was carried out using the Immun-Star WesternC Chemiluminescent Kit (catalogue number 170--5070, Bio-Rad) following the manufacturer's instructions. Blocking solution was decanted and the membrane was washed once in 1X TTBS (5 minutes), followed by incubation in 25 mL of primary antibody solution (PAS) (1 μL rabbit anti-CRMP2 protein antibody; catalogue number ab62661; 100 μg, 2 mg/mL; Abcam, <http://www.abcam.co.jp>) for 1 h, as above. The membrane was then washed with 25 mL of 1X TTBS for five times. After decanting the last TTBS wash, the membrane was incubated in 25 mL of secondary antibody solution (SAS; 0.5 μL of Amersham, ECL anti-rabbit IgG, HRP linked species-specific whole antibody (from Donkey); catalogue number NA 934; GE Healthcare) for 1 h, with slow shaking at RT. The 1X TTBS wash step was repeated five times. For blot development, the luminol/enhancer and peroxide buffer solutions were mixed in a 1:1 ratio (1 mL:1 mL; one membrane volume) and spread over the membrane and incubated at RT for 5 minutes. Excess solution was drained by touching one end of the membrane on a KimWipe paper towel, and the signal (cross-reacting protein bands) was visualized by x-ray film (Kodak, Tokyo, Japan) and on a ChemiDoc XRS+ imaging system. The western blot analysis was repeated at least three times, and representative data from the image obtained using an X-ray film is shown. Immunofluorescent staining -------------------------- The brains were removed and immersed for 1 day in 0.1 M phosphate buffer (PB, pH 7.2) containing 2% paraformaldehyde and then for 2 days in 0.1 M PB containing 20% sucrose, at −80°C (see Additional file [7](#S7){ref-type="supplementary-material"}: Figure S5) \[[@B22],[@B23]\]. The brains were then embedded in the mixture of 20% sucrose in 0.1 M PB and Tissue Tek Optical Cutting Temperature (OCT) solution (2:1; Miles Inc., Elkhart, IN, USA), frozen on dry ice, and stored at −80°C until required. Sections, 8 μm thick, were subsequently cut with a cryostat (HYRAX C50, Microedge Instruments, Tokyo, Japan), and mounted onto gelatin-coated glass slides. Immunofluorescent staining was carried out as described previously \[[@B23],[@B24]\]. Prior to detection of the desired protein, the 8-μm frozen sections were immersed in 0.01 M PBS solution and washed for three times 5 minutes each wash, followed by blocking in 5% normal horse serum (in 0.01 M PBS). Sections were incubated overnight at 4°C in mixtures of primary antibodies, consisting of rabbit anti-CRMP2 antibody (1:2000) with mouse anti-NeuN antibody (1:400) for neuronal cell marker. Post-incubation, the sections were washed with 0.01 M PBS as above. Immunoreactivity for CRMP2 and NeuN was detected using an Alexa 488-labeled goat anti-mouse IgG and an Alexa 546-labeled goat anti-rabbit following 90-minute incubation at RT. After washing with 0.1 M PBS as above, the sections were incubated for 5 minutes with 4′, 6-diamidine-2-phenylindole dihydrochloride (DAPI), 1:10,000 (in 200 mL 0.01 M PBS, add 20 μL DAPI) (Roche Diagnostics, IN, USA) as a nuclear stain. Labeling was imaged with a fluorescence microscope (Axio Imager Z1 with Apotome, Zeiss, Germany). The CRMP2-positive cells were examined in three different regions: healthy, penumbra, and ischemic core in a coronal section of the ipsilateral hemisphere (see Additional file [8](#S8){ref-type="supplementary-material"}: Figure S6). To note, edema (swelling caused by a collection of fluid in the third space surrounding the tissue or organ) was discernible, mostly at the 24 h time period, and is therefore marked in the figure. To assess the specificity of the antibodies, negative control sections in which the primary antibody is absent were prepared to check for background staining levels. Immunostaining were performed on sections obtained from the brains of three mice in each group. Results and discussion ====================== Selection of PMCAO mice with neurological grades between 1 and 2 and preparation of brain tissues for grinding in liquid nitrogen --------------------------------------------------------------------------------------------------------------------------------- Mice exhibiting neurological deficiency 6 and 24 hours after PMCAO were graded according to a routinely used methodology in our laboratory \[[@B13],[@B18]\]. Sixteen mice that exhibited neurological grade (NG) 1 and 2 were used for further experiments; two mice died, and one had NG 0 (no PMCAO) and one had NG 3 (extreme neurological symptoms) (Additional file [1](#S1){ref-type="supplementary-material"}: Table S1). The selected mice were decapitated and the brains were dissected on ice. The ipsilateral (right hemisphere, non-injected) and contralateral (left hemisphere; injected with saline or PACAP38) hemispheres without the olfactory bulb (OB) and cerebellum were quickly separated, placed in 2-mL Eppendorf tubes, and deep frozen in liquid nitrogen followed by storage at −80°C. Brains were ground to a very fine powder in liquid nitrogen; aliquots of the finely powdered tissue samples were stored at −80°C, and used for extraction of total RNA or protein thus giving a comparative data expression analysis in the same sample. Overview of the brain genomic response to PACAP38 injection in the ischemia brain (ipsilateral hemisphere) by DNA microarray analysis ------------------------------------------------------------------------------------------------------------------------------------- ### DNA microarray analysis and confirmatory RT-PCR To investigate global changes in gene expression in the ischemic hemisphere, the previously optimized protocols for total RNA extraction, critical for any downstream analysis, and DNA microarray analysis in mice were followed (Ogawa *et al*., 2011; \[[@B14],[@B20]\]). The ischemic hemisphere (ipsilateral) consists of the infarct core, penumbra, and non-ischemic region (Additional file [8](#S8){ref-type="supplementary-material"}: Figure S6). Thus, the present experimental design used to perform the microarray analysis provides an overall picture of the ischemic hemisphere rather than one specific ischemic region. Nevertheless, and as will be discussed below, an additional experiment was carried out to check for the expression of proteins in each of the aforementioned regions in the ipsilateral hemisphere. Prior to DNA microarray analysis, the expression of *GAPDH* and *β-actin*, as positive controls, was confirmed at both hemispheres at 6 and 24 h in the PMCAO samples, with or without PACAP; saline was used as control. Results revealed the mRNAs for *GAPDH* and β-actin were expressed almost uniformly across the tested conditions (Figure [2](#F2){ref-type="fig"}). {ref-type="supplementary-material"}: Table S2.](1742-2094-9-256-2){#F2} DNA microarray analysis gave a wide change in gene expressions as anticipated, and for confirmation of alterations in gene expression by DNA microarray, we randomly selected 14 upregulated genes and 6 downregulated genes with annotated functions; one gene that was the most highly upregulated at 6 h post-ischemia, but with no known function, was also selected. Using RT-PCR, the mRNA expression profiles are presented in Figure [2](#F2){ref-type="fig"}, which validate the microarray experiment. Two trends can immediately be observed in the upregulated gene candidates examined, namely early (6 h) and late (24 h) expression of target genes by PACAP38 treatment. For example, the gamma-aminobutyric acid (GABA) A receptor, subunit alpha 6, *Gabra6*, was found to be strongly upregulated at 6 h in the PACAP-treated ischemic hemisphere but was not expressed at 24 h. This was confirmed by RT-PCR (Figure [2](#F2){ref-type="fig"}). This is the first report of *Gabra6* regulation by PACAP. Why PACAP affects *Gabra6* gene expression is unclear, but it may be related to GABA and glutamate levels, which, though not been investigated in this study, are of future interest for investigation. A gene with unknown function, *C030005H24Rik*, showed the most increased mRNA expression fold at 6h, and we analyzed its expression by design-specific primers based on the available gene sequence in the NCBI database. RT-PCR results revealed only a slight increase in mRNA abundance at 6 h (Figure [2](#F2){ref-type="fig"}). However, as this gene still remains unannotated and there are no functional domains identified so far, the meaning of this gene expression under ischemia remains a mystery. Interestingly, a gene neurofibromatosis 1 (*Nf1*), which is known to regulate PACAP-mediated signaling in astrocytes \[[@B25]\] was found to be slightly induced at 24 h. Therefore, it would be tempting to speculate a role for *Gabra6* and *Nf1* in PACAP-mediated neuroprotection. Similarly, *Il6*, *S100a5*, *Il22*, *Il1b*, *Igf1*, and *Ccl2* were highly expressed at 6 h in the PACAP-treated ischemic brain, whereas their expression level decreased at 24 h compared to the PACAP effect alone (Figure [2](#F2){ref-type="fig"}). On the other hand, *Fgf21*, *Pitpnc1*, and *Epha3* genes showed an increase in expression at 24 h (Figure [2](#F2){ref-type="fig"}). These results suggest a clear demarcation in the expression of genes to the administered PACAP38 with time, and that may be linked to the progression of ischemia itself. However, further detailed studies will be necessary to provide concrete evidence for the hypothesis. Here, we would like to emphasize an early first study on neuroprotection by endogenous and exogenous PACAP following stroke that was performed by the Lee E Eiden's group at the NIH (Bethesda, MD, USA). In that study the cerebrocortical transcriptional response in the MCAO model was studied in PACAP-deficient mice and in wild-type mice using a 36K cDNA microarray chip for transcriptome profiling of gene expression changes in the cortex at 1 and 24 h post-ischemia \[[@B8]\]. A large percentage of 142 up-regulated genes at 24 h were shown to require endogenous PACAP, and the authors suggest a more prominent role for PACAP in later response to injury than in the initial response \[[@B8]\]. Although 40 pmol of PACAP was used as an exogenous treatment, compared to the 1 pmol used in our present study, we looked at any similarities in gene expression changes with our present data at both 6 and 24 h post-ischemia (with or without PACAP). Only three upregulated genes and one downregulated gene were found to be common at 6 h, whereas two upregulated genes and twenty-four downregulated genes were found to be common at 24 h by Chen *et al*. \[[@B8]\]. The lack of similarity between the Chen et al. \[[@B8]\] study and ours may lie in 1) MCAO versus PMCAO and KO mice versus wild-type, 2) use of cDNA versus the whole genome microarray, and 3) different time points used (1 versus 6 h). Nevertheless, we believe that as the two studies have identified unique changes in gene expression, these data are complementary and represent the growing number of potential PACAP-regulated gene transcripts in the ischemic brain. ### Tabulation of the differentially expressed genes The next step was the analysis of the vast microarray data (gene lists). As we had previously reported an inventory of ischemia-related genes in the ipsilateral brain hemisphere \[[@B14]\], we wanted to know at first how many and what kind of genes PACAP alone influences in the brain (ipsilateral hemisphere). Using the processed data from the previous study \[[@B14]\] and this study on changes in gene expression \> 1.5-fold for upregulated and \< 0.75-fold for downregulated gene expression, at both 6- and 24-h time points, we were able to list the ischemia-related and PACAP-related genes. The first two columns of Table [1](#T1){ref-type="table"} show the results of the comparison, and it is immediately clear that PACAP influences the expression of a lower number of genes (263 to 743; PACAP-related) genome-wide over the large number of genes (622 to 2,759; ischemia-related) showing changed expression under ischemia alone. Interestingly, it was seen that the PACAP38 injection influenced a greater number of gene expressions early, that is, at 6 h post-ischemia, rather than at 24 h, suggesting that PACAP indeed has a role in regulating gene expression in the brain post-injection, a character of the neuropeptide. We next asked how many genes are influenced by PACAP under ischemia, and eliminating the ischemia alone-related genes, we could narrow down our inventory to a considerably lower number of genes (third column, Table [1](#T1){ref-type="table"}). But this is again a significantly high number of genes to discuss in single study. Nonetheless, all these genes expression data are open and freely available on the GEO website at NCBI (GSE 28201 \[[@B14]\]; GSE 37565, this study) for the scientific community. ###### Changes in gene expression in the ipsilateral hemisphere revealed by comparison of genes expressed under ischemia with or without pituitary adenylate cyclase-activating polypeptide (PACAP)-38 injection over saline control **Sampling time (expression change)** **Ischemia-related** **PACAP-related** **PACAP in ischemia-related** **Common: ischemia-related and PACAP in ischemia-related** **Common: PACAP-related and PACAP in ischemia-related** --------------------------------------- ---------------------- ------------------- ------------------------------- ------------------------------------------------------------ --------------------------------------------------------- 6 h (upregulated) 1,237 435 839 171 43 6 h (downregulated) 622 743 498 31 24 24 h (upregulated) 2,759 263 831 49 30 24 h (downregulated) 2,104 271 1468 28 75 Results are presented as number of genes. To further narrow down our search, we also looked at the common genes between two sets of data, that is, ischemia-related and PACAP in ischemia-related (fourth column, Table [1](#T1){ref-type="table"}) and PACAP-related and PACAP in ischemia-related (fifth column, Table [1](#T1){ref-type="table"}). The gene numbers decreased drastically to 171 and 31 (upregulated and downregulated at 6 h, respectively; Additional file [9](#S9){ref-type="supplementary-material"}: Table S3), and 49 and 28 (upregulated and downregulated at 24 h, respectively; Additional file [10](#S10){ref-type="supplementary-material"}: Table S4) for the predominantly ischemic group, and to 43 and 24 (upregulated and downregulated at 6 h, respectively; Additional file [11](#S11){ref-type="supplementary-material"}: Table S5), and 30 and 75 (upregulated and downregulated at 24 h, respectively; Additional file [12](#S12){ref-type="supplementary-material"}: Table S6) for the predominantly PACAP group. For details, see the genes listed in Additional files [9](#S9){ref-type="supplementary-material"}: Table S3, 10: Table S4, 11: Table S5, 12: Table S6. ### Functional categorization of differentially expressed genes and candidate genes regulated by PACAP38 with potential links to ischemic neuroprotection To overcome the challenge of explaining the large gene dataset, we also functionally categorized the obtained gene lists (Additional files [13](#S13){ref-type="supplementary-material"}: Table S7, [14](#S14){ref-type="supplementary-material"}: Table S8, [15](#S15){ref-type="supplementary-material"}: Table S9 [16](#S16){ref-type="supplementary-material"}: Table S10; 6 and 24 h up/downregulation) using the pathway- or specific disease states-focused gene classifications available on the QIAGEN website (SABiosciences; <http://www.sabiosciences.com>) in order to reveal the trend of predominant pathways affected by PACAP in the ischemic hemisphere at 6 and 24 h post-ischemia. Results show the distribution of genes among multiple categories (Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}). At 6 h after PACAP treatment, the number of upregulated gene expressions is evident in contrast to the large number of downregulated gene expression at 24 h. This indicates that PACAP influences the regulation of numerous gene pathways in the ischemic hemisphere quite early, and that might have a role in neuroprotection. Based on the listing of the gene candidates and their functional categorization, we discuss potential PACAP-regulated genes, including a novel gene in the context of a role for PACAP as a neuroprotective peptide in limiting ischemia. {#F3} {ref-type="fig"} in the SABiosciences PCR array list (QIAGEN; <http://www.sabiosciences.com>) for *Mus musculus*. The numbers in the y-axis represent the number of genes in each category, which are indicated on the x-axis.](1742-2094-9-256-4){#F4} ### PACAP influences early regulation/induction of interleukins The predominant gene with annotated function influenced by PACAP38 injection was interleukin 6 (*IL6*; 4.3-fold upregulated at 6 h; Additional file [9](#S9){ref-type="supplementary-material"}: Table S3) upon analyzing the genes with focus on the keywords ischemia and PACAP (fourth column, Table [1](#T1){ref-type="table"}). By RT-PCR, the *IL6* mRNA abundance was also found to be strongly increased under PACAP treatment at 6 h in the ischemic hemisphere over both the sham and PMCAO alone expression levels (Figure [2](#F2){ref-type="fig"}). On the other hand, when focusing on the genes under the keywords PACAP and ischemia (fifth column, Table [1](#T1){ref-type="table"}), no interleukins were found to be upregulated (Additional files [11](#S11){ref-type="supplementary-material"}: Table S5 and 12: Table S6). This is because in the ischemic hemisphere at 24 h, the PACAP plus PMCAO expression level was reduced compared to the sham plus PACAP control. This suggests that interleukin is only one target of PACAP action under ischemia, and PACAP might regulate different gene sets at different stages of ischemia progression. In other words, the influence of PACAP on interleukins should logically be early rather than late, to see the neuroprotective effects of PACAP38 administration to the brain. *IL6* is one of the most well-known and critical inflammatory cytokines in the ischemic brain \[[@B26]\]. Further, this result also re-confirms, and at the same time differentiates, the finding of our previous study on ischemic brain gene expression profiles \[[@B14]\], where *IL6* was induced to very high levels (10.35-fold and 89.23-fold upregulated at 6 and 24 hours, respectively). It has been reported that PACAP stimulates IL-6 expression from neurons and astrocytes after ischemia \[[@B12]\], and the neuroprotective effect of PACAP mediates the IL-6 pathway \[[@B13]\]. This evidence supports our findings of increased *IL6* expression in acute response after PACAP injection. Surprisingly, the *IL6* gene was not among the 49 common genes at 24 h post-treatment, suggesting that PACAP negatively influences *IL6* expression after acute response. *IL6* seems to respond and rebound after the effects of PACAP deficiency at 24 hours after injection. Using this analytical approach, we can begin to see a certain function of PACAP38 in regulating the expression of the IL inflammatory cytokines, namely, PACAP activates the differential and early induction of ILs during ischemia progression. Further, among the numerous cytokines found to be induced in the ischemic hemisphere \[[@B14]\], only *IL1b*, *IL8rb*, *ILl1rap*, and *IL11* were found to be upregulated at 6, but not at 24 h (Additional files [9](#S9){ref-type="supplementary-material"}: Table S3 and 10: Table S4). IL-11, a member of the IL-6 family of proinflammatory cytokines, can be a potential target for PACAP-mediated neuroprotection. IL-11 also has a role in the response to oxidative stress and compensatory proliferation \[[@B27]\], and has recently been proposed as a candidate cytokine clinically available for cardioprotection therapy \[[@B28],[@B29]\]. Another potential target is the IL-1 receptor accessory protein gene *IL1rap*. The IL-1RAP is a necessary part of the interleukin 1 receptor complex that initiates signaling events to activate the IL-1-responsive genes. Further, the *IL1rap* is also associated with IL-1 beta (IL-1b), and an early study had revealed that IL-1RAcP is necessary for centrally mediated neuroendocrine and immune responses to IL-1beta \[[@B30],[@B31]\]. These reports imply that IL-11 and IL-1 pathway could connect with PACAP-induced neuroprotection after brain ischemia. ### Novel induction of the cytotoxic and regulatory T cell molecule (Crtam) gene at 6 h We identified *Crtam*, also called the MHC class-1-restricted T-cell-associated molecule, as a 4.2-fold upregulated gene at only 6 h (Additional file [9](#S9){ref-type="supplementary-material"}: Table S3). To our knowledge, this is the first report of *Crtam* gene expression by PACAP. The *Crtam* gene was initially identified as a new member of the immunoglobulin superfamily (Ig-SF) from the molecular analysis of natural killer T (NKT) cells, by Kennedy *et al*. \[[@B32]\]. CRTAM is also expressed by other types of activated T cells, CD4^+^ and CD8^+^. Interestingly, CRTAM maintains the T cell's polarity, which regulates T cell activation and movement through the lymphatic system and tissues \[[@B33]\]. Moreover, CRTAM regulates IFN and IL-22 production \[[@B34]\]. In our RT-PCR results, the *IL22* mRNA expression was also strongly upregulated in PACAP administration at 6 h compared to the control (Figure [2](#F2){ref-type="fig"}), suggesting that PACAP affects activation and invasion of T cells in the ischemic brain. A recent review has stated that T lymphocytes are activated, infiltrated into the brain, and appear to release cytokines to contribute to the early inflammation and brain injury following an ischemic stroke \[[@B35]\]. PACAP-induced neuroprotection may mediate T cell activation and infiltration to regulate the acute phase of inflammation. During the initial finding of this gene in lymphoid tissues, the *Crtam* RNA was also found in the brain as expected, but its significance has remained unclear \[[@B32]\]. In 2006, CRTAM was characterized and found to be highly expressed in the human cerebellum, particularly the Purkinje neurons, and where it was reported that CRTAM/Necl-2 binding may contribute to neuronal interactions \[[@B36]\]. Recently, the *Crtam* gene was also reported as an ethanol-responsive gene, and associated with alcohol preference in mice \[[@B37]\]. Although, we do not know the neuronal CRTAM function, based on the few reports available to date, it is tempting to speculate a neuroprotective role under PACAP influence in the ischemic hemisphere. Moreover, as with the induction of *IL* genes, it is of interest to note that *Crtam* is also regulated by PACAP38 at 6 but not 24 h, suggesting that PACAP effects at the molecular level can be seen within hours of administration to the brain. This result is also in accordance with our observation with the RT-PCR data that the effect of PACAP on gene expression can be divided into an early response (with potential involvement in neuroprotection) and late response (secondary effects) in the ischemic hemisphere. Nevertheless, additional experiments by functional approaches using the candidate genes are necessary to demonstrate the mechanism behind the action of PACAP. Two-dimensional gel electrophoresis western blotting, and immunofluorescent staining analysis identifies CRMP2 as a candidate protein factor in PACAP-influenced neuroprotection in the ischemic hemisphere ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Total soluble protein profiles were first examined by one-dimensional gel electrophoresis (SDS-PAGE). Staining the separated proteins with Flamingo fluorescent stain showed no significant differences in polypeptide patterns among all the samples examined (Additional file [6](#S6){ref-type="supplementary-material"}: Figure S4). Therefore, we proceeded to two-dimensional analysis. Flamingo-stained two-dimensional gel protein spots clearly revealed a major and newly appeared spot with a molecular weight of approximately 60 kDa and a pI of 5.5, in the ischemic hemisphere at 6 h (lower left two-dimensional gel image, Figure [5](#F5){ref-type="fig"}A) over the sham saline sample; this spot was also observed in the PACAP38-treated ischemic brain (lower right two-dimensional gel, Figure [5](#F5){ref-type="fig"}A) over the sham PACAP sample. Interestingly, the same spot was also observed at a much stronger intensity (more than 2-fold) in the ischemic hemisphere at 24 h (lower left two dimensional gel image, Figure [5](#F5){ref-type="fig"}B) but not in the sham saline sample. Two additional minor spots appeared at approximately the same molecular weight range but with a slightly acidic pI (isoelectric point). Surprisingly, and in contrast to the 6 h results, the major and minor spots were not found in the PACAP38-treated brain (lower right two-dimensional gel image, Figure [5](#F5){ref-type="fig"}B) over the sham PACAP sample. To identify the protein in these newly appearing spots, we excised these three spots from the stained gels and after in-gel trypsin digestion analyzed the peptides by MALDI-TOF-MS, as described in Methods. The MASCOT search revealed the major abundant protein to be a dihydropyrimidinase-related protein 2 (DPYL2_MOUSE) or collapsin response mediator protein 2 (CRMP2_MOUSE). No significant data (protein IDs) could be obtained for the two protein spots with the lowest abundance appearing exclusively at 24 h post ischemia. The CRMP2 protein, as it more popularly known, is involved in axonal growth and neuronal differentiation \[[@B17]\]. Although there is no information to date on the effects of PACAP on the CRMP2 protein, a few studies have already shown that the CRMP2 protein is induced in the brain after focal cerebral ischemia, in old mouse brains, and in the cerebral cortex of rats \[[@B38]-[@B40]\]. {#F5} We next examined by western blot analysis the CRMP2 protein profile on SDS-PAGE. Using a specific CRMP2 protein antibody three main bands of approximately 70, 65, and 63 kDa were detected (Figure [6](#F6){ref-type="fig"}). No significant change in protein expression was seen among the sham control and PMCAO samples, with or without PACAP38 treatment. However, in the PMCAO samples only, at an approximate molecular weight of 56 kDa, a cross-reacting protein band was seen. At 6 h post-ischemia, the 56 kDa protein was increased in abundance over the minus-PACAP sample. At 24 h post-PACAP treatment the 56 kDa protein band was found at very low levels. This result matches the induction profile of the major protein spot seen on the two-dimensional gels. Further, using two anti-phosphorylated CRMP2 proteins antibodies, approximately 70 and 65 kDa and 70 kDa cross-reacting bands were observed. However, no significant change in abundance of protein bands was observed (Additional file [17](#S17){ref-type="supplementary-material"}: Figure S7). No specific band at the 56 kDa molecular weight was seen in the PMCAO group. These data rule out the possibility of protein phosphorylation as a post-translational modification for the appearance of the 56 kDa cross-reacting protein band. {ref-type="supplementary-material"}: Figure S4.](1742-2094-9-256-6){#F6} Immunofluorescent staining using anti-CRMP2 antibody revealed that the CRMP2 protein is localized to the cytoplasm in neuronal cells as is evident from the healthy region of the ipsilateral hemisphere (Figure [7](#F7){ref-type="fig"}). Interestingly, in the penumbra, the CRMP2 protein appears to be more abundant in the 6-h PACAP group. In the core region, the CRMP2 protein is reduced in abundance, in particular under PACAP38 treatment, especially prominent by almost no presence of CRMP2 at 24 h after PACAP treatment to the ischemic brain. Comparing these three data, it can be stated that 1) two-dimensional gel electrophoresis and western blot analysis confirmed the presence of a newly appearing 56 kDa protein, 2) PACAP treatment influences its induction at 6 h post-treatment in the ischemic hemisphere, and 3) a very low level of the CRMP2 protein on two-dimensional gel at 24 h post-ischemia under PACAP treatment correlates well with the results seen from western blot analysis data and immunofluorescent staining of the core region. In other words, our results show that PACAP treatment affects the expression of CRMP2 protein, but how PACAP and CRMP2 are involved in neuroprotection of the brain remains to be answered. {ref-type="supplementary-material"}: Figure S6.](1742-2094-9-256-7){#F7} Conclusions =========== To our knowledge the omics-based study demonstrates significant change in gene and protein expressions after PACAP38 injection in the ischemic brain. Results presented in the study highlight the usefulness of global gene expression profiling in searching for changes in gene expression, and delineating the molecular events in a defined experimental PMCAO model. Furthermore, the proteomics approach proved to be critical in identifying the CRMP2 protein, which was surprisingly the most abundant and newly appearing protein spot on the two-dimensional gel. Post-confirmation of the two-dimensional gel spot by MS and western blot analyses, immuno-double staining revealed the possibility of CRMP2 as a controlling factor in PACAP-regulated control of ischemic neuroprotection. Forward or reverse genetic approaches involving certain target genes, including the most promising candidate CRMP2 protein, will be necessary to increase our understanding of the functional role of various genes in response to an ischemic injury and to clarify the precise mechanism behind PACAP38-induced neuroprotection. Furthermore, to further enhance our understanding of the functional roles of the many identified gene and protein candidates, future work such as utilizing the integrated pathway analysis (IPA) software for predicting the possible pathways and underlying mechanisms of PACAP-mediated molecular changes, and experiments, especially with the PACAP knockout mice, will be essential. Finally, the numerous specific PACAP target transcripts that were unannotated have not been discussed here, and will be also a target for future studies. Abbreviations ============= CAN: Acetonitrile; CNS: Central nervous system; CRMP2: Collapsin response mediator protein-2; Crtam: Regulatory T cell molecule; DAPI: 4′ 6-diamidine-2-phenylindole dihydrochloride; DPYL2: Dihydropyrimidinase-related protein 2; IL: Interleukin; MALDI-TOF-MS: Matrix-assisted laser desorption/ionization-time of flight-mass spectrometry; MS: Mass spectrometry; NG: Neurological grade; PACAP: Pituitary adenylate cyclase-activating polypeptide; PB: Phosphate buffer; PBS: Phosphate-buffered saline; PMCAO: Permanent middle cerebral artery occlusion; PVDF: Polyvinyldifluoride; TFA: Trifluoroacetic acid; VIP: Vasoactive intestinal polypeptide. Competing interests =================== The authors declare they have no competing interests. Authors\' contributions ======================= MH, TN, RR, JS, and SS discussed and designed the study plan. MH and TN performed the animal experiments. MH, TN, RR, JS, and TA designed and performed the genomics and proteomics experiments. MH, TN, RR, JS, and TO analyzed the data. MH and RR wrote the paper. MH, TN, RR, TT, KT, and SS checked, revised and finalized the paper. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Table S1.** The number of mice used for this experiment. A total of 23 mice were prepared, and 3 each were selected randomly based on neurological grade (NG). Mice with score 0 and 3 were not selected for sampling the brains. ###### Click here for file ###### Additional file 2 **Table S2.** Primer combinations used for RT-PCR. ###### Click here for file ###### Additional file 3 **Figure S1.** Preparation of LB-TT. ###### Click here for file ###### Additional file 4 **Figure S2.** ProteoExtract Protein Precipitation Kit Protocol (Illustrated). ###### Click here for file ###### Additional file 5 **Figure S3.** Two-dimensional gel electrophoresis. (2-DGE) ###### Click here for file ###### Additional file 6 **Figure S4.** One-dimensional gel electrophoresis (SDS-PAGE). ###### Click here for file ###### Additional file 7 **Figure S5.** Dissected Brain Storage and Sectioning Protocol (Illustrated). ###### Click here for file ###### Additional file 8 **Figure S6.** Dissected Regions as Ischemic core, Penumbra and Healthy of the Ipsilateral Hemisphere (Illustrated). ###### Click here for file ###### Additional file 9 **Table S3.** Effect of PACAP 38 injection on expression of genes at 6 h with a focus on ischemia and PACAP. ###### Click here for file ###### Additional file 10 **Table S4.** Effect of PACAP 38 injection on expression of genes at 24 h with a focus on ischemia and PACAP. ###### Click here for file ###### Additional file 11 **Table S5.** Effect of PACAP 38 injection on expression of genes at 6 h with a focus on PACAP and ischemia. ###### Click here for file ###### Additional file 12 **Table S6.** Effect of PACAP 38 injection on expression of genes at 24 h with a focus on PACAP and ischemia. ###### Click here for file ###### Additional file 13 **Table S7.** The functionally categorized (pathway- or specific disease states-focused gene classifications) upregulated genes at 6 h. ###### Click here for file ###### Additional file 14 **Table S8.** The functionally categorized (pathway- or specific disease states-focused gene classifications) downregulated genes at 6 h. ###### Click here for file ###### Additional file 15 **Table S9.** The functionally categorized (pathway- or specific disease states-focused gene classifications) upregulated genes at 24 h. ###### Click here for file ###### Additional file 16 **Table S10.** The functionally categorized (pathwayor specific disease states-focused gene classifications) downregulated genes at 24 h. ###### Click here for file ###### Additional file 17 **Figure S7.** Western blot analysis of the anti-phosphorylated CRMP2 proteins cross-reacting proteins. ###### Click here for file Acknowledgements ================ MH gratefully acknowledges the members of Professor Seiji Shioda's laboratory (Department of Anatomy I) for their support and encouragement during this study. This work was supported in part by Grant-in-Aid for Scientific Research (TN and SS) and High-Technology Research Center Project (SS) from the Ministry of Education, Culture, Sports, Science and Technology, Japan, and Support Program for the Strategic Research Foundation at Showa University, 2008 to 2012.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Microorganisms in nature engage in a variety of interactions with other species in their environment. Syntrophy is one such type of inter-species interaction in which one species lives off the metabolic by-products of another \[[@pcbi.1004364.ref001]--[@pcbi.1004364.ref003]\]. Synthetic methanogenic communities \[[@pcbi.1004364.ref004]\] are typically tightly constrained by thermodynamics, as the oxidation reactions carried out by the first community member are thermodynamically unfavorable unless the degradation products are maintained at low levels by the second community member \[[@pcbi.1004364.ref005]\]. In anaerobic syntrophic communities, electrons are transferred from one partner to the other through direct contact or small molecule diffusion \[[@pcbi.1004364.ref006]\]. Traditional biochemistry has elucidated intracellular electron transport mechanisms \[[@pcbi.1004364.ref003],[@pcbi.1004364.ref007]--[@pcbi.1004364.ref009]\], but it is difficult to evaluate these pathways in their metabolic and environmental context. Genome-scale metabolic models (GEMs) \[[@pcbi.1004364.ref010]--[@pcbi.1004364.ref012]\] and constraint-based methods are powerful computational tools for understanding individual pathways in a broader metabolic context including both isolated microbial species \[[@pcbi.1004364.ref013]--[@pcbi.1004364.ref015]\] and simple microbial communities \[[@pcbi.1004364.ref016]--[@pcbi.1004364.ref026]\]. One of the earliest microbial community models used flux balance analysis (FBA, \[[@pcbi.1004364.ref027]\]) to investigate formate and H~2~ exchange between the sulfate-reducing bacterium *Desulfovibrio vulgaris* and the methanogenic archaeon *Methanococcus maripaludis* \[[@pcbi.1004364.ref016]\]. In this study, each organism was modeled as a compartment within a larger community-scale model. Compartmentalized approaches have been used to study the origins of cooperation and competition \[[@pcbi.1004364.ref017]--[@pcbi.1004364.ref019]\], as well as specific communities \[[@pcbi.1004364.ref020]--[@pcbi.1004364.ref024]\]. These approaches \[[@pcbi.1004364.ref016]--[@pcbi.1004364.ref024]\] have often used a single (joint) objective function to capture community behavior. OptCom \[[@pcbi.1004364.ref025],[@pcbi.1004364.ref026]\] instead uses a multi-level optimization framework, to capture the trade-offs between individual and community fitness, with separate objective functions for the individual species and the community. In addition, community FBA (cFBA) \[[@pcbi.1004364.ref028]\] extends compartmentalized approaches \[[@pcbi.1004364.ref016]--[@pcbi.1004364.ref024]\] to specifically account for individual species' biomass abundance. Genome-scale models can also be used to study the relationship between thermodynamics and metabolism, by ensuring that network predictions are consistent with thermodynamic principles \[[@pcbi.1004364.ref029]--[@pcbi.1004364.ref034]\]. In this study, we used thermodynamics-based metabolic flux analysis (TMFA) to develop a thermodynamic, coculture model of the syntrophic association between the anaerobic bacterium *Syntrophobacter fumaroxidans* and the methanogenic archaeon *Methanospirillum hungatei*. In association with *M*. *hungatei*, *S*. *fumaroxidans* converts propionate to acetate, CO~2~, and H~2~ \[[@pcbi.1004364.ref035]--[@pcbi.1004364.ref037]\]. CO~2~ and H~2~ can be interconverted to formate \[[@pcbi.1004364.ref038]--[@pcbi.1004364.ref040]\], with H~2~ and formate serving as the electron carriers between the two species. H~2~ and formate production are only observed during syntrophic growth. Using a thermodynamic, constraint-based model, we set out to test the proposed hypothesis that this behavior is governed by thermodynamics \[[@pcbi.1004364.ref003],[@pcbi.1004364.ref005],[@pcbi.1004364.ref006],[@pcbi.1004364.ref008],[@pcbi.1004364.ref009]\]. We developed genome-scale metabolic reconstructions of both microorganisms, and verified proposed mechanisms of ATP production within each individual species. Additional constraints and a cellular objective function were identified to predict the proper flux through experimentally characterized carbon and electron transport pathways during monoculture and syntrophic (i.e., coculture) growth. Our analysis revealed that thermodynamic constraints alone are insufficient to explain why *S*. *fumaroxidans* does not produce H~2~ in monoculture. We also extended TMFA to model the syntrophic association between the two microorganisms. The association is modeled as a continuous coculture system with constraint-based models for each microbe and a mass balance around the reactor. Similar to cFBA \[[@pcbi.1004364.ref028]\], the coculture model accounted for the biomass concentrations of each species. We predicted the behavior of this syntrophic association under a variety of dilution rates, and identified regimes of behavior consistent with experimental observations. Results {#sec002} ======= Metabolic reconstructions of *S*. *fumaroxidans* (*i*Sfu648) and *M*. *hungatei* (*i*Mhu428) were tested and parameterized using experimental data from growth on single substrates in monoculture and coculture. The *i*Sfu648 and *i*Mhu428 models were used to verify proposed mechanisms of energy conservation within each species. A coculture model was then developed to identify conditions where H~2~ and formate, produced by *S*. *fumaroxidans*, is fully metabolized by *M*. *hungatei*. Testing and Parameterizing the *i*Mhu428 Metabolic Model {#sec003} -------------------------------------------------------- The *i*Mhu428 reconstruction of *M*. *hungatei* was built from the *i*MB745 reconstruction of *M*. *acetivorans* \[[@pcbi.1004364.ref041]\]. A preliminary draft reconstruction was built from *i*MB745 using the RAVEN Toolbox \[[@pcbi.1004364.ref042]\] and the KEGG SSDB \[[@pcbi.1004364.ref043]\]; however, *M*. *hungatei* orthologs were found for only 428 of the 745 genes in *i*MB745. To avoid extensive gapfilling, reactions from the *i*MB745 were copied into the *M*. *hungatei* reconstruction, with modifications to reflect key metabolic features of *M*. *hungatei* (see [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}). As a consequence, the *i*Mhu428 reconstruction is a draft reconstruction requiring further evaluation. A thermodynamic model for *i*Mhu428 was built and TMFA was used to predict ATP generating mechanisms in minimal media monoculture conditions (see S1 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"} for constraints used). Experimental evidence suggests that *M*. *hungatei* is able to generate 0.5 mole ATP per mole of CO~2~ converted to CH~4~ \[[@pcbi.1004364.ref007]\], via the metabolic route shown in [Fig 1](#pcbi.1004364.g001){ref-type="fig"}. In order for this route to be thermodynamically feasible, the Δ~*r*~*G*^\'0^ of one reaction (*FMFTSPFT*, formylmethanofuran-tetrahydromethanopterin formyltransferase) had to be allowed to vary within a 99% confidence interval of its estimated standard transformed Gibbs free energy of reaction ($\Delta_{r}G_{est}^{'0}$) (rather than the 95% interval used for all other reactions) in order to carry flux in the proper direction. The reactions for carbon source utilization in *M*. *hungatei* are well-characterized \[[@pcbi.1004364.ref007],[@pcbi.1004364.ref044]--[@pcbi.1004364.ref047]\], but uncertainty remains about the stoichiometry of small ion transport \[[@pcbi.1004364.ref007]\]. Na^+^ transport stoichiometries associated with tetrahydromethanopterin S-methyltransferase (*MTSPCMMT_CM5HBCMT*, E.C. 2.1.1.86) and a Na^+^/H^+^ antiporter (*NAT3_1*) were selected to give an ATP yield matching the experimental estimates: two Na^+^ ions exported by *MTSPCMMT_CM5HBCMT*, and a one Na^+^ per H^+^ transported by *NAT3_1*. However, different stoichiometries for these reactions are also thermodynamically possible (see [Discussion](#sec011){ref-type="sec"}). {ref-type="supplementary-material"}.](pcbi.1004364.g001){#pcbi.1004364.g001} Experimental measurements of growth rates, yields, and maintenance costs were also used to identify substrate uptake rates (SUR) for CO~2~ and formate, and the growth- (GAM) and non-growth-associated (NGAM) ATP maintenance requirements for *M*. *hungatei*, as described in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. NGAM represents the amount of energy spent to maintain the cell (i.e., maintenance energy), while GAM represents energy spent on growth-related functions (e.g., protein synthesis). For the *i*Mhu428 model, the NGAM was estimated to be 0.6 mmol ATP/gDW/day, GAM was estimated to be 47 mmol ATP/gDW, SUR~CO2~ was estimated to be 75.7 mmol/gDW/day, and SUR~formate~ was estimated to be 955 mmol/gDW/day. Testing and Parameterizing the *i*Sfu648 Metabolic Model {#sec004} -------------------------------------------------------- The *i*Sfu648 reconstruction of *S*. *fumaroxidans* was built from the KEGG database using the RAVEN Toolbox \[[@pcbi.1004364.ref042]\]. The resulting draft reconstruction was manually refined (see [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}), with particular attention paid to ATP production mechanisms. A number of studies have identified gene clusters encoding a variety of hydrogenases, dehydrogenases, and other electron transport enzymes \[[@pcbi.1004364.ref008],[@pcbi.1004364.ref009],[@pcbi.1004364.ref048]--[@pcbi.1004364.ref051]\], whose expression levels vary across growth conditions \[[@pcbi.1004364.ref051]\]. All told, 17 enzymes which catalyze 12 different electron transport reactions have been identified (S3 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"}). In many cases, the proposed reactions catalyzed by these enzymes differ between studies; a brief description of each reaction and justification for each annotation is given in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. The draft reconstruction was updated to be consistent with the reported carbon utilization and electron transport reactions, and the resulting stoichiometric model was converted to a thermodynamic model. ATP Production Mechanisms in *S*. *fumaroxidans* {#sec005} ------------------------------------------------ Experimental studies have elucidated five growth modes for *S*. *fumaroxidans*: four in monoculture and one in coculture with *M*. *hungatei* ([S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}) \[[@pcbi.1004364.ref036],[@pcbi.1004364.ref048],[@pcbi.1004364.ref052]\]. This work examines the three most commonly studied growth modes ([Table 1](#pcbi.1004364.t001){ref-type="table"}): monoculture growth on fumarate, monoculture growth on fumarate plus propionate, and coculture growth on propionate. 10.1371/journal.pcbi.1004364.t001 ###### Experimentally observed and computationally predicted extracellular flux distributions for *S*. *fumaroxidans* examined in this study. {#pcbi.1004364.t001g} Growth Mode[\*](#t001fn002){ref-type="table-fn"} Experimental Stoichiometry (No Growth)[^\#^](#t001fn003){ref-type="table-fn"} Predicted Stoichiometry (Max Growth) Observed Growth (1/days) Predicted Growth (1/days) -------------------------------------------------- ------------------------------------------------------------------------------- ------------------------------------------------------------------- -------------------------- --------------------------------------------- Fumarate, Monoculture 7 fumarate → 6 succinate + 4 CO~2~ 7 fumarate → 5.11 succinate + 4.75 CO~2~ + 0.25 acetate 0.33 0.33 Propionate + Fumarate, Monoculture propionate + 3 fumarate → acetate + CO~2~ + 3 succinate propionate + 3 fumarate → 0.59 acetate +1.25 CO2 + 2.83 succinate 0.73 0.41[^\$^](#t001fn004){ref-type="table-fn"} Propionate, Coculture propionate → acetate + CO2 + 3 H~2~ propionate---\> 0.99 acetate + 0.86 CO~2~ + 2.69 H~2~ 0.22 0.22 Experimental measurements from fumarate-only and propionate-only conditions were used to parameterize the model. \*Monoculture simulations were performed with qualitative reaction direction constraints which prevent H~2~ production. ^\#^Experimental stoichiometries were calculated from experimental measurements assuming no biomass growth \[[@pcbi.1004364.ref036],[@pcbi.1004364.ref052]\]. ^\$^This growth rate was predicted after the model was modified to include an explicit constraint on the propionate to fumarate uptake ratio. A variety of experimental findings were synthesized to develop theoretical flux distributions for these three growth modes ([Fig 2](#pcbi.1004364.g002){ref-type="fig"} and [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}). These experimental findings suggested additional regulatory and flux-coupling constraints for the *i*Sfu648 model, such as coupling between fumarate reductase and the cytosolic hydrogenase due to co-localization in the membrane (see S2 Table in [S1 Dataset](#pcbi.1004364.s010){ref-type="supplementary-material"} and [S1](#pcbi.1004364.s010){ref-type="supplementary-material"} Text for the full set of constraints and their justification). {ref-type="table"}. (A) Monoculture growth on fumarate alone. (B) Monoculture growth on fumarate and propionate. (C) Syntrophic growth on propionate. Substrates and products are indicated with purple and blue ovals, respectively, while reactions with and without estimates are indicated with green and orange arrows, respectively. Arrow thickness indicates relative flux values. Metabolite and reaction abbreviations are given in [S4 Dataset](#pcbi.1004364.s004){ref-type="supplementary-material"}.](pcbi.1004364.g002){#pcbi.1004364.g002} During monoculture growth on fumarate alone ([Fig 2A](#pcbi.1004364.g002){ref-type="fig"}), one mole of fumarate gets fully oxidized to CO~2~, while six moles of fumarate get reduced to succinate \[[@pcbi.1004364.ref048],[@pcbi.1004364.ref052]\]: $$\left. 7\ \text{fumarate}\rightarrow 6\ \text{succinate}\mspace{2mu} + \mspace{2mu} 4\ \text{CO}_{2} \right.$$ The oxidation of one fumarate to CO~2~ generates one ATP and five reducing equivalents (three NADH and two pairs of reduced ferredoxin) \[[@pcbi.1004364.ref048],[@pcbi.1004364.ref052]\], while the reduction of additional fumarate to succinate by fumarate reductase (*FRD*) consumes reducing equivalents (menaquinol) \[[@pcbi.1004364.ref048]\]. Electrons are transferred from NADH and reduced ferredoxin to menaquinone through the combined action of the Rnf complex (*RNF*), the ferredoxin-oxidizing hydrogenase (*frH*~*2*~*ase*), the cytosolic hydrogenase (*cytH*~*2*~*ase*) and formate hydrogen lyase (*FHL*). The reduction of fumarate to succinate also generates the proton motive force (PMF) responsible for driving the *RNF* reaction and producing ATP. During monoculture growth on fumarate plus propionate ([Fig 2B](#pcbi.1004364.g002){ref-type="fig"}), one mole of propionate gets oxidized to succinate, while one mole of fumarate gets oxidized to acetate and CO~2~. Two additional moles of fumarate get reduced to succinate \[[@pcbi.1004364.ref036],[@pcbi.1004364.ref048]\]: $$\left. \text{propionate}\mspace{2mu} + \mspace{2mu} 3\ \text{fumarate}\rightarrow\text{acetate}\mspace{2mu} + \mspace{2mu}\text{CO}_{2}\mspace{2mu} + \mspace{2mu} 3\ \text{succinate} \right.$$ The oxidation of fumarate to acetate and CO~2~ produces one NADH and one pair of reduced ferredoxin, while the reduction of fumarate to succinate by *FRD* consumes menaquinol. Electrons are transferred from NADH and reduced ferredoxin to menaquinone through the combined action of the confurcating hydrogenase (*cH*~*2*~*ase*) and *cytH*~*2*~*ase*. Oxidation of propionate to succinate produces one ATP, while *FRD* generates the PMF necessary for additional ATP production. During coculture growth on propionate ([Fig 2C](#pcbi.1004364.g002){ref-type="fig"}), propionate gets oxidized to acetate and CO~2~ via the methylmalonyl-CoA pathway \[[@pcbi.1004364.ref048],[@pcbi.1004364.ref052]\]: {#pcbi.1004364.e004g} propionate → acetate \+ CO 2 \+ 3 H 2 ATP is generated during the oxidation of propionate to succinate, and this ATP establishes the PMF necessary to drive the endergonic oxidation of succinate to fumarate (*SDH*), producing menaquinone. *cytH*~*2*~*ase* then transfers electrons from menaquinol to two protons, generating H~2~. The oxidation of fumarate to acetate and CO~2~ produces one NADH and one pair of reduced ferredoxin, and *cH*~*2*~*ase* couples NADH and ferredoxin re-oxidation with H~2~ production. Unlike in the monoculture growth modes, the H~2~ is not consumed intracellularly and must diffuse outside the cell. It has been proposed that the net production of H~2~ by *S*. *fumaroxidans* is only thermodynamically favorable at the low H~2~ concentrations maintained by methanogens, thereby explaining why *S*. *fumaroxidans* only produces H~2~ during coculture growth. *S*. *fumaroxidans* exhibits considerable flexibility in its ATP production mechanisms during coculture growth (\[[@pcbi.1004364.ref008],[@pcbi.1004364.ref009],[@pcbi.1004364.ref048],[@pcbi.1004364.ref051]\]), and can produce formate instead of CO~2~ ([Fig 3](#pcbi.1004364.g003){ref-type="fig"}) yielding an overall transformation of: $$\left. \text{propionate}\rightarrow\text{acetate}\mspace{2mu} + \mspace{2mu}\text{formate}\mspace{2mu} + \mspace{2mu} 2\ \text{H}_{2} \right.$$ {ref-type="disp-formula"}. Substrates and products are indicated with purple and blue ovals, respectively, while reactions with and without estimates are indicated with green and orange arrows, respectively. Arrow thickness indicates relative flux values. Metabolite and reaction abbreviations are given in [S4 Dataset](#pcbi.1004364.s004){ref-type="supplementary-material"}.](pcbi.1004364.g003){#pcbi.1004364.g003} In one mechanism ([Fig 3A](#pcbi.1004364.g003){ref-type="fig"}), activity of the cytosolic formate hydrogenase (*cytFDH*) substitutes for the activity of *cytH*~*2*~*ase*. In a second mechanism ([Fig 3B](#pcbi.1004364.g003){ref-type="fig"}), the confurcating formate dehydrogenase (*cFDH*) substitutes for *cH*~*2*~*ase*. Here, *cFDH* couples NADH and ferredoxin re-oxidation with the conversion of CO~2~ (from propionate oxidation) to formate. Model Predictions of ATP Production by *S*. *fumaroxidans* {#sec006} ---------------------------------------------------------- Experimental evidence and conceptual models of *S*. *fumaroxidans* energy metabolism suggest that the carbon and electron transfer pathways shown in [Fig 2](#pcbi.1004364.g002){ref-type="fig"} provide the sole source for ATP production in *S*. *fumaroxidans*, either by substrate-level phosphorylation or through establishment of a proton gradient used by ATP synthase \[[@pcbi.1004364.ref005]\]. To test the computational model's predictions, TMFA was used to maximize ATP production under each of the three growth modes. When flux was restricted to a reduced network containing all the reactions shown in [Fig 2](#pcbi.1004364.g002){ref-type="fig"} (listed in S4 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"}), the *i*Sfu648 model correctly predicted the flux distributions shown in [Fig 2](#pcbi.1004364.g002){ref-type="fig"}. However, when flux was allowed throughout the entire network, additional flux distributions with higher ATP yields were identified. Additional reaction direction constraints were developed to ensure model-predicted flux distributions matched experimental observations (see S2 Table in [S1 Dataset](#pcbi.1004364.s010){ref-type="supplementary-material"} and [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"} for details). However, the resulting flux distributions are not fully consistent with the hypothesis that *S*. *fumaroxidans* has adapted to maximize its energy yield. Evaluating Thermodynamics of H~2~ Production {#sec007} -------------------------------------------- Experimental studies of *S*. *fumaroxidans* have shown that H~2~ is not produced during growth in monoculture \[[@pcbi.1004364.ref036],[@pcbi.1004364.ref052],[@pcbi.1004364.ref053]\], and it is widely thought that H~2~ production is only thermodynamically favorable at low partial pressures \[[@pcbi.1004364.ref003],[@pcbi.1004364.ref006],[@pcbi.1004364.ref008],[@pcbi.1004364.ref009]\]. In particular, methanogens in syntrophic communities enable sustained H~2~ production by consuming H~2~ and keeping its partial pressure low \[[@pcbi.1004364.ref003],[@pcbi.1004364.ref005],[@pcbi.1004364.ref006],[@pcbi.1004364.ref008],[@pcbi.1004364.ref009]\]. Indeed, when H~2~ production was observed in monoculture, H~2~ production ceased at a partial pressure of approximately 10 Pa \[[@pcbi.1004364.ref053]\]. However, when maximizing H~2~ production under monoculture conditions, simulations reveal that H~2~ production remains thermodynamically feasible. For example, during monoculture growth on fumarate, the *i*Sfu648 model predicts that H~2~ can be produced via the following mechanism: $$\left. \text{fumarate}\rightarrow 4\ \text{CO}_{2} + \mspace{2mu} 6\ \text{H}_{2} \right.$$ In this scenario, H~2~ molecules produced by the ferredoxin-oxidizing hydrogenase are exported outside the cell, instead of serving as substrates for the cytosolic hydrogenase. As a result, no PMF is generated by fumarate reductase, and the net ATP yield is zero. Thus, while H~2~ production remains thermodynamically possible, H~2~ production is only associated with sub-optimal mechanisms of ATP generation. This suggests that thermodynamic considerations alone may not explain the absence of H~2~ production during monoculture growth, but that the observed flux distribution may instead be driven by demands for energy generation. While H~2~ production was not predicted for monoculture conditions when ATP production was maximized, H~2~ production was initially predicted when growth was instead maximized. To eliminate monoculture H~2~ production in the TMFA model, we first sought to constrain ratios of metabolite concentrations with an approach similar to that used to correct TMFA growth predictions \[[@pcbi.1004364.ref034]\] (see [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"} for details). While metabolite ratio constraints could be identified to prevent some H~2~ production mechanisms, H~2~ production during monoculture growth could not be completely eliminated. If thermodynamics prevents H~2~ production in monoculture, then the current thermodynamic model may contain too much uncertainty in its Gibbs free energy estimates (see [Discussion](#sec011){ref-type="sec"}). Regulatory effects could also potentially prevent H~2~ production in monoculture conditions. To correct the model, all subsequent monoculture simulations were performed by preventing H~2~ production. Parameterization of the *i*Sfu648 Metabolic Model {#sec008} ------------------------------------------------- Model parameters were estimated after reaction direction constraints were added to the *i*Sfu648 model (to be consistent with reported ATP generation and H~2~ production mechanisms). Experimental measurements of growth rates, yields, and maintenance costs were used to identify the SURs, GAM, and NGAM parameters for *S*. *fumaroxidans*. These parameters were estimated using data from monoculture growth on fumarate alone and coculture growth on propionate alone (see [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}). For the *i*Sfu648 model, the following parameters resulted in the best fit of the model to the experimental data: NGAM = 3.36 mmol ATP/gDW/day, GAM = 22.8 mmol ATP/gDW, SUR~propionate~ = 37.7 mmol/gDW/day, and SUR~fumarate~ = 27.6 mmol/gDW/day. Using these parameter values, the *in silico* growth rates under each growth condition were predicted ([Table 1](#pcbi.1004364.t001){ref-type="table"}). Not surprisingly, the predicted growth rates for fumarate alone and propionate alone conditions agree with experimental observations (since these were used to estimate the parameter values). However, the model significantly under-predicts the measured growth rate during monoculture growth on fumarate plus propionate (0.55 days^-1^ predicted, 0.73 days^-1^ observed). This discrepancy could be caused by differences in uptake rates or maintenance costs in the fumarate plus propionate condition compared to the conditions with propionate alone or fumarate alone. Uptake and Secretion at Maximal Growth {#sec009} -------------------------------------- When maximizing biomass production on the entire network, the *i*Sfu648 model predicted a wide range of product secretion rates. When the enzyme cost (i.e., total flux) was minimized at the maximum growth (pTMFA \[[@pcbi.1004364.ref054]\], see [Methods](#sec015){ref-type="sec"}) the model-predicted product yields closely matched reported values for two of the three growth modes ([Table 1](#pcbi.1004364.t001){ref-type="table"})---monoculture growth on fumarate alone and coculture growth on propionate alone. These results indicate that the majority of carbon is diverted to fermentation products, consistent with the expectation that high fluxes through the low-energy fermentation pathways are needed to meet cellular energy demands. However, for monoculture growth on fumarate plus propionate, the model failed to predict that fumarate and propionate should be consumed at the observed ratio of approximately three fumarate per propionate \[[@pcbi.1004364.ref036]\]. Instead, the model predicted both substrates would be consumed at their maximum SURs, resulting in a ratio of 0.73 fumarate per propionate. The experimentally observed 3:1 ratio is thought to arise due to coupling within the metabolic network ([Fig 2B](#pcbi.1004364.g002){ref-type="fig"}), since oxidation of one fumarate produces one CO~2~ (used to oxidize one propionate) and two pairs of electrons (used to reduce two fumarate). While this coupling arises naturally on the reduced network, the full metabolic network enables alternative coupling mechanisms (not shown) that permit other fumarate to propionate ratios. Since propionate oxidation generates carbon precursors and ATP for biomass, maximizing biomass production results in the model under-predicting the fumarate to propionate uptake ratio. Increasing the ratio of fumarate to propionate uptake (to 3:1) decreases the predicted propionate SUR and growth rate (S1 Fig in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}), implying the experimental ratio is sub-optimal with respect to growth maximization. Instead of constraining SURs (since values were not reported in the literature), we incorporated a fumarate to propionate SUR ratio constraint for this condition. While the predicted product yields closely matched experimental observations (after imposing the SUR ratio constraint), the pTMFA-predicted intracellular flux distribution during monoculture growth on fumarate and propionate substantially deviated from that shown in [Fig 2B](#pcbi.1004364.g002){ref-type="fig"}. Further constraints on reaction directions were required so that propionate and fumarate were metabolized in the model via the pathways shown in [Fig 2B](#pcbi.1004364.g002){ref-type="fig"} (results not shown). Taken together, the need for constraints on reaction directions and fumarate to propionate uptake ratio suggests that neither maximization of biomass nor minimization of enzyme cost are sufficient to explain the fluxes of *S*. *fumaroxidans* growing in this monoculture condition. Behavior of *M*. *hungatei* and *S*. *fumaroxidans* in Coculture {#sec010} ---------------------------------------------------------------- During growth in coculture, *S*. *fumaroxidans* converts propionate to acetate, H~2~, and CO~2~ or formate \[[@pcbi.1004364.ref036],[@pcbi.1004364.ref052]\], while *M*. *hungatei* consumes acetate, CO~2~, H~2~, and formate and produces CH~4~ \[[@pcbi.1004364.ref055]\]. *M*. *hungatei* can also optionally interconvert excess CO~2~ and H~2~ to formate via a formate dehydrogenase \[[@pcbi.1004364.ref056]\]. Cocultures of *M*. *hungatei* and *S*. *fumaroxidans* have been grown in both batch and continuous (chemostat) systems. During batch growth, H~2~ pressure rose during the lag phase and became constant during exponential growth \[[@pcbi.1004364.ref053]\]. Continuous cultures also exhibited constant H~2~ partial pressure \[[@pcbi.1004364.ref053]\]. However, to the best of our knowledge, measurements for the relative ratios of *M*. *hungatei* to *S*. *fumaroxidans* at constant H~2~ pressure (where H~2~ consumption and production rates are balanced) have not been reported. Instead, an overall reaction for the coculture of $$\left. \text{propionate}\rightarrow\text{acetate}\ + \ 0.25\ \text{CO}_{2} + \mspace{2mu} 0.75\ \text{CH}_{4} \right.$$ is frequently discussed \[[@pcbi.1004364.ref005],[@pcbi.1004364.ref053]\], which can occur at a ratio of three *M*. *hungatei* to four *S*. *fumaroxidans*. Since initial conditions for batch experiments were not reported, a continuous culture model was constructed and first evaluated using a 3:4 relative biomass ratio (*M*. *hungatei*: *S*. *fumaroxidans)*. The continuous coculture model included constraint-based models for each microbe and mass balances around the reactor (S2 Fig in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}), and accounted for the biomass concentrations of each species. Both species were constrained to grow at the dilution rate, and the model minimized the species-weighted total flux through the two metabolic networks (pTMFA). Propionate was the only substrate in the reactor feed (i.e., it had a net flux into the reactor), ensuring that all carbon and electrons used by *M*. *hungatei* were produced by *S*. *fumaroxidans*. Predicted yields around individual species were first evaluated ([Fig 4A](#pcbi.1004364.g004){ref-type="fig"}). At low dilution rates, *S*. *fumaroxidans* was predicted to convert propionate to acetate, H~2~, and CO~2~/formate and *M*. *hungatei* was predicted to convert CO~2~/formate to CH~4~ ([Fig 4B and 4C](#pcbi.1004364.g004){ref-type="fig"}, which show alternate solutions with CO~2~ or formate being exchanged). As the reactor dilution rate increased, the predicted species yields of H~2~ and CO~2~/formate (*S*. *fumaroxidans*) and CH~4~ (*M*. *hungatei*) decreased slightly ([Fig 4B and 4C](#pcbi.1004364.g004){ref-type="fig"}). In addition, species' uptake rates of propionate and CO~2~/formate increased with dilution rate, as biochemical transformation of these substrates provides the energy needed for cellular growth and maintenance. {#pcbi.1004364.g004} Species' uptake rates, secretion rates, and relative biomass ratio affects overall bioreactor yields, and these bioreactor yields were subsequently investigated ([Fig 4D](#pcbi.1004364.g004){ref-type="fig"}). At a 3:4 relative biomass ratio (*M*. *hungatei*: *S*. *fumaroxidans)*, *M*. *hungatei* did not fully utilize all of the H~2~ and CO~2~/formate produced by *S*. *fumaroxidans*, even at high dilution rates ([Fig 4E and 4F](#pcbi.1004364.g004){ref-type="fig"}). The net H~2~ production by the community indicates that *S*. *fumaroxidans* produces more H~2~ than *M*. *hungatei*'s needs and suggests the community can maintain higher *M*. *hungatei* to *S*. *fumaroxidans* ratios or that *S*. *fumaroxidans* could support faster growth of *M*. *hungatei*. At a dilution rate of 0.05 days^-1^, *S*. *fumaroxidans* produces H~2~ in excess of *M*. *hungatei*'s energy needs until the relative *M*. *hungatei* to *S*. *fumaroxidans* biomass ratio reaches approximately 1.6:1 ([Fig 5](#pcbi.1004364.g005){ref-type="fig"}). These simulations suggest that invariant external H~2~ concentration requires high ratios of *M*. *hungatei* to *S*. *fumaroxidans* (higher than has been proposed in the literature based on overall reaction stoichiometries), *M*. *hungatei* growing at faster rates than *S*. *fumaroxidans* (e.g., in batch culture), or a combination of the two. {ref-type="fig"}, formate could be exchanged in the place of CO~2~ and H~2~.](pcbi.1004364.g005){#pcbi.1004364.g005} Furthermore, studies have shown that in coculture, *S*. *fumaroxidans* passes electrons to *M*. *hungatei* via formate, as well as H~2~ \[[@pcbi.1004364.ref038],[@pcbi.1004364.ref039],[@pcbi.1004364.ref056]\]. Coculture simulations predicted that formate could be exchanged in lieu of CO~2~ ([Fig 4B and 4C](#pcbi.1004364.g004){ref-type="fig"}), without affecting the predicted bioreactor yields or species-weighted total flux (pTMFA objectives). When formate is exchanged the formate dehydrogenases of *S*. *fumaroxidans* and *M*. *hungatei* facilitate the interconversion of formate to CO~2~ and H~2~. Finally, steady-state metabolite concentrations in the coculture were predicted using thermodynamic variability analysis \[[@pcbi.1004364.ref033],[@pcbi.1004364.ref034]\] ([S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}). Similar to a previous study of *E*. *coli* \[[@pcbi.1004364.ref034]\], the majority of steady-state metabolite concentrations were not constrained by thermodynamics (i.e., the concentration ranges were the global concentration bounds of 0.01mM and 20 mM). However, hypotheses for some extracellular ([Table 2](#pcbi.1004364.t002){ref-type="table"}) and intracellular (S5 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"}) metabolite concentrations could be made. The model predicts that propionate in the media must be greater than 0.36 mM for coculture growth to occur, and that acetate and CO~2~ concentrations must be less than 4mM and 0.65 mM, respectively. The model also predicts critical concentrations for H~2~ (0.0032 mM) and formate (0.0020 mM, when formate is being exchanged), which must be maintained in order for methanogenesis to occur. All of these predictions were insensitive to both the dilution rate and biomass ratio (*M*. *hungatei*: *S*. *fumaroxidans*). 10.1371/journal.pcbi.1004364.t002 ###### Model-predicted steady-state metabolite concentrations for select external metabolites in coculture. {#pcbi.1004364.t002g} Metabolite Default Lower Bound Default Upper Bound Predicted Lower Bound Predicted Upper Bound ---------------- ------------------------------------------- --------------------- ----------------------- ----------------------- Propionate 0.01 mM 20 mM 0.36 mM 20 mM Acetate 0.01 mM 20 mM 0.01 mM 4.0 mM Carbon Dioxide 0.01 mM 20 mM 0.27 mM 0.65 mM H~2~ None[\*](#t002fn001){ref-type="table-fn"} 20 mM 0.0032 mM 8.1 mM Formate None[\*](#t002fn001){ref-type="table-fn"} 20 mM 0.0020 mM 4.0 mM Methane 0.01 mM 20 mM 0.01 mM 20 mM \* No lower bounds were imposed on H~2~ or formate concentrations, which can fall below 0.01 mM, based on experimental measurements. Discussion {#sec011} ========== The *i*Mhu428 and *i*Sfu648 thermodynamic models were successfully used to verify proposed carbon and ATP production pathways in *M*. *hungatei* and *S*. *fumaroxidans;* however, some more efficient pathways (based on ATP or biomass yields) were found in some cases. The computational results highlighted topics that necessitate further discussion relating to (1) the stoichiometry of transport reactions, (2) the thermodynamics of H~2~ production, and (3) the carbon and electron shuttles between species in coculture. Ion Transport in the *i*Mhu428 Metabolic Model {#sec012} ---------------------------------------------- The majority of the *i*Mhu428 model content comes from the *i*MB745 reconstruction of *M*. *acetivorans* and still needs to be verified. Changes to model content could affect the conclusions drawn about *M*. *hungatei* behavior. Despite containing a complete methanogenesis pathway, the *i*Mhu428 model was unable to identify the H^+^/Na^+^ transport stoichiometry of the energy-converting (Eha- or Ehb-type) hydrogenase (*EHA*, 1.12.7.2), which is thought to pump H^+^/Na^+^ while reducing ferredoxin \[[@pcbi.1004364.ref007]\]. The heterodisulfide reductase (*HDR*, 1.8.98.1) can also reduce ferredoxin, and the *i*Mhu428 model predicted *HDR* to be the only ferredoxin-reducing reaction required for methanogenesis. This observation is consistent with the observation that the expression of Eha/Ehb is considerably lower than that of *HDR* \[[@pcbi.1004364.ref057]\]. Additionally, different stoichiometries for other ion transport reactions important to methanogenesis remain thermodynamically possible. For example, the group contribution method predicted that tetrahydromethanopterin S-methyltransferase (*MTSPCMMT_CM5HBCMT*, E.C. 2.1.1.86) could drive transport of up to 4 Na^+^ ions under standard conditions, instead of the 2 Na^+^ ions used in the *i*Mhu428 reconstruction. Furthermore, some studies suggest the archaeal A~1~A~0~ ATP synthase is coupled to Na^+^ instead of H^+^ translocation \[[@pcbi.1004364.ref007],[@pcbi.1004364.ref058]\]. When the *i*Mhu428 model was modified to reflect this coupling, the model predicted the Na^+^/H^+^ antiporter was no longer active, as Na^+^ ions from *MTSPCMMT_CM5HBCMT* were directly used for ATP synthesis. Thus, while the modeled methanogenesis pathway is consistent with available data, it is not the only possibility. H~2~ Production in the *i*Sfu648 Metabolic Model {#sec013} ------------------------------------------------ Analysis of the *i*Sfu648 model revealed that H~2~ production is thermodynamically feasible in monoculture, implying there may be other biological reasons why H~2~ production is not normally observed under this condition, or that tighter estimates of thermodynamic parameters are needed. The *i*Sfu648 thermodynamic model also has some important limitations. In particular, the *i*Sfu648 model does not contain enough thermodynamic information to predict the directions of important electron transport reactions that involve ferredoxin, including the confurcating hydrogenase and formate dehydrogenase, the ferredoxin-oxidizing hydrogenase and formate dehydrogenase, and the RNF-type oxidoreductase (S3 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"}). This is because the group contribution method is unable to estimate the standard transformed Gibbs free energy of formation (Δ~*f*~*G*^\'0^) of ferredoxin, resulting in no Δ~*r*~*G*^\'0^ estimates for these reactions. Fortunately, new quantum chemical approaches for estimating the thermodynamics of metabolism \[[@pcbi.1004364.ref059]\] may potentially provide additional Δ~*f*~*G*^\'0^ estimates. This work also raises important questions about the appropriate mathematical basis for representing thermodynamic constraints. Previous studies used the Δ*G*^\'0^ of groups directly when modeling thermodynamics \[[@pcbi.1004364.ref034]\], and found that introducing uncertainty into a thermodynamic model of *E*. *coli* made the model computationally difficult to solve. In this work, using either the Δ*G*^\'0^ of molecules or groups to model thermodynamics proved computationally difficult (results not shown). Instead, only using Δ~*r*~*G*^\'0^ as the basis for thermodynamic calculations enabled uncertainties in free energy estimates to be handled without any computational difficulties. However, using Δ~*r*~*G*^\'0^ as a basis for thermodynamic calculations leads to larger uncertainties in Δ~*r*~*G*^\'0^ and greater network flexibility, as it does not account for thermodynamic interconnectivity between reactions with shared metabolites. As a result, the model-predicted feasible Δ~*r*~*G*^\'0^ range through a linear combination of reactions considerably exceeds the group-contribution predicted Δ~*r*~*G*^\'0^ range of the overall reaction. In the future, thermodynamic interconnectivity should be captured using the Δ*G*^\'0^ of molecules or groups, and optimization techniques are needed to improve the runtime performance of the resulting thermodynamic models. However, the use of Δ~*r*~*G*^\'0^ as a basis for thermodynamic calculations is insufficient to explain why the *i*Sfu648 model could still produce H~2~ under monoculture growth conditions. As described in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}, Δ~*f*~*G*^\'0^ was used as a basis for thermodynamic calculations when attempting to find additional constraints which would prevent H~2~ production. The failure to find such constraints indicates either that thermodynamics does not explain the absence of H~2~ production in monoculture, or that current thermodynamic models cannot capture this phenomenon. If it is the latter, more accurate group contribution methods with smaller error estimates may eventually be able to explain the role of thermodynamics in syntrophic associations. Formate and H~2~ Transfer in Coculture {#sec014} -------------------------------------- The thermodynamic coculture model of the syntrophic association between these species confirmed the role of formate and H~2~ in electron transfer in the community, and led us to hypothesize that total H~2~ consumption by the community indicates that *M*. *hungatei* cells are more abundant and/or faster growing than the *S*. *fumaroxidans* cells. The coculture model correctly predicted that both H~2~ and formate could shuttle electrons between members of this community. Formate may be preferred over H~2~ for electron transfer for thermodynamic reasons, as the Δ~*r*~*G*^\'0^ of [Eq 4](#pcbi.1004364.e005){ref-type="disp-formula"} is more favorable (less positive) than that of [Eq 3](#pcbi.1004364.e004){ref-type="disp-formula"} in which formate is not exchanged. By exchanging formate in place of CO~2~ and H~2~, *S*. *fumaxoridans* could sustain propionate oxidation at higher extracellular concentrations of formate than of H~2~. Formate exchange could also be preferred due to differences in kinetics, diffusion, and/or volatility. These scenarios could stabilize the syntrophic association by enabling faster shuttling of electrons to *M*. *hungatei*. This work highlights some important obstacles to successful modeling of microbial consortia. In order for computational models of microbial communities to make meaningful predictions, individual species models must be integrated into a community model in a biologically relevant manner. Such integration will require an understanding of the objectives and constraints governing the behavior of each community member, and this work demonstrates that identifying the proper constraints and objectives requires extensive experimental characterization of the community. For example, neither maximization of growth rate nor maximization of ATP yield were sufficient for the *i*Sfu648 model to predict the observed behavior of *S*. *fumaroxidans*. Additional constraints on reaction directions and flux ratios were required before *i*Sfu648 could be combined with *i*Mhu428 model to simulate the coculture. In addition, the *i*Sfu648 model relied on data from gene expression and ^13^C NMR experiments, suggesting that constraint-based approaches will complement traditional top down ('omics') approaches \[[@pcbi.1004364.ref060]\] by enabling a mechanistic understanding of microbial interactions \[[@pcbi.1004364.ref024]\]. Methods {#sec015} ======= Reconstruction of the *i*Mhu428 Metabolic Model {#sec016} ----------------------------------------------- The *i*Mhu428 reconstruction of *M*. *hungatei* was built from the *i*MB745 reconstruction of *M*. *acetivorans* \[[@pcbi.1004364.ref041]\]. A preliminary draft reconstruction was built based on sequence homology (using the RAVEN Toolbox \[[@pcbi.1004364.ref042]\]), but the reconstruction contained less than 200 genes ([S2 Dataset](#pcbi.1004364.s002){ref-type="supplementary-material"}). Instead of performing extensive gapfilling, reactions from the *i*MB745 *M*. *acetivorans* reconstruction were copied into the *M*. *hungatei* reconstruction, with modifications to reflect key metabolic features of *M*. *hungatei* (see [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}). Results from the RAVEN Toolbox and the KEGG SSDB \[[@pcbi.1004364.ref043]\] were used to map genes in *M*. *acetivorans* to *M*. *hungatei* and identify those reactions which have genomic evidence ([S2 Dataset](#pcbi.1004364.s002){ref-type="supplementary-material"}). Finally, blocked reactions lacking genomic evidence were removed from the reconstruction. The final *i*Mhu428 reconstruction contains 720 reactions, 428 genes (associated with 493 reactions), and 639 metabolites. Of the 428 genes, 351 were added based on sequence homology, and 77 were added manually. The reconstruction is available in [S2 Dataset](#pcbi.1004364.s002){ref-type="supplementary-material"} and [S3](#pcbi.1004364.s003){ref-type="supplementary-material"} Dataset in Excel and SBML formats. Reconstruction of the *i*Sfu648 Metabolic Model {#sec017} ----------------------------------------------- The *i*Sfu648 reconstruction of *S*. *fumaroxidans* was built from KEGG \[[@pcbi.1004364.ref043]\] ([S6 Dataset](#pcbi.1004364.s006){ref-type="supplementary-material"} and [S7 Dataset](#pcbi.1004364.s007){ref-type="supplementary-material"}) using the RAVEN Toolbox \[[@pcbi.1004364.ref042]\], which uses protein homology to identify the KEGG Orthology (KO) ID for each gene in a genome. The reactions and genes corresponding to that KO ID are then imported into the reconstruction. The resulting draft reconstruction was manually refined as described in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. The final *i*Sfu648 reconstruction contains 874 reactions, 648 genes (associated with 770 reactions), and 893 metabolites. The reconstruction is available in [S4 Dataset](#pcbi.1004364.s004){ref-type="supplementary-material"} and [S5](#pcbi.1004364.s005){ref-type="supplementary-material"} Dataset, in Excel and SBML formats. Thermodynamics-Based Metabolic Flux Analysis (TMFA) {#sec018} --------------------------------------------------- Flux-balance analysis (FBA) \[[@pcbi.1004364.ref027]\] is a constraint-based technique for predicting the state of a metabolic network consistent with physiochemical principles. FBA identifies a flux distribution which maximizes cellular growth (or some other objective function), subject to steady-state mass-balance and enzyme capacity constraints. Thermodynamics-Based Metabolic Flux Analysis (TMFA, \[[@pcbi.1004364.ref033],[@pcbi.1004364.ref034]\]) extends FBA via the introduction of thermodynamic constraints, which require that the transformed Gibbs free energy of a reaction (Δ~*r*~*G*^\'^) and its flux (*v*) have opposite signs. Estimates ($\Delta_{r}G_{est}^{'0}$) and uncertainties (${SE}_{\Delta_{r}G_{est}^{'0}}$) of Δ~*r*~*G*^\'0^ for the reactions in the reconstructions were obtained using a group contribution method \[[@pcbi.1004364.ref061]\] via the von Bertalanffy 2.0 Toolbox \[[@pcbi.1004364.ref062]\]. TMFA was implemented as previously described \[[@pcbi.1004364.ref034]\], with additional details given in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. The mol files for metabolites in *i*Mhu428 and *i*Sfu648 are provided in [S1 File](#pcbi.1004364.s008){ref-type="supplementary-material"} and [S2](#pcbi.1004364.s009){ref-type="supplementary-material"} File, respectively. Parsimonious TMFA (pTMFA) {#sec019} ------------------------- pFBA \[[@pcbi.1004364.ref054]\] is a constraint-based approach which maximizes cellular growth while also minimizing total flux through the network (a proxy for minimizing the total mass of enzymes required to sustain optimal growth through the network). pTMFA uses the same assumptions as pFBA while implementing the thermodynamic constraints of TMFA. pTMFA was implemented as a two-stage optimization process. In the first stage, growth rate is maximized via TMFA. In the second stage, the growth rate is fixed and the total flux through the network is minimized, subject to the same constraints as TMFA. Additional details on implementation can be found in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. Coculture Model {#sec020} --------------- For the coculture simulations, a community model of growth in a continuous stirred-tank reactor was developed that accounts for the biomass concentrations of each species. The model minimizes the species-weighted total flux through the metabolic networks subject to TMFA constraints for each species and mass balances around the entire reactor. Details on the specific implementation used in this work can be found in [S1 Text](#pcbi.1004364.s010){ref-type="supplementary-material"}. To avoid solving a mixed-integer non-linear program (MINLP), the dilution rate and biomass concentrations for each species were fixed, resulting in a MIP. To explore the community behavior under a variety of operating conditions, the reactor dilution rate was systematically changed, while allowing unlimited propionate uptake by the reactor. Simulations {#sec021} ----------- The uptake fluxes for carbon and other nutrients used in the simulations are given in S1 Table in [S1 Dataset](#pcbi.1004364.s001){ref-type="supplementary-material"}. All simulations were performed using CPLEX 12 (IBM, Armonk, NY) accessed via the General Algebraic Modeling System, Version 23.9.5 (GAMS, GAMS Development Corporation, Washington, DC). Estimates ($\Delta_{r}G_{est}^{'0}$) and uncertainties (${SE}_{\Delta_{r}G_{est}^{'0}}$) of thermodynamic parameters were obtained using version 2.0 of von Bertalanffy and Matlab R2012b (The MathWorks, Inc., Natick, MA). Supporting Information {#sec022} ====================== ###### Supporting tables. This file contains Supporting Tables S1 to S7 as described in the text. (XLS) ###### Click here for additional data file. ###### Excel version of the *i*Mhu428 model. (XLS) ###### Click here for additional data file. ###### SBML version of the *i*Mhu428 model. (XML) ###### Click here for additional data file. ###### Excel version of the *i*Sfu648 model. (XLS) ###### Click here for additional data file. ###### SBML version of the *i*Sfu648 model. (XML) ###### Click here for additional data file. ###### Excel version of the balanced KEGG database. (XLS) ###### Click here for additional data file. ###### SBML version of the balanced KEGG database. (XML) ###### Click here for additional data file. ###### Molfile structure files for all metabolites in the iMhu428 GEM. (ZIP) ###### Click here for additional data file. ###### Molfile structure files for all metabolites in the iSfu648 GEM. (ZIP) ###### Click here for additional data file. ###### Supporting text. This file contains Supporting Methods, Results, and Figs to complement the main text. (PDF) ###### Click here for additional data file. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JJH JLR. Performed the experiments: JJH MCC. Analyzed the data: JJH JLR. Wrote the paper: JJH JLR.

Pancreatic adenocarcinoma is a common cancer with extremely poor prognosis. In the past few years, gemcitabine, a novel pyrimidine nucleoside analogue, has become the standard chemotherapeutic agent used in patients with pancreatic cancer ([Rothenberg *et al*, 1996](#bib24){ref-type="other"}; [Burris *et al*, 1997](#bib3){ref-type="other"}). However, even with this drug, most pancreatic adenocarcinomas exhibit an inexorable disease progression, and the clinical impact of gemcitabine remains modest owing to a high degree of inherent and acquired chemoresistance ([Carmichael *et al*, 1996](#bib4){ref-type="other"}; [Li *et al*, 2004](#bib17){ref-type="other"}). No clinical molecular markers have previously been shown that can predict a benefit from gemcitabine chemotherapy, and patients are treated empirically until disease progression or worsening therapy performance. Therefore, it is extremely important to determine predictive molecular markers of gemcitabine resistance for more effective treatment of these tumours. Gemcitabine has a complex pathway of metabolism, and there are many mechanisms that can contribute to gemcitabine cytotoxicity and/or chemoresistance ([Habiro *et al*, 2004](#bib12){ref-type="other"}; [Koizumi *et al*, 2005](#bib16){ref-type="other"}). In recent studies performed on human cancer cell lines, human equilibrative nucleoside transporter-1 (hENT1) was found to be the major gemcitabine transporter ([Garcia-Manteiga *et al*, 2003](#bib10){ref-type="other"}). If gemcitabine is not transported into the cell via hENT1 it cannot inhibit cell growth ([Mackey *et al*, 1998](#bib18){ref-type="other"}; [Rauchwerger *et al*, 2000](#bib22){ref-type="other"}), but increased hENT1 abundance facilitates efficient cellular entry of gemcitabine and confers increased cytotoxicity ([Mackey *et al*, 1999](#bib19){ref-type="other"}; [Ritzel *et al*, 2001](#bib23){ref-type="other"}). Inside the cell, gemcitabine is phosphorylated by deoxycytidine kinase (dCK) in a rate-limiting step. Deficiency in dCK activity has been considered to be one of the main mechanisms responsible for the development of resistance to gemcitabine. Another factor in gemcitabine resistance is the overexpression of ribonucleotide reductase (RR). Ribonucleotide reductase is mainly responsible for the conversion of ribonucleosides to deoxyribonucleoside triphosphates (dNTPs), which are essential for DNA polymerisation and repair ([Cory and Sato, 1983](#bib5){ref-type="other"}; [Thelander and Berg, 1986](#bib27){ref-type="other"}; [Fan *et al*, 1996](#bib7){ref-type="other"}; [Zhou *et al*, 1998](#bib29){ref-type="other"}). RR consists of the dimerised large and small subunits, M1 and M2, respectively. The M1 subunit possesses a binding site for enzyme regulation (regulatory subunit), and the M2 subunit is involved with RR activity (catalytic subunit). Although it has been proposed that the genes for gemcitabine transport and metabolism are involved in the mechanism of cellular resistance to gemcitabine, it is not fully understood how gemcitabine influences its own transport and metabolism in the process of acquired resistance. Understanding alterations in expression of genes, which characterise the response of cancer cells to gemcitabine treatment in the process of acquired resistance, would allow us to improve therapeutic strategies for pancreatic cancer. In this study, we established various gemcitabine-resistant subclones of human pancreatic cancer cell lines, and investigated changes in gene expression associated with gemcitabine transport and metabolism, that is, *hENT1*, *dCK*, *RRM1*, and *RRM2* mRNA, in the development of gemcitabine resistance. Quantitative RT--PCR analysis showed that the balance of the four gene expression levels is associated with inherent and acquired resistance to gemcitabine in pancreatic cancer cells. Resistance of cancer cells to gemcitabine is determined by the ratio of these gene expression levels, but not predicted by that of a single gene. The ratio of gene expression might be useful as a predictive marker for the efficacy of gemcitabine therapy in pancreatic cancer patients. MATERIALS AND METHODS ===================== Chemicals --------- Gemcitabine was a gift from Eli Lilly Pharmaceuticals (Indianapolis, IN, USA). All other chemicals were of analytical grade and commercially available. Cell culture and establishment of gemcitabine-resistant pancreatic cancer cells ------------------------------------------------------------------------------- Eight human pancreatic adenocarcinoma cell lines, PK1, PCI43, KLM1, PK8, PK9, MIAPaCa2, KP1N, and BxPC3, were used in this study. PK1, KLM1, PK8, and PK9 cell lines were obtained from the Cell Resource Center for Biochemical Research (Tohoku University, Sendai, Japan). KP1N and MIAPaCa2 cell lines were purchased from the Health Science Research Resources Bank (Osaka, Japan). The PCI43 cell line was provided by Dr H Ishikura at Hokkaido University (Sapporo, Japan). PK1, PCI43, KLM1, PK8, PK9, KP1N, and BxPC3 were grown in RPMI 1640 media (Gibco, Paisley, Scotland) supplemented with 10% heat-inactivated fetal bovine serum in a humidified 5% CO~2~ incubator at 37°C. The MIAPaCa2 cell line was cultured in Dulbecco\'s modified Eagles medium (DMEM). Gemcitabine-resistant cells were generated by exposing the PCI43, PK1, and KLM1 cell lines to incrementally increasing gemcitabine concentrations starting at 3 n[M]{.smallcaps}. As the cells adapted to the drug, the gemcitabine concentration was doubled. The intermediate resistant variants were cultured for at least 4 weeks. The cell lines were named as follows: G for gemcitabine, followed by the n[M]{.smallcaps} concentration at which the cell line grew logarithmically. The most resistant variants were PCI43-G4000, PK1-G4000, and KLM1-G4000 and were resistant to continuous exposure to gemcitabine at 4000 n[M]{.smallcaps}. Experiments were performed using cells in the exponential phase of growth. Drug cytotoxicity assay ----------------------- The relative cytotoxicity of gemcitabine in each cell line was assessed with a WST-1 assay using a Cell Counting Kit (Dojindo Laboratories, Kumamoto, Japan). This assay is based on the reduction of a tetrazolium compound to a soluble derivative by the dehydrogenase enzymes of metabolically active cells. The absorbance (450 nm) is directly proportional to the number of living cells in culture. Cells were added to 96-well tissue culture plates (3 × 10^3^ cells/well) overnight and exposed to increasing concentrations (10^−2^ 10^3^ *μ*[M]{.smallcaps}) of gemcitabine for 72 h, after which the number of remaining living cells was determined according to the manufacturer\'s instructions. Chemosensitivity was expressed as the drug concentration that inhibited cell proliferation by 50% (IC~50~ values) and was determined from concentration--effect relationship. Quantitative LightCycler RT--PCR -------------------------------- Total RNA was extracted from each cell line and gemcitabine-resistant subclones using the RNeasy Protect Mini Kit (Qiagen, Hilden, Germany) with the RNase-Free DNase Set (Qiagen) according to the manufacturer\'s instructions. Complementary DNA (cDNA) was produced from 1 *μ*g of RNA using an Oligo (dT)~12−18~ Primer (Invitrogen, Carlsbad, CA, USA) and MMLV reverse transcriptase (Promega, Madison, WI, USA). Primers for *hENT1*, *dCK*, *RRM1*, *RRM2*, and *glyceraldehyde-3-phosphate dehydrogenase (GAPDH)* were based on the sequence of each gene (Entrez-PubMed) and designed by the program Primer 3. Oligonucleotides used as PCR primers are summarised in [Table 1](#tbl1){ref-type="table"}. Quantitative RT--PCR was performed in a LightCycler system (Roche Molecular Biochemicals, Mannheim, Germany) using SYBR Green fluorescence. In this system, all reactions were run in glass capillaries with a total volume of 20 *μ*l. The reaction mixture consisted of 2 *μ*l of FastStart DNA Master SYBR Green I, SYBR Green I dye, and 10m[M]{.smallcaps} MgCl~2~ (Roche Diagnostics GmbH, Mannheim, Germany). Primers were added to a final concentration of 3 4 *μ*[M]{.smallcaps}. In each experiment, 1 *μ*g of extracted RNA from the cells was reverse transcribed to generate cDNAs that were then diluted 1:10. Finally, 5 *μ*l of the diluted cDNA was added to a capillary tube. One positive control, one negative control, and standards were included in each run. The PCR programs started with a preincubation step for activation of the FastStart enzyme, then continued with amplification, and ended with melting curve analysis. The temperature transition rate was 0.1°C s^−1^. The preincubation and amplification programmes for each line were as follows: *hENT1*, preincubation at 95°C for 10 min and amplification with 40 cycles of 95°C for 10 s, 60°C for 10 s, and 72°C for 10 s; *dCK*, preincubation at 95°C for 10 min and amplification with 40 cycles of 95°C for 10 s, 56°C for 10 s, and 72°C for 17 s; *RRM1*, preincubation at 95°C for 10 min and amplification with 40 cycles of 95°C for 10 s, 58°C for 10 s, and 72°C for 10 s; *RRM2*, preincubation at 95°C for 10 min and amplification with 40 cycles of 95°C for 10 s, 60°C for 10 s, and 72°C for 6 s; *GAPDH*, preincubation at 95°C for 10 min and amplification with 40 cycles of 95°C for 10 s, 55°C for 10 s, and 72°C for 13 s. All programs were then followed by a heating step at 95°C and a cooling step at 65°C for 15 s each. Each experiment using quantitative RT--PCR was performed in triplicate. The expression of *hENT1*, *dCK*, *RRM1*, and *RRM2* mRNA was quantified relative to *GAPDH* expression. The Roche software uses the second derivative maximum method to calculate the fractional cycle numbers where the fluorescence rises above background (crossing point, *C*~p~), that is, the point at which the rate of change of fluorescence is fastest. For the standard curve, CpS are plotted vs log concentration for the standards. This standard curve is used to estimate the concentration of each sample. The standard curves were saved in a coefficient file that was used by the relative quantification software from Roche to calculate the mRNA levels relative to *GAPDH*. This program also corrected for the differences in efficiency of the PCR reaction for each target. Statistical analysis -------------------- Data are expressed as means±s.d. Analysis was performed using the Mann--Whitney *U*-test (two-tailed) for nonparametric data. Correlations between nonparametric data were analysed by the Spearman correlation test. RESULTS ======= Development of acquired gemcitabine resistance in pancreatic cancer cell lines, PCI43, PK1, and KLM1 ---------------------------------------------------------------------------------------------------- Resistance to gemcitabine was successfully induced in the PCI43, PK1, and KLM1 cell lines via exposure to stepwise increases in gemcitabine concentration. Gemcitabine-resistant PCI43, PK1, and KLM1 cells were developed by continuous exposure to increasing concentrations of gemcitabine over a period of 6 months, starting with an initial concentration of 3 n[M]{.smallcaps}. The resistant cells obtained, PCI43-G4000, PK1-G4000, and KLM1-G4000, were viable in medium containing 4000 n[M]{.smallcaps} gemcitabine. Resistant cells showed no apparent morphologic differences or difference in growth rate compared with the parental cells. The IC~50~ values and resistance ratios for parental cells and gemcitabine-resistant cells are listed in [Table 2](#tbl2){ref-type="table"}. The IC~50~ values of gemcitabine for the parental PCI43, PK1, and KLM1 cells were 350, 160, and 80 n[M]{.smallcaps}, respectively. PCI43-G4000, PK1-G4000, and KLM1-G4000 cells were 157-, 625-, and 2625-fold less sensitive to gemcitabine than the parental cell lines, respectively. Analysis of *hENT1*, *dCK*, *RRM1*, and *RRM2* mRNA expression by real-time light cycler-PCR in different gemcitabine-resistant cell lines ------------------------------------------------------------------------------------------------------------------------------------------ To evaluate the expression of *hENT1*, *dCK*, *RRM1*, and *RRM2* mRNA, real-time light cycler-PCR was performed in a quantitative manner. The mRNA was extracted from each gemcitabine-resistant cell line and analysed via light cycler. Glyceraldehyde-3-phosphate dehydrogenase was used as an internal control. The quantitative data were summarised from the means of the data gathered from the three experiments ([Figure 1](#fig1){ref-type="fig"}). Real-time light cycler-PCR demonstrated that PCI43-G4000 cells had approximately 4- and 10-fold increases in the levels of *RRM1* and *RRM2* mRNA compared with the parental cells, respectively. PK1-G4000 cells had approximately a two-fold increase in levels of *RRM1* mRNA, but there was no increase of *RRM2* mRNA compared with parental cells. Similarly, KLM1-G4000 cells had approximately a 42-fold increase in levels of *RRM1* mRNA, but no increase of *RRM2* mRNA compared with parental cells. *Human equilibrative nucleoside transporter*-1 gene expression was significantly increased compared with parental cells in PCI43-G4000 and KLM1-G4000 cells, but not in PK1-G4000 cells. Expression of *dCK* mRNA was not detected by real-time light cycler-PCR in either PCI43-G4000 or PK1-G4000 cells, and no changes in *dCK* mRNA expression were observed in KLM1-G4000 cells. To determine the cellular modification responsible for *dCK* mRNA downregulation, seven exons of the *dCK* gene as well as the 5′-untranslated regions were amplified by PCR as described previously ([Galmarini *et al*, 2004](#bib8){ref-type="other"}). In PCR products obtained using genomic DNA from cells, as shown in [Figure 2](#fig2){ref-type="fig"}, a partial deletion of the *dCK* gene was amplified in both PCI43-G4000 and PK1-G4000 cells. No deletion of the *dCK* gene was detected in parental cell lines or KLM1-G4000 cells. Gene expression changes in the process of acquired gemcitabine resistance by real-time light cycler-PCR in pancreatic cancer cells ---------------------------------------------------------------------------------------------------------------------------------- It is not fully understood how gemcitabine influences its own transport and metabolism in the development of gemcitabine resistance. To clarify the changes of gene expression in the process of acquired resistance, we established seven gemcitabine-resistant subclones of the PCI43, PK1, and KLM1 cell lines. Examination of the gemcitabine-resistant PCI43 subclones revealed a proportional increase in *RRM2* message with increasing gemcitabine exposure ([Figure 3](#fig3){ref-type="fig"}). *RRM1* mRNA showed a two- to four-fold up regulation of expression levels. Significant increases in *RRM1* and *RRM2* gene expression induced by gemcitabine were detected in the subclones PCI43-G30 and -G3, respectively. No *dCK* mRNA could be detected in subclones PCI43-G3000 and -G4000, which had higher resistance to gemcitabine, but the *dCK* gene was slightly increased at lower concentrations of gemcitabine. The *hENT1* gene was slightly increased in the latter phase of development of acquired resistance. In the gemcitabine-resistant PK1 subclones, a slight increase in *RRM1* mRNA was detected, but no increase in the *RRM2* gene was observed. No *dCK* mRNA could be detected in subclones PK1-G300 to -G4000. The gemcitabine-resistant KLM1 subclones showed marked upregulation in *RRM1* mRNA, but there was no increase in the *RRM2* gene. *dCK* mRNA could be detected in all KLM1 subclones. These results suggest that acquired resistance of pancreatic cancer cells to gemcitabine may be determined by the balance of these four factors, but is not predicted by that of a single gene. Decreased *hENT1* or *dCK* has been reported to promote gemcitabine resistance ([Ruiz van Haperen *et al*, 1994](#bib25){ref-type="other"}; [van der Wilt *et al*, 2000](#bib28){ref-type="other"}; [Garcia-Manteiga *et al*, 2003](#bib10){ref-type="other"}; [Jordheim *et al*, 2003](#bib13){ref-type="other"}; [Galmarini *et al*, 2004](#bib8){ref-type="other"}). In contrast, increased expression of RR has been reported to be associated with gemcitabine resistance in human tumour cells ([Goan *et al*, 1999](#bib11){ref-type="other"}; [Jung *et al*, 2001](#bib14){ref-type="other"}). For these reasons, we calculated the ratio of *hENT1* × *dCK*/*RRM1* × *RRM2* gene expression in gemcitabine-resistant subclones. As shown in [Figure 4](#fig4){ref-type="fig"}, the ratio of *hENT1* × *dCK*/*RRM1* × *RRM2* expression progressively decreased in the process of acquired gemcitabine resistance. Correlation of the hENT1 × dCK/RRM1 × RRM2 ratio with gemcitabine chemosensitivity in pancreatic cancer cell lines ------------------------------------------------------------------------------------------------------------------ To determine whether the *hENT1* × *dCK*/*RRM1* × *RRM2* ratio correlated with gemcitabine sensitivity, we examined the relative mRNA expression of *hENT1*, *dCK*, *RRM1*, and *RRM2* to *GAPDH* in eight human pancreatic cancer cell lines. Correlations of IC~50~ values and relative levels of gene expression in each cell lines are summarised in [Table 3](#tbl3){ref-type="table"}. We found that the IC~50~ values of gemcitabine did not significantly correlate with relative expression levels of *hENT1*, *dCK*, *RRM1*, or *RRM2*. Next, the ratio of *hENT1* × *dCK*/*RRM1* × *RRM2* expression was calculated and the correlation with gemcitabine sensitivity was determined ([Figure 5](#fig5){ref-type="fig"}). Cells with a higher *hENT1* × *dCK*/*RRM1* × *RRM2* expression ratio showed higher gemcitabine chemosensitivity, whereas cells with a lower ratio showed higher chemoresistance. The *hENT1* × *dCK*/*RRM1* × *RRM2* ratio significantly correlated with the gemcitabine IC~50~ in eight pancreatic cancer cell lines (*P*=0.0029). Neither the *RRM1* × *RRM2 ratio*, the *hENT1*/*RRM1* × *RRM2* ratio, nor the *dCK*/*RRM1* × *RRM2* ratio correlated with sensitivity to gemcitabine. DISCUSSION ========== Chemoresistance is a major cause of pancreatic adenocarcinoma treatment failure with gemcitabine. The majority of patients with gemcitabine-treated pancreatic adenocarcinoma become resistant after consecutive treatments, and fail to derive benefit from chemotherapy. Therefore, it is extremely important to clarify the mechanism behind chemoresistance and to identify predictive markers of inherent and acquired chemoresistance to gemcitabine for better treatment of these tumours. It has been shown that modulation of cellular enzymes of gemcitabine transport and metabolism influences drug activity *in vitro* ([Mackey *et al*, 1998](#bib18){ref-type="other"}, [1999](#bib19){ref-type="other"}; [Goan *et al*, 1999](#bib11){ref-type="other"}; [Ritzel *et al*, 2001](#bib23){ref-type="other"}). Cellular enzymes of gemcitabine transport and metabolism, that is, hENT1, dCK, RRM1, and RRM2, are well documented. Moreover, experimental data may improve the success of pancreatic cancer treatment either by selecting responsive patients or by modulation of gemcitabine effect with rationally selected drug combinations. For these reasons, this study addressed the transcription analysis of *hENT1*, *dCK*, *RRM1*, and *RRM2* in parental cell lines and in isolated variable resistant subclones continuously exposed to gemcitabine to determine possible predictive markers for gemcitabine resistance. Our data demonstrate for the first time that acquired and inherent chemoresistance of pancreatic cancer cells to gemcitabine is determined by the balance of *dCK, RRM1, RRM2*, and *hENT1* gene expression, but not to that of any of the individual genes. The *hENT1* × *dCK*/*RRM1* × *RRM2* expression ratio significantly correlates with resistance to gemcitabine in pancreatic cancer cells, including acquired gemcitabine-resistant cells, suggesting that a decrease of this ratio reflects inherent and acquired chemoresistance of pancreatic cancer cells to gemcitabine and may be a key to understanding the variable effectiveness of gemcitabine among individual patients. The expression ratio is a novel, informative marker for predicting and monitoring the responses of pancreatic cancer patients to gemcitabine. A recent study performed on cultured cancer cell lines indicated that hENT1 is the major gemcitabine transporter in human pancreatic cancer cells ([Garcia-Manteiga *et al*, 2003](#bib10){ref-type="other"}). It is speculated that populations of cells with lower hENT1 abundance may be relatively gemcitabine resistant owing to reduced intracellular accumulation. In fact, pharmacological inhibition of hENT1 in cells has been reported to render them gemcitabine resistant ([Mackey *et al*, 1998](#bib18){ref-type="other"}). However, the current study shows that expression of the *hENT1* gene was not reduced in the development of gemcitabine resistance, and did not correlate with IC~50~ values of gemcitabine in eight pancreatic cancer cell lines. These data suggest that *hENT1* expression alone does not reflect inherent and acquired resistance to gemcitabine. The increase in *hENT1* expression in the PK1-G4000, PCI-G4000, and KLM1-G4000 subclones may be regarded as a compensatory adaptation to higher chemoresistance to gemcitabine. Previous reports in cells and animal models cited *dCK* mutation/deficiency as the main mechanism for gemcitabine resistance in cells with an acquired resistance ([Ruiz van Haperen *et al*, 1994](#bib25){ref-type="other"}; [van der Wilt *et al*, 2000](#bib28){ref-type="other"}; [Galmarini *et al*, 2001](#bib9){ref-type="other"}; [Jordheim *et al*, 2003](#bib13){ref-type="other"}). In this study, expression of the *dCK* gene was slightly increased in subclones that were resistant to lower concentrations of gemcitabine, and was undetectable in subclones with high resistance to gemcitabine, that is, PCI43-G4000 and PK1-G4000. Furthermore, we found that expression of the *dCK* gene alone does not correlate with sensitivity to gemcitabine in eight pancreatic cancer cell lines. Deficiency of dCK described in previous reports was mainly based on studies performed on highly gemcitabine-resistant clones, and quite different from the clinical setting. Our data indicate that dCK deficiency is involved in a higher grade of acquired gemcitabine resistance, but not in a lower grade of resistant cells or in parental cells. Expression of RR has been reported to be one of the determinants of gemcitabine chemoresistance in human tumour cells ([Jung *et al*, 2001](#bib14){ref-type="other"}). In fact, artificial overexpression of RRM2 results in a further increase in gemcitabine chemoresistance ([Goan *et al*, 1999](#bib11){ref-type="other"}). An increased level of RR expands the size of the dNTP pools, which competitively inhibits the incorporation of gemcitabine triphosphate into DNA ([Plunkett *et al*, 1996](#bib21){ref-type="other"}). The expanded dNTP pools further downregulate the activity of dCK via a negative-feedback pathway. Results from this study demonstrate that *RRM1* mRNA expression was markedly increased in the KLM1 gemcitabine-resistant cells over that of the parental cells. This result supports the previous report that RRM1 is the marker predicting resistance to gemcitabine in lung cancer cell lines ([Davidson *et al*, 2004](#bib6){ref-type="other"}). In this study, KLM1 gemcitabine-resistant cells had no *dCK* deficiency in contrast to the lack of *dCK* in PCI43 and PK1 gemcitabine-resistant cells. Therefore, these data demonstrate that RRM1 may correlate with acquired resistance to gemcitabine, especially in cells without dCK deficiency. *RRM2* mRNA was increased in PCI43 gemcitabine-resistant cells, but not in PK1 and KLM1 gemcitabine-resistant subclones. In the eight pancreatic carcinoma cells tested, the cells with higher inherent resistance to gemcitabine did not show higher levels of *RRM1* and *RRM2* expression. These results suggest that an increase in either *RRM1* or *RRM2* expression alone correlates with acquired chemoresistance, but does not reflect inherent chemoresistance to gemcitabine. In the development of gemcitabine resistance, increases in either *RRM1* or *RRM2* expression, in addition to a decrease in *dCK*, may indicate that cells are regulated to avoid damage by gemcitabine incorporation in gemcitabine-resistant cells. Ribonucleotide reductase plays a role as a central enzyme controlling the rate of dNTP synthesis ([Plunkett *et al*, 1996](#bib21){ref-type="other"}). It has recently been reported that the physiological factor for ribonucleotide reduction is thioredoxin, which is required for the RR reaction *in vitro* ([Koc *et al*, 2006](#bib15){ref-type="other"}). In yeast, mutants lacking thioredoxin had significantly lower dNTP levels, supporting the idea that thioredoxin functions as an RR reductant *in vitro*. Interestingly, thioredoxin is identified as a gene whose basal expression is increased in pancreatic cancer cells in which Smad7 is commonly overexpressed ([Arnold *et al*, 2004](#bib2){ref-type="other"}). Thioredoxin is downstream of smad7 in a pathway that acts to promote growth and induce apoptosis resistance in pancreatic cancer cells. Future studies are necessary to investigate whether or not thioredoxin is downregulated in gemcitabine-resistant pancreatic cancer cells. Previous genomic analysis of pancreatic cancer has been performed exclusively with surgical and autopsy specimens, owing to the difficulty of tissue sampling without surgery. To overcome this difficulty, endoscopic ultrasonography-guided fine-needle aspiration biopsy (EUS-FNAB) is applied to obtain tumour cells as an effective and safe method for tissue diagnosis of pancreatic cancer ([Agarwal *et al*, 2004](#bib1){ref-type="other"}). More recently, several investigators have demonstrated that genetic analysis using EUS-FNAB specimens is possible for determination of cancer stage or to improve the accuracy of diagnosis ([Tada *et al*, 2002](#bib26){ref-type="other"}; [Pellise *et al*, 2004](#bib20){ref-type="other"}). The current study shows that detection and quantitation of *hENT1*, *dCK*, *RRM1*, and *RRM2* mRNA can be performed rapidly and reliably using light cycler-PCR with SYBR Green fluorescence. The analytical strategy using light cycler-PCR with EUS-FNAB specimens will enable us to evaluate the chemoresistance of pancreatic cancer before and after gemcitabine treatment in the clinic. In summary, quantitative RT--PCR analysis showed that the balance of four gene expression levels correlates with inherent and acquired resistance to gemcitabine in pancreatic cancer cells. We propose that resistance of cancer cells to gemcitabine is determined by the ratio of expression of these genes in pancreatic cancer cells. The ratio of gene expression may be a useful marker for predicting and monitoring the efficacy of gemcitabine therapy in pancreatic cancer patients. Further studies are currently underway using pancreatic cancer tissues obtained by EUS-FNAB to elucidate the importance of the expression ratio in gemcitabine therapy for pancreatic cancer patients. This research was supported in part by a Grant-in-Aid for Scientific Research provided by the Ministry of Education, Culture, Sports, Science and Technology, Japan. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} ###### Sequences of primers used in reverse transcription--PCR **Gene** **Forward primer (5′--3′)** **Reverse primer (5′--3′)** ---------- ----------------------------- ----------------------------- hENT1 AAAGGAGAGGAGCCAAGAGC GGCCCAACCAGTCAAAGATA dCK CCCGCATCAAGAAAATCTCC TCATCCAGTCATGCCAGTC RRM1 GGAGGAATTGGTGTTGCTGT GCTGCTCTTCCTTTCCTGTG RRM2 CCCGCTGTTTCTATGGCTTC CCCAGTCTGCCTTCTTCTTG GAPDH ATGACCACAGTCCATGCCAT TTGAAGTCAGAGGAGACCAC Abbreviations: dCK=deoxycytidine kinase; GAPDH=glyceraldehyde-3-phosphate dehydrogenase; hENT1=human equilibrative nucleoside transporter-1. ###### IC~50~ values of gemcitabine in parental and gemcitabine-resistant cells of PCI43 and PK1 cell lines **Cell lines** **IC~50~ (n[M]{.smallcaps}) of GEM** **Resistant ratio** ---------------- -------------------------------------- --------------------- PCI43 350   PCI43--G4000 55 000 157 PK1 160   PK1--G4000 100 000 625 KLM1 80   KLM1--G4000 210 000 2625 ###### Relative hENT1, dCK, RRM1, and RRM2 mRNA expression to GAPDH in human pancreatic cancer cell lines     **Relative mRNA expression to GAPDH** --------------------- ----- --------------------------------------- ------------ ----------- ----------- PCI43 350 1.00±0.12 0.36±0.0.3 0.43±0.02 0.50±0.04 PK1 160 1.00±0.13 1.00±0.16 1.00±0.13 1.00±0.21 KLM1 80 3.88±0.22 0.81±0.02 1.08±0.04 1.46±0.13 PK9 70 1.12±0.17 0.99±0.05 0.31±0.02 0.63±0.06 PK8 40 3.45±0.28 2.51±0.07 0.71±0.04 1.82±0.19 MIAPaCa2 40 2.64±0.42 0.42±0.08 0.42±0.07 0.47±0.04 KPIN 40 23.26±1.33 2.59±0.04 1.58±0.02 4.31±0.41 BxPC3 30 3.52±0.51 1.20±0.21 0.70±0.08 0.84±0.16 *P*-value to IC~50~   NS NS NS NS NS: not significant. Abbreviations: dCK=deoxycytidine kinase; GAPDH=glyceraldehyde-3-phosphate dehydrogenase; hENT1=human equilibrative nucleoside transporter-1.

Introduction ============ Research in quality of life traditionally relies on paper-and-pencil questionnaires. Internet surveys may have advantages compared to the traditional paper-and-pencil surveys with respect to turn-a-round time, expenses, and data management \[[@ref1]\]. However, Internet-based data collection may differ from traditional methods with respect to response rate and data quality as well as validity and reliability of the involved scales. Only a few studies have systematically evaluated Internet-based survey methods. The main questions have addressed validity \[[@ref2]-[@ref7]\], response rate, response speed, and completeness of data \[[@ref1],[@ref6]-[@ref14]\]. Most studies report small differences in answers obtained in Internet and paper-and-pencil versions of questionnaires \[[@ref2]-[@ref7]\]. Pealer et al found no significant difference in response rates, the Internet version having a response rate of 62% compared to 58% for the paper-and-pencil version \[[@ref12]\]. Ritter et al found a high response rate in both groups of a study population recruited on the Internet: 87% in the Internet group, and 83% in the paper-and-pencil group \[[@ref9]\]. These studies either recruited their participants on the Internet or invited only participants with a known active email account, and, as a consequence, the results from these studies are not valid for a general population of patients. A Swedish study conducted in a general population sample obtained a response rate of 50% in the Internet group and 64% in the paper-and-pencil group. The method included two reminders, including a contact by telephone \[[@ref6]\]. However, in a workplace health survey, a poor response rate was observed among the Internet group (19%) compared to the paper-and-pencil group (72%), but this study did not include a reminder procedure \[[@ref1]\]. Overall, the results with respect to differences in response rate are inconsistent, which may reflect differences in methodology and populations. We have not identified any randomized studies comparing Internet and paper-and-pencil questionnaires in patient populations unselected with respect to Internet access. Therefore, we aimed to evaluate an Internet survey method in comparison to paper-and-pencil with respect to response rate and completeness of data in a randomized controlled design among women referred for mammography. Methods ======= Participants were women referred for mammography from September 2004 to April 2005 in the Department of Radiology at the public hospital, Randers Regional Hospital. The municipality of Randers has around 62000 inhabitants. Patients were referred by their family doctor. A consultant at the Department of Radiology assigned the referred patients to one of three categories: acute, subacute, or nonacute. Subsequently, a letter was sent to the woman informing her about the date, location, and other details of the mammography. The women were randomized to be invited to answer either an Internet version or a paper-and-pencil version of a questionnaire. We only invited women up to retirement age (67 years in Demark) who did not have a history of breast cancer. Patients from all categories (acute, subacute, or nonacute) were invited until February 2005, whereafter only patients in the acute and subacute groups were invited to participate. Nonrespondents in both groups were mailed a reminder after 10 days, given that the date of their mammography was not reached. The reminder informed the woman that she was free to answer the opposite version of the originally requested questionnaire if she so desired. Only questionnaires filled in before the date of the mammography were included in the analysis. The procedure is outlined in [Figure 1](#figure1){ref-type="fig"}. There were no incentives to promote the survey response. Figure 1Flow of participants through the randomized trial The letter to women randomized to answer the paper-and-pencil version included a paper questionnaire and a prepaid return envelope, while the letter to women randomized to answer the Internet version included a guideline on how answer the Web-based version. Access to the Internet questionnaire required entry of a unique five-letter username. No password was needed since the first letter in the username was a redundancy code. The layout of the Internet version was as close to the paper version as possible (see [Figure 2](#figure2){ref-type="fig"} and [Figure 3](#figure3){ref-type="fig"}). In the Internet version, the participants were reminded of missing answers if they tried to leave a page incomplete. However, after pressing an "OK" button, they were allowed to continue even if there were still missing answers \[[@ref15]\]. The questionnaire consisted of 17 pages and 119 items and included Short Form-36 \[[@ref16]\], Multidimensional Fatigue Inventory-20 \[[@ref17]\], and The Hospital Anxiety and Depression Scale \[[@ref18]\]. Questions regarding social status, education level, occupation, and access to the Internet were also asked. All respondents were interviewed by telephone 1 month after they had their mammogram. They were invited to join a follow-up study and were asked to select the version of questionnaire they preferred. Figure 2Screenshot of the Internet version of the questionnaire Figure 3Photograph of the paper-and-pencil version of the questionnaire The sample size was calculated to provide a statistical power of at least 90% to detect a true difference in response rate of 15%. The actual power was 93.8%. Women had an equal probability of assignment to the two groups. The randomization code was developed using a computer random number generator. We tested the significance of categorized variables by the chi-square test and compared continuous variables by risk differences with 95% confidence intervals. Homogeneity across strata was tested with the Mantel-Haenszel test. Results ======= The characteristics of the invited women are shown in [Table 1](#table1){ref-type="table"}. Approximately 80% of the women were between 30 and 59 years old. The distributions within the two randomized groups were similar with respect to age, place of residence, and category of referral. ###### Characteristics of patients, by randomization group ------------------------------------------------------------------------------- Internet Group, %\ Paper-and-Pencil Group, %\ (n = 257) (n = 276) ----------------------------- -------------------- ---------------------------- **Age (years)** 20-29 5.1 7.4 30-39 25.0 21.0 40-49 29.7 29.6 50-59 26.8 27.6 60-67 13.4 14.4 **Place of Residence^\*^** Rural 60.1 53.3 Village/suburb 26.5 27.2 Urban 13.4 19.5 **Category of Referral^†^** Acute 50.0 47.9 Subacute 18.5 20.6 Nonacute 31.5 31.5 ------------------------------------------------------------------------------- ^\*^Defined by postal code ^†^The acute group was called in for mammography within 3-14 days, the subacute group within 1 to 3 weeks, and the nonacute group, not before 5 months. The response rate before the reminder was 17.9% in the Internet group compared to 73.2% in the paper-and-pencil group, corresponding to a 55.3% difference in response rate in favor of the paper-and-pencil version ([Table 2](#table2){ref-type="table"}). The same tendency was found in all strata with respect to age, place of residence, and category of referral (see [Table 2](#table2){ref-type="table"}). After the reminder, the response rate improved distinctly in the group originally randomized to the Internet ([Table 3](#table3){ref-type="table"}). Among the women assigned to the nonacute group, who had the longest respite before their mammogram, the response rate was even higher in the group randomized to the Internet version. The completeness of answers in the two versions is summarized in [Table 4](#table4){ref-type="table"}. The Internet version produced significantly more complete questionnaires than the paper-and-pencil version. For the paper-and-pencil version, there was a tendency toward more incomplete scales the longer the scales were. ###### Response rate before reminder, by randomization group ------------------------------------------------------------------------------------------------------------ Internet Group, %\ Paper-and-Pencil Group, %\ Difference, % (95% CI)^\*^ (n = 257) (n = 276) ----------------------------- -------------------- ---------------------------- ---------------------------- **Total** 17.9 73.2 55.3 (48.3-62.3) **Age (years)** 20-29 10.5 85.7 75.2 (52.2-98.1) 30-39 16.7 72.5 55.8 (41.3-70.3) 40-49 23.7 68.3 44.6 (30.7-58.5) 50-59 18.3 75.7 57.4 (44.1-70.7) 60-67 10.8 75.7 64.9 (47.8-81.9) *χ*^2^~4~ = 4.8, *P* = .30 **Place of Residence^†^** Rural 17.5 74.7 57.2 (48.0-66.49) Village/suburb 22.9 71.2 48.4 (34.1-62.7) Urban 12.0 70.3 58.3 (41.0-75.5) *χ*^2^~2~ = 2.2, *P* = .33 **Category of Referral^‡^** Acute 23.9 72.5 48.9 (38.3-59.5) Subacute 13.2 70.6 57.4 (41.9-72.9) Nonacute 12.4 75.9 63.5 (52.0-75.0) *χ*^2^~2~= 5.1, *P* = .08 ------------------------------------------------------------------------------------------------------------ ^\*^With Mantel-Haenszel test of homogeneity ^†^Defined by postal code ^‡^The acute group was called in for mammography within 3-14 days, the subacute group within 1 to 3 weeks, and the nonacute group, not before 5 months. ###### Response rate after reminder, by randomization group -------------------------------------------------------------------------------------------------------------------------- Internet Group\ Paper-and-Pencil Group\ Difference in %\ (n = 257) (n = 276) (95% CI) ----------------------------- ----------------- ------------------------- ------------------ ---------- ------------------ **Total** 64.2 75 (159) 76.5 59 (44) 12.2 (4.5-20.0) **Category of Referral**^†^ Acute 49.6 58 (55) 74.6 39 (15) 25.0 (13.6-36.5) Subacute 67.9 76 (35) 76.5 53 (8) 8.5 (−8.6-25.7) Nonacute 84.0 97 (69) 79.3 100 (21) −4.6 (−16.3-7.0) -------------------------------------------------------------------------------------------------------------------------- ^\*^The percentage of primary nonrespondents who were reminded ^†^The acute group was called in within 3-14 days, the subacute group within 1 to 3 weeks, and the nonacute group, not before 5 months. ###### Completeness of the three scales, by version ------------------------------------------------------------------------------------------------------------------------------------ Internet Version, %\ Paper-and-Pencil Version, %\ Difference, % (95% CI) (n = 46) (n = 202) ----------------------------------------------------- ---------------------- ------------------------------ ------------------------ Total 97.8 63.4 34.5 (26.6-42.3) Short Form-36 \[[@ref16]\]\ 100.0 71.3 28.7 (22.5-35.0) (36 items) Multidimensional Fatigue Inventory-20 \[[@ref17]\]\ 97.8 90.6 7.2 (1.4-13.1) (20 items) Hospital Anxiety and Depression Scale \[[@ref18]\]\ 100.0 92.6 7.4 (3.8-11.0) (14 items) ------------------------------------------------------------------------------------------------------------------------------------ The scores for the eight subscales of Short Form-36 are displayed in [Table 5](#table5){ref-type="table"}. There were no statistically significant differences between the two versions. ###### Scores for subscales of Short Form-36, by randomization group ------------------------------------------------------------------------------------------------------------------- Subscale of Short Form-36 \[[@ref16]\] Internet Version,\ Paper-and-Pencil Version,\ Difference Mean (SD) Mean (SD) ---------------------------------------- -------------------- ---------------------------- ------------ ----- ----- Physical Function 91.4 (15.1) 90.1 (15.6) 1.3 0.5 .60 Role Physical 85.9 (30.6) 81.9 (30.3) 3.9 0.8 .43 Bodily Pain 81.2 (22.2) 76.3 (21.1) 4.9 1.4 .16 General Health 81.3 (14.7) 77.1 (18.7) 4.2 1.4 .16 Vitality 65.4 (22.7) 64.1 (22.2) 1.3 0.4 .71 Social Function 89.9 (16.4) 87.3 (19.0) 2.6 0.9 .39 Role Emotional 86.2 (25.9) 78.7 (32.8) 7.5 1.4 .14 Mental Health 75.2 (17.3) 71.8 (19.8) 3.4 1.1 .28 ------------------------------------------------------------------------------------------------------------------- During the telephone interview with the respondents 1 month after they had their mammogram, they were invited to participate in the follow-up part of the study. They were asked to select the version of future questionnaires they preferred. The majority (55.4%) preferred the paper-and-pencil version, while 32.4% preferred the Internet version. The remaining 17.1% declined further participation. Among the 46 respondents from the Internet group, 73.2% preferred to continue on the Internet compared to 17.1% who preferred to change to a paper-and-pencil version. Access to Internet, estimated by answers from the paper-and-pencil group, is displayed in [Table 6](#table6){ref-type="table"}. ###### Internet access among the paper-and-pencil group ------------------------------------------------------------------------- No. At Home,\ Other^\*^,\ None,\ % % % -------------------------------- ----- ----------- ------------- -------- **Total** 198 68.7 9.6 21.7 **Age (years)** 20-29 12 58.3 25.0 16.7 30-39 48 70.8 8.4 20.8 40-49 56 82.1 3.6 14.3 50-59 55 63.6 18.2 18.2 60-65 27 51.9 0.0 48.1 *χ*^2^~8~= 25.6, *P* = .001 **Place of Residence^†^** Rural 120 62.5 12.5 25.0 Village 52 86.5 7.7 5.8 Urban 26 61.5 0 38.5 *χ*^2^~4~= 17.1, *P* = .002 **Education Level (years)^‡^** 7-10 33 39.4 6.1 54.5 10-12 47 68.1 4.3 27.6 13-17 113 77.0 13.3 9.7 *χ*^2^~4~= 33.2, *P* \< .001 ------------------------------------------------------------------------- ^\*^At work, local library, etc ^†^Defined by postal code ^‡^According to International Standard Classification of Education Discussion ========== We found an initial response rate of only 17.9% in the Internet group compared to 73.2% in the paper-and-pencil group. However, after a reminder, when the participants were free to choose between versions, the total response rate was similar in the two randomized groups. The quality of data regarding completeness was superior in the Internet version for all the involved scales. We did not identify any differences in Short Form-36 subscales. However, even in a randomized study, caution should be exercised when comparing the distribution of answers between the two groups since the distributions depend on differences in the two methods as well as selection bias, especially when the response rate in one of the groups is very low. The population was unselected with respect to Internet access and experience. According to the 2005 Statistics Denmark survey, 77% of Danish women had access to the Internet \[[@ref19]\]. Based on answers from the paper-and-pencil group, we estimate that 70% of the women in the present study had access to the Internet at home. Access was closely associated with level of education. The geographic area surrounding the public hospital includes rural locations as well as the fifth largest city in Denmark. We consider our sample representative for female patients in Denmark. The most prominent weakness of the Internet version was a low response rate, and we could not identify any single determinant factor. However, as expected, the response rate was highest in the age group with greatest access to the Internet. After a reminder letter, which stated that participants were free to fill out their preferred version of the questionnaire, the total response rates were nearly the same. However, women in the acute and subacute groups had less time to complete the questionnaire before their mammogram, which in some cases prevented the reminder. Response rates to Internet questionnaires reported in the literature vary a lot between studies \[[@ref1],[@ref6]-[@ref14]\]. It is evident that studies conducted in populations with known access to the Internet are supposed to have higher response rate than studies of populations without known access, like the present study. However, differences in response rate may also be attributed to methodology and other characteristics of the population. A Swedish study compared the same paper-and-pencil questionnaire in two different versions with respect to ordering of questions and level of difficulty and found that the proportion of completers varied significantly \[[@ref20]\]. It is plausible that populations of patients and general population samples may react differently to an invitation to complete a Web questionnaire about health-related issues. The fact that only 17.1% of respondents in the Internet group preferred to shift to the paper-and-pencil version when asked to join the follow-up study indicates that Internet versions may be more feasible in follow-up studies. One advantage of the Internet version is a high degree of completeness, and the design of Internet questionnaires allows the researcher to compensate for human error among participants who enter inconsistent answers or accidentally skip an item or even a page. At present, Internet questionnaires can hardly stand alone as the method of data collection in studies of patients. Access to the Internet still depends on socioeconomic factors, and results obtained solely from Internet users may be biased. The general population must become more familiar with the Internet before an online survey can be the first choice of researchers, although it is worthwhile considering within selected populations of patients as it saves resources and provides more complete answers. An Internet version may be combined with a traditional version, and it may be more feasible to offer Internet versions in follow-up studies. This work was supported by a grant from the Danish Cancer Society. None declared.

1. Introduction {#sec1-materials-11-02420} =============== In tissue engineering, different structures or scaffolds, which reproduce the shape of tissues, are obtained \[[@B1-materials-11-02420]\]. The design of the architecture of the scaffolds at a macro, micro, and nano level, is essential for the structure, nutrient transport, and cell-matrix interaction conditions \[[@B2-materials-11-02420]\]. Scaffolds must fill the defect cavities and they have to be biocompatible. They require a certain surface morphology to enhance cell attachment and proliferation. They need to be sufficiently strong and have similar stiffness to that of the surrounding tissues. They also require appropriate porosity, pore size, and ratio of surface to volume, to assure permeability \[[@B3-materials-11-02420],[@B4-materials-11-02420]\]. The modulus of elasticity of the bulk material can be reduced by the use of porous structures of the same material. A network structure is a microstructure that is formed by the cells of a block. The lattice materials are categorized as Stretching Dominated Lattice Material (SDLM) and Bending Dominated Lattice Materials (BDLM). In the case of the SDLM, a much stiffer and resistant material per unit mass can be provided by governing the nodal connectivity. This material crumbles with the stretching of its struts. The microscopic structure analysis showed that there is a direct relationship between their stiffness and resistance values, due to an increase in the density ratio \[[@B5-materials-11-02420]\]. In addition, the lower-density network materials undergo failure with plastic bending, even if the macroscopic load is imposed under a uniaxial tensile load \[[@B6-materials-11-02420]\]. The BDLM materials are composed of struts connected at nodes. Its main characteristic is the low-joint connectivity (the number of struts that meet in the same node). In recent years, 3D printing has eased the production of microstructural elements in complex shapes \[[@B7-materials-11-02420]\]. Wettergreen used the Finite Element Analysis (FEA) to estimate microstructural network properties \[[@B8-materials-11-02420]\]. Williams et al. \[[@B9-materials-11-02420]\] found that the experimental modulus of the elasticity for Polycaprolactone (PCL) grid networks produced by Selective Laser Sintering (SLS), was higher than the one obtained by the Finite Element Methods (FEM). This was attributed to the differences in packing of pores and in strut sizes. Luxner et al. \[[@B10-materials-11-02420]\] studied different structures, such as the Simple Cubic (SC), Gibson Ashby (GA), Body Centered Cubic (BCC), and the Reinforced Body Centered Cubic (RBCC). Both Digital Light Processing (DLP) and SLS were used in these experimental tests. Similar values between experimental and FEM tests were reported for all the structures, except for the SC. This was attributed to the high directional sensitivity of the SC structure. Studies have been performed on the additively-manufactured cubic scaffolds composed of several unit cells of Schoen's F-RD surface, a central chamber with tubes to the alternating corners of a cube (bending dominated) and Schoen's I-WP surface, a central chamber with tubes to the corners of a cube (stretching dominated) architectures. The possibility of modifying the mechanical properties by controlling, both, the pore architecture and the relative density of the structures, has been demonstrated \[[@B11-materials-11-02420]\]. Fernandez et al. investigated three different structures---rectilinear, honeycomb, and line---with different densities and different infill percentages, using the Fused Deposition Modelling (FDM) technology. The rectilinear pattern with a 100% infill had the highest tensile strength of 36.4 MPa \[[@B12-materials-11-02420]\]. The experimental tests by Ahmed et al. showed that the average modulus of elasticity of the 3D-printed Acrylonitrile Butadiene Styrene (ABS) specimens was lower than that of the regular plastic ABS \[[@B13-materials-11-02420]\]. A study of different PLA-printed scaffolds showed that the change in mechanical strength was slight and the surface area improved significantly by reducing the scaffolds' cell size \[[@B14-materials-11-02420]\]. Arabnejad et al. have shown that the stretch-dominated cell topologies can be used in load-bearing orthopedic applications for bone replacement and that the octet-truss structure promoted bone ingrowth in a canine model \[[@B15-materials-11-02420]\]. In the same line, Egan et al. tested four types of scaffolds, taking into account the modulus of elasticity and the surface-volume ratio. The scaffolds with the beam-based unit cells showed higher elastic moduli, while the truss-based scaffold tended to attain a greater surface-volume ratio \[[@B16-materials-11-02420]\]. Hollister et al. stated that the higher the density of the scaffolds, the normalized elastic modulus is higher and the permeability is lower. Different scaffold microstructures also showed different stiffness \[[@B17-materials-11-02420]\]. The open-porous titanium scaffolds presented a good agreement for cell growth of a sheep's bone, in vivo. In addition, a higher amount of the newly-formed bone was found surrounding the less-stiff implants \[[@B18-materials-11-02420]\]. As for the octet-truss structure, recent studies by O'Masta \[[@B19-materials-11-02420]\] have shown that its cellular structures, such as the Ti-6Al-4V alloy octet lattice, have higher compressive strengths (between 20 and 70 MPa) and higher elastic moduli, when creating low-density metallic trusses. Bonatti and Mohr studied four different structures---solid octet truss (SOT), hollow sphere assembly (HAS), hollow octet truss (HOT), and hybrid truss-sphere assembly (HTS). They found the lowest modulus of 7.1 GPa, for the SOT structures, and the highest modulus of 10.4 GPa, for the HTS structure. In addition, the experimental results were lower than the FEA-simulated results, for all the structures \[[@B20-materials-11-02420]\]. However, changes in the density of structures can present a sudden drop in mechanical properties \[[@B21-materials-11-02420]\]. The mechanical properties of the structures can be influenced by several factors, such as the cell size, the length of the struts, the radius of struts, and the degree of mutuality of the octet lattice \[[@B22-materials-11-02420]\]. In the present paper, the effect of the struts radius on the structure stiffness has been assessed. The structures density was measured and compression tests were carried out. Prismatic samples were printed with a width to height ratio W/H of 2 and 4. FEM analysis was performed and compared to the experimental results. 2. Materials and Methods {#sec2-materials-11-02420} ======================== In the octet-truss model, the regular octahedral unit cell core is surrounded by eight tetrahedrons, distributed on its eight faces ([Figure 1](#materials-11-02420-f001){ref-type="fig"}a). The cell has the shape of a Face Centered Cube (FCC), along with a lattice structure, which has cubic symmetry and isotropic properties \[[@B23-materials-11-02420]\]. The nodes in this structure are connected to twelve cell elements, in the same manner. In the related literature, the octet-truss network structure is conducted with a solid circular cross-section of struts, especially in the lattice structures, whose cell topology is integrated for constructing an open-cell foam \[[@B24-materials-11-02420]\]. In the present paper, circular cross-section of struts has also been used. 2.1. Printing Tests {#sec2dot1-materials-11-02420} ------------------- Printing tests were performed by means of the Fused Filament Fabrication (FFF) technology, with a Sigma printer, from BCN3D. In the present work, Polylactic acid (PLA) was used, which is a synthetic polymer that does not produce toxic fumes when printing. In the literature, PLA, PLA/calcium phosphate glass and chitosan have been used for manufacturing scaffolds that help in tissue repair and regeneration \[[@B25-materials-11-02420]\]. PLA was also employed for simulating the trabecular structures of the bone tissue \[[@B26-materials-11-02420]\]. For designing the structure, the Grasshopper plug-in \[[@B27-materials-11-02420]\] and the Rhinoceros 3D software \[[@B28-materials-11-02420]\] were used together. The Grasshopper allowed the pieces to be created, through Rhinoceros, with random patterns and forms that could not be created using other drawing software. Two structures were created; each of them were studied separately. The structures were created by Grasshopper, with a cell size of 4 mm and in the toggle graphics mode. As seen in [Figure 1](#materials-11-02420-f001){ref-type="fig"}b, the length of each member of the struts is equal to $L\sqrt{2}$, where $L$ is half the length of the cell size (Equation (1)):$$L~ = ~\frac{Length~of~cell}{2}$$ In this structure, the cell is a cube with a side of 4 mm, which encloses the octal structure. Octet-truss structures were designed within a square base prism, with length to width ratios of H/W = 2 and H/W = 4 ([Figure 1](#materials-11-02420-f001){ref-type="fig"}c). A toggle-false feature was selected so that the program could make the surfaces of the structure, continuous, in the prism, and connected to each other. Preliminary tests showed a higher strength for those prismatic structures that had toggle-false features than those with toggle-true. Therefore, the toggle-false structures have been used in this paper. After creating the octet structure, it was saved as a stereolithographic (STL) file. Moreover, in order to import this structure into a printer, the file had to be stored in the G-code, in laminate programs, such as Cura Ultimaker Ver. 2.7 \[[@B29-materials-11-02420]\], used to prepare an appropriate file for printing. The printer is able to read the file as a G-code. The point to be noted is that the printing parameters should also be included in the STL file, prior to saving as a g-code file. The parameters---nozzle size, layer height, infill, and print speed, were selected, with values of 0.4 mm, 0.1 mm, 95% and 40 mm/s, respectively. The other main parameters selected for this work were---printing temperature, initial layer thickness, infill speed, and shell thickness, with values of 200 °C, 100%, 60 mm/s and 0.8 mm, respectively. Each specimen was manufactured in accordance with Section 6.2 of the ASTM D695-02a standard \[[@B30-materials-11-02420]\]. It was prismatic in shape, with its length being twice its principal width. Initially, the dimensions were 10 × 10 × 20 mm^3^. However, when the elastic modulus is desired, the test specimen should be of such dimensions that the slenderness ratio is in the range from 11 to 16. In this case, the preferred specimen sizes were 10 × 10 × 40 mm^3^ ([Figure 1](#materials-11-02420-f001){ref-type="fig"}c and Equations (2)--(5)). $${\lambda\ = \ }\frac{H}{i}$$ where *λ* is the slenderness ratio, *H* is the specimen height, and *i* is the radius of gyration:$$i{\ = \ }\sqrt{\frac{I}{A}}{\ = \ }\sqrt{\frac{\frac{1}{12}W^{4}}{W^{2}}}{\ = \ }\frac{W}{\sqrt{12}}$$ where *i* is the second moment of area, *A* is the area of the cross-section, and *W* is the width of specimen (length of each side of the cross-sectional area). For H/W = 2:$${\lambda\ = \ }\frac{20\ {mm}}{(\frac{10\ {mm}}{\sqrt{12}})} \cong {\ 7}$$ For H/W = 4 specimens:$${\lambda\ = \ }\frac{40\ {mm}}{(\frac{10\ {mm}}{\sqrt{12}})} \cong {\ 14} \in \left\lbrack 11,\ 16 \right\rbrack$$ Thus, ratio H/W = 4 complies with the specifications of the standard. 2.2. Determination of Density {#sec2dot2-materials-11-02420} ----------------------------- The theoretical density of the octet structure was calculated from the following formulae (Equations (6) and (7)):$$\overline{\rho}\ = \ \frac{6\sqrt{2}A}{l^{2}}$$ $$\overline{\rho}{\ = \ \frac{6\sqrt{2}A}{l^{2}}}{\ - \ c\ \left( \frac{r}{l} \right)^{3}}$$ where $\overline{\rho}$ is the ratio of the lattice material density to the solid material density \[[@B31-materials-11-02420]\], *A* is the cross-section area of each strut, $l$ is the length of each strut, and $r$ is the strut radius. In addition, *c* depends on the detailed geometry of the nodes. The first formula (Equation (6)) was used for struts with a small radius. This formula is a first-order approximation and gives relatively higher results than the second one (Equation (7)), since it calculates the volume of the node, twice. In the second formula (the high-order approximation of Equation (6)), the *c* value should be calculated according to the regular shape of the nodes. In this study, the nodes could not be drawn with a spherical shape; the shape of the intersection among the struts was used instead. Due to the irregular structure, the Rhinoceros software \[[@B28-materials-11-02420]\] was used to obtain the density ratio, instead of Equations (6) and (7). First, a 4 × 4 × 4 mm^3^ cube was designed, using this software, and then the characteristics of the PLA were entered into the program. The unit-cell of the octet structure was also designed as a 4 × 4 × 4 mm^3^ cube, with the same material properties. Then, dividing the density of the octet structure ($\rho)$ by the density of the solid in the same cell size ($\rho^{*})$, according to Equation (8), the density ratio was obtained for each of the three different radii 0.4, 0.5 and 0.6 mm. $$\overline{\rho} = \frac{\rho}{\rho^{*}}$$ The specimens were weighed after printing and the experimental density was calculated, based on the actual volume and mass. Three samples of each condition were used for calculating the experimental density. 2.3. Compression Tests {#sec2dot3-materials-11-02420} ---------------------- The compression tests of the specimens were performed on an Instron 3366 universal testing machine ([Figure 2](#materials-11-02420-f002){ref-type="fig"}a), according to the ASTM D695-02a standard protocol. This standard describes the methodology to be followed in determining the mechanical properties of rigid plastics under compression. Three prismatic specimens of 10 × 10 × 20 mm^3^ were printed for each condition (H/W = 2). Five prismatic specimens of 10 × 10 × 40 mm^3^ were printed for each condition (H/W = 4). Both types of specimen were tested. The results are shown for comparison. The test speed was 1.3 ± 0.3 mm/min, which is the test speed given by the standard. [Figure 2](#materials-11-02420-f002){ref-type="fig"}a shows an example of a compression test. 2.4. FEM Analysis {#sec2dot4-materials-11-02420} ----------------- An academic version of the ANSYS v15 FEM code \[[@B32-materials-11-02420]\] was used for the FEM simulation. In order to simulate the deformation of the octet-truss structure, it was first drawn through the coordinates of its key corners. Each one of the octahedrons and the tetrahedrons were placed within a given 4 mm-sided cube ([Figure 2](#materials-11-02420-f002){ref-type="fig"}b). The stiffness of the octet depends significantly on the size of this cubic pattern. Three radii *r* of the circular cross-section were analyzed: 0.4, 0.5, and 0.6 mm. The struts lying on the outer faces of the unit-cell were cut in half, due to the symmetry boundary condition. The ten-noded solid element (Solid 187 in Ansys) (Canonsburg, Pennsylvania, United States) was used to mesh the truss members of the lattice structure. This finite element model was then subjected to the following uniaxial compressive boundary conditions ([Figure 2](#materials-11-02420-f002){ref-type="fig"}b). The nodes at the bottom face of the octet-truss structure were fixed in the Y direction (Y-symmetry condition). All nodes lying on the top plane were coupled with each other, in the Y direction, i.e., all nodes were forced to share the same UY displacement. In this way, the uniaxial load could be easily applied through the "master node" and the nodes were kept as a horizontal flat surface. The nodes on the XY and YZ rear planes were fixed, in terms of the Z and X directions, respectively, because of the symmetry condition. Likewise, the nodes lying on the XY and YZ front-planes 1 and 2 were coupled with regards to the X and Z directions respectively, in order to keep them as vertical flat surfaces (symmetry conditions). These boundary conditions were verified by the displacement results in [Figure 2](#materials-11-02420-f002){ref-type="fig"}b. A linear structural analysis was performed for this simulation. By prescribing a vertical displacement ($\Delta l_{s}$ (of −0.5 mm on the master node in the negative Y direction, the full reaction force (*F*) was obtained, in this single node, as a computational result. Then, the longitudinal Young's modulus (*E*) could be determined for every radius *r* from Equation (9):$$E~ = ~\frac{\sigma}{\varepsilon}\ = \ \frac{\frac{F}{A}}{\frac{\Delta l_{s}}{l_{s}}}$$ where $\sigma$ is the normal Y engineering stress, i.e., the vertical reaction force *F* divided by the initial area *A* of the upper surface of the cubic cell. $\varepsilon$ is the longitudinal Y strain, i.e., the prescribed displacement of the upper surface in the Y direction ($\Delta l_{s}$) divided by the initial length of the cell ($l_{s}$). 3. Results {#sec3-materials-11-02420} ========== 3.1. Density {#sec3dot1-materials-11-02420} ------------ The theoretical and experimental densities of the samples are shown in [Table 1](#materials-11-02420-t001){ref-type="table"}. As can be seen, similar values were obtained for both theoretical and experimental densities. The bigger the strut radius, higher the theoretical and experimental relative densities of the prisms. The average experimental density values were similar to the theoretical density values. 3.2. Compression Tests {#sec3dot2-materials-11-02420} ---------------------- [Figure 3](#materials-11-02420-f003){ref-type="fig"} shows an example of the linear regression at the initial slope of the stress--strain curve, to obtain the modulus of elasticity. The results of the uniaxial compressive tests are presented in [Figure 4](#materials-11-02420-f004){ref-type="fig"} and [Figure 5](#materials-11-02420-f005){ref-type="fig"}, for ratios H/W = 2 and H/W = 4, respectively. The stress--strain curves showed that the greater was the radius of the struts (r), the higher were the curves and higher was the linear elastic modulus. On the other hand, the stress--strain curves of the specimens with H/W = 4, were higher than those of the specimens with H/W = 2. According to [Figure 4](#materials-11-02420-f004){ref-type="fig"} and [Figure 5](#materials-11-02420-f005){ref-type="fig"}, if the same radius of the struts is considered, compressive strength for samples with H/W = 4 is higher than that for H/W = 2. In addition, the curves obtained showed that the slope of the stress--strain curves for the longer samples was higher than that for the shorter ones; samples with a greater length, met the elastic stage sooner. The regression equations of the five samples for each structure with H/W = 4 and the three samples with H/W = 2 are presented in [Table 2](#materials-11-02420-t002){ref-type="table"} and [Table 3](#materials-11-02420-t003){ref-type="table"}, respectively. For each experiment, the mean value of the modulus of elasticity was calculated. The slope of the regression equations corresponded to the modulus of elasticity. The average values are shown in [Table 4](#materials-11-02420-t004){ref-type="table"}. The greater was the radius of the struts, the higher was the average equivalent Young's modulus; as was expected. Lower Young's modulus and higher standard deviation values of the Young's modulus, were found for H/W = 2 than for H/W = 4. Additionally, the ultimate compressive strength (UCS) obtained from the experimental results is shown in [Table 5](#materials-11-02420-t005){ref-type="table"}. The greater the radius of the struts, the higher was the UCS. Regarding the average values observed, the compressive strength increased along with an increase in the radius of the struts. It was also found that by increasing the radius by 0.1 mm, the compressive strength of the structure was, approximately, doubled, due to the increase in density from 42.1(%) to 57.1(%). Moreover, it could be concluded that the compressive strength for the specimens with H/W = 4 was higher than that for specimens having an H/W = 2. Standard deviation values ranged from 0.252 to 0.468. 3.3. FEM Results {#sec3dot3-materials-11-02420} ---------------- Linear elastic simulations were performed on the above described single-cell FEM models. The full-solid PLA properties were used. Although the elastic modulus value for the commercial amorphous PLA, commonly found in the literature, is 3500 MPa \[[@B33-materials-11-02420]\], other values have also been reported, within a certain range. In this study, the reference values chosen were E = 2500 MPa and ν = 0.36, according to the results of preliminary experimental tests with H/W = 2. [Figure 6](#materials-11-02420-f006){ref-type="fig"}a shows the Y displacement map (UY) of the octet-truss structure. As shown in [Figure 6](#materials-11-02420-f006){ref-type="fig"}b, the experimental results for a radius of 0.4 mm were very similar to the simulated ones. For radii 0.5 and 0.6 mm, the experimental values were lower than the simulated ones. The results of H/W = 4 matched the simulation, better, than the results of H/W = 2. This agreed with the recommendation of the standard---increasing the sample height to prevent the effect of cross-section deformation, due to the compression test. Furthermore, a higher variability was found for H/W = 2 than for H/W = 4. In [Table 6](#materials-11-02420-t006){ref-type="table"}, the equivalent Young's modulus, obtained from the FEM calculations, for each radius value, is presented. The greater the radius of struts, the higher is the obtained equivalent Young's modulus. As an illustration of the specimen densification process, the stress--strain curves are presented in [Figure 7](#materials-11-02420-f007){ref-type="fig"}. These were obtained from a compression test, until extreme deformation of the specimens occurred. Three curves are shown---a solid specimen and two octet specimens---with different radii of the struts. It should be noted that the specimens have been manufactured under conditions different from the previous ones, so the numerical values could not be compared. The curves of the octet structures show a linear elastic area, followed by a plateau, while the curves of the solid structure shows a descending curve, after the linear elastic area. 4. Discussion {#sec4-materials-11-02420} ============= In order to replace natural human bones with implants, surgeons assess different criteria. Modulus of elasticity is one of the main items for the selection of material and structure. In addition, the modulus of scaffolds can vary with time because of the possible cell infiltration and extracellular matrix (ECM) deposition \[[@B34-materials-11-02420]\]. Analytical and clinical studies are required to determine the modulus of elasticity that will result in the best patient outcomes \[[@B35-materials-11-02420]\]. Some studies show that, for bone growth and interconnectivity of the cellular structures, the octet-truss has better results, compared to other porous structures, and it is reconfigurable to match the bone elasticity \[[@B36-materials-11-02420]\]. This fact has also been observed in the present study. Junchao et al. worked on a simplified porous structure, which proved that the ultimate strength and elasticity modulus have a nearly linear increase with the growth of the struts \[[@B37-materials-11-02420]\]. The type of the unit cell is a key factor for the biological and mechanical properties of the porous structures, with regards to the bone tissue regeneration \[[@B38-materials-11-02420]\]. With the help of the compression test, the compressive strength and elastic modulus of the printed structures can be assessed so that they match the requirements of different structures \[[@B39-materials-11-02420]\]. The study of the bone matrix showed that the increase or decrease in bone mass were related to its stress--strain state \[[@B40-materials-11-02420]\]. Different plastic materials have been used for tissue regeneration \[[@B41-materials-11-02420]\]. For example, poly([d]{.smallcaps},[l]{.smallcaps}-lactic-co-glycolic acid) (PLGA) and bothpoly(ε-caprolactone) (PCL) are used in bone tissue regeneration \[[@B42-materials-11-02420]\]. PLGA provided a faster polymer degradation rate than the PCL. The decrease of the molecular weight for PLGA was about 50%, in vitro, and 70%, in vivo, after 28 days, while for the PCL it was about 30%, in vitro, and 40%, in vivo, after the same time \[[@B43-materials-11-02420]\]. In the present study, PLA was used, which is more stable than the PLGA and the PCL \[[@B44-materials-11-02420]\]. On the other hand, mechanical properties of the PLA are suitable for application in bone tissue engineering \[[@B45-materials-11-02420]\]. The PLGA, however, needs to be combined with other polymers to improve its relatively poor mechanical properties (especially its Young's modulus) \[[@B46-materials-11-02420]\]. Another important advantage of these materials is that they have been approved for clinical use \[[@B47-materials-11-02420]\]. Along with polymers, other types of materials are used in tissue engineering. Metals can be employed for manufacturing scaffolds because of its high strength and safety, for use in vivo. However, they also have some disadvantages, such as the limited availability of printing technology and the toxicity of metals when released to the body. Different metals have been printed, for example, stainless steel, cobalt--chromium alloys, titanium alloys, nitinol, etc. \[[@B41-materials-11-02420]\]. Tantalum has also been used for the same purpose. For example, the Young's modulus of the porous Ta structures is low, ranging from 1.5 to 20 GPa \[[@B47-materials-11-02420]\], and can be used in hip and knee-joint reconstruction \[[@B48-materials-11-02420]\]. Metal matrix syntactic foams (MMSFs) are closed-cell porous structures with high stiffness-to-weight ratio. For example, Katona et al. reported modulus of elasticity values around 20 GPa for two different types of aluminium foams \[[@B49-materials-11-02420]\]. Linul et al. tested stiffness of aluminium foam panels with stainless steel reinforcement, at extreme temperatures, between −196 °C and 250 °C. They observed that, the higher the temperature, the lower the stiffness. All stress--strain graphs showed a short elastic area, followed by a long plateau and a final densification area. At 25 °C, they reported modulus of elasticity values up to 336 MPa, with longitudinal reinforced foams \[[@B50-materials-11-02420]\]. As for ceramics, they show high mechanical strength and biocompatibility \[[@B41-materials-11-02420]\]. In addition, some ceramics, like hydroxyapatite (HA) show properties similar to those of natural bones \[[@B51-materials-11-02420]\] and can be used for bone regeneration. Leukers et al. printed HA scaffolds with internal surfaces with a 45° inclination to enhance cell proliferation \[[@B52-materials-11-02420]\]. Calcium silicate has also been employed to print scaffolds with high bone-healing properties \[[@B53-materials-11-02420]\]. However, printing technologies for ceramics are less developed than the printing technologies for polymers. As for modulus of elasticity of biomaterials, modulus of elasticity of ceramics is generally higher than the modulus of metals, and this is higher than the modulus of polymers \[[@B54-materials-11-02420]\]. In addition, the stress--strain curve for ceramics only shows the elastic zone, while for metals, the elastic zone, the plastic zone, and necking are shown. For polymers, the curves show, both, elastic and plastic zone. In this study, the radius of the struts was varied, in order to investigate their effect on the modulus of elasticity of the structure. 5. Conclusions {#sec5-materials-11-02420} ============== In this study, square base prisms were printed with an octet-truss structure. The density of specimens was calculated. Compression tests were carried out and the experimental modulus of elasticity of the octet structure was compared to the results from the FEM calculations. The main conclusions are as follows. The FEM is a useful tool for predicting the behavior of printed octet structure with a low strut radius, when the structure deforms mainly by stretching. For a greater strut radius, the experimental results of the Young's modulus were lower than that predicted by FEM, probably because of bending. Both FEM and the experimental results showed that increasing the radius of the struts in the PLA structures, increases the stiffness and, therefore, the modulus of elasticity of the structures. The compression strength also increases with the radius of the struts. The results for H/W = 4 were more similar to the simulated ones, than the results for H/W = 2. Thus, the results agree with the recommendation of the standard ASTM D695-02a, of using H/W = 4, when determining the modulus of elasticity. Due to its mechanical properties, this structure can be used for printing structures for tissue regeneration. Mechanical properties, such as elasticity modulus can be varied from the selection of the different radius of the struts. In the future, other geometric variables of the structure, such as cell size or length of struts, can also be varied, in order to obtain different structures. The authors acknowledge Ramón Casado-López and Francesc J. García-Rabella for their help in the experimental tests. Conceptualization, A.B., F.R.F. and M.F.B.; Methodology, M.M.P. and I.B.-C.; Writing---original draft preparation, A.B.; Writing---review and editing, I.B.-C., M.M.P. and M.F.B.; Supervision, F.R.F. This research was funded by the Spanish Ministry of Economy, Industry, and Competitiveness, Grant Number DPI2016-80345-R. The authors declare no conflict of interest. FDM fused deposition modeling FFF fused filament fabrication PLA polylactic acid ABS acrylonitrile butadiene styrene W width of specimen H height of specimen R radius of struts l length of struts l s initial length of the cell for FEM {#materials-11-02420-f001} {#materials-11-02420-f002} {#materials-11-02420-f003} {#materials-11-02420-f004} {#materials-11-02420-f005} {#materials-11-02420-f006} {#materials-11-02420-f007} materials-11-02420-t001_Table 1 ###### Comparison between the theoretical and experimental relative density of the octet structure, for the different radii of the struts. Reference Radii of Struts (mm) Theoretical Relative Density (%) Experimental Relative Density (%) Average Experimental Relative Density (%) Standard Deviation of Experimental Relative Density (%) ------------------------------ ---------------------- ---------------------------------- ----------------------------------- ------------------------------------------- --------------------------------------------------------- \[[@B1-materials-11-02420]\] 0.4 34.7 34.6 35.2 0.513 35.3 35.6 \[[@B2-materials-11-02420]\] 0.5 48.3 49.1 48.0 1.018 47.8 47.1 \[[@B3-materials-11-02420]\] 0.6 61.8 58.7 58.8 0.891 57.9 59.7 materials-11-02420-t002_Table 2 ###### Linear regression equations for H/W = 4. Reference of Structures Radius (mm) Regression Equations ------------------------- ---------------------- ---------------------- 1a 0.4 Y = 255.02X − 0.6123 1b Y = 248.08X − 0.5865 1c Y = 247.11X − 0.3492 1d Y = 247.23X − 0.2129 1e Y = 256.24X − 0.4689 2a 0.5 Y = 452.26X − 2.7196 2b Y = 443.74X − 3.3128 2c Y = 462.55X − 1.8626 2d Y = 452.26X − 2.7196 2e Y = 462.81X − 1.8526 3a 0.6 Y = 700.15X − 1.5756 3b Y = 701.18X − 1.9425 3c Y = 698.99X − 1.0848 3d Y = 701.70X − 1.0791 3e Y = 698.68X − 1.1595 materials-11-02420-t003_Table 3 ###### Linear regression equations for H/W = 2. Reference Radius (mm) Regression Equations ----------- ---------------------- ---------------------- 1a 0.4 Y = 162.86X − 1.8463 1b Y = 186.14X − 2.6123 1c Y = 191.27X − 2.0990 2a 0.5 Y = 361.74X − 2.2872 2b Y = 339.97X − 3.3646 2c Y = 365.58X − 3.0434 3a 0.6 Y = 473.57X − 3.6558 3b Y = 451.52X − 2.9425 3c Y = 467.35X − 3.0848 materials-11-02420-t004_Table 4 ###### Young's modulus equivalent to each structure of the octet-truss, given by the compression test. Reference Radius (mm) Average Equivalent Young's Modulus (MPa) for H/W = 4 Standard Deviation of Equivalent Young's Modulus (MPa) for H/W = 4 Average Equivalent Young's Modulus (MPa) for H/W = 2 Standard Deviation of Equivalent Young's Modulus (MPa) for H/W = 2 ------------------------------ ------------- ------------------------------------------------------ -------------------------------------------------------------------- ------------------------------------------------------ -------------------------------------------------------------------- \[[@B1-materials-11-02420]\] 0.4 250.7 4.504 180.1 15.140 \[[@B2-materials-11-02420]\] 0.5 454.7 8.053 355.8 13.811 \[[@B3-materials-11-02420]\] 0.6 700.1 1.320 464.1 11.369 materials-11-02420-t005_Table 5 ###### Ultimate Compressive Strength (UCS) equivalent to each structure of the octet-truss obtained from the compression test. Reference Radius (mm) Compressive Strength (MPa) for H/W = 4 Average Compressive Strength (MPa) for H/W = 4 Standard Deviation of Compressive Strength (MPa) for H/W = 4 Compressive Strength (MPa) For H/W = 2 Average Compressive Strength (MPa) for H/W = 2 Standard Deviation of Compressive Strength (MPa) for H/W = 2 ----------- ------------- ---------------------------------------- ------------------------------------------------ -------------------------------------------------------------- ---------------------------------------- ------------------------------------------------ -------------------------------------------------------------- 1a 0.4 6.87 6.76 0.318 5.07 5.56 0.430 1b 6.40 5.76 1c 6.48 5.86 1d 7.18 \- 1e 6.85 2a 0.5 11.93 12.17 0.252 11.45 11.12 0.468 2b 11.93 10.59 2c 12.53 11.34 2d 12.23 \- 2e 12.25 3a 0.6 19.03 18.65 0.465 17.13 14.41 0.369 3b 18.57 17.83 3c 18.32 17.28 3d 19.21 \- 3e 18.11 materials-11-02420-t006_Table 6 ###### Equivalent Young's modulus obtained from the FEM. Reference of Structure Radii of Struts (mm) Equivalent Young's Modulus (MPa) ------------------------ ---------------------- ---------------------------------- 1 0.4 184.9 2 0.5 508.6 3 0.6 886.2

INTRODUCTION ============ Arteriosclerotic occlusive disease of the ilio-femoral arteries is a common cause of ischemic symptoms in the lower limbs. Revascularization is the treatment of choice for patients when either limiting intermittent claudication unresponsive to clinical treatment or critical limb ischemia is present [@b1-cln_70p675]-[@b4-cln_70p675]. Iodine contrast medium (ICM) is considered the gold standard for endovascular treatments worldwide. An alternative that has been used within the last 15 years is carbon dioxide (CO~2~), which is used for patients with renal impairment or hypersensitivity to iodine because it is non-nephrotoxic and non-allergenic [@b5-cln_70p675]. With improvements in both imaging equipment and injection techniques, CO~2~ has even been used in patients without contraindications to ICM for the detection of endoleaks [@b6-cln_70p675], femoral-popliteal angioplasty [@b7-cln_70p675],[@b8-cln_70p675], cavography for vena cava filter placement [@b9-cln_70p675] and investigation of the biliary tree [@b10-cln_70p675]. To date, no randomized prospective studies have compared the outcomes of CO~2~ contrast media and ICM use for endovascular revascularization of the ilio-femoral segment in patients with no contraindications to ICM. Therefore, to determine whether CO~2~ is as effective as iodine contrast media for ilio-femoral angioplasties, we performed a randomized prospective study involving patients with ilio-femoral occlusive disease who were eligible for either type of contrast (ICM or CO~2~). We analyzed the feasibility of the procedures, the surgical and clinical outcomes, the procedure lengths, the endovascular material costs, the contrast costs and the quality of the angiographic images. MATERIAL AND METHODS ==================== From August 2012 to August 2014, 21 consecutive patients with ilio-femoral disease (all with atherosclerotic disease) with arterial lesions classified as Trans-Atlantic Inter-Society Consensus Document on Management of Peripheral Arterial Disease (TASC) A, B, C or D (identified via preoperative angiotomography) and adequate runoff underwent endovascular ilio-femoral revascularization in a prospective randomized controlled study. This study was approved by the Ethics Committee for Analysis of Research Projects on Human Experimentation of our institution (Clinical Trials Identifier: NCT01861327). None of these patients exhibited severe chronic obstructive pulmonary disease, kidney failure, heart failure, or pregnancy The patients were randomized into the following two groups: a CO~2~ group and an ICM group, according to the contrast medium selected for the procedure. All patients provided informed consent and agreed to the use of either contrast medium. Randomization was performed using a computer-generated list immediately prior to the beginning of the surgery. Eighteen of the 21 patients had no palpable femoral pulses in the limb that was revascularized. Two patients in the CO~2~ group and one patient in the ICM group had weak palpable femoral pulses pre-operatively. [Table 1](#t1-cln_70p675){ref-type="table"} presents the demographic, clinical and laboratory profiles of the patients in our study sample who were subjected to either ICM or CO~2~. All procedures were performed in an endovascular suite operating room utilizing a Philips Allura Xper FD system (Philips Healthcare, Best, the Netherlands) under general anesthesia with cardiac monitoring, invasive arterial pressure monitoring and bladder catheterization. The patients underwent ilio-femoral angioplasties by the same surgical team (four surgeons) using the same surgical technique throughout the study in the same hospital (Hospital Israelita Albert Einstein, São Paulo, Brazil). An ipsilateral puncture of the common femoral artery was used for access in 19 of the 21 patients. In one case, an ipsilateral puncture was unsuccessful; therefore, we opted for contralateral femoral access. In another case, a contralateral puncture was the initial choice due to the presence of an atherosclerotic lesion in the external iliac artery. Standard surgical technique was used in all cases. After an ultrasound guided arterial puncture and the insertion of a 6-French introducer access sheath, systemic heparinization was performed followed by initial angiography with the appropriate contrast medium. The lesion was crossed using catheters and hydrophilic guide wires followed by the selective delivery of a self-expanding stenting and pre- and post-dilatation with an angioplasty balloon. At the end of the procedure, the heparinization was reversed, and local manual compression was performed for 30 minutes after removal of the introducer. All of the endovascular materials were provided by Cook Medical, Inc.®. The CO~2~ injection was performed manually without a specific pump system. A cylinder with medicinal CO~2~, a particle filter (Millex® Durapore® hydrophilic 0.22 μm pore) and a three-way stopcock were used for CO~2~ aspiration, and the entire procedure was performed inside a bowl with saline solution. After capturing the required volume of CO~2~, an additional 3 to 5 ml of saline solution was aspirated into the syringe to provide a water seal as the tip was held down. This process allowed for the creation of a physical barrier between the room air and the carbon dioxide, which was independent of the manual compression and safe from air contamination [@b11-cln_70p675]. Twenty-milliliter syringes were used for the CO2 intra-arterial contrast injection, which was performed using either a femoral introducer or a catheter. The injection of ICM was performed manually using 10-ml syringes with 3 ml of iodinated contrast media and 7 ml of saline solution per injection. The ICM used in all cases was Omnipaque 300 (Iohexol), a nonionic low osmolar contrast agent routinely used by our service. Immediately after the operation, all patients were referred to the intensive care unit for at least 24 hours in accordance with our institutional research protocol for vascular surgery and received intravenous fluids following a fixed protocol for renal protection, regardless of the contrast medium used. The protocol for renal protection included intravenous fluids pre-procedure with endovenous acetylcysteine and intravenous fluid maintenance for 24 hours. Thereafter, the patient remained hospitalized for the time needed, with daily renal function, blood count and electrolyte monitoring for at least 72 hours. The endovascular material used in each intervention and the volumes of ICM or CO~2~ were precisely recorded for further analysis. The endovascular material costs included puncture needles, sheaths, angioplasty balloons, catheters, insufflating syringes and stents used during the endovascular procedure. The contrast costs included the total cost of the contrast used during the procedure. It is important to emphasize that because ICM is required for balloon filling during angioplasties, we included the price of one 20-ml flask of ICM for each patient in the CO~2~ group (irrespective of ICM supplementation). Procedure length was measured based on the amount of time elapsed from the arterial puncture until the end of manual compression. We analyzed creatinine clearance levels between the groups during the pre- and post-operative periods as well as creatinine clearance variations between the post- and pre-operative periods (delta clearance). All procedures were recorded on DVD and the recorded angiography procedures (of both groups) were analyzed by two observers who did not participate in the intervention and had no experience with the use of CO~2~ contrast. The observers assigned a score ranging from 1 to 3 to each image (scoring was only based on the iliac-femoral segment; therefore, 11 CO2 images were scored, i.e., one image per patient). A score of 1 was considered *poor* and was assigned when there was significant loss of definition within the vessels or collateral circulation, which precluded performing the procedure. A score of 2 was considered *fair* and was assigned when there was some loss of definition within the vessels or collateral circulation, which did not preclude the intervention. A score of 3 was considered *good* and was assigned when there was good contrast within the vessels or collateral circulation. We evaluated the following clinical outcomes in both groups: the feasibility of the procedures, the complications, the surgical outcomes (femoral pulses and ankle brachial index (ABI)), the pre- and post-operative creatinine clearance, the contrast volume, the quality of the angiographic images, the procedure costs and the contrast media costs. Statistical Analysis -------------------- Categorical variables were expressed as absolute frequencies and percentages and compared between groups via Fisher\'s exact test. Numerical variables were expressed as medians and interquartile ranges and compared via the nonparametric Mann-Whitney U test. The level of significance was 5%. RESULTS ======= Endovascular treatment was performed in 21 patients (11 patients in the CO~2~ group and 10 patients in the ICM group) with ilio-femoral disease. The percentages of overall technical success in the CO~2~ and ICM groups were 90.9% and 90%, respectively, with only one procedure failure per group. These failures were the only patients who did not receive a self-expanding stent. No procedure-related deaths were noted and none of the patients presented with major clinical or surgical complications. Conversion to open surgery was not necessary in any of the cases and no contrast media-related complications were noted in either group (CO~2~ or ICM). The surgical outcomes were satisfactory in both groups. The presence of palpable femoral pulses measured immediately following each procedure indicated that the technical success rate was similar for both groups, as presented in [Table 2](#t2-cln_70p675){ref-type="table"}. (The three patients with weak femoral pulses pre operatively presented with strong palpable femoral pulses post operatively and were considered technical successes.) Variability (difference between post-operative and pre-operative ABI) was higher for the ICM group (0.40 *vs*. 0.16, *p*=0.044). For the ICM group, the median iodine contrast volume was 34.25 ml (range: 9 to 69 ml); for the CO~2~ group, the median CO2 volume was 60 ml (range: 23 to 130 ml). In four (36.4%) CO~2~ patients, the use of iodine contrast was necessary to complete the procedure; however, none of these patients required more than 15 ml of iodine (range: 3 to 15 ml). The median iodine contrast volume used for the CO~2~ patients who required iodine supplementation was 6.5 ml. Creatinine clearance during the postoperative period did not differ significantly between the groups and remained stable during the post-operative period in both groups, as presented in [Table 3](#t3-cln_70p675){ref-type="table"}. The delta clearance (creatinine clearance after the procedure minus clearance before the procedure) was positive for the CO~2~ group (2.01) and zero for the ICM group; however, this difference was not statistically significant (*p*=0.44). Procedure length did not differ significantly between the groups (median length was 75 minutes for the CO~2~ group and 70 minutes for the ICM group, *p*=0.92), indicating that CO~2~ does not increase procedure duration. We also analyzed the procedure lengths for the different TASC classifications in each group and observed no significant differences. The costs of the endovascular materials were similar for both groups, indicating that CO~2~ does not increase procedure costs. However, the contrast medium costs were significantly increased for the ICM group compared with the CO~2~ group (see [Table 4](#t4-cln_70p675){ref-type="table"}). Regarding angiographic image quality, which is presented in [Table 5](#t5-cln_70p675){ref-type="table"}, 82% of the CO~2~ angiography images were graded as either good or fair by both observers, with high inter-observer image quality concordance. All iodine arteriograms were graded as good by both observers. [Figure 1](#f1-cln_70p675){ref-type="fig"} depicts an example of CO~2~ angiography. DISCUSSION ========== Treating patients with renal impairment or iodine related hypersensitivity remains a challenge when performing both diagnostic and therapeutic endovascular procedures. The use of CO~2~ as an alternative contrast medium to ICM has provided a solution for these situations. Given that CO~2~ is a good option for these patients, we aimed to investigate the possibility of the more widespread use of CO~2~ by treating patients with no contraindications to either CO~2~ or ICM. In this study, we observed that the use of CO~2~ did not change surgical outcomes given that the patients in the CO~2~ group exhibited no differences in post-operative femoral pulses compared with the patients who received iodine contrast. The variation in ABI between the groups may be explained by differences in distal outflow (i.e., patients presented with different patterns of femoro-popliteal and infra-genicular stenosis/occlusion). In addition, procedure length was not extended in the CO~2~ group. Pain during the injection of CO~2~ was not assessed because the patients were under general anesthesia. However, some of the patients could have received local anesthesia (and this assessment could have been undertaken). Due to the institutional research protocol (which mandated general anesthesia and an overnight stay on ICU for every vascular surgery patient), we were unable to compare discomfort (e.g., pain) during the procedure between the groups. No significant differences were noted between the patients treated with CO~2~ or iodine with respect to renal function during the post-operative period. In the ICM group, we did not observe any changes. The lack of alteration was most likely because of either a small sample size or the administration of a low volume of iodine contrast, as numerous studies have demonstrated the nephrotoxicity of ICM. Endovascular material costs were similar for both groups, demonstrating that the use of CO~2~ does not increase procedure costs. However, the contrast media costs for the ICM group were significantly higher compared with the CO~2~ group, even after including the price of one 20-ml flask of ICM for balloon filling for each patient in the CO~2~ group. The evaluation of the angiographic images by the observers indicated that most of the images in the CO~2~ group (82%) were evaluated as either good or fair; therefore, it may be possible to perform the procedure with only CO~2~ in these cases. It is noteworthy that 63% of images for observer 1 and 45% for observer 2 were considered good with CO~2~ and that all images with iodine were rated as good. However, our aim was to assess the feasibility of C0~2~ not its superiority to iodine contrast media. Images were considered poor with CO~2~ in patients with long occlusions (2 patients with TASC D lesions for whom the filling of the distal vessels with CO~2~ was difficult to assess). Additional poor images were observed at the beginning of our series, at which time we supplemented patients with iodine to verify vessel patency. Although the learning curve of this procedure may be overcome with experience, it is possible that CO~2~ may not replace iodine in TASC D lesions. This limitation might be a problem for allergic patients but may also benefit patients with borderline renal function for whom iodine contrast media usage may be significantly reduced. The CO~2~ contrast agent is a gas; therefore, its manipulation requires the use of an unusual technique. Most vascular surgeons are not accustomed to working with a gaseous contrast medium, as its manipulation requires care to prevent air contamination [@b12-cln_70p675]. In addition, most surgeons lack the training needed to obtain good quality images. Some CO~2~ injection pumps have been developed [@b13-cln_70p675]-[@b15-cln_70p675], but they are expensive and difficult to use. We have developed and published a simple homemade delivery system for CO~2~ in endovascular procedures, which uses materials that may be found in any hospital and are both cheap and reproducible. We believe that systems such as ours may lead to more widespread use of CO~2~ as a contrast agent in endovascular procedures. When performed by experienced physicians familiar with its use, CO~2~ may provide angiographic image quality comparable to that achieved using iodinated contrast media. We consider CO~2~ as effective as ICM within the femoro-popliteal territory. We demonstrated that CO~2~ is a viable contrast option for ilio-femoral lesion treatment and observed outcomes similar to those noted with iodine contrast. Complications following the use of intra-arterial CO~2~ have been reported, including *vapor-lock*, which occurs upon entrapment of CO~2~ at the origin of a vessel, thereby impairing flow and potentially resulting in ischemia. This condition may be mild and cause only transient pain or may present as significant bowel ischemia [@b16-cln_70p675],[@b17-cln_70p675]. However, no adverse events were observed in our series secondary to the use of this particular contrast media. We routinely waited 3 minutes between CO~2~ injections to avoid *vapor-lock* and any subsequent complications. We believe that vascular surgeons should be trained during their residency in vascular surgery to acquire the technical ability to use CO~2~. The technique is unusual, but its learning curve is not steep. The technique may be useful for vascular surgeons searching for the best contrast option for each patient, as opposed to being limited to the use of only one type of contrast. Given this study\'s small sample size and the absence of a longer follow up period with respect to renal function, additional studies supporting our findings are encouraged. Our study demonstrated that CO~2~ is a good alternative to ICM in the endovascular treatment of ilio-femoral occlusive disease (except in TASC D patients, for whom iodine was often but not always necessary to complete the procedure). CO~2~ is characterized by similar endovascular material costs, procedure lengths and surgical outcomes as well as and reduced contrast medium costs compared with iodine contrast. This work was supported by the Ministério da Saúde do Brasil. No potential conflict of interest was reported. {#f1-cln_70p675} ###### Demographic and clinical profiles. GROUPS ------------------------------- ---------------- ---------------- ----------- Age average (IIQ) 48 (43.5-60.0) 69 (60.0-75.0) 0.005 \* Age range (years) 37-73 58-79 Median BMI (kg/m^2^) 21.72 24.17 0.26 \* Diabetes (%) 6 (54.5) 4 (40.0) 0.67 \# Hypertension (%) 7 (63.6) 8 (80.0) 0.63 \# Dyslipidemia (%) 4 (36.4) 4 (40.0) \>0.99 \# Coronary disease (%) 0 (0.0) 1 (10.0) 0.48 \# History of tobacco use (%) 10 (90.9) 7 (70.0) 0.42 \# Intermittent claudication (%) 3 (27.3) 3 (30.0) \>0.99 \# Critical ischemia 8(72.7) 7(70.0) 0,63 \# TASC A lesion (%) 6 (54.5) 2 (20.0) 0.24 \# TASC B lesion (%) 2 (18.2) 2 (20.0) 0.24 \# TASC C lesion (%) 0 (0.0) 3 (30.0) 0.24 \# TASC D lesion (%) 3 (27.3) 3 (30.0) 0.24 \# \# Fisher\'s exact test, \* Mann-Whitney U test, ICM: Iodine contrast media CO~2~: Carbon dioxide, BMI: Body mass index. ###### Surgical outcomes. GROUPS -------------------------------------- ----------- --------- ----------- **Palpable femoral pulse pos n (%)** 8 (88.9%) 8 (80%) \>0.99 \# **Median ABI variation** 0.40 0.16 0.044 \* \# Fisher exact test, \* Mann-Whitney U test, ICM: Iodine contrast media CO~2~: Carbon dioxide, Pos: Immediate post-operative period, ABI: Ankle brachial index. ###### Comparison of pre- and post-operative creatinine clearance and delta clearance between the groups. ICM GROUP n=10 CO~2~ GROUP n=11 *p*-value\# ------------------------------------------- ---------------------- ---------------------- ------------- **Median creatinine clearance pre (IIQ)** 90.20 (64.04-112.65) 79.40 (74.29-108.74) 0.94 **Median creatinine clearance pos (IIQ)** 93.05 (75.55-99.16) 88.40 (69.93-131.88) 0.62 **Median delta clearance** 0 (-16.9-10.28) 2.01 (-7.65-13.54) 0.44 \# Mann-Whitney U test, IIQ: Interquartile interval, ICM: Iodine contrast media, CO~2~: Carbon dioxide, Pre: Before procedure, Pos: Post-operative period, Delta clearance: Creatinine clearance after the procedure minus clearance before the procedure. ###### Endovascular material costs and contrast costs between the CO~2~ and iodine contrast media groups. GROUPS -------------------------------------- ---------------------------- --------------------------- --------- **Endovascular material cost (IIQ)** 1527.69 (1527.6 9-1789.23) 2302.69 (1540.19-2984.07) 0.168 **Contrast cost (IIQ)** 10.12 (10.12-10.12) 25.0 (25.0-25.0) \<0.001 \# Mann-Whitney U test, IIQ: Interquartile interval, ICM: Iodine contrast media, CO2: Carbon dioxide, values in American dollars. ###### Evaluation of CO~2~ angiography by observers. Ilio-femoral angiography ---------------- -------------------------- --- --- **Observer 1** 7 2 2 **Observer 2** 5 4 2 [^1]: Mendes CA and Wolosker N conceived the study, participated in its design and drafting, collected the data, were responsible for analysis and interpretation of the samples, manuscript writing and statistical analysis. Martins AA, Teivelis MP and Yasbek AM collected the data, wrote the manuscript and helped with the statistical analysis. Kuzniec S and Fioranelli A participated in the analysis, interpretation of the samples and data collection.

Introduction ============ The state of cerebral autoregulation is related to functional outcome as has been shown in several studies including ischemic stroke \[[@B1]\], spontaneous intracerebral hemorrhage (ICH) \[[@B2],[@B3]\], subarachnoid hemorrhage (SAH) \[[@B4]\] and traumatic brain injury (TBI) \[[@B5]\]. With the pressure reactvity index (PRx), Czosnyka *et al*. have introduced a method for continuous monitoring of cerebrovascular reactivity which can be used at the bedside \[[@B6]\]. Of importance, the PRx can be used to calculate the 'optimal cerebral perfusion pressure' (CPPopt) which is defined as the cerebral perfusion pressure (CPP) associated with the lowest values of PRx \[[@B5]\]. Studies of traumatic brain injury \[[@B5],[@B7]\] and subarachnoid hemorrhage \[[@B8]\] have shown a relationship between outcome and the difference between the actual CPP and CPPopt. Just recently an algorithm to continuously assess CPPopt in regularly updated four-hour intervals has been published \[[@B7]\]. However, for spontaneous ICH there are only limited data available on dynamic autoregulation \[[@B2],[@B3],[@B9],[@B10]\] and CPP management \[[@B11],[@B12]\] and, so far, there are no data regarding autoregulation-based CPP management. However, current guidelines recommend maintaining CPP between 50 and 70 mmHg (Class IIb, Level C), depending on the state of autoregulation \[[@B13],[@B14]\], conceding that these recommendations are entirely based on data from TBI patients. Here, we (1) continuously assessed the state of cerebrovascular reactivity in patients with severe spontaneous ICH treated on a Neurocritical Care Unit (NCCU) and (2) investigated if the CPPopt concept is applicable for ICH patients. Furthermore, the association of impaired autoregulation and CPPopt to functional outcome as reported for TBI \[[@B5]\] and SAH \[[@B8]\] patients was exploratively investigated. Methods ======= Patients -------- Between 2007 and 2009, 55 non-consecutive patients with spontaneous ICH admitted to the Heidelberg NCCU were prospectively included in our monitoring database. Inclusion criteria were: (1) spontaneous ICH and (2) the need for intracranial pressure (ICP) measurement. Our clinical standard protocol includes ICP measurement for all ICH patients who are analgosedated and mechanically ventilated. Furthermore, if patients have an indication for placement on an extraventricular drainage due to intraventricular extension of the hemorrhage, we usually aim to place a Raumedic NEUROVENT (RAUMEDIC AG, 95233 Helmbrechts, Germany) probe, allowing ICP measurement during simultaneous cerebrospinal fluid (CSF) drainage (see below). All patients required intubation and mechanical ventilation and were continuously sedated using midazolam, sufentanil, ketamine and/or propofol. A total of 18 patients had been included in a previous study \[[@B2]\]. Therapy was aimed to keep ICP \<20 mmHg and CPP \>60 mmHg according to current guidelines and local standards \[[@B13]\]. All patients were ventilated to achieve partial arterial oxygen pressures above 80 mmHg and a partial pressure of carbon dioxide in the blood (P~a~CO~2~) of 30 to 40 mmHg. ICH severity and neurological deficit on admission were assessed by the National Institutes of Health Stroke Scale (NIHSS). Hematoma volume was calculated from the first computed tomography (CT) scan using the axbxcx0.5 method \[[@B15]\]. The amount of intraventricular blood was estimated using the Graeb score \[[@B16]\]. The SAPS II (Simplified Acute Physiology Score II) on admission was extracted from the hospital patient database as a measure for overall disease severity. Outcome at three months was assessed using the modified Rankin Scale (mRS) based on a telephone interview or evaluation of the reports from the rehabilitation clinic. The study was part of a larger neuromonitoring project. The retrospective analysis of the monitoring data was approved by the local ethics committee (Ethics Committee University of Heidelberg, S-139/2007, Amendment IV). For retrospective data analysis, the need for informed consent was waived. Neuromonitoring and data recording ---------------------------------- Intracranial pressure was measured using a Raumedic NEUROVENT probe, providing a combined probe allowing ICP measurement during simultaneous CSF drainage. In six patients, ICP was measured by a standard external ventricular drain (EVD). These patients were excluded from the current analysis since ICP values are not correctly measured via an open EVD. Blood pressure was measured from the radial artery (Dräger, Medical Deutschland GmbH, Lübeck, Germany). ICP, systolic and diastolic blood pressure, mean arterial pressure (MAP), CPP and heart rate were synchronously recorded with a sampling frequency of 1 Hz using a commercially available software (ICU pilot, M Dialysis AB, Johanneshov, Sweden). The data were stored on a bedside computer. Assessment of pressure reactivity index (PRx) --------------------------------------------- Data files were cleaned from epochs containing incomplete data recordings. Incomplete recordings were caused by disturbed interaction between the monitoring system and the recording software or by complete disconnection of the patient from the monitoring system (for example, in-house transportations). Next, data were re-sampled to obtain one value every 10 seconds. Then, PRx was calculated as a moving linear (Pearson) correlation between 30 consecutive values (=five-minute time window) of MAP and ICP as described by Aries *et al*. \[[@B7]\]. For calculation of PRx only, data points fulfilling the criteria MAP between 50 and 120 mmHg, ICP \>0 mmHg and CPP between 30 and 120 mmHg were included into the analysis, thereby aiming to further eliminate artifacts and outlier values, for example, due to flushing of the arterial system or shorter disconnections of one probe. According to the literature, a PRx \>0.2 was defined as impaired cerebrovascular reactivity \[[@B6]\]. For each patient the time spent with a PRx \>0.2 was calculated as the percentage of the total monitoring time. Furthermore, PRx values were averaged using Fisher-Z-Transformation to obtain an average value for each patient. All calculations were performed using Matlab (version R2007b, Mathworks, Ismaning, Germany). Assessment of CPPopt -------------------- The optimal CPP (CPPopt) was calculated according to the algorithm published by Aries *et al*. \[[@B7]\]. Shortly, all recorded CPP values of each patient were divided into bins of 5 mmHg and corresponding PRx values were averaged using Fisher-Z-transformation within these groups. Fisher-Z-transformation was used because of the expected non-normal distribution of PRx values. CPPopt in general is defined as the CPP value associated with the lowest average value of PRx (see Figure [1](#F1){ref-type="fig"} for an example). For continuous calculation and update of CPPopt values, the calculation was repeated every hour using a four-hour moving time window. An automatic algorithm to identify CPPopt was used as published by Aries *et al*. \[[@B7]\]. Furthermore, the difference between the real CPP and CPPopt was continuously calculated within the four-hour moving time window. A median value of the continuously calculated difference between the actual CPP and CPPopt was recorded for each patient and used for further analysis (Δ). The percentage of the total monitoring time during which CPPopt was calculable was recorded for each patient. In addition, the time each patient spent within (CPPopt ± 5 mmHg), below (CPP \< CPPopt-5 mmHg) and above (CPP \> CPPopt + 5 mmHg) the continuously updated CPPopt was calculated as the percentage of the total monitoring time. Finally, CPPopt was identified based on the entire monitoring data available for each patient as described by Steiner *et al*. \[[@B5]\]. {#F1} Statistical data analysis ------------------------- Spearman rank correlation was used in univariate analysis to assess the relationship among PRx, time spent within, below and above the CPPopt range and outcome at three months. In addition, patients were grouped according to their median difference between real CPP and CPPopt (three groups: median Δ ±5 mmHg within CPPopt range, median Δ \>5 mmHg above CPPopt and median Δ \>5 mmHg below CPPopt). The distribution of dichotomized outcome parameters (mortality, acceptable (mRS 1 to 4) and favorable (mRS 1 to 3) outcome) was compared using Fisher's exact test. Finally, separate forward stepwise logistic regression models were calculated to predict mortality, and favorable and acceptable outcome at three months. A stepwise model was chosen due to the explorative nature of the analysis. Due to the limited number of patients, only the known outcome predictors including age (years), initial hemorrhage volume (ml), presence of intraventricular blood in addition to percentage of time with PRx \>0.2 and percentage of time within the CPPopt range were included in the model. To account for general disease severity in this special subset of ICU patients, the SAPS II score was also included in the model. There was no significant correlation between the included covariates. Statistical analysis was performed using the SPSS software package (IBM Deutschland GmbH, Ehningen, Germany). Results ======= Patients' characteristics and outcomes -------------------------------------- In total, 55 patients with spontaneous ICH were prospectively monitored. Of these, four were excluded because of monitoring time \<12 hours or CPP below 30 mmHg, one because of early withdrawal of care, one because of terminal stage malignoma and one because of generalized brain edema in a follow-up CCT. Additionally, the five patients with purely intraventricular hemorrhage were excluded as well as the patients who had standard EVDs without continuous ICP monitoring. All patients in whom full therapeutic measures initially were undertaken have been included in our analysis. This left 38 patients for the current analysis. During their ICU stay, in three patients it was decided not to escalate therapeutic measures (days 7, 8, 17), two of these died. The mean age was 58 years (±SD 16, range 18 to 84 years). Median NIHSS at admission was 30.5 (IQR 20, range 7 to 39). The median SAPS II on admission was 13 (IQR 5.3, range 6 to 32). Median hematoma size was 36 ml (IQR 50, range 3 to 144 ml), 34 patients (89.5%) had intraventricular hemorrhage extension, the median Graeb score was 6.5 (IQR 5, range 9 to 11). A total of 32 patients (84.2%) had deep hemorrhage, 5 (13.2%) lobar and 1 brainstem and cerebellar hemorrhage (2.6%). With respect to etiology, 23 patients (60.5%) had hypertensive hemorrhage, 4 had hemorrhage associated with coagulopathy (10.5%), 5 had hemorrhage of other etiologies (13.2%) and for 6 patients no etiology could be determined (15.8%). Median total monitoring time was 78 hours (IQR 74, range 21 to 190 hours). A total of 12 (31.6%) patients underwent hematoma evacuation, 2 (5.3%) had hemicraniectomy and 1 (2.6%) had hematoma evacuation and hemicraniectomy (for this patient only data before surgery were available). At three months, 3 patients (8.1%) attained a mRS of 2, 7 patients (18.9%) a mRS of 3, 11 patients (29.7%) a mRS of 4, 5 patients (13.5%) a mRS of 5. In-hospital mortality was 23.7% (n = 9), mortality at three months was 29.7% (n = 11). For one patient outcome at three months was not available. Analysis of cerebrovascular pressure reactivity ----------------------------------------------- The mean PRx was suggestive for impaired pressure reactivity defined as PRx \>0.2 in 47.4% (n = 18) of patients. The median time spent in PRx ranges \>0.2 was 45.4% of the total monitoring time (IQR 23.0, range 15.9 to 95.1). The mean PRx and the time spent with a PRx \>0.2 were significantly correlated to mRS at three months (r = 0.50, *P* = 0.002 and r = 0.46, *P* = 0.004). In our cohort, autoregulatory failure was not correlated to ICH severity: neither the mean total PRx nor the mean PRx on the first day of monitoring were significantly correlated to admission status assessed by NIHSS (r = −0.23, *P* = 0.159 and r = -0.22, *P* = 0.181) or hemorrhage volume (r = 0.14, *P* = 0.398 and r = 0.09, *P* = 0.605). Analysis of optimal CPP (CPPopt) -------------------------------- Figure [1](#F1){ref-type="fig"} shows two exemplary patients to depict the CPPopt concept. CPPopt was calculable during a median 57.1% of the total monitoring time (IQR 27.3, range 0 to 92.9). In a single patient, CPPopt was not calculable at all. The median of all continuously calculated CPPopt values was 83 mmHg (IQR 10, range 68 to 98). The median of all continuously calculated actual CPP values was 78 mmHg (IQR 8, range 62 to 92). The median CPPopt based on the CPP and PRx values of the entire monitoring time was 93 mmHg (IQR 12.5, range 68 to 113). The median CPP of the entire monitoring time was 79 mmHg (IQR 9, range 61 to 92). Median ICP was 10 mmHg (IQR 3.5, range 2 to 22). Patients spent 10.5% of their total monitoring time within the continuously updated CPPopt range (IQR 10.8, range 0 to 33%). The percentage of time above the CPPopt range was 15% (IQR 13.1, range 0 to 64) and the time spent below the CPPopt range was 26.7% (IQR 23.6, range 0 to 59) in the median. The continuously calculated difference between CPP continuous CPPopt (Δ) was -5.6 mmHg (IQR 9.0, range -12.9 to 9.4) in the median. CPPopt and outcome ------------------ There was no linear correlation between mRS at three months and the times spent within, above or below the continuously updated CPPopt. Furthermore, functional outcome was not correlated to total monitoring time (r = -0.045, *P* = 0.791). The correlation between median Δ and outcome was not assessed since we did not assume a linear relationship. Figure [2](#F2){ref-type="fig"} instead depicts the relationship among mortality, acceptable and favorable outcome, and the individual median of the continuously calculated difference between real CPP and CPPopt (mmHg). A total of 36 patients were grouped according to their median Δ (n = 36 since for 1 patient outcome was not available, and in 1 patient CPPopt was not determinable): 21 patients (55.3%) lay 5 mmHg or more below their CPPopt values (median Δ -7.9 mmHg, range -5.5 to -12.9, IQR 4), 10 patients (26.3%) lay ±5 mmHg within their individual CPPopt range (median Δ -1.05 mmHg, range -4.9 to 4.3, IQR 5.6) and 6 patients (15.8%) were 5 mmHg or more above their CPPopt range (median Δ 6.7 mmHg, range 5.2 to 9.4, IQR 2.7). {#F2} Mortality was lowest in the group of patients within the CPPopt range (10%), while it was almost equal in those more than 5 mmHg below (38.1%) or above (33.3%) CPPopt (*P* = 0.241, Fisher's exact test, Figure [2](#F2){ref-type="fig"}A). The percentage of patients with acceptable outcome was highest when CPP was within the CPPopt range (70%), compared to 47.6% of those below and 50% of those above CPPopt (*P* = 0.523, Fisher's exact test, Figure [2](#F2){ref-type="fig"}B). In contrast, patients who were more than 5 mmHg above their CPPopt range had the highest percentage of favorable outcome (33.3%), compared to 20% of patients within the CPPopt range and 28.6% of those more than 5 mmHg below CPPopt (*P* = 0.769, Fisher's exact test, Figure [2](#F2){ref-type="fig"}C). The distribution of the NIHSS scores or hemorrhage volume did not differ significantly among the three groups (*P* = 0.630 and *P* = 0.287, Kruskal-Wallis-test). Outcome prediction - multivariate models ---------------------------------------- Mortality at three months was predicted by age and initial hemorrhage volume (Table [1](#T1){ref-type="table"}). The only significant predictor of good outcome (mRS 1 to 3) at three months was younger age. Independent predictors for acceptable outcome at three months were lower percentage of time with PRX \>0.2 and lower hemorrhage volume. The variable 'time spent within the CPPopt range' did not reach significance in any of the models. ###### Stepwise logistic regression models to predict outcomes at three months   **B (SE)** **OR** **95% CI** ***P*** ------------------------------------------------------------ -------------- -------- -------------- --------- **Mortality at three months**^**1**^ Age 0.08 (0.04) 1.08 1.00 to 1.17 0.042 Hemorrhage volume 0.03 (0.01) 1.03 1.01 to 1.06 0.101 **Good outcome at three months (mRs 1 to 3)**^**2**^ Age -0.13 (0.04) 0.88 0.81 to 0.96 0.004 **Acceptable outcome at three months (mRS 1 to 4)**^**3**^ \% time PRx \>0.2 -0.06 (0.03) 0.94 0.90 to 1.00 0.038 Hemorrhage volume −0.02(0.1) 0.98 0.96 to 1.00 0.054 Included variables: age \[years\], hemorrhage volume \[ml\], presence of intraventricular hemorrhage, percentage of time with PRx \>0.2, percentage of time within the CPPopt range, SAPS II score (^1^Nagelkerkes R^2^ 0.47, ^2^Nagelkerkes R^2^ 0.50, ^3^Nagelkerkes R^2^ 0.38), *B*, regression coefficient; CI, confidence interval; OR, odds ratio; SE, standard error. Discussion ========== To our knowledge, this series is the largest available on dynamic assessment of cerebral autoregulation in severe ICH patients. Furthermore, it is the first study assessing the concept of an autoregulation-oriented CPP management in ICH. We have found that impairment of pressure reactivity occurs in almost half of the cases following spontaneous ICH. The PRx and the time spent with a PRx \>0.2 were significantly correlated to mRS at three months. The amount of time a patient had impaired pressure reactivity (PRx \>0.2) was an independent predictor for acceptable outcome at three months. These results confirm the findings of a previous pilot study \[[@B2]\]. Importantly, the CPPopt concept seems applicable in ICH patients. In our retrospective, explorative analysis, calculation of CPPopt was feasible in all but one patient and during 57% of the monitoring time. This number is almost identical to the number indicated by Aries *et al*. who reported that identification of CPPopt was on average feasible during 55% of the recording period in their cohort of 326 TBI patients \[[@B7]\]. The median CPPopt value was 83 mmHg and thereby higher than the threshold or 70 mmHg recommended in current guidelines \[[@B13],[@B14]\]. This is potentially interesting since a recent study by Ko *et al*. applying microdialysis and brain tissue oxygenation study in 18 ICH patients showed that CPP values between 50 and 70 mmHg were associated with brain tissue hypoxia, depicting the possibly detrimental effects of hypoperfusion \[[@B11]\]. In our series, however, we did not find a significant relationship between functional outcome at three months and any of the CPPopt variables. Aries *et al*. have found in their much larger TBI cohort that hypoperfusion was associated with mortality while hyperperfusion was correlated with severe disability. In contrast, in our cohort of ICH patients, both hypo- as well as hyperperfusion seemed harmful as could be concluded from the mortality rates in the different CPP groups. Mortality was lowest for patients close to their individual CPPopt. At the same time, patients in this group had the highest percentage of acceptable outcome. These findings, however, were not statistically significant. In addition, the results for favorable outcome remain inconclusive. In our cohort, the highest percentage of patients with favorable outcome was in the group of those with 5 mmHg or more above their individual CPPopt while at the same time patients with a CPP close to CPPopt had the least percentage of favorable outcomes. This directly contrasts the above mentioned findings with respect to mortality. These inconsistencies may be due to the small sample size and the total lack of patients with a delta CPP above +10 mmHg. Patient numbers in total and especially in the above CPPopt group (n = 6) were too small to draw any definitive conclusions. Furthermore, the variable 'time spent within the CPPopt' range was not an independent predictor for any of the outcome variables. Our study has various important limitations. First, sample size was still relatively small and the study included a highly selected population of ICH patients. As stated above, especially the results of the outcome analyses comparing the different groups based on their proximity to CPPopt did not render any statistically significant results and have to be weighed critically. Only speculative conclusions can be drawn so far. Furthermore, the study was an explorative retrospective analysis of prospectively monitored, non-consecutive patients. Sample size was not predetermined. Second, the amount of available data differed greatly among different individuals, possibly adding further bias. However, reassuringly, total monitoring time was not correlated to outcomes, PRx or time of impaired PRx. Third, CPPopt was calculated retrospectively and the amount of time spent within the CPPopt range was relatively small (11%, range 0 to 33%). In addition, calculation of CPPopt based on the currently used automated algorithm was on average only possible during half of the monitoring time. This may be due to the fact that not all CPP values were covered by each patient, rendering identification of CPPopt impossible since it possibly lay outside the recorded values. This limits the generalizability of the association between the difference of 'real CPP minus optimal CPP' and outcome after ICH. However, as stated above, the same percentage (55%) was found in one other retrospective trial using continuous calculation of CPPopt in TBI patients. This number certainly has to be improved by carefully evaluating the reasons for failure of determination of CPPopt and possibilities to increase the numbers. This should be done in a prospective trial. Only after this issue has been addressed can the CPPopt concept be safely introduced for further clinical use. We are aware that a prospectively conducted study, trying to optimize cerebral perfusion based on an improved CPPopt algorithm may render entirely different results due to the availability of larger amounts of data and broader CPP variations. Finally, as a technical limitation, times during CSF drainage as well as patients following surgery and specifically two patients following hemicraniectomy have been included in our analysis. While the NEUROVENT probe allows reliable ICP measurement also during CSF drainage, it is unclear whether this will affect the reliability of PRx assessment. Calculation of PRx is based on the principle of a closed system, with ICP being a surrogate for changes in intracerebral blood volume. However, considering the fact that we clearly found positive correlations between MAP and ICP and hence positive PRx values, we argue that the system still reacts during times of drainage. The same holds true for the two patients following hemicraniectomy. A comparison of PRx values and CPPopt calculation before and after surgery suggests no systematic difference in these two patients (compare Additional file [1](#S1){ref-type="supplementary-material"}: Table S1). If specific patients during CSF drainage would entirely have to be excluded, this would drastically reduce the applicability of PRx for continuous assessment of pressure reactivity in collectives like ICH or SAH. A study on this issue is thus far missing. However, there are data indicating that in situations with high brain compliance, for example, following surgery, the PRx may not be an ideal parameter, suggesting a different index PAx as an alternative, especially in situations with lower ICP \[[@B17]\]. Furthermore, a non-invasive alternative for continuous determination of cerebrovascular reactivity and CPPopt may be the near-infrared spectroscopy based index THx. The index has been evaluated in TBI patients \[[@B18]\]. However, the applicability of THx in the setting of high brain compliance has not been addressed. Furthermore, the correlation between the autoregulation indices varies around 0.5 \[[@B18]\] and 0.6 \[[@B17]\], indicating that different techniques may represent distinct aspects of cerebrovascular reactivity and autoregulation. So far, the PRx remains the best studied parameter, especially with respect to calculation of CPPopt and probably would be the parameter of choice in a prospective trial on CPPopt management. Despite these important limitations, the study encourages further research on autoregulation-based CPP management in ICH. Moreover, our finding of impaired dynamic autoregulation following ICH is corroborated by two other independent studies using the Doppler flow velocity-based index Mx \[[@B3],[@B9]\]; one of them also reports a relationship between impaired autoregulation and worse functional outcome \[[@B3]\]. Further indirect arguments in favor of individualized thresholds have just been provided in a recent study on ICP management in TBI patients \[[@B19]\]. Chestnut and colleagues have compared an ICP-based treatment protocol to a protocol solely based on clinical observations. The outcomes in both groups did not differ. The authors stress that their findings do not argue against the use of ICP monitoring in general but only argue against the monitoring-based interventional algorithm of lowering ICP \>20 mmHg. They suggest that the lack of efficacy of the ICP-based protocol may be attributable to the use of a universal threshold for intracranial pressure. Currently, a protocol for a prospective multicenter trial comparing standard CPP management versus autoregulation-based CPP management is being designed. Conclusions =========== Due to a lack of ICH specific data, current guidelines for CPP management in ICH recommend taking the functioning of cerebral autoregulation into account but are entirely based on data from TBI patients. Here we could show that cerebrovascular pressure reactivity is impaired in almost 50% of patients with severe ICH and impairment of pressure reactivity was related to poor functional outcome at three months. The CPPopt concept seems applicable to ICH patients: CPPopt could be determined in all but one patient and during 57% of the monitoring time. The median CPPopt was 83 mmHg and, therefore, considerably higher than the cutoff value of 70 mmHg recommended in the current guidelines. So far, however, a significant relationship between functional outcome and CPPopt could not be demonstrated in ICH patients. This may be due to sample size and the retrospective approach. Larger, prospective trials are needed to assess the validity of the CPPopt concept for severe ICH patients. Key messages ============ ● Cerebrovascular pressure autoregulation was impaired in half of the patients treated for severe ICH. ● Impairment of pressure reactivity is independently associated with poor functional outcome at three months. ● The CPPopt concept seems applicable in ICH: CPPopt was determinable in 57% of the total monitoring time. ● The median CPPopt was 83 mmHg and thereby considerably higher than the guideline recommended value of 70 mmHg. ● Mortality was lowest in the group of patients with a CPP close to their individual CPPopt. Abbreviations ============= B: Regression coefficient; CI: Confidence interval; CPP: Cerebral perfusion pressure; CPPopt: Optimal cerebral perfusion pressure; CSF: Cerebrospinal fluid; CT: Computed tomography; EVD: External ventricular drain; ICH: Intracerebral haemorrhage; ICP: Intracranial pressure; IQR: Interquartile range; MAP: Mean arterial pressure; mRS: Modified Rankin Scale; NCCU: Neurocritical Care Unit; NIHSS: National Institutes of Health Stroke Scale; OR: Odds ratio; PaCO2: Arterial pressure of carbon dioxide; PRx: Pressure reactivity index; R: Correlation coefficient; SAH: Subarachnoid hemorrhage; SAPS II: Simplified Acute Physiology Score II; SE: Standard error; TBI: Traumatic brain injury; Δ: Delta. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== JD contributed to conception and design, data collection and analysis, manuscript writing and final approval of the manuscript. ES and SP contributed to data collection and analysis, critical revision and final approval of the manuscript. MS contributed to conception and design, data collection and analysis, critical revision and final approval of the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1: Table S1 Comparison of PRx values and CPPopt assessment before and after surgery for the two patients who had hemicraniectomy. ###### Click here for file Acknowledgements ================ JD was supported by the Olympia-Morata scholarship of the University of Heidelberg. There was no special funding for the study.

{#sp1 .129} {#sp2 .130} {#sp3 .131} {#sp4 .132} {#sp5 .133} {#sp6 .134} {#sp7 .135} {#sp8 .136} {#sp9 .137} {#sp10 .138} {#sp11 .139} {#sp12 .140} {#sp13 .141} {#sp14 .142} {#sp15 .143}

1. Introduction {#sec1} =============== Advances in perinatal imaging and diagnostic tools often allow for recognition of complex, rare, and even life-threatening congenital malformations prior to birth. Prenatal diagnosis of these conditions provides time for earlier counseling and planning for perinatal management options. Guiding a family through this process can be difficult for the medical team, but is an attempt to improve the overall outcome and experience for everyone involved. We present a case of diprosopus associated with multiple congenital malformations which were prenatally diagnosed. The parents received extensive multidisciplinary, prenatal counseling allowing both the family and the medical providers to be well prepared for the birth and postnatal management. 2. Case {#sec2} ======= A 29-year-old gravida 2 para 1 Caucasian female was referred for maternal-fetal-medicine consultation at 26-week gestation due to suspected fetal anomalies. Obstetrical ultrasound examination, confirmed by fetal magnetic resonance imaging (MRI), demonstrated craniofacial duplication, several abnormalities of the brain and skull, thoracolumbosacral dysraphism with neural tube defect and likely Chiari II malformation, large congenital diaphragmatic hernia (CDH) with liver and bowel noted in the left chest, hypoplastic left lung, and possible horseshoe kidney ([Figure 1](#fig1){ref-type="fig"}). Fetal echocardiogram findings were consistent with Tetralogy of Fallot (TOF). The parents had a multidisciplinary consultation with maternal-fetal medicine, neonatology, genetics, cardiology, radiology, and palliative care. Given the multiple, severe congenital anomalies, the medical team and family planned for limited resuscitative efforts and anticipated comfort care after birth. A 2440 g male infant was delivered via repeat cesarean section at 36-week gestation secondary to preterm labor. At birth, the infant was pale and cyanotic with no spontaneous cry and poor tone. He had shallow spontaneous respirations with poor aeration on auscultation. The left mouth was small and fixed. No glottis was visualized when the laryngoscope was inserted into the right mouth. The infant was then wrapped in a warm blanket and given to the family to hold. He died in his parents\' arms a few minutes after birth. The parents consented to a complete postmortem evaluation. Physical exam demonstrated complete facial duplication with tetrophthalmos (four eyes), two noses, two mouths, two chins, and one fully formed ear on each side of the head with a hypoplastic pinna in the midline. Two anterior and two posterior fontanels were palpable. There was also a ruptured myelomeningocele measuring 6.3 cm × 3.9 cm in the thoracolumbar region and a scaphoid abdomen. The first and second toes of each foot overlapped and the nails were hypoplastic. Dissection of the cranial cavity revealed brain duplication with fusion of the parietal and occipital lobes of each brain. The right and left brains fused at the mesencephalon and brainstem. Each brain had an optic chiasm, a pituitary gland, and a single large ventricle with loss of the ependymal lining and no third ventricle identified. There was a single midbrain, rudimentary cerebellar tissue and fourth ventricle, atresia of the cerebral aqueduct with rudimentary and hypoplastic cerebral peduncles, bilateral absence of the corpus callosum, bilateral polymicrogyria, and dysplasia of the right and left cortices. Dissection of the thorax and abdomen revealed left CDH with intestine, stomach, spleen, and the left lobe of the liver herniated into the left pleural cavity with concomitant pulmonary hypoplasia. Cardiac findings included TOF (pulmonary atresia, ventricular septal defect, overriding aorta, and pulmonary arteries), atrial septal defect, patent ductus arteriosus, superior pulmonary veins draining to the left atrium (no inferior pulmonary veins), dilated right atrium, and hypoplastic left atrium. A common oropharynx connected two separate nasopharynxes and two separate oral cavities. There was a single larynx and esophagus. The intestines were malrotated with the appendix located in the left upper abdominal quadrant. Ectopic right and left kidneys were located in the lower abdomen and an accessory spleen was present. Testes were undescended bilaterally and the right testis was atrophic. The placental pathology revealed normal fetal membranes and chorionic villi consistent with third trimester gestation with a two-vessel umbilical cord ([Figure 2](#fig2){ref-type="fig"}). Whole-genome microarray-based comparative genomic hybridization (aCGH) using Oxford Gene Technology revealed a male karyotype (46 X,Y), a 983 kb deletion on chromosome 4q34.3, a 562 kb gain on chromosome Xp22.31p22.2, and a 32 kb gain on chromosome 13q12.11. Quantitative polymerase chain reaction of maternal blood indicated the duplications on chromosomes Xp22.31p22.2 and 13q12.11 were maternally inherited. The 4q34.3 deletion was not maternally inherited, however. Unfortunately, due to the family\'s financial burden, genetic testing was not completed on the father. 3. Discussion {#sec3} ============= Diprosopus or craniofacial duplication is the rarest form of conjoined twinning, with an incidence of approximately 0.4% of all types of conjoined twins \[[@B1]\]. Historically, two main theoretical embryologic explanations have been considered: either a "fusion" of two parallel notochords in close proximity occurs or a "fission" of a single notochord occurs during the first few weeks after conception \[[@B2]\]. More recent theories include duplication of neural crest cell derivatives and mutations of the*Dix* homeobox gene \[[@B3], [@B4]\]. A spectrum of diprosopus exists from a duplication of only the nose to complete facial duplication similar to that of our patient \[[@B5]\]. Many other congenital anomalies have been reported to occur in conjunction with diprosopus, including various neurologic, cardiac, pulmonary, skeletal, and gastrointestinal system defects \[[@B5]--[@B8]\]. Conjoined twins have a high mortality, especially if they have other major congenital anomalies \[[@B8], [@B9]\]. There have been no previous reports of genetic mutations associated with diprosopus. The clinical significance of the maternally inherited duplications at Xp22.31p22.2 and 13q12.11 is unknown. There is a theoretical possibility of an X-linked recessive inheritance pattern since the X chromosome duplication has caused no disease or malformation in the mother, but this is unlikely as no reports of familial recurrence have previously been reported in the literature. The 4q34.3 deletion has been associated with cardiovascular abnormalities such as TOF \[[@B10], [@B11]\], but none of these genes are located in the deletion region in our case. Given the rare incidence of this disease and our inability to complete genetic testing of both parents, we are unable to conclude that any of the genetic findings are associated with our patient\'s congenital malformations and the family was counseled accordingly. However, a lack of genetic association for diprosopus continues to support an embryologic theory of abnormal twinning. It should be noted that the prenatal diagnosis at 26 weeks somewhat limited perinatal management options. However, the combination of detailed ultrasound and MRI imaging provided accurate prenatal diagnoses so that the family and medical providers were able to develop a comprehensive plan of care in advance of the delivery. Almost three months before the birth, medical providers compassionately informed the family that a successful resuscitation was unlikely. The parents asked that providers to make an attempt to resuscitate, but if unsuccessful, they wanted to hold and baptize their son. During follow-up counseling, the family stated that the birth and death experiences were more peaceful than anticipated. They felt that the medical, emotional, and spiritual support provided by the multidisciplinary team significantly reduced their anxiety, facilitated their decision-making, and, ultimately, aided them in coping with their loss. Conflict of Interests ===================== The authors declare no conflict of interest. {#fig1} {#fig2} [^1]: Academic Editor: Nan-Chang Chiu

INTRODUCTION {#rjz005s1} ============ The prevalence of small bowel diverticulitis is small; prevalence increases with age, characteristically in males in the sixth to eighth decades of life \[[@rjz005C3], [@rjz005C4]\]. Jejunal diverticula are thought to be pulsion diverticula secondary to intestinal dyskinesia \[[@rjz005C3]\]. Although jejunal diverticula are not an unusual finding within themselves, they create a nidus for possible complications including perforation, bleeding, abscess and obstruction \[[@rjz005C4]\]. The effects of these possible outcomes may lead to clinically confusing presentations, such as iron deficiency anemia due to diverticular bleeding \[[@rjz005C5]\]. These complications are rare but can be serious and potentially require surgical management in some cases. Therefore, it is imperative for providers to consider small bowel diverticula when managing any patient presenting with symptoms such as abdominal pain, nausea and fever. There is little literature providing research on the presentation or management of small bowel diverticulitis, increasing the risk of the underlying pathology of patients with such symptoms being undetected. Small bowel diverticulitis, while rare, is a complication that must be addressed by providers to prevent significant morbidly and mortality in these patients. CASE REPORT {#rjz005s2} =========== A 65-year-old male with a history of Barrett's esophagus presented to the emergency department with acute pain across his lower abdomen, nausea and fever of 39.4°C. The lower abdominal pain was described as sharp and most notably in the left lower quadrant. On physical exam, the pain was present upon palpation of the abdomen, but no guarding or rebound was noted. Bowel sounds were normal and abdominal distension was noted. The patient stated that he had never experienced symptoms similar to these previously. The patient's WBC count was 15.10 k/μL (3.80--10.80 k/μL) and all other laboratory data was within normal limits. A chest x-ray was unrevealing. An abdominal and pelvis CT with IV contrast revealed a small collection of air and debris adjacent to a loop of mildly thick-walled small bowel in the left lower quadrant measuring roughly 2.3 × 1.9 × 2.0 cm^3^, with a small adjacent focus of apparent extraluminal air (Fig. [1](#rjz005F1){ref-type="fig"}). The CT also did reveal extensive colonic diverticulosis; however, the radiographic evidence was suggestive of the possible sequelae of an acute perforated small bowel diverticulitis as opposed to a perforated colonic diverticulitis. {#rjz005F1} The patient was medically treated with ceftriaxone (2 g every 24 h), metronidazole (500 mg IV every 8 h), ondansetron (4 mg every 6 h, PRN) and pantoprazole (40 mg IV daily). Following a surgical consult with the patient and his family, it was decided to pursue surgical treatment with an open laparotomy with possible small bowel resection. The operation was performed under general anesthesia with endotracheal, nasogastric (previously placed) and Foley catheter intubation. The abdomen was widely shaved, prepped and draped in a sterile fashion. A midline incision was made from above the umbilicus to just below the umbilicus through a previous supraumbilical scar from his prostatectomy, and subcutaneous tissue was dissected with cautery and fascia was opened under direct visualization along with the peritoneum sharply. Exploration was done manually and visual for the entirety of the abdominal cavity. During the exploratory laparotomy, the segment of small bowel with the inflammatory phlegmon was noted and withdrawn from the surgical site for inspection. An isolated segment of bowel measuring \~20 cm was resected between and a side-to-side functional and end anastomosis was created. The surgical findings included a solitary diverticulum at the mesenteric aspect of the bowel with marked erythema, induration and slight exudate at the proximal jejunum. There was no evidence of diverticular perforation. Pathology revealed the serosal surface of the jejunum was notable for a 2.5 cm region of protrusion associated with congestion, exudation and possible hemorrhage; when the resected segment of small bowel was opened, there was a prominent fecal filled diverticulum corresponding to the focus of serosal protrusion (Figs [2](#rjz005F2){ref-type="fig"}--[4](#rjz005F4){ref-type="fig"}). {#rjz005F2} {#rjz005F3} {#rjz005F4} The patient had an unremarkable recovery with no complications for the remainder of his admission. The patient was discharged home 5 days after surgery when ambulating, having bowel function, tolerating diet. DISCUSSION {#rjz005s3} ========== We present a case of small bowel diverticulitis in a 65-year-old male who presented with abdominal pain, nausea and fever. Initial medical management was utilized to stabilize the patient, and surgical management was then pursued. In this case, surgical management of small bowel diverticulitis was successful in this patient. The success of this method could likely be replicated in other patients with similar presentations and pathologies. However, there is a lack of research of the efficacy of medical management compared to surgical management of this condition, which leaves the provider to decide between the two interventions. In cases of diverticular bleeding, there is an indication for surgical management to remove the affected area of small bowel \[[@rjz005C5]\]. The area is first localized via intraoperative endoscopy or CT angiography. Failure to satisfactorily remove the region of bleeding could expose the patient to complications including anemia and shock. In this situation, surgical management is more clearly chosen. In patients without acute bleeding from a diverticulum, the choice may be less clear on how to proceed with treatment. Both medical and surgical management options offer potential solutions, and each is with risks and benefits. When proposing a surgical intervention, a patient's overall health status and personal medical history must be considered. CONFLICT OF INTEREST STATEMENT {#rjz005s4} ============================== None declared.

BM is one of most actively proliferating tissues of the body. In humans, ∼10^11^--10^12^ mature blood cells are generated per day ([@bib14]) from short-lived highly proliferative hematopoietic progenitors which arise from a rare population of hematopoietic stem cells (HSCs) with multilineage differentiation and self-renewal capacity ([@bib22]). Transplantation studies in mice have shown that a single HSC is capable of reconstituting and maintaining all hematopoietic lineages in lethally irradiated recipients for their respective lifetime ([@bib33]; [@bib19]), or even longer in assays of serial transplantation ([@bib15]; [@bib3]); however, physiological hematopoiesis is at least oligoclonal at any given time, indicating that it is not maintained by offspring of one single HSC ([@bib16]; [@bib26]). Adult BM provides a specialized microenvironment for HSCs, the so-called "niche," that allows HSC maintenance at homeostatic levels. This process is controlled by both intrinsic and extrinsic signals, as demonstrated in multiple animal models with genetic alterations in nonhematopoietic and hematopoietic tissues ([@bib32]). Steady-state HSC cycling kinetics have been evaluated by in vivo labeling assays with BrdU ([@bib7]; [@bib20]) and biotin ([@bib30]) and, in recent studies, using both BrdU labeling and histone 2B GFP (H2B-GFP) transgenic mouse models ([@bib46]; [@bib12]). BrdU and biotin labeling suggested that all HSCs divide, on average, every 17.8 d, whereas combined BrdU and H2B-GFP experiments revealed two HSC populations, one actively dividing about every 9--36 d and one dividing approximately every 56--145 d ([@bib46]; [@bib12]). Assuming conserved division rates in these populations in steady state during a 2-yr laboratory mouse lifetime, the former experiments suggest that HSCs divide ∼42×, whereas the latter suggest that one population would divide ∼20 times and the other 5 times. This might imply that slow dividing HSCs would not contribute to relevant amounts of mature blood cells and could be a dormant reserve, only recruited upon hematopoietic challenge ([@bib46]; [@bib12]). However, the faster cycling HSCs repopulated lethally irradiated animals only in the short term, whereas the dormant HSCs showed long-term repopulation ([@bib46]; [@bib12]). Therefore, these findings face difficulty in explaining how high-throughput hematopoiesis is maintained by only a small HSC fraction with long-term potential but a very slow division rate. An alternative explanation would be that a single steady-state HSC pool exists that contains stochastically changing slow and fast dividing HSCs, with most HSCs within the pool dividing longitudinally at similar rates. Although in steady-state adult mice more than two-thirds of HSCs are in a resting, i.e., G0/G1 phase of cell cycle ([@bib7]), HSC division can be induced in situations of hematopoietic challenge, for example, chemotherapeutic treatment, irradiation, and BM transplantation ([@bib42]). Recently, it was demonstrated that HSC-enriched populations express Toll-like receptors (TLRs), which recognize bacterial or viral molecules, and that TLR ligation causes proliferation and enhances production of innate immune cells such as macrophages and dendritic cells for respective host defense ([@bib29]; [@bib25]). Also, some cytokines produced by immune cells in response to viral or bacterial infection can activate quiescent HSC ([@bib11]; [@bib39]; [@bib4]). However, it is thus far unclear if these naturally occurring stimulants trigger only differentiation or also accelerated HSC self-renewal. To address the questions of how often HSCs divide and, thus, possibly contribute simultaneously, sequentially, or repetitively to steady-state hematopoiesis, of whether there is a relationship between divisional history and blood forming or repopulating ability, and if HSC self-renewal is increased in hematopoietic stress upon severe infectious challenge, we set up an in vivo HSC divisional tracking system using CFSE labeling. RESULTS ======= In vivo CFSE dilution reveals steady-state divisional heterogeneity of Lin^−^c-kit^+^Sca-1^+^ (LKS) cells --------------------------------------------------------------------------------------------------------- Because ∼1% of HSCs are estimated to circulate in steady-state blood, and some of them physiologically rehome into the ∼1% unoccupied BM HSC niches ([@bib48]; [@bib5], [@bib6]), we reasoned that i.v. transfer of HSC into nonirradiated mice resembles this process competitively and thus allows evaluation of steady-state HSC divisional dynamics. To track HSC division with high resolution, we labeled cells with CFSE, a fluorescent dye which is equally distributed to daughter cells upon each cellular division ([@bib45]; [@bib24]). CFSE labeling did not impair HSC function, as CFSE-labeled LKS cells containing HSCs successfully engrafted nonirradiated animals with stable chimerism of ∼1% over at least 20 wk (unpublished data). Transplantation of 10^5^ LKS into nonirradiated mice revealed that 0×-divided Lin^−^c-Kit^+^ cells with an equivalent CFSE intensity to that of naive CD4^+^CD62L^+^ T cells transferred to determine zero cell division CFSE intensity. When mice were irradiated before LKS transfer, all donor-derived LKS cells had divided ≥5× in 3 wk and high BM and blood chimerism was established as expected ([Fig. S1](http://www.jem.org/cgi/content/full/jem.20101643/DC1)). Weekly BM analysis of nonirradiated recipient mice transplanted with CFSE-labeled LKS showed that most of the donor cells divided within 4 wk, whereas a small fraction did not divide over 21 wk, the longest period observed ([Fig. 1 A](#fig1){ref-type="fig"}). 0×-divided cells were maintained in the BM, but not in the spleen, and included LKS cells with a CD150^+^ and CD34^−^ phenotype, which are reported to be highly enriched for quiescent HSC ([@bib33]; [@bib19]; [Fig. S2, A and B](http://www.jem.org/cgi/content/full/jem.20101643/DC1)). These data demonstrate that in steady state the majority of LKS cells divide actively and a small fraction is quiescent. {#fig1} To compare the HSC-tracking method established in this study with BrdU labeling ([@bib7]; [@bib20]), mice were treated with BrdU starting 1 wk after CFSE-labeled LKS cell transfer. BrdU treatment led to more LKS divisions, confirming a mitogenic effect of BrdU ([@bib20]; [@bib46]; [Fig. 1 B](#fig1){ref-type="fig"}). However, BrdU incorporation neither correlated linearly with divisions nor allowed divisional resolution. Most importantly, BrdU did not recruit all LKS cells into division, leaving behind a fraction of 0×-divided non--BrdU-labeled LKS. These data demonstrate that CFSE labeling provides high-resolution single-division HSC tracking and reveal that BrdU labeling alters cell cycle kinetics but, as expected, does not mark the remaining quiescent cells. This fact might have had a profound impact on readout and interpretation of previous studies using the BrdU labeling and chase assay to evaluate HSC turnover, as cells are recruited into division and only recently divided and labeled cells are subsequently monitored, and most 0×-divided quiescent cells are not included in the chased population ([@bib7]; [@bib46]). To estimate the turnover and loss rate of total LKS cells, i.e., a population which contains but is not exclusively composed of HSCs ([@bib34]), we developed mathematical models assuming that the LKS cell population is composed of one, two, three, or four subpopulations with differing rates of proliferation and loss. Fitting these mathematical models to the CFSE dilution data revealed at least three subpopulations with different division and differentiation kinetics (see Materials and methods and [Table S1](http://www.jem.org/cgi/content/full/jem.20101643/DC1)). One population was characterized by a high rate of loss and no proliferation, and the remaining two populations roughly correspond to the fast-cycling and dormant populations identified previously ([@bib46]; [@bib12]). Expression of HSC and lineage-differentiation markers by LKS cells upon division -------------------------------------------------------------------------------- To correlate divisional history with HSC and progenitor cell--associated surface marker expression, we analyzed lineage negative cells in BM 3 wk after transfer. 0--2×-divided donor Lin^−^c-kit^+^ cells were positive for Sca-1 (\>98%) and the thrombopoietin receptor c-Mpl (\>93%), both of which are highly expressed in HSC ([@bib41]; [@bib35]; [@bib49]), and their expression decreased with division. 40--50% of 0--2×-divided cells were positive for CD150 ([@bib20]) and negative for CD34 ([@bib33]), and Fms-like tyrosine kinase 3 (Flt3) expression ([@bib8]; [@bib17]) was low in 0--2×-divided and up-regulated in 3--6×-divided cells ([Fig. 2](#fig2){ref-type="fig"}). Some Lin^−^c-Kit^+^ cells started to express markers of lineage commitment like M-CSFR ([@bib31]), IL-7Rα ([@bib21]), and FcγRII/III ([@bib1]; [@bib22]) after more than three divisions (Fig. S2, C and D). Thus, surface marker expression pattern analysis demonstrates that at 3 wk after transfer, 0--2×-divided cells displayed a HSC immunophenotype. Furthermore, LKS cell differentiation to lymphoid or myeloid lineage-committed progenitors required at least three divisions in steady state. {#fig2} Heterogeneous cycling activity of functional HSCs ------------------------------------------------- As HSCs represent a minority within LKS cells ([@bib34]) and LKS cell division does not necessarily reflect HSC divisional dynamics, we determined HSC frequency by in vivo single-cell and limiting dilution repopulation and serial transplantation assays. We reisolated defined numbers of donor LKS cells based on their divisional history at different time points after primary transfer ([Fig. 3 B](#fig3){ref-type="fig"}) and transplanted these into lethally irradiated secondary recipient mice along with irradiation-protective whole BM cells, followed by monthly analysis for donor engraftment ([Fig. 3 A](#fig3){ref-type="fig"}). Transplantation of single 0×-divided LKS cells, isolated 3--14 wk after primary transplantation, yielded multilineage reconstitution in 22--28% of irradiated mice for 16 wk ([Fig. 3, C and D](#fig3){ref-type="fig"}; [Table I](#tbl1){ref-type="table"}; and [Fig. S3](http://www.jem.org/cgi/content/full/jem.20101643/DC1)). Up to 250 LKS cells that had divided ≥5× in 3 wk in primary recipients showed no engraftment, whereas 50--250 ≥5×-divided LKS cells at 7 or 12--14 wk after primary transfer contained long-term multilineage repopulating HSCs ([Fig. 3 E](#fig3){ref-type="fig"}, [Table I](#tbl1){ref-type="table"}, and Fig. S3). Based on in vivo limiting dilution transplantation, the frequency of HSC in ≥5×-divided LKS cells at 7 and 12--14 wk was 1:293 and 1:153, respectively (Fig. S3 D). BM LKS CD34^−^ cell analysis revealed maintenance of donor chimerism within this HSC-enriched cell fraction over 6 mo ([Fig. S4](http://www.jem.org/cgi/content/full/jem.20101643/DC1)). {#fig3} ###### Transplantation of 0×-, 0--2×-, or ≥5×-divided LKS cell into lethally irradiated mice reveals steady-state cycling heterogeneity in HSC Cells injected Cell dose Secondary recipient mice with long-term multilineage reconstitution ---------------- ----------- --------------------------------------------------------------------- ----------- ----------- 0×-divided 1 24 (5/21) 22 (4/18) 28 (5/18) 0--2×-divided 10 67 (4/6) 100 (5/5) ND 0--2×-divided 20 100 (7/7) ND ND ≥5×-divided 10 0 (0/4) 0 (0/5) 11 (1/9) ≥5×-divided 20 0 (0/7) 0 (0/7) 10 (1/10) ≥5×-divided 50 0 (0/7) 20 (1/5) 36 (4/11) ≥5×-divided 250 ND 67 (2/3) 71 (5/7) The indicated numbers of LKS cells at the indicated time points after primary transplantation were deposited into individual wells of 96-well plates. The contents of each well were injected into lethally irradiated animals along with 2 × 10^5^ total BM cells. After 4 mo, PB of transplanted mice was analyzed on flow cytometry to identify mice that were multilineage reconstituted by donor cells (above background: \>0.1% in B220^+^, CD3ε^+^, and B220^−^CD3ε^−^CD11b^+^Gr-1^+^ cells). The numbers in the time point columns represent the percentage of engrafted mice, and the parenthetical numbers represent the number of engrafted mice per number of transplanted mice. ND, not determined. We next tested serial long-term hematopoietic potential of 0×- and ≥5×-divided LKS cells by transferring whole BM cells from engrafted secondary recipients into lethally irradiated tertiary recipient mice ([Fig. 3 A](#fig3){ref-type="fig"}). Both 0×- and ≥5×-divided LKS cells were able to reconstitute BM compartments of irradiated mice for a total of 52--64 wk, i.e., \>1 yr, over serial transplantation. Donor chimerism varied depending on individual mice but not on divisional history of donor cells from primary transplants ([Fig. 3 F](#fig3){ref-type="fig"}). Given that nonconditioned i.v. transfer reflects steady-state physiological HSC circulation and BM rehoming ([@bib48]; [@bib5]; [@bib27]), these findings, in contrast to previous studies ([@bib46]; [@bib12]), directly demonstrate cycling heterogeneity in steady-state HSC; some HSCs are quiescent over months and consequently do not contribute to blood production, whereas others possess high cycling activity and, thus, likely contribute to blood cell production during this time. To determine the mean turnover rate of HSC, we calculated the number of HSC in division classes 0 and ≥5 from the repopulation assay data ([Fig. 4 A](#fig4){ref-type="fig"} and [Table S2](http://www.jem.org/cgi/content/full/jem.20101643/DC1)) and fitted a mathematical model to the number of HSC, assuming that their division rate is equal to the rate of loss, and, hence, the total HSC number in this steady-state model is constant over time (see Materials and methods). We estimate that HSCs divide, on average, every 39 d (95% confidence interval 20--50 d), leading to a total of ∼18 divisions during a lifetime of a mouse. As there was no statistical evidence for heterogeneous division rates in the functional biologically defined HSC population, this indicates that HSCs with long-term reconstitution capacity are not necessarily permanently split into subpopulations with different cycling kinetics ([@bib46]; [@bib12]). . (B) Experimental scheme of steady-state serial transplantation of cycling LKS cells into nonirradiated mice. (C) Representative dot plots of BM from secondary transplants gated on donor (CD45.2^+^) or host (CD45.1/2^+^) Lin^−^ cells at 6 wk after secondary transplantation. Dashed lines represent zero division.](JEM_20101643R_RGB_Fig4){#fig4} Changing divisional frequencies of LKS cells revealed by steady-state serial transplantation -------------------------------------------------------------------------------------------- Virus-mediated HSC marking suggested that hematopoiesis is maintained by stem cell clones that asynchronously self-renew and differentiate ([@bib23]; [@bib26]). Consistent with those findings, our serial transplantation experiments showed that the donor contribution to mature blood cells and LKS cells varied and increased in some tertiary recipients ([Fig. 3 F](#fig3){ref-type="fig"}, arrowheads). Secondary and tertiary recipients contained blood-forming cells derived from single 0×-divided HSCs and, according to limiting dilution transplantation experiments, single or double ≥5×-divided HSCs. Thus, the fluctuating hematopoietic contribution indicates that some HSC clones or their offspring vary in contribution to blood production over time. This is consistent with recent publications indicating reactivation of HSCs from quiescence over time ([@bib28]). To address the question of whether divisional frequency is kept constant or if steady-state fast-dividing LKS cells naturally vary turnover rates, we performed steady-state serial transplantation with fast-cycling cells. 4,000--6,500 LKS cells that had divided ≥5× at 14 wk after primary transplantation were reisolated, labeled again with CFSE, and subsequently transplanted into secondary nonirradiated animals ([Fig. 4 B](#fig4){ref-type="fig"}). Assuming constant divisional frequency, transferred cells should divide at least ≥2× within 6 wk. However, some donor LKS cells remained in the zero- or onefold division, demonstrating that fast-cycling cells or their progeny can slow down the divisional rate in steady state ([Fig. 4 C](#fig4){ref-type="fig"}). Increased dormancy of aged LKS CD150^+^ and expanded LKS cells -------------------------------------------------------------- Next, we examined whether steady-state aging or massive hematopoietic system expansion after lethal irradiation and BM transplantation with low HSC numbers, i.e., both states with overall increased HSC divisional history, affects cell cycle kinetics of HSC-enriched cells. CFSE-labeled LKS cells from aged and young mice were transplanted into steady-state young mice. The frequency and number of 0×-dividing or slow-dividing LKS CD150^+^ was substantially increased in mice transplanted with aged cells ([Fig. 5, A and B](#fig5){ref-type="fig"}). Similarly, CFSE-labeled LKS cells from mice that were transplanted 16 wk previously with low numbers of HSCs showed an increased ratio of slow-dividing and quiescent to fast-dividing cells (unpublished data). This data demonstrates that LKS cells, containing HSCs with an increased divisional history, have a tendency to return to quiescence. If this reflects a general biological feature of HSCs, steady-state fast-dividing HSCs will slow down their divisional rate, once given the opportunity in a permissive environment. {#fig5} Recruitment of HSC into division upon in vivo LPS challenge ----------------------------------------------------------- Immunophenotypically defined HSCs have been shown to recognize viral or bacterial components through TLRs and, subsequently, undergo division at peripheral sites to give rise to innate immune cells ([@bib29]; [@bib25]). However, there is no direct evidence that quiescent HSCs in BM increase division and self-renewal to contribute to blood cell production upon demand in vivo. We tested if LPS challenge, as a surrogate for gram-negative infection which induces massive myeloid cell production, would recruit quiescent HSCs into cycle. Mice that had been previously transplanted with CFSE-labeled LKS cells were repetitively treated with LPS and analyzed 8 d after final LPS injection ([Fig. 6 A](#fig6){ref-type="fig"}). Although 0×-divided control T cells were detectable, 0×-divided LSK cells were absent, and the number of ≥5×-divided LKS cells was increased in LPS-treated mice ([Fig. 6, B and C](#fig6){ref-type="fig"}), indicating cell loss or recruitment of LKS cells into cell cycle. To evaluate HSC potential in 2--4×- or ≥5×-divided LKS cells from PBS- or LPS-treated mice, cells were reisolated and transplanted into lethally irradiated secondary recipients ([Fig. 6 A](#fig6){ref-type="fig"}). 4 mo after transplantation, donor chimerism in peripheral blood (PB) and contribution to different lineage cells were examined. Consistent with previous experiments ([Fig. 3 E](#fig3){ref-type="fig"}), donor cell engraftment was rarely detected in secondary recipients transplanted with 2--4×- and ≥5×-divided LKS cells from PBS-injected control mice. In marked contrast, both 2--4×- and ≥5×-divided LKS cells from LPS-injected mice contributed to the generation of multilineage hematopoiesis in more than half and in some of the recipients, respectively, after 16 wk ([Fig. 6 D](#fig6){ref-type="fig"}). Lineage distribution analysis in secondary transplanted animals showed no lineage-biased repopulation, indicating that LPS challenge does not affect cell fate or differentiation potential of HSC (unpublished data). Thus, this data provides the first evidence that a naturally occurring hematoimmunological challenge, such as gram-negative bacterial infection, induces proliferation and self-renewal of HSCs, a mechanism which might have evolved to enhance fitness to rapidly amplify innate immune responses upon demand. {#fig6} DISCUSSION ========== Using a newly established HSC tracking method that utilizes in vivo CFSE dilution, we communicate three major findings on HSC turnover and contribution to blood formation in steady state and upon inflammation. The first finding is that in steady state, HSCs with equivalent life-long multilineage repopulation potential are contained in both frequently cycling cell populations that divide ≥5× in 7 wk, i.e., about every 1.4 wk, and in quiescent cells that do not divide over 14 wk. The finding is not consistent with the previous observations that only quiescent cells possess serial reconstitution capacity and fast-diving cells have limited self-renewal ([@bib46]; [@bib12]). This contradiction might arise from technical issues, as numbers of cells transferred into secondary transplants, and from sensitivity and resolution of divisional tracking methods. Another possibility would be that the dividing HSCs are not included in the population with LKS CD48^−^CD150^+^ phenotype as subsets of HSC might not express CD150 ([@bib44]). Although it was assumed that both BrdU incorporation and H2B-GFP transgenic animal models allow us to follow seven cellular divisions, our comparative analysis demonstrates that staining intensity for BrdU detection reaches a limit at two to three divisions and that BrdU staining is not linear, suggesting a lower divisional resolution of BrdU retention than expected, whereas, in contrast, CFSE dilution can distinguish at least five divisions with high resolution. Furthermore, the fact that BrdU has mitogenic activity, which has been shown in this paper and previously ([@bib20]; [@bib46]), has a substantial impact on the experimental readout because BrdU labeling changes cell cycle state as well as, potentially, consecutive function of cells. Also, most quiescent 0×-dividing HSCs are ignored in the BrdU assay, as our data shows remaining CFSE-high cells that do not incorporate BrdU. Thus, with some HSCs being deeply quiescent and inactive in DNA replication and protein synthesis, uniform labeling of all HSC by DNA labeling or marker protein expression might not be achieved at the starting point of chase. This contrasts with the CFSE labeling established in this paper that ensures highly uniform labeling and high-resolution divisional tracking of cells without impairing HSC function, independent of cell cycle activity during the labeling process. The second finding is that steady-state fast-cycling populations can slow down over time in steady-state serial transplantation and that LKS CD150^+^ cells containing HSCs with extensive proliferative history---i.e., HSCs that have gone through extended proliferation in aging or after in vivo challenge by transplantation---are prone to return to quiescence. As demonstrated in this paper, divisional activity is not associated with HSC function in young adult mice. Furthermore, HSCs have the capacity to engraft and constitute long-term hematopoiesis over several serial transplantations, indicating that high divisional history does not lead to immediate loss of HSC function ([@bib2]). HSC cycling activity is a result of extrinsic and intrinsic regulation ([@bib32]). Based on our data, we would like to suggest that steady-state fast-cycling or enhanced turnover with aging or irradiation- and transplantation-induced proliferation might activate an intrinsic HSC program that, based on divisional history, drives toward quiescence. Thus, an intrinsic cell memory effect to prevent HSC exhaustion might be counterbalanced by an extrinsic drive for proliferation. The underlying mechanisms will need to be determined in the context of environmental cues such as availability of adhesion molecules and growths factors in the putative BM HSC niche. Third, we show that in vivo TLR4 agonist challenge recruits in vivo functional quiescent HSCs into proliferation and self-renewal with nonbiased lineage repopulation capacity. Although TLR ligation on HSCs has been shown to induce cellular division and myeloid lineage-skewed differentiation ex vivo, there was no direct evidence for enhanced self-renewal of HSC in BM ([@bib29]; [@bib25]). It is not clear from our experiments if LPS executes its effect on cell cycle regulation directly via TLR4 expressed on HSCs, via an indirect pathway, or via a combination of both. Hematopoietic and nonhematopoietic cell--secreted interferons have been recently identified by several studies to drive HSC in cycle upon artificial addition or in chronic infection ([@bib11]; [@bib39]; [@bib4]). Our study extends these findings and directly demonstrates that correlates of gram-negative infections, or possibly self-damage ([@bib37]), can signal from the periphery to primary hematopoietic sites in BM and have an impact on divisional behavior of HSCs. This mechanism likely allows adequate hematopoietic responses and, at the same time, prevents loss of HSCs by differentiation. Our mathematical simulation reveals that HSCs with different cycling activity can be contained in one HSC population with relatively broad cycling variation and that, on average, HSCs divide 18× during a 2-yr lifespan of a laboratory mouse. Two principle models have been posed for the maintenance of hematopoiesis by stem cells. The clonal maintenance model suggests that all HSCs give rise to mature blood cells continuously throughout life and, thus, all HSCs should divide similarly to produce cells that contribute to blood formation ([Fig. S5 A](http://www.jem.org/cgi/content/full/jem.20101643/DC1); [@bib16]; [@bib7]; [@bib26]; [@bib20]; [@bib30]). The clonal succession model proposes that some HSCs divide frequently, contribute to hematopoiesis, and fully differentiate or die subsequently and are followed by previously quiescent HSCs that then meet the same fate (Fig. S5 B; [@bib18]; [@bib10]; [@bib46]; [@bib12]). Based on our data, we suggest a "dynamic repetition" model, where some HSCs dominate blood formation for a time, subsequently enter a quiescent state in which other HSCs increase hematopoietic contribution, and get reactivated again and contribute to blood formation in repetitive cycles (Fig. S5 C). Our data do not suggest how long active and resting phases might last or how many HSC clones at any given time contribute to hematopoiesis. However, the model of steady-state reversible change between proliferation and quiescence in HSCs over time is consistent with virtual single cell--based simulation models ([@bib13]). Furthermore, the dynamically changing cycling activity likely results in a similar turnover of the entire HSC pool, indicating a homogeneous divisional history for all HSCs at the end of life, a suggestion which would be coherent with linear telomere shortening observed in the human aging HSC pool ([@bib38]). The findings reported in this paper might represent a biological principle that could hold true for other somatic stem cell--sustained organ systems and might have developed during evolution to ensure equal distribution of work load, efficient recruitment of stem cells during demand, and reduction of risk to acquire genetic alterations by alternating fractions of stem cells in quiescence at any given time. MATERIALS AND METHODS ===================== ### Mice. C57BL/Ka-Thy1.1 (CD45.1^+^ or CD45.2^+^) mice were bred and maintained at the Institute for Research in Biomedicine animal facility. CD45.1/2^+^ F1 mice were generated by crossbreeding. 8--12-wk-old or 64--72-wk-old mice were used in the studies. Mice were treated in accordance with the guidelines of the Swiss Federal Veterinary Office, and experiments were approved by the Dipartimento della Sanità e Socialità and Gesundheitsdirektion Kanton Zürich, Veterinäramt. ### Cell isolation and sorting. Total BM cells were harvested from long bones, red blood cells were lysed with ACK lysis buffer (150 mM NH~4~Cl, 10 mM KHCO~3~, and 0.1 mM EDTA), and debris was removed with 70-µm cell strainers (BD). Cells were stained with PE-Cy5--conjugated antibodies against the lineage antigens B220 (RA3-6B2), CD3ε (145-2C11), Ter119 (Ter119), and Gr-1 (RB6-8C5) and immunomagnetically enriched for lineage-negative cells using anti-Cy5/anti--Alexa Fluor 647 Microbeads (Miltenyi Biotec). The lineage-depleted cells were subsequently stained with allophycocyanin (APC)-Cy7--conjugated antibody to c-Kit (2B8) and FITC-conjugated antibody to Sca-1 (D7), and LKS or Lin^−^c-Kit^+^Sca-1^−^ (LK) cells were sorted using a FACSAria (BD). Naive CD4^+^CD62L^+^ T cells were enriched from spleen cells with PE-Cy5--conjugated anti-CD4 antibody (GK1.5) and anti-Cy5/anti--Alexa Fluor 647 Microbeads, and were then stained with PE-conjugated anti-CD62L antibody (MEL-14) followed by sorting of CD4^+^CD62L^+^ cells on a FACSAria. All antibodies were obtained from eBioscience. ### CFSE staining and cell transplantation into nonirradiated or sublethally irradiated mice. Sorted cells were labeled for 7 min at 37°C with 2 µM CFSE (Invitrogen) in Dulbecco's PBS (D-PBS; Invitrogen) supplemented with 2% FBS (Invitrogen). The same volume of ice-cold D-PBS with 10% FBS was then added to stop the reaction. After wash with Mg^2+^/Ca^2+^-free PBS, 10^5^ CFSE-labeled LKS cells (CD45.2^+^) or 2 × 10^6^ CFSE-labeled CD4^+^CD62L^+^ naive T cells (CD45.2^+^) were i.v. transplanted into nonirradiated recipient F1 (CD45.1/2^+^) mice. In some cases, mice were sublethally irradiated with 6.5 Gy from a Cesium 137 source (BIOBEAM 8000; STS GmbH) at 3.75 Gy/min before transplantation. ### In vivo LPS challenge. 10^5^ CFSE-labeled LKS cells were transplanted into nonirradiated animals. 1 wk after transplantation, mice were i.p. injected with PBS or 35 µg LPS from *Escherichia coli* (Ultrapure; InvivoGen) 4×, each with a 2 d-interval, and were analyzed 6 d after the final injection. ### Flow cytometric analysis. Lineage-negative cells from BM were enriched from transplanted mice and stained with additional monoclonal antibodies as follows: APC-Cy7--conjugated c-Kit, PE-Cy7-conjugated anti--Sca-1, Pacific blue--conjugated anti-CD45.1 (A20), APC-conjugated anti-CD45.2 (104), PE-conjugated anti-CD150 (TC15-12F12.2; BioLegend), PE-conjugated anti-CD34 (Raw34), PE-conjugated anti-CD16/32 (2.4G2), PE-conjugated IL-7α (A7R34), PE-conjugated anti-Flt3 (A2F10.1), biotinylated anti--c-Mpl (AMM2 donated by Kyowa Hakko Kirin Co., Ltd.), PE-conjugated anti--M-CSFR (AFS98), or PE-Cy5--conjugated antilineage antibody. The biotinylated antibody was detected with PE-conjugated streptavidin. Each single division was determined as follows: a gate for zero division was set on the CFSE peak of the undivided naive T cell control, and subsequent divisions were determined according to reduced fluorescence intensity of peaks in respective histograms. The background fluorescence was determined from nontransplanted BM sample. For evaluation of peripheral engraftment, thymus, spleen, lymph nodes, and PB were harvested and stained with FITC-conjugated anti--Gr-1, PE-conjugated anti-CD11b, PE-Cy5--conjugated CD3ε, APC-Cy7--conjugated B220, PE-Cy7--conjugated CD45.1, and APC-conjugated CD45.2. Before analysis, cells were resuspended in PBS containing 2 µg/ml Hoechst 33342 (Invitrogen). The percentage of donor contribution was assessed by measuring CD45.1^+^ or CD45.2^+^ cells in B220^+^ B-lineage, CD3ε^+^ T-lineage, or B220^−^CD3ε^−^CD11b^+^Gr-1^+^ myeloid lineage cells. Cutoff for donor engraftment was determined based on the percentage of nonspecific stain in nontransplanted secondary transplanted recipient. All antibodies were purchased from eBioscience except when otherwise specified. ### BrdU retention assay with CFSE labeling. 1 wk after transfer of CFSE-labeled LKS cells into nonirradiated mice, animals were i.p. injected with a single dose of 180 µg BrdU (Sigma-Aldrich) and were fed water containing 800 µg/ml BrdU and 5% glucose for the next 14 d. Mice were then sacrificed and BrdU staining was performed using a BrdU labeling kit (BD). Cells from PBS-treated animals were used as a negative control for BrdU staining. ### Single cell and whole BM transplantation into lethally irradiated mice. 3--14 wk after primary transplantation with CFSE-labeled cells, Lin^−^ cells were immunomagnetically enriched from long bones of mice, as described in Flow cytometric analysis and stained with antibodies to CD45.1, CD45.2, c-Kit, and Sca-1. 1--250 donor-derived LKS cells (CD45.2^+^) were sorted into individual wells of 96-well plates using a FACSAria. Single cell deposition was confirmed microscopically. 2 × 10^5^ total BM cells from CD45.1^+^ or CD45.2^+^ mice were suspended in Mg^2+^/Ca^2+^-free Hank's buffered salt solution supplemented with 2% FBS and plated into each well. The content of each well was i.v. injected into mice that were lethally irradiated with 2 × 6.5 Gy in a 4-h interval. Mice were bled monthly, blood was subjected to red blood cell lysis, and donor engraftment was analyzed as described in Flow cytometric analysis. Recipients with \>0.1% donor chimerism in B-lineage (B220^+^), T-lineage (CD3ε^+^), and myeloid-lineage (B220^−^CD3ε^−^CD11b^+^Gr-1^+^) population was considered to be multilineage repopulated. The frequency of HSC in LKS cells at the indicated time points after primary transfer was determined according to Poisson statistics ([@bib40]) or L-Calc (STEMCELL Technologies Inc.). For the tertiary transplantation, one to three million cells of total BM cells harvested from secondary transplants were injected into lethally irradiated mice (CD45.1^+^ or CD45.2^+^). ### Statistical analysis. The significance of differences was determined by an unpaired Student's *t* test. ### Mathematical modeling and analysis of LKS cell division kinetics. The simplest model that can be used to describe the division kinetics of CFSE-labeled cells is given by the following differential equations ([@bib9]): d X 0 / d t = − ( λ \+ d ) X 0 d X i / d t = 2 λ X i − 1 − ( λ \+ d ) X i , i = 1 , ... , 4 d X 5 \+ / d t = 2 λ ( X 4 \+ X 5 \+ ) − d X 5 \+ . Here, *X*~*i*~ denotes the number of LKS cells that gave undergone *i* divisions, and *X*~5+~ denotes the number of LKS cells that have divided five or more times. The parameter λ is the rate constant for cell division, and *d* is the rate constant for LKS cell loss. Several processes contribute to this loss of LKS cells. These are differentiation, which leads to a loss of LKS marker, cell death, or loss of LKS marker not related to differentiation. For simplicity, we refer to *d* as the differentiation rate constant. Because the experiment involved the transfer of 5,000 CFSE-labeled LKS cells into the mice, we assume that at time *t* = 0, all cells are in the division class 0, corresponding to the highest CFSE label, and their number is *X*~0~(0) = 5,000. (In mathematical immunology, the division of cells is often described by the Smith-Martin model rather than this simple model above \[[@bib9]\]. The Smith-Martin model describes the cell cycle in more detail introducing a parameter for the time a cell requires for dividing once it committed to division. However, because the division kinetics of LKS cells is very slow, occurring on the time scale of weeks rather than hours, the time required for cell division can be ignored.) We constructed extensions of this basic model to study the heterogeneity of the division kinetics of the LKS populations. In particular, we developed versions of the basic model in which we assume two or three subpopulations, each with their own division and differentiation rates. These mathematical models are given by the following set of differential equations: d X 0 ( s ) / d t = − ( λ \+ d ) X 0 ( s ) d X i ( s ) / d t = 2 λ X i − 1 ( s ) − ( λ \+ d ) X i ( s ) , i = 1 , ... , 4 d X 5 \+ ( s ) / d t = 2 λ ( X 4 ( s ) \+ X 5 \+ ( s ) ) − d X 5 \+ ( s ) . Hereby, the parameter *s* denotes the subpopulation and is an integer ranging from 1 to the number of subpopulations considered in the model. We assume that at time *t* = 0, all cells are undivided and their number is ${\sum_{s}{X_{0}^{(s)}\left( 0 \right)}} = 5000.$ It is useful to define the initial fraction *f*~*s*~ of undivided cells in the *s*th subpopulation by $X_{0}^{(s)}\left( 0 \right) = f_{s}{\sum_{s}{X_{0}^{(s)}\left( 0 \right)}}.$ Obviously, ${\sum_{s}f_{s}} = 1.$ Unlike the models used in [@bib46] and [@bib43], this model does not assume that fast-cycling cells arise from dormant cells by differentiation. Rather, we assume that each subpopulation is independently dividing and differentiating. The most straightforward way to analyze the LKS CFSE cell data would be to successively fit the one-, two-, and three-subpopulation versions of the mathematical model and to assess if increasing the number of subpopulations increases the goodness of fit (as measured by the residual sum of squares) significantly (as determined by an *F*-test). However, a straightforward fit of a model with more than one subpopulation, in which the parameters of the model are unconstrained, does not result in consistent goodness of fit and consistent parameter estimates. We have therefore adopted a two-step procedure to estimate the parameters of the mathematical models. First, we estimate the differentiation rates *d* of the first two subpopulations from the normalized cell counts (defined in [Eq. 3](#fd3){ref-type="disp-formula"}). This leads to a constraint for the differentiation rates. Second, we estimate the division rates and the differentiation rates of the remaining subpopulations from fitting the model to the LKS cell count in the different division classes. Normalized cell counts are a way to eliminate the division kinetics from CFSE data ([@bib9]). The normalized cell count is defined as: N ( t ) = ∑ i = 0 4 X i ( t ) / 2 i . Assuming a single, homogeneous subpopulation, *N*(*t*) is an exponentially decaying function which is independent of the division rate λ: N ( t ) = N ( 0 ) e − d t . Thus, in a plot of ln(*N*(*t*)) versus the time, *t*, the slope is *d*. Assuming two subpopulations, *N*(*t*) is a bimodally decaying function that is independent of the division rates λ~1~ and λ~2~: N ( t ) = N ( 0 ) ( f 1 e − d 1 t \+ ( 1 − f 1 ) e − d 2 t ) . We fitted [Eqs. 4](#fd4){ref-type="disp-formula"} and [5](#fd5){ref-type="disp-formula"} to the normalized cell counts by minimizing the squared difference between log-transformed normalized cell counts and the predictions of [Eqs. 4](#fd4){ref-type="disp-formula"} and [5](#fd5){ref-type="disp-formula"}. We found that the two-subpopulation model fits significantly better than the one-subpopulation model (*F*-test, *df*~1~ = 28, *df*~2~ = 26, P = 2.2 × 10^−4^). Thus, there is a strong statistical signal that the LKS cell population is heterogeneous with respect to its differentiation rate. The best estimates for the parameters of [Eq. 5](#fd5){ref-type="disp-formula"} are: f 1 = 0.70 ± 0.06 d 1 = ( 1.82 ± 0.90 ) / w k d 2 = ( 0.127 ± 0.022 ) / w k . The death rates correspond to a mean life-span of 3.8 and 55 d, respectively. (Fitting [Eq. 4](#fd4){ref-type="disp-formula"}, we obtain *d* = (0.254 ± 0.020)/wk.) In the next step, we fitted the full two- and three-subpopulation model and estimated the remaining parameters of these models. The fitting routine involved minimizing the squared difference between log-transformed LKS cell counts and the predictions of the two- or three-subpopulation models. The last division class, five and greater, was omitted in the fitting procedure because we found, on the basis of simulated data, that omitting the last division class resulted in less biased parameter estimates. We found that there is evidence for at least three subpopulations in the data. A three-subpopulation model, assuming Poisson division and loss of LKS cells, fits significantly better than a model assuming two subpopulations (*F*-test, *df*~1~ = 153, *df*~2~ = 150, P = 2.4 × 10^−14^). Estimates of the parameters that characterize the three subpopulations are listed in Table S1. Hereby, *f*~1~, *f*~2~, and *f*~3~ are the initial fractions of subpopulation one, two, and three, respectively. λ~*i*~ is the division rate, and *d*~*i*~ is the loss rate of the *i*th subpopulation (as a result of differentiation, loss of marker, or death). Thus, we have evidence for a subpopulation that is lost at the high rate of *d*~1~ = 1.8/ *wk* and does not divide. Subpopulations two and three initially constitute 22 and 7.7% of the LKS population and divide once every 12 and 97 d, respectively. Subpopulations two and three correspond to the slowly and fast cycling subpopulations that were identified previously ([@bib46]; [@bib12]; [@bib43]). ### Mathematical modeling and analysis of division kinetics of biologically functional HSC. To estimate the turnover rate of biologically functional HSC from the repopulation data, we first estimated the fraction of biologically functional HSC in the LKS cell population in division class zero and five or higher by a maximum likelihood procedure. If the fraction of biologically functional HSC is *f*~HSC~, and we transfer *i* LKS cells to *m* mice, the probability to observe repopulation of the hematopoietic system in *r* mice is: ( m m − r ) ( 1 − f H S C ) i ( m − r ) ( 1 − ( 1 − f H S C ) i ) r . If we conduct *E* such repopulation experiments, the best estimate for the fraction of biologically functional HSC in the LKS population maximizes the following likelihood: L ( f H S C ) = ∏ k = 1 E ( m k m k − r k ) ( 1 − f H S C ) i k ( m k − r k ) ( 1 − ( 1 − f H S C ) i k ) r k . In this expression, the variables *m*~*k*~ or *r*~*k*~ denote the number of mice used or repopulated in experiment *k*, respectively, and *i*~*k*~ is the number of LKS cells transferred in experiment *k*. We used the repopulation data to estimate the fraction of biologically functional HSC in division classes zero and five or higher at 3, 7, 12, and 14 wk after primary transfer of the LKS cell population. Combining these estimates with the number of LKS cells in each division class, we obtain an estimate for the number of biologically functional HSC in the division classes zero and five or higher. Table S2 shows these estimates. We then used the homogeneous cell division model ([Eq. 1](#fd1){ref-type="disp-formula"}) to estimate the turnover rate of biologically functional HSC. We assumed that the number of biologically functional HSC stays constant over the 14 wk. This constant number of biologically functional HSCs can be interpreted as the number of stem cell niches that are occupied after the transfer of CFSE-labeled cells. Assuming a constant number of biologically functional HSCs is mathematically equivalent to setting the division rate equal to the rate of differentiation or loss of biologically functional HSC. We estimated a division rate of 0.18/wk, which corresponds to one division every 39 d. The 95% confidence interval of this estimate is 0.14--0.35/wk, which corresponds to a range of one division every 20--50 d. This is strong evidence that biologically functional HSCs do not divide only every 150 d, as has been suggested by other studies ([@bib46]; [@bib47]; [@bib42]). The constant number of biologically functional HSCs in the system (i.e., the number of occupied stem cell niches) is estimated as 388 cells (95% confidence interval: 193--559 cells). We investigated whether relaxing some of the model assumptions alters the estimate of biologically functional HSC dramatically but found that the estimate of the turnover rate is robust. If we relax the assumption that the division rate equals the rate of differentiation and loss, we obtain a division rate of 0.26/wk, which lies within the 95% confidence interval of the estimate assuming equal division and differentiation rates. This means that the two estimates are not significantly different. If we set the number of niches to 200 rather than estimating this quantity from the data, we obtain a division rate of 0.24/wk, which is, again, not significantly different from the estimate based on flexible number of niches. Lastly, a two-population model does not significantly improve the fit to the data on biologically functional HSC (Table S2): *F*-test, *df*~1~ = 29, *df*~2~ = 27, P = 0.15. Thus, there is no evidence for a heterogeneously dividing population of biologically functional HSC. ### Implementation. All the mathematical models and the analysis described in the previous two sections were implemented in the R language of statistical computing ([@bib36]). ### Online supplemental material. Fig. S1 shows homogenous CFSE labeling and CFSE label--retaining cell populations upon nonconditioned transplantation of LKS but not LK. Fig. S2 shows 0×-divided cells that are maintained in BM, but not spleen, and included in LKS CD34^−^CD150^+^ population. Fig. S3 shows donor contribution to lymphoid lineage in secondary transplantation with 0×- or ≥5×-divided LKS cells, and HSC frequency determined by limiting dilution transplantation. Fig. S4 shows donor chimerism in HSC-containing population in BM of secondary recipient reconstituted with 0--2×- or ≥5×-divided LKS cells. Fig. S5 shows hypothetical models for steady-state hematopoiesis. Table S1 shows estimates of the parameters of the three-subpopulation model on LKS turnover. Table S2 shows estimate of biologically functional HSC. Online supplemental material is available at <http://www.jem.org/cgi/content/full/jem.20101643/DC1>. The authors thank D. Jarrossay for FACS sorting, and D. Bossi and C. Borsotti for assistance with some experiments. This work was supported in part by a Postdoctoral Fellowship of the Japanese Society for the Promotion of Science for Research Abroad to H. Takizawa, the Swiss National Science Foundation (310000-116637), the Oncosuisse (OCS-02019-02-2007), and the Promedica Foundation to M.G. Manz. The authors declare no competing interest. Author contributions: H. Takizawa designed research, performed the experiments, and wrote the manuscript; C.S. Boddupalli performed experiments; R.R. Regoes and S. Bonhoeffer generated mathematical models; and M.G. Manz directed the study and wrote the manuscript. Abbreviations used:APCallophycocyaninFlt3Fms-like tyrosine kinase 3H2B-GFPhiston 2B GFPHSChematopoietic stem cellLKLin^−^c-Kit^+^Sca-1^−^LKSLin^−^c-kit^+^Sca-1^+^PBperipheral bloodTLRToll-like receptor [^1]: H. Takizawa and R.R. Regoes contributed equally to this paper.

Introduction {#sec1-1} ============ Vesico-vaginal fistula is commonly caused by obstructed labor, gynecological surgery, occasionally due to malignancy and rarely foreign body; in most cases being ring pessaries.\[[@ref1]\] Cases of a foreign body in the vagina have been reported mostly in developed countries.\[[@ref1][@ref2][@ref3][@ref4]\] Foreign body within the vagina can result in major complications if left unmanaged for a long period of time.\[[@ref5][@ref6]\] We present a case of forcefully inserted foreign body within the vagina that resulted in extensive vesico-vaginal fistula. Case Report {#sec1-2} =========== A nulliparous 19-year-old female presented with a 3 week history of a foreign body in the vagina. Her uncle allegedly forcefully inserted the foreign body after she refused a pre-arranged marriage. The presence of this foreign body was associated with purulent and offensive vaginal discharge, dysuria and leakage of urine from the vagina. On examination she was afebrile with stable vital signs. The patient was found to have suprapubic tenderness on palpation. Examination of the genitalia could not be performed due to severe pain. A plain pelvic radiograph showed an opacified object within the pelvic cavity \[[Figure 1](#F1){ref-type="fig"}\]. {#F1} The clinical radiographic impression was a foreign body in the vagina with associated bladder injury. The patient was examined in theatre under general anaesthesia. The findings were perineal excoriation and clearly identifiable leakage of urine from the vagina. A plastic container approximately 4 cm in diameter was found invading into the vaginal musculature. With the successful removal of the container, a mid-sized vesico-vaginal fistula was noted measuring 2 cm in diameter with significant surrounding fibrosis and debris. A small stone was visualized in the bladder and removed vaginally through the fistula. The patient was then transferred to the ward for interim treatment of a Sitz bath to clean the fistula for later repair. After 10 days the fistula was deemed operable. A 4 cm vesico-vaginal fistula and considerable associated fibrotic tissue were identified \[[Figure 2](#F2){ref-type="fig"}\].\[[@ref5]\] The ureteric orifices were far from the fistula margins. Hydro-dissection was performed using a specially formulated solution (jungle juice). Fistula margins were dissected with scalpel, then scissors to separate the bladder from the vaginal tissue. Some scarred tissues were excised. Bladder tissue was freed from tension. Repair of the bladder was commenced in a transverse direction, inverting the tissue inward using interrupted sutures (polysorb \#2-0). Water-tight repair of the bladder was achieved by single layer closure and confirmed by negative dye test (methylene blue). Then the repair of the vaginal wall with polysorb suture \#0 was performed in a transverse direction. The post-operative course was unremarkable. The Foley catheter was removed on day 14 with no leakage of urine through the vagina. Despite involvement of a counselor, psychiatrist and hospital social worker, the patient chose not to initiate legal proceedings. {#F2} She was discharged with advice to abstain from intercourse for 3 months and we have recommended elective caesarean section should she fall pregnant in the future. Discussion {#sec1-3} ========== Similar cases have been described by other institutes.\[[@ref2][@ref3][@ref4][@ref5][@ref6]\] In these cases foreign bodies in the vagina have been noted to not only cause fistulas, but have also resulted in ureteric obstruction due to formation of bladder stones and ascending infection with possible peritonitis.\[[@ref4][@ref5][@ref6][@ref7]\] Since our patient presented late, the foreign body embedded deep within the vaginal tissue causing significant fibrosis. This increased the difficulty of repair.\[[@ref2][@ref3]\] During the initial examination under anesthesia the fistula appeared to be small, but we subsequently discovered the actual deformity to be more than twice the anticipated size after clearing of debris and desquamation. This incident has significant implications for the patient physically, psychologically and socially. We have endeavored to provide support for her psycho-social rehabilitation but with limited engagement. In future pregnancy we have recommended elective caesarean section as the mode of delivery. As a nulliparous woman, predicting the outcome of her vaginal delivery would be uncertain (untried pelvis). Even an uncomplicated vaginal delivery may result in the breakdown of tissue and recurrence of this fistula. First repair of fistula has a significantly better outcome than repeated repair.\[[@ref8]\] The finding of fibrotic tissue in this case means that if one had to perform a repeat repair, successful closure is not guaranteed. When a physician is faced with a case of the foreign body within the vagina, we recommend urgent removal of the foreign body, preferably under general anesthesia. However, if the history or physical examination reveals prolonged exposure, repair should be delayed to allow for adequate debridement in order to prevent any life-threatening complication. **Source of Support:** Nil. **Conflict of Interest:** None declared.

The limiting factor for improving solar-cell efficiency lies in the simple physics that single-photon absorption generates one electron-hole pair[@b1]. The possibility of generating multiple charge carriers per photon, known as carrier multiplication (CM) or multiple exciton generation (MEG), is of crucial importance for developing efficient solar-cell devices[@b2][@b3][@b4][@b5][@b6][@b7][@b8]. Semiconductor quantum dots (QDs) represent well-defined structures to explore the quantum limit of harnessing solar-conversion efficiency[@b9][@b10][@b11][@b12][@b13]. By engineering the sizes of QD composites, it has been demonstrated that not only the optical properties[@b14][@b15], but also the MEG efficiency in QDs can be modified[@b16]. MEG in a photo-excited QD system is a prominent route for enhancing the conversion efficiency because carriers confined in spatial dimensions that are smaller than the bulk exciton Bohr radius lead to the formation of discrete excitonic states such that efficient MEG is possible either by suppressing the ultrafast electron-phonon relaxation[@b4][@b17][@b18] or by enhancing the Coulomb interactions via reduced dielectric screening at the QD surface[@b19]. Numerous investigations have shown that the kinetic origin of MEG dynamics in QDs is intrinsically complex because the photo-generated single exciton initially suffers from extremely fast intra-band relaxation[@b20][@b21][@b22], whose interaction time-scale is typically in the range of a few ps[@b6]. To enhance the MEG efficiency, it is desirable to circumvent the ultrafast energy-loss intra-band process[@b23][@b24]. Recent studies suggest that the MEG is an instantaneous phenomenon occurring before the intra-band energy relaxation[@b25] via virtual single excitonic[@b26] or biexcitonic optical transition[@b27] or coherent superposition among multi-exciton states[@b12]. Other investigation suggests that the intra-band relaxation rate competes with the MEG formation rate[@b6]. The above mentioned photo-physical complexity of MEG is largely due to the nature of intrinsic multi-particle (or multi-exciton) interaction[@b28]. When more than two excitons are created under high-energy excitation condition, the lowest lying energy state is not the single exciton; the mutual interaction between two excitons results in the formation of a Coulomb-correlated two excitonic state, called biexciton[@b29][@b30][@b31][@b32][@b33]. The biexciton is energetically more stable than the single exciton such that it exists below the single-exciton state[@b32][@b34]. Recent studies have reported that the final biexciton density strongly influences the solar-conversion efficiency[@b25][@b26][@b35]. Although it is important to study the impact of the MEG on the transient biexciton spectra, no experimental investigations have been provided to compare the MEG-induced biexciton dynamics with the intra-band relaxation dynamics. The key experimental observation in this study is that the optically-induced MEG is an extremely fast process, arising before the intra-band relaxation. By exploring the lowest observable biexciton dynamics, we directly measure that the biexciton bleaching comes from early emergence of the photo-induced MEG, in which the effect of extra-exciton generation is manifested by the increased broadening of the biexciton linewidth via multi-exciton interaction. Note that, in contrast to the conventional single-exciton MEG spectroscopy[@b11][@b36][@b37][@b38][@b39], our ultrafast time-resolved experiments were performed both in the MEG and in the non-MEG regimes via photon-energy and density-controlled measurements on the single- and biexciton spectra. Results ======= Single-exciton MEG dynamics --------------------------- [Figure 1a](#f1){ref-type="fig"} shows data for the broadband optical absorption of the colloidal semiconductor PbS QDs and [Fig. 1b](#f1){ref-type="fig"} shows a schematic for the ultrafast pump-probe measurements (See method for the detailed description of sample preparation and ultrafast spectroscopy). The lowest single-exciton bandgap energy *E*~x~ is identified as 0.93 ± 0.01 eV, and the ground-state biexciton energy *E*~xx~ is estimated to be 0.87 ± 0.03 eV[@b30][@b31][@b32][@b33]. Before the discussion on the biexciton dynamic, it is instructive to present detailed measurements on the intra-band relaxation dynamics because the linewidth broadening of single excitons and biexcitons is necessary related to the competing relaxation rate between the MEG and the intra-band dynamics, in which the time scale of the intra-band relaxation is typically a few ps[@b16][@b40][@b41], comparable with the MEG time scale. In the experiment, the colloidal semiconductor PbS QD sample was pumped by two different pump-photon energy *E*~pump~ with 1.55 eV and 3.10 eV, and the average number of initially photo-generated excitons per QD 〈*N*~0~〉, or initial exciton occupancy, was controlled from 0.1 to 2.2 to investigate the photon density-dependent *E*~x~ dynamics. In order to determine the intra-band relaxation rate, we measured the *E*~x~ dynamics in a short Δ*t* range between −1 ps and 7 ps as shown in [Figs. 2a and b](#f2){ref-type="fig"}. By examining the rising edge of the *E*~x~ peak, we show that the relaxation process is completed at pump-probe delay Δ*t* = 1 ps for 1.66*E*~x~ excitation (non-MEG regime) and Δ*t* = 2 ps for 3.3*E*~x~ excitation (MEG regime). This 2 ps time constant is consistent with prior experimental studies of hot-carrier MEG dynamics in PbS quantum dots, where the reported value of intra-band relaxation is in the range of 2--2.5 ps[@b16][@b40][@b41]. [Figure 2c](#f2){ref-type="fig"} shows the *E*~x~ transients excited by low *E*~pump~ ( = 1.66*E*~x~). The observed step-like signals with a small *A*/*B* ratio (amplitude ratio of the early to late pump-probe delay Δ*t*) are not attributed to the MEG transients, because the MEG typically requires *E*~pump~ greater than a few *E*~x~. When the QDs are excited by high *E*~pump~ ( = 3.3*E*~x~), we observed fast (90 ps) and slow decay (\~100 ns) components with a large *A*/*B* ratio, as depicted in [Fig. 2d](#f2){ref-type="fig"}. The experimentally determined *A*/*B* ratio of the QD occupancy was modelled via Poisson statistics ([Fig. 2e](#f2){ref-type="fig"})[@b42]. Since multiple excitons generated by the MEG decay via Auger recombination, the amplitude at long Δ*t* (denoted by *B* in [Fig. 2c and d](#f2){ref-type="fig"}) provides a scaling factor for calculating the exciton multiplicity 〈*N~x~*〉 = *A*/*B*, where *A* is the amplitude of single-exciton population immediately after pump excitation (denoted by *A* in [Fig. 2c and d](#f2){ref-type="fig"}). By comparing the measured *A*/*B* ratios in the limit of 〈*N*~0~〉 → 0, a strong indication of the MEG for the 3.3*E*~x~ pump was identified[@b43]. As reported previously[@b16][@b36][@b38][@b40][@b44], these observations confirm that the typical MEG dynamics are observable via probing the *E*~x~ dynamics. Transient Stark shift and biexciton linewidth broadening -------------------------------------------------------- The central issue to address in this paper is to investigate how the biexciton dynamics is influenced by the early formation of MEG. [Figures 3a and b](#f3){ref-type="fig"} display the biexciton transients for the 1.66*E*~x~ pump and 3.3*E*~x~ pump as a function of Δ*t* with controlled excitations from 〈*N*~0~〉 = 0.22 to 〈*N*~0~〉 = 2.2. Immediately after pump excitation, the photo-induced absorption (PA) exhibits rapid bleaching at *E*~xx~ within the first Δ*t* = 400 fs with a much larger PA peak for the 3.3*E*~x~ pump than the 1.66*E*~x~ pump. While both signals decay non-exponentially, the signals pumped by 1.66*E*~x~ decay to zero after a few ps, and the transients pumped by 3.3*E*~x~ change their signs from positive to negative near Δ*t* = 2 ps. In a strong quantum-confinement regime, the pump-created local electric field induces a large transient shift of absorption, a phenomenon known as transient Stark shift[@b42][@b45]. This effect is more considerable with increasing photo-generated carriers, which in turn produces a stronger local field and complicates the ultrafast PA spectra as schematically shown in [Fig. 3c](#f3){ref-type="fig"}. Note that the increased carrier density is reflected both by the carrier-induced Stark shift and by the absorption linewidth Γ that leads to a broader feature[@b46][@b47]. As discussed later, this broadened Γ directly determines the effect of MEG on the biexciton dynamics through extra-exciton generation. It is expected that high *E*~pump~ excitation, larger than *E*~x~, enhances the Γ broadening due to the extra-exciton generation. Immediately after the pump (Δ*t* = 400 fs), we clearly observe that the biexciton Γ is broader for the 3.3*E*~x~ excitation case than for the 1.66*E*~x~ one, as shown in [Figs. 3d and e](#f3){ref-type="fig"} with two different excitations of 〈*N*~0~〉 for each *E*~pump~ excitation. Thus, the observed transient PA dynamics can be understood by combined effects of both the carrier-induced transient Stark shift and the MEG-induced biexciton Γ broadening. We additionally notice that the spectrally-integrated areas of the broadened biexciton absorption remain the same regardless of 〈*N*~0~〉 as shown [Fig. 3f](#f3){ref-type="fig"}. This constraint indicates that the broadening is determined by the number of excitons, and it ensures that the biexciton PA peak is reduced by the exciton-exciton collision-induced broadening rather than the phase-space filling argument[@b46]. Quantitative analysis of the MEG-induced biexciton broadening and the early emergence of MEG -------------------------------------------------------------------------------------------- The entire pump-induced changes of the absorption spectra can be faithfully fit via the following third-order susceptibility function[@b31][@b33], where *E~L~* is the electric field of the pump, Δ*~XX~* is the biexciton binding energy, and *μ~X~* and *μ~XX~* are the transition dipole moments from the ground state to *E*~x~ and to *E*~xx~, respectively. The first term represents the bleaching at *E*~x~ and the second term represents the PA at ground-state *E*~xx~. For the PA dynamics measured at Δ*t* = 400 fs ([Figs. 3d and e](#f3){ref-type="fig"}), because the intra-band relaxation time (2 ps) is longer than Δ*t* of 400 fs, the absorption change measured at *E*~x~ was not induced by the single-exciton state filling. In addition, Auger recombination and impact ionization (Auger processes) can be neglected because the time-scale of Auger processes is much slower (100 \~ 200 ps) than the intra-band relaxation. On the other hand, the difference in Γ, obtained from a fit of [equation (1)](#m1){ref-type="disp-formula"} to the measured PA spectra, shows that the broadening is associated with the MEG-induced biexciton broadening. For quantitative analysis, the biexciton Γ is plotted as a function of the average number of total excitons per QD 〈*N~x~*〉, and the results are displayed in [Fig. 3g](#f3){ref-type="fig"}. Here, we note that the definition of 〈*N~x~*〉 (obtained from the measured *A*/*B* ratios in [Fig. 2c](#f2){ref-type="fig"}) differs from that of 〈*N*~0~〉 in a sense that 〈*N~x~*〉 includes both the average number of initially photo-generated excitons and the MEG-induced excitons per QD; 〈*N*~0~〉 is the average number of photo-generated exciton per QD[@b11]. In other words, the biexciton broadening is directly related to the total number of excitons 〈*N~x~*〉, not by the initial exciton occupancy 〈*N*~0~〉. By plotting the Γ as a function of 〈*N~x~*〉, we obtain a linear relationship of where *γ* ( = 6.8 meV per exciton) is the Γ broadening parameter per exciton. Because Γ(0) represents the linewidth broadening in the absence of photo-generated excitons, the value should corresponds to the *E*~x~ broadening in [Fig. 1a](#f1){ref-type="fig"}. A simple Gaussian fit shows that the *E*~x~ broadening in [Fig. 1a](#f1){ref-type="fig"} is 100 ± 5 meV, well corroborated with the fitted Γ(0) = 98 meV of the biexciton broadening. The characteristic broadening of Γ with increasing 〈*N~x~*〉 entails the effect of MEG, i.e. as more excitons are injected, more broaden feature of biexciton Γ is expected. Discussion ========== The early emergence of the MEG is substantiated by measuring the single- and biexciton spectra before/after the intra-band relaxation of 2 ps. It is expected that Γ should be large if Δ*t* is shorter than the intra-band relaxation time, i.e. if the MEG-induces exciton-exciton scattering occurs earlier than the intra-band relaxation, Γ before the intra-band relaxation is larger than Γ after intra-band relaxation. [Figures 4 (a) and (b)](#f4){ref-type="fig"} show the PA signals at Δ*t* = 1 ps. As expected, the Γ broadening at Δ*t* = 1 ps is smaller than at Δ*t* = 400 fs, but larger than at Δ*t* = 2 ps. [Figures 4c and d](#f4){ref-type="fig"} show the spectra at Δ*t* = 2 ps for the 1.66*E*~x~ pump and for the 3.3*E*~x~ pump, respectively. The Γ at 2 ps for 3.3*E*~x~ with 〈*N*~0~〉 = 2.2 is 110 meV while the Γ at Δ*t* = 400 fs with same condition is 123 meV. Indeed, we clearly see that Γ at Δ*t* = 2 ps is smaller than that of before intra-band relaxation both for the two 〈*N*~0~〉 excitations (see [Fig. 3e](#f3){ref-type="fig"} and [Figs. 4b and d](#f4){ref-type="fig"}). Long after the intra-band relaxation finishes, the carrier-induced Stark shift becomes weak, and the single-exciton state filling is dominant ([Figs. 4e--h](#f4){ref-type="fig"}). As schematically shown in [Fig. 4i](#f4){ref-type="fig"}, the weak Stark shift is rendered as the absence of PA signals at 0.85 eV, but the effect is not completely vanished; negative PA peaks appear at 0.97 eV instead of the single-exciton energy of 0.93 eV in [Fig. 1a](#f1){ref-type="fig"}. Because the PA peak is proportional to the generated exciton numbers, the magnitude of bleaching is larger for the case of 3.3*E*~x~ pump than the 1.66*E*~x~ pump case. We note that the chosen two *E*~pump~ (1.66*E*~x~ and 3.3*E*~x~) set the below and upper limit on the occurrence of MEG such that the observed two dynamics (before and after the intra-band relaxation) are distinguishable in comparing the MEG-induced biexciton lineshape and the single-exciton-dominated one. There is a possibility that significant re-shaping of single-exciton spectra can be observed at longer Δ*t*, which may occur when as many as 50% of QDs are occupied by multiple electron-hole pairs (i.e. 〈*N*~0~〉 \< 1). This scenario can be excluded in our investigation because the PA peaks at Δ*t* = 2 ps show negligible energy shifts[@b48] even when 〈*N*~0~〉 \> 1. To investigate the effect of Auger and single-exciton recombination on Γ, we compare the PA spectra at Δ*t* of 10 ps and 500 ps. We noted that the single-exciton decay dynamics consists of two relaxation components (see [Figs. 2c and d](#f2){ref-type="fig"}): one is "fast" Auger recombination (known as biexcitonic relaxation component[@b6]) and another is "slow" single-exciton recombination (referred to as excitonic background[@b6]). [Figures 4e and f](#f4){ref-type="fig"} display the PA spectra at Δ*t* = 10 ps. Because the Auger recombination is not completed, Γ at Δ*t* = 10 ps is smaller than Γ at Δ*t* = 2 ps. After the Auger recombination is finished, 〈*N~x~*〉 at Δ*t* = 500 ps approaches one both for the 1.66*E*~x~ and 3.3*E*~x~ pump cases. Because nearly one exciton is left at Δ*t* = 500 ps, Γ for both *E*~pump~ ([Figs. 4g and h](#f4){ref-type="fig"}) is identical with Γ of 100 meV, representing negligible effect of single-exciton recombination on Γ. The measured data are summarized in [Fig. 4j](#f4){ref-type="fig"}. Two main aspects are addressed. First, Γ at Δ*t* = 400 fs is the largest compared to the Γ at Δ*t* \> 400 fs, providing an evidence for the large biexciton Γ broadening in early Δ*t*. Second, by observing the fact that the decreasing slope of Γ with Δ*t* for 3.33*E*~x~ excitation is steeper than the 1.66*E*~x~ excitation up to Δ*t* = 2 ps, we can find that the effect of MEG on Γ is strongly influenced by extra-exciton generation before the intra-band relaxation. To conclude, we have investigated the transient dynamics of biexciton, located below the single-exciton energy, and have explored the impact of MEG on the biexciton spectra. Our ultrafast spectroscopy shows that the linewidth broadening of the biexciton spectra provides direct evidence on the early emergence of the MEG compared to the intra-band relaxation time. We additionally have presented quantitative analysis that the broadening parameter Γ per exciton increases linearly with increasing the total number of excitons. For detailed time-resolved spectral analysis, the PA spectra are compared with single-exciton ones at Δ*t* = 400 fs and longer delays. The comparison underscores that Γ broadening before Δ*t* = 2 ps is larger than the Γ after Δ*t* = 2 ps, corroborating that the MEG indeed occurs before the intra-band relaxation. Methods ======= Synthesis of PbS quantum dots ----------------------------- Our PbS colloidal quantum dots are capped using eleic acid and dispersed in toluene. The synthesis of the sample followed a procedure that used standard air-free solution based technique[@b49]. In a typical synthesis, 2.0 mmol of PbO (0.445 g), 8.0 mmol (2.25 g) of oleic acid (OA), and 9.9 mmol (2.5 g) of 1-octadecene (ODE) are placed in a flask and heated to 100°C under vacuum, and then nitrogen was introduced. The temperature was controlled to the appropriate injection temperature (100 to 150°C) to obtain the desired particle size. The sulfur precursor was prepared by mixing bis(trimethylsilyl)sulfide with ODE. Removal of excess ligand was completed by repeated the followings: precipitation in acetone, centrifugation of the particles, and dispersion in toluene. PbS QDs and ultrafast spectroscopy ---------------------------------- The sample used in this experiment is semiconductor colloidal PbS QDs dispersed in toluene with an average diameter of approximately 5.1 nm. The broadband optical absorption is measured by a Fourier transform infrared (FTIR) spectrometer (Bomem DA8). For the ultrafast pump-probe spectroscopy, the colloidal PbS QDs are maintained in a 3-mm cell contained in the toluene liquid with two optically-transparent MgO windows, and the samples are actively stirred using a magnetic stirrer to ensure that photo-charging does not occur during the measurements ([Fig. 1b](#f1){ref-type="fig"})[@b50]. Using a 250 kHz Ti-sapphire regenerative amplifier (Coherent RegA 9050), the samples are excited by 50 fs pulses with a pump-photon energy *E*~pump~ of 1.55 eV and its second harmonic *E*~pump~ of 3.10 eV for investigating the MEG photo-dynamics. A fraction of the amplifier output is used as a probe pulse with photon energy *E*~probe~ of 0.93 eV for the lowest *E*~x~ and 0.87 eV for the *E*~xx~. Both probe pulses are delivered from wavelength-tunable optical parametric amplifier (Coherent OPA 9850). Author Contributions ==================== H.C. and Y.H.L. developed the original experimental ideas. Y.C. and S.S. performed the ultrafast pump-probe measurements. Y.C. and S.C.L. prepared the colloidal QD samples and analyzed the data. The manuscript was written through contributions of all authors. The work at Yonsei was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (NRF-2011-0013255, NRF-2011-220-D00052, NRF-2011-0028594, NRF-2009-0083512, NRF-2012R1A1A2043180) and the LG Display Academic Industrial Cooperation Program. S. C. Lim and Y. H. Lee at SKKU are grateful for the support by Institute for Basic Science (grant number EM 1304). {#f1} ![Single-exciton MEG dynamics.\ Single-exciton dynamics in a short pump-probe range between −1 and 7 ps for 1.66*E*~x~ (a) and 3.3*E*~x~ excitations (b). The differential transmission signals probed at *E*~x~ with the 1.66*E*~x~ pump (c) and the 3.3*E*~x~ pump (d) are displayed with 〈*N*~0~〉 ranging from 0.11 to 2.2. Here, 〈*N*~0~〉 is estimated via 〈*N*~0~〉 = *j~p~σ~a~*, where *j~p~* is the pump fluence in unit of photons per cm^2^ and *σ~a~* is the absorption cross-section in unit of cm^2^ [@b14]. (c) The experimentally measured *A/B* ratio is shown for the 1.66*E*~x~ pump (filled circle) and for the 3.3*E*~x~ pump (filled square). The calculated *A/B* ratio is obtained via Poisson distribution of the QD occupancies for the 1.66*E*~x~ pump (solid line) and for the 3.3*E*~x~ pump (dashed line).](srep03206-f2){#f2} {ref-type="disp-formula"}. The obtained biexciton Γ for the 1.66*E*~x~ pump with 〈*N*~0~〉 = 1.1 and 2.2 are 106 meV and 107 meV, respectively, and those for the 3.3*E*~x~ pump with 〈*N*~0~〉 = 1.1 and 2.2 are 118 meV and 123 meV, respectively. (f) Spectrally integrated areas of the broadened biexciton absorption. (g) The biexciton Γ broadening linearly increases with increasing the total number of excitons. The experimentally determined Γ (black filled squares) from the Figs. 3d and e are compared to the [equation (2)](#m2){ref-type="disp-formula"}.](srep03206-f3){#f3} {#f4}

Introduction ============ Wagner's Chronic Care Model (CCM) \[[@B1]\] (extended in 2003) \[[@B2]\] has been used as a best practice model in the delivery of care to patients with chronic and complex conditions. The CCM provides guidance on what elements should be adopted to improve chronic care management in practice with the CCM being particularly focused on primary care \[[@B3]\]; paying attention to factors at micro (patient/carer and provider), meso (organization) and macro (system/policy) levels. While outcomes related to CCM \[[@B4]\] and implementation of CCM \[[@B3]\] have been explored, there is limited understanding of factors facilitating the adoption of the CCM in ***integrated community-based primary health care (ICBPHC) settings*** to appropriately care for patients with complex care needs. With their multiple co-occurring chronic illnesses and social complexity \[[@B5]\], this patient population is most in need of a patient-centred approach to chronic disease management \[[@B6][@B7]\], however strategies to address multi-morbidity have mainly focused on formal health service provision through general practice and secondary care that are often fragmented with poor coordination of care \[[@B8][@B9][@B10][@B11]\]. Linking together multiple episodes of care offered by providers across the health system (such as the services identified in the CCM) has been identified as central to supporting patient-centredness \[[@B7]\]. Furthermore, core components of patient-centredness, such as integration of medical and non-medical care, coordination and continuity \[[@B12]\] are supported through integrated models. In order to better understand the implementation of innovative models of ICBPHC, the *Implementing Integrated Models of Primary Health Care for Older Adults with Complex Needs (iCOACH)* research team takes a comprehensive "whole systems" approach, incorporating a policy analysis, organizational, provider and patient/carer level perspectives on implementing and delivering integrated care. "Whole systems" approaches recognize that relationships between different hierarchical levels will impact on effectiveness of the overall health system \[[@B13]\], and have been identified as essential to support our understanding of implementation of new models of care \[[@B14]\]. Each level of analysis is led by one of four research teams who have developed robust conceptualizations of pertinent issues relevant to their level of analysis and disciplinary approach. Three theoretical frameworks are brought to bear on this research question including: 1) Ham's Ten Characteristics of the High-Performing Chronic Care System \[[@B15]\] (which captures key elements of the CCM) to capture patient/carer and provider level perspectives; 2) Organizational Context and Capabilities for Integrating Care (CCIC) framework \[[@B16]\] to capture organizational and provider level perspectives; and 3) the Health Policy Monitor (HPM) \[[@B17]\] survey methodology to capture system and policy level perspectives. See Wodchis et al. \[[@B18]\] and Breton et al. \[[@B19]\] in this issue for an overview of the research program and case study summaries. The challenge, however, was in the identification of how frameworks used in parallel to study the same cases of ICBPHC overlapped. What is required is conceptual linkage when looking across different theoretical frameworks and disciplinary approaches. Conceptual linkage, we argue, is essential in studies of complex interventions like models integrated care which often rely on multiple *mechanisms of action* that may relate to different theories and disciplinary perspectives \[[@B20]\]. It is also particularly important for studies engaged in "whole systems" health services research, which requires that we draw on multiple disciplines. For the iCOACH team, we required clarification of conceptual linkages in order to: *1) ensure that data collection at the case sites efficiently captures important variables across levels*; and *2) guide data analysis across perspectives and disciplinary approaches*. Concept mapping \[[@B24]\] offers a useful approach to identify how seemingly disparate theoretical frameworks can be brought together in a single study. The **aim** of the paper is to link concepts across different theoretical framework to guide the analysis in "whole systems" on integrated care for complex patients. As such, we used a concept mapping method to answer the research question: *How are concepts across different theoretical frameworks used to guide the iCOACH study linked?* Theoretical approaches used by the iCOACH team ============================================== Team structure -------------- The iCOACH research team is organized into four teams (patient/carer, provider, organization, and policy) comprised of both Canadian and New Zealand researchers. Each brought forward a disciplinary perspective and theoretical lens with which to study implementation of ICBPHC models. Each team developed a protocol for data collection that would capture issues pertinent to their level of analysis. While four protocols were then generated, data collection at case sites occurred concurrently, requiring careful consideration of how protocols overlapped in terms of the data collection and analysis. This section provides an overview of the theoretical approach used by each team. Patient/carer research team --------------------------- The patient/carer research team sought to understand ICBPHC from the perspective of the patients and carers being served by the model. The team adopted Ham's ***Ten Characteristics of the High-Performing Chronic Care System framework*** \[[@B15]\]; a heuristic model developed to identify factors enabling the implementation of the CCM. The model was developed through an exploration of the evidence on implementation of the CCM, and includes a range of factors that relate to access, disease management, and service design and delivery. To capture key elements of the Ham's framework and the CCM related to the patient and carer perspectives, the Patient Assessment of Chronic Illness Care (PACIC) instrument \[[@B21]\] was adopted, and supplemented by instruments that enquired in more depth about financial deprivation and holistic assessment of health outcomes from health care, including physical, mental, and spiritual relationships (related to patient and carer demographics). Key elements from Ham's framework related to the patient and carer perspective captured using the PACIC include: 1. *Self-management and prevention*: Patient's perspective on their role in their health care and whether they have adequate supports to self-manage. Also related to chronic disease management and illness prevention activities. 2. *Care coordination*: Patient perspective on alignment of care across different providers. 3. *Access to services (and types of services delivered)*: Patient's perceived ability to get support in accessing needed services in the clinic, by specialists and services in the community. The provider research team -------------------------- The provider research team was also guided by Ham's framework in their approach to understanding implementation of ICBPHC. In order to collect key variables from this framework pertinent to the provider perspective, the team used the *Assessment of Chronic Illness Care* (ACIC) tool, a survey designed to capture key components of the CCM \[[@B22]\]. Domains from Ham's framework were thus captured along the lines of the 6 key components of the ACIC tool \[[@B23]\] including: 1. *Organization of delivery system*: Orientation of the health system in relation to delivery of chronic illness care. 2. *Self-management support*: Empowering and preparing patients to manage their health and health care through emphasizing the patient's central role in managing their health, use of effective self-management support strategies that include assessment, goal-setting, action planning, problem-solving and follow-up, and organizing internal and community resources to provide ongoing self-management support to patients. 3. *Community linkages*: Mobilizing community resources to meet the needs of patients. Includes activities like encouraging patients to participate in community program, forming partnerships with community organizations to support and develop interventions that fill gaps in needed services, and advocating for policies to improve patient care \[[@B2]\]. 4. *Decision support*: Promoting clinical care that is consistent with scientific evidence and patient preferences. Can include embedding evidence-based guidelines into daily clinical practice, sharing evidence-based guidelines and information with patients to encourage their participation, using proven provider education methods, and integrating specialist expertise and primary care. 5. *Delivery system design*: Organizational practice in the provision of care including aspects of team functioning, team leadership, clinic processes (appointments, follow-up, planned visits), and continuity of care. 6. *Clinical information systems*: Organizing patient and population data to facilitate efficient and effective care, including sharing information with patients and providers in support of care coordination \[[@B2]\]. The organization research team ------------------------------ The organization research team adopted the ***Organizational Context and Capabilities for Integrating Care (CCIC) Framework \[[@B16]\]*** to guide data collection and analysis. The CCIC Framework is a theoretically grounded model that captures key organizational factors that influence successful implementation of integrated models of care. The Framework includes 17 factors (or concepts) grouped into three categories: 1. *Basic structures of the organization*: Including physical features, resources, accountability mechanisms, IT, and organizational/network design. 2. *People and values associated with the organization*: Including leadership approach, clinician engagement and leadership, organizational/network culture, focus on patient-centredness and engagement, commitment to learning, work environment, and readiness for change. 3. *Key organizational processes*: Including partnerships, delivering care, measurement of performance, and quality improvement practices. The Framework includes attention to how the external environment (such as characteristics of the intervention and target patient population) may influence each factor, as well as proximal and distal outcomes likely to be related to adopting models of integrated care. The policy research team ------------------------ The policy research team used the ***Health Policy Monitor (HPM)*** \[[@B17]\] survey methodology to explore the role of health and social policy in the implementation of ICBPHC. The HPM is designed to enable comparative policy analysis, capturing information regarding current and historical health policies and reform efforts, policy ideas, as well as change management activities and processes. The methodology focuses on policy actors who have been integral to policy reform and change efforts. The questionnaire is intended to be used to focus on particular policy developments, in this case, the policy developments around ICBPHC. The HPM questionnaire is organized on a two-dimensional matrix capturing 15 issue categories (matrix 1) as they develop over 7 process stages (matrix 2). It aims to capture and categorize the issue being assessed as well as its implementation. Rather than adopting the questionnaire directly, it is used to guide development of semi-structured interview guides which are organized to capture data at three system levels: ***case site*** (may be within a single organization or set of organizations that constitute a loosely aligned network), ***service matrix (network)*** (organizations or networks like regional health boards that influence how services are delivered, funded and governed), and ***policy subsystem*** (referring to higher system level oversight which may include state as well as non-state actors such as Ministries of Health and accreditation bodies). Each of the three levels of analysis includes key domains or questions that are explored. At the case site level, four areas of inquiry are explored through directed questions including: 1) *Why the model of care is delivered*; 2) *what services are being provided*; 3) *who is being targeted by the model of care*; and, 4) *how services are delivered*. At the service matrix (or network) key aspects of *delivery arrangements, financial arrangements* and *governance/accountability* structures are captured. Finally, *financial arrangements* and *governance/accountability* structures that occur at the policy subsystem level are examined in this approach. Identifying the linkages: A game of 3-D chess --------------------------------------------- The overlap between disciplinary and theoretical approaches can be understood much like a game of 3-D chess. In this particular game, we have four boards at different levels, each consisting of multiple squares (or concepts). While some squares remain stable at their level, others squares allow our game pieces to jump between levels, creating a series of conceptual trap-doors that connect two or more boards together. By developing a clear understanding of the boards *and* trap-doors, we can comprehensively capture all factors relevant to ICBPHC from a multidisciplinary, whole system perspective. We adopted a concept mapping approach to both locate and understand the nature of these trap-doors. Methods ======= Concept mapping was initially developed by Joseph D. Novak \[[@B24]\] to graphically represent the relationships between concepts using labels, nodes, boxes and links \[[@B25][@B26][@B27]\]. Concept maps were originally designed to support learning for school-aged children but have increasingly been used as a tool in research, supporting data collection and analysis \[[@B25][@B26][@B27][@B28]\]. This newer use of concept mapping is well aligned with the aims of our study to guide data collection and analysis while ensuring all critical concepts are captured. Trochim \[[@B28]\] suggests the concept mapping process can be viewed as an "artistic procedure which yields interpretable, suggestive conceptual pictures and a scientific one based upon sound evidence regarding its validity, reliability, and theory-enhancing value" (p. 87), and further suggests that concept maps provide "suggestive devices, useful for their stimulative or creative value" (p. 88). Here we adopt the "hard art" approach to concept mapping suggested by Trochim, whereby we use rigorous qualitative methods to develop a map to inform both data collection and analysis for the iCOACH study. To this end we adopt two key methods. First is the generation of a concept map to align the four disciplinary approaches and theoretical models used to guide the iCOACH study. Second is a preliminary exploration of one key overlap point using data from one Ontario case site. We focus on a single overlap point, care coordination, to act as an initial "test" of the map and to demonstrate its utility. Further testing of the full map will be done by the research team as we move forward with our data analysis. Concept mapping of perspectives on integrated community based primary health care --------------------------------------------------------------------------------- Novak and Cañas suggest a six-step approach to concept mapping: 1) clearly defining the focus/question; 2) identification of relevant concepts; 3) construction of a preliminary map; 4) map revision; 5) identifying cross-links; and 6) map revision after links identified \[[@B26]\]. In this section we describe the activities of our research team aligned with each of the 6 steps. The iterative analysis approach conducted by multiple members of the research team supports credibility and dependability of the initial map \[[@B29]\] through crystallization of multiple viewpoints (see \[[@B30]\] on crystallization as a method to enhance credibility in qualitative research). Step 1: Defining the focus/question The focus for the iCOACH project is to explore implementation of ICBPHC systems, and the question we are aiming to answer with the concept map is *what macro, meso and micro level factors are relevant to implementing these models?* The aim then for this concept map is to produce a descriptive guide for our data collection and analysis approach. Step 2: Identification of relevant concepts Each of the four research teams identified relevant theories and tools that point to key concepts to be understood at each level. The previous section describes the theoretical frameworks and approaches identified by each team. Step 3: Constructing the preliminary map The following process was followed to construct the preliminary map. Prior to constructing the map, the process described below was introduced in a team meeting and agreed upon by the research team: 1. The lead author initially looked over the concepts within each of the theoretical frameworks and identified conceptual similarity across models. This analysis involved both looking at the description of concepts from the original theoretical models, along with an examination of the operationalization of those concepts in the form of data collection tools (e.g. survey and interview questions, document collection) as well as the analysis strategies set out by the research teams in the iCOACH protocol. 2. Four concept tables were created (one for each team) which linked to theoretical framework and data collection methods. 3. Concept tables were sent to iCOACH team leads (and other team members) for feedback to ensure tables adequately reflected theoretical frameworks. Concept tables were reviewed by 8 team members. Meetings were held with each team lead to discuss the tables and approach. The types of feedback provided varied by team. In some cases, research team leads (and/or other members) reviewed the table to check that concepts and items were adequately reflected. Other team leads provided broader feedback with regard to their overall approach, conceptual framework, and anticipated methodology. 4. Based on this feedback, the concept tables, and protocols, conceptual overlaps between team protocols were identified. Concepts were grouped when they: 1) directly overlapped in terms of what was being captured; and/or 2) were fully or partially conceptually related (e.g. Patient characteristics are related to "who" the ICBPHC project is targeting). Table [1](#T1){ref-type="table"} overviews concept groupings. ###### Concept overlap across four levels: trap doors. ---------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Concept groups Analytic level Concept Definition System structure and design Policy Policy subsystem -- funding and governance Applicable legislation, regulation, guidelines. Eligibility criteria for patients to enter the program. Patient-carer Access to services (universal coverage) Patients have access to all basic necessary health care services when needed which are free at the point of care. Related to system level concept of university coverage captured in the Ham model. System governance Policy Funding and governance arrangement -- all levels How services are delivered in terms of governance structures and payment mechanisms. Organization Organizational infrastructure and design: Resources and Governance Resources: Amount and sources of funding Governance: Board composition, activities and oversight. Organization structure and leadership Organization Leadership, strategy and clinician engagement Leadership style of the organization and level to which clinicians and frontline providers are engaged in decision-making (e.g. centralized or distributed/shared, responsive approaches; receptivity to new ideas). Provider Organization of delivery system -- leadership Elements of the organization of the delivery system specifically focused on leadership and support for frontline providers. IT systems design and utilization Organization Organizational infrastructure and design: IT systems Information technology systems available in and across organizations (e.g. decision support tools, electronic information systems, electronic health records, interoperability). Provider Provider use of IT system What information technology systems (in particular clinical information systems) are used by providers in clinical practice. System level values and perspectives Policy Why deliver this model of care -- case site level The organizational impetus for delivering this particular model of care and the nature of the care provided. Organization Social-Psychological context: Organizational culture The overarching focus and goals of the organization, as well as the character of the organization, managerial style and employee support and overall cohesion. Includes organizational climate concepts such as openness and flexibility in the organization, change readiness, and employee burnout/stress. Care Coordination Policy Delivery arrangements -- case site and matrix level Who delivers care in terms of direct care delivery (in-house staff), contracting out, partnerships, and/or referrals. How services are delivered -- case site and matrix level The extent to which models use integrating/coordinating mechanisms (e.g. inter-disciplinary teams, integrated care plans, information technology used to integrate). Organization Teamwork and collaboration Use of multiple modes of communication within and between teams, feelings of teamwork, cooperation, information sharing and goal alignment, staff input in decision-making. Inter-organizational linkages Formal and informal connections made between different organizations in a network. Including connections between different teams. Focus on process of working across organizational boundaries. Provider Delivery system design Team level functioning and leadership and continuity of care. Community linkages Providers ability to link patients to outside resources and partnerships with community organizations. Patient-carer Care coordination Patient and carer perceptions of level to which their care from different providers is aligned. Person-centred design Policy Who is targeted -- case site level Which individuals and communities are being served by the project defined in terms of their acute and chronic illness profile, their geography, culture or needs. Provider Delivery system design -- population management Adopting a systematic approach to population management, and seeking ways to meet community needs. Patient-carer Patient and carer demographics Information regarding patient and carer personal information (i.e. gender/sex, age), health and social status, culture and ethnicity, beliefs and values. Program design and monitoring Policy What services are delivered -- case site level What types of services are provided by the project. Governance structure -- matrix and subsystem levels How agencies fund and oversee projects (governance and accountability arrangements). Organization Clinical processes Use of clinical practice guidelines (adoption, availability and organizational support for adoption. Continuous quality improvement (CQI) Adoption of quality improvement processes (strategies, committees, patient experience and outcome collection measures). Provider Types of services Services delivered to patients and quality improvement processes. Self-management support Services specifically around supporting self-management, assessment and documentation. Decision-support Use and availability of decision-support tools, enabling use of evidence based practice. Patient-carer Types of services used Services used by both patients and carers. Self-management and prevention. Reported self-management support received, as well as support for patients to engage in their care and do more self-care activities, such as engaging in goal-setting. ---------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Step 4: Revising the map An initial map was constructed which grouped concepts together in boxes rather than linking using lines. This initial map was presented to the full iCOACH team at a team meeting where it was discussed and modified based on team feedback (approximately 20 members of the research team were in attendance). CMapTools software was then used to visually depict the revised concept map. Step 5: Seeking cross-links A sub-group of the broader iCOACH team (the co-authors on this paper) was brought together to identify cross-links, do a final revision, and publish findings. Co-authors reviewed and revised the map, and agreed upon preliminary cross-links for the map. We agreed as a team that the links named in the map would need to be tested and validated using data from the iCOACH study. Step 6: Revising the map We used CMapTools to add in cross-links agreed upon in step 5. This version was reviewed again by the co-authors of this paper and consensus was reached with regard to the visual map and content. Results ======= Concept map ----------- Figure [1](#F1){ref-type="fig"} visually presents the concept map. Each analytic level is represented down the center of the map and is given a unique colour: policy level is green, organization level (written as healthcare organizations and networks) is orange, provider level is purple, and patient/carer level is blue. The central boxes of each level are linked to their related concepts in the same coloured boxes via similarly coloured lines. {#F1} Conceptual overlap across levels is depicted through black lines linking concepts from different levels. As noted in the methods section these concepts are similarly defined and operationalized (e.g. similar survey and/or interview questions are used to capture the concept), however the nature of their relationship in practice is yet to be determined. To help us make sense of the map we include labels where concepts intersect that relate to what we see as 8 overarching concepts: 1. System structure and design 2. System governance 3. Organization structure and leadership 4. IT systems design and utilization 5. System level values and perspectives 6. Care coordination 7. Person-centred design 8. Program design and monitoring We acknowledge that this is a preliminary attempt to categorize concept groups which will need to be explored through the process of data analysis. A full exploration of these overlaps is beyond the scope of this paper and will be conducted through analysis of iCOACH data over the course of future papers. Exploring interconnections: the example of care coordination ------------------------------------------------------------ To demonstrate the utility of the concept map, data from the iCOACH study was used to explore the ***care coordination*** theme and the related concepts across the four perspectives. Care coordination is a critical component in the integration of services \[[@B31]\] and as such is a useful place to start our analysis. One case from Ontario was selected to extract findings. Transcribed interviews from policy makers, organizational managers, front-line providers and patients and carers in each jurisdiction had been coded as part of the first stage of analysis along common coding structures specific to each level of analysis. This initial coding process was conducted by different members of the iCOACH research team from Ontario, Quebec and New Zealand through a multi-phased iterative approach in which coding was validated through multiple rounds of double coding, discussion, and consensus building as is consistent with qualitative analysis methods used to guide analysis of the iCOACH study \[[@B32][@B33]\]. This approach created a set of rigorously coded data, allowing for the team to easily extract data on similar concepts across cases and jurisdictions. Coded data that related to overlapping concepts related to care coordination were extracted from the one Ontario case for the analysis presented in this paper. Care coordination is a widely explored concept with multiple theoretical understandings and activities described in the literature. The Agency for Healthcare Research and Quality in the United States has defined care coordination as: > "... the deliberate organization of patient care activities between two or more participants (including the patient) involved in a patient's care to facilitate the appropriate delivery of health care services. Organizing care involves the marshaling of personnel and other resources needed to carry out all required patient care activities and is often managed by the exchange of information among participants responsible for different aspects of care." \[[@B34]\] (p. 6) In the concept map, we see an important grouping of concepts around care coordination. While care coordination is explicitly included in the patient/carer theoretical enquiries, it is not explicitly written into the frameworks used by the provider, organizational and policy research teams. The concept map offers insight into where care coordination fits into other theoretical perspectives, and where we might see differing conceptualizations and approaches to this concept. Similarities in concepts arising from different theoretical approaches and levels of analysis is a key advantage of using a concept map approach, allowing for a multidisciplinary analytic approach which recognizes multiple definitions and conceptualizations of similar concepts. From the policy perspective, *how services are delivered* and *delivery arrangements* are concepts used by the research team to capture the extent to which studied models use integrating or coordinating mechanisms. While this is primarily a descriptive concept, the policy perspective approach suggests we explore historical trends, and related policy subsystems relative to how services are delivered. Current and historical policies that shape the system and health care institutions that can create barriers or enablers to *how services are delivered* are included in this concept from a policy perspective. From the organizational perspective the map shows how *inter-organizational linkages, collaboration* and *teamwork* are linked to the concept of care coordination. The connection here is a bit more obvious as inter-organizational linkages and partnerships create processes such as sharing administrative and clinical information, referral systems, and joint accountabilities that are important to cross-organizational coordination of care \[[@B16]\]. Similarly, at the provider level, *community linkages* between individual providers can enable coordination of patient care across health and social care boundaries \[[@B35]\]. *Delivery system design* specifically around team functioning, leadership and continuity are additionally key components of care coordination at the provider level. Care Coordination in Ontario ---------------------------- Elsewhere in this special issue Breton et al. \[[@B19]\] offer detailed descriptions of case examples from Ontario and Quebec in Canada, and New Zealand. The Ontario case of focus for the analysis presented in this paper is the Integrated Client Care Program, a collaborative model of care based on a partnership between a Family Health Team and Community Care Access Centre located in Toronto with strong partnerships with a local hospital and other community agencies. This program focuses on high needs older adults with the aim to coordinate care delivery across the multiple providers in primary, community and acute care settings. ### The policy perspective on care coordination: How services are delivered and delivery arrangements Organizational managers from the Integrated Client Care Program identify significant system fragmentation which results in difficulties when different sectors (e.g. primary, acute and community care) seek to coordinate patient care. One hospital manager clearly defines the issue of fragmentation with regard to patients with complex care needs transitioning from hospital to home: > "When the reality is \[the hospital\], maybe, see that patient for like 1% of their health care journey and the majority of the care that they receive is actually in primary care. And yet, primary care often is the one sector that doesn't get engaged in the delivery, or in, actually, in defining what the care's going to look like for that patient post discharge from the hospital. And because primary care is not engaged in that ward, that patient often \[...\] fall through the gaps all the time." (SE-05 -- Hospital manager) This sentiment of fragmentation is echoed by a policy-level informant from a CBPHC partner who identifies that while excellent care delivery occurs, it does so in silos: > "We have pockets of excellent primary care and pockets of excellent homecare. They almost never integrate \[...\] So I think that is the vision is to integrate them." (M5) In the Ontario example, new legislation, *The Excellent Care for All Act*, 2010 \[[@B36]\], establishes quality reporting and delivery process requirements for health care providers to enable patient-centred care delivery, which is funded and governed through regional bodies (at the service matrix level). A manager/clinician from the Family Health Team in this case describes how legislation provides an opportunity to shift how services are delivered through development of a new program that could work to improve integration: > "... the good luck of the Excellent Care for All Act really coming right at the time when we're thinking about improving care for patients and making patients the centre of care. So that was exactly what the Virtual Ward was, um, intending to do. And so a major healthcare policy shift allowed us to really have the driver, um -- it's, you know, at hand to also bring CCAC onboard, because it aligned, um, with, um, major healthcare policies." (SE-13 -- Family Health Team manager/clinician) ### The organizational perspective of care coordination: Organizational linkages and teamwork In seeking to adopt new programs to address fragmentation and improve coordination, managers seek to develop new inter-organizational linkages and teams. Having a shared vision of integration between organizations as well as clear strategies related to collaboration and inter-organizational linkages are identified as key enablers to building connections. A manager from the Community Care Access Centres (CCACs) in the Ontario case reflects: > "We have one strategy across the province for all CCACs around building meaningful relationships with primary care and building integrated teams in local geographies across the province. One such example is with the \[family health team\]." (SE-14 -- CCAC manager) Beyond just the strategy and vision of collaboration, organizational managers from the case identify the importance of having a culture of teamwork to develop organizational linkages and support integrated care delivery: > "Um, the teamwork and the organizational culture are interesting because, um, for example in my team, it was really the team itself. I was somehow very lucky to have just wonderful, wonderful team players. Um, and so that set the culture, because you know, it wasn't that it was just simply my family health team's culture or CCAC's culture, or \[the hospital's\] organizational culture. But it was the team that established it as a culture that we are all committed to that same outcome and vision." (SE-13 --Family Health Team manager/clinician) ### The provider perspective of care coordination: Delivery system design and establishing community linkages The culture of teamwork and organizational strategy aimed at developing relationships creates an environment to support the development of community linkages and inter-professional teams. One physician reflects on how coming together on patient care serves to solidify relationships with community partners: > "I think it was just, you know, through providers that have provided care for mutual patients that that's sort of how the relationships have formed. They've matured into sort of more collaborative groups. I think they started informally and now have sort of formalized." (SE-01 -- Family Health Team clinician) Built relationships and a shared culture of collaboration across organizations creates a supportive environment that allows inter-professional teams to communicate and work together to establish processes to coordinate care. Information sharing processes and integrated systems, as well as activities like case conferences allow Family Health Team providers to work closely with other home and community care providers. Another Family Health Team clinician in an allied health discipline describes how case conferences are a particularly helpful process: > "So, it's kind of like a lot of back and forth. Which can just be time consuming and a little bit frustrating, but rather have everybody at the table, hear the patient's concerns, come up with a plan together -- that way everybody's sort of the same page rather than the back and forth. I don't know, I think that's been helpful." (SE-02 -- Family Health Team clinician) ### The patient and carer perspective of care coordination: Perceiving coordinated care Patients and carers want to be confident they are referred for expert care and advice when needed, that the information they have provided to one party is passed on as appropriate to others responsible for components of their care, that they are listened to, respected, and provided with the information and resources they need to manage their conditions. The inter-professional work facilitated through the Integrated Client Care Program seems to provide an integrated experience for patients and carers in that the whole team is aware of the patient needs. As one carer reflects: > "So there's always (sighs), they're always there, and they all seem to know what's going on with Jim, and they come when they're needed." (SCa-02 -- carer) ### A whole system perspective of care coordination in the Integrated Client Care Program in Ontario Using the concept map to approach analysis we can see how different disciplinary lenses assign relevance to different aspects of a similar concept. In the above example from Ontario, patients and carers identify that a key aspect of care coordination for them is that they have a team of health care providers who are available to meet their needs when required. For providers, care coordination is about building relationships in teams that exist within and across different organizations. The data presented above would suggest that managers and policy makers see care coordination as being more about creating linkages and addressing systems gaps that could *enable* care coordination at a clinical level. These different meanings of care coordination for different actors suggest there are multiple ways to enable and measure care coordination. In terms of measurement, if we were to simply count the number of organizational linkages that occurred since adoption of new policy, we may miss out on whether these linkages have come along with development of strong relationships and trust needed to ensure functioning of teams from a provider perspective. We may also miss whether these new linkages are in fact helping meet needs of patients and carers, or whether these add to the confusion of where services can be accessed. Similarly, different meanings of care coordination suggest there are likely different drivers as well. Where strong relationships and teamwork are important to providers, aligned vision and shared culture is more important to create linkages between organizations. These differing perspectives are not issues of semantics. Rather, they identify a key point in the implementation process of integrated models of care. The critical issue is about perspective. Starting at the perspective of any one of these four levels will lead to a different design response to the issue of 'care coordination'. This highlights the importance of developing a "whole system" framework that can work across the four levels of patient/carer, provider, organization, and policy. Diving more deeply into these trap-doors is beyond the scope of this paper. This preliminary analysis does, however, demonstrate how a multidisciplinary approach requires attention to new, potentially unexpected areas of inquiry that can better inform implementation of ICPBHC. Discussion ========== Using a concept map to guide multidisciplinary and cross-level analysis demonstrates the value that comes from linking across levels to better understand integrated models of care, and how different perspectives understand and assign value to similar ideas. While concepts from different theoretical perspectives may be similar in definition, the cross-level analysis shows how looking at these concepts through different perspectives reveals different realities which can be brought into focus. The analysis guided by the concept map points to three important dimensions of any concept which requires our attention. First, *what is the meaning of the concept from different perspectives*? In the Ontario case example, it is found that what care coordination means to the patient is very different then what it means to health care organization managers and policy makers. Second, *how do different actors value and measure the concept?* How does a policy maker know care coordination has been achieved? When does a provider feel like the care they deliver is coordinated across their inter-professional team? And finally, *what are the perceived drivers of the concept across levels?* This last point regarding drivers of different concepts is particularly important in our understanding of implementation of ICBPHC. In the Ontario example we see how what is important to providers is having strong relationships with inter-disciplinary team members, supported by agreed roles, responsibilities, boundaries, funding and reporting, in order to coordinate care delivery. For policy-makers, however, the interest is in achieving high-quality care delivery as captured through performance indicators. The risk is in the implementation of these activities. Relationship development in inter-professional teams "requires time, interaction, and focused attention" \[[@B37]\], which can potentially take time away from direct patient care. Performance indicators in primary care often focus on access time for patients \[[@B38]\], leaving little room for providers to work on team and relationship building outside of patient visits. Other conflicts may arise for organizations who seek to develop new partnerships within a system that includes competitive procurement. Previous studies have shown that competitive procurement processes (intended to meet policy objectives of quality and efficiency) can act as a significant barrier to partnership \[[@B39]\]. Limitations ----------- What is presented in this paper is a methodological approach to multidisciplinary whole systems studies using concept mapping. The emphasis in the presented work is on the development of the map with only an initial look into analysis. An important limitation here is that the analysis of the Ontario case with regard to care coordination is a preliminary exploration. Future steps in this work will require a more in-depth study of conceptual overlap of care coordination, as well as the other 7 conceptual areas of the map using cross-disciplinary theories to tease out some of the questions that have been posed by this work. Additional testing of the map using empirical data will enhance credibility and dependability of the map. The map should further be validated through empirical testing. Future analysis, much of which is currently underway, will require the work of multidisciplinary research teams working across different perspectives to adequately tease apart meanings, measures and drivers of shared concepts. Conclusion ========== The concept map and analysis presented in this paper marks an important first and foundational step towards a more in-depth cross-disciplinary analysis of the integrated models of CBPHC to be conducted as part of the iCOACH study. The map in its current form requires a clearer understanding of exactly how concepts are related in practice. We present an initial exploration of these relationships, identifying three dimensions of analysis that could be adopted in future explorations of the map using data from the iCOACH study. The methods and preliminary analysis presented in this paper can be used as a guide by other multidisciplinary research teams exploring implementation of integrated models of CBPHC and/or who are using multiple theoretical models to engage in "whole systems" health services research. The collaborative nature of iCOACH across New Zealand and Canada illustrates that the approach is also a viable strategy for alignment and comparison of international models of service delivery. The concept map should be viewed as a living document, one that grows and develops as analysis of data continues. Importantly, identifying which concepts may be more important to the implementation of ICBPHC models should be explored with particular attention to the drivers of similar concepts for different stakeholders across the system. We would like to acknowledge all participants in the iCOACH study. We would additionally like to thank the large iCOACH team of investigators, collaborators and trainees, many of whom provided feedback and thoughts on the concept mapping process, and whose efforts were instrumental in supporting data collection and analysis for the broader project. This study is supported by grants from the Canadian Institutes of Health Research (CIHR) (Funding Reference Number TTF-128263) and the New Zealand Health Research Council. The views expressed in this paper are the views of the authors and do not necessarily reflect those of the funders. Reviewers ========= Roelof Ettema PhD, Principal Lecturer Institute for Nursing Studies, Postdoc Researcher Chair Chronic Illnesses and Chair Methodology of Applied Sciences, Staff Lecturer Master Integrated Care Design (MICD)\|**HU** University of Applied Sciences, Utrecht, The Netherlands. Jill Manthorpe, Professor of Social Work, Director of the Social Care Workforce Research Unit, King's College London, UK. Competing Interests =================== The authors have no competing interests to declare.

INTRODUCTION ============ Topoisomerase II (topo II) and heat shock protein 90 (Hsp90) both contain a conserved ATPase domain and belong to the same family, namely, GHKL (gyrase, Hsp90, histidine kinase, and MutL) domain ([@b10-bt-24-453]; [@b6-bt-24-453]). ATPase domain in both of these proteins requires ATP to exert important cellular functions such as cell cycle progression, proliferation and survival. Therefore, inhibitors targeting the ATP binding site of these two proteins through binding in an ATP-competitive manner were searched and characterized in this study. Another important biological implication in topo II and Hsp90 is that they are both overexpressed in proliferating cancer cells and have been attractive targets for the development of anti-cancer drugs ([@b33-bt-24-453]; [@b34-bt-24-453]). Topo II is very important in cellular processes such as transcription and replication by introducing transient breaks in DNA double strand ([@b35-bt-24-453]). Topo II requires ATP binding for its conformational change to solve topological problems in DNA. Recently, there are much efforts in developing catalytic inhibitors of topo II in order to overcome the side-effects of topo II poisons such as etoposide ([@b39-bt-24-453]). Hsp90 is a molecular chaperone which has diverse client proteins involved in tumor growth and survival. Therefore, Hsp90 also has been an attractive target for chemotherapeutic development and phase II clinical trial was conducted for Hsp90 inhibitor, 17-allyaminogeldanamycin (17-AAG) ([@b45-bt-24-453]). In 2006, Jenkins and coworkers reported that the topo II and Hsp90 form a complex, and co-treatment of 17-AAG showed synergistic efficacy by enhancing the activity of topo II poison ([@b2-bt-24-453]; [@b54-bt-24-453]). From these findings, development of inhibitors that target both ATPase domains of topo II and Hsp90 can be a promising research area. There are many advantages of multi-target drugs since they can simultaneously inhibit multiple pathways and escape an undesirable drug-drug interaction which may encounter with co-treatment of single-target drugs ([@b38-bt-24-453]). In this review, the topo II and Hsp90 inhibitors that bind to the ATPase domain of each of topo II and Hsp90 are analyzed and the possibility of designing dual inhibitor is explored through molecular modelling studies. METHODS ======= Molecular docking studies ------------------------- The 3D structures of the inhibitors of topo II and Hsp90 were sketched using Sybyl X-2.1.1 (Certara L.P., St. Louis, MO, USA). All the structures were energetically minimized using Tripos force field and Gasteiger-Hückel charges. The structures of ATPase domain of topo II and Hsp90 were retrieved from RCSB Protein Data Bank (PDB entry code: 1ZXM and 3EKR) ([@b50-bt-24-453]; [@b25-bt-24-453]). The ligands were extracted and water molecules were removed from the initial x-ray crystal structure. The docking was carried out for both of the topo II and Hsp90 inhibitors to topo II and Hsp90 using Surflex-Dock ([@b17-bt-24-453]). The protomol was generated using ligand mode, which used the ligand, extracted from the crystal structure occupying the ATP binding site, to ensure that the inhibitor could bind to the ATP binding site. Then polar hydrogens were added to the structure. After the protomol generation, ligands were docked using Surflex-Dock Geom and GeomX modes using default parameters. Pharmacophore hypothesis generation ----------------------------------- Compounds PU3, 3t, AUY922 and comp. 14 were used as input molecules to generate pharmacophore model using GASP module implemented in Sybyl X-2.1.1. Four molecules selected were used as the data set for the pharmacophore model generations. All the features on each of the molecules were used and the default GA parameters were used. The parameters used for the calculations were as follows; 100 population size, 1.1 selection pressure, 100000 max operations, 6500 operation increment and 0.01 fitness. RESULTS ======= Comparison of the structural similarities of ATPase domain of topo II and Hsp90 ------------------------------------------------------------------------------- Among structures of ATPase domain of topo II and Hsp90 deposited in Protein Data Bank (PDB), 1ZXM for topo II and 3EKR and 1BYQ for Hsp90 were chosen for structure comparison and docking. The length of the two proteins is 376 and 217 amino acid residues for topo II and Hsp90, respectively. The similarity of the two proteins were compared by sequence alignment using BioEdit ([Fig. 1](#f1-bt-24-453){ref-type="fig"}) ([@b13-bt-24-453]). Although the sequence identity between the two ATPase domains is 15.8 %, which is rather small value, the overall fold has high similarity where they are superimposable ([Fig. 2](#f2-bt-24-453){ref-type="fig"}). The ATP binding sites of the two proteins can be suggested to have similar environment. The amino acid residues involved in binding with the ligand adenylyl-imidophosphate (AMPPNP) or ADP for topo II and Hsp90, respectively, do not coincide exactly, however the properties of each amino acids are conserved. For example, in topo II, Asn120 forms hydrogen bond with the N6 amino group of adenine ring, whereas in Hsp90, Asp93 is involved in the hydrogen bond interaction. The hydrophobic residue Ile125 in topo II corresponds to the residue of Met98 in Hsp90 which is also hydrophobic. Additionally, the size of the ATP binding site of each protein was calculated with Computed Atlas of Surface Topography of proteins (CASTp, <http://sts.bioe.uic.edu/castp/>) ([@b28-bt-24-453]). As listed in [Table 1](#t1-bt-24-453){ref-type="table"} and shown in [Fig. 3](#f3-bt-24-453){ref-type="fig"}, the calculated area and volume of topo II ATP binding site were 792.2 Å^2^ and 1077.6 Å^3^, respectively. The ATP binding site's area and volume of Hsp90 were slightly smaller than topo II, 628.9 Å^2^ and 971.0 Å^3^, respectively. The mouth opening of the binding pocket was also identified and characterized with CASTp. Although the overall pocket size was slightly larger for topo II, the area and the circumcircle of the mouth opening of Hsp90 were larger than topo II, with the values of 167.6 Å^2^, 78.6 Å and 70.9 Å^2^, 52.6 Å, respectively. Topo II inhibitors that bind to the ATPase domain ------------------------------------------------- The topo II inhibitors that bind to the topo II ATPase domain were searched. The inhibitors can be largely divided into two categories, purine analogues and non-purine analogues. [Table 2](#t2-bt-24-453){ref-type="table"} lists the topo II inhibitors and gives information about their structures and ATPase inhibition activity where applicable. Purine analogue inhibitors contain the purine ring and have substitutions on the 2, 6, or 9 positions. In order to develop novel topo II catalytic inhibitors, 1,990 compounds from the National Cancer Institute (NCI) diversity set library was screened and *S^6^*-substituted thioguanine analog, NSC35866 was identified ([@b19-bt-24-453]). This finding was further expanded to discover more potent ATPase inhibitors by screening 40 substituted purine or purine-like compounds in the NCI database and several compounds including NSC348400 were identified from this screening ([@b18-bt-24-453]). Compounds 1 and 2 were searched from the Novartis compound collection to specifically target the ATP binding site of topo II ([@b11-bt-24-453]). The hydrogen bond forming residues of topo II were Asn120 and Asn91 which were identical in three topo II complexes with compounds 1, 2 and AMPPNP, however the purine ring of compounds 1 and 2 adopted different orientation compared to that of ATP. Compounds 1 and 2 were further optimized by considering these interactions with the binding site and obtained a purine analogue with substitution of an ethyl group at position C6 and a morpholino-ethoxy group in the quinolone substituted on position N^2^ (called quinoline amino-purine, QAP1) ([@b7-bt-24-453]). QAP1 showed improvement in topo II ATPase inhibitory activity with the half maximal inhibitory concentration (IC~50~) of 128 ± 21 nM. 3t has a new scaffold, aloisine moiety, which is similar to purine ring ([@b26-bt-24-453]). In contrast to compounds 1 and 2, the aloisine ring was aligned with the purine ring of ATP from docking study. 2c is a organoplatinum(II) complex with an attachment of 2-amino-6-chloropurine ([@b49-bt-24-453]). 2c inhibited topo II by preventing ATP entering into the ATPase domain. Although 2c is a purine analog, its purine moiety did not occupy the ATP purine ring binding site, but the *tert*-butyl groups of the terpyridine scaffold occupied on it, determined by molecular docking study. 8-chloro-adenosine (8-Cl-Ado) is an anti-cancer agent currently undergoing phase I/II clinical trial. 8-Cl-Ado converted into 8-Cl-ATP in cells and it competed with ATP to inhibit topo II ([@b52-bt-24-453]). There are some topo II inhibitors originated from natural products. Daurinol is a lignan isolated from *Haplophyllum dauricum*, whose structure is similar to etoposide ([@b23-bt-24-453]). Etoposide is a well-known cytotoxic anti-cancer drug functioning as a topo II poison. Daurinol occupied the same binding site with AMPPNP, which was shown by molecular docking study, suggesting it inhibited topo II by targeting the ATPase domain. Gambogic acid (GA) is a natural product isolated from *Garcinia hanburi* tree. GA was shown to be a catalytic inhibitor of topo II by binding to the ATPase domain, determined by surface plasmon resonance (SPR) analysis and by molecular docking ([@b41-bt-24-453]). Diphyllin was extracted from *Justicia procumbens*, which showed tumoricidal effects. D11 is a novel acetylated D-quinovose diphyllin analogue exhibiting potent topo II inhibitory activity and binding to the ATPase domain ([@b12-bt-24-453]). When the compound binds to the ATP binding site it is not always a topo II catalytic inhibitor, as it is the case with salvicine and emodin. Salvicine is a derivative of diterpenequinone isolated from *Salvia prionitis*, which bound to the ATPase domain validated by SPR and molecular docking, and acted as a topo II poison generating double strand breaks ([@b14-bt-24-453]). Emodin is an anthraquinone isolated from *Rheum emodi* and also from molds, lichens and fungi. Emodin also generated DNA double strand breaks and stabilized the topo II-DNA cleavage complex ([@b27-bt-24-453]). Last group of topo II inhibitors are synthesized compounds designed from different scaffolds known to be biologically active or potent anti-cancer agents. Thiosemicarbazone (TSC) is one of the scaffolds, and their derivatives including TSC24 showed catalytic inhibition of topo II ([@b15-bt-24-453]). TSC24 directly bound to the ATPase domain which was confirmed by competitive inhibition assay, SPR and molecular docking studies. Baviskar and coworkers designed and synthesized bicyclic N-fused aminoimidazole which had similar structure to reported topo II inhibitors and marketed drugs such as zolpidem and zolimidine ([@b4-bt-24-453]). From the synthesized compounds, comp. 5 is a non-intercalating topo II catalytic inhibitor that bound to the ATP binding site. Compounds 14 and 14mod are xanthone derivatives that bound to the topo II ATPase domain, which was confirmed by ATPase competitive inhibition assay and molecular docking ([@b22-bt-24-453]; [@b36-bt-24-453]). There were several topo II catalytic inhibitors containing quinone moiety that bound to the ATP binding site. Pyranonaphthoquinone, comp. 3a, was shown to be a topo II catalytic inhibitor and suggested to bind to the ATPase domain through docking ([@b21-bt-24-453]). Naphthoquinone fused cyclic aminoalkyl-phosphonates and aminoalkyl-phosphonic monoester were synthesized and tested for their topo II activity ([@b48-bt-24-453]). Some of them including comp. 18 were catalytic topo II inhibitors and these compounds were docked into the ATP binding site of topo II ([@b29-bt-24-453]). Hsp90 inhibitors that bind to the N-terminal ATPase domain ---------------------------------------------------------- Geldanamycin (GDA) and radicicol (RDC), antibiotic isolated from natural product ([@b42-bt-24-453]) were the first discovered Hsp90 inhibitors that target the N-terminal ATPase domain. Due to poor solubility and hepatotoxicity of GDA and RDC, GDA and RDC derivatives were designed and synthesized to have good physical properties and stability with improved potency. 17-AAG is a GDA derivative that improved the toxicity and stability of GDA itself ([@b44-bt-24-453]). The co-crystal structure of 17-DMAG and Hsp90 N-terminal ATPase domain was solved ([@b20-bt-24-453]). GDA derivatives were also genetically engineered to produce GDA analogs, such as KOSN1559, showing better binding affinity than GDA ([@b37-bt-24-453]). Another group of Hsp90 inhibitors are RDC analogs. KF25706, KF29163 and KF58333 were chemically synthesized and their biological activities were assessed ([@b46-bt-24-453]; [@b1-bt-24-453]). Various RDC analogs were further synthesized such as aigialmycin D, c-RDC, pochonin A, pochonin D and O-(piperidinocarbonyl) methyloxime derivative of RDC. The GDA and RDC analogs are rather big in size and their poor properties in solubility and toxicity led to designing and synthesizing purine analogs for inhibiting Hsp90 by binding to the ATP binding site. PU3 is one of them and it competed with GDA for Hsp90 binding and when treated in cancer cells, HER2 level decreased ([@b8-bt-24-453]). Other small Hsp90 inhibitors include pyrazole analogs such as CCT018159 and G3130. CCT018159 was searched from high-throughput screening compound collection of more than 56,000 compounds utilizing the ATPase activity assay ([@b43-bt-24-453]). The crystal structure of G3130 bound to the N-terminal ATP binding domain of Hsp90 was solved and the value of K~d~ was 280 nM determined by SPR ([@b24-bt-24-453]). SNX0723 is one of the synthetic compound having a novel scaffold containing benzamide moiety which was discovered to bind to the ATPase domain of Hsp90 by screening a compound library ([@b40-bt-24-453]). Resorcinol moiety was also identified to be an important scaffold for ATPase binding in Hsp90. AUY922, AT-13387 and CPUY201112 are the Hsp90 inhibitors that have resorcinol moiety which plays an important role in hydrogen bonding and hydrophobic interactions with the receptor ([@b9-bt-24-453]). Hsp90 inhibitors targeting its N-terminal ATP binding site reviewed in the current study are listed in [Table 3](#t3-bt-24-453){ref-type="table"}. Molecular docking studies ------------------------- The similar molecular environment in the ATP binding sites of topo II and Hsp90 led us to assess whether reported inhibitors targeting either topo II or Hsp90 could function as a dual inhibitor. The listed topo II and Hsp90 inhibitors mentioned above were subjected for docking against both topo II and Hsp90. [Tables 4](#t4-bt-24-453){ref-type="table"} and [5](#t5-bt-24-453){ref-type="table"} list the docking results of topo II and Hsp90, respectively. Surflex-Dock gives total score, crash and polar values for each of conformers. Generally, the inhibitors targeting their own binding partner scored high total score. Interestingly, the best scoring inhibitor for topo II was PU3, which was reported as an Hsp90 inhibitor with purine ring. The inhibitors showing good docking score for Hsp90 did not perform well with topo II ATP binding site. This may be due to smaller mouth opening in ATP binding pocket of topo II than Hsp90 which was calculated from CASTp. The typical Hsp90 inhibitors are bulkier compared to topo II inhibitors, therefore it would be difficult for bulky Hsp90 inhibitors to enter into the topo II ATP binding pocket. PU3, 8-Cl-ATP and compound 3t in the docking of topo II, all showed high docking score which are all purine analogs. Fascinatingly, PU3, an Hsp90 inhibitor, scored the highest when docked to topo II. PU3 had hydrogen bonding interactions with Asn91, Asn120, Ala167 and Thr215, where they are key residues that formed hydrogen bonds with ATP ([Fig. 4A](#f4-bt-24-453){ref-type="fig"}). Also, PU3 had hydrophobic interactions with the residues comprising the ATP binding pocket, namely, Asn91, Asp94, Arg98, Asn120, Ile125, Ile141, Phe142, Ser149, Asn150, Thr159, Gly161, Arg162, Gly164, Ala167, Lys168 and Thr215. Compound 3t also occupied the ATP binding site and interacted with residues Asn91, Ala92, Asn95, Asn120, Pro126, Ile141, Phe142, Ser149, Gly164, Tyr165, Gly166, Ala167, Lys168, Thr215 and Ile217 ([Fig. 4B](#f4-bt-24-453){ref-type="fig"}). However, 3t had only one hydrogen bond interaction with residue Asn91. AUY922 is an Hsp90 inhibitor with isoxazole moiety. There are two hydroxyl substituents from the phenyl ring and amide group that can form hydrogen bonds with residues Asn95, Asn120 and Ser149. Since AUY922 is rather big molecule compared to PU3, 3t, comp. 14 and 8-Cl-ATP, larger number of residues are involved in van der Waals interaction, namely, Ile88, Asn91, Ala92, Asn95, Arg98, Ile118, Asn120, Il2125, Pro126, Ile141, Phe142, Ser149, Asn150, Gly161, Gly164, Tyr165, Gly166, Ala167, Lys168, Thr215 and Ile217. Recently from AstraZeneca, compound with benzisoxazole scaffold, ETX0914, was discovered as a novel DNA gyrase inhibitor undergoing phase II clinical trial for the treatment of uncomplicated gonorrhea ([@b3-bt-24-453]). There is no known topo II inhibitor reported with isoxazole scaffold to the best of our knowledge. Therefore our docking results suggest AUY922 may act as a topo II inhibitor with novel scaffold. Comp. 14 is a small compound with xanthone moiety which competed with ATP. Comp. 14 had hydrogen bond interactions with Asp94 and Thr215 and hydrophobic interactions with Asn51, Ser52, Ala55, Asp93, Ile96, Gly97, Met98, Asn106, Phe138, Val150, Thr184 and Val186. In the case of docking topo II and Hsp90 inhibitors to Hsp90, the bulky Hsp90 inhibitors were high in rank as mentioned above. However, AUY922, a rather smaller isoxazole derivative compared to classical Hsp90 inhibitors such as GDM or RDC, scored highest. AUY922 also had hydrogen bond interactions with five residues in the ATP binding site of Hsp90, Asn51, Lys58, Asp93, Gly97 and Phe138. The residues involved in hydrophobic interactions are Asn51, Ala55, Lys59, Asp93, Ile96, Gly97, Met98, Asp102, Leu107, Gly135, Val136, Gly137, Phe138, Val150, Thr184 and Val186. PU3, a purine analog Hsp90 inhibitor showed good result in Hsp90 along with 3t, another purine analog topo II inhibitor. Two purine derivatives of PU3 and 3t and two non-purine compounds of AUY922 and comp. 14 were selected for further comparison in depth since they ranked high in docking study of both topo II and Hsp90. The binding interactions of topo II and Hsp90 with compounds of PU3, 3t, AUY922 and 14 are shown in [Fig. 4](#f4-bt-24-453){ref-type="fig"}, [5](#f5-bt-24-453){ref-type="fig"}. The hydrogen bonding residues are labeled and the bonds are displayed as light blue dashed lines. The selected common four compounds and high scoring compounds 8-Cl-ATP and KOSN1599 for topo II and Hsp90, respectively, were analyzed in detail. [Table 6](#t6-bt-24-453){ref-type="table"} summarizes the residues involved in hydrophobic and hydrogen bond interactions. Pharmacophore model analysis ---------------------------- Compounds PU3, 3t, AUY922 and comp. 14 were further evaluated for their dual targeting features by generating pharmacophore models. The pharmacophore model was generated using Genetic Algorithm Similarity Program (GASP) module implemented in Sybyl X-2.1.1. From the four compounds, four models were generated by GASP. The fitness score for each model ranged from 2589.82 to 2689.23 and model 2 was chosen as the best model ([Table 7](#t7-bt-24-453){ref-type="table"}). Model 2 consists of two hydrophobic regions (HY, cyan), one acceptor atom (AA, green) and one donor site (DS, green) as shown in [Fig. 6](#f6-bt-24-453){ref-type="fig"} with PU3 as the template. The two hydrophobic regions are about 5 Å apart and the hydrophobic region 1 and the acceptor atom is 2.5 Å apart. The pharmacophore model suggested here can be used as a template to further optimize the design of the dual inhibitor of topo II and Hsp90. CONCLUSIONS =========== In this study, the inhibitors reported to target each ATPase domain of human topo II and Hsp90 were investigated. The structures of ATPase domains of topo II and Hsp90 were compared to evaluate how similar the environment of the receptor sites were. The topo II and Hsp90 inhibitors known to target the ATP binding site were searched and the possibility to function as a dual inhibitor was investigated *in silico*. All the inhibitors searched were docked to both topo II and Hsp90. Through the analysis of docking results, four candidate compounds were selected as possible dual inhibitors. These compounds were used as a template to generate pharmacophore model. This suggested pharmacophore model will be useful in developing dual inhibitor of topo II and Hsp90 by constructing 3D query for virtual screening using publically available database such as ZINC (<http://zinc.docking.org/>). This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MEST) (NRF-2013R1A1A2060408), the Korean Health Technology R&D Project funded by Ministry of Health & Welfare, Republic of Korea (HI14C2469), and by the grant of the Bio & Medical Technology Development Program (NRF-2014M3A9A9073 908) of the National Research Foundation of Korea (NRF), funded by the Korean government (Ministry of Science, ICT & Future Planning). {#f1-bt-24-453} {#f2-bt-24-453} {#f3-bt-24-453} {#f4-bt-24-453} {#f5-bt-24-453} {#f6-bt-24-453} ###### The characterization of the active sites of topo II and Hsp90 by CASTp Protein Pocket Mouth --------- -------- -------- --- ------- ------ topo II 792.2 1077.6 2 70.9 52.6 Hsp90 628.9 971.0 1 167.6 78.6 Number of mouth openings for the pocket. Each has to be large enough to allow the solvent probe to pass through. ###### Topo II inhibitors that bind to the ATPase domain Name Structure IC~50~[\*](#tfn2-bt-24-453){ref-type="table-fn"} Type Reference --------------- ----------------------- -------------------------------------------------- ----------------------------------------------------------------------------------------- ------------------ Comp. 1  1.7 µM Purine analog [@b11-bt-24-453] Comp. 2  8.4 µM Purine analog [@b11-bt-24-453] NSC35866  50 µM Purine analog [@b19-bt-24-453] NCS348400  0.39 µM Purine analog [@b18-bt-24-453] QAP1  128 nM Purine analog [@b7-bt-24-453] 2c  K~i~=1.25 µM in ATP competition assay Purine analog with platinum [@b49-bt-24-453] 8-Cl-ATP  [@b52-bt-24-453] 3t  1,3-benzoazolyl-substituted pyrrolo\[2,3-b\]pyrazine derivatives [@b26-bt-24-453] Daurinol  Natural product [@b49-bt-24-453] TSC24  K~d~=18.3 µM compared to ATP (615 µM) Thiosemicarbazone [@b15-bt-24-453] Comp. 5  52.77 µM, K~i~=75 µM N-fused imidazole [@b4-bt-24-453] Comp. 14  Xanthone [@b22-bt-24-453] Comp. 14mod  Xanthone [@b36-bt-24-453] Comp. 18  Naphthoquinone fused cyclic aminoalkyl-phosphonates and aminoalkyl-phosphonic monoester [@b49-bt-24-453] Salvicine  K~d~=74.3 µM Natural product [@b14-bt-24-453] D11  K~d~=37.7 µM Diphyllin glycoside [@b12-bt-24-453] Gambogic acid  K~d~=3.32 µM Natural product [@b41-bt-24-453] Emodin  Anthraquinone analog [@b27-bt-24-453] IC~50~ values for the compound otherwise noted; inhibition constant (K~i~), dissociation constant (K~d~). ###### Hsp90 inhibitors that bind to the ATPase domain Name Structure IC~50~ Type Reference ------------------------------------------------------ ----------------------- ------------------------------------------------------------ ------------------------- ------------------ GDA  K~d~=1.2 µM (determined from isothermal calorimetry (ITC)) [@b42-bt-24-453] Radicicol  23 nM, K~d~=19 nM (ITC) [@b42-bt-24-453] 17-AAG  Geldanamycin derivative [@b44-bt-24-453] 17-DMAG  Geldanamycin derivative [@b20-bt-24-453] KOSN1559  K~d~=16 nM Geldanamycin derivative [@b37-bt-24-453] KF25706  RDC analog [@b46-bt-24-453] KF29163  RDC analog [@b1-bt-24-453] c-RDC  RDC analog [@b53-bt-24-453] Aigialmycin D  RDC analog [@b53-bt-24-453] Pochonin A  90 nM RDC analog [@b30-bt-24-453] Pochonin D  RDC analog [@b31-bt-24-453] KF58333  RDC analog [@b46-bt-24-453] o-(piperidinocarbonyl) methyloxime derivative of RDC  RDC analog [@b16-bt-24-453] PU3  K~d~=15∼20 µM Purine derivative [@b8-bt-24-453] PU3  K~d~=15∼20 µM Purine derivative [@b8-bt-24-453] CCT018159  8.9 µM Pyrazole [@b43-bt-24-453] G3130  K~d~ =280 nM (SPR) Pyrazole [@b24-bt-24-453] SNX0723  Benzamide [@b40-bt-24-453] AUY922  Resorcinol [@b5-bt-24-453] AT-13387  Resorcinol [@b32-bt-24-453] CPUY201112  Resorcinol [@b51-bt-24-453] ###### Topo II docking results of combined inhibitors Name Target Total Score[^1^](#tfn3-bt-24-453){ref-type="table-fn"} Crash[^2^](#tfn4-bt-24-453){ref-type="table-fn"} Polar[^3^](#tfn5-bt-24-453){ref-type="table-fn"} Similarity[^4^](#tfn6-bt-24-453){ref-type="table-fn"} ---------------- --------- -------------------------------------------------------- -------------------------------------------------- -------------------------------------------------- ------------------------------------------------------- PU3 Hsp90 13.1154 −0.9961 4.5667 0.541 8-Cl-ATP Topo II 11.2764 −2.0944 11.0551 0.527 3t Topo II 10.4766 −0.6886 1.7276 0.397 Comp. 14mod Topo II 10.2868 −2.5358 3.0286 0.407 Comp. 14 Topo II 9.927 −2.1249 3.0481 0.427 AUY922 Hsp90 9.8884 −3.5617 3.1905 0.463 Salvicine R Topo II 8.5495 −2.5071 2.9333 0.333 CCT018159 Hsp90 8.0916 −0.9934 2.739 0.382 SNX0723 Hsp90 8.034 −5.2877 2.8872 0.443 NSC348400 Topo II 7.9712 −3.1936 5.9702 0.568 Comp. 2 Topo II 7.9074 −0.4726 2.1046 0.418 Daurinol Topo II 7.564 −1.1849 4.962 0.459 Comp. 5 Topo II 7.4758 −1.6766 2.5831 0.404 QAP1 Topo II 7.3683 −2.2725 2.9142 0.372 G3130 Hsp90 7.1866 −0.3941 3.5543 0.336 Comp. 1 Topo II 7.1634 −1.3298 3.7455 0.235 Salvicine S Topo II 7.0436 −2.892 3.1112 0.461 NSC35866 Topo II 6.9393 −1.0221 2.0358 0.458 KF58333 Hsp90 6.7973 −4.5147 2.4482 0.380 CPUY201112 Hsp90 6.7935 −2.242 1.4817 0.528 AT13387 Hsp90 6.2906 −9.1331 4.0913 0.415 Emodin Topo II 6.279 −0.6391 3.2784 0.440 2c Topo II 5.9724 −5.5461 0.0714 0.342 Pochonin D Hsp90 5.8197 −3.0107 1.6854 0.406 o-RDC Hsp90 5.743 −6.3103 1.714 0.393 Pochonin A Hsp90 5.6544 −3.0449 1.7182 0.431 c-RDC Hsp90 5.571 −3.8203 2.7142 0.353 Comp. 18 Topo II 5.2181 −0.9817 2.1991 0.381 KF25706 Hsp90 4.9578 −3.5256 2.2759 0.467 KF29163 Hsp90 4.8178 −2.5315 2.4504 0.406 RDC Hsp90 4.2342 −2.6456 1.1539 0.421 Aigialomycin D Hsp90 4.1481 −4.3331 4.1964 0.433 TSC24 Topo II 4.1158 −0.8441 0.0514 0.392 Gambogic acid Topo II 3.8992 −7.3216 0.844 0.385 D11 Topo II −1.4678 −14.195 1.1678 0.361 KOSN1559 Hsp90 −2.5462 −14.8183 2.1768 0.428 GDM Hsp90 −4.9652 −18.4653 2.1288 0.358 17-AAG Hsp90 −6.3308 −17.5419 0.7688 0.478 17-DMAG Hsp90 −8.4678 −21.6041 1.4689 0.508 Total Score represents the total Surflex-Dock score expressed as --log(*K*~d~), Crash is the degree of inappropriate penetration by the ligand into the protein between ligand atoms that are separated by rotatable bonds. Crash scores close to 0 are favorable, Polar values show the contribution of the polar interactions to the total score, Similarity indicates the difference between the top scoring pose and the original ligand (AMPPNP) used as the reference. ###### Hsp90 docking results of combined inhibitors Name Target Total Score Crash Polar Similarity ---------------- --------- ------------- --------- -------- ------------ AUY922 Hsp90 11.3281 −2.9912 5.8905 0.566 KOSN1559 Hsp90 10.5386 −4.2057 6.0038 0.497 GDM Hsp90 9.0405 −4.3587 5.7304 0.506 3t Topo II 8.9981 −1.9006 2.2646 0.451 KF58333 Hsp90 8.8035 −2.216 3.7782 0.475 PU3 Hsp90 8.6779 −1.3914 2.3162 0.465 Comp. 14 Topo II 8.0945 −2.9386 4.3295 0.547 o-RDC Hsp90 7.4652 −3.5968 3.1916 0.421 CCT018159 Hsp90 7.4339 −0.7895 3.2601 0.521 SNX0723 Hsp90 7.1997 −2.4905 1.3399 0.526 Salvicine S Topo II 7.1362 −1.4093 1.6745 0.475 AT13387 Hsp90 7.1259 −2.0725 1.7217 0.519 Comp. 14mod Topo II 7.0102 −2.3685 3.184 0.519 G3130 Hsp90 6.8551 −0.5655 4.089 0.514 QAP1 Topo II 6.7837 −2.2874 0.9711 0.497 CPUY201112 Hsp90 6.7467 −2.5928 2.7711 0.612 NSC348400 Topo II 6.7386 −1.4421 3.7083 0.511 RDC Hsp90 6.6884 −2.5576 3.4177 0.666 NSC35866 Topo II 6.649 −1.1619 2.6058 0.486 Salvicine R Topo II 6.4918 −2.4288 2.1968 0.510 Comp. 1 Topo II 6.4641 −1.627 2.2857 0.561 2c Topo II 6.4036 −3.4089 0.0014 0.329 KF25706 Hsp90 6.185 −2.8315 3.6404 0.647 Gambogic acid Topo II 6.0092 −2.2055 1.3905 0.297 Comp. 2 Topo II 5.952 −1.2823 1.4835 0.524 Daurinol Topo II 5.8292 −0.4604 1.3936 0.556 KF29163 Hsp90 5.7498 −2.3134 1.6213 0.530 17-DMAG Hsp90 5.718 −3.6373 1.0756 0.318 c-RDC Hsp90 5.6849 −3.5861 3.0525 0.691 Comp. 5 Topo II 5.6657 −1.6574 1.9241 0.430 D11 Topo II 5.6512 −0.8901 2.5246 0.390 Emodin Topo II 5.6113 −0.429 1.9987 0.398 8-Cl-ATP Topo II 5.5933 −1.2132 4.7994 0.399 Pochonin D Hsp90 5.5041 −1.5103 2.1344 0.550 TSC24 Topo II 5.477 −1.2353 0.5795 0.418 Pochonin A Hsp90 5.4494 −1.3695 2.6113 0.279 17-AAG Hsp90 5.2911 −2.6657 0.545 0.301 Aigialomycin D Hsp90 4.8757 −3.236 3.4461 0.549 Comp. 18 Topo II 4.553 −0.4098 0.9114 0.427 ###### Docking analysis of selected inhibitors Topo II Hsp90 ---------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------ ---------- -------------------------------------------------------------------------------------------------------------------------- ------------------------------------ PU3 Asn91, Asp94, Arg98, Asn120, Ile125, Ile141, Phe142, Ser149, Asn150, Thr159, Gly161, Arg162, Gly164, Ala167, Lys168, Thr215 Asn91, Asn120, Ala167, Thr215 PU3 Asn51, Asp54, Ala55, Asp93, Ile96, Gly97, Met98, Asn106, Leu107, Gly135, Val136, Phe138, Val150, Thr184, Val186 Asp54, Thr184 3t Asn91, Ala92, Asn95, Asn120, Pro126, Ile141, Phe142, Ser149, Gly164, Tyr165, Gly166, Ala167, Lys168, Thr215, Ile217 Asn91 3t Asn51, Ala55, Lys58, Ile96, Gly97, Met98, Asn106, Phe138, Val150, His154, Thr184, Val186 Lys58, Gly97 Comp. 14 Ile88, Asn91, Ala92, Asp94, Asn95, Ile118, Asn120, Ile125, Asn150, Gly161, Gly164, Tyr165, Ala167, Lys168, Thr215, Ile217 Asp94, Thr215 Comp. 14 Asn51, Ser52, Ala55, Asp93, Ile96, Gly97, Met98, Asn106, Phe138, Val150, Thr184, Val186 Asp93, Gly97, Thr184 AUY922 Ile88, Asn91, Ala92, Asn95, Arg98, Ile118, Asn120, Ile125, Pro126, Ile141, Phe142, Ser149, Asn150, Gly161, Gly164, Tyr165, Gly166, Ala167, Lys168, Thr215, Ile217 Asn95, Asn120, Ser149 AUY922 Asn51, Ala55, Lys58, Asp93, Ile96, Gly97, Met98, Asp102, Leu107, Gly135, ValL136, Gly137, Phe138, Val150, Thr184, Val186 Asn51, Lys58, Asp93, Gly97, Phe138 8-Cl-ATP Asn91, Asp94, Asn95, Arg98, Lys123, Gly124, Ile125, Ser149, Asn150, Gly161, Arg162, Asn163, Gly164, Tyr165, Gly166, Ala167, Gln376, Lys378 Asn91, Asp94, Asn150, Arg162, Tyr165, Gly166, Lys378 KOSN1599 Asn51, Ser52, Asp54, Ala55, Lys58, Asp93, Ile96, Met98, ASP102, Asn106, Leu107, Phe138, Thr184, Val186 Ser52, Asp54, Phe138 ###### Results of pharmacophore hypothesis generated by GASP Model Fitness Size[^a^](#tfn7-bt-24-453){ref-type="table-fn"} Hits[^b^](#tfn8-bt-24-453){ref-type="table-fn"} Dmean[^c^](#tfn9-bt-24-453){ref-type="table-fn"} Features[^d^](#tfn10-bt-24-453){ref-type="table-fn"} ------- --------- ------------------------------------------------- ------------------------------------------------- -------------------------------------------------- ------------------------------------------------------ 1 2676.46 4 4 5.7693 DS, AA, HY1, HY2 2 2689.23 4 4 3.5916 DS, AA, HY1, HY2 3 2589.82 4 4 3.1774 DS, AA, HY1, HY2 4 2663.25 2 4 4.5547 HY1, HY2 Number of features in the model, Number of molecules that matched during the search, Average interpoint distance, Pharmacophore features. DS: donor site, AA: acceptor site, HY: hydrophobic.

Tumor tissue consists not of only cancer cells, but also various stromal cells, such as cancer-associated fibroblasts (CAF), immune inflammatory cells, myeloid progenitor cells and vascular endothelial cells.[@b1] The heterogeneity of the complicated environment within tumor tissue, or the tumor microenvironment, plays an important role in cancer malignancy progression.[@b1] Recent evidence suggests that not only the tumor microenvironment but also cancer cells themselves within tumor tissue are heterogeneous by representing numerous subpopulations with both genetic and non-genetic variations.[@b2],[@b3] Cancer metastasis disease, one of the major contributing factors to the high mortality rate in cancer patients, involves multiple biological steps, including intravasation, attachment to vessels, extravasation, angiogenesis and subsequent growth in distal tissues of the primary tumor. Among those steps, the initial acquisition of cellular invasiveness is likely a key stage in metastatic dissemination from the primary tumor site, and the process of epithelial-to-mesenchymal transition (EMT) is known to play an important role during this process.[@b4]--[@b7] In accordance with the dynamic yet transient morphological and phenotypic alteration of cancer cells during the EMT process, such mesenchymal-transitioned cancer cells are often seen at the invasive front of tumor tissue beside neighboring epithelial cancer cells.[@b8]--[@b13] Even though the importance of transforming growth factor (TGF)-β in the initiation of EMT has been demonstrated,[@b14],[@b15] there are several reports that primary cancer specimens acquire mesenchymal features and develop metastatic disease even in the presence of the deletion of in Smad4, which is a key component of the TGF-β signaling pathway.[@b16],[@b17] These observations might imply that there are alternative pathways to maintain the EMT phenotype other than TGF-β pathway within the tumor microenvironment. Interestingly, the cooperation of mesenchymal-transitioned and surrounding epithelial cancer cells in establishing spontaneous metastasis in a mouse model has been reported.[@b18] It has also been revealed that the heterogeneity of cancer cells and the intra-tumoral cross-talk between distinct types of cancer cells might contribute to the induction of abnormal proliferation and metastasis;[@b19]--[@b21] however, the exact mechanism of how those distinct cancer cell types interact with each other is not yet understood. In the present study, we demonstrate that the coexistence of mesenchymal-transitioned cancer cells with epithelial cancer cells induces the invasive ability and the metastatic potential of epithelial cancer cells *in vitro* and *in vivo*. Furthermore, we identified WNT3 and WNT5B as the secretory factors from TGF-β-induced mesenchymal-transitioned cancer cells that induce the invasion of neighboring epithelial cancer cells and secondary EMT phenotype. Collectively, these results strongly implicate secretory WNT ligands as critical soluble factors mediating the invasion instigation of epithelial cancer cells derived from mesenchymal-transitioned cancer cells, and further targeting those secretory WNT proteins could be a new approach to prevent cancer invasion and subsequent metastasis. Materials and Methods ===================== Cell culture and inhibitors --------------------------- Human lung adenocarcinoma A549 and human pancreatic ductal adenocarcinoma Panc-1 cells were obtained from the American Type Culture Collection. A549 cells were maintained in RPMI1640, and Panc-1 cells were maintained in DMEM, containing 10% FBS, 1 mM L-glutamine and antibiotics (100 units/mL penicillin and 100 mg/mL streptomycin) in a humidified atmosphere of 95% air and 5% CO~2~ at 37°C. To establish labeled A549 and Panc-1 cells, A549 or Panc-1 cells were transfected with pGL4.50/Luc2 (Promega, Madison, WI, USA) or pEGFP-C1 (Clontech, Palo Alto, CA, USA), selected, and cloned in growth medium containing 200 μg/mL hygromycin B or 1 mg/mL G418, respectively. The reagents used were: WNT secretion inhibitor, IWP-2 (Sigma-Aldrich, St. Louis, MO, USA) and TGF-β receptor kinase inhibitor (TβRI) (Calbiochem, Dermstadt, Germany). Preparation of E-cells and M-cells ---------------------------------- In this study, non-stimulated epithelial A549 and Panc-1 cells were used as E-cells. To prepare the mesenchymal-transitioned cancer cells (M-cells), A549 or Panc-1 cells were treated with 5 ng/mL recombinant TGF-β (Pepro Tech, Rocky Hill, NJ, USA) for 48 h then washed with fresh culture medium twice and harvested for subsequent experiments. Direct and separated co-culture experiment ------------------------------------------ For the direct co-culture experiment, E-cells and M-cells were re-seeded into 60 mm culture dish according to the indicated cell number and co-cultured for 24 h. For the separated co-culture experiment, 1 × 10^5^ E-cells and 3 × 10^5^ cells of M-cells were seeded into lower or upper compartments of the transwell chamber with 1 μm pore diameter (BD Falcon, Bedford, MA, USA) for 24 h. Generation of conditioned medium -------------------------------- After preparation of E-cells and M-cells, the cells were further cultured in fresh growth medium for an additional 48 h. Finally, these culture supernatants were collected and diluted with fresh growth medium at a ratio of 2:1. The freshly prepared supernatant was used as the conditioned medium (CM) in each experiment. For preparation of WNT3-depleted and WNT5B-depleted CM by siRNA, parental A549 and Panc-1 cells were transfected with siControl (siGENOME Control Pool Non-targeting siRNA\#2), siWNT3 and/or siWNT5B (ON-TARGETplus SMARTpool siRNA) (Thermo Fisher Scientific, Rockford, IL, USA) by Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA, USA) 48 h prior to induction of M-cells. For preparation of WNT-depleted CM by IWP-2, E-cells and M-cells were cultured in fresh medium containing IWP-2 for 48 h. The protein level of WNT3 and WNT5B in CM was determined by a specific ELISA (CUSABIO Biotech, Wuhan, China) according to the manufacturer\'s protocol. Matrigel invasion assay ----------------------- **C**ancer cell invasion through reconstituted basement membrane (Matrigel \[BD Biosciences\], San Jose, CA, USA) was assayed as previously described.[@b22] After fixing the filter and staining with H&E, the invaded cells were counted manually under a microscope at ×100. For detection of luciferase activity in invaded A549/Luc2 cells, the filters were soaked in passive lysis buffer (Promega) and luciferase activity was determined. For detection of EGFP^+^ invaded cells, filters were fixed with 4% paraformaldehyde and stained with VECTASHIELD mounting media with DAPI (Vector Laboratories, Burlingame, CA, USA). Western blotting ---------------- Whole cell lysates and nuclear protein extracts were prepared as described previously.[@b23] The primary antibodies used were Epithelial-Mesenchymal transition (EMT) Antibody Sampler Kit (\#9782, Cell Signaling Technology, Beverly, MA, USA), antibodies against WNT3 (ab32249) and WNT5B (ab94914) from Abcam, and antibodies against PCNA (PC10), β-actin (C11), Lamin B (C-20) and α-tubulin (D-10) from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Experimental lung metastasis experiment --------------------------------------- C.B-17/lcrHsd-*Prkdc*^scid^ mice were purchased from Japan SLC (Hamamatsu, Japan). All experiments were approved and performed according to the guidelines of the Care and Use of Laboratory Animals of the University of Toyama. Cells were inoculated intravenously (2 × 10^6^ cells/200 μL PBS/mouse) into mice and the lungs were removed 24 h after the tumor inoculation. Mice were intraperitoneally injected with 200 μL of luciferin (1.5 mg/mL \[VivoGlo; Promega\]) 20 min prior to subject bioluminescent assay by using an *in vivo* imaging system (IVIS Lumina II, Caliper Life Sciences, Hopkinton, MA, USA). The data was presented as the mean luminescence ± SEM. Microarray data analysis ------------------------ The datasets (GSE17708 and GSE23952) were reanalyzed on GenePattern.[@b24] Briefly, the differential expression level of all genes between TGF-β-treated samples and non-treated samples was computed and the top 5% of upregulated genes in TGF-β-treated cells compared with control cells were selected by using the "Comparative Marker Selection" tool from each dataset. Finally, the genes coding the secreted proteins were picked up from the commonly upregulated genes in both datasets. Gene set enrichment analysis (GSEA) was performed using javaGSEA application v2.0.13 (GSEA, Broad Institute, Boston, MA, USA). These pathway gene sets were provided by the Molecular Signatures Database (MSigDB \[[http:\\\\www.broadinstitute.org/gsea/msigdb](http:\\www.broadinstitute.org/gsea/msigdb)\]). Statistical analysis -------------------- Statistical significance was calculated using Microsoft Excel. More than three means were composed using one-way [anova]{.smallcaps} with the Bonferroni correction, and two means were composed using unpaired Student\'s *t*-test. *P *\< 0.05 were considered statistically significant. Results ======= Epithelial cancer cells acquire metastatic potential by the co-culture with mesenchymal-transitioned cancer cells ----------------------------------------------------------------------------------------------------------------- To directly evaluate the role of heterogeneity of cancer cells in their invasive potential, we performed Matrigel invasion assays in human epithelial lung adenocarcinoma A549 and pancreatic ductal adenocarcinoma Panc-1 cell co-culture at various ratios of epithelial cancer cells (E-cells) and TGF-β-induced mesenchymal-transitioned cancer cells (M-cells), because TGF-β is known to trigger EMT in both A549 and Panc-1, which is the most potent EMT-inducing cytokine (Fig. S1).[@b25],[@b26] After 24 h co-culture of E-cells and M-cells, the invasive potential of the E-cell and M-cell mixture was more enhanced in both A549 and Panc-1 cell lines, in an M-cell dose-dependent manner, than that in M-cells alone (Fig.[1](#fig01){ref-type="fig"}a). To determine the cell population, either E-cells or M-cells, responsible for the enhanced invasive potential after the co-culture, we employed luminescence-labeled E-A549 (E-A549/Luc2) or fluorescent-labeled E-Panc-1 (E-Panc/EGFP) to distinguish E-cells from M-cells within the co-culture. By measuring the invasiveness of labeled E-cells in either A549 or Panc-1, we have found that the invasive potential of E-cells co-cultured with M-cells was higher than that of E-cells alone in both A549 and Panc-1 (Fig.[1](#fig01){ref-type="fig"}b) and such induction of invasiveness was observed in an M-cell dose-dependent manner (Fig.[1](#fig01){ref-type="fig"}b). We have also found that the invasion potential of E-cells was not altered without the co-culture with M-cells (Fig. S2). Of note, the invasive potential of M-Panc cells was not affected by the co-culture with E-Panc cells as compared with M-Panc cells alone (data not shown). These results clearly indicate that the co-culture of E-cells with M-cells selectively enhances the invasiveness of E-cells in both A549 and Panc-1 cell lines. {#fig01} To further investigate the significance of the interaction between E-cells and M-cells *in vivo*, the metastatic potential of E-A549 cells to the lungs was examined by injecting E-A549/Luc2 cells upon co-culture with either E-A549 or M-A549. The metastatic spread of E-A549/Luc2 in the lungs was much higher after the co-culture with M-A549 cells compared to with E-A549 cells (Fig.[1](#fig01){ref-type="fig"}c), therefore indicating the potential of M-cells to promote the metastatic ability of neighboring E-cells *in vivo*. Mesenchymal-transitioned cancer cells induce the metastatic potential of the neighboring epithelial cancer cells in a cell--cell contact independent mechanism -------------------------------------------------------------------------------------------------------------------------------------------------------------- To determine whether the direct cell--cell interaction is required for promoting the metastatic potential of E-cells by neighboring with M-cells, we first examined the expression of EMT-related proteins in A549 or Panc-1 cells under the co-culture of E-cells with M-cells in the presence or absence of direct cell--cell contact (Fig.[2](#fig02){ref-type="fig"}a,b). Given that the upregulation of the mesenchymal marker (Snail and Vimentin) and the downregulation of the epithelial marker (E-cadherin) were markedly induced by both direct and separated co-culture of E-cells and M-cells, the soluble factor(s) derived from M-cells was at least sufficient for the induction of the secondary EMT phenotype of E-cells upon co-culture with M-cells. We further confirmed that the invasion of E-A549 cells was remarkably enhanced after the culture with M-A549-derived conditioned medium (M-A549-CM) compared with E-A549-CM (Fig.[2](#fig02){ref-type="fig"}c and Fig. S3). In concert with the enhanced invasive ability of E-A549 cells after the cultivation with M-A549-CM, the secondary EMT phenotype in E-A549 cells was also induced after the culture with M-A549-CM. Similar results were obtained in the invasion of E-Panc cells after the cultivation with M-Panc-CM (data not shown). {#fig02} Importantly, the CM from M-Panc cells was able to introduce enhanced invasive ability and secondary EMT phenotype in E-A549 cells (Fig.[2](#fig02){ref-type="fig"}d and Fig. S3), indicating that the common soluble factor(s) derived from M-A549 cells and M-Panc cells are likely to be involved in this process. Considering that the induction of secondary EMT in E-cells by M-cell-CM was not affected by TGF-β receptor kinase inhibitor (data not shown), the involvement of the TGF-β signaling pathway is less likely. Collectively, these data indicate that mesenchymal-transitioned cancer cell-derived soluble factor(s), which is common in both M-A549 and M-Panc cells, play a significant role in the induction of invasive ability and secondary EMT phenotype in the neighboring epithelial cancer cells. WNT3 and WNT5B derived from mesenchymal-transitioned cancer cells are the soluble factors that induce metastatic potential in the neighboring epithelial cancer cells --------------------------------------------------------------------------------------------------------------------------------------------------------------------- In order to identify the common soluble factor(s) that is secreted from mesenchymal-transitioned A549 and Panc-1 cells, we analyzed the published cDNA microarray datasets (GSE17708 and GSE23952) representing A549 and Panc-1 gene expression following the TGF-β stimulation for 72 or 48 h, respectively. There are 55 candidate genes as the top 5% of encoding secretory proteins that are commonly upregulated in both A549 and Panc-1 cells (Fig.[3](#fig03){ref-type="fig"}a and Table S1). By using Gene Set Enrichment Analysis, we further selected candidate pathway gene sets that are significantly enriched in phenotype of TGF-β as shown in Table S2. Among those candidate pathways, WNT pathway was commonly enriched in both M-A549 and M-Panc. Thus, we further focused on WNT3 and WNT5B molecules in the induction of secondary EMT in epithelial cancer cells. WNT3 and WNT5B are known to be a ligand for activating both canonical and non-canonical WNT pathways.[@b27] As shown in Fig.[3](#fig03){ref-type="fig"}b, we confirmed the higher expression of WNT3 and WNT5B at protein level in both M-cells compared to E-cells. Consistent with the upregulation of WNT3 and WNT5B, the secretion of these WNT ligands was detected in CM of M-A549 by ELISA (Fig.[3](#fig03){ref-type="fig"}c). We also confirmed higher nuclear β-catenin expression and β-catenin transcriptional activity in E-cells with M-cell-CM, indicating that E-cells received the WNT signals from M-cells (data not shown). {#fig03} To further examine whether WNT3 and WNT5B are the molecules responsible in M-cell-derived CM for the induction of invasiveness and secondary EMT phenotype in E-cells, we used the siRNA of WNT3 and WNT5B or the chemical WNT ligand secretion inhibitor, IWP-2, during the preparation of CM. The knockdown efficiencies or the inhibition of secretion were confirmed by qRT-PCR, western blotting and ELISA (Fig. S4). While the M-cell-CM derived from single knockdown of either WNT3 or WNT5B did not completely diminish the induction of invasive potential and Vimentin/Snail expression of E-A549 and E-Panc cells, the knockdown of both WNT3 and WNT5B completely abrogated the activity of M-cell-CM in the induction of invasive potential and secondary EMT phenotype of E-A549 and E-Panc cells (Fig.[4](#fig04){ref-type="fig"}a). Importantly, the M-cell-CM prepared in the presence of IWP-2 also completely diminished its activity to induce invasion and secondary EMT phenotype in both E-A549 and E-Panc cells (Fig.[4](#fig04){ref-type="fig"}b,c). These results strongly indicate that WNT3 and WNT5B are likely to be key soluble factors produced by mesenchymal-transitioned cancer cells to instigate the metastatic potential of neighboring epithelial cancer cells by enhancing their invasiveness and inducing secondary EMT phenotype. ![Critical requirement of secretory WNT3 and WNT5B from mesenchymal-transitioned cancer cells for inducing secondary epithelial-to-mesenchymal transition (EMT) phenotype of epithelical cancer cells. (a, b) E-A549 (left) and E-Panc (right) cells were treated with WNT-depleted CM derived from E-cells or M-cells either by siRNAs against WNT3/WNT5B (a) or by 10 μM WNT secretion inhibitor (IWP-2) (b) for 48 h. The cells were subjected to Matrigel invasion assay and western blotting. \*\**P *\< 0.01 by one-way [anova]{.smallcaps} with the Bonferroni correction. (c) E-A549 or E-Panc cells were treated with WNT-depleted CM derived from E-cells or M-cells by the indicated dose of IWP-2 for 48 h and the expression of proteins were determined by western blotting.](cas0105-0281-f4){#fig04} Finally, we have tested whether the mesenchymal-transitioned cancer cells can instigate metastatic spread of neighboring epithelial cancer cells through providing secretory WNT3 and WNT5B ligands. Consistent with the induction of invasiveness and secondary EMT phenotype *in vitro*, the lung colonization of E-A549/Luc2 cells co-cultured with M-A549 cells was largely enhanced compared to that of E-A549/Luc2 cells co-cultured with E-A549 cells (Fig.[5](#fig05){ref-type="fig"}). Such increased lung colonization was fully abrogated when E-A549/Luc2 cells co-cultured with M-A549 cells in the presence of IWP-2. Collectively, these results strongly indicate that the secretory WNT3 and WNT5B derived from mesenchymal-transitioned cancer cells are able to enhance the metastatic potential of neighboring epithelial cancer cells *in vivo*. {#fig05} Discussion ========== Considering the significance of the *in vivo* metastatic ability of epithelial cancer cells upon co-culture with mesenchymal-transitioned cancer cells (Figs[1](#fig01){ref-type="fig"} and [5](#fig05){ref-type="fig"}), there are additional effects on epithelial cancer cells other than inducing invasive ability and secondary EMT phenotype through the secretion of WNT ligands by neighboring mesenchymal-transitioned cancer cells. First, it is reported that mesenchymal-transitioned cancer cells play a unique role in escorting epithelial cancer cells to metastatic organ *in vivo*.[@b18] Second, the interaction with other host stromal cells, such as platelets or fibroblasts, might be involved in cancer metastasis by preventing cancer cells from cellular stresses, inducing EMT within the blood stream,[@b28] or leading to collective cell invasion through gap junctions or integrins.[@b29],[@b30] Third, WNT ligands are known to not only enhance the invasiveness or secondary EMT of epithelial cancer cells but also affect the multiple steps in cancer metastasis and progression. Along with the paracrine WNT signaling as seen in this study, the autocrine WNT signaling of mesenchymal-transitioned cancer cells has been reported to contribute to the maintenance of mesenchymal phenotypes and stem cell-like states in breast cancer.[@b31] It has also been reported that WNT signaling is involved in the expression of matrix metalloproteinases for digesting extracellular matrix during intra-vasation or extra-vasation of the metastasis process,[@b32] niche formation,[@b33] and enhancement of anchorage-independent sphere formation to increase the metastatic ability of pancreatic cancer cells.[@b34] Collectively, this evidence might indicate that not only the cell-contact independent interaction but also the involvement of direct cell--cell interaction between epithelial cancer cells and mesenchymal-transitioned cancer cells play important roles in the regulation of the metastasis process. Although WNT3 and WNT5B secreted from mesenchymal-transitioned cancer cells are indispensable in inducing invasiveness and secondary EMT in neighboring epithelial cancer cells (Fig.[4](#fig04){ref-type="fig"}), we have also identified other gene candidates of secretory proteins, which include other known-inducers of cancer cell invasiveness and EMT, such as CXCL12,[@b35] LOX/LOXL2[@b36] and HB-EGF[@b37]. Those secretory proteins from mesenchymal-transitioned cancer cells might also contribute to induce invasiveness and secondary EMT in epithelial cancer cells by cooperating with WNT ligands; therefore, further studies are required in this context. Transforming growth factor-β is one of the most potent inducers of EMT and metastasis; however, TGF-β pathway could be often genetically abrogated in relatively late-stage tumors because of the deletion or mutation of Smad4 (*DPC4*) or TGF-β receptors.[@b16],[@b17],[@b38] In addition, TGF-β -induced EMT is reversible unless there is long-term exposure to TGF-β as reported previously,[@b39],[@b40]. We also clarified that M-cell-CM-dependent EMT was also a reversible process, at least *in vitro* (data not shown). Thus, the heterogeneity of epithelial and mesenchymal-transitioned cancer cells might be maintained within the tumor microenvironment through such dynamic cellular transition between E-cell and M-cell states. Considering TGF-β receptor kinase inhibitor did not affect the induction of invasive ability and secondary EMT phenotype in this study (data not shown), the paracrine WNT stimulation can be an alternative inducer of EMT and metastasis to TGF-β in the cross-talk between mesenchymal-transitioned cancer cells and epithelial cancer cells. In this study, we focused on the WNT ligands secreted from mesenchymal-transitioned cancer cells; however, other stromal cells in the cancer microenvironment might also produce WNT ligands and, therefore, be involved in the cancer metastasis process. Although we did not observe the induction of WNT3 and WNT5B by TGF-β stimulation in mouse NIH3T3 fibroblast or primary human lung fibroblasts (data not shown), it has been reported that upregulation of WNT3A in CAF could result in the aggressive progression of prostate tumor.[@b41] Besides secretion of WNT ligands, the hyperactivation of WNT signaling pathway has been observed in highly metastatic lung adenocarcinoma, colon cancer[@b42],[@b43] and pancreatic cancer.[@b34] In the context of the clinical significance, WNT3 was reported to be associated with poor prognosis of non-small cell lung cancer,[@b44] and to promote EMT in HER2-overexpressing breast cancer cells.[@b45] Although we cannot exclude the possibility that WNT5B need to be coordinated with WNT3 to induce cellular invasion, we believe WNT5B could be solely responsible for impaired instigation considering that the non-canonical WNT pathway through WNT5B is reported to be involved in inducing tumor invasion.[@b46],[@b47] Furthermore, WNT5A, a paralog of WNT5B, and its receptor (FZD3) are known to be involved in the promotion of cell motility through the activation of paracrine non-canonical WNT signaling in skin cancer.[@b48] Even though IWP-2 is a pan Wnt ligand secretion inhibitor, our presented data by knockdown both WNT3 or 5B with siRNA almost completely diminished the activity of M-cell CM to induce invasion of E-cells; therefore, these results strongly suggest that both WNT3 and WNT5B from M-cells are important for the induction of E-cell invasion. Given that E-cell CM in the presence of IWP-2 downregulated Snail or nuclear β-catenin expression in E-A549 cells (Fig.[4](#fig04){ref-type="fig"}c and Fig S5), we speculate that even E-cells may produce substantial levels of WNT ligand, by which Snail or β-catenin expression of E-cells is maintained in an autocrine manner. Collectively, WNT ligands derived from cancer stromal cells as well as mesenchymal-transitioned cancer cells and subsequent activation of WNT-signaling pathway may play a significant role in the malignant behavior of cancer cells, including metastatic spread to distant organs. In conclusion, the intra-tumoral heterogeneity has been considered to be one of hallmarks in cancer malignancy and we have newly identified that secretory WNT ligands from mesenchymal-transitioned cancer cells instigate the invasion of neighboring epithelial cancer cells. This novel function of WNT signaling in the cancer microenvironment could be an attractive target not only for the new therapeutic opportunity but also for the new biomarker candidate in metastatic disease. This work was supported in part by Grants-in-aid for Challenging Exploratory Research 24659348 (IS), by Grant-in-Aid 24700971 for Young Scientists (B) (SY) from the Ministry of Education, Culture, Sports, Science, and Technology, Japan, and by Grant-in-Aid for JSPS Fellows (DC2) 2510159 from Japan Society for the Promotion of Science (SK). We thank members of the Saiki laboratory for discussions and suggestions. Disclosure Statement ==================== The authors have no conflict of interest. Funding information =================== Grants-in-aid for Challenging Exploratory Research (24659348) and Grant-in-aid for Young Scientists (B) (24700971) from Ministry of Education, Culture, Sports, Science, and Technology, Japan. Grant-in-Aid for JSPS Fellows (DC2) (2510159) from Japan Society for the Promotion of Science, Japan. Supporting Information ====================== Additional supporting information may be found in the online version of this article. ###### Expression of epithelial-to-mesenchymal transition (EMT)-related proteins following transforming growth factor (TGF-β-induced EMT. ###### Effect of seeded cell number on invasion potential in E-cells. ###### Quantification of western blotting bands related to Fig. [2](#fig02){ref-type="fig"} by densitometry. ###### mRNA expression of WNT3 and WNT5B upon siRNA and transforming growth factor (TGF-β treatment. ###### Quantification of western blotting bands related to Fig. [4](#fig04){ref-type="fig"}a and c by densitometry. ###### Common upregulated genes (top 5%) in both mesenchymal-transitioned A549 and Panc-1 cells by the stimulation of transforming growth factor (TGF-β for 7 and 48 h, respectively. ###### Enriched pathways in transforming growth factor (TGF- -treated cells. ###### Materials and Methods.

1.. Introduction {#s1} ================ Animals and humans go to great lengths to obtain rewards, such as food and water, and to avoid punishment, such as bodily damage and pain. Essential to these processes is the learning of cues predictive of such actual or first-order reinforcement. Critically, predictive cues not only acquire learned valence but, once predictive relationships are established, also can confer learned valence themselves; i.e. they can serve as second-order reinforcement \[[@RSBL20190084C1]--[@RSBL20190084C3]\]. In humans, for example, learning that money can buy food establishes money as a second-order reward. In general, second-order conditioning may underlie chains of predictions and early anticipatory behaviour in humans and animals. Indeed, the capacity for second-order conditioning is widely distributed across the animal kingdom, including insects \[[@RSBL20190084C4]--[@RSBL20190084C7]\], and is implemented in many computational models of associative learning \[[@RSBL20190084C8]\]. In flies, presenting odour A with an electric shock punishment and odour B without punishment leads to learned avoidance of A in a subsequent choice test. This learning of an odour as a predictor of electric shock takes place in the Kenyon cells (KCs) of the mushroom body ([figure 1](#RSBL20190084F1){ref-type="fig"}*a*) \[[@RSBL20190084C9]--[@RSBL20190084C12]\]. The mushroom body provides a sparse, combinatorial representation of the sensory environment, including odours. Along their long axonal fibres, the KCs further receive intersecting input from neurons mediating internal reinforcement, many of which are dopaminergic (DANs). The coincidence of activation by odour and of DAN signalling can lead to presynaptic plasticity at the cholinergic synapse between the KCs and the output neurons of the mushroom body (MBONs). Arborizations from DANs and MBONs overlap and are regionally confined along the KC fibres, establishing a characteristic compartmental organization. In the case of the PPL1-01 DAN mediating an internal punishment signal, synaptic strength between the odour-coding KCs and the approach-promoting MBON-11 is reduced \[[@RSBL20190084C18],[@RSBL20190084C19]\]. For the punished odour, the innate balance between approach and avoidance is thus tilted in favour of avoidance. In other words, activity in PPL1-01 can provide first-order punishment, and an odour that predicts first-order punishment leads to reduced activity in MBON-11. We therefore wondered whether, in experimentally naive flies, optogenetically silencing MBON-11 might be an analogue of a punishment-predicting odour such that it confers a punishing effect of second-order upon an actually present odour associated with such silencing (also see \[[@RSBL20190084C20]\])---and whether in turn optogenetically activating MBON-11 might have a rewarding effect. Figure 1.(*a*) Simplified account of odour--shock associative learning in flies (after \[[@RSBL20190084C9]--[@RSBL20190084C14]\]). Odour presentation in untrained animals mediates balanced approach and avoidance tendencies of mushroom body output neurons (MBONs). Coincidence of odour-evoked activity in the mushroom body Kenyon cells (KCs) and activity of the dopaminergic neuron PPL1-01 evoked by the electric shock leads to a depression of the synapses from these KCs to an approach-promoting MBON. In a subsequent test, this allows avoidance tendencies through non-depressed KC-MBON synapses in parallel compartments to prevail. The organization of innate olfactory, punishment- and reward-related behaviour largely bypasses the mushroom body. For simplicity KC--KC, KC--DAN, DAN--MBON and MBON--MBON synapses are omitted from this figure \[[@RSBL20190084C15],[@RSBL20190084C16]\]. Cloud: odour; star: depressed/non-depressed KC-MBON synapse. A possible feedback from the MBONs towards the DANs is indicated. Note the multiple targets of MBON-11 within the ipsi- and contralateral mushroom body, as well as outside the mushroom body sketched in (*d*). (*b*) Presenting odour (cloud) with green light (star) leads to aversive associative memory in flies expressing the green-light-gated anion-channel GtACR1 in MBON-11, but not in genetic controls. (*c*) As in (*b*), using three training trials with an inter-trial interval of 3 min. (*d*) Sketch of connectivity of MBON-11; Greek letters refer to mushroom body lobes. Target regions of MBON-11 outside the mushroom body include MBON-01, the crepine (CRE) and the superior medial, intermediate and lateral protocerebrum (SMP, SIP, SLP) (after \[[@RSBL20190084C13]\]). Postsynaptic partners of the contralateral branch of MBON-11 include PPL1-01 \[[@RSBL20190084C17]\]. All these target regions could contribute to the reinforcing effects of manipulating the activity of MBON-11. Data are displayed as box plots (middle line: median; box boundaries and whiskers: 25/75% and 10/90% quantiles, respectively). Data were analysed across groups by Kruskal--Wallis tests at *p* \< 0.05, followed in the case of significance by pairwise comparisons with Mann--Whitney *U*-tests at *p* \< 0.05 with Bonferroni--Holm correction (asterisk). Underlying preference scores can be found in the electronic supplementary material, figure S1. Sample sizes and statistical results can be found in the electronic supplementary material, table S1. A '+' below box plots indicates the presence of the respective transgene. (Online version in colour.) 2.. Material and methods {#s2} ======================== Procedures follow \[[@RSBL20190084C21]\], unless mentioned otherwise. *Drosophila melanogaster* were maintained on standard food, with 60--70% relative humidity, at 25°C, and in constant darkness to prevent unintended optogenetic effects. Flies aged 1--3 days after hatching were collected and kept at 18°C for up to four additional days. MB320C and MB085C (Fly Light Split-GAL4 Driver Collection) \[[@RSBL20190084C13]\] as driver strains covering the PPL1-01 and MBON-11 neurons, respectively, were crossed to UAS-ChR2-XXL (Bloomington stock number: 58374) \[[@RSBL20190084C22]\] or UAS-GtACR1 as effectors for optogenetic activation or silencing, respectively. To generate the latter strain, the GtACR1 DNA was synthesized (Thermo Fisher Scientific) according to the published sequence \[[@RSBL20190084C23]\] with codon usage optimized to *D. melanogaster*. The synthesized GtACR1 DNA with a C-terminal YFP was inserted into the expression vector pJFRC7. Embryo injection (BestGene Inc.) was performed to establish flies carrying UAS-GtACR1. Crosses for genetic controls yielded animals heterozygous for either construct. Synonyms for PPL1-01 are PPL1-γ1pedc and MB-MP1; synonyms for MBON-11 are MBON-γ1pedc\>α/β and MB-MVP2. Behavioural experiments used a set-up from CON-ELEKTRONIK (Greussenheim, Germany) and took place at 23--25°C and 60--80% relative humidity. Training was performed in red light, which is invisible to flies, and testing in darkness. As odorants, 50 µl benzaldehyde (BA) and 250 µl 3-octanol (OCT) (CAS 100-52-7, 589-98-0; both from Fluka, Steinheim, Germany) were applied to 1 cm-deep Teflon containers of 5 and 14 mm diameter, respectively. Flies were presented with both odours during training, but only one was paired with light for optogenetic activation (465 nm) or silencing (520 nm), whereas the other odour was presented alone (see electronic supplementary material, figure S2, for more details). The flies were then tested in a T-maze for their choice between the two odours. From the number of flies choosing each odour (\#), the relative preference was calculated as$${{BA}\,{Preference}}\, = \left( \frac{{\#\,_{BA}} - \#\,_{OCT}}{\#\,{Total}} \right) \times 100.$$The presentation of BA and OCT with or without the light (\*) was alternated between repetitions of the experiment, allowing an associative memory score to be obtained from reciprocally trained sets of flies as$${{Memory}\,{score}} = {\frac{{{BA}\,{Preferenc}}e_{{BA} \ast} - {{BA}\,{Preferenc}}e_{{OCT} \ast}}{2}.}$$ Data were analysed with Kruskal--Wallis tests (KW-tests) to compare more than two groups, Mann--Whitney *U*-tests (*U*-test) for pairwise comparisons, one-sample sign-tests for comparisons to chance level (i.e. zero), in all cases with Bonferroni--Holm corrections of *p* \< 0.05 significance levels as appropriate, using Statistica 11.0 (StatSoft, Hamburg, Germany) and R 2.15.1 ([www.r-project.org](www.r-project.org)). 3.. Results {#s3} =========== Presenting an odour together with optogenetically silencing MBON-11 via the green-light-gated anion-channel GtACR1 established aversive memory for the odour ([figure 1](#RSBL20190084F1){ref-type="fig"}*b*). This effect was replicated using three training cycles ([figure 1](#RSBL20190084F1){ref-type="fig"}*c*). Consideration of the genetic controls suggests a weak appetitive olfactory memory through the pairing of odour with the green light, which is visible to the flies. Critically, relative to either genetic control, silencing MBON-11 had a punishing effect. Conversely, does activating MBON-11 have a rewarding effect? Presenting an odour together with optogenetically activating MBON-11 via the blue-light-gated cation-channel ChR2-XXL established appetitive memory for the odour ([figure 2](#RSBL20190084F2){ref-type="fig"}). Corresponding to what is typically observed for primary food rewards such as sugar \[[@RSBL20190084C24]\], this appetitive memory appeared slightly stronger under starved conditions ([figure 2](#RSBL20190084F2){ref-type="fig"}*c*; indeed starvation was shown to facilitate MBON-11 activity \[[@RSBL20190084C25]\]). In the case of blue light too, the data from the genetic controls suggest a weakly rewarding effect. We further note that relative to the respective genetic controls, the punishing effect of silencing MBON-11 ([figure 1](#RSBL20190084F1){ref-type="fig"}*c*) appears to be stronger than the rewarding effect of activating it ([figure 2](#RSBL20190084F2){ref-type="fig"}*b*). Figure 2.(*a*,*b*) The same as in [figure 1](#RSBL20190084F1){ref-type="fig"}*b*, *c* but using ChR2-XXL to activate MBON-11 (star). This leads to stronger appetitive learning in the experimental genotype than in genetic controls. (*c*) Same as the experiment in (*a*), but with an initial 18 h period of wet starvation, which improves appetitive learning \[[@RSBL20190084C24]\]. Underlying preference scores can be found in electronic supplementary material, figure S3. Sample sizes and statistical results can be found in electronic supplementary material, table S1. Other details as in the legend of [figure 1](#RSBL20190084F1){ref-type="fig"}. (Online version in colour.) We conclude that silencing/activating MBON-11 has a punishing/rewarding effect. 4.. Discussion {#s4} ============== MBON-11 is GABAergic \[[@RSBL20190084C13]\]. It targets premotor circuitry outside the mushroom bodies, and hetero-compartmental regions in the ipsi- and the contralateral mushroom body, and furthermore features a homo-compartmental and contralateral feedback loop onto the dopaminergic, punishing PPL1-01 neuron ([figure 1](#RSBL20190084F1){ref-type="fig"}*d*) \[[@RSBL20190084C13],[@RSBL20190084C17],[@RSBL20190084C25],[@RSBL20190084C26]\]. All of these regions could contribute to reinforcement through manipulation of MBON-11 activity, and we expressly do not draw a conclusion as to which of these regions is indeed involved in these reinforcing effects. One scenario is that silencing MBON-11 lifts inhibition from PPL1-01, promotes PPL1-01 activity and thus exerts a punishing effect (but see \[[@RSBL20190084C20]\]). Accordingly, the observation that activating MBON-11 has just a mild rewarding effect ([figure 2](#RSBL20190084F2){ref-type="fig"}) would suggest that spontaneous activity in PPL1-01 is moderate, and thus that silencing PPL1-01 would have less effect than activating it. Indeed, as previously reported, activating PPL1-01 is very strongly punishing (electronic supplementary material, figure S4B) \[[@RSBL20190084C14],[@RSBL20190084C19]\], whereas silencing it is of no measureable rewarding effect (electronic supplementary material, figure S4C) (see \[[@RSBL20190084C27]\] for a punishing effect of silencing the DAN of the γ3 compartment). This scenario would therefore suggest that targets other than PPL1-01 are responsible for the rewarding effect of activating MBON-11 (also see \[[@RSBL20190084C20]\]). Interestingly, the pathway from MBON-11 onto the glutamatergic MBON-01 neuron of the γ5 compartment and further from MBON-01 to the rewarding DANs of that compartment is critical for extinction learning after aversive training (\[[@RSBL20190084C26]\]; also see \[[@RSBL20190084C25]\]) (synonyms for MBON-01 are MBON-γ5β′2a and MB-M6). According to the scenario put forward in \[[@RSBL20190084C26], fig. 7E-F\], odours presented with MBON-11 silencing should lift inhibition from MBON-11 to MBON-01 and should thus drive the rewarding DANs of the γ5 compartment. This indirect, hetero-compartmental connection would thus support appetitive learning through MBON-11 silencing, whereas aversive learning would result for odours presented with MBON-11 activation---which is the *opposite* of what we report here! To reconcile this contradiction, consider that during second-order conditioning a stimulus X is first paired with primary reinforcement, and then X is presented together with a novel stimulus A in the absence of primary reinforcement. Whereas during AX training the effects of X as a reinforcement-predicting, second-order reinforcer will initially dominate, extended AX training will extinguish the X-with-reinforcement association. The above scenario would thus suggest that the opposing effects of second-order reinforcement and extinction learning, well known to practitioners of this paradigm, are related to homo- versus hetero-compartmental processes. We note that placing the behavioural effects of manipulating MBON-11 activity into an experimental psychology framework of secondary reinforcement processing also encompasses the effect labelled 'BGAM' (for blockade of MBON-γ1pedc-induced aversive memory) \[[@RSBL20190084C20], fig. 3B,C\], obtained by blocking synaptic output from MBON-11 (also see \[[@RSBL20190084C28]\]). Critically, the present framework suggests that silencing MBON-11 or preventing synaptic output from it leads to aversive learning about the odour paired with such treatment, whereas \[[@RSBL20190084C20], p. 569\] suggests that synaptic output from MBON-11 is necessary to prevent aversive learning about odours presented in an unpaired manner (for a discussion of paired and unpaired learning, see \[[@RSBL20190084C29]\]). We think that it is interesting that activity in a cell such as MBON-11 can be an analogue of second-order reinforcement, because this is the earliest site efferent to the memory trace in the presynaptic terminals of the mushroom body KCs for such an effect. This might inform the search for such analogues of secondary reinforcement in other species. It also raises the question of how much further down efferent pathways such analogues of second-order reinforcement can be observed, and indeed what the relation of action to valence is. Supplementary Material ====================== ###### Figure S1 Supplementary Material ====================== ###### Figure S2 Supplementary Material ====================== ###### Figure S3 Supplementary Material ====================== ###### Figure S4 Supplementary Material ====================== ###### Table S1 We thank R. Kittel, University of Leipzig, Germany and G. Nagel, University of Würzburg, Germany, for providing the effector strains for GtACR1 and ChR2-XXL expression, and R. D. V. Glasgow, Zaragoza, Spain, for language editing. Ethics {#s5} ====== All experiments comply with applicable law and ethics regulations. Data accessibility {#s6} ================== For data and statistical report, see the electronic supplementary material, table S1. Authors\' contributions {#s7} ======================= B.G. and C.K. conceived and coordinated the project; S.G. generated UAS-GtACR1; other authors collected and/or analysed the data. All authors contributed to the preparation of the manuscript by drafting or critically revising it, gave their final approval, and are accountable for its content. Competing interests {#s8} =================== We declare we have no competing interests. Funding {#s9} ======= This study was supported by Deutsche Forschungsgemeinschaft CRC 779-B11, GE1091/4-1 and FOR 2705 (to B.G.), as well as Leibniz Institute for Neurobiology, Magdeburg. [^1]: Electronic supplementary material is available online at <https://dx.doi.org/10.6084/m9.figshare.c.4534184>.

Introduction {#s1} ============ Alzheimer\'s disease (AD) is one leading cause of dementia and affects more than 35 million people in the world (Gregori et al., [@B20]; Hamilton et al., [@B21]). AD is a kind of neurodegenerative disorder and mainly characterized by extracellular senile plaques and intracellular neurofibrillary tangles. The high incidence and financial global burden of AD manifestations greatly reduce the quality of elderly life. However, the etiology of AD is complex and medicine therapy is still the mainstay for AD patients. Cholinesterase inhibitor and memantine (1-amino-3,5-dimethyl-adamantane) are often used for the conventional therapy of AD patients (Ahn et al., [@B1]; Dysken et al., [@B14]; Evans et al., [@B15]). However, all the medicine has significant side effects. Cholinesterase inhibitors cause the side effects, such as bradycardia (Leikin et al., [@B32]), hypotension (van Beek et al., [@B54]), and low intraocular pressure (Sawada et al., [@B52]). Memantine causes common adverse effects, such as confusion, dizziness, drowsiness, headache, insomnia and hallucinations, and some less common adverse effects including vomiting, anxiety, hypertonia, cystitis and libido (Rossom et al., [@B50]; Jain et al., [@B25]). Furthermore, it can induce reversible neurological impairment in sclerosis (an inflammatory disease in the brain and spinal cord) patients, resulting in the halt of a clinical trial (Villoslada et al., [@B56]). High cost of the medicine and serious unwanted adverse effects, have limited its utilization. Therefore, it is highly demanded to explore low-cost non-pharmaceutical method for AD patients. Brain exercises are widely known approaches to prevent dementia, increase neurogenesis and protect neurons (Gatz, [@B17]). Physicians often advise the elderly to engage in a mentally challenging activity to reduce dementia risk. The elderly spend time solving crossword puzzles, anagrams and figural logic puzzles, and may feel well if the activities are both challenging and successfully completed. More education has been found to be related to a lower incidence of AD. Typically, the risk of AD is 2--4 times higher in the persons with less education than those with more education (Qiu et al., [@B44]). On the other hand, the elderly participate in leisure activities, especially for mental stimulation, will also have a lower incidence of AD (Provencher et al., [@B42]). In addition, longitudinal studies have showed older adults have less dementia if they participate in intellectually challenging activity (Hultsch et al., [@B23]). The GO game, a kind of Chinese chesses, has been practiced as a brain activity for more than 5000 years. Playing the game involves many aspects of cognition and improves mental health (Kim et al., [@B27]). A GO game involves the changes associated with many cognitive functions, including learning, abstract reasoning, and self-control, which facilitate cognitive behavioral therapy (Lee et al., [@B30]). However, the functional role of a GO game in AD patients remains elusive. Furthermore, brain derived neurotrophic factor (BDNF) is often deficient in AD patients or animal AD models (Voineskos et al., [@B57]; Lee et al., [@B31]; Boiocchi et al., [@B8]). BDNF protects cerebral cortical neurons against beta-amyloid 25--35 (the major AD toxic peptide) and inhibits beta-amyloid 25--35 aggregation in the brain (Xiao et al., [@B59]; Zhang et al., [@B60]). A GO game may ameliorate AD patients by affecting the levels of BDNF since BDNF plays a critical role in learning and memory functions (Fan et al., [@B16]). Thus, we explored the functions of a GO game in AD patients and measured BDNF levels in AD patients before and after the study. Materials and methods {#s2} ===================== Participants ------------ All the protocols were approved by the Institutional Ethics Committee from the Fourth Affiliated Hospital of China Medical University and conducted in accordance with the Declaration of Helsinki (Palacios, [@B39]). All subjects were Han Chinese from Shenyang city (China). An informed consent was obtained either from each subject or from his or her guardian. All AD patients were diagnosed with NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer\'s Disease and Related Disorders Association) criteria according to a previous report (Tamaoka, [@B53]). All familial cases of AD were excluded in this study. The patients were excluded if they met any of the following criteria: (1) a history of suicidal behavior; (2) substance abuse; (3) verbal communication difficulty; (4) a family history of AD; (5) brain injury or other brain disorders; (6) disliking playing a GO game. From July 7th, 2014 to January 8th, 2015, a total of 147 AD patients were recruited at the Fourth Affiliated Hospital of China Medical University (Shenyang, China). Study design ------------ After scanning all AD patients, 147 patients were eligible to enter designated 6-month follow-up (Figure [1](#F1){ref-type="fig"}). A randomized and controlled trial was conducted using three groups: control (without GO-game intervention, *n* = 49), short-time GO-game intervention (SGGI, 1 h daily, *n* = 49) and long-time GO-game intervention (LGGI, 2 h daily, *n* = 49) groups. The three groups have the similar gender ratio, age, BMI (body mass index, the ratio of weight (kg) to height in meters squared), years of education, alcohol drinking, cigarette smoking, and diabetes. Each of them has a spouse and there is a good mutual relation. Among the AD patients, some subjects have complex disorders such as comorbid diabetes, depression, and overeating (Brownley et al., [@B9]). {#F1} To avoid the interference of the experiences for playing a GO game, none of them had ever played a GO game before the study. Before game training, all subjects from GO-game groups were invited to learn the rules for playing the game from websites, such as <http://en.wikipedia.org/wiki/GO_(game)>. All participants were trained by a game player in the same club. After mastering the basic rules, each person from GO-game groups can finish the game as Figure [2](#F2){ref-type="fig"} showed. In GO groups, two partners were combined randomly. If one patient could not find a partner, he or she watched the game played by other participants. Otherwise, he or she would play with one of GO-game staff. {#F2} Neuropsychological tests ------------------------ According to Diagnostic and Statistical Manual of Mental Disorders (DSM), all the cases were diagnosed by licensed neurological physicians including neurological examination, blood tests, and neuroimaging data (computed tomography or magnetic resonance imaging). Clinical Dementia Rating (CDR) scale is often used to detect memory and executive functions (Inoue et al., [@B24]). Furthermore, CDR is used not only to detect memory impairment but also to quantify dementia severity. CDR is also used to assess domains of cognition (memory, orientation and problem solving) and the domains of functions (community affairs, home, and hobbies and personal care) (Cedarbaum et al., [@B11]). Furthermore, the Mini Mental State Examination (MMSE) has been widely used for testing memory problems. Clinicians often diagnose dementia and assess its progression and severity by MMSE scores. Depression is a frequent condition for AD patients, so the depressive severity was examined based on the Montgomery and Asberg Depression Rating Scale (MADRS) (Alonzo et al., [@B2]). There are 10 items for MADRS scoring from 0 to 60. The severities of depression were regarded as severe, moderate and mild based on different cut-off scores (Birnbaum et al., [@B7]). Furthermore, all participants was evaluated with the Kimberley Indigenous Cognitive Assessment of Depression (KICA-Dep) (Salvatore et al., [@B51]). There are 11 items for KICA-Dep, each of which is assigned as from "never" to "sometimes," "a lot" and "all the time" based on a frequency scale. Some AD patients suffer from anxiety disorder (Ramakers et al., [@B46]; Mormont et al., [@B37]) and anxiety was measured based on Hospital Anxiety and Depression Scale (HADS) (Hinz et al., [@B22]). The scores of HADS are presented as from 0 to 21, and higher scores stand for more severe anxiety. General functioning was evaluated by a Global Assessment of Functioning (GAF) scale (Mello et al., [@B36]). Life quality was assessed using RAND-36 (Mattila et al., [@B34]), which counts for wellbeing and functioning in eight dimensions. Alexithymia is a psychological problem and was evaluated using the Toronto Alexithymia Scale--20 (TAS-20) (Melin et al., [@B35]). ELISA analysis of BDNF ---------------------- To determine serum levels of BDNF, 5 ml of blood samples were taken (8:00--11:00 a.m.) by venipuncture into a free-anticoagulant vacuum tube at before and after 6-month follow-up. The blood samples were centrifuged at 3000 g for 10 min, and serum was isolated and kept at −80°C until next step. The serum levels of BDNF were measured using a human BDNF ELISA kit (Adipo Bioscience, Santa Clara, CA, USA). Statistical analyses -------------------- All statistical analyses were performed with a SPSS 20.0 package (IBM China Company Limited, Beijing, China). The AD patients were compared by *t*-tests between GO groups and the control group without a GO intervention. One-Way ANOVA was used to explore the effect of a GO game on the levels of BDNF in AD patients. Spearman\'s rank correlation coefficient was conducted to identify the correlation between the depressive severity (Montgomery-Asberg Depression Rating Scale or RAND-36) and BDNF levels. There were statistically significant differences if *P* \< 0.05. Results {#s3} ======= The baseline characters of AD patients -------------------------------------- All 147 AD patients had medical histories consistent with AD and took neurological examinations (Supplemental Table 1). Controls were matched well to AD patients by gender and age at onset (*P* \> 0.05). A previous report has found that BMI is associated with the risk of AD (Besser et al., [@B6]). To avoid the effects of BMI on present results, all patients were selected to guarantee that there were no statistically significant differences among three groups (*P* \> 0.05) (Supplemental Table 1). Diabetes is an increasing epidemic and affects millions of the elderly worldwide. The chronic disease also affects brain function and contributes to an AD risk (Butterfield et al., [@B10]). Similarly, high blood pressure and coronary heart disease are also contributing factors for the development of AD (Qiu et al., [@B45]; Lattanzi et al., [@B29]). Furthermore, cigarette smoking has been reported to be also associated with incipient AD (Chang et al., [@B12]). Alcohol drinking can impair brain function, and cause dementia and geriatric cognitive disorders (Wiscott et al., [@B58]). The patients with long-period education have better cognitive functions than the AD patients with short-period education (Pradier et al., [@B41]). Furthermore, the anterior temporal lobes play an important role in melody recognition, and that music can significantly affect AD patients (Johnson et al., [@B26]). To avoid these interfering factors, all the elements were carefully examined to exclude their effects on final results (Supplemental Table 1). In neuropsychological tests, all AD patients had MMSE scores less than 26 and the CDR scale of 0.5 or over 0.5 (Supplemental Table 1). The results indicated that all AD subjects had clinical dementia. Similarly, all AD patients had the similar psychiatric test values and the prevalence of KICA-dep items in AD patients among LGGI, SGGI, and control groups (*P* \> 0.05) (Supplemental Table 1 and Table [1](#T1){ref-type="table"}). Therefore, the baseline characters of patients were basically similar among three groups. ###### **The comparison for Prevalence of KICA-Dep items among AD patients**. **KICA-Dep Items** **Response** **Control group** **SGGI group** **LGGI group** ------------------------------------------------------------------ -------------- ------------------- ---------------- ---------------- ---------- ---------- ------- ---------- ---------- ------- Felt down, sad, no good Never 5(10.2) 5(11.3) 0.587 5(10.2) 10(20.8) 0.001 5(10.2) 12(25.0) 0.001 Sometimes 30(61.2) 26(59.1) 0.182 29(59.2) 30(62.5) 0.183 28(57.1) 31(64.6) 0.043 A lot 5(10.2) 5(11.3) 0.210 5(10.2) 3(6.3) 0.002 6(12.2) 2(4.2) 0.001 All the same 9(18.4) 8(18.2) 0.859 10(20.4) 5(10.4) 0.001 10(20.4) 2(4.2) 0.001 Felt like doing things that you usually like doing Never 5(10.2) 5(11.4) 0.066 6(12.2) 4(8.3) 0.006 5(10.2) 4(8.3) 0.004 Sometimes 25(51.0) 22(50.0) 0.393 25(51.0) 20(41.6) 0.002 25(51.0) 19(39.6) 0.002 A lot 6(12.2) 5(11.4) 0.299 6(12.2) 7(14.6) 0.015 6(12.2) 8(16.7) 0.005 All the same 13(26.5) 12(27.3) 0.180 12(24.5) 17(35.4) 0.001 13(26.5) 17(35.4) 0.001 Had trouble going to sleep, staying asleep, or sleeping too much Never 5(10.2) 5(11.4) 0.804 5(10.2) 18(37.5) 0.001 5(10.2) 20(41.7) 0.001 Sometimes 30(61.2) 26(59.1) 0.185 29(59.2) 20(41.7) 0.001 30(61.2) 18(37.5) 0.001 A lot 4(8.2) 3(6.1) 0.119 4(8.2) 3(6.3) 0.048 5(10.2) 4(8.2) 0.008 All the same 10(20.4) 6(22.7) 0.083 11(22.4) 7(14.6) 0.001 9(18.4) 6(12.5) 0.001 Felt tired or slack, and had no energy Never 7(15) 10(20.0) 0.854 13(16.3) 21(27.5) 0.085 13(16.3) 21(27.5) 0.085 Sometimes 24(50) 28(57.5) 0.341 39(48.7) 37(49.3) 0.874 39(48.7) 37(49.3) 0.874 A lot 12(25) 8(17.5) 0.246 19(23.8) 14(18.8) 0.440 19(23.8) 14(18.8) 0.440 All the same 6(10) 2(5.0) 0.230 9(11.2) 3(3.8) 0.072 9(11.2) 3(3.8) 0.072 Eating too much or eating only a little bit. Never 5(10.2) 4(9.1) 0.197 5(10.2) 15(31.3) 0.001 5(10.2) 16(33.3) 0.001 Sometimes 4(8.2) 4(9.1) 0.133 5(10.2) 25(52.1) 0.001 4(8.2) 25(52.1) 0.001 A lot 16(32.7) 14(31.8) 0.210 15(30.6) 4(8.3) 0.001 16(32.7) 4(8.3) 0.001 All the same 24(49) 22(50) 0.512 24(49) 4(8.3) 0.001 24(49) 3(6.3) 0.001 Felt bad or shamed that you let yourself or your family down. Never 8(16.3) 7(15.9) 0.514 7(14.3) 17(35.4) 0.001 7(14.3) 18(37.5) 0.004 Sometimes 4(8.2) 4(9.1) 0.614 5(10.2) 15(31.3) 0.001 6(12.2) 16(33.3) 0.001 A lot 17(34.7) 15(34.1) 0.693 17(34.7) 8(16.7) 0.001 16(32.7) 7(14.6) 0.001 All the same 20(40.8) 18(40.9) 0.875 20(40.8) 8(16.7) 0.001 20(40.8) 7(14.6) 0.001 Had trouble paying attention or concentrating on things Never 6(12.2) 5(11.4) 0.449 7(14.3) 16(33.3) 0.001 7(14.3) 17(35.4) 0.001 Sometimes 10(20.4) 9(20.5) 0.526 11(22.4) 21(43.8) 0.001 11(22.4) 22(45.8) 0.001 A lot 14(28.6) 13(29.5) 0.514 13(26.5) 6(12.5) 0.001 14(28.6) 4(8.3) 0.001 All the same 19(38.8) 17(38.6) 0.948 18(36.7) 5(10.4) 0.001 17(34.7) 5(10.4) 0.001 Been told that you are speaking or moving too slowly or fast Never 5(10.2) 4(9.1) 0.194 5(10.2) 15(31.3) 0.001 5(10.2) 15(31.3) 0.001 Sometimes 6(12.2) 5(11.4) 0.445 6(12.2) 18(37.5) 0.001 6(12.2) 18(37.5) 0.001 A lot 12(24.5) 11(25) 0.551 12(24.5) 7(14.6) 0.001 13(26.5) 8(16.7) 0.001 All the same 26(53.1) 24(54.5) 0.303 26(53.1) 8(16.7) 0.001 25(51) 7(14.6) 0.001 Had thoughts that you would be better off dead Never 3(6.1) 3(6.8) 0.204 4(8.2) 17(35.4) 0.001 4(8.2) 18(37.5) 0.001 Sometimes 6(12.2) 5(11.4) 0.752 6(12.2) 16(33.3) 0.001 6(12.2) 17(35.4) 0.001 A lot 12(24.5) 11(25) 0.822 11(22.4) 9(18.8) 0.001 11(22.4) 7(14.6) 0.001 All the same 28(57.1) 25(56.8) 0.730 28(57.1) 6(12.5) 0.001 28(57.1) 6(12.5) 0.001 Thought of hurting yourself Never 4(8.2) 4(9.1) 0.105 4(8.2) 14(29.2) 0.001 4(8.2) 15(31.3) 0.001 Sometimes 4(8.2) 4(9.1) 0.138 4(8.2) 19(39.6) 0.001 4(8.2) 19(39.6) 0.001 A lot 14(28.6) 12(27.3) 0.357 15(30.6) 7(14.6) 0.001 15(30.6) 7(14.6) 0.001 All the same 27(55.1) 24(54.5) 0.548 26(53.1) 8(16.7) 0.001 26(53.1) 7(14.6) 0.001 Felt angry Never 2(4.1) 2(4.5) 0.627 2(4.1) 18(37.5) 0.001 3(6.1) 17(35.4) 0.001 Sometimes 3(6.1) 3(6.8) 0.322 3(6.1) 17(35.4) 0.001 2(4.1) 18(37.5) 0.001 A lot 24(49) 22(50) 0.512 23(46.9) 7(14.6) 0.001 23(46.9) 6(12.5) 0.001 All the same 20(40.8) 17(38.6) 0.142 21(42.9) 6(12.5) 0.001 21(42.9) 7(14.6) 0.001 *n = 49 in each group*. Statistical analyses for the outcomes at 6-month follow-up ---------------------------------------------------------- Statistical analyses showed that a GO-game intervention significantly decreased the mean score of MADRS of 4.72 (95% CI, 0.69--9.12) compared with those in a control group after 6-month follow-up (*P* \< 0.05) (Supplemental Table 2). In the similar way, a GO-game intervention also reduced mean score of HADS of 1.75 (95% CI, 0.17--3.68) when compared with those from controls (*P* \< 0.05) (Supplemental Table 2). In contrast, a GO-game intervention significantly increased the mean score of GAF of 4.95 (95% CI, −1.37--9.18) and RAND-36 of 4.61 (95% CI, −2.75--11.32) when compared with a control group (*P* \< 0.05) (Supplemental Table 2). For TAS-20, there was no statistically significant difference in three groups (*P* \> 0.05) (Supplemental Table 2). The results may be caused by alexithymic personality traits, which are closely related with gender, advanced age, educational level and social intelligence (Koelkebeck et al., [@B28]). A GO game improves the symptoms of AD ------------------------------------- The analysis for KICA-Dep items showed that GO game relieved the symptoms of AD patients compared with those from a control group (Table [1](#T1){ref-type="table"}). In the 11 items of KICA-Dep, a control group could not attenuate the symptoms of AD patients and there was no statistically significant difference for most items (*P* \> 0.05) except of two ones (Table [1](#T1){ref-type="table"}). Comparatively, a GO game ameliorated the symptoms of depression and there were statistically significant differences for most items (*P* \< 0.05) except of two items (Table [1](#T1){ref-type="table"}). After a 6-month GO-game intervention, the severity of AD was decreased significantly (*P* \< 0.05) (Table [1](#T1){ref-type="table"}). The levels of BDNF are negatively related with stratification of CDR -------------------------------------------------------------------- When AD severity was stratified by CDR scales at onset (χ^2^ = 6.8, *P* = 0.03), significant differences of BDNF levels were observed in lower (CDR 0.5), middle (CDR 1), and higher CDR (CDR 2^+^). There were statistically significant differences for BDNF levels between CDR 0.5 and 2^+^ (*P* \< 0.05). BDNF levels were negatively related with CDR scales, suggested that low levels of BDNF were related with AD risk since the BDNF levels were higher in lower CDR scores compared to those in higher CDR scores (χ^2^ = 214, *P* \< 0.01) while CDR scores have been used to evaluate the severity of AD (Clark et al., [@B13]). Association between an AD episode and the protein levels of BDNF ---------------------------------------------------------------- Elevated levels of BDNF were related with a decrease in MADRS scores and an increase in RAND-36 scales (*P* \< 0.01). There was a strong negative relationship between the severity of AD and the levels of BDNF because the rho values were less than −0.5 based on the calculation of Spearman\'s rank correlation coefficient. A GO game enhances the serum protein levels of BDNF in AD patients ------------------------------------------------------------------ The study consisted of 147 AD patients before and after 6-month follow-up. The socio-demographic was list in Supplemental Table 1 and balanced based on age, gender, educational levels, life habits and medicine therapy. The results showed that BDNF levels were similar in three groups before GO game intervention. After 6-month intervention, serum BDNF levels were significantly higher in subjects receiving SGGI and LGGI (respectively, 24.02 ± 7.16 and 28.88 ± 4.12 ng/ml, *t* = 0.345, *P* = 0.051) when compared with controls (17.28 ± 7.75 ng/ml) (*t* = 3.423, *P* = 0.001) (Figure [3](#F3){ref-type="fig"}). All these results implied that a GO game promotes the production of BDNF. {#F3} Discussion {#s4} ========== A GO game is very popular in China, Japan, and Korean, and can be widely played. More importantly, a GO game has been proved to be very interesting for its rich strategies with simple rules and attracts many elderly adults. The game is more suitable to be developed to prevent the progression of AD. We investigated the effects of the game on AD and related molecular mechanisms. On the other hand, physical activities can certainly delay the loss of autonomy in AD patients and are also useful strategies for delaying the complications of AD (Rolland et al., [@B49]). During the recruitment, some AD patients like physical exercises but they dislike playing GO game. In the similar cases, some AD patients like playing the game but they dislike physical exercises. Only a few AD patients like playing a GO game and physical exercises. The sample size will be a problem for considering both physical exercises and a GO game. Furthermore, the effects of physical exercises on AD patients have been reported. Thus, we only consider the effects of a GO game on AD here. The patients will be excluded if they dislike playing the game. Importantly, a GO game showed effective anti-depression results with few side effects. Meanwhile, we interpreted the possible outcome by using GO-game-specific and computational methods with a particular motivation to detect therapeutically-relevant phenomena. The strength of the paper is its technique in clinical practice. Furthermore, each person in GO groups could play the game well with basic rules. On the other hand, the combination of various techniques was utilized. Brain imaging and self-ratings, computational game analyses and psychiatric assessment was undertaken, resulting in complementary and myriad measures access to various underlying determinants of an intervention. Some brain imaging techniques, such as fMRI and PET, can provide distinct spatial information, but which cannot provide time and event-related correlations. Depression has been reported to be correlated with a hypo-activation of left brain activity and a hypo-inactivation of right brain activity (Richieri et al., [@B48]), which can be ameliorated with the time going after playing the game. MADRS is one of the most effective tools for detecting the depression in AD patients and higher MADRS score suggests severe depression (Müller-Thomsen et al., [@B38]). We used the MADRS to measure the changes of depression severity. The results showed that a GO game significantly decreased mean score of MADRS of 4.72 (95% CI, 0.69--9.12) when compared with controls after 6-month follow-up (*P* \< 0.05, Supplemental Table 2). HADS is also an important tool to detect the depression of AD (Pietrzak et al., [@B40]), so we also measured the HADS scores here. In the similar results, a GO game decreased the mean score of HADS of 1.75 (95% CI, 0.17--3.68) when compared with controls in the same period (*P* \< 0.05, Supplemental Table 2). Furthermore, KICA-Dep has been reported to very useful for detecting the depressive symptoms in AD patients. The results indicated that a GO game improved 9 of 11 items of KICA-Dep compared with controls after 6-month follow-up (Table [1](#T1){ref-type="table"}). On the other hand, GAF has been used to evaluate the overall improved quality of life in AD patients (Lu et al., [@B33]). Thus, we measured GAF and found that a GO game increased the mean score of GAF of 4.95 (95% CI, −1.37--9.18) and RAND-36 of 4.61 (95% CI, −2.75--11.32) when compared with those in a control group after 6-month follow-up (*P* \< 0.05, Supplemental Table 2). Although a GO game showed effective results for ameliorating the symptoms of AD, the molecular mechanism remains unknown. Previous work demonstrates that AD patients have reduced levels of BDNF in the brain and serum. Additionally, animal-based studies also showed protective effects of BDNF against Amyloid beta-protein (Abeta)-induced neurotoxicity (Psotta et al., [@B43]). Abeta1-42 exhibits neurotoxicity and induces neural death. Abeta1-40 can inhibit the fibril formation of Abeta1-42 (Zou et al., [@B61]). In contrast, Abeta1-16, Abeta25-35, and Abeta40-1 prevent neither the fibril formation of Abeta1-42 nor Abeta1-42-induced neural death. Here, we measured serum levels of BDNF in AD patients and found that a GO game increased serum BDNF levels in SGGI and LGGI groups (24.02 ± 7.16 and 28.88 ± 4.12 ng/ml respectively, *t* = 0.345, *P* = 0.051) compared those in controls (17.28 ± 7.75 ng/ml) (*t* = 3.423, *P* = 0.001) after 6-month follow-up. Furthermore, the serum levels of BDNF showed a negative relation with MADRS and a positive relation with RAND-36 scores (*P* \< 0.01). A GO game ameliorates AD patients by up-regulating BDNF levels. Thus, A GO game is effective to prevent the development of AD. Alexithymia is considered for its defects in regulating feelings (Goerlich et al., [@B19]) and typically co-occurs with depression (Gilbert et al., [@B18]). However, alexithymia is a complex brain disorder with a cluster of deficits in the recognition (Reker et al., [@B47]). Here we found that a GO game was not an effective way for the treatment of alexithymia compared with that from a control group (*P* \> 0.05) (Table [1](#T1){ref-type="table"}). In any way, a GO game can improve the life quality in other aspects (*P* \< 0.05) (Supplemental Table 2). In addition, evoking and dealing with emotions is usually related with a GO game, which fit the treatment of emotional disorders just like depression. Although all AD patients wanted to learn and play a GO game before the study, someone kept active exercises and a few kept passive exercises for the game with time going. We found that most winners were active for playing the game while a few losers were passive for playing the game. We would let the game trainers play with these passive losers and let them win on purpose. Most of them could become active again for playing the game finally. In order to gain an insight into the effectiveness of the game, we also aimed to develop novel analytical methods for clinical purposes, such as kinematics, electromyographic and electroencephalographic (EEG) data (Barlaam et al., [@B5]). The new concept is also needed to understand the mechanisms by studying symptoms, behaviors, or biomarkers, which are different to traditional classification for the mental disorders (Badcock and Hugdahl, [@B3]). Meanwhile, the study will provide valuable information for the therapy of depression. Limitations {#s5} =========== A GO game affects the progression of depression but the molecular mechanism remains unknown. Much evidence indicates that chronic stress and low levels of BNDF are the key causes of depression (Banerjee et al., [@B4]; Vásquez et al., [@B55]). However, the clinical relevance of a GO game intervention and the levels of BDNF remain unknown. Here, we firstly reported that a GO game ameliorated the depression by affecting the serum levels of BDNF. We found the relationships between SGGI or LGGI and levels of BDNF. The results showed that lower concentration of BDNF might be specific for the depressive state in AD patients. BDNF is a potential biomarker for adjuvant diagnosis and therapy of an AD episode. However, further studies are necessary to define a functional role of BDNF in preventing the development of depression. Certainly, the present trial still has its limitations in a statistical sense. The sample size was sufficient to detect an effect in the primary outcome at 6-months, but not at longer period although we found the effect tended to persist. Second, the present trial was the GO game intervention. It remained unknown whether a long-term GO game was better than a short-term GO game. At least, there are no reports showing such indication yet. However, we still believe that a GO game intervention may be more suitable for a few populations only received medicine therapy. Finally, a major problem with the design is that the control group did not receive any treatment. Therefore, it is not possible to know whether the changes reported are because the treatment groups received attention from and interacted with colleagues and staff or because of the game itself. A "placebo" procedure will be necessary to enable solid conclusions. Unfortunately, the "placebo" group is hard to be designed because no game can be matched GO game well according our experiences. Being interesting is very important to play a game. Gomoku, also called Five in a Row, can be played on a GO board. From May 6th, 2012 to August 12th, 2013, a total of 50 AD patients were recruited. All the participants did not have the experiences of playing Gomoku and showed interests for the game before the study. However, more than 50% patients lost their passions for the game after 2-month follow-up and the work was unavoidably stopped. Fortunately, less than 10% patients lost enthusiasm for GO game even after 6-month. Furthermore, these patients still could show enthusiasm again for GO game when they won in subsequent matches, which were designed on purpose by the GO game staff. We think that a GO game shows its beneficial functions on AD patients since the enthusiasm still can be maintained after 6-month follow-up. Clinical and mechanistic implications {#s6} ===================================== A GO game can improve the life quality of AD patients. None complained of playing a GO game and no routine neurological exam could identify any adverse effect. Playing a GO game should be developed a public activity since the game intervention can reduce the symptoms of AD. Its mechanistic implication may be that many strategies involve in the play and the number of possible games is vast (there are about 10^120^ possible results for playing the game), although the rules of game are very simple. Given the substantial changes in the game, the game attracts an ever-growing older population. Playing a simple and inexpensive game can improve the cognitive functions of AD patients. Millions of AD patients are looking for a technologically-oriented game to enhance their cognitive skills, which can prevent the development of AD. In sum, a GO game is feasible and effective to improve life quality of AD patients by reducing their depression. A GO game intervention reduces AD severity by increasing the level of BDNF. Thus, a GO game should be developed as a new method for the therapy of AD. 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. We are very grateful to all anonymous reviewers for their critical and strategic comments, which have significantly improved the quality of the paper. Supplementary material {#s7} ====================== The Supplementary Material for this article can be found online at: <http://journal.frontiersin.org/article/10.3389/fnagi.2015.00163> ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Vadim Fraifeld, Ben Gurion University of the Negev, Israel [^2]: Reviewed by: Changiz Geula, Northwestern University, USA; Junming Wang, University of Mississippi Medical Center, USA

All relevant data are within the paper. Introduction {#sec005} ============ Thrombotic thrombocytopenic purpura (TTP) and hemolytic uremic syndrome (HUS) are rare diseases characterized by microangiopathic hemolytic anemia (MAHA) and thrombocytopenia. HUS is characterized by three clinical signs: kidney failure, hemolytic anemia, and thrombocytopenia while TTP has been defined by a diagnostic pentad of thrombocytopenia, hemolytic anemia, fever, neurologic changes and renal compromise \[[@pone.0127744.ref001]\]. However, the pentad associated with TTP is today mostly academic since most patients do not present with all symptoms at once \[[@pone.0127744.ref002], [@pone.0127744.ref003]\]. At the center of the pathology initial reports listed deficiency in the von Willebrand Factor (vWF) metalloprotease ADAMTS13 as a marker of disease and as a potential factor that may aid in the differentiation of TTP from HUS \[[@pone.0127744.ref004], [@pone.0127744.ref005]\]. However, even though ADAMTS13 differences were originally described, not all patients presenting with the clinical diagnosis of TTP have a deficiency in this metalloprotease leading to the suggestion that it lacks importance in the diagnostic workup of these disease entities \[[@pone.0127744.ref006]\], and as predictor of clinical response to therapy which should be based mostly on clinical criteria \[[@pone.0127744.ref007]\]. Ever since the results of the clinical trial by the Canadian Apheresis Study Group showed the benefits of therapeutic plasma exchange (TPE) vs. simple high volume plasma infusion, the survival of TTP patients has improved significantly from the high mortality seen a quarter of a century ago \[[@pone.0127744.ref008]\], and has led to a seven fold improvement in outcomes including those TTP patients with severe renal compromise \[[@pone.0127744.ref009], [@pone.0127744.ref010]\]. In another study evaluating the efficacy of TPE vs. plasma infusion, the former was shown to be more beneficial in patients receiving disease-relief by plasma infusion but who did not respond well to the increased volume leading to the use of TPE in this subgroup \[[@pone.0127744.ref011]\]. Complications can occur in response to TPE, and the disease can still be fatal and patients can relapse despite aggressive therapy. However, reports for the last two decades have shown markedly differing survival and relapse rates during the critical treatment period with TPE. In the present study we determined the 30-day mortality rate and relapse of TTP-HUS patients at a large tertiary academic medical center and compared these to published literature reporting similar data. Materials and Methods {#sec006} ===================== Study design and data collection {#sec007} -------------------------------- All patient information was anonymized and de-identified prior to analysis. This retrospective study evaluated the 30-day survival and relapse rates in patients with suspected TTP-HUS at University Hospitals Case Medical Center (UHCMC), a tertiary academic medical center, from January 1^st^ 2008 to December 31^st^ 2012. All individuals given a presumptive diagnosis of TTP-HUS by the hematology clinical service, referred to the Apheresis Center (AC) at UHCMC, and who were transferred to be treated with TPE by the AC were included in the study. None of the patients received TPE at transferring institutions. Patients who had undergone apheresis for other clinical indications were excluded. The inpatient and outpatient records of all patients were identified through AC records and the hospital's electronic medical record. Two reviewers using standardized forms that we developed, independently collected and verified the accuracy of the data gathered which included: demographic information, duration of TPE treatment, number of TPE procedures, volume replaced, complete blood count (platelet count, hemoglobin/ hematocrit), creatinine at presentation and end of therapy, ADAMTS13 activity, neurological symptoms and/ or fever, 30 day-mortality and relapse, and presence of other co-morbidities. In order to ascertain and compare disease severity, a Clinical Severity Score (CSS) that has been previously described was used \[[@pone.0127744.ref012]\]. Briefly, the four parameters used in clinically scoring patients were: neurologic symptoms, renal insufficiency, platelet counts and hemoglobin concentration, if available at the time of presentation. Score ranged from 0 to 8 points. Clinical diagnostic criteria followed recommendations previously outlined \[[@pone.0127744.ref013]\]. Patients were grouped into two categories: idiopathic if they had no other illness that is recognized to be associated with the symptoms at presentation; or secondary TTP-HUS if the patients had an identified trigger leading to the symptoms (e.g. stem cell transplantation, pregnancy/ postpartum, drugs, infections, autoimmune diseases, or malignancy-associated). The study was approved by the Institutional Review Board of UHCMC, Cleveland, OH. Patient outcome definitions {#sec008} --------------------------- Remission, response and relapse were defined as previously described \[[@pone.0127744.ref014], [@pone.0127744.ref015]\]. Briefly, remission was defined as an increase in platelet counts within a 30-day period after completion of TPE. Complete response was defined as a platelet count greater than 100 x 10^9^/L for two consecutive days (used for data comparison since three of four studies found in literature review use this platelet count to define recovery \[[@pone.0127744.ref012], [@pone.0127744.ref014], [@pone.0127744.ref016]\]); partial response as a platelet count in the 50--100 x10^9^/L range; and no response as platelet counts below 50 x10^9^/L. Survival was defined as achievement of remission. Mortality from TTP was defined as occurring within 30-days of stopping TPE. Relapse was defined as a decrease in platelet count below 100x10^9^/L, consistent with TTP after achievement of remission. All patients received corticosteroids at different times during their clinical presentations. Therapeutic Plasma Exchanges {#sec009} ---------------------------- All procedures were performed as previously described \[[@pone.0127744.ref017]\]. Briefly, all patients were initiated on daily TPE as soon as the presumed diagnosis of TTP-HUS was clinically made. TPE was performed daily with 1--1.5 plasma volume exchange using the COBE Spectra Apheresis System (Terumo BCT, Lakewood, CO). When indicated, patients were transfused to a hematocrit of ≥ 23% to increase the efficiency of TPE. Pre- and post-TPE fibrinogen, pro-thrombin time, partial thrombin time, ionized calcium and daily CBC were obtained. Patients were pre-medicated with 650 mg acetaminophen and 25 mg of diphenhydramine prior to TPE. All procedures used ABO-type-specific fresh-frozen plasma as replacement fluid and citrate dextrose-A solution as anticoagulation. ADAMTS13 activity {#sec010} ----------------- Forty one out of 48 (85.4%) patients had a sample collected for baseline ADAMTS13 activity prior to initiation of TPE but activity results were not available at the time of TPE initiation. ADAMTS13 activity assay was performed at the Blood Center of Wisconsin (BCW) (Milwaukee, WI) by the fluorescence resonance energy transfer (FRET) assay for ADAMTS13 activity as previously described \[[@pone.0127744.ref018]\]. When appropriate, metalloprotease inhibitor level was also measured by BCW. Literature search and study selection {#sec011} ------------------------------------- A comprehensive literature search strategy was conducted to identify potential articles published in English on PUBMED from 1970 to 2013. Following search keyword combinations were used: TTP-HUS, mortality/outcome, and relapse. Our search algorithm is shown in [Fig 1](#pone.0127744.g001){ref-type="fig"}. Selected articles were examined by two reviewers independently. Inclusion criteria were: 1) Patients with initial diagnosis of TTP-HUS, 2) 30-days mortality rate, and 3) relapse rate reported. {#pone.0127744.g001} Statistical Analysis {#sec012} -------------------- All statistics were performed using Prism 6 (GraphPad Software Inc., La Jolla, CA). Since the number of studies we found in the literature review was small and we did not have all of their raw numerical data, we were unable to perform a meta-analysis so we limited our analysis to comparisons of their results to ours. Mortality and relapse rates comparisons with similar combined data from five relevant articles were performed using a two-tailed Fisher test with a *p*\< 0.05 set for significance. Results {#sec013} ======= UHCMC patients {#sec014} -------------- A total of 59 patients were analyzed; 11 patients had TPE discontinued as requested by the clinical team due to other diagnoses that explained these patients' presentations. These patients were eventually diagnosed with disseminated intravascular coagulation, acute myeloid leukemia with marrow involvement, scleroderma, idiopathic thrombocytopenic purpura, malignant hypertension and sepsis/ bacteremia. Patients responded to treatment modalities which addressed each of these diagnoses after discontinuation of TPE. The remaining 48 patients were diagnosed with TTP-HUS and received daily TPEs. The demographic characteristics of these patients at diagnosis are summarized in [Table 1](#pone.0127744.t001){ref-type="table"}. Idiopathic TTP-HUS was diagnosed in 36 patients (75%) and secondary causes in 12 patients (25%). Thirty-three (69%) of patients were female, mean age of 49 (range 16--85); while 15 (31%) patients were males with mean age of 47.5 (range 28--76). There were a number of co-morbidities in our patient cohort which were as follows: two patients were pregnant/ postpartum (4%); 3 patients had a drug-associated presentation (6%); 2 patients had autoimmune disease (4%); 1 patient had a systemic infection (2%); 4 patients had malignancy/ metastasis (8%); and 1 patient had received a renal allograft (2%). Thirty-day mortality among TTP-HUS patients was 1/48 patients (2%). However, overall mortality for our patient cohort was 4/59 (6.7%) since 3 additional patients died of causes unrelated to TTP-HUS: one of these patients died due to complications of engraftment failure and graft vs. host disease (outside 30 day study period); a second died of multi-organ failure due to sepsis and bacteremia in the setting of disseminated terminal Burkitt's lymphoma complicated by renal allograft rejection; and a third died of heart failure due to worsening ejection fraction and heart disease. Only one of the four patients who died in our cohort had an autopsy with findings that were consistent with a TTP presentation \[[@pone.0127744.ref019]\] and had ADAMTS13 deficiency (\<5%). One male patient was treated with two cycles of TPE 3 months apart due to relapse and a second male was treated three times with TPE within a year from initial diagnosis (data not shown). 10.1371/journal.pone.0127744.t001 ###### UHCMC study patients' characteristics at initial presentation with TTP-HUS. {#pone.0127744.t001g} Mean or number (Percent) Range ---------------------------------- -------------------------- ------------------------------------------------- Age 49 16--85 Female 33 (69) Male 15 (31) Hemoglobin (g/dl) 9.0 2.5--13.4 Platelet count (x 10^9^/L) 4.6 5--307.000[\*](#t001fn001){ref-type="table-fn"} Serum creatinine (mg/dL 2.6 0.5--11.3 Idiopathic 35 (74) Pregnancy/Postpartum 2(4) Drug associated 3 (6) Autoimmune disorder (SLE and MG) 2 (4) Systemic infection 1 (2) Systemic Malignancy 4 (8) Renal transplant 1 (2) \*Patient with platelet count of 307,000 x 10^9^/Lwas diagnosed with HUS due to Factor H mutation/ deficiency. Patients' laboratory findings {#sec015} ----------------------------- As shown in [Table 2](#pone.0127744.t002){ref-type="table"}, mean platelet count at presentation was 46 x 10^9^/ L (range 5 x 10^9^/ L---307 x 10^9^/ L). Mean hemoglobin was 9.0 g/dl (range 2.5--13.4 g/dl); mean creatinine concentration was 2.6 mg/dL (range 0.5--11.3 mg/dL). Mean CSS for our cohort was 4. Of note, baseline values and clinical severity score at our institution may not reflect the values/score at initial presentation because a few patients were transferred to our institution after steroid therapy was initiated at outside hospitals. 10.1371/journal.pone.0127744.t002 ###### Patient outcomes at UHCMC and comparison with relevant studies describing 30-day mortality and relapse. {#pone.0127744.t002g} Study UHCMC2008--20125 years Levandovsky, et al.1978-200224 years Kim, et al.1998-200810 years Lara, et al.1978-199820 years George, et al.1989-200314.5 years Roberts, et al.1984-19906 years --------------------------------------------------------------- --------------------------------------------------- ------------------------------------------------------------- ----------------------------------------------------- ------------------------------------------------------------------------- ------------------------------------------------------------------------------------------- --------------------------------------------------- Total cases (total cases used for mortality rate calculation) 59 (59) 178 (167)[^a^](#t002fn004){ref-type="table-fn"} 52 (52) 126 (124)[^c^](#t002fn006){ref-type="table-fn"} 290 (214)[^e^](#t002fn008){ref-type="table-fn"} 14 (14)     Idiopathic cause (%) 75 72 27 71 39.3 43     Secondary cause (%) 25 28 73 29 60.7 57     Age 49[\*](#t002fn012){ref-type="table-fn"} 49[\*](#t002fn012){ref-type="table-fn"} 47[\*](#t002fn012){ref-type="table-fn"} 49[\*](#t002fn012){ref-type="table-fn"} 36[\*\*](#t002fn013){ref-type="table-fn"} [^f^](#t002fn009){ref-type="table-fn"} 45[\*](#t002fn012){ref-type="table-fn"}     Gender (Female, %) 70 68 61.5 66 71 50 At presentation     Hemoglobin (mg/dL) 9[\*](#t002fn012){ref-type="table-fn"} 9.2[\*](#t002fn012){ref-type="table-fn"} 7.6[\*\*](#t002fn013){ref-type="table-fn"} 8.9 [\*](#t002fn012){ref-type="table-fn"} NS 9.8[\*](#t002fn012){ref-type="table-fn"}     Serum creatinine (mg/dL) 2.6[\*](#t002fn012){ref-type="table-fn"} 3.2[\*](#t002fn012){ref-type="table-fn"} 2.5[\*\*](#t002fn013){ref-type="table-fn"} 3,4[\*\*](#t002fn013){ref-type="table-fn"} 1.2 [\*\*](#t002fn013){ref-type="table-fn"} [^f^](#t002fn009){ref-type="table-fn"} 5.7 [^g^](#t002fn010){ref-type="table-fn"}     Platelet count 46 x 10^9^/L[\*](#t002fn012){ref-type="table-fn"} 49 x 10^9^/L[\*](#t002fn012){ref-type="table-fn"} 30 x 10^9^/L[\*\*](#t002fn013){ref-type="table-fn"} 44 x 10^9^/L[\*](#t002fn012){ref-type="table-fn"} 11 x10^9^/L[\*\*](#t002fn013){ref-type="table-fn"} [^f^](#t002fn009){ref-type="table-fn"} 50 x 10^9^/L[\*](#t002fn012){ref-type="table-fn"}     Clinical Severity Score 4[\*](#t002fn012){ref-type="table-fn"} 4.4 NS 5 NS NS Complete response (%) 81.2 65 51.9 56 61[^h^](#t002fn011){ref-type="table-fn"} NS 30days-Relapse rate (%) 18.6 18 NS 13 12 28.6 TTP-HUS related deaths 30-days (number and %) 4 (6.7%) 23 (14%) [^b^](#t002fn005){ref-type="table-fn"} 18 (34.6%) 12 (10%) [^d^](#t002fn007){ref-type="table-fn"} 38 (17%) 1(7%) TPE as principal treatment (%) 100 96 100 97 100 100 Others principal treatment N/A FFP infusion and Staphylococcal protein A absorption column N/A FFP infusion, protein adsorption column therapy N/A N/A Number of TPE procedures 12[\*](#t002fn012){ref-type="table-fn"} 8[\*\*](#t002fn013){ref-type="table-fn"} 5[\*\*](#t002fn013){ref-type="table-fn"} NS 20 [\*\*](#t002fn013){ref-type="table-fn"} [^f^](#t002fn009){ref-type="table-fn"} NS TPE-Technique used COBE Spectra NS COBE SpectraMFM (Plasauto) Fenwal CS-3000 Blood cell separator,Haemonetics model V50, COBE Spectra NS NS NS = Not specified N/A = Not applicable MFM = membrane filtration method \(a\) 11 patients did not have death information and were excluded from the mortality rate calculation \(b\) Total of 27 patients died but 4 patients died of causes unrelated to TTP-HUS and were excluded from the mortality rate calculation \(c\) 2 patients' charts were unavailable for review \(d\) 1 patient died of an HIV-related infection and was excluded from the mortality rate calculation \(e\) 76 patients had additional/alternative disorder and were excluded from the mortality rate calculation \(f\) Detailed data from only 18 patients with severe ADAMTS13 deficiency were presented \(g\) Calculated from mmol/L \(h\) Defined as \>150 x10^9^/L platelet count \* = mean \*\* = median All patients started receiving and underwent TPE at our institution performed according to the manufacturer's recommendations as well as our standard operating procedure. The daily procedures were done until achieving the target platelet level. Study patients received a mean of 12 TPEs (range 1--42); complete response was obtained in 81.2% of patients (N = 39/48) (platelet count \> 100 x 10^9^/L); partial response in 8.3% of patients (N = 4/48) and 10.4% had no response (N = 5/48). Forty percent (N = 19/48) of the patients obtained platelet counts greater than 150 x 10^9^/L. Response relative to ADAMTS13 activity {#sec016} -------------------------------------- In the overall cohort of 59 patients, those with severe ADAMTS13 deficiency (≤10%) required more TPE than those patients with enzyme deficiency \>10% (mean 19.1 vs. 7.2). If the patients with ADAMTS13 deficiency are further subdivided into moderate deficiency (11%-40%), mild deficiency (41%-66%) and no deficiency (≥67%) there was no difference in the number of TPE in each subgroup (data not shown). For those patients who had a diagnosis of TTP-HUS and enzyme activity \>10% the mean number of TPE received was 8.2. In our patient cohort, 8/59 (13.6%) patients had no ADAMTS13 measured at start or throughout TPE regimen. Of the 51/59 (86.4%) patients who had ADAMTS13 activity measured, 18/59 (30.5%) had severe metalloprotease deficiency (\<10%); 16/59 (27.1%) patients had moderate to mild deficiency; and 17/59 (28.8%) patients had no deficiency in ADAMTS13 activity. Comparison with published literature {#sec017} ------------------------------------ We pooled the data of five studies that met the selection criteria ([Fig 1](#pone.0127744.g001){ref-type="fig"} and [Table 2](#pone.0127744.t002){ref-type="table"}). These five studies \[[@pone.0127744.ref012], [@pone.0127744.ref014], [@pone.0127744.ref016], [@pone.0127744.ref020], [@pone.0127744.ref021]\] described a total of 660 patients, 571 of which were used to calculate the mortality rate. Clinical Severity Score was derived only for two of the studies. Mean number of TPE are reported by 3/5 studies and practically all patients used in the statistical analysis received TPE. As shown in [Fig 2](#pone.0127744.g002){ref-type="fig"}, our overall mortality rate was significantly lower than the combined reported rate of all studies (4/59 (6.7%) vs. 92/571 (16%) respectively; *p* = 0.04). Our relapse rate, 11/59 (18.6%), was not significantly different from that which has been reported for the other studies used in our analysis (range 12--28%). However, when only those patients with TTP-HUS are taken into account the mortality rate (1/48) was more significant (*p* = 0.0053). {#pone.0127744.g002} Discussion {#sec018} ========== We have defined the clinical outcome of relapse and 30-day mortality in a population of 48 patients with a diagnosis of TTP-HUS. TPE was used as therapy for all patients whether the condition was idiopathic or secondary. Of interest, our 30-day mortality rate was significantly lower than what has been previously published but our relapse rate was similar to published reports. The marked difference in mortality rates may be difficult to dissect but in order to minimize confounding biases we compared our cohort findings to studies also reporting 30-day mortality rates and similar patient populations to ours since we were unable to perform a meta-analysis due to the small number of studies and limited raw data provided in these publications An exhaustive search for treatment modalities/ interventions in published literature which included patients with either TTP or HUS found that patients with TTP tend to respond to TPE vs. patients with typical/ diarrheal HUS who respond best to supportive therapy that includes dialysis \[[@pone.0127744.ref022]\]. At our institution, all patients were given immunosuppression in the form of steroids which was given and discontinued at different intervals during TPE regimen as previously described \[[@pone.0127744.ref017]\]. This is similar to what has been described for a similar size patient cohort \[[@pone.0127744.ref023]\]. Therefore, the combination of immunosuppression and TPE may account for our improved survival. However, it did not have a role in decreasing relapse rates. In any case this is the therapeutic approach taken by most institutions (standard of care) and it is unlikely to explain our mortality differences with published reports. As a result, inherent differences in the patient cohorts which lead to marked differences in mortality may begin to explain our observations. Studies which have grouped patients under TTP-HUS as a single clinical entity find that responses differ depending on the elements identified in the clinical presentation and patients' gender \[[@pone.0127744.ref016]\]. Likewise, ABO blood group may be an independent risk factor for patients with TTP who have severe ADAMTS13 deficiency \[[@pone.0127744.ref024]\]. As a result, potential reasons for these reported differences include heterogeneity of patients diagnosed with TTP-HUS and serious underlying co-morbid conditions. Since the majority of our patients had idiopathic TTP-HUS of which half had severe ADAMTS13 deficiency (TTP diagnosis), may make our better survival even more significant. Of interest, a recent report from our group showed that the response to TPE in patients with ADAMTS13 deficiency follows similar kinetics in platelet count recovery \[[@pone.0127744.ref025]\]. There is overlap between the two syndromes and the distinction between HUS and TTP based on routine laboratory and clinical presentation is at times challenging which has led to the unifying diagnosis of TTP-HUS to describe this patient population. For these reasons the term unifying both diagnoses into a single entity has been introduced \[[@pone.0127744.ref026]\]. However, if two potentially pathologically different disease entities were grouped, it could lead to differences in mortality rates and give creed to alternative suggestions for the markedly lower mortality rate that we observed in our patient cohort. In addition, unlike other studies included in our analysis, most of our patients had ADAMTS13 activity measured which was helpful in sub-stratifying our patient cohort. However, even subdividing patients according to ADAMTS13 activity still yielded a lower mortality rate than those prior reports. The grouping of TTP and HUS into a unifying single pathologic entity may not represent a uniform way of studying, reporting and comparing these patients that is reproducible. A great deal of debate has placed by some investigators these two potential diseases under the same clinical umbrella as manifestations of a single disease continuum \[[@pone.0127744.ref013], [@pone.0127744.ref026], [@pone.0127744.ref027]\]. However, grouping these two clinical entities may oversimplify these two complex syndromic presentations that fails to recognize previously reported markedly different histopathologic findings in a large cohort of autopsied patients with either TTP or HUS which showed non-overlapping pathological signs, presentations that were distinct, and do not support the view that they represent a continuum of the same disease process \[[@pone.0127744.ref019]\]. Similarly, a meta-analysis looking at both clinical entities found that TTP patients respond to TPE leading to restoration of platelet count while HUS patients do not fully benefit from plasma therapy \[[@pone.0127744.ref028]\]. Molecularly, ADAMTS13 deficiency has to happen in the context of a "second hit" which is necessary for the clinical presentation to occur \[[@pone.0127744.ref029]\]. These could come in the form of both inhibitory and non-inhibitory antibodies to ADAMTS13 which could begin to explain why not all patients present with severe metalloprotease deficiency and that may also depend on the subjects' genetic susceptibility as seen in different strains of ADAMTS13 knockout mice \[[@pone.0127744.ref029]\]. However, all ADAMTS13 deficient mice have shown a pro-thrombotic phenotype. The second hit could be in the form of an increase in vWF as shown in a TTP mouse model in which disease severity improved with the use of recombinant human ADAMTS13 \[[@pone.0127744.ref030]\]. Likely, disease dynamics of TTP and HUS may prove to be independent of each other with few if any intermediates mediating both pathologic presentations. In the case of TTP the correlation with ADAMTS13 deficiency may be stronger than the metalloprotease's link to HUS \[[@pone.0127744.ref012]\]. Of note, though there is a wide variety of animal models that are used to determine the pathologic mediators behind TTP and HUS, these models have not yielded any shared molecule that is involved in the pathology of both \[[@pone.0127744.ref029]\]. More basic research is needed to dissect these possibilities further. Additionally, promptness of TPE initiation is essential for patient survival \[[@pone.0127744.ref031]\] since in those whose diagnosis is delayed are more likely to die than those promptly treated \[[@pone.0127744.ref023]\]. Another possibility for our better outcomes is that at our institution TPE could be initiated earlier than others, based on clinical practice that patients with MAHA and thrombocytopenia of unknown etiology is suspected of TTP-HUS until proven otherwise. This is however unlikely. Along these lines, it has been proposed that adults with idiopathic disease characterized by unexplained thrombocytopenia, MAHA, normal international normalized ratio, partial thromboplastin time and D-dimer may have TTP-HUS \[[@pone.0127744.ref032]\]. However, this algorithm may be more based in caution than in actual evidence that these two disease entities are isoforms of a single syndromic presentation due to valid concerns that failure to timely treat a TTP diagnosis will lead to a potentially poor and complicated clinical outcome \[[@pone.0127744.ref032]\]. Differences in survival could also be due to lack of uniformity in plasma exchange practice, apheresis equipment used (COBE, membrane filtration method \[[@pone.0127744.ref016]\]; Staphylococcal protein A absorption column;\[[@pone.0127744.ref012]\] Fenwal CS-3000 Blood Separator and Haemonetics model V50 \[[@pone.0127744.ref014]\]; use of plasma infusion only \[[@pone.0127744.ref020]\]; and as adjunctive therapy \[[@pone.0127744.ref033]\]. Our institution has a TPE protocol, which is universally applied for all patients presenting with suspected TTP-HUS; additionally, as a tertiary referral center for others in our region, patients may begin adequate treatment prior to transfer or may not have survived their disease prior to transfer. These findings may begin to explain the observed discrepancies among studies. In summary, our findings reinforce that early diagnosis; promptness of TPE initiation and use of TPE as standard treatment in the setting of TTP-HUS is essential to avoid the high 30-day mortality rate. Differences in survival may be due to lack of uniformity among the patient cohorts reported by us and others. Survival of TTP-HUS patients continues to improve; however, the marked differences in survivability need to be addressed with prospective studies that take into account the actual differences between these two disease presentations and negate the grouping of both under a single pathological entity. Data included in this manuscript was presented by CEB at the 66^th^ annual AABB 2013 meeting, in Denver CO (Transfusion 2013; 53(2S):SP149). We would like to thank Ms Donna Hendrix and Ms Vanessa Griffin in the Apheresis Center of UHCMC for their assistance; and to Ms Iryna Maitta and Ms Annette Graves for helping in the preparation of the manuscript. Additionally, the authors would like to thank Dr Hollie M. Reeves for critical review of the manuscript. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CEB JPH RWM. Performed the experiments: CEB JPH RWM. Analyzed the data: CEB RWM. Wrote the paper: CEB RWM.

Introduction ============ Lymphangiomas are a heterogeneous group of vascular malformations of the lymphatic channels composed of cystically dilated lymphatics. According to Landing and Farber \[[@B1]\] , those benign malformations, are classified in four categories: capillary lymphangioma, cavernous lymphangioma, cystic lymphangioma (hygroma) and hemo-lymphangioma (combination of hemangioma and lymphangioma). The latter, congenital malformation, can remain asymptomatic for a long period of time. On the other hand, it may grow rapidly, surrounding or infiltrating the neighboring tissues or other major structures, thus making the excision a real challenge for the physician \[[@B2],[@B3]\] . Hemo-lymphangiomas are in most cases detected at birth or early in a child\'s life, usually before the age of two years. Alternatively, with the introduction of the prenatal ultrasound, the diagnosis can be placed in the uterus \[[@B4]\] . This essay is a case report of two children with hemo-lymphangiomas of the lower extremities and its purpose is to define the most effective therapeutic approach of those lesions. Written parental permission was obtained to allow the use of confidential information held in the hospital\'s records, as Institutional Review Board (IRB) does not exist in our country. Case 1 ====== A two month old female infant presented to the orthopedic examination room with a palpated mass on the anteromedial side of the proximal left tibia. On the physical examination the lesion was found to be oval in shape, soft, compressible and painless. Anteroposterior and lateral plain radiographs demonstrated swelling of the soft tissue at this part of the tibia without signs of bony erosion (Fig. [1](#F1){ref-type="fig"}). The ultrasound (Doppler sonography) detected a cystic lesion with blood flow, measured 12 mm × 4 mm, while the integrity of the bone cortex was confirmed. The most possible diagnosis was hemo-lymphangioma. In the following six months, despite the fact that the size of the mass increased, the patient remained asymptomatic. A new ultrasound was performed (Fig. [2](#F2){ref-type="fig"}) and measured the mass 23 mm × 6 mm; the origin remained cystic. In the yearly follow-up the lesion\'s size increased to 44 mm × 37 mm. {#F1} {#F2} Due to the continuous augmentation of the mass, surgical excision was decided. An MRI (Fig. [3](#F3){ref-type="fig"}) was performed, pre-operatively, in order to establish the extent of the tumor and define the relationship to the surrounding structures. A well-defined extra-articular cystic malformation with a maximum diameter of 45 mm was viewed, located on the anteromedial side of the left tibia. The lesion had fine adhesions to the surrounding tissues and the cortex of the bone was found intact. {#F3} En bloc resection was performed releasing the specimen from the adhesions to the subcutaneous tissues and the medial head of the gastrocnemious. Macroscopically, it was found oval in shape with harsh features and spotty surface (Fig. [4](#F4){ref-type="fig"}). {#F4} Histological examination described the resected tumor as a fibro-lipomatous mass containing dense fibrous conjunctive tissue with vascular areas of lymphatic cells and vessels filled in with red blood cells (Fig. [5](#F5){ref-type="fig"} &[6](#F6){ref-type="fig"}). The definitive histological diagnosis was hemo-lymphangioma. {#F5} {#F6} Ten days post-operatively the trauma area was swelling; the seroma formed was aspirated and gave 20 ml of sterile, bloody fluid. A week later, the swelling recurred; surgical exploration and lavage was performed, while a drainage was placed. Intravenous Netilmicin and Cefamandile Nafate were administered for a total period of ten days, whereas the leg was immobilized on an above knee splint. The swelling was progressively decreased. A week after, the trauma area was examined by the ultrasound and 1,58 ml\'s of fluid were demonstrated. A full plaster was placed with the knee joint flexed in 90° for three weeks. At the end of that period of time the volume of the fluid was less that 1 ml. After two months of follow-up by clinical and Ultrasonography examination, the quantity minimized to zero, while in the yearly follow-up no recurrence was observed. Case 2 ====== The second case regards a 5 year-old male patient, with a palpable, painless mass on the anterior side of the proximal left tibia, located 2 cm below the tibial tubercle. The tumor, within a two month follow-up, was doubled in size; although it remained asymptomatic. A soft tissue swelling and intact bone cortex were demonstrated by the plain radiographs (anteroposterior and lateral views). The ultrasound performed depicted a cystic mass with moderate vascularization, measured 21 mm × 18 mm × 5 mm (Fig. [7](#F7){ref-type="fig"}). {#F7} The treatment decided was en bloc resection of the tumor, followed by rigorous cauterization of the blood and lymphatic vessels; a drainage was placed. Microscopically, the tumor examined, was described as a soft tissue cystic mass consisted of lymphatic and blood vessels; the stroma was infiltrated by lymphocytes. The histological diagnosis was hemo-lymphangioma. Cephalosporin was administered, in 3 doses totally, pre and post-operatively, while the limb was immobilized on a splint with the knee joint flexed in 30° for a total period of three weeks. In the trauma area no seroma was formed thereby, we removed the drainage. At the yearly follow-up no recurrence was observed, verified by the use of Ultrasonography (Fig. [8](#F8){ref-type="fig"}). {#F8} Discussion ========== Hemo-lymphangiomas are rare benign tumors that appear to arise from congenital malformation of the vascular system. The formation of that tumor may be explained by obstruction of the venolymphatic communication, between dysembrioplastic vascular tissue and the systemic circulation \[[@B5]\] . Hemo-lymphangiomas are mostly presented as cystic or cavernous lesions. Very few cases of hemo-lymphangiomas of the extremities have been reported in the literature. A retrospective study of one hundred and eighty-six (186) patients\' presenting one hundred and ninety-one (191) hemangiomas was published; their anatomical location was: 48% in the head and neck, 42% in the extremities and 10% in internal or visceral locations. Histological examination revealed that only three of them had combined hemo-lymphangiomas \[[@B6]\] . Macroscopically, complete excision gave the best results with lower recurrence rate. On the contrary, aspiration and injection of sclerotic agents gave the highest risk of recurrence. Among the most important risk factors for recurrence are: anatomical location of the lesion, size, complexity and surgical technique. In that study, the average follow-up period was three years, while 95% of the patients completed a sufficient period of twelve months. The incidence of hemo-lymphangiomas varies from 1,2 to 2,8 per 1000 newborns\[[@B7]\] however, only 6,5% of them are located on the extremities \[[@B8],[@B9]\] ; both sexes get equally affected. The diagnosis in most cases (90%) is placed before the age of two years \[[@B2]\] , while 60% of those are present at the time of birth. The clinical on set of hemo-lymphangiomas can vary from a slowly growing cyst over a period of years to an aggressive enlarging tumor, without invasive ability. Their size varies due to the anatomical location and relationship to the neighboring tissues. Small tumors are usually superficial, while the large ones are located deeper and have cystic texture. The most common complications are random or traumatic hemorrhage, rupture, infection and lymphorrhya. In the clinical examination they are usually described as soft and compressible masses, loculated in pattern. Histologically, hemo-lymphangiomas consist of dense fibrous tissue that grows in bands between the numerous vascular spaces and invades the subcutaneous fat. Some of those are blood vessels and the others lymphatic. Imaging modalities, Ultrasonography, Computed Tomography and Magnetic Resonance, are useful in confirming the diagnosis, describing the margins of the mass and planning the surgical strategy \[[@B10]\] . The ultrasound, that is a low cost modality, may demonstrate the solid or cystic nature of the lesion, therefore should be ordered in a routine base; the extension and the relationship of the tumor to the surrounding tissues is rather depicted by the MRI \[[@B11]\]. The differential diagnosis includes hemangioma, lipoma, teratoma, congenital lymphedema, dermoid cyst and neurofibroma. Needle aspiration or biopsy can also be useful in the differentiating hemo-lymphangioma to other fluid-filled masses. As far as it concerns the therapeutic approach of localized hemo-lymphangiomas, the treatment of choice is complete surgical excision, which also presents the lowest recurrence rate. An important issue regards the time of surgery. Some authors express the aspect that the mass should be excised promptly after being diagnosed, while others prefer to wait and examine the size regularly. In the latter case, the risk of infection and hemorrhage remains, while complete surgical eradication may become more difficult. Other techniques available are: aspiration and drainage, cryotherapy, injection of sclerotic agents and radiotherapy (by radium, roentgen ray or radon seed), but none of them produced acceptable results. Radiotherapy is used when surgical excision is not feasible; the radio-sensitivity of hemo-lymphangiomas is not well understood, however in the past they were considered to be radio-resistant. In children, radiotherapy may lead to tumor retardation or to malignant transformation \[[@B12]\]. When surgical eradication fails and the mass recurs, conservative treatment methods may be applied. The recurrence rates vary depending on the complexity of the mass, the anatomical location and the adequacy of the excision. However, lesions that have been completely excised, present 10-27% recurrence, while those being partially resected may recur in 50-100%. An important issue remains the continuous lymphorrhya during the instant post-operative period; attentive en bloc excision of the tumor, followed by rigorous cauterization of the blood and lymphatic vessels may reduce or eliminate the risk of recurrence. Other auxiliary measures are the application of a bandage under pressure, the prolonged drainage and immobilization of the extremity. Conclusions =========== In the treatment of hemo-lymphangioma, surgical excision appears to be the best choice of treatment, especially when the tumor increases in size, creating pressure to the surrounding tissues. Eradicate attentive excision offers the minimum risk of recurrence. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= IK, who is the corresponding author, was the surgeon of the first patient, gathered the articles used as references and compiled the manuscript. M.V was the surgeon of the second patient. A.K helped with the editing of the manuscript. KF, as the director of the orthopaedic department, guided us and helped in finalizing the manuscript. All authors have read and approved the final manuscript Consent ======= Written informed consent was obtained from the patients for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

1. Introduction {#sec1} =============== MD is a rare clinical disease which is characterized by abnormal hypertrophy of gastric folds; causing hypoproteinemia due to protein losing gastropathy \[[@B1]\]. It also causes hypochloremia and associated with increased chances of gastric adenocarcinoma. The pathophysiology is not exactly clear, but it is hypothesized that up regulation of Transforming growth factor (TGF) is the underlying mechanism of cellular changes in gastric mucosa \[[@B2]\]. It is also noted that enormous secretions of thick mucus and transforming growth factor alpha (TGF-*α*) contribute towards gastric hypertrophy \[[@B3]\]. Some authors have discussed its linkage to *Helicobacter pylori* (*H. pylori*) infection \[[@B4]\]. The disease commonly affects the middle-aged males but younger pediatric population is also involved \[[@B5], [@B6]\]. The clinical features include upper abdominal pain and discomfort, nausea, vomiting, weight loss, and swelling of the body due to hypoalbumemia. Severe disease lead to intractable protein loss, upper gastrointestinal bleeding or gastric cancers \[[@B7]\]. Endoscopically giant gastric folds are visualized and on microscopic examination hyperplasia of the mucus-secreting surface cells is seen commonly in the gastric pits which is called foveolar epithelium. The hypertrophy of the foveolar epithelium results in replacement of the normal gastric glandular structure. These changes mainly occur in the gastric body where gastric glands containing chief cells are located \[[@B4]\]. The disease is diagnosed with clinical features and further laboratory testing including serum albumin levels, EGD, and testing for *H. pylori* \[[@B4], [@B7]\]. It is commonly treated with proton pump inhibitors, high protein diet, *H. pylori* eradication, cetuximab (monoclonal antibody), and somatostatin analog injections octreotide long-acting release \[[@B4], [@B6], [@B8]\]. In severe progressive disease and medically resistant cases, total and partial gastrectomy are required for management. 2. Case Presentation {#sec2} ==================== A 58-year-old male with comorbidities of hypertension, chronic obstructive pulmonary disease, benign prostatic hyperplasia, and gastro-esophageal reflux disease was initially referred to the gastroenterology clinic by primary care physician for the surveillance colonoscopy, and evaluation of dark colored stools. Patient noted dark colored stools during the Christmas weekend associated with nausea, vomiting, and abdominal pain. The abdominal pain was epigastric, gradual onset, severe in intensity, nonradiating, aggravated with emesis with no specific relieving factors. Denies any hematemesis. He had colonoscopy 10 years ago and had history of colonic polyps. At the time of assessment in the clinic he denied any nausea, vomiting, and worsening abdominal pain or other symptoms. Physical examination of cardiovascular, respiratory, gastrointestinal, and neurological system was unremarkable. Vitals were within normal limits. Recent laboratory workup showed hemoglobin of 14.7 g/dl with normal renal and hepatic parameters including normal albumin levels. He has past medical history of anemia and esophagogastroduodenoscopy (EGD) and Colonoscopy which revealed arteriovenous malformations (AVMs). He also has history of Alcoholic pancreatitis, alcohol dependence, smoking but no history of any drug abuse. Family history was negative for the colon cancer and stomach cancer. For gastroesophageal reflux disease, he was prescribed PPIs as needed but he was not taking it regularly due to little or no relief from them. For the evaluation of GI bleed, he underwent colonoscopy in January 2016 which showed normal ileum, few 2--4 mm hyperplastic polyps in the sigmoid colon, and few 3 mm polyps in the rectum which had reactive lymphoid aggregate and focal epithelial hyperplasia on biopsy. Other findings included diverticulosis in the sigmoid colon and in the descending colon and nonbleeding internal hemorrhoids. In January 2016 an EGD showed normal esophagus but enlarged gastric folds and erythematous mucosa in the antrum and stomach biopsy showed gastric antral mucosa with focal mild increase of eosinophils and congestion. Three nonbleeding angioectasias were seen in the duodenum which were treated with argon plasma coagulation (APC). Mucosa biopsies were negative for *H. pylori* infection. For evaluation of the AVMs, Small bowel enteroscopy was done on February 2016 which showed mucosal changes in the jejunum and biopsy revealed small bowel mucosa with dilated lymphatics. EGD was repeated for polypectomy in February 2016 which showed enlarged gastric folds and were ligated. There was normal duodenal bulb and mucosal resection was successfully performed. The biopsy was consistent with fundic gland polyps. He was referred to the gastroenterology clinic after eighteen months for abnormal abdominal computerized tomography findings of gastric wall thickening which he had for abdominal pain. This epigastric pain was gradual in onset, severe in intensity, radiating to right upper quadrant, and progressively worsened. To rule out gastric malignancy, EGD and EUS were done. EGD in September 2017 showed giant gastric folds, much larger than seen on the prior endoscopies. This is shown in [Figure 1](#fig1){ref-type="fig"}. The biopsied enlarged gastric folds revealed gastric fundic mucosa with foveolar hyperplasia, dilated fundic glands and chronic gastritis which is shown in [Figure 2](#fig2){ref-type="fig"}. Another gastric fundus nodule biopsy revealed gastric fundic mucosa with foveolar hyperplasia and mild chronic inflammation. With the background of thickened gastric folds and episodes of recurrent pancreatitis, the patient underwent upper endoscopy ultrasound for the evaluation of the pancreas as well as the stomach on March 2018 which showed wall thickening in the body of the stomach. The thickening appeared to be primarily within the submucosa (Layer 3) of the stomach ([Figure 3](#fig3){ref-type="fig"}). There was no sign of significant pathology in the common bile duct, gall bladder, and the main pancreatic duct. Stomach biopsy results were consistent with Menetrier disease. Patient was followed in the clinic and explained the risks of adenocarcinoma associated with this disease. Our patient has worsening of the disease in one and half years. 3. Discussion {#sec3} ============= MD is a diagnostic challenge due to the rare occurrence of the disease, complex pathophysiology and the lack of precise diagnostic criteria \[[@B7], [@B9]\]. Our case is a unique example of MD that is without hypoalbuminemia and without gastric colonization with *H. pylori* like other similar cases reported in literature \[[@B5], [@B10]\]. This points out to the fact that further research is needed in understanding the pathophysiology of the disease in addition to *H. pylori* infection and low serum albumin levels. It is well known that adult MD occurs in the setting of *H. pylori* infection of the stomach \[[@B11], [@B12]\]. In 1993, Bayerdorffer et al. evaluated about 138 cases of hypertrophic gastropathy and found more than 90% of patient having biopsies positive for *H. pylori* \[[@B13]\]. It is hypothesized cyclooxygenase (COX)-2 induced by *H. pylori* infections can result in antiapoptotic effect on the mucus secreting gastric cells resulting in hypertrophy of the glandular epithelium causing excessive protein and chloride secretion \[[@B14]\]. Although there are case reports of MD in the absence of *H. pylori* infection in the biopsied gastric tissues. This suggests that *H. pylori* may not be the causative agent but possibly a contributor towards the development of foveolar hyperplasia and gastric hypertrophy \[[@B14]\]. Low albumin level is an important finding for the diagnosis of MD \[[@B15]\] but in literature normal albumin levels in MD are mentioned in adults and children \[[@B5], [@B6], [@B10]\]. In our case, the serum albumin level of the patient was 4.1 g/dl without associated edema and never reduced during the follow up clinical visits. This speculates that our patient might have a new subtype of MD. The pathophysiology of MD is explained by the upregulated signaling of epidermal growth factor receptor (EGFR) which is the effect of increased production of TGF-*α*). This finding is based on transgenic mice with MD who were found to have elevated TGF-*α* levels in gastric mucosa. Endo et al. also found high levels of TGF-*α* in the hypertrophied foveolar gastric mucosa. These stimulate EGFR which are usually found mainly in the parietal cells are expressed on the lumen side of foveolar cells \[[@B3], [@B14]\]. Ligand binding to EGFR results in a cascade of proliferative events; ultimately accounting for hypertrophic gastropathy. The inducers of the TGF-*α*/EGFR signaling pathway are not well known but certain infectious agents are mentioned in literature. This include cytomegalovirus (CMV) and *H. pylori*. Wang et al. reported that CMV proteins stimulate the pathway of hypertrophy and proliferation \[[@B16]\]. This pathway is common in but rarely present in adults except in immunocompromised states \[[@B17]\]. However, in our case *H. pylori* was not found. Different possible etiologies of the large gastric fold as mentioned in [Figure 1](#fig1){ref-type="fig"}. This disease commonly manifests with epigastric pain and low albumin levels due to protein losing gastropathy \[[@B18]\]. Other symptoms include nausea, vomiting, diarrhea, weight loss, anorexia, early satiety, gastrointestinal bleed, fatigue, and edema \[[@B5]\]. The disease is progressive and affects men more than women with average age of diagnosis of 55 years \[[@B3]\]. The disease typically involves the whole fundus, body, and upper part of the gastric mucosa with sparing of the antrum \[[@B19]\]. However, there are case reports of involvement of the antrum \[[@B20]\]. MD mimics a wide variety of other diseases. These consist of different forms of polyps and polyposis syndromes like hyperplastic polyps or juvenile polyps or juvenile polyposis syndrome. These are differentiated from MD with detailed history including family history, additional clinical manifestations, endoscopic imaging, and histopathological analysis and genetic analysis. PPI can also cause hyperplasia of the gastric mucosa and mimics MD but in these cases histological examinations reveal predominance of parietal cells along with occasional dilation of the oxyntic glands. In sporadic cases, the apex of the parietal cell snout within the lumen of the gastric glands \[[@B21]\]. These findings were not seen in our histological examination which differentiate PPI induced hyperplastic glands from MD. In addition, gastrin producing tumors like Zollinger--Ellison syndrome also shows parietal cell hyperplasia in histopathological examination. MD in contrast has decreased appearance of parietal cell mass. It is essential to rule out other differentials of thickened gastric walls like malignancies (lymphomas, gastric cancers, gastrointestinal stromal tumors), infections (tuberculosis), or other infiltrative diseases \[[@B22]\]. Management of MD includes *H. pylori* eradication, use of octreotide for the modulation of the TGF-*α*/EGFR signaling, antibody binding and inhibiting the EGFR or surgical treatments. Many studies have mentioned positive results with *H. pylori* eradication using proton pump inhibitors and antibiotics \[[@B12], [@B23]\]. However, in *H. pylori* negative cases it is assumed that the disease will not respond to the eradication regimen of *H. pylori*. Some evidence suggests the somatostatin analogs like octreotide help in relieving symptoms but effect on the disease progression and development of malignancies is not clear \[[@B24], [@B25]\]. It is found that chronic cases of MD respond well to the monoclonal AB-cetuximab \[[@B8]\]. But cases refractory to the medical therapy, progressively worsening disease and for prevention of gastric cancer, gastrectomy is the best option. Partial gastrectomy is done with the use of laparoscopic technique \[[@B26]\]. Total gastrectomy is preferred despite MD sparing the antrum as risk of leakage from the anastomotic inflamed tissue is very high \[[@B7], [@B10]\]. However, there are cases of successful limited gastric resection in cases of MD localized to the gastric antrum \[[@B5]\]. This technique needs further exploration in these cases. Also, the percentage of risk of gastric adenocarcinoma with limited gastrectomy need to be further determined. It is very important to regularly follow up cases of MD as there is high risk of gastric tumors likely adenocarcinomas and lymphomas \[[@B15], [@B27], [@B28]\]. MD is considered as a premalignant condition and it reasonable to have regular follow up endoscopies to monitor the developed for cancers in this precancerous state. Cases are mentioned in literature where surveillance endoscopies were done 1-2 years to monitor the MD \[[@B15]\]. There is no definite time internal for the developed of the cancer in MD. In literature, cases are mentioned of simultaneous occurrence of MD and gastric malignancy \[[@B29], [@B30]\]. Based on the limited data of case reports, the cancer developments were mentioned from 3.5 to 16 years since the initial diagnosis of MD \[[@B15], [@B28], [@B31]\]. Therefore, it is reasonable to monitor malignant transformation of MD for enough time. Patient should be notified about the possibility of cancer in MD and patient who are very concerned may opt to definitive treatment in the form of partial or total gastrectomy \[[@B15]\]. 4. Conclusion {#sec4} ============= This disease should be considered in patients coming with upper abdominal symptoms and endoscopic findings of large abnormal gastric folds with or without evidence of *H. pylori* and hypoalbuminemia. Polypectomy is essential for the diagnosis and patient can be managed medically initially. Surgical treatment is considered in cases refractory to medical management or in patients with suspected gastric malignancy or obstructive symptoms. As there is high evidence of malignant transformations in MD, regular endoscopic surveillance of the gastric mucosa is required. Consent ======= Informed consent for participation was obtained from this patient. Conflicts of Interest ===================== None of the authors have any financial conflicts of interest. {#fig1} {#fig2} {#fig3} [^1]: Academic Editor: Matteo Neri

INTRODUCTION ============ Blueberries are flowering plants in the genus of *Vaccinium*. Many species of blueberries have various beneficial properties. Among *Vaccinium*, the wild blueberry (*Vaccinium angustifolium*) has consistently shown higher levels of phenolics and anthocyanins, and demonstrated a wide variety of health-relevant bioactivities including anti-cancer, anti-diabetic, anti-hypertensive, anti-inflammatory effects, and protection against chronic diseases ([@b1-pnfs-23-030]--[@b8-pnfs-23-030]). We previously reported that *V. angustifolium* root extract (VAE) inhibited A23187 and phorbol myristate acetate (PMA)-induced degranulation via down-regulation of protein kinases C (PKC) translocation ([@b9-pnfs-23-030]). Moreover, the expression of FcɛRI, a high affinity IgE receptor, was down-regulated by VAE ([@b10-pnfs-23-030]). However, the regulation of protein tyrosine kinases (PTK) and nuclear factor kappa-B (NF-κB) expression by VAE has not been examined. The high affinity IgE receptor, FcɛRI, plays a crucial role in IgE-mediated allergic reactions, and it is expressed on the surface of effector cells such as basophils and mast cells ([@b11-pnfs-23-030],[@b12-pnfs-23-030]). Binding of allergen and IgE antibody complexes to FcɛRI causes the activation of a signaling cascade, which triggers the elevation of intracellular calcium levels and the secretion of various inflammatory mediators from activated basophils and mast cells, and causes allergic diseases such as asthma, allergic rhinitis and atopic dermatitis ([@b13-pnfs-23-030],[@b14-pnfs-23-030]). We previously reported that VAE inhibited FcɛRI-mediated calcium influx and degranulation ([@b10-pnfs-23-030]). Degranulation of mast cells and basophils is induced by various stimuli such as calcium ionophore, antigens, and anti-FcɛRI α chain antibody (CRA-1), which is accompanied by production of reactive oxygen species (ROS). Moreover, ROS generation depended on the activation of PTK such as Lyn and Syk, and PI3K in FcɛRI-signaling ([@b15-pnfs-23-030]). Activation of the signaling cascade after cross-linking of FcɛRI-bound IgE antibody with allergens determines the interaction of FcɛRI with Src kinases, Lyn and subsequent activation of Syk, other tyrosine kinases, and mitogen-activated protein kinases (MAPK) such as extracellular regulated kinases (ERK)-1/2, c-jun N-terminal kinase (JNK), and p38 MAPK ([@b16-pnfs-23-030]--[@b19-pnfs-23-030]). Moreover, NF-κB activation is regulated by MAPK and contribute to the expression of inflammatory mediators in allergic reactions ([@b17-pnfs-23-030]). We previously reported that VAE negatively regulated degranulation through inhibition of PKC translocation and FcɛRI expression through inhibition of ERK-1 activation in human basophilic KU812F cells ([@b18-pnfs-23-030], [@b19-pnfs-23-030]). To identify the suppressive molecular activities of VAE on FcɛRI-mediated allergic reactions, we evaluated the regulation of FcɛRI-mediated PTK involving Syk and Lyn, and NF-κB activities in anti-human FcɛRI α chain antibody, in CRA-1-stimulated KU812F cells. MATERIALS AND METHODS ===================== Chemicals --------- RPMI-1640 medium and fetal bovine serum (FBS) were purchased from HyClone Laboratories (Logan, UT, USA). CRA-1 was acquired from Kyokuto (Tokyo, Japan). Antibiotics and antimycotics were purchased from Gibco BRL (Gaithersburg, MD, USA). Protease inhibitor cocktail was obtained from Roche Diagnostics GmbH (Penzberg, Germany). β-Actin, anti-phosphorylated Syk, Lyn, and NF-κB, and horseradish peroxidase (HRP)-conjugated secondary antibody were bought from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Chemiluminescence detection reagents were acquired from Perkin Elmer (Waltham, MA, USA), and polyvinylidene difluoride (PVDF) membrane was purchased from Millipore (Bedford, MA, USA). 2′7′-dichlorofluorescin-diacetate (DCF-DA) was obtained from Sigma Chemicals (St. Louis, MO, USA). Protease inhibitor cocktail was purchased from Roche (Penzberg, Germany). Enhanced chemiluminescence detection reagents were procured from Perkin Elmer. Cell culture, treatment, and stimulation ---------------------------------------- The human basophilic KU812F cells were obtained from the American Type Culture Collection (Manassas, VA, USA) and cultured in Roswell Park Memorial Institute (RPMI)-1640 supplemented with 10% heat-inactivated FBS, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (10 mM), penicillin (100 U/mL), and streptomycin (100 μg/mL) at 37°C in a humidified atmosphere with 5% CO~2~, and passaged every 3\~4 days. KU812F cells were treated with various concentrations of VAE in FBS-free RPMI-1640, and were induced by CRA-1. Extract preparation ------------------- The *V. angustifolium* roots were obtained from Quebec, Canada, and the dried. For extraction, 10 volumes of methanol was added to the powered *V. angustifolium* roots. The supernatant of the mixture was condensed in a vacuum, and lyophilized. The VAE was stored at −20°C in dimethyl sulfoxide. Total phenolic content (TPC) assay ---------------------------------- The TPC of the VAE was assayed using the Folin-Ciocalteau method, with slight modifications ([@b20-pnfs-23-030]). A 20 μL aliquot of the extract was added to 100 μL Folin-Ciocalteau reagent and 300 μL 20% Na~2~CO~3~ solution, and distilled water was added to a final volume of 2 mL. After 2 h, the absorbance was measured at 765 nm, and the concentration of TPC expressed as gallic acid equivalents (GAE) was determined using a calibration curve with gallic acid as a standard polyphenol. Intracellular ROS analysis -------------------------- The intracellular ROS activity was measured by the ROS-specific fluorescent probe, DCF-DA ([@b21-pnfs-23-030]). Cells were pretreated with VAE for 24 h, and then stimulated with CRA-1 for 30 min. The cells were treated with DCF-DA for 30 min, and the absorbance was measured at 485 nm for excitation wavelength and at 528 nm for emission wavelength. β-Hexosaminidase release assay ------------------------------ The β-hexosaminidase activity in the supernatant of treated and stimulated cells was determined spectrophotometrically. Briefly, the sample was aliquoted and 100 μL of 2 μM NP-GlcNAc (in 0.4 M citrate and 0.2 M phosphate buffer, pH 4.5) was added. The color was formed, after which the reaction was finished by adding 200 μL of 0.2 M glycine-NaOH, pH 10.7, and the absorbance was measured at 405 nm. The cells were subsequently lysed with 0.1% Triton X-100, after which the β-hexosaminidase activity of the VAE was measured. Western blot analysis --------------------- Phosphorylation of Lyn and Syk, and NF-κB was examined by Western blot analysis. Briefly, induced cells were lysed in cell lysis buffer containing 20 mM Tris-Cl (pH 8.0), 137 mM NaCl, 10% glycerol, 1% Triton X-100, 1 mM Na~3~VO~4~, 1 mM NaF, 2 mM ethylenediaminetetraacetic acid, and a protease inhibitor cocktail. The proteins were separated by 10% sodium dodecylsulfate-polyacrylamide gel electrophoresis, transferred to polyvinylidene difluoride membrane, and blocked with 10% skim in plain buffer (50 mM Tris-HCl, pH 7.5, 34 mM NaCl, and 0.001% Tween 20). The membrane was incubated with primary antibodies followed by anti-HRP conjugated secondary antibodies. Then, the chemoreactive proteins were visualized by enhanced detection reagents according to the manufacturer's instructions, and the membrane was then exposed to X-ray film, after which it was quantified. Statistical analysis -------------------- All experiments were carried out independently in triplicate. The data were presented as the mean±standard deviation (SD). Statistical differences between the control and VAE were resolved by a Student's *t*-test using the statistical software, SPSS (version 12.0; SPSS Inc., Chicago, IL, USA). *P*-values \<0.05 were considered statistically significant. RESULTS AND DISCUSSION ====================== Effects of VAE on FcɛRI-mediated PTK, Syk, and Lyn activation ------------------------------------------------------------- *V. angustifolium* contains many polyphenolic compounds, that can be extracted with methanol, which is the most suitable solvent for plant extraction. In a previous study, the TPC of VAE was 1,700±1.9 mg GAE/g ([@b9-pnfs-23-030],[@b10-pnfs-23-030]). Human basophilic KU812F cells express a high affinity for the IgE receptor, and they are therefore used as a cell line in FcɛRI expression research ([@b22-pnfs-23-030]). We previously found that VAE exerted no cytotoxicity at ≤20 μg/mL (data not shown) ([@b9-pnfs-23-030],[@b10-pnfs-23-030]). Therefore, the VAE concentrations of 1\~20 μg/mL were selected for further experiments. Mast cells and basophils are major allergic and immune effector cells in FcɛRI-mediated allergic reactions. A high affinity IgE receptor, FcɛRI, expressed on the surface of mast cells and basophils, has an important role in IgE-mediated allergic reactions ([@b23-pnfs-23-030],[@b24-pnfs-23-030]). The transcriptional signaling results in the activation by Fc receptors in basophils and mast cells have been extensively characterized, and the initial FcɛRI stimulation activates a transcriptional signaling cascade that involves activation of PTK such as Syk and Lyn and MAPK such as ERK 1/2, p38, and JNK ([@b25-pnfs-23-030],[@b26-pnfs-23-030]). We previously reported that VAE suppressed FcɛRI expression through inhibition of ERK-1 phosphorylation ([@b10-pnfs-23-030]). However, the effects of VAE on the regulation of FcɛRI-mediated PTK such as Syk and Lyn in FcɛRI-mediated allergic responses have not been characterized. Therefore, in this study, KU812F cells were pretreated with VAE for 24 h, and then stimulated with CRA-1 for 30 min. Western blot analysis showed that activation of Syk and Lyn was profoundly and dose-dependently inhibited by VAE ([Fig. 1](#f1-pnfs-23-030){ref-type="fig"}). Effects on FcɛRI-mediated ROS production ---------------------------------------- Activation of PTK, Lyn, and Syk stimulates ROS production ([@b27-pnfs-23-030]). To examine the effects of VAE on FcɛRI-mediated ROS production, KU812F cells were pretreated with VAE for 24 h, and then stimulated with CRA-1 for 30 min. VAE concentration dependently inhibited CRA-1-induced ROS production ([Fig. 2](#f2-pnfs-23-030){ref-type="fig"}). The produced ROS participate in the regulation of calcium mobilization. We previously reported that VAE negatively regulated FcɛRI-mediated intracellular calcium levels ([@b10-pnfs-23-030]). The degranulation process depends on increases in cytosolic calcium concentrations, and ROS play a crucial role in FcɛRI-dependent signaling, which results in degranulation of mast cells and basophils ([@b15-pnfs-23-030]). The histamine content and β-hexosaminidase activity are powerful markers of allergic disorders such asthma, atopic dermatitis, and rhinitis. In a previous study, we found that VAE inhibited FcɛRI-mediated histamine release (data not shown) ([@b10-pnfs-23-030]). Effects on FcɛRI-mediated β-hexosaminidase release -------------------------------------------------- Degranulation of effector cells such as mast cells and basophils through FcɛRI-cross linking or various stimuli leads to release of histamine and β-hexosaminidase, which causes the symptoms of allergic disorders ([@b23-pnfs-23-030]--[@b29-pnfs-23-030]). To assess the suppressive effects of inflammatory mediator secretion, KU812F cells were treated with VAE at 0, 1, 5, 10, and 20 μg/mL for 24 h, and then stimulated with CRA-1 for 30 min. VAE suppressed CRA-1-induced β-hexosaminidase release in a dose-dependent manner ([Fig. 3](#f3-pnfs-23-030){ref-type="fig"}). The results demonstrate that the secretion of inflammatory mediators from activated KU812F cells was negatively regulated by VAE, and it showed that VAE suppresses basophils degranulation. VAE inhibits FcɛRI-mediated NF-κB activation -------------------------------------------- The activation of MAPK and NF-κB is intimately associated with the expression of inflammatory mediators. MAPK involving ERK-1, p38, and JNK are important signal transcription factors that mediate cellular reactions including the allergic response, and are key signaling molecules in cell growth, development, differentiation, inflammation, and apoptosis ([@b16-pnfs-23-030]--[@b18-pnfs-23-030]). Moreover, MAPK have been investigated as rational targets for drug design for the treatment of allergic diseases. Activated MAPK regulate the transcriptional activity of many genes involved in the maintenance of cellular homeostasis. We previously reported that VAE down-regulated FcɛRI-mediated phosphorylation of ERK-1 ([@b10-pnfs-23-030]). NF-κB plays a crucial role in allergic reactions, and it is a transcriptional factor required for the expression of proinflammatory cytokines such as interleukin (IL)-1β, IL-6, and tumor-necrosis factor-α ([@b28-pnfs-23-030]). To examine the effects of NF-κB activation, cells were treated with VAE at various concentrations, and then stimulated with CRA-1 for 30 min. Western blot analysis revealed that VAE down-regulated the activation of NF-κB in a dose-dependent manner ([Fig. 4](#f4-pnfs-23-030){ref-type="fig"}). We previously reported that VAE showed potent inhibition of degranulation through down-regulation of PKC translocation in A23187 and PMA stimulated KU812F cells ([@b9-pnfs-23-030]). Moreover, VAE inhibited FcɛRI expression via suppression of ERK-1 activation in FcɛRI-mediated KU812F cells ([@b10-pnfs-23-030]). In the present study, we showed that VAE negatively regulated degranulation via inhibition of the activation of Lyn, Syk, and NF-κB. Further studies investigating the molecular mechanism of FcɛRI α chain gene regulation by multiple transcription factors are needed. Nevertheless, the results of this study suggest that the effects of VAE occur via down-regulation of FcɛRI-mediated allergic reactions, which contributes to its therapeutic activity. Accordingly, VAE may be useful for protection against allergic disorders. **AUTHOR DISCLOSURE STATEMENT** The author declares no conflict of interest. {#f1-pnfs-23-030} {#f2-pnfs-23-030} {#f3-pnfs-23-030} {#f4-pnfs-23-030}

Background {#Sec1} ========== Steroid resistant nephrotic syndrome (SRNS) remains one of the most common intractable causes of end-stage renal disease (ESRD) in children with 50--70% reaching ESRD within 5--10 years of diagnosis \[[@CR1], [@CR2]\]. Discovery of pathogenic variants in at least 40 genes responsible for maintenance of podocyte structure and functions with sporadic SRNS helped in identifying a distinct subgroup of SRNS \[[@CR3], [@CR4]\]. This group of children are generally unresponsive to immunosuppressive medications, and the disease does not recur post-transplantation \[[@CR5], [@CR6]\]. Thus, mutational analysis in SRNS is important because it helps in preventing unnecessary exposure to immunosuppressants and their adverse effects, besides establishing a molecular diagnosis and clear prognosis. WT1 mutations are associated with a spectrum of renal and extra-renal manifestations. Mutations in *WT1* (Wilm's tumour 1) are detected in 5--9% of children with SRNS with a higher frequency in those with congenital or infantile onset of nephrotic syndrome and in children with Diffuse Mesangial Sclerosis \[[@CR7]--[@CR9]\]. Most of the *WT1* mutations occur mainly in exon 8 and 9, which code for zinc finger domains 2 and 3, respectively which can lead to isolated SRNS, two distinct clinical syndromes that are associated with SRNS (1) Denys-Drash syndrome (DDS) (2) Frasier syndrome (FS), and conditions without Nephrotic syndrome such as Wilm's tumor, Meacham syndrome and somatic Mesothelioma (Online Mendelian Inheritance in Man, <http://www.ncbi.nlm.nih.gov/omim> database). DDS, mostly caused by mutations in exon 8 or 9 of WT1 is characterized by congenital/infantile NS, ambiguous genitalia, and a high risk for Wilms tumor while FS caused by mutations in the donor splice site at intron 9 is characterized by SRNS due to focal segmental glomerulosclerosis (FSGS), gonoadoblastoma and 46 XY disorder in sex development with sex reversal \[[@CR10]--[@CR12]\]. The current approach used for identifying pathogenic variants in the gene is Sanger's (Direct) sequencing. It requires additional steps after amplification of the target by polymerase chain reaction (PCR) leading to a higher turnaround time. Sequencing also adds to the cost, a major constraint in emerging countries in the management of this disease. Since incidence of *WT1* mutation in SRNS is lower in Asian population, as compared to western countries, we expect that majority of the patients tested in the Indian population will have normal (wild type) sequence \[[@CR13]--[@CR17]\]. Hence a fast and cost effective high throughput screening assay to identify individuals with normal sequence is an unmet clinical need. The proposed two -- step screening method will help in reducing the sequencing burden by elimination of samples with normal sequence and proceeding for Sanger's sequencing of only those samples suspected of carrying a mutation. The approach will also aid in quickening the clinical decision making process. One novel technique that is increasingly being used as screening tool for identification of normal or abnormal sequence pattern in the entire amplicon is the high resolution melting (HRM) analysis \[[@CR18]\]. The principle of HRM is that a single base change in the amplicon influences the thermodynamic stability of the duplex resulting in a slight change in the melting temperature (Tm) and the fluorescence absorbance behaviour during the melting of the DNA double strand to single strands. HRM is a rapid, closed tube high throughput system wherein PCR amplification and subsequent analysis are sequentially performed in the same tube and therefore more convenient and less labor intense compared to other mutation scanning methods such as denaturing high performance liquid chromatography (DHPLC) and fluorescent multiplexed-PCR analysis (FMPA). Being a closed tube method, the risk of contamination is low. In addition, the sample identified to have a probable mutation using HRM can be further processed for sequencing thereby avoiding additional PCR. HRM also has better sensitivity and specificity than DHPLC \[[@CR19]\]. Besides, HRM is not capital intensive since it can be performed directly with optimized primers without any other probe on a PCR instrument that collect fluorescent data with fine temperature resolution at the end of the PCR. The aim of our study is to report for the first time the frequency and spectrum of *WT1* mutations in Indian children with sporadic SRNS from a single center. Since we observed low prevalence of pathogenic variants in exons 8 and 9 of *WT1*, we developed and validated a simple assay based on high-resolution melting analysis to quickly identify normal sequence profiles of *WT1* exon 8 and 9 in a clinical setting to exclude them from being further processed for Sanger sequencing. Methods {#Sec2} ======= Subjects {#Sec3} -------- One hundred children diagnosed with initial SRNS or congenital nephrotic syndrome (NS) as defined by standard guidelines were included after informed consent and clinical data were recorded \[[@CR20]\]. Children with secondary nephrotic syndrome or those who had mutation in Podocin (*NPHS2*) gene were excluded. Blood samples for mutational analysis were obtained and genomic DNA was extracted from peripheral blood leukocytes by the phenol chloroform method \[[@CR21]\]. The quantity and quality of DNA was estimated using Nano-Drop ND-1000 (Thermo Scientific, India). To differentiate the mutations from polymorphisms, DNA from 50 healthy young adults, were used as controls. Mutation analysis by Sanger sequencing {#Sec4} -------------------------------------- Exons 8 and 9 including the splice site at intron 9 of *WT1* gene of all the recruited subjects were amplified using oligos (*WT1* \_E8_F\_A and *WT1* \_E8_R\_D for exon 8; *WT1* \_E9_F\_A and *WT1* \_E9_R\_D for exon 9) located on the intron-exon boundaries (Table [1](#Tab1){ref-type="table"}). Oligos were designed Primer 3 plus software \[[@CR22]\]. Bi-directional sequencing of the amplified products was performed using Big Dye Terminator v3.1 Cycle Sequencing Kit and analysed on an ABI 3730xl genetic analyser (Applied Biosystems, Foster City, CA) (Eurofins Genomics India Pvt Ltd, Bangalore, India). Sequences were evaluated for variants using the FinchTV 1.4.0 (Geospiza, Inc.; Seattle, WA, USA; [http://www.geospiza.com](http://www.geospiza.com/)) and NCBI BLAST.Table 1Mutagenic oligos (Round 1 PCR) used for site directed mutagenesis in exon 8 and 9 of the *WT1* geneMutationOligo NameOligo sequence 5'--3'Annealing temperature (°C)Exon 8\ c.1079G \> A;C360Y (hg19 - c.1283G \> A; C428Y)*WT1*\_E8_F \_A\ **E8_C428Y_R\_C**TTCCCCAAGGTGAGAAACCA\ ACCTTCGTTCATAGTCCTTGAAG56**E8_C428Y_F\_ B**\ *WT1*\_E8_R\_DCTTCAAGGACTATGAACGAAGGT\ GCTGCCAGCAATGAGAAGTG58Exon 8\ c.1119C \> A;H373Q (hg19 - c.1323C \> A;H441Q)*WT1*\_E8_F \_A\ **E8_H441Q_R\_C**TTCCCCAAGGTGAGAAACCA\ TGTCTCCTTTGTTGTCTTTTGAG58**E8\_ H441Q \_F\_ B**\ *WT1*\_E8_R\_DCTCAAAAGACAACAAAGGAGACA\ GCTGCCAGCAATGAGAAGTG58Exon 9\ c.1180C \> T;R394W (hg19 - c.1384C \> T;R462W)*WT1*\_E9_F\_A\ **E9_R462W_R\_C**TTGTTAGGGCCGAGGCTAGA\ TGGTCGGACCAGGAGAAC56**E9_R462W_F \_B**\ *WT1*\_E9_R \_DGTTCTCCTGGTCCGACCA\ AGTGCGTAAACTTTTCTTCACAT60Exon 9\ c.1190A \> C;H397P (hg19 - c.1394A \> C; H465P)*WT1*\_E9_F\_A\ **E9\_ H465P_R\_C**TTGTTAGGGCCGAGGCTAGA\ GTCTTCAGGGGGTCGGACC60**E9\_ H465_F\_B**\ *WT1*\_E9_R \_DGGTCCGACCCCCTGAAGAC\ AGTGCGTAAACTTTTCTTCACAT52Exon 9\ c.A1200C \> T;H401Y (hg19 - c.1405C \> T;H469Y)*WT1*\_E9_F\_A\ **E9\_ H469Y_R\_C**TTGTTAGGGCCGAGGCTAGA\ GTCCTGGTGTGAGTCTTCAGGTG58**E9\_ H469Y_F\_B**\ *WT1*\_E9_R \_DCACCTGAAGACTCACACCAGGAC\ AGTGCGTAAACTTTTCTTCACAT56Oligos in bold indicate mutagenic oligos Construction of mutated sequence by site directed mutagenesis {#Sec5} ------------------------------------------------------------- We used the site directed mutagenesis technique to introduce a mutation in each of the exons 8 and 9 in order to be able to validate the HRM technique as described by Heish et al., 2013 \[[@CR23]\]. The following single nucleotide change from the wild type were introduced: c.1079G \> A;p.C360Y (hg19 - c.1283G \> A;p.C428Y) and c.1119C \> A;p.H373Q (hg19 - c.1323C \> A; p.H441Q) in Exon 8 and c.1180C \> T;p.R394W (hg19 - c,1384C \> T;p.R462W), c.1190A \> C;p.H397P (hg19 - c.1394A \> C; p.H465P) and c.A1200C \> T;p.H401Y (hg19 - c.1405C \> T;p.H469Y) in exon 9 \[[@CR24]--[@CR27]\]. Round 1 PCR was carried out using mutagenic oligos (2 oligos per mutation; Table [1](#Tab1){ref-type="table"}). In the second round of PCR, both the first round amplicon products were mixed in 1:1 ratio and re-amplified to produce the full length PCR products of exon 8 and 9 of the *WT1* gene using the same oligos that were used for Sanger sequencing. The insertion of point mutations was verified by Sanger sequencing. High resolution melt analysis for *WT1* exon scanning {#Sec6} ----------------------------------------------------- The oligos sequence that was used for Sanger sequencing was also used for HRM. Standardization of the assay to determine the optimal annealing temperature and characterize the melt curve profiles of exons 8 and 9 of *WT1* gene was performed on five healthy control samples that were previously sequenced and confirmed to have wild type sequence. The standardization runs for each exon was initially performed separately and subsequently, the conditions were standardized to amplify both the exons in the same run. In brief, each reaction was performed in a final volume of 10 μl containing 20 ng of DNA, 250nM of each oligo (forward or reverse) and 1 × Melt Doctor Master mix (Thermofisher Scientific, USA). The PCR reaction conditions were as follows: initial denaturation step at 95 °C for 10 min followed by 45 cycles of 95 °C for 10s, 56 °C for 15 s, and 72 °C for 15 s. All HRM runs were performed in duplicate on Step one plus instrument (Thermofisher Scientific, USA). For the melt analysis, the PCR products obtained after amplification were first heated to 95 °C for 1 min and then rapidly cooled to 40 °C at 1.6 °C per second to allow heteroduplex formation. The PCR products were then reheated to 95 °C at 0.01 °C per second. Melt curve was monitored (by fluorescence emission) from 40 °C to 95 °C. The melt curve analysis was performed by the resolution melt (HRM) software version 2.0 (Thermofisher Scientific, USA) using default setting. Briefly, the melt curve data for the amplicons are normalized between 0 and 100% fluorescent intensity and temperature shifted (to eliminate temperature difference between samples) for comparisons. The normalized, temperature-adjusted plot allows visual comparison of melting curve shapes to detect heterozygous variants. For the difference plot, one melting curve is chosen as a reference, and the difference between each curve and the reference is plotted against temperature to give a "fluorescence difference". The original reference curve becomes a horizontal line at zero. Significant differences in fluorescence from the horizontal baseline indicate variations in the dissociation pattern of the amplicons. Differences are judged as significant if the technical duplicate fall outside the range. The derivative plot allowed direct visualization of the melting temperatures (Tm) and was used to differentiate between normal sequences of exon 8 and 9. To test for reproducibility, a single analyst prepared samples in duplicate and performed the HRM assay and the same analyst repeated the procedure on a different day. In addition, the melt curve prediction for both exons was carried out by the POLAND software using the nearest neighbor thermodynamic modeling, and the Blake and Delcourt algorithm \[[@CR28]\]. For validation of the assay, five samples with mutations that were generated by site directed mutagenesis and verified by sequencing and 3 samples from healthy controls for each exon were used as positive and negative controls respectively along with three randomly selected patient samples. To further test the clinical applicability of the assay, we included additional 91 previously sequenced samples from the cohort for HRM testing and a blinded assessment was performed. The healthy control samples used in standardization and validation experiment for exon 8 and 9 were used as reference samples and the melting profiles of the amplicons were compared with that of reference samples for each exon. Results {#Sec7} ======= Clinical profile and *WT1* mutations in the cohort {#Sec8} -------------------------------------------------- One hundred children with sporadic SRNS (50 females, 50 males) were included in the study (see Table [2](#Tab2){ref-type="table"} for clinical profile). Majority of the patients were from the southern region of India (90%). The median age at the onset of nephrotic syndrome was 2.5 years (IQR 1.2--6.2 years). Parental consanguinity was found in 16% of the families. There were no children with DDS or FS. Among the patients who underwent renal biopsy, focal segmental glomerulosclerosis, minimal change disease and mesangial hyper cellularity was seen in 37, 21 and 12 patients respectively. None of the children biopsied had diffuse mesangial sclerosis (DMS). Most children received multiple immunosuppressants with variable response. Among the children who were recorded to have received calcineurin inhibitors (Cyclosporine and/or Tacrolimus), three had complete response, 11 had partial response and 40 children were calcineurin resistant. Of the 76 children with follow up data, 50 children maintained normal renal function (three complete remission, 47 had persistent proteinuria), 17 developed CKD stages II -- IV, nine progressed to CKD stage V or end stage renal disease (6 received renal transplant and 3 died). The median duration for progression to CKD stage V from onset of disease was 2.7 years. Genomic DNA of all these samples were screened for mutation in exon 8 and 9 of *WT1* gene by Sanger sequencing. The samples did not show any mutation or single nucleotide polymorphism (SNP) in *WT1* gene.Table 2Clinical profile of the cohortCharacteristicPatients (*n*)100 (50females/50 males)Ethnic background (Southern region/Northern region/Eastern region; %)90/9/1Consanguinity (%)12Onset of NS (congenital/infantile/childhood primary SRNS/Unknown; %)4/4/80/12Age at diagnosis^a^ (years; Median, IQR)2.5 (1.2--6.2)Edema74%Proteinuria (yes/unavailable; %)81/19Hematuria (Yes/No/unavailable; %)13/52/35Hypertension (yes/no/unavailable)35/45/20Serum albumin (g/l; Median, Range)1.7 (1.2--2.3)Histopathology subtype (%) Minimal change disease37 Focal segmental glomerulosclerosis21 Mesangial hypercellularity12 Others9 Not performed21Renal outcome (%) Remission4 Persistent relapse62 Chronic Kidney disease Stage II-IV22 End-stage renal disease12^a^incomplete data on 11/100 patients Standardization and validation of HRM assay for identifying normal sequences {#Sec9} ---------------------------------------------------------------------------- The derivative curves, aligned melt curves and the difference plots of *WT1* exon 8 and 9 of 5 healthy control samples used in the standardization run are presented in Fig. [1](#Fig1){ref-type="fig"}. Exon 8 and 9 produced unique melt curves and were visually discriminated from each other and they could be detected in the same PCR run. The melting temperature (standard deviation) of exon 8 and 9 were 78.8 °C (±0.15) and 80.5 (±0.1) respectively. We observed high reproducibility of the melting temperature (Tm) values for a given exon. The melt profiles of *WT1* exon 8 and 9 obtained in the standardization run was in concordance with the melt profile obtained using the nearest neighbor thermodynamic modeling and the Blake and Delcourt algorithm (POLAND software) (Fig. [2](#Fig2){ref-type="fig"}).Fig. 1HRM data for exon 8 and 9 of *WT1* gene (standardization assay). The melt plots show five wild type control samples, which have been previously sequenced exons 8 and 9 . For each exon the Derivative plot (dF/dT) (**a**, **b**), Normalized melt curve (**c**, **d**) and Fluorescence difference plots (Δ Threshold/Temperature) (**e**, **f**) is presented Fig. 2POLAND Melt profile prediction (Blake and Delcourt algorithm) for *WT1*- Exon 8 and 9. The differentiated hypochromicity at 260 and 280 nm (dA/dT) vs. temperature plot is depicted. Theta is the calculated fraction of base pairs remaining in helical state. Both exon 8 and 9 were predicted to melt completely at a single temperature and was in concordance with the experimental melt profiles The method was validated using the five positive mutated sequences obtained by site directed mutagenesis methodology along with three healthy controls (wild type) and three randomly chosen patient samples. All the samples with mutation had a distinctive melting pattern which was visually discriminated when compared with those of the wild-type sequences indicating the ability of the HRM assay to discriminate mutated from wild type samples (Fig. [3](#Fig3){ref-type="fig"}). The HRM profile of patient samples was similar to that of healthy controls which were verified by sequencing.Fig. 3HRM Validation assay showing derivative melt plot, normalized melt curve and difference plots for exon 8 and 9 of the *WT1* gene. In *WT1* exon 8, the wildtype (*blue*) and the blinded patient sample (*maroon*) grouped together indicating that they have the same genotype. The mutant samples c.1079G \> A; p.C360Y (*orange*) and c.1119C \> A; p.H373Q (*green*) were distinguished from the wildtype sample (Fig. 3 **a**, **c** and **e**). Similary in *WT1*exon 9, the wildtype (*green*) and the blinded patient sample (*maroon*) grouped together and the mutant samples c.1180C \> T; p.R394W (*red*), c.1190A \> C; p.H397P (*pink*) and c.A1200C \> T; p.H401Y (*black*) were easily distinguished from the wild type (Fig. 3 **b**, **d** and **f**) To further test the clinical application of this assay, we screened 91 patient samples that had been previously sequenced for Exon 8 and 9 of *WT1* gene with the healthy control samples as reference samples. The analysis of the HRM data was done independently by two observers in the following way: the curves were first auto grouped by the software based on the melt transition. The observers, who were blinded to the identity of the samples, confirmed that the Tm difference obtained between the exon 8 and 9 was same as that obtained in the standardization assay. The heterozygotes if any were to be identified by the shapes of their melting curves. There was no change in the classification by the observers as compared to that grouped by the software. The Tm of the amplicons, exon 8 and 9 were unambiguously distinguished from each other in the derivative plot (Fig. [1a](#Fig1){ref-type="fig"}). The two exons were also readily distinguishable in the aligned melting curve and in the difference plots (amplitude and/or shape of the curves: Fig. [1b and c](#Fig1){ref-type="fig"}). Exons 8 and 9 of all the patient samples had similar melt profiles as that of their respective reference sample and reliably sorted into one of the two distinct groups (Additional file [1](#MOESM1){ref-type="media"}: Figure S1). The melt profile results were identical to the Sanger's sequencing results. The HRM results were obtained in 2 h 15 min, while sequencing required a minimum of 5 days to obtain the final result. The cost analysis calculated in the present conditions showed that conventional PCR followed by Sanger sequencing was three times costly when compared to high-resolution melting assay (Additional file [2](#MOESM2){ref-type="media"}: Table S1). Discussion {#Sec10} ========== SRNS is a genetically heterogeneous disorder with a broad phenotypic spectrum \[[@CR29], [@CR30]\]. Our previous study indicated that the prevalence of Podocin mutation in Indian children with SRNS was low as compared to the Western countries \[[@CR31]\]. The present study from a single centre in India comprising largely of patients from the southern region of the country (90%) was conducted to determine whether *WT1* mutations could be the cause of sporadic SRNS in a subset of patients who did not have Podocin mutation. In our cohort, no subject tested positive for mutation in *WT1*. Based on the number of children tested, the prevalence of WT-1 mutation in Indian children with SRNS with 95% confidence is ≤ 3% \[[@CR32]\]. The prevalence of *WT1* mutation in children with SRNS varies according to the population origin and is summarised in Table [3](#Tab3){ref-type="table"}. Data from large cohort studies that included children with SRNS from various ethnic background like central and west European, Turkish, American, African-American, Hispanic, South Asian, Arabic, or African, shows that the prevalence of *WT1* mutation varies between 1.7 and 9% \[[@CR9], [@CR26], [@CR33]\]. Studies limited to specific ethnic groups also show wide variation in the prevalence of *WT1* mutation with low prevalence in Japanese and Chinese population and high prevalence in Greek, Spanish and Korean population \[[@CR13]--[@CR16], [@CR34], [@CR35]\]. The location and type of WT1 mutations can predict clinical features, histology and renal outcomes. In the largest cohort study of 61 children with WT1 mutations, Lipska et al. observed that while missense mutations were associated with DMS, early onset SRNS and rapid progression to ESRD, the truncating mutations were seen in those who had late onset of SRNS and these children were also at high risk for Wilm's tumor \[[@CR9]\]. In the same study isolated SRNS was associated with intronic (KTS) mutations and slow progression to ESRD. A similar observation was also reported by Chernin et. al., who observed that intronic mutations (KTS) cause NS with slower progression to ESRD compared with missense mutation \[[@CR36]\]. In another retrospective analysis of 53 mid-European patients, those with missense mutations required dialysis significantly earlier that those with truncating mutations \[[@CR37]\]. We could not perform genotype-phenotype correlation due to absence of mutations in our study. The average age of onset of nephrotic syndrome in our cohort is very similar to that reported in other studies suggesting that age cannot explain the low prevalence *WT1* mutation in our cohort \[[@CR14]--[@CR17], [@CR26], [@CR33], [@CR34], [@CR38]\]. As shown in Table [3](#Tab3){ref-type="table"} the presence of syndromic SRNS (DDS or FS or patient with genitourinary tract malformations) can also influence the frequency *WT1* mutation. In the study by Mucha et al. the prevalence of *WT1* mutation decreased from 6. 3 to 3.9% when patients with genitourinary tract malformations were excluded \[[@CR33]\]. Being a single centre study with most of the patients from the southern region, our findings cannot to be generalized to the Indian population. The present cohort comprised of children with only isolated SRNS and none of the patients had DMS on biopsy. The probability of detecting mutations in WT 1 are higher in children with DDS and FS or in those with DMS. In addition, we screened only the hot spot regions of exon 8 and 9 that account for majority of mutations and not the complete open reading frame of WT1 gene. These reasons could potentially explain the absence of *WT1* mutation observed.Table 3Detection rates of *WT1* mutations in children with SRNS in different ethnic groupsEthnic backgroundTotal (Female/Male)Age of Onset (Median; Years)Number of cases with *WT1* mutation (%)Prevalence of *WT1* mutation in isolated SRNS (%)Korean \[[@CR35], [@CR43]\]424 (39/31)4.725 (5.9%)8 (1.9%)Japanese \[[@CR44]\]13 (10/3)0--1 month00Chinese \[[@CR14]\]52 (14/38)3.51 (1.9%)1 (1.9%)Italian \[[@CR17]\]64 (32/32)6.84 (6.2%)2 (3%)Spanish \[[@CR45]\]125 (NA)1.35 (4%)NAGreek \[[@CR16]\]27 (17/10)84 (14.8%)3 (11%)Europe and Middle East \[[@CR9]\]7612.061 (8%)17 (2.23%)Worldwide cohort^a^ \[[@CR33]\]167 (80/87)5.515 (9%)11 (6.6%)^b^Worldwide \[[@CR7]\]2016 (943/1067)3.435 (1.7%)32 (1.6%)Indian (Current study)100 (50/50)2.60 (0%)0 (0%)*NA* not available^a^Central European, Turkish, African-American, Hispanic, or Asian^b^Central Europe, Turkey, and India The heterogenic nature of SRNS makes mutation analysis costly and time consuming. Since the prevalence of the *WT1* mutation is observed to be low, we standardized a simple and economical assay that will be useful in the clinic to identify quickly and accurately subjects with no mutation. We chose HRM assay as a screening tool because it has been found to be a very useful high throughput mutation scanning method in various human diseases \[[@CR39]--[@CR41]\]. The oligo designed for Sanger sequencing were also suitable for HRM analysis, and hence there was no need to design separate set of oligos. Further, the PCR conditions were optimized in such way that both exons can be tested in a single run. Performing simultaneous analysis of the two exons in the same run saves time as well as cost. The melting curves generated in this study allowed unambiguous differentiation of exon 8 and 9 and were highly reproducible. This is the first report wherein a single HRM run can be used for screening of both exons 8 and 9 of *WT1* gene. Our HRM results were in complete concordance with conventional Sanger sequencing suggesting that the HRM gene scanning method can be used to screen for *WT1* gene mutation. In contrast to Sanger sequencing, the use of HRM allows to considerably lower the costs of gene mutation screening in the clinic by reducing the number of samples subjected to Sanger sequencing. Use of this method could reduce costs to 9\$ per sample when compared to Sanger sequencing which is 30\$ per sample, thus proving to be a cost effective method for gene scanning. We also noted that the turnaround time is much lower with HRM. The entire analysis, including the preparation of the reaction mixture, amplification and melting of the products takes a little more than 2 h and interpretation of the result takes 30 min (Additional file [2](#MOESM2){ref-type="media"}: Table S1). Further, HRM analysis require less manpower as single technician can perform the test and screen large number of samples in a single assay. Since the assay only requires real time PCR equipment, clinical laboratories can initiate similar screening strategy using our standardized protocol. The same technique could be used in screening other genes and risk alleles associated with SRNS in a similar way. The *APOL1* G1 and G2 risk alleles in the African American population with FSGS is significantly enriched as compared to the general population. HRM will be an ideal technique to rapidly screen for these alleles in our population \[[@CR42]\]. In summary, no pathogenic variant in exon 8 and 9 of *WT1* gene was found in the present cohort. We developed a simple, high throughput approach for identification of normal variants in *WT1* gene which is six times faster and three times less cost compared with traditional direct sequencing. The results obtained by HRM on the clinical samples further support the feasibility of applying the assay as a screening technique in a clinical setting. Additional files {#Sec12} ================ Additional file 1: Figure S1.HRM analysis of cohort for exon 8 and 9 of *WT1* gene. The melt plots show wild type control samples and representative patient samples, which have been previously sequenced exons 8 and 9 . For each exon the Derivative plot (dF/dT) (A,B), Normalized melt curve (C, D) and Fluorescence difference plots (Δ Threshold/Temperature) (E, F) is presented. (TIF 3577 kb) Additional file 2: Table S1.Summary of the properties of High Resolution Melt and Direct sequencing. (DOCX 12 kb) CKD : Chronic kidney disease DDS : Denys-Drash syndrome DHPLC : Denaturing high performance liquid chromatography DMS : Diffuse mesangial sclerosis ESRD : End-stage renal disease FMPA : Fluorescent multiplexed-PCR analysis FS : Frasier syndrome HRM : High resolution melt NS : Nephrotic syndrome PCR : Polymerase chain reaction SNP : Single nucleotide polymorphism SRNS : Steroid resistant nephrotic syndrome Tm : Melting temperature *WT1* : Wilm's tumor 1 We thank all the patients who participated in the study. We also thank Dr. Jose Remacle for his valuable suggestions. Funding {#FPar1} ======= This work was supported by a grant from Council of Scientific & Industrial Research (CSIR), India, (No.27 (0288)/13/EMR-II). We also acknowledge Indian Council of Medical Research (3/1/2/6-RCH; IRIS ID No.2012-26950) for their support. Availability of data and materials {#FPar2} ================================== All data reported in this study is available from Dr Anil Vasudevan. Authors' contribution {#FPar3} ===================== AS was involved in patient recruitment, designed primers, performed the assay, analysed data and wrote the manuscript. VCP was involved in manuscript preparation. SR and ANV were involved in patient recruitment^,^ prepared the blinded samples. AV originated the study, reviewed data analysis and co-wrote 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} ======================= Written informed consent was obtained from the patient for publication. A copy of the written consent is available for review by the editor. Ethics approval and consent to participate {#FPar6} ========================================== The study has been approved by the institutional ethics committee (St. John's National Academy of Health Sciences Institutional Ethics committee, Bangalore, India, reference no 171/2012). Written informed consent was obtained from all patients.

ANNOUNCEMENT {#s1} ============ Nontyphoidal *Salmonella* is one of the leading causes of human foodborne gastroenteritis in North America, with public health costs of approximately \$695 million annually in the United States ([@B1]). According to the National Outbreak Reporting System (NORS) database (Centers for Disease Control and Prevention), 38% of all salmonellosis outbreaks recorded in the United States between 2012 and 2016 were associated with poultry (<https://wwwn.cdc.gov/norsdashboard/>). Salmonella enterica subsp*. enterica* serotype Heidelberg is the most common serotype isolated in all poultry breeder types and throughout all levels of the production chain in the United States and Canada ([@B2][@B3][@B4]). Rapid adaptation of Salmonella to the poultry host and the farm environment where the birds are raised reduces the antimicrobial efficacy of control methods implemented by the poultry industry ([@B5]). However, little is known about the impact of the farm environment on the genome content of Salmonella spp., which may facilitate adaptation to different environments. To this end, we sequenced the genomes of 31 Salmonella Heidelberg isolated from environmental samples obtained from chicken or turkey farms ([@B6]). Nineteen and 12 isolates were collected from environmental booties of 16 chicken and 8 turkey breeder farms, respectively, and 2 isolates were collected from hatchery debris in 2 different turkey farms ([Table 1](#tab1){ref-type="table"}). For environmental bootie samples, fabric shoe covers premoistened with skim milk were used and walked on the floor inside the farm over a distance of 305 m. Samples were collected under the supervision of the Minnesota Board of Animal Health between April and July 2015 as part of the National Poultry Improvement Plan (NPIP) Salmonella monitoring program. Salmonella Heidelberg was isolated on a brilliant green agar with novobiocin, xylose lysine tergitol-4, and Millier-Mallinson medium. The identities of the 31 isolates were further confirmed by a PCR assay using one set of primers specific to the Salmonella genus (for OMPC, ATCGCTGACTTATGCAATCG \[forward\] and CGGGTTGCGTTATAGGTCTG \[reverse\]; amplicon length, 204 bp) and another set specific to the Salmonella. Heidelberg serotype (for GenBank accession number [ACF69659](https://www.ncbi.nlm.nih.gov/protein/ACF69659), TGTTTGGAGCATCATCAGAA \[forward\] and GCTCAACATAAGGGAAGCAA \[reverse\]; amplicon length, 216 bp) and compared to a known Salmonella Heidelberg strain ([@B7][@B8][@B9]). ###### Metadata of the 31 *Salmonella enterica* subsp. *enterica* serotype Heidelberg isolates from chicken and turkey farm environments ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Isolate Sample origin Sample type NCBI genome\ NCBI SRX\ No. of\ Contig *N*~50~[^*a*^](#ngtab1.1){ref-type="table-fn"} GC content\ Total no. of\ accession no. accession no. contigs (%) sequencing reads ---------- --------------- ---------------------- ------------------------------------------------------------------- ----------------------------------------------------------- --------- ------------------------------------------------------- ------------- ------------------ C_NS001 Chicken Environmental bootie [PGWO00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWO00000000) [SRX3441541](https://www.ncbi.nlm.nih.gov/sra/SRX3441541) 53 412,084 51.5 948,617 C_NS002 Chicken Environmental bootie [PGWN00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWN00000000) [SRX3441539](https://www.ncbi.nlm.nih.gov/sra/SRX3441539) 51 412,084 51.5 1,025,661 C_NS003 Chicken Environmental bootie [PGWM00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWM00000000) [SRX3441540](https://www.ncbi.nlm.nih.gov/sra/SRX3441540) 52 412,084 51.8 838,953 C_NS004 Chicken Environmental bootie [PGWL00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWL00000000) [SRX3441522](https://www.ncbi.nlm.nih.gov/sra/SRX3441522) 71 412,084 51.8 622,170 C_NS005 Chicken Environmental bootie [PGWK00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWK00000000) [SRX3441521](https://www.ncbi.nlm.nih.gov/sra/SRX3441521) 93 412,084 51.7 491,774 C_NS006 Chicken Environmental bootie [PGWJ00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWJ00000000) [SRX3441528](https://www.ncbi.nlm.nih.gov/sra/SRX3441528) 52 291,725 52.3 715,859 C_NS007 Chicken Environmental bootie [PGWI00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWI00000000) [SRX3441527](https://www.ncbi.nlm.nih.gov/sra/SRX3441527) 76 144,340 52.5 906,054 C_NS008 Chicken Environmental bootie [PGWH00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWH00000000) [SRX3441530](https://www.ncbi.nlm.nih.gov/sra/SRX3441530) 49 411,223 52.2 1,038,062 C_NS009 Chicken Environmental bootie [PGWG00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWG00000000) [SRX3441529](https://www.ncbi.nlm.nih.gov/sra/SRX3441529) 65 291,730 52 1,006,172 C_NS010 Chicken Environmental bootie [PGWF00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWF00000000) [SRX3441524](https://www.ncbi.nlm.nih.gov/sra/SRX3441524) 74 184,236 52.2 826,444 C_NS011 Chicken Environmental bootie [PGWE00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWE00000000) [SRX3441523](https://www.ncbi.nlm.nih.gov/sra/SRX3441523) 63 276,222 51.4 796,049 C_NS020 Chicken Environmental bootie [PGWD00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWD00000000) [SRX3441526](https://www.ncbi.nlm.nih.gov/sra/SRX3441526) 71 412,018 52.9 667,956 C_NS024 Chicken Environmental bootie [PGWC00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWC00000000) [SRX3441525](https://www.ncbi.nlm.nih.gov/sra/SRX3441525) 88 203,724 51.5 1,151,538 C_NS025 Chicken Environmental bootie [PGWB00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWB00000000) [SRX3441542](https://www.ncbi.nlm.nih.gov/sra/SRX3441542) 58 411,826 51.3 887,634 C_NS026 Chicken Environmental bootie [PGWA00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGWA00000000) [SRX3441543](https://www.ncbi.nlm.nih.gov/sra/SRX3441543) 53 411,820 52.2 769,371 C_NS027 Chicken Environmental bootie [PGVZ00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVZ00000000) [SRX3441531](https://www.ncbi.nlm.nih.gov/sra/SRX3441531) 51 411,985 52.2 761,794 C_NS028 Chicken Environmental bootie [PGVY00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVY00000000) [SRX3441532](https://www.ncbi.nlm.nih.gov/sra/SRX3441532) 71 235,527 52.1 937,797 C_NS029 Chicken Environmental bootie [PGVX00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVX00000000) [SRX3441533](https://www.ncbi.nlm.nih.gov/sra/SRX3441533) 95 170,930 52.1 787,619 C_NS030 Chicken Environmental bootie [PGVW00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVW00000000) [SRX3441534](https://www.ncbi.nlm.nih.gov/sra/SRX3441534) 68 291,725 52.2 646,325 T_NS-012 Turkey Environmental bootie [PGVV00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVV00000000) [SRX3441535](https://www.ncbi.nlm.nih.gov/sra/SRX3441535) 70 248,810 52.2 884,633 T_NS-013 Turkey Environmental bootie [PGVU00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVU00000000) [SRX3441536](https://www.ncbi.nlm.nih.gov/sra/SRX3441536) 56 424,424 52.5 1,043,609 T_NS-014 Turkey Environmental bootie [PGVT00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVT00000000) [SRX3441537](https://www.ncbi.nlm.nih.gov/sra/SRX3441537) 78 231,397 52.4 883,270 T_NS-015 Turkey Environmental bootie [PGVS00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVS00000000) [SRX3441538](https://www.ncbi.nlm.nih.gov/sra/SRX3441538) 88 231,397 52.4 972,069 T_NS-016 Turkey Hatchery debris [PGVR00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVR00000000) [SRX3441519](https://www.ncbi.nlm.nih.gov/sra/SRX3441519) 71 291,727 52.6 1,099,328 T_NS-017 Turkey Environmental bootie [PGVQ00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVQ00000000) [SRX3441520](https://www.ncbi.nlm.nih.gov/sra/SRX3441520) 72 233,820 51.7 1,128,127 T_NS-018 Turkey Environmental bootie [PGVP00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVP00000000) [SRX3441515](https://www.ncbi.nlm.nih.gov/sra/SRX3441515) 60 423,224 52.1 920,022 T_NS-019 Turkey Hatchery debris [PGVO00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVO00000000) [SRX3441516](https://www.ncbi.nlm.nih.gov/sra/SRX3441516) 102 298,617 51.9 611,230 T_NS-031 Turkey Environmental bootie [PGVL00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVL00000000) [SRX3441511](https://www.ncbi.nlm.nih.gov/sra/SRX3441511) 78 240,110 53 775,776 T_NS-032 Turkey Environmental bootie [PGVK00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVK00000000) [SRX3441512](https://www.ncbi.nlm.nih.gov/sra/SRX3441512) 100 175,042 53 1,139,971 T_NS-033 Turkey Environmental bootie [PGVJ00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVJ00000000) [SRX3441513](https://www.ncbi.nlm.nih.gov/sra/SRX3441513) 94 154,373 52.9 1,511,607 T_NS-034 Turkey Environmental bootie [PGVI00000000](https://www.ncbi.nlm.nih.gov/nuccore/PGVI00000000) [SRX3441514](https://www.ncbi.nlm.nih.gov/sra/SRX3441514) 153 82,339 52.6 1,356,089 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Sequence length of the shortest contig at 50% of the total genome length. Genomic DNA was extracted using an E.Z.N.A. bacterial DNA kit (Omega Bio-tek, Norcross, GA). Sequencing libraries were prepared using 1 ng of genomic DNA with a Nextera XT DNA sample prep kit (Illumina, Inc., San Diego, CA). PCR amplification was done in a Veriti 96-well thermal cycler machine (Thermo Fisher Scientific, Waltham, MA) using Nextera XT Index 1 primers (N7XX) from the Nextera XT index kit (FC131‐1001). Illumina paired-end sequencing was performed on a MiSeq platform using 2 × 250 paired-end sequencing chemistry. The raw data files were demultiplexed and converted to FASTQ files using Casava v.1.8.2. (Illumina, Inc., San Diego, CA). Quality control of the raw reads was performed using FastQC (Babraham Bioinformatics, Cambridge, MA). The reads were trimmed and assembled with BBDuk (DOE Joint Genome Institute, Walnut Creek, CA) and SPAdes (SPBU, St. Petersburg, Russia), respectively. The coverages of the assembled genomes were evaluated with BBMap (DOE Joint Genome Institute, Walnut Creek, CA). The sizes of the genomes ranged between 4,766,150 and 5,104,082 bp, their coverage values ranged between 46.52× and 135.54×, and the numbers of contigs per assembly ranged between 62 and 174. Significant differences in the nucleotides and gene contents were detected between the isolates obtained from turkey environments and those obtained from chicken farm environments. Therefore, the availability of these genome sequences may facilitate understanding of *Salmonella* adaptation to different niches in poultry production ([@B6]). Data availability. {#s1.1} ------------------ Draft genomic sequences have been deposited in the NCBI Sequence Read Archive (SRA accession number [SRP126070](https://www.ncbi.nlm.nih.gov/sra/SRP126070)) and NCBI GenBank (BioProject number [PRJNA417775](https://www.ncbi.nlm.nih.gov/bioproject/PRJNA417775)) and are listed in [Table 1](#tab1){ref-type="table"}. We thank Saranga Wijeratne and Benjamin Lorentz for assistance with genome analysis. We thank Kakambi V. Nagaraja, Muhammad Nisar, Dale Lauer, and Shauna Voss for providing S. Heidelberg isolates and The Ohio State University for providing assistance with sequence analysis. Research in the Rajashekara laboratory is supported by funds from the National Institute of Food and Agriculture, a United States Department of Agriculture grant, and the Ohio Agricultural Research and Development Center. [^1]: **Citation** Deblais L, Scaria J, Rajashekara G. 2018. Draft genome sequences of *Salmonella enterica* subsp. *enterica* serotype Heidelberg from chicken and turkey farm environments. Microbiol Resour Announc 7:e01204-18. <https://doi.org/10.1128/MRA.01204-18>.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ In recent years, a lot of attention has been given to the victimization that same-sex attracted middle school- and high school-aged adolescents face because of its associations with suicidal behavior and substance use \[[@pone.0129976.ref001],[@pone.0129976.ref002]\]. This has resulted in researchers, educators, and policy makers emphasizing same-sex attraction as a key risk factor for being victimized and experiencing negative health outcomes \[[@pone.0129976.ref003]--[@pone.0129976.ref006]\]. While the attention given to same-sex attracted adolescents is warranted given the elevated risk of experiencing negative health outcomes for sexual minority adolescents \[[@pone.0129976.ref001],[@pone.0129976.ref007]\], same-sex attracted adolescents have heterogeneous experiences with being victimized. Work exploring victimization among same-sex attracted adolescents has demonstrated that they may be targeted for both sexual orientation-based victimization (victimization based on individuals' perceived or actual sexual orientation) and gender-variant-based victimization (victimization based on the way others perceive individuals to convey masculine, feminine, and androgynous characteristics through their appearance, mannerisms, and behaviors), with same-sex attracted adolescents who are targeted for gender-variant-based victimization experiencing more negative health outcomes than same-sex adolescents who are not targeted for this type of victimization \[[@pone.0129976.ref008],[@pone.0129976.ref009]\]. However, in all of this, little work on gender-variant-based victimization has addressed the outcomes experienced by other-sex attracted students who also experience this type of victimization, but are not part of the sexual- or gender-minority (i.e., other-sex attracted students who are not transgender, and who do not identify as gender nonconforming). To address this limitation, the current work explores the association between gender-variant-based victimization in other-sex attracted adolescents and health-risking suicidal behaviors and substance use. Additionally, to help put the impact of this type of victimization into perspective, this work separates out the effects of attraction and gender-variant-based victimization to allow for comparisons between other-sex attracted adolescents and same-sex attracted adolescents who experience gender-variant-based victimization. Being Targeted for Gender-Variant-Based Victimization {#sec002} ----------------------------------------------------- Adolescents may be targeted for victimization by their peers because they are perceived to be violating gender norms. This type of victimization is a form of gender policing, which uses victimization to rebuke both male and female peers who do not conform to traditional gender norms \[[@pone.0129976.ref010],[@pone.0129976.ref011]\]. In recent years, much of the work exploring the association between gender variance and negative outcomes has focused on sexual-minority adolescents because identifying as same-sex attracted is viewed as a violation of traditional gender norms \[[@pone.0129976.ref008],[@pone.0129976.ref009]\]. However, any adolescents that are not perceived as expressing their gender in ways that meet prescribed social norms are potential targets for this type of victimization \[[@pone.0129976.ref011]\]. Indeed, research has shown that transgender and gender non-conforming adolescents, regardless of sexual orientation, experience high rates of victimization \[[@pone.0129976.ref009],[@pone.0129976.ref012],[@pone.0129976.ref013],[@pone.0129976.ref014]\], including high rates of sexual orientation-based and gender-variant-based victimization \[[@pone.0129976.ref015],[@pone.0129976.ref016]\] even though the adolescents may be other-sex attracted. However, this work is limited by focusing on adolescents who are gender non-conforming, rather than exploring the consequences of any adolescent being victimized for being perceived as gender-variant. Examining the association between gender and attraction best highlights the potential for gender-variant-based victimization to impact all adolescents, regardless of same- or other-sex attraction. Gender norms in the United States, and in many places in the world, are heteronormative in that they are based on an assumption that all typical people should want to attract an opposite sex romantic/sexual partner (e.g., males need to behave in ways consistent with masculine stereotypes so that they can attract female romantic/sexual partners) \[[@pone.0129976.ref017]\]. Thus, being in any way perceived as same-sex attracted is automatically viewed a violation of expected gender norms. However, there are many aspects of a person including self-presentation, interests, and behaviors that can be judged to violate expected gender norms \[[@pone.0129976.ref018],[@pone.0129976.ref019]\]. For example, an other-sex attracted female adolescent who plays sports, expresses an interest in being a welder, and prefers to have very short hair would likely be viewed by her peers as violating expected gender norms and would likely be targeted for gender-variant-based victimization, regardless of whether she is attracted to males. If a same-sex attracted male adolescent has the exact same profile of behavior, interests, and self-presentation, his peers would view him as conforming to expected gender norms and would likely not target him for gender-variant-based victimization, as long as they do not find out that he is attracted to males. Therefore, there is nothing about simply being same-sex attracted that would make an adolescent a target for gender-variant-based victimization, but rather it is being publicly identified as being same-sex attracted that alerts his or her peers to a violation of expected gender norms. This assertion is supported by the previous work suggesting that same-sex attracted adolescents are not equally targeted for gender-variation based victimization. Same-sex attracted adolescents who most fit the prescribed feminine gay or masculine lesbian stereotypes report the highest levels of victimization \[[@pone.0129976.ref008],[@pone.0129976.ref009]\]. Thus, being attracted to the same-sex may be associated with gender-variant-based victimization based on the self-presentation or self-disclosure of some same-sex attracted people, but it is not the only factor that drives this type of victimization. Thus, there may be a large number of youth who experience gender-variant-based victimization who are not identified in research or reached through interventions that target sexual-minority adolescents. The Impact of Gender-Variant-Based Victimization {#sec003} ------------------------------------------------ Gender-variant based victimization has a large impact on adolescents because gender plays a central role in their sense of self and well-being. For example, it has been found that both males and females who feel that they are gender typical are more likely to have a reduced sense of loneliness \[[@pone.0129976.ref018]\], higher sense of self-worth \[[@pone.0129976.ref018],[@pone.0129976.ref019]\], and better overall psychological adjustment \[[@pone.0129976.ref018]\] compared to those who feel that they are perceived as not expressing their gender appropriately. Importantly, these effects are not based on objective assessments of gender typicality, but rather on how individuals feel like their peers perceive them. Work focused on males has shown that males who are perceived to express more feminine, or gender atypical, traits than their peers report higher levels of victimization, loneliness, and distress \[[@pone.0129976.ref020]\]. This results in adolescent males' feeling the need to constantly monitor their self-presentation to defend themselves from being victimized for being perceived as having feminine qualities or not being masculine enough \[[@pone.0129976.ref016]\]. Thus, managing other peoples' impressions of ones' gender expression becomes a constant part of navigating social interactions. The connection between gender-variant-based victimization and suicidal behaviors and substance use can be understood through the Minority Stress Model \[[@pone.0129976.ref021],[@pone.0129976.ref022]\] and the Interpersonal Theory of Suicide \[[@pone.0129976.ref023]\]. The Minority Stress Model posits that mental health issues and self-medication via substance use can result from the stress people experience when faced with a hostile social environment and social exclusion due to being part of a minority group. While this model has most often been used with traditional minority groups such as sexual minorities \[[@pone.0129976.ref021],[@pone.0129976.ref022]\], the key component of this model is that people feel that they are being targeted for an aspect of who they are, and that who they are in someway separates them from the majority group. Thus, an other-sex attracted student who is victimized because some of her peers do not think she acts feminine enough may feel excluded from the majority group because of who she is and how she normally expresses herself. As outlined in a similar theoretical framework connecting the high risk of suicide and substance use among trans people \[[@pone.0129976.ref024]\], this feeling of social exclusion can result in high levels of stress, which people may try to cope with through substance use. Additionally, this feeling of social exclusion can result in feelings of social isolation, which the Interpersonal Theory of Suicide suggests is associated with being at higher risk for expressing suicidal behaviors \[[@pone.0129976.ref023],[@pone.0129976.ref024]\]. Thus, simply experiencing gender-variant-based victimization, a type of victimization directed at people who are seen as not conforming to the majority expectations for gender expression, may result in feelings of social exclusion, which has potential ramifications for any adolescents' well-being. The Current Study {#sec004} ----------------- Much of the previous work exploring gender-variant-based victimization has focused on adolescents who were identified as being part of a sexual- or gender-minority group \[[@pone.0129976.ref001]\]. The current study takes a broader approach to exploring gender-variant-based victimization by focusing on the outcomes of adolescents who report any amount of victimization based on their gender expression, regardless of attraction. To highlight the point that this type of victimization is not just an issue for a subset of youth based on their same-sex attraction, the current study separates out attraction and gender-variation-based victimization as being independently associated with suicidal behavior and substance use. Additionally, the present study compares the association of gender-variant-based victimization with suicidal behavior and substance use between other-sex and same-sex attracted adolescents. In sum, this study addresses two novel research questions. First, is gender-variant-based victimization related to suicidal behavior and substance use among other-sex attracted adolescents? Second, is the relationship between gender-variant-based victimization and these health-risking behaviors similar between same- and other-sex attracted adolescents? Based on the above review highlighting the negative effects of gender-variant-based victimization, three hypotheses were proposed and examined: Other-sex attracted students who experience gender-variant-based victimization will be more likely to report suicidal behaviors and substance use, compared to other-sex attracted students who do not experience gender-variant-based victimization.Attraction (i.e., same-sex as compared to other-sex) and experiencing gender-variant-based victimization will both independently be associated with higher levels of suicidal behaviors and substance use.The relationship between gender-variant-based victimization and these health-risking behaviors will not differ as a function of attraction (i.e., same-sex as compared to other-sex). Method {#sec005} ====== Procedures and Ethics Statement {#sec006} ------------------------------- Students completed an anonymous, 30-minute online survey under teacher/staff supervision during school hours in May 2012 as part of a district needs assessment. As part of the district needs assessment process, the school administrators ensured that parents/guardians were: informed of the study, could review the survey, and could opt their child out of participation, in writing, prior to survey administration (fewer than 1% withdrew). As per standard practice in United States School districts, choosing not to opt their child out of participation was taken as parent/guardian consent for their child to participate in the needs assessment. The students themselves were not formally asked to consent to participate in the needs assessment, but they were informed that they could opt out of participating at any point by ceasing to respond to questions and were only required to provide their biological sex and school attended to complete the survey (93% of students participated). The Colorado State University Institutional Review Board approved the procedures used to attain consent and implement the needs assessment, and approved the use of the anonymous data for the purposes of research. This work was deemed as exempt from the requirements of human subject protections by the Colorado State University Institutional Review Board based on the grounds that the research was being conducted in "established or commonly accepted education settings, involving normal education practices, such as a) research on regular and special education strategies, or b) research on the effectiveness of or the comparison among instructional techniques, curricula, or classroom management methods." Measures {#sec007} -------- ### Gender-variant-based victimization {#sec008} Based in part on previous work \[[@pone.0129976.ref025]\], males were asked, "*Do kids ever make fun of you for not being masculine enough/too feminine*?", and females were asked, "*Do kids ever make fun of you for not being feminine enough/too masculine*?" Participants responded using a Likert-scale (1 = *Never*, 2 = *Sometimes*, 3 = *Often*, 4 = *Always*). A dichotomous variable was created (having never experienced gender-variant-based victimization (0) vs. experiencing gender-variant-based victimization at least sometimes or more often (1). ### Attraction {#sec009} Attraction to members of the same- and other-sex was assessed using a single item (*When you are older*, *who might you be interested in dating*? *If you are already dating*, *who are you interested in dating*?), with participants selecting each option that applied (*boys*, *girls*, *or not sure*). A dichotomous variable was created, i.e., participants indicating any same-sex attraction were categorized as same-sex attracted (1), whereas participants indicating only other-sex attraction were categorized as other-sex attracted (0). All participants (*n* = 36), who selected *not sure*, also selected at least one other option allowing for categorization as either same-sex (*n* = 20) or other-sex attracted (*n* = 16). ### Suicidal ideation {#sec010} Suicidal ideation during the previous week was assessed using 4 items (α = .87; e.g., *I had thoughts about death*, *I felt that I would kill myself if I knew a way*) \[[@pone.0129976.ref026]\]. Participants responded to all items using a 4-point scale (1 = *Rarely or none of the time (less than 1 day)*, 2 = *(1--2 days)*, 3 = *(3--4 days)*, 4 = *Most or all of the time (5--7 days)*). A dichotomous variable was created so that all of the participants who reported a 1 on all 4 items were coded as 0, and all participants who reported a 2 or higher on at least 1 of the 4 items were coded as 1. ### Suicide planning {#sec011} Suicide planning since the beginning of the school year (past 35--38 weeks) was assessed using 2 items (α = .74; *Since the beginning of the school year have you ever seriously considered attempting suicide (i*.*e*., *tried to kill yourself)*? *and Since the beginning of the school year have you made a plan about how you would attempt suicide*?). Participants responded to each question with either a *Yes* (1) or a *No* (0). A dichotomous composite variable was created by combing the responses to the two questions, i.e., all of the participants who reported a 0 on both items were coded as 0, while all participants who reported a 1 on at least 1 of the 2 items were coded as 1. ### Suicide attempts {#sec012} Suicide attempts were assessed using a single item (*Since the beginning of the school year*, *how many times have you actually attempted suicide*?). Participants responded to this item using a 4-point scale (1 = *0 times*, 2 = *1 time*, 3 = *2 or 3 times*, 4 = *4 or more times*). A dichotomous variable was created so that no suicide attempts since the beginning of the school year were indicated with 0; any suicide attempts since the beginning of the school year were indicated with 1. ### Substance use {#sec013} Alcohol use (*How often in the past 3 months have you*: *been drunk*?), marijuana use (*How often in the past 3 months have you*: *used marijuana (pot*, *hash*, *reefer)*?), and other drug use (α = .88; *How often in the past 3 months have you*: *used meth*?, and *How often in the past 3 months have you*: *used other drugs*?) were each assessed. Participants responded to each item using an 8-point scale (1 = *Never*, 2 = *Once or twice*, 3 = *About once per month*, 4 = *A few times a month*, 5 = *About once a week*, 6 = *A few times a week*, 7 = *Almost every day*, 8 = *More than once a day*). A dichotomous variable for each type of substance use was created, i.e., never drunk in the past 3 months (0) vs. drunk at any point during the past 3 months (1). Regular cigarette smoking (*How often in the past 3 months have you*: *smoked cigarettes*?) was also assessed using the same 8-point scale. A dichotomous variable was created i.e., did not smoke cigarettes regularly in the past 3 months (0; 1 = *Never*, 2 = *Once or twice*, 3 = *About once per month*, 4 = *A few times a month*, 5 = *About once a week*, 6 = *A few times a week*) vs. did smoke cigarettes regularly in the past 3 months (1; 7 = *Almost every day*, 8 = *More than once a day*). Analysis {#sec014} -------- All models were tested using a logistic regression model in SAS, Version 9.3. For hypothesis 3, we estimated both the multiplicative and additive interaction (i.e., the Relative Excess Risk due to Interaction (RERI)). In logistic regression, the logit coefficient associated with a product of two variables (x\*z, for example, bullying based on gender expression and attraction) is an estimate of interaction on a multiplicative scale and a significant interaction denotes that the combined effect of x and z is greater than the *product* of the individual effects of x and z \[[@pone.0129976.ref027]\]. It has been argued that assessment of the interaction on an additive scale when the outcome is categorical is also useful, particularly if assessment of the interaction has public health relevance \[[@pone.0129976.ref027],[@pone.0129976.ref028]\]. Assessment of the additive interaction based on a logistic regression model is commonly done through the calculation of the RERI. A significant RERI denotes that the combined effect of x and z is greater than the *sum* of the individual effects of x and z. For hypotheses 2 and 3, we report the OR (i.e., exp(*B*)) and 95% CI for all models, a 95% CI that does not contain 1 indicates statistical significance, *p*\<.05. For the RERI estimates for hypothesis 3, we also report the 95% CI calculated using the delta method \[[@pone.0129976.ref028]\], for these, a 95% CI that does not contain 0 indicates statistical significance, *p*\<.05. Results {#sec015} ======= Sociodemograph Characteristics and Preliminary Analyses {#sec016} ------------------------------------------------------- Participants were from one school district in the Northeastern United States. A total of 2,944 students started the survey; but only 2,438 (82.8%) completed both the attraction and gender-variant-based victimization questions. Approximately half (50.1%) of the participants were female. The median age was 15 with 51.8% of the students attending one of two middle schools, and the remainder attending the high school. Approximately half were eligible for free/reduced lunch at the time of the survey. Most students, based on school records, were either Black (37%) or Hispanic (35%); about one-quarter of the students (24%) were White, non-Hispanic. Individuals who reported interest in dating a member of the same sex (35 males, 33 females), or both sexes (27 males, 80 females), were categorized as same-sex attracted. This categorization is supported by the similarity between the proportion of both- and same-sex attracted students reporting health-risking behaviors, in conjunction with the dissimilarity between the proportions of both- and other-sex attracted students reporting the same behaviors. Tables [1](#pone.0129976.t001){ref-type="table"} and [2](#pone.0129976.t002){ref-type="table"} present the sample size, sample proportions, and prevalence of each outcome as a function of attraction. Tables [3](#pone.0129976.t003){ref-type="table"} and [4](#pone.0129976.t004){ref-type="table"} present the sample size, sample proportions, and prevalence of each outcome as a function of sex for each attraction group. 10.1371/journal.pone.0129976.t001 ###### Prevalence of suicidal and substance using behaviors as a function of attraction, separating out both-sex attracted students. {#pone.0129976.t001g} Other-Sex Attracted Only Same-Sex Attracted Only Both-Sex Attracted --------------------------- -------------------------- ------------------------- --------------------------------------------------------------------------------------- *n* 2263 (92.8%) 68 (2.8%) 107 (4.4%) Health-risking behavior Suicidal ideation 0.24 0.40 0.44[^a^](#t001fn001){ref-type="table-fn"} Suicide attempt 0.07 0.32 0.27[^a^](#t001fn001){ref-type="table-fn"} Suicide planning 0.12 0.32 0.45[^a^](#t001fn001){ref-type="table-fn"} Alcohol intoxication 0.22 0.07 0.51[^a^](#t001fn001){ref-type="table-fn"} Regular cigarette smoking 0.03 0.05 0.19[^a^](#t001fn001){ref-type="table-fn"} ^,^ [^b^](#t001fn002){ref-type="table-fn"} Marijuana use 0.16 0.29 0.51[^a^](#t001fn001){ref-type="table-fn"} ^,^ [^b^](#t001fn002){ref-type="table-fn"} Other drug use 0.06 0.08 0.25[^a^](#t001fn001){ref-type="table-fn"} ^a^Denotes a significant difference in the proportion of the health-risking behavior compared with students who are other-sex attracted only. ^b^Denotes a significant difference in the proportion of the health-risking behavior compared with students who are same-sex attracted only. 10.1371/journal.pone.0129976.t002 ###### Prevalence of suicidal and substance using behaviors as a function of attraction. {#pone.0129976.t002g} Other-Sex Attracted Same-Sex Attracted --------------------------- --------------------- -------------------------------------------- *n* 2263 (92.8%) 175 (7.2%) Health-risking behavior Suicidal ideation 0.24 0.42[^a^](#t002fn001){ref-type="table-fn"} Suicide attempt 0.07 0.29[^a^](#t002fn001){ref-type="table-fn"} Suicide planning 0.12 0.40[^a^](#t002fn001){ref-type="table-fn"} Alcohol intoxication 0.22 0.50[^a^](#t002fn001){ref-type="table-fn"} Regular cigarette smoking 0.03 0.14[^a^](#t002fn001){ref-type="table-fn"} Marijuana use 0.16 0.43[^a^](#t002fn001){ref-type="table-fn"} Other drug use 0.06 0.24[^a^](#t002fn001){ref-type="table-fn"} ^a^Denotes a significant difference in the proportion of the health-risking behavior compared with students who are other-sex attracted. 10.1371/journal.pone.0129976.t003 ###### Prevalence of suicidal and substance using behaviors as a function of sex for same-sex attracted students. {#pone.0129976.t003g} Males Females --------------------------- ------------ ------------- *n* 62 (35.4%) 113 (64.6%) Health-risking behavior Suicidal ideation 0.44 0.42 Suicide attempt 0.34 0.27 Suicide planning 0.39 0.41 Alcohol intoxication 0.44 0.53 Regular cigarette smoking 0.18 0.12 Marijuana use 0.36 0.47 Other drug use 0.32 0.20 ^a^Denotes a significant difference in the proportion of the health-risking behavior compared with students who are males. 10.1371/journal.pone.0129976.t004 ###### Prevalence of suicidal and substance using behaviors as a function of sex for other-sex attracted students. {#pone.0129976.t004g} Males Females --------------------------- -------------- -------------------------------------------- *n* 1154 (51.0%) 1109 (40.0%) Health-risking behavior Suicidal ideation 0.20 0.27[^a^](#t004fn001){ref-type="table-fn"} Suicide attempt 0.08 0.06[^a^](#t004fn001){ref-type="table-fn"} Suicide planning 0.09 0.16[^a^](#t004fn001){ref-type="table-fn"} Alcohol intoxication 0.23 0.20 Regular cigarette smoking 0.05 0.01[^a^](#t004fn001){ref-type="table-fn"} Marijuana use 0.19 0.12[^a^](#t004fn001){ref-type="table-fn"} Other drug use 0.07 0.04[^a^](#t004fn001){ref-type="table-fn"} ^a^Denotes a significant difference in the proportion of the health-risking behavior compared with students who are males. Overall Descriptives {#sec017} -------------------- [Table 5](#pone.0129976.t005){ref-type="table"} presents the sample size, sample proportions, and prevalence of each outcome as a function of attraction (1 = same-sex, 0 = other-sex) and experience with gender-variant-based victimization (1 = experienced gender-variant-based victimization, 0 = did not experience gender-variant-based victimization). Within each attraction group, we indicate whether the prevalence of each health-risking behavior is significantly different between students who report gender-variant-based victimization and those who do not. 10.1371/journal.pone.0129976.t005 ###### Prevalence of suicidal and substance using behaviors as a function of attraction and gender-variant-based victimization. {#pone.0129976.t005g} Other-Sex Attracted Students Same-Sex Attracted Students --------------------------- ------------------------------ --------------------------------------------------- -------------------------------------------------- ------------- ------------ -------------------------------------------------- *n* 1876 (82.9%) 387 (17.1%) 108 (61.7%) 67 (38.3%) Females 941 (84.9%) 168 (15.1%)[^a^](#t005fn001){ref-type="table-fn"} 74 (65.5%) 39 (34.5%) Males 935 (81.0%) 219 (19.0%)[^a^](#t005fn001){ref-type="table-fn"} 34 (55.8%) 28 (45.2%) Significant Significant Health-risking behavior Difference[^b^](#t005fn002){ref-type="table-fn"} Difference[^b^](#t005fn002){ref-type="table-fn"} Suicidal ideation 0.20 0.41 *p*\<.05 0.34 0.55 *p*\<.05 Suicide attempt 0.05 0.16 *p*\<.05 0.20 0.43 *p*\<.05 Suicide planning 0.10 0.26 *p*\<.05 0.31 0.55 *p*\<.05 Alcohol intoxication 0.21 0.23 NS 0.46 0.55 NS Regular cigarette smoking 0.03 0.05 *p*\<.05 0.12 0.18 NS Marijuana use 0.15 0.16 NS 0.40 0.48 NS Other drug use 0.05 0.08 *p*\<.05 0.21 0.29 NS ^a^Denotes a significant difference in the proportion of other-sex attracted students experiencing victimization, *Χ* ^2^ (1, n = 2263) = 5.85, *p* \<.05. ^b^Denotes a significant difference in the prevalence of the health-risking behavior by victimization status within attraction type, NS = not significant. Overall, 18.6 percent of students reported experiencing gender-variant-based victimization. A larger proportion of same-sex attracted students (38.3%) reported experiencing gender-variant-based victimization than other-sex attracted students (17.1%), *Χ* ^2^ (1, n = 2438) = 48.11, *p* \< .05. Additionally, a larger proportion of males (20.3%) reported experiencing gender-variant-based victimization than females (16.9%), *Χ* ^2^ (1, n = 2438) = 4.58, *p* \< .05. However, there was not a significant difference between middle school (18.1%) and high school (19.2%) students in the proportion of students who reported experiencing gender-variant-based victimization, *Χ* ^2^ (1, n = 2438) = .44, n.s. Hypothesis Tests {#sec018} ---------------- Congruent with hypothesis 1, other-sex attracted students who experienced gender-variant-based victimization were more likely than other-sex attracted students who did not experience gender-variant-based victimization to report suicidal ideation in the past week, suicidal attempts since the beginning of the school year, suicide planning since the beginning of the school year, regular cigarette smoking in the past 3 months, and other drug use in the past 3 month. However, gender-variant-based victimization was not associated with alcohol intoxication and marijuana use in the past 3 months among other-sex attracted youths. A similar pattern is observed for same-sex attracted youths in terms of suicide-related behaviors. That is, same-sex attracted youths who reported gender-variant-based victimization were more likely report suicidal ideation in the past week, and both suicidal attempts and suicide planning since the beginning of the school year. The association between substance use and gender-variant-based victimization is in the expected direction, although it doesn't meet traditional criteria for statistical significance. To test hypothesis 2, we regressed each of the outcomes on attraction and experiencing gender-variant-based victimization to determine if each is uniquely associated with self-reports of the health-risking behaviors. These results are presented in [Table 6](#pone.0129976.t006){ref-type="table"}, Model 1. Controlling for attraction, students who experienced gender-variant-based victimization were more likely than students who did not experienced gender-variant-based victimization to report suicidal ideation in the past week, suicidal attempts since the beginning of the school year, suicide planning since the beginning of the school year, and regular cigarette smoking and other drug use in the past 3 months. However, gender-variant-based victimization was not associated with alcohol intoxication in the past 3 months or marijuana use in the past 3 months. Controlling for experiencing gender-variant-based victimization, same-sex attracted students were more likely than their other-sex attracted peers to report all of the suicide-related behaviors and all forms of substance use. These findings suggest that gender-variant-based victimization and same-sex attraction each uniquely associated with health-risking behaviors. These results provide general support for hypothesis 2. 10.1371/journal.pone.0129976.t006 ###### Results of logistic regression models to test hypothesis 2 and 3. {#pone.0129976.t006g} Model 1 Model 2 ------------------------------------ --------- --------- ------ ------ ------- ------ ------ ------- ------- Health-risking behavior Suicidal ideation   Intercept 0.25 -0.23 0.28 0.25 -0.22 0.28   Attraction 1.95 -1.40 2.70 2.08 -1.38 3.14   Victimization 2.75 -2.22 3.42 2.81 -2.23 3.55   Attraction\*Victimization 0.84 -0.43 1.64 1.03 -1.54 3.60 Suicide attempt   Intercept 0.05 -0.04 0.06 0.05 -0.04 0.06   Attraction 4.52 -3.09 6.62 5.05 -3.02 8.44   Victimization 3.60 -2.64 4.89 3.78 -2.68 5.33   Attraction\*Victimization 0.79 -0.37 1.68 7.24 -0.60 15.07 Suicide planning   Intercept 0.11 -0.09 0.13 0.11 -0.09 0.12   Attraction 3.91 -2.78 5.50 4.17 -2.69 6.46   Victimization 3.17 -2.46 4.08 3.25 -2.47 4.28   Attraction\*Victimization 0.85 -0.43 1.70 5.11 -0.78 11.01 Alcohol intoxication   Intercept 0.27 -0.24 0.30 0.27 -0.24 0.30   Attraction 3.49 -2.54 4.79 3.16 -2.13 4.69   Victimization 1.13 -0.89 1.43 1.08 -0.83 1.40   Attraction\*Victimization 1.33 -0.68 2.58 1.28 -1.22 3.78 Regular cigarette smoking   Intercept 0.03 -0.02 0.03 0.03 -0.02 0.03   Attraction 4.89 -2.96 8.08 5.21 -2.73 9.94   Victimization 1.77 -1.11 2.83 1.86 -1.07 3.23   Attraction\*Victimization 0.86 -0.31 2.37 2.24 -3.84 8.32 Marijuana use   Intercept 0.18 -0.16 0.20 0.18 -0.16 0.21   Attraction 4.02 -2.90 5.56 3.62 -2.41 5.43   Victimization 1.08 -0.83 1.42 1.02 -0.76 1.38   Attraction\*Victimization 1.35 -0.68 2.68 1.36 -1.43 4.15 Other drug use   Intercept 0.06 -0.05 0.07 0.06 -0.04 0.07   Attraction 4.78 -3.21 7.11 4.96 -3.00 8.22   Victimization 1.59 -1.11 2.28 1.63 -1.08 2.47   [Attraction\*Victimization]{.ul}       0.90 -0.40 2.05 1.73 -2.84 6.30 Notes: exp([*B*]{.ul}) = exponentiated estimate, CI = confidence interval, RERI = relative excess risk due to interaction. Attraction is coded 1 if same-sex attracted, 0 if other-sex attracted. Victimization is coded 1 if victimized, 0 if not victimized. 95% CI exp([*B*]{.ul}) is statistically significant when the CI doesn\'t include 1. 95% CI RERI is statistically significant when the CI doesn\'t include 0. To test hypothesis 3, we enhanced the models specified to test hypothesis 2 by adding an interaction term between victimization and attraction to determine if the effect of experiencing gender-variant-based victimization differed by attraction. The results are presented in [Table 6](#pone.0129976.t006){ref-type="table"}, Model 2. Across all of the outcomes, none of the interaction terms were statistically significant, and this holds for both the multiplicative and additive interaction terms. These findings suggest that the effect of experiencing gender-variant-based victimization does not significantly differ as a function of attraction, providing support for hypothesis 3. Discussion {#sec019} ========== This study set out to examine the association between gender-variant-based victimization and negative outcomes (i.e., suicidal behavior, substance use) in other-sex attracted adolescents, and to highlight gender-variant-based victimization and attraction as uniquely associated with suicidal behavior and substance use for adolescents. As hypothesized, it was found that other-sex attracted adolescents who experienced gender-variant-based victimization were more likely to report suicidal behaviors than other-sex attracted adolescents who did not experience this type of victimization. Additionally, as predicted, same-sex attraction and experiencing gender-variant-based victimization were both found to be associated with suicidal behaviors, and the effect of gender-variant-based victimization was not found to be modified by attraction. However, our hypotheses were only partially supported for the substance use measures. Other-sex attracted adolescents who experienced gender-variant-based victimization were more likely to report regular cigarette smoking and other drug use than those who did not experience this type of victimization. Controlling for attraction, gender-variant-based victimization was associated with self-reports of regular cigarette smoking and other drug use. However, the association between experiencing gender-variant-based victimization and suicidal behavior and substance use does not vary as a function of attraction. Thus, overall, our findings suggest that for both other- and same-sex attracted adolescents experiencing gender-variant-based victimization is associated with an increased likelihood of reporting suicidal behaviors, and some types of substance use, especially for other-sex attracted adolescents. The results of the present study are consistent with previous findings suggesting that expressing gender atypical characteristics during adolescence \[[@pone.0129976.ref008],[@pone.0129976.ref009]\], or experiencing gender-variant-based victimization \[[@pone.0129976.ref008],[@pone.0129976.ref025],[@pone.0129976.ref029]\], are related to lower levels of psychological well-being, regardless of being same- or other-sex attracted. Extending from these studies, the present study highlights the association between gender-variant-based victimization and suicidal behavior and substance use in other-sex attracted adolescents. Thus, this study helps to move the present literature beyond almost exclusively focusing on negative outcomes associated with gender-variant-based victimization in specific minority groups (i.e. sexual- and gender-minorities), and provides an opportunity to discuss the broader issue of gender policing in adolescents. Implications {#sec020} ------------ The current work has potential implications for interventions addressing types of bullying that have gender-variant-based victimization components. For example, over the last several years, homophobic bullying has been identified as a key risk factor for suicidal behavior, substance use, and other negative health outcomes in adolescents \[[@pone.0129976.ref003],[@pone.0129976.ref029],[@pone.0129976.ref030]\]. This has resulted in research mostly focusing on identification as lesbian, gay, bisexual (LGB), questioning, or being same-sex attracted as risk factors for negative health outcomes \[[@pone.0129976.ref003],[@pone.0129976.ref004]\], or for experiencing victimization \[[@pone.0129976.ref031],[@pone.0129976.ref032]\]. However, this work does little to address the motives behind homophobic bullying and the forms of victimization that fall under its broad umbrella. For example, same-sex attracted adolescents may be target for victimization solely because of their sexual orientation, but often times they are targeted for victimization because their identity as being same-sex attracted is not viewed by their peers as being consistent with traditional gender norms \[[@pone.0129976.ref015],[@pone.0129976.ref016]\]. In these instances, victimization is gender-variant-based and used as a form of peer gender policing. Thus, same-sex attracted adolescents who experiences homophobic bullying may be targeted exclusively for sexual orientation-based victimization, but may also targeted for gender-variant-based victimization. While exploring gender-variant-based victimization specifically as a component of homophobic bullying is beyond the scope of this the current work, others have noted that homophobic bullying is made up of both sexual orientation-based victimization and gender-variant-based victimization \[[@pone.0129976.ref015],[@pone.0129976.ref016],[@pone.0129976.ref033],[@pone.0129976.ref034]\]. Even though these two types of victimization are very closely linked, and both likely disproportionately impact sexual- and gender-minority adolescents, their underlying motives have implications for who and how interventions are targeted. For example, if homophobic bullying is indeed mostly sexual orientation-based, then focusing on identifying and supporting sexual-minority students and promoting acceptance of the diversity in sexual orientation is likely the best approach. Yet, if homophobic bullying is mostly driven by policing gender through the use of gender-variant-based victimization, the issues becomes much broader because any adolescents, regardless of sexual orientation, may become targets for homophobic bullying based on the way they express their gender. Indeed, some previous work demonstrates the effects of homophobic bullying on the psychological and social wellbeing of adolescents, regardless of sexual orientation \[[@pone.0129976.ref028]\]. The current work does not capture whether the gender-variant-based victimization that other-sex students experienced was homophobic in nature, but it does demonstrate the broad impact that gender-variant-based victimization can have on both other- and same-sex adolescents. This study also identifies gender-variant-based victimization as being uniquely associated with health-risking behavior, especially suicidal behavior. Further work is needed to understand the specific mechanisms that underlie this association. However, if this relationship is found to be explained by feelings of social exclusion created by this type of victimization as previous work suggests \[[@pone.0129976.ref024]\], then our study would support creating broad-based interventions that focus on general acceptance of gender diversity. Focusing on promoting acceptance for individual differences in gender expression is a qualitatively different approach than targeting interventions to sexual- and gender- minority adolescents who are stereotyped as being gender atypical and experiencing this type of victimization. Strengths & Limitations {#sec021} ----------------------- This work has several strengths. First, this study utilizes a large, diverse sample of both middle school and high school students in the Eastern United States. Given the relatively progressive nature of this region of the country, the presence of gender-variant-based victimization and its association with negative outcomes as serious as suicidal behavior and substance use is notable. Additionally, this work is strengthened by its investigation of gender-variant-based victimization across the entire sample of adolescents. Most previous work addressing gender-variant-based victimization has focused on subsets of the adolescent population that are stereotyped as expressing gender-atypical traits (i.e. lesbian, gay, and bisexual adolescents) or who are identified as being gender-atypical (e.g., transgender or gender nonconforming). The current work, instead, asked all students if they had experienced victimization based on their gender expression. Another key strength of this work is that it separates the effects of gender-variant-based victimization from the effects of attraction in their association with suicidal behavior and substance use. This helps to highlight the similar negative effects experienced by both other-sex and same-sex attracted youth who are targeted for gender-variant-based victimization. Additionally, this analysis demonstrates that gender-variant-based victimization is not something that is universally experienced by same-sex attracted youth, helping to account for some of the heterogeneity among same-sex attracted adolescents in their experiences with and outcomes related to victimization. One of the primary limitations to the current work is that the investigators did not ask participants to describe the nature of the victimization that they identified as being based on their gender expression. Thus, there is no way to know exactly what types of behaviors and language were identified as being gender-variant-based victimization in this work. However, for this study, the researchers were intentional about providing a broad question that was devoid of any terms related to sexuality, attraction, or other identities so as to better account for the experiences of all students, not just students who are same-sex attracted or identify as transgender or gender nonconforming. This study is also limited in that the sample was only obtained from one school district, but due to the large sample size and the diversity of the students who participated, the effects of this limitation on the outcome of this study are believed to be minimal. Additionally, despite our large sample size, our study may have still been underpowered when it comes to investigating the association between gender-variant-based victimization and substance use for same-sex attracted adolescents. This is evidenced by results for the substance use outcomes trending in the predicted direction. It should also be noted that our study may have been underpowered for detecting an interaction between attraction and experiencing gender-variant-based victimization. The statistical analyses conducted did not find a multiplicative or an additive interaction for any of the health-risking behaviors assessed, but it is not possible for the current work rule out the existence of an interaction between these two constructs. However this study does suggest that the effects of any potential relationships between an interaction and the assessed health-risking behaviors are weaker than the effects of attraction and gender-variant-based victimization in of themselves. Future studies with larger, multi-school district samples will be needed to help fully tease apart these relationships and provide more clarity with regard to the nuances of the lesser effects. This work is also limited by the way that same-sex attracted adolescents were identified and categorized. The actual identification item itself focused on future dating intentions, and therefore it does not provide a direct assessment of self-identification as a sexual minority, limiting this work in the broader literature on sexual minority adolescents. However, this single item provided a developmentally appropriate way to capture attraction from adolescents spanning both middle school and high school. Additionally, adolescents who expressed attraction to both males and females were coarsely grouped with adolescents who expressed exclusive same-sex attraction to create a single comparison group. While this grouping combines two distinctly different groups, this decision was based on preliminary analyses demonstrating exclusively same-sex attracted adolescents and adolescents expressing both same- and other-sex attracted were similar to one another, and similarly different from exclusively other-sex attracted adolescents, on all of the variables assessed in this study. Thus, this decision did not alter the overall findings of the study. A similar potential limitation comes from grouping both males and females together within each of the attraction groups; however, this decision was also based on preliminary analyses indicating that the general pattern of results held true for both males and females. Therefore, combining males and females did not alter the overall findings of the study. In the end, these findings are based on a large, diverse sample, and show strong relationships, despite only focusing on suicidal behavior and substance use over a relatively short amount of time, supporting the assertion that these findings are not just due to chance or particularly unique conditions. Future Research {#sec022} --------------- Future work may benefit from following-up with a subset of participants in focus groups to better understand the specific actions and language that they believe constitutes gender-variant-based victimization. The current study identifies gender-variant-based victimization as a factor associated with health-risking behavior through the use of a question asking students whether they have been made fun of for how they express their gender (i.e., *Do kids ever make fun of you for not being feminine enough/too masculine*?), but this simple item does not provide any information about why the adolescents perceived the victimization as being based on their gender expression. Collecting additional information through focus groups could help to identify the types of behaviors and language that are associated with this form of victimization. Additionally, this work may benefit from being replicated across multiple school districts in different parts of the country to ensure the findings are not just an artifact of the specific culture and experiences of students in this particular school district. Finally, it is recommended that interventions developed to reduce bullying and its negative outcomes are designed with an understanding that gender-variant-based victimization is an issue for all students, and not just specific target populations. Thus, adolescents might benefit from creating a school culture that is accepting of the diversity of all students' expressions of gender, regardless of their actual or perceived sexual orientation or gender identity. However, this is an empirical question that will need to be addressed in further work. Conclusions {#sec023} =========== The current work uniquely contributes a link between gender-variant-based victimization and suicidal behavior and substance use. For all adolescents (both other- and same-sex attracted), gender-variant-based victimization was associated with a higher odds of suicidal thoughts and behaviors. In addition, for other-sex attracted adolescents, gender-variant-based victimization was associated with increased odds of smoking cigarettes regularly and using other drugs. These findings suggest that gender-variant-based victimization has potentially serious implications for the psychological wellbeing and substance use of other-sex attracted adolescents, not just same-sex attracted adolescents. Thus, gender-variant-based victimization has implications for all adolescents, not just a subset of adolescents that are typically considered to be at risk for this this type of victimization based on their identity. Taken together, the findings of this study highlight the need to better understand the influence that gender-variant-based victimization has on all adolescents. From a prevention and intervention standpoint, this work suggests that additional attention may need to be paid to all adolescents who do not conform to gender norms or who are bullied based on their gender expression. Additionally, educators, researchers, and providers may need to focus on the broader issue of creating acceptance for a diverse array of gender expressions, instead of focusing on specific minority groups, in order to reduce negative outcomes such as suicidal behaviors and substance use. This may be especially true when addressing homophobic or gender-based bullying because much of this victimization stems from the policing of gender expression through the use of gender-variant-based victimization. Supporting Information {#sec024} ====================== ###### This the Dataset File. This file contains all of the data for the analyses conducted in this study. (XLS) ###### Click here for additional data file. We thank the students and staff of the participating school district, and K. Foula Dimopoulous for his insightful feedback on earlier drafts of this work. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MI KLH PYC KPC RGT. Performed the experiments: KLH PYC KPC RGT. Analyzed the data: MI KLH. Wrote the paper: MI KLH PYC KPC RGT. [^3]: ‡ These authors also contributed equally to this work.

Introduction ============ Pericardial abscess is a rare condition which results from hematogenous spread, direct extension from an adjacent infectious focus, trauma, or surgery. A pericardial abscess by *Staphylococcus aureus* is rarer[@B1]-[@B3] and to our knowledge, this is the first case report of a pericardial abscess as a complication of staphylococcal bacteremia in Korea. Case ==== A 72-year-old woman presented to the emergency room with dyspnea and myalgia that developed 2 weeks after undergoing acupuncture therapy on both knees due to arthralgia. Vital signs on admission showed hypotension (80/40 mm Hg), tachycardia (118 beats per minute), tachypnea (20 per minute), and hypothermia (35℃). On physical examination, the patient had multiple needle scars on both knees with dappled rashes on her entire body ([Fig. 1A](#F1){ref-type="fig"}). Her laboratory results showed elevated white blood cell counts (10100/mm^3^), with neutrophil 88%, elevated high sensitive C-reactive protein level of 36 mg/dL (reference range, 0-0.5 mg/dL), blood urea nitrogen/creatinine 71/4.5 mg/dL, myoglobin 5169 ng/mL (reference range, 16.3-96.5 ng/dL), creatine kinase myocardial band 9.7 U/L (reference range, 0-3.6 U/L), Troponin I 0.28 ng/mL (re-ference range, 0-0.1 ng/mL). A chest radiograph showed cardiomegaly and an electrocardiography showed atrial fibrillation with rapid ventricular response. A transthoracic echocardiogram demonstrated concentric left ventricular hypertrophy with fluid collection in the posterolateral wall of the pericardium with no evidence of valvular vegetation or tamponade physiology ([Fig. 1B](#F1){ref-type="fig"}). Pericardial aspiration of the fluid revealed a bloody material ([Fig. 1C](#F1){ref-type="fig"}) and cultures grew *Staphylococcus aureus*. Blood cultures showed staphylococcal bacteremia ([Fig. 1D](#F1){ref-type="fig"}). Lab analysis of aspiration fluid showed elevated white blood cell counts \>50000/mm^3^), with polymorph-onuclear neutrophil 90%, pH 7.3, glucose 5 mg/dL, lactate dehydrogenase 12397 U/L, albumin 2.4 g/dL and total protein 6.2 g/dL. Percutaneous drainage and empiric antibiotic treatment were started immediately. However, the patient expired due to refractory sepsis and organ failure. Discussion ========== Pericardial abscess is a serious, life-threatening illness associated with high mortality. A pericardial abscess is an extremely unusual complication of *Staphylococcus aureus* bacteremia.[@B1]-[@B4] The mechanism of purulent pericarditis by *Staphylococcus aureus* is unknown. Possible explanations include hematogenous seeding or direct extension into a pre-existing pericardial cyst or purulent pericarditis occurring in a patient with old pericardial adhesions.[@B5][@B6] Other microorganisms causing pericardial abscess include *Mycobacterium tuberculosis*, Gram-negative bacilli, *Streptococcus species*, and *Aspergillus*. In Korea, only two cases of tuberculous pericardial abscess and Bacteroides fragilis have been reported.[@B7][@B8] Because delayed diagnosis of pericardial abscess may lead to debilitating complications, early echocardiography is important. To-mography provides useful information on the extent of the pericardial abnormality when the echocardiographic picture is not clear.[@B5] The primary treatments for pericardial abscess include percutaneous or surgical drainage and pericardiectomy with prompt administration of appropriate antibiotics. Although we cannot verify the pathogenesis of this patient\'s infection, based on the multifocal acupuncture therapy history of this patient and the absence of previous pericardial disease, the pericardial abscess may have been caused by hematogenous spread of *Staphylococcus aureus* from the soft tissue infection of the knees. However, it remains to be determined whether acupuncture treatment severely increases the risk of bacteremia, or whether this case is simply a coincidence implicating acupuncture. The authors have no financial conflicts of interest. {#F1}

Introduction {#Sec1} ============ EPSEs are drug-induced movement disorders. These symptoms include dystonia (continuous spasms and muscle contractions), akathisia (motor restlessness), parkinsonism (characteristic symptoms such as rigidity and slowness of movement), tremor, and tardive dyskinesia (irregular, jerky movements). Most patients with drug-induced EPSEs are associated with lipophilic dopamine D2 receptor blockers prescribed for psychotic disorders and depression \[[@CR1]\]. The first case of fz-induced parkinsonism was reported in 1984 \[[@CR2]\]. From that moment, many related cases have been reported \[[@CR3]--[@CR9]\]. Fz and cz are both calcium channel blockers with a similar structure and pharmacologic effects. Fz is a di-fluorinated derivative of cz and has a 2.5--15 time stronger potency than cz \[[@CR10]\]. Other than the calcium entry blocking effect, fz and cz also have antihistaminic, antiserotoninergic, and antidopaminergic activities. Because of the D2 receptor blocking activity \[[@CR11]\], both fz and cz have EPSEs including parkinsonism, orobuccolingual dyskinesia, dystonia, and akathisia \[[@CR5], [@CR12]\]. Previous case series showed that elderly woman and history of essential tremor may probably be risk factors for the development of EPSE following cz or fz exposure \[[@CR4], [@CR5]\]. Until now, only few studies have used a large sample size to investigate fz- and cz-induced extrapyramidal symptoms. In Taiwan, fz and cz are frequently prescribed for vertigo, migraine prophylaxis, and cerebrovascular blood flow insufficiency. The aim of this study was to investigate the incidence and occurrence time of cz- and fz-related EPSEs. Materials and methods {#Sec2} ===================== Data source {#Sec3} ----------- This retrospective cohort study used data from the Longitudinal Health Insurance Databases (LHIDs) 2005 and 2010. Both databases contain registration data and medical claims of 1,000,000 individuals randomly sampled from the 23.68 million beneficiaries registered in the National Health Insurance Research Database (NHIRD). The NHIRD contains a comprehensive health care information including diagnoses, prescriptions, and information about inpatient and outpatient care during 1996--2011. It covers over 99% of the 23 million residents in Taiwan. This study was approved by the Institutional Review Board of Chung-Shan Medical University Hospital in Taiwan. Patient identification {#Sec4} ---------------------- Included in this study were patients who were prescribed cz or fz from 2002 to 2011 for more than 1 month. Patients who had been prescribed cz or fz in 2001 were excluded. A total of 68,924 patients who took either fz (*n* = 54,052), cz (*n* = 13,027), or both (*n* = 1665) were identified. The first prescription date of either cz or fz medication was defined as the index date. Excluded from the study were 10,961 patients who were diagnosed with one or more of the several diseases mentioned below before the index date. They included dementia (International Classification of Diseases, 9th revision, clinical modification (ICD-9-CM) 290.0∼290.43), neurodegeneration (ICD-9-CM 333.0, 333.4, 334.0--334.9), hydrocephalus (ICD-9-CM 331), subdural hemorrhage (ICD-9-CM 432.1), brain tumor (ICD-9-CM 191), Wilson's disease (ICD-9-CM 275.1), hypoparathyroidism (ICD-9-CM 252.1, 252.8, 252.9, 275.49A), pantothenate kinase-associated neurodegeneration (ICD-9-CM 277.9I), human immunodeficiency virus infection (ICD-9-CM 042, 079.53, 795.71), neurosyphilis (ICD-9-CM 094.89, 094.9), progressive multifocal leukoencephalopathy (ICD-9-CM 046.3), toxoplasmosis (ICD-9-CM 130.0, 130.7), and stroke (ICD-9-CM 431, 432.9, 434, 436). Also excluded were 12,297 patients below the age of 45 and 5728 patients diagnosed with parkinsonism, dyskinesia, or dystonia (ICD-9-CM 332.0, 332.1, 333.90, 333.99, 333.7, and 333.8) before the index date. In addition, patients who used fz and cz simultaneously (*n* = 1016) were excluded. Definition of outcomes {#Sec5} ---------------------- Drug-related EPSEs was defined as a diagnosis of parkinsonism, dyskinesia, or symptomatic dystonia (ICD-9-CM 332.0, 332.1, 333.90, 333.99, 333.7, and 333.8) from the index date to 3 months after stopping either fz or cz \[[@CR8]\]. To determine the hazard ratio of extrapyramidal symptoms, a control group matched by sex, age, and duration of follow-up was included in the analysis (exposure: control group = 1:2). Control subjects were excluded due to fulfillment of the exclusion criteria listed above. Five thousand six hundred three patients were excluded from the study because there were no suitable control individuals. The potential confounders included sex, age, low income, urbanization, and antipsychotics usage 2 years before the index date (anatomical therapeutic chemical code N05A), co-morbidities (DM, ICD-9-CM: 250), chronic kidney disease (CKD, ICD-9-CM: 585), severe liver dysfunction (ICD-9-CM: 572.2, 571.5, 572.2--572.4), history of essential tremor (ICD-9-CM: 333.1), history of other movement disorders (ICD-9-CM: 333.2, 333.3, 333.5, 333.6), and cardiovascular disease (ICD-9-CM: 410--414, 433, 444)). Statistical analysis {#Sec6} -------------------- Data were analyzed using the SAS software. ANOVA was used to compare the mean difference among groups for continuous variables while chi-square test was used for nominal variables. The hazards ratio (HR) and 95% confidence interval (CI) were estimated using the Cox proportional hazard model. A *P* value \< 0.05 was considered statistically significant. Results {#Sec7} ======= Table [1](#Tab1){ref-type="table"} shows basic characteristics of study subjects. In the final analysis, 26,133 and 7186 individuals were found to have been prescribed fz or cz, respectively. The matched controls involved 66,638 participants. The incidence rates of fz- and cz-induced EPSEs were 21.03 and 10.3 per 10,000 person-months, respectively (Table [2](#Tab2){ref-type="table"}). Figure [1](#Fig1){ref-type="fig"} shows the cumulative incidence proportion of the EPSEs. Both fz and cz patients had a higher cumulative incidence of EPSEs than their control individuals (Fig. [1](#Fig1){ref-type="fig"} a, b, *p* \< 0.001). Patients who were prescribed fz had a higher cumulative incidence than those who were prescribed cz (Fig. [1](#Fig1){ref-type="fig"} c, *p* \< 0.001). The incidences of EPSEs in both treatment groups were higher in the second than the first year of exposure (i.e., 45.59 vs 18.62 per 10,000 person-months for fz; 13.89 vs 9.41 per 10,000 person-months for cz). The incidence rates of parkinsonism or dystonia/dyskinesia are listed in supplementary Table [1](#MOESM1){ref-type="media"}. Because many patients had antipsychotics or metoclopramide concomitantly, the risk of cz- or fz-related EPSEs might have been overestimated. A sensitivity analysis which excluded patients concomitantly taking antipsychotics, reserpine, or metoclopramide with cz or fz showed that the incidence rates of fz (*n* = 10,020) and cz (*n* = 2567) induced EPSEs were 13.05 (10.74--15.86) and 8.10 (5.04--13.04) per 10,000 person-months.Table 1Basic characteristics of the study participantsFlunarizineCinnarizineControl*p* value*N*26,133718666,638Initial age60.94 ± 10.2362.29 ± 10.1661.23 ± 10.23\<.0001Sex0.56 Male9449(36.16%)2549(35.47%)23,996(36.01%) Female16,684(63.84%)4637(64.53%)42,642(63.99%)Low income\<.0001 No25,891(99.07%)7135(99.29%)66,231(99.39%) Yes242(0.93%)51(0.71%)407(0.61%)Urbanization\<.0001 Urban14,905(57.04%)4185(58.24%)41,320(62.01%) Normal8026(30.71%)1994(27.75%)18,693(28.05%) Rural3202(12.25%)1007(14.01%)6625(9.94%)Comorbidity at baseline CKD584(2.23%)180(2.5%)1075(1.61%)\<.0001 Severe liver dysfunction424(1.62%)107(1.49%)766(1.15%)\<.0001 Essential tremor history139(0.53%)40(0.56%)131(0.20%)\<.0001 Other movement disorder64(0.24%)19(0.26%)56(0.08%)\<.0001 DM5888(22.53%)1738(24.19%)9936(14.91%)\<.0001 CVD7419(28.39%)1985(27.62%)9347(14.03%)\<.0001Medication used before baseline Antipsychotic2411(9.23%)541(7.53%)1796(2.7%)\<.0001*N* number, *CKD* chronic kidney disease, *DM* diabetes mellitus, *CVD* cardiovascular disease Table 2Incidence rate of parkinsonism, dyskinesia, or dystonia in the exposed and control subjectsFlunarizineCinnarizineControlCaseIncidence rate (per 10,000 person months)CaseIncidence rate (per 10,000 person months)CaseIncidence rate (per 10,000 person months)All period43621.03 (19.15--23.1)6110.30 (8.02--13.24)1272.36 (1.99--2.81)Within 1 year30118.62 (16.63--20.85)419.41 (6.93--12.78)922.23 (1.82--2.74)From 1 to 2 years10245.59 (37.55--55.35)913.89 (7.23--26.69)132.21 (1.28--3.80) Fig. 1Kaplan-Meier plot of cumulative incidence proportion of parkinsonism or dyskinesia/dystonia by groups Table [3](#Tab3){ref-type="table"} shows the HRs for parkinsonism, dyskinesia, or dystonia after exposure to drugs. After adjusting for age, sex, income, urbanization, baseline antipsychotic use, and other comorbidities, the HRs for EPSEs among patients who were exposed to fz and cz were 8.03 (95% CI 6.55--9.84) and 3.41 (95% CI 2.50--4.63), respectively. Table [4](#Tab4){ref-type="table"} shows the HRs for parkinsonism, dyskinesia, or dystonia for different periods. The risk for EPSEs among subjects who took fz was significantly higher in the second (HR = 18.55, 95% CI 10.32--33.35) than the first year (HR = 7.82, 95% CI 6.17--9.93).Table 3Cox regression analysis to estimate the hazard ratios of parkinsonism, dyskinesia, or dystoniaHR95% C.I.*p* valueIntervention (ref: control) Flunarizine8.036.55--9.84\<.0001 Cinnarizine3.412.50--4.63\<.0001Sex (ref: Female) Male0.940.79--1.100.42Age (per 1 year)1.061.05--1.07\<.0001Low income (ref: No) Yes1.670.92--3.030.094Urbanization (ref: Urban) Normal1.110.94--1.320.219 Rural0.880.69--1.130.31Comorbidity at baseline (ref: Without) CKD1.310.87--1.980.20 Severe liver dysfunction0.400.15--1.080.071 Essential tremor history6.143.91--9.62\<.0001 Other movement disorders3.971.64--9.580.002 DM1.140.95--1.360.16 CVD1.331.13--1.580.0008Medication used before baseline (ref: No) Antipsychotic1.821.44--2.31\<.0001*CKD* chronic kidney disease, *DM* diabetes mellitus, *CVD* cardiovascular disease Table 4Cox regression model to estimate the hazard ratios of parkinsonism, dyskinesia, or dystonia by groups and observation periodFlunarizineCinnarizineHR^a^ (95% C.I.)*p* valueHR^a^ (95% C.I.)*p* valueAll period (*n* = 99,957)8.03 (6.55--9.84)\<.00013.41 (2.50--4.63)\<.0001Within 1 year (*n* = 99,957)7.82 (6.17--9.93)\<.00013.45 (2.38--5.00)\<.0001From 1 to 2 years (*N* = 13,116)18.55 (10.32--33.35)\<.00014.95 (2.11--11.62)0.0002Adjusted for age, sex, low-income household, urbanization, comorbidities (chronic kidney disease, severe liver dysfunction, history of essential tremor or other movement disorder, and cardiovascular disease), and antipsychotic use at baseline*HR* hazard ratio^a^Reference group is control group Discussion {#Sec8} ========== To our knowledge, there are few epidemiologic studies focusing on fz- and cz-induced EPSEs. The incidence rates of EPSEs in patients who took fz and cz were 21.03 and 10.3 per 10,000 person-months, respectively. Patients who took fz and cz were 8.03 and 3.41 times more likely to develop these symptoms. Previously, Fabinai et al. created a point prevalence analysis including 26 patients and found that 50% of chronic fz and cz users had EPSEs \[[@CR12]\]. According to reports from Negrotti \[[@CR8]\] and Martí-Masso \[[@CR5]\] et al., the diagnosis of fz or cz induced parkinsonism should exclude other causes including other offending drugs. For this reason, we performed a sensitivity analysis which excluded patients concomitantly taking antipsychotics, reserpine, or metoclopramide with cz or fz. The incidence rates of fz- and cz-induced EPSEs were 13.05 and 8.10 per 10,000 person-months. The risks for EPSEs were higher among subjects who took fz (HR = 10.59, 95% CI = 7.36--15.22) and cz (HR = 5.49, 95% CI = 3.11--9.70) than their matched control individuals. We did not use the population of sensitivity analysis as our main result because of the following reasons. First, the NHIRD does not contain clinical data; therefore, we could not clarify distinctly which drug was the major cause of EPSEs. More than half of the patients were concomitantly treated with either antipsychotics or metoclopramide and fz or cz. Further investigations are required to understand the interactions between these drugs. Second, the control group may have selection bias when restricting all these medications. To avoid such biases and to eliminate the influence of antipsychotics, adjustments were made for baseline antipsychotic use. In this study, the fz users had a higher risk of developing EPSEs than the cz individuals. Fz and cz both block striatal dopamine D2 receptors at clinically used doses \[[@CR11]\]. Loss of tyrosine hydroxylase in monoaminergic presynaptic neuron leading to dopamine deletion might also be responsible for these side effects \[[@CR4], [@CR5], [@CR8]\]. Since fz is 2.5 to 10 times more potent than cz \[[@CR10]\], it is no surprise that the risk of EPSEs was 2 to 3 times higher in subjects who took fz. The incidence of EPSEs 1--2 years after fz prescription (45.59 per 10,000 person-months) was significantly higher than that within the first year (18.62 per 10,000 person-months). The HR increased from 7.82 (95% CI 6.17--9.93) to 18.55 (95% CI 10.32--33.35) among subjects who took fz for more than 1 year. However, this was not obvious in cz individuals. In Fig. [1](#Fig1){ref-type="fig"}, the risk of fz-induced extrapyramidal side effects rose rapidly 2--3 years after the drug exposure. Previous literatures have reported a positive relationship between the duration of fz or cz use and the onset of parkinsonism \[[@CR12], [@CR13]\]. The present study has strengthened this concept and has also demonstrated the differences that exist between fz and cz. Potential benefits and risks should be weighed when considering long-term use of these calcium channel blockers. This study has several limitations. First, the NHIRD does not contain information about clinical examination. This study design may have included patients with clinical presentations of parkinsonism who have never sought medical advice. In addition, some authors included clinical improvement after drug withdrawal as one of the diagnostic criteria of DIP \[[@CR5]\]. However, such information was not also available in the NHIRD. However, more researchers found that patients did not fully recover after being exposed to cz and fz \[[@CR8], [@CR14]\]. Diseases that may cause parkinsonism or dystonia were excluded from this study. Adjustments were made for possible confounders such as antipsychotic use. Second, patients with a diagnosis of Parkinson's disease or Parkinson plus syndrome might have been included in the analysis since ICD-9-CM codes 332.0 and 332.1 were considered as outcomes. However, according to previous studies, it is hard to distinguish between fz- or cz-induced parkinsonism and idiopathic Parkinson's disease (IPD) based on clinical presentations \[[@CR3]--[@CR6], [@CR8]\]. One report showed that 43% patients with fz- or cz-induced parkinsonism had a clinical pattern similar to IPD patients. Dopaminergic treatment was also effective in patients with fz- or cz-induced parkinsonism \[[@CR15]\]. Furthermore, more evidence supports that patients with fz- or cz-induced parkinsonism were not fully recovered after stopping medications. In addition, some ultimately developed IPD \[[@CR5], [@CR8]\]. In this study, ICD-9-CM codes for Parkinson's disease were included into analysis considering that fz- or cz-induced parkinsonism may be misclassified as IPD. Third, the incidence of akathisia, one of the EPSEs induced by fz or cz \[[@CR12], [@CR14]\] could not be obtained. There is no specific ICD-9-CM code for akathisia. Most patients with akathisia had concomitant parkinsonian symptoms \[[@CR12]\]; therefore, the incidence could not have been underestimated in this study. In conclusion, cz and fz significantly increased the risk of extrapyramidal symptoms. Potential risks should be weighed when considering long-term use of these drugs. Electronic supplementary material ================================= {#Sec9} ESM 1(DOCX 24 kb) fz : Flunarizine cz : Cinnarizine EPSE : Extrapyramidal side effect HR : Hazard ratio DIP : Drug-induced parkinsonism LHID : Longitudinal Health Insurance Database NHIRD : National Health Insurance Research Database ICD-9-CM : International Classification of Diseases, 9th revision, clinical modification CKD : Chronic kidney disease CI : Confidence interval IPD : Idiopathic Parkinson's disease **Electronic supplementary material** The online version of this article (doi:10.1007/s00228-017-2247-x) contains supplementary material, which is available to authorized users. An erratum to this article is available at <http://dx.doi.org/10.1007/s00228-017-2268-5>. This study is based in part on data from the NHIRD provided by the bureau of National Health Insurance. The descriptions or conclusions in this document do not represent the viewpoint of the Bureau. Source of funding {#FPar1} ================= None. Conflict of interest {#FPar2} ==================== The authors declare that they have no competing interests.

Introduction {#s1} ============ Epilepsy is one of the most prevalent chronic brain diseases worldwide and is characterized by abnormal, excessive, and synchronous neuronal activity ([@B1]). It is estimated that more than 60 million individuals are affected by epilepsy worldwide, making up 0.6% of the global economic cost of the disease, as this disease severely reduces quality of life and increases the risk of mortality ([@B2]). Cognitive dysfunction occurs in approximately 70--80% of epileptic individuals ([@B3]). In addition, recent research has demonstrated that 50% of newly diagnosed and untreated epileptic patients had cognitive impairments, and cognitive impairments were present in 64.5% of epileptic children with abnormal brain imaging ([@B4], [@B5]). The significant association between the frequency of epileptic seizures and cognitive impairment was identified by various studies ([@B4], [@B6]). In addition, the cognitive status of epileptic individuals is also affected by other factors, such as age of onset, type of epilepsy, duration, use of antiepileptic drugs (AEDs) and so on ([@B7]--[@B9]). Event-related potentials (ERP) are an objectively non-invasive approach for studying information processing and cognitive brain functions, such as attention, learning, memory, and decision-making, that are characterized by their positive or negative polarity, latency, and high temporal resolution ([@B10], [@B11]). Sutton et al. first reported a evoked potential component that reached its peak amplitude at approximately 300 ms and found a significant association between this ERP component and cognitive function ([@B12]). Subsequently, it has been identified by many studies that cognitive impairment in various brain diseases, such as Alzheimer\'s disease, Parkinson\'s disease, epilepsy, and stroke, can be objectively assessed by ERPs ([@B13]--[@B17]). Mild cognitive impairment (MCI) refers to a prodromal stage of Alzheimer\'s disease (AD) that exceeds typical "age-related" reduction in cognition but does not fulfill the criteria for AD ([@B18]). A recent meta-analysis indicated the effectiveness of ERPs in the early diagnosis of MCI. The P300 is an important and extensively explored late component of ERPs that is widely applied to assess cognitive function in humans ([@B12]). The amplitude and latency of the P300 component provides information about cognitive processes in the brain, such as memory, attention, concentration and speed of mental processing ([@B13]). Many previous studies have focused on the application of the P300 component in the assessment of cognitive impairment in epileptic patients; however, consistent conclusions were not reached for various reasons. In addition, some studies did not have sufficient statistical power due to a limited sample size. Considering these factors, we conducted a comprehensive systematic review and meta-analysis of P300-related studies to assess the latency and amplitude of the P300 in epileptic patients. This was the first meta-analysis to summarize all of the epilepsy-related P300 studies. We hypothesized that the P300 may be sensitive to detecting cognitive impairment in epileptic patients. Methods {#s2} ======= Search Strategy --------------- PubMed, EMBASE, and Cochrane Library databases were systematically searched for eligible studies by two independent reviewers (Zhong and Chen) from inception to 1st Feb 2019. The search terms were used as follows: ("event-related potential" OR "ERP" OR "P300" OR "P3") AND ("epilepsy" OR "seizure"). The publication language was restricted to English. The reference lists of eligible studies were also reviewed to identify additional articles. Two authors independently reviewed the title and abstract of each paper. After excluding obviously irrelevant articles, the authors examined the full text of the selected articles and decided the exact list of literature to be included in the present meta-analysis. The authors resolved disagreements by discussion. Selection Criteria ------------------ An eligible study should fulfill the following criteria: (1) The study design was a case-control study, cross-section study or cohort study; (2) the trial included epileptic patients and healthy controls; (3) the amplitude and the latency of the P300 component were measured by the ERP technique and were pooled; and (4) the amplitude and latency of the P300 component were compared between epileptic patients and controls. Duplicates were removed. Reviews, case reports, meta-analyses and conference abstracts were also excluded. Two reviewers (Zhong and Chen) independently carried out study selection. Data Extraction and Quality Assessment -------------------------------------- The following data were extracted by two independent reviewers: first author, publication year, population based on age group (children or adults), sample size, electrode location and stimuli (Hz). We also extracted the amplitude and latency of the P300 component in epileptic patients and controls. Two reviewers (Zhong and Chen) independently extracted data from each eligible study. The Newcastle-Ottawa Scale (NOS) was used to assess the quality of each eligible study. Studies with a score of six or higher were categorized as being high quality studies. Studies that scored between zero and five were regarded as being low quality. Two reviewers independently evaluated the quality of the included studies. Statistical Analysis -------------------- The standard mean difference (SMD) and the 95% confidence interval (CI) were calculated as the effect size of the P300 component. We used Cochran\'s Q statistic and the *I*^2^ statistic to assess heterogeneity across studies. *P* \< 0.1 or *I*^2^ \> 50% indicated significant heterogeneity. We used the random effects model if heterogeneity was present in the data; otherwise, we chose the fixed effects model. Subgroup analysis based on age-group population and type of epileptic seizure was carried out to explore potential sources of heterogeneity. Sensitivity analysis was also conducted to test the stability of the results by removing one study at a time. Publication bias was described using funnel plots. The cognitive status of epileptic patients was affected by various factors, such as the type of epilepsy, duration, and use of AEDs. Thus, the differences in the amplitude and latency of the P300 component across different types of epileptic cases were also calculated by meta-analysis. The Review Manager Software Package (RevMan 5.3) was used to conduct all statistical analyses. Results {#s3} ======= Search Results and Study Characteristics ---------------------------------------- A total of 1,359 potentially eligible references were initially identified based on our search strategy. A total of 521 articles were removed due to being duplicates, and 838 articles remained. Then, we excluded 796 articles after screening the title and abstract. A total of 42 articles remained for full text reading. Eventually, 27 eligible studies ([@B13], [@B19]--[@B44]) containing 1,513 epileptic patients and 1,124 healthy controls were included in our meta-analysis. [Figure 1](#F1){ref-type="fig"} shows the study selection process. {#F1} A total of 27 eligible studies were included in the present meta-analysis. Of these studies, 10 included children and adolescent patients with epilepsy, and 17 studies were based on adult epileptic cases. The locations of the electrodes and the stimuli (Hz) used in the tasks were substantially variable across studies. The Cz electrode and 1,000/2,000 Hz stimuli were most frequently used in these studies. Most included studies had a quality score over 6, according to the Newcastle-Ottawa Scale, indicating good quality in general. The main basic characteristics and quality assessments of the included studies are summarized in [Table 1](#T1){ref-type="table"}. ###### The main basic characteristics and quality assessment of the included studies. **References** **Age group** **Number of** **Site** **Frequent/rare stimuli(HZ)** **Quality** --------------------------------- --------------- --------------- ---------- ------------------------------- ------------- --- Elsawy et al. ([@B19]) Adults 30 30 Cz 2,000/8,000 6 Fosi et al. ([@B20]) Children 22 22 Cz, Fz, T3, T4 1,000/2,000 8 Takhirovna et al. ([@B21]) Adults 75 30 C3, C4 1,000/2,000 7 Casali et al. ([@B13]) Children 20 16 Fz 1,000/2000 7 Boscariol et al. ([@B22]) Children 19 16 Fz 1,000/2,000 7 Artemiadis et al. ([@B23]) Adults 16 43 CPz, AFz 1,000/2,000 5 Watanabe et al. ([@B24]) Adults 14 14 Cz, Fz, Pz 1,000/1,050 7 Tumay et al. ([@B25]) Adults 53 20 Cz, Pz 1,000/2,000 6 Rocha et al. ([@B26]) Adults 12 12 C2, C4 1,000/1,500 6 Ivetic et al. ([@B27]) Adults 35 20 Cz, Fz 1,000/2000 5 Sun et al. ([@B28]) Adults 30 15 C3, C4, P3, P4 NR 7 Duman et al. ([@B29]) Children 21 21 Cz, Fz 1,000/2,000 7 Chayasirisobhon et al. ([@B30]) Adults 30 30 Cz, Fz, Pz 1,000/2,000 6 Soyuer et al. ([@B31]) Adults 73 31 Cz, Fz 1,000/2,000 7 Gokcay et al. ([@B32]) Children 30 25 Cz, Fz 1,000/8,000 7 Celebisoy et al. ([@B33]) Children 120 25 Cz 1,000/8,000 6 Turkdogan et al. ([@B34]) Children 50 21 Cz, Fz, Pz 1,000/8,000 6 Chen et al. ([@B35]) Adults 27 60 P3, P4 125/750 6 Caravaglios et al. ([@B36]) Adults 108 32 Cz, Fz, Pz 1,000/2,000 6 Soysal et al. ([@B37]) Adults 116 15 Cz, Fz 1,000/8,000 6 Kubota et al. ([@B38]) Adults 120 78 Cz, Fz, Pz 1,000/2,000 7 Wu et al. ([@B39]) Adults 50 38 Cz, Fz,C3, Oz 1,000/2,000 6 Shimono et al. ([@B40]) Adults 12 9 Cz, Fz, Pz 1,000/2,000 6 Naganuma et al. ([@B41]) Children 72 67 Pz 1,000/2,000 6 Konishi et al. ([@B42]) Children 129 53 Pz 1,000/2,000 6 Naganuma et al. ([@B43]) Children 23 54 Pz 1,000/2,000 6 Triantafyllou et al. ([@B44]) Adults 68 30 Cz, Fz, Pz 1,000/2,000 6 *EG, epilepsy group; CG, control group; NR, not reported; Site, location of electrodes*. Meta-analysis Results for the Latency and Amplitude of the P300 in Epilepsy --------------------------------------------------------------------------- Twenty-six studies reported the P300 latency in epileptic patients. The pooled results showed a statistically significant difference in the P300 latency between epileptic patients and healthy controls (SMD: 0.78; 95% CI: 0.57, 0.99; *I*^2^ = 83%). In other words, individuals with epilepsy tended to have a longer P300 latency than controls. Twenty studies reported the P300 amplitude in epileptic patients. The combined analysis showed a lower P300 amplitude in individuals with epilepsy than in controls (SMD: −0.39; 95% CI: −0.59, −0.18; *I*^2^ = 71%) ([Figure 2](#F2){ref-type="fig"}). The heterogeneity was significant for both the latency and amplitude of the P300. {#F2} Subgroup Analysis Based on Age Group ------------------------------------ Of the studies reporting the P300 latency, 10 studies were conducted in children and adolescents, and 16 studies focused on adults. Both children (SMD: 0.54; 95% CI: 0.42, 0.67; *I*^2^ = 19%) and adults (SMD: 1.00; 95% CI: 0.64, 1.37; *I*^2^ = 90%) with epilepsy had a longer P300 latency than controls in our meta-analysis. Of the studies reporting the P300 amplitude, six studies were carried out in children, and 14 were in adults. Compared with healthy controls, a lower P300 amplitude was observed in both children (SMD: −0.27; 95% CI: −0.51, −0.04; *I*^2^ = 0) and adults (SMD: −0.43; 95% CI: −0.69, −0.16; *I*^2^ = 78%) with epilepsy. Subgroup Analysis Based on Type of Epilepsy ------------------------------------------- Subgroup analyses based on the type of seizure were also conducted with the aim of exploring the change in the P300 latency across different epileptic types compared with that in controls. The types of epileptic seizure mainly included temporal lobe epilepsy, idiopathic epilepsy, symptomatic epilepsy, generalized epilepsy and partial epilepsy. Compared with normal controls, a significantly longer P300 latency was observed in the temporal lobe epilepsy group (SMD: 0.86; 95% CI: 0.25, 1.47; *I*^2^ = 78%; 6 studies), idiopathic epilepsy group (SMD: 1.37; 95% CI: 0.46, 2.27; *I*^2^ = 93%; 7 studies), symptomatic epilepsy group (SMD: 1.46; 95% CI: 0.71, 2.21; *I*^2^ = 90%; 4 studies), generalized epilepsy group (SMD: 1.52; 95% CI: 0.46, 2.58; *I*^2^ = 95%; 7 studies) and partial epilepsy group (SMD: 0.87; 95% CI: 0.39, 1.35; *I*^2^ = 87%; 7 studies). The heterogeneity was still significant in all types of epileptic seizures after subgroup analysis. Sensitivity Analysis -------------------- We carried out sensitivity analysis by removing one study at a time with the aim of assessing the influence of each included study on the overall results. The significantly longer P300 latency and lower P300 amplitude seen in the epilepsy group never changed throughout our sensitivity analysis. Therefore, the results of the present meta-analysis were reliable and stable. Publication Bias ---------------- A funnel plot showing the association between the P300 latency and the P300 amplitude ([Figure 3](#F3){ref-type="fig"}) and epilepsy were drawn. In general, it seems that no obvious publication bias was observed in our meta-analysis. {#F3} Discussion {#s4} ========== The main results of the present meta-analysis indicated that epileptic patients have a longer P300 latency and a lower P300 amplitude than controls. Subgroup analysis based on age group demonstrated that these differences can also be observed in both children and adult patients compared with healthy controls. In addition, the P300 latency was longer in patients with the five main types of epileptic seizures than in controls. The P300 is one of the most important ERP components that is used to evaluate cognitive function, such as attention, working memory, and concentration ([@B12], [@B45]). The P300 component has been considered a potential marker of cognitive dysfunction, and it is mostly elicited by the oddball paradigm and appears approximately 300 ms after an infrequent stimulus. The waveform of the P300 component is described by its amplitude and latency. The P300 amplitude represents the degree of information processing, which is also associated with the amount of attentional resources allocated to a task and the level of superior cognitive function ([@B46]). Previous studies have demonstrated that the process of selective attention, stimulus evaluation time and working memory updating may be associated with the P300 latency. The prolongation of the P300 latency reflects bad cognitive performance ([@B47], [@B48]). The prevalence of cognitive impairment in epilepsy is estimated to reach 70--80% ([@B3]). The early diagnosis and intervention of cognitive impairment in epilepsy is essential for patients to improve their quality of life. Currently, psychological test scales are the main tools for evaluating cognitive dysfunction in epileptic patients. Over the past three decades, many studies have focused on the application of the P300 in epilepsy, with the aim of evaluating neurocognitive function in patients; however, the results have been inconsistent and even contradictory. A previous investigation showed that brain damage in regions important for cognitive development in epileptic children may lead to neural network impairment and extratemporal abnormalities, which are associated with language impairment and deficits in auditory processing ([@B49]). Such abnormalities may have a major influence on cognitive processes and behavior ([@B50]). In P300-related studies in epileptic children, Naganuma et al. found that epileptic children had a significantly longer P300 latency than controls ([@B43]). Similar conclusions were replicated by subsequent studies ([@B34], [@B51]). However, in several other previous studies, no differences were observed between the epilepsy group and controls in terms of the cognitive potential P300 ([@B22], [@B30], [@B52]). The results of P300-related studies based on epileptic adults were also controversial and conflicting. Additionally, the latency and amplitude of the P300 in epileptic patients has been identified as being associated with the type of seizure. However, there were inconsistent results across studies. Takhirovna et al. found that the P300 latency in patients with generalized seizures was significantly longer than that in patients with partial seizures ([@B21]). Kubota et al. found that there were no significant differences in terms of the P300 latency and P300 amplitude between either the unmedicated partial seizure group or the unmedicated generalized seizure group and the control group ([@B38]). The sample size of some studies was too small to have enough statistical power, which may account for the different results. Thus, a total of 27 studies containing 1,513 epileptic patients and 1,124 healthy controls were involved in the present meta-analysis, which can overcome this limitation. The results of our study confirmed the significant role of the P300 latency and P300 amplitude in the cognitive assessment of epileptic patients. Compared with controls, a prolonged P300 latency and a reduced P300 amplitude were found in epileptic patients, both in children and adults. We also found that longer P300 latency and lower P300 amplitude were more obvious in epileptic adults than in children. Thus, we hypothesize that cognitive impairment in adults with epilepsy is more severe than that in epileptic children. In addition, the latency and amplitude of the P300 may be associated with the type of epileptic seizure. Casali et al. found that children with temporal lobe epilepsy (TLE) had a significantly longer P300 latency than controls; however, children with benign childhood epilepsy with centrotemporal spikes (BECTS) did not have a longer P300 latency ([@B13]). The latency and amplitude of the P300 may vary largely due to patients having different types of epileptic seizures. Thus, subgroup analyses based on epileptic seizure type were carried out. We found that the P300 latency was significantly longer in patients with all types of epilepsy than in controls. In the present meta-analysis, the prolongation of the P300 latency was most obvious in the generalized epilepsy group, indicating that the cognitive impairment was most severe in generalized epilepsy. Additionally, recent progress revealed the effects of AEDs on cortical activity and connectivity which could be measured by EEG ([@B53]). This finding may partly explain the prolongation of p300 latency and reduction of p300 amplitude in epileptic patients due to AEDs treatment. The heterogeneity was significant in the present meta-analysis. We did not find the potential source of heterogeneity by subgroup analysis or sensitivity analysis. Previous studies have found that the waveform of the P300 can be easily influenced by various factors, such as the complexity of the target stimulus, type of response required, modality, drugs and IQ ([@B54], [@B55]). The location of the electrodes, stimulus frequency and stimulus intensity were widely variable across studies, which may also contribute to the heterogeneity. Recent progress has indicated that the P300 may be a attractive approach to assess cognitive impairment in epilepsy. Our meta-analysis further confirmed this hypothesis. However, there is still a long way to go. The paradigm, target stimulus, location of the electrodes, stimulus frequency and stimulus intensity should be unified first before the P300 can be applied in the cognitive assessment of epilepsy. Several limitations should be acknowledged in the explanation of our results. First, a meta-analysis may be biased when the literature search fails to identify all relevant studies. However, access to unpublished articles remains difficult, which might be a potential limitation of our study. Second, subgroup analysis based on drug type, disease duration and treatment course was not performed due to the limited studies. Third, the heterogeneity was persistently significant in our study. Conclusion {#s5} ========== In conclusion, this work revealed that the P300 latency was significantly longer and the P300 amplitude was significantly lower in the epileptic group than in the controls, both in adults and children. This prolongation of the P300 latency can be observed across different types of seizures, such as temporal lobe epilepsy, idiopathic epilepsy, symptomatic epilepsy, generalized epilepsy and partial epilepsy. This study revealed that epileptic patients have abnormalities in the P300 component of event-related potentials, which may reflect deficits in cognitive function. Thus, the P300 may be a potential objective approach for evaluating cognitive function in epileptic patients. Note that more well-designed studies with a large sample size are requested to test our findings. Author Contributions {#s6} ==================== RZ, QC, and WL designed the study, reviewed the literature, conducted the statistical analysis, and drafted of the manuscript collectively. JL, GL, and RZ performed summary tables, edited pictures, and discussed on the manuscript. ML and RZ contributed significantly to the revision of the final manuscript. 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.** This study was supported by The Excellent Youth Talents Foundation of Jilin Province (No. 3D518NP13428). [^1]: Edited by: Jeremy Daniel Slater, University of Texas, United States [^2]: Reviewed by: Gabriele Zanirati, Institute of the Brain of Rio Grande do Sul (InsCer), Brazil; Giovanni Assenza, Campus Bio-Medico University, Italy [^3]: This article was submitted to Epilepsy, a section of the journal Frontiers in Neurology

Introduction {#s1} ============ Melon (*Cucumis melo*) belongs to the *Cucurbitaceae* family that contains about 800 species mainly distributed in tropical and subtropical regions [@pone.0015776-Jeffrey1]. Cucurbitaceae includes several other economically important cultivated plants, such as cucumber (*C. sativus*), watermelon (*Citrullus lanatus*), squash and pumpkin (*Cucurbita* spp.). Aside tomato (*Solanum lycopersicum*), watermelons, cucumbers and melons are the most cultivated vegetable species (<http://faostat.fao.org>). For decades, melon has played key roles in the field of plant molecular biology and plant physiology, serving as an excellent model organism for investigating sex determination [@pone.0015776-Boualem1], [@pone.0015776-Martin1] and ripening processes [@pone.0015776-Ezura1], [@pone.0015776-Pech1]. Melon is also a reference model to investigate vascular fluxes, as xylem and phloem saps can be readily collected [@pone.0015776-Haritatos1], [@pone.0015776-Gomez1]. Melon is a diploid species (2n = 2x = 24) with an estimated genome size of 450 Mb [@pone.0015776-Arumuganathan1]. Research efforts have been undertaken in order to gather genetic and genomic informations on melon, such as EST sequences [@pone.0015776-GonzalezIbeas1], BAC libraries [@pone.0015776-vanLeeuwen1], [@pone.0015776-Morales1], oligo-based microarrays [@pone.0015776-MascarellCreus1] and high resolution genetic maps [@pone.0015776-Perin1]--[@pone.0015776-Deleu1]. Melon has truly entered in the genomic era by the sequencing of the highly syntenic cucumber genome [@pone.0015776-Huang1]. With the completion of the melon genome sequencing project in the near future [@pone.0015776-Gonzalez1], a major challenge is to determine gene functions. In plants, the most common techniques to produce altered or loss of function mutations are T-DNA or transposon insertional mutagenesis [@pone.0015776-Alonso1] and RNA interference [@pone.0015776-Waterhouse1]. However, because these methods are mainly based on the ability of a given plant to be transformed, their usefulness as general reverse genetics methods is limited to very few plant species. On the other hand, ethyl methanesulfonate (EMS) mutagenesis is a simple method to saturate a genome with mutations [@pone.0015776-Greene1]. TILLING (Targeting Induced Local Lesions IN Genomes) combines advantages of random chemical mutagenesis and high throughput mutation discovery methods [@pone.0015776-Triques1] and generates allelic series of the targeted genes which makes it possible to dissect the function of the protein as well as to investigate the role of lethal genes. This technique has been successfully applied to a large variety of organisms including plants and animals [@pone.0015776-Greene1], [@pone.0015776-Dalmais1]--[@pone.0015776-Gilchrist1]. To establish a reverse genetics platform in melon and to identify novel alleles of agronomic importance, we have set up a melon TILLING platform and performed a screen for mutations in genes that control fruit ripening. Fruit ripening is a process characterized by a number of biochemical and physiological processes that alter fruit firmness, colour, flavour, aroma, and texture [@pone.0015776-Brady1], [@pone.0015776-Brummell1]. Ripening patterns are conserved across fleshy fruit species. Characterizations of homologous genes involved in fruit ripening in different species suggest that genetic mechanisms are also conserved [@pone.0015776-AdamsPhillips1]. In the fleshy fruit model, tomato, the plant hormone, ethylene, plays a central role in ripening [@pone.0015776-Giovannoni1]--[@pone.0015776-Barry1]. Several fruit ripening mutants have been characterized in tomato [@pone.0015776-Giovannoni3]. The *rin* (*ripening-inhibitor*), *nor* (*non-ripening*) and *cnr* (*colorless non-ripening*) genes were shown to act upstream the ethylene signaling in ripening [@pone.0015776-Giovannoni3]--[@pone.0015776-Manning1]. Mutations in the ethylene receptor, the *Nr* (*Never-ripe*) gene, were also shown to affect fruit ripening [@pone.0015776-Wilkinson1]. Ethylene biosynthesis requires the conversion of aminocyclopropane-1-carboxylic acid (ACC) to ethylene by the ACC oxidase (ACO). ACC oxidases are encoded by multigene families in plants [@pone.0015776-Alexander1], [@pone.0015776-Lin1] and have been described to be involved in ripening, growth, and development. In melon, *CmACO1* silencing inhibits fruit ripening and extends fruit shelf life [@pone.0015776-Ayub1]. Melon is an attractive model for investigating fruit ripening. Unlike tomato, melon has two different ripening patterns, climacteric and non climacteric. To investigate further the role of ethylene in melon fruit ripening, we have developed a reference EMS mutant population under controlled conditions and established a TILLING platform. Then, we screened for mutations in 11 genes, mainly involved in ripening processes, and characterized *CmACO1* TILLING mutants. This work yields a missense mutation in *CmACO1* that inhibits fruit ripening and extends fruit storage life. The use of TILLING as a translational research tool is discussed. Results {#s2} ======= Production of CharMono mutant collection {#s2a} ---------------------------------------- The melon inbred line CharMono is a monoecious climacteric Charentais type cultivar (*Cucumis melo* L. subsp. *melo* var *cantalupensis*) that bears male flowers on the main stem and female flowers on axillary branches. The success of the TILLING approach relies on the construction of high quality mutant libraries. Ideally the mutant population must produce a mutation frequency that is conducive to high-throughput screening but is below a threshold that causes extensive sterility and plant development alteration. To optimize the mutagenesis, we first conducted a "kill-curve" analysis on batches of 100 seeds, using a dose range from 1% to 3% EMS ([Table 1](#pone-0015776-t001){ref-type="table"}). Most M1 treated plants exhibited growth retardation at early seedling stage, but all of them recovered, except for the 3% EMS fraction for which only 40 plants were recovered. The M1 plants were assessed for seed production and seed germination. Low seed production was observed at 3% EMS with 37% of the plants producing less than 40 seeds per fruit. At 2% EMS, only 9% of the plants yielded less than 40 seeds per fruit ([Table 1](#pone-0015776-t001){ref-type="table"}). Thus, the highest EMS doses allowing good seed set and seed germination, 1.5% and 2%, were retained and tested on large batches of seeds. The EMS treated seeds were sown in soil and seedlings were grown to fruit maturity in insect-proof plastic tunnels to avoid cross-pollination. Female flowers were hand pollinated with male flowers from the same plants, bagged and the fruits left to develop to maturity. M2 seeds were harvested from individual M1 plants. We also produced M2 seeds from 617 and 40 M1 plants treated with 1% and 3% EMS, respectively. In total, 4260 M2 seed stocks were collected. 10.1371/journal.pone.0015776.t001 ###### Impact of EMS concentration on M2 seed setting and germination. {#pone-0015776-t001-1} EMS dose M1 plants Fruits with less than 40 seeds (%) Seed germination (%) ----------- ----------- ------------------------------------ ---------------------- **1%** 100 0 99 **1.50%** 100 1 99 **2%** 100 9 98 **3%** 40 37 97 To assess the quality of the mutagenesis, we investigated the rate of appearance of depigmentation mutants at the cotyledon stage and developmental alterations at fruit maturity. Albino and chlorotic plants, the most frequently observed phenotype in mutagenized populations [@pone.0015776-Dalmais1], [@pone.0015776-Wu1], occurred at the rate of 1,3% of the M2 families. This rate is similar to the rate reported for other well characterized mutant collections [@pone.0015776-Chawade1]--[@pone.0015776-Wang1]. The most commonly observed developmental phenotypes were related to cotyledon number and morphology, dwarfism, shoot branching and plant architecture. At the flowering stage, we observed mutants with flower sex type transition, abnormal flower morphology and architecture. At the fruiting stage, altered fruit shape and size, ovary number and flesh color and thickness variations were also observed. Examples of the observed phenotypes are shown in [Figure 1](#pone-0015776-g001){ref-type="fig"}. {#pone-0015776-g001} Melon TILLING platform {#s2b} ---------------------- To set up the TILLING platform, DNA samples were prepared from 4023 M2 families, each representing an independent M1 family and organized in pools of 8 families as described previously [@pone.0015776-Piron1]. One key factor in TILLING is the availability of an annotated genomic sequence of the gene to be tilled, which is facilitated by the ongoing sequencing of melon genome (International Cucurbit Genomics Initiative, [www.icugi.org](http://www.icugi.org)), the availability of a large melon EST collection [@pone.0015776-GonzalezIbeas1] and the 95% sequence similarity over coding regions between melon and the available cucumber genome sequence [@pone.0015776-Huang1]. Primer design and PCR amplifications were performed as described previously [@pone.0015776-Dalmais1]. Mutations were detected, in the amplified targets, using the mismatch-specific endonuclease ENDO1 [@pone.0015776-Triques1]. Individual mutant lines were identified following a pool deconvolution step, and the mutated bases were identified by sequencing. To assess the quality of CharMono mutant collection and to estimate the mutation density, we screened for induced mutations in 11 genes related to fruit quality. In total, we identified and confirmed by sequencing 134 induced mutations in 18,3 kb total length of tilled amplicons ([Table 2](#pone-0015776-t002){ref-type="table"}). Most induced mutations were as expected, G/C to A/T transitions [@pone.0015776-Greene1], with the exception of the three following mutations, G/C to C/G, T/A to A/T and C/G to A/T ([Table S1](#pone.0015776.s003){ref-type="supplementary-material"}), which have also been reported in other TILLING studies [@pone.0015776-Minoia1], [@pone.0015776-Cooper1], [@pone.0015776-Till3], [@pone.0015776-Caldwell1]. 10.1371/journal.pone.0015776.t002 ###### Tilled genes and mutation frequency in CharMono mutant population. {#pone-0015776-t002-2} Targets Function Amplicon size (pb) GC content (%) Identified mutants Screened M2 families Mutation frequency ---------- --------------------------- -------------------- ---------------- -------------------- ---------------------- -------------------- *CmDET1* light signaling pathway 3193 35.11 24 2538 1/337 kb *CmDHS* eIF5A activation 2434 36.48 19 2538 1/325 kb *CmACO1* ethylene biosynthesis 1469 40.16 7 3306 1/694 kb *CmNOR* fruit ripening process 1272 39.70 9 3306 1/467 kb *CmEXP1* cell-wall modification 1168 46.23 11 3306 1/351 kb *CmSP1* Inflorescence development 1055 33.93 3 4023 1/1415 kb *CmTCTP* cell growth 1512 36.24 11 4023 1/553 kb *CmCNR* fruit ripening process 1333 40.36 11 4023 1/487 kb *CmSGR* chlorophyll degradation 1383 37.24 13 4023 1/428 kb *CmACS7* sex determinism 1680 38.57 8 4023 1/845 kb *CmWIP1* sex determinism 1812 31.95 18 4023 1/405 kb **Total** **18311** **37.82** **134** **-** **1/573 kb** The size of the tilled amplicons, the GC content, the number of induced alleles and the mutation frequency per amplicon are shown. The average mutation frequency in the collection is estimated to one mutation per 573 kb and is calculated as described previously [@pone.0015776-Greene1]. Sequence analysis of the exonic induced mutations showed that 31,3% were silent, 65,1% were missense, 2,4% were stop and 1,2% were splice junction mutations ([Figure S1](#pone.0015776.s001){ref-type="supplementary-material"}, [Table 3](#pone-0015776-t003){ref-type="table"}). The number of stop mutations was lower than the CODDLE predicted proportions. In contrast, silent, missense, and splice junction mutations were recovered in an equivalent proportion as predicted ([Table 3](#pone-0015776-t003){ref-type="table"}). As many tilled amplicons harbored non coding segments, some recovered mutations could potentially affect the splicing or the stability of the mRNA, such impacts are unpredictable. Thus, non coding mutations were not characterized further. In contrast, the many non-synonymous mutations discovered were investigated to dissect the function of the proteins as they may lead to gain- or loss-of-function phenotypes. We estimated the mutation density based on Greene *et al*. [@pone.0015776-Greene1] to 1 mutation every 573 kb. This observed density is proportional to the EMS dose used, except for the 1.5% EMS fraction showing a lower mutation frequency may be due to some experimental variations ([Table 4](#pone-0015776-t004){ref-type="table"}). 10.1371/journal.pone.0015776.t003 ###### Expected and observed frequencies of induced mutation types in tilled gene-coding regions. {#pone-0015776-t003-3} missense nonsense splicing silent ---------- ---------- ---------- ---------- -------- expected 64.0% 5.0% 1.7% 29.3% observed 65.1% 2.4% 1.2% 31.3% The percentage of the expected mutations were calculated using the CODDLE program as described previously [@pone.0015776-Dalmais1]. 10.1371/journal.pone.0015776.t004 ###### Effect of EMS dose on mutation frequency in CharMono mutant collection. {#pone-0015776-t004-4} EMS dose M1 plants Induced mutations Mutation frequency ----------- ----------- ------------------- -------------------- **1%** 617 19 1/588 kb **1.50%** 2130 45 1/848 kb **2%** 1473 67 1/356 kb **3%** 40 3 1/146 kb **Total** **4260** **134** **1/573 kb** Characterization of *CmACO1* TILLING mutants {#s2c} -------------------------------------------- Fruit softening is a major factor that determines fruit quality and shelf life. In melon and tomato, silencing of enzymes or regulators of the ethylene biosynthesis pathway inhibits fruit ripening and extend fruit storage life [@pone.0015776-Ayub1], [@pone.0015776-Silva1]--[@pone.0015776-Xiong1]. To identify melon lines with improved fruit shelf life, we screened for mutations in the CmACO1 enzyme that catalyses the last step of ethylene biosynthesis ([Figure 2A](#pone-0015776-g002){ref-type="fig"}). In this TILLING screen, we identified seven independent point mutations among which two led to L124F and G194D missense mutations. The L124F and G194D changes occurred in a highly conserved amino acid positions and may therefore affect the function of the protein ([Figure 2B](#pone-0015776-g002){ref-type="fig"}, [@pone.0015776-Ng1]). However, X-ray crystallography studies determined that L^124^ is located in the α-5 helix of the protein away from the active site and thus, it is predicted to not affect the function of the protein ([Figure 2C--E](#pone-0015776-g002){ref-type="fig"}, amino acid indicated in green; [@pone.0015776-Zhang1]). In contrast, the residue G^194^ is located in the β-7 strand, one of the eight β strands (β-4 to β-11) that form the distorted double-stranded β helix (DSBH or jellyroll) core of the active site, common to all members of the 2-OG-oxygenases [@pone.0015776-Zhang1]--[@pone.0015776-Elkins1]. The residue G^194^ is located in the inner face of the active site, and thus, a mutation at this position is predicted to affect the function of the protein ([Figure 2C-E](#pone-0015776-g002){ref-type="fig"}, amino acid indicated in red; [@pone.0015776-Zhang1]). ![Sequence and structural analysis of *CmACO1.*\ (A) Schematic diagram of *CmACO1* gene structure. The numbers indicate the size of the 4 exons (filled boxes) and the 3 introns (black lines) in bps. Intronic, silent and the L124F and G194D missense mutations are represented by black, grey, green and red triangles, respectively. (B) Amino acid alignment of CmACO1 (*Cucumis melo*, Q04644) and homologous proteins from *Cucumis sativus* (Cs, BAA33377), *Arabidopsis thaliana* (At, NP_171994), *Solanum lycopersicum* (Sl, P05116), *Petunia x hybrida* (Ph, Q08506), *Vitis vinifera* (Vv, XP_002273430), *Medicago truncatula* (Mt, AAL35971), *Populus trichocarpa* (Pt, XP_002320487) and *Oryza sativa* (Os, NP_001063330). Numbers above the alignment indicate the amino acid positions along the CmACO1 protein. L124F and G194D EMS-induced mutations are shown above the alignment in green and red, respectively. (C--D) 3D structure model of CmACO1. (C) Superposition of the Petunia ACO structure determined by X-ray crystallography [@pone.0015776-Zhang1], indicated in pink and the 3D model of CmACO1, indicated in white. The melon model was determined using the Geno3D server (<http://geno3d-pbil.ibcp.fr>). (D and E) Zoom in of the ACO1 active site. Yellow sticks represent Ile171, His177, Asp179 and His234 residues involved in the Fe (II) cofactor binding (orange ball). Blue sticks represent the ligating phosphate or sulfate ion. Amino acids L^124^ and G^194^ are represented in green and red sticks, respectively. (F--K) Fruit-related traits of the L124F and G194D EMS mutants. Data for the duration between the pollination and the fruit ripening (days, F), firmness (Durofel index, G), shape (ratio length/width in mm, H), soluble sugars (° Brix, I) and flesh (J) and rind (K) color (b value) are shown. G124D mutant and its relative wild type control and L124F and its relative wild type control are indicated by red, black, green and open bars, respectively. Asterisks indicate significant differences.](pone.0015776.g002){#pone-0015776-g002} To test whether induced mutations in *CmACO1* could affect melon fruit ripening, L124F and G194D TILLING mutants were evaluated for different fruit traits ([Figure 2F--K](#pone-0015776-g002){ref-type="fig"}). The L124F mutant did not show any difference from its wild type control for the evaluated fruit traits. This is consistent with the conservative L124F mutation, two amino acids with hydrophobic side chains, and the position of the L^124^ residue away from the active site of the protein ([Figure 2F--K](#pone-0015776-g002){ref-type="fig"}, white and green bars). In contrast, the duration between the pollination and the fruit ripening as well as the fruit firmness were significantly increased in the G194D mutant compared to the wild type control ([Figure 2F and G](#pone-0015776-g002){ref-type="fig"}, black and red bars). G194D mutation was also associated with a significant change in the fruit rind color leading to a delayed fruit yellowing ([Figure 2K](#pone-0015776-g002){ref-type="fig"}). Fruit shape, soluble sugar content and flesh color showed no differences between the G194D mutant line and the wild type control ([Figure 2H--J](#pone-0015776-g002){ref-type="fig"}). No differences were also observed for the fruit peduncle abscission layer that is activated at fruit maturity and lead the fruit to drop from the plant. To confirm the observed phenotypes, G194D heterozygote mutant lines were selfed and 40 homozygote mutant and wild type segregants were assessed for the above fruit traits. Homozygous wild type plants for this locus had no alteration of any fruit traits described above and behaved as the CharMono wild type parent. As predicted, plants homozygous for the G194D mutation presented an increased duration between the pollination and the fruit ripening, a better fruit firmness and a delayed fruit yellowing without any changes in fruit shape, peduncle abscission layer, soluble sugar content and flesh color. These data confirmed that the observed phenotypes resulted from G194D induced mutation in *CmACO1*. To investigate the robustness of the observed phenotypes, we repeated the phenotypic characterization of the mutant lines in two different locations, Montfavet and Saint Rémy de Provence, and confirmed the observed phenotypes ([Figure S2](#pone.0015776.s002){ref-type="supplementary-material"}). Discussion {#s3} ========== In the genomic era and in the absence of efficient tools for homologous recombination, TILLING has become an obligate technology to dissect gene function as well as to engineer alleles of agronomic importance in crops. To set up the melon TILLING platform, we first developed a reference EMS mutant collection under controlled conditions. To assess the quality of the mutagenesis, we estimated the rate of occurrence of chlorotic and albino phenotypes. The observed frequency of 1,3%, comparable to well characterized mutant collections, confirmed the quality of the mutagenesis [@pone.0015776-Chawade1]--[@pone.0015776-Wang1]. Genomic DNA was prepared from the mutant lines and organized in pools for bulked screening using the mismatch specific endonuclease, Endo1 [@pone.0015776-Triques1]. The value of this melon mutant collection was then confirmed by screening for mutation in 11 genes. On average, we identified 8 alleles per tilled kilobase, calculated from the tilled 18,5 kb and the 134 identified alleles. We also estimated the overall mutation rate as one mutation every 573 kb in our CharMono mutant collection. This mutation frequency is two fold lower than the rate of one mutation per 200 kb reported for *Pisum sativum* [@pone.0015776-Dalmais1], two fold higher than the rate of one mutation per megabase reported for barley [@pone.0015776-Caldwell1] and equivalent to the rate of mutations described in tomato [@pone.0015776-Minoia1], [@pone.0015776-Piron1]. A much more saturated mutation density has been observed in polyploid species which withstand much higher doses of EMS without obvious impact on plant survival (1/40 kb in tetraploid wheat and 1/24 kb in hexaploid wheat [@pone.0015776-Slade1]). Fruit ripening and softening are key traits that determine the shelf life of fleshy climacteric fruits. Post-harvest losses can reach 70% in developing countries because of the lack of post-harvest infrastructure to store and to retail the commodities. A wide list of approaches, ranging from the application of growth regulators to delay ripening, fruit storage in controlled atmospheres and breeding or genetic engineering of leader lines have been used to control fruit ripening and softening. Ethylene has been identified as the major hormone that controls fruit ripening and softening in climacteric fruits. Investigations in the model species as well as in crops have accumulated considerable evidences at the genetic, physiological, biochemical and molecular levels that pointed out the ethylene function at various levels. This includes ethylene biosynthesis, its perception by the target cells through specific receptors, signal transduction pathway involving both positive and negative regulators and finally regulation of target genes expression by transcription factors such as ethylene response factors [@pone.0015776-Lin1]. The final step of ethylene biosynthesis in plants is catalyzed by the ACC oxidase which converts ACC to ethylene. ACC oxidases belong to a multigene family and members have been shown to be highly expressed during fruit ripening in climacteric fruits. ACC oxidase role in ripening was demonstrated by co-suppression of the fruit ACC oxidases in fleshy fruit plants. The ACC oxidases silenced fruits exhibited non-softening phenotypes and thus, an extended shelf life [@pone.0015776-Ayub1], [@pone.0015776-Silva1]--[@pone.0015776-Xiong1]. The TILLING of *CmACO1* identified a mutation at the conserved G^194^ residue, located in the active site of the enzyme. Detailed phenotypic characterization of the TILLING mutants showed that the G194D mutation mimic the phenotype of *CmACO1* antisense line [@pone.0015776-Ayub1]. The G194D mutant line showed longer fruit maturation, enhanced fruit firmness and delayed fruit yellowing. As reported for the antisense ACC oxidase melon lines [@pone.0015776-Ayub1], there are also no difference in the fruit sugar content, shape or flesh color. Besides, we noticed that the peduncle abscission layer that is activated at fruit maturity in wild type plant was also observed on the G194D mutant fruits in contrast with the antisense melon lines that failed to drop even at a very late developmental stage (65 days) [@pone.0015776-Ayub1]. This difference could be explained by the antisense strategy that can not, with complete conviction, exclude that other related homologous enzymes are affected in the peduncle abscission layer. In conclusion, to investigate fruit maturation and fruit softening we tilled a list of candidate genes and isolated key mutants that enhanced melon fruit quality (data not shown). Through the TILLING approach we knocked out a specific gene, *CmACO1* and pointed out an important amino acid, Glycine at the position 194, for the fruit maturation, not reported previously. Moreover, we created a new allele for the melon post-harvest management (maturation duration) and transport (firmness), which is an economically attractive trait for breeders. This is even more important regarding the transgenic strategies that face strong concern in public acceptance. The characterization of the other mutants will likely deliver alleles that may be exploited in melon breeding. Cucurbits are also an important model species in many key areas of plant research, including sex determination [@pone.0015776-Boualem1], [@pone.0015776-Martin1], [@pone.0015776-Boualem2], fruit maturation [@pone.0015776-Ezura1], [@pone.0015776-Pech1] and the investigation of vascular trafficking of molecules [@pone.0015776-Haritatos1], [@pone.0015776-Gomez1]. Hence, by opening the TILLING platform to the scientific community, we hope to fulfill the expectations of both crop breeders and scientists who are using melon as their model of study. Materials and Methods {#s4} ===================== Plant material and EMS mutagenesis {#s4a} ---------------------------------- Experiments were carried out using the melon inbred line CharMono, a monoecious climacteric Charentais type cultivar (*Cucumis melo* L. subsp. *melo* var *cantalupensis*). Mature seeds from CharMono line were immersed in bottles containing 250 ml of EMS diluted at right concentration in deionized water, and placed on a rotary shaker overnight (16 h) at 24°C in the dark. The EMS treatment was stopped by addition of 100 ml of Na~2~SO~3~ at 0,1 M. Treated seeds were then transferred into a new solution of Na~2~SO~3~ 0,1M for 15 minutes. The EMS solution was then removed and seeds were washed extensively. Treated seeds were sown in soil and grown under insect-proof plastic tunnels according to standard melon agronomic practice. At the end of the fruit ripening phase, M2 seeds were collected from individual M1 plants and stored. Genomic DNA extraction and pooling {#s4b} ---------------------------------- Sixteen melon leaf discs (diameter 10 mm) from eight individual plants per M2 family were collected in 96-well plates containing 2 steel beads (4 mm) per well, and tissues were ground using a bead mill. Genomic DNA was isolated using the Dneasy 96 Plant Kit (Qiagen, Hilden, Germany). DNAs were quantified on a 1% agarose gel using λ DNA (Invitrogen, Carlsbad, USA) as a concentration reference. DNA samples were then diluted ten fold and pooled eight fold in a 96-well format. PCR amplification and mutation detection {#s4c} ---------------------------------------- PCR amplification is based on nested-PCR. The first PCR amplification is a standard PCR reaction using target-specific primers ([Table S2](#pone.0015776.s004){ref-type="supplementary-material"}) and 4 ng of melon genomic DNA. One µl of the first PCR served as a template for the second nested PCR amplification, using combination of specific primers carrying M13 tail and M13 universal primers, M13F700 (5′-CACGACGTTGTAAAACGAC-3′) and M13R800 (5′-GGATAACATTTCACACAGG-3′), labelled at the 5′end with infra-red dyes IRD700 and IRD800 (LI-COR®, Lincoln, Nebraska, USA), respectively. Mutation detection was carried out as described previously [@pone.0015776-Dalmais1]. The identity of the mutations was determined by sequencing. TILLING request should be addressed to the corresponding author. Sequence analysis tools {#s4d} ----------------------- CODDLE (Codons Optimized to Discover Deleterious Lesions, <http://www.proweb.org/coddle/>) was used to identify regions of the target gene in which G/C to A/T transitions are most likely to result in deleterious effects on the protein. PARSESNP (Project Aligned Related Sequences and Evaluate SNPs, <http://www.proweb.org/parsesnp/>) was used to illustrate the distribution of mutations within the gene, and to indicate the nature of each single mutation. SIFT (Sorting Intolerant from Tolerant, <http://sift.jcvi.org/www/SIFT_seq_submit2.html>) was used to predict the impact of the mutation on the protein. Multiple sequence alignment of full-length protein sequences was performed using ClustalW (<http://www.ebi.ac.uk/Tools/clustalw2>). Protein structure modeling {#s4e} -------------------------- The CmACO1 three-dimensional structures were generated using the Geno3D server (<http://geno3d-pbil.ibcp.fr>). Superposition of the petunia ACO structure (1W9Y.pdb) determined by X-ray crystallography [@pone.0015776-Zhang1], and our CmACO1 model, was carried out and visualized using Chimera (<http://www.cgl.ucsf.edu/chimera>). TILLING mutant phenotyping {#s4f} -------------------------- Mutant and wild type plants were grown in two locations, Montfavet and Saint Rémy de Provence, in France, under greenhouse conditions. Fruits were harvested at maturity, according to the harvest indicators usually observed in Charentais type fruits, namely: the development of abscission layer, yellow rind and death of the first leaf beside the peduncle. At maturity, the duration between pollination and harvesting was evaluated. Mature fruit shape was measured as the ratio of the polar diameter (peduncle to blossom end), referred to as fruit length, to the equatorial width of the fruit (measured halfway between the peduncle and blossom end). Soluble sugar content of the flesh was measured with a hand refractometer. Firmness was measured in two different areas of the flesh after cutting the fruit lengthwise, using a Durofel electronic penetrometer or with Gullimex penetrometer equipped with an 8 mm diameter probe. Rind and flesh colors were recorded using a Minolta colorimeter at two different sites per fruit. The (b) parameter was selected for monitoring changes in the yellow hue [@pone.0015776-Voss1]. Supporting Information {#s5} ====================== ###### **Graphic representations of genes screened for mutations.** Black boxes represent the exons. Lanes linking exons indicate introns. Dashed lines in red indicate the genomic regions screened for mutations. Triangles pointing up indicate mutations in coding regions, whereas those pointing down indicate mutations in intron-exon splicing sites. Red, black and grey triangles represent alterations causing truncations, missense and silent mutations, respectively. Mutations in introns, 5′ and 3′UTR are not shown. (TIF) ###### Click here for additional data file. ###### **Multi-location evaluation of fruit traits for the G194D EMS mutant.** (A--F) Fruit-related traits tested at INRA Avignon experimental station in Montfavet, France and (G--H) at a second experimental station at Saint Rémy de Provence, France. Data for the duration between the pollination and the fruit ripening (days, A and G), firmness (Durofel index, B and kg/0,5 cm^2^, H), shape (ratio length/width in mm, C and I), soluble sugars (° Brix, D and J) and flesh (E and K) and rind (F) color (b value) are shown. G124D mutant and its relative wild type control are indicated by red and black bars, respectively. Blue stars indicate significant difference. Values represent the means of measurements obtained for at least 10 fruits from different plants. (TIF) ###### Click here for additional data file. ###### **Missense, STOP and splicing induced mutations.** Nucleotide substitution, amino acid changes and the predicted impact on the protein function are reported for each EMS-mutant line. Asterisks indicate non conventional EMS mutations. The amino acid substitution is predicted damaging if the score is \< or  = 0.05 and tolerated if the score is \>0.05. (XLS) ###### Click here for additional data file. ###### **Primers used in this study.** Asterisk indicate primers labeled with infra-red dyes IRD700 (forward primers) and IRD800 (reverse primers). (XLS) ###### Click here for additional data file. The authors thank N. Giovinazzo, D. Besombes, V. Chareyron, P. Audigier, C. Lepage and the INRA Avignon experimental team for the plant care, seed production and technical assistance. We thank F. de Langen and D. Potey (Clause Company, Saint Rémy de Provence, France) for their contribution to the mutants phenotyping, J. Eleblu, B. Lasseur and C. Clepet for comments on the manuscript. We are grateful to the seed companies, ASL, Gautier Semences and Takii France for their help to create the melon mutagenized population. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work was supported by INRA-KSU Plant biotechnology network and the MELRIP Plant-KBBE program. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: AB AB CD AAA AAA. Performed the experiments: FDM CT SL CD. Analyzed the data: AB AB FDM CT SL CD. Contributed reagents/materials/analysis tools: AB AB FDM CT SL CD. Wrote the paper: AB AB FDM.

Introduction ============ The morbidity, mortality and costs related to diabetes are escalating epidemically. A US citizen born in 2000 will have a 33% lifetime risk of developing diabetes.[@b1-clep-1-125]--[@b3-clep-1-125] Eye complications are important adverse effects of diabetes. Reduced intraocular blood flow in diabetic patients causes secondary nerve cell damage and leads to diabetic retinopathy.[@b4-clep-1-125] Diabetes may also cause a thicker central corneal thickness[@b5-clep-1-125] and a rise in intraocular pressure (IOP)[@b6-clep-1-125],[@b7-clep-1-125] made worse by poor long-term control of diabetes.[@b6-clep-1-125] Long-term control of diabetes can be assessed with measurements of glycosylated hemoglobin (HbA~1c~). [@b8-clep-1-125],[@b9-clep-1-125] Glaucoma is the most common cause of irreversible blindness worldwide.[@b10-clep-1-125] The visual field loss in glaucoma caused by loss of retinal nerve cells can be worsened by nerve cell apoptosis, which is increased in diabetic patients. The previous epidemiologic studies that examined diabetes as a potential risk factor for glaucoma had conflicting conclusions.[@b7-clep-1-125],[@b11-clep-1-125]--[@b16-clep-1-125] Moreover, while hyperglycemia is known to play a role in the development of diabetic retinopathy, to our knowledge no study has examined the association between the level of glycemic control and risk of glaucoma. We hypothesized that diabetes is a risk factor for glaucoma, and that poor long-term glycemic control and longer duration of diabetes may increase the risk of glaucoma. We conducted a large case-control study using population-based Danish medical databases. Methods ======= We conducted the study in the former Danish counties of Northern Jutland and Aarhus, with a total population of 1.14 million mixed rural and urban inhabitants, representing 20% of the Danish population.[@b17-clep-1-125] The population is 93% Caucasian, and less than 2% migrate annually from or into the counties.[@b17-clep-1-125] The Danish National Health Service provides tax-financed primary medical care and free access to hospitals to all citizens. Furthermore, the National Health Service partially reimburses the costs of prescription medications, including those for glaucoma and diabetes. We linked data from several Danish population-based databases and medical registries. The unambiguous individual-level linkage is possible thanks to the Danish Civil Registration System, whereby a unique 10-digit civil registration number (CRN) encoding date of birth and sex is assigned to every Danish citizen at birth. The Civil Registration System[@b18-clep-1-125] was established in 1968. It is updated daily and tracks vital status, residence, and migrations. The CRN is used in all public databases to uniquely identify individuals.[@b18-clep-1-125] Cases of medically treated glaucoma ----------------------------------- Glaucoma treatment cases were identified and categorized using the Danish National Registry of Patients (DNRP)[@b19-clep-1-125] and prescription databases of the two former counties merged into a research database at Aarhus University. Established in 1977, the DNRP tracks all hospitalizations and, from 1994 onwards, all outpatient visits. Recorded data include dates of admission and discharge, surgical procedures performed, and up to 20 diagnoses per visit or hospitalization. The diagnoses are coded by medical doctors using the 8th revision of the International Classification of Diseases (ICD-8) through the end of 1993 and the 10th revision (ICD-10) thereafter. The prescription databases hold information on every sale of prescription medication in the counties, including date of dispensing and type of drug coded according to the Anatomical Therapeutic Chemical (ATC) classification system. We defined an incident case of glaucoma treatment as a person filling three or more prescriptions for a glaucoma medication (ATC code group S01E) over 365 days or less during the study period, from January 1st, 2001 to December 31st, 2006. We required a minimum of three filled prescriptions on separate dates in order to exclude cases of possible short-term secondary glaucoma associated with prior ocular surgery or uveitis. Glaucoma medication is sold only by prescription in Denmark and is prescribed for primary or secondary glaucoma or for ocular hypertension (OHT), but not for any other condition. We did not consider as cases patients with prescription for Acetazolamide tablets (ATC code S01 EC01), since these are used for other indications, including high-altitude sickness; moreover, they are rarely used for long-term glaucoma treatment. The date on which the patient filled his or her third prescription for a glaucoma medication was considered the glaucoma case index date. In order to be eligible for inclusion in the study, persons had to reside in one of the counties for at least five years before the index date. In order to only include patients with incident glaucoma, we excluded persons with prescriptions for a glaucoma medication filled within five years prior to the beginning of the study period and patients with a previous hospital contact with angle-closure glaucoma in the DNRP (ICD-10 diagnosis H40.2). We also excluded patients with glaucoma surgery or laser treatment prior to index date (procedure codes KCHD all, or KCHF 05-10-15-20-30 or 99, obtained from the DNRP). We identified incident glaucoma cases based on glaucoma medications rather than ICD-10 codes in our study because the majority of glaucoma patients in Denmark are treated in private practices that do not report diagnosis codes to the DNRP. Population controls ------------------- Using data from the Civil Registration System to ascertain vital status and residence at the time of sampling, we randomly selected 10 population controls for each glaucoma case, matching birth-year, sex, and residence in one of the two counties for at least five years prior to the case's index date. Control selection was done by risk set sampling,[@b20-clep-1-125],[@b21-clep-1-125] whereby controls had to be alive and at risk of filling their third prescription for a glaucoma medication on the index date of their corresponding case. Odds ratios (OR) calculated from samples drawn using risk-set sampling provide estimates of the incidence rate ratio that would be observed in the underlying cohort.[@b22-clep-1-125] In this study OR was used as a measure of relative risk. The same eligibility and exclusion criteria were applied to cases and controls. Identification of diabetes -------------------------- Presence of diabetes for all subjects was ascertained by a validated method using prescription and hospital data.[@b23-clep-1-125] We used the DNRP to identify all persons with a hospitalization or outpatient diagnosis of diabetes mellitus recorded since 1977 (ICD-8 codes 249--250 and ICD-10 codes E10--14). From the counties' prescription databases we identified all persons with at least one prescription for insulin or an oral antidiabetic drug. Diabetes was classified as type 1 diabetes mellitus (T1DM) for patients first diagnosed before the age of 30 years, treated with insulin monotherapy, and without history of taking an oral antidiabetic medication; the remaining diabetes patients were considered to have type 2 diabetes mellitus (T2DM). The use of insulin and oral antidiabetic medication was ascertained by selecting prescriptions with ATC codes A10A and A10B recorded in the prescription databases.[@b23-clep-1-125],[@b24-clep-1-125] Duration of diabetes was computed as the time elapsed from the first record of diabetes treatment or hospital diagnosis until the index date. Information on HbA~1c~ levels was obtained by linkage of the regional laboratory databases.[@b25-clep-1-125] These databases contain information on all specimens submitted for analysis by hospitals and practitioners, including the exact time of blood sample collection. The most recent measurement was used for analysis. We only could identify 27 T1DM patients in the case group; of these, 25 (93%) had vitreoretinal surgery recorded before the index date, indicating secondary glaucoma. We refrained from making individual analysis on T1DM patients as this subgroup was too small and analyzed all diabetes patients together. Covariates ---------- We retrieved data on variables that are putative risk factors for glaucoma and also may be associated with diabetes. A recent review article identified cardiovascular events, hypertension, thyroid disease and migraine as risk factors for glaucoma.[@b26-clep-1-125] We considered as potential confounders the following conditions, which were considered present if they were identified by a diagnosis in DNRP or by a relevant medication in the prescription database within five years prior to the index date for cases and matched controls: cardiovascular disease (ICD-10: E78; I20--26, 35.0--35.2, 60--61, 63--67, 69, 70, 74, 82), hypertension (ICD-10: I10--15), thyroid disease (ICD-10: E01--07, O90.5 and ATC: H03), migraine (ICD-10: G43 and ATC: N02C), autoimmune disorders (ICD-10: M05--9, M30--35, D86, K50--51, MO2.3, M45.9, M02.9, M03.2, M03.6, I73), alcoholism-related diagnoses (ICD-10: F10, K86.0, Z72.1, R78.0, T51, K 29.2, G62.1, G72.1, G31.2, I42.6), cataract (ICD-10: H25--26, 28.0 and procedure codes KCJA,C,D,E), retinal detachment (ICD-10: H33.0 and procedure codes KCKC60,70, KCKD25,30) and uveitis (ICD-10: H20, 22, 30, 32.0 and ATC-codes S01BA). Furthermore, we ascertained use of potential IOP-lowering systemic drugs: oral beta-adrenoceptor blockers (ATC: C07), statins (ATC: C10AA), and angiotensin-converting enzyme (ACE) inhibitors (ATC: C02E, C02L, C09A, C09B, C08DA51).[@b27-clep-1-125]--[@b30-clep-1-125] Statistical analysis -------------------- We used conditional logistic regression to compute the crude and adjusted OR with 95% confidence intervals (CI) for glaucoma according to presence of diabetes. Diabetes exposure was categorized by duration of diabetes (\<5 years, ≥5--\<10 years, ≥10 years) and HbA~1c~ levels (\<7%, ≥7%--\<8%, ≥8%--\<9%, ≥9%, unknown).[@b8-clep-1-125],[@b31-clep-1-125] We conducted analysis stratified by sex and age group: 0--50 years, 51--60 years, 61--70 years, 71--80 years, 81--90 years, \>90 years. We adjusted for cardiovascular disease, hypertension, thyroid disease, other autoimmune disorders, migraine, alcoholism-related disorders, and use of ACE inhibitors, beta-blockers, and statins. Analyses were conducted both with and without adjustment for retinal detachment, cataracts, and uveitis, as these conditions could represent intermediate steps between diabetes and development of glaucoma. All analyses were performed using Stata software (version 9.2; StataCorp, College Station, TX, USA). Results ======= From January 1st, 2001 to December 31st, 2006, 9,295 patients filled at least one prescription for glaucoma medication in the counties' pharmacies; of these, 6,737 filled at least three prescriptions on different dates. Of these, 268 patients had more than 365 days between the first and the third glaucoma medication prescription, 384 patients moved away from the counties, and 94 patients had a history of angle-closure glaucoma or other glaucoma, laser treatment or surgery, and were therefore excluded, leaving 5,991 patients with medically treated incident glaucoma for analysis. [Table 1](#t1-clep-1-125){ref-type="table"} has descriptive data of the 5,991 cases with treated incident glaucoma and the 59,910 age- and sex-matched control subjects. Seven hundred and four patients using glaucoma medication had diabetes (11.8%) compared with 3,975 (6.6%) of the 59,910 controls. Patients using glaucoma medication were substantially more likely than controls to have a history of uveitis, cataract, and retinal detachment. Most other comorbidities and use of medications were slightly more prevalent among glaucoma cases than among matched controls ([Table 1](#t1-clep-1-125){ref-type="table"}). The crude OR for using glaucoma medication in patients with diabetes was 1.95 (95% CI: 1.79--2.12) ([Table 1](#t1-clep-1-125){ref-type="table"}). After adjusting for confounding factors, the OR decreased slightly to 1.81 (95% CI: 1.65--1.98). When the eye disorders of retinal detachment, cataracts, and uveitis were added to the model, the OR further decreased slightly, to 1.75 (95% CI: 1.60--1.92). One or more measurement of HbA~1c~ within the year prior to index date was available for 535 patients treated with glaucoma medication (73.3%) with diabetes and for 2,535 of the controls (63.0%) with diabetes. Diabetes was associated with about twofold increased risk for glaucoma at all levels of HbA~1c~: adjusted ORs among diabetic subjects with an HbA~1c~ level of \<7%, ≥7%--\<8%, ≥8%--\<9%, and ≥9% were 1.97 (95% CI: 1.69--2.30), 2.24 (95% CI: 1.86--2.69), 2.05 (95% CI: 1.62--2.60), and 2.20 (95% CI: 1.73--2.80), respectively ([Table 1](#t1-clep-1-125){ref-type="table"}). Fifty-six percent of the diabetes patients who used glaucoma medication had had diabetes for more than five years. The duration of diabetes did not modify relative risk estimates considerably: OR = 1.88 (95% CI: 1.66--2.14) for diabetes duration under five years, OR = 1.64 (95% CI: 1.41--1.92) for duration between five and 10 years, and OR = 1.87 (95% CI: 1.60--2.18, n = 200) for duration of more than 10 years. The strength of association between diabetes and risk of using glaucoma medication decreased with age, with OR decreasing from 3.35 (95% CI: 3.50--5.08) for patients aged 60 years or younger to 1.12 (95% CI: 0.91--1.31) for patients older than 80 years. The effect of diabetes was greater among men with OR = 2.04 (95% CI: 1.79--2.31) than among women with OR = 1.61 (95% CI: 1.42--1.83) ([Table 2](#t2-clep-1-125){ref-type="table"}). Discussion ========== We found that diabetes mellitus was associated with a 1.8-fold increased risk of medically treated glaucoma. The increased risk was independent of diabetes duration or of level of glycemic control. The relative effect associated with diabetes was greatest among persons aged under 60 years, possibly due to the lower baseline glaucoma risk in younger individuals. The effect was greater in men than in women. The effect of diabetes on the risk of using glaucoma medication estimated in our study was similar to that estimated in the only other study of comparable size and design. In the Scottish Tayside study (2000), a cohort of all T2DM patients (n = 6,631) in the Tayside region was followed for two years between 1993 and 1995.[@b32-clep-1-125] The investigators used prescription data and operation codes as surrogate measures for glaucoma, while estimating, based on a case notes review, the extent of misclassification thus introduced. Based on 65 incident glaucoma cases in the T2DM cohort and 963 incident cases in the general Tayside population, relative risk was estimated at 1.57 (95% CI: 0.99--2.48) for the association between T2DM and glaucoma. Four large cross-sectional studies found positive associations between diabetes and glaucoma including the Beaver Dam study (1994; OR = 1.84; n = 4,926, glaucoma cases = 105);[@b11-clep-1-125] the Rotterdam study (1995; OR = 3.11; n = 4,178, glaucoma cases = 37);[@b7-clep-1-125] the Blue Mountain Eye study (1997; OR = 2.12; n = 3,654, glaucoma cases = 108);[@b12-clep-1-125] and the Los Angeles Latino Eye Study (LALES, 2008; OR = 1.4; n = 5,894, glaucoma cases = 288).[@b13-clep-1-125] Other cross-sectional studies, including the Baltimore Eye Survey (n = 5,308, glaucoma cases = 161)[@b15-clep-1-125] and the Ocular Hypertension Treatment Study (n = 1,636, glaucoma cases = 119)[@b14-clep-1-125] showed no or even a negative association between diabetes and glaucoma. A meta-analysis that included most of the case-control and cross-sectional studies concluded that diabetes patients are at a significantly increased risk of glaucoma (OR = 1.5, 95% CI: 1.16--1.93).[@b33-clep-1-125] Two recent prospective population-based cohort studies found no association between diabetes and incident glaucoma: the Barbados Eye Study cohort (n = 3,222, glaucoma cases = 125, OR diabetes = 1.2) was followed for nine years[@b16-clep-1-125] and the Rotterdam cohort (n = 3,387, glaucoma cases = 87, OR diabetes = 0.82,) was followed for seven years.[@b34-clep-1-125] In two subsequent letters, Ellis and Quigley propose that a potential surveillance bias in diabetic patients in previous studies may fully explain the apparent increased risk of glaucoma in diabetes patients.[@b35-clep-1-125],[@b36-clep-1-125] Our study is the first to demonstrate the equal risk of glaucoma associated with diabetes at all levels of glycemic control and duration of diabetes. Our findings must be interpreted in the context of the study's methodological strengths and weaknesses. The universal health coverage of Denmark's national health care system enabled us to conduct a large, truly population-based study, while using population-based data sources with independently and routinely recorded data. These features tend to reduce the risks of selection and information bias. As in previous studies of this issue, it is difficult to rule out surveillance bias stemming from greater likelihood of diabetic patients to undergo ophthalmological investigations compared with general population.[@b11-clep-1-125],[@b32-clep-1-125],[@b34-clep-1-125]--[@b36-clep-1-125] As shown in the Beaver Dam study, diabetes patients with glaucoma were significantly more likely than glaucoma patients without diabetes to have consulted an ophthalmologist over a two-year period (49.2 vs 39.6%).[@b11-clep-1-125] Some evidence for such bias in our data comes from the observation that glaucoma patients with diabetes tended to be slightly younger than other glaucoma patients (median age of 68.8 years vs. 71.1 years). Moreover, our case group is likely to include patients treated for OHT as in the Tayside Study.[@b32-clep-1-125] Our data did not allow us to differentiate primary from secondary glaucoma, neither could we differentiate between different types of glaucoma (pseudoexfoliation, angle closure, pigment dispersion, etc). Also, glaucoma diagnoses in this population-based study were made by approximately 60 different ophthalmologists who may have had different thresholds for commencement of medical treatment for glaucoma. Quigley and colleagues suggested that it is not safe to conclude that most physicians prescribe glaucoma medication according to preferred practice patterns, and that many patients fill their prescriptions so irregularly that they keep "reappearing" as incident glaucoma cases.[@b37-clep-1-125] We aimed to solve this potential problem with our defined medication algorithm. Ascertainment of medication use and of data on covariates through independent administrative registries reduced potential patient-related (differential recall) or investigator-related information bias, which might affect studies with primary data collection. In order to become a glaucoma case in this study, patients had to buy glaucoma medication at least three different times within one year. These cases are very likely to actually use the medication and to have glaucoma rather than a temporary increase in IOP. Prescriptions can only be filled at a pharmacy, and all Danish pharmacies report complete data to the prescription database. Glaucoma medications purchased in pharmacies are partially refunded by the national health insurance, and the share of Internet-based or foreign purchases is negligibly small. We were able to adjust our analysis for several potential confounders measured by data from the medical databases that aim to track complete prescription and medical history, including all surgical procedures performed. Known strong risk factors for glaucoma are older age, family history, and race. Hispanics[@b13-clep-1-125] and blacks[@b16-clep-1-125],[@b38-clep-1-125] are known to have a higher risk than Caucasians for both diabetes and glaucoma, but are hardly represented in this study performed in a very homogenous population of Caucasians. Most of the recently reviewed risk factors for glaucoma[@b26-clep-1-125] were more prevalent among cases than among controls in our study, but adjustment for their effects explained only 15%--20% of the observed diabetes effect. Of note, cardiovascular disease did not confound our analysis. While cardiovascular diseases and hypertension were associated with diabetes, they were equally prevalent among glaucoma cases and controls, which suggests that only a small part of the risk increase conferred by diabetes was mediated by these disorders. In conclusion, our findings, representing the everyday practice pattern of Danish ophthalmologists, and based on a large population-based sample, corroborate and extend the existing evidence of diabetes mellitus as a risk factor for medical glaucoma treatment. This study received financial support from Væm om Synet and from the Clinical Epidemiological Research Foundation at Aarhus University Hospital. The funding organization had no role in the design or conduct of this study. **Disclosures** The authors report no conflicts of interest in this work. ###### Adjusted and crude ORs for glaucoma according to presence of diabetes mellitus **Risk factor** **Cases** **Controls** **Crude OR (95% CI)** **Adjusted OR[a](#tfn1-clep-1-125){ref-type="table-fn"} (95% CI)** --------------------------------- ------------------- ------------------- ----------------------- -------------------------------------------------------------------- **Sex** Male 2,492 (41.6) 24,920 (41.6) Female 3,499 (58.4) 34,990 (58.4) Age, years, median (IQR) 70.4 (60.8--78.4) 70.3 (60.8--78.4) **Diabetes** Absent 5,287 (88.25) 55,935 (93.37) 1.0 (ref) 1.0 (ref) Present 704 (11.75) 3,975 (6.63) 1.95 (1.79--2.12) 1.81 (1.65--1.98) Thyroid disease 439 (7.3) 3,603 (6.0) 1.24 (1.12--1.38) 1.20 (1.08--1.34) Migraine 203 (3.4) 1,568 (2.6) 1.31 (1.13--1.53) 1.32 (1.13--1.54) Auto immune disorders 244 (4.1) 1,915 (3.2) 1.29 (1.12--1.47) 1.25 (1.09--1.43) Cardiovascular risk factors 765 (12.8) 7,919 (13.2) 0.96 (0.88--1.04) 0.76 (0.70--0.84) Hypertension 593 (9.9) 4,772 (8.0) 1.28 (1.17--1.40) 1.19 (1.07--1.31) Alcohol related diagnoses 47 (0.8) 494 (0.8) 0.95 (0.70--1.29) 0.91 (0.68--1.24) ACE inhibitors 1,042 (17.4) 8,616 (14.4) 1.26 (1.18--1.36) 1.08 (1.00--1.17) Oral beta adrenoceptor blockers 1217 (20.3) 11,695 (19.5) 1.05 (0.98--1.13) 0.96 (0.89--1.03) Statins 828 (13.8) 6,307 (10.5) 1.05 (0.98--1.13) 1.29 (1.18--1.41) Cataract 686 (11.5) 3,995 (6.7) 1.90 (1.74--2.08) 1.42 (1.29--1.56) Retinal detachment 135 (2.3) 159 (0.3) 8.65 (6.86--10.90) 6.15 (4.82--7.85) Uveitis 523 (8.7) 1,349 (2.3) 4.21 (3.79--4.68) 3.60 (3.22--4.02) **Duration of diabetes** Diabetes absent 5,287 (88.25) 55,935 (93.37) 1.0 (ref) 1.0 (ref) \<5 years 320 (5.34) 1,689 (2.82) 1.88 (1.66--2.14) 2.03 (1.79--2.29) ≥5--\<10 years 200 (3.34) 1,210 (2.02) 1.64 (1.41--1.92) 1.77 (1.52--2.06) ≥10 years 211 (3.52) 1,111 (1.85) 1.87 (1.60--2.18) 2.02 (1.74--2.35) **HbA**~**1c**~ Diabetes absent 5,287 (87.3) 55,935 (93.37) 1.0 (ref) 1.0 (ref) HbA~1c~\< 7% 213 (3.6) 1,074 (1.79) 1.97 (1.69--2.30) 2.12 (1.83--2.47) HbA~1c~ ≥ 7%--\<8% 150 (2.5) 660 (1.10) 2.24 (1.86--2.69) 2.44 (2.04--2.92) HbA~1c~ ≥ 8%--\<9% 88 (1.5) 421 (0.70) 2.05 (1.62--2.60) 2.24 (1.77--2.82) HbA~1c~ ≥ 9% 84 (1.4) 380 (0.63) 2.20 (1.73--2.80) 2.36 (1.86--3.00) HbA~1c~ unknown 196 (3.3) 1,475 (2.46) 1.34 (1.15--1.57) 1.41 (1.21--1.65) **Notes:** Data on cases and controls are n (%). OR adjusted by conditional logistic regression for thyroid disease, migraine, autoimmune disorders, cardiovascular events, hypertension, alcohol-related disorders, angiotensin-converting enzyme inhibitors, oral beta adrenoceptor blockers and statins. **Abbreviations:** ACE, angiotensin-converting enzyme; CI, confidence interval; HbA~1c~, glycosylated hemoglobin; IQ, interquartile range; OR, odds ratio. ###### Odds ratio (OR) for glaucoma according to presence of diabetes stratified by age and sex **Crude OR (95% CI)** **Adjusted OR[a](#tfn3-clep-1-125){ref-type="table-fn"} (95% CI)** -------------- ----------------------- -------------------------------------------------------------------- **Age** 0--60 years 4.22 (3.50--5.08) 3.35 (2.72--4.14) 61--70 years 2.55 (2.17--2.99) 2.26 (1.90--2.69) 71--80 years 1.54 (1.32--1.79) 1.47 (1.25--1.73) \>80 years 1.08 (0.89--1.33) 1.12 (0.91--1.37) **Sex**   Female 1.74 (1.54--1.96) 1.61 (1.42--1.83)   Male 2.20 (1.95--2.49) 2.04 (1.79--2.31) **Notes:** Odds ratio adjusted by conditional logistic regression for thyroid disease, migraine, autoimmune disorders, cardiovascular events, hypertension, alcohol related disorders, ace-inhibitors, oral beta adrenoceptor blockers and statins. **Abbreviation:** CI, confidence interval.

Introduction {#s1} ============ Total creatine (TCr) is the sum of creatine (Cr) and creatine phosphate (CrP). TCr stores in adult rodent and human skeletal muscle are derived predominantly from extracellular Cr uptake via the activity of Na^+^/Cl^−^-dependent creatine transporter (CrT) proteins (Slc6a8) located on the sarcolemmal membrane (Snow and Murphy, [@B18]). The extracellular Cr is derived from the diet and endogenously synthesized mostly by the kidney and liver (Wyss and Kaddurah-Daouk, [@B24]). However, a small proportion of adult skeletal muscle TCr stores could be produced from its own Cr biosynthesis given that adult skeletal muscle is known to have low, but detectable *in vitro* activity of both of the enzymes involved in this process (Van Pilsum et al., [@B20]; Daly, [@B3]; Cullen et al., [@B2]). Cr biosynthesis is a two-step reaction process involving the enzymes L-arginine:glycine amidotransferase (AGAT) and guanidinoacetate methyltransferase (GAMT). AGAT produces ornithine and guanidinoacetate (GAA) from arginine and glycine. GAMT catalyses S-adenosyl-L-methionine-dependent methylation of GAA to form Cr and S-adenosyl-L-homocysteine. Interestingly, McClure et al. ([@B13]) have demonstrated that muscle AGAT and GAMT gene expression and GAMT protein expression are elevated in skeletal muscle from *mdx* mice, a model of human Duchenne muscular dystrophy. This suggests that *mdx* mice may have an enhanced capacity to synthesize Cr in skeletal muscle and that this adaptation may help to maintain muscle Cr levels and limit energy failure in *mdx* skeletal muscle (McClure et al., [@B13]). Up regulation of AGAT gene and protein activity has also been observed in failing human heart muscle (Cullen et al., [@B2]). Cullen et al. ([@B2]) hypothesized that elevated AGAT mRNA and enzyme activity in heart failure was an attempt to increase local Cr synthesis to alleviate reductions in cardiomyocyte Cr content typically observed in this disease. In apparent contradiction to the concept that skeletal muscle could up regulate its Cr biosynthesis in circumstances where Cr content was compromised, no Cr was detectable in muscle from a ubiquitous CrT knockout (CrT KO) mouse (Skelton et al., [@B17]). However, it is possible that the failure to detect Cr in the muscle of these mice was due to the use of a relatively insensitive colorimetric analytical technique. The aim of the present study was therefore to re-analyse the CrT KO mouse muscle for Cr and CrP content using more specific and sensitive techniques. Furthermore, if Cr was found to be present in CrT KO muscle we aimed to investigate if at least some of this Cr could be produced by an up regulation of muscle Cr biosynthesis. We hypothesized that CrT KO muscle would have low, but detectable levels of Cr and CrP, and that this muscle would display elevated levels of AGAT and GAMT expression resulting in an increased *in vitro* capacity to produce Cr. Methods {#s2} ======= Animals ------- Gastrocnemius muscle samples were obtained from ubiquitous CrT knockout mice (CrT^−/y^) and control mice (CrT^+/y^) bred and raised as described by Skelton et al. ([@B17]). The generation of these mice and tissue extraction was approved by the Cincinnati Children\'s Research Foundation Institutional Animal Care and Use Committee, protocol \#1C06058. The muscle samples were subsequently transported to Deakin University (Australia) on dry ice and stored at −80°C until analysis. Metabolite analysis ------------------- A portion of each gastrocnemius muscle sample was freeze dried for 24 h, powdered and then weighed. The powdered samples were extracted on ice with 0.5 M perchloric acid and 1 mM ethylenediaminetetraacetic acid and neutralized with 2.1 M potassium hydrogen carbonate. Extracts were analyzed in triplicate for ATP, CrP, and Cr using enzymatic analysis with fluorometric detection as described previously (Harris et al., [@B6]; Febbraio et al., [@B4]). TCr was determined by summing the Cr and CrP content. Percent co-efficient of variation for repeated measurement of aqueous standards (200 to 500 μM) for these assays in our laboratory are 2.7, 3.6, and 7.4% for ATP, Cr, and CrP, respectively. The variation of muscle metabolites due to analytical procedure and error employing the same extraction process and assay system is reported to be less than 4% (Harris et al., [@B6]). Gene analysis ------------- RNA was isolated from another portion of the gastrocnemius muscle using TRI-Reagent® Soln. (Ambion Inc., Austin, TX) according to the manufacturer\'s protocol. RNA concentration was determined by the absorbance at a wavelength of 260 nm using the NanoDrop® ND-1000 spectrophotometer (Thermo Scientific, Rockford, IL). Sample RNA (1.0 μg) was treated with DNAse amplification I (Invitrogen, Carlsbad, CA) before being reverse transcribed to synthesize first strand cDNA using the high-capacity RNA-cDNA reverse transcription kit (Applied Biosystems, Forster, CL). All cDNA was diluted to a working concentration of approximately 5 ng/μl. Efficiency of the RT was determined by Quant-iT™ OliGreen® ssDNA Reagent Kit (Invitrogen). Real-time quantitative polymerase chain reaction (QPCR) was used to measure mRNA expression as described previously (Wallace et al., [@B22]). The primers for the CrT gene were designed using the web based software Primer3plus (Untergasser et al., [@B19]) from the mouse gene sequence (Ensembl Genome Browser). AGAT and GAMT primers were designed against a homologous region across human, mouse, and rat as previously described Ireland et al. ([@B9]). The specific primer sequences are described in Table [1](#T1){ref-type="table"}. ###### **Sequence of forward and reverse primers for genes of interest**. **Gene** **Sense** **Anti-sense** ---------- ------------------------------- ----------------------------------- CrT 5′ GCC GGC AGC ATG AAT GTC 5′ GTG ATT GTC TTC TAC TGC AAC AC AGAT 5′ TCA CGC TTC TTT GAG TAC CG 5′ TCA GTC GTC ACG AAC TTT CC GAMT 5′ TGG CAC ACT CAC CAG TTC A 5′ AAG GCA TAG TAG CGG CAG TC *CrT, creatine transporter; AGAT, L-arginine:glycine amidinotransferase; GAMT, guanidinoacetate methyltransferase*. QPCR was performed using a Stratagene Mx3000p QPCR run by MxPro QPCR Software (Stratagene, La Jolla, CA) with SYBR Green PCR Mastermix (Applied Biosystems). Each 20 μl reaction contained 5 μl template and 0.2 μM of each forward and reverse primer. A 3-step QPCR was used to amplify mRNA; initial template denaturing of 95°C for 10 min, followed by 40 cycles of 95°C for 30 s, 60°C for 60 s, and 72°C for 30 s. Fluorescence readings were measured during the last step of cycling and normalized to cDNA content. Protein analysis ---------------- Western blotting technique was used to measure AGAT and GAMT protein expression in the gastrocnemius muscles of the CrT^−/y^ and CrT^+/y^ mice. GAMT was detected with an affinity purified mouse monoclonal antibody (Monash Antibody Technologies Facility, Melbourne, Australia) made through injection of the antigenic peptide N-terminal aa 125--145. AGAT (also known as GATM) antibody was purchased from Biorbyt Ltd. The antibodies detected a positive band at expected molecular masses (GAMT, 26 kDa; AGAT, 46 kDa) as previously determined (Braissant et al., [@B1]; McClure et al., [@B13]). Muscle protein was extracted using radioimmunoprecipitation assay (RIPA) buffer (Millipore, North Ryde, NSW) containing protease inhibitor cocktail I (Sigma-Aldrich, Sydney, Australia) and Halt® phosphatase inhibitor cocktail (Thermo Scientific, Rockford, IL). Protein concentrations were determined using the bicinchoninic acid (BCA) assay (Pierce Biotechnology, Rockford, USA). Electrophoresis was performed using a 4--12% NuPAGE® Novex Bis-Tris Gel in NuPAGE® SDS MOPS Running Buffer (Invitrogen). Proteins were transferred to a PVDF membrane using a Bjerrum buffer containing 50 mM Tris, 17 mM glycine, and 10% methanol. For GAMT, membranes were blocked with 5% milk powder in PBS for one hour followed by overnight incubation at 4°C with the GAMT primary antibody diluted 1:500 in 5% milk powder in PBS. For AGAT, membranes were blocked with 5% BSA/PBS for 1 h followed by overnight incubation at 4°C with the AGAT primary antibody diluted 1:70 in 5% BSA/PBS. Following washing, all GAMT membranes were incubated for 1 h with rabbit anti-mouse IgG AlexaFluor® 680 (Invitrogen) diluted 1:5000 in PBS containing 50% Odyssey® blocking buffer (LI-COR Biosciences, Lincoln, USA) and 0.01% SDS. AGAT membranes were similarly incubated but with an anti-rabbit secondary antibody. After washing, the proteins were exposed on an Odyssey® Infrared Imaging System (LI-COR Biosciences) and individual protein band optical densities were determined using ImageJ Software (National Institutes of Health, Bethesda, USA). The blots were normalized against the tubulin protein (Sigma-Aldrich). *in vitro* creatine biosynthesis -------------------------------- Gastrocnemius muscle from CrT^+/y^ and CrT^−/y^ mice (\~20 mg each) were homogenized in potassium phosphate buffer (0.066 M, pH 7.4)(1 mg tissue: 4.5 μl buffer) on ice using an electric homogenizer (Lab-Serv-D-130). The homogenate was subsequently freeze-thawed three times to lyse cell membranes. Rates of Cr synthesis were measured using an adaptation of the methods described (Grazia Alessandri et al., [@B5]; Ide et al., [@B8]). One aliquot of the muscle homogenate (33 μl) was incubated in a standard reaction media containing Tris--HCl (5 μl, 100 mM, pH 7.5), arginine (10 μl, 1mM), glycine (5 μl, 1 mM), S-adenosylmethionine (10 μl, 0.25 mM), dithiotheritol (DTT) (5 μl, 2 mM), and 32 μl of potassium phosphate for 2 h at 37°C. After the incubation period the samples were spun twice at 10,000 g, 4°C for 10 min (Eppendorf centrifuge 5402) and the subsequent supernatant filtered (syringe filter 0.2 μm---Phenomenex, USA) and stored on ice prior to Cr analysis. To obtain a reaction blank a second aliquot (33 μl) from each muscle homogenate was added to the reaction media and immediately centrifuged, filtered and stored on ice prior to Cr analysis. A third aliquot (10 μl) from each muscle homogenate sample was analyzed for protein concentration using the BCA Protein Assay Kit (Pierce Biotechnology, Rockford, IL) according to the manufacturer\'s protocol, and absorbance was measured on a Synergy 2 Microplate Reader (BioTek, Winooski, VT). Creatine concentration of known Cr standards, the reaction blanks and the two hour samples were measured using high performance liquid chromatography (Agilent 1100 Series system), based on an adaption of the method by Moore et al. ([@B14]). The separation column (i.d. 250 mm × 4.6 mm) was a Luna 10 μm -- C18 (2) 100 A (Phenomenex, USA) and the mobile phase was an aqueous solution (pH = 1.5) of 10 mM Na~2~SO~4~, 5 mM H~2~SO~4~, and 10 mM sodium-1-hexane sulfonate (Sigma Aldrich-USA). The mobile phase flow rate was 1 mL.min^−1^ for the first 8 min and then increased to 1.4 mL.min^−1^ for a further 22 min. Elution of Cr was detected by a spectrophotometer (Agilent 1100 G1365B) at a wavelength of 230 nm. Statistical analysis -------------------- All data are presented as mean ± s.e.m., and were analyzed using a two-tailed Student\'s two sample *t*-test. Significance was set at *P* ≤ 0.05. Results {#s3} ======= Muscle metabolites ------------------ Muscle ATP (*P* = 0.002), Cr (*P* = 0.002), CrP (*P* = 0.02), and TCr (*P* = 0.0000014) content were all markedly lower in the CrT^−/y^ mice compared with controls (Table [2](#T2){ref-type="table"}). Importantly, Cr and CrP were detected in CrT^−/y^ mice muscle. ###### **Gastrocnemius metabolite content in CrT^+/y^ and CrT^−/y^ mice**. **CrT^+/y^** **CrT^−/y^** ---------- -------------- ----------------------------------------------- ATP 33.7 ± 0.7 15.9 ± 2.4[^\*\*^](#TN2){ref-type="table-fn"} PCr 36.1 ± 6.9 2.3 ± 0.3[^\*^](#TN1){ref-type="table-fn"} Cr 88.9 ± 8.6 20.1 ± 4.2[^\*\*^](#TN2){ref-type="table-fn"} Total Cr 125.0 ± 3.3 22.3 ± 4.3[^\*\*^](#TN2){ref-type="table-fn"} *Values are means ± s.e.m., n = 4 per group. Metabolite content is expressed in mmol.kg^−1^ dry mass. Cr, creatine; PCr, phosphocreatine; TCr, total creatine*, *P \< 0.05*, *P \< 0.01*. Genes ----- As expected there was no detectable CrT gene expression in the CrT^−/y^ mice muscle (Figure [1A](#F1){ref-type="fig"}). Interestingly, there was a different expression response for the two Cr biosynthesis genes AGAT and GAMT. AGAT gene expression was elevated (*P* = 0.00007), but GAMT gene expression was unchanged (*P* = 0.20) in CrT^−/y^ muscle compared with CrT^+/y^ muscle (Figures [1B,C](#F1){ref-type="fig"}). {#F1} Protein expression and *in vitro* skeletal muscle creatine biosynthesis ----------------------------------------------------------------------- The protein expression of AGAT was increased three fold (*P* = 0.01), however GAMT protein was not different (*P* = 0.28) in the CrT^−/y^ mice compared with controls (Figure [2](#F2){ref-type="fig"}). The *in vitro* Cr biosynthesis rate was 1.5 fold greater (*P* = 0.05) in the CrT^−/y^ mice compared with CrT^+/y^ (Figure [3](#F3){ref-type="fig"}). {#F2} {#F3} Discussion {#s4} ========== The major findings of the present study were that Cr and CrP were detected in CrT^−/y^ skeletal muscle (e.g., TCr content was 18% of normal muscle) and that *in vitro* Cr biosynthesis rates, as well as AGAT gene and protein expression were elevated in the CrT^−/y^ muscle compared with CrT^+/y^. In contrast to our hypothesis, muscle GAMT gene and protein expression in the CrT^−/y^ were not different to CrT^+/y^ mice. Our findings suggest that at least some of the Cr detected in CrT^−/y^ muscle may be synthesized within the muscle. The total Cr content determined for CrT^+/y^ gastrocnemius mouse muscle in the present study is similar to that reported previously (Kushmerick et al., [@B11]), however due to slow sampling and freezing of the muscle the ratio of CrP to Cr we report in Table [2](#T2){ref-type="table"} is not likely to accurately represent that typically found in resting skeletal muscle (Kushmerick et al., [@B11]). Importantly, we have clearly demonstrated that Cr and PCr can be detected in CrT^−/y^ gastrocnemius muscle in contrast to that reported previously (Skelton et al., [@B17]). This discrepancy is best explained by the different methods employed to measure these metabolites. In the present study Cr and CrP was analyzed using a widely accepted specific and sensitive enzymatic fluorometric technique on freeze dried muscle (Harris et al., [@B6]; Febbraio et al., [@B4]), whereas Skelton et al. ([@B17]) used a less sensitive colorimetric method to analyse wet muscle. The analytical method employed in the present study is likely to be more sensitive. Firstly, because we analyzed freeze dried muscle that is known to concentrate the metabolite of interest by approximately five fold and secondly, because we used a fluorometric detection system as opposed to a spectrophotometric system. Depending upon the analyte, fluorometry is known to have 1000 to 500,000 fold better limits of detection as compared to spectrophotometry. Of note, the mean gastrocnemius muscle Cr content of wild type mice (CrT^+/y^) reported by Skelton et al. ([@B17]) of 21.82 mmol.kg^−1^ wet muscle (or 94.8 mmol.kg^−1^ dry muscle assuming muscle is 77% water (Hultman and Sahlin, [@B7]) was very similar to that measured in the present study (88.9 mmol.kg^−1^ dry muscle). The discrepancy between the studies therefore revolves around the muscle Cr content in the gastrocnemius muscle of the CrT^−/y^ mice. The fact that both studies produce similar results when muscle Cr content is high, but not when Cr content is expected to be low lends to support to the contention that Skelton et al\'s ([@B17]) analysis method using wet muscle tissue was relatively insensitive and therefore unable to detect Cr in CrT^−/y^ mice muscle. The muscle TCr content measured in this study is expected to be almost exclusively found within the myocytes as extracellular Cr content is calculated to contribute ≤1% of the total Cr measured. This assumes resting muscle extracellular fluid volume is 10% of total muscle water (Hultman and Sahlin, [@B7]) and serum Cr concentration is 300 μM (Skelton et al., [@B17]). Of note, the ATP and TCr levels reported for the CrT^−/y^ mouse muscle in the present study are similar to those reported for other models of Cr deficiency such as β-guanidinoproprionic feeding of rats or GAMT-deficient knockout mice (Kan et al., [@B10]; Williams et al., [@B23]). Two interesting questions arise from the presence of Cr in CrT^−/y^ mice muscle. First, via what mechanism does CrT^−/y^ muscle obtain the small but measurable amount of Cr? Secondly, why is this alternative mechanism unable to fully compensate for the deficit in Cr content? Clearly, knockout of the CrT gene should block most, if not all, transport of extracellular Cr into the myofibers. Unfortunately, the present experiment cannot rule out that some Cr enters the muscle via other non-specific transport mechanisms. In support of this possibility, Loike et al. ([@B12]) reported that approximately 10% of total Cr transport into cultured rat L6 muscle cells could occur via a sodium independent mechanism and therefore, at least in this muscle cell culture system, all sarcolemmal Cr transport could not be attributed to CrT activity alone. Further research using radiolabeled Cr is required to determine the actual existence and extent of non-CrT dependent Cr uptake into skeletal muscle of the CrT^−/y^ mice. In addition to, or alternatively, skeletal muscle may up regulate its own Cr biosynthesis when myocyte Cr content is compromised (Cullen et al., [@B2]; McClure et al., [@B13]). The elevated rates of *in vitro* Cr biosynthesis coupled with increased AGAT gene and protein expression in the CrT^−/y^ mouse muscle strongly indicates that skeletal muscle Cr synthesis is switched on to a greater extent and may be responsible for at least some of the muscle Cr content in this model. This contention is supported by the fact that AGAT is considered to be the rate limiting enzyme of Cr biosynthesis (Walker, [@B21]). Therefore any increase in GAA production by AGAT should lead to Cr production in skeletal muscle, provided GAMT is present in this tissue, which we have clearly demonstrated to be the case (see Figure [2](#F2){ref-type="fig"}). The finding that AGAT gene and protein expression can be up regulated in tissues not normally expected to rely on its own Cr production is consistent with that reported for the failing human heart muscle and *mdx* mouse skeletal muscle (Cullen et al., [@B2]; McClure et al., [@B13]). Interestingly, GAMT gene and protein were expressed to a similar extent in both wild type and CrT^−/y^ muscle contrary to our hypothesis and in apparent contrast to that reported by McClure et al. ([@B13]) when studying *mdx* mouse muscle. Based on previous research we expected that GAMT gene and protein expression would be up regulated in CrT^−/y^ mice because the observed disturbance in high energy phosphagen content (see Table [2](#T2){ref-type="table"}) should activate AMP kinase (Williams et al., [@B23]). AMP kinase is known to activate the p53 transcription factor, which is able to increase GAMT gene transcription (Ide et al., [@B8]). It should be noted that Cr levels do not directly regulate GAMT gene expression, at least in liver cells (Wyss and Kaddurah-Daouk, [@B24]). It is unclear why GAMT gene and protein expression in the CrT^−/y^ muscle was not up regulated as hypothesized. Nevertheless this did not prevent an increase in Cr biosynthesis capacity when muscle homogenates were incubated with arginine, glycine, and S-adenosylmethionine, suggesting that AGAT, but not GAMT activity, was rate limiting in the *in vitro* assay system employed in the present study. The Cr synthesis rate we observed in CrT^+/y^ mice gastrocnemius muscle (1.06 ± 0.07 nmol Cr/mg protein/h) was slightly higher than reported for rat quadriceps muscle (Daly, [@B3]) GAMT activity (0.39 and 0.82 nmol Cr/mg protein/h), but markedly lower than that reported for rat liver (Wyss and Kaddurah-Daouk, [@B24]) GAMT activity (12-20 nmol Cr/mg protein/h). Even though CrT^−/y^ muscle displayed increases in AGAT protein expression and *in vitro* Cr synthesis capacity, there still remained a marked deficit (\~80%) in muscle Cr content in these mice when compared with CrT^+/y^ mice. This observation indicates that the elevation in endogenous muscle Cr synthesis capacity and/or non-CrT dependent muscle Cr uptake capacity is insufficient to overcome the loss of Cr transport across the sarcolemma associated with the absence of CrT protein expression. In the case of insufficient Cr synthesis capacity it remains unclear why this occurs, but may be related to insufficient intramuscular substrate supply to activate AGAT and/or GAMT to high enough rates, endogenous inhibition of one or both these enzymes and/or an elevated Cr leakage rate from the muscle counteracting against any increase in endogenous Cr production rates. Clearly the functional significance of CrT gene knockout on skeletal muscle was not investigated in the present study. We observed that gastrocnemius mass was considerably smaller in CrT^−/y^ mice compared with CrT^+/y^ mice suggesting that muscle morphology was affected. Although speculative, TCr depletion in the CrT^−/y^ mice muscle would probably trigger similar adaptations to that observed in other models of chronic muscle Cr depletion such as β-guanidinoproprionic acid feeding and AGAT or GAMT gene knockout models (Kan et al., [@B10]; Nabuurs et al., [@B15]; Oudman et al., [@B16]). For example chronic muscle Cr depletion has typically resulted in ATP depletion, muscle fiber atrophy, increased mitochondrial content, altered contractile performance and a shift toward a greater proportion of slow twitch fibers. In conclusion, the present experiment clearly demonstrated that Cr was present in CrT^−/y^ mice gastrocnemius muscle. At least some of this Cr may result from an increased muscle capacity to synthesize Cr, however non-specific Cr transport across the sarcolemmal cannot be excluded as a possible contributor to the Cr content observed in CrT^−/y^ mice muscle. Author contributions ==================== All authors were either involved conception and design of the work (Aaron P. Russell, Rodney J. Snow, Matthew R. Skelton) or the acquisition, analysis or interpretation of the data (Séverine Lamon, Aaron P. Russell, Lobna Ghobrial, Craig R. Wright, Erin L. Brown, Michihiro Kon). All authors were involved in critically evaluating drafts of the manuscript and provided final approval of the submitted version to be published. All authors have agreed to be accountable for all aspects of the work and will ensure that questions related to accuracy and integrity will be appropriately investigated. 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. [^1]: Edited by: Wataru Aoi, Kyoto Prefectural University, Japan [^2]: Reviewed by: Robert W. Wiseman, Michigan State University, USA; Lucas Guimarães-Ferreira, Federal University of Espirito Santo, Brazil [^3]: This article was submitted to Striated Muscle Physiology, a section of the journal Frontiers in Physiology.

Salem L, Malouvier A, Blatchford J, Rivero‐Ferrer E, Deltour N, Jacquot E. Ivabradine drug utilization study in five European countries: A multinational, retrospective, observational study to assess effectiveness of risk‐minimization measures. Pharmacoepidemiol Drug Saf. 2019;28:1470--1479. 10.1002/pds.4880 31486198 ###### Key Points Following the SIGNIFY study, the benefit‐risk ratio of ivabradine was reassessed in 2014, and risk minimization measures (RMM) were recommended.Compliance to RMM was evaluated in a drug utilization study (DUS) across five European countries.The study results show that the heart rate at treatment initiation, ivabradine dosing at initiation and during follow‐up, and concomitant use of verapamil or diltiazem were in line with the updated summary of product characteristics (SmPC).In conclusion, the RMMs were well implemented across the five participating countries confirming a favorable benefit‐risk balance of ivabradine in chronic stable angina pectoris. 1. INTRODUCTION {#pds4880-sec-0005} =============== Ivabradine hydrochloride (Procoralan/Corlentor) is a selective inhibitor of the cardiac pacemaker if current, with corresponding reductions in cardiac workload and myocardial oxygen consumption.[1](#pds4880-bib-0001){ref-type="ref"} Ivabradine is indicated in Europe from 2005 for the symptomatic treatment of chronic stable angina pectoris in patients with normal sinus rhythm and a contraindication to, or intolerance of beta blockers. Extensions of the indication were subsequently approved in combination with beta‐blockers in patients inadequately controlled despite an optimal beta‐blocker dose and heart rate (HR) \> 60 bpm (October 2009) and in chronic heart failure (HF) in patients with sinus rhythm and HR ≥ 75 bpm (February 2012). The recommended ivabradine starting dose for these indications was 5‐mg bid, with consideration of 2.5‐mg bid for patients aged 75 years and older. The recommended maintenance dose was 7.5‐mg bid. The SIGNIFY study[2](#pds4880-bib-0002){ref-type="ref"} was a randomized clinical trial evaluating ivabradine at a starting dose of 7.5 mg bid (5‐mg bid if age ≥ 75 years) and a maintenance dose of 10‐mg bid in 19 102 patients with coronary artery disease without HF. Results showed an increase of cardiovascular events, possibly because of bradycardia in a subgroup of patients with angina of Canadian Cardiovascular Society Class II or higher. These findings triggered a benefit‐risk reevaluation by the European Commission[3](#pds4880-bib-0003){ref-type="ref"} in May 2014, and a direct health care professional communication (DHPC) was disseminated in Europe in June 2014 to inform prescribers and remind them of the current conditions of use of the product in patients with angina pectoris. Ivabradine benefit‐risk ratio was reassessed by the Pharmacovigilance Risk Assessment Committee (PRAC) in November 2014 and was found to remain positive for its authorized indications.[4](#pds4880-bib-0004){ref-type="ref"} The PRAC recommended to increase the resting HR threshold of patients with angina pectoris from greater than 60 to greater than or equal to 70 bpm before treatment initiation, contraindicate concomitant use of ivabradine with verapamil or diltiazem, and reinforce current posology including initial (5‐mg bid) and maintenance (7.5‐mg bid) maximal doses. Previous information found in the summary of product characteristics (SmPCs) regarding HR monitoring and warning of use in patients with atrial fibrillation was also reinforced. As routine and additional risk minimization measures (RMMs), the SmPC was updated accordingly, and a second DHPC was distributed to inform prescribers in Europe from December 2014. Following the PRAC recommendations, a drug utilization study (DUS) was conducted to evaluate how ivabradine is used in patients with chronic stable angina pectoris in routine clinical practice and the consistency of ivabradine prescribing with the aforementioned PRAC recommendations. 2. METHODS {#pds4880-sec-0006} ========== 2.1. Study design overview {#pds4880-sec-0007} -------------------------- This retrospective cohort study collected data from medical records (chart review) of patients initiating ivabradine for chronic stable angina pectoris in routine clinical practice in five European countries. Data, from start of treatment until 6 months, describing ivabradine new users\' characteristics and ivabradine patterns of use were collected retrospectively from patients\' charts by the physicians. The study design included two study periods: pre‐ and post‐RMM (Figure [1](#pds4880-fig-0001){ref-type="fig"}): Pre‐RMM: before implementation of the new RMM, from January 2010 to December 2013.Post‐RMM: after implementation of the new RMM, from end of June 2015 to end of June 2016. {#pds4880-fig-0001} For each study period, ivabradine initiation was defined as the first date in which a patient was treated with ivabradine in the considered study period, provided that the patient had not received ivabradine during the previous 6 months. The targeted countries were France, Germany, Italy, Spain, and the United Kingdom (UK). Countries\' selection was based on ivabradine volume sales, geographic representation of the European Union (EU), and ability to represent a variety of medical practices, in term of specialty and practice settings. A pilot study was conducted to identify potential challenges in the study design such as the shared‐care management (SCM) (ie, when patient care was shared between different physicians and not only the participating physician), to evaluate practical aspects of implementation, and test the data collection form. The protocol was registered and made publicly available on the European Medicine Agency electronic Register of Post‐Authorization Studies (EU PASS 19522). 2.2. Physician and patient enrollment {#pds4880-sec-0008} ------------------------------------- A total of 600 patients per study period were targeted taking into account the clustering effect and other contingencies such as missing data, allowing for an absolute precision of at least 5%. Patients were identified across a variety of physicians\' specialties, including general practitioners (GPs) and specialists (cardiologists or internists) practicing in outpatient settings (private practices or hospital outpatient clinics). The physicians were targeted to allow the recruitment of patients whose treatment modalities reflected, to the extent possible, prescribing patterns in each country. The targeted patients\' distribution between specialties within each country was defined according to the distribution of national sales and information on SCM obtained during the pilot study. There were 70 targeted active sites. Sampling of participating physicians was performed on IMS Health lists (Sponsor\'s list for France) of physicians. The source population included all patients who had initiated ivabradine treatment for chronic stable angina in regular clinical practice in one of the study periods. The initiation could have been done in the same participating site or elsewhere, provided that key data at initiation were present in the patients\' medical records. To be included, patients were required to have documented initiation of ivabradine treatment during one of the study periods, chronic stable angina as the indication for ivabradine initiation, and provision of informed consent for study participation, where applicable. Exclusion criteria included ivabradine prescribed for an indication other than chronic stable angina, documented ivabradine use in the previous 6 months, and participation in an ivabradine clinical trial simultaneously. Sites began identifying patients when all appropriate approvals from competent authorities and ethics committees were received. To avoid a cluster effect and to ensure a sufficient number of participating sites allowing to assess different practice modalities, the physicians were informed that they could not include greater than 20 patients by period for a specialist and greater than 10 patients by period for a GP. To avoid selection bias, if a site had a larger number of patients than the maximum threshold, physicians had to organize the eligible patients in alphabetical order by surname and to start patient data abstraction in ascending or descending order whether the last number of the site ID was odd (eg, XX1) or even (eg, XX2). 2.3. Analysis {#pds4880-sec-0009} ------------- Primary outcome was the compliance with the composite of the four criteria in the SmPC assessed before and after RMMs implementation: (1) use in patients with a HR threshold greater than or equal to 70 bpm at initiation; (2) no doses higher than 5‐mg bid at treatment initiation; (3) no doses higher than 7.5‐mg bid during 6‐month follow‐up; and (4) no concomitant use of verapamil or diltiazem at treatment initiation or during 6‐month follow‐up. The primary analysis evaluated the proportion of patients prescribed ivabradine according to the four criteria of the SmPC. The secondary analyses compared the demographics and specific baseline comorbidities of ivabradine new users in routine clinical practice, pre‐, and post‐RMM. Total treatment duration was defined as the time between date of treatment initiation and date of treatment discontinuation or date of censorship for patients who did not discontinue treatment during the 6‐month follow‐up. Primary and secondary analyses were stratified by study period. Primary analyses were also performed by country and by specialty. The primary analysis was conducted in the patients\' set with complete data. However, a patient who was not compliant with one of the four SmPC criteria was deemed noncompliant, regardless of the completeness of data for the other criteria. Confidence intervals (CI) and *P* values for the difference in proportions of patients satisfying each criterion were estimated using the Wilson score with continuity correction (Newcombe score[5](#pds4880-bib-0005){ref-type="ref"}). For the primary outcome, sensitivity analyses were performed for missing data, based on patients\' set with missing data in the denominator for proportions, and by physicians\' initiator status, either the participating physician is an initiator or a subsequent prescriber, to explore the impact of SCM on the primary outcome and the missing data proportion. Analyses were conducted using SAS statistical software (SAS Institute, Cary, North Carolina), version 9.4. 3. RESULTS {#pds4880-sec-0010} ========== 3.1. Disposition {#pds4880-sec-0011} ---------------- Of the 138 741 physicians in the source list with relevant contact information, 60 675 physicians were contacted. The percentage of interested physicians overall was 0.86%. Of these, 13% participated in the study. A total of 68 physicians were active (included at least one eligible patient). Active physicians\' distribution by country and specialty is presented in Table [1](#pds4880-tbl-0001){ref-type="table"}. ###### Total number of physicians recruited by country and specialty Variable France Germany Italy Spain UK Total ----------------------------------------------------------------------------------------- ------------ ----------- ------------ ------------ ----------- ------------ No. of physicians contacted 12531 9304 21238 13300 4302 60675 GP 6996 6789 16039 10041 3167 43032 Specialists 5535 2515 5199 3259 1135 17643 No. of interested physicians (n \[%\])[a](#pds4880-note-0002){ref-type="fn"} 119 (0.95) 83 (0.89) 176 (0.83) 107 (0.80) 35 (0.81) 522 (0.86) GP 52 (0.74) 45 (0.66) 67 (0.42) 63 (0.63) 25 (0.79) 252 (0.59) Specialists 63 (1.14) 38 (1.51) 97 (1.87) 37 (1.14) 5 (0.44) 240 (1.36) Missing 4 (0.03) 0 12 (0.06) 7 (0.05) 5 (0.12) 30 (0.05) No. of qualified physicians with signed agreement[b](#pds4880-note-0003){ref-type="fn"} 22 26 11 15 11 85 GP 9 15 1 6 9 40 Specialists 13 11 10 9 2 45 No. of active physicians[c](#pds4880-note-0004){ref-type="fn"} 17 18 11 12 10 68 GP 7 11 1 4 8 31 Specialists 10 7 10 8 2 37 Abbreviations: GP, general practitioner; RMM, risk minimization measures; UK, United Kingdom. Percentages presented between parentheses are based on the number of contacted physicians within each country. Total includes two physicians with missing country. Physicians were qualified for activation if they had the potential to contribute the minimum number of patients treated with ivabradine in at least one study period. Active physicians were participating physicians who had included at least one eligible patient in the study. Data of 1326 patients were entered in the Case Report Form (CRF) (Figure [2](#pds4880-fig-0002){ref-type="fig"}). From the 1217 (91.8%) eligible patients, 711 were included in pre‐RMM period and 506 in post‐RMM period. Patient characteristics are summarized in Table [2](#pds4880-tbl-0002){ref-type="table"}. Patients in both study periods were comparable except for history of HF and of hypertension, which were more frequent in patients with data collected during the post‐RMM period, and history of sinus bradycardia was less frequent in the post‐RMM period. Mean (standard deviation) total treatment duration during the 6‐month follow‐up period was similar for both periods (18.0 \[9.41\] weeks in the pre‐RMM period and 17.0 \[9.70\] in the post‐RMM period). {#pds4880-fig-0002} ###### Patient characteristics Variable Pre‐RMM (N = 711) Post‐RMM (N = 506) *P* value ------------------------------------------------------------ ------------------- -------------------- ----------- Sex (n \[%\]) Male 442 (62.2) 318 (62.8) .8213 Female 269 (37.8) 188 (37.2) Age at ivabradine initiation (y), (n \[%\]) 18‐44 21 (3.0) 11 (2.2) .5856 45‐64 226 (31.8) 178 (35.2) 65‐74 217 (30.5) 162 (32.0) 75‐84 195 (27.4) 119 (23.5) ≥85 52 (7.3) 36 (7.1) Under 75 464 (65.3) 351 (69.4) .2551 75 and over 247 (34.7) 155 (30.6) Smoking status (n \[%\]) Current smoker 91 (12.8) 70 (13.8) .7993 Past smoker 267 (37.6) 192 (37.9) Nonsmoker 283 (39.8) 191 (37.7) Unknown 70 (9.8) 53 (10.5) Medical history (n \[%\]) Hypertension 541 (76.1) 410 (81.0) .0313 Hyperlipidaemia 488 (68.6) 329 (65.0) .2142 Coronary angioplasty 310 (43.6) 229 (45.3) .8224 Diabetes mellitus (Type 1 or 2) 280 (39.4) 199 (39.3) .8007 Overweight or obese 242 (34.0) 196 (38.7) .0983 Peripheral vascular disease 142 (20.0) 113 (22.3) .4943 Coronary artery bypass 135 (19.0) 84 (16.6) .2573 Heart failure 122 (17.2) 147 (29.1) \<.0001 Sinus bradycardia 33 (4.6) 10 (2.0) .0146 Other conduction disorders 32 (4.5) 27 (5.3) .4631 Atrial fibrillation or atrial flutter---Paroxysmal 29 (4.1) 29 (5.7) .1623 Atrioventricular block 26 (3.7) 10 (2.0) .1804 Other supraventricular arrhythmias 22 (3.1) 10 (2.0) .2428 Ventricular tachycardia---Not sustained 21 (3.0) 16 (3.2) .7949 Atrial fibrillation or atrial flutter---Persistent 13 (1.8) 6 (1.2) .3008 Ventricular tachycardia--- Sustained 12 (1.7) 6 (1.2) .3881 Other ventricular arrhythmias 11 (1.5) 8 (1.6) .9370 QT interval prolongation 5 (0.7) 5 (1.0) .5789 Ivabradine initiation status (n \[%\]) Initiated in the participating practice/clinic 553 (77.8) 385 (76.1) Initiated outside the participating practice/clinic 158 (22.2) 121 (23.9) Status of subsequent prescription(s) (n \[%\]) Written by physician in the practice/clinic 400 (56.3) 268 (53.0) Written by physician outside of the practice/clinic 168 (23.6) 148 (29.2) Physician has no information on subsequent prescription(s) 143 (20.1) 90 (17.8) Abbreviation: RMM, risk‐minimization measures. ### 3.1.1. Ivabradine dosing at treatment initiation and during follow‐up {#pds4880-sec-0012} Overall, 660 patients (92.8%) and 475 patients (94.1%) in the pre‐RMM and post‐RMM periods had initial ivabradine doses less than or equal to 5‐mg bid, in line with the SmPC. Results were similar regardless of age (92.0% of patients aged \<75 years in the pre‐RMM and 92.9% in the post‐RMM periods and 94.4% of patients aged ≥75 years in the pre‐RMM and 96.1% in the post‐RMM periods had initial ivabradine doses of ≤5‐mg bid). In the post‐RMM period, the percentages of patients prescribed the lowest initial dose of ivabradine (≤2.5‐mg bid) increased compared with the pre‐RMM period. This increase was greater for patients aged greater than or equal to 75 years (29.6% \[n = 73\] in the pre‐RMM period to 36.1% \[n = 56\] in the post‐RMM period) relative to those aged less than 75 years (25.0% \[n = 116\] in the pre‐RMM period to 30.2% \[n = 106\] in the post‐RMM period). During follow‐up, there were no ivabradine prescriptions that exceeded 7.5‐mg bid in either pre‐ or post‐RMM periods, in line with SmPC. ### 3.1.2. Concomitant medication use {#pds4880-sec-0013} The proportion of patients with no concomitant use of verapamil or diltiazem increased in post‐RMM period (96.1% and 99.2% in the pre‐RMM and post‐RMM periods, difference: 3.2; 95% CI, 1.3‐5.0; *P* value = .0007). ### 3.1.3. HR at initiation {#pds4880-sec-0014} The proportion of patients with HR ≥ 70 bpm at treatment initiation increased in post‐RMM period compared with pre‐RMM period (521 patients \[79.4%\] in the pre‐RMM period and 396 patients \[85.2%\] in the post‐RMM period; *P* value = .0141). 3.2. Overall patterns of use pre‐ and post‐RMM {#pds4880-sec-0015} ---------------------------------------------- The overall proportion of patients treated with ivabradine according to the composite of the four criteria of SmPC increased in the post‐RMM period (70.6% in the pre‐RMM period and 78.4% in the post‐RMM period, difference: 7.8; 95% CI, 2.5‐12.9; *P* value = .0035) (Table [3](#pds4880-tbl-0003){ref-type="table"}). As per country, the proportion of patients treated with ivabradine according to the composite of the four criteria of SmPC increased in the post‐RMM study period compared with pre‐RMM period in France, Germany, Spain, and the UK but not in Italy mainly because of a lower‐HR threshold adherence (84.2% in pre‐RMM vs 79.8% in post‐RMM). Similar increase in compliance to the RMM was observed in both GPs and specialists (Table [4](#pds4880-tbl-0004){ref-type="table"} **)**. ###### Compliance with RMMs: Overall pattern of ivabradine prescribing Criteria Pre‐RMM (N = 711) (n \[%\]) Post‐RMM (N = 506) (n \[%\]) Difference (95% CIs) *P* value ------------------------------------------------------------------------------------------------------------------------------------------ ----------------------------- ------------------------------ ---------------------- ----------- Heart rate Heart rate at treatment initiation ≥70 bpm[a](#pds4880-note-0007){ref-type="fn"} 521 (79.4) 396 (85.2) 5.7 (1.0‐10.3) .0141 Unknown/missing (%) 55 (7.7) 41 (8.1) SmPC dose No ivabradine dose higher than the SmPC doses at treatment initiation and during follow‐up 660 (92.8) 475 (94.1) 1.2 (−1.8 to 4.1) .3957 Unknown/missing (%) 0 (0) 1 (0.2) No ivabradine dose higher than the SmPC doses at treatment initiation 660 (92.8) 475 (94.1) 1.2 (−1.8 to 4.1) 0.3957 Unknown/missing (%) 0 (0) 1 (0.2) No ivabradine dose higher than the SmPC doses during follow‐up (among patients with renewals data)[b](#pds4880-note-0008){ref-type="fn"} 382 (100) 276 (100) 0.0 (−1.2 to 1.7) NC Patients with renewals data 382 276 Missing dose among renewals recorded (%) 0 (0) 0 (0) Verapamil/diltiazem use No concomitant use of verapamil or diltiazem at ivabradine treatment initiation and during follow‐up 683 (96.1) 502 (99.2) 3.2 (1.3‐5.0) .0007 Unknown/missing (%) 0 (0) 0 (0) Treated according to current SmPC^c^ (four criteria) 466 (70.6) 366 (78.4) 7.8 (2.5‐12.9) .0035 Unknown/missing (%) 51 (7.2) 39 (7.7) Abbreviation: bpm, beats per minute; CI, confidence interval; NC, not calculated; RMM, risk minimization measures; SmPC, summary of product characteristics. Based on the latest heart rate measurement available prior to or on ivabradine initiation date. If multiple values are available for the same date, this criterion is satisfied if the mean is greater than or equal to 70 bpm. Patients with recorded renewals that had missing dose are classified as noncompliant. Corresponds to patients prescribed ivabradine according to the heart rate recommendation, no doses higher than the SmPC doses at treatment initiation and during follow‐up (if available) and no concomitant use of verapamil or diltiazem during the study period. Patients with missing data for a given criterion but who failed to satisfy another criterion are classified as noncompliant (and not counted in the missing data row). ###### Compliance with RMMs: Overall pattern of ivabradine prescribing by country and by specialty Treated According to the Four Criteria of Current SmPC[a](#pds4880-note-0012){ref-type="fn"} ----------------- ---------------------------------------------------------------------------------------------- ---------------- ---------------- Country France n patients/N (%) 150/218 (68.8) 69/94 (73.4) Unknown/missing 5 5 Germany n patients/N (%) 112/143 (78.3) 156/187 (83.4) Unknown/missing 3 1 Italy n patients/N (%) 108/134 (80.6) 71/95 (74.7) Unknown/missing 20 29 Spain n patients/N (%) 65/105 (61.9) 33/46 (71.7) Unknown/missing 8 2 UK n patients/N (%) 31/60 (51.7) 37/45 (82.2) Unknown/missing 15 2 Specialty GPs n patients/N (%) 114/169 (67.5) 123/153 (80.4) Unknown/missing 18 7 Specialists n patients/N (%) 352/491 (71.7) 243/314 (77.4) Unknown/missing 33 32 *Note*. The denominator is patients who reported information for each of the criteria (patients with no subsequent prescriptions are included in the denominator). Abbreviations: GP, general practitioner; N, total number of patients in the strata; RMM, risk minimization measures; SmPC, summary of product characteristics. Corresponds to patients prescribed ivabradine according to the heart rate recommendation, no doses higher than the SmPC doses at treatment initiation and during follow‐up (if available) and no concomitant use of verapamil or diltiazem during the study period. 3.3. Sensitivity analysis {#pds4880-sec-0016} ------------------------- Except one patient at initiation in post‐RMM period, doses (at initiation and during follow‐up) and concomitant use with verapamil and diltiazem were completely reported in the patients\' chart in both study periods. The percentage of missing key data was higher for HR at initiation (55 patients \[7.7%\] and 41 patients \[8.1%\] in pre‐ and post‐RMM periods, respectively) than for the other criteria. The increase in compliance with the new RMMs remained significant when patients with missing data were included in the denominator (*P* value = .0121). The sensitivity analysis taking into account the physicians\' initiator status showed a lower proportion of missing data on HR at initiation when the participating physician was the initiator (4.2% and 3.4% in the pre‐RMM and post‐RMM periods, respectively) than when the participating physician was the subsequent prescriber (20.3% and 23.1%, respectively). The compliance with the four criteria increased between the pre‐ and post‐RMM periods among patients for whom the initiator (ie, who initially prescribed ivabradine) was the participating physician (from 70.4% to 80.2%). No increase was observed in patients recruited by subsequent prescribers. 4. DISCUSSION {#pds4880-sec-0017} ============= The overall objective of this PASS was to assess in five European countries how ivabradine is used in patients with chronic stable angina pectoris in routine clinical practice and to evaluate the compliance with the new RMMs. Patients were identified across a variety of physician specialties, including specialists from private and hospital settings, as well as GPs. Overall, the RMMs were well implemented across the participating countries and among both specialists and GPs. One of the specific study aims was to check that appropriate ivabradine doses were prescribed. These recommended doses had not changed following the benefit‐risk assessment, explaining the compliance being already high in the pre‐RMM period. Absence of dosing higher than 7.5‐mg bid in both periods is also reassuring compared with the SIGNIFY[2](#pds4880-bib-0002){ref-type="ref"} maintenance dose of 10‐mg bid. Similarly, concomitant use of ivabradine with verapamil or diltiazem, already not recommended before RMM, rarely occurred in both periods. Regarding HR, the threshold in patients with angina changed from greater than 60 bpm to greater than or equal to 70 bpm with RMM implementation. Although 79.4% of patients had a HR ≥ 70 bpm in the pre‐RMM period, this proportion increased significantly post‐RMM. This study aimed to ensure selection of a diverse and generally representative physicians\' sample and their treated patients. However, as for most studies conducted for regulatory reasons,[6](#pds4880-bib-0006){ref-type="ref"} this study faced a very low interest rate among physicians and recruitment challenges. Having a total of 68 active physicians across the five countries needs to be considered when interpreting the results. The study was not powered to assess compliance with RMMs at country and specialty levels. This is particularly true for Italian GPs and UK specialists (one and two active physicians, respectively). The assessment of any potential difference between participating and nonparticipating physicians was not possible because data on nonparticipating physicians were very scarce. No robust conclusion could be drawn on representativeness of participating physicians compared with the general medical population. However, physicians\' sampling was done based on very large source lists, iteratively through six waves of recruitment. A total of 60 675 physicians were invited to participate in the study of whom 522 (0.86%) were interested. This low rate was driven by the strategy of mass mailing, chosen to optimize the absolute number of interested sites and to meet regulatory timelines. Given that invited physicians were informed of the study objectives and that physician participation was voluntary, there was potential for bias towards more participation of physicians who prescribed ivabradine as recommended in the SmPC. However, physicians\' recruitment activities started after the post‐RMM period was over, ensuring that prescriptions issued during this period were not influenced by physicians\' awareness of the study objectives. The initial target of 600 patients per period was not reached for the post‐RMM period, which was shorter than the pre‐RMM period; yet, the inclusion of 506 patients was acceptable with a sufficient level of absolute precision to meet the study objective. Patients\' characteristics were in line with those usually observed in populations with chronic stable angina pectoris in terms of age and sex distribution and most frequent comorbidities.[7](#pds4880-bib-0007){ref-type="ref"} HF was significantly more frequent among patients in the post‐RMM period, in line with the new indication of chronic HF granted in 2012. While low proportions of patients with missing key data were observed during the study, the proportions were higher for HR at initiation (7.7% and 8.1% in the pre‐RMM and post‐RMM periods, respectively). Missingness of data was further impacted by SCM. HR values at initiation were more likely to be missing when the participating physician was a subsequent prescriber. However, it does not imply that the HR measure had not been performed at initiation. Moreover, it can be assumed that patients with missing HR at initiation and included by the subsequent prescriber had HR distribution similar to the one observed among patients with recorded HR and included by the initiator. Measuring RMMs\' effectiveness is an important aspect of a drug benefit‐risk evaluation. This study design is aligned with the guideline on good pharmacovigilance practices (GVP) module XVI,[8](#pds4880-bib-0008){ref-type="ref"} where a medical chart abstraction is considered as a valuable option to assess clinical knowledge and prescribing behavior and is preferred to surveys or self‐reported data. The pre‐post retrospective study design, widely used among similar PASS and DUS,[9](#pds4880-bib-0009){ref-type="ref"}, [10](#pds4880-bib-0010){ref-type="ref"}, [11](#pds4880-bib-0011){ref-type="ref"}, [12](#pds4880-bib-0012){ref-type="ref"}, [13](#pds4880-bib-0013){ref-type="ref"}, [14](#pds4880-bib-0014){ref-type="ref"} allows for showing the immediate impact of short‐term programs such as RMMs.[15](#pds4880-bib-0015){ref-type="ref"}, [16](#pds4880-bib-0016){ref-type="ref"} A review of 29 studies in the European Union electronic Register of Post‐Authorization Studies found that only four studies used retrospective medical files review---others were surveys---and only 10 studies were conducted within a 12--18‐month timeframe after RMM implementation.[17](#pds4880-bib-0017){ref-type="ref"} The main strength of this PASS remains its conclusiveness on the effectiveness of RMMs on the prescribing of ivabradine since EU marketing authorization. A review of studies registered in EU PAS Register showed that only half of the effectiveness indicators were reported as successful and conclusive.[18](#pds4880-bib-0018){ref-type="ref"} In addition, the current study did examine the compliance with each individual component of the RMMs and the composite. Overall, although most patients were already prescribed ivabradine according to RMMs, the study was able to show a significant improvement in compliance to these RMMs in the post‐RMM period. 5. CONCLUSION {#pds4880-sec-0018} ============= Overall, the study results show that the RMMs for ivabradine were well implemented across the participating European countries. Ivabradine prescribing patterns have significantly changed to be in line with newly implemented RMMs and the updated SmPC, confirming a favorable benefit‐risk balance of ivabradine in chronic stable angina pectoris and maintaining the ivabradine EU marketing authorization. ETHICS STATEMENT {#pds4880-sec-1023} ================ All required national or local ethical committees approvals were obtained. CONFLICT OF INTEREST {#pds4880-sec-0020} ==================== Linda Salem, Nicolas Deltour, and Emmanuelle Jacquot are employees of Institut de Recherches Internationales Servier. Alexandre Malouvier and Jon Blatchford are employees of PRA Health Sciences, the Contract Research Organization to which the research project was contracted by Servier, the company that owns ivabradine. E Rivero‐Ferrer works for RTI Health Solutions, a business unit of RTI International, which has been compensated for work done for this research. RTI International is an independent, nonprofit research organization that conducts work for government, public, and private organizations, including pharmaceutical companies. AUTHORS CONTRIBUTION {#pds4880-sec-0021} ==================== Authors E.J., N.D., and E.R.F. planned the study. E.J., N.D., L.S., and E.R.F. conducted feasibility evaluation and drafted the study protocol. E.J., N.D., and L.S. made contributions to the final design and final approved version of the protocol. J.B. undertook the statistical analysis of the study. J.B. completed the analysis, and all authors contributed in the interpretation of the results. All authors contributed to the writing of the first draft of the manuscript. All authors contributed to and have approved the final manuscript. The authors thank RTI‐Health Solutions particularly Dr Susana Perez‐Gutthann for their contributions in planning the study, developing the study protocol, and conducting the pilot study. The authors also thank Mr Jean‐Christian Nachef and Mr Didier Bessam for their implication in the implementation and the conduction of the study. Medical writing support was provided by PRA Health Sciences.

1. Introduction {#sec1} =============== Endothelial dysfunction is one of the most critical risk factors of diabetes leading to morbidity and is characterized by insulin deficiency or impaired insulin signaling \[[@B1], [@B2]\]. Increasing amounts of evidence have shown that oxidative stress is associated with the pathogenesis of many diseases, including diabetes \[[@B3]--[@B5]\]. Although there has been a significant amount of study into the precise mechanisms of oxidative stress in vascular dysfunction in diabetes \[[@B6]--[@B8]\], there is a dearth of effective strategies to decrease the incidence of diabetic vascular disease. More progress in the understanding of endothelial dysfunction at the molecular level is urgently needed in order to develop new therapies for clinical application. Profilin-1, a member of the profilin family, was first characterized as a small actin-binding protein and is widely distributed in various types of cells \[[@B9]\]. Profilin-1 has been found to be increased in the endothelial cells and macrophages of atherosclerotic lesions in diabetic individuals, and oxidized LDL (oxLDL) was capable of triggering profilin increases \[[@B10]\]. Further investigation has shown that a knockdown of profilin-1 was protective against endothelial dysfunction triggered by oxLDL in cultured aortic endothelial cells \[[@B11]\]. Overexpression of transgenic profilin 1 resulted in the elevation of blood pressure via medicating vascular remodeling \[[@B12]\]. Profilin-1 also plays a crucial role in hypertension-induced artery remodeling through the p38-iNOS-peroxynitrite pathway \[[@B13]\]. Recently, profilin-1 was reported to be involved in endothelial injury induced by AGEs \[[@B14]\]. RhoA/ROCK1 has been reported to downregulate eNOS, which results in impaired endothelial function and vascular relaxation \[[@B15]--[@B17]\]. These data suggest that profilin-1 plays a critical role in regulation of oxidative stress and vascular remodeling. However, the exact roles and mechanisms of profilin-1 in endothelial abnormalities and vascular disease remain largely unknown. In this work, we assume that profilin-1 is induced by AGEs and that AGE-stimulated profilin-1 activation is necessary for ROS production and endothelial injury. The aim of the present study was to clarify the molecular mechanisms of profilin-1 in endothelial injury, with the goal of identifying a potential target for genetic therapy of diabetes-induced arterial remodeling. 2. Methods and Materials {#sec2} ======================== 2.1. Cell Culture and Treatments {#sec2.1} -------------------------------- Human umbilical vein endothelial cells (HUVECs) were cultured as described previously \[[@B18]\]. Cell-permeative Rho inhibitor C-3 transferase (CT-04), a RhoA inhibitor, was purchased from Cytoskeleton (Denver, CO, USA). Y-27632, Rho kinase inhibitor, and Tiron (4,5-dihydroxy-1,3-benzene disulfonic acid-disodium salt), an ROS scavenger, were purchased from Sigma-Aldrich (St. Louis, MO, USA) and dissolved in DMSO. 2.2. Plasmids and Transfection {#sec2.2} ------------------------------ Human profilin-1 cDNA was subcloned into pCDH-CMV-MCS-EF1-copGFP lentiviral vector (System Biosciences, Mountain View, CA, USA). Lentiviral vectors harboring the profilin-1 expressing plasmid accompanied with the pPACKH1 HIV Lentivector Packaging Kit (System Biosciences) were transfected into HEK-293 T cells according to manufacturer\'s instructions. HUVECs were infected with profilin-1 in the presence of 4 *μ*g/ml polybrene (Sigma-Aldrich, St. Louis, MO, USA) and selected using a Coulter EPICS Elite ESP flow cytometer (Beckman, Miami, Florida, USA) based on the expression of GFP. For knockdown assays, profilin-1, siRNA targeting profilin-1, and the scrambled nontarget control (NC) were purchased from GenePharma (Shanghai, China) and transfected into HUVECs using HiPerFect transfection reagent (Qiagen, Valencia, CA, USA) according to manufacturer\'s instructions. 2.3. Apoptosis Assay {#sec2.3} -------------------- Apoptosis of cells was determined by flow cytometric analysis using PE Annexin V Apoptosis Detection Kit I (BD Biosciences, Franklin Lakes, NJ, US) as described previously \[[@B18]\]. 2.4. Cell Migration Assay and Tube Formation Assay {#sec2.4} -------------------------------------------------- Cell migration potential was assessed by transwell assays as previously reported \[[@B18]\]. For the tube formation assay, after appropriate treatment, approximately 4 × 10^4^ HUVECs (in 400 *μ*l) were added into a 24-well plate, precoated with Matrigel (BD Biosciences), and incubated for 6 h. Random photographs were taken under a light microscope (Olympus, Tokyo, Japan) at 100x magnification to assay for the formation of capillary-like structures. 2.5. Intracellular ROS Measurement {#sec2.5} ---------------------------------- Intracellular ROS was measured using 2,7-dichlorodihydrofluorescein diacetate dye (H2DCFDA, Life Technologies) as described previously \[[@B19]\]. 2.6. Western Blots and RhoA Pull-Down Assay {#sec2.6} ------------------------------------------- Western blotting was performed using standard methods as previously reported \[[@B19]\]. Antiprofilin-1 antibody (\#3237), anti-ROCK1 (\#4035), and anti-p-eNOS antibodies (\#9570) were purchased from Cell Signaling Technology (Beverly, MA, USA). Anti-*β*-actin antibody (sc-47778) was obtained from Santa Cruz Biotechnology (Santa Cruz, CA, USA). RhoA activation was assessed using the RhoA Activation Assay Biochem Kit (Cytoskeleton, Denver, CO, USA) according to previously described methods \[[@B18]\]. 2.7. Statistical Analysis {#sec2.7} ------------------------- The data are presented as the mean ± SD (standard deviation) of at least three independent experiments. Student\'s *t*-test was used to compare two groups, and one-way analysis of variance (ANOVA) was used to compare multiple variables. *P* \< 0.05 was considered as significant. 3. Results {#sec3} ========== 3.1. AGEs Upregulate Profilin-1 Expression, Endothelial Injury, and ROS Generation in HUVECs {#sec3.1} -------------------------------------------------------------------------------------------- Previous studies have shown the effects of AGEs on HUVECs \[[@B14], [@B20]\], but the results were inconsistent. We first confirmed the functions of AGEs on HUVECs. ROS production in HUVECs was significantly increased after a 24 h treatment with 200 *μ*g/ml AGEs ([Figure 1(a)](#fig1){ref-type="fig"}). Flow cytometric analysis was performed to assess the effects of AGEs on cell apoptosis. Results indicated that cell apoptosis was significantly increased after AGE treatment ([Figure 1(b)](#fig1){ref-type="fig"}). The tube formation assay also showed that treatment with AGEs significantly decreased the ability to form tubes ([Figure 1(c)](#fig1){ref-type="fig"}). In addition, the ability of cells to migrate was also significantly hampered after AGE treatment ([Figure 1(d)](#fig1){ref-type="fig"}). Western blot analysis revealed that AGEs increased the expression of profilin-1 in HUVECs exposed to AGEs ([Figure 1(e)](#fig1){ref-type="fig"}). 3.2. Inhibition of Profilin-1 Attenuates AGE-Induced Endothelial Damage and ROS Generation in HUVECs {#sec3.2} ---------------------------------------------------------------------------------------------------- Profilin-1 has been shown to be associated with endothelial abnormalities and ROS production in HUVECs \[[@B14]\]. We examined whether profilin-1 was involved in the activation of ROS production by AGEs. Profilin-1 expression was significantly reduced in cells transfected with profilin-1 siRNA compared with blank (not transfected) or NC cells ([Figure 2(a)](#fig2){ref-type="fig"}). Inhibition of profilin-1 in HUVECs resulted in a drastic decrease of ROS ([Figure 2(b)](#fig2){ref-type="fig"}). Flow cytometry showed that knockdown of profilin-1 reversed the apoptotic effects caused by AGEs ([Figure 2(c)](#fig2){ref-type="fig"}). The average number of tubular structures was increased when profilin-1 was silenced ([Figure 2(d)](#fig2){ref-type="fig"}). Migration assays showed that HUVECs transfected with profilin-1 siRNA displayed a notable increase when compared with NC ([Figure 2(e)](#fig2){ref-type="fig"}), followed by a decrease in the expression of ROCK1 and p-RhoA, and an increase in the expression of p-eNOS ([Figure 2(a)](#fig2){ref-type="fig"}). 3.3. Overexpression of Profilin-1 Promotes Endothelial Damage and ROS Generation under AGE Treatment {#sec3.3} ---------------------------------------------------------------------------------------------------- To further explore whether profilin-1 was required for AGE-induced endothelial damage and ROS production, we investigated the effects of overexpression of profilin-1 on the biological behavior of HUVECs. HUVECs were transduced with profilin-1, and expression was confirmed by Western blot ([Figure 3(a)](#fig3){ref-type="fig"}). Compared with vector-only cells, HUVECs transduced with profilin-1 had significantly increased ROS production based on DCFDA assays ([Figure 3(b)](#fig3){ref-type="fig"}). Flow cytometry analysis of HUVECs showed that overexpression of profilin-1 induced apoptosis when compared with the vector-only cells ([Figure 3(c)](#fig3){ref-type="fig"}). Tube formation assays indicated that overexpression of profilin-1 could lead to a decrease in tube formation ([Figure 3(d)](#fig3){ref-type="fig"}). Moreover, transwell assays showed that increased profilin-1 expression inhibited HUVEC migration ([Figure 3(e)](#fig3){ref-type="fig"}). All the above results were obtained from HUVEC cells treated with 50 *μ*g/ml AGEs. However, there was no significant change in HUVECs under normal state (data not shown). Consistent with cell data, Western blot analysis demonstrated that profilin-1 increased the expression of ROCK1 and p-RhoA and decreased the expression of p-eNOS in HUVECs treated with 50 *μ*g/ml AGEs ([Figure 3(a)](#fig3){ref-type="fig"}). These results suggested that profilin-1 requires the presence of AGEs to induce endothelial damage and ROS generation. 3.4. Involvement of RhoA/Rock1 in Profilin-1-Mediated Endothelial Injury {#sec3.4} ------------------------------------------------------------------------ RhoA/Rock1 is known to induce endothelial injury \[[@B21], [@B22]\]. To investigate whether RhoA/Rock1 is involved in profilin-1-mediated endothelial injury, we treated cells that stably overexpress profilin-1 with CT-04 (1 *μ*g/ml) or Y-27632 (10 *μ*M). As shown in [Figure 4(a)](#fig4){ref-type="fig"}, CT-04 and Y-27632 were able to increase the expression of p-eNOS that was downregulated by profilin-1 ([Figure 4(a)](#fig4){ref-type="fig"}). Interestingly, this suppressive effect of profilin-1 was prevented when HUVECs were pretreated with CT-04 or Y-27632, which suggested the involvement of RhoA/ROCK1 in the profilin-1-mediated inhibitory effects (Figures [4(b)](#fig4){ref-type="fig"}--[4(e)](#fig4){ref-type="fig"}). 3.5. ROS Were Required for Profilin-1 to Activate the RhoA/Rock1 Pathway in HUVECs {#sec3.5} ---------------------------------------------------------------------------------- Previous reports have shown that ROS could activate RhoA/ROCK1 and cause endothelial cell injury \[[@B17], [@B23], [@B24]\]. Here, we determined whether ROS exerted their inhibitory effects on endothelial injury through the RhoA/ROCK1 signaling pathway. HUVECs stably overexpressing profilin-1 were exposed to AGEs (50 *μ*g/ml) for 24 h in the presence and absence of Tiron, an ROS scavenger. As shown in [Figure 5(a)](#fig5){ref-type="fig"}, we observed that CT-04 prevented profilin-1-induced increases of p-RhoA and ROCK1 and decreases of p-eNOS. Moreover, profilin-1 significantly increased ROS production and Tiron (5 mM) prevented the increase of ROS in HUVECs exposed to AGEs ([Figure 5(b)](#fig5){ref-type="fig"}). The alteration of cell apoptosis, tube formation, and migration induced by profilin-1 was reverted by Tiron treatment (Figures [5(c)](#fig5){ref-type="fig"}--[5(e)](#fig5){ref-type="fig"}). Taken together, these observations indicate that profilin-1 activated the RhoA/ROCK1 pathway through ROS. 4. Discussion {#sec4} ============= Numerous studies have suggested that diabetes is associated with various cardiovascular complications, which are the principal causes of morbidity and mortality of patients with diabetes \[[@B2], [@B3], [@B25]\]. Increasing amounts of evidence have indicated that larger amounts of ROS, produced by electron transport chain activation, result in oxidative stress and further aggravate the progression of diabetes and its complications \[[@B26], [@B27]\]. Profilin-1 has been found to regulate vascular remodeling via vascular inflammation and oxidative stress \[[@B11], [@B12], [@B14]\]. In the current study, we demonstrated that AGEs could induce profilin-1 expression and profilin-1 mediated endothelial damage and oxidative stress via the RhoA/ROCK1 signaling pathway and that the inhibition of profilin-1 attenuates RhoA/ROCK1 pathway activity and oxidative stress in HUVECs exposed to AGEs. Previous studies have demonstrated that AGEs cause vascular dysfunction by interacting with their receptors and activating complex signaling pathways, resulting in the increased generation of ROS \[[@B14], [@B28], [@B29]\]. We investigated the influence of AGEs on profilin-1 in HUVECs to explore the underlying mechanisms of intracellular ROS production by AGEs. Our data showed that profilin-1 was significantly increased in HUVECs treated with AGEs and accompanied with release of ROS. These results indicated that AGEs have the ability to cause endothelial injury and increase the expression of profilin-1. In contrast, inhibition of profilin-1 with siRNA reversed AGE-induced ROS production and alteration of apoptosis. The RhoA/ROCK1 signaling pathway is known to play an important role in endothelial cell injury caused by hyperglycemia in vitro \[[@B30], [@B31]\]. In the present study, we found that AGE-induced RhoA and ROCK1 were markedly decreased after profilin-1 silencing. The above results revealed that profilin-1 was participating in the AGE-induced endothelial damage and ROS generation. Additionally, overexpression of profilin-1 promoted endothelial cell injury and ROS production, followed by the increase of p-RhoA and ROCK1 in HUVECs treated with 50 *μ*g/ml AGEs. These results suggest that profilin-1 requires the presence of AGEs to induce endothelial damage and ROS generation. Furthermore, profilin-1-mediated cell functions in HUVECs were partly blocked with the treatment of CT-04 or Y-27632. These findings provide evidence to support the fact that the RhoA/ROCK1 signaling pathway plays an important role in the generation of excess ROS by profilin-1 in HUVECs after exposure to extrinsic AGEs. It is well documented that increased ROS production contributes to endothelial injury in diabetes \[[@B3], [@B32]\]. Further investigation has shown that ROS could activate the Rho/ROCK1 signaling pathway, which is a key modulator of the actin cytoskeleton and plays an important role in oxidative stress \[[@B17], [@B31]\]. Whether ROS were involved in profilin-1-mediated Rho/ROCK1 signaling pathway was not known. Similar to the profilin-1 knockdown, we found that CT-04 prevented profilin-1-inducted ROS production and apoptosis and inhibited vascular-like tube formation and migration. These results indicated that ROS was required for profilin-1 activation of the Rho/ROCK signaling pathway. Shao et al. have shown that ROCK1 could mediate phosphorylation of profilin-1 at Ser-137 in neurodegeneration \[[@B33]\]. Thus, we concluded that increased profilin-1, induced by AGEs, resulted in ROS release and subsequently activated the Rho/ROCK1 signaling pathway ([Figure 6](#fig6){ref-type="fig"}). In conclusion, our data provide new molecular insights into the regulation of the endothelial dysfunction induced by AGEs and may provide novel therapeutic strategies for clinical treatment. The study was supported by grants from the Shanghai Municipal Health Bureau (no. 20144Y0078) and the Shanghai Science and Technology Committee of Qingpu District (no. QKF 2012-1). Conflicts of Interest ===================== The authors declare that they have no competing interests. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} [^1]: Academic Editor: Vladimir Jakovljevic

Introduction ============ There are no prospective studies comparing outcomes between restrictive or liberal strategies in cardiac surgery. This study is a double-blind randomized study to determine whether a restrictive strategy of red cell transfusion and a liberal strategy produced equivalent results in patients submitted to cardiac surgery. Methods ======= Until November 2009 we enrolled 380 patients undergoing elective cardiac surgery and randomly assigned 185 patients to a restrictive strategy of transfusion, in which red cells were transfused if the hematocrit concentration dropped below 24%, and 195 patients to a liberal strategy, in which transfusions were given when the hematocrit fell below 30%. Both transfusion strategies were followed into the operative room and during the ICU stay. We compared the rates of complications and death from all causes during the hospital stay and clinical outcomes of patients after 90 days. Adult patients, after written informed consent, were allocated if they would be submitted to elective primary surgery or were redone adult cardiac surgical patients for coronary artery bypass grafting, valve procedure or combined procedures. Results ======= Overall, hospital mortality was similar in the two groups (4.7% vs 5.3%, *P*= 0.11). The rates of complications were similar in the two groups (28.3% in the restrictive-strategy group and 26.1% in the liberal-strategy group, *P*= 0.33). The ICU stay was significantly lower in the restrictive-strategy group (2.2 days vs 4.1 days, *P*= 0.05). There was no difference in 90-day mortality between groups. Conclusions =========== A restrictive strategy of red cell transfusion is at least as effective as and possibly superior to a liberal transfusion strategy in patients submitted to cardiac surgery.

All relevant data are within the paper and its Supporting Information file. Introduction {#sec005} ============ Prostate cancer (PCa) is the most common malignancy among men and a remarkable public health challenge in Western countries. In United States, more than 230 000 new cases and 29 000 PCa related deaths was expected to be diagnosed in 2014 \[[@pone.0140671.ref001]\]. The major risk factors for PCa include age, positive family history and African-American race \[[@pone.0140671.ref002]\]. The majority of patients are currently treated with radical prostatectomy and/or radiotherapy while active surveillance may be beneficial for low risk patients. In most of the cases, the disease is local and only \~7% of patients treated with radical prostatectomy (RP) die from PCa during a 15-year follow-up \[[@pone.0140671.ref003]\]. Furthermore, it seems clear that the treatment of low-grade cancer (Gleason score 6 or below) confers no survival benefit when compared to other causes of death. The use of prostate-specific antigen (PSA) has revolutionized the diagnostics of PCa during the past 25 years and more recently systematic PSA screening has been shown to decrease PCa-related mortality \[[@pone.0140671.ref004]\]. However, PSA lacks the specificity to detect clinically significant cancers and the drawback is over-diagnosis of cancers that would not have effect on the quality and expectancy of life if left untreated \[[@pone.0140671.ref005]\]. This in turn leads to overtreatment with potential side effects. In order to avoid over-treatment, there is an indisputable need for reliable biomarkers to identify the potentially aggressive and lethal cases from those that remain local. Nuclear lamins are type V intermediate filament (IF) proteins. They are the major components of the nuclear lamina, a fibrous meshwork of proteins that underlies the inner nuclear membrane \[[@pone.0140671.ref006]\]. Lamins are devided into two subgroups, A-type and B-type lamins. A single gene, *LMNA*, encodes all A-type lamins that include lamin A, lamin C, a testis-specific isoform lamin C2 and minor isoform lamin AΔ10. The major B-type lamins, lamin B1 and lamin B2, are encoded by two genes (*LMNB1* and *LMNB2*, respectively) \[[@pone.0140671.ref007]\]. In addition, a germ cell specific lamin B3 is generated from *LMNB2* through alternative splicing \[[@pone.0140671.ref008]\]. At least one B-type lamin is expressed in all metazoan cells, while A-type lamins are primarily expressed in differentiated cells \[[@pone.0140671.ref007]\]. The lamins have multiple functions in the cell nucleus. They provide mechanical support for the nucleus and play a role in DNA replication and repair, transcription, chromatin organization and gene regulation \[[@pone.0140671.ref007]\]. Numerous mutations in *LMNA* cause a group of inherited human diseases or disorders commonly called the laminopathies. These include, for example, cardiac and skeletal myopathies, lipodystrophies, peripheral neuropathy and premature aging disease known as progeria syndrome \[[@pone.0140671.ref009]\]. While the fundamental pathobiology of these diseases is mostly poorly understood, abnormalities of nuclear shape (e.g. herniations and lobulations) are commonly reported in cells carrying disease-associated lamin mutations. Interestingly, similar morphological changes of the nucleus are frequently observed in cancer cells including PCa cells \[[@pone.0140671.ref010]\]. Therefore, it is possible that altered expression and modifications in lamins contribute malignant transformation in cancer cells \[[@pone.0140671.ref011]\]. Only a few retrospective studies have substantiated the association between altered lamin expression and patient outcome. In stage I-III colorectal cancer (CRC), the patients with lamin A expressing tumors had significantly increased risk to die from CRC when compared to patients with lamin A negative tumors \[[@pone.0140671.ref012]\]. Based on *in vitro* studies, the authors further concluded that this is due to increased cell motility, invasiveness and stem cell-like phenotype promoted by lamin A/C. On the contrary, Belt et al. reported that loss of lamin A/C in stage II-III colon cancer is associated with disease recurrence \[[@pone.0140671.ref013]\]. In nodal diffuse large B-cell lymphoma inactivation of lamin A/C gene by CpG island promoter hypermethylation is associated with poor survival \[[@pone.0140671.ref014]\]. The role of lamins in PCa has been studied more recently. Using mass spectrometry (MS) and immunohistochemistry, Skvortsov et al. reported decreased expression of lamin A in low grade (Gleason score 6) tumors relative to paired samples from histologically normal tissue while an increased expression was detected in high risk (Gleason score 8) tumors \[[@pone.0140671.ref015]\]. Furthermore, lamin A/C is overexpressed at the invasive front of PCa tissue and promotes cell growth, migration and invasion through the PI3K/AKT/PTEN pathway under culture conditions \[[@pone.0140671.ref016]\]. The expression of B-type lamins, on the other hand, is increased in cancerous prostate tissue and strongly correlates with Gleason score \[[@pone.0140671.ref017]\]. MS analysis further suggests that lamin B1 is hyperphosphorylated in androgen independent PC-3 cells \[[@pone.0140671.ref018]\]. Interestingly, chromosomal regions associated with PCa risk localize to nuclear lamin B-deficient microdomains (LDMD) that exhibit reduced gene transcription (10). While the nature of these structures is currently unclear, the frequency of LDMDs correlates with PCa cell line aggressiveness, cell motility and Gleason grade \[[@pone.0140671.ref010]\]. In summary, lamins undergo significant alterations during malignant transformation but to date it is unclear whether any of these changes have predictive value in PCa, especially in high grade cancer. In the present study we studied the expression of different lamins in PCa in detail using tissue microarray (TMA) material covering over 500 patients treated with radical prostatectomy and lymph node dissection. Our results show that decreased expression of A-type lamins is associated with elevated risk for lymph node metastasis and disease specific death while increased expression of lamin B1 is associated with BCR and local spread. Additionally, low lamin B2 expression predicted higher risk of PCa mortality. We also discuss the potential cell biological mechanisms and their significance for clinical risk stratification. Materials and Methods {#sec006} ===================== Tissue material {#sec007} --------------- Paraffin embedded formalin fixed tissue material from 501 patients operated with radical prostatectomy and limited pelvic lymphadenectomy in the Turku University Central Hospital between January 2000 and September 2005 were used for constructing the TMA blocks as previously described \[[@pone.0140671.ref019]\]. In brief, three to twelve (median 3.0) adjacent cores of 1 mm in diameter were obtained from the index carcinoma lesion that was considered the most significant (based on the Gleason grade pattern, the volume of cancer lesion and/or extra-prostatic extension). In addition, one core from histologically benign prostate tissue was obtained from the same patient. These tissue cores were transferred from the donor blocks to the recipient TMA blocks that contained in total 63 cores per block including control tissues from normal liver. All cases in the final TMA blocks were graded based on the International Society of Urological Pathology updated Gleason grading system \[[@pone.0140671.ref020]\]. At the time of surgery, no consent was obtained from the participants for this specific study. At the time of the study, the use of tissue material and the corresponding clinicopathological and follow-up data was approved by the Ethics Committee of the Hospital District of the Southwest Finland (130/180/2008) and the National Authority for Welfare and Health (Valvira 394/05.01.00.06/2009) according to the national legislation. Patient information was anonymized and de-identified prior to analysis. Clinical data {#sec008} ------------- None of the patients received androgen deprivation or radiotherapy prior to or immediately after the surgery. The follow-up was conducted by digital rectal examination and post-operative PSA measurement at least three times during the first year after the surgery and at least once a year during the following years. Of the 501 patients, 158 (31%) had a biochemical recurrence (BCR) defined as PSA level of 0.2 μg/L or above after the surgery. The staging of the PCas was performed according to the WHO pTNM classification system \[[@pone.0140671.ref021]\]. 28 patients (4.4%) died from PCa during the follow-up, while 39 (7.8%) died of other causes. Due to long time interval between operations, each patient's original clinical Gleason score (GS) was re-evaluated by experienced genitourinary pathologist according to the current Gleason grading system criteria \[[@pone.0140671.ref020]\] using hematoxylin-eosin sections from the original radical prostatectomy specimens. This score was used in the survival analyses and is presented in [Table 1](#pone.0140671.t001){ref-type="table"}. 10.1371/journal.pone.0140671.t001 ###### Clinicopathological variables of the patient cohort (n = 501). {#pone.0140671.t001g} Variable Number of patients --------------------------------------------------- --------------- -------------------- Age \<60 years 169 (34%) 60--65 years 187 (37%) \>65 years 143 (29%) PSA group (μg/L) \<10 306 (61%) 10--20 116 (23%) \>20 38 (8%) pT-category[^a^](#t001fn001){ref-type="table-fn"} pT2 209 (42%) pT3a 160 (32%) pT3b 52 (10%) Gleason group \<7 181 (36%) 7 199 (40%) \>7 96 (19%) Extraprostatic extension Yes 203 (41%) Surgical margin Positive 190 (38%) Lymph node status N+ 21 (4%) BCR Yes 157 (31%) Alive 444 (89%) Death due Pca 20 (4%) Death, other 37 (7%) Follow-up time (months) Median 91 Range 1--167 ^a^ Pathological T (pT) category refers to the stage of primary tumor in the TNM tumor staging system. pT2 = Tumor confined within prostate, pT3a = Extracapsular extension, pT3b = tumor invading seminal vesicle(s). Immunohistochemistry and microscopy {#sec009} ----------------------------------- In order to perform immunohistochemical (IHC) assessment, paraffin was first removed with xylene and the sections were rehydrated with graded series of alcohol. All the antibodies were tested on prostate TMA test blocks prior to use in the final TMA material in order to optimize the dilution and the most suitable pre-treatment. Epitope unmasking was carried out by microwaving the tissue slides in either Tris-HCl buffer (pH 9; lamin B1, B2 and C antibodies) or Citrate based buffer (pH 6; lamin A antibody) for 10 min. The primary antibodies used were mouse monoclonal anti-lamin A (1:1000, clone 133A2, Abcam), goat polyclonal anti-lamin B1 (1:4000, C-20, Santa Cruz), mouse monoclonal anti-lamin B2 (1:500, LN43, Abcam) and rabbit polyclonal anti-lamin C (1:150, RalC, Novus). Due to variation in the staining between different patches of lamin C antibody, only 199 patients stained with the same patch were included in the final statistical analysis. The stainings were carried out using Ventana automated staining machine and the primary antibodies were detected with Vectastain anti-mouse or anti-goat HRP-conjugated secondary antibodies. The slides were counterstained with hematoxylin and observed with Olympus BX60 microscope (Olympus Optical Co., Ltd., Tokyo, Japan). The TMA slides were graded visually for IHC grades, without the knowledge of clinicopathological characteristics of the patients. The staining intensities of each core were graded as 0 (no specific staining in carcinoma cells, intense staining in benign glands and stromal cells), 1 (low intensity nuclear lamina staining in carcinoma cells, more intense staining in benign glands or stromal cells), 2 (moderate intensity and clearly visible lamina in carcinoma cells, yet more intense staining in benign glands and stromal cells) or 3 (high intensity lamina staining equivalent to, or exceeding the intensity of normal glands and stromal cells). Patients whose tissue samples were stained poorly (based on internal controls) or detached from the slides during staining process were excluded from the analysis. The mean intensity values of 3 or more carcinoma cores were determined for each patient and eventually the patients were dichotomized into low (IHC grade 0 and 1) and high (IHC grades 2 and 3) expressing groups for further statistical analysis (except for lamin B1 and C, see [Table 2](#pone.0140671.t002){ref-type="table"}). Dichotomization was based on the distribution of different subgroups for each lamin staining in order to have comparable groups with sufficient number of patients (see [Table 2](#pone.0140671.t002){ref-type="table"}). 10.1371/journal.pone.0140671.t002 ###### The distribution of expression for each lamin in the patient cohort. {#pone.0140671.t002g} Number of patients ---------- ------------------ -------------------- ---------------- ----- Lamin A IHC grade 0--1 IHC grade 2--3 Entire group 251 (52.0%) 232 (48.0%) 483 Subgroup (GS\>6) 141 (50.0%) 141 (50.0%) 282 Lamin C IHC grade 0 IHC grade 1--3 Entire group 41 (22.9%) 138 (77.1%) 179 Subgroup (GS\>6) 22 (21.4%) 81 (78.6%) 103 Lamin B1 IHC grade 0--2 IHC grade 3 Entire group 142 (30.3%) 327 (69.7%) 469 Subgroup (GS\>6) 82 (30.0%) 191 (70.0%) 273 Lamin B2 IHC grade 0--1 IHC grade 2--3 Entire group 169 (37.5%) 282 (62.5%) 451 Subgroup (GS\>6) 88 (33.5%) 175 (66.5%) 263 \*GS = Gleason score Statistical analysis {#sec010} -------------------- The statistical analysis was performed with SPSS 20 (IBM). The correlations between clinicopathological variables and biomarkers were analyzed using chi-square test. The correlation between mean lamin expression in benign and cancerous prostate tissue from the same patients were analyzed with independent-samples T test. The Kaplan-Meier method, Log-rank test and Cox proportional hazards regression model were used to analyze the association of staining intensity to outcome. For outcome analyses, the time for BCR-free survival was calculated from the day of surgery to the day of detection of PSA of 0.2 μg/L or above. The disease specific survival time was calculated from the date of surgery to the date of the last follow-up visit or death. The survival analyses were performed for the entire cohort (except for lamin C, see above) and for a high risk subpopulation including patient with GS 7--10 tumors. All statistical tests were two-sided and p-values ≤0.05 were considered statistically significant. Results {#sec011} ======= In order to study whether expression of different lamins is associated with PCa biochemical progression and disease specific survival (DSS), TMA material containing at least triplicate cancerous samples and one morphologically benign sample from 501 radically operated PCa patients was stained with antibodies detecting different lamins (A, C, B1 and B2) and the staining intensities were compared to clinicopathological variables. The demographics of patients are presented in the [Table 1](#pone.0140671.t001){ref-type="table"}. In the Kaplan-Meier estimation analysis, the well-known clinicopathological features such as high GS and preoperative PSA, T category, extraprostatic extension (EPE) and seminal vesicle invasion (SVI) were statistically significantly associated with BCR indicating that our cohort is highly representative to test the correlation between the proteins of interest and patient outcome (data not shown). Similarly, high GS, T category, positive surgical margin, EPE, SVI and lymph node (LN) positivity predicted shortened time for disease-specific death (data not shown). In multivariable Cox regression analysis that took other prognostic markers into account, high GS, preoperative PSA and T-category were independent predictors of BCR ([Table 3](#pone.0140671.t003){ref-type="table"}) while high GS, T-category and LN positivity predicted death from PCa ([Table 4](#pone.0140671.t004){ref-type="table"}). 10.1371/journal.pone.0140671.t003 ###### Uni- and multivariate Cox regression analysis for risk of biochemical recurrence. In addition to entire study cohort, a subpopulation of patients with GS\>6 tumor was analyzed separately. {#pone.0140671.t003g} Entire group Subgroup (GS\>6) --------------------------------------------------- ------------ -------------- ------------------ ---------------------------------------------- ------ ------------ ---------------------------------------------- ------ ------------ ---------------------------------------------- ------ ------------ ---------------------------------------------- Gleason score \<7 ref ref na na 7 1.9 1.2--3.0 \<0.003[\*](#t003fn002){ref-type="table-fn"} 1.3 0.8--2.3 0.32 ref ref \>7 4.8 3.0--7.5 \<0.001[\*](#t003fn002){ref-type="table-fn"} 2.6 1.5--4.7 0.001[\*](#t003fn002){ref-type="table-fn"} 2.4 1.7--3.5 \<0.001[\*](#t003fn002){ref-type="table-fn"} 2.0 1.0--3.3 \<0.002[\*](#t003fn002){ref-type="table-fn"} sPSA Continuous 1.06 1.05--1.08 \<0.001[\*](#t003fn002){ref-type="table-fn"} 1.04 1.02--1.06 \<0.001[\*](#t003fn002){ref-type="table-fn"} 1.05 1.04--1.07 \<0.001[\*](#t003fn002){ref-type="table-fn"} 1.04 1.02--1.06 \<0.001[\*](#t003fn002){ref-type="table-fn"} pT category[^a^](#t003fn003){ref-type="table-fn"} pT2 ref ref ref ref pT3 3.1 2.1--4.6 \<0.001[\*](#t003fn002){ref-type="table-fn"} 2.2 1.3--3.5 0.002[\*](#t003fn002){ref-type="table-fn"} 2.3 1.4--3.7 0.001[\*](#t003fn002){ref-type="table-fn"} 1.7 1.0--2.9 0.053 Lamin A Low ref ref ref ref High 1.0 0.7--1.3 0.85 0.81 0.5--1.2 0.29 0.9 0.6--1.3 0.58 0.7 0.5--1.1 0.15 Lamin B1 Low ref ref ref ref High 1.5 1.0--2.2 0.04[\*](#t003fn002){ref-type="table-fn"} 1.8 1.1--2.9 0.023[\*](#t003fn002){ref-type="table-fn"} 1.7 1.0--2.7 0.036[\*](#t003fn002){ref-type="table-fn"} 1.6 0.9--2.9 0.077 Lamin B2 Low ref ref ref ref High 1.1 0.8--1.6 0.48 0.95 0.6--1.4 0.81 0.95 0.6--1.4 0.82 0.85 0.5--1.4 0.49 Lamin C Low ref ref ref ref High 1.0 0.5--1.9 0.88 0.70 0.3--1.7 0.41 1.1 0.5--2.5 0.9 0.58 0.2--1.7 0.32 HR, hazard ratio (describes the relative risk of the event at any given time based on comparing the group with the reference group); 95% CI, 95% confidence interval (A 95% confidence interval means that 95% of the values of an unobservable parameter of interest would set in the interval if the experiment is repeated); na, not available; ref, reference \* Significant p-value ^a^ Pathological T category refers to the stage of primary tumor in the TNM tumor staging system. pT2 = Tumor confined within prostate, pT3 = Tumor invading extraprostatic tissue and/or seminal vesicle(s). 10.1371/journal.pone.0140671.t004 ###### Uni- and multivariate Cox regression analysis for risk of prostate cancer-specific death. In addition to entire study cohort, a subpopulation of patients with GS\>6 tumor was analyzed separately. {#pone.0140671.t004g} Entire group Subgroup (GS\>6) --------------------------------------------------- ------------ -------------- ------------------ -------------------------------------------- ------ ------------ -------------------------------------------- ------ ------------- ---------------------------------------------- ----- ----------- -------------------------------------------- Gleason score \<7 ref ref na na 7 8.8 1.0--71.0 0.042[\*](#t004fn002){ref-type="table-fn"} 9.3 1.1--77.4 0.039[\*](#t004fn002){ref-type="table-fn"} ref ref \>7 28.9 3.7--224.7 0.001[\*](#t004fn002){ref-type="table-fn"} 37.6 4.8--296.2 0.001[\*](#t004fn002){ref-type="table-fn"} 3.3 1.4--8.0 0.008[\*](#t004fn002){ref-type="table-fn"} 3.0 0.9--9.3 0.063 sPSA Continuous 1.80 0.99--3.24 0.051 1.02 0.98--1.07 0.31 pT category[^a^](#t004fn003){ref-type="table-fn"} T2 ref ref T3 60.3 1.4--2688.4 0.034[\*](#t004fn002){ref-type="table-fn"} 41.4 0.6--2642.6 0.079 N category[^b^](#t004fn004){ref-type="table-fn"} N0 ref ref N1 6.5 2.5--16.9 \<0.001[\*](#t004fn002){ref-type="table-fn"} 5.5 2.1--14.6 0.001[\*](#t004fn002){ref-type="table-fn"} Lamin A Low ref ref ref ref High 0.3 0.2--1.5 0.27 0.4 0.16--1.0 0.052 0.4 0.1--1.0 0.052 0.4 0.2--1.2 0.11 Lamin B1 Low ref ref ref ref High 1.5 0.4--5.1 0.55 1.6 0.5--5.7 0.44 1.4 0.4--5.1 0.57 1.7 0.5--6.1 0.40 Lamin B2 Low ref ref ref ref High 0.5 0.2--1.2 0.11 0.41 0.17--0.99 0.047[\*](#t004fn002){ref-type="table-fn"} 0.4 0.2--1.0 0.058 0.6 0.2--1.6 0.29 Lamin C Low ref ref ref ref High 0.4 0.2--1.2 0.095 0.2 0.1--0.6 0.004[\*](#t004fn002){ref-type="table-fn"} 0.3 0.1--0.8 0.023[\*](#t004fn002){ref-type="table-fn"} 0.3 0.1--0.9 0.03[\*](#t004fn002){ref-type="table-fn"} HR, hazard ratio (describes the relative risk of the event at any given time based on comparing the group with the reference group); 95% CI, 95% confidence interval (A 95% confidence interval means that 95% of the values of an unobservable parameter of interest would set in the interval if the experiment is repeated); na, not available; ref, reference \* Significant p-value ^a^ pathological tumor (pT) category refers to the stage (i.e. size and extent) of primary tumor in the TNM tumor staging system. pT2 = Tumor confined within prostate, pT3 = Tumor invading extraprostatic tissue and/or seminal vesicle(s). ^b^ N category describes in TNM system weather (N1) or not (N0) the cancer has spread to the regional lymph nodes. Staining with lamin A specific antibody showed that lamin A is expressed in both basal and luminal epithelial cells of normal prostatic glands and is enriched at the nuclear lamina region as expected ([Fig 1A](#pone.0140671.g001){ref-type="fig"}). However, there were significant differences in staining intensities between carcinomas from individual patients (from grade 0 to 3). The mean staining intensities of carcinoma cell nuclei were significantly higher when compared to normal epithelium in benign samples from the same patients (paired-samples t-test; mean staining intensities +/- SEM 1.97+/-0.03 and 1.45+/-0.03 for carcinoma and benign, respectively, p\<0.001). Low lamin A expression (grade 0--1) was statistically significantly associated with lymph node positivity (Pearson Chi-Square test, p = 0.009) but no other clinicopathological variables tested. In Kaplan-Meier estimation analysis, there was no correlation between lamin A expression and BCR (data not shown). When DSS was used as an end point, a trend between low lamin A expression and poor outcome was seen in the whole population but this difference failed to show statistical significance ([Fig 1B](#pone.0140671.g001){ref-type="fig"}; p = 0.27). However, in the subpopulation of GS \>6 tumors with potentially aggressive behavior, low lamin A expression was statistically significantly associated with poor outcome ([Fig 1C](#pone.0140671.g001){ref-type="fig"}; p = 0.044). These results suggest that diminished amount of lamin A may promote lymphovascular invasion and progression of potentially metastatic disease. {#pone.0140671.g001} Lamin C is another major A-type lamin encoded by *LMNA* through alternative splicing. This isoform is 92 amino acids shorter than lamin A and carries a unique C-terminal tail that is removed from pre-lamin A during posttranslational modification. The staining with lamin C antibody (detecting the lamin C specific C-terminal tail domain only) showed that lamin C is similarly expressed and present at the nuclear lamina in both normal basal and luminal epithelial cells as well as in the majority of (but not all) carcinomas with low to moderate intensities ([S1A Fig](#pone.0140671.s001){ref-type="supplementary-material"}). A paired T-test showed that the mean staining intensities for lamin C were significantly higher in carcinoma samples (1.30+/-0.06) relative to normal epithelium (1.04+/-0.05, p = 0.001). There was no correlation between lamin C expression and clinicopathological variables (data not shown). In the Kaplan-Meier analysis, we found no correlation between lamin C expression and BCR (data not shown) similar to lamin A. There was a trend between undetectably low lamin C expression (grade 0) and less favorable DSS when patients with all the different Gleason groups were studied ([S1B Fig](#pone.0140671.s001){ref-type="supplementary-material"}; p = 0.14). Moreover, in the GS \>6 subpopulation the patients with lamin C negative tumors (grade 0) had statistically significantly higher risk to die from PCa ([S1C Fig](#pone.0140671.s001){ref-type="supplementary-material"}; p = 0.009). These results suggest that lamin A and C, both produced from *LMNA* gene, may have similar biological effects in malignant transformation of PCa. The expression of lamin B1 is generally detected in all the eukaryotic cells and its loss has severe consequences during embryogenesis as highlighted with developmental abnormalities of lung and brain and early perinatal death in *LMNB1* knockout mice \[[@pone.0140671.ref022]\]. In our material, all the histologically benign samples showed mild to moderate staining intensity for lamin B1. All the carcinoma samples were lamin B1 positive as well ([Fig 2A](#pone.0140671.g002){ref-type="fig"}) but 65% of the carcinomas showed high expression (grade 3) of lamin B1. Paired t-test revealed that there was significantly more lamin B1 staining in carcinomas relative to normal benign samples (2.7+/-0.03 vs. 1.7+/-0.03, p\<0.001). No significant correlation was found between lamin B1 expression and clinicopathological variables tested. In the Kaplan-Meier analysis, high expression of lamin B1 was associated with BCR in the whole cohort ([Fig 2B](#pone.0140671.g002){ref-type="fig"}; p = 0.038) and in the subpopulation of GS \>6 (p = 0.034, data not shown). However, there was no correlation between lamin B1 expression and DSS in either the whole cohort ([Fig 2C](#pone.0140671.g002){ref-type="fig"}; p = 0.55) or in the subpopulation of patients with GS \>6 tumors (p = 0.57, data not shown). In multivariable Cox regression analysis high lamin B1 expression remained an independent predictor of BCR in the entire cohort when adjusted for preoperative PSA and Gleason grade (HR 1.8, 95% CI 1.1--2.9, p = 0.023; [Table 3](#pone.0140671.t003){ref-type="table"}). These results suggest that high expression of lamin B1, for unknown reasons, increases risk for PCa recurrence but has no significant effect on PCa survival. {#pone.0140671.g002} Lamin B2 was expressed in all the normal and malignant prostate glands ([Fig 3A](#pone.0140671.g003){ref-type="fig"}). However, similar to other lamins, there was statistically significantly higher IHC intensity of lamin B2 in the cancerous glands relative to normal benign glands (1.74+/-0.03 vs. 1.48+/-0.03, p\<0.001). Low lamin B2 expression was associated with lymph node positivity (Chi-square; p = 0.005) and there was a trend between low lamin B2 expression and T category (p = 0.076) and EPE (p = 0.098). In Kaplan-Meier analysis, there was no correlation between lamin B2 expression and BCR (data not shown). However, When DSS was used as an end point, there was a trend between low lamin B2 expression and poor DSS in the whole cohort ([Fig 3B](#pone.0140671.g003){ref-type="fig"}; p = 0.099) and a borderline statistical significance in GS \>6 subpopulation ([Fig 3C](#pone.0140671.g003){ref-type="fig"}; p = 0.051). These results show that lamin B1 and lamin B2 predict different outcome in PCa which may be due to differential roles in cancerous prostate cells. {#pone.0140671.g003} Discussion {#sec012} ========== The nuclear lamins are highly conserved among species and phylogenetic analysis suggests that they represent ancestors of whole IF protein family \[[@pone.0140671.ref023]\]. There is accumulating evidence that lamins not only support physical shape of the nucleus but also contribute several fundamental nuclear functions such as genome organization and transcription through interactions with chromatin and multiple transcription factors \[[@pone.0140671.ref007]\]. Dysfunction in these mechanisms may have deleterious effects *in vivo* as evidenced by various hereditary human diseases due to lamin mutations \[[@pone.0140671.ref009]\]. However, the role and alterations of lamins in malignant diseases are less well understood. In the present study we investigated the expression of different lamins in PCa and their potential use in estimation of patient outcome using an extensive TMA material and a detailed clinical patient data with a median follow-up time of more than 8 years. While the expression of all the major lamins was detected in prostatic epithelium, the statistical analysis revealed some interesting changes that are relevant in the light of recent *in vitro* findings about lamin function. Swift et al. have shown that the amount of lamin A and the ratio of A- and B-type lamins correlates with tissue stiffness (e.g. muscle has more lamin A than brain) to stabilize nucleus and chromatin from physical stress \[[@pone.0140671.ref024]\]. Furthermore, low levels of A-type lamins increase nuclear deformability and enhance cell migration through microfluidic constrictions *in vitro* \[[@pone.0140671.ref025]\] and *in vivo* in a mouse tumor model \[[@pone.0140671.ref026]\]. This is consistent with our results showing that reduced amounts of lamin A/C are associated with lymph node metastasis and unfavorable DSS, presumably due to increased lymphovascular invasion. The association between low lamin A expression and unfavourable DSS was statistically significant in Kaplan-Meier analysis carried out on patients with potentially aggressive GS\>6 disease ([Fig 1C](#pone.0140671.g001){ref-type="fig"}). However, in the multivariate Cox regression analysis, that also takes other clinicopathological variables into account, lamin A slightly failed to reach statistical significance (p = 0.052, [Table 4](#pone.0140671.t004){ref-type="table"}) suggesting that it is not an independent predictive marker different from lamin B1. The data from others suggest that lamin A is overexpressed in the invasive front in PCa and may increase local migration and invasion \[[@pone.0140671.ref016]\]. If this were the case, one would expect that high expression levels of lamin A/C correlate with BCR but we were unable to confirm such phenomenon. These differences may be, at least partly, due to variation in sampling. Despite the fact that we obtained three triplicate samples from the index tumor, these tissue cores represent overall tumor expression levels and were not intentionally targeted to areas of invasive front. The results from this and other studies suggest that the changes in lamin expression and their impact on tumor biology and clinical outcome may be complex and organ-specific. Previously the prognostic value of lamins in cancer progression has been established in colorectal cancer (CRC) where lamin A expression predicts poor outcome when compared to lamin A negative cancers \[[@pone.0140671.ref012]\]. However, little is known about the prognostic role of lamins in other solid tumors and our results suggest opposite role for lamin A in PCa progression ([Fig 1C](#pone.0140671.g001){ref-type="fig"}). A fundamental difference between these two types of adenocarcinomas is that 30% of CRCs were virtually devoid of lamin A \[[@pone.0140671.ref012]\] but at least some lamin A expression was detected in all PCas included in the present study. The results from *in vitro* studies suggest that silencing of lamin A may actually sensitize the cells for mechanical stress and reduce cell survival \[[@pone.0140671.ref026]\]. Therefore, presence of lamin A in CRC may mechanically protect the migrating cancer cells and reduce tumor cell death when compared to lamin A negative tumors while in the case of PCa moderately low amount of lamin A may simultaneously provide an ideal nuclear deformability and sufficient protection against mechanical stress in invading cells. A- and B-type lamins have differential roles in normal cells. While A-type lamins are mostly expressed in terminally differentiated cells, one or both of the B-type lamins are expressed in all animal cells. Mice lacking lamin B1, lamin B2 or both have defects in lung and brain development and these mice die at birth suggesting an essential developmental function \[[@pone.0140671.ref022]\]. Interestingly, increased production of one B-type lamin does not rescue the loss of the other B-type lamin indicating that lamin B1 and B2 have also differential cellular functions \[[@pone.0140671.ref027]\]. Similarly, our results suggest differential roles for lamin B1 and B2 in PCa progression. PCa tissues contained significantly more lamin B1 than benign tissues from the same patients. High lamin B1 expression associated with BCR in the entire cohort, as well as in the GS\>6 subpopulation ([Fig 2B](#pone.0140671.g002){ref-type="fig"}), and in multivariable Cox regression analysis lamin B1 appeared an independent prognostic factor for BCR ([Table 3](#pone.0140671.t003){ref-type="table"}). Nevertheless, there was no statistically significant correlation between lamin B1 expression and DSS ([Fig 2C](#pone.0140671.g002){ref-type="fig"} and [Table 4](#pone.0140671.t004){ref-type="table"}). Although our cohort is relatively large, it is important to notice that only 4.4% of patients died from PCa during the follow-up while 31% showed BCR. In most cases, the latter may be due to minimal local residual tumor that has no impact on DSS after salvage radiation therapy \[[@pone.0140671.ref028]\]. Therefore, high lamin B1 expression is more likely promoting local growth and/or proliferation, rather than enabling metastatic behavior through lymphovascular invasion. In support for this, a recent study has shown that moderate reduction of lamin B1 in cancer cells delays cell cycle progression \[[@pone.0140671.ref029]\]. Furthermore, silencing of lamin B1 in normal diploid WI-38 lung cells causes replicative senescence while overexpression increases the proliferation rate \[[@pone.0140671.ref030]\]. Coradeghini et al. have previously reported that lamin B1 expression correlates with GS and could be used as a biomarker in tumor differentiation and prognostics \[[@pone.0140671.ref017]\]. We were unable to confirm the correlation with GS but even more importantly our results clearly suggest that lamin B1 is an independent predictor that could be used to determine the risk of local growth and proliferation. Little is known about role of lamin B2 and its association with cancer. In the present study, low lamin B2 expression predicted LN positivity, similar to lamin A and C. Additionally, there was a trend between low expression of lamin B2 and less favorable DSS ([Fig 3B and 3C](#pone.0140671.g003){ref-type="fig"}) but little in common with lamin B1. Recently, Kuga et al. provided evidence that lamin B2 plays a role in mitotic spindle formation \[[@pone.0140671.ref031]\]. Furthermore, knockdown of lamin B2 resulted in chromosomal instability in CRC cells \[[@pone.0140671.ref031]\]. Although a more precise role of lamin B2 in spindle function still needs to be addressed it is tempting to speculate that PCa tumors with low amounts of lamin B2 are more prone to chromosomal instability, aneuploidy and eventually progression towards aggressive and potentially metastatic phenotype. In summary, our results suggest that lamins have differential roles in PCa progression and that they could be used as prognostic biomarkers in PCa diagnostics and risk stratification. For example, immunohistochemical lamin A staining from prostatic biopsy material might help in identifying the high risk patient who would benefit from lymph node dissection. However, the use semi-quantitative methods like immunohistochemistry in routine clinical diagnostic work may be challenging and would require sensitive and well validated control samples. Due to relatively high frequency of BCR and low frequency of mortality in PCa, further studies with even more extensive and multi-centered patient cohorts would be needed to validate the prognostic role of lamins in PCa. In addition, it would be of great interest to study the expression of A- and B-type lamins in PCa LN and distant metastases in the following studies. Supporting Information {#sec013} ====================== ###### Low lamin C expression in PCa predicts shortened DSS. \(A\) Representative examples of TMA slides stained for lamin C with immunohistochemistry. Low and high power field images from both low and high expressing tumors are shown. (B-C) Kaplan-Meier analysis shows a trend between low lamin C expression and shortened DSS in the entire cohort (B; p = 0.14). However, in the subpopulation of patients with Gleason score \>6 tumors, there is statistically significant difference (C; p = 0.009). (TIF) ###### Click here for additional data file. Sinikka Collanus is acknowledged for technical assistance with immunohistochemistry and Kalle Alanen, Miia Pavela and Elina Aaltonen for their help in constructing tissue microarray. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: IS TM PJB PT. Performed the experiments: IS PJB PT. Analyzed the data: IS HS PJB PT. Wrote the paper: IS TM HS PJB PT.

Introduction {#s001} ============ M[itochondria are essential]{.smallcaps} for neuronal health and synaptic function. Mitochondrial dysfunction is well-described during the minutes-to-days after a traumatic brain injury (TBI), but there is a critical knowledge gap concerning mitochondrial bioenergetic deficits beyond the acute period after a brain injury. Substantial evidence supports the involvement of mitochondria in the neurological sequelae of acquired brain injuries and neurodegenerative diseases.^[@B1],[@B2]^ Mitochondria isolated from pre-synaptic nerve terminals have been shown to be sensitive uniquely to the damaging effects of a TBI.^[@B3]^ A clinical history of TBI is known to increase the risk of neurodegenerative disease and dementia.^[@B4],[@B5]^ The commonalities of mitochondrial dysfunction after TBI and in neurodegenerative diseases suggest a potential mechanism whereby TBI induces a premature aging of mitochondria---particularly synaptic mitochondria. Using a mouse model of mild TBI, induced by a closed head injury (CHI), we found strong support for our hypothesis that a mild TBI would result in a premature aging of synaptic mitochondria. Specifically, we found a decrease in mitochondrial bioenergetics at one month post-injury, particularly in neuronal synaptic mitochondrial bioenergetics, which corresponded temporally with deficits in spatial memory as measured in the radial arm water maze. To extend the translational potential of our findings, we tested in our mouse mild TBI model two surrogate radiological measures for bioenergetics, which could be used in head-injured patients. The field of neuroimaging is far more mature in humans than in rodent models. Multi-modal magnetic resonance (MR) changes in humans after a mild TBI have been well-described,^[@B6]^ while there is a paucity of studies in mild TBI models in rodents. Technological improvements, such as high-field strength scanners (7--11 Tesla), are providing the spatial and/or temporal resolution necessary to image the rodent brain after a mild TBI. We sought to reverse translate two advanced neuroimaging modalities, which have shown utility clinically in mild TBI^[@B7]^ to a mouse model of mild TBI. The MR methods included pseudo-continuous arterial spin labeling (pCASL) to measure cerebral blood flow and proton magnetic resonance spectroscopy (^[@B1]^H-MRS). The current results demonstrate that ^[@B1]^H-MRS deficits in phosphocreatine, a surrogate radiological biomarker of mitochondrial dysfunction, occur at the same chronic time point where deficits in mitochondrial bioenergetics were confirmed in isolated mitochondria. Our results provide strong support for a novel mechanism of chronic mitochondrial deficits after TBI and a translationally relevant radiological method to monitor this mechanism. Methods {#s002} ======= Animals {#s003} ------- All procedures were approved by the Institutional Animal Care and Use Committee of the University of Kentucky, and that were conducted in accordance with the standards of proper experimentation in the Guide for the Care and Use of Laboratory Animals and ARRIVE guidelines. Experiments used four-month-old C57BL/6J mice and included an equal ratio of male and female for all end points. This project used a total of 204 mice: 36 mice were used for mitochondrial isolation; 19 mice were used for behavioral assays; 37 mice were used for pCASL and ^[@B1]^H-MRS experiment 1; 112 mice were used for ^[@B1]^H-MRS experiment 2; 22 of the 204 mice were not included in the data analysis for reasons described in the following method sections for each end point. The final number mice used for each end point are indicated in the figure or figure legend. Mice were randomized and assigned groups before the start of the experiment. The study was completed with multiple batches of mice using a block experimental design, with each batch including all experimental groups; the order of the group was randomized for each block, so that the order of sham and CHI surgery varied for each cage. Each cage of mice included more than one experimental group. Each subject was given a unique identification number, which does not identify the experimental group. The persons conducting the mitochondrial bioenergetics (HV), behavior (TM), and neuroimaging (DNL, VB) were blinded to the treatment conditions and were not involved on the surgery or postoperative care of the mice, which could potentially lead to unblinding of the experimental groups. CHI {#s004} --- The CHI was performed as described previously.^[@B12],[@B13]^ Briefly, mice were anesthetized with isoflurane (3--5%). The head was stabilized in a astereotaxic frame before a midline sagittal incision was made. A 1 mL latex pipette bulb filled with water was placed under the head. The stereotaxic electromagnetic impactor^[@B14]^ used a 5.0 mm flat steel tip (Leica Biosystems). It delivered a closed-skull midline impact (coordinates: mediolateral, 0.0 mm; anteroposterior, 1.5 mm) 1.0 mm deep with a controlled velocity (5.0 ± 0.2 m/sec) and a dwell time of (100 msec). No mice were eliminated because of surgery complications. Mice were randomized to a group (sham or injured), with each group containing approximately 50% males and females. Sham mice received the incision, but not the impact and were included for the three day post-injury and 28 day post-injury time points. Immunohistochemistry (IHC) {#s005} -------------------------- Mice were deep anesthetized with 5% isoflurane before transcardial perfusion with ice-cold phosphate buffered saline (PBS) for 5 min. The brains were rapidly removed, dissected, processed for IHC end points as described previously.^[@B12],[@B13]^ Primary antibodies used included: rabbit anti-βAPP (LifeTechnologies cat\# 51--2700 (1:2000)) and rabbit anti-glial fibrillary acidic protein (GFAP) (Dako Cat\# Z0334; (1:10,000). Aperio ScanScope XT digital slidescanner was used to image the entire stained slide at 20x magnification to create a single high-resolution digital image. Mitochondrial bioenergetics {#s006} --------------------------- Total mitochondria were isolated from the cortex and hippocampus as described previously.^[@B15],[@B16]^ Synaptic and nonsynaptic mitochondria were isolated from the cortex as described previously.^[@B3]^ Mitochondrial bioenergetics were measured using the Seahorse XF-24 Flux Analyzer as described previously.^[@B16]^ One mitochondria sample from the cortex was lost because of technical error. Six-arm radial arm water maze {#s007} ----------------------------- A modified four-day version of the previously reported radial arm water maze (RAWM)^[@B13],[@B17]^ was used as described previously.^[@B12],[@B13]^ The RAWM protocol consisted of four days of training, which occurred on nonconsecutive days (that is, day 14, 15, 16, and 21 post-injury). The total trials/day were 15, and each trial was up to 60 sec long. EthoVision (Noldus Information Technology) was used for video recording and scoring behavior. A memory retention 60 sec probe trial was performed seven days after the last training day (28 days post-injury). Time spent in the goal arm and the percentage of entries in the goal were recorded. MR imaging {#s008} ---------- MR imaging was performed on a 7-Tesla Clinscan scanner (Bruker, Billerica, MA) using a 2 × 2 Bruker brain surface coil or a Bruker quadrature, cryocoil at the University of Kentucky Magnetic Resonance Imaging and Spectroscopy Center, as described previously.^[@B18]^ Heart rate (90--130 bpm), respiration rate, and rectal temperature (37 ± 0.5°C) were monitored continuously. A water bath with circulating water at 45--50°C was used to maintain the body temperature. The order of mice scanned was randomized between sham and CHI groups for each acquisition period. T2-weighted structural images were acquired with field of view (FOV) = 18 × 18 mm^2^, matrix = 256 × 256; slice thickness = 1 mm, six slices, repetition time (TR) = 1500 msec, and echo time (TE) = 35 msec. Quantitative cerebral blood flow (CBF) (units of mL/g per min) was measured using MR based pCASL. A whole-body volume coil was used for transmission, and a mouse brain surface coil was placed on top of the head for receiving. Paired control and label images were acquired with a train of Hanning window-shaped radiofrequency pulses of duration and spacing 200/200 μsec and the following parameters: flip angle = 25 degrees, slice-selective gradient = 9mT/m, labeling duration = 2100 msec, echo spacing = 0.33 msec. A two-dimensional multi-slice spin-echo planner imaging sequence was used with a FOV: 18.0 × 18.0mm^2^, matrix = 128 × 128, slice thickness = 1.0 mm, six slices, TR = 4000 msec, and TE = 20 msec with 1200 repetitions.^[@B19]^ Manual shimming was applied after automatic shimming with the lowest possible frequency obtained before the image was acquired. Ten measurements were obtained before the usable 120 measurements as a test run to optimize the phase adjustment between two successive tagging radiofrequency pulses. ^[@B1]^H-MRS was acquired in the first setup experiments using the 2 × 2-surface coil, a single ^[@B1]^H-voxel of interest was defined as the bilateral dorsal hippocampus with a size of 12.48 mm^3^ (2.0 mm × 5.23 mm × 1.2 mm). Acquisition parameters for applied PRESS spectroscopy sequence include: TR = 1500 msec, TE = 135 msec, flip angle (FA) = 90 deg, rotation (R = 0 deg), spectral width (SW = 60 Hz), and averages (A = 400). Manual shimming was applied after automatic shimming on a shimming volume 30% larger than the spectroscopy voxel to maximize the suppression of outer-voxel water. A second acquisition with 10 averages and no water suppression was acquired and used to calculate metabolite concentrations relative to water. ^[@B1]^H-MRS was acquired in the second set of experiments using the cryocoil; acquisition parameters for the LASER spectroscopy sequence^[@B20]^ for the cryocoil include: TR = 2500 msec, TE = 21 msec, FA = 90, R = 0 deg, SW = 250, and A = 200. The LASER sequence was used for the second set of experiments to maximize the suppression of outer-voxel water and use a short TE. This maximizes signal to noise, allows for the measurement of metabolites not observable at higher TEs, and increases sensitivity to metabolite concentrations. The single ^[@B1]^H-voxel of interest was defined as the bilateral dorsal hippocampus. Manual shimming was applied after automatic shimming on a shimming volume 30% larger than the spectroscopy voxel to maximize the suppression of outer-voxel water. The lowest possible full width half maximum (FWHM) was obtained, and any spectra with a FWHM greater than 40 Hz as determined by the LCmodel processing software was eliminated. A second acquisition with 10 averages and no water suppression was acquired and used to calculate metabolite concentrations relative to water. pCASL data analysis {#s009} ------------------- MANGO software (version 3.8, UT Health, San Antonio, TX) was used to quantify the blood perfusion to the brain. pCASL images were overlaid on top of the T2-weighted structural images in MANGO software to visualize the regions of interest (ROI). Labeling efficiency (e.g., \>0.85) was used to verify the quality of the image series. Four mice were excluded because of a poorly acquired image. The pCASL images were analyzed using a custom, in-house developed MATLAB (Natick, MA) script, based on published equations.^[@B21]^ A single image slice was used for the CBF quantification with approximate coordinates of 1.5--1.7 mm anterposterior). ^[@B1]^H-MRS data analysis {#s010} -------------------------- The concentrations of the metabolites were derived from the Linear Combination model (LCModel) on a Linux operation system.^[@B22],[@B23]^ The concentrations of the metabolites (in μM) were computed using the following equation: $$\documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*} \left[ m \right] = \left( { { \frac { { S_m } } { { S_ { water } } } } } \right) \left[ { water } \right] { C_n } { C_ { av } } \end{align*} \end{document}$$ where \[*m*\] is the concentration of the specific metabolite, S~m~ is the metabolite intensity obtained from ^[@B1]^H-MRS, S~water~ is the water intensity, \[water\] is the water concentration (55.14 mM at 310 K), C~n~ is the correction for the number of equivalent nuclei for each resonance, and C~av~ is the correction for the number of average.^[@B18],[@B24]^ For each spectra, a signal to noise ratio was calculated by LCModel; the FWHM estimate of 0.049 was used as a cutoff for the cyrocoil ^[@B1]^H-MRS results. Nine mice were excluded because of a FWHM greater than 0.049. Statistical analysis {#s011} -------------------- JMP Software version 12.0 (SAS institute, Cary, NC) was used for statistical analysis. A repeated-measures analysis of variance (ANOVA) was used for RAWM. For all other end points, a one-way ANOVA was used comparing injury groups. If a significant main effect was found, *post hoc* analysis was used to compare groups. The normality assumption was assessed using the Shapiro-Wilk test, and heterogeneity of variances was assessed with the Levene test. Differences between mean were considered significant at α = 0.05. Graphs were generated using GraphPad Prism version 7.0. Values are expressed as mean ± standard error of the mean (SEM), unless otherwise noted. Scatter plots represent individual mice. No difference was found between the sham group three days post-surgery compared with the sham group at 28 days post-surgery; therefore, these groups were collapsed into a single sham group for all end points. Number of mice used for each end point are indicated in the figure or figure legend. Results {#s012} ======= At least 75% of the 1.7 million TBIs that occur in the United States each year are classified as a mild TBI.^[@B25],[@B26]^ In the present study, we used a CHI to model mild TBI. As reported previously,^[@B12],[@B13]^ and confirmed here ([Fig. 1](#f1){ref-type="fig"}), the CHI model does not cause gross structural lesions to the brain. At a one day post-injury time point, we find markers of axonal injury in the neocortex ([Fig. 1A, B](#f1){ref-type="fig"}). At 28 days post-CHI, we find diffuse astrogliosis ([Fig. 1C, D](#f1){ref-type="fig"}), with the highest GFAP staining corresponding to the same region of the brain with the greatest amyloid precursor protein (APP) staining at one day post-CHI. {#f1} C57Bl/6J mice show persistent deficits in mitochondrial bioenergetics after a mild TBI {#s013} -------------------------------------------------------------------------------------- Total mitochondria were isolated from the cortex and hippocampus, and State-III (adenosine triphosphate \[ATP\] synthesis capacity), State-V~C1~ (complex I driven maximal electron transport), and State-V~C2~ (complex II driven maximal electron transport) were measured using the Seahorse Extracellular Flux Analyzer. In the cortex, we found a stepwise decrease with time post-injury in both State-III and State-V~C-I~ mitochondrial respiration after CHI compared with the sham mice. No statistical significant difference in mitochondrial bioenergetics was found at three day post-CHI compared with sham-injured mice in the cortex ([Fig. 2A](#f2){ref-type="fig"}) or hippocampus ([Fig. 2B](#f2){ref-type="fig"}). By 28 days post-CHI, mice had a significant decrease in State-III and State-V~C-I~ mitochondrial respiration compared with the sham-injured mice ([Fig. 2A](#f2){ref-type="fig"}). In the hippocampus, we found a decrease in State-III, State-V~C-I~, and State-V~C-II~ mitochondrial respiration at 28 days post-CHI compared with the sham-injured mice ([Fig. 2B](#f2){ref-type="fig"}). {#f2} A brain-region heterogeneity was evident in the CHI-induced mitochondrial bioenergetics dysfunction between the hippocampus and cortex. In the hippocampus at 28 days post-CHI, the deficit in mitochondrial respiration compared with shams was 35% in State-III and 32% State-V~C-I~. In contrast, the cortex at 28 days post-CHI compared with shams was found to have a 16% deficit in State-III and 20% deficit in State-V~C-I~. It is unknown whether the regional heterogeneity in mitochondrial bioenergetics is a reflection of different sensitivity of the mitochondria in the brain regions to the effect of the injury, or differences in the degree of the primary injury in the two brain regions. After a CHI, the cortex but not the hippocampus has profound gliosis (microglia and astrocytes) and immune cells infiltrates (macrophages and neutrophils).^[@B12],[@B13]^ We hypothesized that the infiltration of immune cells and proliferation of glia in the cortex could mask some deficits in neuronal mitochondrial bioenergetics, compared with that seen in the hippocampus. To test this hypothesis, in a second cohort of mice, we performed differential isolation of neuronal presynaptic mitochondria and nonsynaptic mitochondria (neuronal cell bodies, glia, and infiltrating immune cells) in the cortex at 28 days post-CHI. In the synaptic mitochondria fraction, we found a 17% deficit in State-III and 31% deficit in State-V~C-I~ (Fig 2C). In the nonsynaptic, there was a 27% deficit in State-III, while State-V~C-II~ was unchanged from sham levels ([Fig 2D](#f2){ref-type="fig"}). Mild TBI causes impaired hippocampal-dependent learning and memory {#s014} ------------------------------------------------------------------ In light of the persistent suppression of hippocampal and neuronal mitochondrial bioenergetics after a CHI, we hypothesized that injured mice would have impairments in hippocampal-dependent learning and memory. To measure the cognitive status of the injured mice, we used the six-arm RAWM test of spatial learning and memory. The CHI-injured mice were found to have a delayed learning curve over the four days of training compared with the sham-injured mice ([Fig. 3A](#f3){ref-type="fig"}). Importantly, both groups, CHI and sham, were eventually able to find the location of the hidden platform while making less than two errors by the fourth day of training ([Fig. 3A](#f3){ref-type="fig"}). ](fig-3){#f3} Long-term memory consolidation was measured by using a probe trial, where the hidden platform was removed, and the amount of time and number of correct entries into the goal arm were considered an indicator of a memory of the platform location. The probe trial was conducted seven days after the last training session, on day 28 post-injury. As shown in [Figure 3B](#f3){ref-type="fig"}, CHI mice showed no preference for the goal arm, whereas sham-injured mice were found to have a preference for the goal arm. CHI mice spent less time in the target arm ([Fig. 3C](#f3){ref-type="fig"}) and made fewer entries into the target arm ([Fig. 3D](#f3){ref-type="fig"}) compared with sham-injured mice. Regional variations in CBF after CHI {#s015} ------------------------------------ Multimodal neuroimaging has the potential to be used as surrogate measures of mitochondrial bioenergetics. We sought to determine in a pilot experiment whether CBF and ^[@B1]^H-MRS would be sensitive biomarkers of metabolic disturbance after CHI in mice. We chose to measure CBF and ^[@B1]^H-MRS because these two MR modalities are commonly available clinically and could be completed in one scanning session. To measure CBF, we used the pCASL method. pCASL has higher signal to noise ratio and better labeling efficiency than other arterial spin labeling techniques. Also, pCASL allows for regional analysis to look for heterogeneity in CBF.^[@B21],[@B27]^ Examples of the CBF perfusion maps are shown in [Figure 4A](#f4){ref-type="fig"}. To quantify the regional heterogeneity in CBF after injury, we chose to analyze four distinct ROIs: (i) proximal cortex (closest to injury); (ii) adjacent cortex; (iii) hippocampus; and (iv) thalamus ([Fig. 4B](#f4){ref-type="fig"}). The ROIs were based on our previous studies, which defined the patterns of neuropathological changes in the CHI model.^[@B12],[@B13]^ ](fig-4){#f4} We observed subtle regional and temporal changes in the pattern of the CBF. Specifically, a 16.4% decrease in CBF was found in the three days post-CHI group in the region of the cortex most proximal to the site of injury compared with the sham and 28 days post-CHI group ([Fig. 4C](#f4){ref-type="fig"}). In the cortex adjacent to the site of injury, an increase in CBF was found in the three days post-CHI group (15.8%) and 28 days post-CHI group (21.4%) compared with the sham group ([Fig. 4C](#f4){ref-type="fig"}). Similar to the proximal cortex, a small decrease was found in CBF in the hippocampus in the three days post-CHI group (-9.4%) compared with the sham or 28 days post-CHI group ([Fig. 4C](#f4){ref-type="fig"}). The thalamus was found to have a slight increase in CBF in the 28 days post-CHI group (12.1%) compared with the sham or three days post-CHI group ([Fig. 4C](#f4){ref-type="fig"}). Apparent from the scatter plots in [Figure 4, a](#f4){ref-type="fig"} sizable amount of variability in CBF was seen within the groups. While patterns of CBF changes after a CHI are evident, none of the changes were found to be statistically significant. The amount of variability between the brain regions was relatively proportional to the size of the ROI, indicating that the smaller ROIs were pushing the limits of the scanner. A goal of the study was to determine the feasibility of using pCASL to detect CHI-induced changes in CBF. Results of *post hoc* power analysis showed that to detect an effect size of 10--20% change in CBF in the neocortex or hippocampus caused by a mild TBI would require an unrealistic number of mice per group (*n* = 144--56, respectively). The number of mice needed to detect a similar magnitude change in the thalamus was reasonable (*n* = 21). The results suggest that further improvements in the MR methods to increase the sensitivity of CBF in small ROIs may allow for MR to be used to detect changes in CBF after a mild brain injury in mice. A mild TBI decreases hippocampal metabolites associated with energy metabolism (PCr) and neuron function (NAA and tCho) {#s016} ----------------------------------------------------------------------------------------------------------------------- The second multi-modal neuroimaging sequence acquired in the same scanning session with pCASL was ^[@B1]^H-MRS in the bilateral dorsal hippocampus using a Bruker brain surface coil. Previous studies have demonstrated reliable detection of metabolites with a voxel of interest covering the bilateral dorsal hippocampus.^[@B18]^ While we anticipated a larger change in metabolites in the cortex immediately adjacent to the site of injury, this region was not selected as the ROI because the size and shape of the region would result in a voxel of interest that we believed would be too small to measure metabolites accurately. [Figure 5A](#f5){ref-type="fig"} shows a representative spectrum of the metabolites acquired by ^[@B1]^H-MRS in the dorsal hippocampus of a mouse. ![Closed head injury (CHI)-induced changes in hippocampal metabolites measured by brain surface coil ^[@B1]^H-MRS. (**A**) Representative example of the proton magnetic resonance spectroscopy (^[@B1]^H-MRS) spectrum acquired in the dorsal hippocampus of mice. A significant decrease was found in (**B**) phosphocreatine (PCr) (F~2,27~ = 4.63, *p* = 0.0195), (**C**) N-acetylaspartate (NAA) (F~2,27~ = 3.87, *p* = 0.0345), and (**D**) total choline (tCho) (glycerophosphorylcholine + phosphorylcholine) (F~2,27~ = 4.50, *p* = 0.0214) in CHI mice compared with sham-injured mice. *Post hoc* comparisons were made using a Dunnett test comparing all CHI groups with sham. Each symbol on the scatter plot represents a single animal. Sham (*n* = 13), three days post-CHI (*n* = 9), and 28 days post-CHI (*n* = 6).](fig-5){#f5} A total of five metabolites were acquired by ^[@B1]^H-MRS: phosphocreatine (PCr), N-acetylaspartate (NAA), glutamate (Glu) taurine (Tau) and total choline (tCho). Of the five metabolites, three metabolites were found to be significantly decreased after the injury. PCr was found to be reduced in the hippocampus at three days post-CHI (-26%) and 28 days post-CHI (-27.5%) compared with the sham-injured group ([Fig. 5B](#f5){ref-type="fig"}). NAA was reduced at three days post-CHI (-18.2%) compared with the sham-injured group ([Fig. 5C](#f5){ref-type="fig"}). tCho was found to be reduced in the hippocampus at three days post-CHI (-15.1%) and 28 days post-CHI (-10.1%) compared with the sham-injured group ([Fig. 5D](#f5){ref-type="fig"}). In comparison with the pCASL, power analysis on the pilot study of ^[@B1]^H-MRS determined that a reasonable sample size would be needed in future studies to detect CHI-induced changes in PCr, NAA, and tCHo by ^[@B1]^H-MRS in the dorsal hippocampus (*n* = 15, 7, 4; per group for PCr, NAA, tCho, respectively, for 20% effect size). To confirm the changes in hippocampal metabolites found in the pilot study, we generated a second cohort of sham and CHI mice to repeat the ^[@B1]^H-MRS. To increase the signal to noise, we used a Bruker cryocoil. The cryocoil provided a 2.5-fold increase in signal to noise, which is a 6.25-fold improvement in imaging time. In addition, we used an improved water suppression sequence called LASER,^[@B20]^ which allowed us to reduce the echo time while still suppressing the outer voxel water ([Fig. 6A](#f6){ref-type="fig"}). These improvements in cryocoil and LASER sequence increased the total number of metabolites detected from five to 18 ([Table 1](#T1){ref-type="table"}). ![Closed head injury (CHI)-induced changes in hippocampal metabolites measured by cryocoil ^[@B1]^H-MRS. (**A**) Representative example of the proton magnetic resonance spectroscopy (^[@B1]^H-MRS) spectrum acquired in the dorsal hippocampus of mice. A significant decrease was found in (**B**) phosphocreatine (PCr) (F~2102~ = 7.569, *p* = 0.0009), (**C**) N-acetylaspartate (NAA) (F~2,102~ = 6.294, *p* = 0.0027), and (**D**) total choline (tCho) (glycerophosphorylcholine + phosphorylcholine) (F~2,102~ = 5.281, *p* = 0.0066) in CHI mice compared with sham injured mice. *Post hoc* comparisons were made using a Tukey-Kramer test comparing all groups. Each symbol on the scatter plot represents a single animal. Sham (*n* = 54), three days post-CHI (*n* = 23), and 28 days post-CHI (*n* = 26).](fig-6){#f6} ###### [Hippocampal Metabolites Detected by Cryocoil ^1^H-MRS in the Hippocampus of Mice after Sham or Closed Head Injury Surgical Procedure]{.smallcaps}   *SHAM (mean ± SD μM)* n* = 54* *3D CHI (mean ± SD μM)* n* = 23* *28D CHI (mean ± SD μM)* n* = 26* ------- -------------------------------- ---------------------------------- ----------------------------------- Ala 1.097 ± 0.381 1.267 ± 0.455 1.073 ± 0.757 Cr 2.407 ± 0.51 2.206 ± 0.383 2.603 ± 0.89 PCr 4.975 ± 0.524 4.816 ± 0.898 4.326 ± 0.824 GABA 4.143 ± 0.598 3.753 ± 0.524 3.841 ± 0.884 Gln 3.527 ± 1.235 3.764 ± 1.375 3.593 ± 2.053 Glu 10.187 ± 1.089 9.788 ± 1.176 9.748 ± 1.583 GSH 1.295 ± 0.249 1.27 ± 0.266 1.296 ± 0.402 Ins 6.588 ± 0.847 6.127 ± 0.931 6.494 ± 1.161 Lac 3.018 ± 1.003 2.961 ± 0.836 2.415 ± 1.105 NAA 6.508 ± 0.54 6.125 ± 0.732 5.983 ± 0.841 NAAG 0.723 ± 0.369 0.658 ± 0.382 0.65 ± 0.345 Tau 9.986 ± 1.351 9.552 ± 1.534 9.629 ± 2.036 CrCH2 0.82 ± 0.398 0.962 ± 0.43 0.661 ± 0.354 tCho 1.469 ± 0.215 1.317 ± 0.153 1.508 ± 0.275 MM09 7.122 ± 1.132 6.989 ± 0.804 7.516 ± 1.456 Lip20 0.027 ± 0.127 0.053 ± 0.176 0.079 ± 0.227 MM20 8.128 ± 1.905 7.953 ± 2.762 7.947 ± 2.374 MM14 6.189 ± 2.401 5.86 ± 1.346 5.619 ± 1.64 CHI, closed head injury; SD, standard deviation; Ala, alanine; Cr, creatine; PCr, phosphocreatine; GABA, gamma-aminobutyric acid; Glu, glutamate; Gln, glutamine; GSH, glutathione; Oms, myoinositol; Lac, lactate; NAA, N-acetylaspartate; NAAG, N-acetylaspartylglutamate; tau, taurine; --CrCH2, creatine methylene group; tCHo, total choline; Lip, mobile lipids; and MM, macromolecules. The pattern of hippocampal metabolite changes seen with the cryocoil ^[@B1]^H-MRS was in agreement with the changes seen using the brain surface coil; however, the magnitude of the changes in metabolites was smaller, which likely reflects a decrease in variability associated with greater sample size and reduced variance from the 2.5 times more signal-to-noise using the cryocoil compared with the brain surface coil. We found that PCr was decreased 3% at three days post-CHI and 13% by 28 days post-CHI compared with sham levels ([Fig. 6B](#f6){ref-type="fig"}). NAA was decreased by 6% at three days post-CHI compared with sham levels ([Fig. 6C](#f6){ref-type="fig"}). In contrast to the brain surface coil where NAA levels had normalized by 28 days post-CHI, we found an 8% decrease in NAA at 28 days post-CHI compared with sham levels ([Fig. 6C](#f6){ref-type="fig"}). The most consistent changes between the brain surface coil and cryocoil ^[@B1]^H-MRS were found with tCho. A 10% decrease in tCho was found at three days post-CHI compared with sham levels ([Fig. 6D](#f6){ref-type="fig"}), with levels of tCho returning to sham values by 28 days post-CHI. Cryocoil ^[@B1]^H-MRS as a radiological biomarker of TBI-induced deficits in brain metabolism and neuronal function {#s017} ------------------------------------------------------------------------------------------------------------------- We next sought to determine the usefulness of cryocoil ^[@B1]^H-MRS as a surrogate indicator of deficits in brain metabolism and neuronal function. We chose to use two threshold values with varying degrees of stringency. We wanted to assess how accurately we could separate injured mice from sham mice while maintaining the robustness to identify injured mice with deficits in surrogate indicators of brain metabolism or neuronal function. One standard deviation below mean sham levels of the metabolites was used as our low stringency threshold. For high stringency, we used a threshold just below the lowest sham value. For PCr, the one standard deviation threshold was able identify 35--58% of the three days post-CHI and 28 days post-CHI mice, respectively; while only falsely identifying 13% of the sham-injured mice as having a deficit in PCr ([Fig. 7A](#f7){ref-type="fig"}). With the high stringency threshold, we eliminated false detection of PCr deficits, and still maintained 13--19% of three days post-CHI and 28 days post-CHI mice, respectively, identified with deficits in PCr ([Fig. 7B](#f7){ref-type="fig"}). NAA was able to identify 42--43% of the 28 days and three days injured mice, with a false positive rate of 19% in the sham-injured mice when a one standard deviation threshold was used ([Fig. 7C](#f7){ref-type="fig"}). The detection rate of brain-injured mice with deficits in NAA was reduced to 13--27% of three days and 28 days post-CHI mice, respectively, when all false positives were eliminated ([Fig. 7D](#f7){ref-type="fig"}). The discrimination potential of tCho was low, only identifying 27% of three days post-CHI mice while having a false positive rate of 14% in the sham-injured mice ([Fig. 7E](#f7){ref-type="fig"}). The discrimination potential of tCHO to identify brain-injured mice was less than 10% when all false positives were eliminated ([Fig. 7F](#f7){ref-type="fig"}). ![Discriminative potential of ^[@B1]^H-MRS as a surrogate indicator of deficits in brain metabolism and neuronal function. The percentage of mice with a PCr (phosphocreatine) value less than 4.44 μM (i.e., one standard deviation (SD) of sham mean levels (**A**) or below the lowest sham PCr value (3.57 μM) (**B**) is shown. The percentage of mice with a NAA value less than 5.97 μM (i.e., one SD of sham mean levels (**C**) or below the lowest sham NAA (N-acetylaspartate) value (3.57 μM) (**D**) is shown. The percentage of mice with a tCHo (total choline) value less than one SD of sham mean levels (1.26 μM) (**E**) or below less lowest sham tCho value (1.01 μM) (**F**) is shown. Sham (*n* = 54), three days post-closed head injury (CHI) (*n* = 23), and 28 days post-CHI (*n* = 26).](fig-7){#f7} Female mice show increased vulnerability to CHI-induced changes in hippocampal metabolites {#s018} ------------------------------------------------------------------------------------------ To assess sources of variability that contribute to changes in hippocampal metabolites, we stratified the data by sex ([Table 2](#T2){ref-type="table"}). In female mice, we found six metabolites that were showing a significant effect of CHI; however, in male mice, no statistically significant changes in hippocampal metabolites were observed. The metabolites that were found to be significantly affected by the CHI included PCr ([Fig. 8A](#f8){ref-type="fig"}), NAA ([Fig. 8B](#f8){ref-type="fig"}), and tCho ([Fig. 8C](#f8){ref-type="fig"}), which showed a similar injury-induced pattern of changes in both the female and male mice. In addition, three metabolites were found to be changed by the injury only in female mice, including gamma-aminobutyric acid (GABA) ([Fig. 8D](#f8){ref-type="fig"}), lactate ([Fig. 8E](#f8){ref-type="fig"}), and myoinositol ([Fig. 8F](#f8){ref-type="fig"}). ![Stratification by sex identifies closed head injury (CHI)-induced changes in six hippocampal metabolites measured by cryocoil ^[@B1]^H-MRS. A significant change was found in (**A**) phosphocreatine (PCr) (F~2,51~ = 7.354, *p* = 0.0016), (**B**) N-acetylaspartate (NAA) (F~2,51~ = 6.061, *p* = 0.0044), (**C**) total choline (tCho) (glycerophosphorylcholine + phosphorylcholine) (F~2,51~ = 5.881, *p* = 0.0051), (**D**) gamma-aminobutyric acid (GABA) (F~2,51~ = 6.061, *p* = 0.0044), (**E**) lactate (Lac) (F~2,51~ = 4.511, *p* = 0.0159), and (**F**) myoinositol (Ins) (F~2,51~ = 5.295, *p* = 0.0083) in female mice. No statistically significant changes were found in male mice. *Post hoc* comparisons were made using a Dunnett test comparing all CHI groups with sham. Each symbol on the scatter plot represents a single animal. Female sham (*n* = 27), female three days post-CHI (*n* = 11), and female 28 days post-CHI (**n** = 14). Male sham (*n* = 27), male three days post-CHI (*n* = 12), and male 28 days post-CHI (*n* = 12).](fig-8){#f8} ###### [Hippocampal Metabolites Detected by Cryocoil ^1^H-MRS in the Hippocampus in Female versus Male Mice after Sham or Closed Head Injury Surgical Procedure]{.smallcaps}   *♀ SHAM (mean ± SD μM)* n* = 27* *♀ 3D CHI (mean ± SD μM)* n* = 11* *♀ 28D CHI (mean ± SD μM)* n* = 14* *♂ SHAM (mean ± SD μM)* n* = 27* *♂ 3D CHI (mean ± SD μM)* n* = 12* *♂ 28D CHI (mean ± SD μM)* n* = 12* ------- ---------------------------------- ------------------------------------ ------------------------------------- ---------------------------------- ------------------------------------ ------------------------------------- Ala 0.88 ± 0.22 1.258 ± 0.558 1.146 ± 0.446 1.297 ± 1.069 1.276 ± 0.361 1.049 ± 0.303 Cr 2.4 ± 0.518 2.106 ± 0.41 2.299 ± 0.468 2.84 ± 1.17 2.298 ± 0.347 2.516 ± 0.535 PCr 4.154 ± 0.9 4.579 ± 0.86 5.047 ± 0.53 4.526 ± 0.71 5.033 ± 0.913 4.904 ± 0.517 GABA 3.607 ± 0.71 3.609 ± 0.455 4.159 ± 0.53 4.114 ± 1.014 3.886 ± 0.567 4.126 ± 0.668 Gln 4.069 ± 2.718 4.23 ± 1.578 3.52 ± 0.641 3.037 ± 0.47 3.337 ± 1.051 3.533 ± 1.643 Glu 9.386 ± 1.562 9.522 ± 1.155 10.288 ± 0.946 10.169 ± 1.565 10.032 ± 1.192 10.086 ± 1.225 GSH 1.166 ± 0.263 1.137 ± 0.24 1.241 ± 0.27 1.448 ± 0.488 1.392 ± 0.236 1.349 ± 0.217 Ins 6.174 ± 0.967 5.758 ± 0.797 6.662 ± 0.72 6.867 ± 1.296 6.465 ± 0.947 6.515 ± 0.965 Lac 2.36 ± 0.979 3.305 ± 0.728 3.313 ± 1.119 2.48 ± 1.278 2.645 ± 0.829 2.723 ± 0.786 NAA 5.842 ± 0.828 6.011 ± 0.668 6.578 ±H 0.531 6.148 ± 0.862 6.229 ± 0.8 6.438 ± 0.55 NAAG 0.637 ± 0.329 0.746 ± 0.398 0.797 ± 0.311 0.665 ± 0.377 0.578 ± 0.364 0.649 ± 0.411 Tau 8.818 ± 1.874 8.788 ± 1.426 9.72 ± 1.361 10.576 ± 1.86 10.253 ± 1.317 10.253 ± 1.312 CrCH2 0.671 ± 0.329 0.793 ± 0.344 0.811 ± 0.41 0.649 ± 0.396 1.117 ± 0.454 0.83 ± 0.393 tCho 1.438 ± 0.225 1.264 ± 0.138 1.486 ± 0.17 1.591 ± 0.313 1.365 ± 0.155 1.452 ± 0.255 MM09 7.14 ± 0.776 6.853 ± 1.061 7.155 ± 1.26 7.955 ± 1.927 7.113 ± 0.483 7.089 ± 1.011 Lip20 0.037 ± 0.133 0 ± 0 0.032 ± 0.156 0.128 ± 0.302 0.102 ± 0.237 0.022 ± 0.094 MM20 7.829 ± 2.693 8.78 ± 2.635 7.939 ± 1.789 8.086 ± 2.049 7.195 ± 2.762 8.317 ± 2.031 MM14 5.217 ± 1.11 5.938 ± 1.829 6.697 ± 3.108 6.088 ± 2.053 5.789 ± 0.755 5.681 ± 1.247 CHI, closed head injury; SD, standard deviation; Ala, alanine; Cr, creatine; PCr, phosphocreatine; GABA, gamma-aminobutyric acid; Glu, glutamate; Gln, glutamine; GSH, glutathione; Oms, myoinositol; Lac, lactate; NAA, N-acetylaspartate; NAAG, N-acetylaspartylglutamate; tau, taurine; --CrCH2, creatine methylene group; tCHo, total choline; Lip, mobile lipids; and MM, macromolecules. Discussion {#s019} ========== Two key findings from the studies are reported here. First, lasting deficits in mitochondrial-driven energy metabolism were found after a single, comparatively mild, diffuse brain injury. The injury-induced changes in mitochondrial metabolism were strongest in the hippocampus and synaptic nerve terminals. Unexpectedly, the deficits in mitochondrial bioenergetics worsened with time. TBI is one of the most well-described risk factors for the development of dementia.^[@B4],[@B5],[@B28]^ The chronic effects of TBI on mitochondrial bioenergetics could be part of the mechanism contributing to the enhance risk of dementia developing, potentially associated with a premature aging of mitochondria. The second relevant finding was the ability of MR spectroscopy to provide a way to monitor energy metabolism and neuronal health after a TBI noninvasively. Our results expand the utility of ^[@B1]^H-MRS beyond humans and large animals toward the use of ^[@B1]^H-MRS in mouse models of mild TBI. The ability to use the same efficacy biomarker in pre-clinical and clinical studies can provide a translational bridge in the important experiential medicine stage of drug development.^[@B29]^ A mild TBI induces changes to mitochondrial bioenergetics similar to those seen with aging {#s020} ------------------------------------------------------------------------------------------ The mechanisms by which a TBI hastens the onset of dementia are poorly defined. Aging is known to cause global decreases in brain metabolism, including both oxidative phosphorylation^[@B30]^ and aerobic glycolysis.^[@B31]^ Mitochondria are the primary source of oxygen radicals, and this underlies the mitochondrial free radical theory of aging.^[@B32]^ Mitochondrial enzymes are at risk of damage when the antioxidant buffering capacity does not match demand. Increases in mitochondrial oxidative damage are known to occur with age.^[@B33]^ Similarly, after TBI, elevated levels of intracellular Ca^2+^ are buffered by mitochondria, protecting the cell but potentially damaging mitochondria by the generation of oxidative stress.^[@B34]^ Mitochondrial oxidative damage after TBI preferentially targets pyruvate dehydrogenase, the gate-keeper of mitochondrial respiration and ATP production.^[@B35]^ Once pyruvate dehydrogenase is damaged, pyruvate cannot be converted into acetyl CoA, stopping the flow of electrons into the electron transport system by nicotinamide adenine dinucleotide and flavin adenine dinucleotide. Prior work has shown that TBI suppresses mitochondrial respiration up to 72 h after the brain injury.^[@B36]^ It has been assumed that with time after TBI, the damaged mitochondria would be eliminated by mitophagy, or the damaged mitochondria would lead to the death of the host cell. Our results contradict this assumption and suggest that in the case of a mild TBI, the deficits in mitochondrial bioenergetics propagate across time. Our finding suggests that TBI leads to changes in mitochondria akin to those seen with aging. Future studies are warranted to experimentally test whether long-term mitochondrial dysfunction after a mild TBI contributes to TBI-induced deficits in cognitive function and enhanced risk for dementia, and to determine whether therapeutic interventions targeting chronic mitochondrial dysfunction could reduce that risk. Hippocampal mitochondria are selectively vulnerable to the effects of a mild TBI {#s021} -------------------------------------------------------------------------------- Selective neuronal vulnerability is a hallmark of neurodegenerative diseases. Evidence suggests that neurons in general, and particularly neurons in the CA1 region of the hippocampus, have increased vulnerability to oxidative stress^[@B37]^ and have decreased ability to buffer Ca^2+^, which contributes to the greater neuronal mitochondria dysfunction after a TBI.^[@B3],[@B38],[@B39]^ In agreement with neuronal selective vulnerability, we found that a CHI resulted in a statistically significant suppression of mitochondrial respiration in the neuronal pre-synaptic mitochondria. Our previous results had suggested that the neocortex, most proximal to the direct forces of the CHI, was associated with the most pathology as determined by reactive gliosis.^[@B12],[@B13]^ The larger decrease in mitochondrial bioenergetics in the hippocampus compared with the neocortex, therefore, was unexpected. The decrease in the hippocampal energetic capacity did at least temporally correlate with hippocampal dependent learning and memory impairments in the brain-injured mice in agreement with previous studies from our group assessing mitochondrial function in the basal ganglia and its correlation to motor function both in TBI and aging.^[@B40],[@B41]^ We tested the possibility that gliosis in the cortex was masking the neuronal decrease in bioenergetics. However, our results do not support this argument because both synaptic and nonsynaptic mitochondria were found to have decreased respiration. There are two additional explanations for the selective vulnerability of the hippocampal mitochondria to TBI. The first possibility is that neurons in the hippocampus are more sensitive to the damaging effects of oxidative stress, and this resulted in increased damage to the mitochondria. The second possibility is that biomechanical forces of the CHI affected the hippocampus more than the cortex. Studies are ongoing to test these two possibilities. Head injuries are strongly correlated with cognitive deterioration and neurodegenerative dementia, including Alzheimer disease.^[@B42]^ It is assumed widely that the effects of a mild TBI, such as dizziness and headaches, are temporary. Evidence suggests that this is not the case, and even a mild TBI can result in progressive brain atrophy.^[@B43]^ All blows to the head do not lead to neurodegeneration; however, they do increase the likelihood that dementia may develop in the person\'s lifetime. Our work highlights a potential mechanism by which a mild TBI could lead to cognitive decline through a process of premature mitochondrial aging. Therefore, we sought to test translatable biomarkers that could be used to identify patients with mild TBI who are showing signs of TBI-induced mitochondrial aging. Advanced neuroimaging is useful at detecting changes in surrogate markers brain metabolism and neuronal health that are altered after a mild TBI {#s022} ------------------------------------------------------------------------------------------------------------------------------------------------ Recent positive acute stroke intervention clinical trials highlight the importance of using advanced neuroimaging-based selection criteria in the successful outcomes of the trials.^[@B44]^ The Stroke Imaging Research (STIR) group has identified that: (i) Neuroimaging can be used as an efficacy/biomarker or as an outcome assessment in clinical trials; (ii) no one single imaging approach will address all clinical requirements; (iii) use of imaging outcomes as a restrictive selection criterion may reduce the heterogeneity in the study population allowing for studies with smaller sample size requirements.^[@B45]^ The guidelines outline by the STIR group are exceedingly relevant for mild TBI, because the number of cases of mild TBI are numerous, but the percentage of cases that will have clinical difficulties associated with the mild TBI represent a small proportion of the total number of cases (15%).^[@B46],[@B47]^ A primary goal of our study was to reverse translate two advanced neuroimaging surrogate markers of brain metabolism, which have shown utility clinically in mild TBI ^[@B7]^ to a mouse model of mild TBI. We chose to use MR methods instead of positron emission tomography (PET) to measure brain metabolism, because of the wider availability of MR scanners in nonacademic clinical centers, and the advantage of not exposing persons to potentially toxic PET ligands. Our two primary MR methods used were pCASL to measure CBF and ^[@B1]^H-MRS to measure hippocampal metabolites. In agreement with previous human studies,^[@B48]^ we were able to detect changes in CBF in mice after a mild TBI. The pCASL sequence used in this study has greater specificity in analyzing CBF than other common methods.^[@B21],[@B27]^ We found substantial amounts of variability in all groups, however. Because we found that a TBI caused regionally based changes in CBF, with some areas showing hypoperfusion and adjacent areas showing hyperperfusion, it was necessary to use small ROIs. Despite the increased sensitivity of the 7-Tesla scanner and the pCASL sequence, we believe that the small ROIs used in the study were at least in part associated with the substantial variability in CBF seen between animals. In support of our assumption that the size of the ROI was at least partially associated with the increased variability, we found the lowest within group variance in the thalamus and hippocampus, which had the largest ROIs. Improved MR sequences and scanners hold the potential to overcome the spatial limitations in the ability to define regional variations in CBF in mice after a mild CHI. This is only a limitation for pre-clinical studies, however. While we only see a trend in our mouse models after TBI, the results are nevertheless promising that pCASL could be useful in humans or larger animal models of TBI to identify metabolic disturbances caused by mild TBI. The second neuroradiological method used to measure metabolic disturbances was ^[@B1]^H-MRS. In our pilot experiment, we identified three hippocampal metabolites that were significantly decreased from sham levels, namely PCr, NAA, and tCHo. In a second large cohort of mice, we used a more sensitive ^[@B1]^H-MRS method to validate these three metabolite changes and determine the potential of ^[@B1]^H-MRS to be a surrogate indicator of brain metabolism and neuronal health. PCr is directly associated with mitochondrial energy metabolism and is used as a source of energy to maintain ATP when levels of ATP are limited.^[@B52]^ In agreement with the decreases in mitochondrial respiration found in the hippocampus at 28 days after the brain injury, we also found decreases in PCr in the hippocampus at 28 days after injury by ^[@B1]^H-MRS. Our data are in accordance with a recent study in athletes who were exposed to a mild brain injury.^[@B53]^ We found that a significant proportion of the head-injured mice decreased PCr below all sham-injured mice, suggesting that PCr levels accurately define injured versus uninjured animals with metabolic disturbances. In addition, we found that NAA, a marker for neuronal viability, which has been observed to decrease in patients in the acute phases after mild TBI,^[@B8],\ [@B54]^ was also able to accurately define a subset of injured versus uninjured animals. In contrast, while we found a statistically significant decrease in tCho levels at three days post-injury, tCho values were not able to accurately separate injured from uninjured animals. In agreement with our findings, the peak decrease in tCho after a moderate TBI in rats occurred three days post-injury, after which point the levels of tCho began to rise.^[@B55]^ Changes in tCho have been reported in persons after a mild to moderate head injury; however, the levels of tCho were found to be increased after injury.^[@B54],[@B56]^ Similarly, we found that tCho were slightly elevated over sham values at 28 days post-injury. The results suggest a dynamic pattern of tCho values depending on the time after injury when measurements are made. The temporal changes and limited diagnostic potential limits the usefulness of tCho as an efficacy biomarker of intervention studies. The changes in hippocampal metabolites reported in the current study are in contrast to those of a recent study in a rat model of mild TBI, which failed to find differences in cortical metabolites after the injury.^[@B57]^ Differences between the two studies are numerous and include species, TBI model, brain region, sample size, and spectroscopy method. We believe that the higher field strength scanner, improved water suppression sequence,^[@B20]^ and the larger sample size used in the current study were all important for our ability to detect changes in brain metabolites after the mild TBI. Limitations of the current study {#s023} -------------------------------- Our studies included an equal proportion of female and male mice. Most of our end points were not sufficiently statistically powered to be disaggregated and reported by sex. The sample size available for the ^[@B1]^H-MRS study was adequate to stratify the data by sex for exploratory analysis. For PCr, NAA, and tCho, we found a similar pattern of CHI-induced changes between the female and male mice; however, the CHI-induced changes in the metabolites did not reach statistical significance in the male mice, likely because of decreased statistical power with the smaller sample size. Interestingly, the exploratory analysis found that GABA, lactate, and myoinositol were decreased by the CHI in the female mice only. Future studies will be needed to determine how sexually dimorphic characteristics, chromosomal differences, and sex hormone levels may contribute to the sex-specific effects on GABA, lactate, and myoinositol. Multi-modal MR is noninvasive, but in mice, it does require extended periods of anesthesia, with a range of 30--90 min, depending whether one or two sequences are acquired. Using isoflurane, we were able to regulate the depth of the anesthesia to maintain respiration. In addition, each imaging session required the mice to be under anesthesia. Studies have shown that isoflurane can be protective after injury.^[@B58]^ Other studies have also demonstrated that prolonged isoflurane can impact CBF and mitochondrial function.^[@B59]^ There is an advantage to repetitive scans in the same mouse, including establishing a pre-injury baseline. We do not know the effect that multiple rounds of extended anesthesia have on the recovery of TBI and brain metabolite levels. Therefore, we did not attempt to establish the time course of MR changes in the same mouse. For these same reasons, we used separate cohorts of mice for our mitochondrial bioenergetics measurements. Conclusion {#s024} ========== In this report, we find that a mild TBI results in a mitochondrial bioenergetics response that is akin to accelerated aging. We found a regional heterogeneity in mitochondrial bioenergetics that suggests selective vulnerability of neurons in different regions of the brain to the effect of TBI. Finally, we validated ^[@B1]^H-MRS as a surrogate indicator of bioenergetic and neuronal health after a mild TBI. Acknowledgments {#s025} =============== The corresponding author, Adam Bachstetter, PhD, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. DNL was supported in part by a Kentucky Spinal and Head Injury Trust trainee fellowship. Research reported in this publication was supported by National Institutes of Health under award numbers R00 AG044445 (ADB) and P30 GM110787 (ADB), and pilot grant funding from the University of Kentucky (ADB). This work was supported in part by a Merit Review Award \# I01BX003405 to (PGS) from the United States Department of Veterans Affairs Biomedical Laboratory Research and Development Program. The content is solely the responsibility of the authors and does not represent the official views of the funding organizations. Author Disclosure Statement {#s026} =========================== No competing financial interests exist.

{#sp1 .247}

{#sp1 .130}

1. Introduction {#sec1} =============== Nanostructured electrodes with gold nanoparticles (AuNPs) have demonstrated unique properties in the development of nucleic acid electrochemical biosensors with improved analytical performances. The catalytic activity of the AuNPs together with their large surface area and excellent biocompatibility make these nanostructured surfaces more conductive and allow the immobilization of large probe loadings with optimal orientation and spacing for efficient hybridization, keeping their biological activity and storage stability.^[@ref1]−[@ref4]^ The resulting bioplatforms exhibit also interesting advantages in terms of simplicity, cost, assay time, and applicability in different environments. These characteristics make AuNPs-involved bioplatforms especially attractive compared to conventional methodologies for their implementation in routine determinations. These properties have been successfully harnessed to develop electrochemical platforms for the determination of biomarkers of high clinical relevance, such as miRNAs,^[@ref5],[@ref6]^ a class of endogenous and small noncoding RNAs that play a relevant role in many diseases, particularly in human cancers.^[@ref7],[@ref8]^ Alterations in miRNA expression levels are involved in the initiation and progression of human cancers, and minimal signature profiles have been identified for various types of human cancers, where miRNAs can be regulated both upward and downward. Therefore, the determination of regulatory patterns is of great importance since they can provide relevant information about tumor initiation, invasion, and metastasis processes. However, the determination of miRNAs is highly complex due to their short sequence, low concentration level, and high sequence similarity between members of the same family. Conventional available methods for miRNAs determination, such as northern blotting,^[@ref9],[@ref10]^ in situ hybridization, reverse transcription polymerase chain reaction,^[@ref11]−[@ref13]^ and miRNA microarrays,^[@ref14]−[@ref17]^ suffer from important limitations (use in centralized laboratories, expensive, laborious, require specialized personnel, and poorly portable instrumentation), which make the development of more simple and rapid alternative strategies highly desirable. In recent years, several electrochemical platforms have been reported for the single or simultaneous determination of miRNAs either using integrated formats^[@ref4],[@ref18]−[@ref22]^ or involving magnetic beads (MBs).^[@ref23]−[@ref27]^ However, many of these biosensors require amplification strategies, which demand complex and/or long protocols to achieve the required sensitivity. This is an important restriction for their wide use in clinical practice by unskilled personnel. In this context, particularly attractive and relatively simple electrochemical strategies have been reported recently with acceptable sensitivity using commercial capture and detector bioreceptors with high affinity toward RNA homohybrids (viral protein p19^[@ref4],[@ref28]−[@ref31]^) and heterohybrids (anti-DNA--RNA hybrid antibodies^[@ref19]−[@ref21],[@ref23],[@ref25],[@ref26]^). This paper describes a novel electrochemical bioplatform for miRNAs determination, which combines the simplicity of a direct DNA--RNA hybridization format with the advantages of disposable AuNPs-nanostructured platforms and the high specificity and affinity of anti-DNA--RNA hybrid antibodies. To achieve high sensitivity, an easily implementable amplification strategy using a multienzyme bacterial protein able to recognize the Fc region of the anti-DNA--RNA hybrid antibody is used. The hybridization reaction is monitored amperometrically with the H~2~O~2~/hydroquinone (HQ) system. The resulting biosensor exhibits interesting performance in terms of sensitivity, selectivity, and reduced fabrication and assay times. It was successfully applied for the reliable and accurate determination of the selected target miRNA-21 (a relevant onco-miRNA^[@ref32]^) directly in a five-times diluted human blood serum from breast cancer patients as well as in breast cancer cells using the lowest amount of total extracted RNA reported to date for an electrochemical biosensor. 2. Results and Discussion {#sec2} ========================= The developed strategy is displayed in [Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}. It involves direct RNA--DNA hybridization implemented on disposable carbon surfaces nanostructured with AuNPs. The target miRNA was selectively captured on the nanostructured surfaces by efficient hybridization with a complementary thiolated DNA probe. The resulting heterohybrid was recognized with an antibody with high affinity for these heteroduplexes, which was labeled with a bacterial binding antibody protein (ProtA) conjugated with a homopolymer containing 40 horseradish peroxidase (HRP) units. The extent of the hybridization reaction and thus of the miRNA-21 concentration was monitored by amperometry in the presence of the H~2~O~2~/HQ system. Interestingly, both the nanostructured disposable surfaces and all necessary reagents involved in this strategy are commercially available. {#sch1} [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"} shows the comparison of the amperometric traces recorded in the absence and presence of the target miRNA at unmodified AuNPs/SPCEs and modified with HS-DNA and HS-RNA capture probes. Negligible and similar amperometric responses were observed in the absence of miRNA-21 on HS-DNA/AuNPs/SPCEs and in the presence of 20 pM miRNA-21 on AuNPs/SPCEs with no immobilized HS-DNA capture probe immobilized and on complementary HS-RNA/AuNPs/SPCEs ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}b). These results confirmed that the amperometric responses were due to the efficient hybridization of the HS-DNA capture probe with the target miRNA, the negligible nonspecific adsorption of anti-DNA--RNA hybrid antibody and ProtA-polyHRP40 on the AuNPs/SPCEs surface, and the specificity of the antibody for the selective detection of DNA--RNA heterohybrids.^[@ref33]−[@ref35]^ {#fig1} 2.1. Optimization of the Experimental Variables {#sec2.1} ----------------------------------------------- The most relevant experimental variables affecting the behavior of the designed bioelectrode were optimized. The current measured in the presence of 20 pM (signal, S) and in the absence (noise, N) of the target miRNA-21 as well as the corresponding ratio (S/N ratio) were used as selection criteria. The results obtained are shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"} and summarized in [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}. Apart from these, other variables used involved in the modification of AuNPs/SPCE, such as the HS-DNA probe concentration, the hybridization time with the target miRNA,^[@ref4],[@ref22]^ the pH value, the detection potential, and the concentrations of H~2~O~2~ and HQ, were taken from previous works.^[@ref36]−[@ref38]^ {#fig2} ###### Evaluation of Different Experimental Variables Involved in the Preparation and Functioning of the Developed Biosensor variable evaluated range optimal value selected ------------------------------------------------------------------------------- ----------------- ------------------------ anti-DNA--RNA hybrid antibody (dilution) 1/5000--1/250 1/1000 ProtA-polyHRP40 (dilution) 1/500--1/5 1/50 incubation time with mixture DNA--RNA hybrid antibody + ProtA-polyHRP40 (min) 0--90 60 number of steps[a](#t1fn1){ref-type="table-fn"} 1--3 2 Starting from the preparation of HS-DNA/AuNPs/SPCEs. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a,b shows the effect of the concentration of bioreagents used to carry out the heterohybrid recognition and the enzymatic labeling on the biosensor response. The 1/1000 and 1/50 dilutions of the anti-DNA--RNA hybrid antibody and ProtA-polyHRP40 conjugate provided larger S/N ratios, and therefore, they were selected for further work. In an attempt to simplify the whole working protocol and reduce the assay time, the influence of the number of steps involved in the preparation of the biosensor on its analytical behavior was checked; 30 min incubation steps were assayed in all cases starting from the preparation of HS-DNA/AuNPs/SPCEs. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c shows a comparison between the results obtained with the following working protocols:One single incubation step with a mixture solution consisting of the target miRNA, anti-DNA--RNA antibody, and ProtA-polyHRP40.Two successive incubation steps: first with the target miRNA and thereafter with a mixture solution containing anti-DNA--RNA antibody and ProtA-polyHRP40.Three successive incubation steps with the target miRNA, anti-DNA--RNA antibody, and ProtA-polyHRP40. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}c shows clearly that the protocol involving two successive incubation steps provided a larger S/N current ratio. These results suggest that the mixing of all of the bioreagents worsened the efficiency of the hybridization process, whereas the anti-DNA--RNA antibody was better recognized by ProtA-polyHRP40 when both reagents were incubated in homogeneous solution. Once the optimal working protocol was selected, the influence of the incubation time with the mixture of anti-DNA--RNA antibody and ProtA-polyHRP40 to label the DNA/miRNA heteroduplex captured on the electrode was evaluated. As it is shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}d, there were no significant differences for the nonspecific responses, but the specific current increased notably with the incubation time up to 60 min. Accordingly, this incubation time ensured an efficient labeling of the immobilized heterohybrids and was selected for further studies. 2.2. Analytical Characteristics {#sec2.2} ------------------------------- [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"} shows the amperometric responses obtained for different concentrations of the synthetic target miRNA as well as the corresponding calibration plot. As expected for a direct hybridization assay, the Δ*i* values (measured as described in [Section [4.5](#sec4.5){ref-type="other"}](#sec4.5){ref-type="other"}) increased linearly (correlation factor *r*: 0.998) with the miRNA-21 concentration from 0.096 to 25 pM, with a slope value of 378 ± 7 nA pM^--1^. The limit of detection (LOD), defined as the concentration of the analyte that gives a signal that is significantly different from the blank, was estimated from the signal of the blank plus 3 times the standard deviation of the signal values for 10 independent amperometric measurements obtained without target miRNA. The resulting LOD was 29 fM (0.29 amol in 10 μL sample). {#fig3} A comparison of the analytical performance of the developed biosensor with that of other electrochemical biosensors reported for the determination of miRNAs using also DNA--RNA heterohybrid-specific antibodies is shown in [Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}. The LOD value provided by the developed biosensor is notably better than that achieved with biosensors involving the use of magnetic microbeads^[@ref23],[@ref26]^ and similar to that claimed using electrodes nanostructured with carbon nanomaterials.^[@ref19],[@ref20]^ Importantly, no application is reported for this latter approach. The achieved LOD is also higher than that given for the sensor using AuNP-modified glassy carbon electrode (GCE) and applied for the determination in RNA~t~ extracted from rice seedlings.^[@ref21]^ However, both the fabrication and assay time of the developed sensor are remarkably shorter than those required using these more sensitive strategies ([Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}). ###### Analytical Characteristics of Electrochemical Biosensors Reported so far for miRNAs Determination Using DNA--RNA Hybrid Antibodies electrode role of DNA--RNA hybrid antibody electrochemical technique target miRNA I.L. LOD selectivity against 1-m sequences (%)[b](#t2fn2){ref-type="table-fn"} biosensor or MBs fabrication/assay time[a](#t2fn1){ref-type="table-fn"} sample (amount) ref ------------------------------------------------- --------------------------------------------------------------------------- --------------------------- --------------------------- ------------------------ -------------------------- ----------------------------------------------------------------------- ------------------------------------------------------------------------- ------------------------------------------------------------------------------- ------------ carbon paper/reduced graphene oxide (RGO)/GCE detector SWV miRNA-29b-1 and miRNA-141 1--1 × 10^6^ fM 8 fM   12 h/2 h 30 min   ([@ref19]) oxidized multiwalled carbon nanotubes/RGO/GSPEs detector SWV miRNA-29b-1 and miRNA-141 10--1 × 10^6^ fM 10 fM   12 h/2 h 30 min   ([@ref20]) SPCE capture bioreceptor immobilized onto ProtG-MBs amperometry (H~2~O~2~/HQ) miRNA-21 and miRNA-205 (8.2--250) × 10^3^ fM 2.4 × 10^3^ fM (60 amol) 51 45 min/2 h RNA~t~ extracted from cancer cells (500 ng) and human tumor tissues (1000 ng) ([@ref23]) SPCE detector bioreceptor for DNA--RNA heterohybrid immobilized onto Strep-MBs amperometry (H~2~O~2~/HQ) miRNA-21 (1.0--100) × 10^3^ fM 0.4 × 10^3^ fM (10 amol) 48 30 min/1.5 h RNA~t~ extracted from cancer cells (250 ng) and human tumor tissues (250 ng) ([@ref26]) AuNPs/GCE detector DPV miRNA-319a 0.5--500 fM 0.40 fM 14 13 h/5 h 10 min RNA~t~ extracted from rice seedlings ([@ref21]) AuNPs/SPCE detector bioreceptor amperometry (H~2~O~2~/HQ) miRNA-21 (0.096--25) × 10^3^ fM 29 fM (0.29 amol) 30 9 h/1 h 30 min RNA~t~ extracted from cancer cells (50 ng) this work Once modified the electrode or the MBs with the capture bioreceptor. \% of the response obtained in comparison to that of the target miRNA. Interestingly, the designed bioscaffold provides about 7-times enhanced sensitivity (55.3 vs 378 nA pM) and a 34-times lower LOD value (0.29 vs 10 amol) than the performance exhibited by a magnetic microbeads-based biosensor involving the same bioreagents.^[@ref26]^ The improved performance can be attributed to the inherent conductivity and catalytic activity of AuNPs and the enhanced immobilization of nucleic acid probe with an adequate orientation and spacing to improve the target accessibility at the nanostructured surfaces.^[@ref1]−[@ref4]^ Indeed, the LOD achieved with this biosensor is rather similar to the values claimed for other biosensors reported recently for the amperometric determination of miRNAs at AuNPs/SPCEs using competitive and direct RNA/RNA hybridization assays using Strep-HRP and the viral protein p19, respectively.^[@ref4]−[@ref22]^ Importantly, as shown below, the sensitivity achieved is sufficient to allow quantification using a significantly small amount of extracted RNA~t~. Moreover, the higher stability of DNA probes compared to RNA probes can be considered as another relevant practical advantage of the developed methodology.^[@ref39]^ The reproducibility of the measurements carried out with the developed bioplatform was checked by comparing the currents obtained for 1 pM miRNA-21 with nine different biosensors prepared in the same manner. A relative standard deviation value of 2.9% was calculated to show an excellent reproducibility in both the biosensor preparation and amperometric measurement procedures. Regarding the storage stability of the bioplatforms, the currents measured with HS-DNA/AuNPs/SPCEs, which were stored at 4 °C in a humidified chamber (after incubation with 1% (w/v) bovine serum albumin (BSA)), for 0 and 1.0 pM miRNA-21 showed that they remained within the set control limits over a 2 month period (no longer times were tested) ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}). This attractive storage stability, in good agreement with the behavior observed for bioelectrodes involving thiolated RNA probes at disposable AuNP-nanostructured scaffolds,^[@ref4],[@ref22]^ is attributed to the great biocompatibility and major role played by AuNPs in keeping the activity of the biomolecules immobilized onto their surface. {#fig4} 2.3. Selectivity {#sec2.3} ---------------- Due to the high sequence homology between miRNA family members, a very important feature of miRNAs biosensing platforms is the selectivity toward other nontarget and mismatched miRNA sequences. [Figure [5](#fig5){ref-type="fig"}](#fig5){ref-type="fig"} displays amperometric traces provided by the HS-DNA/AuNPs/SPCEs in the absence and presence of the synthetic target miRNA, a single-central base mismatched (1-m) and a fully noncomplementary miRNA (miRNA-451). A very small response similar to that measured without target miRNA was obtained for the noncomplementary sequence. The 1-m sequence provided a 30% response of that obtained for the target miRNA. These findings are of particular interest for the applicability of the method in challenging samples, such as raw cellular RNA~t~, where the target miRNA is present with many other nontarget miRNAs. The selectivity achieved toward 1-m sequences (30% hybridization efficiency assigning 100% for the target miRNA) is better than that demonstrated for other electrochemical biosensors using anti-RNA--DNA hybrid antibodies (48^[@ref26]^ and 51%^[@ref23]^) requiring similar assay times, while it is slightly worse (30 vs 14%) than that claimed for the strategy requiring an assay time 3 times longer.^[@ref21]^ {#fig5} 2.4. Determination of the Endogenous Content of miRNA-21 in Serum from Breast Cancer Patients and RNA~t~ Extracted from Breast Cancer Cells {#sec2.4} ------------------------------------------------------------------------------------------------------------------------------------------- The practical suitability of the developed methodology was tested by determining the endogenous content of miRNA-21 directly in blood serum from breast cancer patients and in raw RNA~t~ extracted from breast cancer cells. [Figure [6](#fig6){ref-type="fig"}](#fig6){ref-type="fig"}a shows the amperometric traces recorded with the developed biosensors for the three serum samples analyzed (one from a healthy individual and two from patients diagnosed with breast cancer). The samples were five-times diluted and heated, as described in [Section [4.6](#sec4.6){ref-type="other"}](#sec4.6){ref-type="other"}. These results are in agreement with those reported by other authors^[@ref40]^ showing that miRNA-21 expression is about 3--4 times larger in breast cancer patients that in healthy individuals using an indirect highly liquid chromatography--mass spectrometry (MS)/MS method.^[@ref41]^ No significant differences were found for the slope values of the calibration plot constructed with miRNA-21 standards ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}, 378 ± 7 nA pM^--1^) and of the calibration graphs constructed for five-times diluted and heated sera from the healthy individual (374 ± 8 nA pM^--1^) and the breast cancer patients 353 ± 8 nA pM^--1^ spiked with growing amounts of synthetic miRNA-21 up to 20 pM. This made the determination of the endogenous concentration of the target miRNA in serum feasible by simple interpolation of the Δ*i* values measured for the diluted samples into the miRNA standards calibration plot. The concentrations of miRNA-21 obtained were 2.0 ± 0.2, 8.0 ± 0.3, and 7.3 ± 0.4 pM in the serum samples from the healthy individual and breast cancer patients 1 and 2, respectively. {#fig6} Regarding the determination of miRNA-21 in cellular extracted RNA~t~, [Figure [6](#fig6){ref-type="fig"}](#fig6){ref-type="fig"}b shows the amperometric traces provided by the HS-DNA/AuNPs/SPCEs showing the expected miRNA-21 3--4 times higher expression in the tumorigenic (MCF-7) compared to the nontumorigenic (MCF-10A) cells according to the oncogenic role of this miRNA.^[@ref41],[@ref42]^ Similar to that observed in diluted and heated serum samples, no statistically significant differences were found between the slope values of the corresponding calibration plots constructed with the synthetic target miRNA-21 both in buffered solutions and in the presence of 10 ng of RNA~t~ extracted from MCF-10A cells 354 ± 9 nA pM^--1^, thus confirming the absence of apparent matrix effect in the cellular RNA~t~ analysis. Interpolation in the miRNA-21 standard calibration plot provided concentrations of 0.82 ± 1.05 and 2.97 ± 3.22 amol ng^--1^ RNA~t~ in MCF-10A and MCF-7 cells, respectively. These results are in complete agreement with those reported previously with concentrations ranging from 0.8 to 1.0 amol ng^--1^ RNA~t~ in MCF-10A and from 2.3 to 3.3 amol ng^--1^ RNA~t~ in MCF-7 cells.^[@ref4],[@ref23],[@ref43],[@ref26]^ These results demonstrate the bioplatform usefulness for the accurate and selective determination of the target miRNA in the presence of many other nontarget miRNAs. Importantly, this determination can be accomplished using the lowest amount of raw RNA~t~ reported so far with electrochemical biosensors,^[@ref4],[@ref24],[@ref26],[@ref27],[@ref29],[@ref30],[@ref44]^ which required between 50 and 1000 ng of RNA~t~ extracted from cells to perform the target miRNA determination. 3. Conclusions {#sec3} ============== This work reports a novel amperometric bioplatform to determine miRNA-21. Combining the inherent advantages of disposable surfaces nanostructured with gold nanoparticles with the use of selective antibodies to DNA--RNA heteroduplexes and bacterial immunoglobulin-binding protein conjugated to multiple HRP units made this determination possible with an LOD as low as 29 fM and good discrimination toward sequences with a single unpaired base. Interestingly, this bioplatform is able to accurately determine the endogenous content of miRNA-21 directly in human blood serum from breast cancer patients and in breast cancer cells using very small amount of RNA~t~. It should be emphasized that no target miRNA amplification steps are required with this strategy. In addition, all of the reagents and disposable electrodes used are commercially available, making this bioscaffold particularly attractive to be handled by nonspecialized personnel and even on an outpatient basis for routine determinations. 4. Experimental Section {#sec4} ======================= 4.1. Apparatus and Electrodes {#sec4.1} ----------------------------- Amperometric measurements were performed with a Metrohm Autolab PGSTAT M204 electrochemical workstation using Nova v1.11 software. The transducers, purchased from Dropsens S.L. (Oviedo, Spain), were AuNP-modified screen-printed carbon electrodes (ref DRP-110GNP) consisting of a 4 mm diameter working electrode, a carbon counter electrode, and a Ag pseudoreference electrode and the specific cable connector with the potentiostat (ref DRP-CAC). Scanning electron microscopy (SEM) characterization studies carried out both by us and by the electrode supplier company confirm the presence of AuNPs homogeneously distributed over the whole surface with a mean diameter of 30 nm. A Bunsen AGT-9 Vortex, a Raypa steam sterilizer, a Telstar Biostar biological safety cabinet, an Optic Ivymen System incubator shaker (Comecta S.A., Sharlab), and a NanoDrop ND-1000 spectrophotometer were employed. 4.2. Reagents and Solutions {#sec4.2} --------------------------- Highest available grade reagents were used. NaH~2~PO~4~, Na~2~HPO~4~, NaCl, and KCl were purchased from Scharlab. Tris-(hydroxymethyl)aminomethane hydrochloride (Tris--HCl), mercaptohexanol (MCH), hydroquinone, H~2~O~2~ (30%, w/v), ethylenediaminetetraacetic acid (EDTA), and bovine serum albumin (BSA) were purchased from Sigma-Aldrich (Germany). ProtA-polyHRP40, a native ProtA labeled with a homopolymer containing 40 HRP molecules (antibodies-online), and an anti-DNA--RNA hybrid antibody (AbS9.6) from Kerafast (Boston, MA) were commercially available. A commercial blocker solution, consisting of a phosphate-buffered saline (PBS) containing 1% w/v purified casein, from Thermo Scientific (Waltham, MA) was also used. The buffer solutions were prepared with Milli-Q water (18 MΩ cm at 25 °C) and sterilized after their preparation to avoid RNase degradation. Phosphate-buffered saline (PBS) consisted of 0.01 M phosphate-buffered solution containing 137 mM NaCl and 2.7 mM KCl (pH 7.5), Tris--EDTA buffer (pH 8.0) formed by mixing 10 mM Tris--HCl, 1 mM EDTA, and 0.3 M NaCl (pH 8.0), and phosphate buffer (0.05 M, pH 6.0) were used. The synthetic oligonucleotides used (sequences shown in [Table [3](#tbl3){ref-type="other"}](#tbl3){ref-type="other"}) were purchased from Sigma-Aldrich (Germany). Once received, they were reconstituted in nuclease-free water to give a final concentration of 100 μM and stored at −80 °C divided into small aliquots. ###### DNA and RNA Used Synthetic Oligonucleotides oligonucleotide sequence (5′ → 3′) ------------------------------- ----------------------------------- HS-capture DNA probe (HS-DNA) ACATCAGTCTGATAAGCTA-\[thiol-C6\] HS-capture RNA probe (HS-RNA) ACAUCAGUCUGAUAAGCUA-\[tThiol-C6\] target miRNA-21 UAGCUUAUCAGACUGAUGU 1-mismatched miRNA-21 (1-m) UAGCUUAU[A]{.ul}AGACUGAUGU miRNA-451 (NC) UUGAGUCAUUACCAUUGCAAA 4.3. Preparation of HS-DNA/AuNPs/SPCEs {#sec4.3} -------------------------------------- Immobilization of HS-DNA onto AuNPs/SPCEs was carried out following the protocol described in our previous works.^[@ref4],[@ref22]^ Briefly, a 10 μL aliquot of a 0.05 μM DNA capture probe solution was incubated over the working electrode in a humidified chamber (8 h, 4 °C). After washing thoroughly with water and drying with nitrogen, the HS-DNA/AuNPs/SPCEs was treated with 10 μL of 0.1 mM MCH and 1% (w/v) BSA solutions (both prepared in Tris--EDTA, pH 8.0) for 5 min and 1 h, respectively. 4.4. Hybridization with the Target miRNA and Labeling of the DNA/RNA Heteroduplex with Anti-RNA--DNA Antibodies and ProtA-PolyHRP40 {#sec4.4} ----------------------------------------------------------------------------------------------------------------------------------- A 10 μL aliquot of the synthetic target miRNA or the sample (heated and diluted serum or RNA~t~ extracted from cells, all in sterilized PBS, pH 7.5) solution was incubated onto the HS-DNA/AuNPs/SPCEs at room temperature for 30 min to hybridize the target miRNA with the immobilized DNA probe. Thereafter, electrodes rinsed with sterilized water and dried with nitrogen were 60 min incubated with 10 μL of a mixture solution containing anti-RNA--DNA hybrid antibody and ProtA-polyHRP40, 1/500 and 1/25 diluted, respectively. This mixture solution, prepared in blocker casein solution, was preincubated for 30 min in an incubator shaker (30 °C, 950 rpm). The modified bioelectrodes were rinsed with sterilized water and dried with nitrogen. To prevent miRNA degradation by RNases, all manipulations, except the amperometric measurements, were made in a laminar flow cabinet. 4.5. Amperometric Detection {#sec4.5} --------------------------- Amperometric detection at an applied potential of −0.20 V versus Ag pseudoreference electrode was carried out in stirred solutions by immersing the prepared bioelectrodes into an electrochemical cell containing 10 mL of 0.05 M phosphate buffer (pH 6.0) and freshly prepared 1.0 mM HQ. Once the background current was stabilized, 50 μL of a 0.1 M H~2~O~2~ solution was added and the current was recorded for about 100 s until reaching the steady-state current. The Δ*i* values given corresponded to the difference between the currents measured in the presence and absence of the target miRNA and are the average of three replicates (α = 0.05). 4.6. Application for the Analysis of Serum and Breast Cells-Extracted RNA~t~ {#sec4.6} ---------------------------------------------------------------------------- The developed bioplatform was used for the determination of the endogenous miRNA-21 content directly in human serum samples (without previous RNA~t~ extraction) and in RNA~t~ extracted from epithelial nontumorigenic and tumorigenic breast cells. Blood serum samples were provided by Prof. Besma Loueslati (Department of Biology, University of Tunis El Manar) and collected from informed consenting patients according to ethical guidelines. Serum samples corresponded to healthy individuals and breast cancer patients. A serum aliquot was five-times diluted with the sterilized PBS (pH 7.5) solution and heated for 15 min at 95 °C (in a Stuart block heater) to release the miRNAs from microparticles and exosomes.^[@ref18]^ Subsequently, a 10 μL aliquot of the diluted and heated sample was cast onto HS-DNA/AuNPs/SPCEs and the protocol described in [Section [4.4](#sec4.4){ref-type="other"}](#sec4.4){ref-type="other"} for the synthetic target miRNA was followed. Cell culture and RNA~t~ extraction were performed following protocols described previously.^[@ref29],[@ref30]^ The quality and concentration of the extracted RNA~t~ were confirmed by measuring the absorbance at appropriate wavelengths. ^∥^ Departamento de Química Analítica, Facultad de CC. Químicas, Universidad Complutense de Madrid, E-28040 Madrid, Spain (J.M.P.). The authors declare no competing financial interest. University of Tunis El Manar is acknowledged for the mobility grant (Bourse d'Alternance) awarded to M.Z. The financial support from the Spanish Ministerio de Economía y Competitividad Research Project (CTQ2015-64402-C2-1-R) and the NANOAVANSENS Program from the Comunidad de Madrid (S2013/MT-3029) are gratefully acknowledged. The authors also thank Drs. J.M. Sánchez-Puelles and E. López-Hernández for kindly providing us the RNA~t~ samples extracted from cell lines and Prof. Besma Loueslati for the blood serum samples. They also acknowledge Dropsens S.L. company for SEM characterization of the working electrodes.

Chronic leg wounds have a considerable impact on quality of life. It has been reported that over 70% of patients experience pain associated with their wounds which can range from moderate to severe. Wound pain and intensity is highly variable, it is not an accurate predicator to make clinical assumptions that specific wound types or wound size will define the type of pain the patient is experiencing. Pain intensity can be stable over time, vary day-to-day, and may increase. Wound pain is an indicator of ineffective wound management, the underlying causal pathology has not being identified nor treated or infection is present. Ineffective wound pain management can result in delayed healing and lack of compliance by the patient. Wound pain can be caused by skin damage, nerve damage, blood vessel injury, infection and ischaemia. It can lead to hypoxia which impairs wound healing and increase infection rates. As tissue oxygen decreases, there is a decrease in the production of leucocytes which provides opportunity for bacteria to colonise the wound thus leading to infection Nerve damage is constantly occurring in the wound, due to the biochemical processes occurring in the wound and external stimuli such as wound debridement, cleansing or dressing changes. It is normally the underlying pathology or aetiology of chronic wounds which dictates the sort of pain that the patient may experience. In order to provide appropriate and effective treatment for addressing wound pain it is necessary to understand the aetiology of the wound, treat the underlying cause and eliminate the noxious stimuli. This presentation will provide an overview of common leg wound aetiologies and related pain symptoms.

1. Introduction {#sec1} =============== Recent advances in multiparametric magnetic resonance imaging (mp-MRI) that have expanded its role in the diagnosis of organ confined prostate were focused mainly on a lack of standardized reporting methods, which led to widely variable diagnostic performance (different Likert scales) [@bib1], [@bib2]. In 2012, the European Society of Urogenital Radiology (ESUR) introduced a set of consensus guidelines to standardize prostate magnetic resonance imaging (MRI) interpretations, referred to as Prostate Image Reporting and Data System (PI-RADS) [@bib3]. However, due to a lack of integration of the PI-RADS scores from different imaging sequences, there was heterogeneity in the suspicion threshold across studies [@bib4]. Prostate Image Reporting and Data System Version 2.0 (PIRADSv2) was published in April 2015, and provided standardized recommendations on the technical specifications of image acquisition, reporting methods, and integration of mp-MRI scores according to prostate zonal anatomy ([Table 1](#tbl1){ref-type="table"}) [@bib5]. However, there are limited reports in literature on the efficacy of PI-RADSv2, with some initial data suggesting it had limited efficacy [@bib6].Table 1Prostate Imaging Reporting and Data System Version 2 (PI-RADSv2) grading method.Table 1Peripheral zoneTransition zoneDWIT2WDCEPI-RADS scoreT2WDWIDCEPI-RADS score1AnyAny11AnyAny12AnyAny22AnyAny23Any--33≤4Any3+45Any44AnyAny44AnyAny45AnyAny55AnyAny5[^1] The purpose of this study was to evaluate the diagnostic performance of mp-MRI, using PI-RADSv2 recommendations in real-life clinical practice. The following outcomes were analysed:(1)Correlation between mp-MRI PI-RADS score and pathological Gleason score (GS);(2)Size detection limit of mp-MRI for malignant lesions;(3)Location of tumours not detected by mp-MRI;(4)Impact of the time interval between needle biopsy of the prostate and mp-MRI accuracy in detecting tumours. 2. Materials and methods {#sec2} ======================== 2.1. Patient cohort {#sec2.1} ------------------- Following institutional review board approval (No. CIRB 2009/743/D), all patients who underwent radical prostatectomy as primary treatment for biopsy proven primary adenocarcinoma of the prostate gland, between January 1, 2014 and December 30, 2014, were identified. A prospectively maintained clinical registry provided data on the clinical characteristics of these patients. Only those who had mp-MRI within 180 days of surgery were included. Those who had prior pelvic irradiation or neo-adjuvant androgen deprivation therapy were excluded from this study. From this, a total of 39 patients with a median age of 64 years (range 45--74 years) were identified. All relevant mp-MRI images and fully embedded whole mount histopathology were centrally reviewed by one dedicated uro-radiologist and one uro-pathologist respectively. The reference landmarks used for imaging and histopathological correlation were the prostatic apex, base and urethra. The mp-MRI images were analysed using axial cuts to facilitate visualization of the tumour lesions marked on whole mount pathology. In this study, tumour size was defined using the cross sectional area, which was calculated as a product of the length radius (measured parallel to a line drawn from right to left of the gland through the middle of the urethra divided by 2, R1), and the width radius (R2) which was measured from the anterior to posterior of the gland in a line drawn perpendicular to the length, divided by 2. The mathematical formula was based on the area of an ellipse (π × R1 × R2). 2.2. Mp-MRI acquisition {#sec2.2} ----------------------- All patients underwent a high field mp-MRI examination, which was obtained with a 3-T MRI imaging system (Magnetom Verio; Siemens, Erlangen, Germany), using a multi-channel pelvic phased array coil. The MRI protocols included high spatial resolution T2-weighted imaging in the axial, sagittal and coronal planes (turbo spin echo sequences), diffusion weighted imaging in the axial plane (b-values: 0--50, 500 and 1000 s/mm^2^) and dynamic contrast-enhanced (DCE) images. The acquisition parameters were summarized in [supplementary Table S1](#appsec1){ref-type="sec"}. Apparent diffusion coefficient maps were generated from the diffusion weighted images by using the mono-exponential model on a voxel-wise basis, fitting the b-value data. For the DCE MRI, gadoterate meglumine (DOTAREM^®^; Guerbet LLC, Villepinte, France), was administered via an automatic power injector (Medrad, PA, Warrendale, USA) at a dose of 0.1 mmol/kg body weight at a rate of 3 mL/s. An endorectal coil was not used. A single dedicated genitourinary radiologist (with 8 years of experience in reading prostate MRI) retrospectively evaluated and scored all lesions. All assessments were made on a commercial PACS workstation (Carestream, Rochester, NY, USA). Each suspicious lesion detected on mp-MRI was assigned a PI-RADS score in accordance with the PI-RADSv2 criteria. 2.3. Histopathological preparation {#sec2.3} ---------------------------------- All prostatectomy specimens were macroscopically examined and weighed. The base and apex were sampled using the conical method [@bib7]. The remaining prostate surfaces (anterior, posterior, right and left) were inked in different colours for orientation and for determining the surgical margins. A positive surgical margin was defined as tumour reaching inked surfaces. The specimens were sequentially sliced from apex to base of the glands at 3--5 mm intervals, and embedded in paraffin as whole-mount sections. The entire specimen was submitted for histological examination by a senior uro-pathologist (with at least 10 years experience in genitourinary pathology). The sections were stained with hematoxylin and eosin. Each tumour focus was marked out on the glass slide and localized onto histological maps, recording its Gleason grade, location and dimensions. Tumours of similar GS, and less than 1 mm apart in the same plane, were considered part of the same lesion. All malignant foci were included in this study regardless of size. The radiologist and pathologist were blinded to their mutual findings prior to data analysis. 2.4. Data analysis {#sec2.4} ------------------ The data analysis was performed using the presumption that each tumour lesion was discrete and independent. The correlation between PI-RADS scores and Gleason grades was calculated using generalized logistic regression. The limits of tumour detection on mp-MRI were determined using binary logistic regression by comparing mp-MRI detection rate against tumour size. We used Fisher\'s exact test for the correlation between mp-MRI detected lesions and pathological cancer foci, and weighted Kappa coefficient for the association between MRI staging and pathological staging. Statistical significance was defined as *p* \< 0.05. The SAS 9.3 software (SAS Institute, Cary, NC, USA) was used to perform statistical analyses. 3. Results {#sec3} ========== 3.1. Patient and tumour characteristics {#sec3.1} --------------------------------------- There were 39 suitable patients with a median age of 64 years. Most of the patients had MRI scans performed post prostate biopsy (90%, *n* = 35). The median time interval between prostate biopsy and mp-MRI was 30 days (range 7--176 days). There were a total of 93 tumour foci identified on whole mount histology, comprising lesions of GS 6 (*n* = 28, 30%), GS 7 (*n* = 54, 58%) and GS ≥ 8 (*n* = 11, 12%). The patient and tumour foci characteristics are summarized in [Table 2](#tbl2){ref-type="table"}.Table 2Summary of patient demographics and lesion characteristics.Table 2Patient and tumour characteristicsValuesPatient characteristics Number of patients39 Median age, year (range)64 (45--74) Post-biopsy MRI, *n* (%)35 (90%) Pre-biopsy MRI, *n* (%)4 (10%) Median interval from biopsy to MRI,  days (range)30 (7--176)Lesion characteristics MRI suspicious lesions, *n*81 Malignant on histology, *n* (%)63 (78%) Benign on histology, *n* (%)18 (22%)PI-RADSv2 score 3, *n* (%)18 (22%) 4--5, *n* (%)63 (78%)Cancer foci identified on pathology, *n*93 Mp-MRI detected, *n* (%)63 (68%) Mp-MRI missed, *n* (%)30 (32%)Gleason score 6, *n* (%)28 (30%) 7, *n* (%)54 (58%) ≥8, *n* (%)11 (12%)[^2][^3] 3.2. Evaluation of mp-MRI in detecting prostate cancer {#sec3.2} ------------------------------------------------------ Among the 93 lesions documented on pathology, 63 (68%) were detected and 30 (32%) were missed (false negatives) on mp-MRI. Therefore, the sensitivity and positive predictive values of mp-MRI in detecting prostate cancer in this study were 68% (95% CI: 57%--77%) and 78% (95%CI: 67%--86%), respectively. 3.3. Correlation of mp-MRI and pathological findings {#sec3.3} ---------------------------------------------------- Of the 63 lesions detected on mp-MRI, they comprised GS 6 (*n* = 13, 21%), GS 7 (*n* = 41, 65%) and GS ≥ 8 (*n* = 9, 14%). Of the GS 6 lesions, they were classified as PI-RADS 3 in six and PI-RADS 4--5 in seven. Of the GS 7 lesions, the mp-MRI grading were PI-RADS 3 in eight and PI-RADS 4--5 in 33. All lesions with GS ≥ 8 were graded PI-RADS 4--5 on mp-MRI. The mp-MRI detection rates in relation to tumour grades were: 48% (13/27) for GS 6, 75% (41/55) for GS 7, and 82% (9/11) for GS ≥ 8 lesions. Higher Gleason grade cancer lesions were associated with higher PI-RADS scores (odds ratio \[OR\] = 3.53, 95% CI: 0.93--13.45, standard error \[SE\] = 0.68, *p* = 0.064, for GS 7 lesions correlating with PI-RADS 4--5). In associating radiological and pathologic stage, the weighted Kappa value was 0.69 (*p* \< 0.0001). When analysing all lesions as a cohort, there was no significant relation between tumour detection and its location (*p* = 0.62 by Fisher\'s exact test). 3.4. Determination of tumour size detection limit on mp-MRI {#sec3.4} ----------------------------------------------------------- In estimating the limits of tumour size detection on mp-MRI in this study population, the probability of cancer detection on mp-MRI was compared against tumour size measured on whole mount histology ([Fig. 1](#fig1){ref-type="fig"}). For a cancer lesion with a surface area of 20 mm^2^ (approximately 5 mm × 5 mm in diameter), the probability of detection was 50%; for a lesion 100 mm^2^ in size (approximately 11 mm × 11 mm in diameter), its probability of detection was 75%. The overall median tumour size in this study was 57 mm^2^ (range 1.6--955.0 mm^2^).Figure 1Size detection limits of cancer by multi-parametric magnetic imaging resonance (mp-MRI). Probability of tumour detection was calculated using binary logistics regression. Tumour size was defined as the largest cross sectional areas of cancer foci on pathology in square millimeters.Figure 1 3.5. Characteristics of tumour lesions missed on mp-MRI {#sec3.5} ------------------------------------------------------- There were 30 cancer foci identified on whole mount histology that were not detected on mp-MRI, giving a false negative rate of 32%. These comprised GS 6 (n = 15), GS 7 (*n* = 13) and GS ≥ 8 (*n* = 2) lesions. The median size of these missed lesions was 19 mm^2^ (range 2--220 mm^2^). The locations of these missed lesions were apex (*n* = 16, 54%), mid-gland (*n* = 7, 23%) and base (*n* = 7, 23%). In a subgroup analysis of cancer foci larger than 10 mm in diameter (in any dimension), there were 51 such lesions, of which eight (16%) were missed on mp-MRI. Of these missed large lesions, 88% (7/8) involved the apical location. On histology, 50% (4/8) had narrow footprints on cross section ([Table 3](#tbl3){ref-type="table"}).Table 3Characteristics of missed large lesions.Table 3Patient No.GSLength (mm)Width (mm)Area (mm^2^)Location1714222AM2711325AM3718455AM4636257A5812985AM671810137A771415165MB872810220AM[^4] 3.6. The impact of time interval between prostate needle biopsy and acquisition of mp-MRI images {#sec3.6} ------------------------------------------------------------------------------------------------ Within the 35 patients who had mp-MRI performed after prostate biopsy, a total of 83 cancer foci were identified on pathology. These were dichotomized into two groups: Less than 6 weeks (*n* = 26) and greater than or equal to 6 weeks (*n* = 57). The false negative rates were 31% (8/26) for less than 6 weeks, and 33% (19/57) for greater or equal to 6 weeks. 4. Discussion {#sec4} ============= This study represents the first few in literature correlating the performance of PI-RADSv2 in detecting prostate cancer using radical prostatectomy specimens as a standard of reference. Previous studies using needle biopsy tissue diagnosis and localization may have outcomes influenced by technical constraints of the biopsy process, including anatomical locality and the sensitivity of tumour detection [@bib8], [@bib9], [@bib10], [@bib11], [@bib12], [@bib13]. These limitations were overcome, in this study, by using radical prostatectomy specimens as reference standard for anatomical and pathological correlation. Of note, anatomical distortion during mp-MRI image acquisition was minimized as there was no utilization of an endorectal coil. There was also awareness that post-processing changes occurring during specimen preservation may confound the correlation between imaging and histopathology. These potential confounders were largely overcome by adopting fixed anatomical points of reference comprising the urethra, prostatic base and apex, which were unlikely to be modified after paraffin embedding. It was well recognized that post-biopsy artefacts may confound the imaging analysis, and hence there has been a preference towards acquisition of pre-biopsy mp-MRI images [@bib14], [@bib15]. However, it is still the predominant practice to perform mp-MRI after histological diagnosis of cancer. Within this clinical context, the outcomes from this study could guide clinicians in disease localization and planning of treatment. This is especially relevant in pre-operative surgical planning for laterality of nerve-sparing. The risk of missing a significant apical lesion in this series was 35%. It was recognized that positive surgical margins were largely apical in location and, therefore, where a pre-biopsy MRI was not possible, adequate apical sampling during needle biopsy would be warranted [@bib16], [@bib17]. The lesions missed on mp-MRI also tend to have a narrow footprint on cross sectional imaging, which can make radiological detection challenging in the transverse cross section. The existing recommendations propose a minimum time interval of 6 weeks between performing the needle biopsy and mp-MRI [@bib18]. However, an exploratory analysis in this study comparing the proportion of lesions missed on mp-MRI, which performed within or after a 6 week time intervals from biopsy, showed no difference in the false negative rates of tumour detection (30%). In addition, hemorrhagic artefacts could be seen at the scans performed 6 weeks after biopsy, degrading the interpretation of the mp-MRI data ([Fig. 2](#fig2){ref-type="fig"}). These suggested that the optimal time interval for mp-MRI after prostate biopsy should be longer than 6 weeks; this will require elucidation in a larger patient cohort comparing pre- and post-biopsy images. There may also be a suggestion that the presence of hemorrhage in post-biopsy scans may not compromise the detection of extra-prostatic disease in the early post-biopsy time period, but only if the reporting radiologists were aware of tumour location(s) at biopsy [@bib19]. In this study, there was no attempt to quantify the extent of post-biopsy hemorrhage, as there is no objective methodology for doing so. The subjective assessment of hemorrhage can also vary between image readers and, thus, is of limited value.Figure 2Persistent haemorrhagic changes on magnetic resonance imaging obtained 51 days post transrectal ultrasound needle biopsy. (A) Axial T2-weighted image of prostate at the level of the mid-gland demonstrating an ill-defined, mildly hypointense lesion in the right posterior peripheral zone (arrow). (B) Axial apparent diffusion coefficient map of the prostate at the same level demonstrating the lesion as a mildly hypointense signal (arrow). The lesion was isointense on diffusion weighted images (not shown). (C) Axial T1-weighted image demonstrates hemorrhage appearing as a hyperintense signal (arrow), limiting the assessment of dynamic contrast enhanced images. Histology showed a corresponding Gleason score 6 (3 + 3) lesion.Figure 2 The sensitivity and positive predictive values (PPV) for our study were 0.68 and 0.78 respectively, which were similar to the diagnostic performance of the original PI-RADS, based on a meta-analysis of 14 studies by Hamoen et al. [@bib4]. In that review, the pooled sensitivity was 0.78 with PPV ranging from 0.50 to 0.83. However, most of the histological correlation was made with prostate biopsy outcomes, rather than a gold standard of prostatectomy histopathology. More recently, some authors have reported an improved diagnostic accuracy when mp-MRI included magnetic resonance spectroscopy imaging (MRSI) [@bib20], [@bib21]. The authors reported a high negative predictive value of 0.93 in a cohort with prostate specific antigen (PSA) values of \<10 ng/mL, suggesting a utility of mp-MRI in identifying patients who will have negative biopsies [@bib22]. When comparing studies with similar methodology in a meta-analysis by Kirkham et al. [@bib23], the diagnostic efficacy was similar when using both T2 and DCE images (sensitivity 0.73--0.89) referenced to histological outcomes at prostatectomy, with inclusion of all tumour lesions. However, it was not clear if these studies looked at pre- or post-biopsy cohorts. In a study comparing PI-RADSv2 with version 1, using MRI-guided biopsy results as the reference standard, Polanec et al. [@bib24] reported a higher sensitivity for lesions in the transitional zone (TZ), but lower in the peripheral zone (PZ); the overall sensitivity using PI-RADSv2 was close to 100%. In another retrospective study by Vargas et al. [@bib25], the authors correctly classified 95% of tumours of any grade ≥0.5 mL in size with PI-RADSv2, which decreased to 20%--26% for GS ≥ 7 tumours that were ≤0.5 mL. This study was not designed to be a diagnostic accuracy study, and pathologically detected tumours were used as a template to classify MRI detected lesions. The characteristics of missed lesions in this study were those of small physical dimensions (median size being 19 mm^2^, which is approximately 5 mm in diameter), and lower GS. This was also similarly shown in earlier studies using different radiological reporting systems [@bib26], [@bib27]. In those studies, lesions located in the apex were more likely to be missed. The inclusion of all visible lesions on histopathology allowed for determination of the size detection limit for mp-MRI in the post-biopsy setting. This was in contrast to other studies where small lesions (\<0.5 mL volume) of lower grade were excluded [@bib27]. In the subgroup analysis of large lesions that were missed on mp-MRI (defined as at least 10 mm in any one dimension), the majority were located in the apex. Furthermore, half of these had narrow footprints on cross section ([Fig. 3](#fig3){ref-type="fig"}). This finding suggested that a good sampling of the apex during needle biopsy of the prostate gland is important, and should be strongly recommended. Careful attention must therefore be paid when dissecting the apex during radical prostatectomy.Figure 3Disc-shaped tumour lesion missed on multiparametric-magnetic resonance imaging. (A) Wholemount histopathology section at the level of mid-gland, showing a tumour focus at the peripheral zone in the left lateral position measuring 18 mm × 4 mm (black arrow); (B) Histology revealed Gleason score 7 (3 + 4) cancer (Haemotoxylin and Eosin \[H&E\] stain, × 40 magnification); (C) and (D) Axial T2-weighted and diffusion-weighted images of the same prostate gland at the same level did not demonstrate the lesion seen on histology.Figure 3 There are limitations in this study. The inherent shortcomings of a limited size cohort include the phenotypic expression of cancer lesions, where a group of tumour foci from the same prostate specimen may share MRI characteristics that could affect analysis on a per lesion basis. The population studied was limited by the selection criteria of surgical candidates. The majority of lesions were of pathological GS 7, which reflect a contemporary practice of offering active surveillance for GS 6 tumours [@bib16]. Patients operated in our institution, but with mp-MRI scans performed not within our protocol, were excluded. However, it is not evident that these patients are inherently different in characteristics from the study population. The radiologist analysing the mp-MRI images was not blinded to the diagnosis of malignancy, although there was no prior knowledge of the exact anatomical location of each tumour foci on whole mount histology. Mp-MRI interpretation by a single radiologist reader directly impacts on the outcome of this study. It is difficult to quantify the effect of the learning curve using PI-RADSv2 on sensitivity and specificity. However, the significant background clinical experience of the reader is expected to shorten the learning process. In this predominantly post-biopsy cohort of organ confined prostate cancer patients, mp-MRI using PI-RADSv2 showed good correlation with the histological and anatomical profile of tumours detected on whole mount prostatectomy histopathology, providing predictive information on tumour grade, location and size. The post-biopsy changes may post a challenge to the mp-MRI detection of smaller tumours, especially those apical in locations and with narrow footprints on cross sections. Nevertheless, this study still provides a useful guide for planning treatment strategy in a practice where mp-MRI scans are commonly done after tissue diagnosis of prostate cancer is made. These results may not be generalizable to practices where a dedicated radiological service observing stringent protocols for image acquisition and a dedicated reader to interpret mp-MRI images are not available. 5. Conclusion {#sec5} ============= Mp-MRI scans of the prostate gland using PIRADSv2 reporting provides predictive information on tumour grade, location and size, and even in a post-biopsy setting. The findings of this study can help guide urologists in the planning of local treatment strategy to optimize patient outcomes. Author contributions {#sec6} ==================== *Study concept and design*: Edwin Jonathan Aslim, Yan Mee Law, Lui Shiong Lee. *Data acquisition*: Edwin Jonathan Aslim, Yan Mee Law, Viswanath Anand Chidambaram. *Drafting of manuscript*: Edwin Jonathan Aslim, Yan Mee Law, Viswanath Anand Chidambaram, Lui Shiong Lee. *Critical revision of the manuscript*: Puay HoonTan, John Carson Allen Jr, Lionel Tim-Ee Cheng, Li Yan Khore, Benjamin Yongcheng Tan, Christopher Wai Sam Cheng, John Shyi Peng Yuen, Henry Sun Sien Ho, Lui Shiong Lee. *Final approval of manuscript*: Yan Mee Law, Puay Hoon Tan, Lui Shiong Lee. Conflicts of interest {#sec7} ===================== The authors declare no conflict of interest. Appendix A. Supplementary data {#appsec1} ============================== The following is the supplementary data related to this article:Multimedia component 1Multimedia component 1 Peer review under responsibility of Second Military Medical University. Supplementary data related to this article can be found at [https://doi.org/10.1016/j.ajur.2018.05.008](10.1016/j.ajur.2018.05.008){#intref0010}. [^1]: Mp-MRI scores from all sequences were integrated into a single PI-RADS score, based on zonal anatomy of the lesion. DCE, dynamic contrast enhanced imaging; DWI, diffusion weighted imaging; T2W, T2 weighted imaging. Any denotes a score of 1--5. "+" denotes the presence of early enhancement on DCE, while "−" denotes the absence of it [@bib5]. [^2]: Analyses were performed on a per lesion basis. [^3]: Mp-MRI, multiparametric MRI; MRI, magnetic resonance imaging; PI-RADSv2, Prostate Imaging Reporting and Data System version 2.0. [^4]: All measurements taken were based on cross sectional whole-mount sections of the prostate glands at 3--5 mm intervals. Large lesions were those that were at least 10 mm in any one dimension. A, apex; AM, apex to mid-gland; GS, Gleason score; MB, mid-gland to base.

Terms used in this study are based on the following definitionsNausea refers to the unpleasant feeling of throwing up without retching or evacuation of the stomach\'s contents.Vomiting refers to pushing out the stomach\'s contents through the mouth or nose.Retching (dry heaving) is the reverse movement (peristalsis) of the stomach and esophagus without vomiting.Early PON and POV refer to nausea and vomiting up to 2 hours post post-operation.Delayed PON and POV refer to nausea and vomiting.Rescue therapy (liberation therapy) refers to the use of at least one dose of ondansetron 2 mg within 24 hours after surgery.^\[[@R1]\]^Cupping (Suction position): stimulation of a point on skin to create suction and desired effect.^\[[@R2]\]^ 1. Introduction =============== Postoperative nausea and vomiting (PONV) is one of the most common complications of general anesthesia.^\[[@R3]\]^ About 30% of surgical cases report unpleasant experiences after general anesthesia.^\[[@R4]\]^ PONV is one of the most common concerns reported by patients' preoperation visits, even more so than pain,^\[[@R5]\]^ as well as a cause of patient dissatisfaction post-operation.^\[[@R6]\]^ Moreover, PONV is associated with other serious complications, such as aspiration, wound dehiscence, pneumothorax, subcutaneous emphysema, and even esophageal rupture.^\[[@R7],[@R8]\]^ Therefore, anesthesiologists and surgeons consider the control of PONV an important treatment priority. There are a number of drugs to reduce the risk of PONV; however, in addition to their costs, their adverse effects limit their usage in routine clinical practice.^\[[@R9]--[@R11]\]^ For example, Droperidol is black-boxed because it is a risk factor for cardiac arrhythmias.^\[[@R12]\]^ As a result, there is great need for nonpharmacologic techniques (NPTs) and complementary and alternative medicine (CAM) to treat PONV. Acupuncture stimulation of the point P6 (acupuncture points) has been shown to significantly control nausea and vomiting in a World Health Organization (WHO) study.^\[[@R13]\]^ Moreover, stimulated P6 also has analgesic effects.^\[[@R14]\]^ The location of point P6 point is between the flexor carpi radialis and the palmaris longus muscle tendons, about 2 inches proximal to the distal crease of the wrist (Fig. [1](#F1){ref-type="fig"}).^\[[@R15]\]^ Stimulation of this point has been tested in several ways, including in acupuncture, acupressure, electrical stimulation, acoustic stimulation, and so on.^\[[@R15]\]^ {#F1} Cupping therapy is a 2000-year-old form of CAM, and depending on its application is classified as dry or wet cupping.^\[[@R16]\]^ Wet cupping involves bloodletting, that is, the evacuation of morbid humor from affected areas.^\[[@R16]\]^ Dry cupping involves diverting morbid matters from one site to another by applying quick, vigorous, rhythmical strokes on intact skin (without bloodletting). Therefore, dry cupping is considered to be a noninvasive and inexpensive technique.^\[[@R17]\]^ More specifically, in this technique, the underlying tissues are pulled into the suctioning cupping glass by heat production to increase the local blood and lymphatic circulation.^\[[@R18]\]^ Although this technique has been used in the treatment of numerous conditions including excessive menstrual bleeding, edema, scrotal hernia, sciatica, hydrocele, postpartum perineal pain, chronic neck pain, and low back pain,^\[[@R19]--[@R24]\]^ we are not aware of any previous studies testing the effectiveness of dry cupping in the treatment of PONV. Considering the fewer side effects of this therapy compared to other medicinal therapies and the lack of available clinical trials on dry cupping and PONV,^\[[@R24]\]^ the present study we aimed to test the preventive effects of dry cupping through stimulation of the P6 on postoperative nausea (PON) and postoperative vomiting (POV), as well as reducing the number of cases requiring rescue therapy (that is, need to treat \[NNT\]). This is the first time dry cupping therapy has been employed in treatment of PONV. 2. Methods ========== 2.1. Study design, participants, sample size -------------------------------------------- This study was a single-center observer-blind randomized clinical trial that was carried out between January and December 2014 at Imam Reza hospital, a university hospital affiliated with Kermanshah University of Medical Sciences in Iran. The inclusion criteria were selected based on previous studies using the Apfel scale and included the following: female sex; age \>18 years; ASA Class I-II; scheduled surgery (laparoscopic cholecystectomy); and type of anesthesia (general anesthesia).^\[[@R7],[@R25],[@R26]\]^ Exclusion criteria include: change in the type of surgery, that is, from laparoscopic cholecystectomy to laparotomy, and ASA-classification III or more. A general practitioner who was part of the study group screened 228 patients. Twelve patients were excluded because of having an ASA III or more. This resulted in 216 consenting patients, 10 of whom were excluded because of a change in their surgical procedures, which resulted in a final sample size of 206 patients (CONSORT Flow Diagram, Fig. [2](#F2){ref-type="fig"}). Subsequently, the study therapist (a physician with several years of experience in cupping) obtained the randomization allocation by the research coordinator who used a computer program to generate the random allocation sequence and had no contact with study patients. Only the therapist knew the randomization result. The patients and the observer of the endpoints were not informed about the allocation. Using a sham technique ensured blinding of the patients before receiving general anesthesia. {#F2} According to the findings of an earlier observer-blind, randomized controlled PONV trial^\[[@R27]\]^ (P1 = 0.63, P2 = 0.33) and a power of 95%, a minimum sample size for each group was 53 and a total of 106 cases and control. We increased the sample size to 206 to increase the quality of the study and to control for the effect of confounding variables such as anxiety. Ten minutes before the induction of anesthesia, all patients received midazolam 1 mg/iv/stat as premedication. This was followed by an infusion of 100 mL of Ringer lactate over 10 minutes, after which the induction of anesthesia started by sodium thiopental (5 mg/kg), sufentanil (0.2 μg/kg), and atracurium (0.5 mg/kg). The maintenance continued with isoflurane (end-tidal between 0.8 and 1.6) and oxygen (7 L/min without N~2~O). During the operation, patients were observed carefully while receiving crystalloids (Ringer lactate). Ten minutes before the end of surgery, all patients received morphine (0.05 mg/kg) or meperidine (0.5 mg/kg). After the surgery, 10 minutes before leaving the post-anesthesia care unit (PACU), analgesic therapy (morphine 0.05 mg/kg) was repeated for all patients. Rescue therapy with ondansetron 2 mg was used for all patients with moderate or severe nausea episodes or if patients requested to have treatment for nausea or vomiting.^\[[@R27]\]^ In both groups, after injection of premedication (midazolam 1 mg, 10 minutes before induction), a 30- to 40-cc cup, fitting the patient\'s wrist, was placed on the dominant hand P6 point. In the experimental group, after the induction of anesthesia, negative pressure (60--100 mmHg) was induced. In the sham group, the cup was remained inactive without negative pressure. The visual analogue scale (VAS) was used to evaluate PONV in patients at 2, 6, and 24 hours after surgery. Anxiety levels before receiving premedication (before injection of midazolam), as well as 2, 6, and 24 hours after surgery were assessed by VAS and both groups were matched for anxiety. The study protocol was reviewed and approved by the local ethics committee of Kermanshah University of Medical Sciences in Imam Reza hospital and is listed in the Iranian registry of clinical trials (IRCT2011020131ON6). All data are presented as mean and standard deviation (SD) unless otherwise indicated. Parametric variables (anxiety, nausea, vomiting, retching, recuse therapy, analgesic satisfaction) were compared using an unpaired Student *t* test. Categorical variables were compared using the *χ*^2^ test (history of motion sickness, migraine headache, or PONV). *P* values for the risk of PONV or the need of rescue therapy were set at \>0.05. The overall, PONV was defined as at least 1 episode of nausea, retching, or vomiting during the observation time of 24 hour. Rescue therapy was defined as at least 1 dosage of tropisetron during the observation time of 24 hour. Adjusted logistic regression analysis was performed in a stepwise backward fashion with the indicated variables as covariates. Differences were regarded statistically significant with an alpha error of \>0.05. All statistical analyses were 2-sided and were performed using SPSS, version 21.0 (SPSS, Chicago, IL). 3. Results ========== The average age of patients in the intervention group was 44.9 ± 13.9 years and in the control group was 47.9 ± 16.3 years, and the difference between the 2 groups was not statistically significant (*P* \> 0.05). As indicated in Table [1](#T1){ref-type="table"}, the 2 groups were similar with respect to smoking status, history of motion sickness, migraine headache, anxiety, nauseas, pain, retching or vomiting, and education (*P* \> 0.05). ###### Patient characteristics at baseline by allocation groups.  According to the Table [2](#T2){ref-type="table"}, dry cupping had significant effect on PONV in the intervention group compared to the control group (sham group). More specifically, the means for nausea in the 2 hours (2.02 vs. 3.89), 6 hours (1.10 vs. 3.33), and 24 hours (0.15 vs. 1.15) were significantly reduced in the intervention group versus the control group (*P* \< 0.001), and the most reducing effect was evident in the 6 hours post-surgery. The means for vomiting also reduced significantly in the experimental group compared to the control group in 2 hours (0.83 vs. 1.84), 6 hours (0.45 vs. 1.46), and 24 hours (0.04 vs. 0.47) (*P* \< 0.001), and the most reduction occurred in the 2 hours post-surgery. Furthermore, the means for rescue therapy at the 2 hours (0.50 vs. 1.04), 6 hours (0.27 vs. 1.07), and 24 hours (0.03 vs. 0.28) reduced in the intervention group compared to the control group (*P* \< 0.001), and the most reduction occurred in the 6 hours post-surgery. ###### Means (standard deviations), mean difference, 95% CI, and *P* value for intervention and control groups and comparing both groups (2-tailed tests).  Table [3](#T3){ref-type="table"} illustrates the predictive role of known risk factors on nausea, vomiting, and rescue therapy, as well as the overall predictive role of intervention on these factors. Of the variables in the model, being in the intervention group significantly reduced the likelihood of nausea (odds ratio \[OR\]: 2.76, confidence interval \[CI\]: 2.15--3.53), vomiting (OR: 1.56, CI: 1.30--1.87), and need for rescue therapy (OR: 1.29, CI: 1.17--1.42), compared to the control group, controlling for other variables in the model. ###### Adjusted logistic regression analysis for known risk factors and dry-cupping on the development of PONV and the requirement of rescue therapy.  4. Discussion ============= Nausea and vomiting are the costly "little big problem" postoperative side effect that compromises patient treatment outcome.^\[[@R28],[@R29]\]^ Nonpharmacological CAM techniques have traditionally helped to relieve nausea and vomiting by stimulating pericardium 6 (P6 nei-guan), located a 3-finger span below the wrist on the inner forearm between the 2 tendons.^\[[@R24],[@R26],[@R28]\]^ In the present study, the stimulation of P6 by dry cupping had a meaningful effect in reducing nausea, vomiting, and the need for rescue therapy 2 hours, 6 hours, and 24 hours after surgery in the intervention group compared to the control group. The most meaningful effect of dry cupping on nausea was observed in the 6 hours, for vomiting in the first 2 hours, and for need for rescue therapy in 6 hours, post-surgery. When we controlled for PONV known risk factors,^\[[@R29]\]^ patients who received dry cupping were nearly 3 times less likely to report nausea, over one and a half times less likely to report vomiting or retching, and 1.3 times less likely to need rescue therapy, post-surgery. As one of the modalities of acupuncture, cupping is valued for its potential to strengthen body resistance, to eject pathogenic factors, and to promote blood circulation. These restore the balance between Yin (negative/passive/dark/water) and Yang (positive/active/bright/fire) and promote the flow of "Qi," which signifies power and movement similar to energy.^\[[@R16],[@R30]\]^ Several suggestions have been made regarding the possible mode of action in dry cupping. These include adjustment in skin blood flow,^\[[@R31]\]^ influencing biomechanical properties of tissues under treatment,^\[[@R32]\]^ increasing anaerobic metabolism in subcutaneous tissue,^\[[@R33]\]^ modulation of cellular part of the immune system,^\[[@R34]\]^ and generally improving microcirculation. This prompts capillary endothelial cell repair, accelerating granulation and angiogenesis in regional tissues.^\[[@R18]\]^ This is the first study testing the effectiveness of dry cupping in reducing PONV. The technique used in dry cupping is similar to acupuncture. We therefore use the acupuncture literature to inform us of possible underlying mechanisms of P6 point stimulation in controlling nausea and vomiting.^\[[@R27],[@R28]\]^ It has been suggested that acupuncture may result in low-frequency electrical stimulation of the skin. This causes nerve activity for A-β and A-δ fibers, which may have an effect in nerve transmission in the dorsal horn and upper neurons. Clement-Jones, McLoughlin et al (1980) also reported that the internal opioid system may be involved in the release of enkephalin, endorphins, beta-endorphin, and dynorphin.^\[[@R35],[@R36]\]^ It is also possible that the effect of P6 point stimulation is because of the inhibition of gastric acid secretion and improvement in stomach motion.^\[[@R15],[@R36]\]^ Confirmatory benefits of sensory effect of stimulation on the wrist in reducing nausea and vomiting compared to placebo device have been reported in pregnant women,^\[[@R37]\]^ in outpatients undergoing laparoscopic cholecystectomy procedures,^\[[@R38]\]^ and in children undergoing anesthesia and surgery.^\[[@R39],[@R40]\]^ In Zarate\'s study, 221 patients were randomized into: active involvement in the P6; sham group; and placebo. The authors reported that intervention significantly decreased (*P* \< 0.05) the incidence of moderate to severe nausea up to 9 hours after surgery compared to the sham and the placebo groups. No difference was detected in the incidence of vomiting and the need for treatment (escape medication). In the study of 120 children undergoing tonsillectomy, the authors found that electrical P6 acupuncture, although patients were anesthetized, significantly reduced the incidence of PONV in the intervention group compared to the sham puncture and the control group (63% vs. 88% vs. 93%; *P* \< 0.001).^\[[@R40]\]^ Further studies are needed to delineate the underlying palliative mechanism of dry cupping on PONV. Overall, our findings support previous acupuncture and acupressure findings in that pressure at the P6 point by dry cupping may have a specific therapeutic effect in preventing the incidence of PONV in patients undergoing laparoscopic cholecystectomy surgery. It is also noteworthy that none of the patients in our study reported unusual side effects. In a systematic review of cupping therapy for acute and chronic pain management Cao et al reported that of 10 randomized clinical trials reported severe adverse effect related to cupping. However, 6 trails reported mild to moderate hematoma, pain, soreness or tingling at the treated site, which are considered common side effects.^\[[@R24]\]^ 4.1. Limitations and future research ------------------------------------ Our study had several limitations. First, it is a single-center study. A multicenter study is essential in supporting generalizability of findings. Second, because we included only females in the study, our findings suffer from converge bias in that they are limited to female population. Future studies are needed to replicate our study with males, and different populations such as patients with different body mass indices and those who were administered different anesthetic agents or underwent different anesthesia procedures. Gastric tube decompression, for example, has no apparent effect on PONV, but the use of nasogastric tubes is associated with higher incidence of nausea.^\[[@R41]\]^ The third limitation involves our lack of control over response bias; high response expectancies may compromise the internal validity of the study.^\[[@R42],[@R43]\]^ As part of the Prophetic Medicine, cupping is worshiped by many.^\[[@R44]\]^ It is suggested that this may involuntarily mediate a participant\'s response to treatment prerandomization as in obtaining informed consent. Current scientific evidence is mixed in supporting this assertion, because of the lack of standardized measures to assess response expectancies.^\[[@R42]\]^ However, the importance of assessing and comparing groups in terms of their expectancies appears significant. Expectations of benefit may interfere with trial validity in studies involving sham medical devices by making it difficult to detect between-group differences.^\[[@R45]\]^ This further limits the validity of our findings in that we did not investigate, determine, and measure sham credibility, and relied on its applicability based on previous studies. To overcome these biases, studies with larger samples are needed. These allow for stratified randomization by strength of response expectancy across both control and sham group.^\[[@R42]\]^ In addition to response expectancies, post-surgery women with real cupping may notice distinctly colored skin over the P6 region because of the negative pressure created during the process. This, of course, would not occur in the control group. Although our patients were blinded to their group assignment, this discoloration effect could introduce additional response bias. Lastly, our study was not designed to establish whether dry cupping was noninferior to acupunctures. Doing so would require studies with larger sample size. 5. Conclusion ============= Our findings are promising in that they suggest dry cupping treatments at the acupressure P6 can prevent the incidence of nausea, vomiting, and need for rescue therapy after laparoscopic cholecystectomy surgery. Additional clinical studies of dry cupping are needed to investigate its effect in preventing and treating PONV in men, and children undergoing similar surgery. Also, considering that dry cupping is a noninvasive procedure with fewer side effects compared to other medicinal therapy, its prophylactic antiemetic therapy should be investigated for different patients, other types of surgeries and different anesthetic agent or anesthesia procedures. More randomized, controlled multicenter studies are needed. They may provide firm evidence of the effectiveness of dry cupping, and demonstrate its underlying analgesic mechanism in preventing PONV. Acknowledgments =============== The authors thank Kermanshah University of Medical Sciences for supporting this project. Abbreviations: ASA class = American Society of Anesthesiology patient classification status, ASA I = Normal healthy patient, ASA II = patient with mild systemic disease; no functional limitation, ASA III = patient with severe systemic disease; definite functional impairment, CAM = complementary and alternative medicine, CI = confidence interval, CONSORT = Consolidated Standards of Reporting Trials, NNT = the number need to treat (NNT), NPT = nonpharmacologic techniques, OR = odds ratio, PACU = post-anesthesia care unit, PONV = post-operative nausea and vomiting, SD = standard deviation, VAS = visual analogue scale, WHO = World Health Organization. This study was registered in the World Health Organization, International Clinical Trials Registry Platform with registration number = IRCT201102011310N6. This article is the result of an anesthesiology residency project \# 89172. This work was performed in partial fulfillment of the requirements for Specialty of Anesthesiology (MD), of Mohammad Kameli, in faculty of Medicine, Kermanshah University of Medical Sciences, Kermanshah Iran. The authors have no conflicts of interest to report.

Introduction {#S1} ============ Adaptive immunity depends upon both antigen-restricted cell-cell interactions and environmental niches, which enable and coordinate cellular communication. In mice, two-photon excitation microscopy (TPEM) has revolutionized our understanding of immune cell architectures and their contribution to normal immunity. By visualizing cells and structures in live hosts, TPEM provides both a quantitative and dynamic picture of immune processes^[@R1],\ [@R2],\ [@R3],\ [@R4],\ [@R5]^. While the gold standard for understanding the organization of immunity, TPEM has several limitations. Cells must be fluorescently labeled to be visualized^[@R6],\ [@R7]^ and, therefore, manipulated systems must be used^[@R8]^. Only small volumes of tissue can be assessed and this must be done over sufficient time to capture cellular dynamics. These restraints limit the number of measurements that can be practically obtained using TPEM. Furthermore, only tissue that can be exposed in live mice is generally amenable to TPEM. While TPEM has a maximal effective depth of 1.6 mm^[@R9]^, most applications are limited to less than 500 μm. Therefore, immune processes occurring within the interior of some organs cannot be visualized. Finally, with few exceptions^[@R10],\ [@R11]^, TPEM cannot be used to directly study human disease. Great strides have been made in multiparameter imaging of fixed-human tissue such that 36 or more markers can be assayed simultaneously^[@R12],\ [@R13],\ [@R14],\ [@R15]^. With these and other approaches^[@R16],\ [@R17]^ one can identify infiltrating cell subsets and describe their relative regional behaviors. Such studies have revealed that the cellular constituency of inflammation is very complex^[@R16],[@R18]^ and the organization of immune cells can be both characteristic of disease states^[@R13]^ and define prognosis^[@R14]^. However, it is difficult to know why different cell populations appear together. Cells such as T cells and antigen-presenting cells (APCs), can engage in cognate interactions that drive local adaptive immunity and inflammation^[@R19],[@R20]^. Alternatively, cells can just be bystanders of inflammation with different populations coalescing because they are responding to similar environmental cues such as chemokines^[@R21]^. There are limited tools to discriminate between these states in human tissue. Previously, we demonstrated that quantitative analysis of human frozen tissue samples, imaged by multicolor confocal microscopy, could be used to characterize interactions between T follicular helper (T~FH~) cell populations and B cells^[@R19]^. In these investigations, we observed that when T~FH~ cells formed cognate interactions with B cells, their nuclei became tightly apposed. These data indicate that distances between nuclear borders can discriminate between cognate interactions and when T and B cells are merely in close proximity. Therefore, by mapping relative distances between T and B cells in tissue (CDM), we could identify functional relationships. However, the fixed filters and algorithms used in CDM to segment signals within tissue were insufficient for defining positions of larger complex objects such as stains associated with DCs. Furthermore, CDM did not accurately capture object shape. We postulated that this might be important, as T cells adopt different shapes when scanning for antigen and after recognizing peptides in the context of MHC^[@R22],\ [@R23],\ [@R24],\ [@R25],\ [@R26],\ [@R27],\ [@R28]^. In the latter case, T cells flatten against the APC to form a stable synapse. In contrast, T cells scanning for antigen or those engaged in brief antigen-specific interactions (kinapses), do not undergo the same changes in T cell shape and polarity^[@R29]^. We hypothesized that using computational tools that accurately captured T cell shape features and DC boundaries, we could identify stable synapses and, thus, discriminate between cognate and non-cognate T cell:APC interactions in human tissue. Therefore, we implemented a deep convolutional neural network (DCNN) that precisely measured both cell position and shape. The DCNN output was then analyzed with a tuned neural network (TNN) to identify the combination of distance and cell shape features that best discriminated between different T cell populations relative to DCs. The use of a TNN allowed us to restrict our analysis to fundamental morphological features of T cell:DC interactions conserved in mice and humans^[@R22],\ [@R23],\ [@R24],\ [@R25],\ [@R26],\ [@R27],\ [@R28]^. We refer to this analysis pipeline as CDM version 3 (CDM~3~). Herein, we demonstrate that in both mice and humans, CDM~3~ can discriminate between *in situ* cognate and non-cognate T cell:DC interactions^[@R30]^. Results {#S2} ======= Identifying cognate T cell:dendritic cell interactions using TPEM {#S3} ----------------------------------------------------------------- To develop better computational tools to study immunity in fixed tissue, we first used an established murine model of T cell and dendritic cell interactions^[@R30]^. Briefly, three cell populations were prepared ([Fig. 1a](#F1){ref-type="fig"}): CD11c^+^ DCs from CD45.1^+^ mice stimulated *in vitro* with lipopolysaccharide (LPS), loaded with pigeon cytochrome C peptide (PCC, 10 μM), and labeled with the fluorescent dye CMF2HC (blue); polyclonal CD45.1^+^ CD4^+^ T cells (wild-type) labeled with CMFDA (green); TCR-transgenic 5CC7 CD45.2^+^ CD4^+^ T cells labeled with CMTPX (red). Previous studies have identified a peptide recognition rate of 0.1--0.3% for wild-type cells in this model system^[@R31],[@R32]^. Cells were then transferred into wild-type recipient mice and, after 12 h, popliteal lymph nodes (LNs) were imaged by TPEM. These same LNs were then frozen, sectioned, stained for cell nuclei with TO-PRO-3, and imaged by confocal microscopy. TPEM revealed clear differences in the behavior of 5CC7 antigen-specific and wild-type T cells relative to antigen-pulsed DCs ([Supplementary Movie 1](#SD4){ref-type="supplementary-material"}). Many of the PCC-specific 5CC7 T cells engaged in prolonged interactions with DCs, while wild-type T cells were more motile and only interacted briefly. Furthermore, wild-type T cells were spherical when interacting with DCs ([Supplementary Movie 1](#SD4){ref-type="supplementary-material"} and [Fig. 1b](#F1){ref-type="fig"}), while 5CC7 T cells tended to flatten against DCs ([Fig. 1c](#F1){ref-type="fig"}). Quantitative analysis of TPEM data ([Supplementary Table 1](#SD1){ref-type="supplementary-material"}) revealed that measures of cellular motion discriminated between wild-type and 5CC7 T cells relative to DCs. Plotting the mean T cell arrest coefficient per mouse ([Fig. 1d](#F1){ref-type="fig"}) revealed that the motility of 5CC7 cells was less than that of wild-type cells. However, when plotting values per cell across mice there was substantial overlap between the two populations ([Fig. 1e](#F1){ref-type="fig"}). Mean T cell interaction time provided better separation with larger relative differences between wild-type and 5CC7 T cells on a per mouse basis and less overlap when individual cells were plotted ([Fig. 1f,g](#F1){ref-type="fig"}). The cellular mean velocity of T cells, plotted both per mouse and per cell, provided intermediate separation between groups with moderate overlap between individual cell values ([Fig. 1h,i](#F1){ref-type="fig"}). In contrast, in response to very low doses of antigen (DCs pulsed with 0.01 μM PCC) there was relatively little difference between wild-type and 5CC7 cells by most TPEM measures ([Supplementary Fig. 1](#SD2){ref-type="supplementary-material"}). These data indicate that TPEM measures can accurately discriminate between wild-type and antigen-specific T cell populations based on their behavior relative to antigen-pulsed DCs. CDM~3~ captures cell shape {#S4} -------------------------- We then used a custom three-dimensional deep convolutional neural network (DCNN) to analyze confocal images of LN frozen sections, extracting cell type, position, and shape. The DCNN was implemented in Tensorflow (<https://www.tensorflow.org>), with 10 convolutional layers, three maximum pooling layers, and 701,000 trainable variables ([Fig. 2a](#F2){ref-type="fig"} and [Supplementary Fig. 2](#SD2){ref-type="supplementary-material"}). Rather than down-sampling the feature maps (FM) at each maximum pooling layer, we increased the sparsity of subsequent convolutional and maximum pooling layers. This allowed us to capture features at different scales yet retain information critical for accurate segmentation. Training was conducted on patches of pixels (184×184×5) with a model field of view (FOV) of 85×85×5, where FOV refers the size of the region around a single pixel that goes into predicting the cell type of that pixel. Each pixel was assigned the cell type with the maximum predicted probability from the DCNN output. This produced solid nuclear and dendritic cell segmentations on which shape-based object analysis was performed. To train the DCNN, we manually segmented confocal images using ICY Bio Image Analysis software and ImageJ. All segmentations were independently validated by a blinded observer. For the murine experiment above, the total data set of 295 randomly collected high-powered fields (HPFs), corresponding to regions of interest (ROIs) containing all three cell types, was segmented for wild-type T cells, 5CC7 T cells, DCs, and corresponding cell nuclei ([Supplementary Table 1](#SD1){ref-type="supplementary-material"}). Training batches consisted of 4 image patches drawn randomly from the entire dataset, each belonging to four different classes (5CC7 T cells, wild-type T cells, DCs, and background). The DCNN was trained for 200,000 iterations at which point cross-entropy error was stable and small ([Supplementary Figure 2b](#SD2){ref-type="supplementary-material"}). Confocal images segmented by the DCNN were compared for agreement against the manually segmented images using 5-fold DCNN model-based cross validation. The DCNN output as compared to manual segmentation of input images revealed excellent agreement between the two analysis methods ([Fig. 2b](#F2){ref-type="fig"}). To assess the sensitivity and specificity of localization and segmentation, a segmented cell was considered detected if the intersection over union (IOU) of the manually segmented cell with the automatically segmented cell was greater than or equal to 0.5. Overall, across all cells, the DCNN had an average sensitivity of 88%, specificity of 92%, and an IOU of 0.85. ([Supplementary Tables 1 and 2](#SD1){ref-type="supplementary-material"}). From the DCNN output, we extracted relative distances between T cell populations and DCs, as well as features of T cell shape. For the latter, we used seven independent measures of two-dimensional shape that include major and minor axis lengths, convex and regular perimeters, convex and regular areas, and equivalent diameter^[@R33]^ ([Fig. 2c](#F2){ref-type="fig"}). These data were imported into R statistical software^[@R34]^ and analyzed to determine which combination of variables and variable weights best discriminated between the 5CC7 and wild-type T cell populations relative to DCs. Our approach included the use of the following algorithms: logistic regression, support vector machines (SVM), and neural networks. For the latter, three different neural network models (simple, tuned, and linear output) were generated. The performance of each algorithm was assessed as measured by parameters of classification accuracy, error, and receiver operator characteristic (ROC) curve performance for correctly predicting cell type ([Supplementary Table 3a](#SD1){ref-type="supplementary-material"}). An actively tuned NN (TNN) ([Fig. 2d](#F2){ref-type="fig"}) consistently displayed the best performance among neural network models ([Supplementary Table 3b](#SD1){ref-type="supplementary-material"}) at the expense of increased number of steps and computation time ([Supplementary Fig. 3a-c](#SD2){ref-type="supplementary-material"}). Therefore, we used a DCNN followed by a TNN in the CDM~3~ pipeline. Comparison of CDM~3~ to TPEM in mice {#S5} ------------------------------------ Random forest analysis revealed that minimum distance to a DC provided the best discrimination between 5CC7 and wild-type T cells ([Fig. 3a](#F3){ref-type="fig"}). Simply plotting relative distance to closest DC provided excellent discrimination between wild-type and 5CC7 T cells with 5CC7 T cells being, on average, much closer to the nearest DC ([Fig. 3b](#F3){ref-type="fig"}). We next plotted the true-positive rate (sensitivity) versus the false-positive rate (1-specificity) in a receiver-operator curve (ROC) to determine how distance performed as a test to discriminate between wild-type and 5CC7 T cells ([Fig. 3c](#F3){ref-type="fig"}). This analysis revealed that cellular distance provides good discrimination with an area under the curve (AUC) of 0.70 (95% CI: 0.62--0.74, *P* \< 5 × 10^−5^). Among the T cell shape variables, minor-axis length was the most promising for discriminating wild-type from 5CC7 T cells. Comparison of T cell minor axis length at distances of less than 5 versus 75 μm or greater from DCs, revealed that 5CC7 T cells had a longer minor axis compared to wild-type T cells at close distances ([Fig. 3d](#F3){ref-type="fig"}). Plotting individual cell minor axes as a function of distance ([Fig. 3e](#F3){ref-type="fig"} and [Supplementary Fig. 4a](#SD2){ref-type="supplementary-material"}) clearly revealed a subpopulation of 5CC7 cells displaying high values, especially when contacting a DC (0 μm). This difference was diminished by 25 μm and largely lost at distances greater than 50 μm. These changes are consistent with the 5CC7 cells flattening against the antigen-pulsed DCs resulting in an increase in minor cell axis for some cells in the two-dimensional plane of the confocal micrograph. In contrast, the minor axis of wild-type T cells in some cells decreased as a function of distance from DCs. This latter trend is consistent with wild-type T cells becoming more spherical upon contacting DCs. Plotting T cell cross-sectional area as a function of distance revealed similar results ([Fig. 3f,g](#F3){ref-type="fig"} and [Supplementary Fig. 4b](#SD2){ref-type="supplementary-material"}) with a subset of 5CC7 T cells having relatively large cross-sectional areas when very close to or abutting DCs. Wild-type T cell area decreased at close DC distances. In contrast, 5CC7 and wild-type T cells have similar shape characteristics at distances of 75 μm or greater from DCs. These data suggest that there were no substantial intrinsic differences between the T cell populations in shape or size. Rather, antigen-specific 5CC7 and wild-type cells diverge in their shape properties in proximity to antigen-pulsed DCs. These data suggest that CDM~3~ can capture changes in T cell shape that occur upon recognition of antigen presented by DCs. We next determined how well CDM~3~ could discriminate between the 5CC7 and wild-type T cell populations by simultaneously incorporating both distance relationships and T cell shape. We focused our analysis to T cell populations within 25 μm of DCs, as this was the distance at which we observed substantial differences in T cell shape. Furthermore, we examined close distances because we were interested in discriminating between T cells that recognize antigen from those that are scanning peptide-MHC class II complexes, looking for antigen. The full CDM~3~ output, which integrates distance and T cell shape variables, provided an AUC of 0.84 (95% CI 0.80 -- 0.90, *P* \< 5 × 10^−5^) and was substantially better than distance alone for all measurements ([Fig. 3h](#F3){ref-type="fig"}). In contrast, at distances of greater than 75 μm, the two T cell populations were indistinguishable ([Fig. 3i](#F3){ref-type="fig"}). Within 25 μm, the use of the minimum distance variable by itself could also discriminate between 5CC7 and wild-type cells (data not shown). However, differences between populations were less robust (AUC=0.65, 95% CI: 0.59--0.72, *P* = 0.008). These data indicate that CDM~3~, by combining measurements of both cell distance and T cell shape, provides excellent discrimination between T cell populations that are scanning for antigen versus those that have recognized antigen. We next examined how well measures obtained by TPEM discriminated between 5CC7 and wild-type cells interacting with antigen-pulsed DCs. Therefore, we took the TPEM outputs described above ([Fig. 1](#F1){ref-type="fig"} and [Supplementary Table 1](#SD1){ref-type="supplementary-material"}) and subjected them to the same statistical modeling by plotting their true-positive rate versus false-positive rate. The arrest coefficient provided good discrimination with an AUC of 0.74 (95% CI: 0.72--0.82 *P* \< 5 × 10^−5^; [Fig. 3j](#F3){ref-type="fig"}). Cell velocity was more robust with an AUC of 0.86 (95% CI: 0.78--0.90, *P* \< 5 × 10^−5^, [Fig. 3k](#F3){ref-type="fig"}). Only cell interaction time, with an AUC of 0.95 (95% CI: 0.94--0.97, *P* \< 5 × 10^−5^), substantially outperformed CDM^[@R3]^ ([Fig. 3l](#F3){ref-type="fig"}). These data indicate that CDM~3~ performs as well as many TPEM measures in identifying antigen-specific T cell interactions with DCs. T cell nuclei as a surrogate for cell shape {#S6} ------------------------------------------- We next sought to apply CDM~3~ to the analysis of multichannel confocal images of human tissue. However, immunofluorescence with antibodies to surface markers is often inadequate for identifying the exact boundaries of lymphocytes in dense infiltrates. In our original studies using CDM, nuclear stains were necessary to define lymphocyte position^[@R19]^. As the nucleus constitutes the majority of a lymphocyte's volume^[@R35]^, we postulated that nuclear shape would approximate cell shape. Therefore, 5CC7 and wild-type T cell nuclei were segmented and analyzed using CDM~3~. Evaluation of representative images of cell trackers compared to TO-PRO-3 nuclear staining revealed extensive overlap between the two staining signatures ([Fig. 4a](#F4){ref-type="fig"}). Analysis of individual shape parameters revealed similar relationships between 5CC7 and wild-type T cell nuclear shape and distance to closest DCs as observed for their cell tracker counterparts ([Fig. 4b-d](#F4){ref-type="fig"}, [Supplementary Fig. 4d,e](#SD2){ref-type="supplementary-material"} and [Supplementary Table 1](#SD1){ref-type="supplementary-material"}). That is, in a subset of 5CC7 cell nuclei, shape parameters increased close to DCs while in wild-type nuclei they did not. Interestingly, a small population of 5CC7 nuclei displayed a decrease in two-dimensional T cell shape parameters close to DCs. This distance-dependent increase in nuclear shape variability is consistent with 5CC7 cells becoming more irregular (less spherical) upon contacting DCs. This is expected when a cell (sphere) flattens against a DC (surface). Application of the composite distance and T cell nuclear shape scores revealed similar discrimination between 5CC7 and wild-type T cell interactions with DCs as that observed for whole cells ([Fig. 4e](#F4){ref-type="fig"}, AUC=0.82, 95% CI: 0.77--0.91, *P* \< 0.005). Similarly, at distances greater than 25 μm, the two T cell nuclei populations were indistinguishable (data not shown, AUC=0.52, 95% CI: 0.45--0.72). These data indicate that nuclear shape alone can be used to approximate T cell shape for the purpose of discriminating cognate from non-cognate T cell--DC interactions. Finally, to validate our mouse experiments, we independently repeated the adoptive cell transfer experiment in mice and obtained 233 additional ROIsHere, DC and T cell nuclei were subjected to the same segmentation and analysis using CDM~3~ as described for the original experiment. These data revealed similar discrimination as before between 5CC7 and wild-type T cell nuclei with an AUC of 0.82 (95% CI: 0.72 to 0.87, *P* \< 5 × 10^−5^) at distances less than 25 μm from DCs. Likewise, for distances greater than 25 μm, the two T cell nuclei populations were similar (AUC=0.54, 95% CI: 0.45--0.58). Application of CDM~3~ to human lupus nephritis {#S7} ---------------------------------------------- We next sought to understand the relationships between DCs and T cells in human tissue. Specifically, we examined the relative abilities of mDCs and pDCs to present antigen to T cells in lupus nephritis tubulointerstitial inflammation (TII). Conventionally, mDCs are considered a professional APC^[@R36],\ [@R37]^, while the function of pDCs is thought to be the secretion of interferon-α and other cytokines^[@R38]^. However, some subpopulations of pDCs can present antigen to CD4^+^ T cells^[@R39],\ [@R40]^. Single longitudinal sections were captured by either tiling HPFs across the entire renal biopsy (*n* = 10 biopsies), or by capturing those with at least one mDC or pDC (*n* = 12 biopsies). A total of 243 ROIs, from a data set of 687 ROIs, were manually segmented and used for training the DCNN, as described above ([Supplementary Table 4](#SD1){ref-type="supplementary-material"}). An example of raw input images, the output from the DCNN, and a comparison of cell segmentation using DCNN versus manual approaches is depicted ([Fig. 5a](#F5){ref-type="fig"}). Overall, there was an excellent agreement between the cross-validated DCNN and manual segmentation with a sensitivity of 0.72, specificity of 0.86, and an IOU of 0.70 for all cell types ([Supplementary Table 5](#SD1){ref-type="supplementary-material"}). The lowest sensitivity was observed for CD3^+^CD4^--^ T cells, which coincided with the most technically difficult staining, as compared to the other cell types. Another example of the DCNN output for pDCs and CD4^+^ T cell nuclei ([Fig. 5b](#F5){ref-type="fig"}) illustrates the DC outlines and T cell nuclear edges used to calculate minimum distances. Cells that could be categorically assigned with 90% or greater confidence were used for the subsequent analysis. We then compared the DCNN segmentation output to the output obtained using the original CDM^[@R19]^. With detections computed at an IOU of 0.50, the sensitivity and specificity for CD3^+^CD4^+^ nuclei was only 0.07 and 0.17, respectively. The sensitivity and specificity for CD3^+^CD4^--^ nuclei was 0.29 and 0.05, respectively. The segmentation of DCs by CDM was so poor a sensitivity and specificity could not be calculated. Therefore, the original CDM, which used fixed segmentation filters, was poor at capturing nuclear and cell shape. Twenty five lupus nephritis biopsies, scored for degree of TII (integer scale: 0--3^[@R41]^), were stained with antibodies specific for CD11c, BDCA1, BDCA2, CD123, CD3, CD4, and DAPI. In individual longitudinal biopsy sections, the numbers of mDCs (CD11c^+^BDCA1^+^) and pDCs (CD123^+^BDCA2^+^) were determined using CDM~3~. Myeloid DCs were present in all degrees of TII with no statistically significant differences among the groups. In contrast, the majority of pDCs occurred in severe TII (*P* \< 0.05)([Fig. 5c,d](#F5){ref-type="fig"}, [Supplementary Table 6](#SD1){ref-type="supplementary-material"}). CD4^+^ T cells appeared more frequent around pDCs than mDCs; CD4^--^ T cells exhibited similar behavior, albeit they were less frequent ([Fig. 5e](#F5){ref-type="fig"} and [Supplementary Figure 5](#SD2){ref-type="supplementary-material"}). However, the number of CD4^+^ and CD4^--^ T cells per either DC population was similar when considering whole biopsies or HPFs ([Supplementary Table 5](#SD1){ref-type="supplementary-material"}). These data indicate that pDCs are more common than mDCs in severe TII and are associated with a proportional increase in local T cell infiltrates. Both pDCs and mDCs could just be co-segregated with T cells in areas of active inflammation or they could be contributing to inflammation by locally presenting antigen to CD4^+^ T cells. To discriminate between these two possibilities, we analyzed the distance and shape characteristics of CD3^+^CD4^+^ T cell nuclei relative to each DC population in biopsies with severe TII (*n* = 8). Local CD3^+^CD4^--^ T cells provide a non-MHC class II restricted bystander control population ([Fig. 5f](#F5){ref-type="fig"}). Analysis of relative distances revealed that CD3^+^CD4^+^ T cells were, on average, closer to both pDCs and mDCs than CD3^+^CD4^−^ T cells ([Fig. 5g](#F5){ref-type="fig"} and [Supplementary Fig. 6](#SD2){ref-type="supplementary-material"}). Analysis of T cell nuclear shape revealed differences between CD3^+^CD4^+^ and CD3^+^CD4^--^ cells relative to each DC population, with different measures of T cell shape making differential contributions to accuracy relative to either pDCs or mDCs ([Fig. 5h](#F5){ref-type="fig"}). Examination of T cells relative to pDCs for both convex area and equivalent diameter revealed that some of those CD3^+^CD4^+^ cells close to pDCs tended to be larger on cross-section ([Fig. 5i,j](#F5){ref-type="fig"} and [Supplementary Fig. 4f,g](#SD2){ref-type="supplementary-material"}). Similar relationships were observed for T cell equivalent diameter and major axis relative to mDCs ([Fig. 5k,l](#F5){ref-type="fig"} and [Supplementary Fig. 4h,i](#SD2){ref-type="supplementary-material"}). In contrast, some CD3^+^CD4^--^ cells tended to get smaller in cross-section as a function of distance from either DC population. These same trends were observed for 5CC7 and wild-type murine T cells relative to antigen pulsed DCs. However, in contrast to the murine experiment, CD3^+^CD4^--^ cells were larger than CD3^+^CD4^+^ cells when not in proximity to a DC ([Supplementary Fig. 4f--i](#SD2){ref-type="supplementary-material"}). These data demonstrate that by examining changes in T cell shape as a function of distance, one can compare T cell populations that are intrinsically different in size. Plotting the composite CDM~3~ output of distance and T cell shape features revealed clear discrimination between CD3^+^CD4^+^ and CD3^+^CD4^--^ T cells within 25 μm for both mDCs ([Fig. 6a](#F6){ref-type="fig"}, AUC=0.63, 95% CI: 0.55--0.71, *P* \< 0.01) and pDCs ([Fig. 6b](#F6){ref-type="fig"}, AUC=0.65, 95% CI: 0.61 -- 0.69, *P* \< 0.0005). Use of minimum distance alone yielded inferior results in cell type discrimination (data not shown; AUC=0.63, 95% CI: 0.50--0.69, *P* = 0.05, and AUC=0.57, CI: 0.52--0.61, *P* = 0.03 for CD3^+^CD4^+^ and CD3^+^CD4^--^ T cells within 25 μm of mDCs and pDCs, respectively). These data are consistent with distance and T cell shape relationships observed in murine cells and suggest that both mDCs and pDCs present antigen to CD4^+^ T cells. However, as there are more pDCs than mDCs in severe TII, pDCs appear to make a larger contribution to *in situ* CD4^+^ T cell activation. In peripheral blood, a subpopulation of pDCs expressing the surface markers AXL and SIGLEC6 can function as APCs^[@R40]^. However, the lupus intrarenal pDCs are AXL^--^SIGLEC6^--^ (data not shown). Therefore, to confirm that intrarenal pDCs were an important APC *in vivo*, lupus TII biopsies were stained with antibodies specific for CD3, CD4, CD43, BDCA2, and tubulin to visualize the microtubule organizing center (mTOC), as well as DAPI. We then performed three-dimensional confocal imaging on representative T cell:pDC conjugates. As seen, there is polarization of the TCR:CD3 complex towards the interface with the pDC ([Fig. 6c,d](#F6){ref-type="fig"}). Likewise, the T cell mTOC is oriented towards the interface with the pDC. In contrast, CD43 is accumulated at the distal pole complex - all consistent with a canonical, mature T cell:APC synapse^[@R42],\ [@R43]^. To quantify mTOC polarization, lupus renal biopsies with pDCs were stained with antibodies specific for CD3, CD4, BDCA2, tubulin, and DAPI then subjected to confocal microscopy to obtain *z*-stack images as described in Methods. We scored the number of CD3^+^CD4^+^ and CD3^+^CD4^--^ T cells abutting pDCs (186 pDCs across three biopsies), and how many of these T cells had their mTOCs polarized towards the interface with pDCs. CD3^+^CD4^+^ were almost 6-times more likely to be abutting pDCs (a two to three-fold enrichment compared to total T cell numbers ([Fig. 6e](#F6){ref-type="fig"}, [Supplementary Table 6](#SD1){ref-type="supplementary-material"}). Of these, 40% had a mTOC polarized towards pDCs while only 10% of abutting CD3^+^CD4^--^ cells did. Therefore, there was an overall 24-fold difference in conjugate rate between CD4^+^ and CD4^--^ T cells (*P* \< 0.0005). These data confirm that CDM~3~ accurately detects T cell:pDC conjugates. Discussion {#S8} ========== The study of immune cell dynamics in tissue is challenging and presents several trade-offs. Techniques such as TPEM provide direct visualization of precisely labeled and characterized cell populations and quantifies their interactions with other cells over time. However, TPEM can only be used on some tissues at limited organ depths. It is also difficult to apply directly to the study of human disease^[@R10],\ [@R11]^. In this work, we demonstrate that by utilizing multi-channel confocal microscopy and a novel analytic pipeline that we term CDM~3~, we approach the performance measures of TPEM in discriminating stable cognate from non-cognate T cell--DC interactions in mice. These data indicate that a quantitative analysis of many static two-dimensional images can approximate much of the information obtained from time-lapse 3D videos of the same phenomenon. Additionally, as CDM~3~ is performed on single plane images of fixed tissue, we could use CDM~3~ to study human disease and identify important *in situ* APCs. Beyond applicability to the study of both human disease and animal models, not readily accessible to TPEM, CDM~3~ offers several additional advantages. It provides higher throughput, which enables larger sample sizes and robust statistical confidence. CDM~3~ does not require experimental manipulations to label cells and, therefore, can be performed on native systems, minimizing experimental artifact. Furthermore, it can be performed on any tissue and at any depth. CDM~3~ cannot assess the kinetics of cellular interactions and, hence, cannot be used to address some questions. Regardless, for many experimental applications in animal models, CDM~3~ might be a preferred approach. However, the major advantage of CDM~3~ is that it can be performed on single sections from frozen tissue. Therefore, it is ideally suited to human studies and can be applied to biopsies that are routinely obtained as part of clinical care. In lupus nephritis, we were able to identify putative *in situ* cognate T cell--DC interactions and, furthermore, assign relative importance of different DC populations in presenting antigen to CD4^+^ T cells. Overall sensitivities and specificities in human disease were less than those observed in transgenic mice. This could reflect inherent differences in the imaging approaches used in mice and humans. More likely, the lower signal observed in human disease reflects underlying heterogeneity in the cells and antigens driving *in situ* adaptive immunity. However, by capturing more events than is practical with TPEM, we overcame this heterogeneity to make statistically robust conclusions. The sequential use of a DCNN followed by a TNN in CDM~3~ is critical for application to human disease. In mice, we can dictate the antigen specificity of T cell populations and, therefore, use a DCNN to best discriminate between antigen-specific and non-antigen- specific interactions with DCs. We have an ideal training set. In humans, we cannot control T cell antigen specificity. However, by using mice to learn how to extract fundamental features of how T cells interact with DCs, that are conserved across mice and humans^[@R22],\ [@R23],\ [@R24],\ [@R25],\ [@R26],\ [@R27],\ [@R28]^, we can apply the CDM~3~ pipeline to human samples and identify stable cognate interactions. Thus, our sequential pipeline helps overcome major limitations of machine learning including transparency, in-depth understanding of how machine-learning algorithms function, and what newly created predictors or intermediate variables represent. These ambiguities often make it difficult to relate machine learning outputs to meaningful biological variables or behavior^[@R44][@R45]^. In CDM~3~, we designed and trained a highly customized DCNN from scratch because our application was so different than mainstream applications designed for photographic images. Particularly, we were interested in making predictions on multichannel images where there was significant bleed-through between stain channels. We also wanted to classify nuclei based on surface stains that were not spatially co-localized between image channels. To do this we used 3D convolutional kernels to relate information across the channel dimension. This not only allowed us to relate non-spatially co-localized information between image channels, but it also allowed the DCNN to use features from one image channel to mitigate noise or ambiguity in another. Additional customization of our DCNN included selecting an appropriate number of neural network parameters for the CNN such that it was complex enough to learn features in our labeled image data set, but not so complex that it overfit and simply memorized the training set. Training images were also carefully selected to be representative of the whole dataset and representative of variations in image quality. The classes of training examples used were balanced to ensure equal performance in classifying the different cell types. The scale of the dilated convolutions used in the layers of the neural network were selected such that they acted on spatial scales where features relevant to classifying the image data were most likely to exist. Data were normalized using the standard score (z-score) applied separately to each channel of each ROI. These design decisions were made so that learned features from the training data were robust and transferrable to other experimental data. Our studies were limited to single plane confocal images and simple pair-wise cell:cell interactions. However, CDM~3~ is adaptable to three-dimensional images that would provide more definitive measures of *in situ* APC function. Furthermore, the general approach illustrated by CDM~3~ is applicable to the study of complex cellular networks containing three or more cell types. A quantitative understanding of the cellular architectures of *in situ* adaptive immunity and inflammation in human disease will provide new insights into the pathogenic mechanisms of autoimmunity and features of immunity effective against cancer. Methods {#S9} ======= Mice {#S10} ---- B10.A CD45.2^−^ and B10.A CD45.2^+^ 5CC7 TCR-transgenic *Rag2*^−/−^ mice were obtained from Taconic Laboratories through a special NIAID contract. All animal experiments were conducted under a protocol approved by the NIAID Animal Care and Use Committee (LSB-1E) and the University of Chicago Animal Resource Center. All animal experiments were conducted in compliance with all relevant ethical regulations. Adoptive transfer {#S11} ----------------- Dendritic cells were purified from mouse spleens using anti-CD11c beads (Miltenyi). Purified dendritic cells were activated *in vitro* with LPS (1 μg/ml) and pulsed with pigeon cytochrome C peptide (Bachem, sequence corresponding to amino acids 88--104) at high (10 μM) or low (0.01 μM) concentration for 4 h at 37 °C. Activated DCs were labeled with Cell Tracker Blue (CMF2HC, Invitrogen), then injected (1 × 10^6^ per recipient) into the right rear footpad of recipient mice. Polyclonal and 5CC7 TCR-transgenic T cells were isolated from the lymph nodes of B10.A CD45.2^--^ wild-type and B10.A CD45.2^+^ 5CC7 TCR-transgenic *Rag2*^--/--^ mice, respectively, and purified using a CD4^+^ T cell isolation kit (Miltenyi). Polyclonal T cells were then labeled with Cell Tracker Green (CMFDA, Invitrogen) and 5CC7 T cells with Cell Tracker Red (CMTPX, Invitrogen). 2 × 10^6^ of each T cell population were then co-injected IV into recipient mice 18 h post transfer of DCs. 12 h post T cell transfer, mice were subjected to two-photon emission microscopy studies as previously described^[@R46]^. Isoflurane was used to anesthetize mice prior to exposure of popliteal LN (Baxter; 2.5% for induction, ∼1%--1.5% for maintenance, vaporized in an 80:20 mixture of O~2~ and air), and subsequent TPEM was performed as described^[@R46]^. Briefly, imaging was conducted on a Bio-Rad/Zeiss Radiance 2100MP, configured with a Nikon 600FN upright microscope equipped with a 203 water immersion lens (NA 0.95, Olympus) and LaserSharp acquisition control software. Anesthetized mice were maintained in environmental chambers warmed by heated air with the surgically exposed LN kept at 36--37 °C with warmed PBS. Upon completion, mice were euthanized and draining popliteal lymph nodes were isolated, cured overnight in 30% sucrose, and frozen at −80 °C. The tissue was subsequently sectioned at 5 μm thickness and prepared for confocal microscopy. Imaging of mouse tissue {#S12} ----------------------- Mouse tissue sections were prepared and stained with TO-PRO-3 Iodide (Invitrogen) to visualize nuclei and avoid interference with the fluorescence spectrum of transferred cell trackers. Single-fluorochrome controls were utilized to ensure no cross-bleeding was present in between fluorescent channels. Images were acquired at 12-bit depth, 1024×1024 pixel size, at 400× and 630× magnifications utilizing either the SP5 Tandem Scanner Spectral 2-photon confocal microscope or the SP8 3D 3-color STED laser scanning confocal microscope with time gating (Leica). Each region of interest (ROI) was 144.74 μm/1024 pixels wide, corresponding to an average absolute resolution size of 0.28 μm, based on Nyquist sampling. Regions of interest, containing all three transferred cell populations, were selected for acquisition. Raw images were stored in manufacturer-specified .lif format. Lif files were converted to multi-channel .tif images and used as input for DCNN analysis. Renal biopsy staining {#S13} --------------------- This study used a total of 25 renal biopsies from de-identified patients, obtained from the University of Chicago Human Tissue Resource Center (HTRC), Department of Pathology. The tissue was fresh-frozen in OCT Tissue Plus (Thermo-Fischer) and stored at −80 °C. The study protocol was approved by the University of Chicago Institutional Review Board (IRB\#15--0727) and did not require informed consent as no patient data were used. All human experiments were conducted in compliance with all relevant ethical regulations. Confirmation of the diagnosis of lupus nephritis as well as grading of the severity of tubulointerstitial inflammation was performed by a blinded reading nephropathologist (A.C.) as previously described^[@R47]^. In addition, de-identified tonsil samples were utilized from the University of Chicago Pathology Core Facility for antibody testing and validation. Two distinct antibody panels were utilized to stain 3--4 μm thick tissue sections; for pDC analysis -- CD3 (clone SP7, Abcam or clone CD3--12, AbD Serotec), CD4 (clone YNB46.1.8, Abcam), BDCA2 (clone AC144, Miltenyi), and CD123 (clone 6H6, eBioscience); mDC analysis -- CD3, CD4, BDCA1 (clone L161, Beckman Coulter), and CD11c (clone EP1347Y, Abcam). DAPI (Hoechst 33342, Invitrogen) was used with the above to visualize tissue nuclei. Images were acquired using a SP5 or SP8 confocal microscope as described above. In addition to selecting individual ROIs for analysis, selected biopsies underwent tiling, wherein the entire available tissue was imaged and a composite stitched image obtained based on default manufacturer settings (SP8). All images were stored in .lif format. Lif files were converted to multi-channel .tif images and used as input for DCNN analysis. mTOC localization quantification {#S14} -------------------------------- Three 3 μm-thick severely inflamed lupus nephritis biopsies (TII grade 3) were stained with antibodies to CD3, CD4, BDCA2, and DAPI as per the above protocol. Anti-tubulin (clone YL1/2, Abcam) was added to visualize the mTOC. Automated *z*-stack protocol images were obtained from regions containing pDCs using the Leica SP8 laser scanning confocal microscope. The images were processed in Fiji using the 3D viewer plug-in (<https://imagej.net/3D_Viewer>) to obtain a maximal projection in 2D. Manual counting of mTOCs was then performed in a blinded fashion by a single observer (J.A.). All CD3^+^CD4^--^ and CD3^+^CD4^+^ T cells abutting a pDC were quantified along with those, featuring an mTOC at the junction between the T cell and APC. DCNN {#S15} ---- Manual segmentation of confocal images was conducted using ICY Bio Image Analysis software, version 1.8.6.0 (<http://icy.bioimageanalysis.org/>) and Fiji/ImageJ, version 2.0.0-rc-43/1.50i (<http://imagej.nih.gov/ij/>). Image channels were rigidly registered using the 'multimodel' configuration of the imregister function in Matlab^[@R48]^. All segmentations were independently validated by a blinded observer (V.M.L.). The total data set for the murine analysis consisted of 295 ROIs, which were segmented for wild-type T cells, 5CC7 T cells, transferred DCs, and corresponding cell nuclei. The standard score, *z* = (*x* -- μ)/σ, of the ROIs was taken for every individual channel within every ROI independently. All training and inference was performed on *z*-score transformed ROIs. The DCNN was trained with a batch size of four 184×184×6 image patches distributed across 4 Tesla GPUs with 12GB memory per card, system memory of 128 GB with 2 Intel E5--2680v4 CPUs @ 2.4 GHz on the University of Chicago Research Computing Center GPU2 nodes on Midway2^[@R49]^. Image patches were sampled from the entire labeled dataset. Each of the four patches for a training iteration represented one of the four classes to be segmented. Class membership of a patch was determined by the class of the center pixel. Where this 184×184 patch around pixels extended beyond the ROI border, mirror padding was used. Sampling was implemented in Tensorflow (<https://www.tensorflow.org>)^[@R50]^ using a list of all pixel locations by class, stored in system memory. The image, label, and data for each ROI was stored in binary TFrecord format in 32-bit float format and accessed by Tensorflow queue runners for active data augmentation (rotation and mirroring) of training examples onto a queue. Error was computed for all classes within the patch. Softmax binomial cross-entropy was used to compute neural network error over an output image patch of 100×100 pixels, reduced input size of 184×184. This dense output was important because it reduces training redundancy and increases training stability. All convolutional and pooling layers were allowed to shrink the input by filter overlap at each layer in *x* and *y*, known as valid padding. Convolutional layers were padded in the channel dimension to keep channel dimensions the same. All weights were initialized with Xavier initialization^[@R51]^ and all biases were initialized at zero. Gradients were averaged across GPUs for each variable at each iteration. Stochastic gradient descent was used for optimization with learning rate of 0.001 with no decay of the learning rate. The DCNN was trained for a total of 200,000 iterations. Inference was performed on 1024×1024 ROIs with boundary mirroring to fit the field of view. The final segmentation result is taken by assigning each pixel the label of the class with the max predicted probability from the softmax output. Segmentation performance was assessed using 5-fold cross validation, in which a full DCNN model is trained on 4/5^ths^ folds of the data set and tested on the 5^th^ fold, for all 5 folds. We assessed the sensitivity and specificity of our cell detection and computed the mean intersection over union (IOU) - defined as: IOU(A, B) = \|A∩B\|/\|A∪B\|, for each class of detected cells. All object analysis in this work was conducted using scikit-image^[@R52]^. For the purposes of computing sensitivity and specificity, a cell was considered to be detected if the IOU with the truth, based on manual segmentation, was ≥ 0.5. Following segmentation by the DCNN, minimum and mean minimum distances in between cells or nuclei of interest, the convex and regular areas, circularity and eccentricity, convex and regular perimeters, equivalent diameter, major and minor axis lengths, aspect ratio, pixel size, solidity, perimeter/circularity ratio, and probability of belonging to the designated class for each object were computed for detected CD3^+^CD4^+^ and CD3^+^CD4^--^ T cells. In addition, for every T cell the minimum distance to the nearest DC was computed in two ways: the first was simply the minimum Euclidean distance found between all the pixels in the T cell and the nearest DC pixel; an additional mean minimum distance was computed by averaging the distances from all pixels in the T cell to the nearest dendrite. Data Analysis and Algorithm Comparison {#S16} -------------------------------------- Original CDM analysis based on minimum distance^[@R48]^ and preliminary TPEM data was performed with GraphPad Prism 5.0a software for Mac (GraphPad Software). All subsequent dataset analysis was performed using R statistical software (version 3.4.1, The R Foundation for Statistical Computing) and RStudio (version 1.0.153, RStudio), running on MacOS 10.13.1 High Sierra, powered by a 3.5 GHz 6-core Intel Xeon E5 CPU with 32 GB of RAM. Mouse adoptive transfer and lupus nephritis data were subjected to multivariate logistic regression, support vector machine (SVM), random forest, and neural network analyses. The following R packages were used for modeling: randomforest (ver 4.6--14), e1071 (1.7--0), xgboost (0.71.2), rpart (4--1.13) and rpart.plot (3.0.4), neuralnet (1.33), glmnet (2.0--16). The caret package was utilized when multiple instances or tunes were required to be generated and compared. Each instance of SVM, neural network, and random forest model generation was preceded by explicitly invoking a specific kernel seed to allow for result reproducibility. Modeling was performed on log-transformed and normalized input data with a binary outcome variable, representing classification (either wild-type or 5CC7, or CD3^+^CD4^+^ or CD3^+^CD4^--^ cells). To control for incomplete or incorrect segmentation, all objects with areas \< 3 and ≥ 100 square pixels were removed. Test and train datasets were defined as random 1/3:2/3 selections of input data. All *n*-fold cross-validations were performed with an *n* = 5. SVM analysis was subjected to linear, radial, and sigmoid kernels when comparing among models. Random forest analysis was used to define the relative importance of predictors as follows: a default of 500 trees were generated for each datapoint using the randomforest package and the optimal Cp parameter was selected based on minimum square error (MSE) optimization. The resultant RF plots were visualized and the relative importance of each split versus mean decrease in accuracy was recorded. The features across each experiment were compared to determine predictor hierarchy. Results of logistic regression were cross-referenced to ensure data consistency and agreement. ROCR and pROC packages were used to generate AUC curves and confidence interval on all included ROC plot figures with the following parameters: 10,000 bootstrap replicates, stratification, curve smoothing, and a confidence interval alpha of 0.90, corresponding to a type I error of 0.05. *P* values for comparison of neural network algorithm output with the null hypothesis (AUC = 0.5) were obtained with the verification R package using the roc.area function, and are separate from the indicated confidence intervals. Mann-Whitney U-test (unpaired Wilcoxson rank sum test) was utilized whenever group comparisons were performed. A Benjamini-Hochberg correction was applied for large datasets exceeding 5,000 observations with an FDR set to 5%, which was applicable for human lupus nephritis analysis. TNN generation {#S17} -------------- After segregating the data based on minimum distance cutoffs, neural network models were generated using the following independent predictors: convex and regular areas, convex and regular perimeters, equivalent diameter, and major and minor axis lengths. A total of 3 different models of neural networks were generated and their performance compared: a single-layer model with (*n*+1) nodes in the hidden layer, where *n* is number of predictors; an actively tuned model, with 3:(*n*−1) nodes in the first hidden layer with the addition of 0:2 additional hidden layers consisting of a maximum of 2 and 1 additional hidden nodes, respectively; and a linear output neural network model with the specification of no hidden layers, constant weights, and linear output. In order to avoid result skewing due to unbalanced numbers of cell types in relation to distance, each instance of neural network analysis was performed by taking the minimum number of cells (wild-type and 5CC7 cells for mouse, and CD3^+^CD4^+^ and CD3^+^CD4^--^ cells for lupus nephritis), and randomly sampling an equal number of cells from the second population. Every network analysis performed was specified with a threshold of 0.1, stepmax of 1 × 10^8^, default (logistic) activation function, cross entropy error differentiable function, and otherwise default package parameters for learning rate, starting weights, and number of repetitions. When possible, the performance of each algorithm was compared based on the parameters of total error, classification accuracy, and receiver operator curve performance of correctly predicting cell type. Data and materials availability {#S18} ------------------------------- The data that support the findings of this study are available from the corresponding authors upon request. Supplementary Material {#SM1} ====================== The research reported in this publication was supported by the National Institute of Allergy and Infectious Diseases (NIH) under Award Number U19 AI082724 (M.R.C.) and the National Institute of Arthritis and Musculoskeletal and Skin Diseases under Award Number K08 AR068421 (V.M.L.). Funding was also provided by R01 AR055646 (M.R.C.), T32 EB002103 (A.S.), and U01 CA195564 (M.L.G.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Computational support was provided by the SIRAF cluster with special thanks to Chun-Wai Chan, M.Sc (Cancer Center Support Grant P30 CA014599). GPU computation was supported by the University of Chicago Research Computing Center. Computational support on the Beagle supercomputer was provided by the NIH through resources provided by the Computation Institute and the Biological Sciences Division of the University of Chicago and Argonne National Laboratory, under grant 1S10OD018495-01, with thanks to Lorenzo Pesce and Joe Urbanski. All imaging was performed at the University of Chicago Integrated Light Microscopy Facility, with thanks to Vytas Bindokas, and Christine Labno. We thank Rebecca Abraham for careful reading of this manuscript. Competing Interests: R.N.G. is a full-time employee of the National Institutes of Health. There are no other applicable disclosures for the authors. Accession Codes Not applicable. No data with mandated deposition were utilized in this work. {#F1} {ref-type="supplementary-material"}) illustrating input of five layers (three fluorescent channels, TO-PRO-3 nuclei, and differential interference contrast \[DIC\]), four subsequent levels of convolution, and 4 classifier outputs (fluorescent signatures for WT T cell nuclei, 5CC7 T cell nuclei, DCs, and background). Across each convolution level, sparsity is increased, resulting in progressively larger kernel sizes (increasingly larger 3×3×3 boxes with spacing between elements). The ROI size stays constant throughout the DCNN. In the fully connected (FC) layer, all features (600) are integrated at each pixel, which are then ascribed classifier probabilities in the softmax layer (not shown). Each pixel is assigned to the cell type with maximum predicted probability to produce solid objects (segmentations). (**b**) Example of DCNN input and output. Five input channels are processed through the DCNN with an output of solid objects (middle right) for 5CC7 T cells (red), WT T cells (green), and DCs (blue). Far right, DCNN segmentations are shown in colored lines, and the corresponding manual segmentations shown in white lines. (**c**) Illustration of independent cell shape parameters. Convex measurements can be thought of as applying a theoretical rubber band around an object (dashed gray line) and are important in identifying local concave shape changes. The equivalent diameter represents the diameter of a circle, possessing an identical area as a non-circular object. (**d**) Outline of the neural network models used to analyze lymphocyte distance and cell shape data. After segregation by minimum distance, the indicated seven measures of cell shape were scaled, normalized, and used as input into three neural network models (simple, tuned, and linear output) in R statistical software, as described in Methods. The arrows and circles indicate data flow from a single input node throughout the network (applied weights omitted). The maximum potential number of hidden layers and nodes, used for active tuning, is indicated by dashed gray circles, with data flows in between steps denoted by dashed gray lines. Dark circles and lines denote obtained optimal parameters used at the completion of network tuning. *n* denotes the number of input nodes. Scale bars, 10 μm.](nihms-1518196-f0002){#F2} {ref-type="sec"}). (**b**) Cumulative frequency of either 5CC7 (grey) or WT (black) T cells as a function of distance from antigen-pulsed DCs (*P* \< 0.005). (**c**) Plot of sensitivity and specificity of distance alone in discriminating between 5CC7 and WT T cells (AUC 95% CI: 0.62--0.74, *P* \< 5 × 10^−5^). (**d**) Comparison of minor axis length for 5CC7 (grey) or WT (black) T cells at indicated distances. (**e**) Minor axis length for each cell plotted as a function of distance from DCs. (**f,g**) 5CC7 (grey) and WT (black) T cell cross-sectional area as an average at indicated distances (**f**) or for each cell as a function of distance (**g**). (**h**) Plot of sensitivity and specificity of CDM~3~ for discriminating between 5CC7 and WT T cells (AUC 95% CI: 0.80--0.90, *P* \< 5 × 10^−5^). (**i**) Comparison of AUC and 95% confidence interval performance shown in (**h**) with values derived from analysis of cells at distances ≥ 75 μm. (**j,k**) Plot of sensitivity and specificity of TPEM measures for discriminating between 5CC7 and WT T cells including cellular arrest coefficient (**j**, AUC 95% CI: 0.72--0.82 *P* \< 5 × 10^−5^), cellular velocity (**k**, AUC 95% CI: 0.78--0.90, *P* \< 5 × 10^−5^) and cellular interaction time (**l**, AUC 95% CI: 0.94--0.97, *P* \< 5 × 10^−5^). \* *P* \< 0.05, \*\* *P* \< 0.005, two-sided Mann-Whitney U test. Center values denote the mean (**b,d,f,h,i**) and error bars denote standard deviation (**b**), standard error of the mean (**i**), or cross-validation error (**c,h,j-l**). Center lines in box plots (**d,f**) denote median value with upper and lower hinges denoting first (Q1) and third (Q3) quartile values, vertical bars corresponding to values 1.5x of the inter-quartile range (IQR) for Q1 and Q3, respectively, and dots representing outlier values, not included in the above. Diagonal lines in (**c,h,j-l**) denote AUC of 0.5, which represents a random probability ( *P* = 0.5). *n* = 2 independent experiments for all panels.](nihms-1518196-f0003){#F3} {#F4} ![Identification of pDCs as an antigen-presenting cell in lupus nephritis.\ (**a**) Example of DCNN input and output. Utilizing a similar framework as described for the murine model, six input channels were processed through the DCNN with an output of solid objects (middle) for CD3^+^CD4^+^ T cells (red), CD3^+^CD4^--^ T cells (green), and plasmacytoid DCs (blue). Right, DCNN segmentations are shown in colored lines and the corresponding manual segmentations shown in white lines. Scale bar = 10 μm. (**b**) Example of processed image showing outlines of pDCs (red outlines), CD4^+^, and CD4^--^ T cell nuclei (blue). (**c,d**) Frequency of mDCs (**c**) and pDCs (**d**) per biopsy by TII grade (0 = none, 1 = mild, 2 = moderate, 3 = severe^[@R41]^). (**e**) Representative images of mDCs and pDC (green) with CD4^+^ T cells (red) in lupus TII. Nuclei are blue. Scale bar = 40 μm. (**f**), Schematic of CD3^+^CD4^+^ and CD3^+^CD4^--^ T cells relative to DCs. (**g**) Minimum distances between indicated DC populations and CD3^+^CD4^+^ (grey) and CD3^+^CD4^--^ (black) T cells. (**h**) Hierarchy of contribution of distance and T cell shape parameters to accuracy based on random forest analysis of pDC and mDC data sets. (**i,j**) Plot of convex area (**i**) and equivalent diameter (**j**) per T cell \[CD3^+^CD4^+^ (grey) and CD3^+^CD4^--^ (black)\] as a function of distance from pDCs (*P* \< 0.0005 for both groups at all distances). Random 10% of total values plotted for visualization. (**k,l**) Equivalent diameter (**k**, *P* \< 0.05) and major axis (**l**, *P* \< 0.005) per T cell as a function of distance from mDCs. Random 10% of total values plotted for visualization. \* *P* \< 0.05, \*\*\* *P* \< 0.0005, 2-sided Mann-Whitney U test; Benjamini-Hochberg correction with FDR of 5% was used for (**i-l**). Center values denote the mean and error bars denote standard standard error of the mean (**c,d**). Center lines in box plot (**g**) denote median value with upper and lower hinges denoting first (Q1) and third (Q3) quartile values, vertical bars corresponding to values 1.5x of the inter-quartile range (IQR) for Q1 and Q3, respectively, and dots representing outlier values, not included in the above. *n* = 2 independent experiments for all panels.](nihms-1518196-f0005){#F5} {#F6} [^1]: authors made equal contributions. [^2]: Author Contributions V.M.L. prepared the manuscript and was involved in CDM~3~ development (automated and manual approaches), development of the tuned neural network analytic pipeline, assisted with convolutional network development, supervised all manual classification of cell types, and performed all data analyses; A.S. was responsible for the development, training, and testing of the image segmentation and the Tensorflow-based deep convolutional neural network, and conducted with data analysis; N.V.P. performed all two-photon animal experiments and related data analyses; J.A. performed all immunofluorescent staining of samples and 2D and 3D image acquisition, along with use of Imaris analysis; A.C. performed nephropathological scoring and categorization of all biopsy specimens and provided de-identified human tissue samples; D.K. performed the repeat mouse transfer experiments at the University of Chicago; M.M. performed manual classification of cell types; R.N.G. oversaw all experiments and data analyses relating to two-photon animal experiments; M.L.G. provided expert advice on development of CDM~3~, image segmentation, use of DCNN-based approaches to classification of cells algorithms, and method of evaluation; M.R.C. conceived of the project, oversaw its progress, preparation of the final manuscript, and development of CDM~3~ as a tool to predict cognate interactions.

Introduction ============ International medical graduates (IMGs) are an important source of new physicians to Canada. Although many IMGs are able to immigrate to Canada and immediately practice on restricted or provisional licenses ([@R2], [@R3]; [@R7]; [@R25]), other IMGs must complete post-graduate medical education (PGME) training, typically in the form of a residency. The term "IMG" includes Canadian citizens and residents who graduated from medical school abroad (Canadians who study abroad \[CSAs\]). Compared to Canadian medical graduates (CMGs), a larger proportion of CSAs are male, older, have more post-secondary education, have a parent who is a physician and have applied fewer times for admission to medical schools in Canada ([@R5]). Compared to immigrant IMGs, a larger proportion of CSAs are male, younger than 35 years old, single and without children ([@R28]). Like most IMGs, CSAs must complete PGME training to qualify for examinations needed for full licensure in Canada. Most CSAs have few opportunities to complete PGME training where they went to medical school and thus return to Canada to complete PGME training and practice ([@R5]). Recent studies have found that CSAs are more likely to obtain PGME training positions than their immigrant IMG counterparts ([@R31]), and that there is some evidence to support the suggestion of bias in the selection process ([@R1]). Despite having less clinical experience, CSAs appear to be more successful than immigrant IMGs in obtaining a residency position ([@R28]). Thomas and Cohn (2011b) attribute the difference to initial screening criteria that emphasize recent medical school graduates and clinical experience in North America. Unlike immigrant IMGs, CSAs are closer to their undergraduate medical training (many CSAs apply for residency during the fourth year of medical school) and are more likely to have done clinical rotations in North America as part of their undergraduate medical program ([@R5]; [@R30]). [@R28] suggest that interviews may play an influential role in the selection process and may disadvantage immigrant IMG applicants. Post-graduate training positions for IMGs are a limited resource in the medical education system, and there has been increasing pressure to accommodate CSAs by providing post-graduate training spots for them ([@R31]), thereby enabling them to work in Canada. CSA advocates argue that given their existing ties to Canada, CSAs are more likely than immigrant IMGs to work in rural communities and provide a more long-term solution to physician shortages in these communities. However, we were unable to find any data on whether CSAs remain in Canada, and in rural communities, longer than immigrant IMGs. Do the retention patterns of CSAs and immigrant IMGs differ? This study compares the retention patterns of CSAs and immigrant IMGs who completed PGME training in Canada. Specifically, we describe the proportion and predictors of: (1) working in Canada in 2015 and (2) working in rural communities in Canada in 2015. We hypothesize that a larger proportion of CSAs than immigrant IMGs will work in Canada and in rural communities in 2015. Only a handful of studies have examined the retention and migration patterns of IMGs who begin practice in Canada, and most have focussed on IMGs who work immediately in Canada on restricted licenses ([@R4]; [@R27]). Retention rates are higher among IMGs who complete residency training in Canada than provisionally licensed physicians; compared to provisionally licensed physicians, 78% of IMGs who completed residency in Canada remained in the country five years after completing training ([@R32]). To date, no study has examined whether CSAs display the same retention and migration patterns as immigrant IMGs. This study is part of a larger project on CSAs in Canada. The study responds to calls from policy makers and medical educators for more information about CSA and immigrant IMG performances at the various stages of the credentialing and licensing process and their contribution to the supply of physicians in Canada ([@R12]; [@R23]). The study provides information to guide policies related to the limited numbers of IMG residency positions in Canada. Methods ======= The Newfoundland and Labrador Health Research Ethics Board (HREB\#14.154) approved this study. Sources of Data --------------- We linked data from the Canadian Post MD Education Registry\'s (CAPER) National IMG Database with data from Scott\'s Medical Database (SMD). The National IMG Database was funded from 2005 to 2011; it captures longitudinal data to track IMGs as they qualify for licensure and join the physician workforce, and it includes data from various agencies that are involved in the training, assessment, certification and licensing of IMGs ([@R12]). SMD is a listing of physicians in Canada and is the most comprehensive database available to track physician locations ([@R11]; [@R22]). As part of its ongoing monitoring of work locations of PGME residents (Canadian Post-MD Education Registry 2015), CAPER had data from SMD for 2015. Variables --------- Using a cohort study design, we examined two dichotomous (Y/N) outcomes in the analyses: (1) work in Canada in 2015; and among those working in Canada, (2) work in rural communities in 2015. We identified rural communities (population \<10,000) based on 2011 census data ([@R26]), and accounted for Metropolitan Influence Zone (MIZ). Statistics Canada defines a strong MIZ as municipalities where "at least 30% of the CSD\'s \[census sub-division, i.e., municipality\] resident employed labour force commute to \[larger city\]" ([@R26]). This population cut-off is used in studies of rural communities in Canada ([@R16]). The independent variable was type of IMG. We defined CSAs as IMGs who were born in Canada and/or who are Canadian citizens or permanent residents before entering medical school. Immigrant IMGs were IMGs who were neither born in Canada nor Canadian citizens or permanent residents. Preliminary analyses suggested that IMGs who graduated from medical school in Western (the UK, Ireland, Western Europe, New Zealand and Australia) or Caribbean countries have different outcomes than IMGs who did not graduate from medical schools in these countries ([@R19] In Press). We therefore created an independent variable that captured both legal status and training site variables included in the National IMG Database. We examined four groups of IMGs: - Canadian (citizens/permanent residents) who graduated from Western and Caribbean medical schools (CSA/Western). - Canadian (citizens/permanent residents) who graduated from other (not Western or Caribbean) medical schools (CSA/non-Western). - Immigrant international medical graduates of Western and Caribbean medical schools (OtherIMG/Western). - Immigrant international medical graduates of other (not Western or Caribbean) medical schools (OtherIMG/non-Western). Co-variates were gender (male/female), age at start of PGME, years between medical degree and start of PGME, participation in a skills assessment/training program (Y/N), specialty type (family medicine/specialty), PGME region (Ontario/non-Ontario), first PGME rank (resident/fellow) and eligibility for a full license (Y/N). We coded the age at start of PGME variable as younger than 30 years old and 30 years and older. For years between medical degree and PGME, we calculated the difference between the year in which an IMG first entered a Canadian post-graduate medical education program and the year of graduation from medical school and coded the IMG as either a recent graduate (0--5 years) or older graduate (6+ years). After a preliminary analysis, we coded PGME region as Ontario and outside Ontario based on the location of the medical school. Eligibility for a full license refers to whether an IMG passed both the Medical Council of Canada Qualifying Examination Part 2 (MCCQE2) and specialty examinations. Specialty examinations refer to the Canadian College of Family Physicians examination and any of the examinations through the Royal College of Physicians and Surgeons of Canada. For first rank, we coded trainees whose first ranks ranged from PG1 to PG7 as residents (in a residency program) and those whose first rank was PG9 or higher as fellows (in a fellowship program) ([@R14]). Sample ------ To be eligible for the study, IMGs had to have first entered a family medicine PGME program between 2005 and 2009, or have entered a specialty PGME program in 2005 or 2006. These cut-off periods allow sufficient time to qualify for the exams. IMGs who first entered PGME programs before these dates may have passed the MCCQE2 before 2005 (before the start of the Database) and would be coded as not passing the examination. CAPER data in the National IMG Database include all enrolment in PGME programs from 1988 onward. Moreover, the cut-off period allows IMG trainees sufficient time to complete their programs and write specialty examinations. We included fellows because a substantial number of IMGs use fellowship programs to qualify for practice in Canada (Mathews et al. In Press). We excluded US graduates because a number of agencies in Canada (e.g., CaRMS, Collège des médicins du Québec) consider graduates of accredited US medical schools as CMGs ([@R17]; [@R28]). We excluded visa trainees because they are funded by their home countries to complete their residency training in Canada with the expectation that they will return to practice in the home country ([@R13]; [@R15]). Statistical Analysis -------------------- Using SPSS (version 23.0), we described the characteristics of the sample. We used chi-square tests to compare the characteristics of the four groups of IMGs and identify potential interactions. Where the four-group test was significant, we conducted post hoc chi-square tests comparing two groups at a time. We also used chi-square tests between each outcome and relevant predictors, and multiple logistic regression to identify significant (*p* \< 0.05) predictors for each outcome. Variables were examined for possible co-linearity a priori. If variables were correlated (e.g., "age at start of PGME" and "years between medical degree and PGME"), we included only one of the variables in the model. Predictors were removed from the model if they were not significant (based on the Wald test) and if they did not significantly improve the change in the −2 log-likelihood value ([@R29]). The tables list the variables included in the final regression models. Results ======= There were 1,214 IMGs who first entered a family medicine PGME program between 2005 and 2009, or who first entered a specialty PGME program in 2005 or 2006. More than half the sample was an OtherIMG/non-Western IMG (50.3%), was female (55.4%), was an older graduate (61.9%), had not participated in a skills assessment/training program (75.1%), was a family physician (70.2%), had trained outside Ontario (50.5%), had the qualifications to be eligible for a full license (71.3%) and had first entered a PGME program as a resident (92.5%) (Table [1](#T1){ref-type="table"}). Most IMGs (88.0%) were working in Canada in 2015. Of the IMGs working in Canada, 9.1% worked in a rural community in 2015. ###### Characteristics of IMGs who entered family medicine PGME training in 2005--2009 or specialist PGME training in 2005--2006, by full cohort and by group type (*N* = 1,214) Characteristics Full cohort, *n* (%) IMG type *p*-value ----------------------------------------------------------- ---------------------- ------------ ------------ ----------- ------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Gender  Female 673 (55.4) 79 (40.3) 234 (61.3) 19 (76.0) 341 (55.8) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[\*\*](#TFN1-7){ref-type="table-fn"}  Male 541 (44.6) 117 (59.7) 148 (38.7) 6 (24.0) 270 (44.2) Age at start of PGME  Under 30 237 (19.5) 96 (49.0) 56 (14.7) 11 (44.0) 74 (12.1) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[^‡^](#TFN1-5){ref-type="table-fn"}^,^[\*\*](#TFN1-7){ref-type="table-fn"}  30+ 977 (80.5) 100 (51.0) 326 (85.3) 14 (56.0) 537 (87.9) Years between MD and PGME  Recent graduate (0--5 years) 462 (38.1) 189 (96.4) 130 (34.0) 14 (56.0) 129 (21.1) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[^‡^](#TFN1-5){ref-type="table-fn"}^,^[^⋄^](#TFN1-6){ref-type="table-fn"}^,^[\*\*](#TFN1-7){ref-type="table-fn"}  Older graduate (6+ years) 752 (61.9) 7 (3.6) 252 (66.0) 11 (44.0) 482 (78.9) Had skills assessment  Yes 302 (24.9) 9 (4.6) 94 (24.6) 5 (20.0) 194 (31.8) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[^⋄^](#TFN1-6){ref-type="table-fn"}  No 912 (75.1) 187 (95.4) 288 (75.4) 20 (80.0) 417 (68.2) Specialty type  Family medicine 852 (70.2) 168 (85.7) 292 (76.4) 17 (68.0) 375 (61.4) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[^⋄^](#TFN1-6){ref-type="table-fn"}  Specialist 362 (29.8) 28 (14.3) 90 (23.6) 8 (32.0) 236 (38.6) PGME region  Ontario 601 (49.5) 72 (36.7) 203 (53.1) 17 (68.0) 309 (50.6) 0[^§^](#TFN1-2){ref-type="table-fn"}^,^[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}  Other 613 (50.5) 124 (63.3) 179 (46.9) 8 (32.0) 302 (49.4) Full license eligible  Yes 865 (71.3) 151 (77.0) 272 (71.2) 21 (84.0) 421 (68.9) 0.077  No 349 (28.7) 45 (23.0) 110 (28.8) 4 (16.0) 190 (31.1) First rank  Residents 1,123 (92.5) 195 (99.5) 371 (97.1) 20 (80.0) 537 (87.9) 0[^¶^](#TFN1-3){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}^,^[^‡^](#TFN1-5){ref-type="table-fn"}^,^[^⋄^](#TFN1-6){ref-type="table-fn"}  Fellows 91 (7.5) 1 (0.5) 11 (2.9) 5 (20.0) 74 (12.1) In Canada in 2015  Yes 1,068 (88.0) 179 (91.3) 342 (89.5) 22 (88.0) 525 (85.9) 0.146  No 146 (12.0) 17 (8.7) 40 (10.5) 3 (12.0) 86 (14.1) In rural Canada in 2015[\*](#TFN1-1){ref-type="table-fn"}  Yes 97 (9.1) 26 (14.5) 27 (7.9) 1 (4.5) 43 (8.2) 0.045[^§^](#TFN1-2){ref-type="table-fn"}^,^[^†^](#TFN1-4){ref-type="table-fn"}  No 971 (90.9) 153 (85.5) 315 (92.1) 21 (95.5) 482 (91.8) CSA = Canadian who studies abroad; CSA/non-Western = Canadian citizen, not Western/Caribbean graduate; CSA/Western = Canadian citizen, Western/Caribbean graduate; IMG = international medical graduate; MD = medical degree; OtherIMG/non-Western = non-citizen, not Western/Caribbean graduate; OtherIMG/Western = non-citizen, Western/Caribbean graduate; PGME = post-graduate medical education. of IMGs working in Canada. Post-hoc tests: CSA-Western significantly differs from CSA/non-Western; CSA-Western significantly differs from OtherIMG/Western; CSA/Western significantly differs from OtherIMG/non-Western; CSA/non-Western significantly differs from OtherIMG/Western; CSA/non-Western significantly differs from OtherIMG/non-Western; OtherIMG/Western significantly differs from OtherIMG/non-Western. As expected by variable definitions, all CSA-Western and OtherIMG/Western IMGs had graduated from medical schools in Western and Caribbean countries (Table [1](#T1){ref-type="table"}). The largest proportion of CSA/non-Western IMGs graduated from medical school in Eastern Europe and Asia, whereas the largest proportions from the OtherIMG/non-Western group graduated from medical schools in the Middle East and North Africa and Asia. Female IMGs made up a lower proportion of CSA-Western IMGs than the three other groups. Those from the OtherIMG/Western group had a larger proportion of females than OtherIMG/non-Western IMGs. IMGs from Western medical schools (CSA/Western and OtherIMG/Western) had larger proportions of IMGs who were under 30 years of age at the start of their PGME programs than non-Western IMGs (CSA/non-Western and OtherIMG/non-Western). CSA/Western had the largest proportion of younger graduates (96.4%), followed by OtherIMG/Western (56.0%), CSA/non-Western (34.0%) and OtherIMG/non-Western (21.1%). Almost one-quarter (24.9%) of those in the study participated in a skills assessment/training program. A smaller proportion of CSA/Western IMGs participated in skills assessment and training programs than each of the other groups. A larger proportion of CSA/Western IMGs was in family medicine programs than each of the other groups. A smaller proportion of CSA/Western IMGs entered PGME programs in Ontario than all other groups. For retention outcomes, a larger proportion of CSA/Western IMGs than OtherIMG/non-Western IMGs worked in rural communities in 2015. Compared to IMGs who did not work in Canada in 2015, a larger proportion of IMGs who worked in Canada were Canadian citizens, were family physicians, were eligible for a full license and were a resident when they first entered PGME (Table [2](#T2){ref-type="table"}). After controlling for other significant predictors, IMGs who were eligible for a full license were 4.28 times more likely to work in Canada in 2015 (Table [3](#T3){ref-type="table"}). Fellows were 0.38 times as likely as residents to work in Canada. ###### IMGs who worked and did not work in Canada in 2015 and rural Canada in 2015, among a cohort of IMGs who entered family medicine PGME training in 2005--2009 or specialist PGME training in 2005--2006 Variable In Canada in 2015 In rural Canada in 2015[\*](#TFN1-1){ref-type="table-fn"} ------------------------------- ------------------- ----------------------------------------------------------- ----------- ------------ ------------ ------- Type of IMG  CSA/Western 179 (16.8) 17 (11.6) 0.146 26 (26.8) 153 (15.8) 0.045  CSA/non-Western 342 (32.0) 40 (27.4) 27 (27.8) 315 (32.4)  OtherCSA/Western 22 (2.1) 3 (2.1) 1 (1.0) 21 (2.2)  OtherCSA/non-Western 525 (49.2) 86 (58.9) 43 (44.3) 482 (49.6) Gender  Female 596 (55.8) 77 (52.7) 0.485 43 (44.3) 553 (57.0) 0.017  Male 472 (44.2) 69 (47.3) 54 (55.7) 418 (43.0) Age at start of PGME  Under 30 209 (19.6) 28 (19.2) 0.911 18 (18.6) 191 (19.7) 0.792  30+ 859 (80.4) 118 (80.8) 79 (81.4) 780 (80.3) Years between MD and PGME  Recent graduate (0--5 years) 413 (38.7) 49 (33.6) 0.233 48 (49.5) 365 (37.6) 0.022  Older graduate (6+ years) 655 (61.3) 97 (66.4) 49 (50.5) 606 (62.4) Had skills assessment  Yes 264 (24.7) 38 (26.0) 0.732 19 (19.6) 245 (25.2) 0.219  No 804 (75.3) 108 (74.0) 78 (80.4) 726 (74.8) Specialty type  Family medicine 773 (72.4) 79 (54.1) 0 82 (84.5) 691 (71.2) 0.005  Specialist 295 (27.6) 67 (45.9) 15 (15.5) 280 (28.8) PGME region  Ontario 538 (50.4) 63 (43.2) 0.102 50 (51.5) 488 (50.3) 0.809  Other 530 (49.6) 83 (56.8) 47 (48.5) 483 (49.7) Full license eligible  Yes 810 (75.8) 55 (37.7) 0 77 (79.4) 733 (75.5) 0.393  No 258 (24.2) 91 (62.3) 20 (20.6) 238 (24.5) First rank  Residents 1,010 (94.6) 113 (77.4) 0 93 (95.9) 917 (94.4) 0.551  Fellows 58 (5.4) 33 (22.6) 4 (4.1) 54 (5.6) CSA = Canadian who studies abroad; CSA/non-Western = Canadian citizen, not Western/Caribbean graduate; CSA/Western = Canadian citizen, Western/Caribbean graduate; IMG = international medical graduate; MD = medical degree; OtherIMG/non-Western = non-citizen, not Western/Caribbean graduate; OtherIMG/Western = non-citizen, Western/Caribbean graduate; PGME = post-graduate medical education. ###### Predictors of IMGs who worked and did not work in Canada in 2015 and rural Canada in 2015, among a cohort of IMGs who entered family medicine PGME training in 2005--2009 or specialist PGME training in 2005--2006 Variable In Canada in 2015 In rural Canada in 2015[\*](#TFN2-1){ref-type="table-fn"} ----------------------- ------------------- ----------------------------------------------------------- ------ ------- Full license eligible  Yes 4.28 (2.92, 6.27) 0 NS NS  No 1.00 NS First rank  Residents 1.00 0 NS NS  Fellows 0.38 (0.23, 0.63) NS Specialty type  Family medicine NS NS 1.00 0.003  Specialist NS 0.43 (0.24, 0.75) Gender  Female NS NS 1.00 0.009  Male NS 1.77 (1.16, 2.70) 95% CI = 95% confidence interval; IMGs = international medical graduates; NS = not significant predictor (not included in model); OR = odds ratio; PGME = post-graduate medical education. IMGs working in Canada. In total, 97 (9.1%) of the 1,068 IMGs who worked in Canada in 2015 worked in a rural community. Compared with those who did not work in a rural community, a larger proportion of IMGs who worked in a rural community were a CSA/Western IMG, were male, were a recent graduate and were a family physician (Table [2](#T2){ref-type="table"}). After controlling for other significant predictors, male IMGs were 1.77 times more likely to work in a rural community in 2015 than female IMGs (Table [3](#T3){ref-type="table"}). Specialists were 0.43 times as likely as family physicians to work in a rural community. Discussion ========== The vast majority of IMGs in the study were working in Canada in 2015 (88.0%), up to nine years following their training. Among IMGs who worked in Canada, a small proportion of IMGs worked in rural Canada in 2015 (9.1%). The rural retention rates of IMGs (97 of 1,214) is 8.0%; lower than the rate for CMGs reported in the literature (11.2%) ([@R16]). Studies have highlighted the importance of cultural communities in the work locations of IMG physicians; small rural communities may not offer the cultural network that IMG physicians prefer ([@R21]). Male IMGs were more likely to work in rural communities in 2015 than female IMGs. This finding likely reflects traditional gender roles; men may be more likely to head single-income households or have more influence in work locations. Moreover, given the importance of opportunities for spousal employment in the rural communities, non-physician male spouses may have greater difficulty finding suitable employment in rural communities ([@R21]; [@R24]). Family physicians are more likely to work in small rural communities, which may not have sufficient patient size to support a sub-specialist. Similar findings have been reported for CMGs ([@R19]). Almost three-quarters of the IMGs in the cohort qualified for a full license. Although eligibility for full licensure was a predictor of working in Canada in 2015, one-quarter of the study physicians who worked in Canada in 2015 were not eligible for a full license. Roughly one-fifth of IMGs in the study cohort exit PGME programs without full license credentials (Mathews et al. In Press); however, data from the National IMG Database do not reveal why. For example, studies have shown that IMGs do not perform as well as CMGs in PGME programs ([@R1]; [@R8]; [@R17]; [@R30]). Alternatively, IMGs who did not intend to remain in Canada may have chosen not to write the MCCQE2 or specialty examinations. Obtaining a residency position is competitive and many qualified IMGs are unable to secure a training position. Further research is needed to understand and improve IMG performance in PGME. There were no differences in the work location patterns of different groups of IMGs. Contrary to our hypotheses, CSAs were not more likely to remain in Canada or work in rural communities than immigrant IMGs. These findings refute suggestions that CSAs are more likely to be a solution to rural physician shortages than other IMGs. In response to earlier reports that highlighted biases in the selection of IMG post-graduate applicants ([@R31]), medical schools in Canada have strengthened policies around applicant screening, ranking and selection ([@R9]). The findings from this study lend further weight to these initiatives. Although return-of-service agreements are part of rural physician recruitment and retention programs, we are unable to assess the exact role of return-of-service agreements in the work location patterns of IMGs in the study because these data are not included in the National IMG Database. Return-of-service agreements provide physicians with financial incentives (usually during training) in exchange for a commitment to work in an underserved community following training ([@R6]). Before 2007, return-of-service agreements were generally voluntary for IMGs. Starting in 2007, return-of-service agreements were mandatory in most provinces for IMGs who entered the first match of the residency matching process ([@R18]). Because most IMGs in the study would have participated in the CaRMS match before 2007, they are unlikely to have had a mandatory return-of-service agreement. An evaluation of return-of-service agreements in Newfoundland and Labrador found that a large proportion of IMGs (most with mandatory agreements) did not fulfill service commitments ([@R18]); however, this study examined agreements implemented before changes to the residency match (and return-of-service agreements) in 2007. A 2013 study of family physicians initially located in rural areas indicated that, although a greater proportion of new IMGs go to rural areas than CMGs, they migrate out of rural areas much sooner than the CMGs who do not typically have return-of-service agreements ([@R10]). Further research is needed to assess how mandatory agreements have influenced work location and retention of IMGs who completed residency training after 2007. The study also demonstrates the utility of data sets, such as the National IMG Database, that link data from various agencies involved in the training, credentialing and licensing of the health professionals. Research using these data sets provides a valuable means to inform policy debates and support health workforce planning. The study has a number of limitations. As previously described, we created a proxy variable to identify IMGs who qualified for a full license, and used SMD to identify work locations. Some physicians who work in Canada may not be included in SMD, and therefore, we may underestimate retention rates. Our sample size was limited to the period covered by the National IMG Database (2005--2011). Our analyses identify IMGs who are Canadian citizens or permanent residents, but we are unable to describe how long they have lived in the country or when they became citizens or permanent residents relative to their entry to medical school. Moreover, the National IMG Database does not include information on rural background or province of origin -- known predictors of work locations among Canadian medical trainees ([@R24]). Conclusion ========== We linked data from the National IMG Database and SMD to examine the retention of IMG PGME trainees. Most IMGs were working in Canada in 2015 (88.0%); 9.1% of these physicians were working in a rural community. There was no difference in the work location patterns of CSA and immigrant IMGs. CSAs are as likely to remain in Canada or work in rural communities as their immigrant IMG counterparts. The study findings support the equal treatment of all IMG physicians (CSA and immigrant) in PGME selection and training. It also highlights the need to improve IMG PGME training so that more IMGs are able to obtain credentials for independent practice. Finally, it reinforces the need to retain IMG PGME trainees.

1. Introduction {#sec1} =============== Skeletal muscle infarction is ischemic necrosis of striated muscles of the skeleton due to long-term poorly controlled diabetes mellitus, alcoholism, compartment syndrome after blunt trauma or fracture, and vascular diseases such as acute artery thrombosis or embolism and arteriosclerosis obliterans (ASO). The typical clinical symptom is severe abrupt-onset pain in the region of the affected muscle. Herein, we describe a rare case of diabetic muscle infarction (DMI) localized in the region of the tibialis anterior (TA) and extensor hallucis longus (EHL) muscles. The patient\'s clinical symptom and magnetic resonance imaging (MRI) findings were atypical and interesting compared to those of previously reported patients with DMI. The patient was aware of a palpable mass in the lower left leg for 6 months but neglected it owing to a lack of pain. A palpable mass was suggestive of a malignant soft-tissue tumor; therefore, we needed to distinguish the lesion from a malignant soft-tissue tumor. 2. Case Report {#sec2} ============== A 64-year-old man was referred to our hospital with the diagnosis of a soft-tissue tumor in the lower left leg. He had a 10-year history of type 2 diabetes mellitus, and he was consuming oral hypoglycemic agents. However neither nephropathy nor retinopathy was pointed out by clinical examination such as blood test, urine test, and funduscopy before, and he did not complain of sensory disturbance. He also had a history of reconstruction of an acute aortic dissection. He was aware of a palpable mass in his lower left leg but had neglected it for 6 months because he did not feel pain at the site of the lesion. The mass gradually expanded and developed large blisters. Physical examination revealed swelling without tenderness on the anterior aspect of the lower left leg. Blood tests revealed the following: C-reactive protein level, 0.05 mg/L; white blood cell count, 4,580/mm^3^; CK level, 164 IU/L; and casual blood glucose level, 298 mg/dL. Radiography revealed no bony abnormalities and calcification of soft tissue, but there was evidence of moderate soft-tissue swelling. Computed tomography (CT) revealed a large mass (maximum diameter, 7.2 cm; length, 21.0 cm) with cystic lesions in the region of the TA and EHL muscles ([Figure 1](#fig1){ref-type="fig"}). Neither the calcification nor ossification was detected inside and around the mass. The mass and cystic lesions were not contrasted but the anterior tibialis artery was clearly observed by using enhanced CT. MRI revealed that the mass exhibited heterogeneous signals on T1- and T2-weighted images and the cystic lesions exhibited homogeneous signals on T1- and T2-weighted images; perifascial, intramuscular, and/or subcutaneous edema was not seen ([Figure 2](#fig2){ref-type="fig"}). Fat suppressed gadolinium-enhanced T1-weighted images revealed slight heterogeneous enhancement within the affected muscle with focal hypointense nonenhancing areas. The ankle brachial index (ABI) was normal. To distinguish muscle necrosis from a malignant soft-tissue tumor, a needle biopsy (Tru-Cut biopsy needle, 14-gauge, CareFusion, CA, USA) was performed. However, we did not collect full volume of the muscle tissue; therefore, open surgical muscle biopsy was performed. Histopathological analysis revealed mostly necrotic muscle tissues, but neoplastic cells were not observed in the soft tissue resected from the lesion. Because the lesion was a remnant of a previous muscle infarction with several large blisters producing discharge, we resected the affected muscles ([Figure 3](#fig3){ref-type="fig"}). Excisional specimens revealed large areas of muscles with coagulative necrosis, fibrosis, and hemorrhage ([Figure 4](#fig4){ref-type="fig"}). Two weeks after surgery, the wound healed, and the patient was discharged. There is no evidence of any diseases including local recurrence at 9 months after surgery. 3. Discussion {#sec3} ============= The most common causes of skeletal muscle infarction are vascular diseases such as artery thrombosis or embolism and ASO. Another cause is a rare complication of diabetes mellitus. It is very important to distinguish these diseases correctly because methods of treatment for each disease are different. DMI is a rare complication associated with long standing and poorly controlled diabetes mellitus; it was first reported by Angervall and Stener in 1965 \[[@B1]\]. A systematic review by Trujillo-Santos and colleagues in 2003 showed that DMI was more frequent in women (61.5%), with a mean age at presentation of 42.6 years and a mean duration of diabetes of 14.3 years \[[@B2]\]. Approximately 60% of the patients had had type 1 diabetes, and approximately 24% had type 2 diabetes \[[@B2]\]. However, a systematic review by Horton and colleagues in 2015 showed there was not type 1 diabetes or female predominance in patients with DMI \[[@B3]\]. The most frequently affected site of DMI was the thigh (71.2--83.7%), while the calf was the second most commonly affected site (15.3--19.3%) \[[@B2], [@B3]\]. Horton reported that the incidence rate of DMI in TA and EHL muscles was 2.9% and 0.7%, respectively, and so our reported case of DMI localized in TA and EHL muscles was very rare. Vascular complications of diabetes were reported in most patients, mostly nephropathy (71.1%), retinopathy (56.6%), and neuropathy (54.5%) \[[@B2]\]. Our patient had a 10-year history of poorly controlled type 2 diabetes mellitus, but neither nephropathy nor retinopathy was pointed out by clinical examination such as blood test, urine test, and funduscopy before and he did not complain of sensory disturbance. The pathogenesis of DMI may involve atherosclerotic occlusion, hypoxia-reperfusion injury, vasculitis with thrombosis, arterial embolism of small vessels, and a hypercoagulable state \[[@B2]--[@B7]\]. However, it is not fully understood and is controversial. In diabetic microangiopathy, high blood glucose levels cause the endothelial cell lining of the blood vessels to absorb more glucose than normal, resulting in vascular fragility. In our case, enhanced CT revealed that the tibialis anterior artery was clearly visualized, and the ABI was normal. Therefore, we assume that atherosclerotic occlusion and arterial embolism of small vessels caused by microangiopathy led to the onset of DMI. Typical symptoms of DMI can include local swelling, abrupt onset of pain in the affected muscle, and a palpable painful mass \[[@B2]--[@B7]\]. If the area of necrosis is wide, muscle weakness may occur in the area where the muscle tissue necrotizes. Patients with diabetes often have diabetic peripheral neuropathy, a complication in which the peripheral nervous system breaks down, and they lose sensation in their extremities. Therefore, DMI can become severe because patients with diabetes neglect their lesions owing to a lack of pain. Interestingly, our patient had no complaint of pain in his lower leg even though he was aware of a palpable mass for 6 months. According to the clinical course and physical findings, the lesion simulated a malignant soft-tissue tumor more closely than a muscle infarction. Other differential diagnoses except soft-tissue tumor included muscle infection (cellulitis, abscess, and necrotizing fasciitis) and calcific myonecrosis which is characterized by a slowly progressive enlarging, painless soft-tissue mass with occasional tenderness in the lower limb. The most common area of calcific myonecrosis is the anterolateral part of the lower leg. Although this condition is not fully understood, it is assumed that these lesions most likely result from posttraumatic ischemia and cystic degeneration of the muscle. Radiographs typically show a fusiform mass with peripherally oriented plaque-like amorphous calcifications. In our patient, the calcification was not seen by radiographs and CT, and so, we ruled out this disease. In cellulitis, subcutaneous swelling is seen but with no muscle involvement \[[@B8]\]. Necrotizing fasciitis has MR findings including muscle swelling, edema, and inflammatory changes similar to DMI. However, compared to DMI, necrotizing fasciitis has less pronounced muscle involvement and more extensive fascial involvement \[[@B8]\]. Additionally, gas bubbles and fluid collection may be seen in the tissues \[[@B8]\]. To confirm the correct diagnosis, the clinical course, physical findings, CT and MRI features, and histopathological analysis are important. Muscle biopsy is the gold standard for diagnosis and exclusion of other diseases such as pyomyositis, necrotizing fasciitis, and soft-tissue tumors \[[@B8]--[@B10]\]. However, recent studies reported that surgical biopsy would either prolong the disease in the individual or acutely exacerbate this condition \[[@B2], [@B3], [@B9]\]. Chason and Kiers mentioned that MRI was the most sensitive modality, as it was very useful for diagnosis and could avoid the need for biopsy \[[@B9], [@B10]\]. As such, a needle biopsy may be reserved for cases in which the clinical presentation is atypical or the diagnosis is uncertain. In our patient, MRI revealed localization of the lesion only in the TA and EHL muscles, suggesting a previous muscle necrosis or a malignant soft-tissue tumor. To rule out a malignant soft-tissue tumor, we performed a muscle biopsy. Histopathological analysis showed that the necrotic and atrophic skeletal muscle had been infiltrated by necrotic neutrophils. On the basis of the above results, we diagnosed the lesion as a DMI. Because the lesion was a remnant of a previous muscle infarction with several large blisters producing discharge, we resected the affected muscles. In summary, herein, we reported a rare case of DMI in the TA and EHL muscles. The typical symptom of DMI is severe abrupt-onset pain in the region of the affected muscle, but our patient did not complain of pain. The diagnosis of and treatment for DMI were delayed, and widespread irreversible muscle necrosis requiring resection developed. It is true that MRI is the most sensitive test for making the diagnosis of DMI; however, in some cases, MRI finding of DMI can be similar to that of the malignant soft-tissue tumor. So, it is necessary to consider the malignant soft-tissue tumor as one of the differential diagnoses of DMI. Consent ======= The patient\'s family was informed that data from the case would be submitted for publication and gave their consent. Conflict of Interests ===================== The authors declare that they have no conflict of interests. {#fig1} {#fig2} {#fig3} {#fig4} [^1]: Academic Editor: John Nyland

1. Introduction {#sec1-micromachines-10-00216} =============== Answering the question "What is life?" is a persistent challenge in science \[[@B1-micromachines-10-00216],[@B2-micromachines-10-00216]\]. In general, creating a system that is the same or similar to the system we want to understand can help us to deepen our knowledge of it. In this context, creating cell-like structures that mimic the functions of a living cell in a confined space (also known as artificial cells) may give us clues to answer the question. In addition, because artificial cells have a defined molecular composition, they can provide useful platforms to investigate biological phenomena under defined chemical conditions and can serve as tools for multiple biological applications, such as therapeutic techniques, fermentation by artificial cells, the construction of artificial red blood cells, and simplified cell models \[[@B3-micromachines-10-00216],[@B4-micromachines-10-00216],[@B5-micromachines-10-00216],[@B6-micromachines-10-00216]\]. Biochemical reactions inside compartments surrounded by a cellular membrane or at the cellular interface play an important role in regulating many biological functions. In the study of artificial cells, cellular functions are mimicked or reconstructed using the same or similar biochemical components. In the past few decades, artificial cells exhibiting a part of cellular functions including DNA replication, translation, transcription, division, and morphological changes, have been developed using artificial compartments \[[@B7-micromachines-10-00216],[@B8-micromachines-10-00216],[@B9-micromachines-10-00216],[@B10-micromachines-10-00216],[@B11-micromachines-10-00216],[@B12-micromachines-10-00216],[@B13-micromachines-10-00216],[@B14-micromachines-10-00216],[@B15-micromachines-10-00216],[@B16-micromachines-10-00216],[@B17-micromachines-10-00216],[@B18-micromachines-10-00216],[@B19-micromachines-10-00216],[@B20-micromachines-10-00216]\]. Many of the studies have employed water-in-oil (W/O) emulsions (or W/O droplet), polymersomes, or giant unilamellar vesicles (GUVs) as the compartments \[[@B7-micromachines-10-00216],[@B8-micromachines-10-00216],[@B9-micromachines-10-00216],[@B10-micromachines-10-00216],[@B11-micromachines-10-00216],[@B12-micromachines-10-00216],[@B13-micromachines-10-00216],[@B14-micromachines-10-00216],[@B15-micromachines-10-00216],[@B16-micromachines-10-00216],[@B17-micromachines-10-00216],[@B18-micromachines-10-00216],[@B19-micromachines-10-00216],[@B20-micromachines-10-00216],[@B21-micromachines-10-00216],[@B22-micromachines-10-00216],[@B23-micromachines-10-00216]\]. A W/O emulsion is a water drop in oil, which is covered with a monolayer of amphiphilic molecules such as artificial surfactants or lipid molecules. A polymersome is a vesicle comprising amphiphilic synthetic block copolymers. A GUV is a capsule structure formed by lipid bilayer membranes; it has been widely used as a compartment for artificial cells owing to its high similarity to the cellular membrane. Although these compartments can be constructed by simple methods, such as the mechanical agitation (sonication, pipetting, vortexing, or tapping) of a water drop in an oil, hydration of dry films of amphiphilic molecules, or emulsion phase transfer \[[@B24-micromachines-10-00216],[@B25-micromachines-10-00216],[@B26-micromachines-10-00216],[@B27-micromachines-10-00216]\], microfluidics-based fabrication of such compartments can further accelerate the study of artificial cells because of its technological benefits. Microfluidics is an interdisciplinary research field based on microfabrication technology, involving physics, chemistry, biology, and nanotechnology. It typically deals with small fluid volumes in a "chip" device or a tiny capillary \[[@B28-micromachines-10-00216],[@B29-micromachines-10-00216],[@B30-micromachines-10-00216]\]. In studies involving artificial cells, microfluidics provides a powerful means to efficiently produce a large number of compartments of a consistent size, within a scale ranging from a few dozen to several hundred micrometers. It is not easily achieved by traditional fabrication methods of the compartments. Furthermore, it enables the creation of multi-cellular compartments and the fabrication of compartments/reactors resembling living cells on the chip. These technological benefits in microfluidics will promote artificial-cell studies and lead to the design and construction of additional bio-inspired systems. In this review, we introduce and discuss the microfluidics technologies that can produce artificial cells. We first outline typical methods for creating the compartments and newer methods with further advantages. Then, we describe chemical communication inside multi-compartments or among multiple compartments. We also describe unique studies with microfluidics, including artificial cells containing artificial organelles, living cells as organelles for artificial cells, and on-chip artificial cells. Finally, we discuss future perspectives in the field. 2. Microfluidic Fabrication of Compartments for Artificial Cells {#sec2-micromachines-10-00216} ================================================================ 2.1. Typical Fabrication Methods {#sec2dot1-micromachines-10-00216} -------------------------------- Compartments with a defined size and producing a large number of the compartments allow investigations of biochemical reactions at defined volumes and efficient evaluations of the function of artificial cells. Microfluidics fabrication provides size controllability (several hundred nm to over hundreds µm) and a high-production rate (a few Hz to over thousands Hz) of the compartments. The methods used to create W/O emulsions using microfluidic devices were developed in the 2000s: the T-junction method \[[@B31-micromachines-10-00216]\], the flow-focus method \[[@B32-micromachines-10-00216]\], and the co-flowing method \[[@B33-micromachines-10-00216],[@B34-micromachines-10-00216]\] ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}a). All methods are categorized as droplet microfluidics technologies. Two immiscible phases, an aqueous phase containing components for artificial cells and an oil phase with surfactant molecules, flow in channels at designated flow rates. By adjusting these flow rates, W/O emulsions can be formed at a chosen size as a product of shear stress by the oil flow. Viscosity of oil and interfacial tension between two immiscible solutions are also important parameters for fabrication of the emulsions. Readers can gain a further understanding of the droplet microfluidics in previous review articles \[[@B30-micromachines-10-00216],[@B35-micromachines-10-00216],[@B36-micromachines-10-00216],[@B37-micromachines-10-00216]\], in which the usage of droplet microfluidics is described in more detail. In this section, we introduce typical methodologies employed to fabricate the compartments for artificial cells based on the droplet microfluidics. Droplet microfluidics technology can be used to fabricate not only W/O emulsions, but also other compartments for artificial cells, including multi-component double emulsions \[[@B38-micromachines-10-00216],[@B39-micromachines-10-00216],[@B40-micromachines-10-00216]\], droplet interface bilayers (DIBs) \[[@B41-micromachines-10-00216],[@B42-micromachines-10-00216],[@B43-micromachines-10-00216],[@B44-micromachines-10-00216],[@B45-micromachines-10-00216],[@B46-micromachines-10-00216]\], polymersomes \[[@B26-micromachines-10-00216],[@B47-micromachines-10-00216]\], and GUVs \[[@B48-micromachines-10-00216],[@B49-micromachines-10-00216],[@B50-micromachines-10-00216],[@B51-micromachines-10-00216],[@B52-micromachines-10-00216],[@B53-micromachines-10-00216],[@B54-micromachines-10-00216],[@B55-micromachines-10-00216],[@B56-micromachines-10-00216],[@B57-micromachines-10-00216],[@B58-micromachines-10-00216]\]. In the case of multi-compartment double emulsions, smaller emulsions are further encapsulated into larger emulsions. The methods reported by Adams et al., Che et al., and Chu et al. allow the encapsulation of a large number of smaller, inner emulsions into the outer emulsion \[[@B38-micromachines-10-00216],[@B39-micromachines-10-00216],[@B40-micromachines-10-00216]\]. Regarding DIBs, Elani et al. and Walsh et al. have reported DIBs in a W/O emulsion containing multiple droplets \[[@B41-micromachines-10-00216],[@B42-micromachines-10-00216]\]. In addition, Baxani et al. have developed a robust platform for creating artificial cells with a DIB \[[@B43-micromachines-10-00216]\], combining the T-junction and co-flowing methods ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}b). The chassis is covered with an alginate gel compartment containing droplets that formed the DIB. The chassis was found to be stable in an aqueous, an oil, and even in an air phase, and was also mechanically stable ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}b (ii) and (iii)). Polymersomes and GUVs are fabricated by similar approaches \[[@B26-micromachines-10-00216],[@B51-micromachines-10-00216]\]. In this review, we focus on GUVs more than on polymersomes because of the higher similarity of GUVs to cellular membranes. Researchers have developed multiple fabrication methods for GUVs, primarily W/O/W double emulsion-templated \[[@B48-micromachines-10-00216],[@B49-micromachines-10-00216],[@B50-micromachines-10-00216],[@B51-micromachines-10-00216]\], emulsion-transfer \[[@B52-micromachines-10-00216],[@B53-micromachines-10-00216],[@B54-micromachines-10-00216],[@B55-micromachines-10-00216],[@B56-micromachines-10-00216]\], and jetting methods \[[@B57-micromachines-10-00216],[@B58-micromachines-10-00216]\]. The double emulsion-templated method was developed by Shum et al. in 2008 \[[@B48-micromachines-10-00216]\]. The double emulsions were generated using a microcapillary fluidic device that combines the co-flowing and the flow-focus methods. Arriaga et al. have reported GUVs with ultrathin oil shells using the double emulsion method ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}c) \[[@B49-micromachines-10-00216]\]. A double emulsion with an ultrathin shell was created based on previous work \[[@B59-micromachines-10-00216]\] by the same group ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}c (i)). The ultrathin shell was achieved by exploiting the affinity of the fluid for the glass capillaries. The double emulsions with ultrathin oil shells were converted into GUVs with the formation of an oil pocket and lipid reservoir by the evaporation of organic solvents ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}c (ii)). After the formation of the unilamellar membrane, the GUVs exhibited phase separation of the membrane (membrane composition: DOPC/DPPC/cholesterol = 35:35:30 mol%) ([Figure 1](#micromachines-10-00216-f001){ref-type="fig"}c (iii)). Using a similar approach, Deshpande et al. have demonstrated that GUVs were spontaneously formed from double emulsions ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}d) \[[@B50-micromachines-10-00216]\]. They used 1-octanol as the oil phase for lipids. The 1-octanol oil pocket was formed in lipid vesicles, but it completely budded off from the double emulsions owing to the minimization of interfacial energy, resulting in the formation of GUVs. An emulsion transfer method was first reported by Pautot et al. in 2003 using a test tube and a centrifuge, which enabled them to fabricate GUVs with asymmetric bilayer membranes \[[@B60-micromachines-10-00216],[@B61-micromachines-10-00216]\]. This technique has been expanded to microfluidics to prepare GUVs. Abkarian et al. have developed a continuous-droplet interface-crossing encapsulation (cDICE) method ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}e) \[[@B52-micromachines-10-00216]\], consisting of a glass capillary and a centrifuge cylindrical chamber with a hole in the lid. Dispersing aqueous solution (DAS), lower density lipid-in-oil solution (LOS) (mineral oil), and decane were poured into the rotating-cylindrical chamber in this order. The encapsulated aqueous solution (EAS) was injected from the capillary in the decane phase. W/O emulsions were formed in the decane and moved to the LOS phase, where they were covered with a lipid monolayer. GUVs were formed when the W/O emulsions passed through the LOS/DAS interface, where a lipid monolayer was added. Detailed characteristics of the cDICE methods were further investigated by Blosser et al. \[[@B53-micromachines-10-00216]\]. In addition to the cDICE method, Hu et al. have reported GUV production using emulsion transfer, in which W/O emulsions were fabricated using microfluidics and then poured onto a lipid monolayer at an oil/water interface \[[@B54-micromachines-10-00216]\]. Furthermore, Matosevic et al. and Karamdad et al. have reported microfluidic devices that can produce GUVs from W/O emulsions in a single microfluidic chip using emulsion transfer \[[@B55-micromachines-10-00216],[@B56-micromachines-10-00216]\]. In the jetting method, vesicles were generated by applying a pulse jet flow against a planer lipid bilayer from a small glass capillary \[[@B57-micromachines-10-00216],[@B58-micromachines-10-00216]\]. The pulse jet induced prominence of the lipid bilayer and the prominence was detached from the bilayer, resulting in the formation of vesicles. In addition to the methods described above, other approaches have also been investigated. For example, Ota et al. developed a unique method to generate GUVs \[[@B62-micromachines-10-00216]\]. Lipid bilayers were formed in a small microfluidic chamber that was connected to a large chamber. Outward flow from the large to the small chamber was generated by the irradiation of a laser to the large chamber. The bilayer-covered solution was pushed from the small chamber and shear forces from a continuous fluid stream led to the fission of the bilayer-covered solution, resulting in the generation of GUVs. In a method reported by Tan et al., W/O lipid emulsions in oleic acid prepared by a microfluidic device were poured in an ethanol/water mixture \[[@B63-micromachines-10-00216]\]. The oleic acid dissolved in ethanol, and lipids were forced to rearrange around the emulsion to assemble into GUVs. Matosevic et al. developed a layer-by-layer approach in which W/O emulsions trapped in the microfluidic chamber were further covered with a lipid monolayer \[[@B64-micromachines-10-00216]\]. It resulted in the formation of GUVs and allowed multilamellar vesicles to form. Readers can find more details of microfluidic-fabrication methods for lipid vesicles in a review article \[[@B65-micromachines-10-00216]\]. In the methods described in this section, microfluidic fabrication of the compartments for artificial cells contributes advantages including highly efficient production, monodispersity, size-controllability, and well-encapsulated yields of components into the inner aqueous phase. 2.2. Novel Microfluidic Fabrication Methods for Compartments with Additional Advantages {#sec2dot2-micromachines-10-00216} --------------------------------------------------------------------------------------- In the past few years, attempts to improve microfluidics technologies have followed several routes: simplicity at a low cost, stability of the compartments during the fabrication process, asymmetricity of the lipid bilayer without an organic solvent, and reduction of the consumed sample volume. Here, we summarize recent studies improving microfluidics technologies for artificial cell studies. Valet et al. developed a simple method to produce aqueous droplets in an oil, which they named the "capillary-trap method" ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}a) \[[@B66-micromachines-10-00216]\]. In their device, a glass capillary is cyclically moved up and down across the oil-air interface (left in [Figure 2](#micromachines-10-00216-f002){ref-type="fig"}a). The capillary is immersed in the oil and an aqueous droplet is formed by flowing the aqueous solution in the capillary. When the capillary is vertically pulled out, the droplet detaches from the capillary because of the minimization of surface energy in the oil-water-air phase (right in [Figure 2](#micromachines-10-00216-f002){ref-type="fig"}a). This method produced droplets of a minimum radius of approximately 30 µm. The authors stated that this method can produce about 1 µm droplets in principle. In addition to the development of the method, the authors also developed a low-cost and easy-to-implement setup for their technique. According to the authors' evaluation, the maximum cost is about a few hundred euros \[[@B66-micromachines-10-00216]\]. Although GUVs are often used due to their similarities to biological membranes, their chemical and mechanical instabilities make the creation of complex and functionalized artificial cells difficult. Weiss et al. addressed these points by developing a microfluidic method to generate stable GUVs termed "droplet-stabilized giant unilamellar vesicles (dsGUVs)" \[[@B67-micromachines-10-00216]\] ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}b). In their method, a continuous lipid bilayer is formed at the inner interface of an amphiphilic block/copolymer-stabilized water-in-oil droplet. Multiple biomolecules can be loaded into dsGUVs using picoinjection \[[@B68-micromachines-10-00216]\]. The authors have demonstrated the loading of transmembrane proteins (F~1~F~o~ ATP-synthase and integrin) and cytoskeletal proteins (G-actin and tubulin) into dsGUVs. In addition, the authors showed that GUVs in stabilizing polymer droplet shells can be released into an aqueous phase from the oil phase. They designed a microfluidic chamber with passive trapping structures. The dsGUVs are separated from each other at the T-junction and flow in the oil with destabilizing surfactants; then, the dsGUVs are decelerated at the trapping structures and gradually approach the oil-water interface. When the destabilized droplets contact the interface, GUVs are released into the aqueous phase. This release mechanism can be applied to dsGUVs containing actin filaments (bottom right in [Figure 2](#micromachines-10-00216-f002){ref-type="fig"}b) and integrin. Eukaryotic cellular membranes have asymmetrical inner and outer leaflets. This asymmetry is hypothesized to play an important role in biological processes \[[@B69-micromachines-10-00216],[@B70-micromachines-10-00216]\]. Therefore, artificial cells with asymmetric lipid bilayers can serve as a platform to investigate the role of the asymmetricity in living cells. Pautot et al., Hu et al., and Funakoshi et al. have created asymmetric GUVs using microfluidics; however, in their methods, a large amount of organic solvents remained in the bilayer, which will affect important membrane dynamics, such as lipid flip-flop phenomena. Kamiya et al. developed a method to fabricate asymmetric GUVs containing little organic solvent in their membranes \[[@B71-micromachines-10-00216]\] with a pulse-jetting method \[[@B57-micromachines-10-00216]\] ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}c). They prepared a planar asymmetric lipid bilayer membrane using a droplet contact method \[[@B72-micromachines-10-00216]\], and microfluidic flow was applied to the membrane, resulting in the formation of a lipid tube ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}c (i)). They found that large (100--200 µm) and small (3--20 µm) GUVs were formed during the deformation process of the lipid tube due to Rayleigh--Plateau instability ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}c (ii)). Using a Raman scattering microscope, they showed that small GUVs with the asymmetric bilayer contained only a very small amount of organic solvents. They also investigated the time-course of flip-flop dynamics of lipid molecules and influence of peptides on the flip-flop. Microfluidics generally requires a large amount of sample solution because of tubing connected to a microfluidic chamber, adjustment of the flow rate, or preparing steady flow. This is an important issue since artificial cells often include rare and expensive reagents. Thus, reduction of the sample volume is one of the important issues in microfluidics. A novel method for the production of GUVs (or W/O emulsion) in extremely small sample volumes (0.5 µL) has been devised by Morita et al. \[[@B73-micromachines-10-00216]\]. They fabricated a centrifugal capillary-based microfluidic device employing the W/O emulsion transfer method. Water microdroplets were discharged from the tip of a glass capillary toward an oil phase containing lipids, and the lipids stabilized the W/O emulsion transfer through an oil-water interface, creating GUVs ([Figure 2](#micromachines-10-00216-f002){ref-type="fig"}d). Although large (\~100 mm) and small (\~15 µm) droplets were discharged from the tip, only cell-sized GUVs were formed due to the size-filtration effect. The detailed mechanism is described in the article. They called this method "droplet-shooting and size-filtration" (DSSF). This method allows the production of hundreds of GUVs from a single microliter of sample without wastage. Such low-sample volume microfluidic techniques will further accelerate the production of artificial cells. 3. Communication Among Artificial Cells {#sec3-micromachines-10-00216} ======================================= Intercellular chemical communication, in which secreted molecules are diffused or are transported to other cells, regulates multiple cellular functions, including differentiation and morphogenesis \[[@B74-micromachines-10-00216],[@B75-micromachines-10-00216]\]. This has important roles, not only in multicellular organisms, but also in the collective behavior of single-celled organisms \[[@B76-micromachines-10-00216]\]. Therefore, communication between artificial cells and communication between artificial cells and living cells has the potential to lead to the construction of artificial tissue structures, artificial cells capable of controlling their function based on individual interactions, and "development of life like technology" \[[@B77-micromachines-10-00216]\]. Lipid vesicle networks connected via lipid nanotubes \[[@B78-micromachines-10-00216],[@B79-micromachines-10-00216]\] and a multi-compartment of W/O emulsions \[[@B80-micromachines-10-00216]\] have been fabricated by manual manipulation. Moreover, microfluidics has the potential to automate the fabrication of multicellular structures with a large number of single compartments containing arbitrary molecules in each compartment \[[@B81-micromachines-10-00216],[@B82-micromachines-10-00216],[@B83-micromachines-10-00216],[@B84-micromachines-10-00216]\]. In this section, we describe chemical communication among artificial cells based on microfluidics technology. Elani et al. have reported a construction method for multi-compartment vesicles based on the W/O emulsion \[[@B85-micromachines-10-00216],[@B86-micromachines-10-00216]\]. Multiple W/O droplets were tandemly loaded in a single tube and deposited onto interfacial lipid monolayers from the oil phase, resulting in the formation of a multi-compartment through phase transfer ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}a (i)) \[[@B86-micromachines-10-00216]\]. The authors demonstrated a cascading chemical reaction in this system, in which each reaction step was isolated in a distinct compartment by a lipid bilayer ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}a (ii)). The dewetting phenomenon of W/O/W double emulsions can also be used to fabricate multicellular structures \[[@B83-micromachines-10-00216],[@B87-micromachines-10-00216]\]. Deng et al. have developed a method for assembling multicompartment vesicles from a double emulsion ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}b) \[[@B87-micromachines-10-00216]\]. The double emulsion was fabricated using a microfluidic device combining co-flow and flow-focusing methods ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}b (i)). The oil phase and outer aqueous phase (W2 in [Figure 3](#micromachines-10-00216-f003){ref-type="fig"}b) contain egg L-α-phosphatidylcholine and a triblock copolymer, Pluronic F-68. They demonstrated that the spontaneous dewetting process of double emulsions can be controlled by adjusting the interfacial tension between aqueous and oil phases, utilizing a triblock copolymer to ensure complete dewetting of the double emulsion ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}b (ii)). The authors also demonstrated molecular transfer to the outer aqueous solution through the multi-compartments ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}b (iii)). They loaded small and large fluorescent molecules (calcein and RTIC-Dextran, respectively) to each compartment, where αHL pores were formed at the one compartment loading large molecules. As a result, small molecules passed through to another compartment and eventually diffused to the outer solution, while the large fluorescent molecule was retained. H. Bayley's group successfully constructed a huge tissue-like multicompartment by printing a W/O emulsion \[[@B88-micromachines-10-00216],[@B89-micromachines-10-00216],[@B90-micromachines-10-00216],[@B91-micromachines-10-00216]\]. They demonstrated that each compartment comprising the artificial tissue can communicate with others via pores. In addition to communications between artificial molecular components, chemical communication between a cell-free system and living organisms (bacteria) has been achieved ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}c) \[[@B92-micromachines-10-00216]\]. Schwarz-Schilling et al. used amphiphilic molecules of N-acyl-l-homoserine lactones (AHLs) and isopropyl-β-d-thio-galactopyranoside (IPTG) as communication signals among W/O emulsions. They designed a genetic AND gate that responded to AHL and IPTG, leading to the expression of the green fluorescent protein (GFP) and a sender circuit that produced AHL when stimulated by the IPTG signal. The AND gate or sender plasmid was cloned in Escherichia coli (*E. coli*) or mixed with a cell-free gene expression system. The *E. coli* and the cell-free system were encapsulated into separate droplets, which were then loaded into a capillary ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}c (i)). They demonstrated communication between the cell-free system and bacteria using their AND gate and sender circuits ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}c (ii)). IPTG, cell-free sender circuits, and the bacteria AND gate were encapsulated into droplets, which were shown with red, no-color (shown as "S" in the figure), and green color markers, respectively. In this system, IPTG diffused to both bacterial and the sender-circuit droplets. The sender droplets that received IPTGs produced AHLs and the produced AHLs diffused to other droplets, including the bacterial droplets. When IPTGs and the produced AHLs diffused into the bacterial droplets, the AND gate was switched and the bacteria expressed GFP. Furthermore, the direction of communication could be reversed by encapsulating bacteria with sender circuits and the cell-free system with the AND gate into droplets. The GFP expression level depended on the distance from IPTG and sender droplets because the concentration of diffused molecules in droplets depends on their distance from the diffusion source \[[@B92-micromachines-10-00216]\]. Quorum sensing (QS) is a common mode of chemical communication in bacteria for controlling specific gene transcription and collective behavior, depending on population density \[[@B93-micromachines-10-00216]\]. Since microfluidics can generate a large number of cell-like compartments of controlled sizes, it enables the mimicking of QS-type chemical communication \[[@B94-micromachines-10-00216],[@B95-micromachines-10-00216]\]. Niederholtmeyer et al. have shown this using artificial nuclei ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}d) \[[@B95-micromachines-10-00216]\]. They fabricated artificial cells with a flow-focusing method ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}d (i)). These artificial cells had clay-DNA nuclei with genetic information. A porous polymer membrane allowed the introduction of large molecules, such as cell-free transcription/translation reagents, from the outside to inside. The authors demonstrated that the artificial cells exchanged proteins with their neighbors ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}d (ii)). They prepared activator and reporter artificial cells containing gene templates for T3 RNA polymerase (RNAP) and for the T3 RNAP-driven synthesis of the TetR-sfGFP reporter, as well as a tetO array plasmid. The T3 RNAPs produced from the activator were successfully transferred to the reporter, which exhibited green fluorescence. By using a similar technique, the authors produced artificial QS ([Figure 3](#micromachines-10-00216-f003){ref-type="fig"}d (iii)). The artificial cells containing both the activation and reporter constructs exhibited fluorescence as a function of density. The threshold density they observed was 400 artificial cells in 4.5 µL volume. 4. Artificial Organelle in Artificial Cells {#sec4-micromachines-10-00216} =========================================== In a typical approach toward the creation of artificial cells, components are mixed in a batch reaction solution; however, living cells spatially divide functional modules as organelles including the nucleus, Golgi apparatus, endoplasmic reticulum, and others. Construction of functional modules in a cell-mimicking compartment is a promising strategy for the realization of artificial cells. Many researchers have demonstrated the usefulness of microfluidics for the creation of artificial organelles in artificial cells \[[@B96-micromachines-10-00216],[@B97-micromachines-10-00216],[@B98-micromachines-10-00216],[@B99-micromachines-10-00216],[@B100-micromachines-10-00216],[@B101-micromachines-10-00216],[@B102-micromachines-10-00216]\]. Deng et al. fabricated liposome-in-liposomes (vesosomes) using microfluidics and the dewetting mechanism \[[@B96-micromachines-10-00216]\]. The inner and outer liposomes in the vesosomes contain an in vitro translation system and in vitro translation/transcription system, respectively ([Figure 4](#micromachines-10-00216-f004){ref-type="fig"}a). The liposome successfully translated RNA only in the inner, "nuclear" liposome and expressed monomeric red fluorescent protein (mRFP) in the outer, "cytoplasmic" liposome. Multiple artificial organelles in artificial cell models have also been demonstrated. L. Aufinger and F. C. Simmel developed hydrogel-based artificial organelles ([Figure 4](#micromachines-10-00216-f004){ref-type="fig"}b) \[[@B98-micromachines-10-00216]\]. They immobilized gene-length DNA in agarose gels and prepared the gel-based artificial organelles using microfluidics. The organelles were equipped with different functions: gene transcription, translation, and RNA localization. They investigated each artificial organelle function and demonstrated translation and transcription by the organelles in W/O droplets. This gel-based organelle fabrication approach can provide enzymatic cascade reactions via each organelle. Wang et al. have recently reported a multienzyme system with this feature in hydrogel microcapsules \[[@B99-micromachines-10-00216]\]. They immobilized enzymes in inverse opal particles (IOPs), which were encapsulated into a hollow alginate hydrogel ([Figure 4](#micromachines-10-00216-f004){ref-type="fig"}c). The multienzyme cascade reaction was demonstrated using horseradish peroxidase, ß-glucosidase, and glucose oxidase, which were immobilized in different IOPs in the capsule. As reaction substrates, octyl ß-D-glucopyranoside, o-dianisidine, and hydrogen peroxide were used. Detailed cascade reaction pathways were described in the article \[[@B99-micromachines-10-00216]\]. Interestingly, this cascade reaction within the microcapsule exhibited a higher performance than in the absence of encapsulation. The authors explained that this was due to the shorter distance between the IOPs in the capsule. The use of living cells as organelles for artificial cells has also been reported \[[@B100-micromachines-10-00216]\]. Encapsulation of living cells into a compartment such as a W/O droplet has been widely studied because such techniques can be used for the high-throughput screening of cells \[[@B103-micromachines-10-00216],[@B104-micromachines-10-00216],[@B105-micromachines-10-00216]\]. Elani et al. constructed GUVs with embedded living cells as organelles \[[@B100-micromachines-10-00216]\]. The authors encapsulated living cells into GUVs using a microfluidics technique, and the embedded cells served an organelle-like function in a cascading enzymatic reaction ([Figure 4](#micromachines-10-00216-f004){ref-type="fig"}d). The reaction was demonstrated in a hybrid cellular bionic system. In addition to the organelle-like behavior, they showed cell growth in the GUVs. Morita et al. also showed bacterial growth in GUVs fabricated by a microfluidic device \[[@B101-micromachines-10-00216]\]. These results pave the way for the creation of living cell-artificial cell hybrid organelle systems. 5. On-chip Bio-inspired Molecular System {#sec5-micromachines-10-00216} ======================================== In the above sections, we described artificial-cell studies based on cell-mimicking compartments fabricated using microfluidic devices. Microfluidics can also provide a confined space for mimicking biological reactions. Therefore, a combination of biochemical reactions and microfluidic devices can produce artificial cells that work "on-chip". In fact, some researchers have created such artificial cells or cell-mimicking biochemical reactions in microfluidic channels. Bar-Ziv et al. created on-chip biological systems capable of mimicking metabolism, programmable protein expression, communication, synchronization, and pattern formation \[[@B106-micromachines-10-00216],[@B107-micromachines-10-00216],[@B108-micromachines-10-00216]\]. In their system, DNA-encoding genetic networks were patterned on the surface of circular compartments carved in silicon ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}a) \[[@B106-micromachines-10-00216]\]. The DNA compartments were connected to a main channel in which cell-free extracts flowed via a narrow diffusive channel. This design allowed gene-network dynamics to be controlled, depending on the compartment geometry and biological reaction; i.e., controlling reaction-diffusion dynamics. The constructs for an oscillatory gene network were DNA in compartments. The authors demonstrated the oscillation of protein expression in 15 different compartments ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}a (i)). Furthermore, to couple the oscillations, the authors connected each compartment at an appropriate geometry, which allowed the diffusive transportation of newly synthesized proteins ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}a (ii)). As a result, in the 15 compartments, the frequency and phase of protein expression were successfully synchronized ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}a (ii)). This research group has also reported the propagation of gene expression on similar on-chip artificial cells \[[@B107-micromachines-10-00216]\]. They designed a one-dimensional array of DNA compartments with a bistable genetic circuit for propagation. They showed front propagation of gene expression along the artificial cell arrays based on reaction-diffusion dynamics and investigated the behavior of the gene circuit at monostable, bistable, and transition points. Such reaction-diffusion dynamics have an important role in pattern formation during development. The on-chip biological system exhibited the pattern formation of gene expression \[[@B106-micromachines-10-00216]\]. In addition, Zadorin et al. have shown a French flag pattern formed by a DNA chemical reaction network in a flow channel \[[@B109-micromachines-10-00216]\]. The reaction network comprised the PEN DNA tool box \[[@B110-micromachines-10-00216]\]. They demonstrated that two orthogonal bistable chemical reaction networks could produce a French flag pattern of DNA concentration in the flow chamber. Besides reaction-diffusion dynamics on a chip, electrical control of microfluidic channels provides a powerful means to investigate and mimic biochemical reactions \[[@B111-micromachines-10-00216],[@B112-micromachines-10-00216],[@B113-micromachines-10-00216],[@B114-micromachines-10-00216],[@B115-micromachines-10-00216]\]. Niederholtmeyer et al. have developed a microfluidic device in which an in vitro translation/transcription system functions at a steady state by exchanging the solution via multi-flow channels and computer-controlled valves ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}b) \[[@B113-micromachines-10-00216]\]. The microfluidic device has circular reactors, with the inlet and outlet controlled by a computer ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}b (i)). They confirmed that their device can keep the system at a steady state and that the concentration of each reaction component can be controlled by regulation parameters, including replaced-volume and dilution rate and residence time. The authors designed a biological oscillator network ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}b (ii)) and investigated its oscillation behavior. As they expected, steady-state conditions were necessary to produce oscillations. The oscillator network exhibited multiple states: damped oscillations, a stable steady state, oscillations with different periods, and one peak pulse ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}b (iii)). Sugiura et al. have addressed chemical oscillation far from equilibrium using a droplet open-reactor system in a microfluidic system ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}c) \[[@B114-micromachines-10-00216]\]. Their method is similar to the molecular transportation systems in living cells; namely, chemical fluxes were controlled by the fusion/fission of transporter droplets regulated by electrical stimulation ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}c (i)). The detailed fusion/fission methods for the reactor system were described in the previous article \[[@B116-micromachines-10-00216]\]. The authors investigated the controllability of chemical reaction dynamics far from equilibrium using the system of bromate-sulfite-ferrocyanide pH oscillation \[[@B117-micromachines-10-00216]\]. They confirmed that this reaction generated pH oscillation when chemical fluxes were maintained appropriately ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}c (ii)). Furthermore, autonomous feedback control of the chemical reaction was demonstrated using a feedback control system ([Figure 5](#micromachines-10-00216-f005){ref-type="fig"}c (iii)). The state of the reactor was monitored and classified, and then, the fusion/fission interval of the transporter was changed to the designated one. As shown in the figure, the state of the chemical reaction was successfully adjusted to the oscillation state from the steady state in the droplet open-reactor system. 6. Conclusions and Future Perspectives {#sec6-micromachines-10-00216} ====================================== In this review, we have described microfluidics as a powerful technology for the study of artificial cells, particularly for the creation of compartments for artificial cells because of its high production efficiency and size-controllability. Although there are still significant differences between the living cellular membrane and artificial compartments, microfluidics techniques have the potential to fill the gap. As we described in this review, microfluidics provides methods for the fabrication of oil-less GUVs, asymmetric membranes, and protein-embedded compartments. These methods will lead to the creation of compartments more similar to living cells. We have also noted that microfluidics enables us to fabricate multi-compartments, liposome-in-liposomes, and compartments containing artificial organelles. The contribution of microfluidics techniques to fabricating multi-compartments containing artificial organelles will be key for the construction of artificial tissues, which is one of the practical applications of artificial cells. The studies of on-chip artificial cells described in this review aimed to mimic cellular functions. Although there are some differences at the boundary between the on-chip systems and the vesicle-based systems, such as the diffusion of molecules on the surface of the boundary, covering the chip with biological membranes (lipid monolayers or supported lipid bilayers) will fulfill the gap and expand the applicability of the on-chip artificial cells. Because therapeutic applications of artificial cells have been proposed \[[@B3-micromachines-10-00216]\], on-chip artificial cells are likely to be applicable for biomedical devices, such as diagnosis, therapeutic, or monitoring devices. A parallel research field to that of artificial cells has recently emerged, called molecular robotics. In this field, researchers aim to create artificial molecular systems comparable to or outperforming the corresponding natural living system \[[@B118-micromachines-10-00216],[@B119-micromachines-10-00216]\]. The realization of conformational changes and locomotion are two of the major targets in this field. From the perspectives of the reconstruction of cells and molecular robotics, researchers have so far reported locomotion and shape changes of W/O droplets \[[@B120-micromachines-10-00216],[@B121-micromachines-10-00216],[@B122-micromachines-10-00216]\], oil droplets in water \[[@B123-micromachines-10-00216]\], and GUVs \[[@B12-micromachines-10-00216],[@B13-micromachines-10-00216],[@B14-micromachines-10-00216],[@B15-micromachines-10-00216],[@B124-micromachines-10-00216],[@B125-micromachines-10-00216],[@B126-micromachines-10-00216]\] ([Figure 6](#micromachines-10-00216-f006){ref-type="fig"}a,b). In 2017, the robotic behavior of GUVs, as in sensing signal molecules, processing the signals, and exhibiting actuation, was achieved. Sato et al. showed an amoeba-type molecular robot that changes shape in response to signal molecules produced inside the robot via photo-irradiation ([Figure 6](#micromachines-10-00216-f006){ref-type="fig"}c) \[[@B127-micromachines-10-00216]\]. As described above, GUVs are utilized as a chassis for the molecular robot \[[@B119-micromachines-10-00216],[@B127-micromachines-10-00216],[@B128-micromachines-10-00216]\]. Therefore, microfluidics will also contribute to the creation of the chassis for molecular robots, as well as artificial cells and their subcompartments. In addition, we expect two additional contributions of microfluidics to molecular robotics. The first is as an assembly factory. The pico-injection method \[[@B68-micromachines-10-00216]\] will allow us to sequentially assemble robot parts in the intended order in the W/O emulsions (GUV precursors) flowing in microfluidic channels. Sorting techniques for a single cell in microfluidic devices \[[@B129-micromachines-10-00216]\] can serve as a quality control system for removing misassembled robots. Thus, microfluidic devices with injectors and sorters could make a large contribution to the development of molecular robotics. The second is the fabrication of testing grounds to evaluate a robot's function. Potential applications of molecular robots are in biomedical and pharmacological fields \[[@B118-micromachines-10-00216],[@B119-micromachines-10-00216]\]. Microfluidic cell-culture systems \[[@B130-micromachines-10-00216]\] and blood-vessel mimics on a chip \[[@B131-micromachines-10-00216]\] would be applicable as testing grounds (mimics of inside the body) for molecular robots having medical functions to investigate the efficacy of drug transport to the cells, to study the effects of the stimulation of cells by robots, and others. Such uses of microfluidics can also be adapted to the future application of artificial cells in biomedical fields. In conclusion, microfluidics technology has clearly promoted the study of artificial cells. Although the early stage of microfluidics only provided cell-sized compartments, the great progress in the past years allows us to more closely study artificial cells. The progress has accelerated and contributed to the development of biological and engineering fields. We believe that microfluidics will become an indispensable technology from fundamentals to applications in multiple fields, including synthetic biology, artificial cell studies, and molecular robotics. Y.S. designed and wrote the manuscript. M.T. critically reviewed the manuscript. This research was supported by JSPS KAKENHI Grant Numbers No. JP18J00720, JP17H01813, JP18K19834. Y.S. is a JSPS Research Fellow (SPD). The authors declare no conflict of interest. ![Methods for the production of compartments for artificial cells. (**a**) Typical design of flow channels for the fabrication of W/O emulsions: T-junction, flow-focusing, and co-flowing methods from left to right. (**b**) Droplet interface bilayers (DIBs) from the contact of two W/O emulsions in the large W/O emulsion covered with a gel-shell \[[@B43-micromachines-10-00216]\]. (i) Illustration of microfluidic fabrication of gel-encapsulated DIBs. (ii) Schematic illustrations of the gel-encapsulated DIBs in an aqueous, an oil, and an air phase. (iii) Gel-encapsulated DIBs in aqueous, oil, and air phase and manipulated using a tweezer. Scale bars: 1 mm. Reproduced with permission from \[[@B43-micromachines-10-00216]\]. (**c**) Fabrication of giant unilamellar vesicles (GUVs) with an ultrathin oil shell using the double emulsion method \[[@B49-micromachines-10-00216]\]. (i) An illustration and image for the fabrication of double emulsions with an ultrathin shell. (ii) Illustrations and microscopic images of GUVs with ultrathin shells. (iii) Phase separation of the GUV prepared by this method. Red and cyan colors indicate liquid-disordered and liquid-ordered phases, respectively. Reproduced with permission from \[[@B49-micromachines-10-00216]\]. (**d**) Fabrication of GUVs by the octanol-assisted method \[[@B50-micromachines-10-00216]\]. Reproduced with permission from \[[@B50-micromachines-10-00216]\]. (**e**) A continuous droplet interface crossing encapsulation (cDICE) method for GUV fabrication \[[@B52-micromachines-10-00216]\]. Scale bars: 10 µm. Reproduced with permission from \[[@B52-micromachines-10-00216]\].](micromachines-10-00216-g001){#micromachines-10-00216-f001} ![Novel and unique microfluidic technologies improving the fabrication of artificial cell-compartments. (**a**) A low-cost capillary-based fabrication method of W/O emulsion \[[@B66-micromachines-10-00216]\]. (i) An illustration of the low-cost setup. (ii) Snapshots of processes of droplet fabrication. Scale bar: 400 µm. Reproduced with permission from \[[@B66-micromachines-10-00216]\]. (**b**) A microfluidic chip for the fabrication of droplet-stabilized giant unilamellar vesicles (dsGUVs) and for the release of GUVs from the droplet \[[@B67-micromachines-10-00216]\]. (i) A schematic of the dsGUV. (ii) The microfluidic device used to release the GUVs from the surrounding stabilizing polymers into an aqueous phase. (iii) A fluorescence microscopic image of GUVs encapsulating actin filaments (red). Scale bars: 20 µm. Reproduced with permission from \[[@B67-micromachines-10-00216]\]. (**c**) A jetting-based fabrication method for asymmetric GUVs containing little organic solvent in the membrane \[[@B71-micromachines-10-00216]\]. (i) and (ii) Schematics and microscopic images of the formation process, reproduced with permission from \[[@B71-micromachines-10-00216]\]. (**d**) Droplet-shooting and size-filtration (DSSF) method for the fabrication of the compartment \[[@B73-micromachines-10-00216]\]. (i) Schematics of the device and formation process of GUVs. (ii) Differential interference contrast (DIC) and confocal laser scanning microscopic (CLSM) images of GUVs, and size-distribution of GUVs. The white arrow in the DIC image indicates a multilamellar vesicle. Reproduced with permission from \[[@B73-micromachines-10-00216]\].](micromachines-10-00216-g002){#micromachines-10-00216-f002} ![(**a**) Multi-compartment formation via the phase transfer of multiple droplets and cascadic chemical reaction in the compartment \[[@B86-micromachines-10-00216]\]. (i) Schematics of the fabrication method and a microscopic image of a multi-compartment vesicle. (ii) Three-step cascadic-chemical reaction in the multi-compartment. Scale bar in microscopic images: 250 µm. Reproduced with permission from \[[@B86-micromachines-10-00216]\]. (**b**) Multi-compartment GUVs and selective molecular release from them \[[@B87-micromachines-10-00216]\]. (i) Schematics and microscopic images of the fabrication process of the multi-compartments via dewetting of W/O emulsions. (ii) Schematics and microscopic images of the multi-compartments consisting of two, three, four, and five compartments. (iii) Size-dependent molecular release from the multi-compartment to an outer solution through alpha-hemolysin pores. Reproduced with permission from \[[@B87-micromachines-10-00216]\]. (**c**) Chemical communication between a cell-free system and bacteria \[[@B92-micromachines-10-00216]\]. (i) Schematics of the experimental system. (ii) A kymograph of a fluorescence microscopic image of chemical communication between cell-free and bacteria droplets. Reproduced with permission from \[[@B92-micromachines-10-00216]\]. (**d**) Quorum sensing (QS) of artificial cells \[[@B95-micromachines-10-00216]\]. (i) Microfluidic fabrication of artificial cells capable of gene expression in the nuclei. (ii) Protein exchanges between neighboring artificial cells. (iii) QS of artificial cells with fluorescence accumulation, depending on the artificial cell density. Reproduced from \[[@B95-micromachines-10-00216]\].](micromachines-10-00216-g003){#micromachines-10-00216-f003} ![(**a**) Schematics of the fabrication method for "liposome-in-liposome", and in vitro transcription and transcription/translation in "artificial nuclei" and "cytoplasm" \[[@B96-micromachines-10-00216]\]. Scale bar: 50 µm. Reproduced with permission from \[[@B96-micromachines-10-00216]\]. (**b**) Artificial gel-based organelles in W/O emulsion that exhibit transcription and translation in each artificial organelle \[[@B98-micromachines-10-00216]\]. Reproduced with permission from \[[@B98-micromachines-10-00216]\]. (**c**) Schematics of a biomimetic enzyme cascade reaction in the microcapsule and a microscopic image of the microcapsule with three-cores \[[@B99-micromachines-10-00216]\]. Scale bar: 400 µm. Reproduced with permission from \[[@B99-micromachines-10-00216]\]. (**d**) A schematic of the enzymatic pathway containing a living cell as an organelle-like bioreactor and sequential microscopic images representing a successful reaction with the aid of the cell \[[@B100-micromachines-10-00216]\]. Reproduced with permission from \[[@B100-micromachines-10-00216]\].](micromachines-10-00216-g004){#micromachines-10-00216-f004} ![(**a**) On-chip DNA compartments as artificial cells \[[@B106-micromachines-10-00216]\]. (i) An image of expressed green fluorescence protein (GFP) (green) and DNA patterns (white), and a graph of oscillation dynamics in 15 different compartments. Scale bar: 100 µm. (ii) An image of three coupled compartments and a graph of the synchronized oscillations in the coupled compartments. Reproduced with permission from \[[@B106-micromachines-10-00216]\]. (**b**) In vitro transcription/translation system in a microfluidic nanoreactor \[[@B113-micromachines-10-00216]\]. (i) and (ii) Schematics of the nanoreactor (i) and an oscillator network (ii). (iii) Oscillation behavior of the network in the nanoreactor controlled by microfluidic parameters. Reproduced with permission from \[[@B113-micromachines-10-00216]\]. (**c**) Chemical reactions far from equilibrium in a droplet open-reactor system \[[@B114-micromachines-10-00216]\]. (i) Schematic diagrams of the experimental system in which fusion/fission of droplets were electrically controlled. (ii) pH Oscillation reaction depending on the ratio of fusion states in the fusion--fission process (shown as *q* in the figure). Scale bar: 500 µm. (iii) Feedback control of pH oscillation using the droplet open reactor system: scheme of feedback control (left) and time course of *q* and oscillation behavior during the feedback control. Reproduced with permission from \[[@B114-micromachines-10-00216]\].](micromachines-10-00216-g005){#micromachines-10-00216-f005} ![(**a**) Autonomous motility of the W/O droplet containing microtubule-kinesin bundles \[[@B120-micromachines-10-00216]\]. A bright-field image of the droplet (left) and a fluorescence image of microtubules (right). Scale bars: 80 µm (left) and 100 µm (right). Reproduced with permission from \[[@B120-micromachines-10-00216]\]. (**b**) Morphological changes of GUVs encapsulating tubulins that are polymerized to microtubules in response to hydrostatic pressures \[[@B14-micromachines-10-00216]\]. Reproduced with permission from \[[@B14-micromachines-10-00216]\]. (**c**) An amoeba-type molecular robot capable of changing its shape in response to signal DNA molecules \[[@B127-micromachines-10-00216]\]. Reproduced with permission from \[[@B127-micromachines-10-00216]\].](micromachines-10-00216-g006){#micromachines-10-00216-f006}

{#sp1 .35} {#sp2 .36}

Introduction {#s1} ============ Common Neuropsychiatric Disorders (NPDs) are a major cause of morbidity and impairments in the quality of life around the world,^[@R1]^ with very large societal costs in the order of trillions of dollars.^[@R2]^ Several common NPDs (such as schizophrenia, bipolar disorder, Alzheimer's disease) have large heritabilities, with a possible shared pathophysiology for some of them, highlighting the importance of the study of genetic factors.^[@R3],[@R4]^ In recent years, dozens of genome-wide association studies (GWASs) have been carried out for NPDs.^[@R5]--[@R7]^ However, very few consistent and strong associations with NPDs have been found.^[@R3],[@R8]^ GWASs have been based in the common disease/common variant (CDCV) hypothesis. This hypothesis proposes that a single common variant could be a risk factor for a large fraction of patients.^[@R9]^ Several researchers have proposed that an alternative hypothesis, the common disease/rare variants (CDRV) could lead to the identification of additional risk factors for complex diseases, such as neuropsychiatric disorders.^[@R9]^ In the context of the CDRV hypothesis, a special interest is focused on functional variants, such as non- synonymous SNPs (nsSNPs), which are a minor fraction of the total number of SNPs. nsSNPs represent the most well understood group of genetic variants of possible functional importance.^[@R10]^ Alterations in dozens of genes involved in brain plasticity are responsible for increasing or decreasing the risk of NPDs.^[@R6]--[@R8],[@R11]--[@R13]^ In the present study, we screened *in silico* a large number of brain-expressed genes, identified 5 nsSNPs located in genes of neuropsychiatric relevance, developed novel genotyping assays and reported their frequencies in a sample of Colombian healthy subjects. Methods {#s2} ======= Bioinformatic analysis of candidate genes {#s2a} ----------------------------------------- Several dozens of human brain-expressed genes were screened *in silico* for the presence of nsSNPs, using the BioMart tool.^[@R14]^ These genes correspond to several well-known functional categories, such as acetylcholine receptors, adaptor proteins, adhesion molecules, calcium channels, dopamine receptors, GABA receptors, glutamate receptors, potassium channels, protein kinases, protein phosphatases, serotonin receptors, signaling proteins, sodium channels, synaptic vesicle and transporters.MAF (\>0.01) in HapMap populations^[@R15]^ was screened to identify nsSNPs. The catalog of GWAS from the National Human Genome Research Institute (<http://www.genome.gov/gwastudies>),^[@R5]^ the public database *BioGPS* (<http://biogps.org>),^[@R16]^ the Stanley Brain Database --SBD- (<https://www.stanleygenomics.org>)^[@R17]^ and the HuGeNavigator tool (<http://hugenavigator.net/>)^[@R18]^ were used for further functional annotation. The GWAS catalog contains information for available GWAS for diseases and phenotypes, BioGPS contains an extensive set of genome-wide expression data for a wide range of normal human tissues, SBD provides differential expression data for BP and SZ postmortem human brains and the HuGeNavigator contains data for existing meta-analysis of genes and diseases ([Tab. 1](#tbl_1){ref-type="table"}). ###### General information about brain-expressed candidate genes Gene Name Gene symbol Position Size (bp) Gene Expression in brain Expression Changes Meta-Analysis GWAS ---------------------------------------------------------- ------------- -------------------------------- ----------- -------------------------- -------------------- --------------- ----------- Clock circadian regulator CLOCK chr 4: 56,294,068-56,315,663 21,596 **PFC, PL, PN** -- -- PERS Neuronal PAS domain protein 2 NPAS2 Chr 2: 101,436,614-101,613,291 176,679 **PFC, PL, PLB** BP †, SZ † -- CFJD, CFS Neuregulin 1 NRG1 Chr 8: 31,496,902-32,622,548 1,125,739 **PFC, PL, PLB** -- SZ NAR Solute carrier family 18 (vesicular monoamine), member 1 SLC18A1 Chr 8: 20,002,366-20,040,717 38,352 **PFC, PL, STN** -- -- ALS, MD WW and C2 domain containing 1 WWC1 Chr 5: 167,718,656-167,899,308 180,653 **HT, THAL, CN** -- EWM MP DNA sequences for flanking regions for the nsSNPs was obtained from the UCSC genome browser (<http://genome.ucsc.edu>).^[@R19]^ Primers were designed with the WASP^[@R20]^ (for allele- specific PCR, -AS-PCR-), Batch Primer3 and Primer1 online programs^[@R21],\ [@R22]^ (for Tetra Primer Allele-Specific PCR, -T-ARMS-PCR-). Oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA, USA). Genotyping of nsSNPs {#s2b} -------------------- DNA samples, extracted from peripheral blood using a salting out method,^[@R23]^ from 171 unrelated Colombian subjects (122 women; mean age = 21.2; SD = 2.92) were genotyped for the 5 nsSNPs. None of the subjects had personal history of neuropsychiatric disorders. Written informed consent was obtained from all subjects and this study was approved by the institutional ethical committee^[@R24]^ and was conducted according to the Declaration of Helsinki Principles. Conventional PCR (AS-PCR and T-ARMS-PCR) assays were carried out in Labnet MultiGene 96-well thermal cyclers (Labnet Internation al Inc, Edison, NJ, USA). Amplification reactions were performed in a total volume of 20 μl containing: 2 μl of genomic DNA (\~ 50 ng), 2.0 mM MgCl~2~, 10 X reaction Buffer, 1 mM of dNTPs, 1 M of Betaine and 0.8 U of Taq polymerase (Bioline, London, United Kingdom) (See [Table III](#tbl_3){ref-type="table"} for primers concentrations). PCR products were separated in a 2% agarose gel, stained with SYBR Safe (Invitrogen, Carlsbad, CA, USA) and visualized on a UV Transilluminator. Genotyping process for *NPAS2* gene was carried out in a CFX96 Touch Real-Time PCR system (BioRad, Hercules, CA, USA). To verify the consistency in the genetic results, a random subsample (10% of subjects) was re-genotyped for each nsSNP. In addition, two different investigators confirmed and validated the results, independently checking all genotypes. NPAS2-rs2305160 was analyzed using a specific TaqMan SNP Genotyping Assay (Applied Biosystems, Foster City, CA, USA). It was performed as follows: 2 μl (20 ng) genomic DNA, 1 X TaqMan^®^ Genotyping Master Mix (Applied Biosystems), 1 X of TaqMan Pre-Designed SNP Genotyping Assay (C\_\_15976652_10, Applied Biosystems) and water in a total volume of 10 μl. The amplification protocol consisted of 10-min denaturation step at 95°C (1 cycle), 95°C for 15s and 60°C for 90s (50 cycles). Statistical analysis {#s2c} -------------------- PLINK program was used for Hardy-Weinberg equilibrium analysis of genotype and allele frequencies.^[@R25]^ Allele frequencies for the five nsSNPs were compared using a 2 × 2 contingency table between Colombian subjects and HapMap samples: Utah Residents with Northern and Western European ancestry (CEU); Yoruba in Ibadan, Nigeria (YRI); Han Chinese in Bejing, China (CHB); Japanese in Tokyo, Japan (JPT); and Mexican ancestry from Los Angeles, USA (MXL). Results were considered statistically significant at *p*\<0.05 Results {#s3} ======= An *in silico* screening of several dozens of human brain-expressed genes identified 5 nsSNPs (rs6855837; p.Leu395Ile, rs2305160; p.Thr394Ala, rs10503929; p.Met289Thr, rs2270641; p.Thr4Pro and rs3822659; p.Ser735Ala), with higher probabilities of being true positives (MAF\>0.01 in HapMap samples) ([Table 2](#tbl_2){ref-type="table"}). These nsSNPs are located in genes (CLOCK, NPAS2, NRG1, SLC18A1 and WWC1) with a high relevance for neuropsychiatric disorders and endophenotypes ([Table 1](#tbl_1){ref-type="table"}), supported by multiple lines of experimental evidence, such as genome-wide expression and association studies and meta-analysis of genetic association studies. ###### Details for novel nsSNPs in brain-expressed genes Gene SNP Location Reference Allele Minor Allele Amino acid change MAF-CEU MAF-YRI MAF-CHB MAF-JPT MAF-MEX ------------- ------------ ----------------- ------------------ -------------- ------------------- --------- --------- --------- --------- --------- **CLOCK** rs6855837 chr4: 56319244 G T p.Leu395Ile 0.004 0.326 0.006 0.006 0.040 **NPAS2** rs2305160 chr2: 101591304 C T p.Thr394Ala 0.296 0.027 0.220 0.198 0.330 **NRG1** rs10503929 chr8: 32613983 T C p.Met289Thr 0.173 0.000 0.000 0.000 0.120 **SLC18A1** rs2270641 chr8: 20038466 T G p.Thr4Pro 0.376 0.106 0.256 0.333 0.286 **WWC1** rs3822659 chr5: 167858372 T G p.Ser735Ala 0.035 0.350 0.207 0.186 0.070 We designed novel cost-effective genotyping assays, based in allele-specific PCR methods, for four of these nsSNPs ([Table 3](#tbl_3){ref-type="table"}). We genotyped these five nsSNPs in a sample of healthy subjects (CLB)**.** [Table 4](#tbl_4){ref-type="table"} shows the allele and genotype frequencies found in our samples. Minor allele frequencies for the five SNPs ranged between 4.1% (WWC1) and 24.3% (SCL18A1). All markers were in Hardy--Weinberg equilibrium (HWE) (p value \>0.05). ###### Primers and PCR conditions for genotyping assays, based in allele-specific PCR, for novel nsSNPs in brain-expressed genes Method Allele bp Primer sequence Conc (μM) Primer sequence Conc (μM) ------------- ----------------- ----------- ------------------ ---------------------------------------------- ----------------- ---------------- --------------------------------------------- ------ CLOCK AS-PCR T:122 Wildtype forward 5-TCA GCA GCT GTC TCA GGA GG-3 1.0 Common reverse 5-TTT AAA CGA ATG ACC AGA CA-3 1.0 p.Leu395Ile 53°C -35 cycles G:122 Mutant forward 5-TCA GCA GCT GTC TCA GGA GT-3 1.0 NRG1 T-ARMS-PCR A:805 566 Forward 1 5-GCA GAG CCT TCG GTC TGA ACG AAA CA AGA C-3 1.0 Reverse 1 5-GTG AGG CCC ATT GGC AAT GTT CAG CA-3 1.0 p.Met289Thr 63°C -35 cycles G:805 269 Forward 2 5-CTC CCT TTC TTA TGT CCA GGA AAC AGC GG-3 0.2 Reverse 2 5-TCA AGG CGT CTG AAC AAA GGA GAG AGG CTC-3 0.2 SLC18A1 T-ARMS-PCR T:315 212 Forward 1 5-GCA ACC GCT GGG GAG CAT CCA GAA TAGG-3 1.0 Reverse 1 5-CAG TCC GGC CAT CAC CAT GCT CCA GA-3 1.0 p.Thr4Pro 63°C -35 cycles G:315 157 Forward 2 5-TCA AGC AGG GTG TAC ACT GCC TGC TGG G-3 0.4 Reverse 2 5-TTT CCT TCG TCT GGG CAG ACT CCC TCC C-3 0.4 WWC1 T-ARMS-PCR T:427 201 Forward 1 5-CAA TGA GGT GTT CTG GGT ATC CAC GT-3 0.75 Reverse 1 5-GTC TTC TGG TGA AGG GCT GGA TAA GC-3 0.75 p.Ser735Ala 63°C -35 cycles G:427 278 Forward 2 5-TGA TTC CCC AGA TTC CTC ACT TCT ACA AA-3 0.75 Reverse 2 5-CAG GGT AGA CCA ATC CCC ATA AAC ACA G-3 0.75 ###### Allele and genotype frequencies for five novel nsSNPs *Clock* rs6855837 *NPAS2* rs2305160 *NRG1* rs10503929 *SLC18A1* rs2270641 *WWC1* rs3822659 ------------- ------------------- ------------------- ------------------- --------------------- ------------------ ----- ------- ------ ----- ------- ------ ----- ------- ------ ----- GENOTYPE N \% N \% N \% N \% N \% G/G 128 74.9 C/C 98 57.3 T/T 142 83.0 T/T 100 58.5 T/T 157 91.8 G/T 41 24.0 C/T 66 38.6 T/C 28 16.4 T/G 59 34.5 T/G 14 8.2 T/T 2 1.1 T/T 7 4.1 C/C 1 0.6 G/G 12 7.0 G/G 0 0.00 Total 171 100 Total 171 100 Total 171 100 Total 171 100 Total 171 100 ALLELE N \% N \% N \% N \% N \% G 297 86.8 C 262 76.6 T 312 91.2 T 259 75.7 T 328 95.9 T 45 13.2 T 80 23.4 C 30 8.8 G 83 24.3 G 14 4.1 Total 342 100 Total 342 100 Total 342 100 Total 342 100 Total 342 100 HWE p value 0.81 0.60 0.96 0.73 0.86 We compared MAFs for these five nsSNPs found in the Colombian population with available data for samples from European, African, Asian and Latin American populations (extracted from HapMap). The YRI samples was the only population with significant differences for all genetic markers, with a *p* value less than 0.01 (see [Figure 1](#fig_1){ref-type="fig"}). For CHB and JPT samples, we observed significant differences for rs10503929 (*NRG1*) and rs3822659 (*WWC1*) SNPs (p\<0.002). Finally, the CLB samples showed allelic frequencies similar to those of MEX and CEU samples for all SNPs, with only one exception, rs6855837 SNP (MEX *p* = 0.039 and CEU *p*\<0.001). {#fig_1} Discussion {#s4} ========== Alterations in dozens of genes involved in brain and synaptic plasticity are responsible for a large number of common NPDs.^[@R11]--[@R13],[@R26]^ Genome-wide association studies, mainly focused on non-functional SNPs, have found a few consistent and strong associations with NPDs.^[@R3]^ nsSNPs represent the most well understood group of genetic variants of possible functional importance.^[@R10]^ Due to their impact on function of the encoded proteins, nsSNPs are not common in brain-expressed genes.^[@R10]^ We identified five nsSNPs in brain-expressed genes of high relevance for neuropsychiatric disorders: *CLOCK, NRG1, NPAS2*, *SLC18A1* and *WWC1*. *CLOCK* is a key regulator of the circadian system, and regulates the transcription of *PER1, PER2, PER3, CRY1* and*CRY2* genes. *NRG1* is known to mediate cell--cell interactions in the nervous system and has been implicated in the etiology of schizophrenia.^[@R27]^ *NPAS2* also has a key role the regulation of circadian mechanisms in the brain.^[@R28]^ *SLC18A1* encodes the vesicular amine transporter 1 (also known as VMAT1), important for the functioning of monoaminergic systems.^[@R29]^ WWC1 is involved in long-term potentiation mechanisms and synaptic plasticity.^[@R30]^ There is previous experimental evidence, arising from genome-wide expression and association studies, implicating the involvement of these genes in several neuropsychiatric disorders and endophenotypes, such as schizophrenia, mood disorders or memory performance.^[@R27]--[@R30]^ However, a large part of the previously studied variants in these genes were neutral polymorphisms.^[@R5]^ Similary the SNP detection method has wide applications and can be used for diagnosis of other untreated neurodegenerative disorders such as ALS, Parkinson's and AMD as shown.^[@R31]--[@R37]^ Cost-effective genotyping assays described in this work for these nsSNPs could be interesting methodological alternatives for researchers in developing and developed countries. These five nsSNPs in brain-expressed genes represent interesting candidates to analyze in future genetic studies of neuropsychiatric disorders and related endophenotypes, a focused synaptogenomics approach.^[@R12],[@R13],[@R26]^ This study was supported by grants from Universidad Antonio Nariño (VCTI-UAN). This article complies with International Committee of Medical Journal Editor's uniform requirements for manuscript. Conflict of interest: None. Competing interests: None, Source of funding: Universidad Antonio Nariño (VCTI-UAN)

Findings ======== Assembly of the CARMA1-BCL10-MALT1 (CBM) complex is an essential step in the signal transmission from the T cell receptor (TCR) to the activation of canonical IκB kinase (IKK)/NF-κB signaling \[\[[@B1]\]\]. After TCR/CD28 co-stimulation, receptor-proximal signaling events at the immunological synapse lead to activation of protein kinase C θ (PKCθ), which in turn phosphorylates the scaffold protein CARMA1. Hereby a conformational change of CARMA1 is induced that enables the recruitment of pre-assembled BCL10-MALT1 complexes \[\[[@B2]\],\[[@B3]\]\]. This process is accompanied by the association of many other factors to the CBM complex such as TRAF6, Caspase8, CK1α, CSN5, A20 and PP2A that control CBM activity and downstream signaling \[\[[@B4]\]\]. CARMA1 belongs to the family of MAGUK (membrane-associated guanylate kinase) proteins comprising PDZ, SH3 and GUK domains in its C-terminus. CARMA1 expression is restricted to lymphoid cells where it is associated to the cytosolic membrane. To identify novel CARMA1 interaction partners, we had previously performed yeast-two-hybrid (Y2H) screens using C-terminal CARMA1 constructs as baits \[\[[@B5]\]\]. Besides the protein phosphatase PP2A, we identified the aryl hydrocarbon receptor (AHR) interacting protein (AIP) as a new interaction partner of the C-terminal PDZ-SH3 domain of CARMA1 in yeast (data not shown). Using co-immunoprecipitation (co-IP) after transfection of HEK293 cells, we confirmed the association of CARMA1 and AIP (Figure [1](#F1){ref-type="fig"}A--C; see Additional file [1](#S1){ref-type="supplementary-material"} for detailed Methods description). Full length HA-CARMA1 and FLAG-AIP interacted after anti-HA or anti-FLAG IP, respectively. In agreement with the data from the Y2H screen, AIP bound to C-terminal CARMA1 fragments that contain the PDZ-SH3 domains, but not to the CARMA1 N-terminus (Figure [1](#F1){ref-type="fig"}A). Vice versa mapping of the CARMA1 interaction surface on AIP revealed that the N-terminal peptidyl-prolyl *cis/trans* isomerase (PPI) domain is binding to the CARMA1 PDZ-SH3, while the C-terminal tetratricopeptide repeats (TPRs) are dispensable (Figure [1](#F1){ref-type="fig"}B). {#F1} Initially, AIP has been proposed to regulate the aryl hydrocarbon receptor (AHR) localization, stability and ligand receptivity \[\[[@B6]\],\[[@B7]\]\]. AIP was shown to bind Hsp90 and AHR primarily through the TPR and suggested to act in complex with Hsp90 to regulate ligand-triggered AHR responses \[\[[@B8]\]-\[[@B10]\]\]. The N-terminal FKBP-like PPI domain of AIP that we identified as the CARMA1 interaction surface does not confer enzymatic activity \[\[[@B11]\]-\[[@B13]\]\]. Since extensive intramolecular restructuring of the CARMA1 MAGUK region is required to initiate downstream signaling after T cell stimulation \[\[[@B14]\]\], we asked if AIP can support conformational changes involved in CBM formation. For MAGUK proteins like DLG (discs large) and PSD95 an inhibitory intramolecular association between the SH3 domain and the C-terminal GUK domain has been demonstrated \[\[[@B15]\],\[[@B16]\]\]. Strep-Tactin pull-down (PD) revealed that the CARMA1 GUK domain directly binds to the PDZ-SH3 domain in HEK293 cells (Figure [1](#F1){ref-type="fig"}D). Since no direct association between AIP and CARMA1 GUK was obtained, we asked whether the CARMA1 GUK and AIP may compete for the same binding surface on CARMA1 and co-expressed HA-GUK and HA-AIP together with Strep-FLAG-PDZ-SH3 (Figure [1](#F1){ref-type="fig"}E and F). Again, CARMA1-GUK or AIP alone interacted with CARMA1-PDZ-SH3. Upon co-transfection, increasing concentrations of HA-GUK or HA-AIP led to a dose dependent loss of PDZ-SH3 association to AIP or CARMA1 GUK, respectively. Thus, AIP and CARMA1 GUK compete for the same surface on CARMA1, suggesting that their binding is mutually exclusive. In this setting, the CARMA1 GUK seemed to have a higher affinity for PDZ-SH3, which could keep CARMA1 in an inactive state. However, AIP binding may either facilitate an opening of this intramolecular interaction or stabilize the open conformation to alleviate CARMA1 activation and downstream signaling. To investigate if the CARMA1-AIP association is relevant for T cell signaling, we first confirmed an endogenous interaction in Jurkat T cells (Figure [2](#F2){ref-type="fig"}). Co-IPs using anti-CARMA1 or anti-AIP antibodies showed a transient interaction in the initial phase of T cell stimulation after PMA/Ionomycin (P/I) treatment or by CD3/CD28 co-ligation (Figure [2](#F2){ref-type="fig"}A-C). CARMA1 recruits BCL10-MALT1 to assemble the CBM complex and we asked if AIP is associated with the CBM holo-complex by performing anti-BCL10 IPs (Figure [2](#F2){ref-type="fig"}D). Indeed, AIP is also precipitated with BCL10 after T cell stimulation and this interaction was not seen in CARMA1 deficient JPM50.6 Jurkat T cells \[\[[@B17]\]\], revealing that the AIP - CBM association is mediated through CARMA1. To obtain further evidence that AIP could be involved in CBM regulation, we performed parallel time course analyses of AIP-CARMA1 and BCL10-CARMA1 association after P/I stimulation of Jurkat T cells (Figure [2](#F2){ref-type="fig"}E). CARMA1 co-precipitated with AIP and BCL10 at early time points after T cell stimulation. CARMA1-AIP binding was lost after pro-longed treatment, when the CBM complex was destroyed due to BCL10 degradation \[\[[@B18]\],\[[@B19]\]\]. Next we asked if the CARMA1-AIP association is also relevant in primary T cells. AIP is expressed in primary human as well as mouse T cells as detected by Western Blotting (Additional file [2](#S2){ref-type="supplementary-material"}A-B) and AIP mRNA levels did not significantly change in response to CD3/CD28 stimulation (Additional file [2](#S2){ref-type="supplementary-material"}C). By co-IPs we could verify a stimulus dependent interaction of CARMA1 and AIP in murine CD4 T cells (Figure [2](#F2){ref-type="fig"}F). Thus, just like the competition experiments of AIP and CARMA1 GUK for CARMA1 PDZ-SH3, the endogenous association studies suggest that AIP is predominantly bound to the open and active CARMA1 conformation within the CBM complex. {#F2} To address if AIP is involved in CBM complex formation and downstream signaling in Jurkat T cells, we performed knockdown experiments of AIP using a panel of three independent siRNAs (siAIP1, siAIP2 and siAIP3). A crucial step to initiate CBM assembly represents PKC dependent phosphorylation at Ser645 within the linker region of CARMA1 \[\[[@B3]\],\[[@B5]\]\]. We downregulated AIP and detected Ser645 phosphorylated CARMA1 by a phospho-specific antibody (Figure [3](#F3){ref-type="fig"}A) \[\[[@B20]\]\]. There was no difference in the extent of CARMA1 S645 phosphorylation upon CD3/CD28 stimulation, implying that AIP did not impair initial PKC-mediated CARMA1 linker phosphorylation. To see if AIP knockdown has an influence on CBM complex formation, we directly determined CARMA1 association after anti-BCL10 co-IP in response to P/I or CD3/CD28 stimulation (Figure [3](#F3){ref-type="fig"}B and C). In all cases there was a significant decrease in CARMA1-BCL10 association after AIP knockdown, demonstrating the necessity of AIP for proper CBM complex formation after T cell stimulation. {#F3} CBM complex formation is the key step for activation of canonical IKK/NF-κB signaling in response to TCR/CD28 co-engagement \[\[[@B1]\]\]. To assess if AIP also affects canonical NF-κB signal transmission downstream of the CBM complex, we determined IKK activation (Figure [3](#F3){ref-type="fig"}D), IκBα phosphorylation and degradation (Figure [3](#F3){ref-type="fig"}E) and NF-κB activation (Figure [3](#F3){ref-type="fig"}F) in AIP knockdown cells using different siRNAs. Clearly, IKKα/β T loop phosphorylation and thus IKK activity was severely reduced in AIP depleted Jurkat T cells. Further, less IKK activation was accompanied by reduced IκBα phosphorylation and degradation and decreased nuclear NF-κB DNA binding in response to P/I or CD3/CD28 stimulation. To address if AIP controls selectively TCR signaling, we compared NF-κB activation after CD3/CD28 and TNFα stimulation in AIP knockdown cells. Indeed, reduction of AIP selectively diminished NF-κB activity after CD3/CD28, but not after TNFα stimulation (Figure [3](#F3){ref-type="fig"}G), suggesting that AIP is modulating IKK/NF-κB upstream of IKK by regulating CBM-complex formation. Further, we determined the role of AIP for activation of the MAPKinases ERK and JNK in response to P/I stimulation (Additional file [3](#S3){ref-type="supplementary-material"}). Whereas ERK phosphorylation was largely independent of AIP, phosphorylation of JNKp54 isoform was slightly decreased, which is in line with previous findings that CARMA1-BCL10 are involved in JNKp54 activation \[\[[@B21]\]\]. To assess the downstream consequences of diminished TCR signaling, we determined interleukin-2 (IL-2) production as a hallmark of T cell activation after siRNA mediated AIP downregulation. As measured by quantitative RT-PCR, upregulation of *IL-2* mRNA in response to T cell stimulation by P/I or CD3/CD28 stimulation was significantly impaired by AIP knockdown (Figure [4](#F4){ref-type="fig"}A). Congruently, P/I induced IL-2 production and secretion was also decreased in AIP depleted Jurkat T cells as determined by ELISA (Figure [4](#F4){ref-type="fig"}B). IL-2 induction in T cells does not only rely on IKK/NF-κB signaling, but also on activation of the transcription factors NF-AT and AP-1 \[\[[@B22]\]\]. Therefore, we also assessed NF-AT and AP-1 DNA binding in nuclear extracts of Jurkat T cells after AIP knockdown (Figure [4](#F4){ref-type="fig"}C). NF-AT was severely reduced and AP-1 was slightly diminished in AIP downregulated Jurkat T cells, highlighting that AIP augments TCR/CD28 downstream pathways that contribute to optimal IL-2 induction. Effects on AP-1 may be downstream of NF-κB, because NF-κB activation can contribute to expression of Jun/AP-1 transcription factors such as JUNB and JUND \[\[[@B23]\]\]. Intriguingly, the MAGUK family member DLGH1 was shown to be an essential factor for TCR-triggered NF-AT activation, which may indicate that AIP could be a general regulatory factor for MAGUK dependent signaling events \[\[[@B24]\]\]. {#F4} Taken together, our data reveal a novel and unexpected role of AIP as a positive regulator of CBM complex formation and canonical IKK/NF-κB as well as NF-AT signaling in activated T cells. The observed effects of AIP are independent of AHR, as we and others could not detect AHR expression in Jurkat T cells \[\[[@B25]\]\]. In its inactive state, CARMA1 was suggested to adopt a double-closed conformation with the N-terminal CARD bound to the coiled-coil and the C-terminal GUK associated to the SH3 \[\[[@B14]\]\] (scheme Figure [4](#F4){ref-type="fig"}D). Whereas BCL10-MALT1 association to the CARD opens up the N-terminus of CARMA1, AIP interacts with the PDZ-SH3 and may thereby facilitate loss of C-terminal GUK-SH3 interaction to support opening of the C-terminal MAGUK region and downstream signaling. In this model, AIP binding may support the opening of CARMA1 or stabilize the signal competent active conformation. Since CARMA1 conformation is regulated by a multistep process, AIP may also function in signal amplification and/or positive feedback loop \[\[[@B26]\]\]. Recent data reveal that within the CBM complex CARMA1 acts as a molecular seed that initiates the assembly of BCL10 filamentous fibers \[\[[@B27]\]\]. It is tempting to speculate that AIP as a cofactor may guide the complex assemblies of such higher order molecular clusters that are initiated by rearrangements of MAGUK family members. Abbreviations ============= AHR: Aryl hydrocarbon receptor AIP: AHR interacting protein CARD: Caspase recruitment domain CC: Coiled-coil CBM: CARMA1-BCL10-MALT1 GUK: Guanylate kinase IKK: IκB kinase IL-2: Interleukin 2 IP: Immunoprecipitation LR: Linker region MAGUK: Membrane-associated guanylate kinase NF-AT: Nuclear factor of activated T-cells NF-κB: Nuclear factor-κB P/I: PMA/Ionomycin PDZ: PSD95, DLG1, ZO1 PKCθ: Protein kinase C θ PMA: Phorbol 12-myristate 13-acetate PPI: Peptidyl-prolyl *cis/trans* isomerase TCR: T-Cell receptor TPR: Tetratricopeptide repeat SH3: SRC homology 3 Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== GS designed and conducted most experiments, analyzed the data and wrote the manuscript. ACE, MV and KD designed and conducted experiments and analyzed the data. SH and TK generated and contributed important material. DK conceived the project, analyzed the data and wrote the manuscript. All authors read and approved the final version of the manuscript. Additional files ================ Supplementary Material ====================== ###### Additional file 1: Methods description. ###### Click here for file ###### Additional file 2: AIP is expressed in primary mouse and human T cells. ###### Click here for file ###### Additional file 3: Effects of AIP on MAP kinase activation. ###### Click here for file Acknowledgements ================ We thank Maria Henker for excellent technical assistance. Atufect lipofection reagent was a kind gift from Silence Therapeutics, Berlin. We thank Elisabeth Kremmer for gifting anti-HA antibody and Xin Lin for JPM50.6 cells. This work was supported by a grant within the DFG SPP1365 to DK.

{#sp1 .708} {#sp2 .709} {#sp3 .710} {#sp4 .711} {#sp5 .712} {#sp6 .713} {#sp7 .714} {#sp8 .715} {#sp9 .716} {#sp10 .717} {#sp11 .718}

{#sp1 .124}

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ The genus *Cystoseira* C. Agardh, brown algae belonging to the order Fucales, is distributed along the Mediterranean and Atlantic coasts from the intertidal to the lower sublittoral. This genus is ecologically relevant as an "ecosystem engineer" \[[@pone.0193011.ref001]\], and plays a key functional role in controlling spatial habitat heterogeneity, productivity, and nutrient cycling in temperate rocky reefs. In particular, *Cystoseira* forests provide refuge and food for many invertebrates and fishes and modulate the structure of the associated benthic community \[[@pone.0193011.ref002]\]. Currently, some *Cystoseira* populations (depending on species and location) are declining/lost throughout the Mediterranean \[[@pone.0193011.ref003],[@pone.0193011.ref004],[@pone.0193011.ref005],[@pone.0193011.ref006],[@pone.0193011.ref007], [@pone.0193011.ref008], [@pone.0193011.ref009]\], largely due to multiple human impacts such as urbanization, overfishing and climate change. Consequently, many systems have shifted from complex and productive assemblages to simpler, less-productive habitats such as barrens, turf-forming algae and other ephemeral opportunistic seaweeds, thus impacting the provision of ecosystem services \[[@pone.0193011.ref003],[@pone.0193011.ref010]--[@pone.0193011.ref016]\]. *Cystoseira* species are listed as "of community interest" according to the Habitat Directive (92/43/EEC) \[[@pone.0193011.ref017]\], and are indicators of environmental quality in Mediterranean coastal waters according to the Water Framework Directive (2000/60/EC) \[[@pone.0193011.ref018]\] (i.e., EEI \[[@pone.0193011.ref019]\] and CARLIT \[[@pone.0193011.ref020], [@pone.0193011.ref021]\]. Several species are protected by the Bern Convention, recognized as a priority by the Barcelona Convention and considered vulnerable by several international organizations (i.e. IUCN, RAC/SPA, MedPan). Despite the implementation of significant conservation efforts, most degraded systems have not recovered, emphasizing the urgency to develop an active intervention to restore endangered habitats \[[@pone.0193011.ref016]\]. The threat of losing *Cystoseira* species is magnified by their low dispersal capacity due to rapid egg fertilization and zygote sinking \[[@pone.0193011.ref022]--[@pone.0193011.ref025]\], which hampers natural recovery in the absence of adults, even if in some *Cystoseira* species the potential dispersal distance can be enhanced by the transport in floating rafts \[[@pone.0193011.ref026],[@pone.0193011.ref027]\]. As a result, interest in habitat restoration is increasing according to the Biodiversity Strategy to 2020 (Target 2; European Commission, 2011), which recommends the reintroduction of relevant species into areas where they were present historically and where the factors that led to their loss have been removed. Small-scale *Cystoseira* transplants have been attempted utilizing several techniques \[[@pone.0193011.ref028]--[@pone.0193011.ref031]\]. The most frequently tested method is the transplantation of juveniles or adult thalli: the only major challenge to this approach is the appropriate fixing of individuals or installation in the target area. Outplanting, which consists of producing recruits from fertile material in hatcheries for placement in the sea, has been explored for the genus *Cystoseira* to a lesser degree \[[@pone.0193011.ref029],[@pone.0193011.ref032]\]. In contrast, many studies have been performed using other large fucoid seaweeds \[[@pone.0193011.ref033]--[@pone.0193011.ref041]\] with a particular focus on the long-term maintenance of seedlings in culture \[[@pone.0193011.ref041]--[@pone.0193011.ref047]\]. Usually, the need for large numbers of germlings for outplanting represents a bottleneck in the design of large-scale restoration actions, so it is especially challenging to plan an efficient, effortless and cost-effective seedling production system that fits the breeding needs of a specific species. Considering the high potential of *Cystoseira* to generate gametes and zygotes under optimal conditions, the cultivation of germlings starting from fertile receptacles represents a sustainable option for restoring endangered species without depleting natural populations. From this perspective, the development of effective cultivation protocols tailored to the eco-physiological needs of different species is a compulsory milestone. The aim of this study was to develop an *ex situ* protocol for the restoration of *Cystoseira amentacea* var. *stricta* Montagne, a sensitive caespitose intertidal Mediterranean species whose reduction/loss has primarily been recorded in several locations in the NW Mediterranean \[[@pone.0193011.ref006],[@pone.0193011.ref048]\]. The protocol aimed to maximize zygote settlement, minimize embryo development time and generate a dense coverage of healthy germlings for outplanting. Based on these objectives, we tested the effects of easily adjustable variables (temperature, light and substratum) on the settlement and growth of early life stages to develop best practices for the restoration of this sensitive species. Materials and methods {#sec002} ===================== In June 2016, during the reproductive period of *C*. *amentacea* var. *stricta*, healthy apical fronds of ca. 3 cm in length holding mature receptacles were collected in the intertidal zone at Bogliasco, Genoa (NW Italy, 44°22\'40.37\"N---9°4\'35.14\"E) ([Fig 1](#pone.0193011.g001){ref-type="fig"}). No specific permits were required for collecting specimens in the selected location because it is not part of a protected or private area. Additionally, non-destructive sampling was performed, as only apical branches were collected. In particular, the site is characterized by a tide in the range of 30 cm (the barometric tide may dominate the water level) and an average spring temperature of 20°C. After sampling, apices wrapped with seawater-wetted towels were delivered within 24 h under dark, cold and humid conditions to the laboratory in Trieste (NE Italy) ([Fig 1](#pone.0193011.g001){ref-type="fig"}) for culture in environmentally controlled rooms. {#pone.0193011.g001} The temperature and photoperiod were selected to reflect typical seasonal conditions at the sampling site during the reproductive phase of *C*. *amentacea* var. *stricta* (from late spring to summer). The photoperiod was set to a 15:9 h light:dark cycle, and light intensities were chosen to mimic two possible scenarios occurring in nature during the gamete release, fertilization and the early life growth stages of *Cystoseira*: in the absence of a canopy, as on barren ground (higher light intensity) or in the understory (lower light intensity). Light was provided by LED lamps (AM366 Sicce USA Inc., Knoxville, USA), and irradiance was measured with a LI-COR LI-190/R Photometer (LICOR-Biosciences, Lincoln, NE, USA). Light irradiance (L) was set at 125 μmol photons m^−2^s^−1^ (L^-^) or at 250 μmol photons m^−2^s^−1^ (L^+^), while temperature (T) was set at 20°C (T^-^) or at 24°C (T^+^) The medium used in the experiments was Stosch\'s enriched filtered and autoclaved seawater (VSE) \[[@pone.0193011.ref049],[@pone.0193011.ref050]\]. Aquaria were filled with 4 L of culture medium, renewed every 4 days to minimize any possible effects of nutrient limitation and continuously aerated by air pumps. Two substrata with differing natures and roughness were tested: flat polished pebbles and rough clay tiles. A factorial laboratory experiment was performed that combined irradiance, temperature and substratum. Four combinations of culturing conditions consisting of two crossed levels of each environmental condition (L^+^T^+^, L^+^T^-^, L^-^T^+^, L^-^T^-^) and two substrata (Pebbles and Tiles) were tested in a two-way crossed design. Fertile apices were gently cleaned with a brush and rinsed with sterile seawater to remove adherent biofouling and surface detritus. Fronds were then placed in individual aquaria: three apices with mature receptacles on each substratum in separate aquaria per condition (in triplicate). Three additional replicates were placed on glass slides under each condition to observe zygote development with an inverted microscope (Leica, DM IL LED), and photographs were obtained with a Canon Powershot G9, avoiding stress on the treatment replicates. Data analysis {#sec003} ------------- ### Egg release and settlement {#sec004} After 2 h, gametes were released on all substrata under all conditions. Next, the receptacles were removed, and their fresh weight (FW) was measured. Due to the high density of eggs released on each substratum and to reduce manipulation stress as much as possible, counts were performed by processing photographic data. For each substratum, eggs were counted in 5 randomly selected 1x1 cm quadrants in photos obtained from a Leica MZ6 stereo microscope with a Nikon Coolpix 4500 camera. The number of eggs per unit of receptacle FW was analyzed as a response variable to compare settlement on different substrata under different conditions. Two-way crossed ANOVA was performed using both factors and their interaction as fixed factors. The data were square-root transformed to satisfy the assumptions of normality and homoscedasticity. ### Embryo development {#sec005} At the end of the first week, replicate embryos in all four growth stages were visible on glass slides and were counted: I-embryos with 4 primary rhizoids, II-embryos with 8 rhizoids, III-embryos with short apical hair/s, and IV-embryos with long apical hair/s ([Fig 2](#pone.0193011.g002){ref-type="fig"}). To analyze the differences among conditions, a PERMANOVA was applied using the percentage of individuals at each stage as a response variable and conditions as factors. Pairwise comparisons of significant terms were performed. {#pone.0193011.g002} ### Embryo survival {#sec006} After 24 h (T0), the number of zygotes was counted on each substratum (5 1x1 cm quadrants) by processing photographic data. Counts were then repeated at week 1 and week 2 to calculate the germling survival rate. We applied an ANCOVA for both week 1 and week 2 with unequal slopes for survival rate as a response variable and density (i.e., number of fertilized eggs) as a covariate, and substrata, conditions and their interactions were used as fixed factors. Assumptions were validated after applying the arcsine square root transformation (suitable for proportional data). Post hoc SNK tests were performed on significant interaction terms. ### Subsequent germling growth {#sec007} At week 2, three subsequent developmental stages were identifiable: I-round-shaped, II-elongated, and III-elongated with branch ([Fig 3](#pone.0193011.g003){ref-type="fig"}). The area of ten randomly chosen individuals per shape was measured in each replicate substratum and used as a response variable. The area was quantified by processing photographic data using ImageJ software \[[@pone.0193011.ref051]\]. Conditions and substrata were used as crossed fixed factors in a PERMANOVA, and pairwise comparisons were performed on significant terms. {#pone.0193011.g003} Results {#sec008} ======= Morphogenesis {#sec009} ------------- *C*. *amentacea* var. *stricta* is a monoic species with female and male gametes produced in the same conceptacle ([Fig 4A](#pone.0193011.g004){ref-type="fig"}). In our trials, gamete release began soon after the receptacles were placed in the aquaria, and the mean diameter of the eggs was 122±3 μm (n = 20). Fertilization occurred externally, and the development of a fecundation membrane around the zygote facilitated its adhesion to a substratum ([Fig 4B](#pone.0193011.g004){ref-type="fig"}). The zygote cytoplasm, which was initially homogeneous, became metabolically differentiated (polarization) with the establishment of a vertical growth axis (connecting the rhizoid and thallus pole). Twelve hours after fertilization (AF), the first division perpendicular to the growth axis was observed, leading to the formation of two equally sized cells ([Fig 4C](#pone.0193011.g004){ref-type="fig"}). The second division, which was parallel to the first, occurred in the lower cell 20--22 h AF ([Fig 4D](#pone.0193011.g004){ref-type="fig"}), while the third division, perpendicular to the first, appeared in the upper cell ([Fig 4D](#pone.0193011.g004){ref-type="fig"}). Within 32--34 h AF, many divisions occurred without an increase in embryo volume. {#pone.0193011.g004} Within the first week, the rhizoids developed as follows: via perpendicular divisions, the rhizoid mother cell gave rise to four cells that differentiated into four primary rhizoids (Figs [4E](#pone.0193011.g004){ref-type="fig"} and [2](#pone.0193011.g002){ref-type="fig"}) that grew further, forming long filaments (ca. 150--200 μm long). After detachment of the fecundation membrane, the length of the embryo increased through subsequent cell divisions, and secondary rhizoids were formed ([Fig 4F](#pone.0193011.g004){ref-type="fig"}). Thus, the embryo assumed an erect position, and an invagination with hyaline hairs appeared in the apical region ([Fig 4H and 4G](#pone.0193011.g004){ref-type="fig"}). At week 1 AF, the more developed embryos were 353±26 μm long and 259±37 μm wide (n = 20). At week 2 AF, germlings with numerous rhizoids grew further \[466±26 μm long and 275±28 μm wide (n = 20)\] and small lateral branches with some cryptostomata began to appear. At week 3 AF, numerous cryptostomata were observed ([Fig 4I](#pone.0193011.g004){ref-type="fig"}), and iridescence, which is typical of adult plants, was visible on the thallus surface. At this time point, the germlings were 1.38±0.13 mm long and 0.46±0.06 mm wide (n = 20). At the end of the third week, few tiles were transported back in the field at Bogliasco. Juveniles were 4.73±0.05 mm long and 0.81±0.09 mm wide after 1 month in the field, and they grew up to 9 cm in 9 months (April 2017). Egg release and settlement {#sec010} -------------------------- The number of settled eggs was higher on Tiles (avg = 5226, SE = 566) than on Pebbles (avg = 2429; SE = 199), highlighting a significant effect of substratum roughness (p\<0.0001; [S1 Table](#pone.0193011.s001){ref-type="supplementary-material"}). Conversely, no significant differences were detected between conditions or within the interaction term. Early embryo development {#sec011} ------------------------ The percentage of individuals was calculated for each of the four stages observed in the glass slide replicates ([Fig 2](#pone.0193011.g002){ref-type="fig"}). An MDS ordination plot ([Fig 5](#pone.0193011.g005){ref-type="fig"}) showed three different groups: L^+^T^-^, L^+^T^+^ and one group comprising L^-^ conditions (L^-^T^-^ and L^-^T^+^). PERMANOVA confirmed significant differences among these groups (p\<0.001; [S2 Table](#pone.0193011.s002){ref-type="supplementary-material"}). Furthermore, a bar plot ([Fig 6](#pone.0193011.g006){ref-type="fig"}) revealed a higher percentage of embryos in stage IV (embryos with long apical hair/s) under L^-^conditions. {#pone.0193011.g005} {#pone.0193011.g006} Embryo survival {#sec012} --------------- The ANCOVA results indicated strong significant differences among conditions and small differences between the two substrata at week 1 (p\<0.0001 and p\<0.03, respectively; [S3 Table](#pone.0193011.s003){ref-type="supplementary-material"}). At week 2 the interaction term (conditionXsubstratum) was significant (p\<0.001; [S4 Table](#pone.0193011.s004){ref-type="supplementary-material"}): on both substrata, the extreme conditions did not differ from each other (L^+^T^+^ = L^-^T^-^), but they differed significantly from the other two conditions (L^+^T^-^; L^-^T^+^). On Tiles, L^+^T^-^ and L^-^T^+^ did not differ significantly, although they differed on Pebbles. Additionally, ANCOVA indicated that L^+^T^-^ was significantly different between substrata, while survival was slightly higher on Pebbles. As shown in boxplots ([Fig 7](#pone.0193011.g007){ref-type="fig"}), the survival rate at week 1 was higher under extreme conditions (L^+^T^+^; L^-^T^-^) compared to the other two conditions (L^+^T^-^; L^-^T^+^). The survival rate from week 1 to week 2 conspicuously decreased (between 50 and 95%) for the L^+^T^+^, L^+^T^-^ and L^-^T^+^ conditions, while the survival rate under L^-^T^-^ remained more stable with a mortality below 30%. {#pone.0193011.g007} Subsequent germling growth {#sec013} -------------------------- PERMANOVA performed on the germling area at different stages ([Fig 3](#pone.0193011.g003){ref-type="fig"}) at week 2 showed significant differences among conditions (p = 0.001) with L^-^T^-^condition different from the others conditions ([S5 Table](#pone.0193011.s005){ref-type="supplementary-material"}; [Fig 8](#pone.0193011.g008){ref-type="fig"}). {#pone.0193011.g008} Discussion {#sec014} ========== Given the worldwide concern over the loss of key habitat-forming organisms, such as the large brown macroalgae of the order Fucales, and the downstream cascade effects on the services provided by such organisms, there is an urgent need to develop best practices and restoration strategies. Studies that provide sound information on how to best undertake habitat restoration are crucial for managing coastal ecosystems. Outplanting appears to be an ecologically sustainable approach that consists of two main steps: culturing germlings in the laboratory and transferring them into the field \[[@pone.0193011.ref037],[@pone.0193011.ref041],[@pone.0193011.ref045],[@pone.0193011.ref052]\]. For *Cystoseira*, outplanting appears to be a feasible management option that can provide many healthy specimens for re-introduction to the environment without impacting natural populations \[[@pone.0193011.ref029],[@pone.0193011.ref032],[@pone.0193011.ref053]\]. In this study, we focused on the first step of the outplanting process: developing an effective protocol for cultivating the early stages of *C*. *amentacea* var. *stricta*. This approach is challenging because most eco-physiological studies of *Cystoseira*, both in the field and in the laboratory, have focused on the adult stages. Nevertheless, the single/few-celled stages are characterized by simplicity and sensitivity, so compared to adults, any environmental variable will exert greater effects on germling mortality and growth rate \[[@pone.0193011.ref054],[@pone.0193011.ref055]\]. Thus, the needs of these ontogenetic stages must be understood because findings related to the macrothallus stages cannot be extrapolated to the microscopic stages \[[@pone.0193011.ref056]\]. Thus, species-specific best practices for the cultivation of germlings must be developed to implement a successful large-scale restoration strategy. First, we tested whether it was possible to collect samples far from the breeding facility (ca. 600 km) without damaging the reproductive materials to exclude the possible negative effects related to the distance of the target site from the hatchery. Transporting under dark and cold conditions allowed immediate gamete release, thus avoiding thermal and light shock in the laboratory; indeed, receptacles that were placed in aquaria soon after their arrival in the laboratory immediately released gametes. Starting from eggs fecundation, we described the morphogenesis and successive germlings development of *C*. *amentacea* var. *stricta*. Based on zygotes segmentation and number of primary rhizoids, three groups of *Cystoseira* species have been identified by \[[@pone.0193011.ref022]\]. *C*. *amentacea* var. *stricta* belongs to the first group, that is characterized by spherical eggs, zygotes that adhere to substrata by the fecundation membrane, and four primary rhizoids \[[@pone.0193011.ref057]\]. Nutrient limitation affects many processes, such as photosynthetic capacity \[[@pone.0193011.ref058]\], protein content \[[@pone.0193011.ref059],[@pone.0193011.ref060]\], photoprotection mechanisms \[[@pone.0193011.ref061]--[@pone.0193011.ref063]\], egg behavior and settlement, embryonic development and growth rate \[[@pone.0193011.ref064]--[@pone.0193011.ref069]\], so the culture medium was enriched to allow the germlings to invest their photosynthetic energy in growth processes \[[@pone.0193011.ref070],[@pone.0193011.ref071]\]. Better growth of *C*. *amentacea* var. *stricta* germlings with nutrient enrichment has also been observed by \[[@pone.0193011.ref057]\]. The positive effect of nitrate supply on *C*. *stricta* growth rate has been demonstrated in adults cuttings cultivation, although with small differences between apical and subapical segments \[[@pone.0193011.ref072]\]. Together with culture conditions, the choice of substrata must also favor the adhesion of gametes and zygotes and their successive development. Regardless of the culture conditions, rougher tiles showed higher zygote settlement than smoother pebbles, although the substrata did not affect successive germling growth or survival under any of the tested conditions. Embryonic mortality after one week was elevated under all conditions (50% on average), which was expected given the very high stochastic gamete and zygote mortality of *Cystoseira* in the natural environment. At the end of the first week, embryonic survival was positively affected by two of the tested conditions: L^+^T^+^ and L^-^T^-^. At week 2, survival was still higher under the L^-^T^-^ condition but significantly decreased under L^+^T^+^ treatment. The other two combined conditions showed the lowest embryonic survival throughout the entire experiment. Low-light and low-temperature conditions also favor higher embryonic survival rate in *Sargassum vachellianum* Greville cultivations \[[@pone.0193011.ref073]\]. Lower light intensity also reduced the time required for embryo development, allowing a greater number of individuals to reach developmental stage IV (larger embryos) within the first week regardless of temperature. After two weeks at low irradiance, lower temperature also strongly determined the growth of larger embryos. Our findings corroborate that environmental conditions (specifically light and temperature) may interact and exert synergistic or antagonistic effects on physiological responses in unpredictable ways that differ according to developmental stage. The distribution of *C*. *amentacea* var. *stricta* is restricted to the intertidal, so this species is naturally exposed to high levels of irradiance that potentially exceed its light energy requirements, as has been reported for other species that live close to the water surface \[[@pone.0193011.ref070],[@pone.0193011.ref074]--[@pone.0193011.ref076]\]. Generally, sun-adapted species \[*sensu* [@pone.0193011.ref077]\] develop efficient photoprotection mechanisms to tolerate light stress in addition to dynamic photoinhibition \[[@pone.0193011.ref075],[@pone.0193011.ref076],[@pone.0193011.ref078]--[@pone.0193011.ref086]\].Our study highlighted the light-shade adaptation of *C*. *amentacea* var. *stricta* germlings, which showed enhanced growth at lower irradiance (125 μmol photons m^−2^ s^−1^). Other *Cystoseira* species have been cultivated under different conditions that have primarily depended on laboratory facilities such as *Cystoseira susanensis* Nizamuddin (16±1°C \| 40 μmol photons m^−2^ s^−1^ \[[@pone.0193011.ref087]\] and *C*. *barbata* (Stackhouse) C. Agardh (16--17°C \| 120 μmol photons m^−2^ s^−1^ \[[@pone.0193011.ref029]\]. The morphological development of *C*. *amentacea* var. *stricta* embryos cultivated at 18±1°C and an average light intensity of 70 μmol photons m^−2^s^−1^ has been described by \[[@pone.0193011.ref057]\]. After ca. 2 months in these conditions, embryos cultivated in seawater were 332.60 ± 22.3μm long, while in VSE medium they were 5.14 ± 0.08 mm \[[@pone.0193011.ref057]\]. In our experiment germlings cultivated in VSE were 1.38±0.13 mm long after three weeks and 4.73±0.05 mm long after one month in the field. We observed cryptostomata and lateral branches after two weeks of cultivation, while in \[[@pone.0193011.ref057]\] study cryptostomata were observed after 52 days and ramifications appeared after 106 days. Studies examining adult thalli of *Cystoseira* have demonstrated the absence of photosynthetic inhibition, even with very high irradiance \[[@pone.0193011.ref088]\]. In *C*. *barbata* (Stackhouse) C. Agardh f. *aurantia* (Kuetzing) Giaccone, photoinhibition only occurs at irradiances higher than 1500 μmol photons m^−2^s^−1^ \[[@pone.0193011.ref089]\], while photosynthesis in *C*. *mediterranea* Sauvageau is not saturated at an irradiance of 1600 μmol photons m^−2^s^−1^\[[@pone.0193011.ref075]\]. Notably, the light requirements of adults should not be extrapolated to the microscopic stages because the presence of non-photosynthetic tissues in complex thalli increases the need for light energy \[[@pone.0193011.ref076]\], and the irradiance reaching the embryos is restricted by adult fronds in nature \[[@pone.0193011.ref035],[@pone.0193011.ref042],[@pone.0193011.ref090]\]. *Cystoseira* zygotes and germlings settle under adult plants, where they find a protective screen against high irradiance and other stressors. In nature, such community self-protection could be particularly important during spring-summer, when *C*. *amentacea* var. *stricta* produces new recruits in the study area. Conversely, the lower irradiance requirements of germlings permit high-density cultures because self-shading is not a restricting factor. At higher temperatures (24°C), the proliferation of biofouling was enhanced, particularly at lower light intensity, progressively affecting the development of *C*. *amentacea* var. *stricta* embryos. Based on these results, we determined that the lower values tested for irradiance (125 μmol photons m−2 s−1) and temperature (20°C) were the best hatchery conditions to accelerate the development of high numbers of healthy, large embryos. Further studies are required to improve the second step in the outplanting process to increase the number of juveniles that can reach the adult stage once they are reintroduced into the field. Grazing pressure, timing and density dependent effects need to be considered to achieve the best restoration results. Supporting information {#sec015} ====================== ###### Two-way crossed ANOVA performed on the number of eggs per gram. Significant effects are in bold. (PDF) ###### Click here for additional data file. ###### One-way PERMANOVA performed at week 1 based on the percent composition of three embryonic developmental stages under the different conditions. Significant effects are in bold. ^a^Pairwise comparisons among conditions: L^+^T^-^≠L^+^T^+^≠ L^-^T^-^ = L^-^T^+^. (PDF) ###### Click here for additional data file. ###### ANCOVA performed at week 1 using survival, substratum and condition as factors and density as a covariate. Significant effects are in bold. ^a^SNK test among conditions: L^+^T^+^≠L^-^T^-^≠L^+^T^-^≠L^-^T. (PDF) ###### Click here for additional data file. ###### ANCOVA performed at week 2 using survival as a response variable, substratum and condition as factors and density as a covariate. Significant effects are in bold. ^a^SNK test among substrata within condition: Cond. L+T-, T≠S; all other Cond., T = S. ^b^SNK test among conditions within substratum: Sub. S, (L+T+ = L-T-)≠L+T-≠L-T+; Sub. T, (L+T+ = L-T-)≠(L+T- = L-T+). (PDF) ###### Click here for additional data file. ###### Two-way PERMANOVA performed on the germling area at different stages at week 2. **Condition and substratum are crossed fixed factors.** Significant effects are in bold. ^a^Pairwise comparisons among conditions: L^-^T^-^≠L^+^T^+^ = L^-^T^+^ = L^+^T^-^. (PDF) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Taxonomy {#s1} ======== Domain *Bacteria*, phylum *[Firmicutes](http://doi.org/10.1601/nm.3874)*, class *[Bacilli](http://doi.org/10.1601/nm.4854)*, order *[Bacillales](http://doi.org/10.1601/nm.4855)*, family *[Bacillaceae](http://doi.org/10.1601/nm.4856)*, genus *[Bacillus](http://doi.org/10.1601/nm.4857)*, species *subtilis*. Properties {#s2} ========== *[B. subtilis](http://doi.org/10.1601/nm.10618)* is a fast-growing, Gram-positive, aerobic bacterium with rod-shaped cells that are typically 2--6 µm long and just less than 1 µm in diameter. The optimal growth temperature is about 30--35 C, giving a doubling time of as little as 20 min. Under some growth conditions the cells have a tendency to form long chains connected by uncleaved septal wall material. Under starvation conditions the cells can undergo a complex-2 cell-differentiation process leading to the formation of an endospore, which is released by lysis of the enveloping mother cell. The vegetative cells can be motile. Alternatively, they can form biofilms and 'fruiting bodies' containing spores. Genome {#s3} ====== The widely studied *[B. subtilis](http://doi.org/10.1601/nm.10618)* strain 168 is a tryptophan auxotroph isolated in the 1950s. Its was one of the first bacteria to be fully genome sequenced, revealing a 4.2 Mbp chromosome with about 4100 genes \[[@R1]\]. The genome of *[B. subtilis](http://doi.org/10.1601/nm.10618)* remains one of the best annotated, through a series of updates, most recently that of \[[@R2]\]. A comprehensive database, 'SubtiWiki' (<http://subtiwiki.uni-goettingen.de/>), provides a reliable and user-friendly interface to the latest data. The database includes a hugely comprehensive set of data listing transcriptional units, promoters and regulatory RNAs from the work of Nicolas *et al*. \[[@R3]\]. Complete lists of essential genes have been obtained in a series of global projects, most recently identifying 257 genes required for growth in LB at 37°C \[[@R4]\]. Analysis of the complete genome sequences of 36 diverse *[B. subtilis](http://doi.org/10.1601/nm.10618)* isolates has revealed a 'pan genome' (total gene set) of about 6250 genes, and a 'core genome' (conserved gene set) of about 2500 genes \[[@R2]\]. Notable gene classes include about 300 genes required for endospore formation, and multiple prophages or phage remnants. Conclusions from an overview of gene content are consistent with the notion that *[B. subtilis](http://doi.org/10.1601/nm.10618)* is adapted for life on plants or in the rhizosphere. Phylogeny {#s4} ========= *[B. subtilis](http://doi.org/10.1601/nm.10618)* is the 'type strain' of the order *[Bacillales](http://doi.org/10.1601/nm.4855)* and the defining organism of the whole *[Firmicutes](http://doi.org/10.1601/nm.3874)* phylum, having been first described in detail by Ferdinand Cohn in 1872 \[[@R5]\]. Cohn's organism was probably identical to an organism isolated even earlier, in 1832, by Ehrenberg. The history of the discovery and characterization of *[B. subtilis](http://doi.org/10.1601/nm.10618)*, and controversies over its taxonomic status, are summarized in an interesting article by Soule \[[@R6]\]. The most recent version of Bergey's Manual lists 141 species of *[Bacillus](http://doi.org/10.1601/nm.4857)* \[[@R7]\]. A wide range of traits are used to distinguish between *[B. subtilis](http://doi.org/10.1601/nm.10618)* and other species in the Genus. Most prominent among these are types of murein (peptidoglycan) cross bridging; ability to hydrolyse and utilize various carbon sources; colony, cell and spore morphology; and tolerance of salt, pH and temperature variation. Key features and discoveries {#s5} ============================ *[B. subtilis](http://doi.org/10.1601/nm.10618)* has a long history, being first described in the nineteenth century. The origins of the standard lab strain, 168, are poorly documented, but its place in the annals of genetics was cemented by experiments in the late 1950s showing that it was naturally transformable with linear DNA (see \[[@R8]\]). *[B. subtilis](http://doi.org/10.1601/nm.10618)* emerged as the Gram-positive model organism of choice largely because endospore formation became popular as a marvellously tractable system for studying fundamental aspects of cellular development and differentiation. Processes such as the decision to initiate sporulation, asymmetric cell division, cell fate determination and cell morphogenesis were all worked out in molecular detail at a time when it was very difficult to dissect these processes in higher organisms. A pivotal problem in understanding spore development lay in discriminating between events occurring simultaneously in the developing prespore and mother-cell compartments, which have identical chromosomes but very different gene expression profiles. This problem powered the adaptation of digital fluorescence imaging for use in bacteria, which was then a major factor ushering in the modern field of bacterial cell biology. Later, these methods were applied to many other important problems, especially central bacterial cell processes of cell division, chromosome segregation, and cell growth and morphogenesis. Progress in understanding these processes now runs almost in parallel between *[B. subtilis](http://doi.org/10.1601/nm.10618)* and its Gram-negative comparator, *Escherichia coli. Bacillus* genetics and cell-biology methods have also made the organism popular for more general studies of cell physiology and biochemistry, as well as alternative morphogenic processes, such as biofilm formation. Another major driver of interest in *[B. subtilis](http://doi.org/10.1601/nm.10618)* is based on its importance as an industrial organism, mainly through its prodigious ability to secrete various important hydrolytic enzymes directly into the culture medium but also as a producer of fine chemicals, such as riboflavin. Its attractiveness as a safe host for production of natural and engineered products has been helped by its long standing use in 'natto', a Japanese dish made from fermented soy bean curd and also as a probiotic. As mentioned above, *[B. subtilis](http://doi.org/10.1601/nm.10618)* appears to be adapted to life in association with plants, either as an epiphyte or in the rhizosphere, and historically it has typically been isolated from decaying vegetative matter such as hay. Adaptation to this ecological niche may help explain a third important industrial use of *[B. subtilis](http://doi.org/10.1601/nm.10618)*, as a plant growth promoter, through production of specialized metabolites, niche exclusion of pathogens and other probably various other factors. Open questions {#s6} ============== *[B. subtilis](http://doi.org/10.1601/nm.10618)* and the wider *Firmicute* lineage are usually referred to as Gram-positive, but some *[Firmicutes](http://doi.org/10.1601/nm.3874)* have a genuine outer membrane. What is the explanation for this profound evolutionary conundrum? Can we use *[B. subtilis](http://doi.org/10.1601/nm.10618)* to solve the fundamental mechanistic questions surrounding cell morphogenesis and cell division? *[B. subtilis](http://doi.org/10.1601/nm.10618)* is a wonderful host for production of hydrolytic enzymes from other bacteria but can the secretion mechanism be made to work for heterologous, high-value proteins? What ecological factors and molecular mechanisms underlie the plant growth promoting properties of *[B. subtilis](http://doi.org/10.1601/nm.10618)*? Work in the Errington lab is funded by major grants from the European Research Council (670980) and the Wellcome Trust (209500). The authors declare that there are no conflicts of interest.

India is home to approximately 2.1 million people living with HIV ([@CIT0001]). Of the 0.6 million people living with HIV and on antiretroviral therapy (ART) in 2012, approximately 5,503 (0.9%) were receiving second-line ART and none were on third-line ART under the National AIDS Control Programme. Because a certain percentage of patients living with HIV fail their ART regimens every year, and there is the possibility of transmission of drug-resistant HIV ([@CIT0002], [@CIT0003]), a gradual rise in the number of patients failing second-line ART regimens can be expected, as well as a growing need for access to third-line ART in the future. Furthermore, the cost of a third-line regimen represents a major challenge, because it is nearly 15 times higher than that of a first-line ART regimen and over six times that of a typical second-line regimen ([@CIT0004]). Routine HIV viral load (VL) testing is an indispensable requirement in any ART programme. The World Health Organization (WHO) recommends VL monitoring as the preferred approach to provide an early and more accurate indication of treatment failure, thereby reducing the accumulation of drug-resistance mutations and improving clinical outcomes ([@CIT0005]). However, VL testing is not routinely offered in many countries owing to its cost and lack of availability. To date, there are limited data describing experiences of routine VL monitoring in the contexts of patients living with HIV who have suspected second-line ART failure and the use of third-line ART in resource-limited settings. To address this knowledge gap, this study describes the experiences and programmatic challenges during treatment and VL testing of patients living with HIV who are suspected of second-line ART failure and those needing third-line ART in Mumbai within a non-governmental organization (NGO)-run programme that provides ART and VL testing free of charge. Methods {#S0001} ======= Study design {#S0001-S20001} ------------ This was a retrospective, observational cohort study of patients living with HIV who had suspected failure of second-line ART using data collected under routine programmatic conditions. Setting and study population {#S0001-S20002} ---------------------------- Since 2006, Médecins Sans Frontières (MSF) has been operating an HIV/tuberculosis (TB) programme in Mumbai, India ([@CIT0006]), for those unable to access essential HIV care and treatment services elsewhere. During the study period, the size of the entire clinic cohort was approximately 290 patients. One of the main objectives of the MSF programme has been to provide free second-line or third-line ART as appropriate to patients living with HIV who are not able to access these through national ART centres. These patients living with HIV are referred from government ART centres, public--private ART centres, and a network of community NGOs. The programme initially started providing third-line ART in 2011. Patients were eligible for inclusion in this study if they were suspected of failing a second-line ART regimen that included a protease inhibitor (PI) and 2--3 nucleoside or nucleotide reverse transcriptase inhibitors (NRTIs), and they were followed for at least 12 months between January 2011 and March 2014. Definitions and treatment protocol {#S0001-S20003} ---------------------------------- Second-line ART was defined as the regimen used for treatment of patients living with HIV who failed a first-line regimen, and typically it would consist of a PI (e.g. atazanavir or ritonavir) and two or three NRTIs (e.g. lamivudine and tenofovir±zidovudine). Third-line ART regimen in the MSF programme included the second-generation PI darunavir boosted with ritonavir (DRV/r) and the integrase inhibitor raltegravir (RAL), together with one or more NRTIs likely to be effective on the basis of HIV genotyping results. As part of routine clinical care, HIV VL and CD4 cell count testing was carried out for each patient every 6 months in the MSF programme. An undetectable VL was defined as a result with fewer than 50 copies/ml. Treatment failure was defined as a VL result \>5,000 copies/ml in two consecutive results in a 3-month time frame, as previously recommended by WHO ([@CIT0007]). Patients in the programme were suspected of failing second-line ART if they had received and failed a first-line regimen, were subsequently treated with a second-line regimen for at least three consecutive months, and then had a single VL test result measuring \>5,000 copies/ml. All patients suspected of failing second-line ART underwent a thorough assessment, including structured client-centred adherence counselling and repeated VL measurement after 3 months of good adherence. The assessment involved a multidisciplinary team consisting of an HIV physician, psychologist, social worker, and nurse. Each member of the team had special predefined tasks and responsibilities ([Box 1](#T0002){ref-type="boxed-text"}). The time to virological suppression, if achieved, was measured and reported. ###### Predefined tasks and responsibilities of members of the multidisciplinary team assessing those suspected of failing second-line ART. ***HIV Physician:*** Physicians perform a detailed history and examination at the first visit, and they arrange for a number of laboratory tests, including complete blood count, alanine transaminase, serum creatinine, serum bilirubin, blood sugar, serum lipids, Western blot (for HIV-1 and 2), HIV-1 viral load (VL), CD4 count, chest X-ray, abdominal ultrasound, and fundoscopy, as well as a pregnancy test in females. Second-line ART is restarted only after ruling out opportunistic infections. Any treatment-related adverse events are symptomatically managed. Initially, the patient is assessed every two weeks for the first three visits and then monthly. At every visit, adherence is reinforced, and after 2--3 months of good adherence, VL testing is repeated. VL and CD4 count testing are repeated every 6 months. If VL suppression is not achieved, genotyping is performed, and a treatment regimen is designed based on the genotyping results. ***Psychologist:*** Every patient is evaluated at baseline, including assessment with the PHQ9 Patient Depression questionnaire. Baseline HIV knowledge is tested using 30 basic HIV questions: a score of \<10 is defined as poor knowledge, 11--20 as average knowledge, and 21 or more as good knowledge. Barriers for adherence are identified. The patients see the psychologist every two weeks for the first three visits, and then every month. ***Social Worker:*** Every patient receives a social assessment at their first visit, during which social barriers for adherence are identified and solutions discussed with the patient. ***Nurses:*** Dosage and timing of all pills are explained to all patients. A pill count is done at every visit, and the percentage of pills taken is entered into a file. Pillboxes are used for patients who have difficulties with timing of their ARVs. The adherence levels of all patients were assessed with patient self-reports. These levels and any adherence concerns were recorded by the team following individual interviews. In patients observed to have adherence issues, the team provided education to the patient about HIV, VL and CD4 testing, and treatment failure. Psychological and emotional support were offered to the patients during treatment, and they were also referred to other NGOs and services as needed. A two-log decrease in the repeat VL measurement 3 months after interventions to support adherence in those suspected of failing a second-line regimen was defined as virological suppression, and these patients were continued on second-line ART. Those patients who failed to achieve virological suppression were investigated for antiretroviral (ARV) resistance using genotyping with the ViroSeq HIV-1 genotyping system method (Celera Diagnostics, Alameda, CA, USA); the instrument used was the ABI 310 Genetic Analyzer (Applied Biosystems, Carlsbad, CA, USA). The Stanford genotypic resistance interpretation algorithm was used to analyse resistance of HIV in these patients. Patients in need of a third-line regimen were prescribed the regimen based on genotyping results. Data collection and analysis {#S0001-S20004} ---------------------------- Demographic and clinical information of all patients living with HIV was recorded in patient files. The clinical data that were routinely collected for each patient, including treatment and laboratory data, were entered into an electronic database. A full-time data manager routinely supervised data entry for accuracy and completeness. Data from all patients suspected of second-line failure between January 2011 and March 2014 were included in the analyses. Descriptive statistics were used to analyse the data of these patients using SPSS (Release 20, 2011; SPSS, Inc., Chicago, IL, USA). Ethics {#S0001-S20005} ------ The study satisfied the criteria for reports using routinely collected programmatic data set by the MSF independent Ethics Review Board in Geneva, Switzerland. As this was a study of routinely collected monitoring data, informed consent from the patients was not obtained. The named ethics committee specifically approved the study and waived the need for consent. Results {#S0002} ======= Patient characteristics {#S0002-S20001} ----------------------- A total of 47 patients living with HIV were suspected to have failed second-line ART during the study period. The median age of patients was 40 years (interquartile range (IQR): 35--46), and almost three-fourths of them (35/47) were male. The median baseline CD4 count for these patients at the time of enrolment was 110 cells/µl (IQR: 47.0--208.8). The monthly family income was less than 7,000 Indian rupees in 27/44 (61.4%) cases, which is less than the cost of second-line ART in a pharmacy in Mumbai ([@CIT0008]). Thirty-one (66%) patients were in WHO clinical stage IV at the time of enrolment ([@CIT0009]). Three patients were co-infected with HIV and multidrug-resistant TB (MDR-TB). The median duration of exposure to ART of these patients prior to enrolment in this study was 7 years (IQR: 5.0--9.0). Twenty-two (47%) patients had suffered from ARV medication--related grade III or IV toxicity ([@CIT0010]). Treatment outcomes {#S0002-S20002} ------------------ Of the 47 patients suspected of second-line ART failure, 29 patients (62%) responded to enhanced adherence support and had a two-log decrease in their level of HIV on subsequent VL testing ([Fig. 1](#F0001){ref-type="fig"}). The other 18 patients (38%) did not respond to enhanced adherence support and were eventually switched to third-line ART. The median duration of prior ART exposure of these 18 patients ([Table 1](#T0001){ref-type="table"}) was 9.0 years (IQR: 7.0--10.3). Out of 18 patients who switched to third-line ART, 13 had VL follow-up before the end of March 2014, of whom 11 achieved viral suppression after a median duration of 3 months (IQR: 2.5--3.0). Two patients failed to achieve viral suppression, one of whom died in April 2013 due to *Pneumocystis jirovecii* pneumonia (PCP). {#F0001} ###### Clinical characteristics of 18 patients living with HIV and started on third-line ART, Mumbai, India, January 2011--January 2014 Age Sex WHO stage Prior duration of ART (years) Comorbidity Drugs in third-line ART regimen Duration on third-line ART (months) Adverse events ---- ----- ----- ----------- ------------------------------- ------------- --------------------------------- ------------------------------------- ----------------------- 1 40 M III 8 DL, DM DRV/R, RAL, ATV 30 Transient rise in ALT 2 40 M IV 8 DL DRV/R, RAL, EFV, TDF, 3TC 33 No adverse event 3 38 M IV 8.5 -- DRV/R, RAL, ABC, 3TC 33 No adverse event 4 32 F III 7 DL DRV/R, RAL, EFV 32 No adverse event 5 60 F IV 10 DL, HTN DRV/R, RAL, ABC 32 No adverse event 6 38 M IV 7 -- DRV/R, RAL, TDF, 3TC 18 No adverse event 7 40 M II 10 DL DRV/R, RAL, TDF, 3TC 30 No adverse event 8 45 M IV 10 HTN DRV/R, RAL, TDF, 3TC 15 No adverse event 9 38 F IV 9 DL DRV/R, RAL, TDF, 3TC 17 No adverse event 10 49 M IV 13 DL DRV/R, RAL, ABC, 3TC 16 No adverse event 11 46 M IV 4 HBV DRV/R, RAL, ABC, 3TC 3 No adverse event 12 45 M IV 16 DL DRV/R, RAL, ABC, 3TC, EFV 10 No adverse event 13 16 F IV 6 -- DRV/R, RAL, ABC, 3TC 3 No adverse event 14 40 M IV 15 DM DRV/R, RAL, TDF, 3TC 5 No adverse event 15 48 M IV 9 DM DRV/R, RAL, ABC, 3TC 6 No adverse event 16 47 M IV 11 -- DRV/R, RAL, TDF, 3TC 1 No adverse event 17 42 F IV 6 -- DRV/R, RAL, TDF, 3TC 4 No adverse event 18 52 M IV 9 -- DRV/R, RAL, TDF, 3TC 3 No adverse event ART: antiretroviral therapy; DM: diabetes mellitus; DL: dyslipidemia; HTN: hypertension; HBV: hepatitis B; ATV: atazanavir; ABC: abacavir; DRV/R: darunavir boosted with ritonavir; EFV: efavirenz; RAL: raltegravir; TDF: tenofovir; 3TC: lamivudine; ALT: alanine aminotransferase. Discussion {#S0003} ========== As a result of enhanced adherence support and routine HIV VL monitoring, the majority of patients living with HIV who had suspected second-line ART failure in the MSF programme in Mumbai did not have to be switched to third-line ART. In fact, most of the patients suspected to be failing second-line ART in this study had adherence issues rather than HIV strains that were resistant to the ARV drugs, a finding which is in line with the results of previous studies, including a similar South African cohort and several other programmatic cohorts in resource-limited settings ([@CIT0011]--[@CIT0013]). One of the main reasons behind the lack of adherence reported by patients was the high cost of second-line ART (prior to enrolment in the MSF programme). These patients had initially opted to receive their HIV treatment in private facilities rather than public ones, in an effort to reduce HIV-related stigma and discrimination; however, they were eventually no longer able to afford the medications and were forced to stop the treatment. Patients co-infected with HIV and TB had a high pill burden, which has the potential to negatively affect adherence. Additional reasons for poor adherence in our cohort included anxiety, in part due to fear of disclosure of their HIV status, and alcohol abuse. Routine VL monitoring is an important tool to identify poor adherence and ART failure at an early stage, and with proper management this will prevent the accumulation of further resistance mutations and preserve treatment options ([@CIT0014]). Routine HIV VL monitoring is not currently available in the public sector in India; instead, CD4 count monitoring continues to be used as the main test to identify treatment failure. VL testing is available within the public sector on only a case-by-case basis, after being approved by an ART centre expert committee. The authors believe that routine HIV VL monitoring should no longer be considered as a luxury, but a necessity in HIV ART programmes, and that it should be rapidly scaled up in the public sector. Efforts should be made to ensure that VL testing is affordable and easily accessible in all resource-limited settings, including the removal of any barriers, such as reliance on a committee to sanction each and every VL test, as this can result in significant delay. Until such time that VL testing is performed routinely in those on ART, actual treatment failure is likely to be grossly underdiagnosed ([@CIT0015]). HIV resistance surveillance or genotyping in patients ([@CIT0016], [@CIT0017]) assists in designing appropriate treatment regimens for patients needing third-line ART. Thus, genotyping should be considered before regimen preparation for all those requiring a switch to a third-line ART regimen. The lack of access to affordable antiretroviral drugs that can be used in robust combinations is another barrier to management of those failing second-line ART. National ART programmes need to be prepared to offer third-line ART ([@CIT0018]), and they should already start preparing by making supplies of darunavir and raltegravir accessible. Furthermore, management of patients needing third-line ART can be complicated and may require palliative care ([@CIT0019]). A medical dilemma exists for those failing third-line ART regarding whether they should continue costly treatment that consists of many pills and is only weakly effective or switch to another ARV regimen that has a lower pill burden and is less expensive. The small number of patients in the study is one of the limitations, and therefore generalisation of the results may not be possible. However, this study is one of the few reports from resource-limited settings describing operational feasibility and programmatic challenges in relation to management of patients living with HIV who have suspected second-line ART failure and patients requiring third-line ART. There may be additional factors influencing treatment adherence in such patients; however, it would be difficult to comment on these factors on the basis of routine programme data. Conclusion {#S0004} ========== Early detection of ART failure through routine VL monitoring and HIV resistance surveillance should be prioritised in settings where ART programmes have been successfully established. Adequate adherence support and extensive counselling should be provided to all patients suspected of second-line ART failure since the underlying reason for suspected failure is often not genetic mutation of HIV but improper adherence by patients. HIV viral load testing should be offered after 3 months of enhanced adherence support in all patients with suspected failure based on a single VL measurement. With intensive psychosocial and medical intervention in those suspected to be failing second-line ART, an unnecessary switch to more expensive and often unavailable third-line ART can be averted in the majority of cases. However, there is still an urgent need for improved access to third-line ART regimens in India, for which the national ART programme should be prepared. The cost of such medications and the additional costs of VL monitoring and genotyping are currently major barriers to the optimal management of patients failing second-line ART. The authors are grateful for the contribution of healthcare workers from the MSF clinic and the Infectious Diseases Department, Mahatma Gandhi Medical College, as well as the patients living with HIV and their families. Authors' contributions {#S0005} ====================== SK and PI conceived and designed the experiments. SK and AA performed the experiments. PI, MD, and AA analysed the data. HM, AD, PI, and PS contributed reagents, materials, and analysis tools. SK, PI, MD, and AA wrote the paper. AD, HM, PS, PI, MD, and AA reviewed the final manuscript. Conflict of interest and funding {#S0006} ================================ The authors declare no conflicts of interest, and they have not received any funds for the study.

1. Introduction {#sec0005} =============== Pyoderma gangrenosum (PG) is a chronic, idiopathic, and rapidly evolving cutaneous ulcerative condition \[[@bib0005]\]. PG is considered a neutrophilic dermatosis that can affect any age group, however it has a peak incidence between ages of 20--50 \[[@bib0010]\]. It equally affects both sexes. Pathophysiology of the disease is not completely understood \[[@bib0015]\], however it is thought to involve up-regulation and aberrant trafficking of polymorphonuclear neutrophils (PMNs) with subsequent release of PMN-stimulating cytokines \[[@bib0020]\]. Although they can occur anywhere, the painful cutaneous lesions of PG most commonly occur on the lower extremities and trunk \[[@bib0015]\]. Typically, they begin as a tender pustule that becomes ulcerated with a raised, violaceous, undermined border and a cribriform base. Fever and malaise are often concurrent \[[@bib0025]\]. Although non-specific, common lab abnormalities in patients with PG can include leukocytosis, elevated erythrocyte sedimentation rate, and elevated C-reactive protein \[[@bib0015]\]. There have been a few reports of an elevated ANA titer \[[@bib0005]\]. A sterile incisional biopsy from the ulcer border including subcutaneous tissue is required to rule out an infectious etiology. Histopathology reveals mixed cellular inflammation with neutrophil predominance \[[@bib0030]\]. A biopsy from erythematous skin adjacent to the ulcer may reveal superficial and deep vascular injury with necrosis and a dense lymphocytic infiltrate \[[@bib0030]\]. Pathergy plays a key role in the pathogenesis and development of PG \[[@bib0035]\]. Pathergy is the induction or exacerbation of skin lesion after trauma. More severe trauma, such as a surgical procedure, can result in induction of PG through the release of cytokines and PMN chemotaxis \[[@bib0040]\]. Herein, we present three cases of post-operative PG, whose course would likely have been curtailed by prompt diagnosis and initiation of appropriate treatment. This work has been reported in line with the SCARE criteria \[[@bib0045]\]. 2. Presentation of cases {#sec0010} ======================== 2.1. Case 1 {#sec0015} ----------- A 41-year-old male presented to the hospital with progressive painful ulcerations on the left lower extremity and right arm following endovenous varicose vein ablation. The lesions progressively worsened over a 4-week period despite broad-spectrum antibiotics and surgical debridement. Physical exam revealed an ulcer on left lower extremity measuring 65 × 25 × 5 cm and on the right ventral forearm measuring 7 × 3.8 × 1.3 cm. Ulcerations exhibited deep violaceous undermined borders and granulation tissue and exposed muscle at base. The patient subsequently developed leukocytosis and fever. Blood, urine, and wound bacterial and fungal cultures didn't yield any infectious organisms. Four weeks after his initial endo-venous procedure, dermatology was consulted and performed an incisional skin biopsy. The histopathology showed epidermal ulceration with underlying superficial and deep neutrophilic inflammation extending to the deep subcutaneous adipose tissue lobules and connective tissue septae, findings suggestive of PG. Patient was managed with wound care, systemic steroids, and cyclosporine. Progression of ulcerations was quickly halted following initiation of treatment ([Fig. 1](#fig0005){ref-type="fig"}).Fig. 1Case 1. A: Ulceration shows violaceous undermined border B: Ulceration at site of venipuncture C: Histopathology at low magnification shows superficial and deep suppurative inflammation extending to deep subcutaneous tissue D: Histopathology at high magnification shows abundant neutrophils.Fig. 1 2.2. Case 2 {#sec0020} ----------- A 55-year-old Caucasian male with a history of multiple amputations of the right lower extremity due to peripheral vascular disease presented to the hospital with a progressive painful ulceration involving the right lower extremity stump following right below knee amputation. Physical exam revealed a 14 × 12 cm ulcer with a rolled border, granulation tissue and a black eschar at the base. The ulceration was progressively enlarging despite antibiotic treatment, surgical debridement and negative tissue and blood cultures. One year after his initial amputation dermatology was consulted. Skin biopsy revealed a neutrophilic infiltrate and leukocytoclastic vasculitis suggestive of PG. Patient was immediately started on systemic steroids and noticed significant improvement of pain. The size of ulcer decreased gradually during treatment ([Fig. 2](#fig0010){ref-type="fig"}).Fig. 2Case 2. A: Ulceration at stump site shows undermined erythematous border and a cribriform base B: Histopathology at low magnification shows superficial and deep suppurative inflammation extending to subcutaneous tissue C: Histopathology at high magnification shows mainly neutrophils and leukocytoclastic vasculitis.Fig. 2 2.3. Case 3 {#sec0025} ----------- A 30-year-old Caucasian male presented to the hospital with a history of recurrent wounds and abscesses on his face and prior surgical sites on his trunk over a period of 3.5 years. Ulcers continued to progress despite broad spectrum antimicrobial coverage, multiple attempts of incision and drainage and surgical debridement. He subsequently underwent successful skin grafting; however, surgical sites still remained complicated by non-healing ulcerations. Physical exam revealed large ulcers with undermined wound edges and surrounding violaceous, blue grey erythema on the mid sternum, left flank, and bilateral pre-auricular regions. Tissue aspirates and blood cultures only grew staphylococcus epidermidis, which was considered a contaminant. Three and a half years after developing his initial lesions, dermatology was consulted. A tangential biopsy was performed and showed a diffuse neutrophilic infiltrate consistent with PG. He was then started on systemic steroids with subsequently rapid improvement of his condition. Colonoscopy was performed and the patient was diagnosed with IBD. Patient was successfully managed with Adalimumab for both his PG and IBD ([Fig. 3](#fig0015){ref-type="fig"}).Fig. 3Case 3. A: Ulceration at previous surgical sites show undermined erythematous ulcers with surrounding violaceous, blue grey erythema B: Histopathology at low magnification shows suppurative inflammation extending into the deep subcutaneous tissue C: Histopathology at high magnification shows sheets of neutrophils with focal necrosis and hemorrhage.Fig. 3 3. Discussion {#sec0030} ============= Post-surgical PG (PSPG) has been noted to develop at the site of surgery within 7--11 days of surgical intervention \[[@bib0005],[@bib0050]\]. It may be mistaken for necrotizing fasciitis due to the initial erythema and disproportional pain \[[@bib0055], [@bib0060], [@bib0065], [@bib0070], [@bib0075]\]. Both misdiagnosis, most commonly as infection, and aggressive surgical intervention lead to rapid progression of PSPG \[[@bib0060]\]. Baranska-Rybak et al. found that 5 out of 12 patients underwent surgical intervention before the diagnosis of PSPG was made \[[@bib0065]\]. In over 50% of cases, PG is associated with systemic diseases such as inflammatory bowel disease, rheumatoid arthritis, and myeloproliferative disorders \[[@bib0080]\]. PSPG, on the other hand, has been found to be associated with systemic disease in 22%--35% of cases, the most common comorbidity being myeloproliferative disorders \[[@bib0045]\]. This leaves the majority of patients with no predisposing comorbidity other than surgery. Correctly and promptly diagnosing PSPG is of the utmost importance in reducing its often destructive course. Ahronowitz et al. \[[@bib0085]\] advised obtaining a complete patient history and physical examination, culturing the tissue to exclude infectious causes, and collecting skin biopsies containing the active border of the ulcer and penetrating deep into the subcutaneous tissue. Histology of the wound will show a non-specific inflammatory infiltrate without bacteria, however it is important to remember that there may be colonization of the wound or secondary bacterial infection that can make it challenging to differentiate from primary infection \[[@bib0060]\]. Binus et al. demonstrated that 74 patients with PG had tissue samples evaluated and 30% contained at least one single colony of bacteria \[[@bib0035]\]. A lack of response to antibiotics and surgical debridement along with repeated negative bacterial cultures should prompt consideration of PG. Management of PSPG is similar to classic PG. Surgical debridement is not generally recommended because it may induce pathergy and persistence and/or expansion of the ulcers. In addition to wound care and pain management, systemic corticosteroids and cyclosporine alone or in combination are considered first-line therapies \[[@bib0050]\]. Hyperbaric oxygen, cyclophosphamide, methotrexate, mycophenolate mofetil, sulfasalazine, and azathioprine are considered other treatment options \[[@bib0070]\]. More recently targeted therapies are proven effective in management of PG, namely the tumor necrosis factor alpha antagonists, infliximab and adalimumab \[[@bib0090]\]. There have been few reports in the literature of successful grafts and flaps to cover primary defect after institution of immunosuppression. However, this treatment option remains highly controversial \[[@bib0075]\]. 4. Conclusion {#sec0035} ============= PG remains a diagnosis of exclusion and one not often seen by surgical specialties outside of Dermatology. It can mimic many other cutaneous conditions including bacterial infection, vascular occlusive disease and chronic non-healing wounds, which often lead to misdiagnosis and aggressive surgical debridement. The diagnosis is often delayed despite inadequate response to antibiotics, surgical debridement, and negative bacterial cultures. Recognizing the clinical features of PG and its pathergenic nature while ensuring timely management is fundamental to preventing severe destruction and deformity. Conflict of interest {#sec0040} ==================== No conflict of interest for any of the authors. Funding {#sec0045} ======= No funding provided. Ethical approval {#sec0050} ================ Ethical approval was obtained from Ochsner Health Care System and University of Texas Medical Branch Health Care System. Consent {#sec0055} ======= Written consent was obtained from all three patients reported in this case series to have their case and pictures published in the journal. Author contribution {#sec0060} =================== Rawaa Almukhtar, M.D. contributed to study concept, collection of clinical history, physical exam, biopsy, any other investigations, interpretation of data, and manuscript writing and review. Julie Martin, M.D. contributed to study concept, collection of clinical history, physical exam, interpretation of data, and manuscript review. Bethany Vincent, M.D. contributed to study concept, interpretation of data, and manuscript review. Brain Lee, M.D. contributed to study concept, interpretation of data, and manuscript review. Andrew Armenta contributed to study concept, collection of clinical history, physical exam, biopsy, any other investigations, interpretation of data, and manuscript writing and review. Brandon Goodwin collection of clinical history, physical exam, biopsy, any other investigations, interpretation of data, and manuscript review. Mara M. Dacso contributed to study concept, collection of clinical history, physical exam, biopsy, any other investigations, interpretation of data, and manuscript writing and review. Guarantor {#sec0065} ========= Rawaa Almukhtar, Andrew Armenta, Brain Lee, Julie Martin, Brandon Goodwin, Bethany Vincent, and Mara M. Dacso.

Background {#Sec1} ========== In contemporary clinical management of patients with breast cancer, prognostications and therapeutic decisions are based on the assessment of clinicopathological factors as well as on the expression status of biomarkers with established clinical validity (i.e., human epidermal growth factor receptor 2 \[HER2\]; estrogen receptor \[ER\]; progesterone receptor \[PgR\]; and Ki67, a marker of cell proliferation) \[[@CR1], [@CR2]\]. Currently, the most commonly applied method for the determination of these four markers is immunohistochemistry (IHC), which allows for the semiquantitative assessment of the protein expression levels on histological slides \[[@CR3], [@CR4]\]. For HER2, an additional analysis of the amplification status of the corresponding gene *ERBB2* by fluorescence in situ hybridization (FISH), chromogenic in situ hybridization (CISH), or silver in situ hybridization (SISH) can also be applied in selected cases. The quality of the determination of these markers in terms of accuracy and reproducibility is essential for effective therapeutic interventions. However, the inter- and intraobserver variability of IHC is of concern \[[@CR3]--[@CR9]\]. For HER2, ER, and PgR, several studies have reported discrepancies of up to 20% \[[@CR5]--[@CR7]\], but most prominent and challenging is the inconsistency regarding Ki67 \[[@CR8], [@CR9]\]. Ki67 is a marker of the proliferative activity of the tumor cells and thereby carries valuable prognostic information \[[@CR10]--[@CR12]\]. In addition, Ki67 may have a direct impact on therapeutic decisions by assisting in the distinction between luminal A and luminal B breast cancer and therefore may aid in the selection of cytotoxic chemotherapy in addition to endocrine treatment \[[@CR2], [@CR13]\]. The variability in Ki67 is due mainly to the subjectivity of the visual estimation method and the choice of areas of evaluation on the histological slides and, to a lesser extent, the technical variations in the IHC staining process \[[@CR9], [@CR14]\]. Efforts to standardize Ki67 scoring resulted in considerable improvements, but interobserver agreement is still unsatisfactory \[[@CR15], [@CR16]\]. In addition, implementation of these methodological advances in clinical routine laboratories is challenging, and clinical validity of the new methods remains to be shown. For these reasons, the Ki67 determined by IHC is not currently included in the American Society of Clinical Oncology/College of American Pathologists guidelines for routine clinical use \[[@CR1], [@CR17]\]. There remains an urgent need for alternative, more robust, standardized, and precise assays with proven analytical and clinical validity for Ki67, HER2, ER, and PgR in routine breast cancer diagnostics \[[@CR17], [@CR18]\]. The MammaTyper® (BioNTech Diagnostics, Mainz, Germany) is a novel CE-marked in vitro diagnostic test that quantifies the messenger RNA (mRNA) expression of the four key marker genes *ERBB2*, *ESR1*, *PGR*, and *MKI67* on the basis of reverse transcription-quantitative real-time polymerase chain reaction (RT-qPCR), which differs from the currently applied standard of protein-based semiquantitative assessment by IHC. The main goal in using this technology is to provide a precise and reproducible assessment of the four biomarkers. Similarly to IHC, the MammaTyper® test can be integrated into the local laboratory setup because it supports analysis on widely accessible qPCR platforms using total RNA extracted from clinical routine formalin-fixed, paraffin-embedded (FFPE) breast cancer samples from resections or core needle biopsies. In this study, we assessed the precision of the MammaTyper® test with a focus on reproducibility \[[@CR19]\]. We adopted a multicenter design to fully evaluate the inter- and intrasite components of precision as well as other sources of imprecision, including preanalytical factors. Ten international pathology institutions, all with expert-level background in the field of breast cancer diagnostics, participated in the study. Each site carried out the same technical procedures according to a predefined study plan based on the EP05-A3 guideline for precision evaluation of quantitative measurement methods issued by the Clinical & Laboratory Standards Institute \[[@CR20]\]. To our knowledge, a similar study has not been conducted to date. Methods {#Sec2} ======= Study objectives {#Sec3} ---------------- The precision (reproducibility) of the MammaTyper® test was evaluated on multiple levels according to the following parameter definitions:Intermediate precision, here also referred as *interrun precision*, as the variability of quantitative results across repeated measurements over several days by the same operator, in the same laboratory, and using the same instrument; this parameter also included repeatability, the variance component due to simple replicates (intrarun)Intersite reproducibility, as the most comprehensive demonstration of precision, including the variability introduced by different laboratories, operators, and instrumentsPreanalytical and lot-to-lot variabilityAgreement of binary single-marker results and subtypesInterclass correlation coefficient (ICC) as the agreement of quantitative results Study design {#Sec4} ------------ A prospective, two-stage study was designed with the participation of ten international pathology institutions (*see* authors' affiliations 1--9 and 14). Prior to the study start, one operator per site was trained on the correct use of the preanalytical RNA extraction kit RNXtract® (BioNTech Diagnostics) and the MammaTyper® test within a 2-day standard training phase carried out by the manufacturer. This training also included qualification of the local qPCR instrument for use with the MammaTyper®, which in this study was the LightCycler® 480 instrument II (Roche Molecular Diagnostics, Pleasanton, CA, USA). The training was followed by a familiarization period consisting of at least four MammaTyper® runs on 3--4 days using BioNTech Diagnostics' reference material, carried out by the operator without supervision. During the study, each site performed repeated MammaTyper® measurements on different days according to a predefined study plan using RNA extracts from clinical FFPE breast cancer tissues. The same MammaTyper® lot was used at all sites, and only one site repeated study arm 1 using a second lot of MammaTyper®. The study comprised 8 days in total (consecutive or nonconsecutive days), as illustrated in the study design (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Study design. \*One site measured additionally a second set of samples 1--8 using a second lot of MammaTyper®. *FFPE* Formalin-fixed, paraffin-embedded ### Study arm 1 {#Sec5} RNA was extracted at a central laboratory (BioNTech Diagnostics), and eight different RNA pools, each containing RNA from a single tumor sample, were provided as single-use aliquots to the study sites (samples 1--8). Samples were measured repeatedly on 4 different days using MammaTyper®. ### Study arm 2 {#Sec6} Ten-micrometer sections of 16 FFPE tissue samples from different breast tumors (samples 9--24) were provided to each study site for local RNA extraction using RNXtract®. After extraction, each RNA eluate was split into three single-use aliquots for repeated MammaTyper® measurements on 3 different days. Samples {#Sec7} ------- The samples used in the study were prepared from clinical FFPE breast cancer tissue blocks by BioNTech Diagnostics and were distributed to study sites as RNA aliquots (samples 1--8) or 10-μm FFPE whole-tissue sections (samples 9--24). The 24 FFPE tissue samples were selected from a series of clinical routine breast cancer cases (*n* = 43) kindly provided by PSi. A summary of the clinicopathological characteristics of these patient samples is given in Additional file [1](#MOESM1){ref-type="media"}: Table S1. The use of archived samples was approved by the ethics committee of the University of Heidelberg (206/2005). Patient informed consent (and a specific approval for this study) was not necessary, because the ethics approval covers the use of samples for retrospective analysis. The selection process was based on sample validity (RNA amount) and *ERBB2*, *ESR1*, *PGR*, and *MKI67* marker expression obtained using the MammaTyper®. Different histological subtypes such as ductal, lobular, tubular, and micropapillary breast carcinomas were included in the cohort. To encompass the entire range of clinically anticipated expression levels, six samples were selected for each marker equally distributed from the lowest to the highest expression value (sextiles), including values close to the cutoff. From among the selected 24 FFPE tissue samples, 8 were chosen for preparation of the RNA pools; the other 16 were used for preparation of the FFPE tissue sections. Nineteen consecutive 10-μm sections were prepared per paraffin block, and homogeneity of the sections was ensured in a prequalification process by MammaTyper® analysis of the first, middle, and last sections (Additional file [2](#MOESM2){ref-type="media"}: Figure S1). RNA extraction {#Sec8} -------------- Total RNA was purified from 10-μm FFPE tissue sections using the paramagnetic particle-based RNXtract® RNA Extraction Kit (reference 90040; BioNTech Diagnostics GmbH) according to the manufacturer's instructions. The RNXtract® kit has been validated as a preanalytical RNA extraction method for the MammaTyper® by the manufacturer. MammaTyper® test {#Sec9} ---------------- The MammaTyper® (reference 90020; BioNTech Diagnostics GmbH) is a molecular in vitro diagnostic RT-qPCR test for the quantitative detection of the mRNA expression status of the genes *ERBB2*, *ESR1*, *PGR*, and *MKI67* in human FFPE breast cancer tissue from resection or core needle biopsies with at least 20% tumor cell content using whole-tissue sections without macrodissection. Primary analysis outputs are the normalized, quantitative single-marker results given as 40^−∆∆Cq^ (quantification cycle) values on a continuous scale \[[@CR21]\]. The test also provides the status of each marker as a binary category (positive or negative) based on clinically validated marker- and device-specific cutoff values. The combination of the four binary single-marker results can be further translated into the molecular subtype of the given breast cancer sample according to the St. Gallen classification \[[@CR2]\] (i.e. luminal A-like, luminal B-like \[HER2-negative\], luminal B-like \[HER2-positive\], HER2-positive \[nonluminal\], and triple-negative \[ductal\]). The MammaTyper® test was performed according to the Instructions for Use 150528-90020 revision 3.0, applying the following cutoffs for the LightCycler® 480 II instrument \[40^−∆∆Cq^\]: *ERBB2* = 41.10, *ESR1* = 38.00, *PGR* = 35.50, and *MKI67* = 35.50. Statistical analysis {#Sec10} -------------------- The results were analyzed according to a predefined statistical analysis plan using SAS version 9.4 software (SAS Institute, Cary, NC, USA). The number of measurements (sample size) of the study was determined using simulations to achieve predefined levels of uncertainty using results of a previous method validation \[[@CR21]\]. On each study day, exported raw C~q~ values were directly transferred by the operator to the statistician. To reflect a realistic estimate of the test precision, statistical outliers were not excluded from the analyses. The precision of the quantitative single-marker assessments (40^−∆∆Cq^ values) was estimated by a random effects model II analysis of variance (ANOVA) with site as a random factor \[[@CR20]\]. Because the variability does not depend on 40^−∆∆Cq^ values, the sample was also included as a (trivial) random factor in the model, which allows averaging of the variance components over the samples:The intermediate precision referring to *interrun/day SD* is obtained as the residual SD in the ANOVA.The reproducibility was calculated as the intersite SD summarizing the condition of different sites, operators, and instruments. The total SD is also presented, calculated as the square root of the sum of residual and intersite variance components. Because the total SD is the precision as experienced in clinical practice, we decided to report this parameter as the main result as a conservative approach.The variance introduced by a different MammaTyper® lot obtained in a separate experiment was given as the interlot SD.Agreement of the categorical marker results and the breast cancer biological subtypes across all sites was evaluated using Fleiss' kappa statistics \[[@CR22]\]. According to the method of Landis and Koch \[[@CR23]\], the strength of the agreement was defined as follows: kappa \< 0.00 = poor, 0.00--0.20 = slight, 0.21--0.40 = fair, 0.41--0.60 = moderate, 0.61--0.80 substantial, and 0.81--1.00 = almost perfect.The ICC was estimated for the continuous scaled quantitative marker results and was used to evaluate the reproducibility and intermediate precision in relationship with the intersample variance using the approach proposed by Eliasziw et al. \[[@CR24]\]. Thus, and different from the kappa statistic, the ICC determines the agreement of measured quantitative values over the whole measurement range, independent of any cutoff point \[[@CR25]\]. The agreement is generally interpreted as follows: ICC \< 0.40 = poor, 0.40--0.74 = fair to good, and 0.75--1.00 = excellent \[[@CR26]\]. More stringent thresholds were recommended by Kirkegaard et al. \[[@CR25]\] for IHC assessments, with an ICC level of 0.7 regarded as the minimum acceptable standard, 0.8 as good, and ≥0.9 as excellent. The latter thresholds were applied in this study. In a final analysis, kappa and ICC values were simulated in a larger sample cohort using quantitative data from 769 breast cancer cases of the FinHer trial that had been measured previously by MammaTyper® \[[@CR27]\]. ICC values were calculated using the intersample variance of the larger cohort along with the intersite and residual variance of the present study. To estimate the kappa values for this cohort, 1000 simulated pairs of datasets were created by adding random noise to the 40^−∆∆Cq^ values according to the marker-specific total variance observed in this study. For each pair (2 × 769 values), kappa values for binary marker results and subtypes were calculated, resulting in 1000 kappa values, of which the median kappa as well as the 2.5% and 97.5% percentiles are reported. Results {#Sec11} ======= Intermediate precision {#Sec12} ---------------------- On the basis of MammaTyper® measurements of study arm 1 (Fig. [1](#Fig1){ref-type="fig"}), quantitative single-marker results were obtained as 40^−∆∆Cq^ values for *ERBB2*, *ESR1*, *PGR*, and *MKI67* and are presented in Fig. [2](#Fig2){ref-type="fig"} as box plots for each marker, sample, and study site. The intermediate precision for each marker at the individual site was computed over all samples, presented as interrun SD (Fig. [2](#Fig2){ref-type="fig"}, graphs at the *bottom*).Fig. 2Box plots depicting inter- and intrasite reproducibility. The box plots represent the distribution of the four MammaTyper® measurements of the eight centrally extracted RNA samples for each marker at the ten different study sites A--J (study arm 1). The box plots indicate the median 40^−∆∆Cq^ values by the *horizontal line* dividing the boxes, the first and third quartiles by the *lower* and *upper borders* of the boxes, and the minimum and maximum values by the *whiskers*. The graphs at the *bottom* show the interrun SD over all samples for each marker at the respective site, as well as the corresponding 95% CI. *ERBB2* (HER2) Human epidermal growth factor receptor 2, *ESR1* (ER) Estrogen receptor, *MKI67* Marker of proliferation Ki67, *PGR* (PgR) Progesterone receptor Intersite reproducibility {#Sec13} ------------------------- As indicated by the side-by-side box plots in Fig. [2](#Fig2){ref-type="fig"}, the 40^−∆∆Cq^ quantitative single-marker results of each individual sample were highly consistent across all ten study sites. The total SD of the measurements of the eight centrally extracted RNA samples (samples 1--8) was as low as 0.18 C~q~ for *PGR*, 0.29 C~q~ for *ERBB2* and *MKI67*, and 0.44 C~q~ for *ESR1* (Table [1](#Tab1){ref-type="table"}, *upper panel*). As demonstrated by the variance component analysis, the factor site (intersite SD) had less impact on the total imprecision (total SD) than the interrun/day variability within one laboratory (residual SD) (Table [1](#Tab1){ref-type="table"}, *upper panel*).Table 1Intersite reproducibility of centrally (samples 1--8) and locally (samples 9--24) extracted RNA samples as total SD, as well as intersite and residual SD (quantification cycle values)AnalyteIntersiteResidualTotalSD95% CISD95% CISDRNA pools Samples 1--8*ERBB2*0.160.12--0.210.240.22--0.270.29*ESR1*0.180.13--0.290.400.37--0.440.44*PGR*0.090.07--0.130.160.15--0.170.18*MKI67*0.210.17--0.260.210.19--0.230.29Self-extracted RNAs Samples 9--24*ERBB2*0.220.19--0.260.230.21--0.240.31*ESR1*0.250.20--0.310.360.34--0.390.44*PGR*0.120.10--0.160.190.18--0.200.23*MKI67*0.160.14--0.200.190.18--0.210.25RNA pools and self-extracted RNAs Samples 1--24*ERBB2*0.200.17--0.230.230.22--0.250.31*ESR1*0.220.19--0.270.380.36--0.400.44*PGR*0.110.10--0.140.180.17--0.190.21*MKI67*0.180.16--0.210.200.19--0.210.27*Abbreviations: ERBB2* (HER2) Human epidermal growth factor receptor 2, *ESR1* (ER) Estrogen receptor, *MKI67* Marker of proliferation Ki67, *PGR* (PgR) Progesterone receptor Intersite reproducibility including preanalytical variances {#Sec14} ----------------------------------------------------------- The total variance of marker results (40^−∆∆Cq^ values) in the self-extracted samples (study arm 2, samples 9--24) was almost identical to the variance seen for the RNA pool aliquots (samples 1--8) (Table [1](#Tab1){ref-type="table"}, *middle panel*). There was no additional variance or bias introduced by RNA extraction at local sites. Therefore, the intersite reproducibility was again computed on the whole sample set (samples 1--24), leading to a similar approximation of the total variability of single-marker 40^−∆∆Cq^ assessments with SDs between 0.21 and 0.44 C~q~ (Table [1](#Tab1){ref-type="table"}, *lower panel*). Performing the analysis of the eight RNA pool samples with a different MammaTyper® lot resulted in comparable quantitative values (Additional file [3](#MOESM3){ref-type="media"}: Figure S2). The interlot SD was almost completely covered by the existing interrun/day variability (residual SD), and its impact on the total variance was negligible (Additional file [4](#MOESM4){ref-type="media"}: Table S2). Individual laboratory deviations for all samples are also shown with Bland-Altman plots (Fig. [3](#Fig3){ref-type="fig"}). The average deviation at the respective site was in all cases close to zero, with values ranging from C~q~ −0.13 to 0.16 for *ERBB2*, −0.11 to 0.20 for *ESR1*, −0.15 to 0.19 for *PGR*, and −0.22 to 0.31 for *MKI67*.Fig. 3Bland-Altman plots representing inter- and intrasite reproducibility. Bland-Altman plots show the differences of the measured 40^−∆∆Cq^ values per sample, site, and marker (*y*-axis) against the respective mean over all measurements (sites and days) (*x*-axis). The 24 samples are represented by either 4 or 3 measurements for the centrally extracted (study arm 1) or self-extracted (study arm 2) RNA samples, respectively. The average deviation of measurements at the respective site is indicated by the *dashed horizontal line*. The *continuous vertical line* indicates the marker-specific cutoff. *ERBB2* (HER2) Human epidermal growth factor receptor 2, *ESR1* (ER) Estrogen receptor, *MKI67* Marker of proliferation Ki67, *PGR* (PgR) Progesterone receptor. **a**-**j** correspond to the ten study sites Binary single-marker and subtype agreement {#Sec15} ------------------------------------------ The binary single-marker results (positive/negative) for all measurements at the ten sites are displayed as counts for each sample in Table [2](#Tab2){ref-type="table"}, revealing a very high concordance. The 24 samples showed 100% concordance for *ERBB2* and for *PGR* and *ESR1* an equivocal assignment in only one and two samples, respectively. These cases exhibited a marker expression level near the cutoff, as indicated by the distance to cutoff value (Table [2](#Tab2){ref-type="table"}). This also explained the divergent measurements seen for *MKI67*, which biologically exhibits more samples near the cutoff because of its continuous distribution \[[@CR28]\]. Nevertheless, *MKI67* showed a high agreement because for most discrepant cases only 1 of 30 determinations was classified differently (Table [2](#Tab2){ref-type="table"}). Calculating the overall agreement of the categorical marker assessments resulted in kappa values of 1.00, 0.91, 0.94, and 0.94 for *ERBB2*, *ESR1*, *PGR*, and *MKI67*, respectively. Corresponding subtype assessments resulted in an almost perfect agreement with a kappa value of 0.90 (Table [3](#Tab3){ref-type="table"}). Discrepancies were observed for example for the luminal A-like and luminal B-like (HER2-negative) subtype, where discrimination using St. Gallen guidelines relies on *MKI67* marker expression \[[@CR2]\], which for the discrepant cases was very close to the cutoff (as described above).Table 2Binary single-marker agreements as positive/negative counts, and corresponding kappa values (distance of mean 40^−∆∆Cq^ value to marker-specific cutoff is shown for each sample)*ERBB2ESR1PGRMKI67*SampleCountsDistance to cutoff (C~q~)CountsDistance to cutoff (C~q~)CountsDistance to cutoff (C~q~)CountsDistance to cutoff (C~q~)NegativePositiveNegativePositiveNegativePositiveNegativePositive1**40**0−1.10**40**1.50**40**3.2**40**0−3.22**40**0−1.10**40**1.70**40**1.8**40**0−0.830**40**0.60**40**1.90**40**2.50**40**0.84**40**0−1.20**40**3.00**40**4.60**40**0.75**40**0−1.90**40**1.60**40**5.60**40**0.66**40**0−2.90**40**1.40**40**6.00**40**0.67**40**0−3.2**40**0−4.0**40**0−3.00**40**2.280**40**1.7**1228**0.20**40**2.90**40**1.390**30**1.3**30**0−3.2**30**0−2.30**30**1.1100**30**3.1**30**0−2.0**30**0−1.70**30**1.511**30**0−1.60**30**2.10**30**4.9**426**0.212**30**0−1.80**30**2.70**30**4.70**30**1.113**30**0−1.80**30**2.40**30**4.6**129**0.414**30**0−2.00**30**1.40**30**3.2**30**0−0.815**30**0−1.30**30**1.10**30**1.0**291**−1.216**29**0−1.50**29**2.50**29**2.90**29**0.717**30**0−1.10**30**1.40**30**0.80**30**1.118**30**0−1.30**30**2.50**30**4.2**30**0−1.219**30**0−1.50**30**2.80**30**2.8**129**0.420**30**0−2.6**30**0−0.90**30**3.3**291**−0.621**30**0−2.8**2010**0.00**30**3.4**291**−0.522**30**0−3.1**30**0−3.4**30**0−3.10**30**2.123**30**0−3.1**30**0−2.3**1614**0.00**30**2.2240**30**1.9**30**0−5.2**30**0−3.3**129**0.4kappa Statistic1.000.910.940.94*Abbreviations: ERBB2* (HER2) Human epidermal growth factor receptor 2, *ESR1* (ER) Estrogen receptor, *MKI67* Marker of proliferation Ki67, *PGR* (PgR) Progesterone receptorBoldface values represent summary of measurements Table 3Subtype agreement as counts, and corresponding kappa valueSampleNot definedHER2-positive (nonluminal)Luminal B-like (HER2-positive)Luminal B-like (HER2-negative)Triple-negative (ductal)Luminal A-like100000**40**200000**40**300**40**0004000**40**005000**40**006000**40**0070000**40**08**12**0**28**00090**30**0000100**30**000011000**26**0**4**12000**30**0013000**29**0**1**1400000**30**15000**1**0**29**16000**29**0017000**30**001800000**30**19000**29**0**1**20**30**0000021**20**0000**10**220000**30**023**14**000**16**0240**30**0000kappa Statistic0.90*Abbreviations: HER2* Human epidermal growth factor receptor 2Boldface values represent summary of measurements Intra- and interclass correlation {#Sec16} --------------------------------- ICC estimates of all markers were between 0.976 and 0.996 for the intralaboratory assessment (ICC_intra), and between 0.980 and 0.998 for the intersite reproducibility (ICC_inter) (Table [4](#Tab4){ref-type="table"}, *upper panel*), reflecting excellent agreement of the quantitative data. To exclude any effect of the sample selection on ICC results, the ICCs were again computed using the intersample variance observed in the 769 breast cancer cases of the FinHer trial \[[@CR27]\] along with the intersite and residual variance extracted from the present reproducibility study. The marker-specific ICCs in this analysis were nearly identical (Table [4](#Tab4){ref-type="table"}, *lower panel*).Table 4Interclass correlation coefficient based on depicted variance components, and summary of kappa valuesAnalyteIntersite varianceIntersample varianceResidual varianceICC_intraICC_interkappa Statistic*ERBB2*0.0392.8860.054**0.9820.9871.00***ESR1*0.0495.8890.144**0.9760.9920.91***PGR*0.0138.2790.031**0.9960.9980.94***MKI67*0.0321.5650.039**0.9760.9800.94**SubtypeN/A**0.90**AnalyteIntersite variance (present study)Intersample variance (FinHer data)Residual variance (present study)ICC_intraICC_interkappa Statistic (median)2.5--97.5% percentile*ERBB2*0.0393.7230.054**0.9860.9900.95**0.92--0.97*ESR1*0.0495.9410.144**0.9770.9920.97**0.96--0.99*PGR*0.0137.4610.031**0.9960.9980.97**0.96--0.99*MKI67*0.0321.7390.039**0.9780.9820.85**0.82--0.88SubtypeN/A**0.91**0.89-0.93*Abbreviations: ERBB2* (HER2) Human epidermal growth factor receptor 2, *ESR1* (ER) Estrogen receptor, *MKI67* Marker of proliferation Ki67, *PGR* (PgR) Progesterone receptor, *N/A* Not applicable*Note*: Intersite variance and residual variance are squares of SD presented in lower part of Table [1](#Tab1){ref-type="table"}. The *lower panel* of this table is a simulated analysis using the intersample variance of data from the FinHer trial cohort (described in detail in the text)Boldface values highlight ICC and kappa statistic Finally, the kappa values were simulated in this larger sample cohort and were found to be comparable to the ones observed in the smaller sample size of the present study (Table [4](#Tab4){ref-type="table"}, *lower panel*). Discussion {#Sec17} ========== This study addressed the question whether the recently developed molecular in vitro diagnostic MammaTyper® test could improve the reproducibility of the assessment of the four key routine breast cancer biomarkers *ERBB2* (HER2), *ESR1* (ER), *PGR* (PgR), and *MKI67* (Ki67). The routine diagnostic assessment of these markers, as well as the corresponding subtyping, is currently performed by semiquantitative IHC and FISH, CISH, and SISH assays \[[@CR1]--[@CR4]\]. IHC assays suffer from considerable inter- and intralaboratory variability, which particularly applies to the assessment of the valuable biomarker Ki67 \[[@CR8], [@CR9], [@CR15], [@CR16]\]. Therefore, it is of importance that new technologies carrying the potential for more accurate, reliable, and precise analysis of Ki67 expression are brought under consideration to overcome the persisting inconsistencies \[[@CR17], [@CR18]\]. In this multicenter study, we demonstrated that MammaTyper® shows excellent inter- and intralaboratory precision, both for the continuous quantitative single-marker measurements (40^−∆∆Cq^ values) and for the categorical positive/negative status and the breast cancer molecular subtype classification. These data therefore confirm the high analytical performance of the MammaTyper® that was previously reported in the original technical validation of the test \[[@CR21]\] but was shown in this study in a more comprehensive and challenging methodological setting. In our study, ten different laboratories were able to generate consistent and highly concordant test results after an initial training and a relatively short familiarization period. Overall, the test results were found to be independent of the preanalytical process and not influenced by the MammaTyper® lot. The source of imprecision of the quantitative measurements was related mainly to the general run-to-run variability rather than to the variance introduced by different laboratories. These observations were in line with the original technical validation report \[[@CR21]\]. The ICC values for all markers were above 0.976, which signifies excellent agreement of the quantitative data and suggests improved inter- and intrasite reproducibility achieved with qPCR compared with what has been documented previously for IHC \[[@CR9], [@CR16]\]. In studies on IHC reproducibility, the intersite agreement for Ki67 IHC displays an ICC of 0.59, which is below the minimum acceptable standard proposed by Kirkegaard et al. \[[@CR9], [@CR25]\]. Even when standardized Ki67 scoring methods on centrally stained histological slides were tested, the interobserver reliability on resection specimens reached ICCs of only 0.40 to 0.74, with kappa values ranging from 0.29 to 0.58 \[[@CR16]\]. Only training and precise calibration resulted in better Ki67 assessment on centrally stained tissue microarray slides (ICC 0.94) or centrally stained core-cut biopsies using a standardized scoring method (ICC 0.87), but this process is difficult to implement in routine clinical practice, and clinically important discrepancies persisted in the critical range of 10% to 20% Ki67-positive nuclei staining \[[@CR15], [@CR29]\]. The challenges in the standardization of Ki67 assessment include the variability in the selection of the tumor areas to be assessed, the technique used for nuclei counting, and the dilemma of the numerical cutoff for positivity, especially for large tissue sections \[[@CR8], [@CR9], [@CR15], [@CR16]\]. The highly promising reproducibility of the MammaTyper® was confirmed by a simulated analysis using MammaTyper® data obtained from 769 samples from the FinHer trial cohort \[[@CR27]\], verifying that the study samples were representative of the whole spectrum of routine clinical samples. The high values of the various reproducibility metrics in the present study are a result of both the underlying high degree of standardization of the MammaTyper® test, which minimizes the main sources of variability, and the adaptation of a fully objective assessment method (i.e., qPCR). Thus, the MammaTyper® assay has the potential to overcome the substantial and varying rates of inter- and intraobserver variability that may occur with IHC. This applies especially in samples where high-quality IHC is not readily available for diagnostic purposes. Analytical validity, such as reproducibility, is a prerequisite for accurate diagnostics, and its formal evaluation is required along with a test's clinical performance to allow conclusions on its potential use in clinical practice \[[@CR17], [@CR18]\]. In a previous clinical performance evaluation study, good concordance was shown between MammaTyper® single-marker assessments and IHC (or IHC/CISH for HER2), using 769 archived breast cancer cases available from the FinHer trial \[[@CR27]\]. Only for *MKI67* mRNA expression was the correlation moderate, most likely because of the analytical restriction of Ki67 IHC. The multivariable analysis revealed that *MKI67* expression assessed by MammaTyper®, but not Ki67 IHC, was an independent predictor of distant disease-free survival (DDFS), indicating the superiority of MammaTyper® compared with IHC with respect to *MKI67*/Ki67 determination \[[@CR27]\]. Furthermore, this clinical performance evaluation study demonstrated that the mRNA-based subtyping by MammaTyper® resulted in clinically meaningful prognostic and predictive information with regard to DDFS, overall survival, and response to taxane-based chemotherapy in the luminal B-like (HER2-negative) subtype \[[@CR27]\]. These published data provide evidence for the clinical validity of the test, and additional clinical performance evaluation studies would help further strengthen its clinical value for routine applications. As is true of all molecular assays that use tissue homogenates, one may also consider low tumor cell content or high lymphocytic infiltration as potential sources of error. For MammaTyper®, a minimum tumor cell content of 20% was required to generate stable test results when compared with the paired macrodissected sample \[[@CR21]\]. Adjacent nontumor tissue had no major influence on test results, likely because of the reduced metabolic activity and low RNA content in the surrounding tissue compared with the invasive tumor \[[@CR30], [@CR31]\]. As reported previously, nontumor components may have a stronger impact on multigene tests that analyze a recurrence score based on genes with partially notable expression in normal tissue \[[@CR32], [@CR33]\]. Nevertheless, further validations of MammaTyper® on samples with problematic characteristics (i.e., varying amounts of ductal carcinoma in situ and lymphocytic infiltrates) should be envisaged. The reliable quantification of single-marker expression by MammaTyper® has the potential to become part of a predictive marker panel in breast cancer diagnostics with further refined implications for clinical management. The expression of the single markers *ERBB2*, *ESR1*, *PGR*, and *MKI67* is obtained on a continuous scale covering a much broader dynamic range (up to 5 orders of magnitude) than can be achieved by IHC (up to 2 orders of magnitude; 0--100% positive cells or H-score 0--300) \[[@CR21], [@CR34]\]. An ongoing challenge in clinical management is the decision whether to treat patients with luminal breast cancer with systemic chemotherapy when other clinicopathological factors, such as nodal status, are not decisive \[[@CR17]\]. Various multigene assays address exactly this diagnostic dilemma. Oncotype DX® (Genomic Health, Redwood City, CA, USA), MammaPrint® (Agendia, Irvine, CA, USA), EndoPredict® (Myriad Genetics, Salt Lake City, UT, USA), and Prosigna® (NanoString Technologies, Seattle, WA, USA) provide risk scores with prognostic information for distant recurrence in patients with luminal A-like and luminal B-like (HER2-negative) breast cancers. High and low risk scores, even though they are mainly prognostic, are frequently used to decide for or against chemotherapy in the ER/PgR-positive, HER2-negative subgroup \[[@CR17]\]. Limitations exist, however, because the appropriate course of action remains unclear for patients with intermediate risk cancers or for node-positive patients. Prospective studies on multigene tests in breast cancer showing predictive values are limited \[[@CR35]--[@CR37]\]. The randomized phase III Microarray in Node-Negative and 1 to 3 Positive Lymph Node Disease May Avoid Chemotherapy (MINDACT) trial recently demonstrated that chemotherapy can be omitted in around 46% of clinically high-risk cases with low MammaPrint® scores \[[@CR37]\]. Similarly, the recurrence score measured with Oncoytpe DX® in the prospective randomized phase III Plan B study demonstrated excellent 3-year survival by omitting chemotherapy in clinically high-risk but recurrence score low-risk cases \[[@CR35]\]. Further prospective studies on Oncotype DX® testing revealed excellent 5-year survival (98%) in low-risk score cases treated with hormone therapy alone \[[@CR36], [@CR38]\]. However, these tests come with high costs. In addition, some of these tests require sending the samples to a central laboratory, such as Oncotype DX® and MammaPrint®. In this respect, it is interesting that a similar score, namely the immunohistochemical 4 (IHC4) score, with comparable prognostic value was generated by using just the four IHC markers HER2, ER, PgR, and Ki67 \[[@CR39], [@CR40]\]. However, insufficient standardization and considerable interlaboratory variability of IHC suggest that the IHC4 algorithm cannot easily be transferred to other laboratories, although it was successfully validated in an independent study cohort \[[@CR41]\]. On the basis of data obtained in this study, MammaTyper® could provide a highly reproducible and reliable assessment of these four markers. It is tempting to suggest that in the future a similar approach might be applicable for the MammaTyper® to generate additional prognostic information to guide personalized treatment options at much lower cost than multigene expression tests. Conclusions {#Sec18} =========== The MammaTyper® test has the potential to improve the quality of primary breast cancer molecular diagnostics. The test showed reliable reproducibility in the quantitative assessment of the single markers *ERBB2*, *ESR1*, *PGR*, and *MKI67*, as well as in the subtype determination, and thereby overcomes the variability known on the basis of diagnostic experience with IHC. The low intersite and intrasite variance of the MammaTyper® test enables pathology institutions to perform this assay in-house and integrate the technology into routine diagnostic services. However, additional clinical performance evaluation studies in larger cohorts are necessary to confirm the clinical utility of the test and to extract further predictive information for more personalized clinical management of patients with breast cancer. Additional files ================ {#Sec19} Additional file 1: Table S1.Clinicopathological characteristics of the breast cancer patient samples used in this study, including HER2, ER, PgR, and Ki67 marker status. (DOC 61 kb) Additional file 2: Figure S1.Prequalification of FFPE sections for study arm 2. Presented are 40-∆∆C~q~ values of section numbers 1, 10, and 19 for each marker to confirm homogeneous expression between sections. (PDF 553 kb) Additional file 3: Figure S2.Box plots depicting interlot reproducibility. The box plots represent the distribution of the four MammaTyper® measurements of the eight RNA pool samples using two different MammaTyper® lots at one site. The box plots indicate the median 40^−∆∆Cq^ values by the horizontal line dividing the boxes, the first and third quartiles by the lower and upper border of the boxes, and the minimum and maximum values by the whiskers. (PDF 1359 kb) Additional file 4: Table S2.Interlot reproducibility of centrally extracted RNA samples (samples 1--8) as total SD, and the variance components interlot and residual SD (in C~q~). (DOC 31 kb) ANOVA : Analysis of variance CISH : Chromogenic in situ hybridization C~q~ : Quantification cycle DDFS : Distant disease-free survival *ERBB2* (HER2) : Human epidermal growth factor receptor 2 *ESR1* (ER) : Estrogen receptor FFPE : Formalin-fixed, paraffin-embedded FISH : Fluorescence in situ hybridization ICC : Interclass correlation coefficient IHC : Immunohistochemistry IHC4 : Immunohistochemical 4 score *MKI67* : Marker of proliferation Ki67 mRNA : Messenger RNA *PGR* (PgR) : Progesterone receptor RT-qPCR : Reverse transcription-quantitative real-time polymerase chain reaction SISH : Silver in situ hybridization John Bartlett and Giuseppe Viale shared last authorship **Electronic supplementary material** The online version of this article (doi:10.1186/s13058-017-0848-z) contains supplementary material, which is available to authorized users. We thank Caroline Elzner (ACOMED statistik, Leipzig, Germany) for statistical analyses. We thank Cécile Aube, Amandine Bonhomme, Cheryl Crozier, Susanne Dettwiler, Britta Emming, Chunxiao Fu, Dr. Sabine Glombitza, Peng Guo, Stefanie Herlein, Ke Huang, and Fabiola Prutek for excellent technical assistance. This work was supported by BioNTech Diagnostics. We thank Karin Rohleder for managing the performance evaluation study. Funding {#FPar1} ======= The study was sponsored by BioNTech Diagnostics GmbH. Employees of BioNTech Diagnostics GmbH were involved in study design and management. Availability of data and materials {#FPar2} ================================== The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request. Authors' contributions {#FPar3} ====================== ZV, JB, GV, TK, ML, and US made substantial contribution to the conception and design of the study. ZV, PSc, AL, AT, HB, AH, RS, FPL, GV, EGR, FS, XT, RvW, and JB participated at study sites and performed RT-qPCR measurements. PSi provided the FFPE tissue samples. CG, ML, and KS prepared sample material and provided technical support during study execution. TK performed the statistical analysis. HJ provided data generated previously on samples of the FinHer trial. ZV drafted the manuscript. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== ZV provides consultancies or has an advisory role for Genomic Health Inc. and Roche Pharma. AL has received honoraria from Roche Pharma and Novartis; has performed a consulting or advisory role for Genomic Health Inc., Roche Pharma, and Menarini Diagnostics; and has received institute research funding from Sysmex Europe and BioNTech Diagnostics. AH has received honoraria from and performed a consulting and advisory role for Roche, MSD, Bristol-Myer Squibb, Novartis, AstraZeneca, Pfizer, Zytomed, and Medac and has received sponsoring of research studies from Novartis, Illumina, Archer, QIAGEN, and BioNTech Diagnostics. AT has received institute research funding from Sysmex Europe and BioNTech Diagnostics. TK has received payments from BioNTech Diagnostics for statistical design and analysis of the experiments. PSi provides consultancies or has an advisory role for Genomic Health Inc. and Roche Pharma. JB provides consultancies for Due North, BioNTech Diagnostics, and BioTheranostics Inc. US is the founder and chief executive officer of BioNTech Diagnostics GmbH. CG, ML, and KS are employees of BioNTech Diagnostics GmbH. The other authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== The use of archived samples was approved by the ethics committee of the University of Heidelberg (206/2005). Patient informed consent (and a specific approval for this study) was not necessary, because the ethics approval covers the use of samples for retrospective analysis. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1-ijerph-16-00603} =============== The increasing prevalence of unhealthy eating habits and sedentary lifestyles among adolescents, which often continues into adulthood, is a global public health problem \[[@B1-ijerph-16-00603],[@B2-ijerph-16-00603]\]. Increased intake of foods that are high in energy density, fat, and sugar and in combination with sedentary lifestyles (being physically inactive) and low levels of physical activity (PA) are major contributors to obesity in adolescents \[[@B3-ijerph-16-00603],[@B4-ijerph-16-00603],[@B5-ijerph-16-00603]\]. Negative longitudinal and secular trends have been reported in dietary behaviours among adolescents, such as an adverse change in dietary patterns and a decline in dietary quality during the transition from childhood to adolescence, and particularly decreases in fruit, vegetable, milk, and fruit juice consumption, and an increase in sugar-sweetened beverage (SSB) consumption \[[@B6-ijerph-16-00603],[@B7-ijerph-16-00603]\]. Furthermore, portion sizes and frequency of fast food and snack intake have increased in the past decade \[[@B8-ijerph-16-00603],[@B9-ijerph-16-00603]\]. There is an increasing trend of unhealthy eating among adolescents worldwide, which is contributing to an increase in the incidence of obesity \[[@B7-ijerph-16-00603]\] and the development of chronic diseases such as diabetes and cardiovascular disease \[[@B10-ijerph-16-00603],[@B11-ijerph-16-00603]\]. It has been revealed that the eating patterns of children and adolescents are related to characteristics of both the social and physical environment \[[@B12-ijerph-16-00603]\]. They are eager to eat foods that are easily available and reachable, and they have a tendency to eat higher amounts when larger food portions are prepared. Moreover, socioeconomic and sociocultural factors such as parents' educational level, ethnicity, time constraints, and mealtime structure including whether families eat together, the source of foods (e.g., schools, restaurants), and TV-viewing during meals are related to the eating patterns of children and adolescents \[[@B12-ijerph-16-00603]\]. In a recent review, feeding strategies and parental food habits were the most dominant determinants of a child's eating behaviour and food choice \[[@B12-ijerph-16-00603]\]. In addition, gender has also been found to be associated with dietary behavior in adolescents \[[@B2-ijerph-16-00603]\]; girls have greater or more frequent consumption of fruits and vegetables compared to boys \[[@B13-ijerph-16-00603]\]. An association between obesity and low intakes of fruits and vegetables and overweightness has been reported among this age group \[[@B14-ijerph-16-00603]\]. There were also positive relationships noted between parental encouragement, family rules, home availability, and fruit and vegetable consumption \[[@B15-ijerph-16-00603]\]. Globally, physical inactivity and low fitness in children and adolescents are contributing to a rising health burden \[[@B16-ijerph-16-00603]\]. Several systematic reviews have reported that adolescents spend most of their time engaged in sedentary activities \[[@B17-ijerph-16-00603],[@B18-ijerph-16-00603],[@B19-ijerph-16-00603]\]. Regular participation in PA has been shown to produce significant health benefits for adolescents such as obesity prevention, improved psychological well-being, cardiovascular fitness and bone health \[[@B20-ijerph-16-00603],[@B21-ijerph-16-00603]\]. Additionally, PA behaviours adopted during adolescence are likely to be maintained into adulthood \[[@B22-ijerph-16-00603],[@B23-ijerph-16-00603]\]. The global trend shows that girls are less active than boys \[[@B24-ijerph-16-00603]\] and there is a greater reduction in PA during adolescence for girls compared to boys \[[@B25-ijerph-16-00603]\]. Despite the health benefits linked to PA, a recent literature review has revealed that PA levels decline across the lifespan and particularly during adolescence. This is in line with findings reported in Malaysian adolescents \[[@B26-ijerph-16-00603]\]. Inadequate PA and increased prevalence of obesity among Malaysian adolescents, especially for rural female adolescents, is a public health concern \[[@B27-ijerph-16-00603]\]. In Malaysia, there are 5.5 million adolescents aged 10--19, which equates to 18.9% of the total population \[[@B28-ijerph-16-00603]\]. The prevalence of unhealthy lifestyles as well as the proportion of overweight and obese Malaysian adolescents have both increased in recent decades \[[@B29-ijerph-16-00603]\]. Thus, the promotion of PA and healthy eating in this age group has become an increasingly important priority in order to promote health, prevent disease, and reduce the prevalence of obesity \[[@B30-ijerph-16-00603]\]. In recent years, several observational studies have been conducted to understand which factors are related to the dietary patterns and PA of Malaysian adolescents \[[@B27-ijerph-16-00603],[@B31-ijerph-16-00603],[@B32-ijerph-16-00603]\]. However, there seems to be a lack of well-conducted studies and the various methodological approaches that have been adopted thus far provide insufficient evidence-based information regarding the dietary patterns and PA of Malaysian adolescents. Therefore, the main aim of this systematic review was to summarize the evidence from observational studies related to diet and PA in Malaysian adolescents in secondary schools (13--18 years old). Another aim was to understand the determinants of diet and PA and their associations with these behaviours. Hopefully, the findings of this systematic review will be used to develop interventions that might improve the eating patterns and levels of PA, as well as reduce the sedentary behaviours among adolescents in Malaysia. 2. Methods {#sec2-ijerph-16-00603} ========== The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement for systematic reviews was used to structure the present review and increase its integrity ([Table S1](#app1-ijerph-16-00603){ref-type="app"}).The review protocol was registered in the International Prospective Register of Systematic Reviews database (PROSPERO) (registration number: CRD42017074556). 2.1. Literature Search {#sec2dot1-ijerph-16-00603} ---------------------- One reviewer implemented the search strategy to search for relevant articles published from August 1990 up through August 2017 on the PubMed, Science Direct, Cochrane Review and Web of Science. Two reviewers then independently screened the identified studies against inclusion criteria and before extracting data. The search strategy was inclusive and concentrated on three key elements: population (e.g., Malaysian adolescents, secondary school), observation (e.g., diet and PA) and outcomes (e.g., associations with diet and PA behaviours) ([Text S1](#app1-ijerph-16-00603){ref-type="app"}). 2.2. Selection of Studies {#sec2dot2-ijerph-16-00603} ------------------------- This systematic review included all observational studies that looked at the determinants of diet and/or PA and their associations with these behaviours among Malaysian adolescents. The studies selected for analysis were required to meet the following inclusion criteria: (i) reported associations between diet and/or PA factors and diet and/or PA behaviours; (ii) involved healthy Malaysian adolescents aged between 13 and 18 years; (iii) were fully published articles. Studies were excluded based on the following criteria: (i) population studies that included children younger than 13 years of age or adults older than 18 years of age; (ii) papers that were not peer-reviewed or were only abstracts; (iii) studies that recruited adolescent subjects on the basis of disease, e.g., obese, diabetic, etc.; (iv) studies that did not report on the associations between diet/PA determinants and the diet/PA behaviour as the outcomes; (v) studies that included subjects that were not Malaysian. In cases where the selected papers did not present all the necessary information, the corresponding authors were contacted to obtain further details in order to establish the eligibility of the paper for inclusion in this systematic review. To identify the appropriate studies, one reviewer (S.M.) reviewed all the titles and the abstracts retrieved through the database searches. Then, an initial screening of the title and the abstract was conducted by two reviewers (S.M. and Z.T.) against the above-mentioned exclusion criteria. Next, an evaluation of full-text articles was conducted by the review team (S.M. and Z.T.) to refine the results based on the inclusion criteria. Disagreements between the reviewers were resolved through discussion with L.J. and H.A.M. Among the 25 studies that were initially captured, seven studies were excluded because they reported mixed age ranges or some adolescents below 13 years old. 2.3. Quality Assessment {#sec2dot3-ijerph-16-00603} ----------------------- The quality of the studies was assessed using the modified Newcastle-Ottawa Scale for cohort and cross-sectional studies \[[@B33-ijerph-16-00603]\] ([Text S2](#app1-ijerph-16-00603){ref-type="app"}). It was developed to assess the quality of studies in three domains selection of study groups, comparability of groups and ascertainment of exposure and outcomes. Two reviewers (S.M. and Z.T.) did pilot testing of the quality assessment tools independently on four of the included studies, and then completed the evaluation of the quality of the remaining studies, independently. Where there were disagreements, the two reviewers discussed the issues until consensus was achieved. Studies were determined as being good, fair, or poor. "Good quality "comprised three or four stars in the selection domain, one or two stars in the comparability domain, and two or three stars in the outcome/exposure domain. "Fair quality" had two stars in the selection domain, and one or two stars in the comparability domain and two or three stars in the outcome/exposure domain. "Poor quality" had zero or one star in the selection domain, zero stars in the comparability domain, and zero, or one star in the outcome/exposure domain \[[@B33-ijerph-16-00603]\]. [Tables S2 and S3](#app1-ijerph-16-00603){ref-type="app"} provide a summary of the quality assessment. 2.4. Data Extraction {#sec2dot4-ijerph-16-00603} -------------------- Data was extracted by two independent reviewers (S.M. and Z.T.) into a standardized Excel spreadsheet (Microsoft Corporation, Albuquerque, NM, United States) and included a range of study characteristics like author, year of publication, study setting, study design, methodology, sample size, participants'demographics (e.g., age, ethnicity, maternal education, income), physical activity, dietary assessment, covariates, as well as the main exposures and outcomes. Data was also extracted on specific diet/PA factors, associations between the potential determinants of diet/PA and diet/PA behaviours, and characterization of outcomes (diet, PA). In addition, differences between reviewers were infrequent (concordance \> 95%). Discrepancies were resolved through discussion by consensus. 2.5. Data Synthesis {#sec2dot5-ijerph-16-00603} ------------------- Associations were considered significant when the *p*-value in the study was \< 0.05. For those studies that performed both univariate and multivariate analyses, only the multivariate results were reported in this review. This was done because the results of multivariate analyses provide more accurate associations, and because they control for other potential confounding variables. Also, conceptually similar correlates and determinants of diet and PA were combined to reduce the number of variables considered in this review. 3. Results {#sec3-ijerph-16-00603} ========== The initial multi-database search yielded 2410 publications. From this, 18 were included in this systematic review ([Figure 1](#ijerph-16-00603-f001){ref-type="fig"}). The selected studies focused on associations between the determinants of diet and PA as well as their associations with diet and PA behaviours in Malaysian adolescents. Seventeen studies were cross-sectional \[[@B31-ijerph-16-00603],[@B34-ijerph-16-00603],[@B35-ijerph-16-00603],[@B36-ijerph-16-00603],[@B37-ijerph-16-00603],[@B38-ijerph-16-00603],[@B39-ijerph-16-00603],[@B40-ijerph-16-00603],[@B41-ijerph-16-00603],[@B42-ijerph-16-00603],[@B43-ijerph-16-00603],[@B44-ijerph-16-00603],[@B45-ijerph-16-00603],[@B46-ijerph-16-00603],[@B47-ijerph-16-00603],[@B48-ijerph-16-00603],[@B49-ijerph-16-00603]\]; only one was a prospective cohort study \[[@B27-ijerph-16-00603]\]. [Table 1](#ijerph-16-00603-t001){ref-type="table"} provides a summary of each study. From the table, it can be seen that the sample sizes of the studies ranged from 156 to 40,011 adolescents. Seventeen studies included both boys and girls (*n* = 17); one study was conducted among girls \[[@B31-ijerph-16-00603]\]. The majority of the studies included adolescents from different ethnicities, i.e., Malay, Indian and Chinese (*n* = 12). Nearly half of the studies were conducted in central Malaysia and Kuala Lumpur (*n* = 8). Where diet was measured, this was done using food frequency questionnaires (*n* = 3), and 24-h diet recalls (*n* = 3) followed by diet histories (*n* = 1) and other nutrition-related questionnaires. As for PA, this was self-reported via the Physical Activity Questionnaire for Children (PAQ-C) in nearly half of the studies (*n* = 8). Five studies were conducted in urban areas and three took place in both urban and rural areas. The rest of the studies (*n* = 10) did not specify the location of the data collection. The majority of studies (*n* = 9) did not report maternal education; for the nine studies that did report on this maternal education, the majority of mothers had a secondary school level of education. Eight studies did not report on household income. The other ten studies reported that the majority of adolescents lived in households with a moderate level of household income. Gender was the most frequently adjusted covariate in only four studies \[[@B34-ijerph-16-00603],[@B40-ijerph-16-00603],[@B44-ijerph-16-00603],[@B49-ijerph-16-00603]\]. Based on the quality assessment criteria, only one study was rated as good \[[@B34-ijerph-16-00603]\] while the others were rated as low quality ([Table S1](#app1-ijerph-16-00603){ref-type="app"}). 3.1. Dietary Determinants of Dietary Behaviours {#sec3dot1-ijerph-16-00603} ----------------------------------------------- [Table 2](#ijerph-16-00603-t002){ref-type="table"} summarizes the associations between the determinants of diet and dietary behaviours. The factors related to dietary behaviours were investigated in nine studies. The outcomes were grouped into three categories: energy and nutrients, dietary patterns, and foods. Gender and ethnicity were the most investigated correlates of diet. "Dietary pattern" was the most commonly assessed dietary behaviour in association with diet. ### 3.1.1. Energy and Nutrients {#sec3dot1dot1-ijerph-16-00603} According to the included studies, the significant correlates of energy and nutrient intakes were gender, place of residence, and meal and snacking patterns. • Energy Intake Five studies investigated the association between energy intake and the following potential determinants of diet: gender, place of residence, meal and snacking patterns. Four of the five studies made no adjustment for confounders \[[@B37-ijerph-16-00603],[@B38-ijerph-16-00603],[@B39-ijerph-16-00603],[@B48-ijerph-16-00603]\]. Only one of the five studies focused on the possibility of a link between energy intake and eating away from home but found no significant association \[[@B49-ijerph-16-00603]\]. The three studies that focused on the potential association between energy intake and gender reported that males consumed significantly more energy than female adolescents \[[@B37-ijerph-16-00603],[@B38-ijerph-16-00603],[@B48-ijerph-16-00603]\]. One study investigated the relationship between energy intake and place of residence and reported that energy intake was significantly higher among rural adolescents compared to their urban counterparts \[[@B48-ijerph-16-00603]\]. One study showed with the greater frequency of having snacks, the amount of energy consumed was greater \[[@B39-ijerph-16-00603]\]. • Macronutrient Intakes Three studies investigated the association between macronutrient intakes of carbohydrates, protein, total fat and unsaturated and saturated fat, as well as some potential determinants of diet (i.e., meal and snacking pattern, gender and place of residence) \[[@B39-ijerph-16-00603],[@B48-ijerph-16-00603],[@B49-ijerph-16-00603]\]. Two of the three studies showed significant associations between macronutrient intakes of carbohydrates, proteins and fats with meal and snacking pattern \[[@B39-ijerph-16-00603],[@B49-ijerph-16-00603]\] and gender and place of residence without adjustment for confounders \[[@B48-ijerph-16-00603]\]. However, fat intakes were not associated with place of residence \[[@B48-ijerph-16-00603]\]. It was also reported that male adolescents had significantly greater macronutrient intakes compared to females and that rural adolescents had higher carbohydrate and protein intakes than adolescents from urban areas \[[@B48-ijerph-16-00603]\]. In addition, there was a significant difference between the meal and snacking patterns and macronutrient intakes. Moreover, more frequent snack intakes led to more carbohydrate intake as compared to protein and fat intake \[[@B39-ijerph-16-00603]\]. Furthermore, one study did not show any significant association between eating away from home and carbohydrate and protein intakes while the energy-adjusted fat intake was significantly higher in more frequent eating away from home after adjusting for confounders such as gender, ethnicity, household income and body mass index (BMI) \[[@B49-ijerph-16-00603]\]. • Other Nutrients One study examined unadjusted associations between fibre, sugar and dietary cholesterol with gender and adolescents' place of residence \[[@B48-ijerph-16-00603]\]. The study found no association between the intake of sugar and fibre with gender and their place of residence. However, it did show that rural adolescents had a significantly higher intake of dietary cholesterol compared to their urban counterparts. ### 3.1.2. Foods {#sec3dot1dot2-ijerph-16-00603} In the reviewed studies, foods were grouped into two categories: SSBs and food groups. • Sugar-Sweetened Beverage (SSB) Intake One cross-sectional study focused on the association between SSB intake and ethnicity, gender and maternal education without adjusting for covariates. The study found no association with gender and maternal education. In this study based on unadjusted associations, only significant differences were observed for SSB intake with ethnicity, where SSB intake was highest among Malay adolescents followed by Indian and Chinese adolescents \[[@B36-ijerph-16-00603]\]. • Food Groups One study examined the association between food groups and gender, in which the component scores of fish, fruit, vegetables, milk and milk products for the healthy eating index (HEI) were found to have significant mean and median differences between male and female adolescents without any adjustment for confounders \[[@B35-ijerph-16-00603]\]. However, there was no association between gender and the following food groups: cereals and grains, legumes, poultry, meat and eggs. • Dietary Patterns In the studies considered in this review, dietary patterns were grouped into five categories: diet quality, healthy, Western and local dietary pattern score, meal skipping behaviours, snacking and meal frequency. The significant correlates of dietary patterns included the availability of healthy foods, ethnicity, gender, age, self-efficacy for healthy eating, PA, eating out, fast food consumption, maternal education, snacking practices, nutritional supplement consumption, breakfast skipping, soft drink consumption, household income and eating companion. • Diet Quality Only one study looked for correlates of diet quality by using the HEI for Malaysia in order to evaluate the diet quality of Malaysian adolescents \[[@B35-ijerph-16-00603]\]. Without adjusting for covariates, the study found that five correlates (i.e., the availability of healthy foods, ethnicity, gender, age and self-efficacy for healthy eating) were significantly associated with diet quality while the frequencies of consuming breakfast, lunch and dinner were not associated with diet quality. The study also found that male adolescents had significantly lower mean composite scores on the HEI in comparison to females, indicating that males had a poorer diet quality \[[@B35-ijerph-16-00603]\]. In addition, Malay adolescents had a significantly lower composite HEI score compared to Indian adolescents \[[@B35-ijerph-16-00603]\]. The study also showed that there was a significant positive correlation between age and diet quality of the participants \[[@B35-ijerph-16-00603]\]. Moreover, the total score for self-efficacy for healthy eating was correlated weakly and positively with the diet quality of the adolescent participants \[[@B35-ijerph-16-00603]\]. Lastly, the study concluded that the low availability of healthy foods may have been contributed to the poor diet quality noted among the adolescents \[[@B35-ijerph-16-00603]\]. • Healthy, Western and Local Dietary Pattern Scores One study investigated the associations between the dietary patterns and diet correlates of Malay and Chinese Malaysian adolescents \[[@B34-ijerph-16-00603]\]. The study found that Malay adolescents had significantly higher scores for the Western-based food pattern and the local-based food pattern, whereas the Chinese adolescents had higher scores for the healthy-based food pattern. The multivariate analyses that were conducted by the study after adjusting for potential confounders revealed that, in the case of the Malay adolescents, age (Beta = 0.141, SE = 0.033; *p* \< 0.001) and PA level (Beta = 0.142, SE = 0.036; *p* \< 0.001) were positively associated with the healthy-based food pattern, whereas higher frequencies of eating out (away from home) (Beta = −0.088, SE = 0.036; *p* = 0.014) and fast food consumption (Beta = −0.166, SE = 0.081; *p* = 0.041) were negatively associated. Also, high frequencies of weekly breakfast skipping (Beta = 0.476, SE = 0.129; *p* \< 0.001) and eating out (Beta = 0.109, SE = 0.036; *p* = 0.003) were positively associated with the Western-based pattern, whereas age (Beta = −0.136, SE = 0.033; *p* \< 0.001) and household income (Beta = −0.078, SE = 0.027; *p* = 0.005) were negatively associated. Moreover, a higher frequency of daily snacking (*p* = 0.013) was positively associated with the local-based food pattern. As for the Chinese adolescents, age (Beta = 0.165, SE = 0.029; *p* \< 0.001), PA level (Beta = 0.10, SE = 0.024; *p* \< 0.001) and maternal education level (Beta = 0.242, SE = 0.114; *p* = 0.035) showed positive associations with the healthy-based pattern, whereas high frequencies of eating out (Beta = −0.086, SE = 0.026; *p* = 0.001) and fast food intake (Beta = −0.223, SE = 0.068; *p* = 0.001) were negatively associated. Also, higher weekly frequencies of eating out (Beta = 0.072, SE = 0.026; *p* = 0.007), fast food intake (Beta = 0.156, SE = 0.068; *p* = 0.023), soft drink consumption (Beta = 0.080, SE = 0.035; *p* = 0.023), and daily snacking practice (Beta = 0.157, SE = 0.055; *p* = 0.004) were positively associated with the Western-based food pattern, whereas age (Beta = −0.084, SE = 0.029; *p* = 0.004) was negatively associated. • Meal Skipping Behaviours One study investigated the unadjusted association between meal skipping behaviour and dietary correlates \[[@B31-ijerph-16-00603]\] and found that ethnicity and eating companion were associated with meal skipping behaviours while living arrangement was not significantly associated. Those who usually skipped meals were commonly Malays; while more than half of the Chinese and Indian participants never skipped any meals, more than half of the Malays skipped at least one meal \[[@B31-ijerph-16-00603]\]. The results also indicated that those who had a meal companion (either ate with family or peers) had a lower probability of practising meal skipping behaviours; the majority of adolescents who had a meal companion never skipped any meals while one-fourth of those who ate alone skipped all three main meals in a day \[[@B31-ijerph-16-00603]\]. • Meal and Snacking Frequency Two studies examined the effect of gender on the meal and snacking frequency and both found that gender was a significant correlate in unadjusted associations \[[@B35-ijerph-16-00603],[@B37-ijerph-16-00603]\]. In one of the studies, adolescent girls had significantly higher snacking frequency compared to boys \[[@B37-ijerph-16-00603]\]. The other study reported that the frequency of breakfast intake was significantly higher among male adolescents compared to female \[[@B35-ijerph-16-00603]\]. 3.2. Physical Activity Determinants of Physical Activity Behaviours {#sec3dot2-ijerph-16-00603} ------------------------------------------------------------------- [Table 3](#ijerph-16-00603-t003){ref-type="table"} summarizes the associations between the potential determinants of PA and PA behaviours. The factors related to PA determinants were investigated in 12 studies. The PA determinants were grouped into four categories: demographics, physical-environmental, social-environmental and behavioural. The PA score and PA level were the most common outcomes related to PA behaviour. Gender and ethnicity were the most commonly studied correlates. ### 3.2.1. Demographic Determinants of Physical Activity {#sec3dot2dot1-ijerph-16-00603} In the reviewed studies, eight potential demographic determinants of PA were considered: age, gender, ethnicity, maternal employment, paternal and maternal education, household income, household size and parents' marital status. • Age Two studies investigated the association between age and PA \[[@B40-ijerph-16-00603],[@B44-ijerph-16-00603]\]. One of the studies showed that there was no significant association \[[@B44-ijerph-16-00603]\]. In the other study, after adjusting for confounders (i.e., age, gender, breakfast intake, BMI and school session), the results indicated that each additional year of age raised the odds of being physically inactive (OR = 1.2 (95% CI: 1.16--1.23); *p* \< 0.001) and thus younger adolescents were more physically active \[[@B40-ijerph-16-00603]\]. • Gender Ten studies \[[@B27-ijerph-16-00603],[@B37-ijerph-16-00603],[@B38-ijerph-16-00603],[@B40-ijerph-16-00603],[@B41-ijerph-16-00603],[@B42-ijerph-16-00603],[@B43-ijerph-16-00603],[@B44-ijerph-16-00603],[@B46-ijerph-16-00603],[@B47-ijerph-16-00603]\] focused on the association between gender and PA and level of PA. Significant gender differences in PA behaviour were reported in nine studies. Six studies found a consistent positive association between gender and PA and reported that boys were significantly more physically active than girls \[[@B37-ijerph-16-00603],[@B40-ijerph-16-00603],[@B41-ijerph-16-00603],[@B42-ijerph-16-00603],[@B44-ijerph-16-00603],[@B47-ijerph-16-00603]\]. However, among these, only two studies made adjustments for confounders \[[@B40-ijerph-16-00603],[@B44-ijerph-16-00603]\]. In one study, there was no significant association between gender and PA duration while girls reported spending longer each day on PA before and during school \[[@B46-ijerph-16-00603]\]. In contrast, two studies reported that adolescent boys had significantly higher daily PA levels and moderate to vigorous physical activity levels compared to girls \[[@B38-ijerph-16-00603],[@B43-ijerph-16-00603]\]. In the cohort study, PA among girls residing in rural areas dropped significantly in a cohort study from baseline in 2012 to the first follow-up in 2014 \[[@B27-ijerph-16-00603]\]. • Ethnicity Five studies \[[@B34-ijerph-16-00603],[@B41-ijerph-16-00603],[@B42-ijerph-16-00603],[@B44-ijerph-16-00603],[@B47-ijerph-16-00603]\] examined the association between ethnicity and PA. The significant associations between ethnicity and PA behaviour were consistent in four studies \[[@B34-ijerph-16-00603],[@B41-ijerph-16-00603],[@B44-ijerph-16-00603],[@B47-ijerph-16-00603]\], but the associations were adjusted for covariates in only one of these studies \[[@B44-ijerph-16-00603]\]. In one of the studies, race was one of the factors that differed significantly between the active and inactive group (Malays vs others), where Malays were more active than other ethnicities \[[@B41-ijerph-16-00603]\]. In another study, the PA scores of the Indian adolescents were slightly lower followed by the Malays. In contrast, it has been reported that Chinese adolescents were the least active \[[@B47-ijerph-16-00603]\]. In addition, it was reported in another study that, compared to Malays, Chinese spend less time in PA \[[@B44-ijerph-16-00603]\]. • Maternal Employment One study reported a significant association between PA behaviour and maternal employment \[[@B41-ijerph-16-00603]\] Based on the result of a logistic regression analysis, one study reported that adolescents with an unemployed mother were more likely to be physically inactive compared to those with an employed mother (OR = 2.167 (95% CI: 1.263--3.717); *p* = 0.005) \[[@B41-ijerph-16-00603]\]. • Paternal and Maternal Education Two studies focused on the associations between PA and paternal and maternal education \[[@B42-ijerph-16-00603],[@B44-ijerph-16-00603]\]. In one study, there was a positive but weak association between paternal education and PA score \[[@B42-ijerph-16-00603]\]. The other two studies did not find any association with paternal education. Only one study out of three found that adolescents who had a primary- or secondary-educated mother were less likely to participate in PA compared to adolescents who had a tertiary-educated mother \[[@B44-ijerph-16-00603]\]. • Household Income One study examined the association between PA and household income, and did not find any association \[[@B42-ijerph-16-00603]\]. • Household Size Two studies focused on household size and PA \[[@B42-ijerph-16-00603],[@B44-ijerph-16-00603]\]. One study did not show any significant associations \[[@B42-ijerph-16-00603]\]. The results of the other study showed that household size could increase the likelihood of being physically active. Specifically, at the conditional level, an additional household member increased the time spent by the adolescent on PA by 0.083 days. At the unconditional level, an increase in household size raised the time spent on PA by 0.062 days. The results also showed that there was an approximately 0.9% increase (based on the estimated coefficients) in the probability of participating in PA when the household size increased by one member \[[@B44-ijerph-16-00603]\]. • Parents' Marital Status One study looked at the association between parents' marital status and PA, and did not find any association \[[@B44-ijerph-16-00603]\]. ### 3.2.2. Physical-Environmental Determinants of Physical Activity {#sec3dot2dot2-ijerph-16-00603} The physical-environmental determinants of PA that were considered by the studies included in this systematic review were school session and place of residence, as well as a number of environmental barriers to the use of PA facilities, namely, hot weather, availability of equipment, distance between facility and home, usage level of facilities and facility support. • School Session One study, after controlling for other factors, indicated that adolescents attending afternoon school sessions had higher odds of being physically inactive than those attending morning sessions (AOR = 1.3; 95% CI: 1.13--1.44) \[[@B40-ijerph-16-00603]\]. • Place of Residence Two studies related to one cohort study examined the association between place of residence and physical activity without adjustment for covariates \[[@B27-ijerph-16-00603],[@B47-ijerph-16-00603]\]. One did not find any significant association at the cross-sectional level \[[@B47-ijerph-16-00603]\] but observed a downward trend in the PA level among all adolescents with a significant reduction among all rural students from 2012 (baseline) to 2014 (first follow-up) \[[@B27-ijerph-16-00603]\]. • Environmental Barriers to Use of PA Facilities Three studies examined the association between environmental barriers and the use of PA facilities without any adjustment for covariates \[[@B41-ijerph-16-00603],[@B45-ijerph-16-00603],[@B46-ijerph-16-00603]\]. One study did not show any associations \[[@B46-ijerph-16-00603]\]; however, one study showed a significant association with environmental barriers \[[@B41-ijerph-16-00603]\] and one identified significant associations with the environmental characteristics of the facilities themselves \[[@B45-ijerph-16-00603]\]. In one study, it was reported that there were significant differences between the active and inactive group with regards to the effect of personal environmental factors such as hot weather, availability of equipment and distance between facility and home, which were categorized as barriers to PA \[[@B41-ijerph-16-00603]\]. In the other study, it was found that facility support and usage level of facilities had a significant but poor relationship with adolescents' involvement in physical activities \[[@B45-ijerph-16-00603]\]. ### 3.2.3. Social-Environmental Determinants of Physical Activity {#sec3dot2dot3-ijerph-16-00603} Three studies examined the social-environmental determinants of PA, namely, family without exercise, family and peer influence and physical education \[[@B41-ijerph-16-00603],[@B42-ijerph-16-00603],[@B44-ijerph-16-00603]\]. • Family Without Exercise One study indicated that PA was significantly lower among adolescents who lived in a family without exercise \[[@B41-ijerph-16-00603]\]. • Family and Peer Influence One study showed that there was a positive and significant relationship between peer influence and PA score without adjustment for covariates \[[@B42-ijerph-16-00603]\]. It also showed that family influence had a positive but weak impact on PA score. • Physical Education One study reported that after adjusting for covariates (age, gender, and ethnicity) an additional day spent in attending a physical education class increased the time spent on PA by 0.200 (unconditional) and 0.151 (conditional) days. The probability of participating in PA also increased by 2.2% when the time spent attending physical education class increased by one day \[[@B44-ijerph-16-00603]\]. ### 3.2.4. Behavioural Determinants of Physical Activity {#sec3dot2dot4-ijerph-16-00603} Two studies evaluated the behavioural determinants of PA \[[@B40-ijerph-16-00603],[@B41-ijerph-16-00603]\]: one considered breakfast intake, while the other considered the effects of six personal factors: preferring to watch TV, being embarrassed, being lazy, finding exercise too troublesome, only exercising when having ample time, and seeing stretching as important before exercise. • Breakfast Intake The study on breakfast intake and PA, which made adjustments for covariates, found that the odds of adolescents who did not consume breakfast being inactive increased 1.9 times, which was considered significant (AOR = 1.9; 95% CI: 1.74--2.13) and among those who had irregular breakfast intake the odds increased by 1.4 times (AOR = 1.9; 95% CI: 1.33--1.55) compared to those who consumed breakfast daily \[[@B40-ijerph-16-00603]\]. • Personal Barriers The study that examined several personal barriers to PA did not adjust for covariates. The results of the study showed that there were significant differences between the active and inactive group with regards to the following personal factors: stretching is important before exercise, exercise when having ample time, prefer to watch TV, embarrassed, being lazy and too troublesome \[[@B41-ijerph-16-00603]\]. In addition, the time constraint was associated with inactivity, where those who reported that time constraints prevented them from doing PA were 2.5 times more at risk of becoming inactive. Those who stated that they preferred to do exercise when the time was available were 2.5 times more likely to be inactive and those who stated that stretching was necessary before exercise were 3.7 times more likely to be inactive \[[@B41-ijerph-16-00603]\]. 4. Discussion {#sec4-ijerph-16-00603} ============= This systematic review identified 18 studies on the potential factors influencing the dietary and PA behaviours of Malaysian adolescents. This review clearly demonstrates that for many variables, the evidence is insufficient and this is mainly due to the limited number of studies. Also, the associations that were found were often small or inconsistent, with few studies controlling for confounding factors. Many of the factors identified by this systematic review have been reported in other systematic reviews \[[@B2-ijerph-16-00603],[@B50-ijerph-16-00603],[@B51-ijerph-16-00603]\]. This review found that most of the diet-related studies focused on dietary patterns and nutrient analysis, while only a few of them reported results concerning specific food groups. The most common PA-related outcomes identified by this review were PA score and PA intensity. It was not possible to establish strong relationships between some of the significant associated diet/PA correlates and diet/PA behaviours because these were evaluated in only one study. The finding of this systematic review with respect to gender differences in dietary intake suggest that male adolescents consumed more energy \[[@B37-ijerph-16-00603],[@B38-ijerph-16-00603],[@B48-ijerph-16-00603]\] and macronutrients \[[@B48-ijerph-16-00603]\] compared to females while the intakes of carbohydrate and protein were higher among rural adolescents than their urban counterparts. In addition, male adolescents had a lower diet quality \[[@B35-ijerph-16-00603]\], snacking frequency \[[@B35-ijerph-16-00603]\] and higher meal frequency than females \[[@B37-ijerph-16-00603]\]. This finding is in agreement with a previous systematic review, which reported that most girls had a desire to be thin as they perceived that a thin figure is the ideal female body image. It also found that girls had higher nutrition knowledge scores than boys, particularly because girls were more likely to read nutritional food labels \[[@B52-ijerph-16-00603]\]. The previous systematic review also revealed that many other factors such as PA level, socioeconomic status (SES), diet, individual and social factors contribute to either a higher or a lower energy intake among adolescents \[[@B52-ijerph-16-00603]\]. In regards to such factors, this review only found weak evidence for the relation between ethnicity and eating and meal skipping behaviours. Specifically, the studies included in this review showed that the adolescents who usually skipped meals were Malay or those who usually ate alone \[[@B31-ijerph-16-00603]\]. A previous systematic review indicated that there was a significant positive association between peer influence and eating habits, meaning that the higher the peer pressure, the unhealthier the students' dietary intake \[[@B52-ijerph-16-00603]\]. The previous review also suggested that the presence of peers and friends increases the energy intake of children and adolescents \[[@B52-ijerph-16-00603]\]. In this systematic review, only weak evidence was found for the effect of ethnic differences on dietary patterns, where Malays had higher SSB consumption \[[@B36-ijerph-16-00603]\] and lower diet quality than Indian and Chinese adolescents \[[@B35-ijerph-16-00603]\]. In addition, one of the reviewed studies showed that Malay adolescents were more likely to have Western-based and local-based dietary patterns, whereas Chinese adolescents were more likely to follow a healthy-based food pattern \[[@B34-ijerph-16-00603]\]. An earlier systematic review indicated that the dietary habits of ethnic populations can be affected by many factors such as the availability of food, income level, food beliefs, religion, cultural patterns and customs \[[@B53-ijerph-16-00603]\]. Thus, the ethnic differences in eating patterns may possibly be a reflection of socio-cultural differences related to food preferences \[[@B54-ijerph-16-00603]\]. This systematic review also found that age (being older) and higher maternal educational status were associated with a healthy-based food pattern among Malaysian adolescents \[[@B34-ijerph-16-00603]\]. On the other hand, from the reviewed studies it was found that Malay adolescents in lower-income households had a higher intake of Western-based food \[[@B34-ijerph-16-00603]\]. Thus this review presents consistent evidence that SES factors (e.g., parental education level, household income) were associated with dietary behaviours, which is in line with the findings of other systematic reviews \[[@B51-ijerph-16-00603],[@B55-ijerph-16-00603]\]. However, a previous systematic review determined that interpersonal factors also had a major role to play in dietary behaviours, whereas school, neighbourhood or societal factors were not consistently associated with dietary behaviours \[[@B56-ijerph-16-00603]\]. In addition, it was reported in another systematic review that adolescents with low SES may have poorer dietary patterns compared to high-SES adolescents because low-SES families may not be able to buy nutritious foods \[[@B57-ijerph-16-00603]\]. Moreover, the review argued that it is possible that low-SES families have less information of the nutritional content and the daily recommendations of food groups \[[@B58-ijerph-16-00603]\]. Furthermore, low-SES families may have more fast food consumption and have a lower intake of healthy options \[[@B59-ijerph-16-00603]\]. A systematic review of studies related to high, medium and low development countries also suggested that, as countries develop economically, there is a higher adherence to an unhealthy diet in adolescents of parents with lower education level \[[@B60-ijerph-16-00603]\]. The results of this systematic review suggest that the significant correlates of PA for adolescents, with limited evidence, are gender (male), ethnicity (Malay), paternal and maternal education (higher), household size (bigger), family and peer influence and physical education. The reviewed studies showed that higher PA intensity was related to gender (male). In addition, physical inactivity was found to be significantly associated with age (older), breakfast (no intake), school session (afternoon), personal barriers (e.g., family without exercise, unemployed mother), environmental barriers (e.g., hot weather, unavailability of equipment and distance between facility and home) and place of residence (rural). One consistent finding of this systematic review, which is similar to that reported in other systematic reviews, was that boys are more active than girls and that PA decreases with increasing age. On the other hand, the results reported in this review regarding the influence of ethnicity show that it correlates with adolescents' PA \[[@B61-ijerph-16-00603]\], whereas in the other systematic review the evidence was inconclusive \[[@B62-ijerph-16-00603]\]. In this review, the evidence related to the effect of age on PA was weak; however, there was a relationship. The above findings of this review are consistent with those presented in a previous review, which identified that cluster patterns were different among adolescents according to age and gender, where a higher proportion of girls and older adolescents were in the clusters defined by low levels of PA \[[@B2-ijerph-16-00603]\]. A reduction in PA related to age during adolescence has also been reported in a recent systematic review of European children and adolescents \[[@B63-ijerph-16-00603]\]. This decrease in PA may be due to the fact that older adolescents have to focus more on academic activities and achievements. In the Malaysian context, adolescents (13--18 years) in middle/high school are under pressure to pass high school/college entrance examinations. In addition, it has been indicated in other studies that female adolescents were more concerned about others seeing their bodies while performing PA. They were also less interested in vigorous PA, which was more favoured by males \[[@B64-ijerph-16-00603]\]. In addition, girls were more responsible for doing housework, thereby limiting the time they had for PA \[[@B65-ijerph-16-00603]\]. Furthermore, this review found weak evidence that school session (afternoon session) and breakfast intake (no breakfast and irregular breakfast) were significant factors related to physical inactivity among Malaysian adolescents \[[@B40-ijerph-16-00603]\]. It was explained in one of the reviewed studies that skipping breakfast can lead to mental and physical fatigue as well as energy and nutrient deficiencies that affect the ability to perform physical activities \[[@B66-ijerph-16-00603]\]. Finally, it was indicated in a previous systematic review that PA in adolescents was unrelated to the copying of friends and siblings, support of parents for PA, and family size and that instead it was more consistently related to school and neighbourhood characteristics than to interpersonal and societal environments \[[@B56-ijerph-16-00603]\]. However, in another systematic review, it was indicated that most studies on the environmental determinants of PA rely on self-reports of environmental factors and thus represent the perceived, rather than the actual features of the physical environment \[[@B51-ijerph-16-00603]\]. 4.1. Limitations {#sec4dot1-ijerph-16-00603} ---------------- However, some issues and limitations should be considered when interpreting the findings. First, all of the included studies except one were cross-sectional and cannot be considered good-quality, thus the conclusions about the direction and possible associations may not be precise. Second, the majority of studies relied on self-reported data, and it was difficult to judge the validity and reliability of the PA and diet measurement tools that used to collect these data because this information was not reported in several of the studies and some used invalid or poorly validated tools. Also, the variation in the age of the adolescents involved in the reviewed studies may also have interfered with the appropriateness of the methods used for measuring exposures and outcomes and may also have affected the magnitude of the errors in the information and measurements. In addition, the literature reviewed presented very heterogeneous and inconsistent results with regard to the correlates of diet and of PA in adolescents. Furthermore, this inconsistency remained even when the results of previous reviews were compared with those of this review. Possible reasons for this are the lack of consistency between studies in regards to the methods used for measuring dietary intake and PA and the presence of methodological errors as well as the use of inaccurate and imprecise techniques. For instance, in the reviewed studies only questionnaires were used to assess PA; no accelerometers or other related electronic devices were used. Furthermore, it is notable that one previous systematic review concluded that no questionnaires with acceptable reliability and validity were available for the assessment of PA among adolescents \[[@B67-ijerph-16-00603]\]. Also, in this systematic review, in order to compile and analyse the results of the included studies, conceptually similar determinants were placed into the same category, even though some of the potential determinants in the same category were often different or measured in different ways. Insufficient control for confounders also appeared to be a problem in the majority of studies. For instance, in some studies, the analyses were not adjusted for relevant socio-demographic confounders such as gender, age and/or SES. Additionally, many studies used samples that were non-representative or only representative of a limited geographical area. The majority of studies focused on specific states such as Kuala Lumpur and Selangor and Pahang, whereas only a few of the included studies were conducted in the northeast regions of Malaysia. This indicates that the development of scientific research on the diet/PA behaviours of Malaysian adolescents has yet to cover all regions of the country. Lastly, the reviewed studies were heterogeneous in terms of their measurement, samples and analyses, and therefore it was not probable to evaluate the overall strength of the associations. Thus, an improvement in the quality of further studies could lead to more consistent studies and greater sureness in the identified correlates and determinants of diet and PA. 4.2. Recommendations for Further Research {#sec4dot2-ijerph-16-00603} ----------------------------------------- Regardless of the heterogeneity in the measures, specific behaviours and study areas, this review was able to highlight the factors that were or were not related to PA and dietary behaviours in Malaysian adolescents. It is therefore hoped that the results of this review will help in identifying potentially effective mediators that can be used in interventions to promote a healthy lifestyle among this population group. The results suggest that interventions could incorporate structured physical activities before and after school or during breaks; improved accessibility of PA facilities nearby the school environment; increased physical education and PA sessions; improved availability of healthy food options especially in school canteens and government-subsidized healthy foods in schools. Currently, the Malaysian government only subsidizes foods in primary schools. However, extending this subsidy to high schools may prepare more opportunities for low-SES adolescents to eat healthy foods. Despite there are some studies that is available and mostly well conducted from the higher income countries, some of the policy that was developed for healthy lifestyle may not be suitable to be adapted in Malaysian local context. It has been stated in a systematic review the main structural factors that affect adolescent health include economic and political systems, the education system, wealth and its distribution within a country, poverty, migration and cultural factors such as gender and ethnic equality, as well as factors such as climate change and war or conflict \[[@B68-ijerph-16-00603]\]. National wealth and income dissemination affect adolescent-health outcomes across countries. There is considerable evidence that income inequality within countries has the impact on many aspects of adolescent health, principally among middle-income and high-income countries \[[@B24-ijerph-16-00603]\]. Another systematic review also revealed that in low-income countries or in countries with low human development index (HDI), the relation between SES and obesity seems to be positive for both men and women: the wealthier and/or those with higher educational level tend to be more likely to be obese. However, in middle-income countries or in countries with medium HDI, the relation becomes mostly mixed for men and negative for women \[[@B69-ijerph-16-00603]\]. School-based interventions have the potential to improve adolescents' dietary and PA behaviours because adolescent students spend on average at least six hours in school per day during term-time. Potentially the healthy lifestyle subjects should be incorporated in a more interactive curriculum/ manner. The availability of well maintain playing area/ park in addition to the availability of healthy food is crucial to ensure PA program can be conducted. It is also recommended that gender-specific strategies are identified for use in further interventions to improve the eating patterns and PA among Malaysian adolescents. Lifestyle and health-related interventions that focus on PA and healthy eating practices require attention from all the stakeholders. High-quality longitudinal studies using electronic devices to assess daily PA is also needed to improve understanding of the dietary and PA behaviours of Malaysian adolescents. 5. Conclusions {#sec5-ijerph-16-00603} ============== This systematic review is the first to summarize the determinants associated with diet and PA behaviours among adolescents in Malaysia. However, the significance of these associations was often small or inconsistent. This review highlights there is a lack of longitudinal observational research to support the causal role of specific factors in improving diet and PA behaviour in Malaysian adolescents. However, this review summarizes the best available evidence for local policymakers and public health practitioners so that they can consider incorporating these findings into the development of intervention protocols for improving diet, PA and health in Malaysian adolescents. The authors would like to thank Laura Johnson (L.J.) and Zoi Toumpakari (Z.T.) (Centre for Exercise, Nutrition and Health Sciences, School for Policy Studies, University of Bristol, United Kingdom) for their valuable cooperation and support during manuscript preparation. The following is available online at <http://www.mdpi.com/1660-4601/16/4/603/s1>, Text S1: Search strategies; Text S2: Newcastle-Ottawa Scale for cross-sectional and cohort studies; Table S1: PRISMA checklist; Tables S2 and S3: Quality assessment of the studies. ###### Click here for additional data file. S.M. and H.A.M. conceived and designed the protocol for this review. S.M., T.T.S., M.Y.J., M.D., M.N.A.M. and H.A.M. wrote the paper. All authors critically reviewed drafts of manuscript and approved the final version. This review was undertaken as part of the MyHeART BEaT (Malaysian Health and Adolescents Longitudinal Research Team) (Behavioural Epidemiology and Trial) project (IF017-2017) in Malaysia funded by the Academy of Science Malaysia (Newton Ungku Omar Fund) and the UK Medical Research Council between the University of Malaya and University of Bristol. The other authors declare no conflict of interest. {#ijerph-16-00603-f001} ijerph-16-00603-t001_Table 1 ###### Characteristics of the included studies. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Author, Year \[Ref\] Setting/Urbanity Sample Age (y) Mean ± SD Ethnicity Maternal Education Income (RM) Diet Measure Diet Outcome PA Measure PA Outcome Covariates ------------------------------------------------------- --------------------------------------------- ------------- ------------------- ------------------------------- --------------------------------------------------- --------------------------------- -------------------- ------------------------------------------------------------ ------------ ------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------- Chin & Mohd Nasir 2009 \[[@B31-ijerph-16-00603]\] Kuantan in Pahang/NR 407 (♀) 15.2 ± 1.9 Malay, Chinese, Indian secondary: 57.0% Mean ± SD\ OQ (EBQ) Meal skipping behaviours NA NA NR RM 3266 ± 2566 Abdullah et al. 2016 \[[@B34-ijerph-16-00603]\] Kelantan/NR 454 (♂ ♀) 15.3 ± 1.9 Malay, Chinese Malay; secondary: 67.8% Chinese; secondary: 72.5% Malay: 70% Chinese: 68%\ FFQ Healthy, Western & Local\ PAQ-C PA Age, gender, ethnicity, SES, breakfast skipping, snacking, eating out, fast food intake, soft drink intake, dietary supplement, PA levels, screen viewing low income (\<RM 2300) dietary pattern score Rezali et al. 2015 \[[@B35-ijerph-16-00603]\] Kuala Lumpur/Urban 373 (♂ ♀) 14.3 ± 1.2 Malay, Chinese, Indian, other NR NR 2 × 24R & OQ (EBQ) Diet quality, food groups, meal frequency NA NA NR Loh et al. 2017 \[[@B36-ijerph-16-00603]\] Kuala Lumpur/Urban 873 (♂ ♀) 13 \* Malay, Chinese, Indian, other Secondary: 61.4% NR OQ (CNQ) Sugar sweetened beverages intake NA NA NR Nurul-Fadhilah et al. 2013 \[[@B37-ijerph-16-00603]\] Kota Bharu in Kelantan/NR 236 (♂ ♀) 15.3 ± 1.9 Malay NR (Mean ± SD)\ FFQ Energy intake, frequency of eating out, snacking frequency PAQ-C PA NR RM 2191 ± 2553 Teo et al. 2014 \[[@B38-ijerph-16-00603]\] Kota Bharu in Kelantan/NR 454 (♂ ♀) 15.3 ± 1.9 Malay, Chinese NR NR FFQ Energy intake PAC-C PA & MVPA NR Boon et al. 2012 \[[@B39-ijerph-16-00603]\] Kuala Lumpur/Urban 156 (♂ ♀) 14.1 ± 0.8 Malay, Chinese, Indian NR Moderate (RM 2000--5999): 57.1% 1 × 24R Energy & macronutrients intake NA NA NR Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] \* Kuala Lumpur, Selangor, Perak/Urban & rural 794 (♂ ♀) 12.86 ± 0.3 Malay, Chinese, Indian, other Secondary: 66% Low SES: 49% 7DH Energy & macronutrients intake NA NA NR Cynthia et al. 2013 \[[@B49-ijerph-16-00603]\] Puchong in Selangor/Urban 408 (♂ ♀) 13.74 ± 0.56 Malay, Chinese, other Upper secondary: 38.9% 41.9% \< RM 3999 2 × 24R Energy & macronutrients intake NA NA Gender, ethnicity, BMI Baharudin et al. 2014 \[[@B40-ijerph-16-00603]\] National/NR 40011 (♂ ♀) 13.48 ± 2.24 NR NR NR NA NA PAQ-C Physical inactivity Age, gender, breakfast intake, BMI, School session Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] Petaling in Selangor/Urban 519 (♂ ♀) 14 and 16 Malay, Chinese, Indian NR Father---41.6%\ NA NA IPAQ Physical inactivity, PA NR Mother---80.9%\ Low income (\<RM 1500) Dan et al. 2011 \[[@B42-ijerph-16-00603]\] Kuantan in Pahang/NR 400 (♂ ♀) 13.23 ± 0.31 Malay, Chinese, Indian Total years\ \>RM 3000: 38.5% NA NA PAQ-C PA NR (Mean ± SD): 12.29 ± 3.39 Farah Wahida et al. 2011 \[[@B43-ijerph-16-00603]\] Kuantan in Pahang/NR 360 (♂ ♀) 13.2 ± 0.3 Malay, Chinese, Indian, other Secondary: 50.6% NR NA NA PAQ-C PA level, MVPA NR Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] NR/NR 2991 (♂ ♀) 15.88 ± 0.71 Malay, Chinese, Indian, other Secondary: 64.39% NR NA NA OQ PA Age, gender, ethnicity Abd-Latif et al. 2012 \[[@B45-ijerph-16-00603]\] Seremban, Muar, Kota, Star, Kuantan/NR 913 (♂ ♀) 13--17 NR NR Medium SES: 44% NA NA OQ PA involvement NR Cheah et al. 2012 \[[@B46-ijerph-16-00603]\] Kuching in Sarawak/NR 316 (♂ ♀) 14--16 Malay, Chinse Secondary: 63.9% Mean ± SD\ NA NA OQ PA NR RM 3652.9 ± 3740.6 Su et al. 2014 \[[@B47-ijerph-16-00603]\] \* Kuala Lumpur, Selangor, Perak/Urban & rural 1327 (♂ ♀) 12.9 ± 0.3 Malay, Chinese, Indian, other NR NR NA NA PAQ-C PA NR Abdul Majid et al. 2016 \[[@B27-ijerph-16-00603]\] \* Kuala Lumpur, Selangor, Perak/Urban & rural 820 (♂ ♀) 15 Malay, Chinese, Indian, other NR NR NA NA PAQ-C PA NR -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Note: ♂, Male; ♀, Female; SES, Socio-Economic Status; NR, Not Reported; NA, Not Available; RM, Malaysian Ringgit (currency); MVPA, Moderate to Vigorous Physical Activity; BMI, Body Mass Index; FFQ, Food Frequency Questionnaire; PA, Physical Activity; X, Number; X × 24R, 24-h Recall completed over X days; XDH, X days Diet History; OQ, Other Questionnaire; EBQ, Eating Behaviours Questionnaire; CNQ, Child Nutrition Questionnaire; PAQ-C, Physical Activity Questionnaire for Children; \*, MyHeARTs study. ijerph-16-00603-t002_Table 2 ###### Summary of associations between the determinants of diet and dietary behaviours. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Author, Year \[Ref\] Outcome Correlate Association *p*-Value ------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------- -------------- Energy & Nutrients Teo et al. 2014 \[[@B38-ijerph-16-00603]\] Energy intake (kcal/day) Gender Male vs. Female (Median (95%, CI) Malay ♂ 2408 (2255--2437) vs. ♀ 2178 (2058--2246) *p* \< 0.01 Chinese ♂ 1860 (1792--1970) vs. ♀ 1649 (1642--1828) *p* \< 0.05 Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 1774.0 (1730.8--1817.3) vs. ♀ 1595.2 (1567.4--1623.1) Nurul-Fadhilah et al. 2013 \[[@B37-ijerph-16-00603]\] Male vs. Female (Mean ± SD)\ *p* \< 0.01 ♂ 2346 ± 468 vs. ♀ 2152 ± 547 Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Place of residence Urban vs. Rural (Mean (95% CI)\ *p* \< 0.001 1612.3 (1581.6--1643.1) vs. 1706.1 (1668.7--1743.4) Boon et al. 2012 \[[@B39-ijerph-16-00603]\] Meal patterns (Mean ± SD)\ *p* \< 0.05 3M + 3S: 1952 ± 411 vs. 3M + 2S: 1883 ± 456 vs. 3M + 1S: 1687 ± 426 vs. 3M: 1405 ± 426 vs. ≤2M + 2,3S: 1414 ± 335 vs. ≤2M + 0,1S: 1340 ± 252 Snacking patterns (Mean ± SD)\ *p* \< 0.05 3M + 3S: 793 ± 246 vs. 3M + 2S: 514 ± 207 vs. 3M + 1S: 259 ± 176 vs. 3M vs. ≤2M + 2: 259 ± 176 vs. 3S: 0 vs. ≤2M + 0: 459 ± 209 vs. 1S: 247 ± 244 Cynthia et al. 2013 \[[@B49-ijerph-16-00603]\] Eating out 0--2 times vs. 3--6 times vs. ≥7 times (Mean ± SE)\ *p* = NS 1984 ± 65 vs.1915 ± 97 vs. 2077 ± 100 Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Carbohydrate intake (g/day) Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 245.2 (238.6--251.8) vs. ♀ 220.0 (215.7--224.2) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* \< 0.001 221.6 (216.8--226.3) vs. 236.4 (230.8--242.0) Cynthia et al. 2013 \[[@B49-ijerph-16-00603]\] Eating out 0--2 times vs. 3--6 times vs. ≥7 times (Mean ± SE)\ *p* = NS 131.10 ± 1.69 vs. 132.02 ± 2.03 vs. 126.9 ± 2.58 Boon et al. 2012 \[[@B39-ijerph-16-00603]\] Meal pattern (Mean ± SD)\ *p* \< 0.05 3M + 3S: 258.9 ± 49.5 vs. 3M + 2S: 253.7 ± 72.6 vs. 3M + 1S: 225.2 ± 64.1 vs. 3M: 187.0 ± 40.0 vs. ≤2M + 2,3S: 200.2 ± 50.8 vs. ≤2M + 0,1S: 168.1 ± 43.7 Snacking practices (Mean ± SD)\ *p* \< 0.05 3M + 3S: 111.5 ± 34.4 vs. 3M + 2S: 74.0 ± 29.8 vs. 3M + 1S: 37.0 ± 24.8 vs. 3M: 0 vs. ≤2M + 2,3S: 71.8 ± 28.5 vs. ≤2M + 0,1S: 38.4 ± 37.9 Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Protein intake (g/day) Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 65.7 (63.6--67.7) vs. ♀ 58.7 (57.4--59.7) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = 0.001 59.3 (57.9--60.6) vs. 63.1 (61.1--64.8) Cynthia et al. 2013 \[[@B49-ijerph-16-00603]\] Protein intake (g/day) Eating out 0--2 times vs. 3--6 times vs. ≥7 times (Mean ± SE)\ *p* = NS 42.37 ± 0.74 vs. 40.08 ± 0.89 vs. 43.73 ± 1.13 Boon et al. 2012 \[[@B39-ijerph-16-00603]\] Meal pattern (Mean ± SD)\ *p* \< 0.05 3M + 3S: 79.7 ± 21.4 vs. 3M + 2S: 73.7 ± 17.9 vs. 3M + 1S: 72.1 ± 26.2 vs. 3M: 58.7 ± 15.8 vs. ≤2M + 2,3S: 49.7 ± 12.4 vs. ≤2M + 0,1S: 54.5 ± 14.4 Snacking practices (Mean ±SD)\ *p* \< 0.05 3M + 3S: 25.4 ± 10.3 vs. 3M + 2S: 16.4 ± 8.7 vs. 3M + 1S: 8.1 ± 9.60 vs. 3M: 0 vs. ≤2M + 2,3S: 11.3 ± 8.1; ≤2M + 0,1S: 8.1 ± 11.5 Cynthia et al. 2013 \[[@B49-ijerph-16-00603]\] Fat (g/day) Eating out 0--2 times vs. 3--6 times vs. ≥7 times (Mean ± SE)\ *p* = 0.043 34.05 ± 0.64 vs. 34.54 ± 0.77 vs. 34.86 ± 0.98 Boon et al. 2012 \[[@B39-ijerph-16-00603]\] Meal pattern (Mean ± SD)\ *p* \< 0.05 3M + 3S: 66.7 ± 21.4 vs. 3M + 2S: 64.3 ± 21.2 vs. 3M + 1S: 55.3 ± 18.2 vs. 3M: 47.1 ± 17.0 vs. ≤2M + 2,3S: 46.4 ± 18.0 vs. ≤2M + 0,1S: 50.0 ± 15.9 Snacking practices (Mean ± SD)\ *p* \< 0.05 3M + 3S = 27.7 ± 14.1 vs. 3M + 2S = 17.2 ± 10.3\ 3M + 1S = 8.8 ± 7.6 vs. 3M = 0 vs. ≤2M + 2 vs. 3S = 14.4 ± 10.6 vs. ≤2M + 0,1S = 6.9 ± 8.2 Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 59.7 (57.6--61.7) vs. ♀ 53.2 (52.0--54.3) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 54.6 (53.2--56.0) vs. 56.4 (54.8--57.9) Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Cholesterol (mg/d) Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 248.8 (236.5--261.0) vs. ♀ 209.1 (201.6--216.7) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* \< 0.001 202.6 (194.2--211.1) vs. 244.1 (234.1--254.0) Mono-unsaturated fatty acid (g/d) Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 9.0 (8.5--9.5) vs. ♀ 7.9 (7.6--8.1) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 8.3 (7.9--8.6) vs. 8.3 (7.9--8.6) Poly-unsaturated fatty acid (g/d) Gender Male vs. Female (Mean (95% CI)\ *p* = 0.005 ♂ 6.4 (6.1--6.7) vs. ♀ 5.8 (5.6--6.0) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 5.9 (5.7--6.2) vs. 6.2 (5.9--6.4) Saturated fatty acid (g/d) Gender Male vs. Female (Mean (95% CI)\ *p* \< 0.001 ♂ 12.0 (11.2--12.7) vs. ♀ 10.3 (9.9--10.7) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 10.8 (10.4--11.3) vs. 10.9 (10.4--11.5) Sugar (g/d) Gender Male vs. Female (Mean (95% CI)\ *p* = NS ♂ 34.7 (32.5--36.8) vs. ♀ 34.1 (32.7--35.5) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 34.1 (32.3--35.8) vs. 34.5 (33.0--36.1) Abdul Majid et al. 2016 \[[@B48-ijerph-16-00603]\] Crude fiber (g/d) Gender Male vs. Female (Mean (95% CI)\ *p* = NS ♂ 2.9 (2.7--3.1) vs. ♀ 3.0 (2.8--3.1) Place of residence Urban vs. Rural (Mean (95% CI)\ *p* = NS 3.0 (2.8--3.1) vs. 2.9 (2.7--3.1) Foods Rezali et al. 2015 \[[@B35-ijerph-16-00603]\] Cereals and grains (HEI score) Gender Male vs. Female (Mean ± SD)\ *p* = NS ♂ 5.5 ± 1.9 vs. ♀ 5.4 ± 2.1 Fish (HEI score) Male vs. Female (Mean ± SD)\ *p* \< 0.05 ♂ 1.6 ± 2.0 vs. ♀ 3.4 ± 3.6 Fruit (HEI score) Male vs. Female (Median)\ *p* \< 0.05 ♂ 0 vs. ♀ 0 Legumes (HEI score) Male vs. Female (Mean ± SD)\ *p* = NS ♂ 1.8 ± 2.9 vs. ♀ 1.6 ± 2.6 Vegetables (HEI score) Male vs. Female (Mean ± SD)\ *p* \< 0.05 ♂ 3.7 ± 2.5 vs. ♀ 3.1 ± 2.4 Poultry, meat & egg (HEI score) Male vs. Female (Mean ± SD)\ *p* \> 0.05 ♂ 8.0 ± 2.9 vs. ♀ 8.6 ± 2.6 Milk and milk products (HEI score) Male vs. Female (Median)\ *p* \< 0.05 ♂ 0 vs. ♀ 1 Loh et al. 2017 \[[@B36-ijerph-16-00603]\] Sugar sweetened beverages (SSB) (mL/day) Ethnicity Malay vs. Chinese vs. Indian vs. others (mean ± SE)\ *p* = 0.03 0.76 ± 0.04 vs. 0.44 ± 0.05 vs. 0.55 ± 0.10 vs. 0.63 ± 0.21 Gender Male vs. Female (mean ± SE)\ *p* = NS ♂ 0.68 ± 0.08 vs. ♀ 0.67± 0.03 Maternal education Primary vs. Secondary vs. Tertiary (mean ± SE)\ *p* = NS Primary: 0.62 ± 0.07 vs. Secondary: 0.74 ± 0.05 vs. Tertiary: 0.61 ± 0.06 Dietary Patterns Rezali et al. 2015 \[[@B35-ijerph-16-00603]\] Diet quality (HEI score) Availability of healthy foods Beta = 0.351 *p* \< 0.05 Ethnicity Malay; Beta = −2.416 *p* \< 0.05 Gender Male vs. Female (Mean ± SD)\ *p* \< 0.05 ♂ 34.2 ± 8.2 vs. ♀ 39.9 ± 9.0; Beta ♂ = −5.883 Age r = 0.123 *p* \< 0.05 Self-efficacy for healthy eating Beta = 0.242 *p* \< 0.05 Frequency of breakfast r = 0.038 *p* = NS Abdullah et al. 2016 \[[@B34-ijerph-16-00603]\] Healthy dietary pattern score Age Malay vs. Chinese Beta = 0.141, SE = 0.033 *p* \< 0.001 Beta = 0.165, SE = 0.029 *p* \< 0.001 PA Malay vs. Chinese Beta = 0.142, SE = 0.036 *p* \< 0.001 Beta = 0.10, SE = 0.024 *p* \< 0.001 Eating out Malay vs. Chinese Beta = −0.088, SE = 0.036 *p* = 0.014 Beta = −0.086, SE = 0.026 *p* = 0.001 Ethnicity Malay vs. Chinese (Mean ± SD)\ *p* = 0.039 −0.101 ± 0.957 vs. 0.094 ± 1.03 Fast food consumption Malay vs. Chinese Beta = −0.166, SE = 0.081 *p* = 0.041 Beta = −0.223, SE = 0.068 *p* = 0.001 Maternal education Chinese; Beta = 0.242, SE = 0.114 *p* = 0.035 Local dietary pattern score Eating out Chinese; Beta = 0.067, SE = 0.022 *p* = 0.003 Fast food consumption Chinese; Beta = 0.133, SE = 0.057 *p* = 0.021 Snacking practices Malay vs. Chinese Beta = 0.158, SE = 0.063 *p* = 0.013 Beta = 0.254, SE = 0.096 *p* = 0.009 Ethnicity Malay vs. Chinese (Mean ± SD)\ *p* \< 0.001 0.399 ± 1.05 vs. −0.427 ± 0.73 Nutritional supplements consumption Chinese;\ *p* = 0.027 Beta = −0.216, SE = 0.097 Western dietary pattern score Breakfast skipping Malay;\ *p* \< 0.001 Beta = 0.476, SE = 0.129 Eating out Malay vs. Chinese Beta = 0.109, SE = 0.036 *p* = 0.003 Beta = 0.072, SE = 0.026 *p* = 0.007 Fast food consumption Chinese;\ *p* = 0.023 Beta = 0.156, SE = 0.068 Snacking practices Chinese;\ *p* = 0.004 Beta = 0.157, SE = 0.055 Abdullah et al. 2016 \[[@B34-ijerph-16-00603]\] Western dietary pattern score Soft drink consumption Chinese;\ *p* = 0.023 Beta = 0.080, SE = 0.035 Household income Malay;\ *p* = 0.005 Beta = −0.078, SE = 0.027 Age Malay vs. Chinese Beta = −0.136, SE = 0.033 *p* \< 0.001 Beta = −0.084, SE = 0.029 *p* = 0.004 Ethnicity Malay vs. Chinese (Mean ± SD)\ *p* \< 0.001 0.224 ± 1.04 vs. −0.239 ± 0.89 Nurul-Fadhilah et al. 2013 \[[@B37-ijerph-16-00603]\] Frequency of eating out (times/week) Gender Male vs. Female (%)\ *p* = NS Daily: 8 vs. 7\ 4--6 times/week: 22 vs.32\ 1--3 times/week: 74 vs. 93 Chin & Mohd Nasir 2009 \[[@B31-ijerph-16-00603]\] Meal frequency (meals/daily) Eating companions Family vs. Peer vs. Alone (%)\ *p* \< 0.05 Never skip any meals: 38.7 vs. 33.3 vs. 12.9\ Skipped at least one meal: 52.7 vs. 47.6 vs. 61.3\ Skipped all three meals daily: 12.9 vs. 61.3 vs. 25.8 Ethnicity Malay vs. Chinese vs. Indian (%)\ *p* \< 0.05 Never skip any meals: 27.1 vs. 51.3 vs. 57.7\ Skipped at least one meal: 56.8 vs. 45.2 vs. 3.5\ Skipped all three meals daily: 16.2 vs. 3.5 vs. 0 Living arrangement Staying with family vs. In school hostel (%)\ *p* = 0.051 Never skip any meals: 36.8 vs. 31.4\ Skipped at least one meal: 53.4 vs. 48.6\ Skipped all three meals daily: 9.8 vs. 20 Nurul-Fadhilah et al. 2013 \[[@B37-ijerph-16-00603]\] Snacking frequency (times/day) Gender Snacking frequency\ *p* \< 0.001 Male vs. Female (Mean ± SD)\ ♂ 1.86 ± 1.0 vs. ♀ 2.4 ± 1.1 Rezali et al. 2015 \[[@B35-ijerph-16-00603]\] Snacking frequency (days/week) Male vs. Female (Mean ± SD) Breakfast: ♂ 5.2 ± 2.1 vs. ♀ 4.7 ± 2.6 *p* \< 0.05 Lunch: ♂ 5.9 ± 1.8 vs. ♀ 5.8 ± 2.0 *p* = NS Dinner: ♂ 6.0 ± 1.9 vs. ♀5.8 ± 2.0 *p* = NS --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Note: ♂, Male; ♀, Female; (3M + 3S), 3 meals + 3 snacks; (3M + 2S), 3 meals + 2 snacks; (3M + 1S), 3 meals + one snack; (3M), 3 meals; (≤2M ± 2,3S), meal skippers consumed snacks frequently; (≤2M ± 0,1S), meal skippers consumed snacks only one time or never; HEI, healthy eating index; SE, standard error; NS: Not Statistically Significant (*p* \> 0.05); CI: Confidence interval; OR: Odd ratio. ijerph-16-00603-t003_Table 3 ###### Summary of associations between physical activity determinants and physical activity behaviours. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Correlate Author, Year \[Ref\] Outcome Association *p*-Value ------------------------------------------------------- ------------------------------------------------------------------------------------- ------------------------------------------- ----------------------------------------------------- -------------- Demographics Age Baharudin et al. 2014 \[[@B40-ijerph-16-00603]\] Physical inactivity Inactive vs. Active, OR (95% CI) 1.2 (1.16--1.23) *p* \< 0.001 Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] PA −0.075 (0.101) *p* = NS Gender Baharudin et al. 2014 \[[@B40-ijerph-16-00603]\] Physical inactivity Female vs. Male (ref), OR (95% CI) 2.9 (2.66--3.10) *p* \< 0.001 Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] Female vs. Male (ref), OR (95% CI) 2.176 (1.225--3.866) *p* = 0.008 Farah Wahida et al. 2011 \[[@B43-ijerph-16-00603]\] PA level, MVPA Male vs. Female: (%)\ *p* \< 0.001 Low: ♂ 65.0 vs. ♀ 82.7\ Moderate: ♂ 35.0 vs. ♀ 17.3\ High: ♂ 0 vs. ♀: 0 Dan et al. 2011 \[[@B42-ijerph-16-00603]\] PA Male, Beta: 2.366 *p* = 0.0001 Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] Male vs. Female (ref): ♂ 0.603 (0.062) *p* \< 0.01 Nurul-Fadhilah et al. 2013 \[[@B37-ijerph-16-00603]\] Male vs. Female (Mean ± SD)\ *p* \< 0.001 ♂ 2.1 ± 1.7 vs. ♀ 1.3 ± 0.9 Su et al. 2014 \[[@B47-ijerph-16-00603]\] Male vs. Female Mean (95% CI)\ *p*\< 0.001 ♀ 2.02 (1.91--2.12) vs. ♂ 2.46 (2.29--2.64) Abdul Majid et al. 2016 \[[@B27-ijerph-16-00603]\] Female Median (IQR) Rural: 2.09 (1.72--2.43) in 2012 *p* = 0.006 1.93 (1.56--2.28) in 2014 *p* = NS Cheah et al. 2012 \[[@B46-ijerph-16-00603]\] Male vs. Female (Mean ± SD) Before school: 26.1 ± 22.08 vs. 26.7 ± 23.71 *p* = NS During school: 37.7 ± 36.42 vs. 38.6 ± 36.70 *p* = NS After school: 47.4 ± 37.60 vs. 43.8 ± 35.62 *p* = NS Total time: 111.1 ± 77.70 vs. 109.1 ± 75.45 *p* = NS Teo et al. 2014 \[[@B38-ijerph-16-00603]\] Male vs. Female (Median (95%, CI) Malay ♂1.7 (1.8--2.4) vs. ♀1.1 (1.2--1.5) *p* \< 0.001 Chinese ♂ 1.4 (1.6--2.4) vs. ♀ 0.8 (1.0--1.5) *p* \< 0.01 MVPA duration (h/day) Male vs. Female (Median (95%, CI) Malay ♂1.3 (1.5--2.1) vs. ♀ 0.4 (0.5--0.8) *p* \< 0.001 Chinese ♂ 1.0 (1.4--2.1) vs. ♀ 0.4 (0.6--1.0) *p* \< 0.001 Ethnicity Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] PA Inactive vs. Active (%)\ *p* = 0.007 Malay 17.3 vs. 82.7;\ Others 27.3 vs. 72.7 Dan et al. 2011 \[[@B42-ijerph-16-00603]\] Malay vs. Chinese (%)\ *p* = NS Low: 38.2 vs. 32.1\ Moderate/High: 61.8 vs. 67.9 Su et al. 2014 \[[@B47-ijerph-16-00603]\] Malay vs. Chinese vs. Indian vs. Others (Mean (95% CI)\ *p* \< 0.05 2.21 (2.18--2.24) vs. 1.92 (1.72--2.17) vs. 2.31 (2.03--2.59) vs. 2.50 (2.31--2.68) Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] Chinese & Indian/other vs. Malay(ref) Chinese: −0.496 (0.086) *p* \< 0.01 Indian/other: −0.042 (0.115) *p* = NS Abdullah et al. 2016 \[[@B34-ijerph-16-00603]\] Malay vs. Chinese (Mean ± SD)\ *p* = NS 2.8 ± 1.7 vs. 3.0 ± 2.3 Maternal employment Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] Physical inactivity Not working vs. Working (ref), OR (95% CI)\ *p* = 0.005 2.167 (1.263--3.717) Paternal education Dan et al. 2011 \[[@B42-ijerph-16-00603]\] PA r = 0.105 *p* \< 0.05 Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] Primary vs. Secondary vs. Tertiary(ref)\ *p* = NS, *p* = NS −0.106 (0.131) vs. −0.052 (0.084) Maternal education Dan et al. 2011 \[[@B42-ijerph-16-00603]\] PA r = 0.08 *p* = NS Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] Primary vs. Secondary vs. Tertiary (ref)\ *p* \< 0.1, *p* \< 0.01 −0.248 (0.0130) vs. −0.293 (0.090) Household income Dan et al. 2011 \[[@B42-ijerph-16-00603]\] PA r = 0.08 *p* = NS Household size r = 0.03 *p* = NS Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] 0.062 (0.016) *p* \< 0.01 Parent's marital status Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] Married vs. Divorced/widowed (ref)\ *p* = NS 0.059 (0.137) Physical-Environmental School session Baharudin et al. 2014 \[[@B40-ijerph-16-00603]\] PA Noon vs. Morning (ref), OR (95% CI)\ *p* \< 0.001 1.3 (1.13--1.44) Place of residence Su et al. 2014 \[[@B47-ijerph-16-00603]\] Rural vs. urban (Mean (95% CI)\ *p* = NS 2.14 (1.95--2.32) vs. 2.34 (2.25--2.43) Abdul Majid et al. 2016 \[[@B27-ijerph-16-00603]\] Rural (Median (IQR)\ *p* = 0.013 Rural: 2.24 (1.90--2.70) in 2012\ 2.12 (1.70--2.64) in 2014 Hot weather Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] PA Inactive vs. Active (%)\ *p* = 0.031 Yes: 25.3 vs. 74.7\ No: 20 vs. 80 Equipment not available Inactive vs. Active (%)\ *p* = 0.023 Yes: 26.9 vs. 73.1\ No: 20.7 vs. 79.3 Facility far from home Inactive vs. Active (%)\ *p* = 0.026 Yes: 25.6 vs. 74.4\ No: 19.7 vs. 80.3 Traffic safety Cheah et al. 2012 \[[@B46-ijerph-16-00603]\] r = −0.15 *p* = NS Residential density r = 0.072 *p* = NS Land-use mix diversity r = 0.074 *p* = NS Land-use mix access r = 0.43 *p* = NS Street connectivity r = −0.03 *p* = NS Infrastructure for walking r = −0.078 *p* = NS Aesthetics r = −0.041 *p* = NS Safety from crime r = −0.046 *p* = NS Neighborhood satisfaction r = −0.009 *p* = NS Facility support Abd-Latif et al. 2012 \[[@B45-ijerph-16-00603]\] PA involvement r = 0.069 *p* = 0.038 Usage level of facilities r = 0.094 *p* \< 0.05 Safety r = 0.002 *p* = NS Social-Environmental Family without exercise Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] PA Inactive vs. Active (%)\ *p* = 0.005 Yes: 27.1 vs. 72.9\ No: 16.9 vs. 83.1 Physical education Cheah et al. 2016 \[[@B44-ijerph-16-00603]\] 0.151 (0.018) *p* \< 0.01 Social influence Dan et al. 2011 \[[@B42-ijerph-16-00603]\] Peer; Beta = 0.339 *p* = 0.0001 Family influence r = 0.298 *p* \< 0.001 Behavioral Breakfast intake Baharudin et al. 2014 \[[@B40-ijerph-16-00603]\] Physical inactivity None, Irregular vs. Daily(ref), OR (95% CI) 1.9 (1.74, 2.13) *p* \< 0.001 1.4 (1.33, 1.55) *p* \< 0.001 Stretching is important before exercise Aniza et al. 2009 \[[@B41-ijerph-16-00603]\] PA No vs. Yes (ref), OR (95% CI)\ *p* = 0.004 3.747 (1.540--9.118) Time constraint Yes vs. No (ref), OR (95% CI)\ *p* = 0.004 2.473 (1.335--4.579) Exercise when having ample time No vs. Yes (ref) OR (95% CI)\ *p* = 0.002 2.482 (1.413--4.360) No skills to participate in PA Inactive vs. Active (%)\ *p* = NS Yes: 27.3 vs. 72.7\ No: 19.3 vs. 80.7 Prefer to watch TV Inactive vs. Active (%)\ *p* = 0.005 Yes: 24.9 vs. 75.1\ No: 9.1 vs. 90.9 Embarrassed Inactive vs. Active (%)\ *p* = 0.028 Yes: 30 vs. 70\ No: 21.4 vs. 78.6 Being lazy Inactive vs. Active (%)\ *p*=\<0.0001 Yes: 30 vs. 70\ No: 21.4 vs. 78.6 Too troublesome Inactive vs. Active (%)\ *p* = 0.005 Yes: 32.5 vs. 67.5\ No: 20.6 vs. 79.4 -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Note: ♂, Male; ♀, Female; SE, standard error; NS, Not Statistically Significant (*p* \> 0.05); CI, Confidence interval; OR, Odd ratio; IQR, Interquartile Range; PA, Physical Activity.

Introduction {#s1} ============ Xenotropic murine leukemia virus (MLV)-related virus (XMRV) is a new human retrovirus having a 8.65 kbp genome and shares up to 95% overall nucleotide sequence identity with other known MLVs [@pone.0010255-Urisman1]. XMRV was first reported to be associated with prostate cancer from patients homozygous for a defective variant of RNase L (R462Q), a regulated endoribonuclease for single-stranded RNA that functions in the antiviral action of interferon (IFN) [@pone.0010255-Urisman1], [@pone.0010255-Silverman1]. The Arg to Gln substitution at amino acid position 462 (R462Q) of RNase L is a common missense variant (35% allelic frequency), resulting in a 3-fold decrease in catalytic activity compared with the wild-type enzyme [@pone.0010255-Casey1], [@pone.0010255-Xiang1]. Consistent with the observation that the virus is associated with patients having the homozygous mutant *RNASEL* genotype, XMRV replication *in vitro* is sensitive to IFN-β inhibition [@pone.0010255-Dong1]. The link between XMRV and prostate cancer suggests that inherited defects of RNase L may enhance susceptibility to XMRV, leading to tumorigenesis. However, detection of XMRV has recently been reported in prostate samples independent of the *RNASEL* genotype [@pone.0010255-Schlaberg1]. XMRV has also been detected in the blood of patients with chronic fatigue syndrome [@pone.0010255-Lombardi1]. The causal relationships of XMRV infection to prostate cancer and chronic fatigue syndrome, as well as the mechanism for virus pathogenicity, have yet to be established. Additionally, several studies have failed to detect XMRV in different European cohorts of patients with either prostate cancer [@pone.0010255-Hohn1] or with chronic fatigue syndrome [@pone.0010255-Erlwein1], [@pone.0010255-Groom1], [@pone.0010255-vanKuppeveld1], suggesting that either population differences or environmental factors may modulate the incidence of XMRV infections. Integration of the cDNA copy of the viral RNA genome is essential for retroviruses to establish a productive infection (for reviews, see reference [@pone.0010255-Brown1]). However, because of its nonspecific nature, retroviral DNA integration is inherently a mutagenic event. Many retroviruses, especially members of the gammaretrovirus genus, can induce tumors as a consequence of integrating their viral genome into the host cell chromosome and activating proto-oncogenes via promoter or enhancer insertion, a mechanism referred to as proviral insertional mutagenesis [@pone.0010255-Mikkers1]. XMRV is a member of the gammaretrovirus family, and does not encode host-derived oncogenes [@pone.0010255-Urisman1]. Genome-wide analyses of XMRV integration sites in a human prostate cell line, DU145, and prostate cancer tissues showed that XMRV integration favors gene-dense regions and genomic features frequently associated with structurally open, transcriptional regulatory regions of a chromosome, such as transcription start sites, CpG islands, and DNase hypersensitive sites [@pone.0010255-Kim1]. The XMRV integration sites in prostate cancer tissues are further associated with cancer breakpoints, common fragile sites, and microRNA genes. However, no common integration site or integration hotspot has been detected within or near known proto-oncogenes and tumor suppressor genes in both acutely infected cells and cancer tissues [@pone.0010255-Kim1]. Due to the relatively few integration sites (a total of 14) analyzed thus far in prostate cancer tissues, the role of XMRV infection in causing prostate cancer by insertional mutagenesis is still unclear. Integration of retroviral DNA is catalyzed by the viral enzyme integrase (IN) and involves sequential steps of DNA breaking and joining reactions ([@pone.0010255-Brown1]; and see [Fig. 1A](#pone-0010255-g001){ref-type="fig"}). During integrative recombination, the two ends of the linear viral DNA genome are joined in a concerted fashion to staggered sites on the opposite strands of the target DNA. Gap repair of the integration intermediate results in the formation of a provirus that is flanked by short direct repeats of target DNA, a hallmark of retroviral DNA integration [@pone.0010255-Yoder1], [@pone.0010255-Li1]. The length of the direct repeats, which varies from 4 to 6 bp among the retroviruses but is invariant for each particular retrovirus, presumably corresponds to the spacing of the staggered target DNA sites that are attacked by IN during integration. Analyses of various proviruses together with the associated flanking DNA sequences have revealed high integration fidelity. For instance, 15 of 15 human immunodeficiency virus type 1 (HIV-1) integration sites [@pone.0010255-Muesing1], [@pone.0010255-Vincent1], [@pone.0010255-Vink1], 8 of 8 MLV integration sites [@pone.0010255-Shoemaker1], [@pone.0010255-Shoemaker2], [@pone.0010255-Horowitz1], and 7 of 7 spleen necrosis virus integration sites [@pone.0010255-Shimotohno1], [@pone.0010255-Shimotohno2] have the correct length of the target site duplication. However, certain mutations of the viral genome or reaction conditions can lead to uncoordinated integration of the two viral ends and result in deletions, insertions, or other rearrangements of the host DNA [@pone.0010255-Moreau1], [@pone.0010255-Taganov1], [@pone.0010255-Oh1], [@pone.0010255-Oh2], [@pone.0010255-Vatakis1]. Therefore, in addition to insertional mutagenesis, uncoordinated integration of the two viral ends during integrative recombination may constitute another mechanism that can cause genomic alterations and initiate deleterious events in the infected cell. In this study, we have cloned and determined host DNA sequences flanking XMRV proviruses. We found that integration of XMRV DNA proceeds with high fidelity, and consistently produces a 4-bp direct repeat at the virus-target DNA junctions. Analysis of the 4-bp direct repeats reveals a weak consensus integration sequence. {#pone-0010255-g001} Results {#s2} ======= Fidelity and length of target site duplication during XMRV integration {#s2a} ---------------------------------------------------------------------- The IN-catalyzed integration of retroviral DNA involves sequential DNA breaking and joining steps ([Fig. 1A](#pone-0010255-g001){ref-type="fig"}). To determine the length of the target-site duplication during XMRV integration, we sequenced the stretches of host cell DNA flanking the long terminal repeat (LTR) at each end of a given provirus, and then searched for these flanking sequences within the human genome. To facilitate the analysis, a human prostate cancer cell line DU145 was infected by XMRV and then clonally expanded. Ten infected cell clones were analyzed, and a total of 15 integration site sequences flanking both ends of the XMRV provirus were determined and mapped ([Table 1](#pone-0010255-t001){ref-type="table"}). Three cell clones (C-6, -7, and -8) contained multiple XMRV proviruses, which may have resulted from multiple integration events within the same cell clone or from mixed clonal populations. 10.1371/journal.pone.0010255.t001 ###### Positions of XMRV integration sites and lengths of the target site sequence duplication. {#pone-0010255-t001-1} Cell Clones Integration Site[\*](#nt101){ref-type="table-fn"} (chromosome; nucleotide position) Duplication Length (bp) ------------- ------------------------------------------------------------------------------------- -------------------------------------- C-1 13; 77,016,416 (+) 4 C-3 2; 33,211,657 (+) 273 [†](#nt102){ref-type="table-fn"} C-4 5; 34,622,591 (+) 4 C-5 10; 25,254,665 (+) 4 C-6 1; 19,788,033 (+) 4 2; 19,118,533 (+) 4 C-7 4; 109,005,770 (−) 4 5; 64,073,721 (+) 4 9; 94,680,941 (−) 4 19; 2,119,434 (+) 4 C-8 1; 8,643,694 (+) 4 1; 9,804,426 (+) 4 [ψ](#nt103){ref-type="table-fn"} C-9 2; 109,669,551 (−) 4 C-10 6; 30,858,925 (+) 5 C-12 16; 67,648,746 (−) 4 \*The nucleotide position corresponds to the position of viral DNA insertion at the top strand of the chromosome indicated. Symbols + and -- within the parenthesis indicate the orientation of the viral transcription is the same and opposite, respectively, to the polarity of the top strand. GenBank accession numbers for the integration site sequences are GU816075 to GU816104. †The left LTR of the provirus contains a 5-bp deletion that includes the conserved CA dinucleotide at the viral end. ψThe target DNA contains a T to A transversion immediately adjacent to the left LTR (position 4). Of the 15 XMRV integration sites analyzed, 13 had a 4-bp target site duplication, one site had a 5-bp duplication (clone C-10), and one had a 273-bp duplication (clone C-3). Examination of the viral DNA sequence of the provirus with the 273-bp target site duplication revealed that the left LTR contained a 5-bp deletion at the U3 end that includes a CA dinucleotide that is highly conserved in retroviruses [@pone.0010255-Brown1]. Deletion or mutation of the CA-dinucleotide in the viral donor DNA substrates significantly reduces the efficiency of coordinated integration of two donor molecules into a target DNA [@pone.0010255-Oh2], [@pone.0010255-Aiyar1], [@pone.0010255-Goodarzi1], [@pone.0010255-Hindmarsh1], [@pone.0010255-Li2]. The U3 end deletion in the left LTR might cause an uncoordinated integration of the two XMRV DNA ends, resulting in staggered breaks that were 273-bp apart. For the 13 proviral integration sites with a 4-bp duplication, 12 had duplication sequences that matched correctly with human genomic DNA sequences. The remaining integration site (from clone C-8) contained a T to A transversion at the position 4 within the direct repeat flanking the left LTR (5′-TAAA), while the direct repeat flanking the right LTR (5′-TAAT) matched correctly with the human genomic DNA (5′-TAAT). Since mismatches in the genome would most likely be repaired by host enzymes before integration, we speculate that the transversion was produced by base misincorporation during gap filling or aberrant processing of the unpaired nucleotides at the viral 5′ end, followed by mismatch repair that fixed the mutation ([Fig. 1B](#pone-0010255-g001){ref-type="fig"}). In addition to the length of the direct repeats, analysis of the 15 integration site sequences showed that all viral sequences, with the exception of the left LTR end of the proviral clone C-3, were terminated with the conserved CA dinucleotide at the 3′ end (data not shown), indicating that the viral DNA ends were correctly cleaved by IN [@pone.0010255-Brown1]. Based on our analysis that 87% (13 of 15) of the proviruses had a correct 4-bp direct repeat at the integration site, we conclude that the majority of XMRV integration reactions involve a concerted joining of two viral DNA ends that are spaced 4 bp apart on the target DNA. Base composition surrounding XMRV integration sites {#s2b} --------------------------------------------------- Genome-wide analyses of virus-target DNA junctions reveal a weak consensus integration sequence that is nonetheless unique for each retrovirus examined [@pone.0010255-Holman1], [@pone.0010255-Lewinski1], [@pone.0010255-Berry1], [@pone.0010255-Kim2], [@pone.0010255-Derse1]. This consensus integration sequence is generally palindromic. For instance, the consensus integration sequence for HIV-1 and MLV are 5′-GTWAC and 5′-VTAB, respectively (using standard International Union of Biochemistry base codes: B  =  C, G, or T; V  =  A, C, or G; W  =  A or T) [@pone.0010255-Holman1], [@pone.0010255-Lewinski1], [@pone.0010255-Kim2], [@pone.0010255-Derse1]. To determine the base composition surrounding the XMRV integration site, the target DNA sequences flanking the proviruses were aligned relative to the integration site (between position −1 and D1; [Fig. 2](#pone-0010255-g002){ref-type="fig"}), and the nucleotide frequency of the 4-bp direct repeat (positions D1 to D4; [Fig. 2](#pone-0010255-g002){ref-type="fig"}) and the positions 10 bp upstream (positions −1 to −10) and 10 bp downstream (positions +1 to +10) of the direct repeat were calculated. In addition to the 13 integration site sequences from the cell clones, the analysis included a dataset containing 472 XMRV integration sites from acutely infected DU145 cells and 14 integration sites from human prostate cancer tissues [@pone.0010255-Kim1]. {ref-type="table"}), 472 integration sites from acutely infected DU145 cells (GenBank accession numbers EU981292 to EU981799) and 14 integration sites from human prostate cancer tissues (GenBank accession numbers EU981800 to EU981813) [@pone.0010255-Kim1]. Integration occurs between positions −1 and D1 on the top strand, and between positions D4 and +1 on the bottom strand (blue arrows). Any base in a position that is significantly overrepresented than the random dataset (*P*\<0.0001) is highlighted in green, while any base in a position that is significantly underrepresented than the random dataset (*P*\<0.0001) is highlighted in red.](pone.0010255.g002){#pone-0010255-g002} Comparison of the nucleotide frequency at each position to the value of a random dataset generated *in silico* led to identification of a 5′-CTVB consensus sequence (*P*\<0.0001). Among all the retroviruses analyzed, the consensus integration site sequence of XMRV is most similar to that of MLV [@pone.0010255-Holman1], [@pone.0010255-Lewinski1], [@pone.0010255-Derse1]. Both XMRV and MLV generate a 4-bp target site duplication with thymine favored at the D2 position and adenine disfavored at the D4 position. In addition, thymine was disfavored at the D1 position for both XMRV and MLV. At position D3 of the XMRV integration site sequence, although the only statistical significance at *P*\<0.0001 was the underrepresentation of thymine, adenine was significantly favored at *P*\<0.005. In addition to the 4-bp direct repeat, many positions upstream and downstream of the direct repeat had nucleotide frequencies that were significantly overrepresented (e.g. cytosine and guanine at positions +3 and +9, respectively) or underrepresented (e.g. guanine at position −2) when compared to the random *in silico* control. Furthermore, some of the positions with significantly different representation showed symmetry, such as adenine being favored at position +2 and the corresponding thymine being favored at position −2. Other positions exhibiting a distinct nucleotide preference, however, did not show this symmetry; for example, cytosine was favored at position +3, but guanine was not favored at position −3. Discussion {#s3} ========== XMRV is a newly discovered gammaretrovirus that has been associated with prostate cancer and chronic fatigue syndrome in humans [@pone.0010255-Urisman1]. An important question is whether XMRV has a causal role in initiation or progression of either of these two diseases. In this study, we investigated if integration of XMRV DNA into the host cell chromosome can cause genetic alterations that may subsequently lead to human disease. During integration, the two ends of the linear viral DNA are joined to staggered sites on the opposite strands of the target DNA [@pone.0010255-Brown1]. Subsequent strand separation and gap repair lead to the presence of short direct repeats flanking the proviral DNA [@pone.0010255-Yoder1], [@pone.0010255-Li1]. Therefore, the length of the direct repeats presumably corresponds to the spacing of the two viral ends on target DNA during integrative recombination catalyzed by IN. Analyses of various proviruses have revealed that the length of target site duplication, though varying from 4 to 6 bp among the different retroviruses examined, is invariant for each particular retrovirus [@pone.0010255-Brown1], [@pone.0010255-Shimotohno1], [@pone.0010255-Lewinski2]. The high fidelity of the direct repeat length supports the notion that IN multimers form a stable complex with viral and target DNA and catalyze coordinated processing and integration of the two viral DNA ends [@pone.0010255-Aiyar1], [@pone.0010255-Li2], [@pone.0010255-Murphy1], [@pone.0010255-Moreau2], [@pone.0010255-Li3]. In addition, reaction conditions *in vitro* and *in vivo* that promote uncoordinated integration of the two ends often produce deletions and duplications of various lengths in the target DNA [@pone.0010255-Moreau1], [@pone.0010255-Taganov1], [@pone.0010255-Oh1], [@pone.0010255-Oh2], [@pone.0010255-Vatakis1], [@pone.0010255-Goodarzi1], [@pone.0010255-Lewinski2], [@pone.0010255-Sinha1]. Since the majority of the integrated XMRV contain viral sequences that terminate with the conserved CA dinucleotide and are flanked by a 4-bp direct repeat of target DNA sequence, we conclude that the two viral DNA ends are correctly processed and joined in a coordinated manner to target DNA by IN during XMRV integration. Although retroviruses can access most of the host genome for integration, selection of particular target sites is not random, and the frequency of use of specific sites varies considerably, with some sites being preferred up to several hundred times greater than random [@pone.0010255-Craigie1], [@pone.0010255-WithersWard1], [@pone.0010255-Schroder1]. The mechanism that determines target site specificity is not well understood, and is likely affected by multiple factors [@pone.0010255-HolmesSon1], [@pone.0010255-Bushman1]. Both *in vitro* and *in vivo* studies have implicated IN as one important determinant in specifying a chromosomal or DNA site for integration. INs of different retroviruses exhibit significant differences in the distribution and preference of integration into an identical target substrate *in vitro* [@pone.0010255-Pryciak1], [@pone.0010255-Katzman1], [@pone.0010255-Shibagaki1], and *in vivo*, a chimeric HIV that encodes IN from MLV integrates preferentially into chromosomal features favored by MLV (i.e. transcription start sites and CpG islands) instead of transcription units as favored by HIV-1 [@pone.0010255-Lewinski2]. Although primary DNA sequence is likely not a dominant factor in determining target site specificity, genome-wide analyses of virus-target DNA junctions reveal the presence of weak consensus integration sequences, which are generally palindromic and unique for each retrovirus [@pone.0010255-Holman1], [@pone.0010255-Lewinski1], [@pone.0010255-Berry1], [@pone.0010255-Kim2], [@pone.0010255-Derse1], [@pone.0010255-Hacker1], [@pone.0010255-Kang1], [@pone.0010255-Moalic1], [@pone.0010255-Nowrouzi1]. A weak palindromic consensus sequence is also detected among the XMRV integration sites. We hypothesize that integration of retroviral DNA into a host DNA site depends on the specific interaction between IN and target DNA sequences, resulting in each retrovirus having its own unique, though weak, consensus sequence. The consensus sequence for each retrovirus may be a result of favorable interactions between the DNA bases and certain amino acid residues of IN, or may reflect the amenability of the sequence in adopting particular DNA structures favorable for IN binding. For instance, a common mechanism for stimulating HIV-1 integration is DNA bending, which creates a widened major groove at the outer curved face that is favorable for integration [@pone.0010255-Pryciak1], [@pone.0010255-Muller1], [@pone.0010255-Taganov2], [@pone.0010255-Bor1], [@pone.0010255-Wang1]. The site and fidelity of integration have significant implications for the fate of both the virus and the host cell. Although the present study shows that XMRV integration proceeds with high fidelity, further analysis of additional XMRV integration sites in human tissues would be necessary to clarify whether insertional mutagenesis plays a pathogenic role during XMRV infection. Many viruses from the gammaretrovirus genus of the *Retroviridae* family, such as MLV, feline leukemia virus, and koala retrovirus, are responsible for leukemogenesis and other diseases in their respective host species [@pone.0010255-Rosenberg1]. Therefore, the recent evidence of authentic infections of humans by XMRV and the association of XMRV infection with prostate cancer and chronic fatigue syndrome [@pone.0010255-Urisman1], [@pone.0010255-Schlaberg1], [@pone.0010255-Lombardi1] are alarming and warrant further investigations to determine the causal relationship and pathogenic mechanisms. Materials and Methods {#s4} ===================== Host DNA sequences flanking the XMRV provirus {#s4a} --------------------------------------------- To determine the length and base composition of the target sequence duplication produced by XMRV integration, ten single-clonal (isogenic) populations of XMRV-infected cells were prepared. Plasmid VP62/pcDNA3.1(−) containing the molecular clone of XMRV [@pone.0010255-Dong1] was transfected with Lipofectamine 2000 (Invitrogen) into DU145 cells. The transfected cells were cultured with complete RPMI 1640 media for 3 weeks, trypsinized, diluted, and plated in 96-well plates so that the calculated number of cells per well on average would be 0.15, 0.45, 1.5, 4.5 and 15. The media from wells with a single colony were assayed for reverse transcriptase (RT) activities after 17 to 24 days. Based on high RT activities, ten clones were chosen for integration site analysis. For each clonal population, the cellular DNA sequence at the right LTR-host DNA junction was determined using the linker ligation-mediated PCR assay as described below. Based on the sequence information of the right LTR-host DNA junction, the left LTR-host DNA junction was amplified by nested PCR using forward primers that anneal to positions upstream of the left LTR-host DNA junction and reverse primers that anneal to sequences downstream and within the left LTR. XMRV613R (5′-GATCGCCGGCCGGCTTA), which is complementary to nt positions 597 to 613 of XMRV, and XMRV165R (5′-CCTGACTACAGATATCCTGTTT), which is complementary to nt positions 143 to 165, were used as reverse primers for the first and second PCRs, respectively. The PCR product was electrophoresed on a 1.5% agarose gel, and the expected size of DNA band was excised from the gel and extracted using a gel extraction kit (Qiagen). Extracted DNA was cloned into a pCR-Blunt vector using a Zero Blunt PCR Cloning Kit (Invitrogen). Linker ligation-mediated PCR assay for cloning XMRV integration sites {#s4b} --------------------------------------------------------------------- The genomic DNA from XMRV-infected cells was isolated with a QIAamp DNA Mini Kit (Qiagen) following the manufacturer\'s instruction. The assay for determining XMRV integration sites in DU145 cells was performed as described previously [@pone.0010255-Kim1]. Briefly, genomic DNA from XMRV-infected DU145 cells was digested with *Pst* I, which cuts once in the XMRV genome at nucleotide (nt) position 7,534 and produces on average 4-kbp DNA fragments. After digestion, DNA was denatured and annealed with a biotinylated primer, bXMRV7938 (5′-biotin-ATCCTACTCTTCGGACCCTGT), which is complementary to nt positions 7,938 to 7,958 within the *env* gene (about 160 bp upstream of the right LTR). The annealed primer was extended using the PicoMaxx High Fidelity PCR system (Stratagene) to produce biotinylated double-stranded DNA containing the viral-human DNA junction region. The biotinylated DNA product was then isolated by binding to streptavidin-agarose Dynabeads (Dynal), and digested with *Taq^α^*I (5′-T↓CGA), a 4-bp cutter that does not cleave the viral DNA portion of the biotinylated DNA. Digestion of the human genomic DNA with *Taq^α^*I produces on average 1.9-kbp DNA fragments [@pone.0010255-Gabriel1]. After digestion, the integration site-containing DNA was ligated with TaqLinker, which was prepared by annealing BHLinkA (5′-**CG**GATCCCGCATCATATCTCCAGGTGTGACAGTTT) with TaqLinkS (5′-CACCTGGAGATATGATGCGGGATC). The TaqLinker contains a 2-nt 5′-overhang (in bold type) complementary to the *Taq^α^*I -digested biotinylated DNA. The linker-ligated DNA product was amplified by a two-step PCR process. The first PCR was carried out using primers XMRV8415F (5′-AACCAATCAGCTCGCTTCTC) and Linker1 (5′-TAACTGTCACACCTGGAGATA) in a final volume of 300 µl with 0.5 µM of each primer, 0.2 mM of dNTPs, and 12 U *Pfu* DNA polymerase (Stratagene) under the following condition: 2 min of preincubation at 94°C, followed by 29 cycles at 94°C for 30 s, 58°C for 30 s, and 72°C for 4 min. The PCR product was purified using a PCR Purification Kit (Qiagen), and was used as the template for the second PCR with two nested primers, XMRV8535F (5′-CGGGTACCCGTGTTCCCAATA) and Linker2 (5′-TAGATATGATGCGGGATCCG), which anneal downstream of XMRV8415F and Linker1 binding sites, respectively. The condition for the second PCR was identical to the first PCR except being conducted with only 18 cycles. The second PCR product was electrophoresed on a 1.5% agarose gel and DNA bands between 200 bp to 2 kbp were extracted and cloned into a pCR-Blunt vector using a Zero Blunt PCR Cloning Kit (Invitrogen). Integration site sequence determination and data analysis {#s4c} --------------------------------------------------------- The sequence of the cloned DNA was determined by dideoxy sequencing, and sequencing ambiguities were resolved by repeated sequencing on both strands. The authenticity of the integration site sequence were verified by the following criteria: (i) the sequence contained both XMRV LTR and linker sequence, (ii) a match to the human genome begining after the end of the LTR (5′-...CA-3′) and ending with the linker sequence, and (iii) the host DNA region (containing 20 or more nucleotides) from the putative integration site sequence showed 96% or greater identity to the human genomic sequence. The authenticated integration site sequences were then mapped to the human genome hg18 \[University of California, Santa Cruz (UCSC) March 2006 freeze; National Center for Biotechnology Information (NCBI) Build 36.1\] using BLASTN program (<http://www.ensembl.org>) or BLAT (UCSC; <http://genome.ucsc.edu>). To determine nucleotide preference at integration sites, the target DNA sequences flanking the viral-host DNA junctions were aligned relative to the point of viral DNA integration. The XMRV integration site datasets used to determine nucleotide preference include the 13 correct integration sites listed in [Table 1](#pone-0010255-t001){ref-type="table"} (GenBank accession numbers GU816075, GU816076, GU816079 to GU816100, GU816103, GU816104), 472 integration sites from acutely infected DU145 cells (GenBank accession numbers EU981292 to EU981799) [@pone.0010255-Kim1], and 14 integration sites from human prostate cancer tissues (GenBank accession numbers EU981800 to EU981813) [@pone.0010255-Kim1]. The nucleotide frequency at each position was calculated and compared to values obtained from a set of 10,000 random positions generated *in silico* by choosing a random number between 1 and 3,093,120,360, which represents the total length of the 22 autosomal chromosomes plus the X-sex chromosome of the human genome. The nucleotide frequencies of the random dataset are 29.8%, 20.4%, 20.5%, and 29.3% for A, C, G, and T, respectively. Statistical difference of nucleotide frequency between XMRV integration site sequences and the random dataset was analyzed at each position using a chi-square test at *P*\<0.0001. Nucleotide sequences accession numbers {#s4d} -------------------------------------- The GenBank accession numbers for integration site sequences from the ten XMRV-infected cell clones listed in [Table 1](#pone-0010255-t001){ref-type="table"} are GU816075 to GU816104. We thank Thomas A. Wilkinson for comments on the manuscript. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work was supported by a National Institutes of Health (NIH) Grant CA68859 and The Margaret E. Early Medical Research Trust Grant to S.A.C., and by grant number W81XWH-07-1-338 from the U.S. Department of Defense Prostate Cancer Research Program, NIH Grant CA103943, the Charlotte Geyer Foundation, and the Mal and Lea Bank Chair to R.H.S. S.K. is partly supported by a Dissertation Year Fellowship Award from the UCLA Graduate Division. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: RHS SAC. Performed the experiments: SK AR BD RR. Analyzed the data: SK AR SAC. Contributed reagents/materials/analysis tools: BD RHS. Wrote the paper: SK RHS SAC.

Background {#Sec1} ========== Today, with scientific advances in health and treatment, we are seeing an increasing increase in chronic diseases \[[@CR1]\]. Chronic diseases are principally due to four health risk behaviors: physical inactivity, poor nutrition, tobacco use, and excess alcohol consumption \[[@CR2]\]. In this regard, the importance of exercise is to such an extent that the World Health Organization (WHO) has identified it as the first indicator of health \[[@CR3]\]. Inappropriate physical activity is one of the risk factors for chronic diseases \[[@CR4]\] and is one of the leading causes of mortality worldwide which has doubled the risk of cardiovascular disease (CVDs)**)** \[[@CR5], [@CR6]\], and 23% of these deaths Causes of such diseases \[[@CR7]\]. Low physical activity is one of the causes of overweight and obesity and obesity is a risk factor for non-communicable diseases such as hypertension and cardiovascular disease \[[@CR8]\]. Studies have shown that exercising for 120--120 min at moderate intensity throughout the week can significantly reduce the risk factors for coronary artery disease \[[@CR9]\], although in many countries physical activity levels are poor \[[@CR10]--[@CR12]\]. including Iran \[[@CR13], [@CR14]\]. Exercise is a form of organized physical activity that is regularly, repeatedly, and programmed with the aim of playing and having fun, gaining more fitness, health or fitness \[[@CR15]\]. Its important role in preventing chronic diseases such as CVDs, diabetes, cancer and musculoskeletal diseases has been proven \[[@CR16]\]. Promoting exercise is important because cardiovascular disease is the leading cause of death in Iran and the average age of Iranians with such diseases is 20 years younger than in the European countries and the USA \[[@CR17]\]. Therefore, implementation of appropriate educational interventions based on health education models and theories is necessary. One of the commonly used models in preventing diseases is health belief model (HBM), exclusively created for behaviors related to health \[[@CR18]\]. This model has been the basis of many educational interventions to promote healthy behaviors including exercise \[[@CR19]\]. HBM explains the quality of behavior change in relation to people's health and helps educators evaluate and describe people's health behaviors by understanding their beliefs about health \[[@CR20]\]. Researchers believe that any kind of education leads to learning, but the depth and sustainability of learning varies in different ways \[[@CR21], [@CR22]\], and technology plays a major role in today's knowledge sharing \[[@CR23]\]. Technology-based learning is very useful in creating and developing geographically restricted learning environments and people are affected by information technology \[[@CR23]\]. The use of modern educational technology is increasing so that in one study, both telegram and Instagram were used to assess the impact of continuing education on patients \[[@CR24]\]. In spite of the above, few studies have examined the impact of using virtual networks and messaging programs in education in Iran, while these technologies are not limited to a specific time and place and have a favorable spread and influence among people. Therefore, due to the increasing growth of telegram use and its features in Iran during this study, the research team decided to evaluate the impact of HBM-based educational interventions using telegram on exercise among health staff of Ilam University of Medical Sciences. This article is part of a study. In the previous article extracted from the study referred to the psychometric section of the measurement instrument was reported \[[@CR25]\] In the present article the relevant intervention section is reported. Methods {#Sec2} ======= The present semi-experimental study was conducted in 2017 in the city of Ilam, Iran. Data were collected using a valid and reliable questionnaire (HBM-ISCS) \[[@CR25]\] at two stages of before and 6 months after the educational intervention. The main purpose of the present study was to investigate the effect of HBM-based education on exercise as a CVD preventive factor. In addition, blood biochemical markers (including FBS, TG, Cholesterol, HDL and LDL) were used as measurable markers to indicate changes due to the effect of training intervention on exercise. The conceptual framework {#Sec3} ------------------------ HBM is suitable for preventing chronic diseases when a researcher wants to educate people who are not sick. According to HBM, adopting a health behavior depends on people themselves believing in health problems (inactivity), accepting the reality, being sensitive to its impact on health, feeling threatened, listed as a problem. Take their health seriously and understand it. Different side effects (CVD) in different aspects of their health. Then, with guidance from their surroundings, they are convinced that preventive activities (such as doing exercise) are much less expensive than treating related diseases such as CVDs. As a result, they quickly take such precautionary measures. It is important that we understand why and how people adopt new behaviors, how they have changed behavior, and what has caused them. According to HBM, eople change their behavior when they realize that their illness is serious because they would otherwise become less healthy. The measure {#Sec4} ----------- Data were collected using a valid and reliable questionnaire (HBM-ISCS (at two stages of before and 6 months after the educational intervention. The questionnaire includes 14 items based on the constructs of HBM (two items of perceived susceptibility, three items of perceived severity, two items of perceived benefits, three items of perceived barriers and four items of self-efficacy) to assess the health beliefs of health care workers about the impact of exercise on CVDs. The behavior contained two items for assessing daily and weekly exercise (in minutes). The validity and reliability of HBM-ISCS has been verified by the research team of the present study \[[@CR25]\]. Cronbach's alpha coefficients of 0.715 to 0.816 were calculated for the subscales, perceived susceptibility (0.725), perceived severity (0.715), perceived benefits (0.768), perceived barriers (0.727), self-efficacy (0.816) and HBM-ISCS (0.746). Sampling {#Sec5} -------- The target population of this study consisted of health care workers in the healthcare networks of Ilam province. In order to prevent the delivery of provided information in educational interventions, the healthcare workers in two healthcare networks (draw from 8 health networks) with more than 120 km distance from each other were selected randomly, and by random allocation (by the draw method), the workers of one network were enrolled as intervention group and the workers of the other network were enrolled as control group. First, two cities in Ilam Province were selected randomly and then a list of all Health Care Workers with inclusion terms was prepared by the administrative affairs department. The research team went to the workplace of health care workers and explained the study's objectives and working methods and get a written informed consent from those applicants who wanted to participate in the research. In the following, all applicants were visited by a general physician and their health regarding CVDs was confirmed. At this stage, census method was employed to determine the sample size and for sampling, and all healthcare workers, who had no exercise (the criterion for not having exercise included not having at least 30 min of exercise frequently and with moderate severity for 4 days in a week), were listed to participate in the study if they agreed. It is to be noted that none of the participants in the interventional study was present in the descriptive section. Figure [1](#Fig1){ref-type="fig"} displays the process of sampling, random allocation and follow-up. Fig. 1Flow diagram of participants through the study ### Inclusion/exclusion criteria {#Sec6} Inclusion terms were having literacy, being formal or contractual employee, lack of developing chronic diseases or those leading to limit movement, not having daily exercise, having access to the internet and using social networks for education and completing informed consent form. Exclusion terms were resignation, dismissal, and failure to attend educational programs or developing a disease or condition limiting the movement. ### Educational intervention {#Sec7} The educational intervention was carried out non-person and using the Telegram messaging application. First, a meeting was held with the participants to introduce the program and explaining the main issues of study (including the introduction of research team, objectives, schedule, expectations from the participants, how to communicate with research team, and alike). The non-person education program had two parts: sending the educational package via Telegram to the intervention group, and establishing a group in Telegram for all members of the intervention group. In this group, questions according to the content of educational package were provided while sending the package to provide the opportunity for discussion and interlocution of the group members. The duration of intervention was 2 weeks. The content of educational package was sent to the intervention group in three parts with the aim of providing a thinking opportunity for the participants and also using the reminding role of educational package (each part of the educational content was send every 2 weeks). The group members could send the pictures of their own exercise to the Telegram group. The follow-up duration in the present study was 6 months, and after completing the education course and also during the follow-up, members of the intervention group used to receive reminding messages through the Telegram group to encourage doing exercise. The content of educational intervention was designed based on descriptive study results, which was performed by the research group and it was shown that all constructs of HBM have the ability to predict the participants' exercise behavior. The results of the descriptive study to evaluate the ability of HBM constructs to predict the participants' exercise behavior using structural equation modeling (SEM) showed that the strongest predictive constructs were self-efficacy (43%), perceived susceptibility (41%) and perceived barriers (20%), orderly. The predictive power of the constructs of perceived severity and benefits was 6 and 5%, respectively (Table [1](#Tab1){ref-type="table"}, Fig. [2](#Fig2){ref-type="fig"}). Table 1Fit indices in path analysis (SEM)Fit indexesDegrees of Freedom (df)89χ^2^/ df2.08*P* value0.001Root Mean Square Error of Approximation (RMSEA)0.066Goodness of Fit Index (GFI)0.91Non-Normed Fit Index (NNFI)0.94Comparative Fit Index (CFI)0.95Root Mean Square Residual (RMR)0.073Standardized RMR (SRMR)0.060Incremental Fit Index (IFI)0.95Fig. 2Structural Equation Model (SEM) Diagram Therefore, in addition to defining exercise, types of exercise, and other related topics, the training package included ways to increase self-efficacy, perceived susceptibility, and suggestions for reducing perceived barriers to daily exercise. The content was provided to intervention group only via Telegram and control group members did not receive any training course during the study. In the present study, the female participants had access to the gym and aerobic exercise daily under the supervision of a sports instructor and the men had access to the gym for volleyball and futsal and mountaineering. Participants could at any time perform other sports such as hiking and running. Controlling group exercise as well as its duration and severity was performed by the researcher or one of the colleagues on behalf of the researcher. However, due to the impossibility of controlling individual exercise, participants' daily and weekly exercise data collection was ultimately self-reported. The information in this section was collected with three questions; the first question to examine daily exercise by the participants was answered yes or no. The second and third questions asked about the amount of exercise activities such as fast walking, running, football, futsal, volleyball, mountaineering and etc. in minutes and days per week. The blood samples of the participants were taken in addition to completing the questionnaire at the two stages of before and 6 months after intervention. Some indices were measured such as fasting blood sugar (FBS), Cholesterol, Triglycerides (TGs), LDL and HDL, and were included in the study as objective evidence. ### Data analysis {#Sec8} SPSS (ver. 16) was used to study data analysis. In addition to descriptive methods, we used chi-square, t-tests and repeated measures ANOVA (RMANOVA) in SPSS software for statistical analysis of data. Significance level in all measurements was considered *P* \< 0.05. LISREL (ver. 8.8) was used structural equation model (SEM). Results {#Sec9} ======= Fifty percent of the participants were men and 58.77% of them had BSc degree. As indicated, there is no statistically significant difference between the two groups in terms of gender, age, marital status, education, work experience, daily and weekly exercise, in terms of family history of CVDs and death due to CVDs in the family (*P* ≥ 0.05).In other words, the two groups are homogenous (Tables [2](#Tab2){ref-type="table"} & [3](#Tab3){ref-type="table"}). Table 2Survey of the groups' consistency in terms of individual variables before interventionVariablesIntervention group (*N* = 59)Control group (*N* = 55)*P*-valueN%N%Gender Female2949.22850.90.500 Male3050.82749.1Marriage Married5288.141740.092 Single711.91426Education Diploma813.6610.9 Associate Degree1220.3916.30.516 Bachelor3152.53665.4 Master's degree813.647.2Exercise (daily) Yes0000 No5910055100VariableMean ± SDMean ± SDExercise (weekly)13.3 ± 18.5624.75 ± 20.180.627Job experience6.03 ± 13.86.03 ± 13.790.088Age (years)4.65 ± 37.646.14 ± 37.50.168Table 3CVD history among the participantsVariablesIntervention Group (*N* = 59)Control Group (*N* = 55)*P*-valueN%N%Family history of CVDs Yes2135.62545.40.519 No3355.92545.4 I do not know58.559.2The history of death from CVDs in the family Yes610.2814.50.404 No5288.14480 I do not know11.735.5 Based on the results of the repeated measures analysis of variance test for HBM structures in intervention group, there was no significant difference in perceived barriers (before and after the intervention) (*P* ≥ 0.05); but in other structures, this difference was significant (*P* \> 0.05). The above changes were not significant for all constructs in the control group (P ≥ 0.05). Given the significance of the interaction between time and group in perceived susceptibility, the results of the main effects test may be misleading. Therefore, t-test (Bonferroni correction) was used to compare groups at different times. There was no significant difference between groups before intervention for perceived susceptibility (*P* = 0.154), but there was a significant difference after intervention (*P* = 0.036) (Table [4](#Tab4){ref-type="table"}). Table 4Comparison of HBM structures between the intervention and control groups before and after intervention by RMANOVAVariablesPre-interventionPost-intervention*P*-valueFMean ± SDMean ± SDPerceived susceptibility^a^ Intervention Group0.81 ± 9.570.18 ± 9.97Sig. = 0.018F(1,112) = 5.718 Control Group0.77 ± 9.740.54 ± 9.76Perceived severity Intervention Group1.02 ± 13.851.82 ± 12.78Sig. = 0.000F(1,112) = 20.390 Control Group1.77 ± 12.931.9 ± 13.14Perceived benefits Intervention Group0.58 ± 9.631.06 ± 10.9Sig. = 0.010F(1,112) = 6.913 Control Group1.13 ± 8.891.36 ± 9.03Perceived barriers Intervention Group3.01 ± 9.462.22 ± 8.24Sig. = 0.634F(1,112) = 0.998 Control Group3.21 ± 8.713.12 ± 9.11Self-efficacy Intervention Group2.99 ± 16.781.32 ± 19.18Sig. = 0.024F(1,112) = 0.955 Control Group3.16 ± 16.212.94 ± 16.41^a^ The interaction between time and group was significant Based on the results of the repeated measures analysis of variance test, there was significant difference between daily and weekly exercise before and after the intervention in intervention group (*P* \> 0.05). The above changes were not significant in the control group (*P* ≥ 0.05). Given the significance of the interaction between time and group in daily exercise and weakly exercise, the results of the main effects test may be misleading. Therefore, t-test (Bonferroni correction) was used to compare groups at different times. There were no significant difference between groups before intervention for daily exercise and weakly exercise (*P* ≥ 0.05), but there were a significant difference after intervention in both them (*P* \> 0.05) (Table [5](#Tab5){ref-type="table"}). Table 5Comparison of the average regular activity between the intervention and control groups before and after intervention by RMANOVAVariablesPre-interventionPost-intervention*P*-valueFMean ± SDMean ± SDDaily exercise (Minute)^b^Intervention Group0^a^1.72 ± 25.23Sig. = 0.001F(1,112) = 105.010Control Group0^a^4.16 ± 1.09Weakly exercise (Minute)Intervention Group31.34 ± 18.5541.74 ± 129.76Sig. = 0.001F(1,112) = 94.182Control Group24.75 ± 20.183.85 ± 24.54^a^The groups without daily exercise^b^The interaction between time and group was significant Based on the results of the repeated measures analysis of variance test for blood biochemical markers of the participants in intervention group, there were no significant difference in TG, Cholesterol, LDL and HDL (before and after the intervention) (*P* ≥ 0.05); but in FBS, this difference was significant (*P* \> 0.05). Given the significance of the interaction between time and group in FBS, the results of the main effects test may be misleading. Therefore, t-test (Bonferroni correction) was used to compare groups at different times. There was no significant difference between groups before intervention for FBS (*P* = 0.604), but there was a significant difference after intervention (*P* = 0.003) (Table [6](#Tab6){ref-type="table"}). Table 6Comparison of the mean blood indices between the intervention and control groups before and after intervention- by RMANOVAVariablesPre-interventionPost-intervention*P*-valueFMean ± SDMean ± SDFBS (Mg/dl) Intervention Group7.92 ± 77.917.19 ± 76.23Sig. = 0.771F(1,112) = 0.085 Control Group8.72 ± 78.729.03 ± 80.94TG (Mg/dl) Intervention Group24.83 ± 166.0123.41 ± 160.29Sig. = 0.732F(1,112) = 0.118 Control Group27.46 ± 164.6525.41 ± 171.02Cholesterol (Mg/dl) Intervention Group30.16 ± 185.0226.96 ± 181.15Sig. = 0.822F(1,112) = 0.051 Control Group29.04 ± 186.3228.25 ± 192.09LDL (Mg/dl) Intervention Group21.78 ± 11520.07 ± 112.25Sig. = 0.982F(1,112) = 0.0001 Control Group21.01 ± 116.1621.30 ± 117.65HDL (Mg/dl) Intervention Group6.06 ± 37.015.26 ± 38.56Sig. = 0.872F(1,112) = 0.026 Control Group5.86 ± 37.253.79 ± 35.95 Discussion {#Sec10} ========== The present research results showed that before the educational intervention, there was no significant difference in the mean of HBM constructs and also in the amount of the participants' daily and weekly physical activities, but the educational intervention using Telegram led to significant changes in the HBM constructs, and consequently, the participants' daily and weekly exercise. Perceived susceptibility is one of the very effective factors in adopting preventive behaviors \[[@CR26], [@CR27]\]. In the study by Zeinali conducted on the evaluation of impact of educational intervention on promoting preventive behaviors for CVDs among people with natural Angiography \[[@CR26]\], and also in the study by Amirzade, the educational intervention led to the change of mean score of perceived susceptibility of women that is consistent with the results of present study \[[@CR28]\]. In the study by Karimi, there was a significant change of all HBM constructs, including perceived susceptibility before educational intervention \[[@CR29]\]. In this study, the increase of perceived susceptibility was directly related to the increase of preventive behavior (i.e. doing exercise) for CVDs. Not having sufficient understanding about the seriousness of CVDs risk (i.e. low perceived severity) can be a barrier in the way of changing the lifestyle of people and preventing such diseases \[[@CR30]\]. In the present study, we observed the significant increase of the mean scores of perceived severity in the intervention group after intervention, which can be one of the reasons of behavior change among the participants about exercise; this result is supported by various studies. For instance, in the study of Tahernia, the mean scores of perceived severity had significant increase after intervention with the aim of promoting preventive behaviors for CVDs \[[@CR31]\]. Also Mardani and Shamsi observed significant changes of HBM constructs after intervention that are same as the present study \[[@CR32], [@CR33]\] . Individual preventive behavior is defined as the perception of an individual about the benefits and barriers of behaviors and balance between them \[[@CR34]\]. On the other hand, the more is the perceived benefits of adopting preventive behavior in a person's opinion, the greater will be the possibility of doing that behavior. In a study by Abood, the educational intervention led to increasing the perceived benefits effectively in the intervention group \[[@CR35]\], and in the study of Ghaderian, which was conducted with the aim of evaluating the impact of educational intervention on promoting preventive behaviors for CVDs among health care workers in the health centers of the west of Ahvaz city, Iran, there was a significant increase in the scores of perceived benefits \[[@CR36]\]. On the other hand, one of the defining factors of doing exercise is the barriers a person encounters with while taking these behaviors, and in contrast, his/her ability to cope with the barriers of doing physical activity is significantly and positively related to the increase of exercise \[[@CR37], [@CR38]\], to the extent that some researchers have reported that perceived barriers are among the most important component of HBM in conducting recommended behaviors \[[@CR39]\]. The importance of perceived barriers and benefits is especially observed when the goal of suggested behavior is to prevent an unhealthy behavior \[[@CR40]\]. In the present study, the educational intervention led to the significant decrease of perceived barriers in the intervention group for doing exercise, and the reverse relationship between the scores of perceived barriers and exercise was observed. In the study of Peyman with the aim of evaluating the impact of educational intervention on the lifestyle of girl students, there was a significant decrease of perceived barriers among the participants about exercise and healthy diet \[[@CR41]\]. Also the findings of Mohseni-pouya support the results of this study, implying educational intervention significantly decreases barriers of people's perception about preventing CVDs \[[@CR42]\]. In this study, the mean score of perceived barriers was decreased significantly as one of the predictive factors for exercise behavior after intervention. This decrease, besides the increase of perceived benefits, highlighted the benefits of exercise among the participants in the intervention group and increased the amount of daily and weekly exercise significantly in this group. Self-efficacy is among those variables that results in the increase of exercise in people \[[@CR43]\] and among the effective factors for taking healthy behaviors \[[@CR44]\]. The results of the present study showed that there is a direct relationship between self-efficacy and exercise, which is consistent with the results of other studies \[[@CR45]--[@CR47]\]. Also the increase of self-efficacy after educational intervention could increase the mean amount of the participants' daily and weekly exercise. Our results showed that doing daily exercise and the mean of daily and weekly exercise were also increased in the control group though this increase was not significant. We talked with three members of the control group who began daily exercise and found out that two of them according to the suggestion of their physician and one because of concerns about overweight have begun daily exercise in a limited amount. These results are consistent with the results of other studies about exercise. For example, the educational intervention in the study of Tahernia using HBM led to a significant statistical difference in the physical activity amount of the participants \[[@CR31]\]; this result is in agreement consistent with findings of Hadad in Jordan \[[@CR48]\]. Furthermore, various other studies have reported a significant increase in the amount of physical activity after educational intervention \[[@CR49]--[@CR53]\]. In simple terms, educational intervention for doing exercise resulted in controlling FBS, TG, Cholesterol and LDL in the intervention group and also decreased the mean level of blood biochemical markers between 1.67 and 5.72 units, while the lack of exercise in the control group led to the increase of the mean level of these indices for about 1.49 to 6.36 units. In other words, exercise could improve the mean of blood biochemical markers in the intervention group and not increased or worsened the mean of blood biochemical markers in the control group. This shows that though exercise could not make any significant changes in the intervention group but led to control and minor the improvement of these indices, which is very important. Meanwhile, the mean of HDL for the intervention group members was increased by as much as 1.54 units post-intervention; in other words, exercise resulted in improving the blood level of HDL in the intervention group though this change was not significant. On the other hand, the mean of HDL in the control group decreased by 1.31 units; in other words, not having exercise worsened the HDL level in the control group. As mentioned above, intervention based on exercise could not make any significant changes in the blood biochemical markers of the intervention group, and the research team tried to explain its reason(s). Thus, it was tried to identify similar studies and compare their results with the results of the present study; for example, in the study of Sardar and colleagues \[[@CR54]\], 8 weeks of aerobic exercise led to decrease the FBS, Cholesterol, LDL and HDL but the changes were not significant statistically. Ramalho \[[@CR55]\] and Herbst \[[@CR56]\] also observed similar results to the findings of Sardar \[[@CR54]\]. Alizadeh et al. also studied the impact of 40-min walking with moderate severity for 5 days a week in three groups (40 min continuously, 40 min periodic without physical activity); the results showed that these three groups had no significant difference in changes of Cholesterol, FBS and TG statistically \[[@CR57]\]. An important point in Alizadeh's study was that all the three groups had limited daily calorie and received their dietary advice from a dietitian \[[@CR57]\]. The results of all of the above studies are consistent with the present study. These studies have shown that exercise alone is not sufficient to make a significant change in the blood biochemical markers and requires reviewing the intervention program. This review can be done on the amount of exercise, its severity, follow-up period or addition of other sections such as diet to the intervention. This has been confirmed in numerous studies where the interventions were based on lifestyle or combination of healthy diet and physical activity, which were done on people from different age groups and in different countries around the world with different cultures; in these studies, the interventions led to improve the levels of FBS, TG, Cholesterol, LDL and HDL significantly \[[@CR58]--[@CR62]\]. The results of this section of the study, assuming a same diet among the study groups, found that exercise can control and make minor improvement in the mean of FBS, TG, Cholesterol and HDL in the intervening group, and the lack of exercise worsens their level in the control group; this shows the impact of exercise in controlling the blood indices. Even though the intervention based on exercise could not make any significant changes in the blood indices of the intervention group, it led to decrease/control the blood indices in them, which is of high importance itself. Because, for example, 1 % decrease in the amount of Cholesterol and LDL can decrease the amount of risk factors of CVDs accordingly \[[@CR63]\]. Limitations {#Sec555} ----------- Among the limitations of this study, we can refer to the impact of some factors including personality traits and psychic moods when answering the questionnaire as well as the inability to control the dietary status of the participants. Extracting information based on self-reporting is also among the shortcomings of this study. Conclusions {#Sec11} =========== The results of this study indicated that exercise is closely related to the health care worker's beliefs, and applying HBM can be well effective in this regard. Therefore, implementing educational interventions with emphasis on health beliefs and using HBM can be an appropriate strategy to change health care worker's behavior (doing exercise) in order to prevent CVDs. In addition, exercise is effective in controlling the level of FBS and stopping its retardation towards an undesirable level of health though this variable alone is not enough to improve the above indices. The results also showed that Telegram is a perfect tool to deliver educational materials, especially in the difficult conditions of holding in-person classes. Therefore, despite the lack of access to this application in Iran in the current situation, the results of the present study showed that modern messaging technologies in the form of messengers, media and social networks can be used to deliver information and, in some cases, the desired health skills in order to achieve the ultimate goal of promoting the health care workers' health level and quality of life. HBM : Health Belief Model CVDs : Cardiovascular diseases SEM : Structural Equation Modeling ISCS : Impact of Sport on the Cardiovascular Diseases Scale **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We are grateful to all participants who made this study possible. Authors 'contributions {#FPar1} ====================== RJ, FG and MT contributed to design, analysis and wrote the first draft. AAHM: contributed to analysis. All authors read and approved the final manuscript. The author(s) received no specific funding for this project. All data generated and/or analyzed during the current study are available by the responsible author if needed. Research Council and Medical Ethics Committees of Tarbiat Modares University approved the study (Code no. IR.TMU.REC.1394.148). All participants signed the written consent form. Not applicable. The authors declare that they have no competing interests.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec006} ============ Children and adolescents with low levels of physical activity (PA) are at increased risk of becoming inactive adults, and of diseases which result from inadequate activity \[[@pone.0179429.ref001], [@pone.0179429.ref002]\]. In addition to low PA, there is increasing concern that high levels of sedentary time (ST) may also be common, and both low PA and high ST are important risk factors for chronic disease \[[@pone.0179429.ref003], [@pone.0179429.ref004]\]. Multiple national and international bodies have recommended a minimum of 60 minutes of moderate-to-vigorous intensity physical activity (MVPA) every day for school-age children and adolescents \[[@pone.0179429.ref005], [@pone.0179429.ref006]\]. These recommendations for the general population are usually also considered to be applicable to children and adolescents with chronic disease \[[@pone.0179429.ref007]--[@pone.0179429.ref009]\], with an understanding that usual levels of MVPA might be lower in such sub-groups, and achievement of the MVPA recommendation would be a slower and more gradual process than in the healthy population \[[@pone.0179429.ref007], [@pone.0179429.ref010]--[@pone.0179429.ref012]\]. Objective techniques such as accelerometry currently represent the most accurate methods for measuring the amount and intensity of PA and amount of ST \[[@pone.0179429.ref013], [@pone.0179429.ref014]\]. While there have been many studies on the levels and adequacy of MVPA and ST in healthy children and adolescents \[[@pone.0179429.ref015]--[@pone.0179429.ref018]\], there are surprisingly few such studies in children and adolescents with chronic disease. In fact, numerous previous studies and national PA surveillance programs have actually excluded children and adolescents with chronic disease. The primary aim of the present study was therefore to examine whether children and adolescents with chronic disease met the current MVPA recommendation \[[@pone.0179429.ref001], [@pone.0179429.ref006], [@pone.0179429.ref019]\]. Secondary aims were to examine the amount of accelerometer-measured ST in children and adolescents with chronic diseases, and to determine whether accelerometer measured MVPA and ST in children and adolescents with chronic disease were different from those in healthy control or comparison groups. This systematic review provides evidence on whether levels of MVPA are adequate and ST excessive in children and adolescents with chronic disease. Methods {#sec007} ======= Review governance and registration {#sec008} ---------------------------------- A systematic review of the literature was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines \[[@pone.0179429.ref020]\] as shown in [S1 File](#pone.0179429.s001){ref-type="supplementary-material"}. The review protocol was registered on PROSPERO (registration number CRD42015016783), the international prospective register for systematic reviews (<http://www.crd.york.ac.uk/NIHR_PROSPERO>) see [S2 File](#pone.0179429.s002){ref-type="supplementary-material"}. Study eligibility {#sec009} ----------------- **Inclusion criteria:** To be eligible for inclusion in the review, papers had to meet all of the following criteria as per PICOS principles: [**P**]{.ul}opulation (children and adolescents aged from 0--19 years); [**I**]{.ul}ntervention or exposure: chronic childhood disease (chronic disease defined as any physical health problem that lasts three months or more). The chronic diseases included were decided on following a scoping review and were cardiovascular disease, respiratory disease, diabetes type 1 or type 2, and malignancies); [**C**]{.ul}omparison (where applicable): healthy children matched for relevant criteria (in particular age, gender); [**O**]{.ul}utcome (accelerometer measured MVPA and/or ST of at least 3 days and 6 hours/day). All [**s**]{.ul}tudy designs were considered eligible (cross-sectional, longitudinal, case-control studies and intervention studies if pre-intervention data could be extracted). We looked for original research studies, published in English, in peer-reviewed journals; a detailed description of the study eligibility criteria is given in [S1 Table](#pone.0179429.s003){ref-type="supplementary-material"}. **Exclusion criteria:** Studies that included participants with co-morbid acute or chronic medical diseases or conditions that may have impacted their physical activity were excluded. The present study aimed to examine the subtle impact of chronic disease on MVPA and ST, not the more obvious impacts from co-morbidities that preclude physical activity (e.g. arising from injury or acute illness requiring bed rest, and chronic physical limitations from e.g. cerebral palsy). Because of the on going debate about whether obesity is a disease, studies in children with obesity were also excluded and are the subject of a separate report. Since the aim of the review was to examine habitual levels of MVPA and ST, studies that measured these variables for less than 6 hours per day or over two days or less were excluded. Recommendations currently exist for habitual (overall) MVPA rather than MVPA during specific domains (e.g. the after school period) and so studies that focused only on specific periods of the day (e.g. school activity only, or outdoor activity only, or weekend or weekday activity only, or after-school only) were also excluded. ### Search strategy {#sec010} The literature search was conducted using the five most relevant electronic databases: MEDLINE OVID; Cochrane library; EMBASE; SPORTSDiscus and CINAHL. We searched from the year 2000 (to increase generalisability, since levels of MVPA and/or ST might be different now than in the past, and because accelerometry became more widely used in research from the early 2000's) up to March 2017. The literature search strategy used in MEDLINE is given in [Table 1](#pone.0179429.t001){ref-type="table"}, and was adapted as required for the other four databases. Full literature search details are available from the corresponding author on request. The electronic search was complemented by reference tracking (forward and backward) of the eligible studies. 10.1371/journal.pone.0179429.t001 ###### Search strategy used for MEDLINE database. {#pone.0179429.t001g} -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 1\. exp Child/ 2\. exp Adolescent/ 3\. (child\* or adolesc\* or teen\* or youth or girl\* or boy\*).tw. 4\. (young adj1 (person or people)).tw. 5\. 1 or 2 or 3 or 4 6\. exp Exercise/ 7\. exp Motor Activity/ 8\. exp Sports/ 9\. exercis\*.tw. 10\. physical\* activ\*.tw. 11\. (active adj2 (living or lifestyle)).tw. 12\. sedentary behavi?r.tw. 13\. exp Sedentary Lifestyle/ 14\. ((sedentary or sitting or screen or TV or television or computer or PC or video games) adj2 time).tw. 15\. 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 16\. exp Obesity/ or exp Overweight/ 17\. (overweight or obes\*).tw. 18\. exp Accelerometry/ 19\. exp Actigraphy/ 20\. acceleromet\*.tw. 21\. actigraph.tw. 22\. activity monitor\*.tw 23\. (objective adj1 (measure\* or monitor\* or assess\*)).tw. 24\. 18 or 19 or 20 or 21 or 22 or 23 25\. exp cardiovascular abnormalities/ or exp heart diseases/ 26\. exp Cardiovascular Abnormalities/ 27\. exp Heart Diseases/ 28\. \"congenital heart disease\".tw. 29\. \"Atrial Septal Defect\".tw. 30\. \"Complete Atrioventricular Canal Defect \".tw. 31\. \"Ventricular Septal Defect\".tw. 32\. (Tetralogy adj2 Fallot).tw. 33\. exp Asthma/ 34\. asthma.tw. 35\. exp Respiratory Tract Diseases/ 36\. exp Respiratory Hypersensitivity/ 37\. exp Cystic Fibrosis/ 38\. (respiratory adj2 allerg\*).tw. 39\. \"cystic fibrosis\".tw. 40\. wheez\*.tw. 41\. exp Bronchopulmonary Dysplasia/ 42\. lung diseases/ or exp alpha 1-antitrypsin deficiency/ or exp \"cystic adenomatoid malformation of lung, congenital\"/ or exp hepatopulmonary syndrome/ or exp hypertension, pulmonary/ or exp lung diseases, fungal/ or exp lung diseases, interstitial/ or exp lung diseases, obstructive/ or exp lung diseases, parasitic/ or exp lung injury/ or exp lung neoplasms/ or exp lung, hyperlucent/ 43\. \"chronic lung disease\".tw. 44\. \"chronic respiratory disease\".tw. 45\. exp diabetes mellitus, type 1/ or exp diabetes mellitus, type 2/ 46\. (diabetes adj1 mellitus).tw. 47\. exp Leukemia/ 48\. Leukemia.tw. 49\. exp Lymphoma/ 50\. Lymphoma.tw. 51\. exp Neuroblastoma/ 52\. Neuroblastoma.tw. 53\. \"Wilms' tumor\".tw. 54\. exp Central Nervous System Neoplasms/ 55\. exp Sleep Apnea Syndromes/ 56\. \"sleep apnea\".tw. 57\. 16 or 17 or 25 or 26 or 27 or 28 or 29 or 30 or 31 or 32 or 33 or 34 or 35 or 36 or 37 or 38 or 39 or 40 or 41 or 42 or 43 or 44 or 45 or 46 or 47 or 48 or 49 or 50 or 51 or 52 or 53 or 54 or 55 or 56 58\. 5 and 15 and 24 and 57 59\. exp Adult/. 60\. 58 not 59 61\. limit 60 to (english language and yr = \"2000 -Current\") -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ### Study selection {#sec011} Titles, abstracts, and full-text articles were screened in duplicate for eligibility and disagreements were resolved through discussions with other reviewers when required. Reference lists of eligible studies were examined for potentially eligible studies. Reasons for exclusion are summarised in the study flow diagram and available in details from the corresponding author on request. ### Data extraction and data synthesis {#sec012} This review used a standard form for extracting relevant information from the eligible studies. The systematic review identified that the eligible studies fell logically into four categories: cardiovascular disease; respiratory disease; diabetes; malignancy. A fifth category (obesity) was identified but this is reported separately as noted above. Obesity was not included here because of the on-going debate about whether obesity is a disease or not. International recommendations for school-age children and adolescents specify at least 60 minutes of MVPA every day \[[@pone.0179429.ref005], [@pone.0179429.ref006]\], but in the eligible studies the achievement of MVPA recommendations was never operationalised in this way. In most studies, only the mean or median daily MVPA was provided (rather than achievement of MVPA recommendations on 7/7 days), and so this was used as a proxy for achievement of guideline recommendations in the present study. Where suitable data for patients and healthy controls were reported, mean and standard deviation of MVPA and ST in minutes per day, and sample sizes for similar chronic disease conditions were combined in a random effects model accounting for heterogeneity between studies. Given the differing methods of determining MVPA levels obtained from accelerometers, differences in MVPA between patients and controls were generated as weighted standardised mean difference (SMD). While methodology (e.g. accelerometer model, accelerometry cut point and/or epoch) varied substantially between studies, within study comparisons are all based on the same methods. Separate meta-analyses were performed for MVPA and ST. Review Manager 5.2 was used for the quantitative analysis \[[@pone.0179429.ref021]\]. Some eligible studies recruited healthy control participants and measured MVPA and/or ST in the same way as in their patient group and at the same time (referred to here as studies with controls), while other studies compared patient data with other studies (e.g. published data) and are referred to here as studies with comparison groups; some studies simply reported patient data in relation to physical activity recommendations. ### Quality assessment {#sec013} Eligible articles were assessed for methodological quality using a 15-item quality assessment scale as shown in [S2 Table](#pone.0179429.s004){ref-type="supplementary-material"}, collapsed to 6 items for scoring, with higher scores suggesting higher study quality. Each eligible study was assessed independently by two authors (RE, JJR), and disagreements were resolved by discussion. The quality assessment scale was modified from the methodological quality assessment scale of Tooth *et al*. \[[@pone.0179429.ref022]\]. This is a reliable and valid 30-item tool for assessing the quality of observational studies, and was considered for use without modification initially. After careful reflection, modifications to the original scale were made to focus quality assessment on issues of particular importance to accelerometry measurement of PA. A modified Tooth tool has been used previously with several recent systemic reviews of PA studies with 8--17 items, that were usually collapsed to a smaller number of items for scoring \[[@pone.0179429.ref023]--[@pone.0179429.ref026]\]. Results {#sec014} ======= Identification of eligible studies {#sec015} ---------------------------------- The PRISMA flow diagram with numbers of included and excluded articles at each step of the review process is provided in [Fig 1](#pone.0179429.g001){ref-type="fig"}. Tables [2](#pone.0179429.t002){ref-type="table"}--[5](#pone.0179429.t005){ref-type="table"} provide a brief summary of all studies included in this systematic review. Of the 1592 identified records from the five databases, 504 were selected for full text screening and of these, 24 met the inclusion criteria. Additionally, 1 study was identified and deemed eligible through searching references of eligible studies, bringing the final total to 25 eligible studies (7 in cardiovascular disease; 7 in respiratory disease; 8 in diabetes; 3 in malignancies) and 11 of these 25 studies were suitable for inclusion in meta-analysis (4 in those with respiratory disease; 5 in those with diabetes; 2 in those with malignancies). All eligible studies measured MVPA, and 16 out the 25 eligible studies compared levels of MVPA between patients with chronic disease and a healthy control group (referred to here as studies with controls), while the other 8 eligible studies compared data from patients with data from previously published studies of healthy children and adolescents (referred to here as studies with comparison groups); and two studies simply reported patient data in relation to recommendations; 14 of the 25 studies also provided data on ST, and 10 of these 14 studies compared ST in those with chronic disease with a healthy control group, while another 4 eligible studies compared data from patients with data from previously published studies of healthy children and adolescents comparison groups. {#pone.0179429.g001} 10.1371/journal.pone.0179429.t002 ###### Descriptive characteristics and levels of moderate-to-vigorous intensity physical activity and sedentary time in children with cardiovascular disease. {#pone.0179429.t002g} ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Reference[\*](#t002fn002){ref-type="table-fn"} Place and publication year Sample group\ Control group\ Measurement Accelerometry methods MVPA Sedentary time n, male% and age (yrs) n, male% and age (yrs) --------------------------------------------------------------------------------- ------------------------------------------------------- -------------------------------------------------------------------- ------------------------------------------------------------------ ------------------------------------------------------------------------------ ------------------------------------------------------------ ------------------------------------- -------------------------------------------------------- **Banks et al** \[[@pone.0179429.ref027]\] Canada, 2012 n: 20, (60% male) N/A Actigraph MTI, worn above the iliac crest for 4 days including 1 weekend day Epoch = 60s Median 8 (Q1 4---Q3 11) min/day N/A Age: mean 11.8 (SD 3) MVPA cut-points not clearly reported MVPA lower in patients than healthy comparison groups All patients failed to meet an average of 60 min/day MVPA **Banks et al** \[[@pone.0179429.ref028]\] Canada, 2013 n: 50, (59% male) N/A Actical, worn above the iliac crest for 7 days Epoch = 15s Mean 52 (SD 20) min/day N/A Age: range 6--12 MVPA ≥ 1600 cpm MVPA lower in patients than healthy comparison groups All patients failed to meet an average 60 min/day MVPA **Duncombe et al** \[[@pone.0179429.ref029]\] Canada, 2016 n: 90, (54% male N/A Actigraph GT3X, worn on right hip for 7 days during waking hours Epoch = 15s, Evenson cut-point \[[@pone.0179429.ref030]\] Median 43 (IQ range 29--60) min/day Median 70 (IQ range 61--76) % of waking time sedentary Age: mean 13.6 (SD 2.7) MVPA ≥ 2296 cpm MVPA similar in patients and healthy comparison groups Sedentary time similar in patients and healthy comparison groups Sedentary time \< 100 cpm 8% of patients achieved an average 60 min /day MVPA **Ewalt et al** \[[@pone.0179429.ref031]\] USA, 2012 n: 21, (24% male) n: 21, matched for age and gender Actigraph 7164, worn on right hip for 7 days during waking hours Epoch = 30s Patients mean 71 (SD 50) min/day Patients mean 399 (SD 107) min/day Age: mean 10.7 (SD 3.2) MVPA cut-points not clearly reported Control group mean 61 (SD 30) min/day Control group mean 406 (SD 90) min/day Sedentary time ≤ 50 counts /30s MVPA difference not significant (p = 0.2) Sedentary time difference not significant (p = 0.7) 33% of patients and 5% of control group achieved an average 60 min /day of MVPA **Gardner et al** \[[@pone.0179429.ref032]\] Canada, 2016 n: 30, (46% male) N/A Actigraph GT3X, worn on right hip for 7 days during waking hours Epoch = 15s, Evenson cut-point \[[@pone.0179429.ref030]\] Median 40 (IQ range 27--57) min/day Median 68 (IQ range 61--76) % of waking time sedentary Age: mean 10.7 (SD 3.2) MVPA ≥ 2296 cpm 25% of patients achieved an average 60 min /day of MVPA Sedentary time \< 100 cpm Longmuir et al \[[@pone.0179429.ref033]\] Canada, 2011 n: 63, (60% male) N/A Actical, worn above the iliac crest for 7 days during waking hours Epoch = 15s Mean 51 (19) min/day N/A Age: range 5--11 MVPA ≥ 1,600 cpm MVPA 4 to 5 times lower in patients than healthy comparison groups 3% of patients achieved an average of 60 min per day of MVPA **McCrindle et al** \[[@pone.0179429.ref034]\] USA and Canada. 2007 n: 147, (62% male) N/A Actigraph MTI, worn for 4 days include 1 weekend day during waking hours Epoch = 60s, Freedson cut-point \[[@pone.0179429.ref035]\] Absolute MVPA not given N/A Age: range 6--18 MVPA lower in patients than healthy comparison groups 38% of patients achieved an average of 60 min per day of MVPA ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- cpm: counts per minutes; MVPA: moderate-to-vigorous physical activity; n: Number; N/A: No data included; S: Second; Data are expressed as mean (SD) unless otherwise; *Freedson MVPA cutpoint* \[[@pone.0179429.ref035]\] *calculated using the following equation*: *METS = 2*.*757 + (0*.*0015 x counts/min)--(0*.*08957 x age (yr))--(0*.*000038 x counts/min x age (yr))* \*One study \[[@pone.0179429.ref031]\] recruited healthy control and patient participants at the same time and measured MVPA and ST in the same way as (referred to here as studies with control group); 5 studies \[[@pone.0179429.ref027]--[@pone.0179429.ref029], [@pone.0179429.ref033], [@pone.0179429.ref034]\] compared patient data with other studies-previous published studies (referred to here as studies with comparison groups); and1 study \[[@pone.0179429.ref032]\] reported patient data in relation to recommendations. 10.1371/journal.pone.0179429.t003 ###### Descriptive characteristics and levels of moderate-to-vigorous intensity physical activity and sedentary time in children with chronic respiratory diseases. {#pone.0179429.t003g} ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Reference[\*](#t003fn002){ref-type="table-fn"} Place and publication year Sample group\ Control group\ Measurement Accelerometry methods MVPA Sedentary time n, male% and age (yrs) n, male% and age (yrs) ----------------------------------------------------------------------------------- --------------------------------------------------------- ---------------------------------------------------------------------- ------------------------------------------------------------ ------------------------------------------------------------------------------------- ------------------------------------------------------------ ----------------------------------------------------------------- --------------------------------------------- **Aznar et al** \[[@pone.0179429.ref036]\] Spain, 2014 n: 47, (51% male) **n**: 39; (59% male) Actigraph GT3X, worn on right hip for 7 days during waking hours Epoch = 15s, Evenson cut-point \[[@pone.0179429.ref030]\] Patients mean 44 (SD 28) min/day Patients mean 362 (SD 67) min/day Age: mean 12.0 (SD 3.0) Age: mean 12.0 (SD 2.0) MVPA ≥ 2296 cpm Control group mean 54 (SD 15) min/day Control group mean 484 (SD 85) min/day Sedentary time \< 100 cpm MVPA significantly lower in patient (p \< 0.02) Sedentary time was lower in patients (p \< 0.001) 2% of patients and 34% of control group achieved an average 60 min/ day of MVPA **Kilbride et al** \[[@pone.0179429.ref037]\] Ireland, 2012 n: 16, (56% male) n: 99, (48% male) RT3 worn for 3 days Epoch and MVPA cut-points not clearly reported Patients mean boys 15 (SD 3); girls 16 (SD 3) min/day N/A Age: range 10--12 Age: range 10--12 Sedentary time = 0--99cpm Control group mean boys 23 (SD 6); girls 15 (SD 3) min/day MVPA similar in patients and control group All patients failed to meet an average of 60 minutes/day MVPA **Smith et al** \[[@pone.0179429.ref038]\] Germany, 2016 n: 94, (56% male) n: 590, (40% male) Actigraph GT3X, worn on right hip for 7 days during waking hours Epoch = 60s. Freedson cut-point \[[@pone.0179429.ref035]\] Patients mean boys 41 (35) 11, 89; girls 43 (36) 14, 71 min/day N/A Age: mean 15.6 (SD 0.5) Age: mean 15.7 (SD 0.5) Control group mean boys 46 (42) 19, 89; girls 38 (34) 13, 70 min/day **Tsai et al** \[[@pone.0179429.ref039]\] USA, 2012 n: 27, (70% male) n: 27, (59% male) actigraph (Actiwatch 64 MM), worn non-dominant wrist for 7 days during waking hours Epoch = 60s Patients mean 265 (SD 83) min/day Patients mean 87 (SD 48) min/day Age: range 9--11 Age: range 9--11 MVPA ≥ 700 cpm Control group mean 308 (SD 97) min/day Control group mean 77 (SD 27) min/day Sedentary time = 0--49 cpm MVPA similar in patients and control group (p = 0.09) Sedentary time difference not significant (p = 0.3) Patients and control group achieved an average of 60 min per day of MVPA **Vahlkvist et al** \[[@pone.0179429.ref040]\] Denmark, 2010 n: 55 n: 154 RT3 worn for 4 weeks during 24 h a day Epoch and cut-points not clearly reported Patients mean of 32 (95% CI 5) min/day Patients mean 1270 (95% CI 15) min/day Age: range 6--14 Age: range 6--14 Control group mean 34 (95% CI 3) min/day Control group mean 1261 (95% CI 9) min/day MVPA similar in patients and control group Sedentary time was similar in patient and control group Patients and control group failed to achieve an average of 60 min per day of MVPA **Van- Gent et al** \[[@pone.0179429.ref041]\] Netherlands, 2007 n: 81, (58% male) n: 202, (50% male) Pam AM 100, worn on hip for 5 days during waking hours Epoch = 60s Patients mean 99 (95%, CI 80, 118) min/day N/A Age: mean 9.4 (SD 0.8) Age: mean 9.4 (SD 0.7) MVPA cut-points not clearly reported Control group mean 98 (95%, CI 85, 106) min/day MVPA similar in patients and control group Patients and control group achieved an average of 60 min per day of MVPA **Yiallouros et al** \[[@pone.0179429.ref042]\] Cyprus, 2015 n: 36, (64% male) n: 99, (60% male) Actigraph worn on wrist for 7 days during waking hours Epoch not clearly reported Patients mean 15 (95% CI 10--21) min/day Patients mean 939 (95% CI 915--963) min/day Age: range 8--9 Age: range 8--9 MVPA \> 3200 cpm Control group mean 16 (95% CI 14--19) min/day Control group mean 927 (95% CI 915--938) min/day Sedentary time \< 100 cpm Similar in patients and control group Similar in patients and control group ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- cpm: counts per minutes; MVPA: moderate-to-vigorous physical activity; n: Number; S: Second; Data are expressed as mean (SD) unless otherwise in Smith et al \[[@pone.0179429.ref038]\]: MVPA calculated as mean (Geometric mean) 5^th^, 95^th^ percentile, and Yiallouros et al \[[@pone.0179429.ref042]\]: MVPA Calculated as geometric means (95% CI) in min/day.; *Freedson MVPA cutpoint* \[[@pone.0179429.ref035]\] *calculated using the following equation*: *METS = 2*.*757 + (0*.*0015 x counts/min)--(0*.*08957 x age (yr))--(0*.*000038 x counts/min x age (yr))* \*Studies are recruited healthy control participants and measured MVPA and ST in the same way as in their patient participants and at the same time(referred to here as studies with control group). 10.1371/journal.pone.0179429.t004 ###### Descriptive characteristics and levels of moderate-to-vigorous intensity physical activity and sedentary time in children with diabetes mellitus. {#pone.0179429.t004g} -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Reference[\*](#t004fn002){ref-type="table-fn"} Place and publication year Sample group\ Control group\ Measurement Accelerometry methods MVPA Sedentary time n, male% and age (yrs) n, male% and age (yrs) ----------------------------------------------------------------------------------- ------------------------------------------------- --------------------------------------------------------------------- ----------------------------------------------------------------------- --------------------------------------------------------------------- ------------------------------------------------------------- ----------------------------------------------------------------- -------------------------------------------------------------------- **Cuenca-Garcia et al** \[[@pone.0179429.ref043]\] UK, 2012 n: 60, (67% male) n: 37, (54% male) Actigraph GT1M, worn for 7 days during waking hours Epoch = 60s Patients mean 28 (SD 21) min/day N/A Age: mean 12.5 (SD 2.3) Age: mean 12 (SD 2.5) MVPA ≥ 3200 cpm Control group mean 20 (SD 11) min/day MVPA difference not significant (p = 0.06) **Kriska et al** \[[@pone.0179429.ref044]\] USA, 2013 n: 669, (51.5% male) N/A Actigraph AM7164, worn on waist for 7 days during waking hours Epoch = 60 s. Freedson cut-point \[[@pone.0179429.ref035]\] 10--14 year old mean boys 35 (SD 26); girls 27 (SD 18) min/ day 10--14 year old mean boys 495 (SD 144); girls 479 (SD 141) min/day Age: range 10--17 Sedentary time \< 100 cpm 15--18 year old mean (boys 26 (SD 24); girls 8 (SD 9) min/day 15--18 year old mean boys 526 (SD 143); girls 546 (SD 143) min/day MVPA low in patients than healthy comparison groups All patients failed to reach an average of 60 min/day MVPA **MacMillan et al** \[[@pone.0179429.ref045]\] UK, 2014 n: 40, (50% male) N/A Actigraph GT3X, worn on waist for 7 days during waking hours Epoch = 15s Patients mean of 43 (SD 24) min/day Patients mean 612 (SD 102) min/day MVPA ≥ 3200 cpm MVPA similar in patients and healthy comparison groups Age: mean 11.1 (SD 2.7) Sedentary time \< 100 cpm 5% of patients achieved an average of 60 min per day of MVPA **Maggio et al** \[[@pone.0179429.ref046]\] Switzerland, 2010 n: 45 n: 85 Actigraph 6471, worn for 7 days during waking hours Epoch not clearly reported Patients mean 54 (SD 7) min/day Patients mean 77% waking time sedentary Age: mean 10.7 (SD 0.4) Age: mean 10.1 (SD 0.3) MVPA \> 2000 cpm Control group mean 71 (SD 5) min/day Control group mean 70% waking time sedentary Sedentary time \< 500 cpm The difference not significant (p = 0.07) Significantly higher in patient (p \< 0.01) 39% of patient and 60% of control group achieved an average of 60 min/day of MVPA **Nguyen et al** \[[@pone.0179429.ref047]\] Canada, 2015 n: 16; (n = 8) good glycemic control, (n = 8) poor glycemic control n: 8 Actigraph GT1, worn on the right hip for 7 days during waking hours Epoch = 3s, Evenson cut-point \[[@pone.0179429.ref030]\] Patients with good glycemic control mean 46 (SD 16) min/day N/A Age: range 8--16 Age: range 8--16 MVPA ≥ 2296 cpm Patients with poor glycemic control mean 47 (SD 8) min/day Sedentary time \< 100 cpm Control group mean 54 (SD 28) min/day The difference not significant (p = 0.07) **Sarnblad et al** \[[@pone.0179429.ref048]\] Sweden, 2005 n: 26, (100% female) n: 49 Actigraph 6471 worn on the hip for 7 days during waking hours Epoch = 60s Patients mean of 56 (SD 20) min/da Patients mean 443 (SD 60) min/day Age: range 12--19 Age: range 12--19 MVPA \> 1952 cpm Control group mean 60 (SD 23) min/day Control group mean 390 (SD 27) min/day Sedentary time \< 100 cpm The difference not significant (p = 0.07) Significantly higher in patient than control group (*P* = 0.002) **Sundberg et al** \[[@pone.0179429.ref049]\] Sweden, 2012 n:24, (50% male) n: 26, (46.2% male) Actiheart, for 7 days Epoch = 60s Absolute MVPA not given Absolute sedentary time not given Age: mean boys 4.3 (SD 1.6); girls 4.7 (SD 1.9) Age: mean boys 4.9 (SD 1.4); girls 4.4 (SD 1.8) MVPA cut-point not clearly reported MVPA significantly lower in patients than the control group (p = 0.02 Significantly higher in patient than control group (p = 0.03) Sedentary time \< 100 cpm **Trigona et al** \[[@pone.0179429.ref050]\] Switzerland, 2010 n: 32, (53% male) n: 42, (40% male) Actigraph 6471, worn on hip for at least 4 days during waking hours Epoch = 60s Patients mean 53 (95% CI 33--74) min/day Age: range 6--17 Age: range 6--17 MVPA \> 2000 cpm Control group mean 77 (95% CI 58--97) min/day Significantly lower in patients than the control group (p \< 0.008) 35% of patients and 57% control group achieved an average of 60 min/day of MVPA -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- cpm: counts per minutes; MVPA: moderate-to-vigorous physical activity; n: Number; S: Second; Data are expressed as mean (SD) unless otherwise; *Freedson MVPA cutpoint* \[[@pone.0179429.ref035]\] *calculated using the following equation*: *METS = 2*.*757 + (0*.*0015 x counts/min)--(0*.*08957 x age (yr))--(0*.*000038 x counts/min x age (yr))*. \*6 studies \[[@pone.0179429.ref043], [@pone.0179429.ref046]--[@pone.0179429.ref050]\] recruited healthy control participants and measured MVPA and ST in the same way as in their patient participants and at the same time (referred to here as studies with control group), and 2 studies \[[@pone.0179429.ref044], [@pone.0179429.ref045]\] compared patient data with other studies (referred to here as studies with comparison groups. 10.1371/journal.pone.0179429.t005 ###### Descriptive characteristics and levels of moderate-to-vigorous intensity physical activity and sedentary time in children with malignancies. {#pone.0179429.t005g} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Reference[\*](#t005fn002){ref-type="table-fn"} Place and publication year Sample group\ Control group\ Measurement Accelerometry methods MVPA Sedentary time n, male% and age (yrs) n, male% and age (yrs) ---------------------------------------------------------------------------- ----------------------------------------------------------------------- ------------------------------------------------------------------- ---------------------------------------- -------------------------------------------------------------- ------------------------------------------------------------ ---------------------------------- ---------------------------------------------- **Aznar et al** \[[@pone.0179429.ref051]\] Spain, 2006 n: 7, (57.1% male) n: 7, (57.1% male Actigraph MTI, worn on waist for 7 days during waking hours Epoch = 60s, Freedson cut-point \[[@pone.0179429.ref035]\] Patients mean 47 (SD 15) min/day Patients mean 41 (SD 18) % of time sedentary Age: range 4--7 Age: range 4--7 Sedentary time ≤ 100 cpm Control group mean 72 (SD 25) min/day Control group mean 42 (SD 11) % of time sedentary Significantly lower in patient group than control group (p = 0.04) Sedentary time was similar in patient and control group (p = 0.07) 0% of patient, 57% control group achieved an average of 60 min/day of MVPA **Gotte et al** \[[@pone.0179429.ref052]\] Germany, 2017 n: 28, (57% male) N/A Step Watch 3^™^ Monitor, worn for 7 days during waking hours Epoch = 60s Patients mean 4 (SD 5) min/day N/A Age: range 11--15 MVPA cut-point not clearly reported 3% (n = 1) of patient achieved an average of 60 min/day of MVPA **Tan et al** \[[@pone.0179429.ref053]\] Malaysia, 2012 n: 38 n: 38 Actical, worn on hip for 7 days during all the day Epoch = 15s Patients mean 20 (SD 28) min/day Patients mean 1295 (SD 119) min/day Age: range 3--12 Age: range 3--12 MVPA cut off points not reported Control group mean 168 (SD 56) min/day Control group mean 925 (SD 111) min/day Sedentary time \< 100 cpm Significantly lower in patient group than control group (*p* \< 0.01) Significantly higher in patients than control group (*p* \< 0.01) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- cpm: counts per minutes; MVPA: moderate-to-vigorous physical activity; n: Number; S: Second; Data are expressed as mean (SD) unless otherwise; *Freedson MVPA cutpoint* \[[@pone.0179429.ref035]\] *calculated using the following equation*: *METS = 2*.*757 + (0*.*0015 x counts/min)--(0*.*08957 x age (yr))--(0*.*000038 x counts/min x age (yr))* \*2 studies \[[@pone.0179429.ref051], [@pone.0179429.ref053]\] recruited healthy control participants and measured MVPA and ST in the same way as in their patient participants and at the same time (referred to here as studies with control group), 1 study \[[@pone.0179429.ref052]\] reported patient data in relation to recommendations. ### Study characteristics {#sec016} Study samples: Eligible study sample sizes ranged from 14--699 with a total of 2062 participants with chronic disease and 1523 participants from healthy control groups. All studies were from high-income, developed nations. Measurement methods: A total of 17 out of the 25 eligible studies used the ActiGraph accelerometer to measure habitual MVPA and/or ST, though with a variety of different ActiGraph models and approaches to data collection and reduction. Of the remaining studies: three used the Actical \[[@pone.0179429.ref028], [@pone.0179429.ref033], [@pone.0179429.ref053]\]; two the RT3 "Triaxial Research Tracker" \[[@pone.0179429.ref037], [@pone.0179429.ref040]\]; one the PAM "Physical Activity Monitor" B.V. type AM 100 \[[@pone.0179429.ref041]\]; one the Actiheart (which combines accelerometry and heart rate monitoring) \[[@pone.0179429.ref049]\]; and one the Step Watch 3^™^ \[[@pone.0179429.ref052]\]. MVPA in children and adolescents with chronic disease {#sec017} ----------------------------------------------------- The mean reported daily MVPA accumulated by children and adolescents with chronic disease across the 25 eligible studies ranged between 4 (SD 4) minutes/day \[[@pone.0179429.ref052]\] to 265 (SD 83) minutes/day \[[@pone.0179429.ref039]\]. **Children and adolescents with cardiovascular disease:** Seven of the 25 eligible studies (n = 442) examined MVPA in children and adolescents previously diagnosed with a congenital heart defect. This included children who had received different types of cardiac surgery \[[@pone.0179429.ref029], [@pone.0179429.ref031], [@pone.0179429.ref032]\], including complex surgery such as a fontan repair \[[@pone.0179429.ref028], [@pone.0179429.ref033], [@pone.0179429.ref034]\], or cardiac transplantation \[[@pone.0179429.ref027]\]. In all cases the patients were studied at least 6 months after surgery while well, clinically stable, free of acute illness, and living in the community. As summarised in [Table 2](#pone.0179429.t002){ref-type="table"}, average daily MVPA in these studies ranged from a low of 8 (range, 4--11) min/day \[[@pone.0179429.ref027]\] to a high of 49 (range, 34--60) min/day \[[@pone.0179429.ref033]\]. In 6/7 of eligible studies reported mean daily time spent in MVPA in minutes and in six studies mean daily MVPA failed to reach the recommended 60 minutes. Only 1 out of 7 eligible studies included data from healthy control group, showing a higher MVPA level in the patient group compared to healthy control group although differences were not significant \[[@pone.0179429.ref031]\]. **Children and adolescents with chronic respiratory diseases:** Seven of the 25 eligible studies (n = 1013) investigated children and adolescents with chronic respiratory diseases; five studies in patients with asthma \[[@pone.0179429.ref038]--[@pone.0179429.ref042]\] and two in patients with cystic fibrosis \[[@pone.0179429.ref036], [@pone.0179429.ref037]\]. In all cases, the children were studied while clinically stable, free of acute illness, and while living in the community. The mean daily MVPA reported ranged from 15 (SD 3) min/day \[[@pone.0179429.ref037]\] to 265 (SD 8) min/day \[[@pone.0179429.ref039]\] as summarised in [Table 3](#pone.0179429.t003){ref-type="table"}. In 2/7 eligible studies \[[@pone.0179429.ref039], [@pone.0179429.ref041]\] mean daily reported MVPA reached or exceeded the 60 minutes recommended, though this included one study with the exceptionally high reported levels of MVPA (30). Meta-analysis of all 4 studies indicated lower MVPA levels in the patient group compared to the healthy control group, approaching statistical significance. The standardised mean difference (SMD) was 0.39 (95% CI -0.80 to 0.02, p = 0.06). The heterogeneity was substantial with an I^2^ statistic of 68%, as shown in [Fig 2](#pone.0179429.g002){ref-type="fig"}. {#pone.0179429.g002} **Children and adolescents with diabetes:** Eight of the 25 eligible studies (n = 1323) involved children and adolescents with diabetes mellitus; 7 studies in children with type 1 diabetes \[[@pone.0179429.ref043], [@pone.0179429.ref045]--[@pone.0179429.ref050]\] and 1 in children with type 2 diabetes \[[@pone.0179429.ref044]\]. Again, in all cases the patients were studied while clinically stable and free of acute illness or diabetes complications, and while living in the community. As summarised in [Table 4](#pone.0179429.t004){ref-type="table"}, the average daily MVPA reported for diabetic patients ranged from a low of 8 (SD 9) min/day \[[@pone.0179429.ref044]\] to a high of 56 (SD 20) min/day \[[@pone.0179429.ref048]\]. Of the 8 eligible studies, 7 reported mean daily MVPA in minutes and in all 7 of these studies MVPA was \< 60 minutes. Patient MVPA was compared to healthy peers in 5 studies \[[@pone.0179429.ref043], [@pone.0179429.ref046]--[@pone.0179429.ref048], [@pone.0179429.ref050]\], all of which included patients with type 1 diabetes as shown in [Fig 3](#pone.0179429.g003){ref-type="fig"}. There was no evidence of a statistically significant difference in MVPA in patients compared to healthy controls (SMD -0.70, 95% CI -1.89 to 0.48, p = 0.25, n = 400). Case-control evidence on MVPA levels appears to be lacking for patients with type 2 diabetes. {#pone.0179429.g003} **Children and adolescents with malignancies:** Three studies, (n = 118) examined MVPA in those with malignancies including one study in acute lymphoblastic leukaemia on maintenance treatment \[[@pone.0179429.ref051]\], one in acute leukemia undergoing induction or consolidation chemotherapy \[[@pone.0179429.ref053]\] and a third in children and adolescents with different types of childhood malignancies \[[@pone.0179429.ref052]\]. In all of these studies, the participants had no other co-morbid conditions that would have been a contraindication for PA such as anemia, fever, or other difficulties with mobility. The mean daily MVPA achieved during these studies ranged from a low of 4 (SD 4) min/day \[[@pone.0179429.ref052]\] to a high of 47 (SD 15) min/day \[[@pone.0179429.ref051]\], as summarised in [Table 5](#pone.0179429.t005){ref-type="table"}. In all three studies mean daily MVPA failed to reach the recommended 60 minutes. Two out the three studies included data from healthy participants and in both of these studies, the level of MVPA was significantly lower in children and adolescents with malignancies \[[@pone.0179429.ref051], [@pone.0179429.ref053]\]. [Fig 4](#pone.0179429.g004){ref-type="fig"} shows the combined result, which suggests a standardised mean difference of 2.2 (95% CI -4.08 to -0.26, p = 0.03). Despite the apparent similarity between the studies the statistical heterogeneity was considerable with an I^2^ statistic of 88%. The heterogeneity noted could be due to differences in sample sizes, age and place of the studies, differences in the stage of treatment, or differences in accelerometer methodology. {#pone.0179429.g004} Sedentary time in children and adolescents with chronic disease {#sec018} --------------------------------------------------------------- In this systematic review, 14 out of the 25 eligible studies reported on accelerometer measured ST, with a total of 1870 participants (those with chronic disease n = 1325; healthy control group n = 545). Of the 14 studies that measured ST, the chronic diseases studied were: cardiovascular disease---3 studies \[[@pone.0179429.ref029], [@pone.0179429.ref031], [@pone.0179429.ref032]\]; chronic respiratory diseases---4 studies \[[@pone.0179429.ref036], [@pone.0179429.ref039], [@pone.0179429.ref040], [@pone.0179429.ref042]\]; diabetes---5 studies \[[@pone.0179429.ref044]--[@pone.0179429.ref046], [@pone.0179429.ref048], [@pone.0179429.ref049]\]; malignancy---2 studies \[[@pone.0179429.ref051], [@pone.0179429.ref053]\]. As summarised in Tables [2](#pone.0179429.t002){ref-type="table"}--[5](#pone.0179429.t005){ref-type="table"}, the mean daily time spent sedentary in these eligible studies ranged from a low of 87 (SD 48) min/day \[[@pone.0179429.ref039]\] to a high of 1295 (SD 119) min/day \[[@pone.0179429.ref053]\]. In 10 out of the 13 eligible studies, there was a healthy control group, and in 4/10 studies ST was significantly higher in those with chronic disease than in the healthy control groups \[[@pone.0179429.ref046], [@pone.0179429.ref048], [@pone.0179429.ref049], [@pone.0179429.ref053]\], in one study ST was significantly lower in the patient group compared to the healthy control group \[[@pone.0179429.ref036]\], and in 5 studies there was no significant group difference \[[@pone.0179429.ref031], [@pone.0179429.ref039], [@pone.0179429.ref040], [@pone.0179429.ref042], [@pone.0179429.ref051]\]. Suitable summary data for combining individual study data were only available for three studies (n = 355) in patients with chronic respiratory disease with findings indicating no statistically significant group difference in time spent sedentary (SMD -0.40, 95% CI -1.53 to 0.74, p = 0.49, I^2^ = 95%). Study quality assessment {#sec019} ------------------------ Study quality assessment summaries are given in [S3 Table](#pone.0179429.s005){ref-type="supplementary-material"}: 3 studies scored 4/6; 13 scored 5/6; 9 scored 6/6 on study quality. Thus in general, studies were high methodological quality. Discussion {#sec020} ========== This systematic review provides evidence that children and adolescents with some chronic childhood diseases have lower than recommended levels of MVPA. In most of the eligible studies, daily MVPA averaged less than the 60 minutes/day recommended. When comparing MVPA level between patients and healthy control or comparison groups, the findings indicated, within the limits of the available data, no marked differences for patients with type 1 diabetes, CVD and chronic respiratory diseases. In patients with leukemia compared to healthy control or comparison groups daily MVPA was significantly lower. With the respect to sedentary time the present review found that studies fairly consistently reported that children and adolescents with chronic disease accumulated a high amount of ST during their waking hours. It should be noted that recommendations for MVPA state that 60 minutes per day is a minimum every day (e.g. usually operationalized as all 7 days in a week), but adherence to recommendations was not operationalized in this way in any of the 25 eligible studies. We therefore used a mean or median daily MVPA of 60 minutes as a proxy for compliance, though this is conservative because in many cases where 60 minutes/day was reached as an average, levels of MVPA would have fallen below 60 minutes/day on at least one of the monitored days. Reasons for lower than recommended levels of MVPA are unclear. Children and adolescents with chronic disease may experience an over-protective care environment, a lack of supervised facilities/ opportunities for PA, and/or insufficient knowledge and self-efficacy about the types of PA suitable for the specific disease condition \[[@pone.0179429.ref054], [@pone.0179429.ref055]\]. Such socio-environmental influences could contribute to low daily MVPA and high ST. Healthcare professionals, parents/caregivers and schools may need to be provided with adequate information and training to be able to encourage and support children with chronic disease to engage in regular and appropriate MVPA. However, it should also be noted that reported levels of MVPA among healthy peers were also generally low in the eligible studies, so it may be that any constraints on PA which apply to healthy children and adolescents also apply equally to those with chronic disease. We believe that the present study is the first systematic review to ask whether or not levels of MVPA are adequate in children and adolescents with chronic childhood disease. There are therefore no directly comparable studies, but we note that in healthy children, and particularly in healthy adolescents, there is concern that levels of MVPA are generally much lower than recommended. A global analysis by Hallal et al suggested that less than 20% of 13--15 year olds meet the recommendation of 60 minutes/ day of MVPA \[[@pone.0179429.ref056]\]. A recent pooling of international accelerometry data from nearly 21,000 healthy children and adolescents showed typically very low levels of adherence to the 60 minutes/day recommendation for MVPA \[[@pone.0179429.ref057]\], so it is perhaps not surprising that levels of MVPA among those with chronic disease were also found to be generally low in the present study. We are also unaware of any previous systematic reviews of accelerometry measured ST among children and adolescents with chronic disease. Interpreting sedentary time data is even more problematic than interpreting the MVPA data in the present study because there are currently no evidence-based recommendations for accelerometer-measured ST. Our review had a number of strengths. It was the first review to investigate objective levels of MVPA and ST in children and adolescents living with childhood chronic disease. Secondly, there are several methodological strengths to this study: in particular, studies were identified from an extensive search of the published literature conducted in a range of databases. The broad definition of search terms applied across multiple databases enabled the searching and identification across many potential studies. Restricting eligibility to accelerometry studies was important in increasing confidence in the objective measurement of MVPA \[[@pone.0179429.ref016], [@pone.0179429.ref058]--[@pone.0179429.ref060]\]. Finally, all included studies were in general rated as being of high or very high quality. However, there are several weaknesses worth highlighting. Firstly, as studies had to be published in peer-reviewed journals in English, this may have excluded some relevant evidence. Studies included in our review investigated MVPA and ST in children and adolescents with chronic childhood disease. However, we excluded some other common medical conditions where significant alterations in activity might have been expected because of the nature of the condition e.g. musculo-skeletal and neurological disease, and we also excluded studies of patients with acute illness or injury requiring or associated with confinement or bed-rest. Future reviews should consider these other groups, and also consider the PA and ST of children and adolescents with the many chronic diseases not included in the present review. Our initial scoping review found that objectively measured PA data were available for only a few chronic disease groups and so the present review focused on those. All eligible studies were from high-income developed nations. We therefore lack data from low-middle income countries where the prevalence of many childhood chronic disease will be common and lack of resources may limit medical care \[[@pone.0179429.ref061]\]. Most of the included studies were based on relatively small samples of children with chronic disease (n 14--699) and their power to estimate habitual MVPA, or to distinguish between MVPA of patients and comparison group participants, might have been limited, and their representativeness was rarely clear. Our method for assessing the quality of eligible studies has been used in variously adapted forms in a number of other recent accelerometry systematic reviews \[[@pone.0179429.ref023]--[@pone.0179429.ref026]\], and used 15 items, but the process of collapsing these 15 items to a six-item scale might have reduced the possibility of identifying differences in quality between studies. Eligible studies made comparisons with healthy peers in a wide variety of ways (control groups; comparison groups; reference to recommendations). Use of control groups was considered ideal, but restricting our synthesis to only those studies would have reduced a small evidence base to an even smaller evidence base, so this was not done. Further, the eligible studies varied substantially in terms of the accelerometers used, and even where the same accelerometer was used the methods varied in a number of potentially important accelerometer data reduction decisions e.g. the definition of a monitoring epoch \[[@pone.0179429.ref043], [@pone.0179429.ref059]\]; the number of hours and days of data constituting a valid data set \[[@pone.0179429.ref044], [@pone.0179429.ref046]\]; MVPA and ST accelerometer cut-points; and criteria for the inclusion or exclusion of non-wear time \[[@pone.0179429.ref036], [@pone.0179429.ref045]\]. These differences between studies are likely to have produced meaningful differences in MVPA and ST estimates \[[@pone.0179429.ref013]\] and they make it difficult to compare across studies. For the present review, the level of heterogeneity between eligible studies was high when combining data in meta-analysis across studies. However, in the case of all eligible studies the methods used for patient and control/comparison groups were identical, so comparison *within-studies* remain meaningful. An example of how the choice of accelerometer cut-point could affect conclusions reached by individual studies is that studies using lower accelerometer cut-points to define MVPA tended to report higher levels of habitual MVPA than those which used lower accelerometer cut-points to define MVPA. Tsai et al \[[@pone.0179429.ref039]\], for example used an Actigraph accelerometer cut-off of ≥ 700 counts per minute to define MVPA in children with asthma. This cut-off point is well below the cut-points used more commonly and which are more evidence based (based on calibration studies such as the Evenson et al cut-point of 2296 counts per minute \[[@pone.0179429.ref030]\]; or the Puyau et al cut-point of 3200 counts per minute \[[@pone.0179429.ref062]\]). The very low accelerometer cut-point used by Tsai et al almost certainly led to the very high estimate of 265 (SD 83) minutes of daily MVPA \[[@pone.0179429.ref039]\], and could lead to the erroneous conclusion that levels of MVPA among children with asthma are extremely high. Conclusions {#sec021} =========== In summary, this systematic review found that overall (habitual) MVPA levels are well below international recommendations in at least some groups of children and adolescents with chronic childhood diseases. The present review suggests that management of pediatric chronic conditions should place greater emphasis on MVPA, and patients with at least some chronic diseases are probably not currently benefiting from the health and non-health benefits that MVPA can bring. Time spent sedentary is often higher than in the comparison groups, and probably too high in many patients, but this is difficult to interpret in the absence of health-related recommendations for accelerometer measured ST in children and adolescents. This valuable information about the MVPA and ST levels in children with chronic disease may help to stimulate improving PA guidelines, and improving PA for these children. The need for more extensive research in this area, including intervention studies of the impact of increased MVPA levels on health related outcomes, is clear. Supporting information {#sec022} ====================== ###### PRISMA 2009 checklist preferred reporting items for systematic reviews and meta-analyses. (DOC) ###### Click here for additional data file. ###### Systematic review protocol. (DOCX) ###### Click here for additional data file. ###### Inclusion and exclusion criteria for selection of studies. MVPA: Moderate-to-Vigorous Intensity Physical Activity; PA: physical activity; ST: sedentary time. (DOCX) ###### Click here for additional data file. ###### Study quality assessment criteria, modified from Tooth *et al* \[[@pone.0179429.ref022]\]. MVPA: Moderate-to-Vigorous Intensity Physical Activity. (DOCX) ###### Click here for additional data file. ###### Methodological quality assessment of the included studies. \+ Indicates that a criterion was satisfied; − indicates that a criterion was not satisfied. 1, described of Sample recruitment?; 2, description of the sample.?; 3, Attrition of sample?; 4, Data collection and reduction?; 5, MVPA definition given?; 6, MVPA Results given?; \* Studies are listed based on diseases groups. (DOCX) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceptualization:** RE AM JYP JJR.**Data curation:** RE AM.**Formal analysis:** RE AM.**Investigation:** RE JJR.**Methodology:** RE AM JYP JJR.**Project administration:** RE AM JJR JYP.**Supervision:** JJR JYP.**Visualization:** RE AM JJR JYP.**Writing -- original draft:** RE.**Writing -- review & editing:** RE AM JJR JYP.

Introduction {#S0001} ============ In the mission intended for green and eco-friendly ways of synthesizing nanomaterials for the development of new products, plant biomasses have proven useful in the manufacture of sustainable nanoparticles (NPs),[@CIT0001],[@CIT0002] which are frequently used in biological approaches[@CIT0003] Due to their unique properties and prospective applications, bio-synthesis of silver nanoparticles (AgNPs) has received considerable attention in current centuries[@CIT0004] NPs with controllable shapes and sizes have usually been synthesized by means of chemical or physical methods[@CIT0005] However, usage of toxic chemicals and expensive physical procedures and tools[@CIT0006] has led to the emergence of environmentally friendly approaches. NPs synthesized through natural sources are the nontoxic, recyclable and most effective technique which is a harmless and nontoxic alternative[@CIT0007] As per to a study of collected works and literature, naturally existing organic foodstuff and agronomic unwanted resources have not yet been broadly studied for the synthesis of several kinds of NPs[@CIT0005] AgNPs amongst other metal NPs have intense application in the pharmaceutical fields. It has a unique thermal, electrical and optical properties which gives it great significance to use[@CIT0008] It has reported earlier that AgNPs retains excellent anticancer, antiviral, antimicrobial and anti-inflammatory potential with enhanced bio-degradability and less toxicity.[@CIT0008],[@CIT0009] Bio-synthesis of silver NPs is economic and environmentally safe approach[@CIT0010] *Ipomoea batatas* (Ib) or sweet potato is the sixth most significant food crop worldwide and is an extremely versatile and delicious vegetable.[@CIT0011]--[@CIT0013] *Ipomoea batatas* comprises high contents of various nutrients and phytochemicals that are beneficial for improving and maintaining human health.[@CIT0011],[@CIT0014] Currently Ib is well known as a valuable source of exclusive natural bioactive compounds and for its high nutritional value,[@CIT0015],[@CIT0016] counting few of which can be utilized in the development of medicines against various diseases and also in the making of manufacturing products[@CIT0012] lb plant also has abundant therapeutic significance as an imminent antidiabetic, anti-inflammatory and anticancer agent.[@CIT0012],[@CIT0014],[@CIT0017] The essential primary metabolites like organic acids, amino acids, sugars, sugar alcohols and health beneficial phytochemicals including phenolic acids, flavonoids, anthocyanins and carotenoids and hydrophilic primary metabolites were identified in colored flesh Ib such as white orange and purple in an earlier study[@CIT0017] As per the previous study, carotenoids like α-carotene, β-carotene, zeaxanthin and lutein were identified in white-fleshed Ib[@CIT0017] Furthermore, it was also reported that orange-fleshed Ib contains the highest level and white-fleshed Ib contains the lowest level of carotenoid[@CIT0017] Purple Ib was reported to have various phenolic compounds, and most of them were found to be anthocyanins[@CIT0018] Hence, Ib is recognized as a basis of unique and valuable active compounds that can be exploited in the improvement of natural drugs against several diseases[@CIT0012] It acts as a vital source in the production of sugar, alcohol and starch in the field of food manufacturing. The high cost for the right process of discarding the wastes generated from these processes is the foremost cause of loss in the field of food manufacturing[@CIT0011] Recently. public concerns about food waste and its economic impact on social life are improved to develop research strategies which can promote food waste utilization in an effective way[@CIT0019] Consequently, the discovery of new techniques to recycle agro-waste materials efficiently is essential. Therefore, it is of great curiosity to get benefit of this excellent vegetable peel to bio-synthesize AgNPs as a new therapeutic agent. The current study aimed at the green synthesis of AgNPs using the outer peel waste of two Korean sweet potato (Ib) varieties, Korean red skin sweet potato and Korean pumpkin sweet potato, which have are termed as Ib1 and Ib2, respectively. Following the bio-synthesis of the AgNPs from these two sweet potato varieties, their antidiabetic, cytotoxicity, antioxidant and antibacterial actions were compared. Materials and methods {#S0002} ===================== Preparation of extract of Ib waste peel {#S0002-S2001} --------------------------------------- Two varieties of Ib (sweet potato), Korean name bam-goguma (Ib1) and Korean name hobak-goguma (Ib2), were purchased from the market in Goyang, Republic of Korea ([Figure 1A](#F0001){ref-type="fig"} and [B](#F0001){ref-type="fig"}). The sweet potatoes were washed thoroughly with double-distilled water (DDH~2~O), dried and the unwanted outer nonedible parts were peeled off and cut into small portions. Aliquots (220 g) of the cut peels were retained in separate Erlenmeyer flasks (1000 mL) containing 550 mL DDH~2~O. The mixtures were boiled under constant stirring and then cooled to room temperature and stored at 4°C for further use.Figure 1(**A**) Sweet potato (*Ipomoea batatas*) varieties Korean red skin (Ib1) and Korean pumpkin (Ib2) vegetable and waste part. (**B**) Gradual color transition of Ib extract during synthesis of Ib1-AgNPs and Ib2-AgNPs between 0 and 24 hrs. (**C**) UV-Vis spectra of the biosynthesized Ib1-AgNPs and Ib2-AgNPs under laboratory condition. Biosynthesis of AgNPs using outer peel of Ib (Ib1 and Ib2) {#S0002-S2002} ---------------------------------------------------------- Ib1-AgNPs and Ib2-AgNPs were biosynthesized under laboratory conditions. Briefly, separate solutions of the Ib1 and Ib2 extracts were prepared by placing 100 mL of aqueous 1 mM AgNO~3~ in an Erlenmeyer flask (500 mL volume) and adding 10 mL of extract (Ib1 and Ib2, respectively) dropwise with continuous stirring at room temperature[@CIT0020] The green synthesis of Ib1-AgNPs and Ib2-AgNPs was monitored by the gradual change in the color of the reaction solutions at regular time intervals. After the final and complete synthesis, each reaction solution was centrifuged (at 10,000 rpm, 30 mins). The pellets were washed properly with DDH~2~O and again centrifuged. Finally, the pellets were dehydrated at 55°C and put in storage in a vial for future study. Characterization of biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0002-S2003} ---------------------------------------------------------- The Ib1-AgNPs and Ib2-AgNPs were characterized by means of UV-Vis spectroscopy, X-ray powder diffraction (XRD), field emission-scanning election microscopy (SEM), energy-dispersive X-ray spectroscopy (EDX) and Fourier-transform infrared (FT-IR) analyses using specific instruments, and standard procedures are as described in details in the previous publications.[@CIT0021],[@CIT0022] Antidiabetic action of bio-synthesized Ib1-AgNPs and Ib2-AgNPs {#S0002-S2004} -------------------------------------------------------------- The α-glucosidase enzyme from *Saccharomyces cerevisiae* (≥30 U/mg; Cat. No. G5003), and other chemicals were obtained from Sigma--Aldrich, St. Louis, MI, USA. The Ib1-AgNPs and Ib2-AgNPs samples were liquefied in methanol at 10 mg/mL with the assistance of a sonicator water bath. The α-glucosidase inhibition assay was accomplished by the standard procedure[@CIT0023] In total, 10 µg/mL of the test samples were aliquoted into 96-well plates and diluted with 0.02 M sodium phosphate buffer (pH 6.9), successively. To the final volume (50 µL), 50 µL of α-glucosidase (0.5 U/mL) was added. After keeping for 10 mins at room temperature, 50 µL of *p*-nitrophenyl- glucopyranoside (3.0 mM) was added as the substrate and the reaction solutions further incubated at 37°C for 20 mins. Next, 50 µL of Na~2~CO~3~ (0.1 M) was added, and the reaction was at stationary phase. Absorbance was documented through a plate reader at 405 nm wavelength. The plate also contained the positive standard (enzyme, buffer, and substrate). The inhibition percentage of α-glucosidase action was evaluated as follows: $$\documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document} $${\rm{\% Inhibition = }}\left[ {{{{\rm{Ab}}{{\rm{s}}_{{\rm{Control}}}}{\rm{ - Ab}}{{\rm{s}}_{{\rm{Test}}}}} \over {{\rm{Ab}}{{\rm{s}}_{{\rm{Control}}}}}}} \right]\,{\rm{{\rm X}}}\,{\rm{100}}$$ \end{document}$$ Antibacterial action of biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0002-S2005} -------------------------------------------------------------- The biosynthesized Ib1-AgNPs and Ib2-AgNPs were separately tested for their antibacterial effect against five pathogenic bacteria, including *E. faecium* DB01, *S. enteritisca* KCCM 11806, *L. monocytogenes* ATCC 19111, *B. cereus* KCTC 3624 and *S. aureus* ATCC 13565, by following the standard disc diffusion technique with trivial modification.[@CIT0024],[@CIT0025] Antioxidant potential of biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0002-S2006} --------------------------------------------------------------- The biosynthesized Ib1-AgNPs and Ib2-AgNPs were, respectively, examined for their antioxidant activity by the ABTS, DPPH and NOx radical scavenging analyses, and reducing power assay, as detailed by Patra et al[@CIT0005] Cytotoxicity activity of biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0002-S2007} --------------------------------------------------------------- The biosynthesized Ib1-AgNPs and Ib2-AgNPs were dissolved separately in Dulbecco's phosphate-buffered saline (Welgene, Gyeongsan-si, Gyeongsangbuk-do, Republic of Korea) at 1 mg/mL and sterilized through an 0.22 µm syringe filter (Millipore, Billerica, MA, USA). For treating HepG2 cells, the Ib1-AgNPs and Ib2-AgNPs were prepared separately at different concentrations (0.1, 0.01, 0.001, 0.0001, 0.00001 mg/mL) usingDMEM (Welgene) supplemented with 10% (v/v) fetal bovine serum and 1% penicillin--streptomycin (Gibco, Carlsbad, CA, USA)[@CIT0026] The HepG2 cells were acquired from a cell line bank (Seoul, Republic of Korea) and cultured in complete DMEM in a CO~2~ (5%) humidified incubator (maintained at 37°C). The harvested, completely grown cells were trypsinized with trypsin--EDTA (Gibco) and transferred to 96-well plates (100 µL/well, with 5×10^4^ cells per well). The HepG2 cells viability was evaluated by the trypan blue exclusion test, and the cells were kept in a humidified incubator at 37°C with 5% CO~2~ and 95% air for 24 hrs. The medium was removed after 24 h incubation, and the cells were inoculated to 0.001, 0.010, 0.100 and 1.000 mg/mL Ib1-AgNPs and Ib2-AgNPs, which were distributed in DMEM. Again, the cells were kept for incubation under 5% CO~2~ at 37°C (24 h-)[@CIT0026] The cell cytotoxicity action of bio-synthesized Ib1-AgNPs and Ib2-AgNPs was evaluated by means of an EZ-Cytox Kit (DoGenBio Co., Ltd., Seoul, Republic of Korea) according to the manufacturer's procedure. The morphology and viability of cells exposed to Ib1-AgNPs and Ib2-AgNPs were evaluated by the trypan blue exclusion test[@CIT0026] Statistical analysis {#S0002-S2008} -------------------- Data are represented by means of the mean ± standard deviation. One-way ANOVA was performed, followed by Duncan's test (at 5% level of significance) using SPSS version 23.0 software (IBM Corp., Armonk, NY, USA). Results {#S0003} ======= Biosynthesis of AgNPs using Ib1 and Ib2 outer peel extract {#S0003-S2001} ---------------------------------------------------------- In this study, Ib1-AgNPs and Ib2-AgNPs were synthesized under normal laboratory light using the outer peel extracts (which is typically a food waste) of two diverse varieties of Ib ([Figure 1A](#F0001){ref-type="fig"}). Biosynthesis of Ib1-AgNPs ([Figure 1B](#F0001){ref-type="fig"}) and Ib2-AgNPs ([Figure 1B](#F0001){ref-type="fig"}) progressed under the laboratory condition, which was visualized by the gradual change in the color of the reaction solution from without color to reddish-brown. The Ib2-AgNPs synthesis was rapid, starting within an hour, whereas the Ib1-AgNPs synthesis was slow, beginning slightly at 6 hrs and becoming rapid after 12 h of reaction ([Figure 1B](#F0001){ref-type="fig"}). Characterization of biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0003-S2002} ---------------------------------------------------------- Once the bio-synthesis of Ib1-AgNPs and Ib2-AgNPs was visually confirmed by the gradual color change of the reaction mixtures, the samples were categorized through UV-Vis spectroscopy, FT-IR spectroscopy, SEM, EDX and XRD. The UV-Vis spectra of both Ib1 and Ib2 reaction mixtures were recorded at altered time intervals for up to 24 hrs. The highest absorbance peaks of the solution mixtures occurred at 452 and 442 nm for Ib1 and Ib2, respectively ([Figure 1C](#F0001){ref-type="fig"}). The FT-IR results of extracts (Ib1 and Ib2) and their corresponding AgNPs are presented in [Figure 2](#F0002){ref-type="fig"}. The absorption peaks appeared at 3321.40, 2118.56, 1634.03, 1089.17 and 685.71 cm^−1^ for the Ib1 extract and at 3285.56, 2924.08, 1620.66, 1362.15, 1017.54 and 571.70 cm^−1^ meant for the Ib1-AgNPs ([Figure 2A](#F0002){ref-type="fig"} and [B](#F0002){ref-type="fig"}). For Ib2 extract, the absorption peaks occurred at 3315.75, 2358.00, 1862.16, 1634.03, 1087.29 and 680.06 cm^−1^, and the corresponding shifts for the Ib2-AgNPs were observed at 3275.54, 2924.71, 1630.42, 1364.69, 1018.51, 925.01 and 519.89 cm^−1^ ([Figure 2C](#F0002){ref-type="fig"} and [D](#F0002){ref-type="fig"}).Figure 2FT-IR spectra of Ib1-AgNPs (**A, B**) and Ib2-AgNPs (**C, D**). Considering the UV-Vis spectra and FT-IR results, the Ib1-AgNPs and Ib2-AgNPs were confirmed to be extremely stable and further analyzed by SEM--EDX for their morphology and chemical composition. The external morphology of the Ib1-AgNPs and Ib2-AgNPs was agglomerated in nature ([Figure 3A](#F0003){ref-type="fig"} and [B](#F0003){ref-type="fig"}). For both the Ib1-AgNPs and Ib2-AgNPs, the EDX results ([Figure 3C](#F0003){ref-type="fig"} and [D](#F0003){ref-type="fig"}) displayed a high peak at 3 keV, conforming to the Ag region, thereby identifying the existence of AgNPs in both Ib1 and Ib2. In Ib1-AgNPs, the Ag element accounted for 22.45% of the total composition and the corresponding content in the Ib2-AgNPs was 34.92% ([Figure 3C](#F0003){ref-type="fig"} and [D](#F0003){ref-type="fig"}). The spectral signals for other elements, like oxygen and chlorine, were also noted in the AgNPs, amounting to 7.60% and 11.37% in the Ib1-AgNPs, and 13.66% and 10.40% in the Ib2-AgNPs, respectively ([Figure 3C](#F0003){ref-type="fig"} and [D](#F0003){ref-type="fig"}).Figure 3SEM images (**A, B**) and EDX data (**C, D**) of Ib1-AgNPs and Ib2-AgNPs. The XRD pattern of the biosynthesized Ib1-AgNPs and Ib2-AgNPs is shown in [Figure 4](#F0004){ref-type="fig"}. Three well-resolved diffraction peaks were apparent in Ib1-AgNPs, with peaks (2θ angles) at 38.43°, 46.23° and 76.91°, equivalent to (111), (200) and (311), respectively, along with a few unknown peaks at 32.18°, 54.92° and 57.60°. For the Ib2-AgNPs, there were four well-resolved diffraction peaks (2θ angles) at 38.22°, 46.23°, 64.69° and 76.99°, corresponding to (111), (200), (220) and (311), respectively, besides some unknown signals at 32.29° and 43.27° ([Figure 4](#F0004){ref-type="fig"}).Figure 4XRD patterns of the biosynthesized Ib1-AgNPs and Ib2-AgNPs. Potential of the biosynthesized Ib1-AgNPs and Ib2-AgNPs {#S0003-S2003} ------------------------------------------------------- Both of the biosynthesized Ib1-AgNPs and Ib2-AgNPs were evaluated for their possible antidiabetic, antibacterial, cytotoxic and antioxidant activities. The α-glucosidase inhibition effects of both AgNPs indicated their promising antidiabetic potential ([Figure 5](#F0005){ref-type="fig"}). The Ib1-AgNPs were highly effective and had the highest α-glucosidase inhibition percentage, displaying 3.2%, 70.10% and 97.73% inhibition, respectively, at the three concentrations tested (0.25, 0.50 and 1.00 µg/mL). In comparison, the Ib2-AgNPs showed zero (at 0.25 µg/mL), 7.4% (at 0.5 µg/mL) and 64.81% inhibition (at 1.00 µg/mL). Further, the IC~50~ values were also calculated and presented in [Table 1](#T0001){ref-type="table"}. The IC~50~ values of Ib1-AgNPs and Ib2-AgNPs were found out to be 0.36 and 0.77 µg/mL, respectively.Table 1IC~50~ values of antioxidant and antidiabetic assaysParametersIC~50~ value (µg/mL) Ib1IC~50~ value (µg/mL) Ib2Antioxidant assayDPPH120.14107.46ABTS327.41323.12NO~X~389.54291.96Reducing (IC~0.5~ value)430.09363.80Antidiabetic assayα-glucosidase0.360.77 Figure 5α-Glucosidase activity of Ib1-AgNPs and Ib2-AgNPs. The cytotoxicity of the Ib1-AgNPs and Ib2-AgNPs in HepG~2~ cancer cells (after 24 h of exposure) ([Figure 6A](#F0006){ref-type="fig"}) indicated that the cell-line viability increased with decreasing concentration in the reaction medium of the Ib1-AgNPs and Ib2-AgNPs. Likewise, while detected under an inverted microscope, it was visualized that the control cells presented a higher number of well-attached live cells than the AgNPs-treated HepG2 cells ([Figure 6B](#F0006){ref-type="fig"}). Both AgNPs were extremely toxic to HepG2 cells at a higher concentration ([Figure 6C](#F0006){ref-type="fig"} and [D](#F0006){ref-type="fig"}). Significant variation in morphology due to cell death, along with limited scattering patterns and a higher number of dead cells (black arrow) were observed at the upper concentrations of the AgNPs ([Figure 6C](#F0006){ref-type="fig"} and [D](#F0006){ref-type="fig"}), whereas, consistent morphology and well-attached cells (white arrow) were seen at a lower concentration of the AgNPs ([Figure 6C](#F0006){ref-type="fig"} and [D](#F0006){ref-type="fig"}).Figure 6Cytotoxicity of Ib1-AgNPs and Ib2-AgNPs in HepG2 cancer cells. (**A**) Control. (**B**) HepG2 cell viability treated with AgNPs for 24 h. (**C, D**) Black arrow indicates dead cells, and white arrow indicates live cells (**C, D**). The antioxidant prospective of the bio-synthesized Ib1-AgNPs and Ib2-AgNPs was examined by the 2,2ʹ-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS) and 1,1-diphenyl-2-picrylhydrazyl (DPPH) scavenging analyses, nitrite/nitrate oxide (NOx) assay and reducing power test, all at three concentrations (25, 50 and 100 µg/mL), and the end results are shown in [Figure 7](#F0007){ref-type="fig"}. The ABTS scavenging activity of Ib1-AgNPs was 3.98--8.83% and 4.18--12.28% in the case of Ib2-AgNPs. These values were lower than the synthetic antioxidant tertiary butyl-hydroquinone (BHQ; control), which were in between the range of 17.06--73.79% at the same concentrations ([Figure 7A](#F0007){ref-type="fig"}). The DPPH scavenging potentials were 26.30--41.62% (Ib1-AgNPs), 29.92--46.53% (Ib2-AgNPs) and 82.61--85.68% (BHQ) ([Figure 7B](#F0007){ref-type="fig"}). The reducing powers of BHQ, Ib1-AgNPs and Ib2-AgNPs were 0.3573--0.6210, 0.0594--0.0715 and 0.0598--0.0852, respectively ([Figure 7C](#F0007){ref-type="fig"}). The NOx results for the Ib1-AgNPs and Ib2-AgNPs were 4.3--10.16 and 5.53--12.94%, respectively, and 17.06--73.79 for the synthetic antioxidant BHQ at the same three concentrations ([Figure 7D](#F0007){ref-type="fig"}). Further, the IC~50~ values of all the antioxidant assays were calculated and presented in [Table 1](#T0001){ref-type="table"}.Figure 7Antioxidant potential of the biosynthesized Ib1-AgNPs and Ib2-AgNPs. (**A**) ABTS radical scavenging activity. (**B**) DPPH-free radical scavenging activity. (**C**) Reducing power assay. (**D**) NOx scavenging activity. Difference in the superscript letters indicate statistical significance at *P*\<0.05. The antibacterial effect of Ib1-AgNPs and Ib2-AgNPs was tested against five pathogenic foodborne bacteria (*Staphylococcus aureus* ATCC 13565, *Salmonella enteritica* KCCM 11806, *Bacillus cereus* KCTC 3624, *Listeria monocytogenes* ATCC 19111 and *Enterococcus faecium* DB01). For Ib1-AgNPs, the inhibition zone was in between the range of 8.67--11.23 mm against the four pathogenic bacteria tested and the corresponding range for the Ib2-AgNPs was 8.74--11.52 mm. Ib2-AgNPs exerted an inhibitory action against all the five tested pathogenic bacteria while Ib1-AgNPs was inactive against *S. aureus* ATCC 13565 ([Table 2](#T0002){ref-type="table"}).Table 2Antibacterial potential of the biosynthesized Ib1 & Ib2-AgNPs against pathogenic bacteriaPathogenic bacteriaIb1-AgNPs\*Ib1-AgNPs\**Enterococcus feacium* DB0109.75^c^±0.0.01\*\*09.58^c^±0.0.53*Salmonella enteritica* KCCM 1180610.56^b^±0.3411.52^a^±0.0.12*Listeria monocytogenes* ATCC 1911111.23^a^±0.2910.43^b^±0.0.37*Bacillus cereus* KCTC 362410.56^b^±0.1808.74^d^±0.0.20*Staphylococcus aureus* ATCC 1356500.00^e^±0.0009.32^c^±0.0.47[^1] Discussion {#S0004} ========== Currently, to formulate nanoparticles of specific size and shape, the green methods are used to advance their features.[@CIT0010],[@CIT0027],[@CIT0028] During the past few years, rules and regulations for solid waste supervision, and environmental concerns surrounding their disposal have been growing day by day. Thus, the emergence of novel uses of vegetable and fruit wastes, such as their unwanted outer peels, in nano-technological applications is rising.[@CIT0029]--[@CIT0031] Nanotechnology is rapidly growing and utilized in a broad array of applications in the agriculture, biomedical (drug delivery, tissue scaffolds, wound dressing) and food industries, among others.[@CIT0010],[@CIT0031]--[@CIT0033] In the arena of nanotechnology, AgNPs have shown impressive performance and are utmost studied among several notable inorganic metal nanoparticles.[@CIT0010],[@CIT0032],[@CIT0034] AgNPs display high antimicrobial efficacy and are nontoxic in nature. To avoid any destructive environmental impact, the green synthesis approaches of nanoparticles are best, as of the absence of any toxic chemicals during the sample preparation and synthesis processes[@CIT0035] Vegetable and fruit peels are generally waste food products and generated in large quantities by the food and agriculture sectors. Polyphenols is a large class amongst the natural metabolites. It is found in a wide variety of vegetables, fruits and various foods[@CIT0018] Different vegetable and fruit peels comprise a plethora of diverse bioactive compounds, like phenolic compounds, carotenoids, dietary fibers, enzymes, vitamins, flavonoids, natural pigments and antioxidant essential oils. Phytochemicals might be utilized in the health industry for medications and medicines and also in the food sector for the development of functional foods.[@CIT0036]--[@CIT0038] On the approach to sustainable improvement, utilization of food waste products for the manufacture of various vital bioactive elements is a vital step. Ib L. Lam. correspond to the family Convolvulaceae. It is enriched with various nutrients, like vitamins, minerals and carbohydrates. It is rich in chlorogenic acid, vitamin C, rutin, quercetin and caffeic acid[@CIT0039] Besides, it is also reported to be utilized as an anti-asthmatic, antidiabetic, antitumor, anti-inflammatory and anti-microbial agent.[@CIT0014],[@CIT0040],[@CIT0041] A number of essential primary metabolites such as organic acids, amino acids, sugars, sugar alcohols and health beneficial phytochemicals including phenolic acids, flavonoids, anthocyanins and carotenoids and hydrophilic primary metabolites were identified in various Ib varieties.[@CIT0013],[@CIT0017],[@CIT0042] Due to its health benefits, *I*Ib is considered a functional food.[@CIT0014] Subsequently, the outer peels of Ib might be expected to have these same compounds, and these could be credited to the bioreduction and capping of Ib1-AgNPs and Ib2-AgNPs. Hence, in the current research, the outer peel of two varieties of Korean Ib , which are termed as Ib1 and Ib2 used, for green synthesis of Ib1-AgNPs and Ib2-AgNPs, respectively ([Figure 1A](#F0001){ref-type="fig"}). The alteration in color of the reaction-solution evidenced the establishment of Ib1-AgNPs and Ib2-AgNPs ([Figure 1B](#F0001){ref-type="fig"})[@CIT0043] The Ibplant originates from Central America and southern Mexico[@CIT0044] It is broadly recognized that each plant parts, like the stems, roots, leaves and tubers, comprise essential metabolites, like coumarins, batatins and triterpenes, with distinct functional actions.[@CIT0045]--[@CIT0047] Indeed, Ib is a tuber vegetable. The peels of Ib are enriched with phenolic compounds, which are associated with almost all health benefits, due to their scavenging (free-radical) prospective.[@CIT0013],[@CIT0048],[@CIT0049] The Ib proteins also possess various nutraceutical potentials. The currently named Caiapo Potato Powder is an extract from a sweet potato cultivar promoted as an antidiabetic supplement.[@CIT0050]--[@CIT0052] The sweet potato peels are abundant in storage proteins, identified as sporamins, which are resistant to simulated gastric digestion and present promising antioxidant and antiproliferative properties.[@CIT0053]--[@CIT0056] Sporamins function as trypsin inhibitors, also after incubation with pepsin or trypsin. After biosynthesizing the Ib1-AgNPs and Ib2-AgNPs, the reaction kinetics were monitored by UV-Vis absorption spectroscopy between the wavelength ranges of 340--640 (Ib1) and 280--680 nm (Ib2). Generally, as of the free electrons, AgNPs may display a surface plasmon resonance band at 440--558 nm[@CIT0057] In our present study, the Ib1-AgNPs and Ib2-AgNPs surface plasmon resonance values were detected at around 458 and 446 nm, respectively ([Figure 1C)](#F0001){ref-type="fig"}. Similar results have been previously reported for synthesized AgNPs.[@CIT0008],[@CIT0057]--[@CIT0059] This result suggests that phytochemicals existing in Ib[@CIT0017] peels work as reducing and capping agents. FT-IR allows the analysis of samples up to \~11 mm in diameter[@CIT0060] In the current study, FT-IR analysis was accomplished to identify the active bio-molecules in the Ib1 and Ib2 peel extracts accountable for reducing and capping the bio-reduced AgNPs. Changes in the functional groups might suggest their involvement in the green synthesis of Ib1-AgNPs and Ib2-AgNPS. The FT-IR data ([Figure 2](#F0002){ref-type="fig"}) evidenced a variation in peaks with different stretching modes between the Ib1 and Ib2 extracts ([Figure 2A](#F0002){ref-type="fig"} and [B)](#F0002){ref-type="fig"} in comparison with their corresponding AgNPs ([Figure 2C](#F0002){ref-type="fig"} and [D](#F0002){ref-type="fig"}). According to the FT-IR analysis, it is probable that the peaks at 3321.40, 1634.09 and 1089.17 cm^−1^ in Ib1 extract shifted to 3285.56, 1620.66 and 1017.54 cm^−1^, respectively, in the Ib1-AgNPs. Similarly, the peaks at 3315.75, 1634.03 and 1087.29 cm^−1^ in Ib2 extract likely shifted to 3275.54, 1630.42 and 1018.51 cm^−1^ in Ib2-AgNPs. The peaks or points at 3285.56 (Ib1) and 3275.54 cm^−1^ (Ib2) indicate the presence of O--H stretch, H--bonded bonds, which belong to the alcohols, phenols functional group.[@CIT0061],[@CIT0062] The peaks at 1620.66 (Ib1) and 1630.42 cm^−1^ (Ib2) specify the presence of N-H bend and it belongs to the functional group 1° amines. Similarly, the peaks at 1017.54 cm^−1^ (Ib1-AgNPs) and 1018.51 cm^−1^ (Ib2-AgNPs) specify the presence of C--O stretch which belongs to the alcohols, carboxylic acids, esters, ethers functional groups[@CIT0061],[@CIT0062] . The minor change in the bands of Ib1-AgNPs and Ib2-AgNPs could be credited to the decline, stabilization and capping procedures in the course of synthesizing the AgNPs[@CIT0043] Visualization of the AgNPs by SEM revealed the spherical form of the bio-synthesized Ib1-AgNPs and Ib2-AgNPs in the nanometer range. The biosynthesized nanoparticles were equally agglomerated ([Figure 3A](#F0003){ref-type="fig"} and [B](#F0003){ref-type="fig"}), which may be correlated to the removal of solvent during the synthesis process.[@CIT0005],[@CIT0043] The elemental composition identified by EDX verified the existence of oxygen and chlorine in both Ib1-AgNPs and Ib2-AgNPs. This finding might be accredited to the sweet potato outer peels utilized in the green-synthesis process, in which the proteins and flavonoids existing in the sweet potato peel could have participated in the capping of Ib1-AgNPs and Ib2-AgNPs[@CIT0043] The high percentage of Ag noted in the EDX end results evidenced that the particles were mostly AgNPs ([Figure 3C](#F0003){ref-type="fig"} and [D](#F0003){ref-type="fig"}). The XRD analysis exposed three distinct peaks in Ib1-AgNPs and four distinct peaks in Ib2-AgNPs. [Figure 4](#F0004){ref-type="fig"} reveals that the peaks were equivalent to the face-centered cubic (fcc) phase of Ag0 standard (JCPDS Card No. 04-0783).[@CIT0043],[@CIT0063],[@CIT0064] Identical values were found in prior studies.[@CIT0043],[@CIT0065] The XRD data also affirmed the crystalline type of the bio-synthesized Ib1-AgNPs and Ib2-AgNPs. Besides, some unknown peaks were noticed ([Figure 4](#F0004){ref-type="fig"}), which suggest that the crystallization of the bioorganic phase could have happened on the peripheral of the biosynthesized Ib1-AgNPs and Ib2-AgNPs, as mentioned in an earlier article[@CIT0066] After their biosynthesis, the antidiabetic, cytotoxic, antioxidant and antibacterial actions of the Ib1-AgNPs and Ib2-AgNPs were accomplished. To therapeutically decrease hyperglycemia, the carbohydrate-digesting enzymes, such as α-glucosidase and α-amylase, need to be constrained or inhibited, thereby inhibiting the breakdown of carbohydrates into monosaccharides, which are the main contributors to high blood sugar levels.[@CIT0067],[@CIT0068] Thus, to manage diabetes, emerging compounds with an inhibitory potential toward carbohydrate-hydrolyzing enzymes might be advantageous. In the current study, the Ib1-AgNPs and Ib2-AgNPs both displayed promising antidiabetic activity (in a dose-dependent manner). A comparatively higher antidiabetic potential was displayed by the Ib1-AgNPs than Ib2-AgNPs in the concentration range of 0.25--1.00 µg/mL ([Figure 5](#F0005){ref-type="fig"}). More than 97% inhibition of α-glucosidase by the Ib1-AgNPs at 1.00 µg/mL was observed while the Ib2-AgNPs exhibited nearly 65% inhibition of α-glucosidase at the same concentration ([Figure 5](#F0005){ref-type="fig"}). Analogous results for AgNPs against α-glucosidase have been stated in the research literature .[@CIT0069]--[@CIT0071] The significant potential of both Ib1-AgNPs and Ib2-AgNPs seen in the current research as evident from its significantly less IC~50~ values ([Table 1](#T0001){ref-type="table"}) is an inspiring sign of their effectiveness in the formulation of active drugs for the management of diabetes. Cancer is an unusual disorder leading to the wild cell division and damages the body tissues. The structural features of NPs make them outstanding approach for targeting the unusual cell growth instigated by cancer[@CIT0034] Nano-based medicines are highly effective for the diagnosis and treatment of cancer and related diseases[@CIT0010] When the anticancer activity was studied, both the Ib1-AgNPs and Ib2-AgNPs showed high cytotoxicity potential against the HepG2 cell line ([Figure 6](#F0006){ref-type="fig"}). In the validation of cytotoxic potential, it was noticed that both AgNPs displayed a high level of cytotoxic potential against the cancer cells, but the AgNPs of Ib2 were more active than the Ib1-AgNPs at low concentrations (1, 10 and 100 µg/mL). Moreover, it was detected that as the concentration of AgNPs increased, the anticancer potential increased ([Figure 6](#F0006){ref-type="fig"}). The current result is similar with a previously reported result.[@CIT0010],[@CIT0022],[@CIT0072] The cytotoxicity potential of Ib1-AgNPs and Ib2-AgNPs can be credited to the initiation of intracellular oxidative stress.[@CIT0022],[@CIT0026] According to the earlier studies, the cytotoxicity activity of Ib1-AgNPs and Ib2-AgNPs might be the end result of oxidative stress induced in the cells rather than the toxic effect of Ag^+^ ions.[@CIT0022],[@CIT0026],[@CIT0073]--[@CIT0079] The bio-synthesized Ib1-AgNPs and Ib2-AgNPs exhibited moderate ABTS, DPPH and NOx scavenging activities ([Figure 7](#F0007){ref-type="fig"}, [Table 1](#T0001){ref-type="table"}). These results might arise from the interference of several functional groups existing in the Ib1 and Ib2 extracts, which might have played a vital part in the capping and stabilization procedure during Ib1-AgNPs and Ib2-AgNPs synthesis.[@CIT0022],[@CIT0072],[@CIT0080] The Ib1-AgNPs and Ib2-AgNPs showed higher scavenging effect against DPPH-free radicals in comparison to ABTS radical cations, as evident from less IC~50~ values ([Table 1](#T0001){ref-type="table"}), which is consistent with previous reports.[@CIT0022],[@CIT0072],[@CIT0081] However, Ib2-AgNPs demonstrated slightly more potent antioxidant potential than the Ib1-AgNPs in entire four assays ([Table 1](#T0001){ref-type="table"}), which may be related with the combination of additional oxidants on the external surface area of Ib2-AgNPs.[@CIT0022],[@CIT0043],[@CIT0072] The reducing power of Ib1-AgNPs and Ib2-AgNPs may possibly be due to the presence of functional groups on the external surface of the particles ([Figure 7C](#F0007){ref-type="fig"}). NOx is an essential bioregulatory molecule in the immune, nervous and cardiovascular systems.[@CIT0022],[@CIT0071],[@CIT0072] In the present study, the biosynthesized Ib1-AgNPs and Ib2-AgNPs displayed a concentration-dependent radical scavenging potential ([Figure 7C](#F0007){ref-type="fig"}). The relationships between AgNPs and NOx under anhydrous, anaerobic conditions and at room temperature help acceptance of electron from AgNPs[@CIT0071] However, the Ib1-AgNPs and Ib2-AgNPs showed moderate NOx scavenging activity ([Figure 7D](#F0007){ref-type="fig"}), which might be due to the difference in the reaction mechanism. Currently, the growth rate of multidrug-resistant pathogens is widespread, which may have an antagonistic effect on human well-being.[@CIT0082],[@CIT0083] Here, the Ib1-AgNPs and Ib2-AgNPs exhibited positive antibacterial activity against four pathogenic Gram-positive bacteria and one pathogenic Gram-negative bacteria ([Table 2](#T0002){ref-type="table"}), corroborating previous reports.[@CIT0009],[@CIT0071],[@CIT0079] The positive results are credited to the lesser size of the AgNPs, which may facilitate their entry into the bacterial cell walls, in turn, causing damage to bacterial proteins that eventually results in the cell demise of the bacteria[@CIT0084] The inhibition zone might be due to the better diffusion of nanoparticles by diffusion of Ag^+^ ions in the nutrient media[@CIT0010] Conclusion {#S0005} ========== The Ib1-AgNPs and Ib2-AgNPs were biosynthesized successfully using the outer peels of two Ib varieties. Both the Ib-AgNPs displayed significant antidiabetic and cytotoxicity activities. Similarly, both the AgNPs exhibited moderate free-radical scavenging and antibacterial activity. Considering all the biological activities tested, the Ib2-AgNPs were considered as the most effective AgNPs, which might be credited to the greater richness of bioactive compounds present in Ib2 peel extract. Based on the findings, both the biosynthesized AgNPs could be utilized as promising candidates in various fields such as in the treatment of several diseases like diabetes, cancer, antibacterial wound dressing and for food preservation. The current supporting results might offer a harmless begin for utilization of eco-friendly silver NPs in the pharmaceutical industries. G Das, JK Patra, and HS Shin are grateful to Dongguk University-Seoul, Republic of Korea, for their support. This work was supported by Korea Institute of Planning and Evaluation for Technology in Food, Agriculture, Forestry and Fisheries (IPET) through High Value-added Food Technology Development Program funded by Ministry of Agriculture, Food and Rural Affairs (MAFRA)(118056-2) and funded by Korea Environmental Industry & Technology Institute (A117-00197-0703-0). The authors also wish to thank Prof. Hojun Kim and Dr. AbuZar Ansari of the Department of Rehabilitation Medicine of Korean Medicine, Dongguk University, Goyang, Republic of Korea, for their help in cytotoxicity analysis. CN Vishnuprasad is grateful for the support provided by the TDU, Bengaluru, India. The authors also acknowledge the technical help received from Ms. Ashwini Elango, a Research Fellow from TDU, Bengaluru, India. Ethics approval and informed consent {#S0006} ==================================== Not applicable. Consent for publication {#S0007} ======================= Not applicable. Data availability {#S0008} ================= All data generated or analyzed during this study are included in this manuscript. Abbreviation list {#S0009} ================= Ib1, Korean red skin *Ipomoea batatas*; Ib2, Korean pumpkin *Ipomoea batatas*; NPs, Nano-particles. Disclosure {#S0010} ========== The authors declare there are no competing interests in this work. [^1]: **Notes:** \*Inhibition zone in mm; \*\*values with different superscript are significant at *P*\<0.05.

Introduction {#s1} ============ Ectopic ovaries are rare embryologic abnormalities with an estimated prevalence between 1:29,000 and 1:93,000 gynecologic admissions ([@B1], [@B2]). Because patients are asymptomatic, it is difficult to diagnose ectopic ovaries ([@B3]). Gastrointestinal stromal tumors (GISTs) are rare tumors which can arise anywhere within the GI tract ([@B4]). Herein, we report the first case of a patient with an ectopic ovary presenting as a GIST and provide the differential diagnosis for GIST. Case presentation {#s2} ================= A 33-year-old female sought evaluation in our Department of General Surgery with a 2-year history of sporadic abdominal pain that had become aggravated during the past week. The character of pain became sharp and frequent. The pain was localized to the left lower abdomen. There was no nausea and vomiting. There was no history of abdominal trauma. The patient had a congenital anomaly of the kidneys and uterus; there was no menstruation. The patient had undergone an appendectomy in the past. On physical examination, the patient was afebrile. The abdominal examination revealed pain and a mass in the left lower quadrant area upon palpation. The mass was approximately 4 × 5 cm in diameters and was not circumscribed. The patient had no rebound tenderness and muscle rigidity. Laboratory testing revealed the following: white blood cell count, 7.13 × 10^9^/L; neutrophilic granulocytes, 76.8%; hemoglobin, 120 g/L; and platelet count, 322 × 10^9^/L. Computed tomography (CT) revealed an intestinal stromal tumor ([Figure 1](#F1){ref-type="fig"}) and pelvic kidneys ([Figure 2](#F2){ref-type="fig"}). Digestive tract radiography showed possible extraintestinal involvement ([Figure 3](#F3){ref-type="fig"}). An intestinal stromal tumor was diagnosed and an abdominal laparotomy was performed; however, the intestinal tract was normal and a mass was noted in the sigmoid flexure. The tumor exhibited exophytic growth without infiltration and was 6.0 × 5.0 × 3.0 cm in size. The tumor and colon (proximal and distal length, 10 cm; \~25 cm) were excised. A rapid frozen section pathologic examination revealed a solitary fibrous tumor (SFT). A colon anastomosis was performed and the patient had fully recovered 7-days post-operatively. The final diagnosis was an ectopic ovary with corpora lutea bleeding ([Figure 4](#F4){ref-type="fig"}). The patient recovered well after surgery and there were no post-operative complications. The patient was doing well at the 11-month follow-up visit. Written informed consent was obtained from the patient and The Third People\'s Hospital of Dalian had approved the study (NO. 2018-LW-001). {#F1} {#F2} {#F3} {#F4} Discussion {#s3} ========== The patient presented to the hospital for evaluation of aggravated abdominal pain, and the CT scan revealed an intestinal stromal tumor. Intra-operatively, a mass located in the colon was thought to be a colon stromal tumor; however, the final pathologic diagnosis was an ectopic ovary with corpora lutea bleeding. Ectopic ovaries can be classified as congenital and acquired ([@B5]). The present case belongs to the congenital type. A developmental error occurring during the formation of genital canals and external genitalia in women may induce ectopic ovaries ([@B6], [@B7]). The methods by which ectopic ovaries are diagnosed include MRI and surgery; however, surgery is the gold standard ([@B3], [@B8]). MRI can be used to diagnose genital tract and renal system abnormalities ([@B8]). Controlled ovarian stimulation (COH) is thought to aid in the diagnosis of ectopic ovaries; magnetic resonance imaging (MRI) more accurately identifies undescended ovaries in the upper abdomen after COH ([@B9], [@B10]). In the present study, because the CT scan revealed an intestinal stromal tumor, an MRI was not performed. Ectopic ovaries are usually accompanied by maldevelopment of the genital system and renal tract ([@B11]). The present case had similar maldevelopments: congenital abnormal development of the ovaries and ectopic kidneys. Ectopic ovaries may lead to menstrual disorders, infertility, or abdominal pain ([@B3]). In the present case, because of uterine dysfunction and amenorrhea, an ectopic ovary was not suspected. Ectopic ovaries can be found in the upper abdomen, near the pelvic brim or neighboring inguinal canal. The location of the ovary in the current case was the colon, which is the first such reported case. GISTs are gastrointestinal mesenchymal tumors accounting for 0.2% of all gastrointestinal tumors ([@B12]). GISTs can originate anywhere in the gastrointestinal tract. Therefore, the present case was initially suspected to be a GIST. Ectopic ovaries can present as primary infertility ([@B13]), a hernia or cyst in the inguinal canal ([@B14]), acute appendicitis ([@B3]), ovarian malignancy ([@B15]), a Brenner tumor ([@B16]), a Wilms\' tumor ([@B17]), as well as a GIST. Ectopic ovaries can cause irregular menses and pain ([@B18]), and are often accompanied by an abnormal urinary system ([@B11]) or a mature teratoma ([@B19]). Patients with developmental anomalies need close attention. An abnormal urinary system is usually accompanied by an abnormal genital system. Although it is difficult for the diagnosis of ectopic ovaries pre-operatively, additional examinations should be performed, such as ultrasonography, MRI, and endoscopy. A multiple disciplinary team (MDT) is also advised. Conclusion {#s4} ========== The present case is an ectopic ovary mimicking a GIST. Maldevelopment of the genital tract can lead to an ectopic ovary and surgery is a good management choice. We have shared our clinical experience to help guide the management of similar cases and offer a differential diagnosis of GISTs. Data Availability {#s5} ================= All datasets generated for this study are included in the manuscript and/or the supplementary files. Consent {#s6} ======= Written informed consent was obtained from the patient for publication of this case report and the accompanying images. Author Contributions {#s7} ==================== JP and SW: conceptualization. HW: data curation. ZF: investigation, validation, and writing of the original draft. 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. We thank International Science Editing (<http://www.internationalscienceediting.com>) for editing this manuscript. **Funding.** This study received financial support from the National Natural Science Foundation of China (NO. 81701965), Natural Science Foundation of Liaoning Province (NO. 20180550116), and Dalian Medical Science Research Project (NO. 1711038). [^1]: Edited by: Zongbing You, Tulane University, United States [^2]: Reviewed by: Qingli Li, Sichuan University, China; Lunxu Liu, Sichuan University, China [^3]: This article was submitted to Surgical Oncology, a section of the journal Frontiers in Oncology [^4]: †These authors have contributed equally to this work

Introduction {#s1} ============ Onchocerciasis, commonly referred to as "river blindness" is classified by the World Health Organization (WHO) as a neglected tropical disease, afflicting approximately 37 million people in Africa, Central and South America and Yemen, with 89 million more at risk [@pntd.0000834-Joint1]. Symptoms of the disease include acute dermatitis and blindness, the result of which is the loss of 1 million disability-adjusted life years (DALYs) annually [@pntd.0000834-World1]. The causative agent, the filarial nematode *Onchocerca volvulus*, is transmitted in its larval stage between human hosts through the bite of a *Simulium* (sp.) black fly. Once these parasites have matured into the adult form, they can live for approximately 14 years in subcutaneous nodules within a human host [@pntd.0000834-Plaisier1]. The drug ivermectin (Mectizan) has served as the principal means of onchocerciasis control [@pntd.0000834-Richards1], however, after initially reducing the number of microfilariae, within a year, the microfilariae return to levels of 20% or higher than that prior to treatment [@pntd.0000834-Awadzi1]. The combination of the lack of effect of annual ivermectin treatment on adult worm survival and the fecundity of adult females, along with significant fly and human migration patterns has helped to perpetuate the disease. In Africa, where onchocerciasis control programs have been in place since the founding of the Onchocerciasis Control Programme in West Africa (OCP, 1974--2002) and are currently being conducted by the African Programme for Onchocerciasis Control (APOC, 1995-present), diagnosis is an essential aspect of the determination of treatment and distribution of medication. In the Western hemisphere, accurate and robust diagnostics are essential for attaining the goal of disease elimination. Twice yearly dosage of ivermectin, through the efforts of the Onchocerciasis Elimination Program for the Americas (OEPA, 1992-present), has lead to a minimization of infection to 13 foci within six countries in Central and South America. Although mass treatment of onchocerciasis foci in the Western hemisphere is slated to be suspended in 2012 [@pntd.0000834-Pan1], achieving the goal of elimination is contingent upon continued surveillance of the disease. However, proper surveillance is directly dependent on the availability of robust diagnostic technologies used for infection assessment. This need is further underscored in studies of antibiotic treatments being investigated for targeting *Wolbachia* endosymbiotic bacteria [@pntd.0000834-Hoerauf1]--[@pntd.0000834-Specht1] as well as reports of sub-optimal response to ivermectin treatment [@pntd.0000834-Churcher1], [@pntd.0000834-OseiAtweneboana1]. In both of these cases an accurate diagnostic is critical for the analysis of drug efficacy and patient drug response. Currently, multinational control and elimination programs primarily rely on various techniques for diagnosis including: entomological studies of *Simulian* flies, *Ov* specific antigen tests, antibody tests, analysis of microfilariae in skin snips, nodule palpation and quality of those nodules that can be excised. There are a number of technical concerns with each technique including: a lack of sensitivity and reproducibility, invasiveness, and the inability to distinguish past from present infection or between filarial diseases [@pntd.0000834-Boatin1]--[@pntd.0000834-Boatin2]. A small molecule/metabolite based test has the potential for reflecting a more accurate picture of infection status, as it is a comprehensive measure of the effects of posttranslational modification and regulation. Furthermore, small molecules are frequently constitutively produced (e.g., excretory-secretory products), diffuse easily and are inherently non-immunogenic *in vivo*, thus avoiding some of the technical challenges associated with DNA and protein-based diagnostics. Although adult *O. volvulus* worms do not reside directly in the blood, the highly vascularized subcutaneous nodules of the human host allow for the potential diffusion of adult parasite-derived compounds into the blood where compounds involved in host response to infection might also be present. Since the microfilariae (mf) and third infective larval stage (L3) of the *O. volvulus* life cycle do come in contact with the vascular system during vector transmission, it is additionally possible that some mf or L3 produced compounds might also be localized to this biological sample. Certainly, as a starting point, the blood matrix serves as an easy to obtain, chemically complex data rich matrix for metabolite analysis [@pntd.0000834-Walsh1], [@pntd.0000834-Ritchie1]. However, a technical challenge of analyzing a large number of metabolites stems from the shear size and complexity of the resulting data set. Initially devised and applied to the analysis of highly dimensional gene micro-array data, a number of machine learning approaches have been expanded and used for identifying patterns of biomarkers resulting from the multidimensional analysis of genes, proteins, and metabolites that can be linked to early detection [@pntd.0000834-GomezRavetti1], survival prediction [@pntd.0000834-Bovelstad1], and disease outcomes [@pntd.0000834-Dolce1]. Although identification of a single biomarker "smoking gun" is perceived as the ideal scenario, more attention is being focused on the use of multiple markers for improving overall diagnostic accuracy [@pntd.0000834-GomezRavetti1], [@pntd.0000834-Zhang1], [@pntd.0000834-Karlsson1] and model stability [@pntd.0000834-Jornsten1]. Herein, we report a liquid chromatography-mass spectrometry (LC-MS) based approach to the discovery of a set of molecules that, in combination, provide a statistically relevant characteristic of onchocerciasis infection. An initial untargeted analysis was applied to the profiling of *O. volvulus* infected and uninfected blood plasma and serum samples representing a variety of geographic regions and disease states, including other tropical diseases. This analysis resulted in a set of statistically significant mass features identified for their potential as onchocerciasis-specific biomarkers. Using multivariate statistics and machine learning algorithms, these metabolic signatures were further evaluated for their ability to discriminate *O. volvulus*-infected and uninfected individuals, therefore, creating the basis of a small molecule-based diagnostic for onchocerciasis. Methods {#s2} ======= Ethics statement {#s2a} ---------------- The use of human serum and plasma samples in the study was approved by the Scripps Health Human Subjects Committee. Samples with geographic origins outside of the United States of America (USA) consisted of pre-existing, unidentifiable diagnostic specimens collected with written informed consent and in cases of illiteracy, a literate witness signed and a thumbprint was made by the participant. These samples were determined by the Scripps Health institutional review board (IRB) to be exempt from formal review under 45 CFR 46 101. *O. volvulus* negative controls from the USA consisted of serum and plasma samples and were obtained with written informed consent from healthy donors through The Scripps Research Institute Normal Blood Donor Service and approved by the Scripps Health IRB. All patient codes have been removed in this publication. Diagnostic sample origin {#s2b} ------------------------ Onchocerciasis positive samples were collected in characterized endemic areas and their status confirmed by either positive skin snip (mf +) or nodule palpation (nodule +). Several sample groups used in this analysis were collected during previously published studies including serum from Liberia [@pntd.0000834-Taylor1], [@pntd.0000834-Taylor2] and Ghana collected in 2003 [@pntd.0000834-Hoerauf3]. The Ghana sera collected in 1986 and 1991 were obtained from the College of Public Health, University of South Florida. Cameroon samples were obtained as part of a nodulectomy campaign conducted in villages surrounding Kumba, Cameroon in 2006 and consist of plasma from *O. volvulus*-positive individuals (nodule + with nodules containing live females), *O. volvulus*-negative individuals (skin snip - volunteers with no current or prior symptoms of *O. volvulus* infection), and ambiguous samples (nodules contained either dead, calcified worms or lipomas with no evidence of worms, or for which there were no particular disease symptoms recorded). Guatemala sera were obtained as part of a nodulectomy campaign conducted by the Guatemala Ministry of Health and the Centro de Estudios en Salud, Universidad del Valle de Guatemala in several villages within the Guatemalan Central Endemic Zone from 2007--2008. Nodules were surgically removed from all individuals sampled, and nodule dissection was conducted to assess worm viability on nodules from five of the 21 individuals whose serum was analyzed in this study. Of those dissected, no live worms were found. Leishmaniasis positive, Chagas disease positive, and onchocerciasis negative sera were obtained from the Centro de Estudios en Salud, Universidad del Valle de Guatemala. Indian lymphatic filariasis positive plasma samples were obtained from the Laboratory of Parasitic Diseases, U.S. National Institutes of Health. A detailed summary of the samples analyzed in this study is presented in [Table 1](#pntd-0000834-t001){ref-type="table"}. 10.1371/journal.pntd.0000834.t001 ###### Summary of samples analyzed in this study. {#pntd-0000834-t001-1} Country of origin Matrix Number/description Clinical pathology Nodulectomy results Average mf per mg ------------------- -------- ---------------------------------------------- ------------------------------- --------------------------------------- ------------------------------------- Cameroon Plasma 16 *O. volvulus* infected Palpable nodule(s) All nodules contained live worms ---[a](#nt101){ref-type="table-fn"} 18 uninfected controls --- NA 0[b](#nt102){ref-type="table-fn"} 1 calcified worm Palpable nodule(s) 1 dead, calcified worm --- 2 ambiguous lipoma Palpable nodule(s) Fat deposit --- 1 indeterminant infection status --- --- --- Ghana Serum 15 *O. volvulus* infected Palpable nodule(s) --- 12[c](#nt103){ref-type="table-fn"} 10 *O.volvulus*, mf+ Acute papular onchodermatitis --- \+[d](#nt104){ref-type="table-fn"} Liberia Serum 10 *O. volvulus* infected --- --- 527[e](#nt105){ref-type="table-fn"} Guatemala Serum 21 *O. volvulus* infected Palpable nodule(s) 24% of nodules sampled all worms dead --- 17 uninfected controls --- NA --- 6 *Leishmania braziliensis* --- NA --- 6 *Leishmania mexicana* --- NA --- 5 *Trypanosoma cruzi* --- NA --- India Plasma 4 *Wuchereria bancrofti* --- NA --- USA Plasma 3 *O.volvulus* uninfected, no known disease --- NA --- Serum 3 *O. volvulus* uninfected, no known disease --- NA --- --- indicates no measurement made. no mf were measured in samples collected from four different anatomical locations from the 18 control individuals included in this analysis. Average number of mf measured per patient in original study [@pntd.0000834-Hoerauf3] using two skin snips [@pntd.0000834-Hoerauf4]. Samples are known to be mf+, but records are not available. Average number of mf collected from six different anatomical locations from the 10 patients included in this analysis. Sample preparation and metabolite extraction {#s2c} -------------------------------------------- Solvents used were of high performance (HPLC) grade. A methanol precipitation of proteins was conducted by adding 400 µl aliquots of ice cold methanol to 100 µl aliquots of serum and plasma samples. The samples were immediately vortexed for 30 sec and allowed to rest on ice for 20 min. After centrifugation at 13,780×g for 5 min, the metabolite containing supernatent was removed from the precipitated protein pellet and transferred to fresh tubes. The supernatent samples were dried in a GeneVac EX-2 Evaporation System (GeneVac Inc., Valley Center, New York, USA) at ambient temperature and then resuspended to a 50 µl volume in water: acetonitrile (95∶5), vortexed for 30 sec and then centrifuged again at 13,780×g for 5 min. After being transferred to LC vials, samples were stored at 4°C and transferred to the LC-MS thermostated autosampler (6°C), typically within 48 h of their preparation. Chromatographic workflow {#s2d} ------------------------ In order to minimize instrumental drift, sample sequences were composed of a single injection of each sample in randomized order. To monitor any potential instrument irreproducibility and to confirm the absence of sample carry over within the chromatographic run, a mobile phase blank and an external standard were injected every 24 h throughout the duration of the analysis. LC-ESI MS analysis {#s2e} ------------------ Experiments were performed with an electrospray-ionization time-of-flight (ESI-TOF) MS (Agilent 1200 LC, TOF 6210, Agilent Technologies, Santa Clara, CA, USA). Each sample analysis consisted of an 8 µl injection of extracted sample with chromatographic separation across a reverse phase C18 column (Zorbax 300SB C18 Capillary, 3.5 µm, 1 mm×150 mm; Agilent Technologies, Santa Clara, CA, USA) at a capillary pump flow rate of 75 µl/min. Mobile phase A was composed of water with 0.1% formic acid, and mobile phase B was acetonitrile with 0.1% formic acid. Each sample was analyzed over a 60 min run time with a gradient consisting of a 45 min linear gradient from 5% to 95% B and 15 min isocratic hold at 95% B. Between sample injections a wash step was used to minimize carry over. It consisted of a saw-tooth linear gradient beginning with a hold at 95% A for 10 min. Then, linear ramping between 5% and 98% B for 5 minute increments throughout the 35 min wash cycle was followed by a 20 minute final re-equilibration of the column with an isocratic hold at 95% A. Mass spectrometric conditions {#s2f} ----------------------------- Consistent mass accuracy (\<2 ppm) was maintained through the constant infusion (2 µl/min) of reference masses via a second nebulizer. Data were collected in positive electrospray ionization (ESI) mode scanning in centroid mode from 75 to 1,100 *m*/*z* with a scan rate of 1.0 spectrum per second in 2 GHz extended dynamic range. The capillary voltage was 3,500 V; the nebulizer pressure, drying gas flow and gas temperature were set to 20 psig, 12 l/min and 350°C, respectively. Data pre-processing, pattern determination, and statistical analysis {#s2g} -------------------------------------------------------------------- All mass spectral data was collected in .d format and converted to .mzData using the Mass Hunter Qualitative Analysis software version B.03.01 (Agilent Technologies, Santa Clara, CA, USA). XCMS [@pntd.0000834-Smith1] software was used for peak matching, non-linear retention time alignment and quantitation of mass spectral ion intensities across all .mzData mass spectral files. Statistical comparison of the intensity data was conducted using the XCMS built in Welch\'s *t*-test. False discovery rate (FDR) analysis was conducted with the q-value program [@pntd.0000834-Dabney1] in R version 2.9.0 [@pntd.0000834-Team1]. Principal Components Analysis (PCA) was conducted with Statistica software version 8.0 (StatSoftInc., Tulsa, OK, USA), machine learning algorithms were implemented using Weka Explorer version 3.6.0 [@pntd.0000834-Hall1] with 10 fold cross-validation settings. Compound formula assignment {#s2h} --------------------------- The molecular formula assignment made for the 10 selected small molecule biomarkers was conducted through a combination of LC-MS/MS fragmentation using a quadrupole- TOF MS (QTOF 6510, Agilent Technologies, Santa Clara, CA, USA) and sub-2 ppm accurate mass measurements using a Bruker Daltonics Apex II 7.0 Tesla Fourier transform ion cyclotron (FT-ICR) MS (Bruker Daltonics., Billerica, MA,USA). For the QTOF analysis chromatographic conditions were identical to those reported for the profiling experiment and serum plasma samples from either the Scripps normal blood or pooled patient samples were used for the analysis. The average *m/z* and retention times of each of the biomarkers obtained through XCMS analysis, were used for targeted MS/MS analysis with a starting collision-induced dissociation energy of 20eV. Fragmentation patterns were analyzed with the Agilent Mass Hunter Qualitative Analysis software version B.03.01 using the targeted MS/MS and formula generation algorithms and compared with the MS/MS fragment data in the METLIN database [@pntd.0000834-Smith2]. The FTMS system was equipped with a custom machined electrospray source with two nebulizers for dual spray ionization. The main orthogonal nebulizer was used for LC-eluent, while the second nebulizer was used to introduce a calibration mixture containing two compounds (aminoantipyrine at 204.1132 *m/z* and quinidine 325.1911 *m/z*) at 3 mM concentration mixed with 1∶10 dilution of Agilent low concentration tune mix. A linear calibration fit was used in the narrow range to internally calibrate individual mass spectra. The chromatographic conditions were identical to those reported for the profiling experiment with an additional analysis using a smaller i.d. column with the same stationary phase composition (Zorbax 300SB C18 Capillary, 3.5 µm, 0.3 mm×150 mm; Agilent Technologies, Santa Clara, CA, USA) at a capillary pump flow rate of 4µl/min. Pooled serum and plasma samples from either the Scripps normal blood or patient samples were used for the analysis. Results {#s3} ======= A metabolomic approach was developed to address the need for improvement of diagnostics in onchocerciasis detection. Profiling of blood biomarkers is much less invasive than skin snipping or nodulectomy, and should have the added advantage of increased sensitivity. Antigen tests have been attempted in the past, however, the immunogenicity of the proteins has been a consistent deterrent [@pntd.0000834-Bradley1], [@pntd.0000834-Park1]. An advantage of profiling low molecular weight compounds is that they are typically not immunogenic (i.e., M.W. \<1,100 amu) and therefore not subject to such a limitation. It is important to note that profiling molecules with molecular weight less than 1,100 amu will also include peptides and/or protein fragments, expanding the pool of available analytes that can be detected. Mass spectrometric biomarker selection {#s3a} -------------------------------------- The most important aspect of any clinical analytical study resides with the quality of the samples used; here representative serum and plasma samples from a variety of subject populations were incorporated to minimize the effects of non-relevant metabolic variation (e.g., nutrition, sex, age, race) and magnify those metabolic differences that are not only statistically significant between specific populations, but relevant in identifying the changes in metabolism that can be directly attributable to infection. One of the analytical limitations with an untargeted LC-MS metabolomics approach is that of inter-sequence reproducibility (i.e., sample preparation, instrument drift, column and mass spectral baseline variation) when comparing samples directly between analytical sequences. Such inter-sequence variability can introduce shifts in ion intensities that can interfere with the accuracy of downstream statistical analysis. Therefore, this study was conducted with single injections of each sample, analyzed in randomized order consecutively within one analytical sequence ([Figure 1](#pntd-0000834-g001){ref-type="fig"}). Due to such analytical constraints, small groups of representative samples were selected from various sample classes (e.g., *O. volvulus*-infected and uninfected individuals from various geographic regions and individuals infected with other parasitic diseases). XCMS analysis of the sample mass spectral data files (n = 136) resulted in the measurement of a total of 2,350 mass features. Testing the overall reproducibility of the analysis, the coefficient of variation (CV) was found to be 15.9% as calculated from all mass feature intensity values compared across triplicate injections of a single plasma sample analyzed throughout the analytical sequence. This value is comparable to previous studies of analytical variation within plasma and serum analysis by our laboratory and consistent with a number of other LC-MS based metabolomics studies [@pntd.0000834-Crews1], [@pntd.0000834-Smilde1]. Statistical comparison between all onchocerciasis positive samples (n = 76) and all onchocerciasis negative samples (n = 56), including those infected with other tropical diseases, by Welch\'s *t*-test resulted in 194 features with a *p*\<1×10^−4^; with a false discovery rate FDR of 54%. To reduce the number of potentially erroneous markers and focus on those mass features with the most potential in distinguishing disease, the top 35 mass features (*p*\<1×10^−7^) were chosen for more stringent analysis through assessment of the quality of the resulting extracted ion chromatograms (EICs) ([Figure S1](#pntd.0000834.s001){ref-type="supplementary-material"}). While XCMS pre-processing software contains a robust retention time correction and peak alignment algorithm, an important aspect of this study is the statistical quantitation of biomarkers, therefore any features with questionable quantitation, observed as imperfect alignment or inconsistent peak boundaries across samples were ruled out of further analysis. Additionally, since several mass features may redundantly describe one chemical metabolite due to the presence of in-source fragments, adducts, or multiply charged species and overlapping retention time. The features were separated into unique peak groups and representative ions with the highest overall abundance were included in a subset of 14 features for further analysis ([Table 2](#pntd-0000834-t002){ref-type="table"}). Interestingly, the majority of these features were detected at lower levels in infected individuals relative to those without onchocerciasis. Analysis of the selected biomarkers with MS/MS and FTMS analysis has provided molecular masses and assigned molecular formulas that could be used to classify the biomarkers into distinct chemical classes; of the 14 markers identified 10 were small molecules and four were protein fragments or small peptides. {#pntd-0000834-g001} 10.1371/journal.pntd.0000834.t002 ###### Characteristics of the 14 candidate biomarkers. {#pntd-0000834-t002-2} Compound Classification p-value RT (min) XCMS average *m/z* Fold change Molecular formula MS/MS major fragments (% abundance)[a](#nt106){ref-type="table-fn"} FTMS accurate mass ----------------------------- -------------- ---------- -------------------- ------------- ------------------- --------------------------------------------------------------------- ------------------------------------- Fatty acid/Sterol lipid 6.32×10^−13^ 45.7 521.4197 −3.36 C~32~H~56~O~5~ 111.0451(39.15)503.4123(29.7) 521.4190 Fatty acid/Sterol lipid 2.06×10^−11^ 45.7 469.3872 −3.71 C~28~H~52~O~5~ 415.357(100)291.2331(48.57) 469.3888 Sterol lipid 2.16×10^−11^ 41.4 425.3611 −3.55 C~26~H~48~O~4~ 389.3432(100.0)139.1107(12.8) 425.3625 Protein 3.59×10^−10^ 31.6 979.9368 −3.99 \- Protein 6.53×10^−10^ 31.5 986.2677 −5.65 \- Hexacosenoic acid 4.01×10^−10^ 50.7 395.3867 −2.77 C~26~H~50~O~2~ 71.0859(100.0)57.0709(81.99) 395.3804 Pentacosenoic acid 7.75×10^−10^ 49.1 381.3710 −2.43 C~25~H~48~O~2~ 71.0858(100.0)57.0719(58.75) 381.3728 Fatty alcohol/aldehyde 2.39×10^−8^ 48.5 241.2505 1.54 C~16~H~32~O 55.0551(77.61)83.0877(46.81) ---[b](#nt107){ref-type="table-fn"} Fatty acid 1.40×10^−9^ 39.0 367.2840 −2.28 C~22~H~38~O~4~ 331.2649(100.0)79.0547(33.89) 367.2828 Hydroxy-octadecenoic acid 1.52×10^−9^ 46.1 299.2581 −2.54 C~18~H~34~O~3~ 95.0851(100)71.0863(82.07) 299.2592 Phosphorylated sphingolipid 4.83×10^−9^ 30.0 352.2256 −1.55 C~16~H~34~NO~5~P 236.2366(100.0)184.0694(25.99) 352.2247 Sterol lipid 1.44×10^−8^ 45.5 447.3470 −2.19 C~28~H~46~O~4~ 429.3376(43.06)411.3276(32.91) 447.3470 Protein 2.05×10^−8^ 33.3 966.5938 −3.09 \- Protein 5.24×10^−8^ 31.7 1086.2922 −2.76 \- Fragments collected under a collision-induced dissociation energy of 20 eV. FTMS accurate mass was not obtained for this compound. This formula is based on TOF-MS mass accuracy. Statistical values such as p-value and fold change were determined by XCMS analysis of the *O. volvulus* + and *O. volvulus* − mass spectral data files. Retention time (RT), and mass to charge value (*m/z*), fold change and the direction of overall ion intensity change, represents the average value across all files. Molecular formula and compound class identifier as determined by MS/MS and FTMS analysis is provided. Multivariate analysis of African serum and plasma biomarkers {#s3b} ------------------------------------------------------------ Beginning with a subset of the larger sample set, the mass spectral data for the top 14 candidate biomarkers were investigated for their ability to discriminate *O. volvulus* infected individuals (n = 55) from healthy controls (n = 18) from the African serum and plasma samples. PCA of the effect of these 14 biomarkers was used to visualize the variation between these samples groups ([Figure 2A](#pntd-0000834-g002){ref-type="fig"}). A distinct clustering of the *O. volvulus* infected versus the healthy individuals was observed across the x-axis of the PCA score plot, implying that principal component 1 (PC1) contained the variance of the data set required to distinguish these two sample groups. The next greatest amount of variation within the data set appeared to have little effect on discriminating infection or even geographic differences, but appears to be more representative of the heterogeneity present among healthy controls. {#pntd-0000834-g002} Multi-region multivariate analysis {#s3c} ---------------------------------- The top 14 candidate biomarkers were also applied to a larger sample set comprised of multiple geographic regions including *O. volvulus*-infected individuals (n = 76) and healthy and disease controls (n = 56). PCA of these 14 biomarkers ([Figure 2B](#pntd-0000834-g002){ref-type="fig"}) revealed the inherent complexity encountered when employing a metabolite profiling approach to diagnostic development. As with the initial African samples, there is general clustering of the onchocerciasis positive individuals with the variance contained in PC1 having good discriminatory power. The disease and healthy controls cluster separately from the onchocerciasis positive individuals, however, there is some overlap between one of the Chagas disease and two of the leishmaniasis positive individuals. Interestingly, the lymphatic filariasis samples, infected with the closely related filarial parasite *Wuchereria bancrofti*, cleanly cluster with the healthy controls. Ideally serum and plasma samples would not be directly compared against each other as the two matrices have distinct chromatographic differences ([Figure S2](#pntd.0000834.s002){ref-type="supplementary-material"}). However, given the nature of onchocerciasis sample banks that have been collected over the past 20 years, it was important to determine if the resulting biomarker results would be biased to one biological sample type over another. Importantly, our results show that the plasma samples from Cameroon as well as the Indian lymphatic filariasis plasma samples consistently align as expected with the multi-region serum sample set in distinguishing onchocerciasis infected from uninfected individuals. Guatemala central endemic zone metabolic signature variability {#s3d} -------------------------------------------------------------- As evidenced in [Figure 2C](#pntd-0000834-g002){ref-type="fig"}, there is little clustering of the Guatemalan individuals initially classified as onchocerciasis positive; rather there appears to be a continuum of onchocerciasis disease variation within those samples. However, dissections of excised nodules at the time of nodulectomy revealed no live worms, as opposed to the results of the Cameroon samples where infection status was confirmed by the extraction of live *O. volvulus* worms. Machine learning algorithm implementation {#s3e} ----------------------------------------- Although tools such as PCA provide a graphical means of distinguishing between sample groups, they do not have the ability to provide a quantitative diagnostic assessment as would be needed nor are they intended to be used for field applications of an onchocerciasis diagnostic. Alternatively, machine learning algorithms do provide the necessary binary output, as well as calculate confidence intervals of a given classification. The mass spectral intensity values for the onchocerciasis serum and plasma data set were used as inputs in a collection of machine learning algorithms. The algorithms were chosen to provide a survey of the various types of machine learning algorithms that could be used with mass spectral data in diagnostic assessments, either alone or in combination in more sophisticated algorithms. Results of this analysis are summarized in [Table 3](#pntd-0000834-t003){ref-type="table"} where sensitivity (true positive rate) and specificity (1--false positive rate) are displayed. The receiver operating characteristic (ROC) areas present a numerical value description of the relationship between sensitivity and specificity for a given diagnostic test [@pntd.0000834-Swets1], [@pntd.0000834-Lasko1]. In the context of a binary classification problem as presented here, a value of 0.5 indicates there is no discrimination within the test and shows any result is essentially the same as a random guess, while a value of 1.0 indicates a perfect test prediction. Based upon the data, it is clear that the inclusion of the Guatemala samples within the sample analysis dramatically increases the number of reported false positives, compromising the accuracy of the test overall. However, it is important to note that within the context of the Africa sample set, the ROC area approaches, or is equal to, a perfect test prediction in numerous cases, and in the case of the functional trees classification tree algorithm, perfect sensitivity and specificity can be achieved. 10.1371/journal.pntd.0000834.t003 ###### Summary of the diagnostic accuracy of the machine learning algorithm analysis. {#pntd-0000834-t003-3} Algorithm Classifier type Entire sample set Africa samples ----------------------- -------------------------------- ------------------- ---------------- ------- ------ ------ ------- BayesNet Bayesian network 84.8 87 0.929 97.3 99.1 1 NaiveBayes Bayesian network 88.6 88.3 0.930 94.5 98.2 1 Logistic Logistic regression 85.6 85.2 0.898 98.6 99.6 0.999 IB1 Nearest neighbor 87.1 83.9 0.855 98.6 95.8 0.972 OneR Minimum error attribute 76.5 76.6 0.766 89.0 85.2 0.871 Multilayer perceptron Backpropagation classification 87.9 85.9 0.921 98.6 95.8 1 FLR Fuzzy lattice reasoning 81.1 83.7 0.824 98.6 99.6 0.991 Functional trees Classification tree 84.1 83.6 0.861 100 100 1 Random forest Classification tree 88.6 89.3 0.954 97.3 95.4 0.997 The mass spectral ion intensities of the top 14 candidate onchocerciasis biomarkers from onchocerciasis infected and uninfected samples were compared between the multi-region sample set and the African blood samples. All results were obtained using a 10 fold cross validation analysis. Discussion {#s4} ========== Metabolomics, or the measurement of all the metabolites present in an organism, and metabolite profiling, in which a smaller subset of metabolites are measured, have become established as useful tools in the "real-time" measurement of organismal metabolism. For infectious disease, previous metabolomics approaches have included mice challenged with the protozoan parasites *Trypanosoma brucei brucei* [@pntd.0000834-Wang1] and *Plasmodium berghei* [@pntd.0000834-Li1], trematode parasites *Echinostoma caproni* [@pntd.0000834-Saric1] and *Schistosoma mansoni* [@pntd.0000834-Wang2] and some viruses [@pntd.0000834-Vinayavekhin1]. This study represents the first investigation of a metabolomic approach to the discovery of biomarkers and creation of a diagnostic test for identifying and classifying onchocerciasis infection. Through the use of multivariate statistics and machine learning algorithms, the potential of metabolomic analysis has been demonstrated for uncovering biomarkers for specific determination of not only onchocerciasis infection but holds promise for the diagnosis of other parasitic diseases. Specifically, this was demonstrated by the clustering of the *W. bancrofti* infected samples with those individuals that were not infected with *O. volvulus* in the multivariate PCA. This clustering showed the potential specificity of the biomarkers for the discrimination of onchocerciasis from other filarial diseases. Although this analysis consists of only four representative lymphatic filariasis samples, the distinct clustering of these samples with uninfected individuals is noteworthy and argues for future analysis that includes other filarial disease pathogens (e.g., *Brugia malayi*, *Loa loa*). The 14 candidate biomarkers showed excellent performance in the African specific sample set with up to 99--100% sensitivity and specificity when examined with the single machine learning algorithms. With 99% of onchocerciasis disease prevalence in Africa [@pntd.0000834-Basanez1] and the presence of multiple regions of ongoing transmission [@pntd.0000834-Dadzie1], this is the most clear test of the biomarker strategy. When applied to a multi-region sample set, the multivariate PCA of the biomarker analysis resulted in a wide spread of results across the range of infected and uninfected individuals. This observation raises several questions regarding the unique epidemiological challenges of measuring onchocerciasis in the Americas. In the context of the PCA, the Guatemalan patients did not classify as expected if nodule presence alone is used as an indicator of infection. However, nodule presence as a diagnostic is known to have exceedingly poor sensitivity and specificity. A possible explanation of this data is that the observed heterogeneity is related to microfilarial load. Unfortunately, skin snip samples with mf counts were not collected for the Guatemala sample set. Nonetheless, if this observed spread of data were correlated with variation in the presence of the mf, then in a region such as the Guatemalan Central Endemic Zone (CEZ) where biannual dosage of ivermectin reaches high coverage levels [@pntd.0000834-World2], mf should be nearly absent and we would expect to see no spread of the data but rather a distinct cluster with or near the uninfected individuals. Alternatively, the observation that a quarter of the nodules from these infected individuals from the Guatemalan CEZ did not contain living worms, indicates that these biomarkers may be sensitive to not only the presence, but also the viability of the infective worms. The results of this PCA are consistent with an increasing body of evidence that biannual ivermectin treatments, as are received in the Guatemalan CEZ, have an effect on the viability of adult female worms and ultimately on the elimination of parasites [@pntd.0000834-Cupp1]--[@pntd.0000834-RodriguezPerez1]. Since the Guatemalan *O. volvulus* positive samples do not segregate along clear lines with the clinically confirmed samples from Africa, it is possible that the continuum seen in the PCA plot reflects a range of infection that could be correlated qualitatively or quantitatively to the health of the worms (e.g., live healthy, dying, and dead) *in vivo*. Given that an individual with dead or dying worms does not need further treatment in the context of ivermectin mass drug administration, this finding is particularly valuable in the context of onchocerciasis elimination progress. Ideally, a biomarker determination study would involve independent sample sets for training, validation, and testing. Due to sample limitations inherent to onchocerciasis and many neglected tropical diseases in general, we have chosen to use an approach that trains on the majority of the sample set, and through the 10-fold cross validation machine learning analyses, conduct tests on small subsets of the full sample set [@pntd.0000834-Baek1]. In this study, we report only those features detected in positive ion mode with the highest statistical significance and the most accurate intensity values by XCMS analysis. Consistent among these 10 small molecule features is that they are all fatty acids and related fatty acid derivatives. Further investigations into the biological roles of these fatty acids and fatty acid sterols in onchocerciasis disease progression and potential interaction with the down-regulated proteins is of distinct interest, not only in the development of a diagnostic but also to more clearly understand the biology of this disease. Almost certainly, other biomarkers could be discovered and validated simply by altering the chromatographic (e.g., HILIC) and/or ionization conditions (e.g., negative mode ESI, APCI). It is possible that additional markers can be eventually be added to the repertoire of biomarkers used for onchocerciasis detection, further increasing assay specificity. The achievement of the goals of elimination and eradication of onchocerciasis and of the neglected tropical diseases in general, ultimately depends upon the ability to measure and track the progress of disease elimination and recrudescence. Our study highlights advantages of a metabolomics based diagnostic over onchocerciasis diagnostics currently implemented including: sensitivity, reproducibility, invasiveness, and the potential for multiplexing with biomarkers for other filarial and/or neglected tropical diseases. Fine calibration of this test in the Western Hemisphere would require characterized samples from individuals with confirmed active infection. Unfortunately, these samples are rapidly becoming a rarity due to the success that has been achieved by OEPA. Further refinement and validation of this metabolomic based diagnostic approach calls for an expansion of the mass spectral analysis with larger sample sets, while inclusion of a greater demographic representation will allow for further validation of the test in specific populations (e.g., children, adults, different genetic backgrounds). Eventually, the optimized biomarkers can be ported into field-based technologies (e.g., immuno-chromatographic or micro-fluidic-based tests) for use as a point-of-care diagnostic, a determinant for the distribution and duration of treatment, and ultimately for long-term disease surveillance. Supporting Information {#s5} ====================== ###### Extracted Ion Chromatograms of the 14 candidate biomarkers as determined from XCMS analysis of *O. volvulus* +(−) and *O. volvulus* −(−) mass spectral data files. (0.56 MB TIF) ###### Click here for additional data file. ###### An overlay of representative serum (−) and plasma (−) TICs (total ion chromatogram) collected from TSRI normal blood. (0.05 MB TIF) ###### Click here for additional data file. We thank Sara Lustigman, Peter Enyong, Thomas Unnasch, Thomas Nutman, Achim Hoerauf, Nidia Rizzo, Nancy Cruz-Ortiz, Mauricio Sauerbrey, Eduardo Catú, Frank Richards and all of the field workers for their assistance with sample collection. Additionally, we thank the TSRI Mass Spectrometry facility, for their assistance with the analytical method and the FTMS analysis. The authors have declared that no competing interests exist. This study was supported by the Worm Institute for Research and Medicine (WIRM) and the Skaggs Institute for Chemical Biology at The Scripps Research Institute. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: **¤:** Current address: Discovery Biology, Takeda San Diego, Inc., San Diego, California, United States of America [^2]: Conceived and designed the experiments: JRD MSH TJD KDJ. Performed the experiments: JRD AAKN. Analyzed the data: JRD AAKN TJD. Wrote the paper: JRD TJD KDJ.

Introduction ============ Adrenocortical carcinoma (ACC) is a rare and highly aggressive malignancy with an annual incidence of 0.7-2.0 cases per million individuals. The initial manifestations of ACC are Cushing\'s syndrome and/or virilization presenting as an abdominal mass or the feeling of abdominal fullness. Acute pancreatitis has never been reported as an initial manifestation of ACC. There are various causes of acute pancreatitis, including alcohol, choledocholithiasis, hypertriglyceridemia, pancreatic divisum, sphincter of Oddi dysfunction, and pancreatic tumors. Metastasis-induced acute pancreatitis (MIAP) is very rare and initially manifests as cancer. In cases of MIAP, the original cancer is usually in the lung, especially small cell lung cancer (SCLC), although renal cell carcinoma has also been reported. We herein report the first case of a patient with ACC presenting as an initial manifestation of MIAP and reveal that endoscopic ultrasonography-guided fine-needle aspiration (EUS-FNA) can be used to diagnose metastasis from ACC. Case Report =========== A 26-year-old woman visited our hospital with upper abdominal pain. Her medical history and family history were not remarkable. She worked as a bar hostess and so consumed alcohol habitually (\>100 g ethanol per day) and smoked. Computed tomography (CT) showed a left adrenal tumor (110×77 mm in diameter), and imaging revealed acute pancreatitis (enlargement of the pancreatic parenchyma and stranding of the surrounding fat). Her serum amylase (1,084 U/L; normal range: 37.0-125 U/L), white blood cell count (9,940/mm^3^; normal range: 3,500-8,000/mm^3^), and C-reactive protein (18.0 mg/dL; normal range: 0.0-0.3 mg/dL) values were elevated. We diagnosed her with acute pancreatitis caused by alcohol and an incidental adrenal tumor. To diagnose the adrenal tumor, we conducted an endocrinologic examination. She had no typical symptoms of Cushing\'s syndrome, such as moon face or central obesity, pheochromocytoma, or virilization. There was no ^123^I metaiodobenzylguanidine uptake in the left adrenal tumor, and urine metanephrine and normetanephrine levels were within the normal range. Serum adrenocorticotropic hormone (ACTH) levels were low (\<1.0 pg/mL; normal range: 7.2-63.3 pg/mL), and dehydroepiandrosterone sulfate levels were high (2,493 μg/dL; normal range: 92-399 μg/mL). There was a lack of circadian rhythm of cortisol, and overnight low-dose dexamethasone (1 mg) suppression tests revealed a lack of cortisol suppression (before, 18.4 μg/dL; after 19.4, μg/dL). Examinations revealed that she had non-ACTH-dependent Cushing\'s syndrome, not pheochromocytoma. Thereafter, we performed contrast-enhanced CT and positron emission tomography (PET) studies, which revealed a left adrenal tumor \[93×78×78.5 mm; maximum standardized uptake value (SUV~max~), 9.5\] with a solitary right lung metastasis (10×9.8 mm; SUV~max~, 3.9) ([Fig. 1](#g001){ref-type="fig"}). {#g001} After her condition recovered with intravenous infusion, she underwent adrenalectomy with open surgery on hospital day 8. A pathologic examination revealed that the adrenal tumor was ACC according to the Weiss system. The tumor met 5 of the 9 criteria for ACC \[mitotic rate \>5/50 high-power fields, \<25% clear cells, diffuse architecture (\>1/3 of the tumor), necrosis, and sinusoidal invasion\], and the Ki67/MIB1 labeling index (Ki67/MIB1LI) of the tumor was 56% ([Fig. 2](#g002){ref-type="fig"}). ![Adrenocortical carcinoma of the left adrenal gland. a: Resected specimen. b: Clusters of tumor cells with eosinophilic cytoplasm and \<25% clear cells \[Hematoxylin and Eosin (H&E) staining, ×40\]. c: Sinusoidal invasion (H&E staining, ×200). d: Mitotic rate \>5/50 high-power fields (H&E staining, ×400).](1349-7235-58-2645-g002){#g002} While preparing for pulmonary resection, she suffered a second upper abdominal attack on hospital day 22. Laboratory data showed that her serum amylase level and white blood cell count had increased (1,629 U/L and 11,260/mm^3^, respectively). CT showed acute pancreatitis and proximal main pancreatic duct (MPD) dilatation. Magnetic resonance cholangiopancreatography (MRCP) also revealed proximal MPD dilatation and diminishment of the distal MPD; however, neither imaging modality revealed a distal pancreatic mass ([Fig. 3](#g003){ref-type="fig"}). Therefore, we performed EUS to examine the pancreas on hospital day 27, which revealed a solitary pancreatic mass (8 mm in diameter) that obstructed the proximal pancreatic duct. Subsequently, we performed EUS-FNA with a 22-G needle (EZ shot 3 plus; Olympus, Tokyo, Japan) ([Fig. 4](#g004){ref-type="fig"}). Pathologic findings revealed a pancreatic mass that was a solitary pancreatic metastasis from ACC ([Fig. 5](#g005){ref-type="fig"}). We diagnosed the patient with acute pancreatitis attacks manifesting as MIAP. To prevent recurrent acute pancreatitis caused by pancreatic metastasis, we performed endoscopic pancreatic stenting (EPS) via the minor papilla through the MPD on hospital day 35. Thereafter, she did not experience acute pancreatitis, but the pancreatic metastasis increased in size. Synchronous multiple metastases from ACC required the initiation of mitotane-based chemotherapy (etoposide, doxorubicin, cisplatin, and mitotane) without resection of other organs. She underwent 12 cycles of mitotane-based chemotherapy, however, she ultimately died of the progression of ACC at 26.5 months after the first admission. {#g003} {#g004} {#g005} Discussion ========== This case is the first report of MIAP from ACC that was diagnosed by EUS-FNA according to a search of the PubMed database using the key words "metastasis-induced acute pancreatitis", "adrenocortical carcinoma" and "endoscopic ultrasonography-guided fine-needle aspiration". ACC is a rare and highly aggressive malignancy, with an annual incidence of 0.7-2.0 cases per million individuals. ACC can occur at any age but has a peak incidence at 40-50 years, and women are more often affected than men, accounting for 55-60% of all cases ([@B1]). Complete surgical resection is the only potentially curative treatment for ACC ([@B2]). However, metastasis is common in patients with ACC, with approximately one-third of patients having synchronous metastasis at the diagnosis ([@B3]). The most common metastatic sites are lung, liver, and bone. After initial resection, locoregional recurrence becomes a challenge with pelvic, peritoneal, or retroperitoneal metastases ([@B4]). In this case, liver metastasis was found three months after adrenalectomy, and she ultimately died of multiorgan disfunction from liver and lung metastases and peritoneal dissemination. Pancreatic metastasis is rare compared with liver and lung metastasis. Adsay et al. reported secondary tumors of the pancreas in 81 (1.6%) of 4,955 adult autopsy cases and 38 (3.9%) of 973 surgical cases ([@B5]). The lung (34 cases), gastrointestinal tract \[stomach (10), colon (5), esophagus (4), rectum (1)\], kidney (4), breast (3), and liver (2) were the most frequent sites of primary in the autopsy series, and most were non-Hodgkin\'s lymphomas (11) in the surgical series ([@B5]). According to a large single-center experience of diagnosing pancreatic metastases with EUS-FNA, the primary tumor sites were the kidney (21 cases), lung (8), skin (6), colon (4), breast (3), small bowel (2), stomach (2), liver (1), ovary (1), and bladder (1) in 49 patients. In addition, EUS-FNA confirmed pancreatic metastases in 47/49 patients (95.9%) without complications ([@B6]). The adrenal gland was not mentioned as a secondary tumor of the pancreas. Baur et al. reported a case of pancreatic metastasis from ACC. Their case was one of recurrent ACC with a solitary pancreatic metastasis that occurred six years after resection of the primary tumor ([@B7]). The current case was MIAP. Two weeks before the second attack of acute pancreatitis, PET-CT showed no proximal pancreatic duct dilatation or pancreatic mass. We did not perform MRCP at the first attack of acute pancreatitis because we considered the etiology to have been alcohol, given the patient\'s history of alcohol intake. We might have been able to detect narrowing of the MPD or slight upstream dilatation had we performed MRCP after the first attack. At the second attack, CT and MRCP revealed proximal pancreatic duct dilatation but no solitary pancreatic mass. In contrast, EUS showed an obvious solitary pancreatic mass and EUS-FNA revealed pancreatic metastasis from ACC on pathological findings. EUS is now regarded as the most sensitive imaging modality for the detection of pancreatic lesions. According to a systematic review, the overall sensitivities of EUS, CT, US, and MRI for the detection of pancreatic tumors were 94%, 74%, 67%, and 79%, respectively ([@B8]). For the detection of pancreatic tumor ≤20 mm, EUS had a higher sensitivity than contrast-enhanced CT (94.4% vs. 50.0%, n=36) ([@B9]). Furthermore, EUS-FNA is the best modality for obtaining a tissue diagnosis; it has a sensitivity of 85-86.8% and specificity of 95.8-98% for determining the etiology of a solid pancreatic mass ([@B10], [@B11]). There are no standard treatment approaches for MIAP due to the rarity of this disease and its poor prognosis. However, several authors reported that MIAP in patients with SCLC was treated successfully with chemotherapy to achieve pancreatic metastasis shrinkage and loss without recurrent acute pancreatitis ([@B12], [@B13]). In the current case, ACC showed poor responsiveness to chemotherapy and radiotherapy; therefore, we inserted a pancreatic stent through the pancreatic metastasis from the minor papilla and decompressed the proximal pancreatic duct, similar to the EPS performed for MPD strictures in patients with chronic pancreatitis ([@B14]). On reviewing the literature, Oku et al. reported that EPS was previously performed for MIAP to prevent recurrent acute pancreatitis ([@B15]). In the current case, we were unable to determine whether alcohol or pancreatic metastasis caused the first attack of pancreatitis. However, her proximal MPD was dilated slightly on CT performed after the first attack, and had had no episodes of acute pancreatitis despite being a heavy drinker. Therefore, we suspect that pancreatic metastasis caused the acute pancreatitis. In conclusion, we should consider the possibility of pancreatic tumor when we encounter cases of acute pancreatitis with uncertain causes. EUS is a useful modality for detecting small pancreatic masses, and we emphasize the efficacy of EUS-FNA for the diagnosis of pancreatic metastasis. **The authors state that they have no Conflict of Interest (COI).** [^1]: Correspondence to Dr.　Hiromu Kondo, <hkondo@higashi-hosp.jp>

INTRODUCTION ============ *ASXL1* (*additional sex combs like 1*) gene mutations frequently occur in a spectrum of myeloid malignancies, including myelodysplastic syndrome (MDS), chronic myelomonocytic leukemia (CMML), myeloproliferative neoplasms (MPNs), and acute myeloid leukemia (AML) ([@R1]--[@R4]). *ASXL1* mutation is a poor prognostic marker for MDS, CMML, and AML ([@R5]--[@R7]), suggesting an important role of *ASXL1* mutations in disease initiation and progression. The *ASXL1* gene encodes ASXL1, one of the polycomb group proteins. These proteins are necessary for the maintenance of stable repression of homeotic genes and other gene loci ([@R8]--[@R10]). We and others have reported that loss of *Asxl1* leads to the development of MDS-like diseases in mice, which can progress to bone marrow (BM) failure or MPN ([@R11], [@R12]). ASXL1 has also been shown to regulate the self-renewal and differentiation of mesenchymal stromal cells and erythropoiesis ([@R13], [@R14]). In addition, loss of *Asxl1* in hematopoietic stem cells (HSCs)/hematopoietic progenitor cells (HPCs) reduces global levels of histone H3 lysine 27 trimethylation (H3K27me3) and H3K4me3, and alters the expression of genes implicated in apoptosis ([@R11]). Cohesin is a multiple-subunit protein complex that is highly conserved in mammalian cells ([@R15]). The cohesin complex consists of four major subunits: RAD21, SMC1A, SMC3, and STAG1/STAG2 ([@R16], [@R17]). The core cohesin proteins form a triangular ring, which embraces sister chromatids and prevents their premature separation ([@R18], [@R19]). Besides its major function in sister chromatid cohesion, the cohesin complex participates in many other cellular processes, such as transcriptional regulation through long-range cis interactions ([@R20]--[@R26]). Recently, clinical studies have discovered recurrent mutations or deletions in the cohesin genes in a variety of myeloid malignancies, including MDS, AML, CMML, and chronic myelogenous leukemia ([@R16], [@R27], [@R28]). Furthermore, cohesin mutations occur in a mutually exclusive manner ([@R16], [@R29], [@R30]). A more recent study by Merkenschlager and Odom ([@R31]) suggests that cohesin associates with enhancers, promoters, and sites defined by CTCF (CCCTC-binding factor) binding to form regulated networks of long-range interactions that can promote cell type--specific transcriptional programs. Here, we report that ASXL1 interacts with the core proteins of the cohesin complex. Chromatin immunoprecipitation followed by DNA sequencing (ChIP-seq) analysis revealed a significant overlap between the ASXL1, RAD21, and SMC1A binding sites on the genome, mainly located at the promoter regions of genes. Loss of *Asxl1* decreased RAD21 and SMC1A occupancy on the genome and altered expression of their target genes. Deletion of *Asxl1* results in a significantly higher frequency of impaired telophase chromatid disjunction in hematopoietic cells, congruent with the previous finding by Díaz-Martínez *et al.* ([@R32]); the silencing of RAD21 leads to the nuclear bridging of HeLa cells. Collectively, these data demonstrate a novel biological function of ASXL1 in transcriptional regulation via interaction with cohesin complex proteins, in addition to the maintenance of normal chromatid separation. These findings provide strong evidence that ASXL1 is essential for the maintenance of gene expression and for preventing dysplastic morphology formation via cohesin complex in HSC/HPCs. RESULTS ======= ASXL1 interacts with cohesin complex proteins --------------------------------------------- Located in the nucleus and capable of regulating gene expression, ASXL1 should exert its function in concert with its interacting partners. To identify functional ASXL1 interacting proteins, we performed protein affinity purification using an anti-FLAG antibody (fig. S1A) and nuclear extracts prepared from human embryonic kidney (HEK) 293T cells engineered to overexpress FLAG-tagged ASXL1 (fig. S1B). The ASXL1 interacting proteins were identified by liquid chromatography--tandem mass spectrometry (LC-MS/MS) analysis (table S1). We found that ASXL1 associates with all core members of the cohesin complex, including RAD21, SMC1A, SMC3, STAG1, and STAG2 ([Fig. 1A](#F1){ref-type="fig"}). Co-immunoprecipitation (co-IP) and Western blotting analyses revealed that ASXL1 interacts with SMC1A, SMC3, and RAD21 in FLAG-ASXL1 overexpressing HEK293T cells ([Fig. 1B](#F1){ref-type="fig"}). Furthermore, reciprocal IP confirmed the interaction of ASXL1 with SMC1A, SMC3, and RAD21 in FLAG-ASXL1 overexpressing HEK293T cells ([Fig. 1](#F1){ref-type="fig"}, C to E). To study the endogenous interaction between ASXL1 and cohesin proteins, we carried out co-IP using primary BM cells of wild-type (WT) mice and showed that ASXL1 associates with endogenous SMC1A, SMC3, and RAD21 ([Fig. 1F](#F1){ref-type="fig"} and fig. S1C). {#F1} To determine whether ASXL1 and cohesin complex members coexist in a complex, we fractionated the nuclear extracts of ASXL1 overexpressing HEK293T cells on size exclusion chromatography. Western blot analysis showed that ASXL1 coeluted with SMC1A, SMC3, and RAD21 in high--molecular weight fractions ([Fig. 1G](#F1){ref-type="fig"}), suggesting that ASXL1 is part of a large multiprotein complex that includes cohesin complex proteins. To map the interacting region of ASXL1 with the cohesin complex, we generated a series of constructs that encode various FLAG-tagged ASXL1 truncations ([Fig. 2A](#F2){ref-type="fig"}). We then overexpressed each of the ASXL1 truncates in HEK293T cells and performed Western blots on the anti-FLAG IP. ASXL1 full-length amino acids 1 to 1010 and 1 to 587, but not amino acids 1 to 420, successfully pulled down RAD21, SMC1A, and SMC3 ([Fig. 2](#F2){ref-type="fig"}, B to D), indicating that the region spanning amino acids 420 to 587 is important for cohesin binding. Convincingly, amino acids 401 to 587 were capable of pulling down the cohesin complex ([Fig. 2E](#F2){ref-type="fig"}). These data indicate that ASXL1 and cohesin form a complex in the nucleus, and the amino acid 401 to 587 region of ASXL1 mediates its interaction with the cohesin complex. ![Mapping the region of ASXL1 that mediates its binding to the cohesin complex.\ (**A**) Schematic diagram of the full-length (FL) ASXL1 and the truncated variants of *Asxl1* \[amino acids (aa) 1 to 1010, 1 to 420, 1 to 587, and 401 to 587\]. Binding affinity was determined by the pull-down efficiency of IP with anti-FLAG and Western blotting with cohesin antibodies. NLS, nuclear localization signal. (**B** to **E**) Western blotting analysis of nuclear fractions and anti-FLAG immunoprecipitates from pcDNA3.1^+^, or each truncated ASXL1 transfected HEK293T cells using antibodies against FLAG, SMC1A, SMC3, or RAD21.](1601602-F2){#F2} ASXL1 interacts with the cohesin complex to maintain the normal cell morphology and telophase chromatin disjunction ------------------------------------------------------------------------------------------------------------------- Cohesin complex proteins embrace sister chromatids by forming a ring-like structure; the defective function of any of the core cohesin proteins disrupts the sister chromatid separation ([@R32], [@R33]). Myeloid cells with depleted *Asxl1* exhibit a specific dysplastic feature as a pseudo--Pelget-Hüet anomaly ([@R11]). Examination of the peripheral blood (PB) smear and BM of *Asxl1*^*+/−*^ and *Asxl1*^*−/−*^ mice revealed an increased frequency of cells with nuclear bridging and prominent disrupted sister chromatid separation in myeloid cells ([Fig. 3](#F3){ref-type="fig"}, A and B, and fig. S2A). Consistently, significantly higher frequencies of cells with nuclear bridging and impaired telophase chromatid disjunction were observed, such as *Asxl1*^+/*−*^ and *Asxl1*^*−*/*−*^ cultures from Lin^*−*^cKit^+^ (LK) cells in the presence of a cocktail of growth factors including stem cell factor (SCF), interleukin-3 (IL-3), macrophage colony-stimulating factor, and thrombopoietin ([Fig. 3](#F3){ref-type="fig"}, C and D). {#F3} To determine whether the increased frequency of cells with disrupted chromatin disjunction is a direct consequence of *Asxl1* loss, we transfected human ASXL1-specific short hairpin RNA (shRNA-hASXL1) and/or mouse *Asxl1* (*mAsxl1*) complementary DNA (cDNA) into HeLa cells stably expressing green fluorescent protein (GFP)--H2B (HeLa^GFP-H2B^). The mRNA levels of *hASXL1* and *mAsxl1* were determined by quantitative polymerase chain reaction (qPCR) with primers specific for *hASXL1* and *mAsxl1*, respectively. Transduction of *shRNA*-*hASXL1* successfully decreased the expression of *hASXL1* mRNA by more than 40% but did not interfere with the expression of *mAsxl1* (fig. S2B). Fluorescence microscopy was used to quantify the morphology of the cells with or without bridging in the nuclear. Knockdown (KD) of *hASXL1* induced a markedly higher frequency of dysplastic nuclear bridging cells ([Fig. 3E](#F3){ref-type="fig"} and fig. S2C), and reintroducing mASXL1 significantly reduced the frequency of cells with premature sister chromatid separation ([Fig. 3E](#F3){ref-type="fig"} and fig. S2C). A parallel study was carried out to knock down *SMC1A* or *RAD21* in HeLa^GFP-H2B^ cells. SMC1A KD resulted in a decreased expression of RAD21 and vice versa (fig. S2, D to F). This finding is consistent with those of previous studies by other groups using different cell systems that the protein levels of subunits of the cohesin complex are reduced upon depletion of other subunits of the complex, probably because of degradation ([@R34]--[@R36]). Consistently, both *SMC1A* and *RAD21* KD led to higher frequency of cells with nuclear bridging in HeLa cells ([Fig. 3](#F3){ref-type="fig"}, F and G). However, Western blot analysis showed that the levels of SMC1A, SMC3, and RAD21 proteins are comparable in WT and *Asxl1*^*−/−*^ LK cells (fig. S2G). These results indicate that ASXL1, SMC1A, and RAD21 are required for the maintenance of normal cell morphology. *Asxl1* loss--mediated cell bridging and premature chromatid separation are likely associated with an impaired cohesion function rather than dysregulating SMC1A or RAD21 expression. To further determine whether ASXL1 maintains normal cell morphology through its interaction with the cohesin complex, we disrupted cohesin/ASXL1 interaction by expressing ASXL1 amino acids 401 to 587, the cohesin binding region of ASXL1 in HeLa^GFP-H2B^ cells ([Fig. 3H](#F3){ref-type="fig"}), and measured the frequency of cells with nuclear bridging. Expression of ASXL1 amino acids 401 to 587 markedly increased the frequency of cells containing premature sister chromatid separation compared to cells expressing full-length ASXL1 or vector control ([Fig. 3I](#F3){ref-type="fig"}). This effect is presumably mediated by its disruption of the interaction between the endogenous WT ASXL1 and cohesin. These results indicate that ASXL1 is critical for the function of the cohesin complex to maintain normal cellular function during mitosis. *Asxl1* loss decreases the genomic occupancy of cohesin in LK cells ------------------------------------------------------------------- The cohesin complex binds to the genomic DNA sequence and regulates gene expression ([@R37]--[@R39]). To examine whether the genomic binding sites of ASXL1 and the cohesin proteins overlap, we next performed ChIP-seq using WT and *Asxl1*^*−/−*^ BM LK cells, as well as antibodies against ASXL1, RAD21, and SMC1A. The genomic binding sites of ASXL1, SMC1A, and RAD21 significantly overlap ([Fig. 4A](#F4){ref-type="fig"}). The percent binding cases at promoter regions for each of the three proteins are 93.45% (ASXL1), 42.61% (SMC1A), and 40.39% (RAD21), respectively. This result further confirmed the association of ASXL1 with the cohesin complex on the genome. Analysis of the genomic features showed that the ASXL1/SMC1A/RAD21 overlapping binding sites were enriched at promoter regions (93%), whereas 5 and 2% of these sites were located at the gene body and intergenic regions, respectively ([Fig. 4B](#F4){ref-type="fig"}). {#F4} Comparison of the ChIP-seq peaks in WT and *Asxl1*^*−/−*^ cells showed that loss of *Asxl1* markedly reduced cohesin occupancy on the genome (based on the SMC1A and RAD21 peaks; [Fig. 4](#F4){ref-type="fig"}, C and D). Deletion of *Asxl1* reduced SMC1A and RAD21 overlapping peaks by 35% ([Fig. 4D](#F4){ref-type="fig"}). Note that the DNA sequence recognized by SMC1A or RAD21 (similar to the CTCF DNA binding sequence) remained unchanged regardless of the presence or absence of ASXL1 ([Fig. 4E](#F4){ref-type="fig"}). These results suggest a role of ASXL1 in stabilizing, but not recruiting, RAD21 and SMC1A onto the genome. ASXL1 regulates gene expression via the cohesin complex ------------------------------------------------------- It has been reported that loss of *Asxl1* alters gene expression in LK cells ([@R11], [@R12]). To determine whether alterations in the genomic binding sites of ASXL1, SMC1A, and RAD21 are correlated with the changes of gene expression in *Asxl1*^*−/−*^ LK cells, we performed an integrated analysis of ChIP-seq data with RNA sequencing (RNA-seq) data by focusing on the \~3000 genes with ASXL1/RAD21/SMC1A overlapping peaks in the promoter regions in WT LK cells. Among the \~1400 genes with RAD21/SMC1A overlapping peaks in both WT and *Asxl1*^−/−^ LK cells, 9.3 and 4.4% were significantly up-regulated and down-regulated, respectively, in *Asxl1*^−/−^ LK cells compared to WT cells ([Fig. 5A](#F5){ref-type="fig"}). Of the \~1600 genes with loss of RAD21 and/or SMC1A occupancy in *Asxl1*^−/−^ LK cells, 14.9 and 4.1% of genes were significantly up-regulated and down-regulated, respectively, in *Asxl1*^−/−^ LK cells compared to WT cells ([Fig. 5B](#F5){ref-type="fig"}). Gene ontology (GO) analysis of these up-regulated genes revealed associations with cell differentiation, regulation of programmed cell death, myeloid differentiation, RNA polymerase II activating transcription factor (TF) binding, and negative regulation of gene expression ([Fig. 5C](#F5){ref-type="fig"}). In contrast, the down-regulated genes were associated with positive regulation of metabolic process, transcription from RNA polymerase II promoter, regulation of cell death, regulation of cell differentiation, and positive regulation of cell proliferation ([Fig. 5D](#F5){ref-type="fig"}). An enrichment map was used for visualizing the networks of these GO terms enriched with up-regulated and down-regulated genes in *Asxl1*^*−/−*^ LK cells (fig. S3). We identified a set of up-regulated genes that are enriched in the biological process termed "regulation of programmed cell death" between *Asxl1*^*−/−*^ and WT cells, such as *Atf4*, *Klf4*, and *Btg1*. Up-regulated ATF4 has been reported to induce the transcriptional initiation of the apoptosis-related *chop* gene ([@R40]). In addition, among the down-regulated genes in *Asxl1*^*−/−*^ relative to WT cells, we identified multiple genes that have a positive impact on cell proliferation, such as *Myc*, *Nfya*, and *Slc*. Patients with gene mutations of *ASXL1* and/or the cohesin complex are found in all forms of myeloid malignancies, including MDS, MDS/MPN, and AML. These cellular phenotypes might be associated with changes of multiple genes. Further studies are warranted to validate these data in human primary cells. Two recent studies using *Stag2* KD or *Smc3* deletion demonstrated that cohesin proteins are required for the maintenance of normal gene expression in HSCs, and deletion of either *Stag2* or *Smc3* altered transcriptional output, leading to differentiation skewing of HSCs ([@R35], [@R41]). {#F5} A number of ASXL1/SMC1A/RAD21 target genes in *Asxl1*^*−/−*^ LK cells are dysregulated, which are implicated in myeloid cell development and/or the pathogenesis of myeloid malignancies ([Fig. 5E](#F5){ref-type="fig"}). For example, ASXL1, SMC1A, and RAD21 co-occupy *Cbfb* and *Fus* gene promoters in LK cells ([Fig. 5](#F5){ref-type="fig"}, F and G). *Asxl1* loss reduced both SMC1A and RAD21 occupancy at *Cbfb* and *Fus* promoters. qPCR confirmed the alteration of the gene expression levels, including *Cbfb*, *Fus*, and *Stat3* ([Fig. 5H](#F5){ref-type="fig"}). These data suggest that ASXL1 acts in concert with cohesin to regulate their target gene expression and that loss of *Asxl1* decreases cohesin occupancy, leading to alterations in gene expression. DISCUSSION ========== High frequencies of *ASXL1* mutations occur in multiple forms of myeloid malignancies, and *ASXL1* mutations are associated with poor prognosis, suggesting a driving role of *ASXL1* mutations in the progression of disease ([@R1], [@R3], [@R4], [@R42]--[@R45]). However, the mechanisms by which *ASXL1* loss leads to myeloid malignancy remain to be elucidated. ASXL1 has been shown to interact with various proteins to exert its regulatory function. ASXL1 recruits polycomb repressive complex 2 (PRC2) to exert its transcriptional repression through H3K27me3 ([@R46], [@R47]). ASXL1 has also been shown to interact with the retinoid acid receptor TF ([@R48]). Identification of the key ASXL1 binding partners would facilitate the investigation of mechanisms through which ASXL1 mediates its tumor-suppressive functions. Our protein purification and LC-MS/MS analyses identified cohesin proteins (including RAD21, SMC1A, SMC3, and STAG1/STAG2) as the major binding partners of ASXL1. Similar to ASXL1, cohesin genes are frequently mutated in multiple forms of myeloid malignancies ([@R16], [@R30], [@R49]). Although ASXL1 and cohesin form a big protein complex, the *ASXL1* mutations and cohesin gene mutations are not strictly mutually exclusive in these patients ([@R16], [@R28]), indicating that ASXL1 and the cohesin complex have both overlapping and unique functions. The cohesin complex has been shown to be critical for sister chromatid cohesion ([@R23], [@R33]) and gene expression regulation ([@R37]--[@R39]). Loss of any cohesin protein leads to premature sister chromatid separation ([@R32], [@R50]). Similar to cohesin loss, we show that *Asxl1* loss also causes disruption of telophase chromatin disjunction, indicating that ASXL1 in concert with cohesin participates in this biological process. Furthermore, disruption of ASXL1 and the cohesin complex interaction by ASXL1^aa401--587^ expression, the region of ASXL1 mediating cohesin complex binding, impaired telophase chromatin separation. These results further demonstrate that ASXL1 is essential for the integral function of the cohesin complex in the maintenance of chromatin separation during cell division. We and others have reported that *Asxl1* loss leads to MDS-like disease in mice ([@R11], [@R12]). Here, we provide further evidence that *Asxl1* loss increased frequencies of dysplastic myeloid cells with disrupted chromatin separation in the PB and BM. Recently, two independent groups reported that cohesin loss leads to the development of myeloid malignancies in mice ([@R35], [@R41]). Here, we show that ASXL1 binds to the cohesin complex and plays an essential role in maintaining normal chromatin separation during cell division, suggesting an overlapping molecular mechanism that underlies the pathogenesis of the myeloid disorders driven by alterations of *ASXL1* or cohesin genes. In addition to maintaining normal sister chromatid separation, the cohesin complex has also been shown to be important in multiple processes regulating transcription and gene expression ([@R26], [@R51]--[@R54]). We also explored whether the occupancy of ASXL1 and cohesin overlaps on the genome in WT LK cells and whether the deletion of *Asxl1* alters the occupancies of cohesin on the genome. Analysis of the ChIP-seq data of ASXL1, RAD21, and SMC1A in WT and *Asxl1*^*−/−*^ LK cells showed significant overlapping in ASXL1/SMC1A/RAD21 binding sites, which are enriched at promoter regions (93%). Furthermore, *Asxl1* deletion reduced the occupancy of cohesin on the genome. The Database for Annotation, Visualization, and Integrated Discovery functional analysis of the up-regulated ASXL1 target genes revealed associations with cell differentiation, regulation of programmed cell death, myeloid differentiation, RNA polymerase II activating TF binding, and negative regulation of gene expression. Deletion of *Asxl1* in LK cells dysregulated a number of ASXL1/SMC1A/RAD21 common target genes, including *Stat3, Cbfb,* and *Fus*, which are implicated in myeloid cell development and/or the pathogenesis of myeloid malignancies. These results suggest a role of ASXL1 in stabilizing, but not recruiting, RAD21 and SMC1A onto the genome. Cohesin proteins have been shown to be important for chromatin topology and facilitate enhancer-promoter looping to regulate gene transcription ([@R55], [@R56]). Future study is warranted to investigate whether *ASXL1* mutations interfere with normal long-range chromosome interactions, altering gene expression and leading to the pathogenesis of myeloid malignancies. In this study, we have identified a novel biological link that underlies the similar clinical features in MDS that are mediated by mutations in either *ASXL1* or its major partners in cohesin genes. We also have shown that the ASXL1-cohesin interaction on the genome is important for regulating gene transcription in hematopoietic cells, establishing a novel mechanism of gene regulation by ASXL1 via the cohesin complex. Our data reinforce the view that ASXL1 has multifaceted functions in gene regulation by assembling epigenetic regulators and TFs to specific genomic loci. MATERIALS AND METHODS ===================== Mouse models ------------ *Asxl1*^+/−^ mice were generated as previously reported ([@R11]). Here, all mice were bred on a C57BL/6 genetic background. All animal experiments were conducted in accordance with the *Guide for the Care and Use of Laboratory Animals*. All animal protocols were approved by the University of Miami Institutional Animal Care and Use Committee. Plasmid constructs and shRNA ---------------------------- The cDNA of full-length *mAsxl1* and its truncated variants (amino acids 1 to 1010, 1 to 587, 1 to 420, and 401 to 587) were tagged with 3×FLAG and engineered into pcDNA3.1^+^ (Invitrogen). The shRNA plasmids of *ASXL1* (TG306527), *SMC1A* (TL513033), and *RAD21* (TL501846) were purchased from OriGene. Cell culture, retroviral transduction, and morphological analysis ----------------------------------------------------------------- HEK293T, HeLa, and HeLa^GFP-H2B^ cells ([@R57]) were cultured and transfected with full-length *Asxl1* and its truncation variant plasmids using Calcium Phosphate Transfection Kit (Invitrogen). The freshly isolated LK cells from fetal liver were cultured in RPMI 1640 medium (Gibco) supplemented with 10% fetal bovine serum, SCF (100 ng/ml), and IL-3 (10 ng/ml) (PeproTech). For morphological analysis, PB was collected and smeared for May-Grünwald-Giemsa staining. Morphological analysis and cell differentiation of BM and fetal liver LK cells were performed on cytospins (5 × 10^5^ cells per sample), followed by May-Grünwald-Giemsa staining. HeLa^GFP-H2B^ cells were cultured on ultrathin glass coverslips. Images were acquired on a DeltaVision deconvolution microscope (Applied Precision) equipped with a 60× or 100× objective. All images were obtained and processed identically. IP assay and LC-MS/MS --------------------- IP was performed using nuclear fraction buffer and washed with IP buffer \[20 mM tris-HCl (pH 7.5), 150 mM NaCl, 1% Triton X-100, 5 mM EDTA, 2 mM sodium orthovanadate, 1 mM phenylmethylsulfonyl fluoride, 2 mM NaF, and protease inhibitor cocktail (Roche)\] for four times. For FLAG-tagged ASXL1, the IPs were eluted with 3×FLAG peptide (100 ng/ml; Sigma-Aldrich, F4799) in phosphate-buffered saline for 30 min at room temperature. All the IPs were performed with nuclear extraction. FLAG-ASXL1 was immunoprecipitated from the nuclear extracts with anti-FLAG antibody--conjugated beads (Sigma-Aldrich), and the associated proteins were eluted from the beads by FLAG peptides. The eluates were then resolved on NuPAGE 4 to 12% Bis-Tris Gel (Invitrogen) followed by Coomassie brilliant blue staining, and lanes were excised for MS analysis by the Taplin Biological Mass Spectrometry Facility (Harvard Medical School). qPCR analysis ------------- Total RNA was isolated with TRIzol reagent (Invitrogen). The cDNA was synthesized using QuantiTect Reverse Transcription Kit (Qiagen). qPCR was performed using an ABI StepOnePlus with Fast SYBR Green Master Mix (Applied Biosystems). PCR amplifications were performed in triplicate for each gene of interest along with parallel measurements of β*-actin*. The primers used for the amplification of each gene are shown in table S2. ChIP assays ----------- BM LK cells were fixed with 1% formaldehyde for 15 min and quenched with 0.125 M glycine. Chromatin was isolated and sonicated to an average length of 300 to 500 bp. Genomic DNA regions of interest were isolated using antibodies against ASXL1 (Santa Cruz Biotechnology, sc-85283), SMC1A (Abcam, ab133643), and RAD21 (Abcam, ab154769). Illumina sequencing libraries were prepared and sequenced on a NextSeq 500. Raw sequence reads from the FASTQ files of the six ChIP-seq samples were trimmed using sickle (<https://github.com/najoshi/sickle>) ([@R58]) with the Phred quality score threshold of 20 bases and the length threshold of 50 bases. Next, the trimmed reads were mapped against the mouse reference genome mm9 using bowtie 1.1.2 ([@R59]), allowing two mismatches. The promoter region was defined as the region 1000 bp upstream and 1000 bp downstream of the first transcription start site of a transcript cluster constructed using UCSC (University of California, Santa Cruz) known gene annotation of mm9. The uniquely mapped reads were counted in each promoter region. These counts of all samples were tabulated, and the differential bindings of ASXL1, SMC1A, and RADad21 in the promoter regions were tested using the Poisson test. The Benjamini-Hochberg procedure ([@R60]) was used to correct for multiple hypothesis testing. The differentially bound promoters were identified with a false discovery rate (FDR) cutoff of 0.05. To evaluate and identify statistical significance on the association between SMC1A, RAD21, and ASXL1 binding profiles in WT cells, we applied a hypergeometric test and calculated *P* values (table S3). Specifically, the CTCF-like motifs were discovered on the basis of their high occurrences with significant *E* values, in the sequences where peaks were identified on each of the four ChIP-seq data sets: SMC1A in WT cells (3.3 × 10^*−*61^), RAD21 in WT cells (2 × 10^−145^), SMC1A in *Asxl1*^*−/−*^ cells (7.1 × 10^*−*146^), and RAD21 in *Asxl1*^*−/−*^ cells (2.9 × 10^*−*183^). The matches between these motifs and the known CTCF DNA binding motif were also highly significant, measured against the likelihood of the same matches by random sequences, with all significant FDR-adjusted *P* values for SMC1A in WT cells (4.12 × 10^*−*13^), RAD21 in WT cells (1.35 × 10^*−*10^), SMC1A in *Asxl1*^*−/−*^ cells (2.05 × 10^*−*13^), and RAD21 in *Asxl1*^*−/−*^ cells (2.64 × 10^*−*9^). RNA-seq and analysis -------------------- Total RNA was isolated from BM LK cells of WT or *Asxl1*^*−/−*^ mice (18- to 21-day-old mice) following standard protocol with TRIzol reagent (Invitrogen) followed by mRNA library preparation with the TruSeq strand-specific mRNA sample preparation system (Illumina). The library was sequenced (PE100bp) using the Illumina HiSeq 2500. Raw sequence reads from the FASTQ files of the four RNA-seq samples were mapped against the mouse reference genome mm9 using STAR2.3.1t with the default parameters ([@R61]). Only the uniquely mapped reads were used to count against the UCSC known gene annotation of mm9 to calculate the numbers of reads per gene. The counts of all samples were tabulated, then analyzed using DESeq ([@R62]) for normalization and identification of differentially expressed genes between the control and knockout samples using a standard workflow, as previously described ([@R63], [@R64]). The Benjamini-Hochberg procedure ([@R61]) was used to correct for multiple hypothesis testing. The differentially expressed genes were identified with an FDR cutoff of 0.05. GO analysis of differentially expressed genes of all comparisons was performed using Fisher's exact test. Statistical analysis -------------------- Differences between experimental groups were determined by Student's *t* test or analysis of variance (ANOVA) followed by Newman-Keuls multiple comparison tests, as appropriate. Supplementary Material ====================== ###### http://advances.sciencemag.org/cgi/content/full/3/1/e1601602/DC1 We thank C. Walczak (Indiana University, Bloomington, IN) for HeLa^GFP-H2B^. We also thank the histological processing and analysis services provided by the Satellite Histological Core of Sylvester Comprehensive Cancer Center Core Facility. **Funding:** This work was supported by the NIH (grant numbers 1R01CA172408-01 and 1R21CA185751-01). **Author contributions:** Z.L., P.Z., A.Y., Z.G., Y.B., J.L., S.C., H.Y., Y.H., J.L., Y.G., W.Z., E.H., H.A., and D.F. designed and performed the experiments and analyzed the data. M.X. and F.-C.Y. reviewed the blood smears and histopathologic sections. J.W.H., Y.R., S.D.N., P.Y., X.C., M.X., and F.-C.Y. designed and supervised the studies, performed the experiments, analyzed the data, wrote the manuscript, and were responsible for its final draft. **Competing interests:** The authors declare that they have no competing interests. **Data and materials availability:** All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Raw and processed next-generation sequencing data have been deposited in the Gene Expression Omnibus GSE83962. Additional data related to this paper may be requested from the authors. Supplementary material for this article is available at <http://advances.sciencemag.org/cgi/content/full/3/1/e1601602/DC1> fig. S1. ASXL1 forms a complex with the cohesin complex. fig. S2. Reintroducing *mAsxl1* rescued the premature sister chromatid separation in HeLa cells with ASXL1 KD. fig. S3. Enrichment map was used for visualizing the network of selected GO terms enriched with up-regulated and down-regulated genes in *Asxl1*^*−/−*^ LK cells. table S1. List of ASXL1 interaction proteins identified by MS in HEK293T cells transfected with FLAG-ASXL1. table S2. qPCR primer sequences. table S3. Statistical evidence for binding between SMC1A, RAD21, and ASXL1.

Background {#S1} ========== Mentorship is pivotal to the professional development of medical students.^[@ref1]^ Residents play an important role in mentoring medical students, impacting their education and career choice.^[@ref2]^ However, formal resident-student mentorship has not been well explored. Pediatric Longitudinal Experience with Residents (PedLER) is a unique program at the University of Toronto, designed to foster formal mentorship between pediatric residents and first-year medical students. The program objectives for students cover early clinical exposure, learning about life in residency, exploring pediatrics, and receiving career guidance. For residents, the program provides an opportunity to gain mentorship skills and experience balancing teaching with clinical work. Program development and structure {#S1_1} --------------------------------- Interested first-year medical students apply and are chosen by lottery. Total program size is based on the number of available resident volunteers. Students and residents participate in separate orientation sessions during which expectations are explained and information sheets are provided. Student-resident pairs are created randomly and then expected to meet at least four half-days over six months (the longitudinal period), which may involve shadowing in a variety of clinical settings, or mentorship meetings. The opportunity exists for longer-term mentorship should the pair choose. At the end of each academic year, students and residents evaluate the program via an online survey containing questions about feasibility, perceived benefit and impact, and suggestions for improvement. Participants do not formally evaluate each other as mentors and mentees. A committee of faculty and resident leaders oversees the program and the university provides administrative support. As per institutional Quality and Risk Management policies, this project is a routine part of educational design and delivery and did not require Quality Improvement approval. Based on preliminary experience and results, Research Ethics Board (REB) approval was obtained in 2017 for ongoing study. Program data {#S1_2} ------------ From 2012 through 2018, 210 resident-student pairs have participated in PedLER. Quantitative data is presented in [Table 1](#T1){ref-type="table"}. Students reported positive and non-intimidating relationships with residents. They felt comfortable asking questions and exploring the benefits and challenges of pediatrics without feeling the need to "impress" their mentor. One student commented, "Participating in PedLER was my extracurricular highlight of the year." Residents enjoyed the opportunity to meet and mentor enthusiastic students, and felt better equipped to teach students and balance mentoring with working in a busy environment. Some residents commented that participation in PedLER made them reflect on their own practice to ensure they set a good example. Negative comments revolved mostly around scheduling difficulties. ###### PedLER participation data and quantitative survey results. Surveys were initiated in 2013-2014 (residents only) and became more detailed over time. (N/A = data is not available) 2012-2013 2013-2014 2014-2015 2015-2016 2016-2017 2017-2018 ---------------------------------------------------- -------------------------------------------- ----------- ----------- ----------- ----------- ----------- ---------- Number of Students Eligible (1^st^ year students) 254 254 254 260 258 261 Applied 74 91 89 104 139 115 Placed (% of applied) 15 (20%) 32 (35%) 35 (39%) 43 (41%) 40 (29%) 45 (39%) Responded to survey N/A N/A 25 41 9 23 Number of Residents Eligible [\*](#TF1-1){ref-type="table-fn"} 62 61 54 76 80 86 Participated (% of eligible) 15 (24%) 32 (52%) 35 (65%) 43 (57%) 40 (50%) 45 (52%) Responded to survey N/A 23 19 37 11 24 Number of pairs who met [†](#TF1-2){ref-type="fn"} N/A 23 (100%) 19 (100%) 37 (100%) 11 (100%) 24 (100%) Number of pairs who met four of more times† N/A 17 (74%) 13 (68%) 20 (54%) 5 (45%) 10 (42%) Anticipate continuing a mentorship relationship Students N/A N/A N/A 26 (63%) 4 (44%) 15 (65%) Residents N/A 13 (57%) 11 (58%) 22 (59%) 5 (45%) 15 (63%) Students' interest in pediatrics Increased N/A N/A N/A 19 (46%) 4 (44%) 11 (48%) Decreased 2 (5%) 0 3 (13%) Unchanged 20 (49%) 5 (56%) 9 (39%) Eligible residents were PGY-2 to PGY-4 from 2012-2013 to 2014-2015. Starting in 2015-2016, PGY-1 residents were included. According to resident survey responses. Conclusions and future directions {#S1_3} --------------------------------- PedLER provides the opportunity for focused resident-student mentorship beyond a single rotation. The program allows residents to benefit from mentoring experience, and medical students to gain early clinical exposure and insight into residency life. Mentoring relationships between students and residents have the potential to continue beyond the formal program. It would have been helpful to initiate surveys at the program's inception to obtain data over a longer period of time. Based on our preliminary data suggesting the impact PedLER may have on the career choices of medical students, we plan to undertake further and more rigorous research. We also hope to develop recommendations for future longitudinal resident mentorship programs. Conflicts of interest: There are no conflicts of interest for any of the authors. Funding: None. [^1]: \*co-senior authors

***Background.*** Hospital stay and antimicrobial therapy are the major risk factors for developing *Clostridium difficile*infection (CDI). Community outpatient parenteral antimicrobial therapy (CoPAT) allows patients to reside in the community while being treated with parenteral antimicrobials. The purpose of this study was to evaluate the incidence and clinical outcomes of community onset CDI (CO-CDI) in patients treated with CoPAT. ***Methods.*** All patients ≥ 18 years, discharged home with CoPAT from January- December 2013 were retrospectively reviewed. Patient who developed symptomatic diarrhea with positive stool toxin PCR assay within 4 weeks of CoPAT initiation in the community were identified as CO-CDI. A review of the electronic medical records was done to identify known CDI related risk factors, severity, treatment, recurrence, readmission and attributable mortality. ***Results.*** During the study period, 2401 patients were discharged on CoPAT with 680 patients through the Cleveland Clinic Home Care agency. Five patients (0.74%) developed CO-CDI with an estimated incidence of 5 cases per 1,000 CoPAT courses. The mean age of CO-CDI patients was 61.2 (SD ± 16) years and 4/5 patients (80%) were men. In patients with CO-CDI, the most frequently administered antimicrobials (2/5) were pipercillin/tazobactam and amikacin with a median duration of 12 days (IQR: 6.5-28). Four of 5 patients completed CoPAT prior to developing CO-CDI. The median duration from CoPAT completion to developing CDI was 9.5 days (IQR: 3-13). All five patients had a recent exposure to a healthcare facility excluding follow-up office visits. The median duration from exposure to development of CO-CDI was 8 days (IQR: 2-11.5). Of the 5 patients, 2 had hospital re-admissions and 3 had outpatient procedures. Four of 5 (80%) patients were on concomitant acid suppressive therapy. All patients had mild-moderate CDI and responded to medical therapy. Two patients had a hospital readmission but none were CDI related. There was no history of CDI recurrence or attributable mortality. ***Conclusion.*** Patients receiving CoPAT had a low incidence of CO-CDI with no major complications. Most patients who developed CO-CDI had a recent healthcare exposure and were on concomitant acid suppressive therapy ***Disclosures.*** **N. Shrestha**, Forest: Speaker\'s Bureau, Speaker honorarium; Merck: Speaker\'s Bureau, Speaker honorarium; The Medicines Company: Scientific Advisor, Consulting fee [^1]: **Session:** 203. *Clostridium difficile* Infection: Epidemiology, Presentation, Treatment [^2]: Saturday, October 11, 2014: 12:30 PM

Introduction {#sec1_1} ============ The incidence of ipsilateral breast tumor relapse (IBTR) in patients previously treated for breast carcinoma through breast-conserving surgery, systemic adjuvant therapy and radiation therapy, is estimated to be less than 6%. However, as each case has different clinical relevance (evolution, prognosis and therapeutic management), it is necessary to distinguish in these patients between a true local recurrence (TR) of a previous tumor, and the appearance of a new primary tumor (NPT) \[[@B1], [@B2], [@B3], [@B4]\]. To do this, we must study the location of the new tumor (distance from the previous tumor), staging (by imaging and biopsy) and determine their histological subtype (through immunohistochemical studies, and the determination of HER-2 and hormone receptor expression). Case Report {#sec1_2} =========== A 35-year-old premenopausal woman with no relevant medical history of interest, except for a family history of a maternal cousin with breast cancer at 28 years, was referred to the oncologic gynecology department by the palpation of a 1-cm nodule in the right axilla. It was nontender to palpation, but fixed to surface skin structures. Fine needle aspiration cytology was performed, with the pathological result of suspicion of malignancy. In February 2004, the patient underwent a lumpectomy with axillary lymph node dissection. The pathological diagnosis was moderately differentiated invasive ductal carcinoma (IDC) of the right breast of 0.8 cm in size, with no metastasis detected in 20 lymph nodes removed (pT~1~ pN~0~ M~0~). Immunohistochemical study of the tumor cells showed positive staining for both estrogen receptors (ER) and progesterone receptors (PR) in 50 and 15%, respectively, and showed negative membrane staining of Her-2 marker. Fluorescence in situ hybridization (FISH) analysis found no HER-2 amplification in the primary tumor. The computer tomography performed after surgical intervention revealed no other organs compromised or distant metastatic extension. Preoperative serum levels of the marker CA 15-3 were normal. With this diagnosis, the patient was referred to our medical oncology department where she received postoperative adjuvant chemotherapy consisting of 4 cycles of adriamycin 60 mg/m^2^ and cyclophosphamide 600 mg/m^2^ every 3 weeks. Then, adjuvant radiotherapy (50 Gy) was performed on right-breast residual tissue, followed by adjuvant hormone therapy with tamoxifen 20 mg daily during the following 4 years. Meanwhile, the patient continued routine follow-up visits in our department. In January 2008 (4 years after surgery), the patient noticed the presence of periareolar thickening in the right breast with redness and heat. In the physical examination there was neither a lymph node palpable nor an underlying tumor. A mammography showed the presence of breast architectural distortion with high suspicion of malignancy (BIRADS-4). The result of the biopsy of the lesion confirmed the presence of an IDC consistent with the diagnosis of inflammatory breast cancer (pT~4~ N~X~ M~0~) in the same breast where the previous tumor was treated. Immunohistochemical study of biopsy sample showed negative staining for both ER and PR, and showed positive membrane staining of Her-2 marker. FISH analysis found HER-2 gene amplification. Then, the patient received treatment with a weekly schedule of paclitaxel (80 mg/m^2^) plus carboplatin (AUC = 2) 3 weeks on and 1 week off, in combination with weekly trastuzumab (initially 4 mg/kg followed by 2 mg/kg every week). She received a total of 11 cycles of chemotherapy schedule and 15 cycles of trastuzumab. During that time period, CA 15-3 serum levels remained normal. After completing treatment, there was a progressive decrease of inflammatory signs, with disappearance of the redness and heat from the affected area. This allowed the realization of a radical mastectomy in June 2008. The pathology report showed no signs of malignancy with a complete pathological response. After surgery, the patient received adjuvant chemotherapy consisting of 8 cycles of weekly paclitaxel (80 mg/m^2^) in combination with weekly trastuzumab (2 mg/kg), followed by trastuzumab monotherapy (6 mg/kg every 3 weeks) up to 1 year of treatment. Later, the patient underwent breast reconstruction surgery, and currently, after 4 years and 8 months of follow-up after the last surgery, the patient remains in complete clinical remission, developing her life normally. However, given the age of the patient at the time of first diagnosis and due to the presence of a second primary tumor in the same breast, a genetic study was performed with negative results for currently known breast cancer-associated genetic mutations. Discussion {#sec1_3} ========== Breast cancer is a major public health problem for women throughout the world \[[@B5]\], and IDC is the most frequent form of invasive breast cancer. It accounts for 70--80% of all cases of invasive carcinomas, and, in global, is the histologic type of breast cancer with the worst prognosis of all \[[@B6], [@B7]\]. Its treatment is based on a multidisciplinary approach consisting in primary tumor surgery, adjuvant chemotherapy and radiotherapy, and hormone therapy if indicated. Currently, the treatment of choice for IDC is a combination therapy based on neoadjuvant chemotherapy, targeted therapy, surgery, adjuvant chemotherapy, hormone therapy and radiation therapy. Numerous randomized prospective studies have shown that breast-conserving treatment in IDC is as effective as mastectomy in terms of overall survival, disease-free survival and long-term disease-free survival \[[@B8], [@B9]\]. Inflammatory breast carcinoma is a rare and very aggressive form of locally advanced breast carcinomas. It represents 0.5--5% of primary invasive breast tumors, being more common in Caucasian women under 60 years. In these cases, it is important to make a differential diagnosis with locally advanced 'noninflammatory' breast carcinomas that subsequently evolve with inflammation, as well as other non-neoplastic diseases (mastitis and breast abscess), by using biopsies and imaging tests to confirm the differential clinical diagnosis. After breast-conserving treatment, IBTR may represent two distinct types of lesion that it is important to define, a TR or an NPT. TR and NPT have different natural histories, prognosis, and in turn different implications for therapeutic management \[[@B1], [@B2], [@B3], [@B4]\]. In a retrospective study, designed by Bouchardy et al. \[[@B10]\], about second primary contralateral breast cancer, it is concluded that women with ER-positive first tumors have a decreased risk of second breast cancer occurrence (standardized incidence ratio (SIR): 0.67; 95% confidence interval (CI), 0.48--0.90), whereas patients with ER-negative primary tumors have an increased risk limited to ER-negative second tumors (SIR: 7.94; 95% CI, 3.81--14.60). Patients with positive family history had a tenfold higher risk of ER-negative second tumor which increased to nearly 50-fold when the first tumor was ER negative. So the risk of second ER-negative breast cancer is really very high after a first ER-negative tumor, particularly in women with strong family history \[[@B10]\]. Certain studies provide additional evidence on differences between ER-positive and ER-negative breast cancer, not only in presentation, prognosis, and treatment, but also in etiology and natural history \[[@B10]\]. Overall, the risk of developing a second breast cancer among women diagnosed with a first breast cancer of any ER status was similar to the risk of developing a first breast cancer in the general population. But it is important to analyze the first breast cancer by means of immunohistochemical markers, because risk of a second tumor depends on ER status, period of diagnosis, and family history of breast and/or ovarian cancer; and consequently, surveillance and prevention of second cancer occurrence should consider these risk factors. The risk of a second breast cancer is the same whether the patient had previously underwent a mastectomy or BCT. And with BCT, the risk for developing a second primary breast cancer in the preserved breast is similar to the contralateral one. Unlike disease recurrence, second primary breast cancer often occurs after the first 36 months, with an average range from 34 to 60 months, requiring a long-term follow-up \[[@B10]\]. Secondary malignancies of the breast are rare with a reported frequency of 0.4 to 2.16%. According to the National Surgical Adjuvant Breast and Bowel Project (NSABP), there is a 14.3% cumulative incidence of IBTR over 20 years since primary operation. IBTR can be defined as the re-emergence of tumor in the previously treated breast. IBTR rate is under 1% per year, but it is about 5 to 10% at 5 years and 10 to 15% at 10 years \[[@B11]\]. When evaluating IBTR, it is important to consider whether it is a TR or an NPT. On the one hand, TR are cases consistent with the regrowth of malignant cells not removed by surgery or not killed by radiotherapy, as Veronesi et al. \[[@B8]\] defined them. On the other hand, NPTs are de novo cases of malignancies arising from mammary epithelial cells of the residual breast tissue \[[@B4], [@B8]\]. The complex behavior of IBTR may be related to the fact that the IBTR patient population is composed by these two different entities. Each IBTR can be classified as either TR or as NPT based on the following criteria: Histological Criteria {#sec2_1} --------------------- According to pathologic features, NPT is designated when it is a distinct histology type (e.g. from an infiltrating ductal carcinoma to an infiltrating lobular, tubular, or medullar carcinoma), or when it has a change from a more invasive to a less invasive carcinoma. By contrast, a change in histology from ductal carcinoma in situ to an infiltrating ductal carcinoma is considered as TR because it is consistent with the natural progression of the disease. Location Criteria {#sec2_2} ----------------- Depending on whether the relapse occurs at or near the original site of the primary tumor. It is classified as NPT if it is located in a different quadrant. DNA Flow Cytometry Criteria {#sec2_3} --------------------------- A tumor is classified as NPT if it changes from an aneuploid primary to a diploid relapse; otherwise, it would be considered as a probable progression representing the natural history. The best way to differentiate NPT and TR would be genetic sequencing (to establish true clonality). Recently, some molecular techniques such as DNA finger printing, loss of heterozygosity pattern or allelic imbalances profile have been used to distinguish NPT from TR, but the classification criteria are not standardized yet. Patients with NPT were significantly younger at initial diagnosis than those who experienced TR. However, there are no age differences at relapse. The rate of NP in the ipsilateral breast does not differ significantly from the rate of primary contralateral tumors. Currently, salvage mastectomy is the standard treatment for all types of IBTR, which provides locoregional control in 90% of patients. This recommendation is based on an elevated risk of further in-breast relapse with conservative surgeries. But, there is no conclusive evidence of its superiority compared to conservative surgery, the number of repeat lumpectomies is small and the follow-up is not long enough to draw definitive conclusions. Further studies are needed to determine the indications \[[@B12]\]. After the surgical management of IBTR, the optimal systemic therapy is also unknown. TR has a poor prognosis in terms of survival rates and development of other metastases, and may benefit from more aggressive adjuvant treatment, with additional radiotherapy, hormone therapy and chemotherapy. By contrast, patients with NPT generally have a favorable prognosis, and therapeutic decisions concerning systemic therapy should be similar to those in patients with de novo breast carcinoma, according to the equivalent stage. However, because of the higher risk of developing contralateral breast carcinoma (genetic predisposition) there is a need for better chemoprevention strategies. Two trials have shown that patients with NPT benefit from adjuvant tamoxifen, if indicated, because it reduces contralateral and ipsilateral disease recurrence with minimal side effects \[[@B13]\]. Due to declining mortality rates that are, in part, attributable to the use of screening mammography and effective adjuvant therapy, more women are surviving their breast cancer. The care of breast cancer survivors is an important issue that requires an understanding of relapse patterns, establishing appropriate follow-up visits and screening tests. Nowadays, there exists a good follow-up among cancer survivors, with higher screening rates.

While prevalence is more readily measured, incidence estimates are critical for determining disease risk. These estimates are particularly useful for etiologic studies, and are also relevant for pharmacovigilance and population disease surveillance. Although several studies have reported the prevalence of medical comorbidities within the multiple sclerosis (MS) population, few have reported the incidence of medical comorbidities in this population.^[@R1]^ Few population-based studies of incident comorbidities in MS have focused on common conditions such as hypertension,^[@R1]^ and none has reported age-specific incidence estimates, even though the burden of comorbidity increases with age.^[@R2]^ Given the recognized adverse effects of vascular comorbidities (diabetes, hypertension, hyperlipidemia, ischemic heart disease \[IHD\]) on outcomes in MS such as disability progression,^[@R3]^ their epidemiology requires attention as a better understanding of the associations between comorbidity and MS is necessary to support studies to evaluate the pathophysiology of these associations. We aimed to assess the age-specific incidence of vascular comorbidity in MS, temporal trends in the incidence of vascular comorbidity, and differences in the incidence of comorbidity according to sex because of the strong female predominance in MS. We compared these findings in the MS population to those in matched controls. METHODS {#s1} ======= Administrative data {#s1-1} ------------------- We conducted this cohort study in 4 Canadian provinces---British Columbia, Manitoba, Quebec, and Nova Scotia---over the years 1995--2005. These provinces capture nearly 43% of the Canadian population,^[@R4]^ and each has procedures for accessing their anonymized, administrative (health) data, which capture nearly all residents in their jurisdictions. The data accessed included population registries, and hospital and physician claims for the years 1990--2010, except in British Columbia, where data extended to 2008.^[@R5][@R6][@R7]^ These datasets can be linked within provinces using a unique identification number. The population registries capture sex, region of residence (postal code), and dates of birth and death. Dates of health care coverage are recorded, including when an individual migrates in or out of the province. Hospital claims include dates of admission and discharge, and diagnosis codes classified using the International Classification of Disease (ICD)--9 or ICD-10 system, depending on the year. Physician claims include service date and diagnosis. Due to provincial privacy regulations, which prevent line-level data from leaving the province of origin, we performed analyses in parallel at each site, adopting the approach of the Canadian Network for Observational Drug Effect Studies.^[@R8]^ Standard protocol approvals, registrations, and patient consents {#s1-2} ---------------------------------------------------------------- The Research Ethics Boards at each participating site approved the study. The relevant body within each province (British Columbia Ministry of Health, Manitoba Health Information Privacy Committee, Commission d\'Acces a l\'Information du Quebec, Health Data Nova Scotia) approved administrative data access. Study populations {#s1-3} ----------------- Using a validated case definition,^[@R9]^ we identified all persons with MS in each province as those with ≥3 hospital or physician claims for MS (ICD-9/10 = 340/G35). We selected a matched cohort from the general population, excluding individuals with any diagnostic codes (ICD-9/10) for demyelinating disease (see appendix e-1 at [Neurology.org/cp](http://cp.neurology.org/lookup/doi/10.1212/CPJ.0000000000000230)). Statistical efficiency is optimized at 4 to 6 matches, therefore we identified up to 5 matches for each case, matched on sex, year of birth, and region of residence (postal \[mailing\] code or first 3 digits of the postal code if unable to match on the full postal code). For each person with MS, we assigned the date of the first health claim for demyelinating disease as the date of diagnosis. The same date (index date) was assigned to their matched controls. Comorbidities {#s1-4} ------------- In each province, we applied case definitions for diabetes, hypertension, hyperlipidemia, and IHD that were validated in Manitoba and Nova Scotia (appendix e-1).^[@R10]^ To estimate incidence, we required a 5-year run-in period preceding the first comorbidity claim to ensure that comorbidity cases were truly incident. In the MS cohort, a comorbidity case was considered incident if the first comorbidity claim occurred after the date of MS diagnosis. For the matched cohort, a comorbidity case was considered incident if the first comorbidity claim occurred after the index date assigned to their matched case. Artifactual drops in incidence can occur at the end of the study period because new cases lack sufficient follow-up time to meet the case definition. Therefore we report incidence for 1995--2005. Due to small cell sizes in some provinces, age was categorized as 20--44, 45--59, and ≥60 years and we examined age effects using average annual age-specific incidence over the study period. We age-standardized findings to the 2001 Canadian population (census year closest to study midpoint) using the direct method, and calculated 95% confidence intervals (CI) assuming a Poisson distribution. Cell sizes \<5 were suppressed to comply with privacy requirements, limiting the ability to model crude rates. Therefore we modeled age-standardized incidence using Poisson regression, adjusting for year and sex.^[@R11]^ This approach controls for age effects without introducing age in the model as a covariate.^[@R11]^ We pooled province-specific estimates using random-effects meta-analysis. We report *I*^2^ (measure of heterogeneity) and *τ*^2^ (measure of between study variance) with forest plots to illustrate variation in estimates across provinces (see appendix e-1). Analyses were performed using SAS V9.3 (SAS Institute Inc., Cary, NC) and an Excel spreadsheet for meta-analyses.^[@R12]^ RESULTS {#s2} ======= We identified 44,452 MS cases, 31,757 (71.4%) of whom were female, and 220,849 matched controls ([table](#T1){ref-type="table"}). All incidence rates are pooled estimates from random effects meta-analysis, reported per 100,000 persons per year (95% CI). In 1995 and 2005, the highest crude incidence rates in both populations were for hypertension ([figure 1, A and B](#F1){ref-type="fig"}). The incidence of all comorbidities increased with age in the MS (figures e-1A--e-4A) and matched (figures e-1B--e-4B) populations. ###### Data sources, number of cases and controls, age, and sex  {#F1} Diabetes {#s2-1} -------- In 2005, the crude incidence of diabetes in the MS population was 765.9, while it was 740.7 in the matched population ([figure 1, A and B](#F1){ref-type="fig"}). Adjusting for year and sex, the age-standardized incidence of diabetes did not differ between populations over the study period (incidence rate ratio \[IRR\] 0.90; 0.79--1.02, [figure 2](#F2){ref-type="fig"}). The age-standardized incidence of diabetes was 25% lower in women than men (IRR 0.75; 0.72--0.78, figure e-5). Average annual age-specific incidence rates also did not differ between the MS and matched populations (data not shown). There was an interaction between population and year (*p* = 0.04); the age-standardized incidence of diabetes increased more (IRR 1.06; 1.03--1.08, figure e-6) in the MS population than in the matched population (IRR 1.02; 1.01--1.03, figure e-7). {#F2} Hypertension {#s2-2} ------------ In 2005, the crude incidence of hypertension in the MS population was 1,553.7, while it was 1,796.2 in the matched population ([figure 1, A and B](#F1){ref-type="fig"}). Adjusting for year and sex, the age-standardized incidence of hypertension was lower in the MS population than in the matched population over the study period (IRR 0.87; 0.75--0.99, [figure 3](#F3){ref-type="fig"}). The age-standardized incidence of hypertension did not differ between women and men (IRR 0.98; 0.84--1.15, figure e-8), and was stable over time (IRR per year 1.02; 0.99--1.05, figure e-9). {#F3} The lower overall incidence of hypertension in the MS population appeared to reflect differences in the incidence rates between the 2 populations for persons aged ≥60 years (*p* for interaction = 0.021). The incidence of hypertension did not differ between the populations for those aged 20--44 years (IRR 0.98; 0.90--1.08), or those aged 45--59 years (IRR 0.94; 0.85--1.04). However, among those aged ≥60 years, the incidence of hypertension was lower in the MS population (IRR 0.75; 0.60--0.95, figure e-10). Hyperlipidemia {#s2-3} -------------- In 2005, the crude incidence of hyperlipidemia in the MS population was 701.8, while it was 1,045.2 in the matched population ([figure 1, A and B](#F1){ref-type="fig"}). Adjusting for year and sex, the age-standardized incidence rate of hyperlipidemia was 36% lower in the MS population than in the matched population over the period 1995--2005 (IRR 0.64; 0.49--0.84, [figure 4](#F4){ref-type="fig"}). This finding was consistent across age groups (data not shown). The age-standardized incidence of hyperlipidemia was 43% lower in women than men (IRR 0.66; 0.57--0.77, figure e-11). The age-standardized incidence of hyperlipidemia increased over time in both populations (IRR per year 1.04; 1.01--1.08, figure e-12). {#F4} Ischemic heart disease {#s2-4} ---------------------- In 2005, the crude incidence of IHD in the MS population was 690.5, while it was 610.2 in the matched population ([figure 1, A and B](#F1){ref-type="fig"}). Adjusting for year and sex, the age-standardized incidence of heart disease did not differ between the populations (IRR 1.00; 0.94--1.06, figure e-13), but was 56% lower in women than men (IRR 0.54; 0.50--0.60, figure e-14). The age-standardized incidence of IHD was stable or declined very slightly over time (IRR 0.99; 0.98--1.00, figure e-15). Age-specific incidence rates differed between the populations. Among those aged 20--44 years, the incidence of IHD was 59% higher in the MS population (IRR 1.59; 1.19--2.11, figure e-16), but did not differ among those aged ≥60 years (IRR 1.01; 0.97--1.06). DISCUSSION {#s3} ========== In this population-based study, we estimated the incidence of 4 major conditions---diabetes, hypertension, hyperlipidemia, and IHD---in MS and reference cohorts. Consistent with expectations for the Canadian population,^[@R2]^ the crude incidence rates of all comorbidities increased with age in the matched populations, and were higher in men than women. Findings were similar in the MS population. The incidence of vascular comorbidity varied across provinces, as previously observed in the Canadian population,^[@R13]^ likely reflecting differences in health behaviors and social factors. The incidence of diabetes and hyperlipidemia rose in the MS and matched populations over the 10-year study period. Although the rising rates of diabetes in the matched populations were consistent with those of the Canadian population,^[@R14]^ the incidence rates of diabetes and hyperlipidemia rose faster in the MS population. In 2005, the crude incidence of diabetes in the MS population was 765.9 per 100,000 persons, slightly exceeding that of the matched population that year (740.7/100,000). Our estimate falls within the bounds of 2 prior studies, although their estimates of the incidence of diabetes in persons with MS varied between 1 and 1,010 per 10,000 persons.^[@R1]^ One Danish study reported the incidence of hypertension to be 373 per 100,000 population in an incident MS cohort,^[@R15]^ an estimate substantially lower than we observed. However, the Danish study used data from hospitalizations rather than outpatient care for much of the study period, potentially underestimating the incidence of hypertension. In 2005, we found that the crude incidence of IHD in the MS population was 690.5 per 100,000 persons, consistent with our prior report in Manitoba for an incident, rather than prevalent, MS cohort.^[@R16]^ Two Nordic studies reported the incidence of myocardial infarction in incident MS cohorts to range from 236 to 275 per 100,000 population (reviewed in [@R1]). These lower estimates reflect the restriction of cases to myocardial infarction, and may also reflect worldwide variation in the burden of IHD.^[@R17]^ Over the 10-year study period, the incidence of diabetes did not differ between the MS and matched populations, while the incidence rates of hypertension and of hyperlipidemia were lower in the MS population, but these associations changed over time. The incidence of diabetes rose 3 times faster in the MS than in the matched population over time, such that by 2005 the incidence of diabetes was higher in the MS population. This rising incidence might reflect improved ascertainment in the MS population over time as compared to the general population. However, this should have also been expected for the other comorbidities studied, and yet the findings clearly differed among them. Consistent with the aforementioned Danish study,^[@R15]^ we found that the incidence rate of hypertension remained lower in the MS population than the matched population throughout the study period. However, this difference was primarily due to a lower incidence of hypertension for persons with MS aged ≥60 years. Over the 10-year study period, the incidence of IHD did not differ between populations, but age-specific differences existed. Persons with MS in the lowest age group had a substantially increased risk of IHD that was attenuated at older ages. One possible explanation may be more frequent health care contacts for those with MS leading to earlier or more frequent diagnosis of IHD. However, physician visits have declined faster in the MS population over time than in the general population,^[@R18]^ and most of the Canadian general population has regular contact with the health system. Further, this was only observed for IHD in younger persons with MS. True differences in risk may also exist. Such was suggested by previous findings that the risk of myocardial infarction was increased during the first year after MS diagnosis (IRR = 1.84; 95% CI 1.28--2.65) in an incident cohort.^[@R15]^ The associations between MS and IHD may reflect shared environmental and genetic factors, or be secondary to MS. Smoking and obesity are risk factors for MS^[@R3]^ and IHD,^[@R19]^ and physical inactivity after MS onset may also increase the risk of IHD. Persons with MS are more likely to smoke, be overweight or obese, and be less physically active than the general population.^[@R3]^ Other immune-mediated diseases such as rheumatoid arthritis increase the risk of IHD independent of vascular risk factors including diabetes, hypertension, hyperlipidemia, and smoking, presumably due to inflammation.^[@R20]^ HLA alleles are associated with the risk of MS and are also implicated in the increased risk of IHD in rheumatoid arthritis^[@R21]^; such associations should be evaluated in MS. The rising incidence of comorbidity with age and over time is highly relevant to clinical care in MS. Growing evidence suggests that vascular comorbidities are associated with more rapid disability progression in patients with MS.^[@R3]^ Risks of adverse effects for some therapies are also increased in the presence of comorbidities such as cardiac disease or diabetes; thus the risks of therapies may change as individuals develop new comorbidities. As some disease-modifying therapies also increase the risk of comorbidities such as hypertension, it is important to determine the incidence of vascular comorbidities before these therapies are widely used, so as to better identify their effects in real-world safety and effectiveness studies, beyond clinical trials. Limitations of this study should be considered. It was not feasible to include all Canadian provinces; however, we captured 43% of the Canadian population. Administrative data are not collected for research purposes, but we have previously validated our case definitions for the comorbidities studied.^[@R10]^ The need to suppress small cell sizes for privacy requirements and the naturally lower incidences in younger individuals meant our age groups were broad. Nonetheless, our study provides the age-specific incidence estimates lacking in the literature. We did not evaluate the reasons for the observed differences between populations, or the potentially complex relationships between these comorbidities and health behaviors. Finally, we lacked information regarding clinical characteristics of the MS population such as subtype and disease duration and could not evaluate associations between clinical characteristics and the incidence of vascular comorbidities. Within the MS population, as in the general population, men and those at older ages are at increased risk of developing vascular comorbidities. This will become increasingly important as the peak prevalence of MS shifts to older ages.^[@R9]^ The rising incidence rates of diabetes and hyperlipidemia are concerning given their potential adverse effects on outcomes.^[@R3]^ Programs to systematically prevent and treat these conditions in the MS population are needed and will require specific attention to health behaviors, using collaborative models of care. Supplementary Material ====================== ###### Data Supplement ###### Coinvestigators Supplemental data at [Neurology.org/cp](http://cp.neurology.org/lookup/doi/10.1212/CPJ.0000000000000230) The results and conclusions presented are those of the authors. No official endorsement by Manitoba Health, Population Data BC, or the Regie d\'Assurance Maladie du Quebec is intended or should be inferred. Some data used in this report were made available by Health Data Nova Scotia of Dalhousie University. Although this research uses data obtained from the Nova Scotia Department of Health and Wellness, the observations and opinions expressed of those of the authors and do not represent those of either Health Data Nova Scotia or the Department of Health and Wellness. The authors thank the following contributors: Patricia Caetano, PhD (University of Manitoba, policy consultant); Nicholas Hall, BSc (University of Manitoba, study coordinator); Feng Zhu, PhD (University of British Columbia, analytic support); Elaine Kingwell, PhD (University of British Columbia, study coordination support), Karen Stadnyk, MSc (Dalhousie University, study coordinator), Yan Wang (Dalhousie University, analytic support), Bin Zhu (McGill University, analyst); Aruni Tennakoon (University of Manitoba, analyst); Stella Leung (University of Manitoba, analyst). AUTHOR CONTRIBUTIONS ==================== The corresponding author (RAM) takes responsibility for the integrity of the data and the accuracy of the data analysis. The analysts and principal investigators at each site had full access to the data at each site (British Columbia: Helen Tremlett, Aruni Tennakoon, Stella Leung; Manitoba: Ruth Ann Marrie, Aruni Tennakoon; Quebec: Christina Wolfson, Bin Zhu; Nova Scotia: John Fisk, Yan Wang). Ruth Ann Marrie, John Fisk, Christina Wolfson, Helen Tremlett, and Sharon Warren designed the study and obtained funding. All authors contributed to the interpretation of the data. Ruth Ann Marrie drafted the manuscript. All authors revised the manuscript and approved of the final version to be published. STUDY FUNDING ============= Supported in part by the Canadian Institutes of Health Research (CIBG 101829), the Rx & D Health Research Foundation, a Don Paty Career Development award from the Multiple Sclerosis Society of Canada (to RAM), and Manitoba Research Chair from Research Manitoba (to RAM). The funding sources had no role in the study design, collection, analysis, or interpretation of the data, or in the decision to submit the article for publication. DISCLOSURES =========== R.A. Marrie serves on the editorial boards of *Neurology*® and *Multiple Sclerosis Journal* and has received research support from Sanofi-Aventis, Canadian Institutes of Health Research (CIHR), Public Health Agency of Canada, Research Manitoba, Multiple Sclerosis Society of Canada, National Multiple Sclerosis Society, Multiple Sclerosis Scientific Foundation, Rx & D Health Research Foundation, and Consortium of Multiple Sclerosis Centers. J. Fisk has received travel expenses and honorarium from EMD Serono for workshop participation; has received research support from CIHR, MS Society of Canada, and National MS Society; and receives license fee payments for a Fatigue Impact Scale and Daily Fatigue Impact Scale (MAPI Research Trust). H. Tremlett is funded by the Multiple Sclerosis Society of Canada (Don Paty Career Development Award) and is a Michael Smith Foundation for Health Research Scholar and the Canada Research Chair for Neuroepidemiology and Multiple Sclerosis. She has received research support from the National Multiple Sclerosis Society, the Canadian Institutes of Health Research, and the UK MS Trust, and speaker honoraria and/or travel expenses to attend conferences from the Consortium of MS Centres, the National MS Society, Bayer Pharmaceuticals, Teva Pharmaceuticals, ECTRIMS, UK MS Trust, the Chesapeake Health Education Program, US Veterans Affairs, Novartis Canada, Biogen Idec, and the American Academy of Neurology. Unless otherwise stated, all speaker honoraria are either donated to an MS charity or to an unrestricted grant for use by her research group. C. Wolfson has received a speaker honorarium from Novartis; serves on the editorial board of the *Journal of Military, Veteran and Family Health*; and receives research support from CIHR, Canada Foundation for Innovation, National Multiple Sclerosis Society, and Multiple Sclerosis Society of Canada. S. Warren receives research support from CIHR and has provided an expert opinion paper on trauma and MS on behalf of a legal firm. J. Blanchard receives research support from the Multiple Sclerosis Society of Canada, CIHR, BMGF, Bill & Melinda Gates Foundation, Canadian International Development Agency, and the United States Agency for International Development. S.B. Patten has received honoraria for reviewing investigator-initiated grant applications submitted to Lundbeck and Pfizer; has received speaker honoraria from Teva and Lundbeck; serves as Editor-in-Chief for the *Canadian Journal of Psychiatry*, as Senior Associate Editor for *Epidemiology and Psychiatric Sciences*, and on the editorial advisory boards of *Chronic Diseases in Canada* and *Clinical Practice and Epidemiology in Mental Health*; receives publishing royalties for *Epidemiology for Canadian Students* (Brush Education, 2015); is the recipient of a salary support award (Senior Health Scholar) from Alberta Innovates, Health Solutions; and receives research support from CIHR, Hotchkiss Brain Institute, Pfizer Canada, Alberta Innovates/Health Solutions, Alberta Health Services, and the Alberta Collaborative Research Grants Initiative. Full disclosure form information provided by the authors is available with the [**full text of this article at Neurology.org/cp**](http://cp.neurology.org/lookup/doi/10.1212/CPJ.0000000000000230).  [^1]: Funding information and disclosures are provided at the end of the article. Full disclosure form information provided by the authors is available with the [**full text of this article at Neurology.org/cp**](http://cp.neurology.org/lookup/doi/10.1212/CPJ.0000000000000230). The Article Processing Charge was paid by Canadian Institutes for Health Research. [^2]: Coinvestigators are listed at [Neurology.org/cp](http://cp.neurology.org/lookup/doi/10.1212/CPJ.0000000000000230).

Introduction {#s1} ============ Improper targeting of membrane proteins causes many diseases. Often point mutations to cysteine hinder the delivery of membrane proteins to the cell surface [@pone.0047693-Peng1], [@pone.0047693-Comoletti1], [@pone.0047693-Koeppen1], [@pone.0047693-Liu1], [@pone.0047693-Bartoli1], [@pone.0047693-Reuter1], [@pone.0047693-Dhermy1], or to the correct side of polarized cells [@pone.0047693-Toye1], [@pone.0047693-Rungroj1]. Because cysteine is a readily reactive amino acid, in principle it should be possible to recover proper trafficking by modifying its chemical structure in order to mimic the side chain of the wild type amino acid. As a proof of principle, we have studied two naturally occurring cysteine mutations in a cyclic nucleotide-gated channel (CNGA3) responsible for hereditary cone photoreceptor disorders: Y181C linked to incomplete achromatopsia and R277C linked to complete and incomplete achromatopsia or cone dystrophy [@pone.0047693-Wissinger1], [@pone.0047693-Patel1]. We have chosen these mutations because proper surface CNG channel expression can be easily assayed using electrophysiological techniques, and because both mutations, which cause channel retention in the endoplasmic reticulum (ER) [@pone.0047693-Patel1], [@pone.0047693-Faillace1], change wild type amino acids of drastically different chemistries. CNG channels open a cationic selective permeation pathway in response to intracellular cyclic nucleotides [@pone.0047693-Zagotta1], [@pone.0047693-Kaupp1]. In the visual system, CNG channels are key players in the transduction of light into electrical signals [@pone.0047693-Stryer1]. In native cells, these channels are formed by the coassembly of four homologous subunits [@pone.0047693-Peng2], [@pone.0047693-Zhong1], [@pone.0047693-Zheng1], [@pone.0047693-Zheng2], [@pone.0047693-Weitz1], [@pone.0047693-Pages1], each containing six transmembrane segments. Functional homotetramers can be formed by the CNGA1, A2 or A3 subunits [@pone.0047693-Bonigk1], [@pone.0047693-Dhallan1], [@pone.0047693-Kaupp2], and these channels are usually studied as homotetramers in heterologous systems. We have introduced both achromatopsia-related cysteines in a cysteine-less CNGA1 channel [@pone.0047693-Flynn1], and used them as a target for specific chemical modification with hydroxybenzyl- (MTSHB) and aminoethyl-methanethiosulfonate (MTSEA). These reagents readily attach to the side chain of cysteines and mimic the chemistry of tyrosine and arginine, respectively ([**Fig. 1**](#pone-0047693-g001){ref-type="fig"}). Although Y181C and R277C caused ER retention, after chemical modification both mutants were targeted to the cell surface, providing a unique opportunity for their functional characterization. {#pone-0047693-g001} Materials and Methods {#s2} ===================== Mutagenesis and Expression {#s2a} -------------------------- cDNA of a cysteine-less CNGA1 channel was kindly provided by William Zagotta (University of Washington, Seattle, WA). Cysteine mutations were introduced in this background using a QuickChange kit (Stratagene). Amino acid substitutions, as well as the integrity of the entire coding region of each channel, were confirmed by DNA sequencing (NINDS sequencing facility). A CNGA1-Green Fluorescent Protein (CNG-GFP) was created with standard PCR techniques. cRNAs were synthesized with a T7 promoter-based *in vitro* transcription protocol (Ambion). *Xenopus* oocytes were injected with 50 nl (500 ng/µl) of cRNA and incubated in ND96 solution (in mM: 96 NaCl, 2 KCl, 1 MgCl~2~, 1.8 CaCl~2~, 5 HEPES, pH 7.6) at17°C for two to three days to allow channel to express. To assess surface expression by fluorescence, we engineered two cysteine-less background constructs: CNGA1-GFP and CNGA1-FLAG. GFP and the FLAG epitope (DYKDDDDK) were inserted in frame immediately before the stop codon. In general, CNG channels tolerate these tags at the carboxy-terminal remarkably well [@pone.0047693-Peng1], [@pone.0047693-Zheng1], [@pone.0047693-Zheng2]. This study was approved by National Institute Neurological Disorders and Stroke/National Institute on Deafness and Other Communication Disorders Animal Care and Use Committee Protocol Number 1253-09. Chemical Treatment {#s2b} ------------------ R272C mutant channels were rescued using aminoethyl-methanethiosulfonate (MTSEA), a compound which readily permeates the membrane of cells, including that of *Xenopus* oocytes, in its uncharged form [@pone.0047693-Holmgren1]. Treatments were performed 48 hrs after cRNA injection. MTSEA (final concentration 2 mM) was prepared in ND96 solution and added into wells containing oocytes. Incubation was performed at 17°C for six hours. Fresh ND96 solution with MTSEA was replaced every 30 minutes. These prolonged treatments were not readily tolerated by all oocytes. We found that after around 4 hours of MTSEA exposure, about 50% of the oocytes began to show signs of deterioration, as the animal pole became pale. Those oocytes were not used for any experiments. Y176C mutant channels were treated with hydroxybenzyl-methanethiosulfonate (MTSHB). Oocytes were injected once with 50 nl of a 40 mM MTSHB stock solution, which was prepared in an ethanol-DMSO mix (50/50), and incubated at 17°C overnight. After each treatment, some oocytes were used to analyze the extent of protein trafficking by fluorescence or immunocytochemistry, membrane protein expression assays, and others were used for electrophysiological characterization. Biotinylation of Membrane Surface Protein {#s2c} ----------------------------------------- Six to eight oocytes were incubated for 1 hr at 4°C in ND96, supplemented with 50 µg/ml gentamicin and 1.0 mg/ml sulfo-NHS-LC-biotin (Pierce). Then, oocytes were washed several times with ND96 supplemented with 100 mM glycine and lysed in 200 µl of buffer H (1% Triton X-100, 100 mM NaCl, 20 mM Tris-HCl, pH 7.4 with protease inhibitors (SIGMA) by trituration. Lysates were rocked at room temperature for 15 min and then centrifuged at 13,000 rpm for 3 min. The pellet was discarded and the supernatant was divided in two equal samples, one containing total proteins and the other to be used for preparation of cell membrane proteins. For cell membrane protein isolation, 50 µl of NeutrAvidinTM Agarose Resin (Thermo Scientific) was added to the sample and rocked gently at 4°C for at least 30 min. Resin was washed at least 6 times with buffer H, and finally eluted in buffer H (supplemented with 10% 2-ME and 50 mM DTT) to the same volume as the sample containing total proteins, and incubated for 5 min at 95°C. All samples were deglycosylated using PNGase F (New England BioLabs) before being loaded into a SDS-PAGE gel and transferred to a polivinylidene difuoride (PVDF) membrane. Blots were probed with an anti-GFP mouse monoclonal antibody at a dilution of 1∶5000 (Clontech). Primary antibodies were detected with a secondary, goat anti-mouse antibody conjugated to horseradish peroxidase used at a dilution of 1∶10000 (Pierce). Membranes were developed by SuperSignal WestFemto (Thermo Scientific) and visualized by chemioluminiscence using a FluorChem E Imager (Cell Biosciences). Analysis was performed using Image AlphaView software (Cell Biosciences). Immunofluorescent Staining {#s2d} -------------------------- The oocyte's vitelline layer was removed to reduce background fluorescence. Oocytes were permeabilized with 0.03% saponin in ND96 solution for 20 min, washed with ND96 solution, blocked with 1% BSA in ND96 for 30 min. and incubated overnight with a 1∶250 dilution of a FLAG polyclonal antibody (Santa Cruz Biotechnology. Inc.) in ND96 at 4°C. Oocytes were washed several times with ND96 and incubated for 1 hr in a 1∶500 dilution of secondary antibody (Texas Red conjugated donkey anti-goat IgG; Santa Cruz Biotechnology. Inc.) at 4°C. After several rinses with ND96, oocytes were imaged using a Zeiss LSM 510 confocal microscope. Electrophysiology {#s2e} ----------------- The recording solution consisted of (in mM): 120 NaCl, 2 EDTA, 10 HEPES (pH = 7.4). All reagents were obtained from SIGMA. Currents from inside-out excised patches [@pone.0047693-Perozo1] were acquired using an Axopatch 200B amplifier (Molecular Devices) and a Digidata 1322 acquisition board (Molecular Devices), and were sampled between 2.5 to 10 kHz using a low-pass filter at 1 or 2 kHz. Patch electrodes with tip diameters between 8 and 15 µm were made with borosilicate glass pipettes. Macroscopic data analysis was performed with pClamp 9 (Molecular Devices) and Origin 8 (Microcal Software) software. Solutions were changed using a computer-controlled rapid solution changer (RSC-200; Biologic Science Instruments). Results {#s3} ======= Recovering Cell Membrane Expression of CNG Channels Containing the R272C Mutation {#s3a} --------------------------------------------------------------------------------- To assess whether chemical modification could restore cell surface expression and functionality, we used the well-characterized bovine CNGA1 channel which lacks all native cysteines [@pone.0047693-Matulef1]. Position R272 in the bovine CNGA1 channel is equivalent to position R277 in the human CNGA3 channel and is located within a large domain known as the voltage sensor. From the crystal structure of a mammalian voltage-activated potassium (K~V~) channels [@pone.0047693-Long1], a cousin of CNG channels, this position is part of the fourth transmembrane segment (S4) where the critical charges that sense the transmembrane voltage are spaced every three amino acids [@pone.0047693-Aggarwal1], [@pone.0047693-Papazian1], [@pone.0047693-Seoh1]. Little is known about the role of the voltage sensor in CNG channels due to technical difficulties, such as poor surface expression of channels harboring S4 mutations [@pone.0047693-Liu1], [@pone.0047693-Faillace1]. In all channels within the superfamily of voltage-activated ion channels, which includes CNG channels, it is known that S4 is important for maturation [@pone.0047693-Liu1], [@pone.0047693-Faillace1], [@pone.0047693-Papazian2], [@pone.0047693-Deutsch1], [@pone.0047693-Mannikko1]. To verify that the R272C mutation impedes surface expression in the cysteine-less CNGA1-GFP background, we injected this construct into *Xenopus* oocytes. After two days, we could not observe channels at the cell surface using fluorescence microscopy. Further, using functional expression as an indicator, we were unsuccessful in detecting cGMP-activated ionic currents in more than 40 excised inside-out membrane patches (not shown). However, by incubating injected oocytes with a solution containing 2 mM MTSEA (a membrane permeant reagent [@pone.0047693-Holmgren1] that leaves a moiety mimicking the side chain of arginine, as shown in [Fig. 1](#pone-0047693-g001){ref-type="fig"}) we restored the proper channel trafficking. [Fig. 2A](#pone-0047693-g002){ref-type="fig"} shows confocal images of an oocyte in which R272C CNGA1-GFP channels were rescued after ∼6 hr of MTSEA treatment. [Fig. 2B](#pone-0047693-g002){ref-type="fig"} shows the time course of fluorescence detection at the cell surface of six oocytes. A similar MTSEA treatment to cysteine-less CNGA1-GFP channels has no effect on cell surface expression ([Fig. S1](#pone.0047693.s001){ref-type="supplementary-material"}). {#pone-0047693-g002} Modified channels respond normally to the presence of agonists. [Fig. 3A](#pone-0047693-g003){ref-type="fig"} shows cGMP-activated currents carried by control MTSEA-treated cysteine-less CNGA1-GFP channels in response to voltage steps between −80 and +80 mV from a holding potential of 0 mV in the presence of saturating \[cGMP\]. Under similar conditions, ionic currents from MTSEA-modified R272C mutant channels are comparable ([Fig. 3D](#pone-0047693-g003){ref-type="fig"}). Dose-response curves for cGMP at +60 mV for wild-type channels ([Fig. 3B](#pone-0047693-g003){ref-type="fig"}) and modified R272C channels ([Fig. 3E](#pone-0047693-g003){ref-type="fig"}) were also similar. Solid lines represent Hill equation fits in which K½ and n values were 19±1 µM and 2.2±0.2 for wild-type channels and 43±0.5 µM and 1.9±0.3 for modified 272C channels. Another property of CNGA1 channels is their sensitivity to saturated concentrations of the various cyclic nucleotide agonists. In general they barely open with cAMP, open more with cIMP and open with a high probability with cGMP. Both, cysteine-less CNGA1-GFP ([Fig. 3C](#pone-0047693-g003){ref-type="fig"}) and modified R272C ([Fig. 3F](#pone-0047693-g003){ref-type="fig"}) channels maintained the same relative efficacy among these agonists. Taken together, these results demonstrate that attaching a primary amine moiety to cysteine 272 can successfully mimic the role played by the side chain of arginine in wild-type channels. Specifically, this modification restores targeting to the cell surface and produces channels that function relatively normal. ![Functional characterization of CNGA1-GFP channels at the cell surface.\ A, B & C. Cysteine-less CNGA1-GFP channels. A, Ionic currents in the presence of saturating \[cGMP\] (2 mM). cGMP-activated currents shown were acquired in response to 100 ms voltage steps from −80 to +80 mV (every 20 mV) from a holding potential of 0 mV. B. Dose response for cGMP at +60 mV. Solid line represents a normalized Hill equation fit to the data. The best fit parameter values for K½ was 19±1 µM, and for the hill coefficient (n) was 2.1±0.2. Average values for K½ and n were 24±3 µM and 2.16±0.46, respectively (n = 4 oocytes). C. Efficacies for different agonists. Wild type CNGA1 channels displayed larger than 95% maximal probability of opening with saturated concentrations of cGMP, less with cIMP, and much less with cAMP. Nucleotide-activated current records shown were obtained from the same excised inside out patch with saturating concentrations of cGMP (2 mM; *black*), cIMP (16 mM; *red*) and cAMP (16 mM; *blue*). D, E & F. Rescued R272C CNGA1-GFP channels. D. Ionic current carried by MTSEA-modified R272C channels. In response to a comparable experimental protocol as in A, rescued channels were able to conduct ionic current with similar properties as wild type channels. E. Dose response of modified R272C channels at the cell surface for cGMP. Solid line corresponds to a normalized Hill equation fit to the data. The best fit parameter values for K½ and n were 43.4±2.5 µM and 1.9±0.3, respectively. Average values for K½ and n were 59±10 µM and 1.48±0.22, respectively (n = 14 oocytes). F. Agonist efficacies for rescued R272C mutant channels. Nucleotide-activated current records shown were acquired from the same patch, using the equivalent agonist concentration as in C. Similar observations were made in 3 different patches.](pone.0047693.g003){#pone-0047693-g003} How efficiently can a MTSEA-modified R272C CNGA1-GFP channel be rescued? To approach this question, we assessed total and cell surface protein expression from pools of six to eight oocytes (see [Methods](#s2){ref-type="sec"}). For cysteine-less CNGA1-GFP channels, ∼45% of the total membrane protein is at the cell surface ([Fig. 4](#pone-0047693-g004){ref-type="fig"}; WT). Untreated R272C channels cannot be detected at the cell surface ([Fig. 4](#pone-0047693-g004){ref-type="fig"}; R272C), consistent with previous observations [@pone.0047693-Liu1], [@pone.0047693-Faillace1]. However, treatment with MTSEA successfully rescued R272C CNGA1 channels to the cell surface, at comparable proportions as cysteine-less CNGA1-GFP channels ([Fig. 4](#pone-0047693-g004){ref-type="fig"}; R272C+MTSEA). These results suggest that the added moiety to cysteine 272 restored the proper channel conformation allowing it to pass the various trafficking checkpoints. {#pone-0047693-g004} {ref-type="fig"}. D. Dose response for cGMP at +60 mV. Solid line represents a normalized Hill equation fit to the data. The best fit parameter values for K½ and n were 16±1 µM and 1.30±0.05, respectively. From a total of 8 oocytes, the average values for K½ and n were 16±3 µM and 1.6±0.1, respectively. E. Agonist efficacies of rescued Y176C mutant channels. Nucleotide-activated current records shown in the presence of saturating concentrations of each agonist (same as [Fig. 3C](#pone-0047693-g003){ref-type="fig"}) were acquired from the same excised patch. Rescued Y176C channels showed a similar efficacy pattern as cysteine-less CNGA1-GFP channels. Similar observations were observed in 4 patches.](pone.0047693.g005){#pone-0047693-g005} Our approach was also successful with other positively charged residues. Unlike the case for K~V~ channels, the substitution of cysteine for the positively charged, voltage sensing amino acids within the S4 transmembrane segment of CNG channels leads to immature products that get trapped in the ER [@pone.0047693-Faillace1]. We tested the ability of MTSEA to rescue surface expression of R269C, R275C and R278C. In each case, mutants were efficiently redirected to the cell membrane, demonstrating the general applicability of the technique ([Fig. S2](#pone.0047693.s002){ref-type="supplementary-material"}). Restoring Proper Targeting to Y176C Mutant CNG Channels {#s3b} ------------------------------------------------------- Position Y176 in CNGA1 channels corresponds to position Y181 in hCNGA3 channels and is located within the first transmembrane segment. As in hCNGA3 channels [@pone.0047693-Patel1], the Y176C mutation impeded Y176C CNGA1-GFP channels from reaching the cell membrane ([Fig. 5A](#pone-0047693-g005){ref-type="fig"}). Chemical modification with MTSHB would attach a moiety to these cysteines that resembles the original tyrosine side chain ([Fig. 1A](#pone-0047693-g001){ref-type="fig"}). This reagent, however, is very hydrophobic and precipitates in aqueous solutions. We were unable to find an experimental condition where we could incubate oocytes in MTSHB to restore function. However, by directly injecting a 40 mM MTSHB solution into the oocytes, we were able to recover cell surface expression ([Fig. 5B](#pone-0047693-g005){ref-type="fig"}). Likely, the rather oily environment of an oocyte's yolk allowed the reagent to stay in solution at sufficiently high concentrations to modify Y176C. A direct reagent injection has been successfully used before to probe proton channel function in a tryptophan mutant [@pone.0047693-Tang1]. Modified Y176C CNGA1-GFP channels responded normally to saturating concentrations of cGMP ([Fig. 5C](#pone-0047693-g005){ref-type="fig"}), although current levels were consistently smaller than cysteine-less and rescued R272C CNGA1 channels. Likely, a single MTSHB injection (c.f. freshly applied MTSEA every 30 min for R272C CNGA1 channels) and the relatively fast hydrolysis of MTS reagents contributed to the lower levels of rescued Y176C CNGA1 channels. The dose-response for cGMP at +60 mV for modified Y176C CNGA1-GFP channels was characteristic of CNGA1 channels ([Fig. 5D](#pone-0047693-g005){ref-type="fig"}). The solid line through the data corresponds to a Hill equation fit with K½ and n values of 16±1 µM and 1.3±0.05. Finally, rescued channels showed a similar efficacy for saturating concentrations of different agonists ([Fig. 5E](#pone-0047693-g005){ref-type="fig"}) as for the cysteine-less CNGA1-GFP channels ([Fig. 3E](#pone-0047693-g003){ref-type="fig"}). These studies demonstrate that Y176C mutant channels retained in intracellular compartments can be successfully targeted to the cell surface by adding a side-chain to a cysteine that mimics that of tyrosine. Once at the cell surface, these channels behave normally. Discussion {#s4} ========== We describe a method to restore proper maturation and trafficking of membrane proteins that have been retained within intracellular organelles due to single point mutations to cysteine. Because the side chain of cysteine is highly reactive, we reasoned that modification with reagents that restored the original chemistry could drive proper maturation. We successfully restored both trafficking and normal function to CNGA1 mutant channels R272C and Y176C, both responsible for hereditary cone photoreceptor disorders [@pone.0047693-Wissinger1], [@pone.0047693-Patel1]. Maturation of any protein is a complex, multi-step process involving a network of intracellular proteins and organelles. Surely, all genetic mutations leading to defective maturation cannot be repaired by a single strategy. Thus far, a variety of experimental approaches have been shown to recover proper maturation. For example, cell surface expression of mutant HERG and CNGA3 channels [@pone.0047693-Reuter1], [@pone.0047693-Zhou1], as well as deficient lysosomal glucocerebrosidase [@pone.0047693-Sawkar1], can be restored simply by lowering the temperature of incubation. Based on their ability to stabilize proper folding conformations in the ER, drugs and lipid chaperones are emerging as new strategies to restore protein maturation [@pone.0047693-Perlmutter1], [@pone.0047693-Yu1], [@pone.0047693-Dowhan1], [@pone.0047693-vanKlompenburg1], [@pone.0047693-Duricka1]. As with other rescue methods, ours has disadvantages: it is restricted to cysteine mutants and it is not specific since MTS reagents will modify any accessible cysteine in a protein. Nevertheless, outside of therapeutics, this method has applications that could, in principle, offer relevant structural and functional information about diseases. For example, it could be used to better understand the chemical nature of the protein folding failure since many different MTS reagents are available that attach moieties resembling different amino acid side chains. Another potential use could be for kinetic studies of folding. For example, if a cysteine mutation impairs proper folding, modification reactions by MTS reagents are sufficiently fast to permit the temporal resolution of downstream conformational changes, providing kinetic information on folding steps. A third application we envision is to use the disease related mutant in the same way that biophysicists use engineered cysteines as a tool to study state dependent accessibility. This will provide information of conformational changes at the site of the cysteine mutation. Supporting Information {#s5} ====================== ###### **Cell surface expression time course of cysteine-less CNGA1-GFP channels exposed to MTSEA.** A. Representative confocal images of one oocyte before (T0) and after 2 mM MTSEA exposure. Media with fresh MTSEA was exchanged every 30 min. MTSEA treatment does not affect cell surface expression of "wild-type" channels. (TIF) ###### Click here for additional data file. ###### **Cell surface targeting of S4 cysteine mutants by chemical modification.** All images shown were obtained after immunocytochemical labeling of oocytes expressing CNGA1 arginine to cysteine mutations in the S4 transmembrane segment. The absence of fluorescence at the oocytes' cell membrane (left column) indicate that these arginine to cysteine mutations in the S4 segment render immature channels that are unable to reach the cell surface. After 6 h MTSEA treatment, we were able to restore proper trafficking to these mutant channels (right column). Representative of \>10 cells in each panel. (TIF) ###### Click here for additional data file. We thank Dr. William Zagotta for kindly supplying the CNGA1 cysteine-less construct, Dr. Joshua Rosenthal for helpful discussions, Deepa Srikumar for technical assistance, Pat Curran and Dan Silverman for assistance with western blot preparations, Dr. Jeet Kalia for preparing Fig. 1 and the DNA-sequencing facility of National Institute Neurological Disorders and Stroke where all DNA constructs were sequenced. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AL MH. Performed the experiments: AL. Analyzed the data: AL MH. Contributed reagents/materials/analysis tools: AL MH. Wrote the paper: AL MH.

Introduction {#sec1} ============ Age-related macular degeneration (AMD) is a leading cause of irreversible blindness among the elderly of industrialized countries, affecting approximately 50 million people worldwide.[@bib1], [@bib2] AMD presents itself as two distinct subtypes, dry and wet AMD. Dry AMD is characterized by geographic atrophy of the central retina, whereas wet AMD is identified by the formation of choroidal neovascularization (CNV).[@bib3] In wet AMD patients, development of CNV can lead to macular edema, which results in the destruction of retinal structures and irreversible loss of vision. Currently, the main strategy for the treatment of wet AMD is based on neutralizing vascular endothelial growth factor (VEGF), either as a monotherapy or in conjunction with photodynamic therapy.[@bib4], [@bib5] However, in addition to the economic burden of these repeated monthly treatments posing an obstacle to patient compliance,[@bib6] the more challenging issue of therapy resistance means a need exists for the development of a new, long-lasting, disease-modifying therapy. Mammalian target of rapamycin (mTOR) has been suggested as a therapeutic target for various human diseases, including cancer, type 2 diabetes, obesity, and neurologic disorders.[@bib7], [@bib8], [@bib9] mTOR is a serine-threonine protein kinase that exists in nearly every eukaryotic cell and is a master regulator of cellular metabolism and cell growth. The mTOR pathway signals through two distinct complexes, mTOR complex 1 (mTORC1) and mTORC2. mTORC1, which consists of RAPTOR, PRAS40, and mLST8 proteins, is sensitive to rapamycin and is responsible for cell growth and proliferation, angiogenesis, synthesis of protein and lipid, and autophagy regulation.[@bib10], [@bib11] mTORC2 contains RICTOR, PROCTOR1/2, and mLST8 as companion proteins and is closely related to the AKT pathway, because it phosphorylates AKT.[@bib11] While mTORC1 signaling is relatively well understood, the functions of mTORC2 have not been widely investigated, and only recently has its critical roles in cytoskeletal organization, regulation of cell survival, and promotion of neovascularization been revealed.[@bib12], [@bib13], [@bib14] Furthermore, deregulation of the mTOR signaling pathway leads to the development of several human diseases, because it is directly related to abnormalities in cell proliferation and survival processes.[@bib7] A number of studies have attempted to elucidate the role of the mTOR pathway in retinal cells. mTOR was shown to be a critical factor in the angiogenesis processes of various retinal pathologic conditions, including AMD, retinopathy of prematurity, and diabetic retinopathy.[@bib15], [@bib16], [@bib17] Other recent studies demonstrated that autophagy induced by mTOR inhibition promoted the survival of retinal pigment epithelial (RPE) cells under toxic conditions,[@bib18], [@bib19], [@bib20] indicating that the development of new therapeutic modalities for managing AMD may crucially include mTOR signaling pathway regulation. Recombinant adeno-associated viruses (rAAVs) have shown promise as a vector for the in vivo delivery of gene-based drugs, and recent clinical successes with AAV vectors in treating Leber congenital amaurosis suggest that they may serve as a suitable platform for treating various ocular diseases.[@bib21], [@bib22], [@bib23], [@bib24] Using mouse models, we previously demonstrated the enhanced transduction of intravitreally administered AAVs to various retinal cells, including RPE cells, upon retinal laser photocoagulation,[@bib25], [@bib26] showing that the transduction of AAV vectors may be enhanced by the inflammatory or pathological state of the retina. As such, the efficient transduction of AAV vectors may be possible in diseased retinas, even for deep-lying RPE cells. Recently, we designed a multispecies-compatible mTOR small interfering RNA (siRNA) sequence using a self-developed program, CAPSID (Convenient Application Program for siRNA Design),[@bib27] that directly interferes with the mTOR signaling pathway by blocking both mTORC1 and mTORC2, and demonstrated in vivo that antitumor effects can be achieved by this rAAV-delivered mTOR-inhibiting short hairpin RNA (shRNA) (rAAV-mTOR shRNA).[@bib28] Here, rAAV-mTOR shRNA was used to investigate whether mTOR inhibition suppresses laser-induced CNV in mouse chorioretinal tissue, and in addition to a substantial suppression of CNV, rAAV-mTOR shRNA-treated retinas exhibited decreased local inflammation and enhanced autophagic activity in CNV lesions. Results {#sec2} ======= Regression of CNV Leakage {#sec2.1} ------------------------- Fundus fluorescein angiography (FFA) was performed 5 days after laser photocoagulation to observe vascular leakage and the establishment of new vessels. This was repeated at 13 days after laser photocoagulation to observe the therapeutic effects of rAAV-mTOR shRNA, which was injected intravitreally 5 days after laser photocoagulation ([Figure 1](#fig1){ref-type="fig"}A). FFA 5 days post-laser photocoagulation confirmed the establishment of CNV lesions, seen as well-defined hyperfluorescent leaking spots ([Figures 1](#fig1){ref-type="fig"}B--1D). These persisted in animals injected with either 0.1% PBS or rAAV-scrambled shRNA ([Figures 1](#fig1){ref-type="fig"}E and 1F), whereas FFA images taken from those animals treated with rAAV-mTOR shRNA showed a marked regression of leakage from the CNV lesions ([Figure 1](#fig1){ref-type="fig"}G).Figure 1Experiment Protocol Schematic and Fundus Fluorescent Angiography Images Captured Pre- and Post-injectionAll mice underwent laser photocoagulation (LP) at day 0. FFA images were captured 5 days post-LP immediately before intravitreal viral vector administration. (A) 13 days post-LP, FFA was performed again and the animals were sacrificed 1 day later. (B--D) Pre-injection FFA exhibited dye leakage at laser spots for all three groups (B, control; C, scrambled; D, mTOR shRNA), indicating CNV progression. (E--G) Eight days after vector administration, images from the control (E) and rAAV-scramble shRNA-injected mice (F) retained fluorescence leakage around CNV lesions, whereas the rAAV-mTOR shRNA-treated group exhibited CNV regression (G). Transduction of Vector to Endothelial Cells in CNV {#sec2.2} -------------------------------------------------- Intravitreally administered self-complementary adeno-associated virus serotype 2 (scAAV2) vectors are known to transduce inner retinal cells in wild-type mice, including retinal ganglion cells and cells in the inner nuclear layer.[@bib25] This is confirmed here, and in addition to the inner retinal cells, CD31-positive endothelial cells in the laser-induced CNV lesions also exhibited the transduction of rAAV-mTOR shRNA expressing GFP ([Figure 2](#fig2){ref-type="fig"}).Figure 2Efficient Vector Transduction Exhibited in Transverse Retinal Sections(A--H) GFP expression was detectable throughout the retinal layers (A--D) (original magnification ×2 in E--H) from retinal ganglion cells to CD31-positive endothelial cells present in areas where choroidal neovascularization (CNV) was induced via laser photocoagulation (arrows in E and F), demonstrating the efficient transduction of rAAV-mTOR shRNA into cells of the outermost retinal layer. Scale bars, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. mTOR Expression in Laser-Induced CNV {#sec2.3} ------------------------------------ To determine the effect of the mTOR shRNA on mTOR activity, we first evaluated mTOR expression in all four groups of mice, including the negative control. Compared with normal chorioretinal tissue ([Figures 3](#fig3){ref-type="fig"}A--3C), mTOR expression was upregulated throughout the entire neural retina and the subretinal CNV lesion induced via laser photocoagulation ([Figures 3](#fig3){ref-type="fig"}D--3F). While this mTOR expression pattern did not change upon the intravitreal administration of rAAV-scrambled shRNA ([Figures 3](#fig3){ref-type="fig"}G--3I), it was, however, substantially reduced in the chorioretinal tissue of mice treated with rAAV-mTOR shRNA ([Figures 3](#fig3){ref-type="fig"}J--3L).Figure 3mTOR Expression Resulting from Laser-Induced CNV(A--L) Compared with normal chorioretinal tissue, mTOR expression was upregulated throughout the neural retina and subretinal CNV lesions for both 0.1% PBS-injected control eyes (A--F) and rAAV-scrambled shRNA-injected eyes (G--I), whereas mice treated with mTOR shRNA showed substantially reduced mTOR expression throughout (J--L). (M and N) Quantitative analyses were conducted using RT-PCR. mTOR mRNA levels were substantially increased upon the formation of CNV and subsequently reduced significantly upon the intravitreal administration of rAAV-mTOR shRNA. Data are presented as mean ± SEM. Asterisk indicates statistically significant (\*p \< 0.05) differences versus the control. Cross indicates statistically significant (^†^p \< 0.05) differences versus the rAAV-mTOR shRNA-treated group. Scale bar, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. qRT-PCR analyses showed that mTOR mRNA levels were substantially higher in retinas wherein CNV occurred, as compared with the control ([Figures 3](#fig3){ref-type="fig"}M and 3N), with significant reduction observed 14 days after rAAV-mTOR shRNA administration (p \< 0.05; n = 5 for each group). Anti-angiogenic Effect of rAAV-mTOR shRNA {#sec2.4} ----------------------------------------- The anti-angiogenic effect of the rAAV-mTOR shRNA was evaluated by immunostaining the areas where CNV occurred with phalloidin and anti-CD31. We found that animals treated with rAAV-mTOR shRNA exhibited markedly reduced areas of CNV activity when compared with the control group and mice treated with rAAV-scrambled shRNA (p = 0.028 and p = 0.026, respectively; n = 5 for each group) ([Figure 4](#fig4){ref-type="fig"}). To more precisely analyze the therapeutic effects of rAAV-mTOR shRNA treatment, we evaluated transverse-sectioned chorioretinal samples of the laser-induced CNV lesions for CD31-positive staining, which showed that, compared with the two other experimental groups, CD31-positive signals were markedly reduced in mice treated with rAAV-mTOR shRNA ([Figure 5](#fig5){ref-type="fig"}).Figure 4Immunohistochemistry of Phalloidin and CD31 for Endothelial Cells Resulting from Laser-Induced CNVThe extensiveness of the CNV was measured via whole-mount phalloidin (red) and CD31 (yellow) immunostains. (A--I) Whereas the areas where CNV occurred were widespread for the control and scrambled shRNA-treated groups (A--F), CNV was substantially reduced for mice treated with mTOR shRNA (G--I). Bar graph shows CNV extent, as measured by pixels per lesion, using the Kruskal-Wallis test with post hoc analysis (J, mean ± SEM). Asterisk indicates statistically significant differences versus the control (\*p \< 0.05). Cross (†) indicates statistically significant differences versus the scrambled shRNA-treated group. Scale bars, 100 μm.Figure 5Transverse Retinal Sections Immunolabeled with CD31 and GFP(A--L) CD31 (red) expression was detectable throughout the laser-induced CNV lesions in the retinas of the control- (A--D) and rAAV-scrambled shRNA-treated groups (E--H), while CD31 expression was markedly reduced in the rAAV-mTOR shRNA-treated group (I--L). (F and J) Substantial GFP (green) expression exhibited in the rAAV-scrambled shRNA- and rAAV-mTOR shRNA-treated groups (F and J) indicated the effective transducing qualities of the viral vector. Scale bar, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. Anti-inflammatory Effect of rAAV-mTOR shRNA {#sec2.5} ------------------------------------------- Inflammatory cells, such as macrophages, monocytes, and microglia, play major roles in the development of AMD, particularly when CNV is implicated. As such, we evaluated whether mTOR inhibition via rAAV-mTOR shRNA modulates the proliferation and/or infiltration of inflammatory cells in the development and maturation of laser-induced CNV using anti-CD11b and -F4/80 antibodies on transverse sections of the retina 14 days after laser treatment. Abundant CD11b- and F4/80-positive cells were observed around CNV lesions in the subretinal and intraretinal regions of the mice injected with 0.1% PBS and rAAV-scrambled shRNA ([Figures 6](#fig6){ref-type="fig"}A--6F and 6J--6O). However, significantly less inflammatory cell infiltration was observed in retinas treated with rAAV-mTOR shRNA when compared with the control group and with mice treated with rAAV-scrambled shRNA (CD11b: p = 0.036 and p = 0.016, respectively; F4/80: p = 0.027 and p = 0.022, respectively; n = 5 for each group; [Figures 6](#fig6){ref-type="fig"}G--6I and 6P--6R). Specifically, the number of F4/80-positive cells was 42.4 ± 10.4 in rAAV-mTOR shRNA-treated retinas, 82.8 ± 10.0 in rAAV-scrambled shRNA-treated retinas, and 84.4 ± 17.0 in the untreated group; the number of CD11b-positive cells was 90.0 ± 11.6, 127.6 ± 14.4, and 123.8 ± 13.0, respectively ([Figures 6](#fig6){ref-type="fig"}S and 6T).Figure 6Representative Confocal Images of CD11b and F4/80 Immunostaining and Quantitative Analyses of Inflammatory Cell Infiltration(A--F and J--O) Immunohistochemistry with CD11b and F4/80 revealed the extensive presence of inflammatory cells in the retinas of mice injected with 0.1% PBS and rAAV-scrambled shRNA. (G--I and P--R) In contrast, significantly fewer CD11b- and F4/80-positive inflammatory cells were found in the animals treated with rAAV-mTOR shRNA. (S and T) Significantly fewer CD11b- (S) and F4/80-positive cells (T) were seen in the rAAV-mTOR shRNA-treated group when compared with the control and rAAV-scrambled shRNA-treated groups (CD11b: p = 0.036 and p = 0.016, respectively; F4/80: p = 0.027 and p = 0.022, respectively). Data are presented as mean ± SEM. Asterisk indicates statistically significant differences versus the control (\*p \< 0.05). Cross (†) indicates statistically significant differences versus the rAAV-scrambled shRNA-treated group. Scale bars, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. Autophagy in Laser-Induced CNV {#sec2.6} ------------------------------ To evaluate whether the induction of autophagy via mTOR inhibition is involved in CNV regression, we immunostained chorioretinal tissues for the autophagy markers LC3B and ATG7. Activation of autophagy was detected via both LC3B and ATG7 immunostaining 14 days after laser photocoagulation in the CNV lesions of retinas treated with rAAV-mTOR shRNA. In 0.1% PBS and rAAV-scrambled shRNA-treated mice, only background LC3B and weakly positive ATG7 signals were observed ([Figure 7](#fig7){ref-type="fig"}).Figure 7Immunolabeling for Autophagy Using Autophagy Markers LC3B and ATG7(G--I and P--R) Immunostaining against LC3B and ATG7 was used to detect autophagy, which was mainly found in the rAAV-mTOR shRNA-treated group. (A--F and J--O) On the other hand, only background LC3B and weak, lesion-confined ATG7 expression were detected in the control (A--C and J--L) and rAAV-mTOR scrambled-treated groups (D--F and M--O). Scale bars, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. Cell Apoptosis around Laser-Induced CNV {#sec2.7} --------------------------------------- At 14 days after laser photocoagulation, the terminal deoxynucleotidyl transferase-mediated biotinylated deoxyuridine triphosphate (dUTP) nick end labeling (TUNEL) assay was used to determine the number of apoptotic cells observed in the outer nuclear layer (ONL) and around the CNV of all three experimental groups. Significantly fewer TUNEL-positive cells were found in the ONL of rAAV-mTOR shRNA-treated retinas when compared with mice injected with 0.1% PBS and rAAV-scrambled shRNA (p = 0.038 and p = 0.024, respectively; n = 5 for each group). Specifically, the number of TUNEL-positive cells was 8.4 ± 3.0 for rAAV-mTOR shRNA-treated retinas, 19.4 ± 4.0 for rAAV-scrambled shRNA-treated retinas, and 17.8 ± 4.8 in untreated control retinas ([Figure 8](#fig8){ref-type="fig"}).Figure 8Analyses of Apoptotic Cells via TUNEL Assay(A--I) TUNEL-positive cells were found in the outer nuclear layer (ONL) and around the CNV lesions of the control (A--C), rAAV-scrambled shRNA-treated (D--F), and rAAV-mTOR shRNA-treated groups (G--I), with the latter containing noticeably fewer apoptotic cells. (J) As seen in the summarized data, significantly fewer TUNEL-positive cells were detected in the ONL of rAAV-mTOR shRNA-treated retinas when compared with the control and rAAV-scrambled shRNA-treated groups (p = 0.038 and p = 0.024, respectively). Data are presented as mean ± SEM. Asterisk indicates statistically significant differences versus the control (\*p \< 0.05). Cross (†) indicates statistically significant differences versus the rAAV-scrambled shRNA-treated group. Scale bars, 100 μm. ONL, outer nuclear layer; RPE, retinal pigment epithelium. Discussion {#sec3} ========== Building upon previous research targeting the mTOR signaling pathway as a therapeutic for various diseases, a number of studies have investigated the role of mTOR in the pathological conditions of the retina.[@bib15], [@bib16], [@bib17], [@bib19], [@bib29], [@bib30] Using mTOR pathway inhibitors, recently it has been shown that blocking mTOR signaling conferred protection against the formation and progression of CNV.[@bib16], [@bib18], [@bib19], [@bib30] Most of these studies, however, utilized rapamycin and/or temsirolimus, a rapamycin analog, both of which inhibit mTORC1 while activating mTORC2 via a negative feedback loop. These opposing effects hinder the comprehensive investigation of the therapeutic potential of mTOR inhibition. In comparison, we were here able to block both mTORC1 and mTORC2 simultaneously by utilizing siRNA to selectively target mTOR itself. To the best of our knowledge, the use of these siRNAs, incorporated into rAAV as an shRNA, represents the first study to simultaneously affect both mTORC complexes to demonstrate the therapeutic effects of mTOR inhibition on CNV development. We found that intravitreally administered rAAV-mTOR shRNA successfully transduced various cells in CNV lesions and suppressed the progression of CNV. Significantly elevated autophagy levels were detectable in rAAV-mTOR shRNA-treated mouse retinas when compared with animals injected with an rAAV-scrambled shRNA or 0.1% PBS as a control. Furthermore, the retinas of rAAV-mTOR shRNA-treated mice exhibited markedly reduced apoptosis activity. Among the well-established roles of mTOR are the regulation of cellular metabolism and cell growth during development or after injury. Neuronal mTOR activity decreases over the course of development in the murine CNS, and only limited activity is observable in the adult CNS.[@bib31], [@bib32] mTOR expression is restricted to the ganglion cell layer and inner nuclear layer (INL) of normal mouse retinas,[@bib33], [@bib34] and our results correspond with these earlier findings. Furthermore, we confirm here that mTOR expression is significantly increased in the CNV-induced retinas, particularly the inner plexiform retinal layer and the CNV lesion itself. Because mTOR immunoreactivity was shown to be downregulated in normal mouse retinas, we speculate that shRNA-mediated mTOR inhibition does not affect the viability of normal retinal cells, and that it has inhibitory effects only upon cells with upregulated mTOR expression. One of the major novel findings in this study is that intravitreally administered rAAV-mTOR shRNA successfully transduced cells in CNV lesions in mouse retinas induced via laser photocoagulation. Although numerous studies have shown that the AAV-mediated delivery of therapeutic genes may inhibit CNV formation,[@bib35], [@bib36], [@bib37] none has demonstrated the direct transduction of AAV into cells of CNV lesions. Here, we report that rAAV-mTOR shRNA effectively transduces endothelial cells, the main cellular components of CNV. Combined with other results, we posit that the inhibition of the mTOR pathway in endothelial and inflammatory cells may play a crucial role in limiting CNV formation, with long-lasting therapeutic effects achievable via a single intravitreal administration of rAAV-mTOR shRNA capable of directly transducing cells in the CNV lesions themselves. Several pathophysiological factors have previously been linked with AMD, primarily inflammation and angiogenesis.[@bib38], [@bib39] A number of previous studies have documented that both mTORC1 and mTORC2 are involved in angiogenesis and pro-inflammatory processes, and that the inhibition of the mTOR pathway may have anti-inflammatory and anti-angiogenic effects.[@bib12], [@bib14], [@bib30], [@bib40], [@bib41], [@bib42], [@bib43] The rAAV-mTOR shRNA used in this study inhibits both mTOR complexes, leading to the profound suppression of inflammation and angiogenesis, confirmed by immunohistology utilizing CD11b, F4/80, and CD31. These revealed that rAAV-mTOR shRNA-treated mouse retinas exhibited a reduction in the extent of CNV lesions and significantly less infiltration by monocytes and macrophages. Additionally, as the secretion of various cytokines and chemokines, such as IL-1β and TNF-α,[@bib44] further activates endothelial cells in the early stages of inflammation, the anti-inflammatory effects of mTOR inhibition may also be implicated in reducing angiogenesis. To date, there is strong evidence to support that impaired autophagy in the RPE leads to the accumulation of lipofuscin and a reduced ability to clear intracellular debris,[@bib45], [@bib46], [@bib47], [@bib48] and that the activation of autophagy, modulated by the rapamycin-induced inhibition of mTORC1, is able to prevent this harmful AMD-related aging of RPE cells.[@bib18], [@bib19], [@bib20] As seen in the expression of LC3B and ATG7, selective markers for autophagic activity, we demonstrate here that this specific mode of autophagy activation significantly reduces CNV. LC3B expression was detected mainly in the CNV lesions themselves, and with the clearly observable transduction of rAAV-mTOR shRNA into the lesions, it is highly likely that this site-specific activation of autophagy was due to the inhibitory effects of mTOR shRNA on mTOR signaling. Moreover, by increasing autophagy in the CNV lesions, mTOR inhibition may not only limit CNV progression, but also reverse previous CNV activity. Therefore, treatments resulting in enhanced autophagy via mTOR inhibition may help overcome the limitations of the anti-VEGF therapies currently used, which work by decreasing vascular permeability and inhibiting new vessel formation, but have a limited effect on stable mature vessels covered with pericytes.[@bib49], [@bib50], [@bib51] In addition to CNV formation, AMD is characterized by RPE degeneration and the loss of photoreceptors. It was previously reported that mTOR inhibition by rapamycin prevents pathological changes in RPE cells and protects photoreceptors from oxidative stress.[@bib52] Correspondingly, the TUNEL assay showed that rAAV-mTOR shRNA treatment, which unlike the variable activity of rapamycin with respect to the two mTOR complexes, affects both mTORC1 and mTORC2, and significantly reduced apoptosis activity in ONL cells around the CNV when compared with controls. Although beyond the scope of the current manuscript, if mTOR inhibition is directly responsible for protecting photoreceptors and is not due to the effects to rapamycin, this additional aspect of rAAV-mTOR shRNA treatment may increase its efficacy as an AMD therapeutic. In summation, we demonstrate here that mTOR inhibition mediated by rAAV-mTOR shRNA in a mouse model for AMD resulted in increased autophagic activity, reduced inflammatory activity, and reduced angiogenesis in laser-induced CNV lesions, as well as an overall reduction in the extent of CNV. In addition to being able to effectively transduce CNV lesions because of its scAAV2 packaging vector, the mTOR-targeting shRNA was designed to be multispecies compatible. Therefore, rAAV-mTOR shRNA may serve as the foundation for the development of therapeutics to replace or to be used in conjunction with the anti-VEGF therapies currently used to treat AMD. Further preclinical studies will be necessary to determine the optimal concentration of the therapeutic vector to maximize efficacy and to evaluate ocular and systemic toxicity with regard to dose escalation. Materials and Methods {#sec4} ===================== Animals {#sec4.1} ------- Eight-week-old male C57/BL6 mice (The Orient Bio) were used in this study. All animal care and experiments were performed in accordance with the guidelines in the Association for Research in Vision and Ophthalmology Resolution on the Use of Animals in Ophthalmic and Vision Research and overseen by the Institutional Animal Care and Use Committee of Soonchunhyang University Hospital Bucheon. Laser-Induced CNV {#sec4.2} ----------------- After anesthetizing the animals via the intraperitoneal (i.p.) injection of a mixture of 40 mg/kg zolazepam/tiletamine (Zoletil; Virbac) and 5 mg/kg xylazine (Rompun; Bayer Healthcare), the pupils were dilated with 0.5% tropicamide and 2.5% phenylephrine (Mydrin-P; Santen). Laser photocoagulation (200 μm spot size, 0.02 s duration, 100 mW) was performed using a PASCAL diode ophthalmic laser system (neodymium-doped yttrium aluminium garnet \[Nd:YAG\], 532 nm; Topcon Medical Laser Systems). Only the right eye of each mouse was exposed to laser photocoagulation to induce CNV. Five to six laser spots were applied around the optic nerve head of said eye. A gaseous bubble formed at each laser spot, indicating the rupture of Bruch's membrane. Preparation of rAAV-mTOR shRNA-EGFP and Intravitreal Injections {#sec4.3} --------------------------------------------------------------- All recombinant AAV vectors were derived from scAAV2 vectors. The mTOR siRNA (5′-GAAUGUUGACCAAUGCUAU-3′) was designed from the completely conserved multi-species region found in humans (NM_004958), monkeys (XR_014791), rats (NM_019906), and mice (NM_020009) to establish rAAV-mTOR shRNA-EGFP. A scrambled control siRNA (5′-AUUCUAUCACUAGCGUGAC-3′) was prepared to make rAAV-scrambled control shRNA-EGFP (rAAV-scrambled shRNA-EGFP). Both of the scAAV2 vectors use the H1 promoter to express either mTOR siRNA or the scrambled control siRNA, whereas EGFP expression is driven by the cytomegalovirus promoter. All rAAV vectors were supplied by CdmoGen. Intravitreal injections of the vector were performed in the right eyes of the mice, with pupil dilation, 5 days after laser photocoagulation under anesthesia using 35G blunt needles with Nanofil syringes (World Precision Instruments). One microliter of the viral vectors at a concentration of 5.0 × 10^10^ viral genomes (vg)/mL was used per injection. Laser photocoagulation was used to induce CNV in three groups of mice (n = 20 per each group) before being injected with the following: 0.1% PBS for the first group, rAAV-scrambled shRNA for the second, and rAAV-mTOR shRNA for the third. As a negative control, 10 mice were not treated with laser photocoagulation or intravitreal injections. Five mice from each group were used for qRT-PCR. FFA {#sec4.4} --- FFA was performed using a scanning laser ophthalmoscope (Heidelberg Retina Angiograph 2; Heidelberg Engineering) as previously described.[@bib53] In brief, the animal was anesthetized and the pupil dilated to observe the retina. FFA images were captured 3--5 min after an i.p. injection with 0.1 mL of 2% fluorescein sodium (Fluorescite; Akorn). FFA was performed at 5 and 13 days post-laser photocoagulation. Tissue Preparation {#sec4.5} ------------------ The mice were deeply anesthetized via an intraperitoneal injection of a 4:1 mixture of zolazepam/tiletamine (80 mg/kg) and xylazine (10 mg/kg), then intracardially perfused with 0.1 M PBS containing 150 U/mL heparin, followed by an infusion of 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB). After ocular enucleation, the anterior segment, including the cornea and lens, was removed to generate eyecups. For RPE whole mounts, the neural retina was removed as well. The RPE-choroid complex was fixed with 4% PFA in 0.1 M PB (pH 7.4) for 2 hr and prepared with four equidistant cuts. To prepare frozen sectioned samples, we fixed eyecups with attached neural retinas with 4% PFA in 0.1 M PB (pH 7.4) for 2 hr. The eyecups were then transferred to 30% sucrose in PBS, incubated overnight, and embedded in optimal cutting temperature (OCT) compound (Sakura Finetek). Using the embedded eyecups, serial sagittal sections 10 μm thick were prepared and the sections mounted on adhesive microscope slides (HistoBond; Marienfeld-Superior). Transverse sections with CNV lesions were selected among the samples by visually scanning all serial sections. Immunohistochemistry {#sec4.6} -------------------- Immunohistochemistry was performed for both whole mounts and transverse sections of the retina. To immunostain whole mounts, we incubated RPE-choroid tissues overnight at 4°C with anti-CD31 (550274, 1:200; BD Pharmingen) diluted in PBS containing 1% Triton X-100 (PBST; Sigma-Aldrich). The tissues were washed three times for 10 min apiece with PBST and incubated for 2 hr at room temperature with Alexa Fluor 532-conjugated goat anti-mouse (A11002; Thermo Fisher Scientific) and rhodamine-conjugated phalloidin (A22287; Thermo Fisher Scientific). For transverse sections, slides were incubated overnight at 4°C with primary antibodies for either anti-CD11b (MCA711G, 1:200; Serotec), anti-F4/80 (MCA497GA, 1:200; Serotec), or anti-CD31 (550274, 1:200; BD Pharmingen) to detect monocyte, macrophage, and endothelial cells, respectively. To visualize EGFP expression and verify transduction by the AAVs, we used an anti-GFP antibody (ab6556, 1:200; Abcam). mTOR or autophagy activity was observed using anti-mTOR (AF15371, 1:200; R&D Systems) and anti-LC3B IgG (NB110-2220, 1:200; Novus Biologicals), respectively. For the TUNEL assay, the tissue was stained in accordance with the protocol provided by the manufacturer (12156792910, In Situ Cell Death Detection Kit, TMR red; Roche Diagnostics). After washing three times with PBST, the samples were incubated with Alexa Fluor 488-, 546-, and 647-conjugated (Thermo Fisher Scientific) secondary antibodies and stained with DAPI (D9542; Sigma-Aldrich) to visualize the cell nuclei. Both the whole mounts and transverse sections were examined via fluorescence confocal microscopy (LSM 700; Carl Zeiss Microscopy), with images captured using image-capture software (LSM Image Browser; Carl Zeiss Microscopy) at ×100 and ×200 magnification. RT-PCR {#sec4.7} ------ The eyeball was enucleated from deeply anesthetized mice, and the cornea, lens, and RPE-choroid complex were removed. Total RNA was prepared from the neural retina without the RPE-choroid complex using TRIzol reagent (Invitrogen). RNA (2 μg) was reverse transcribed into cDNA using Superscript III (Invitrogen). Samples were analyzed for mRNA levels using SYBR Green kits (Invitrogen), and fold changes in mRNA expression were determined using the 2^−ΔΔCt^ method, normalizing the results to the expression of the control *GAPDH*. PCR was carried out in triplicate, with amplification performed utilizing a pair of primers specific for mTOR (forward: 5′-CCACTGTGCCAGAATCCATC-3′, reverse: 5′-GAGAAATCCCGACCAGTGAG-3′). Image Analysis and Statistical Analysis {#sec4.8} --------------------------------------- Data collection for quantitative comparisons of the extent of CNV and cell counts for monocyte, macrophage, and TUNEL-positive cells was conducted using ImageJ software (NIH). For counting cells, five transverse sections from each CNV lesion were selected and immunostaining-positive cells counted at ×100 magnification. Five laser-induced CNV lesions from each group were included for each analysis, with statistical analyses performed using SPSS software (version 20.0 for Windows; SPSS). The Kruskal-Wallis test with a post hoc analysis (Bonferroni's method) was used for the comparison, and significant difference was determined at p \< 0.05. Author Contributions {#sec5} ==================== Conceptualization, T.K.P. and K.P.; Methodology, T.K.P. and J.S.C.; Investigation, T.K.P., S.H.L., J.S.C., S.K.N., H.J.K., and H.Y.P.; Writing -- Original Draft, T.K.P. and S.H.L.; Writing -- Review & Editing, T.K.P., K.P. S.H.L., S.H.S.L., and J.S.C.; Supervision, T.K.P., H.L. and K.P. Conflicts of Interest {#sec6} ===================== No potential conflicts of interest exist for all authors. This work was supported in part by the Soonchunhyang University Research Fund. [^1]: These authors contributed equally to this work.

Background ========== With more than 240.000.000 registered players (17% female), soccer (i.e., European "football") is the most popular sport in the world today \[[@B1]-[@B4]\]. Although the players and their mastery of the sport are the primary reasons for soccer's popularity \[[@B5],[@B6]\], the soccer referees are an inseparable factor of this sport \[[@B7]-[@B10]\]. The Fédération Internationale de Football Association (FIFA) reported more than 800.000 registered soccer referees and assistant referees \[[@B11]\]). The referees' activities during the game are regularly studied and reported \[[@B12]-[@B15]\]. In addition, their fitness level is a highly important factor, and it is officially tested \[[@B16]-[@B20]\]. An appropriate fitness status is a mandatory prerequisite for any referee at advanced competitive levels \[[@B17],[@B21],[@B22]\]. The overall physical demands on the referees are considered similar to those on the soccer players. In short, a referee covers a distance of 9 to 13 km, with 4%--18% of physical activity consisting of high-intensity running; elite referees may perform up to 1270 changes in activity during a game and make more than 130 decisions \[[@B14],[@B15],[@B23]\]. However, high-level referees are 15--20 years older than the players and, in most cases, are not professionals and cannot be substituted during the game. Evidently, an appropriate health status is one of the most important aspects of successful refereeing. The high demands of the game and the stress it places on the musculoskeletal and cardio-pulmonary systems increase the risk of injury. One of the first studies that investigated this problem was that of Fauno and his associates \[[@B24]\], where the authors recognized the problem and investigated the effect of shock-absorbing heel inserts on the incidence of soreness. Bizzini et al. \[[@B25]\] studied referees selected for the 2006 FIFA World Cup and reported the types of injuries, as well as the location, characteristics, type, frequency, and consequences of the complaints and the type of treatment offered. More than 40% of the referees reported an injury, more than 60% had musculoskeletal complaints during their career, and there was a mean rate of 20.8 injuries per 1000 match hours. Almost 50% of all female soccer referees participating in the Women's World Cup 2007 reported that they had been injured during their career, with a rate of 34.7 match injuries per 1000 match hours during the World Cup \[[@B26]\]. However, those data are only partially generalizable to referees from lower levels. Studies performed on Swiss referees reported a rate of 22.5% to 44% of injured referees \[[@B27],[@B28]\]. In a recent review Weston et al. concluded that high training loads combined with increasing age could, in part, explain the incidence of non-contact match injuries (18 injuries per 1000 match hours), with lower leg muscle strains being the most common type of non-contact injuries in referees \[[@B29]\]. In one of the rare prospective studies on this topic, performed on Irish referees \[[@B30]\], the authors found 8.8 injuries/1000 hours of training and 16.4 injuries/1000 hours for match officiating. However, the risk of injury varies considerably based on the standards of the game of which a referee has been placed in charge. Additionally, previous studies did not report data on injuries that happened as a result of official physical fitness testing of soccer referees. The aim of this study was to describe the frequency, type and consequences of injuries among Croatian soccer referees. Specifically, we studied the problems regarding (A) the various levels of refereeing quality, (B) the differences between the main and assistant referees, and (C) the injuries that occurred during games and those that occurred during physical fitness testing. Methods ======= Design and subjects ------------------- Prior to this retrospective study, we contacted the Croatian Football Federation and obtained formal permission for the study. A self-administered questionnaire that had been previously used in similar studies and obtained from the authors of those studies \[[@B25]-[@B28]\] was used as a measurement tool (Additional file [1](#S1){ref-type="supplementary-material"}). Based on previous studies \[[@B25]-[@B28]\], the content validity of the questionnaire is therefore presumed. Additionally, following translation into Croatian, the questionnaire was checked for its clarity and content validity by two MDs (both sports-medicine specialists) and one high-level soccer referee (first author of the paper). The first phase of the experiment consisted of an evaluation of the reliability of the translated questionnaire. Twenty-two referees were asked to complete the questionnaire twice (a test-retest procedure, with a period of 10 days between the test and re-test). The reliability was checked throughout the test-retest correlation (for ordinal variables) and the analysis of equally answered questions (for ordinal and nominal variables) \[[@B31]\]. The reliability test showed the high reliability of the questionnaire translated into Croatian. Briefly, the test-retest correlation for ordinal variables ranged from 0.80 (for injury consequences for the whole career) to 0.99 (refereeing experience). The consistency of the test-retest answers was 90 to 100%, and differences were only found for some of the questions. Because the most common and the greatest disagreements between the test and retest results were found for injuries and consequences throughout the whole refereeing career, in this investigation, we have focused on the previous year, the most recent refereeing match and physical fitness testing. The discriminative validity of the questionnaire was estimated throughout this investigation by discriminating MRs and ARS and differentiating referees of different levels \[[@B32]\]. The survey was administered at seven specialized soccer referee seminars from April to July of 2011. Using the multimedia presentation (see Additional file [2](#S2){ref-type="supplementary-material"}) the first author informed subjects about the purpose of the study, explained them the questionnaire design, and other important issues (definition of injuries; anonymity, testing protocol, etc.). The participants consisted of 342 soccer referees (all males, engaged in outdoor soccer; mean age 32.9), comprising more than 90% of all licensed Croatian soccer referees. Practically all subjects who were present at the seminars from April to July of 2011 (90% of all Croatian national level referees) were asked to participate in the study, and the response rate was more than 95%. Only a few examinees did not answer all questions. The study objective was to assess a representative sample of Croatian soccer referees of various competitive levels. Subjects completed the questionnaire in groups of 15 to 20 respondents, and the examiner (the first author) was available at all times for questions and answers. The participants were informed that they could refuse participation and withdraw from the study at any time, for any reason, and their informed consent was obtained. Participation was anonymous (except for those examinees included in the test-retest procedure for reliability testing), and no personal data regarding date of birth, city of residence and/or any other data that could be directly connected to an individual were included in the questionnaire. The answer options were presented as multiple-choice, closed-ended responses for most of the questions. Prior to the study, ethical approval was obtained from the Ethical Board of the Faculty of Kinesiology, University of Split, Croatia. Variables --------- The questionnaire consisted of the following sections: (1) morphological-anthropometric data; (2) refereeing variables; and (3) musculoskeletal injuries together with the consequences of injury. Morphological anthropometric data: Subjects reported body height (BH) and body mass (BM) according to the official measurements performed during physical fitness testing (a day or two before the questioning was performed), and we calculated the body mass index (BMI). Refereeing variables: The subjects were grouped into main and assistant referee groups. We asked the subjects about their experience in refereeing (e.g., how long ago did they receive their official license), the highest current level of refereeing achieved (four point scale: UEFA level, 1^st^ national, 2^nd^ national, 3^rd^ national level), the number of games refereed during the last year, and the average training hours per week (during the preseason and during the season, two separate questions). Musculoskeletal injuries and consequences: The respondents were asked about the occurrence of injuries and musculoskeletal complaints for the most recent match, the last year, and the entire career, as originally suggested by the authors \[[@B25]-[@B28]\]. However, in this study we have reported only data on injuries for the most recent match and the last year. In addition to the items on the original questionnaire, the subjects were asked if they had any injuries or complaints during the physical testing. If the answer was yes, the subject was asked for the location of the injury (22 choices, such as head, neck, etc.), the type of injury (15 choices, such as dislocation, contusion, tendinitis, etc.), the length of recovery (how many days the consequences of the injury were felt), and absence from training (days) and refereeing (days). In the case of multiple injuries, all injuries were reported separately. An injury was defined as "any physical complaint sustained as a result of refereeing and training, irrespective of the need for medical attention or time lost from activity (refereeing or training)", which was adopted from Fuller et al. \[[@B33]\]. To calculate the ratio of injuries per 1000 hours of match during the last year (see enclosed questionnaire; page 5, section 4; 2^nd^ question), the hours were determined as the total number of games refereed throughout the last year x 1.5 hours (90 minutes of official game with no extra time). The ratio of injuries per 1000 hrs of last year refereeing is calculated separately for MRs and ARs according to their quality level. Statistics ---------- For parametric variables (age, experience, etc.), the means and standard deviations were reported. For nonparametric measures (i.e., nominal and ordinal variables), the frequencies and proportions, together with 95% confidence intervals in some cases \[[@B34]\], were calculated. Depending on the parametric/nonparametric nature of the variables, we applied a univariate analysis of variance (ANOVA) with the Schefee post-hoc analysis and/or the Kruskal-Wallis ANOVA \[[@B35]\] to establish the differences between (a) the main and assistant referees and (b) the referees of different levels (separately for main and assistant referees). The responses from the open-ended questions and other comments by the subjects were transcribed and further analyzed for common themes. A value of P \< 0.05 was considered statistically significant. The statistical analyses were performed using Statsoft's Statistica version 10. Results ======= Of 342 subjects, the 157 were main referees (MRs; mean age 31.5 years), and 185 were assistant referees (ARs; mean age 34.1 years). The ARs were significantly (P \< 0.05) older than the MRs, but only when considering those involved in the 1^st^ and 3^rd^ national leagues. The 1^st^ and 2^nd^ national league MRs were significantly (P \< 0.05) taller than the ARs. The ARs from the 2^nd^ and 3^rd^ league had significantly (P \< 0.05) longer experience than their MR peers. The UEFA referees were the most experienced in refereeing, followed by the 1^st^ league MRs. The MRs were significantly (P \< 0.05) taller than the ARs. Among the ARs, the UEFA referees were the oldest. The 3^rd^ league ARs had the highest BMI of all the groups (Table [1](#T1){ref-type="table"}). ###### **Descriptive statistics for general data of examinees (M-mean, SD-standard deviation), and ANOVA differences between match referees (MR) and assistant referees (AR) within the same competitive level (\* denotes significant differences), and within MR and AR between the different competitive levels (**^**¥**^**denotes significant differences)**     **Total (n = 342)** **UEFA (n = 18)** **1**^**st**^**(n = 78)** **2**^**nd**^**(n = 91)** **3**^**rd**^**(n = 155)** -------------------- -------------- --------------------- ------------------- --------------------------- --------------------------- ---------------------------- -------------- ---------------- ----------------- ------------------- Age (years) 31.4 ± 4.88 34.11 ± 5.08 34.67 ± 2.5 38.25 ± 4.45 33.65 ± 4.46 35.8 ± 4.15\* 30.59 ± 4.42 31.98 ± 4.97 30.88 ± 5.16^¥^ 33.71 ± 5.11\*^¥^ Experience (years) 10.79 ± 3.26 12.52 ± 3.43\* 17 ± 2.28 17 ± 2.98 13.24 ± 2.57 14.33 ± 2.62 9.47 ± 2.87 11.31 ± 3.82\* 10.15 ± 2.71^¥^ 11.4 ± 2.54\*^¥^ BH (cm) 183.4 ± 5.75 181.41 ± 6.23\* 181 ± 3.52 183.17 ± 8.11 185.69 ± 6.07 182.78 ± 6.28\* 183.6 ± 5.08 180.4 ± 5.69\* 182.59 ± 5.96 180.91 ± 6.158 BM (kg) 83.44 ± 7.24 82.04 ± 7.45 81.5 ± 2.35 81.5 ± 7.45 85.93 ± 7.46 83.89 ± 7.09 82.06 ± 7.35 79.49 ± 7.04 83.47 ± 7.19 82.5 ± 7.62^¥^ BMI (kg/m2) 24.77 ± 1.35 24.9 ± 1.44 24.88 ± 0.42 24.28 ± 1.37 24.88 ± 1.23 25.07 ± 0.95 24.31 ± 1.45 24.41 ± 1.71 25 ± 1.32 25.17 ± 1.46^¥^ LEGEND: Age-age of the subjects, Experience-time involved in the soccer refereeing, BH-body height, BM-body mass, BMI-body mass index. In general, 60 (17%), 101 (29%) and 56 (16%) referees reported injuries during the last match, during the last 12 months and during physical fitness testing, respectively, corresponding to 0.19, 0.41 and 0.22 injuries, respectively, per subject. Excluding the UEFA MRs, who were the least commonly injured, the higher level referees were more frequently injured than their peers who participate in matches at the lower levels. When comparing the ARs for injuries between levels, the 1^st^ league referees were the most commonly injured during fitness testing, with a significant difference (P \< 0.05) between the 1^st^ and 3^rd^ leagues. There was an overall significant (P \< 0.05) Kruskal Wallis difference in the 12-month injury status between the different levels of ARs, with an observable trend toward fewer injuries in national lower level refereeing (Table [2](#T2){ref-type="table"}). ###### **Prevalence of injuries (frequency-f; percentage-%; 95% CI-95% confidence interval) among soccer referees during the last match, the last year and during fitness testing, and Kruskal Wallis analysis of the differences (\* denotes significant differences) between qualitative levels (UEFA referees-UEFA, 1**^**st**^**National league-1**^**st**^**, 2**^**nd**^**National league-2**^**nd**^**, 3**^**rd**^**National league), within match referees (MR), and within assistant referees (AR)**   **Last match** **Last year** **During fitness testing** ---------- ---------------- --------------- ---------------------------- ------------ ---------- ------------ MR 29 (18) 0.13--0.25 46(29)\* 0.23--0.37 21(13) 0.05--0.14 AR 31 (17) 0.12--0.22 55(30)\* 0.23--0.37 35(19)\* 0.14--0.25 UEFA MR 0(0) \- 3(50) 0.19--0.81 0(0) \- UEFA AR 0(0) \- 1(8) 0.01--0.35 3(25) 0.09--0.53 1^st^ MR 8(26) 0.13--0.43 12(39) 0.24--0.56 7(23) 0.11--0.40 1^st^ AR 11(23) 0.13--0.37 21(45) 0.31--0.58 17(36) 0.24--0.50 2^nd^ MR 8(17) 0.01--0.31 17(38) 0.25--0.52 8(18) 0.09--0.31 2^nd^ AR 8(17) 0.09--0.30 14(30) 0.19--0.45 9(20) 0.10--0.33 3^rd^ MR 13(17) 0.10--0.27 14(19) 0.11--0.29 6(8) 0.04--0.16 3^rd^ AR 12(15) 0.12--0.29 19(24) 0.16--0.34 6(8) 0.03--0.15 As a result of injuries that occurred during the last match, the 1^st^ league MRs abstained from training and refereeing for the shortest time. For the injuries that occurred during the last year, the UEFA referees abstained from training for the shortest time (18 days on average). The longest break because of injuries that occurred during physical fitness testing was noted for the 3^rd^ league (79 days), followed by the 1^st^ league (41 days) and 2^nd^ league referees (22 days). It must be stressed that UEFA MRs did not experience any injuries during physical fitness testing (Table [3](#T3){ref-type="table"}). ###### **Descriptive statistics for consequences of injury which occurred during the last match, last year and during fitness testing among match referees (n-number of subjects in each group, M-mean, SD-standard deviation) for different competitive levels (UEFA referees-UEFA, 1**^**st**^**National league-1**^**st**^**, 2**^**nd**^**National league-2**^**nd**^**, 3**^**rd**^**National league)**   **Total (n = 342)** **UEFA (n = 18)** **1**^**st**^**(n = 78)** **2**^**nd**^**(n = 91)** **3**^**rd**^**(n = 155)** --------------------------- --------------------- ------------------- --------------------------- --------------------------- ---------------------------- Last match (injured) n = 29 n = 0 n = 8 n = 8 n = 13 Recovery (days) 36.5 ± 43.42   30.38 ± 29.14 52.5 ± 45.17 33.11 ± 55.91 No training (days) 36.38 ± 48.47   21.43 ± 18.49 39 ± 32.86 46.73 ± 66.62 No refereeing (days) 58.38 ± 64.11   42.33 ± 42.85 54.67 ± 57.14 74.67 ± 83.96 Last year (injured) n = 46 n = 3 n = 12 n = 17 n = 14 Recovery (days) 53.55 ± 73.43 76 ± 90.16 73.4 ± 104.73 51.63 ± 68.5 35.46 ± 46.5 No training (days) 35.35 ± 35.31 17.5 ± 4.95 34.5 ± 22.48 38.46 ± 34.49 35.5 ± 44.96 No refereeing (days) 42.59 ± 48.67 20.5 ± 0.71 20.4 ± 8.76 36.11 ± 35.99 78.17 ± 74.38 Fitness testing (injured) n = 21 n = 0 n = 7 n = 8 n = 6 Recovery (days) 39.94 ± 57.69   36.57 ± 28.56 59 ± 101.29 22 ± 16.37 No training (days) 32.25 ± 25.73   31.33 ± 22.37 37.67 ± 32.21 25.5 ± 25.04 No refereeing (days) 43.11 ± 47.13   41 ± 33.53 22.33 ± 13.28 78.5 ± 101.12 For the ARs, the recovery from injuries that occurred during the last match ranged from 30 (2^nd^ league) to 81 days (3^rd^ league) on average. As a result of injuries that happened during the last year, the shortest period with no training was presented by the ARs at the UEFA level. Injuries occurring during the last refereed game did not result in any pause from refereeing for the UEFA ARs. However, an average of 21, 37 and 87 days elapsed before the 3^rd^, 1^st^ and 2^nd^ league ARs resumed their duties, respectively. As a result of fitness testing injury, the 1^st^ league ARs abstained from their training for a relatively brief period of time (26 days) when compared with their 2^nd^ league colleagues, who abstained from their training for 40 days (Table [4](#T4){ref-type="table"}). ###### **Descriptive statistics for the consequences of injury which occurred during the last match, last year and during fitness testing among assistant referees (n-number of subjects in each group, M-mean, SD-standard deviation) for different competitive levels (UEFA referees-UEFA, 1**^**st**^**National league-1**^**st**^**, 2**^**nd**^**National league-2**^**nd**^**, 3**^**rd**^**National league)**   **Total (n = 342)** **UEFA (n = 18)** **1**^**st**^**(n = 78)** **2**^**nd**^**(n = 91)** **3**^**rd**^**(n = 155)** --------------------------- --------------------- ------------------- --------------------------- --------------------------- ---------------------------- Last match (injured) n = 31 n = 0 n = 11 n = 8 n = 12 Recovery (days) 54.91 ± 57.96   43.43 ± 40.41 29.83 ± 30.31 80.56 ± 75.45 No training (days) 41.64 ± 53.23   33.67 ± 30.21 44.5 ± 57.23 44.75 ± 67.08 No refereeing (days) 78.33 ± 102.61   66.6 ± 64.67 104 ± 174.27 71 ± 84.76 Last year (injured) n = 55 n = 1 n = 21 n = 14 n = 19 Recovery (days) 48.5 ± 87.19 6 ± 0 75 ± 119.55 37.75 ± 50.07 12.43 ± 9.9 No training (days) 36.85 ± 56.88 6 ± 0 49.15 ± 78.83 45.2 ± 50.5 15.6 ± 11.06 No refereeing (days) 45.87 ± 74.26   37.09 ± 29.84 87.33 ± 139.27 20.5 ± 9.01 Fitness testing (injured) n = 35 n = 3 n = 17 n = 9 n = 6 Recovery (days) 44.86 ± 91.31 13.67 ± 7.51 69.6 ± 120.7 13.83 ± 12.86 22 ± 9.24 No training (days) 26.75 ± 34.13 10.33 ± 6.35 26 ± 20.21 40.14 ± 62.61 18.25 ± 7.85 No refereeing (days) 48.74 ± 49.95 24.5 ± 14.85 59.44 ± 50.15 55.4 ± 70.36 21.67 ± 8.02 The Achilles tendon, knee, ankle and lower leg were the most commonly injured body regions. The MRs injured the Achilles tendon, thigh, and ankle most commonly, while the ARs experienced more lower leg, lower back and shoulder injuries. There was no evident difference between the type of injuries that occurred in the last 12 months and those that occurred during the last match refereed. During fitness testing, thigh and groin injuries prevailed (Table [5](#T5){ref-type="table"}). ###### **Location of injuries (frequencies-f; percentage-%) among soccer referees on the last match, during the last year and during fitness testing, for match referees (MR), and assistant referees (AR); and for different competitive levels (UEFA referees-UEFA, 1**^**st**^**National league-1**^**st**^**, 2**^**nd**^**National league-2**^**nd**^**, 3**^**rd**^**National league**   **Total n = 342** **MR n = 157** **AR n = 185** **UEFA n = 18** **1**^**st**^**n = 78** **2**^**nd**^**n = 91** **3**^**rd**^**n = 155** ------------------------------------------ ------------------- ---------------- ---------------- ----------------- ------------------------- ------------------------- -------------------------- Last match (n = 67 injuries)               Elbow 1 (1) 0(0) 1(3) 0(0) 0(0) 0(0) 1(4) Lower back 4 (6) 0(0) 4(11) 0(0) 1(4) 1(6) 2(8) Abdominal, hip and groin area 10(14) 4(13) 6(17) 0(0) 1(4) 5(28) 4(16) Thigh 5 (7) 2(7) 3(8) 0(0) 2(9) 2(11) 1(4) Knee 10 (15) 4(14) 6(16) 0(0) 3(13) 3(17) 4(15) Lower leg 9 (13) 2(7) 7(19) 0(0) 4(17) 0(0) 5(19) Achilles tendon 17 (25) 8(28) 9(24) 0(0) 6(26) 6(33) 5(19) Ankle and foot 11 (16) 9(31) 1(3) 0(0) 6(26) 1(6) 4(15) Last 12 months (n = 120 injuries)               Head 1 (1) 1(2) 0(0) 0(0) 0(0) 0(0) 1(3) Shoulder 3 (3) 0(0) 3(5) 0(0) 1(3) 1(2) 1(3) Lower back 14 (12) 4(7) 10(17) 0(0) 4(11) 7(16) 3(9) Abdominal, hip and groin area 17 (14) 9(14) 8(13) 2(34) 0(0) 7(14) 5(15) Thigh 10 (8) 8(13) 2(3) 1(17) 4(11) 3(7) 2(6) Knee 15 (13) 8(13) 7(12) 0(0) 2(5) 9(20) 4(12) Lower leg 21 (18) 5(8) 16(27) 2(33) 5(14) 5(11) 9(27) Achilles tendon 22 (18) 15(25) 7(12) 1(17) 12(32) 6(14) 3(9) Ankle and foot 17 (14) 11(18) 6(10) 0(0) 6(16) 6(13) 5(15) During fitness testing (n = 57 injuries)               Head 3 (5) 0(0) 3(9) 0(0) 2(8) 1(6) 0(0) Lower back 4 (7) 2(10) 2(6) 0(0) 1(4) 2(12) 1(8) Abdominal, hip and groin area 13 (23) 4(20) 9(26) 2(22) 4(17) 5(30) 2(16) Thigh 15 (26) 8(38) 7(20) 1(11) 8(33) 3(18) 3(25) Knee 4 (7) 1(5) 3(9) 0(0) 2(8) 1(6) 1(8) Lower leg 8 (15) 2(10) 6(17) 0(0) 1(4) 4(24) 3(25) Achilles tendon 8 (14) 3(14) 5(14) 0(0) 6(25) 1(6) 1(8) Ankle 1 (2) 1(5) 0(0) 0(0) 0(0) 0(0) 1(8) Muscle strain was more common during fitness testing than during a match, but other types of injury prevailed during a match (Table [6](#T6){ref-type="table"}). ###### **Type of injuries (frequencies-f; percentage-%) among soccer referees on the last match, during the last year and during fitness testing, for match referees (MR), and assistant referees (AR); and for different competitive levels (UEFA referees-UEFA, 1**^**st**^**National league-1**^**st**^**, 2**^**nd**^**National league-2**^**nd**^**, 3**^**rd**^**National league**   **Total n = 342** **MR n = 159** **AR n = 183** **UEFA n = 18** **1**^**st**^**n = 78** **2**^**nd**^**n = 91** **3**^**rd**^**n = 155** ------------------------------------------ ------------------- ---------------- ---------------- ----------------- ------------------------- ------------------------- -------------------------- Last match (n = 65 injuries)               Fracture 1 (2) 0(0) 1(3) 0(0) 0(0) 0(0) 1(4) Dislocation 1 (2) 0(0) 1(0) 0(0) 0(0) 1(5) 0(0) Sprain 8 (12) 7(22) 1(0) 0(0) 3(9) 1(5) 4(17) Ligament injury 8 (12) 4(13) 4(0) 0(0) 1(3) 3(15) 4(17) Lesion of meniscus or cartilage 5 (8) 2(6) 3(0) 0(0) 2(6) 1(5) 2(8) Muscle rupture/tear 4 (6) 2(6) 2(0) 0(0) 2(6) 1(5) 1(4) Muscle strain 11 (17) 3(9) 8(0) 0(0) 5(16) 2(10) 4(17) Tendon injury/strain 16 (25) 10(31) 6(0) 0(0) 2(6) 8(40) 6(25) Tendinitis/bursitis 8 (12) 4(13) 4(0) 0(0) 5(16) 1(5) 2(8) Contusion/hematoma 2 (3) 0(0) 1(0) 0(0) 1(3) 1(5) 0(0) Laceration 1 (2) 0(0) 1(0) 0(0) 0(0) 1(5) 0(0) Last 12 months (n = 110 injuries)               Concussion 1 (1) 0(0) 1(2) 0(0) 0(0) 0(0) 1(3) Fracture 1 (1) 0(0) 1(2) 0(0) 1(3) 0(0) 0(0) Dislocation 3 (3) 1(2) 2(3) 0(0) 0(0) 1(2) 2(6) Sprain 12 (10) 8(15) 4(6) 0(0) 3(8) 4(10) 5(14) Ligament injury 9 (7) 2(4) 7(11) 0(0) 1(3) 4(10) 4(11) Lesion of meniscus or cartilage 7 (6) 4(7) 3(5) 0(0) 2(5) 2(5) 3(9) Muscle rupture/tear 9 (8) 5(9) 4(6) 2(33) 3(8) 2(5) 2(6) Muscle strain 32 (29) 11(20) 21(33) 2(33) 10(27) 12(29) 8(23) Tendon injury/strain 20 (18) 15(27) 5(8) 2(33) 6(16) 8(20) 4(11) Tendinitis/bursitis 18 (16) 8(15) 10(16) 0(0) 8(22) 6(15) 4(11) Contusion/hematoma 3 (3) 1(2) 2(3) 0(0) 1(3) 2(5) 1(3) Nerve injury 3 (3) 0(0) 3(5) 0(0) 2(5) 0(0) 1(3) During fitness testing (n = 56 injuries)               Concussion 1 (2) 0(0) 1(3) 0(0) 1(4) 0(0) 0(0) Dizziness 2 (4) 0(0) 2(6) 0(0) 1(4) 1(6) 0(0) Fracture 1 (2) 1(5) 0(0) 0(0) 1(4) 0(0) 0(0) Sprain 1 (2) 1(5) 0(0) 0(0) 0(0) 0(0) 1(8) Lesion of meniscus or cartilage 2 (4) 1(5) 1(3) 0(0) 2(8) 0(0) 0(0) Muscle rupture/tear 3 (5) 1(5) 2(6) 0(0) 1(4) 2(12) 0(0) Muscle strain 33 (59) 13(59) 20(59) 3(100) 12(50) 11(65) 7(58) Tendon injury/strain 3 (5) 2(9) 1(3) 0(0) 2(8) 1(6) 0(0) Tendinitis/bursitis 6 (11) 1(5) 5(15) 0(0) 4(17) 0(0) 2(17) Contusion/hematoma 2 (4) 1(5) 1(3) 0(0) 0(0) 1(6) 1(8) Nerve injury 2 (4) 1(5) 1(3) 0(0) 0(0) 1(6) 1(8) The injury incidence per 1000 hours shows that there was no evident difference between the MRs and ARs. Among the MRs, the lowest rate was found for the 3^rd^ league, while for the ARs, the lowest prevalence was found among the UEFA referees (Table [7](#T7){ref-type="table"}). ###### Number of injuries per 1000 hours of match during the last year   **Total** **UEFA** **1**^**st**^**league** **2**^**nd**^**league** **3**^**rd**^**league** -------------------- ------------------- -------------------- ------------------------- ------------------------- ------------------------- Match referees 5.29 (2.23--8.30) 3.85 (-0.93--8.33) 12.21 (-0.71--25.09) 6.24 (1.48--10.99) 1.90 (0.47--3.34) Assistant referees 4.58 (2.63--6.54) 2.01 (-3.72--6.52) 8.37 (3.27--13.46) 5.42 (-0.54--11.39) 2.11 (0.79--3.43) TOTAL 4.92 (3.17--6.69) 3.33 (-0.44--7.11) 9.90 (4.14--15.66) 5.89 (2.33--9.45) 2.00 (1.05--2.96) Discussion ========== The Croatian soccer referees included in this study were somewhat younger than those previously studied \[[@B9],[@B14],[@B20],[@B27]\], but this is in concordance with the recent UEFA suggestions regarding the need for decreasing the referees' ages at all competitive levels. In addition, our subjects were taller and had a higher BM than Chilean \[[@B7]\] and Spanish referees \[[@B8]\], while they had higher BM and BMI values than Swiss referees \[[@B28]\]. The rate of approximately 5 injuries per 1000 match hours is similar to those previously reported in retrospective studies performed for Swiss referees (1.6 and 6.8 injuries/1000 hours for AR and MR, respectively) \[[@B27]\]. Meanwhile, prospective investigations show a much higher incidence of injury (up to 20 injuries per 1000 hours of refereeing) \[[@B25],[@B30]\], which is explained by the recall-bias \[[@B36]\]. Our result of 1--2 injuries/1000 hours among UEFA referees is similar to the retrospective data reported for top-level international referees \[[@B25]\]. In the following section, we will focus on the issues that have not been extensively examined in previous studies. These issues mainly relate to (a) the differences between the type and location of injuries occurring during fitness testing and games, (b) the types of injuries and the consequences of those injuries in relation to the level of refereeing, and (c) the comparisons between the main and assistant referees with regard to the type and location of the injuries. Fitness testing injuries vs. game injuries ------------------------------------------ The injury rate of 29% during the last year is similar to rates that have been previously reported for Swiss referees \[[@B28]\]. One of every six subjects suffered from an injury during fitness testing, and therefore, although it is performed in a "controlled environment" (e.g., the tests are known in advance, there are no tests with quick changes of direction, tests are performed on a well-prepared field), fitness testing should be considered potentially dangerous in terms of injury risk. It is especially interesting to compare the number of tests per year (i.e., 4) and the average number of games per year (i.e., 15), which leads us to conclude that physical fitness testing carries a higher risk of injury among soccer referees than match refereeing. There are several possible explanations for these findings. First, participation and successful achievement in fitness testing are the main prerequisites for match refereeing. Therefore, referees are highly concerned about their achievement at every examination. During a game, the referee can "tailor" his method of refereeing and thereby pace the physical demands of refereeing, whereas throughout testing, standards for both tests done throughout fitness testing must be achieved. Most likely, a lack of fitness level and poor preparation are the main risk factors for injury occurrence during fitness testing. This is indirectly demonstrated by the fact that the UEFA referees, who must achieve the highest fitness standards, were less commonly injured during physical fitness testing (i.e., none of the UEFA MRs reported injuries that occurred during fitness testing). The lower leg region (calf and tendon, represents 31% of all injuries) and the upper leg region (quadriceps and hamstrings, represents 26%) were equally injured during physical fitness testing. During games, however, almost 50% of all injuries were lower leg injuries. The main reason for the dissimilarity between games and fitness testing with regard to the location of injuries is based on the types of activities in these two events. Previous studies noted that soccer referees sprint (run faster than 18 km/h) approximately 1 km in each half of the game \[[@B14],[@B15]\]. Given that one of the fitness tests is completed at maximum running speed, the running intensity is much higher during fitness testing. The upper leg (quadriceps and hamstring) is biomechanically and anatomically more challenged during sprinting than during lower-intensity running \[[@B37]\]. It is therefore logical that studies have found that the quadriceps and hamstring regions are more commonly injured among sprinters than among middle- and long-distance runners \[[@B38],[@B39]\]. This directly explains the higher rate of injury to the upper leg region during fitness tests than during match refereeing. The same logic applies when explaining the relatively higher injury rate for the groin area during physical fitness testing. Although not reported among soccer referees, this problem has been reported in soccer players where more intensive sprinting was related to a higher risk of groin injury in \[[@B40]\]. In contrast to the upper leg and groin area, which are more commonly injured during fitness testing, the knee is injured twice as often during games compared with during fitness testing, which is most likely a result of the characteristic movement patterns during the game. Throughout a soccer match, referees perform numerous changes of direction, some of which are executed during high-intensity runs. In contrast, during fitness testing, such risky movement patterns are not as common, which places the referees at a lower risk for knee injury. Furthermore, throughout testing, the referees focus exclusively on the fitness task, while during a match, they must focus on the game. This focus can place the referees at risk of inappropriate movement patterns, such as a spin turn \[[@B41]\], which consequently increases the possibility of knee injuries. The risk is amplified during the game because refereeing is performed in soccer boots (during the spin-turn the foot is "locked" in the ground), and this is not the case during fitness testing. Prevalence and consequences of injuries: are there any differences between the various levels of refereeing? ------------------------------------------------------------------------------------------------------------ There is an obvious increase in injuries concurrent with an increase in the referee level. The highest prevalence of injuries is found in referees involved in the 1^st^ league, and the prevalence decreases at lower levels, with the lowest prevalence observed for the 3^rd^ league referees. However, among UEFA referees, the injury rate is somewhat lower, although these referees regularly participate in the national 1^st^ league competitions. Previous studies \[[@B27]\] reported a trend similar to that found in our study. The highest prevalence of match injuries was noted for the highest competitive referee level (5.54 injuries per 1000 match hours). The prevalence rate was lower for the lower level competitions (0.59, 2.57 and 1.97 injuries per 1000 match hours for junior, low-level amateur and high-level amateur referees, respectively). It has been suggested \[[@B27],[@B42]\] that one of the risk factors for injury is the age of the subjects. The lowest level referees, in our case, the 3^rd^ national league, are the youngest. Another explanation for the higher injury rates for higher-level referees may be found in the more advanced level of play and game dynamics, which increases with competitive level \[[@B5],[@B6]\]. However, as stated before, this trend of higher injury rates for higher competitive levels is not evident among the top-level referees (UEFA referees). Even more intriguing is that the UEFA referees are the oldest of all referees studied. There are several possible reasons for their low injury rate. First, the injury rate is one of the selective factors for advancement in refereeing. All soccer referees begin their career at low-level competitions, and their progress depends on the quality of their performance. Those referees who are less frequently injured are more likely to be recognized as successful, which leads to their promotion to higher-level competitions. Most likely even more important is the fact that the UEFA referees are the most strictly monitored for their fitness and training status and fitness achievements \[[@B21]\]. They are obligated to provide a weekly/monthly report of their training throughout the season to the UEFA authorities. This report includes not only subjective reports but also precise and detailed analyses of the overall training volume and intensity and includes heart rate monitoring diaries. Furthermore, their training is prescribed and monitored by UEFA experts. Apart from the already proven superior fitness status \[[@B21]\], such a supervised exercise regime almost certainly places the referees at a lower risk of injury. This explanation is indirectly proven by the fact that none of the UEFA MRs reported any injuries during physical fitness testing, which is a clear indicator of their excellent physical readiness and overall fitness status. Comparison between the main and assistant referees with regard to the type and location of injuries --------------------------------------------------------------------------------------------------- The injury rates of the Croatian soccer referees are similar to those previously reported for Swiss referees (29% of the subjects were injured during the last 12 months, with approximately 5 injuries per 1000 hours) \[[@B28]\]. However, the Swiss study found a higher percentage of injured MRs (36%) compared with ARs (20%), most likely because the authors sampled only top-division referees. The ratio between the MRs and ARs for injured ankles is approximately 8 to 1, which is explainable by knowing the differences in the MR and AR activities during the game. The movement of the ARs is characterized by short, intense sprints of 5 to 40 meters, interrupted by relatively long, low-intensity periods \[[@B21],[@B43]\]. The MRs sprint at a high intensity over a longer distance than the ARs (up to 80 meters). Additionally, the ARs are placed outside of the official playing field, where the consistency and quality of the turf is far better than within the playing field, where the MRs are positioned. Surface quality has been recognized as an important factor in sports injury rates \[[@B44]\]. These factors (surface quality and differences in the sprint distance) most likely also contribute to the higher prevalence of Achilles tendon injuries among MRs. Our findings are in accordance with previous studies \[[@B25],[@B26]\] that noted a higher prevalence of lower back injuries among ARs than among MRs. Although not studied in detail in the questionnaire, which was based on previous studies relating lower-back pain to the advanced age of the subjects \[[@B45]\], we believe that the problem is age-related (i.e., ARs are older than MRs). This is supported by the fact that lower-back problems are more prevalent in ARs from the qualitative groups in which assistants are significantly older than their match peers (e.g., 1^st^ and 3^rd^ national leagues). Although not frequent, shoulder injuries are noted exclusively among ARs, and this is almost certainly related to the ARs' use of flags. Study limitations ----------------- There are several limitations of the study. First, the study is retrospective in nature. Consequently, the data should be assessed with regard to their possible low reliability. For example, the subjects may have forgotten some previous injuries and musculoskeletal disorders or the consequences of these injuries \[[@B36]\]. However, prior to the testing, we performed a pilot study in which the reliability of the questionnaire was shown to be appropriate (see Methods for more details). Moreover, the study included subjects from only one country, and therefore, the data are of questionable generalizability. However, the sample was relatively large and included more than 90% of all national level referees from the country. In addition, the response rate was high (more than 95%), which is one of the most important factors when studying health-related issues on a self-reported basis \[[@B46],[@B47]\]. Additionally, the investigation was based on self-reports, and the subjects might not have told the truth if they felt uncomfortable. However, we believe that the anonymity and design of the study (e.g., testing in large groups with no pressure on the examinees) decreased this possibility. The study was approved by the highest national soccer organization (Croatian Football Federation) and was therefore officially accepted. The testing was performed by the first author, who is a 1^st^ league MR and a UEFA additional assistant and is therefore an "insider". This fact almost certainly had a positive effect on the honesty of the examinees, as the examiner was familiar with the problems, and therefore, the subjects had no reason to be untruthful \[[@B35],[@B47]\]. Finally, and with regard to previous studies we must acknowledge that significant limitation should be observed in the fact that we studied only injuries, and did not report complaints, which almost certainly must be emphasized in forthcoming studies. Despite the study's limitations, we believe that the results (although not the final word on this topic) contribute to knowledge in the field. Conclusion ========== At the national refereeing levels, there is an increase in injuries concurrent with an increased competitive level. However, the low prevalence of injuries among UEFA level referees may be explained by their greater physical fitness and by officially prescribed and supervised UEFA training, overall professionalism, and regular medical assistance. Although fitness testing is performed in a controlled environment, testing should be considered to be a risk factor for injuries. The differences between match injuries and fitness testing injuries may be due to the characteristics of activities during matches and fitness tests (e.g., maximal sprints are more common during testing). The unevenness in the playing field together with frequent quick changes of direction are most likely the main contributors to ankle injuries among MRs. The ARs are older, which most likely influences the prevalence of lower back problems among ARs. In the future, we suggest using a prospective design to provide a deeper insight into the health-related problems and consequences of soccer refereeing. Abbreviations ============= MR: Match referees; AR: Assistant referees; FIFA: Fédération Internationale de Football Association; UEFA: Union of European Football Associations; ANOVA: Analysis of variance. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== GG designed the study, tested the subjects, performed the statistical analysis and discussed the data; KI designed the study and drafted the manuscript; MO reviewed the previous studies and discussed the results; BN reviewed the previous studies and discussed the data; and DS leaded the project reviewed the statistics, discussed the data and drafted the manuscript. All authors have read and approved the final version. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/14/88/prepub> Supplementary Material ====================== ###### Additional file 1 Questionnaire. ###### Click here for file ###### Additional file 2 Multimedia presentation of the project. ###### Click here for file Acknowledgements ================ Support of the Ministry of Science, Education and Sport of the Republic of Croatia is gratefully acknowledged (Project No 315-1773397-3407). The authors express their deepest gratitude to Dr Mario Bizzini (Schulthess Clinic, Zurich, Switzerland) for kindly providing the original questionnaire, which was adapted for the purpose of this study, translated into Croatian and applied accordingly, and for his overall support to this investigation. Additionally, we deeply appreciate the support of the Croatian Football Federation and the professional courtesy of all participating colleague referees. We are also particularly grateful for the reviewers of the article, Dr Wilson, Dr Robertson, and Dr Bizzini, for their cooperation, critiques and suggestions, which have definitively improved the quality of the investigation and the paper.

Background {#Sec1} ========== Genomic region and single-base level data retrieval and processing, which represent fundamental steps in genomic analyses such as copy number estimation, variant calling and quality control, still constitute one of the major bottlenecks in NGS data analysis. To deal with the computationally intensive task of calculating depth of coverage and pileup statistics at specific chromosomal regions and/or positions, different tools have been developed. Most of them, including specific modules of SAMtools \[[@CR1]\] and BEDTools \[[@CR2]\] and the most recent Mosdepth \[[@CR3]\], only measure and optimize the computation of depth of sequencing coverage. Few others, like the *pileup* modules of SAMtools, Sambamba \[[@CR4]\], GATK \[[@CR5]\] and ASEQ \[[@CR6]\] provide instead statistics at single-base resolution, which is essential to perform variant calling, allele-specific analyses and exhaustive quality control. Although most of these tools offer parallel computation options, scalability in terms of memory and multiple processes/threads usage is still limited. To enable an efficient exploration of large scale NGS datasets, here we introduce PaCBAM, a tool that provides fast and scalable processing of targeted re-sequencing data of varying sizes, from WES to small gene panels. Specifically, PaCBAM computes depth of coverage and allele-specific pileup statistics at regions and single-base resolution levels and provides data summary visual reporting utilities. PaCBAM introduces also an innovative and efficient *on-the-fly* read duplicates filtering approach. While most tools for read duplicates filtering work on SAM/BAM files sorted by read name \[[@CR1], [@CR7]\] or read position (Tarasov et al., 2015, [broadinstitute.github.io/picard](http://broadinstitute.github.io/picard)) and generate new SAM/BAM files, PACBAM performs the filtering directly during the processing, not requiring the creation of intermediate BAM/SAM files and fully exploiting parallel resources. Implementation {#Sec2} -------------- PaCBAM is a command line tool written in C programming language that combines multi-threaded computation, SAMTools APIs, and an ad-hoc data structures implementation. PaCBAM expects as input a sorted and indexed BAM file, a sorted BED file with the coordinates of genomic regions (namely the *target*, e.g. captured regions of a WES experiment), a VCF file specifying a list of SNPs of interest within the *target* and a reference genome in FASTA format. PaCBAM implements a multi-threaded solution that optimizes the execution time when multiple cores are available. The tool splits the list of regions provided in the BED file and spawns different threads to execute parallel computations using a shared and optimized data structure. The shared data structure collects both region and single-base level information and statistics which are processed and finally exposed through four different output options. Each output mode provides the user with only the statistics of interest, generating a combination of the following text output files: a) *depth of coverage of all genomic regions*, which for each region provides the mean depth of coverage, the GC content and the mean depth of coverage of the sub-region (user specified, default 0.5 fraction) that maximizes the coverage peak signal, to account for the reduced coverage depth due to incomplete match of reads to the captured regions (Additional file [1](#MOESM1){ref-type="media"}: Figure S1); b) *single-base resolution pileup*, which provides for each genomic position in the target the read depth for the 4 possible bases (A, C, G and T), the total depth of coverage, the variants allelic fraction (VAF), the strand bias information for each base; c) *pileup of positions with alternative base support*, which extracts the pileup statistics only for positions with positive VAF, computed using the alternative base with highest coverage (if any); d) *pileup of SNPs positions*, which extracts the pileup statistics for all SNPs specified in the input VCF file and uses the alternative alleles specified in the VCF file for the VAF calculation and the genotype assignment (Additional file [1](#MOESM1){ref-type="media"} for details). All output files are tab-delimited text files and their format details are provided in the Additional file [1](#MOESM1){ref-type="media"}. PaCBAM allows the user to specify the minimum base quality score and the minimum read mapping quality to filter out reads during the pileup processing. In addition, we implemented an efficient *on-the-fly* duplicated reads filtering strategy which implements an approach that is similar to the Picard MarkDuplicates method but that applies the filter during region and single-base level information retrieval and processing without the need of creating new BAM files (Additional file [1](#MOESM1){ref-type="media"}). The filtering strategy, which fully exploits multi-core capabilities, uses single or paired read alignment positions (corrected for soft-clipping at the 5′ end) and total mapping size information to identify duplicates and implements ad-hoc data structures to obtain computational efficiency. PaCBAM package also includes a Python script to generate visual data reports which can be directly used for quality control. Reports include plots summarizing distributions of regions and per-base depth of coverage, SNPs VAF distribution and genotyping, strand bias distribution, substitutions spectra, regions GC content (Additional file [1](#MOESM1){ref-type="media"}: Figure S2-S8). Results {#Sec3} ======= PaCBAM performances were tested on an AMD Opteron 6380 32-cores machine with 256 GB RAM. To mimic different application scenarios, we measured the execution time and memory used by PaCBAM to compute pileups from multiple input BAM files spanning different depth of coverage and different target sizes (Additional file [1](#MOESM1){ref-type="media"}: Table S1) using an increasing number of threads. We compared PACBAM performances against pileup modules of SAMtools, Sambamba and GATK (SAMtools offer no parallel pileup option). In terms of runtime, as shown in Fig. [1](#Fig1){ref-type="fig"}a and Additional file [1](#MOESM1){ref-type="media"}: Figure S9-S11, PaCBAM and Sambamba are the only tools that scale with the number of threads used. PaCBAM outperforms all other tools in all tested conditions. Of note, while PaCBAM pileup output files are of constant size, output files of SAMtools, Sambamba and GATK have a size that is function of the coverage; among all the experiments we run in the performance analyses, PaCBAM output is up to 17.5x smaller with respect to outputs generated by the other tested tools. Fig. 1PaCBAM performances. Time (**a**) and memory (**b**) required by PaCBAM to perform a pileup compared to SAMtools, GATK and Sambamba, using increasing number of threads. The figure focuses on the analysis of a BAM file with \~300X mean coverage and \~30Mbp target size using 30 threads. Note that parallel pileup option is not available for SAMtools and red lines in panel **a** and **b** refer to the average of single thread executions While GATK and PaCBAM, as shown in Fig. [1](#Fig1){ref-type="fig"}b and Additional file [1](#MOESM1){ref-type="media"}: Figure S12-S14, have a memory usage that depends only on the target size, Sambamba usage depends on both target size and number of threads and SAMtools usage is constant. Above 8 cores, PaCBAM beats both GATK and Sambamba in all tested conditions in memory usage. As an example of performance comparison, when analyzing a BAM file with \~300X mean coverage and \~30Mbp target size using 30 threads (Fig. [1](#Fig1){ref-type="fig"}a-b), PaCBAM improves execution time of 4.9x/5.27x and requires 80%/82% less memory compared to Sambamba/GATK. Of note, in the sequencing scenarios here considered, PaCBAM demonstrates up to 100x execution time improvement and up to 90% less memory usage with respect to the single-base pileup module of our previous tool ASEQ (Additional file [1](#MOESM1){ref-type="media"}: Figure S15). Depth of coverage and pileup statistics of PaCBAM pileup were compared to GATK results on a BAM file with \~300X average coverage and \~64Mbp target size observing almost perfect concordance (Fig. [2](#Fig2){ref-type="fig"}a-b). Fig. 2Comparison of PaCBAM results with other tools. **a** Comparison of PaCBAM and GATK depth of coverage (left) with zoom in the coverage range \[0,500\] (right); number of positions considered in the analysis and correlation results are reported. **b** Comparison of allelic fraction of \~ 40 K positions annotated as SNPs in dbSNP database v144 and having an allelic fraction \> 0.2 in both PaCBAM and GATK pileup output. **c** Single-base coverage obtained by running either *Picard MarkDuplicates + PaCBAM pileup* or *PaCBAM pileup with duplicates filtering* option active (left) with zoom in the coverage range \[0,500\] (right). **d** Regional mean depth of coverage obtained by running either *Picard MarkDuplicates + PaCBAM pileup* or *PaCBAM pileup with duplicates filtering* option active PaCBAM duplicates removal strategy was tested by comparing PaCBAM pileups obtained from a paired-end BAM file first processed with Picard MarkDuplicates or parallel Sambamba markdup, to PaCBAM pileups obtained from the same initial BAM file but using the embedded on-the-fly duplicates filtering. As shown in Fig. [2](#Fig2){ref-type="fig"}c-d and Additional file [1](#MOESM1){ref-type="media"}: Figure S16, both single-base and region level statistics results are strongly concordant, with single-base total coverage difference (with respect to Picard) that in 99.94% of positions is \< 10X, single-base allelic fraction difference that in 99.95% of positions is \< 1% and region mean coverage difference that in 99.96% of regions is \<10X (Additional file [1](#MOESM1){ref-type="media"}: Figure S17). In addition, PaCBAM strategy improves overall execution time of 2.5x/1.7x with a single thread and of 25x/3x with 30 threads compared to Picard and parallel Sambamba, respectively (Additional file [1](#MOESM1){ref-type="media"}: Table S2, Fig. [2](#Fig2){ref-type="fig"}c, Additional file [1](#MOESM1){ref-type="media"}: Figure S16A). Overall, these analyses demonstrate that PaCBAM exploits parallel computation resources better than existing tools, resulting in evident reductions of processing time and memory usage, that enable a fast and efficient coverage and allele-specific characterization of large WES and targeted sequencing datasets. The performance analysis is completely reproducible using an ad-hoc Debian-based Singularity container (Additional file [1](#MOESM1){ref-type="media"}: Table S3). Conclusion {#Sec4} ========== We presented PaCBAM, a fast and scalable tool to process genomic regions from NGS data files and generate coverage and pileup statistics for downstream analysis such as copy number estimation, variant calling and data quality control. Although designed for targeted re-sequencing data, PaCBAM can be used to characterize any set of genomic regions of interest from NGS data. PaCBAM generates both region and single-base level statistics and provides a fast and innovative *on-the-fly* read duplicates filtering strategy. The tool is easy to use, can be integrated in any NGS pipeline and is available in source/binary version on Bitbucket and containerized from Docker and Singularity hubs. Availability and requirements {#Sec5} ============================= Project name: PaCBAM Project home page: [bcglab.cibio.unitn.it/PaCBAM](http://bcglab.cibio.unitn.it/PaCBAM) Operating system(s): Platform independent Programming language: C, Python License: MIT Additional file =============== {#Sec6} **Additional file 1: Figure S1.** Genomic region mean coverage computation. **Figure S2.** Cumulative coverage distribution report. **Figure S3.** Variant allelic fraction distribution report. **Figure S4.** SNP allelic fraction distribution report. **Figure S5.** Alternative bases distribution report. **Figure S6.** Strand bias distribution report. **Figure S7.** Genomic regions depth of coverage distribution report. **Figure S8.** Genomic regions GC content distribution report. **Figure S9.** Run time comparison at 150X depth of coverage. **Figure S10.** Run time comparison at 230X depth of coverage. **Figure S11.** Run time comparison at 300X depth of coverage. **Figure S12.** Memory usage comparison at 150X depth of coverage. **Figure S13.** Memory usage comparison at 230X depth of coverage. **Figure S14.** Memory usage comparison at 300X depth of coverage. **Figure S15.** Memory usage comparison among PaCBAM pileup and pileup module of ASEQ. **Figure S16.** Comparison of PaCBAM duplicates filtering strategy to Sambamba markdup and Picard MarkDuplicates modules. **Figure S17.** Performance of PaCBAM duplicated reads filtering. **Table S1.** Mean depth of coverage and target sizes of all BAM files used to test PaCBAM performance.**Table S2.** Time and memory usage of duplicates filtering performance analyses. **Table S3.** Versions of the tools used in performance evaluation analysis. NGS : Next-Generation Sequencing SNP : Single Nucleotide Polymorphism VAF : Variant(s) Allele Frequency WES : Whole-Exome Sequencing **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s12864-019-6386-6. Not applicable. AR designed and implemented PaCBAM. SV designed and implemented visual reporting scripts and performed all performance analyses. TF and FD contributed with tool testing, access to computational resources and performance analyses. AR supervised the project. All authors contributed to the writing and editing of the manuscript and approved the manuscript. The research leading to these results has received funding from AIRC under MFAG 2017 - ID. 20621 project - P.I. Romanel Alessandro - for the design, implementation and performance analyses and from NCI P50 CA211024-01 Weill Cornell Medicine Prostate Cancer SPORE - Demichelis Francesca - for testing and performance analyses. All data and analysis scripts supporting the results of this article are available at bcglab.cibio.unitn.it/PaCBAM_Performance_Analysis. Not applicable. Not applicable. The authors declare that they have no competing interests.

Infection with the Asian lineage of highly pathogenic avian influenza (HPAI) A(H5N1) virus (prototype strain A/goose/Guangdong/1/96 \[Gs/GD\]) has resulted in substantial losses in the poultry industry and poses a threat to public health worldwide. According to the World Health Organization (WHO)/World Organisation for Animal Health (OIE)/Food and Agriculture Organization of the United Nations (FAO) H5N1 Evolution Working Group, 10 distinct clades of these viruses (0--9) were initially designated in 2008 to characterize the phylodynamics of the hemagglutinin (HA) gene of the Gs/GD-like viruses circulating during 1996--2007 ([@R1]). On the basis of these nomenclature criteria, new second-, third-, and fourth-order clades have been identified within the previously defined clades in the phylogenetic analyses that were updated in 2009 and 2011 ([@R2],[@R3]). Therefore, as HPAI A(H5N1) virus continues to undergo substantial evolution, extensive genetic divergence is expected to periodically accumulate to form novel monophyletic groups. To identify continued divergence of clade 2.3.4 other than the recognized subclades 2.3.4.1, 2.3.4.2, and 2.3.4.3, we characterized 7 HPAI A(H5N1) viruses isolated from poultry during 2009--2012 in China ([@R2]). The Study ========= As part of continuous national avian influenza virus surveillance, we performed a monthly collection of cloacal swabs from various poultry species (chicken, duck, goose, quail, and pigeon) at a wholesale live-bird market (LBM) in Yangzhou, Jiangsu Province, in eastern China. Birds offered for retail sale in the LBM were mainly from local farms in Jiangsu and the neighboring provinces in eastern China; some were transported from regions in southern or northern China. Virus isolation and identification were conducted as described ([@R4]). During December 2009--September 2012, avian influenza virus isolates belonging to 8 HA subtypes (H1, H3--H6, H9--H11) were identified; 7 of the isolates were subtype H5N1 ([Table 1](#T1){ref-type="table"}). ###### Results of HI assays using Re-5 and k0602 antiserum for 7 avian influenza A(H5N1) viruses isolated in China, 2009--2012\* Isolate Isolation date Antibody titer, log~2~ ------------------------------------------- ---------------- ------------------------ -------- A/chicken/Jiangsu/WJ/2009(H5N1) 2009 Dec 4 8 A/chicken/Jiangsu/XZ/2010(H5N1) 2010 Mar 4 8 A/chicken/northern China/k0602/2010(H5N1) 2010 May 4 **10** A/chicken/Shandong/k0603/2010(H5N1) 2010 Jun 4 10 A/chicken/eastern China/ZG56/2011(H5N1) 2011 Dec 4 5 A/chicken/eastern China/JX/2011(H5N1) 2011 Dec 3 7 A/chicken/eastern China/AH/2012(H5N1) 2012 Sep 4 1 Re-5 diagnostic antigen† NA **10** 4 \*Re-5 and k0602 antiserum were generated by vaccinating specific-pathogen free chickens with the commercial Re-5 vaccine (Qingdao Yebio Bioengineering Co., Ltd, Qingdao, China) and the oil-emulsified inactivated A/chicken/northern China/k0602/2010(H5N1) vaccine, respectively. HI titers against the homologous antigen/virus are shown in **boldface**. HI, hemagglutination inhibition; NA, not applicable.
†Qingdao Yebio Bioengineering Co., Ltd, China. To characterize these 7 isolates, we sequenced the HA genes to determine clade distribution. In the reconstructed phylogenetic tree ([Figure](#F1){ref-type="fig"}) using reference sequences retrieved from the GenBank database and partially recommended by WHO/OIE/FAO ([@R2]), the 7 isolates belonged to clade 2.3.4 (the "Fujian-like" sublineage), which has been prevalent in China since 2005 ([@R5]). However, none of the isolates could be further classified into previously identified subclades 2.3.4.1, 2.3.4.2, or 2.3.4.3. Six of the viruses closely resembled A/peregrine falcon/Hong Kong/810/2009, and the remaining virus was highly homologous with recent H5 viruses bearing various neuraminidase (NA) subtypes (N1, N2, N5, and N8). ). Numbers above or below the branch nodes denote bootstrap values [\>]{.ul}60% with 1,000 replicates. Numbers on the right are existing (2.3.3, 2.3.4.1, 2.3.4.2, 2.3.4.3, 2.5) and proposed (2.3.4.4, 2.3.4.5, 2.3.4.6) virus subclades. Black triangles indicate the 7 variants identified in this study; GenBank accession numbers for their HA genes are KC631941--KC631946 and KC261450. Scale bar indicates nucleotide substitutions per site.](13-0340-F){#F1} According to WHO/OIE/FAO guidelines ([@R1]--[@R3]), new clades (including subclades) were specified not only with a bootstrap value of [\>]{.ul}60 at the clade-defining node in which sequences monophyletically arose from a common ancestor but also with average between-clade and within-clade nucleotide divergences of \>1.5% and \<1.5%, respectively. Apart from subclades 2.3.4.1, 2.3.4.2, and 2.3.4.3, we found 3 additional monophyletic categories---the A/peregrine falcon/Hong Kong/810/2009-like viruses, the HPAI subtype H5N5--like reassortants, and the HPAI subtype H5N2/H5N8--like reassortants---that grouped clearly within the tree ([Figure](#F1){ref-type="fig"}). The bootstrap values and average within-clade and between-clade distances for these 3 groups were 81, 1.0%, 4.2%; 100, 1.0%, 5.2%; and 100, 1.3%, 5.3%, respectively. Because of the compulsory vaccination practice against HPAI in China ([@R6]), we examined serologic cross-reactivity between the 7 subtype H5N1 isolates and the diagnostic antigen of the widely used inactivated reassortant H5N1/PR8 vaccine Re-5 ([Table 1](#T1){ref-type="table"}). Although Re-5 derived its HA and NA genes from a clade 2.3.4 representative virus A/duck/Anhui/1/2006, the hemagglutination inhibition (HI) titers of Re-5 antiserum against the 7 subtype H5N1 viruses were as much as 6--7 log~2~ lower than that against the homologous antigen. In contrast, the antiserum of A/chicken/northern China/k0602/2010 (k0602) showed limited reaction to Re-5 and A/chicken/eastern China/AH/2012. Moreover, antigenic variation also existed among the 6 A/peregrine falcon/Hong Kong/810/2009-like viruses, as highlighted by the HI assay using k0602 antiserum ([Table 1](#T1){ref-type="table"}). To explore whether these antigenic variations can be translated into protection efficacy difference in vivo, we selected A/chicken/northern China/k0602/2010 (k0602) and A/chicken/Shandong/k0603/2010 (k0603) viruses to evaluate the bivalent inactivated reassortant H5N1/PR8 vaccine Re-4/Re-5 (the HA and NA genes of Re-4 are from a clade 7 virus A/chicken/Shanxi/2/2006). This vaccine has been extensively used to control the prevalence of clade 2.3.4 and clade 7 viruses in China since 2008 ([@R6]). In addition, a reassortant rk0602 virus, which carries the HA and NA genes of k0602 virus and the internal genes of PR8, was recovered by using reverse genetics and the inactivated rk0602 vaccine was applied to evaluate the homologous protection. Four-week-old specific-pathogen free chickens were vaccinated with Re-4/Re-5 or the rk0602 vaccine and readily developed specific antibodies against the component viruses by day 28 after vaccination ([Table 2](#T2){ref-type="table"}). The birds were then intranasally challenged with 10^6.0^ 50% egg infectious dose of k0602 or k0603 virus. During the 10-day observation period, the Re-4/Re-5 vaccinated birds displayed clinical signs including severe depression, ruffled feathers, huddling, decreased feed and water consumption, and diarrhea; moreover, only 14.3% (1/7 birds in the k0602 group) and 10% (1/10 birds in the k0603 group) of the challenged chickens survived, reflecting poor protection by the Re-4/Re-5 vaccine. In addition, shed virus was detected in tracheal and cloacal swabs from most of the tested chickens on 3 and 5 days postchallenge. By contrast, the rk0602-vaccinated chickens all survived the challenge, and no virus was recovered from tracheal and cloacal samples ([Table 2](#T2){ref-type="table"}). ###### Efficacy of vaccines against highly pathogenic avian influenza virus A(H5N1) clade 2.3.4 variants in chickens, China\* Virus and vaccine type HI titer ±SD, log~2~ Challenge test results, by swab type, no. positive birds/no. tested (mean titer ±SD)† No. surviving birds/total no. ------------------------ ---------------------- ------------ --------------------------------------------------------------------------------------- ------------------------------- ------------------ ------------------ -- ------------------ ------------------ ------- k0602 Re-4/Re-5 8.10 ±0.97 6.20 ±0.31 2.25 ±0.45 6/6 (3.79 ±1.46) 4/6 (2.25 ±2.39) 2/3 (2.17 ±2.40) 2/3 (2.92 ±2.55) 1/7 rk0602 2.10 ±0.37 3.27 ±0.33 9.35 ±0.75 0/10 0/10 0/10 0/10 10/10 Control‡ ND ND ND ND ND ND ND 0/5 k0603 Re-4/Re-5 7.70 ±0.26 5.70 ±0.29 1.93 ±1.21 8/8 (3.15 ±1.30) 8/8 (3.50 ±0.25) 3/3 (3.50 ±0.43) 3/3 (2.33 ±0.58) 1/10 rk0602 2.05 ±0.85 2.75 ±0.57 9.25 ±0.71 0/10 0/10 0/10 0/10 10/10 Control‡ ND ND ND ND ND ND ND 0/5 \*Chickens were immunized with the Re4/Re5 or the inactivated rk0602 vaccine (the HA and NA genes of rk0602 were derived from subtype H5N1 k0602 virus; the internal genes were from PR8), and HI antibody titers were determined on day 28 postvaccination. HI, hemagglutination inhibition assay; dpc, days postchallenge; ND, not done.
†Birds were challenged with 10^6.0^ 50% egg infectious dose (EID~50~) of k0602 or k0603 virus; virus titers are expressed as log~10~ EID~50~/0.1 mL.
‡Two groups of 5 mock-vaccinated chickens served as controls; all died within 3 dpc. Conclusions =========== The location of the 7 HPAI A(H5N1) virus variants in the HA gene tree ([Figure](#F1){ref-type="fig"}) suggests that novel monophyletic subclades other than the previously identified 2.3.4.1, 2.3.4.2, and 2.3.4.3 subclades continue to emerge within clade 2.3.4. As a result of our findings, we suggest that these groups should be assigned new fourth-order clades of 2.3.4.4, 2.3.4.5, and 2.3.4.6 to reflect the wide divergence of clade 2.3.4 viruses. In China, 1 of the 6 countries to which subtype H5N1 virus is endemic ([@R7]), multiple distinct clades (2.2, 2.5, 2.3.1, 2.3.2, 2.3.3, 2.3.4, 7, 8, and 9) were identified by surveillance during 2004--2009 ([@R5]). In particular, clades 2.3.2, 2.3.4, and 7 viruses have gained ecologic niches and have continued circulating by further evolving into new subclades ([@R2]). In addition, various NA subtypes of H5 viruses (H5N5, H5N8, and H5N2) bearing the genetic backbone of clade 2.3.4 A(H5N1) viruses have been detected in ducks, geese, quail, and chickens ([@R8]--[@R12]). These findings highlight the importance of periodic updates of the WHO/OIE/FAO classification of Asian A(H5N1) viruses by continuous surveillance to better understand the dynamic nature of the viral evolution. Our findings have implications for the effectiveness of vaccination of chickens against HPAI A(H5N1) viruses. The results of cross-HI assays ([Table 1](#T1){ref-type="table"}) and vaccine efficacy experiments ([Table 2](#T2){ref-type="table"}) indicate antigenic drift in subtype H5N1 variants, as compared with the vaccine strain specifically designed to control the prevalent clade 2.3.4 virus infection in poultry. Although previous studies by Tian et al. ([@R13]) and Kumar et al. ([@R14]) proposed that vaccinated chickens with HI antibody titers of \>4 log~2~ could be protected from virus challenge, our data demonstrate that vaccine efficacy is substantially influenced by antigenic matching between the vaccine strain and circulating viruses in preventing the replication and transmission of influenza virus, especially when the induced antibodies are of moderate titers. *Suggested citation for this article*: Gu M, Zhao G, Zhao K, Zhong L, Huang J, Wan H, et al. Novel variants of clade 2.3.4 highly pathogenic avian influenza A(H5N1) viruses, China. Emerg Infect Dis \[Internet\]. 2013 Dec \[*date cited*\]. <http://dx.doi.org/10.3201/eid1912.130340> This work was supported by the Major State Basic Research Development Program of China (973 Program; (grant no. 2011CB505003), the Jiangsu Provincial Natural Science Foundation of China (grant no. BK20130442), the earmarked fund for Modern Agro-industry Technology Research System (nycytx-41-G07), and the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). Dr Gu is a faculty member at the College of Veterinary Medicine, Yangzhou University. Her primary research interest is the epidemiology and evolutionary mechanism of avian influenza viruses.

Introduction {#s1} ============ From a biological point of view, anticipating the emotional content of an upcoming event according to environmental cues may help individuals in preparing adaptive reactions to approach the benefit and to avoid the harm. However, people are living in an ever-changing world; there is often uncertainty about which emotional content of the event will actually occur. This uncertainty has been found to be associated with anxiety (Grupe and Nitschke, [@B10]). Identifying the processing of uncertainty may contribute to understanding the detrimental effects of uncertainty on well-being and psychological symptoms. Studies have indicated that uncertainty modulates neural activity to emotional pictures before their occurrences (during the anticipation period; e.g., Onoda et al., [@B24]; Lin et al., [@B15]). In our previous Event-related potential (ERP) study (Lin et al., [@B15]), for example, neutral symbols (e.g., arrows; cues) that uncertainly as compared to certainly signify the emotional content of the upcoming picture generally evoked larger N2 amplitudes. When the following picture was negative, early contingent negative variation (CNV)[^1^](#fn0001){ref-type="fn"} was reduced for uncertain as compared to certain cues. Meanwhile, several studies investigated the uncertainty effects after the occurrences of the emotional pictures. However, results are under debate in both functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) studies. In terms of the fMRI studies, Sarinopoulos et al. ([@B31]) showed that neural activity in insula and amygdala was stronger for negative pictures following uncertain as compared to certain cues; whereas this uncertainty effect was not found in Onoda et al.\'s ([@B25]) study, regardless of the emotional contents (positive and negative) of the pictures. For EEG studies, Onoda et al.\'s ([@B23]) visual evoked magnetic fields (VEF) study found larger M120 amplitudes for uncertain compared to certain negative pictures. Using the same paradigm, however, uncertain and certain emotional (positive and negative) pictures were found to have similar lower-1, lower-2, and upper alpha[^2^](#fn0002){ref-type="fn"} in event-related desynchronization/synchronization (ERD/ERS; Onoda et al., [@B24]). Accordingly, the authors suggested that the sensory and cognitive processing of emotional pictures are unaffected by uncertainty (Onoda et al., [@B24]). In the above-mentioned fMRI studies (Onoda et al., [@B25]; Sarinopoulos et al., [@B31]), the temporal resolution of BOLD fMRI was \~ 2 s (i.e., TR = 2 s). However, the effect of uncertainty on emotional stimuli seems to occur only in the first second relative to the onset of the stimuli (Onoda et al., [@B23]; Gole et al., [@B8]; Yang et al., [@B38]). In this case, the temporal resolution may be not high enough to precisely and accurately capture the time course of the uncertainty effect, which results in altering the uncertainty effects. Regarding to EEG studies, while VEF and ERD/ERS are thought to have high temporal resolution, it is still unclear whether the indices (i.e., lower-1, lower-2, and upper alpha) used in Onoda et al.\'s ([@B23], [@B24]) studies are suitable to investigate the effects related to emotional anticipation. Emotional pictures following anticipatory cues as compared to these pictures without any anticipatory cues were shown to be similar in VEF (Onoda et al., [@B23]) and to be differential in lower-2 alpha only at late stages of processing (e.g., \>300 ms; Onoda et al., [@B24]). Using ERPs that also have high temporal resolution, however, we found the anticipation effects at both early and late stages (Lin et al., [@B14], [@B16]). Due to temporal resolution and indices, ERPs are ideal to investigate neural responses during the perception of pictures which emotional contents are certain or uncertain. To the best of our knowledge, two ERP studies have investigated whether uncertainty modulates the processing of emotional pictures (Gole et al., [@B8]; Yang et al., [@B38]). Gole et al. ([@B8]) showed that emotionally uncertain compared to certain pictures evoked greater early posterior negativity (EPN; overlapping with N2) and smaller late positive potential (LPP) amplitudes, regardless of the emotional contents (negative and neutral) of the pictures. Using facial pictures only, Yang et al. ([@B38]) found larger P2 amplitudes for emotionally (fearful and neutral) uncertain as compared to certain faces. For fearful faces, the N2 amplitudes were smaller after uncertain as compared to certain cues. In Gole et al.\'s ([@B8]) study, however, there was a long anticipatory phase (i.e., inter-trial interval, ITI) between the cue and the picture (6 s). Participants may have enough time to prepare for the upcoming event during the phase, which may reduce the effect of uncertainty (Lin et al., [@B15]). Different from previous studies in which certain and uncertain cues were used by neutral symbols (Onoda et al., [@B23], [@B24], [@B25]; Sarinopoulos et al., [@B31]; Gole et al., [@B8]); Yang et al. ([@B38]) used emotional pictures and neutral symbols (i.e., "+") as certain and uncertain cues, respectively. For fearful faces, certain and uncertain cues differed not only in the meanings but also in some other factors (e.g., the emotional contents and composition of the cues). Therefore, it is still unclear whether the uncertainty effects observed in the study were relevant in the meanings of the cues or these factors. Therefore, our present study aimed to further investigate whether uncertainty about the emotional content of an upcoming picture modulates the ERPs to the picture. To address this issue, EEG was recorded while participants viewed emotional pictures. Each trial started with a cue, followed by an anticipatory interval and subsequently, a picture. The cue either indicated the emotional content of the following picture or not. All the cues were used by neutral symbols (e.g., arrows). The ITI between the cue and the picture was short (about 2 s) to enhance the effect of uncertainty. Furthermore, as emotional valence and arousal may influence the uncertainty effect and we were interested in the modulation of valence on the uncertainty effect, the present study used positive and negative pictures which differed in valence but not in arousal. Based on the above-mentioned studies, we expected that uncertain as compared to certain emotional pictures might evoke greater P2 and smaller LPP amplitudes. However, it was still unclear whether uncertainty modulates the N2 amplitudes for emotional pictures. Methods {#s2} ======= Participants ------------ Nineteen undergraduate and postgraduate students (11 females and 8 males; age: *M* ± *SD* = 22.42 ± 1.85) were recruited in South China Normal University via advertisements in return for the compensation of 30 RMB for this study. All participants were right-handed, as assessed through the Edinburgh Handedness Inventory (Oldfield, [@B21]). Participants had normal or corrected-to-normal vision. No participants reported any medical, neurological or psychiatric disorders. Participants were told that the study was to investigate the neural activity to emotional pictures and written informed consent was given prior to the experiment. Ethical approval was obtained from the Ethics Committee of School of Psychology, South China Normal University. Stimuli ------- Stimuli were the same as those in our previous studies (Lin et al., [@B14], [@B15],[@B16]). Stimuli were 160 colored pictures (80 positive and 80 negative) that were obtained from Chinese Affective Picture System (CAPS; Bai et al., [@B2]). All the pictures were converted to gray-level, were of the same size (11 × 8 cm, 6.30 × 4.58°) and resolution (100 pixels per inch) and were aligned in the non-emotional features (e.g., luminance, contrast, and composition). The pictures were rated on valence and arousal using a 9-point scale ranging from "1" (extremely unpleasant) to "9" (extremely pleasant) and "1" (low arousal) to "9" (high arousal), respectively, by another 22 undergraduate and postgraduate students (11 females, 19--25 years, *M* ± *SD* = 21.24 ± 1.75). The ratings of positive and negative pictures showed significant difference in valence \[*t*~(21)~ = 10.55, *p* = 7.48E-10, 6.88 ± 1.03 vs. 3.06 ± 1.28\], but not in arousal \[*t*~(21)~ = −0.03, *ps.* \> 0.05, 5.36 ± 1.51 vs. 5.38 ± 1.77\]. Procedure --------- Participants were seated in a soundproof and dimly room approximately 1 m directly in front of a computer monitor. Stimuli were presented using E-prime 1.0 software (Psychology Software Tools, Inc., Pittsburgh, PA, USA) on a black screen in the center of a 17″ monitor with a screen resolution of 1024 by 768 pixels. All stimuli were presented against a gray background. Every trial started with a black fixation cross for 500 ms, replaced by a random blank screen varying from 1000 to 2000 ms (*M* = 1500 ms). A bidirectional, a left or a right arrow (1.03 × 0.23°) was randomly presented for 200 ms. The bidirectional arrow served as the uncertain cue in that it was randomly followed by a positive or negative picture with equal conditional probabilities. The right and the left arrow served as the certain cue for a subsequent positive and a negative picture, respectively. A positive or a negative picture was shown at the center of the screen for 1000 ms after another blank screen of random duration between 1600 and 2000 ms (*M* = 1800 ms). Participants were told the meaning of the cue before the formal experiment and were asked to view the cues and the pictures during their presentations. After another blank screen was shown for 200 ms following the picture, participants were asked to rate the pleasantness of the picture on a 9-point scale ranging from "1" (extremely unpleasant) to "9" (extremely pleasant). There was no time limit for the response. The subsequent trial started after another blank screen for 1500 ms. Each picture described in the Stimuli Section was used twice, once after uncertain cues and once after certain cues. That is, the task comprised a total of 320 trials. Before the formal experiment, there were 20 trials of practice to familiarize with the experimental procedure. The pictures used in the formal experiment were not used in the practice. EEG recording ------------- EEG was continuously recorded using a NEUROSCAN Synamps2 AC-amplifier (Neuroscan, Inc., Sterling, VA, USA). The Ag/AgCl electrodes were placed upon the scalp by a 32 channel Quick-Cap in accordance with the international extended 10--20 system (FP1, FP2, F7, F3, Fz, F4, F8, FT7, FC3, FCz, FC4, FT8, T7, C3, Cz, C4, T8, TP7, CP3, CPz, CP4, TP6, P7, P3, Pz, P4, P8, O1, Oz, and O2). EEG electrodes were connected to ground and were referenced to the right mastoid online. The horizontal electrooculogram (EOG) was recorded from two electrodes at the outer canthi of both eyes, and the vertical EOG was monitored bipolarly from electrodes above and below the left eye. The EEG was amplified using a 0.05--100 Hz band-pass and sampled at 1000 Hz/channel with a 50 Hz notch filter. Electrode impedances were maintained below 5 kΩ. The EEG data were analyzed offline using the SCAN 4.3 software. Raw EEG signals were digitally re-referenced to the average of two mastoids. Ocular movements were inspected and removed based on the default parameter of the SCAN 4.3 ocular artifact tool. EEG was then segmented from −200 ms until 1000 ms relative to the onset of the picture, with the first 200 ms serving as a baseline. Artifact rejection was carried out using an amplitude threshold of 100 μV. Trials were averaged separately for each channel and each experimental condition and averaged ERPs were digitally low-pass filtered at 30 Hz (24 db/oct, zero phase shift, Butterworth). ERPs were quantified using mean amplitudes for P2 (130--180 ms), N2 (220--300 ms), early LPP (350--450 ms), and late LPP (550--1000 ms). P2 and N2 were measured at frontal (F3, Fz, F4), frontal-central (FC3, FCz, FC4), and central (C3, Cz, C4) electrodes. Early and late LPP were measured at frontal (F3, Fz, F4), frontal-central (FC3, FCz, FC4), central (C3, Cz, C4), central-parietal (CP3, CPz, CP4), and parietal (P3, Pz, P4) electrodes. The time window for P2 was chosen to correspond with peaks identified in the grand waveforms across all conditions (155 ms) and previous studies (Gole et al., [@B8]; Lin et al., [@B14], [@B16]) and time windows for N2, early, and late LPP were chosen based on visual inspection of the grand waveforms and previous studies (Gole et al., [@B8]; Lin et al., [@B14], [@B16]). The electrodes of interest for all the components were based on visual inspection of the grand waveforms and previous studies (Gole et al., [@B8]; Lin et al., [@B14], [@B16]; Giglio et al., [@B7]; Richards et al., [@B27]). The average number of trials was shown in Table [1](#T1){ref-type="table"}. ###### **Mean number of trials for each experimental condition**. **Uncertain positive** **Certain positive** **Uncertain negative** **Certain negative** ------------------------ ---------------------- ------------------------ ---------------------- 72.42 71.63 74.11 74.37 Data analysis ------------- For statistical analysis of behavioral results, the ratings were analyzed in 2 × 2 repeated measures analyses of variance (ANOVAs) with uncertainty (uncertain vs. certain) and emotion (positive vs. negative) as within-subject factors. Mean ratings and their *SD* across conditions are presented in Figure [1](#F1){ref-type="fig"}. {#F1} For ERPs, the above-mentioned ANOVAs were performed separately for P2, N2, early, and late LPP. The analysis for P2 and N2 included site (Frontal vs. Frontal-Central vs. Central) and hemisphere (Left vs. Middle vs. Right) as within-subject factors. The analysis for early and late LPP included site (Frontal vs. Frontal-Central vs. Central vs. Central-Parietal vs. Parietal) and hemisphere (Left vs. Middle vs. Right) as within-subject factors. Grand-average waveforms and topographical maps of all the components are presented in Figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, respectively. *M* and *SD* of the mean amplitudes for these components are presented in Tables [2](#T2){ref-type="table"}--[5](#T5){ref-type="table"}. {#F2} {#F3} ###### **Mean (*M*) and standard deviation (*SD*) for P2 mean amplitudes (μV)**. **Uncertain positive** **Certain positive** **Uncertain negative** **Certain negative** ----- ------------------------ ---------------------- ------------------------ ---------------------- ------ ------ ------ ------ F3 4.65 4.94 4.72 4.91 1.55 4.44 2.67 5.18 Fz 4.80 5.64 4.88 5.60 1.58 4.99 2.62 5.73 F4 4.66 5.59 4.65 5.37 1.60 4.96 2.65 5.64 FC3 4.83 5.08 4.85 5.07 1.91 4.49 2.99 5.26 FCz 5.31 5.84 5.37 5.68 2.08 5.13 3.11 5.89 FC4 4.62 5.75 4.75 5.43 1.78 5.09 2.85 5.67 C3 4.59 4.42 4.56 4.58 2.17 3.95 3.06 4.83 Cz 5.89 5.80 5.84 5.80 2.97 5.06 3.60 5.89 C4 4.49 5.40 4.49 4.94 0.97 7.19 2.68 5.26 ###### ***M* and *SD* for N2 mean amplitudes (μV)**. **Uncertain positive** **Certain positive** **Uncertain negative** **Certain negative** ----- ------------------------ ---------------------- ------------------------ ---------------------- ------- ------ ------- ------ F3 −4.19 3.37 −4.45 3.69 −5.85 3.41 −4.97 3.85 Fz −5.23 4.17 −5.73 4.27 −7.16 4.01 −6.01 4.37 F4 −4.65 4.44 −5.43 4.56 −6.60 4.53 −5.41 4.63 FC3 −3.42 3.62 −3.68 3.96 −4.74 3.50 −3.96 4.25 FCz −4.96 4.52 −5.33 4.63 −6.60 4.19 −5.72 4.88 FC4 −4.28 4.63 −4.76 4.67 −5.90 4.66 −4.66 4.73 C3 −1.80 3.89 −2.02 4.37 −2.48 3.65 −1.88 4.83 Cz −3.47 5.16 −3.93 5.21 −4.46 4.56 −3.77 5.38 C4 −2.90 4.97 −3.72 5.00 −5.71 7.42 −3.12 5.18 ###### ***M* and *SD* for early LPP mean amplitudes (μV)**. **Uncertain positive** **Certain positive** **Uncertain negative** **Certain negative** ----- ------------------------ ---------------------- ------------------------ ---------------------- ------- ------ ------- ------ F3 2.40 5.50 1.72 5.04 −1.32 5.46 −0.52 5.48 Fz 1.91 6.35 0.95 5.77 −2.26 5.90 −1.24 6.04 F4 1.12 5.97 0.01 5.40 −2.33 5.86 −1.54 5.85 FC3 3.67 5.52 2.90 5.04 −0.06 5.33 0.67 5.51 FCz 2.95 6.18 2.01 5.48 −1.43 5.57 −0.55 5.95 FC4 1.91 5.72 1.11 4.98 −1.52 5.40 −0.63 5.44 C3 5.23 4.61 4.55 4.28 1.94 4.57 2.58 5.16 Cz 4.77 6.25 3.84 5.60 0.51 5.31 1.03 5.88 C4 3.45 4.75 2.78 4.27 −0.97 6.55 1.29 5.18 CP3 8.34 4.84 7.87 4.84 5.15 4.60 5.89 5.48 CPz 8.48 5.88 7.71 5.72 4.63 5.29 5.07 6.47 CP4 6.96 4.76 6.32 4.75 3.95 4.57 4.50 5.46 P3 10.80 5.22 10.59 5.23 8.55 5.14 9.09 5.56 Pz 12.14 6.40 11.72 6.19 9.04 5.65 9.44 6.48 P4 10.03 4.64 9.68 4.74 7.96 4.45 8.39 5.17 ###### ***M* and *SD* for late LPP mean amplitudes (μV)**. **Uncertain positive** **Certain positive** **Uncertain negative** **Certain negative** ----- ------------------------ ---------------------- ------------------------ ---------------------- ------- ------ ------ ------ F3 3.64 3.78 3.98 4.24 1.09 3.38 2.75 3.91 Fz 2.76 3.82 2.85 4.15 −0.03 3.48 2.06 4.10 F4 3.86 3.50 3.58 4.24 1.36 3.64 3.26 4.67 FC3 4.00 3.69 4.10 4.57 1.52 3.24 3.14 3.79 FCz 2.63 3.98 2.78 4.57 0.38 3.08 2.05 4.04 FC4 3.76 3.56 3.95 3.88 1.57 3.37 3.50 4.29 C3 4.70 2.98 4.81 3.63 2.93 2.98 4.03 3.48 Cz 4.03 4.24 4.20 4.75 2.08 3.98 3.14 4.65 C4 5.21 3.92 5.34 4.19 2.31 4.06 5.04 4.94 CP3 6.19 2.77 6.31 3.39 4.38 2.75 5.34 3.36 CPz 6.18 4.34 6.22 4.60 4.68 3.89 5.32 5.03 CP4 6.29 3.47 6.40 3.73 4.68 3.40 5.82 4.47 P3 6.14 3.02 6.44 3.56 5.14 2.93 5.70 3.60 Pz 6.18 4.65 6.61 4.91 5.44 4.23 5.80 4.81 P4 5.58 3.53 5.84 4.21 4.82 3.70 5.45 4.49 Degree of freedom and *p*-values of repeated measurements were corrected by Greenhouse-Geisser and *p*-values of *post-hoc* tests were corrected by Bonferroni correction. Please note that the effects which failed to reach statistical significance would not be reported (*ps.* \> 0.05). Results {#s3} ======= Behavioral results ------------------ The analysis showed a significant main effect of emotion \[*F*~(1,\ 18)~ = 289.85, *p* = 1.52E-12, $\eta_{p}^{2} = 0.94$\], with larger ratings for positive than negative pictures. The interaction between emotion and uncertainty was also significant \[*F*~(1,\ 18)~ = 17.37, *p* = 0.001, $\eta_{p}^{2} = 0.49$\]. For positive pictures, the ratings were higher in the certain as compared to the uncertain condition \[*t*~(18)~ = 4.34, *p* = 4.00E-4\]; for negative pictures, however, the ratings were higher in the uncertain as compared to the certain condition \[*t*~(18)~ = −3.47, *p* = 0.003\]. ERP results ----------- ### P2 components The analysis of P2 amplitudes revealed main effects of uncertainty \[*F*~(1,\ 18)~ = 4.97, *p* = 0.039, $\eta_{p}^{2} = 0.22$\], emotion \[*F*~(1,\ 18)~ = 67.09, *p =* 1.75E-7, $\eta_{p}^{2} = 0.79$\], and hemisphere \[*F*~(2,\ 36)~ = 4.15, *p* = 0.024, $\eta_{p}^{2} = 0.19$\]. The P2 amplitudes were larger for pictures in the certain as compared to the uncertain condition and for positive compared with negative pictures. The amplitudes were larger for the Middle as compared to the Right (*p* = 0.010), but they were similar in the Left as compared to the Middle and in the Left compared to the Right (*ps.* \> 0.05). There was a significant interaction between site and hemisphere \[*F*~(2,\ 30)~ = 6.01, *p* = 0.010, $\eta_{p}^{2} = 0.25$\]. No hemisphere effect was found at the Frontal and the Frontal-Central site (*ps.* \> 0.05). For the Central site, the effect of hemisphere was significant \[*F*~(2,\ 36)~ = 6.68, *p* = 0.003, $\eta_{p}^{2} = 0.27$\], with larger amplitudes for the Middle compared to the Left (*p* = 0.025), and for the Middle compared to the Right (*p* = 0.008). The interaction between uncertainty and emotion was significant \[*F*~(1,\ 18)~ = 10.32, *p* = 0.005, $\eta_{p}^{2} = 0.36$\]. Further analyses showed that for negative pictures, the amplitudes were larger in the certain compared to the uncertain condition \[*t*~(18)~ = 3.32, *p* = 0.004\]; for positive pictures, however, the effect of uncertainty did not reach statistical significance (*ps.* \> 0.05). ### N2 components The ANOVA for N2 amplitudes showed main effects of emotion \[*F*~(1,\ 18)~ = 16.99, *p* = 0.001, $\eta_{p}^{2} = 0.49$\], site \[*F*~(1,\ 20)~ = 18.77, *p* = 2.00E-4, $\eta_{p}^{2} = 0.51$\], and hemisphere \[*F*~(2,\ 36)~ = 9.51, *p* = 5.00E-4, $\eta_{p}^{2} = 0.35$\]. Negative pictures elicited larger amplitudes than did positive pictures. Pictures were larger in amplitudes at the Frontal as compared to the Frontal-Central (*p* = 0.048) and the Central (*p* = 0.001) site and at the Frontal-Central as compared to the Central site (*p* = 2.00E-4). The amplitudes were also larger at the Middle as compared to the Left (*p* = 5.00E-4). Furthermore, there was an interaction between uncertainty and emotion \[*F*~(1,\ 18)~ = 9.32, *p* = 0.007, $\eta_{p}^{2} = 0.34$\]. Follow-up analyses showed larger amplitudes for negative pictures preceded by uncertain as compared to certain cues \[*t*~(18)~ = 2.66, *p* = 0.016\], but this uncertainty effect did not reach statistical significance for positive pictures (*ps.* \> 0.05). ### Early LPP components The analysis showed a main effect of emotion \[*F*~(1,\ 18)~ = 59.95, *p* = 3.89E-7, $\eta_{p}^{2} = 0.77$\], with larger amplitudes for positive as compared to negative pictures. The main effect of site was significant \[*F*~(1,\ 22)~ = 71.57, *p* = 5.24E-9, $\eta_{p}^{2} = 0.80$\]. The amplitudes were larger for the parietal site as compared to the other sites (Frontal: *p* = 2.86E-7; Frontal-Central: *p* = 2.95E-7; Central: *p* = 4.25E-8; Central-Parietal: *p* = 1.04E-7), for the Central-Parietal site as compared to the Frontal (*p* = 1.10E-5), Frontal-Central (*p* = 1.00E-5), and Central sites (*p* = 1.00E-6), for the Central site as compared to the Frontal (*p* = 0.001) and Frontal-Central sites (*p* = 0.004) and for the Frontal-Central as compared to the Frontal site (*p* = 0.001). While the main effect of hemisphere was also significant \[*F*~(2,\ 36)~ = 3.63, *p* = 0.037, $\eta_{p}^{2} = 0.17$\], pairwise comparisons did not show any significant effects among all conditions (*ps.* \> 0.05). There was an interaction between uncertainty and emotion \[*F*~(1,\ 18)~ = 5.58, *p* = 0.030, $\eta_{p}^{2} = 0.24$\]. For positive pictures, the early LPP was larger in the uncertain as compared to the certain condition \[*t*~(18)~ = −2.37, *p* = 0.029, $\eta_{p}^{2} = 0.24$\]. For negative pictures, however, the effect of uncertainty did not reach statistical significance (*ps.* \> 0.05). The interaction between emotion and site was significant \[*F*~(1,\ 24)~ = 5.33, *p* = 0.023, $\eta_{p}^{2} = 0.23$\]. The early LPP was more shifted in the positive direction for positive as compared to negative pictures at all sites, though to different extents \[Frontal: *t*~(18)~ = 7.96, *p* = 2.61E-7; Frontal-Central: *t*~(18)~ = 8.36, *p* = 1.31E-7; Central: *t*~(18)~ = 7.70, *p* = 4.22E-7; Central-Parietal: *t*~(18)~ = 7.04, *p* = 1.00E-6; Parietal: *t*~(18)~ = 4.92, *p* = 1.10E-4\]. The interaction between emotion and hemisphere was significant \[*F*~(1,\ 24)~ = 6.91, *p* = 0.009, $\eta_{p}^{2} = 0.28$\]. The early LPP amplitudes were larger for positive as compared to negative pictures at all levels of hemisphere, though to different extents \[Left: *t*~(18)~ = 7.34, *p* = 8.23E-7; Middle: *t*~(18)~ = 7.87, *p* = 3.10E-7; Right: *t*~(18)~ = 6.60, *p* = 3.00E-6\]. We also found an interaction between site and hemisphere \[*F*~(4,\ 78)~ = 3.00, *p* = 0.021, $\eta_{p}^{2} = 0.14$\]. Separate analysis for each hemisphere showed that the site effect was significant at all levels of hemisphere, though to different extents \[Left: *F*~(1,\ 25)~ = 54.05, *p* = 1.45E-8, $\eta_{p}^{2} = 0.75$; Middle: *F*~(1,\ 23)~ = 59.20, *p* = 1.44E-8, $\eta_{p}^{2} = 0.77$; Right: *F*~(1,\ 27)~ = 77.36, *p* = 7.78E-11, $\eta_{p}^{2} = 0.81$\]. For the Left, the early LPP was larger for the Parietal site as compared to the other sites \[Frontal: *p* = 1.00E-6; Frontal-Central: *p* = 2.00E-6; Central: *p* = 8.71E-7; Central-Parietal: *p* = 8.00E-5\], for the Central-Parietal site as compared to the Frontal (*p* = 3.70E-5), Frontal-Central (*p* = 7.90E-5), and Central sites (*p* = 1.20E-5), for the Central site as compared to the Frontal (*p* = 4.49E-4) and Frontal-Central sites (*p* = 0.003) and for the Frontal-Central as compared to the Frontal site (*p* = 0.002). For the Middle, the early LPP was larger for the Parietal site as compared to the other sites (Frontal: *p* = 1.00E-6; Frontal-Central: *p* = 6.33E-7; Central: *p* = 1.15E-7; Central-Parietal: *p* = 2.66E-7), for the Central-Parietal site as compared to the Frontal (*p* = 6.40E-5), Frontal-Central (*p* = 3.60E-5), and Central sites (*p* = 4.00E-5), for the Central site as compared to the Frontal (*p* = 0.007) and Frontal-Central sites (*p* = 0.009) and for the Frontal-Central as compared to the Frontal site (*p* = 0.043). For the Right, the early LPP was larger for the Parietal site as compared to the Frontal (*p* = 1.15E-7), Frontal-Central (*p* = 1.15E-7), Central (*p* = 4.64E-8), and Central-Parietal sites (*p* = 4.07E-7), for the Central-Parietal site as compared to the Frontal (*p* = 3.00E-6), Frontal-Central (*p* = 2.00E-6), and Central sites (*p* = 9.64E-7) and for the Central (*p* = 0.008) and Frontal-Central (*p* = 0.012) sites as compared to the Frontal site. ### Late LPP components The AVONA showed main effects of anticipation \[*F*~(1,\ 18)~ = 5.43, *p* = 0.032, $\eta_{p}^{2} = 0.23$\], emotion \[*F*~(1,\ 18)~ = 11.54, *p* = 0.003, $\eta_{p}^{2} = 0.39$\], and site \[*F*~(1,\ 22)~ = 16.19, *p* = 2.74E-4, $\eta_{p}^{2} = 0.47$\]. The late LPP was more shifted in the positive direction in the certain as compared to the uncertain condition and for positive as compared to negative pictures. The late LPP was also larger for the Parietal site as compared to the Frontal (*p* = 0.017), Frontal-Central (*p* = 0.008), and Central sites (*p* = 0.050), for the Central-Parietal site as compared to the Frontal (*p* = 0.002), Frontal-Central (*p* \< 0.001), and Central sites (*p* = 0.002) and for the Central site as compared to the Frontal (*p* = 0.042) and Frontal-Central sites (*p* = 0.003). The interaction between emotion and site was significant \[*t*~(1,\ 24)~ = 3.97, *p* = 0.048, $\eta_{p}^{2} = 0.18$\]. Positive as compared to negative pictured evoked larger late LPP amplitudes at the Frontal \[*t*~(18)~ = 4.13, *p* = 0.001\], Frontal-Central \[*t*~(18)~ = 3.43, *p* = 0.003\], Central \[*t*~(18)~ = 3.30, *p* = 0.004\], Central-Parietal sites \[*t*~(18)~ = 3.00, *p* = 0.008\], thought to different extents. However, the emotional effect was not significant at the Parietal site (*ps.* \> 0.05). The interaction between site and hemisphere showed statistical significance \[*F*~(4,\ 72)~ = 4.54, *p* = 0.002, $\eta_{p}^{2} = 0.20$\]. Separate analysis for each hemisphere showed that the site effect was significant at all levels of hemisphere, though to different extents \[Left: *F*~(1,\ 23)~ = 12.65, *p* = 0.001, $\eta_{p}^{2} = 0.41$; Middle: *F*~(1,\ 25)~ = 17.85, *p* = 8.30E-5, $\eta_{p}^{2} = 0.50$; Right: *F*~(1,\ 27)~ = 10.90, *p* = 0.001, $\eta_{p}^{2} = 0.38$\]. For the Left, the late LPP was larger for the Parietal as compared to the Central (*p* = 0.035), Frontal-Central (*p* = 0.029) and Frontal (*p* = 0.024) sites and for the Central-Parietal as compared to the Central (*p* = 0.002), Frontal-Central (*p* = 0.007) and Frontal (*p* = 0.009) sites. For the Middle, the late LPP was larger for the Parietal as compared to the Central (*p* = 0.009), Frontal-Central (*p* = 0.001), and Frontal (*p* = 0.009) sites, for the Central-Parietal as compared to the Central (*p* = 0.004), Frontal-Central (*p* = 2.46E-4), and Frontal (*p* = 0.004) sites and for the Central as compared to the Frontal-Central site (*p* = 0.017). For the Right, the amplitudes were larger for the Central-Parietal as compared to the Central (*p* = 0.029), Frontal-Central (*p* = 0.001), and Frontal sites (*p* = 0.001) and for the Central site as compared to the Frontal-Central (*p* = 0.013) and Frontal (*p* = 0.036) sites. More importantly, the three-way interaction among uncertainty, emotion, and site was also significant \[*F*~(2,\ 40)~ = 9.48, *p* = 2.89E-4, $\eta_{p}^{2} = 0.35$\]. Separate analysis for each site showed that for the Frontal and the Frontal-Central site, the late LPP was larger in the certain as compared to the uncertain condition \[Frontal: *F*~(1,\ 18)~ = 6.05, *p* = 0.024, $\eta_{p}^{2} = 0.25$; Frontal-Central: *F*~(1,\ 18)~ = 7.10, *p* = 0.016, $\eta_{p}^{2} = 0.28$\] and for positive as compared to negative pictures \[Frontal: *F*~(1,\ 18)~ = 17.01, *p* = 0.001, $\eta_{p}^{2} = 0.49$; Frontal-Central: *F*~(1,\ 18)~ = 11.77, *p* = 0.003, $\eta_{p}^{2} = 0.40$\]. More importantly, the interaction between uncertainty and emotion was significant \[Frontal: *F*~(1,\ 18)~ = 7.63, *p* = 0.013, $\eta_{p}^{2} = 0.30$; Frontal-Central: *F*~(1,\ 18)~ = 4.75, *p* = 0.043, $\eta_{p}^{2} = 0.21$\]. For negative pictures, the late LPP was larger in the certain as compared to the uncertain condition \[Frontal: *t*~(18)~ = 3.06, *p* = 0.007; Frontal-Central: *t*~(18)~ = 2.94, *p* = 0.009\]; whereas the uncertainty effect did not reach statistical significance for positive pictures (*ps.* \> 0.05). For the Central site, the late LPP was generally larger in the certain as compared to the uncertain condition \[*F*~(1,\ 18)~ = 6.26, *p* = 0.022, $\eta_{p}^{2} = 0.26$\] and for positive as compared to negative pictures \[*F*~(1,\ 18)~= 10.88, *p* = 0.004, $\eta_{p}^{2} = 0.38$\]. In addition, there was a trend for the interaction between uncertainty and emotion \[*F*~(1,\ 18)~ = 3.92, *p* = 0.063, $\eta_{p}^{2} = 0.18$\]. The late LPP was larger for certain as compared to uncertain negative pictures \[*t*~(18)~ = 2.59, *p* = 0.019\], whereas this effect was not significant for positive pictures (*ps.* \> 0.05). For the Central-Parietal site, the analysis only showed a main effect of emotion \[*F*~(1,\ 18)~ = 9.02, *p* = 0.008, $\eta_{p}^{2} = 0.33$\], with larger amplitudes for positive as compared to negative pictures. For the Parietal site, however, no main effects or interaction reach statistical significance (*ps.* \>0.05). Discussion {#s4} ========== The present study further investigated whether uncertainty about the emotional content of an upcoming picture modulates ERPs to the picture. Results showed that uncertain as compared to certain negative pictures evoked smaller P2 and late LPP and larger N2 amplitudes. For positive pictures, early LPP was greater in amplitude in the uncertain compared to the certain condition. Taken together, the findings suggest that uncertainty modifies the ERPs to emotional pictures and that the uncertainty effects are altered by emotional valence of the pictures. P2 is a positive component that peaks over anterior sites around 200 ms following stimulus onset. The P2 is related to selected attention in early sensory processes (e.g., Yuan et al., [@B39]; van Hooff et al., [@B33]), with enhanced amplitudes for certain stimuli (e.g., Kanske et al., [@B13]). In a subsequent time range, N2 is a negative deflection over anterior scalp sites at \~ 200--300 ms. The N2 is supposed to be relevant in attention allocation in the final stages of sensory processing (Olofsson et al., [@B22]; Ernst et al., [@B6]), with larger N2 amplitudes for uncertain as compared to certain emotional pictures (Gole et al., [@B8]). Therefore, our findings suggest that uncertainty reduces the attention toward negative pictures during early sensory processes, but this uncertainty effect is reversed in late sensory processes. Surprisingly, the uncertainty effects related to attention are different in different sensory processes of negative pictures. Knowing about the negative content of the upcoming picture is supposed to activate attention mechanism before the occurrence of the picture (e.g., Böcker et al., [@B3]; Erk et al., [@B5]; Lin et al., [@B14], [@B15], [@B17]), which is facilitated in enhancing the attention toward the pictures shortly after they occurred (i.e., early sensory stages of picture processing; Lin et al., [@B14]). In the present study, as participants had not known about the negative content of the pictures in the preceding uncertain anticipation phase, the attentional resources allocated to uncertain negative pictures were very limited at early sensory stages of picture processing. Just at these stages, however, participants clearly knew about the emotional contents of the pictures (as the emotional effects started in the time range of P2). In order to process uncertain negative pictures in a better way, participants may enhance the allocation of attentional resources to uncertain negative pictures to a large extent in later stages of sensory processing (e.g., N2), resulting in observing the enhanced attention toward these pictures in these stages. For positive pictures, we did not find the uncertainty effect on P2 and N2, indicating that uncertainty does not modulate attention toward positive pictures during sensory processes. Due to negativity bias, participants may overestimate the frequency of negative pictures but underestimate that of non-negative pictures in uncertain circumstances (Sarinopoulos et al., [@B31]; Grupe and Nitschke, [@B9]). The occurrences of positive pictures may slightly violate the expectations and thus, enhance the attention at early stages of sensory processing (e.g., Qin et al., [@B26]). Therefore, the difference in certain as compared to uncertain positive pictures is reduced. In addition, as the attention toward uncertain positive pictures has been enhanced at early stages of sensory processing, it may be unnecessary to enhance the attention again to modulate the pictures at later stages, resulting in failing in observing the uncertainty effect in N2. However, our findings were inconsistent with a previous study by Yang et al. ([@B38]), which showed greater P2 and smaller N2 amplitudes for uncertain compared to certain fearful faces. One possible reason for the discrepancies may be related to stimuli serving as cues. In Yang et al.\'s ([@B38]) study, simple neutral symbols (e.g., "+") and emotional pictures served as uncertain and certain cues, respectively. In the fearful face condition, uncertain and certain cues were different not only in the meanings but also in some other factors, such as the emotional contents and composition of the cues. A recent study by Guan et al. ([@B11]) reported that the P2 to target faces was reduced by emotional compared to neutral primes. Therefore, it is possible that P2 amplitudes for faces were also reduced by the emotional contents of certain cues in Yang et al.\'s ([@B38]) study, which resulted in altering the uncertainty P2 effects. In addition, stimuli with complex composition are found to enhance N2 amplitudes (Wiens et al., [@B36]). As the composition was more complex for certain cues (emotional pictures) compared to uncertain cues (simple symbols) in Yang et al.\'s ([@B38]) study, the uncertainty N2 effects may be altered. In addition, Gole et al. ([@B8]) did not report the uncertainty P2 effect, regardless of the emotion. One possible reason is that the anticipatory intervals were long (6 s) in Gole et al.\'s ([@B8]) study. Such long intervals may allow participants to have an appropriate preparation even in the uncertain condition (Lin et al., [@B15]), resulting in reducing the uncertainty effect at early sensory stages. Early LPP (350--450 ms; often overlapping with P3), which is widely distributed over frontal, central, and parietal scalp sites, is supposed to reflect motivated attention (e.g., Nieuwenhuis et al., [@B20]; Olofsson et al., [@B22]). High motivated stimuli (i.e., attractive stimuli) were shown to enhance early LPP (Marzi and Viggiano, [@B19]; Righi et al., [@B28]). Therefore, our findings indicate that uncertain as compared to certain positive pictures capture more motivated attention. Anselme et al. ([@B1]) proposed that uncertainty enhances the attractiveness of the consequences and motivated attention as a result. Specifically, when people are certain that a positive stimulus is upcoming, they begin to adapt to it, primarily by reaching an understanding of what the stimulus means and why it occurs, and as a result, the stimulus loses some of its force. On the contrary, without knowing the exact emotion and adapting to it in advance, the stimulus produces strong attractive affect, which results in enhancing the motivated attention (Whitchurch et al., [@B35]). For negative pictures, we did not find the effect of uncertainty on the early LPP. According to previous studies (e.g., Nieuwenhuis et al., [@B20]; Olofsson et al., [@B22]), the findings indicate that motivated attention is similar to uncertain as compared to certain negative pictures. One possible reason is that negative pictures are dangerous and threatening and thus, are perceived as unattractive regardless of the uncertainty. The late LPP develops around 500 ms after the onset of a stimulus and sometimes lasting for a few seconds. While this component is repeatedly shown to be largest over parietal sites, the anticipation effect seemed to be reflected over anterior sites (Lin et al., [@B14]; Richards et al., [@B27]). The late LPP has been found to reflect attentional allocation during the evaluation processes, with larger amplitudes for emotional pictures that are emotional as compared to non-emotional evaluation (Hajcak et al., [@B12]). Accordingly, our findings may imply that certain compared to uncertain negative pictures are evaluated more negatively and thereby, capture more attentional resources. Previous studies suggested that stimuli are evaluated more negatively when the stimuli are preceded by negative stimuli than when they are preceded by neutral stimuli (de Jong et al., [@B4]; Tomarken et al., [@B32]). In the present study, while both certain and uncertain cues were used by neutral symbols; cues definitely indicating the upcoming negative pictures may be perceived as negative in some extent, as these cues have been found to activate the brain regions (e.g., amygdala) associated with negative emotion (e.g., Onoda et al., [@B25]). In this case, certain cues may enhance the negative evaluation to the negative pictures and the attention as a result. However, we did not find the effect of uncertainty on the late LPP to positive pictures. Therefore, our findings may indicate that uncertainty does not alter the evaluation and the attention toward positive pictures as a consequence. While according to the above-described studies (de Jong et al., [@B4]; Tomarken et al., [@B32]), preceding certain cues may enhance the positive evaluation toward positive pictures; the positive evaluation toward the pictures may also be enhanced by uncertainty (Wilson et al., [@B37]). In this case, differential positive evaluation toward uncertain as compared to certain positive pictures may be decreased, resulting in showing similar attention toward these two pictures. While the present study found that emotional valence of the pictures altered the uncertainty ERP effects, it is still unclear whether the uncertainty effect can be modulated by the arousal of the pictures or not. If this is really the case, then is the uncertainty effect modulated by both of the factors? Further studies should be devoted to investigate the issues in more detail. Conclusion {#s5} ========== The present study showed that negative pictures evoked smaller P2 and late LPP and larger N2 amplitudes in the uncertain as compared to the certain condition. For positive pictures, early LPP was larger in the uncertain as compared to the certain condition. Taken together, our findings indicate that uncertainty about the emotional contents of the pictures modulates the attention to the pictures, and that this modulation is altered by the valence of the pictures. Author contributions {#s6} ==================== HL was involved in study design, execution, data analysis, and manuscript drafting and revises. HJ was involved in study design and manuscript revises. JL, RY, TL, and YW were involved in data analysis and manuscript revises. We have read and approved the manuscript and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. 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. This work was supported by the Project of Key Research Institute of Humanities and Social Science Department of Province Construction in China (13JJD190007). ^1^The early CNV, which starts appropriately 1 s after the cue onset and is maximal at frontal-central scalp sites, is often related to the alerting and attention properties of the cue (e.g., Weerts and Lang, [@B34]; Rohrbaugh et al., [@B29]; Sakamoto et al., [@B30]). More importantly, this early CNV component has been found to reflect the processing of uncertainty during the anticipation phase, with reduced amplitudes when the task requirement was uncertain (Linssen et al., [@B18]). Therefore, it is important to investigate the relationships between this component and emotionally uncertain cues. For more detail, please refer to our previous study (Lin et al., [@B15]). ^2^As the alpha ERD/ERS was calculated by individual alpha frequency (IAF), the bandwidth ranges of lower-1, lower-2, and upper alpha were different across participants. [^1]: Edited by: Seung-Lark Lim, University of Missouri -- Kansas City, USA [^2]: Reviewed by: Tae-Ho Lee, University of Illinois at Urbana-Champaign, USA; Bahar Güntekin, Istanbul Kultur University, Turkey [^3]: This article was submitted to Emotion Science, a section of the journal Frontiers in Psychology