Hepatic DPP4 DNA Methylation Associates With Fatty Liver

Dipeptidyl peptidase 4 (DPP4) is an ubiquitously expressed cell-surface protease that cleaves a variety of peptides carrying a proline or alanine residue in the second N-terminal position (1–3). Membrane-located DPP4 can also be released into the circulation upon cleavage by matrix metalloproteases (4). DPP4 has recently been described as a novel adipokine that is upregulated in obesity (5,6). It has been suggested to be a potential link between obesity and the metabolic syndrome and was shown to induce insulin resistance in a paracrine and endocrine fashion (5–8). Furthermore, soluble DPP4 has been linked to cardiovascular disease because it induces inflammation in human vascular smooth muscle cells via pathways mediated by mitogen-activated protein kinase and nuclear factor-κB (9). Elevated expression of DPP4 was detected in the liver in patients with nonalcoholic fatty liver disease (NAFLD) (10). The trigger of hepatic DPP4 expression apparently is hyperglycemia, because elevated glucose levels increase DPP4 expression in cultured HepG2 cells (10). In line with this, prolonged periods of hyperglycemia increase plasma levels of systemic DPP4, indicated by a robust correlation of circulating DPP4 and HbA_1c levels in patients with type 2 diabetes (11,12). Furthermore, hepatic DPP4 protein content positively correlates with the grade of hepatosteatosis and DPP4 activity in the circulation (13). Associations of liver damage markers, such as γ-glutamyl transferase and alanine aminotransferase, with serum DPP4 levels (14), together with the increased hepatic expression of DPP4 in patients with hepatosteatosis, suggest that increased circulating DPP4 in NAFLD patients is of hepatic origin (15). However, current data do not clarify whether increased hepatic DPP4 expression is a cause or consequence of hepatic steatosis.

The incretins glucagon-like peptide 1 (GLP-1) and glucose-dependent insulinotropic polypeptide (GIP) are well known for their insulinotropic action and are inactivated upon DPP4-mediated cleavage (16). Accordingly, genetic inactivation (17) and inhibition of DPP4 (18) improves glucose tolerance and β-cell function caused by sustained incretin action in mice and humans. However, data are accumulating that indicate the beneficial effects of DPP4 inhibition in patients with diabetes are not solely a result of enhanced incretin action on pancreatic β-cells. For example, inhibition of DPP4 in diet-induced obese (DIO) C57BL/6 mice revealed effects on liver function, reflected by a reduction of hepatic fat content and improved insulin sensitivity (19,20). Although the latter observations might indicate the consequences of hepatic DPP4 inhibition, the data do not allow discrimination between direct and secondary effects.

Besides nutrition during adulthood, intrauterine exposure to a high-fat diet (HFD) and early childhood obesity promote the development of fatty liver and, therefore, are risk factors for metabolic disease (21). Epigenetic mechanisms such as DNA methylation and histone modifications integrate these environmental cues into gene programs (22). In the current study, we investigated the epigenetic regulation of DPP4 expression in the livers of very young DIO mice and of obese patients with NAFLD and nonalcoholic steatohepatitis (NASH) and analyzed the capacity of murine livers to release DPP4 compared with that of adipose tissue.

The animal welfare committees of the German Institute of Human Nutrition (DIfE) and the local authorities (Landesamt für Umwelt, Gesundheit und Verbraucherschutz, Brandenburg, Germany) approved all animal experiments in this study. Male C57BL/6J mice were fed an HFD (60 kcal% from fat; D12492, Research Diets) ad libitum directly after weaning at 3 weeks of age. Mice had free access to water and food and were housed in a temperature-controlled room (22 ± 1°C) on a 12:12 h light/dark cycle in accordance with the Principles of Laboratory Animal Care (23). Blood glucose was measured from tail blood. At the age of 6, 13, and 20 weeks, mice were sacrificed after a 6-h fast.

Hepatic triglyceride content was measured using the commercial TR-210 kit (Randox) according to the manufacturer’s protocol.

Plasma insulin concentrations were quantified from vena cava blood using a Mouse Ultrasensitive Insulin ELISA kit (Alpco). Soluble DPP4 was determined using a Mouse DPP4 ELISA Kit (R&D Systems).

DPP4 release from tissue explants was measured using a modified method described earlier (24,25). Briefly, 13-week-old mice were anesthetized and perfused with 20 mL perfusion buffer (Earle's balanced salt solution, 0.5 mmol/L EGTA). Biopsy specimens from the liver (middle lobe), gonadal white adipose tissue (gWAT), and subcutaneous WAT (scWAT) were rinsed in prewarmed saline and minced into small pieces (∼10 mg). Ten pieces were incubated in 850 µL DMEM containing 25 mmol/L glucose, 10% (volume-for-volume [v/v]) FCS, and 1% (v/v) penicillin/streptomycin at 37°C and 5% CO₂ for 2 h or 24 h. The medium was centrifuged (500g for 5 min), and the supernatant was analyzed for DPP4 concentration. Remaining tissue was washed in saline and homogenized for protein measurement. DPP4 concentration in the medium was normalized to protein content of the explant tissues.

Primary hepatocytes were isolated from 12-week-old C57BL/6J mice fed a normal chow diet by a collagenase perfusion method, as previously described (26). Isolated hepatocytes were cultured in DMEM (5.5 mmol/L glucose) supplemented with 10% (v/v) FCS, 1% (v/v) nonessential amino acids, and 1% (v/v) penicillin/streptomycin at 37°C and 5% CO₂. After 4 h, cells were infected with adenovirus (Ad) coding for murine Dpp4 (Ad-Dpp4) or Gfp (Ad-Gfp) (Vector BioLabs) and cultured for 24 h or 48 h until further analysis.

DPP4 activity was measured by the conversion of glycin-prolin-p-nitroanilide (Sigma-Aldrich) to p-nitroanilide. Then, 20 µL plasma, 90 µL cell supernatants, and 45 µL cell homogenates were filled to 90 µL with assay buffer (50 mmol/L glycine, 1 mmol/L EDTA, pH 8.7) and supplemented with 10 µL glycine-proline-p-nitroanilide (5 mmol/L). Production of p-nitroanilide was measured by the absorbance at 405 nm in a kinetic measurement at 37°C.

Primary hepatocytes were virus infected and cultivated for 24 h in DMEM. The cells were incubated with 2 mmol/L oleic acid in 0.2% BSA for another 24 h and subsequently homogenized in Hanks’ balanced buffer. Triglyceride content was measured as depicted above. De novo triglyceride synthesis was determined as specified before (27). Briefly, virus-infected cells were cultured for 48 h and incubated with [¹⁴C]-labeled glucose (0.025 µCi/well) overnight. Triglycerides from cells were isolated by chloroform-methanol extraction and measured in a scintillation counter (Beckman Coulter).

Western blotting of 20 µg of protein lysates was performed as described previously (28). Incubation with primary antibodies against DPP4 (1:2,000; ab129060, Abcam) and tubulin (Sigma-Aldrich) was done overnight.

Cells were fixed, permeabilized, and stained as described earlier (29). Antibody against murine DPP4 was from Abcam (1:200; ab129060). Secondary antibody was labeled with Alexa-488 and used in the presence of TO-PRO-3 iodide (Invitrogen) for nuclei staining. Microscopy was performed with the confocal Leica-DMi8 laser scan microscope (Leica Microsystems).

Total RNA from tissues of mice was extracted, and cDNA synthesis and TaqMan gene expression assays were performed as described previously (28).

Genomic DNA (gDNA) from the livers was isolated with the Invisorb Genomic DNA Kit II (STRATEC) according to the manufacturer’s instructions. The EpiTect Fast Bisulfite Kit (QIAGEN) was used as recommended for conversion of 500 ng gDNA.

Implementation of direct bisulfite sequencing PCR (dBSP) and pyrosequencing was performed as described before (30).

The murine hepatocyte cell line AML12 (American Type Culture Collection CRL-2254) was cultured as previously described (31). A fragment of the murine Dpp4 gene ranging from 500 to 1,500 bp was cloned upstream of the Firefly luciferase gene into the CpG-free pCpGL-Basic vector generating pCpGL-Dpp4. For in vitro methylation, pCpGL-Dpp4 was incubated for 1 h with M.SssI, HpaII, or HhaI CpG methyltransferases (New England BioLabs), and AML12 cells were cotransfected with 500 ng pCpGL-Dpp4 plasmid and 5 ng thymidine kinase promoter–Renilla luciferase reporter plasmid (Promega). Firefly activity was measured 48 h after transfection and normalized to Renilla luciferase activity.

The protocols for all clinical studies were approved by local ethics committees, and all participants gave written consent. In study 1, 96 obese women with Caucasian ethnicity (age: 46.8 ± 7.0 years, BMI: 47 ± 7.5 kg/m², no diabetes) from the ABOS cohort (Biological Atlas of Severe Obesity, clinical trial reg. no NCT01129297 at clinicaltrials.gov) underwent bariatric surgery.

A laparoscopic liver wedge biopsy was taken and snap-frozen for RNA and DNA analysis. A needle biopsy sample was simultaneously taken and routinely stained with hematoxylin and eosin saffron and Masson’s trichrome, Sirius red, and Perl’s staining. Steatosis was quantified as previously described (32). DPP4 mRNA expression of all individuals was analyzed using the HumanHT-12 v4.0 Whole-Genome DASL HT Assay (Illumina), and subjects were classified into quartiles by DPP4 expression. Subgroups with the lowest (n = 24) and highest (n = 24) DPP4 mRNA levels were used for DNA methylation analysis of indicated CpG positions by pyrosequencing. Briefly, gDNA was bisulfite converted using the EZ DNA Methylation Kit (Zymo Research). Fragments containing CpGs of interest were amplified using PyroMark PCR Kit (QIAGEN) with primers designed by PyroMark Assay Design software. Before measurement of patient samples, assays were checked for their performance with universal methylated human standards (EpigenDx). DNA methylation analysis was conducted on a PyroMark Q24 pyrosequencer (QIAGEN).

In study 2, obese individuals were selected from an ongoing study recruiting all subjects undergoing bariatric surgery (Kuopio Obesity Bariatric Study [KOBS]), as previously described (33,34). Liver samples obtained from 95 obese individuals (age: 49.5 ± 7.7 years, BMI: 43 ± 5.7 kg/m², type 2 diabetes: n = 35) as a wedge biopsy during a Roux-en-Y gastric bypass operation were investigated. The liver phenotype was normal in 34 individuals, 35 had simple steatosis, and 26 had NASH.

DPP4 mRNA expression was analyzed from 42 individuals (14 with normal liver, 13 with simple steatosis, and 15 with NASH) using the Human HT-12 Expression BeadChip (Illumina), as described before (35). DNA methylation was analyzed using the Infinium HumanMethylation450 BeadChip (Illumina), as described previously (35). To identify differences in DPP4 mRNA expression in the liver associated with NASH, a linear regression model was used including NASH diagnosis, steatosis, diabetes, sex, BMI, and age as covariates and mRNA expression as the dependent variable. Spearman correlations were used to relate DNA methylation and mRNA expression or BMI.

Statistical analysis was performed by Student t test, one- or two-way ANOVA, and linear regression analysis using the GraphPad Prism 6 software. Significance levels were set for P values of <0.05, <0.01, and <0.001.

To investigate the role of hepatic DPP4 during DIO, we measured hepatic Dpp4 expression and metabolic parameters in very young (6-week-old) and in adult (20-week-old) C57BL/6J mice receiving an HFD. Although being inbred, C57BL/6J mice display large heterogeneity in their response to the HFD (36). At both time points, expression of hepatic Dpp4 correlated with body weight (P < 0.001) (Fig. 1A). When the body weight tertiles were compared at 6 weeks, high body-weight gainers (HG, >24 g) displayed increased Dpp4 expression compared with low body-weight gainers (LG, <22 g) and intermediate gainers (IG, 22–24 g; P < 0.05) (Fig. 1B). This trend continued until week 20, with LG (<35 g) and IG (35–41 g) mice not being different but HG mice (>41 g) displaying increased Dpp4 expression levels (P < 0.01) (Fig. 1C). At both time points, hepatic Dpp4 expression also correlated to blood glucose levels (P < 0.01 and P < 0.001, respectively) (Fig. 1D). HG mice also displayed elevated plasma insulin levels and a positive correlation of Dpp4 expression with HOMA (Fig. 1E and F). In contrast, liver triglyceride content was independent of Dpp4 expression level at 6 weeks of age (Fig. 1G). However, at 20 weeks, liver triglyceride concentrations were increased in mice with elevated Dpp4 expression levels (P < 0.05) (Fig. 1G). Together, these data indicate that insulin resistance and later ectopic lipid accumulation in the liver are preceded by increased levels of hepatic Dpp4 expression.

Previous studies suggested DPP4 as an adipokine that is elevated in obesity (5,6). To investigate whether the liver also contributes to circulating DPP4 and to compare this to adipose tissue, explants from the liver, gWAT, and scWAT of 13-week-old mice were analyzed for their capacity to release DPP4. In parallel, RNA was isolated to examine Dpp4 expression. Dpp4 mRNA content and ex vivo protein release were both highest in the liver, followed by scWAT and gWAT (Fig. 2A and B). Total DPP4 release by the organs was calculated by tissue weight. This analysis indicates that the liver largely contributes to systemic DPP4 levels (89.6 ± 3.8 ng/h) compared with WAT, which releases less DPP4 (10.9 ± 0.6 ng/h) (Fig. 2C). Surprisingly, Dpp4 mRNA expression and DPP4 release correlated in the liver (Fig. 2D) but not in adipose tissues (Supplementary Fig. 1A and B). Moreover, DPP4 release was higher in liver explants of HG mice at 3 h (P = 0.178) and 24 h (P < 0.05) (Fig. 2E) but not different in adipose tissue depots from LG and HG animals (Supplementary Fig. 1C and D). In summary, ex vivo data show a clear correlation between hepatic Dpp4 expression and DPP4 release, both being elevated in obesity and contributing to circulating levels.

Ad-mediated overexpression of DPP4 (Ad-Dpp4) in primary hepatocytes led to a stable increase of Dpp4 mRNA 24–48 h after transduction compared with noninfected and Ad-Gfp–treated cells (Fig. 3A). DPP4 protein (Fig. 3B and C) and DPP4 activity (Fig. 3D) were markedly increased after 48 h. Furthermore, DPP4 was largely targeted to the plasma membrane (Fig. 3E). Because the extracellular part of plasmalemmal DPP4 can be released into the circulation by shedding (4), supernatants of infected cells were analyzed for DPP4 content and activity, both showing to be increased (Fig. 3F and G). Accordingly, plasma DPP4 activity was increased from 6- to 20-week-old animals (4.78 ± 0.07 vs. 6.06 ± 0.19, P < 0.001) and correlated with hepatic Dpp4 expression (Fig. 3H).

The data from young LG and HG mice indicated that Dpp4 expression did not depend on elevated hepatic fat mass. To evaluate the direct effect of DPP4 on triglyceride storage, primary hepatocytes overexpressing DPP4 were treated with oleic acid for 24 h. This increased cellular triglyceride content in Ad-Dpp4–infected cells (+40.5%, P < 0.01) to the same extent as in Ad-Gfp–treated control cells (+42.4%, P < 0.01) (Fig. 4A). Moreover, de novo lipogenesis, measured by [¹⁴C] incorporation into the triglyceride pool, was similar in Ad-Gfp– and Ad-Dpp4–treated hepatocytes (Fig. 4B), indicating that DPP4 does not affect hepatic lipid accumulation in a direct manner.

The HG compared with LG mice displayed elevated hepatic Dpp4 expression levels already at a young age (Fig. 1B). To test whether differential DNA methylation of the Dpp4 gene might be involved in the early variation, we screened a genomic region of 4 kb (2 kb of the promoter, the 5′-untranslated region/exon 1, exon 2, and intronic regions) (Fig. 5A) for CpG sites as targets for DNA methylation (37) and identified 86 CpG sites. To increase the visibility of methylation differences, we compared methylation in HG mice (highest Dpp4 expression) and LG mice (lowest Dpp4 expression) at the age of 6 weeks (Fig. 1B). Although large parts of the depicted genomic region were not differentially methylated, a region containing intron 1, exon 2, and parts of intron 2 showed lower levels of DNA-methylation in HG compared with LG mice at four positions (Fig. 5B and Supplementary Fig. 2). This finding was validated by pyrosequencing at positions CpG₈₇₇, CpG₁₂₀₄, CpG₁₂₅₃, and CpG₁₂₅₅ (Fig. 5C–F).

Next, we correlated total CpG methylation with Dpp4 expression and found strong negative correlations of DNA methylation and expression for CpG₈₇₇ (P = 0.0089), CpG₁₂₅₃ (P = 0.0064), and CpG₁₂₅₅ (P = 0.0015) and a weak correlation for CpG₁₂₀₄ (P = 0.0926) (Fig. 5G–J). To identify factors possibly involved in differential CpG methylation, we performed correlation analyses. Body weight correlated with methylation levels at all four positions (Table 1). Fasted blood glucose levels correlated only with CpG₈₇₇, and plasma insulin correlated with CpG₁₂₅₃ and CpG₁₂₅₅ but not with CpG₈₇₇ and CpG₁₂₀₄.

Because glucose appears to be a regulator of Dpp4 expression (10), we investigated whether DNA methylation affects glucose-dependent Dpp4 expression. We generated a luciferase reporter construct (pCpGL-Dpp4) carrying 500–1,500 bp downstream of the transcription start site, including 27 CpG sites of Dpp4 in a CpG-free backbone vector (pCpGL) (Fig. 6A). We in vitro methylated pCpGL-Dpp4 using HpaII (CCGG motif), HhaI (GCGC motif), and M.SssI (CG motif) methyltransferases and quantified luciferase activity in AML12 hepatocytes. Partial methylation of pCpGL-Dpp4 with HpaII and HhaI reduced luciferase activity by 47.1 ± 13.3% and 58.7 ± 11.6%, respectively. Full methylation by M.SssI almost completely abolished luciferase activity (Fig. 6B). The glucose sensitivity of Dpp4 expression apparently was preserved in the sequence cloned into pCpGL-Dpp4 because luciferase activity increased with increasing concentrations of glucose (Fig. 6C). To test whether this glucose-dependent Dpp4 expression was affected by changes in DNA methylation, we in vitro methylated pCpGL-Dpp4 and measured luciferase activity in the presence (17.5 mmol/L) or absence of glucose. Although basal rates were decreased, glucose sensitivity of luciferase activity was preserved after HpaII treatment (P < 0.05) (Fig. 6D). However, upon HhaI or M.SssI treatment, the glucose-stimulated increase in luciferase activity was blunted (Fig. 6D). Together, these data indicate that glucose-stimulated Dpp4 expression is modulated by changes in DNA methylation.

We next studied DPP4 expression in human liver biopsy specimens from the ABOS cohort. Individuals with high hepatic DPP4 mRNA expression exhibited increased levels of hepatosteatosis compared with patients with lower DPP4 expression (P = 0.062) (Fig. 7A). To test whether alterations in DNA methylation correlate with different DPP4 expression in human livers, we analyzed those CpG sites that are conserved between mice and humans (CpG₈₇₇, CpG₁₂₀₄, and CpG₁₂₅₃). We found lower DNA methylation at CpG₁₃₆₂ (corresponding to murine CpG₁₂₀₄) in patients with high DPP4 expression (P = 0.060) (Fig. 7B). However, methylation at human CpG₁₀₀₁ (murine CpG₈₇₇) and CpG₁₄₁₀ (murine CpG₁₂₅₃) was not different between individuals with low and high hepatic DPP4 mRNA expression (3.94 ± 0.19% vs. 3.76 ± 0.17% [P = 0.485] and 40.29 ± 1.43% vs. 38.43 ± 1.01% [P = 0.299], respectively).

In another human cohort (KOBS), subjects with NASH tended to have higher DPP4 mRNA expression in the liver compared with subjects without NASH (6.72 ± 0.37 vs. 6.57 ± 0.30, P = 0.090). DNA methylation of a CpG site annotated to DPP4 correlated negatively with DPP4 expression in human liver (CpG₄₉ [cg12335708]; R = −0.340, P = 0.030). In addition, liver DNA methylation of a CpG site annotated to DPP4 correlated negatively with BMI (CpG₉₉₆ [cg20539283]; R = −0.330, P = 0.001). Taken together, human data from two independent cohorts confirm the relationship of DPP4 methylation and expression and show a positive correlation of hepatic DPP4 and hepatosteatosis/NASH.

This study demonstrates that in mice 1) the hepatic Dpp4 expression is weight-dependently regulated already at very young age, 2) the liver releases DPP4 and contributes markedly to the soluble DPP4 pool, and 3) increased expression of hepatic Dpp4 is linked to decreased methylation of four distinct CpG sites flanking exon 2, and our human data show that 4) one of three homologous CpG sites (CpG₁₂₀₄) is less methylated in subjects exhibiting increased hepatic DPP4 mRNA expression and higher level of steatosis.

Miyazaki et al. (10) showed earlier that DPP4 expression in human HepG2 cells is increased by glucose. The in vitro reporter assays of our study indicated that the glucose-stimulated Dpp4 expression is modulated by DNA methylation. We therefore believe that hypomethylation of the Dpp4 gene enhances its glucose-mediated expression and release by the liver and thereby increases the amount of circulating DPP4. This might lead to metabolic deteriorations in a para- and autocrine fashion as shown before (5,9). However, our data demonstrate that DPP4 does not directly affect hepatic lipid accumulation: 1) the HG mice exhibited already higher Dpp4 mRNA levels than the LG mice at week 6 when liver triglyceride concentrations were comparable between both groups, and 2) Dpp4 overexpression in hepatocytes did not directly increase ectopic lipid accumulation.

Previous studies suggested DPP4 to be an adipokine linking obesity with the metabolic syndrome and insulin resistance (5–7). DPP4 is upregulated in visceral versus subcutaneous adipose tissue of obese humans, and circulating DPP4 levels are elevated in obese insulin-resistant patients (5,6). To our surprise, the present mouse data show higher Dpp4 expression in scWAT than in gWAT, but neither the expression nor the release of DPP4 detected in adipose tissue explants differed between the LG and HG mice. In addition, total DPP4 release was much higher from the liver compared with that of WAT depots. Thus, our data clearly demonstrate a strong contribution of the liver to systemic DPP4 levels. These findings, together with the in vitro results of primary hepatocytes, give direct functional evidence that hepatic DPP4 is released into the circulation and thereby might participate in the pathophysiology of metabolic diseases.

In contrast to human studies, genetic differences or diet effects can be ruled out as sources for the interindividual variation in hepatic Dpp4 expression observed in our mouse study. It is more likely that differences in the epigenetic imprinting during the perinatal period forms the basis for variation in body weight development and metabolic health early in life (22). Accordingly, it is well accepted that early childhood obesity and intrauterine exposure to HFDs promote hepatic fat accumulation (21). Our data suggest that these weight gain dynamics early in life are closely linked to changes in hepatic Dpp4 expression, which are in turn associated with variation in Dpp4 methylation around exon 2 and linked to body weight and blood glucose levels. Interestingly, our human data support a similar relationship of DNA methylation at CpG₁₃₆₂ (murine CpG₁₂₀₄) and hepatic DPP4 expression as we observed in obesity-prone and -resistant mice. The human data also show a correlation of hepatic DPP4 expression and the grade of hepatosteatosis and inflammation as well as a negative correlation with DNA methylation.

According to our results, methylation in exon 2 of human DPP4 has been previously shown to correlate negatively with its expression in visceral adipose tissue of obese women (38). Furthermore, DPP4 methylation in adipose tissue correlated positively with HDL cholesterol levels in the blood, indicating a role of adipose tissue DPP4 in lipidemia (38). However, a follow-up study in obese men and women did not reveal any association of DPP4 methylation in adipose tissue with the status of the metabolic syndrome but confirmed the association with HDL cholesterol levels (39).

In summary, our data reveal an early weight-dependent regulation of hepatic Dpp4 expression and a strong contribution of the liver to circulating DPP4 levels. Furthermore, we show that DNA-methylation around exon 2 is likely to be involved in the regulation of hepatic Dpp4 expression by glucose and that perturbations in this mechanism might play a role in the development of fatty liver disease.

Acknowledgments. The authors thank Christine Gumz, Andrea Teichmann, and Kathrin Warnke of the German Institute of Human Nutrition Potsdam-Rehbrüecke for their skillful technical assistance and Dr. Michael Rehli of the University Hospital Regensburg for providing the CpG-free pCpGL vectors used in the luciferase assays.

Funding. This work was supported by the German Ministry of Education and Research (DZD grant 01GI0922) and the German Research Foundation (GK1208). The human liver data (KOBS) was supported by grants from the Swedish Research Council and Avtal om Läkarutbildning och Forskning.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.B., S.S., and A.K. performed data acquisition and analysis and drafted the article. C.B. and A.S. reviewed and edited the manuscript. M.J., L.S., D.H., M.C., S.L., R.C., V.R., F.P., E.N., J.P., C.L., and P.F. performed data acquisition and analysis. A.S. performed study conception and design and critically revised the manuscript. R.W.S. was responsible for study conception and design, performed data analysis, and drafted the article. All authors approved the final version of the manuscript. A.S. and R.W.S. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.