Genetic Determinants of Adiponectin Regulation Revealed by Pregnancy

Objective: This study investigated genetic determinants of adiponectin during pregnancy to reveal novel biology of adipocyte regulation.
Methods: A genome-wide association study was conducted in 1,322 pregnant women from the Hyper- glycemia and Adverse Pregnancy Outcome Study with adiponectin measured at 28 weeks of gestation. Variants reaching P < 531025 for de novo genotyping in two replication cohorts (Genetics of Glycemic
regulation in Gestation and Growth N 5 522; ECOGENE-21 N 5 174) were selected. Results: In the combined meta-analysis, the maternal T allele of rs900400 located on chr3q25 (near LEKR1/ CCNL1) was associated with lower maternal adiponectin (b 6 standard error [SE] 5 20.18 6 0.03 standard deviation [SD] of adiponectin per risk allele; P 5 1.531028; N 5 2,004; multivariable adjusted models). In contrast, rs900400 showed only nominal association with adiponectin in a large sample of nonpregnant women (b 6 SE 5 20.012 6 0.006; P 5 0.05; N 5 16,678 women from the ADIPOgen consortium). The off- spring rs900400 T risk allele was associated with greater neonatal skinfold thickness (b 6SE 5 0.19 6 0.04 SD per risk allele; P 5 4.131028; N 5 1,489) and higher cord blood leptin (b 6 SE 5 0.28 6 0.05 log-leptin per risk allele; P 5 8.231029; N 5 502), but not with cord blood adiponectin (P 5 0.23; N 5 495). The T allele of rs900400 was associated with higher expression of TIPARP in adipocytes. Conclusions: These investigations of adipokines during pregnancy and early life suggest that rs900400 has a role in adipocyte function.

Introduction
Adipose tissue is a key regulator of insulin sensitivity, partially through the endocrine functions of adipokines. Healthy, “metabolically flexible” adipose tissue is characterized by smalladipocytes that secrete high levels of adiponectin, while large hyper- trophic adipocytes in macrophage-infiltrated adipose tissue produce less adiponectin and high levels of leptin (1). In human studies, lowadiponectin levels have been associated with lower insulin sensitivity and increased risk of type 2 diabetes (T2D) and gestational diabetes mellitus (2-4). The most recent genome-wide association study (GWAS) of adiponectin levels identified 10 loci (5), highlighting ADIPOQ as the strongest genetic determinant of adiponectin levels, confirming candidate gene investigations (6). Despite GWAS (5,7) and candidate gene (6) investigations of adiponectin, we know very little about the regulation of adipocytes’ endocrine function. Investi- gating adipokine genetics in the context of physiologic challenge could increase our understanding of adipose tissue “flexibility.”Pregnancy is characterized by major physiologic changes, including a marked decrease in insulin sensitivity. White adipose tissue expresses lower amounts of adiponectin in late gestation, and levels decrease over the course of pregnancy (8). Pregnancy may unmask metabolic risk, e.g., women with gestational diabetes mellitus are more likely to develop T2D (9). We have previously found genetic determinants of glycemic traits in pregnant women that were not identified in much larger studies of nonpregnant adults (10). Given that, we hypothesized that pregnancy-induced metabolic changes would enhance adipose tissue dysfunction in genetically predisposed women and allow detection of novel genetic determinants of adipo- nectin levels. Using an agnostic genome-wide discovery approach followed by replication, we investigated genetic determinants of adi- ponectin in three prospective cohorts of mother-newborn dyads. We pursued our main finding for associations with adiposity-related traits and other adipokines in mothers and newborns.Hyperglycemia and Adverse Pregnancy Outcome Study— discovery GWAS.

Detailed methods for recruitment and pheno- typing of participants in the Hypoglycemia and Adverse Preg- nancy Outcome (HAPO) Study were published previously (11). In brief, pregnant women 18 years old were eligible if they were at less than 32 weeks of gestation, had a singleton pregnancy, and had no history of diabetes. All women had a 75-g oral glucose tolerance test (OGTT) between 24 and 32 weeks. All pregnant women gave written consent, and an external Data Monitoring Committee provided oversight across sites. The protocol was approved by the institutional review board at each field center. The original HAPO Study enrolled women from diverse ancestry groups; main analyses for the present study include 1,322 women of European ancestry who had consented to genetic studies and were included in a biomarkers substudy in which adiponectin lev- els were measured. Newborns’ weight, length, and skinfolds were measured within 72 hours of birth using standardized procedures (12). Skinfolds were measured in duplicate at three sites (flank, subscapular, and triceps) and summed; the average of two meas- urements at each site was used for analyses. Cord blood samples were collected at delivery, including circulating cells to obtain DNA. Only offspring whose mothers consented to genetic analy- ses are included in this analysis.Adiponectin was measured using LuminexVR technology (Luminex Corp., Austin, Texas) in stored (2808C) maternal fasting samples collected at the time of the OGTT; the interassay coefficient of vari- ation (standard deviation/mean) for low and high controls includedwith each assay was 11.3% and 15.1%, respectively (13). DNA was prepared using the automated Autopure LS (Gentra Systems, Minne- apolis, Minnesota).Genetics of Glycemic regulation in Gestation and Growth cohort (replication)Women planning to deliver at the Centre Hospitalier Universitaire de Sherbrooke (CHUS) were recruited between 6 and 15 weeks of pregnancy.

Exclusion criteria were age <18 or >40 years old, multi- ple pregnancies, pregestational diabetes (type 1 or 2), diabetes dis-covered at first trimester, or medical conditions or medications that would affect glucose regulation. The CHUS ethical review board approved the project, and all women provided written consent before inclusion in the study. This analysis includes up to 522 women from the Genetics of Glycemic regulation in Gestation and Growth (Gen3G) cohort with adiponectin levels and genetic consent.Details of Gen3G methods during pregnancy were published previ- ously (4,14). Between 24 and 30 weeks of gestation, maternal anthropometry was measured using standardized procedures, and each participant had a fasting 75-g OGTT. At delivery, cord blood samples were collected in addition to late pregnancy and peripartum events from electronic medical records. Skinfolds were measured in duplicate at four sites (triceps, biceps, subscapular, and suprailiac) within 72 hours of delivery in a subsample using standardized procedures.Following collection, blood samples were maintained at 48C and then centrifuged and stored at 2808C. Plasma glucose levels were measured by glucose hexokinase (Roche Diagnostics, Indianapolis, Indiana). Adiponectin was measured using radioimmunoassay (Milli-pore Corp., Billerica, Massachusetts). Leptin in maternal and cord blood was measured using Luminex technology (Human Milliplex, Millipore Corp., Billerica, Massachusetts). Intra- and interassay coefficients of variation were all <10%. DNA was extracted frommaternal blood and from cord blood samples using the Gentra Pure-gene Cell Kit (Qiagene, Valencia, California).ECOGENE-21 birth cohort (replication). Women with a sin- gleton pregnancy in their first trimester were recruited from a foun- der population of French-Canadian origin (Saguenay area, Canada) and followed until delivery. Women over 40 years old and those with pregestational diabetes or other disorders known to affect glu- cose metabolism were excluded.

The Chicoutimi Hospital Ethics Committee approved the project. All women provided written informed consent before inclusion in the study; 174 women who provided genetic consent are included in this analysis. Maternal anthropometric measurements were performed using standardized procedures. Glucose tolerance was assessed using a 75-g OGTT per- formed at 24 to 30 weeks’ gestation. Blood glucose levels were measured on fresh serum samples using a Beckman analyzer (model CX7; Fullerton, California). Serum adiponectin levels were measured by ELISA (B-Bridge International, Santa Clara, California). DNA was extracted from maternal blood samples using the Gentra Puregene Cell Kit (Qiagene, Valencia, California). Newborns’ char- acteristics were collected at birth from clinical records.Genome-wide genotyping of HAPO participants. DNA sam- ples were genotyped using genome-wide arrays (IlluminaHuman610-Quad v1 B at the Broad Institute), as previously reported (10). Genotype data that passed initial quality control (QC) were released to the Gene Environment Association Studies Coordinating Center, National Center for Biotechnology Information’s Database of Genotypes and Phenotypes (dbGaP), and HAPO Study teams, who collectively performed QC using procedures previously described by the Gene Environment Association Studies consortium (15). Poorly performing samples and single-nucleotide polymor- phisms (SNPs) were removed based on mis-specified sex, chromo- somal anomalies, unintended sample duplicates, sample relatedness, low call rate, high number of Mendelian errors, departures from Hardy-Weinberg equilibrium, duplicate discordance, sex differences in heterozygosity, and low minor allele frequencies, as detailed previ- ously (10,16). Complete QC reports are available through dbGaP (http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id5 phs000096.v4.p1).cohorts. Independent loci demonstrating an association with mater- nal adiponectin levels at P < 131025 in the HAPO GWAS (total nine loci) were identified for replication in Gen3G maternal samples. OneSNP (rs4943768) failed genotyping QC criteria and was excluded from meta-analyses.

Among selected SNPs, the best candidate SNP with the lowest P value after combining HAPO and Gen3G (rs900400) was fur- ther genotyped using the ECOGENE-21 maternal samples. In Gen3G and ECOGENE-21 cohorts, selected SNPs were genotyped on a quan- titative reverse transcription-polymerase chain reaction (qRT-PCR) (model 7500 Fast, Applied BiosystemsVR ) using Applied Biosystems TaqManVR probes and primers following the manufacturers’ recommen- dations (Life Technologies Inc., Burlington, Ontario, Canada).Genetic associations with maternal adiponectin. We used a z score transformation for adiponectin in all cohorts. First, we per- formed a discovery GWAS of maternal adiponectin using SNPTEST v2 in 1,322 HAPO women. We then performed a meta-analysis of HAPO and replication cohort(s): additive genetic linear regression models between maternal genotypes and maternal adiponectin levels (after z score transformation) were adjusted for the following: Model 1: ancestry (for HAPO participants, using first two principal compo- nents), parity, maternal age, gestational age at OGTT, and neonatal sex; Model 2: Model 1 covariates and maternal mean arterial pressure,height, and BMI at OGTT. Inverse variance-weighted meta-analysis results with P < 5.031028 were considered statistically significant.Maternal rs900400 and adiposity/glycemia traits. To extend our understanding of our top finding on maternal adiposity and glu- cose regulation in pregnancy, we examined maternal rs900400 for association with maternal BMI, fasting, 1-hour, and 2-hour glucose, and insulin sensitivity (z score) (17) using meta-analysis across all three cohorts. Maternal leptin (log-transformed) was examined in Gen3G, and adiponectin measured at both the first and second trimes- ters was evaluated in Gen3G and ECOGENE-21.Offspring rs900400 and neonatal adiposity-related traits. Given prior reports of rs900400’s association with birth weight and neonatal anthropometric measures (18,19), HAPO and Gen3G data were used to explore associations between offspring rs900400 geno- type and neonatal adiposity-related traits not previously reported: cordblood C-peptide (z score), adiponectin, and leptin (only in Gen3G).

We also confirmed associations with birth weight, birth length, pon- deral index, and skinfolds (z score).Additive genetic models were adjusted for the following: Model 1: maternal age, parity, gestational age, and newborn sex; Model 2: Model 1 covariates and maternalBMI. P < 0.05 was considered nominally significant; P < 0.007 wasconsidered statistically significant (Bonferroni corrected 0.05 dividedby seven neonatal traits). We conducted secondary analyses, addition- ally adjusting for maternal genotype at rs900400.Public data searches in nonpregnant individuals’ databases and functional data. To compare our findings in pregnant women with nonpregnant adults, genetic associations for rs900400 were explored using publicly available GWAS databases of anthropo- metric measures (Genetic Investigation of ANthropometric Traits [GIANT] consortium) (20,21), glycemic-related traits (Meta-Analyses of Glucose and Insulin-related traits Consortium, MAGIC), and adipo- nectin levels (ADIPOgen). The potential function of rs900400 in adi-pocytes was examined by extracting expression quantitative trait loci (eQTL) with false discovery rate q < 0.01 from the publicly available Multiple Tissue Human Expression Resource (MuTHER) (22).

Results
Characteristics of HAPO, Gen3G, and ECOGENE-21 participants are presented in Table 1. All three cohorts were population-based with relatively similar characteristics: about half of the women were primiparous, and mean midpregnancy BMI was in the overweight range. About half of newborns were male, and, by design, only term deliveries are included. We found weak correlations between mater- nal adiponectin levels and birth weight (HAPO r 5 20.11; P < 0.001; Gen3G r 5 20.09; P 5 0.048).Genetic associations with maternal adiponectin The discovery GWAS in 1,322 HAPO women revealed nine inde- pendent loci associated with maternal adiponectin at the second tri- mester at P < 1.031025 (Supporting Information Figure S1), and we meta-analyzed eight loci using HAPO and Gen3G data (Table 2). Meta-analysis of all three cohorts (N 5 2,004 women) revealed that the maternal T allele at rs900400 located at chr3q25 (Figure 1a) was associated with lower second-trimester adiponectin (b 6 standard error [SE] 5 20.177 6 0.031 standard deviation [SD] of adiponectin per risk allele; P 5 1.4531028 in Model 2; Table 3). The direction and effect sizes of the maternal rs900400 T allele on adiponectin levels (SD per risk allele) were consistent in all three cohorts: b 6 SE 5 20.189 6 0.038 in HAPO, 20.113 6 0.065 in Gen3G, and20.252 6 0.104 in ECOGENE-21 (all Model 2). Secondary analyses including the fetal genotype in models slightly reduced the effect size, but the association remained strong in the same direction of effect(Model 2: b 6 SE 5 20.141 6 0.038; P 5 2.531024; N 5 1,261 mother-child pairs in HAPO). Maternal rs900400 seemed to have a smaller effect size for its association with adiponectin at the first trimester (Model 2: b 6 SE 5 20.118 6 0.064 SD of adipo- nectin per risk allele; P 5 0.07) versus second trimester (Model 2: b 6 SE 5 20.166 6 0.065 SD of adiponectin per risk allele; P 5 0.01) in 498 women with levels measured at both time points (Gen3G and ECOGENE-21), but the difference in effect sizes was not statistically significant (P 5 0.58). We did not find a significantaAll maternal characteristics were measured at the time of OGTT, except for maternal age in Gen3G and ECOGENE-21 cohorts, which was collected at first trimester.bAbsolute values differ because of bioassay-specific characteristics; all values z score transformed before meta-analyses.cSum of three folds in HAPO; sum of four folds in Gen3G; z score transformed for analyses.association between maternal rs900400 and the change in adiponectin between the first and second trimester (Model 2: b 6 SE 5 20.218 6 0.259 ug/mL adiponectin per risk allele;P 5 0.40).At the ADIPOQ locus, the maternal G allele at rs17300539 (pro- moter region) was associated with lower adiponectin just below genome-wide significance (b 6 SE 5 20.260 6 0.058 SD of adipo-nectin per risk allele; P 5 9.1431026 in Model 2; N 5 1,842, Figurereported (rs17451107) is in strong linkage disequilibrium (LD) with rs900400 (r2 5 0.932 in CEU). In MAGIC (up to 46,186 individu- als), we found no association with fasting glucose (P 5 0.37) or fast- ing insulin (P 5 0.21).We searched for eQTLs in the chr3q25 region in an adipose tissue expression dataset of the publicly available MuTHER data set (22). In this region, 789 SNPs were significantly associated (false discov- ery rate ≤ 0.01) with the expression of 11 protein-coding genes. The T allele of rs900400 was associated with higher expression of TIPARP (P 5 6.75 310258) but with no other transcript.For potential functionality, we searched relevant Encyclopedia of DNA Elements regulation tracks and findings from 3D chromatin contact partitions (23) for the chr3q25 region (Supporting Informa- tion Figure S2). According to 3D chromatin contact partitions defined in the GM12878 lymphoblastoid cell line using the DNA proximity ligation assay Hi-C, rs900400 and the promoters of the protein-coding genes CCNL1, LEKR1, TIPARP, and SSR3 colocalizeto the same genomic subcompartment of the A2 type, which is asso- ciated with high gene density, high expression, and activating chro- matin marks. While the GM12878 cell line is not representative of adipose tissue, chromatin contact is largely stable across cell lines.

Discussion
Our findings support that a genetic variant at 3q25 influences adipo- cyte function differently at diverse life stages. Starting from a genome-wide agnostic investigation, we demonstrated that the T allele at rs900400 is associated with lower adiponectin levels, spe- cifically during pregnancy. This is the same genetic variant for which the T allele in offspring was associated with higher birth weight in a prior meta-analysis from the EGG consortium (16,19) and greater newborn adiposity in HAPO newborns (16). It is also notable that rs900400 was previously associated with leptin levels (b 6 SE 5 0.030 6 0.005 log-leptin per risk allele; P 55.631029 unadjusted for BMI; N 5 51,139 adults) (24) and with age at men- arche (b 6 SE 5 0.03 6 0.005 year per risk allele; P 5 2.3310211) (25), likely reflecting the role of adiposity in timing of puberty in women. Our current analyses reveal that the offspring T risk allele at rs900400 is strongly associated with higher cord blood leptin in newborns (P 5 8.2331029; N 5 502). Therefore, the same risk allele is associated with varying phenotypes related to adipocyte function at different times over the life course.Our findings support the concept that the association between rs900400 and adiponectin is enhanced by pregnancy-induced physio- logic changes or that we have identified a genetic determinant of pregnancy-specific mechanisms of adiponectin regulation (8). First, ADIPOgen data demonstrate only a modest association in a large sample of nonpregnant adults with similar effect sizes in men and women, arguing against a sex-specific effect. Second, the strength of association of adiponectin during pregnancy with maternal genotype at rs900400 compares favorably with rs17300539 in the promoter of ADIPOQ, the strongest genetic determinant of adiponectin in non- pregnant adults (5,6). Our observations suggest that pregnancy indu- ces a “metabolic stress test” on adipocyte function reflected by lower adiponectin and further indicate a lack of adipose tissue flexi- bility in rs900400 risk allele carriers. On the other hand, our find- ings could also be interpreted as women carrying the T allele having a stronger physiologic response of pregnancy-related hypoadiponec- tinemia, a potential adaptive mechanism to deliver more nutrients to the fetus (8). Adiponectin is exclusively produced by adipocytes, even in pregnancy (26), in contrast to leptin, which is highly expressed by the placenta (27). Pregnancy is characterized by an increase in multiple cytokines and hormones—estrogens, prolactin, cortisol, leptin—likely contributing to insulin resistance. Future functional studies may indicate whether some pregnancy-related cytokines/hormones mechanistically influence expression of adipo- nectin by interacting with rs900400.

Intriguingly, the risk variant for rs900400 in newborns demonstrates no association with cord adiponectin but did demonstrate strong association with cord leptin (b 6 SE 5 0.277 6 0.047 log-leptin per risk allele; P 5 8.2331029; N 5 502). Previous GWAS of leptin lev- els in more than 50,000 adults also revealed rs900400 as a genetic determinant of leptin levels but with a more modest effect size (b 6 SE 5 0.030 6 0.005 log-leptin per risk allele; P 5 5.631029 unad- justed for BMI) (24). These observations are puzzling in the context of adipocyte biology. Newborn adiponectin levels are positively cor- related with adiposity at birth but inversely correlated with excess weight later in life. Leptin levels reflect overall adiposity in both adults and newborns. Adipocytes secrete a greater amount of leptin as they differentiate and grow larger, even when overfilled with tri- glycerides (28). Small well-differentiated adipocytes produce high levels of adiponectin, but adiponectin secretion decreases as adipo- cytes become hypertrophic (1). Given our findings, we hypothesize that rs900400 T allele carriers have adipocytes that allow greater fat accumulation within adipocytes, leading to higher adiposity and lep- tin levels at birth but to dysfunction of adipocytes and lower adipo- nectin in the face of specific “environmental factors” such as pregnancy-induced physiologic changes. In contrast to nonpregnant individuals (24), we did not find an association between maternal genotype and maternal leptin during pregnancy. This lack of associ- ation in our population of pregnant women could be related to the fact that circulating leptin during pregnancy is substantially derived from placental production, which might not be under the same genetic influence as adipose tissue.

Nominal associations of the maternal T risk allele rs900400 with greater insulin resistance and higher glycemia during pregnancy could be downstream effects of adipocyte function, either as an adaptive pregnancy-specific mechanism to deliver nutrients to the fetus or as a sign of adipose tissue maladaptation. On one hand, the T allele at rs900400 was nominally associated with lower risk of T2D in the Nurses’ Health Study and Health Professionals Follow-up Study (29), suggesting a beneficial metabolic adaptation, yet this was not reported in the larger GWAS (30). On the other hand, associations with adi- posity distribution indices in GIANT participants are in line with adi- pocyte dysfunction, as a lack of flexibility in peripheral adipose tissue is believed to lead to central fat accumulation, represented by higher WHR. It is notable that we found absolutely no association of mater- nal rs900400 with BMI in our pregnant women or in more than 233,000 GIANT participants, supporting the idea that this variant likely influences adipocyte function and is not an obesity locus per se. In previous reports from GIANT, most WHR loci were not associated with BMI, and many genes at WHR loci pointed to adipogenesis, embryonic development, and angiogenesis (21).In eQTL analyses, we found that the rs900400 T allele was associ- ated with higher expression of TIPARP in adipocytes. TIPARP resides 374 kb upstream from rs900400 and colocalizes to the same genomic subcompartment (Supporting Information Figure S2) (23). TIPARP suppresses glucose production, possibly by depleting NAD1 levels, which may depress SIRT1 and ultimately PGC1a activity (31). PolyADP-ribose polymerase enzymes are emerging as coregulators of adipogenesis and glucose metabolism (32). Among all tissues in the European Bioinformatics Institute Gene Expression Atlas, TIPARP is most highly expressed in adipose tissue (33), and in the Genotype-Tissue Expression pilot project data, TIPARP is highly expressed in visceral adipose tissue (34).

Among loci that passed our initial discovery threshold but did not reach genome-wide significance after replication, we identified a few interesting biologic candidates, including PPP1R3A, FOXO1, and FADS1. PPP1R3A has been associated with rare severe insulin resistance disorders (with combined defect in PPARG) (35) and is part of the same family as PPP1R3B, which was recently associated with glycemic traits in pregnant women (10) and nonpregnant adults (36,37). FOXO1 may regulate adipocyte differentiation (38) and mediate insulin action in adipose tissue and hepatocytes (39). The FADS1 locus has been associated with fasting glucose (36,37) and multiple lipids and metabolites (40). It is likely that our relatively small sample size limited our power to detect associations with adi- ponectin levels at these loci, but our observations suggest that pregnancy-induced physiologic changes enhance genetic associations with adipocyte function, lipids, or insulin sensitivity pathways that otherwise necessitate a much larger sample size to observe.Our study was limited by sample size for some traits, and analyses were limited to women of European descent. Interpretation of gene expression microarray MuTHER data is limited by poor coverage of noncoding transcripts. Nevertheless, our study has numerous strengths. We tested genetic associations with many adiposity-related pheno- types in pregnant women and newborns using three population-based cohorts with prospective data/sample collection and standardized pro- tocols. Moreover, we expanded our results by accessing publicly available databases, including expression in adipocytes.

Conclusion
In conclusion, our findings suggest that rs900400 is implicated in adipocyte biology. T allele carriers at rs900400 have higher leptin levels and adiposity at birth, and female carriers show pregnancy- specific lowering of adiponectin levels. Investigating genotype- phenotype associations during pregnancy and early life permits the discovery of new biology not captured in genetic association studies conducted in general adult populations and sheds new light on adi- pocyte endocrine AZD5004 function.