Influences of Gestational Obesity on Associations between Genotypes and Gene Expression Levels in Offspring following Maternal Gastrointestinal Bypass Surgery for Obesity

Maternal obesity and excess gestational weight gain with compromised metabolic fitness predispose offspring to lifelong obesity and its comorbidities. We demonstrated that compared to offspring born before maternal gastrointestinal bypass surgery (BMS) those born after (AMS) were less obese, with less cardiometabolic risk reflected in the expression and methylation of diabetes, immune and inflammatory pathway genes. Here we examine relationships between gestational obesity and offspring gene variations on expression levels. Methods Whole-genome genotyping and gene expression analyses in blood of 22 BMS and 23 AMS offspring from 19 mothers were conducted using Illumina HumanOmni-5-Quad and HumanHT-12 v4 Expression BeadChips, respectively. Using PLINK we analyzed interactions between offspring gene variations and maternal surgical status on offspring gene expression levels. Altered biological functions and pathways were identified and visualized using DAVID and Ingenuity Pathway Analysis. Results Significant interactions (p ≤ 1.22x10-12) were found for 525 among the 16,060 expressed transcripts: 1.9% of tested SNPs were involved. Gene function and pathway analysis demonstrated enrichment of transcription and of cellular metabolism functions and overrepresentation of cellular stress and signaling, immune response, inflammation, growth, proliferation and development pathways. Conclusion We suggest that impaired maternal gestational metabolic fitness interacts with offspring gene variations modulating gene expression levels, providing potential mechanisms explaining improved cardiometabolic risk profiles of AMS offspring related to ameliorated maternal lipid and carbohydrate metabolism.


Introduction
Epidemiological studies demonstrate that parental obesity increases obesity risk in offspring and suggest an important role of the intrauterine environment owing to stronger associations between maternal than paternal body mass index (BMI) with offspring obesity [1,2]. Maternal obesity, excess gestational weight gain, high inter-pregnancy BMI and gestational diabetes increase risks of offspring obesity, type 2 diabetes mellitus (T2DM), cardiovascular disease (CVD) and fatty liver [3][4][5]. Environmental and genetic factors mediate the link between parental obesity and increased risk of obesity in offspring [2,6]; family and twin studies demonstrate heritability of obesity and CVD risk factors [7,8].
Large-scale genome-wide association studies (GWAS) have consistently revealed the presence of specific genes in metabolic diseases such as type 1 and T2DM [9,10] and obesity [7,11]. GWAS on expression traits identified variations regulating gene expression (expression quantitative trait loci; eQTL) and demonstrated that gene expression levels show complex inheritance patterns [12,13]. Such studies elucidate basic processes of gene regulation and may identify the pathogenesis of prevalent diseases adding information to associations identified by GWAS.
Gene expression levels are greatly affected by genetic and environmental factors [14] where gene variations have the potential to attenuate or amplify environmental effects. An adverse intrauterine environment has long been known to contribute to metabolic and cardiovascular diseases [15] where differences in expression levels between offspring born under different maternal conditions were reported for specific genes and at genome-wide level [16][17][18][19]. Several loci associated with specific traits interact with intrauterine environment [20][21][22]. A striking example of such gene-environment interaction is the association of SIRT1 SNPs with lower prevalence of type 2 diabetes observed in individuals prenatally exposed to famine in utero but not in those not exposed to famine.
Bariatric bypass operations improve glucose and lipid metabolism and treat and/or prevent hypertension, dyslipidemia, T2DM and fatty liver disease [23][24][25]. Similar to weight loss [26,27], bariatric surgery results in changes in gene expression levels [28,29]. Our studies uniquely demonstrated that offspring born after maternal gastrointestinal bypass surgery (AMS) exhibit lower prevalence of severe obesity, greater insulin sensitivity and improved lipid profiles compared to offspring born before maternal surgery (BMS) [30,31]. Recently, we demonstrated that these improvements are associated with differences in gene expression and methylation of genes involved in diabetes and immune and inflammatory pathways [17,32].
In order to further explore the role of the intrauterine environment in the determination of offspring phenotype and to provide molecular mechanisms explaining changes in cardiometabolic risk markers of AMS vs. BMS offspring, we studied the combined influence of maternal surgical status and offspring gene variations on offspring gene expression levels.

Materials and Methods Subjects
Women from Quebec City and surrounding areas (administrative regions of Capitale-Nationale, Mauricie and Chaudière-Appalaches) who had given birth before and after biliopancreatic diversion with duodenal switch [25] for severe obesity were eligible. We recruited a subset of 19 unrelated mothers aged 34-51 years having offspring aged 2-23 years, 22 born before and 23 after maternal operations. Between July and October 2010 mothers and offspring visited the Quebec Heart and Lung Institute (Quebec City, Quebec, Canada) or a regional hospital for clinical evaluation and blood sampling. There were 15 mothers with siblings born before and after surgery (21 BMS and 18 AMS), one with BMS offspring only (1 BMS) and 3 mothers with only AMS offspring (5 AMS).
Maternal pre-surgical data were obtained from medical records. At the office visit weight and percent body fat were determined for individuals aged 6 years or more (BMS, N = 21; AMS, N = 15) using bioelectric impedance analysis (Tanita; Arlington Heights, IL). Height and resting systolic (SBP) and diastolic (DBP) blood pressure were obtained using standardized procedures. BMI was calculated for mothers and adults and BMI percentiles for children were obtained from the National Health and Nutrition Examination Survey 2000 chart [33]. BMI Z-score was calculated for children using charts from the Centers for Disease Control and Prevention [34]. Fasting whole blood samples were collected from an antecubital vein into tubes containing EDTA and PAXgene Blood RNA collection tubes (Qiagen, Valencia, CA, USA). Plasma lipid, glucose and insulin concentrations were measured as previously described [35]. Lipid and glucose levels values from 3 AMS non-fasting offspring were excluded. The homeostatic model assessment of insulin resistance (HOMA-IR) index was calculated as fasting glucose x insulin/22.5. Levels of high-sensitivity C-reactive protein (CRP) were measured with a BN ProSpec nephelometer (Siemens Canada Limited, Oakville, Ontario, Canada) [36]. CRP values under the detection limit (< 0.17 mg/L) were arbitrarily set at detection limit.

Ethics Statement
This study was approved by the Quebec Heart and Lung Institute Ethics Committee. Written informed consent was obtained from mothers and adult offspring and assent from minor offspring were obtained from mothers.

Gene expression analysis
Gene expression levels of the 45 offspring analyzed here were obtained from previous studies from our group evaluating differences in gene expression and methylation of genes involved in diabetes, immune and inflammatory pathways [17,32]. Briefly, total RNA was isolated and purified from offspring whole blood using PAXgene Blood RNA Kit (Qiagen). The quality and integrity of the purified RNA was assessed using both the NanoDrop (Thermo Scientific, Wilmington, DE, USA) and the 2100 Bioanalyzer (Agilent Technologies, Cedar Creek, TX, USA). Expression levels were measured using the HumanHT-12 v4 Expression BeadChip (Illumina Inc., San Diego, CA) with 250 ng of total RNA and processed at the McGill University and Genome Quebec Innovation Centre (Montreal, Canada). Expression data were visualized and analyzed using the FlexArray software [37] (version 1.6) and the lumi R package was used for expression data analysis and normalization. To be considered as expressed, a probe had to show a detection p-value 0.05 in at least 25% of samples of a group. Among the 47,323 probes on the microarray, 16,060 (33.9%) showed significant gene expression in blood and were used as dependent expression phenotypes (expression traits) for analysis of interactions between offspring gene variations and maternal obesity status (GEO accession number GSE44407).

DNA extraction and genome-wide genotyping
Genomic DNA was isolated from offspring blood buffy coat using the GenElute Blood Genomic DNA kit (

SNPs and sample quality control
Calculations of allele frequencies and tests of SNP data for Hardy-Weinberg equilibrium (HWE) were performed using PLINK [38] (version 1.07). Standard quality control exclusion criteria for the SNPs were used: call rate < 95%, genotype distribution deviating from Hardy-Weinberg Equilibrium (p-values less than 10 -7 ) and monomorphic SNPs or those with a minor allele frequency (MAF) < 0.01 [39]. A total of 1,751,034 SNPs were excluded leaving 2,550,297 SNPs for statistical analyses. All samples were tested for call rate (> 90%), ethnicity (Caucasian; HapMap) and gender mismatch based on genotyping data. No subjects were excluded: all 45 samples were used in further analysis.

Statistical analysis
Anthropometric and clinical data were expressed as mean ± SD. Maternal treatment effect on anthropometric-, blood pressure-, lipid profile-and glucose-related variables was assessed using a within-subject, paired t-test. Differences between BMS and AMS offspring were tested using analysis of variance (general linear model, type III sum of squares) and adjusted for the effects of sex and puberty. BMI percentile and BMI Z-score being obtained from age-and sexspecific charts, no further adjustments for age and sex were made to test for differences between BMS and AMS offspring for those adiposity measurements. Severe obesity in offspring was defined as BMI percentile > 98% and Z-score > 3. Transformations were applied to nonnormally distributed variables (log10 transformed for insulin and HOMA-IR; negative inverse transformed for C-reactive protein). In the absence of Tanner scores, we arbitrarily defined puberty as 12 years for female and 14 years for male offspring based on Canadian sex-specific probabilities of having entered puberty [40]. Differences in severe obesity between BMS and AMS offspring were evaluated using BMI percentile and BMI Z-score and tested using Fisher's exact test. P-values for CRP were adjusted for the effects of sex, puberty and BMI percentile. Statistical analyses were done using the SAS software version 9.2 (SAS Institute Inc). Statistical significance was defined as p 0.05. Interactions between offspring gene variations and maternal surgical status were tested on offspring gene expression levels in whole blood using PLINK. Differences in regression slopes obtained from additive model were then tested between BMS and AMS offspring. Bonferroni correction was applied to correct for multiple testing of offspring gene variation x environment interactions thus leading to a p-value cutoff of p 1.22x10-12 (as calculated with 0.05/ (2,550,297 SNPs x 16,060 transcripts)) to claim statistical significance. Linkage disequilibrium (LD; r 2 ) between SNPs demonstrating significant interactions was calculated using Haploview [41] to assess the number of independent (nonlinked) SNPs. The tagger algorithm implemented in Haploview was used to identify tag SNPs among the significant polymorphisms (r 2 threshold = 0.8).

Gene functions and pathways analysis
Two independent function and pathway analysis tools were employed to identify potentially enriched functions and overrepresented pathways from the list of transcripts demonstrating significant interactions, namely the Database for Annotation, Visualization, and Integrated Discovery (DAVID; http://david.abcc.ncifcrf.gov) bioinformatics resources [42,43] and Ingenuity Pathway Analysis (IPA). DAVID provided annotation for the list of transcripts and computed annotation term enrichment to highlight the most relevant functions from the list of transcripts. Similarly, IPA classified each transcript from this list according to function and pathway. Using a right-tailed Fisher's exact test, IPA measured the likelihood that transcripts from the list participate in each function/pathway solely due to chance and calculated p-values. Enriched functions and overrepresented pathways were then obtained from transcripts showing interactions.
Offspring ages varied between 2 years 8 months and 23 years 9 months, with similar sex distributions in the two groups (41% male in BMS vs. 43% in AMS; Table 1). BMS offspring were born 3 years 4 months (40.2 ± 28.0 months) before and AMS 3 years 9 months (44.9 ± 26.6 months) after maternal surgery. BMS offspring were older than AMS at follow-up (14.5 ± 5.7 vs. 9.0 ± 5.0 years; p = 0.001; BMS range: 5 years 9 months to 23 years 9 months; AMS range: 2 years 8 months to 19 years 6 months). Severe obesity was less prevalent in AMS using BMI percentile (p = 0.01) or BMI Z-score (p = 0.02). Adjusting for sex and puberty, AMS offspring exhibited trends toward lower fasting insulin levels and HOMA-IR index, and lower diastolic blood pressure (p < 0.10 for all).
Statistically significant offspring SNP-by-maternal surgical status interactions identified for the 48,156 unique SNPs involved 525 unique transcripts (3.3%) thus implying that a single transcript might be under multiple genetic constraints. Indeed, 375 of these 525 unique transcripts (71.4%) demonstrated multiple (2) significant interactions. The most highly represented transcripts from significant interactions are shown in Table 3, including transcripts encoding genes involved in regulation of gene expression per se, immune and inflammatory Values are presented as mean ± SD.

Gene functions and pathways
Clustering of the 525 transcripts showing significant interactions based on ontology using DAVID identified 5 over-represented functional categories (-log group enrichment score > 1.30; p-value for group enrichment score < 0.05): 1) transcription, 2) metabolic process, 3) guanine nucleotide exchange factor, 4) death/ZU5 domain and 5) zinc finger domain (S4 Table). IPA analysis revealed infectious disease, inflammatory response, gene expression, and cellular growth and proliferation among the over-represented functional categories from the list of transcripts demonstrating statistically significant interactions, thus highlighting transcription and cellular metabolism functions using both function analysis tools. Similarly, IPA revealed 18 pathways enriched for these transcripts, including 12 related to cellular stress and signaling, immune response and inflammation, and growth, proliferation and development. Importantly, DNA Double-Strand Break Repair by Homologous Recombination was the most overrepresented pathway (p = 0.001) and carbohydrate and lipid biosynthesis/degradation pathways were also identified ( Fig. 1).

Discussion
In a unique offspring cohort born discordant for maternal biliopancreatic bypass surgery affecting maternal metabolic fitness we extend observations that an adverse dysmetabolic intrauterine environment is associated with subsequent obesity and cardiometabolic risk [30,31] related to gene expression levels [16,18,19]. Using a cohort in which gene expression and methylation levels were previously evaluated in regards to metabolic differences between BMS and AMS offspring [17,32], the current study focusing on a different aspect (gene-environment interactions) demonstrated modulatory effects of maternal fitness on the association between genotype and gene expression in offspring. In this study we tested gene expression levels (expression traits) for interactions between offspring SNPs and maternal treatment. By analyzing the offspring genotypes and gene expression at the genome-wide level combined with the impact of maternal status, we provide an objective insight into the relation between maternal status and offspring cardiometabolic risk profile and have the potential to elucidate the functional basis of known associations identified previously.
Overlapping SNPs identified here with those previously associated with specific phenotypes and metabolic variables in GWAS has the potential to elucidate mechanisms for previously reported associations. Systematic comparison of SNPs and regulated transcripts with results from GWAS was then conducted. Our top interactions (Table 2) identified SNP rs57573303 in the gene LINGO2 and SNP rs3803712 located in HS3ST4. The former was associated with  BEX5 expression levels, a brain expressed X-linked gene family member regulating differentiation of dopamine neurons involved in food reward signaling [44]. The latter we found was associated with CEACAM1 gene expression levels encoding a cell-cell adhesion molecule involved in differentiation, apoptosis and modulation of innate and adaptive immune response consistent with decreased liver CEACAM1 expression reported in severely obese patients [45]. In regards to the interaction identified here, increased CEACAM1 expression was observed in heterozygous AMS offspring while BMS rare allele carriers demonstrated lower CEACAM1 expression. SNPs in LINGO2 and HS3ST4 were previously found to be associated with children BMI z-score and % fat mass, respectively [11]. Associations between rs10968576, also in LINGO2, and fasting plasma cholesterol levels and BMI were found in different populations [46,47]. We identify here rs115462216 SNP (kgp10871570) in NCAM2 for which rs11088859 has been associated with waist circumference [48]. In addition, the presence of neurological (ERCC8, KISS1, IDO1, OTOF, SEPT4, SERPING1, SOCS7, STAT2, TGIF1, TIAM2) and endocrine system development (CHRM2, SC5D, SOCS7) genes among the regulated transcripts ( Table 2) suggest a mechanistic role in offspring programming under varying maternal conditions [19,49]. Taken together, the identification of SNPs located in genes previously reported by others to be associated with obesity traits support our results and is consistent with studies showing that SNPs associated with complex traits are more likely to be eQTLs [50,51]. The limited number of transcripts for which we found offspring gene variation-by-maternal status interactions and the large number of SNPs suggest that single transcripts may be under multiple genetic constraints, coherent with previous studies conducted on larger cohorts [52][53][54]. Among the numerous SNPs we identified with significant interactions, many were specifically associated with the inflammatory, insulin-resistant, dysmetabolic diathesis of diabesity such as SNPs in the IFNAR1-IL10RB gene region forming the class II cytokine receptor gene cluster (S3 Table). Those genes involved in IL10-induced signal transductions were associated with inflammatory diseases and ischemic stroke with hypertension [55][56][57]. We also observed significant interactions for SNPs located near the GPR175 (TPRA1) gene, expression level of which was previously demonstrated to be associated with plasma lipid levels [58]. Furthermore, we found interactions of SNPs near BARX2, a member of the homeobox transcription factor family known to influence cellular processes controlling cell adhesion and remodeling actin cytoskeleton. Others previously found such associations with T2DM and end-stage renal disease [59].
Gene function analysis conducted with two independent tools highlighted genes related to transcription and cellular metabolism. Combined with other overrepresented functional categories (guanine nucleotide exchange factor, death/ZU5 domain, zinc finger domain), these results suggested some potential effects of maternal metabolic fitness on offspring at both the cellular and transcriptional levels. Our pathway analysis also identified pathways related to cellular growth, proliferation and development, stress and signaling as well as carbohydrate and lipid metabolism similar to others' findings. Global gene expression analysis of amniotic fluid cell-free fetal RNA identified lipid (apolipoprotein D) and transcriptional regulators (FOS and STAT3) as well as apoptotic cell death-related genes among differentially expressed genes between fetuses of obese vs. lean pregnant women [16]. In conjunction with influences of maternal obesity before conception reported on gene expression profiles of rat embryos, genes related to cell cycle, carbohydrate metabolism, DNA repair and transcriptional regulator were altered [60]. These results strengthen our observations pertaining to an overrepresentation of cellular processes (cellular growth, proliferation and development), carbohydrate and lipid metabolism pathways identified. The study on rat embryos [60] also supports overrepresentation of immune response and inflammatory genes as well as DNA double-strand break repairrelated genes among transcripts with significant interactions. Similar to changes in expression previously observed in obese patients early after having undergone weight loss or bariatric surgery [26,27,29], we found that inflammation-related transcripts were overrepresented. Globally, overrepresentation of cellular signaling, carbohydrate metabolism and inflammatory pathways identified here from the list of transcripts showing significant offspring gene variations by maternal surgical status interactions are in line with previous results from our group comparing BMS and AMS offspring at gene methylation and expression levels [17,32]. In addition, our results are consonant with murine studies showing effects of maternal gestational obesity and high-fat diet on offspring with differences in gene expression levels for genes related to inflammation and glucose homeostasis [61,62] and for pathways related to cellular stress, signaling, growth, proliferation, development and regulation of lipogenic pathways [63] of significant pathogenic importance for the dysmetabolic diathesis of diabesity. Our pathway analysis demonstrated involvement of carbohydrate and lipid metabolism pathways, in agreement with results from maternal weight loss studies in sheep demonstrating an impact on insulin signaling, glucose transport and glycogen synthesis pathways in offspring' skeletal muscle [64] as well as with previous studies demonstrating gene-by-maternal diet interactions in offspring [21,65].
Some of the modulating effects of gestational metabolic fitness may be confounded. Young age of the offspring limits the potential contribution of different postnatal environments. However, it does not allow extrapolation over the life span stretching into mature adulthood when most pathology emerges through the cumulative effects of environmental exposures. Nevertheless, the preponderance of literature on developmental origins of adult disease, specifically for cardio-metabolic outcomes related to our findings demonstrates durability of effects over the life-span [15]. The rarity of gastrointestinal biliopancreatic bypass surgery, low pregnancy rates before and after maternal surgery, constraints of study design and the exclusive nature of the molecular analyses all limited the size of our offspring population. The size of the study sample limited the number of adjustments made to correct for confounding factors relating to offspring and maternal condition during pregnancy (breastfeeding, smoking, etc.). Adjustments for confounding factors were thus limited to sex and puberty. Although metabolic parameters in the offspring cohorts were not statistically significantly different owing to sample size, the differences were clinically significant particularly for insulin resistance, dyslipidemia and CRP. Studies from our group conducted on larger cohorts have previously demonstrated robust group differences between BMS and AMS offspring [30,31]. Our gene analyses were performed on blood, more convenient to obtain and to justify sampling than other tissues in healthy juvenile offspring. We and others have reported partial inter-tissue correlations [54,[66][67][68]. The multicellular nature of blood constitutes an inherent limitation in our study; tissue heterogeneity potentially influenced measurement of gene expression levels [69,70]. Nonetheless, we assessed gene expression as representative of systemic biological differences between BMS and AMS offspring to which multiple organs and tissues have contributed. Causality of identified variants cannot be determined: identified variants might be markers of genomic regions or loci in which causal variants lie and allelic heterogeneity cannot be ruled out, together necessitating much larger population studies than our unique but relatively small cohort study.
Strengths of our study are the unique genetically and phenotypically characterized offspring cohort discordant for maternal gestational metabolic fitness, the efficacious standardized currently performed metabolic operation as a tool to alter the intrauterine milieu and an exceptionally high follow-up rate enabled by the national health insurance system. The biliopancreatic bypass operation selectively increases steatorrhea, lowering maternal plasma free fatty acids, reducing fatty infiltration of metabolically active tissues, reducing lipid peroxidation and systemic lipotoxic inflammation altogether improving insulin action and glucose disposal approximating pre-obese levels. Although these durable effects were not replicated quantitatively or qualitatively by other current bariatric operations, they add critical insight into molecular mechanisms associated with the gene expression levels presented here.
Our results demonstrated influences of the intrauterine metabolic environment on associations between offspring genotype and gene expression levels. The lower prevalence of obesity and cardiometabolic risk observed in AMS offspring argues for implementation of maternal weight loss and improved metabolic fitness before pregnancy and provides potential mechanisms for physiological improvements through regulation of gene expression in offspring.