Genome-wide Association of Single-Nucleotide Polymorphisms With Weight Loss Outcomes After Roux-en-Y Gastric Bypass Surgery

Abstract

Context:

Roux-en-Y gastric bypass (RYGB) is among the most effective treatments for extreme obesity and obesity-related complications. However, despite its potential efficacy, many patients do not achieve and/or maintain sufficient weight loss.

Objective:

Our objective was to identify genetic factors underlying the variability in weight loss outcomes after RYGB surgery.

Design:

We conducted a genome-wide association study using a 2-stage phenotypic extreme study design.

Setting:

Patients were recruited from a comprehensive weight loss program at an integrated health system.

Patients:

Eighty-six obese (body mass index >35 kg/m²) patients who had the least percent excess body weight loss (%EBWL) and 89 patients who had the most %EBWL at 2 years after surgery were genotyped using Affymetrix version 6.0 single-nucleotide polymorphism (SNP) arrays. A second group from the same cohort consisting of 164 patients in the lower quartile of %EBWL and 169 from the upper quartile were selected for evaluation of candidate regions using custom SNP arrays.

Intervention:

We performed RYGB surgery.

Main Outcome Measures:

We assessed %EBWL at 2 years after RYGB and SNPs.

Results:

We identified 111 SNPs in the first-stage analysis whose frequencies were significantly different between 2 phenotypic extremes of weight loss (allelic χ² test P < .0001). Linear regression of %EBWL at 2 years after surgery revealed 17 SNPs that approach P < .05 in the validation stage and cluster in or near several genes with potential biological relevance including PKHD1, HTR1A, NMBR, and IGF1R.

Conclusions:

This is the first genome-wide association study of weight loss response to RYGB. Variation in weight loss outcomes after RYGB may be influenced by several common genetic variants.

Roux-en-Y gastric bypass (RYGB) surgery is among the most successful interventions for long-term weight loss in extreme obesity, inducing 60% to 70% excess body weight loss (EBWL) within the first 12 months (1). However, the degree of weight loss after surgery is variable with an estimated 20% of patients failing to achieve or maintain sufficient weight loss (generally defined as ≥50% EBWL) (2). Factors affecting after RYGB weight loss include higher initial body mass index (BMI) and type 2 diabetes (3–8). Studies of twins and close relatives have provided evidence of a genetic component to dietary and surgical weight loss (9, 10). For example, a recent study reported that the difference in average percent EBWL (%EWL) between 2 first-degree relatives was 9.9% versus 25.4% observed between a given pair of unrelated individuals in response to RYGB (11). A few studies have examined the influence of single-nucleotide polymorphisms (SNPs) in known obesity genes on weight loss after bariatric surgery, but none has used a genome-wide association approach.

We employed a genome-wide approach using a high-density SNP array, as well as detailed clinical information gathered on a large cohort of RYGB patients, to identify common genetic variants associated with EBWL after RYGB surgery.

Patients and Methods

A cohort of 1143 Caucasian patients underwent either open or laparoscopic primary RYGB by 1 of 3 surgeons from September 2004 to May 2008 and were enrolled into an institutional review board-approved study (12). DNA and 2 years of postoperative weight loss data were available for 1008 of these patients (DNA and complete weight loss data were not available for 76 and 59 patients, respectively). Patients underwent a standardized preoperative assessment providing comprehensive clinical and laboratory measures as well as a blood sample for DNA analysis. Weight measurements were obtained at each visit. The %EBWL was calculated as the difference of an estimate of fat mass using BMI and an idealized BMI of 25 kg/m². Eighty-six patients who had the least EBWL at 2 years after RYGB and 89 patients who had the most EBWL after RYGB were selected for the first-stage cohort of a case-control genome-wide association study (GWAS). An additional 164 and 169 patients of the remaining upper and lower quartiles of postsurgery EBWL were selected for the second-stage cohort.

Whole-blood genomic DNA was isolated as previously described (13). First-stage discovery cohort samples were genotyped using the Affymetrix version 6.0 SNP platform (Santa Clara, California). Custom SNP assays were designed using KBioscience competitive allele-specific PCR (KASPar) chemistry (http://www.lgcgenomics.com/kasp-overview) for 111 SNPs for testing in the second-stage cohort samples.

Principal components analysis was conducted using HapMap3 data that identified only 1 of the 175 first-stage patients as a significant outlier from the CEU population (Utah residents with Northern and Western European ancestry from the CEPH collection). The genomic inflation factor for the first-stage cohort was 1.11 (PLINK software: http://pngu.mgh.harvard.edu/purcell/plink/), indicating no strong stratification of the cohort. Finally, a Q-Q plot of the first-stage cohort SNPs (Supplemental Figure 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org) showed that the study was well controlled for population confounders.

All genetic association analyses were performed using PLINK. The χ² association analyses were performed on allele counts, and linear regression analysis was performed on each independent SNP as a predictor (with or without other predictors) with year-2 %EBWL as the dependent variable. Haplotypes were identified (using PLINK “blocks” command) and analyzed using linear regression analysis with year-2 %EBWL.

All clinical and demographic data were analyzed with the 2 weight loss extremes using Pearson χ² for categorical data or t tests for continuous data (SPSS version 19). All clinical factors were also tested independently using linear regression with year-2 %EBWL as the dependent variable in a pooled cohort of both first-stage and second-stage patients. The significant clinical variables (P < .05) were then incorporated as covariates in multivariate analyses with each of the top SNPs using year-2 %EBWL as the phenotype (PLINK). Clinical covariates that were no longer significant in multivariate models were removed, and remaining clinical covariates were tested with each SNP in each cohort.

For the top SNPs, individual and multivariate P values for each cohort are reported, as well as a Fisher meta-analysis P value combining the 2 cohorts. Cross-validation by bootstrapping was done with 1000 iterations, randomly selecting new first-stage and second-stage cohorts (175 and 333 individuals, respectively, matching the original cohorts) and calculating a Fisher meta P value at each iteration (R software). The 95% confidence intervals (CI) are reported.

Results

First-stage and second-stage cohort subjects were selected based on plots of %EBWL from 1 year before surgery up to 4 years after surgery (Figure 1, A and B). Those patients with the least %EBWL (least successful) and those who had the most %EBWL (most successful) at 2 years after RYGB were selected for first-stage analysis. The 89 most successful patients had 101% ± 9.88% EBWL at 2 years after RYGB; ie, they essentially returned to a normal BMI of 25 kg/m², whereas the 86 least successful had just 43.2% ± 8.85% EBWL (Figure 1C). Similarly for the second-stage group, recruited from the remaining cohort, the 169 most successful patients had 96.6% ± 15.9% EBWL and the 164 least successful patients had 49.9% ± 10.4% EBWL at 2 years after RYGB.

Figure 1.

Loss of excess body weight after RYGB. Percent excess body weight was measured for patients every 6 months, starting 12 months before RYGB surgery (time of surgery indicated by vertical gray line) until at least 24 months after surgery. Patients were classified by high degree of EBWL (below dashed line) or low amount of EBWL (above dashed line) at 24 months after surgery for both the first-stage (A) and the second-stage (B) cohorts. Patient data are compared between most successful and least successful weight loss outcomes (C).

The first-stage cohort was genotyped for 730 767 SNPs. The most significant 111 SNPs (allelic-χ² P < 1 × 10⁻⁴; Supplemental Figure 2 and Supplemental Table 1) were carried forward for further evaluation using a lenient statistical threshold to minimize false-negative results due to small sample size. Of the 111, 17 SNPs approaching P < .05 in the second-stage cohort had odds ratios in the same direction as those observed with the first-stage cohort. Supplemental Table 2 shows the pooled minor allele frequencies and odds ratios for these SNPs. Because most the patients were female (Figure 1C), data excluding males are shown separately as well. Table 1 shows linear regression P values for the 17 SNPs in each cohort as well as meta P values as a measure of validation that these SNPs are associated with year-2 %EBWL in both cohorts (P < .05). These 17 SNPs cover 6 genes/regions: PKHD1, IPO11/HTR1A, CITED2/NMBR, GUCY1A2, IGF1R, and CENPF/KCNK2. To further estimate the robustness of statistical association, cross-validation was performed by randomly selecting subgroups from the first-stage and second-stage cohorts and recalculating significance levels. Confidence intervals from the bootstrapping analysis are reported (Table 1). Importantly, one PKHD1 SNP (rs2894788; P = 7.40 × 10⁻⁸) approached genome-wide significance and another SNP (rs728996; P = 4.00 × 10⁻⁸) as well as the 6-SNP PKHD1 haplotype CCAACT (Supplemental Table 3; P = 5.00 × 10⁻⁸) reached genome-wide significance in females.

Table 1.

Top SNPs Associated With %EBWL 2 Years After RYGB

Gene/Region and SNP	Variant	Sex	Linear Regression P values			95% CI
			First Stage	Second Stage	Fisher Meta
PKHD1
rs728996	C	Females	5.95 × 10−6	3.19 × 10−4	4.00 × 10−8	9.46 × 10−9, 1.48 × 10−7
		All	5.95 × 10−6	4.68 × 10−3	5.12 × 10−7	3.58 × 10−7, 4.81 × 10−6
rs2894788	C	Females	6.68 × 10−6	5.42 × 10−4	7.40 × 10−8	1.87 × 10−8, 2.50 × 10−7
		All	6.68 × 10−6	7.71 × 10−3	9.16 × 10−7	6.71 × 10−7, 9.90 × 10−6
rs6458777	A	Females	2.08 × 10−5	8.72 × 10−4	3.41 × 10−7	7.18 × 10−8, 9.52 × 10−7
		All	2.08 × 10−5	1.09 × 10−2	3.70 × 10−6	2.14 × 10−6, 2.88 × 10−5
rs4715233	A	Females	1.48 × 10−5	5.74 × 10−4	1.66 × 10−7	3.79 × 10−8, 5.05 × 10−7
		All	1.48 × 10−5	8.56 × 10−3	2.14 × 10−6	1.27 × 10−6, 1.74 × 10−5
rs9296661	C	Females	4.74 × 10−5	7.75 × 10−3	5.81 × 10−6	1.08 × 10−6, 1.57 × 10−5
		All	4.74 × 10−5	7.50 × 10−2	4.82 × 10−5	3.75 × 10−5, 5.03 × 10−4
rs1326589	T	Females	5.46 × 10−5	4.31 × 10−3	3.83 × 10−6	5.36 × 10−7, 8.95 × 10−6
		All	5.46 × 10−5	4.33 × 10−2	3.30 × 10−5	1.81 × 10−5, 2.49 × 10−4
IPO11/HTR1A
rs349487	C	Females	7.88 × 10−5	7.63 × 10−2	7.83 × 10−5	1.91 × 10−5, 2.36 × 10−4
		All	7.88 × 10−5	5.64 × 10−2	5.92 × 10−5	2.71 × 10−5, 3.10 × 10−4
rs448936	T	Females	6.77 × 10−5	4.71 × 10−2	4.35 × 10−5	1.17 × 10−5, 1.25 × 10−4
		All	6.77 × 10−5	2.08 × 10−2	2.04 × 10−5	7.51 × 10−6, 9.08 × 10−5
rs369747	C	Females	7.88 × 10−5	2.35 × 10−2	2.63 × 10−5	4.11 × 10−6, 5.72 × 10−5
		All	7.88 × 10−5	1.37 × 10−2	1.59 × 10−5	4.39 × 10−6, 5.68 × 10−5
rs424672	C	Females	8.67 × 10−5	3.27 × 10−2	3.90 × 10−5	6.03 × 10−6, 8.72 × 10−5
		All	8.67 × 10−5	1.71 × 10−2	2.14 × 10−5	6.49 × 10−6, 7.84 × 10−5
CITED2/NMBR
rs1414839	G	Females	1.52 × 10−3	0.12	1.75 × 10−3	2.65 × 10−4, 3.25 × 10−3
		All	1.52 × 10−3	1.09 × 10−2	1.99 × 10−4	3.51 × 10−5, 3.72 × 10−4
rs967790	T	Females	1.12 × 10−3	7.14 × 10−2	8.34 × 10−4	1.19 × 10−4, 1.54 × 10−3
		All	1.12 × 10−3	6.03 × 10−3	8.72 × 10−5	1.45 × 10−5, 1.63 × 10−4
GUCY1A2
rs7943998	C	Females	1.88 × 10−4	4.85 × 10−2	1.15 × 10−4	2.24 × 10−5, 2.50 × 10−4
		All	1.88 × 10−4	1.80 × 10−2	4.60 × 10−5	1.57 × 10−5, 1.49 × 10−4
rs7950959	T	Females	6.02 × 10−4	0.13	8.18 × 10−4	1.95 × 10−4, 2.01 × 10−3
		All	6.02 × 10−4	4.66 × 10−2	3.22 × 10−4	1.13 × 10−4, 1.01 × 10−3
IGF1R
rs8031382	T	Females	7.54 × 10−4	1.86 × 10−2	1.71 × 10−4	2.60 × 10−5, 2.46 × 10−4
		All	7.54 × 10−4	6.85 × 10−2	5.61 × 10−4	2.26 × 10−4, 2.00 × 10−3
rs12439557	T	Females	4.91 × 10−4	1.96 × 10−2	1.21 × 10−4	2.80 × 10−5, 2.38 × 10−4
		All	4.91 × 10−4	7.22 × 10−2	3.99 × 10−4	2.50 × 10−4, 2.30 × 10−3
CENPF/KCNK2
rs1858650	A	Females	2.04 × 10−5	5.85 × 10−2	1.75 × 10−5	8.11 × 10−6, 8.62 × 10−5
		All	2.04 × 10−5	0.10	2.88 × 10−5	2.95 × 10−5, 4.04 × 10−4

Clinical data collected before surgery were analyzed independently with year-2 %EBWL; 28 variables were determined to be associated with response to RYGB (linear regression P < .05; Supplemental Table 4). These 28 clinical variables were then used as covariates with each individual SNP in multivariate analyses of year-2 %EBWL. All clinical covariates dropped out as predictors of RYGB response except for initial BMI, biguanide use, and PUBScore (a quality of life instrument measure of public distress). Multivariate analysis inclusive of these 3 covariates and each SNP revealed only initial BMI and biguanide use as significant covariates with SNPs in association with year-2 %EBWL. Adjustment for initial BMI and biguanide use did not completely negate the association between each of the SNPs and response to RYGB in the first-stage analysis but rendered many of SNPs nonsignificant in the second-stage cohort (Supplemental Table 5). However, meta P values of both cohorts show that all 17 SNPs maintain P < .05 even when adjusting for initial BMI and biguanide use.

Exploratory predictive models were constructed to evaluate the potential clinical utility of the top 17 SNPs and the 2 clinical variables that were associated with response to RYGB, initial BMI, and biguanide use (Supplemental Methods). The accuracy of genetic or clinical predictors alone as measured using area under the receiver operating characteristic curve was 0.69 (95% CI, 0.65, 0.74) and 0.78 (95% CI, 0.74, 0.82), respectively (Supplemental Figure 3). Combining genetic and clinical predictors improved the accuracy of the predictive model for RYGB response to 0.82 (95% CI, 0.78, 0.85).

Discussion

A number of loci have been identified through GWAS to be associated with obesity. No studies to date have used a GWAS approach to identify variants associated with outcomes of bariatric surgery. We conducted a GWAS using an extreme-trait sampling followed by second-stage evaluation to identify 17 SNPs associated with %EBWL 2 years after RYGB surgery. To minimize population stratification and potential confounders, subjects were selected from a cohort of over 1000 Caucasian patients treated at one institution in a uniform clinical program 2 years after undergoing a standardized RYGB bariatric procedure.

Although we were limited by sample size, cross-validation results and Fisher meta P values, even in the presence of clinical covariates, revealed some SNPs that were persistently associated with %EBWL at 2 years after RYGB. Moreover, the Fisher meta P value of one of the PKHD1 SNPs, rs2894788, approached genome-wide significance whereas another SNP, rs728996, and the 6-SNP haplotype CCAACT did reach genome-wide significance in females.

Some potentially relevant biological information is known about the genes/regions where the SNPs reside. Six of the SNPs are intronic within PKHD1, a gene implicated in antipsychotic drug-induced weight gain, potentially through roles in fat and glucose metabolism (14). Serotonin receptor genes, especially 5-HTR2C, were also suggested to play a role in antipsychotic-induced weight gain. We identified 4 SNPs upstream of a member of this family known to have a role in appetite, HTR1A (15). We also identified 2 SNPs upstream of NMBR, a gene previously associated with waist circumference (16). Another gene, GUCY1A2, has copy number variations that have been associated with severe early-onset obesity (17). Additionally, IGF1R (18), CENFP (19), and CITED2 (20) have been implicated in insulin and glucose metabolism.

There is a noticeably strong relationship between year-2 %EBWL and initial BMI, and the patients in the 2 extremes have significantly different initial BMIs. Other studies have also observed this trend (7, 8); therefore, it was important that we adjust for such an important confounder. Indeed, despite a large number of available clinical measurements taken before surgery, only initial BMI and biguanide use were still associated with %EBWL at 2 years after RYGB when combined with the SNPs in multivariate analyses. The fact that some SNPs/haplotypes have small P values in multivariate analysis with adjustment of initial BMI and biguanide use suggests that the SNPs/haplotypes identified may explain extra portions of variation in year-2 %EBWL beyond those explained by initial BMI.

The weight loss SNPs identified may have potential clinical utility for predicting response to RYGB. In exploratory modeling, combining SNP genotypes with the use of biguanides (a surrogate for type 2 diabetes and/or insulin resistance) and initial BMI resulted in an area under the curve of 0.82. Because biguanide use and BMI are both readily available parameters, the addition of SNPs to increase the predictive power may add new information that could be used to guide therapy. Future studies in other cohorts will be needed to fully delineate the interaction among genetic and clinical factors influencing outcomes after RYGB surgery.

Conclusions

This study is the first to apply GWAS discovery and clinical information to weight loss response to RYGB and represents a proof-of-principle for future studies.