Genetic Predisposition to Weight Loss and Regain With Lifestyle Intervention

Analyses From the Diabetes Prevention Program and the Look AHEAD Randomized Controlled Trials

George D. Papandonatos; Qing Pan; Nicholas M. Pajewski; Linda M. Delahanty; Inga Peter; Bahar Erar; Shafqat Ahmad; Maegan Harden; Ling Chen; Pierre Fontanillas; GIANT Consortium, Lynne E. Wagenknecht; Steven E. Kahn; Rena R. Wing; Kathleen A. Jablonski; Gordon S. Huggins; William C. Knowler; Jose C. Florez; Jeanne M. McCaffery; Paul W. Franks

Disclosures

Diabetes. 2015;64(12):4312-4321. 

In This Article

Research Design and Methods

DPP

The DPP was a 27-site multicenter parallel arm RCT that assessed the effects of metformin treatment or an intensive lifestyle intervention on type 2 diabetes incidence in persons with prediabetes at baseline. The full details of the study are published elsewhere.[6,7] Briefly, a total of 3,234 overweight or obese adults with elevated fasting and postchallenge glucoses were randomized to placebo control, metformin treatment (850 mg twice daily), or intensive lifestyle intervention (primarily fat gram, calorie, and physical activity goals) aimed at ~7% weight loss. The intensive lifestyle intervention included 16 individual sessions within the first 6 months with in-person or phone follow-up at least monthly thereafter. In years 2 and beyond, group classes and campaigns were offered to reinforce lifestyle changes. In the placebo arm, standard lifestyle recommendations (annually) and inactive tablets were given. The trial's primary outcome was the development of diabetes. Numerous other phenotypes, including weight, waist circumference, and abdominal adipose tissue distribution (from CT scans), were measured at baseline and intermittently as the trial progressed; this report focuses on weight change and weight regain in the placebo control and lifestyle arms only. Analyses for a small subset of single nucleotide polymorphisms (SNPs) examined here have been published previously in the DPP.[8]

DPP Participants. Prior to initiating the study protocol, each participant provided written informed consent (88% consented to genetic analyses), and each study center obtained approval from its respective institutional review board. The analytic sample for the weight-change analyses consists of 1,824 comparison and lifestyle arm participants who provided genetic consent, for whom follow-up data were available in years 1–4, and whose genotype data passed quality-control procedures. The sample used for the weight-regain analyses consisted of 834 participants who lost at least 3% of their baseline weight at 1 year and attended at least one follow-up assessment in years 2–4, consistent with prior research in the DPP[8] and Look AHEAD trials.[9,10]

Look AHEAD

Look AHEAD is a 16-site multicenter parallel arm RCT designed to determine whether weight loss achieved through lifestyle change of diet and physical activity reduces cardiovascular disease morbidity and mortality among 5,145 ethnically diverse overweight or obese adults with type 2 diabetes. The full details of the study are published elsewhere.[11–14] Briefly, at baseline participants were randomized to either an intensive lifestyle intervention or diabetes support and education (DSE) arm. Both the lifestyle and DSE groups were provided one session of education on diabetes and cardiovascular risk. In addition, lifestyle participants received an intensive lifestyle program (focused on achieving ~7% weight loss through calorie, fat gram, and physical activity goals) adapted from the DPP intervention. The lifestyle intervention included one individual and three group meetings per month for 6 months, followed by one individual and two group meetings per month through year 1. In years 2–4, lifestyle participants were seen individually at least monthly, contacted another time each month by telephone or e-mail, and offered a variety of group classes and campaigns, as in the DPP. The DSE group received the option of attending three sessions per year on nutrition, physical activity, and social support with no explicit weight-loss goals.

Look AHEAD Participants. Prior to initiating the study protocol, each participant provided written informed consent (84% consented to genetic analyses), and each study center obtained approval from its respective institutional review board. The analytic sample for the weight-change analyses consists of 3,906 DSE and lifestyle arm participants who provided genetic consent, for whom follow-up data were available in years 1–4, and whose genotype data passed quality-control procedures. The sample used for the weight-regain analyses consisted of 2,116 participants who lost at least 3% of their baseline weight at 1 year and attended at least one follow-up assessment in years 2–4, consistent with prior research in the DPP and Look AHEAD.[8–10] Analyses for a small subset of SNPs examined here have been published previously in Look AHEAD.[9,10] Both the DPP and Look AHEAD trials were conducted in accordance with the Declaration of Helsinki.

Neither the active intervention nor comparison arms of the DPP and Look AHEAD are identical. However, the active intervention in both studies represents considerably higher-intensity lifestyle interventions than were administered in the comparison arms. Therefore, for the sake of simplicity, we refer to the nonactive intervention arms of the DPP (placebo control) and Look AHEAD (DSE) as comparison and the active intervention arms as lifestyle from here on.

Genotyping

Ninety-one independent loci, characterized by 93 SNPs, identified or confirmed recently by the GIANT Consortium[5] were genotyped using the MetaboChip genotyping array (Illumina, San Diego, CA) in both studies. To ensure quality control, study participants with failed genotyping, sex inconsistency, or cryptic familial relatedness were excluded. SNPs with within-study genotyping call rates <95% or marked deviation from Hardy-Weinberg equilibrium (P < 1.2 × 10−4) in any ethnic group were also excluded. After quality control, the residual genotyping success rate for the 93 SNPs was >99.2% in the DPP and >99.7% in Look AHEAD (Supplementary Table 1 http://diabetes.diabetesjournals.org/content/64/12/4312/suppl/DC2).

Statistical Analysis

After excluding SNPs in linkage disequilibrium (r2 >0.30), EIGENSTRAT was used to compute principal components from all SNPs on the MetaboChip to control for population stratification in regression analyses. Four primary racial/ethnic groups were distinguished: non-Hispanic white, African American, Hispanic, and Asian.

Longitudinal trajectories of 1) weight change (baseline to 1-, 2-, 3-, and 4-year postrandomization) and 2) weight regain (year 1 to 2-, 3-, and 4-year postrandomization) among participants experiencing ≥3% weight loss from baseline to year 1 were first analyzed separately by study. Three-way interaction models of individual SNP markers (0, 1, and 2 copies of the minor allele; additive model) with study arm (lifestyle vs. comparison) and measurement time were estimated using the generalized least squares capabilities of S-Plus 8.2.[15] Three distinct types of SNP effects are presented, which can be interpreted as the effect per copy of the corresponding minor allele on 1) time-specific weight change within lifestyle, 2) time-specific weight change within the comparison arm, and 3) lifestyle versus comparison arm differences in relation to time-specific weight change. Longitudinal regression models for weight-change and -regain outcomes also included adjustment for age, sex, genetic ancestry (top three principal components), clinic site, and baseline weight (for weight-loss analyses) or year 1 weight (for regain analyses). With the exception of clinic site, all the aforementioned covariates were fully interacted with time, treatment, and time by treatment, so as to allow their effects to vary across study arm and/or time point. Correlation among repeated measures on study participants and variance heteroscedasticity across time points were accommodated using an unstructured covariance matrix whose parameters were estimated via restricted maximum likelihood.

Meta-analysis

Cross-Study Differences in SNP Effects. To determine if study-specific results could be pooled, we tested for DPP versus Look AHEAD (LA) differences in SNP effects using χ2 tests that combined information across all years of follow-up. Let βLA = (βLA,1LA,2LA,3LA,4 and βDPP = (βDPP,1DPP,2DPP,3DPP,4) be the study-specific parameter vectors for a particular type of SNP effect from the weight-change analysis, with elements corresponding to year-specific parameters (years 2–4 only for weight-regain analysis). We tested H 0: Δ = 0 vs. H 0: Δ ≠ 0, where Δ = βLA − βDPP. If ΣLA and ΣDPP represent the covariance matrices of their estimates, and , respectively, then the estimated between-study difference is and the estimated variance of this difference is , as the estimates from the two studies are independent. Using these estimates, we used a approximation of the distribution of the heterogeneity test statistic, , where ν = 4 for weight loss and ν = 3 for weight regain.

Cross-Study Averages of SNP Effects. When no significant between-study heterogeneity was evident, pooled estimates of SNP effects across the DPP and Look AHEAD were obtained using matrix-weighted averages, which weigh study-specific estimates by their relative precision. Individual precision matrices are simply the inverses of the covariance matrices given above. On the basis of matrix-weighted averaging, the pooled estimate of each type of SNP effect was calculated as Alternatively, letting and be the corresponding weight matrices, we see that with estimated variance .

Univariate Tests of Pooled SNP Effects. To test whether the pooled estimate of a particular type of SNP effect was significantly different from zero at a specific time, H 0 : β pooled,j = 0 vs. H 1 : β pooled,j ≠ 0 for j = 1, …, 4 for weight change (or j = 2, …, 4 for weight regain), we used a normal approximation of the distribution of the pooled test statistic, where is the square root of the j th diagonal element of . Identical methodology was used to test the significance of within-study effects.

Multivariate Tests of Pooled SNP Effects. To test whether the pooled estimates of a particular SNP effect were significantly different from zero across all years of follow-up, H 0 : β pooled,j = 0 vs. H 1 : β pooled,j ≠ 0 for j = 1, …, 4 for weight change (or j = 2, …, 4 for weight regain), we used a χ2 approximation of the distribution of the test statistic, , where ν = 4 for weight change and ν = 3 for weight regain. Identical methodology was used to test the significance of within-study effects.

Study-wide Significance Threshold. In addition to the nominal P < 0.05 significance level, a multiplicity-adjusted significance threshold of P < 5.8 × 10−4 was derived via the Li and Ji method,[16] taking into account the effective number of 88 uncorrelated markers under consideration (compared with 93 correlated ones). We note that this significance threshold corrects for the number of independent loci considered but assumes no prior probability of an effect. Hence, this may be a conservative estimate, given that these loci have known effects on BMI.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....