Plant-based dietary patterns and genetic susceptibility to obesity in the CARTaGENE cohort
Abstract
Objective
The study's objective was to examine whether adherence to three plant-based dietary indices (PDIs) mediated or moderated genetic susceptibility to obesity.
Methods
Baseline participants were 7037 adults (57% women, aged 55.6 ± 7.7 years) from the CARTaGENE cohort of Quebec adults. Two polygenic risk scores for BMI (PRS-BMI), 92 single-nucleotide polymorphisms and 2 million single-nucleotide polymorphisms, and three plant-based scores were calculated (overall, healthy, and unhealthy). Follow-up participants were 2258 adults with data on obesity outcomes, measured 6 years later. General linear models were used to examine the relationships between PRSs and PDI scores on obesity outcomes. Causal mediation analyses were conducted to assess mediation and interaction models.
Results
The overall- and healthy-PDIs and PRSs were significantly associated with obesity outcomes. Adherence to PDIs did not mediate or moderate genetic susceptibility to obesity. Associations between PRSs and obesity outcomes were partly mediated by meat intake cross-sectionally and whole grains intake among males both cross-sectionally and longitudinally. Higher meat intake had a positive association with obesity outcomes, whereas higher whole grains intake had an inverse association.
Conclusions
These findings suggest that components of a plant-based diet and a shift away from animal products, specifically meat, might be beneficial for nutrition interventions, particularly among individuals with higher genetic risk of obesity.
Study Importance
What is already known?
- Plant-based dietary patterns are increasingly promoted for their human and planetary health benefits.
- Dietary patterns have been reported to modulate genetic susceptibility to obesity, but evidence is lacking regarding plant-based dietary patterns and genetic risk of obesity.
What does this study add?
- This investigation demonstrated that overall adherence to healthful (but not unhealthful) plant-based dietary patterns was inversely associated with obesity outcomes, but this did not mediate or moderate genetic susceptibility to obesity.
- However, lower meat intake and higher whole grains intake showed favorable associations with obesity outcomes and were observed to mediate genetic susceptibility to obesity.
How might these results change the direction of research or the focus of clinical practice?
- Adherence to healthful plant-based dietary patterns is inversely associated with obesity outcomes irrespective of genetic risk of obesity.
- The consumption of certain food groups (meat and whole grains) partly mediates the genetic susceptibility to obesity. Thus, components of a healthy plant-based diet may be beneficial targets for obesity interventions, especially among individuals with high genetic risk.
INTRODUCTION
The mechanisms by which obesity susceptibility genes exert their influence are not fully understood. Some genes are expressed in the central nervous system [(1)] whereas others are expressed in adipose tissue [(2)]. With the advancement of genomics research, several mediation and interaction studies have analyzed relationships between dietary patterns and obesity genes. Mediation analyses aim to identify how an intermediate variable may explain the effects of an observed association [(3)], whereas interaction studies aim to assess the modifying effects of an intermediate variable on the effects between an exposure on an outcome [(4)]. In other words, mediation studies may be useful for identifying targets for intervention, whereas interaction studies may identify subgroups that are responsive to a particular intervention [(4)]. Prior studies have shown that genetic risk to obesity was mostly mediated by eating behaviors [(5)] although the genetic susceptibility to obesity was also mediated by other factors such as gray matter volume [(6, 7)], early life stress [(8)], leptin levels [(9)], or education, physical activity, conscientiousness, and depressive symptoms [(10)]. Moreover, two recent studies have shown that body mass index (BMI) mediated the association between obesity genes and eating behaviors [(11, 12)].
Despite growing evidence of different lifestyle, demographic, and psychosocial factors as mediators in the obesity genetic pathway, only a small number of studies have investigated diet as a mediator of the genetic susceptibility to obesity. A previous study reported that an eating behavior pattern characterized by a lower diet quality score and skipping meals mediated the genetic susceptibility to obesity in Finnish adults [(13)]. Furthermore, a recent study of Quebec adults found that a poor diet quality score partly mediated the associations between obesity genes and BMI and waist circumference (WC) [(14)]. Thus, prior studies provide supportive evidence for a potential mediation and moderation effect of diet on the genetic susceptibility to obesity, but plant-based dietary patterns in particular have not been comprehensively evaluated. The aim of this investigation was to examine whether adherence to plant-based dietary patterns mediated or moderated the associations between obesity susceptibility genes and anthropometric outcomes in the CARTaGENE cohort. We hypothesized that increased adherence to plant-based dietary patterns would attenuate associations between obesity genetic risk and obesity outcomes, while also partially mediating the genetic risk.
METHODS
Study population
Data for this study were derived from the population-based CARTaGENE cohort (www.cartagene.qc.ca) that aims to investigate the genetic, environmental, and lifestyle factors that contribute to the health of ~40,000 men and women, aged 40 to 69 years old, from Quebec, Canada [(15)]. Participants were randomly selected between August 2009 and October 2010 and December 2012 and February 2015. Participants' data were collected via questionnaires, physical measurements, and biological samples. For this study, we used a subgroup of participants who had provided dietary information through a semiquantitative food frequency questionnaire (FFQ; n = 9696). Participants who did not provide responses to at least half of the FFQ were excluded as shown in Figure S1 (yielding n = 7313). We also excluded participants with no information on anthropometric data or genetic data (yielding n = 7037). Follow-up examinations were conducted 6 years later in a subset of participants who completed a health questionnaire as part of a complimentary study. For the present investigation, n = 2258 participants who had reliable data on BMI and WC from the follow-up, as well as genotype and dietary data from the initial assessment, were included in longitudinal analyses (Figure S1). The study was approved by the CARTaGENE Sample and Data Access Committee and was conducted in accordance with the ethical standards of the Declaration of Helsinki, as approved by the Research Ethics Board from McGill University Faculty of Agriculture and Environmental Sciences. Informed consent was obtained from all study participants prior to their participation.
Dietary assessment and calculation of the plant-based dietary indices
Dietary intake was assessed cross-sectionally in 2012 by using the Canadian adaptation of the US NIH Diet History Questionnaire (C-DHQ II) semiquantitative FFQ. The questionnaire contained 164 food and beverage items and has been validated for use in the Canadian population [(16)]. To evaluate adherence to plant-based dietary patterns, three plant-based dietary indices (PDIs) were constructed in this study, labeled overall, healthy, and unhealthy, following the approach described by Satija et al. [(17)]. A total of 18 food groups were included: whole grains, fruits, vegetables, nuts, legumes, vegetable oils, tea and coffee, fruit juices, refined grains, potatoes, sugar-sweetened beverages, sweets and desserts, animal fat, dairy, eggs, fish or seafood, meat, and miscellaneous. In this study, the food group labeled animal fat included margarine, as the FFQ question asked about the consumption of both butter and margarine together. Food groups, in servings per day, were divided into quintiles and assigned a numerical score ranging from 1 to 5 based on the quintile category. Additional details of the scoring system and the different PDIs (overall, healthy, or unhealthy) can be found in online Supporting Information Methods and Table S1.
Genotyping and polygenic risk scores
Blood and saliva samples were collected for DNA extraction as part of study participation. Genotyping was conducted using the Illumina Infinitum Global Screening Array. Genetic imputation was conducted with Minimac4 software [(18)]. Prior to genetic imputation, a quality control following the procedure of Anderson et al. [(19)] was performed. After imputation, PLINK 2.0 (www.cog-genomics.org/plink/2.0/) was used to exclude variants with low imputation scores (rsq < 0.03) and to calculate genetic principal components.
Two polygenic risk scores (PRSs) for BMI were calculated using pgsc_calc, which is a workflow that matches genetic variants between genotype data and existing PRSs published in the open database PGS Catalog [(20)]. The PRS for main analyses was a genome-wide PRS that incorporated all available single-nucleotide polymorphisms (SNPs) irrespective of their genome-wide significance. It was constructed as proposed by Khera et al. [(21)], had more than 2 million SNPs, and explained 10% of BMI variance in this study sample (PRS-Khera). For sensitivity analyses we used a more conservative PRS based on genome-wide significant loci as proposed by Locke et al. [(22, 23)], which matched with 92 SNPs and explained 3% of BMI variance in this study sample (PRS-Locke). Emerging evidence has suggested that the loci identified by Locke et al. interact with dietary factors on obesity outcomes [(24, 25)]. More information on the number of SNPs included in PRSs and density plots can be found in Table S2 and Figure S3, respectively.
Assessment of obesity outcomes and covariates
At baseline, participants' weight (kilograms), height (centimeters), WC (centimeters), and body fat percentage (%) were directly measured using noninvasive procedures at study assessment centers. Body fat percentage was measured with bioelectrical impedance. BMI was calculated as weight in kilograms divided by height in squared meters (kg/m2). At follow-up, participants self-reported their weight and self-measured their WC twice. The average of the two measurements was used in analyses. Questionnaires were used to obtain self-reported sociodemographic and lifestyle data for age, sex, ethnicity, alcohol intake, smoking, education, sleeping time, income, and anxiety, as previously described [(15)]. Physical activity and energy misreporter status was calculated as described previously [(26, 27)] (more information in online Supporting Information).
Statistical analyses
Statistical analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, North Carolina) and R-Studio (www.r-project.org). Missing data on covariates (missing for ~12% of participants) were imputed using chained random forest imputation, assuming values were missing at random [(28)]. Imputation was performed using the package missRanger (www.CRAN.R-project.org/package=missRanger). The imputation included 200 trees and controlled for a variety of additional predictors.
Non-normally distributed variables were log-transformed prior to analyses. Participant characteristics were reported as means and SDs for continuous variables and as numbers and frequencies for categorical variables. Differences between males and females were compared using the t test for continuous variables and Pearson χ2 for categorical variables. Differences across PRS-Khera quintiles were determined using p values for trend. General linear models (GLMs) were used to assess the main effect associations between PRSs and BMI, body fat percentage, and WC. Main effect associations between PRSs and PDIs (and their individual food groups) were also assessed. All exposures and outcomes were standardized to facilitate interpretation, and sex-stratified analyses were conducted to determine whether patterns varied between males and females. All models were adjusted for age, sex, principal components of ancestry (or ethnicity where appropriate), income, education, alcohol intake, smoking, physical activity, sleeping time, anxiety, energy intake, and energy misreporter status.
Causal mediation analyses (CMA) examined whether PDIs or their food group items mediated or moderated the association between PRS-Khera and BMI, body fat, and WC (Figure S2), based on the counterfactual approach from VanderWeele [(29)]. Significant models were also tested using a longitudinal approach where PRS-Khera was the exposure, PDIs or food groups at baseline were the mediators/moderators, and BMI and WC measured 6 years later were the outcomes. We tested two longitudinal models: Model 1, which adjusted for all the above-mentioned covariates, and Model 2, which included the baseline outcome measure (BMI or WC) as a covariate. We used the CAUSALMED Procedure that allows for a 4-way decomposition of the total effect (the sum of the direct and indirect effects) that unifies methods to attribute effects to interactions and methods to assess mediation. The 4-way decomposition decomposes the total controlled direct effect (the effect without the mediator/moderators' influence), pure indirect effect (mediation effect), reference interaction (interaction effect), and mediated interaction (effect due to mediation and interaction). Sample bias correction estimates and 95% confidence intervals (CIs) were calculated using the bootstrapping approach of 5000 draws.
Sensitivity analyses were performed using two approaches. First, the same CMA models were repeated using PRS-Locke as the exposure, which consisted of only genome-wide association study significant loci. Second, we assessed interaction effects with a more traditional method, a GLM with an interaction term added to the model, because CMA were mainly focused to evaluate mediation. To control the proportion of false positives due to multiple testing, a false discovery rate (FDR) correction was applied at q = 0.10. Results with an FDR-adjusted p value < 0.10 (unadjusted p value < 0.02) were considered statistically significant.
RESULTS
General characteristics of the study sample
Table 1 summarizes the general characteristics of the study sample. More than half of the study population were females, and mean BMI, body fat percentage, and WC at baseline were 26.7 ± 5.2 kg/m2, 31.1% ± 8.4%, and 92.8 ± 13.9 cm, respectively. Females had a greater adherence to the healthy-PDI (55.7 ± 7.9 points females vs. 51.9 ± 8.2 points males) and males had a greater adherence to the unhealthy-PDI (52.7 ± 8.2 points females vs. 55.3 ± 8.1 points males). General characteristics presented by quintiles of PRS-Khera are shown in Table S3. Compared with the lowest PRS-Khera quintile, participants at the highest quintile had higher BMI, body fat, and WC and a higher adherence to the healthy-PDI.
| Characteristics | Overall sample, n = 7037 | Female, n = 3981 | Male, n = 3056 | p value |
|---|---|---|---|---|
| Age, mean (SD) (y) | 55.6 (7.7) | 55.3 (7.7) | 56.0 (7.8) | 0.0002 |
| WC, mean (SD) (cm) | 92.8 (13.9) | 87.8 (13.0) | 99.3 (12.4) | <0.0001 |
| WC follow-up, mean (SD)b (cm) | 92.7 (13.9) | 90.1 (41.5) | 95.8 (14.0) | <0.0001 |
| BMI, mean (SD)a (kg/m2) | 26.7 (5.2) | 26.7 (5.2) | 28.0 (4.5) | <0.0001 |
| BMI follow-up, mean (SD)b (kg/m2) | 26.2 (4.4) | 25.4 (4.5) | 27.2 (4.1) | <0.001 |
| Body fat percentage, mean (SD)a | 31.1 (8.4) | 35.3 (7.2) | 25.4 (6.3) | <0.0001 |
| Khera PRS, mean (SD) | 38.8 (0.2) | 38.8 (0.2) | 38.8 (0.2) | 0.005 |
| Locke PRS, mean (SD) | 11.4 (0.8) | 11.4 (0.8) | 11.4 (0.8) | 0.8 |
| Overall-PDI, mean (SD) | 53.9 (5.9) | 54.0 (5.8) | 53.8 (6.1) | 0.3 |
| Healthy-PDI, mean (SD) | 54.1 (8.2) | 55.7 (7.9) | 51.9 (8.2) | <0.0001 |
| Unhealthy-PDI, mean (SD) | 53.8 (8.3) | 52.7 (8.2) | 55.3 (8.1) | <0.0001 |
| Energy intake, mean (SD) (kcal) | 2008.1 (892.6) | 1823.1 (758.8) | 2249.0 (991.0) | <0.0001 |
| Ethnicity, n (%) | ||||
| White | 6251 (92.7) | 3741 (94.0) | 2780 (91.0) | <0.0001 |
| Non-White | 414 (5.9) | 189 (4.7) | 225 (7.3) | |
| Unsure | 102 (1.5) | 51 (1.3) | 51 (1.7) | |
| Alcohol, n (%) | ||||
| Monthly | 2429 (34.5) | 1526 (38.3) | 903 (29.6) | <0.0001 |
| Weekly | 2555 (36.3) | 1452 (36.5) | 1103 (36.1) | |
| Daily | 1741 (24.7) | 785 (19.7) | 956 (31.3) | |
| Missing | 312 (4.4) | 218 (5.5) | 94 (3.1) | |
| Income, n (%) | ||||
| Low | 2181 (31.0) | 1347 (33.8) | 834 (27.3) | <0.0001 |
| Low-medium | 2621 (37.2) | 1432 (36.0) | 1189 (38.9) | |
| Medium-high | 1627 (23.1) | 813 (20.4) | 814 (26.6) | |
| High | 233 (3.3) | 128 (3.2) | 105 (3.4) | |
| Missing | 375 (5.3) | 261 (6.6) | 114 (3.7) | |
| Smoking, n (%) | ||||
| Never | 2921 (41.5) | 1766 (44.4) | 1155 (37.8) | <0.0001 |
| Past | 3035 (43.1) | 1622 (40.7) | 1413 (46.2) | |
| Occasional | 268 (3.8) | 142 (3.6) | 126 (4.1) | |
| Daily | 787 (11.2) | 434 (10.9) | 353 (11.6) | |
| Missing | 26 (0.4) | 17 (4.3) | 9 (0.3) | |
| Education, n (%) | ||||
| High school or less | 1650 (23.4) | 958 (24.1) | 692 (22.6) | 0.01 |
| College | 2260 (32.1) | 1315 (33.0) | 945 (30.9) | |
| University or higher | 3099 (44.0) | 1691 (42.5) | 1408 (46.1) | |
| Missing | 28 (0.4) | 17 (0.4) | 11 (0.4) | |
| Sleep, n (%) | ||||
| <6 h | 1352 (19.2) | 726 (18.2) | 626 (20.5) | <0.001 |
| 7–8 h | 4921 (69.9) | 2774 (69.7) | 2147 (70.3) | |
| <9 h | 730 (10.4) | 462 (11.6) | 268 (8.8) | |
| Missing | 34 (0.5) | 19 (0.5) | 15 (9.7) | |
| Anxiety, n (%) | ||||
| Never | 6156 (87.5) | 3408 (85.6) | 2748 (89.9) | <0.0001 |
| Several days | 633 (9.0) | 404 (10.1) | 229 (7.5) | |
| More than half of the days | 155 (2.2) | 107 (2.7) | 48 (1.6) | |
| Almost every day | 70 (1.0) | 48 (1.2) | 22 (0.7) | |
| Missing | 23 (0.3) | 165 (4.1) | 9 (0.3) | |
| Energy reporter status, n (%) | ||||
| Underreporter | 2017 (28.7) | 920 (23.1) | 1097 (35.9) | <0.0001 |
| Plausible reporter | 4251 (60.4) | 2522 (63.4) | 1729 (56.6) | |
| Overreporter | 597 (8.5) | 374 (9.4) | 223 (7.3) | |
| Missing | 172 (2.4) | 165 (4.1) | 7 (0.2) | |
| Physical activity level, n (%) | ||||
| Sedentary | 4496 (63.9) | 2628 (66.0) | 1868 (61.1) | <0.0001 |
| Low active | 1110 (15.8) | 605 (15.2) | 505 (16.5) | |
| Active | 757 (10.8) | 376 (9.4) | 381 (14.5) | |
| Very active | 503 (11.2) | 207 (5.2) | 296 (9.7) | |
| Missing | 171 (2.4) | 165 (4.1) | 6 (0.2) | |
- Note: p values were determined by t test for continuous variables and Pearson χ2 for categorical variables.
- Abbreviations: PDI, plant-based dietary index; PRS, polygenic risk score; WC, waist circumference.
- a Lower sample size due to missing values, BMI: n = 7022 for the overall sample, n = 3972 for females, n = 3056 for males; body fat percentage: n = 6651 for the overall sample, n = 3817 for females, n = 2834 for males.
- b Participants at follow-up: n = 2258 for the overall sample, n = 1249 for females, n = 1009 for males.
Main effect associations between PDIs, food groups, and obesity outcomes
A higher overall-PDI was inversely associated with BMI, body fat, and WC (β = −0.09, per one unit increase in overall-PDI, for BMI and WC; β = −0.07, per one unit increase in overall-PDI, for body fat; Table 2). The healthy-PDI was also significantly associated with obesity outcomes (β = −0.09, per one unit increase in healthy-PDI, for BMI; β = −0.08, per one unit increase in healthy-PDI, for body fat; β = −0.11, per one unit increase in healthy-PDI, for WC). When stratifying analyses by sex, main effect associations were slightly higher among females than males, although 95% CIs were overlapping. The unhealthy-PDI was not significantly associated with obesity outcomes, among either males or females.
| Sample | WC, n = 7037 | BMI, n = 7022 | Body fat percentage, n = 6651 | ||||||
|---|---|---|---|---|---|---|---|---|---|
| β (95% CI) | p value | FDR-adjusted p value | β (95% CI) | p value | FDR-adjusted p value | β (95% CI) | p value | FDR-adjusted p value | |
| Overall sample | |||||||||
| Overall-PDI | −0.09 (−0.11 to −0.07) | <0.0001 | 0.003 | −0.09 (−0.11 to −0.06) | <0.0001 | 0.002 | −0.07 (−0.09 to −0.05) | <0.0001 | 0.002 |
| Healthy-PDI | −0.11 (−0.13 to −0.09) | <0.0001 | 0.003 | −0.09 (−0.11 to −0.06) | <0.0001 | 0.002 | −0.08 (−0.10 to −0.06) | <0.0001 | 0.002 |
| Unhealthy-PDI | 0.02 (0.00 to 0.04) | 0.07 | 0.3 | −0.01 (−0.03 to 0.02) | 0.5 | 0.8 | 0.00 (−0.02 to 0.03) | 0.7 | 0.9 |
| Females | |||||||||
| Overall-PDI | −0.10 (−0.13 to −0.07) | <0.0001 | 0.002 | −0.10 (−0.14 to −0.07) | <0.0001 | 0.002 | −0.08 (−0.10 to −0.05) | <0.0001 | 0.002 |
| Healthy-PDI | −0.13 (−0.16 to −0.10) | <0.0001 | 0.002 | −0.11 (−0.15 to −0.08) | <0.0001 | 0.002 | −0.09 (−0.12 to −0.06) | <0.0001 | 0.001 |
| Unhealthy-PDI | 0.02 (−0.01 to 0.05) | 0.2 | 0.5 | −0.01 (−0.05 to 0.02) | 0.5 | 0.8 | −0.01 (−0.04 to 0.02) | 0.7 | 0.9 |
| Males | |||||||||
| Overall-PDI | −0.09 (−0.12 to −0.06) | <0.0001 | 0.001 | −0.08 (−0.11 to −0.05) | <0.0001 | 0.001 | −0.07 (−0.10 to −0.04) | <0.0001 | 0.001 |
| Healthy-PDI | −0.08 (−0.11 to −0.05) | <0.0001 | 0.001 | −0.05 (−0.08 to −0.01) | 0.005 | 0.04 | −0.06 (−0.09 to −0.03) | <0.0001 | 0.001 |
| Unhealthy-PDI | 0.03 (0.00 to 0.06) | 0.07 | 0.3 | 0.00 (−0.03 to 0.04) | 0.8 | 0.9 | 0.03 (0.00 to 0.06) | 0.08 | 0.3 |
- Note: Data are β-coefficients per one unit increase in PDIs, from general linear models with respective 95% CIs. All models were adjusted for: age, sex, ethnicity, alcohol intake, smoking, education, sleeping time, income, anxiety, physical activity, energy intake, and energy misreporter status.
- Abbreviations: FDR, false discovery rate; PDI, plant-based dietary index; WC, waist circumference.
Obesity outcomes were associated with most of the PDI food items (Table S4). The strongest association was observed for meat intake (β = 0.17, 95% CI: 0.14–0.19 for BMI; β = 0.15, 95% CI: 0.13–0.17 for body fat; and β = 0.16, 95% CI: 0.14–0.18 for WC). Similar results were observed among males and females (data not shown).
Main effect associations between PRSs, PDIs, and obesity outcomes
At baseline, obesity outcomes were significantly associated with PRS-Khera (Table 3). The association between PRS-Khera and BMI was slightly higher (β = 0.29, 95% CI: 0.27–0.32) than the associations between PRS-Khera and body fat (β = 0.21, 95% CI: 0.19–0.23) and WC (β = 0.23, 95% CI: 0.21–0.25). The obesity outcomes were also associated with PRS-Locke, although associations were weaker. Similar results were observed at follow-up 6 years later in Model 1 (Table S5), but the association between WC and PRS-Khera was weaker (β = 0.14, 95% CI: 0.09–0.18). When adjusting for the baseline outcome measure in longitudinal Model 2, several associations between obesity outcomes and PRSs were attenuated. Analyses by sex described similar associations between obesity outcomes and PRSs among males and females (Table 3 and Table S5).
| Sample | n | PRS-Khera | PRS-Locke | ||||
|---|---|---|---|---|---|---|---|
| β (95% CI) | p value | FDR-adjusted p value | β (95% CI) | p value | FDR-adjusted p value | ||
| Overall sample | |||||||
| WC | 7037 | 0.23 (0.21 to 0.25) | <0.0001 | 0.002 | 0.06 (0.04 to 0.08) | <0.0001 | 0.002 |
| BMI | 7022 | 0.29 (0.27 to 0.32) | <0.0001 | 0.002 | 0.08 (0.06 to 0.11) | <0.0001 | 0.002 |
| Body fat percentage | 6651 | 0.21 (0.19 to 0.23) | <0.0001 | 0.002 | 0.05 (0.04 to 0.07) | <0.0001 | 0.002 |
| Overall-PDI | 7037 | −0.02 (−0.04 to 0.01) | 0.2 | 0.5 | 0.00 (−0.02 to 0.02) | 0.9 | <1.0 |
| Healthy-PDI | 7037 | 0.01 (−0.01 to 0.04) | 0.2 | 0.5 | 0.01 (−0.01 to 0.03) | 0.5 | 0.769 |
| Unhealthy-PDI | 7037 | −0.03 (−0.05 to 0.00) | 0.03 | 0.16 | −0.03 (−0.05 to −0.01) | 0.007 | 0.047 |
| Females | |||||||
| WC | 3981 | 0.22 (0.19 to 0.25) | <0.0001 | 0.001 | 0.05 (0.02 to 0.08) | <0.001 | 0.008 |
| BMI | 3972 | 0.32 (0.28 to 0.35) | <0.0001 | 0.001 | 0.09 (0.05 to 0.12) | <0.0001 | 0.001 |
| Body fat percentage | 3817 | 0.22 (0.19 to 0.25) | <0.0001 | 0.001 | 0.06 (0.04 to 0.09) | <0.0001 | 0.001 |
| Overall-PDI | 3981 | −0.02 (−0.05 to 0.01) | 0.2 | 0.5 | 0.00 (−0.03 to 0.03) | 0.9 | <1.0 |
| Healthy-PDI | 3981 | 0.03 (0.00 to 0.06) | 0.08 | 0.3 | 0.01 (−0.01 to 0.04) | 0.3 | 0.6 |
| Unhealthy-PDI | 3981 | −0.03 (−0.06 to 0.00) | 0.05 | 0.2 | −0.04 (−0.07 to −0.02) | 0.002 | 0.02 |
| Males | |||||||
| WC | 3056 | 0.24 (0.20 to 0.27) | <0.0001 | 0.001 | 0.07 (0.04 to 0.10) | <0.0001 | 0.001 |
| BMI | 3050 | 0.27 (0.23 to 0.30) | <0.0001 | 0.001 | 0.08 (0.05 to 0.11) | <0.0001 | 0.001 |
| Body fat percentage | 2834 | 0.19 (0.16 to 0.22) | <0.0001 | 0.001 | 0.04 (0.02 to 0.07) | 0.001 | 0.009 |
| Overall-PDI | 3056 | −0.02 (−0.06 to 0.02) | 0.4 | 0.7 | 0.01 (−0.03 to 0.04) | 0.8 | 0.9 |
| Healthy-PDI | 3056 | 0.00 (−0.04 to 0.03) | 0.8 | 0.9 | 0.00 (−0.03 to 0.03) | 0.9 | <1.0 |
| Unhealthy-PDI | 3056 | −0.02 (−0.05 to 0.02) | 0.4 | 0.7 | −0.01 (−0.04 to 0.02) | 0.6 | 0.8 |
- Note: Data are β-coefficients per one unit increase in PRSs, from general linear models with respective 95% CIs. All models were adjusted for the following: age, sex, principal components of ancestry, alcohol intake, smoking, education, sleeping time, income, anxiety, physical activity, energy intake, and energy misreporter status.
- Abbreviations: FDR, false discovery rate; PDI, plant-based dietary index; PRS, polygenic risk score; WC, waist circumference.
Associations between PDIs and PRSs are presented in Table 3, showing an inverse association between PRS-Locke and unhealthy-PDI in the overall sample (β = −0.03, 95% CI: −0.05 to −0.01). When stratifying by sex, this association only remained significant among females (β = −0.04, 95% CI: −0.07 to −0.02).
Results from CMAs
Mediation was not observed for any of the three PDIs in the genetic susceptibility to obesity (Tables 4–6). Within food groups, CMA showed that a higher intake of meat mediated the association between PRS-Khera and body fat percentage (mediation effect 2.26%, p = 0.01, q = 0.07). Similar patterns were observed for meat as a mediator of the association between PRS-Khera and WC as well as BMI, although these results were attenuated upon FDR correction. When stratifying analyses by sex, mediation was observed by meat intake among females in the association between PRS-Khera and BMI (mediation effect 2.54%, p = 0.02, q = 0.10) and WC (mediation effect 3.31%, p = 0.02, q = 0.09), and borderline significant mediation was observed for body fat percentage (mediation effect 3.08%, p = 0.02, q = 0.11). Among males, whole grains mediated the association between PRS-Khera and BMI (mediation effect 2.06%, p = 0.007, q = 0.05), body fat percentage (mediation effect 3.19%, p = 0.006, q = 0.04) and WC (mediation effect 2.74%, p = 0.006, q = 0.04). Results from CMA did not indicate moderation for any of the PDIs or food groups in the association between PRS-Khera and BMI, WC, or body fat percentage.
| Mediator | Total effect | Controlled direct effect | Pure indirect effect | Reference interaction | Mediated interaction | Percentage mediated |
|---|---|---|---|---|---|---|
| Overall-PDI | 1.343 (1.469 to 1.225)*** | 1.336 (1.220 to 1.447)*** | 0.007 (−0.003 to 0.018) | 0.000 (−0.001 to 0.001) | 0.001 (−0.008 to 0.015) | - |
| Healthy-PDI | 1.349 (1.232 to 1.461)*** | 1.351 (1.237 to 1.463)*** | −0.007 (−0.019 to 0.003) | 0.000 (−0.002 to 0.002) | 0.004 (−0.002 to 0.024) | - |
| Unhealthy-PDI | 1.359 (1.240 to 1.473)*** | 1.345 (1.229 to 1.458)*** | 0.000 (−0.004 to 0.003) | 0.000 (−0.003 to 0.003) | 0.014 (0.001 to 0.042) | - |
| Whole grains | 1.344 (1.221 to 1.458)*** | 1.339 (1.217 to 1.451)*** | 0.006 (0.000 to 0.015) | 0.000 (−0.001 to 0.001) | −0.001 (−0.018 to 0.009) | - |
| Fruits | 1.346 (1.230 to 1.462)*** | 1.344 (1.230 to 1.459)*** | −0.002 (−0.008 to 0.001) | 0.000 (−0.002 to 0.002) | 0.005 (−0.002 to 0.026) | - |
| Vegetables | 1.355 (1.239 to 1.471)*** | 1.343 (1.226 to 1.457)*** | 0.000 (−0.003 to 0.005) | 0.000 (−0.002 to 0.002) | 0.012 (0.000 to 0.039) | - |
| Nuts | 1.369 (1.253 to 1.488)*** | 1.366 (1.251 to 1.483)*** | 0.002 (−0.005 to 0.010) | 0.000 (−0.001 to 0.001) | 0.001 (−0.004 to 0.013) | - |
| Legumes | 1.338 (1.220 to 1.456)*** | 1.337 (1.218 to 1.455)*** | −0.001 (−0.006 to 0.004) | 0.000 (−0.003 to 0.003) | 0.002 (−0.008 to 0.019) | - |
| Vegetable oils | 1.331 (1.209 to 1.443)*** | 1.329 (1.209 to 1.442)*** | 0.000 (−0.004 to 0.001) | 0.000 (−0.001 to 0.001) | 0.002 (−0.004 to 0.020) | - |
| Tea and coffee | 1.356 (1.233 to 1.470)*** | 1.343 (1.220 to 1.456)*** | 0.004 (−0.001 to 0.011) | 0.000 (−0.002 to 0.002) | 0.009 (−0.002 to 0.037) | - |
| Fruit juices | 1.359 (1.236 to 1.482)*** | 1.346 (1.225 to 1.464)*** | 0.003 (−0.001 to 0.010) | 0.000 (−0.002 to 0.002) | 0.010 (−0.011 to 0.040) | - |
| Refined grains | 1.337 (1.216 to 1.450)*** | 1.340 (1.220 to 1.451)*** | 0.001 (−0.001 to 0.006) | 0.000 (−0.001 to 0.002) | −0.005 (−0.024 to 0.002) | - |
| Potatoes | 1.346 (1.225 to 1.461)*** | 1.347 (1.230 to 1.463)*** | −0.007 (−0.017 to 0.002) | 0.000 (−0.002 to 0.002) | 0.006 (−0.002 to 0.027) | - |
| Sugar-sweetened beverages | 1.395 (1.263 to 1.527)*** | 1.388 (1.254 to 1.518)*** | −0.005 (−0.013 to 0.000) | 0.000 (−0.002 to 0.003) | 0.011 (−0.001 to 0.044) | - |
| Sweets and desserts | 1.342 (1.229 to 1.456)*** | 1.343 (1.229 to 1.456)*** | 0.000 (−0.005 to 0.003) | 0.000 (−0.002 to 0.001) | 0.000 (−0.005 to 0.012) | - |
| Animal fat | 1.372 (1.252 to 1.487)*** | 1.368 (1.248 to 1.481)*** | −0.002 (−0.008 to 0.000) | 0.000 (−0.002 to 0.002) | 0.005 (−0.003 to 0.026) | - |
| Dairy | 1.344 (1.233 to 1.464)*** | 1.334 (1.222 to 1.452)*** | 0.002 (−0.001 to 0.007) | 0.000 (−0.003 to 0.003) | 0.008 (−0.003 to 0.030) | - |
| Eggs | 1.348 (1.230 to 1.474)*** | 1.338 (1.129 to 1.460)*** | 0.006 (−0.001 to 0.015) | 0.000 (−0.002 to 0.002) | 0.004 (−0.005 to 0.026) | - |
| Fish or seafood | 1.365 (1.246 to 1.478)*** | 1.366 (1.247 to 1.479)*** | 0.000 (−0.003 to 0.001) | 0.000 (−0.001 to 0.001) | 0.000 (−0.008 to 0.006) | - |
| Meat | 1.340 (1.221 to 1.458)*** | 1.316 (1.198 to 1.429)*** | 0.020 (0.003 to 0.039) | 0.000 (−0.001 to 0.001) | 0.004 (−0.008 to 0.023) | 1.52 (0.26 to 2.92) |
| Miscellaneous | 1.341 (1.223 to 1.453)*** | 1.341 (1.223 to 1.453)*** | 0.000 (−0.006 to 0.007) | 0.000 (−0.001 to 0.001) | 0.000 (−0.009 to 0.006) | - |
- Note: Results are presented as β-coefficients and their respective 95% CIs. All models were adjusted for: age, sex, principal components of ancestry, alcohol intake, smoking, education, sleeping time, income, anxiety, physical activity, energy intake, and energy misreporter status.
- Abbreviations: PDI, plant-based dietary index; PRS, polygenic risk score.
- *** FDR-corrected p < 0.01.
| Mediator | Total effect | Controlled direct effect | Pure indirect effect | Reference interaction | Mediated interaction | Percentage mediated |
|---|---|---|---|---|---|---|
| Overall-PDI | 1.034 (0.930 to 1.151)*** | 1.025 (0.921 to 1.140)*** | 0.008 (−0.003 to 0.019) | 0.000 (−0.001 to 0.001) | 0.002 (−0.004 to 0.018) | - |
| Healthy-PDI | 1.035 (0.935 to 1.151)*** | 1.042 (0.939 to 1.155)*** | −0.008 (−0.021 to 0.005) | 0.000 (−0.002 to 0.002) | 0.003 (−0.002 to 0.021) | - |
| Unhealthy-PDI | 1.041 (0.938 to 1.158)*** | 1.037 (0.932 to 1.153)*** | −0.004 (−0.010 to 0.000) | 0.000 (−0.002 to 0.003) | 0.008 (−0.002 to 0.029) | - |
| Whole grains | 1.036 (0.925 to 1.143)*** | 1.028 (0.917 to 1.136)*** | 0.007 (0.000 to 0.016) | 0.000 (−0.001 to 0.001) | 0.001 (−0.009 to 0.015) | - |
| Fruits | 1.034 (0.921 to 1.142)*** | 1.035 (0.921 to 1.143)*** | −0.004 (−0.012 to 0.003) | 0.000 (−0.002 to 0.002) | 0.003 (−0.002 to 0.023) | - |
| Vegetables | 1.039 (0.934 to 1.157)*** | 1.036 (0.931 to 1.151)*** | −0.003 (−0.009 to 0.000) | 0.000 (−0.001 to 0.002) | 0.006 (−0.004 to 0.029) | - |
| Nuts | 1.048 (0.935 to 1.160)*** | 1.045 (0.932 to 1.157)*** | 0.003 (−0.005 to 0.011) | 0.000 (−0.001 to 0.001) | 0.001 (−0.004 to 0.013) | - |
| Legumes | 1.029 (0.921 to 1.138)*** | 1.029 (0.920 to 1.137)*** | −0.001 (−0.007 to 0.004) | 0.000 (−0.002 to 0.002) | 0.001 (−0.005 to 0.017) | - |
| Vegetable oils | 1.008 (0.895 to 1.112)*** | 1.009 (0.898 to 1.114)*** | −0.001 (−0.005 to 0.000) | 0.000 (−0.001 to 0.001) | 0.000 (−0.013 to 0.008) | - |
| Tea and coffee | 1.036 (0.925 to 1.148)*** | 1.030 (0.919 to 1.140)*** | 0.001 (−0.004 to 0.006) | 0.000 (−0.001 to 0.001) | 0.005 (−0.014 to 0.029) | - |
| Fruit juices | 1.053 (0.943 to 1.165)*** | 1.037 (0.928 to 1.147)*** | 0.003 (−0.002 to 0.009) | 0.000 (−0.002 to 0.002) | 0.013 (−0.005 to 0.044) | - |
| Refined grains | 1.027 (0.917 to 1.135)*** | 1.030 (0.922 to 1.137)*** | 0.003 (−0.002 to 0.009) | 0.000 (−0.002 to 0.002) | −0.006 (−0.024 to 0.003) | - |
| Potatoes | 1.035 (0.930 to 1.143)*** | 1.036 (0.930 to 1.144)*** | −0.007 (−0.018 to 0.003) | 0.000 (−0.002 to 0.002) | 0.006 (−0.001 to 0.027) | - |
| Sugar-sweetened beverages | 1.067 (0.946 to 1.193)*** | 1.064 (0.942 to 1.189)*** | −0.004 (−0.012 to 0.000) | 0.000 (−0.002 to 0.002) | 0.008 (−0.003 to 0.037) | - |
| Sweets and desserts | 1.032 (0.923 to 1.127)*** | 1.032 (0.924 to 1.139)*** | 0.000 (−0.003 to 0.001) | 0.000 (−0.001 to 0.001) | 0.000 (−0.004 to 0.011) | - |
| Animal fat | 1.048 (0.939 to 1.156)*** | 1.044 (0.935 to 1.151)*** | −0.003 (−0.009 to 0.000) | 0.000 (−0.002 to 0.002) | 0.007 (−0.001 to 0.027) | - |
| Dairy | 1.032 (0.923 to 1.147)*** | 1.028 (0.918 to 1.141)*** | 0.002 (−0.001 to 0.007) | 0.000 (−0.001 to 0.001) | 0.003 (−0.002 to 0.019) | - |
| Eggs | 1.033 (0.924 to 1.151)*** | 1.022 (0.913 to 1.136)*** | 0.005 (0.000 to 0.014) | 0.000 (−0.003 to 0.003) | 0.006 (−0.003 to 0.028) | - |
| Fish or seafood | 1.054 (0.937 to 1.159)*** | 1.054 (0.937 to 1.159)*** | 0.000 (−0.002 to 0.003) | 0.000 (−0.001 to 0.001) | 0.000 (−0.006 to 0.006) | - |
| Meat | 1.030 (0.917 to 1.138)*** | 1.006 (0.898 to 1.111)*** | 0.021 (0.004 to 0.040) | 0.000 (−0.001 to 0.001) | 0.003 (−0.008 to 0.021) | 2.02 (0.36 to 2.38) |
| Miscellaneous | 1.032 (0.921 to 1.140)*** | 1.032 (0.923 to 1.140)*** | 0.000 (−0.007 to 0.007) | 0.000 (−0.002 to 0.002) | 0.000 (−0.012 to 0.010) | - |
- Note: Results are presented as β-coefficients and their respective 95% CIs. All models were adjusted for the following: age, sex, principal components of ancestry, alcohol intake, smoking, education, sleeping time, income, anxiety, physical activity, energy intake, and energy misreporter status.
- Abbreviations: PDI, plant-based dietary index; PRS, polygenic risk score.
- *** FDR-corrected p < 0.01.
| Mediator | Total effect | Controlled direct effect | Pure indirect effect | Reference interaction | Mediated interaction | Percentage mediated |
|---|---|---|---|---|---|---|
| Overall-PDI | 0.931 (0.834 to 1.029)*** | 0.926 (0.829 to 1.023)*** | 0.005 (−0.003 to 0.015) | 0.000 (−0.001 to 0.001) | 0.000 (−0.011 to 0.007) | - |
| Healthy-PDI | 0.937 (0.839 to 1.034)*** | 0.940 (0.844 to 1.037)*** | −0.008 (−0.019 to 0.002) | 0.000 (−0.002 to 0.002) | 0.005 (−0.002 to 0.023) | - |
| Unhealthy-PDI | 0.939 (9.841 to 1.036)*** | 0.934 (0.836 to 1.031)*** | −0.002 (−0.007 to 0.001) | 0.000 (−0.001 to 0.002) | 0.007 (−0.004 to 0.027) | - |
| Whole grains | 0.932 (0.830 to 1.032)*** | 0.927 (0.828 to 1.028)*** | 0.006 (−0.001 to 0.014) | 0.000 (−0.001 to 0.001) | −0.001 (−0.017 to 0.007) | - |
| Fruits | 0.932 (0.834 to 1.035)*** | 0.934 (0.836 to 1.037)*** | −0.004 (−0.011 to 0.001) | 0.000 (−0.003 to 0.003) | 0.003 (−0.003 to 0.019) | - |
| Vegetables | 0.933 (0.833 to 1.030)*** | 0.932 (0.834 to 1.027)*** | 0.000 (−0.004 to 0.003) | 0.000 (−0.001 to 0.001) | 0.001 (−0.011 to 0.017) | - |
| Nuts | 0.937 (0.833 to 1.036)*** | 0.936 (0.833 to 1.034)*** | 0.001 (−0.005 to 0.008) | 0.000 (−0.001 to 0.001) | −0.001 (−0.012 to 0.003) | - |
| Legumes | 0.921 (0.823 to 1.020)*** | 0.921 (0.822 to 1.020)*** | 0.000 (−0.005 to 0.003) | 0.000 (−0.002 to 0.002) | 0.001 (−0.008 to 0.014) | - |
| Vegetable oils | 0.921 (0.819 to 1.020)*** | 0.919 (0.818 to 1.019)*** | 0.000 (−0.002 to 0.003) | 0.000 (−0.001 to 0.001) | 0.002 (−0.002 to 0.003) | - |
| Tea and coffee | 0.930 (0.833 to 1.026)*** | 0.920 (0.823 to 1.016)*** | 0.003 (−0.001 to 0.010) | 0.000 (−0.001 to 0.001) | 0.007 (−0.011 to 0.034) | - |
| Fruit juices | 0.945 (0.837 to 1.048)*** | 0.934 (0.829 to 1.034)*** | 0.003 (−0.001 to 0.008) | 0.000 (−0.002 to 0.002) | 0.008 (−0.010 to 0.034) | - |
| Refined grains | 0.925 (0.830 to 1.024)*** | 0.927 (0.832 to 1.026)*** | 0.001 (−0.001 to 0.005) | 0.000 (−0.002 to 0.002) | −0.003 (−0.018 to 0.002) | - |
| Potatoes | 0.937 (0.837 to 1.039)*** | 0.939 (0.839 to 1.040)*** | −0.008 (−0.018 to 0.001) | 0.000 (−0.002 to 0.002) | 0.006 (−0.001 to 0.024) | - |
| Sugar-sweetened beverages | 0.977 (0.867 to 1.088)*** | 0.975 (0.862 to 1.084)*** | −0.005 (−0.013 to 0.000) | 0.000 (−0.002 to 0.002) | 0.007 (−0.003 to 0.032) | - |
| Sweets and desserts | 0.931 (0.832 to 1.033)*** | 0.931 (0.832 to 1.033)*** | 0.000 (−0.002 to 0.002) | 0.000 (−0.001 to 0.001) | 0.000 (−0.008 to 0.006) | - |
| Animal fat | 0.962 (0.862 to 1.060)*** | 0.958 (0.859 to 1.055)*** | −0.003 (−0.008 to 0.000) | 0.000 (−0.002 to 0.002) | 0.007 (−0.002 to 0.027) | - |
| Dairy | 0.931 (0.828 to 1.027)*** | 0.926 (0.825 to 1.020)*** | 0.001 (0.000 to 0.005) | 0.000 (−0.002 to 0.002) | 0.004 (−0.002 to 0.020) | - |
| Eggs | 0.917 (0.812 to 1.017)*** | 0.915 (0.811 to 1.015)*** | 0.004 (0.000 to 0.011) | 0.000 (−0.001 to 0.001) | −0.002 (−0.019 to 0.006) | - |
| Fish | 0.944 (0.844 to 1.050)*** | 0.944 (0.844 to 1.050)*** | 0.000 (−0.002 to 0.001) | 0.000 (−0.001 to 0.001) | 0.000 (−0.005 to 0.008) | - |
| Meat | 0.927 (0.822 to 1.026)*** | 0.904 (0.802 to 1.002)*** | 0.021 (0.005 to 0.038)* | 0.000 (−0.001 to 0.001) | 0.002 (−0.011 to 0.018) | 2.26 (0.51 to 4.07) |
| Miscellaneous | 0.930 (0.828 to 1.029)*** | 0.930 (0.829 to 1.030)*** | 0.000 (−0.005 to 0.006) | 0.000 (−0.001 to 0.001) | 0.000 (−0.007 to 0.006) | - |
- Note: Results are presented as β-coefficients and their respective 95% CIs. All models were adjusted for the following: age, sex, principal components of ancestry, alcohol intake, smoking, education, sleeping time, income, anxiety, physical activity, energy intake, and energy misreporter status.
- Abbreviations: PDI, plant-based dietary index; PRS, polygenic risk score.
- * FDR-corrected p < 0.10.
- *** FDR-corrected p < 0.01.
Because mediation was observed by meat intake and whole grains at the cross-sectional level, longitudinal analyses were conducted for these food groups and available obesity outcomes at follow-up (BMI and WC). Longitudinal results from Model 1 showed that whole grains intake at baseline mediated the association between PRS-Khera and BMI measured 6 years later among males (6.4%, p = 0.004, q = 0.03) and borderline significant mediation was observed for WC (10.9%, p = 0.02, q = 0.11; Table S6). However, when adjusting for baseline BMI or WC in mediation Model 2, no significant results were observed (Table S7). No significant mediation effects were observed for meat intake in the longitudinal analyses.
Sensitivity analyses
Statistical interactions from GLMs examining associations between PRS-Khera and PDIs and individual food groups for obesity outcomes are presented in Table S8. Fruits, vegetables, and sugar-sweetened beverages significantly moderated the association between PRS-Khera and BMI. Figures S4–S6 show interaction plots for BMI categories and PRS-Khera expressed in quintiles. The interaction plots showed associations in the expected direction. Specifically, fruits and vegetables attenuated the association between PRS-Khera and BMI, with a more pronounced pattern among the lower risk PRS quintiles. On the other hand, sugar-sweetened beverages accentuated the association between PRS-Khera and BMI across all PRS quintiles.
In CMA using PRS-Locke, no significant mediation or moderation effects were observed between PRS and PDIs or individual food groups on obesity outcomes (Tables S9–S11). However, when examining statistical interactions using GLMs, interactions were observed between PRS-Locke and tea and coffee intake on BMI and WC and between PRS-Locke and dairy intake on BMI (Table S11). Interaction plots for BMI are shown in Figures S7 and S8. The interaction patterns varied according to level of genetic risk of obesity. Coffee and tea intake appeared to attenuate the association between PRS-Locke and BMI among the lower risk quintiles, whereas it accentuated the association among the higher risk quintiles. Similarly for dairy intake, the association between PRS-Locke and BMI was attenuated among the lowest risk quintile but was accentuated as PRS quintiles increased.
DISCUSSION
Findings from our investigation partially support our hypothesis that adherence to plant-based dietary patterns would moderate and mediate associations between obesity genetic risk and obesity outcomes. Although overall adherence did not moderate associations, we observed that genetic susceptibility to obesity was partly mediated by meat intake in the overall sample and females and by whole grains in males both cross-sectionally and longitudinally. These findings extend the current knowledge of dietary factors as plausible mediators of the genetic pathway for obesity. Several studies have analyzed the associations between different PDIs and obesity outcomes, showing similar associations to our observations. Overall- and healthy-PDIs were inversely associated with obesity outcomes [(17, 30, 31)], and the unhealthy-PDI was positively associated [(17)] or did not associate with obesity outcomes [(30)], similar to the present findings when genetic susceptibility was not considered. The healthfulness of the dietary pattern is an important consideration when studying the relationship between plant-based diets and obesity. According to a recent review, it is unclear whether a stricter plant-based dietary pattern might be beneficial to reduce obesity [(32)]. However, in our study, animal-based food products were associated with an increased BMI, WC, and body fat percentage, suggesting that a more strictly plant-based dietary pattern may be beneficial for obesity.
In this investigation, intake of meat explained a small, but statistically significant, proportion of the genetic susceptibility to obesity cross-sectionally. Additionally, we observed that whole grains intake was found to partly explain a proportion of the genetic susceptibility to obesity in males both cross-sectionally and longitudinally, although not when the baseline outcome measure was adjusted for. Evidence suggests that genetic susceptibility to complex disease varies over the life course, and genetic susceptibility to obesity appears to be more pronounced in the earlier stages of life, which may partly explain our observation [(33)]. These results contribute to emerging evidence that diet may mediate the genetic susceptibility to obesity [(13, 14)]. They also suggest that there may be sex differences in gene-diet relationships with obesity outcomes, which requires additional investigation. The mediation percentages for food groups observed in the present study were similar to those found in a previous study of Quebec adults from a different cohort (1.5%–4.8%), which reported mediation of obesity genetic risk by diet quality and food groups [(14)].
Statistical interactions between the PRSs and dietary factors on obesity outcomes were only observed when examining interaction terms from regression models, but not from CMA. CMA is mainly used to assess mediation rather than interaction, which could explain this inconsistency [(34)]. Currently, only one previous study has analyzed gene-diet interactions on obesity outcomes using the healthy-PDI [(35)]. In contrast with the present results, they reported that adherence to the healthy-PDI attenuated the association between a 75-SNP PRS-BMI and obesity outcomes in adults from the UK Biobank [(35)]. In the present study, gene-diet interactions were not observed with PDI scores but were observed for individual food groups when modeled with a traditional approach. Fruit and vegetable intake attenuated, whereas sugar-sweetened beverage consumption accentuated, the genetic risk for obesity, reinforcing current dietary recommendations for all individuals irrespective of genetic susceptibility.
The differences in observed results when considering the two different PRSs requires discussion. In addition to the difference in BMI variance explained by the PRSs (10% and 3% for the genome-wide PRS and 92-SNP PRS, respectively), another relevant consideration is the difference in biological pathways that would be captured by the PRSs. For example, a previous study analyzed gene-diet interactions between different dietary pattern scores and a 97-SNP PRS-BMI, which was also divided into two sub-PRSs: 54 SNPs related to the central nervous system and 43 SNPs unrelated to the central nervous system [(25)]. Significant interactions were observed with the 54-SNP sub-PRS and the overall 97-SNP PRS-BMI, but not with the sub-PRS that was unrelated to the central nervous system. This highlights the importance of using a PRS that captures the relevant underlying biological pathways between the exposure and outcome. Previous mediation studies have also reported that eating behaviors mediated the genetic susceptibility to obesity [(13, 36, 37)]. Therefore, it will be important for future work to evaluate eating behaviors in addition to dietary intake as factors implicated in obesity genetic risk, particularly because well characterized obesity genes are mostly expressed in brain areas involved in appetite modulation [(22)]. Finally, the genome-wide PRS from Khera et al. explained 23% of BMI variance in the cohort from which it was derived [(21)], but substantially less in the present study. The CARTaGENE cohort largely comprises French Canadian individuals, a founder population, indicating that the performance of the PRS varies between populations. Hence, continued efforts to develop a population-specific PRS, considering different ancestries and stages of life, will be essential to improve the scientific understanding of gene-diet relationships with obesity and cardiometabolic outcomes.
Although the present study focused on evaluating adherence to plant-based dietary factors as possible mediators or moderators of genetic susceptibility to obesity, consideration of the underlying nutrient intake patterns is warranted. Prior interaction studies have reported that different fatty acids moderated the genetic susceptibility to obesity [(38-40)]. Indeed, some of the top target obesity genes, including fat mass and obesity-associated protein (FTO gene), melanocortin 4 receptor (MC4R gene), and leptin (LEP gene), are associated with lipid metabolism and energy regulation [(41)], suggesting that fats may have a particularly relevant role in the obesity genetic pathway. Although the present investigation did not report interactions for food groups rich in fatty acids, the findings of mediation by meat (typically a mixture of fatty acids and protein) and whole grains (low in fat) are notable. A recent genome-wide association study of macronutrient intake in the UK Biobank discovered that certain macronutrient intake-related genes expressed in the brain, particularly in the cerebellum, were enriched for dietary intake signals [(42)]. Hence, as a future direction, it will be useful to identify the regions and/or mechanisms of expression of obesity genes to construct more phenotype-specific PRSs that are composed of genetic variants implicated in specific pathways, such as metabolism, adipose tissue, appetite regulation, and eating behavior. This approach may help to improve the precision of identifying dietary factors and behaviors that are implicated in specific pathways of genetic susceptibility to obesity.
This study is not without limitations. First, data were analyzed from a cohort largely composed of French Canadian individuals, hence the results may not be generalizable to other populations. In addition, dietary data were self-reported, which is prone to measurement error, particularly among subgroups that are likely to underreport their intakes and body weight (e.g., females and individuals with overweight) [(43, 44)]. Most participants did not provide follow-up data, so there may have been a lack of statistical power for longitudinal analyses as well as attrition bias. Last, effect sizes in observed associations were small and results varied when evaluating the two different PRSs. Thus, consideration is needed of the most suitable PRS for specific research questions. Despite these limitations, the present study also possessed important strengths, including the availability of measured anthropometric variables at baseline and availability of longitudinal data that improved the quality of the observational evidence.
CONCLUSION
The present findings support prior evidence that adherence to healthful plant-based dietary patterns is associated with a lower BMI, WC, and body fat percentage. The findings suggest that intake of certain food groups, including meat and whole grains, partly mediated the genetic susceptibility to obesity, with lower meat and higher whole grains consumption having beneficial patterns. Moreover, higher consumption of fruits and vegetables and lower consumption of sugar-sweetened beverages may attenuate the relationship between genetic risk and obesity outcomes. Hence, promoting components of plant-based dietary patterns could help to mitigate obesity risk, particularly among individuals with higher genetic susceptibility to obesity. Future studies may be improved by assessing more biologically informed PRSs based on gene coexpression data from specific tissues/regions [(45)] and comparing genome-wide versus non-genome-wide PRSs to understand which approach is most suitable to characterize the obesity genetic risk.
AUTHOR CONTRIBUTIONS
The authors’ responsibilities were as follows—Guiomar Masip, Daiva E. Nielsen: conceptualization; Guiomar Masip, Daiva E. Nielsen: methodology; Guiomar Masip, Atheer Attar: formal analysis; Guiomar Masip: data curation; Guiomar Masip, Daiva E. Nielsen: writing original draft; Guiomar Masip, Atheer Attar, Daiva E. Nielsen: draft review and editing; and all authors: reading and approving the final manuscript.
ACKNOWLEDGMENTS
The authors thank the CARTaGENE cohort for their contribution to the scientific community. We thank Professor Louis Pérusse for providing feedback on this manuscript and Nacho Lasheras for valuable coding assistance.
FUNDING INFORMATION
Guiomar Masip received funding support from the Institut sur la nutrition et les aliments fonctionnels (INAF) and Fonds de Recherche du Québec–Santé Postdoctoral Fellowship. Atheer Attar is supported by a doctoral scholarship from King Abdulaziz University. Daiva E. Nielsen is supported by a William Dawson Scholar award from McGill University.
CONFLICT OF INTEREST STATEMENT
The authors declared no conflict of interest.
Open Research
DATA AVAILABILITY STATEMENT
Codebooks for the statistical analyses conducted in this research are available from the authors upon reasonable request. Requests for data access must be submitted to the CARTaGENE Samples and Data Access Committee for review.
