CC Genotype at TCF7L2 Diabetes Risk Locus rs7903146 Directs a Coordinated Fatty Acid Response to a Mediterranean Diet Intervention: A Randomized Controlled Trial

TCF7L2, encoding an essential transcription factor in Wnt signaling, has been studied extensively owing to its consistent association with type 2 diabetes (T2D) susceptibility in humans [1]. Among many TCF7L2 variants, rs7903146 has received the most attention, with numerous studies demonstrating the T allele as a significant risk for T2D. This allele has implications beyond T2D, namely relating to disturbances in glucose homeostasis and obesity-related metrics, showing effects across various tissues. For instance, in both rodent and human pancreatic islets, the risk T-allele correlated with increased TCF7L2 expression, leading to reduced insulin content and secretion [2], a phenomenon independent of adiposity [3]. Similarly, in peripheral blood mononuclear cells, allele-specific upregulation has been observed [4]. TCF7L2 is instrumental in regulating glucose tolerance by modulating glucagon and beta-cell insulin secretion [5] and has vital functions in controlling proglucagon gene expression and glucose balance in the gut and brain [6]. Moreover, this risk allele is associated with higher glucose production in the liver [7], and studies in rat hepatic cells have revealed TCF7L2 binding to essential glucose metabolism genes [8]. Yet, knowledge is scant on the mechanisms by which these associations affect T2D outcomes.

Although plasma glucose and acetylated hemoglobin remain standard diagnostics for T2D, levels of free FAs (FFAs) and other lipids can indicate potential metabolic dysregulation leading to insulin resistance [9‒14]. For example, when hepatic Tcf7l2 is inactive in mice, liver steatosis develops by prioritizing carbohydrate metabolism, thereby initiating de novo lipogenesis [15]. Similarly, deleting Tcf7l2 in mouse adipocytes resulted in disrupted lipid metabolism and impaired glucose tolerance [16]. Metabolomics analyses in human cohorts further show that elevated plasma saturated fatty acid (SFA) and monounsaturated fatty acid (MUFA) concentrations are indicators of T2D risk [17, 18]. Thus, while TCF7L2 is central to regulating glucose and lipid metabolism, how the strength of its actions varies by genotype during a dietary intervention is not well characterized.

The Mediterranean (Med) diet has demonstrated efficacy in reducing T2D risk and other metabolic complications in certain populations [14, 19, 20], potentially exceeding the benefits of an LF diet [21‒23]. However, gaps remain in understanding how specific diets interact with TCF7L2 genotypes and the broader dynamics of glucose and lipid metabolism. The impact of lifestyle, in conjunction with other molecular factors like DNA methylation, on TCF7L2 in a genotype-specific context remains poorly explored [24]. Upon this foundation, our study aimed to investigate the diet-induced, genotype-related changes in metabolic responses among individuals with different T2D susceptibilities. Using a crossover design comparing LF and Med diets over 1-week intervention periods, we aimed to delineate the metabolic changes associated with the T2D risk locus TCF7L2 rs7903146, using blood metabolite data from the beginning and end of each intervention period.

For this genetics-based dietary intervention study, a methods supplement details the experimental approach (for all online suppl. material, see https://doi.org/10.1159/000542468). A schematic depicting the study design is presented in online supplementary Figure S1, and the study was registered at ClinicalTrials.gov: https://www.clinicaltrials.gov/study/NCT03458494. Prospective participants were recruited to the study via an internal list of former participants, social media, the research center’s website, and informational flyers, and after a telephone screening interview were invited to submit a sample for genotyping and basic screening (see Participant Characteristics and Genotyping, below). Inclusion criteria were men and women (not pregnant) aged 18 years or older, and BMI between 27 and 34 kg/m². Exclusion criteria included evidence of active liver disease, severe renal dysfunction, alcohol consumption above 2 drinks/day, preexisting cardiovascular disease, uncontrolled T2D (fasting glucose >126 mg/dL), uncontrolled hypertension (systolic blood pressure >180 mm Hg or diastolic blood pressure >100 mm Hg), thyroid diseases, taking lipid-lowering or diabetes medications, smoking, pregnancy, BMI below 27 or greater than 34 kg/m², any extreme dietary habits, multiple food allergies, extreme levels of physical activity, or changes in body weight >20 lbs during the last 6 months, plus other criteria listed in the supplement. Persons expressing interest in the study were screened, and each provided signed consent forms. The principal investigator was blinded to the randomization scheme until study completion. The study was approved by the Institutional Review Board at Tufts University (12691) and adhered to the ethical principles of the Helsinki Declaration of 1975 as revised in 1983.

Of 146 individuals who expressed interest in the study and were genotyped at rs7903146, the genotype distribution was 91 (62%), 16 (11%), and 39 (27%) for CC, TT, and heterozygous CT individuals, respectively. Forty individuals comprised the initial study cohort, 25 with CC genotype and 15 with TT. All participants completed the 3-week dietary intervention period, except five individuals, all having CC genotype (online suppl. Fig. S1), for a variety of reasons, and their details are presented in the supplement. For the 35 individuals who completed the entire study, 15 (43%) were female, the age range was 18–70 years (mean 46.3, SD 18.5), and the BMI range was 26.38–33.95 kg/m² (mean 30.33, SD 2.3) (Table 1). The BMI range, representing the 50th to 75th percentile of the USA population, was chosen to target individuals at higher metabolic risk. No statistically significant differences based on genotype were noted for anthropometric, clinical, or dietary measures. Furthermore, the five participants who did not complete the study were not significantly different from those who did in terms of baseline characteristics, except for one individual who presented with three of four diastolic blood pressure measurements (two left, one right) that were elevated and beyond the 95% confidence interval of the mean of the group that finished the study. This individual did not complete the study because all participant-facing activities at the research site ceased owing to the COVID-19 pandemic.

Saliva for DNA analysis was collected with DNA Genotek #OGR-500 (DNA Genotek, Ottawa, ON, Canada) per the vendor’s protocol. Purified DNA pellets were dissolved in 100 µL of TE. The TCF7L2 T/C rs7903146 SNP was genotyped with TaqMan assays (Applied Biosystem QuanStudio 6 Flex; ThermoFisher Scientific, Waltham, MA, USA) [25]. A goal in recruiting the study participants was 40 participants, comprising 20 of each genotype, CC and TT. Because the cohort size was small and because TT is the rarer genotype at the rs7903146 variant (∼8%–11% of the US population), it was important to match TT and CC individuals by certain influential characteristics, namely age and sex.

The intervention diets were Med or low-fat (LF), each of 2,200 total kilocalories and designed to maintain body weight. The Med diet had the following percent calories: 41% fat (9% total energy from saturated fat), 42% carbohydrate, 17% protein, and 26 g total dietary fiber. The LF diet had the following percent calories: 30% fat (9% total energy from saturated fat), 53% carbohydrate, 17% protein, and 23 g total dietary fiber. Regarding dietary fatty acids, the Med diet provided 41% total fat, with 9% from saturated fats, and the remaining fat intake comprising primarily MUFA and polyunsaturated fats (PUFA). The LF diet provided 30% total fat, also with 9% from saturated fats, but proportionally less from MUFA and PUFA. Example meals and other details of the intervention are described in the supplement. Each intervention spanned 7 days. Participants were randomly assigned one diet and, after a ∼10-day washout period, were placed into the other diet group for the second 7-day intervention. During the washout, participants consumed their own food and their typical diet. Compliance was monitored by providing all meals to participants and checking for adherence through self-reports.

Physical examinations included height and weight. Participants were requested to fast overnight for at least 12 h prior to venous blood draw (50 mL) for metabolomics analysis, taken both prior to and upon completion of each intervention. Samples were stored on ice immediately after collection and during transport to the on-site laboratory. Blood was fully processed within 1 h of collection. Plasma and serum aliquots were frozen at −80°C until analysis. All analyses were performed as a batch.

Metabolomics profiling of plasma was performed as described [26] by Nightingale Health (Helsinki, Finland). The high-throughput NMR metabolomics platform was deployed on beginning-of- and end-of-intervention plasma samples from all participants. That platform quantified 249 metabolic biomarkers: 168 were directly measured and 81 were ratios, including amino acids, fatty acids, lipids, ketone bodies, and other low-molecular-weight metabolic biomarkers, plus lipoprotein subclass distribution, particle size, and composition. Samples were evaluated for batch effects and quality control, and data were returned in absolute relative units (mmol/L, %).

Metabolomics data were loaded into R (version 4.2.2) [27] and RStudio (version 2023.03.1+446) (Posit Software, Boston, MA, USA). For each participant, delta intervention or simply delta values under each diet were calculated for each metabolic factor as end-of-intervention value minus beginning-of-intervention value. Delta values represent the change in a metabolite in response to the 1-week intervention for that participant. Log (natural) transformation reduced skewness in the data. Four correlation matrices were determined in R (rcorr function, Pearson correlation type, Hmisc package), with data split by genotype and diet. Correlation plots were made with corrplot and hierarchically clustered (hclust); clustering shuffles the metabolic factors, grouping similarly correlating factors on the subsequent plot. Flattened correlation matrices were filtered to yield pairs of metabolic factors, beginning of intervention, and delta intervention, that were very highly correlated and significant in one genotype and not the other. In the event of observing discordant responses to one diet based on genotype, the filtering was implemented because genotype was the basis for the study. We evaluated the ratio of log₁₀ of the p value of correlation coefficient r in the CC genotype group to the corresponding log₁₀ of the TT group p value, naming this ratio log(pCC)/log(pTT). Using a ratio of p values and an aggregate score of shared correlating factors was derived in part from the use of ratios in a study on acetaminophen metabolism [28] and the Coronary Event Risk Test (CERT2), which bundled ceramide and phosphatidylcholine levels plus ratios to predict cardiovascular risk [29]. Please see Results for details regarding the application of the aggregate score to the findings in this study. An additional rationale for filtering the correlation data was to reduce the statistical burden of numerous regression tests. The work presented here concentrates on 23 lipid factors – 18 FAs and 5 other lipids. Thus, any correlation matrix prior to filtering began with 529 correlation pairs as 23 beginning-of-intervention values relating to 23 delta values. For comparison purposes, all correlation pairs that passed the above filtering criteria in one diet-genotype group were retained in the other three diet-genotype matrices in their respective correlation plots. Tests for genotype-based interactions of a beginning-of-intervention metabolic factor affecting the delta-intervention value of any metabolite were conducted with linear regression using fixed effects in R. Covariates included age, BMI, sex, and intervention diet order. Details on plotting data and the packages used are in the supplement.

Functions mean and sd provided the mean and standard deviation. Standard and Welch’s T tests were run with the t.test function, var.equal = TRUE/FALSE, respectively. For correlation coefficients generated in a 23 × 23 matrix, a strict p value of significance was defined by 0.05/23², or 9.45E−05. For association tests and subsequent genotype by metabolite interaction tests, the significance threshold is 0.05 after Bonferroni correction, depending on the number of fully independent tests performed on each outcome.

The intervention diets produced a significant change from the habitual diet in percent energy from fat, which was 36.13% (SD = 9.60) and 36.30% (SD = 9.32) in the CC and TT genotype groups, respectively. Comparing baseline diets to the Med diet intervention, the mean increase in percent energy from fat (4.8%) was significant (p = 0.0034, T test). Comparing baseline diets to the LF diet intervention, the mean decrease in percent energy from fat (6.2%) also was significant (p = 0.00021, T test). Of note, both diets had 9% of total energy from SFA, but this amount was proportionally more total calories from fat for LF (30%, 9/30) than for the Med diet (22%, 9/41).

We sought to examine TCF7L2 genotype effects on the FA and other lipid metabolites in response to two intervention diets. Our reasons were fat type and content in the diet and FA profiles in blood are influential T2D risk factors [9‒14, 30‒33], and blood metabolite profiles from participants in this study were assessed for a variety of FA variables. Specifically, we were interested in responses to the dietary interventions, i.e., identifying the influence of beginning-of-intervention levels of a metabolite on the delta-intervention levels of any metabolite, including itself, in either a genotype- or diet-specific manner. The correlation matrices relating beginning-of-intervention values to delta values were filtered as described in the Methods to identify the strongest correlations in any diet-genotype group, prior to illustrating the relationships in correlation plots.

Correlation matrix plots (Fig. 1) illustrate that with the Med diet intervention (panels a and b), there are many more strongly positive or negative correlations of beginning-of-intervention to delta values in CC individuals than in TT. This shows a lack of tight regulation between the metabolite variables assayed in the TT individuals, which then suggests that there is increased loss of coordinated FA metabolism and homeostasis with the TT genotype on the Med diet. The observed correlations in these plots, or lack thereof, suggest a type of network-based approach for investigating genotype by metabolite interactions affecting delta-metabolite outcomes. Thus, for each of four diet-genotype correlation matrices, we filtered the correlation data by the p value of the correlation coefficient r, whereby the log(pCC)/log(pTT) ratio was either <−3.0 or >3.0 to identify in one diet beginning-of-intervention to delta relationships that were robust for one genotype but essentially nonsignificant in the other genotype. This process identified 72 relationships in the Med diet data that were promoted to further analysis, including networks, regression analysis, and tests for genotype-diet interactions (online suppl. Table S1).

Differences between the correlation coefficients r for a given beginning-of-intervention/delta-metabolite pair in CC vs. TT groups ranged from 0.265 to 0.843 (mean, 0.524), with CC exhibiting the stronger correlation in all 72 relationships. For example, the beginning-of-intervention ratio of triglycerides to phosphoglycerides correlated with delta-SFA values with r of −0.754 (p = 7.87E−05) in CC but with r of −0.315 (p = 0.27) in TT – difference in r values equals 0.439. Other examples include the largest difference in r values, where the beginning-of-intervention n3-PUFA level correlated with delta-MUFA with an r of −0.689 (p = 5.51E−04) in CC but 0.154 (p = 0.60) in TT (log (pCC)/log (pTT) = −3.036, difference in r values = 0.843), and the largest difference in log of p values of r with the pre-MUFA level on delta-SFA having r −0.896 (p = 4.07E−08) in CC but −0.318 in TT (p = 0.27; log (pCC)/log (pTT) = −6.818).

In contrast to the 72 correlations that were strongly genotype-specific in the Med diet group, the LF data presented no such compelling differences in the pre-to delta-correlations (Fig. 1c, D; online suppl. Table S2). In fact, the four strongest signals in the LF correlation matrix all involve delta-MUFA as influenced by beginning-of-intervention values of MUFA, n6-PUFA, PUFA, and total FAs, where the log(pCC)/log(pTT) ratio was between −2.24 and −2.32. These four corresponding values in the Med diet matrix were between −3.04 and −5.77 (online suppl. Table S2). Although these four relationships concerning delta-MUFA levels at the end of the LF intervention are stronger in CC on LF than in TT, the LF data did not produce genotype-based differences in correlations across FAs and other lipids that were sufficiently stark to suggest genotype interactions. Nonetheless, we tested for an interaction of genotype by beginning-of-LF intervention metabolite levels affecting delta-MUFA as outcome under a fixed-effect model with age, BMI, diet order, and sex as covariates. For beginning-of-intervention levels of n6-PUFAs on delta-MUFA, the p value of the genotype by n6-PUFA interaction was 0.056 (beta = 0.408).

Within the set of relationships of beginning-of-intervention affecting delta values that passed our filter criteria, we observed 14 very tight correlations (Pearson correlation coefficient r > 0.802 or <−0.696, all p < 4.62E−04) involving delta-SFA levels. These correlations were observed in the CC genotype and with the Med diet, encouraging scrutiny by testing for genotype interactions. Thus, as this set showed the highest number of such genotype- and diet-specific relationships converging on the delta of a single metabolite, we elected to investigate first the associations and genotype interactions for delta-SFA. Of 14 metabolic factors highly correlated with delta-SFA in CC individuals on the Med diet, 11 negatively correlated with delta-SFA, indicating decreased SFA in blood with increasing levels of the other FA or lipid factor. Tests for interactions of genotype by single metabolic factor acting on delta-SFA indicated a range of relationships including several that failed to reach significance (online suppl. Table S3). The most significant of these interactions was for MUFA, p = 0.0059, beta = 0.57. The strong correlations observed with CC individuals on the Med diet were not detected with TT individuals nor with either genotype on the LF diet.

With 14 metabolic factors converging on the change in a single blood measure, namely SFA, we explored an approach to test for the genotype interaction of these factors in aggregate on delta-SFA. To achieve this, we devised an aggregate score for each participant. The aggregate score was derived in a two-part process: first, for each metabolic factor, the level of that factor was standardized to the mean of the cohort; then, standardized values were summed to yield a participant’s aggregate score. This was done separately for the 11 negatively correlating and the three positively correlating factors. We termed the two derived variables score-neg11 and score-pos3, whereby respective means across the population were 11.00 and 3.00. The score-neg11 factor was then used in place of a single metabolic factor to test for the genotype interaction on delta-SFA as an outcome. Results showed a significant genotype by score interaction (p = 4.64E−03, beta = 0.209, SE = 0.068, adjusted R² = 0.657) (Fig. 2). No single metabolic constituent of the set of factors used to derive the aggregate score showed as strong an interaction (see online suppl. Table S3). Adding diet order as a covariate to the interaction tests yielded an interaction p value of 6.25E−03.

To identify the most and least impactful factors influencing delta-SFA, we serially removed one of the eleven components and recalculated the aggregate negative score for the remaining 10. This leave-one-out analysis showed that the ratio of triglycerides to phosphoglycerides (or glycerophospholipids) was the most influential component of the 11 beginning-of-intervention FAs and other lipids that correlate strongly and negatively with delta-SFA during the Med diet intervention. Removing this metabolic factor from the 11-member set resulted in both the p value of interaction and the adjusted R² changing to the highest degree (Table 2). The least influential component was the level of plasma n3-PUFAs, where its omission from the aggregate score decreased the p value of interaction to 3.14E−03 and increased R² to 0.678. Leaving out any of the other nine components of the score-neg11 derived variable yielded p values essentially equivalent to that of the full set of 11 components (range: 4.32E−03 to 4.91E−03). This was also observed for the adjusted R² values of the model. From the LF intervention, deriving score-neg11 values and then testing for a genotype interaction on delta-SFA gave a p value of 0.27.

Initial correlation analyses highlighted three metabolic factors that correlated strongly and positively with delta-SFA in CC individuals on Med diet, namely the PUFA to total FA ratio, the PUFA to MUFA ratio, and the omega-6 PUFA to total FA ratio. Deriving an aggregate score of these three factors, identical to the description above for the 11 negative factors, and using that score as a term in the genotype interaction affecting delta-SFA yielded a p value of 0.041. Applying data from the LF intervention to this interaction test gave a p value of 0.92.

Under the filtering criteria of genotype- and diet-specific correlation analyses described above, we also observed other highly networked components of the FA and other lipid factors: delta values of the ratio n6-PUFA to total FA (10 metabolic factors), MUFA (10 factors), total FAs (9 factors), and the ratio PUFA to total FA (8 factors). Individual beginning-of-intervention metabolite levels were standardized to the cohort mean, and aggregate scores were determined, as described above. The aggregate scores served as terms in the genotype by aggregate-metabolite interaction tests acting on the delta value. None of these delta values showed genotype by aggregate score interactions as strong as the coalesced negative-correlating factors acting on delta-SFA. Yet, some interactions qualify as statistically significant, including the aggregate of nine negatively correlating factors and genotype acting on delta-MUFA (p = 7.22E−03) and nine negatively correlating factors and genotype acting on delta-total FAs (p = 1.65E−02). All results are presented in online supplementary Table S4 and online supplementary Figures S2, S3.

We observed nine factors that negatively correlated with delta-MUFA (Table 3). Testing for a genotype by aggregate score interaction for these nine factors acting on delta-MUFA in the LF intervention trended toward significance (p = 0.074, beta = 0.197). Although merely a trend, this is the best p value of genotype interaction in the LF intervention among all networked FA relationships initially observed in the Med diet. The nine metabolic factors jointly contributing to affect delta-MUFA include four factors (subset 1) with the absolute greatest differences in log(p value) among all 529 pairs in the LF correlation matrix and with more robust relationships to delta-MUFA in CC genotype than TT. These four beginning-of-intervention factors are MUFA, n6-PUFA, PUFA, and total FAs. Using only those four factors in generating an aggregate score and testing for a genotype by score interaction on delta-MUFA yielded p of 0.0147 in Med and p of 0.040 in LF. Together, these results indicate that CC individuals on either a LF or Med diet exhibit stronger negative correlations between baseline levels of several FAs and the intervention-induced change in MUFA but also show that the same metabolic factors at beginning-of-intervention values contribute to delta-MUFA but with different strengths. A larger follow-up study is warranted to distinguish the most important baseline factors affecting delta-MUFA, or if several factors are equally important, and to verify the relationships in LF with delta-MUFA.

The current study was designed to provide more concrete evidence of the mechanisms by which a common T2D-associated DNA variant leads to observed outcomes. Thus, we organized a human intervention trial based on the TCF7L2 rs7903146 genotype and two diets in a randomized crossover design followed by metabolomics data analysis to characterize the genotype- and diet-specific responses. The diets investigated here – LF and Med – were selected because analysis of data from the PREDIMED study in Spain identified a number of gene-diet interactions affecting T2D incidence and risk [34, 35]. Overall, we observed that the CC genotype, the most common globally and conferring lower T2D risk, responded to the Med diet, and that response was a coordinated influence on the metabolism of FAs and other lipids. In contrast, the TT genotype, the allele conferring elevated T2D risk, did not respond so uniformly across these FAs and other lipid measures. Specifically, the observed correlations across the delta-SFA network under the Med diet intervention, and the score-neg11 components in particular, are lost or severely weakened by a short-term LF diet intervention.

The study design used a BMI range of 27–34, chosen to target individuals at higher metabolic risk. This allowed the exploration of the interaction between TCF7L2 genotype and diet in a population more likely to experience variation in metabolic responses. While it is possible that results could differ in individuals with a lower or healthy BMI, our study focused on a population with BMI in the overweight to obese range, which better reflects the target population for dietary interventions addressing type 2 diabetes and obesity.

Although the two intervention diets used in the current study did not differ in total calories from SFA and protein, there were differences in the percent energy from total fat and carbohydrates. Thus, aspects of any diet-specific response observed must consider the different dietary fats. Although not designed specifically as dietary intervention studies, certain reports examined dietary and other interactions involving TCF7L2 rs7903146 and various cardiometabolic outcomes. One, an analysis of a cohort from Lebanon showed that individuals who were homozygous TT (at TCF7L2 rs7903146) had significantly higher BMI and body fat despite lower intakes of saturated fat and a trend to lower muscle mass with higher intakes of saturated fat when compared to CC individuals [36]. Two, in the LIPGENE-SU.VI.MAX study, metabolic syndrome risk was associated with T-allele carriers, but in an interaction, high SFA intake (≥15.5% energy) worsened MetS risk (OR 2.35, 95% CI 1.29–4.27, p = 0.005) and was associated with additional impaired insulin sensitivity relative to CC homozygotes (p = 0.025) [37]. No significant genotype effect on MetS risk or insulin sensitivity was evident among low-SFA consumers [37]. However, it must be acknowledged that the observed interaction derived from lower intake of other perhaps healthier fats. Three, the diabetes-associated allele of rs7903146 was shown to impair glucose tolerance through effects on both glucagon and insulin secretion, especially during an accompanying elevation in FFAs [5].

Several reports presented differential responses to the diet based on the rs7903146 genotype or rs7903146 genotype by diet interactions regarding associations with cardiometabolic outcomes. Those longitudinal or observational studies examined the effects of various diets, including high-carbohydrate hypoenergetic, high-fat hypoenergetic, Med, Nordic, and others [38‒41]. Yet, these analyses examined the effects of the rs7903146 genotype retrospectively and not within an intervention specially based on the genotype at this TCF7L2 variant. T risk-allele carriers in the Boston Puerto Rican Health Study with a high Med diet score had lower weight and waist circumference compared with CC participants [38]. Another study compared the effects of Med and Nordic diets, where marginally significant interactions between Med diet score and rs7903146 on weight gain suggested a beneficial effect of Med diet score only in carriers of one or two risk alleles, and no differences with the Nordic diet [39]. Our metabolomics data do not show evidence in the blood metabolites for beneficial effects of the TT genotype in response to the Med diet, and a 1-week intervention is not designed to test for changes in obesity anthropometrics. In addition, the protective effect of whole-grain intake on the risk of developing T2D was lost in persons who carried the TCF7L2 rs7903146 T allele [40], likely because this diabetogenic allele decreases glucagon-like peptide-1 expression and actions [41]. One effect of this is glucagon-like peptide-1, known to slow gastric emptying [40], potentiating insulin secretion from beta cells.

Working from a set of correlations that were robust in one genotype or diet group but weak or nonexistent in others, we built a lipid and FA network of beginning-of-intervention values of specific metabolic factors that joined to affect delta values of other lipid factors. The strongest of this type of beginning-of-intervention by genotype interaction was for delta-SFA, then delta-MUFA and delta-total FAs (Fig. 2; online suppl. Table S3; Table 2; online suppl. Table S4). For these three interactions, the genotype-based interaction was observed at the end of the Med diet intervention and with a collection of the single metabolic factors that were strongly negatively correlated with the respective delta outcome. Higher values of the individual beginning-of-intervention factors correlated with greater decreases in the outcomes. These relationships were not statistically significant with the Med diet intervention in TT individuals and not with the LF diet for either genotype (online suppl. Table S4). On one side, the T2D non-risk genotype exhibits statistically significant interactions across a network of FAs and lipid metabolic factors, both single entities and ratios. This suggests a healthier metabolic profile for lipid homeostasis. On the other side, the weakening or outright loss of the relationships among these lipid factors under the other intervention scenarios tested here (TT on Med diet, both genotypes on LF) suggests, albeit perhaps weakly, that there are impactful disruptions to lipid homeostasis. Consequences of these disruptions are fatty liver [31, 42], hepatic and adipose insulin resistance [30, 32], and impaired mitochondrial function, membrane composition, and metabolic capacity [33, 43]. Moreover, elevated TG and decreased levels of various phospholipids have been noted in persons with overweight/obesity or diabetes [44], conditions which can define metabolic-dysfunction-associated fatty liver disease (MAFLD) [45]. In addition, it must be acknowledged that the LF diet may be classified as a high-carbohydrate diet as well as differing in the actual foods consumed.

Given the combination of genotype, anthropometric measures, disease status, and metabolic profile, precision nutrition attempts to formulate a diet that will elicit changes in responsive phenotypes, aiming to improve the health status of the individual or group of highly similar individuals. Thus, in this study of admittedly short duration we sought to identify factors that alter metabolic endpoints after a dietary intervention period and ascribe those changes either to a genetic difference, the diet or both in concert. Clinical and disease outcomes require interventions of longer duration. Adopting a version of the Med diet has been shown to lower the risk of T2D or increase the probability of remission [19, 20], while less impact was reported with LF diets [19, 21‒23]. Yet such interventions, even those lasting years, did not produce absolute or consistent results. It is well recognized that there is no universal response to the adoption of these or other diets, which translates to incomplete achievement of health goals. One implication of these results is the need to fold genetic and epigenetic variation as well as lifestyle and cultural factors into models that develop diet and lifestyle (e.g., exercise regimens) changes predicted to yield greater success in reducing disease risk. Such models require approaches steeped in machine learning methods [46‒49], particularly because it is not feasible to conduct interventions based on a single-locus one at a time as such studies are limited and all pertinent loci for a single phenotype along with relevant perturbations (i.e., diet or other lifestyle interventions) are numerous. Future dietary intervention studies will progress to use both the findings and the boundaries uncovered by single-locus studies and will be based on predictive models that have been advanced with machine learning approaches.

An important limitation of this study is its small sample size of 35 participants with a genotype distribution of 21 CC and 14 TT. Because caloric needs of different individuals may vary, the 2,200 daily calories provided to participants for this short-term intervention may not have met the needs of all individuals. It is important to note that no significant weight changes were observed during the short-term dietary period. The fiber content of the Med diet as provided in this study is below certain standards, which could contribute to some of the observed metabolic differences, but is more representative of a real-world dietary pattern. While this study was designed in the context of precision nutrition, the range of metabolomics data within the study limits the discovery and interpretation of the full extent of the metabolic disruptions and homeostasis that can be attributed to the genotype difference and the two diets tested. The work presented here has a focus on FAs and other lipids because many inter-related changes to lipids were observed in both genotype- and diet-specific ways. Pertinent to the focus on FAs is the varied duration of the washout across all participants and our acknowledgment that lipid metabolism might stabilize over a longer period such that a fixed, longer washout period could provide more robust results. Additionally, lipids are central in the onset and progression of the loss of insulin-controlled glucose homeostasis [9, 16, 50], but the relationships of these lipid and FA factors with other metabolites also assessed as a part of this study (e.g., citrate, pyruvate, ketone bodies, and amino acids) were not investigated in this context. Another limiting factor was the short intervention period, but one designed to capture acute dietary responses. Longer exposure to these two or any other diets certainly would actuate different, perhaps more nuanced or more stabilized, changes to blood metabolite patterns. In addition, any intervention study benefits from accurate testing for adherence, which was not performed. In this study, all food was provided, and participants were asked about compliance.

A two-part dietary intervention based on the T2D risk locus TCF7L2 showed that individuals with genotype CC at rs7903146 taking the Med diet displayed the most robust coordination of lipid factors. This relationship was most apparent regarding delta-SFA and delta-MUFA. The diminished interdependencies among the analyzed lipid factors observed with the TT genotype or the LF intervention have the potential to elevate T2D risk via a combination of disruptions to lipid homeostasis and tissue-specific insulin resistance.

The authors acknowledge the effort and contributions made by the study participants, as well as HNRCA staff, who facilitated volunteer participation and sample and data collection.

This study protocol was reviewed and approved by Tufts Health Sciences (HS) IRB Office Approval No. 12691. Included in the screening process for interested volunteers was disclosure of key aspects of the study; all participants provided signed, written informed consent forms.

All other authors have no conflicts of interest to declare.

The Allen Foundation funded this work in part, and the US Department of Agriculture funded this work in part under project 8050–51000-107-00D. The USDA had no part in the design of this project, collection, analysis and interpretation of data, or composing the manuscript. Mention of trade names or commercial products is solely for providing specific information and does not imply recommendation or endorsement by the USDA. The USDA is an equal opportunity provider and employer.

J.M.O. conceived the study and established the analyses. L.D.P. and C.-Q.L. contributed to the design of analyses, analyzed the data, and wrote the manuscript. L.D.P., C.-Q.L., C.H., J.J.C., and J.M.O. interpreted the results and critically reviewed the manuscript.

The data that support the findings of this study are not publicly available because certain information in those datasets could compromise the privacy of volunteer research participants. The data may be available from the corresponding author [J.M.O.] upon reasonable request.