Lifestyle precision medicine: the next generation in type 2 diabetes prevention?

Two contrasting approaches have been used most in studies of genetics and lifestyle in complex disease traits. The dominant strategy has been the use of genome-wide association studies (GWAS), which leverage massively parallel genotyping technologies to interrogate variation across the genome, in a way that is agnostic to prior knowledge about genes, lifestyle, or disease. The second strategy involves studies focused on animal or human biology, through which genes and pathways have emerged as subsequent targets of epidemiological studies and/or clinical trials. The published findings from the latter are generally far less reproducible than those from studies using GWAS-based approaches.

The use of large cohort collections within which genetic variation, self-reported measures of macronutrient intake, and other relevant factors are included have facilitated the discovery of variants associated with protein and carbohydrate intake. The CHARGE consortium reported robust associations between FTO variants and dietary protein intake in 177,000 adults [112], whereby each copy of the rs9939609 ‘A’ allele was associated with a mean of 0.08% (95% CI 0.06–0.10%, P = 2.4 × 10^–16) higher protein intake; weak associations between the ‘A’ allele and lower total energy intake and lower dietary carbohydrate intake were also reported. The latter contradicts data from studies in children using objective measures of total energy intake, where the ‘A’ allele was associated with higher total energy intake [113].

There is compelling evidence that the hepatokine fibroblast growth factor 21, and variation at the gene that encodes it (FGF21), helps regulate energy homeostasis [114] and influences macronutrient and alcohol preference in animals and humans [115‐118]. The FGF21 hormone is induced by metabolic stress, including ketogenesis and dietary carbohydrate consumption. FGF21 was initially suggested as a therapeutic agent for diabetes [114] by the pharmaceutical giant Eli Lilly and Co. Exogenous FGF21 infusion lowers blood glucose and triglyceride concentrations and improves pancreatic beta-cell function in db/db mice [119], yet endogenous FGF21 concentrations are positively associated with a range of dysmetabolic biomarkers, including blood glucose, insulin, and c-peptide concentrations in humans [120], which may reflect FGF21 resistance or other metabolic feedback processes. Nevertheless, despite its therapeutic potential, the FGF21 protein has rapid renal clearance [121], although the development of FGF21 analogues have helped circumvent this problem [122]. A phase I placebo-controlled clinical trial testing the effects of one such analogue (LY2405319) improved blood lipids and reduced body weight and fasting insulin, but showed no marked impact on blood glucose concentrations [123]. An alternative therapeutic approach might focus on dietary interventions designed to limit sugar-sweetened foods and beverages in carriers of susceptibility variants at FGF21. However, the extent to which each risk allele is associated with increased sugar consumption is negligible at an individual level [115], and is unlikely to be of any value for targeted dietary interventions in clinical practice.

There are many studies exploring interactions between gene variants and diet or exercise in T2D (see [124]), yet most are relatively small, some are cross-sectional, and few findings have been replicated. The largest, most comprehensive epidemiological analysis was performed by the InterAct consortium [107], where interactions between 58 established T2D variants and lifestyle factors were assessed in a nested case-cohort comprised of 12,403 incident cases of diabetes and 16,154 non-diabetics. In analyses focused on a gene-score, no statistically significant interactions were observed with physical activity or dietary habits assessed by a Mediterranean diet score; although several individual variants showed nominal evidence of gene–lifestyle interactions, none was significant after correction for multiple-testing. The largest and most comprehensive clinical trial analysis was performed in the DPP (n = 2843), which focused on a genetic risk score and its interactions with lifestyle intervention (vs. standard of care) [104]. There was no evidence of a gene–lifestyle interaction in diabetes incidence, but there was nominal evidence of an interaction in regression from impaired to normal glucose regulation. Follow-up analyses in the DPP trial, focusing on gene variants previously associated with insulin resistance, found that these did not influence the effects of lifestyle on insulin sensitivity after 1 year of intervention [125]. Overall, there is little robust evidence from epidemiological studies or clinical trials showing that variants previously associated with T2D or insulin resistance modify the effects of lifestyle in diabetes incidence.

TBC1D1 and 4 encode Rab-GTPase-activating proteins that regulate muscle glucose transport and fatty acid oxidation in response to insulin and exercise (see [126]). Abundant animal data implicate coding variation at Tbc1d1 in exercise-related modulation of muscle glucose uptake and weight change, and in vitro perturbation of TBC1D1-transfected mouse myocytes by AICAR (an exercise mimetic) was shown to impact palmitate oxidation [127]; however, it remains unknown whether coding variation at TBC1D1 in humans influences glucose and lipid metabolism. Evidence in humans of how TBC1D4 variation impacts diabetes risk is more concrete. Homozygote carriers of the nonsense p.Arg684ter allele at TBC1D4, common within the Greenlandic Inuit but rare elsewhere, have a roughly 10-fold increased odds of T2D [128]. An exaggerated early insulin response was reported elsewhere in family members with acanthosis nigricans who carried a heterozygous substitution of thymine for cytosine at nucleotide position 1087 in exon 3, resulting in the substitution of a premature stop codon (TGA) for arginine (CGA) at codon 363 [129]. The mechanism appears to involve muscle-selective loss of the long isoform of TBC1D4, leading to a much reduced GLUT4-mediated insulin-stimulated glucose uptake into muscle and marked postprandial (but not fasting) hyperglycemia. What remains unknown is whether exercise-induced AS160 phosphorylation (the protein encoded by TCB1D4) and GLUT4 translocation (a key feature of insulin-dependent and non-insulin-dependent glucose transportation regulated by AS160 phosphorylation) also differ by TBC1D4 genotypes in outbred populations and, if so, whether exercise might be sufficient to offset the impairments in GLUT4 sequestration attributable to TBC1D4 isoform restriction.

The nuclear genome encodes biological processes that are necessary to maintain normal physiological function. The transcription and translation of genetic code can be perturbed by extrinsic and intrinsic environmental stimuli, and by chemical modifications of DNA (broadly termed ‘epigenetics’). In some instances, it may be that diet and exercise interact with epigenetic features, such that the physiological consequences of a lifestyle exposure are determined in part by the presence or absence of an epigenetic mark; in other cases, it may be that diet and exercise causes an epigenetic mark to emerge or disappear.

There is extensive literature on the effects of exercise or diet interventions on biomarkers of gene transcription and translation. The transcription (mRNA production) and translation (protein synthesis) of metabolic regulator genes, particularly those involved in mitochondrial biogenesis and mitochondrial function (e.g., PPARGC1A, AMPK, and SIRT1), have long been the focus of diet and exercise studies, particularly in the context of energy flux and substrate metabolism (see [130]). Although there are many intervention studies designed to test whether perturbation by diet or exercise affects molecular processes, most are not RCTs; this is an important limitation, as the absence of a control arm makes it impossible to determine that the intervention’s effects are not confounded by other unmeasured variables. This problem was highlighted in a study where many genes thought to be ‘exercise-induced’ were shown to change in both control and intervention arms, i.e., effects were not specific to exercise [131].

Most evidence linking lifestyle with changes in gene expression and epigenetic marks originates from cross-sectional cohorts. By contrast, RCTs focused on these issues are exceptionally rare. One of the few RCTs within which the effects of diet on metabolites and methylation marks have been studied is the LIPOGAIN trial [132]. In this double-blind, randomized, parallel-arm intervention trial, young adults (21–38 years; n = 41) were randomized to receive one of two types of high-energy content muffins, supplemental to their habitual diet, for 7 weeks; muffins contained either refined palm oil (rich in the major SFA palmitic acid (16:0)) or refined sunflower oil (rich in the major PUFA linoleic acid (18:2 n–6)). Abdominal subcutaneous adipose tissue was biopsied before and after the dietary intervention, liver fat was assessed using magnetic resonance imaging, and DNA and RNA were extracted for genome-wide methylation and mRNA analyses, respectively. Analysis of 37 of the 41 randomized participants explored changes in gene expression. Three genes, carbonic anhydrase 3, connective tissue growth factor, and aldehyde dehydrogenase 1 family member A1, were determined to be differentially expressed over time and between intervention arms. In a second report from a subset of the LIPOGAIN study (n = 31) [133], multiple changes in DNA methylation of individual genes and CpG sites were reported when focusing on both intervention arms combined or separately within intervention arms. In the within-arm analyses controlling for changes in cell composition and multiple testing, methylation levels changed at 309 sites within the SFA-rich muffin arm and at 4662 sites in the PUFA-rich muffin arm. However, whether expression and methylation were of greater or lesser magnitude between intervention arms was not reported, and there was no control arm.

Metabolites are intermediary compounds produced by naturally occurring enzyme-catalyzed reactions within cells, bearing the parent compound’s characteristics until fully degraded, and generated to control the rate of energy turnover and to perform other functions within cells. The measurement of metabolites in the body (usually in blood or urine) provides a read-out of specific metabolic processes. Examples of metabolites include amino acids within proteins, glucose molecules in glycogen, fatty acids within membrane lipids, and nucleotides in DNA [134].

Metabolomic profiling of diet and exercise signatures has yielded potentially useful tools to objectively assess these traits. For example, differential metabolomic profiles were determined using nuclear magnetic resonance spectroscopy in an observational study of twins discordant for physical activity, such that one twin of each pair reported no structured leisure-time physical activity, whereas the other reported more than 5 years regular leisure-time physical activity [135]. Twins who reported being physically active had lower serum concentrations of isoleucine, α1-acid glycoprotein, and glucose; moreover, the fatty acid profile was less saturated. In the INTERMAP UK cohort [136], ‘diet discriminative’ metabolomic profiles were generated within a randomized cross-over diet intervention trial by profiling urine samples taken following each 72-h diet intervention using proton nuclear magnetic resonance spectroscopy. In both studies, physical activity and diet profiles were validated by demonstrating associations between metabolomic signatures and disease outcomes known to be associated with diet and physical inactivity.

The human microbiota is comprised of four micro-organisms (bacteria, fungi, archaea, and viruses) that live in or on the body and serve multiple essential functions [137]. The ‘microbiome’ is the genetic material derived from all four micro-organisms taken from a given microbial ecosystem.

In a study of 800 young adults, Zeevi et al. [138] assessed glycemic responses to almost 50,000 meals. They found high between-participant variability in response and, using machine learning methods, they were able to derive and validate a prediction algorithm that harnessed data on the intestinal microbiome; they later designed personalized diet interventions and showed that these could be used to regulate glycemic excursions following meals and robust changes in the configuration of gut microbiota.