Lifestyle and precision diabetes medicine: will genomics help optimise the prediction, prevention and treatment of type 2 diabetes through lifestyle therapy?

The major developments in genomic technologies and their application to large, well characterised collections of samples have led to the generation of extensive new knowledge about disease biology. This has inspired new avenues for type 2 diabetes prevention, treatment and cure that are inherent to the concept of precision medicine. According to the National Research Council [1], precision medicine is not intended to involve the complete personalisation of medical devices and therapies; instead, it should focus on the classification of ‘individuals into subpopulations that differ in their susceptibility to a particular disease, in the biology and/or prognosis of those diseases they may develop, or in their response to a specific treatment’ with the expectation that ‘preventive or therapeutic interventions can then be concentrated on those who will benefit, sparing expense and side effects for those who will not’. By this and most other definitions, precision medicine focuses on applying biomarker technologies to the individual patient to help improve prediction and assessment of: (1) risk-factor susceptibility; (2) disease stratification; (3) prognosis; and (4) treatment response.

Our understanding of type 2 diabetes pathobiology has improved dramatically recently, owing largely to a quantum leap in human genome sequencing, precipitated by ground-breaking achievements made in the preceding century, including mapping of the Drosophila genome [2] and the structural characterisation of DNA [3]. These were the foundations for human genome sequencing [4, 5] and affordable technologies for high-resolution characterisation of the genome, metagenome, epigenome, transcriptome, proteome, and metabolome. Combined with novel bioinformatics and the emergence of large global collaborative networks, exciting possibilities have emerged for the prediction and prevention of disease in ways that are more personal and precise than ever before.

Genetic variation is quite literally the starting point of the biological cascade that underpins phenotypic expression (known as the ‘central dogma of molecular biology’) [6]. But for complex diseases like type 2 diabetes, genetics is by no means the absolute determinant, and thus an enormous amount of downstream work remains before we will adequately understand how genetic and lifestyle factors (e.g. nutrition, exercise, medications and stress) work jointly to affect gene transcription and translation, and phenotypic expression (Fig. 1). The model is thus one of primers and catalysts: environmental triggers in the context of genetic predisposition.

Given that significant changes in human genetic variation manifest over many generations, the consensus is that the global surge in type 2 diabetes prevalence is caused predominantly by the rapid and widespread adoption of obesogenic lifestyles. Metabolic dyshomeostasis is a common consequence of unhealthy lifestyles, driven by disturbed substrate production and/or metabolism in the liver, skeletal muscle and adipose tissue, or by interfering with the synthesis, secretion or action of insulin. However, diabetes pathophysiology is complex and heterogeneous with multiple feedback loops, such that people vary in susceptibility to risk factors and response to therapies, and the molecular defects that cause the disease in a given patient are rarely known. Nevertheless, the measurement or prediction of primordial factors might facilitate more effective diabetes prevention if they helped determine the specific risk factors to which a person is susceptible and the therapies they are likely to respond well to.

In most people at risk of or with type 2 diabetes, prognosis can be improved by enhancing peripheral insulin sensitivity. Although oral glucose-lowering agents are often used for this purpose, reducing energy intake and increasing non-resting energy expenditure (physical activity) are highly effective first line therapeutic options, with associated reduced lipid content in or around adipose, liver and muscle tissue being pivotal to this process.

Although reduction of energy intake and an increase in physical activity can both cause weight loss, they do so through contrasting states of low and high metabolic turnover respectively, which involve the activation of different molecular pathways and processes that encompass both common and unique corresponding health benefits. Accordingly, some patients will respond well and others poorly to the same intervention, with response being determined in part by individual biology, but also by psychosocial factors that influence adherence and perceived success.

Precision medicine in type 2 diabetes is very much at a theoretical stage, particularly as it relates to personalised lifestyle therapy. However, its successes in other diseases and with pharmacotherapy offer a glimpse of how tailored lifestyle advice for type 2 diabetes prevention, guided by personal genomic data, might be achievable (Fig. 2 and text box: Precision lifestyle medicine in type 2 diabetes):

Early twin and family studies showed that response to diet and exercise interventions vary to a significantly greater extent between sibships and pedigrees than within them, suggesting a heritable component to some treatment response phenotypes [24‐26]. A 100-day overfeeding protocol conducted in 12 pairs of monozygotic (MZ) male twins showed, for example, that the gain in fat mass was roughly three times more similar within the MZ twin pairs than between non-twins [26]. Elsewhere, a study conducted to investigate the response to a negative energy balance in seven pairs of MZ male twins found that the variance in change in fat mass was 14.1 times more similar within than between twin pairs [25].

When used to model interactions between common genetic variants and lifestyle exposures, data from observational studies can generate hypotheses relevant to treatment response and risk-factor susceptibility. However, the inherent limitations of epidemiology (chance, bias, confounding), the difficulties in accurately and precisely assessing phenotypes and lifestyle behaviours in free-living populations, as well as challenges that specifically hinder the estimation of interaction effects in observational data (e.g. heteroscedasticity and scale dependency) necessitate caution when interpreting the causal relevance of observational data. In interaction studies, the imperative of replication has been especially difficult to achieve owing to winner’s curse, heterogeneous study designs, environmental idiosyncrasies, etc. (see [27]). Nevertheless, a complete absence of replication in the face of concerted, adequately powered attempts undermines the value of the original findings, as they likely reflect idiosyncratic effects or false-positives.

There is little extensively replicated epidemiological evidence of gene–lifestyle interactions, with the exception of those at the FTO locus (see text box: What we know about the FTO–physical activity interaction). Soon after the finding that FTO variation affects obesity risk, data emerged from two epidemiological studies showing that FTO variation may modify the relationship between physical activity and estimates of adiposity in European [28, 29] and North American [30] adults. An analysis of clinical trial data from the Diabetes Prevention Program (DPP) found no evidence that FTO modified the effects of lifestyle intervention on weight loss, although there was weak evidence that an interaction might affect subcutaneous adipose mass [31]. Many studies followed, but with mixed results; therefore, we undertook a large meta-analysis (N = 220,000) to test the hypothesis, which confirmed the presence of an interaction [32]. Those findings were recently extended in the UK Biobank [33], where statistically robust interaction effects were reported between the FTO variant and physical activity, frequency of alcohol consumption, sleep duration, diet and salt consumption. Cross-sectional analyses focusing on lifestyle interactions with genetic risk scores comprised of other obesity-associated loci have also been widely replicated, although the extent to which those interactions are driven by FTO is rarely described.

While the epidemiological data seem compelling, it is predominantly cross-sectional and evidence of gene–lifestyle interactions in large observational datasets may be biased or confounded. Hence, epidemiological evidence of gene–lifestyle interactions should not only be replicated but also supported by other types of causal evidence before exploiting the findings for the purposes of genotype-guided lifestyle prescription. There are some functional data positioning FTO as a plausible candidate for lifestyle interactions (e.g. FTO encodes a 2-oxoglutarate oxygenase that is highly expressed in hypothalamic nuclei in mice and humans [34], and phosphocreatine and inorganic phosphate levels have been shown to recover more rapidly following exhaustive exercise in FTO (rs9939609) risk allele carriers [35]), but the discovery of long-range activation of IRX3 and IRX5 by FTO [36, 37] suggests that simple interpretations about how FTO and lifestyle interact are unlikely to be accurate.

Nevertheless, while understanding the functional basis of interactions is desirable, the absence of this knowledge does not preclude using evidence of gene–lifestyle interactions for precision medicine, whereas demonstrating cause and effect in trials is essential. A recent systematic review of selected published trial data concluded that FTO variants modify weight loss in response to lifestyle interventions [38]. However, the meta-analysis of published data for gene–lifestyle interactions may be biased [39], and a subsequent meta-analysis of clinical trial data, which adopted a more rigorous and more inclusive approach involving de novo, standardised interaction analyses, found no evidence of interaction between FTO and lifestyle [40]. A recent clinical trial analysis (DPP and Look AHEAD [The Action for Health in Diabetes] studies) that focused on the spectrum of BMI-associated variants reached similar conclusions for FTO and almost all other variants assessed; the exception was for MTIF3, which showed evidence of gene–lifestyle interactions in the trials [41], as well as in epidemiological cohorts [42].

Although the starkly opposing conclusions drawn about FTO from the observational and clinical trial data appear contradictory, they are potentially reconcilable. For example, the observational analyses [32] are cross-sectional and may reflect the modifying effects of very long-term lifestyle exposures and outcomes, whereas the clinical trials [40] are prospective and confined to a relatively short intervention period (ranging from 8 weeks to 3 years). Moreover, the epidemiological studies are large (N = 100,000–200,000) and apparently adequately powered, whereas the trials analysis is an order of magnitude smaller (N = 9563), may have been underpowered to detect interaction effects, and was conducted in very selected populations compared with the observational studies (Fig. 3). Nevertheless, the absence of an interaction effect in the trials indicates that, for the benefit of enhancing weight loss for diabetes prevention, there may be little clinical value in tailoring common lifestyle interventions to FTO genotype.

Determining the causal effects of lifestyle, even in randomised controlled trials, is much more challenging than for drug therapies, not least because most lifestyle interventions cannot be masked and no placebo exists. Thus, a participant’s behaviours during a lifestyle trial that affect the trial’s outcomes may change as a consequence of the intervention in ways that are not expected or measured. This phenomenon was first described in a study of older men and women (58–78 years old) who underwent an 8 week supervised aerobic training programme (cycling three times per week) [43]. Objective assessments of total energy expenditure were made using doubly labelled water, resting metabolic rate by respiratory gas exchange, and physical activity energy expenditure using the latter values combined with information about sleep and exercise. The authors found that despite an increase in exercise energy expenditure of 628 kJ/day (150 kcal/day), non-exercise-activity energy expenditure decreased proportionately and no overall change in total energy expenditure was hence observed.

When one considers that exercise interventions deployed in clinical trials typically occupy about 150 mins per week, or about 1.5% of the total time, it is possible that compensatory behaviours underlie much of the apparent heterogeneity in response. Even in tightly controlled inpatient studies, unintended variations in the participants’ behaviours can hinder data interpretation. For example, in one of the studies cited above, 84 days of overfeeding with restricted physical activity resulted in an average weight gain of 8.1 kg (range: 4.3–13.3 kg) [26]. The very lowest and highest weight gains suggest that the diet and exercise regime was not strictly followed, as a weight gain of 4.3 kg with this level of overfeeding accounts for only ∼40% of the predicted weight gain for that participant and would require energy expenditure equivalent to running ∼17–29 km each week during the intervention. By contrast, a weight gain of 13.3 kg would require complete indolence during the protocol. While individual biological variation in energy metabolism might explain some of these differences in weight gain, it seems probable that lack of adherence to the study protocol is the predominant explanation.

In current research and practice, lifestyle behaviours are frequently assessed through self-report methods, despite some key limitations [44]. However, numerous wearable technologies exist that are relatively inexpensive, suitable for long-term monitoring, and valid, albeit also with important caveats [45]. Nevertheless, modern wearables, used in combination with self-report methods have tremendous potential to characterise health-related behaviours and exposures, as the continuous assessment of movement, sleep, temperature, blood glucose and other relevant factors is possible without apparently impacting behaviour [46]. The use of such devices in epidemiology and clinical trials is likely to be necessary to delineate biological drivers of risk-factor susceptibility and treatment response from other sources of error and bias that cause heterogeneity. The continuous assessment of quantitative phenotypes germane to a clinical trial’s outcomes would also help overcome a further major limitation of some lifestyle trials that only assessed outcomes at enrolment and the trial’s end. In those settings, regression dilution, which occurs when changes in quantitative traits are inferred from too few data points, generates error. Thus, continuous (or frequently repeated) assessments of the trial’s quantitative outcomes would further help isolate response variability from error. In a recent eloquent study that focused on personalised diets, Zeevi and colleagues collected detailed diet data and objectively assessed physical activity and blood glucose in 800 Israeli adults during a 1 week observational phase. In combination with gut microbiota data, the authors were able to predict individual postprandial glucose excursions to specific foods and designed personalised diet interventions that improved glucose control [47]. This approach had the added benefit of allowing the background heterogeneity in the participants’ lifestyles to be factored into the intervention design, further reducing error.

Although evidence of specific genetic loci that modulate risk-factor susceptibility and treatment response in type 2 diabetes is weak, the rationale, which is supported by data from twin and family studies, remains strong, motivating continued research in this area. For example, the National Institutes of Health’s (NIH’s) Common Fund initiative called the Molecular Transducers of Physical Activity (http://​commonfund.​nih.​gov/​MolecularTransdu​cers), is a multimillion dollar funding programme focused on defining ‘optimal physical activity recommendations for people at various stages of life’ and developing ‘precisely targeted regimens for individuals with particular health needs.’ Nevertheless, this topic is clearly challenging, and to realise the vision of the NIH and others will likely require new study designs and analytical methods that overcome the major barriers to precision lifestyle medicine in type 2 diabetes.