From nutrigenomics to personalizing diets: are we ready for precision medicine?

Since the completion of the human genome project >15 y ago, a plethora of genetic variants have been identified that affect people's susceptibility to disease (1). With a growing understanding of the genetic basis of diseases and of their risk factors, the hope and expectation is that genetics will soon revolutionize health care. Knowing a patient's genome would enable us to predict risk of future disease more accurately and to prescribe personalized treatment and prevention strategies, as opposed to the traditional “one-size-fits-all” approach. More than 3 y ago, President Obama launched the Precision Medicine Initiative as a bold new effort to support researchers, patients, and health care providers to work together towards the development of individualized care that takes into account people's unique characteristics. Although environment and lifestyle were both considered when personalizing care, a strong emphasis was placed on the importance of people's genomes.

Long before the Precision Medicine Initiative, however, online genomic companies had already been offering genetic testing directly-to-consumers (DTC). Their business model is straightforward: customers send saliva and pay a few hundred dollars for DNA extraction and for the genotyping of thousands of genetic variants across their genome. Four to 6 wk later, they receive a “personalized” report that informs them of their genetic predispositions—all without involving health care providers or genetic counselors. The conditions covered by the DTC companies vary from specific life-changing diseases (e.g., Parkinson disease, phenylketonuria, celiac disease) to general traits of appearance, senses, and behavior (e.g., eye and hair color, wake-up time, taste preference). A key question, however, is whether there is sufficient scientific evidence for “personalized” recommendations.

Several DTC companies focus on diet, nutrition, physical performance, and fitness. They claim that, based on the customers’ genotype data, they can design “genetically matched diets” to help them lose weight more easily, determine what nutrients their body favors during exercise, and what type of training is most effective, among other things. However, scientists have urged caution because, for many of these outcomes, research has not generated the evidence to support such bold claims and customers do not have sufficient insight to interpret the reports (2–4).

Most popular DTC genetic testing companies aim to provide personalized recommendations on common multifactorial outcomes (e.g., to lose weight, improve health, nutrition, fitness, or performance), which are only partially determined by genetic variation. Lifestyle, environment, sociodemographic factors, and other nongenetic factors are equally—or sometimes even more—important. Thus, even the best genetic test will never predict a multifactorial outcome perfectly, nor will it allow for a personalized tailoring of treatment strategies. In addition, the genetic contribution is polygenic, meaning that multiple—sometimes hundreds or more—genetic variants play a role, all with small effects. Although detailed information is hard to find, most DTC companies seem to base their recommendations on only 1, or just a few, genetic variants, which can result in incomplete or potentially misleading reports. Furthermore, a common misconception is that a genetic variant can predict a disease when it is associated with the disease. However, many genome-wide association study–identified genetic variants show highly significant associations with a trait, but explain only a fraction of the susceptibility to a disease. Therefore, it is no surprise that the predictive ability of 1, or even 100, variants combined is very limited, and insufficient to inform individuals about their future risk of disease. Finally, another common mistake is made when results from cross-sectional studies are interpreted as if they were longitudinal. For example, a genetic variant may be associated with higher body weight when consuming a high-fat diet (cross-sectional), but this does not mean that a person carrying that given variant will lose weight when they reduce their fat intake (longitudinal). Evidence from longitudinal studies is critical for genetic testing companies, in particular when their focus is on improving (i.e., changing over time) their customers’ health, fitness, performance, etc. Although there are rather few studies that examine the genetic underpinning of changes in response to dietary intervention, The American Journal of Clinical Nutrition’s current edition reports on 2 studies that did exactly that (5, 6).

The first study (5) aimed to develop a genetic risk score (GRS) to accurately predict whose triglyceride (TG) concentrations would decrease in response to omega-3 fatty acid (n–3 FA) supplementation. The 208 participants of the Canadian Fatty Acid Sensor population were divided into “responders,” whose TG concentrations decreased after a 6-wk n–3 FA supplementation, and “non-responders,” whose TG concentrations increased or did not change. The GRS, representing a person's innate likelihood to benefit from n–3 FA supplementation (i.e., to be a “responder”) or not, explained almost 50% of the variation in TG response and had an almost perfect discriminatory ability (area under the receiver operator curve = 0.94). Specifically, knowing a person's GRS allows us to determine with great accuracy whether they will benefit from a 6-wk n–3 FA supplementation or not, at least among the Fatty Acid Sensor participants. However, when the same GRS was used in the European FINGEN study, in which 310 participants underwent a 12-wk n–3 FA supplementation, the findings were remarkably different. The GRS explained only 3.7% of variation in TG response and did not allow identifying who would benefit from supplementation (area under the receiver operator curve = 0.64). This shows that a genetic test may work in one population but not necessarily in another population of different sex, age, ancestry, diet, baseline values, etc.

The second study (6) examined whether the beneficial effect of dietary linoleic acid (LA) on health is attenuated by people's Fatty Acid Desaturase 1(FADS1) genotype (rs174550). Cross-sectionally, in 1337 Finnish men of the Metabolic Syndrome in Men (METSIM) study, higher plasma LA concentrations were associated with lower glucose concentrations, but only among carriers of the rs174550 T-allele, not in CC homozygotes. In a subsequent 4-wk dietary intervention, the effects of an LA-enriched diet on a range of metabolic markers were examined in 26 individuals with the TT genotype and 33 individuals with the CC genotype. Although changes in lipid mediator and inflammatory marker concentrations in response to the 4-wk LA-enriched diet differed by FADS1 genotype, the differences between genotypes were generally small.

Both studies support the notion that a person's genotype may influence their response to changes in diet, but the results do not provide evidence that genotype data can be used to predict their response or tailor their diet. For a genetic variant or a GRS to be useful in precision medicine, their effect sizes would need to be large and their penetrance high. They would need to be studied in the context of people's full genetic background, not in isolation. Furthermore, they should also be studied in a longitudinal setting (to study response to intervention) and, importantly, observations will need to be validated across a diversity of populations. Given the importance of nongenetic factors (lifestyle, environment, demographics, etc.) in many of the health outcomes covered by the DTC companies, precision medicine that focuses solely on people's genome is doomed to fail. Inaccurate “genotype-specific” recommendations may be misleading and have potentially negative consequences. It is clear that to predict disease accurately, or to personalize treatment, an integrated approach that accounts for people's unique genetic and nongenetic characteristics is needed. Research efforts such as the Accumulating Data to Optimally Predict Obesity Treatment project, which aims to understand interindividual variability in response to obesity treatment (7), are needed for other common disease outcomes. In the meantime, it seems that genotype-based recommendations from DTC genetic testing companies are likely as effective as the “one-size-fits-all” recommendations.

Notes

Supported by National Human Genome Research Institute grant U01HG007417 and National Institute of Diabetes and Digestive and Kidney Diseases grants R01DK110113 and R01DK107786.

Abbreviations used: DTC, directly-to-consumers; GRS, genetic risk score; LA, linoleic acid; n–3 FA, omega-3 fatty acid; TG, triglyceride.