Despite its widespread use as the first-line agent for the treatment of type 2 diabetes, it has become clear that metformin does not work optimally for everyone. Elucidating who are the likely metformin responders and non-responders is hampered by our limited knowledge of its precise molecular mechanism of action. One approach to achieve the related goals of stratifying patients into response subgroups and identifying the molecular targets of metformin involves the deployment of agnostic genome-wide approaches in cohorts of appropriate size to attain sufficient statistical power. While candidate gene studies have shed some light on the role of genetic variation in influencing metformin response, genome-wide association studies are beginning to provide additional insight that is unconstrained by prior knowledge. To fully realise their potential, much larger samples need to be assembled via international collaboration, preferably involving the academic community, government and the pharmaceutical industry.
KeywordsGenome-wide association studies Metformin Pharmacogenetics Review Single nucleotide polymorphisms Type 2 diabetes
Diabetes Prevention Program
Genetics of Diabetes and Audit Research Tayside Study
Genome-wide association studies
Multidrug toxin and extrusion
Metformin Genetics (Consortium)
Organic cation transporter
Plasma monoamine transporter
UK Prospective Diabetes Study
Metformin is the undisputed queen among type 2 diabetes drugs. She is enthroned atop all professional society algorithms as the first-line drug to be prescribed as soon as the diagnosis is made [1, 2, 3], in the absence of contraindications such as severe renal or hepatic insufficiency. This is because among the 12 available drug classes for type 2 diabetes, it alone embodies the three cardinal virtues expected of good medicines: it is eminently safe, cheap and effective [4, 5]. Indeed, metformin therapy is well tolerated, with mild gastrointestinal side effects in ~30% of patients; only ~5% of users stop medication because of severe intolerance [6, 7]. In general, metformin is highly efficacious and is one of the few type 2 diabetes oral agents that promote weight loss rather than weight gain [6, 8, 9]. In addition, this weight loss is particularly beneficial and similar to exercise-induced weight loss, in that it does not reduce resting energy expenditure ; it is also sustained and dependent on adherence . Thus, it has become one of the most commonly used medications for type 2 diabetes worldwide, with more than 48 million prescriptions filled in the USA in 2010. As described elsewhere in this issue of Diabetologia, metformin is also used for diabetes prevention  and the treatment of polycystic ovary syndrome , and there is hope that its use will expand to other indications, such as cancer treatment and/or prevention .
Despite all of these favourable qualities, metformin is not a panacea. Clinical practice suggests, and published trials confirm, that metformin monotherapy is often insufficient in achieving glycaemic control. Very frequently, escalation of therapy is required  and as many as 21% of patients initiated on metformin treatment fail to meet glycaemic goals in the first 5 years of therapy . The failure rate exceeds 50% in youth with newly diagnosed type 2 diabetes, a group in whom diabetes incidence is rising at alarming rates . The long-term follow-up of the Diabetes Prevention Program (DPP) showed that participants randomised to the effective metformin preventive intervention group continued to develop diabetes over time, even while on treatment and despite an adherence rate near 80%, heightening the concern that metformin may only delay (by about 12 months on average), rather than prevent, diabetes incidence .
The metformin scenario thus illustrates a key barrier in modern therapeutics: by focusing on the ‘average patient’, as typified by the ideal responder in a successful clinical trial, we neglect to capture and describe the full range of pharmacological responses and thereby fail to identify those in whom a specific therapy might not be appropriate. Precious time might be wasted in trying to achieve glycaemic control in people in whom such a goal might be unreachable, and some patients may be unnecessarily exposed to an unacceptable level of side effects. To realise the promise of ‘precision medicine’ we need to transcend the law of averages and begin to characterise the subgroup of patients who would benefit from an alternative to metformin upfront, either because they do not tolerate it or because they fail to achieve glycaemic targets.
Why are we so limited in our ability to define a group of metformin non-responders a priori? In part, this is due to our ignorance around the molecular target of metformin and its precise mechanism of action. While its physiological effects in humans have been described and we are gaining new insights on the cellular pathways affected by the drug , we are still unclear as to the specific molecular entity that metformin physically engages with to modulate metabolism. This knowledge gap hampers the design of experiments aiming to describe likely non-responders but it also presents an opportunity to conduct agnostic screens that might yield fundamental discovery.
The genetic approach as a potential solution
Some of the individual differences that underlie the variation in response to metformin are likely genetic in nature. For example, epidemiological studies suggest that ethnic disparities exist in metformin response , and a genetic component is likely to underlie some of these disparities. Nevertheless, environmental factors may also be at play, either directly or by interacting with the genetic background; to guide further exploration it would be useful to quantify the relative contribution of each. Typically, this is done via classical heritability studies but these present obvious challenges in finding sufficient numbers of close relatives treated with metformin and phenotyped for therapeutic response. More recently, the availability of dense genome-wide genotyping in cohorts in which metformin response has been quantified allows investigators to measure the degree of genetic relatedness with adequate precision and estimate its contribution to the variance in the trait of interest .
Heritability estimates of various measures of glycaemic response to metformin obtained from a genome-wide association study in GoDARTS
Heritability (h 2 )
Absolute reduction in HbA1c a
Proportional reduction in HbA1c b
Adjusted reduction in HbA1c c
Achieved target HbA1c d
Prior to the development of GWAS approaches to interrogate the entire genome, investigators were limited by pre-existing biological knowledge; for metformin, research had centred on the cellular transporters that controlled its metabolism, which could now be queried via candidate gene association studies. If robust associations were found they might subserve the classification of patients into responders and non-responders, but no new insights on metformin’s targets would be gained; the panorama would be radically altered with the advent of GWAS. The next sections describe these efforts chronologically.
First phase: candidate gene studies
The most studied transporter regarding the impact of genetic variation on metformin action has been OCT1. The gene encoding OCT1, SLC22A1, is highly polymorphic, with a number of coding missense SNPs that affect its activity [36, 37, 38, 39]. The Giacomini group pioneered some of these studies, showing that the presence of at least one of four reduced-function variants (R61C/rs12208357, G401S/rs34130495, M420del/rs72552763 and/or G465R/rs34059508) attenuates the glycaemic effects of a short course of metformin , likely by an impairment of intracellular transport resulting in higher maximal and AUC plasma concentrations . However, in a large retrospective observational clinical cohort of 1531 individuals assembled by GoDARTS, the two most common reduced-function SNPs (R61C and M420del) were not associated with four different measures of metformin glycaemic response, namely initial HbA1c reduction, odds of achieving a target HbA1c of <7%, average HbA1c on metformin monotherapy, or hazard of monotherapy failure, defined by the addition of a second agent by the treating clinician . Whether this is a true negative result, or a false negative finding resulting from insufficient statistical power or the limitations inherent to observational studies (where treatment decisions are left to the discretion of the treating provider), is not clear. In the prospective South Danish Diabetes Study of 105 participants , both trough steady-state metformin concentrations and the change in HbA1c after 6 months were lower with increasing number of reduced-function alleles at the four reduced-function variants, R61C, G401S, M420del and G465R . A more recent study using radioisotope labelling has demonstrated that accumulation of metformin in human hepatocytes is diminished in carriers of the M420del and R61C variants in SLC22A1, without changes in circulating levels of the drug . A study of side effects by the GoDARTS investigators in 251 metformin-intolerant individuals and 1915 metformin-tolerant individuals showed that the presence of two or more reduced-function alleles at R61C, C88R/rs55918055, G401S, M420del or G465R increased the odds of metformin intolerance by more than twofold, presumably by inducing accumulation of metformin in enterocytes ; the effect was additive with the concomitant use of OCT1-inhibiting medications, increasing the odds by fourfold. An initial report of association of the SLC22A1 intronic SNP rs622342 with metformin response in a retrospective Rotterdam study of 102 patients  has not been corroborated by subsequent independent publications [43, 47, 48]. Similarly, no associations have been found between reduced-function variants in OCT2 and metformin response [43, 47, 48, 49]. The Rotterdam group found an association of the intronic SNP rs2289669 in SLC47A1 (encoding MATE1) with metformin response, with the minor allele enhancing the reduction in HbA1c; though this finding is supported by similar observations by the DPP  and Tkáč et al , it was not seen in the South Danish Diabetes Study  nor in a large meta-analysis recently conducted by the Metformin Genetics (MetGen) Consortium . In conclusion, the most solid evidence seems to support a role for coding missense variants in OCT1 with metformin response and/or intolerance.
Second phase: GWAS
Conclusions and future directions
To tailor the use of metformin to the most appropriate segment of the population, an improved understanding of its molecular mechanism of action is necessary. Given the documented heritability of metformin response in humans, agnostic genomic searches can facilitate both goals, but this requires the assembly of very large cohorts in order to detect likely modest effects. The MetGen Consortium represents one step in the right direction. Despite its most recent GWAS of >10,000 samples, the combined impact of the ATM and SLC2A2 loci on metformin response was shown to be minimal, suggesting that many other genetic determinants of metformin action remain to be discovered, requiring much larger sample sizes and greater precision in pharmacogenetic measurements, such as drug doses, participant adherence and glycaemic response. However, most samples of participants treated with metformin and followed for glycaemic endpoints in a controlled fashion reside in the standard-of-care arms of randomised clinical trials for new type 2 diabetes agents, typically housed in the realm of pharmaceutical companies. Though non-disclosure by the pharmaceutical industry of pharmacogenetic outcomes for proprietary compounds might be defensible, it would be in everyone’s best interest to share samples and data of individuals randomised to the active standard comparator (usually metformin), as the identification of metformin non-responders would open the door to a new patient population for whom newer agents could be marketed. While guarding against potential heterogeneity, a pharmacogenetic study that used every possible sample of a person treated with metformin, whether in healthcare systems, publicly funded studies or industry-sponsored clinical trials, would surely reveal new associations, leading to a richer understanding of metformin action, the potential design of superior agents, and better categorisation of patient subgroups based on their likelihood to benefit from this time-honoured treatment.
In practice, once robust pharmacogenetic associations are discovered and confirmed, they can be aggregated into genetic risk scores that explain a substantial proportion of the variance in glycaemic response or the appearance of side effects. They could be included into multi-trait genotyping arrays that capture all known clinically actionable genetic variants for most common diseases and approved drugs. This ‘megachip’ would only need to be deployed once in the lifetime of an individual at an affordable cost and the information could become part of that person’s electronic medical record. Statistical algorithms that define the likelihood of response could be created, tested, refined and automatically triggered once a prescription order is initiated; decision support tools elaborated by experts would then inform the clinician, at the point of care, whether this patient is a good candidate for the selected agent. The methods and expertise exist to realise this vision but it will not come to fruition unless we generate the required knowledge base, the data stand rigorous scrutiny and cost-effectiveness analyses demonstrate that the benefit of clinical outcomes outweighs the expense incurred.
JCF is a Massachusetts General Hospital Research Scholar. Work in the author’s laboratory is supported by NIDDK R01 DK072041, NIDDK U01 DK105554, NIGMS R01 GM117163, NIDDK R01 DK105154, and NIDDK K24 DK110550.
Duality of interest
JCF has received consulting honoraria from Merck, Boehringer-Ingelheim and Intarcia Therapeutics.
JCF conceived and wrote the entire manuscript, and is the sole contributor to this paper.
- 1.Nathan DM, Buse JB, Davidson MB et al (2009) Medical management of hyperglycaemia in type 2 diabetes mellitus: a consensus algorithm for the initiation and adjustment of therapy: a consensus statement from the American Diabetes Association and the European Association for the Study of Diabetes. Diabetologia 52:17–30CrossRefPubMedGoogle Scholar
- 3.Inzucchi SE, Bergenstal RM, Buse JB et al (2015) Management of hyperglycaemia in type 2 diabetes, 2015: a patient-centred approach. Update to a position statement of the American Diabetes Association and the European Association for the Study of Diabetes. Diabetologia 58:429–442CrossRefPubMedGoogle Scholar
- 12.Aroda VR, Knowler WC, Crandall JP, Group DPPR (2017) Metformin for diabetes prevention: insights gained from the Diabetes Prevention Program/Diabetes Prevention Program Outcomes Study. Diabetologia doi: 10.1007/s00125-017-4361-9
- 13.Sam S, Ehrmann DA (2017) Metformin therapy for the reproductive and metabolic consequences of polycystic ovary syndrome. Diabetologia doi: 10.1007/s00125-017-4306-3
- 14.Heckman-Stoddard B (2017) Repurposing metformin for the prevention of cancer and cancer recurrence. Diabetologia doi: 10.1007/s00125-017-4372-6
- 15.Turner RC, Cull CA, Frighi V, Holman RR (1999) Glycemic control with diet, sulfonylurea, metformin, or insulin in patients with type 2 diabetes mellitus: progressive requirement for multiple therapies (UKPDS 49). UK Prospective Diabetes Study (UKPDS) Group. JAMA 281:2005–2012CrossRefPubMedGoogle Scholar
- 18.Rena G, Hardie DG, Pearson ER (2017) The mechanisms of action of metformin. Diabetologia doi: 10.1007/s00125-017-4342-z
- 21.GoDARTS and UKPDS Diabetes Pharmacogenetics Study Group, Wellcome Trust Case Control Consortium, Zhou K et al (2011) Common variants near ATM are associated with glycemic response to metformin in type 2 diabetes. Nat Genet 43:117–120Google Scholar
- 44.Sundelin EI, Gormsen LC, Jensen JB et al (2017) Genetic polymorphisms in organic cation transporter 1 attenuates hepatic metformin exposure in humans. Clin Pharmacol Ther doi: 10.1002/cpt.701
- 46.Becker ML, Visser LE, van Schaik RH, Hofman A, Uitterlinden AG, Stricker BH (2009) Genetic variation in the organic cation transporter 1 is associated with metformin response in patients with diabetes mellitus. Pharm J 9:242–247Google Scholar