In this study, we have defined the response to metformin in 1,531 patients with type 2 diabetes. Contrary to our hypothesis, we show that two established loss-of-function variants in SLC22A1, encoding OCT1, did not impair initial glycemic response to metformin, the mid-term A1C control, or the rate of metformin monotherapy failure.
In our models, the A1C reduction is primarily determined by the baseline A1C. In addition, metformin response is increased with enhanced adherence and reduced creatinine clearance, consistent with the renal clearance of metformin. These covariates would act to increase availability of metformin, and this demonstrates the utility of these models in detecting factors such as genotypes that might alter drug availability. The nonsignificant but negative correlation of the dose with response is likely to reflect a higher dose given to those who seem less likely to respond. We show no genotypic effect of the two SLC22A1
variants tested in any model. Our study was adequately powered to detect a clinically useful difference in metformin response (QUANTO,http://hydra.usc.edu/GxE/
). Assuming an additive model for the 420del genotype (minor allele frequency [MAF] 0.20) and 1,500 subjects, the linear regression analysis had a 90% power (α = 0.05) to detect a difference in A1C reduction of 0.15% per copy of the variant allele. For the R61C genotype (MAF 0.07), the analysis had an 80% power to detect a difference in A1C reduction of 0.2% per copy of the variant allele.
Our study design is observational, and the results could be prone to bias. In particular, the decision to commence metformin and the rate of dose titration were determined by the patients and their physicians. We had no direct measure of intolerance. However, we have shown that there is no variation in genotype frequency in those who did not achieve 6 months of metformin prescriptions compared with those who were included in the final cohorts; thus, any potential bias due to drug withdrawal is minimal. A further limitation is the fact that we are unable to measure serum metformin concentrations or to look at change in other measures such as insulin sensitivity. However, we believe that the study sample size of 1,531 patients provides adequate power to see a true genotypic effect, and we have used a similar approach successfully to show that TCF7L2
variants affect sulfonylurea response (13
In the treatment-naive group, our logistic regression found an increase of treatment success in the C-allele carriers of an R61C variant, though only of nominal significance without multiple test correction. It is worth noting that those C-allele carriers also showed a consistent trend to better metformin response in the other A1C outcome models: higher A1C reduction in the linear regression model and lower mean A1C outcome from the repeated-measurement analysis. This effect is in the direction opposite that in the findings of Shu et al. and, although not definitive, would certainly suggest that our inability to show loss of metformin response does not simply reflect lack of power.
In mice, Slc22a1 was shown to be an important transporter of metformin into the liver and, to a lesser extent, into the small intestine (7
). Subsequently, compared with wild type, mice lacking Slc22a1 have reduced lactic acidosis (14
) and loss of glucose reduction (10
) when treated with metformin. Shu et al. (10
) have also identified a number of rare and relatively common SLC22A1
polymorphisms in humans and demonstrated that 12 human, glucose tolerant carriers of these variants showed a reduced efficacy of metformin in lowering glucose excursions in an oral glucose tolerance test compared with that in eight control subjects. It is interesting that in a subsequent pharmacokinetic study, the serum metformin concentrations are increased in these same individuals (15
), which in mice is explained by decreased hepatic clearance. In contrast to the findings of the study by Shu et al., in a small study of patients with type 2 diabetes defined by response or nonresponse to metformin, variation in SLC22A1
was not consistently associated with response (16
The difference between our study and that of Shu et al. could be explained by key differences in the sample ascertainment and response models. We have studied patients with type 2 diabetes and in a real-world setting assessed A1C reduction in association with prolonged oral metformin use; this compares with a more controlled study in a few subjects with normal glucose tolerance who are given just two doses of oral metformin and assessed by response to oral glucose. It is generally accepted that metformin primarily acts to suppress hepatic glucose output (17
) and, therefore, that fasting glucose may be a better measure of efficacy than the dynamic response to oral glucose used by Shu et al. There is an ongoing controversy over the pharmacological mechanism of metformin. Metformin has been shown to increase non–insulin-mediated glucose clearance, and this effect is responsible for as much glucose reduction as the effect on hepatic glucose production (17
). Furthermore, metformin has also been shown to augment insulin-mediated glucose uptake into the periphery (18
) and decrease glucose absorption from the gut (19
). Given that carriers of the loss-of-function SLC22A1
variants had increased serum metformin concentrations (15
), this may augment the nonhepatic actions of metformin—an effect that might be more apparent in patients with diabetes compared with control subjects. This is an area where further study is required.
Although in this observational study there is no apparent reduction in the ability of metformin to lower A1C in patients with type 2 diabetes carrying the R61C or 420del loss-of-function polymorphisms in SLC22A1,
this does not rule out an effect of SLC22A1
variation on metformin response. A more complete investigation of SLC22A1
variants would be required to fully assess the effect of the gene on metformin response given that variants with a more severe loss of function have been described (20
). However, we have studied the two most common loss-of-function variants described in Europeans, and any additional known or unknown variants would be very rare. The observational pharmacogenetic approach that we have used in this study requires a large sample size, and the study of rarer loss-of-function polymorphisms would lack power in our current dataset. A more carefully controlled, intensively phenotyped prospective study of metformin response ideally selected by SLC22A1
genotype will have less variation in the response phenotype and, hence, would be better powered to detect subtle effects of SLC22A1
variation on metformin response.
In conclusion, we have shown no clinically evident reduction in the ability of metformin to lower A1C in patients with type 2 diabetes with two common loss-of-function polymorphisms in SLC22A1
. A number of other OCTs have recently been implicated in the transport of metformin:SLC22A2 (OCT2
), and PMAT
). These are clearly of interest in the regulation of metformin response; in particular, SLC22A2
is involved in renal excretion of metformin and has been shown to have a greater affinity for metformin than SLC22A1
). In addition to a further study of SLC22A1,
a detailed pharmacogenetic study of these metformin transporters is also required.