Common variants near ATM are associated with glycemic response to metformin in type 2 diabetes.

Metformin is the most commonly used pharmacological therapy for type 2 diabetes. We report a genome-wide association study for glycemic response to metformin in 1,024 Scottish individuals with type 2 diabetes with replication in two cohorts including 1,783 Scottish individuals and 1,113 individuals from the UK Prospective Diabetes Study. In a combined meta-analysis, we identified a SNP, rs11212617, associated with treatment success (n = 3,920, P = 2.9 × 10(-9), odds ratio = 1.35, 95% CI 1.22-1.49) at a locus containing ATM, the ataxia telangiectasia mutated gene. In a rat hepatoma cell line, inhibition of ATM with KU-55933 attenuated the phosphorylation and activation of AMP-activated protein kinase in response to metformin. We conclude that ATM, a gene known to be involved in DNA repair and cell cycle control, plays a role in the effect of metformin upstream of AMP-activated protein kinase, and variation in this gene alters glycemic response to metformin.

In treating Type 2 diabetes metformin is recommended as first line therapy in most national and international guidelines1,2. Despite the clinical use of metformin for over 50 years, its mechanism of action has not been fully elucidated. Whilst it is established that metformin activates AMP-activated protein kinase (AMPK)3 by inhibition of the mitochondrial respiratory chain4, causing an increase in cellular AMP5, it remains uncertain whether AMPK is its sole therapeutic target.

There is considerable variability in glycaemic response to metformin. No clinical phenotype usefully predicts response6 yet there has been little pharmacogenetic investigation of metformin with no consistently replicated genetic variant established. We hypothesised that a GWA approach could be applied to the glycaemic response to metformin to gain insight into the mechanism of metformin’s action in humans, and to identify variants that may be useful clinically to predict efficacy or adverse outcome.

As part of the Wellcome Trust Case Control Consortium 2 study (WTCCC2), a GWA study of 15 complex traits and disorders, we carried out the first GWA study on metformin response in patients with type 2 diabetes, using a large Scottish observational genetic cohort (GoDARTS) of European ancestry. As our principal outcome phenotype, we used the ability to reduce HbA1c (the most widely used measure of medium term glycaemic control) in the first 18 months of therapy to below 7%, this being a key measure of success in many treatment algorithms. Covariates shown to alter metformin response, such as baseline HbA1c and creatinine clearance were included in a logistic regression model (supplementary methods). Full details of the cohorts and models used are available in the online methods and baseline characteristics of the cohorts are shown in supplementary figure 1 and supplementary table 1. Samples were genotyped using the Affymetrix 6.0 microarray. After strict quality control we analysed 705,125 SNPs in 1024 metformin treated patients (supplementary methods). The quantile-quantile plot is shown in supplementary figure 2; the genomic inflation factor was 1.003. The Manhattan plot is shown in supplementary figure 3. We found that 14 SNPs with a p-value <1*10−6 mapped to a 340 kb strong LD block on chromosome 11q22 (figure 1). No stronger association was observed around this locus after imputing the data to the 2.2 million HapMap II CEU panel. SNPs at other loci, that are potentially associated with metformin response, did not achieve a p-value lower than 1*10−6 and have not been followed up in the current study (supplementary table 2). The minor allele (C) of the most strongly associated SNP, rs11212617, had a frequency of 44% and was associated with treatment success (achieving an HbA1c below 7%) with an allelic odds ratio of 1.64 (95% Confidence interval 1.37 to 1.99, p=1.9*10−7) (table 1). The full model is shown in supplementary table 3a.

Figure 1

Regional association plots around the ATM locus for the logistic regression analysis. The solid and open triangles are from directly typed and imputed SNPs respectively

Table 1

Association analysis results between rs11212617 and glycaemic response to metformin in the discovery and internal replication cohorts, and the combined sample. The reference allele for rs11212617 is A. For the UKPDS samples, results are for rs609261, which was genotyped in this cohort due to technical difficulties, but was a proxy for rs11212617 (r2=0.997 in WTCCC2 controls). The logistic regression analysis shows allelic odds ratio (OR) for the ability to achieve a treatment HbA1c <=7% in the 18 months after starting metformin. The linear regression analysis shows the per-allele increase in treatment HbA1c in the treatment period after starting metformin. Covariates included in the model were baseline HbA1c, gap between treatment starting and baseline HbA1c, dose, adherence, creatinine clearance, and treatment group. Full models are shown in supplementary table 3. 95% confidence intervals of the beta or OR are shown in square brackets.

Study	Samplesize	Logistic		Linear
		OR	p	beta	p
Discovery	1024	1.64 [1.37,1.99]	1.9E-07	−0.18 [−0.26,−0.10]	1.8E-05
Replication 1	1783	1.21 [1.05,1.38]	0.007	−0.07 [−0.13,−0.01]	0.022
Replication 2(UKPDS)	1113	1.37 [1.10,1.72]	0.006	−0.12 [−0.23,−0.02]	0.021
Combined	3920	1.35 [1.22,1.49]	2.9E-09	−0.11 [−0.16,−0.07]	6.6E-07

Our primary analysis used a binary treatment target as its endpoint. To check the robustness of this, we also analysed the treatment HbA1c as a quantitative trait in a linear regression. In parallel with the primary analysis we found the C allele of rs11212617 was associated with lower treatment HbA1c (per allele Beta −0.18% [95% confidence intervals −0.26 to −0.1], p=1.8*10−5) (table 1).

Two replication cohorts were used. SNP rs11212617 was genotyped in an independent GoDARTS cohort of 1783 metformin-treated patients with type 2 diabetes (replication 1). The minor allele (C) of rs11212617 was associated with treatment success (allelic OR 1.21 95%CI 1.05 to 1.38; p=0.007) (table 1). The second replication cohort was 1113 UK patients prospectively treated with metformin in the UKPDS (UK Prospective Diabetes) cohort (replication 2). The UKPDS was a prospective randomised clinical trial of intensive vs conventional treatment in type 2 diabetes. In the UKPDS, where, for technical reasons, we typed the proxy SNP rs609261 (r2=0.997 with rs11212617 in 5197 WTCCC2 controls) the minor allele was associated with treatment success (allelic OR 1.37 95%CI 1.1 to 1.72; P=0.006) (table 1). The combined p-value achieved significance of p=2.9*10−9 for 3920 metformin-treated patients (table 1). The full models for each cohort are shown in supplementary table 3a-c. In the combined linear regression, each copy of the rs11212617 minor allele C is associated with 0.11% (p=6.6*10−7) lower absolute treatment HbA1c (table 1).

Metformin can be used as monotherapy alone, or added in to other therapies. However, current prescribing practice is for metformin to be used first line; we therefore analysed the monotherapy subgroup separately. Most of the association signal at rs11212617 with metformin response for the full group arises from the monotherapy subgroup (supplementary table 4). In a meta-analysis of this monotherapy group (n=2264) the combined odds-ratio for treatment success was 1.42 (95% CI 1.26 to 1.62), p=4*10−8. To assess the clinical impact of rs11212617, we studied the UKPDS cohort that was randomly assigned to metformin monotherapy (n=284) and followed up prospectively, and therefore not prone to treatment selection bias. In this subgroup, the 19% of patients with two copies of the C allele at rs11212617 have a 3.3-fold greater likelihood of achieving an HbA1c <=7%. This equates to a model adjusted difference in treatment HbA1c of 0.61% between those who are CC vs AA at this SNP. Adding genotype to the full linear regression model in this UKPDS group randomised to metformin increased the variance in the treatment HbA1c explained by the model from 27.5% to 30% (p=0.007).

To ensure that the genotypic effect on metformin response was not related to an effect on HbA1c per se, and to assess the association with fasting insulin and HOMA derived insulin resistance, we analysed summary statistics for rs11212617 from the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC)7. We found no association of rs11212617 with HbA1c (p=0.82), HOMA-IR (p=0.99) and fasting insulin (p=0.73) in at least 35,914 non-diabetic individuals (supplementary methods). We found no association between baseline HbA1c and rs11212617 in the GoDARTs discovery or replication dataset. In addition there was no association of rs11212617 with lipid parameters, blood pressure, height, weight, BMI, adiponectin and leptin in up to 6148 Scottish controls (supplementary table 5); nor with type 2 diabetes risk in a case-control study of 5788 Scottish GoDARTS patients with type 2 diabetes and 6357 non-diabetic Scottish GoDARTS controls (p=0.64).

The SNP rs11212617 falls within a large block of linkage disequilibrium that includes the genes CUL5, ACAT1, NPAT, ATM, C11orf65, KDELC2, EXPH5. Of these, ATM was considered a possible candidate gene. Firstly, homozygous loss of function mutations in ATM cause Ataxia Telangiectasia (A-T; OMIM #208900) which is a neurodegenerative disorder characterized by loss of muscle coordination and progressive ataxia, radiosensitivity, immunodeficiency and a predisposition to cancer8; additionally, patients with A-T have been reported to have marked insulin resistance and increased risk of diabetes9,10. Secondly, previous laboratory reports suggest that activation or inhibition of ATM alters AMPK activation11-13. None of the other genes at the locus have been reported to be associated with diabetes or insulin action.

ATM encodes a 370 kDa protein that is a Ser/Thr protein kinase of the atypical phosphoinositide 3-kinase-related protein kinase (PIKK) family. ATM is activated by double-stranded DNA breaks, and acts to induce cell-cycle arrest and facilitate DNA repair14. To investigate if ATM was the causal gene affecting the glycaemic response to metformin we studied the effects of a selective ATM inhibitor, KU-55933, on the activation of AMPK by metformin in rat hepatoma (H4IIE) cells. Figure 2 shows that KU-55933 markedly reduced AMPK activation by metformin. Similarly, figure 3 shows that phosphorylation of AMPK and its downstream target, ACC, by metformin was inhibited by KU-55933. These results are supported by previous reports of ATM involvement in the activation of AMPK by stimuli other than metformin11-13. We conclude that ATM acts upstream of AMPK, and is required for a full response to metformin. ATM is also reported to be involved in insulin signalling and pancreatic β-cell dysfunction, both of which may influence metformin action: apoE −/− mice heterozygous for loss of atm function were insulin-resistant compared to apoE−/− mice with normal atm15; however, mice lacking atm develop diabetes due to β cell dysfunction16.

Figure 2

Effect of KU-55933 on AMPK activation by metformin

H4IIE cells were pre-treated with or without 10 μM KU-55933 for 30 min and then with various concentrations of metformin for 1 hr, and AMPK activity measured; Results are mean ± S.D. (n = 2); **significantly different from incubation without KU-55933 by 2-way ANOVA (p<0.01).

Figure 3

A Western Blot comparing the phosphorylation status of Thr-172 of AMPK and Ser-79 of ACC (a well characterized marker of AMPK activation). H4IIE cells were pre-treated with or without 10 μM KU-55933 (KU) for 1 hour and then for 3 hours with or without 2.5mmol/L metformin. Metformin induced phosphorylation of AMPK and subsequent phosphorylation of ACC was partially reduced by KU-55933.

Potential functionality of all the 98 SNPs with strong linkage disequilibrium (r2 >0.8 according to the HapMap CEU panel) to rs11212617 was assessed (supplementary table 6). SNP rs228589, which is in intron 1 of the NPAT gene, is in a predicted promoter of ATM17. Two SNPs, rs227092 and rs4585, are located in the ATM 3′UTR region. Variant rs4585 lies 24bp downstream of a polyadenylation site, and is predicted to alter efficiency of polyadenylation (supplementary table 6). The 3′UTR region of ATM is among the longest known mammalian 3′UTRs and has been suggested to influence ATM mRNA translation allowing rapid response to stimuli at the post-transcriptional level18,19.

Type 2 diabetes is associated with increased cancer risk, and an overlap between genes predisposing to prostate cancer and type 2 diabetes has been described20. Metformin has been shown in epidemiological studies to be associated with decreased cancer risk21, and to decrease tumour burden in pten deficient mice22. Activation of AMPK by metformin requires the known tumour suppressor LKB1. In this study the implication of ATM, a gene known to be involved in DNA repair and cancer, in the glycaemic response to metformin establishes a further link between cancer pathways, type 2 diabetes and metformin activation of AMPK.

In this study, we have established the utility of a genome wide approach to study the pharmacogenomics of metformin response, and the utility of large genetics resources linked to routinely collected clinical data for pharmacogenetic studies. We have identified the first robustly replicated variant to be associated with metformin response. Whilst this observation may not be of immediate clinical utility, explaining only 2.5% of variance in metformin response, this study is an example of how genome-wide association studies can be applied to pharmacogenomic models to identify novel pathways and mechanisms, and has established an unexpected link between glucose homeostasis and the DNA damage response.