Research ArticleMetabolism

Epigenetic markers associated with metformin response and intolerance in drug-naïve patients with type 2 diabetes

See allHide authors and affiliations

Science Translational Medicine  16 Sep 2020:
Vol. 12, Issue 561, eaaz1803
DOI: 10.1126/scitranslmed.aaz1803

How to mete out metformin

Metformin is the most commonly used drug to treat type 2 diabetes (T2D), though not all patients respond to it, and still, others do not tolerate it. García-Calzón et al. analyzed genome-wide DNA methylation in the blood of drug-naïve patients who were recently diagnosed with T2D. They found that DNA methylation at specific loci associated with future metformin response or tolerance, respectively, across multiple cohorts. These epigenetic markers may have theranostic potential regarding which patients should receive metformin.

Abstract

Metformin is the first-line pharmacotherapy for managing type 2 diabetes (T2D). However, many patients with T2D do not respond to or tolerate metformin well. Currently, there are no phenotypes that successfully predict glycemic response to, or tolerance of, metformin. We explored whether blood-based epigenetic markers could discriminate metformin response and tolerance by analyzing genome-wide DNA methylation in drug-naïve patients with T2D at the time of their diagnosis. DNA methylation of 11 and 4 sites differed between glycemic responders/nonresponders and metformin-tolerant/intolerant patients, respectively, in discovery and replication cohorts. Greater methylation at these sites associated with a higher risk of not responding to or not tolerating metformin with odds ratios between 1.43 and 3.09 per 1-SD methylation increase. Methylation risk scores (MRSs) of the 11 identified sites differed between glycemic responders and nonresponders with areas under the curve (AUCs) of 0.80 to 0.98. MRSs of the 4 sites associated with future metformin intolerance generated AUCs of 0.85 to 0.93. Some of these blood-based methylation markers mirrored the epigenetic pattern in adipose tissue, a key tissue in diabetes pathogenesis, and genes to which these markers were annotated to had biological functions in hepatocytes that altered metformin-related phenotypes. Overall, we could discriminate between glycemic responders/nonresponders and participants tolerant/intolerant to metformin at diagnosis by measuring blood-based epigenetic markers in drug-naïve patients with T2D. This epigenetics-based tool may be further developed to help patients with T2D receive optimal therapy.

INTRODUCTION

Metformin is commonly prescribed as a first-line pharmacotherapy for type 2 diabetes (T2D) (1). However, ~30% of patients with T2D do not respond to metformin (2) and ~20 to 30% experience intolerable side effects, including gastrointestinal symptoms that warrant discontinuation of metformin treatment in ~5% of patients (3). To our knowledge, there are no ways to successfully predict the glycemic response or intolerance to metformin (4). Genetics explain only a modest proportion of metformin response and intolerance (412). Therefore, additional studies are needed to identify markers that determine whether patients with T2D will respond to or tolerate metformin or whether other therapies should be prioritized. We and others have demonstrated that epigenetics, specifically DNA methylation, contribute to T2D (1318). We also identified blood-based epigenetic markers that mirror the methylation pattern in human islets and associate with future insulin secretion and T2D (16). Epigenetic markers could provide valuable tools for precision medicine however whether blood-based epigenetic markers associate with future drug response and intolerance in patients with T2D remains to be tested.

We aimed to investigate whether DNA methylation in blood associates with future glycemic response and intolerance to metformin therapy in multiple cohorts of drug-naïve patients with T2D from ongoing prospective studies. We further explored cross-tissue methylation patterns of sites associated with future glycemic response or intolerance to metformin in human adipose tissue (14). In addition, we studied whether genes to which the identified DNA methylation markers are annotated to affect phenotypes related to metformin therapy in hepatocytes.

RESULTS

Epigenetic markers associate with future glycemic response to metformin

As part of the prospective ANDIS (All New Diabetics In Scania) study (19) we carried out a pharmacoepigenetic study for diabetes to identify blood-based epigenetic markers that associate with changes in glycated hemoglobin (∆HbA1c) or future metformin response in drug-naïve patients with T2D (Fig. 1). Using the 850K array, we analyzed DNA methylation in blood of the discovery and replication cohorts for metformin response (tables S1 and S2 and figs. S1 to S3). We assessed whether methylation status before taking metformin was associated with ∆HbA1c in the full discovery cohort after ~1.5 years of therapy and whether epigenetic markers could discriminate between nonresponders and responders to metformin in a subset of patients fulfilling the American Diabetes Association (ADA) criteria for glycemic response (20).

Fig. 1 Study design.

DNA methylation was analyzed genome-wide to identify blood-based epigenetic markers that could associate with change in HbA1c and discriminate future glycemic response and intolerance to metformin therapy. Discovery and replication cohorts from ANDIS, ANDiU, and OPTIMED were included. A fixed meta-analysis was performed to select individual methylation markers associated with future metformin response or intolerance. Methylation risk scores (MRSs) were calculated and used to stratify patients with T2D into glycemic responders/nonresponders and metformin tolerant/intolerant. We then assessed whether these blood-based epigenetic markers mirror DNA methylation in human adipose tissue, a central tissue of diabetes. Last, functional in vitro follow-up experiments in hepatocytes tested whether genes annotated to the identified epigenetic markers might influence phenotypes related to metformin therapy such as expression of metformin transporters and regulators of gluconeogenesis as well as AMPK activity.

First, we explored whether DNA methylation associated with change in HbA1c after ~1.5 years of metformin treatment in newly diagnosed patients with T2D in the ANDIS discovery cohort for metformin response (Fig. 2A and table S1). Methylation of 2583 sites was significantly associated with ∆HbA1c after ~1.5 years of metformin [false discovery rate (FDR) < 5%, q < 0.05]. Moreover, 2577 sites remained significant (FDR < 5%) after adjusting for cell composition (table S3) (21). Methylation of each site seemed to explain a proportion of variation in ∆HbA1c as the adjusted R-squared ranged between 0.13 and 0.61. In addition, methylation of all these sites except one was associated with ∆HbA1c when adjusting regression models for fewer covariates (table S4), suggesting that these covariates did not substantially influence the association. Methylation of 499 and 48 sites was also associated with baseline HbA1c and creatinine clearance estimated glomerular filtration rate (eGFR), respectively (table S3). We proceeded with replication testing of sites associated with ∆HbA1c in a cohort of 204 newly diagnosed participants with T2D, the ANDIS replication cohort for metformin response (table S1). We found that methylation of 132 CpGs was also associated with ∆HbA1c (P < 0.05) in the replication cohort (n = 204), with beta-coefficients in the same direction as in the discovery cohort (table S5).

Fig. 2 Associations between DNA methylation and metformin response and intolerance in the ANDIS discovery cohorts.

(A) Associations between methylation and ∆HbA1c in 63 drug-naïve participants with T2D after adjusting for basal HbA1c, eGFR, and time gaps (between baseline HbA1c and methylation measurements and the start of metformin) displaying 2583 significant CpG sites (FDR < 5%, q < 0.05). (B) Associations between methylation and metformin response in 21 nonresponders and 26 responders after adjusting for basal HbA1c, eGFR, and time gaps in baseline HbA1c and methylation displaying 7973 significant CpG sites (FDR below 5%, q < 0.05). Beta-coefficients in the volcano plot are shown when comparing glycemic nonresponders versus responders to metformin. (C) Associations between methylation and metformin intolerance in 66 tolerant and 17 intolerant participants with T2D after adjusting for basal HbA1c, eGFR, and time gap in baseline methylation displaying 12,579 significant CpG sites (FDR below 5%, q < 0.05). Beta-coefficients in the volcano plot are shown when comparing metformin-intolerant versus metformin-tolerant participants. Blue dashed lines and red lines indicate methylome-wide significance (q < 0.05).

We next selected two well-defined groups of 26 glycemic responders (HbA1c after ~1.5 years < 48 to 53 mmol/mol and reduction in HbA1c ≥ 11 mmol/mol) and 21 nonresponders (HbA1c after ~1.5 years ≥ 48 to 53 mmol/mol and reduction in HbA1c < 11 mmol/mol) to metformin treatment from the ANDIS discovery cohort (table S2 and fig. S1) and tested whether baseline methylation discriminated these patient groups. In this case-control set, 7973 sites showed significant (FDR < 5%) differences in methylation between glycemic responders and nonresponders, and 7916 sites remained significant (FDR < 5%) when adjusting for cell composition (Fig. 2B and table S6) (21). In addition, methylation of 7542 sites was associated with glycemic response when adjusting for less covariates in regression models (table S7), suggesting that these covariates did not substantially influence the association. We then performed replication testing of sites associated with glycemic response (table S6) using two independent cohorts of 48 responders and 39 nonresponders selected from the ANDIS replication cohort as well as 47 responders and 31 nonresponders from the European replication cohort (table S2 and figs. S2 and S3). Among the significant sites (FDR < 5%) that we identified in the discovery cohort, methylation of 601 and 329 sites was associated with glycemic response also in the ANDIS and European replication cohorts, respectively, with directional consistency (tables S8 and S9). Furthermore, methylation of 33 sites was associated with glycemic metformin response in the discovery cohort (FDR < 0.05) and in the two replication cohorts (P < 0.05; table S10). In a combined meta-analysis of the discovery and replication data, 11 of these 33 methylation markers reached epigenome-wide significance after Bonferroni correction (P < 6.1 × 10−8, 0.05/816,000) for the association with glycemic metformin response (Table 1). Higher methylation values of all 11 sites were associated with a higher risk of not responding to metformin with odds ratios (OR) ranging between 1.43 and 2.46 per 1-SD increase in methylation (Fig. 3A).

Table 1 Differentially methylated sites associated with future metformin response or intolerance in the combined meta-analysis of discovery and replication cohorts (P < 6.1 × 10−8).

Beta-coefficients (SEM) are estimated from a linear model in the given cohort with either nonresponders versus responders to metformin (top) or intolerant versus tolerant to metformin (bottom). Models were adjusted for basal HbA1c, eGFR, and time gaps (time gap in baseline HbA1c and time methylation gap) for metformin response and eGFR and time methylation gap for metformin intolerance. Time gap in baseline HbA1c was defined as the number of days between the measurement of baseline HbA1c and the start of metformin therapy, whereas time gap in baseline methylation was defined as the number of days between the measurement of DNA methylation in blood and the start of metformin therapy. Inverse variance fixed meta-analysis of discovery and replication cohorts was performed and Bonferroni-corrected. P < 6.1 × 10−8 was considered significant. 5′UTR, 5′ untranslated region.

View this table:
Fig. 3 Risk for not responding to or not tolerating metformin of methylation markers associated with glycemic response or intolerance.

Logistic models were performed for each CpG site in all participants with T2D included in the discovery and replication cohorts for metformin response (n = 212) (A) and in all participants included in the discovery and replication cohorts for metformin intolerance (n = 151) (B). OR are shown per 1-SD increase in methylation for each CpG site. CI, confidence interval.

We proceeded to generate combined weighted methylation risk scores (MRSs) (22) based on these 11 sites and examined whether these scores could discriminate between glycemic responders and nonresponders to metformin. Using the CpG-specific effect sizes (beta-coefficients from logistic models) from the ANDIS discovery cohort, we found that MRSs adequately discriminated metformin responders from nonresponders in the two replication cohorts (Fig. 4). Receiver operating characteristic (ROC) curves showed that MRSs discriminated between metformin responders and nonresponders with an area under the curve (AUC) of 0.80 for the ANDIS replication cohort and 0.89 for the European replication cohort (Fig. 4). We next used the CpG-specific effect sizes from ANDIS or the European replication cohorts to calculate and evaluate MRSs in the other two cohorts. These MRSs also allowed adequate discrimination of metformin responders and nonresponders with AUCs ranging between 0.80 and 0.98 (fig. S4 and S5). In addition, these MRSs explained 68 to 73% of the variation in glycemic response to metformin in the ANDIS discovery cohort, 19 to 20% in the ANDIS replication cohort, and 38 to 42% in the European replication cohort (based on R-squared McFadden). These data support the notion that blood-based epigenetic markers may be useful for stratification of metformin response in drug-naïve patients with T2D.

Fig. 4 Combined MRSs discriminate between glycemic responders and nonresponders to metformin in drug-naïve participants with T2D.

The MRSs include the 11 CpG sites associated with future metformin response (see Table 1). CpG-specific effect sizes (beta-coefficients from logistic models) from the ANDIS discovery cohort for metformin response (n = 47) (A) were used to calculate and evaluate the MRSs in the ANDIS (n = 87) (B to D) and European (n = 78) (E to G) replication cohorts for metformin response. Boxplots show significantly different MRSs between glycemic responders and nonresponders to metformin in both the ANDIS replication (P for Mann-Whitney U test = 6.6 × 10−7) (B) and the European replication (P for Mann-Whitney U test = 1.6 × 10−10) (E) cohorts. Histogram plots show distributions of the MRSs stratified by response to metformin in the ANDIS (C) and the European (F) replication cohorts. Red bars represent nonresponders; orange bars represent responders to metformin. The ROC curves show the discrimination between responders/nonresponders based on MRSs. The AUC for metformin response was 0.80 in the ANDIS replication cohort (D) and 0.89 in the European replication cohort (G). Beta, beta-coefficients.

In line with previous findings (4), age, body mass index (BMI), baseline HbA1c, and eGFR were not associated with future glycemic response to metformin in our cohorts (fig. S6A). Moreover, other factors that might affect glycemic control such as ongoing treatment with lipid-lowering or antihypertensive medication, blood pressure, as well as albumin or creatinine in urine were not associated with future glycemic response either (fig. S7).

Epigenetic markers associate with future metformin intolerance

We next evaluated whether methylation in blood taken before treatment could discriminate patients with T2D who experienced intolerable side effects (metformin-intolerant) from those who were able to tolerate metformin (metformin-tolerant). We analyzed the methylation of ~850,000 sites in blood samples from the discovery and replication cohorts for metformin intolerance comprising drug-naïve patients with T2D (Fig. 1 and table S11).

DNA methylation of 12,579 sites was associated with intolerable side effects in the ANDIS discovery cohort (FDR < 5%) (Fig. 2C). In addition, 9676 sites remained significant after adjusting for cell composition (FDR < 5%) (table S12) (21) and 9673 sites were significant when adjusting for less covariates in regression models (P < 0.05) (table S13), suggesting that these confounders did not substantially influence the association. Most sites showed higher methylation (7865 CpGs) in metformin-intolerant versus metformin-tolerant patients. We next performed replication testing of the sites associated with metformin intolerance (table S12) in two independent cohorts, the ANDIS and European replication cohorts. We found that methylation of 235 and 352 CpGs was associated with metformin intolerance in the ANDIS and European replication cohorts, respectively, with directional consistency (tables S14 and S15). Overall, seven methylation markers were associated with metformin intolerance in the discovery cohort (FDR < 0.05) and in both replication cohorts (P < 0.05) (table S16). In a combined meta-analysis of the discovery and replication data, four of these seven methylation markers reached epigenome-wide significance after Bonferroni correction (P < 6.1 × 10−8, 0.05/816,000) for association with metformin intolerance (Table 1). Higher methylation values of each of these four sites were associated with a higher risk of metformin intolerance with ORs ranging between 1.65 and 3.09 per 1-SD increase in methylation (Fig. 3B).

We then generated combined MRSs (22) based on the data from these four sites to assess whether these scores could discriminate metformin-tolerant from metformin-intolerant drug-naïve participants with T2D. Using the CpG-specific effect sizes (beta-coefficients from logistic models) from the ANDIS discovery cohort, we calculated and evaluated MRSs in the two replication cohorts and found a separation between metformin-tolerant and metformin-intolerant participants (Fig. 5), with an AUC of 0.94 for the ANDIS replication cohort and 0.89 for the European replication cohort (Fig. 5). We next used CpG-specific effect sizes from the ANDIS replication cohort or the European replication cohort to calculate and evaluate MRSs in the other two cohorts. These MRSs did also give a good separation between metformin-intolerant and metformin-tolerant participants, with AUCs ranging between 0.85 and 0.93 (figs. S8 and S9). In addition, these MRSs explained 50 to 51% of the variation in metformin intolerance in the ANDIS discovery cohort, 51 to 54% in the ANDIS replication cohort, and 32 to 33% in the European replication cohorts (based on R-squared McFadden). In line with previous findings (4), age, BMI, baseline HbA1c and, eGFR were not associated with future intolerance to metformin in all participants from our discovery and replication cohorts (fig. S6B).

Fig. 5 Combined MRSs discriminate metformin tolerance and intolerance in drug-naïve participants with T2D.

The MRSs include the four CpG sites associated with future metformin intolerance (see Table 1). CpG-specific effect sizes (beta-coefficients from logistic models) from the ANDIS discovery cohort for metformin intolerance (n = 83) (A) were used to calculate and evaluate the MRSs in the ANDIS (n = 48) (B to D) and European (n = 20) (E to G) replication cohorts for metformin intolerance. Boxplots show significantly different MRSs between tolerant and intolerant participants to metformin in both the ANDIS replication (P for Mann-Whitney U test = 4.5 × 10−7) (B) and the European replication (P for Mann-Whitney U test = 1.5 × 10−2) (E) cohorts. Histogram plots show distributions of the MRSs stratified by intolerance to metformin in the ANDIS (C) and the European (F) replication cohorts. Red bars represent intolerant participants; orange bars represent tolerant participants to metformin. The ROC curves show the discrimination between tolerant/intolerant participants based on MRSs. The AUC for metformin intolerance was 0.94 in the ANDIS replication cohort (D) and 0.89 in the European replication cohort (G). Beta, beta-coefficients.

Associations between genetic variants and epigenetics for discriminating metformin response and intolerance

Some studies have previously performed associations between single-nucleotide polymorphisms (SNPs) and glycemic response or tolerance to metformin, but the degree of confidence in the reported results varies (59, 12, 2331). Nevertheless, we selected 26 SNPs previously associated with metformin response (8, 9, 12, 2330) or intolerance (57, 31) to test whether genetics together with epigenetics could better discriminate between metformin response/nonresponse or tolerance/intolerance. We extracted these SNPs from genome-wide Illumina array data available in the ANDIS discovery and replication cohorts.

We first assessed whether any of these SNPs were associated with DNA methylation of any of the epigenetic marks that we identified as discriminating between metformin response/nonresponse or tolerance/intolerance (Table 1). After correcting for multiple testing, we found only one significant association between a SNP in SCL22A1 (rs628031) and DNA methylation of cg05151280 (PANOVA = 0.001, q = 0.028). Here, A/A genotype carriers had lower methylation (83.6 ± 2.3%) compared to carriers of the G/G (85.3 ± 1.9%, P = 0.002) and G/A (85 ± 1.8%, P = 0.006) genotypes in 132 participants from the ANDIS discovery and replication cohorts. Lower methylation of this CpG site was associated with a better glycemic response to metformin (Table 1). One previous study found a greater reduction in HbA1c in response to metformin in A/A compared to G-allele carriers of rs628031 (30), whereas other studies found no association between this polymorphism and metformin response (23, 27).

We also evaluated the extent to which these SNPs alone and in addition to our MRSs discriminated between metformin responders/nonresponders and tolerant/intolerant participants in ANDIS (tables S17 and S18). The ability of each SNP to discriminate metformin response and intolerance was generally low, with AUCs ranging from 0.50 to 0.66. Moreover, there was no significant improvement of the AUC regarding metformin response or intolerance after adding each SNP on the top of the MRSs in the ANDIS discovery and replication cohorts (P > 0.05). These data support that the association between our epigenetic markers and future metformin response or intolerance occurs independently of these 26 SNPs.

We also performed regression analyses to test whether any of these SNPs were associated with glycemic response or intolerance to metformin in participants from ANDIS discovery and replication cohorts. Here, T-allele carriers of rs8192675 (SLC2A2) had a nominally higher risk of not responding to metformin compared with homozygous CC carriers [CT + TT versus CC, OR = 4.9(2.2), P = 0.04], which is in line with previous data (12). We also found a nominal association between rs12208357 (SLC22A1) and metformin intolerance where T-allele carriers had a lower risk of intolerance [CT + TT versus CC, OR = 0.13(2.85), P = 0.05].

Cross-tissue methylation in blood and human adipose tissue

Next, we investigated whether methylation in blood of the 11 sites (8 sites available in the 450K array) associated with metformin response and the 4 sites (2 sites available in the 450K array) associated with intolerance reflected methylation in human adipose tissue. Here, we used available methylation data on the 450K array in blood and adipose tissue; as for these cells, we had access to methylation data from the same participants (tables S19 and S20) (14, 32). We found that DNA methylation of three sites in the blood positively correlated with methylation in adipose tissue after correcting for multiple testing (fig. S10). These findings suggest that methylation in blood associated with metformin response and intolerance may also have a biological role in key tissues for T2D.

Functional follow-up experiments in hepatocytes cultured in vitro

We asked whether genes to which the identified CpG sites associated with metformin response and intolerance (Table 1) are annotated to have a functional role in liver cells (HepG2 cells) (33, 34). For functional follow-up experiments, we focused on genes that have previously been related to phenotypes involved in diabetes. On the basis of these criteria, we selected five genes (OR4S1, SEPT11, CST1, FOXA2, and PGM1) (3541) for functional experiments, and we elucidated their effects on expression of two metformin transporters (SLC22A1, encoding OCT1, the main transporter for metformin uptake into hepatocytes, and SLC47A1, encoding MATE1, the main efflux transporter of metformin to the bile), 5′ adenosine monophosphate–activated protein kinase (AMPK) activity and expression of key regulators of gluconeogenesis (PCK1 and G6PC) in liver cells untreated and treated with metformin. We silenced the expression of these five genes in HepG2 cells using small interfering RNA (siRNA), which resulted in an 82 to 98% reduction in expression of all the genes (Fig. 6A) except for OR4S1 whose mRNA expression was undetectable, probably because of low expression in this liver cell line. OR4S1 was therefore excluded from further experiments.

Fig. 6 Silencing of genes associated with metformin response (SEPT11 and CST1) or intolerance (FOXA2 and PGM1) in hepatocytes affects expression of metformin transporters, AMPK activity, and expression of key regulators of gluconeogenesis.

(A) Quantification of siRNA-mediated knockdown of SEPT11, CST1, FOXA2, and PGM1 (siSEPT11, siCST1, siFOXA2, and siPGM1) compared with negative control siRNA (siNC) in Hep2G cells. (B) AMPK activity and expression of key regulators of gluconeogenesis (PCK1 and G6PC) and metformin transporters (SLC22A1 and SLC47A1) in metformin-treated (exposed to 5 and 2.5 mM metformin, respectively) compared to nonmetformin exposed HepG2 cells. (C to F) mRNA expression of metformin transporters (SLC22A1 and SLC47A1), AMPK activity, and mRNA expression of key regulators of gluconeogenesis (PCK1 and G6PC) in Hep2G cells deficient for SEPT11 (C), CST1 (D), FOXA2 (E), or PGM1 (F) expression compared to cells transfected with siNC after nonmetformin exposure or metformin treatment overnight. For all panels, data are means ± SEM of four independent experiments performed in different passages of Hep2G cells, with two technical replicates for each condition. P values were calculated using paired t tests, #P = 0.056 to 0.075, *P < 0.05, **P < 0.01, ***P < 0.001. Two-sided P values were calculated using paired t tests of logged values for all the analyses.

As expected, metformin treatment activated AMPK and decreased PCK1 and G6PC gene expression in cultured HepG2 cells, confirming the pharmacological effect of metformin in the inhibition of gluconeogenesis (Fig. 6B) (42). Moreover, metformin did not alter the expression of metformin transporters (Fig. 6B) (43).

To investigate glycemic response to metformin, we silenced two genes (SEPT11 and CST1) located near CpG sites associated with metformin response in newly diagnosed patients with T2D (cg01070242 and cg07511259, respectively) (Table 1). We found that SEPT11-deficient HepG2 cells had lower SLC47A1 expression (Fig. 6C), which could result in lower efflux and higher metformin concentration in the hepatocytes associated with a greater pharmacologic response (44). In addition, SEPT11 deficiency resulted in lower G6PC expression (Fig. 6C), a mechanism previously associated with decreased gluconeogenesis and lower hepatic glucose output (42). CST1-deficient HepG2 cells had increased SLC47A1 expression (Fig. 6D), which could result in higher efflux and lower metformin concentration in the hepatocytes associated with a lower pharmacologic response (44). Moreover, CST1-deficient cells had nominally decreased AMPK activity and increased expression of PCK1 and G6PC (Fig. 6D), associated with increased gluconeogenesis and elevated hepatic glucose output (42).

Regarding intolerance to metformin, we silenced two genes, FOXA2 and PGM1, near cg12356107 and cg02994863, respectively, associated with metformin intolerance in newly diagnosed patients with T2D (Table 1). Both FOXA2- and PGM1-deficient HepG2 cells had higher SLC47A1 expression (Fig. 6, E and F), which could result in higher excretion and therefore lower metformin concentration in the hepatocytes (44, 45). Moreover, FOXA2 and PGM1 deficiency resulted in higher PCK1 and G6PC expression (Fig. 6, E and F), associated with increased gluconeogenesis, lower lactate production, and therefore a better tolerance to metformin (34). Regarding AMPK activity, FOXA2-deficient cells had nominally reduced AMPK phosphorylation (Fig. 6E), which is in line with the increase in gluconeogenesis and therefore a better tolerance to metformin (34). However, PGM1 deficiency did not change AMPK activity (Fig. 6F), suggesting that AMPK-independent mechanisms may be involved to activate gluconeogenesis in these cells (42, 46) and hence reduced lactate concentration. Overall, these experiments support that several genes annotated to CpG sites associated with response or intolerance to metformin have a functional role in liver cells where they affect metformin transporters and key regulators of gluconeogenesis.

Increased methylation in promoter and CpG island regions has been associated with decreased expression (17). We used a luciferase assay to study the impact of increased DNA methylation in transcriptional regulation of SAP130, a gene annotated to a CpG site (cg16240962) located in promoter and CpG island regions and that is associated with glycemic response to metformin (Table 1). The promoter sequence for SAP130 was inserted into a luciferase expression plasmid and mock-methylated or methylated with the methyltransferase SssI (methylating 158 methylation sites, including cg16240962). Our data show that increased promoter methylation suppressed the transcriptional activity of SAP130 in HepG2 cells (fig. S11), supporting that DNA methylation of sites associated with metformin response may mediate gene regulation.

DISCUSSION

We performed a pharmacoepigenetic study for T2D in which we identified and validated blood-based epigenetic markers associated with future glycemic response and intolerance to metformin therapy in drug-naïve participants with T2D. Metformin-nonresponsive and metformin-intolerant patients with T2D should be prescribed other glycemic lowering drugs to achieve treatment goals of ADA and the European Association for the Study of Diabetes (EASD) (1, 20). However, there are currently no biomarkers available for identifying these patients at diagnosis (412). Our study found that DNA methylation at 11 and 4 specific loci was associated with future glycemic response and intolerance to metformin, respectively, in discovery and replication cohorts. Patients with higher degrees of methylation at these sites were up to 2.5 times more likely to not respond to, and up to 3 times more likely to not tolerate, metformin because of severe side effects. Moreover, methylation at these sites used in weighted MRSs was different between responders/nonresponders and tolerant/intolerant to metformin. AUCs for these MRSs ranged between 0.80 and 0.98 in the cohorts for metformin response and between 0.85 and 0.94 in the cohorts for metformin intolerance. Although more studies are needed to validate these markers in other populations, our results support further development of epigenetic markers for stratification of nonresponsive and intolerant patients to metformin already at diagnosis. Such stratification may help patients with T2D receive an optimal therapy and could be a step toward personalized medicine. Future studies in additional cohorts may optimize these MRSs further and may add or replace some markers. In the current study, the MRSs were slightly different depending on which CpG-specific effect sizes were estimated from one cohort and evaluated it in the other two. However, a single MRS would be useful for clinical use and therefore should be further developed and optimized in independent cohorts. In addition, it would be useful to clinically validate epigenetic markers in a randomized clinical trial. To this end, collection of larger cohorts should be prioritized.

There are several reasons supporting the potential use of DNA methylation markers when deciding whether to prescribe metformin therapy. Analyzing epigenetic markers in blood is noninvasive, safe, quick and, cost effective. Methylation is quite stable, can persist over time, and is inherited through cell divisions (47). Moreover, the combination of identified methylation sites associated with response and intolerance to metformin using MRSs shows AUCs >0.80, which is a requirement for a useful clinical discrimination (48, 49). In addition, giving an optimal therapy to newly diagnosed patients with T2D by using epigenetic markers could potentially decrease costs related to poor glycemic control, reducing visits to the doctor, sick leave, exhaustion, and vascular complications. The average cost of vascular complications in T2D was estimated to $47,240 per patient over 30 years (50). This can be compared with an estimated cost of $200 for measuring DNA methylation in blood. Although this is a pharmacoepigenetic study for diabetes, there is already a commercial liver cancer test available that analyses SEPT9 methylation (51) supporting the feasibility of clinical epigenetic markers.

This study has potential limitations. CpG sites were selected on the basis of their significance in the discovery and two replication cohorts, which may result in overestimated AUCs. To mitigate this, we used CpG-specific effect sizes (beta-coefficients from logistic models) in the discovery cohorts, calculating and evaluating the MRSs in the replication cohorts. This approach gave similar AUCs for all three cohorts, supporting further development of epigenetic markers for discrimination of response and intolerance to metformin. Associations were present regardless of baseline HbA1c and eGFR, providing additional support for the robustness of the findings. Also, the association was present when adjusting for cell composition (21). Whereas case-control cohorts for metformin response were balanced for basal HbA1c, eGFR, age, sex, and BMI, the full discovery cohort for metformin response included participants with a continuum of these variables. Subsequently, one would not expect these two different analyses to give the same result. Eight hundred eighty-eight methylation sites were significant in both the case-control and full discovery cohort for metformin response, whereas some sites were only significant in one of these analyses (FDR < 5%). Our study was carried out in Caucasians, and validation in other ethnicities is strongly needed. However, we are not aware of any additional cohorts with blood samples available in drug-naïve newly diagnosed patients with T2D at this time point. The fact that we had to change the inclusion criteria slightly to find participants for replication may have reduced the possibility of replicating significant sites. However, the replicated sites were statistically robust and were found in several cohorts. Last, in line with several previous metformin studies (8, 10, 12), we used pharmacy registers to identify patients who were on metformin therapy. One potential limitation with this design is that we cannot examine patient medication adherence. However, for metformin intolerance, we called the patients or checked in their clinical history records for the reason why they stopped metformin therapy so we could confirm the intolerance status of these patients to metformin therapy.

Environmental factors such as exercise and diet as well as obesity and weight change might affect DNA methylation (22, 5254). However, neither baseline BMI nor weight change had an impact on methylation of the sites associated with glycemic response or intolerance to metformin in respective cohorts in our study. We also examined whether methylation of our significant markers changed in other studies where the impact of environmental factors was investigated. Here, methylation of only 1 site (of the 11 and 4 sites associated with metformin response and intolerance, respectively) changed in adipose tissue after 5 days of high fat diet, and none of these sites changed in adipose tissue after exercise (53, 54). Together, these data suggest that lifestyle factors have minor effects on methylation of the sites associated with either response or intolerance to metformin.

Some of the identified blood-based epigenetic markers associated with metformin response or intolerance mirror the methylation pattern in adipose tissue, a metabolically relevant tissue for T2D (14). Blood-based epigenetic markers might hence reflect what it is happening in the central tissues of diabetes. In addition, silencing nearby genes (SEPT11, CST1, FOXA2, and PGM1) annotated to these CpG sites in hepatocytes resulted in altered expression of metformin transporters and key enzymes affecting gluconeogenesis, supporting biological functions of these epigenetic markers in the pharmacological effect of metformin. For example, we found that FOXA2 has a functional role in hepatocytes which might alter tolerance to metformin. Similarly, it has been shown that FOXA2 mediates an effect of metformin on bile acid metabolism, which is a likely cause of adverse gastrointestinal effects (55). The genes that we selected for these functional experiments have previously been shown to affect diabetes-related phenotypes (3541), for example, CST1 has been proposed as a promising biomarker for both diabetic neuropathy and breast cancer (56, 57). Our functional data further support that some of these epigenetic markers can regulate gene transcription. Overall, we shed light on some potential biological mechanisms related to our epigenetic markers and their link to metformin response or intolerance.

In conclusion, our study provides potential blood-based epigenetic markers for stratification of newly diagnosed patients with T2D into metformin nonresponsive/responsive and metformin intolerant/tolerant. Further research is warranted to develop this panel of epigenetic markers to aid clinical decision-making in T2D therapy by assigning newly diagnosed patients to receive either metformin or other glycemic-lowering medication, which may reduce suffering for patients.

MATERIALS AND METHODS

Study design

This study was designed to identify blood-based epigenetic markers that could discriminate between glycemic responders/nonresponders and tolerant/intolerant patients with T2D to metformin. The glycemic response of the participants to metformin treatment was based on the change in HbA1c values after ~1.5 years of therapy according to ADA and EASD guidelines, and metformin intolerance was assessed by the presence of intolerable side effects in clinical history records. Discovery and replication cohorts from ANDIS, ANDiU (All New Diabetics in Uppsala County), and OPTIMED (Optimized program of personalized treatment of type 2 diabetes) were included. Study size was not prespecified, and results are reported for all patients with T2D who fulfilled the criteria of being a responder or nonresponder or tolerant or intolerant to metformin therapy within this population at the time of the study. Individual methylation markers associated with future metformin response or intolerance were selected on the basis of genome-wide significance in a fixed meta-analysis and then combined in MRSs to better discriminate patients into responders/nonresponders and tolerant/intolerant participants with T2D. Moreover, genetic-epigenetic interaction analyses, cross-tissue methylation patterns, and functional follow-up experiments in hepatocytes in vitro were performed to better understand the role of these epigenetic markers in metformin response or intolerance. More cohort and method details are available in the Supplementary Materials.

Study populations

All participants included in this study were newly diagnosed drug-naïve patients with T2D with available blood samples before the start of metformin therapy. They were selected from the following cohorts: The ANDIS (19) cohort, an ongoing prospective cohort, which aimed to register all new cases of diabetes in Scania for improvement of diagnosis and treatment strategies; The ANDiU cohort (19) (www.andiu.se/), an ongoing prospective cohort, which included anyone who was diagnosed with diabetes and resides in the County of Uppsala, Sweden; and The OPTIMED cohort, which includes patients with T2D from Latvia (58). ANDIS, ANDiU, and OPTIMED were performed in accordance with the Declaration of Helsinki, and written informed consent was obtained from all participants. These cohorts were divided into discovery and replication cohorts (figs. S1 to S3 and tables S1, S2, and S11) and are further described in the Supplementary Materials.

Statistical analyses

All statistical analyses were performed using R Statistical Software. Two-sided P values were used for all the analyses. To evaluate differences between clinical variables, Mann-Whitney U and χ2 tests were performed as appropriate. To assess the association between genome-wide DNA methylation and metformin response or intolerance, linear regression models were fit prior transformation of those variables that were not normally distributed to achieve normality. Linear regression models were fit to identify and replicate epigenetic markers associated with metformin response and intolerance. Methylation markers were selected from the discovery stage to replication testing if FDR was below 5% (q < 0.05). Methylation markers were considered if P < 0.05 in the replication cohorts with directional consistency. We performed combined analyses of discovery and replication data using fixed meta-analysis, and here, we required epigenome-wide significance after Bonferroni correction. Logistic regression models based on these methylation markers were used to assess the risk of not responding to or not tolerating metformin.

We also evaluated whether MRSs could discriminate between glycemic responders and nonresponders as well as between participants tolerant and intolerant to metformin. MRS was calculated as the sum of standardized methylation values at each site associated with metformin response or intolerance, weighted by CpG-specific effect size (22). To validate our findings and control for overfitting, we performed three evaluations, calculating the MRS using CpG-specific effect sizes (beta-coefficients from the logistic models) estimated from one cohort (either the ANDIS discovery or replication cohort, or the European replication cohort) and evaluated in the other two using ROC curves and AUCs.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/12/561/eaaz1803/DC1

Materials and Methods

Fig. S1. Flowchart and participant selection criteria of the ANDIS discovery cohort for metformin response.

Fig. S2. Flowchart and participant selection criteria of the ANDIS replication cohort for metformin response.

Fig. S3. Flowchart and selection criteria of ANDiU and OPTIMED participants for investigation of glycemic response to metformin therapy: “The European replication cohort for metformin response.”

Fig. S4. Combined MRSs discriminate between glycemic responders and nonresponders to metformin in drug-naïve participants with T2D from the ANDIS discovery and the European replication cohorts.

Fig. S5. Combined MRSs discriminate between glycemic responders and nonresponders to metformin in drug-naïve participants with T2D from the ANDIS discovery and the ANDIS replication cohorts.

Fig. S6. ROC curves for response and intolerance to metformin incorporating different clinical baseline phenotypes in all subjects from discovery and replication cohorts for metformin response and intolerance.

Fig. S7. ROC curves for response to metformin incorporating additional clinical baseline phenotypes in ANDIS participants from the discovery and replication cohorts for metformin response combined.

Fig. S8. Combined MRSs discriminate between tolerant and intolerant participants to metformin in drug-naïve participants with T2D from the ANDIS discovery and the European replication cohorts.

Fig. S9. Combined MRSs discriminate between tolerant and intolerant participants to metformin in drug-naïve participants with T2D from the ANDIS discovery and the ANDIS replication cohorts.

Fig. S10. Correlations between DNA methylation in blood and DNA methylation in adipose tissue (n = 28) from the same participant (monozygotic twin cohort).

Fig. S11. In vitro methylation of the SAP130 promoter resulted in decreased transcriptional activity.

Table S1. Clinical characteristics of the full discovery and replication cohorts for metformin response including drug-naïve and newly diagnosed participants with T2D from the ANDIS cohort.

Table S2. Clinical characteristics of case-control discovery and replication cohorts including patients who fulfill the criteria of being glycemic responders and nonresponders to metformin therapy.

Table S3. CpG sites with a significant association (FDR < 5%) between DNA methylation in whole blood before taking metformin and the change in HbA1c after ~1.5 years on metformin in drug-naïve participants with T2D from the discovery cohort (n = 63) (Excel).

Table S4. Comparison of the 2577 significant CpG sites (FDR < 5%) with an association between DNA methylation and the ∆HbA1c after ~1.5 years in drug-naïve participants with T2D from the discovery cohort, with two other linear models (Excel).

Table S5. CpG sites with DNA methylation associated with the change in HbA1c (ΔHbA1c) in both the discovery cohort and in the ANDIS replication cohort for metformin response (Excel).

Table S6. CpG sites exhibiting differences in DNA methylation in whole blood between glycemic responders (n = 26) and nonresponders (n = 21) to metformin therapy in drug-naïve participants with T2D from the discovery cohort (Excel).

Table S7. Comparison of the 7916 significant CpG sites (FDR < 5%) between metformin responders and nonresponders in drug-naïve participants with T2D from the discovery cohort, with three other linear models (Excel).

Table S8. Methylated CpG sites associated with response to metformin in the discovery cohort and in the ANDIS replication cohort for metformin response (Excel).

Table S9. Methylated CpG sites associated with response to metformin in the discovery cohort and in the European replication cohort for metformin response (Excel).

Table S10. Methylated CpG sites associated with response to metformin in the discovery cohort and in both the ANDIS and the European replication cohorts for metformin response (Excel).

Table S11. Clinical characteristics of drug-naïve and newly diagnosed patients with T2D included in the metformin intolerance discovery and replication cohorts.

Table S12. CpG sites exhibiting differences in DNA methylation in whole blood between metformin-intolerant (n = 17) and metformin-tolerant (n = 66) drug-naïve participants with T2D from the discovery cohort (Excel).

Table S13. Comparison of the 9676 significant CpG sites (FDR < 5%) between metformin-tolerant and metformin-intolerant drug-naïve participants with T2D from the discovery cohort, with two other linear models (Excel).

Table S14. CpG sites with DNA methylation associated with intolerance to metformin in the discovery cohort and in the ANDIS replication cohort for metformin intolerance (Excel).

Table S15. CpG sites with DNA methylation associated with intolerance to metformin in the discovery cohort and in the European replication cohort for metformin intolerance (Excel).

Table S16. CpG sites with DNA methylation associated with intolerance to metformin in the discovery cohort and in both the ANDIS and the European replication cohorts for metformin intolerance (Excel).

Table S17. Assessing discrimination between glycemic responders and nonresponders to metformin using SNPs and MRSs associated with metformin response.

Table S18. Assessing discrimination between tolerant and intolerant participants to metformin using SNPs and MRSs associated with metformin intolerance.

Table S19. Clinical characteristics of all study participants in the monozygotic twin cohort (MZ).

Table S20. Available data from the monozygotic twin cohort used for cross-tissue methylation analysis in human tissues in the present study.

Table S21. Promoter sequence upstream of SAP130 inserted into the CpG-free firefly luciferase reporter vector (pCpGL-basic) and used for luciferase experiments.

Data file S1. Tables S3 to S10 and S12 to S16 (Excel).

Data file S2. Raw data from figures (Excel).

References (5970)

REFERENCES AND NOTES

Acknowledgments: We thank Y. Wessman, M. Sterner, E. Nilsson, the SCIBLU genomics facility at Lund University and the Genome Database of the Latvian Population for providing biological material and data for the OPTIMED cohort. We thank E. Pearson for help in the design of statistical models. Funding: This work was supported by grants from the Novo Nordisk Foundation, Swedish Research Council, Region Skåne (ALF), ERC-Co grant (PAINTBOX, no. 725840), H2020-Marie Skłodowska-Curie grant agreement no. 706081 (EpiHope), The Swedish Heart Lung Foundation, EFSD, Exodiab, Swedish Foundation for Strategic Research for IRC15-0067, Swedish Diabetes Foundation, Albert Påhlsson Foundation, and ERC-CoG-2015_681742_NASCENT. The group of Swedish twins was recruited from the Swedish Twin Registry, which is supported by grants from the Swedish Research Council. The funders had no role in study design, data collection, analysis and interpretation, decision to publish, or preparation of the manuscript. Author contributions: S.G.-C., L.G., E.A., and C.L. contributed to the conception of the work. S.G.-C., S.K., P.W.F., M.M., M.U., I.E., J.P., A.V., L.G., J.K., E.A., and C.L. contributed to the data collection. S.G.-C., A.P., M.M., S.K., K.B., E.A., and P.V. contributed to the data analysis. S.G.-C., S.K., and K.B. performed experiments. S.G.-C. and C.L. drafted the article. All authors contributed to the interpretation of data and critical revision of the article. All authors gave final approval of the version to be published. S.G.-C. and C.L. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. S.G.-C. and C.L. are guarantors. Competing interests: S.G.-C., A.P., and C.L. have a pending patent (“DNA methylation level of specific CpG sites for prediction of glycemic response and tolerance to metformin treatment in type 2 diabetic patients,” P5066SE00) on using the degree of DNA methylation of specific sites for prediction of glycemic response and tolerance to metformin treatment in patients with T2D. P.W.F. has received research funding from Boehringer Ingelheim, Eli Lilly, Janssen, Novo Nordisk A/S, Sanofi Aventis, and Servier; received consulting fees from Eli Lilly, Novo Nordisk, and Zoe Global Ltd.; and has stock options in Zoe Global Ltd. P.V. works now at the pharmaceutical company AstraZeneca. The other authors declare that they have no competing interests. Data and materials availability: All data used to generate figures 3, 4, 5, and 6 are available in data file S2. DNA methylation data are deposited at the LUDC repository (www.ludc.lu.se/resources/repository) under the following accession numbers and are available upon request: LUDC2020.08.1 (discovery cohort for metformin response), LUDC2020.08.2 (ANDIS replication cohort for metformin response), LUDC2020.08.3 (discovery cohort for metformin response case control), LUDC2020.08.4 (ANDIS replication cohort for metformin response case control), LUDC2020.08.5 (European replication cohort for metformin response case control), LUDC2020.08.6 (discovery cohort for metformin intolerance), LUDC2020.08.7 (ANDIS replication cohort for metformin intolerance), and LUDC2020.08.8 (European replication cohort for metformin intolerance).
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article