A Two-Stage Study Identifies Two Novel Polymorphisms in PRKAG2 Affecting Metformin Response in Chinese Type 2 Diabetes Patients

Introduction

Type 2 diabetes (T2D) is a common chronic metabolic disease that is harmful to public health. The 2019 International Diabetes Federation (IDF) Diabetes Atlas reported 116.4 million diabetics aged 20 to 79 years in China, making it the country with the highest number of diabetes sufferers in the world.¹ Among adults in China, the estimated overall prevalence of diabetes is 10.9%, including diagnosed and undiagnosed cases.² Yet, according to the latest epidemiological studies, only 25.8% of definitely diagnosed patients were receiving antidiabetic therapy, and only about 40% of patients were under favorable glycemic control.³

Oral antidiabetic drugs (OADs) can be classified as follows: biguanide (metformin is the only biguanide in general use), second-generation sulfonylureas (SUs), meglitinides, thiazolidinediones (TZDs), α-glucosidase inhibitors, dipeptidyl peptidase-4 (DPP-4) inhibitors, and sodium-glucose cotransporter 2 (SGLT2) inhibitors.⁴ Among these, metformin is the most widely used agent owing to its high efficacy, neutral/mild weight loss, low cost, and rare side effect of hypoglycemia.⁴ The American Diabetes Association continued to advise in 2020 that metformin is the preferred initial pharmacologic agent for type 2 diabetes and should be used up to contraindication or intolerance.⁵ Individual differences in glycemic response to metformin in antidiabetic treatment exist widely. Less than half of the T2D patients treated with metformin could reach their HbA_1c target (˂7%) and 30% experienced an adverse gastrointestinal reaction.^6–8

Metformin is not metabolized in vivo and is excreted unchanged in urine. Pharmacogenomics of metformin previously focused mainly on genetic variants of its transporters. SNPs within organic cation transporter (OCT) 1–3 (encoded by SLC22A1, SLC22A2, SLC22A3, separately)^9–12 and multidrug and toxin extrusion (MATE) 1/2-k (encoded by SLC47A1/ SLC47A2),^13,14 plasma monoamine transporter (PMAT; encoded by SLC29A4),¹⁵ serotonin reuptake transporter (SERT; encoded by SLC6A4),^8,16 as well as thiamine transporter (THTR-2; encoded by SLC19A3)¹⁷ were reported to take part in the drug disposal process of metformin. Thus, genetic variants of transporters mentioned above probably have an impact on metformin pharmacokinetics, accompanied or not by an influence on pharmacodynamics.

Over the past few decades, about 50 single nucleotide polymorphisms (SNPs) have been found likely to affect its glycemic response, including several genetic variants identified by genome-wide association study (GWAS). These were rs11212617 closed to ATM (a regulator of the target of metformin, AMPK),¹⁸ rs8192675 in SLC2A2 (the coding gene of a glucose transporter, GLUT2),¹⁹ rs254271 in PRPF31 (pre-mRNA processing factor 31) and rs2162145 in CPA6 (carboxypeptidase A6).²⁰ Taking rs11212617 near ATM as an example, several investigators attempted to conduct replication and meta-analysis of this locus to confirm its influence, but the results were inconsistent.^21–23 Moreover, these high throughput screening researches were all conducted in a multiethnic population, among which Asians made up a small proportion or were not included.

We used a candidate gene approach, involving thousands of SNPs, to explore the characteristic genetic variants that affect metformin’s glycemic response in Chinese T2D patients.

Methods

Study Participants

Data for this study were obtained from two trials. One is the “Glibenclamide” arm of the Xiaoke Pill Trial, described in detail by Ji et al²⁴. The other is a group of newly diagnosed T2D patients that received metformin monotherapy.

A total of 365 patients were recruited for the “Glibenclamide” arm. Among these, 182 received a combination treatment of metformin plus glibenclamide. We called it the “combination treatment group”, or “discovery group”. For this group, glycometabolism measurements were assessed at baseline and then every 12 weeks until the trial’s termination. Glibenclamide doses were adjusted according to changes of FPG level every four weeks, and metformin doses remained unchanged throughout the trial. Another 183 patients were treatment naïve T2D cases who received glibenclamide monotherapy. We named this as the “glibenclamide monotherapy group”, or “exclusion group”. Dose adjustment was similar to the above (Trial no. ChiCTR-TRC-08000074).

As for the metformin group, 145 newly diagnosed and drug-naïve T2D patients received metformin monotherapy for 16 weeks. We called it the “metformin monotherapy group” or “validation group”. Glycometabolism measurements were evaluated at baseline and at the ending point (Trial no. NCT00778622).

Phenotype Definitions

Referring to Zhou et al¹⁸ we used the change of HbA_1c level (on-treatment HbA_1c level minus pre-treatment HbA_1c level), adjusted by known clinical covariates, as the glycemic response phenotype. On-treatment HbA_1c was defined as the minimum recorded HbA_1c achieved within 36 weeks after the index date in the “combination treatment group” and “glibenclamide monotherapy group”. The covariates included age, sex, weight, serum creatinine (Scr), baseline HbA_1c level, and drug doses. If the first four covariates were all available, the creatinine clearance rate (Ccr) would be recommended as a whole instead of being adjusted separately. The Ccr was calculated as the following equation: (140-age) × weight (in kg) × (0.85 if female)/(0.818 * Scr (in μmol/L). Drug dose was defined as the average daily dose during the three months prior to the minimum HbA_1c being achieved.

Genotyping

Infinium iSelect HD Custom Genotyping BeadChips and Illumina GoldenGate genotyping platforms were used for patient genotyping. SNPs were primarily selected on the basis of pharmacokinetics and pharmacodynamics, as well as reported disease-related SNPs, such as diabetes, obesity, glucose, and lipid metabolism. The 20 top-ranked GO (Gene Ontology) biological process and KEGG (Kyoto Encyclopedia of Genes and Genomes) Pathway analysis of SNP lists are presented in Figure S1. For iSelect BeadChip, SNP selection was based primarily on the DMET (Drug Metabolizing Enzymes and Transporter; Affymetrix) chip, with some extension. As for GoldenGate BeadChip, SNPs were selected mainly direct to metformin. Genes likely to affect metformin pharmacokinetics and pharmacodynamics, confirmed or speculated, were enrolled. In total, 2986 SNPs were included in the iSelect BeadChip, while 768 SNPs were customed into the GoldenGate BeadChip. For comparability between the two chips, 15 SNPs were customed into both. Because the “combination treatment group” was at the discovery stage, genotyping by both chips was undertaken. Subsequently, the “glibenclamide monotherapy group” used the iSelect chip only because glibenclamide-related genes were involved in this chip, while the “metformin monotherapy group” utilized the GoldenGate Chip for the same reason.

Statistical Analysis

Before genetic association analysis, SNP quality control (QC) and sample QC were performed in three groups. For each SNP, simultaneously satisfying call rate ≥90% and MAF (minor allele frequency) ≥0.05 and Hardy–Weinberg equilibrium (HWE) test P values > 0.5 were filtered. For each sample, a genotyping call rate ≥90% was retained for subsequent analyses. Stepwise linear regression was utilized to select clinical covariates of potential effects. Linear regression model was performed to test associations between each SNP and drug efficacy. The Bonferroni correction was used for multiple testing corrections to adjust raw P values. All the above analyses were achieved by using plink 1.07 (http://pngu.mgh.harvard.edu/purcell/plink/) and SPSS 20.0 (SPSS Inc., Chicago, Illinois, USA).

Results

Results of SNP Selection and Genotyping

A total of 2986 SNPs and 768 SNPs were included in iSelect BeadChip and GoldenGate BeadChip, separately. The accordance ratio of the 15 reduplicative SNPs was over 98%. In the “combination treatment group”, 551 SNPs in iSelect chip and 645 SNPs in GoldenGate Chip passed SNP and sample filtering. In the “glibenclamide monotherapy group”, 545 SNPs in iSelect chip passed filtering, while in the “metformin monotherapy group”, 644 SNPs in GoldenGate chip passed filtering. The screening process is shown in Table S1.

Characteristics of Study Populations

Detailed demographics are shown in Table 1. After strictly excluding patients who did not meet the entry criteria but were recruited, there were 176 patients (90 males and 86 females) in the combination treatment group (discovery group), 181 patients (106 males and 75 females) in the glibenclamide monotherapy group (exclusion group) and 143 patients (84 males and 59 females) in the metformin monotherapy group (validation group). Baseline age, weight, BMI, and waist/hip ratio are listed in Table 1. A relatively higher proportion of overweight and obese individuals were observed in the validation group. The baseline FPG levels were, respectively, 9.36±1.71, 9.02±1.57, and 8.50±1.80 mmol/L in the discovery, exclusion, and validation groups in sequence. The baseline HbA_1c levels in turn were 8.47±1.26, 8.34±1.22, and 8.32±0.82%. The on-treatment HbA_1c refers to the minimum HbA_1c level during visits, and the level was 6.78±0.99, 6.56±0.90, 6.53±0.54% in sequence. Correspondingly, medication daily dose was the average daily dose for three months prior to the minimum HbA_1c being achieved. For the discovery group, the glibenclamide daily dose was 2.50 mg (2.08–5.00 mg) (IQR, 25th and 75th percentile, the same as below) and the metformin daily dose was 1000 mg (750–1500 mg). For the exclusion group, the glibenclamide daily dose was 3.75 mg (2.50–5.00 mg). For the validation group, the metformin daily dose was 1500 mg (1500–2000 mg).

Table 1 Demographics of Study Populations

Results of Genetic Association Analysis

The integrated workflow is shown in Figure 1.

Figure 1 The design workflow of this clinical study.

First, we established the association between genotypes and drug response in the discovery group. We merged genotyping data of two platforms and redid SNP and sample QC. 1245 SNPs passed filter, including 14 reduplicative loci between the two platforms, so that the number of enrolled SNPs was 1231. After adjustment for baseline HbA_1c level, Ccr, and medication daily dose, 50 SNPs were found to be associated with the change of HbA_1c value (P˂0.05, shown in Table 2). Among these probably positive loci, 60% were from the GoldenGate chip.

Table 2 SNPs Associated with Phenotype in Discovery Group (Stage 1)

Next, associations between genotype and phenotype in the exclusion group were analyzed. 19 of 545 SNPs were found related to glibenclamide response (P˂0.05, shown in Table 3), among which was rs1800545 in ADRA2A (adrenoceptor alpha 2A) with P value less than 0.05 in both groups above. Our preference is that this variant is the most likely to affect glibenclamide response.

Table 3 SNPs Associated with Phenotype in Exclusion Group (Stage 1)

Because most SNPs found in the discovery group were derived from the GoldenGate chip, only GoldenGate genotyping was performed on validation group patients using metformin monotherapy. In this group, 27 of 644 SNPs were found to be correlated with metformin glucose-lowering efficacy (P˂0.05, shown in Table 4). Compared with SNPs identified in the discovery group, two variants of the PRKAG2 (protein kinase AMP-activated non-catalytic subunit gamma 2) gene were validated (bold in Tables 2 and 4). One was PRKAG2 rs2727528 (discovery group: β=−0.212, P=0.046; validation group: β=−0.269, P=0.028). The other was PRKAG2 rs1105842 (discovery group: β=0.205, P=0.048; validation group: β=0.273, P=0.025). C allele carriers (W/M+M/M, W=wild type; M=mutation type) of rs2727528 and C allele carriers (W/W+W/M) of rs1105842 would have a larger difference of HbA_1c level when using metformin (shown in Figure 2). Meanwhile, we were concerned that in the metformin monotherapy group, there were 5 SNPs located in the PRKAG2 gene found to be associated with metformin response. Except for the two SNPs mentioned above, the other three were rs1029946 (β=0.306, P=0.001), rs6964824 (β=−0.347, P=0.013), and rs2727551 (β=0.296, P=0.042). Linkage disequilibrium analysis showed that the linkage among the five SNPs was relatively low (shown in Figure 3). In addition, rs11212617 near C11orf65 or ATM, identified by the first metformin GWAS, was repeated in the metformin monotherapy group (β=−0.255, P=0.035), while C allele carriers benefited more in our research.

Table 4 SNPs Associated with Phenotype in Validation Group (Stage 2)

Figure 2 Proportional reduction in HbA1c by PRKAG2 rs2727528 and rs1105842 genotypes as represented by violin plots. Proportional reduction in HbA1c was calculated as (on-treatment HbA1c level minus pre-treatment HbA1c level)/pre-treatment HbA1c level. (A) Proportional reduction in HbA1c among PRKAG2 rs2727528 different genotypes in discovery group; (B) Proportional reduction in HbA1c among PRKAG2 rs2727528 different genotypes in validation group; (C) Proportional reduction in HbA1c among PRKAG2 rs1105842 different genotypes in discovery group; (D) Proportional reduction in HbA1c among PRKAG2 rs1105842 different genotypes in validation group.

Figure 3 Linkage disequilibrium analysis of 5 SNPs (rs1029946, rs1105842, rs2727528, rs2727551, rs6964824) in PRKAG2 in validation group. (A) D’ of the 5 SNPs in PRKAG2; (B) r2 of the 5 SNPs in PRKAG2.

Discussion

To our knowledge, the current study is the first to use high-throughput genotyping chips to identify candidate SNPs, which may affect metformin response in Chinese T2D patients through a two-stage study. Three groups totaling 500 patients met the final selection criteria and were analyzed. Previous studies on metformin pharmacogenomics were mostly carried out in patients receiving combination therapy, with at least one more antidiabetic drug being added to metformin. Even if subjects were metformin monotherapy patients, or considering monotherapy patients as a subgroup, the sample size was usually relatively small. This was understandable for at least two reasons. First, many T2D patients have progressed to such a degree that a single drug could not well control at the time of diagnosis. That is why we emphasize screening for diabetes. Second, as described in the introduction, a proportion of patients do not respond well to metformin or cannot tolerate its side effects. To cripple interference from the combined drugs, we individually recruited commensurate patients for treatment with the specified antidiabetic drug. We validated our results at the discovery stage in metformin monotherapy patients. All the above was to strengthen the credibility of verified SNPs in affecting metformin response in Chinese T2D patients.

In the discovery group, we screened out 50 SNPs nominally associated with the change of HbA_1c value. Although the one with the lowest P value (10⁻³ level) did not pass the Bonferroni test (Bonferroni P value should be less than 4.06×10⁻⁵), potential impacts could be masked. Furthermore, due to the combination of metformin and glibenclamide, we did not know the contribution of each drug in glucose lowering. To minimize the influence, 19 SNPs were identified for association with glibenclamide response in the glibenclamide monotherapy group. It was not surprising that some of them were located in or near “known” genes to affect pharmacokinetics or pharmacodynamics of sulfonylureas, such as IRS1,²⁵ CYP51A1,²⁶ ADRA2A,²⁷ and so on. Due to racial differences in allele frequency, some crucial variants of sulfonylureas like *2 variant (Arg144Cys, rs1799853), *3 variant (Ile359Leu, rs1057910) of CYP2C9²⁸ were rare mutations in Chinese patients, so that they either were not selected in the genotyping chip originally, or did not pass MAF filtering. By comparing the results of the two groups above, one repeated locus was regarded as associated with glibenclamide, but not metformin. Over 60% of the remaining 49 SNPs came from the GoldenGate chip, which was targeted at metformin’s intracorporal process and efficacy. Thus, we decided to verify the remaining SNPs using only the GoldenGate chip. A total of 27 SNPs with a raw P value less than 0.05 and PRKAG2 rs2727528 and rs1105842 were duplicated in both discovery and validation groups. Further analysis indicated that C allele carriers of rs2727528 and C allele carriers of rs1105842 would have a larger difference of HbA_1c level when using metformin. This could mean that patients with prepotent genotype will obtain more benefit from metformin in glucose control. Meanwhile, we were concerned that in the metformin monotherapy group, five SNPs located in the PRKAG2 gene were nominally associated with metformin response, and were in poor linkage with each other. This suggests that PRKAG2 and its variants may contribute more to metformin efficacy than we recognize.

Metformin has been shown to act via both AMP-activated protein kinase (AMPK)-dependent and AMPK-independent mechanisms.²⁹ AMPK is a heterotrimeric complex consisting of a catalytic subunit (α, encoded by PRKAA1 and PRKAA2) and two regulatory subunits (β, encoded by PRKAB1 and PRKAB2; γ, encoded by PRKAG1, PRKAG2, and PRKAG3).^30,31 The γ-subunit harbors nucleotide-binding sites and plays an important role in AMPK regulation in response to cellular energy levels. In mammals, there are three isoforms of the γ-subunit, and these respond differently to regulation by nucleotides.^32,33 A recent study has further reported that humans carrying the R302Q mutation in γ2 have increased adiposity and slightly raised fasting glucose levels compared with unaffected individuals, owing to chronic activation of γ2 AMPK when mutation exists.³⁴ This suggests that mutation could change the state of activation. Genome-wide association studies show that PRKAG2 is significantly associated with diabetes incidence.³⁵ In addition, methylation signatures of cg24061580 (PRKAG2) correlate with insulin resistance.³⁶ Polymorphisms in encoding genes of other subunits, PRKAA1 (encode α1), PRKAA2 (encode α2), and PRKAB2 (encode β2), have been found to affect metformin glucose-lowering effect.³⁷ However, PRKAG2 has been extensively studied mainly for its mutations, which could cause human cardiomyopathy characterized by hypertrophy, Wolff-Parkinson-White syndrome, conduction system disease, and glycogen storage in the myocardium.³⁸ Recent studies have revealed the molecular pathogenesis of cardiac abnormality owing to PRKAG2 mutation. PRKAG2 mutant patients and model mice displayed anomalous atrioventricular conduction related to cardiac glycogen overload. Most likely, the increased AMPK activity caused by active mutation enhanced glycogen synthesis through robust glucose uptake.^39,40 That is, glucose-6-phosphate and the abundant substrate functioned as allosteric activators of glycogen synthase, thus promoting the influx of glucose by AMPK activation to synthesize glycogen. However, because of insulin deficiency and glucagon-induced insulin resistance, diabetics cannot store glucose as liver glycogen, either directly (glycogen synthesis from dietary glucose after meals) or indirectly (glycogen synthesis from “de novo” synthesis of glucose).

Our study found that PRKAG2 rs2727528 and rs1105842 could affect the hypoglycemic effect of metformin in Chinese Han T2D patients. We speculate that the mutation in PRKAG2 might change the conformation or activity of γ2 AMPK, thus altering the rate of gluconeogenesis, glycogen cycling, and hepatic glucose output. Coincidentally, metformin acts primarily by decreasing hepatic glucose output, largely by inhibiting gluconeogenesis.⁴¹ The interaction between metformin and PRKAG2 mutation is fascinating. However, our hypothesis needs to be verified by cell and animal experiments.

There were certain limitations to our study. First, superabundant trivial loci were enrolled when designing the genotyping chip, especially those with very low allele frequency in Chinese people. Second, due to differences in visit times, only 16-week glycometabolism and lipometabolism measures were collected in the metformin monotherapy group. A different course of treatment compared with the discovery group may mask the effects of some meaningful gene variants.

Conclusion

Nevertheless, this is progressive research with a more rigorous grouping and a larger population to screen genetic variants that could affect metformin response in Chinese T2D patients. By correlating the change of HbA_1c levels with thousands of related SNPs, we found that PRKAG2 rs2727528 and rs1105842 polymorphisms may affect metformin response in Chinese T2D patients. The mechanisms of their influence need further research.