Magnesium is an essential mineral derived primarily from dietary sources, including whole-grains, green leafy vegetables, legumes, and nuts 1. Adequate magnesium intake is critical in maintaining magnesium balance and normal magnesium-dependent cellular reactions in the human body. Magnesium deficiency has been correlated with a cluster of metabolic abnormalities and associated chronic diseases such as oxidative stress, systemic inflammation, endothelial dysfunction, insulin resistance, hypertension, type 2 diabetes, and coronary heart disease 23456. Although the precise mechanisms underlying magnesium metabolism are unclear, magnesium homeostasis in the human body seems to be tightly regulated via the balance between intestinal absorption and renal excretion. Recent studies suggest that transient receptor potential membrane melastatin 6 and 7 (TRPM6 and TRPM7), two members of the "transient receptor potential" (TRP) family of cation channels, may play a central role in the regulation of magnesium homeostasis 78910.

The TRPM6 gene, located on chromosome 9q21.13, comprises 39 exons spanning 163 kb of genomic sequence and encodes 2022 amino acids 11. TRPM6 is thought to be an ion channel subunit primarily expressed in intestinal epithelia and kidney tubules that may play an important role in intestinal and renal magnesium handling 1213. Several loss-of-function mutations in TRPM6 have been identified among patients with autosomal-recessive familial hypomagnesemia with secondary hypocalcemia 121314. TRPM7, located on chromosome 15q21, spans 127 kb of DNA sequence and encodes a protein of 1865 amino acids. TRPM7 is ubiquitously expressed in various tissues or cell lines 89 and may be part of a magnesium sensing and/or uptake mechanism underlying cellular magnesium homeostasis 89. Although TRPM6 and TRPM7 share 52% homology 15, their biological functions are not redundant, and they can form functional heterometic channels with different permeability 16. Thus, these two genes are candidate genes for evaluating the genetic basis of hypomagnesemia-related metabolic traits such as the common form of type 2 diabetes. To date, no study has assessed common genetic variation in these two genes or their contribution to susceptibility to chronic metabolic disorders highly correlated with hypomagenesemia. Furthermore, a magnesium-deficient diet was shown to upregulate TRPM6 mRNA expression in mice 17. Low serum magnesium levels caused by TRPM6 mutations among HSH patients can be ameliorated by oral supplementation of high doses of magnesium 18. Thus, it is important to further examine in human population data whether magnesium intake modifies any common genetic effects of TRPM6 and TRPM7 on diabetes risk.

We therefore conducted a prospective nested case-control study among postmenopausal women in the Women's Health Study to investigate the association between common variations of the TRPM6 and TRPM7 genes and the risk of type 2 diabetes. We specifically tested the following hypotheses: (1) common genetic variants in TRPM6 and TRPM7 are associated with risk of developing type 2 diabetes because of their critical functions on maintaining magnesium homeostasis; and (2) The genetic association, if any, is more pronounced among women with inadequate magnesium intake.

Basic characteristics of the population by case-control status are shown in Table 2. Overall, diabetes cases had a higher prevalence of traditional diabetes risk factors at baseline than controls. The location, minor allele, and allele frequency for all 25 SNPs are shown in Table 1. Their observed genotype frequencies, except for rs7867868 (P = 0.03), were within HWE among the controls. We examined whether each of the SNPs chosen was significantly associated with type 2 diabetes under three genetic models (additive, recessive, and dominant models). As showed in Table 1, only one SNP in TRPM6 was significantly associated with diabetes risk in the recessive model (rs2769195, P = 0.03). Of the 5 TRPM7 SNPs, rs3109881 and rs3098198 showed significant associations with diabetes risk (rs3109881, additive P = 0.04; and rs3098198, additive P = 0.03 and dominant P = 0.03). These results were likely to be false positives, because they did not remain statistically significant after further adjustment for more covariates or multiple comparisons.

The LD structure and haplotype blocks were shown for 20 SNPs in TRPM6 (Figure 1) and 5 SNPs in TRPM7 (Figure 2) among controls. The 20 SNPs in TRPM6 fall into six blocks with high LD and different sizes. Notably, the two non-synonymous SNPs (rs3750425 and rs2274924) in TRPM6 were 762 bp apart and yet not in complete LD (D' = 0.98; r² = 0.46). For TRPM7, rs8042919 (Ile1482Thr in exon 29) was in tight LD with all other SNPs in TRPM7 (D' = 1.00; r² = 0.10–0.59).

Haplotype analyses of the 20 SNPs spanning the TRPM6 region did not reveal any significant associations. There was no evidence for haplotypic associations with all derived haplotypes based on all 20 TRPM6 SNPs and 5 TRPM7 SNPs. Haplotype analysis in each of these blocks did not provide evidence of statistically significant associations with diabetes risk (data not shown). Overall, a global test for differences in risk according to common haplotypes (≥ 1%) inferred from 20 SNPs over TRPM6 was not statistically significant (P = 0.31). There was no significant association with haplotypes in any single block: p values for global testing of haplotype associations were 0.69 for block 1, 0.96 for block 2, 0.12 for block 3, 0.92 for block 4, 0.85 for block 5, and 0.33 for block 6. Neither was there any evidence of association between common TRPM7 haplotypes and diabetes risk (P for global test = 0.43).

We conducted a sliding window (window width = 2) haplotype-based analyses to examine the genetic associations of TRPM6 with risk of type 2 diabetes. There were a total of 19 window frames for the 20 SNPs. Overall, there were no significant associations for any 2-SNP "sub-haplotype" in any case-control samples (data not shown). When we stratified all cases and controls by magnesium intake, only one 2-tSNP "sub-haplotype", rs3750425-rs2274924 (χ² = 24.14, d.f. = 7, p = 0.0011), reached global significance (i.e. -log₁₀ p > 2.88) for women with magnesium intake in the lowest 20% (<250 mg/day) (Figure 3). These two SNPs were both non-synonymous and in a tight LD block, with the frequency of carriers for both rare alleles of these two SNPs (the A allele of rs3750425 and the G allele of rs2274924) representing 9.3% in all the control samples. As shown in Table 3, there were no significant associations between any individual non-synonymous SNP and diabetes (under dominant models), among women with either low or high magnesium intake. Only among women with low magnesium intake was the haplotype based on 2SNPs in TRPM6 significantly associated with diabetes risk, although statistical power in our subgroup analyses became weak. Among women with low magnesium intake, ORs and 95% CIs in a comparison with carriers of both common alleles (G-A) were 4.92 (1.05–23) for the haplotype A-G and 0.56 (0.09–3.52) for the haplotype G-G. Adjustment for more potential confounders did no materially change the results.

In addition, our bioinformatics analyses provided some indirect evidence for prediction of the effects of an amino acid substitution SNP on protein structure and function (Figure 4 and Table 4). One non-synonymous SNP (rs2274924; K1584E) in TRPM6 seemed to be evolutionarily conservative and were predicted to be neutral using PMut prediction algorithm; however, measures of Polyphen and SIFT scores predicted that rs3750425 (V1393I) in TRPM6 and rs8042919 (T1482I) in TRPM7 might have a negative impact on protein activity. It should be noted that these algorithms are unable to predict the impact of multiple amino acid substitutions in a subunit on the activity of the protein. Thus, we cannot completely rule out the possibility that haplotypes based on multiple coding SNPs are functionally important.

Overall, we observed no significant associations between the TRPM6 and TRPM7 genetic variants, individually or jointly, with risk of type 2 diabetes in women. However, women who were homozygous carriers of both the rare alleles of two non-synonymous SNPs in TRPM6 and had low magnesium intake had a significantly higher risk of type 2 diabetes, although these findings may have been due to chance and further large-scale studies are needed to confirm these subgroup analyses.

To the best of our knowledge, this is the first population study to comprehensively investigate TRPM6 and TRPM7 as potential candidate genes for the common form of type 2 diabetes. On the basis of the publicly accessible dataset, we focused on common SNPs including intronic SNPs, synonymous SNPs, and non-synonymous SNPs to characterize genetic variation spanning the TRPM6 gene, including at least 30 kb upstream and downstream of the coding regions. There may be some unknown causal SNPs including splicing SNPs and SNPs in the promoter region of TRPM6 that likely regulate gene expression. However, we could not identify regulatory SNPs without functional data and did not find any common SNP in the promoter region. Although we did not sequence genomic regions in TRPM6 in our study samples, empirical data have demonstrated that tSNPs selected from the HapMap population samples can effectively capture common variation and provide good power to detect a modest association in many other independent samples 38. We cannot completely rule out the possibility of rare but functional variants (with an allele frequency <5%). Nevertheless, such rare alleles may not be major contributors to the most common late-onset form of type 2 diabetes with a high prevalence and incidence in the general population.

Our most significant statistical finding focused on two coding SNPs in TRPM6 among women with insufficient magnesium intake (at the lowest 20 percentile) (i.e. <250 mg/d), close to three quarters of the RDA for adult women. Among women, only those with low magnesium intake were found to have increased risk of type 2 diabetes. In our study population, approximately 33% are heterozygous for the rare G allele of rs2274924 and 18% heterozygous for the allele A of rs3750425. Both SNPs are also very common in populations with various genetic backgrounds; the HapMap population frequencies are 0.07–0.21 for the rs3750425 A allele (1393 Ile) and 0.15–0.36 for the rs2274924 G allele (1584Glu) in four ethnic groups. Two changed amino acids are located between the coiled region and kinase near the C-terminal 14. Although such two amino acid substitutions in this region are unlikely responsible for direct regulation of the TRPM6 channel trans-membrane structures and kinase activity, they may alter protein conformation and thus reduce TRPM6 channel activity. Our bioinformatics approaches indicated that these coding SNPs in highly conserved region might be functionally significant, although further functional analysis should be warranted. There is also the possibility that these SNPs are in high LD with untyped SNPs that directly affect channel activity. Alternatively, two amino acid substitutions may be necessary to influence channel protein structure and function. Individuals with both rare alleles, whose TRPM6 channel function is deficient in magnesium absorption, may be more prone to magnesium deficiency through low intake and in turn to an increased risk of diabetes. If our finding is replicated in future studies, it will suggest that common genetic variation in the TRPM6 locus known to harbor severe mutations causing monogenic magnesium deficiency confers a modest susceptibility to the risk of type 2 diabetes in a small subgroup of the general population.

We found no significant evidence for an association between TRPM7 genetic variants and diabetes. Given the limited number of genotyped SNPs (n = 5) across TRPM7 (128 kb) from the reference panel (HapMap database), it is likely that they are not sufficient to capture the vast majority of the genetic variability of TRPM7. Biologically, TRPM7 is ubiquitously expressed and its constitutive activation is required for cellular survival. Animal studies showed that dietary magnesium restriction did not alter TRPM7 mRNA expression in mouse kidney and colon 17. We therefore hypothesize that TRPM7, as a housekeeping gene regulating cellular magnesium metabolism, may truly have limited genetic variability.

The precise mechanisms for the regulatory role of TRPM6 and TRPM7 in magnesium homeostasis remain largely undefined. TRPM6 is a magnesium-permeable channel protein localized to the apical domain of the distal convoluted tubule and the brush-border membrane of the absorptive cells in intestine and plays an important role in active magnesium handling in the small intestine and kidney. Rare mutations in TRPM6, either incomplete or complete loss-of-function, have been associated with cases of autosomal-recessive hypomagnesemia with secondary hypocalcemia 1213. Clinical data show that sufficient magnesium intake can partially compensate for the severe magnesium deficiency caused by genetic defect in the TRPM6 gene 13. It has been hypothesized that intestinal magnesium absorption occurs via two different pathways: an active transcellular transport and a passive paracellular passive transport 13. High magnesium concentrations in intestinal tracts may overcome the genetic defects of magnesium absorption and independently increase magnesium absorption via the passive paracellular pathway. In contrast, when dietary magnesium intake is inadequate, the function of TRPM6 in active intestinal magnesium absorption and renal reabsorption become very important. Further well-designed functional studies are needed to investigate the molecular consequences of common genetic variability of TRPM6 and TRPM7 on mRNA expression, splicing, and degradation and metabolic profiles.

The strengths of our study include a prospective study design with up to 10 years of follow-up, careful collection of DNA samples and baseline variables, and an effective matching strategy used in a well-characterized study population. However, several limitations deserve further consideration. First, our study had relatively small sample size and did not have the power to assess modest genetic effects or interactions. The associations we identified could have been caused by chance due to statistical fluctuation; however, consistency with laboratory findings makes our results plausible. Second, population stratification may be a concern but unlikely to explain our findings, because our populations are racially homogeneous, with the majority of the participants being white (>92.5%). Third, we lacked stable and reliable measures of extracellular or intracellular magnesium status. Serum magnesium may be a good indicator for severe magnesium deficiency but is not sensitive to suboptimal magnesium status. Finally, findings in this cohort of women of predominantly European descent could not address population-specific genetic effects.

Our results suggest that two common non-synonymous TRPM6 coding region variants might interact with magnesium intake in determining the risk of type 2 diabetes. However, our findings could be caused by chance due to multiple comparisons and insufficient statistical power; additional studies are needed to confirm these results and to further characterize the effects of genetic variation in TRPM6 on magnesium homeostasis.

The authors declare that they have no competing interests.

YS was responsible for the conception and design of the study. JEB and SL were responsible for acquisition of data. YS carried out the statistical analyses and drafted the manuscript. YHH and TN oversaw all the population genetics analyses and bioinformatics analyses and data interpretation. YHH, TN, JEM, JEB, and SL critically revised the manuscript for important intellectual content. YS, JEM, JEB, and SL obtained funding and provided administrative, technical, and material support. All authors read and approved the final manuscript.