Genetic Evidence Finds No Clear Vitamin D–Diabetes Link in South Asians

Genetically predicted vitamin D levels show no statistically significant association with the risk of type 2 diabetes in South Asian populations, according to an exploratory two-sample Mendelian randomization study. The primary analysis, which used inverse variance weighting, produced an odds ratio of 1.06 per one standard deviation increase in genetically predicted serum 25-hydroxyvitamin D, with a 95 percent confidence interval of 0.83 to 1.36 and a p-value of 0.630. Because that confidence interval spans both protective and harmful directions, the result indicates an absence of a detectable association within the precision afforded by current genetic instruments, rather than definitive evidence that no causal effect exists. The finding matters because South Asian populations carry one of the world’s heaviest burdens of both vitamin D deficiency and type 2 diabetes, and observational studies have long suggested that low vitamin D status might contribute to diabetes risk in this group.

The motivation for the study rests on a striking overlap of two public health problems. A systematic review and meta-analysis pooling South Asian data estimates that roughly 73 percent of people in Pakistan have vitamin D deficiency, with a 95 percent confidence interval of 63 to 83 percent, although substantial between-study heterogeneity, reflected in an I-squared value of approximately 99 percent, means individual cohorts vary widely. That deficiency reflects dietary insufficiency, cultural practices that reduce sun exposure, and lower cutaneous vitamin D synthesis associated with darker skin pigmentation. At the same time, Pakistan faces one of the highest burdens of type 2 diabetes worldwide, with approximately 26 percent of adults affected, or more than 33 million people. The co-occurrence of these conditions has fueled speculation that correcting vitamin D status could help curb the diabetes epidemic, but correlation alone cannot settle the question.

Biologically, the hypothesis is plausible. Vitamin D receptors are expressed in pancreatic beta cells, suggesting a role in beta-cell function and insulin secretion, and experimental evidence indicates that vitamin D may improve peripheral insulin sensitivity and reduce systemic inflammation, both central processes in type 2 diabetes pathogenesis. Observational studies across diverse populations have accordingly reported inverse associations between serum 25-hydroxyvitamin D concentrations and diabetes risk. Yet such associations are vulnerable to residual confounding by obesity, physical inactivity, and poor diet, and to reverse causation, since people with diabetes or its complications may spend less time outdoors and therefore have lower vitamin D levels. Biological plausibility and observational consistency together still cannot establish causation, which is where genetic instruments enter the picture.

Mendelian randomization addresses these limitations by using germline genetic variants as instrumental variables. Because variants are fixed at conception, they are largely immune to the confounding and reverse causation that plague conventional observational studies. The researchers employed a two-sample design, drawing single nucleotide polymorphism to exposure associations from one genome-wide association study and single nucleotide polymorphism to outcome associations from another. The exposure data came from the largest available South Asian vitamin D genome-wide association study, drawn from UK Biobank and comprising 7,380 participants, while the outcome data came from a South Asian type 2 diabetes meta-analysis of 50,533 participants, including 16,677 cases, which combined 16 cohorts. Cases were defined by a history of type 2 diabetes or its treatment, fasting plasma glucose of at least 7.0 millimoles per liter, two-hour oral glucose tolerance test glucose of at least 11.1 millimoles per liter, or HbA1c of at least 6.5 percent.

The instrument set was small. Only one variant, rs3755967 in the GC region encoding vitamin D-binding protein, met the conventional genome-wide significance threshold for circulating 25-hydroxyvitamin D, with a p-value of 7.4 times ten to the minus nine. To enable pleiotropy-robust sensitivity analyses, which require at least three variants, the researchers added two further variants meeting a relaxed exploratory threshold of p less than one times ten to the minus six: rs2459139 near CYP2R1, the 25-hydroxylase gene, and rs72744878 near DHCR7, involved in vitamin D synthesis. All three F-statistics exceeded the conventional weak-instrument threshold of 10, ranging from 26.4 to 33.4 with a mean of 29.6. However, the total proportion of variance in vitamin D levels explained by the instruments was only 1.06 percent, far below the 5 to 10 percent typically achieved in larger European datasets using hundreds of variants, which widens the causal confidence intervals considerably.

The primary result was null, and the sensitivity analyses largely agreed. The weighted median estimate yielded an odds ratio of 1.06, the simple mode 1.29, and the weighted mode 0.93, all statistically non-significant and close to the null. Cochran’s Q test found no significant heterogeneity across variant-specific estimates, and the MR-Egger intercept provided no evidence of directional pleiotropy, though with only three instruments these tests have very limited statistical power and a non-significant intercept cannot be taken as reassurance that pleiotropy is absent. The MR-Egger point estimate itself was wildly imprecise, an odds ratio of 32.0 with a confidence interval stretching from 0.11 to 9,103, reflecting estimation instability with a single residual degree of freedom rather than support for any particular effect. When the researchers repeated the analysis with relaxed instrument-selection thresholds yielding 68 and 18 variants, results remained directionally consistent with the primary analysis, and the larger instrument sets produced more informative diagnostics showing no significant heterogeneity or directional pleiotropy.

A pre-specified sensitivity analysis using a BMI-adjusted version of the outcome dataset produced essentially identical results, with an inverse variance weighted odds ratio of 1.06 and a confidence interval of 0.83 to 1.36, suggesting that conditioning on body mass index did not materially alter the estimate. The power calculations, however, explain why caution is warranted. Given the outcome sample size and the 1.06 percent variance explained, the study had approximately 18 percent power to detect an odds ratio of 1.1, roughly 51 percent power for an odds ratio of 1.2, and greater than 99 percent power only for an odds ratio of 1.5. In other words, the analysis was underpowered for small effects and moderately powered for moderate ones, which is precisely why the confidence interval spans both protective and harmful directions despite the large outcome sample.

The findings align with prior genetic evidence. A large European-ancestry Mendelian randomization analysis found no significant causal effect of circulating 25-hydroxyvitamin D on incident type 2 diabetes, and a previous study using a composite vitamin D genetic score found no significant association in South Asian or European populations, though a signal from the DHCR7 locus was noted. The randomized trial literature is more nuanced. The D2d trial, which enrolled adults with prediabetes regardless of baseline vitamin D status rather than selecting participants for vitamin D insufficiency, found that vitamin D3 supplementation did not significantly reduce diabetes risk in its primary analysis. Yet a subsequent individual-participant-data meta-analysis pooling D2d with two other prediabetes prevention trials reported an adjusted 15 percent relative reduction in diabetes risk with supplementation among adults with prediabetes. These literatures address different questions: Mendelian randomization evaluates lifelong, genetically predicted differences across the population distribution, whereas supplementation trials test a clinical intervention over a limited period in high-risk participants, so a benefit concentrated in people with prediabetes would not necessarily appear as a linear population-wide genetic effect.

Several limitations temper the conclusion. The South Asian vitamin D exposure dataset is small, forcing a relaxed selection threshold for two of three instruments and raising the possibility of winner’s curse, in which variant effects and instrument strength appear inflated because they were selected from the same data used to estimate them. Standard linear Mendelian randomization also cannot detect non-linear or threshold effects, which may be especially relevant where severe deficiency is common, and the effect of vitamin D might only emerge at very low 25-hydroxyvitamin D concentrations. Sample overlap is another concern: the exposure data come entirely from UK Biobank, and UK Biobank was one of the 16 cohorts contributing to the outcome meta-analysis. A bounding analysis estimated that, under a conservative worst-case overlap proportion of 0.146 and an assumed phenotypic correlation of minus 0.12, the relative bias from overlap would be approximately 0.32 percent, about 5.6 percent of the observed estimate, suggesting overlap is unlikely to materially change the interpretation, though this remains a theoretical assessment. The UK-resident South Asian participants may also differ systematically from populations in South Asia through healthy migrant selection and dietary acculturation.

For public health, the implications are measured. Supplementation programs specifically aimed at reducing the incidence of type 2 diabetes are not supported by current causal evidence, and resources may be more efficiently directed toward interventions with established benefit, including lifestyle modification, dietary reform, and pharmacological prevention in high-risk individuals. The findings do not preclude vitamin D benefits for other outcomes such as bone health or immune function, though the large VITAL randomized trial did not find a reduction in cardiovascular events with supplementation, so cardiovascular benefit should not be assumed. Taken together, the results suggest that the inverse relationship between low vitamin D and type 2 diabetes observed in South Asian populations may partly reflect confounding or reverse causation, and that correcting deficiency is unlikely to substantially reduce diabetes incidence unless future evidence identifies a specific high-risk subgroup that benefits. Larger South Asian genome-wide association studies, non-linear Mendelian randomization approaches, and analyses of continuous metabolic traits such as fasting glucose and insulin resistance indices will be needed to clarify whether small, threshold-dependent, or subgroup-specific effects exist at the deficiency levels most prevalent in the region.

Subject of Research: Mendelian randomization analysis of genetically predicted vitamin D levels and type 2 diabetes risk in South Asian populations

Article Title: Association Between Genetically Predicted 25‐Hydroxyvitamin D and Type 2 Diabetes Risk in South Asians: An Exploratory Two‐Sample Mendelian Randomization Study

Article References: Zaka, A., Khan, S., Shah, A., Khalid, A. A., Khalil, I., & Kamrul‐Hasan, A. B. M. (2026). Association Between Genetically Predicted 25‐Hydroxyvitamin D and Type 2 Diabetes Risk in South Asians: An Exploratory Two‐Sample Mendelian Randomization Study. Endocrinology, Diabetes & Metabolism, 9(6), Article e70359. https://doi.org/10.1002/edm2.70359

Image Credits: AI Generated

DOI: 10.1002/edm2.70359

Keywords: vitamin D, type 2 diabetes, Mendelian randomization, South Asian populations, 25-hydroxyvitamin D, genetic instruments, GWAS, Pakistan, insulin sensitivity, observational confounding, vitamin D supplementation, instrument strength