Diabetes reduces Vitamin D - Mendelian randomization

Type 2 Diabetes Causally Reduces Circulating Vitamin D Levels: A Multi-Ancestry Mendelian Randomization Study

Nutrients, 16 June 2026, https://doi.org/10.3390/nu18121944

Madhusmita Rout, Piers Blackett, Dharambir K. Sanghera (University of Oklahoma Health Sciences Center)

PDF

Summary by Claude - June, 2026

The practical takeaway: this large genetic study argues the diabetes–vitamin D arrow points "backwards" from the popular assumption. Rather than low vitamin D causing type 2 diabetes (T2D), the genetic susceptibility to T2D appears to cause low vitamin D.

This is a bidirectional Mendelian randomization (MR) study — a method that uses inherited gene variants as natural "instruments" to probe causality while sidestepping confounders like diet, sun, and supplementation. Sample size was very large: 471,861 people from the UK Biobank (European, South Asian, African) plus 3,486 Asian Indians from the Sikh Diabetes Study, with serum 25(OH)D measured by immunoassay.

Two findings stand out. First, a polygenic score loaded with vitamin D–raising alleles did not lower the risk of T2D or cardiovascular disease in any ancestry — echoing the disappointing supplementation trials (D2d, VITAL). Second, and conversely, a higher genetic risk score for T2D strongly predicted vitamin D deficiency: per 1-SD increase in T2D genetic risk, 25(OH)D fell by about 9.1 nmol/L (≈3.6 ng/mL; 95% CI −8.9 to −9.3; p = 9.5 × 10⁻³¹). The effect held across Europeans, South Asians, and Africans and survived MR-Egger, weighted-median, and leave-one-out sensitivity checks. In the Sikh cohort, 25(OH)D dropped from ~44 to ~29 nmol/L as T2D duration exceeded 11 years.

Proposed mechanisms (not tested here): sequestration of vitamin D in adipose tissue, VDR/CYP27B1/CYP2R1 downregulation and hypermethylation, and urinary loss of D-binding-protein–bound vitamin D via a disrupted Lrp2–Cubn–Dab2 renal complex.

The authors frame low 25(OH)D in diabetics as a marker for secondary cardiovascular prevention rather than a causal lever — though they note good vitamin D status may still modify endocrine and cardiovascular health.

What this does NOT show / limitations: MR tests genetically determined lifetime vitamin D, not supplementation — it cannot rule out benefit from correcting a deficiency in someone who is already deficient, in T2D, in pregnancy, or for non-cardiometabolic outcomes (immune, autoimmune, etc.). Over 80% of the sample was European, weakening the non-EU genetic instruments; the African cohort (n = 3,346) was underpowered for sensitivity analysis. Samples were collected ~2006–2010, and the headline "−9.1 nmol/L" is presented alongside small standardized betas (≈−0.09 on the log scale), so treat the exact magnitude cautiously. Mechanisms remain unproven. No conflicts of interest declared; NIH-funded; CC-BY open access.

Related in VitaminDWiki

The MR result is real but mechanism-blind — at least three non-exclusive pathways could each subtract from 25(OH)D, and they likely stack:

#1 Diabetes ==> downregulates CYP2R1 ==> less D3 converted to 25(OH)D The most "global" lever: CYP2R1 hydroxylates D3 from both skin and gut, so throttling it lowers 25(OH)D regardless of source — fitting a signal seen across populations with very different sun exposure.

#2 Diabetes ==> poor gut ==> less oral/dietary vitamin D absorbed Narrower than #1 — it only affects the oral fraction (skin-made D3 never enters the gut), so it best explains why diabetics are refractory to ordinary oral dosing.

#3 Diabetes/obesity ==> adipose sequestration & volumetric dilution ==> lower blood 25(OH)D Often the largest single contributor, and the main confounder of the "causal" claim: many T2D risk genes act through adiposity, and fat mass dilutes/sequesters vitamin D independently of diabetes.

Strongest evidence that this is acquired (disease state), not just shared genetics: in the Sikh cohort, 25(OH)D fell from ~44 nmol/L at fresh T2D onset to ~29 nmol/L after >11 years of diabetes (p = 5.6 × 10⁻¹⁰) — a duration-dependent erosion that pleiotropy alone would not produce. Honest caveat: the MR signal is about lifetime genetic liability, while this duration effect is about acquired disease, so the two lines of evidence are pointing at the same conclusion from different directions.

How to tell #1 from #2 (a testable prediction): measure the D3-to-25(OH)D ratio. High circulating D3 with low 25(OH)D = a hydroxylation bottleneck (#1, CYP2R1). Low in both = an intake/absorption problem (#2, gut). Route-of-administration does the same job: if sublingual/topical/UVB (all gut-bypassing) still leave 25(OH)D blunted, the limit is enzymatic, not absorptive.