Association of iron homeostasis biomarkers in type 2 diabetes and glycaemic traits: a bidirectional two-sample Mendelian randomization study

Abstract

Background

Mendelian randomization (MR) studies show iron positively associated with type 2 diabetes (T2D) but included potentially biasing hereditary haemochromatosis variants and did not assess reverse causality.

Methods

We assessed the relation of iron homeostasis with T2D and glycaemic traits bidirectionally, using genome-wide association studies (GWAS) of iron homeostasis biomarkers [ferritin, serum iron, total iron-binding capacity (TIBC), transferrin saturation (TSAT) (n ≤ 246 139)], T2D (DIAMANTE n = 933 970 and FinnGen n = 300 483), and glycaemic traits [fasting glucose (FG), 2-h glucose, glycated haemoglobin (HbA1c) and fasting insulin (FI) (n ≤ 209 605)]. Inverse variance weighting (IVW) was the main analysis, supplemented with sensitivity analyses and assessment of mediation by hepcidin.

Results

Iron homeostasis biomarkers were largely unrelated to T2D, although serum iron was potentially associated with higher T2D [odds ratio: 1.07 per standard deviation; 95% confidence interval (CI): 0.99 to 1.16; P-value: 0.078) in DIAMANTE only. Higher ferritin, serum iron, TSAT and lower TIBC likely decreased HbA1c, but were not associated with other glycaemic traits. Liability to T2D likely increased TIBC (0.03 per log odds; 95% CI: 0.01 to 0.05; P-value: 0.005), FI likely increased ferritin (0.29 per log pmol/L; 95% CI: 0.12 to 0.47; P-value: 8.72 x 10^–4). FG likely increased serum iron (0.06 per mmol/L; 95% CI: 0.001 to 0.12; P-value: 0.046). Hepcidin did not mediate these associations.

Conclusion

It is unlikely that ferritin, TSAT and TIBC cause T2D although an association for serum iron could not be excluded. Glycaemic traits and liability to T2D may affect iron homeostasis, but mediation by hepcidin is unlikely. Corresponding mechanistic studies are warranted.

Iron, ferritin, type 2 diabetes, glycaemic traits, hepcidin, Mendelian randomization

Key Messages

• Most iron homeostasis biomarkers [ferritin, total iron-binding capacity (TIBC), and transferrin saturation] do not affect type 2 diabetes (T2D) risk and the association of serum iron in T2D is unclear.

• Liability to T2D and glycaemic traits affected iron homeostasis biomarkers such as increasing ferritin, TIBC and serum iron. Hepcidin did not appear to mediate these associations.

• Monitoring iron homeostasis biomarkers in people with type 2 diabetes may be useful to mitigate associated iron-related diseases.

Introduction

Understanding the effect of iron homeostasis on type 2 diabetes (T2D) is of global health importance because iron is an essential nutrient included in many supplements.¹ Observationally higher meat consumption, a key dietary source of iron, is positively associated with T2D risk,^2,3 with similar findings for iron homeostasis biomarkers such as ferritin and transferrin.⁴ However, these studies are vulnerable to confounding, making interpretation difficult. Furthermore, iron homeostasis biomarkers may have different aetiological roles in T2D, which have not often been explored.

Mendelian randomization (MR) is a potentially more credible study design to investigate the causal role of exposures in diseases than conventional observational studies, since it uses genetic predictors randomly allocated at conception, similarly to randomization in randomized controlled trials (RCTs), and hence is less susceptible to confounding.⁵ Furthermore, MR studies tend to be more consistent with RCTs than conventional observational studies.^6,7 Two previous MRs have assessed systemic iron status on T2D risk with inconsistent findings, possibly due to lack of study power in the earlier MR study.^8,9 However, both studies only used a few genetic instruments and included variants related to the predominant type of hereditary haemochromatosis (HH) (rs1800562 and rs1799945 in HFE) which has substantial pleiotropic effects, such as liver disease which increases T2D risk, so the findings could be the effects of HH rather than of iron.^10,11 These studies also did not evaluate the role of different iron homeostasis biomarkers in T2D or the possibility that hyperglycaemia affects iron homeostasis.^12,13

To provide a more comprehensive assessment of the role of iron homeostasis in T2D and hence inform dietary recommendations, we used genetic instruments from a much larger genome-wide association study (GWAS) of iron homeostasis biomarkers,¹⁴ applied to the largest suitable GWAS of T2D and glycaemic traits, taking HH into consideration. We included haemoglobin (Hgb) as a positive control outcome because iron increases Hgb synthesis.¹⁵ We also assessed the possibility that genetic liability to T2D or glycaemic traits affected iron homeostasis. Last, we investigated whether hepcidin, a key regulator of iron homeostasis encoded by hepcidin antimicrobial peptide (HAMP), mediated any associations found of liability to T2D or glycaemic traits in iron homeostasis.¹⁶

Methods

This study is reported as per Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guideline (Supplementary Table S1, available as Supplementary data at IJE online).¹⁷

Study design

This is a bidirectional, two-sample MR study using summary autosomal genetic associations to estimate the causal effects of four iron homeostasis biomarkers on T2D, glycaemic traits and Hgb, and vice versa. MR relies on three assumptions: relevance, independence and exclusion-restriction.¹⁸ Relevance implies the genetic instruments, i.e. single nucleotide polymorphisms (SNPs), predict the exposure. Independence implies the genetic instruments are not confounded and are independent of confounders of exposure on outcome. Exclusion-restriction implies the genetic instruments are independent of the outcomes given the exposure, i.e. no horizontal pleiotropy or selection bias (Figure 1a and b).

Figure 1.

Study design and directed acyclic graph of this bidirectional mendelian randomization (MR) study and two-step Mendelian randomization study. (a) Study design of this study including research questions, analytical plans, data sources and statistical analyses. (b) Directed acyclic graph illustrating these bidirectional MR study assumptions. (c) Directed acyclic graph illustrating these two-step MR study assumptions. The grey dashed line indicates the causal associations to be assessed. SNP, single nucleotide polymorphism; T2D, type 2 diabetes; FG, fasting glucose; 2hGlu, 2-h glucose; FI, fasting insulin; HbA1c, glycated haemoglobin; BMI, body mass index; Hgb, haemoglobin; TIBC, total iron-binding capacity; TSAT, transferrin saturation; SD, standard deviation; GWAS, genome-wide association study; IVW, inverse variance weighting; WM, weighted median; RAPS, robust adjusted profile score; MR-PRESSO, Mendelian Randomization Pleiotropy RESidual Sum and Outlier; r², linkage disequilibrium (LD); DIAMANTE, DIAbetes Meta-ANalysis of Trans-Ethnic association studies; FinnGen, FinnGen consortium; MAGIC, the Meta-Analyses of Glucose and Insulin-related traits Consortium; MR, Mendelian randomization; cis-pQLT, cis-protein quantitative trait; HAMP, hepcidin antimicrobial peptide, which encodes hepcidin. Iron homeostasis biomarker, ferritin, serum iron, TIBC or TSAT individually; systemic iron status, concurrent increase in ferritin, serum iron and TSAT, and decrease in TIBC

Data sources

Iron homeostasis biomarkers

The four iron homeostasis biomarkers considered were ferritin, serum iron, total iron-binding capacity (TIBC) and transferrin saturation (TSAT), which represent different domains of iron homeostasis, including stored iron (ferritin) and iron transport (serum iron, TIBC and TSAT) (Supplementary Table S2, available as Supplementary data at IJE online).¹⁹

Genetic instruments strongly (P <5 x 10^–8), and independently (linkage disequilibrium r²<0.001) predicting the four biomarkers [ferritin (n = 246 139), serum iron (n = 163 511), TIBC (n = 135 430) and TSAT (n = 131 471)] were extracted from summary statistics of a meta-analysis of GWAS of participants of European ancestry from the deCODE genetics project from Iceland, the INTERVAL study from the UK and the Danish Blood Donor Study (DBDS) from Denmark.¹⁴ Measurement of biomarkers varied by study, as described in Supplementary Table S3 (available as Supplementary data at IJE online). Genetic associations with ferritin, serum iron, TIBC and TSAT, in standard deviation (SD), were obtained from sex-specific, generalized, additive models adjusted for age in all studies, with additional adjustment for menopausal status, ABO blood group, body mass index (BMI), smoking, alcohol use and iron supplementation status in the INTERVAL study.¹⁴

To reduce bias from horizontal pleiotropy, we excluded genetic instruments from ABO (rs532436) and HFE (rs1800562 and rs1799945) which relate to blood group and HH, respectively, and may violate the exclusion-restriction assumption.^11,20 After exclusion of these variants, 61, 27, 32 and 26 SNPs for ferritin, serum iron, TIBC and TSAT, respectively, were considered (Supplementary Tables S3–S5, available as Supplementary data at IJE online).

T2D and glycaemic traits

The primary outcome was T2D, from the DIAbetes Meta-Analysis of Trans-Ethnic association studies (DIAMANTE) consortium in people of European descent (cases = 80 154, controls = 853 816).²¹ The GWAS was adjusted for age, sex and study-specific covariates. We also used summary statistics for T2D from FinnGen r7 (released 1 June 2022) (cases = 44 266, controls = 256 217), which is a combination of Finnish registry data and existing cohorts. The genetic associations were adjusted for age, sex, principal components and genotype batch (Supplementary Tables S3 and S6, available as Supplementary data at IJE online).²²

The secondary outcomes were glycaemic traits from the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC) in participants of European descent.²³ These included fasting glucose (mmol/L) (FG, n = 209 605), 2-h glucose (mmol/L) (2hGlu, n = 64 469), fasting insulin (log pmol/L) (FI, n = 158 550) and glycated haemoglobin (%) (HbA1c, n = 149 006).²³ The GWAS excluded participants with type 1 diabetes, T2D, diabetes-related medications use records, FG >7 mmol/L, 2hGlu ≥11.1 mmol/L or HbA1c ≥6.5%. Genetic associations were adjusted for age, sex, study-specific covariates, BMI (except for HbA1c) and principal components.²³ Adjustment for BMI had little collider bias effect, as reported in the original GWAS (Supplementary Tables S3 and S6).²³

Hgb (positive control outcome)

For the positive control outcome of Hgb (g/dL) (n = 563 085) the GWAS from the Blood Cell consortium excluded participants with diseases or treatments which affect Hgb, such as blood cancer, chemotherapy, congenital or hereditary anaemia and those with extreme measurements of blood cell traits such as Hgb >20 g/dL and haematocrit >60%.²⁴ Detailed exclusion criteria are described in the original publication. Genetic associations were specific to people of European descent and adjusted for age, age squared, sex, principal components and cohort-specific covariates, such as study centre (Supplementary Table S3).²⁴

Statistical analyses

We aligned genetic instruments for exposures and outcomes on effect allele and effect allele frequency (EAF).²⁵ Palindromic genetic instruments were discarded if EAF could not be unequivocally determined (42% < EAF < 58%). For each exposure, we estimated the variance (R²) explained by the genetic instruments, and the overall F statistic.²⁶ A larger F statistic indicates that weak instrument bias is unlikely.⁵ We used inverse variance weighting (IVW) with multiplicative random effects as the main analysis. IVW estimates for T2D from the DIAMANTE consortium and FinnGen were meta-analysed using fixed effects. IVW assumes balanced pleiotropy which may be violated by invalid instruments.²⁷ As such, we also assessed heterogeneity of the Wald ratios (i.e. genetic variants on outcomes divided by genetic variants on exposures) using I², where high I² implies the possibility of invalid instruments (Figure 1a).²⁸

Sensitivity analyses

We conducted sensitivity analyses with different assumptions; consistency across estimates strengthens evidence of causation. Sensitivity analyses included the weighted median (assumes majority valid), robust adjusted profile score (assumes pleiotropic effects normally distributed about zero), MR-PRESSO (outlier-robust) and MR-Egger (assumes instrument strength independent of direct effects).²⁹ A significant MR-Egger intercept (P <0.05) indicates possible violation of the exclusion-restriction assumption via overall horizontal pleiotropy. We also approximated the magnitude of regression dilution via the ${\mathrm{I}}_{\text{GX}}^2$ statistic, where a value close to 100% indicates that dilution bias affecting the MR-Egger estimates is unlikely (Figure 1a).³⁰

To further reduce potential pleiotropic effects of the instruments, we searched for pleiotropic effects of the genetic instruments using PhenoScanner, a curated genotype to phenotype cross-reference.^31,32 We excluded genetic variants likely directly associated with phenotypes related to glycaemic traits and adiposity traits (BMI and arm/leg/trunk/body fat mass or fat-free mass) in sensitivity analyses (Figure 1a, Supplementary Tables S4 and S7, available as Supplementary data at IJE online).

Exploratory analyses concerning the influence of HH variants

To explore whether findings from previous MR studies could be due to the inclusion of genetic instruments for HH, we additionally conducted the following exploratory analyses (Figure 1a).

• i)  Only included rs1800562 and/or rs1799945 as genetic instruments to identify the potential pleotropic effects of predominantly HH-related instruments (Supplementary Tables S4 and S5);

• ii) Only included genetic instruments with concordant associations across biomarkers (rs1800562, rs855791 and rs57659670) which predict increased ferritin, serum iron and TSAT but decreased TIBC or vice versa, as previously (Supplementary Tables S4 and S5).^9,33

Assessment of liability to T2D and glycaemic traits with iron homeostasis biomarkers

We also assessed the association of liability to T2D and of glycaemic traits with iron homeostasis biomarkers, using instruments from DIAMANTE (T2D, cases = 80 154, controls = 853 816), and MAGIC (FG (mmol/L, n = 209 605), 2hGlu (mmol/L, n = 64 469), FI (log pmol/L, n = 158 550) and HbA1c (%, n = 149 006)) applied to GWAS summary statistics of iron homeostasis biomarkers (ferritin (n = 246 139), serum iron (n = 163 511), TIBC (n = 135 430) and TSAT (n = 131 471)), with exclusion of ABO and HFE (rs1800562 and rs1799945) variants where appropriate (Figure 1a, Supplementary Tables S8–S10, available as Supplementary data at IJE online).^14,21,23

Mediation analysis using two-step Mendelian randomization

We used two-step MR to assess whether hepcidin mediated the observed associations of liability to T2D or glycaemic traits in iron homeostasis biomarkers, as implicated in previous studies.^16,34 In Step 1, we assessed the causal effects of the exposures, liability to T2D or glycaemic traits on the mediator (hepcidin, SD) using summary statistics from a proteins GWAS (n = 35 559),³⁵ excluding ABO and HFE (rs1800562 and rs1799945) variants. Where any of these exposures affected hepcidin, we assessed the causal effects of hepcidin on the outcomes (iron homeostasis biomarkers) using the cis-protein quantitative trait loci (pQTL, rs185716928 in HAMP) as an instrument to reduce the likelihood of horizontal pleiotropy (Figure 1a and c).^35,36

Power calculation

Based on the significance level of 0.05 and the smallest R² of the four iron homeostasis biomarkers explained by the genetic instruments in the main analysis (i.e. R² of ferritin: 1.9%), our study had 80% power to detect an odds ratio of ≤0.941 or ≥1.063 for associations with T2D, and an effect size of ≤-0.045 or ≥0.045 for associations with glycaemic traits (Supplementary Table S11, available as Supplementary data at IJE online).^37,38

All analyses were performed using R version 4.0.2 with the R packages (‘TwoSampleMR’ version 0.5.6,³⁹ ‘meta’ version 6.2–1, ‘MRPRESSO’ version 1.0,⁴⁰ and ‘forestplot’ version 3.1.1).

Results

Genetic instruments for iron homeostasis biomarkers

After excluding genetic instruments not available for the outcomes, up to 59 SNPs for ferritin, 26 SNPs for serum iron, 31 SNPs for TIBC and 26 SNPs for TSAT were used in the main analysis. The overall F statistics were 80 for ferritin (R²: 1.9%), 166 for serum iron (R²: 2.6%), 164 for TIBC (R²: 3.5%) and 192 for TSAT (R²: 3.5%), implying that weak instrument bias is unlikely (Supplementary Tables S4 and S5).

The association of iron homeostasis biomarkers in type 2 diabetes

Most iron homeostasis biomarkers were unrelated to T2D risk in DIAMANTE and FinnGen and in meta-analyses of the IVW estimates (Figure 2a). However, serum iron was potentially associated with higher T2D risk in DIAMANTE, with stronger evidence from sensitivity analyses, but not in FinnGen. Supplementary Figure S1 (available as Supplementary data at IJE online) shows the scatter plots of genetic variants' association for iron homeostasis biomarkers against association for T2D from DIAMANTE and FinnGen. Heterogeneity was greater in DIAMANTE than in FinnGen, although evidence for horizontal pleiotropy was only found for serum iron and TSAT with T2D in DIAMANTE. These associations were generally consistent across different sensitivity analyses. Exclusion of additional pleiotropic variants reduced heterogeneity (Supplementary Table S7) but gave similar findings as the main analyses with wider confidence intervals (Supplementary Figure S3a, available as Supplementary data at IJE online).

Figure 2.

The associations of genetically predicted iron homeostasis biomarkers in type 2 diabetes and glycaemic traits with non-ABO and non-HFE (rs1800562/rs1799945) variants using Mendelian randomization and meta-analysis. (a) The association of iron biomarkers in type 2 diabetes. (b) The association of iron biomarkers in glycaemic traits. Sample size of DIAMANTE [Mahajan et al. (2022)) is 933 970 (80 154 cases and 853 816 controls), FinnGen r7 (release on 1 June 2022) is 300 483 (44 266 cases and 256 217 controls)]. No. of SNPs, number of single nucleotide polymorphisms; I², degree of heterogeneity; OR, odds ratio; CI, confidence interval; IVW, inverse-variance weighting; WM, weighted median; RAPS, robust adjusted profile score; MR-PRESSO, Mendelian Randomization Pleiotropy RESidual Sum and Outlier; DIAMANTE, DIAbetes Meta-Analysis of Trans-Ethnic association studies; FinnGen, FinnGen consortium; FG, fasting glucose; FI, fasting insulin; 2hGlu, 2-h glucose; HbA1c, glycated haemoglobin. *P-value <0.05; **P-value <0.01, ***P-value <0.001

Figure 2.

(Continued)

The association of iron homeostasis biomarkers in glycaemic traits

Ferritin, serum iron and TSAT were inversely associated with HbA1c whereas TIBC was positively associated with HbA1c, with directionally consistent associations from sensitivity analyses (Figure 2b). No association of iron homeostasis biomarkers in other glycaemic traits were found. Supplementary Figure S2 (available as Supplementary data at IJE online) shows the scatter plots of genetic variants' association for iron homeostasis biomarkers against association for glycaemic traits. Heterogeneity was high although evidence for overall horizontal pleiotropy for most analyses was limited. Results were consistent after exclusion of potentially pleiotropic variants (Supplementary Figure S3b).

Analyses restricted to HFE variants or variants with concordant effects on iron homeostasis biomarkers

Using only HH genetic variants from HFE (rs1800562 and/or rs1799945), ferritin, serum iron and TSAT were positively associated with T2D whereas TIBC was inversely associated with T2D from DIAMANTE and in the meta-analyses but not in FinnGen (Supplementary Figure S4a, available as Supplementary data at IJE online). Directionally consistent associations were found using variants with concordant effects on iron homeostasis biomarkers (which included the HFE variant, rs1800562) in DIAMANTE but not in FinnGen (Supplementary Figure S5a, available as Supplementary data at IJE online). Meta-analyses showed ferritin associated with higher T2D and TIBC associated with lower T2D. These exploratory analyses gave associations for iron homeostasis biomarkers with HbA1c consistent with the main analyses (Supplementary Figures S4b and S5b).

The association of iron homeostasis biomarkers with the control outcome of haemoglobin

Ferritin, serum iron and TSAT were positively associated with Hgb, and TIBC was inversely associated with Hgb, with directionally consistent estimates in sensitivity analyses and exploratory analyses using HFE (rs1800562 and/or rs1799945) variants and variants representing systemic iron status (Supplementary Table S12, available as Supplementary data at IJE online).

The association of liability to type 2 diabetes and glycaemic traits in iron homeostasis biomarkers

In the analyses of liability to T2D and glycaemic traits on iron homeostasis biomarkers, there were up to 178 SNPs for liability to T2D and 66, 37, 11 and 70 SNPs for FG (R²: 4.6%), FI (R²: 1.3%), 2hGlu (R²: 1.1%) and HbA1c (R²: 5.3%) (Supplementary Tables S8–S10). Liability to T2D was positively associated with TIBC, FG was associated with higher serum iron, and FI was associated with higher ferritin (Figure 3). HbA1c was associated with lower ferritin, iron and TSAT and possibly with higher TIBC. These associations were directionally consistent in sensitivity analyses. There was evidence of heterogeneity, although evidence of horizontal pleiotropy was not strong for most analyses (Figure 3).

Figure 3.

The associations of liability to type 2 diabetes and glycaemic traits in iron homeostasis biomarkers with non-ABO and non-HFE (rs1800562/rs1799945) variants using Mendelian randomization. (a) The association of type 2 diabetes from DIAMANTE in iron homeostasis biomarkers. (b) The association of glycaemic traits in iron homeostasis biomarkers. No. of SNPs, number of single nucleotide polymorphisms; I², degree of heterogeneity; CI, confidence interval; TIBC, total iron-binding capacity; TSAT, transferrin saturation; IVW, inverse variance weighting; WM, weighted median; RAPS, robust adjusted profile score; MR-PRESSO, Mendelian Randomization Pleiotropy RESidual Sum and Outlier; SD, standard deviation; FG, fasting glucose; FI, fasting insulin; 2hGlu, 2-h glucose; HbA1c, glycated haemoglobin. *P-value <0.05; **P-value <0.01; ***P-value <0.001

Mediation analysis using two–step MR

Liability to T2D, FG and FI were not associated with hepcidin (Figure 4). Hepcidin had the expected inverse association with serum iron but not with TIBC or ferritin (Supplementary Figure S6, Supplementary Tables S13 and S14, available as Supplementary data at IJE online).

Figure 4.

The associations of liability to type 2 diabetes, fasting glucose and fasting insulin in hepcidin. No. of SNPs, number of single nucleotide polymorphisms; I², degree of heterogeneity; CI, confidence interval; T2D, type 2 diabetes; FG, fasting glucose; FI, fasting insulin; SD, standard deviation; IVW, inverse variance weighting; WM, weighted median. *P-value <0.05; **P-value <0.01; ***P-value <0.001

Discussion

To the best of our knowledge, this is one of the largest bidirectional MR studies concerning iron homeostasis biomarkers and T2D/glycaemic traits, as well as taking into account potential pleiotropy from HH and the mediating role of hepcidin which may have biased the previous MR study.⁹ In meta-analyses of IVW estimates, we found that it is unlikely that iron homeostasis biomarkers affect T2D, although a positive association of serum iron with T2D in DIAMANTE is possible. These findings are inconsistent with the previous MR study which only considered systemic iron status as an overall exposure, making direct comparisons difficult.⁹ More importantly, our study adds by showing that glycaemic traits and liability to T2D influence iron homeostasis, as previously suggested,¹² although mediation by hepcidin is unlikely.

Previous observational studies of iron homeostasis biomarkers and T2D have shown circulating ferritin in the normal reference range, dietary total iron, non-haeme iron or iron supplementation not associated with T2D,^3,41 consistent with our findings. Observational studies have also suggested red meat consumption increases T2D,³ possibly due to red meat containing not only iron but also other dietary factors likely causally related to T2D, such as branched chain amino acids.⁴² We observed a possible positive association of serum iron with T2D in DIAMANTE, but not in FinnGen (Figure 2a). Reasons accounting for such differences are unclear, but could be due to differences in selection criteria or case definition, or the relevance of iron instruments in Europeans in general (DIAMANTE) compared with Finnish only (FinnGen) given possibly different genetic architecture.⁴³ Alternatively, pleiotropic instruments [rs144861591 (strongly relates to rs1800562 in HFE, r²=0.98), rs12633819 (LRIG1, affects adipocyte morphology)⁴⁴ and rs10822143 (associated with sex hormone binding globin in UK Biobank, P-value = 7.78 x 10^–301)] may drive these discrepancies.⁴⁵ However, exclusion of these pleiotropic instruments did not change the conclusion (Supplementary Figure S3a). The underlying mechanisms linking serum iron with T2D may be via insulin resistance,¹² which needs to be explored in future investigations.

Systematic reviews of observational studies have shown iron deficiency anemia increases HbA1c,⁴⁶ which is consistent with our findings for iron homeostasis biomarkers on HbA1c (Figure 2b). The lack of effect of iron homeostasis biomarkers on FG and T2D (Figure 2b) suggests any effect on HbA1c is via erythrocytic properties rather than glucose.^47,48 This also indirectly highlights the importance of considering iron levels when using HbA1c for T2D diagnosis or management, as well as considering the affect of sex in the relation of iron with HbA1c as suggested previously.⁴⁹

Here, we found liability to T2D positively associated with TIBC, FI positively associated with higher ferritin and FG positively associated with serum iron (Figure 3). Previous studies suggest glucose triggers hepcidin release,¹⁶ a key iron homeostasis regulator, which would reduce iron availability. However, in our mediation analyses, liability to T2D, FI and FG were unrelated to hepcidin (Figure 4;Supplementary Figure S6, available as Supplementary data at IJE online). Alternatively, glucose metabolism may affect iron homeostasis through the gut microbiota.¹³ For example, previous diabetic rat studies showed reduction in insulin caused hepcidin reduction and led to increased expression of intestinal ferroportin and iron absorption,⁵⁰ whilst other in vitro studies suggested high glucose reduced hepcidin expression and secretion.⁵¹ Future studies taking into account of inflammatory factors and ferroportin would be useful in exploring the association of glycaemic traits and type 2 diabetes in iron.⁵²

Despite using a design which is less susceptible to confounding than conventional observational studies, limitations exist. First, MR relies on the three assumptions of relevance, independence and exclusion-restriction.⁵ Although we used genetic instruments strongly and independently associated with iron homeostasis biomarkers and glycaemic traits, MR estimates may be biased by inclusion of invalid instruments leading to horizontal pleiotropy. However, we used different statistical methods with different assumptions, as well as exclusion of potentially pleiotropic instruments, which gave similar findings. Our exploratory analyses using only concordant SNPs to proxy systemic iron status, as previously, and restricting to HFE variants, also suggest previous MR studies could be biased by including HFE variants.⁹ Whether findings in previous MR studies are driven by horizontal pleiotropy or by substantial iron overload due to HH requires additional investigation. Second, we did not explore sex-specific effects of iron due to the lack of sex-specific GWAS summary statistics. However, this would be worth exploring in future studies, given men have higher iron than women,⁵³ and testosterone increases iron but protects against T2D.^54,55 Third, MR studies are also susceptible to selection bias due to inclusion of only survivors in many GWAS.^18,56 However, it is unlikely to have such bias for T2D.¹⁸ Fourth, some summary statistics were adjusted for heritable covariates, for example genetic associations with iron homeostasis biomarkers were adjusted for BMI in one GWAS (INTERVAL),¹⁴ which could induce bias.⁵⁷ Future MR studies using GWAS without adjustment for heritable covariates would help verify our findings.

No strong evidence that iron homeostasis biomarkers influence T2D risk was found. However, the positive association of glycaemic traits and liability to T2D in iron homeostasis biomarkers, such as ferritin, indicates that T2D may increase risk of iron-related diseases, such as liver disease or ischaemic stroke.^58,59 As such, controlling these iron homeostasis biomarkers in T2D may mitigate the risk of iron-related diseases.

Conclusion

In conclusion, our study suggests ferritin, TSAT and TIBC are not associated with T2D although a possible positive association of serum iron in T2D needs to be confirmed. Our study additionally shows a possible effect of liability to T2D on TIBC, FG on serum iron and FI on ferritin, unlikely to be mediated by hepcidin. Additional mechanistic studies, such as the mediating roles of inflammatory factors or ferroportin, would be helpful to delineate the association of glucose in iron homeostasis biomarkers.

Ethics approval

This study only used publicly available GWAS summary statistics and hence no ethics approval was required. Respective ethics approvals have been obtained by the relevant GWAS investigators.

Data availability

All data used in this study can be found in the cited references, and URLs in the Acknowledgements and Supplementary Material (available as Supplementary data at IJE online).

Supplementary data

Supplementary data are available at IJE online.

Author contributions

S.L.A.Y and Y.L. designed the study. Y.L. wrote the analysis plan, performed the data analyses and interpreted the results, with feedback from S.L.A.Y. T.H.T.W. cross-checked the results. Y.L. wrote the first draft of the manuscript, with critical feedback and revisions from S.L.A.Y., S.L., T.H.T.W., B.H. and C.M.S. All authors gave final approval of the version to be published. Y.L. is the guarantor of work.

Funding

This study was partly funded by the Health and Medical Research Fund, Food and Health Bureau, HKSAR Government, Hong Kong, China (CFS-HKU1). The funder had no role in the design, analyses, interpretation of results or writing of the paper.

Acknowledgements

Summary data on iron homeostasis have been contributed by S. Bell et al. (2021) and have been downloaded from [https://www.decode.com/summarydata/]. Summary data on type 2 diabetes have been contributed by DIAMANTE investigators and have been downloaded from [http://diagram-consortium.org]. Summary data on fasting glucose, 2-h glucose, fasting insulin and glycated haemoglobin have been contributed by MAGIC investigators and have been downloaded from [http://www.magicinvestigators.org/]. Summary data on haemoglobin have been contributed by D. Vuckovic et al. (2020) and have been fetched from GWAS Catalog [https://www.ebi.ac.uk/gwas/publications/32888494]. We acknowledge the participants and investigators of the FinnGen study, where summary data on type 2 diabetes from FinnGen have been downloaded from [https://www.finngen.fi/en]. Summary data on hepcidin have been contributed by Ferkingstad et al. (2021) and have been obtained from [https://www.decode.com/summarydata/]. We also thank Mr Jacky Man Yuen Mo for cross-checking the results related to hepcidin.

Conflict of interest

None declared.