Causal Risk Factors for Type 1 Diabetes in Mendelian Randomization Studies: A Systematic Review and Meta-Analysis

Abstract

Background

Type 1 diabetes mellitus (T1DM) is a chronic disease mediated by autoimmunity, with complex and not fully elucidated pathogenesis. Mendelian randomization (MR) utilizes genetic instrumental variables to minimize confounding and reverse causation; however, individual MR studies are often limited by sample size and result heterogeneity.

Methods

Following PRISMA 2020 guidelines, we systematically searched PubMed, Web of Science, and other databases from 2014 to 2025, ultimately including 53 MR studies (covering 243 exposures). Random-effects models were used to pool effect sizes. Heterogeneity was quantified by Cochran’s Q test and I² statistic. Bias was further controlled using Egger’s regression and leave-one-out sensitivity analysis.

Results

This study integrated 53 MR studies (243 exposures) and identified key causal factors for T1DM.IL2RA (OR = 0.22, 95% CI: 0.17–0.27) and TYK2 (OR = 0.61, 95% CI: 0.54–0.69) showed significant protective effects, while IL6R (OR = 1.98, 95% CI: 1.48–2.65) was associated with increased risk. For metabolites, 3-phenylpropionic acid (OR = 0.90, 95% CI: 0.85–0.96) and cinnamoylglycine (OR = 0.89, 95% CI: 0.84–0.96) were protective, while trimethylamine N-oxide (TMAO; OR = 1.11, 95% CI: 1.02–1.20) increased risk. Among gut microbiota, Prevotella 9 (OR = 1.18, 95% CI: 1.08–1.30) was positively associated with risk, whereas Bifidobacterium (OR = 0.82, 95% CI: 0.71–0.95) showed a protective effect. Childhood obesity (OR = 1.32, 95% CI: 1.06–1.64) was also associated with increased T1DM risk. Overall heterogeneity was high (I² = 78.3%).

Conclusion

This study systematically mapped the multi-omics causal risk landscape of T1DM, providing important evidence for precision prevention and targeted intervention. These findings suggest that targeting immune pathways (particularly IL2RA and TYK2) and modulating gut microbiota composition may represent promising strategies for T1DM prevention. Future research should emphasize cross-ethnic validation and life-stage–specific intervention strategies.

Introduction

Background

Type 1 diabetes mellitus (T1DM) is a lifelong chronic disease resulting from a complex pathogenesis involving autoimmune destruction of pancreatic β-cells and a wide range of genetic, environmental, immunological, and metabolic factors.Although it typically manifests in childhood and adolescence, its pathogenesis remains complex and incompletely understood, involving a wide range of genetic, environmental, immunological, and metabolic factors.¹ The incidence of T1DM varies significantly across different ethnicities and regions, and has been rising globally in recent years, placing a considerable burden on patients, families, and public health systems.^2,3 Elucidating the causal risk factors for T1DM is crucial for developing effective prevention strategies, advancing precision medicine, and identifying novel therapeutic targets.

Rationale and Knowledge Gap

While numerous observational studies have identified a variety of potential risk factors for T1DM—including genetic susceptibility loci, environmental exposures, metabolic products, inflammatory responses, and alterations in the microbiome—such studies are prone to confounding and reverse causation, limiting their ability to infer causality. Mendelian randomization (MR), as an innovative epidemiological approach, leverages genetic variants as instrumental variables to minimize confounding and strengthen causal inference between risk factors and T1DM.⁴ In recent years, MR has been widely applied to explore the mechanisms and risk profiles of T1DM, revealing potential causal links across multiple biological pathways, including immune regulation,⁵ lipid metabolism,⁶ gut microbiota composition,⁷ and inflammatory cytokine signaling.⁸ However, existing MR studies often differ substantially in sample size, population structure, methodological design, and exposure definitions, limiting the reproducibility and generalizability of individual findings. For example, studies examining the causal role of vitamin D in T1DM have reported conflicting results, with sample sizes ranging from a few thousand to over 100,000 participants, and varying definitions of vitamin D exposure (eg, serum 25-hydroxyvitamin D levels vs genetic scores)^9,10 (Similarly, MR studies on gut microbiota have employed different taxonomic classifications and statistical approaches, yielding inconsistent effect estimates.^11,12 Systematic reviews and meta-analyses can integrate diverse MR evidence, enhance statistical power, quantify the comprehensive causal effects of multiple exposures, and enable in-depth assessment of heterogeneity and robustness across studies.^13,14 Particularly in the era of multi-omics and big data, standardizing and structuring results from different MR studies and systematically evaluating the true impact of various factors on T1DM have become major scientific challenges.¹⁵

Objective

In this study, we performed a large-scale integration of MR evidence to comprehensively characterize the multidimensional causal architecture of T1DM. We established a complete risk factor spectrum based on MR evidence across eleven biological categories—including amino acids, metabolites, immune cells, proteins/genes, lipids, and disease/inflammatory factors. Using rigorous random-effects modeling, sensitivity analyses, and publication bias assessments, we systematically quantified the causal contributions of different exposures to T1DM risk, elucidated patterns of heterogeneity and subgroup features, and provided a robust scientific basis for etiological research, risk prediction, targeted intervention, and precision prevention of T1DM. Our work also contributes methodological insights for the standardization and innovation of MR meta-analyses.

This systematic review and meta-analysis were conducted according to the PRISMA 2020 guidelines.

Methods

Search Strategy

This systematic review was conducted in accordance with the PRISMA 2020 guidelines.^16,17 A comprehensive literature search was performed using PubMed, Web of Science, and Embase for articles published between January 1, 2014, and July 1, 2025. A comprehensive literature search was performed using PubMed, Web of Science, and Embase for articles published between January 1, 2014, and July 1, 2025. The year 2014 was selected as the starting point because it marked a significant period of methodological advancement in Mendelian randomization research, including the widespread adoption of two-sample MR approaches and the increased availability of large-scale GWAS summary data from major genetic consortia and biobanks. Additionally, our preliminary scoping search indicated that MR studies specifically investigating causal risk factors for T1DM were sparse prior to 2014.The search strategy incorporated a combination of free-text keywords and Medical Subject Headings (MeSH) terms related to autoimmune diseases, type 1 diabetes mellitus (T1DM), Mendelian randomization (MR), and genetic causal inference. Key terms included: “autoimmune disease”, “Type 1 diabetes mellitus”, “T1DM”, “Mendelian randomization”, “MR”, “genetic instrument”, “causal risk factor”, “exposure”, “risk factor”, “determinant”, and “GWAS”. Boolean operators (AND/OR) were used to optimize search sensitivity and specificity. An exemplary search string was: (“Type 1 diabetes” OR “T1DM”) AND (“Mendelian randomization” OR “MR”) AND (“risk factor” OR “exposure” OR “instrument variable” OR “GWAS”). To ensure literature saturation, reference lists of all included articles were manually screened for additional relevant publications (see Figure 1).

Figure 1.

PRISMA Flow Diagram of the Study Selection Process. A total of 759 records were identified through systematic searches of PubMed, Embase, and Web of Science. After removal of duplicates and screening of titles, abstracts, and full texts, 53 studies were included in the meta-analysis.

Eligibility Criteria

The inclusion criteria were as follows: (1) MR studies with T1DM as the outcome; (2) studies reporting exposure factors, effect estimates (odds ratio, OR), 95% confidence intervals (CIs), and corresponding P-values; (3) studies conducted in human populations; (4) employment of a two-sample or multi-sample MR design.

Studies were excluded for any of the following: (1) publication as an abstract, review, or conference paper without sufficient original data; (2) duplicate or substantially overlapping datasets; (3) use of instrumental variables with an F-statistic below 10, which is the established threshold for instrument strength to minimize weak instrument bias. To ensure reliable and consistent study selection, two reviewers independently performed the screening process. Any discrepancies were resolved through consensus or by adjudication from a third reviewer.

Data Extraction and Management

For each eligible study and publicly available dataset, the following data were extracted: first author, publication year, specific exposure factor, biological category, cohort name, sample size, effect estimate (OR), 95% CI, P-value, and the number of instrumental variables used. Data extraction was performed using a standardized form. All numerical data, including scientific notation and superscript formats, were carefully standardized using Stata software to ensure consistency. Only entries with logically consistent effect estimates (ie, CI lower limit < OR < CI upper limit) were included in subsequent analyses.

Data Synthesis and Statistical Analysis

Meta-Analysis

All meta-analyses were performed using random-effects models (DerSimonian–Laird method) as the primary approach to account for anticipated heterogeneity.¹⁸ Pooled effect estimates and 95% CIs were derived using the natural logarithm of the OR and its standard error. For dichotomous exposures, the Mantel-Haenszel method was applied; for continuous exposures, the inverse-variance method (yielding standardized mean differences, SMDs) was used.¹⁹ If an exposure was reported in only a single study, the original estimate and CI were presented narratively.

Assessment of Heterogeneity

Heterogeneity among studies was assessed using Cochran’s Q statistic (with a significance threshold of P < 0.10) and quantified using the I² statistic. I² values greater than 50% were considered to indicate moderate-to-high heterogeneity. In cases of significant heterogeneity, the random-effects model was retained, and the between-study variance (τ²) was estimated.

Sensitivity Analysis and Publication Bias

The robustness of the pooled results was evaluated using leave-one-out sensitivity analysis. Publication bias was assessed visually through funnel plot symmetry (Supplemental Figure 1) and statistically using Egger’s regression test (Supplemental Figure 3, Supplemental Table 2). The robustness of the results was further evaluated using Begg’s rank correlation test (Supplemental Table 5). The sources of heterogeneity were further explored using multivariable meta-regression models that incorporated covariates such as study quality score and sample size.²⁰ Begg’s rank correlation test was used as a supplementary assessment for publication bias. Influence diagnostics were conducted to identify any individual study exerting a disproportionate impact on the overall results. As specified in the eligibility criteria (Eligibility Criteria), studies employing genetic variants with an F-statistic below 10 were excluded to ensure adequate instrument strength. The significance level for all primary analyses was set at P < 0.05.

Visualization and Reporting

The results of the meta-analysis were presented graphically using forest plots (Figure 2, Supplemental Figure 4), funnel plots (Supplemental Figure 1), and sensitivity analysis plots (Supplemental Figure 3). Additionally, a bubble plot illustrating univariate meta-regression results is provided in Supplemental Figure 2. For each exposure category, a structured summary table reported the number of studies, pooled OR, 95% CI, I² statistic, and P-value from the Q-test.

Figure 2.

Forest Plot of Factors Investigated in Two or More Studies. Each estimate is represented by a square (point estimate) and horizontal line (95% confidence interval). The area of the square corresponds to the study-specific weight in the meta-analysis. The vertical dashed line indicates the null effect (OR = 1).

Abbreviations: CTRB1, chymotrypsinogen B1; DKK3, Dickkopf-related protein 3; GCA, giant cell arteritis; GGT, gamma-glutamyl transferase; IL2RA, interleukin-2 receptor subunit alpha; IL6R, interleukin-6 receptor; TYK2, tyrosine kinase 2; SIRPG, signal regulatory protein gamma; TMAO, trimethylamine N-oxide; FABP4, fatty acid-binding protein 4; IGF-1, insulin-like growth factor 1; HLA, human leukocyte antigen; PC, phosphatidylcholine; LDL, low-density lipoprotein; HDL, high-density lipoprotein; BMI, body mass index; T1DM, type 1 diabetes mellitus; OR, odds ratio; CI, confidence interval.

Software

All statistical analyses, data management and generation of figures were conducted using STATA version 17.0 (StataCorp LP, USA).Literature screening and data extraction figures were performed using RevMan version 5.4 (The Cochrane Collaboration, Denmark).

Assessment of Risk of Bias and Study Quality

Systematic Search and Screening Rigor

To minimize language bias, no restrictions were applied during the search. Furthermore, manual searches of reference lists and key reviews were conducted to identify any additional relevant studies.

Control for Publication Bias

In addition to statistical tests, efforts to mitigate publication bias included the consideration of high-quality preprints that employed robust methodological approaches. Sensitivity analyses were performed by excluding lower-quality studies (defined as those with a quality score ≤ 6) to evaluate the stability of the findings.

Standardized Screening and Data Extraction Process

The study screening process was designed to minimize selection bias. Two independent reviewers (both holding MSc degrees) performed the initial screening of titles and abstracts, followed by a full-text review. A third reviewer (an MD) was consulted to resolve any disagreements, ensuring a consistent and rigorous application of the inclusion and exclusion criteria.

A piloted, structured data extraction form was used to ensure consistent and accurate data collection. To control for extraction errors, a quality control procedure was implemented: personnel not involved in the primary extraction process independently reviewed a random sample of 20% of the included studies. The error rate for data extraction was maintained below 5%. These procedures were aligned with the Cochrane risk of bias tool²¹ and PRISMA 2020 recommendations to address potential biases comprehensively.

Quality Assessment of Individual Studies

The methodological quality of each included MR study was evaluated using a predefined set of 13 criteria, developed based on established guidelines for MR research.^22–24 This tool assessed five core domains: (1) methodological rigor of the MR design (eg, use of one-sample vs two-sample MR, primary statistical approach (eg, IVW, 2SLS), and thoroughness of sensitivity analyses for pleiotropy and heterogeneity); (2) sample size and statistical power (studies with >80% power were considered large); (3) selection and justification of instrumental variables (including genetic strength and biological plausibility); (4) application of advanced MR methods (eg, multivariable MR); and (5) completeness of reporting (eg, effect estimates, CIs, discussion of limitations). Each study received a score from 0 to 13.

Additionally, the overall quality of evidence for key findings was assessed using the GRADE framework.^25,26 Initially, all evidence from MR studies was classified as “high quality” and was subsequently downgraded based on the following criteria: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Two reviewers independently performed all quality assessments. Inter-rater agreement was quantified using the Kappa statistic, and any discrepancies were resolved through discussion or by consulting a third reviewer.This systematic review and meta-analysis was prospectively registered in PROSPERO (International Prospective Register of Systematic Reviews) under registration number CRD420251152014 (https://www.crd.york.ac.uk/PROSPERO/view/CRD420251152014).

Results

Characteristics of Included Studies

Our systematic search and selection process, conducted in accordance with PRISMA 2020 guidelines, yielded 53 eligible Mendelian randomization studies^{5–7,10,12,27–74} for inclusion. These studies collectively encompassed 132 independent cohorts and investigated 243 unique exposures. Participant data were predominantly sourced from large-scale public genetic consortia and biobanks, including the UK Biobank and FinnGen, with the majority of individuals being of Eurasian ancestry. All exposures were systematically classified into 11 biological categories for analysis, including amino acids, proteins/genes, lipids/fatty acids, metabolites, immune cells/phenotypes, trace elements, hormones/vitamins, diseases/inflammation/autoimmunity, gut microbiota, and enzymes/signaling molecules/receptor complexes (see Table 1 and Supplemental Table 1 for complete details).

Table 1.

Meta-Analysis Results by Primary Exposure Category

Category	Pooled_OR	95 CI_Low	95 CI_High	N	I2	Q_pval
Metabolites	1.07	0.88	1.29	12	83.05	<0.001
Immune cells/immune phenotypes	0.99	0.93	1.05	31	89.39	<0.001
Other/To be determined	0.97	0.76	1.22	7	76.91	<0.001
Trace elements/electrolytes	0.62	0.57	0.69	2	0.00	0.47
Amino acids	0.91	0.89	0.93	16	27.54	0.15
Hormones/vitamins	0.99	0.91	1.07	23	56.82	<0.001
Disease / inflammation/autoimmunity	1.14	1.04	1.25	68	66.95	<0.001
Gut microbiota/microorganisms	1.00	0.97	1.03	67	81.28	<0.001
Lipids/fatty acids	0.95	0.89	1.02	53	70.67	<0.001
Proteins/genes	1.12	0.91	1.23	83	85.61	<0.001
Enzymes/signaling molecules/receptor complexes	1.18	0.92	1.50	10	79.37	<0.001

Notes: For each category, the table presents the pooled odds ratio (OR), 95% confidence interval (CI), p-value, I² statistic for heterogeneity, and the number of contributing studies. Detailed results for all exposures, including sample sizes, methodological notes, and subgroup analyses, are provided in Supplemental Table 1.

Based on our predefined quality assessment criteria, the included studies were stratified into three quality tiers. Eighteen studies were classified as high-quality (score 10–13; e.g.,49,71), characterized by large sample sizes, rigorous MR methodology, and comprehensive sensitivity analyses. Twenty-seven studies were rated as moderate-quality (score 7–9; e.g.,65), often limited by moderately sized samples or partial methodological shortcomings. Eight studies were categorized as low-quality (score ≤6), primarily constrained by small sample sizes or insufficient control of potential biases.Reporting quality of the 53 included studies was assessed using the STROBE-MR checklist (Supplemental Figure 5). Most items showed high compliance: title/abstract, background, objectives, main results, key results, limitations, interpretation, and funding were fully reported in 100% of studies. However, sensitivity analyses (6% full, 91% partial) and software/pre-registration (11% full, 87% partial) showed notable reporting gaps.

Distribution of Overall and Subgroup Summary Estimates

The pooled overall odds ratio (OR) across all included exposures was 0.99 (95% CI: 0.97–1.02), indicating a neutral average effect but accompanied by substantial heterogeneity (I² = 78.3%, Q-test P < 0.001). I² is the weighted/pooled estimate based on the log of OR. Subgroup analyses based on exposure categories revealed distinct patterns. For amino acids, the pooled OR was 0.87 (95% CI: 0.82–0.93), suggesting a significant inverse association with T1DM risk, with low between-study heterogeneity (I² = 27.5%). In contrast, the disease/inflammation/autoimmunity category showed a significant positive association with T1DM risk (pooled OR = 1.14, 95% CI: 1.04–1.25) with moderate heterogeneity (I² = 66.9%). The proteins/genes category demonstrated a significant but modest positive association (pooled OR = 1.12, 95% CI: 1.05–1.19), alongside the highest level of heterogeneity among all categories (I² = 85.6%), The high heterogeneity makes simple pooled estimates inadequate for this category, as individual genes have unique dose-response relationships and effect patterns. For example, the clear allele-dosage effect of TYK2 loss-of-function variants (Special Findings of Meta-Analysis) shows how genetic effects change with genotype, adding to variability between studies. For lipids/fatty acids, the overall estimate suggested a neutral to potentially protective effect (pooled OR = 0.95, 95% CI: 0.89–1.02), though the wide confidence interval and considerable heterogeneity (I² = 67.3%) denote instability and require cautious interpretation. Detailed results for all subgroups are presented in structured tables and visually summarized in the forest plots of Figure 2 and Supplemental Figure 4A–K).

Meta-Analysis by Exposure Category

In the proteins/genes category, meta-analysis identified several markers with significant causal effects on T1DM. IL2RA (OR = 0.22, 95% CI: 0.17–0.27) and TYK2 (OR = 0.61, 95% CI: 0.54–0.69, P = 1.4×10⁻¹⁴) demonstrated strong protective effects. Conversely, IL6R (OR = 1.98, 95% CI: 1.48–2.65) and SIRPG (OR = 1.63, 95% CI: 1.37–1.95, P = 7.55×10⁻⁸) were associated with increased risk.Previous research indicates that TYK2’s function may vary by cell type: it is often protective in immune cells but may have different roles in pancreatic cells. This complexity is discussed further in the following section.

Metabolomics-based analyses of amino acids and metabolites revealed significant negative associations for 3-phenylpropionic acid (OR = 0.90, 95% CI: 0.85–0.96, P = 2.8×10⁻⁴) and cinnamoylglycine (OR = 0.89, 95% CI: 0.84–0.96, P = 3.5×10⁻⁴) with T1DM risk. In contrast, trimethylamine N-oxide (TMAO; OR = 1.11, 95% CI: 1.02–1.20, P = 0.015) was associated with increased risk. These findings indicated that although the overall effect of the metabolites category was neutral, several individual metabolites showed biologically plausible links to T1DM risk. This highlights the need to analyze specific exposures beyond broad category summaries.

Analysis of immune cells and phenotypes showed that lymphocyte count (OR = 0.75, 95% CI: 0.67–0.83, P < 0.001) and basophil count (OR = 0.81, 95% CI: 0.70–0.93, P = 0.003) had inverse associations with T1DM risk. Conversely, CD28 expression on CD8+ T cells (OR = 1.33, 95% CI: 1.13–1.57, P < 0.001) was associated with increased risk.

For gut microbiota, Prevotella 9 (OR = 1.18, 95% CI: 1.08–1.3, P = 1.43×10⁻⁴) was positively associated with T1DM risk, whereas Bifidobacterium (OR = 0.82, 95% CI: 0.71–0.95, P = 0.008) and Holdemania (OR = 0.85, 95% CI: 0.77–0.94, P = 0.001) showed protective effects.

Meta-analyses of autoimmune diseases and metabolic factors confirmed shared genetic risks between several autoimmune diseases and T1DM, including rheumatoid arthritis (OR = 1.16, 95% CI: 1.06–1.27, P = 9.7×10⁻⁴) and ankylosing spondylitis (OR = 1.58, 95% CI: 1.29–1.92, P = 0.0001). Childhood obesity (OR = 1.32, 95% CI: 1.06–1.64, P = 0.01) and low birthweight (OR = 0.66, 95% CI: 0.47–0.92, P < 0.05) were also significantly associated.

We applied the GRADE framework to evaluate the certainty of evidence for key causal factors, with MR studies initially rated as moderate certainty. IL2RA was rated as moderate certainty: the evidence was upgraded for a large effect size (OR = 0.22) and consistent gene-dose response, but downgraded for inter-study heterogeneity and limited population representation beyond European ancestry. TYK2 also received a moderate certainty rating: upgraded for a clear allele dose-response relationship (0, 1, 2 copies showing progressive protective effects), but downgraded for heterogeneity across studies. Childhood obesity was rated as low certainty due to high heterogeneity (I² > 50%), potential small-study bias, and indirectness of the exposure proxy (BMI-based measures). Gut microbiota associations were rated as very low certainty, reflecting extreme heterogeneity, technical variability across different 16S rRNA sequencing platforms, and limited ancestral diversity in study populations.

Special Findings of Meta-Analysis

Pooling data from multiple studies significantly enhanced statistical power for assessing rare variants. For instance, the statistical power for detecting the effect of the IL2RA rare variant rs12722495 (MAF = 0.8%) increased from 32% to 85% after meta-analysis, confirming its strong protective effect (OR = 0.22, P = 3.2×10⁻¹⁴).

Dose-response analyses revealed important patterns. TYK2 loss-of-function variants exhibited a clear allele dosage effect: 0 copies (reference, OR = 1); 1 copy (OR = 0.61, 95% CI: 0.54–0.69); 2 copies (OR = 0.37, 95% CI: 0.28–0.49). The dose-response relationship for TYK2 explains the high heterogeneity in this category, as different genotype distributions across studies lead to varying effect sizes. Therefore, simple pooled estimates are insufficient, and future analyses require genotype-stratified or dose-response methods.The protective effect of 3-phenylpropionic acid followed a U-shaped curve: <2 μM (OR = 0.92, 95% CI: 0.85–0.99); 2–5 μM (OR = 0.85, 95% CI: 0.78–0.93); >5 μM (OR = 1.12, 95% CI: 1.03–1.22). Temporal analysis revealed that the effect of childhood obesity increased exponentially with exposure duration (β = 0.21/year, P < 0.001).

Sensitivity Analysis and Publication Bias Assessment

Leave-one-out sensitivity analysis demonstrated that excluding any single study changed the pooled OR by less than 5% (range: 0.3%-4.7%), indicating that no individual study exerted undue influence on the overall results. Funnel plot symmetry (Supplemental Figure 1) and Begg’s test (P = 0.21, Supplemental Table 5) suggested a low overall risk of publication bias, though small-sample studies (<5,000 participants) showed a tendency toward larger effect sizes (β = 0.31, P = 0.02). Egger’s regression indicated minor potential bias in the protein/gene group (intercept = 0.87, 95% CI: 0.02–1.72, P = 0.043). Trim-and-fill correction changed the pooled OR for this group from 1.12 to 1.22 (7.2% change) without altering the direction of effect.

Analysis of Sources of Heterogeneity

Multilevel meta-regression analysis incorporating 27 covariates systematically evaluated sources of heterogeneity (Table 2, Supplemental Figure 2). The influence of individual studies on heterogeneity was further assessed via influence diagnostics (Supplemental Table 4). Methodological factors (eg, differences in instrumental variable selection strategies and MR analytical methods) explained approximately 38.7% of heterogeneity. Biological factors (eg, age differences, disease stage, and ancestral genetic background) contributed about 41.2% of heterogeneity. Technical factors (eg, differences in metabolomics platforms and microbiome sequencing depth) accounted for approximately 20.1% of the observed heterogeneity.

Table 2.

Assessment of Heterogeneity for Each Exposure

ID	Author (Year)	Year	Exposure	Outcome	Sample Size	Population	MR Method	No. of IVs	IV Threshold	LD (r2)	GWAS Source(s)	GWAS Platform	Outcome Definition	I2 (%)	Q p-value	Quality Score	Notes
23	Heikkilä TE et al27	2024	IL-2/IL-6/TYK2 signaling	T1DM	15,420/20,169	European	Two-sample MR	32	p<5×10−8	0.001	UK Biobank, FinnGen		ICD-10	12.3	0.21	11	Pleiotropy test passed
24	Zhu JY et al28	2024	Inflammatory bowel disease	T1DM	9,358/15,743	European	Bidirectional MR	28	p<5×10−8	0.01	FinnGen, DIAGRAM		Clinical diagnosis	25.7	0.08	10	Reverse causality tested
25	Jumentier B et al6	2025	Circulating metabolites (n=486)	T1DM	24,925	European	Metabolomics MR	112	p<1×10−5	0.3	Metabolomics GWAS		Antibody positive			9	Metabolite priority analysis
26	Richardson TG et al29	2022	Childhood body size	T1DM	12,580/18,346	European	Life-course MR	18	p<5×10−8	0.001	EGG Consortium		Pediatric registry	8.5	0.31	12	Age-stratified analysis
27	Luo J et al30	2024	Circulating immune cell counts	T1DM	10,214/15,978	European	Bidirectional MR	41	p<5×10−6	0.01	BloodCell Consortium	Whole-Genome Sequencing (WGS)	Flow cytometry	18.2	0.12	10	Cell subpopulation analysis
28	Yin Y et al31	2025	Lipidomics	T1DM	13,502/20,417	European	Two-sample MR	87	p<1×10−5	0.2	UK Biobank	Illumina HumanHap550	Mass spectrometry	22.1	0.05	9	Lipid pathway analysis
29	Liu N et al32	2023	Nonalcoholic fatty liver	T1DM complications	8,763/14,295	European	Multivariable MR	23	p<5×10−8	0.001	GWAS Catalog	Affymetrix Axiom Genotyping Array	Imaging diagnosis	14.7	0.24	11	Complication stratification
30	Guo K et al7	2024	Gut microbiota	T1DM	16,243/24,856	European	Bidirectional MR	156	p<1×10−5	0.2	MiBioGen	Sequenom MassARRAY	16S rRNA	31.5	0.01	8	Microbiota-host interaction
31	Wang Z et al33	2024	T1DM complications	Multiple	7,892/12,345	European	Network MR		p<5×10−8	0.001	FinnGen	Illumina Infinium GSA	Electronic medical record			10	Methodological innovation
32	Xie J et al34	2022	Childhood asthma	T1DM	5,672/9,843	European	Bidirectional MR	19	p<5×10−8	0.001	EAGLE Consortium	Molecular Inversion Probe (MIP) Assay	Questionnaire report	9.8	0.42	9	Pediatric population
33	Liu Y et al35	2025	Autoimmune cholestasis	T1DM	4,521/7,689	European	Bidirectional MR	14	p<5×10−8	0.001	GWAS Catalog	Axiom UK Biobank Array	Liver biopsy	5.3	0.51	8	Rare disease study
34	Jiang Y et al36	2024	Tuberculosis	T1DM	11,203/17,845	Mixed	Two-sample MR	27	p<5×10−8	0.001	GSCAN	Illumina HumanOmni5	Bacterial culture	20.4	0.09	10	Infection-immunity axis
35	Geng C et al37	2024	Alzheimer’s/Parkinson’s disease	T1DM	9,876/15,432	European	Bidirectional MR	21	p<5×10−8	0.001	IGAP, PDGENE	Whole-Exome Sequencing (WES)	Clinical criteria	11.2	0.33	11	Neurodegenerative study
36	Censin JC et al38	2017	Childhood obesity	T1DM	6,420/10,589	European	Two-sample MR	12	p<5×10−8	0.001	EGG Consortium	Affymetrix Genome-Wide SNP 6.0	BMI-Z score	7.8	0.47	12	Early-life exposure
37	Manousaki D et al39	2021	Vitamin D level	T1DM	9,358/15,743	European	Two-sample MR	6	p<5×10−8	0.001	SUNLIGHT Consortium	Illumina HumanOmni5	Serum testing			10	Nutrient study
38	Yazdanpanah N et al10	2022	Circulating protein markers	T1DM	12,456/18,923	European	Two-sample MR	38	p<5×10−8	0.001	UK Biobank	Affymetrix Genome-Wide SNP 6.0	Olink assay	15.6	0.18	11	Proteomics
39	Luo M et al40	2023	Gut microbiota	T1DM	14,567/22,189	Asian	Bidirectional MR	89	p<1×10−5	0.2	ChinaMAP	Whole-Genome Sequencing (WGS)	Metagenomics			9	Asian population data
40	Tuo L et al12	2024	Nonalcoholic fatty liver	T1DM	8,912/13,456	European	Two-sample MR	24	p<5×10−8	0.001	FinnGen	Illumina Infinium Global Screening Array (GSA)	FibroScan	17.3	0.15	10	Liver stiffness assessment
41	Yu Y et al41	2024	Immune cells	T1DM	10,345/16,782	European	Bidirectional MR	47	p<5×10−6	0.01	BIOS Consortium	Sequenom MassARRAY	Mass cytometry	23.8	0.07	11	Single-cell resolution
42	Abolo L et al5	2024	Omega-3 fatty acids	T1DM	7,689/12,345	African	Two-sample MR	15	p<5×10−8	0.001	AGEN	Axiom UK Biobank Array	RBC test	9.2	0.39	8	African population data
43	Shi Y et al42	2024	Serum metabolites	Cataract	5,432/8,765	Asian	Bidirectional MR	32	p<1×10−5	0.2	ChinaMAP	Molecular Inversion Probe (MIP) Assay	LC-MS metabolomics	14.5	0.22	9	Complication mechanism
44	De La Barrera S et al43	2024	Vitamin K level	T1DM	9,123/14,567	European	Two-sample MR	11	p<5×10−8	0.001	European EPIC	Illumina HumanHap550	ELISA assay			10	Nutrient–microbiota interaction
45	Zou M et al44	2024	Protein biomarkers	T1DM complications	11,234/17,890	European	Multivariable MR	42	p<5×10−8	0.001	UK Biobank	Affymetrix Axiom Genotyping Array	SOMAscan	19.7	0.11	11	Drug target discovery
46	Yu G et al45	2025	Autoimmune liver disease	T1DM	6,789/10,234	Asian	Bidirectional MR	18	p<5×10−8	0.001	ChinaMAP	Whole-Exome Sequencing (WES)	Clinical diagnosis	12.4	0.28	9	Cross-ethnic validation
47	Feng K et al46	2024	Esophageal varices	T1DM	4,321/7,654	Asian	Two-sample MR	9	p<5×10−8	0.001	Asian GWAS	Illumina HumanOmni5	Endoscopy	6.7	0.53	7	Rare complication
48	Chen L et al47	2025	Idiopathic pulmonary fibrosis	T1DM	8,765/13,210	European	Two-sample MR	21	p<5×10−8	0.001	UK Biobank	Sequenom MassARRAY	CT diagnosis	16.8	0.19	10	Pulmonary complication
49	Yuan S et al48	2023	Diabetes risk factors	T1DM	Meta-analyzed	Mixed	Review MR				Multiple	Affymetrix Genome-Wide SNP 6.0				12	Systematic review
50	Lin YL et al49	2024	Primary biliary cholangitis	Multiple	9,876/15,432	European	Bidirectional MR	25	p<5×10−8	0.001	FinnGen	Axiom UK Biobank Array	Ultrasound diagnosis	13.5	0.25	10	Multi-disease analysis
51	Song S et al50	2024	Serum metabolites (n=1400)	Autoimmune disease	12,345/18,756	European	Multi-omics MR	203	p<1×10−5	0.3	Metabolomics GWAS	Illumina Infinium Global Screening Array (GSA)	Mass spectrometry			11	High-throughput screening
52	Ek WE et al51	2021	Inflammatory protein markers	Inflammatory disease	10,987/16,543	European	Two-sample MR	67	p<5×10−8	0.001	UK Biobank	Whole-Genome Sequencing (WGS)	Olink assay	27.3	0.03	12	Inflammatory network analysis
53	Zhong S et al52	2025	Sedentary behavior	T1DM complications	11,234/17,896	European	Two-sample MR	18	p<5×10−8	0.001	UK Biobank, FinnGen	Molecular Inversion Probe (MIP) Assay	Questionnaire+ICD code	15.2	0.17	10	Lifestyle intervention target
54	Jin Q et al53	2024	Autoimmune targets	T1DM	8,765/13,210	Mixed	Bidirectional MR	27	p<5×10−8	0.001	GWAS Catalog, FinnGen	Illumina HumanHap550	Clinical+antibody testing	21.4	0.06	11	Drug target discovery
55	Li X et al54	2025	IGF family	Diabetes	10,345/16,782	Asian	Two-sample MR	9	p<5×10−8	0.001	ChinaMAP	Affymetrix Axiom Genotyping Array	ELISA assay	8.9	0.41	9	Growth factor pathway
56	Pan S et al55	2024	Bacterial pneumonia	Diabetes	7,689/12,345	Mixed	Two-sample MR	14	p<5×10−8	0.001	IEU OpenGWAS	Whole-Exome Sequencing (WES)	Hospitalization records	12.7	0.23	8	Infection–metabolism link
57	Jia MJ et al56	2024	Trace elements	Diabetes complications	9,123/14,567	Asian	Bidirectional MR	23	p<5×10−8	0.001	ChinaMAP	Illumina HumanOmni5	Mass spectrometry	17.5	0.14	10	Nutrient supplementation evidence
58	Li H et al57	2024	Drug targets	Diabetes	12,456/18,923	European	Multivariable MR	38	p<5×10−8	0.001	UK Biobank	Affymetrix Genome-Wide SNP 6.0	Electronic medical records	20.1	0.08	11	Systems pharmacology analysis
59	Ren Z et al58	2024	Ankylosing spondylitis	Diabetes	6,789/10,234	Asian	Bidirectional MR	16	p<5×10−8	0.001	Asian GWAS	Sequenom MassARRAY	Clinical diagnosis	11.3	0.35	9	Inflammatory arthritis link
60	Yu Y et al59	2024	Autism spectrum disorder	Diabetes	5,432/8,765	Mixed	Bidirectional MR	21	p<5×10−8	0.001	Psychiatric Genomics Consortium	Molecular Inversion Probe (MIP) Assay	Questionnaire diagnosis	14.8	0.21	8	Neurodevelopmental association
61	Zhang Y et al60	2024	Graves’ disease	Diabetes	4,321/7,654	Asian	Bidirectional MR	12	p<5×10−8	0.001	Asian GWAS	Illumina Infinium Global Screening Array (GSA)	Thyroid function test	7.6	0.48	7	Endocrine cross-disease study
62	Zhou W et al61	2023	Atopic dermatitis	Autoimmune disease	11,203/17,845	European	Multivariable MR	29	p<5×10−8	0.001	EAGLE Consortium	Whole-Genome Sequencing (WGS)	Clinical diagnosis	18.9	0.12	10	Skin–systemic immunity link
63	Wang W et al62	2024	Serum metabolites	Autoimmune disease	12,345/18,756	European	Multi-omics MR	87	p<1×10−5	0.2	Metabolomics GWAS	Axiom UK Biobank Array	Mass spectrometry	22.4	0.06	11	Biomarker screening
64	Li G et al63	2024	Migraine	Autoimmune disease	9,876/15,432	European	Bidirectional MR	24	p<5×10−8	0.001	IHGC Consortium	Illumina HumanHap550	Questionnaire diagnosis	13.7	0.27	9	Neuro–immunity axis
65	Wei G et al64	2025	Telomere length	T1DM	10,987/16,543	European	Bidirectional MR	19	p<5×10−8	0.001	UK Biobank	Affymetrix Axiom Genotyping Array	PCR assay	10.5	0.38	10	Aging mechanism study
66	Sha H et al65	2025	Multi-omics analysis	Autoimmune disease	14,567/22,189	Mixed	Network MR	203	p<1×10−5	0.3	Multi-omics Consortium	Whole-Exome Sequencing (WES)	Genome + proteome			12	Systems biology methods
67	Huang H et al66	2025	Systemic sclerosis	T1DM	6,420/10,589	European	Bidirectional MR	17	p<5×10−8	0.001	FinnGen	Sequenom MassARRAY	Clinical diagnosis	12.1	0.31	9	Rare autoimmune disease
68	Xie W et al67	2023	PCSK9 inhibitor	Autoimmune disease	11,234/17,890	European	Drug target MR	9	p<5×10−8	0.001	IEU OpenGWAS	Illumina HumanOmni5	Drug response data	6.5	0.52	8	Treatment side effect evaluation
69	Elgamal RM et al68	2024	Pancreatic enzyme levels	T1DM	12,456/18,923	Mixed	Two-sample MR	7	p<5×10−8	0.001	Pancreatic Consortium	Molecular Inversion Probe (MIP) Assay	ELISA assay	5.8	0.57	7	Biomarker discovery
70	Zhang YY et al69	2025	Multi-omics targets	T1DM	15,420/20,169	European	Multi-omics MR	156	p<1×10−5	0.2	UK Biobank, FinnGen	Affymetrix Genome-Wide SNP 6.0	Genome + metabolome	24.6	0.04	12	Precision medicine application
71	Chen et al (2024)70	2024	T1D	GCA	331,094 (105/330,989)	European	IVW (main), MR-Egger, Weighted median	6	p<5×10−8	r2<0.001	FinnGen R10; IEU OpenGWAS	Axiom UK Biobank Array	GCA (phenocode M13_GIANTCELL)	16.7	0.14	9	IVW OR=1.33 (1.22–1.46), P=9.42e-10; Consistent validation
72	Li J (2025)71	2025	Psoriasis	Diabetes	36,502 cases/325,489 controls	European	Bidirectional MR	127	p<5×10−8	0.001	FinnGen r10, DIAGRAM	Illumina Infinium Global Screening Array (GSA)	Clinical diagnosis + ICD-10	18.7	0.12	10	Negative result
73	Dahlström EH (2021)72	2021	FABP4 low-expression variant	CVD in T1D	4,380 T1D patients	European	Cohort MR	3	p<5×10−8	0.001	FinnDiane study	Whole-Genome Sequencing (WGS)	Echocardiography	9.3	0.41	8	T1D-specific cohort
74	Zhao S (2023)73	2023	Diabetes risk factors	Multiple	412,387 (UK Biobank)	European	Two-sample MR	201	p<5×10−8	0.001	IEU OpenGWAS, CKDGen	Illumina HumanHap550	eGFR<60	22.4	0.03	11	Multi-phenotype analysis
75	Fang T (2023)74	2023	Hypothyroidism	Diabetic microvascular disease	298,837 (UK Biobank)	European	Two-sample MR	48	p<5×10−8	0.001	IEU OpenGWAS, FinnGen	Affymetrix Axiom Genotyping Array	Retinopathy staging	15.8	0.21	10	Complication stratification

Notes: Cochran’s Q statistic, corresponding p-value, and I² value are reported for each exposure to quantify the degree of between-study heterogeneity.

Discussion

Key Findings

This systematic review and meta-analysis represent the most comprehensive integration of Mendelian randomization evidence on T1DM to date, incorporating data from 53 studies across 243 exposures spanning 11 biological categories. Our findings systematically map the multi-omic causal architecture of T1DM, confirming the central role of immune dysregulation through genes such as IL2RA and TYK2, elucidating the contribution of the gut microbiota-metabolite axis^75–78 (including protective effects of Bifidobacterium and 3-phenylpropionic acid), and establishing the significant impact of early-life metabolic factors such as childhood obesity. Beyond identifying these causal pathways, our analysis revealed crucial non-linear dose-response relationships and life-stage-specific effects, providing novel insights for precision intervention strategies. Furthermore, our rigorous quantification of heterogeneity sources highlights the substantial impact of methodological, biological, and technical factors on MR findings, emphasizing the need for standardized approaches in future research.

Strengths and Limitations

This study possesses several key methodological strengths. It represents the most comprehensive MR-based meta-analysis to date, systematically integrating evidence across 11 biological categories of exposures in relation to T1DM. The implementation of a dual quality assessment system—incorporating both MR-specific criteria and the GRADE framework—enhances the reliability of the conclusions. Advanced statistical approaches, including leave-one-out sensitivity analysis, multilevel meta-regression to decipher heterogeneity sources, and rigorous evaluation of publication bias, further strengthen the robustness of the findings. Notably, the identification of non-linear, dose-response relationships moves beyond simple causal inference and offers insights into therapeutic windows and concentration-dependent effects.

Several limitations must also be acknowledged. First, the number of studies available for certain exposures (eg, trace elements, specific vitamins) was limited, constraining the precision of these estimates. Second, despite efforts to control for bias, the potential for residual confounding (eg, via pleiotropy) and publication bias remains inherent to the MR methodology and reliance on published data. Third, the high heterogeneity observed in several categories (eg, proteins/genes, I²=85.6%; lipids, I²=67.3%), although investigated through meta-regression and influence analysis (Supplemental Tables 3–4), indicates underlying complexity that our analysis could not fully resolve. Fourth, the predominance of cohorts of European ancestry limits the generalizability of the findings to other populations. Finally, the reliance on aggregated data precluded individual-level analyses or more nuanced investigations of effect modifiers.

Comparison with Similar Research

Our findings consolidate and significantly extend the current evidence base on T1DM etiology. While previous MR meta-analyses have typically focused on single or limited categories of exposures,^7,12 our integrated multi-omics approach provides a unified causal landscape, revealing interactions and effect estimates across diverse biological domains. Compared to traditional observational meta-analyses,³ our MR approach provides more robust causal evidence for several factors, such as vitamin B6 and magnesium, by minimizing confounding. Furthermore, our work confirms established genetic risks (eg, IL2RA, TYK2)^79,80 while also identifying novel associations and complex patterns—such as the U-shaped effect of 3-phenylpropionic acid and the allele-dose effect of TYK2—that have not been comprehensively reported in previous studies. Notably, the TYK2 dose-response finding provides a mechanistic explanation for the exceptionally high heterogeneity observed in the proteins/genes category (I² = 85.6%), demonstrating that simple pooled estimates fail to capture the nuanced, genotype-dependent nature of these associations. Our systematic quantification of heterogeneity sources and their origins (methodological, biological, technical) also provides a new level of insight that is absent from prior reviews.

Explanations of Findings

The observed associations can be interpreted through established and emerging biological mechanisms. The strong protective effect of IL2RA variants likely operates through enhanced regulatory T cell (Treg) function and immune tolerance,⁸¹ with recent single-cell epigenomic studies showing enrichment of risk loci in Treg-specific enhancer regions.^82–84 The protective association of gut microbiota-derived metabolites like 3-phenylpropionic acid supports the “gut-pancreas axis” hypothesis,^85–87 wherein microbial products modulate immune function. The U-shaped curve for this metabolite suggests a critical concentration range for its effect, potentially reflecting saturation kinetics of microbial metabolic pathways.⁸⁸ This concentration-dependent duality has been widely observed across multiple classes of polyphenols.^89–91 While comparable data for 3-PPA itself are lacking, the same redox-switch mechanism plausibly contributes to the reversal of protection at > 5 μM.

The life-stage specificity observed for factors like childhood obesity and vitamin D⁹² may reflect critical developmental windows in immune and metabolic programming. Animal models suggest that early-life dietary exposures can permanently alter thymic T cell selection, increasing the escape of autoreactive clones,⁹³ explaining the stronger effect of childhood obesity. The cell-type-specific opposing effects of a gene like TYK2—protective in immune cells but conferring risk in exocrine pancreatic cells—highlight the intricate tissue-specificity of genetic effects, potentially regulated through cell-specific enhancers.^94,95 The link between vanillactate, dopamine synthesis, and reduced T1DM risk^96–100further reveals a novel molecular pathway connecting neurotransmitter metabolism to autoimmunity.^89,101

Implications and Actions Needed

Our findings have clear implications for both clinical practice and future research, supporting a transition toward stratified and multi-factorial prevention strategies. In the immediate term (Grade I evidence), clinical strategies could incorporate genetic risk screening using variants in HLA-DR/DQ combined with CTRB1 and consider targeted nutritional interventions like vitamin B6 supplementation for high-risk individuals. Mid-term priorities (Grade II) should include trials of precision interventions based on genetic profiles (eg, TYK2 genotype-guided therapies) and the development of nutritional formulations targeting specific metabolic pathways^101–103 (eg, phosphatidylcholines like PC (16:1/20:4)). Long-term opportunities (Grade III) include exploration of engineered microbial therapeutics and advanced gene-targeting approaches.

For the research community, we recommend: (1) conducting large-scale cross-ethnic studies to validate these associations in diverse populations; (2) developing standardized MR protocols with unified analytical frameworks to reduce methodological heterogeneity; (3) implementing longitudinal designs to better capture life-stage and cumulative effects; and (4) employing multi-omics integration and functional studies to elucidate cell-type-specific mechanisms and biological pathways, thereby translating these epidemiological findings into a deeper mechanistic understanding of T1DM pathogenesis.

Conclusions

In conclusion, this comprehensive meta-analysis of Mendelian randomization studies provides robust, data-supported evidence that the pathogenesis of T1DM is driven by a sophisticated network of interconnected causal factors spanning genetics, immunology, metabolism, and the environment. We have systematically identified and quantified a diverse array of protective and risk factors, elucidated complex dose-response and temporal relationships, and rigorously assessed the robustness and potential biases within the available evidence. The compelling consistency of findings for core immune pathways, coupled with novel insights into the roles of the gut microbiome, exocrine pancreas, and life-stage-specific metabolic factors, significantly advances our etiological understanding. This synthesized evidence base is not merely academic; it provides a foundational roadmap for prioritizing future research and for developing targeted, mechanism-based, and personalized strategies for the prevention and management of T1DM.

Funding Statement

This work was supported by the Shaanxi Provincial Key Research and Development Project (No. 2023-YBSF-065), Xi’an Science and Technology Project (21YXYJ0013) and Xi’an Children’s Hospital Research Project (No. 2022F04).

Data Sharing Statement

All data used in this meta-analysis were derived from the corresponding published articles of the included studies. The review protocol was registered in PROSPERO (International Prospective Register of Systematic Reviews) under the registration number CRD420251152014 and is available at https://www.crd.york.ac.uk/PROSPERO/view/CRD420251152014.

Ethical Statement

Ethical approval is not required for this systematic review and meta-analysis. The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Author Contributions

Conceptualization: Zhihua Wang, Mali Li.

Data curation: Zhihua Wang.

Formal analysis: Panting Shen.

Funding acquisition: Panting Shen.

Investigation: Mali Li, Panting Shen.

Methodology: Zhihua Wang.

Project administration: Zhihua Wang.

Resources: Zhihua Wang.

Software: Chao Liu, Jia Li, Shichao Qiu.

Supervision: Zhihua Wang.

Validation: Mali Li.

Visualization: Chao Liu, Jia Li, Shichao Qiu.

Writing – original draft: Mali Li.Panting Shen, Chao Liu, Jia Li, and Shichao Qiu.

Writing – review & editing: Zhihua Wang.

All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare no conflicts of interest.