Integrative Proteogenomic Analyses Provide Novel Interpretations of Type 1 Diabetes Risk Loci Through Circulating Proteins

Type 1 diabetes is an autoimmune disease characterized by the destruction of pancreatic $β$-cells (1–3), which are responsible for producing insulin. Although traditionally considered a disease of children and adolescents, type 1 diabetes can be diagnosed at any age and can affect a significant proportion of the global population (1–3). Despite continuing efforts to develop risk predictors (4,5), few effective preventive measures for type 1 diabetes have been implemented in public health practice. Following diagnosis, preserving residual $β$-cell function and delaying type 1 diabetes–associated autoimmunity present challenges due to the multifactorial and heterogeneous nature of the disease (1–3). Consequently, there is an urgent need for new biomarkers and drug targets for type 1 diabetes.

Circulating molecules participate in various biological processes and play essential roles encompassing immune responses, signaling cascades, and regulatory mechanisms (6–9). These molecules may be promising biomarkers or drug targets because their abundances are measurable and possibly modifiable. Autoantibodies to insulin, glutamic acid decarboxylase, islet antigen-2, zinc transporter 8, and other circulating proteins have emerged as biomarkers for characterizing type 1 diabetes risk (10–12). Yet establishing the causal roles of novel proteins is difficult. The feasibility of conducting randomized controlled trials for these proteins remains limited. Meanwhile, observational studies can encounter several pitfalls, including uncontrolled confounding factors as well as reverse causation.

Mendelian randomization (MR) is an instrumental variable framework that can effectively mitigate biases arising from confounding and reverse causation (13,14). MR employs genetic variants as instruments for an exposure (i.e., a circulating protein), and evaluates the potential causal effect of the exposure on a disease outcome. MR relies on three core instrumental variable assumptions (13,14). First, a genetic instrument should strongly predict the exposure, known as the relevance assumption. Second, the genetic instrument should not be associated with confounders of the instrument-disease outcome relationship, known as the independence assumption. Third, the genetic instrument should not act on the disease outcome through alternative pathways other than the instrumented exposure, known as the no horizontal pleiotropy assumption. Recent large-scale proteogenomic studies have identified ideal genetic instruments for circulating protein abundances, which are lead variants in cis-protein quantitative trait loci (cis-pQTLs) (6–8). These genetic variants are unlikely to be associated with confounders, because of the randomization of alleles at conception. Moreover, their proximity to protein-coding genes suggests a direct influence on protein abundances, thereby reducing the risk of horizontal pleiotropy. Previous studies have implemented MR to investigate potential effects of circulating protein abundances on the risks of several complex diseases (15–17).

While cis-pQTL–facilitated MR can pinpoint target proteins, it is important to note that circulating proteins originate from various sources (6–8). These include but are not limited to endocrine cell secretion, cellular turnover and apoptosis, immune and inflammatory response, and diet. To further understand the underpinning disease mechanisms and open new avenues for diagnostic and therapeutic advancements, it is crucial to identify candidate tissues and cell types where the target proteins are primarily produced.

In this study, we conducted integrative proteogenomic analyses to systematically identify potential biomarkers and drug targets for type 1 diabetes. We first capitalized on genetic associations from large-scale genome-wide association studies (GWAS) to conduct MR, to assess the associations between circulating protein abundances and type 1 diabetes risk. We then prioritized target proteins through multiple sensitivity and colocalization analyses, examinations of horizontal pleiotropy, and replications. Furthermore, we identified candidate tissues and cell types through enrichment analyses, using both bulk tissue and single-cell gene expression profiles. Our findings underscore circulating proteins that exhibit a potential effect on the risk of type 1 diabetes.

GWAS for circulating protein abundances and type 1 diabetes based on individuals predominantly of European ancestry and selection of genetic instruments are described in the Supplementary Material.

Two-sample MR was performed based on GWAS summary statistics to test associations between the genetically predicted circulating abundance of each protein and type 1 diabetes risk (18). Independent cis-pQTL variants within 500 kb of the protein-coding genes identified in the Fenland study were used as genetic instruments after linkage disequilibrium (LD) clumping (P value < 5 × 10⁻⁸ and LD r² < 0.001). Trans-genetic variants distal to the protein-coding genes likely act on other genes; thus, to mitigate the risk of horizontal pleiotropy, they were not used. If a cis-pQTL variant was unavailable in the type 1 diabetes GWAS summary statistics, we attempted to identify a proxy as the genetic instrument using the LDlink R package (19). The proxy should be in high LD (r² > 0.8) with the cis-pQTL variant based on the LD reference panel consisting of non-Finnish European ancestry populations in the 1000 Genomes Project phase 3 (20). GWAS summary statistics for genetic instruments were harmonized with forward strand alleles inferred using allele frequency information. Palindromic variants with high minor allele frequency (>0.42) were discarded to avoid allele mismatches.

Wald ratio estimates were derived for proteins with only one cis-genetic instrument, while inverse variance weighted estimates were derived for proteins with two or more cis-genetic instruments. Associations with a P value < 3.2 × 10⁻⁵ were considered significant, representing the Bonferroni-corrected significance threshold to account for 1,560 tests. This significance threshold may be overly conservative because of possible correlation and functional relevance between proteins but should effectively control the false positive rate.

MR analyses were conducted using the TwoSampleMR R package version 0.5.6 (21).

Sensitivity and follow-up analyses are detailed in the Supplementary Material. These analyses include 1) MR sensitivity analyses using alternative methods, 2) colocalization analyses to mitigate the risk of confounding due to LD, 3) functional annotation of genetic instruments and phenome-wide association studies to mitigate the risk of horizontal pleiotropy, 4) replication of findings using GWAS for circulating protein abundances and type 1 diabetes in other cohorts, 5) reverse MR to assess potential reverse causation, 6) assessment of the associations between circulating protein abundances and the risk of type 2 diabetes, 7) quantification of tissue-specific gene expression, 8) evaluation of genetic effects on tissue-specific gene expression based on cis-expression and cis-splicing quantitative trait loci (eQTL and sQTL) analyses, 9) single-cell gene expression profiling in the immune system, 10) identification of clinically relevant variants affecting target protein-coding genes, and 11) assessment of observational associations with incident disease outcomes in the UK Biobank.

All results generated in this study are included in the manuscript and the Supplementary Material. Full summary statistics of MR analyses are provided in Supplementary Table 2.

An overview of this study is presented in Fig. 1A. After identification of cis-genetic instruments and data harmonization, associations between abundances of 1,560 circulating proteins and type 1 diabetes risk were assessed using MR. MR analyses of 135 (8.7%) of these proteins used LD proxies of cis-genetic instruments (RESEARCH DESIGN AND METHODS and Supplementary Text and Supplementary Table 1).

A total of 12 associations between circulating protein abundances and type 1 diabetes risk reached the Bonferroni-corrected significance threshold (P value < 3.2 × 10⁻⁵) (Fig. 1B and Supplementary Fig. 1), excluding proteins whose coding genes map to the major histocompatibility complex (MHC) region. These significant associations had a minimal F statistic of 46.6, indicating a low risk of weak instrument bias (Supplementary Table 2).

Of these 12 proteins, circulating abundances of CTSH, ANXA2, and CCL25 were instrumented using three different cis-genetic instruments. Results obtained using weighted median, penalized weighted median, weighted mode, and MR-Egger methods were highly consistent with those obtained using the inverse variance weighted method (Supplementary Table 3). MR-Egger intercepts largely overlapped with the null, suggesting a low risk of directional horizontal pleiotropy (Supplementary Table 3).

Colocalization analyses and horizontal pleiotropy assessment were performed to verify MR assumptions for these 12 proteins (Supplementary Material). Strong (posterior probability of two traits sharing the same causal variants, PP.H4 > 80%) or suggestive (PP.H4 50–80%) evidence of colocalization between circulating protein abundance and type 1 diabetes risk was observed for CTSH, RHOC, IL27RA, ANXA2, SIRPG, CCL25, and PGM1. Conversely, colocalization evidence was limited for EBI3-IL27 complex, IL15RA, ERBB3, WARS, and ALDH2 (Figs. 1B and 2 and Supplementary Fig. 2 and Supplementary Table 4).

Among these target proteins supported by colocalization evidence, the genetic instruments for both RHOC and ANXA2 were predicted to have stronger functional connections to neighboring genes (i.e., ST7L and ICE2, respectively) than their respective coding genes, as indicated by variant-to-gene scores (24), which measure the functional connection between a variant and a nearby gene (Supplementary Text and Supplementary Table 5). In addition, the genetic instruments for RHOC, ANXA2, SIRPG, and CCL25 have also been associated with the expression or splicing of other neighboring genes, which introduces an elevated risk of horizontal pleiotropy. In contrast, the genetic instruments for CTSH, IL27RA, and PGM1 demonstrated the strongest functional connection to their respective coding genes, were not associated with the expression, splicing, or translation of other neighboring genes, and had not been associated with other known risk factors for type 1 diabetes in the Open Targets Genetics database, thereby mitigating the risk of horizontal pleiotropy (Fig. 1B and Supplementary Tables 5 and 6).

Seven significant associations supported by colocalization evidence were reevaluated when cis-genetic instruments could be identified in two other proteomic studies: deCODE (7) or the UK Biobank Pharma Proteomics Project (UKB-PPP) (22), or when the associations could be assessed based on the type 1 diabetes GWAS meta-analysis by Robertson et al. (23) (Fig. 1B and Supplementary Text and Supplementary Table 7). Six of the seven associations were replicated using these additional resources with both consistent effect direction and similar magnitudes of effects as obtained in the primary analyses (Supplementary Table 8). However, based on the cis-genetic instrument identified in the UKB-PPP study, a 1-SD increase in genetically predicted circulating abundance of CCL25 was not associated with the risk of type 1 diabetes (odds ratio [OR] 1.01; 95% CI 0.96–1.05; P value = 0.82) (Supplementary Table 8).

Following these assessments, we prioritized CTSH, IL27RA, SIRPG, and PGM1 as target proteins (Figs. 1B and 2). We caution that there is a moderate risk of horizontal pleiotropy affecting the genetic instrument for circulating SIRPG abundance. Specifically, genetically predicted circulating abundances of CTSH, IL27RA, SIRPG, and PGM1 were associated with increased odds of developing type 1 diabetes, with ORs of 1.17 (CTSH: 95% CI 1.10–1.24, P value = 9.3 × 10⁻⁷, PP.H4 = 99.6%), 1.13 (IL27RA: 95% CI 1.07–1.19, P value = 2.3 × 10⁻⁵, PP.H4 = 92.7%), 1.37 (SIRPG: 95% CI 1.26–1.49, P value = 4.3 × 10⁻¹³, PP.H4 = 86.6%), and 1.66 (PGM1: 95% CI 1.40–1.96, P value = 3.9 × 10⁻⁹, PP.H4 = 71.3%) per 1 SD increase, respectively. Notably, these associations were not detected in MR analyses using a GWAS for type 1 diabetes conducted in an East Asian ancestry population (25), likely because of the insufficient statistical power (Supplementary Text and Supplementary Table 9).

There was no evidence of reverse causation affecting these target proteins in reverse MR analyses considering the genetic liability to type 1 diabetes as the exposure and circulating protein abundances as outcomes (P value > 0.05) (Supplementary Text and Supplementary Table 10).

Interestingly, unlike CTSH, IL27RA, and SIRPG, the genetically predicted circulating abundance of PGM1 was significantly associated with decreased odds of developing type 2 diabetes (26), with an OR of 0.84 (95% CI 0.82–0.87; P value = 1.5 × 10⁻²³) per 1 SD increase (Supplementary Text and Supplementary Table 11).

For each of the target protein-coding genes, enrichment of gene expression in 54 tissue sites profiled by the Genotype-Tissue Expression (GTEx) Consortium was assessed to identify potential candidate tissues (Supplementary Text and Supplementary Table 12). As a result, the expressions of CTSH, IL27RA, and SIRPG were enriched in whole blood. with tissue-specific enrichment z scores >10 (Fig. 3). Furthermore, CTSH expression was enriched in Epstein-Barr virus-transformed lymphocytes, and SIRPG expression was enriched in the spleen, while IL27RA expression was enriched in both of these tissues (Fig. 3). In contrast, the expression of PGM1 exhibited enrichment in skeletal muscle, heart (left ventricle), and liver (Fig. 3).

Whole blood–specific cis-eQTL of CTSH and liver-specific cis-eQTL of PGM1 demonstrated strong evidence of colocalization with the genetic associations with type 1 diabetes risk, while tissue-specific cis-eQTLs of IL27RA and SIRPG did not show evidence of colocalization (Supplementary Fig. 3 andSupplementary Table 13). Importantly, the cis-genetic instruments for circulating abundances of IL27RA and SIRPG were not strongly associated with their mRNA abundances in these candidate tissues (Supplementary Fig. 4). Meanwhile, there was strong evidence of colocalization between genetic associations with multiple isoforms of both CTSH and SIRPG in whole blood and genetic associations with the risk of type 1 diabetes (Supplementary Table 14). These cis-sQTLs also overlapped with cis-pQTLs of CTSH and SIRPG (Supplementary Fig. 5), respectively.

Given the enrichment of CTSH, IL27RA, and SIRPG expression in immune system–related tissues, we further examined cell type–specific gene expression based on single-cell transcriptomic profiling of 329,762 immune cells, consisting of 45 curated cell types (Supplementary Material and Supplementary Fig. 4A). Among these immune cells, it was evident that CTSH expression was enriched in B cells, excluding pro-B cells and pre-B cells, as well as in myeloid cells (Fig. 4B, Supplementary Figs. 6A and 7–9 and Supplementary Table 15). On the other hand, CTSH expression was depleted in T cells, albeit with modest expression observed in effector memory CD4⁺ T cells (Teffector/EM_CD4) and tissue-resident memory T-helper 1 and T-helper 17 cells (Trm_Th1/Th17). Meanwhile, the expression level of IL27RA was moderate and relatively consistent across most cell types (Fig. 4C and Supplementary Figs. 6B and 7–9 and Supplementary Table 15). In contrast, SIRPG expression was enriched in T cells and natural killer cells, and depleted in B cells and myeloid cells (Fig. 4D and Supplementary Figs. 6C and 7–9 and Supplementary Table 15).

Among the four target proteins, PGM1 has been implicated in congenital disorder of glycosylation type 1t (CDG1T), an autosomal recessive disorder caused by PGM1 deficiency due to pathogenic homozygous or compound heterozygous mutations affecting PGM1 (OMIM#614921) (Supplementary Table 16). However, the relationship between CDG1T and type 1 diabetes has not been characterized. The other target proteins did not have known implications in Mendelian disorders.

In the UK Biobank, observational associations between circulating protein abundances and incident disease outcomes were available for CTSH, although type 1 diabetes was not included as an outcome because of the small number of cases. Over 16 years of follow-up, a 1-SD increase in circulating CTSH abundance was associated with a 1.14-fold increased hazard of mortality (95% CI 1.10–1.17; P value = 1.7 × 10⁻¹⁷) and, interestingly, a 1.16-fold increased hazard of type 2 diabetes (95% CI 1.11–1.21; P value = 4.0 × 10⁻¹²) (Supplementary Text, Supplementary Fig. 10, and Supplementary Table 17).

Type 1 diabetes impacts millions of individuals worldwide, causing acute and chronic complications that profoundly deteriorate the quality of life and increase mortality rates (1–3). Managing type 1 diabetes typically requires insulin injections for glycemic control, resulting in a significant socioeconomic burden (27). There is an urgent need for innovative strategies to prevent, intervene early in, and manage the disease. In this study, we conducted MR-guided target discovery to systematically examine circulating proteins that may play a crucial role in the etiology of type 1 diabetes. We also identified candidate tissues and cell types enriched for target protein-coding gene expression. Our study presents a curated selection of candidate proteins with potential as biomarkers or drug targets.

Our integrative proteogenomic analyses prioritized four target proteins, CTSH, IL27RA, SIRPG, and PGM1. Increased circulating abundances of these proteins were predicted to increase the risk of type 1 diabetes. Specifically, CTSH (cathepsin H) is a lysosomal cysteine protease involved in the degradation of lysosomal proteins. CTSH in pancreatic islets may affect $β$-cell survival and insulin secretion by modulating apoptotic signaling pathways and transcription factors (28). The genomic locus within the CTSH gene has previously been associated with the risk of type 1 diabetes as well as early-onset type 1 diabetes (29–31). In this study, we observed that CTSH expression was enriched in B cells and myeloid cells, implying a potential role of CTSH in antigen presentation and antibody-mediated immunity. Furthermore, although the genetic risk might be conferred by differences in gene expression, colocalization between the genetic associations with multiple isoforms of CTSH in whole blood and the risk of type 1 diabetes suggests that alternative splicing of CTSH may contribute to the disease pathogenesis. In the UK Biobank, increased circulating CTSH abundance was linked to higher mortality and risks of common autoimmune diseases. Importantly, while the observational association analyses did not encompass type 1 diabetes as an outcome because of the limited number of cases, increased circulating CTSH abundance was associated with an increased risk of type 2 diabetes based on health records. This suggests the possibility of misdiagnosing type 1 diabetes as type 2 diabetes (1) within the adult population of the UK Biobank, or, less likely, a shared molecular basis of type 1 diabetes and type 2 diabetes. Increased CTSH expression has also been associated with early-onset type 1 diabetes in children diagnosed under age 7 years (31) and rapid decline of $β$-cell function in other cohort studies (28). Taken together, our findings strongly encourage functional follow-up studies to explicate the role of CTSH in type 1 diabetes and to evaluate its potential as a biomarker or drug target.

IL27RA (α subunit of the interleukin 27 receptor) binds to IL27, a heterodimeric cytokine composed of IL27p28 and EBI3 subunits. IL27 has both proinflammatory functions by mediating T-helper 1 cell differentiation and increasing interferon $γ$-production, and anti-inflammatory functions by inhibiting proinflammatory cytokines in T cells and promoting the production of anti-inflammatory cytokines (32). However, because of the lack of colocalization evidence and potential horizontal pleiotropic effects, we were unable to determine the effect of IL27. On the other hand, the association between circulating abundance of IL27RA and the risk of type 1 diabetes was substantiated by multiple lines of evidence. The results of our enrichment analyses align with the involvement of IL27RA in cell-mediated and antibody-mediated immunity by mediating IL27 signaling in various immune cells (33). While the functions of IL27RA and IL27RA-mediated IL27 signaling in type 1 diabetes remain to be explored, we posit that IL27 and IL27RA may regulate both innate and adaptive immune responses that attack the pancreatic $β$-cells.

SIRPG (signal-regulatory protein $γ$⁠) is a receptor protein involved in the negative regulation of receptor tyrosine kinase-coupled signaling processes. It has been suggested that SIRPG signaling may play an immunoregulatory role in maintaining peripheral immune tolerance and preventing autoimmunity (34). In line with existing studies, our analyses demonstrated that SIRPG expression was enriched in T cells and natural killer cells, where blocking of the SIRPG-CD47 interaction has been found to inhibit superantigen-induced T-cell proliferation (35,36). These findings imply the potential significance of investigating SIRPG as a T-cell–specific target for type 1 diabetes, although it should be noted that the genetic instrument for circulating SIRPG abundance was subject to a moderate risk of horizontal pleiotropy due to its associations with the expression, splicing, or translation of neighboring genes encoding other signal-regulatory proteins, including SIRPB1, SIRPB2, and SIRPD. While our MR analyses did not detect any association between circulating SIRPB1 abundance and the risk of type 1 diabetes, the potential effects of circulating abundances of SIRPB2 and SIRPD could not be estimated because of lack of cis-genetic instruments.

PGM1 (phosphoglucomutase 1) is an enzyme that catalyzes the reversible conversion between glucose 1-phosphate and glucose 6-phosphate, which are important intermediates in glucose metabolism. The Mendelian disorder of PGM1 deficiency can result in congenital disorder of glycosylation. Given the crucial functions of insulin in the uptake of glucose into cells and the regulation of glycogen synthesis and breakdown, we hypothesize that PGM1 may play a role in type 1 diabetes through influencing the balance between glycogen storage and glucose utilization, or by affecting glycolytic intermediates involved in autoimmune response that leads to $β$-cell destruction, particularly in muscle and liver tissues. Notably, previous GWAS have suggested that the PGM1-increasing allele of the genetic instrument, which increases the risk of type 1 diabetes, may have a protective effect against type 2 diabetes (26,37). Elucidating the precise involvement of PGM1 in diabetes necessitates further efforts.

Although deprioritized proteins, such as IL15RA and ERBB3, have been previously linked to the risk of type 1 diabetes in different contexts (38,39), substantiating whether these associations truly denote causal effects requires future efforts. Notably, the genetic associations with type 1 diabetes risk near IL15RA appear to more strongly implicate IL2RA (Supplementary Fig. 2). This is consistent with previous fine-mapping efforts that prioritized IL2RA, rather than the flanking genes, as the likely causal gene within this locus (40). Although our MR analyses did not include IL2RA because of lack of cis-genetic instruments, soluble IL2RA is a well-established biomarker for immune activation (41). Additionally, analyses using the genetic instruments identified in the UKB-PPP study failed to replicate the association between circulating CCL25 abundance and type 1 diabetes risk, highlighting the potential influence of protein detection platform and study population.

Our study has important limitations. First of all, our findings rely on fundamental assumptions of MR, where risks of confounding and horizontal pleiotropy may not be eliminated. In particular, sensitivity analyses using alternative MR methods could not be performed when only one genetic instrument was available for an exposure. Our results require further triangulation through multiple lines of evidence, including experimental validation. Meanwhile, our study leverages assays that may have measured only a subset of isoforms for each protein. This aspect could be the focus of follow-up studies, especially regarding CTSH and SIRPG. Second, the definitions of type 1 diabetes may vary across participating cohorts of the meta-analysis of GWAS. Future investigations may characterize the disease heterogeneity by examining variations in protein abundances among patients with distinct autoantibody profiles, C-peptide levels, or disease durations, or those classified as having latent autoimmune diabetes in adults or fulminant type 1 diabetes (42,43). Longitudinal or cross-sectional studies with detailed phenotypic data may further assess the associations between circulating protein abundances and the clinical severity and progression of type 1 diabetes. Third, our analyses were restricted to populations predominantly of European ancestry, while the East Asian ancestry-based GWAS had a much smaller sample size. Given the substantial variability in the prevalence and strong heterogeneity of type 1 diabetes across different populations in different countries (1–3), it is important to exercise caution when generalizing our findings to populations of non-European ancestries. Fourth, it should be noted that all cis-pQTLs used in this study were identified in middle-aged and older adults, whereas the type 1 diabetes GWAS included patients who typically develop disease at a younger age. However, we posit that the cis-genetic regulation of circulating protein abundances is likely consistent across age distributions. Nevertheless, we strongly advocate for similar analyses to be conducted across populations of diverse ancestries and demographic characteristics. Fifth, although existing protein detection platforms have enabled the measurement of circulating abundances for nearly 5,000 proteins, the possibility remains that potential target proteins lack valid genetic instruments. Genetics-guided target discovery based on proteogenomic studies featuring increased sample sizes and enhanced coverage of the circulating proteome should be pursued in the future. Last but not least, because of the high variability, intricate LD structure, and strong pleiotropy of the MHC region (44), we did not prioritize MHC gene-coded proteins, despite significant associations identified through MR. Considering the well-established role of the MHC region in the pathogenesis and progression of type 1 diabetes, future efforts should be dedicated to elucidating the functional impacts of these proteins.

In conclusion, through integrative proteogenomic analyses, we identified significant associations between circulating protein abundances and the risk of type 1 diabetes, which further suggested possible effects of CTSH, IL27RA, SIRPG, and PGM1. The roles of CTSH, IL27RA, and SIRPG in the immune system are underscored, with enrichment of CTSH expression in B cells and myeloid cells, and SIRPG expression in T cells and natural killer cells. In contrast, PGM1 may influence the risk of type 1 diabetes through its impact on glucose metabolism, particularly in muscle and liver tissues. Exploration of these target proteins as biomarkers or viable candidates for drug targeting strategies while considering the candidate tissues and cell types may be warranted in the context of type 1 diabetes.

Funding. T.L. has been supported by a Schmidt AI in Science Postdoctoral Fellowship and start-up funding from the Office of the Vice Chancellor for Research and Graduate Education, School of Medicine and Public Health, and Department of Population Health Sciences at the University of Wisconsin-Madison.

The funder has no role in study design; collection, management, analysis and interpretation of data; or the decision to submit for publication.

Duality of Interest. T.L. was employed by 5 Prime Sciences Inc. until September 2023 and has been providing consulting services to 5 Prime Sciences Inc. since December 2023. The research presented in this paper was conducted independently, and 5 Prime Sciences Inc. was not involved in the design, execution, analysis, or interpretation of the study. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. T.L. conceived and designed the study, acquired data, performed data analysis, interpreted the results, and wrote the first draft of the manuscript. D.M., L.S., and A.D.P. contributed to the interpretation of results. All authors revised the manuscript critically and approved the final version of the manuscript. T.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This work was presented orally at the International Genetic Epidemiology Society Annual Conference, 5–7 November 2023, Nashville, TN.