Abstract
Hypothesis/Aim Type 1 diabetes (T1D) is associated with excess coronary artery disease (CAD) risk even when known cardiovascular risk factors are accounted for. Genetic perturbation of hematopoiesis that alters leukocyte production is a novel independent modifier of CAD risk. We examined whether there are shared genetic determinants and causal relationships between leukocyte counts, T1D, and CAD.
Methods Genome-wide association studies (GWAS) summary statistics were utilized to perform pairwise linkage disequilibrium score regression (LDSC) and heritability estimation from summary statistics (ρ-HESS) to respectively estimate the genome-wide and local genetic correlations, and two-sample Mendelian randomization (MR) to estimate the causal relationships between leukocyte counts (335 855 healthy subjects), T1D (18 942 cases, 501 638 controls), and CAD (122 733 cases, 424 528 controls).
Results There was significant genome-wide genetic correlation (rg) between T1D and CAD (rg = 0.088; P = 9.0e-03) and both diseases shared significant genome-wide genetic determinants with eosinophil count (rg(T1D) = 0.093, P = 7.20e-03; rg(CAD) = 0.092, P = 3.68e-06) and lymphocyte count (rg(T1D) = −0.052, P = 2.80e-02; rg(CAD) = 0.1761, P = 1.82e-15). Sixteen independent loci showed stringent Bonferroni significant local genetic correlations between leukocyte counts, T1D and/or CAD. Cis-genetic regulation of the expression levels of genes within loci that are shared between T1D and CAD were associated with both diseases as well as leukocyte counts, including SH2B3, CTSH, MORF4L1, CTRB1, CTRB2, CFDP1, and IFIH1. Genetically predicted lymphocyte, neutrophil, and eosinophil counts were associated with T1D and CAD (lymphocyte odds ratio (OR)T1D = 0.667, P = 2.02e-19; ORCAD =1.085, P = 2.67e-06; neutrophil ORT1D = 0.82, P = 5.63e-05; ORCAD = 1.17, P = 5.02e-14; and eosinophil ORT1D: 1.67, P = 4.45e-25; ORCAD: 1.07; P = 2.02e-03).
Conclusions/Interpretations This study sheds light on shared genetic mechanisms that underlie T1D and CAD, which may contribute to their co-occurrence through regulation of gene expression and leukocyte counts. This study also identifies molecular targets for further investigation for disease prediction and potential drug discovery.
What is already known about the subject?
Genetic factors have been shown to contribute to the occurrence of coronary artery disease (CAD) in type 1 diabetes (T1D), but the mechanisms are unknown.
Genetic perturbation of hematopoiesis that alters leukocyte production is associated with CAD risk and clinical observations have documented altered leukocyte counts in T1D. However, it is unknown whether altered leukocyte frequencies contribute to T1D and the co-occurrence of T1D and CAD or these reflect reverse causation.
What is the key question?
Do T1D and CAD share genetic determinants with leukocyte counts, and if so, are genetically predicted leukocyte counts associated with risk of T1D or CAD and their co-occurrence?
What are the new findings?
T1D and CAD share significant genetic architecture, and both diseases share significant genetic determinants with eosinophil and lymphocyte counts.
Genetically predicted eosinophil, lymphocyte, and neutrophil counts are associated with risk of T1D and CAD.
Genetic regulation of the expression levels of genes in shared loci between T1D and CAD are associated with both diseases and leukocyte counts.
How might this impact on clinical practice in the foreseeable future?
Genetic heritability for T1D is shared with CAD risk and leukocyte counts, and the counts of eosinophils, lymphocytes, and neutrophils are associated with both diseases.
Introduction
Type 1 diabetes (T1D) results from autoimmune-mediated destruction of insulin-producing pancreatic beta cells. Epidemiological studies suggest that coronary artery disease (CAD) occurs in patients with T1D more than in any other disease [1, 2]. Even when hyperglycemia and other known cardiovascular risk factors are accounted for, patients with T1D still have an increased risk of CAD relative to the general population [3–5]. The residual risk factors and their mechanisms that contribute to CAD in T1D are poorly defined, which complicates disease prediction and treatment. Population-based genetic studies have shown that T1D is genetically linked to CAD independent of traditional cardiovascular risk factors [3]. However, how genetic factors contribute to the comorbidity between T1D and CAD remains undefined. Genetic perturbation of hematopoiesis that alters the production of leukocytes is a novel independent modifier of several clinical outcomes, including autoimmune diseases and CAD [6–8]. For instance, genetic mutations in hematopoietic stem and progenitor cells (HSPCs) that alter leukocyte production known as clonal hematopoiesis of indeterminate potential (CHIP) can independently increase the risk of CAD up to four-fold [8]. Interestingly, integrated genetic and epigenomic analyses have shown that T1D risk variants are enriched in open chromatin of multiple hematopoietic cell types, including HSPCs [9–11]. Analyses of atherosclerotic lesions in patients also revealed that CAD risk variants are enriched in open chromatin of multiple cell types, including multiple immune cell types [12, 13].
Genetic variants that impact multiple hematopoietic cell types normally affect progenitor cells (i.e., HSPCs) to direct the production of distinct lineages [14, 15]. Thus, the above observations suggest that genetic perturbation of the production of leukocytes may contribute to T1D and CAD and potentially, their co-occurrence. In line with this hypothesis, clinical observations have documented altered leukocyte counts that correlate with the progression of T1D [16–19] and CAD [20]. In particular, the counts of neutrophils, which constitute the bulk of circulating leukocytes and are short-lived, requiring continuous replenishment by hematopoiesis [21], are consistently diminished in genetically at-risk individuals for T1D and in those diagnosed with the disease [16–19]. However, it is unknown whether these altered leukocyte counts causally contributes to disease and the co-occurrence of T1D and CAD, or these altered leukocyte counts are simply a consequence of disease. Unlike observational studies, association studies that employ inherited genetic variants as instruments are not confounded by environmental or lifestyle and disease. In this study, we utilized large-scale genome-wide association studies (GWAS) summary statistics to perform statistical genetic analyses to assess the genetic correlations and the potential causal relationships between leukocyte counts, T1D, and CAD.
RESEARCH DESIGN AND METHODS
GWAS datasets and study design
The overall study design is illustrated in Figure 1A. For leukocyte counts, we used summary statistics of our previously published blood cell traits GWAS based on 335 855 UK Biobank (UKB) European participants who had no conditions or diseases that may influence complete blood counts [22]. GWAS summary statistics for T1D (18 942 cases; 501 638 controls) [9] and CAD (122 733 cases; 424 528 controls) [23], whose participants were also of European ancestry, were obtained from the GWAS catalog (accession numbers GCST90014023 and GCST005195, respectively).
(A) Schematic of study design. (B) Genome-wide genetic correlations (±se) between phenotypes estimated by LDSC. (C) Local genetic correlations between leukocyte counts, T1D, and/or CAD. (D-G) Local genetic correlations between T1D, CAD, and leukocyte (eosinophil and lymphocyte) counts at genes of apriori interests. Manhattan plots showing local genetic correlations between T1D, CAD, and leukocyte (EOS and LYM) counts at the (D) SH2B3/CUX2/PTPN11 and (C) IFIH1 (D) MORF4L/CTSH and (E) CTRB2/CTRB1/CFDP1 loci. For each gene locus, local heritability and genetic correlation estimates are shown for each pair of traits for the LD-independent block containing the gene, with corresponding p values (−log10 P).
Genetic correlation analyses
We conducted estimation of the genome-wide genetic correlations (rg) between leukocyte counts, T1D, and CAD using linkage disequilibrium score regression (LDSC). Analyses were performed using the HapMap3/1000 Genomes reference panel, restricting to variants with minor allele frequency (MAF) >0.01 outside of the major histocompatibility complex (MHC) region. Leukocyte counts included basophil count (BAS), eosinophil count (EOS), lymphocyte count (LYM), monocyte count (MON), and neutrophil count (NEU), yielding 11 phenotype pairs, i.e., T1D/CAD, T1D/BAS, T1D/EOS, T1D/LYM, T1D/MON, T1D/NEU, CAD/BAS, CAD/EOS, CAD/LYM, CAD/MON, and CAD/NEU.
In addition to genome-wide genetic correlation, locus-specific or local genetic correlation between phenotypes could provide further evidence of shared etiology. For instance, there may be some regions in the genome where genetic correlations between phenotypes differ in magnitude or sign from the genome-wide genetic correlation [24]. We performed local genetic correlation analyses using heritability estimation from summary statistics (ρ-HESS) [25], which uses a fixed-effect model to estimate heritability and genetic covariances within specific regions of the genome, while accounting for linkage disequilibrium (LD) and sample overlap [25]. Local genetic correlations between T1D, CAD, and leukocyte counts were estimated at 1453 LD-independent blocks across the genome, excluding the MHC region [26]. We conducted two types of analyses of the ρ-HESS results. We first undertook an agnostic genome-wide analysis approach to identify the blocks for each phenotype pair that had heritability and genetic covariance significantly different than 0 at a stringent Bonferroni threshold of p<0.05/1453 blocks (i.e., p<3.44e-05). In the second approach, we adopted a candidate gene approach, examining the local genetic correlations at blocks containing pre-selected genes based on previous studies that implicate these genes in either T1D, CAD or hematopoietic traits [6, 27, 28]. Local genetic correlations were only computed for candidate gene blocks with significant local heritability observed for both paired-phenotypes. After filtering for loci with non-significant local heritability, we performed 36 significance tests for local genetic covariance at candidate gene blocks and used a Bonferroni threshold of p < 0.05/36 (i.e. p < 1.34e-3).
Two-sample Mendelian randomization (MR) analyses
All MR analyses were performed using the TwoSampleMR package in R.
Causality of shared gene loci between T1D and CAD
To infer the putative causality of loci that showed local genetic correlation between T1D and CAD, we used cis-expression quantitative trait loci (cis-eQTL) of the nominated/overlapping genes in the shared loci as instruments to perform the MR analyses because genetic variants underlying complex diseases such as T1D and CAD predominantly act as cis-regulatory elements [9, 29, 30]. We extracted cis-eQTLs (eSNPs) that significantly (p<1.0e-05; FDR<0.05) regulate the expression levels of the nominated genes from the Genotype-Tissue Expression (GTEx) consortium portal in disease-relevant tissue whenever available, including whole blood, pancreas, and EBV-transformed lymphoblastoid cell lines (LCLs). Using these eSNPs as instrumental variable (IVs), we performed two-sample MR by calculating the Wald ratios to estimate the effect of the expression levels of each gene on the risk of T1D, CAD, and leukocyte counts. The multiplicative random effects inverse variance weighted (IVW-mre) approach was performed to combine the Wald ratios of multiple cis-eQTLs to obtain an overall MR estimate for each gene.
Association between leukocyte counts, T1D, and CAD
To investigate the association between genetically inferred leukocyte counts and risk of T1D and CAD, we performed bi-directional MR analyses by first treating leukocyte counts as the exposure, followed by the reverse analyses, wherein leukocyte counts were outcomes and genetic predisposition to T1D and CAD were the exposures. IVs for each exposure were LD-independent SNPs that reached genome-wide significance (p< 5.0e-08) for the given phenotype. We have previously described the IVs for leukocyte counts [22]. The IVs for T1D and CAD were clumped with an LD threshold of r2<0.001 based on the European 1000 Genomes LD reference to obtain independent IVs. MR makes several strong and untestable assumptions. For instance, horizontal pleiotropy, which occurs when a genetic instrument influences the outcome directly or through any pathway other than through the exposure, is a major threat to valid inference in MR. Therefore, we used multiple analytic approaches that make different assumptions to assess the stability of the results. The IVW-mre method, which accounts for some heterogeneity due to the presence of potentially invalid instruments, was used as the primary method to estimate the causal relationships between phenotypes. Heterogeneity was assessed using the Cochran’s Q statistic, whereby a Cochran Q test p value <0.05 indicated the presence of significant heterogeneity. If there was evidence of significant heterogeneity, outliers contributing to heterogeneity were pruned using Radial IVW regression [31]. We then looked for consistency in the IVW-mre results based the weighted median test [32], which is less sensitive to the influence of outliers, and the MR-Egger test, which tests for the presence of directional pleiotropy and corrects the effect estimate if pleiotropy is detected.
Unless otherwise stated, the MR results are presented as odds ratio (OR) and 95% CI (CI) of the risk of the outcome per 1-standard deviation (1-SD) increase in genetically predicted levels of the exposure.
RESULTS
Genetic correlations between T1D and CAD, and leukocyte counts
Shown in Figure 1B, there was significant positive genome-wide genetic correlation (rg) between T1D and CAD (rg = 0.088, P=0.009). Both T1D (rg = 0.093, P = 7.20e-03) and CAD (rg = 0.092, P = 3.68e-06) had significant positive genome-wide genetic correlations with eosinophil count. In contrast to eosinophil count, T1D had a significant negative genome-wide genetic correlation with lymphocyte count (rg = −0.052, P = 2.80e-02), whereas CAD had a significant positive genome-wide genetic correlation with lymphocyte count (rg = 0.1761, P = 1.82e-15; Figure 1B). T1D also showed a negative genome-wide genetic correlation with neutrophil count, albeit statistically insignificant (rg = −0.0115, P = 5.080e-01; Supplemental (Suppl) Table 1). CAD had a significant positive genome-wide genetic correlation with all other leukocyte counts, including, basophil, monocyte, and neutrophil counts (rg ≥ 0.0858, P<0.0011; Figure 1B and Suppl Table 1).
Only eosinophil and lymphocyte counts were included in the local genetic correlation analyses because only these showed significant genome-wide genetic correlations with both T1D and CAD. The agnostic scan across the 1453 LD-independent blocks identified 16 genomic loci with Bonferroni significant (p<3.44e-05) local genetic correlation and local SNP heritability between phenotype pairs, i.e., between eosinophil and/or lymphocyte counts and T1D and /or CAD (Figure 1C; Suppl Data 1). There were an additional 55 genomic loci that showed nominally significant (p<0.05) local genetic correlations and local SNP heritability between phenotype pairs (Suppl Data 1).
Of the 16 significantly locally correlated loci, five were at an LD block on chromosome 12, of which Chr12:110336719-113263518 that harbors rs3184504 and rs7137828, respectively overlapping SH2B3 and ATXN2, showed significant local genetic correlation across all four phenotypes, i.e., eosinophil count, lymphocyte count, T1D, and CAD (Figure 1D; Suppl Data 1). In addition to the SH2B3/ATXN2 locus, two other genomic loci showed local genetic correlation between T1D and CAD at nominal significance, including Chr15:78516053-80860978 that harbors rs345593439 and rs7173743 that overlap CTSH and MORF4L1 respectively (local rg(T1D-CAD) = −0.4709, Pρ-HESS = 1.87e-02; Figure 1E; Suppl Data 1), and Chr16:74971503-75977954, which harbors rs55993634 and rs8046696 that respectively overlap CTRB2 and CFDP1 (local rg(T1D-CAD) = −0.6556, Pρ-HESS = 1.93e-02; Figure 1F; Suppl Data 1). Of the 16 significantly shared loci (Figure 1C), Chr15:67094767-69017999 (rs72743461-SMAD3 and rs17293632-SMAD3) was shared between CAD and eosinophil count only, while the rest of the 16 loci showed local genetic correlations between T1D and leukocyte counts (eosinophil and lymphocyte counts) only. These include IKZF4/SUOX (12:55665837-57548860) CLEC16A (Chr16:10426032-11520161), HORMAD2 (22:29651799-31439918), IRF1-AS1(5:129519025-132139649), BACH2 (6:89973052-91843196), PTPN22 (1:113273306-114873845), BLTP1/ADAD1/IL21(4:122657987-124286481), ZBTB9/TCP11/BAK1 (6:33236497-35455756), IL2RA (10:5983762-7171484).
The ρ-HESS analysis at 23 apriori selected gene loci corroborated the strong local genetic overlap and covariance across T1D, CAD, and eosinophil and lymphocyte counts at the LD-independent block on chromosome 12 that contains genes including SH2B3, PTPN11, and CUX2 (local rg>0.8; P<1.0e-08; Figure 1D; Suppl Data 1). The IFIH1 locus on Chr 2 also showed local genetic correlation across eosinophil count, lymphocyte count, T1D and CAD at nominal significance (Figure 1G; Suppl Data 1), which was also identified by the agnostic ρ-HESS analysis, i.e., Chr2: 161769733-163503551 harboring rs2111485-IFIH1 and rs2075302-IFIH1 (local rg(T1D-EOS) = −0.5328, Pρ-HESS = 3.00e-03; local rg(T1D-LYM) = −0.4626, Pρ-HESS = 6.40e-03; Suppl Data 1). However, while the local genetic correlation between T1D and CAD was statistically significant at the IFIH1 locus (Pρ-HESS = 1.67e-02), the local heritability for CAD did not reach the significance threshold (Suppl Data 1). The CTRB1 locus on Chr16:75255860 also showed nominally significant local genetic correlation between T1D and CAD (local rg = −0.6456, Pρ-HESS = 1.93e-02). Other loci that showed local genetic correlation with either CAD and leukocyte counts or T1D and leukocyte counts include TYK2-Chr19:10476280.5 (local rg(T1D-LYM) = −0.5982, Pρ-HESS = 8.42e-05), JAK1-Chr1:65365549.5 local rg(T1D-EOS) = 0.7810, Pρ-HESS = 5.25e-05), and IKZF3-Chr17:37970819.5 (local rg(T1D-LYM) = −0.7399, Pρ-HESS = 2.73e-03). Of note, these ρ-HESS analyses results do not include several cases where significant local genetic covariance was observed in the setting of non-significant local heritability, as we elected conservatively not to interpret such cases (point estimates are included in Suppl Data 1).
Genetically determined expression of genes in loci shared between T1D and CAD is associated with risk of both diseases and leukocyte counts
The results for the effects of cis-eQTL-mediated expression of shared genes between T1D and CAD are summarized in Table 2. Overall, genetically determined expression levels of genes in shared loci between T1D and CAD were associated with risk of both diseases, either in the same or opposing directions, as well as altered leukocyte counts. Cis-eQTL-mediated upregulation of blood expression levels of SH2B3, whose locus positively overlapped across T1D, CAD, and leukocyte counts (EOS and LYM), were associated with reduced risk of both T1D (OR = 0.30, CI = 0.25-0.37, P = 3.50e-31) and CAD (OR = 0.80, CI = 0.72-0.88, P = 5.69e-06), and altered counts of all leukocyte types except monocytes (Table 2). The results for genetically determined increased expression levels of SH2B3 in pancreas were similar to that for blood (Table 2). Genetically determined increased blood and pancreas expression levels of CTSH, MORF4L1, and CFDP1, whose loci showed negative local genetic correlation between T1D and CAD, also had opposing causal effects on the two diseases. Thus, while genetically determined increased blood and pancreas expression levels of CTSH were associated with increased T1D risk (ORblood = 1.45, CI = 1.43-1.47, P <2.23e-308; ORpancreas = 1.27, CI = 1.26-1.29, P = 6.30e-221), the risk of CAD was reduced (ORblood = 0.81, CI = 0.80-0.82, P = 4.10e-221; ORpancreas = 0.92, CI = 0.92-0.93, P = 8.10e-122). Likewise, genetic upregulation of blood and pancreas expression levels of CFDP1 were associated with increased risk of T1D (ORblood = 1.23, CI = 1.23-1.24, P <2.23e-308; ORpancreas = 1.38, CI = 1.36-1.40, P <2.23e-308) but reduced CAD risk (ORblood = 0.80, CI = 0.79-0.80, P <2.23e-308; ORpancreas = 0.81, CI = 0.80-0.81, P <2.23e-308). Similarly, while cis-genetic upregulation of the expression levels of MORF4L1 in blood and pancreas were associated with increased T1D risk (ORblood = 1.86, CI = 1.81-1.90, P <2.23e-308; ORpancreas = 1.20, CI = 1.19-1.21, P = 3.10e-301), the risk of CAD was reduced (ORblood = 0.54, CI = 0.53-0.55, P <2.23e-308; ORpancreas = 0.87, CI = 0.86-0.88, P = 9.60e-211). Unlike CTSH where genetic upregulation in blood, pancreas, and LCLs had the same direction of effects on T1D and CAD, upregulation of CFDP1 and MORF4L1 in LCLs had an opposing direction of effect compared to that in blood and pancreas, wherein genetically determined higher levels of CFDP1 and MORF4L1 in LCLs were associated with reduced T1D risk (ORCFDP1 = 0.96, CI = 0.95-0.97, P = 2.50e-10; ORMORF4L1 = 0.96, CI = 0.95-0.97, P = 2.50e-10) and increased CAD risk (ORCFDP1 = 1.15, CI = 1.11-1.20, P = 2.64e-15; ORMORF4L1 =1.22 CI = 1.22-1.23, P <2.23e-308). Whereas cis-genetic upregulation of pancreas expression levels of CTRB1 were associated with reduced risk of both T1D (OR = 0.83, CI = 0.80-0.86, P = 4.59e-23) and CAD (OR = 0.92, CI = 0.91-0.93, P = 1.53e-70), upregulation of pancreas expression levels of CTRB2 were associated with increased risk of both T1D (OR = 1.07, CI = 1.06-1.08, P = 5.87e-34) and CAD (OR = 1.02, CI = 1.02-1.03, P = 2.04e-25). We did not find cis-eQTLs for CTRB1 and CTRB2 in blood or LCLs from GTEx. Genetically determined expression levels of CTSH, MORF4L, CFDP1, CTRB1, CTRB2 were also associated with leukocyte counts, except blood expression of CTSH that did not associate with eosinophil count; pancreas expression of CTSH that did not affect basophil and lymphocyte counts; and pancreas expression of CTRB2, which did not affect lymphocyte count.
Genetically determined increased blood expression levels of IFIH1 were associated with increased risk of both T1D (OR = 1.12, CI = 1.11-1.14, P = 3.28e-28) and CAD (OR = 1.24, CI = 1.04-1.49, P = 1.67e-02), as well as altered leukocyte counts. However, while genetically predicted transcript levels of IFIH1 in the pancreas were associated with risk of CAD (OR = 1.02, CI = 1.02-1.02, P = 9.11e-73), the risk of T1D was insignificant (OR = 0.99, CI = 3.45e-02). Given that IFIH1 is an established T1D susceptibility gene, the finding that cis-genetic upregulation of IFIH1 expression levels in blood increases both T1D and CAD risk led us to investigate whether genetic variants in IFIH1 that are associated with T1D risk also influence CAD risk. Using T1D GWAS significant (p<5.0e-08) SNPs that reside within the IFIH1 gene and impact transcript and/or protein function [9, 11, 33, 34] as genetic instruments, we found that IFIH1 variants that increase T1D risk also increase CAD risk (OR = 1.16, CI = 1.11-1.21, P = 5.83e-10; Figure 2) and alter leukocyte counts, including decreased counts of basophils (OR = 0.96, CI = 0.95-0.98, P = 5.97e-06), eosinophils (OR = 0.92, CI = 0.89-0.96, P = 2.64e-05), lymphocytes (OR = 0.89, CI = 0.87-0.91, P = 5.98e-31), and neutrophils (OR = 0.95, CI = 0.92-0.97, P = 6.68e-05), except monocytes (OR = 1.00, CI = 0.99-1.01, P = 8.40e-01).
(A) Annotation of GWAS variants in IFIH1 associated with T1D risk (p<5.0e-8) used as genetic instruments for MR analysis (left panel) and MR estimates of their effects on CAD and leukocyte counts (right panel). (B-G) Scatter plot showing the effect of T1D risk variation in IFIH1 on (B) CAD, (C) basophil count, (D) eosinophil count, (E) lymphocyte count, and (G) neutrophil count.
Genetically determined eosinophil, lymphocyte, and neutrophil counts is associated with T1D and CAD risk
The primary IVW-mre MR analysis showed that genetically predicted increased eosinophil count is associated with increased risk of T1D (OR = 1.67, CI = 1.52-1.85, P = 5.45E-25) and CAD (OR = 1.07; CI = 1.03-1.11, P = 2.02e-04; Table 2 and Suppl Data 2). These results were consistent in the MR-Egger test, except in the Weighted median test wherein eosinophil count did not associate significantly with CAD risk (OR = 1.02, CI = 0.96-1.09, P = 4.70e-01). In contrast to eosinophil count, genetic upregulation of the counts of basophils, neutrophils, and lymphocytes were inversely associated with T1D risk (ORBAS = 0.70, CI = 0.58-0.83, P = 5.57e-05; ORNEU = 0.82, CI = 0.75-0.90, P = 5.63e-05; ORLYM = 0.67, CI = 0.61-0.73, P = 2.02e-19), while positively associating with CAD risk (ORBAS = 1.23, CI = 1.13-1.35, P=2.47e-06; ORNEU = 1.17, CI = 1.12-1.21, P= 5.02e-14; ORLYM = 1.09, CI = 1.05-1.12, P = 2.67e-06; Table 2 and Suppl Data 2). Genetic predisposition to T1D was not associated with CAD and vice versa. The reverse MR analyses showed that genetic risk for T1D was associated with decreased lymphocyte count (OR = 0.98, CI = 0.98-0.99, P = 1.34e-08) and decreased neutrophil count (OR = 0.99, CI = 0.99-1.00, P = 1.38e-02), while increasing eosinophil count (OR = 1.01, CI = 1.01-1.02, P = 5.21e-05). Genetic predisposition to CAD did not have significant association with leukocyte counts.
T1D is considered a lymphocyte-mediated disease, therefore the apparently protective effect of increased lymphocyte counts was unexpected. To probe this finding further, we expanded our analyses to other related autoimmune diseases, including multiple sclerosis (MS) and rheumatoid arthritis (RA). Unlike T1D, genetic upregulation of lymphocyte count was associated with increased odds of MS (OR = 1.22, CI = 1.09-1.37, P = 7.61e-04) and no significant association with RA (OR = 0.97, CI = 0.85-1.11, P = 6.59e-01; Suppl Data 2).
Discussion
This study demonstrates that T1D and CAD share a positive genetic architecture and that both diseases share significant genetic determinants with eosinophil and lymphocyte counts at the genome-wide level. Importantly, not only do T1D and CAD share genetic determinants with eosinophil and lymphocyte counts, but genetically determined counts of these leukocytes as well as neutrophils were also associated with both diseases in the same direction as the genetic correlations. At the locus-specific level, several loci were identified that had positive or negative genetic correlations between T1D and CAD, including loci that overlap SH2B3, CTSH, MORF4L1, CFDP1, CTRB1, CTRB2, and IFIH1. Cis-genetic regulation of the expression levels of these genes were also associated both diseases as well as leukocyte counts. GWAS had previously associated CTSH, CTRB1, CTRB2, and IFIH1 with T1D [9, 11], and CFDP1 and MORF4L1 with CAD risk [23]. By analyzing summary level data, this study demonstrates that the loci that harbor these genes are shared between T1D and CAD and that genetic regulation of the expression levels of these is associated with both diseases. Overall, these findings suggest that shared genetic basis between T1D and CAD that affect leukocyte counts may link the two diseases. While genetically determined expression levels of CTSH, MORF4L1, and CFDP1 had opposing associations with T1D and CAD, that for SH2B3, CTRB1, CTRB2, and IFIH1 had the same direction of effect on both diseases and could represent predictive markers or targets for both diseases. Deleterious coding variants in CTSH were associated with reduced risk of T1D [9], which is in line with the results in this study that genetic upregulation of CTSH expression levels in blood and pancreas increases T1D risk. This is also similar in the case of SH2B3, wherein GWAS SH2B3 loss-of-function variants have been associated with increased T1D risk [9, 11] and atherosclerosis in murine models [35], while our analysis showed that genetic upregulation of SH2B3 expression protects against T1D and CAD. Variants in IFIH1 have been associated with many autoimmune diseases. In particular, loss-of-function variants have been associated with protection against T1D whereas gain-of-function variants increase T1D risk [11, 33]. Interestingly, not only did we find that cis-eQTL-mediated increased expression of IFIH1 levels increase T1D and CAD risk, but also that GWAS variants in IFIH1 that increased T1D risk also increased CAD risk while altering leukocyte counts. MR analysis that used cis-protein quantitative trait loci (cis-pQTL) as instruments associated genetic upregulation of chymotrypsinogen B1 (encoded by CTRB1) with reduced T1D risk [36]. Our studies using cis-eQTLs as instruments also showed that increased expression of CTRB1 against T1D as well as CAD. Unlike CTRB1, genetic upregulation of CTRB2 showed the opposite association with T1D and CAD, increasing the risk of both diseases.
Given that T1D is largely considered a lymphocyte-mediated disease, it was counterintuitive that T1D showed a negative genome-wide genetic correlation with lymphocyte count. Interestingly, genetically determined lymphocyte count had an inverse association with T1D risk, and vice versa, which is consistent the genetic correlations. T1D risk variants were shown to have a predominant enrichment in lymphoid enhancers, but it was unknown whether this relates to lymphocyte counts [10]. In addition to a genome-wide genetic correlation, several T1D risk loci, including PTPN22, IL2RA, HORMAD2, KLRG1/CD69, and ZBTB9/TCP11 showed strong local genetic correlations with lymphocyte counts. Similar to lymphocyte counts, neutrophil count had an inverse association with T1D risk and displayed a negative genome-wide genetic correlation with T1D, albeit statistically insignificant. Clinical observations have documented reduced lymphocyte and neutrophil counts in patients with of T1D [16–19, 37]. Our findings thus suggest that these altered counts could contribute to disease.
Eosinophil count displayed significant genome-wide positive genetic correlation with both T1D and CAD and genetic upregulation of eosinophil count was positively associated with both diseases. Observational studies have associated eosinophil counts with CAD [38, 39] and shown that eosinophils from patients with T1D express high levels of myeloid alpha defensins and myelopoxidase, which could be pathogenic [40]. Thus, eosinophils may play a causative role in both T1D and CAD. In line with this proposition, we identified several T1D and/or CAD risk loci that showed significant local genetic correlations with eosinophil count, including CLEC16A, IKZF4/SOUX, BLTP1/ADAD1, BACH2, ATXN2, and SMAD3.
In summary, this study identifies shared genetic basis between T1D and CAD and implicate genetic regulation of leukocyte counts and gene expression levels as potential contributors to both diseases and their co-occurrence. A potential limitation to the findings in this study relates to potential confounders in MR analysis, such as genetic pleiotropy, which makes it problematic to conclude that the outcome is entirely due to the effect of the instruments on the exposure. Experimental perturbation of the shared genes between T1D and CAD in hematopoietic cells would also enable a better understanding of the genetic mechanisms that contribute to the co-occurrence of the two diseases through leukocytes.
Data Availability
The source of publicly available data used for analyses in this study are described in the manuscript. All data produced in the present study are available upon reasonable request to the authors.
Conflict of interest
All authors on this manuscript declare no conflict of interest.
SUPPLEMENTARY TABLES AND FIGURES
Suppl Data 1: Local genetic analyses by ρ-HESS. (Sheet 1) Genome-wide agnostic scan for significant local genetic correlations. Local heritability and genetic correlation estimates are shown for LD-independent blocks and pairs of traits with nominally significant local heritability and genetic correlation estimates. Estimates in red correspond to cases where the local genetic correlation p value exceeds a Bonferroni threshold for the number of LD-independent blocks in analysis (i.e. 0.05/1453). We recommend individually interpreting these Bonferroni-significant loci, and using the larger list for downstream bioinformatic analyses where a higher false discovery rate is tolerable. (Sheet 2). Local genetic correlation analysis at preselected gene loci. Local heritability and genetic correlation estimate at each gene locus are shown for each pair of traits for the LD-independent block containing the gene, with corresponding p values (Bonferroni significance 0.05/36). Genetic correlation values are set to NA in cases where there is not nominally significant heritability for at least one of the two traits. SNPs with lowest p value per trait are also shown.
Suppl Data 2. Mendelian randomization estimates of association between cis-genetic regulation of the expression levels of shared genes between T1D and CAD on both diseases, and bi-directional MR estimates of the associations between genetically determined leukocyte counts and T1D and CAD.
ACKNOWLEDGEMENTS
These studies were supported by JDRF (2-SRA-2022-1276-S-B).
Abbreviations
- T1D
- Type 1 diabetes
- CAD
- Coronary artery disease
- GWAS
- Genome-wide association studies
- HSPCs
- Hematopoietic stem and progenitor cells
- LD
- Linkage disequilibrium
- LDSC
- Linkage disequilibrium score regression
- p-HESS
- Heritability estimation from summary statistics
- eQTL
- Expression quantitative trait loci
- BAS
- Basophils
- EOS
- Eosinophils
- LYM
- Lymphocytes
- Mon
- Monocytes
- NEU
- Neutrophils
- pQTL
- Protein quantitative trait loci
REFERENCES
