Leveraging data from large-scale GWAS and multiple quantitative trait locus datasets, we investigated the genetic evidence for the efficacy of seven candidate drugs in prevention of type 1 diabetes. Using co-localisation and Mendelian randomisation, we found genetic evidence to support the role of IL-2 and IL-6 signalling in the pathogenesis of type 1 diabetes. In addition, the investigation of functional missense variants suggested that TYK2 signalling is involved in the aetiology of type 1 diabetes.

While the original GWAS of immune cell subsets did not report the eQTL of IL2RA in regulatory T cells (Tregs), our evidence for the protective effect of blood IL2RA expression on the risk of type 1 diabetes could be interpreted as supporting the role of Tregs as a natural protection against type 1 diabetes. A low but sufficient level of IL-2 is crucial for the survival and function of Tregs, which constitutively express IL2RA, IL2RB and IL2RG to produce α-, β- and γ-chains, respectively, required for trimeric high-affinity IL-2 receptors [17]. Naive T cells express IL2RB and IL2RG, required for the intermediate-affinity IL-2-receptors, but they express IL2RA only transiently when stimulated by antigen-presenting cells and are thus less stimulated by IL-2 when not activated. Thus, higher blood IL2RA expression could be a sign of increased quantity and function of Tregs, which maintain a level of tolerance towards self-peptides and decrease the risk of type 1 diabetes [18]. A previous small study showed that rs12722495, which is in strong LD (r2=0.89) with our lead IL2RA eQTL and type 1 diabetes risk locus rs61839660 near IL2RA, decreased IL2RA expression in Tregs as well as their sensitivity to IL-2 [19].

Alternatively, as IL-2 signalling increases the proliferation of the conventional T cells and Tregs alike, IL-2 signalling might not decrease the risk of type 1 diabetes only by increasing tolerance but also by promoting appropriate responses to pathogens. This is supported by our single-cell-level results, in which we observed evidence for co-localisation between IL2RA expression and the risk of type 1 diabetes only in CD8+ central memory T cells (eQTL of Tregs were not available as they were not distinguished from other T cells). In a birth cohort study of children at high genetic risk of type 1 diabetes (TEDDY study), presence of enteroviral DNA in stool was associated with the risk of islet autoimmunity and the association was stronger in persistent infections indicated by prolonged shedding of enteroviral DNA [20]. Likewise, children who developed islet autoimmunity presented longitudinal transcriptional signatures consistent with a less-robust immune response against enteroviral infections compared with matched control children [21]. Since the increased IL-2 signalling in CD8+ T cells during viral infections prioritises robust immune response against production of long-lived memory cells, this explanation might also explain why rs61839660 near IL2RA strongly co-localised between the risk of type 1 diabetes and the eQTL of IL2RA in effector memory T cells and CD8+ naive/central memory T cells.

The co-localisation of IL2RA expression in CD8+ effector memory T cells with type 1 diabetes risk is further supported by the previous reports of IL-2 impairment leading to CD8+ T cell exhaustion, potentially driven by both acute and chronic viral infections [22, 23]. Interestingly, our lead rs61839660 near IL2RA, which was associated with increased IL2RA expression and decreased risk of type 1 diabetes, was previously shown to be associated with higher risk of Crohn’s disease [24] and lower risk of type 1 diabetes [25]. This suggests that the optimal balance between effector and regulatory T cell function may vary between autoimmune diseases [17]. Regardless of the possible mechanism, our findings support the rationale of conducting type 1 diabetes prevention trials with low-dose IL-2.

Consistent with our finding of the protective effect of IL-2 signalling, we found that IL-6 signalling increased the risk of type 1 diabetes. The secretion of IL-6 from macrophages rapidly in response to infections and tissue damage promotes various acute phase responses [26]. IL-6 signalling occurs as classic signalling through a cell-membrane-bound IL-6R, trans-signalling through soluble circulating IL6R and membrane-bound gp130 (encoded by IL6ST) and trans-presentation by dendritic cells, in which IL-6 is presented to membrane-bound gp130 in T cells via dendritic cell-membrane-bound IL-6R (ESM Fig. 3) [27]. IL-6 inhibits the development and function Tregs and promotes the development of pathogenic T helper 17 (Th17) cells [28], which is inhibited by IL-2. Pronounced IL-6 signalling may alter the balance of Treg/Th17, a proposed causative factor in autoimmune diseases such as rheumatoid arthritis [28] and possibly also type 1 diabetes [29].

While, IL-6 trans-signalling and trans-presentation may be a more potent inducer of autoimmunity than classic signalling [12], our finding that blood IL6R and IL6ST expression are associated with increased risk of type 1 diabetes may be explained by any of the three signalling modalities. However, since IL6ST expression in CD4+ and CD8+ naive or central memory T cells co-localised with type 1 diabetes and IL6R expression did not, it is tempting to speculate that trans-signalling might be more important than classic IL-6 signalling in the pathogenesis of type 1 diabetes.

In contrast to our findings, tocilizumab (a monoclonal antibody against IL6R), which blocks all three IL-6 signalling modalities, did not significantly affect the decline in residual beta cell function in individuals with newly diagnosed type 1 diabetes in a randomised, placebo-controlled, double-blind clinical trial [30]. However, this discrepancy may be partially explained by the timing of the intervention. Genetic polymorphisms typically exert life-long influence on risk of diseases, including every stage of type 1 diabetes, whereas the intervention in the study by Greenbaum et al [30] took place after the diagnosis of type 1 diabetes, at which stage a sharp fall in beta cell function has already taken place [31].

Whole-blood TYK2 expression and the risk of type 1 diabetes did not co-localise, whereas when using the missense mutation rs2304256 in TYK2 as an instrument in Mendelian randomisation, TYK2 expression was associated with type 1 diabetes risk. The absence of evidence for co-localisation may reflect violations of the one-causal-variant assumption. Indeed, the missense mutation rs2304256 is in very high LD (r2=1.00) with the lead TYK2 eQTL rs34725611. Moreover, rs2304256 is only in moderate LD (r2=0.10) with rs144309607, the lead variant on type 1 diabetes liability in co-localisation, suggesting two independent signals. Of note, a known missense variant rs34536443 is not available in the eQTL data and therefore could not be used in co-localisation. Despite the one-causal-variant assumption, ‘coloc’ is relatively robust to multiple causal variants, and co-localisation methods allowing for multiple causal variants are highly sensitive to LD misspecifications in the reference panel [6]. Therefore, in the absence of an accurate LD reference, we proceeded with the ‘coloc’ method for our co-localisation while acknowledging its limitations.

Overall, our TYK2 findings are consistent with previous studies suggesting that TYK2 signalling is associated with the risk of type 1 diabetes [32]. Promising results have been found in clinical trials targeting TYK2 in autoimmune diseases, supporting the potential of drug repurposing in type 1 diabetes [33]. Other reasons for the co-localisation discordance could be that TYK2 is not activated until IFN-α binding to IFNAR1 and TYK2 RNA expressions are known to have low tissue and cell type specificity [34, 35].

It is important to note that Mendelian randomisation estimates are only valid if the instrumental variable associations (relevance, independence, exclusion restriction) are met. The strong associations between the assessed genetic polymorphisms and the studied exposures suggested that the genetic instruments studied were relevant for the studied exposures. The strategy of selecting instruments from within the cis-region of the exposure of interest is an established approach for investigating drug effects [36]. cis-Mendelian randomisation studies are by design less prone to horizontal pleiotropy and exposure misspecification (which may lead to violations of independence and exclusion restriction assumptions), as genetic variants typically exert the strongest influence on nearby genes and therefore most effects are secondary to reading of the nearby genes. However, restricting the instruments to cis-variants comes at the expense of potentially missing strong trans-variants that associate with the exposure. Furthermore, our co-localisation results suggest that associations between whole-blood IL2RA and IL6R expression and the risk of type 1 diabetes are unlikely to be caused by LD with a genetic variant that primarily influences the reading of other genes in the vicinity of IL2RA or IL6R loci.

Some aspects of generalisability of our results are also worth mentioning. While our study did not explicitly exclude individuals from non-European ancestries, the original GWAS studies primarily included individuals of European descent, which limits the spectrum of rare variants and the generalisability of our findings to other ancestries. Furthermore, our Mendelian randomisation estimates represent the influence of small changes in IL2RA, IL6R and TYK2 expression during the entire life course before the diagnosis of type 1 diabetes. Therefore, these effect sizes cannot be directly extrapolated to clinical trials in which the doses are larger and exposures shorter, and possibly outside a key sensitive window for disease development. Thus, natural history studies and clinical prevention trials should pinpoint the optimal stage of pathogenesis at which to interfere with IL-2, IL-6 or TYK2 signalling to prevent type 1 diabetes. Finally, even if up to 50% of variability in genetic risk of type 1 diabetes is attributable to HLA-II locus [37], we could not analyse the interactions between SNPs reported here and the HLA genotype or genetic risk scores on risk of type 1 diabetes, or any sex-specific effects, as we did not have access to individual-level data.

In conclusion, our results provide genetic evidence that IL-2, IL-6 and TYK2 signalling are associated with type 1 diabetes risk. Our findings suggest that clinical trials investigating the efficacy of drugs such as tocilizumab (IL-6R antagonist that targets all IL-6 signalling modalities), olamkicept (soluble gp130Fc that blocks IL-6 trans-signalling) and low-dose aldesleukin (IL-2 analogue) may be promising candidates for the prevention of type 1 diabetes.