Narcolepsy risk loci are enriched in immune cells and suggest autoimmune modulation of the T cell receptor repertoire

Type 1 narcolepsy (T1N) is a neurological condition, in which the death of hypocretin-producing neurons in the lateral hypothalamus leads to excessive daytime sleepiness and symptoms of abnormal Rapid Eye Movement (REM) sleep. Known triggers for narcolepsy are influenza-A infection and associated immunization during the 2009 H1N1 influenza pandemic. Here, we genotyped all remaining consented narcolepsy cases worldwide and assembled this with the existing genotyped individuals. We used this multi-ethnic sample in genome wide association study (GWAS) to dissect disease mechanisms and interactions with environmental triggers (5,339 cases and 20,518 controls). Overall, we found significant associations with HLA (2 GWA significant subloci) and 11 other loci. Six of these other loci have been previously reported (TRA, TRB, CTSH, IFNAR1, ZNF365 and P2RY11) and five are new (PRF1, CD207, SIRPG, IL27 and ZFAND2A). Strikingly, in vaccination-related cases GWA significant effects were found in HLA, TRA, and in a novel variant near SIRPB1. Furthermore, IFNAR1 associated polymorphisms regulated dendritic cell response to influenza-A infection in vitro (p-value =1.92*10−25). A partitioned heritability analysis indicated specific enrichment of functional elements active in cytotoxic and helper T cells. Furthermore, functional analysis showed the genetic variants in TRA and TRB loci act as remarkable strong chain usage QTLs for TRAJ*24 (p-value = 0.0017), TRAJ*28 (p-value = 1.36*10−10) and TRBV*4-2 (p-value = 3.71*10−117). This was further validated in TCR sequencing of 60 narcolepsy cases and 60 DQB1*06:02 positive controls, where chain usage effects were further accentuated. Together these findings show that the autoimmune component in narcolepsy is defined by antigen presentation, mediated through specific T cell receptor chains, and modulated by influenza-A as a critical trigger.


Main Text
examined using LD Score Regression 33 , the shared heritability was largest with type-1 diabetes 1 (T1D) (r g =0.3261 (se=0.1015), p-value = 0.0013).  We also discovered strong overlap between T1N and T1D at CTSH pp=0.998 and SIRPG 11 pp=0.999, as well as evidence for partial sharing at IL27 pp=0.71, while signals were independent 12 for P2RY11 (pp=0.02). T1D is also the only autoimmune trait besides narcolepsy where any 13 association was seen near the TRA locus, although the T1D signal (rs7145202, beta = 0.1, p-14 value = 4*10 -6 ) 41 is independent from the narcolepsy signal (r 2 <0.5) and located ~100 kb all cohorts) and with combined vaccination related narcolepsy sample (p-value = 7.96*10 -10 ). 1 ( Table 2, and Supplementary Table 8  Similarly to GRS evidenced shared signal, we found GWA significant signal with HLA-4 DQB1*06:02, TRA rs1154155 and a variant between SIRPB1-SIRPG locus (rs76958425, OR= 5 2.49 [1.82 -3.41], p-value = 1.12*10-8, Table 2) not present in regular cases (rs76958425, p-6 value=0.15, beta = -0.0694, OR=0.93). The overall association of GRS and two shared loci 7 indicate that vaccination related narcolepsy is fundamentally the same disorder as idiopathic T1N. IL27 (rs181206 L119P) as well as variants marking different HLA-alleles. 16 17 We confirmed that variants within CTSH are also important in the predisposition of T1N. Among 18 immune cells, CTSH is only expressed in Class II positive antigen presenting cells (B cells, 19 dendritic cells and monocytes), and is known to process antigen for HLA presentation, thus 20 furthering a role for HLA-DQ presentation in T1N. Of note, we also observed a sub threshold  In addition, we discovered associations is in signal-regulatory protein gamma SIRPG (rs6110697, 6 V263A) a receptor-type transmembrane glycoprotein known to interact with CD47, an anti-7 autophagy signal for the immune system that has shown success in cancer immunotherapy 59 . 8 Although V263 is conserved in all SIRP family members, it is also located within an alternate 9 exon. Unlike other members of the SIRP family, SIRPG is almost exclusively expressed in CD4 + 10 and CD8 + T cells. Furthermore, the SNP is also a strong eQTL in thymus and whole blood 60 .

11
Interestingly, vaccination-associated cases displayed an additional GWAS significant association Overlap of risk with cell-type specific chromatin regions. We examined whether associations 4 with narcolepsy were enriched genome-wide on specific enhancer elements using stratified LD 5 score regression on Epigenome Roadmap cell type specific annotations (n=216 cell types) 71 .

6
Partitioned heritability by functional categories enriched in the hematopoietic cell lines 7 (Supplementary Fig. 2b and 2c, Supplementary Fig 8.). Consistent with our model, association  Fig. 3). Furthermore, the effect was stronger in individuals with T1N that 1 had significantly lower expression level of TRA J28 than healthy controls (beta = -0.20, p-value = 2 0.027). Similarly, the effect of rs1154155 on J24 usage was also similar population cohort (beta = 3 0.33, p-value<0.001). We also confirmed that these effects were cis mediated, and the ratio of   suggests specificity for the autoimmune pathology through the T cell receptors. The co-2 localization of signal at the population sample with expression suggests a direct effect on the 3 specific usage of TRAJ28 expression coding effect on TRAJ24 (F8L) variation as well as TRBV4-4 2 gene expression. This was also is seen specifically in T cell receptor sequencing in CD4+ T 5 cells and is stronger in patients (p<0.05) suggesting for direct causal effect for disease 6 pathophysiology through expression and autoantigen recognition. fine map this association, we imputed HLA haplotypes using HIBAG 76 and HLA IMP:02 77 . We 12 then performed ethnic specific HLA association and combined them using fixed effects meta-13 analysis. As expected 5,6 , the strongest association was with the DQA1*01:02~DQB1*06:02 14 (DQ0602) haplotype.

16
To look for additional independent signal, we performed conditional analysis using stepwise 17 forward regression. We detected (1)  In this study, we explored genetic risk for narcolepsy and potential disease mechanisms of 2 identified genetic risk factors. The strongest associations were seen with the HLA region. In 3 addition, we confirmed six previously described risk loci (TRA, TRB, CTSH, IFNAR1, ZNF365 and 4 P2YR11) and discovered five novel associations in PRF1, CD207, SIRPG, IL27 and ZFAND2A.

5
Analysis of functional consequences of these loci in a multi-ethnic sample discovered remarkable 6 association with immune loci evidenced by individual associations and partitioned heritability 7 enrichment. A notable example is the effect of both missense and regulatory variants in the TRA 8 and TRB regions that had a substantial effect on the T cell receptor chain usage. All these 9 findings strongly suggest specific risk factors in genes controlling immune reactions. (HLA-A, PRF1), sketching a remarkably narrow disease pathway (Fig. 4). Accordingly, a direct  This association is unique to T1N and has not to our knowledge been seen with other 26 autoimmune diseases. 27 In addition, a strong functional connection with Influenza A infection in dendritic cells was found at 1 IFNAR1, furthering the role of this virus as a common trigger for the disease. We also discovered  Based on these observations, we propose that narcolepsy is the result of an autoimmune process 16 triggered primarily by influenza-A on an HLA-DQA1*01:02~DQB1*06:02 (DQ0602) background.

17
The involvement of influenza-A is likely to explain why the genetic associations we found are 18 universal. Indeed, influenza is one of few viruses that act worldwide on a seasonal basis. Other autoimmune diseases, unlike narcolepsy, are also associated with a plethora of 4 autoantibodies and known autoantigen targets. For example Insulin, GAD, IA-2 and ZNT8 are 5 involved in T1D and β-cell antigen targeting, suggest that these other diseases involve multiple B 6 and T cell mechanisms and antigens, likely explaining the weaker and more complex HLA effects 7 and a lack of association with any specific TCR polymorphisms. It is our hypothesis that the 8 strong effects of TCR polymorphisms in narcolepsy likely represent the fact autoimmunity in this 9 disease is oligoclonal and limited to one or a few hypocretin cell antigen epitopes. These epitopes 10 may bind DQ0602 specifically and involve a few αβTCR receptors containing TRAJ24, TRAJ28 or 11 TRBV4-2 (Fig 4). Although it is conceivable NK cells could be involved, the most likely explanation is involvement 2 of CD8 + T cell in hypocretin cell killing in collaboration with CD4 + T cells or microglia. This was 3 also supported by CTSC association, an enzyme of critical importance to cytotoxic CD8 detectable autoantibodies has made objective demonstration of autoimmunity challenging, but will 23 likely made the eventual discovery of the culprit T cell antigen even more informative to our 24 understanding of T cell immunity in the brain. and Swedish vaccination related cases and European Narcolepsy Network samples, which were 5 analyzed by respective study teams using exactly the same analysis. Genome-wide association 6 analysis was first performed in each case control group separately using SNPTEST v.2.5.2 99 . We 7 used linear regression implemented in SNPTEST method score adjusting for ten first principal association was set to genome-wide significance (p-value<5*10 -8 ) controlling for multiple testing. 16 Overall test statistics showed no genomic inflation. GCTA was used for heritability and gene 17 based tests 101 . Coloc analysis was done using coloc package in R version 3.4.2 (2017-09-28) 40 , 18 Manhattan and QQ-plots were created with QQman or FUMA 97 . Shared heritability was 19 estimated using LD score regression 32 . Analysis of HLA variants: HLA effects in narcolepsy were analyzed as described before 6 . We V and J usage estimated from total peripheral blood RNA sequencing as described before 73 .

Methods
in T cells from 60 individuals with narcolepsy and 60 healthy individuals from using total CD4+ T 1 cells, CD4+ T memory and CD8+ T cell populations. We used fastqc to infer quality and trimmed 2 low quality reads. We then performed barcode demultiplexing, after which local blast was used to 3 align and extract CDR3s. Linear regression was fit for TRA usage ~ Genotype adjusting for age 4 and gender, RNA sequencing lane and case/control status as covariates. We also analyzed 5 separately coding consequences for each TRAJ24 containing productive CDR3 fragment as one 6 of the most significantly associating SNPs was a coding SNP (rs1483979) was changing an