Estimation of pleiotropic enrichment
In the stratified quantile-quantile plots for ALS conditional on association P values with each obesity-related trait, successive enrichment was found for BMI, BFP, HDL-C, LDL-C and T2D (Figure 1), indicating the proportion of non-null SNPs in ALS increase with higher levels of association with these traits. In contrast, minimal or no enrichment was found for WHR, BW, TC, TG, FG and FI. For progressively stringent P value thresholds, we could observe approximately 13-fold enrichment conditional on BMI, 13-fold enrichment on BFP, 10-fold enrichment on LDL-C, 8-fold enrichment on HDL-C and 8-fold enrichment on T2D, while minimal enrichment on the other traits, suggesting selective genetic overlap between ALS and obesity-related traits (Figure 2).
ALS-associated loci identified with conditional FDR
To discover genetic variants associated with ALS, we performed the conditional FDR statistical analysis. A total of 20 risk loci were identified with conditional FDR < 0.01 (SupplementaryTable 2), including 14 novel loci which were not significant (P<5E-08) in the original ALS GWAS[6]. In these 20 loci, 11 were suggestively significant (P<1E-04) in the replication GWAS results, namely rs10463311 (GPX3,TNIP1), rs7813314 (LOC101927815), rs7864502 (C9orf72), rs58854276 (ACSL5), rs12810996 (TBK1), rs447614 (G2E3), rs229150 (SCFD1), rs978220 (ATXN3), rs35714695 (SARM1), rs2285642 (GGNBP2) and rs12608932 (UNC13A) (SupplementaryTable 2). Among these replicated genes, GPX3, TNIP1, C9orf72, TBK1, SCFD1, ATXN3 and UNC13A have been described as risk genes for ALS by earlier GWAS, while the others were novel risk genes, including LOC101927815, ACSL5, G2E3, GGNBP2 and SARM1.
Loci shared between ALS and obesity-related traits
To identify shared risk loci between ALS and obesity-related traits, we further performed conjunctional FDR analysis. A total of 9 shared risk loci were identified with conjunctional FDR < 0.05, including rs62333164 (CLCN3), rs170663 (G2E3), rs8018993 (SCFD1), rs11160036 (TRIP11), rs978220 (ATXN3) and rs12603276 (GGNBP2) between ALS and obesity traits BMI/WHR/BFP, rs3849942 (C9orf72) and rs1976704 (PGS1) between ALS and HDL-C/LDL-C, and rs68069258 (DENND6B) between ALS and T2D (Table 1). In these 9 risk loci, 6 were suggestively significant (P<1E-04) in the replication GWAS results, namely rs62333164 (P=3.84E-06), rs3849942 (P=1.20E-22), rs170663 (P=2.22E-06), rs8018993 (P=8.00E-06), rs978220 (P=2.61E-05) and rs12603276 (P=9.21E-06) (Table 1). Among these replicated genes, CLCN3, G2E3, and GGNBP2 were newly discovered risk genes for ALS, while the others have been described as risk genes for ALS by previous GWAS. No shared risk loci were found for ALS conditional on TC, TG, FG, FI and BW, which was consistent with the stratified quantile-quantile plots and fold-enrichment plots with no apparent enrichment observed. Notably, in the 6 shared risk loci identified between ALS and BMI, 5 were in reverse effect direction for ALS and BMI in the original GWAS. This is in line with recent Mendelian randomization results[37] that premorbid higher BMI contributes to decreased ALS risk. In contrast, the effect direction of the two shared risk SNPs between ALS and HDL-C were not consistent.
Functional interpretation of shared risk loci
To determine the functional effects of the shared risk loci, we evaluated cis-eQTL in human brains free of neuropathologic characteristics. The pleiotropic risk SNPs in GGNBP2 were associated with expression of MYO19 (P=3.10E-04) and GGNBP2 (P=8.40E-12), both of which were risk genes associated with BMI. Pleiotropic SNPs in C9orf72 affect expression of TEK (P=3.40E-05), which is also associated with blood protein level. Pleiotropic SNPs in ATXN3 were related to expression of SLC24A4 (P=1.10E-03) and ATXN3 (P=1.10E-06), particularly in cerebellum. Interestingly, SLC24A4 is risk gene for Alzheimer disease and ATXN3 is risk gene for ALS, suggesting this region might have some common effect in neurodegenerative diseases. In addition, pleiotropic SNPs in PGS1 affect expression of PGS1 (P=6.60E-11) and USP36 (P=2.00E-04) (Table 2).
To determine the biological pathways represented by shared risk genes and the genes identified with eQTL analysis, we conducted pathway overrepresentation analysis. Two pathways were enriched, namely membrane trafficking (P=0.004) and vesicle-mediated transport (P=0.005), both of which have been identified as involved in the pathogenesis of ALS. Membrane trafficking has been implicated in virtually every aspect of neuronal function and, in particular, neuronal maintenance and degeneration[38], and intracellular membrane trafficking defects affecting key neuronal functions may be an early determinant of motor neuron loss in ALS[39]. Meanwhile, the molecular regulation of intracellular and extracellular vesicle trafficking is an important pathway in ALS pathogenesis[40], and mutation in the vesicle-trafficking protein has been implicated to cause late-onset spinal muscular atrophy and amyotrophic lateral sclerosis. Additionally, 7 GO sets were identified, most of which are related to the membrane trafficking and cytoskeleton (SupplementaryTable 3).