Shared polygenic risk and causal inferences in amyotrophic lateral sclerosis

Abstract

Objective: To identify shared polygenic risk and causal associations in amyotrophic lateral sclerosis (ALS).

Methods: Linkage disequilibrium score regression and Mendelian randomization were applied in a large-scale, data-driven manner to explore genetic correlations and causal relationships between >700 phenotypic traits and ALS. Exposures consisted of publicly available genome-wide association studies (GWASes) summary statistics from MR Base and LD-hub. The outcome data came from the recently published ALS GWAS involving 20,806 cases and 59,804 controls. Multivariate analyses, genetic risk profiling, and Bayesian colocalization analyses were also performed.

Results: We have shown, by linkage disequilibrium score regression, that ALS shares polygenic risk genetic factors with a number of traits and conditions, including positive correlations with smoking status and moderate levels of physical activity, and negative correlations with higher cognitive performance, higher educational attainment, and light levels of physical activity. Using Mendelian randomization, we found evidence that hyperlipidemia is a causal risk factor for ALS and localized putative functional signals within loci of interest.

Interpretation: Here, we have developed a public resource (https://lng-nia.shinyapps.io/mrshiny) which we hope will become a valuable tool for the ALS community, and that will be expanded and updated as new data become available. Shared polygenic risk exists between ALS and educational attainment, physical activity, smoking, and tenseness/restlessness. We also found evidence that elevated low-desnity lipoprotein cholesterol is a causal risk factor for ALS. Future randomized controlled trials should be considered as a proof of causality. Ann Neurol 2019;85:470-481.

Figure 1

Flow chart of analysis. The ALS Research Resource is an interactive tool where the user can explore genetic correlations and causal associations across more than 700 traits. GWAS exposures used for LD score regressions are available in LD hub at http://ldsc.broadinstitute.org/ldhub/. GWAS exposures used for the Mendelian randomization analyses are available in MR Base at http://www.mrbase.org/. (A) The inclusion criteria used for LD score regression analyses comprise traits with heritability estimates within normal boundaries. (B) The inclusion criteria used for Mendelian randomization includes (1) GWAS with at least two associated SNPs with p values <5.0 × 10^–8; (2) SNPs present in both the exposure and outcome (ALS) data sets or when not present their linkage‐disequilibrium (LD) proxies (R² value > = 0.8); and (3) independent SNPs (R² < 0.001 with any other associated SNP within 10 Mb), considered as the most stringent clumping threshold used when performing MR analyses. (C) LD score regression analyses included 751 publicly available GWASes considered as exposures of interest versus the most recent ALS GWAS as an outcome, and (D) MR analyses were performed considering two phases. Phase I includes 345 available GWASes in the public domain as exposures of interest while phase II includes unpublished UK Biobank GWAS data. (E) Significantly associated GWASes with ALS at inverse variance weighted (p < 0.05). (F) Significantly associated GWASes with ALS at weighted median and MR Egger (p < 0.05). (G) Causally linked GWASes with ALS after performing reverse causality, sensitivity, and directionality analyses. (H) Multivariate analyses used to explore how each related exposure of interest independently contributes to ALS. ALS = amyotrophic lateral sclerosis; GWAS = genome‐wide association study; kb = kilobases; R² = clumping threshold; LDL = low‐density lipoprotein.

Figure 2

Bayesian colocalization plots. A plot and B plot represent two independent LDL‐cholesterol–associated regions with posterior probability greater than 95% of sharing a causal variant involved in ALS. Panels in column A show the region spanning chr5:73656720‐75651786 where rs182826525 is likely the shared causal variant with a posterior probability of nearly 100%. Panels in column B show the region spanning chr5:155390511‐157388284 where rs116226146 is likely the shared causal variant with a posterior probability of 96%. The first row displays the p values from the LDL GWAS for each region. Color is coded by p values in the ALS GWAS. The second row displays the p values from the ALS GWAS for the same regions. Color is coded by p values in the LDL GWAS. The third row shows local gene positions (with strands denoted by ±), as well as recombination rates measured in cM/Mb.38 The bottom row shows the posterior probabilities of a shared causal variant between LDL cholesterol and ALS. ALS = amyotrophic lateral sclerosis; GWAS = genome‐wide association study; kb = kilobases; LDL = low‐density lipoprotein.