Genome-wide study of DNA methylation shows alterations in metabolic, inflammatory, and cholesterol pathways in ALS

Abstract

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease with an estimated heritability between 40 and 50%. DNA methylation patterns can serve as proxies of (past) exposures and disease progression, as well as providing a potential mechanism that mediates genetic or environmental risk. Here, we present a blood-based epigenome-wide association study meta-analysis in 9706 samples passing stringent quality control (6763 patients, 2943 controls). We identified a total of 45 differentially methylated positions (DMPs) annotated to 42 genes, which are enriched for pathways and traits related to metabolism, cholesterol biosynthesis, and immunity. We then tested 39 DNA methylation-based proxies of putative ALS risk factors and found that high-density lipoprotein cholesterol, body mass index, white blood cell proportions, and alcohol intake were independently associated with ALS. Integration of these results with our latest genome-wide association study showed that cholesterol biosynthesis was potentially causally related to ALS. Last, DNA methylation at several DMPs and blood cell proportion estimates derived from DNA methylation data were associated with survival rate in patients, suggesting that they might represent indicators of underlying disease processes potentially amenable to therapeutic interventions.

Fig. 1.. EWAS meta-analysis.

EWAS on 6763 patients and 2943 controls. (A and C) Manhattan plot comparing (A) LB (linear model + bacon) and (C) OSCA MOA association P values [−log₁₀(P), y axis] and genomic location (x axis). The dashed line indicates the genome-wide significance threshold (9 × 10⁻⁸). Sites were annotated with the nearest protein-coding gene in ensembl [some gene labels in (A) could not be clearly displayed; all labels are presented in fig. S10]. (B and D) Volcano plots showing (B) LB and (D) OSCA MOA estimated effect sizes (x axis) and association P values [−log₁₀(P), y axis]. Ninety-five percent confidence intervals are shown for DMPs, and the nearest genes are shown for the top 10 DMPs identified with the LB algorithm and for all DMPs identified with the MOA algorithm. (E and F) Quantile-quantile plot showing observed (E) LB and (F) OSCA MOA P values [−log₁₀(P), y axis] against the expected distribution under the null (x axis).

Fig. 2.. EWAS database enrichments.

Significant overlap (Fisher’s exact test, FDR < 0.05) between traits included in the MRC-IEU EWAS database and ALS-associated positions identified using the LB model. (A) Network showing the traits that significantly (FDR < 0.05) overlap with the ALS-associated positions. Nodes indicate the overlap between ALS-associated positions and positions associated with indicated traits, with larger nodes indicating more overlap, and lighter shades of blue indicating stronger associations. Edges indicate probe overlap between the traits, with thicker lines indicating more overlapping probes. Colored surfaces indicate the clusters (cholesterol, metabolic, and inflammatory) identified using the Louvain clustering algorithm. (B) Identification of independent clusters of traits. The first iteration shows the traits that significantly overlap with the ALS-associated probes at FDR < 0.05. In subsequent iterations, the probes belonging to the trait with the lowest-enrichment P value were excluded, and enrichment tests were performed using the remaining traits. This algorithm was repeated, retaining traits that were nominally significant (P < 0.05, indicated in bold), until at most one trait remained nominally significant. At the third iteration, no traits remained nominally significant (P < 0.05), showing that both BMI and related traits (including triglycerides and HDL-c) and IgE and related traits (atopy) show independent overlap with the ALS-associated positions. IgE, total serum IgE; TG, triglycerides; sTG, serum triglycerides; WC, waist circumference; sHDL-c, serum HDL-c: HW, hypertriglyceridemic waist; FG, fasting glucose; AF, atrial fibrillation; BMIc, BMI change; PL, postprandial lipemia; GGT, gamma-glutamyl transferase; fINS, fasting insulin; AC, alcohol consumption per day; 2hINS, 2-hour insulin; ATP, atopy; sIgE, high serum IgE; pAN, plasma adiponectin; T2D, type 2 diabetes; CKD, chronic kidney disease; HOMA-IR, homeostatic Model Assessment of Insulin Resistance.

Fig. 3.. Polymethylation score analyses on disease risk and patient survival.

Polymethylation scores (PMSs) were determined as proxies for various traits, exposures, proteins, and WBC proportions, calculated as weighted sums based on probes and weights derived from published papers, respectively. Case-control association analyses were performed on 6763 patients and 2943 controls; survival analyses were performed within 5162 patients. (A) Explained variance of PMSs calculated in samples for which both DNA methylation data and biomarker/clinical data were available (N = 800 of 2000). Reduced R² represents the variance explained by the null model, whereas the incremental R² represents the additional variance explained by the PMS over the null model. Last, the explained variance of the univariate model of the respective PMS is displayed (see Materials and Methods). The asterisk indicates that the PMS was used in the association tests. (B and C) The top panel shows association P values from logistic regression [−log₁₀(P), y axis] for each PMS (x axis). (B) WBC proportions and (C) various traits and exposures, colored by whether a higher score is associated with increased (black) or decreased (gray) disease risk. The bottom panel shows the Cox PH P values [−log₁₀(P), y axis] for each PMS (x axis), colored by whether a higher score is associated with decreased (black) or increased (gray) survival, respectively. The dashed line indicates the significance threshold (1.3 × 10⁻³). (D) Original P values [−log₁₀(P), x axis] compared to P values after including all PMSs as fixed covariates in the logistic regression model [−log₁₀(P), y axis] for the ALS-associated traits/exposures. (E and F) Association P values [−log₁₀(P), y axis] upon incrementally adding principal components (PCs) as fixed covariates in the logistic regression model. HGF, hepatocyte growth factor; EN.RAGE, extracellular newly identified RAGE-binding protein; GDF8, growth/differentiation factor 8; OSM, Oncostatin-M, SKR3, Serine/threonine-protein kinase receptor R3; TNFSF14, tumor necrosis factor ligand superfamily member 14; VEGFA, vascular endothelial growth factor A; nPCs, number of principal components.