Multi-omics analyses identify potential epigenetic loci associated with survival in amyotrophic lateral sclerosis across diverse populations

Summary

Background

Amyotrophic lateral sclerosis (ALS) is a severe motor neuron disease, with highly diverse survival time. However, genetic and epigenetic factors influencing ALS survival across diverse populations remain unclear.

Methods

We performed whole-genome sequencing (WGS) and DNA methylome array in blood DNA of patients with ALS. For survival analysis, we used Cox proportional hazards model for genetic variants, DNA methylation (DNAm) of CpG sites or CpG-SNPs in Chinese and Canadian cohorts, followed by meta-analysis. We performed pathway enrichment analysis for candidate genes inferred from DNAm events associated with survival. In paired genome and methylome data, we analysed the effect of the candidate CpG-SNP genotypes on DNAm status.

Findings

Genome-wide cross-population meta-analysis of common variants in 511 patients with ALS showed a suggestive association of CAV1/CAV2 rs117002347 genotypes with survival. Epigenome-wide cross-population meta-analysis in 459 patients revealed that ALS survival was significantly linked to DNAm of 88 CpGs on 40 genes, and highlighted the AMPK and cytoskeleton pathways. Epigenome-wide cross-population meta-analysis of CpG-SNPs in 459 patients identified 8 loci on 4 genes, including BAG6 (cg27014438/rs28732154), which was further validated in another 204 patients with ALS. Moreover, analysis of paired genome/epigenome data (n = 454) indicated that BAG6 rs28732154 genotypes may modulate cg27014438 methylation, which is also a cis-eQTM of BAG6 expression in blood.

Interpretation

Our study identified BAG6 cg27014438 methylation as a potential epigenetic modifier of ALS survival. BAG6 cg27014438 methylation is modulated by rs28732154 genotypes, and linked to BAG6 expression. Our findings extended our understanding of epigenetic modifiers in ALS survival.

Funding

This work was supported by the National Natural Science Foundation of China (82071430, 82371878) (MZ), Shanghai Municipal Natural Science Foundation General Program (22ZR1466400) (MZ), the Fundamental Research Funds for the Central Universities (MZ), the G. Harry Sheppard Memorial Research Fund, and Canadian Consortium on Neurodegeneration in Aging (ER).

Research in context.

Evidence before this study

Identification of survival modifiers is the basis for understanding disease mechanisms and designing drugs. Previous studies have identified genetic and epigenetic factors associated with disease risk and survival in patients with ALS of mainly European ancestry. However, current knowledge of genetic and epigenetic factors associated with ALS survival in Chinese population is limited.

Added value of this study

Our multi-omics study revealed potential epigenetic factors that are associated with ALS survival, which implicated an epigenetic survival-modifying BAG6 locus in both Chinese and Canadian patients.

Implications of all the available evidence

Our findings extended our knowledge of the cross-talk between genetic variants and epigenetic modifications, and indicated a potential ALS survival modifier across diverse populations.

Introduction

Amyotrophic lateral sclerosis (ALS) is a severe disease characterised by degeneration of motor neurons in the brain and spinal cord. Patients with ALS have highly heterogenous phenotypes, such as age of onset and survival time.^1,2 Most patients survive for 2–5 years, but some patients can survive for up to 20 years.³ For instance, patients with ALS carrying the pathogenic C9orf72 G₄C₂ repeat expansion showed diverse disease duration of 0.5–22 years.⁴ Moreover, investigations of identical twin pairs discordant for ALS carrying the same genetic mutations in C9orf72 or SOD1 indicate the involvement of both genetic and epigenetic factors in modifying phenotypes.⁵

Identification of survival modifiers in ALS is the basis for understanding disease mechanisms and designing drugs. However, current knowledge of genetic and epigenetic factors associated with ALS survival are limited to patients with European ancestry. For example, a genome-wide association study (GWAS) indicated that the CAMTA1 variant rs2412208 was associated with ALS survival.⁶ Also, the GRN variant rs34424835 was associated with faster disease progression and shorter survival in ALS.⁷ The UNC13A rs12608932 minor allele is linked to ALS risk⁸ and a shorter survival time in patients with ALS or frontotemporal dementia (FTD).9, 10, 11, 12 Additionally, DNA methylation (DNAm) of 4 CpGs (affecting FKBP5, ATP8B2, SPIDR and DHCR24) were associated with ALS survival in a large ALS cohort with European ancestry (n = 5138).¹³ Genetic and epigenetic survival modifiers were reported in Japanese cohorts,^14,15 but their roles in Chinese patients are largely unknown. Identification of common survival modifiers across diverse populations is important for precision medicine.

Genetic variants have close interactions with epigenetic modifications. For example, allele-specific DNAm, primarily contributed by CpG-SNPs,¹⁶ have been identified within promoter regions, transcription factor binding sites, or DNase I hypersensitive sites,¹⁷ and play a crucial role in regulating gene expression and are involved in schizophrenia risk.¹⁸ We previously reported that rs4970944 genotypes were associated with DNAm status and linked to cerebellar expression of CTSS, suggesting a role for antigen presenting processes in modifying ALS age of onset.¹⁹ However, the link between epigenetic factors and ALS survival remains to be discovered.

In the current study, we aim to dissect the link between genetic/epigenetic variants and ALS survival in a multi-population cohort of patients with ALS. We conducted genome-wide and epigenome-wide survival analyses in Chinese and Canadian patients. We found that a CAV1/CAV2 genetic variant was suggestively linked to ALS survival at a genome-wide level. Furthermore, we identified a BAG6 CpG-SNP showing a significant association between its DNAm level and survival in our multi-population ALS cohort as well as in an independent Chinese ALS cohort.

Methods

Participants

In the discovery stage, we included a total of 558 patients from China (n = 294) and Canada (n = 264) diagnosed with ALS by a neurologist from 2019 to 2022 at Huashan Hospital (Shanghai) or from 2007 to 2019 at Sunnybrook Health Sciences Centre (Toronto) based on the revised El Escorial criteria.²⁰ Most (∼95%) Canadian patients are of White ancestry. Clinical information included sex, age at last assessment, age of disease onset and family history (Supplementary Table S1). Sex information were self-reported by study participants. All patients were free from known pathogenic mutations in C9orf72, SOD1, FUS or TARDBP.

We included 289 patients from China and 222 patients from Canada for whole-genome sequencing (WGS) (Supplementary Table S2). Most of them (249 Canadian and 210 Chinese patients) were also analysed on the EPIC DNAm array (Illumina). Clinical characteristics of the ALS samples for the DNAm study can be found in Supplementary Table S3.

In the replication stage, we included 204 patients with ALS (PUTH-ALS).²¹ The patients were recruited from 2003 to 2013 at Peking University Third Hospital, and diagnosed as ALS according to the revised El Escorial criteria.

Ethics

The study has been reviewed and approved by the Ethics Committees of Huashan Hospital (ID: KY2016-005, KY2022-1061) (Shanghai, China), the University of Toronto (ID: 35402) (Toronto, Canada), Peking University Third Hospital (IRB00006761-M2020461),²¹ and Ministry of Science and Technology of China (2023-CJ0834). Written informed consent was obtained from all participants.

Whole-genome sequencing

Peripheral blood DNA was isolated using the QIAGEN FlexiGene DNA Kit, followed by PCR-free library construction using the Illumina DNA PCR-Free Library Prep kit (Illumina, USA). We performed WGS using the Illumina NovaSeq6000 sequencing platform (Illumina, USA). The raw-fastq files were filtered using Fastp (version 0.23.2). The clean fastq files were aligned to the human reference genome sequence (hg19) using the Burrows-Wheeler-Aligner (BWA, version 0.7.17). We used GATK (version 4.2.2.0) to call SNPs and insertions/deletions (Indels). We used ANNOVAR (version 2019-10-24) for annotations of SNPs/Indels. We screened for rare missense variants (MAF <0.0001) on 44 reported ALS genes as previously reported.²²

DNAm array

We performed bisulfite conversion of gDNA using the EZ DNA Methylation-Lightning kit (Zymo Research), and profiled the DNAm status of ∼850,000 CpG sites using the Infinium Methylation EPIC array (Illumina). The raw data were filtered, quality controlled, and normalised using the R (ChAMP) package.²³ We used β values to estimate the DNAm level at each CpG site based on the ratio of intensities between methylated and unmethylated alleles. We analysed CpG sites that are not influenced by common SNPs with minor allele frequency (MAF) > 0.05, as well as CpG-SNPs (CpGs mapped to common SNPs).

Genome-wide survival analyses

We first extracted common SNPs/Indels from WGS data of patients with ALS (MAF >0.05). To perform the genome-wide association study, we used Hardy–Weinberg Equilibrium (HWE) to filter the variant loci by using the hwe parameter of PLINK (version 1.9) with a threshold of 1 × 10⁻⁶ to exclude loci that significantly deviate from HWE to minimise the interference of typing errors or technical bias. To perform the survival analyses, we used a Cox proportional hazards model (Wald test) with the gwasurvivr R package,²⁴ and analysed the association of each SNP with survival time under an additive model (genotypes coded as 0/1/2), adjusting for sex and principal components (PC1-PC2) (Supplementary Figure S1). Subsequently, we performed a cross-ethnic Meta-analysis using the GWAMA software.²⁵ Firstly, the effect allele orientation was unified between the Chinese and Canadian populations. Then the random-effects model with a DerSimonian and Laird approach was used to combine the effect sizes. The genome-wide significance threshold for Meta-analysis was set as P < 5 × 10⁻⁸.

Survival status was assessed as reported previously²⁶; at the time of last follow-up for 389 patients with ongoing status or time at death for 122 deceased patients. Manhattan plot was used to prioritise significant variants at the genome-wide level (P < 5 × 10⁻⁸). We used the qqman R (v4.2.1) package²⁷ to visualise the Quantile–quantile (QQ) plot and highlight potential confounders.

We used the survminer R (v4.2.1) package²⁸ to draw a Kaplan Meier (KM) curve to visualise survival time in different groups, and obtain the median age of survival, which represents the time when survival probability is 50%. We also used the Log-rank test to assess the difference of survival probability curves across different groups.

Epigenome-wide survival analyses

To measure the effect of every 10% change of DNAm level (β values) on survival time, we classified CpG DNAm into 11 groups with an β value interval of 0.1 (β ranges of 0–0.05, 0.05–0.15, 0.15–0.25, 0.25–0.35, 0.35–0.45, 0.45–0.55, 0.55–0.65, 0.65–0.75, 0.75–0.85, 0.85–0.95, 0.95–1). Additionally, we removed reported age-related CpGs²⁹ to avoid the impact on survival analysis. To assess whether DNAm of CpGs was associated with survival time, we performed survival analyses using Cox proportional hazards model (Wald test) (R survival and survminer packages)²⁸ for the Chinese and Canadian cohorts, adjusted for sex, experimental batch, PCs (PC1-PC10) and cell types (CD4T, CD8T, B cell and NK cell). Subsequently, we performed a cross-ethnic Meta-analyses using the meta R package³⁰ with a random-effects model (DerSimonian and Laird approach). To control for multiple testing errors, analyses were corrected using the Bonferroni correction and Benjamini-Hochberg method (Bonferroni<0.05; FDR<0.05). Manhattan plot was used to prioritise significant variants. We used a Quantile–quantile plot to assess coefficient inflation. All P-values presented were adjusted for sex.

To define DNAm status for candidate CpG sites, we used the surv_cuttpoint function in the R (survival) package to select the optimal cutoff value for DNAm as a continuous variable. We classified DNAm status into high and low groups based on this cutoff value. We used a KM curve to demonstrate the link between DNAm status and survival time, with hazard ratio (HR) and 95% confidence interval (CI).

Evaluation of the association between DNAm of CpG-SNPs and ALS survival

In an independent ALS cohort (PUTH-ALS, n = 204),²¹ we further evaluated the link between DNAm level of candidate loci and ALS survival. DNAm of gDNA samples were profiled using the Illumina 450 k methylome array.

Gene expression analyses of TargetALS data

We extracted BAM files from the TargetALS RNA-seq dataset, including five CNS tissues: frontal cortex (n = 140, 123 ALS and 17 control), motor medial cortex (n = 121, 107 ALS and 14 controls), motor lateral cortex (n = 124, 111 ALS and 13 control), cervical spinal cord (n = 146, 130 ALS and 16 control), and lumbar spinal cord (n = 137, 122 ALS and 15 control). We used Cox proportional regression to analyse the link between BAG6 expression in CNS tissues and ALS survival time, adjusted for sex.

Pathway enrichment analyses

We conducted KEGG pathway and GO biological process enrichment analyses (biological Process) for selected genes using the ‘clusterProfiler’ package (v4.6.2) in R. The default parameters (pvalueCutoff = 0.05, qvalueCutoff = 1) were utilised, and only the top 10 significant pathways are shown.

Cis-eQTM analyses

Amena Keshawarz et al.³¹ performed the eQTM analysis in RNA-seq and methylome data characterised by Infinium HumanMethylation450 or EPIC BeadChip in 2115 Framingham Heart Study (FHS) participants (48% women, age 54 ± 15 years). We then assessed the association between cg27014438 methylation and expression of 110 transcripts within the 250 kb window of cg27014438 by adjusting for age, sex, white blood cell count, blood cell fraction, platelet count, five gene expression PCs, and ten DNAm PCs.³¹ The Bonferroni corrected P-value <0.05 was accepted as statistical significance.

Statistics

The data used for statistical analysis are derived from patients diagnosed as ALS according to the revised El Escorial criteria. To assess the proportional hazards assumption, we used the Schoenfeld residuals test for top candidate SNPs, CpG-SNPs and CpGs linked to ALS survival. We also used the Wilcoxon rank sum test to assess the effect of genotypes of CpG-SNPs on DNAm level. P < 0.05 was considered statistically significant. All statistical analyses were performed in R v4.2.1.

Role of funders

The funders of this study were not involved in study design, sample and data collection, data analyses and interpretation, or manuscript writing.

Results

Clinical characteristics of patients with ALS

A total of 294 Chinese and 264 Canadian patients participated in the current study (Supplementary Table S1). For WGS, we included 289 Chinese and 222 Canadian patients. The median age of last assessment for survival status is 55.75 years (IQR: 45–64 years) in the Chinese cohort and 63 years (IQR: 54–71.9 years) in the Canadian cohort (Supplementary Table S2). We also identified 7 patients carrying rare variants (MAF <0.0001) on reported ALS genes (DCTN1, SETX, SIGMAR1 and OPTN) (Supplementary Table S3). For DNAm array, we included 210 Chinese patients and 249 Canadian patients. The median age of last assessment for survival status is 54.4 years (IQR: 41.6–64.4 years) in the Chinese cohort and 64 years (IQR: 54.3–73 years) in the Canadian cohort (Supplementary Table S4). In both cohorts, we observed more male patients than female as expected.

Genome-wide analyses of genetic variants associated with ALS survival

The workflow of the study can be found in Fig. 1. To investigate the link between common genetic variations and survival in ALS, we performed WGS in a total of 511 mainly sporadic patients with ALS, including 289 Chinese and 222 Canadian patients (Supplementary Table S2). Quality control analyses showed population derived genetic heterogeneity (Supplementary Figure S2). We then performed survival analyses and found no significant correlation between common variants and overall ALS survival in the Chinese cohort after the genome-wide correction (P > 5 × 10⁻⁸, adjusted for sex and PC1-PC2, Cox proportional hazards model) (Supplementary Figure S3). In the Canadian cohort, one variant is significantly associated with ALS survival (rs715064, P < 4.55 × 10⁻⁸, HR = 4.56, 95% CI = 2.64–7.87, adjusted for sex and PC1-PC2, Cox proportional hazards model) (Supplementary Figure S4), which was not located to a known gene. Quantile–quantile plots showed no significant population stratification in either population group (λ = 1.017; λ = 1.036, Supplementary Figure S5).

Fig. 1.

Workflow of the study. The discovery stage includes the genome-wide survival analysis of whole-genome sequencing data from Chinese (CHN, n = 289) and Canadian (CAN, n = 222) patients with ALS, and the epigenome-wide survival analysis of CpG methylation and CpG-SNPs methylation in Chinese (CHN, n = 210) and Canadian (CAN, n = 249) patients with ALS, followed by a meta-analysis. The replication stage includes validation of the link between candidate CpG-SNPs methylation and survival in an independent PUTH-ALS cohort (n = 204); and assessment of the effects of BAG6 rs28732154 genotypes on cg27014438 methylation in a paired dataset of 454 patients.

We then conducted a cross-ethnic meta-analyses for two cohorts by using a random-effects model. A total of 3,300,056 common variant loci were included in the analyses after harmonising the direction of risk allele effects. We found no significant correlation between common variants and overall survival time in ALS at a genome-wide significance level of P < 5 × 10⁻⁸ (Fig. 2, Table 1, Supplementary Table S5). No genomic inflation was observed (λ = 1.015, Fig. 2B). To test the reported survival modifiers, we identified a significant link of CAMTA1 rs2412208 genotypes⁶ with ALS survival in our cohort (P = 0.007, HR = 1.43, 95% CI: 1.10–1.86, n = 511, random-effects model). However, we found no link of ZNF521B rs2275294³² or UNC13A rs12608932^11,33 genotypes with ALS survival (P > 0.05). That is probably due to the limited sample size in the current study compared to previous ones.

Fig. 2.

Genome-wide association between common SNPs/Indels and disease survival in the cross-population Meta-analysis of patients with ALS. (A) Manhattan plot showing the association between 3,300,056 common SNPs/Indels and survival in Chinese and Canadian patients with ALS (n = 511). Trends of genome-wide statistical significance are represented by a red line (P < 5 × 10⁻⁸) and blue line (P < 5 × 10⁻⁷) (meta-analysis with a random-effects model). (B) QQ-plot for the meta-analysis results of genome-wide survival analysis in patients with ALS.

Table 1.

Information of meta-analysis results on SNPs associated with ALS survival in Chinese and Canadian cohorts (n = 511, P < 5 × 10⁻⁷).

SNP	Position (hg19)	Gene	Region	ALT	REF	P	HRa(95%CI)
rs117002347	7:116101079	CAV1/CAV2	Intron2	G	A	7.18 × 10−8	2.27 (1.68–3.05)
rs17533945	19:17257802	MYO9B	Intron2	C	T	1.01 × 10−7	2.05 (1.57–2.66)
rs72672232	4:111991274			A	G	3.66 × 10−7	2.45 (1.65–3.07)

^aRepresents the effect of the minor allele.

Notably, in the meta-analysis, we identified a suggestive link of CAV1/CAV2 rs117002347 (P = 7.18 × 10⁻⁸, HR = 2.27, 95% CI = 1.68–3.05, random-effects model) with ALS survival at genome-wide level. In addition, rs117002347 is still linked to survival after removing 7 carriers of known ALS genes (P = 6.12 × 10⁻⁷, random-effects model). Rs117002347 is located in intron-2 of the CAV1/CAV2 locus. CAV1 was previously reported to be associated with ALS risk.³⁴ Cox proportional hazards model analysis indicated that the rs117002347 GG or GA-genotype is significantly associated with a shorter survival time in our combined cohort (P = 3.57 × 10⁻⁷, HR = 2.14, 95% CI = 1.60–2.86), the Chinese cohort (P = 0.038, HR = 1.76, 95% CI = 1.03–3.02) and the Canadian cohort (P = 5.01 × 10⁻⁶, HR = 2.25, 95% CI = 1.58–3.18), suggesting that it is a potential survival modifier in both Chinese and Canadian patients. KM curves of survival probability indicated that the median survival time difference between AA and GG/GA carriers is 8 years (78 years vs 70 years) in the combined cohort (Supplementary Figure S6A), 7 years (79 years vs 72 years) in the Chinese cohort (Supplementary Figure S6B) and 9 years (78 years vs 69 years) in the Canadian cohort (Supplementary Figure S6C), which is corresponding to the significant difference of survival probability curves across rs117002347 genotypes (P = 3 × 10⁻⁶ for combined cohort, P = 0.02 for Chinese cohort, P = 2 × 10⁻⁶ for Canadian cohort, Log-rank test).

DNAm levels of CpG sites are associated with ALS survival

To identify the link between genome-wide non-genetic controlled DNAm of CpGs and ALS survival, we profiled DNAm status of 715,305 CpGs that do not overlap common SNPs in our combined ALS cohort (n = 459). Quality control analyses showed population heterogeneity derived from DNAm patterns; and most CpG sites clustered at hyper- or hypo-methylated levels (Supplementary Figure S7).

In the Chinese cohort, we identified five CpGs (cg020528, DLD cg18382273, cg21230007, cg04005341, C2CD3 cg26119740) showing significant associations between their DNAm levels and ALS survival (n = 210, P_bonferroni <0.05, Cox proportional hazards model, Supplementary Figure S8). In addition, DNAm levels of 158 CpGs mapped to 123 genes were nominally associated with ALS survival (P_FDR <0.05, Cox proportional hazards model, Supplementary Figure S8). KEGG pathway analysis of these 123 genes showed no significant enrichment, but GO pathway analysis (hypergeometric test) indicated significantly over-representative pathways, such as cell projection membrane (P_FDR = 0.0223), cytoplasmic side of plasma membrane (P_FDR = 0.0419), positive regulation of exocytosis (P_FDR = 0.0113) and dopamine receptor binding (P_FDR = 0.0211) pathways (Supplementary Figure S9). In the Canadian cohort (n = 249), we identified 8 CpGs (cg18627729, cg22232737, cg09726879, cg21757266, cg19921093, cg26638716, cg1713155, cg13017319) with significant association between their DNAm levels and ALS survival (P_bonferroni<0.05, Cox proportional hazards model, Supplementary Figure S10). In addition, DNAm levels of 366 CpGs mapped to 248 genes were nominally associated with ALS survival in Canadian patients (P_FDR<0.05, Cox proportional hazards model, Supplementary Figure S10). KEGG and GO pathway analysis of these 248 genes showed no significant enrichment. These findings suggest population differences for epigenetic survival modifiers. Notably, we observed left-shift inflation on QQ-plots for both cohorts, but the inflation factor is small (λ = 1.05 and λ = 1.01, Supplementary Figure S11). This inconsistency might be linked to the fact that epigenome wide association study is sensitive to environmental exposures (such as smoking and ageing) that also affect DNA methylome.

We then performed cross-ethnic Meta-analysis of the two cohorts using a random-effects model. A total of 715,305 CpG loci were included after coordinating the direction of CpG methylation effects. DNAm levels of 88 CpGs mapped to 40 genes are significantly associated with ALS survival (P_bonferroni <0.05, random-effects model, Fig. 3A, Supplementary Tables S6–S7). A moderate inflation of P-values was observed (λ = 1.17, Fig. 3B). In addition, DNAm levels of 3311 CpGs mapped to 2403 genes are nominally associated with ALS survival (P_FDR <0.05, random-effects model, Fig. 3). KEGG pathway analysis of these 2403 genes indicated a significant enrichment of 13 pathways, such as Cytoskeleton in muscle cells and the AMPK signalling pathway (P_FDR <0.05, hypergeometric test, Supplementary Figure S12A). Additionally, GO biological process pathway analysis of these 2403 genes revealed significant enrichment of 303 biological processes, such as regulation of neuron projection development and synapse organisation (P_FDR <0.05, hypergeometric test, Supplementary Figure S12B). Notably, enhanced AMPK activation in motor neurons corresponding to energy depletion and hypermetabolism,³⁵ disrupted cytoskeleton³⁶ and abnormal synapse organisation³⁷ were previously reported in patients with ALS.

Fig. 3.

The association between genome-wide DNA methylation of CpGs with disease survival in cross-population Meta-analysis of patients with ALS. (A) Manhattan plot showing the association between DNAm levels of 715,305 CpGs and survival in Chinese and Canadian patients with ALS (n = 459). The red line (P_bonferroni<0.05) and the blue line (P_FDR <0.05) represent the trend of statistical significance (meta-analysis with a random-effects model). Top 10 CpGs were labelled. (B) QQ-plot for the meta-analysis results of epigenome-wide survival analysis in patients with ALS.

DNAm levels of CpG-SNPs are associated with ALS survival

To investigate how the genetic controlled DNAm are linked to ALS survival, we performed genome-wide survival analyses on the DNAm of 8096 common CpG-SNPs in the combined ALS cohort (n = 459). Quality control analyses showed population heterogeneity derived from DNAm patterns; and CpG-SNPs clustered at hyper-, moderate- or hypo-methylated levels (Supplementary Figure S13), implicating the effect of three genotypes on DNAm.

In the Chinese cohort (n = 210), we identified one CpG-SNP showing significant association between its DNAm and survival (EXOC3L2 cg12478440, P_FDR = 0.047, HR = 0.712, Cox proportional hazards model) (Supplementary Figure S14). To validate this finding, we analysed the association between DNAm of EXOC3L2 cg12478440 and survival time in an independent Chinese ALS cohort (PUTH-ALS, n = 204).²¹ We found that DNAm of EXOC3L2 cg12478440 was not significantly associated with ALS survival time (P = 0.138, adjusted for sex, Cox proportional hazards model). In the Canadian cohort (n = 249), we found that DNAm of 4 CpG-SNPs were significantly associated with survival (PCGF3 cg11725415: P_FDR = 0.001, HR = 1.59; cg10235702: P_FDR = 0.023, HR = 2.94; cg26153182: P_FDR = 0.023, HR = 0.74; cg00759354: P_FDR = 0.037, HR = 0.71, Cox proportional hazards model) (Supplementary Figure S15). No inflation of P-values was observed based on quantile–quantile plots in both cohorts (λ = 1.06, λ = 1.08, Supplementary Figure S16). These findings suggest population differences in CpG-SNP survival modifiers (such as the PCGF3 cg11725415).

We then performed the cross-population meta-analysis and found significant associations between DNAm levels of 8 CpG-SNPs and ALS survival (P_FDR <0.05), four of which mapped to known genes, namely PCGF3 (cg11725415, P_FDR = 8.15 × 10⁻⁵, HR = 1.62), BAG6 (cg27014438, P_FDR = 7.26 × 10⁻³, HR = 1.74), MCC (cg07044588, P_FDR = 3.23 × 10⁻², HR = 0.80) and TPSD1 (cg10516012, P_FDR = 3.23 × 10⁻², HR = 2.23) (random-effects model) (Fig. 4A, Table 2, Supplementary Table S8). No significant inflation of P-values was observed (λ = 0.954, Fig. 4B).

Fig. 4.

The association between genome-wide DNA methylation of CpG-SNPs with disease survival in cross-population Meta-analysis of patients with ALS. (A) Manhattan plot showing the association between DNAm levels of 8096 CpG-SNPs and survival in Chinese and Canadian patients with ALS (n = 459). The red line (P_bonferroni <0.05) and the blue line (P_FDR <0.05) represent the trend of statistical significance (meta-analysis with a random-effects model). (B) QQ-plot for the meta-analysis results of survival analysis of CpG-SNP methylation in patients with ALS.

Table 2.

Information of meta-analysis results on CpG-SNPs showing significant association between their DNA methylation levels and survival in Chinese and Canadian ALS cohorts (n = 459, P_FDR <0.05).

CpG	SNP	Position (hg_19)	Gene	Region	HRa	HRa (95% CI)	P	P_FDR
cg11725415	rs55990625	4:736485	PCGF3	Intron7	1.62	1.38–1.92	1.01 × 10−8	8.15 × 10−5
cg14042541	rs4432238	15:81693224			5.72	2.83–11.55	1.13 × 10−6	4.55 × 10−3
cg27014438	rs28732154	6:31607674	BAG6	Intron23	1.74	1.38–2.19	2.70 × 10−6	7.27 × 10−3
cg10516012	rs3865205	16:1306346	TPSD1	Exon1	0.80	0.72–0.89	2.02 × 10−5	3.23 × 10−2
cg07044588	rs76637203	5:112498185	MCC	Intron1	2.23	1.85–2.70	2.06 × 10−5	3.23 × 10−2
cg03145999	rs8108978	19:58629828			0.16	0.07–0.38	2.69 × 10−5	3.23 × 10−2
cg08748626	rs55641734	4:12725896			1.68	1.32–2.14	2.80 × 10−5	3.23 × 10−2
cg19300401	rs6903672	6:16962712			1.15	1.07–1.23	4.73 × 10−5	4.78 × 10−2

^aRepresents the effect of every 10% DNA methylation level increase.

DNAm level of the BAG6 locus is linked to ALS survival

In our combined ALS cohorts, one CpG-SNP (BAG6 cg27014438/rs28732154) showed consistent association between their DNAm levels and survival in both the Chinese (n = 210, P = 0.00043, HR = 1.15, for cg27014438, Cox proportional hazards model) and Canadian ALS cohorts (n = 249, P = 0.0005, HR = 1.23, for cg27014438, Cox proportional hazards model). KM curve of survival probability showed that the high DNAm subgroup of BAG6 cg27014438 had a shorter survival time in the combined ALS cohort (n = 459, HR = 2.706, P = 5.09 × 10⁻⁷, 95% CI: 1.83–4.00, Cox proportional hazards model, Fig. 5A). The median survival time is 12.3 years longer in the low methylation group than the high methylation group of cg27014438 (83.3 vs 71 years).

Fig. 5.

The DNA methylation of BAG6 CpG-SNPs is linked to ALS survival. The DNAm level (β) of 0.47 was set as the cutoff to classify low and high methylation subgroups. (A) Kaplan Meier curve of survival probability stratified by DNAm level of cg27014438 in 459 patients with ALS (P = 5.09 × 10⁻⁷, HR = 2.706, 95% CI: 1.83–4.00, Cox proportional hazards model). (B) Kaplan Meier curve of survival probability stratified by DNAm level of cg27014438 in 210 Chinese patients with ALS (P = 7.96 × 10⁻⁵, HR = 5.032, 95% CI: 2.26–11.227, Cox proportional hazards model). (C) Kaplan Meier curve of survival probability stratified by DNAm level of cg27014438 in 249 Canadian patients with ALS (P = 1.15 × 10⁻⁵, HR = 2.976, 95% CI: 1.83–4.844, Cox proportional hazards model).

To validate this finding, we analysed the link between DNAm of BAG6 cg27014438 with survival time in an independent Chinese ALS cohort (PUTH-ALS, n = 204).²¹ We found that DNAm of BAG6 cg27014438 still showed significant association with ALS survival (P = 0.031, HR = 1.148, 95% CI = 1.01–1.304, adjusted for sex, Cox proportional hazards model).

Population subgroup analyses showed that the hypermethylation subgroup of cg27014438 (with a DNAm cut-off of 0.47) was significantly associated with a shorter survival time in both Chinese (P = 7.96 × 10⁻⁵, HR = 5.032, 95% CI: 2.26–11.23, Cox proportional hazards model, Fig. 5B); and Canadian cohorts (P = 1.15 × 10⁻⁵, HR = 2.97, 95% CI: 1.83–4.84, Cox proportional hazards model, Fig. 5C). The median survival time difference is 9.2 years (73.1 years vs 63.9 years) and 9.3 years (83.3 years vs 74 years) in Chinese and Canadian cohort.

Cg27014438 methylation is dysregulated in different BAG6 rs28732154 genotypes and is linked to transcript expression

To further validate the cross-talk of genetic variant and DNAm on the BAG6 locus, we analysed the paired genome and methylome data from our Chinese (n = 205) and Canadian (n = 249) patients. We found that the rs28732154 CC genotype increased DNAm status of cg27014438 on BAG6 in both Chinese (P = 1.934 × 10⁻⁶, Wilcoxon rank sum test, Fig. 6A) and Canadian patients (P = 0.0001387, Wilcoxon rank sum test, Fig. 6B). In the combined cohort (n = 454), the BAG6 rs28732154 CC-genotype significantly increased DNAm status of cg27014438 (P = 9.453 × 10⁻¹⁴, Wilcoxon rank sum test, Fig. 6C). Additionally, reanalysis of a previous dataset further supports rs28732154 as a mQTL in blood (Supplementary Table S9).³⁸ However, the rs28732154 genotype is not directly linked to ALS survival in our combined cohort (P = 0.56, Cox proportional hazards model, n = 454), the Chinese (P = 0.844, n = 205) or the Canadian cohort (P = 0.26, Cox proportional hazards model, n = 249).

Fig. 6.

BAG6 cg27014438 methylation is altered in patients carrying different rs28732154 genotypes. (A) Violin plot showing that cg27014438 methylation is increased in rs28732154 CC genotype group compared to CT/TT genotype group in Chinese patients with ALS (P = 1.934 × 10⁻⁶, Wilcoxon rank sum test, median methylation level: 0.39 vs 0.26). (B) Violin plot showing that cg27014438 methylation is increased in rs28732154 CC genotype group compared to CT/TT genotype group in Canadian patients with ALS (P = 0.0001387, Wilcoxon rank sum test, median methylation level: 0.47 vs 0.15). (C) Violin plot showing that cg27014438 methylation is increased in rs28732154 CC genotype group compared to CT/TT genotype group in the combined patients with ALS (P = 9.453 × 10⁻¹⁴, Wilcoxon rank sum test, median methylation level: 0.45 vs 0.25).

To evaluate whether DNAm of cg27014438 is linked to gene expression, we extracted the results of cis-eQTM analysis from 2115 Framingham Heart Study individuals for genes located within 250 kb from the candidate CpG site.³¹ We found that cg27014438 methylation is significantly linked to the expression of two BAG6 transcripts (ENST00000435080.5, P_adj = 3.29 × 10⁻⁵, slope = 0.80; ENST00000424176.5, P_adj = 4.57 × 10⁻⁵, slope = 0.65; linear regression) in blood (Supplementary Table S10). These findings suggest that hypermethylation of cg27014438 is linked to increased blood expression of BAG6 transcripts and a shorter survival time. However, in the TargetALS dataset, BAG6 expression is not associated with survival in multiple CNS tissues (P > 0.05, Cox proportional hazards model, Supplementary Table S11). It is unclear whether DNAm of cg27014438 also regulate BAG6 expression in brain tissues.

Discussion

In the current study, our multi-omics study nominated both genetic and epigenetic factors associated with ALS survival, implicating a previously reported ALS-linked gene (CAV1), and an epigenetic survival-modifying BAG6 locus. Currently, most epigenome studies in patients with ALS are from populations with European ancestry.^6,13,26,39 Our multi-population study revealed both population specific and common survival modifiers in Chinese and Canadian patients, providing a unique dataset.

The CAV1/CAV2 locus (tagged by rs117002347) is nominally linked to ALS survival, which corresponds to a previous report that pointed to CAV1 as an ALS-associated gene.³⁴ In motor neurons, CAV1 and CAV2 are co-expressed in hetero-oligomeric complexes within plasma membrane lipid rafts and plays a critical role in intercellular signal transduction.^40,41 Hence, our and previous studies support the potential role of CAV1 in modifying pathological process of ALS. The rs117002347 variant was not reported in a previous survival study⁶ including 4256 patients with ALS. This might be related to the difference of (1) patients’ ethnicity (95% White vs unclear ethnicity in the USA samples), (2) genotyping methodology (WGS vs GWAS arrays), and (3) screening of known ALS mutations (a complete C9orf72 expansion screening vs an incomplete C9orf72 expansion screening).

The interactions between genetic variants and epigenetic modifications may affect disease risk or phenotypes in neurodegenerative diseases. For example, two CpG-SNPs at the C6orf10/LOC101929163 locus are significantly associated with age of onset in C9orf72 patients.³⁹ Here, we showed that DNAm of a BAG6 CpG-SNP (cg27014438/rs28732154) is linked to ALS survival across Chinese and Canadian patients, implicating the BAG6 locus as an ALS survival modifier. BAG6 encodes BAG cochaperone 6, which may prevent intracellular aggregation of TDP-43, a key CNS pathology in ALS. As a molecular ‘staging partner’, BAG6 stabilises aggregation-prone substrates and prevents their self-assembly into toxic oligomers or insoluble aggregates.⁴² In ALS, exposure of misfolded TDP-43 to hydrophobic domains drives pathological aggregation, whereas BAG6 reduces cytoplasmic aggregation by sensing its solvent-exposed hydrophobicity and promoting ubiquitin-proteasomal degradation.⁴² Abnormal BAG6 function might be involved in regulating ALS survival by modulating TDP-43 toxicity in CNS tissues.

BAG6 cg27014438 methylation, but not rs28732154 genotypes are associated with ALS survival, suggesting that the BAG6 methylation may regulate gene expression driving the modifying effect of survival time, while rs28732154 is not an eQTL based on GTEx. The cg27014438 methylation may also be regulated by other ALS risk related environmental factors (such as smoking^43,44 and head injury^45,46). Particularly, cg27014438 hypermethylation located on intron-2 is linked to an increased BAG6 expression, which is corresponding to a reported phenomenon that gene body methylation may up-regulate gene expression by regulating transcriptional elongation or splicing.⁴⁷

Our cross-ethnicity study shows that the cg27014438 methylation might be a potential prognostic marker to predict ALS survival time, but warrants further validation in other independent cohorts. The current study lacks mechanistic evidence of how BAG6 cg27014438 methylation regulates ALS survival, which may hamper the usage of this gene as a drug target. Future studies are needed to use iPSC motor neurons and animal models to test how cg27014438 methylation regulate gene expression and TDP-43 pathology (such as aggregation and mis-localisation) in motor neurons, brain and spinal cord tissues, and assess whether BAG6 can be a therapeutic target to rescue ALS phenotypes in mouse models (such as TDP-43^tg/tg mouse⁴⁸).

Our study included a moderate sample size for identifying genetic modifiers, which may hamper the identification of survival modifiers with a small effect size. However, we validated the previously reported CAMTA1 locus as a survival modifier,⁶ suggesting that our cohort has the power to identify modifiers with a similar effect size as the CAMTA1 locus. Future studies should include independent cohorts to validate our findings. Notably, we provided a unique dataset with paired WGS and methylome data from a total of 454 patients with ALS, which made it possible to elucidate the interactions between genetic variants and DNAm modifications and explore their roles in disease progression and survival. Future functional assessments could also evaluate whether BAG6 knock-down/overexpression damages or rescues neuronal functions to further clarify the molecular mechanism underlying the modification of disease survival. Additionally, the current study reported blood epigenetic alterations enriched to AMPK, cytoskeleton and synapse organisation pathways, which might partially reflect systemic inflammation or metabolic disturbances in ALS.¹³ However, whether epigenetic changes linked ALS survival is similar between blood and CNS tissues remain unknown. Although blood and CNS has different methylome profile, the previous study showed that some inter-individual variation was reflected across brain and blood, indicating that peripheral tissues may have some utility in studying neurobiological phenotypes.⁴⁹ Future epigenetic and functional studies in motor neurons or brain tissues might clarify the link of our findings between blood and CNS tissues.

In conclusion, our multi-omics study highlighted BAG6 cg27014438 methylation as a potential epigenetic modifier of ALS survival, which is modulated by rs28732154 genotypes and linked to BAG6 expression. Our findings enhanced our understanding of the interplay between genetic variants and DNAm, and suggested a potential survival modifier (BAG6 locus) in patients with ALS across diverse populations.

Contributors

M.Z, Y.G contributed to the conception and design of the study; Y.C, X.C, E.R, L.Z, C.S, Y.D, D.S.F and G.B.C contributed to the collection of clinical samples and clinical information. X.T, W.Y, Y.C, X.C, J.G, Y.G, J.H, J.L, Y.L.C and Y.R contributed to the acquisition and analysis of data. M.Z and Y.G contributed to drafting the manuscript. M.Z and Y.G have the full access to all data related to the study. All authors read and approved the final version of the manuscript.

Data sharing statement

The data (WGS and DNAm array) that support the findings of this study are available on reasonable request from the corresponding author. The genomic data and epigenomic data of Chinese patients have been deposited in the Genome Sequence Archive for human and Genome Variation Map (GVM) (GVM001120) and OMIX (OMIX009601) in the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/). The RNA-seq from TargetALS is available on request at the Human Postmortem Tissue Core resources, https://www.targetals.org/.

Declaration of interests

Y.L.C is a visiting Professor of Shanghai JiaoTong University, and a founder of Synphatec (Shanghai) Biopharmaceutical Technology Co., Ltd. The other authors have no conflict of interest to declare.