Conserved DNA methylation combined with differential frontal cortex and cerebellar expression distinguishes C9orf72-associated and sporadic ALS, and implicates SERPINA1 in disease

Abstract

We previously found C9orf72-associated (c9ALS) and sporadic amyotrophic lateral sclerosis (sALS) brain transcriptomes comprise thousands of defects, among which, some are likely key contributors to ALS pathogenesis. We have now generated complementary methylome data and combine these two data sets to perform a comprehensive “multi-omic” analysis to clarify the molecular mechanisms initiating RNA misregulation in ALS. We found that c9ALS and sALS patients have generally distinct but overlapping methylome profiles, and that the c9ALS- and sALS-affected genes and pathways have similar biological functions, indicating conserved pathobiology in disease. Our results strongly implicate SERPINA1 in both C9orf72 repeat expansion carriers and non-carriers, where expression levels are greatly increased in both patient groups across the frontal cortex and cerebellum. SERPINA1 expression is particularly pronounced in C9orf72 repeat expansion carriers for both brain regions, where SERPINA1 levels are strictly down regulated across most human tissues, including the brain, except liver and blood, and are not measurable in E18 mouse brain. The altered biological networks we identified contain critical molecular players known to contribute to ALS pathology, which also interact with SERPINA1. Our comprehensive combined methylation and transcription study identifies new genes and highlights that direct genetic and epigenetic changes contribute to c9ALS and sALS pathogenesis.

Introduction

Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) are devastating neurodegenerative disorders seen comorbidly in up to 50% of patients, with only a small proportion of ALS patients reporting a family history of either disease[8]. Genetic studies have implicated more than two dozen genes in disease pathogenesis[36,40], the most common mutation being a GGGGCC (G₄C₂) hexanucleotide repeat expansion in the C9orf72 gene that is carried by 6% and 34% of sporadic (sALS) and familial (fALS) cases, respectively[34]. Since sALS accounts for up to 95% of all ALS cases[7,13], and only ∼11% of these cases can be attributed to known genetic risk factors, there is an intensifying need to reveal sALS etiology. The motor cortex is the brain region primarily affected in ALS, though increasing evidence suggests the frontal cortex and cerebellum play a role in both c9ALS and sALS[33,12]. Our recent RNA sequencing (RNAseq) study demonstrated distinct misregulated brain transcriptome profiles in the cerebellum and frontal cortex for both sALS patients, and ALS patients carrying the C9orf72 G₄C₂ repeat expansion (c9ALS)[33]. Transcriptome defects, including differential expression and alternative splicing, were abundant in both the frontal cortex and cerebellum[33]. While transcriptome data provide critical disease insight, understanding the mechanism underlying these distinct transcriptome profiles may be even more important. Here, we contrast epigenetic modifications in frontal cortex and cerebellum between genetically unexplained sALS with c9ALS cases to: (1) further understand the mechanisms driving the distinct transcriptome profiles and disease, and (2) filter and identify new genes driving disease development and progression.

Substantial evidence suggests epigenetic changes may be regulating the changes observed in ALS. Xi et al. previously reported ALS-discordant monozygotic twins carrying the G₄C₂ C9orf72 repeat expansion[43], suggesting the C9orf72 repeat expansion, alone, may not have driven disease in the affected twin. Similarly, Lam et al. recently demonstrated significant DNA methylation differences between 50-year-old, ALS-discordant monozygotic twins[21]. Both studies suggest epigenetics may be a catalyst for disease, even with the highly penetrant C9orf72 genetic variant. Recent ALS epigenetic studies also showed patients with c9ALS have expression-altering epigenetic modifications of the C9orf72 locus, including repressive histone marks and DNA methylation in promoter and repeat regions[4,45,42,44]. Liu et al. further demonstrated that hyper-methylation in the C9orf72 promoter region moderates: (1) RNA foci accumulation, (2) C9orf72 protein burden, and (3) vulnerability to oxidative stress, supporting an interaction between aberrant epigenetic control, RNA misregulation and protein toxicity in c9ALS[23]. Additionally, hairpin and G-quadruplex structures adopted by the C9orf72 repeat expansion RNA[11,38,15] may interact with chromatin remodeling factors, as similar secondary structures have been shown to attract several chromatin-modifying complexes[17]. In fact, there is now compelling evidence that RNA not only serves as an intermediary between DNA and protein, but also mediates chromatin remodeling and nuclear architecture through epigenetic regulation[2,24,16,26,6]. Evidence further supports altered RNA-mediated regulation as a central pathological epigenetic mechanism in many diseases[14,31,33,35,5]. While some studies have demonstrated epigenetic involvement in ALS[25], identifying the C9orf72 repeat expansion provided even more support, as other repeat expansion disorders lead to chromatin remodeling [4,3,42,45,44,5].

To clarify the molecular mechanisms initiating RNA misregulation in ALS, we sought to begin characterizing the relationship between methylation and disease by: (1) evaluating whether aberrant DNA methylation contributes to the distinct c9ALS and sALS transcriptomes previously described[33], and (2) identifying interesting epigenetic mediators of RNA regulation in ALS, and (3) assess the role of ALS-associated genes in an interconnected gene network context. We generated DNA methylome profiles for frontal cortex and cerebellum from c9ALS, sALS, and non-disease control individuals from our initial RNAseq study[33] and found that genes in critical ALS pathways have both transcriptome defects (e.g., differential expression and alternative splicing) and differential methylation. We present a valuable characterization of methylation profiles in c9ALS and sALS across the frontal cortex and cerebellum, and we validate new gene targets that may be critical in disease. This is the first full methylome characterization across the frontal cortex and cerebellum in ALS, and we combine it with our previous RNASeq data for a comprehensive “multi-omic” approach to better understand what drives disease.

Materials and Methods

Human Tissue

Human cerebellum and frontal cortex tissues were obtained with written informed consent from participants or authorized family members and processed under the approved Mayo Clinic Institutional Review Board and Ethics Committee on Human Experimentation 09-008148 and 12-007795 protocols. Information about ALS subject recruitment, diagnosis, genotyping, pathological assessment procedure and brain tissue sampling were previously described[33]. Among the c9ALS, sALS and control subjects used for the RNAseq study, four subjects per group were matched based on clinical information and selected for methylome profiling. We validated differential gene expression using a total of 32 cases carrying a pathogenic C9orf72 repeat expansion (16 ALS, 3 ALS-FTD, 11 FTD), 28 cases carrying a normal C9orf72 repeat size (17 ALS, 3 ALS-FTD, 8 FTD), and 8 disease controls. Some samples included in cerebellum were excluded in frontal cortex because of RNA quality. We included FTD cases to increase sample sizes during validation. The controls were confirmed pathologically normal, and have no history of motor neuron disease or dementia, as previously described[33]. Patients were classified into ALS, ALS-FTD, and FTD groups based on the neuropathological diagnosis. Information on all subjects used for this study can be found in online resource, Table 12. All subjects were confirmed negative for protein-coding mutations in SOD1, TARDBP, and FUS.

Next Generation RNA sequencing (RNAseq)

All RNASeq results are available online at the NCBI Gene Expression Omnibus under accession number GSE67196 [33].

DNA and RNA Preparation

DNA was isolated from 20–40mg of sampled frontal cortex and cerebellum frozen tissue using the Wizard genomic DNA purification kit (Promega, Madison, WI, USA) per manufacturers' instructions, and re-suspended in TE buffer. Concentration was assessed using a Qubit 3.0 Fluorometer (ThermoFisher Scientific, Waltham, MA), and 50ng/ul of DNA in a 10ul volume was sent to the Mayo Clinic Medical Genome Facility in Rochester (Minnesota, USA) for processing and sequencing.

RNA was isolated from 20–40mg of sampled frontal cortex and cerebellum frozen tissue using a two-step tissue homogenization method. Total RNA was extracted using miRNeasy Mini Kit (QIAGEN, Hilden, Germany), per manufacturer's instructions. RNA integrity (RIN>6.6) was verified on an Agilent 2100 bioanalyzer (Santa Clara, CA, USA).

Reduced Representation Bisulfite Sequencing (RRBS)

Reduced representation bisulfite sequencing (RRBS) library construction and sequencing were conducted at the Mayo Clinic Medical Genome Facility in Rochester (Minnesota, USA). Briefly, genomic DNA (250ng) was digested with the methylation insensitive restriction digest enzyme, Msp1, which cleaves the DNA at CCGG, creating fragments high in CpG content. The DNA was then purified using QIAGEN (Hilden, Germany) MiniElute columns. Samples were end-repaired and A-tailed using a Klenow fragment. The end repair A-tailing allowed for selective ligation of the Illumina (San Diego, CA, USA) TruSeq™ indexed adaptors. TruSeq™ adaptors were ligated to the sample DNA using T4 DNA ligase. Size selection was performed with Ampure Xp™ beads. Selection range was between 150 and 400bp, allowing for insert size of 40 to 280bp and including many CpG islands and shores. Each sample was bisulfite modified using the Zymo EX DNA Methylation Kit (Zymo Research Corp, Irvine, CA; PN D5001) under the following conditions: 55 cycles of 95°C for 30 seconds followed by 50°C for 15 minutes, then held at 4°C. The samples were then PCR amplified (15 cycles) to enhance for CpG regions. Concentration and size distribution of the libraries were determined using Agilent (Santa Clara, CA, USA) Bioanalyzer DNA 1000 chip, Qubit fluorometry (Invitrogen, Carlsbad, CA), and qPCR. Libraries were pooled at equimolar concentrations and loaded onto a paired end flow cell at concentrations of 7-8pM to generate cluster densities of 600,000-700,000/mm² following Illumina's standard protocol, using the Illumina cBot and HiSeq Paired end cluster kit version 3 (San Diego, CA, USA). Because bisulfite treated samples are unbalanced due to an under representation of C bases, a lane of PhiX genomic DNA was run to allow the Illumina data collection software to call bases correctly. The flow cells were sequenced as 51×2 paired end reads on an Illumina HiSeq 2000 using TruSeq SBS sequencing kit version 3 and HiSeq data collection version 2.0.12.0 software. Base-calling was performed using Illumina's RTA version 1.17.21.3. We assessed bisulfite conversion success rate by calculating the percentage of non-CpG cytosines across all samples, where we would expect nearly all said cytosines to convert. Conversion percentages were >99% for all samples in the CHG context, and ranged from 98.6% to 99.5% in the CHH context, where ‘H’ represents any non-G nucleotide.

RRBS Alignment, Annotation, Duplicate Removal, and Bioinformatic Analysis

Paired-end reads were aligned to the hg19 reference genome and cytosine counts were determined using Bismark (v0.13.1)[20] and Bowtie 2 (v2.2.3)[22]. We tested for differential methylation between sample groups with methylKit's[1] calculateDiffMeth function, using logistic regression and a sliding linear model for false discovery rate adjustment. Sites were filtered for minimum read coverage of 10, and to be within the bottom 99.9% of coverage for all sites. Each differentially methylated cytosine (DMC) was then annotated using methylKit's get Association With TSS function according to (1) the nearest downstream transcription start site or (2) the “gene body” (located between the transcription start site and end of the 3′UTR) in which the DMC was located, according to UCSC genome browser hg19 refseq annotations (https://genome.ucsc.edu/). A DMC was considered significant if the q-value was less than 0.01 and the estimated methylation in the c9ALS or sALS group was greater than 25% different from the control group. Duplicate removal was conducted in the following way: (1) if a DMC was located upstream from the TSS of more than one gene, it was counted as multiple events; (2) if a DMC was located upstream or in the gene body of different transcript variants of the same gene, it was counted as one single event; (3) if a DMC was located one base pair away from another methylation event on the opposite strand and annotated to the same gene, it was considered a paired symmetrical methylation and was counted as one single event; (4) if a DMC was located one base pair away from another DMC on the opposite strand and annotated to a known antisense gene, it was counted as two different events.

RT-PCR and qRT-PCR

Using the High Capacity cDNA Transcription Kit (Applied Biosystems, Foster City, CA, USA) per manufacturer's instructions, 500ng of RNA was used to generate cDNA by reverse transcription polymerase chain reaction (RT-PCR). To quantify the RNAs, quantitative real-time PCRs (qRT-PCRs) were conducted using cDNA generated from the High Capacity cDNA Transcription Kit and the SYBR Green PCR Mastermix (Applied Biosystems, Foster City, CA, USA), per manufacturer's instructions. All primer sequences are available in online resource, Table 13. Samples were run in triplicate on a QuantStudio 7 Flex PCR System (Applied Biosystems, Foster City, CA, USA). Relative quantification was determined using the ΔΔCt method and the median for each sample was normalized to the endogenous control RPLP0. The non-parametric Kruskal-Wallis test was used to test for significant differences between cases and controls. The false discovery rate was controlled using the two-stage linear step-up procedure of Benjamini, Krieger and Yukutieli. Comparisons reached significance with p-values <0.05 (*p<0.05; **p<0.01; ***p<0.001).

Combined RRBS and RNASeq analysis and validation

Because our data are a comprehensive methylation profile in ALS, we filtered out many genes to identify specific targets to validate. Specifically, we focused on genes that were differentially expressed (p < 0.05, |log2FC| > 1) and had differential methylation patterns (p < 0.05) with >25% difference between c9ALS and controls, or between sALS and controls. From that list, we looked for genes that were common between either c9ALS and sALS, or between cerebellum and frontal cortex, and tested for association between expression and methylation in these genes using logistic regression (glm, family = “binomial”) in R (v. 3.3.2). To preserve power, samples were sorted into two groups, fully methylated and partial/fully unmethylated, and tested for association with expression levels. Genes that are common across these groups may be critical to disease development and progression. We also used public data sets from The Genotype-Tissue Expression (GTEx) Project (https://gtexportal.org/home/) and 10× Genomics (https://support.10xgenomics.com/single-cell-gene-expression/datasets/1.3.0/1M_neurons) to assess gene expression for top gene(s) across human tissues and across individual mouse brain cell types. GTEx is supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. The data used for the analyses described in this manuscript were obtained from the GTEx Portal on July 12, 2017.

Hierarchical clustering

Methylation differences from the top 2,000 sites in c9ALS and sALS frontal cortex and cerebellum contrast by magnitude were clustered using complete-linkage clustering and Euclidian distance.

Data Availability

We deposited the RRBS data in the NCBI Gene Expression Omnibus (GEO) database under accession number # GSE97106.

Results

Differentially methylated cytosines are abundant in ALS patient brains

We found thousands of differentially methylated cytosines (DMCs) in the frontal cortex and cerebellum of c9ALS and sALS cases, when compared to control brains. Overall, there were 4,430 total frontal cortex DMCs in c9ALS patients, where 1,487 were in a gene body and 2,943 were upstream of a transcription start site (TSS) (Fig. 1a). There were 3,721 frontal cortex DMCs in sALS patients, where 1,267 were in a gene body and 2,454 were upstream of a TSS (Fig. 1a). In the cerebellum, there were 4,454 (1,330 + 3,124) and 6,412 (2,100 + 4,312) across c9ALS and sALS (Fig. 1b), respectively. In strong contrast, DMCs were least abundant in sALS frontal cortex and most abundant in sALS cerebellum. DMC abundance was similar between c9ALS frontal cortex and cerebellum.

Fig. 1. c9ALS and sALS have distinct brain methylome profiles.

(a, b) Pie charts representing the proportion of DMCs identified upstream of a TSS (darker color) or in the body of a gene (lighter color) in c9ALS (blue; n=4) and sALS (yellow; n=4), for both frontal cortex (a) and cerebellum (b). (c, d) Pie charts representing the total number of DMCs located upstream of a TSS in the frontal cortex (c) and cerebellum (d) for c9ALS (n=4) and sALS (n=4), and their distance from the nearest TSS. (e, h) Pie charts representing DMCs upstream of the TSS unique to c9ALS (blue; n=4) and sALS (yellow; n=4), or shared between the two disease groups, in frontal cortex (e) and cerebellum (h). (f, g) Pie charts representing upstream of TSS DMCs unique to frontal cortex (left) and cerebellum (right), or shared between the two brain regions, in c9ALS (n=4) and sALS (n=4). (i, j) Pie charts representing the type of genes affected by upstream of TSS DMCs in frontal cortex (i) and cerebellum (j). (k, l) Hierarchical clustering of genes with the top 2,000 upstream of TSS DMCs in c9ALS (n=4) and sALS (n=4) (4,000 total), in frontal cortex (k) and cerebellum (l).

Approximately 66% of DMCs were located upstream of a TSS in the frontal cortex for both c9ALS and sALS, while the remaining 34% were located in a gene body. While proportions were similar in sALS cerebellum, 70% of c9ALS cerebellum DMCs were upstream of a TSS. Of the 66-70% of DMCs located upstream of the TSS, we observed that DMCs in the frontal cortex were more frequent in c9ALS (2,943) than sALS (2,454; Fig. 1c), while DMCs in the cerebellum were much more abundant in sALS (4,312) than c9ALS (3,124; Fig. 1d). A majority of DMCs, ranging from 95-96%, were located >3kb upstream of the TSS, and DMCs were predominantly located in the cerebellum for both disease groups. Among the upstream DMCs identified, more than 100 were located within 3kb of a TSS for each of the four groups (Fig. 1c and 1d). Among these, more than half were found less than 1.5kb away, a region known to play a crucial role in gene regulation.

c9ALS and sALS patients may have distinct, but overlapping brain methylome profiles

While most upstream DMCs were unique to a certain disease group or brain region, some were also shared between c9ALS and sALS in both frontal cortex (n=385; Fig. 1e) and cerebellum (n=576; Fig. 1h), affecting 687 and 1,014 genes respectively. Some DMCs were also shared between the two brain regions in c9ALS (n=737; Fig. 1f) and sALS (n=673; Fig. 1g), affecting 973 and 954 genes respectively.

Most of the genes with upstream DMCs encode proteins (69-70%), while 13-14% of the total number of genes affected comprise non-coding RNAs (ncRNAs; Fig. 1i and 1j). Approximately 5-6% of the genes consisted of small RNA transcripts (sRNAs) such as miRNAs and PIWI-associated RNAs (piRNAs), whereas the remaining 11-12% of genes for each disease group and brain region mostly consisted of other unclassified genes. A small proportion of the affected regions were in transposable elements and genes encoding RNA-binding proteins (Fig 1i and 1j).

To evaluate the similarity of the methylome profiles included in each disease group, we conducted hierarchical clustering of the top 2,000 upstream DMCs found in c9ALS and the top 2,000 upstream DMCs identified in sALS in both frontal cortex and cerebellum. We observed that samples clustered perfectly per group (Fig. 1k and 1l), suggesting that between-group methylomes are distinct from one another and within-group methylomes are highly similar.

Combined methylation and transcriptome data identify genes with both DMCs and differential transcriptome defects

Combining data between the RRBS and our previous RNASeq data[33], we identified differentially-expressed genes in ALS that also have DMCs. DMCs were less frequent in differentially-expressed genes for c9ALS frontal cortex (n=77, Fig. 2a), and sALS frontal cortex (n=21, Fig. 2b), compared to c9ALS cerebellum (n=91, Fig. 2c) and sALS cerebellum (n=43, Fig. 2d). For both c9ALS and sALS frontal cortex and sALS cerebellum, the most common types of transcriptome changes in genes with DMCs were differential expression and alternative poly-A events, while alternative cassette exon splicing was the most common in c9ALS cerebellum (Fig. 2a-2d).

Fig. 2. DMCs and transcriptome defects are abundant in ALS.

(a-d) The darker left pie charts show the total number of genes found with differential expression (DE) (blue; p<0.05; |log2FC|≥1), alternative cassette exon (CE) splicing (green; FDR<0.05; |dl|≥0.1), intron retention (IR) (orange; FDR<0.05; |dl|≥0.1), and alternative polyadenylation (APA) (purple; FDR<0.05; |ΔPDUI|≥0.2; |dPDUI|≥0.2) in our RNAseq for c9ALS (a, c; n=8) and sALS (b, d; n=10) when compared to control subjects (n=9 in frontal cortex; n=8 in cerebellum), in both frontal cortex (a, b), and cerebellum (c, d). The lighter right pie charts (a-d) represent the proportion of genes from each transcriptome defect category with DMCs. For each of these events, the protruding slices of each color represent DMCs in the predicted effective region. Of note, numbers for genes with DE combine both upstream of TSS and 5′UTR DMCs, while numbers for genes with CE, IR and APA are for gene body DMCs.

We also evaluated the position of DMCs affecting genes with transcriptome defects (Fig. 2a-2d). While many DMCs were located in regions predicted to modulate transcription, most DMCs were located farther away or outside these regions. Regions typically predicted to modulate transcription include regions within 3kb of a TSS or in 5′UTR for differentially expressed genes (Fig. 2a-2d, light blue), in the 3′UTR for alternative poly-A (light purple), or within an alternatively spliced region for cassette exons (light green) and intronic region retention (light orange).

SERPINA1 and other genes are differentially expressed and methylated in the cerebellum and frontal cortex

We selected top differentially expressed genes with significant DMCs that were common across brain regions or disease states to validate (Fig. 2a-2d, protruding pie slices). The gene SERPINA1 had a significant increase in differential expression (DE) and significant decrease in differential methylation (DM) in the frontal cortex for sALS (Fig. 3a; DE vs. control: p = 9.14e^-3, log2FC = 1.69; DM vs. control: q = 2.49e^-6, percent difference = -25.54), and in the cerebellum for both c9ALS (DE vs. control: p = 6.38^e-4, log2FC = 2.29; DM vs. control: q = 1.31e^-8, percent difference = -29.40) and sALS (DE vs. control: p = 4.67^e-2, log2FC = 1.26; DM vs. control: q = 2.37e^-10, percent difference = -33.39). We confirmed SERPINA1 expression is increased in the frontal cortex for sporadic cases (p = 0.036, Fig. 3b), when compared to controls. Additionally, we found C9orf72 expansion carriers also have increased SERPINA1 expression (p = 0.001, Fig. 3b), when compared to controls. We also confirmed increased SERPINA1 expression in the cerebellum for both sporadic (p = 0.048) and C9orf72 expansion carriers (p = 0.007), when compared to controls in our expanded validation cohort (Fig. 3c). SERPINA1 methylation at position 94,849,201 on chromosome 14 is also correlated with SERPINA1 expression (p = 0.051) in our expanded validation cohort.

Fig. 3. SERPINA1 is over-expressed in C9orf72 repeat carriers and non-carriers brains.

(a) Summary of data obtained after RNAseq, RRBS and systems biology analyses for SERPINA1 in c9ALS and sALS frontal cortex and cerebellum. (b-c) Validation of SERPINA1 over-expression in frontal cortex (b) and cerebellum (c). Validation cohort included ALS, ALS-FTD, and FTD cases for both disease groups. Quantifications were performed for C9orf72 repeat carriers (C9orf72, blue) and non-carriers (sporadic, orange) and disease control cases (white). The non-parametric Kruskal-Wallis test was used to test for significant differences between cases and controls. The false discovery rate was controlled using the two-stage linear step-up procedure of Benjamini, Krieger and Yukutieli. Stars represent comparisons with p-values reaching significance (*p-value<0.05; **p-value<0.01); red stars denote comparisons with p-values reaching significance and also qualifying as a true discovery. Results are shown using box-and-whiskers plots. Expression levels are shown as medians normalized to RPLPO endogenous control. (d) Chromatogram showing sequences of a complete bisulfite conversion (Refseq, two unmethylated alleles), completely unconverted cytosines (two methylated alleles, found in control cases,), combination of converted and unconverted cytosines (one methylated and one unmethylated alleles, found in patients from both C9orf72 and sporadic groups), and completely converted cytosines (two unmethylated alleles, found in patients from C9orf72 group only) at position chr. 14:94,849,201 (h19). (e) Graph showing that SERPINA1 is largely unexpressed in most human tissues, except in liver and blood. Information has been retrieved from https://gtexportal.org/home/gene/SERPINA1 in July 2017.

After bisulfite DNA amplification of the SERPINA1 region of interest, we observed fully unconverted methylated cytosines in all but one control samples, whereas fully converted, unmethylated cytosines were observed in C9orf72 expansion carriers (Figure 3d). Both methylation states were found in eight C9orf72 expansion carriers and non-carriers.

To test whether SERPINA1 is commonly expressed in mouse brain, we investigated several existing public data sets that demonstrate SERPINA1 is specifically expressed in the liver and blood (Fig. 3e; https://gtexportal.org/home/gene/SERPINA1, accessed July 2017). SERPINA1 is largely unexpressed in both the human (Fig. 3e) and mouse brain (online resource Fig. 1); mouse homologue of the SERPINA1 transcripts (Serpina1a through Serpina1f) were measured in only approximately four E18 mouse brain cells in the 1.3million cell 10× Genomics single-cell expression data set (18,500 reads per cell; https://support.10xgenomics.com/single-cell-gene-expression/datasets/1.3.0/1M_neurons; accessed July 2017),while SERPINA1 transcripts were measured in zero E18 mouse brain cells in the 9k single-cell expression dataset (42,000 reads per cell; https://support.10xgenomics.com/single-cell-gene-expression/datasets/1.3.0/neuron_9k).

We also attempted to replicate twelve other genes that were both differentially expressed and methylated, but were not in common across brain regions or disease groups (online resource Table 1). Three of the twelve genes could not be assessed because of very low expression. Four genes were differentially expressed in our expanded cohort (online resource Fig. 2a-2d). As expected, DCAF4L1 (p = 0.017) and VAMP5 (p = 0.026) were over expressed in the cerebellum for C9orf72 expansion carriers when compared to controls, and COL27A1 (p = 0.017) and PODNL1 (p = 0.003) were over expressed for C9orf72 expansion carriers in the frontal cortex, when compared to controls. DCAF4L1 and COL27A C9orf72 expansion carriers were also over expressed compared to sporadic cases in cerebellum and frontal cortex, respectively (p = 0.026 and 0.035; online resource Fig. 2a and 2c), while VAMP5 and PODNL1 sporadic cases were also over expressed when compared to controls in cerebellum and frontal cortex, respectively (p = 0.040 and 0.011; online resource Fig. 2b and 2d). The other five genes (H19, CDKN1A, CIRBP-AS1, TJP3, and LRG1) did not reach significance in our expanded validation cohort.

Genes exhibit alternative cassette exon splicing and methylation

We also tested five genes that exhibited significantly different alternative cassette exon splicing and methylation: FBLN2, C14orf80, IFI27, TRPM2, and NAGPA. Two of the genes, C14orf80 and TRPM2, replicated. As expected, both C14orf80 and TRPM2 demonstrated significantly less inclusion of exons 6 (p = 0.001; transcript variant 1, NM_001134875.1; online resource Table 2 and online resource Fig. 2e) and 27 (p = 0.007; transcript variant 1, NM_003307.3; online resource Table 2 and online resource Fig. 2f), respectively, in the frontal cortex for C9orf72 expansion carriers, when compared to controls. Exon 6 of C14orf80 was also included less in sporadic cases (p = 0.042; online resource Fig. 2e), when compared to controls.

Common biological functions in ALS

Considering the distinct methylomes identified in c9ALS and sALS, we assessed whether differentially-expressed genes found with DMCs upstream of TSS or within the gene-body in each group (1) have similar or distinct cellular and molecular gene functions and (2) are part of similar or distinct biological networks. Using Ingenuity Pathway Analysis (IPA) software, we evaluated the cellular and molecular functions of differentially-expressed genes (p <0.05) that also have DMCs, and the functions of their associated biological networks (online resource, Tables 3-6) in c9ALS and sALS frontal cortex and cerebellum. Of the top five most significantly affected cellular and molecular gene functions in each group, 2-4 were shared with at least one other brain region or disease group, including “cell cycle” and cellular function and maintenance, movement, morphology, assembly, and organization.

Top affected biological networks were identified in each disease group and brain region based on the significance of aberrant methylation upstream of genes involved in associated networks. The commonly top-affected network functions, considering all affected genes together, were highly similar to the individual molecular and cellular gene functions previously identified (online resource, Tables 3-6). Of the top ten hyper-and hypo-methylated genes for each disease group, 60% and 40% were part of the top five affected networks for c9ALS and sALS across both brain regions, respectively (online resource, Table 7). These networks are connected to other networks containing genes with methylation and expression changes. Therefore, a change in a gene can affect not only members of the network to which it belongs, but also members of connected networks, expanding the impact range of a given defect (online resource, Tables 4 and 6).

SERPINA1 contributes to many critical cellular and molecular functions, and interacts with ALS-associated molecules

The most significant network in c9ALS cerebellum includes SERPINA1, and is also part of networks 2 and 4 (Fig. 3a, online resource Table 4 and 6). All three networks are interconnected through SERPINA1 and have “cellular movement” as a common biological function. Within network 1, there are seven molecules previously associated with ALS, including FUS, VEGF, histones H3 and H4, ACTIN, estrogen receptor, and NFκB (Fig. 4). The network also includes KHDRBS1, a gene encoding SAM68, which is sequestered by CGG-containing RNA foci in Fragile X-associated tremor/ataxia syndrome (FXTAS). Among members of the same network, CDC25B, FBLN2, and PPP2R2A were part of the top ten hyper-methylated genes, while ALDH1A1, COL4A2, LIG1, and PRMT6 were part of the top hypo-methylated genes. Five additional genes in network 1 (FBLN2, GSN, H19, POLE, and TNFAIP8) were differentially expressed and differentially methylated within the region expected to modulate transcription (Fig. 2a-2d, part of light-blue protruding pie slices).

Fig. 4. Schematic representation of network 1 in c9ALS cerebellum after conducting IPA network analysis of genes with expression changes and DMCs.

Nodes in green represent hypomethylated genes with expression p-value<0.05, and nodes in red represent hypermethylated genes with expression p-value<0.05. Nodes in gray represent upstream regulators. Some of the top 10 hypermethylated and 10 hypomethylated genes identified in our IPA analysis were part of network 1 (dark blue font). Genes or molecules associated with ALS (red font) and another neurodegenerative disease (purple font) are also part of network 1. Network 1 also include genes with differential expression (p-value<0.05; |log2FC|≥1) or alternative cassette exon splicing (FDR<0.05; |dl|≥0.1) and DMCs in regions predicted to modulate transcription (light blue font). Of note, FBLN2 was both part of the top 10 hypermethylated group and the alternative cassette exon splicing (FDR<0.05; |dl|≥0.1) and DMCs in regions predicted to modulate transcription group. FBLN2 is shown in light blue font.

Key ALS-associated genes appear to accumulate transcriptome and methylome changes

While additional studies are necessary to clarify, we nonetheless observed transcriptome and methylome changes in key ALS genes. We analyzed DNA methylation and transcriptome defects in 172 candidate genes (76 ALS-associated and 96 neuro-associated genes) carefully selected based on their involvement in ALS and other neurological diseases (full list of genes and complete results available in online resource, Tables 8-11). We found that, while transcriptome defects were not common in these selected genes, except for cassette exons in c9ALS cerebellum, DNA methylation events were much more frequent (Table 1). We identified an upstream DMC in C9orf72 for both c9ALS frontal cortex and cerebellum. We also identified upstream DMCs for TARDBP in both c9ALS and sALS, in frontal cortex and cerebellum. RANGAP1 also has upstream DMCs in c9ALS and sALS frontal cortex, and in sALS cerebellum, as well as in the RANGAP1 gene body in c9ALS frontal cortex (Table 1). IGHMBP2, involved in Charcot-Marie-Tooth disease, and KHDRBS1, involved in FXTAS, were also found to have upstream DMCs in frontal cortex and cerebellum in both c9ALS and sALS. Gene body DMCs were also found in IGHMBP2 for c9ALS and sALS frontal cortex and cerebellum, further supporting a role for this gene in ALS (Table 1).

Table 1. ALS and neuro-associated genes with DMCs and/or transcriptome defects in ALS.

Frontal Cortex		DE p<0.05 |log2FC|≥1	CE FDR<0.05 |dl|≥0.1	IR FDR<0.05 |dl|≥0.1	APA FDR<0.05 |ΔdPDUI|≥0.2 |dPDUI|≥0.2	Upstream/promoter q-value<0.01 methyl diff >25%	Gene Body q-value<0.01 methyl diff >25%
c9ALS	ALS-associated	PRPH, SERPINH1, SUSD2	ATXN2	n/a	CYP27A1, DNMT3A, DNMT3B	C9orf72, CAMTA1, RANGAP1, SPAST, TARDBP, UNC13A	ATXN2, CAMTA1, GLE1, RANGAP1, VAPB
	neuro-associated	PFN1P2, TBPL2	MAPT	n/a	KHDRBS1	ADARB1, CACNA1C, ELAVL4, ENOX1, IGHMBP2, KHDRBS1, PLA2G6, PSEN2, RALY, SRSF1	ADARB1, ELAVL4, ENOX1, IGHMBP2, NOVA2, PARK2, PARK7, PLA2G6
sALS	ALS-associated	C7orf57	EWSR1, SLC1A2	n/a	DNMT3A, DNMT3B, SUSD2	C21orf2, CAMTA1, DAO, RANGAP1, TARDBP	ATXN2, C21orf2, CAMTA1, MOB3B
	neuro-associated	BCL2L2-PABPN1, TBPL2	HNRNPA1L2	n/a	ATXN2L, CACNA1C, ENOX1, HNRNPA1L2, KHDRBS1, RBFOX1, SRSF1 0	ADARB1, ADARB2, ATXN8OS, CACNA1C, CELF1, ENOX1, IGHMBP2, JPH3, KCNMA1, KHDRBS1, NRXN1, NRXN3, PARK2, PARK7, PURA	ADARB2, BCL2L2-PABPN1, CACNA1C, CELF1, CELF5, ENOX1, IGHMBP2, NRXN2, PABPN1, PARK2, PSEN1
Cerebellum
c9ALS	ALS-associated	C9orf72, IFIH1, SERPINH1	ANG, ATXN2, C21orf2, CAMK1G, CAMTA1, DCTN1, DNMT1, ELP3, EWSR1, FUS, KIFAP3, MATR3, SCFD1, SPAST, SS18L1, TAF15, SUPT4H1, TIA1, VAPB, VEGFA	DNMT1, EWSR1, SIGMAR1, TAF15	ALS2, ATXN2,CCNF, LMNB1, VAPB	C9orf72, FUS, HNRNPA2B1, OPTN, TARDBP	MOB3B, ZNF512B
	neuro-associated	n/a	ADARB1, ATXN3, ATXN7, CACNA1C, CELF1, ELAVL2, FBXO7, GIGYF2, GRIN1, HNRNPA1L2, HNRNPH1, IGHMBP2, KCNMA1, MBNL1, MBNL2, NOVA1, NRXN1, NRXN2, NRXN3, PRMT10, PTBP2, RBFOX1, RBFOX2, RBMXL1, SRSF10, SRSF11	GRIN2C, HDAC6, RBMX, SRSF1	ADARB2, ATXN7, KCNMA1, LMNA, LRRK2, RBFOX1	ADARB2, ATXN1, ATXN10, CACNA1C, IGHMBP2, KHDRBS1, NRXN2, PLA2G6, SLC1A6, SNCA	ADARB2, DMPK, IGHMBP2, NRXN2, PLA2G6
sALS	ALS-associated	n/a	GRIA2	n/a	CAMTA1, ITPR2, SPAST, STX12, TUBA4A	CAMTA1, CHCHD10, PFN1, RANGAP1, SERPINH1, SPAST, TARDBP, UNC13A	CAMTA1, SLC1A2, UNC13A
	neuro-associated	ATXN8OS	CELF1, KCNMA1, MBNL1	HNRNPH1	ADARB2, KCNMA1, LRRK2, PABPN1L, PARK7, PINK1, RBFOX1	ADARB1, ADARB2, APP, ATXN10, HTT, IGHMBP2, JPH3, KCNMA1, KHDRBS1, MBNL2, NEFL, NOVA2, NRXN3, PABPN1L, PLA2G6, PRMT10, SNCB	ADARB2, APP, ATXN7, CACNA1C, IGHMBP2, JPH3, KCNMA1, NRXN3, PARK2, PLA2G6, SPG7

DE: differential expression; CE: alternative cassette exon splicing; IR: retention of intronic region; APA: alternative polyadenylation.

Discussion

The methylation data presented herein provides a comprehensive methylation overview between controls, sALS, and c9ALS patients across the frontal cortex and cerebellum. We also combined it with our transcriptome data to identify and validate new genes that may be heavily involved in ALS. To our knowledge, no other report broadly addresses methylation and transcriptome effects in sALS and c9ALS across two important brain regions. While the sample sizes for our methylation data are small, the general trends are still valuable, and assessing both overlapping and distinct DMCs, combined with transcriptome data, will help clarify what drives sALS vs c9ALS. Our data also reports individual cytosine methylation events, where most studies combine methylation events across large regions. Characterizing individual methylation events will be critical to determining what role each methylation event plays in gene regulation, as some events are likely far more influential than others.

Increasing evidence suggests cerebellar dysfunction plays a role in ALS

The broad methylation and expression overview we presented suggests generally distinct, but overlapping methylome profiles between sALS and c9ALS, where disease phenotypes may converge on common biological mechanisms or pathways, and also supports increasing evidence that the cerebellum plays a role in ALS[33,12]. We identified top DMCs and differentially expressed genes for sALS and c9ALS when compared to controls that distinguish between groups, and we identified approximately 800 more DMCs in the cerebellum of sALS cases compared to c9ALS patients—nearly twice as many DMCs as was found in sALS frontal cortex (Fig. 1c and 1d). There were many overlapping DMCs and differentially expressed genes, however. These data support cerebellar involvement in disease, and further suggest that, while sALS and c9ALS methylome and expression profiles are largely distinct, there may be some genes in common that play a central role in disease.

The cerebellum has not generally been considered important in ALS since there is no substantial neuronal loss[32,39], but cell death may not be the only mechanism driving ALS phenotypes. For example, cellular functions may be inhibited without actually killing the cells, and the excessive number of DMCs in the cerebellum may simply indicate general gene dysregulation in the cerebellum that might modulate downstream effects in other brain regions.

Aberrant DMCs may be a common link amongst sALS cases

While it is unsurprising that sALS had many more DMCs than c9ALS, given its sporadic nature, it was unexpected that many common DMCs were within sALS. These data suggest that while there is currently no known common cause for sALS, there may actually be a core set of biological pathways that explain this group of patients (Fig. 1k and 1l). More studies are necessary to fully assess these common features, but these results may have big implications in sALS disease therapies.

Both proximal and distal DMCs may affect the transcriptome

Surprisingly, approximately 95% of all DMCs were located >3kb upstream of the affected gene's TSS, providing some support that distal DMCs may regulate transcriptome modifications. While proximal epigenetic modifications are the most likely cause for a given gene's altered transcription, DMCs that modulate transcription have been reported as distant as 1Mbp[37], which our data further supports. Just as a few critical DMCs may be more influential than an abundance of trivial DMCs to modulate expression, methylation of key sites located more than 3kb away from the TSS may be as imperative as changes closer to the TSS. Precisely, our results support a role for DNA methylation at distant regulatory regions, such as at enhancer sites, which are responsible for transcription regulation through transcription factor-binding[30].

Epigenetics is still a relatively young field, however, and additional work is necessary to clarify what role distant DMCs may have on transcriptome regulation. Similarly, most work to date has focused on DMCs within a region as a whole, rather than characterizing how influential individual DMCs (proximal or distal) are on a gene's transcription regulation; more work is needed to clarify the individual DMC contributions.

Brain cells may not tolerate increased SERPINA1 expression

While the comprehensive methylation and transcription data are individually valuable, their true utility is best exploited together as a combined “multi-omics” analysis. As described, we filtered our candidate gene list by including only those that were both differentially expressed and methylated, and identified SERPINA1 as a potentially critical disease gene. Our data suggests SERPINA1 is involved across both sporadic cases and C9orf72 expansion carriers in frontal cortex and cerebellum (Fig. 3a), where overexpression is particularly pronounced in C9orf72 expansion carriers (Fig. 3b and 3c). While most genes are expressed in easily measureable abundance across body tissues, SERPINA1 expression is strictly down regulated throughout the body (including all major brain regions) except in the liver and blood (Fig. 3e). We further verified this finding in public 10x Genomics single-cell expression experiments in E18 mouse brain (online resource Fig 1). We verified this finding in both the 1.3 million and 9,000 single-cell 10x Genomics mouse data sets with 18,500 and 42,000 reads per cell, respectively. While the single-cell sequencing experiments may not sequence enough reads to accurately quantify the low SERPINA1 levels, it does emphasize how tightly SERPINA1 is regulated in the brain. Given how strictly SERPINA1 is regulated throughout the brain and body, over expression may be detrimental to normal neuronal function.

SERPINA1 encodes a serine protease inhibitor that interacts primarily with elastase, but also with trypsin, chymotrypsin, thrombin, and bacterial proteases[9]. Our previous data demonstrated members of the same gene family are also differentially expressed, namely SERPINH1 (increased in c9ALS frontal cortex, p = 0.0017, and cerebellum, p = 0.0250) and SERPINA3 (increased in c9ALS frontal cortex, p = 0.0085, and trending towards significance in cerebellum, p = 0.0527)[33]. All three genes are serine protease inhibitors, and SERPINA3 has also been implicated in Alzheimer's[18,19] and Parkinson's disease[41]. Given that all three genes have increased expression may indicate serine protease inhibitors are obstructing neuronal function in disease.

DCAF4L1, VAMP5, COL27A1, and PODNL1 are also implicated in disease

Differential expression for genes DCAF4L1, VAMP5, COL27A1, and PODNL1 was also previously implicated in our RNASeq data, and we found DMCs in regions predicted to regulate their expression. Differential expression replicated for DCAF4L1, VAMP5, COL27A1, and PODNL1 (online resource Fig. 2a-2d). DCAF4L1 and VAMP5 were differentially expressed in the cerebellum, while COL27A1 and PODNL1 were implicated and validated in the frontal cortex. Little is known about DCAF4L1 and PODNL1, but VAMP5 is associated with muscle formation (myogenesis) and vesicle tracking[46]. COL27A1 is in the fibrillar collagen family[29] which is important in tissue growth and repair. COL27A1 is also highly expressed in the brain, when compared to other body tissues, and most highly expressed in the cerebellum, compared to other brain tissues (https://gtexportal.org/home/gene/COL27A1, accessed July 2017)[27]. Nagase et al. further demonstrated that COL27A1 (referred to as KIAA1870 in their paper) has a primary function in cellular structure and motility[27]. Collagen binding was also one of the top upregulated molecular functions identified in our RNAseq Gene Ontology study for c9ALS frontal cortex[33].

Methylation modifications may be driving TARDBP and RANGAP1 dysregulation

We identified TARDBP and RANGAP1 as two commonly differentially methylated ALS associated genes. TARDBP encodes TDP-43, a key RNA-binding protein known to aggregate in ALS neurons after mislocalizing from the nucleus to the cytoplasm[28]. RanGAP physically interacts with C9orf72 expanded RNA[10] and is believed to be a key regulator of nucleocytoplasmic transport in c9ALS[47]. Since approximately 97% of ALS cases have TDP-43 pathology[28], future studies should assess whether modulating aberrant TARDBP methylation has functional consequences on downstream TDP-43 cytoplasmic aggregation. Also, while alterations in RanGAP function have been reported in c9ALS[47], our data suggest an unpredicted role for RanGAP in sALS as well, both in frontal cortex and cerebellum. In addition to these ALS associated genes, we also found DMCs in IGHMBP2, a gene associated with other neurological diseases, in both cerebellum and frontal cortex tissue from both c9ALS and sALS patients. IGHMBP2 encodes the DNA-binding protein SMBP2, a transcription regulator that predominantly localizes to the cytoplasm of neurons and associates with ribosomes. SMBP2 is an ATP-dependent 5′ to 3′ helicase, responsible for unwinding RNA and DNA duplexes. Our analyses of ALS and neuro-associated genes also further support our findings from the upstream and gene body analyses. Our data points to a complex biological network as being affected in both c9ALS and sALS in which genes found with both altered transcription and aberrant methylation can modulate multiple interconnected genes and molecules within individual networks, all together modulating the biological function of the network as a whole.

Conclusions

ALS is a devastating and fatal disease with no effective therapy to prevent, stop or decelerate neurodegeneration, making comprehensive “mutli-omic” approaches critical to identifying key molecular players in disease etiology. Our study is the first to demonstrate that c9ALS and sALS patients have distinct methylomes, with members within each disease group sharing thousands of DMCs, and the first to combine methylome and expression data for a “multi-omic” analysis in ALS. Our results are particularly important for genetically unexplained sALS cases, as there is currently no known common cause. Aberrant methylation may be key to understanding the disease.

Our data also strongly implicates SERPINA1 in both c9ALS and sALS, as it was differentially methylated and expressed in both disease groups across the frontal cortex and cerebellum. It also interacts with many ALS-associated genes and is strictly down regulated throughout the body, under normal conditions. We propose that functional studies in mouse and cell models will be necessary to determine whether SERPINA1 is driving ALS phenotypes, or whether its dysregulation is one of many downstream effects of the underlying cause.