Meta-analysis of genetic modifiers reveals candidate dysregulated pathways in Amyotrophic Lateral Sclerosis

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a neurodegenerative disease that has significant overlap with frontotemporal dementia (FTD). Mutations in specific genes have been identified that can cause and/or predispose patients to ALS. However, the clinical variability seen in ALS patients suggests that additional genes impact pathology, susceptibility, severity, and/or progression of the disease. To identify molecular pathways involved in ALS, we undertook a meta-analysis of published genetic modifiers both in patients and in model organisms, and undertook bioinformatic pathway analysis. From 72 published studies, we generated a list of 946 genes whose perturbation (1) impacted ALS in patient populations, (2) altered defects in laboratory models, or (3) modified defects caused by ALS gene ortholog loss of function. Herein, these are all called modifier genes. We found 727 modifier genes that encode proteins with human orthologs. Of these, 43 modifier genes were identified as modifiers of more than one ALS gene/model, consistent with the hypothesis that shared genes and pathways may underlie ALS. Further, we used a gene ontology-based bioinformatic analysis to identify pathways and associated genes that may be important in ALS. To our knowledge this is the first comprehensive survey of ALS modifier genes. This work suggests that shared molecular mechanisms may underlie pathology caused by different ALS disease genes. Surprisingly, few ALS modifier genes have been tested in more than one disease model. Understanding genes that modify ALS-associated defects will help to elucidate the molecular pathways that underlie ALS and provide additional targets for therapeutic intervention.

Introduction

Amyotrophic Lateral Sclerosis (ALS) is a progressive disease that results in selective degeneration and death of upper (cortical) and lower (spinal) motor neurons. First described by Jean-Martin Charcot in 1869 (Rowland, 2001), ALS is characterized by muscle weakness, paralysis, respiratory failure, and death typically within 3–5 years of symptom onset. Within the past two decades, over 20 genes have been identified and/or implicated in ALS (Baker et al., 2006; Brenner et al., 2016; Chaussenot et al., 2014; Chen Y. Z. et al., 2004; Chesi et al., 2013; Chow et al., 2009; Cirulli et al., 2015; Couthouis et al., 2012; Cruts et al., 2006; Daoud et al., 2012; DeJesus-Hernandez et al., 2011; Deng et al., 2011; Deng et al., 1993; Elden et al., 2010; Figlewicz et al., 1994; Freischmidt et al., 2015; Greenway et al., 2006; Gros-Louis et al., 2004; Hutton et al., 1998; Johnson et al., 2014a; Johnson et al., 2010; Johnson et al., 2014b; Kabashi et al., 2008; Kenna et al., 2016; Kim H. J. et al., 2013; Kwiatkowski et al., 2009; Leblond et al., 2014; Leung et al., 2004; Maruyama and Kawakami, 2013; Millecamps et al., 2014; Mitchell et al., 2010; Munch et al., 2005; Munch et al., 2004; Nishimura et al., 2004; Parkinson et al., 2006; Pensato et al., 2015; Rademakers and van Blitterswijk, 2014; Renton et al., 2011; Rosen et al., 1993; Skibinski et al., 2005; Skvortsova et al., 2004; Smith et al., 2014; Sreedharan et al., 2008; Sreedharan and Brown, 2013; Takahashi et al., 2013; Teyssou et al., 2013; Teyssou et al., 2014; Ticozzi et al., 2011; Van Deerlin et al., 2008; Vance et al., 2009; Wu et al., 2012; Yang Y. et al., 2001). Together, mutations in superoxide dismutase 1 (SOD1), TAR DNA binding protein (TARDBP), fused in sarcoma (FUS), and chromosome 9 open reading frame 72 (C9orf72) account for approximately 60–70% of ALS cases with a family history. Other genes, including VAMP-associated protein B (VAPB), valosin-containing protein (VCP), and optineurin (OPTN), account for 30–40% of familial cases. The proteins encoded by these genes are involved in a variety of pathways, including oxidative stress (Barber et al., 2006), protein aggregation (Bruijn et al., 1998), and neuroinflammation (Hooten et al., 2015). However, despite the varied roles of these proteins in healthy cell function, disease alleles of the aforementioned genes can lead to ALS. Even though we have made progress on understanding some aspects of ALS, we do not understand how or why mutations in functionally diverse proteins can cause what appears to be a single disease.

Insights into ALS pathological mechanisms came from the discovery that mutations in a subset of these genes can also cause frontotemporal dementia (FTD), with characteristic degeneration of frontal and temporal lobe neurons (Ratnavalli et al., 2002). ALS and FTD share many pathological hallmarks, including ubiquitinated inclusions, which have been observed in lower motor neurons and cortical neurons of patients with ALS. Furthermore, approximately 50% of ALS patients develop FTD-like symptoms and around 40% of FTD patients develop ALS-like symptoms (Ferrari et al., 2011; Ji et al., 2017; Lomen-Hoerth et al., 2002; Lomen-Hoerth et al., 2003; Strong, 2008). These observations suggest that ALS and FTD are related and may share pathways leading to neurodegeneration (Arai et al., 2006; Leigh et al., 1991; Liscic et al., 2008; Mackenzie and Feldman, 2005). One strategy that can be used to delineate shared pathways, is to find “genetic modifiers” or “modifier genes” of ALS and FTD genes, which can reveal pathological mechanisms.

Broadly defined, “modifier genes” are genes with alleles that ameliorate or exacerbate defects caused by an allele of another gene. Modifier genes, in patients, may influence clinical presentation of disease including disease onset, severity, penetrance, or progression. Classical genetic studies in model organisms have extensively used modifier gene analysis to dissect function and dysfunction, contributing to our understanding of neurodegenerative diseases (Alexander et al., 2014; Dimitriadi and Hart, 2010; Gama Sosa et al., 2012; Plantie et al., 2015; Therrien and Parker, 2014; Verbandt et al., 2016). Large scale forward genetic screens for modifiers are possible in small, genetically tractable organisms, such as S. cerevisiae, C. elegans, and D. melanogaster (Chen X. and Burgoyne, 2012; Sin et al., 2014). These can yield unexpected insights into mechanisms and complement hypothesis-driven studies. Most animal models of ALS compare the consequences of expressing a human protein containing the disease mutation versus the wild type form of the protein. These models have been used to identify genetic modifiers of ALS-associated defects and we surveyed their results. Also, ALS alleles may cause loss of function, which may contribute to disease pathology. Therefore, we surveyed the results of previous studies focused on identifying modifiers of either disease models or ALS-gene ortholog loss of function.

Further, ALS modifier genes have also been identified in human populations and may help explain variation in clinical presentation or disease progression. The site of onset (bulbar or spinal), age of onset, progression rate, and level of cognitive impairment can differ between patients even within the same family (Swinnen and Robberecht, 2014). The variability observed in ALS patients may be, in part, due to a result of different alleles of modifier genes that affect progression, penetrance or onset—even if these modifier alleles do not cause disease per se. Risk genes are also of interest, as they may reveal pathways critical for disease, even if risk genes are neither necessary nor sufficient to cause disease. Genome wide association studies (GWAS) and linkage analysis in humans with ALS have been used to identify genetic modifiers (Giess et al., 2002; Gros-Louis et al., 2004; Lee Y. B. et al., 2013).

Here, we undertook a comprehensive literature search and identified 946 genes that act as modifiers of ALS-associated defects in S. cerevisiae, C. elegans, D. melanogaster, M. musculus, or human patients. As shared mechanisms may underlie ALS, we used a gene ontology bioinformatics approach to identify pathways pertinent to disease. This bioinformatic analysis focused on 727 modifier genes that are orthologous to human genes, some of which have been identified in human studies. The results suggest that shared pathways may underlie ALS, regardless of the disease gene involved.

Experimental Methods

Literature Search

We searched the literature in PubMed from September 7, 2016 - December 31, 2016 and identified studies that reported modifier genes in ALS models or modifiers of ALS ortholog loss of function (Fig. 1). Specifically, we examined papers in PubMed reporting genetic modifiers of SOD1, TDP43, C9orf72, FUS, or VAPB. For SOD1, TDP43, C9orf72, and FUS; two independent co-authors searched the literature. The literature review included, but was not limited to, genome-wide screens and candidate genes reported to modify phenotypes in S. cerevisiae, C. elegans, D. melanogaster, M. musculus, and cell culture models. Additionally, we searched for genetic modifiers of ALS in patients, which was reviewed by Ghasemi and Brown (Ghasemi and Brown, 2017). A database was assembled in Microsoft Excel with the NCBI GeneID, modifier gene name, human ortholog name, ALS model used, type of screen (RNAi knockdown, genome-wide screen) and impact of modifier on phenotype. The total number of modifier genes (946) corresponds to all modifier genes identified in the literature survey; we did not reexamine this list to identify and eliminate orthologous genes independently identified in different model organisms, which would modestly reduce this number. Additionally, using OMIM (Online Mendelian Inheritance of Man) a list of genes known to cause ALS and/or FTD was compiled. At least 2 independent coauthors checked each ALS gene after the database was assembled to ensure the accuracy. All data used in the analysis are included in the manuscript and supplemental files.

Fig. 1.

Schematic representation of the workflow used to compile and analyze the list of genetic modifiers of ALS.

Human Ortholog Identification

Human orthologs of modifier genes found in model organisms were identified using best match similarity with BLAST (NCBI) at blast.ncbi.nlm.nih.gov/Blast.cgi based on protein sequences. If there was more than one best match, then up to three were reported in the “other orthologs” column in Supplemental File 1. When genes with identical statistical scores were called as best match, both genes were included in the bioinformatics analysis. For example, Hbr98DE gene is an ortholog of both hnRNPA1 and hnRNPA2B1. Human ortholog identification was verified with DIOPT (http://www.flyrnai.org/diopt) (Hu et al., 2011). If no human ortholog was found, the modifier gene was not included in bioinformatics analysis presented herein.

Gene Ontology Bioinformatic Analysis

Gene ontology (GO) bioinformatic analysis was performed independently for lists of human genes and/or orthologs of modifier genes identified in other species. GO terms that describe Biological Processes are tested to determine if these were over-represented in the curated gene lists, compared to the rest of the transcriptome, using a hypergeometric test implemented in the GOstat package (Falcon and Gentleman, 2007). GO terms with a p-value less than 0.05 after Bonferroni correction were considered overrepresented. In addition, a list of modifier genes associated with more than one ALS genes was assembled and independently subjected to GO analysis. The supplemental files containing lists of modifier genes and other data use in the bioinformatic analysis are available at https://doi.org/10.26300/7asw-k867.

Results

To identify modifier genes associated with ALS, we searched the PubMed literature database at the National Center for Biotechnology Information (NCBI) for modifiers of SOD1, TDP43, FUS, C9orf72, VAPB, and other ALS genes. In total, 72 studies were found reporting modifier genes 1) in ALS models, 2) in human patients, or 3) for loss of function alleles of ALS gene orthologs. The resulting list of genes is available in Supplemental File 1. Here, we provide an abbreviated background for ALS genes that served as the basis for our search, including a brief description of ALS models used in modifier gene studies. For each ALS gene, gene ontology bioinformatic pathway analysis was undertaken and pathways that were enriched in gene ontology analysis are discussed.

SOD1

In 1993, the discovery that point mutations in superoxide dismutase 1 (SOD1) cause ALS revolutionized the field (Deng et al., 1993; Rosen et al., 1993). SOD1 is an evolutionarily conserved, ubiquitously expressed protein that catalyzes breakdown of superoxide radicals into hydrogen peroxide and water. As the second most common gene whose mutation causes familial ALS (fALS), mutations in SOD1 account for approximately 20% of fALS cases and 5% of sporadic ALS (sALS) cases (Kaur et al., 2016). Over 100 mutations have been identified in SOD1, and almost all disease alleles are dominant in patients. From many studies, it seems likely that disease alleles cause a toxic gain of function, but loss of function may contribute to disease pathology (Bruijn et al., 1998; Saccon et al., 2013).

Two non-exclusive hypotheses for SOD1-associated ALS motor neuron degeneration dominate the field: the aggregation hypothesis and the oxidative stress hypothesis. Mutant SOD1 protein aggregates in the cytosol of SOD1 ALS patient cells are thought to confer toxicity or reduce SOD1 enzymatic activity (Stieber et al., 2000; Watanabe M. et al., 2001). Early studies in mice supported a toxic gain of function hypothesis, as SOD1 null mice do not exhibit ALS-like pathology and overexpression of mutant human SOD1 resulted in reduced enzymatic activity (Gurney et al., 1994; Reaume et al., 1996). However, how SOD1 mutations cause ALS is still debated. More recent studies suggest that SOD1 loss of function also contributes to ALS dysfunction and degeneration. One possible mechanism is that mutations in SOD1 cause loss of function by aggregation, causing abnormal buildup of superoxide radicals or hydrogen peroxide, the substrate and byproducts of SOD1 action, respectively (Beckman et al., 1993). SOD1 activity is decreased in patients with ALS, suggesting that SOD1 loss of function may contribute to pathology (Rosen et al., 1993; Watanabe Y. et al., 1997). SOD1-mediated motor neuron death may be caused by a combination of the loss and gain of function consequences of patient alleles of SOD1 (Sahin et al., 2017).

ALS models have been created by overexpressing mutant human SOD1 protein in numerous model organisms and comparing the deleterious consequences of the mutant protein to the consequences of overexpressed wild type human SOD1 protein. Two of the most frequently used patient alleles in SOD1 ALS models are missense mutations that result in a glycine to arginine substitution at position 85 (G85R) or a glycine to alanine substitution at position 93 (G93A). As SOD1 loss of function may also contribute to ALS-associated defects, modifier genes that suppress defects associated with SOD1 loss of function alleles are also of interest. In the literature, we found 33 articles that, in combination, yielded 164 modifier genes in either SOD1 ALS model animals or animals lacking SOD1 ortholog function. These are listed in Supplemental File 1 in the SOD1 tab (Allodi et al., 2016; Bahadorani et al., 2013; Boccitto et al., 2012; Chloupkova et al., 2003; Couillard-Despres et al., 1998; Dadon-Nachum et al., 2015; Dobrowolny et al., 2008; Giess et al., 2002; Hetz et al., 2009; Jablonski et al., 2015; Kieran et al., 2007; Kumimoto et al., 2013; Lambrechts et al., 2003; Lapinskas et al., 1995; Liu et al., 2002; Lobsiger et al., 2005; Lorenzl et al., 2006; Lu et al., 2009; Lunn et al., 2009; Marden et al., 2007; Ohta et al., 2016; Pitzer et al., 2008; Reyes et al., 2010; Riddoch-Contreras et al., 2009; Sharp et al., 2008; Silva et al., 2011; Strain et al., 1998; Teuling et al., 2008; Turner et al., 2014; Wang J. et al., 2009; Yang Y. S. et al., 2009; Zhai et al., 2005).

We hypothesized that shared pathways might link SOD1 modifier genes. To identify these connections, we undertook gene ontology enrichment analysis of the assembled SOD1 modifier gene lists and identified enriched Gene Ontology (GO) pathways. Initially, this analysis was complicated by the diversity of model organisms used for modifier gene identification. To facilitate cross-species comparisons and bioinformatic analysis, the closest human ortholog of each modifier gene was identified based on amino acid similarity using reciprocal BLAST analysis. Proteins that lacked a human ortholog were excluded from bioinformatic analysis. This bioinformatic analysis revealed enrichment of pathways integral for endogenous SOD1 function: “response to reactive oxygen species” (GO:0000302) and “regulation of oxidative stress-induced intrinsic apoptotic signaling” (GO:1902175). The complete SOD1 pathway analysis is presented in Supplemental File 2, SOD1 tab, and top hits are illustrated in Fig. 2A (pathways with odds ratio > 5). The most enriched pathway for SOD1 modifier genes was “neurotransmitter reuptake” (GO:0098810). The SOD1 modifier genes from the literature survey that led to bioinformatic analysis identification of “neurotransmitter reuptake” are shown in Fig. 6.

Fig. 2.

GO terms enriched for SOD1, TDP43 and FUS A) Diagram (top) shows relationship between GO terms enriched in gene ontology analysis of SOD1 genetic modifiers. Arrows indicate related terms that are “nested” inside a broader category. GO terms above odd ratios of 5 or greater are listed (below); the most highly enriched genes are at the top of the list. Darker red hues indicate a higher odds ratio (the magnitude of enrichment). For example, “neurotransmitter reuptake (GO:0098810)” has an odds ratio of 29.86 and is shown in red. This indicates that we observe more genes associated with “neurotransmitter reuptake” in the list of SOD1 genetic modifiers than expected. Though it is enriched in our dataset, “intrinsic apoptotic signaling pathway (GO:0097193)” has an odds ratio of 5.35 and is shown in white. In this case, we still observe more genes than expected with “intrinsic apoptotic signaling pathway” in our dataset, but not to the same extent as “neurotransmitter reuptake” genes. B) GO terms enriched in TDP43 modifiers, presented as in panel A. C) GO terms enriched in FUS modifiers, presented as in panel A.

Fig. 6.

Modifier genes in the most enriched GO term for ALS genes. The most enriched GO term for each ALS genes is listed. Each pie chart contains the names of all human genes (or orthologs) that were associated with the top GO term. Genes are grouped and color coded based on originally reported perturbation of the modifier gene and their impact on ALS-associated defects. For example, knockdown of the NUP98 D. melanogaster ortholog ameliorated ALS-associated defects in a D. melanogaster C9orf72; this gene was classified as “downregulation suppressed”.

TDP43

Ubiquitinated inclusions in affected patient neurons are a frequent pathological hallmark of ALS (Arai et al., 2006; Leigh et al., 1991; Ling et al., 2013; Liscic et al., 2008; Mackenzie and Feldman, 2005; Maekawa et al., 2009; Neumann et al., 2006). In 2006, TAR DNA binding protein 43 (TDP43), was identified as the ubiquitinated protein in intracellular aggregates in both ALS and FTD (Neumann et al., 2006). TDP43, encoded by the TARDBP gene, is a ubiquitously expressed nucleic acid binding protein that play critical roles in RNA splicing and microRNA biogenesis (Buratti and Baralle, 2008). Over 40 missense mutations in TARDBP have been identified in ALS cases (Sreedharan and Brown, 2013). These missense mutations are almost always located in the glycine-rich C-terminal domain of the protein, which has important roles in protein-protein interactions and liquid-liquid phase separation (Wang et al. 2018; Sreedharan et al., 2008; Van Deerlin et al., 2008; Yokoseki et al., 2008).

TDP43 is predominantly found in the nucleus, with a minor fraction of the protein cycling through the cytosol. However, cytosolic TDP43 dramatically increases in patients carrying TARDBP fALS alleles, in many sALS patients, and in a large fraction of fALS patients carrying mutations in other causal genes. TDP43 mislocalization may contribute to the degeneration of motor neurons in ALS/FTD. One hypothesis is that mutant TDP43 acts through a gain of toxic function mechanism by aggregating and inhibiting the endogenous function of normal TDP43. In this model, TDP43 loss of function defects would contribute to neurodegeneration. Alternatively, mutations in TDP43 could alter endogenous RNA splicing and microRNA biogenesis via disruption of functional interactions (Conicella et al., 2016) or mutant TDP43 protein may act in an abnormal cellular compartment, resulting in neurodegeneration and indicative of a gain of toxic function mechanism.

We found eleven published studies that, in combination, reported 93 modifier genes of TDP43 ALS/FTD phenotypes (Supplemental File 1, TDP43 Tab), for either mutant TDP43 overexpression, wildtype TDP43 overexpression, or TDP43 ortholog loss of function (Armakola et al., 2012; Chou et al., 2015; Elden et al., 2010; Figley and Gitler, 2013; Jablonski et al., 2015; Kim H. J. et al., 2014; Kim S. H. et al., 2012; Liachko et al., 2013; Sreedharan et al., 2015; Zhan et al., 2013; Zhan et al., 2015). We undertook bioinformatics analysis, as described above, with these TDP43 modifier genes and found only 4 enriched GO pathways with odds ratio above 5 (Fig. 2B). Pathways are listed in Supplemental File 2, TDP43 tab, and include “G/M2 cell cycle regulation” (GO:0000086, GO:0044839) and “regulation of viral transcription” (GO:0019083, GO:0032897), for which TDP43 roles have already been described (Ignatius et al., 1995; Yamashita et al., 2014). Modifier genes that led to bioinformatic analysis identification of “G/M2 cell cycle regulation” are shown in Fig. 6.

FUS

Originally characterized as a liposarcoma oncogene, mutations in the Fused in Sarcoma gene (FUS) were found in a cohort of 197 British ALS patients in 2009. The FUS protein is a ubiquitously expressed RNA-binding protein involved in splicing and stress granule formation (Lagier-Tourenne and Cleveland, 2009). Mutations in FUS cause approximately 4–5% of all familial ALS cases. Patient mutations can be found throughout the FUS protein, but mutation of the C-terminal nuclear localization signal (NLS) is most frequently observed (Ju et al., 2011; Ling et al., 2013). In some cases, FUS mutations result in FTD, and patients with FUS-linked FTD usually show ALS symptoms (Nolan et al., 2016). These FTD patients present with FUS-immunoreactive inclusions; these inclusions are also present in the motor neurons of FUS ALS patients who lack FTD symptoms (Deng et al., 2010; Hewitt et al., 2010; Rademakers et al., 2010).

In most FUS ALS patients examined, mutant FUS is mislocalized from the nucleus and protein aggregates form in the cytoplasm (Dormann et al., 2010; Vance et al., 2009). Furthermore, cytoplasmic FUS incorporates into membraneless organelles - phase separated liquid structures (e.g. stress granules), which may drive mutant FUS aggregation (Bosco et al., 2010; Burke et al., 2015; Patel et al., 2015). Cytoplasmic aggregation of FUS may inhibit the maturation of RNAs integral for the survival of motor neurons, as nuclear FUS is important for mRNA splicing (Colombrita et al., 2015; Sun S. et al., 2015). Alternatively, mutant FUS may act via a gain of function mechanism where patient mutations may subvert DNA repair mechanisms, leading to cumulative increases in DNA damage (Hill et al., 2016).

We found five articles that identified 72 modifiers of FUS (Armakola et al., 2012; Chen Y. et al., 2016; Farg et al., 2013; Ju et al., 2011; Sun Z. et al., 2011) (Supplemental File 1, FUS tab). Many of these suppressor and enhancer genes were identified in genome-wide modifier screens in yeast expressing mutant human FUS at high levels (Sun Z. et al., 2011). Our bioinformatic analysis identified 34 GO terms/pathways that were enriched (Fig. 2C, Supplemental File 2, FUS tab). Many of these are related to cellular pathways associated with normal FUS protein function, including “RNA processing” (GO:0006396) and “translation” (GO:0006412). The FUS modifier genes associated with the most enriched GO term “nuclear-transcribed mRNA poly(A) tail shortening” (GO:0000289) are shown in Fig. 6.

C9orf72

In 2011, expansion of GGGGCC (G₄C₂) repeats in the non-coding region of chromosome 9 open reading frame 72 (C9orf72) was identified in ALS patients. C9orf72 expansion is one of the most common causes of ALS and FTD and accounts for approximately 40% of fALS cases (Rademakers et al., 2012). The number of G₄C₂ repeats varies dramatically between patients; Southern blot analysis from one family revealed pathogenic repeats ranging from 700–1,600 (DeJesus-Hernandez et al., 2011; Haeusler et al., 2016). In addition to the typical ALS motor neuron functional defects, C9orf72 ALS patients may have earlier disease onset, cognitive and behavioral impairment, and decreased survival compared to other patients (Rademakers et al., 2012).

Why G₄C₂ repeats cause disease is still unclear; studies have suggested the C9orf72 protein has roles in the endolysosomal pathway and vesicle trafficking (Aoki et al., 2017; Corrionero and Horvitz, 2018). Three non-exclusive mechanisms have been proposed: decreased C9orf72 protein expression, toxic expanded G₄C₂ repeat RNAs, and/or toxic dipeptide repeat (DPR) proteins generated by repeat-associated non-AUG (RAN) translation of G₄C₂ repeat RNAs (Haeusler et al., 2016).

Considerable evidence suggests that high level expression of either G₄C₂ repeat-derived RNAs or DPR proteins can be toxic. G₄C₂ repeat RNA may sequester RNA binding proteins and splicing factors, thus disrupting their normal functions and causing neurodegeneration (Lee Y. B. et al., 2013; Mori et al., 2016; Xu et al., 2013). This model is supported by the observation that overexpression of Pur-α, an RNA-binding protein that physically interacts with repeat RNAs, suppresses G₄C₂-mediated neurodegeneration in mouse neuronal cells and D. melanogaster (Xu et al., 2013). However, DPR proteins are also toxic. These are produced through RAN translation of G₄C₂ repeat RNA, which occurs in the absence of an AUG initiation codon and from both sense and antisense G₄C₂ repeat strands (Zu et al., 2013). Different DPR proteins have varying levels of toxicity: the arginine-rich DPRs, poly(GR) and poly(PR) are most toxic (Jovicic et al., 2015; Kwon et al., 2014; Wen et al., 2014), poly(GA) is moderately toxic, and poly(GP) and poly(PA) are the least toxic (Freibaum et al., 2015; Wen et al., 2014).

We found eight articles that identified modifier genes for G₄C₂ RNA and/or DPR toxicity (Supplemental file 1, C9orf72 tab) (Boeynaems et al., 2016; Freibaum et al., 2015; Jovicic et al., 2015; Kramer et al., 2016; Lee K. H. et al., 2016; Mori et al., 2016; Xu et al., 2013; Zhang et al., 2015). Multiple unbiased genetic screens were undertaken in D. melanogaster and S. cerevisiae for modifiers of poly(PR) toxicity (Boeynaems et al., 2016; Jovicic et al., 2015). Screens in a D. melanogaster eye poly(PR) model yielded modifiers encoding proteins that directly interact with poly(GR) and poly (PR) peptides (Lee K. H. et al., 2016). In a candidate-based screen using D. melanogaster expressing (G₄C₂)₃₀ repeats, RanGAP was identified as a suppressor of neurodegeneration (Zhang et al., 2015). No modifiers of C9orf72 loss of function have been reported.

From these eight studies, we assembled a list of 285 genetic modifiers with human orthologs of G₄C₂ toxicity (Supplemental File 1, C9orf72 tab). Gene ontology bioinformatic analysis revealed 98 enriched GO pathways (Supplemental File 2, C9orf72 tab). These include “nuclear pore assembly”, “protein import”, and “protein export” (Fig. 3, Fig. 4). Additionally, “gene silencing by miRNA” (GO:0035195) and metabolism-associated pathways were enriched in this dataset. Genes associated with the most enriched pathway in our bioinformatics analysis, “nuclear pore complex assembly” (GO:0051292), are presented in Fig. 6.

Fig. 3.

Diagram of GO terms enriched for C9orf72. Illustration of the relationship between GO terms enriched in gene ontology analysis of C9orf72 genetic modifiers. Arrows indicate related terms that are “nested” inside a broader category. Darker red hues are GO terms that were more enriched in the modifier list.

Fig. 4.

List of GO terms enriched for C9orf72. GO terms with odd ratios of 5 or more are listed; the most highly enriched genes are at the top of the list.

VAPB and other ALS genes

Most studies that report ALS modifier genes focus on SOD1, TDP43, C9orf72, or FUS. Mutations in other genes also lead to ALS and insights into disease pathogenesis may arise from analysis of these other disease genes. In 2004, a P56S mutation in the Vesicle-Associated Membrane Protein-Associated Protein B/C (VAPB) gene was identified in seven different Brazilian families with afflicted individuals showing ALS and/or late-onset spinal muscular atrophy (Nishimura et al., 2004). VAPB protein interacts with SNARE proteins and regulates vesicular transport. Although the severity, presentation, and progression of disease varies between families, the VAPB P56S mutation was dominant (Nishimura et al., 2004). The P56S mutation lies in the VAPB protein Major Sperm Protein (MSP) domain, which likely mediates protein dimerization and other protein-protein interactions.

The functional consequences of VAPB P56S that lead to ALS are poorly understood, but both gain of function (Kuijpers et al., 2013; Ratnaparkhi et al., 2008) and loss of function mechanisms (Kabashi et al., 2013) have been proposed. In normal cells, VAPB mediates membrane interactions between mitochondria and the endoplasmic reticulum, which are critical for mitochondrial calcium regulation and ATP production (Stoica et al., 2014; Stoica et al., 2016). Mutations in either TDP43 or FUS can disrupt VAPB function, ultimately leading to disrupted mitochondrial calcium uptake and decreasing ATP production (Stoica et al., 2014; Stoica et al., 2016). We found two articles describing genetic modifiers of VAPB (Deivasigamani et al., 2014; Sanhueza et al., 2015), as well as one article describing genetic modifiers of OPTN (Akizuki et al., 2013) and one article describing VCP modifier genes (Ritson et al., 2010) (Supplemental File 1, Other tab). We searched for modifier genes of other fALS-linked genes, but did not uncover additional modifier studies in the published literature. VAPB modifiers were identified in two different D. melanogaster screens. Deivasigamani et al. upregulated or downregulated D. melanogaster VAPB (dVAP) ortholog levels, which results in altered bristles (Deivasigamani et al., 2014). Sanhueza et al., found that high level expression of dVAP[P58S] in the D. melanogaster eye leads to reduced eye size and used this observation to identify 85 modifier genes (Sanhueza et al., 2015). Only one pathway, “single-organism cellular localization” (GO:1902580), was significantly enriched in our gene ontology bioinformatics.

Modifier genes associated with more than one ALS gene

If patient alleles in the genes listed above lead to a single disease, which we call ALS, then one would expect commonalities in disease mechanism and pathological processes. Accordingly, one might expect common pathways to arise from modifier gene analyses. This hypothesis is supported by previous work demonstrating that some modifier genes impact ALS-associated defects in more than one ALS model. In total, 946 modifier genes were identified from the literature with 727 corresponding human orthologs (Supplemental File 1). To look for commonalities between ALS modifier genes, we compiled a list of modifier genes with impact on more than one ALS causal gene (Table 1). For example, if a gene modified defects in both an SOD1 ALS model and a TDP43 ALS model, it was included in Table 1 (and in Supplemental File 1, multiple ALS genes). In addition, Table 1 includes modifier genes identified in ALS patient GWAS or genetic studies that have been validated in ALS models, as these are likely relevant to disease. In total, 43 modifier genes have functional impact on more than one ALS gene and are listed in Table 1.

Table 1.

Genetic modifiers that may modify multiple ALS genes. This list includes the human orthologs that were identified as modifiers of more than one ALS-causal genes (e.g. SRRT orthologs were reported as modifiers in both SOD1 and TDP43 models). Additionally, human genes reported as modifiers through GWAS or linkage analysis studies are also included in this list.

Human Gene	Organism	Modifier for	Reference
ATXN2	Yeast, Human	TDP43, Human	(Elden et al., 2010; Figley and Gitler, 2013; Lee T. et al., 2011)
BMP2	Nematode, Fly	SOD1, C9orf72	(Wang J. et al., 2009; Zhang et al., 2015)
CCNB1	Yeast, Fly	FUS, VAPB	(Sanhueza et al., 2015; Sun Z. et al., 2011)
CCT4	Nematode, Fly	SOD1, C9orf72	(Lee K. H. et al., 2016; Wang J. et al., 2009)
CDC6	Yeast, Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Jovicic et al., 2015)
CHGB	Mouse, Human	SOD1, Human	(Gros-Louis et al., 2009; Ohta et al., 2016)
COA4	Nematode, Yeast	SOD1, FUS	(Sun Z. et al., 2011; Wang J. et al., 2009)
DAZAP1	Fly	C9orf72, VCP	(Ritson et al., 2010; Zhang et al., 2015)
DBR1	Yeast	TDP43, FUS	(Armakola et al., 2012; Figley and Gitler, 2013)
DNAJA1	Yeast, Fly	SOD1, C9orf72	(Lee K. H. et al., 2016; Strain et al., 1998)
FBXW7	Nematode, Yeast	SOD1, FUS	(Sun Z. et al., 2011; Wang J. et al., 2009)
GNAQ	Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Zhang et al., 2015)
HNRNPC	Yeast, Fly	C9orf72, FUS	(Lee K. H. et al., 2016; Sun Z. et al., 2011)
IMPA1	Yeast, Fly	FUS, VAPB	(Deivasigamani et al., 2014; Sun Z. et al., 2011)
IPO5	Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Lee K. H. et al., 2016)
KPNB1	HeLa cells, Fly	TDP43, C9orf72, VAPB	(Boeynaems et al., 2016; Kim S. H. et al., 2012; Sanhueza et al., 2015)
MMP9	Mouse, Human	SOD1, Human	(Kaplan et al., 2014; Lorenzl et al., 2006)
MRPL33	Yeast	TDP43, FUS	(Armakola et al., 2012; Sun Z. et al., 2011)
NAA20	Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Zhang et al., 2015)
NCL	Yeast, Fly	C9orf72, FUS	(Lee K. H. et al., 2016; Sun Z. et al., 2011)
NOB1	Nematode, Yeast	SOD1, C9orf72	(Jovicic et al., 2015; Wang J. et al., 2009)
NUP50	Fly	TDP43, C9orf72	(Boeynaems et al., 2016; Freibaum et al., 2015; Zhan et al., 2013)
NUP107	Fly	FUS, C9orf72	(Boeynaems et al., 2016; Sun Z. et al., 2011)
PDE6D	Nematode, Fly	SOD1, VAPB	(Deivasigamani et al., 2014; Silva et al., 2011)
PELO	Yeast, Fly	TDP43, C9orf72	(Armakola et al., 2012; Jovicic et al., 2015)
PGM1	Yeast, Fly	TDP43, C9orf72	(Kim H. J. et al., 2014; Zhang et al., 2015)
PIAS1	Nematode, Fly	SOD1, VAPB	(Sanhueza et al., 2015; Wang J. et al., 2009)
POLD3	Yeast, Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Jovicic et al., 2015)
PPP1R3C	Fly, Yeast	C9orf72, FUS	(Sun Z. et al., 2011; Zhang et al., 2015)
RAN	Human cells, Fly	TDP43, C9orf72	(Freibaum et al., 2015; Kim S. H. et al., 2012)
RBM10	Fly	C9orf72, VAPB	(Lee K. H. et al., 2016; Sanhueza et al., 2015)
SAMD4A	Yeast	TDP43, C9orf72	(Jovicic et al., 2015; Kim H. J. et al., 2014)
SCARB1	Fly	C9orf72, VAPB	(Sanhueza et al., 2015; Zhang et al., 2015)
SIRT1	Fly	C9orf72, VAPB	(Sanhueza et al., 2015; Zhang et al., 2015)
SLC17A5	Nematode, Fly	SOD1, C9orf72	(Wang J. et al., 2009; Zhang et al., 2015)
SOD1	Fly	SOD1, VAPB	(Deivasigamani et al., 2014; Kumimoto et al., 2013)
SRRT	Fly	C9orf72, VAPB	(Deivasigamani et al., 2014; Freibaum et al., 2015)
TARDBP	Fly	TDP43, C9orf72, VAPB, VCP	(Deivasigamani et al., 2014; Lee K. H. et al., 2016; Ritson et al., 2010; Sreedharan et al., 2015; Wang J. W. et al., 2011)
TNPO1	Fly	C9orf72, VAPB	(Boeynaems et al., 2016; Deivasigamani et al., 2014)
TOP1	Nematode, Fly	SOD1, C9orf72	(Lee K. H. et al., 2016; Wang J. et al., 2009)
UBR5	Yeast, Fly	TDP43, C9orf72	(Kim H. J. et al., 2014; Lee K. H. et al., 2016)
UQCRC2	Nematode, Fly	SOD1, VAPB	(Deivasigamani et al., 2014; Silva et al., 2011)
XPO1	Nematode, Fly	SOD1, C9orf72	(Silva et al., 2011; Zhang et al., 2015)

Two genes, KPNB1 and TARDBP, were identified as genetic modifiers in more than two ALS models. KPNB1 was reported as a modifier of C9orf72, VAPB, and TDP43 ALS models. In an RNAi screen conducted in D. melanogaster C9orf72 model of ALS, decreased KPNB1 function enhanced PR25-mediated eye degeneration (Boeynaems et al., 2016). In HeLa cells, KNPB1 knockdown enhanced the cytosolic localization of TDP43 (Kim S. H. et al., 2012). Additionally, overexpression of the D. melanogaster ortholog of KPNB1, Fs(2)Ket, resulted in suppression of the rough eye phenotype in a VAPB model of ALS. TARDBP was reported as a modifier of C9orf72, VAPB, and VCP. RNAi knockdown of TARDBP in a C9orf72 model suppressed a viability defect and the rough-eye phenotype in a D. melanogaster model (Lee K. H. et al., 2016). TARDBP has also been reported as a suppressor of VCP-related degeneration (Ritson et al., 2010) and acts as a suppressor in an overexpression model of VAPB (Deivasigamani et al., 2014). While TARDBP was intentionally selected for assessment in these studies, the nuclear pore complex protein, KPNB1, was independently identified in less biased screens.

To identify common pathways associated with modifier genes, we undertook bioinformatic analysis with the genes listed in Table 1. Forty-two GO terms were enriched in this analysis. Pathways enriched in in this modifier gene list included GO terms associated with protein transport or metabolic processes. The most enriched GO term was “protein import into nucleus, translocation” (GO:0000060).

Discussion

Modifier gene studies have the potential to dramatically increase our understanding of ALS pathogenesis and to provide insight into variation in patient symptoms, penetrance of disease, and disease progression. Modifier gene studies can provide insights into pathways associated with neuronal dysfunction and neurodegeneration in ALS. Furthermore, common genetic modifiers may link ALS caused by mutations in different genes, suggesting a common mechanism of motor neuron degeneration. Additionally, modifier genes can be used to identify pathways and targets for therapeutic intervention. Many genetic modifiers of ALS have been discovered through hypothesis-driven experiments, forward genetic screens, or genetic studies in human populations. But, to our knowledge, a comprehensive listing and analysis of modifier genes pertinent to ALS has not been undertaken previously.

It is likely that both loss and gain of function mechanisms contribute to ALS pathology. Therefore, we included modifiers of both loss- and gain of function in our survey, as well as overexpression of wildtype or mutant protein. In total, we compiled a list of 727 modifier genes with human orthologs. Many modifier genes were found in S. cerevisiae, C. elegans, D. melanogaster, M. musculus, and cell culture models of ALS. Additionally, genetic modifiers were identified in ALS patients through linkage analysis and genome wide association studies. In total, we identified 72 articles in the published literature that reported 727 modifier genes with human orthologs for SOD1, TDP43, FUS, C9orf72, VAPB, VCP, or OPTN. We searched for modifiers of other ALS-linked genes, but did not find any in the published literature. Interestingly, the 727 genes identified as modifiers of ALS corresponds to roughly 5% of human genes, consistent with the complexity of this disease. We appreciate the enormous effort these original studies represent and we hope to highlight the importance of these studies and leverage their results to identify common pathways pertinent to ALS. We note that additional modifier genes have been reported in subsequent studies, including those by Kramer et al. 2018 (Kramer et al., 2018).

Many of the pathways that were enriched in our bioinformatics analysis are associated with the endogenous functions of genes implicated in ALS and are established as dysregulated in ALS patients, including “response to reactive oxygen species” (GO:0000302) in SOD1 and “RNA processing “(GO:0006396) in FUS. Interestingly, metabolic processes were identified as enriched pathways in all of our modifier gene lists. This commonality highlights the importance of previous studies demonstrating that metabolism is affected in ALS patients (Mattiazzi et al., 2002).

FUS and TARDBP encode RNA-associated proteins and it has been suggested that they act in the same pathways in ALS pathogenesis (Honda et al., 2013). The analysis of FUS and TARDBP modifier genes could be interpreted to support this hypothesis. When GO pathways are examined, 2 of the 12 enriched GO terms found from the list of TDP43 modifier were also included in the list of 60 FUS-modifier enriched GO terms: “cellular macromolecule metabolic process” and “viral transcription”. There were over 2000 GO pathway terms available in our bioinformatic analysis; the small overlap we observed between FUS and TDP43 is significant, but may reflect the importance of RNA-binding proteins in these processes.

The comprehensive literature search reveals that relatively few ALS modifier genes have been tested in other models of ALS or have been identified in more than one independent modifier screen. Overall, only 43 modifier genes are reported to modify more than one ALS gene. Of these 43 genes, KNBP1 and TARDBP were reported to be modifiers in more than two ALS models. An inherent bias against the publication of “failure to suppress cross-species” may partially account for this, as well as a bias against reporting negative results, or a lack of motivation to re-test modifier genes identified in other species/models. To fully understand why mutations in specific genes cause ALS and to identify therapeutic targets, we suggest that modifier genes should be tested in multiple ALS models. This should expose commonalities and differences between ALS caused by mutations in different genes and inform the selection of therapeutic targets.

ALS modifier gene studies have already increased our understanding of pathways that may be dysregulated in this devastating disease. We provide the first comprehensive review of published ALS modifier genes and undertook bioinformatic analysis. These data suggest that common pathways may underlie ALS caused by mutations in different genes. We expect that as additional modifier genes are identified and tested in additional models of ALS, more commonalities between the different ALS genes will be found and additional therapeutic targets will be developed for the treatment of this disease.

Supplementary Material

Supplemental File 1

Supplemental File 1. List of genes reported to modify ALS-associated defects. The page labeled “Modifiers of multiple” contains the genes that were found to modify two or more ALS genes. Pages are labeled with the name of the ALS gene that the genetic modifier was observed to modify. Each tab contains the following information: the originally reported gene, the human ortholog, and the reference.

Supplemental File 2

Supplemental File 2. Gene ontology analysis results listing enriched GO terms for each of the analyses reported in the main text and the Figures. Each page lists the gene ontology term, the odds ratio, p-value, expected counts (the number of times a gene should be in this category given the number of genes on our list), the count (the number of times a gene is associated with the specific GO term here), and size (the number of genes associated with the called GO term).

Supplemental File 3

Supplemental File 3. Genes associated with enriched GO terms. The modifier genes associated with each enriched GO term are reported from the original analysis output.

Fig. 5.

GO terms enriched for genes reported to be modifiers of multiple ALS genes. Diagram (top) shows relationship between GO terms enriched in gene ontology analysis. Arrows indicate related terms that are “nested” inside a broader category. Darker red hues are GO terms that were more enriched in the modifier list. GO terms above odd ratios of 5 or greater are listed (below); the most highly enriched genes are at the top of the list.

Abbreviations

ALS

Amyotrophic Lateral Sclerosis

C9orf72

Chromosome 9 open reading frame 72

fALS

Familial ALS

FTD

Frontotemporal Dementia

FUS

Fused in sarcoma

OPTN

Optineurin

sALS

Sporadic ALS

SOD1

Superoxide dismutase 1

TDP43

TAR DNA binding protein 43

VCP

Valosin-containing protein

VAPB

Vesicle-Associated Membrane Protein-Associated Protein B/C