The complete list of sequenced genes is provided in Supplementary Table 1. Genes were either known ALS genes or genes encoding RNA-binding proteins. The RNA-binding proteins were in two groups—RRM-containing proteins with a PrLD (Couthouis et al., 2011) or those identified binding partners of the G4C2-repeat expansion (Cooper-Knock et al., 2014).

ALS patients were selected to increase the probability of discovering novel genetic variants—they either had a positive family history, or they were relatively young (<50 years old) at presentation or they carried an expansion of C9ORF72. Genomic DNA was extracted from 103 ALS patients from the North of England. The cohort included 34 familial ALS patients in whom a genetic cause had not been identified despite screening for ALS associated mutations in SOD1, C9ORF72, TARDBP, and FUS; 61 young-onset sporadic ALS patients; and thirteen C9ORF72-ALS patients (Table 1). A patient with an identified mutation in FUS was included as a positive control. G4C2-repeat expansions of C9ORF72 expansions were identified by repeat-primed PCR as described previously (Cooper-Knock et al., 2012); all patients were screened for C9ORF72 expansion prior to selection for the screen. The study was approved by the South Sheffield Research Ethics Committee and informed consent was obtained for all samples.

Genomic DNA was enriched for selected RNA-binding proteins and known ALS genes using a custom designed Agilent SureSelect in solution kit. Sequencing was performed using an Illumina HiScan platform according to manufacturers instructions.

Rare deleterious mutations were defined by frequency within the Exome Aggregation Consortium data set of <1/10,000 control alleles (Lek et al., 2016), and a Phred-scaled Combined Annotation Dependent Depletion (CADD) score >10 (Kircher et al., 2014). Comparison of various pathogenicity prediction tools recently supported the sensitivity and specificity of CADD (Salgado et al., 2016). Given that we were focused on exonic changes with an effect on protein function, synonymous changes were excluded. We excluded any changes with a read depth <10 and validated by Sanger sequencing any changes with read depth 10–15 or a novel allele frequency less than one third the reference allele frequency (Supplementary Figure 1).

ExAC defines constrained genes based on an observed frequency of loss of function mutations which is much less than predicted by sequence specific mutation probabilities (Lek et al., 2016). A threshold for “constrained” is set as probability of a gene being loss of function intolerant (PLi) > 0.95.

We identified 12 patients with mutations in nine known ALS genes (Table 2, Supplementary Table 2). This is expected based on reported frequencies of these mutations and served as a validation of our strategy. One patient with a previously identified FUS mutation was included as a positive control. Rare deleterious variants were newly identified in ALS2, DCTN1 (two different variants), ELP3, EWSR1, SETX (two different variants), SOD1 (two different variants), UNC13A, C9ORF72, and VCP.

Several of the mutations we identified in ALS genes affect previously reported amino acids or protein domains. For example, both mutations in DCTN1 were within the dynein associated protein domain, which is consistent with previously reported mutations (Münch et al., 2004); the mutation identified in EWSR1 occurs in the same amino acid as previously reported (Couthouis et al., 2011); one of the SETX mutations we identified lies within a helicase domain which contains several previously reported mutations (Hirano et al., 2011); both SOD1 mutations have been previously described in familial ALS (Orrell et al., 1999); and a mutation in the same amino acid of VCP has been previously identified in another ALS patient (Johnson et al., 2010).

Other variants we identified in known ALS genes are more novel. ELP3 has been previously associated with ALS by GWAS (Simpson et al., 2009), but pathogenic variants have not been identified. The patient identified in this screen demonstrated a nonsense mutation in exon 10 which disrupts all described functional domains of the protein. Similarly variation in UNC13A has been identified as a risk factor for sporadic ALS (van Es et al., 2009) and as a modifier of the clinical phenotype, but pathogenic variants have not been identified. Our patient with a variant in UNC13A has a family history of ALS and no other identified mutation in an ALS gene (or any other gene in our screen). One sporadic ALS patient has a variant in ALS2; given that mutations in ALS2 are usually autosomal recessive and associated with a slowly progressive juvenile onset form of the disease, then this variant is of unknown significance. However, no study has reported this exact change previously (Al-Chalabi et al., 2003; Luigetti et al., 2013). Similarly a rare deleterious variant was identified in C9ORF72 in a patient who also carried a G4C2-repeat expansion; no pathogenic variants have been confirmed in C9ORF72 except the G4C2-repeat expansion in intron 1 and therefore this variant is also of unknown significance.

We identified 19 patients with a rare deleterious variant in a RNA-binding protein with a PrLD of whom three had more than one variant (Table 3, Supplementary Table 2). Fourteen of the patients had died, four patients were still alive and in these cases disease duration was censored to the present date. One patient with a variant in MTHFSD also had a mutation in SOD1. SOD1 mutations are associated with a distinct clinical phenotype and pathology compared to characterized mutations in RNA-binding proteins (Cooper-Knock et al., 2013) and therefore this patient was excluded from further analysis. Of the 21 identified variants remaining, 16 (76%) occurred in either the RRM domain or a low complexity sequence (Table 3). The number of variants per patient correlated with rate of disease progression (Figure 1A, t-test, p = 0.033) but not age of onset. Including C9ORF72 expansions in this analysis did not appear to be synergistic.

The Project MinE browser (http://databrowser.projectmine.com/) was utilized to search for additional evidence of similar variants in these proteins. The Project MinE Consortium has to date reported whole genome sequencing of 1169 ALS cases and 608 controls from the Netherlands. For RBM4B, RBM45, RBMS2, RAVER1, PPARGC1B, and TRNAU1AP, the project Project MinE data identified additional variant(s) within the same exon which were present either exclusively in ALS patients or were more frequent in ALS patients than controls. RBM12, RBM12B, RBM15, RBM15B, and RBM45 are single exon genes, but Project MinE identified ALS cases with disease-associated variant(s) within <25 amino acids in each of these genes. This clustering of cases for each of these genes supports the functional significance of the rare variants we have discovered.

The ALS Variant Server, Worcester, MA (http://als.umassmed.edu/) reports whole exome sequencing from 1,022 familial ALS patients. Within this cohort we identified an additional example of an ALS patient carrying an A622G variant in RBM12 and four ALS patients carrying p.S550G/p.S525G/p.S589G (single case) or p.E395K/p.E370K/p.E434K (3 cases) variants in PPARGC1B.

It is noteworthy that a small number of genes found to contain rare deleterious variants but classified as known ALS genes or G4C2-repeat binding partners are also RRM-containing proteins with a PrLD. This includes EWSR1, HNRNPA3, HNRNPU, and HNRNPUL1. Except for being previously identified as a known ALS gene, EWSR1 is not distinct from the other RRM-containing proteins with a PrLD under consideration; therefore EWSR1 is included in the analysis of synergy detailed above. In contrast HNRNPA3, HNRNPU, and HNRNPUL1 were selected on the basis of an independent hypothesis: that loss of function in the proteins encoded by these genes might mimic sequestration by G4C2-repeat-RNA derived from a C9ORF72 expansion. The majority of variants identified in RRM-containing proteins with a PrLD are located in either the RRM-domain or the PrLD but, consistent with an alternate mechanism, variants identified in HNRNPA3, HNRNPU, and HNRNPUL1 are located distinct functional domains (Table 4). To avoid potentially confounding discrepancy between mechanisms of pathogenicity HNRNPA3, HNRNPU, and HNRNPUL1 were not included in analysis of other variants identified within RRM-containing proteins with a PrLD.

We identified 18 patients with a rare deleterious variant in a G4C2-repeat binding protein (Table 4, Supplementary Table 2). No patients had more than one variant in a G4C2-repeat binding protein. Five of the patients carried a G4C2-repeat expansion in C9ORF72. Fourteen of the patients had died, six patients were still alive and in these cases disease duration was censored to the present date. Patients with a G4C2-binding protein variant in combination with a C9ORF72 expansion had a significantly faster disease course (Figure 1B, t-test, p = 0.025) but age of onset was not significantly different. For one patient with a variant in ILF3, no clinical information was available. In two specific examples the same gene is mutated in patents with and without a C9ORF72 expansion—SLC1A3 and EEF1G. In both cases there is a 50% reduction in disease duration (SLC1A3: 52 months to 27 months; EEF1G: 79 months censored to 22 months) in the patient carrying the C9ORF72 expansion and the mutation.

Sequestration of RNA-binding proteins by G4C2-repeat RNA associated with C9ORF72-ALS would be expected to prevent those proteins performing their normal function. Consequently a mutation which exacerbates this toxicity would be expected to cause loss-of-function. Of the 18 G4C2-repeat binding proteins in which we identified a rare deleterious variant, 67% are encoded by genes which are defined by ExAC as extremely loss-of-function intolerant (ExAC refer to this property as “constrained”) (Lek et al., 2016). This is enriched compared to the total list of G4C2-repeat binding proteins screened (Supplementary Table 1) of which 42% are ExAC constrained (41 constrained from 98 total). This observation supports our proposed mechanism. In comparison, for RRM-containing proteins with a PrLD, the proportion of variants discovered in ExAC constrained genes is only 32%.

The Project MinE browser (http://databrowser.projectmine.com/) was utilized to search for additional evidence of similar variants in these proteins. For SLC1A3, EEF1G, hnRNPU, hnRNPUL1, EZR, and GRSF1 Project MinE identified additional variant(s) within the same exon which were present either exclusively in ALS patients or were more frequent in ALS patients than controls. The ALS Variant Server, Worcester, MA (http://als.umassmed.edu/) reports whole exome sequencing from 1,022 familial ALS patients. Within this cohort we identified an additional example of an ALS patient carrying a p.D403E mutation in EEF1G, a p.P12R mutation in HNRNPA3, and a p.A297P/p.A308P mutation in SRPK2. This clustering of cases for each of these genes supports the functional significance of the rare variants we have discovered.

To calculate whether the observed frequency of rare deleterious variants in RNA-binding proteins in our DNA sequencing screen is higher than expected we utilized ExAC frequencies and CADD scores for the identified changes. CADD scoring is expressed as the observed frequency of variants which are at least as pathogenic as the observed variant. For this analysis we assumed that observed frequency is independent of pathogenicity on the basis that ALS does not usually affect reproductive fitness. We observed 39 rare deleterious variants in 1,223,647 bases of DNA from 103 patients; this is a significant enrichment compared to observed control frequencies (p = 5.31E-18) suggesting that these variants are significantly enriched in ALS patients.

No significant correlation was identified between total number of variants per patient and the clinical phenotype (Pearson correlation, correlation coefficient = −0.20, p = 0.21) (Figure 1C). This was unchanged whether or not C9ORF72 expansions are considered. In contrast, when either RRM-containing proteins with PrLDs or G4C2 binding partners are considered in isolation, then there is a significant synergistic effect on clinical phenotype (Figures 1A,B). We conclude that a synergy is present only between variants in functionally interacting genes/proteins.

SLC1A3 encodes excitatory amino acid transporter 1 (EAAT1) which is a glial glutamate transporter and also a G4C2 binding partner. Mutations of SLC1A3 are a cause of episodic ataxia type 6 (EA6). The proposed mechanism is excitotoxicity via loss of glutamate uptake—excitotoxicity has also been proposed as a pathophysiological mechanisms in ALS (Cooper-Knock et al., 2013). Of the mutations associated with EA6, a p.C186S mutation in transmembrane segment 4 is the closest to both of our identified variants: p.N216S and p.F170L (Table 4). Transmembrane segment 4 has been associated with inter-subunit contact to stabilize the trimeric structure of the transporter (Yernool et al., 2004). The p.N216S mutations occur in a eukaryotic specific insertion between transmembrane domains 4b and 4c. The p.F170L mutation occurs in transmembrane domain 4A. Interestingly, while complete loss of SLC1A3 function leads to a severe phenotype with progressive ataxia (Jen et al., 2005), mutation in transmembrane segment 4 has been associated with partial loss of function and variable penetrance (de Vries et al., 2009) which is consistent with a late onset disease such as ALS. It is noteworthy that Project MinE identified an additional ALS patient with a rare (ExAC frequency <1/10,000 control alleles) mutation within the 4A transmembrane region.

EEF1G encodes a component of the elongation factor-1 (EF1) complex involved in the elongation phase of protein translation which is a G4C2 binding partner. The EEF1G subunit is not proposed to have a direct role in translation (Fan et al., 2010), but co-immunoprecipitates with tubulin (Janssen and Moller, 1988) and has been observed to bind mRNA directly (Al-Maghrebi et al., 2002). This is consistent with a role for EEF1G in anchoring and translation of mRNAs in cytoskeleton bound ribosomes (Corbi et al., 2010). Translation at sites distant from the nucleus is particularly relevant in neurons and in large motor neurons in particular. We have identified two patients with mutations in the C-terminal domain of EEF1G: p.D403E and p.R327C (Table 4). Project MinE identified an additional ALS patient with a T902C variant in exon 8, the same exon as the C979T change we have identified.

XRCC6 is a component of the non-homologous end joining (NHEJ) complex involved in repair of double stranded DNA breaks and is a G4C2 binding partner. Two patients were identified with rare deleterious variants in XRCC6: p.M348T and p.G589R (Table 4). Both variants occur within DNA binding domains, therefore both variants could conceivably lead to loss of function which is consistent with our disease model. Deletion of XRCC6 in mice leads to premature aging without an increased rate of neoplasm (Li et al., 2007). This is consistent with observations in ALS and indeed impairment of NHEJ has been previously implicated in ALS (Sama et al., 2014).

PPARGC1B is a transcription factor with roles in energy metabolism and mitochondrial biogenesis and a RRM-containing protein with a PrLD. We identified three young sporadic patients with rare deleterious variants in PPARGC1B: pP385H, p.S589G, and p.E434K (Table 3). Two of the variants identified lie within exon 4 either within or close to a low complexity region containing glutamic acid repeats. The p.E434K variant is actually within the glutamic acid repeats region and the same genetic change is observed in an additional three familial ALS cases within the ALS Variant Server. It seems likely that the variants we have identified and those found in the ALS Variant Server affect the function of the PrLD within PPARGC1B, leading to an increased risk of pathological aggregation.

A rare predicted deleterious variant was identified in C9ORF72 in a patient who also carries a G4C2-repeat expansion. From a single patient it is not possible to determine whether there was synergy between the variant and the expansion but it is noteworthy that the patient identified suffered rapidly progressive disease: death occurred in 12 months from first symptom onset. In our population this is within the 10% most rapidly progressive C9ORF72-ALS patients (Cooper-Knock et al., 2012). If this variant is pathogenic and synergistic with the G4C2-repeat expansion, then it provides some insight into the pathogenesis of C9ORF72-ALS. A variant in C9ORF72 could not recapitulate the proposed gain-of-function toxicity attributed to the G4C2-repeat, but it could potentially cause loss-of-function highlighting the relative importance of proposed haploinsuffuciency due to G4C2-repeat expansion.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.