The role of de novo mutations in the development of amyotrophic lateral sclerosis

Abstract

The genetic basis combined with the sporadic occurrence of amyotrophic lateral sclerosis (ALS) suggests a role of de novo mutations in disease pathogenesis. Previous studies provided some evidence for this hypothesis; however, results were conflicting: no genes with recurrent occurring de novo mutations were identified and different pathways were postulated. In this study, we analyzed whole-exome data from 82 new patient-parents trios and combined it with the datasets of all previously published ALS trios (173 trios in total). The per patient de novo rate was not higher than expected based on the general population (P = 0.40). We showed that these mutations are not part of the previously postulated pathways, and gene-gene interaction analysis found no enrichment of interacting genes in this group (P = 0.57). Also, we were able to show that the de novo mutations in ALS patients are located in genes already prone for de novo mutations (P < 1 × 10⁻¹⁵). Although the individual effect of rare de novo mutations in specific genes could not be assessed, our results indicate that, in contrast to previous hypothesis, de novo mutations in general do not impose a major burden on ALS risk.

1 |. INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a heterogeneous disease in which many of the contributing genetic factors are still unknown. The identification of the underlying genetic factors is not only important for the understanding the disease pathogenesis, but also for individualized genetic counseling of patients and their families.

In recent years, de novo mutations have been implicated in the pathogenesis of a growing number of different diseases, for example, type 1 Charcot-Marie-Tooth (Motley et al., 2016), specific congenital syndromic disorders (Hoischen et al., 2011), nonsyndromic intellectual disability (de Ligt et al., 2012), and autism (Neale et al., 2012; O’Roak et al., 2012). In contrast to these disorders, ALS manifests later in life with peak incidence in the seventh decade (Huisman et al., 2011). Therefore, it is often not possible to obtain DNA from both parents, and thus it is difficult to acquire a large cohort of patient-parent trios. Recently, two studies (Chesi et al., 2013; Steinberg, Yu, Koboldt, Mardis, & Pamphlett, 2015) suggested a possible role of de novo mutations in the pathogenesis of ALS and hypothesized that genes with a function in either chromatin or transcription regulation might be mutated. In this study, we aimed to verify this role of de novo mutations in ALS, using next-generation sequencing data from, to our knowledge, the largest international cohort of ALS patients with healthy parents (patient-parent trios) and combining this data with the previously published data.

2 |. MATERIALS AND METHODS

2.1 |. Patients

All patients fulfilled the revised El Escorial criteria for possible, probable laboratory-supported, probable or definite ALS (Brooks, Miller, Swash, Munsat, & World Federation of Neurology Research Group on Motor Neuron Diseases, 2000). The patients had no first- or second-degree relatives with motor neuron disease or dementia.

The patients were recruited in The Netherlands from an ongoing prospective, population-based study (Huisman et al., 2011) (21 trios), in Germany by the MND-NET (18 trios), in Italy by the IRCCS Istituto Auxologico Italiano (39 trios), and in Sweden by the Department of Pharmacology and Clinical Neuroscience, Umea University (four trios). All participants gave written informed consent and local ethical committees have prospectively reviewed and approved this research at the participating centers.

Genetic prescreening encompassed at least testing of the C9orf72 repeat expansion. For the Dutch, Italian, German, and Swedish trios mutations in SOD1 and FUS were excluded. In Italian trios, mutations in TARDBP were excluded as well. In one patient from a Dutch trio (trio NL21), a C9orf72 repeat expansion was found, which was inherited by the mother. No first- or second-degree family member of this patient was diagnosed with ALS or frontotemporal dementia. The mother was over 60 years old, and her medical history reported only gastric acid- related problems and thus contained no neurological or psychiatric diseases. She reported no use of medication. Because the C9orf72 repeat expansion is known to co-occur with other mutations in ALS genes (van Blitterswijk et al., 2012; van Blitterswijk et al., 2013), we included this trio for further analysis.

2.2 |. Whole-genome/exome sequencing and variant detection

Genomic DNA was obtained from whole blood using standard procedures. In the Dutch samples, whole-genome sequencing was performed at Complete Genomics (Mountain View, CA) using DNA nanoball arrays and combinatorial probe-anchor ligation reads. Sequence reads were mapped to the National Center for Biotechnology Information (NCBI) reference genome 36. The mean coverage was 57×, and 97.7% of the genome was assessed by >10 individual reads per base. Variant calling confidence scores were provided by Complete Genomics, based on a likelihood model for true variant calls. Variants not passing the standard confidence score threshold were discarded. CGAtools 1.6.0 (Complete Genomics) was used to call de novo variants. Variants that were present in other nonrelated parental genomes or in 69 publicly available Complete Genomics genomes were considered to be platform-specific variants and also discarded. Passing variants were annotated using Annovar software 2013–08-24 (http://www.openbioinformatics.org/annovar) and nonsynonymous exonic SNVs were selected. SNVs had to be called in more than 10% of the reads at that base position. Due to platform-specific false-positive variant calling in simple tandem repeat regions and near indels (Reumers et al., 2012), variants in simple tandem repeat regions (UCSC table browser, http://genome.ucsc.edu) and within five bases of an indel were excluded. Liftover from NCBI build 36 to human reference genome hg19 was performed by the NCBI remap tool (https://www.ncbi.nlm.nih.gov/genome/tools/remap).

In the Italian samples, whole-exome sequencing was performed using the SeqCap EZ Human Exome Library 2.0 (Roche Nimblegen, Madison, WI) on a HiSeq2000 platform (Illumina, San Diego, CA). Reads were aligned to a human reference (hg19) using Burrows-Wheeler Aligner (BWA) and processed using the Genome Analysis ToolKit (GATK). After removal of duplicate reads, indel realignment (GATK IndelRealigner), and base quality recalibration (GATKTableRe-calibration) was performed. The average percentage of target bases with 10x coverage per sample was 93.11%. Variant detection and genotyping were performed using the UnifiedGenotyper (UG) tool from GATK. Variants not passing quality control criteria were eliminated (QD3, MQ60.0, HaplotypeScore> 13.0, MQRankSum<−12.5, ReadPosRankSum<−8.0). Outlying heterozygosity analyses were performed using PLINKv1.0734. This was a two-step procedure where we first established an LD pruned (R2<0.5, window size = 50, step = 5) set of autosomal markers with minor allele frequency (MAF) > 0.01 and P > 0.001 for deviation from Hardy-Weinberg equilibrium. These markers were then used to calculate per sample coefficients of an inbreeding (F), and samples were excluded where these fell outside the expected range (F < −0.1 or F > 0.1) (Cirulli et al., 2015). Samples were also discarded in case of genotypes with low quality (GQ 0.2 or clinically reported) or if the gender determined by genotypes did not match the clinically reported gender.

In the German and Swedish samples, whole-exome sequencing was performed using the enrichment kits SeqCap EZ Human Exome v.2.0 or v3.0 and an Illumina HiSeq2000 sequencer with a paired-end 100 bp protocol or for the last batch SureSelect Human All Exon V6 and an Illumina Hiseq4000 sequencer with a paired-end 75 bp protocol. Data analysis was performed with VARBANK exome pipeline of the Cologne Center for Genomics. In detail, reads were aligned to the human reference genome (hg19) with BWA-aln (version 0.6.2) followed by duplicate marking (Picard version 1.64), base quality score recalibration, and local indel realignment (both GATK version 1.6–11). Mean average coverage was 91.9 and at least 92.9% of the target sequences were covered 10×. For the de novo calling, BCF files were generated from the resulting BAM files for each trio using SAMtools mpileup (version 0.1.18), and variants were called with the open source program DeNovoGear (version 0.5.1; Ramu et al., 2013). The detection filters were set to a posterior probability >0.5. All de novo variants, compound heterozygous rare variants, and homozygous rare variants were uploaded to the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/).

2.3 |. Sanger sequencing

All possible nonsynonymous de novo SNVs were validated by Sanger sequencing. Primer sequences and reaction conditions are available upon request. Sanger sequencing was performed using BigDye Terminator 3.1 sequencing kit (Applied Biosystems, Foster City, CA), DNA Analyzer 3700 or 3730XL (Applied Biosystems). PolyPhred (Nickerson, Tobe, & Taylor, 1997) version 6.18 or the DNASTAR Lasergene software (DNASTAR, Madison, WI) were used for data analysis.

2.4 |. Publicly available datasets

The Exome Aggregation Consortium (ExAC) dataset version 0.3 was downloaded from the Website (exac.broadinstitute.org); European (non-Finnish) subjects were included. The Genome of The Netherlands (GoNL) dataset release 4, and the de novo mutation rate map was downloaded from the Genome of The Netherlands Website (www.nlgenome.nl). A list of human genes with coding region size was obtained from the CCDS project (release 18, http://www.ncbi.nlm.nih.gov/CCDS/CcdsBrowse.cgi).

2.5 |. Functional enrichment analysis

The functional annotation chart of the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 (Huang, Sherman, & Lempicki, 2009) was used for functional annotation, with homo sapiens as background. Pathways with Bonferroni-adjusted P values below 0.05 and false discovery rates below 0.1 were considered significant.

2.6 |. Protein-protein interaction networks

For network analysis between proteins, the Disease Association Protein-Protein Link Evaluator v0.17 (Rossin et al., 2011) (DAPPLE, http://genepattern.broadinstitute.org) was used with genes of interest as input and 1,000 permutations. The ALS Online Database (ALSoD, http://alsod.iop.kcl.ac.uk) was consulted on May 3, 2015.

2.7 |. Statistical analysis

Statistical analysis was performed using the statistical analysis program R version 3.1.0 (http://www.r-project.org/). Independent sample t-tests were used for comparison of de novo mutation rates between studies; a one-sample t-test was used for comparison with the general population. Wilcoxon rank sum tests were used to compare distributions of de novo rate, gene size, and de novo probability. All tests were two-tailed. Power calculations were performed using G*Power (V.3.1; Heinrich Heine Universität, Düsseldorf, Germany).

3 |. RESULTS

3.1 |. De novo mutations

We collected a total of 82 patient-parent trios from four different European populations for whole-exome or whole-genome sequencing. The rate of successful Sanger sequencing validation after computational de novo mutation-filtering steps was 52% for the Dutch cohort, 59% for the German/Swedish cohort, and 61% for the Italian cohort. In total, 69 nonsynonymous de novo mutations were found in the index patients; however, all of them in different genes (Table 1).

TABLE 1.

De novo mutations (DNM)

Trio	DNM	Gene	Position	Transcript	Reference	Variant	Protein change	PolyPhen	ExAC MAF
NL2	1	PACS1	11:66002865	NM_018026.3	G	A	p.R733Q	Possibly damaging	0.000406
NL3	1	PTPN1	20:49195087	NM_002827.3	A	G	p.H208R	Benign	0
NL4	1	ZNF618	9:116811441	NM_133374.2	G	A	p.R527Q	Probably damaging	0.000017
NL6	1	SNRNP200	2:96955598	NM_014014.4	G	A	p.A960V	Probably damaging	0.000033
NL8	1	PIK3C2B	1:204415166	NM_002646.3	C	T	p.A866T	Probably damaging	0
NL11	2	MAP3K6	1:27688646	NM_004672.3	C	T	p.A451T	Benign	0.000008
		ZNF292	6:87970762	NM_015021.1	A	C	p.E2472A	Benign	0
NL12	1	COG5	7:107013177	NM_006348.3	A	C	p.Q264P	Probably damaging	0
NL13	1	MKI67	10:129901675	NM_002417.4	T	A	p.E2472A	Benign	0
NL14	1	FBXL13	7:102553612	NM_145032.3	G	C	p.T310R	Probably damaging	0
NL15	2	PXDNL	8:52321174	NM_144651.4	T	C	p.I1004V	Benign	0
		TLE2	19:3002420	NM_001144761.1	G	A	p.R674C	Probably damaging	0.000076
NL19	2	LRIG2	1:113616050	NM_014813.1	G	A	p.V8L	Probably damaging	0.000041
		OR2T11	1:248789610	NM_001001964.1	A	T	p.F274L	Possibly damaging	0
NL20	1	SREBF1	17:17723001	NM_004176.4	G	C	p.L188V	Possibly damaging	0
G1	1	PLD1	3:171392324	NM_002662.4	A	G	p.Y732C	Probably damaging	0
G2	1	SYT15	10:46963852	NM_181519.2	A	G	p.M371V	Benign	0.001571
G3	2	DOPEY2	21:37618674	NM_005128.2	C	A	p.P1466T	Benign	0
		AHNAK2	14:105416174	NM_138420.2	T	A	p.C1872S	Benign	0.008125
G4	1	HYAL4	7:123508339	NM_012269.2	A	C	p.L4F	Possibly damaging	0
G6	1	SRP72	4:57368013	NM_006947.3	A	G	p.K668E	Probably damaging	0
G8	1	ZNF793	19:38028548	NM_001013659.2	C	T	p.R330X	NA	0
G9	1	C6orf89	6:36887382	NM_152734.3	T	C	p.I292T	Benign	0.000231
G13	1	PRKD3	2:37501677	NM_005813.3	C	G	p.T513S	Benign	0
G14	1	RTEL1	20:62324513	NM_016434.3	C	T	p.R957W	Probably damaging	0.00005
G16	2	CAB39	2:231682556	NM_016289.3	A	G	p.N261D	Possibly damaging	0
		ABL2	1:179086640	NM_005158.4	C	T	p.A397V	Probably damaging	0
G17	1	TMEM131	2:98378517	NM_015348.1	G	C	p.V1624L	Benign	0.000009
G18	1	OR1N1	9:125289536	NM_012363.1	C	T	p.R13G	Benign	0.000018
S2	1	TECTA	11:121016433	NM_005422.2	A	G	p.Y1238C	Benign	0.000016
S4	4	OAS3	12:113379385	NM_006187.2	C	T	p.S63L	Benign	0.000083
		BAHCC1	17:79409915	NM_001080519.2	G	A	p.G514R	Benign	0.000026
		ZNF24	18:32917633	NM_006965.2	C	G	p.P224A	Benign	0
		FBXW8	12:117383291	NM_012174.1	A	C	p.Q116H	Probably damaging	0
IT1	2	NPTXR	22:39222573	NM_014293.3	G	A	p.G344R	Possibly damaging	0.000030
		RBM33	7:155530234	NM_053043.2	A	C	p.E195A	Possibly damaging	0
IT2	1	C10orf90	10:128192697	NM_001004298.2	G	A	p.G455S	Probably damaging	0.000015
IT3	2	ADAM33	20:3651693	NM_025220.2	G	A	p.R734X	NA	0.000563
		PTCRA	6:42893368	NM_138296.2	G	T	p.G265V	Probably damaging	0
IT4	2	LTV1	6:144181653	NM_032860.3	G	A	p.E296K	Possibly damaging	0
		SARDH	9:136535790	NM_007101.3	G	A	p.P804L	Benign	0.000034
IT5	3	ASCC3	6:100988080	NM_006828.2	G	T	p.D1912Y	Probably damaging	0
		C8B	1:57422504	NM_000066.3	G	T	p.C110F	Probably damaging	0
		NEK10	3:27173451	NM_001031741.3	A	G	p.R370X	NA	0
IT6	1	TRPM4	19:49671909	NM_017636.3	G	A	p.G123S	Probably damaging	0
IT7	2	SLC6A17	1:110714820	NM_001010898.2	T	C	p.I142T	Probably damaging	0
		VPS11	11:118949581	NM_021729.4	G	A	p.E714K	Benign	0
IT8	1	CACNA1B	9:141014708	NM_000718.3	G	T	p.R2042L	Probably damaging	0
IT9	1	COL15A1	9:101748312	NM_001855.4	G	T	p.R189L	Probably damaging	0
IT10	1	MFSD12	19:3557319	NM_021731.2	C	T	p.G28D	Probablydamaging	0
IT11	1	ZFP64	20:50704964	NM_199427.2	A	C	p.S399R	Probably damaging	0
IT12	1	MRGPRG	11:3239743	NM_001164377.1	G	C	p.V101L	Benign	0
IT13	2	ARHGEF18	19:7533837	NM_001130955.1	G	A	p.A1015T	Benign	0.000053
		TJP3	19:3730576	NM_001267560.1	G	A	p.R162H	Possibly damaging	0.000083
IT14	1	RRP1B	21:45095000	NM_015056.2	A	G	p.I169V	Benign	0
IT15	4	CLCN2	3:184075179	NM_001171087.2	G	A	p.R290Q	Probably damaging	0
		LTBP4	19:41131902	NM_001042544.1	G	A	p.A1392T	Probably damaging	0
		PDCD1	2:242794902	NM_005018.2	G	A	p.G103R	Benign	0.000030
		SLC9A5	16:67289730	NM_004594.2	C	T	p.R270C	Probably damaging	0
IT16	2	EPHA2	1:16464557	NM_004431.3	A	G	p.Q368R	Benign	0
		PCK1	20:56140150	NM_002591.3	C	T	p.A458V	Probably damaging	0.00003
IT17	1	PKD1	16:2158656	NM_000296.3	C	A	p.A2171D	Probably damaging	0.00019
IT18	2	ODZ3	4:183522220	NM_001080477.1	C	G	p.P219A	Benign	0
		PAPPA	9:118950074	NM_002581.3	G	A	p.V353M	Possibly damaging	0
IT19	1	CEL	9:135939864	NM_001807.4	T	A	p.V50E	Probably damaging	0
IT20	1	PROS1	3:93598107	NM_000313.3	G	A	p.R515H	Benign	0
IT21	1	ZWILCH	15:66821226	NM_017975.3	G	A	p.V336I	Benign	0
IT22	1	SHD	19:4290547	NM_020209.3	C	T	p.L314F	Benign	0
IT23	1	SH2D3A	19:6755290	NM_005490.2	C	T	p.T178M	Benign	0

Note: Trios without DNM are not included.

This resulted in an average of 0.84 coding de novo mutations per trio, which is higher compared with previous studies in ALS with de novo mutation rates of 0.53 (Chesi et al., 2013) and 0.39 (Steinberg et al., 2015) (mean difference with previous studies combined 0.38, 95% confidence interval [CI] 0.14–0.62, P = 0.003), but consistent with the expected distribution in the general population (Samocha et al., 2014) (mean difference 0.14,95% CI −0.06 to 0.34, P = 0.16).

The number of de novo mutations per patient follows a Poisson distribution (Supp. Fig. S1; P = 0.17) indicating that the occurrence of multiple de novo mutations in certain individuals does not differ from what one would expect based on a de novo mutation rate of 0.84. There was no gene in which homozygous mutations were identified more than once when filtering our datasets for exonic variants with a MAF < 0.01 (Supp. Table S1). Compound heterozygous alterations were identified in OBSCN and VCAN in two trios each; however, no other gene had biallelic alterations in more than one trio (Supp. Table S2).

When we combined our trios with the previously published ALS trios (Chesi et al., 2013; Steinberg et al., 2015) (173 trios in total), the overall de novo mutation rate in ALS was 0.64. This also does not differ from the mutation rate in the general population (mean difference −0.058, 95% CI −0.18 to 0.067, P = 0.40).

3.2 |. Pathway enrichment analysis

Next, we investigated whether genes containing de novo mutations were enriched for certain functional pathways as has been shown in previous studies, using DAVID functional annotation analysis (Huang et al., 2009). Our 69 de novo mutation genes showed enrichment for phosphoproteins (P = 0.0086; Supp. Table S3).

We next combined our gene list with the previously published genes with de novo mutations in ALS (Chesi et al., 2013; Steinberg et al., 2015) to determine whether the total set of genes was enriched for specific pathways (111 de novo mutations in 110 genes; Supp. Table S4). The combined genes showed enrichment for phosphoproteins (P = 0.0004); however, the “chromatin,” “transcription regulation,” and “cell cycle” pathways showed no significant association in our or in the combined dataset.

3.3 |. Protein-protein interaction analysis

To analyze whether the de novo mutations have occurred randomly or in genes that are interconnected, we used the Disease Association Protein-Protein Link Evaluator (DAPPLE) (Rossin et al., 2011). This program analyzes the connectivity between genes using known protein interaction networks, and determines by permutation the likelihood of such interaction. Our total set of 110 genes with denovo mutations showed no more direct interactions than one would expect by chance (1,000 permutations, P = 0.57; Fig. 1A). Subsequently, we added major ALS genes from the ALS Online Database (ALSoD) and three recently discovered ALS genes (TUBA4A [Smith et al., 2014], TBK1 [Cirulli et al., 2015; Freischmidt et al., 2015], and NEK1 [Brenner et al., 2016; Kenna et al. 2016]) to the de novo mutation dataset (Supp. Fig. S2A). Although the known major ALS genes do show a network with higher connectivity than expected by chance (P = 0.03; Supp. Fig. S2B), combining these genes with our de novo mutation genes did not lead to an increase in connectivity (P = 0.41; Fig. 1B). This indicated that there is no overall increased interaction between the genes harboring de novo mutations in our study and known ALS genes.

FIGURE 1.

Protein interaction network of genes with de novo mutations by DAPPLE. Interacting genes are connected by a line, color represents the origin of the genes, genes found to harbor de novo mutations in this and previous studies are represented in orange, genes with a known association with development of ALS are represented in blue. A: The connectivity of the de novo mutation genes in the total network is not higher than one would expect by chance (P = 0.57). B: Protein interaction network after addition of the 34 known ALS genes. Two larger networks can be identified: one centered around FUS and one around TUBA4A.The connectivity in the total network is not significant (P = 0.41)

3.4 |. Comparison with de novo mutations in a control population

To further investigate whether the observed de novo mutations in ALS are likely to contribute to the disease, we used de novo mutation data from the GoNL, a whole-genome genotyping database of 769 healthy individuals including 250 child-parents trios from The Netherlands (Genome of the Netherlands Consortium, 2014). Using this database, we examined whether the de novo mutations identified in our study have occurred in genes that are relatively prone to de novo mutations or not. The GoNL mutation rate map (Francioli et al., 2015) was used to calculate the likelihood for each gene of having a de novo mutation in respect to the per-base mutation rate and gene size. We compared our de novo mutation genes to this population-based null distribution using the nonparametric Wilcoxon rank sum test, and found that the 69 genes identified by us have a higher overall probability for the occurrence of de novo mutations (P = 1.7 × 10⁻¹¹). In addition, the previously reported genes with de novo mutations in ALS showed a similar pattern (Chesi et al., 2013; P = 1.4 × 10⁻⁴; Steinberg et al., 2015; P = 3.0 × 10⁻⁴), with both studies implicating genes with intrinsically elevated de novo mutation rates. The three studies combined also showed a clear shift toward genes with a higher de novo mutation rate (P < 1.0 × 10⁻¹⁵; Fig. 2). This demonstrates that the identified de novo mutations in ALS patients are within genes that in general are more likely to harbor de novo mutations, thus challenging the view that these de novo mutations are disease-specific or disease-causing.

FIGURE 2.

Distribution of the genes with de novo mutations compared with all genes in the genome. Genes with de novo mutations are from the combined data from this study, the study by Chesi et al. (2013) and the study by Steinberg et al. (2015). A: Average de novo mutation rate (DNM rate) per base of the genes with de novo mutations (blue) compared with the GoNL de novo database distribution (red); in general, the genes with de novo mutations have a higher per-base mutation rate (P < 1.0 × 10⁻¹⁵). B: Gene size (in base pairs) of the genes with de novo mutations (blue) compared with gene size of all genes from the genome (red); the genes with de novo mutations are on the whole larger (P < 1 × 10⁻¹⁵). C: Probability of de novo mutations per gene (the product of de novo mutation rate per base and gene size, blue) compared with the GoNL de novo database distribution (red); the genes with de novo mutations have, in general, a higher probability of de novo mutations (P < 1.0 × 10⁻¹⁵)

4 |. DISCUSSION

In this large international cohort of ALS patient-parent trios, we discovered that pattern and rates of de novo mutations in ALS patients in general do not differ from the general population. We could not find evidence for the previously reported role for de novo mutations in the development of ALS.

Due to the high mean age of symptom onset in ALS patients, it is more difficult to acquire DNA from unaffected parents compared with disorders with an earlier onset, for example, autism and schizophrenia. Therefore, this study included multiple international cohorts and combined this with previously published ALS trios. With these 173 trios, we could detect a de novo rate increase of 0.15 with a power of 0.8. A limitation of this study is that a potential smaller effect size could be missed due to sample size, although the clinical consequence of such a small effect can be questioned. Therefore, we also used DAVID, DAPPLE, and the GoNL de novo mutation data to further look into these mutations but found no differences between de novo patterns between patients and controls.

Using DAVID pathway analysis, we found no evidence for the previous reported association of de novo mutations with ALS via chromatin regulation, transcription regulation, or cell cycle pathways. This indicates again that the results of DAVID pathway analyses should be interpreted with caution if only a small number of genes are analyzed (Elbers et al., 2009).

De novo mutation rates increase with paternal age (Kong et al., 2012) and indeed a higher mean paternal age could be observed in patients with intellectual disability (Hehir-Kwa et al., 2011) and autism spectrum disorders (Neale et al., 2012; O’Roak et al. 2012). The average paternal age in our 82 trios was 28.8 years (data missing in seven trios), which is lower than in autism and intellectual disability where de novo mutations do play a role. A previous study also showed that paternal age of ALS patients is not higher than the paternal age of controls (de Jong et al., 2013), which is in line with our findings.

The trios in this research are sequenced on either Complete Genomics (cohort from The Netherlands) or Illumina (other cohorts) sequencing platforms. Although these platforms have different methods to assess the genome, comparative research showed these methods to have a large overlap in called SNVs, with a high false-positive rate in the platform-specific calls (Lam et al., 2012). Therefore, SNVs found with these two methods can be combined.

This study does not analyze de novo structural variants. Although structural variants could be important for explaining the missing heritability in ALS, analysis of structural variants from exome data is still challenging. Also, compared with SNVs, structural variants have a larger variation between different sequencing platforms (Lam et al., 2012). Therefore, these structural variants were not included in this study and further research is needed to assess their role on ALS development.

Our results do not rule out that individual de novo mutations in specific genes, as is reported for de novo mutations in FUS (Hübers et al., 2015), can be pathogenic in specific patients. However, these seem only to represent a minority of cases. In general, our results thus challenge a major role of de novo mutations in the development of ALS.