Varseq

Illumina fastq files were quality checked using FastQC,⁹ and low-quality nucleotides and adaptors were trimmed using fastp (Chen et al., 2018 (link)). Reads were then aligned to the reference human genome version GRCh38/hg38 using BWA-MEM v0.7.17.¹⁰ All bam files were cleaned by local realignment around indel sites, followed by duplicate marking and recalibration using Genome Analysis Toolkit v3.8.1.6. BamUtil v1.4.14 was used to clip overlapping regions of the bam file in order to avoid counting multiple reads representing the same fragment. The genotypability of the FMR1 gene was calculated using CallableLoci in GATK v3.8, with a minimum read depth of 10. CollectHsMetrics by Picard v2.17.10 was used to calculate fold enrichment to determine enrichment quality. Variants were called using HaplotypeCaller (GATK v4.1.8.0). Variant filtering was then carried out according to the GATK Best Practices for exomes. Variants were also filtered by quality (filter PASS) and by location within the FMR1 gene. The accuracy of variant calling for each replicate was calculated using SNPSift, comparing their genotypes with the GIAB NA12878_HG001 annotated VCF file,¹¹ based on variants called by at least two different pipelines. Variants were annotated using VarSeq (GoldenHelix, Bozeman, MT, United States) to screen clinical databases of germline mutations: ClinVar and HGMD Professional v2020.1.

LOH analyses took advantage of the predisposing mutation. When the predisposing mutation was a point mutation, LOH analysis was performed utilizing VCP filtered sequencing data (.vcf-files) on the predisposing MMR gene mutation regions obtained from tumor and normal samples by VarSeq (GoldenHelix^®). The ratio of variant allele (Alt) to reference allele (Ref) reads was determined in tumor (T) and matching normal (N) DNA and the LOH ratio calculated using the following formula: R = (Alt:Ref)_T/(Alt:Ref)_N. The thresholds for LOH and putative LOH are specified in Ollikainen et al. [43 (link)] When the predisposing MMR mutation was a large deletion, LOH analysis was performed by MLPA (with SALSA P003-C1 for MLH1 and MSH2 and 072-C1 for MSH6, MRC Holland, Amsterdam, The Netherlands), and the results interpreted according to Zhang et al. [44 (link)].
Data on somatic point mutations in MLH1, MSH2, and MSH6 were obtained as part of the Nimblegen Comprehensive Cancer Panel. The promoters of MLH1, MSH2, and MSH6 were investigated for methylation by MS-MLPA as described in Valo et al. [8 (link)] and Niskakoski et al. [7 (link)].

In silico evaluation of somatic single nucleotide variants (SNVs) was conducted using VarSeq (GoldenHelix^®) (Supplementary Tables 1 and 5). VarSeq includes 6 individual algorithms to predict the effect of amino acid substitution on protein function: SIFT (http://sift.jcvi.org/), PolyPhen-2 [45 (link)], MutationTaster [46 (link)], MutationAssessor (http://mutationassessor.org/r3/), FATHMM [47 (link)–49 (link)], and FATHMM MKL Coding (http://fathmm.biocompute.org.uk/). In the second hit analysis (Supplementary Table 1), splicing consequences of SNVs in splice site regions were predicted using Human Splicing Finder (http://www.umd.be/HSF3/). Somatic mutations in MMR genes were checked against the InSIGHT database (Leiden Open Variation Database, LOVD v. 2.0 Build 36; http://chromium.lovd.nl/LOVD2/colon_cancer/home.php) for pathogenicity classifications (Supplementary Table 1). Somatic mutations in MMR genes as well as those affecting the top 72 colorectal tumor and 10 ovarian tumor-associated genes were also assessed for possible presence in the Catalogue of somatic mutations in cancer (COSMIC v71, GRCh 37; http://grch37-cancer.sanger.ac.uk/cosmic) (Supplementary Tables 1 and 5).

This human study was approved by Institutional Review Boards at Children’s Wisconsin and Einstein Medical Center Philadelphia. Written informed consent including research analysis and photo publication if applicable was obtained for every participant. Exome sequencing was undertaken by Psomagen (Rockville, MD) and analyzed with VarSeq (Golden Helix, Bozeman, MT). In silico analysis of variants of interest included filtering for frequency <0.001 in the general population in gnomAD v2.1.1 [11 (link)] and for predicted effect upon the protein. The effect of missense variants on protein function was further analyzed by two combined analysis tools (CADD phred hg19 and REVEL). Samples were first analyzed for variants in known MAC and ASD genes as previously described [12 (link), 13 (link)]. Trio analysis in negative cases identified ARHGAP35 as a candidate in two families and screening for variants in this gene specifically identified one more case. Sanger sequencing was used to confirm variants and for segregation analysis. An additional case was identified through clinical genome sequencing and Matchmaker Exchange Databases [14 (link)]. Variants in ARHGAP35 were named based on reference sequence NM_004491.4 and human Genome Build hg19 and evaluated according to ACMG/AMP guidelines [15 (link)].

ONDRISeq VCF files of the ONDRI cases were imported into VarSeq® (Golden Helix, Bozeman, MT, United States) and variants were annotated with sequence ontologies. Minor allele frequencies (MAFs) were obtained from the Genome Aggregation Database (gnomAD v.2.0.1v3 non-neuro)^{37 (link)}. Rare (MAF < 0.01), nonsynonymous variants were prioritized. Further assessment of variants was performed to identify those in genes known to contribute to Mendelian forms of the patient’s disease of diagnosis and those classified as pathogenic or likely pathogenic in ClinVar^{38 (link)}, Online Mendelian Inheritance in Man (OMIM)³⁹ , and/or the Alzforum Mutation Database⁴⁰ . All identified variants were considered those likely to be contributing to Mendelian forms of disease.
All samples were genotyped for C9orf72 using both amplicon length analysis and repeat-primed polymerase chain reaction (PCR), as previously described^{41 (link)}. Harbouring >30 repeats is a commonly accepted genetic cause of ALS and FTD^{41 (link),42 (link)}, and therefore was the cutoff used to determine those with pathogenic repeat expansions.

The BAM and VCF files generated for each patient were imported into VarSeq® (version 1.4.8; Golden Helix, Inc., Bozeman, MT) for annotation of each genetic variant. SNVs were identified following methods that have been described in detail previously (1 (link)). Assessment of CNVs in ABCA1, APOA1, and LCAT was performed using the VarSeq-CNV® caller algorithm. To identify CNVs, the algorithm uses depth-of-coverage information contained within each patient BAM file and compares it to coverage information from a set of “reference” samples that have been previously confirmed to not carry any CNVs. CNVs were called based on comparative increases in read-depth, indicating a duplication of genetic material, and comparative decreases in read-depth, indicating a deletion of genetic material. The criteria in which CNVs were called has been described in detail previously (9 (link)).