The characteristics of the 270 individuals are shown in Table 1. The most common phenotype was brain malformation (n=144, 53.3%). In 10 patients, dual phenotypes caused by multiple pathogenic variants were identified. The majority of disease inheritance was autosomal dominant (n=165, 58.9%). A total of 273 SNVs and small indels were detected as pathogenic variants (Figure 1A). TUBA1A variants were the most frequent among these. Additionally, 33 CNVs were found (10.8%, Figure 1B), demonstrating that exome sequencing, including CNV analysis, is useful in the genetic diagnosis of pediatric neurological diseases1,23.
To assess the effectiveness of gnomADv4.0 or 54KJPN in identifying de novo variants in probands, we evaluated the allele frequency of these variants in the databases. A total of 162 de novo variants in autosomal dominant or X-linked dominant genes were confirmed in 164 patients, with one patient having two de novo variants and one patient having three. Five recurrent de novo variants were also observed. Among these, 13 variants (8.0%) were found in the databases (Table 2). An identical variant was found in two unrelated patients. Specifically, two variants were registered in 54KJPN, 11 in gnomADv4.0 exome, and one in both 54KJPN and gnomADv4.0 exome, all with an allele frequency of less than 0.001%. These data demonstrated that pathogenic de novo variants could be observed, albeit very rarely, in the large public cohort databases.
Next, we evaluated the effectiveness of annotation based on ClinVar's pathogenicity classifications. Among the SNVs and small indels identified in this study, 38.1% were registered in ClinVar with P or LP classification (Figure 2A), highlighting great utility of this database. Variants unregistered in ClinVar were account variants for 49.1%.
Among 24 intronic variants, SpliceAI could predict aberrant splicing with delta score equal or above 0.2 in 22 variants (91.7%). Notably, we found four variants which were located more than 10-bp away from the exon-intron boundary and predicted aberrant splicing by SpliceAI (Table 3, and Supplementary Figure 2). Among these variants, the splicing change in WDR37, CEP290 has been confirmed in the previous reports10,24. Three of four variants have been registered as P or LP in ClinVar, including a WDR37 variant which we registered10.
We also evaluated the utility of a phenotype annotation tool, the PhenoMatcher module (https://github.com/liu-lab/exome_reanalysis). Approximately 95% of the candidate genes had maximum PhenoMatch scores of 0.6 or above, and 85.1% of candidate genes have scores of 1.0 or above (Figure 2B). Because maximum PhenoMatch scores of 0.3 is used for a threshold in a previous study22, these data suggest a good correlation between genes and phenotypes, and demonstrate the utility of prioritizing candidate genes.
In this re-analysis, we combined gVCF files of patients by GLNexus. During this process, a FOXG1 variants was filtered out (Supplementary Figure 3), which was called in the gVCF. Multi-sample calling is recommended in GATK best-practice, but we should keep in mind true but low-quality calls might be excluded by the quality filtering step.