RNA splicing refers to the process of removing introns from the initial transcript (preRNA), transcribed from its DNA template, and connecting exons to form a continuous RNA molecule. In eukaryotic cells, sequences near the splicing sites of preRNA are conserved, which include the GT bases at the 5′ donor site of the intron, the AG bases at the 3′ acceptor site, the branch point composed of the polypyrimidine trace and splicing regulatory sequences such as exonic splicing enhancer (ESE), exonic splicing silencer (ESS), intronic splicing enhancer (ISE) and intronic splicing silencer (ISS) (Glisovic et al., 2008; Anna and Monika, 2018). The existence of these conserved sequences ensures the accurate RNA splicing, while mutations in these sequences may lead to structural alteration of the protein products and a variety of genetic disorders (Wang et al., 2012).

According to the human gene mutation database (HGMD), approximately 9% of the pathogenic mutations are splice site mutations (http://www.hgmd.cf.ac.uk/ac/index.php). The splice site mutations can be divided into four categories based on their locations: (I) canonical splice site (CSS) mutations at the intronic +1 and +2 positions of the 5′donor splicing site and the -2 and -1 positions of the 3′ receptor splicing site; (II) mutations in the junction regions from exonic 3bp to intronic 6bp and from intronic 12bp to exonic 2bp, excluding the classic region; (III) exonic missense or synonymous mutations in the ESE, ESS and UTR region; (IV) mutations in deep intron such as the branch point, ISE and ISS (Cartegni et al., 2002). The type II, III, and IV mutations were referred to as noncanonical splice site (NCSS) mutations. Like CSS mutations, NCSS mutations can abolish an existing donor or receptor splicing site, often resulting in loss of a whole exon (exon skipping). It can also indirectly activate a hidden splice site in exons or introns, sometimes resulting partial intron retention or exon shortening (Booth et al., 2018). In some cases, intronic or exonic mutations may create a new splice site, resulting in partial intron retention, exon shortening or formation of pseudo exons (Vaz-Drago et al., 2017).

Detection of NCSS mutations can be difficult, because the targeted sequences are not rigidly conserved as the canonical ones and the locations of mutations are far more variable (Anna and Monika, 2018). To this end, bioinformatic analyzing tools, such as Human Splicing Finder (HSF), Splice Site Prediction by Neural Network (NNSPLICE) and NetGene2, have been developed to evaluate the possible pathogenic effect of the splice site mutations (Anna and Monika, 2018). The results of the computational analysis, however, are only predictive. For NCSS mutations, functional studies such as in vivo RNA sequencing and in vitro minigene analysis, are often needed to verify their exact effect on RNA splicing (Anna and Monika, 2018).

Hereditary deafness has been known for its tremendous genetic heterogeneity (Kremer, 2019; Sheffield and Smith, 2019). Facilitated by recent development and widespread implication of next-generation sequencing (NGS), vast amounts of pathogenic variants in over 100 deafness-causative genes have been documented in recent years (https://hereditaryhearingloss.org/). Variants at NCSS, however, are often classified as variants of unknown significance (VUS) due to limited research methods to readily distinguish the pathogenic ones from the large number of benign polymorphisms (Wang et al., 2012). To this day, NCSS mutations have been reported and functionally verified in only a handful of deafness-causative genes, including a c.1282-12T > A in EYA4 lead to non-syndromic deafness DFNA10 (Hildebrand et al., 2007), a c.6050-15G > A mutation in CDH23 lead to atypical USH1 syndrome (Valero et al., 2019), and a number of NCSS mutations involving splicing of exon 8 in DFNA5 (Booth et al., 2018).

In this study, we identified 6 NCSS and 1 CSS mutations in EYA4, PAX3, DFNA5, PTPRQ and USH1G from 4 multiplex, autosomal dominant families and 2 simplex, autosomal recessive families with hereditary deafness. A workflow has been proposed for analyzing the NCSS variants in genetic hearing loss, including NGS with modified filtering criteria for NCSS mutation, familial segregation and genotype-phenotype correlation analysis, bioinformatic prediction and functional verification.

Targeted or whole-exome NGS has been increasingly employed for mutation screening of heterogeneous diseases such as deafness. The routine sequencing data analysis usually focus on non-synonymous variants in exon and CSS variant only, while the potential NCSS variants are often ignored. Though the exact prevalence of the NCSS pathogenic variants is not clear for hereditary hearing loss, several previous reports and our current study has suggested that it very likely be notable, especially in familial cases with no pathogenic variants identified in known deafness genes (Booth et al., 2018). In this paper, we propose a workflow for detection and verification of NCSS mutations by NGS (Supplementary Figure S1). The modified filtering criteria evaluate all variants, intronic or exonic, synonymous or non-synonymous, for potential negative effect on splicing, which will be followed by bioinformatic analysis and functional verification by either reversed transcriptional PCR or minigene analysis. Other helpful criteria include intrafamilial phenotypic co-segregation, consistency with previously reported gene-specific genotype-phenotype correlation, and identification of pathogenic mutations in both alleles of the recessive causative genes.

In this study, multiplex, autosomal dominant families A-D have pedigrees large enough for effective intrafamilial phenotypic co-segregation analysis. The compound heterozygous (PTPRQ) or homozygous (USH1G) mutations in Family E and F are in agreement with the presumable recessive inheritance as reported by numerous previously studies (Sang et al., 2015; Wu et al., 2018; D'Esposito et al., 2021). Affected members within each family have audiograms that show similar hearing phenotype and are consistent with previous reports for genotype-phenotype correlation of EYA4, DFNA5, PAX3, PTPRQ and USH1G. Family D has additional features perfectly matched to WS3, for which PAX3 is the main causative gene. With these supporting evidences, we were able to narrow down a list of candidate NCSS variants including several that were originally missed by routine NGS filtering process.

In this paper, we predicted the pathogenicity of the six NCSS variants by three computational prediction software HSF, NNSPLICE and NetGene2 (Table 2). Though all six variants were predicted to affect the RNA splicing by at least one computational tool, it remains to be functionally verified as prediction tools alone frequently generate false positive or false negative results. In this aspect, the simplest and most effective approach is targeted sequences reversed transcriptional PCR in tissues of the lesion (Valero et al., 2019). This approach, however, rarely works for deafness-causative genes, as the cochlear tissue is difficult to obtain and none of five genes involved in this study are expressed in the peripheral blood. In this case, minigene analysis provides an in vitro assay for testing RNA splicing effect of NCSS variants, which can be used to confirm whether potential pathogenic variants affect splicing efficiency or activate variable hidden splicing sites (Booth et al., 2018). It also remains to be verified by other hearing-impaired patients carrying the same NCSS variants, which will provide further genetic evidence in support for its pathogenicity.

Based on the minigene analysis results, we concluded that splice site variants at the classical (type I) and junction (type II) region often directly destroy the original acceptor or donor splice site, leading to exon skipping, such as the c.1616+3A > T and c.580G > A variants in EYA4 (Figure 3A). Sometimes NCSS variants can weaken the recognition of the donor or acceptor splicing site and activate a cryptic splicing site in the intron, resulting in exon skipping and/or partial intron retention, such as the c.5426+1G > A and c.6087-3T > G variants in PTPRQ (Figure 3F), or activate a cryptic splicing site in the exon, resulting in shortening of the exon, such as the c.164+5G > A variant in USH1G (Figure 3C). Mutations in the deep intron region (type IV) can destroy the branch point with polypyrimidine tract, resulting in exon skipping, such as c.991-15_991-13del in DFNA5 (Figure 3B), or activate a cryptic splicing site in the deep intron, resulting in partial intron retention, such as the c.322-57_322-8del variant in PAX3 (Figure 3D).

The c.580G > A variant in EYA4 is of particular interest, as under most circumstance it will be interpreted as a p. Asp194Asn missense mutation at a very conservative amino acid position. Though most of the reported EYA4 mutations are truncating mutations leading to haploinsufficiency (Zhang et al., 2004), at least four EYA4 missense mutations have been reported to lead to non-syndromic deafness DFNA10(Tan et al., 2014; Liu et al., 2015; Sun et al., 2015; Cesca et al., 2018). However, our results showed that the c.580C > G variant disrupts the pre-mRNA splicing of EYA4, resulting in exon8 (143 bases) skipping and presumably early termination of the protein translation. In light of this discovery, we designed minigene experiment to analyze the other four previously reported EYA4 missense mutations c.511G > C, c.978C > G, c.1301T > A and c.1643C > G, which all tested negative for any splicing abnormality (Supplementary Figure S2). Nevertheless, it is worth noting that exonic synonymous variants or benign amino acid substitutions may be hidden NCSS variants such as the c.580C > G variant in EYA4, which may deserve re-evaluation and in-depth study.

In conclusion, we identified and functionally verified 6 NCSS variants in EYA4, PAX3, PTPRQ, and USH1G in Chinese Han families with sensorineural deafness. The NCSS variants may be an important cause of genetic hearing loss that demands closer attention in genetic diagnosis.