Retrospective analysis of a clinical exome sequencing cohort reveals the mutational spectrum and identifies candidate disease-associated loci for BAFopathies

Abstract

Purpose

The BRG1/BRM-associated factor (BAF) complex is a chromatin remodeling complex playing a critical role in gene regulation. Defects in the genes encoding the BAF subunits lead to BAFopathies, a group of neurodevelopmental disorders with extensive locus and phenotypic heterogeneity.

Methods

We retrospectively analyzed data from 16,243 patients referred for clinical exome sequencing (ES) with a focus on the BAF complex. We applied a genotype-first approach combining predicted genic constraints to propose candidate BAFopathy genes.

Results

We identified 127 patients carrying pathogenic, likely pathogenic variants or de novo variants of unknown clinical significance (VUS) in 11 known BAFopathy genes. Those include 34 patients molecularly diagnosed through ES reanalysis with new gene-disease evidence (N= 21) or variant re-classifications in known BAFopathy genes (N=13). We also identified de novo or predicted loss-of-function variants in four candidate BAFopathy genes, ACTL6A, BICRA (implicated in BAFopathy during this study), PBRM1, and SMARCC1.

Conclusion

We report the mutational spectrum of BAFopathies in an ES cohort. A genotype-driven and a pathway-based reanalysis of ES data identified new evidence for candidate genes for BAFopathies. Further mechanistic and phenotypic characterization of additional patients are warranted to confirm their roles in human disease and to delineate their associated phenotypic spectrums.

INTRODUCTION

The BRG1/BRM-associated factor (BAF) complex, the mammalian form of SWItch/Sucrose Non-Fermentable (SWI/SNF) complex, is a highly conserved chromatin remodeling complex playing a critical role in gene regulation, tumor suppression, and neural development. More than 29 genes encoding the subunits in the BAF complex have been identified in mammals¹. While the BAF complex is typically composed of 10–15 subunits, the exact subunit composition is variable at different time points during development and is specific to different cell types¹. The combinatorial assembly of the BAF subunits provides the functional specificity required for precise gene regulation in a temporal- and tissue-specific manner. Defects in a subset of BAF-related genes have been shown to cause a heterogeneous group of neurodevelopmental disorders (Supplementary Table 1) collectively termed “BAFopathies”². These disorders are characterized by a wide spectrum of phenotypes including syndromic and nonsyndromic intellectual disability (ID), growth delay, autistic behavior, ectodermal defects, and skeletal anomalies³.

The diagnosis of BAFopathies can be clinically and molecularly challenging. One of the most well-known syndromes among the BAFopathies is Coffin-Siris syndrome (CSS), a neurodevelopmental disorder with extensive genetic and phenotypic heterogeneity. To date, nine genes of the BAF complex have been associated with the CSS clinical phenotypic spectrum, including ARID1A (MIM: 603024), ARID1B (MIM: 614556), ARID2 (MIM: 609539), DPF2 (MIM: 601671), SMARCA4 (MIM: 603254), SMARCB1 (MIM: 601607), SMARCC2 (MIM: 601734), SMARCD1 (MIM: 601735), and SMARCE1 (MIM: 603111). Nicolaides-Baraitser syndrome (NCBRS), caused by missense variants in the helicase domains encoded by SMARCA2 (MIM: 600014), is the most clinically recognizable syndrome within BAFopathies, characterized by a distinct facial gestalt, severe ID, seizures, short stature, prominent interphalangeal joints, and sparse hair^4,5. However, phenotypes of other BAFopathies sometimes make clinical distinction challenging^6–8, suggesting the importance of molecular testing to establish an accurate diagnosis. Recently, “SWI/SNF-related intellectual disability disorders (SSRIDDs)” was proposed to describe the clinical continuum of BAFopathies with syndromic ID and mild CSS at the mildest end, followed by more severe and atypical features in the middle, and NCBRS at the most severe end⁴.

The molecular etiologies of BAFopathies are rapidly expanding as more than half (8/15) of the BAFopathy disease genes were established as disease-associated genes within the last five years^3,9–15. In this study, we investigated the mutational spectrum of genes encoding BAF complex proteins in a cohort of patients (N= 16,243) referred for clinical exome sequencing (ES) with a variety of unselected clinical presentations that often included neurodevelopmental phenotypes. Within this cohort, we first sought to identify variants potentially contributing to molecular diagnosis in BAF-related genes that are known to cause human disorders according to Online Mendelian Inheritance in Man (OMIM). Our results provide evidence that the reanalysis of existing ES data can improve molecular diagnosis among individuals with variable spectrums of neurodevelopmental phenotypes.

Recognizing that genotype-driven approaches have proven to be effective in the identification of novel candidate disease genes, especially the genes involved in crucial developmental pathways such as the cohesin complex¹⁶, we used the genotype-first approach to propose new BAFopathy candidate genes. Specifically, we identified predicted loss-of-function (pLoF: frameshift, nonsense, splicing) variants and de novo missense variants in four candidate genes—ACTL6A, BICRA, PBRM1, and SMARCC1—that encode BAF complex proteins which have not yet or only recently been associated with a specific genetic syndrome.

MATERIALS AND METHODS

Samples

The cohort in this study consists of patients (N= 16,243) referred for ES at Baylor Genetics (BG) between March 2012 and December 2019. The majority of the patients (~87.8%) in the BG ES cohort have neurologic disorders (e.g. DD/ID) based on an estimation from a previous study analyzing 2,000 consecutive patients with ES¹⁷. The informed consent for ES was obtained from all individuals in the cohort. The aggregated analyses of anonymized samples and reporting of de-identified molecular data with minimum clinical information were approved by the Institutional Review Board (IRB) of Baylor College of Medicine (BCM, protocols H-37568 and H-41191). In order to provide additional case-level clinical information, written informed consent (re-consent) was obtained from families of two patients with SMARCC1 variants (Patients N6 and N7) and one patient with a BICRA variant (Patient N4) (protocol H-22769 approved by BCM IRB). Medical records were reviewed for phenotypic characterization in these three patients.

ES and variant interpretation

Trio and proband ES were performed as previously described (Supplementary Information)¹⁷. The exonic coverage of the BAFopathy genes (Supplementary Table 2) and the low coverage regions of these genes (Supplementary Table 3) were documented. The variant classification and interpretation were conducted by a clinical standard based on the ACMG/AMP guidelines¹⁸.

SNP arrays, primarily used as a quality control measurement concurrent to ES, were used to detect large heterozygous deletions or duplications at BG¹⁹. Homozygous/hemizygous deletions were detected using an in-house developed pipeline based on ES read-depth data as previously described²⁰. No homozygous deletions or heterozygous gene-disrupting genic CNV events were detected using the joint SNP array and ES CNV analysis; however, the possibility of the presence of small heterozygous CNVs affecting the BAF complex component genes cannot be excluded.

Chromosome microarray analysis (CMA)

The experimental design and data analysis of CMA were performed according to previously described procedures¹⁹ in Patient N8 and Patient N9.

RESULTS

Mutational spectrum of established BAFopathy genes in the BG ES cohort

We identified variants potentially contributing to a molecular diagnosis of BAFopathy in 127 patients, including 81 with pathogenic variants, 38 with likely pathogenic variants, and eight with de novo variants of uncertain significance (VUS). The VUS(s) included seven de novo missense and one de novo intronic variant (Supplementary Table 4). Most of the variants were heterozygous single-nucleotide variants or small insertion/deletions (SNVs/indels), except for the biallelic compound heterozygous or homozygous variants in ACTL6B in Patient 12 and Patient 15, respectively (Supplementary Table 4). Three patients had decreased ratio of variant reads to total sequencing reads (Patient 21, 18% variant allele fraction (VAF) in ARID1A; Patient 77, 13% VAF in ARID2; and Patient 81, 19% VAF in ARID2), indicating mosaicism, which were confirmed by Sanger sequencing. Mosaic pathogenic variants in ARID1A have been reported in multiple patients with CSS, whose phenotypic severity correlates with the level of mosaicism^6,8. Heterozygous knockout of BAF250a, encoded by Arid1a, is embryonic lethal in mice.²¹ Therefore, truncating variants in ARID1A are reported to be most likely mosaic as a mechanism to escape lethality in liveborn humans.

A genic variant distribution was calculated to determine the per-gene contribution to molecular diagnoses (Fig. 1A). Variants in ARID1B (39%) contributed to the largest portion of molecular diagnoses among the BAFopathy genes, followed by SMARCA4 (10%). Other genes each contributed to less than 10% of the total. Pathogenic variants in ARID1B accounts for 61.5%−68% of resolved cases in phenotype-driven cohorts of CSS patients^8,22. The lower ARID1B contribution (39%) in our study is likely explained by the genotype-driven approach and/or patients with characteristic CSS features might be prioritized to single-gene or panel testing rather than ES.

Fig. 1.

Overview of variants in the genes encoding the BAF complex identified by Baylor Genetics clinical exome sequencing. (A) The distribution of variants by genes in 127 probands. (B) The distribution of variants by variant type. Recurrent variants: 7 missense, 2 nonsense, 1 frameshift, and 1 splicing. (C) The distribution of variants by inheritance at case-level. Patient 12 and Patient 15 carried compound heterozygous and homozygous variants in ACTL6B, respectively, and were considered Biallelic-AR inheritance. (D) Time intervals between the date of original exome analysis and the date of gene-disease association literature publications in the reanalysis cases with updated molecular diagnosis. AD autosomal dominant, AR autosomal recessive.

There were 115 unique variants reported. Eleven of these were recurrently observed in unrelated patients, indicating mutational hotspots (Supplementary Table 4). Nine of these eleven recurrent variants, leading to gain-of-function and/or dominant-negative effect as proposed in the literature, occur at the hypermutable CpG dinucleotide sites inside ACTB, ACTL6B, SMARCA4, or SMARCB1^8,10,23,24. Most of the variants (60%) were pLoF variants, while the missense variants (remaining 40%) appeared to contribute less often to BAFopathies in this study (Fig. 1B). The diminished contribution of missense variants may be explained by: (1) missense variants may be less deleterious than substantial loss-of-function effect caused by pLoF variants, exert less of a burden on cell fitness, and may be better tolerated during organismal development; or (2) missense variants may be underrepresented in the mutational spectrum due to challenges in the interpretation of missense variants in specific genes, particularly SMARCA2, SMARCA4, and SMARCB1, where the reported pathogenic variants are almost exclusively missense variants in specific protein domains⁸. Variant interpretation can be particularly challenging for disorders with extensive phenotypic heterogeneity and/or less established phenotype-genotype correlation.

The variants seen in 103 patients (103/127, 81.1%) occurred de novo in the proband. The variants seen in another 21 patients (21/127, 16.5%), contributing to autosomal dominant (AD) conditions, were not present in the available paternal or maternal sample, but a definitive inheritance pattern could not be established due to the lack of a DNA sample from the other parent (Fig. 1C). The variants seen in the remaining three patients (3/127 = 2.4%) were inherited from the parents: Patients 12 and 15 had biallelic pLoF variants in ACTL6B inherited from the asymptomatic heterozygote parents, consistent with the Mendelian expectation of autosomal recessive (AR) disease association of ACTL6B; Patient 73 inherited a nonsense variant in ARID1B from the father, who was a mosaic heterozygote confirmed by an orthogonal method (Supplementary Figure 1).

Improved molecular diagnosis of BAFopathies by ES reanalysis

Diagnostic variants may not be reported at the time of original analysis due to lack of literature support. We reanalyzed the ES data for possible new molecular diagnoses because of rapidly increased recognition of gene-disease association within the BAF complex during the past decade. The disease associations of eight genes encoding BAF complex proteins, including ACTL6B, ARID2, BCL11A, BCL11B, DPF2, SMARCC2, SMARCD1, and SMARCD2, were established within the past 5 years ^3,9–15. Another seven genes, ACTB, ARID1A, ARID1B, SMARCA2, SMARCA4, SMARCB1, and SMARCE1, were identified as disease-associated genes within the last 10 years^7,24–27. Seventeen percent of patients (21/127) received a molecular diagnosis resulting from newly discovered disease genes. For these 21 patients, the original ES were conducted an average of 27 months (range +6 to −68 months; median −27 months) prior to the publication of the gene-disease association (Fig. 1D and Supplementary Table 4). Eighteen out of the 21 patients (86%) carried variants in the more recently discovered disease genes, including ACTL6B, ARID2, BCL11A, BCL11B, and SMARCC2. Updated molecular diagnoses also resulted from upgraded variant classifications in known disease genes in an additional 13 patients (13/127 = 10.2%), whose original ES were performed on average 29 months (range +100 to −14 months; median +32 months) after the establishments of gene-disease association.

The common underlying conditions for upgrading variant classifications in this cohort included (1) stronger clinical evidence supported by additional patients with the same variant reported in literature or in our internal database (9/13), (2) inheritance classified by familial variant testing (2/13), (3) better clinical correlation enriched by updated clinical information provided by the referrals (1/13), and (4) pathogenicity of a novel missense change supported by an allelic established pathogenic variant of the same amino acid (1/13) (Supplementary Table 4). Notably, 10 out of the 13 patients (77%) who received the upgraded variant classifications carried de novo missense variants, reflecting challenges in the interpretation of de novo missense variants²⁸ especially for newly established genes and the importance of familial variant testing and/or publication of additional patients after the original publication of the gene-disease association.

New supporting evidence for candidate disease genes of BAFopathies

We applied a genotype-first approach in association with predicted genic constraints to investigate candidate disease genes, whose defects may potentially explain the neurodevelopmental phenotypes in undiagnosed patients. We primarily focused on the genes involved in the BAF complex that did not have a clear gene-disease association and were predicted to have loss-of-function intolerance, suggesting possible haploinsufficiency (gnomAD probability of LoF intolerance [pLI] ≥ 0.98). Those genes included ACTL6A, BRD7, BICRA, BICRAL, DPF1, PBRM1, SMARCC1, and SS18L1 (Supplementary Table 1).

Three different de novo heterozygous missense variants in the ACTL6A gene, c.377C>T (p.Pro126Leu), c.1129C>T (p.Arg377Trp), and c.1165C>T (p.Arg389Trp), were identified in Patients N1-N3, respectively (Table 1). All three missense variants were predicted damaging by CADD (Supplementary Figure 2).

Table 1.

Summary of variants in the BAF complex candidate disease genes identified by clinical exome sequencing at Baylor Genetics

Gene (transcript)	Patient	Genomic coordinates (hg 19)	Exon/intron	Nucleotide	Amino acid	Variant type	Zygosity	Inheritance	pLIa	gnomAD allele frequencya
ACTL6A (NM_004301.5)	N1	Chr3: 179291256	exon 4	c.377C>T	p.Pro126Leu	missense	Het	Denovo	1.00	0.00
	N2b	Chr3: 179304340	exon 13	c.1129C>T	p.Arg377Trp	missense	Het	Denovo		0.00
	N3	Chr3: 179304376	exon 13	c.1165C>T	p.Arg389Trp	missense	Het	Denovo		0.00
BICRA (NM_015711.3)	N4c	Chr19: 48197565	exon 8	c.2479_2480delinsA	p.Ala827Thrfs*15	frameshift	Het	Denovo	0.98	0.00
PBRM1 (NM_018313.4)	N5	Chr3: 52651557	intron 14	c.1542–3C>G	N/A	splicing	Het	Denovo	1.00	0.00
SMARCC1 (NM_003074.4)	N6	Chr3: 47651586	exon 26	c.3013C>T	p.Gln1005*	nonsense	Het	Not maternal	1.00	0.00
	N7	Chr3: 47677539	exon 23	c.2463del	p.Asp821Glufs*4	frameshift	Het	Not maternal		0.00
	N8	Chr3:47770487–47779878 (minimun deletion)	exons 4–6	Deletion of exon 4–6	N/A	exonic deletion	Het	Unknown		0.00
		Chr3:47762512–47787230 (maximun deletion)
	N9	Chr3:47779508–47779878 (minimun deletion)	exon 4	Deletion of exon 4	N/A	exonic deletion	Het	Unknown		0.00
		Chr3:47777707–47787230 (maximun deletion)

^a https://gnomad.broadinstitute.org/

^bPatient N2 has been published in PMID_28649782

^cPatient N4 has been published in PMID_33232675

A de novo heterozygous variant, c.2479_2480delinsA (p.Ala827Thrfs*15), in BICRA was identified in Patient N4 with DD/ID, hypotonia, short stature, failure to thrive, feeding difficulties, chronic constipation, and joint laxity. This patient was identified by our genotype-first approach and subsequently included in an independent case series for comprehensive phenotypic comparison with other patients harboring BICRA variants²⁹.

A de novo heterozygous c.1542–3C>G variant in PBRM1 was identified in Patient N5. This variant was predicted by multiple algorithms to activate a cryptic splicing acceptor site that was two base pairs 5’ to the canonical splicing acceptor site (Supplementary Figure 2). SpliceAI also predicted a strong change by loss of the canonical acceptor with a score of 0.91 and gain of a cryptic acceptor 1 bp upstream (5’) to the variant with a score of 0.98. However, RNA samples were not available to demonstrate the exact splicing alteration for this individual.

Interestingly, two heterozygous variants in SMARCC1 were identified in Patients N6 and N7, respectively (Table 1). Patient N6 harbored a heterozygous nonsense variant c.3013C>T (p.Gln1005*) in SMARCC1. RNA study in peripheral blood from Patient N6 suggested nonsense-mediated decay of the SMARCC1 transcript caused by the nonsense variant (Supplementary Figure 3). Patient N7 harbored a heterozygous frameshift variant c.2463del (p.Asp821Glufs*4) in SMARCC1. Both SMARCC1 variants in Patients N6 and N7 were not present in the mothers, while the paternal samples were not available for testing. Extensively variable clinical features were present in Patients N6 and N7. While neurodevelopmental delay may be consistently observed, features involving other systems were variable and often solely observed in isolated cases (Supplementary Table 5).

In an independent cohort consisted of patients with predominant neurodevelopmental disorders and subjected to CMA at BG (N= ~70,000), we identified heterozygous intragenic deletions affecting SMARCC1 in two patients, including an exons 4–6 deletion in Patient N8 and an exon 4 deletion in Patient N9 (Fig. 2). Both deletions were predicted to be out-of-frame. While the collective data of pLoF SNVs/indels (N=2) and copy number deletions (N=2) identified in our clinical cohort suggested that haploinsufficiency of SMARCC1 might be associated with a human disease, the limited inheritance data for the identified variants and non-unifying clinical features observed in a small number of patients precluded a definitive gene-disease association. Therefore, further studies are warranted to establish gene-disease association and shed light on the disease-causing mechanisms.

Fig. 2.

The variants in SMARCC1 reported in the literature and in the current study. (A) SNVs/indels in the SMARCC1 gene that affect its protein product BAF155. The white segment represents the full-length protein, and the black segments represent protein domains. The variants colored in red are reported in the current study. (B) SNVs/indels and exonic deletion variants in the SMARCC1 gene. The white segment represents the full-length gene and the black boxes represent exons. The variants colored in red are reported in the current study. The red segments represent the exonic deletions identified from the BG CMA cohort.

DISCUSSION

In this study, we investigated the mutational spectrum of BAFopathies, a molecularly related group of disorders known for their remarkable genetic and phenotypic heterogeneity, primarily in a clinical ES cohort. We report 127 patients who carried putatively disease-contributing variants in genes with clear association with BAFopathies. We then applied a genotype-first approach on the ES cohort, which revealed genetic evidence complementing recently described diseases gene and supporting disease candidacy for new BAF complex genes.

Novel and recently reported BAFopathy disease genes

Four patients harboring ACTL6A missense/splicing variants were recently reported to have ID/DD, anomalies of the extremities, and dysmorphic features^30,31. RNA and protein studies of an ACTL6A splicing variant in patients’ lymphoblasts suggested a haploinsufficient mechanism³¹, which is consistent with intolerance to pLoF (pLI= 1.00). We identified novel de novo missense variants in ACTL6A in additional patients with neurodevelopmental disorders, substantiating the literature for the disease candidacy of ACTL6A.

We also identified compelling genetic evidence to propose disease candidacy for three other genes involved in the BAF complex, including BICRA, PBRM1, and SMARCC1. Notably, we reported four pLoF variants (two SNV/indels and two exonic deletions) in SMARCC1, encoding a core component BAF155 of the BAF complex. One de novo nonsense variant in SMARCC1 was identified in a pair of monozygotic twins with neural tube defects³². Four truncating variants and one de novo missense variant were recently reported in five unrelated families with congenital hydrocephalus³³ (Supplementary Table 5). Additional manifestations in these patients included neurodevelopmental delay, epilepsy, schizencephalic cleft, and other structural brain abnormalities. Interestingly, incomplete penetrance was noted as three of the truncating variants were inherited from unaffected parents³³. To date, eight pLoF SNVs/indels in SMARCC1 were found to distribute throughout the entire gene (Fig. 2), and the only missense variant was located in SWIRM, a protein domain critical for the protein-protein interaction with BAF47 encoded by SMARCB1 and the protein-DNA interaction with nucleosome DNA³⁴. Although observed at an ultra-low frequency, pLoF variants in SMARCC1 have been identified in databases documenting variations in general populations, including gnomAD (https://gnomad.broadinstitute.org/) and Database of Genomic Variants (DGV; http://dgv.tcag.ca/). Smarcc1 homozygous knockout mice were embryonically lethal, while 20% of the heterozygous mutant embryos displayed exencephaly³⁵. Mice homozygous for a missense variant, syntenic to human SMARCC1 p.Thr417Lys, showed neural tube defect, exencephaly, developmental delay, enlarged pericardium, and died before weaning³⁶. Taken together, our data and recent publications suggested highly variable expressivity and perhaps incomplete penetrance of SMARCC1 variants, which may be attributed to modifying mechanisms involving genetic (oligogenic or modifying factors), epigenetics, environmental, and/or other non-genetic contributors³⁷.

De novo variants in ACTL6A, BICRA, PBRM1, and SMARCC1 were observed in two public databases (Deciphering Developmental Disorders and denovo-db) and one recent publication reporting candidate disease genes from a large cohort of individuals with DD³⁸ (Supplementary Table 6). While the de novo variants from the databases may further support the disease candidacy of these genes, the recent publication using 31,058 parent–offspring trios of individuals with developmental disorders, a patient cohort resembling the BG ES cohort, did not suggest ACTL6A, BICRA, PBRM1, and SMARCC1 to be significantly associated with developmental disorders statistically (Supplementary Table 7). Moreover, a comparison between the abundance of pLoF variants observed in the BG ES cohort and the gnomAD cohort did not suggest an enrichment of pLoF variation in BG ES cohort (Supplementary Table 8). The lack of significance by these studies may be possibly related to the methods and sample cohort being used for the statistical analysis ³⁸. Moreover, the lack of RNA studies for the predicted splicing altering variant in PBRM1 obscured a clear understanding of the biological significance of this variation. Additional clinical and function evidence are necessary to further investigate the disease-association candidacy of these genes.

Penetrance and expressivity in BAFopathies

Extensive phenotypic variability is well known among the BAFopathies. Although the penetrance for CSS appeared to be very high or complete, van der Sluijs et al. discussed the possibility of incomplete penetrance in ARID1B³⁹, the most common causative gene of CSS. Interestingly, patients with IQ in the normal range were reported among 143 patients with ARID1B variants in the same study³⁹. In our study, transmissions of deleterious variants in known BAFopathy genes were observed. Patient 73 inherited a c.6322C>T (p.Gln2108*) nonsense variant in ARID1B from an asymptomatic father who was mosaic for the same variant. The lack of phenotype in the father was possibly due to mosaicism. A de novo c.1582del (p.His528Thrfs*35) frameshift variant in BCL11B was identified in Patient 89, who transmitted the variant to a more severely affected offspring.

ES data reanalysis in rapidly emerging BAFopathy genes

With approximately 250 gene-disease associations and 9200 variant-disease associations being reported annually, timely and rigorous reanalysis of existing clinical genomic data has been demonstrated to effectively improve the diagnostic yield⁴⁰. In this cohort among 127 patients carrying variants in BAFopathy genes, ES reanalysis provided updated molecular diagnoses as a result of newly discovered disease genes in 21 (17%) patients and updated variant classifications in another 13 (10.2%) patients. Most new molecular diagnoses resulted from newly discovered disease genes, which is explained by the fact that many BAFopathy genes were described after 2012. Lack of gene-disease association in the primary database used for gene-disease association (e.g. OMIM) at the time of ES reporting and lag of updated clinical annotation in these databases precluded identification of variants in the most recently discovered disease genes. These variants are not routinely prioritized within the clinical pipeline until the disease association of a gene has been firmly established in the literature. This is consistent with the generally prolonged time interval between the original ES and the first literature report of the BAFopathy (Fig. 1D), suggesting that frequent and periodic update of databases involved in the analytical pipeline is critical to provide the most timely and accurate molecular diagnosis. Rigorous reanalysis embracing the research efforts could facilitate novel disease gene discovering, promoting clinical analyses. This study also showed that updates in the genetic database, literature, and the patient’s clinical phenotype can contribute to an increase in diagnostic yield. These findings, among other reanalysis studies⁴⁰, demonstrated that periodic and timely ES reanalysis may benefit patients and their families. Furthermore, a pathway-based ES reanalysis may promote the discovery of candidate disease genes for future investigation. Although continuous reanalysis may introduce significant burden to clinical practice, it may be improved to be less burdensome and more efficient with the recent advancements using automated or semi-automated approaches⁴⁰.

Challenges in the interpretation of de novo missense variants in BAFopathies

De novo missense variants can be difficult to interpret with respect to pathogenicity and contribution to disease phenotypes. Although de novo inheritance is considered strong supporting evidence for pathogenicity when the patient’s clinical features are consistent with the disease spectrum established by reported genotype-phenotype correlations, a de novo inheritance as a standalone evidence is not sufficient to fulfill the criteria of pathogenic or likely pathogenic variant classification¹⁸. We reported eight patients carrying de novo VUSs (one intronic and seven missense variants that were all absent from gnomAD) in the genes known to be associated with BAFopathies (Supplementary Table 4). The reasoning to classify the de novo variants as VUSs in this cohort included (1) partial phenotype correlation between the gene and clinical features of the patients; (2) the identification of a de novo intronic variant that was predicted to affect splicing yet without other supporting evidence; and (3) the detection of a concurrent potential diagnostic finding in known or other candidate disease genes whose defects also lead to neurodevelopmental disorders that could fit with the patient’s phenotypes. Periodic reanalysis of these de novo variants in light of the accumulated case reports, clinical molecular knowledgebase, and functional studies may help with reclassification in the future.

To summarize, our study contributed to the understanding of the mutational spectrum of BAFopathies within an ES cohort mostly composed of patients with neurodevelopmental disorders. A pathway-focused and genotype-first ES reanalysis enabled improved molecular diagnosis among patients with BAFopathies and identified supporting evidence for novel or recently reported BAFopathy disease genes, namely ACTL6A, BICRA, PBRM1, and SMARCC1, potentially expanding the locus heterogeneity in BAFopathies. Identification of additional patients with potential disease-contributing variants and further elucidation of molecular mechanisms of these candidate disease genes are needed to establish a firm gene-disease relationship.

Data Availability

The datasets supporting the conclusions of this article are included within the article and its additional files. Variants reported in this study have been deposited in ClinVar (https://www.ncbi.nlm.nih.gov/clinvar), with accession numbers SCV001748836.1-SCV001748959.1