Functional Assays Reclassify Suspected Splice-Altering Variants of Uncertain Significance in Mendelian Channelopathies

Abstract

Background:

Rare protein-altering variants in SCN5A, KCNQ1, and KCNH2 are major causes of Brugada Syndrome (BrS) and the congenital Long QT Syndrome (LQTS). While splice-altering variants lying outside 2-bp canonical splice sites can cause these diseases, their role remains poorly described. We implemented two functional assays to assess 12 recently reported putative splice-altering variants of uncertain significance (VUS) and 1 likely pathogenic (LP) variant without functional data observed in BrS and LQTS probands.

Methods:

We deployed minigene assays to assess the splicing consequences of 10 variants. Three variants incompatible with the minigene approach were introduced into control induced pluripotent stem cells (iPSCs) by CRISPR genome editing. We differentiated cells into iPSC-derived cardiomyocytes (iPSC-CMs) and studied splicing outcomes by reverse transcription-polymerase chain reaction (RT-PCR). We used the American College of Medical Genetics and Genomics functional assay criteria (PS3/BS3) to reclassify variants.

Results:

We identified aberrant splicing, with presumed disruption of protein sequence, in 8/10 variants studied using the minigene assay and 1/3 studied in iPSC-CMs. We reclassified 8 VUS to LP, 1 VUS to Likely Benign, and 1 LP variant to pathogenic.

Conclusions:

Functional assays reclassified splice-altering variants outside canonical splice sites in BrS- and LQTS-associated genes.

Introduction

The arrhythmia syndromes Brugada Syndrome (BrS) and congenital Long QT Syndrome (LQTS) are rare autosomal dominant Mendelian diseases, mainly involving variants in cardiac ion channels. BrS is associated with loss-of-function (LoF) variants in the SCN5A sodium channel gene in 20% of patients, while LoF variants in the potassium channel genes KCNQ1 and KCNH2 and gain-of-function (GoF) variants in SCN5A are found in 80% of patients with congenital LQTS ^1,2. These diseases contribute to the >250,000 cases of sudden cardiac death (SCD) each year in the US through fatal ventricular arrhythmias. When affected heterozygotes are recognized early, BrS and LQTS can be clinically managed through medications and/or interventional therapies. However, when not recognized through medical or genetic screening, SCD may present as the sentinel disease manifestation³. Genetic sequencing therefore offers the opportunity to uncover risk in 1) clinically unrecognized heterozygotes identified in unascertained population sequencing studies, or 2) family members of a recognized proband. Multiple approaches are in development to support genome-first approaches and cascade screening, but these efforts are still hampered by issues of clinical interpretability^4,5. The American College of Medical Genetics and Genomics (ACMG) has provided a framework for variant interpretation, spanning benign (B) to pathogenic (P) based on criteria including variant functional evidence, population minor allele frequencies, segregation within families, and computational predictions, among others⁶. While panel sequencing and whole exome/genome sequencing are improving our broader understanding of the genetic basis of disease, a large proportion of detected variants are classified as Variants of Uncertain Significance (VUS) and are therefore not clinically actionable⁷.

Protein-altering variants (including nonsense, frameshift, and missense variants) are a major focus of clinical attention. Variants affecting RNA splicing occupy a comparatively less explored fraction of the genome, despite estimates that they contribute to up to 10% of pathogenic variants in Mendelian diseases⁸. Aberrant splicing from rare variants that alter the 2-bp canonical splice sites flanking each exon is typically characterized as meeting the PVS1 criterion, and therefore often these variants are often classified as P/LP. However, rare splice-altering variants falling outside of these 2-bp sites are more difficult to interpret. Aberrant splicing may arise from more distant intronic or exonic variation by introducing or ablating splice acceptors or donor sequences or by affecting regulatory splicing enhancers or silencers (Figures 1A and 1B; see Supplemental Figure I for position nomenclature)⁹. In silico predictors of splicing consequences have historically had only modest predictive ability.^10,11 However, recent tools leveraging advances in machine learning and large RNA-seq datasets raise the possibility that in silico splicing predictors may be increasingly used to facilitate interpretation of suspected splice-altering variation⁸.

Figure 1. Aberrant splicing from cis-genetic variation and previous variant prioritization.

A) Variants in an exon near the 2-bp splice acceptor (AG) or donor (GT) can disrupt the recognition of the canonical site and lead to exon skipping. Alternatively, variants within the exon may introduce a new splice acceptor or donor site and lead to truncation of the exon, which may introduce frameshifts or damage protein function. This schematic features canonical AG-GT splice sites; rare alternate splice sites are possible. B) Intronic variants adjacent to the 2-bp splice acceptor or donor may disrupt spliceosome recognition of the canonical splice site and lead to exon skipping. Deeper intronic variants may introduce cryptic splice acceptor or donor sites that may lead to transcription of intronic regions (pseudo-exons) that may disrupt the reading frame or compromise protein function. This schematic features canonical AG-GT splice sites; rare alternate splice sites are possible. C) A recent variant interpretation effort curated LQTS and BrS heterozygotes and reported variants. 17 putative splice-altering variants were reported, of which 15 were Single Nucleotide Variants. Of these, 13 were previously not functionally characterized. We analyzed 10 of 13 using a standard minigene approach and 3 using RT-PCR assays from iPSC-CMs.

In this study, we used minigene assays in Human Embryonic Kidney 293 (HEK293) cells to study variant consequences on splicing for 10 putative splice-altering variants in arrhythmia genes identified in a recent cohort of BrS and LQTS patients (Figure 1C). For three variants incompatible with minigene assays, we examined the impact on splicing in induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) that were edited with CRISPR-Cas9¹². We applied these functional splicing assays to reclassify a total of 10 putative splice-altering variants within the ACMG framework, including 8 VUS to Likely Pathogenic.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request. All data drawn from previously published studies used in this analysis received IRB/REC approval. Stem cell work was completed under IRB #9047. Full methods are available online as Supplemental Data.

Results

Minigene Assay Reveals Aberrant Splicing.

We used a minigene construct previously deployed by our group¹³ and others^14,15 to study the effects of SCN5A potential splice-altering variants that have been observed in one or more patients with BrS (Figure 2A). This assay allows the direct comparison of WT versus variant splicing outcomes by studying a specific exon and flanking intronic regions of interest (primers in Supplemental Table I). We first studied the known likely pathogenic (LP) exon-truncating SCN5A synonymous variant c.4719C>T in our system. This variant was previously shown to disrupt splicing in both a minigene assay and RNA isolated from patient peripheral blood lymphocytes¹⁶. Compared to the respective WT minigene construct, the size of the RT-PCR product from the c.4719C>T minigene was consistent with cDNA truncation (Figure 2B). This was confirmed by Sanger sequencing (Figure 2C). All major bands observed by gel electrophoresis were subjected to characterization by Sanger sequencing. For the bands that could be unambiguously identified, we inferred the splice product. For each major band, we inferred the splicing consequence based on reconstruction of the minigene vector and native genomic sequence (Supplemental Table II). We next studied the SCN5A VUS c.1890G>A located at the terminal coding nucleotide position of exon 12 and its WT counterpart. In the minigene assay, c.1890G>A resulted in retention of intron 10 by a splice donor loss as evidenced by RT-PCR product gel band size and confirmatory Sanger sequencing (Figure 2D and 2E). Quantification of the gel band intensity showed significant changes in the WT exon cassette percent spliced in (PSI) in both cases (Figure 2F; all minigene PSI in Supplemental Table III). The effects on transcript composition of these two variants are shown schematically in Figure 2G. This approach was attempted for two additional SCN5A VUS, c.4299G>C and c.4299+6T>C, but the results of the minigene assay were inconclusive due to poor splicing in of the WT exon (Supplemental Figure II)

Figure 2. Minigene assay and studies on SCN5A.

A) Schematic of minigene assay. The PCR product of an exon and segments of flanking introns of interest is cloned into a multiple cloning site (MCS) between two known exons. Mutagenesis inserts the variant. After transfection of the WT and variant plasmids into HEK cells, RNA is isolated, and RT-PCR provides cDNA for analysis of transcript composition. B) Gel of PCR products for the known exon truncating variant c.4719C>T and WT plasmid. The WT band is consistent with higher molecular weight compared to the truncated form induced by the aberrant splice donor gain. All minigene assays are presented in triplicate. C) Sanger sequencing of the exon-exon junctions of WT and truncated RT-PCR products. D) Gel of the WT exon 12 and c.1890G>A RT-PCR products implicating a pseudo-exon gain induced by the variant. E) Confirmatory Sanger sequencing of RT-PCR products for the WT and c.1890G>A VUS. F) Quantification of WT and variant Percent Spliced In (PSI) for 3 independent replicates, respectively. * indicates p < 0.05, independent 2-group Mann-Whitney U Test. G) Schematic of assay results. SCN5A c.4719C>T truncates exon in our assay, whereas the SCN5A variant c.1890G>A induces intron retention.

We studied three KCNH2 VUS c.1128G>A, c.2145G>A, and c.2398+5G>T discovered in the LQTS cohort¹⁷. The variant c.1128G>A induced exon skipping, which is predicted to disrupt the downstream reading frame based on exon size (Supplemental Figure III). The variant c.2145G>A led to an exon skipping event at exon 8 (Supplemental Figure III), similarly resulting in a predicted frameshift. Both nucleotides are located at the last nucleotide position of the coding exon. The intronic variant c.2398+5G>T behaved similarly, leading to exon skipping (Supplemental Figure III). This exon 9 skipping event is also predicted to lead to a frameshift. Quantification of the respective WT exon cassette PSI showed significant changes in all cases (p < 0.05, Mann-Whitney U Test) (Figure 3A). Schematic depictions of the major splicing outcomes on the transcript are shown in Figure 3B. Raw gels and Sanger traces for all major splicing outcomes are shown in Supplemental Figure III.

Figure 3. Minigene analysis of KCNH2 and KCNQ1 splice variants.

A) Quantification of KCNH2 band intensity corresponding to the PSI of the WT and variant bands. 3 independent replicates per group (* indicates p < 0.05, independent 2-group Mann-Whitney U Test). B) Schematic depiction of the splicing outcome associated with the highest intensity band for each KCNH2 variant. C) Quantification of normal exon PSI across all samples. 3 independent replicates per group (* indicates p < 0.05, independent 2-group Mann-Whitney U Test). D) Schematic depictions of the major splicing outcome for each variant.

We studied 5 candidate LQTS splicing variants in KCNQ1. 3 KCNQ1 VUS (c.683+5G>A, c.1032+5G>A, c.1032G>T) and 2 KCNQ1 LP variants (c.477+5G>A and c.1032G>A) were studied. The c.1032 position is the terminal coding position of this exon and is a hotspot for variation. As a control, we first studied c.1032G>A, one of the most common causes of LQTS among Japanese probands (29 cases, 0 gnomAD controls).¹⁸ This variant has been previously shown to affect splicing in vitro and did so in our assay as well, leading to 2 alternative pseudoexon inclusions (Supplemental Figure IV). The adjacent intronic VUS, c.1032+5G>A, had a similar gel electrophoresis profile and led to multiple pseudoexon inclusions with variable respective cryptic donor sites. A second VUS near this splice junction, c.1032G>T, was also functionally assayed. In this case, 3 pseudoexons and an exon skipping event were observed, with a small amount of WT transcript overlapping with a pseudoexon (Supplemental Figure IV). These consequences are predicted to have similarly deleterious effects on the as those previously studied for c.1032G>A¹⁸. These results show that the KCNQ1 exon 9 splice donor site is highly intolerant to variation. The previously unstudied LP variant, c.477+5G>A, was assayed and showed nearly complete loss of WT splicing, with inclusion of 2 unambiguously defined pseudoexons (Supplemental Figure IV). The retained intron sequence contains a stop codon that would disrupt protein function. In our functional studies, introduction of the VUS c.683+5G>A variant completely abrogated WT splicing, while also introducing a pseudoexon and retaining an exon skipping event moderately observed in the WT (Supplemental Figure IV). The retained intronic sequence would also lead to a stop codon inclusion, which is predicted to disrupt protein function. Quantifications of splicing outcomes for the WT PSI are shown in Figure 3C (p < 0.05, Mann-Whitney U Test). Major splicing outcomes for each variant are depicted schematically in Figure 3D. Raw gels and Sanger traces for all major splicing outcomes are shown in Supplemental Figure IV.

CRISPR/Cas9 and iPSC-CMs Reveal Aberrant Splicing.

Many non-standard splice sites exist at a small percentage of splice sites throughout the human genome¹⁹. Although the minigene assay can reliably assess many variant effects on splicing, it is an in vitro assay that best performs with splicing to an exon on both sides with the canonical AG-GT splice sites. Therefore, outcomes may be confounded when assaying the first or last exon of a gene or assaying rare splice sites that do not use the standard AG-GT sites. Three candidate splice-altering VUSs were incompatible with minigene assays: KCNQ1 c.386+6T>G (first exon), SCN5A c.393–5C>T (AC-GT splice site), and SCN5A c.4437+5G>A (AG-AT splice site). These three variants were therefore studied by introducing each variant with CRISPR-Cas9 into healthy control iPSCs (PAM site variant SpliceAI scores in Supplemental Table IV), differentiating into iPSC-CMs, and then assessing splicing consequences by RT-PCR of isolated RNA²⁰.

The RT-PCR products of the SCN5A VUS c.4437+5G>A and WT RNA were approximately the same size by gel electrophoresis (Figure 4A); however, Sanger sequencing of relevant exon junctions showed aberrant splicing of the edited allele (Figure 4B). We used negative and positive controls in all RT-PCR experiments corresponding to no PCR input and a WT SCN5A or KCNQ1 cDNA construct. TA cloning and sequencing revealed 2 major transcripts corresponding to the WT and predominant edited allele product, inducing a frameshifting variant (Figure 4B). Splicing predictions suggested a cryptic donor site that would lead to a frameshift of the transcript, matching the experimentally observed transcript (Figure 4C). The variant SCN5A c.393–5C>T was previously studied by minigene assays and shown to induce exon skipping²¹; however, due to the non-standard splice site usage, this specific implementation of the assay may lead to inappropriate conclusions. In the CRISPR-Cas9 iPSC-CM model, we observed no change in splicing by RT-PCR (Figures 4D and 4E). RT-PCR of the KCNQ1 VUS c.386+6T>G also showed no disruption of splicing by gel electrophoresis or by Sanger sequencing (Figures 4F and 4G). Further investigations into this variant using primers to detect large pseudoexons and NMD inhibition by cycloheximide failed to show aberrant splicing (Supplemental Figure V). Altogether, these studies implicated SCN5A c.4437+5G>A as splice-altering, and SCN5A c.393–5C>T and KCNQ1 C.386+6T>G as not splice-altering.

Figure 4. Variant functional analysis through CRISPR and iPSC-CMs.

A) RT-PCR analysis of population control iPSC-CM (designated C2) vs SCN5A c.4437+5G>A iPSC-CM. Negative control indicates no PCR input, and positive control indicates SCN5A cDNA construct. B) Sanger trace of exon 25 junction shows normal splicing of WT allele and intron retention in the CRISPR-edited allele. * indicates variant DNA base from CRISPR edit. C) Schematic depiction of predicted splicing outcome of c.4437+5G>A, consistent with experimental findings. D) RT-PCR analysis of population control iPSC-CM (designated C2) vs SCN5A c.393–5C>T iPSC-CM.E) Sanger trace of exon 3/exon 4 junction shows WT splicing of both SCN5A alleles c.393–5C and c.393–5T. F) RT-PCR analysis of population control iPSC-CM (designated C2) vs KCNQ1 c.386+6T>G iPSC-CM. G) Sanger trace of exon 1/exon 2 junction shows canonical splicing of both KCNQ1 alleles c.386+6T>G.

SpliceAI Scores of Affected Variants.

Aggregate SpliceAI scores are shown in Table 1 (individual scores are provided in Supplemental Table V). SpliceAI provides probabilities of losing or gaining a splice acceptor or donor site due to cis-genetic variation. Most variants were predicted to introduce and/or ablate a donor site as their major consequence, consistent with their position at the 3’ end of each exon (12/13). SpliceAI results were generally concordant with our experimental findings. For example, the SCN5A c.393–5C>T VUS was not predicted to alter splicing, matching the experimental findings (Figure 4C). 7/11 variants that disrupted splicing in vitro had SpliceAI scores >0.5. One exception is KCNH2 c.2145G>A, which was not highly predicted to disrupt a splice site by SpliceAI (0.25 aggregate score), but was observed in 2 LQTS patients¹⁷ and caused exon skipping in vitro. SpliceAI also successfully identified the exonic cryptic donor splice site introduced by the LP variant SCN5A c.4719C>T (0.96 aggregate score). Plotting the difference between WT and variant PSI for each minigene construct pair versus SpliceAI score revealed a modest but non-significant correlation (Supplemental Figure VI).

Table 1.

Summary of variants with aggregate SpliceAI scores, assay results, case and control frequencies, and ACMG reclassifications.

Gene	Variant	Aggregate SpliceAI	Assay Type	Assay Result	Case / gnomAD	ACMG Post-Assay	Reclassification
SCN5A	c.393–5C>T	0.21	iPSC-CM	Normal	1 / 0	PM2, BS3, BP4	VUS → LB
SCN5A	c.1890G>A	0.72	Minigene	Aberrant	1 / 0	PM2, PS3,	VUS → LP
SCN5A	c.4299+6T>C	0.75	Minigene	N/A	1 / 0	PM2	VUS → VUS
SCN5A	c.4299G>C	0.80	Minigene	N/A	1 / 0	PM2	VUS → VUS
SCN5A	c.4437+5G>A	0.69	iPSC-CM	Aberrant	2 / 0	PM2, PS3	VUS → LP
SCN5A	c.4719C>T	0.96	Minigene	Aberrant	2 / 2	PM2, PS3, PS4_strong, PP3	LP → P
KCNH2	c.1128G>A	0.34	Minigene	Aberrant	2 / 0	PM2, PS3	VUS → LP
KCNH2	c.2145G>A	0.25	Minigene	Aberrant	2 / 0	PM2, PS3, BP4	VUS → LP
KCNH2	c.2398+5G>T	0.45	Minigene	Aberrant	2 / 0	PM2, PS3	VUS → LP
KCNQ1	c.386+6T>G	0.96	iPSC-CM	Benign	1 / 0	PM2, BS3, PP3	VUS → VUS
KCNQ1	c.477+5G>A	0.75	Minigene	Aberrant	7 / 3	PM2, PS3, PS4_strong	LP → P
KCNQ1	c.683+5G>A	0.29	Minigene	Aberrant	3 / 4	PM2, PS3, PS4_moderate	VUS → LP
KCNQ1	c.1032+5G>A	0.95	Minigene	Aberrant	2 / 0	PM2, PS3, PP3	VUS → LP
KCNQ1	c.1032G>T	0.74	Minigene	Aberrant	1 / 0	PM2, PS3	VUS → LP
KCNQ1	c.1032G>A	0.68	Minigene	Aberrant	9 / 0	PM2, PS3, PS4_strong	LP → P

ACMG Reclassification of Putative Splice-altering Variants.

Following functional assays, we applied the PS3 criteria to 8 variants and the BS3 to 2 variants. Computational predictions from SpliceAI were also included: 2 variants meeting BP4 and 3 meeting PP3 criteria. After integrating these functional and computational findings with the previous ACMG criteria calculated by Walsh et al.¹⁷ (Supplemental Table VI), we reclassified 8/12 VUS to LP, 1 VUS to LB, and 1 LP variant to P (Table 1).

Discussion

Splicing as Disease Mechanism.

It is estimated that up to 10% of pathogenic variants may arise from aberrant splicing⁸. In addition to the widespread annotation of variants disrupting the canonical 2-bp splice sites as meeting the PVS1 criterion, the importance of splice-altering variants outside the canonical 2-bp splice sites has also been shown for Mendelian diseases^22,23. While not all variation close to the exon/intron junction should be assumed to affect splicing, functional assays provide expedient evidence for prospective variant reclassification when observed. Although we did not use SpliceAI to help decide which variants to investigate, SpliceAI scores correlated well with functional assay results (Supplemental Figure VI). This recently developed convolutional deep neural network appears to predict aberrant splicing with much higher accuracy than previous computational predictors⁸. Future in vitro splicing studies might therefore prioritize variants that are predicted to be splice-altering by this algorithm. Notably, the LQTS/BrS consortium dataset did not include any examples of deep intronic variation, commonly defined as >100 bp away from the exon/intron junction²⁴.

Functional Assays Reveal Aberrant Splicing.

We applied functional studies to reclassify a set of VUS that have been observed in clinical cohorts of BrS and LQTS patients. The minigene remains a standard assay to address the fidelity of splicing and has been deployed in many applications^25,26. Although a few reports of high-throughput adaptations of the minigene system have been published, these often remain restricted to a small genomic area²⁷ or are limited by in-frame exon triplet cassettes²⁸. A complementary tool is RT-PCR of primary patient tissue or a secondary tissue developed from patient-derived iPSCs when primary tissue is unavailable. Studying candidate arrhythmia variants in a cardiomyocyte context naturally provides more information than the in vitro minigene assays alone. For example, investigators found many more splicing aberrations in an iPSC-CM model than in minigene assays alone for the LQTS-associated splice variant c.1032G>A¹⁸. We observed aberrant splicing in 1/3 of our CRISPR edited iPSC-CM assays. We did not detect aberrant splicing for the SCN5A c.393–5C>T variant in iPSC-CMs, which was discordant with a previous study that showed exon skipping in a minigene assay²¹. As mentioned above, however, the use of non-canonical splice sites in minigene assays may have caused the previously observed exon skipping events. Further supporting our results is the high frequency of the related c.393–5C>A variant: an allele count of 45 in gnomAD, ~8x higher than the 2.5e-5 cutoff that we and others have used²⁹.

Aberrant Splicing-associated Cardiac Morbidity.

Aberrant splicing is increasingly recognized as a driver of cardiovascular disease. Most of such variation has been implicated in cardiomyopathy genes^{22,23,30–32}. A recent functional genomics variant reclassification effort examined aberrant splicing from a set of 56 hypertrophic cardiomyopathy (HCM) probands with MYBPC3 variants. The investigators performed RNA analysis of 9 such variants outside the canonical splice sites from venous blood or myocardial tissue, and were able to reclassify 6 variants (4 VUS -> LP, and 2 LP -> P)³². In a follow up study, the investigators used patient-derived iPSC-CMs followed by RNA-seq to identify 2 known and 1 novel deep intronic splice variants in HCM probands³⁰. The investigators also found that rationally designed antisense oligonucleotides (ASOs) could correct the aberrant splicing of one of the three variants. In a similar study, investigators combined SpliceAI with cohort WGS to identify known splice-altering variants and a novel splice-altering variant in MYBPC3 from studies of peripheral blood and RT-PCR³¹. A complementary study jointly investigated splice-altering variants in LMNA, a dilated cardiomyopathy (DCM) associated gene, and MYBPC3 using computational tools to prioritize variants, and a high-throughput sequencing-based minigene platform²³. The authors also extended this system to study non-canonical splice-altering variants in TTN²². Aberrant splicing in the cardiac arrhythmias is comparatively less studied. A seminal study identified a branch point-altering variant in KCNH2 that elicited the LQTS phenotype in a multigenerational family using patient tissue and minigene assays³³. More recently, a deep intronic splice-altering variant in KCNH2 was identified in a genotype-negative LQTS family and functionally characterized using a similar CRISPR iPSC-CM framework as used in this study³⁴.

In silico Prediction Concordance.

We found that SpliceAI agreed with our functional assays in most cases (Table 1). For example, our RT-PCR studies of the SCN5A c.393–5C>T variant agreed with the benign predictions of SpliceAI to have little effect. The SCN5A c.4719C>T variant was also strongly predicted by SpliceAI to introduce a new cryptic exonic donor site, consistent with the truncation event observed in our study. In contrast, we observed divergent outcomes with the KCNQ1 c.386+6T>G variant, where we observed no splicing abnormality despite high predictions of disruption (0.96 aggregate score). Further studies excluding large pseudoexon retention and NMD inhibition experiments did not yield evidence of aberrant splicing. SpliceAI also was highly predictive of disrupted splicing at the +5 position (4/5 variant studied had >0.5 aggregate SpliceAI score). The +5 position has frequently been invoked in disease and is particularly intolerant to variation in evolutionary studies³⁵. Experimentally, a recent high-throughput study of non-canonical splice-altering TTN variants also found a large enrichment of splice-altering variants at this position²².

Therapeutic Opportunities.

Splice-altering variants comprise an eminently targetable class of genetic variation, with therapeutic modalities spanning ASOs, traditional small molecules, and emerging gene editing platforms. ASO technologies that leverage coding variation benefit by retaining pharmacokinetic and pharmacodynamic properties over a broad range of possible sequence targets³⁶. Indeed, ‘N of 1’ drug discovery has already been implemented clinically³⁷, and has even precipitated new regulatory processes in anticipation of increased translational opportunities³⁸. Complementary mechanism-agnostic gene therapy approaches such as the ‘suppression-replacement’ strategy may be capable of treating a variety of variant classes in a specified gene, including those that alter splicing³⁹. Altogether, the characterization of splice-altering variants may prioritize therapeutic opportunities in the genetic arrhythmias.

Future Directions.

Future studies can apply advances multiplexed assays to enable rapid high-throughput prospective evaluation of splice-perturbing variants. Further characterizing the fraction of splice-altering variants responsible for the Mendelian channelopathies may inspire future investigations in this area.

Limitations.

HEK293 cells do not capture the full environment of a native cardiomyocyte. Trans-acting proteins may also affect splicing outcomes in native cardiomyocytes. Although minigene assays offer clear functional readouts, the exact consequences of the reading frame in a human gene may remain cryptic due to the limited intronic window of the assay. Predicted termination codons beyond the 100 bp of cloned intronic DNA are therefore necessarily inferred from the observed reading frames. The variants in this study represent only a small fraction of the total quantity of possible splice site-adjacent variants in these three arrythmia-associated genes. Splicing is a complex and stochastic process with a continuum of effect. The current ACMG guidelines do not capture the nuance of this behavior, and only provide a dichotomous (yes/no) framework for variant effect interpretation in this context. An additional challenge is the use of in silico predictors for variants adjacent to exons that use non-canonical splice sites, given the relative paucity of training data. A limitation of the CRISPR iPSC-CM approach is the necessity of an additional PAM site-disrupting variant to prevent Cas9 re-cutting. Although the SpliceAI predictions were low for each variant, we cannot exclude the possibility that a PAM variant could influence the splicing outcome (Supplemental Table IV). In the future, technologies such as base editors or prime editors that do not require a PAM-disrupting variant may be preferred⁴⁰.

Conclusions

We deployed functional assays to reclassify 10 variants near canonical splice sites observed in BrS and LQTS probands. This work suggests that splice-altering variation may play a prominent role in the genetic etiology of Mendelian arrhythmias, consistent with recent work defining splice-altering substrates in the cardiomyopathies^{22,23,30–32}. Methods to rapidly assess suspected splice-altering variation will play an important role in the continued realization of personalized medicine for diagnostic, and potentially therapeutic purposes.”

Nonstandard Abbreviations and Acronyms

ACMG

American College of Medical Genetics and Genomics

ASO

Antisense Oligonucleotide

BrS

Brugada Syndrome

DCM

Dilated Cardiomyopathy

GoF

Gain-of-Function

HCM

Hypertrophic Cardiomyopathy

iPSC-CM

induced Pluripotent Stem Cell-Cardiomyocyte

LoF

Loss-of-Function

LQTS

Long QT Syndrome

NMD

Nonsense Mediated Decay

PAM

Protospacer Adjacent Motif

PSI

Percent Spliced In

SCD

Sudden Cardiac Death

VUS

Variant of Uncertain Significance