Saturation Mapping of MSH2 Splicing Effects to Resolve Variants of Uncertain Significance in Lynch Syndrome

ASHG 2026

Oct 20, 2026
Oncology
Presentation

Connor B Ward, Elena Glick, Ashley PL Marsh, Felicia Hernandez, Kristin Kohler, Jacob O Kitzman

Abstract
Resolving variants of uncertain significance (VUS) is essential to advancing genomic medicine because VUS cannot be used to diagnose, guide treatment, or prevent associated diseases. Currently over half of variants in ClinVar are listed as VUS. Multiplexed assays of variant effect (MAVEs) drive VUS resolution by functionally characterizing variants at scale.

 

While MAVEs often focus on protein-level effects, a clinically significant subset of variants disrupt RNA splicing. We used a high-throughput splicing MAVE to systematically identify these splice-disruptive variants (SDVs). We measured splicing effects for 11,100 SNVs across 14 internal coding exons and proximal flanking introns (+/-25 bp) of MSH2 with pooled splicing reporter libraries and ultra-deep RNA-seq in HEK293 cells. We identified known SDVs, including 144 canonical splice site mutations, and 25 additional literature-curated variants. In total, ~3% (n=300) of variants were SDVs, and an additional ~2% (n=219) showed intermediate splice disruption. Of these SDVs, ~7.5% (n=39) were synonymous and may have previously been overlooked.

 

We validated the clinical impact of this map with patient RNA-seq data from 1,820 carriers of MSH2 variants. ~75% of the SDVs we identified were classified as pathogenic/likely pathogenic in this dataset, while an additional 23% were VUS. ~85% of non-SDVs were classified as benign or VUS. Assay effects were well correlated with SpliceAI prediction scores (r=0.61, p<0.001), with discordant results highlighting key limitations of computational prediction. For instance, a cluster of 4 SDVs in exon 14 disrupt a putative SRSF1 motif (an exonic splicing enhancer), and are weakly scored by SpliceAI (mean=0.12), while a cluster of 4 variants with high SpliceAI scores (mean=0.84) near the exon 9 splice donor have minimal splicing effect in our assay and patient RNA-seq. Beyond variant prioritization, we identify specific aberrant isoforms arising from splice disruption, validated by in vivo usage in the SpliceVault database. While ~60% of these isoforms result in frameshift mutations, a subset preserve the native reading frame. We will present LLM predictions and protein validation assays to further classify these in-frame alterations.

 

This comprehensive functional splicing map provides a critical resource for resolving VUS in MSH2, establishing a scalable framework to improve clinical variant interpretation and explore splicing biology across other disease-associated genes.