The ATM Splice-ome: Charting the multifaceted splicing landscape of ATM
ASHG 2026
Laura I. Hudish, Terra Brannan, Kristin Kohler
Background
Pathogenic variants in the DNA damage response gene ATM are associated with hereditary cancer susceptibility and ataxia-telangiectasia. While many pathogenic variants directly result in protein truncation, disruption of splicing represents an additional and underrecognized mechanism of pathogenicity. Splice-altering variants may arise not only at canonical splice sites but also from exonic substitutions that alter splicing regulatory elements or intronic variants that activate pseudoexonization. Because exonic splice variants can be misinterpreted as missense changes and deep intronic variants may escape routine annotation, the landscape of ATM variants that disrupt splicing remains incompletely characterized in clinical diagnostic cohorts.
Methods
Likely pathogenic (LP) and pathogenic (P) single nucleotide variants with a demonstrated splice impact were analyzed. Variants identified during routine clinical diagnostic testing included canonical splice variants, first/last exonic nucleotide variants, exonic missense and synonymous variants, and non-canonical intronic variants. SpliceAI predicted events with scores >0.2 across a 500bp flanking region around each variant were analyzed.
Results
Among 253 eligible ATM splice-altering variants, canonical splice alterations were the most frequent (56.52%), followed by first/last exonic nucleotide variants (10.28%), exonic missense variants (9.88%), and deep intronic variants (7.51%). Splice-altering variants were distributed across the gene but showed exon-level enrichment in CDS 20 and 52, each having 10 events. Exonic missense splice variants occurred throughout ATM but were non-uniformally distributed with CDS 52 showing the highest burden (5 events). Deep intronic variants were also observed across ATM but were not detected beyond CDS 43.
SpliceAI showed strong performance for non-canonical splicing variants. Among exonic missense splice variants 96% had SpliceAI scores >0.2 and 73% exceeded 0.5. Deep intronic variants showed even stronger prediction, with 100% scoring >0.2 and 75% exceeding 0.5.
Conclusions
Over 30% of splice-altering LP/P ATM variants are non-canonical and likely missed without RNA analysis but accurately predicted by SpliceAI. These findings highlight that deep intronic and exonic missense variants can carry a substantial risk of pathogenicity through splice-altering mechanisms and that SpliceAI is a dependable tool for identification.
Related publications