The Alternative but Intolerant Paradox: Non-Canonical (GC-AG and AT-AC) Splice Sites Break the Standard Evolutionary Rules of Splicing

Abstract Background: The prevailing evolutionary paradigm in splicing dictates that constitutive splice sites are under strict purifying selection, while alternative splice sites act as "evolutionary sponges" that tolerate higher levels of genetic variation to facilitate transcript diversity. However, it remains unknown whether non-canonical splice sites (GC-AG and AT-AC) which are predominantly alternativefollow this standard rule or if their specialized roles subject them to distinct population-genetic constraints. Methods: I conducted a systematic analysis of 492,133 unique human splice junctions from GENCODE v50, isolating 8,007 GC-AG and 278 AT-AC junctions. I quantified their transcript usage using a continuous "architecture fraction" metric, nucleotide-level evolutionary conservation (phyloP across 100 vertebrates), and population variant burden using 1000 Genomes Phase 3 data. Results: Non-canonical junctions are overwhelmingly skewed toward alternative usage (Kruskal-Wallis p = 2.94 × 10⁻²⁹⁶). However, contrary to the standard paradigm, they harbor a significantly lower density of standing population genetic variation (median variant counts: 61.0 for GC-AG and 74.0 for AT-AC) compared to canonical GT-AG sites (median = 81.0). At the nucleotide level, conservation profiles reveal striking divergence: GC-AG sites display attenuated donor motif conservation (mean phyloP = 2.00) compared to GT-AG (mean = 4.50), whereas AT-AC sites exhibit intense hyper-conservation (mean = 6.23). Conclusions: Non-canonical splice sites break the standard "alternative-tolerant" evolutionary rule, exhibiting an "alternative but intolerant" phenotype. This paradox reflects their localization in highly constrained, essential genomic regions. These findings provide the first empirical population baseline for non-canonical sites, offering a vital calibration dataset to prevent the misclassification of variants in clinical genomics pipelines trained predominantly on canonical GT-AG sites.