Annotation-aware multidomain validation of NucS and MutS-MutL across 23,435 prokaryotic reference genomes

Abstract Background EndoMS/NucS mediates non-canonical mismatch repair in several archaeal and actinobacterial lineages. A 2017 phylogenomic survey found that NucS and canonical MutS-MutL were largely mutually exclusive, with coexistence concentrated in Halobacteria and Deinococcus-Thermus. This study reassessed this pattern in an expanded prokaryotic reference collection and tested how annotation wording relates to complete multidomain protein architecture. Results GFF3 annotations from 23,435 NCBI reference genomes (22,606 Bacteria and 829 Archaea) were screened and annotation-positive candidates were evaluated with Pfam profile HMMs. Complete NucS required NucS_N (PF21003) and NucS_C (PF01939) on the same protein; canonical MutS required five defined MutS profiles on one protein without a same-protein PF01713 warning; and MutL required PF01119 plus PF08676. Annotation screening yielded 1,145 apparent coexistence genomes, whereas 470 met all three profile criteria: 421 Halobacteria, 48 Deinococcota and one highly contaminated Lysobacter korlensis assembly. Across 6,270 unique NucS candidate proteins, 5,189 carried both NucS profiles, 111 were NucS_N-only, 901 were NucS_C-only and 69 matched neither. The 682-genome generic “NucS domain-containing” tier contained 680 testable genomes and 717 unique proteins; 685 were NucS_C-only, eight NucS_N-only, five carried both profiles and 19 carried neither. At genome level, 663/680 testable generic-tier genomes (97.50%) contained at least one NucS-associated domain, but only five contained a same-protein two-domain NucS candidate. Explicit nucS annotations showed 99.63% two-domain concordance; canonical MutS and MutL annotation-positive genomes met strong profile criteria in 96.78% and 97.14%, respectively. The Jaccard overlap between two-domain NucS and strong MutS-MutL positives was 2.18%. Conclusions The contemporary collection confirms the lineage concentration reported in 2017 rather than identifying a new coexistence lineage. The main advance is an architecture-aware audit showing that broad “NucS domain-containing” annotations are usually valid at the domain level but are poor proxies for complete two-domain NucS. Because annotation-negative genomes were not exhaustively searched, the reported counts are validated positives from an annotation-first workflow, not exhaustive prevalence or gene-loss estimates.