Discovery of novel Salmonella enterica effectors through combining machine-learning and experimental validation

Type III secretion systems (T3SSs) enable symbiotic and pathogenic Gram-negative bacteria to secrete effector proteins and to translocate them directly into host cells, where they modulate host responses to benefit the bacteria. Salmonella enterica encodes two distinct T3SSs (T3SS-1 and T3SS-2) that together translocate at least 49 known type III effectors (T3Es). Despite extensive study, the complete repertoire of S. enterica’ s T3Es remains incomplete, and additional, yet unidentified, effectors are likely expressed by S. enterica serovars. Here, we present an integrated pipeline that combines machine-learning (ML)-based prediction with systematic experimental validation to identify novel effectors in S. enterica serovars. Our model, trained with known effectors and enriched with biologically relevant features, ranked candidate effectors across four S. enterica serovars. Of these, six proteins were confirmed to be translocated into host cells. Three proteins, STM4158, STM3155 and SEN1975, were translocated into host cells by S . Typhimurium in a T3SS-dependent manner, as inactivation of both secretion systems abolished their secretion and translocation. Notably, STM4158 and STM3155, previously uncharacterized proteins, were translocated by both T3SS-1 and T3SS-2, displayed pro-apoptotic activity, and reduced host-cell viability following transient transfection of mammalian cells, highlighting their potential roles in virulence. SEN1975 (TlpA) was translocated exclusively via T3SS-1. In contrast, STM2138 (SrcA), STM1089, and STM4316 were translocated by both Wild-Type and T3SS-1/2 double mutant, indicating translocation independent of the canonical T3SS-1 and T3SS-2 pathways. These findings expand the known arsenal of S. enterica translocated proteins, uncover non-canonical translocation routes, and underscore the value of integrating computational and experimental approaches for effector discovery. The framework presented here can be broadly applied to identify hidden effectors in diverse pathogens.