Filling in the blanks: In silico prediction of novel type III secreted effectors in the Pseudomonas syringae species complex

Abstract

Predicting novel type III secreted effectors (T3SEs) in bacteria remains challenging because of their extensive sequence diversity and our heavy reliance on features derived from previously annotated genes. The available tools for predicting T3SS effectors produce a high number of false positives, which complicates their reliable identification. To address this gap, we developed a new bioinformatics workflow and meta-analysis designed to improve confidence in novel T3SE prediction by integrating multiple complementary approaches, including Effectidor2, Bastion3, EffectiveT3, curated effector databases, all-vs-all BLAST comparisons, comparative genomics, and localization-based features. We applied this approach to 57 complete P. syringae genomes, a diverse species complex whose pathogenicity is closely tied to its type III secreted effector (T3SE) content, allowing us to reduce an initial collection of 283 candidate T3SEs to 15 high-confidence predictions. By applying our newly customized EffRank scoring pipeline, three candidates emerged as high-confidence novel T3SEs in P. syringae strains: PsaNZ45_RS26420 (P. syringae pv. actinidiae ICMP20586), ACOZ4J_RS28825 (P. syringae pv. actinidiae FX219), and RRP28_RS13015 (P. syringae pv. actinidiae Yunnan2.4). Furthermore, our scoring pipeline successfully detected and removed likely false positive T3SEs, such as YenB-like toxins, among the high-confidence candidates that persisted through earlier curation steps, highlighting the challenges and biases of current publicly available prediction tools. Finally, we also characterized the final T3SE repertoires of each strain, which were consistent with phylogroup clustering based on the core-genome. As expected, PG1 strains carried larger and more conserved T3SE repertoires and contributed the majority of the promising novel T3SE candidates, while the likely false-positive YenB-like toxin was detected in PG2.