Co-application of pan-genomics and machine learning uncovers novel insights into the maintenance and evolution of microcystins production trait in Microcystis.

Microcystins (MCs), the potent hepatotoxins produced by toxic strains of Microcystis and other cyanobacteria, pose a major threat to freshwater ecosystems worldwide. However, the regulatory mechanisms, evolutionary origin, and maintenance of this energy-intensive toxigenicity remain largely unresolved. Based on 132 non-redundant Microcystis strains, this study combined pan-genomics analysis, phylogenetic reconstruction, pan-genome-wide association analysis and machine learning approaches to investigate the regulatory and evolutionary basis of MCs production. Results suggest that MCs production likely originated as an ancestral trait in Microcystis, while secondary horizontal gene transfer (HGT) and homologous recombination across the MCs biosynthesis-related (mcy) gene cluster and its flanking regions might have contributed to its distribution among polyphyletic lineages. Enrichment analysis further indicated distinct metabolic strategies between toxic and non-toxic Microcystis strains. Toxic strains are enriched in secondary metabolism, whereas non-toxic strains prioritize core metabolic pathways. Through co-occurrence analysis and multi-method screening, this study identified a candidate type II toxin-antitoxin (TA) system (TumE-TumA) that may be synergistically associated with mcy gene cluster. Structural and energetic analyses predicted a potential interaction between the TumE-family toxin protein and mcyA RNA (ΔiG= -44.6 kcal/mol), suggesting its potential regulatory role in MCs biosynthesis. Taken together, these findings support a proposed co-evolutionary framework in which secondary HGT may contribute to the phylogenetic distribution of mcy gene cluster, while the TA system may form a synergistic network with mcy gene cluster, contributing to the maintenance and evolution of MCs production by balancing metabolic costs with ecological benefits. While these in silico predictions require experimental validation, they provide new insights into the adaptive evolution of cyanobacterial toxigenicity and inform future strategies for managing harmful algal blooms.