Differentiating commensal and disease-associated Enterococcus cecorum isolates in poultry using protein sequences.

Enterococcus cecorum (EC) is an avian enteric commensal, but around the year 2000, it emerged as a frequent causative agent of bacterial chondronecrosis with osteomyelitis (BCO) in older broilers. More recently, around 2020, EC has also been isolated from cases of sepsis (SS) in young birds. Differentiating pathogenic from commensal isolates has therefore become critical for surveillance and control efforts. In this study, we applied amino acid k-mer profiling (k = 5) along with supervised learning algorithms, including Random Forest and Multilayer Perceptron approaches, to identify minimal sets of discriminatory oligopeptides associated with pathogenicity. Using only five k-mers, we distinguished commensal isolates (NONE, n = 80) from all pathogenic isolates (BCO + SS, n = 146) with 86% accuracy. In addition, comparisons between BCO (n = 113) and SS (n = 33) isolates were classified with 90% accuracy using eight k-mers. The discriminatory k-mers mapped to proteins involved in carbohydrate transport, stress response, mobile genetic elements, and metabolic remodeling, reflecting niche-specific evolutionary pressures. Sepsis isolates showed signatures associated with acute systemic pathogenicity, while BCO isolates were enriched for proteins that support chronic persistence in bone tissue. These findings provide new insights into E. cecorum pathogenesis and highlight the potential utility of supervised learning approaches in resolving subtle but functionally relevant proteomic differences. The results have implications for improving targeted interventions and monitoring strategies in poultry production.