Evolutionary Decoding of the Bacillus subtilis Secretome: Insights from Pan-Genomics and Deep Learning

Bacillus subtilis serves as a crucial host for industrial protein production, where the efficiency and regulation of its secretion system represent a central focus of applied research. Despite substantial genomic diversity among strains, current understanding of signal peptides, which are key elements in the secretion process, remains largely based on single model strains, lacking systematic investigation from a pan-genomic perspective. This study integrates pan-genomics and deep learning to construct a pan-secretome profile of 287 B. subtilis strains. Our analysis reveals an open repertoire of signal peptides in B. subtilis. Our results reveal an open pan-signal peptidome in B. subtilis. Core signal peptides exhibit high sequence conservation and primarily direct the localization of housekeeping proteins, whereas accessory signal peptides govern the extracellular secretion of environment-responsive proteins, forming a functional “housekeeping-adaptive” dichotomy. We demonstrated for the first time at a broad scale the widespread evolutionary decoupling between signal peptides and their corresponding mature peptides, uncovering a modular evolutionary mechanism in protein evolution. While the deep learning model achieved high accuracy (88.0%) in discriminating core and accessory gene-encoded full-length proteins, its performance was considerably limited when using signal peptide sequences alone (69.7% accuracy), reflecting the information constraints inherent to short sequences. This study provides a systematic portrayal of the signal peptide evolutionary landscape in B. subtilis at the pan-genome scale, advancing our understanding of the evolutionary principles governing bacterial secretion systems and establishing a theoretical foundation for optimizing industrial protein production through rational signal peptide design.