Molecular basis of reproductive plasticity in termites under environmental stresses and prediction of colony fecundity using machine learning

The reproductive plasticity of social insects provides colonies with tremendous flexibility to respond to environmental changes and adapt to new habitats. However, understanding how gene networks regulate this plasticity to generate complex social phenotypes remains a challenge. Here, we propose an “inhibition-activation-reinhibition” model for the reproductive transformation of female workers using three species of Reticulitermes termites. Through transcriptomic analysis of female workers during this dynamic cycle, we identified functional genes involved in their reproductive transformation. We found that gene expression profiles in female workers responded strongly to queen loss. Furthermore, we confirmed that under severe environmental stress that threatened colony survival, the workers did not activate the gene regulatory networks associated with reproductive transformation. We also demonstrated that piggyBac and tigger transposable element-derived genes in the workers’ brains cooperatively promoted reproductive transformation. Metabolomic changes in female workers were linked to stress signals from queens and the nest environment, primarily involving fatty acid and amino acid metabolism pathways. Finally, we constructed seven machine learning models to elucidate gene expression profiles related to reproductive plasticity and to predict the presence and loss of female reproductives (queens, which are indicators of colony fecundity). Among these, the deep neural network models showed the highest accuracy and practicability for analyzing gene regulatory networks and predicting caste. We propose that female workers, as indicators, carry extensive colony-specific information at the transcriptomic level. These findings provide valuable insights and innovative methodologies for understanding reproductive plasticity in social insects under environmental stress.