Exploring the Temporal Dynamics of Mobile Genetic Elements in Microbial Communities

Abstract

Mobile genetic elements (MGEs) are fundamental drivers of microbial evolution, enabling rapid adaptation to environmental pressures by carrying genes encoding diverse functions, including toxin degradation1–5, virulence6–8, and antibiotic resistance9–14. Despite their critical role in shaping natural microbial communities and their importance in clinical and bioengineering applications, we lack quantitative frameworks for understanding MGE dynamics and predictive models for their effects on microbial hosts. These limitations are manifested in the ongoing antibiotic resistance crisis15 and challenges in precise and stable microbiome engineering16,17.In this dissertation, I addressed these challenges through two complementary studies that reveal fundamental principles of MGE-driven microbial community adaptations in the first and provide a novel data-driven prediction approach for engineering applications in the second. In one work, I investigated the dynamical properties of a special type of MGE interaction, where a transposon is embedded in a plasmid. Despite distinct maintenance mechanisms, different types of MGEs can form nested structures. Using bioinformatics analysis of 14,338 plasmids in the NCBI RefSeq database, my coauthors and I found transposons to be widespread and significantly enriched on plasmids relative to chromosomes. Using engineered transposon-plasmid systems, I then showed that this nested structure enables rapid and tunable responses of transposon-encoded genes in fluctuating environments. Specifically, transposition maintains a reservoir of the encoded genes, while plasmid copy number dynamics further amplifies the dynamic range of gene dosage, thus enhancing the response speed and stability of transposon-encoded traits. Since transposons have been found across the entire tree of life18 and have been observed nested not only in plasmids19–22 but also in phages23,24 and within other transposons22, 25, this dual-layer control mechanism may present a fundamental adaptive benefit that generalizes to other nested MGE architectures. In the other work, I improved prediction accuracy of the effect of MGEs on microbial community responses using a data-driven approach. While fundamental understanding of MGE dynamics is crucial for predictive control and engineering of microbial communities, precise predictions are hard to achieve, given our limited understanding of MGEs and their dynamics. Moreover, practical applications often require direct predictions based on simple inputs like environmental perturbations. One challenge for these applications is that despite advances in high-throughput technologies, large-scale experiments typically remain prohibitively expensive in many cases. I therefore created a machine learning platform, structure-augmented regression (SAR), that exploits the intrinsic structure of each biological system to learn a high-accuracy model with minimal data requirement, from simple inputs like environmental perturbations. Under different perturbations, each biological system exhibits a unique, structured phenotypic response. This structure can be learned based on limited data and once learned, can constrain subsequent quantitative predictions. I demonstrated that SAR requires significantly fewer data comparing to other existing machine-learning methods to achieve a high prediction accuracy, first on simulated data, then on experimental data of various systems and input dimensions. I then showed how a learned structure can guide effective design of new experiments. This approach thus also has implications for integrations of machine learning prediction and experimental design. Together, these findings contribute to both basic principles of MGE-driven microbial community adaptations and a new practical framework for predicting and controlling microbial responses in environmental and clinical settings.