Disentangling gene expression burden identifies generalizable phenotypes induced by synthetic gene networks

Large-scale genetic circuits are rapidly becoming critical components for the next generation of biotechnologies and living therapeutics. However, the relationship between synthetic and host gene expression is poorly understood. To reveal the impact of genetic circuits on their host, we measure the transcriptional response of wild-type and engineered E. coli MG1655 subject to seven genomically integrated circuits and two plasmid-based circuits across 4 growth time points and 4 circuit input states resulting in 1007 transcriptional profiles. We train a classifier to distinguish profiles from wild-type or engineered strains and use the classifier to identify synthetic construct burdened genes, i.e., genes whose dysregulation is dependent on the presence of a genetic circuit and not what is encoded on the circuit. We develop a deep learning architecture, capable of disentangling influence of combinations of perturbations, to model the impact that synthetic genes have on their host. We use the model to hypothesize a generalizable, synthetic cell state phenotype and validate the phenotype through antibiotic challenge experiments. The synthetic cell state results in increased resistance to β -lactam antibiotics in gram-negative bacteria. This work enhances our understanding of circuit impact by quantifying the disruption of host biological processes and can guide the design of robust genetic circuits with minimal burden or uncover novel biological circuits and phenotypes.