Single-cell parameter inference reveals kinetic heterogeneity in synthetic mammalian gene expression

Natural fluctuations in gene expression ("noise") impact many cell processes such as differentiation, proliferation, and apoptosis. Negative feedback gene circuits, a gene self-repressing its own transcription, e.g., through protein binding, are often used to reduce the single-gene expression noise and control its impact on investigated phenomena. However, the detailed, quantitative understanding of how temporal and population-level noise is affected by natural mammalian gene expression processes and how this dependence is modified by transcriptional negative feedback is lacking. Here, we present a comprehensive framework combining live-cell imaging of single-cell arrays with likelihood-free inference methods to characterize stochastic gene expression in the time domain and at single-cell resolution. We investigate the temporal and population-level variability in protein levels driven by a negative feedback circuit and compare the results with constitutive expression where the feedback mechanism is broken. We use inferred kinetic rate parameters to validate several hybrid stochastic/deterministic models at their predictive capacity regarding coefficient of variation (CV) versus mean protein-level relation. Our analysis shows significant differences between temporal and population-level noise profiles. We show how feedback strength and cooperativity quantitatively control not only mean protein levels per cell but also their variation range within cell population (CV_p) and across time in individual cells (CV_t).