Inferring Properties of Transcription from Stochastic Gene Expression

Single-cell experiments reveal considerable fluctuations in the expression of both mRNA and protein. Among genetically identical cells, these fluctuations manifest as remarkably broad distributions of expression. What does the nature of stochastic expression intimate about the mechanisms underlying transcriptional regulation? To understand how specific cis-regulatory mechanisms manifest as noise in expression, we tackled two key methodological deficiencies: (1) a lack of assumption-free approaches for fitting mechanistic models to protein distributions, and (2) methods for isolating biochemical noise (intrinsic) from noise due to cellular environment (extrinsic). To fit the standard stochastic protein model without assumptions, we constrained a search algorithm with solutions to the model\u27s higher order moments. The resulting algorithm enables efficient discovery of solution ensembles representing every kinetic scheme consistent with an observed protein distribution. I find that measurement of protein and mRNA degradation rates should permit estimation of the macroscopic rate constants governing gene ON-OFF transitions, transcription and translation from distribution shape alone. I also found that higher-order moments of intrinsic noise separate naturally from their extrinsic counterparts, in principle enabling intrinsic stochasticity to be estimated by comparing expression of strains containing a variable number of identical genes. To test both frameworks, we assembled S. cerevisiae strains expressing one or multiple copies of GFP reporter genes driven by the heat shock responsive promoter SSA1. In contrast to previous studies, we find that stochastic expression from SSA1 resists decomposition into intrinsic and extrinsic components. Degradation rates appear constant across the population, while transcription rates vary extensively with cellular volume, leading us to predict that a large fraction of noise arises from extrinsic mRNA fluctuations. Consistent with this hypothesis, perturbations to transcription rate dramatically impact the balance of protein noise. Together these data argue for models of stochastic expression that explicitly incorporate fluctuating inputs into transcription