Quantitative investigation of gene expression principles through combinatorial drug perturbation and theory

The regulation of gene expression is one of the most fundamental processes in living systems. In recent years, thanks to advances in sequencing technology and automation, it has become possible to study gene expression quantitatively, genome-wide and in high-throughput. This leads to the possibility of exploring changes in gene expression in the context of many external perturbations and their combinations, and thus of characterising the basic principles governing gene regulation. In this thesis, I present quantitative experimental approaches to studying transcriptional and protein level changes in response to combinatorial drug treatment, as well as a theoretical data-driven approach to analysing thermodynamic principles guiding transcription of protein coding genes. In the first part of this work, I present a novel methodological framework for quantifying gene expression changes in drug combinations, termed isogrowth profiling. External perturbations through small molecule drugs influence the growth rate of the cell, leading to wide-ranging changes in cellular physiology and gene expression. This confounds the gene expression changes specifically elicited by the particular drug. Combinatorial perturbations, owing to the increased stress they exert, influence the growth rate even more strongly and hence suffer the convolution problem to a greater extent when measuring gene expression changes. Isogrowth profiling is a way to experimentally abstract non-specific, growth rate related changes, by performing the measurement using varying ratios of two drugs at such concentrations that the overall inhibition rate is constant. Using a robotic setup for automated high-throughput re-dilution culture of Saccharomyces cerevisiae, the budding yeast, I investigate all pairwise interactions of four small molecule drugs through sequencing RNA along a growth isobole. Through principal component analysis, I demonstrate here that isogrowth profiling can uncover drug-specific as well as drug-interaction-specific gene expression changes. I show that drug-interaction-specific gene expression changes can be used for prediction of higher-order drug interactions. I propose a simplified generalised framework of isogrowth profiling, with few measurements needed for each drug pair, enabling the broad application of isogrowth profiling to high-throughput screening of inhibitors of cellular growth and beyond. Such high-throughput screenings of gene expression changes specific to pairwise drug interactions will be instrumental for predicting the higher-order interactions of the drugs. In the second part of this work, I extend isogrowth profiling to single-cell measurements of gene expression, characterising population heterogeneity in the budding yeast in response to combinatorial drug perturbation while controlling for non-specific growth rate effects. Through flow cytometry of strains with protein products fused to green fluorescent protein, I discover multiple proteins with bi-modally distributed expression levels in the population in response to drug treatment. I characterize more closely the effect of an ionic stressor, lithium chloride, and find that it inhibits the splicing of mRNA, most strongly affecting ribosomal protein transcripts and leading to a bi-stable behaviour of a small ribosomal subunit protein Rps22B. Time-lapse microscopy of a microfluidic culture system revealed that the induced Rps22B heterogeneity leads to preferential survival of Rps22B-low cells after long starvation, but to preferential proliferation of Rps22B-high cells after short starvation. Overall, this suggests that yeast cells might use splicing of ribosomal genes for bet-hedging in fluctuating environments. I give specific examples of how further exploration of cellular heterogeneity in yeast in response to external perturbation has the potential to reveal yet-undiscovered gene regulation circuitry. In the last part of this thesis, a re-analysis of a published sequencing dataset of nascent elongating transcripts is used to characterise the thermodynamic constraints for RNA polymerase II (RNAP) elongation. Population-level data on RNAP position throughout the transcribed genome with single nucleotide resolution are used to infer the sequence specific thermodynamic determinants of RNAP pausing and backtracking. This analysis reveals that the basepairing strength of the eight nucleotide-long RNA:DNA duplex relative to the basepairing strength of the same sequence when in DNA:DNA duplex, and the change in this quantity during RNA polymerase movement, is the key determinant of RNAP pausing. This is true for RNAP pausing while elongating, but also of RNAP pausing while backtracking and of the backtracking length. The quantitative dependence of RNAP pausing on basepairing energetics is used to infer the increase in pausing due to transcriptional mismatches, leading to a hypothesis that pervasive RNA polymerase II pausing is due to basepairing energetics, as an evolutionary cost for increased RNA polymerase II fidelity. This work advances our understanding of the general principles governing gene expression, with the goal of making computational predictions of single-cell gene expression responses to combinatorial perturbations based on the individual perturbations possible. This ability would substantially facilitate the design of drug combination treatments and, in the long term, lead to our increased ability to more generally design targeted manipulations to any biological system.