Fig 1.
(A) Scatter plot between mean and noise (squared coefficient of variation, CV2) in single-cell protein numbers of the data reported in [24]). Blue balls are the 49 genes with unimodal dynamics. The remaining seven genes were found to have ‘bimodal dynamics’ and are highlighted and colored in accordance with the colors of the single-cell distributions in (C). The black line is the best fitting equation. (B) Classification of genes as having bimodal expression using manual classification, Bayesian information criterion (BIC), and peak detection, respectively. Shown is a Venn Diagram depicting which genes were classified as bimodal and by which criteria. (C) Best fits of the empirical single-cell distributions of protein levels of each of the seven genes under standard growth conditions that were classified as bimodal by at least one criterion. Each distribution results from merging three biological repeats (Fig D in S1 Text). Also shown is the distribution of the wild type (WT) strain (grey), which does not code for any YFP. The color scheme facilitates data visualization.
Fig 2.
Robustness of gene expression bimodality to perturbations targeting gene expression.
Best fits of the single-cell distributions of protein expression of genes classified as bimodal in standard growth conditions (green), when subjected to stationary growth (orange), novobiocin (yellow), cold shock (blue), rifampicin (pink) and streptomycin (brown), respectively. The control (WT, absence of genes expressing YFP) is shown for comparison. Each distribution results from merging the data from three biological repeats.
Fig 3.
Single-gene expression levels across generations and growth phases.
Shown are best fits to the single-cell distributions of expression levels of (A) elaB, (B) lldD, (C) manX, (D) metK, (E) paaX, (F) pgi, and (G) pyrH. For these measurements, overnight cells are first placed in fresh media, which over time becomes poor, forcing cells to enter stationary growth phase. At that moment, the same cells were inoculated into fresh media, causing them to re-enter the exponential growth phase. The same steps are repeated a second time. Consequently, the same cells go through exponential growth phase three times and stationary growth phase three times as well (Section 1.2 in S1 Text). Measurements are conducted four times, during each generation (early exponential, exponential, early stationary, and stationary phases) resulting in 12 time points.
Fig 4.
(A) Scatter plot between the levels of YFP expressed by the genes classified as bimodal and GFP expressed by the promoters of those genes on low-copy plasmids. Also shown are the best linear fits along with the corresponding 95% confidence bounds, coefficients of determination (R2) and the p-values of t-tests (Section 1.5 in S1 Text). These lines serve for mapping fluorescence intensities of GFP to fluorescence intensities of YFPs. (B) Best fits to the calibrated single-cell distributions of GFP levels expressed from low-copy plasmids [63], under standard growth conditions. Each distribution results from merging the data from three biological repeats. The color scheme facilitates data visualization and the correspondences to the figures in (A), used for the calibration.
Fig 5.
Detailed model of gene expression that includes transcription locking due to PSB (green region), σ factor regulation (pink region), and translation (orange region).
Also included are effects of rifampicin, which hampers RNAP-promoter escape (red region), and streptomycin, causing the formation of non-functional proteins (yellow region). It is assumed that cold shock mainly affects transcription, similarly to novobiocin [10].
Fig 6.
(A) Reduced stochastic model of bimodal gene expression. The model includes the transition between two states differing in transcription initiation rates, which generates bimodal single-cell proteins numbers. Also included is the reaction for RNA production in each state. Finally, there are reactions for RNA translation of RNAs as well as for RNA and protein degradation, which are not transcription state dependent. Perturbations are simplified by assuming that they change rate-constant values, rather than being modeled by independent reactions. All reactions, elements, and rate constants are described in detail in Section 1.7 in S1 Text. (B) Distribution of single-cell protein numbers resulting from simulations of the reduced stochastic model using the reference parameter values in Table B in S1 Text.
Fig 7.
Single cell distributions of protein numbers estimated from simulations of the reduced model when tuning, respectively:
(A) kL, (B) kH, (C) both kL and kH while keeping kH/kLconstant (D) kbindL, (E) kbindH, (F) kescL, (G) kescH, (H) ktr, (I) kdRNA, (J) kdP, and (K) different number of σ38 and different rates of ktr, kdRNA, kdP to mimic the gradual transition from exponential to stationary growth phases (0% and 100%, respectively).