Generation of Single-Cell Transcript Variability by Repression

Abstract

Gene expression levels vary greatly within similar cells, even within clonal cell populations [1]. These spontaneous expression differences underlie cell fate diversity in both differentiation and disease [2]. The mechanisms responsible for generating expression variability are poorly understood. Using single-cell transcriptomics, we show that transcript variability emerging during Dictyostelium differentiation is driven predominantly by repression rather than activation. The increased variability of repressed genes was observed over a broad range of expression levels, indicating that variability is actively imposed and not a passive statistical effect of the reduced numbers of molecules accompanying repression. These findings can be explained by a simple model of transcript production, with expression controlled by the frequency, rather than the magnitude, of transcriptional firing events. Our study reveals that the generation of differences between cells can be a direct consequence of the basic mechanisms of transcriptional regulation.

Keywords: Dictyostelium; RNA stability; heterogeneity; repression; self-organization; single-cell transcriptomics; stochastic differentiation; stochastic gene expression; transcription.

Figure 1

Dynamics of Gene Expression Heterogeneity during Early Dictyostelium Differentiation (A) Single-cell RNA-seq was carried out on 0-, 3-, and 6-hr differentiated cells. Three replicates were carried out at each stage. (B) The relationship between variance (CV²) and mean (read counts) of transcript levels in single 0-hr cells. The 7,670 genes (dots) with more than ten mean counts per cell are shown, with a running median in red. (C) Global noise increases during development. Data show the running medians from the three time points, averaged over all replicates. (D) The CV² distribution for each time point is shown as box-and-whiskers plots, with the white line denoting the median. (E) No branches in developmental trajectories were detected by Monocle. Cells, colored by time point, are shown in the first two components’ space attained by independent component analysis. The black line shows the longest identified path through the minimal spanning tree. See also Figure S1 and Data S1.

Figure 2

Downregulated Genes Show Greater Transcript Variability than Upregulated Genes Variability is described by DM, the deviation from the expected noise value for a given expression level [14]. See also Figures S2 and S3 and Tables S1 and S2. (A) Downregulated genes are more variable than upregulated genes. Plots show DM versus expression for up- and downregulated genes (black and purple, respectively) at 6 hr development. Data are shown for different thresholds of fold change (|FC|) in expression level of each gene between 0 and 6 hr, averaged over three replicates. Bin borders are every 500 genes within the entire dataset, starting from a mean of ten counts. Mean and SEM within each bin are shown. Numbers of up- and downregulated genes for each threshold are shown below. (B) No correlation between RNA stability and gene expression variance. Expression variability in 6-hr cells is plotted against RNA turnover (Pearson r = −0.009). Each dot represents a gene colored by its mean expression level. Degradation units are the ratio of expression before to the expression after 1-hr actinomycin D treatment [7].

Figure 3

A Simple Model of Transcriptional Dynamics Explains the Global Variance Properties of Up- and Downregulated Genes (A) Two-state model of transcriptional bursting. The gene toggles between active and inactive states, with rates k_on and k_off. When active, transcript production occurs at a rate λ with transcript lifetime τ. Transcript burst frequency (the frequency with which the active state occurs) is k_on, although in most models k_on is scaled by τ. Burst size (the amount of RNA produced per burst) is λ/k_off. (B) Stochastic simulation of transcription based on the model in (A) generates different simulated clouds (i–v) from different pre-set distributions of burst size and frequency (from i, where genes vary predominantly in burst size, to v, where genes vary predominantly in burst frequency, with equal contributions of size and frequency in iii). In (ii)–(iv), where both size and frequency contribute more equally, simulated data more closely resemble the experimental data. (C) Intuitive explanation of how controlling the burst parameters affects the variance of up- and downregulated genes. (i) Schematic shows mean expression is increased by increasing either burst size or frequency. (ii) Noise increases with burst size and decreases with burst frequency. (iii) Restricting the range of possible sizes and frequencies means the gene can only sample a limited range of values of mean and noise. In the example shown, the gene is mainly regulated via frequency, so an increase in expression favors a decrease in noise. (D) Matching the experimental data in Figure 2A using the two-state model. Lower expressed members of random gene pairs are more variable, if transcriptional output is determined by burst frequency rather than burst size. Shown are the simulations of randomized selections of genes constrained to have >2-fold changes in expression, allowing genes to have more variability in (i) burst size and (iii) frequency; (ii) where frequency and size vary equally. Low-expressed genes from simulated pairs (purple) and their high-expressed partners (black) are shown. Mean and SEM within each bin are shown.

Figure 4

Differentiation-Induced Genes Show Elevated Transcript Variability in Undifferentiated Cells (A) Plots of variability versus expression level (read counts) for genes that will be up- and downregulated during differentiation (black and purple, respectively) before differentiation onset (0 hr). Bins are defined as in Figure 2. Mean and SEM within bins are shown for different fold-change thresholds. (B) Negative scaling of change in expression (FC) with the change in transcript variability (ΔDM) during differentiation. Variability falls in upregulated genes and increases in downregulated genes. (C) Summary. Genes induced during development are initially more variable than genes that will be repressed. Genes that are repressed become more variable than induced genes. See also Figure S4 and Data S2.