Single-cell heterogeneity in ribosome content and the consequences for the growth laws

Abstract

Across species and environments, the ribosome content of cell populations correlates with population growth rate. The robustness and universality of this correlation have led to its classification as a "growth law." This law has fueled theories about how evolution selects for microbial organisms that maximize their growth rate based on nutrient availability, and it has informed models about how individual cells regulate their growth rates and ribosomal content. However, due to methodological limitations, this growth law has rarely been studied at the level of individual cells. While populations of fast-growing cells tend to have more ribosomes than populations of slow-growing cells, it is unclear whether individual cells tightly regulate their ribosome content to match their environment. Here, we employ recent groundbreaking single-cell RNA sequencing techniques to study this growth law at the single-cell level in two different microbes, S. cerevisiae (a single-celled yeast and eukaryote) and B. subtilis (a bacterium and prokaryote). In both species, we observe significant variation in the ribosomal content of single cells that is not predictive of growth rate. Fast-growing populations include cells exhibiting transcriptional signatures of slow growth and stress, as do cells with the highest ribosome content we survey. Broadening our focus to non-ribosomal transcripts reveals subpopulations of cells in unique transcriptional states suggestive that they have evolved to do things other than maximize their rate of growth. Overall, these results indicate that single-cell ribosome levels are not finely tuned to match population growth rates or nutrient availability and cannot be predicted by a Gaussian process model that assumes measurements are sampled from a normal distribution centered on the population average. This work encourages the expansion of growth law and other models that predict how growth rates are regulated or how they evolve to consider single-cell heterogeneity. To this end, we provide extensive data and analysis of ribosomal and transcriptomic variation across thousands of single cells from multiple conditions, replicates, and species.

Figure 1.. Population average ribosome levels correlate with population growth rates, but single-cell ribosome levels do not.

a.) Yeast were sampled at four growth rates and bacteria at eight. Cell density (black) is measured in cells/mL*10⁷ in yeast and OD600 in bacteria. Growth rate is given in generations per hour (gray) against time in hours. Yeast were sampled at 0.92, 2.7, 5.7, and 8.6*10⁷ cells per milliliter. Bacteria were sampled at OD600 0.5, 1.0, 1.3, 1.6, 2.0, 3.6, 5.3, and 6.0. Yeast and bacteria cartoon icons were created using Biorender. b.) Mean detected rRNA counts per cell correlate with the population growth rate in generations per hour. Trend lines and R²s were calculated by applying a linear regression model to the mean data points in R. c.) rRNA counts per cell do not correlate with population growth rate and are highly variable. rRNA counts for all cells are represented as violin distributions. Means (black dots) and trendlines are the same as in b. R²s were calculated by first applying a linear regression model to the entire dataset, and then confirming that the residuals were approximately normally distributed (absolute value Pearson median skewness < 0.3).

Figure 2.. rRNA abundance in cells with growth associated transcriptional signatures does not differ from cells with stress signatures.

a.) Means and single-cell distributions of transcriptionally inferred growth rates against population growth rates (in yeast). R²s were calculated by applying a linear regression model to the means (black dots) and all data for each single cell (green dots). b.) Transcriptionally inferred growth rate for each cell does not strongly correlate with corresponding rRNA counts. c.) rRNA counts/cell density distributions for the cells with the 25% highest (green) and lowest (gray) inferred growth rates (in yeast) are largely overlapping. Dotted lines represent the density distribution maximum, or the largest mode. d.) and e.) Similarly overlapping distributions are observed for the cells with the highest percentage of the transcriptome dedicated to ribosomal proteins and ribosomal biogenesis genes (lavender) versus stress genes (gray) as defined in yeast, and a similar cohort of genes for B. subtilis (Supplemental Table 1).

Figure 3.. Variation in single-cell growth rates does not explain the observed variation in rRNA abundance.

a.) Coefficient of variation (σ/μ) of single-cell growth rates estimated using gene expression (Figure 2a–c) for cells in each timepoint decreases with population growth rate. b.) Coefficient of variation of single microcolony growth rates decreases with the average growth rate for each imaging time. c.) Coefficient of variation of ribosome counts for cells in each timepoint increases with population growth rate. Inset plots show the CV for ribosomal protein mRNA counts/cell. Trendlines are shown to indicate the direction of correlation with population growth rate. d.) rRNA counts/cell densities for each timepoint split by ploidy in yeast and biological replicate in bacteria.

Figure 4.. Cells show differential subpopulations and signatures of sample timepoint despite having similar ribosome levels.

a.) UMAP embeddings of cells from the most (i-ii) and least (iii-iv) rRNA abundance quartiles colored by the Louvain cluster contain both cells with transcriptional signatures of growth as well as stress. b.) UMAP embeddings of cells from the most (i-ii) and least (iii-iv) rRNA abundance colored by sample timepoint suggest that having similar amounts of rRNA does not necessarily make cells similar at the rest of the transcriptional level.

Figure 5.. Unexpected subpopulations are observed across the growth curve in both yeast and bacteria.

a.) UMAP embeddings of cells from the first timepoint and the last timepoint of the yeast growth curve and timepoints 2 and 3 in bacteria, colored by Louvain cluster. Green clusters include cells with differentially expressed genes more associated with growth and gray clusters with differentially expressed genes associated with stress. Density plots show that the distributions of rRNA counts/cell for green and gray clusters overlap. b.) UMAP embeddings of cells from the first timepoint and the last timepoint of the yeast growth curve colored by retrotransposon gene transcript abundance. Retrotransposons are mobile genetic elements within the yeast genome typically thought to be activated by stress. However, here we observe cells expressing them not only in the final timepoint of the yeast growth curve, but also in the first timepoint in cells with transcriptional signatures of fast growth. c.) UMAP embeddings of bacterial cells from an independent replicate experiment colored by sample time (left) or mRNA-rRNA ratio (right). Cells in the UMAP visual cluster with distinctly high mRNA-rRNA ratios also upregulate genes related to sporulation, and are present in all timepoints, including when cells are growing in rich media. Cartoon diagrams inspired by^,.