Translation elicits a growth rate-dependent, genome-wide, differential protein production in Bacillus subtilis

Abstract

Complex regulatory programs control cell adaptation to environmental changes by setting condition-specific proteomes. In balanced growth, bacterial protein abundances depend on the dilution rate, transcript abundances and transcript-specific translation efficiencies. We revisited the current theory claiming the invariance of bacterial translation efficiency. By integrating genome-wide transcriptome datasets and datasets from a library of synthetic gfp-reporter fusions, we demonstrated that translation efficiencies in Bacillus subtilis decreased up to fourfold from slow to fast growth. The translation initiation regions elicited a growth rate-dependent, differential production of proteins without regulators, hence revealing a unique, hard-coded, growth rate-dependent mode of regulation. We combined model-based data analyses of transcript and protein abundances genome-wide and revealed that this global regulation is extensively used in B. subtilis We eventually developed a knowledge-based, three-step translation initiation model, experimentally challenged the model predictions and proposed that a growth rate-dependent drop in free ribosome abundance accounted for the differential protein production.

Keywords: Bacillus subtilis; global regulation; growth rate; protein production; translation efficiency.

Figure 1. Total RNA and mRNA abundances increase proportionally to the growth rate

1. Total RNA abundance in B. subtilis (blue dots) at different growth rates (0.25 h⁻¹, 0.40 h⁻¹, 0.60 h⁻¹, 1.2 h⁻¹ and 1.7 h⁻¹; this work). The 95% confidence intervals are shown with horizontal and vertical bars. Previously quantified total RNA abundance in B. subtilis (green dots) from Dauner et al (2001). Data are fit with a first‐order polynomial.

2. Scheme of the experimental design developed to quantify the relative abundance of total mRNA and of each transcript across growth conditions. r ₁ and r ₂ stand for total ribosomal RNA amounts, t ₁ and t ₂ stand for total transfer RNA amounts, and m ₁ and m ₂ stand for total messenger RNA amounts in growth conditions 1 and 2, respectively. S _i corresponds to the amount of the ten different in vitro‐synthesized (spike‐in) transcripts. c ₁, c ₂ and c _Si correspond to the amounts of cDNA reverse transcribed from m ₁, m ₂ and S _i. Note that the amount of S _i added to the total RNA extract of either growth condition is identical, which straightforwardly allows calculating the scaling factor between growth conditions (see Appendix section 1.4).

3. Mean‐normalized scaling factor.

4. First‐order polynomial fit of normalized total mRNA abundance (as inferred from the scaling factor and total RNA abundance). The total mRNA abundance measured at slow growth (μ = 0.4 h⁻¹) was then assigned a value of 1.0 and total mRNA abundances across the other conditions were reported normalized to this value (in arbitrary units, AU; note that AU is a quantity.OD₆₀₀ ⁻¹). The 95% confidence interval is shown with shaded areas.

5. The population of transcripts was sorted into two groups using a slow (μ = 0.4 h⁻¹) and a fast growth condition (μ = 1.7 h⁻¹). The first group contained transcripts for which the abundance increased faster than the growth rate (green), and the second group included all other transcripts (orange). The enriched sub‐group of transcripts at fast growth contained the ribosomal mRNAs (dark green) and several non‐ribosomal mRNAs (light green). The corresponding datasets can be found in Dataset EV1.

Figure EV1. Molecular RNA species as function of the rate of growth

1. Total RNA abundance as function of the growth rate. Data are linearly fit (dashed line).

2. First‐order polynomial fit of total RNA abundance (black) and rRNA abundance (red, as calculated from the rRNA percentage within the total RNA pool; panels C and D). The 95% confidence intervals are shown with shaded areas.

3. Percentage of each rRNA species (filled‐in circles) in total RNA as function of the growth rate. rRNA sub‐species are fit with a second‐degree polynomial (dashed line). The corresponding 95% confidence band is given by the coloured area.

4. Percentage of total rRNA species (filled‐in circles) in total RNA as function of the growth rate obtained by summing every rRNA species (from panel C). Total rRNA is fit with a first‐degree polynomial (dashed line). The corresponding 95% confidence band is given by the coloured area.

5. Distribution histograms of one (out of four replicates) representative expression dataset (spike‐in genome‐wide array) per medium.

6. Plots of a representative gene expression dataset from SE‐, S‐, CH‐ and CHG‐grown cells vs. a representative gene expression dataset from CHG‐grown cells. The red line corresponds to the bisector and the blue lines to linear fits. Blue and red lines are identical when plotting two replicates of the very same growth condition (CHG vs. CHG), while with increasing growth rates, expressions of an increasing number of genes are switched off.

7. Normalized mRNA abundance of four constitutively expressed genes (see Fig 2B). The 95% confidence intervals are shown with bars. The corresponding datasets can be found in Dataset EV1.

Figure 2. Protein abundance and translation efficiency of ^fbaATIR_fbaA drop at fast growth

1. Blue dots represent the GFP abundance in different growth conditions (average of 12 technical replicates); the 95% confidence intervals are given as horizontal and vertical bars. Data were fit with a third‐order polynomial (dashed line), and the 95% confidence interval of the fit is shown in grey.

2. Normalized fbaA transcript abundance (i.e. proportion of fbaA transcript within total mRNA × normalized total mRNA abundance from Fig 1D).

3. Relative abundance (measured by qPCR) of the gfp mRNA to the endogenous fbaA mRNA in strain ^fbaATIR_fbaA as a function of the growth rate. Filled circles are the mean of eight replicates in one medium; the 95% confidence intervals are depicted as bars. Data are fit with a third‐order polynomial (dashed line), and the 95% confidence interval of the fit is shown in grey.

4. Deduced translation efficiency (h⁻¹) of the ^fbaATIR_fbaA construct as a function of the growth rate. Note that translation efficiency in h⁻¹ is a slight abuse of notation since we divided protein abundance (in FAU.OD₆₀₀ ⁻¹) by the normalized transcript abundance (AU, quantity.OD₆₀₀ ⁻¹). The 95% confidence interval is shown in grey. The corresponding datasets can be found in [Link], [Link].

Figure EV2. Relative abundance of the gfp mRNA expressed under the control of the P_fbaA or P_hs promoters as function of the growth rate and the resulting ratios of translation efficiencies

1. A, B

   Relative abundance (qPCR) of the gfp mRNA to the natural fbaA mRNA in the different strains as function of the growth rate. Filled‐in circles correspond to 8 replicates per medium. Data are fit with a third‐degree polynomial (dashed line) and the 95% confidence bands are given by the coloured area.

2. C

   The gfp mRNA produced by the ^hsTIR₀ construct exhibits a differential stability with increasing growth rate as compared to that of ^hsTIR_fbaA, the 95% confidence bounds is given as bar and the data are fit with a third‐degree polynomial polynomial (dashed line).

3. D

   Ratios between the relative abundance of gfp mRNAs from the different strains and that of the ^fbaATIR_fbaA strain are invariant with growth rate. Filled‐in circles are the mean of the 8 replicates (from panel A) per medium; the 95% confidence bounds are given as bars and the data are fit with first‐degree polynomial (dashed line).

4. E

   ${log}_{10}\left(\frac{[P_{gfp}]}{[m_{gfp}]}\right)$ as a function of ${log}_{10}(\mathrm{\mu })$ for the ^fbaATIR_fbaA strain (in blue) using the dataset plotted on Fig 2 and linearly fitted (dashed line). The straight line has a slope of −1 and indicates a linear dependency to μ⁻¹.

5. F

   ${log}_{10}\left(\frac{[P_{gfp}]}{[m_{gfp}]}\right)$as a function of ${log}_{10}(\mathrm{\mu })$ for the ^fbaATIR_modif1 and ^fbaATIR_short (in red and green, respectively) using the dataset plotted in Fig 3 and linearly fitted (dashed lines). The straight line has a slope of ‐1 and indicates a linear dependency to μ⁻¹.

6. G, H

   The black dots represent ratios of translation efficiencies (λ) between two strains (^hsTIR_fbaA gfp vs. ^hsTIR_modif2 gfp on panel G; ^hsTIR_fbaA gfp vs. ^hsTIR _modif1 gfp on panel H) and 95% of confidence is given as vertical bars. The dashed curves are computed using the parameters estimated on Fig 4. The corresponding datasets can be found in Dataset EV2.

Figure 3. Nonlinear translation efficiencies entail differential drops in protein abundances with increasing growth rate

1. A

   Circles represent the GFP abundance (in fluorescence arbitrary units by optical density units: FAU.OD₆₀₀ ⁻¹) measured during exponential growth for 12 replicates at a given growth rate for constructs ^fbaATIR_fbaA gfp, ^fbaATIR_modif1 gfp and ^fbaATIR_short gfp. The GFP levels were fit with a third‐degree polynomial (dashed line); 95% confidence intervals are given in the coloured areas.

2. B, C

   Black dots represent the ratio of the GFP abundance after correction of the different mRNA stability by qPCR (i.e. ratio of translation efficiencies, λ) between two strains (^fbaATIR_short gfp vs. ^fbaATIR_modif1 gfp in panel B; ^fbaATIR_short gfp vs. ^fbaATIR_fbaA gfp in panel C) and the 95% confidence intervals are given as vertical bars (see Appendix section 3.1). The dashed curves are computed using the parameters estimated on Fig 4. The corresponding datasets can be found in Dataset EV2.

Figure 4. R _free abundance drops at fast growth and results in different TIR‐specific translation efficiencies

1. Inference of R _free abundance. R _free abundance was fit with a third‐order polynomial (black line); the 95% confidence interval is given by the grey area (see Appendix section 3.3).

2. Set of {K _1i, K _2i} pairs for six synthetic constructs. The three upper plots correspond to synthetic constructs having a translation efficiency as ${\mathrm{\lambda }}_i=\frac{K_{1i}[R_{free}]}{K_{2i}+[R_{free}]}$ and the three lower plots correspond to synthetic constructs having a translation efficiency as${\mathrm{\lambda }}_i=\frac{K_{1i}}{K_{2i}}[R_{free}]$, with $K_{2i}$ large as compared to $[R_{free}]$. The ratio K ₁ ^hsTIRfbaA/K ₂ ^hsTIRfbaA was scaled to 1 in order to relatively estimate the entire set of K _1i and K _2i constants (for detailed explanation, see Appendix section 3.3). Red curves represent model‐based fits of GFP abundance quantified in more than 120 cultures for each bacterial strain (black dots).

3. Relative translation efficiencies of the six strains from panel (B) (^fbaATIR_fbaA gfp in blue, ^fbaATIR_modif1 gfp in red, ^fbaATIR_short gfp in green and ^hsTIR_fbaA gfp, ^hsTIR_modif1 gfp and ^hsTIR_modif2 gfp in black). The grey area corresponds to the space of possible translation efficiencies, which is bounded by the translation efficiency when K _2i is much larger than R _free (^hsTIR_fbaA gfp, ^hsTIR_modif1 gfp and ^hsTIR_modif2 gfp) and when R _free is much larger than K _2i (theoretical values, dashed black line). The corresponding datasets can be found in Dataset EV2.

Figure 5. Genome‐wide estimation of {K _1i, K _2i} pairs and of over a thousands of growth rate‐dependent, gene‐specific translation efficiencies in B. subtilis

1. Plot of the values of K _1i (in log₁₀) vs. K _2i (in log₁₀) for each of the 698 transcripts (i), denoted R _free‐unsaturated, for which translation efficiencies were in the form of a Michaelis–Menten‐type rate law. For illustration purpose, filled‐in circles correspond to the tufA (in red) and sdpC (in blue) transcripts. The grey area represents the variation in R _free (Fig EV3C). 223 transcripts, denoted R _free‐saturated, exhibited a constant translation efficiency (i.e. K _2i ˜ 0, Fig EV3A) and 81 transcripts, denoted R _free‐undersaturated, exhibited a linear relationship between their translation efficiencies and R _free (i.e. K _2i ≫ R _free, Fig EV3B).

2. Gene‐specific, growth rate‐dependent translation efficiencies estimated for 1,002 transcripts (i, index of transcripts) and normalized to the growth‐rate point 0.25 h⁻¹ (colour bar and z‐axis). The plain lines separate transcripts by groups of fifty. The list of the transcripts corresponding to each panel can be found in Dataset EV3.

Figure EV3. Genome‐wide estimation of the {K _1i, K _2i} pairs of each transcript and of free ribosome in B. subtilis using transcriptome and proteome datasets and a Michaelis–Menten‐type rate law

1. Plot of the K _1i (in log₁₀) of each of the 223 transcripts (i, index of transcripts) that exhibited a constant translation efficiency (R _free‐saturated).

2. Plot of the values of K _1i/K _2i ratio (in log₁₀) of each of the 81 transcripts (i, index of transcripts) that exhibited a linear relationship between their translation efficiencies and R _free (R _free‐undersaturated).

3. Inference of R _free abundance from the genome‐wide proteome and transcriptome datasets (black line); the 95% confidence interval is given by the grey area (see Appendix section 3.4). The free ribosome abundance at slow growth was assigned a value of 1.0 and free ribosome abundance in other conditions is reported normalized to this value. The free ribosome abundance estimated from the synthetic reporter library is represented with the dashed black line. The list of the transcripts corresponding to panels (A) and (B) can be found in Dataset EV3.

4. Complete model of protein (P_i) production derived from Fig 7A, with the translation efficiency in parentheses (for more detail, see Appendix section 2).

5. Putative R _free abundance variation with respect to the addition of a translation inhibitor. The R _free abundance in media with 1 mg ml⁻¹ tetracycline (green line) compared with R _free abundance in media without tetracycline (black line).

Figure 6. Challenging the structural sensitivity of the global regulation of translation with respect to perturbations

1. Inhibition of the elongation phase results in an increase in total ribosome abundance (modified from Scott et al, 2010).

2. Circles represent the ratios of translation efficiencies computed at each growth rate for 12 replicate cultures of the ^fbaATIR_modif1  gfp and ^fbaATIR_short  gfp strains. The dark grey, blue and red dots correspond to the ratios obtained in the presence of 0, 0.5 and 1 mg ml⁻¹ tetracycline, respectively. The green and purple arrows overlay the “ratio vs. μ” variations resulting from the addition of tetracycline in rich (CH, CHG) and intermediate (S, TS) media, respectively. The orange area illustrates the little variation in the ratios in poor media (M9SE and M9P). The corresponding datasets can be found in Dataset EV4. The 95% confidence intervals are given as vertical bars.

Figure 7. Exploring the interdependence of free ribosome abundance and both mRNA and ribosome abundances using a three‐step translation initiation model

1. Knowledge‐based, three‐step translation initiation model (for more information, see Appendix section 2.1). The growth rate‐dependent molecular entities and rates of the translation initiation process are the abundance of the translation initiation complex (R _free), the 50S subunit, the mRNA abundance (mRNA_i) and the protein elongation aggregated parameter (k _pi, Table 2; Bremer & Dennis, 2008). The growth rate‐independent parameters are the binding constant of R _free onto the mRNA_i (k _bi), the release constant (k _‐bi) of mRNA_i‐associated R _free (R _bi), the accommodation constant (k _ai) of R _bi on the start codon, the disaccommodation constant (k_‐ai) of R_bi‐accommodated R_ai from the start codon and the constant of the initiation of translation elongation (k_ti).

2. The elementary three‐step translation initiation model gives rise to translation efficiency in the form of a Michaelis–Menten‐like equation with the two integrative constants, K _1i and K _2i (see Fig EV3).

3. Formal relationship relating total ribosome (R_Tot) and mRNA species (m_i) abundances with the free ribosome abundance (R _free; see Appendix section 2.3). The sum of the Michaelis–Menten‐like equations (blue) is an increasing function of global and gene‐specific variables (ϕ([R _free], [m_i], K _1i, K _2i, k _pi)). n indicates the number of different mRNA species.

4. The left and right parts of the equation from panel (C), respectively, in purple and blue, are plotted vs. the free ribosome abundance, [R _free]. The intersection (①) corresponds to the equilibrium in a given growth condition. Following a growth‐rate increase from μ₁ to μ₂ (μ₂ > μ₁), total ribosome abundance increases (from plain to dashed purple line), and the equilibrium (②) is shifted towards a decrease in R _free abundance. The global growth rate‐dependent transcriptome reorganization (Fig 1E) can directly trigger a drop in R _free abundance by increasing titration if a significant fraction of the transcripts from the first class (green) exhibits lower K _2i values than the second class (orange).

Figure 8. Schematic representation of the interdependence of free ribosome abundance and both mRNA and ribosome abundances

The K ₂‐related global reorganization of mRNA expression leads to a drop in R _free with increasing growth rate, which in turn promotes differential protein production. From low to high growth rates, the total mRNA and total ribosomal abundances as well as the protein elongation aggregated parameter (k _pi, Table 2; Bremer & Dennis, 2008) increase, while the free ribosome abundance drops. Titration of R _free is weaker at low than at fast growth because transcripts exhibiting strong K _1i and low K _2i (green TIR) are relatively upregulated and those exhibiting converse properties (orange TIR) downregulated with increasing growth rate. The ribosome density along the mRNA is the combined outcome of the translation efficiency and the elongation rate. The “red” and “green” proteins are differentially produced with increasing growth rate due to the growth rate‐dependent translation efficiency. See also a more detailed representation in Fig EV5.

Figure EV4. Graphical representation of the diversity of growth rate‐dependent controls regulating protein production

1. A–C

   The ^hsTIR_fbaA gfp transcript abundance increases with increasing growth rate (violet to grey), while the translation efficiency strongly decreases since the K ₂ constant does not compensate for the drop in R _free abundance.

2. D–F

   The ^fbaATIR_short gfp transcript abundance increases with increasing growth rate (violet to grey), while the translation efficiency only slightly decreases because the K ₂ constant partially compensates for the drop in R _free abundance.

3. G–I

   The ^dhaSTIR_dhaS gfp transcript abundance decreases with increasing growth rate (violet to grey), which is the consequence of its chromosomal location at the vicinity of the terminus of replication. The translation efficiency strongly decreases since the K ₂ constant does not compensate for the drop in R _free abundance.

4. J–L

   The ^ackATIR_ackA gfp transcript abundance strongly increases with increasing growth rate (violet to grey) and compensate for the strong drop in translation efficiency. The ackA gene is not supposed to be regulated neither positively nor negatively in the growth media used to reach the four different growth rates.

Data information: Plots showing GFP abundances (A, D, G and J) and GFP abundances (B, E, H and K) of the genetic constructs ^hsTIR_fbaA gfp, ^fbaATIR_short gfp, ^dhaSTIR_dhaS gfp, ^ackATIR_ackA gfp, respectively. Protein abundance (coloured area in C, F, I and L) at slow (violet) and fast (grey) growth depends on the dilution rate, the free ribosome abundance, the cognate transcript abundance and the Michaelis–Menten‐type K _2i constant in growth‐ rate‐dependent manner.

Figure EV5. Schematic representation of the interdependence of free ribosome abundance and both mRNA and ribosome abundances

The K ₂‐related global reorganization of mRNA expression leads to a drop in R _free with increasing growth rate, which in turn promotes differential protein production. From low to high growth rates, the total mRNA and total ribosomal abundances as well as the protein elongation aggregated parameter (k _pi, Table 2; Bremer & Dennis, 2008) increase, while the free ribosome abundance drops. Titration of R _free is weaker at low than at fast growth because transcripts exhibiting strong K ₁ and low K ₂ (green TIR) are relatively upregulated and those exhibiting converse properties (orange TIR) downregulated with increasing growth rate. The ribosome density along the mRNA is the combined outcome of the translation efficiency and the elongation rate. The “red” and “green” proteins are differentially produced with increasing growth rate due to the growth rate‐dependent translation efficiency.