Spurious regulatory connections dictate the expression-fitness landscape of translation factors

Abstract

During steady-state cell growth, individual enzymatic fluxes can be directly inferred from growth rate by mass conservation, but the inverse problem remains unsolved. Perturbing the flux and expression of a single enzyme could have pleiotropic effects that may or may not dominate the impact on cell fitness. Here, we quantitatively dissect the molecular and global responses to varied expression of translation termination factors (peptide release factors, RFs) in the bacterium Bacillus subtilis. While endogenous RF expression maximizes proliferation, deviations in expression lead to unexpected distal regulatory responses that dictate fitness reduction. Molecularly, RF depletion causes expression imbalance at specific operons, which activates master regulators and detrimentally overrides the transcriptome. Through these spurious connections, RF abundances are thus entrenched by focal points within the regulatory network, in one case located at a single stop codon. Such regulatory entrenchment suggests that predictive bottom-up models of expression-fitness landscapes will require near-exhaustive characterization of parts.

Keywords: expression-fitness landscape; multiscale measurements; peptide chain release factors; regulatory entrenchment; translation factors.

Figure 1. Mapping the underlying determinants of the release factor expression‐fitness landscapes

1. Expression‐fitness landscapes, which connect enzyme expression (microscopic variable) to the growth rate (cellular phenotype), can be dictated by direct or indirect effects. Inset (i): direct effects correspond to reduction in the flux cognate to the perturbed enzyme (protein synthesis rate in the case of translation factors). Inset (ii): indirect effects result from pleiotropic propagation across mechanistic, regulatory, and systemic levels.

2. As a case study, the expression of peptide chain release factors (RFs: RF1, RF2, and associated methyltransferase PrmC), involved in the first step of mRNA translation termination, was tuned around endogenous levels.

3. Strains with inducible copies of RFs, and deleted endogenous genes, were used to systematically vary RF expression. The resulting impacts on the cell internal state (RNA‐seq, ribosome profiling) and relative growth rate s (competition experiments) were measured, leading to precise mapping of expression‐fitness landscapes.

Data information: See also Figs EV1 and EV2 for details on strains and measurement platform.

Figure EV1. Details of RF‐inducible expression constructs

1. A

   Schematic tunable expression system. Inducible constructs are added at safe harbor loci, together with a barcode for competition experiments. The endogenous gene copy is then deleted in a scarless fashion.

2. B, C

   Details of loci with tunable expression cassettes for the orthogonally tunable (B) RF2 and PrmC strain, and (C) RF1 and PrmC strain. Disrupted safe harbor endogenous loci are amyE, lacA, and levB. Control strains were also constructed with blank expression cassettes at these locations (Materials and Methods, Fig EV2F, G, I and J). Inducible repressors XylR and LacI, respectively, responsive to IPTG and xylose are shown in black. Resistance cassettes are shown in yellow. The location of the 8‐nt chromosomal barcode in one arm of the amyE homology region is shown in purple.

3. D

   Fold‐change in RF levels as a function of inducers, with fitted Hill curve, serves as a guide to the eye. Endogenous expression is shown as the horizontal dashed line (fold‐change of 1) and the full attainable dynamic range indicated on the right. See Materials and Methods for calibration to proteome fraction.

4. E, F

   3D RF expression space for all phenotypically profiled conditions (shown separately in Fig 2A–D) shown in (E), and orthogonal projection in 2D subspaces in (F). Dashed lines mark endogenous expression levels.

Figure EV2. Details of relative growth rate measurement

1. A

   Schematic of competition experiments. Pools of barcoded strains competed for ≈ 30 generations with five samplings, barcode frequencies are quantified, and changes in barcode frequencies over time determined. Relative growth rate is λ _inducible/λ _WT = 1 + s, where s is the slope of the log₂ barcode ratio vs. time (number of generations). This process was performed for all induction conditions shown in Fig 2A–D.

2. B

   Schematic the barcode readout procedure, carried out in two PCR steps from genomic DNA extracted from pools of competing strains, with UMI and first index added at the first PCR, and a second index added at the second PCR. Details of the final amplicon for barcode readout are in Dataset EV6.

3. C, D

   Examples of barcode frequency ratios over time for isogenic strain pairs from a single competition experiment, (C) wild‐type vs. wild‐type, and (D) RF2 overexpression vs. wild‐type. Representative strain pairs from experiment E1‐C9 are shown. Inferred s: = λ _inducible/λ _WT − 1 from the linear fit (black line) is shown on the graph. Range of slopes s_min–s_max of subsampled bootstraps (gray lines) is reported as $s_{s_{\mathit{min}}}^{s_{\mathit{max}}}$. Dashed lines correspond to 95% confidence interval. Error bars correspond to estimated noise attributable from Poisson counting noise in UMI counts ($\frac{1}{\log (2)}\sqrt{\frac{1}{N_1}+\frac{1}{N_2}}$, where N ₁ and N ₂ are the respective UMI barcode counts for the compared strain pairs at the corresponding time point).

4. E

   Distribution of measured s for pairs of strains with identical genotype apart from barcode across all experiments (n = 1,253 comparisons, e.g., 4/1,253 experimentally determined s are shown in panel C). The shaded gray area corresponds to the ± 2σ _s = ± 1.2% shown in Fig 2E–H.

5. F, G

   Measured fitness difference to wild‐type for strains with blank expression cassettes. (F) Blank P_xyl at amyE & blank P_spankHy at lacA, strains GLB434–437, n = 644 comparisons. (G) Blank P_spankHy at amyE & blank P_xyl at levB, strains GLB446–449, n = 252 comparisons. Both control strain series show minimal effect of ectopic insertions on cell fitness. s values displayed correspond to median with 25^th and 75^th percentile of values across isogenic pairs.

6. H

   Hand mixing experiment with two strains with different barcodes, showing accurate (slopes 1.00 and 0.98 for observed vs. expected ratio of barcodes from two technical replicates) readout of cell frequencies in pool over nearly four orders of magnitude. Median difference between readout from two technical replicates is 20%. Error bars are as in (C and D).

7. I, J

   Representative examples of mRNA level (RNA‐seq) comparison between wild‐type and control strains with blank expression cassette insertions: (I) GLB434 vs. wild‐type (experiment E1‐C5), and (J) GLB446 vs. wild‐type (experiment E2‐C1). Cumulative distributions of fold‐changes are shown as insets as in Fig 3A and B.

Figure 2. Diverse fitness landscapes and physiological trajectories upon RF expression perturbation

1. A–D

   Profiled orthogonal directions of the RF expression subspace, respectively, scanning along the dimension of (A) RF1, (B) RF2, (C) PrmC, and (D) PrmC with RF2 overexpression. Axes correspond to expression levels of RF1, RF2, and PrmC.

2. E–H

   Cell exponential growth rate s measured by competition (relative to wild‐type) at corresponding RF levels (reported in units of proteome fraction, derived from a ribosome profiling calibration, Materials and Methods) shown in (A–D). Endogenous levels of RFs are indicated with dashed vertical lines. Gray shadings mark the precision of our fitness measurement, defined as ± 2σ _s = ± 1.2%, where σ _s is the standard deviation in the measured relative growth rate among isogenic redundantly barcoded strain (see Fig EV2E), with the distribution shown as inset in panel (H). Relative fitness value reported corresponds to the median across isogenic barcoded pairs, and vertical black bars delineate 25^th to 75^th percentile among such pairs (between 21 and 28 isogenic pairs, see Dataset EV6) for a single experiment and is typically smaller than the plot symbol.

3. I–L

   Trajectories following RF perturbation in the space of relative growth rate vs. estimated proteome fraction to translation proteins (translation sector).

Data information: Matched arrows across panels (E–H) and (I–L) show direction of increasing perturbations in the fitness landscape. Lines in (J–L) correspond to least‐square fits. Lines in panels (K and L) have been reproduced in panel (J) to highlight of additivity of trajectories under PrmC perturbation with RF2 overexpression. See also Figs EV1, EV2, EV3, EV4.

Figure 3. σ ^B regulon induction upon PrmC overexpression compresses the translation sector

1. mRNA levels (reads per million mapped reads per kilobase, rpkm, genes with > 5 reads mapped shown) for maximal PrmC overexpression (green arrow) versus unperturbed cells (average across control datasets, Materials and Methods). σ ^B regulon members and translation‐related proteins are marked in red × and dark gray +, respectively. Targets for RT–qPCR measurements of σ ^B induction (ygxB, ywzA, Fig 5) are highlighted in dark red. mRNA levels for RF1 (blue), RF2 (orange), and PrmC (green) are marked by dots (corrected for translation efficiency of ectopic expression constructs, Materials and Methods). σ ^B regulon activation is marked by dashed red polygon. Inset shows cumulative distribution of fold‐changes in mRNA levels (red σ ^B regulon, dark gray translation, pale gray rest of proteins). σ ^B activation and translation compression are highlighted by arrows indicating shift in median expression. As a comparison, distribution of fold‐changes for all genes among unperturbed replicates are show in light blue.

2. Same as (A), but with a strain harboring a deletion of gene sigB, which abrogates σ ^B regulon activation, and restores genome‐wide expression levels despite PrmC overexpression (light gray line in inset).

3. Quantification of the proteome fraction to the σ ^B regulon (σ ^B sector) as a function of PrmC (see Materials and Methods for calibration from transcriptome to proteome fraction). Dashed vertical line marks endogenous PrmC level. Inset reproduces broader context of data in RF expression subspace (Fig 2C).

4. Similar to (C), but quantifying proteome fraction to the translation sector.

5. Proteome fraction of the translation sector as a function of the excess proteome fraction to the σ ^B regulon, denoted ${\varphi }_{\mathrm{U}}$. Dashed line corresponds to growth laws prediction (Appendix Supplementary Methods, equation 1, using parameters ${\kappa }_n$, ${\kappa }_t$, ${\varphi }_{\circ }$, and ${\varphi }_R^{\mathit{max}}$ obtained from fits in Fig 2J and K), full line corresponds to decrease by factor 1 − ϕ _U.

6. Schematic illustration of passive proteome fraction compression under σ ^B activation (increase in regulon expression).

7. Relative growth rate as a function of PrmC level, with and without sigB.

8. Growth difference with and without sigB (corresponding to ∆s panel G), as a function of excess proteome fraction to the σ ^B regulon. Dashed lines are growth law prediction (Appendix Supplementary Methods, equation 2), full line corresponds to −ϕ _U.

Data information: In panels (C–E and G), open light green triangles correspond to cells with sigB, and filled dark green diamonds to cells without sigB (deletion). See also Figs EV3 and EV4.

Figure EV3. Interplay between expression of RF2 and PrmC, fitness and σ ^B regulon activation

1. A–I

   Analogous to Fig 3, but for varying PrmC levels in conjunction with RF2 overexpression (conditions shown in Fig 2D). In (E–H), open light green pentagrams correspond to cells with sigB, and filled dark green hexagrams to cells without sigB (deletion). Blue shadings mark the region of the expression space for which the growth defect is not rescued by sigB deletion, indicating a different underlying cause for the decrease in translation sector.

2. J

   Comparison of expression at maximal PrmC expression for endogenous and overexpressed RF2 levels (respective comparisons to unperturbed conditions in Fig 3A and current panel C), showing highly reproducible σ ^B induction independent of RF2 levels (the two outliers marked by black circles are xylA and xylB, which are responsive to xylose).

3. K

   Expression‐fitness landscape for RF2 and PrmC.

4. L–N

   orthogonal projections from K showing the (L) RF2, and (M, N) PrmC directions. Panel (N) is the PrmC fitness landscape, with the fitness defect caused by RF2 overexpression defect subtracted out (arrows in panels L and M). Fitness defect at overexpressed PrmC is independent of RF2 (dashed black line in N). Knockdown defect is exacerbated by RF2 overexpression (black arrows in N).

5. O, P

   Transcriptome under RF2 expression perturbation. (O) RF2 knockdown shows modest σ ^B induction, whereas (P) maximal RF2 overexpression displays little expression changes.

6. Q

   Growth rate difference for strain with inducible RF2, with and without sigB. A mild but significant (P < 10⁻⁵, bootstrap subsampling, Materials and Methods) improvement in fitness upon sigB deletion at lowest RF2 levels is seen. Measured s for each of 12 strain pairs, inducible RF2 (GLB426 to GLB429) vs. inducible RF2 without sigB (GLB430 to GLB433), are shown for all profiled RF2 levels, with the median and 25^th to 75^th percentile marked by black lines (error bars). Red marks the condition with lowest RF2 level. Inset shows cumulative distribution of relative fitness difference for lowest RF2 level (red) and rest of conditions (black). The fitness rescue upon sigB deletion (median increase in fitness ∆s = 0.009) is commensurate with the estimated excess σ ^B regulon proteome fraction ($\mathrm{\Delta }{\varphi }_{\mathrm{SigB}}$ = 0.0085) in this condition.

Figure EV4. Transcriptomics changes upon RF1 knockdown

1. A

   Profiled RF1 levels in the RF expression subspace (reproduction of Fig 2A).

2. B

   Fitness defect upon modulation of RF1 level, with (open pale blue squares) and without sigB (filled dark blue circles), showing no strong influence of σ ^B.

3. C

   Quantification of various regulons' transcriptome fraction as a function of RF1 levels (shown for strain with sigB). The large fitness decrease upon RF1 knockdown coincides with decrease in motility (SigD, cyan) regulon and autolysin operon (magenta), and increase in biofilm matrix eps (light green) genes production.

4. D, E

   mRNA levels (rpkm, genes with > 5 reads mapped shown) comparison to unperturbed for maximal RF1 knockdown, with regulon members colored following (c), for strains (D) with or (E) without sigB. Median fold‐change is larger or equal to 5 for SigD, eps, and lyt genes, independently from σ ^B. Cumulative distributions of fold‐change for highlighted regulons (rest of genes in gray, all‐to‐all for across unperturbed replicates in pale blue) are shown as insets as in Fig 3A.

Figure EV5. Details on translational profiling following acute RF depletion

1. A–C

   Heatmaps of all data used to generate metagene ribosome queuing plot shown in Fig 4A for (A) wild‐type, (B) RF1/PrmC CRISPRi depletion, and (C) RF2 CRISPRi depletion. Genes are separated by stop codon. Each horizontal line represents a gene, and gene‐normalized ribosome footprint density (center‐mapped) is shown as gray scale, horizontally aligned by the position of the stop codon (5′–3′ left to right). Genes are organized in increasing order of translation efficiency moving up (TE). Queues upstream of stop codons with perturbed RFs can be seen (colored boxes in B and C) and are longer for genes with high TE. Scale bars indicate 25 nt and 100 genes. Caret ▲ marks the position of stop codons.

2. D

   Control analysis of changes in expression stoichiometry under RF depletion for co‐directional genes within 30 bp, but stratified by the stop codon of the downstream gene (RF1/PrmC: FC_UAG = 0.99, P = 0.39; RF2: FC_UGA = 0.98, P = 0.27, P‐value from stop codon reshufflings, Materials and Methods).

3. E

   Similar to (D), but for co‐directional genes separated by more than 30 bp, stratifying by the upstream gene (RF1/PrmC: FC_UAG = 1.01, P = 0.60; RF2: FC_UGA = 1.04, P = 0.96, P‐value from stop codon reshufflings, Materials and Methods).

4. F

   Analysis for expression stoichiometry of gene pairs paralleling Fig 4D, but with Rend‐seq data, showing no effect (RF1/PrmC: FC_UAG = 1.00, P = 0.59; RF2: FC_UGA = 1.02, P = 0.98, P‐value from stop codon reshufflings, Materials and Methods). This further suggests that perturbed expression stoichiometry results from changes in translation.

5. G

   Distributions of fold‐change in mRNA levels between wild‐type and RF depletion stratified by stop codon, showing small (≈ 2%) changes in median for RF‐perturbed stop (RF1/PrmC: FC_UAG = 0.98, P = 0.06; RF2: FC_UGA = 0.98, P = 0.02, P‐value from stop codon reshufflings, Materials and Methods).

6. H

   Schematic illustrating how decreasing termination rate leads to ribosome queues on mRNAs with high translation efficiency (ribosome initiation rate), see Appendix Supplementary Methods.

7. I

   Fold‐change in expression stoichiometry for gene pairs considered in Fig 4E (within 30 bp and stratified by upstream stop codon, subset with measured TE shown) as a function of TE. Significant (F‐test from MATLAB's regress function, P < 0.05) correlations (increasing translation efficiency leading to more severe effect, R ² ≈ 0.1) are seen for genes with stop cognate to the RF perturbation.

8. J

   Cumulative number of co‐directional gene pairs separated by given distance in Bacillus subtilis, stratified by stop codon identity. Overlap A UGA (arrow) is the most common configuration.

9. K

   Same as Fig 4D, but with UGA pairs split between those with A UGA overlap or not, with the overall effect distribution is similar between the two types of overlaps.

Figure 4. Translation termination stress and gene expression changes upon acute RF depletion with CRISPRi

1. A

   Metagene trace (Materials and Methods) of gene‐normalized ribosome footprint read density (center‐mapped, genes with footprint density > 0.5/nt) upstream of stop codons stratified by stop codon. Wild‐type (top), RF1/PrmC depletion (middle), and RF2 depletion (bottom) are shown. Ribosome queues upstream of stop codon cognate to RF perturbations can be seen.

2. B

   Translation readthrough score for isolated genes (Materials and Methods) under acute R1/PrmC and RF2 CRISPRi knockdown. Points in beeswarm plot correspond to individual genes, overlaid box plot highlighting the interquartile range (25^th to 75^th percentile, median red mark). Fivefold increase in translational readthrough is seen for genes terminating with the stop codon cognate to the knocked‐down RF (arrows).

3. C, D

   Comparison of expression (ribosome footprint density, rpkm) for (C) RF1/PrmC depletion, and (D) RF2 depletion, vs. wild‐type. Regulons and RF are marked as in Fig 3A. RF2 depletion leads to σ ^B activation, in contrast to RF1/PrmC depletion.

4. E

   Fold‐change in the downstream‐to‐upstream expression (ribosome profiling) stoichiometry upon RF depletion for co‐directional genes within 30 bp, and stratified by the stop codon of the upstream gene. A mild but systematic and stop codon‐specific decrease in downstream gene expression is seen (RF1/PrmC: median UAG fold‐change compared with UAA (FC_UAG) = 0.88, P < 10⁻⁶; RF2: FC_UGA = 0.82, P < 10⁻⁶, P‐values from stop codon reshufflings, Materials and Methods). We ascribe this effect to obstruction of downstream ribosome initiation by idle upstream terminating ribosomes. Some protein pairs, such as RsbW/RsbV, are affected more strongly.

5. F

   Fold‐change in expression between RF depletion and wild‐type stratified by stop codon of genes. No systematic effect for the different stop codons is seen (RF1/PrmC: FC_UAG = 1.01, P = 0.69; RF2: FC_UGA = 1.02, P = 0.86, P‐values from stop codon reshufflings, Materials and Methods), indicating lack of strong change in expression for genes with compromised translation termination.

Data information: See also Figs EV3 and EV5, Appendix Figs S3 and S4.

Figure 5. Swapping a single stop codon rewires σ ^B induction susceptibility to RF perturbation

1. Schematic of simplified σ ^B regulatory architecture and operon structure.

2. Different rsbV stop codon variants considered. Top row shows endogenous configuration rsbV _UGA, with RF2‐dependent stop codon and A UGA open reading frames overlap. Middle row shows the UAA RF‐agnostic variant rsbV _UAA, obtained by adding AAT (underlined). Bottom row shows the RF1‐dependent UAG allele rsbV _UAG, obtained by inserting AGAT (underlined).

3. Average of reporter gene log₂ fold‐change compared with wild‐type for σ ^B reporter genes (ywzA and ygxB, highlighted in Fig 3A) as quantified by RT–qPCR (2–3 independent biological replicates for each allele/condition, ± indicates s.e.m. over replicates for the two reporter genes, raw data in Dataset EV8, Materials and Methods) showing strong induction of reporter genes in the RF perturbations cognate to the stop codon of the rsbV variant. Rows correspond to rsbV stop variant in panel (B) and columns to RF expression conditions. RF‐inducible measurements were done in strains with orthogonally tunable RF1/PrmC (IPTG) and RF2 (xylose; Materials and Methods).

Data information: See also Appendix Fig S5.

Figure 6. Regulatory entrenchment in gene expression‐fitness landscapes

Perturbation of specific enzyme expression can lead to changes in the expression of a subset of genes through the sensitivity of cis‐regulatory elements. These expression imbalances can further reverberate in the network of regulatory interactions in the cell through focal points such as master regulators, which control the expression of multiple genes (purple box). In the absence of these regulators, downstream responses are muted (teal box). Pleiotropic expression changes exacerbate the fitness defect of perturbing the original enzyme. The cell's susceptibility to expression is then not simple relation of the decreased enzymatic flux, but also depends on the aggregated impacts regulatory susceptibilities of the whole cell. Cells can be liberated from regulatory entrenchment by removing nodes in the network. Data information: See also Figs EV3 and EV4.