A riboswitch gives rise to multi-generational phenotypic heterogeneity in an auxotrophic bacterium

Auxotrophy, the inability to produce an organic compound essential for growth, is widespread among bacteria. Auxotrophic bacteria rely on transporters to acquire these compounds from their environment. Here, we study the expression of both low- and high-affinity transporters of the costly amino acid methionine in an auxotrophic lactic acid bacterium, Lactococcus lactis. We show that the high-affinity transporter (Met-transporter) is heterogeneously expressed at low methionine concentrations, resulting in two isogenic subpopulations that sequester methionine in different ways: one subpopulation primarily relies on the high-affinity transporter (high expression of the Met-transporter) and the other subpopulation primarily relies on the low-affinity transporter (low expression of the Met-transporter). The phenotypic heterogeneity is remarkably stable, inherited for tens of generations, and apparent at the colony level. This heterogeneity results from a T-box riboswitch in the promoter region of the met operon encoding the high-affinity Met-transporter. We hypothesize that T-box riboswitches, which are commonly found in the Lactobacillales, may play as-yet unexplored roles in the predominantly auxotrophic lifestyle of these bacteria.

Hernandez-Valdes et al. studied the regulation of two methionine transporters in the methionine auxotrophic lactic acid bacterium Lactococcus lactis. They found that the high-affinity Met-transporter is heterogeneous expressed at low Met concentrations leading to two subpopulations thatsurprisingly -are stable over multiple generations. Several layers of regulation modulate expression of the transporter and a T-box riboswitch is accountable for heterogeneity. To the best of my knowledge, this is the first report on a riboswitch involved in heterogeneous gene expression. The authors use state-of-the-art technology and their manuscript is well written and easy to follow. A limitation of the study is the absence of any mechanistic insights into how the riboswitch gives rise to the remarkable multi-generational phenotypic heterogeneity (see title).
We thank the reviewer for this supportive feedback. We are also grateful for the constructive criticisms, which we believe has enormously helped to improve our manuscript. As suggested by the reviewer, we have now examined the heterogeneity in met expression in mutants of global regulators as well as targeted mutations in the riboswitch. Altogether, these mutants clarified how the phenotypic heterogeneity comes about. The details of which, we outline below.

Comments and questions:
1. The evidence that the riboswitch (and not the transcription factor CmhR) is responsible for heterogeneity derives from a mutant that lacks the entire riboswitch. I could not find the exact information how much sequence was deleted but it will certainly be a long stretch and its loss might influence various processes, like transcription elongation, RNA polymerase pausing or transcript stability. Ideally, one would like to see results of strains with less invasive mutations, in which the riboswitch is present but mutated in key residues responsible for its regulatory function.
We now highlight in Fig. S17a what part of the DNA sequence was deleted in our mutant that lacks the entire regulatory element. We also agree with the reviewer that it would be interesting to examine less invasive mutations in the riboswitch. Therefore, we now examined four highly-targeted mutations in conserved residues of the T-box riboswitch. The mutations specifically target (i) the stability of the stem I domain, (ii) the formation of the antitermination complex, by either affecting tRNA binding to the T-box or the stability of the antiterminator, and (iii) the formation of the terminator hairpin. Fig. 6 and Fig. S17b visualize the different mutations and show the resulting met expression using both microscopy and flow cytometry. Overall, the results strongly corroborate our previous findings that the riboswitch gives rise to the phenotypic heterogeneity. None of the mutations show phenotypic heterogeneity and two mutations show particularly interesting results. Namely, the mutant preventing tRNA binding to the riboswitch shows expression patterns that match exactly with the GFP-cells in the wild type, whereas the mutant that prevents the formation of the terminator hairpin shows expression patterns that match the GFP+ subpopulations in the wild type. These mutants thereby confirm that the GFP-subpopulation originates from the successful formation of the terminator hairpin, that lowers met expression through transcriptional attenuation, whereas the GFP+ subpopulation originates from the formation of the anti-terminator complex by binding of uncharged tRNAs to the T-box element. Further information regarding the strains is given in Supplementary Table 2. 2. What is the evidence that the authors really look at phenotypic heterogeneity rather than genetic heterogeneity (in other words, point mutations) when the phenotype is stably inherited? It would be comforting to know that the strains indeed are isogenic, as the authors claim. Given that the riboswitch seems to be responsible for heterogeneity, it might be sufficient to sequence across this region in different subpopulations.
We have now sequenced the 5' UTR end of the met operon by PCR in bacterial colonies of each phenotype: two GFP+ colonies, two GFP-colonies and the GFP+ and GFP-sectors of a switching colony. No mutations in the promoter regions of the met operon were detected. To also completely rule out mutations elsewhere in the genome, we have also performed wholegenome sequencing in one GFP+ colony, one GFP-colony and the GFP+ and GFP-sectors of a switching colony. Also in these samples, mutations were absent, which confirms our expectation that the GFP-and GFP+ subpopulations are isogenic. All results are summarized in Supplementary Fig. 5. We will upload all sequencing data upon publication.
3. It is not immediately evident why a part of the population should use the low-affinity transporter when the amino acid is scarce. Apparently, it would be beneficial to have the high-affinity transporter available for acquisition of the limiting nutrient. Here, the stable commitment to an inefficient transporter is most surprising. Instead, one would expect a rapid switch back to the other phenotype. Please comment.
Good point. We first want to emphasize that the GFP-cells do still express the met operon, although on a much lower level than the GFP+ cells (Fig. 2f, 5d). In fact, we know that the GFP-cells partly rely on the expression of the met operon, since a knockout of the met operon does not support growth at the lowest methionine concentrations, which indicates that cells cannot fully compensate for the absence of the met operon by increasing the expression of the low affinity transporter (Fig. S7, S8). We realized that the schematic summary figure was a bit confusing in this regard and have therefore considerably revised this figure to highlight that both the GFP-and GFP+ cells express the met operon, but that their expression levels differ substantially (see new Fig. 7).
Despite the different expression levels of the met operon, the fitness differences between the GFP+ and GFP-colonies are surprisingly small. We therefore hypothesized that the GFP-cells compensate for the lack of met expression by increasing the expression of the low-affinity transporter (BcaP). Indeed, when we knockout CodY (a repressor of bcaP), we lose the subpopulation of cells that highly express the met operon (i.e. GFP+ cells), which suggests that the low-affinity transporter can partly (not entirely) compensate for the high-affinity transporter. From this perspective, one does not necessarily expect a high switching rate, because two alternative routes for methionine uptake only show minimal differences.
To clarify our arguments, we have now substantially changed the text (page 8, lines 197-205), as well as adjusted Fig. 7. 4. In the discussion, the authors provide a reasonable explanation to the question why heterogeneous expression of the Met-transporter might be beneficial. Bet-hedging and division of labor are wellaccepted concepts in the field. What is lacking, however, is a reasonable answer to the question how the riboswitch determines heterogeneity. How can it lock expression in a certain state for multiple generations?
We thank the reviewer for emphasizing this shortcoming. Both in reply to this comment and comment 6, we have evaluated met expression in different knockout mutants of the global regulators. As it has been shown before that heterogeneity in L. lactis can result from cells that are locked in distinct physiological states 3 , we hypothesized that the differential met expression in the GFP-and GFP+ populations could reflect such physiological states as well. Strikingly, the knockouts of all global regulators (codY, rel, ccpA) abolishes the bimodal gene expression of the met operon as seen in the wild type (for details see comment 6 below; Fig.  S11). The rel knockout is particularly intriguing, because it shows expression distributions of the met operon across the different methionine concentrations that correspond exactly to those of the GFP-subpopulation in the wild type. The rel knockout furthermore gave the same expression patterns as the riboswitch mutant that prevents binding of uncharged tRNAs. Given that Rel is activated by uncharged tRNAs 4 , these results suggest that the stringent response is activated in cells with high met expression, which could lock them in this state. A double knockout mutant that we made previously, indeed shows that the rel knockout exerts its effect on met expression via the riboswitch (Fig. S12), as opposed to, for example, the expression or activity of the transcription factors (CmhR).
Altogether, we therefore envision the following scenario. Upon methionine depletion cells need to increase their methionine uptake rates. In some cells, CodY repression of the bcaP operon is released timely, due to which the low-affinity transporter can sequester enough methionine from the environment to support growth (these cells will also increase the expression of the met operon, due to activity of the transcription factor). Consequently, these cells will not experience methionine starvation that leads to uncharged tRNAs (Note that in the absence of the low-affinity transporter all cells will highly express the met operon; see Fig. 4, S18). In contrast, other cells might not be able to increase uptake rates timely and will experience methionine starvation leading to uncharged tRNAs. These tRNAs trigger both high expression of the met operon (via the riboswitch) and the stringent response. We think that the stringent response could be responsible for keeping cells locked in a distinct physiological state (note that in L. lactis the stringent response does not repress CodY activity by changing the GTP levels, in contrast to Bacilli species 5,6 ). This is supported by the fact that GFP+ cells readily lower met expression when exposed to nutrient rich conditions (which suppress the stringent response).
We now extensively adjusted both the results and discussion sections to incorporate the new results.