Referee 1's review:

Positive feedback and noise activate the stringent response regulator Rel in Mycobacteria

Summary: In this manuscript, the authors explore the existence of and basis for phenotypic heterogeneity in mycobacterial populations under stress. They have published evidence for a regulatory loop involving the two component regulator, mprAB, sigE, and ppk1/polyphosphate. Based on this they show that in stationary phase, rel expression is bimodal (and perhaps bistable) and that the proportion of the population exhibiting Hi and Lo levels of rel expression shows hysteresis.

This line of research is enormously interesting as these data suggest that a subpopulation of M. tuberculosis enter a "persistent" state (and are potentially clinically dormant and/or drug tolerant) through a bistable expression loop. While this manuscript is extremely provocative and potentially important, I have a number of technical and conceptual concerns.

(1) In Figure 1, the authors outline the regulatory loop that they feel underlies bistable expression of rel. In their Mol Micro 2007 paper and in this manuscript, the authors have shown that the individual components of this regulatory cascade are functional. However, they have not made the case this is a biologically relevant positive feedback loop governing the stringent response or rel expression. In particular, I am bothered by the fact that in the published studies of the mprAB regulon (Pang et. al. and He et. al.), loss of mprAB does not affect the expression of rel or ppk1---even under conditions of cell wall stress in which the feedback loop should be active.

(2) The authors show bistability by using FACs to assess the expression of gfp transcribed from the rel promoter on an episomal vector. The authors need to control for the following technical problems.

a. First, in our experience, it is extremely difficult to analyze single mycobacteria by FACS because of their tendancy to clump. This is even more of a problem during stationary phase, which is when the authors see the emergence of a Hi-expressing population. It can also be a problem when the bacteria are resuspended in PBS without detergent as the authors did for their FACS analysis. Reliably gating on single cells is difficult and small clumps can appear as a discrete second population of Hi-expressors. If the FSC and SSC of the GFP-bright population is reanalyzed, the brights will be significantly larger than the dims. At a minimum, the authors should use gfp expressed from another, constitutive promoter as a control. One would expect that under similar growth and analysis conditions, a single gfp-bright population would be detected with this construct. Microscopy would also provide good supporting data for the authors' conclusions.

b. Second, it is problematic to use an expression reporter carried on an episomal vector in these studies. The number of plasmids in any given cell will determine the magnitude of gfp expression seen. However, the copy number of the mycobacterial expression vectors will vary significantly in individual cells. Moreover, bacteria carrying more plasmids are likely to give rise to bacteria carrying more plasmids and vice versa. It is plausible that one could have bimodality in plasmid number not bistability in rel expression per se. Ideally, one would use an integrated expression reporter but the control plasmid proposed above might be a sufficient control

c. The most compelling experiments in this paper are those in which the authors show that rel expression exhibits hysteresis depending on prior ppk levels. However, I am concerned about the use of a tetracycline-inducible promoter to control ppk levels. The tet-inducible system clearly might be bistable with hysteresis. If the tet-inducible promoter functions as a switch instead of a rheostat, couldn't this produce similar findings?

Also, modulating the level of ppk expression alters the amount of polyphosphate (which is the inducer in the positive feedback loop the authors propose). Thus, the bacteria go into stationary phase (and the new, intermediate ppk level) with a larger or smaller store of polyphosphate. This clearly might provide cellular memory of the pre-existing ppk level because the level of biologically relevant inducer (poly P) remains different well after the expression of ppk changes. However, is this is truly hysteresis (as it is used in reference to the lac operon in E. coli where bistability allows cells to retain a memory of earlier inducer concentration after the level of inducer itself is shifted )?

(3) Finally, the authors must be careful in their discussion of "persisters" and "persistence". This is clearly a minefield in the TB literature where persistence, dormancy and latency are often confounded.

Rel has been shown to be important for survival of mycobacteria in stationary phase and starvation. However, the E. coli literature would suggest that even in stationary phase, only a portion of the population is antibiotic tolerant, non-growing "persister" cells. It is not clear that rel governs the formation of this narrowly defined "persister" population. Likewise, rel is also important for the survival of M. tuberculosis during the chronic phase of infection in mice. However, the metabolic and replicative state of M. tuberculosis in chronic infection is an area of intense study. Again, it is not clear what portion of the population is in a non-growing, antibiotic tolerant "persister" state and whether rel is specifically required for the survival of this population of bacteria.

It would be especially interesting if the authors tested the antibiotic susceptibility of the bimodal population of bacteria. It would be great if the authors could show that their rel-Hi population is indeed antibiotic tolerant while the rel-Lo bacteria are drug sensitive.

**********
N.B. These are the comments made by the referee when reviewing an earlier version of this paper. Prior to publication the manuscript has been revised in light of these comments and to address other editorial requirements.