Playing dead: how bacteria persist in the face of antibiotics

Something odd happens when you treat a population of bacteria with antibiotics.  Even if the culture is supposed to be homogeneous, and sensitive to the antibiotics you’re using, some fraction of the cells will typically survive.  This is not because the survivors have mutated and developed antibiotic resistance; we know this because if you grow up the survivors and then hit them again with the same antibiotic, the same fraction die, and the same fraction survive.  (This is partly why you’re told to take every last pill in the bottle when your doctor prescribes antibiotics for you: you’re hitting the bacteria repeatedly in the hope of catching the ones that didn’t die in response to the first treatment).  The survivors are called “persisters”, and over the last several decades a great deal of effort has been focused on answering the clinically important question of what on earth is going on.

A key breakthrough came when Nathalie Balaban, working in Stan Leibler’s lab, characterized the phenomenon of resistance as a phenotypic switch between fast-growing and non-growing or slow-growing cells (Balaban et al. 2004, Bacterial Persistence as a Phenotypic Switch, Science 305 1622-1625).  Any individual bacterium can be in either state: if it’s growing rapidly, it’s sensitive to (most) antibiotics; if not, it’s protected.  This led to the concept that the bacterial population is essentially hedging its bets.  If there’s nothing nasty around, the best strategy for any individual bacterium is to grow as fast as possible.  But when life gets more dangerous, you’re better off playing dead: if an antibiotic is suddenly introduced, you’ll survive to reproduce, while the fast-growers die.  By having both fast-growers and possums in the population, the bacterium gets to have the advantages of both strategies.  In a separate paper, Balaban and colleagues explored how this insurance policy works: how does the increased cost of surviving a trauma compensate for the fitness cost of going through a period of slow/no growth?  It turns out that the optimal frequency of shifting into slow growth depends strongly on the expected rate of unpleasant shocks.

Now Balaban is coming back for another run at the problem, this time working from her own lab at the Hebrew University, Jerusalem (Rotem et al., 2010, Regulation of phenotypic variability by a threshold-based mechanism underlies bacterial persistence.  Proc. Natl. Acad. Sci USA 107 12541-12546).  The question Rotem et al. ask in this paper is how the switch from fast-growing to slow-growing cells is accomplished.  Short answer: pure, blind luck.  Now for the longer answer.

About 20 years ago, a search for mutants that show higher than normal levels of persistence came up with a mutation in the hipA gene, hipA7, which increased the fraction of persistent bacteria by about 1,000-fold. HipA is part of the HipA/B toxin/antitoxin module, and toxin-antitoxin modules are quite puzzling little systems.  They’re usually pairs of genes, usually controlled by the same promoter, in which one gene product has a toxic effect and the other binds to the toxin and cancels its effect.  The heterodimer of toxin and antitoxin also binds to DNA and regulates expression of both genes.  The point of all this is hard to understand; the usual explanation is that toxin-antitoxin pairs are an example of selfish genes, which may be true or may be the equivalent of an M.D. saying that your disease is idiopathic [= I have no idea what caused it, but I’m dressing that up in fancy language in the hope of concealing this from you].  If toxin-antitoxin systems are indeed originally selfish, evolution has found a way to exploit at least some of them. [Evolution is good at that.]

Rotem et al. set out to determine how toxin-antitoxin modules could generate phenotypic variability, and why the hipA7 mutation increased the number of persister bacteria.  They started by asking what happens if you decouple HipA expression from HipB expression.  HipA was expressed under the tet promoter and controlled by the addition of varying amounts of tetracycline, while HipB was either left under its own promoter or deleted.  In the absence of HipB, inducing the expression of the HipA toxin causes essentially all cell growth to stop.  In the presence of HipB, the number of colonies you get goes back to normal, even at high levels of HipA — but the more HipA you have, the more variable are the dynamics of growth.  Cells in the persister phenotype take longer to start growing, and so some colonies don’t appear until much later than others.

Switching to single-cell measurements, they were able to show that the cells that grow more slowly are ones where the level of HipA expression is above a certain threshold level; and then they found that varying the level of HipB changes the threshold level of HipA at which slow growth starts.

Putting all of this together in a stochastic simulation, Rotem et al. reach the following conclusions: the persister phenotype is triggered by free HipA.  Because both genes are controlled by the same promoter, the number of HipA molecules is usually about the same as the number of HipB molecules; and since HipA and HipB bind to each other tightly, the amount of HipA available is usually very small.  This means that small fluctuations in the amount of HipB (down) or HipA (up) can have a dramatic effect on the amount of free HipA; once the HipB sink is saturated, all the excess HipA made will be unbound, and available to do whatever it does to trigger the persister phenotype.  Because biological systems are inherently noisy, small fluctuations in production happen all the time.

Johan Paulsson has sometimes used Scrabble as an analogy to show how random variation plus tight binding can lead to wild swings in the composition of a system.  In Scrabble, you typically want to use both vowels and consonants to make a word (leaving aside crwth and cwm, or euoi — which I discover is not allowed in the US, for some reason), but what you pick up is a random selection of letters that isn’t necessarily balanced between the two letter types.  Usually, you use all or almost all of the letters you have fewer of — for example, if you have one vowel, you typically use it.  Making a word takes the letters out of your rack (this is the analogy of tight binding), and if you use two letters and leave yourself with no vowels, it’s not unlikely that you’ll fail to pick up another vowel next go.  So it’s inevitable that you will occasionally find yourself with a rack of letters that contains either only vowels, or only consonants; it’s not an accident, it’s a consequence of the way the game is set up. If you’ve played Scrabble more than a few times in your life, this has certainly happened to you.  In the same way, a cell that is expressing about the same amount of two proteins, with both proteins showing small fluctuations, is certain to go through phases where there’s more of one protein than the other.  In the case of HipA/B, the occasions when there is more A than B result in a change of phenotype: the cell goes into a slow- or no-growth state.

So what seems to be happening is that the cell is exploiting the noise in a toxin/antitoxin system to trigger a phenotypic switch.  Again, this has nothing to do with genetic diversity: even if the cells have identical genomes some cells will still grow fast while others don’t.  Evolution can tune the system to determine how often the slow-growth phenotype pops up, but what happens to individual cells is a matter of chance.  Overall, a population that uses this mechanism has an advantage in an environment where it may meet antibiotics compared to a population that doesn’t.

Now, what about the hipA7 mutant, which increases the number of persisters?  What seems to happen is that HipA7 binds less strongly to HipB than does wild-type HipA, so there is more free HipA7 even in situations where HipA and HipB expression is well balanced.  Experimental evidence for this includes the fact that FRET (fluorescence resonance energy transfer) between the two proteins (when labeled with fluorophores) is lower when the HipA7 mutation is present.  So cells with this mutation simply have a higher chance of having a certain level of free HipA around, and therefore have a larger chance of inducing the persister phenotype.

I’ll leave you with one little wrinkle that made me puzzled/curious.  HipA’s toxic effect includes inhibiting the synthesis of DNA, RNA and protein.  This seems like a good way to kick a cell into a dormant state.  But according to earlier work from other labs, HipA7 is less toxic than HipA: it’s less good at blocking translation than the wild-type version of the protein, even though it’s better at triggering persistence. Apparently the toxic effects of HipA can be separated from its ability to induce the persistence phenotype.  Rather surprisingly, it seems that the mechanism for inducing persistence is still not understood.  Which leaves me wondering — is it relevant that the trigger molecule is a toxin?   Do all toxin/antitoxin pairs do this, and are all the systems that do this toxin/antitoxin pairs?  If you know more about this than I do, please feel free to comment!

Rotem E, Loinger A, Ronin I, Levin-Reisman I, Gabay C, Shoresh N, Biham O, & Balaban NQ (2010). Regulation of phenotypic variability by a threshold-based mechanism underlies bacterial persistence. Proceedings of the National Academy of Sciences of the United States of America, 107 (28), 12541-6 PMID: 20616060