Testing bacterial vulnerabilities

As regular readers of this blog know, I am not looking forward to living in a world without effective antibiotics at all.  (Well, I’m not insane.)  I was therefore interested in a recent paper (Wei et al. 2011.  Depletion of antibiotic targets has widely varying effects on growth.  PNAS doi:10.1073/pnas.1018301108) that takes a small step in the direction of making the discovery of new antibiotics more rational.

There’s a general feeling out there that the best antibiotic targets to go after are probably the ones that are most needed by the cell.  If a bacterium needs 99% of the activity of a protein to grow, then maybe reducing the activity of that protein by just a little bit would be enough to kill the bacterium, or at least make it grow much slower.  To make it easier to find a small molecule lead that inhibits a particular target, you can reduce the expression of the target, for example by using bacteria in which mRNA production from your gene of interest is reduced.  Wei et al. use a conceptually similar but practically different approach: they knock down the levels of the target protein using a clever strategy for inducing protein degradation.  Their strategy goes like this.

There is a protease system in bacteria that is widely conserved and goes by the name of Clp (pronounced clip). Clp will recognize any protein whose C-terminus ends in the sequence ANDENYGLAA (a slightly altered version of the original), but it will not recognize anything that ends in ANDENYGLAAXXXX.  The HIV-2 protease, conveniently [nice to see this virus being useful for once], will recognize a related sequence, GLAAPQFS, and cleave it to produce GLAA.  So you can add ANDENYGLAAPGFS to your protein of interest, and it will produce normal protein levels; then you can express HIV-2 protease under the control of an inducible promoter, and the GLAAPQFS will be cleaved to GLAA; then your protein of interest will become a target for Clp and will be degraded.  While they’re at it, they add a FLAG tag after the PQFS part of the sequence (which will be cleaved off when the HIV-2 protease is induced), so that they can track uncleaved protein, and a c-myc tag before the ANDENY part of sequence (which will stay attached to the protein), so that they can track protein that has been cleaved by HIV-2 protease but not yet degraded by Clp.  They call the whole system the “ID tag”.

Using homologous recombination, Wei et al. added ID tags to 6 genes in Mycobacterium smegmatis, a model for M. tuberculosis (a bacterium that poses major antibiotic resistance challenges).  The products of the genes they chose were all targets for known antibiotics, and included the β subunit of the RNA polymerase gene (RpoB), the target for rifampicin; gyrase A, the target of quinolones; and dihydrofolate reductase (DHFR), the target of trimethoprim. In all cases, the induction of HIV-2 protease down-regulated the levels of these targets to a significant degree; some were essentially gone completely, while others were knocked down to ~20% of normal levels (perhaps these ones are under some type of feedback control?).  And so we can now test the hypothesis that the fact that a protein is targeted by a successful antibiotic implies that it is hard for the cell to do without.

For a couple of targets, the story is just about what you might expect it to be.  Inducing the degradation of RpoB knocks it down to ~20% of normal, which causes strong growth arrest.  One other target behaves similarly, causing cell death when depleted.  But for the four other targets, including gyrase A and DHFR, very strong knock-down has little to no effect on the growth of the bacterium.  For gyrase A, this is relatively easy to explain: quinolones cause DNA damage as well as slowing down the action of the gyrase enzyme, so it’s reasonable that removing the protein doesn’t have the same effect as treatment with drug.  But the observations with DHFR were worrying to the authors: if knocking down the presumed drug target doesn’t phenocopy the effect of the drug, could it be that the target we know about is not the only relevant target?  Or is the effect of the drug more complicated than reducing target activity, for example does the drug–target interaction cause toxicity?  Protein levels of DHFR were knocked down by >97% after induction of the HIV-2 protease, and rather quickly at that (within 3 hours).  So it seems particularly puzzling that this rapid, thorough degradation of an essential gene would not have a major effect on growth.

To find out what was going on, they took a look at how trimethoprim (the drug that targets DHFR) alters the profile of metabolites in the bacterium.  You would expect to see a buildup of DHFR substrates and a depletion of DHFR products, and they do.  They see the same pattern of metabolite buildup/depletion, though less pronounced, when they knock down protein levels; and when they treated DHFR knockdown bacteria with trimethoprim, they saw extreme sensitization to low levels of the drug.  Treating knockdown bacteria with 1 µg/ml of trimethoprim gave the same profile as treating normal bacteria with 50 µg/ml trimethoprim.  So there is no evidence that trimethoprim needs to do anything but reduce the activity of DHFR to have its effects: instead, we are forced to conclude that less than 3% of the normal level of active enzyme is enough for normal bacterial growth.

The authors’ original goal was to look at how “vulnerable” candidate antibiotic targets are to inhibition: all other things being equal, it should be easier to get clinical effectiveness by targeting a protein that shows growth effects when knocked down just a little bit (like RpoB) than one that needs to be knocked down to extremely low levels (like DHFR).  That said, trimethoprim is an effective antibiotic; so maybe it’s not such a great idea to discard targets that need to be strongly inhibited.  From another point of view, this system might also be quite useful for probing questions of which proteins are expressed at levels much higher than needed, and why.  Work from Uri Alon’s group in the lac system in E. coli showed that for this system, the level of protein expression seems to be tuned to just the level required for maximum growth rate; and I must admit I had thought that this would turn out to be a general result in bacteria.  (In yeast, though, most proteins are expressed at more than double the levels needed for normal growth; and Alon’s group showed more recently that the fitness cost of “unneeded” proteins varies depending on the history of the organism).   What determines whether a protein is expressed at a “just right” level, or in excess?  And how does this affect how we think about antibiotic development?

Wei JR, Krishnamoorthy V, Murphy K, Kim JH, Schnappinger D, Alber T, Sassetti CM, Rhee KY, & Rubin EJ (2011). Depletion of antibiotic targets has widely varying effects on growth. Proceedings of the National Academy of Sciences of the United States of America PMID: 21368134