Julie Fournier

I really enjoyed this new paper on ChemRxiv, a Munich/Michigan/Berkeley  collaboration on reactive covalent groups and their profile across different proteins. There have been a number of papers addressing this subject before, but this one is the most comprehensive one I’ve ever seen, and it’s a valuable resource.

Most of the covalent probes (and nearly all of the covalent drugs) that you see are targeting cysteine residues on proteins, and that’s no accident. Cys is definitely the standout nucleophile of all of them, and (as the Cravatt group and others have shown) there’s a population of them that are hypernucleophilic and are ready to party with electrophiles. These tend to be in active sites of enzymes and other specialized spots, and are surrounded by residues that make them more like a full thiolate anion (which as everyone who’s been through sophomore organic chemistry should recall, is one of God’s Own Nucleophiles when it comes to reactivity).

Over the years, though, there have been many searches for reactive groups that will pick up other amino acid residues, and this paper features a pretty comprehensive evaluation of those (54 of them in total!) The technique used is “isoDTB-ABPP”, which stands for “isotopically labelled desthiobiotin azide activity-based protein profiling”, and here’s how it works: you take a broad proteomic sample and treat it with a reactive probe compound, and at the same time set aside a control proteome sample that just gets solvent treatment, no ligand. You then treat both of those with some sort of broadly active alkyne-containing reagent, which means that in the control sample all the residues that can be tagged with it will get labeled, but in the one that’s been treated beforehand with a reactive compound, the residues it labeled will be blocked from reacting with the alkyne. You then come in with your isotopically labeled desthiobiotin azide reagents, one for the experimental sample and a different isotopically-patterned one for the control sample (light and heavy, basically), and you do a copper-catalyzed click reaction on each. Now you mix the two together (since you’ve differentiated them with those light and heavy additions), do a proteolytic digestion to break all the proteins up into small chunks, and do an LC/MS analysis on the whole mixture.

You’ll get peaks on the LC – lots and lots of peaks – and each of them will have mass spec profile. There will be a lot of those protein fragments that had nothing to do with the whole process – they didn’t get labeled with the covalent probe at first, and they didn’t get labeled with the reactive alkyne reagent, either. Those will show up with no isotopic differences in their mass spectra at all, and can be ignored. There will be others that got labeled to the same extent with the isotopically enriched probe, and you’ll see that, but there won’t be any ratio showing up between the pairs and you ignore them as well. But there will also be protein fragments whose mass spec ratios will be way off the baseline, because they had side chains that were blocked by the original reactive probe in the experimental sample, but were wide open for the reactive alkyne in the control sample. Those will be heavily skewed towards the isotopic pattern that you used in the control, and those are then the parts of the proteins that reacted with your original probe.

Now, you can see from this that you need a good selection of those reactive alkynes that will pick up various amino acid residues, and there have been many of these reported (for Cys, Arg, His, Glu, Lys, Met, Asp and others). But these have shown up from different groups, using different protocols and different LC/MS conditions and data analysis, so it was this paper’s intent to get everything under one roof: same experimental conditions, same mass spec analysis, same software workup. That last one is key, as you would imagine. Sorting seventy-eight Godzillion protein fragments (a rough estimate on my part) looking for isotopic mass ratio differences is definitely a job for automated analysis, and the paper presents an optimized version of the FragPipe computational suite for the chemical biology community. The technique is sensitive enough to pick up events like formylation of side chains from the formic-acid containing elution solvents, S-oxidation of the thioether covalent products, and so on, which is a good sign.

Ripping through the proteome of Staphylococcus aureus as an example (plenty of that available!), the team was able to sort out the various probes under controlled conditions. For example, STP-alkyne is a widely used reagent to label lysine residues, and this work confirmed that it’s selective. But it’s not perfect, because nothing is. 9% of the residues it labels are serines, 2% of them are threonines, and 5% of them are the N-terminal amines of the proteins. Looking closer, it turns out that the threonines and serines that were labeled were strongly biased towards having a histidine two residues down, and the serines also showed a preference for a cysteine two residues upstream or an arginine one residue down. So there are local effects on the reactivity of those serine and threonine OH groups that will cause them to poke their heads up for this reagent (and in fact, for all the lysine-directed probes). As it turns out, there aren’t any probes (yet) that are directed towards Ser and The residues per se, which means that you could make a start with these reagents if you like.

This sort of thing is seen for all of the alkyne probes to one degree or another – they are indeed selective for their advertised amino acids, but with some different stuff around the edges. Some of them pick up greater or fewer numbers of their target residues compared to the others, and they all have off-target reactivity to some degree as with the lysine probes above. These are valuable things to know, to calibrate analyses going forwards and to allow everyone to work from a common baseline. It’s important to keep in mind, though, that the residues that you will pick up (even using the whole suite of labeling reagents) are still a select bunch. S. aureus has over 62,000 lysines in its proteome, and all the lysine probes together will only label about 15% of those (the most accessible and the most reactive). Similarly, all the carboxylate-directed probes, put together, label about 7.8% of the Asp and Glu residues, the Tyr probes cover about 12% of the total tyrosines, and the Trp probes about 12% of the available tryptophans.

Those of you who are into this sort of thing might be saying “Hold on, tryptophans?” The paper also contributes some new probes and validates other recently described ones, especially for Trp, His, and Arg residues as well as protein N-terminals, and shows that a new photochemical lysine probe is very selective indeed. All in all, the paper identifies a set of 17 probes (out of the 54 studied) that the authors can recommend for proteomic residue coverage. All of these together label about 54% of the S. aureus proteome (and a much higher fraction of the annotated or essential proteins). A look at a human cell line showed similar results, fortunately, so this seems like it could be a useful standard set going forward. The biggest gaps are still probes for Ser, Thr, and the carboxylates at protein C-terminals. But as reactive groups are developed for these (and new ones for the other potentially reactive side chains), we now have a common platform to evaluate them.

The hope is, of course, that we can use such information about reactivity and selectivity to come up with chemical probes for specific proteins (and protein classes), and with selective drugs towards the ones that are targets in disease. The latter may well need some new chemistries, to dial down the reactivity of the “warhead” groups from what you’d use from these sorts of broad protein-labeling experiments, but there are a lot of ways that you can think of to do that. This could be applied not only to active sites in enzymes, but to allosteric sites, protein-protein interaction surfaces, and more. These techniques can also be used to selectively label proteins for imaging studies in live cells, to conjugate other small molecules to specific proteins for therapeutic use, to covalently link entirely different proteins together for new purposes, and whatever else we might be able to dream up. And that’s a lot.

The post Comprehensive Covalent Probe Time first appeared on In the Pipeline.

Publication date: 19 March 2020

Source: Cell Chemical Biology, Volume 27, Issue 3

Author(s): Hiroyuki Kojima, Yuki Fujita, Ryosuke Takeuchi, Yuka Ikebe, Nami Ohashi, Keiko Yamamoto, Toshimasa Itoh

Nature Communications, Published online: 31 January 2020; doi:10.1038/s41467-020-14494-8

Enzyme self‐immolation : Catalyzing the generation of an epoxide from a 2‐hydroxyethylammonium proform triggers the self‐destruction of a dehydratase enzyme involved in bacterial virulence. Evidences of this lethal enzyme catalyzed process from incubations and MALDI studies with designed ligands, protein X‐ray crystallography and computational studies on the reaction path are provided.

Abstract

Disabling the bacterial capacity to cause infection is an innovative approach that has attracted significant attention to fight against superbugs. A relevant target for anti‐virulence drug discovery is the type I dehydroquinase (DHQ1) enzyme. It was shown that the 2‐hydroxyethylammonium derivative 3 has in vitro activity since it causes the covalent modification of the catalytic lysine residue of DHQ1. As this compound does not bear reactive electrophilic centers, how the chemical modification occurs is intriguing. We report here an integrated approach, which involves biochemical studies, X‐ray crystallography and computational studies on the reaction path using combined quantum mechanics/molecular mechanics Umbrella Sampling Molecular Dynamics, that evidences that DHQ1 catalyzes its self‐immolation by transforming the unreactive 2‐hydroxyethylammonium group in 3 into an epoxide that triggers the lysine covalent modification. This finding might open opportunities for the design of lysine‐targeted irreversible inhibitors bearing a 2‐hydroxyethylammonium moiety as an epoxide proform, which to our knowledge has not been reported previously.

The Schmidt reaction has been an efficient and widely used synthetic approach to amides and nitriles since its discovery in 1923. However, its application often entails the use of volatile, potentially explosive, and highly toxic azide reagents. Here, we report a sequence whereby triflic anhydride and formic and acetic acids activate the bulk chemical nitromethane to serve as a nitrogen donor in place of azides in Schmidt-like reactions. This protocol further expands the substrate scope to alkynes and simple alkyl benzenes for the preparation of amides and nitriles.

Gaining an understanding of the conformational behavior of fluorinated compounds would allow expansion of the current molecular design toolbox. In order to facilitate drug discovery efforts, a systematic survey of a series of diversely substituted and protected fluorinated piperidine derivatives has been carried out using NMR spectroscopy. Computational investigations reveal that, in addition to established delocalization forces such as charge‐dipole interactions and hyperconjugation, solvation and solvent polarity play a major role. This work codifies a new design principle for conformationally rigid molecular scaffolds.

Nature, Published online: 26 March 2020; doi:10.1038/d41586-020-00933-5

Nature, Published online: 24 February 2020; doi:10.1038/d41586-020-00509-3

Nature, Published online: 20 February 2020; doi:10.1038/d41586-020-00018-3

I’ve mentioned it in passing before, but it bears repeating: this is a really unusual moment in drug discovery. We have simultaneously more new modes of action for therapy coming on in the clinic than I can ever recall, and some older ones are getting reworked to join the action. This short overview is a good look at the topic. Things that many of these have in common are new interfaces between small molecule organic chemistry and biomolecules. We still have a lot of enzyme inhibitors and receptor antagonist drugs out there, but these classic mechanisms (which, you will note, depend mainly on our ability to gum up the works in the right spots) are being joined by protein degradation, gene therapy (CRISPR and others), exon skipping, attempts to go after various RNA species, more non-active-site targeting of proteins, a new wave of antibody-drug conjugates, and more.

One way to look at these things is as a triumph of what we now call chemical biology. I have a slide deck that makes the assertion that chem-bio is the way forward in general, so I’m already biased toward that idea (and the fact that I work in a department focused on that stuff has something to do with it, too). But I think there’s a good case that the tools of that field, and the mindset behind it, are important factors. There are a lot more cellular processes that we can imagine targeting than we ever have actually been willing or able to.

To be sure, some of these represent yet more ways to gum up the works. But I think that drug discovery has always been biased in that direction, because we’re inside cells that have had a billion or two years to work out some very slick processes and tune them up to concert pitch. Stepping in and making these perform even better is a real challenge, whereas throwing a spanner wrench into the gears is much more feasible. That has tended to make the central problem of drug discovery “What process should we beneficially shut down?” This accounts for the number of bounce-shot double-negative mechanisms that you see in this business: what we want to do is activate Pathway X, but the way that we might manage to do it is to inactivate some inactivating mechanism (kinase Y or protease Z) and thus set it free. Drug development, often enough, consists of fashioning just the right size and shape of spanner wrench and flinging it just so into the bewildering mass of cellular machinery in order to bring a particular set of gears to a grinding, screeching halt.

Gene therapy is the outlier of that bunch – it doesn’t work through a small molecule agent, and it is generally aimed at directly restoring function. What it shares with the others, though, is a great deal of what can only be called novelty. We really don’t know what’s going to happen when we step in and try to rewrite some little strech of a patient’s genome, in the same way that we don’t know what’s going to happen when we give them a bifunctional molecule that targets some particular protein for degradation. People haven’t had these things done to them before. And if there’s one thing for sure in this work, it’s that we don’t know a lot of the pathways and connections out there in the cells yet. Doing things to human cells (and to human patients) that you’ve never tried before is a way to uncover some of those, and experience has shown that not all the things you uncover are good.

That’s a roundabout way of saying that I hope that these new therapeutic modes actually make it through. Having so many new mechanisms under development at the same time increases the chances that something unexpected and unwelcome will happen. That post the other day on idiosyncratic toxicity is an example, and that’s just one of many possibilities. Some of those we have seen before (but can’t anticipate accurately) and some of them are things that we will be learning about for the first time.

That’s most certainly not an argument to slow down or steer clear; it’s the cost of doing business in early-stage drug research. And there are worse problems than having a lot of new, interesting, and exciting stuff hitting more or less all at once. But I will be very glad to see more degraders, CRISPR attempts, RNA mechanisms and so on make it further into human trials without anything weird happening. We need these new shots on goal, and I hope that we can take then as cleanly as possible. This is a big moment – let’s hope it continues.

It’s our high failure rate in clinical trials that makes the drug industry what it is. And two of the biggest factors in that failure rate are picking the wrong targets/mechanisms, and unexpected toxicity. The first is clearly a failure of our understanding of human biology, and the only remedy I can see for that is for us to understand more about it. A slow process, that. The second would certainly benefit from more understanding as well, and a key question is whether “idiosyncratic tox” really is completely idiosyncratic. That is, are we bumping into a whole collection of unrelated things that are just waiting out there for us to trip over them, or are there some common mechanisms that we could prepare against?

There’s already evidence for the latter. Look at cardiac arrhythmia and its showstopping manifestation as torsade de pointes. This used to be Just One of Those Things That Happens, until we realized the connection to the hERG ion channel. Now hERG testing is a standard part of preclinical drug development. It’s not perfect, but the fit is good enough to be useful and has surely allowed us to not take compounds into the clinic that would have caused trouble later. What we need are more insights that are at least that useful.

This paper has some good background on the subject. It’s looking at idiosyncratic liver injury, the sort of thing that happens at the lower-than-1-per-ten-thousand-patients level, can kick in well after the exposure to the drug, and can also lead to serious damage. In short, a nightmare for drug development and the sort of thing that you might not even be able to notice until late in Phase III or even after the drug hits the market. What’s more, the data in this area can be pretty messy, because that time delay means that such idiosyncratic adverse drug reactions (IADRs) sometimes aren’t even correlated with a particular drug exposure.

There have been many efforts to find markers of this sort of thing, of course, but it’s tricky. Blood-test signs of liver injury (such as  changes in ALT, AST, and bilirubin levels) are only vaguely correlated with these sorts of adverse events, and there are way too many false positives. There are some situations where blood samples from patients who’ve had an IADR will show effects (such as lymphocyte proliferation) on ex vivo exposure to the suspected drug, but that doesn’t always work, either. Such test can take weeks to perform and are pretty uncommon, so they’re not a great source of raw data.

A big reason for all this vagueness is that immunology is involved. It would be, wouldn’t it? Long and variable incubation time, extremely high patient-to-patient variability, sudden severe tissue damage, effects in high-exposure organs like the liver and immunologically-active ones like the skin (all those sudden-rash side effects): of course it’s the immune system. Indeed, there are specific human leukocyte antigen (HLA) alleles that have been associated with reactions to specific drugs, and you can bet that there are a lot more that we haven’t tracked down yet.

One way you can produce a new antigen is by reaction of a reactive covalent compound with some protein – that’s what’s going on with poison ivy, to pick an all-natural example. The urushiols in that plant (and in poison oak, etc.) get oxidized to reactive quinones in vivo, and those react with skin proteins to generate a neoantigen (usually after degradation to shorter peptides). You always want to be on the lookout for reactive metabolites, for just this reason. This sort of thing is of course one of the reasons that deliberately covalent compounds were avoided for so long in drug discovery, and it’s still something to you have to keep a careful eye out for. The less reactive and thus more selective covalent agents have less of a chance for this as opposed to red-hot stuff like quinones, but there are an awful lot of potential reactive sites out there. The fact that IADRs have also been associated in some cases with particular polymorphisms in metabolizing enzymes supports this mechanism.

The graphic at right is the current thinking about what’s going on, and while it makes sense, you’ll also note some rather fuzzy-sounding concepts. What exactly is that “underlying susceptibility to cell stress”, for example? The stress in this case is often oxidative. If the reactive-oxygen-species (ROS) levels in a cell exceed its capacity to deal with them through the normal routes, such compounds can start modifying proteins and generating neoantigens by that route. So anything that decreases the effectiveness of the heat shock proteins, the Nrf2 system, superoxide dismutase levels, and other such responses could be a problem. Nutritional state, co-morbidities, other drugs being taken simultaneously – there are a lot of possibilities.

As shown, it looks like the first step is an innate immune response, which gradually sets off that adaptive immune system (and this helps to account for the delay in IADRs showing up, since all this takes time and perhaps multiple cycles of injury to build up). That adaptive response will naturally get off the ground faster if it’s been primed by previous exposure, and to make things more complicated, there is always the possibility of immune crosstalk, where exposure to one agent also sensitizes things to a different species. Finally, there’s always that arrow from the “cell stress” box right to an IADR.

That’s similar to what you get with an overdose of acetaminophen, for example: direct damage and severe toxicity via a reactive intermediate. Such damage is primed by conditions (alcohol, e.g.) that deplete the glutathione that would normally soak up the reactive metabolite. The reason that acetaminophen isn’t really an IADR, though, is that it is the opposite of idiosyncratic: everyone who takes too much acetaminophen will destroy their liver, and everyone who washed it down with vodka will have accelerated the process. IADRs can be just as bad, but they’re just a lot harder to predict. I think it’s safe to say that that’s because most of them do involve the immune system, but there is always a possibility for a direct-damage route that takes place because of idiosyncratic factors, too. The problems with troglitazone, for example, seem to have been mediated by disruption of bile-acid homeostasis, a complex system involving several steps with opportunities for inter-patient variability.

The paper goes into detail on efforts to come up with predictive assays for this sort of thing. The best guess is that some combination of advanced sequencing (to look for known HLAs and metabolic variants and to expand both of those lists) and ex-vivo immunological assays will work out, but we have quite a ways to go. And by that I mean both to validate such assays (or assay systems) and to get them into a form where they can be much more widely used. Telling someone after they’ve had a bad IADR that we will only need a couple of weeks to figure out why it happened is not so useful, and neither is being able to assign a cause to a failed trial only after it’s failed. The sorts of assays it looks like we’ll need are not ones that have always been easy to reduce and speed up. For drug development purposes, we’ll need to come up with some sort of standard panel that can at least alert us to the more common IADR routes, and there will be quite a few of those to cover. We’re never going to be able to wring all the risk out of taken an investigational drug into humans – but we should be able to do a lot better than we can now.

Figuring out an unusual natural product’s activity can be a difficult but rewarding exercise. Deep evolutionary time has provided us with a bizarre range of chemical structures that are presumably not being synthesized by organisms for the sheer fun of it – these things are acting as signaling molecules, antifeedants, poisons for the competition, pheromones and who knows what else. Many biochemical pathways have been discovered and defined by the actions of some natural product (the opiate receptors and the cardiac glycosides are two of the classic examples) although sometimes this has taken a very long time indeed (as in the cases of salicylic acid and quinine).

Here’s a new paper on a macrolide from a sponge in the Gulf of Mexico, lasolonide A. It’s long been known to be active against several tumor cell lines and with a unique profile (and has been synthesized more than once, although it’s interesting that some of those papers have made one enantiomer and some the other). Its exact mode of action has been obscure, although it has been shown to promote hyperphosphorylation in cells (so it might be taking the brakes off several kinases somehow). In general, sensitive cells show loss of adhesion, sudden chromosome condensation, and “blebbing“. These effects are actually reversible (cells can take a lot of messing with), but on longer-term exposure the compound is cytotoxic.

It’s pretty potent – Hap1 cells, for example, show these effects at 20 nM concentration. As the paper notes, several other related compounds have been isolated, and they show a definite SAR – moving double bonds around or messing with the size of the macrocycle make it less potent or completely inactive. So there’s likely a specific target, but what? The team here (Stanford, Albany, and Genentech) did the sort of experiment that one does in these days of modern times: they generated a huge library of random mutations and treated those with a severe dose of the compound, reasoning that any cells that survived were likely to have inactivating mutations in particularly important proteins. This worked just as planned, and pointed to mutations in the gene for a protein called lipid droplet associated hydrolase (LDAH). Deliberately inactivating that gene made several different cell lines far less sensitive to the compound, and introducing active LDAH into that line brought the toxicity right back, so the connection looks solid.

That’s an interesting result. There are an awful lot of such enzymes in the cell, generally working through a serine hydrolase mechanism and often involved in lipid processing. But how does that relate to the toxic mechanism, and why is it so specific to the one hydrolase? It’s not because lasolonide A is an inhibitor for LDAH; the knockout experiments showed that cells can get along just fine without that enzyme activity. Mass spec experiments showed that a side-chain ester in the compound was apparently being cleaved to give another natural product that had already been isolated in the wild, lasolonide F. A serine hydrolase could certainly do something like that, and sure enough, in the LDAH-disrupted cells the amount of lasF decreased sharply, suggesting that it is the active toxic species. It’s a lot more polar than the greasy lasA, of course, and actually isnt very active against cells directly, which tells you that lasA is acting as a prodrug to release the active compound once it gets into the cytoplasm.

LDAH has been studied in the past, and it’s known (thus the name) to associate with the surface of lipid droplets as they emerge from the endoplasmic reticulum. In fact, a particular protein hairpin motif has been identified that anchors it into that interface, and as it happens, mutating that hairpin abolishes lasA toxicity. So the protein apparently has to be sitting at the lipid droplet in order to hydrolyze the natural product: where is the natural product, then? Accumulating in the lipid droplet! Various techniques that increase the lipid droplet content of cells also increased the toxicity of lasA. Subcellular fractionation and mass spec confirmed this – the compound is also in the cytoplasm (although that might be an artifact, to be honest), but definitely partitions into the lipids, whereas lasF (as you’d expect from its polarity) doesn’t show up in there at all.

It would be reasonable to assume that LDAH is not necessarily some wonderfully selective enzyme for lasA, but rather is hydrolyzing it because there’s a ton of it accumulating with it in its lipid droplets. This local concentration effect takes us right back into the world of intracellular condensate droplets. A separate lipid droplet phase is a lot more familiar to organic chemists and cell biologists, as is the idea of compound partitioning into such droplets. But the idea is the same for the protein/RNA droplets that we now know are all over the cell: they’re promoting particular reactions and pathways by increased local concentration.

On another level, this is a big reason for why enzyme active sites work, and the same principle applies to proteins interacting with each other. A lot of cell biology is arranging things so that they’re colocated (or most definitely not colocated!), thus all the compartments, membranes, and organelles. Concentration gradients are what keep us all alive. The currently fashionable targeted protein degradation work is all about hijacking such propinquity for fun and profit – there are clearly a lot of ubiquitinating enzymes that will go to work on whatever’s brought next to them, so now we’re focusing on getting them together. “Control of local concentration” may be an overarching theme of 21st-century drug discovery.

To zoom back in to the topic at hand, this paper raises some interesting questions Can this lipid-droplet mechanism be exploited for prodrug delivery? Is it happening already, and lasA is just allowing us to notice it? Are there other lipid-droplet-associated enzymes that can be taken advantage of? We certainly make enough hydrophobic molecules in this business!

One of the key advantages bacteria have (versus our strategies to outwit them) is their fast turnover. Bacterial generations come along so quickly that advantageous mutations can spread through a population much faster than we can deal with the changes. And it gets worse: there are many bacterial species that actually increase their mutation rates under stress. This “adaptive mutability” mechanism (increasing the error rates when copying genetic material) is under selection pressure, too, of course – too much easy mutation and you’re bound to run into trouble – but having it in reserve has apparently been a winning strategy for bacterial species in competitive environments.

Now comes evidence that cancer cells can take advantage of the same trick. This new paper shows that colorectal tumor cells treated with the EGFR antibody cetuximab (Erbitux, which long-timers will remember as an Imclone product originally) downregulate expression of several genes in their DNA mismatch-repair system and at the same time up regulate genes that express error-prone DNA polymerases.The signs of DNA damage and mutation increased in a dose-responsive fashion. That combination is just what you see in bacteria – crank up the errors in one area, dial down a system that would fix the errors that accumulate in another. Now, most of the cells die under these conditions, but the persister cells show these adaptations, which again is just like the situation seen when bacteria are exposed to antibiotics. Meanwhile, cell lines that were already known to be resistant to EGFR antibody treatment did not change their mutation rates, presumably because they didn’t have to (they weren’t under stress on exposure to Erbitux).

Even more, tissue samples from patients who had shown partial response to chemotherapy showed these same patterns of expression. The high-fidelity DNA polymerases went down while the error-prone ones went up, and mismatch-repair genes went down as well. Now it’s true that some chemotherapy agents directly damage DNA or interfere with its repair, but EGFR inhibition is not one of those – all of that is downstream. The same sorts of effects were seen on siRNA knockdown of EGFR (and some other well-known targets such as BRAF), which ties things pretty firmly to the mechanism of action and not some general effect of the drug itself.

What kinds of mutations occur under these conditions? Whole-exome sequencing of the cells (both the starting population and the persisters after treatment) didn’t show much difference. But looking at microsatellite regions, which have a lot of replication errors even under normal conditions (and get a lot of DNA repair attention), showed significant changes in their length after EGFR inhibition (there’s a lot of nucleotide-repeat stuff in these areas). It would be interesting to know more about just what the survival benefits are to this.

So the parallels between tumor cells and bacteria have become even more clear. They have very similar strategies to deal with external attempts to shut down their growth, and just as with bacteria, their change in mutation rate is temporary. Once a population of cells establishes that has found a way to evade the chemotherapy agent, things settle back down. These results also highlight some thoughts that people have already had about strategies that directly target DNA repair and the like. If you’re going to try to kill tumor cells via such mechanisms, you need to to a thorough job of it, because if you just mildly interfere with these pathways you could be helping the tumor cell population to mutate its way into a resistant state. Just like trying to kill bacteria with antibiotics!