Skip to Content

Animal Models: How High to Set the Bar?

A key step in all drug discovery programs are the cellular and animal models. The cells are the first time that the compounds are exposed to a living system (with cellular membranes that keep things out). The animals, of course, are a very stringent test indeed, with the full inventory of absorption, metabolism, and excretion machinery, along with the possibility of side effects in systems that you might not have even considered.
So it’s a tricky business to make sure that these tests are being done in the most meaningful way possible. You can knock your project out of promising areas for development if your model systems are too tough – and it’s even easier to water them down in the interest of getting numbers that make everyone feel better. “As stringent as they need to be” is the rule, but it’s a hard one to handle in practice.
Take, for example, the antibacterial field. The first cell assays there are unusually meaningful, since they’re being done on the real live targets of the drugs. (That doesn’t do much to get you past the high barrier of animal testing, though, since you have to see if your compounds that kill bacteria in a dish will still do it in that much more demanding environment). But there are all sorts of strains of bacteria out there, and it’s up to you to choose the ones that will tell you the most about what your compounds can do.
One way that bacteria evade being killed off by our wonder drug candidates is by pumping the compounds right back out once they get in. There are quite a few of the efflux pumps, and wild-type bacteria (particularly the resistant strains) are well stocked with them. You can culture all sorts of mutants, though, with these various transport mechanisms ablated or wiped out completely. If your compound doesn’t work on the normal lines, but cuts a swath through some of these, you have good evidence that your problem is efflux pumping, not some intrinsic problem with your target mechanism.
The problem is, we often don’t have a very good idea of what to do about efflux pumping. These proteins recognize a huge variety of different structures, and there aren’t really many useful ways to predict what they’ll take up versus what they’ll leave alone. In many cases, you just have to throw all sorts of variations at them and hope for the best. (The same goes for the other situations where active transport can be a big factor, such as with cancer cells and the blood-brain barrier).
So, how do you set up your assays? You can run the crippled bacteria first, which will give you an idea of the intrinsic potencies of your compounds, minus the pumping difficulty. That may be the way to go but you’d better follow that up with some things closer to wild-type, or you’re going to end up kidding yourself. Having a compound that infallibly kills only those bacteria that can’t spit it out is probably not going to do you (or anyone else) much good, considering what the situation is like out in the real world.
The same principle holds for other assays, all the way up to rats. If you run a relative pushover model in oncology, you can put up a very impressive plot of how powerful your compounds are. But what does that do for you in the end? Or for cancer patients, whose malignant cells are much more wily and aggressive? The best course, I’d say, is to run the watered-down models if they can tell you something that will help you move things along. But get to the wild-types, the real thing, as soon as possible. Those latter models may tell you things that you don’t want to hear – but that doesn’t mean that you don’t need to hear them.

16 comments on “Animal Models: How High to Set the Bar?”

  1. Kay says:

    Shh … don’t let management know that we use models and Rules that have uncertainties.

  2. Anonymous says:

    From your perspective and possibly also that of the greater big pharma community, how significant/important are efflux pumps when it comes to anti-bacterial resitance? Granted bacteria have a ton of ways of becoming resistant to any one class of drugs, but are efflux pumps what most people view as being the single most important component? And if so, is anyone (in the pharmaceutical industry) really looking for pump inhibitors? Or are they just looking for compounds that are not recognized by these pumps?
    Sorry. Kinda of a long question but I’m really curious about this

  3. Seth Davidson says:

    I’ve wondered about the issue of watering down when it comes to one very important cancer drug, pemetrexed, which is used in the treatment of malignant pleural mesothelioma. The University of Chicago study by Rusthoven and Vogelzang, and sponsored by Eli Lilly, only barely showed statistical significance, and that was only by changing the protocol midway through the trial to add folic acid, which decreased the toxicity of the pemetrexed/cisplatin cocktail. On the basis of that study, the pemetrexed/cisplatin combination is now the FDA’s only standard of care, even though in my law practice we see very few mesothelioma patients getting any benefit from this particular chemo treatment.
    Your post makes me wonder how rigorous the study really was.

  4. Cellbio says:

    Cell and animal model relevancy, science or faith? This is were I spent my time. Here is how we dealt with. First, we thought dosing to effect was meaningless. By that I mean, simply driving up concentrations of drug until one sees endpoint effects (tumor volume, paw swelling) can be very misleading. One needs to develop “on mechanism” measures, usually the most proximal impact, and derive dosing schemes to achieve appropriate target coverage. Then you can test for efficacy. But what does this then tell you? More about all the things you listed (ADME) than efficacy in humans. Animal models are usually very dynamic with accelerated development of signs and symptoms. Anything that mucks up cell function can score an impact by means that will not represent a suitable method of treating human disease.
    With efficacy in rat models and PK, one can then proceed with safety margin calculations, so this is helpful. We ran programs where we selected clinical candidates based upon target impact assays, without any animal efficacy studies until we entered GLP tox.
    We also developed the same assays for use in primate tox and Ph1. These Ph1 assays helped, in some cases, reveal that we were unable to achieve target coverage at all, or with tolerated doses. The key unknown with this approach is: how much target coverage is enough? We aimed for an IC90 or better at trough. That at least allows one to explore a range of doses. Betting that an IC50 or less at trough would be sufficient is much more risky.
    So, I wouldn’t water down the assays at all, but rather make the measures more relevant for the MOA, and devalue efficacy. If the target’s role in human disease is supported by other data, such as genetics, then I would proceed even if efficacy failed, provided the MOA measure was hit. The predictive nature of efficacy in models is very poor.

  5. As far as efflux pumps go, these probably are important for the intrinsic sensitivity of bacteria to antibiotics.
    Stuart Levy’s company Paratek was trying to target tetracycline pumps with modified tetracylines, I believe.

  6. Anonymous BMS Researcher says:

    Efflux pumps are very important, not only in microbiology but also in oncology where cancer cells sometimes develop very similar pumps.

  7. drug_hunter says:

    Perhaps a contentious view, but I think that animal models as used today are mostly irrelevant as an indication of clinical efficacy. I have seen far too many programs fail in the clinic after “success” was achieved in an animal model. Note, the animal models were almost always advertised as highly relevant to the human disease condition, butressed by many good and even compelling reasons. Yes, you can measure PK/PD (say, you can watch TNF drop when you put p38 inhibitors into your CIA mouse model). But mice and rats are simply different enough that it is really dangerous to conclude that you learn much about the likely effect in the clinic. If this is correct, then either (a) we need entirely different kinds of animal models (e.g. humanized mouse cancer models); or (b) we need to do MANY more phenotypic, mechanistic, and other measurements on our animal models and only believe in them if the preponderance of the evidence is consistent; or (c) we should stop using animal models entirely — e.g. replace them with cell/tissue assays and/or carefully designed Phase 1 and Phase 2 clinical trials.
    I’d love to hear opposing views from the group!

  8. Cell-bio says:

    I agree with your view, and argued this point. Everyone was on board, except tox, which is the only reason we ran a model at all prior to Ph1. Did you experience this? Their point was: how were we going to calculate a safety margin? Thought it was a weak position, we could justify doses by other means.
    I say give up the models of disease and do as you suggest, cells and tissues.

  9. poovai says:

    One of the challenges in antibacterials might include ‘protein binding’ nature of the compounds. We have seen a program where the lead compounds were highly active in vitro against several bacterial strains including some resistant bacteria. These compounds had good PK. But they failed to show efficacy in animal model (mouse) due to high protein-binding nature. Has anyone come acros this situation?

  10. Ty says:

    Re: #8, yes, animal study is abs necessary for tox studies before clinical trial, but it’s not what Derek is referring to here, I believe. By animal models, we mean efficacy models (disease models or PK/PD models).
    The problem lies in disease models. There are relatively relevant disease models, but most are forced ones (including the notorious cancer xenograft modles). For the most part, I agree that we can (should) get by any disease models. Just look at PK/PD and go with human (Phase 0 or PoC studies). However, in practice, it takes a tremendous amount of courage from the project team all the way up to management to do that in the current drug research culture… Yeah, everybody is aware of the inadequacy or insufficiency of animal (disease) models, but very few, if at all, has the courage of going without it probably because… you know, everyone needs a fallback argument. If it fails, you at least have something to blame – “it worked great in the model…”

  11. RB Woodweird says:

    Can we not target the pumps as well, or are they too similar to the human to mess with?

  12. Industry GUy says:

    What the weather going to be like in Ridgefield today? lol

  13. Kay says:

    Anonymous BMS: So all the P-gp reversal agents flopped in the clinic because ____?

  14. G says:

    #7: Cells and tissues are great for specific questions. With many diseases though, there is not one unique pathway that can be tweaked to produce efficacy. This is especially true with CNS disorders/diseases. This point was brought up in Derek’s “where are the drugs” post and the comments. Animal disease models have limitations, and more/better models should be developed. But they are the best way to experimentally model complex diseases.

  15. drug_hunter says:

    Re #12: huh??
    Re #14: I agree they *may* be the best way to model complex disease – but they are still so poor that my argument is you may be better off doing nothing in animals. Especially for CNS diseases. I’d love to see references to animal models that anyone thinks are defensible for these diseases.

  16. Chris says:

    I do worry that many animal models simply model symptoms rather than the underlying causes of disease, or they model a particular mechanism of action rather than disease modification.
    That said for proven mechanisms they can give insights into relative improvements in efficacy and/or side-effects.
    I see a more pressing problem being the evaluation of novel mechanisms, perhaps we need to show a novel drug is safe and well tolerated and be willing to try it in man?

Comments are closed.