Skip to main content

Animal Testing

Worst Animal Model: Nominations Are Open

I was talking to someone the other day about animal models, and that got me to thinking: there are several therapeutic areas with reasonably good ones, but which indication has the most useless ones?
Naturally, just getting a compound into mice or what have you is going to tell you a lot that you’d never learn otherwise. (Try predicting oral absorption and let me know how well you make out, for example). That’s the rough equivalent of a Phase I for animal studies. But finding an animal model of disease (the rough equivalent of Phase II) is a lot trickier. (One of the better ones I can think of is diabetes, and even there you have to work carefully, because a mutant db/db mouse really didn’t get to its condition by the same path a human type II patient did).
By “worst animal model”, I mostly mean “least predictive”. There are some that are a major pain to set up and run, but give you some data that you can at least believe in a bit, and I wouldn’t put them in the same class. My nominee are the traditional models that have been used for Alzheimer’s. No rodent (heck, no other animal at all) develops the real AD pathology, so there’s one strike against you. Years of work on mutants of all stripes haven’t (to my knowledge) been able to get around that problem.
And the disease is affecting higher brain functions that are very poorly modeled in any of the small animals, which is strike two. When I used to work in the field, I would occasionally wonder about the relevance of watching a rat ran into one half of his cage or another to a person forgetting an important appointment. Some of the techniques also have the lotsa-work factor going for them, too, like the infamous Morris Swim Maze, which needs its own special room, full of special equipment, and a full-time person trained in its complications to generate the data that you still don’t quite trust.
So, that’s my candidate. Readers are invited to submit their own – remember, arduous but trustworthy doesn’t make the cut. The winner will be arduous and useless.

42 comments on “Worst Animal Model: Nominations Are Open”

  1. eugene says:

    Sorry, I’m not going to submit anything, even thought I’ve worked with a lot of biologists back in the day. I just wanted to point out that it’s the great philosophical insights and hidden metaphors that really make this blog stand out:
    “When I used to work in the field, I would occasionally wonder about the relevance of watching a rat ran into one half of his cage or another to a person forgetting an important appointment.”

  2. anon says:

    How about Necrotizing Enterocolitis?

  3. Biogeek says:

    oh, this is an Excellent Topic! My nomination – psychiatric disease (take your pick) – schizophrenia, psychosis – anything where you require a social/subjective evaluation, it’s gonna be tough with an animal model.
    Also, as far as I’m aware there’s not a really good model for male pattern baldness (not really a disease, but I think pharma is interested in it for $$ reasons).

  4. Rev. Howard Furst says:

    I nominate rodent carcinogenicity studies, especially for nongenotoxic agents such as PPAR agonists. Two years of dosing to obtain information basically irrelevant to human carcinogenic potential before starting phase 3 studies, at least for diabetes drugs. And, as someone at the FDA wryly said, PPARgamma agonists will kill people via heart failure long before they give them cancer…

  5. Jose says:

    How about essentially every animal model for asthma, aside from the wacky sheep system?

  6. datadriven says:

    My nominee is models for obesity. You never know why the animal is not eating … the drug itself could be causing a lot of other effects to make the animal not want to eat. If the animal feels terrible, it won’t eat; if its food tastes funny, it won’t eat, etc. Of course, water intake can be measured to help assess why the animal isn’t eating, but maybe the drug is making the animal feel sick AND very thirsty. If a drug does make it to the point of toxicology studies, the fact that weight loss can be a dose limiting toxicity makes its development more complex.

  7. I’d go with nude mice with a subcutaneous tumor on their back. I just don’t get how that’s much more relevant than killing the cells in a tissue culture dish (which is where the cells in the tumor came from in the first place).

  8. Biogeek says:

    Actually I disagree with datadriven/comment 6 – the rodent obesity models are pretty good, if you do the right controls (energy expenditure, conditioned taste aversion). It is true though that the effects can be small and difficult to measure in rodents.
    In the larger picture though, I agree that obesity drugs can be problematic – until we can sit the rodents in front of a desk 8 hrs a day, and TV another 4 hrs a day while being bombarded by ads and social cues to eat, I don’t think we have an appropriate model of 1st world overnutrition disease.

  9. Anonymous says:

    Brian has it right. A nude mouse w/ an implanted tumor in his back may be the worst model ever. The fraction of tumor weight to mouse body weight is so high it has no significance to human disease.
    Nice PowerPoint slides though, and the photographs are striking.

  10. qetzal says:

    I agree that xenografting tumors in nude mice is close to useless, but it’s relatively easy (as animal models go, anyway).
    Which is probably why it’s so popular, esp. in academic labs and small biotechs trying to discover/develop cancer treatments.

  11. mike says:

    Don’t be so glib about animals predicting oral absorption in man; there is no animal that reliably predicts this at all. At a former company, we had chimps – this was supposed to be our secret weapon (you know, all that “99% shared DNA” stuff). Guess what-The chimp is no better that the rat, mouse, cyno or dog. Sometimes theey predict and sometimes they don’t, which is the same as sayings it’s random.
    There is no animal like the human one. I believe nothing until I see it in people…

  12. Chrispy says:

    The learned helplessness model of depression gets my pick.
    It was classically done in dogs but is now a rodent model mostly. Basically, you get the animal to associate a tone with getting shocked, and then you repeatedly shock the poor bugger with associated tone until it gives up on trying to escape the shock. Put into a cage with a low wall which it could jump over to escape the shock, normal animals will jump over but learned helplessness animals don’t.
    I am not the only one who thinks this is a cruel and useless model, judging by a Google search…

  13. Eric Johnson says:

    Plenty of not-so-wonderful models out there… so how bout a lil poke at modeling and model-based thinking in general?
    Experimental Autoimmune Encephalitis faced a round of real criticism in 2005 after Prineas and Barnett concluded that some early MS demyelination lesions were devoid of leucocytes. This is significantly disanalogous to the demyelination of EAE, which depends on specific Th1 cells.
    Given that autoreactivity hasn’t be proven pathogenic in immune diseases like MS and rheumatoid arthritis (and that even any pathogenic autoreactivity that might exist would be underlain by an essentially unknown etiology), one could arguably criticize the degree of reliance on autoimmunization mouse diseases like MS and CIA. On the other hand, lord knows you’ve got to study something. But what’s perhaps a little sad is that the Prineas and Barnett study on real human MS tissue was basic histology that could have been done decades ago. While EAE has value, everyone should agree that one fact about MS is worth 40 facts about EAE.
    I am also astonished by the degree to which EAE is leaned on by some people as evidence for an autoimmune pathogenesis of MS. Especially considering that there is a mouse virus that causes demyelination.
    Worst is when people rely solely on data from models to make flat statements of fact about a human disease of unclarified etiology. I’ve seen it many a time in papers on things like MS or Crohn’s, whose etiologies (and even pathogeneses) are pretty wide open. Sometimes I’ve had to look up 3 citations to even find out that this kind of gooey rationale is what’s being proffered. I definitely want to give full consideration to maybe drinking the kool aid, but I’ll thank you to label that stuff precisely!

  14. Insider says:

    Mice lie and monkeys exaggerate!

  15. LNT says:

    Jose, what’s the problem with asthma models? We’ve had very reliable results with ovalbumin challenge models. Are you saying that the results are not relevant to asthma?

  16. Kevin P. Foley says:

    Agree totally with Eric, that 2004 Ann Neurol paper is a classic, but many are still ignoring its implications for the EAE model.
    I call this the “gold-standard model” problem. In most diseases, there are certain animal models that over time have become the accepted standard for testing virtually all new drugs, despite in many cases having poorly understood or even inappropriate disease etiologies.
    Oh, and by the way, end of contest, the hands-down winner for worst models are the still ubiquitous mouse subcutaneous xenograft tumor models. Why? I won’t even go into the technical issues associated with trying to model human cancer using tissue-culture adapted, high-passage human tumor cell lines implanted subcutaneously, often with purified basement membrane complex to make them grow more efficiently, in an immuno-deficient heterologous species. More importantly:
    1) Oncology drugs have something like the 2nd highest failure rate in the clinic. The majority of these failures are due to lack of efficacy. Obviously, this is an indictment of the predictive power of xenograft models.
    2) The NCI has been screening drugs in vitro on tumor cells and in vivo in mouse xenograft models since the 1950’s. Dozens of these drugs have reached the clinic. This is by comparison to other diseases, a massive database. And the results are crystal clear: in vitro does not predict in vivo, and in vivo does not predict clinical. If you’d like to see the numbers, check out: Johnson et al., Br J Cancer 2001.
    But mouse cancer models in general, and even subcutaneous xenograft models in particular, are not as useless as they seem. The key is using the right model for the right drug for the right disease. Besides developing new, better models, we also need to understand the models that we have better and use them more appropriately.
    Shouldn’t animals actually be very good models for human biology? Genome sequencing, quantitative trait loci and KO phenotypes all suggest they should be. In fact, one study suggested that 85% of mouse knockout models are predictive of human drug efficacy. This should also be true for PK and toxicology. Again, the problem is not that we use animals to model humans, it is which animal models we use and how we use them! For this reason, improving animal models in a key feature of the NIH Roadmap.
    Say you would like to develop a drug to treat lupus, a major unmet medical need. The first problem is to understand exactly what target/pathway you can modulate that is important in this disease. For example, recent research has illuminated the role of B cells in this disease and suggested that depletion of B cells might ameliorate disease symptoms. So if you would like to develop a drug to deplete B cells to treat lupus, you had better be sure that your animal model of lupus (or whatever autoimmune syndrome you are using as an animal surrogate, whether it is called a lupus model or not) is dependent on B cells. Forget whether it is the “gold standard for lupus” and focus on the biology.
    Of course, the recent push towards early “proof of concept” clinical trials is also an answer to this problem. However, our dependence on animal models will never go away, so new and improved models are still needed if we are to meet the challenges that face us in drug discovery.

  17. david says:

    This is an easy one–partuition (birth). There is no non-human model which mimic the human. Sure,animals do give birth. But the triggers for labor and the like are completely different.

  18. Devices R Us says:

    Isn’t Harvard’s Law “under the most rigorously controlled laboratory conditions, the animals will do what they damn well please”? I would also vote for the metabolic syndrome models of obesity and T2 diabetes. You can of course use metabolic cages but in the normal case, it’s hard to tell if the rats lose weight because they don’t eat or eat a little and vomit. I wish there were good infectious disease models but at least the lack of a model makes going to humans a bit easier.

  19. The Chemist says:

    My old roommate was a veterinarian in charge of taking care of all of the animals being used for the studies of a prominant hospital. She told me that she was taking care of some primates who were involved in an Alzheimer’s study that were subjected to a crude form of labotomy (without replacing the eyeball) and then given a bolus of some drug to see if they recovered any cognitive function. They never did, of course, and the treatment moved on to humans after checking the “primate tested” box. She moved on to private practice because she couldn’t take it anymore. Does anyone else know of this model?

  20. weirdo says:

    to LNT:
    “We’ve had very reliable results with ovalbumin challenge models. Are you saying that the results are not relevant to asthma?”
    By “reliable results”, are you implying said molecules were tested in humans in Phase 2 and actually worked? I haven’t heard of many (any) new asthma drugs in a while. Lots of failures, though.
    And since just about anything works in the “wacky sheep system”, it’s hard to argue that’s a good model, either.
    Hard to argue it’s the “worst”, though. I’ll pony up another vote for mouse xenograft tumor models.

  21. Eric Johnson says:

    I know very little about asthma, but one day I happened across the “poor epithelial barrier function” idea, and thought it was interesting. This notion de-emphasizes the etiologic role of immunologic abnormalities, emphasizing instead an abnormal excess delivery/leakage of environmental antigen across the epithelium of the lungs (due to some epithelial abnormality):
    [scroll down to the last section (S Holgate); subscription probably required]
    Apparantly there are some people who take a very much analogous view of Crohn’s, with enteric xenoantigens taking the place of inhaled ones. Ie, they propose there’s an abnormal amount of enteric xenoantigen being absorbed and presented, as opposed to an abnormal hyperresponse to a normal amount of antigen.

  22. CET says:

    Re: #12 (the learned helplessness model)
    Are we unable to measure the effects of a compound on neurotransmitter levels? Or just unable to extrapolate from NT levels to effects in humans? (Or both?)

  23. Chrispy says:

    CET —
    I read somewhere that the learned helplessness model is closer to a model of trauma than depression.
    With the redundancy of receptors, dirtiness of neurological drugs and quirkiness of human emotion I suspect that this field is just plain difficult to model or predict.
    That lobotomized primates model #19 beats learned helplessness if it’s true, though!

  24. LNT says:

    I can’t comment on it’s usefulness, but the most bizzare animal model that I’ve encountered has to be the mouse “social recognition model”, or as we affectionately refered to it, “the butt sniffing model”. It’s a mouse model to test memory. When new mice encounter one another they sniff each other’s butts. If the mice “know” each other, this happens only briefly. However, if the mice have never encountered one another before, they sniff for longer periods of time. Basically, two mice are introduced and then removed from each other for set periods of time. The amount of “butt sniffing” then do upon re-encountering each other is (in theory) inversely proportional to how well they remember each other. (After a day or so, the mice completely forget each other) I always feel bad for the highly trained scientists sitting over the mouse cage with a clicker or stopwatch watching mice sniff one another….

  25. Morten says:

    For depression I think the best model is probably stressing rats (various random stress e.g. no water/food, noise, light all night). Some of the rats became kinda lethargic and don’t respond to reward as measured by how much sugar water they drink. Some of the rats are fine tho (like people!). There’s a company that “makes” them (yipes, I’m not supposed to advertise here am I? Sorry).
    But considering the intra-human variance in BBB permeability for various drugs I don’t consider any animal models useful. Are there any reverse studies? Like taking all the human CNS active drugs and injecting them in animals to see how much ends up in the brain?
    My second vote goes to lobotomizing primates to model Alzheimer’s – anybody able to verify that story?

  26. Peter Ellis says:

    On the lines of Alzheimer’s – how about aging in general? *Any* model in any species that shows signs of “old age” (however defined) before about 40-50 years of age is by definition following a different pathological course than humans.
    There exist mouse models for the various progeroid syndromes, but mistaking those for models of aging itself is a category error in my opinion.

  27. Derek Lowe says:

    Regarding Chemist’s story in #19, I have to say that I’ve never heard of anything like this. I did Alzheimer’s work for about seven years in the 1990s, and I can tell you that *any* sort of primate study was a very rare event.
    Plus, the sort of thing your roommate is describing doesn’t sound like much of model to me. Perhaps it could be a brute-force nerve growth test, but there would be no reason to do such a thing in a (hugely expensive, very dubiously ethical) primate study. I can’t imagine a justification that would have gotten this past an Animal Care and Use review.

  28. MedChemKid says:

    I learned this past semester in my undergrad about Lipinski’s Rule of 5. I was told it was based on a human model for oral bioavailability. One thing that didn’t get covered is when did this Lipinski fellow came up with this “rule”, and how data came about to back it up. I guess my question to all of the experts who read this, or anyone who cares to chime in, is are there rules like this for other systems, ie. rats/dogs/chimps etc.
    I just found it odd that in the next lecture he named off a bunch of drugs which don’t fit this rule, so just how closely does this type of classification of hit/lead molecules contribute to intial stages of research?

  29. Gil Roth says:

    This month, I received an unsolicited manuscript for the pharma/biopharma outsourcing magazine I edit (and for which Derek writes a column), Contract Pharma. The article was called, “Novel Approaches &  New Frontiers in Primate Toxicology.”
    I only bring this up because it included the great subhead, “The Expanding Role of the Marmoset in Preclinical Testing.” Just thought I’d share.

  30. weirdo says:

    The short answer to your question is that he did a retrospective analysis of marketed drugs. It is not based on a model, but on actual data from real drugs. (BTW, if your professor discussed Lipiniski’s rules without giving you the reference — which would tell you Lipinski published his first article on the subject in 1997 — then you should ask for a refund of your tuition).
    In answer to your second question: why would anyone care to have such guidelines for rats/dogs/etc.? What would be the point?
    And your third question: There are indeed many companies that adhere strictly to “Lipinski’s rules”, but for the most part they are governed by individuals who never read the original paper nor understand its implications. There are even CROs that market “Lipinski-compliant” libraries for screening.

  31. markm says:

    I have to think that the only way the “learned helplessness” model in rats is useless is that the bureaucrats who do similar things to humans in welfare programs and sometimes even schools, are never going to recognize the analogy…

  32. SBC123 says:

    As I remember, the Lipinski’s rules were based on the compounds that have passed phase 2, not only marketed drugs. Their usefulness can be easily checked by looking at the structures of the ~1200 marketed drugs (you can find them in the drug bank). If you want a quicker answer, they are not that useful.

  33. Kim says:

    What about the study of depression (or any other DSM-classified psych condition) where placebo is used in a clinical trial (all humans, the most entertaining animal model)? I’ve drafted so many study reports and publication drafts where we tried to explain the “placebo effect” as an artifact….yech….

  34. Petros says:

    Another model that is pretty useless is the passive cutaneous anaphylaxis (PCA) model that was used to assess potential anti-allergic activity- i.e. like cromoglycate.
    In the late 70s and early 80s there were was rarely an issue of J Med Chem without at least 1 paper using this model. And virtually nothing progressed to the market
    With regard to the comments on asthma models, they are much better albeit imperfect. The sheep model isn’t over useful because almost everything works in it and the model is very difficult to perform succesfully outside of Miami
    The lack of new classes of asthma drugs, since the cysLT1 antagonists, is partly due to the problem of competing with inhaled steroids.

  35. sabrina says:

    I’ve got it! Rat models of female sexual dysfunction! Unlike male sexual dysfunction which involves physiological manifestations such as premature ejaculation and inability to get an erection, female sexual dysfunction is mainly characterized by psychological issues such as lack of libido and arousal. The lordosis model looks at how receptive a female rat is to being mounted by a male rat. Enough said?
    I have got to think that the the motivation of a rat and a human to have “relations” are very different…

  36. MolecularGeek says:

    Now wait a second. I’ll agree that rat models of female sexual dysfunction are pretty useless, but it’s a pretty sweeping statement that male sexual dysfunction is a matter of physiology and female sexual dysfunction is a matter of psychology. In both genders, sexual dysfunction is a complicated melange of both physical and mental factors. Clinical experience with SCI patients certainly demonstrates that reproductive function can be divorced from any pleasurable sensation (or any conscious sensation at all). The bigger problems with rat models for dysfunction are that the endpoint for treatment in males is easily definable and observable (albeit with a magnifying glass) whereas it is much less so in female rats. I suppose that line a of rats could be developed that did not lubricate and demonstrated the symptoms of primary dyspareunia, but that wouldn’t necessarily stop a male in heat, as anyone who has dealt with cats will tell you. If the target diagnosis is anorgasmia, an animal model is in even more trouble, because there is only anecdotal evidence in the literature of laboratory observation of female orgasm in primates, and observing it is a hit or miss activity too. Now, whether that is because of underlying physiological factor, or because our primate cousins are far more psychologically complex than we give them credit for is left as an exercise for the reader 8).

  37. MolecularGeek says:

    Lipinski’s rule isn’t all bad, but it seems to be more useful when you allow one or two of the criteria to be violated, and you allow the chemist to force inclusion of compounds into screening when they look right. Bean counters in management love it because its simple enough for them to think they understand it, and it draws a nice line around a small enough portion of chemical space that they feel like the medicinal chemists won’t get involved in anything more than 2 or 3 steps long, or heaven forbid, designing their own synthetic schemes. The other issue I will point out is that when using the rule of 5 to screen databases, it tends to select for compounds that are drug-like, and not lead-like. Personally, I think that looking for a clinical candidate directly from a screening library is a path to madness. You’ll get further with a structural diverse library to find leads, and letting the synthetic teams do what they do best and optimize it to hell and back.
    Then again, I am not a big fan of trying to drag a sieve through the databases and trying to pull out the 5 compounds that you think will work. I think that the REOS (rapid elimination of swill) method that Charifson and Walters at Vertex described in a paper in Molecular Diversity in 2000 is saner, where really bad compounds (heavy metal containers, michael acceptors, poly-flouros, etc) are set aside, and the scientist requesting the screen is then given data to help select a representative subset of the database hits for screening. To quote someone or other: “computers should compute and people should think”

  38. sabrina says:

    I did not mean to imply that male sexual dysfunction has NO psychological component, only, as you discussed, that there are more physical aspects to measure in males than in females. If you want to measure vaginal contractions and lubrication in female rats (and these things have been done), then you may be able to physically characterize female sexual dysfunction in rats. But are these things really predictive of sexual dysfunction in human females? There are many lubricants on the market, and they haven’t made a dent in curing female dysfunction. Most female dysfunction is classified as lack of libido — how do you measure that in a rat? On the flip side, the sucess of marketed drugs that only cause an erection highlights the fact that male sexual dysfunction is highly physical. My guess is that much of the psychological aspects of male sexual dysfunction comes from not being able to perform physically while the converse is mostly true in females.

  39. JC says:

    Maybe if the male rat listened once in a while, brought some flowers to the cage sometimes & remembered the female rats birthday & their anniversary for once.

  40. Dr. Richard N. Bennett says:

    Basically the large majority of animal models do not provide useful data that can be extrapolated to humans. This is especially true of the metabolic studies for drugs and dietary phytochemicals. Not only are there differences in the fundamental GI tract anatomy/biology and biochemistry (e.g. numbers of isoforms and substrate specificity of transport proteins) but since most rodents eat their faeces their GI tracts are loaded from the buccal cavity to the oesophagus with bacteria; unlike humans where bacteria are primarily restricted to the colon. So bacterial metabolism in rodent models can have a big effect on the results e.g. bioavailability. Other than necessary toxicology studies it is proving of greater value to perform in vitro human cell studies and in the case of dietary components in vivo human intervention studies (taking blood, urine and certain biopsy samples for ADME and gene studies). It is time we moved away from pointless animal models and invested money in developing more relevant human models using the wide variety of cancer and immortalised cell lines.

  41. ujla says:

    i think making a mouse model of SLE is most difficult as its a 5-6 months task.mortality is high in these animals

  42. steve says:

    By the way, the whole “mouse models aren’t good for sepsis” story was overblown and probably wrong. I tried to send the links but they were help up for review. See if these work:

Comments are closed.