Why Drugs That Work in Mice Don't Work in Humans
thelri.org
thelri.org
Regardless of the reason, this seems fine/workable. Animal studies are a step in the funnel. I assume the step from petri dish to mouse is similar.
I wonder how many false negatives get produced. Drugs which would have worked on humans, but don't on mice, for similar reasons.
And the pharmas would just add the cost of 86% failed trials to the other meds that do pass, so they effectively pass the cost of their bad decisions to us.
I worked in a lab that studies antibody therapeutics for almost a decade and can go on a tirade of all the shitty decisions companies seem to make in candidate identification, but fundamentally I really believe that this number won't be this low if the people making the decisions in pharma companies know what they're doing but then the CEO of gsk used to sell lipsticks so who am I to talk right?
My impression is that you are _mostly_ correct (a lot of animal models for disease don't accurately reflect the analogous state in humans).
However, I was privy to a situation similar to that described by the person to whom you replied.
* A start-up that I worked at developed a "candidate drug" to treat a condition (I don't want to get too detailed).
* The drug was good at ameliorating the disease in our simple models, but we also had a strong suspicion that the drug would be toxic in humans, higher animals.
* A decision was made that we would determine if the drug was toxic in three different species of large animal (I think we chose some type of monkey, dogs and I don't remember).
* Our "go/no-go" decision was: If the drug was NON-toxic in at least 2 of the 3 animal species, we would move to clinical (human) trials.
* 6 months later, we got the results: drug was NON-toxic in 1 of the 3 species.
* This looks like "no-go", right?
* Wrong. Board of directors put pressure on the CEO and senior management ("We put 8 figures of our hard-earned $ into round B and you promised us clinical trials over a year ago.").
* CEO folded like an accordion and got everyone to agree that we should proceed w/ Phase I trials (small n of humans tested, primarily to determine toxicity and maximum tolerated dose) despite the results of the animal studies.
* Phase 1 trial starts. Everyone gets sick (maybe 10 patients) after about 5 days of taking the drug. Trial is halted.
* We go back to square 1.
CEO should have never folded under the pressure because it was a waste of money, time, and it put people at risk.
I've only seen this once (going to trial with a risky drug), so I think it is a rare event.
edit2: Ironically, this is an example of a situation where the animal-studies CORRECTLY predicted what would happen in humans.
At the same time, he board of directors should never have exerted the pressure, for the same reasons.
That company was unusual in that there always seemed to be a tension between the board and the company (distrust). And both sides (CEO and board) seemed to have a short-term view of everything (we were going to "flip" the company to a large pharma), which caused a lot of bad decisions to be made.
The other (more successful) start-ups that I worked at had a much more harmonious relationship with their boards (e.g., "We are in this together and we are here for the long-term.")
> which caused a lot of bad decisions to be made.
Unfortunately, if they were trying to flip the company; then they may have made the 'right' (from the perspective of payoff) call.
I can imagine under some models, that if you are 86% confident that it won't work, and only 14% confident that it will work; then the millions wasted may have been easily justified by the potential payoff if it was in the $100 million dollar range.
Normally, we wouldn't bat an eye at such a decision; but the ugly part here is the high risk exposed to the people participating in the study :(. However, even that, a similar utilitarian calculus can be done to justify it. bleh.
Remind me not to be a Pharma CEO, it sounds like an all around unpleasant job if you aren't a complete sociopath.
No idea. Thankfully, all patients stopped the drug and the toxicity went away.
I remember having a company-wide meeting where they showed us a result from the Phase I trial. It was a graph of time on the x-axis and the concentration of an important molecule in each patient's bloodstream on the y-axis. Every patient had the concentration of that molecule drop to dangerously low levels after about 5 days on the drug, and I'm thinking, "Is this a surprise to anyone here?"
Most antibody therapies (which form the bulk of currently explored therapies) have to be tested on modified mice models simply because of the fact that these antibodies are raised against human targets that have to be engineered into the mice. But then such artificial models make lots of complications in their interpretation which people have to spend lots of time interpreting. I've seen on several occasions from just published data that there were glaring warning signs that a potential therapy might have adverse interactions with the immune system or such, and the company would just go ahead with trials anyway.
According to a widely cited paper on the cost of drug development [0], phase 2 has the lowest probability of succes of any stage in the funnel at 30-40%. Phase 2 is generally the first time a drugs effectiveness is studied in humans. It can cost $50-100M+ to get a drug through a phase 2 study
One of the major advances in drug development in the last few years is to reduce the "translational risk", i.e. The risk that animal models and other disease models are not predictive of human outcomes. A recent study suggests that we have actually started to make some improvements in this area which is a huge step in lowering drug costs and getting more new medicines
In general, the AWA is very restrictive in terms of animal testing for scientific research. Getting exceptions to use an animal for research is very intensive and highly regulated, taking years to get the approval. Getting approval for studies that may involve studying intentional pain, harm, etc to an animal can take up to a decade for approval, and with much study and research on other possible methods. Internal Review Boards (IRBs) are required to include local lay people, clergy, other non-scientists, etc in order to approve a study. Vivariums are also highly regulated and frequently inspected.
The US, at least nowadays, takes the welfare of animals extremely seriously with stiff fines and real consequences for breaches and lapses in protocol.
https://www.aphis.usda.gov/animal_welfare/downloads/AC_BlueB...
The Animal Welfare Act specifies the minimum standards of care for research and exhibition of most species of mammals. Birds, rats, and mice aren't covered by the AWA, nor are "cold-blooded" animals. These are covered by other regulations from the Office of Laboratory Animal Welfare (OLAW; for federally-funded projects) or AAALAC (private funding). Farm animals are also not included, unless they are being exhibited or used for research.
The AWA requires research institutes to form an Institutional Animal Care and Use Committee (IACUCs), which includes scientists as well as clergy and members of the community. This is usually different from the Institutional Review Board (IRB), which regulates research involving human subjects. The idea is admittedly similar though.
The AWA does not limit the species involved in research. In theory, you could propose to set up a grizzly bear colony or something and the University of California has (had?) a hyena colony for a while. The IACUC is responsible for making sure that animals receive the care required by the AWA, OLAW, and other standards, so they might turn you down if suitable resources are not available or other, more humane approaches could be used (e.g., use a species that is less affected by captivity).
Robotic automation and synthetic biology help with that a lot. Machine learning can help make sense of more limited data, for example predicting toxicity and other parameters for drugs so that you can better choose what to forward to Human trials.
Nothing can replace certain studies in Humans completely. We can only hope to make those more effective by better choice of candidate drugs.
They haven’t shown to be a good proxy for a whole organism.
On term of disease model, they are typically artificial in the sense that genetically engineer the cell with a specific mutation that may or may not represent the disease or the population.
And if an experiment doesn’t work the way you want, you are a new set and sometimes it works the way you want.
The only two things they have over mice is that they can be from human origin, and sometimes that human origin might be in a disease state that you want to study, which might be the only way to generate a model for complex disease...
Of course, we are talking about the fruits on the drug tree which are potentially reachable by singular small molecule drugs.
Maybe you'd first need peer-reviewed research to prove that...
"He abandons [Dr. B.M. Hegde] all rational thinking and embraces fantastical belief systems. The sad thing is I personally know many people delaying cancer treatment being swayed by his speeches and ending up with incurable metastasis later."
https://www.quora.com/Do-you-agree-with-the-controversial-vi...
An appeal to the ancient wisdom of Ayurveda doesn't really stand on it's own until it's backed by solid peer reviewed science. It's really not that hard, a well designed set of trials proving your point will have the entire world at your feet. Modern science is extremely receptive of new ideas that way.
It’s true that industry is funding some (publicly published) research directly, and that some research publications fail to disclose their funding source and other conflicts of interest. But for the vast majority of biomedical research, especially fundamental research, this simply isn’t the case.
And it requires a complete suspension of one’s critical thinking capacities to imagine that some vast, weird conspiracy encompasses all of public research, to suppress the “truth” that Ayurveda works, contradicting everything we know from modern medicine as well as basic physics and chemistry. And all that just to make a few pharmaceutical company bosses rich? Why would I, lowly researcher on a sub-par salary, contribute to such a conspiracy? It’s completely irrational.
Dramatically increased life expectancy, reduced childhood mortality, etc.
There are problems with the modern health system, but it does seem to produce much better outcomes than what came before.
How is it possible that they could have prevented that information from ending up in the hands of profiteers even after all these years? And no profit hungry pharmaceutical company has ever been able to rediscover and commercialize those techniques, even though the details of them are mostly freely available on the internet?
As for commercialization of Ayurveda, I see this as exactly what's happening now with Patanjali etc trying to cash in on people's trust in their culture without providing a proper research base for their claims.
Another reply to your comment above makes very good points about research funding, and in my experience, funding is to a large extent non partisan, and free from industry influence. There are even rules on disclosure of funding about major studies, which makes it possible to criticize them.
I only wish that Ayurveda is held to the same standards as other medicine, and passes through the fire of testing the same way all modern medicine has. It's how we know that antibiotics work, or about interactions between medicines, or about side effects and complications. It will ultimately benefit the field, and medicine as a whole.
Typing out that you are prepared for downvotes doesn't really change who has drank the Flavor Aid.
I could be wrong about the exact drug but it is something similarly common if it isn't. Hard to google because there are a million studies which pop up.
I couldn't really make sense of the two proposed theories, other than one suggested perhaps it was an advantageous adaption in the presence of malaria.
Phase 2 failure -- ie stuff failing in humans after working in mice -- is the biggest cost driver in drug dev. You spend $50-100M+ to get there and 35% of things fail. If you could eliminate you'd cut a massive chunk of the cost and risk of drug dev
However I don't think ML is the thing to study if you want to create ML to predict whether drugs will work in people. There is not enough data about how the human body works to develop a good model. Better to work in biology to develop good models of disease, or in developing tools to better measure the molecular biology of the living human body. It's unclear if predicting drug effectiveness with ML will be possible in our generation
For more near term applications of ML in drug discovery and development see here: https://www.getrevue.co/profile/nathanbenaich/issues/6-impac...
edit: but it won't do much to optimize P2 trial success specifically, it'll simply help iterate substantially faster in the pre-P2 trial phases.
(Ethical nightmares)
The reproducibility crisis is happening in academia. There isn’t that much room for uncertainty when it comes to industrial drug development. You won’t find drugs making it all the way to market only for “another group” to be incapable of reproducing the effect.
This is sampling to a foregone conclusion. It is guaranteed to yield unreproducible effects.
> "You won’t find drugs making it all the way to market only for “another group” to be incapable of reproducing the effect."
I doubt this from the description above.
More positively we could make a lot more use of domestic pets in research. If we started to treat more pets as research subjects we could really do a lot of great research.
Phase I clinical trial: https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/28374166/
Phase II is in progress.
Interview with Matt Kaeberlein from the project: https://www.leafscience.org/dr-matt-kaeberlein-the-dog-aging...
There is a huge variety of mouse strains in use and under development. Basically anything you would see in a "fancy mouse" you can have as a strain.
Pets, like cats and dogs, owned by private owners, are used for studies all the time. And not just for veterinary purposes. Dogs have a wide variety of tumors, and studies on these can be used to inform research for Human oncology.
Beagles, kept in laboratory populations, are routinely used to study the cardiological safety of drugs. Such studies are rarely fatal, and dogs often have several of them during their time in the lab. They live in "colonies" of about 10 dogs each, and they look and behave just like very happy pets, even if a lot less trained. In the cases of the colonies I have been in contact with, they live in the lab for about 6 years, then they are spayed/neutered, get a complete dental and are adopted by private owners, often vets and vet students whom they already know.
Yes I know pets are used in studies, but we use pets to only a fraction of their potential. We have 10s of millions of pets and we use a few thousand at most every year. I am lamenting the waste.
A way to evaluate drug efficacy that relies less on comparisons of large groups would be a big step, because there are a lot of disorders that are probably multiple mechanisms manifesting in similar ways
Ultimately there is always going to be some trade-off between making a rigorous statement and a generalizeable one, especially because biology seems to have some pretty messy abstractions. As collecting and analyzing large amounts of data becomes more feasible, I don't see why there shouldn't be some efforts to consider genetic (or environmental) heterogeneity in animal models. Ideally I think it'd be cool to approach any given question with parallel methods that can try to address how the hypothesis holds up both in more narrow but well controlled situations, and in the "wild"
Yes we would do fewer experiments since you would need more mice per experiement, but the results of those experiements would be far more robust.
That is, results can vary based on the mice used for reasons not understood (but gut bacteria perhaps).
I know some drugs are extracted by isolating a compound from a traditional remedy. That obviously makes sense.
Then, if it's a chemical target, they'll make thousands of compounds to test in their preliminary tests to make sure the compound very basically attaches to the target. The ones that make it past that might make it into a cell-based assay then rodents. The path each treatment takes through specific assays is different for every project depending on the specifics, but it generally follows that progression.
Almost always, though, you are starting with a particular medical problem in mind, so the first step is to develop some kind of assay for detecting compounds which might be useful in treating it. Ideally, this would be a simple biochemical reaction, but it might be something involving cell culture.
For example, if you wanted to find new painkillers, you might look for chemicals inhibiting cyclooxygenase (as ibuprofen does). You can buy kits for doing that assay commercially [1], where you prepare a solution of the enzyme, add your test chemical, then add a substrate which emits light when the cyclooxygenase breaks it down, and measure the intensity of light produced.
If you wanted to find new anti-cancer drugs, you might look for drugs which cause proliferating cells to get stuck in the metaphase step of the cell cycle (as paclitaxel does). You would plate out some rapidly proliferating cells, add your test chemical, wait twelve hours, then fix them, stain them with a DNA-specific dye, and use a microscope to count the number of cells in metaphase (which is quite distinctive [2]). This is a lot more tedious than the cyclooxygenase assay, but we have robots that can handle liquids and plates of cells, and operate microscopes, and process images, so it can be highly automated, at a cost.
Then you take your assay and go hunting for molecules.
One approach is indeed just to start with a wide range of compounds. You can get libraries of small molecules [3] [4], so you give them to your robots (or graduate students), and put them all through your assay to find which ones work.
You can also start with mixtures of compounds, perhaps obtained from natural sources. For example, you could go and collect twenty species of fungus or sea sponge, grind them up, and put the extracts of each through your assay. If anything works, you then fractionate the extract somehow (eg by chromatography), and put each fraction through your assay. You pick fractions which work, fractionate them further, assay the sub-fractions, and repeat until you have got a pure substance with some activity, which you then characterise. Here, you can knowledge of ecology and biology to pick likely species - for instance, fungi are a good source of antibiotics, because they have to make antibiotics to defend themselves in their natural habitat.
Or you could start with some knowledge of the structure and function of the target (from X-ray crystallography, NMR, and good old fashioned biochemistry), and try to rationally design a molecule which will bind to and inhibit it. Computer simulations are useful here. Combinatorial methods let you design hundreds of molecules which might work, and then put them all through the assay.
Or you could hope that an antibody will do the job, and inject your target protein into some mice, wait for them to make an immune reaction to it, then collect their blood, extract B-lymphocytes, culture them in bulk, purify antibodies from the culture, then assay the antibodies. If something works, split the lymphocytes into single-cell clones, and assay each clone's antibodies one by one.
I don't work in this field, so my knowledge of these techniques is from undergraduate study, and one relative who grinds up sponges. It's possible some of the approaches i mention are obsolete, or were only ever speculative.
[1] https://www.abcam.com/cyclooxygenase-cox-activity-assay-kit-...
[2] https://www.le.ac.uk/bl/phh4/roottip.htm
[3] https://www.tdi.ox.ac.uk/small-compound-libraries
[4] https://wiki.nci.nih.gov/display/NCIDTPdata/Compound+Sets
If you roll a die enough times, eventually you'll get a 6.
Even this bias longevity piece needs to do that. This 4 breeds will all have different immune system, cognitive behavior and lifestyles, which are all central to aging.
We can even argue that Human !== Human due to a million variables that differentiates them at microcellular level.
Very nice to see an approachable post out there outlining a lot of these issues.