Skip to main content

Drug Assays

Phenotypic Neuroscience

This is a good review from AstraZeneca scientists on phenotypic screening in neurodegenerative disease (by which one means Parkinson’s, Huntington’s, frontotemporal dementia, Lewy body dementia, and of course Alzheimer’s). And it’ll serve as a good intro to the challenges in these two fields in general, and to why they intersect. Put simply, we don’t have a detailed enough understanding of neurodegeneration to be able to pick out the key steps (the most important enzymes, receptors, or other proteins) and screen against them individually. You can try (and people most certainly have) but you’re taking a huge, expensive, risky shot into the unknown – which is what the history of Alzheimer’s drugs in clinical trials will tell you. You target a proteins, you develop a clinical candidate, you test it in humans, and it doesn’t do them any good (or, in some cases, does this outright harm instead): so is that target a bad idea? Or did your drug not hit it hard enough, or in quite the right way? Or did it do other things that you weren’t aware of that cancelled out whatever good effects it might have had? Want to drop another half-a-billion dollars to try to find out (no guarantees you’ll understand the second or third answers, either)? Welcome to the field! Reductionism has taken us a long way in all areas of science, but the nervous system and its functions are one of the areas where that particular magic wand is least effective.

So phenotypic screening has a lot of appeal. Grow yourself some neurons that do what they do in the human disease, and test your compounds against them directly. Instead of trying to understand the whole disease process up front, find compounds that alter it and go in both directions: forward into the clinic and backwards into the disease mechanism with your new research tools. Ideally, that is exactly what a phenotypic effort is supposed to do for you.

But experienced screeners and CNS types will note that I have performed a silent broad jump in that last paragraph that more resembles the migration of the Arctic tern: grow yourself some neurons that do what they do in the human disease. Oh, yeah. Just go do that. The problem is that neuronal cell culture is a black art, full of nonphysiological oddities, mostly because in vivo neuronal architecture is so complex and (still) poorly understood. Progress has been made, for sure, with moves into three-dimensional cell culture and co-culturing of more than one kind of cell as a mixture (both of which try to recapitulate important features of actual nervous system tissue). But as can be easily appreciated, both of them introduce a huge number of new choices and variables into the culture conditions.

And over all these considerations is the problem that the cells you’re using may not be able to reproduce human biology very well at all – a line from this review is “Perhaps the historical mainstay of neuroscience-driven phenotypic screens is based on the use of immortalized cell lines“, and the cynical reply is “Yeah, and just look at it”. As the authors go on to discuss, two ways around that problem are to use primary cells or stem-cell-derived cultures, but both of those have their difficulties too, of course. Just to pick one, your primary neuronal cells are almost certainly going to be from rodents (at best) since there is, for obvious reasons, a supply problem with living human neuronal tissue for running screening assays.

That’s the cell-culture end of things. On the other end of the process, it’s not always clear how you’ll even know that you’ve hit on the right conditions that mimic the real in vivo situation. You can try to match up with the larger histological features of the disease (generally known best from the late stages, i.e. autopsy), but the earlier stages are harder to recognize and aim for. You may also find that you’ve made something that looks kind of like the disease state in the end, but got there through a completely different path, and that may do you less than no good at all. There is always the option (reviewed in the current paper as well) of going outside of human cells entirely and screening in a smaller model organism (all the way down to yeast or nematodes). Relating those phenotypes to human neuronal ones, though, takes a peculiar mix of bravery and caution, although there have been some discoveries made nonetheless.

As much as I like a good phenotypic screen, I’m fond of saying that a bad phenotypic screen is the worst of both worlds: harder to do and full of even more uncertainties than ever. It should go without saying that there is (for example) no such thing as a reliable cellular phenotypic screen for Alzheimer’s. I attended a phenotypic screening conference earlier this year and gave a talk that hammered on some points that Jack Scannell and Jim Bosley made in this article (blogged here) – namely, that there is no possible substitute for predictive validity in a screening cascade. Small increases in PV and thus assay “transferability” towards clinical relevance are worth a great deal, although they’re frustratingly hard to measure. The entire idea behind running phenotypic assays in the first place is to have improved predictive validity compared to the alternatives, so if you haven’t done that, you are setting up for a waste of time, money, and resources.

But with all of these concerns, this is still a very worthwhile area of research, because it still has the best possibilities for finding new modes of action and new chemical matter. (That does tell you something about the alternatives). This review has a lot of solid advice about compound screening libraries, secondary assays, hit triage and more, and anyone getting into this sort of work for the first time had better read up to make sure that they haven’t missed some of the issues. As with many other areas of drug discovery, the state of the field is simultaneously a warning signal and an invitation: any advance that can be made will be very welcome indeed, and there are still a huge number of things that haven’t been tried. Go in with a clear head and appropriate skepticism, and you’ll increase your chances of actually finding something – and there’s a lot that needs to be found.

15 comments on “Phenotypic Neuroscience”

  1. Oligodendrocyte expert says:

    “Grow yourself some neurons that do what they do in the human disease, and test your compounds against them directly.”
    Therein lies the problem. What if it is not the neurons that are the initial problem? Perhaps they are a middle of the pack domino in the etiology of Alzheimer’s disease, in particular. As I have pointed out in an earlier comment in response to a post on Derek’s blog (“Your Brain Shift Gears” 9 May 2013): “While understanding the human brain is admittedly a daunting task, paradoxically, it becomes easier once one acknowledges brain function is far more complex than a series of communications among synaptically connected neurons. CNS macroglia (astrocytes, oligodendrocytes, and NG2 glia, the latter a.k.a. oligodendrocyte progenitor cells) express most–if not all–the same neurotransmitter receptors expressed on neurons. Thus, these macroglial cells are intimately and actively involved in the “conversations” that drive brain function.”
    It is possible, I would say likely, the etiology of AD begins with an injury to a non-neuronal cell type. So, phenotypic screening against neurons will be a futile exercise (on more than one level).

    1. colintd says:


      For instance, I’ve seen some very interesting papers over the last year suggesting a pivotal role for gum disease bacteria in a number of conditions including Alzheimer. If they were to be the primary cause (and I’m not saying they are), then any cell model which didn’t include them would be doomed to failure.

  2. Andrew Molitor says:

    Regarding a supply of live human neurons, I wonder if there’s any way through the ethical thicket of “I would like to donate my still-living body to science after some measurable checkpoint of neurological degeneration”

    I have a family history of vascular dementia, so this doesn’t strike me as a completely insane personal choice. I watched my father go, and it wasn’t pretty.

    1. Dave says:

      I had a buddy who died of Pick’s Disease. That was a horrible way to go (as if there’s a good way to go).

  3. Mzero says:

    I know this is a hornet’s nest in many countries (not in China though), but wouldn’t abortions be a source of living human neuronal tissue that is already more than eager to divide?

    1. Ed says:

      Getting lots of human neurons isn’t the problem. It’s actually pretty straightforward to differentiate stem cells (induced pluripotent or embryonic) into a variety of neuron types.

      But that variety is also part of the challenge. Is the best way to model your phenotype in a heterogenous population of whatever neuron subtypes happen to grow best on plastic? Or a single specific subtype? Or a coculture of neurons and glia? Maybe an organoid might best recapitulate the disease state…

      Let’s say you’re confident that your phenotype is best modeled in whateverminergic type 7q neurons, enough try them straight away. Now you have to decide which of the four published differentiation protocols is best. Probably best to try them all and hope one works – and now you’ve spent the better part of a year just in the initial setup and characterization of your phenotype. Now comes the hard work of getting your differentiated neurons to behave consistently enough for a screen…

      Basically there’s not yet any substitute for a lot of brute force empiricism, luck, and what might as well be black magic (blood of an unborn calf, clotted and clarified, on a dish coated with the secretions of a mouse sarcoma, with a pinch of extract of thymus…)

  4. a. nonymaus says:

    A phenotypic screen is likely to recreate the time-course of the disease. That is to say, a good tumor model will grow at the same rate of cell division as a similar tumor in the patient. Of course, for diseases such as Alzheimer’s, you may then have to maintain your neuronal cell culture for decades as whatever slow biochemical process takes place takes place.

    1. Nesprin says:

      That’s a sticking point though- pretty much all tumor models grow faster than the tumors they came from.

  5. Paramus says:

    First things to test are all the treatments that have failed in man. They should be negative in your ‘Alzheimer’s Brain on a Plate’, what could go wrong!

  6. Adiantum says:

    Thank you for an interesting article, Derek (and the reminder that I should read the article you blogged about previously). However, I’m curious: What are the best alternatives that we have available to us at the moment (granting that all models are wrong, even if some are useful)? I’ve seen that people are quite excited about CRISPR-powered screening (more efficient than shRNA screening): is this where better technology will allow us to create more and more complicated models through multiple edits? What progress is being done with patient-derived cells at scale (or is this partially replaceable by CRISPR-edited cells?)? Organoids? What are the best tissue/metabolic models? Or are we stuck with animal models? Or (and sorry for going on) could we afford to be more agnostic about the models themselves if we had a better idea of the underlying biology and what questions we should really be asking?

    1. Barry says:

      On one end of Med. Chem. you can array infectious diseases; the etiology is established. You find a model organism (guinea pig? armadillo?) that’s susceptible to the pathogen, you eradicate (or attempt to eradicate) the infection in you animal model, and you have a good expectation that results in humans will follow the pre-clinical data. There’s a reason that historically, Med. Chem. began with infectious disease (salvarsan)
      On the other end, you have e.g. Alzheimer’s, for which we don’t know the etiology. We don’t even know that the rare familial forms are the same disease. Without knowing the etiology, we can’t know that the pre-clinical model is valid.
      And in between there are messy cases. Took us a long time to figure out that e.g. gastric ulcers are a symptom of infectious disease. And the multifactorial etiology of cancer has meant that correlation of pre-clinical models to human disease response has been spotty.

  7. Uncle R says:

    Degenerative needs regenerative. No access to full text review, but as fully paid up wrinkly can’t help feeling mental decline a one way street we all head down sooner or later. A matter of how fast, how far. Stuck with it. Sorry everyone.

    So bright young things, Seize the Day! Bright old things keep the old grey matter flying. Savour the creative. Avoid the degenerative like the plague. Which in a past life included any amount of neuronal spam (Opal time cards, Focus Surveys, Performance Management, etc, etc, etc, sound familiar anyone?). No doubt since superseded by more modern managerial tools. So it goes.

    Get the captcha right. Seize the morning.

  8. loupgarous says:

    “Just to pick one, your primary neuronal cells are almost certainly going to be from rodents (at best) since there is, for obvious reasons, a supply problem with living human neuronal tissue for running screening assays.”

    Clinical experience with cadaver-origin nerve tissue grafts indicates that it’s all too easy to get Creutzfeld-Jakob Disease-positive nerve tissue that way – we had a worldwide epidemic of iatrogenic CJD borne by cadaver-origin hGH. Why’s it difficult to get human cadaver nerve tissue useful for the study of other neurodegenerative illnesses?

    1. loupgarous says:

      The same reasoning goes for samples of living nerve tissue – has the rate at which brain biopsy is used for diagnosis of neurodegenerative disease fallen off in recent years? The literature of Creutzfeld-Jakob Disease and its iatrogenic spread is full of instances in which re-use of surgical instruments, indwelling electrodes, et cetera between neurosurgical patients was a highly suspected vector for transmission of CJD. I always thought that implied a fairly high rate of brain tissue biopsy.

  9. Tim BERGEL says:

    Trivial typo in the second sentence: good into to -> good intro to?

Comments are closed.