Disease Models and Target Validation: An Interview with Professor Gavin Woodhall

In a recent interview with Proventa, Professor Gavin Woodhall, research scientist and Head of Neuroscience at Aston University, discussed the challenges of modelling human disease and the impact of alternative models on target validation. Professor Woodhall has an extensive academic background, whose research focuses mostly on electrophysiological studies of epilepsy in animal models and human tissue. However recently, he has begun to investigate rodent models of schizophrenia to better understand the complexity of the disease and potential drug targets. 

Proventa International: Why do we need alternative methods of modelling human disease? What factors need to be taken into account when choosing a model, animal or not, for a specific human disease?

Professor Gavin Woodhall: So you’ve probably come across the different validities, predictive validity etcetera – you’re looking at those really. From the epilepsy side of things, for a long time I didn’t think the models were very good. They were essentially models of traumatic brain injury most of the time, and so I was keen to develop a model that had better construct validity, so it looked a lot more like epilepsy.  

With schizophrenia, we’ve looked into all of the different animal models and the one with the best trio of validity is the chronic PCP model. But even with that, there are aspects to it that I don’t like – it is a really unpleasant and difficult model. So, we’re just going down the route of implanting subcutaneous osmotic mini-pumps to deliver the drug for a longer period at a constant rate, with the hope that it works. 

From a drug discovery perspective, we want the best predictive validity. We want to know that our drug will work in man. So with our schizophrenia work for example, we compare new drugs with clozapine, the gold standard treatment. 

Why are animal models important in development treatment for human diseases? Could you explain in the context of the RISE model?

GW: Without an animal model, you can’t do any kind of reasonably good treatment screening, and you can’t experiment on people. You need an animal model to help you select the right targets and the right drugs. If you can make your animal model alongside human tissue, then you can use that to address whether your animal model replicates the human brain and cross-validate. That’s what we tend to do.

How useful are in vitro vs in vivo animal models? Is one better than the other or is it dependent on the disease type?

GW: It depends on the disease. You can use stem-derived IPSc (induced pluripotent stem cells) for example, for certain diseases. There is also a trend in pharma towards disease-agnostic modelling, so you can take a bunch of cells and grow it into some sort of organoid and you can start to look at those. These are showing some promise and do have their place, especially if you’re looking at a molecular mechanism.

However, if you’re looking at more complex stuff like schizophrenia with multiple brain regions, or autism, then you need the complexity of an animal model, rather than an in vitro model. Colleagues of mine are creating a neural network chip, using cultured neurons to create a computational device. That’s a really good example of using in vitro technology to bring advances to neural network computation – to do interesting and exciting things that you can’t do in vivo.

What are the limitations of animal models?

GW: Translatability. There are lots and lots of models where the translation back to the human brain isn’t very good. The methylazoxymethanol model in schizophrenia, or any number of epilepsy models don’t translate well to the syndrome in the human brain. Human brain cells are different from rat brain cells – we don’t know necessarily if they do the same sort of things, or even if the same sort of cells exist. There’s a lot more diversity in the human brain. 

You can humanise a rat brain, that might help, but there are other issues with that from different perspectives. They’re now talking about humanising a chimpanzee brain: there are a few genes you need to flip the switches on and you can get massive frontal cortical expansion just the same as we have. So theoretically, you could create an ape with a seriously human brain. But obviously that throws up its own issues. 

The main disadvantage of animal models is the lack of similarity and in some cases, like Alzheimer’s mice, mice also don’t live very long and so don’t show the same kind of pathology. Some things are just really hard to model in rodents.

How can animal models be used to improve target validation in drug discovery?

GW: (Animal models) are very useful if you use them in the right way. I, for example, am very interested in anticonvulsant drug development. And what we’re doing at the moment is identifying the similar changes across a sweep of different animal models. So we’ve got different ways of inducing epilepsy: inflammation, traumatic brain injury, chemical epileptogenesis, and through autoimmune antibody-mediated epileptogenesis. If in six or seven different models you see the same target appearing in the multiple different models, then that gives you the confidence that it’s an approach or a target worth following up.

Unfortunately this is really difficult to do, and no one really runs more than one or two models of epilepsy in their lab. It is difficult to get funding because it comes across as over-ambitious if you try. If you can find a target that goes across multiple models of a disease, then that is the way forward in my opinion. 

What can computational models offer vs animal models, both in general and in the context of target validation?

GW: If you can get the right computational model, you can manipulate it virtually, and ask questions which you can then follow up in wet experiments. I’m interested in glutamate receptor dynamics, and with the right type of modelling, you can alter the dynamics of receptor trafficking and things like that which give you experimental predictions that you can then test in the lab. 

Obviously with in silico, it’s massively quick. One of the problems I have with my research is that it takes me a year to do an experiment on a rat with epilepsy – it takes a year to become epilpetic, become treated and then record data. That is a significantly difficult time frame. 

One of the things we’ve done is run animal models in parallel. We now take out brains at P9, post-natal day nine from rat pups, and we culture them. The very process of chopping up the brains and culturing them is a traumatic brain injury and causes epilepsy. This is a really rapid model that takes two weeks to develop epilepsy. You can do all the experiments you want to do on this culture model on a massively accelerated timescale. And you can look at things in the brain like maturation and development, because of the massively accelerated development profile. 

So if you’re working in culture at the same time you’re working in the slower in-vivo models, you get the best of both worlds. In an ideal world, you’d have an in silico arm, a culture arm, and an in vivo arm which you would do all at once.

What is your opinion of in silico models replicating human disorders? Or does it depend on the disease?

GW: Obviously it’s a little bit further away from translating into human stuff, however, it is really important in modelling, because it allows you to understand complicated systems. One really interesting thing with silico modelling is that you can take a complicated system and allow it to evolve it naturally, and then you can tweak it to your parameters. You can re-evolve the system and it can tell you the important element of the system. You can do it without even knowing what’s going on.

Deep neural networks are proven to be really good in brain tumour diagnosis. Say you have someone with a glioblastoma and you don’t know how far it has spread, you can feed your MRI images into a deep learning algorithm, and it can tell you useful information about where it is using subtleties of the MRI images. These deep learning structures are made of six or seven layers of neurons, a neural network. You can run them, but no one knows what goes on in the middle, just like a real brain. You get the right answers out of it but no-one knows why.

So in terms of in silico, do you think it has more of an application in medicine than research or was the tumour diagnostics simply an example?

GW: It is equally as useful in research. So if you’re trying to work out something complicated like turbulent flow of a fluid using the Navier-Stokes equations – which are a horrendous series of partial differential equations with infinite dimensional solutions – you can use in silico.

You can put your input into a double-branched deep neuronal network and train the network to solve those partial differential equations in ways that humans can never do. You can end up  with really complicated mathematics that would normally take a computer years to do the calculations, done in a fraction of that time. You can do huge computations by using in silico approaches that mimic brain networks.

Where do you believe research is heading in terms of representing human disorders? Do you think animal models will continue to be relevant? 

GW: I think animal models have to be for the moment, however many places are looking down on their facilities and closing them. Often it is because institutions get into financial trouble, but it is telling that it is this area that is chosen to close. Animal facilities are under constant review and threat, and lots of drug companies now don’t run drug facilities. In fact, drug companies rarely have many experimental people. Instead they move it out to academics and contract research companies.

It’s not because of ethics: It’s  cheap to do it that way. Instead of a building full of scientists and animals which can cost £4 million to build and £1 million a year to run, you just ask other people to do the work for you.

Charlotte Di Salvo, Lead Medical Writer
PharmaFeatures