Repurposing drugs for rare disease (a LinkAGE webinar)

Lianne Gompertz: Hi, everybody Good evening.

My name is Doctor Lianne Gompertz.

I’m currently a fellow working for the Genomics Education programme.

Welcome to our 4th LinkAGE webinar on repurposing drugs for rare disease. Just a few housekeeping items before we begin.

So microphones will be muted and video cameras will be turned off.

At the end of the webinar, they’ll be time to answer any questions on the topic. At the end of the webinar they’ll also be a QR code for our evaluation, which will help us to improve these webinars. So please, please do stay on and complete that for us.

In completing that evaluation, there’s a space where you can tell us whether you want a CPD certificate or not.

So, our speaker today is my lovely colleague Doctor Hassan Shakeel.

He is an amazing academic paediatrics registrar and development lead within the NHS’s National Genomics Education team. He previously undertook an academic clinical fellowship in paediatrics with the Hurles group at the Sanger Institute in Cambridge, and his work within the open targets and Deciphering Developmental Disorders teams looked at developing computational models of disease phenotypes, comparing variant calling between different genomics projects, and identifying new orphan drug agents for use in rare developmental disorders.

So Hassan, over to you.

Hassan Shakeel: Thanks to the introduction and yeah, my name’s Hassan.

I’m one of the academic paediatrics registrars in Cambridge. And I’m also one of the Genomics Education Programme’s development leads.

Today, I’m going to be talking about my academic clinical fellowship work that I did, primarily during COVID, which is the development of a new computational tool and algorithm that allows us to repurpose drugs – that we already clinically use in a wide array of conditions – for use in developmental disorders.

Hopefully, to combat them and as a lot of us know these disorders, by and large, don’t have effective therapy. So there is a significant need to treat them.

This work was primarily done at the Welcome Trust Sanger Institute in Cambridge.

By the end of this 4th LinkAGE webinar, you should be able to describe the Open Targets platform and its utility.

You should also be able to explain what orphan drugs are.

You should be able to appreciate the basis of the Deciphering Developmental Disorders (or as I’ll refer to it, DDD) project.

And the DDG2P database, so the developmental disorders genotype to phenotype database.

You should also be able to understand the process, of developing the tool, that I undertook. And how it is used to identify drugs for potential clinical use in developmental disorders.

And finally, you should also be able to discuss the benefits and drawbacks of having such a tool in the first place.

So target validation is primarily defined as the process of demonstrating the functional role of the identified target in the disease phenotype. So to put it in perspective from a genetics point of view or a genomics point of view, it’s to identify a mutated form of a protein or a protein target that we can target either using drugs or other genetic modification techniques.

So first and foremost, let’s talk about Open Targets.

So Open Targets is a public platform that is a compendium of protein targets.

Open Targets is a publicly available platform (to a degree) that’s maintained by multiple teams – primarily the EMBL-EBI team, also based at the Sanger or in Hinxton next to the Sanger – and the whole idea of Open Targets is that it is a huge skill partnership that uses human genetics and genomics data to allow systematic drug identification, and drug target identification, and allows prioritisation.

So when I earlier talked about target validation, this is a really effective platform to help us with target validation.

The Open Targets platform integrates public domain data. So lots of, for example trial evidence, to enable target identification and prioritization.

The genetics portal within Open Targets identifies targets based on both Genomic Wide Association Studies as well as functional genomic studies.

Is it a really effective method of aiding with target validation. It is really key to the developmental my algorithm.

The next part is orphan drugs. Now orphan drugs are drugs that are intended to treat diseases that are so rare that sponsors are reluctant to develop them under usual marketing conditions because the R&D costs just can’t be recuperated by making drugs for these diseases. Put in perspective, the process of identifying the new drug molecule is hugely long and hugely expensive.

The slide says tens of millions of Euros, but actually sometimes it’s hundreds of millions. And it’s also very uncertain because less than 10% of agents are actually effective and make it through phase 4 trials.

Developing the drug intended to treat a rare disease, therefore, becomes very tricky if you use this approach and it’s better to repurpose drugs rather than to identify new drugs.

And that’s another key part of my algorithm, and we’ll explore that a bit more in later slides.

Orphan drugs can be defined as drugs that are not developed by the pharmaceutical industry for economic reasons which respond to a public health need. And actually the indications of a drug, they may also be considered an orphan since the substance used may be used in the treatment of a frequent disease but also actually may target rare disease. And this is exactly what I was talking about earlier and exactly why it’s of interest to me with this algorithm.

Now to think about the Deciphering Developmental Disorders project, the aim of the DDD study – and this is taken from their website – is to “advance clinical genetic practice for children with developmental disorders by the systematic application of the latest microarray and sequencing methods while addressing the new ethical challenges raised by such methods and by genomics.”

The genomic data from DDD is stored within the DECIPHER database, and that’s also important because this tool is ultimately going to be available through the DECIPHER database.

All of this work is of undertaken at the Sanger Institute, with the Hurles lab.

And a sub-part of DDD is the Developmental Disorders Genotype to Phenotype database (DDG2P), which is a curated list of genes reported to be associated with developmental disorders.

It has both the genes, the type of mutation, whether or not it’s mono or biallelic, and the direction of action. And all of this is extremely important because it’s relevant to the next part of this presentation.

Move on to the actual algorithm. So the idea and the basis of this whole project, is that we’re approaching the ear of individualized therapeutics.

Indeed, this is the 4th LinkAGE talk on gene-targeted therapies and all of these are individualized therapeutics, to a degree.

And there is an evolving international drive to use existing pharmacological agents in a broader range of conditions and not just as orphan drugs but also in other common diseases. So not just rare disease.

So the idea we had was why should we not cross-interrogate clinically employed agents with their known pharmacological targets, by which I mean the proteins that these drugs act on, and compare that direction of action to the mutation that resulted in that protein having a disease state.

I’m sorry if that’s slightly wordy. I will explain it a bit more in subsequent slides.

In a bit more detail, the whole idea of this project, and my ACF, was to accelerate drug discovery for developmental disorders by (a) determining all antagonistic drugs mutation pairs.

What I mean by this is that, for example, if you have an inhibitor and an activating mutation, the inhibitor should work on the activating mutation. And indeed that doesn’t always hold true but that was the purpose of this computational project.

And the second part (B) of this project was to identify those drugs that already have a orphan status. And the reason this is important is because they’re more likely to be granted subsequent orphan statuses if they’ve already been assigned one, by whichever medical regulatory authority is present in that area. For us, it would be the MHRA and NICE. All of these candidate drugs, so all these drug mutation pairs, also went to proof of bioavailability test, which was partially manual.

And the importance of this is by seeing that if these drugs are already used in for example, epilepsy one can therefore infer that these drugs would have activity on the central nervous system. So if the disease that I’m investigating, the developmental disorder in question affects the central nervous system, one might infer that this drug is bioavailable there.

And this is of course also an important step, otherwise we can throw drugs at various parts of the body, but if they don’t work, there isn’t any point using them.

This is my methods – I’m sorry it’s quite wordy – but the main take home messages from this slide are that we are just over 2,500 mutation disease pairs from the DDG2P database, and putting them through Open Targets found that 323 had drugs that interacted with them for a total of just over 2,500 drug mutation pairs.

The reason this number is so high is because a lot of diseases had zero. Indeed, only 323 out of the 2,589 had drugs that interacted with them, but the range of drugs that interact with any one disease can range from 0-48 drugs.

That means that number becomes very large, very quickly.

And then these were then manually curated, as I previously described, to determine their bioavailability at the site in question, to see whether or not these drug target pairs were actually likely to work at the target site in question.

This is a quick explanation of the number of diseases that we were targeting, so as I’ve said, there was just over 2,500 from the DDG2P database.

323 had disease with drug targets.

Of these with novel targets, so ones that we’d newly identified there were some D2 conditions in question.

And I’m sorry the writing hasn’t come up great on this slide, but of the total drug target pairs for those 323 conditions, what we found was that 560 of that just over 2,600 had an opposing direction of action to the mutation.

But what you have to remember is the whole basis of the algorithm, and what was really promising to us was that 70 of these drugs were already used clinically, so that means that these acted as a essentially a positive control for our algorithm and we identified 484 novel candidate drug target pairs.

This is a really big deal to us because it means that only about just under 20% these sort of targets have effective therapies and potentially we have a lot more therapeutic agents we could use in these diseases.

Just to give you a broader review, this is a sort of summary slide of the different types of agents of that 484 sort of novel drug and target pairs we have. And as you can see the vast, vast majority of these are essentially inhibitors, as one would expect.

The second main category is blocker. This is my first result slide. Don’t worry this is nowhere near the end of the presentation.

As I said, we had 484 novel drug target pairs, and of those 76 of the 560 we found essentially acted as positive controls telling us that this pipeline works to a degree.

And then you have to remember, we were also interested in whether or not these drugs had orphan statuses. And we found that 109 of these 484 agents had an orphan status anywhere in the world, by putting them through Orphanet.

And this is important as will become more obvious later in the presentation.

But just a quick slide on limitations at this stage.

The first, and I’m afraid this is going to get very genetic, is biallelic loss of function variants are very unlikely to be salvaged by activating drugs or by stabilizing drugs or positive-modulating drugs.

Notable exception to this is that in KCNJ11-based familial hyperinsulinemia, diazoxide is an orphan target drug that works extremely well. But this is sort of the exception to the rule.

Also the DDG2P database, so the primary source of data for my whole project, isn’t complete yet. It’s not perfect. And as a consequence it means that quite a lot of …

or not quite a lot, but a reasonable number of mutations’ direction of action still remains to be ascertained.

It is also really difficult to suggest biological effect without actual testing and disease models, be these cell models, be these animal models or indeed humans themselves.

Therefore, we decided we had to narrow down our approach.

We therefore checked activating mutations specifically, and particularly those in clinically severe disorders, ones that cause severe morbidity or severe mortality as these are the ones that are likely to have most salvage or also be approved for potential limited trial use through orphan designation.

So that’s what we did.

So we found that 44 diseases were due to activating mutations from 31 different genes; One gene can cause multiple different disorders. Diseases were then sorted by clinical severity, and taking into account things like life expectancy or the degree of morbidity that results from these disorders.

And we found that of these 44 disorders, 376 pharmacological agents could be used to target them through the algorithm and had been through at least phase 1 trials, internationally.

These were the breakdown of the different drug trial phases of the agents that I’ve mentioned on the previous slide, those 376 agents for using these 44 diseases. And what’s really important is to look at the phase 3 and phase 4 agents because they’re the ones that have been trailed in disorders. They’re the ones that have passed things like safety testing, and we know a lot more data about them, and they’re the ones that are actually used in diseases.

And as you can see, they make up the majority of these drugs that we were investigating.

Now, we found 11 severe phenotypic disorders and 14 moderate ones, from activating mutations. And to a degree, we also took into account current treatments that are available for moderate disorders.

As one may say that Crouzon craniosynostosis is actually very severe, but with surgery, its morbidity is drastically reduced. As such, we classified it as moderate.

I’ve talked about the other ways we looked at severity.

So these are the severe disorders. They range from things like microcephalic primordial dwarfism to well-known tumour syndrome, multiple endocrine neoplasia type IIB, and of course things like hyperkalemic periodic paralysis.

The results when we looked at these disorders is that there were twenty-six phase 4 inhibiting, blocking or negative modulating drugs, that had been granted phase 4 trials, and that had an orphan status within these 11 disorders.

There are also two phase 3 orphan drugs for these 11 disorders.

There are also 61 other drugs that have been through phase 4 trials but didn’t yet have an orphan status and were therefore of less interest to us, because they were less likely to be quickly granted orphan status.

Of the 28 drugs in question, so that is these phase 4 and phase 3 agents with the previous orphan designation, eight had at least one published study on PubMed. – the condition that we were examining them in question – leaving 20 novel agent and phenotype matches. Again, those eight almost acted like a positive control for our pipeline.

Those 20 were the ones that we got interested in.

To give you an example, Penttinen-type premature aging syndrome is caused by monoallelic activating mutations of the gene PDGFRB. We found three phase 4 orphan drugs that are inhibitors (nintedanib, pazopanib and regorafenib) and one that is a phase 4 antagonist (glasdebig).

One drug, imatinib, which was also a phase 4 orphan inhibitor, had been used in one month old already with some success. And that spurred us on to see whether this was possible in other disorders.

So we then looked at hyperkalaemic periodic paralysis type 1.

This is due to monoallelic activating mutations of the gene SCN4A.

There’s one known agent already in use, mexiletine, which is a phase 4 orphan drug for potassium aggravated myotonia and works well in a lot of patients.

Again, acts as a positive control for this algorithm.

There are also four other phase 4 or orphan agents that could be used in situations where mexiletine doesn’t work.

These include well-known anti-epileptic agents, such as carbamazepine, and also other drugs, such as riluzole, rufinamide and lamotrigine, well-known mood-stabilising drugs.

And all of these are blockers.

There are also 32 other non-orphan phase 4 drugs that could be investigated further down the line if these were not likely to work.

Then it gets a bit more interesting, I have to say.

So we then started looking at our moderate severity disorders, and these range from things like a subtype of Noonan syndrome through to Curry-Jones syndrome, and as I previously stated, Crouzon syndrome.

The results for the moderate disorders were that there were thirty-nine phase 4 inhibiting, blocking or negative modulating drugs with an orphan status with the 14 activating mutation-based moderate diseases.

One phase 3 orphan drug for the 14 moderate disorders as well, that had an orphan status.

And also 63 other drugs that didn’t have an orphan status, but had been through phase 4.

So we were primarily then interested in these forty phase 3 and phase 4 drugs that had already been given an orphan status. And interestingly, none of them had entries in PubMed when we started.

That’s a very important statement for the latter part of this presentation.

So we then started looking at CLOVES.

CLOVE syndrome is congenital lipomatous overgrowth with vascular malformations and epidermal naevi syndrome. It’s associated with mosaic activating mutations of PIK3CA. We found two novel phase 4 orphan inhibitors, copanlisib and alpelisib.

And the reason I put alpelisib in bold is I presented this presentation and alongside my presentation when I first started talking about this algorithm, a group had started investigating alpelisib’s use and used our algorithm as further evidence that it should work. And what they found is that it drastically reduced the size of the lipomatous overgrowth seen within CLOVEs. And this was really promising to us as it acted as another positive control, essentially, for our pipeline. And this is the reference for this study.

I’ll start talking about the limitations of this approach because I think it’s important for us to think about many of these disorders, particularly in the moderate category, are due to problems that happen in utero. And that’s important because if we give the drug postnatally or a problem that starts antenatally, it’s unlikely to have a significant effect. Particularly if it, the effect antenatally is that it causes things like dysmorphology, or indeed any organs to develop abnormally.

And as I previously mentioned, it is also really difficult to infer biological effect based on a computational algorithm without actually directly studying it, either in model organisms or indeed within patients.

And the final challenge that we are coming across is that it’s really difficult, simply based on a computational algorithm, to assign further orphan status to a drug.

This is something that there are, have been discussions with the MHRA about and will be revisited on a single drug basis as more and more people hopefully use this algorithm and tool.

What does the future hold for us?

As the DDG2P database becomes more robust with better mutation, annotation and drug mapping, hopefully our results will increase the numbers of potential diseases and drug mutation pairs will increase, which means we potentially will have more treatments for developmental disorders.

Also, if you cast your mind back a few slides, we decide to primarily look at activating mutations. But actually, negative mutations, by which I mean loss-of-function mutations, also have several agents which could potentially prove to be of some benefit and result in some salvage in these conditions. We just haven’t examined them in this arena yet.

The results of this algorithm, and actually the tool itself, is now being embedded into the DECIPHER database, which I’m sure a lot of people here already use.

And when you’re searching, either by patient or by gene, if there is a drug target potentially within that disorder, or the protein in question, DECIPHER should now flag it up or will soon flag it up.

Our hope, and my hope, is that this will accelerate research into the treatment of rare diseases and allow us to repurpose a lot more drugs and give us a lot more orphan drugs to combat these diseases in the hopefully not-too-distant future.

Just going to acknowledge a few of my collaborators and my supervisors.

My direct supervisors, professor Matthew Hurles and Doctor Helen Firth at the Hurles Group at the Sanger Institute, oversaw a lot of this project. But it couldn’t, it wouldn’t have been possible without help from Ian Dunham and David Ochoa at the EMBL-EBI Open Targets Group. I’d also like to acknowledge Orphanet and then the National Institute of Health Research for awarding me in ACF in the first place. And to the Genomics Education Programme for allowing me to present my work in this forum.

Now, hopefully at this point, you should now be able to describe the Open Targets platform and its utility. Explain what orphan drugs are. You should be able to appreciate the basis of the DDD project and the DDG2P database. You should be able to understand the process that I undertook to develop the algorithm I have mentioned, and its potential clinical use in the fight against developmental disorders. And you should also be able to discuss both the benefits and limitations of the tool that I have developed.

And that’s it. That’s the end of my presentation. We’ll now move on to the Q&A. Any questions?

Lianne Gompertz: Thank you Hassan for sharing that wonderful piece of work for us. What an amazing, amazing project and thank you so much for sharing that with us.

So I’m going to kickstart the Q&A session if that’s okay with you.

I have a question here. So, at present, many clinical geneticists do not prescribe medications on a regular basis. Do you anticipate that with the incorporation of your algorithm into the DECIPHER database, that this will become more within the realms of a clinical geneticist role? Or is the intention to highlight potential treatments for further evaluation through research?

Hassan Shakeel: So thank you for that question. And it’s extremely important actually.

It’s identified a core problem with the project and one I was trying to eloquently put together earlier.

So yes and no.

I think it’s going to be a very, very long time before a clinical geneticist is going to look in somewhere like DECIPHER and come up with, this is the treatment you need to give, simply based on what DECIPHER says.

I think it this will allow us, hopefully, to research more drugs. But these, when I say research, I don’t simply mean sort of model testing. I think this tool, essentially with drugs that are already known orphan drugs in other conditions, this tool should help allow us to essentially do limited trials within the conditions in question.

The ones, one ones to investigate.

And that is why we came up with that.

Lianne Gompertz: If a clinician were to look at the DECIPHER database and they came across a potential treatment for their patient with a rare disease, what would be the process if they wish to use that medication? Are there any channels that they need to access to try to get their patients involved with these medications?

Hassan Shakeel: So again, a very important question and one I didn’t delve into too much, but is actually very relevant.

So, for some of these conditions, the drugs in question are essentially off target and are very safe.

An example I’ll give you is nifedipine. Um, so up been a lot of sort of the, um, sort of tools that, or a lot of the diseases we looked at, um, and other agents like amlodipine. Now these are blood pressure agents – they have other effects, they have cardiogenic effects – but it could definitely be investigated within safe monitoring.

I think what you’re getting at though is some of the rare drugs in question here.

To use those within a limited clinical trial, one would have to write to the MHRA and propose it as an orphan agent. And that is a sort of stumbling block that we came across is simply based on a computational output, you can’t say that this drug is necessarily going to work. Indeed it could be dangerous until we actually trial it.

What I think is encouraging is a lot of the drugs we already use in these conditions, or have been trialed in these conditions, were thrown up through this pipeline.

So the idea being that hopefully more will be. And especially on my last couple of slides, that’s what we found. There are people are already starting to use it to identify drugs that work in these conditions. So watch this space. Hopefully it’ll work better.

Lianne Gompertz: Brilliant. Something to look forward to, I hope.

We’ve just had a question through on our chat. Um, apologies if I missed this, but can you please clarify the importance of a drug having orphan status? And indeed, exactly what you mean by that? I would’ve thought that repurposing commonly used drugs for novel indications would be useful. I think I’ve missed something here.

Hassan Shakeel: Yeah, so the importance of a drug having orphan status is actually something decided by different medicines regulatory bodies across the world.

So Japan does it differently to how we do it in the UK to how the Americans do it. But all of them collaborate with something called Orphanet.

For example, the MHRA and the FDA all put in data into Orphanet to say what drugs they have approved for orphan use and which conditions. It’s not just enough to say “X is an orphan drug”, for example, imatinib – which people here might know better as Glivec, which is used a lot in CML.

It only is an orphan drug in CML and a couple of other conditions.

It doesn’t mean you can just start using it willy-nilly in any rare disorder.

But what the reason we looked at orphan agents specifically was once a drug has an orphan designation anywhere in the world, it’s actually much, much, much easier to get a limited trial use. IE, another orphan designation for a different disorder.

And that means hopefully it can accelerate research within this field, and accelerate research within these conditions, if you identify a potential target that is already an orphan.

So that’s what I meant by orphan drugs.

And the second part of the question was …

“I would’ve thought that repurposing commonly used drugs for novel indications would be useful.”

So it absolutely is.

But I think we have to do that within a safe space, so to speak.

We can’t just go using these drugs when we don’t fully know what biological effect they’re going to have in that condition, or indeed the systemic effect they’re going to have.

Which is why, even though this tool will allow us to identify ones that may work, we still need to essentially cross the t’s and dot the i’s and trial them – and publish our trials because then hopefully we’ll get a much better compendium over time.

So with alpelisib, for example, if you now look at CLOVES within DECIPHER, or patient with CLOVES, for example, it brings up alpelisib and says that actually it brings up the trial as well and says it works or it should work.

Doesn’t mean it’s always going to work, this is what we actually want the tool to become.

Essentially, almost an open platform for people to report these trials.

And it, you know, makes us feel better that we develop the tool in the first place.

Lianne Gompertz: And speaking of publications actually, so the next question that’s come through the chat is can you point us to publish papers that summarize your work, Hassan?

Hassan Shakeel: Yeah, so, the big paper that we’re going to be …

that we are in the process of publishing will hopefully, hopefully – it’s currently undergoing a third round of reviewer comments in nature genetics. So watch the space. Hopefully it’ll be available soon.

But the CLOVES paper that I talked about, and indeed a couple of the others that are relevant to this, will be posted essentially when this goes live.

So when this is viewable on YouTube, you’ll be able to look at the CLOVES reference I referenced earlier and within the sort of comment section, I’ll also drop in a couple more papers that we’ve had since.

But our big one, we’re still undergoing reviewer commentary.

This work has been published in other fields as well as in, I’ve published it through the Genomics of Rare Diseases Conference, which was held at the Sanger.

They published it as an abstract, but of course, that’s not a detailed account of what the tool is.

Lianne Gompertz: Wonderful. Thank you.

So hopefully that’s something else that we can look forward to in the near future, to to read more about this algorithm that you created and how you’ve done that.

Brilliant.

So I think we’ll draw today’s webinar to a close in that case.

So again, thank you all for attending today’s webinar.

Thank you again to Hassan for sharing that wonderful piece of work for us. And certainly food for thought for the future.

Just one final comment to all of the attendees here. Please, please, once again, do complete that feedback form. I believe the link may now be in the chat to complete that feedback form.

The next webinar is in three weeks time, on the 11th of May, Thursday 11th May. It’s going to be presented by paediatric oncologist and Sanger Institute researcher Doctor Sam Behjati, and he’s going to be talking about targeted chemotherapy agents in children with cancer.

And so we hope to see you there.

Many thanks to everyone and lovely evening.

Take care.