On this episode of Translating Proteomics, co-hosts Parag Mallick and Andreas Huhmer of Nautilus Biotechnology discuss how clinical researchers can leverage proteomics for drug development. Some of the themes covered in this episode include:

·      Proteomics and pre-clinical models

·      How proteomics can drive patient selection

·      Choosing the right end points in clinical trials

Chapters

00:00 – 01:06 – Introduction

01:06 – 06:51 – Proteomics in pre-clinical studies

06:51 – 11:40 – The importance of choosing the right model for preclinical work

11:40 – 17:10 – How proteomics is used in Phase I/II clinical trials

17:10 – 19:29 – Proteomics tools in patient selection

19:29 – 24:33 – Useful information that we get from proteomics that we can’t get from genomics or transcriptomics

24:33 – 28:14 – Proteomics in Phase III clinical trials and picking the best indications of drug efficacy

28:14 – 29:19 - Understanding why clinical trials fail

29:19 – End - Outro

Resources

• The National Cancer Institute’s webpage covering how clinical trials work
• Preprint on the miBrain model - An example of a new in vitro brain model
• Geary et al., 2021 - Discovery and Evaluation of Protein Biomarkers as a Signature of Wellness in Late-Stage Cancer Patients in Early Phase Clinical Trials
• Wasko et al., 2024 - Tumour-selective activity of RAS-GTP inhibition in pancreatic cancer
• The NCI-MATCH trail: Lessons for precision oncology – Report on a large-scale trial using genomic biomarkers to match cancer patients to treatments
• Tumour-selective activity of RAS-GTP inhibition in pancreatic cancer – Study using proteomics to understand mechanisms of resistance to a cancer drug

On this episode of Translating Proteomics, co hosts Parag Malik and Andreas Humer of Nautilus Biotechnology discuss how clinical researchers can leverage proteomics for drug development.

Some of the themes covered in this episode, proteomics and preclinical models, how proteomics can drive patient selection and choosing the right endpoints in clinical trials.

Now, here are your hosts, Parag Malik and Andreas Humer.

Welcome back to Translating Proteomics.

In this episode, we're going to dive into how proteomics can impact drug development.

In a previous episode, we focused on how proteomics can be used to discover new drug targets.

Today, we're going to focus much more on how proteomics can improve the development of preclinical models, how it can be used to recruit patients, and also how we can help drug developers measure efficacy more effectively.

Well, let's dive in and chat about starting at the first phase of drug development, preclinical.

Now, again, we've talked a lot about how proteomics can be used for identifying targets in the first place.

But in the preclinical stage, we have to think about, we have to think about the efficacy of the drug, we have to think about its downstream consequences and we particularly have to try and understand why is it that this particular drug is working in this system and this model and not that one.

What have you seen that has really been the highlight for you of the proteomics used in preclinical research?

So maybe we start with where we've been and maybe then we go where we will go in the next part of the conversation.

But the first, I would say the last 10 years was really very transformative, particularly for proteomics, as a lot of tools have been developed to understand mechanism of action, but also look at off target effects, meaning when you study a drug and then you're trying to understand, does it bind to the target of intended target or does it also do something else?

A lot of work has been done because we can obviously measure protein and protein interactions in a much more discrete way in a much larger scale.

That has been possible with typical traditional Western blot type techniques and immunoassays.

So I think that's a very big step forward.

I would also say that the second phase of that is now emerging where people are using proteomics to study much more complex cellular models or even animal models and then I think require another level of tool set to really understand what's going on when you apply a drug to a particular organism.

Yeah, I think for me there are really two key areas of research that I've seen that have been very exciting.

The first is going beyond the genetically determined response markers and saying, all right, I have a T790M mutation in EGFR and so therefore my kinase inhibitor isn't going to bind or isn't going to work as effectively.

But really what happens beyond that mutation?

How does cetuximab overcome things that Gefitinib can't do and tracing that path to say, all right, well in some cases we aren't even effectively hitting the protein target, in other cases we're hitting the protein target.

But there's compensation from downstream pathways.

So things like in the RAS pathway or the MEK pathway compensating, we see things like CMET overexpression compensating for egfr.

So that's the first layer is just is the drug even able to function?

Is it able to do its job?

The second layer to that is what else is it hitting?

And I think this is what you're talking about.

Most drugs fail not because of lack of efficacy, but because of off target, cross reactive toxicity concerns.

And so maybe you are hitting the target that you intended, that you thought you intended, but the consequence of hitting that target is that three hops down the line you're causing some reaction that's really dangerous and overlaps with a liver pathway or something like that.

Correct.

And chemical biology in that sense has made big contributions in that sense, because as you study much larger number of proteins, at the same time you can actually see does something else happen somewhere in the cellular system or the tissue that I didn't tend to target?

Secondary and tertiary effects are much easier picked up because we have the ability to, you know, look at much number of proteins and basically use the discovery power of modern proteomics to discover these side effects.

And I think historically a lot of these studies were done using transcriptomic approaches, but from the transcriptomic data you didn't see that a protein was overexpressed and moved to the membrane or moved to the nucleus.

You also couldn't tell about protein protein interactions for instance, or changes in those.

And I think when we think about these mechanism of action studies, we really do need to take advantage of am I causing proteomic rearrangements?

Are those proteomic rearrangements significant or are they just little responses?

What other pathways are being impacted?

Are we Hitting cell death pathways, cell growth pathways, communication pathways, and really trying to dive into any given perturbation that we make in a cell.

The obvious target may not actually be where we want to make the change.

It may be that when we hit the top of the cascade, we just have so many downstream consequences that we're headed straight towards toxicity.

And instead, if we can go three or four hops down in the network, that will allow us to get the more specific change that we want to overcome.

A mitogen, for instance.

Yeah, clearly progress in that space.

But I wonder often whether this form of drug discovery is the most efficient way of discovering new drugs.

And I wonder whether we need to take a very different direction in the future.

I think, for example, the whole field of spatial omics, I think, is adding a very interesting aspect to how drugs are distributed.

Maybe there are in hitting the targets they're supposed to be, but some of them essentially get out of the particular tissue and then cause problem in other areas.

Well, and I think one other area that we need to really consider is what model system are we using for what?

Historically, a lot of preclinical studies were done in 2D cell culture, typically monoculture.

So one type of cells, particularly in cancer research or cardiac research, where we would have just cancer cells in a dish, what we've seen over time is we've seen an increase in the complexity.

We've understood that there are interactions between the epithelial cells, the endothelial cells, other stromal cells that can drive the responsiveness of a system.

We've seen that secretions from one cell type can influence the metastatic potential of others.

And so ratcheting up our model systems from monoculture systems on a plastic dish to even monoculture systems in a more representative matrix.

So a Jello, for instance, that has a slightly different elastic modulus can already change the behavior of the cells to ultimately sets of cells grown as co cultures.

And then we now are seeing a tremendous advance in these organoid model systems that have many, many different cell types and actual 3D structure to them.

Yeah, I agree with you.

And it's kind of fascinating when you look how these organoids are essentially created.

In some cases, we don't do anything else, but taking different cell types and bringing them close to each other so they can communicate.

And just through the communication interaction between those cell types, they actually start creating, you know, an organoid of a specific tissue, for example, little brains.

And so I think in this case, the opportunity is to actually take individual cell types, characterize them you know, you may have a mutation that you're interested in asking very specific questions, how it contributes to a disease.

You characterize them on a DNA level, you characterize them on a proteomic level, and then you bring them together with the rest of the tissue.

And then you basically have, you know, an engineered system that allows you now to observe things that you couldn't observe in the context of a truly complex biological sample.

And I think there's a really important role for proteomics to play in this, because one can use the proteomics studies to examine what is the cellular response to perturbation X, how does that change when the cell is alone, when the cell is on plastic, when the cell is on a matrigel or in an organoid, when it has its friends and neighbors around?

And how do those change over time as well?

We've seen some beginning work with proteomics of these different model systems to assess and validate.

Is this a better model system for this particular question?

The only concern I have when I look at the literature is that I still think we do not have enough throughput, scalability and sensitivity.

I think when you have this level of exquisite biological interactions and there's billions of interactions in these models, right, we're still stuck at the level of did some of the protein go up and down, did some of the proteins emerge or disappear?

But we still don't have the scale and the capability to really understand why, when these cells are communicating, do they form an organoid?

Or in the case of a dysfunctional cell line?

Why is that dysfunctional cell line now creating a problem within this, let's say, organoid?

Well, there was a study that our lab did where we were looking specifically at different types of lung cancer cells.

Some that were more aggressive, some that were more benign, some that were drug resistant, some that were.

And what we found is that in 2D culture, the aggressive and benign cells just sort of intermixed completely randomly.

But then when you put them into 3D culture, the more aggressive cells, they were much happier with the aggressive microenvironments, the hypoxic microenvironment in the center.

And so even without a drug, you would get an enrichment of aggressive cells in a harsh microenvironment, forming essentially an aggressive niche.

And that auto assembly occurred just by microenvironment.

When you looked at the proteome, you saw huge changes in expression depending upon where you were in that organoid.

And so it was a really clear case to us where the 2D culture while it was helping us understand in general molecular pathways, it was completely neglecting the impact from the microenvironment.

So we discussed the preclinical side of drug development and maybe we think a little bit about the clinical side of drug development and how proteomics can potentially have an impact there.

I know a lot of customers that I've spoken to often use very sophisticated proteomics tools in the pre clinical space, but then fade them out in favor of more simplistic systems like immunoassays.

What's your experience?

I think just thinking about that next phase.

So phase one, phase two clinical trials.

You're getting past your mouse model, you're getting into humans, and your questions really are about safety, toxicity and dose.

So how much can I give somebody?

What is the minimum effective dose, the maximum tolerated dose?

And on the safety side, on the side effects side, many of these are gross symptoms like are you having a tummy ache?

Are you?

And so I've seen a lot of hesitation to using more sophisticated tools as part of those trials.

What I've seen instead is that people have been collecting samples along the way so that they can do retrospective analyses to understand the causations of toxicity, dose, response, et cetera.

But I agree with you that I haven't seen people writing explicitly into the prospective part of their trial.

We're going to do broad scale proteomics specifically for predicting a response or understanding a response and making a treatment decision on the basis of it.

And I think that hesitation is because the a priori wants to be able to have as simple a correlated variable as possible.

If they're going to write it into the trial and they're going to make a decision based on it, they have to be able to say, oh, okay, if it's this, I'm going to do this.

If it's that, I'm going to do that.

And so I actually think the challenge here is in part from the measurement technologies and the cost, ease of use, reproducibility of them.

But I think it's also on the analysis side of being able to know exactly what you're going to be looking at and being able to predict a priori what you think is going to be able to happen.

Yeah.

What I have seen though, that some drug developers developing a drug, let's say for kidney disease, and then they find out that during the clinical trial that some people actually develop a favorable profile, glucose profile of their blood and then they're really curious, maybe this is the next best diabetes treatment.

And so they come back and say, hey, can someone take all these plasma samples that we have from the clinical trial and then maybe do some preclinical study again on finding out why did all of these patients suddenly show favorable blood profiles in terms of sugar?

So, but again, that's a retrospective analysis, so that's not used as part of the prospective design of the trial.

That's not used as part of how we select patients.

It's used post facto to understand variables.

But do you think it's the problem of the powerful proteomics tools that they basically disclose information that may be difficult to understand?

Or in the end, when you're trying to prove a drug to explain why, did you see these changes, which may be completely innocuous, but there's still changes that can be observed.

Do you think that's one of the major hurdles to use more complex tools?

I think so, because you are required to report.

So if you write a measurement into a study as something that you're looking at as a primary or even a secondary endpoint, then you're required to report out on all of what it finds.

And if you don't know what that is, that can actually potentially get in the way of your FDA approval process.

And I actually think that's probably wrong.

And this is something we should change in the FDA approval process is allowing some more permissiveness around secondary variables that can potentially be part of prospective use.

But maybe in a trial setting.

But today that's very challenging.

Where I have certainly seen it is in places where we are repurposing therapeutics.

So we already have a long history.

So this is less in phase one, phase two, this is something that's already made it through.

I think in the NCI match study, they did use large scale sequencing.

So we've seen examples like that where large scale data has been used for driving patients towards trials and saying, oh, you should be on this or you should be not.

So it's used as part of patient selection.

So it's actually not explicitly written into the case report form, but it comes up front to say, oh, I think you'd be a good match for this trial.

Yeah.

And certainly having a better patient stratification relies on many, many more data points than the ones you're actually targeting.

And certainly I hope that the application of proteomics tools will make clinical trials more efficient and ultimately even go to the point where we have personalized medicine.

Yeah, I think this aspect of who should even be on the trial, we want folks who have sufficient fitness and there's been some work.

Erie et al, 20, 21 made sure that a patient would likely be able to survive through the trial period.

That's a very narrow indication for some of these broad scale screening or proteomic biomarker associated things.

But ultimately it may be that there are correlated variables that we're not looking at that right now.

We would say, oh, they're a smoker.

And so we would either rule them in or rule them out.

Maybe we want to be able to go deeper and say, what's their inflammatory load?

Let's look at in their circulation.

What are we seeing in terms of this set of proteins that are associated with liver damage and being able to get a little bit more acuity around the state of the people coming into the trial beyond the typical measures.

But also what you're arguing for here is that if you had done in your preclinical studies or used modern proteomics tools with a broad perspective and already learned that there is maybe side effects that affect a particular organ, let's call it a kidney, right.

That when you actually recruit patients, you make sure that there's nobody with a kidney issue because, you know, that person will ultimately be counter, you know, counteracting the clinical trial results, right?

That's right.

So a simple example of this is liver cirrhosis.

So if I'm looking for, if I'm doing a hepatocellular carcinoma trial, I, I will often ask people, oh, do you drink alcohol?

But it would be great if I could instead.

And people will look at liver enzymes and things like that in a standard cbc.

But if I could go deeper and understand the extent of the damage and I could understand if it's present or not, what their history is, the proteome capturing that lengthy history, I might be able to do a better job of identifying patients and even just subtyping them to know, okay, this is what this drug does in an instance where the, the patients have significant minimal medial evidence of liver damage or toxicity.

So is there a particular example where proteomics tools are exquisitely used to understand what drives a disease and you are unable to get that same information from, let's say, a DNA or RNA level?

Yeah, I think actually there's really elegant work from Paul Michel's lab.

He's now at Stanford along with some of the Tim Clausey and others at UCLA.

Looking at EGFR V3 in a variety of conditions, it's often expected that overexpression of EGFR will imply responsiveness to EGFR targeted therapies.

However, if, for instance, you have a mutation or a heterozygous mutation, the transcript is unable to resolve that.

And then even often protein staining can be complicated because you need a protein antibody that targets just the insert region and is sensitive enough to tolerate having a heterozygous population.

It's even more interesting in glioblastoma because it's been shown, again, great work by Paul Michel's group, that you can have a extra chromosomal segment that is transferred from cell to cell to cell.

So it's not actually even in the genome, it's just leading to expression of a protein in a non within the genome manner.

And it can be in a small subset of cells or it can explode and be in many cells.

And the proteomic analysis of that helps us look at what is the actual expression of the v3 form that these therapies are going after.

So it's really a critical example where the therapeutic is completely ineffective without having the right form of the protein present.

And you can't find the form of the protein by looking at anything other than the protein.

So does it also mean that maybe some of the work we do right now, the overemphasis on application of genomic tools in that space maybe is the wrong thing to do?

I need to be a little careful here and not say that no one should do genomics ever.

No, that's not what I'm saying.

That's not what you're saying.

Okay.

All right.

I think the reality is that different processes are regulated at different layers.

And so oftentimes we've been doing a lot of genomic analysis because it's accessible and we've seen individual mutations, we've seen changes in chromosomal duplications, et cetera.

Copy number variation has had associations with phenotypes.

Now where that's broken down is again, in heterozygous populations, this can break down.

It can also break down when there are other processes that are driven at the proteomic scale that are simply not measurable.

So changes in degradation pathways, for instance, those are basically invisible to the genome and transcriptome.

And yet HIF1 alpha hypoxia, inducible factor, known factor in hypoxic response, known factor in driving aggressivity in cancer.

Very hard to study any other way than by looking at the protein because the transcript doesn't obey the behavior of if I experience a hypoxic stress response.

So I do think that there is tremendous value in having the genome data, but I think we just have to be really cognizant of the fact that it's incomplete, it doesn't fully capture the dynamics of what's happening in the cell.

And even if we take it over time, we're seeing evolution in the genome, but we're not necessarily getting the complexity of the cell states or the cell state composition of a tumor, which are hugely dynamic.

So I mean, the takeaway would be not to solely rely on DNA or RNA tools, although they're so cheap and accessible that probably you would do the measurement anyway.

But then don't stop there and continue to measure things on the proteomic level to make sure that you haven't overlooked particular mechanism.

Yeah, I think ultimately what you'd like to get to the point is where we understand where the control points for different processes live and then maybe we don't need to measure quite everything, but we need to measure those control points.

We're just not at that place right now.

So it's incumbent upon us to measure as much as possible so that we can learn where those control points are.

So if you had to guess how many diseases or how many mechanisms are dependent on genomic regulation versus proteomic level regulation, Any wild guess?

Well, I don't think we need to make a wild guess because we generally have a sense of, of which inborn errors lead to downstream consequences.

Genetically associated diseases that are one to one and don't have environmental factors.

And I think it's something like 2% of diseases fall into that category.

Yeah.

So it's very little, it's very low.

Yeah.

So far we focused a lot on preclinical studies and phase one, phase two.

One of the things that comes up when we start thinking about phase three is we're really talking about efficacy and we're talking about efficacy relative to standard of care.

There it becomes particularly important to understand are we looking at the relevant thing?

And we have examples where we aren't looking at the relevant thing.

Proteins themselves are not always the relevant thing In Alzheimer's disease.

Amyloid plaques are clearly a component of the disease.

However, there was a series of studies, clinical trial results, Sims et al, Vandyke et al, that they were looking at the prevalence of A beta plaques and or the disappearance of A beta plaques on different drugs, or different drugs impact on them.

And it wasn't super predictive of either neurological burden or of effectiveness of the therapy in improving the the state of the disease.

So while proteins are clearly important and drivers, this question of what should we look at understanding where we should look, when we should look, and what we should look should be our ultimate goal.

Yeah.

I mean, in the case of Alzheimer's, it's a very complex network of proteins that ultimately lead to the cognitive decline.

And so, yeah, it was part of the big revelation the last several years where biomarkers were discovered that actually are tracing, tracking with the progression of the disease and ultimately the cognitive decline.

And there was certainly a lot of work needed and several proteins had to be sort of discarded as good biomarkers.

For example, neurofilaments themselves are not good biomarkers because they don't correlate with the disease, but they're certainly changing a lot with neurodegeneration and similar other proteins as well.

Interesting, though, is that if you think through, for example, in Alzheimer's disease, finding the right protein is definitely a challenge because in some patients you actually may have mutation in some of the proteins involved and certainly then they become the focus of a drug target, whereas in others, where there's more sporadic disease, in this case, you probably don't have to look at particular proteins.

Right.

Yeah.

Well, I think part of it is understanding there are aspects that may be mechanistic and drivers early on in the disease and then later in the disease, they may not be as predictive.

So one of the questions we have when designing a trial is what are our primary endpoints, what are our secondary endpoints, and what are things that we're collecting along the way?

And we've seen examples of therapeutics that have been approved based on primary endpoints that didn't actually improve patient welfare.

And so this is a place where I think we have to be a little bit careful, is making sure that, again, we're measuring.

It may not always be possible to measure cognitive decline.

It's a very narrow.

It's hard to measure.

So having a molecular surrogate is really helpful.

But are we measuring the right molecular surrogate?

And in retrospect, many clinical trials probably would have given better results if the right endpoint would have been picked from the start.

Right.

And that's certainly a very difficult task.

Yeah.

I think the other thing that we really have to be aware of is when clinical trials fail, either phase one, phase two, phase three, are we doing enough to bring that data together?

So many drugs we know fail because of toxicity, some fail because of lack of efficacy.

But when they fail, what's done with that data?

Both all the data that was collected to that point and the data from that specific trial, can we use our proteomics analysis to go back and say, oh, you know, people who in general have failed, they failed because of this panel of cardiac enzymes going up in the blood.

Yeah.

And so maybe there's a big opportunity in the future to use modern AI tools to sort of mine for clues that we have overlooked and find maybe there is indication that or maybe you can learn something from these trials that actually were unsuccessful.

Thanks so much for joining us today on Translating Proteomics.

Today we talked about how proteomics is impacting all throughout the drug development process, and we particularly talked about its impact on preclinical models, how it can help us recruit patients, and how proteomics in general can help drug developers measure efficacy and learn from failures.

We'd love to hear from you if you have stories or experiences where you've used proteomics as part of your drug development efforts, or you have questions about ways that proteomics might be part of the process, or you have frustrations said, wow, my proteomics was great here and terrible there.

We'd love to hear from you.

Hit us back in the comments.

We hope you enjoyed the Translating Proteomics podcast brought to you by Nautilus Biotechnology.

To contact us or for further information, please email Translating Proteomics at Nautilus Bio.