Theresa LaVallee joined the Parker Institute for Cancer Immunotherapy in 2017 and is currently VP of translational medicine and regulatory affairs. As part of the institute’s efforts to bring precise and powerful immunotherapies to cancer patients, LaVallee and colleagues are on the front lines of both drug and biomarker development for a new class of therapies that are challenging the models of precision medicine established with the emergence of genomically targeted therapies.
Immunotherapy response depends on both tumor and immune factors, however, presenting new challenges in the development of tools to predict response and personalize treatment. Researchers and clinicians have recognized that existing biomarkers like PD-L1, microsatellite instability, and tumor mutational burden predict only some of the variability seen in patients’ responses. In turn, this is driving researchers to seek out new methods or combinations of methods to capture the complex interaction between a tumor and the human immune system in a way that can more effectively guide treatment decisions.
LaVallee recently spoke with Precision Oncology News to discuss barriers the field faces and the broad landscape of strategies being pursued to overcome them. Below is an edited transcript of the interview.
To ease in, could you talk a little about what you view as the biggest biomarker challenges immuno-oncology is grappling with right now? In our coverage of this area, it certainly seems like there is some consensus that there are multiple markers, but their imperfection means we’re going to have to combine them in some way. Would you agree with that?
I think I would take a step back on that. Having been involved in biomarker development since it really became in vogue, [we have to remember that] it was hard for people to really even understand a predictive biomarker ten years ago. But it worked well because it was pretty simplistic. You find tumor drivers based on genetic alterations … You match a drug to that signalling pathway, it cuts it off, and the tumor shrinks. So that’s fairly straightforward. But I say that almost tongue in cheek because there are a multitude of examples that haven’t worked. Finding “the” driver has been a challenge. [But if you can make it work in the targeted therapy space], it’s tried and true.
Now we have to think about why we’re so excited about immunotherapy. Immunotherapy is turning on the immune system to attack the cancer, but the immune system is not a single signaling pathway and not a single cell. You have a multi-cell system, with T cell, B cells, myeloid cells … and you have the tumor itself. So you have to serve a multi-parameter multifactorial system.
I like to point out to people that in the tumor driver space it’s only been in the last couple of years that we had a next-gen sequencing panel for biomarkers in lung cancer, which seems like the obvious thing to do, right? Instead of doing five discrete tests and having to have enough sample to do that, you just have one test that tests five markers. But that was really cutting edge when it happened. And now [with immunotherapy, we need] a multi-omic, multi parameter test, potentially with different technologies. We [may] need to look at immunohistochemistry, and genetics, and RNA, and a flow assay, so getting people to have the patience to test each of those and pull it together is what we are facing.
Encouragingly, I think that there are examples where this is showing some light of day, [for example] the Merck paper in Science late last year using TMB and gene expression profiling. They not only took a multi-omic biomarker approach, but they also looked at large data sets: both clinical data that they had in-house across multiple studies, and TCGA. I invited two of our bioinformatics folks to help with an opinion piece that we wrote on that paper earlier this year [which] touches on many of these points.
Breaking down some of the factors that may end up being part of this multifactorial approach, what can you tell me about where we are right now with tumor mutational burden and microsatellite instability?
MSI and TMB are different [in] where they are in their usefulness. MSI is clinically validated and has shown clinical utility. It has a label and patients are getting benefit from its use. TMB is something that has shown that it has merit for informing who to treat, and there are a number of positive studies showing clinical utility where it can really help inform patient care, but it hasn’t hit that same mark yet, although we are seeing that it informs for better outcomes.
Even just being able to define who to treat with a PD-1 single agent versus going into combination treatment would be of huge value [if TMB can work either alone or combined with another marker]. People are getting long-term clinical benefit on what is a pretty safe therapy, and if there are patients in non-labeled indications that have this mutational profile — we all have the anecdotes of the sarcoma patient here, the ovarian patient there that responded — so like MSI if this could be used for patients where there aren’t available therapies, that would be of great value.
But again, I think the Merck approach where they’re taking both TMB plus gene expression profiling [has promise]. TMB is looking at the ability to present neoantigens to the immune system. The gene expression profiling that they’re using is looking at immune readiness … So that’s an example where characterizing both the tumor and immune system, perhaps the two together [create] an algorithm that — although it may not have the same positive predictive value that a mutation driver does — can inform better who [to treat].
There was also a paper that recently came out from a group at Johns Hopkins University where they compared TMB, multiplex IHC, and gene expression profiling, and their take doing this retrospective analysis was also that [multi-modality biomarker strategies] … are the most informative.
Has Parker been involved at all in any of the assay harmonization efforts? That’s mostly focused on TMB, but we have heard from some people in the field that there could be concerns even with MSI where different assays may not categorize patients as high- or low- in the same way.
Friends of Cancer Research is leading this, but it absolutely fits in the Parker model of [collaboration], and we have participated in their TMB harmonization, gone to many of the meetings, and had discussions about how we could help with what we do. They’re working with the FDA on this as well.
One important thing that they are doing is trying to … come up with a reference standard. So, if you did FoundationOne versus MSK-Impact, is there a reference standard that would allow you to harmonize the two assays, and that work is progressing very well. I think we learned from the PD-L1 issues with having multiple tests and multiple drugs that led to confusion for doctors and patients, [and we’re] really trying to get out ahead of it. Pharma and diagnostic companies have been really terrific about collaborating on that, and that work will be a very important piece coming forward.
Another interesting thing we’ve been hearing about is the microbiome. There was some data highlighted at ESMO, for example, on how microbiome composition, or even particular microbial populations appear to influence Immunotherapy efficacy. My understanding is that groups are looking at using probiotics to try to prime the immune system to respond better to cancer … but I also know that Parker has presented data on how probiotics may be interfering with the action of these drugs. This stuff is early days, but any main takeaways or heads up for the oncology community from what you guys have been tracking or hearing about in this area?
We are actively working with many of our investigators in the microbiome space. Jennifer Wargo had the AACR presentation on probiotics, and we have investigators at Stanford and Memorial Sloan Kettering who have a wealth of research in the microbiome. What we’ve mainly done is worked with them to translate some of the preclinical findings to the clinic, and we are currently running a treatment study with microbiome agents and PD-1 inhibitors in a melanoma clinical study.
The findings that have been well published from investigators in the Parker community and outside is … that there is an observational finding that the microbiome profile and [a patient’s] immune fitness have a strong correlation. So if you have an unfavorable microbiome or a non-diverse microbiome, if you take antibiotics for instance, your immune “tone” is not as good. And there are a lot of studies that show antibiotics use with immunotherapy have unfavorable outcomes.
With probiotics, [the issue is] there’s a variety of them. And how they’re manufactured varies since they’re not regulated, so you’re not always exactly sure what you’re getting. And Jen Wargo’s work did show that that was a detriment. It’s looking like anything that you do to affect your microbiome and decrease your immune fitness … [could reduce efficacy].
What we’re trying to do now is look at whether we can alter the microbiome, whether with a live biotherapeutic product or FMT [fecal microbiota transplantation], to change cancer patients’ microbiome to have that good diversity and a favorable signature, and see if that correlates with a better immune tone and response to PD-1 drugs.
So that’s the hope, that you can affect the microbiome in a way that makes people more sensitive. But probably over-the-counter probiotics are not the way to do that?
Well from the biomarkers space, where we started this conversation, the question is can you use biomarkers to decide who to treat and how to treat? So looking at the microbiome as a biomarker, whether it is favorable or unfavorable, the idea is that you could get to a point where you can use that to say anyone that has a favorable microbiome should get PD-1. That could be part of the multi-omic, multi parameter analysis. I call it the immunogram, which comes from the literature, [to envision] an algorithm that asks does the patient have neoantigens? Are they TMB high? Do they have an interferon signature? Do they have T cells? Do they have a good microbiome, and in doing that, you could say this patient is perfectly poised for a PD-1 inhibitor. The devil’s in the details and it all needs to be worked out much more extensively. Then the next question is can you use microbiome as an intervention?
It sounds like you guys are pushing forward with research on both fronts. Hopefully we’ll have more of a readout on that soon?
Absolutely. We have a microbiome collection in all of our clinical studies now, and the more I learn about it the more the data are very impressive and compelling.
Are there any more experimental biomarkers that researchers at the Parker Institute are considering that go beyond/don’t fit into some of the above mainstays?
There are some other things that have gotten some attention but not a tremendous amount. First, poop is good because it’s minimally invasive … and so a stool-based biomarker [as in the microbiome] is relatively simple. But blood based biomarkers are also simple. So I think technologies … where they’re doing machine learning and circulating tumor DNA analysis, again looking at multiple parameters for both immune and tumor fitness, could be very powerful.
And then the other one that’s minimally invasive is imaging-based approaches like what ImaginAb is doing, where they have a CD-8 PET tracer. Most people in the field will tell you a “hot” tumor is characterized by the presence of T cells in the tumor. And that’s one of the things [some of these other markers may be a surrogate of]. TMB will bring T cells in because there’s a lot of neo antigens, and the same thing with the interferon signatures. It’s just telling you there’s a lot of T cells there.
What I love about this imaging approach, and ImaginAb just recently had their Phase I data published, is it has a really good signal to noise ratio and the ability to look at the whole person. We all like to think a biopsy is representative of the disease, but it’s only a single poke at a single lesion when the person can have multiple lesions, dozens, and there may be tumor heterogeneity across the person. This ability to image the whole person without having to take any samples, to infuse the radio tracer and then get a quantitative read of how many T cells are in the tumors, I think will prove to be very useful.
I’m glad you brought up circulating tumor DNA as well, because there is another aspect of that that we have been tracking. Because of the way these drugs work, when they do work they seem to really work, so we have reported on some groups having success with doing circulating tumor DNA monitoring very early in treatment. Ideally, we’d have a pre treatment biomarker that works well, and from what you describe it sounds like we’re going to get there through some combination of things. But does it make sense in the clinic that we might also start to see a different paradigm where you put people on immunotherapy treatment first and then use a liquid biopsy technology to see if they’re responding?
That is one of my goals, and I think that actually might be simpler. A lot of people have a harder time thinking that way, but given the complexities of the multiple things that have to be right in the immune system and in the tumour to activate it, looking at biomarkers of response may be the simpler approach. But I’m hopeful that the machine learning type approaches actually might do both. They may be able to look at baseline, and then after treatment.
The burden of proof is going to have to be high, though. For doctors to take patients off of immunotherapy, the positive predictive value of the failure is going to have to be high.
Finally, it would be interesting to get thoughts on the cost side of things. The trend these days seems to be toward immuno-oncology drug combinations, coupled with moving toward combination biomarker strategies, all of this implying potentially increasing costs on both the drug and diagnostic side of things. How does this not become unmanageable?
Our mission is to bring immunotherapy to patients with urgency. And if biomarkers, even a complex combination, help inform that, [then you can see them having value.] But it goes to what I was saying earlier, that whatever we do has to prove that it’s useful and that it has utility. We know that just treating everyone is not a great way to do it, so we’ve put a lot of work here into trying to get the information to really inform patients treatment: who should get PD-1 single agent? Who should get CTLA-4 for PD-1 combinations, which we know has better activity, but also more toxicity?
And then we need to define, in addition to the the immunogram, the resistogram: so who shouldn’t be treated with PD-1. That’s where things like the ImagineAb technology may be great because any patients that don’t have any T cells in their tumor, or a very low T cell burden, they are going to need some sort of combination treatment to get the T cells in and to have the checkpoint inhibitors work.
We’re just gonna have to continue to do the work to figure out the best way to have utility.