George Netto, MD (Host): Welcome back, and thank you for joining again. Today's episode will focus on an exciting new study that was just published in Nature Communication. The study discusses potential of a new biomarker that may predict chemotherapeutic response in breast cancer patients where their tumor is ER positive, HER2 negative.

The lead investigator in the study is Dr. Zak Kinsella. He's a post-doctoral member of the Royal College of Surgeons in Ireland. And I'm happy to have him here. Thank you for accepting my invitation, Zak.

Zak Kinsella, PhD: Thank you for having me. I'm excited to be on to go through the paper and discuss some of the key findings.

Host: I'm eager to see your take on it. So as we usually start, give our audience a little background.

Zak Kinsella, PhD: ER-positive, HER2-negative breast cancer is the most common, occupying about 70% of diagnoses. And regularly, the determination of risk of those patients helps to decide who really needs therapy and what type of therapies they need. So, it's really crucial that it's kind of assessed properly, accurately, and as quickly as possible.

And there's some factors that help to do it, you know, patient age, other histological factors like their tumor grade, their tumor subtype. But also, more recently, things like genomic variables, which is where like the Oncotype DX test comes in. And it's been very useful, especially on the back of validation from the TAILORx trial, showing that patients in the low recurrence score, so those between 0 and around 11, up 16 or up to 15, are shown not to benefit from chemotherapy and only to benefit from hormone therapies. Whereas those in the higher end of the recurrence score above 26 really are shown to benefit materially from chemotherapy.

It's been extremely helpful. From the back of the TAILORx trial, it's shown there's a potential reduction in chemotherapy use in, I think, up to about 70% percent of patients, which is enormous. Very beneficial to people, obviously. But there is a bit of a gray area. And the gray area is that intermediate risk between the low and the high. The problem there is that they showed that there is a slight benefit of chemotherapy for premenopausal patients. I think it's at the upper end of the intermediates, I think around 5% to 7% benefit. Whereas there was no real benefit of chemotherapy in the postmenopausal.

The issue there is they're essentially using menopausal status as a stratifier of risk on top of genomic risk, which is quite ambiguous. Perimenopause exists for a long time, and it encapsulates a lot of different biological variables that are happening simultaneously.

So yeah, the background was we suspected, is there another way that can potentially stratify that intermediate risk out into different groups, beyond merely the genomic variables and the clinical variables that we have at the moment?

Host: Beautifully put. So, clearly, there must be another way. That's why you're here, and that's why your paper got accepted just this month, published in Nature Communications. So, we're very excited to hear about what you think could be a tool that help you stratify these patients. But before we do that, can you tell me a little bit about what is the current understanding of the immune microenvironment or immune status of this category of tumors? In the introduction, you said hot or not. That's all about if you're hot, you're going to respond to immunotherapy. If you're not, you're in trouble.

So, before this paper, was there any role for counting T cells or anything like that?

Zak Kinsella, PhD: There's been a lot of larger studies that have probably on the back of the success of looking at TILs and immune infiltrates in other subtypes like HER2 positive and triple-negative. There's been a lot of investigation into single markers, or things like TIL counting through pathologists' assessment, in ER-positive disease overall.

And a lot of the studies were in agreement. They basically showed that there was basically no association with outcome. Even single marker studies showed that there was barely any effect and barely any kind of difference in outcome of patients with ER positive HER2 negative disease. And there's a slight issue there, which also helped us with the background of the study, which is that the ER positive subtype more generally is incredibly heterogeneous.

You know, the ASCO guidelines have changed from 10% of ER positivity being inclusive of truly positive patients down to 1%. And they have different immune profiles. They have different TIL infiltrates. And even within different categories of ER positivity, there are different immune infiltrates.

Unlike triple-negative, which as far as I understand, has a more linear association with TIL infiltrate and outcome, ER positive isn't the same. There's a lot of heterogeneity, both in where they're located, whether they're myeloid, lymphoid, whether there are certain polarities or antitumorgenic versus protumorigenic. And that kind of confuses and dilutes any association whenever it's investigated as a whole. More recently, there's been quite a bit of investigation into this. Thankfully, I published something fairly at the start of the curve, but it seems like a lot of evidence is now mounting to suggest, one, it is indeed extremely heterogeneous, and two, there requires like a more multimodal approach that incorporates the clinical variables on top of genomic variables, on top of immune variables, and others like histopathology too. In a word, it's just extremely heterogeneous, and it was previously thought to be relatively cold.

George Netto, MD: You touched a little bit about that, the compartment, right? It's important probably to be more granular, use the intraepithelial or iTILs versus stromal S-TILs, which you did in this study. But it sounded like, among all these markers, only probably the B cells and the tertiary lymphoid structures, the TLS, seem to be the only one before this, thinking maybe there is a little correlation. But in general, like you stated, it wasn't something that seemed to be. And it's interesting how you frame it, that it's so heterogeneous and mixture of things that makes any single marker surface up as useful. So, what did you do? Which population and what kind of technology you used for this study?

Zak Kinsella, PhD: So, we were quite lucky in the sense that we have Irish patients that were previously enrolled into TAILORx. So, we have a population of around 600 patients, including all of their clinical data from the TAILORx trial that were recruited around hospital sites in the Republic. That's really useful obviously, because we have the actual trials oncotype scores, and we have crucially randomization of treatment in that intermediate group to see who really does and doesn't benefit from chemotherapy in the same way as it was done in the large TAILORx trial.

So, we had a really unique sample set. And based on the background and what we know from the field, although I'd probably debate a bit of this, but it seems like a lot of like the genomic clinical and debatably the estrogen or estrogenic kind of profile of the tumors has been taken into account and incorporated into scores that can classify risk and treatment response.

But one of the issues with oncotype, and I know they include a marker in their assay of CD68, but it leaves a massive axis out, and that's the immune axis. And again, it assumes much like the other studies that one marker can do the trick or there is a linearity between TIL infiltrate and outcome across the genomic risk spectrum. And we basically wanted to start by seeing is there a difference.

To do this, we firstly applied—well, we wanted to apply omics, particularly spatial omics, to begin to look at the immune microenvironment. So, we applied a six-plex panel of immunofluorescence. It's from Akoya, and I think they've since changed their name. I can't remember what their name changed to.

Host: No worries. Yeah. Now what are these markers? Because most of our audience are pathologists, so...

Zak Kinsella, PhD: Yeah. So, we have CD4 for T helper cells, FoxP3. So in combination with CD4, we can see T regs. We have CD8 for cytotoxic T cells, CD68 for macrophage, and CD20 for B cells. That's the immune component. And then, we have pancytokeratin for the epithelia and DAPI just for the nuclei.

Host: Excellent. And then, you did the spatial. But you also did spatial transcriptomic, right?

Zak Kinsella, PhD: We did, yeah. So, there we used NanoString's GeoMX technology. We had the whole transcriptome atlas, which basically you can kind of think of it like bulk transcriptomics, but spatially segregated from the peri-tumoral stroma and the actual epithelia. So, you essentially elute only the RNA that's associated with each of the compartments and sequence them separately. So yeah, you get an idea of what's going on transcriptomically in the stroma and then, again, the same in the tumor.

Host: Wow. And then, you integrate both the proteomic and the transcriptomic and try to find some subtypes, which we're going to discuss in a second. This is amazing, and that's why it's so precise and probably one of the strengths of the study that you had all the tissues, you had all the paraffin blocks on this population. And I noticed that we were discussing that before we came on air is how you use tissue microarrays. And that makes it financially feasible, right? Because each one of these transcriptomic slides can cost $5 to $10, I don't know at that time. Hopefully, the price continues to come down. But when you put tissue microarrays several spots, you're doing several cases in one shot and you're not losing much of information that way.

Zak Kinsella, PhD: Unfortunately, prohibitively expensive to do on full face or whole slide images. I hope it does to continue to come down in price, because I would love to do matched whole slide transcriptomics or spatial transcriptomics with proteomics. I think there's a lot valuable to be done there. But yeah, it's just too expensive at the moment.

Host: Wonderful. But even with that, you managed to get some exciting results. So, let's move to that. What were the main findings of the study after you did all these?

Zak Kinsella, PhD: Well, it's essentially that there are different immune states that are present, at least in our cohort. It seems as if there's a range of states which would be intuitive from actually immune cold through some intermediate states to actually immune hot. But those states did not linearly associated with outcome or response to treatment.

Secondly, that those subtypes are not significantly captured by oncotype. Like, I was saying earlier, oncotype fairly, it misses out a big axis of the biology, which is immune, and it seems like we can recapitulate that or the missing out of that information through our proteomics analysis.

So yeah, it does seem like there is an immune axis that's not captured by oncotype. And I'd say thirdly, that those immune types actually do help to stratify risk across the oncotype spectrum.

Host: Yeah, in all three categories of low and intermediate and high, adding that dimension of the immune state. And these were like there was three epithelial subtype and five stromal subtypes that you—

Zak Kinsella, PhD: Three epithelial and four TME. So, the epithelial is literally just low or cold, intermediate, and hot. The TME range from truly cold and truly hot on the upper end. And then, intermediate states in the middle where one had more T cells and the other didn't; had more Tregs, for example.

Host: But clearly, will require a lot thinking of how can I use this every day.

Zak Kinsella, PhD: Exactly.

Host: I'm sure you boiled it down to some potential biomarker that's more amenable to clinical utility. So, what was that?

Zak Kinsella, PhD: Yeah. I wanted to know—well, I actually found that these states did stratify risk essentially across the spectrum, and some of them actually did appear to be predictive of chemotherapy response. But like you said, these technologies aren't going to come to the clinic relatively soon. So, there does need to be an assay that can potentially be translated.

So, I did look along all of the different markers and their combinations to see which one could be predictive and prognostic. And what came out of that was CD8 positive T cells or cytotoxic T cells. There were others. And again, it plays into the understanding that nothing happens in isolation. It's always more of a local and community-based or an environment-based effect. But given the recent findings in immunotherapy trials, it does seem like having something that's linked to cytotoxic T cells will help to, maybe in the future, pass out whether patients need immunotherapies, for example, while also showing potentially those that may or may not benefit from chemotherapy.

So yeah, the one variable I did find that maintained strong significance and also had a predictive ability, particularly in that intermediate recurrence low group, were the cytotoxic T cells.

Host: And those were in specific compartment, right? The stromal T or both?

Zak Kinsella, PhD: Actually, there was an effect in both, but the stronger effect was found for the stromal T cells. That's probably due to the fact that there were simply more T cells there. There are more patients actually having stromal T cells versus Intratumoral T cells, and therefore probably allowed us to have more confidence around effect sizes. But yeah, it does seem like it's mostly stromal T lymphocytes.

George Netto, MD: Perfect. So, you're envisioning, of course—this needs validation because this is one cohort, the TAILORx that you mentioned, it's a significantly big cohort. Nevertheless, validation and reproducibility of—how do you think? So now, of course, they're going to continue to order Oncotype DX and maybe with the stromal T cells or TILs, cytotoxic TILs. You can add a little bit of information, another dimension for the immune axis like you were saying. So just counting it or by counting it in these compartments, is there any clues here? What would be a way to do it? I guess using CD8.

Zak Kinsella, PhD: Yeah. And I think one way that potentially could be really beneficial is to see if you actually do need that immunohistochemistry assay at all, and whether you can call this from H&E. So, one thing I am exploring that will be my future work, hopefully, is, essentially whether I can generalize or translate the results like through the foundational models to basically score similar variables on H&E and whether you even need to do that assay at all.

Host: Here we go. That seems to be the idea.

Zak Kinsella, PhD: That seems to be the idea.

Host: You don't even need that. You know, that seems to be the way a lot of things—I call these digital biomarkers. So, you just scan the H&E, apply a foundation model. I left out in the introduction, you're not only an expert and interested in spatial and omics, but also in AI and multimodal AI. For those who are watching us on YouTube, you'll figure out why. Very young and all these young postdocs know AI like we used to use a calculator in our times. So, I'm not surprised. And that would be amazing because, like you said, pathology is an underestimated field, but our time is coming back.

H&E is important and topography is very important. So, our expertise and, hopefully, the young pathologists with the AI expertise will do what you're doing. I won't be surprised that that works, you know. So, is that your next paper?

Zak Kinsella, PhD: Hopefully. I mean, ideally I'd like to show this effect in an external cohort or in the larger TAILORx trial. It does require a lot of work to make sure the models are generalizable, both to capture the inherent biology that's present from immunohistochemistry, but also whether it can generalize and is robust to things like medical center differences, scanning differences, differences in the processing pipelines that actually produce the blocks themselves.

So, there's a lot of work to see if anything can be produced, and basically cut through that noise. And whether after doing all of that, it does reproduce the results seen in immunohistochemistry. But there's some preliminary evidence to suggest that that might be the case. So yeah, hopefully, it'll probably require a good few years. But hopefully, that'll be my next paper.

Host: Wonderful. We'll look forward to that and bring you back for sure to discuss.

Zak Kinsella, PhD: Oh, I'd love to. That will be brilliant.

George Netto, MD: I was just going to add the reason also, which is a good news, that this will be the way our pathologists will be encouraging clinicians to use such technology. Because at least on our side, in the US, the FDA and also the reimbursement agencies, the insurance agents, because it's all about the money at the end, are recognizing these algorithm now as a service.

So, a new category of software as medical device. So after you do all your due diligence and show that your foundation model working on H&E and all the pre-analytic and explainability and generalizability, like you said, that is robust, you get it as an FDA approval for that algorithm. And then, on the payer side, software as a service, SaaS, is just a new category where you will be able to build, the path lab will be able to build for that.

So, anytime that last portion, you know, the compliance and reimbursement field is progressing toward that, that will encourage more and more people to develop these tests and, most importantly, use them for our patient in a very effective manner, so...

Zak Kinsella, PhD: It's always interesting to see Nature Medicine publications, Nature Cancer publications, even Nature publications quite recently showing agents and things like that in the medical sphere, just how quickly things are moving, and to think that this is probably the least accurate and least helpful they'll ever be.

So yeah, it's good that people are being included and they're actually being used to support pathologists in doing the crucial work that they do. It will be interesting to see how that goes in the future.

Host: Great. Yeah. I mean, accuracy and all that is very important. So, there is a reason why we need to adhere to compliance and federal approvals and on all these agencies. They can be a pain. But in the same time, they're the right thing to do. Wonderful.

Zak, this has been very enjoyable and educational, and congratulations on the paper. And really inspiring to see postdocs like you doing such amazing studies and speak so eloquently about their findings and the sky's the limit. Thank you very much.

Zak Kinsella, PhD: Thank you again for having me. Hopefully, I can keep the momentum going.

Host: You will. You will. All right. Thank you.