Multiplex IHC as biomarkers for cancer immunotherapy

Share :
Published: 5 May 2016
Views: 2652
Dr Paul Tumeh - UCLA Medical Center, Los Angeles, USA

Dr Paul Tumeh speaks with ecancertv at AACR 2016 about determining potential biomarkers using multiplexed immunohistochemistry.

Having created a spatially resolved model of the tumour microenvironment in all its heterogeneity, with different cell types identified by coloured labels, Dr Tumeh describes visible tracking of a tumour responding to immunotherapy.

As such, he explains how multiple biomarkers are a necessary course for increased drug efficacy, and that a negative assay response to one biomarker need not be a prognostic limitation.


AACR 2016

Multiplex IHC as biomarkers for cancer immunotherapy

Dr Paul Tumeh - UCLA Medical Center, Los Angeles, USA

I was asked by Matt Hellman from Memorial Sloan Kettering to give a talk about multiplexed immunohistochemistry and its role or value towards identifying biomarkers, predictive biomarkers, to immunotherapeutic agents. So right now with the shift moving from targeting oncogenes to targeting native proteins those native proteins are expressed by different cell types in the tumour microenvironment and can be in different cellular microenvironments. The problem with that is that because they’re native when you homogenise a sample you don’t know what the cellular sources are. So what multiplex IHC does is it introduces spatial resolution into the tumour microenvironment in order to understand where are these proteins being expressed and by which cell types and in what kind of niche are they in the tumour microenvironment.

How do you sort targets?

The big challenge is that with oncogenes, with cancer cells, you can identify genes that are mutated that you know are derived from a cancer cell. But with immunotherapy and targeting proteins that are native the issue is that the cell types, the immune cell types, they lack mutations. There’s a wide variety of immune cell types and they’re highly plastic which means that they can change their function just based on what area of the tumour microenvironment they’re in. So how do we capture that? What we do is we interrogate a tumour slide, or a tumour, with multiple antibodies that have different colours. Then once we do that we introduce imaging analysis, and what we’ve done is we’ve introduced machine based learning, to capture on the single pixel resolution level what all these different antibodies, where they’re localised in the tumour. So what it essentially does is it generates a map and that map, we bring meaning to that map through imaging analysis. So we can understand, for example, if there are certain immune cell types that were present before getting immunotherapy and then looking at how those cell types evolve, how they’ve changed during immunotherapy. Then we correlate it with treatment outcome, for example, whether patients responded to therapy or patients progressed.

Have you been working with different biomarkers?

We started off with pembrolizumab which is anti-PD-1 and so that’s a monoclonal antibody that targets the PD-1 receptor, the programmed death 1 receptor, on T-cells and natural killer cells. What we found, we published a paper in the end of 2014, and we found that if you had pre-existing T-cell immunity in the tumour, meaning that tumour was recognised by CD8 T-cells in the tumour, when you introduced anti-PD1 therapy into those tumours those patients were more likely to respond to anti-PD-1 therapy. So what we’ve found since then is that there are many other cell types in the tumour that are interacting with those CD8 T-cells as well as interacting with the cancer cells that play a large role and impact whether that tumour is going to respond or not respond to anti-PD-1 therapy.

Tell me about the results you’ve seen so far?

Right now the large focus has been on PD-L1 and what we’re finding is that, considering the complexity of the immune system, how many immune cell types you have, the fact that they’re plastic, they lack mutations and that cancer cells are plastic and you have a lot of stromal cells, that a single marker to accurately capture the complexity and dynamic state of the tumour microenvironment is actually quite challenging. Not only that, but the way of identifying signal in that tumour microenvironment is actually quite challenging. So I think where the field is going to go is that we’re going to be capturing multiple data points about the tumour microenvironment in order to capture the function of that tumour.

How do you think a poor diagnosis could affect the patient’s outcome?

Psychologically speaking it has the ability to really negatively impact patients and ultimately our goal is to serve our patients in the best way possible. Now, when they hear that, for example, let’s say you have a patient that has non-small cell lung cancer, squamous type, and they’re PD-L1 negative according to the assay, which has its drawbacks alone, that can certainly negatively impact the patient and the family. The positive, the twist to this, is that the number of combinations that are coming through the pipeline, I think there’s a lot of hope that, despite them being negative for one marker where they can’t get one specific drug, it’s just a matter of time and in short time we’ll be able to identify combinations for those patients that they can go and enrol into in terms of a phase I clinical trial. So always remain hopeful and positive.

How has machine learning affected your trials?

I think it’s early. When you take a look at who is leading the world of machine learning, artificial intelligence and deep learning, it’s the larger tech companies – Facebook, Google, Elon Musk’s group. The big question is can we leverage the open source code that has been provided by all these companies to integrate massive amounts of large data in order to identify signatures or biological information that we otherwise could not. It hasn’t impacted clinical trial design yet but my expectation is that it will as it goes from a proof of concept to actually becoming a mainstay integrated way of identifying or capturing information that’s important that impacts the way that clinical trials are designed. We’re just not there yet, I think it’s too early.

What’s next for your research?

From my perspective the two big goals, three to five year plan, is to do two things. One, to identify new proteins or targets that are either not emphasised today, undervalued or just not known, that upon either blocking it or stimulating it actually really changes the tumour in terms of its immunogenicity. So drug target discovery is one area, the other is getting more personal with personalised medicine and that’s integrating a model that captures information from the patient’s peripheral blood as well as background clinical history, clinical variables, as well as the tumour microenvironment, integrating them into a predictive model that really accurately predicts the patient’s response to any given intervention. So it’s going to be predictive biomarkers and drug target discovery.

What is your take home message?

The ultimate goal here is to serve patients in innovative ways that change the quality of their life in terms of dealing with the major challenge. We’re in a very innovative, rapidly changing time in immuno-oncology so it’s a wonderful opportunity to serve patients and, with the integration of these new technologies, whether it’s looking at multiplexed IHC or spatially resolved mapping of the tumour microenvironment and machine based learning and integrating neoantigen densities, mutation, I think in the next five years we’re going to see incredible improvements and milestone achievements that are going to change the field of immune-oncology.