Clinical Utility of Molecular Biomarkers Presentation - Johann de Bono

September 3, 2019

Professor de Bono presents the clinical utility of predictive molecular biomarkers in advanced prostate cancer during the first session of the Advanced Prostate Cancer Consensus Conference (APCCC 2019).  


Johann de Bono, MB ChB FRCP MSc Ph.D. FMedSci.  Professor Johann de Bono is Regius Professor of Cancer Research and a Professor in Experimental Cancer Medicine at The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust.

He is also the Director of the Drug Development Unit, overseeing the conduct of phase I trials, with a particular interest in innovative trial designs, circulating biomarkers and prostate cancer. Additionally, he leads the Prostate Cancer Targeted Therapy Group and the Cancer Biomarkers laboratory team.

Read the Full Video Transcript

Johann de Bono: Good morning. Life is getting more interesting. Prostate cancer care is getting more complex. And the devil is in the detail, as you've heard already from Colin. I want to say that all biases in my interpretations are mine and I do have conflicts of interest.

My talk is divided into three sections. This is my introduction. I would love to talk about a lot of different types of biomarkers, but I'm going to focus specifically on predictive biomarkers. But, remember, that biomarkers can be multipurpose. So, a biomarker may not just be predictive, it may also be prognostic. And the detail matters, because actually, it may make the interpretation quite difficult.

With all biomarkers, we've heard from Colin, there are false positives, there are false negatives. I like these statements, "All that glitters is not gold." The "Absence of evidence is not evidence of absence." A PSMA PET showing no lesion in the liver does not mean there are no liver mets. The lower limit of quantitation, in its definition, is critically important in our understanding, as you've heard already from Colin regarding blood assays. Binary variables are suboptimal. AR-V7 positive may mean one CDC is positive out of a thousand. If 999 are negative, does that matter? Yes, it does.

So, we really have to look at detail. Detail matters, particularly, invalidation and qualification. Somebody said to me recently, "There are a thousand papers on PSMA IHC." My reply was, "How many of them have shown validation of the assay in the paper?" We should reject papers that don't have validation of the assay for the biomarker in question that convinces us, because it misleads the community, and this is really key. Clinical qualification matters too, and importantly, we need prospective bespoke trials that are increasingly difficult sometimes to carry out for fiscal and other reasons. But we have to design trials that can address questions, and give us biomarker answers. And very often, qualifying predictive biomarkers requires drugs that work in that subset to predict response.

Predictive biomarkers are not predicting survival. That's prognostic biomarkers, and that is very important. As I said, biomarkers can be multipurpose, and it can be particularly difficult to disentangle prognostic from predictive. And this has been a major issue. For example, with AR-based blood assays, are they prognostic or predictive? Please read our recent letter to JCO on this that was published recently. As already Colin mentioned, we may need multiple orthogonal assays for the same test. Are we satisfied with one result? For example, a Lynch Syndrome patients came to my clinic with a germline mutation with a normal IHC and his tumor for all the four proteins. Now, you can get a mutation of the gene that is functionally relevant without loss of protein and normal IHC. The devil is in the detail. ATM has been broadly discussed by many groups in the last year for PARP inhibition. Don't throw the baby out with the bathwater.

It is really critically important that we look at the detail as single nuclear aberration of ATM, in a blood assay or a tissue assay, does not mean ATM loss of function. This is complex. Cancer care's becoming more complex, and maybe we need multiple orthogonal assays for ATM. And I believe we do based on our data at the Royal Marsden. The other multiplicity issue is that, in one tumor you may have multiple biomarkers. Let's say a tumor has an MMR defect and an ATM defect or an MMR defect and a BRCA2 defect, what's the hierarchy? What do you choose? How do you classify that? Well, the detail is key. Maybe the MMR has caused a second hit on ATM or BRCA2 and the latter are subclonal, and by looking at the detail of the genomics, the aloe frequency, the actual number of mutations in the sample that can be elucidated and this is really quite important.

We also need to keep everything in perspective, and this has already been really mentioned by Colin to some extent, but this is particularly an issue not only for mutations but also for copy number calls. Actually, I think it's even harder for copy number calls because you have to deal with tumor purity stromal contamination, ploidy, particularly in prostate cancer where we often have triploid prostate cancer, and when you're looking particularly in blood and plasma, it is very, very difficult, although more feasible that with low-pass whole-genome, which we're now doing quite extensively on plasma. And we have some very exciting data there on taxane sensitivity biomarkers, but I can't talk about that today. But the other critical thing is an SNA or a mutation or a deletion does not necessarily mean loss of function.

Hair loss may not mean loss of function, but actually, losing one allele may be haploinsufficient. And this has been shown for P10 and RB1 in many papers, somewhat may be controversial sometimes, but there may be a gene dose effect. And this is hard to lose by sequencing and may require FISH. So, we are not a one-trick pony for biomarkers. It's not just about NGS. We have to think about orthogonal assays beyond sequencing. And remember, there may be complexity with regards to homodimerization of proteins like ATM and maybe dominant-negative effects where you have one inactivating mutation where the proteins make them bind together with ATM that knocks out the wild type and then activates it. And this is something to think about. So the devil's in the detail.

Well, what about predictive biomarkers today that will impact care today? Well, actually, I think in the US, you could probably argue that mismatch repair is here today, but I think in, probably, the rest of the world, we still don't have level one evidence that has led to these drugs becoming standard of care. But I think molecular stratification for advanced prostate cancer is coming rapidly, is here in our clinics today, certainly in my clinic, is impacting my patient care, particularly with giving carboplatin to the BRCA-like cancers. And I've seen really impressive responses. I had a patient recently who came to me who was just, performance was three to four in a wheelchair, came to me from Leicester. Oncologist refused to give him anything else. He had BRCA2. I said, "I beg you, give him one shot of Carbo AUC2. See how he gets on." Performance that is won within three weeks of Carbo AUC2. I spoke to him last week. He's now had four cycles. He's now a performance that is much better. His pain's gone away. This drug works in these patients, and so the PARP inhibition, which is probably better tolerated and maybe a better option. But that's a different question.

But what about AR biomarkers? Well, I think the bottom line is that it's complicated, and I don't really have enough time to go through all the papers that many groups have published. But I think that the truth is, that in my mind, we still need definitive studies to use AR biomarkers in the clinic to guide therapy. I think one of the big issues is that the prognostic rPFS always data with these biomarkers is not predictive. We need predictive data that actually predict response, but there is no doubt that some of these assays mean that, at some point, the cancer acquires resistance mechanisms that made it remain AR driven. But remember the sequence things like archeology. It may mean that, at some point, the cancer had that. It's like BRCA2, but it doesn't mean that today it's still AR driven. Although, we can look at AR driven transcripts and elucidate that by looking at the transcriptome. So, this is complicated.

What about PI3K/AKT and PTEN? Well, I hope this is coming. Our recent paper last year with the AKT inhibitor ipatasertib, from genetic growth, is encouraging, but we've been here before. It's randomized Phase II studies that have failed in Phase III. So, who am I to say that Phase III is going to be positive? But I hope that maybe in the future we'll have drugs targeting this pathway, and then PTEN loss and aberrations of this pathway will help us select patients for treatments. Certainly, AKT mutations be PI3K mutations sensitized to AKT blockade in Phase I and Phase II studies, all published data. Can we use it today in our clinics? I don't think so. But there is increasing evidence that these aberrations evolve at progression on hormonal therapy, and we're seeing that. That these aberrations associated with poor prognosis, worse outcome, and poor response or poorer response to the next generation hormonal agents.

Well, what about transformation biomarkers? And I love this picture. I want to argue that we should start using the word transformed prostate cancer, and I'd love to hear what Felix and others think about this, Felix Feng, that is. We have used many different statements about luminal-to-basal transition. We've talked about RB lost, neuroendocrine, but I think, actually, what we're seeing is prostate cancer behaving like a chameleon and switching lineage. And I think the word transform may be the best description of this as the cancer tries to find ways to evade therapy. And we've published recently, this has worked for my former pathologist Daniel Nava Rodrigues, who used to work with us, showing that actually this transformation can be present at diagnosis, certainly at a FISH level.

It's commonly subclonal, and actually, it increases significantly this RB loss with castration resistance and can, but not always, result in the basal cytokeratin expression and the neuroendocrine phenotype, and often, but not always, results in PSMA negative disease. And this is from our paper. I don't really have time to go through this in detail, but the key issue is that yellow is heterogeneity. Figure E, bottom right, is actually the neuroendocrine cancers that almost all, but not all, almost all have RB loss. But actually, if you look at the adenocarcinomas in figure D, yellow is heterogeneous, red is positive, blue is negative, you will see there's a little prostate cancer that has heterogeneous loss. The bottom line here though, is actually the neuroendocrine change with RB loss may result in loss of BRCA2 with the RB, which is actually on the same locus or a gene called RNASEH2B that we published on Nature last year with Daniel Durocher and Andrew Jackson.

And actually, this is known to sensitize PARP inhibition and probably platinum. So, this switch may sensitize the platinum. So, the killer subclones in these patients are evolving may be platinum-sensitive. So, keep that in mind, and this will require future trials to prove. You've already heard a lot from Colin about DNA repair defects. Please, if you can, come to ESMO, the Olaparib Phase III trial will be at ESMO. As will be the hugely exciting, I think, CARD data on cabazitaxel which I think could alter how we think about the standard of care and drug sequencing. These are two important Phase III trials at ESMO, but I think you've heard already that mismatch repair is something we cannot afford to miss. And therefore, I would argue that we may have to think about orthogonal assays. Now, Colin and his team, Stephen Salipante, have really done beautiful work on the MSI NGS analysis that was used really as the only assay in the Merck trial, MSD.

I can tell you the Merck Sharp & Dohme pembrolizumab trial that I call the Emmanuel Antonarakis, patients who we know by IHC had loss of protein were deemed MSI NGS stable, not MSI high. I think the MSI NGS cutoffs are a problem with that assay, and as Colin actually alluded to, really have an issue. The Promega MSI PCR assay does not work well on formalin-fixed, paraffin-embedded tissue. These tumors don't often have high infiltrating lymphocytes, so we really need here multiple assays. And actually, the precision of diagnosis is key particularly for these drugs that can be hugely impactful for patients. This is a recent data unpublished from my lab looking at CD3 counts in prostate cancer, and in blue you have the mismatched repair cancers. And you'll see that some have high CD3s and some don't, which is quite interesting.

I'm trying to move to the next slide. This is a patient with very high CD3 infiltration in brown. This is a patient with mismatched repair without the high infiltration. These are both mismatched repaired defective patients, so in some patients, something is going on causing immunosuppression. These patients can respond beautifully to pembrolizumab, and I have I think, four or five patients now doing really very well on pembro with either MMRD or CDK12 or rearrangements of the PD-L1 PD-L2 locus. But not every patient with MMRD prostate cancer responds to immunotherapy, and this does require further careful consideration through prospective trials. And this is the case, as well, not only with the prostate cancer, but also with bowel cancer. This is the IHC and the multicolor area for immune cells from this responding patients. And you can see high PD-L1 and multicolor immunofluorescence showing multiple T-cells invading this tumor.

But actually what's interesting, actually, is that PD-L1 is generally, with our assay anyway, our IHC assay lab with a cell-signaling antibody PD-L1 expression is actually very low in all the other chambers that are not MMR defective, largely, maybe not entirely, but it's really quite low. And this is a significant issue, but there are also cancers that have high lymphocyte infiltration that are not mismatched repair defective. And we published this in this JCI paper last year. So, other cancers, this is CRPC biopsies are inflamed but are not mismatched repair defective, and this is one of those. And this is CDK12 mutated bi-allelic mutated prostate cancer and especially is now responding to pembrolizumab in my center. And this is work led by Arul Chinnaiyan from last year, and Colin did not mention CDK12 so it does merit some discussion.

CDK12 bi-allelic mutations, we don't see a lot of deletion or we see some, but bi-allelic CDK12 aberrations do sensitize to immunotherapy associated with this tandem duplications in many, but not all, patients with bi-allelic loss and causing increased rearrangements, very high rearrangement levels and therefore rearrange proteins that become the neoantigens. And these are our data on over 500 patients now. About 5% of patients have CDK12 mutations. They seem to have a worse prognosis. They have higher TILs generally, but not always, and this is found primarily with bi-allelic and not mono-allelic loss. What about PARP and DNA repair inhibitors? You've heard about our data from my former fellow who is now in Barcelona, Joaquin Mateo at ASCO. We'll have more data at ESMO. Bottom line, PARP inhibition and platinum works for the DNA repair defective prostate cancers, and can I say, this is not just BRCA2, we have seen BRCA1 responders in prostate.

We've seen multiple PALB2 responders. We've seen ATM responders, whatever anybody says, not only with Olaparib but also with other PARP inhibitors. But remember, if you look at the labs that have done a lot of this work, the sensitization... We've got the purple arrows. By BRCA2 loss, is much better than by ATM, so ATM is different to BRCA2. The actual detail matters. It's a different magnitude of sensitization, and you do need lots of both alleles, which can be hard to detect, particularly in blood. This is one of my patients, is the lovely Nina Tunari, our radiologist. This is a BRCA2 patient responding to Olaparib liver metastases appearing. This patient was on trial for over two years. Here's a TOPARP-B patient. You see a beautiful response in the neck nodes, and you'll see this man was on trial again for over two years.

And this the next one is an ATM patient. ATM loss does sensitize TOPARP-B inhibition, and you'll see paraaortic lymphadenopathy. Pelvic lymphadenopathy. This patient responding. We have at least seven patients responding with ATM SNA in our TOBARP-B trial, and that paper is now submitted. And actually, this is our TOPARP-B data by gene alteration, and you'll see the PSA on the left and the RECIST on the right, the different colors that are gene aberrations are two different doses. Our trial did suggest that the higher dose of 400, this is a randomized trial of two doses, the higher dose was more active than the lower dose, and we did in this trial look at SNA single nuclear aberrations and IHC. And for ATM, I do think that IHC does matter, but the biological context may also matter for, I think, sensitizing TOPARP-B inhibition.

The ATM loss by IHC can also be heterogeneous. So, you may get mixed responses, which we often see in these scans on this patient's one area getting better and one area not getting better. So, what about other things? Very briefly, SPOP mutation sensitize to hormone therapy, and PSMA you're going to hear more about, but the devil remains in the details. So, I'll stop there. In conclusion, we need validation and qualification. We need orthogonal assays. Stratification is coming in this year today, but we do need more data. I think with PSMA, we do have to think about rational biology-based combinations that require our understanding of what PMSA does. But it's an exciting time for us in prostate cancer care. So, thank you.