Emily Grist: Andrea, thank you so much for inviting me to highlight our work and discuss the data in a bit more depth. So I'm going to discuss tumor transcriptome-wide expression classifiers and how they predict treatment sensitivity in advanced prostate cancers. And as Andrea said, we recently published this work in Cell, and you can see the graphical abstract here on the right. I'm a medical oncologist and a clinician-scientist at the UCL Cancer Institute, and I'm funded by both the John Black Charitable Foundation and the Prostate Cancer Foundation via EM Investigator Award. A few declarations of interest. I'm an employee of UCL who could receive a share of commercial revenue from agreements with Veracyte. And I have patents on some of these transcriptome signatures for treatment selection in advanced disease. So I think prostate cancer is a really interesting area to work because even when advanced disease and metastatic, it's clinically extremely heterogeneous. We have an urgent clinical unmet need for two tests, both prognostic tests to determine who needs treatment intensification beyond ADT and an ARPI, which is our current standard of care. And we need predictive tests to determine which cancers are docetaxel sensitive to determine who needs additional therapy with chemo on top of ADT and an ARPI.
To address this, I use the STAMPEDE trial framework, which is a platform study. It recruited metastatic prostate cancer patients as well as high-risk localized patients, or we call them now ultra-high-risk localized patients with either local lymph node disease or two high-risk features of T3-T4 disease, a high PSA more than 40 or Gleason sum of 8 to 10. 3,909 patients were directly randomized to two practice-changing trial arms, either ADT versus ADT and docetaxel, or testing ADT versus ADT and abiraterone. And my aim was to link tumor multi-gene expression signatures to 14-year prospective overall survival follow-up. When we look at this data, I want you to bear in mind that 95% was synchronous metastatic patients. We had a very small proportion of those that had metastatic disease following relapse after radical local treatment. So this was a massive effort to retrieve blocks. So 95% of trial participants wanted to consent for their routine diagnostic tissue to be used in additional research. And we retrieved about 60% of blocks from over 110 trial sites in the UK. These were all blocks were all centrally reviewed and an H&E cut, and we had 2,261, which are all within a digital H&E repository from the abiraterone or docetaxel comparisons.
These were macro dissected, and we did analysis from the H&E. We did orthogonal studies using Ki-67 and PTEN IHC assays, and we also sent sections for expression analysis. We collaborated with Veracyte, so all the data we are looking at is from a clinically accredited laboratory. And we generated gene expression data for 1,523 participants that were randomized to either docetaxel or ABI comparisons. We selected 59 gene signatures. We hypothesized were irrelevant in prostate cancer for prognostic testing, and then we predefined a small number of signatures for predictive testing to reduce our false discovery. These were Decipher PAM50. We also looked at the ARA signature in the abiraterone comparison. We did no inspection of clinical data before this analysis with the aim of generating level 1B evidence, which is the highest level of testing for a predictive biomarker in a retrospective clinical trial cohort. And Decipher and PAM50 were chosen because of some signals from the charted biomarker cohort that luminal B and the highest Decipher groups may be more sensitive to docetaxel treatments, but this was a small biomarker cohort of 160 patients, so underpowered to show a statistically significant interaction. So this is our prognostic data.
The left box is non-metastatic patients. The right is metastatic, and all of our 59 signatures are in these boxes, as well as our Ki-67 scoring from our IHC assay. As you go from left to right, we see an increasing risk of death and from low to high, go from bottom to top, we see increasing statistical significance. And you can see here I've circled Decipher. So this test was strongly prognostic across both non-metastatic and metastatic disease states, but so was Ki-67. This was a surprising result to us, but we saw increased androgen signaling was protective. So the higher the AR signaling, the longer the survival. And we also saw that the clinical impact of some signatures, for example, PORTOS, was dependent on whether you are metastatic or non-metastatic at diagnosis. So PORTOS, a signature of radiotherapy sensitivity was only prognostic in non-metastatic disease and not metastatic disease. This prognostic testing was all adjusted for baseline and prognostic clinical features, which I've outlined at the bottom of the slide. So onto our predictive testing, so Decipher was not predictive for the benefit of abiraterone in the metastatic cohort. Decipher high patients had slightly more benefit from abiraterone, but low Decipher patients also had benefit from the addition of abiraterone. Therefore, we did not see a treatment interaction. We saw no statistically significant interaction on testing Decipher in our non-metastatic patients either, but this is our key results. The left box is patients that are metastatic that had a high Decipher score of 0.8 or more, and the right box is our lower Decipher cohort.
High Decipher patients had a shorter prognosis on ADT and ARPI, but you can see splitting of the curves on the left. So if patients had a high Decipher score, they were more likely to benefit from the addition of docetaxel as compared to those with a lower Decipher score. And this showed biomarker treatment interaction effect of 0.039, so it's strong evidence. The way we select patients for triplet therapy now is often based on counting the number of metastases, and patients are selected when they have high-volume disease. But the direction of effect was the same in both low- and high-volume subgroups. So even low-volume patients with a high Decipher score benefited more than if those patients who had a lower Decipher score. I'm not going to discuss PAM-50 and ARA any further because we did not show that these tests were predictive for the addition of either docetaxel or abiraterone. We then became very interested in this score, which tells us about PTEN inactivity. So you can see in this forest plot, we identified that if your tumor was defined by transcriptome signature to be PTEN inactive, survival was shorter on either ADT or ADT and abiraterone. But actually you can see with the ADT and docetaxel group in the middle there that if you were PTEN inactive, you did not seem to have a shortest prognosis. Therefore, we hypothesized that this was also predictive. So moving on now to not predefined analysis, so this is exploratory. We looked at PTEN inactive metastatic group on the left and PTEN active group on the right. PTEN inactivity associated with shorter survival, but you can see again, far more splitting of the curves between ADT and red, and ADT and docetaxel in green in the left, Kaplan-Meier compared to the right.
So PTEN inactive patients, again, seem to be patients most likely to benefit from the addition of docetaxel. And again, we saw a significant biomarker treatment interaction of 0.002. This again was the direction of effect was the same between both low- and high-volume patients. So even if you were low-volume, if you are PTEN inactive, it appears you derive benefit from docetaxel as compared to if you are PTEN active. We did hypothesis generating work on PTEN by doing an orthogonal assay using IHC to look at PTEN protein loss. And we found that PTEN protein loss by IHC was not predictive. And we saw no biomarker treatment interaction. We saw the same direction of effect and magnitude of benefit from docetaxel in patients who had tumors that were PTEN protein intact or PTEN protein loss. And this then sparked a lot of further work on our behalf to try and understand the difference between these PTEN assays, PTEN IHC assay, and the PTEN inactivity assay. So we wanted to see whether these are the same or different. So we did further scoring and we used PTEN IHC to score over 50% protein loss, so quite low threshold, and a cutoff for more than 90% PTEN protein loss, which we call PTEN homogenous loss.
And you can see here there's a large amount of discordance between the different assays. Concordance of around 50 to 62% was identified. So there's a very high amount of discordance. Some of this is likely to be sample heterogeneity, but it does suggest that these assays might be measuring different properties and PTEN pathway information. And you can see here we've broken down the results by different Decipher scores. So patients in high Decipher compared to low Decipher were more likely to demonstrate PTEN IHC loss when we use the less stringent cutoff. And patients with a high Decipher were more likely to be PTEN inactive. It's not a perfect correlation, so I don't think one is a surrogate for the other, but there's a large amount of overlap between these tests, but again, a large amount of discordance still. So this is a sub-analysis where we restrict this just to metastatic patients and we only look at high Decipher patients, which is a third of the metastatic cohort in the docetaxel comparison who I showed you demonstrated greatest docetaxel benefit. And if we also use the PTEN inactivity score and restrict by that, we see additional splitting of the curves. And you can see here we have a group in red, who are high Decipher and PTEN inactive, with a really striking curve separation where over half the patients died when they were receiving ADT and five years.
Compare that if you were high Decipher but inactive, actually an additional 20% of patients were surviving at five years after receiving ADT with added docetaxel. And you can see some gain of benefit here with docetaxel in the high Decipher but PTEN active group as well. So I think this is a really interesting result in that there may be additional signals in PTEN score to help us understand in addition to Decipher who really should have docetaxel in addition to ADT and ARPI. So we tested this in an external cohort and we worked closely with colleagues in the charted biomarker cohort. So the charted trial, 125 patients were randomized to abiraterone versus placebo, and some of those patients received docetaxel at the clinician's choice. 250 patients were included in a biomarker cohort, and we generated gene expression data on those patients as well. Decipher data was already available in this cohort. But when we made the PTEN signature into the assay, PTEN inactivity was enriched in the docetaxel treated group. So we can't draw any firm conclusions from this, but again, the direction of effect appeared the same. So summary slide, our data provides this level one evidence that Decipher is predictive for added docetaxel benefit in metastatic hormone sensitive prostate cancer.
And I'm hoping that this is now going to help us give clinicians more insight into deciding who we might want to give chemo in addition to ADT and ARPI. I think the data using PTEN inactivity score is more exploratory, but super interesting. We've shown that this is discordant to PTEN IHC. And I think there may be added clinical utility as compared to Decipher alone. But it needs prospective validation, which we're doing in a number of cohorts. We are also hoping to get an improved understanding of how PTEN inactivity score links to the genomic hallmark of homologous recombination deficiency, something we've been doing in collaboration with another lab. And we would like to explore PTEN inactivity in the CRPC setting with PARP inhibitors and platinum chemotherapy. Thank you.
Andrea Miyahira: Thank you so much, Dr. Grist, for sharing your study with us. So I have some questions. What are the potential mechanisms that you think drive higher Decipher and PTEN inactive scores being associated with docetaxel sensitivity?
Emily Grist: That's a great question. So I think one of the challenges here is we don't have a clear biological rationale. I think the Decipher signature incorporates genes from a number of pathways, including cell cycle proliferation, evasion, metastases, and immune activity, but we don't have a clear biological rationale. And that might in part be due to the way it was developed. So it was developed through a machine-learning algorithm. So I think more work needs to be done to understand that. I think PTEN inactivity, PTEN is involved in microtubule formation, chemotherapy, docetaxel is involved with disruption of microtubules, but we don't have a clear explanation of why it interacts with docetaxel effect. I think people have asked me, could this PTEN inactivity signature be a surrogate for genomic instability? Maybe. We just don't know at this point, so it'd be a really interesting area for further research.
Andrea Miyahira: Okay, thank you. And why do you think PTEN IHC and PTEN activity scores are so different and what efforts are needed to get a clear picture of PTEN pathway activity and mechanisms?
Emily Grist: I think there's a number of possible hypotheses or explanations for this. Heterogeneity, but I really think methodologically we did our absolute best to try and be analyzing the same bit of tumor. We used adjacent slides to determine our PTEN inactivity score and PTEN IHC, but we don't have multi-region sampling to be able to determine the difference in these PTEN scoring across different assays across multiple parts of a tumor. These were advanced patients, so there's a lot of extensive tumor within the prostate. I think cutoffs of the assays are really interesting to look at. I think there's a lot of data come through from the AKT inhibitor trials, so both CAPItello-281, which was published recently, and IPATential150, both showed, which is the latter being in the CRPC setting. And this in the CAPItello looked at ADT plus an ARPI and randomized to either placebo or an AKT inhibitor. And this was a biomarker-selected group and they used the IHC PTEN assay. And what we saw is an rPFS benefit in CAPItello. OS benefit is immature, and in our potential, we didn't see an OS benefit. But what I think is interesting here and relevant to this work is that if you use a higher IHC cutoff, so 100% complete PTEN protein loss, you saw more benefit.
Our cutoff for homogenous was more than 90% of cells needed to not display PTEN protein loss. The problem with this is if you use such a stringent cutoff, your subgroup is much smaller. We don't know how PTEN inactivity score associates with NGS PTEN loss in our cohort. These are really challenging samples to sequence genomically. I think the third reason for this discordance may be due to the fact that the PTEN inactivity score is capturing slightly different information. So PTEN protein loss is telling us one part of the story, but potentially this score is capturing other downstream disruption to the PTEN AKT pathway. I think there's a lot to learn from some of the breast cancer trials that looked at AKT inhibition, where stronger OS benefits were seen in patients with AKT inhibition. But their biomarker was less stringent. So they looked at mutations at multiple different genes involved in the PTEN AKT pathway, whereas the prostate cancer trials haven't. And they saw stronger OS benefits than we're seeing so far in prostate trials. So maybe the classifier is telling us about different aspects of pathway alteration. At this stage, we don't know. But again, really interesting area for future work.
Andrea Miyahira: Okay, thanks. And what are your take-home messages surrounding the utility of various transcriptome signatures for research or clinical use?
Emily Grist: So I think this is really robust evidence that advanced prostate cancer can now be stratified into clinically meaningful groups using this transcriptome data, which remember all was collected from routinely collected diagnostic often tissue that was quite old. It was all FFPE. The patients do not have to go through additional invasive testing to get this data, to get these results all in a clinically accredited laboratory. And I think until now, we've only been able to use clinical information, like counting metastases to decide who do we think may benefit from triplet therapy and who doesn't. And this is the first biological data that helps guide those decision-making things around who should I give chemo to in addition to ADT and an ARPI, and who may not benefit so much. So I think it is important data for helping us with those decisions and counseling patients.
Andrea Miyahira: Okay, thanks. And based on these data, how would you propose using these findings on Decipher or PTEN scores be implemented clinically and what are your next steps?
Emily Grist: So the data around the Decipher score is level 1B evidence. So the strongest level of evidence that we can generate from a retrospective trial. This is being used clinically now in some jurisdictions, and we're looking and working with Veracyte to understand how, and if we can expand this into Europe and other areas. The PTEN inactivity score data is more exploratory because it wasn't predefined, whereas Decipher was defined in our statistical analysis and plan, but it does need further validation. We went some way to doing this in the charted cohort, and we saw the same direction of effect that the PTEN inactive patients appeared to derive greater benefit from chemotherapy. But this was a really small sub. It was 1,160 patients as compared to the STAMPEDE biomarker cohort, but still I think useful information. I think there are some limitations to acknowledge here. So STAMPEDE randomized to ADT versus doublet therapy. So ADT versus ADT and docetaxel, or ADT versus ADT and abiraterone. And that is no longer the question we need to answer as clinicians. The question now is who should get docetaxel in addition to ADT and an ARPI? We have no level-one evidence to tell us more about this. So all of the triplet therapy trials, their standard of care was ADT and docetaxel, not an ARPI. So we don't have level-one evidence in this space.
I think I can't think of a biological reason why the addition of an ARPI to ADT and docetaxel would undo the interaction we see with Decipher and PTEN inactivity and chemotherapy. These are a shorter prognosis group that needs better treatment. So I can't think biologically why we would undo that by adding an ARPI to this. But some people may say, well, you didn't test in triplet therapy. I would argue this is the best data we have now. Future directions, there are two trials running or about to open. So there's triple switch. There is also the INTENSIFY trial in the UK, which I hope will open soon. And what these trials are looking at is randomizing patients to doublets with an ADT and ARPI versus triplets in men that do not drop their PSA to less than 0.2 on doublet treatment with an ARPI. So that poorest prognosis group. And both those trial groups are working, collaborating with Veracyte to generate transcriptome data on their diagnostic samples and will stratify their analysis by that. So that might help us and give us a bit more information. And by the end of the year, expecting data from the ENZAMET trial who we've collaborated really closely with, they will generate gene expression data on their samples. And there is a subgroup of men in that trial that received triplet therapy, so it was physician's choice as to whether to give docetaxel. So I think we will have some more useful information here. But I do think even though tested within a doublet trial, it's useful information for clinicians given, again, we have no level-one evidence for triplet therapy versus ADT and ARPI currently. And people are giving chemotherapy in addition to men often with high-volume disease. So I think it's helpful information.
Andrea Miyahira: Okay. Well, thank you so much for sharing this important study. And I look forward to your next papers.
Emily Grist: Thank you, Andrea. Thanks very much. Bye then.