The evolution of metastasis-directed therapy (MDT) in oligometastatic prostate cancer has highlighted the importance of distinguishing patients with limited metastatic disease from those with underlying polymetastatic biology. As discussed at the 2026 ASTRO Annual Meeting, emerging research on genomic biomarkers and digital pathology artificial intelligence (AI) may provide further insights into disease progression, biologic features associated with progressive disease, and treatment selection.

Following this presentation, featured expert Phuoc T. Tran, MD, PhD, was interviewed by Conference Reporter Editor-in-Chief Tom Iarocci, MD. Clinical perspectives from Dr Tran on these findings are presented here.

The concept of oligometastatic disease was initially supported by studies outside of prostate cancer, including randomized trials in limited metastatic non–small cell lung cancer and the SABR-COMET trial, which demonstrated an overall survival benefit with SABR across multiple tumor types. In prostate cancer, the STOMP trial subsequently showed that MDT could delay the initiation of ADT, with the benefit extending to a median of 21 months. This was followed by the ORIOLE trial, which demonstrated improved progression outcomes with SABR in patients with oligometastatic prostate cancer. Importantly, STOMP and ORIOLE evaluated MDT without concurrent systemic therapy, providing clear evidence that localized treatment can alter the natural history of metastatic castration-sensitive prostate cancer, also now known as androgen pathway modulation–sensitive prostate cancer.

An area that my group has looked at is modes of progression and patterns of failure; after their initial treatment, where do patients fail next? We found that, when first treated with MDT, if a patient with oligometastatic castration-sensitive prostate cancer had 5 or fewer lesions and at least 1 was in the bone, the next site of failure was another bone lesion in 86.5% of cases. If the disease was all nodal, most patients failed again in the nodes, although there was a significant minority of patients who also failed in the bone. So, the biology of the disease seems to be that, even with multiple rounds of MDT, it tends to eventually go to the bone. In another study, when we looked at whether genomic alterations were associated with modes of progression and patterns of failure, we found that patients with TP53 and RB1 mutations had a higher likelihood of progression. These associations were statistically significant but await validation.

We want to distinguish mode of progression from patterns of failure because in patients with oligometastatic prostate cancer, there is the concern that what we are treating is really just the “tip of the iceberg” and that patients we think have oligometastatic disease may actually have undetected polymetastatic disease. In our previously mentioned retrospective study on modes of progression after SABR MDT, 40.9% of patients had long-term disease control (ie, no recurrence for at least 18 months after therapy), 36% were oligoprogressors (≤3 lesions at recurrence) and were treated with another round of MDT, and 23.1% were polyprogressors (>3 lesions at recurrence), developing more extensive disease. For those patients with polyprogressive disease, our initial MDT was probably futile because their underlying biology was polymetastatic. Ultimately, the goal is to find biomarkers that will help us identify patients with polyprogressive biology up front who would benefit most from intensification with ADT, rather than with MDT alone.

In a meta-analysis of the STOMP and ORIOLE cohorts, we identified a high-risk mutation signature (ie, ATM, BRCA1/2, RB1, or TP53) that was prognostic and potentially predictive of benefit from MDT. However, it does not tell us whether a patient is more likely to have polymetastatic disease. The KNIGHTS trial was designed around this question, randomizing patients with oligorecurrent metastatic castration-sensitive prostate cancer and these types of high-risk mutations to 6 months of ADT plus SABR MDT vs the same treatment in addition to niraparib and abiraterone acetate plus prednisone therapy. The idea was that these mutations might identify patients with aggressive, potentially polymetastatic biology who would benefit from treatment intensification. Unfortunately, KNIGHTS was closed early, but we are trying to revive the concept because I think that this remains a central question in the field.

We had previously shown that a multimodal AI (MMAI) biomarker could stratify patients with oligometastatic prostate cancer prognostically and might identify patients who would benefit from MDT. However, the algorithm is essentially a “black box”; we knew it worked, but we did not know what features it was using.

This year, we took that process a step further. At ASTRO 2026, my colleague Philip Sutera, MD, presented work aimed at understanding what digital pathology AI is actually seeing (abstract 316). We used attention heat maps to identify the regions of the pathology slides that were driving the AI score, and we overlaid those regions with spatial transcriptomics to see what was happening biologically. We found that the AI was focusing on core drivers of disease aggressiveness and was reading real tumor biology.

So, for the first time, we were able to ascribe some human-interpretable biology to this biomarker. This suggests that this MMAI biomarker is not simply picking up on some spurious feature of the pathology slide—it is identifying biology that is important for disease aggressiveness. I think that this is important for a biomarker that is now being used clinically. That said, this is really just the beginning. There is a tremendous amount of information in spatial transcriptomics, and we have only looked at a small portion so far. I think that there will be a wealth of additional findings as we continue to analyze the information we have available.