Future directions for precision oncology in prostate cancer

Abstract

Clinical genomic testing is becoming routine in prostate cancer, as biomarker-driven therapies such as poly-ADP ribose polymerase (PARP) inhibitors and anti-PD1 immunotherapy are now approved for select men with castration resistant prostate cancer harboring alterations in DNA repair genes. Challenges for precision medicine in prostate cancer include an overall low prevalence of actionable genomic alterations, and a still limited understanding of the impact of tumor heterogeneity and co-occurring alterations on treatment response and outcomes across diverse patient populations. Expanded tissue-based technologies such as whole genome sequencing, transcriptome analysis, epigenetic analysis, and single-cell RNA sequencing have not yet entered the clinical realm and could potentially improve upon our understanding of how molecular features of tumors, intra-tumoral heterogeneity, and the tumor microenvironment impact therapy response and resistance. Blood-based technologies including cell-free DNA, circulating tumor cells, and extracellular vesicles are less invasive molecular profiling resources that could also help capture intra-individual tumor heterogeneity and track dynamic changes that occur in the context of specific therapies. Furthermore, molecular imaging is an important biomarker tool within the framework of prostate cancer precision medicine with a capability to detect heterogeneity across metastases and potential therapeutic targets less invasively. Here, we review recent technological advances that may help promote the future implementation and value of precision oncology testing for patients with advanced prostate cancer.

Introduction

Advances in sequencing technologies over the last decade have allowed for a comprehensive view of the prostate cancer genome, epigenome, and transcriptome, identifying drivers of prostate cancer initiation and progression and diverse mechanisms of treatment resistance.¹ Based on the field’s growing understanding of the prostate cancer genome, predictive biomarkers have translated to clinic practice with two classes of genomically-driven therapies (poly-ADP ribose polymerase (PARP) inhibitors and anti-program cell death protein 1 (PD1) immunotherapy) now FDA-approved for the treatment of select men with metastatic castration resistant prostate cancer (CRPC) harboring specific aberrations in DNA repair genes.

Despite these exciting advances, there are still several barriers that limit the widespread clinical implementation of genomic sequencing, including cost, access, and feasibility based on often limited tissue availability or quality. Further, only a subset of prostate cancers harbor actionable genomic aberrations, and even then not all patients respond durably to genomically-selected therapy.^2–4 This may be due to a number of factors including the presence of co-occurring genomic and non-genomic alterations, intra-patient tumor heterogeneity, and the development of acquired resistance. Integrative analyses combining genomics with other features such as transcriptome and epigenome, as well characteristics of the tumor microenvironment, may provide additional insights into identifying patients most likely to benefit from our current therapies and inform the development of novel biomarker-driven and precision therapy approaches for patients. These types of integrative analyses are still maturing and are not yet applied clinically. Moreover, methods for linking large emerging datasets with clinical information will be essential for the more accurate prediction of treatment response.

The study of advanced prostate cancer and tumor evolution in the context of therapy is particularly challenged compared to other tumor types due to the challenges of accessing metastatic tissue given the predominance of sclerotic bone metastases that are often difficult to biopsy and yield limited tumor for molecular studies. In addition, the analysis of a tumor from one region might fail to capture intra-individual tumor heterogeneity. Furthermore, the genomic profiles of bulk tumors may not capture intra-tumoral heterogeneity, which may contribute to drug sensitivity and influence resistance patterns.^5–7 Another major barrier in the broad implementation of precision medicine in prostate cancer is our currently limited understanding of the impact of molecular, genetic, and environmental factors across diverse racial and ethnic populations. This should be understood because the incidence of prostate cancer and its mortality rates can vary substantially.^8,9 Barriers also exist that that limit the availability of testing and access to targeted therapies worldwide.

In this review, we discuss recent technological advances that may help promote the future implementation and value of precision oncology testing for patients with advanced prostate cancer.

Tissue-based technologies

Since metastatic tissue acquisition in prostate cancer is a challenge, primary archival specimens are often used to assess for actionable alterations involving DNA repair genes, as these alterations tend to be early events in prostate cancer and therefore detectable through primary tumor sequencing analysis.^10,11 Mateo et al.¹⁰ reported a high concordance of DNA repair alterations when comparing primary prostate cancer and metastatic CRPC tissues. All alterations involving CDK12, BRCA2, ATM, PALB2, and MSH6 found in metastatic CRPC biopsies of 9 of 61 patients were detectable in patient-matched, diagnostic, treatment-naive primary tissues. Conversely, increased AR, TP53, RB1, and PI3K/AKT pathway alterations were enriched in CRPC compared to same-patient primary samples, indicating that these likely emerged later. Schweizer et al.¹¹ reported 84% concordance of DNA repair gene mutations between primary prostate cancer samples and paired metastatic tissue or cell-free DNA (cfDNA) from 51 patients. These data further support DNA repair alterations being early events, and the majority (90%) of patients enrolled on the Phase 3 PROfound trial of olaparib were screened for DNA repair aberrations using targeted exome sequencing of primary archival prostate tissues with an overall sequencing success rate of approximately 70%.¹² Therefore, primary tumor targeted sequencing is a viable approach to evaluate for somatic DNA repair aberrations, particularly when metastatic tumor biopsies are not feasible. Circulating tumor DNA (ctDNA) may also be considered a tissue alternative to assess for DNA repair mutations. Tukachinsky et al.¹³ evaluated ctDNA from 3,334 patients with mCRPC, including 1,674 screening samples from rucaparib trials (TRITON 2 and TRITON 3) using the Foundation Medicine ctDNA targeted assay; 94% had detectable ctDNA (median ctDNA fraction 7.5%). Overall, 72/837 patients had BRCA1/2 mutations detected in tissue, 67 (93%) of which were also identified using ctDNA. Limitations of using ctDNA include decreased sensitivity for detection of copy number alterations (e.g., BRCA2 deletions) and the presence of confounding factors in blood such as clonal hematopoiesis alterations that may also involve DNA repair genes (e.g., ATM). When looking for acquired phenotypic changes in prostate cancer (e.g., histologic transformation) and for studying non-genomic treatment resistance pathways, metastatic biopsy is preferred. Tissue-based sequencing is of particular value when considering expanded molecular profiling for research using emerging technologies such as whole genome sequencing, transcriptome analysis, in situ analyses, or integrative epigenetic studies.

Whole genome sequencing (WGS)

Along with technological advances and an overall decline in sequencing costs, WGS is positioned as a possible future tool to identify more detailed genomic aberrations in prostate cancer including those involving non-coding regions and structural variations. Prostate cancer WGS studies have revealed important alterations not detectable by targeted exome approaches. In CRPC, amplification of an upstream enhancer of the androgen receptor (AR) gene leads to resistance to AR pathway inhibitor (ARPI) therapy.^14–17 A comprehensive analysis of nearly 200 patients with metastatic CRPC defined distinct genomic subgroups by WGS, including a cluster with a tandem duplication phenotype that correlated with biallelic CDK12 inactivation and another cluster showing homologous recombination deficiency features with numerous deletions and BRCAness-associated gene alterations.¹⁷ In addition, a mutational signature for homologous recombination deficiency may be informative for PARP inhibitor or platinum response even in those that lack DNA repair mutations,¹⁸ supporting a possible utility of using expanded genomic analysis over targeted exome profiling in the clinic. Studies in other diseases have shown the feasibility of using WGS in the clinic, including WGS of pediatric cancers.¹⁹

Transcriptome analysis

Commercial gene expression classifiers such as Decipher^{® 20–22} and the Oncotype DX^® Genomic Prostate Score²³ have demonstrated prognostic value in patients with primary prostate cancer. Transcriptome sequencing is not currently used clinically in patients with advanced metastatic disease. Bolis et al.²⁴ integrated the transcriptional profiles of prostate cancers at various disease stages ranging from normal prostate tissue to primary prostate cancer and metastatic CRPC. Principal component analysis showed transcriptional changes related to different disease stages. They performed trajectory analysis to characterize disease progression and found that most prostate cancers evolve from normal tissue by continuously increasing AR signaling and increasing pseudotime resulted in a gradual upregulation of cell cycle-related genes and concomitant downregulation of androgen-responsive genes. Recently, the PAM50 clustering model based on gene expression data of primary prostate cancers classified prostate cancer patients into three molecular subtypes (luminal A, luminal B, basal)²⁵; the basal subtype (detected in primary prostate tissue) has been associated with inferior outcomes in patient with advanced disease treated with ARPI or docetaxel chemotherapy versus luminal. Ongoing studies are underway to better understand the predictive and prognostic value of these gene expression signatures. Additionally, an integrated neuroendocrine prostate cancer (NEPC) score, based on the expression of a set of 70 genes,²⁶ may identify NEPC tumors or those patients at high risk for the development of NEPC progression after AR-targeted therapies. Treatment-emergent NEPC shares similar genomics as castration resistant adenocarcinoma, indicating a possible utility of transcriptomics over genomics as a prognostic or diagnostic assay in the context of advanced prostate cancer and NEPC. Other gene signatures and mRNA expression-based analyses may complement genomic testing in the future, and biomarker-driven clinical trials incorporating mRNA profiling such as the Alliance CRPC umbrella trial A032102 (PREDICT) are now being launched.

Transcriptome analysis does not always detect cell type composition within the tissue to distinguish tumor cell populations, immune cells, and other cells of the tumor microenvironment. For the analysis of cell type-specific expression profiles in heterogenous samples, computational deconvolution of bulk transcriptome data is required.²⁷ Wu et al.²⁸ examined gene expression data of primary prostate cancer tissue and normal prostate tissue using CIBERSORT, a tool for deconvolution, and analyzed the proportion of 22 immune cells infiltrating in the microenvironment and their prognostic effects in prostate cancer. They found that the detection of M1 macrophages and neutrophils in prostate cancer tissue was associated with poor prognosis. In metastatic CRPC, transcriptome analysis using CIBERSORT has revealed substantial variation in overall immune infiltrate–related transcripts among tumor biopsy sites, as well as heterogeneity in inferred immune cell populations.²⁹ Single-cell RNA sequencing (scRNA-seq) has demonstrated utility in assessing intratumoral heterogeneity at single cell resolution as well as the complexity of the microenvironment.^30–32 In a recent study by Chen et al.,³³ scRNA-seq revealed that prostate cancer cells modulate infiltrating T cells to express KLK3, which establishes a pre-metastatic niche in lymph nodes. They also showed cancer-associated fibroblast-marker-expressing endothelial cells were enriched in CRPC and promoted cancer cell invasion. These results indicate that the tumor heterogeneity and the microenvironment may play important roles in disease progression and metastasis through close cell-cell communication. Another scRNA-seq analysis of metastatic CRPC revealed that resistance to enzalutamide was associated with cancer cell–intrinsic epithelial–mesenchymal transition and transforming growth factor-β (TGF-β) signaling, suggesting the clinical utility of inhibiting TGF-β.³⁴ In addition, NEPC cells showed divergent expression programs driven by HOXB5, HOXB6 and NR1D2 as well as transcriptional regulators promoting lineage plasticity, which might help inform therapeutic strategies for NEPC. Furthermore, scRNA-seq and single-cell assays for transposase-accessible chromatin (ATAC) sequencing of prostate cancer resistance models revealed pre-existing and treatment-persistent cell subpopulations, which may lead to the prediction of the risk of recurrence and disease progression.³⁵ Spatial transcriptome sequencing is a newly emerged technology, allowing for the profiling and visualization of cells while they remain in their tissues. Understanding cell-cell interactions in a spatial context between tumor cells and within the tumor microenvironment may inform resistance subtypes and therapeutic strategies.³⁶ Brady et al.⁷ performed digital spatial profiling of metastatic prostate tissues to assess and quantify transcript and protein abundance in spatially-distinct regions tissue specimens. Although there was high intratumor concordance for the status of particular gene expression signatures including AR activity and NEPC-associated genes, they showed the juxtaposition of AR⁺/NE⁻ and AR⁻/NE⁺ tumor phenotypes within the same metastasis. Furthermore, digital spatial profiling revealed that most metastatic tumors are devoid of significant inflammatory infiltrates and express low-to-absent immune checkpoint proteins CTLA4, PD1, and PD-L1, supporting the very low response rates of immune checkpoint inhibitors observed in the majority of prostate cancer patients. Therefore, scRNA-seq and spatial transcriptome analyses could contribute to future precision oncology in prostate cancer by characterizing transcriptional diversity in the tumor and its microenvironment.

Epigenetic analysis

Epigenetic alterations, including changes in DNA methylation and histone modifications, influence gene expression and are key factors driving prostate cancer initiation, progression, and treatment resistance.³⁷ There are several techniques for epigenetic analysis that may be applied to clinical specimens, including bisulfite sequencing, 5-hydroxymethylcytosine (5hmC) sequencing, ATAC sequencing, and chromatin immunoprecipitation (ChIP) sequencing.

Bisulfite sequencing is commonly used technology for profiling DNA methylation with single-base resolution. This method is based on the finding that treatment with sodium bisulfite leads to deamination of unmethylated cytosines into uracils, while methylated cytosines (both 5-methylcytosine [5mC] and 5hmC) remain unchanged.³⁸ As early events in prostate cancer, hypermethylation of the promoter regions of several genes have been reported such as APC, CCND2, GSTP1, RARB2, and RASSF1.^39–41 ConfirmMDx is a multiplex epigenetic assay^42,43 that combines three methylation regions including GAS6, GSTP1, and HAPLN3 as a classifier for distinguishing prostate cancer from benign tissues.⁴⁴ In CRPC, unsupervised hierarchical clustering of recurrent hypomethylated regions (rHMRs) identified a novel cluster with significantly higher methylation levels at rHMRs as well as a cluster of treatment-emergent NEPC.⁴⁵ The novel hypermethylated cluster was enriched for mutations involving TET2, IDH1, BRAF, and DNMT3B. Although DNA methylation changes are usually associated with poor clinical outcomes, methylation of the promoter region of SRD5A2 gene has been correlated with better prognosis in CRPC.⁴⁶ Further studies exploring the role of DNA methylation in prostate cancer progression could help develop novel strategies for precision oncology biomarkers and therapies in prostate cancer.

5hmC is the first oxidative product of 5mC catalyzed by ten-eleven translocation enzymes and another important epigenetic regulator of transcription.⁴⁷ Oxidative bisulfite sequencing⁴⁸ and Tet-assisted bisulfite sequencing⁴⁹ are both single-base resolution sequencing strategies which distinguish between 5mC and 5hmC. Additionally, several bisulfite-free methods to detect 5hmC at base resolution have recently been developed.^50,51 Sjöström et al.⁵² reported that 5hmC levels in metastatic CRPC associate with gene expression to a greater degree than promoter methylation or copy number, especially in androgen response genes, and 5hmC has the ability to track disease-specific gene activation.

ATAC sequencing is a method for detecting chromatin accessibility across the genome. It uses Tn5 transposase to cut and tag adapters to regions of accessible chromatin which correspond to transcription factor binding sites and nucleosome positioning.⁵³ For instance, integrative analysis of gene expression and ATAC sequencing of CHD1 loss prostate cancer models revealed substantial changes in open and closed chromatin with associated transcriptomic changes, which resulted in the emergence of plasticity via upregulation of transcription factors that promote non-luminal lineage programs.⁵⁴ ChIP sequencing for histone modification marks such as H3K27ac, H3K4me3, and H3K27me3 might be helpful for detecting distinct prostate cancer subtypes⁵⁵ and/or other epigenetic events associated with prostate cancer progression.⁵⁶

Blood-based technologies

In order to overcome some of the current limitations associated with tumor biopsies, liquid biopsies including analysis of ctDNA, circulating tumor cells (CTCs), and extracellular vesicles (EVs) in the blood are emerging as noninvasive tools for precision oncology.

Circulating tumor DNA

ctDNA is well recognized as an important biomarker tool in several cancer types.^57–59 In prostate cancer, ctDNA tumor fraction in cfDNA is prognostic⁶⁰ and ctDNA is capable of detecting common recurrent prostate cancer aberrations in metastatic CRPC including those involving DNA repair genes, with high concordance with matched tissue biopsies.^13,61–66 Additionally, rapid autopsy studies have shown that ctDNA can capture more driver alterations than multiple randomly selected tissue samples, indicating its utility in detecting intra-patient tumor heterogeneity.⁶⁷ In practice, several commercial targeted ctDNA platforms are available to assess for targetable aberrations in CRPC, particularly in cases when tumor tissue is not available and a new biopsy is not feasible to obtain. While ctDNA is convenient and noninvasive, there are limitations. ctDNA analysis may detect aberrations involving both the tumor as well as other cells in the circulation, such as white blood cells. Normal leukocytes harboring clonal hematopoiesis of indeterminate potential (CHIP) variants may confound ctDNA test interpretation, which is especially relevant if they involve BRCA or ATM or other genes linked to PARP inhibitor approval.^66,68,69 Additionally, ctDNA is diluted by cfDNA from non-cancer cells and declines with therapy response such that not all patients will have detectable ctDNA, leading to difficulties with the identification of mutations and copy number aberrations.^66,70 Further studies exploring baseline and dynamic changes in ctDNA are needed to validate ctDNA as prognostic and response biomarker and will provide additional insights into the evolution of specific clonal and subclonal lesions that occur during prostate cancer disease progression and in the context of therapies.

cfDNA methylation analysis of plasma is not currently used clinically in prostate cancer, but is being developed across cancer types for early detection strategies and tumor classification.^71,72 In prostate cancer, dynamics in cytosine modification profiles of cfDNA has been shown to be a predictive biomarker for abiraterone treatment response.⁷³ Additionally, Mahon et al.⁷³ showed that undetectable methylated GSTP1 is a favorable prognostic biomarker in metastatic CRPC. More recently, whole genome bisulfite sequencing analysis of cfDNA from NEPC patients revealed that the methylation patterns detected in cfDNA reflected those observed in biopsy tissues,⁷⁴ which include NEPC-associated epigenetic changes such as hypermethylation of ASXL3 and SPDEF and hypomethylation of INSM1 and CDH2.²⁶ In a recent study by Berchuck et al,⁷⁵ a NEPC Risk Score was developed using methylated DNA immunoprecipitation coupled with next-generation sequencing (MeDIP-seq) to predict the presence of NEPC using differentially methylated regions detected from NEPC and CRPC tumor samples. MeDIP-seq was then applied to cfDNA which showed that this tissue-informed analysis resulted in high sensitivity and specificity for detecting NEPC. Overall, these results support the possible utility of using cfDNA methylation as a monitoring tool which may be particularly relevant when detecting epigenetically driven subtypes of advanced disease such as NEPC.

Recently, it has been reported that cfDNA fragment characteristics can also help infer nucleosome positioning and transcription factor binding sites.⁷⁶ Ulz et al.⁷⁷ developed an accessibility score to estimate transcription factor activity based on cfDNA sequencing and nucleosome footprint analysis. They analyzed two cfDNA samples from a prostate cancer patient collected in a 12-month interval, during which the adenocarcinoma transdifferentiated to a treatment-emergent NEPC and showed reduced accessibilities of the binding sites of AR, HOXB13, NKX3–1, and REST, indicating that the accessibility score can distinguish NEPC from prostate adenocarcinoma. cfDNA fragment analysis has also been feasible in other cancer types such as early-stage colorectal cancer.⁷⁷ The analysis of nucleosome positioning in cfDNA could overcome some of the limitations of mutation-based ctDNA analysis with a potentially higher detection sensitivity.⁵⁹

Circulating tumor cells

CTCs offer not only quantitative information but also the ability to isolate heterogenous cell populations, quantify gene expression, detect splice variants, and measure specific protein expression. In prostate cancer, the enumeration of CTCs has been shown to be a prognostic biomarker,^78,79 and the detection of AR splice variant 7 (AR-V7) in CTCs may predict resistance to abiraterone and enzalutamide.^80,81 The PROPHECY study, a multicenter, prospective-blinded clinical trial, investigated the impact of CTC AR-V7 detection in men with metastatic CRPC starting ARPI treatment on progression-free survival (PFS) and OS.⁸² Detection of AR-V7 in CTCs by two assays was significantly associated with shorter PFS (median PFS 3.1 vs 6.9 months and 3.1 vs 6.1 months, respectively) and OS (median OS 10.8 vs 27.2 months and 8.4 vs 25.5 months, respectively). On the other hand, a recent study detected the transcriptional profile of CTCs from metastatic prostate cancer patients using a multiplex gene expression biomarker panel including AR splice variants, AR targets, and NEPC markers.⁸³ The result showed that increased expression of AR-regulated genes was independently associated with shorter OS on multivariate analysis, while AR splice variant status was not significant. Additionally, Scher et al.⁸⁴ quantified digital pathology features of CTCs from 179 metastatic CRPC patients. They classified individual CTCs into 15 phenotypic subtypes and revealed that low CTC phenotypic heterogeneity was associated with better OS in patients treated with ARPIs (median OS 28.1 vs. 8.8 months), whereas patients with an increasing heterogeneity score had a higher risk of death on ARPIs relative to taxane chemotherapy. In recent years, several single-cell analyses of CTC have revealed prostate tumor heterogeneity that could contribute to patients’ prognosis.^85–87 Miyamoto et al.⁸⁵ conducted scRNA-seq of 77 CTCs from 13 CRPC patients and showed the activation of noncanonical WNT signaling was associated with resistance to ARPIs. Conteduca et al.⁸⁷ reported a patient with both primary prostate adenocarcinoma and liver metastasis with NEPC morphology at the time of initial presentation, whose CTCs reflected the state of intra-patient tumor heterogeneity. This supports the promise of CTCs in representing the molecular profiles of metastases, though much still remain to be learned about the origin of CTCs and how well they reflect the molecular landscape across heterogeneous tumors. Conteduca et al. performed single-cell copy-number variation analysis of CTCs and found copy-number heterogeneity involving tumor suppressor genes, such as RB1, TP53, and PTEN, associated with differential detection of AR and NEPC marker protein expression in CTCs⁸⁷. These states of heterogeneity were highly similar to those observed in tumor biopsies, indicating the feasibility of extending CTC analysis at the single-cell level to integrate genomics with protein expression. Furthermore, drug sensitivity testing of ex vivo cultured CTCs could contribute to future precision oncology efforts.⁸⁸

Extracellular vesicles (EVs)

EVs are secreted by cells and detected in almost every biological fluid, especially blood. There has been growing interest in cancer EVs due to their unique functions as intercellular messengers and their diagnostic and therapeutic potential.⁸⁹ EVs may serve as biomarkers for the early diagnosis of prostate cancer⁹⁰ and for the detection of advanced disease.⁹¹ Additionally, exosomal specific microRNAs and exosomal AR-V7 have been shown to be potential prognostic biomarkers and for prediction of ARPI response in CRPC patients.^92,93 The commonly used techniques for isolation of EVs include ultracentrifugation, size-exclusion chromatography, ultrafiltration, and immunoaffinity capture.⁹⁴ However, due to their small diameter and the co-existence of different types of vesicles, their isolation is challenging. Furthermore, since the current isolation technologies are usually time-intensive, the development of reliable and efficient isolation procedures would be mandatory for the clinical applications of EVs.

Molecular imaging

Molecular imaging allows for the visualization and quantification of specific markers or biological processes across anatomic disease sites.^{95 18}F-Fluciclovine positron emission tomography (PET) with computed tomography (CT) imaging detects amino acid transporter upregulation in prostate cancer versus surrounding tissues, and was FDA approved in 2016 for the detection of recurrent prostate cancer.⁹⁶ Prostate specific membrane antigen (PSMA) PET imaging is now recognized for its even higher sensitivity and specificity for identifying prostate cancer recurrence and metastasis⁹⁷ and is rapidly replacing ¹⁸F-Fluciclovine PET. In 2020, the FDA approved Ga-PSMA-11 PET/CT for the initial diagnosis and staging of prostate cancer patients with suspected metastases and the imaging of patients with biochemical recurrence after prostatectomy or radiation therapy. Piflufolastat F 18 was also approved as the second PSMA-targeted PET imaging agent in 2021 for the same prostate cancer imaging indications as Ga-PSMA-11.

In metastatic CRPC, PSMA/PET imaging is also useful to identify candidates for the PSMA-directed radionuclide therapy Lu-PSMA-617. The Phase III VISION trial comparing Lu-PSMA-617 plus standard of care versus standard of care alone improved progression free survival and overall survival for men with metastatic CRPC previously treated with ARPI and taxane chemotherapy. All patients had PSMA-positive disease identified by Ga-PSMA-11 PET/CT. In the VISION trial, PSMA positivity was defined as at least one PSMA-positive metastatic lesion with PSMA uptake greater than liver, and no PSMA negative soft tissue or visceral lesions ≥1cm or lymph nodes ≥2.5cm. Lu-PSMA-617 was approved by the FDA in March 2022 for patients with metastatic CRPC after progression on ARPI and docetaxel, and Ga-PSMA-11 was also approved as a companion diagnostic imaging test. This will expand the number of patients with advanced disease receiving PSMA PET/CT scans. Understanding how baseline PSMA-PET correlates with clinical features, PSMA dynamics on therapy, and PSMA PET characteristics at progression may help refine PSMA PET as biomarker in the context of Lu-PSMA-617. Patterns at progression may influence future sequencing of other PSMA-targeted drugs in development. Beyond targeting PSMA, PSMA PET/CT may also be useful in assessing tumor dynamics and response in the context of other prostate cancer therapies such as ARPI and chemotherapy. PSMA expression is indirectly regulated by the AR, and a subset of CRPC tumors may lose PSMA expression in later stages of the disease in conjunction with loss of AR. In the VISION trial, 12.6% did not meet inclusion criteria based on PSMA-imaging. Several studies incorporating dual-tracer PET/CT have reported that metastatic CRPC patients with low PSMA expression or PSMA-negative fluorodeoxyglucose (FDG)-positive discordant lesions have poor prognosis.^98,99 This could be due to NEPC transformation or AR-negative prostate cancer, which is supported by a recent study showing the positive correlation between levels of FDG uptake-associated genes with NEPC gene signatures in PSMA-suppressed tumors.¹⁰⁰ In addition, Wang et al.¹⁰¹ also demonstrated that 24% of PSMA-negative, FDG-positive disease was found in patients with an early PSA progression during castration. These results suggest that dual-tracer PET/CT might enable the earlier diagnosis of metabolically active PSMA-suppressed disease for earlier or more aggressive management. Fluorodihydrotestosterone F18 ([¹⁸F]-FDHT) PET/CT directly images AR-expressing tissues.¹⁰² A recent study analyzing 133 metastatic CRPC patients using molecular imaging with FDHT and FDG PET/CT also showed that 49% of patients had at least one FDHT-negative, FDG-positive lesion, which was the most potent imaging phenotype with respect to adverse prognosis.¹⁰³ Therefore, dual-tracer with FDHT and FDG imaging could also be a future diagnostic or prognostic biomarker.

The presence of PSMA-negative or AR-negative CRPC lesions may lead to the suspicion of treatment emergent NEPC differentiation, but it is not a strategy to uniquely identify NEPC. Recently, Puca et al.¹⁰⁴ found that delta-like protein 3 (DLL3), which is an inhibitory ligand of the Notch signaling pathway,¹⁰⁵ is aberrantly expressed on the cell surface of the majority of NEPC.¹⁰⁴ DLL3 is also aberrantly expressed in small cell lung cancer (SCLC). ImmunoPET imaging with ⁸⁹Zr-labeled SC16 antibody is capable of detecting DLL3 positive SCLC and NEPC in preclinical models.^106,107 Korsen et al.¹⁰⁷ performed in vivo ⁸⁹Zr-SC16 PET imaging and biodistribution studies using xenograft models of NCI-H660, which is a DLL3-positive NEPC cell line, and DU145, which is a DLL3-negative AR independent prostate cancer cell line. They showed ⁸⁹Zr-SC16 PET imaging can uniquely detect NEPC lesions, indicating that this technology might be useful for the early detection of NEPC in the future and for selection for DLL3-targeted therapies such as T cell engagers.

Although there are several barriers for translating novel molecular imaging tools into daily clinical practice including expense and time to validate novel tracers, a lack of established framework for multicenter trials, and variable quality of imaging acquisition and analysis,¹⁰⁸ molecular imaging may play an important role in future precision oncology in prostate cancer and has great potential to guide more effective and less invasive target detection.

Data integration

Most clinical data as well as molecular information are not well integrated, which provides cumbersome datasets. Since genes, transcripts, proteins, metabolites, and other molecules interact with each other to regulate cellular processes, integrative analysis of multi-omics data is needed for better disease classification, prediction of biomarkers, and understanding of disease biology.¹⁰⁹ Ramazzotti et al.¹¹⁰ established a new cancer subtyping method integrating multi-omics data, called Cancer Integration via Multikernel Learning (CIMLR). They applied CIMLR to multi-omics data from 36 cancer types including 490 primary prostate tumors. CIMLR found three clusters in prostate cancer, and one cluster showed significantly worse outcomes compared to the other clusters, characterized by loss of TRIM35, reduced expression of RHOBTB2, high promoter methylation, and high prevalence of TP53 mutation and/or loss.

Deep learning-based multi-omics data integration has also been developed.^111,112 In prostate cancer, a recent study investigated the Cancer Genome Atlas (TCGA) prostate adenocarcinoma dataset using deep learning and similarity network fusion.¹¹³ From the two models, six multi-omics biomarkers, TELO2, ZMYND19, miR-143, miR-378a, and methylation status of two CpG loci, were selected for multi-omics panel construction. This panel was shown to be a potential biomarker for the early detection of prostate cancer patients at high recurrence risk. Elmarakeby et al.¹¹⁴ developed a deep-learning predictive model named P-NET to predict cancer state in prostate cancer patients on the basis of biological information including mutations, copy number alterations, methylation, and gene expression. P-NET accurately classified metastatic CRPC versus primary prostate cancers. Moreover, this visible neural network model revealed novel alterations which strongly contributed to predictive performance in genes, such as MDM4 and FGFR1. Recently, other novel computational methods have been applied to multi-omics integrative methods,¹¹⁵ including models incorporating histopathology imaging,¹¹⁶ although there are few multimodal studies to date incorporating radiology. Deep learning systems have demonstrated high proficiency at Gleason grading of prostate biopsy specimens¹¹⁷ and have provided support for computational three-dimensional histology analysis.¹¹⁸ Advances in computational methods will further enable the integration of multimodal data including molecular data, histopathology, radiology, and clinical data such as race/ethnicity, tumor grade, recurrence, treatment response, and long term outcomes. This could lead to the development of data-driven novel biomarkers for prostate cancer and a better understanding of its complex nature (Figure).

Figure.

Potential workflow for the future implementation of precision oncology testing for patients with advanced prostate cancer. Tools for implementing precision oncology in advanced prostate cancer include tissue-based technologies, blood-based technologies, and molecular imaging. Clinical information as well as molecular features obtained from the novel technologies are integrated for the detection and application of novel biomarkers and targets.

Conclusions

Precision oncology in prostate cancer is a rapidly evolving field. However, there remain substantial challenges for implementing precision oncology more effectively and more broadly. We have focused on novel technologies and findings that could be used to overcome certain barriers in the advanced prostate cancer setting. Multiple different layers of information including genome, transcriptome, and epigenome could help refine predictive biomarkers and define subclasses that will better predict patient outcomes. The integration of these multi-omics data and clinical information with computational methods including deep learning will also be important. Additionally, integrative analysis might help understand the clinical impact of co-occurring alterations and rare molecular aberrations, leading to broader application of precision oncology for prostate cancer patients. Liquid biopsies and molecular imaging are less invasive technologies that can capture intra-individual heterogeneity, which could affect the treatment response and prognosis. Several biomarker-driven targets are emerging in prostate cancer and clinical trials evaluating the agents against these targets are ongoing. The tremendous progress in the field has only been possible because of large multidisciplinary collaboration and patient engagement.