Figure S14. Luminal B phenotype is associated with higher MYC signaling and lower radiation response scores.

Figure S7. Expression of neuroendocrine genes across CTC phenotypes.

Figure S9. Serum PSA and disease category by CTC phenotype for all samples.

Androgen-targeted therapy, chemotherapy and 177Lu-PSMA radioligand therapy have improved metastatic prostate cancer (mPC) patient survival. However, most patients develop resistance to these therapies. Studying genomic and transcriptomic changes that cause resistance requires repeated sample collection and is not suited to traditional invasive tissue biopsy techniques. We have developed a liquid-biopsy based technique to isolate cell-free DNA (cfDNA) and circulating tumor cell (CTC) RNA from mPC patients enabling genomic and transcriptomic profiling. Transcriptomic profiling of 146 high tumor purity mPC CTC samples from 70 patients identified 4 transcriptional phenotypes driven by changes in luminal, AR and proliferative signaling pathways: Low Proliferation (LP, n= 12, low luminal/AR/proliferation) Luminal A (LumA, n=24, high luminal/AR, low proliferation). Luminal B (LumB, n=31, high luminal/AR/proliferation) and Neuroendocrine Prostate Cancer (NEPC, n=3, low luminal/AR, high proliferation). Compared to LP and LumA, the LumB phenotype is associated with shorter overall survival and early progression on 177Lu-PSMA highlighting an unmet need for treatment options in patients with LumB CTCs. Here, we use mPC CTC transcriptional profiling to identify drivers of proliferation and treatment resistance in the LumB phenotype. Blood was collected from patients with mPC receiving standard of care treatment. CTCs were purified using immunomagnetic capture on a microfluidic platform, subjected to RNA-seq analysis and assessed for transcriptional phenotypes and gene expression signatures. We compared expression of gene sets related to established tumor suppressors or oncogenic drivers, DNA damage repair by homologous recombination repair (HRR), and cell cycle drivers in LP, LumA and LumB CTC phenotypes. LumB CTC samples display elevated expression of cell cycle genes related to S/G2/M phase progression (CCNE1, CCNA2 and CCNB1), but not G1 phase progression (CCND1 or CCND3), suggesting that increased S/G2/M progression is contributing to increased proliferation. In preclinical models, CCNE1 overexpression causes genomic instability and activation of DNA damage repair, cell cycle checkpoints and a S/G2/M phase transcriptional program mediated by the MYBL2-FOXM1 transcription factor complex. Indeed, we also observe increased expression of HRR, cell cycle checkpoint and MYBL2-FOXM1 transcriptional signatures in LumB subtypes. CTC transcriptional profiling has identified an aggressive and treatment resistant LumB phenotype characterized by high expression of genes involved in S/G2/M cell-cycle progression, HRR and cell cycle checkpoints. These findings suggest that elevated cell cycle progression and upregulated DNA repair capability may underlie the aggressive behavior and therapeutic resistance observed in patients with LumB CTCs. Ongoing efforts are focused on developing preclinical models of the mPC CTC transcriptional phenotypes to identify therapeutic vulnerabilities specific to the LumB subtype. David Gallo, Jamie M. Sperger, Amy K. Taylor, Kristen Rosche, Viridiana Carreno, Alex H. Chang, Emily Abella, Kaitlin Durnen, Muhammad Dar, Charlotte Linebarger, William M. Stump, Kendra Marr, Kyle T. Helzer, Matthew L. Bootsma, Grace C. Blitzer, John M. Floberg, David Kosoff, Rana R. McKay, Wei X. Xiao, Shuang G. Zhao, Joshua M. Lang, Marina N. Sharifi. Identifying therapeutic vulnerabilities in metastatic prostate cancer transcriptional phenotypes [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86(2_Suppl):Abstract nr B023.

Figure S15. Longitudinal FOLH1 expression and pathway analysis of CTCs in the 177Lu-PSMA-617 sub-study cohort.

Increased comorbidities and unique service-related exposures among veterans may contribute to unique tumor biology that can influence tumor behavior and response to treatment. Research that incorporates live cells from veteran donors can support investigation of the genomic and transcriptomic biology of veteran tumors as well as how those molecular alterations affect tumor function, such as growth, metastasis, and treatment sensitivity. We conducted a systematic literature review to assess veteran-focused prostate cancer studies published in the last decade that incorporate live primary cells. We found that in the past 10 years, 105 prostate cancer studies were published that utilized live primary cells from the general population compared with zero studies utilizing live primary cells from veterans. Analysis of veterans enrolled in a study utilizing live primary cells at the William S Middleton Veterans Affairs demonstrated that 94% of veterans who were presented with a live cell biospecimen donation study ultimately enrolled. Furthermore, survey responses from veterans enrolled in this study showed that 83% of veterans found their research engagement to be a meaningful experience and 70% reported increased healthcare satisfaction as a result of their participation. These findings suggest that the lack of live cell prostate cancer research in veterans seems to be due to a lack of research opportunities for veterans and not a lack of veteran interest in participation Finally, we demonstrate that implementation of an informatics-based patient screening strategy can potentially support more veteran-focused studies by reducing participant screening time by 89%. SIGNIFICANCE:Live primary cells are a unique model for prostate cancer research that provides both molecular and functional readouts to support advances in veteran prostate cancer care. We detail the shortage of live primary research in veterans and address potential barriers to future studies incorporating this model.

Figure S17. Concordance between protein and RNA expression of cell surface targets in prostate CTCs.

Figure S4. Sample processing is not different between CTC phenotypes.

Treatment with androgen receptor (AR)-targeted therapy, chemotherapy and 177Lu-PSMA-617 radioligand therapy have improved patient outcomes for patients with metastatic prostate cancer (PCa). However, development of treatment resistance remains universal, occurring through AR dependent mechanisms driving constitutive AR signaling, or lineage state transitions that bypass AR signaling and culminate in highly proliferative tumors, including those with luminal B (LumB) and neuroendocrine phenotypes. We recently described two distinct luminal phenotypes with high luminal and AR signaling gene expression but distinguished by high (LumB) or low (luminal A (LumA)) proliferation signature scores, with the LumB phenotype associated with shorter survival compared to the LumA phenotype. We also identified a low proliferation phenotype characterized by low AR/luminal signature and low proliferation, and a neuroendocrine phenotype characterized by high neuroendocrine and proliferation signature scores. Evolution of tumor biology has been difficult to monitor due to challenges in collecting serial tissue biopsies. Blood-based liquid biopsy is non-invasive and well suited for repeat sampling. RNA sequencing of circulating tumor cells (CTCs) provides an accessible means to perform longitudinal transcriptional profiling to track mechanisms of resistance over time. CTCs were isolated from patients with PCa during standard of care treatments including baseline and progression timepoints with automated microfluidic technology integrating negative and positive selection. CTCs and RNA were captured immunomagnetically. RNAseq data was assessed for CTC RNAseq phenotype, signature scores including CCP-31, signatures of tumor suppressor loss (p53, RB1, PTEN) and MYC signatures and genes previously identified as expressed in AR+ PCa or NEPC. We analyzed longitudinal samples including 27 sets of matched baseline and post-treatment samples from 22 patients with metastatic PCa during treatment with AR targeted therapies, chemotherapy and 177Lu-PSMA-617. In 30% (8/27) of these patients, we observed a switch of CTC phenotype at treatment progression from pretreatment baseline sample. The most common switch (n=5) was from the LumA phenotype to LumB phenotype, which was associated with increased RB1 loss signature score (p=0.016). We observed a patient transitioning from a LumB phenotype to neuroendocrine at progression on 177Lu-PSMA-617 radioligand therapy. In this patient, coinciding with the phenotype shift, we detected decreased expression of KLK3 and FOLH1 (PSMA)and increased expression of ASCL1, INSM1 and SYP consistent with a shift from a luminal adenocarcinoma to a neuroendocrine phenotype. To better understand tumor evolution, it is essential to monitor molecular changes during treatment. RNAseq of CTCs enables longitudinal tracking of lineage phenotype transitions. Ongoing studies are investigating mechanisms that drive resistance and plasticity in larger, uniformly treated patient cohorts. Jamie M. Sperger, Marina N. Sharifi, Amy K. Taylor, Krisitin L. Rosche, David Gallo, Viridiana Carreno, Alex H. Chang, Emily Abella, Kaitlin Durnen, Muhammad Dar, Charlotte Linebarger, William M. Stump, Kendra Marr, Kyle T. Helzer, Grace C. Blitzer, John Floberg, David Kosoff, Rana R. McKay, Xiao X. Wei, Shuang G. Zhao, Joshua M. Lang. Monitoring the evolution of treatment resistance by transcriptional profiling of circulating tumor cells with RNAseq [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86(2_Suppl):Abstract nr B071.

Table S1. Clinical characteristics of the 177LuPSMA-treated cohort

Purpose Cell surface-targeted therapies (CSTs) are a rapidly expanding class of cancer treatments with high specificity and reduced toxicity. Matching patients who express specific targets to CST clinical trials remains challenging because of complex eligibility criteria, diverse targets, and the absence of centralized, up-to-date trial databases. These gaps limit patient access and contribute to poor trial accrual. Methods We developed a large language model (LLM)-driven pipeline to identify and annotate CST clinical trials. Using a two-pronged approach, LLMs extracted target information from ClinicalTrials.gov and the National Cancer Institute Drug Database. Eight LLMs, including GPT-4o and several open-source models, were benchmarked against manually curated data sets of 814 CST trials and 814 non-CST trials. We evaluated model performance at target and trial levels and analyzed sources of error. We also provide an up-to-date database of open CST trials and their targets from the >100,000 total oncology clinical trials in ClinicalTrials.gov. Results GPT-4o achieved the highest accuracy in identifying CST trials (96.5%) and their targets (89.5%). Combining data sources improved performance, and accuracy increased with later trial phases. Most errors stemmed from vague therapy descriptions or string-matching issues. The model matched 94% of US trials and >95% of trials globally, with exceptions in China and New Zealand. In predicting cell surface localization, Gemma 3:27b and MedLlama3 correctly labeled all known clinical cell surface targets although performance varied beyond the most well-known CSTs. Conclusion Our LLM-based approach enables real-time, automated matching of patients to CST clinical trials, addressing major barriers to enrollment and expanding trial accessibility. Errors were uncommon, and performance is poised to improve as LLMs evolve. Optimizing patient-trial matching for CSTs can improve both patient benefit and trial success.

Figure S12. Pretreatment CTC and PSMA-PET characteristics in the 177Lu-PSMA-617 sub-study cohort.

While treatment options for castration-resistant prostate cancer (mCRPC) patients have historically been limited, theranostic approaches targeting prostate specific membrane antigen (PSMA) have shown promise, including the FDA approval of 177Lu-PSMA radioligand therapy in mCRPC patients with PSMA positive disease on positron emission tomography scan (PET). While LuPSMA improves clinical outcomes in mCRPC, intrinsic and acquired resistance remain common, the mechanisms of which remain poorly understood. Crosstalk between the PI3K pathway and AR signaling plays a role in resistance to androgen therapies, and may also modulate PSMA expression. We hypothesized that heterogeneity in PSMA expression driven by crosstalk with PI3K pathway signaling could contribute to LuPSMA resistance. Here, we utilized in vitro models as well as circulating tumor cell (CTC) molecular profiling from mCRPC patients receiving LuPSMA to evaluate the role of PSMA protein-expression heterogeneity and PSMA-PI3K signaling crosstalk in LuPSMA resistance. 77 CTC samples were collected from 35 patients with mCRPC receiving standard of care 177Lu-PSMA. CTCs were isolated using microfluidic chip based immunomagnetic capture and protein staining followed by fluorescent imaging. CTC enumeration and single cell fluorescent quantification of PSMA and phospho-ribosomal protein S6 (S6) quantification was compared between patients with early progression (EP) on LuPSMA (within the first 3 cycles/18 weeks) versus those without EP. A mCRPC cell line, 22RV1, was exposed to PI3K pathway inhibitors, before Western blot analysis of PSMA and PI3K expression. 23/35 patients had CTCs detected at baseline (median: 5, range: 1-17), with CTC PSMA fluorescent intensities (PSMA-FI) ranging from less than 100 to greater than 5000 (median: 158.93 range: 69.37-5449.44). There was a significantly increased proportion of EP among patients with low CTC PSMA expression (3/3 with EP) versus high CTC PSMA expression (2/13 with EP) detected at baseline (p=0.018). CTCs from patients with EP exhibited a significant increase in phospho-rpS6 expression from baseline to cycle 2 compared to responders (59.11 vs -124.87, n=10, p=0.017). There was no difference in baseline CTC number (7 vs 5 CTCs/7.5ml, p=0.750) between patients with and without EP. Among PSMA-FI high patients, CTC PSMA heterogeneity (PSMA-FI median absolute deviation 0.606 vs 0.843, p=0.769) did not significantly differ by EP status. In 22RV1 cells, treatment with PI3K pathway inhibitors Everolimus (mean fold change = 1.94, n=3, p = 0.002) and Gedatolisib (mean fold change = 2.78, n =3, p= 0.046), but not Ipatasertib, resulted in a significant increase in PSMA expression. CTC PSMA expression is variable in mCRPC patients with PSMA-PET-positive disease, low CTC PSMA expression correlates with non-response to LuPSMA. CTCs from patients with EP have increased PI3K pathway signaling compared to responders. Crosstalk between PSMA and the PI3K signaling pathway impacts PSMA expression in vitro. William M. Stump, Alex H. Chang, Viridiana Carreno, Matthew L. Bootsma, Jamie M. Sperger, David Gallo, Shannon R. Reese, Emily Abella, Kaitlin Durnen, Kristin Rosche, Muhammad Dar, Charlotte Linebarger, Amy K. Taylor, Kendra Marr, Katharine E. Tippins, Kyle T. Helzer, Grace C. Blitzer, John Floberg, David Kosoff, Rana R. McKay, Xiao Wei, Shuang G. Zhao, Joshua M. Lang, Marina N. Sharifi. Circulating Tumor Cell (CTC) single cell protein quantification reveals inter-patient heterogeneity in PSMA expression associated with LuPSMA response in a Metastatic Castration Resistant Prostate Cancer (mCRPC) patient cohort [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86(2_Suppl):Abstract nr B073.

Figure S13. CTC FOLH1 expression and transcriptional phenotype and PSMA-PET characteristics.

Supplementary Genesets includes prostate cancer specific genesets curated from published literature

Figure S8. MKI67 expression in each CTC phenotype cluster for all samples.

Figure S5. Tumor fraction and immune content across CTC phenotypes.

FOXA1 is a prostate lineage-specifying transcription factor that is frequently dysregulated or mutated in prostate cancer (PCa). While FOXA1 has been reported to exhibit both PCa-promoting and -inhibitory functions, its role within an immune-proficient PCa context remains unclear. Here, we show that prostate-specific deletion of Foxa1 in Pten-deficient mice drives tumor progression by reprogramming luminal PCa cells toward a basal/squamous-like state and promoting an immunosuppressive tumor microenvironment. Histological and transcriptomic analyses reveal aggressive tumors with extensive basal/squamous features, a reactive stroma, and disorganized tissue architecture. Mechanistically, FOXA1 directly represses basal/squamous and inflammatory genes, which become activated upon its depletion. This is accompanied by an accumulation of immunosuppressive myeloid cells, dysfunctional T cells, and immunosuppressive cytokine signaling. Together, these findings demonstrate a tumor-suppressive role for FOXA1 as an enforcer of luminal identity, such that its loss drives basal/squamous de-differentiation, inflammatory response, and immunosuppression.

Figure S1. Calibration of ESTIMATE tumor content prediction.