Transcriptomic Signatures of Monocyte‐Derived Macrophages Associate With Androgen Receptor Pathway Inhibitor Progression in Prostate Cancer

ABSTRACT

Background

Androgen receptor pathway inhibitors (ARPIs) have significantly improved clinical outcomes for patients with metastatic prostate cancer (PC) but acquired ARPI resistance remains universal. Maximizing ARPI treatment duration is crucial to optimal clinical outcomes, but current clinical tools to detect acquired ARPI resistance, including serum Prostate Specific Antigen (PSA) and radiographic disease monitoring, are limited in both sensitivity and specificity. Since prostate cancer disease progression is associated with an increase in systemic inflammation, we hypothesized that circulating monocytes and monocyte‐derived macrophages (MDMs) in patients with PC would express an increased pro‐inflammatory phenotype in the context of disease progression.

Methods

Monocytes and MDMs were isolated from peripheral blood samples from 16 patients with PC who were receiving ARPI therapies and performed transcriptomic and functional analysis both alone and in ex vivo coculture with prostate tumor cells utilizing a novel microscale coculture platform.

Results

We identified a pro‐inflammatory transcriptional signature in MDMs cultured with tumor cells that was associated with current, recent, and impending disease progression. Furthermore, we found that the pro‐inflammatory phenotype of MDMs derived from patients with clinical progression was associated with paracrine anti‐tumorigenic signaling that sensitized tumor cells to ARPI treatment in vitro. Finally, a transcriptional score generated from the MDM transcriptional signature of progressing patients could accurately identify current treatment response status as well as patients with recent/impending changes in response status.

Conclusions

Disease progression in patients with prostate cancer receiving ARPI therapy is associated with a pro‐inflammatory gene signature in peripheral monocyte‐derived macrophages. We were able to develop a scoring signature based on this pro‐inflammatory gene signature that has the potential to identify patients with recent and impending changes in disease response status that is not detectable using conventional disease assessment criteria. Further research will be needed to validate these findings.

1. Introduction

Androgen receptor pathway inhibitors (ARPIs) are effective and well‐tolerated therapies that have significantly improved survival for men with metastatic prostate cancer (PC) [1, 2]. However, as with all therapies in the metastatic setting, patients with PC receiving ARPI therapies eventually develop treatment resistance [2]. Since ARPIs are both tolerable and effective, maximizing treatment duration without treating beyond progression is a crucial component of clinical care for metastatic PC. Clinical determination of ARPI resistance currently relies on serum Prostate Specific Antigen (PSA) in conjunction with radiographic disease monitoring to determine when ARPI resistance has developed [3]. However, discordance between PSA trends and disease status has become increasingly recognized, which limits the reliability of PSA, particularly in later stages of disease [4, 5, 6]. While radiographic response patterns can also be utilized, impactful changes on imaging often take months to develop, resulting in delays in treatment decisions [6]. Optimizing the utility of ARPIs, therefore, depends on the development of new biomarkers that can improve clinical assessment of ARPI disease response status.

Disease progression in PC has been associated with an increase in systemic inflammation, which can occur very early in the course of progression [7, 8]. The increase in inflammation is a consequence of tumor cell dependence on pro‐inflammatory growth and angiogenesis factors as well as the expression of novel tumor antigens in the context of oncogenic and genomic changes related to treatment resistance [9, 10, 11]. This inflammatory response is driven by the multidirectional interactions between tumor, immune, and stromal cells within the tumor microenvironment (TME), which secrete pro‐inflammatory cytokines (i.e., IL‐6, IL‐8, TNF‐alpha) and chemokines (CCL2, CXCL9, CXCL10, etc.) as tumors progress [8, 12, 13, 14]. Circulating monocytes are a myeloid‐derived immune cell population in the peripheral blood that is highly sensitive to the changes in inflammatory signals that occur in conditions such as cancer [15, 16, 17, 18]. In response to increases in systemic inflammation, circulating monocytes express a pro‐inflammatory phenotype that can be assessed through peripheral blood draws [16, 17, 18]. These pro‐inflammatory changes also have the potential to impact the subsequent pro‐versus antitumor roles of these monocytes in cancer progression as well as the phenotype and roles of monocyte‐derived macrophage (MDM) cell populations, including tumor‐associated macrophages (TAMs), which have established roles in the development of therapeutic resistance [19, 20, 21, 22]. However, it is not yet known whether analysis of pro‐inflammatory phenotypic changes in circulating monocytes and MDMs could be utilized as a biomarker of disease response status for metastatic prostate cancers receiving ARPI treatment.

In this study, we performed transcriptomic and functional analysis of circulating monocytes and MDMs from 16 patients with PC who were receiving ARPI therapies. We hypothesized that patient‐derived monocytes and MDMs would express an increased pro‐inflammatory phenotype in the context of disease progression. By leveraging a previously validated microscale cell culture platform, known as Stacks, we were able to utilize a single 15 mL blood draw from each patient to perform multi‐analyte transcriptomic and functional analysis of monocytes and monocyte‐derived cell populations alone and in coculture with prostate tumor cells ex vivo [23, 24, 25, 26]. Our analysis identified a pro‐inflammatory signature in MDMs cultured with tumor cells that was associated with current, recent, and impending disease progression. Furthermore, we identified that MDMs derived from patients with clinical progression increased in vitro responses to enzalutamide treatment in tumor cell lines as compared to MDMs derived from patients with clinical response.

2. Methods

2.1. Patient Cohort

Blood samples were collected from patients with histologically confirmed metastatic prostate cancer at the University of Wisconsin Carbone Cancer Center (UWCCC) or William S. Middleton Memorial Veterans Hospital. The study was conducted in compliance with the Declaration of Helsinki. Patients were enrolled under an institutional IRB‐approved biospecimen protocol (1202–1214). Written informed consent was obtained from all participants before enrollment.

2.2. Stacks Devices

Before culture, Stacks plates (Protolabs, Maple Plain, MN, US #1121‐5161‐007) were prepared by sonication in 100% isopropanol for 60 min and washed in deionized water [26]. Before use, 3D holders, Nunc Omnitrays (Thermo Fisher Scientific, Waltham, MA, USA), and non‐tissue culture‐treated BioAssay dishes (245 mm square; Corning Inc., Corning, NY, USA) were sterilized by 70% ethanol wash followed by 20 min germicidal UV light treatment on each side. Cell suspensions were placed as a droplet onto the hydrogel matrix within the Stacks device and allowed to adhere or migrate through the matrix as applicable. When stacking two or more devices, media was removed from the top and bottom leaving only a small volume of residual media to prevent gas bubble formation during stacking. Stack devices were placed in a three‐layer humidifying chamber including a sterile sponge soaked in sterile ddH₂O in a Nunc Omnitray.

2.3. PBMC Isolation

Blood samples were collected from patient donors with prostate cancer after receiving written informed consent under a protocol approved by the Institutional Review Boards at the University of Wisconsin‐Madison (#2014‐1214) and at the William S. Middleton Memorial Veteran's Hospital, Madison, WI (#WI‐018). Research has been performed in accordance with the Declaration of Helsinki. Blood specimens were collected in vacutainer tubes (BD Biosciences, Franklin Lake, NJ, USA) with EDTA anticoagulant. Whole blood was diluted 1:1 with Hank's balanced salt solution (HBSS, Lonza Group, Basel, Switzerland) before being underlaid with 10 mL of Ficoll‐Paque PLUS (GE Healthcare, Cat# 45‐001‐750) for gradient centrifugation. CD14⁺ monocytes were enriched from PBMCs using LS MACS columns following incubation with anti‐CD14 magnetic beads (Miltenyi Biotec Inc., Bergisch Gladbach, North Rhine‐Westphalia, Germany). Monocyte purity was verified through analysis of bulk RNA‐seq using the NSCLC PBMC reference matrix provided by CIBERSORTx.

2.4. Cell Culture

22Rv1 and C4‐2B (C42B) cells were acquired from ATCC. The 22Rv1 cells were cultured in RPMI1640 media with l‐Glutamine (Corning Thermo Fisher Scientific, Waltham, MA, USA), 10% FBS (GibcoTM, Thermo Fisher Scientific, Waltham, MA, USA), and 2% penicillin/streptomycin (HycloneTM, VWR, Radnor, PA, USA). The C42B cells were cultured in DMEM/F12 at a 4:1 ratio (GibcoTM, Thermo Fisher Scientific, Waltham, MA, USA, #11966‐025 and #11330‐032, respectively) with 10% FBS, 0.1 µg/mL insulin (GibcoTM, Thermo Fisher Scientific, Waltham, MA, USA, #12585‐014), 275 ng/mL Triiodothyronine (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA #T5516), 88.6 ng/mL apo‐Transferrin (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA #T1147), 4.9 ng/mL d‐Biotin (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA #B4639), 251.8 ng/mL Adenine (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA #A‐3159), and 2% penicillin/streptomycin. Patient‐derived monocytes were cultured in RPMI1640 media with l‐Glutamine, 10% FBS, 5% Glutamax (Gibco, Thermo Fisher Scientific, Waltham, MA, USA), and 2% penicillin/streptomycin.

22Rv1, C42B, and monocyte derived macrophages (MDMs) were cultured in Stacks by seeding as a monolayer on a collagen‐fibronectin matrix, consisting of 79% collagen I (Advanced BioMatrix, Carlsbad, CA, USA #5005), 1.5% fibronectin (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA #F1141) prepared according to manufacturer guidelines. Moisture was retained by storing Stacks inside of a humidifying chamber. On Day 1, CD14⁺monocytes were seeded in Stack and differentiated into macrophages using 50 ng/mL colony stimulating factor 1 (CSF1) (Tonbo Biosciences, Cytek, San Diego, CA USA) with an initial seeding concentration of 3 × 10⁶ cells per mL per established protocols [24]. To maintain an unpolarized/M0 phenotype, 50 ng/mL CSF1 was readministered on Day 4 or 5 for an additional 3 days. On Day 4 or 5, single cell suspensions of C42B and 22Rv1 cells were harvested at log phase and were seeded onto Stacks hydrogel wells at a concentration of 1.5 × 10⁵ or 3 × 10⁵ cells per mL, respectively. On Day 7 or 8 of the culture period, an MDM or blank containing only hydrogel Stack was placed onto the 22Rv1 and C42B cultures. From this point forward, cultures had growth media appropriate for their respective cells (MDM mono‐cultures with RPMI1640 plus additives plus MCSF, 22Rv1 mono‐ and co‐cultures with RPMI1640 plus additives, C42B mono‐ and co‐cultures with DMEM, plus DMEM/F12, plus additives).

2.5. ARPI Cytotoxicity Assay

On Day 8 or 9 of the culture period, the culture media was exchanged with fresh media containing either 120 µM (22Rv1) or 90 µM (C42B) Enzalutamide, DMSO vehicle in media (1.2 to1000 for 22Rv1 or 0.9 to1000 for C42B), or only media. Treatment lasted for 72 h at which point the Stacks layers were separated for subsequent cell extraction and analysis. Upon separation of plates, tumor cell microwells were washed with 1× PBS to remove treatment, and then stained with a viability dye, Calcein‐AM (Invitrogen #C3099), in‐chip. Subsequently, cells were extracted from microwells by digesting the biomatrix with 1 mg/mL collagenase from Clostridium histolyticum (Sigma‐Aldrich #C9697‐50MG) and then transferring the released cells to a 384‐well low‐volume microplate. Individual Stacks microwells were transferred to their own unique microplate well. A CLARIOstar Plate Reader (BMG Labtech) in the UW‐Madison Small Molecule Screen Facility was used to capture the Calcein‐AM fluorescence intensity for each microplate well. At least three wells were run per condition, with each condition's average being used to calculate Relative Viability: (Experimental Intensity Average)/(DMSO Control Intensity Average) = Relative Viability.

2.6. Nucleic Acid Extraction

RNA was isolated from CD14⁺ cells and matched MDM cultures. CD14⁺ cells were subjected to QIAshredder cell lysis followed by RNA isolation using the Qiagen RNeasy Mini Kit and protocol (Qiagen, Hilden, Germany, #79656 and #74106, respectively). MDMs were extracted from hydrogel matrix following digestion with 1 mg/mL collagenase I (Sigma‐Aldrich, Millipore Sigma, Burlington, MA, USA) and RNA was isolated with the Qiagen RNeasy Micro Kit (Qiagen, Hilden, Germany, #74034), and QIAshredder columns. Three microwells of each condition, per replicate, were combined to obtain a sufficient number of cells. RNA for both cell extractions were stored at −80° C.

2.7. RNA Sequencing

RNA sequencing was performed by the UW Madison Biotechnology Core. Raw reads were aligned with STAR (v2.7.10a) and transcript abundance calculated using feature counts and TPM normalization [27, 28]. Gene expression data was used to perform gene set variation analysis (GSVA v1.52.2), using the Hallmark pathways from MSigDB, thereby providing biologically interpretable, pathway‐centric results with increased noise reduction and dimensionality reduction [29, 30].

2.8. Differential Gene Expression Analysis

DESeq. 2 (v1.38.3) was used to identify differentially expressed genes between responding and progressing coculture samples. Genes were ranked by log2FC for Gene Set Enrichment Analysis (GSEA) of the Hallmark pathways using the fgsea package (v1.24.0).

2.9. Identification of Signaling Pathways Related to Clinical Disease Status

We applied PCA‐based unsupervised feature selection (PCAUFE) to identify gene‐sets related to patient response as previously validated [31]. PCA loadings were calculated for an input table where features were patients, samples were cell‐specific pathways, and values were enrichment scores (Supporting Information S1: Figure S1). A two‐sided t‐test was applied to these loadings to specify which principal components (PC) exhibited statistical differentiation between patient responses, which were then selected for feature extraction. Finally, a chi‐squared test applied to the PC scores identified which scores, and therefore features, were outliers; these were subsequently selected as features for downstream analysis. Hierarchical clustering was applied to selected features to evaluate the relative expression for each patient and identify those with similar expression patterns.

2.10. Statistical Analysis

Gene expression analysis, plotting and statistical tests were performed using R v.4.0.4 and GraphPad Prism v9. All statistical tests described in the article were two‐sided. Hierarchical clustering was performed using Euclidean distance. Correlations were analyzed using Pearson's method.

3. Results

3.1. Transcriptomic Analysis of Patient‐Derived Monocytes/MDMs Identifies RNA Expression Patterns Associated With Clinical Disease Status on ARPI Therapy

Monocytes were isolated from the peripheral blood of patients with metastatic prostate cancer receiving ARPI therapy. Clinical status of each patient was determined by investigators, defined as clinically progressing if patient was experiencing a biochemical, radiographic, or symptomatic progression at the time of sample collection, or clinically responding if no evidence of biochemical, radiographic or symptomatic progression at the time of sample collection. A total of 16 patients were enrolled, seven in the clinically progressing group and nine in the clinically responding group (Tables 1 and S1). The majority of patients had high‐grade disease, and both CSPC and CRPC were represented, with a higher proportion of CSPC in the responding group (5 out of 9) than in the progressing group (2 out of 7). Sites of disease spread, primarily lymph nodes and bone, were similar between groups. All patients were receiving either abiraterone (6/7 with clinical progression and 3/9 with clinical response) or enzalutamide (1/7 with clinical progression, 6/9 with clinical response).

Table 1.

Clinical Characteristics.

Characteristic	Total cohort	Progressing	Responding
No. of patients	16	7	9
PSA at blood draw median (range), ng/mL	1.377 (< 0.02–283)	4.63 (0.33–283)	0.22 (< 0.02–1.52)
Grade group n (%)
2	1 (6)	1 (14)	0 (0)
3	1 (6)	0 (0)	1 (11)
4	4 (25)	2 (29)	2 (22)
5	7 (44)	3 (43)	4 (44)
Unknown	3 (19)	1 (14)	2 (22)
Type of disease (%)
Hormone sensitive	7 (43)	2 (28)	5 (55)
Hormone resistant	9 (56)	5 (72)	4 (44)
Metastatic sites, n (%)
Lymph node only	2 (12)	0 (0)	2 (22)
Bone only	7 (44)	4 (57)	3 (33)
Bone and lymph node	6 (37)	3 (43)	3 (33)
Visceral	1 (6)	0 (0)	1 (11)
ARPI, n (%)
Abiraterone	9 (56)	6 (86)	3 (33)
Enzalutamide	7 (43)	1 (14)	6 (66)

Abbreviation: ARPI, androgen receptor pathway inhibitor.

Given the potential for phenotypic changes at level of the circulating monocyte as well as in MDMs within the TME, we interrogated the transcriptional signatures of patient‐derived monocytes and MDMs to identify whether there were any global differences in monocyte/MDM transcriptional phenotypes between patients with clinical progression vs. response. RNA was extracted and sequenced from monocytes/MDMs in the following conditions 1) CD14⁺ monocytes directly after isolation (Monocyte), 2) monocytes differentiated into MDMs ex vivo as a monoculture (MDM mono), 3) MDMs Cocultured with 22rv1 prostate tumor cells for 96 h (MDM Co‐22) or 4) MDMs Cocultured with C42B prostate tumor cells for 96 h (MDM Co‐C4). 22rv1 and C42B prostate tumor cell lines were selected for coculture experiments as they are both androgen‐independent, which was consistent with the primary clinical context for ARPI use at the initiation of the study (Figure 1).

Figure 1.

Experimental Workflow. Patient‐derived CD14⁺ monocytes were isolated from peripheral blood specimens using Ficoll gradients and CD14 bead enrichment. Monocytes were then separated into two subsets. Subset 1 was processed for RNA purification while subset 2 was differentiated into monocyte‐derived macrophages (MDMs) in STACKs wells. MDMs were Cocultured with prostate cancer cells for 24 h before 72 h treatment with Enzalutamide or DMSO control (No Treatment). Following treatment, RNA was purified from MDMs and tumor cells were isolated for viability analysis. RNA from CD14⁺ monocytes and matched MDM sub‐cultures were subjected to bulk RNA sequencing. Tumor cell viability was measured using Calcein AM uptake. [Color figure can be viewed at wileyonlinelibrary.com]

We first evaluated for differences in selected genes associated with M1 (CXCL9, CXCL10, CXCL11, TNF) and M2 (CCL18, CCL22, IL10, MRC1) macrophage polarization between samples from patients with clinical progression vs. response (Supporting Information S1: Figure S2). This analysis demonstrated a complex pattern of gene expression that varied both by the response status of the patient as well as differentiation and coculture status of the cell population. Specifically, in the monocytes, we found higher expression of TNF, as well as trends towards increased expression of other M1‐associated genes including CXCL9 and CXCL11 in the responding patients. However, we also found trends towards increased expression of the M2‐associated genes, CCL22, IL10, and MRC1. When the monocytes were differentiated into macrophages, the MDMs from responding patients continued to trend towards increased expression of M1‐associated genes, including CXCL10, CXCL11, and TNF. However, the expression of M2‐associated genes became more mixed. MDMs from progressing patients expressed higher levels of CCL22 and trends towards higher expression of CCL18 and IL10, but lower expression of MRC1 than MDMs from responding patients. When the MDMs were then cultured with tumor cells, we again found a shift in the polarization profiles compared to the other monocyte/macrophage populations. In this setting, there was clear trends towards increased expression of the M1‐associated genes in CXCL9, CXCL10, and CXCL11 in the progressing patients and equalization of TNF expression between the two groups. However, the Cocultured MDMs from progressing patients also expressed higher levels of the M2‐gene, CCL22, as well as trended higher in CCL18 and IL10 expression. This complex pattern of gene expression in monocytes and macrophages, which does not clearly fall within a defined M1 or M2 phenotype, has been well established and is likely due to the presence of both pro‐inflammatory as well as anti‐inflammatory factors that affect individual gene expression in these cells [32, 33, 34].

To more globally assess differences in transcriptional phenotype between progressing and responding patients, gene set enrichment analysis was performed on all four monocyte/MDM samples from each patient utilizing the Hallmark gene sets from MSigDB. Unsupervised principal component analysis (PCA) of mean pathway enrichment scores across all samples and conditions demonstrated clustering of monocyte/MDM samples from patients with clinical progression across conditions, while samples from patients with clinical response were more heterogeneous (Figure 2). The clustering of the progressing patients suggested that disease progression on ARPIs was associated with a distinct transcriptional phenotype in the monocytes/MDMs from these patients. However, further investigation of gene set data was required to identify the specific cell conditions and gene sets driving the transcriptional patterns in monocytes/MDMs derived from progressing and responding patients.

Figure 2.

Principal component analysis of Hallmark pathway scores of MDMs across conditions in patients progressing on ARPI treatment. Principal component analysis scores for patients (points), as determined by enrichment scores for Hallmark pathways. Patient response to androgen receptor pathway inhibitors indicated by point color (Progressing = red, Responding = blue). [Color figure can be viewed at wileyonlinelibrary.com]

3.2. Progression on ARPI Therapy Is Associated With Increased Inflammatory Response and Decreased Proliferation Signatures in Patient‐Derived Mdms Cocultured With Tumor Cells

Since prostate cancer progression is associated with increased systemic inflammation, we hypothesized that the transcriptomic clustering of progressing patients was due to an increase in inflammatory pathway expression. We therefore performed PCAUFE across all sample conditions using the Hallmark pathways. This analysis demonstrated that progressing patients had significantly higher enrichment scores than responding patients for six hallmark pathways related to inflammatory response in at least one of the four monocyte/MDM conditions (Figure 3A), while responding patients had significantly higher enrichment scores than progressing patients in four hallmark pathways related to cell proliferation in at least one cell condition (Figure 3B). Specifically, significant differences were identified in all 10 of these pathways in the MDM Co‐22 condition while three of the inflammatory gene sets and one of the proliferation gene sets were significantly different in the MDM Co‐C4 condition. One proliferation gene set (MYC_TARGETS_V2) was significantly different in the MDM mono condition, and none were significantly different in undifferentiated monocytes (Supporting Information S1: Figure S4A,B). Gene Set Enrichment Analysis (GSEA) of all genes ranked by differential expression between responding vs. progressing patients for the Hallmark pathways identified very similar results to our PCAUFE pathway analysis, with enrichment of inflammatory gene sets in the progressing samples and proliferation gene sets in the responding samples (Supporting Information S1: Figure S5). Taken together, these findings indicate that while clinical progression on ARPIs was associated with complex alterations in pro‐ and anti‐inflammatory gene expression in monocytes and MDMs, there was an overall upregulation of inflammatory gene sets in MDMs from progressing patients and an overall increase in cell proliferation gene sets in the MDMs from responding patients when cocultured with tumor cells.

Figure 3.

Cocultured MDM Enrichment Scores associate with ARPI treatment response. Distributions of enrichment scores for indicated Hallmark pathways, grouped by patient response category. Enrichment scores are shown for MDMs that were cocultured with either C42B (Co‐C4) or 22Rv1 (Co‐22) tumor cells. (A) Gene sets associated with inflammatory response. (B) Gene sets associated with cell proliferation. Patient response to androgen receptor pathway inhibitors indicated by point color (Progressing = red, Responding = blue). Boxes represent mean enrichment scores for MDMs from each group. Whiskers represent standard error of the mean. * = p < 0.05. [Color figure can be viewed at wileyonlinelibrary.com]

3.3. Inflammatory Proliferation Scores of Cocultured MDMs Identifies Recent, Current, and Impending Disease Progression

Hierarchical clustering of the inflammatory and proliferation gene sets that were significantly different between samples from patients with ARPI progression vs. response identified three distinct clusters (Figure 4). MDM samples from patients in cluster A had overall low enrichment of inflammatory gene sets and high enrichment of cell proliferation gene sets. Cluster C had overall high enrichment of inflammatory gene sets and low enrichment of cell proliferation gene sets. Cluster B had intermediate enrichment of both gene set categories. To quantify these differences, we generated a consensus score reflecting the relative inflammatory and proliferation enrichment score (IP Score) for each patient by subtracting the average enrichment score of the proliferation gene sets from the average enrichment score of the inflammatory gene sets (Table 2). Thus, a higher IP score indicates a more inflammatory, less proliferative phenotype whereas a lower IP score indicates a less inflammatory, more proliferative phenotype. This scoring system provided a single quantitative description of the relative interaction between immune and proliferation signatures for each patient.

Figure 4.

Principal Component Analysis of patient‐specific enrichment scores. Clustered heatmap of patient‐specific enrichment scores (scaled by column) for features selected using PCA‐based unsupervised feature selection. Patients (columns) and cell conditions (rows) are annotated at the leaves of the dendrogram. [Color figure can be viewed at wileyonlinelibrary.com]

Table 2.

Immune proliferation index.

Patient ID	Cluster	Immune score	Proliferation score	IP index
463	A	−4.43	3.44	−7.87
707	A	−5.38	1.18	−6.56
686	A	−2.77	2.94	−5.71
016	A	−4.42	1.22	−5.65
397	A	−2.95	1.03	−3.98
693	B	−0.40	2.03	−2.42
610	B	−1.60	‐0.68	−0.91
532	B	−0.15	‐0.90	0.75
474	B	1.68	0.70	0.98
654	B	1.80	‐0.13	1.93
579	B	2.29	‐0.11	2.41
172	C	2.15	‐2.46	4.61
298	C	3.02	‐2.71	5.73
669	C	3.16	‐2.62	5.77
003	C	3.33	‐3.36	6.69
361	C	4.32	‐2.68	7.00

Note: Table 2: Inflammatory‐proliferation index scores and associated Cluster for each patient.

We next evaluated whether this consensus MDM IP Score was associated with ARPI progression vs. response status (Figure 5). We found that IP Scores above 0 were observed in all but one of the patients with clinical progression, while patients with clinical response were more likely to have IP Scores below 0. Importantly, while four patients had an IP Score that did not fit this pattern (one with progressive disease and three with responding disease), in every case this was found to be in the setting of a recent or impending change in clinical status. Specifically, among the three patients in the responding group with IP Scores above 0, two had experienced disease progression within the prior 4 weeks and were currently responding to treatment with radiotherapy to sites of progression. The third patient developed symptomatic and radiographic progression of his disease within 2 weeks of the sample collection. This progression was later confirmed to be a transition to small cell neuroendocrine prostate cancer (NEPC).

Figure 5.

Distributions of IP Index Scores by clinical response category. Patients were grouped by clinical response on x‐axis. Left y ‐axis represents the IP Index score for each patient. Right y‐axis represents clinical status as predicted by IP Index score. Colored horizontal line represents the average IP Index Scores for the patients in the corresponding conventional response category. Whiskers represent standard error of the mean. Blue color corresponds to Responding group and Red color corresponds to progressing group. [Color figure can be viewed at wileyonlinelibrary.com]

In the progressing group, the one patient with an IP Score below 0 was receiving a first‐generation androgen receptor inhibitor (bicalutamide) in addition to abiraterone therapy and was experiencing biochemical progression. Following discontinuation of the anti‐androgen, PSA levels decreased to undetectable, indicating that PSA rise was due to anti‐androgen effect and not due to the development of treatment resistant disease. This data suggests that the patient‐derived MDM IP Score may be able to identify patients with recent and impending progression as well as patients who are currently progressing on ARPIs. It also suggests that pseudo‐progression related to anti‐androgen therapy treatment may result in a different systemic inflammatory signature than true disease progression and that the IP Score may be able to identify this signature.

3.4. Macrophages Derived From Patients With Clinical Progression Increase Tumor Cell Sensitivity to Treatment

Macrophages with diverse phenotypes are found in the TME and have been shown to influence tumor progression and treatment response. In particular, pro‐inflammatory macrophages in the prostate TME have been shown to be tumor suppressive, while anti‐inflammatory macrophages have been shown to play a tumor promoting role [32, 35, 36, 37]. Since our data demonstrated an increase in pro‐inflammatory gene set expression in MDMs from patients with clinical progression after coculture with prostate tumor cell lines, we assessed whether these pro‐inflammatory MDMs, which would be anticipated to have a tumor suppressive function, would therefore increase tumor cell sensitivity to ARPI treatment in vitro.

Each patient‐derived MDM coculture model was treated with enzalutamide for 72 h followed by relative viability analysis of the cocultured tumor cells. We then plotted the IP Score of each patient‐derived MDM sample against the combined average viability of the C42B and 22rv1 cells following coculture with each patient's MDM and enzalutamide treatment (Figure 6). This analysis demonstrated a clear correlation (Pearson: ρ = −0.5815437, p = 0.001465, Spearman: ρ = −0.6567576, p = 0.0001985) between the IP Score of the MDMs and the enzalutamide sensitivity of cocultured tumor cells, with higher IP Scores associated with lower tumor cell viability. This same pattern, while not statistically significant, was also evident when the viability of each tumor cell line was analyzed independently (Supporting Information S1: Figure S6). These collective findings suggest that in patient‐derived MDMs, higher IP Scores are directly related to MDM secretion of tumor suppressive paracrine factors that increase tumor sensitivity to enzalutamide.

Figure 6.

Impact of MDM IP index score on tumor cell sensitivity to enzalutamide. 22Rv1 and C42B cells were individually cocultured with MDMs from each patient and treated with enzalutamide for 72 h. Wells were then analyzed for relative viability (3 replicates were analyzed for each patient). Viability data from the 22Rv1 and C42B cells were combined to generate an average tumor cell viability that reflected data from each coculture condition. Points represents the Immune Proliferation Index Score of the MDMs for each patient (x‐axis) and corresponding average viability of the cocultured tumor cells following enzalutamide treatment (y‐axis). Pearson's ρ and p‐value annotated in top right. Point colors correspond to hclust group as indicated by legend. [Color figure can be viewed at wileyonlinelibrary.com]

4. Discussion

In this study, we isolated peripheral monocytes from a cohort of 16 patients with metastatic prostate cancer receiving ARPI therapy to investigate the transcriptional phenotypes of monocytes and MDMs. Through in vitro coculture with two separate prostate tumor cell lines, we identified pro‐inflammatory transcriptional signatures in MDMs at the time of clinical progression on ARPI therapies. To leverage this association of MDM phenotype with clinical treatment response status, we generated an IP Score that reflected the transcriptional MDM signature of progressing patients. In our cohort, we found that utilization of the IP Score with a threshold of 0 for progression (above) and response (below) could sensitively and accurately identify current treatment response status as well as identify patients with recent/impending changes in response status. Importantly, this included 4/16 patients who were experiencing either (1) early disease progression not yet detected with standard clinical assessment, (2) treatment response in the setting of recent progression and (3) serum PSA rise due to treatment related effect. These findings provide preliminary evidence that the data from the IP score could supplement currently available biomarkers in the clinic to enhance assessment of disease response status.

Prior studies have already established that a higher proportion of pro‐inflammatory (also known as M1‐like) TAMs within the TME is associated with increased tumor sensitivity to treatment and better survival outcomes in patients [20, 36, 37, 38, 39]. In addition, several reports have also demonstrated that an array of cancer therapies are associated with an increase in pro‐inflammatory TAMs, which can enhance treatment response [22, 35, 40, 41]. However, we are not aware of any studies that have compared how the phenotypes of TAMs change during the development of disease progression. Our data suggests that as disease progression occurs, increases in systemic inflammation may precondition circulating monocytes to a pro‐inflammatory tumor suppressive phenotype that secretes paracrine factors that can sensitize tumor cells to treatment. This concept would be consistent with the established data demonstrating increased systemic inflammation during PC progression as well as the known association of pro‐inflammatory TAMs with enhanced treatment response [7, 8, 22, 35]. The pro‐inflammatory preconditioning of monocytes could potentially be an adaptive immune response to attempt to suppress disease progression. Consistent with this hypothesis, we found that higher IP Scores in cocultured MDMs were associated with a tumor suppressive MDM phenotype that increased tumor cell sensitivity to enzalutamide treatment through paracrine signaling.

Interestingly, we found that systemic factors in progressing patients did not induce an increased pro‐inflammatory phenotype in their circulating monocytes, but did induce a pro‐inflammatory phenotype in the macrophages that were derived from those monocytes. One possible explanation is that there were inflammatory changes in the monocytes but that we were not able to detect them because we performed transcriptomic analysis on all peripheral monocytes as a single population. It is well established that there is an array of monocyte subpopulations that comprise the total monocyte population in the blood and multiple studies have previously demonstrated that systemic factors in the setting of cancer results in expansion of specific monocyte subpopulations rather than global changes in the entire circulating monocyte population [42, 43, 44, 45, 46]. Although there is variation in the subpopulation of monocytes that expands in each disease setting, these expanded subpopulations can each have unique gene expression profiles [45, 46, 47]. Therefore, there may have pro‐inflammatory transcriptomic changes in certain monocyte subpopulations in progressing patients that we did not detect in our evaluation of the entire CD14+ cell population. It is possible that such inflammatory monocyte subpopulations in the blood then differentiated into macrophages with augmented inflammatory responses in the presence of tumor cells, which is why we were able to identify differences in progressing and responding patients at the macrophage level. However, a larger prospective study with single cell analysis of monocyte and macrophage subpopulations is needed to explore in more detail the heterogeneity of circulating monocyte/macrophage phenotypes and clinical response in the absence of tumor cell coculture.

Our findings also do not explain the paradox of why pro‐inflammatory MDMs enhance enzalutamide sensitivity but are associated with disease progression in patients. While additional research will certainly be necessary to explain this paradox, the answer may lie in the differences between the in vivo tumor conditions and the in vitro models that were employed in this study. In addition to tumor cells, TAMs in vivo communicate with an array of additional TME cell populations, include fibroblasts, endothelial cells, lymphocytes, and other innate immune cells [32]. These diverse interactions are known to play vital roles in TAM phenotype and function within tumors [32, 41, 48, 49, 50, 51, 52]. Tumors also contain an array of chemical and mechanical stimuli, such as pH, oxygen tension, and extracellular matrix components that are known to influence TAM phenotype as well [34, 53, 54, 55, 56]. In this study, our models included only the macrophages and tumor cells, and did not include other TME cell populations or tumor‐specific chemical and mechanical stimuli. Therefore, the disconnect between the tumor‐destructive functions of the TAMs in our assays and the tumor‐progressive status of the associated patients may be due to the impact of the additional TME cell populations and chemical/mechanical stimuli in vivo. In addition, our prior data has demonstrated that in the presence of pro‐inflammatory tumor T cells, M2‐polarized MDMs upregulate expression of M1 cytokines while retaining elevated expression of M2 cytokines. In the context of T cell responses, these MDMs behaved like M1 macrophages despite the elevated expression of both M1‐ and M2‐associated genes [57]. These findings therefore indicate that the role of MDMs in the context of tumors cannot easily be defined by analysis of select genes. In either scenario, our data does suggest that MDMs derived from progressing patients have the potential to support the antitumor effects of hormone therapies. Therefore, there may be a targetable pathway in MDMs that could be exploited to promote the treatment effects of these therapies to induce a pro‐inflammatory shift in TAM/MDM phenotype.

The data from our study suggests that the IP Score may hold promise as a potential novel noninvasive biomarker that could provide additional clinical information to guide ARPI treatment decisions. In addition, our findings have identified a compensatory response to disease progression that involves pro‐inflammatory preconditioning of monocytes. However, definitive conclusions based on this data are limited by the small and heterogeneous nature of our patient cohort. Additionally, androgen‐independent prostate tumor cell lines were chosen for coculture because at the time of enrollment, ARPIs were primarily utilized in the CRPC setting. Due to emerging data, this practice shifted during the course of enrollment to include ARPI treatment for CSPC patients as well [1, 2]. To validate our findings and establish the IP Score as a biomarker that can provide clinically meaningful information on disease status that is not already provided by current biomarkers, a large prospective trial will be required. It will also be necessary to characterize whether androgen‐sensitive tumor cell lines as well as matched patient tumor cells from progressing patients can induce the same MDM pro‐inflammatory/antitumor phenotype. Finally, further studies are needed to identify the underlying mechanisms by which tumor cell coculture induces this MDM phenotype, including the role of secreted paracrine factors or direct cell‐cell interactions.

Author Contributions

E.H., M.T.B., E.E.R., and D.K. designed experiments. E.E.R., M.T.B., E.H., M.L.B., M.N.S., and D.K. performed data acquisition and analyzed data. M.N.S. and D.K. drafted the text and figures. M.N.S., E.H., M.L.B., E.E.R., M.T.B., A.K.T., S.G.Z., A.M.L., and D.K. revised the manuscript for intellectual content.

Ethics Statement

The study was conducted in compliance with the Declaration of Helsinki. Patients were enrolled under institutional IRB‐approved biospecimen protocols (1202‐1214 and 2020‐0915).

Consent

Written informed consent was obtained from all participants before enrollment.

Conflicts of Interest

S.G.Z. reports unrelated patents licensed to Veracyte, and that a family member is an employee of Artera and holds stock in Exact Sciences. M.N.S. reports unrelated research support from Novartis.

Supporting information

Sharifi et al Prostate R1 supplementary.

Data Availability Statement

Our institutional biospecimen collection protocol does not allow unrestricted public access to the raw sequencing data to maintain protection of patient privacy. Therefore, data sharing requests must be submitted to the University of Wisconsin‐Madison for review and approval.