CX3CL1-CX3CR1 signaling orchestrates malignant progression of prostate cancer through luminal progenitor-macrophage crosstalk

All animal experiments were conducted in accordance with the ethical guidelines approved by the Institutional Animal Care and Use Committee (IACUC) of Tsinghua University (Approval No. 13-LZH1) and complied with relevant national and international guidelines governing the use of laboratory animals. Male C57BL/6J mice (8–12 weeks old) were housed under specific pathogen-free (SPF) conditions in a controlled environment (21 ± 1 °C, 50 ± 10% humidity, 12 h light/12 h dark cycle) with free access to standard rodent diet and autoclaved water. Animal group assignments and sample sizes for each experiment are detailed in the corresponding figure legends. Adult (8–12 weeks) male and female CX3CR-1^GFP mutant mice (Stock# 005582, Jackson Laboratories) were used for this study. PbsnCre; Pten^f/f mice were generously provided by Professor Geng Liu from Nanjing University.

RM-1: The castration-resistant prostate cancer (CRPC) cell line RM-1 (Cat# CL-0198, Pricella Life Science & Technology, Wuhan, China) was cultured in Dulbecco’s Modified Eagle Medium (Cat# 11965092, Gibco, New York, USA) supplemented with 10% fetal bovine serum (FBS, Cat# 10091148, Gibco, Auckland, New Zealand) and 1% penicillin/streptomycin (Cat# 15140122, Gibco, New York, USA). Cells were maintained at 80–90% confluence and routinely passaged using 0.25% trypsin (Cat# 15090046, Gibco, New York, USA).

Raw264.7: Mouse monocyte-macrophage leukemia cell line (Cat# CL-0190, Pricella Life Science & Technology, Wuhan, China) was cultured in RPMI 1640 (Cat# 11875093, Gibco, New York, USA) supplemented with 10% fetal bovine serum (FBS, Cat# 10091148, Gibco, Auckland, New Zealand) and 1% penicillin/streptomycin (Cat# 15140122, Gibco, New York, USA). Cells were maintained at 80–90% confluence and routinely passaged using 1–3 mL of complete culture medium to dislodge adherent cells from the flask walls.

For stimulation experiments, BMDMs were treated with 100 ng/mL recombinant Fractalkine/CX3CL1 (Cat# HY-P72686, MCE, USA) for 2 h in RPMI 1640 containing 10% FBS, 1% penicillin/streptomycin, and 20 ng/mL M-CSF. Subsequently, cells were exposed to either 100 ng/mL LPS (Cat# 437627, Sigma–Aldrich, USA) or 20 µg/mL IL-4 (Cat# ab9729, Abcam, USA), and RNA was extracted 24 h later.

RM-1 cells were transfected with siRNA using Lipofectamine 2000 (Cat# CA92008, Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Briefly, cells were seeded in 12-well plates at a density of 2 × 10⁵ cells per well and cultured overnight to reach 60–70% confluence. For each well, 97.5 µL of RPMI 1640 medium was mixed with 2.5 µL of Lipofectamine 2000 reagent. Simultaneously, 97.5 µL of RPMI 1640 was mixed with 2.5 µL of Cx3cl1 siRNA (Cat# HY-RS16603, MCE, USA) or negative control siRNA. The Lipofectamine 2000 mixture and siRNA mixture were then combined and incubated at room temperature for 15–20 min to allow complex formation. The transfection complexes (200 µL total volume) were added dropwise to cells in fresh culture medium. The cells were incubated at 37 °C in a humidified atmosphere with 5% CO₂. The transfection efficiency was assessed at 12- and 24-hours post-transfection by RT–qPCR analysis of target gene expression, and cells were then harvested for subsequent experiments.

For in vivo studies, RM-1 cells were harvested, washed, and resuspended in phosphate-buffered saline (PBS) at a density of 2 × 10⁶ cells/mL. In the ectopic tumor model, 100 µL of cell suspension (containing 2 × 10⁵ cells) was injected subcutaneously into the left flank of anesthetized C57BL/6J mice. For the orthotopic model, a midline laparotomy was performed to expose the ventral prostate, and 20 µL of cell suspension (1 × 10⁵ cells) was injected into the prostate capsule using a 30-gauge needle. The abdominal wall was sutured with 5 − 0 silk, and mice were monitored until recovery. All animals were euthanized three weeks post-injection; tumors were excised, and volumes were calculated as (length × width²) / 2.

Total RNA was extracted using Trizol reagent (Cat# 15596026CN, Invitrogen, California, USA) according to the manufacturer’s protocol. cDNA was synthesized using the PrimeScript RT Reagent Kit (Cat# RR037A, TaKaRa, Dalian, China). Quantitative PCR was performed using SYBR Premix Ex Taq (Cat# RR420A, TaKaRa, Dalian, China) on an ABI7500 Real-Time PCR system (Cat# ABI7500, Life Technologies, California, USA). Thermal cycling conditions were as follows: 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 34 s. Gene expression was normalized to Gapdh and analyzed using the 2^–ΔΔCt method. The sequences of the primers used are listed below:

Tumor tissues were minced and dissociated enzymatically in a digestion cocktail containing 2 mg/mL collagenase type IV (Cat# 17104019, New York, Gibco, USA) and 1 U/µL DNase I (Cat# 18047019, California, Sigma-Aldrich, USA) at 37 °C for 30–45 min. Single-cell suspensions were obtained by filtering through a 70-µm cell strainer (Cat# BS-70-CS, Biosharp, Anhui, China) and subjected to red blood cell lysis using RBC Lysis Buffer (Cat# 420302, Biolegend, California, USA). After washing, cells were resuspended in FACS buffer (PBS containing 2% FBS) and stained with the following fluorescently conjugated antibodies for 30 min at 4 °C in the dark: anti-CD11b-FITC (Cat# 554982, BD Biosciences, New Jersey, USA) and anti-F4/80-APC (Cat# MA5-16625, eBioscience, California, USA). Following two rinses, cells were resuspended in PBS and analyzed on a BD FACS Canto II flow cytometer (BD Biosciences, New Jersey, USA). Data analysis was performed using FlowJo software (v.10.8.1, Tree Star, Oregon, USA).

For immunohistochemical analysis, 6-µm-thick sections from paraffin-embedded tissues were deparaffinized and rehydrated. Antigen retrieval was conducted using citrate buffer (pH 6.0) in an electric steamer. Subsequent staining steps were carried out in accordance with the instructions of the immunohistochemistry kit (Cat# SPN-9002, ZSGB-BIO, Beijing, China). Briefly, endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 15 min, and non-specific binding sites were blocked with 10% normal goat serum for 30 min at room temperature. Sections were then incubated overnight at 4 °C with primary antibodies against Ki67 (1:500; Cat# ab15580, Abcam, USA) or cleaved caspase-3 (1:400; Cat# 9664s, Cell Signaling Technology, USA). After washing, sections were incubated with a biotinylated secondary antibody (Cat# 31460, Invitrogen, USA) for 1 h at room temperature, followed by detection using a streptavidin-biotin-peroxidase complex and a 3,3′-diaminobenzidine (DAB) substrate kit (Cat# PV-6000D, ZSBIO, Beijing, China). Finally, sections were counterstained with hematoxylin, dehydrated through a graded ethanol series, cleared in xylene, and mounted with neutral balsam (Cat#G8590, Solarbio, Beijing, China). Images were acquired using a Nikon Eclipse Ti-S microscope equipped with a DS-Ri2 camera.

For in vitro immunofluorescence, RM-1 cells were plated on glass coverslips at 5 × 10⁴ cells/well, adhered for 24 h, and serum-starved for another 24 h. Cells were fixed in 4% PFA for 10 min, permeabilized with 0.1% PBST, and blocked with 10% goat serum. Subsequent incubations included primary antibodies overnight at 4 °C and fluorophore-conjugated secondary antibodies (1:1000, Cat# 31460, Invitrogen, USA) for 1 h at room temperature. Nuclei were stained with DAPI before mounting with Fluoromount-G (Cat# 010001, SouthernBiotech, Alabama, USA).

For multiplex immunofluorescence staining of paraffin-embedded tissue sections, a tyramide signal amplification (TSA) staining kit (Cat# NECC4100, Histova Biotechnology, Beijing, China) was employed. Sections (6-µm thick) were deparaffinized, rehydrated, and subjected to antigen retrieval. After blocking, samples were incubated overnight at 4 °C with primary antibodies, followed by appropriate secondary antibodies and TSA reagent according to the manufacturer’s protocol. Antibody complexes were eluted between successive staining cycles to enable sequential labeling of multiple antigens. Nuclei were counterstained with DAPI, and slides were mounted with Fluoromount-G (Cat# 010001, SouthernBiotech, Alabama, USA). Imaging was performed using a confocal laser scanning microscope (Nikon A2, Japan). The following primary antibodies were used: anti-F4/80 (1:200, Cat# 70076, Cell Signaling Technology, USA), anti-CD206 (1:300, Cat# 24595, Cell Signaling Technology, USA), anti-CX3CR1 (1:250, Cat# ab250888, Abcam, USA), anti-CX3CL1 (1:200, Cat# 67735-1-Ig, Proteintech, Wuhan, China), anti-CD80 (1:300, Cat# 98958, Cell Signaling Technology, USA), anti-Ki67 antibody (1:500, Cat# ab15580, Abcam, USA), and anti-SOX9 antibody (1:400, Cat# 82630, Cell Signaling Technology, USA), anti-Epcam antibody (1:200, Cat# 21050-1-AP, Proteintech, USA), anti-CD4 antibody (1:200, Cat# 25229, Cell Signaling Technology, USA), anti-CD8 antibody (1:500, Cat# 98941, Cell Signaling Technology, USA), and anti-CD3 antibody (1:500, Cat# 78588, Cell Signaling Technology, USA).

Publicly available genomic datasets were obtained from the following sources: single-cell RNA sequencing (scRNA-seq) data from the Gene Expression Omnibus (GEO) and the Broad Institute Single Cell Portal; bulk RNA sequencing data from The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) cohort via the UCSC Xena browser.

Human datasets: The study incorporated the following human prostate cancer cohorts: [1] GSE181294, containing scRNA-seq data from 18 untreated primary prostate tumors and 5 benign prostate samples; [2] GSE274229, comprising scRNA-seq profiles from 6 metastatic castration-resistant (mCRPC), 25 metastatic hormone-sensitive (mHSPC), and 13 localized prostate cancer samples; [3] SCP1415, a scRNA-seq cohort of human prostate tumors; and [4] TCGA-PRAD, including bulk RNA-seq data from 501 primary tumors and 52 matched normal tissues.

Mouse datasets: Mouse scRNA-seq datasets included: [1] GSE146811, representing healthy prostate tissues under androgen manipulation; and [2] GSE262893, containing profiles from a murine prostate cancer model.

All datasets were processed using standardized quality control, normalization, and analysis pipelines appropriate for each data type before integrative analysis.

Analysis of scRNA-seq data was conducted using the Seurat package (v4.3.0) in R. Low-quality cells and doublets were filtered out using DoubletFinder (v2.0.3). Following normalization and scaling, principal component analysis (PCA) was performed for dimensionality reduction. The top significant principal components were selected for subsequent uniform manifold approximation and projection (UMAP) visualization and graph-based clustering. Cell types were annotated according to established marker genes. Differential expression analysis between defined cell clusters or conditions was carried out using the FindMarkers function with a Wilcoxon rank-sum test. Gene set enrichment analysis (GSEA) and gene ontology (GO) enrichment analysis were performed using the clusterProfiler package (v4.12.6). Cell-cell communication analysis was inferred using CellChat (v1.6.1). Immunosuppression signature scores were computed for individual cells using the Seurat “AddModuleScore” function, based on a curated gene set derived from published immunosuppression-related signatures [26].

All experiments were performed with at least three independent biological replicates, as indicated in the figure legends. Statistical analyses were conducted using R software and GraphPad Prism 8. Data normality was assessed using the Shapiro-Wilk test. Normally distributed data are presented as mean ± Standard Error of the Mean (SEM), and non-normally distributed data are expressed as median with interquartile range. Comparisons between two groups were performed using an unpaired two-tailed Student’s t-test (for parametric data) or the Mann-Whitney U test (for non-parametric data). Comparisons between three or more groups were performed using Tukey’s multiple comparisons test. Statistical significance was defined as a p-value ≤ 0.05. Exact p-values are provided in the figures and figure legends.

Our analysis of TCGA-PRAD data revealed a significant association between the CX3CL1-CX3CR1 axis and prostate cancer progression (Fig. S1A). This finding prompted us to employ an integrative phenomic approach, analyzing public scRNA-seq datasets to define the bidirectional interplay between CX3CL1⁺ and CX3CR1⁺ cells across stages of tumor development (Fig. S1B). The GSE274229 cohort includes 44 PCa samples across multiple disease stages with epithelial cell enrichment [26], and the GSE181294 dataset contains five healthy and 18 localized PCa samples with broad immune cell representation [27] (Fig. 1A, B; Fig. S1C-F).

Analysis of these datasets revealed that CX3CL1 was expressed across stromal and epithelial cell types, with the highest levels detected in the progenitor luminal 2 epithelial subpopulation (Fig. 1A, B; Fig. S1G, I). This subpopulation was defined according to the luminal progenitor gene set described across several independent reports [4, 27, 28]. Supporting this classification, violin plots show the expression of Tacstd2, Krt4, Ly6d, and Sox9, together with the corresponding Karthaus luminal progenitor module scores, confirming the enrichment of CX3CL1 within this luminal progenitor-like population (Fig. S1N-P). Conversely, CX3CR1 expression was predominantly restricted to the immune compartment, particularly within myeloid cells, where over 80% of CX3CR1⁺ cells were macrophages (Fig. 1A, B; Fig. S1E-H). The proportion of CX3CR1⁺ macrophages was significantly elevated in tumor tissues (1.83%, 1,168 of 63,693 cells) compared to normal prostate samples (0.86%, 155 of 17,943) (Fig. 1B, Fig. S1I).

This expression pattern was conserved in murine models. Evaluation of scRNA-seq data from normal mouse prostate (GSE146811) and PbsnCre; Pten^f/f spontaneous tumors (GSE262893) showed a marked expansion of luminal 2 cells in tumors (52.1%, 17,477 of 33,553 cells) relative to normal tissues (2.5%, 429 of 17,356 cells). Concurrently, the proportion of CX3CR1⁺ macrophages increased from 3.22% (559 of 17,356 cells) in normal tissue to 4.40% (1,478 of 33,553 cells) in tumors (Fig. 1C; Fig. S1J, K).

We validated these findings at the protein level by immunofluorescence staining in orthotopic tumors generated from the RM-1 CRPC cell line. In normal prostate epithelium, EpCAM was predominantly localized to the cell membrane and showed minimal co-localization with CX3CL1. In contrast, prostate tumor cells displayed a distinct nuclear localization pattern of EpCAM, consistent with previous reports [29], and a markedly increased fraction of EpCAM⁺ cells exhibited co-localization with CX3CL1 (Fig. 1D). Notably, this co-localization was observed with a tendency toward enrichment in EpCAM⁺ SOX9⁺ prostate cancer progenitor–like (luminal 2) cells (Fig. S1L, M).

Immunofluorescence staining confirmed an approximately threefold increase in F4/80⁺ CX3CR1⁺ macrophages in tumor tissues compared to normal tissues (10.45% vs. 3.56%, p < 0.0001; Fig. 1E-G). Together, these results demonstrate that the PCa TME is characterized by the coordinated accumulation of CX3CL1-expressing luminal 2 cells and their putative targets, CX3CR1⁺ macrophages.

Having established luminal 2 stem-like cells and macrophages as the primary expressors of CX3CL1 and CX3CR1, respectively, we next asked whether these populations interact via the CX3C signaling pathway. Cell-cell communication analysis (CellChat) of the SCP1415 dataset identified CX3C signaling as a dominant network, with luminal 2 cells serving as the main senders and macrophages as the principal receivers (Fig. 2A, B; Fig. S2A). Supporting this, Gene Set Enrichment Analysis (GSEA) of CX3CL1^hi luminal 2 cells in the GSE274229 dataset revealed significant enrichment of pathways related to cell chemotaxis (Fig. 2C, D). Correspondingly, CX3CR1^hi macrophages from spontaneous PCa mouse models showed enrichment for gene sets involved in directed migration and locomotion (Fig. 2E, F).

To functionally validate these predictions, we performed transwell migration assays using BMDMs. We silenced CX3CL1 in RM-1 cells by siRNA and found that conditioned medium from CX3CL1-silenced RM-1 cells markedly reduced macrophage migration at both 12 h and 24 h compared with control RM-1 conditioned medium (both p < 0.0001), demonstrating that tumor-cell-derived CX3CL1 is required for macrophage chemotaxis in vitro (Fig. S2C). Conditioned medium from RM-1 cells induced strong migration of WT BMDMs, whereas migration was significantly impaired in Cx3cr1^−/− BMDMs (Fig. S2B), confirming the necessity of CX3CR1 for tumor-cell-mediated macrophage recruitment.

Upon recruitment into the tumor microenvironment, macrophages encounter diverse polarizing signals, including cytokines from tumor and stromal cells (e.g., IL-4, IL-13, TGF-β), apoptotic cell debris, metabolic products including lactate and lipid intermediates, hypoxia, and direct cell–cell interactions. These signals collectively reprogram transcriptional and metabolic pathways to promote tissue remodeling, angiogenesis, and immunosuppressive functions, thereby favoring M2-like macrophage states. Given this pro-M2 environment, we next sought to characterize the phenotype of recruited macrophages in orthotopic RM-1 tumors.

Immunofluorescence analysis showed that tumor-infiltrating CX3CR1⁺ TAMs were predominantly polarized towards an M2-like, pro-tumoral phenotype, marked by CD206 expression. The proportion of CD206⁺CX3CR1⁺ TAMs was more than tenfold greater than that of CD80⁺ CX3CR1⁺ M1-like TAMs (4.86% vs. 0.43%; Fig. 2G-I). In vitro, addition of recombinant CX3CL1 to WT BMDMs promoted M2-like polarization with little effect on M1-like polarization. Notably, this pro-M2 effect was CX3CR1-dependent, as recombinant CX3CL1 did not enhance M2 polarization in CX3CR1-knockout BMDMs (Fig. S2D).

Together, these data suggested that luminal 2-derived CX3CL1 recruits CX3CR1⁺ macrophages into TME, where they adopt a pro-tumoral phenotype.

To determine how the interaction between CX3CL1^hi luminal 2 cells and CX3CR1^hiTAMs evolves during disease progression, we analyzed scRNA-seq data from the GSE274229 cohort, which includes samples from localized PCa, metastatic hormone-sensitive prostate cancer (mHSPC), and mCRPC. CellChat analysis confirmed active CX3CL1-CX3CR1-mediated communication across all stages (Fig. 3A, B). As the disease advanced, the proportion of CX3CL1^hi luminal 2 cells decreased, from 28.6% in localized PCa to 14.3% in mCRPC (Fig. 3C). Two‑dimensional signaling space visualization revealed a shift in communication patterns: luminal 2 cells served as the dominant signal sender in early-stage disease, whereas CX3CR1^hi TAMs became a prominent signal source in mCRPC (Fig. 3D-F). GO analysis further indicated stage-specific functional adaptations in luminal 2 cells, which were enriched in ribosome biogenesis pathways in mHSPC and shifted toward extracellular signaling and multicellular interaction pathways in mCRPC (Fig. S3A, B).

We next examined these dynamics under androgen deprivation therapy (ADT) pressure using a murine scRNA-seq dataset (GSE146811) that tracks prostate tissue through castration-induced atrophy and subsequent dihydrotestosterone (DHT) -induced regeneration. This model mimics the transition from HSPC to CRPC. CellChat analysis showed sustained unidirectional signaling from luminal 2 cells to Cx3cr1^hi macrophages throughout the castration–regeneration cycle (Fig. 3G, H). Luminal 2 cells expanded during castration-induced atrophy, peaking at day 28 post-castration, whereas Cx3cr1^hi macrophages declined initially but rebounded during the DHT-mediated regeneration (Fig. 3I-L). These results illustrate a coordinated, stage-dependent interplay in which luminal 2 cells dominate early TME organization, while TAMs assume a more active signaling role in advanced, treatment-resistant disease.

We next characterized the functional phenotype of CX3CR1⁺ TAMs across disease stages. In treatment-naïve human PCa, CX3CR1^hi TAMs accounted for 33.4% of all TAMs and were predominantly composed of lipid-associated macrophages (LA-TAMs, 25.1%) and angiogenic macrophages (7.7%) (Fig. 4A, B, Fig. S4A-B). A similar pro-tumoral composition was observed in the Pten-deficient mouse models, where Cx3cr1^hi TAMs expanded to 77.8% of the TAM compartment, with LA-TAMs (40.8%) and angiogenic TAMs (31.0%) representing the major subsets (Fig. 4C-F), indicating cross-species conservation of a pro-metabolic and pro-angiogenic phenotype.

Following ADT, the abundance and phenotype of Cx3cr1^hi TAMs underwent marked changes. Their proportions declined to 20.3%, 7%, and 7.7% in localized, mHSPC, and mCRPC samples, respectively. These TAMs predominantly clustered into LA-TAM and regulatory TAM (Reg-TAM) subtypes (Fig. 4G-J, Fig. S5A). In mCRPC, pathway analysis revealed significant enrichment of cholesterol transport and storage processes in Cx3cr1^hi TAMs (Fig. S5C-D), aligning with their LA-TAMs identity and suggesting a role in metabolic support of tumor growth.

Beyond metabolic reprogramming, these TAMs also displayed strong immunosuppressive properties. Cx3cr1^hi TAMs exhibited significantly higher immunosuppressive signature scores compared to other myeloid subsets (Fig. 4K, L; Fig. S5E). Moreover, their interaction patterns with CD8⁺ T cells resembled those of SPP1^hi TAMs, a population linked to immune checkpoint inhibitor resistance in mCRPC (Fig. S5F).

Together, these findings establish CX3CR1^hi TAMs as multifunctional mediators of disease progression post-ADT, contributing to both metabolic support and immune suppression in the tumor microenvironment.

Given the pro-tumoral roles of the CX3CL1-CX3CR1 axis, we evaluated the therapeutic potential of its disruption using Cx3cr1^−/− mice in both a spontaneous PbsnCre; Pten^f/f model and an RM-1cell-derived CRPC implantation model. In the Pten-deficient model, genetic deletion of Cx3cr1 significantly delayed tumor progression relative to controls (Fig. 5A-C). This was accompanied by a marked reduction in tumor cell proliferation, as shown by decreased Ki67⁺ staining (Fig. 5D-E; Fig. S6A).

The therapeutic effect was more pronounced in the aggressive CRPC model. At 21 days post-implantation, Cx3cr1^−/− mice bearing subcutaneous RM1 tumors exhibited 76.5% and 79.6% reduction in tumor weight and volume, respectively, compared to WT controls (Fig. 5F-H). This suppression was driven primarily by a decreased proliferation (Fig. 5I, K) rather than increased apoptosis (Fig. 5J, L).

Notably, CX3CR1 ablation led to a marked reduction of F4/80⁺ TAMs in the spontaneous prostate cancer TME of Pten conditional knockout mice (Fig. S6B-I). Similarly, in the RM-1 orthotopic tumor model, CX3CR1 deficiency was associated with a significant decrease in F4/80⁺ TAMs. While the total number of CD80⁺F4/80⁺ M1-like macrophages remained largely unchanged, the number of CD206⁺F4/80⁺ M2-like macrophages was substantially reduced (Fig. S6K, L). Moreover, although the overall number of CD3⁺ T cells was not markedly altered, CX3CR1 ablation was accompanied by a decrease in CD4⁺ T cells and a notable increase in CD8⁺ T cells within the TME (Fig. S6M, N).

Importantly, disrupting the CX3CL1-CX3CR1 axis also directly affected the tumor progenitor compartment. The number of SOX9⁺ cells, especially the CX3CL1-expressing subpopulation, was substantially reduced in Cx3cr1^−/− tumors (Fig. 5M-O; Fig.S6O, P). Co-staining of Ki67 and CX3CL1 further revealed a pronounced decrease in the proliferation of CX3CL1⁺ progenitors, indicating that TAM-derived signals contribute to the maintenance of the progenitors’ niche (Fig. 5P-R).

Together, these results support a functional role for the CX3CL1–CX3CR1 axis in maintaining the progenitor-TAM niche and suggest that disruption of this signaling pathway is associated with attenuated PCa progression (Fig. 6).

Graphical summary of the CX3CL1–CX3CR1 axis in prostate cancer. In CX3CR1-competent tumors, CSC-derived CX3CL1 recruits CX3CR1⁺ TAMs, which deliver trophic signals that sustain CSC survival and proliferation, fostering tumor progression, particularly under stress such as androgen-deprivation therapy (ADT). Disruption of CX3CR1 signaling abolishes TAM recruitment and paracrine support, impairing CSC expansion and suppressing tumor growth in androgen-independent prostate cancer. These findings highlight CX3CR1 inhibition as a strategy to dismantle CSC–TAM reciprocity and limit progression to castration-resistant disease.