Evolution of myeloid-mediated immunotherapy resistance in prostate cancer

Lyu, Aram; Fan, Zenghua; Clark, Matthew; Lea, Averey; Luong, Diamond; Setayesh, Ali; Starzinski, Alec; Wolters, Rachel; Arias-Badia, Marcel; Allaire, Kate; Wu, Kai; Gurunathan, Vibha; Valderrábano, Laura; Wei, Xiao X.; Miller, Richard A.; Van Allen, Eliezer M.; Fong, Lawrence

doi:10.1038/s41586-024-08290-3

Download PDF

Article
Open access
Published: 04 December 2024

Evolution of myeloid-mediated immunotherapy resistance in prostate cancer

Nature volume 637, pages 1207–1217 (2025)Cite this article

50k Accesses
47 Citations
57 Altmetric
Metrics details

Subjects

Abstract

Patients with advanced metastatic castration-resistant prostate cancer (mCRPC) are refractory to immune checkpoint inhibitors (ICIs)^1,2, partly because there are immunosuppressive myeloid cells in tumours^3,4. However, the heterogeneity of myeloid cells has made them difficult to target, making blockade of the colony stimulating factor-1 receptor (CSF1R) clinically ineffective. Here we use single-cell profiling on patient biopsies across the disease continuum and find that a distinct population of tumour-associated macrophages with elevated levels of SPP1 transcripts (SPP1^hi-TAMs) becomes enriched with the progression of prostate cancer to mCRPC. In syngeneic mouse modelling, an analogous macrophage population suppresses CD8⁺ T cell activity in vitro and promotes ICI resistance in vivo. Furthermore, Spp1^hi-TAMs are not responsive to anti-CSF1R antibody treatment. Pathway analysis identifies adenosine signalling as a potential mechanism for SPP1^hi-TAM-mediated immunotherapeutic resistance. Indeed, pharmacological inhibition of adenosine A2A receptors (A2ARs) significantly reverses Spp1^hi-TAM-mediated immunosuppression in CD8⁺ T cells in vitro and enhances CRPC responsiveness to programmed cell death protein 1 (PD-1) blockade in vivo. Consistent with preclinical results, inhibition of A2ARs using ciforadenant in combination with programmed death 1 ligand 1 (PD-L1) blockade using atezolizumab induces clinical responses in patients with mCRPC. Moreover, inhibiting A2ARs results in a significant decrease in SPP1^hi-TAM abundance in CRPC, indicating that this pathway is involved in both induction and downstream immunosuppression. Collectively, these findings establish SPP1^hi-TAMs as key mediators of ICI resistance in mCRPC through adenosine signalling, emphasizing their importance as both a therapeutic target and a potential biomarker for predicting treatment efficacy.

ARID1A loss induces polymorphonuclear myeloid-derived suppressor cell chemotaxis and promotes prostate cancer progression

Article Open access 26 November 2022

Characterisation of an autochthonous mouse ccRCC model of immune checkpoint inhibitor therapy resistance

Article Open access 05 June 2025

XBP1 regulates the protumoral function of tumor-associated macrophages in human colorectal cancer

Article Open access 20 October 2021

Main

Prostate cancer is the most prevalent male malignancy, with approximately 290,000 new cases diagnosed and 35,000 deaths per year in the USA⁵. Androgen deprivation therapy (ADT) initially elicits clinical responses, but most patients with advanced prostate cancer eventually progress to mCRPC and succumb to this disease⁶. There is therefore a clinical need to develop more effective treatment options. In recent years, ICIs have been approved for the treatment of multiple cancer types by disrupting checkpoint proteins, including cytotoxic lymphocyte antigen 4 (CTLA-4), PD-1 and PD-L1 (ref. ⁷). However, despite sporadic clinical responses largely restricted to rare molecular subtypes⁸, patients with mCRPC are typically refractory to these modalities^1,2, underscoring the need for more therapeutic strategies that address the mechanisms of resistance in tumours^9,10. However, these approaches have faced substantial problems, largely resulting from our limited understanding of the complex and intricate nature of mCRPC tumours.

The tumour microenvironment (TME), which is established by bidirectional interactions between tumour cells and components of their local environments, is a critical factor in promoting immunotherapeutic resistance across multiple malignancies^11,12. In prostate cancer, numerous cellular components and soluble factors contribute to the establishment of an immunosuppressive niche^13,14. Notably, there is substantial evidence that myeloid cells, particularly tumour-associated macrophages (TAMs) and myeloid-derived suppressor cells (MDSCs), mediate immunosuppression in prostate cancer by multiple mechanisms^3,4. In particular, the abundance of these suppressive myeloid cells is significantly increased after ADT^15,16, implying that they have a role as drivers of immunotherapy resistance in mCRPC. Although targeting myeloid cells has shown promise in enhancing the efficacy of ICIs in preclinical models^17,18, translating these findings into clinical applications, through broad myeloid-targeted interventions such as the blockade of CSF1R^19,20, has not demonstrated significant efficacy in either improving antitumour responses or providing substantial benefits to a wide subset of patients. This finding is attributed, in part, to the inherent heterogeneity of the targeted populations^10,13. We therefore propose that a comprehensive understanding of specific immunosuppressive myeloid subsets that are highly enriched in the advanced stages of prostate cancer could result in more effective disruption of their molecular mechanisms, enhancing the efficacy of immunotherapy.

In recent years, multi-omics single-cell profiling technologies have revolutionized our understanding of the heterogeneity of the TME across multiple malignancies at the single-cell level^21,22,23,24. These techniques have revealed previously unknown cell types and states within the prostate TME that mediate immunosuppression. For example, studies have identified cells such as fibroblasts that produce C-C motif chemokine ligand 2 (CCL2) and C-X-C motif chemokine ligand 12 (CXCL12)²⁵, as well as endothelial cells and pericytes that enhance the dysregulation of angiogenesis²⁶. Single-cell assessment has also been used to investigate the diverse population of tumour-infiltrating myeloid cells in either primary or metastatic prostate cancer^27,28,29,30. This research has identified the molecular mechanisms of immunosuppression mediated by myeloid cells, such as the activation of a CCR6–CCL20 axis by inflammatory monocytes and M2 macrophages residing in bone metastases²⁸. However, previous single-cell immune profiling of the prostate TME has predominantly focused on lymphocytes, particularly T cells, leaving a substantial gap in our understanding of the complexity of myeloid cells. Moreover, our knowledge of the evolution of the myeloid-mediated mechanisms underlying immunosuppression as prostate cancer progresses remains limited.

Here, we report that the myeloid-mediated mechanisms of immunotherapy resistance evolve as prostate cancer progresses. Through single-cell transcriptional profiling of patient biopsies, we identify a distinct macrophage subset characterized by elevated SPP1 transcript levels (referred to as SPP1^hi-TAMs), which becomes increasingly abundant with elevated immune inhibitory molecular programs as the disease advances. Notably, this specific macrophage population expresses reduced levels of CSF1R transcripts, indicating a potential link to the clinical ineffectiveness of CSF1R blockade in prostate cancer treatment. We reverse translate our findings to a syngeneic CRPC mouse model, in which we find an analogous macrophage subset through single-cell assessment. We demonstrate its role as a driver of immunotherapy resistance by computational analysis, functional assays and adoptive transfer experiments. We also find that SPP1^hi-TAMs directly suppress T cells through the activation of the adenosine signalling pathway. Inhibiting this pathway significantly reduces tumour growth and sensitizes tumour cells to ICI therapies in both humans and mice. Consistent with the findings from the mouse model, inhibition of A2AR using ciforadenant in combination with PD-L1 blockade with atezolizumab can induce clinical responses in patients with mCRPC. Collectively, these studies demonstrate that the myeloid-mediated mechanisms that underlie immunotherapeutic resistance evolve over the course of prostate cancer progression. SPP1^hi-TAMs have a key role in suppressing antitumour activity by activating adenosine signalling in prostate cancer, potentially serving as biomarkers to predict therapeutic efficacy.

Single-cell RNA-seq of human prostate cancer

To investigate the myeloid compartment and identify distinct immunosuppressive subsets during disease progression at the single-cell level, we used single-cell RNA (scRNA)-seq through a droplet-based 5′ 10x Genomics platform on tumour biopsies from patients with prostate cancer at various stages, including those with ADT-naive localized disease, metastatic hormone-sensitive prostate cancer (HSPC) on ADT, or mCRPC progressing on ADT (Fig. 1a). After rigorous quality control and data filtering, we obtained 147,174 single-cell transcriptomes. Using differentially expressed genes, we defined tumour cells and the major components of the TME, including immune cells and stromal cells (Fig. 1b and Extended Data Fig. 1a). Unsupervised clustering further identified 14 distinct subsets of tumour-infiltrating myeloid cells, including eight macrophage subsets, two MDSC subsets, three DC subsets and plasmacytoid DCs (pDCs) (Fig. 1b and Extended Data Fig. 1b). Our analysis revealed dynamic changes in the myeloid compartment as the disease progressed. For example, consistent with previous studies^15,16, we observed enrichment of MDSCs and TAMs with elevated expression of CX3CR1 and CD163 (CX3CR1^hi-TAMs; Extended Data Fig. 1b) in mCRPC compared with HSPC, although this trend was not statistically significant (Fig. 1c,d and Extended Data Fig. 1b). We also identified a distinct TAM subset that exhibited elevated enrichment scores for published FOLR2⁺ macrophage signatures³¹, including SELENOP, FOLR2 and SLC40A1 transcripts (referred to as SELENOP ^hi-TAMs in this study), which is associated with CD8⁺ T cell infiltration and improved patient prognosis in human breast cancer³¹, although their abundance did not significantly change with disease progression (Fig. 1c,d and Extended Data Fig. 1b–d). As well as these populations, we found a significant increase in macrophages characterized by elevated SPP1 transcript levels (SPP1^hi-TAMs) during disease progression (Fig. 1c,d and Extended Data Fig. 1b). Their presence was further confirmed by tissue staining from patients with either HSPC or mCRPC (Extended Data Fig. 1e). Droplet-based scRNA-seq is recognized to have technical limitations in capturing fragile populations, such as neutrophils and MDSCs³². This was evident in our comparison of the scRNA-seq data with staining of matched patient tissues using anti-human CD11b and CD15 antibodies (Extended Data Fig. 1f). Nevertheless, we confirmed an increased abundance of SPP1^hi-TAMs during disease progression (Extended Data Fig. 1e,f) through tissue staining of matched patient tissues, which is consistent with the scRNA-seq results. This result indicates that the prevalence of SPP1^hi-TAMs in mCRPC reflects the cellular composition of the prostate TME. Given their increased prevalence, we hypothesized that this macrophage population could have a critical role in mediating immunotherapy resistance in mCRPC. Consistent with this, we found that these SPP1^hi-TAMs exhibited elevated immunosuppression molecular programs relative to other myeloid subsets^21,33,34,35 (Fig. 1e,f and Extended Data Fig. 1g). Furthermore, our analysis of the T cell compartment (Extended Data Fig. 2a–c) revealed that elevated SPP1^hi-TAM gene signatures were significantly correlated with the degree of CD8⁺ T cell exhaustion³⁶, which showed a marked increase as the disease advanced (Fig. 1g and Extended Data Fig. 2d–f). Notably, further transcriptional analysis revealed a significant decrease in CSF1R transcript levels in SPP1^hi-TAMs relative to other myeloid cells (Fig. 1h and Extended Data Fig. 2g), indicating a mechanism that could contribute to the ineffectiveness of CSF1R blockade. Collectively, these results demonstrate that progression of prostate cancer leads to dynamic changes in the myeloid landscape within the TME, where SPP1^hi-TAMs emerge as potential drivers of immunotherapeutic resistance.

**Fig. 1: Single-cell assessment of biopsies from patients with prostate cancer reveals *SPP1*^hi-TAMs with elevated immunosuppression programs prevalent in advanced disease stages.**

Spp1 ^hi-TAMs in mouse prostate cancer

MyC-CaP is a prostate cancer cell line that is dependent on androgens and originates from a male mouse with prostate cancer³⁷. To delve deeper into our findings from patients, we performed droplet-based 5′ scRNA-seq (10x Genomics) with characterization of myeloid cell-surface antigens, including F4/80, CD11c, CD163 and Ly-6G, on a 1:1 mixture of fluorescence-activated cell sorting (FACS)-isolated immune (CD45⁺) and non-immune (CD45⁻) cells from mice subcutaneously engrafted with MyC-CaP, followed by treatment with either degarelix acetate (a gonadotropin-releasing hormone antagonist) or phosphate-buffered saline (PBS) (Fig. 2a). Consistent with previous studies^27,38, tumour burden significantly regressed after degarelix treatment and then progressed as CRPC (Fig. 2b). By using scRNA-seq, we identified 6 main cell types in the TME, along with 11 distinct myeloid subsets (Fig. 2c and Extended Data Fig. 3a–c). Comparative analysis of transcriptomes between humans and mice enabled us to identify a mouse macrophage subset (Spp1^hi-TAMs) that is analogous to human SPP1^hi-TAMs) (Fig. 2d,e), characterized by increased expression of Spp1, Cd9 and Lgals3 transcripts and reduced expression of Csf1r, Mrc1, Cx3cr1 and Cd163 (Fig. 2f and Extended Data Fig. 3d). To validate these transcriptional findings at the protein level, we established CRPC in Spp1-EGFP mice, in which EGFP is expressed under the control of the Spp1 promotor (Extended Data Fig. 4a). By using a multi-parameter flow-cytometry panel (Extended Data Fig. 4b), we confirmed the presence of multiple myeloid subsets identified through scRNA-seq and observed dynamic changes in the myeloid composition (Extended Data Fig. 4c–f). We observed that the cellularity of Spp1^hi-TAMs remained largely consistent between HSPC and CRPC, although their frequency decreased during disease progression, mainly because of significant infiltration by Cx3cr1^hi-TAMs (Extended Data Fig. 4c,d), as reported previously¹⁵. To determine whether Spp1^hi-TAMs are resistant to CSF1R blockade, we administered an anti-CSF1R antibody to mice bearing CRPC (Fig. 2g). In line with our transcriptional findings, macrophages with high expression of Csf1r transcripts (Extended Data Fig. 3d), including CD163^hi-TAMs and CX3CR1^hi-TAMs, were significantly ablated, but Spp1^hi-TAMs remained largely unaffected (Fig. 2h,i). This supports a potential role of SPP1^hi-TAMs in contributing to the therapeutic resistance of CSF1R blockade. Consistent with data from humans, Spp1^hi-TAMs in mouse prostate cancer exhibited significantly elevated immunosuppressive gene signatures compared with other myeloid cells (Fig. 3a,b and Extended Data Fig. 4g), supporting their role as drivers of immunotherapy resistance. We found analogous results with the TRAMP-C2 model³⁹, a syngeneic prostate cancer cell line that, unlike MyC-CaP, is not Myc-driven, treated with either anti-PD-1 or isotype-matched control antibodies (Extended Data Fig. 4h). Consistent with previous studies⁴⁰, anti-PD-1 antibody treatment alone showed minimal efficacy in this model (Extended Data Fig. 4i). Single-cell analysis revealed the presence of Spp1^hi-TAMs with elevated immunosuppressive molecular programs relative to other TAM subsets, in line with the MyC-CaP model (Extended Data Fig. 4j–n). Taken together, through single-cell assessment, our data enabled us to identify analogous Spp1^hi-TAMs with elevated immunosuppressive gene signatures across multiple mouse models of prostate cancer, and we subsequently demonstrated their resistance to CSF1R inhibition.

**Fig. 3: *Spp1*^hi-TAMs have a critical role in promoting immunotherapeutic resistance by inducing exhaustion in CD8⁺ T cells in vivo.**

Spp1 ^hi-TAMs drive immunotherapy resistance

To assess the ability of myeloid cells to functionally suppress T cell activity, multiple myeloid subsets, including MDSCs, CX3CR1^hi-TAMs and Spp1^hi-TAMs, were isolated by FACS from CRPC developed in Spp1-EGFP mice (Extended Data Fig. 5a) and co-cultured with splenic CD8⁺ T cells in the presence of anti-CD3/CD28 stimulation. As previously reported^18,41, MDSCs and CX3CR1^hi-TAMs effectively suppressed the proliferation of T cells in vitro, serving as controls (Extended Data Fig. 5b). Notably, we found that Spp1^hi-TAMs significantly inhibited T cell proliferation in a density-dependent manner (Fig. 3c,d). Furthermore, the presence of Spp1^hi-TAMs resulted in a marked decrease in the frequency of polyfunctional (IFN-γ⁺TNF-α⁺) CD8⁺ T cells (Fig. 3e and Extended Data Fig. 5c), indicating that their immunosuppressive activity can dampen T cell effector function. Next, we tested whether Spp1^hi-TAMs can promote resistance to ICIs in vivo. We first confirmed that a combination of anti-CTLA-4 and anti-PD-1 antibodies results in a more significant decrease in the growth of CRPC than either given alone (Extended Data Fig. 5d), which is consistent with previous studies^27,38. We reasoned that if Spp1^hi-TAMs could mediate immunotherapy resistance, they would decrease the effectiveness of the dual treatment. To test this possibility, we adoptively transferred FACS-purified Spp1^hi-TAMs into CRPC in the presence of the combination treatment, minimizing potential issues with their trafficking to the TME (Fig. 3f). Strikingly, intratumorally transferred Spp1^hi-TAMs resulted in significantly diminished efficacy of the dual treatment and reduced overall survival (Fig. 3g,h). Transferring Spp1^hi-TAMs significantly increased the frequency of exhausted (CD38⁺PD-1⁺) CD8⁺ T cells within ICI-treated tumours compared with control ICI-treated tumours. The levels of these exhausted T cells were similar to those observed in PBS-treated tumours (Fig. 3i and Extended Data Fig. 5e), highlighting the suppressive activity of Spp1^hi-TAMs in vivo. Taken together, these results indicate that Spp1^hi-TAMs have a critical role in driving immunotherapeutic resistance in CRPC.

SPP1 ^hi-TAMs drive suppression through adenosine

To explore the mechanisms by which SPP1^hi-TAMs promote immunotherapeutic resistance, we further analysed our scRNA-seq datasets of human and mouse prostate cancers. Pathway analysis showed that hypoxia was among the top pathways activated preferentially in SPP1^hi-TAMs in patients and mice (Fig. 4a and Extended Data Fig. 6a), which is consistent with previous studies that found that SPP1 is upregulated in macrophages in the hypoxic TME⁴². Hypoxia is known to promote the accumulation of extracellular adenosine in tumours through the upregulation of CD39 and CD73, which are ectonucleotidases that convert ATP to ADP and AMP, and AMP to adenosine, respectively⁴³. Alternatively, this process can also involve ectoenzymes such as CD38 and CD203a, which generate AMP by degrading NAD⁺ and ADPR⁴³. Adenosine is an established mediator of immunosuppression in tumours⁴⁴; its binding to adenosine receptors, particularly A2ARs and A2B receptors (A2BRs), which have higher and lower affinities, respectively, initiates downstream immunosuppressive signalling by the accumulation of intracellular cAMP, leading to the suppression of the antitumour activity of T cells and natural killer (NK) cells^45,46,47. Notably, we observed elevated levels of ADORA2A transcripts, which encode A2ARs, during disease progression in both CD8⁺ T cells and NK cells, whereas ADORA2B transcript levels were increased only in CD8⁺ T cells (Extended Data Fig. 6b). Using the published specific gene signature associated with adenosine signalling⁴⁸, which has been shown to strongly correlate with the extracellular adenosine concentration in the TME, we confirmed a strong correlation between enrichment scores for hypoxia and the adenosine signalling signature in our patient dataset (Fig. 4b), concordant with there being a link between hypoxia and adenosine accumulation. Notably, the expression of the genes associated with the adenosine signalling signature increasingly correlates with SPP1^hi-TAM enrichment scores as the disease progresses (Fig. 4c and Extended Data Fig. 6c), but this trend is not evident in other myeloid populations, including EEF1A1^hi-TAMs (Extended Data Fig. 6d). Consistent with human results, the adenosine signalling signature is elevated in Spp1^hi-TAMs relative to other myeloid subsets in mice (Fig. 4d, Extended Data Fig. 6c). When stimulated splenic CD8⁺ T cells were cultured in transwell plates with Spp1^hi-TAMs isolated by FACS, such that the two cell types were separated by micropores, allowing only soluble factors to pass through (Extended Data Fig. 6e), we observed that T cell proliferation was suppressed (Extended Data Fig. 6f). This finding indicates that the accumulation of soluble factors, including adenosine, may contribute to immunotherapeutic resistance mediated by SPP1^hi-TAMs. Subsequent in vitro assays confirmed that Spp1^hi-TAMs did indeed release extracellular adenosine (Fig. 4e). To further investigate the role of adenosine in immunosuppression mediated by SPP1^hi-TAMs, we assessed the expression levels of CD38, ENTPD1 and NT5E, which encode CD38, CD39 and CD73, respectively, across multiple cell populations with the emphasis on myeloid cells. Consistent with previous studies^43,49, various cell types, including B cells and endothelial cells, exhibit an elevated adenosine signalling signature, with increased expression of NT5E and/or ENTPD1 (Extended Data Fig. 6g, h). Notably, our analysis revealed elevated transcript levels of NT5E, but not ENTPD1 or CD38, in SPP1^hi-TAMs compared with other myeloid subsets in both humans and mice (Fig. 4f and Extended Data Fig. 6i). In particular, NT5E levels in SPP1^hi-TAMs are significantly higher in mCRPC than in earlier stages (Extended Data Fig. 6j). This trend is similarly observed in CD73 protein levels in mice, although the transcript levels exhibited a diminished, but non-significant, change (Extended Data Fig. 7a–c). To test directly whether SPP1^hi-TAMs suppress T cell activity through adenosine, splenic CD8⁺ T cells activated with anti-CD3/CD28 stimulation were co-cultured with Spp1^hi-TAMs in the presence of ciforadenant, a small-molecule inhibitor of A2ARs (Fig. 4g,h), as well as a blocking antibody against CD73 (Fig. 4i,j). In both cases, blocking either the receptor or the ectoenzyme resulted in a significant reduction in suppression of T cells mediated by Spp1^hi-TAMs, indicating that adenosine is closely involved in the immunosuppressive activity of Spp1^hi-TAMs. However, inhibition of the adenosine pathway was not sufficient to fully restore T cell proliferation in culture, indicating a role for further mechanisms by which Spp1^hi-TAMs can drive immunotherapeutic resistance. To investigate such suppressive mechanisms, we carried out further pathway analysis and identified multiple pathways enriched in SPP1^hi-TAMs associated with inflammatory responses in both humans and mice (Extended Data Fig. 7d). These findings were supported by the elevated scores of published gene signatures for myeloid cells expressing proinflammatory soluble factors, such as IL-1β (tumour-promoting inflammation signature)⁵⁰, in SPP1^hi-TAMs across both species, and there was a significant correlation between SPP1^hi-TAM abundance and tumour-promoting inflammation signature enrichment in patients (Extended Data Fig. 7e,f). Notably, blockade of IL-1R significantly diminished Spp1^hi-TAM-mediated T cell suppression in culture (Extended Data Fig. 7g), indicating that IL-1R signalling also has an important role in driving immunotherapy resistance by these macrophages. There was no significant synergistic effect observed with the combined blockade of A2AR and IL-1R in vitro (Extended Data Fig. 7h). Collectively, these findings indicate that SPP1^hi-TAMs dampen T cell activity, at least in part through extracellular adenosine.

**Fig. 4: *SPP1*^hi-TAMs are hypoxic and mediate immunosuppression through adenosine signalling.**

A2AR blockade reverses ICI resistance

Considering that adenosine signalling probably underlies immunosuppression mediated by SPP1^hi-TAMs, we then examined whether treating mice bearing CRPC with ciforadenant could alter the antitumour responses in vivo (Fig. 5a). Consistent with previous studies on mice with different cancer types, such as MC38 and B16 (refs. ^51,52), blockade of A2ARs led to a significant reduction in CRPC growth, potentially resulting from a significantly lower frequency and number of exhausted CD8⁺ T cells (CD38⁺PD-1⁺; Extended Data Fig. 8a–c). Notably, evaluation of myeloid composition revealed alterations in the myeloid compartment. Although the overall number of the main myeloid populations remained largely unchanged (Extended Data Fig. 8d), the inhibition of A2ARs resulted in a significant decrease in both the frequency and number of Spp1^hi-TAMs (Fig. 5c,d and Extended Data Fig. 8e,f), with no significant effect on other TAMs (Extended Data Fig. 8g). Further analyses of our mouse prostate cancer scRNA-seq dataset revealed elevated transcript levels of Adora2a (Fig. 5e) and higher enrichment scores for an adenosine gene signature expression (AdenoSig)⁵³ (Fig. 5f,g), obtained by using a collection of genes with significantly induced expression on adenosine agonists in Spp1^hi-TAMs compared with other macrophages and monocytes. Consistent with this, we identified enriched AdenoSig scores in SPP1^hi-TAMs and observed a significant correlation between the enrichment scores for SPP1^hi-TAM abundance and AdenoSig (Extended Data Fig. 8h,i) in humans. These findings indicate that adenosine signalling could be crucial for the abundance of Spp1^hi-TAMs in CRPC. Given the significant decreases in the abundance of exhausted CD8⁺ T cells and Spp1^hi-TAMs following treatment with ciforadenant, we proposed that A2AR blockade could augment the efficacy of ICIs. To test this possibility, we administered ciforadenant to mice bearing CRPC in combination with anti-PD-1, which showed limited therapeutic effectiveness as a monotherapy^27,38 (Extended Data Fig. 5d), or relevant isotype-matched control antibodies (Fig. 5h). Notably, consistent with previous studies using different cancer types, including B16 and AT-3 (refs. ^54,55), the dual blockade of A2ARs and PD-1 resulted in a significantly greater reduction in tumour growth than did monotherapies (Fig. 5i), indicating that A2AR inhibition contributes to enhancing the efficacy of ICIs in CRPC. Evaluation of the lymphoid compartment revealed that ciforadenant increased the frequency of polyfunctional CD8⁺ T cells, whereas PD-1 blockade enhanced the infiltration of T cells and NK cells into tumours (Extended Data Fig. 9a–c). Importantly, in line with previous studies^54,55, combining PD-1 blockade with A2AR inhibition from ciforadenant increased the density of polyfunctional CD8⁺ T cells relative to monotherapies (Fig. 5j), indicating a mechanism underlying enhanced antitumour activity from combination therapy. Our assessment of myeloid cells indicated that the frequency of Spp1^hi-TAMs was not further reduced by combining PD-1 blockade with ciforadenant compared with ciforadenant alone (Extended Data Fig. 9d). These findings highlight that the increased abundance of activated CD8⁺ T cells has a key role in the enhanced antitumour responses observed with combination therapy. On the basis of these findings, we evaluated the effect of A2AR blockade on immunotherapeutic resistance in humans in a phase 1 clinical trial (NCT02655822). Ciforadenant was administered to patients with mCRPC after failing at least one next-generation androgen blockade (Extended Data Fig. 10a), either alone (n = 11) or in combination with atezolizumab (n = 24) (Fig. 5k). The dosing frequency for this trial was as follows: ciforadenant, 100 mg twice a day for 28 days; and atezolizumab, 840 mg once every 2 weeks. This treatment was well tolerated, with very low frequencies of major side effects (Extended Data Fig. 10b). Of the 24 patients, 6 (25%) had a decrease in prostate-specific antigen levels from the baseline of 30% or more (Extended Data Table 1), which was confirmed 4 weeks later, and tumour regression was observed in some patients with measurable disease (Fig. 5l,m). Mutational analyses of two responders showed that their tumours were microsatellite stable and lacked CDK12 alterations, which could have sensitized the cancer to the ICI treatment (Extended Data Table 1). Moreover, analysis of biopsies from one responder and two non-responders demonstrated limited PD-L1 expression in all tissues (Extended Data Fig. 10c). Importantly, the responder had a higher prevalence of SPP1^hi-TAMs at baseline (Extended Data Fig. 10d). These findings indicate that baseline SPP1^hi-TAM abundance may serve as a potential biomarker for therapeutic efficacy, although further investigation with a larger cohort is warranted. Taken together, therapeutic interventions targeting adenosine signalling could represent a potential strategy to sensitize mCRPC to ICI treatments.

**Fig. 5: Inhibition of adenosine signalling diminishes the abundance of *Spp1*^hi-TAMs and enhances the responsiveness of CRPC to PD-1 blockade in vivo.**

Discussion

Although the establishment of an immunosuppressive niche by tumour-infiltrating myeloid cells in the prostate TME is well recognized, especially after ADT^{3,4,15,16,17,18}, efforts to target these cells to enhance anti-tumour responses in patients with mCRPC have faced problems. For example, using CSF1R inhibition to target macrophages (the predominant myeloid subset in tumours) have shown limited antitumour activity^19,20. Efforts to deplete intratumoral neutrophils or MDSCs by inhibiting myeloid chemotaxis through CXCR2 blockade have shown a reduction in therapeutic resistance to ADT³⁰. Thus, a detailed understanding of the heterogeneity of myeloid cells, as well as the key regulators that govern myeloid programs within tumours⁵⁶, is crucial for addressing therapeutic resistance. Despite numerous studies on the prostate TME at single-cell resolution^25,26,27,28, there is no comprehensive single-cell atlas of myeloid cells across the disease continuum. Our findings, which are derived from patient biopsies and relevant mouse models, highlight the heterogeneity of myeloid cells in prostate cancer. Notably, SPP1^hi-TAMs, which were identified as a prevalent myeloid subset in advanced disease, express diminished CSF1R transcript levels, providing an explanation for the lack of clinical efficacy in targeting this receptor. Comparative transcriptome analyses between humans and mice identified analogous Spp1^hi-TAMs in mouse CRPC in an unbiased manner, which demonstrates their resistance to anti-CSF1R treatment in vivo. During the development of CRPC in mice, Spp1^hi-TAMs significantly decrease in frequency, although their cell number remains consistent, mainly as the result of a substantial increase in both the frequency and number of infiltrating CX3CR1^hi-TAMs (Extended Data Fig. 4c,d). Conversely, in human mCRPC, the SPP1^hi-TAM fraction increases significantly as the disease progress (Fig. 1c,d), becoming at least as prevalent as CX3CR1^hi-TAMs, if not more so. Given the diminished CSF1R transcript levels in SPP1^hi-TAMs (Fig. 1h and Extended Data Fig. 2g), this could partly explain the clinical ineffectiveness of CSF1R antagonism in human patients with cancer compared with pre-clinical models. It will therefore be important to evaluate whether targeting immune inhibitory signals provided by SPP1^hi-TAMs, along with CSF1R blockade, will elicit antitumour responses and augment the efficacy of ICIs in patients.

Although SPP1^hi-TAMs have been identified in other cancer types^21,42,57, their roles in prostate cancer progression, particularly as drivers of immunotherapy resistance and through molecular mechanisms, have not been functionally investigated. In this study, we demonstrate that Spp1^hi-TAMs induce resistance to ICIs through adoptive transfer into CRPC. Our single-cell transcriptional analysis of human and mouse prostate cancers identified adenosine signalling as one of the main pathways preferentially activated in SPP1^hi-TAMs, and functional assays subsequently confirmed that these macrophages contribute to extracellular adenosine accumulation in the prostate TME. Disrupting adenosine signalling, through either A2AR inhibitors or CD73-targeting antibodies, significantly reduced Spp1^hi-TAM-mediated suppression of CD8⁺ T cells in co-culture, indicating that adenosine-associated signals are potential immunotherapeutic targets. However, the observation that T cell proliferation was not fully restored despite A2AR or CD73 blockade implies that there are more mechanisms underlying Spp1^hi-TAM-mediated immunosuppression. Further transcriptional and functional analyses indicated that IL-1R signalling could have a role in SPP1^hi-TAM-mediated resistance (Extended Data Fig. 7d–g), in line with previous findings⁵⁰. Moreover, alterations in metabolic processes, including dysregulated lipid metabolism, potentially mediated by upregulated Trem2, have also been identified in SPP1^hi-TAMs, indicating a link to prostate cancer growth, invasiveness and therapeutic resistance⁵⁸. Therefore, examining T cell modulation by SPP1^hi-TAMs in vivo through various approaches, including spatial transcriptomics, will be a focus of future studies. Moreover, a deeper understanding of the immunosuppressive mechanisms used by these macrophages will be crucial in identifying further therapeutic targets to enhance efficacy.

In our clinical trial, we observed that patients with mCRPC may benefit more from the combination of ciforadenant and atezolizumab than from atezolizumab alone². Although promising, results from another study of AZD4635 (another A2AR antagonist) combined with durvalumab and cabazitaxel in patients with mCRPC (AARDVARC) failed to show a benefit with A2AR antagonism⁵⁹. The discrepancy between the trials could have resulted from chemotherapy inclusion, the use of a different A2AR antagonist and/or patient selection. Despite the improved clinical activity observed with the combination treatment, antitumour responses were evident in only one of four patients in our trial. Moreover, although combined A2AR and PD-1 blockade significantly prolonged survival in a mouse model of CRPC, the mice did eventually die from the disease. These findings indicate that other immunosuppressive elements within the TME would need to be targeted simultaneously for even more effective immunotherapeutic intervention. Previous studies have identified various signals that contribute to the immunosuppressive nature of the prostate TME. For example, prostate cancer-associated fibroblasts promote immunosuppression on T cells by the release of transforming growth factor-β⁶⁰ or by recruiting suppressive myeloid cells through the CCL2 and CXCL12 pathways²⁵. Furthermore, castration-induced CXCL1, CXCL2 and IL-8 from prostate cancer cells mediate myeloid infiltration, particularly of MDSCs, resulting in an immunosuppressive TME^17,18. Thus, a better understanding of further immunosuppressive TME elements beyond myeloid cells and their role in resistance to ICIs could reveal other therapeutic opportunities in mCRPC and provide strategies for patient selection.

Collectively, the data in this study demonstrate that SPP1^hi-TAMs become increasingly abundant during prostate cancer progression and promote immunotherapeutic resistance through adenosine-mediated immunosuppression. Inhibition of A2AR delays CRPC progression and improves the responsiveness of tumour cells to PD-1 blockade. Moreover, our clinical trial shows that a subset of patients with mCRPC may benefit from ciforadenant plus atezolizumab instead of monotherapies. The abundance of SPP1^hi-TAMs could serve as a biomarker to select for patients in future trials. Inhibiting adenosine signalling, as well as targeting chemokine or growth-factor receptor pathways, could further enhance the efficacy of immunotherapy in this and perhaps other refractory cancers.

Methods

scRNA-seq of samples derived from patients with prostate cancer

Tumour tissues were obtained from baseline biopsies of patients participating in clinical trials for localized prostate cancer (NCT03821246), de novo oligometastatic prostate cancer (NCT03007732) and metastatic mCRPC (NCT03248570). Viable cryopreserved tumour tissue samples were digested in Roswell Park Memorial Institute (RPMI) medium containing Collagenase I and II (0.1 mg ml⁻¹, Thermo Fisher Scientific) and DNase I (Thermo Fisher Scientific), minced and then subjected to 1 h digestion using the gentleMACS system (Miltenyi Biotec). Live cell isolation was done using MACS LS columns (Miltenyi Biotec). The 10x Genomics Chromium Controller was used to generate GEM bead emulsions using the Single Cell 5′ Library & Gel Bead Kit (10x Genomics), followed by cDNA synthesis and amplification, and subsequent library preparation steps using 10x Genomics kits. Library sequencing was done by the University of California, San Francisco (UCSF) Institute for Human Genetics core on a NovaSeq 6000 (Illumina), targeting a median read depth of 150,000 reads per cell for total gene expression libraries and 60,000 reads per cell for CITE-seq libraries. All antibodies were obtained from BioLegend unless otherwise indicated. This work was done with informed consent obtained from all human research participants, and the sample procurement and analysis were approved by the institutional review board committees at UCSF.

Human scRNA-seq analysis

The raw data from 10× sequencing were processed using the Cell Ranger pipeline (v.3, Genome build, GRCh38). The raw gene-expression matrices were subjected to processing by CellBender (v.0.1.0)⁶¹ to eliminate ambient RNAs. The filtered gene-expression matrices then underwent doublet detection using the package DoubletDetection (https://doi.org/10.5281/zenodo.2678041) with default parameters. The results were analysed through the SCANPY pipeline⁶². To ensure the retention of high-quality cells, the following filters were applied: first, cells with less than 10% mitochondrial genes were retained; second, the number of detected genes per cell was set between 100 and 2,500 genes; third, genes expressed in at least three cells were kept; and finally, platelets (PF4, unique molecular identifier (UMI) > 0), red blood cells (HBB, UMI > 1) and doublets were removed. The gene-expression matrix was log₂-transformed with the addition of 1 and normalized to 10,000 counts per cell, followed by highly variable gene selection using default parameters with the SCANPY function. The resulting matrix was corrected by regressing out total UMI counts and mitochondria percentage, followed by scaling to a mean of 0 and a variance of 1. Principal component analysis was performed using the top 50 principal components, followed by sample-wise batch correction using the SCANPY-implemented Harmony⁶³. Leiden clustering (default resolution = 1.0) and UMAP plotting were performed, with a resolution of 1.0 applied for both T cell and myeloid cell clustering. Differential expression analysis identified the top-ranked genes that were upregulated in each individual cluster relative to the combination of all other cells, as determined by the SCANPY function tl.rank_genes_groups. Annotation of each unbiased population was achieved through manual inspection of the top-ranked genes of each cluster. Analysis of cell density on the UMAP was carried out using the SCANPY function tl.embedding_density, and boxplots were generated to represent cell population frequencies for each cell type. Gene scores were computed using the SCANPY function tl.score_genes with curated gene lists provided. To calculate gene scores at the sample level, scores were computed for each cell and subsequently combined at the sample level by using the median score of cells within a given sample.

Mice

FVB/NJ and C57BL/6J male mice (from the Jackson Laboratory) were used in the experiments at 6–10 weeks of age. The STOCK Tg(Spp1-EGFP)PD43Gsat/Mmucd (Spp1-EGFP)⁶⁴ mouse strain was sourced from the Mutant Mouse Resource & Research Centers at the University of California, Davis. All mice were housed in a pathogen-free facility under standardized environmental conditions, including a controlled 12 h:12 h light:dark cycle, humidity of 30–70% and a temperature range of 20–26 °C. For experiments, a total of 1 × 10⁶ cells (murine prostate cancer cell line MyC-CaP (CRL3255, American Type Culture Collection (ATCC)) or TRAMP-C2 (CRL-2731, ATCC)) were resuspended in sterile PBS and transplanted subcutaneously in the right flank of either FVB/NJ or C57BL/6J mice, respectively. The identities of MyC-CaP and TRAMP-C2 were authenticated using the Mouse Cell STR Profiling Service (137-XV, ATCC), and mycoplasma contamination was tested before each injection using a mycoplasma PCR detection kit (G238, abm). Sample size was determined using preliminary data and previous publications to ensure reproducibility of the experiment. Tumour volume = (L × W × W)/2 (mm³), with length (L) and width (W) being the longest diameter and shortest diameter, respectively. All experimental procedures were approved by the Institutional Animal Care and Use Committee at UCSF.

The model for CRPC was established by subcutaneously engrafting 1 × 10⁶ MyC-CaP cells into the right flank of male FVB/NJ mice 6–10 weeks old. When the tumour size reached 100–200 mm³, each mouse was injected subcutaneously with 1.875 mg degarelix (Firmagon) in 100 μl PBS, followed by a maintenance dose of 0.625 mg degarelix in 100 μl PBS every 28–30 days to induce CRPC. The development of CRPC was defined as a tumour volume that regressed after degarelix treatment and then grew back to 100–200 mm³. Subsequently, the mice were randomized and treated with the indicated antibodies and/or inhibitors.

Cell line culture

MyC-CaP and TRAMP-C2 cells were cultured in complete DMEM medium comprising Dulbecco’s Modified Eagle Medium supplemented with 10% fetal bovine serum (FBS, Omega Scientific) and 1× penicillin/streptomycin (10,000 ml streptomycin sulfate and 10,000 units ml⁻¹ penicillin G). All reagents were obtained from the UCSF Cell Culture Facility, unless otherwise indicated.

Flow cytometric analysis

Mouse organs were collected and processed as follows. Spleens were mechanically dissociated with FACS wash buffer (PBS supplemented with 2% (v/v) FBS and 0.5 mM EDTA (Teknova)). Tumours were sequentially digested three times with 12 ml of a cocktail of 2 mg ml⁻¹ (w/v) collagenase type IV and 100 Kunitz units per ml DNase I (both from Sigma-Aldrich) for 12 min for each digest. All single-cell suspensions were filtered using 70-μm filters (Fisher Scientific) and subjected to red-blood-cell lysis using ACK Lysing Buffer (Quality Biological). Cells were immunostained by incubating at 4 °C for 30 min with the fluorescently labelled antibodies below (all antibodies were purchased from BioLegend unless otherwise indicated). After staining, cells were washed once or twice in FWB and resuspended in FWB or FWB containing 1 μg ml⁻¹ propidium iodide (PI, BioLegend) to assess viability. All flow cytometric data were acquired using an LSRFortessa X-50 flow cytometer operated with FACSDiva software (BD Biosciences). Post-acquisition data analysis was performed using FlowJo (v.10.10.0, Tree Star). All antibodies used in this study are commercially available and have been validated by the manufacturer or through published literature. On receipt, laboratory testing was conducted with known positive and negative controls to confirm the reliability of each antibody.

For mouse lymphoid staining, we used anti-mouse CD3-Brilliant Ultraviolet 395 (563565, Clone 145-2C11, BD Biosciences, 1:200), CD4-Brilliant Violet 711 (100447, GK1.5, 1:200), CD8-Brilliant Ultraviolet 805 (612898, 53-6.7, BD Biosciences, 1:200), NK-1.1-Alexa Fluor 647 (108719, PK136, 1:200), CD38-PE/Cyanine7 (102717, 90, 1:200), CD39-Brilliant Violet 421 (567105, Y23-1185, BD Biosciences, 1:200), CD45-Brilliant Violet 785 (103149, 30-F11, 1:200), CD279 (PD-1)-PE/Dazzle 594(109115, RMP1-30, 1:200) antibodies were used. For mouse myeloid staining, anti-mouse CD11b-Brilliant Violet 605 (101257, M1/70, 1:200), CD39-Brilliant Violet 421 (567105, Y23-1185, BD Biosciences, 1:200), CD73-PE (12-0731-82, eBioTY/11.8 (TY/11.8), Invitrogen, 1:200), CX3CR1-PE/Cyanine7 (149015, SA011F11, 1:200), F4/80-Alexa Fluor 647 (565853, T45-2342, BD Biosciences, 1:200), I-A/I-E-Alexa Fluor 700 (107621, M5/114.15.2, 1:200), Ly-6G-APC/Cyanine7 (127623, 1A8, 1:200), Podoplanin-PerCP/Cyanine5.5 (127421, 8.1.1, 1:200), Siglec-F-Brilliant Violet 421 or Brilliant Ultraviolet 395 (562681 or 740280, E50-2440, BD Biosciences, 1:200) antibodies. The relevant isotype-matched antibodies (eBRG1, RTK2758, RTK4530 and SHG-1) were used as controls.

For intracellular immunostaining of proteins, single-cell suspensions were labelled with LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (L34957, Invitrogen, 1:1,000) and then treated with eBioscience Foxp3/Transcription Factor Staining Buffer Set (Invitrogen), according to the manufacturer’s protocol designed for intracellular (cytoplasmic) proteins. Cells were then stained with fluorescently labelled antibodies against anti-mouse CD3; Brilliant Ultraviolet 395 (563565, 145-2C11, BD Biosciences, 1:200), CD8-Brilliant Ultraviolet 805 (612898, 53-6.7, BD Biosciences, 1:200), CD11b-Brilliant Violet 605 (101257, M1/70, 1:200), CD45-Brilliant Violet 785 (103149, 30-F11, 1:200), IFN-γ-PE/Cy7 (505825, XMG1.2, 1:100), and TNF-α-Brilliant Violet 421 (506327, MP6-XT22, 1:100). The relevant isotype-matched antibodies (RTK2071) were used as negative controls.

In vitro co-culture of purified myeloid cells with T cells

Complete RPMI cell culture medium was made up of RPMI 1640 supplemented with 10% FBS (Omega Scientific), 1× β-mercaptoethanol (Gibco, 55 µM), 1× glutamine (29.2 g l⁻¹ l-glutamine, 200 mM), 1 mM sodium pyruvate (11 g l⁻¹ sodium pyruvate), 1× MEM non-essential amino acids, 1× penicillin/streptomycin (10,000 μg ml⁻¹ streptomycin sulfate and 10,000 units ml⁻¹ penicillin G). All reagents were obtained from the UCSF Cell Culture Facility, unless otherwise indicated.

For enrichment of mouse CD8⁺ T cells, single-cell suspensions of spleens from CRPC-bearing mice were labelled with BD Violet proliferation dye 450 (Fisher Scientific) and subsequently negatively enriched using the MojoSort Mouse CD8 T cell isolation kit, according to the manufacturer’s instructions. For isolation of specific myeloid subsets, single-cell suspensions from CRPC developed in Spp1-EGFP mice were incubated with the LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (L34957, Invitrogen, 1:1,000), anti-mouse CD11b-Brilliant Violet 605 (101257, M1/70, 1:200), CX3CR1-PE/Cyanine7 (149015, SA011F11, 1:200), F4/80-Alexa Fluor 647 (565853, T45-2342, BD Biosciences, 1:200), I-A/I-E-Alexa Fluor 700 (107621, M5/114.15.2, 1:200), Ly-6G-APC/Cyanine7 (127623, 1A8, 1:200), Podoplanin-PerCP/Cyanine5.5 (127421, 8.1.1, 1:200) and Siglec-F-Brilliant Violet 421 (562681, BD Biosciences, 1:200) antibodies. After immunostaining, cells were washed twice in FWB and resuspended in FWB containing 1 μg ml⁻¹ propidium iodide to assess viability. The cells of interest were FACS-purified using BD FACSAria Fusion operated with FACSDiva software (BD Biosciences).

To determine whether TAM cells, including Spp1^hi-TAMs, CX3CR1^hi-TAMs and CD163^hi-TAMs, mediate immunosuppression, 1 × 10⁴ CD8⁺ T cells, labelled with BD Violet Proliferation Dye 450 (BDB562158, Fisher Scientific, 1:1,000) and stimulated with 1 × 10⁴ Dynabeads Mouse T-Activator CD3/CD28 (Gibco) were cultured in the presence or absence of purified myeloid subsets at a 1:1, 5:1 or 10:1 ratio, respectively, in 200 µl complete RMPI medium in round (U)-bottom 96-well plates at 37 °C, 5% CO₂ for 3 days. T cell proliferation was assessed by flow cytometry.

To determine whether Spp1^hi-TAMs suppress T cell activation, 1 × 10⁴ CD8⁺ T cells, labelled with BD Violet Proliferation Dye 450 (Fisher Scientific) and activated with 1 × 10⁴ Dynabeads Mouse T-Activator CD3/CD28 (Gibco), were cultured with or without purified Spp1^hi-TAMs at a 1:1 ratio in 200 µl complete RMPI medium in round (U)-bottom 96-well plates at 37 °C, 5% CO₂ for 3 days. The cells were subsequently restimulated with 1× eBioscience Cell Stimulation Cocktail (plus protein transport inhibitors, Invitrogen) for 5 h. After washing, cells were stained for intracellular immunostaining of proteins. T cell activation was assessed by flow cytometry.

To determine whether Spp1^hi-TAM-mediated T cell suppression requires adenosine signalling, 1 × 10⁴ CD8⁺ T cells, labelled with BD Violet Proliferation Dye 450 (BDB562158, Fisher Scientific, 1:1,000) and activated with 1 × 10⁴ Dynabeads Mouse T-Activator CD3/CD28 (Gibco), were cultured with or without purified Spp1^hi-TAMs at a 1:1 ratio in 200 µl complete RMPI medium in the presence of ciforadenant (10 μM, Corvus Pharmaceuticals) or InVivoMAb anti-mouse CD73 (10 μg ml⁻¹; TY/23, BioXCell) in round (U)-bottom 96-well plates at 37 °C, 5% CO₂ for 3 days. T cell proliferation was assessed by flow cytometry and compared with cells treated with the equivalent amount of DMSO or isotype-matched control antibodies (2A3, BioXCell).

To determine whether IL-1R signalling is involved in Spp1^hi-TAM-mediated T cell suppression, 1 × 10⁴ CD8⁺ T cells, labelled with BD Violet Proliferation Dye 450 (BDB562158, Fisher Scientific, 1:1,000) and activated with 1 × 10⁴ Dynabeads Mouse T-Activator CD3/CD28 (Gibco), were cultured with or without purified Spp1^hi-TAMs at a 1:1 ratio in 200 µl complete RMPI medium in the presence of either purified in vivo GOLD functional grade anti-mouse IL-1R (10 μg ml⁻¹, JAMA-147) or the relevant isotype-matched control antibody (PIP, both from Leinco Technologies) in round (U)-bottom 96-well plates at 37 °C, 5% CO₂ for 3 days. T cell proliferation was assessed by flow cytometry.

Extracellular adenosine detection

Spp1^hi-TAMs and MDSCs (both 1 × 10⁵ cells) were isolated from the same CRPC and plated in 250 µl complete RMPI medium in flat-bottom 48-well plates at 37 °C, 5% CO₂. After 24 h, supernatants were collected and adenosine levels were measured using an adenosine assay kit (Fluoreometric, ab211094, Abcam) following the manufacturer’s protocol. Fluorescence was measured using a GluoMax plate reader (Promega; Ex/Em = 535/587 nm), and the concentration of accumulated extracellular adenosine was calculated by subtracting the baseline adenosine levels from medium without cells.

Transwell assays

Transwell assays were performed as previously reported⁶⁵. In brief, FACS-isolated Spp1^hi-TAMs or enriched CD8⁺ T cells labelled with BD Violet Proliferation Dye 450 (BDB562158, Fisher Scientific, 1:1,000), from CRPC developed in mice or their spleens respectively, were plated into the top or bottom chambers of Transwell plates containing 6.5 mm cell culture inserts with 0.4 µm pore polyester membrane (CLS3379, Corning), as depicted in Extended Data Fig. 6e. In the top chamber of the inserts, 1 × 10⁴ CD8⁺ T cells alone or those stimulated with 1 × 10⁴ Dynabeads Mouse T-Activator CD3/CD28 (Gibco) were plated, and in the bottom chamber, 6 × 10⁴ Spp1^hi-TAMs or 6 × 10⁴ CD8⁺ T cells with or without anti-CD3/28 stimulation at a 1:1 ratio were plated, each with 100 µl or 600 µl complete RMPI medium, respectively. After 3 days of culture, T cell proliferation in each chamber was assessed by flow cytometry.

In vivo treatment of antibodies or inhibitors

To determine whether Spp1^hi-TAMs are resistant to CSF1R blockade, mice were randomly divided into two groups when they developed CRPC (tumour volume of 100–200 mm³) and were administered intraperitoneally 1 mg anti-mouse CSF1R (AFS98, BioXCell) or the respective isotype-matched control (2A3, BioXCell) antibodies in 200 μl PBS. A maintenance dose of 0.5 mg in 200 μl PBS was given after 5 days. The myeloid composition was analysed by flow cytometry 2 days after the final injection.

For immune checkpoint inhibition, mice with established CRPC (tumour volume of 100–200 mm³) were randomly divided into four groups and subjected to intraperitoneal injection with these antibodies in 200 μl PBS every 3 days for a total of three injections: 200 μg anti-mouse CTLA-4 (24H2)⁶⁶ alone; 400 μg anti-mouse PD-1 (17D2)⁶⁷ alone; a combination of anti-mouse CTLA-4 and PD-1; or the respective IgG2a, κ isotype-matched control. Tumour burden was measured every 2–3 days after the initial injection until it reached 750 mm³, unless otherwise indicated.

To determine whether Spp1^hi-TAMs contribute to resistance to ICIs in vivo, mice with developed CRPC (tumour volume of 100–200 mm³) were randomly divided into three groups. They were administered with: a combination of anti-mouse CTLA-4 and PD-1 in 200 μl PBS injected intraperitoneally along with intratumoral injection of 1 × 10⁵ Spp1^hi-TAMs purified from digested CRPC (more than 350 mm³) of a mouse from the same cohort in 50 μl PBS; a combination of anti-mouse CTLA-4 and PD-1 in 200 μl PBS injected intraperitoneally along with 50 μl of PBS intratumorally; or the respective isotype-matched control antibody in 200 μl PBS injected intraperitoneally along with 50 μl PBS intratumorally. Antibodies were administered every 3 days for a total of three injections, and Spp1^hi-TAMs were adoptively transferred every 5 days for a total of two injections. Tumour growth was measured every 2–3 days after the initial injection until it reached 750 mm³, unless otherwise indicated. The lymphoid composition was analysed by flow cytometry one day after the final injection.

For blockade of adenosine receptors (A2ARs), mice with established CRPC (tumour volume, 100–200 mm³) were randomly divided into two groups. Ciforadenant (10 mg per kg, Corvus Pharmaceuticals) or DMSO vehicle control (Sigma-Aldrich) in 200 μl of injection solution was administered once daily through oral gavage for 10 consecutive days. The injection solution consisted of 10% ciforadenant (or DMSO medium) and 90% corn oil (MedchemExpress). Tumour growth was measured every 2–3 days after the initial injection.

To determine whether A2AR blockade enhances immunotherapy efficacy, mice with established CRPC (tumour volume, 100–200 mm³) were randomly divided into two groups. Ciforadenant (10 mg per kg, Corvus Pharmaceuticals) or DMSO vehicle control (Sigma-Aldrich) in 200 μl of injection solution described above was administered once daily by oral gavage for 10 consecutive days. Simultaneously, mice were injected intraperitoneally with 400 µg anti-mouse PD-1 or the respective isotype-matched control antibodies in 200 µl PBS every 3 days for a total of three injections. Tumour growth was monitored every 2–3 days after the initial injection. The lymphoid and myeloid compositions were analysed by flow cytometry 1–2 h after the eighth injection of ciforadenant (1 day after the final anti-mouse PD-1 antibody injection).

All comparisons within experiments were carried out using age-matched mice (6–10 weeks old) engrafted with the same stock of MyC-CaP throughout the study.

scRNA-seq of prostate cancer mouse samples

For the single-cell assessment of MyC-CaP, a cohort of FVB/NJ mice bearing MyC-CaP were injected subcutaneously with either degarelix (n = 3) or PBS (n = 3), as described above. HSPC or CRPC tissues were collected on reaching a tumour volume of more than 350 mm³. Tumours were processed and single-cell suspensions were prepared as described above. For the cell-surface protein staining, cells were incubated with CD45.1-Brilliant Violet 510 (A20) for 30 min at 4 °C. After immunostaining, cells were washed twice in FWB and resuspended in FWB containing 1 μg ml⁻¹ propidium iodide for viability assessment. Cells were then sorted into CD45⁺ and CD45⁻ populations using FACSAria (BD Biosciences). Each population was transferred into separate 75 mm flow-cytometry tubes, centrifuged for 5 min at 250g at 4 °C and the supernatant was discarded. Cells were then resuspended in 100 μl Fc blocking buffer, consisting of 95 μl FWB + 5 μl of 0.5 mg ml⁻¹ anti-mouse CD16/32 antibody (2.4G2, Cytek Biosciences), followed by a 10-minute incubation at 4 °C. CD45⁺ cells were subsequently stained directly with 2 μl of 0.05 mg ml⁻¹ TotalSeq-C hashtag antibodies 2, 4 and 6 (M1/42, 30-F11) without washing, for 40 min at 4 °C. All sorted populations from each tissue were then pooled to yield a total of 1 × 10⁶ cells and these cells were stained with 100 μl of a cocktail of TotalSeq-C surface antibodies (CD11c (N418), CD163 (S15049I), F4/80 (BM8) and Ly-6G (1A8), each at a concentration of 2.5 μg ml⁻¹) for 30 min at 4 °C. After staining, cells were washed with 1 ml complete RPMI medium and filtered through a 70 µm cell strainer. Cell viability and counting were reassessed and the volumes were adjusted for 10x chromium chip input at a concentration of 1.29 × 10⁶ cells per ml. For scRNA-seq of TRAMP-C2, single-cell suspensions were initially labelled with LIVE/DEAD Fixable Dead Cell Stain Kit (Near-IR; Invitrogen) for 10 min at 4 °C. Subsequently, cells were stained with anti-mouse CD16/32 antibody (2.4G2) and CD45-Brilliant Violet 570 (30-F11) antibodies for 30 min on ice. After immunostaining, cells were washed with FWB and sorted into CD45⁺ and CD45⁻ populations using a FACSAria2 (BD Biosciences). Each sorted population was transferred into separate 75 mm flow-cytometry tubes, centrifuged for 5 min at 250g at 4 °C and the supernatant was discarded. Cells were then resuspended in 100 μl Fc blocking buffer as above, followed by a 10-minute incubation at 4 °C. Cells were then stained directly with 2 μl of 0.05 mg ml⁻¹ TotalSeq-C hashtag antibodies 1 and 2 (M1/42, 30-F11) for 40 min at 4 °C without washing. Equal proportions of cells labelled with hashtags were pooled together, and three individual reactions, each containing a total of 6 × 10⁴ cells, were washed with 1 ml complete RPMI medium and filtered through a 70 µm cell strainer. After reassessing cell viability and counting, cell concentrations were adjusted to 1.29 × 10⁶ cells per ml for loading into the 10x chromium chip. A 10x Genomics chromium controller was used to create GEM bead emulsions using a Single Cell 5′ Library & Gel Bead Kit (10x Genomics), followed by synthesis and amplification of cDNA and subsequent library preparation steps using 10x Genomics kits. The UCSF Institute for Human Genetics core performed library sequencing on a NovaSeq 6000 (Illumina), targeting a median read depth of 150,000 reads per cell for total gene expression libraries and 60,000 reads per cell for CITE-seq libraries. All antibodies were obtained from BioLegend, unless otherwise indicated.

Mouse scRNA-seq analysis

The raw data obtained from 10× sequencing were processed through the Cell Ranger pipeline (v.5, Genome build GRCm38). Subsequent steps in the analysis were the same as those used in human scRNA-seq until batch correction using Harmony, followed by Leiden clustering (resolution = 1.0) and UMAP plotting. A resolution of 1.0 was applied for myeloid-cell clustering. Differential expression analysis was done to identify the top-ranked genes upregulated in each individual cluster compared with the combination of all other cells. This analysis was done using the SCANPY function tl.rank_genes_groups. Annotation of each population was established through manual examination of the top-ranked genes in each cluster. To analyse cell density on the UMAP, the SCANPY function tl.embedding_density was used. Box plots were generated to visually represent cell population frequencies for each cell type. Gene scores were computed using the SCANPY function tl.score_genes with curated gene lists provided. Gene scores were computed using the SCANPY function tl.score_genes for each cell, using curated gene lists. To calculate similarity scores between myeloid subsets in humans and mice, a pseudo-bulk analysis was done to aggregate gene-expression data from the cellular level to the sample level. Z-scores were computed for each gene on the basis of cells in a given sample, and the mean was determined as the representative value for the sample. We then identified the shared genes in both human and mouse datasets, focusing on the top 50 genes within each subset.

Immunostaining and microscopy

For immunostaining of SPP1^hi-TAMs and CD4⁺ or CD8⁺ T cells in human tissues, in situ hybridization was done using RNAscope (Advanced Cell Diagnostics, ACDBio) on FFPE sections 4 μm thick from patients with either HSPC or mCRPC (NCT03007732, NCT03248570 and NCT02655822). Tissues were pretreated with target retrieval reagents and protease to improve target recovery according to the RNAscope Multiplex Fluorescent Reagent Kit v.2 assay protocol (323100, ACD Bio). Probes for human SPP1 and CD68 mRNA (420101-C2 and 560591-C4, respectively; ACDBio) were applied at a 1:50 dilution for 2 h at 40 °C. The probes were then hybridized with Opal 7-Color Manual IHC Kit (NEL811001KT, PerkinElmer) for the detection of SPP1 and CD68 transcripts using Opal 650 and Opal 690, respectively, at a dilution of 1:700. Immunofluorescence staining for human CD4 (MA-12259, 4B12, Invitrogen) and CD8 (ab60076, YTC182.20, abcam) was then done at a 1:100 dilution each. Targets were detected using Alexa Fluor 488-conjugated donkey anti-mouse IgG secondary antibody (ab150105, abcam) at a 1:100 dilution and Alexa Fluor 555-conjugated goat anti-rabbit IgG secondary antibody (4050-32, Southern Biotech) at a 1:100 dilution. Tissues were counterstained with 4′,6diamidino-2-phenylindole (DAPI, ACD Bio) and mounted with ProLong Gold Antifade Mountant (P36930, Invitrogen). Slides were imaged at 63× magnification using a Leica SP8 X white-light laser confocal microscope (Leica Microsystems) with multiple regions of interest from each specimen slide randomly selected for analysis. No staining was observed using negative control probes specific for the bacterial dapB gene (321831, ACD Bio) counterstained with Opal dyes, or with secondary antibodies alone on tonsil tissue.

Immunostaining of PD-L1 expression on EpCAM⁺ cells and CD68⁺ cells in human tissues was done on FFPE tissue sections 4 µm thick from responders and non-responders in clinical trial NCT02655822. This staining was done using a Ventana DISCOVERY ULTRA automated slide stainer and Ventana DISCOVERY ULTRA reagents (Roche Diagnostics), according to the manufacturer’s instructions (UCSF Protocol 3612), unless otherwise indicated. After deparaffinization, antigen retrieval was performed with cell conditioning 1 solution for 64 min at 97 °C. Primary antibodies for human CD68 (PG-M1; Agilent), PD-L1 and EpCAM (E1L3N and D9S3P, respectively, Cell Signaling Technology) were applied at 1:200, 1:100 and 1:50 dilutions for 32 min, respectively, at 36 °C. Goat Ig Block Ventana (760-6008) was applied for 4 min before the secondary antibodies (OmniMap anti-Ms for the anti-CD68 antibody and OmniMap anti-Rb for the anti-PD-L1 and anti-EpCAM antibodies) were incubated for 12 min. A stripping step between each primary was done with cell conditioning 2 solution at 97 °C for 8 min between primary antibodies. Endogenous peroxidase was inhibited using DISCOVERY Inhibitor RUO Ventana (760-4840) for 12 min. The CD68 was visualized using DISCOVERY Rhodamine 6G Kit Ventana (760-244), PD-L1 was visualized with DISCOVERY Cy5 Kit (760-238) and EpCAM was visualized with DISCOVERY FAM Kit (RUO) (760-243) for 8 min each. Finally, slides were counterstained with spectral DAPI (FP1490, Akoya) for 8 min. Slides were scanned using an AxioScan.Z1 in a whole-slide scanner (Zeiss) with a Plan-Apochromat 20×/0.8 M27 objective lens. Images were captured using an Orca-Flash 4.0 v.2 CMOS camera (Hamamatsu).

Immunostaining of mouse tissues was done on 5-μm acetone-fixed cryosections following standard protocols, as previously described⁶⁸. Sections were immunostained with the following antibodies: anti-mouse F4/80-Alexa Fluor 647 (565853, T45-2342, BD Biosciences) at a 1:200 dilution, and Spp1-EGFP was amplified using chicken anti-GFP antibody (ab13970, abcam) at a 1:2,000 dilution, followed by donkey anti-chicken IgY(IgG)-DyLight 405 (703-475-155, Jackson ImmunoResearch) at a 1:500 dilution. After staining, slides were washed, stained with DAPI to detect nuclei and mounted with ProLong Gold Antifade Mountant (P36930, Invitrogen). Images were obtained on a Leica DMi8 microscope with a 63×/1.32 oil objective lens and a Leica DFC9000 GTC digital microscope camera, with LAS X software (v.3.5.7.23225). Images were processed using ImageJ (v.2.14.0/1.54 f) for fluorescent channel overlays and uniform exposure adjustment.

Statistical analysis

Statistical analyses were done using Prism (v.10, GraphPad Software). Normality was determined with the D’Agostino & Pearson or Shapiro–Wilk tests, chosen according to sample size. Statistical significance was determined using two-sided unpaired Student’s t-tests for normally distributed data or the non-parametric Mann–Whitney test, two-sided paired Student’s t-tests, one sample t-tests, ordinary one-way ANOVA with Sidak correction for normally distributed data or the non-parametric Kruskal–Wallis test with Dunn’s correction, ordinary two-way ANOVA with Sidak correction, simple linear regression analyses, Wilcoxon tests with Benjamini–Hochberg correction, or log-rank tests, as indicated in the figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The data generated in this study are available in the article and its supplementary data files. The human and mouse scRNA-seq data have been deposited in the Gene Expression Omnibus database under accession number GSE274229. The human and mouse genome assemblies, GRCh38 and GRCm38, were obtained from the NIH National Library of Medicine website. Source data are provided with this paper.

References

Beer, T. M. et al. Randomized, double-blind, phase III trial of ipilimumab versus placebo in asymptomatic or minimally symptomatic patients with metastatic chemotherapy-naive castration-resistant prostate cancer. J. Clin. Oncol. 35, 40–47 (2017).
Article CAS PubMed Google Scholar
Powles, T. et al. Atezolizumab with enzalutamide versus enzalutamide alone in metastatic castration-resistant prostate cancer: a randomized phase 3 trial. Nat. Med. 28, 144–153 (2022).
Article CAS PubMed PubMed Central MATH Google Scholar
Martori, C. et al. Macrophages as a therapeutic target in metastatic prostate cancer: a way to overcome immunotherapy resistance? Cancers 14, 440 (2022).
Article CAS PubMed PubMed Central MATH Google Scholar
Koinis, F. et al. Myeloid-derived suppressor cells in prostate cancer: present knowledge and future perspectives. Cells 11, 20 (2021).
Article ADS PubMed PubMed Central Google Scholar
Siegel, R. L., Miller, K. D., Wagle, N. S. & Jemal, A. Cancer statistics, 2023. CA Cancer J. Clin. 73, 17–48 (2023).
Article PubMed Google Scholar
Scher, H. I. et al. Increased survival with enzalutamide in prostate cancer after chemotherapy. N. Engl. J. Med. 367, 1187–1197 (2012).
Article CAS PubMed MATH Google Scholar
Ribas, A. & Wolchok, J. D. Cancer immunotherapy using checkpoint blockade. Science 359, 1350–1355 (2018).
Article ADS CAS PubMed PubMed Central Google Scholar
Abida, W. et al. Analysis of the prevalence of microsatellite instability in prostate cancer and response to immune checkpoint blockade. JAMA Oncol. 5, 471–478 (2019).
Article PubMed MATH Google Scholar
Mahoney, K. M., Rennert, P. D. & Freeman, G. J. Combination cancer immunotherapy and new immunomodulatory targets. Nat. Rev. Drug Discov. 14, 561–584 (2015).
Article CAS PubMed Google Scholar
de Almeida, D. V. P., Fong, L., Rettig, M. B. & Autio, K. A. Immune checkpoint blockade for prostate cancer: niche role or next breakthrough? Am. Soc. Clin. Oncol. Educ. Book 40, e89–e106 (2020).
Article Google Scholar
Binnewies, M. et al. Understanding the tumor immune microenvironment (TIME) for effective therapy. Nat. Med. 24, 541–550 (2018).
Article CAS PubMed PubMed Central MATH Google Scholar
Witkowski, M. T., Kousteni, S. & Aifantis, I. Mapping and targeting of the leukemic microenvironment. J. Exp. Med. 217, e20190589 (2020).
Article PubMed MATH Google Scholar
Stultz, J. & Fong, L. How to turn up the heat on the cold immune microenvironment of metastatic prostate cancer. Prostate Cancer Prostatic Dis. 24, 697–717 (2021).
Article CAS PubMed PubMed Central Google Scholar
Ge, R., Wang, Z. & Cheng, L. Tumor microenvironment heterogeneity an important mediator of prostate cancer progression and therapeutic resistance. NPJ Precis. Oncol. 6, 31 (2022).
Article CAS PubMed PubMed Central MATH Google Scholar
Escamilla, J. et al. CSF1 receptor targeting in prostate cancer reverses macrophage-mediated resistance to androgen blockade therapy. Cancer Res. 75, 950–962 (2015).
Article CAS PubMed PubMed Central Google Scholar
Calcinotto, A. et al. IL-23 secreted by myeloid cells drives castration-resistant prostate cancer. Nature 559, 363–369 (2018).
Article ADS CAS PubMed PubMed Central Google Scholar
Lu, X. et al. Effective combinatorial immunotherapy for castration-resistant prostate cancer. Nature 543, 728–732 (2017).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Lopez-Bujanda, Z. A. et al. Castration-mediated IL-8 promotes myeloid infiltration and prostate cancer progression. Nat. Cancer 2, 803–818 (2021).
Article CAS PubMed PubMed Central Google Scholar
Autio, K. A. et al. Immunomodulatory activity of a colony-stimulating factor-1 receptor inhibitor in patients with advanced refractory breast or prostate cancer: a phase I study. Clin. Cancer Res. 26, 5609–5620 (2020).
Article CAS PubMed PubMed Central MATH Google Scholar
Siddiqui, B. A. et al. Immune and pathologic responses in patients with localized prostate cancer who received daratumumab (anti-CD38) or edicotinib (CSF-1R inhibitor). J. Immunother. Cancer 11, e006262 (2023).
Article PubMed PubMed Central Google Scholar
Cheng, S. et al. A pan-cancer single-cell transcriptional atlas of tumor infiltrating myeloid cells. Cell 184, 792–809 (2021).
Article CAS PubMed MATH Google Scholar
Zheng, L. et al. Pan-cancer single-cell landscape of tumor-infiltrating T cells. Science 374, abe6474 (2021).
Article PubMed Google Scholar
Luo, H. et al. Pan-cancer single-cell analysis reveals the heterogeneity and plasticity of cancer-associated fibroblasts in the tumor microenvironment. Nat. Commun. 13, 6619 (2022).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Tang, F. et al. A pan-cancer single-cell panorama of human natural killer cells. Cell 186, 4235–4251 (2023).
Article CAS PubMed MATH Google Scholar
Vickman, R. E. et al. Heterogeneity of human prostate carcinoma‐associated fibroblasts implicates a role for subpopulations in myeloid cell recruitment. Prostate 80, 173–185 (2020).
Article CAS PubMed Google Scholar
Hirz, T. et al. Dissecting the immune suppressive human prostate tumor microenvironment via integrated single-cell and spatial transcriptomic analyses. Nat. Commun. 14, 663 (2023).
Article ADS CAS PubMed PubMed Central Google Scholar
Jiao, S. et al. Differences in tumor microenvironment dictate T helper lineage polarization and response to immune checkpoint therapy. Cell 179, 1177–1190 (2019).
Article CAS PubMed MATH Google Scholar
Kfoury, Y. et al. Human prostate cancer bone metastases have an actionable immunosuppressive microenvironment. Cancer Cell 39, 1464–1478.e8 (2021).
Article CAS PubMed PubMed Central MATH Google Scholar
He, M. X. et al. Transcriptional mediators of treatment resistance in lethal prostate cancer. Nat. Med. 27, 426–433 (2021).
Article CAS PubMed PubMed Central MATH Google Scholar
Guo, C. et al. Targeting myeloid chemotaxis to reverse prostate cancer therapy resistance. Nature 623, 1053–1061 (2023).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Ramos, R. N. et al. Tissue-resident FOLR2⁺ macrophages associate with CD8⁺ T cell infiltration in human breast cancer. Cell 185, 1189–1207 (2022).
Article MATH Google Scholar
Chen, J. et al. PBMC fixation and processing for Chromium single-cell RNA sequencing. J. Transl. Med. 16, 198 (2018).
Article CAS PubMed PubMed Central MATH Google Scholar
Trovato, R. et al. Immunosuppression by monocytic myeloid-derived suppressor cells in patients with pancreatic ductal carcinoma is orchestrated by STAT3. J Immunother Cancer 7, 255 (2019).
Article PubMed PubMed Central MATH Google Scholar
Maire, C. L. et al. Glioma escape signature and clonal development under immune pressure. J. Clin. Invest. 130, 5257–5271 (2020).
Article CAS PubMed PubMed Central MATH Google Scholar
Bi, K. et al. Tumor and immune reprogramming during immunotherapy in advanced renal cell carcinoma. Cancer Cell 39, 649–661 (2021).
Article CAS PubMed PubMed Central MATH Google Scholar
Zhang, L. et al. Lineage tracking reveals dynamic relationships of T cells in colorectal cancer. Nature 564, 268–272 (2018).
Article ADS CAS PubMed MATH Google Scholar
Watson, P. A. et al. Context-dependent hormone-refractory progression revealed through characterization of a novel murine prostate cancer cell line. Cancer Res. 65, 11565–11571 (2005).
Article CAS PubMed MATH Google Scholar
Shen, Y.-C. et al. Combining intratumoral Treg depletion with androgen deprivation therapy (ADT): preclinical activity in the Myc-CaP model. Prostate Cancer Prostatic Dis. 21, 113–125 (2018).
Article CAS PubMed MATH Google Scholar
Foster, B. A., Gingrich, J. R., Kwon, E. D., Madias, C. & Greenberg, N. M. Characterization of prostatic epithelial cell lines derived from transgenic adenocarcinoma of the mouse prostate (TRAMP) model. Cancer Res. 57, 3325–3330 (1997).
CAS PubMed Google Scholar
Chen, W., Bamford, R. N., Edmondson, E. F. & Waldmann, T. A. IL-15 and anti-PD-1 augment the efficacy of agonistic intratumoral anti-CD40 in a mouse model with multiple TRAMP-C2 tumors. Clin. Cancer Res. 28, 2082–2093 (2022).
Article CAS PubMed PubMed Central Google Scholar
Movahedi, K. et al. Different tumor microenvironments contain functionally distinct subsets of macrophages derived from Ly6C(high) monocytes. Cancer Res. 70, 5728–5739 (2010).
Article CAS PubMed MATH Google Scholar
Bill, R. et al. CXCL9:SPP1 macrophage polarity identifies a network of cellular programs that control human cancers. Science 381, 515–524 (2023).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Augustin, R. C. et al. Next steps for clinical translation of adenosine pathway inhibition in cancer immunotherapy. J. Immunother. Cancer 10, e004089 (2022).
Article PubMed PubMed Central Google Scholar
Vijayan, D., Young, A., Teng, M. W. L. & Smyth, M. J. Targeting immunosuppressive adenosine in cancer. Nat. Rev. Cancer 17, 709–724 (2017).
Article CAS PubMed MATH Google Scholar
Ohta, A. & Sitkovsky, M. Role of G-protein-coupled adenosine receptors in downregulation of inflammation and protection from tissue damage. Nature 414, 916–920 (2001).
Article ADS CAS PubMed Google Scholar
Ohta, A. et al. A2A adenosine receptor protects tumors from antitumor T cells. Proc. Natl Acad. Sci. USA 103, 13132–13137 (2006).
Article ADS CAS PubMed PubMed Central Google Scholar
Young, A. et al. A2AR adenosine signaling suppresses natural killer cell maturation in the tumor microenvironment. Cancer Res. 78, 1003–1016 (2018).
Article PubMed MATH Google Scholar
Sidders, B. et al. Adenosine signaling is prognostic for cancer outcome and has predictive utility for immunotherapeutic response. Clin. Cancer Res. 26, 2176–2187 (2020).
Article CAS PubMed MATH Google Scholar
Ferretti, E., Horenstein, A. L., Canzonetta, C., Costa, F. & Morandi, F. Canonical and non-canonical adenosinergic pathways. Immunol. Lett. 205, 25–30 (2019).
Article CAS PubMed Google Scholar
Wang, L. et al. Myeloid cell-associated resistance to PD-1/PD-L1 blockade in urothelial cancer revealed through bulk and single-cell RNA sequencing. Clin. Cancer Res. 27, 4287–4300 (2021).
Article CAS PubMed PubMed Central MATH Google Scholar
Willingham, S. B. et al. A2AR antagonism with CPI-444 induces antitumor responses and augments efficacy to anti–PD-(L)1 and anti–CTLA-4 in preclinical models. Cancer Immunol. Res. 6, 1136–1149 (2018).
Article CAS PubMed Google Scholar
Leone, R. D. et al. Inhibition of the adenosine A2a receptor modulates expression of T cell coinhibitory receptors and improves effector function for enhanced checkpoint blockade and ACT in murine cancer models. Cancer Immunol. Immunother. 67, 1271–1284 (2018).
Article CAS PubMed PubMed Central Google Scholar
Fong, L. et al. Adenosine 2A receptor blockade as an immunotherapy for treatment-refractory renal cell cancer. Cancer Discov. 10, 40–53 (2020).
Article CAS PubMed MATH Google Scholar
Mittal, D. et al. Antimetastatic effects of blocking PD-1 and the adenosine A2A receptor. Cancer Res. 74, 3652–3658 (2014).
Article CAS PubMed MATH Google Scholar
Beavis, P. A. et al. Adenosine receptor 2A blockade increases the efficacy of anti–PD-1 through enhanced antitumor T-cell responses. Cancer Immunol. Res. 3, 506–517 (2015).
Article CAS PubMed MATH Google Scholar
LaMarche, N. M. et al. An IL-4 signalling axis in bone marrow drives pro-tumorigenic myelopoiesis. Nature 625, 166–174 (2024).
Article ADS CAS PubMed MATH Google Scholar
Zhang, L. et al. Single-cell analyses inform mechanisms of myeloid-targeted therapies in colon cancer. Cell 181, 442–459 (2020).
Article CAS PubMed MATH Google Scholar
Masetti, M. et al. Lipid-loaded tumor-associated macrophages sustain tumor growth and invasiveness in prostate cancer. J. Exp. Med. 219, e20210564 (2022).
Article CAS PubMed MATH Google Scholar
Alonso-Gordoa, T. et al. A phase II study (AARDVARC) of AZD4635 in combination with durvalumab and cabazitaxel in patients with progressive, metastatic, castration-resistant prostate cancer. ESMO Open 9, 103446 (2024).
Article CAS PubMed PubMed Central MATH Google Scholar
Matthews, E. et al. Down-regulation of TGF-β1 production restores immunogenicity in prostate cancer cells. Br. J. Cancer 83, 519–525 (2000).
Article CAS PubMed PubMed Central MATH Google Scholar
Fleming, S. J. et al. Unsupervised removal of systematic background noise from droplet-based single-cell experiments using CellBender. Nat. Methods 20, 1323–1335 (2023).
Article CAS PubMed MATH Google Scholar
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Article PubMed PubMed Central MATH Google Scholar
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019).
Article CAS PubMed PubMed Central MATH Google Scholar
Gong, S. et al. A gene expression atlas of the central nervous system based on bacterial artificial chromosomes. Nature 425, 917–925 (2003).
Article ADS CAS PubMed MATH Google Scholar
Lyu, A. et al. Integrin signaling is critical for myeloid-mediated support of T-cell acute lymphoblastic leukemia. Nat. Commun. 14, 6270 (2023).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Pai, C.-C. S. et al. Tumor-conditional anti-CTLA4 uncouples antitumor efficacy from immunotherapy-related toxicity. J. Clin. Invest. 129, 349–363 (2019).
Article PubMed Google Scholar
Pai, C.-C. S. et al. Clonal deletion of tumor-specific T cells by interferon-γ confers therapeutic resistance to combination immune checkpoint blockade. Immunity 50, 477–492 (2019).
Article CAS PubMed PubMed Central Google Scholar
Lyu, A. et al. Tumor-associated myeloid cells provide critical support for T-ALL. Blood 136, 1837–1850 (2020).
Article PubMed MATH Google Scholar

Download references

Acknowledgements

We acknowledge the patients who volunteered to participate in these studies; the University of California San Francisco (UCSF) Genitourinary Medical Oncology and Urology providers involved in screening, enrolment and clinical care of these patients; the UCSF Biospecimen Resources Program for help with tissue acquisition; the Institute for Human Genetics Core and the Center for Advanced Technologies at UCSF for help with sequencing; the UCSF Laboratory Animal Resource Center for technical assistance; and the UCSF Parnassus Flow Cytometry CoLab for help with Flow/Mass Cytometry data through, in part, Diabetes Research Center grant US National Institutes of Health (NIH) P30 DK063720 and NIH S10 Instrumentation grant S10 1S10OD021822-01. We also thank all members of the Fong lab, H. Jung and R. S. Heist for discussion and advice. This work was supported by the Parker Institute of Cancer Immunotherapy, the Prostate Cancer Foundation and the NIH. L.F. was supported by the Parker Institute of Cancer Immunotherapy, the Prostate Cancer Foundation and NIH grants U01CA233100 and R35CA253175. E.M.V.A. was supported by the Parker Institute of Cancer Immunotherapy, the Prostate Cancer Foundation and NIH grants U01CA233100. A. Lyu was supported by a Parker Scholar Award from the Parker Institute of Cancer Immunotherapy.

Author information

These authors contributed equally: Aram Lyu, Zenghua Fan

Authors and Affiliations

Division of Hematology/Oncology, Department of Medicine, University of California, San Francisco, San Francisco, CA, USA
Aram Lyu, Zenghua Fan, Matthew Clark, Averey Lea, Diamond Luong, Ali Setayesh, Alec Starzinski, Rachel Wolters, Marcel Arias-Badia, Kate Allaire, Kai Wu, Vibha Gurunathan & Lawrence Fong
Parker Institute for Cancer Immunotherapy, San Francisco, CA, USA
Aram Lyu, Eliezer M. Van Allen & Lawrence Fong
Immunotherapy Integrated Research Center, Division of Translational Science and Therapeutics, Fred Hutchinson Cancer Center, Seattle, WA, USA
Aram Lyu & Lawrence Fong
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA, USA
Laura Valderrábano, Xiao X. Wei & Eliezer M. Van Allen
Broad Institute of MIT and Harvard, Cambridge, MA, USA
Laura Valderrábano & Eliezer M. Van Allen
Corvus Pharmaceuticals, Burlingame, CA, USA
Richard A. Miller
Helen Diller Family Comprehensive Cancer Center, University of California, San Francisco, San Francisco, CA, USA
Lawrence Fong

Authors

Aram Lyu
View author publications
Search author on:PubMed Google Scholar
Zenghua Fan
View author publications
Search author on:PubMed Google Scholar
Matthew Clark
View author publications
Search author on:PubMed Google Scholar
Averey Lea
View author publications
Search author on:PubMed Google Scholar
Diamond Luong
View author publications
Search author on:PubMed Google Scholar
Ali Setayesh
View author publications
Search author on:PubMed Google Scholar
Alec Starzinski
View author publications
Search author on:PubMed Google Scholar
Rachel Wolters
View author publications
Search author on:PubMed Google Scholar
Marcel Arias-Badia
View author publications
Search author on:PubMed Google Scholar
Kate Allaire
View author publications
Search author on:PubMed Google Scholar
Kai Wu
View author publications
Search author on:PubMed Google Scholar
Vibha Gurunathan
View author publications
Search author on:PubMed Google Scholar
Laura Valderrábano
View author publications
Search author on:PubMed Google Scholar
Xiao X. Wei
View author publications
Search author on:PubMed Google Scholar
Richard A. Miller
View author publications
Search author on:PubMed Google Scholar
Eliezer M. Van Allen
View author publications
Search author on:PubMed Google Scholar
Lawrence Fong
View author publications
Search author on:PubMed Google Scholar

Contributions

A. Lyu designed experiments, performed experiments, analysed results, generated figures and wrote the manuscript; Z.F. analysed results, generated figures and wrote the manuscript; M.C., A. Lea, D.L., A. Setayesh, A. Starzinski, R.W., M.A.-B., K.A., V.G., and X.X.W. performed experiments; K.W. and L.V. analysed results; R.A.M. and E.M.V.A. provided advice and materials; and L.F. designed experiments, analysed results and wrote the manuscript.

Corresponding author

Correspondence to Lawrence Fong.

Ethics declarations

Competing interests

L.F. has received research support from Roche/Genentech, Abbvie, Bavarian Nordic, Bristol Myers Squibb, Dendreon, Janssen and Merck; has served on the scientific advisory boards of Actym, Astra Zeneca, Bioatla, Bristol Myers Squibb, Daiichi Sankyo, Immunogenesis, Innovent, Merck, Merck KGA, Nutcracker, RAPT, Senti, Sutro and Roche/Genentech; and reports ownership interest in Actym, BioAtla, ImmunoGenesis, Nutcracker, RAPT, Senti and Therapaint. R.A.M. is an employee of Corvus Pharmaceuticals. E.M.V.A. has received research support from Novartis, BMS and Sanofi, and served on the scientific advisory boards of Tango Therapeutics, Genome Medical, Genomic Life, Enara Bio, Manifold Bio, Monte Rosa, Novartis Institute for Biomedical Research, Riva Therapeutics and Serinus Bio. The remaining authors declare no competing interests.

Peer review

Peer review information

Nature thanks Johann de Bono and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Single-cell analysis of prostate cancer patient biopsies identifies diverse cell types including multiple myeloid subsets.

(a-b) Bubble plots showing the relative expression levels of signature genes for (a) the indicated major cell types and (b) myeloid subsets. (c-d) (c) UMAP plot showing the enrichment scores for published FOLR2⁺ macrophage signatures and (d) heatmap depicting the normalized expression levels of FOLR2, SELENOP, and SLC40A1 in the indicated myeloid subsets. (e) Representative RNAscope images of HSPC and mCRPC patient biopsy samples. Immunostaining for SPP1^hi-TAMs (SPP1; white, CD68; turquoise, and DAPI; blue) is presented. Scale bars, 30 μm. (f) Quantitative comparison of MDSC (P = 0.004) and SPP1^hi-TAM (P = 0.34) frequency between scRNA-seq analysis versus tissue imaging. Bars show mean + SEM; symbols represent individual patients. (g) Bubble plot showing the relative expression levels of signature genes associated with immunosuppression. Statistical significance was determined by (f) two-sided paired Student’s t-tests; P-values: **P < 0.01. ns, not significant.

Extended Data Fig. 2 CD8⁺ T cells exhibit significant exhaustion in mCRPC, coinciding with a notable increase in the abundance of SPP1^hi-TAMs.

(a-b) UMAP plots showing (a) distinct T-cell subsets and (b) the relative expression levels of CD4 and CD8 transcripts across T cells in human prostate cancer. (c-d) Bubble plots depicting the relative expression levels of signature genes for (c) the indicated T-cell subsets and (d) exhaustion states in CD4⁺ or CD8⁺ T cells across different disease stages, as indicated. (e-f) Quantification of enrichment scores for (e) CD8⁺ T-cell exhaustion and (f) SPP1^hi-TAMs across different disease stages, indicated by localized disease (n = 13), HSPC (n = 24), and mCRPC (n = 6) in gray, blue, and red, respectively. Boxes denote inter-quartile range (IQR), while bars denote 25% - 1.5 x IQR and 75% + 1.5 x IQR, with outliers exceeding 1.5 x the IQR beyond lower and upper quartiles. In (e), P = 0.96, 0.03, and 0.049 for localized vs. HSPC, localized vs. mCRPC, and HSPC vs. mCRPC, respectively. In (f), P = 0.51, 0.16, and 0.01 for localized vs. HSPC, localized vs. mCRPC, and HSPC vs. mCRPC, respectively. (g) UMAP plots showing the relative expression levels of CSF1R and MRC1 transcripts across myeloid cells in human prostate cancer. Statistical significance was determined by (e, f) ordinary one-way ANOVA with the Sidak correction; P-values: *<0.05. ns, not significant.

Extended Data Fig. 3 Single-cell analysis of mouse prostate cancer reveals diverse cell types including multiple myeloid subsets.

(a) Bubble plot showing the relative expression levels of signature genes for the indicated major cell types in mouse prostate cancer. (b) UMAP plots showing the relative expression levels of Cd68, Cd163, Itgax, and Csf3r transcripts (top) and the protein levels of F4/80, CD163, CD11c, and Ly-6G (bottom) across myeloid cells. (c) Bubble plot depicting the relative expression levels of signature genes for the indicated myeloid subsets. (d) UMAP plots showing the relative expression levels of Csf1r and Mrc1 transcripts across myeloid cells.

Extended Data Fig. 4 Flow cytometry analysis of the myeloid composition in MyC-CaP-derived HSPC and CRPC developed in Spp1-EGFP mice, and single-cell assessment of the myeloid landscape in TRAMP-C2.

(a) Experimental schematic depicting the evaluation of the myeloid compartment in CRPC developed in Spp1-EGFP mice following degarelix treatment. (b) Representative sequential gating schemes for evaluation of the indicated myeloid subsets in mice bearing CRPC. The initial plot is pre-gated on live, singlet, myeloid (CD11b⁺) cells. Sequential gating is indicated by arrows. (c-d) Quantification of the (c) number and (d) frequency of the indicated myeloid subsets from MyC-CaP engrafted into mice treated with degarelix (CRPC; red; n = 10) or PBS control (HSPC; blue; n = 8). In (c), P = 0.70, 0.02, 0.36, 0.02, and 0.84 for eosinophil, MDSC, CD163^hi-TAM, CX3CR1^hi-TAM, and Spp1^hi-TAM, respectively. In (d), P = 0.11, 0.02, 0.70, 0.049, and 0.01 for eosinophil, MDSC, CD163^hi-TAM, CX3CR1^hi-TAM, and Spp1^hi-TAM, respectively. Bars represent the mean + SEM from 4 independent experiments; symbols represent individual mice from each experiment. (e-f) (e) Representative flow cytometry plots and (f) quantification of cell surface CX3CR1 (left) or intracellular Spp1 (right) expression levels of the indicated macrophage subsets from CRPC (n = 11). Isotype control stains are in gray. In (f), P < 0.001 for all comparisons, except for Spp1-EGFP MFI between CD163^hi-TAM and Spp1^hi-TAM, where P = 0.002. Bars represent the mean + SEM from the same experiments as in (c, d); symbols represent individual mice from each experiment. (g) Bubble plot showing the relative expression levels of signature genes associated with immunosuppression. (h) Experimental schematic depicting 10x Genomics 5’ scRNA-seq on immune (CD45⁺) and non-immune (CD45^-) cells isolated from mice subcutaneously engrafted with mouse prostate cancer, TRAMP-C2, followed by treatment with anti-PD-1 or isotype antibodies. (i) Graph shows the cumulative growth of TRAMP-C2 engrafted into healthy wild-type mice, where blue and gray represent groups treated with anti-PD-1 antibody (n = 6) and isotype antibody (n = 3), respectively. Symbols represent mean +/- SEM from 2 independent experiments. (j) UMAP plots showing major cell types (left) and distinct myeloid subsets (right) in TRAMP-C2. (k-l) Quantification of (k) signature scores for SPP1^hi-TAMs across macrophages and monocytes (n = 9,460 cells) and (l) the frequency of Spp1^hi-TAMs in TRAMP-C2 engrafted in mice treated with either anti-PD-1 or isotype control antibodies. In (k), P < 0.001 for all comparisons between Spp1^hi-TAM vs. each indicated subset. Boxes denote IQR, while bars denote 25% - 1.5 x IQR and 75% + 1.5 x IQR, with outliers exceeding 1.5 x the IQR beyond lower and upper quartiles. In (l), bars show mean + SEM; symbols represent individual mice (P = 0.93). The red line indicates the median score of Spp1^hi-TAMs. (m-n) (m) UMAP and (n) bar plots showing the quantification of immunosuppression gene signature scores among macrophages and monocytes in TRAMP-C2 (n = 9,460 cells). In (n), P < 0.001 for all comparisons between Spp1^hi-TAM vs. each indicated subset. Boxes denote IQR, while bars denote 25% - 1.5 x IQR and 75% + 1.5 x IQR, with outliers exceeding 1.5 x the IQR beyond lower and upper quartiles. The red line indicates the median score of Spp1^hi-TAMs. Statistical significance was determined by (c, d, l) two-sided unpaired Student’s t-tests, (f) repeated measures one-way ANOVA with the Sidak correction, (i) a log-rank test, and (k, n) Kruskal-Wallis tests with the Dunn’s correction; P-values: *<0.05, **P < 0.01, ***<0.001. ns, not significant.

Extended Data Fig. 5 Functional assays demonstrating the immune suppressive activity of myeloid cells in CRPC, and a significant increase in exhausted CD8⁺ T cells following the transfer of Spp1^hi-TAMs combined with CTLA-4 and PD-1 blockade in vivo.

(a) Representative flow cytometry plots showing the purity of each myeloid subset after FACS sorting. Myeloid subsets were gated based on the strategy shown at the top, with sequential gating strategies indicated by arrows. (b) Quantification of the frequency of proliferating CD8⁺ T cells 3 days after co-culture with either MDSC (P = 0.04) or CX3CR1^hi-TAMs (P = 0.03 and 0.14 for E:T = 1:1 and 1:5, respectively) FACS-isolated from CRPC at the indicated ratios. Results were normalized to cultures with activated T cells alone. Bars show the mean + SEM from 4 independent experiments, each indicated by a distinct color; symbols represent averages of 2-3 technical replicate wells. Red lines indicate the normalized mean frequency of activated CD8⁺ T cells cultured alone. (c) Representative flow cytometry plots showing a decrease in the frequency of polyfunctional (IFN-γ⁺TNF-α⁺) CD8⁺ T cells 3 days after co-culture with Spp1^hi-TAMs at a 1:1 ratio from the same experiments as in Fig. 3e. (d) Graph depicting the cumulative growth of CRPC following treatment with anti-CTLA-4 +/- anti-PD-1 antibodies (P = 0.33, 0.49, and 0.04 for isotype vs. anti-CTLA-4, anti-PD-1, and anti-CTLA-4 + anti-PD-1, respectively). (e) Representative flow cytometry plots showing an increased frequency of exhausted (CD38⁺PD-1⁺) CD8⁺ T cells following the transfer of Spp1^hi-TAMs in combination with anti-CTLA-4 and anti-PD-1 antibodies, from the same experiments as in Fig. 3i. Statistical significance was determined by (b) one sample t-tests, and (d) log-rank tests; P-values: *<0.05, ***<0.001. ns, not significant.

Extended Data Fig. 6 Evaluation of gene signatures related to adenosine signaling pathways across multiple cell types, including SPP1^hi-TAMs, in human and mouse prostate cancers, and functional validation of the role of soluble factors in SPP1^hi-TAM-mediated immunosuppression.

(a) UMAP plots showing enrichment scores for the “Hypoxia” gene signatures across myeloid cells in both human and mouse prostate cancers. (b) Heatmaps depicting the normalized expression levels of ADORA2A and ADORA2B transcripts in the indicated CD8⁺ T cells and NK cells across different disease stages in patients. (c) UMAP plots showing enrichment scores for the “Adenosine Signaling Sig” gene signatures across myeloid cells in both human and mouse prostate cancers. (d) Plots depict the correlations between enrichment scores for the gene signatures “cDC1” (left) or “EEF1A1^hi-TAMs” (right) and enrichment scores for “Adenosine Signaling Sig” across different disease stages in patient samples. Localized disease, HSPC, and mCRPC are in gray, blue, and red, respectively. The best-fit lines are displayed, and individual patient samples are represented by circles. (e) Schematic illustration of transwell assays in which CD8⁺ T cells, with or without α-CD3/28 stimulation, are cultured in the presence or absence of FACS-isolated Spp1^hi-TAMs in opposite chambers to determine whether Spp1^hi-TAMs suppress T-cell proliferation via soluble factors. (f) Quantification of T-cell proliferation in the red-indicated chamber 3 days after culture initiation (P = 0.02, 0.002, and <0.001 for comparisons between the left vs. middle, middle vs. right, and left vs. right, respectively, as shown in the schematic in (e)). Results were normalized to the proliferation of activated T cells cultured alone in the top chamber of the inserts within each experiment. Bars show the mean + SEM from 3 independent experiments, each indicated by a distinct color; symbols represent averages of 2-3 technical replicate wells. Red lines indicate the normalized mean of activated CD8⁺ T cells cultured alone. (g) UMAP plot showing enrichment scores for the “Adenosine Signaling Sig” gene signatures across major cell types in patients. (h) Heatmap depicting the normalized expression levels of ENTPD1, NT5E, and CD38 transcripts in the indicated major cell types in patients. (i) Heatmaps depicting the normalized expression levels of CD38 transcripts in the indicated myeloid subsets both in humans and mice. (j) Heatmaps depicting the normalized expression levels of ENTPD1 and NT5E transcripts in the indicated tumor-associated macrophages across different disease stages in patients. Statistical significance was determined by (d) simple linear regression analyses, and (f) a repeated measures one-way ANOVA with the Sidak correction; P-values: *<0.05, **P < 0.01, ***<0.001. ns, not significant.

Extended Data Fig. 7 Assessment of CD39 and CD73 levels in Spp1^hi-TAMs in mouse prostate cancer, and investigation of the role of the IL-1R pathway in immunosuppression mediated by SPP1^hi-TAMs both in humans and mice.

(a) Heatmap depicting the normalized expression levels of Entpd1 and Nt5e transcripts in the indicated tumor-associated macrophages and monocytes across different disease stages in mice. (b-c) (b) Representative flow cytometry plots and (c) quantification of fold changes in the levels of cell surface CD73 expressed on Spp1^hi-TAMs from CRPC relative to HSPC. HSPC, CRPC, and isotype control stains are shaded in blue, red, and gray, respectively in (b). In (c), P = 0.35, 0.01, and 0.01 for CD163^hi-TAM, CX3CR1^hi-TAM, and Spp1^hi-TAM, respectively. Bars represent the mean + SEM from 4 independent experiments, each indicated by a distinct color; symbols represent individual mice from each experiment. The red line indicates a fold change of 1. (d) Pathways associated with inflammation, significantly enriched in SPP1^hi-TAMs compared to other myeloid cells in both human (red) and mouse (gray) prostate cancers, were identified using the Enrichr bioinformatics tool with GO Biological Process 2023 gene sets. (e) UMAP plots showing enrichment scores for the “Tumor-promoting Inflammation” gene signatures across myeloid cells in both human and mouse prostate cancers. (f) Plot depicting the correlation between enrichment scores for the gene signatures “SPP1^hi-TAMs” and enrichment scores for “Tumor-promoting Inflammation Sig” across different disease stages in patient samples. Localized disease, HSPC, and mCRPC are in gray, blue, and red, respectively. The best-fit line is displayed, and individual patient samples are represented by circles. (g) Bar plots showing decreased suppression of activated splenic CD8⁺ T cells when co-cultured with Spp1^hi-TAMs in the presence of an anti-IL-1R antibody (10 μg/ml) compared to isotype-treated cultures (P = 0.01). Bars show mean + SEM from 3 independent experiments, each indicated by a distinct color; symbols represent averages of 2-3 technical replicate wells. (h) Bar plots showing the percentage change in suppression of activated splenic CD8⁺ T cells mediated by Spp1^hi-TAMs in the presence of either ciforadenant (a A2AR inhibitor; 10 μM), anti-IL-1R antibody, or a combination of both. Bars show mean - SEM from 3 independent experiments, each indicated by a distinct color; symbols represent averages of 2-3 technical replicate wells. Statistical significance was determined by (c, g) two-sided one sample t-tests, (d) Fisher’s exact tests or the hypergeometric tests with the Benjamini-Hochberg correction, and (f) simple linear regression analysis; P-values: *<0.05. ns, not significant.

Extended Data Fig. 8 Evaluation of the lymphoid and myeloid compartments in CRPC treated with ciforadenant or DMSO control.

(a) Quantification of the frequency (left) or number (right) of CD8⁺ T cells in CRPC developed in mice treated with ciforadenant (10 mg/kg; n = 6) or DMSO vehicle control (n = 5) from the same experiments as in Fig. 5b. (b-c) (b) Representative flow cytometry plots and (c) quantification of the frequency of CD8⁺ T cells exhibiting exhaustion states (CD38⁺PD-1⁺) from the same experiments as in Fig. 5b (P = 0.01). (d) Quantification of the number of the indicated major myeloid subsets from the same experiments as in Fig. 5b. (e) Representative immunofluorescent images of Spp1^hi-TAMs in CRPC developed in either Spp1-EGFP negative (left) or positive mice treated with either DMSO (middle) or ciforadenant (right). Immunostaining for F4/80 (turquoise), Spp1 (magenta), and DAPI (blue) is shown. Scale bars, 10 μm. White arrows indicate Spp1^hi-TAMs. (f) Quantification of the frequency of Spp1^hi-TAMs in CRPC developed in mice treated with ciforadenant or DMSO vehicle control (P < 0.001). (g) Quantification of the number of the indicated macrophage subsets in CRPC developed in mice treated with ciforadenant or DMSO vehicle control from the same experiments as in Fig. 5b. (h) UMAP plot showing enrichment scores for the “AdenoSig” gene signatures across myeloid cells in human prostate cancer. (i) Plot depicts the correlations between enrichment scores for the gene signatures “SPP1^hi-TAMs” and “AdenoSig” in patient samples. Localized disease, HSPC, and mCRPC are in gray, blue, and red, respectively. The best-fit line is displayed, and individual patient samples are represented by circles. Bars represent the mean + SEM throughout this figure; symbols represent individual mice. Statistical significance was determined by (a, c, d, f, g) two-sided unpaired Student’s t-tests, and (i) a simple linear regression analysis; P-values: *<0.05, ***<0.001. ns, not significant.

Extended Data Fig. 9 Assessment of lymphoid and myeloid compartments in CRPC following treatment with ciforadenant +/− anti-PD-1 antibody.

(a-b) (a) Representative flow cytometry plots and (b) quantification showing an increase in the frequency of polyfunctional (IFN-γ⁺TNF-α⁺) CD8⁺ T cells following ciforadenant treatment with or without anti-PD-1 antibody, from the same experiments as in Fig. 5j. (c-d) Quantification of the frequency of (c) CD4⁺ T cells, CD8⁺ T cells, NK cells, and (d) Spp1^hi-TAMs in CRPC developed in Spp1-EGFP mice treated with ciforadenant +/− anti-PD-1 antibody. Treatment groups are indicated as follows: DMSO + isotype antibody (gray; n = 3), DMSO + anti-PD-1 antibody (blue; n = 2), ciforadenant + isotype antibody (olive green; n = 3), and ciforadenant + anti-PD-1 antibody (red; n = 3) (P = 0.049, 0.03, and 0.99 for comparisons between DMSO + isotype vs. ciforadenant + isotype, DMSO + isotype vs. ciforadenant + anti-PD-1, and ciforadenant + isotype vs. ciforadenant + anti-PD-1, respectively). Bars represent the mean + SEM throughout this figure; symbols represent individual mice. Statistical significance was determined by (d) ordinary one-way ANOVA with the Sidak correction; P-values: *<0.05. ns, not significant.

Extended Data Fig. 10 Clinical characteristics of the trial evaluating adenosine receptor blockade with ciforadenant +/− PD-L1 blockade using atezolizumab in mCRPC patients, and assessment of biopsies from responders and non-responders.

(a) Patient characteristics from the clinical trial as shown in Fig. 5k. (b) Treatment emergent adverse effects observed in the patients. (c) Representative immunofluorescent images of baseline biopsy samples from responders and non-responders in the clinical trial. Immunostaining for EpCAM (green), PD-L1 (magenta), CD68 (yellow), and DAPI (blue) is presented. Scale bars, 200 μm. (d) Representative images of SPP1^hi-TAMs in the same samples as in (c). Immunostaining for CD4 (red), CD8 (light green), SPP1 transcripts (white), CD68 transcripts (turquoise), and DAPI (blue) is presented. Scale bars, 30 μm. (c) and (d) represent data from one responder and two non-responders, which were the only samples available from this trial.

Extended Data Table 1 Details of prior treatments and genomic testing results for responders from the clinical trial evaluating ciforadenant and atezolizumab

Full size table

Supplementary information

Reporting Summary

Source data

Source Data Figs. 2–5

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Lyu, A., Fan, Z., Clark, M. et al. Evolution of myeloid-mediated immunotherapy resistance in prostate cancer. Nature 637, 1207–1217 (2025). https://doi.org/10.1038/s41586-024-08290-3

Download citation

Received: 19 January 2024
Accepted: 25 October 2024
Published: 04 December 2024
Issue date: 30 January 2025
DOI: https://doi.org/10.1038/s41586-024-08290-3

This article is cited by

The CTSZ-TRA2A-IL32 axis defines a targetable macrophage-dependent pathway in metastatic prostate cancer
- Saipeng Chen
- Bingqian Deng
- Wenhao Shen
Journal of Translational Medicine (2025)
Research progress of CD73-adenosine signaling regulating hepatocellular carcinoma through tumor microenvironment
- Liang Shan
- Mingxu Gong
- Xiongwen Lv
Journal of Experimental & Clinical Cancer Research (2025)
Macrophages in prostate cancer: dual roles in tumor progression and immune evasion
- Ruihong Liu
- Jianxin Lu
- Jingling Lai
Journal of Translational Medicine (2025)
The tumor microenvironment’s role in the response to immune checkpoint blockade
- Konstantinos Aliazis
- Anthos Christofides
- Vassiliki A. Boussiotis
Nature Cancer (2025)
A distinct TAM subset mediates immunotherapy resistance in prostate cancer
- Maria Chiara Masone
Nature Reviews Urology (2025)