Resolving the immune landscape of human prostate at a single-cell level in health and cancer

1The prostate gland is critical for human reproduction, generating prostatic fluid that is high in zinc and citrate. This forms an essential component of seminal fluid that is required for the maintenance of spermatozoa. Prostatic acini are comprised of an outer basal cell layerand inner layers of secretory luminal epithelial (LE) cells; as well as neuroendocrine, stromal, immune, endothelial, and nerve cells. Prostate cancer is a major cause of cancer-related mortality and morbidity in men and is characterized by a reduction in zinc and citrate concentration in both glandular tissue and prostatic fluid. Cancer cells display a luminal phenotype but the cellular origin of prostate cancer is debated with lineage tracing studies in mice indicating that they may arise from both basal and luminal cells.

2There is limited information on the nature and composition of the tissue-resident immune cell compartment in the healthy human prostate. However, several tumor-associated immune cell subsets have been reported, including T and B lymphocytes, regulatory T cells, monocytes, macrophages, dendritic cells (DCs), and natural killer (NK) cells, some of which correlate with a worse prognosis—for example, regulatory T cells and CD163-positive M2 macrophages.

3Prostate cancer is classified according to the ISUP Grade Group system and the Gleason Grading System, based on the extent to which tissue architecture and cellular morphology are disrupted, with higher scores associated with more aggressive tumor growth and worse outcomes. While primary localized disease has a generally good prognosis, men with locally advanced and metastatic disease have much worse 10-year survival rates. Androgen deprivation therapy is the mainstay of treatment for de novo metastatic prostate cancer, but a proportion of patients progress to castration-resistant disease, although the mechanisms underpinning this are unclear. Immune checkpoint blockade with antibodies against cytotoxic-T-lymphocyte-associated protein 4 (CTLA4) or programmed cell death 1 (PD-1)/PD-1 ligand 1 (PD-L1) has been used in these patients, but results have largely been disappointing, consistent with reports that less than a third of tumors show evidence of PD1 or PDL1 expression. Therefore, there is an urgent need to better understand tissue immunity in the healthy prostate and the nature of its perturbation in prostate cancer to inform future therapeutic strategies. Here, we applied single-cell RNA sequencing (scRNaseq) to paired human prostate biopsies collected at the time of cancer diagnosis, to comprehensively profile the cellular landscape of normal human prostate and to determine how this becomes disrupted in cancer. We validated our findings using flow cytometry, immunofluorescence microscopy, spatial transcriptomics; and by cross-species studies of mouse prostate specimens.

4We performed scRNaseq on paired cancer biopsy and adjacent normal prostate tissue in n = 10 subjects aged 50–72 years of age (Figure 1A, Table S1) generating data on 15,492 cells post-QC. We identified 14 cell clusters (Figure 1B), all of which contained cells from both normal and cancer biopsies (Figures 1C and S1A). This included immune cells, endothelial cells, fibroblasts, as well as several epithelial cell subtypes (Figure 1B), that were annotated based on canonical marker expression and comparison with a previously published scRNA seq dataset of young, healthy human prostate tissue (Figures 1D and S1B).

5Immune cell clusters included mononuclear phagocytes (MNPs), mast cells, NK cells, B cells, and T cells (Figure 1D) that were present in both normal and cancer samples with similar frequency (Figure 1E). We used flow cytometric analysis and confocal imaging to validate the presence of the major immune cell subsets in normal prostate tissue in human, and performed a cross-species comparison in mice, using a CD45 antibody administered intravenously premortem to label intravascular cells and confirm bona fide tissue residency (Figures 1F, 1G, 1H, 1I, S1C, S1D, and S1E). Together, these data show that the healthy prostate has a rich immune landscape dominated by T cells and MNPs, and that these cells persist in prostate cancer.

6Among nonimmune cell clusters, we identified basal, hillock, and club cell clusters, as noted in previous single cell analyses of human prostate, but in contrast to published data, we found two distinct clusters of cytokeratin-8+ luminal epithelial (LE) cells, rather than a single LE cell population (Figure 2A). The two LE cell populations were present in normal prostate and prostate cancer samples and were transcriptionally distinct with one cluster expressing high levels of KLK3 (encoding kallikrein related peptidase 3—also known as prostate-specific antigen [PSA]), KLK2, and KLK4 (annotated as LE-KLK3)—and the second cluster expressing high levels of KLK4 and little KLK3 (annotated as LE-KLK4) (Figure 2B). Cells with a high degree of transcriptional similarity to LE-KLK4 were confirmed to be present in the previously published healthy prostate dataset generated from three organs donors aged 18–31 years of age (; Figure S2A), but the limited number of LE-KLK4 cells in this dataset did not enable their identification as a distinct subset in the previously published analysis. Similarly, examination of previously published prostate cancer single-cell datasets also identified cells with high transcriptional similarity to LE-KLK3 and LE-KLK4 (Figures S2B–S2D). LE-KLK3 demonstrated enrichment for several immune pathways, including ‘TNFa via NFKB signaling’, ‘IL6-JAK-STAT3 signaling’ ‘interferon gamma response’, as well as ‘androgen and estrogen response’, with LE-KLK4 showing some enrichment for ‘Myc target’ genes suggestive of proliferative activity (Figure 2C). Furthermore, a subset of LE-KLK4 cells enriched for a prostate tumor-associated proliferation ‘Polaris’ signature (Figure S3A) and included cells with a G2/M cell-cycle profile (Figure S1A). In keeping with the high expression of ‘androgen response’ pathway genes in LE-KLK3, the gene encoding the androgen receptor (AR) was also highly expressed in this cluster, with little AR expression in LE-KLK4 (Figure 2D). Protein atlas data demonstrated the presence of KLK4+ cells between KLK5+ basal cells and KLK3+ cells adjacent to the lumen (Figure 2E). Colocalization of LE-KLK3 and LE-KLK4 cells was also confirmed in a previously published spatial transcriptomics dataset (; Figure S3B). The closer spatial proximity of LE-KLK3 to the lumen compared with LE-KLK4 is consistent with the transcriptional enrichment in immune defense genes observed in LE-KLK3, as these cells present an interface with the external environment and potential pathogen challenge.

7We next considered MNPs in the human prostate in isolation and integrated scRNaseq data from MNPs in a published normal prostate dataset (; Figure S4A). We found six distinct clusters of MNPs that exhibited transcriptional profiles consistent with their identity as monocytes (Mono), conventional DCs (cDCs), proliferating macrophages (MAC-cycling), and macrophages (MAC1, MAC2 and MAC-MT) (Figures 3A, S4B, and S4C). Two major subsets of cDCs are recognized, cDC1 and cDC2. cDC1 express XCR1 and cross-present antigens to CD8 T cells, while cDC2 express CD1c and activate CD4 T cells. In our dataset, we did not find two distinct DC clusters, but both CD1C positive and negative cells were evident within the DC cluster (Figure S4B). Two of the three macrophage populations were transcriptionally similar (MAC1 and MAC2) (Figure 3B), while MAC-MT cells formed a completely distinct cluster (Figure 3A) and expressed high levels of the metallothionein family genes (Figure 3B), encoding cysteine-rich proteins that bind divalent heavy metal ions and are involved in the cellular transport, storage, and metabolism of metal ions. High expression of metallothioneins has been noted in prostate cancer, but the MAC-MT cluster contained cells from both our dataset and that generated from normal, young human prostate tissue (; Figure S4C), confirming that these cells exist in healthy prostate tissue in homeostasis. We also validated the presence of MAC-MT in the three other published scRNaseq datasets of both normal prostate and prostate cancer (Figures 3C and S4D).

8Spatially, MAC-MT cell signatures colocalized with those of LE-KLK3 and LE-KLK4 (Figure S4E), and we confirmed the presence of an MT1+ CD68+ macrophage population in human prostate using both RNA scope and immunofluorescence microscopy, which were localized adjacent to luminal regions (Figure 3D). Zinc accumulation is controlled by two families of zinc transporters, the SLC39 (Zrt- and Irt-like proteins [ZIP]) that increase intracellular zinc, and the SLC30 (ZnT) proteins that lower zinc cellular levels. Interestingly, among prostate MNPs, the MAC-MT subset expressed the highest level of the zinc transporter genes, SLC39A8 (ZIP-8) and SLC30A1 (ZNT-1) (Figure 3E). MAC-MT cells were unique to the prostate, with no transcriptionally similar cells identified in other human organs (Figure S4F). Together, these data suggest that MAC-MT represent a prostate-specific macrophage subset, adapted to residency within the high zinc environment.

9Tissue macrophages arise prenatally from yolk sac (YS) or fetal liver progenitors, but are variably replaced postnatally by monocyte precursors that adopt tissue-specific transcriptional profiles. All three macrophage clusters in human prostate showed transcriptional similarity to YS-derived macrophages, as did the proliferating macrophage cluster, but MAC2 also enriched for the monocyte-derived macrophage signature (Figure 3F). Therefore, MAC1 and MAC-MT may represent prenatally seeded macrophage subsets, while MAC2 may be monocyte-derived, subsequently taking up a tissue-macrophage transcriptional signature. In keeping with this, mouse prostate contained both F4/80highCD11blo and F4/80loCD11bhigh macrophage subsets, (Figure S5A) identified as YS and hematopoetic stem cell (HSC)-derived subsets, respectively, in murine fate-mapping studies. An analysis of a murine prostate scRNaseq dataset also confirmed the presence of macrophage clusters with enrichment for both YS- and monocyte-derived macrophage signatures (Figure S5B). Anatomically, F4/80high macrophages were located among luminal epithelial cells in mouse prostate (Figure 3G). Expression of zinc transporters and metallothionein genes was also evident in mouse prostate macrophages, particularly F4/80hi cells enriching for the YS-derived macrophage signature (Figures S5C and S5D).

10We therefore hypothesized that these prostate macrophages may contribute to zinc homeostasis in the organ. To test this, we used an anti-Csf1r antibody which effectively depleted prostate macrophages, particularly the F4/80hi subset (Figure 3G, 3H, S5E, and S5F). This led to a significant reduction in prostate zinc concentration (Figure 3I), showing that prostate macrophages play a role in maintaining tissue zinc levels. The tissue zinc concentration in other organs was unaffected by macrophage depletion (Figures 3I and S5F), indicating this is a prostate specific macrophage function.

11The lymphoid compartment of normal prostate included CD4 and CD8 T cells, two subsets of NK cells (CD16+ and CD16neg) and B cells (Figures 4A and 4B). Further analysis indicated the presence of naive, tissue-resident memory, and regulatory CD4 T cells, as well as cytotoxic and tissue-resident memory CD8 T-cell clusters (Figures 4A, 4B, and S6A). We categorized the B cells according to CD27 (a marker of memory B cells) and IGHD (a marker of naive B cells) expression and observed that ∼30% and 10% of B cells were memory (CD27+IGHD-) and naive (CD27-IGHD+), respectively, with the remaining fraction mature non-naive B cells (Figures 4C and S6B). The majority of IGHM-expressing cells were CD27+ IgM memory cells, and there were also some class-switched B cells, with only a handful of BLIMP1-expressing plasma cells observed (Figure S6B). In the mouse prostate, extravascular naive and non-naive B cells were also evident adjacent to luminal epithelial cells (Figure 4D).

12The identification of two subsets of NK cells in normal prostate is consistent with previous descriptions of NK cells in blood and other organs; the majority of peripheral blood NK cells are CD56dimCD16+, with a small subset of CD56brightCD16neg NK cells, termed tissue-resident NK cells that are also present in spleen, uterus, and liver. Functionally, these CD16neg tissue-resident NK cells differ from conventional CD16+ NK cells, with reduced or altered cytotoxicity and prominent cytokine and chemokine production. In the prostate, CD16+ NK cells had a higher expression of genes associated with NK cell activation compared to CD16neg NK cells, with significant enrichment of NK cell cytotoxicity and activation gene sets (; Figure 4E). In addition, CD16neg NK cells showed enrichment for a universal lymphocyte tissue-residency signature (; Figure 4E), confirming that this subset represents tissue-resident NK cells and extending the list of tissues in which these cells have been identified in homeostasis. The presence of CD16+ and CD16- subsets of NK cell was confirmed in an independent normal prostate single-cell dataset (Figure S6C), and NK cells were also evident in normal mouse prostate samples (Figure 4F).

13We next sought to determine how prostate tissue immune cells were perturbed in cancer. Although there was no significant difference in immune cell number in prostate cancer samples (Figure 1E), several transcriptional differences in immune cells were evident; Antigen presentation and processing pathway genes were significantly reduced in prostate MNPs in cancer samples compared with normal prostate (Figure 5A), consistent with a widespread attenuation of CD4 T cell activation, a process critical for the generation of antitumor adaptive immune responses. In line with this, the expression of many co-activating receptors was higher in normal prostate MNPs compared with those in prostate cancer, the exception being MAC-MT, which showed higher expression of a subset of co-activating receptors in tumor samples, including CD40 (Figures 5B and S7A). There was increased expression of inhibitory PDL1 in MAC1 and DCs in tumor (Figure 5C and S7A), little PDL1 expression in non-immune –tumor cells, but some PDL2 expression detectable in tumor fibroblasts and endothelial cells (Figures 5C, S7A, and S7B). In naive/Tcm CD4 T cells, there was reduced expression of CD28, ICOS, and OX40 in tumor compared with normal, with little PD1 expression and reduced CTLA4 expression in tumor T cells (Figures 5C and S7A). Overall, in the lymphoid compartment, immune-related GO term genes were downregulated in tumor-associated lymphocyte subsets (Figure S7C), and there was significant enrichment for exhaustion signature genes in cytotoxic CD8 T cells (Figure 5D). Interestingly, in NK cells, we observed a significant enrichment of the ‘cytokine-cytokine receptor interaction’ gene set in the CD16neg NK cell subset in tumor compared to normal (Figure 5E) and the leading-edge genes included several chemokine transcripts related to DC recruitment, including CCL5, XCL1, and XCL2 (Figure 5F). Spatial transcriptomic analysis of prostate cancer confirmed colocalization of the CD16neg NK cell signature and CCL5 transcripts (Figure 5G). This suggests that the CD16neg resident-like NK cells in the prostate may promote the recruitment of tumor antigen cross-presenting cDC1, with potential beneficial anti-tumor effects, as described in melanoma, breast, and colon cancer.

14Analysis of predicted MAC-MT interactions with other immune cells based on receptor-ligand expression using CellPhoneDB demonstrated that MAC-MT within tumor samples had increased expression of the gene encoding BAFF (TNFSF13B) with the potential to support BAFF-R-expressing B cells (Figures 5H and S7D). BAFF protein expression was evident in prostate adenocarcinoma samples (Figure S7E) and BAFF staining colocalized with MHCII staining, marking MNPs (Figure 5I-J). In contrast, analysis of immune-fibroblast interactions highlighted CXCL12 expression by fibroblasts, with the potential to recruit CXCR3/4-expressing CD8 T cells and MNPs that was reduced in tumor (Figure 5K). LE-KLK3 and LE-KLK4 both expressed macrophage inhibitory factor (MIF), a macrophage survival, activation and recruiting factor, with its receptor CD74 expressed by all prostate MNPs, and this interaction was attenuated in cancer samples (Figure 5L). LE-KLK4 also expressed RPS19, with monocyte-recruiting activity, and this too was reduced in tumor samples (Figures 5L and S7F).

15In summary, these data indicate widespread immune transcriptional perturbation in prostate cancer, with reduced antigen presentation gene expression in MNP subsets, increased expression of exhaustion-associated genes in CD8 T cells, and reduced expression of immune-recruiting and activating chemokines and cytokines by fibroblast and epithelial cells in prostate cancer. However, in contrast, CD16neg NK cells in tumor had increased expression of cDC1-recruiting chemokines and MAC-MT higher expression of the B cell survival factor BAFF, both with potential beneficial antitumor effects.

16To further probe the effect of the tumor microenvironment on MNPs, we compared GO term enrichment in normal and tumor MNPs. This analysis showed attenuated expression of all pathways enriched in homeostasis in every MNP subset except for MAC-MT (Figure S8A). Indeed, metallothionein genes were increased in tumor-associated MAC-MT compared with those in normal prostate (Figures 6A and S8B). SLC30A1 expression was increased and SLC39A8 decreased in MAC-MT cells in cancer compared to non-tumor samples (Figure 6B), the overall effect of which may be to increase zinc efflux, potentially counteracting the known decreased zinc concentration associated with prostate cancer.

17Transcriptional alignment of healthy prostate MNPs with human macrophages activated with a variety of stimuli demonstrated that MAC2 enriched for inflammatory LPS- and IFNγ-stimulated M1-like macrophage signatures, as well as IL4 and glucocorticoid-stimulated macrophage signatures (Figure 6C). MAC1 showed little enrichment for any signatures except for the glucocorticoid-stimulated macrophage signature, while MAC-MT were completely distinct, enriching for TNF-, IL10-, and fatty acid–stimulated macrophage signatures (Figure 6C). In contrast to normal prostate, MNPs in tumor tissue demonstrated a global reduction in most macrophage activation signatures, consistent with a broad immune-suppressive effect of the tumor environment (Figure 6C). However, the remarkable exception to this tumor-associated suppression was the MAC-MT subset, which showed a significant increase in the expression in IFNγ-, TNF-, and fatty acid–stimulated macrophage gene signatures (Figure 6C). We observed an increase in cholesterol homeostasis pathway genes (Hallmarks, Figure S8C) and linoleic acid (a fatty acid) metabolism pathway genes (KEGG, Figure 6D) in prostate cancer–associated LE-KLK4 cells, raising the possibility that the fatty acid response gene enrichment in MAC-MT cells may arise due to fatty acid generation by local LE-KLK4 cells. Of note, increased fatty acid production from de novo lipogenesis has been described in prostate cancer and inhibition of lipogenesis reduced cancer growth in vitro.

18STRING analysis of leading-edge genes (Figure S8D) in these macrophage activation signatures, showed that fatty acid stimulation genes upregulated in MAC-MT in tumor were dominated by metallothionine genes, while inflammatory stimulation genes included several chemokines (CXCL9, CXCL10; Figures 6E and 6F). CD8 cytotoxic T cells were the principle immune cell subset expressing CXCR3, the receptor for these chemokines (Figure 6F), suggesting that MAC-MT activation may promote CD8 T cell recruitment to tumors (Figure S8E). Consistent with this, some/many CD8 T cells were located adjacent to MNPs (Figure 6G) and spatial transcriptomic analysis of prostate cancer confirmed colocalization of the MAC-MT signature with CXCL9/10 and with CD8 T cells (Figure 6H).

19Given the pro-inflammatory transcriptional profile of MAC-MT cells, we hypothesized that they may have an antitumor effect, promoting anti-cancer immune responses and counteracting cancer-associated perturbations in zinc concentration. In keeping with this, cellular deconvolution (Figure S9A) of The Cancer Genome Atlas (TCGA) data (Figure S9B) indicated that prostate cancer biopsies with higher MAC-MT enrichment had a lower Gleason score (Figure S9C) and improved disease-free survival (Figure 6I; Table S2). No other individual immune cell subset signature had prognostic significance in this dataset, although the limited number of samples with a high CD16neg NK cell signature precluded a robust analysis of the prognostic association of these cells (Figure S9C).

20Our single-cell analysis of the human prostate delivered several remarkable findings; our sample processing protocol enriched for immune cells enabling us to deliver the most comprehensive overview of the immune landscape of normal human prostate to date. Our prostate immune cell atlas delineates a range of innate and adaptive immune cells, with several CD4 and CD8 T cell subsets, two subsets of NK cells, and a prostate-specific macrophage subset, which we designated MAC-MT. The latter subset was transcriptionally similar to YS-derived murine macrophages, consistent with the conclusion that they may be prenatally seeded, although transcriptional profile does not definitively prove ontogeny. They also expressed high levels of metallothionein and zinc transporter genes, suggesting that they could contribute to zinc homeostasis in the prostate. To test this hypothesis, we depleted prostate macrophages in mice, and confirmed a reduction in tissue zinc levels. In prostatic epithelial cells, the high zinc concentration acts to inhibit mitochondrial aconitase, truncating the Krebs cycle to generate citrate which is secreted into prostatic fluid maintaining spermatozoa. The bioenergetic consequence of this is a reduction in ATP generation; therefore, aerobic glycolysis is increased in prostatic epithelium. MAC-MT also enriched for glycolysis genes and may therefore directly contribute to prostate zinc and citrate via a similar mechanism, although we did not explore this in the current study. Of note, this homeostatic role of MAC-MT in maintaining prostate zinc adds to other examples of prenatally seeded tissue macrophages that contribute to organ physiology and function; for example, in the heart, a subset of macrophages act to buffer calcium ions within the conducting system, and in the intestine, muscularis macrophages regulate the steady-state peristaltic activity of the colon.

21In prostate cancer, the concentration of both citrate and zinc are markedly decreased. We found that in prostate cancer, MAC-MT markedly increased the expression of the zinc efflux transporter SLC30A1, which may represent a mechanism to counteract the disrupted zinc transport present in tumor luminal epithelial cells. MAC-MT were also largely resistant to the immunosuppressive effect of the tumor environment. In fact, they become transcriptionally more inflammatory, expressing the B cell survival factor BAFF and lymphocyte-recruiting chemokines. Our analyses suggest that this response may be driven by the metabolic changes observed in KLK4 cells in tumor, namely the production of fatty acids. This increase in fatty acid metabolism has been previously described in prostate cancer. We found that MAC-MT in tumor showed a transcriptional profile similar to that observed in macrophages stimulated with fatty acids. Remarkably, enrichment of the MAC-MT signature in prostate cancer biopsies was associated with improved disease-free survival, supporting the conclusion that their pro-inflammatory effects could be beneficial in the context of prostate cancer. The effects of MAC-MT enrichment are in contrast to previous studies investigating MNPs in prostate cancer which have associated the presence of myeloid-derived suppressor cells and CD163-positive M2 macrophages with worse survival, and emphasize the value of scRNA seq in delineating distinct cell subsets with important transcriptional and functional differences. Our data have translational relevance as the antitumor effects of MAC-MT could potentially be harnessed as an immunotherapeutic strategy in advanced progressive prostate cancer.

22We also identified a tissue-resident CD16neg NK cell subset in normal prostate. These cells showed increased expression of CCL5, XCL1, and XCL2 in prostate cancer, with the potential to recruit cDC1. This is reminiscent of a recent study in mice showing that NK cells in implanted melanomas, breast, and colon cancers played a critical role in recruiting cDC1 via the production of CCL5 and XCL1, with important antitumor effects. cDC1 express XCR1, as well as CCR1 and CCR5, both of which bind CCL5 and have antitumor functions; they can attract and activate tumor-specific CD8 T cells, and internalize and transport tumor antigens to lymph nodes, where they may cross-prime CD8 T cells. The potential importance of NK cells in antitumor responses in prostate cancer has been suggested by previous studies associating lower numbers of peripheral blood NK cells, particularly CD16neg NK cells, or reduced activation capacity of circulating NK cells, with a higher risk of prostate cancer on biopsy. NK cells within prostate cancer tissue have not been well studied, but enumeration via immunohistochemical staining indicated an association with a lower risk of disease progression, and there is one previous flow cytometric assessment of prostate tumor NK cells, demonstrating an enrichment of the CD56bright subset, relative to blood. Our study confirms the presence of a tissue-resident CD56bright CD16neg NK cell subset in the prostate in health and in cancer, and provides a transcriptional assessment of these cells, revealing a potentially important antitumor function in cDC1 recruitment.

23To date, metastatic prostate cancer has shown variable responsiveness to checkpoint blockade. Evaluation of PD-L1 expression in prostate cancer previously identified its presence in less than 10% of primary cancer and up to a third of metastatic cancers. Our analysis enabled simultaneous assessment of all PD1/2 ligands and receptors across cell types. PDL1 was expression was upregulated in tumor MAC2 and DCs, and we identified de novo PDL2 expression in tumor samples, suggesting that this may be a more relevant therapeutic target in prostate cancer than PDL1. This is consistent with a recent analysis of bulk RNA seq data from prostate cancer biopsies that found an increase in PDL2 transcripts compared with normal tissue, and that higher PDL2 expression was associated with worse outcomes. Our analysis enabled the specific identification of tumor fibroblasts and endothelial cells as key expressors of this immunoinhibitory molecule.

24Aside from tissue-resident immune cells, we also identified a subset of luminal epithelial cells that lack expression of AR that may interact metabolically with the prostate-specific MAC-MT. Our dataset delineates several cell-specific markers that should be investigated for their utility as biomarkers for the early identification of this cell subset to assess their contributions to development of castration-resistant disease.

25In summary, we define the immune cell landscape in normal human prostate and describe its perturbation in cancer. Our study revealed the presence of a prostate-specific macrophage subset, marked by high expression of metallothionein genes that, in contrast to other MNP subsets, increases inflammatory gene expression in cancer with potential beneficial prognostic effects.

26While our study used a tissue processing strategy that enriched for immune cells on n = 10 paired normal and prostate cancer samples and successfully created a single-cell prostate immune atlas, there are several limitations; the patients recruited to our study had predominantly moderate prostate cancer and did not receive androgen deprivation therapy, in contrast to the other prostate cancer single-cell datasets which primarily consist of samples from more advanced disease. Furthermore, we discovered a metallothionein- expressing, zinc-regulating macrophage (MAC-MT) population, the MAC-MT gene signature was associated with improved outcomes, and experimental depletion in mice showed that prostate macrophages contribute to zinc homeostasis in normal prostate. However, further work is needed to investigate the role of these macrophages in the context of prostate cancer.

27Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Menna R. Clatworthy (mrc38@cam.ac.uk).

28This study did not generate new unique reagents.

29Fifteen men (50 – 74 years old) undergoing image guided prostate biopsies for suspicion of prostate cancer were enrolled in the DIAMOND study (NHS National Research Ethics Service reference 03/018) (CI:Gnanapragasam). All participants had previously undergone multi-parametric magnetic resonance imaging (mpMRI) of the prostate on a 3T magnet (Discovery MR750, GE Healthcare), using a 32-channel phased array coil. T2-weighted, contrast-enhanced, and diffusion-weighted imaging was acquired using the LIKERT scoring system according to Prostate Imaging–Reporting and Data System (PI-RADS) guidelines. Only men with a positive MRI were approached and recruited for this study, defined as a PI-RADS score 3 or greater.

30Each participant underwent transperineal biopsy under general anesthesia using the Biopsee fusion platform (Medcom, Darmstadt, Germany) according to the Ginsburg protocol, with a variable number of biopsies cores taken in order to obtain an appropriate tissue diagnosis for that individual. All targets were defined by radiologists pre-procedure using T2-weighted imaging as the primary source images, using Biopsee fusion software. Patient/sample characteristics are summarized in Table S1. Samples from 5 men were used for sample preparation optimization and remaining 10 were used for sequencing experiments.

31All murine research was conducted under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB). Mice were housed at Cambridge Biomedical Services under specific-pathogen-free conditions. Wild-type C57BL/6 male mice aged 8 – 12 weeks were obtained from Jackson Laboratories (Margate, UK),

32Each participant underwent prostate biopsy with a variable number of biopsies (20-30) cores taken in order to obtain an appropriate tissue diagnosis. Samples were taken from systematic and targeted biopsies as standard of care. Men were consented to have additional cores taken from the “Target” (area where cancer was suspected on the MRI) and from an “off-target” area to provide a normal prostate tissue comparator. The number of biopsies taken from both areas ranged from 1 to 6 additional cores. Biopsies were placed into phosphate buffered saline and placed on ice immediately.

33Prostate tissue was received in ice cold PBS, minced into approximately 5 mm3 pieces and digested for 20 min at 37°C with agitation in a digestion solution containing 32.5 μg/mL Liberase TM and 50 μg/mL DNase in RPMI. Following incubation samples were passed through a 100 μm cell strainer using a 1 mL syringe plunger and washed by centrifugation with PBS. Live cells were enriched using a Dead Cell Removal kit (Miltenyi Biotec) as per manufactures instructions. This was followed by a 44% Percoll density-gradient for 30 min at room temperature. Enriched live cells were washed and counted using a haemocytometer with trypan blue. Cells were then blocked with human FcR block (Miltenyi Biotech) prior to surface staining for flow cytometry. Cell counts per gram were calculated with the addition of 123count eBeads.

3410X Chromium Chip Single-cell library generation and preparation were performed on the single-cell suspension according to 10X Chromium 3′ solution (V2 kit) as per manufacturer’s instructions with an aim to capture 5000-10000 cells/channel. Sequencing was performed at the Cancer Research UK Cambridge Institute on the Illumina HiSeq4000 platform.

35Following sequencing BCL files were demutiplexed to Fastq files using CASAVA. Subsequently splitting to single cells and mapping and quantification of genes was carried out using Cellranger software package (10X genomics). This generated count tables of unique molecular identifiers (UMI) for each gene per droplet.

36The left and right anterior prostate lobes were harvested from mice and minced into approximately 15 mm3 pieces. Samples were digested for 20 min at room temperature in a digestion solution containing 0.1 M HEPES, 32.5 μg/mL Liberase TM and 50 μg/mL DNase in RPMI. Following incubation samples were passed through a 100 μm cell strainer using a 1 mL syringe plunger, washed by centrifugation with PBS and blocked with 50:50 mix of normal mouse and rat serum prior to staining. Cell counts per organ / gram of tissue were calculated with the addition of 123count eBeads (Invitrogen).

37After blocking cells were incubated with live/dead cell staining (Live/Dead Aqua 405, Invitrogen) for 15 minutes on ice. Cell surface staining occurred on ice for 30 minutes. All samples were acquired on an LSR 4/5 laser Fortessa (BD) and data analyzed using FlowJo v10. Human antibody: anti-CD14 FITC (61D3, Invitrogen), anti-CD45 APC-eFluor780 (HI30, eBioscience), anti-CD19 eFluor450 (HIB19, eBioscience), anti-CD3 BV785 (OKT3, BioLegend), anti-CD8 PE (HIT8a, BioLegend), anti-CD16 PE-Cyanine7 (CB16, Invitrogen). Murine antibody (myeloid panel): anti-Gr1 FITC (RB6-8C5, BioLegend), anti-CD11b PerCP-Cy5.5 (M1/70, eBioscience), anti-CD3e APC (145-2C11, eBioscience), anti-CD19 APC (1D3, BioLegend), anti-CD45 APC-eFluor780 (30-F11, Invitrogen), anti-I-A/I-E Pacific Blue (M5/114.15.2, Biolegend), anti-CD11c PE (N418, BioLegend), anti-F4/80 PE-Cyanine7 (BM8, Invitrogen). Murine antibody (lymphoid panel): anti-CD4 FITC (GK1.5, eBioscience), anti-CD19 APC (1D3, BioLegend), anti-CD45 APC-eFluor780 (30-F11, Invitrogen), anti-NK1.1- Pacific Blue (PK136, BioLegend), anti-CD8a BV785 (53-6.7, BioLegend), anti-NKp46 PE/Dazzle (29A1.4, BioLegend), anti-CD3 PE-Cyanine7 (17A2, BD PharMingen).

38The single-cell data (10X cellranger output) processed with EmptyDrops and then corrected for ambient RNA expression using SoupX (v1.2.1). Contamination fractions were estimated using the following genes: haemoglobin genes: HBA1, HBA2 and HBB; immunoglobulin genes: IGKC, IGLC1, IGLC2, IGLC3, IGLC4, IGLC5, IGLC6 and IGLC7; sperm genes: STMN1; prostate specific antigen gene: KLK3. SoupX was run with clustering information derived from a generic processing workflow in Seurat. After SoupX, doublet detection was performed using scrublet (v0.2.1) with adaptations outlined in – Briefly, after scrublet was performed, the data was iteratively sub-clustered using standard Seurat-inspired scanpy (v.1.4.5.post2) workflow and a median scrublet score for each sub-cluster was computed. Median absolute deviation (MAD) scores were computed from the cluster scrublet scores and a one tailed t test was performed with Benjamini-Hochberg (BH) correction applied and cells with significantly outlying cluster scrublet scores (BH pval < 0.1) were flagged as potential doublets. The data was then processed using scanpy with standard quality control steps; cells were filtered if number of genes > 2500 or < 200. Percentage mitochondrial content cut-off was set at < 30%. Genes were retained if they are expressed by at least 3 cells. Genes counts for each cell were normalized to contain a total count equal to the median of total counts in cells before normalization. This led to a working dataset of 17,108 cells. We filtered out 1,616 cells that we could annotate as sperm cells from seminal fluid contamination from a single normal sample before resulting in the final set of 15,492 cells. Highly variable genes were selected based on the following parameters: minimum and maximum mean expression are > = 0.0125 and ≤ 3 respectively; minimum dispersion of genes = 0.5. The number of principal components used for neighborhood graph construction and dimensional reduction was set at 50. Batch correction was performed using bbknn with patients as the batch term with all other parameters as per default settings (for all cells and for lymphoid analysis). Clustering was performed using Leiden algorithm with resolution set at 1.0 (for all cells) or 0.5 (for myeloid and lymphoid). Henry et al. dataset was identically processed for comparison except that cell type identities were used as published and sub-clustering of leukocytes was performed to extract MNPs after marker gene identification with Wilcoxon Rank Sum tests.

39In all cases where Uniform Manifold Approximation and Projection (UMAP; v3.10.0) was used for dimensional reduction and visualization, the minimum distance was set at 0.3 and all other parameters as per default settings in scanpy.

40For integration of MNPs, standard SCTransform workflow implemented in Seurat (v3.2.3) was used. Calculation of PCA, UMAP and neighborhood graphs post integration was performed in scanpy using the SCTransform normalized data. Neighborhood graph graphs were constructed with 10 neighbors. For analysis of lymphoid cells, cells from three patients were excluded due to low numbers of cells (< 10 cells; D7 and D14) or low-quality information from cells (D6; insufficient cell-cell heterogeneity). Further sub-clustering was also performed on NK cells, CD8 T cells and non-CD8 T cells (annotated as CD4 T cells) separately to obtain the final lymphoid clusters via specifying the restrict_to option with resolution set at 0.3 in scanpy.

41Differential gene testing was performed using the Wilcoxon test rank sum test implemented in scanpy’s rank_genes_groups module.

42We used a logistic regression approach to test for cell type similarity. This is done with L2-regularised logistic regression (ridge regression) multinomial models with the glmnet R package (i.e., alpha parameter = 0). Models were trained on normalized gene expression data with 10-fold cross-validation to obtain the appropriate lamda coefficient (lambda.1se; within 1 standard error from best model) for prediction. Gene expression values were standardized in both the training and test sets. The average of 50 iterations was used for the final score. To determine if predictions were significant, a median prediction probability score for each cluster was calculated and MAD-outliers were identified using a one-tailed t test. Cells were considered to be significantly similar if the BH p value was < 0.05 and the probability was > 50%. Re-embedding of each relevant dataset into UMAP or tSNE space were performed where possible with standard scanpy workflow and cluster identities were used as published. Label transfer for prostate cancer datasets was performed with default ingest protocol in implemented in scanpy. Pre-processing of these additional datasets were performed as close as possible to the dataset in this manuscript except for the exclusion of ambient RNA correction due to inavailability of raw data.

43Gene module scores of gene sets used were obtained using scanpy’s score_genes module or AUCell. Kruskal-Wallis test or Mann Whitney U tests were performed to test for significance of enrichment where appropriate using Prism software (v8). P value < 0.05 were considered as statistically significant. Typically, gene sets were retrieved and used as published in the original articles; in cases where murine gene sets were used, murine genes were converted to human orthologs using biomaRt.

44Pre-ranked gene set analysis (prGSEA) on hallmark genesets and macrophage stimulation genesets were performed with gseapy (https://github.com/zqfang/GSEApy/).

45Gene ontology and KEGG pathway analyses were performed using fgsea or over-representation analysis implemented in clusterProfiler R package. Genes were pre-ranked according to signed -log10Pvalues for all prGSEA procedures.

46String-DB (v11) analysis was performed using the web browser tool (https://string-db.org/).

47We compared our scRNA-seq analysis with spatial gene expression data generated by spatial transcriptomics protocol for prostate cancer tissues. The tissues were from radical prostatectomy for a patient with adenocarcinoma. We used the tissue section L1.2, which was pathologically annotated as at cancer stage GLEASON score 3+3. For accurately mapping spot expression data to the tissue H&E image, we used spot count matrices with adjusted spot coordinates. The coordinate adjustment was based on the alignment of the H&E image with the corresponding spot-fluorescent image, where each spot was detected by Cy3 fluorescence signal. The alignment accounted for manufacturing variation that caused the differences between expected coordinates and the actual coordinates of spots printed onto the spatial gene-expression slide.

48To classify cell-types in each spot, we used the anchor-based data integration method and calculated probabilistic transfer scores of discrete cell-type labels from cell types information in our reference scRNA-seq data to spot data. For each spot, the class probability of the spot belonging to each of the 12 cell types was calculated. For estimating the abundance of two cell types in each spot as in the pie chart, we calculate the probability of the spot to be of the cell types of interest and scaled each value to 1. The script for anchor-based label transferring and for plotting cell type proportion to tissue is available at https://github.com/BiomedicalMachineLearning/stLearn and described in stLearn package.

49For label transfer of the FFPE visium datasets available from 10X resource page, SCTransform was used to normalize both the reference (prostate single-cell dataset) and spatial data prior to integration as per instructions for Seurat v3.2.3. Expression values plotted are SCTransformed normalized values. For calculation of spatial correlation, k = 5 nearest neighborhoods were extracted from a kNN graph computed from the spatial location of each voxel. Peason’s correlation was then performed on each neighborhood using the gene expression value and cell type prediction value, followed by averaging across neighborhoods. Correlation values will not be returned if expression value was not detected in all neighborhoods or expression value was uniform across all voxels. For the CXCL9 + CXCL10 comparison, the value from Seurat’s AddModuleScore of the two genes was used for computing the correlation.

50Normalized expression values from cell types found in this dataset were subjected to CellPhoneDB analysis (v2.0.0). The minimum threshold was set at 30% and results were considered statistically significant if p < 0.05.

51The Cancer Genome Atlas (TCGA) expression and clinical data for Prostate Adenocarcinoma (PRAD) were downloaded with TCGAbiolinks (v2.15.3). Single-cell deconvolution was performed using Multi-subject Single Cell deconvolution R package (MuSiC, v0.1.1) with raw counts as instructed in the package. Disease free survival indices were extracted from days_to_new_tumor_event contained in the TCGA clinical data; samples without days_to_new_tumor_event entries were removed. Events were considered if days_to_new_tumor_event_dx was annotated as ‘YES’ and all other samples were censored. Outcome for events were generally from biochemical evidence of disease but also included distant metastasis, locoregional recurrence, primary tumor and ‘not available’. All events were considered regardless of treatment received (radiological or pharmaceutical). Kaplan-Meier survival analyses were performed using the survival (v2.41-3) and survminer (v0.4.6) R packages where the deconvolved scores were categorised into a ‘high’ or ‘low’ group, which corresponds to top and bottom 25% MuSiC deconvolved scores. The results/info are tabulated and summarized in Table S2.

52Simultaneous detection of human CD68 and MT1 family genes were performed on FFPE sections using Advanced Cell Diagnostics (ACD) RNAscope® 2.5 LS Multiplex Reagent Kit (Cat No. 322800), RNAscope® LS 4-Plex Ancillary Kit Multiplex Reagent Kit (Cat No. 322830), RNAscope® 2.5 LS Probes (ACD, Hayward, CA, USA) at the histopathology/in situ hybridization core facility at Cancer Research UK – Cambridge Institute. Because it was not possible to design a specific RNAscope probe for Hs-MT1H due to high homology to other genes in the MT1 family, we used a probe that recognizes multiple MT1 genes. Briefly, sections were cut at 3 μm thick, baked for 1 h at 60°C before loading onto a Bond RX instrument (Leica Biosystems). Slides were deparaffinised and rehydrated on board prior to pre-treatments using Epitope Retrieval Solution 2 (Cat No. AR9640, Leica Biosystems) at 95°C for 15 min, and ACD Enzyme from the Multiplex Reagent kit at 40°C for 15 min. Probes were visualized using Opal fluorophores (Opal 570 and Opal 650 Akoya Biosciences Cat No. FP1488001KT and FP1496001KT respectively) diluted to 1:1000 using RNAscope LS Multiplex TSA Buffer. Probe hybridization, signal amplification and detection was performed on the Bond Rx according to the ACD protocol. Slides were then removed from the Bond Rx and mounted using Prolong Diamond (ThermoFisher Cat No P36965). The slides were imaged on the AxioScan (Zeiss) to create whole slide images. Images were captured at 40x magnification, with a resolution of 0.25 μm per pixel. ISH validation was carried out on one slide from the Cambridge 109 Prostate TMA (slide-1), consisting of paired sets of benign and tumor tissue cores that were collected under PrompT ethics (MREC/01/4/061).

535-6 C57BL/6 male mice aged 16 weeks received 0.5 mg of depleting anti-CSF1R mAb (BioXCell, clone AFS98) or isotype control (BioXCell, Rat IgG2a) intraperitoneally on day −7, −4 and −2 prior to euthanasia per experiment (n = 2 independent replicates). Following terminal procedure, kidneys, liver and prostate from each mouse were harvested and divided for either flow cytometry, microscopy or zinc quantification. Tissues for flow cytometry was minced finely and digested in RPMI containing 0.1 mg/mL DNase I, 32.5 mg/mL Liberase TM and 10 mM HEPES for 25 min at room temperature. Organs were then mechanically dissociated through a 70 μm cell strainer, washed in PBS and red blood cell lysis performed. Single cell suspensions were blocked for 30 min with 50 μL normal mouse serum in PBS 2% FBS on ice then stained with live/dead fixable aqua (Invitrogen), anti-CD45 APC-eFluor780 (30-F11, eBioscience), anti-I-A/I-E Pacific Blue (M5/114.15.2, Biolegend), anti-CD11b PerCP-Cy5.5 (M1/70, Invitrogen), anti-F4/80 PE/Cyanine7 (BM8, Invitrogen), anti-CD3e APC (145-2C11, Invitrogen), anti-CD19 APC (1D3, eBioscience) and anti-Ly-6G/Ly-6C APC (Rb6-8C5, Biolegend).

54A single anterior prostate lobe, one kidney and one lobe of the liver per mouse was weighed and homogenized in 300 μL of T-PER lysis buffer using the Precellys homogenizer system. Samples were then centrifuged at 1500 x g for 10 min to remove contaminating material and supernatants used for zinc analysis. Zinc levels were analyzed using the Zinc Quantification Kit (ab102507, abcam) as per the manufacturer’s instructions. Zinc concentration expressed as nmol per gram of tissue.

55Samples were fixed in AntigenFix for 30 min at 4°C, rinsed in PBS for 5 min then transferred into 30% sucrose in PBS for 24 h and embedded in optimal cutting temperature compound for cutting. 30 μm sections were permeabilised and blocked in blocking buffer containing 0.1 M TRIS, 0.1% Triton, 1% normal mouse serum, 1% normal rat serum, 1% BSA for 1 h at room temperature. Staining was performed in blocking buffer for 2 h at room temperature prior to washing in PBS and mounting in Fluoromount-G or Fluoromount-G with DAPI. When required, a secondary staining was performed in blocking buffer for 2 h at room temperature prior to washing and mounting. Images were acquired using a TCS SP8 confocal microscope and raw images were processed using Imaris. Sections for the human sample were obtained from a 15 mm x 1 mm core needle biopsy as per sample collection described above. Sections for mouse sample were obtained from cross section of the prostate across the lateral/ventral and dorsal prostate region. Antibodies used include – mouse: MHC II-Pacific Blue (clone M5/114.15.2, 1/50 dilution, BioLegend), CD11b-PE (clone M1/70, 1/50 dilution, Invitrogen), CD31-AF594 (clone MEC13.3, 1/100 dilution, BioLegend), F4/80-AF647 (clone F4/80, 1/50 dilution, Abcam), CD19-AF594 (clone 6D5, 1/100 dilution, BioLegend), IgD-AF488 (clone 11-26c.2a, 1/50 dilution, Biolegend), NKp46-PE (clone 29A1.4, 1/50 dilution, eBioscience), CD8-FITC (clone 53-6.7, 1/50 dilution, eBioscience), CD3-Pacific Blue (clone 17A2, 1/50 dilution, BioLegend); human: HLA-DR-AF647 (clone TAL 1B5, 1/50 dilution, Abcam), CD206-PE Dazzle (clone 15-2, 1/50 dilution, BioLegend), MT1 (clone UC1MT, 1/50 dilution, Abcam), CD8-PE (clone HIT8a, dilution 1/100, BioLegend), CD3-AF488 (clone UCHT1, dilution 1/100, BioLegend), BAFF (polyclonal rabbit, dilution 1/50, Bioss), Mouse IgG1 kappa monoclonal isotype control (clone 15-6E10A7, 1/50 dilution, abcam), Goat anti-mouse IgG secondary-FITC (polyclonal, 1/200 dilution, Invitrogen). Dyes: Flash Phalloidin 488 (1/300 dilution, BioLegend), Hoechst 33258 (dilution 1/10 000, cat# 40044, Biotum), DAPI (in mounting medium, cat# 00-4959-52, Invitrogen).

56Results were generated using R packages or python modules and organized as figures using Adobe Illustrator. Combined scatter-,box-, violin-plots were generated in R using code based on and violin plots and dot plots were generated using plotting modules implemented in scanpy or with R ggplot2-based functions. Heatmaps were generated using pheatmap (v1.0.12) R package or matplotlib (v3.0.3) modules in python. Bar plots were generated using Prism (v8).

57Statistical analysis was performed using GraphPad Prism software, R, or python and have been described in the relevant methods sections and figure legends accordingly. In general, unless otherwise specified, non-parametric tests were used and p value after false discovery correction procedures < 0.05 were considered statistically significant. Sample sizes for mice experiments can be found in figure legends.

58An interactive version and h5ad files of the single-cell RNaseq data is available at prostatecellatlas.org.