Perivascular macrophages collaborate to facilitate chemotherapy resistance in cancer

A subset of tumor associated macrophages (TAMs) identified by their expression of the lymphatic vessel endothelial hyaluronan receptor-1 (Lyve-1) reside proximal to blood vasculature and contribute to disease progression. Using a spontaneous murine model of mammary adenocarcinoma (MMTV-PyMT), we show that Lyve-1+ TAMs, which co-express heme oxygenase-1, form coordinated multi-cellular ‘nest’ structures in the perivascular niche. We show that TAM nest formation is dependent on IL-6 and a communication axis involving CCR5 and its cognate ligands CCL3/4. We demonstrate that Lyve-1+ TAM nests are associated with CD8+ T-cell exclusion from the tumor and the resistance to immune-stimulating chemotherapeutics. This study highlights an unappreciated collaboration between TAMs and uncovers a spatially driven therapeutic resistance mechanism of these cells in cancer which can be therapeutically targeted.


Introduction
Macrophages are a phenotypically and functionally diverse population of innate immune cells which become exploited by tumors to facilitate disease progression and therapy resistance (1-8). Heterogeneity within the tumor associated macrophage (TAM) population arises from their site of origin (9) and the influence of environmental cues within the tumor microenvironment (TME) (3,10,11), which can be guided by both spatial (12,13) and temporal (11) parameters. One subset of TAMs reside in close proximity to blood vasculature (<15-20 µm) (14)(15)(16) and are termed perivascular TAMs (PvTAMs). PvTAMs support a variety of pro-tumoral functions including neo-angiogenesis (15,16), metastasis (3,17,18) and facilitate tumor recurrence post chemotherapy (19). Recently, we demonstrated that a subset of PvTAMs expressing lymphatic vessel endothelial hyaluronan receptor 1 (Lyve-1) form a pro-angiogenic Pv niche with a population of pericyte-like cancer associated fibroblasts (CAFs) and orchestrate the platelet-derived growth factor-C (PDGF-C)-dependent expansion of the CAF population with the growing tumor (16). Lyve-1 was traditionally considered a marker of lymphatic endothelium (20), but has emerged as selective marker for a sub-population of Pv tissue-resident macrophages (21)(22)(23)(24)(25) and TAMs (16,26). Depleting Lyve-1 + PvTAMs in the spontaneous mouse mammary tumor virus polyomavirus middle T antigen (MMTV-PyMT) murine model of breast cancer using liposome-based approaches resulted in tumor control, highlighting the importance of the TAM subset in cancer progression and formation of the Pv niche (16).
PvTAMs have been demonstrated to develop from a CCR2 + monocyte origin (11,27). These infiltrating monocytes express CXCR4 within the TME in response to tumor-derived tumor growth factor-beta (TGF-β) and subsequently traffic to the Pv space via CXCL12 expressed by a population of PvCAFs (11). The CXCR4/CXCL12 axis has also been demonstrated to be important for the accumulation of PvTAMs post chemotherapy treatment (19). However, once TAMs reach the endothelium, little is known about the subsequent signaling axes important for niche formation. In the current study we investigate the development of Lyve-1 + TAMs and their Pv niche in MMTV-PyMT tumors and define a new collaborative function of these cells in forming multi-cellular 'nests' which contribute to the resistance to therapeutics through a role in immune exclusion of the TME. This study sheds new light on novel therapeutic strategies to target the immuno-modulatory function of Lyve-1 + TAMs.

Lyve-1 + TAMs express HO-1 and accumulate in Pv 'nests' within the TME
Using the autochthonous MMTV-PyMT murine model of breast cancer (28), we recently demonstrated that Lyve-1 marks a sub-population of pro-angiogenic TAMs in the TME which reside proximal to blood vasculature (16) (Fig. 1A). To investigate whether an analogous TAM population exists in human cancer, we extracted the Lyve-1 + TAM cell cluster as previously described (16) from a scRNA-seq dataset of 9,039 TAMs sorted from 3 individual MMTV-PyMT tumors ( Fig.1B-C). We mapped the murine Lyve-1 + population onto a recently published scRNA-seq atlas for human breast cancer (29). The murine Lyve-1 + TAM population identified with cells within the human myeloid cell cluster ( Fig. 1D and table S1).
Focusing on the myeloid cells within the atlas, 1,444 of the 9,675 myeloid cells were judged to be Lyve-1 + TAM-like in their phenotype ( Fig. 1E-F) and their expression of Lyve1, MRC1 (CD206) and HMOX1 (HO-1) were significantly associated with the identified cells (Fig. 1F), highlighting a conservation of this TAM phenotype between species. HO-1 is a marker that has previously been associated with PvTAMs in murine models of cancer (3,19) and the HO-1 expressing TAM population was also observed in the Pv niche of human cancer (Fig.   1G), highlighting a similar spatial location for these cells in human cancer. HO-1 breaks down heme into the biologically active catabolites biliverdin, ferrous iron (Fe 2+ ) and carbon monoxide (CO) (30, 31) which has pro-tumoral properties including cytoprotection and immune suppression (30, 32-37). HO-1 and Lyve-1 protein co-localized in tumor sections from MMTV-PyMT mice (Fig. 1H) and Lyve-1 + TAMs were the major source of HO-1 within the Pv niche which also includes, endothelial cells (ECs) and pericyte-like αSMA expressing CAFs ( Fig. I-J) (16). Interestingly, from these analyses it was evident that the Lyve-1 + HO-1 + PvTAM subset did not uniformly distribute along the endothelium, but instead was restricted to discrete regions of the vasculature where evidence of clustering could be found (Fig. 1A, G, H, J). These clusters, which we herein refer to as 'nests', highlighted an unappreciated multi-cellular PvTAM structure within the TME. As Lyve-1 is also expressed at high levels by the lymphatic endothelium (20), and HO-1 expression is selectively expressed by this TAM subset ( Fig. 1H-J), we generated a knock-in reporter mouse for the Hmox1 gene to facilitate the study of these cells (Fig. 1K). The reporter consisted of Click Beetle luciferase (Luc) and enhanced green fluorescent protein (eGFP) (38) inserted before the stop codon of the genomic Hmox1 gene locus. HO-1, Luc and eGFP were separated by a self-cleaving P2A sequence (39) to allow equimolar expression of the three proteins ( Fig. 1K-L and fig. S1A) (mouse herein referred to as HO-1 Luc/eGFP ). As HO-1 plays several important functional roles in the TME (3,31,32,40), the arrangement of the reporter construct ensured that the native HO-1 expression was unaffected by the reporter elements. The HO-1 Luc/eGFP reporter mouse was crossed onto the MMTV-PyMT background and tumors were analyzed for their distribution of HO-1/eGFP expression (fig. S1B). TAMs (fig. S1B), and specifically the Lyve-1 + CD206 hi MHCII lo TAM subset ( Fig.1M and fig. S2), were the major tumoral source of HO-1/eGFP. Further characterization of F4/80 + HO-1/eGFP + cells in tumors by immunofluorescence analysis also confirmed the localization of HO-1/eGFP to the Pv space within nest structures (Fig. 1N). These data validate the HO-1 Luc/eGFP reporter mouse as a tool for studying Lyve-1 + PvTAMs and identifies a previously unappreciated multi-cellular nest structure for these cells in the Pv space.

Lyve-1 + macrophages expressing HO-1 can be found populating healthy organs in steady state-conditions
Lyve-1 + macrophages reside in a variety of healthy tissues (16,(21)(22)(23)(24)(25) and in vivo bioluminescence imaging of HO-1/Luc in HO-1 Luc/eGFP reporter mice demonstrated widespread expression of HO-1 in healthy tissues ( Fig. 2A-B). Together this prompted the question whether tissue resident Lyve-1 + macrophages share a phenotypic resemblance to the TAM population and co-expressed HO-1. Analysis of healthy tissues from the HO-1 Luc/eGFP reporter mouse revealed the spleen, lung, mammary gland, visceral adipose, skin and liver to be some of the highest expressors of HO-1 (Fig. 2B). Flow cytometry analysis of enzyme-digested tissues demonstrated tissue resident F4/80 hi macrophages to be the main source of HO-1/eGFP in the healthy tissues analyzed (Fig. 2C). Furthermore, HO-1/eGFP + macrophages expressed Lyve-1 in all organs analyzed apart from the spleen (Fig. 2C-D).
However, the spleen represents a unique tissue as it is the major site of erythrophagocytosis (41) and heme is a major inducer of HO-1 (31). Lyve-1 + macrophages in the lung, mammary gland, visceral adipose, skin and liver accounted for almost all HO-1 + events in these tissues, highlighting a close concordance between these markers (Fig. 2E). However, despite this similarity, there were distinct differences between the Lyve-1 + macrophages between tissues and the tumor in their expression of the markers CD206 and MHCII (Fig.   2D). These data suggest there is a microenvironmental influence on the Lyve-1 + macrophage phenotype. Lyve-1 + macrophages in healthy tissues arise from a recruited monocyte progenitor (25). To confirm the source of Lyve-1 + TAMs we utilized the photoconvertible Kaede mouse (42) crossed to the MMTV-PyMT model. When tumors reached 100mm 3 the TME was photoconverted from Kaede-green to Kaede-red using a UVlight source (Fig. 2F). After 48 h post photoconversion of the TME there was a clear Kaedegreen population of Lyve-1 + TAMs in the TME, indicative of recruitment from a peripheral source, highlighting a monocytic origin for these cells (Fig. 2G). These data highlight a close concordance between Lyve-1 and HO-1 expression associated with this macrophage subset in both healthy tissues and the tumor. Furthermore, we demonstrate that the TME exploits the development of Lyve-1 + TAMs through a microenvironmental influence.

Lyve-1 + TAMs are polarized by IL-6 in the TME
To investigate the development of Lyve-1 + TAMs within TME, we analyzed scRNA-seq data of TAMs from MMTV-PyMT tumors (16). Using the QIAGEN Ingenuity Pathway Analysis software (QIAGEN IPA) (43), IL-6 signaling was predicted to be an upstream polarization signal associated with the transcriptional programs active in the Lyve-1 + TAM subset (Fig.   3A). To investigate the role of IL-6 in polarization of the Lyve-1 + PvTAM subset we crossed MMTV-PyMT mice onto an Il6 -/background and analyzed the tumors in these animals. In the absence of IL-6, tumors were significantly slower to establish, with a median latency of 87 versus 100 days for Il6 +/+ and Il6 -/-MMTV-PyMT mice respectively (Fig. 3B). When tumors reached ~500 mm 3 they were enzyme-dispersed and analyzed by flow cytometry. Despite no overall change in the abundance of TAMs within the TME (Fig. 3C), there was a significant and selective loss of Lyve-1 + TAMs (Fig. 3D-F), highlighting that IL-6 was fundamental to the polarization of the TAM subset. Analysis of bulk RNA-seq data from sorted TME stromal populations within the Pv niche (16), highlighted the αSMA pericyte-like CAF and endothelial cells (but not the TAMs) to express Il6 mRNA (Fig. 3G). More widely, CAFs and endothelial cells were the main tumoral sources of Il6 mRNA expression with no detectable expression in the tumor cells or other stroma subsets ( fig. S3). To identify the spatial location of the Il6 mRNA in the TME we used RNAscope (Fig. 3H). Il6 mRNA was detectable in nearly all endothelial cells but high expression of Il6 mRNA was evident in those endothelial cells that were proximal to HO-1 + cells (Lyve-1 + TAMs) ( Fig. 3H and fig.   S4), although αSMA + CAFs could be found expressing Il6 mRNA in the TME, as indicated by the bulk RNA-seq data, the cells expressing Il6 mRNA were not proximal to the HO-1 + cells ( fig. S4). These data suggest that Lyve-1 + TAMs develop in response to IL-6 that is secreted by the endothelium in the TME.

IL-6 polarizes Lyve-1 + macrophages via a STAT3/c-MAF-dependent signal
To investigate whether IL-6 stimulation alone was sufficient to generate a Lyve-1 + TAM-like phenotype. Bone marrow-derived macrophages (BMDMs) were exposed to IL-6 and their phenotype was assessed. IL-6 stimulated BMDMs upregulated the expression of the three key phenotypic markers; HO-1 ( Fig. 3I and fig. S5), CD206 and Lyve-1 (Fig. 3J) which define the subset. These data suggested that IL-6 plays a dominant role in the polarization identity of Lyve-1 + TAMs and their relative absence in Il6 -/-MMTV-PyMT mice (Fig. 3F) was most likely directly due to the loss of IL-6 from the TME. Recent data demonstrated that the transcription factor c-MAF was vital to the development of Lyve-1 + Pv macrophages in healthy tissues (44), and we sought to establish whether c-MAF signaling may account for the Lyve-1 + macrophage phenotype observed through IL-6 stimulation. IL-6 signaling is associated with the JAK/STAT3 pathway (45), however, STAT3 has been linked to the expression of c-MAF via the basic leucine zipper transcription factor ATF-like, Batf in T follicular helper cells (46). To investigate whether c-MAF and STAT3 signaling were required for the Lyve-1 + macrophage phenotype, we knocked down the expression of either c-MAF or STAT3 in BMDMs prior to stimulation by IL-6 ( Fig. 3I and fig. S5). Loss of either transcription factor was sufficient to prevent the upregulation of Lyve-1, CD206 and HO-1 upon IL-6 stimulation ( Fig. 3I and J and fig. S5). Interestingly, c-MAF appeared to be required for even basal HO-1 expression in BMDMs ( Fig. 3I and Fig. S5B). These data demonstrate that IL-6 is sufficient to generate the key phenotypic markers associated with the Lyve-1 + PvTAM subset through a STAT3 and c-MAF-dependent pathway.

Lyve-1 + TAMs communicate via a CCR5-dependent axis to orchestrate Pv nest formation
Having identified IL-6 as a driver for the Lyve-1 + TAM polarization program (Fig. 3), we investigated whether their polarization program might influence a communication axis between these cells to orchestrate the formation of nests within the Pv niche (Fig.1A). A microarray analysis was performed on monocyte-derived macrophages exposed to IL-6, IL-4 IL-6 stimulated BMDMs (Fig. 4D). Although all TAMs expressed some CCR5, Lyve-1 + TAMs were the highest expressor in the TME ( fig. S6E). CCR5 was functional on the macrophages and IL-6 stimulated BMDMs could migrate towards the CCR5 ligand CCL5 (Fig. 4E). When IL-6 stimulated BMDMs were also placed in close contact in vitro their ability to spread from one another was also reduced (Fig. 4F). To test the role of CCR5 in maintaining Lyve-1 + TAMs in nests within the TME we crossed MMTV-PyMT mice onto a Ccr5 -/background. In the absence of CCR5, tumors were significantly slower to establish, with a median latency of 87 versus 102 days for Ccr5 +/+ and Ccr5 -/-MMTV-PyMT mice respectively (Fig. 4G). As CCR5 expression was downstream of IL-6 signaling in the Lyve-1 + TAM subset, as predicted, the absence of CCR5 did not affect the total prevalence of TAMs (Fig. 4H) or the Lyve-1 + TAM subset ( Fig. 4I-J) in these tumors. However, immunofluorescence imaging of tissue sections from these tumors revealed an increase in the median distance of the Lyve-1 + TAMs to the endothelium and each other (Fig. 4K), without effecting their abundance overall (Fig. 4L). This demonstrated that CCR5 was required to maintain the nest structures of Lyve-1 + TAMs. To demonstrate that CCR5 also played an ongoing active role in maintaining the Lyve-1 + TAM nests post formation, rather than just the initial formation of the structure, we injected MMTV-PyMT mice with maraviroc (Fig. 4M), a drug that is clinically used to inhibit CCR5 (48). Therapeutically blocking CCR5 signaling using maraviroc in tumors with formed Lyve-1 + PvTAM nests did not affect tumor growth ( fig. S7) but did result in an observable dispersion of the Lyve-1 + TAM nests away from the Pv space (Fig. 4N), highlighting that CCR5 represents an ongoing communication axis for the maintenance of the Pv nest structures. These data highlight a new role for CCR5 in the collaborative formation and maintenance of the Lyve-1 + TAM nests in the Pv niche.

Lyve-1 + TAM nests support immune exclusion in the TME
To understand how the absence of Lyve-1 + TAM nests may alter the wider TME we analyzed the composition of the stroma in ~500 mm 3 tumors from WT, Il6 -/and Ccr5 -/-MMTV-PyMT mice (Fig. 5A). The stromal composition in enzyme-dispersed tumors were assessed using flow cytometry ( Fig. 5B and fig. S8). Broadly, the stromal composition of the tumors was highly similar despite the spontaneous nature of the tumor model (Fig. 5B).
However, the only consistent difference for both the Il6 -/and Ccr5 -/-MMTV-PyMT mice compared to WT tumors was a significant increase in tumor infiltrating CD8 + T-cells (Fig.   5B). The CD8 + T-cells had a similar overall proportion of cells which displayed effector function ( Fig. 5C and fig. S9). This suggested that the Lyve-1 + TAM nests could be associated with immune exclusion in the TME. The endothelium expresses adhesion molecules which permit leukocyte rolling, migration and arrest prior to diapedesis into inflamed tissues (49), and it was possible that Lyve-1 + TAMs could modulate the endothelium due to their close proximity to one another. However, there was no evidence for a change in the expression of endothelial VCAM-1, ICAM-1 or pNAD in the absence of Lyve-1 + TAMs or their nests in the TME (fig. S10A). To explore if Lyve-1 + TAMs could play a role in immune exclusion, we established an in vitro assay for creating artificial Pv nests (Fig.   5D). In this assay, M(IL-6) which are analogous to the Lyve-1 + TAMs ( Fig.3I-J), were seeded onto the basolateral side of a transwell insert, and then a basement membrane and an endothelial layer were seeded on the apical side of the insert (Fig. 5D). The presence of M(IL6) had no significant effect on the permeability of the endothelial layer (Fig. 5E). However, when CD8 + T-cells were placed in contact with the endothelial layer with a gradient of the Tcell chemokine CXCL10, there was a significant reduction of T-cell migration across the endothelial layer in the presence of M(IL6) (Fig. 5F). HO-1 has been demonstrated to play a role in vascular biology (31) and, as such, we considered whether the enzyme might play a direct role in the mechanism of CD8 + T-cell exclusion. Pharmacologically inhibiting HO-1 activity using the inhibitor tin mesoporphyrin (SnMP) (50) (Fig. 5G), or a genetic knock out in M(IL6) using BM from a Hmox1 fl/fl mouse crossed with Lyz2 promoter driven Cre recombinase  S11). In the absence of CD8 + T-cells, 5-FU had no effect on tumor growth in Il6 -/tumors demonstrating an immunological basis for the mechanism of tumor control observed. As such, these data highlight a previously unappreciated link between Lyve-1 + PvTAM nests and the immune landscape of the TME which can provide a resistance mechanism for the immune-mediated effects of cytotoxic chemotherapeutics.

Discussion
In this study we describe a coordinated and collaborative role for Lyve-1 + PvTAMs in forming multi-cellular nest structures within the TME which facilitate immune exclusion and the resistance to the immune-stimulating capabilities of cytotoxic chemotherapy. Although a number of pro-tumoral processes have been described to PvTAMs (15), this study highlights an unappreciated mechanism of these cells that is reliant on their collaborative interaction and collusion which promotes cancer progression. This study sheds new light on the development of the Lyve-1 + TAM subset. Using photolabeling approaches we demonstrate that Lyve-1 + TAMs develop from recruitment of a progenitor into the TME, consistent with previous studies that have demonstrated PvTAMs to have derived from a monocyte origin (11,25,27). It is apparent that PvTAMs develop through a sequential signaling program involving their upregulation of CXCR4 in response to tumor cell-derived TGF-β which guides their migration back to the endothelium on a gradient of CXCL12 expressed by a PvCAF population (11). The importance of the CXCR4/CXCL12 axis in PvTAM accumulation at the vasculature has been demonstrated in several studies (19,59). We propose that the mechanisms presented in this study provide insight on the subsequent developmental step of PvTAMs into the Lyve-1 + TAM population post reaching the endothelium. This is supported by the observation that loss of IL-6 resulted in a striking and specific loss of the Lyve-1 + TAM population leaving other TAM polarization states unchanged, suggesting it is important for only the terminal step in polarization. Interestingly, the interaction of angiopoietin-2 expressed on endothelial cells and Tie2 expressed on PvTAMs has also been demonstrated to play an important role in their accumulation at the vasculature (60, 61), however we did not find evidence of Tie2/Tek gene expression in the RNA-seq datasets for the Lyve-1 + TAM in this study.
We demonstrate that a key feature of the Lyve-1 + TAM program was high expression of the enzyme HO-1, a gene which we have previously linked to IL-6 signaling (3). However, surprisingly, in homeostatic tissues, tissue resident Lyve-1 + macrophage populations also expressed HO-1. We identified in vitro that c-MAF signaling was required for Lyve-1, HO-1 and CD206 expression on BMDMs in response to IL-6 and could link these markers on tissue resident macrophages in non-inflammed healthy tissues which have recently been demonstrated also to be c-MAF-dependent (44). Although the signal for c-MAF in healthy tissues is unknown, it could explain the unexpected high expression of HO-1 which is generally considered as a stress-or inflammation-inducible enzyme (30, 62). In this study, using HO-1 Luc/eGFP reporter mice, we find that the Lyve-1 + tissue resident macrophages account for almost all HO-1 expression in healthy tissues which highlights a role for macrophages in the homeostatic functions ascribed to HO-1 in healthy tissues (31).
The identification of the immunological resistance mechanisms to cytotoxic chemotherapy responses is important as it has become apparent that the immune-stimulating properties of these drugs may underlie a significant proportion of their anti-tumoral efficacy (63-68). This study also complements the wider association between TAMs representing a pivotal role in resistance to chemotherapy (6)(7)(8)55) and facilitating relapse after the cessation of treatment (19). This study also helps mechanistically supporting the link between IL-6 suppression and the immune-stimulating effects of cytotoxic chemotherapies (69). There are many studies which have described HO-1 as having pro-tumoral properties having important roles in cytoprotection and immune suppression (30, 32-37). We previously demonstrated that HO-1 plays a major role in the suppression of anti-tumor CD8 + T-cell responses elicited by chemotherapy in MMTV-PyMT mice (55). Many of the effects of HO-1 have been attributed to its production of CO as a catabolite of heme degradation which can modulate several signaling pathways including p38 MAPK (70), STAT1/3 (71) and NFκB (72,73). As such, HO-1 activity can compromise anti-tumor CD8 + T-cell responses in the TME (32). The superior tumor control observed when Lyve-1 + TAMs, which are the exclusive tumoral source of HO-1, could not develop their nest structures (Ccr5 -/-), highlight a spatial parameter associated with its immune-suppressive capabilities and further emphasises the collaborative nature of this suppression. It has been demonstrated that macrophage density can also relate to a 'quorum licensing' of macrophage activation (74), it would be interesting to understand how the nest structures might also influence or refine the effector function of Lyve-1 + TAMs in the context of cancer.
It is clear that PvTAMs reside in unique niche arrangements to support their function, such as the described TMEs of metastasis" (TMEM), where a PvTAM, tumor cell expressing a slice variant of mammalian-enabled protein 'Mena' (75), and an endothelial cell are in direct contact to facilitate transendothelial migration of tumor cells into the blood from the tumor (17,(76)(77)(78)(79). However, the heterogeneity of PvTAMs still requires further investigation, as not all PvTAMs expressed Lyve-1 which may represent a progenitor stage or discrete subset of these cells. In this study we characterize the functionality associated with a multi-cellular Lyve-1 + PvTAM structure and define a new collaborative action of these cells to form multicellular biological units which are associated with immune exclusion of CD8 + T-cells from the TME. Interestingly, PvTAMs have been demonstrated to play an active role in neutrophil recruitment to inflamed skin in response to Staphylococcus aureus infection (80), suggesting that PvTAMs may play a gatekeeper role to modulate the immune landscape of the TME, however in MMTV-PyMT tumors the loss of Lyve-1 + TAMs, or their nests, resulted in a specific increase in the abundance of only CD8 + T-cells in the TME. An IL-6 driven CCR5 expression by Lyve-1 + TAMs provide the means to connect a TAM>TAM communication axis of CCR5-CCL3/4 which maintained their nest structures. Interestingly, blockade of CCR5 using maraviroc has been explored in patients with metastatic colorectal cancer (NCT01736813) where 4/6 patients showed a trend towards an increase in CD8 + T-cells within the TME (81). Also, in this study, the authors identified a partial response (3/5 patients) and stable disease (1/5 patients) when maraviroc was combined with a chemotherapeutic agent (81). Although tentative, it highlights intriguing key clinical parallels with our preclinical observations.
In summary, we show that Lyve-1 + TAMs derive from an IL-6 polarization program in the TME and demonstrate that the Lyve-1 + TAM functions are not always autonomous but can be collaborative through their formation of nests within the Pv space using a CCR5-CCL3/4 axis. We demonstrate that these multi-cellular 'nest' structures are biologically important and associated with immune exclusion in the TME and a new resistance mechanism for cytotoxic chemotherapies. This study sheds new light on the collaborative actions of TAMs and suggests their communication could provide novel therapeutic opportunities for targeting their pro-tumoral functions in cancer.
Authors declare no competing interests relating to this work.     Volume= (S 2 xL)/2. MMTV-PyMT Kaede mice were photolabeled under anesthesia. Individual tumors were exposed to a violet light (405nm wavelength) through the skin for a total of nine 20 second exposure cycles with a short 5 second break interval between each cycle. Black cardboard was used to shield the rest of the mouse throughout the photoconversion procedure. Mice for 0 h time points were culled immediately after photoconversion. This photoconversion approach was adapted from that used to label peripheral lymph nodes (85) and was optimized for MMTV-PyMT tumors (16). Tumor tissue for flow cytometry analyses was enzyme-digested to release single cells as previously described (55,84). In brief, tissues were minced using scalpels, and then single cells were liberated by incubation for 60 mins at 37°C with 1 mg/mL Collagenase I from Clostridium Histolyticum (Sigma-Aldrich) and 0.1 mg/mL Deoxyribonuclease I (AppliChem) in RPMI. Released cells were then passed through a 70 μm cell strainer prior to staining for flow cytometry analyses. Viable cells were numerated using a hemocytometer with trypan blue (Sigma-Aldrich) exclusion. For drug treatments, drugs were freshly prepared on the day of injection and administered by intraperitoneal (i.p.) injection using a 26 G needle. Maraviroc (Cayman) was solubilized in ethanol and diluted with saline and administered to mice i.p. using a bi-daily dose of 10mg/kg. 5-fluorouracil (Sigma-Aldrich) was prepared fresh and dissolved in saline at 6 mg/mL and injected to mice i.p. at 40 mg/kg/4 days. Immune-depleted mice were injected i.p. every 4 days, starting 48 h prior to the commencement of treatment, with 400 µg of anti-CD8α (53-6.7) (Thermo Fisher Scientific).
Spleens were crushed through a 70 µm pore strainer and washed through using RPMI.
Liberated splenocytes were centrifuged at 500 x g for 3 mins and the cell pellet was resuspended in 1 mL of red blood cell lysis buffer (Roche) for 2 mins at RT. Cells were then re-  Following RNAscope, immunofluorescence imaging was performed as previously described above. Images were acquired using a Nikon Eclipse Ti-E Inverted spinning disk confocal. To quantify luminescence, a region of interest (ROI) was drawn around a specific area and total photon flux (PF) (photon/second; p/s) was measured. All data was analyzed using the Living Image Software (PerkinElmer).

Transcriptomic data and analysis
TAM, CAF and endothelial Bulk RNAseq and human and mouse TAM scRNA-seq datasets were previously published and described (16,29) and datasets are publicly accessible (see 'Data Availability' section). Downstream analysis was performed using the Seurat v3 R package (87) and analysis pipeline outlined in (16). For upstream regulator analysis we used the QIAGEN IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA) (43). When comparing scRNA-seq datasets between human and mouse TAMs, the Garnett package was used (88) which has previously been employed to perform mouse-human cross-species comparative analysis (89). Murine data and a marker file specifying LYVE1 were provided to Garnett and the model was trained ( train_cell_classifier() ) with default settings, using the same 2000 genes with highest variance chosen for clustering previously (16). Publicly available human data (29) were then classified ( classify_cells() ) with default settings.
Results were projected and plotted on the associated UMAP coordinates from the same data using a customised R script. For Illumina microarray analysis purified mRNA for the respective polarized splenocyte-derived macrophages were cultured for isolated using the PureLink ® RNA Mini Kit (Ambion) according to the manufacturers' protocol. The purity of the isolated mRNA was assessed using a NanoDrop™ spectrophotometer (Thermo Fisher Scientific) and the quality and integrity using an Agilent 2100 Bioanalyzer (Agilent Technologies). mRNA was converted to cDNA, then subsequently amplified using the Ovation ® PicoSL WTA system V2 (NuGen), biotinylated using the Encore ® BiotinIL Module (NuGen), and then hybridized to MouseWG-6 V2.0 Beadchip microarray (Illumina).
Following hybridization, the arrays were washed, blocked, and stained with streptavidin-Cy3 using the Whole-Genome Gene Expression Direct Hybridisation Assay (Illumina).
Microarrays were run on an Illumina iScan system, raw fluorescence signals were collected using GenomeStudio (Illumina), and the data imported into Partek Genomics Suite for analysis. Background was subtracted from the raw data and fluorescence signals were normalized using the quantiles method (90). All p-values were adjusted for multiple testing using the procedure of Benjamini and Hochberg.

Statistics
Normality and homogeneity of variance were determined using a significant. Statistical analysis of tumor growth curves was performed using the "CompareGrowthCurves" function of the statmod software package (91). No outliers were excluded from any data presented.

Study approval
All experiments involving animals were approved by the Animal and Welfare and Ethical

Data availability
The RNA-seq transcriptomic and microarray datasets that support the findings of this study are available through the Gene Expression Omnibus; GSE160561, GSE160641, GSE113034. The microarray datasets are available at GSE192911. The authors declare that all other data supporting the findings of this study are available within the paper and its supplementary information files.