Myeloid derived Oncostatin M reprogrammes cancer-associated fibroblasts and tumour cells promoting breast cancer progression

Angela M. Araujo1, Andrea Abaurrea1, Peio Azcoaga1, Joanna I. López-Velazco1, Ricardo Rezola2, Iñaki Osorio-Querejeta1, Fátima Valdés-Mora3,4, Juana M. Flores5, Liam Jenkins6, Patricia Fernández-Nogueira7, Nicola Ferrari8, Natalia Martín-Martín9,10, Alexandar Tzankov11, Serenella Eppenberger-Castori11, Isabel Alvarez-Lopez1,2, Ander Urruticoechea1,2, Paloma Bragado12, Nicholas Coleman13, Arkaitz Carracedo9,10,14,15, David Gallego-Ortega3,16 , Fernando Calvo8,17, Clare M. Isacke6, Maria M. Caffarel1,14*#, Charles H. Lawrie1,14,18,#


Introduction
Tumour-promoting inflammation is a hallmark of cancer that influences all stages of tumour progression 1 . Therapies aimed at modulating inflammatory responses in the tumour microenvironment have been of great interest in recent years 2 . The tumour microenvironment (TME) is composed of several different cell types (i.e. fibroblasts, adipocytes, endothelial and infiltrating immune cells) with complex interactions. Cancer cells are constantly communicating with the surrounding stroma and they hijack physiological interactions to overcome immune system surveillance and promote cancer progression 3 . Cancer-associated fibroblasts (CAFs) are a key cell population in the TME 4 . Apart from their very well-known functions in matrix deposition and extracellular matrix remodelling, CAFs have recently been shown to interact with the immune system by responding to and secreting chemokines and However, the interplay between inflammation, cancer cells and other cell types in the TME is not well understood. Cytokines act as central signalling nodes for multicellular interactions.
They are very important in cancer-related inflammatory processes as they are responsible for controlling cell recruitment and modulating immune cell responses in the TME 5 . Here we show that the pro-inflammatory cytokine Oncostatin M (OSM) is a key player in mediating the crosstalk between immune cells, CAFs and cancer cells; and that these immune-stromal-cancer cell interactions favour breast cancer progression and metastasis.
OSM belongs to the IL6 family, which is considered one of the most important cytokine families in the process of tumorigenesis and metastasis 6 . IL6, in particular, is one of the best characterized tumour-promoting cytokines and it is highly up-regulated in many cancers 7 . OSM was first described as an anti-tumoural cytokine due to its anti-proliferative effect in melanoma and other cancer cells 8 . However, in recent years, OSM has been associated with tumour progression as it induces EMT transition, cancer stem-like features, cell migration and metastasis in animal and cellular models 9,10 . High oncostatin M receptor (OSMR) expression correlates with decreased survival in clinical samples of glioblastoma, breast and cervical cancer [11][12][13] . However, the information regarding the role of OSM signalling in the TME is scarce, restricted to reports describing altered mRNA expression in this compartment 14,15 and a role for OSM in macrophage M2 polarization 16 .
By using samples from primary breast tumours, transgenic and orthotopic mouse models of breast cancer, single-cell analysis and in vitro cultures we demonstrate that OSM is a central node for multicellular interactions within the breast TME. OSM is mainly produced by tumourinfiltrating myeloid cells, while its receptor OSMR is highly expressed by CAFs and cancer cells.
OSMR activation in CAFs and cancer cells reprogrammes CAF phenotype and promotes the secretion of inflammatory chemokines, generating a positive feedforward loop that leads to sustained tumour inflammation and consequent breast cancer progression. Importantly, OSMR depletion in both cancer cells and TME, or only in the TME, decreased tumour onset and tumour growth, and reduced metastasis.
This work reveals a clear paracrine signalling between the immune system, CAFs and tumour cells supporting the notion that targeting OSM:OSMR interactions is a potential therapeutic strategy to inhibit tumour-promoting inflammation and breast cancer progression. Of interest, targeting IL6 to block chronic inflammation in cancer is problematic and anti-IL6 drugs have not yielded significant results against solid tumours in clinical trials 17,18 . It may be worth exploring therapeutic targeting of OSM signalling as an alternative. Importantly, humanized antibodies against OSM are being tested in clinical trials for the treatment of inflammatory diseases and have shown to be safe and well tolerated 19 .

Oncostatin M (OSM) expression associates with poor survival in breast cancer.
Although breast cancer is not directly associated with persistent infections or extrinsic chronic inflammation, tumour-elicited inflammation is increasingly being recognized as an important contributor to breast cancer progression 20 . To determine the relevance of the proinflammatory OSM signalling in breast cancer prognosis, we performed immunohistochemistry (IHC) in 141 samples of early breast cancer included in 3 tissue microarrays (TMAs) and observed that high OSM protein levels were associated with decreased overall survival (P=0.029) (Fig. 1a,b). In addition, OSM expression was significantly increased in the cases with high inflammation, assessed by the pathologist as infiltration of inflammatory cells from all lymphoid and myeloid subtypes (Fig. 1b,c). We observed that OSM was mainly expressed by myeloid-like cells as determined by their larger size and more irregular shape (Fig. 1b).
Lymphoid cells, characterized by being smaller and round and by having a round nucleus with little cytoplasm, showed very low or negative OSM expression (Fig. 1b).
We confirmed the association between OSM expression and poor survival in published gene expression profiles of breast cancer, observing that increased OSM mRNA levels associated with decreased disease-free survival ( Fig. 1d and Extended Data Fig.1a) in the Metabric 21 and Wang 22 datasets. By analysing TCGA data by Kaplan-Meier Plotter 23 we observed that high OSM levels were significantly associated with worse overall survival in a large number of cancer types (Extended Data Fig. 1b). These data indicate that OSM associates with poor survival and inflammation in breast cancer and that it could also be relevant in other tumour types.

Deletion of Oncostatin M Receptor (OSMR) in the pre-clinical model MMTV-PyMT impairs tumour progression.
As OSM is mainly secreted by immune cells in the breast cancer microenvironment (Fig. 1b), we decided to use an immunocompetent mouse model to test the relevance of the OSMR signalling pathway in breast cancer progression. We crossed Oncostatin M Receptor (OSMR) deficient mice with MMTV-PyMT as illustrated by the experiment scheme in Fig. 2a. The MMTV-PyMT mouse is a widely used mouse model to study breast cancer progression with an intrinsic tumour microenvironment and immune system 24 . It expresses Polyoma Virus Middle T antigen under the control of the mouse mammary tumour virus promoter/enhancer and develops, from 5-7 weeks of age, spontaneous breast tumours, which metastasize to the lungs within 14 weeks or less. This transgenic model progresses through four distinctly identifiable stages of tumour progression (from benign or in situ lesions to invasive carcinomas) and mimics well the progression of human breast cancer 24 . MMTV-PyMT: OSMR KO females showed a significant delay in tumour onset compared to their WT and HET littermates ( Fig. 2b and Extended Data Fig. 2a,b). In addition, OSMR depletion delayed tumour growth and significantly reduced tumour burden at 14 weeks of age, when WT and HET animals reached a humane endpoint (Fig. 2c,d). Whole mount analysis of mammary glands at 9 week of age revealed reduction of preneoplastic lesions in MMTV-PyMT: OSMR KO compared to control mice (Extended Data Fig. 2c). At 14 weeks of age, most of the control animals (19 out of 21 for WT; and 17 out of 20 for HET) harboured tumours that were malignant carcinomas with necrotic areas, while the KO group contained 7 animals with carcinomas, 7 with pre-malignant adenomas/ mammary intraepithelial neoplasia (MIN) and 1 mouse showed no tumours ( Fig. 2e and Extended Data Fig. 2d, P value = 0,007 for Chi Square test comparing malignant lesions versus pre-malignant lesions or no lesions). Interestingly, when compared to their controls, tumours in OSMR-deficient mice showed decreased levels of the extracellular matrix protein fibronectin, predominantly produced by CAFs 25 (Fig 2f), increased levels of apoptosis, but similar degree of proliferation (Extended Data Fig. 2e). Finally, OSMR deficiency produced a remarkable reduction in the percentage of animals with lung metastases. While 86 and 65 % of WT and HET mice, respectively, developed lung micro or macrometastasis, 80 % of KO mice remained metastasis free (Fig. 2g,h). In summary, our data in the MMTV-PyMT model demonstrate that OSMR signalling is an important contributor of breast cancer progression.
The OSM:OSMR signalling module exhibits a distinct microenvironment-restricted expression.
We performed single cell RNA-seq analysis of mammary tumours from the MMTV-PyMT model to decipher which cells were responsible to produce OSM and to express OSMR in the breast cancer context. Our data indicate that the ligand OSM is almost exclusively expressed by the myeloid cell population (as suggested by the OSM staining in Fig. 1b), while the receptor OSMR is mainly expressed by the fibroblasts and some of the cancer epithelial clusters (Fig. 3a-c). The pattern of expression of OSM:OSMR differs from the one observed for other cytokine-receptor pairs of the same family such as IL6:IL6R (Il6ra) and LIF:LIFR (Fig. 3b,c), supporting that OSM exerts distinct and unique functions from other members of the family 26 . Il6st (GP130) is the common subunit receptor for OSM, IL6, LIF and other cytokines of the family and is ubiquitously expressed (Fig. 3b,c), being the expression of the other receptor subunits more To prove the relevance of our previous findings in human cancer clinical data, we used the xCell 30 and TIMER 31 web resources to analyse the association between OSM and OSMR expression and TME composition in two different clinical breast cancer datasets 32,33 . xCell analysis showed that OSMR mRNA expression significantly correlates with fibroblast enrichment in human breast cancer, while OSM mRNA levels show the most significant associations with macrophage M2-like and common myeloid progenitor signatures (Fig. 4a).
The OSM and OSMR correlations with fibroblasts and myeloid cell infiltration respectively, were validated by TIMER in a different clinical dataset (Extended Fig.4a). This analysis also showed that OSM and OSMR mRNA expression inversely correlated with tumour cell purity. In line with these results, OSM and OSMR mRNA levels were found to be elevated in breast cancer stroma compared to the epithelial compartment in a different dataset (Fig. 4b).
Moreover, both OSM and OSMR are increased in human breast cancer stroma compared to healthy stroma (Fig. 4c). A similar pattern of OSM:OSMR expression was observed in other cancer types including ovarian and colorectal cancer (Extended data Fig. 4b,c). Altogether, our data reveal that OSM and OSMR are stroma-expressed molecules, and point to paracrine OSM:OSMR signalling in cancer, as ligand and receptor are expressed by different cell types in the tumour microenvironment.

Stromal OSMR deletion hampers tumour progression and aggressiveness.
To test the hypothesis that stromal OSM:OSMR signalling contributed to breast cancer progression we injected TS1 cells, derived from a MMTV-PyMT tumour 27 , orthotopically into the mammary gland of syngeneic OSMR deficient (KO) and wild-type (WT) control mice ( As we previously observed that fibroblasts were the cell population with higher levels of OSMR within the stroma (Fig. 3), we performed complementary in vitro and in vivo experiments to assess the effect of OSMR activation in mammary cancer-associated and normal fibroblasts derived from human breast tumours and reduction mammoplasty surgeries 34 , respectively.
The ability to remodel the extracellular matrix is a hallmark of CAFs 25 . Importantly, OSM treatment enhanced the capacity of CAFs (CAF-173 and CAF-318) to contract collagen matrices and, interestingly, the effect was not observed in non-cancerous skin and breast fibroblasts (HS27 and RMF-31, respectively) (Fig. 5a). To further investigate the role of OSM in potentiating CAFs activation, we selected RMF-31 to be used as a model of normal breast fibroblasts and CAF-173 as a model of CAFs. In accordance with the contractility experiments, OSM promoted the growth of 3D CAF spheroids while it did not affect normal mammary fibroblasts 3D spheroids (Fig. 5b). Similarly, OSM induced the expression of classical CAF markers such as FAP, POSTN, VEGF and IL6 25 , only in CAF-173 CAFs, and not in normal RMF-31 fibroblasts (Fig. 5c). Of interest, OSMR was similarly expressed in normal and cancer fibroblasts  Fig. 5k), suggesting reduced levels of CAFs in this experimental group, probably due to impaired CAF proliferation upon OSMR reduction, in line with the increased size of CAF spheres observed after OSMR activation (Fig. 5b). Together, our data prove that OSM:OSMR signalling activates CAFs and that this contributes to cancer progression.

OSM:OSMR signalling in CAFs and cancer cells induces cytokine secretion and myeloid cell recruitment, establishing a positive feedback loop.
In an attempt to understand how OSMR activation in tumours was inducing malignancy we analysed transcriptomic data of CAFs (CAF-173) and breast cancer cells (MDA-MB-231) activated with OSM. Microarray data indicated that pathways related to leukocyte recruitment and neutrophil degranulation were significantly enriched upon OSM activation ( Fig. 6a and Extended Data Fig. 6a). In addition, GSEA analysis showed that the inflammatory response hallmark signature from Molecular Signatures Database (MsigDB) 37 was enriched in both CAF-173 and MDA-MB-231 activated with OSM (Fig. 6b). These data suggested that, upon OSMR activation by OSM, both CAFs and cancer cells could be involved in shaping the tumour microenvironment by recruiting leukocytes to the tumour site. Analysis of a panel of 31 chemokines by antibody array showed that OSM induced expression of important chemoattractants ( Fig. 6c and Extended Data Fig. 6b). Some of these factors were exclusive of CAFs (mainly CXCL10 and CXCL12), others of cancer cells (mainly CXCL7 and CCL20) and some factors, such as CCL2, were common for both cell types. Vascular endothelial growth factor (VEGF) can also modulate tumour immunity by inducing macrophage and myeloid-derived suppressor cells (MDSCs) recruitment 38 and we previously showed that it is an OSMR target 35 .
As seen in Fig. 6d, VEGF levels were increased upon OSM treatment both in CAFs and tumours cells. As some of the OSM-induced chemokines are potent myeloid chemoattractants (i.e. VEGF, CCL2, CXCL12 39,40 ), we sought to determine whether myeloid cell populations were altered in tumours after OSMR signalling abrogation. We performed immunostaining of macrophages and neutrophils, assessed by F4/80 and Ly6G positivity respectively 3,41,42 , and we observed that these two populations were reduced in MMTV-PyMT: OSMR KO tumours compared to OSMR WT tumours (Fig. 6e). Interestingly, VEGF, CXCL1 and CXCL16 levels were reduced in the serum of tumour bearing MMTV-PyMT: OSMR KO mice compared to control mice ( Fig. 6f), all factors being involved in myeloid cell recruitment 38,43,44 . In summary, these results show that macrophages and neutrophils are reduced in OSMR deficient tumours and that chemokines induced by OSM in tumour cells and in stroma may be responsible for myeloid cell recruitment. Our findings are clinically relevant as VEGF, CXCL1 and CXCL16 mRNA expression is associated with decreased recurrence-free survival in breast cancer patients (Fig.   6g), and strongly correlated with OSM and/or OSMR levels (Fig. 6h).
As OSM is mainly expressed by myeloid cells (Figs. 3 and 4), our data point to the existence of a feedback positive loop where OSM signalling in both cancer cells and CAFs induces the recruitment of more myeloid cells which will in turn secrete more OSM within the tumour.
Moreover, conditioned media from cancer cells pre-treated with OSM further increased OSM expression in macrophage-like differentiated HL-60 cells (Fig. 6i). We did not observe this effect with conditioned media from OSM-activated CAFs or with OSM itself. These results suggest that OSMR activation in cancer cells, not only increases the secretion of chemokines involved in myeloid cell recruitment, but also induces OSM expression in these myeloid cells, doubly contributing to the aforementioned feed-forward loop.

Discussion
Cytokines are important players in inflammation, a process associated with tumour progression 5 . Even the cancers not directly associated with persistent infections or chronic inflammation, such as breast cancer, exhibit tumour-elicited inflammation, which has important consequences in tumour promotion, progression and metastasis 6,20 . Understanding how inflammatory signals orchestrate pro-malignant effects in the different cell compartments within the tumour microenvironment is key to design new therapeutic strategies to target tumour-promoting inflammation. Our study identifies the pro-inflammatory cytokine OSM as a crucial mediator of the crosstalk between different cell types within the tumour by activating an intriguing pro-tumoural "ménage-à-trois" between myeloid cells, CAFs and cancer cells (Fig.   7). By analysing multiple datasets of clinical samples and a wide array of animal and cellular models, we demonstrated that myeloid derived OSM promotes tumour progression by activating the stroma. OSM activates CAFs by increasing their contractility and proliferation. In addition, OSM activates a pro-inflammatory signature in CAFs and cancer cells which, in turn, recruit more myeloid OSM-secreting cells and induce OSM expression by macrophages, establishing a feed forward positive loop (Fig. 7).
Our single cell RNA-seq and FACS sorting analyses revealed that OSM and OSMR have a unique expression pattern in breast tumours, compared to other members of the IL6 cytokine family ( Fig. 2 and Extended Data Fig.2). While the ligand OSM was only expressed by the myeloid cell populations, we found that the receptor OSMR was mainly expressed by fibroblasts, cancer and endothelial cells. Our data resonate well with other reports describing OSM secretion by leukocytes (myeloid but also T cells) and OSMR expression in stromal cells (i.e. fibroblasts, endothelial cells, muscle cells, adipocytes and osteoblasts) in different physiopathological contexts 26 . In our hands, the T cell compartment exhibits little or no OSM expression and its production is restricted to myeloid cells. Whether there is one myeloid cell population mainly responsible for OSM production in breast cancer, or whether OSM is secreted by different immune cell types (including neutrophils, TAMs and MDSCs) remains to be determined. Of interest, our analysis of clinical breast cancer samples revealed a significant positive correlation between OSM expression and presence of monocytes, macrophages, and neutrophils, being the latter the cell population that showed the strongest association with OSM (Extended Data Fig. 4a). Importantly, a recent report identified OSM as one of the key signalling mediators of neutrophil-cancer cell interactions in pro-metastatic clusters of neutrophils and circulating tumour cells (CTCs) 45 .
The cell population showing the most significant association with OSMR expression in human breast cancer samples is the cancer-associated fibroblasts compartment, suggesting an important role for this cell type in transducing OSM signalling in the tumour. In fact, our coinjections of human immortalized CAFs with breast cancer cells demonstrated that activation of OSMR by OSM in fibroblasts leads to a more aggressive cancer phenotype. While the protumoural effect of OSMR activation in cancer cells has been extensively described 11,12,46 , little is known about the effects of OSM in the tumour stroma and our results shed light on the effects of OSM signalling in CAFs. It has been previously reported that OSM stimulates actomyosin contractility and matrix remodelling by oral SCC derived CAFs 47 . In addition to an effect in contractility, we observed an increase in proliferation and a pro-inflammatory phenotype of OSM activated CAFs. OSM also induces in CAFs the expression of bona fide activation markers such as FAP and periostin. Importantly, a signature comprising the top 4 OSM target genes in CAFs (SERPINB4, THBS1, RARRES1 and TNC) is associated with decreased recurrence-free survival (Fig.1, Extended Fig.1 and 4e). Tenascin C (TNC) and Thrombospondin 1 (THBS1) are extracellular matrix proteins involved in cell adhesion, and abundant in breast cancer stroma 48 . RARRES1 (also known as TIG1) has been involved in breast cancer progression 49 and SERPINB4 is up-regulated in many cancer types 50 .
In addition to the aforementioned effects, OSM induces in CAFs, but also in cancer cells, the secretion of important myeloid chemoattractants. Some of them are fibroblast-specific (i.e. In addition, we also observed that cancer cells, but not CAFs, contribute further to the OSM- impact. Importantly, we found that OSMR depletion in the MMTV-PyMT breast cancer model has a profound effect in reducing metastasis, compared to its effect in tumour onset and tumour growth (Fig. 2). In view of these results, the role of OSM:OSMR signalling in the premetastatic niche requires further investigation.
Analysis of human breast cancer samples revealed that our results are clinically relevant, showing that OSM and its receptor OSMR are upregulated in breast cancer stroma, and OSM expression and its transcriptional signature are associated with decreased survival. OSM:OSMR interactions could be blocked by antibody based inhibition, a strategy that has had a major impact on cancer 62 , which makes them a promising candidate for therapeutic targeting. TMAs were generated by punching 1 mm spot of each sample. Complete histopathological information, date and cause of death, as well as date of local and/or distant relapse were available for all the patients. Tissue sections were subjected to a heat-induced antigen retrieval step prior to exposure to primary OSM antibody (1:50, HPA029814, Sigma-Aldrich).
Immunodetection was performed using the Roche Ventana BenchMark ULTRA IHC/ISH staining system, with DAB as the chromogen. Cases were reviewed by two independent pathologists and OSM staining was evaluated by the semiquantitative method of H-score (or "histo" score), used to assess the extent of immunoreactivity in tumour samples 63
Collagen cell contraction assays. To assess the collagen remodelling capacity 77 , fibroblasts were treated for 4 days with recombinant human OSM (R&D Systems) at 10 ng/μl or vehicle (PBS) before being embedded in collagen (2mg/ml, Corning). After polymerization, collagen gels were detached, and they were treated with OSM (10ng/ml) or vehicle. Pictures were taken 48 hours later, and the area of collagen disks was analysed using Fiji-Image J software. shOSMR CAFs were co-injected in GFR matrigel. Animals were monitored 3 times a week and tumour growth measured using an external calliper. Animals were culled once tumours reached the maximum allowed size. In all mouse experiments, after animal culling lungs were visually inspected for macroscopic metastases, and mammary glands and lungs were fixed in neutral buffered formalin solution (Sigma-Aldrich). Microscopic metastases were determined by H&E staining of formalin-fixed paraffin-embedded sections. Tumours were divided in portions for 1) preparation of tissue sections for H&E and IHC staining (fixed in formalin) and 2) protein and RNA extraction (snap frozen).
Flow cytometry sorting. TS1 cells were injected orthotopically in FVB mice as described above, and 15 days after injection, freshly obtained TS1 tumours were dissociated into single cell suspension and stained with the antibodies described in Supplementary Table 1 Statistical analyses. Statistical analyses were performed using GraphPad Prism or SPSS software. For Gaussians distributions, the student's t-test (paired or unpaired) was used to compare differences between two groups. Welch's correction was applied when variances were significantly different. One-way ANOVA was used to determine differences between more than two independent groups. For non-Gaussian distributions, Mann-Whitney's test was performed. Chi-square test was used to determine differences between expected frequencies.
For Kaplan-Meier analysis the Log-rank (Mantel-Cox) test was used. P values inferior to 0.05 were considered statistically significant. Unless otherwise stated, results are expressed as mean values +/-standard errors (SEM).     weeks of age. Lung tumour masses were classified as macrometastases when they were visible at dissection and as micrometastases when they were only detectable by hematoxylin and eosin staining. P value was determined comparing animals with metastasis (macro and micro) vs. non-metastasis using Chi-square test. h) Representative pictures of lung metastases at week 14 in OSMR WT, HET and KO animals. ** p < 0,01; *** p < 0,001.  combined in e and g. e) P value was calculated using the Mantel-Cox test. f,g) P values were calculated using the unpaired two-tailed t test and graphs represent mean ± SEM. **, p < 0,01. represents the percentage of animals with detectable qPCR signal and P value was calculated using the Chi-square test. n.s. non-significant. * p < 0,05; *** p < 0,001. P values were calculated using the unpaired two-tailed t test unless specified.