CXCR3-expressing metastasis-initiating cells induce and exploit a fibroblast niche in the lungs to fuel metastatic colonization

Metastatic colonization relies on interactions between disseminated cancer cells and the microenvironment in secondary organs. Here, we show that disseminated breast cancer cells evoke major phenotypic changes in lung fibroblasts to form a metastatic niche that supports malignant growth. Colonization of the lungs by cancer cells confers an inflammatory phenotype in associated fibroblasts, where IL-1α and IL-1β, secreted by breast cancer cells, induce CXCL9 and CXCL10 production in metastasis-associated fibroblasts via NF-κB signaling. These paracrine interactions fuel the growth of lung metastases. Notably, we find that the chemokine receptor CXCR3, that binds CXCL9/10, is specifically expressed in a small subset of breast cancer cells with stem/progenitor cell properties and high tumor-initiating ability when co-transplanted with fibroblasts. CXCR3-expressing cancer cells show high JNK signaling that drives IL-1α/β expression. Thus, CXCR3 marks a population of breast cancer cells that induces CXCL9/10 production in fibroblast, but can also respond to and benefit from these chemokines. Importantly, disruption of this intercellular JNK-IL-1-CXCL9/10-CXCR3 axis significantly reduces metastatic colonization in xenograft and syngeneic mouse models. These data mechanistically demonstrate an essential role for this molecular crosstalk between breast cancer cells and their fibroblast niche in the progression of metastasis.


INTRODUCTION
Metastasis remains the primary threat to the lives of cancer patients with few effective therapeutic options available 1 . In breast cancer, metastases often occur years after the primary tumor has been diagnosed and resected. This indicates that the outgrowth of disseminated cancer cells towards clinically overt metastasis -metastatic colonization -is a rate-limiting step in the metastatic process. Indeed, despite high penetrance across cancer patient populations, successful metastatic colonization is inefficient at the cellular level in that disseminated cancer cells that invade secondary organs confront a suboptimal microenvironment and must cope with strong selective pressure 2 . Metastatic colonization is ultimately not only determined by genetic and epigenetic networks in cancer cells, but also by the microenvironment 3,4 . Indeed, the nontransformed tumor microenvironment affects many aspects of cancer progression. A number of cells within the stroma, such as myeloid progenitor cells, macrophages, neutrophils, endothelial cells and fibroblasts have been implicated in tumor progression and metastasis. Disseminated cancer cells that successfully colonize secondary organs are able to not only withstand the repressive nature of resting microenvironment, but can also re-educate stromal cells to support malignant growth 5 . While the influence of the tumor microenvironment adds to the complexity of cancer initiation and progression, it likely also represents multiple opportunities for therapeutic intervention in metastatic cancer patients.
In cancer, the reactive microenvironment is recognized to have certain features of the microenvironment of healing wounds and regenerative tissues 6 . For example, fibroblasts, that form a heterogeneous group of mesenchymal cells that are commonly found within the connective tissues, play an essential role in tissue regeneration and wound healing 7,8 . Upon tissue injury, fibroblasts alter their phenotype and acquire contractile properties and secrete large quantities of growth factors and extracellular matrix (ECM) proteins. Cancer associated fibroblasts (CAFs), that comprise significant portions of primary tumors, have received considerable attention in recent years and have been shown to regulate both tumor initiation and development 9 . In breast cancer, CAFs provide a cytokine and ECM milieu that promotes growth and progression of primary tumors [10][11][12][13] . Colonization of secondary organs by malignant cells requires modulation of the local microenvironment, leading to the formation of a metastatic niche that supports secondary tumor growth 5 . Whereas the understanding of CAF biology in primary malignancies is substantial and growing, the precise molecular function of stromal fibroblasts at metastatic sites and their effect on metastatic progression is poorly understood. This is particularly relevant when considering the dynamic stromal changes that occur during reprogramming of the microenvironment from early colonization to the growth of overt metastasis.
In this study, we explored the dynamic molecular interactions between disseminated breast cancer cells and fibroblasts during different stages of lung metastasis. We find that fibroblast number and phenotype change dramatically as metastatic nodules grow from micrometastases to macrometastases. Transcriptomic profiling of fibroblasts from metastatic lungs established by breast cancer cells with different metastatic potential reveals that highly metastatic cancer cells can induce early activation in fibroblasts characterized by a major increase in inflammatory and TGFβ signaling as well as proliferation. Two of the most highly induced genes in these metastasis-associated fibroblasts (MAFs) encode the inflammatory cytokines CXCL9 and CXCL10. We find that overexpression of these cytokines confers stem cell properties to breast cancer cells in vitro and promotes metastasis to the lungs in mouse models. Moreover, we find that a subset of breast cancer cells with high metastatic potential expresses the cell surface receptor CXCR3, that binds CXCL9 and CXCL10. Importantly, systemic treatment with an inhibitor of CXCR3 significantly reduces lung metastatic colonization of breast cancer cells in xenograft and syngeneic mouse models. Our data therefore reveals an important crosstalk between breast cancer cells and MAFs that promotes metastatic initiation and progression to overt lung metastasis.

Metastatic breast cancer cells induce early activation and inflammatory signaling in stromal lung fibroblasts
To investigate evolution of fibroblasts during metastatic colonization, we established experimental lung metastases by injecting MDA-MB-231 (MDA) human breast cancer cells or the highly metastatic derivative MDA-MB-231-LM2 (MDA-LM2) cells 14 intravenously into immunocompromised mice. At one week post injection (when lungs harbor primarily micrometastases) and at three weeks post injection (when macrometastases are prominent and widespread), we isolated fibroblasts using fluorescence-activated cell sorting (FACS) (Fig. 1a-1c). Considering the different capacity of MDA and MDA-LM2 cells to grow metastasis in lungs, the experimental approach was designed to address the qualitative difference between MDA and MDA-LM2 associated fibroblasts in metastasis at each time point. Taking advantage of the heterogeneity within both populations, we selected individual mice for analysis that harbored comparable MDA or MDA-LM2 metastatic loads based on in vivo bioluminescence imaging ( Supplementary Fig. 1a). Lung fibroblasts were isolated by FACS using two positive selection markers, PDGFRα and PDGFRβ, and a panel of negative selection markers ( Fig. 1a and Supplementary Fig. 1b). Fibroblasts isolated from lungs with growing metastases were compared with fibroblasts from lungs of healthy, age-matched mice. Interestingly, we observed a substantial increase in the number of fibroblasts in lungs harboring macrometastases derived from both MDA and MDA-LM2 cell lines. In contrast, fibroblast numbers within micrometastases were comparable to those observed in healthy lungs (Fig. 1d). These data suggested that the fibroblast population in lung stroma expands extensively during metastatic colonization of breast cancer cells.
To determine whether stromal lung fibroblasts phenotypically evolve as lung metastases progress, we performed transcriptomic analysis of purified fibroblasts. Principal component analysis (PCA) showed that biological replicates from each experimental group cluster together ( Fig. 1e). Interestingly, fibroblasts from MDA-derived micrometastases, but not MDA-LM2derived micrometastases, clustered close to healthy fibroblasts, whereas fibroblasts from macrometastases by both lines clustered together at a distance from healthy fibroblasts (Fig.   1e). Gene set enrichment analysis (GSEA) showed that, at the micrometastatic stage, MDA-LM2 breast cancer cells uniquely induced fibroblast activation based on early signs of proliferation and inflammation as well as TGFβ-signaling ( Fig. 1f and Supplementary Table 1).
At the macrometastatic stage, however, proliferation and inflammation signatures were strongly induced in MAFs by both breast cancer cell lines (Fig. 1f). Gene Ontology (GO) analysis revealed similar results in that the top 500 genes driving the PCA shift between MDA-LM2-and MDA-associated MAFs were notably involved in cell contraction, proliferation and inflammation ( Supplementary Fig. 1c). In support of these findings, immunofluorescent staining of fibroblasts isolated from lungs harboring micrometastases showed that fibroblast proliferation and expression of alpha smooth muscle actin (αSMA), a marker for reactive fibroblasts, were increased in MDA-LM2-associated fibroblasts compared to MDA-associated fibroblasts ( Fig. 1g and Supplementary Fig. 1d). Importantly, immunohistochemical staining of paraffin sections from human lung metastases from breast cancer patients revealed that 11/12 (92 %) samples exhibited αSMA-expressing fibroblasts ( Supplementary Fig. 2a-c), whereas αSMA expression in healthy lungs was restricted to vessel linings (data not shown), indicating that activated MAFs are also implicated in human metastases. Within metastatic foci, αSMA-positive human fibroblasts were observed in direct contact with cancer cells (Supplementary Fig. 2a). Enhanced cell contractility in MDA-LM2-associated MAFs, suggested by GO analysis (Supplementary Fig.   1c), was functionally confirmed in vitro, as lung fibroblasts demonstrated a significant increase in collagen gel contraction upon stimulation with conditioned medium (CM) from MDA-LM2 cells compared to CM from MDA cells or control medium (Fig. 1h). Inflammatory response signatures were also observed in fibroblasts from MDA-LM2-derived micrometastases and were further enriched in macrometastases (Fig. 1f,i,J). Interestingly, fibroblasts associated with MDA-LM2 micrometastases showed a significant enrichment for genes comprising a stromal-derived "poor outcome" signature from breast cancer patients when compared to fibroblasts from lungs with MDA micrometastases (Fig. 1k). Moreover, this signature was further enriched in fibroblasts isolated from lungs with MDA and MDA-LM2 macrometastases ( Supplementary Fig. 2d). These data support a model in which the phenotype of MAFs is influenced on one hand by the stage of metastatic progression and on the other the metastatic potential of associated cancer cells.
Moreover, these data indicate that transcriptomic changes in MAFs are linked to poor outcome in breast cancer patients.

CXCL9 and CXCL10 are induced in MAFs and promote lung metastasis in mouse models
Our findings led us to hypothesize that changes in stromal fibroblasts during metastatic colonization of the lungs may support the growth of metastasis. To address this, we aimed to identify genes expressed in MAFs that are involved in direct crosstalk with disseminated cancer cells and that are functionally relevant for metastatic growth in the lungs. Transcriptomic analysis of fibroblasts revealed that many genes encoding collagens, ECM glycoproteins or ECM modifying enzymes were markedly induced at macrometastatic stages ( Supplementary   Fig. 3a-c). Several of these genes have been shown to promote cancer and metastasis, such as Tnc, Spp1, Fn1, Thbs2, Lox, and Serpinb2 [15][16][17][18][19] . Based on the link of early transcriptomic changes in MDA-LM2 associated fibroblasts to poor outcome (Fig. 1k), we reasoned that genes induced early in MDA-LM2 associated fibroblasts and further induced in macrometastases would be strong pro-metastatic candidates. Of the 115 genes that were induced in MAFs from MDA-LM2-derived micrometastases, 50 overlapped with genes expressed in MAFs from lungs harboring MDA-and MDA-LM2-derived macrometastases ( Supplementary Fig. 4a,b), and this group comprised a number of genes encoding proteins that are secreted or membrane bound but exposed to the extracellular space. We prioritized these genes for further analysis and identified eight genes that represented candidates for a potential direct crosstalk between metastatic breast cancer cells and MAFs (Fig. 2a).
Among the earliest and most highly upregulated genes encoding secreted proteins were the two inflammatory chemokine (C-X-C motif) ligands 9 and 10 (Cxcl9/10) (Fig. 2a,b and Supplementary Fig. 5a,b). Expression analysis of different cell types isolated from lungs with growing metastases revealed that fibroblasts are the main source of Cxcl9 and Cxcl10 (Fig. 2c).
Importantly, CXCL9 and CXCL10 expression strongly correlated in data sets of dissected distant metastases samples from breast cancer patients (Fig. 2d), indicating co-expression of the cytokines in human metastasis. Therefore, to address the functional role of CXCL9 and CXCL10, we ectopically expressed the genes together in parental MDA breast cancer cells as well as in SUM159 (SUM) cells, a second human breast cancer cell line ( Supplementary Fig.   5c). Combined overexpression of CXCL9 and CXCL10 in MDA and SUM parental breast cancer cells significantly increased their ability to form spheres when cultured under serum-free low adhesive conditions compared to control cells (Fig. 2e). These results suggested that CXCL9 and CXCL10 can confer stem cell properties to breast cancer cells, as the ability to form oncospheres is associated with stem cell features in cultured cells 20 . To analyze the role of these chemokines in metastasis, we intravenously injected breast cancer cells co-expressing CXCL9 and CXCL10 into female, non-obese, diabetic-severe combined immunodeficiency gamma (NSG) mice. Notably, ectopic expression of CXCL9 and CXCL10 significantly promoted lung colonization by both MDA and SUM breast cancer cells (Fig. 2f-i). To address whether both CXCL9 and CXCL10 contribute to oncosphere formation and lung metastasis, we overexpressed the genes individually in MDA or SUM cancer cells ( Supplementary Fig. 5d).
Indeed, both CXCL9 and CXCL10 promoted sphere formation and lung colonization by breast cancer cells ( Supplementary Fig. 5e,f). Together, these data show that CXCL9 and CXCL10 represent components of the metastatic niche that are co-induced in activated MAFs in lungs and support lung colonization.

Metastatic breast cancer cells secrete IL-1α and IL-1β that induce CXCL9 and CXCL10 expression in stromal lung fibroblasts by an NF-κB-dependent mechanism
We next examined how Cxcl9 and CXCL10 are induced in lung fibroblasts during metastatic colonization and whether breast cancer cells directly account for this induction. Gene set enrichment analysis (GSEA) revealed a significant enrichment of pro-inflammatory Interleukin-1 (IL-1) cytokine response and NF-κB signaling in fibroblasts from MDA-and MDA-LM2-derived macrometastases ( Fig. 1f and Table 1). In line with this, we observed a significant correlation between CXCL9/10 and IL1A/B expression in dissected metastases samples from breast cancer patients (Fig. 3b). Therefore, we hypothesized that IL-1α/β present in metastatic lungs may induce Cxcl9/10 expression in lung fibroblasts. Indeed, stimulation with recombinant IL-1α and IL-1β significantly induced expression of CXCL9 and CXCL10 in MRC-5 human lung fibroblasts, and this induction was mediated by NF-κB activity ( Fig. 3c and 3d). Moreover, blockade of IL-1 receptor (IL-1R) signaling through the use of an inhibitory human anti-IL-1R monoclonal antibody blunted the induction of CXCL9 and CXCL10 in fibroblasts by recombinant IL-1α and IL-1β ( Supplementary   Fig. 6a,b). These results indicated that IL-1α and IL-1β mediated induction of CXCL9 and CXCL10 in the fibroblasts via activation of IL-1 receptor and downstream NF-κB activation.
We hypothesized that the cancer cells may be a direct source of the IL-1 ligands that induce CXCL9/10 in MAFs. Indeed, IL1A and IL1B were expressed by MDA and SUM breast cancer cell lines, and IL1A/B expression levels were significantly increased in the respective lung metastatic derivatives MDA-LM2 and SUM-LM1, suggesting an association with metastatic potential (Fig. 4a). To measure protein levels of cancer cell-derived IL-1α and IL-1β in situ, we carried out human-specific enzyme-linked immunosorbent assays (ELISAs) on whole lung homogenates from mice bearing MDA-LM2-derived lung metastases. ELISAs confirmed expression of both cytokines in metastatic lungs (Fig. 4b). Considering IL-1α and IL-1β are secreted cytokines, we investigated whether conditioned medium (CM) of cultured breast cancer cells can drive induction of CXCL9/10 in fibroblasts in vitro. We treated MRC-5 cells with CM from parental (MDA/SUM) or highly metastatic (MDA-LM2/SUM-LM1) breast cancer cells  Supplementary Fig. 6c). Importantly, treatment of lung fibroblasts with CM from MDA-LM2 cells transduced with shRNA against IL-1α/β or treatment of IL-1R knockout (IL1R-KO) fibroblasts with CM from MDA-LM2 cells also prevented upregulation of CXCL10 in the fibroblasts (Fig. 4g-i, Supplementary Fig. 6d). These experiments confirm that IL-1α/β are indeed the factors contained within the CM from metastatic breast cancer cells that drive CXCL10 expression in lung fibroblasts. Importantly, similar effects were observed when we stimulated fibroblasts with CM from patient-derived cancer cells that were collected from pleural effusions of advanced breast cancer patients, as this CM induced CXCL10 in MRC-5 cells in an NF-κB-dependent manner (Fig. 4j,k). To address whether IL-1 cytokines play a functional role in metastatic colonization of the lungs, we injected control and shIL1A/B transduced MDA-LM2 cells intravenously into NSG mice and measured lung colonization by bioluminescence. IL-1α/β knockdown cancer cells showed a significantly reduced ability to colonize the lung, indicating that the IL-1 cytokines are required for the growth of lung metastasis (Fig. 4l,m). Together, these findings suggest that IL-1α/β secreted by breast cancer cells that have reached the lungs directly activate IL-1R on lung fibroblasts to induce NF-κB-dependent CXCL9/10 expression that promotes lung colonization.

Expression of IL1A/B in metastatic breast cancer cells is driven by JNK activity
We previously demonstrated that the JNK signaling pathway promotes lung metastasis via induction of the ECM proteins osteopontin (SPP1) and tenascin C (TNC) 21 . These studies also revealed that JNK activity in breast cancer cells promotes the expression of IL1A and IL1B. As expression levels of both IL1A and IL1B were significantly higher in MDA-LM2 cells compared to their parental line (Fig. 4a), we hypothesized that this may be explained by a higher JNK activity in the metastatic derivative MDA-LM2. Indeed, the JNK response signature 21 was significantly enriched in highly metastatic MDA-LM2 cells compared to MDA parental cells, both in vivo and in vitro ( Fig. 5a and Supplementary Fig. 6e). To confirm regulation of IL1A and IL1B by JNK in breast cancer cells, we determined their mRNA levels in MDA-LM2 cells upon expression of a constitutively active form of JNK, consisting of a protein fusion between JNK1 and its upstream MAPK kinase (MAPKK) activator MKK7 (MKK7-JNK), or a mutated version (MKK7-JNK(mut)), in which the phosphorylation motif Thr180-Pro-Tyr182 in JNK1 is replaced with Ala-Pro-Phe, thereby preventing its activation by MKK7 22,23 . In line with our previous observations 21 , MKK7-JNK expression significantly induced both IL1A and IL1B, and this induction was blunted in MKK7-JNK(mut)-expressing cells (Fig. 5b). Moreover, treatment with a JNK inhibitor (JNKi) significantly reduced endogenous IL1A and IL1B expression in MDA-LM2 cells (Fig. 5c). To determine whether JNK induces IL1A and IL1B via the transcription factor c-Jun, we performed chromatin immunoprecipitation (ChIP), pulling down c-Jun and bound chromatin. qPCR analysis of c-Jun-bound chromatin confirmed that c-Jun binds to IL1A and IL1B promoters in breast cancer cells (Fig. 5d,e and Supplementary Fig. 6f).
Consistent with JNK-driven expression of IL1A/B in cancer cells, CM from MKK7-JNKexpressing MDA-LM2 breast cancer cells increased the production of CXCL10 in fibroblasts compared to CM control, and this increase was blunted when fibroblasts were treated with CM from MKK7-JNK(mut)-expressing cells (Fig. 5f). Importantly, these findings suggested that inhibition of JNK activity in breast cancer cells may alter their ability to induce a pro-metastatic paracrine crosstalk with lung fibroblasts. To test this hypothesis, we pre-treated MDA-LM2 cells overexpressing CXCL9 and CXCL10 or a control vector with JNKi and injected the cells intravenously into NSG mice ( Fig. 5g and Supplementary Fig. 6g). In mice injected with MDA-LM2 control cells, pre-treatment with JNKi significantly reduced metastatic colonization ( Fig. 5h and Ref. 21 ). However, in mice injected with cancer cells overexpressing CXCL9 and CXCL10, no reduction in metastasis was observed in response to JNKi treatment (Fig. 5i). These results indicate that JNK-driven production of IL-1α/β by metastatic cancer cells induces CXCL9/10 in pulmonary MAFs to form a supportive metastatic niche.

CXCR3 marks a subset of breast cancer cells that can both induce and benefit from the pro-metastatic crosstalk with lung fibroblasts
The G-protein coupled receptor CXCR3 is the only cellular receptor known to bind and become activated by CXCL9 and CXCL10 24 . Interestingly, flow cytometric analysis revealed that CXCR3 is expressed in a subpopulation of MDA and SUM cancer cells as well as their metastatic derivatives (Fig. 6a). Moreover, the proportion of CXCR3 + cancer cells was significantly higher when cultured in serum-free sphere conditions, and was higher in the lung metastatic derivative SUM-LM1 compared to the respective parental counterpart (Fig. 6a). Importantly, CXCR3 was also expressed in subsets (range 3.8% -11.1%) of cancer cells isolated from pleural effusions or ascites of four breast cancer patients ( Supplementary Fig. 7a). To further characterize the CXCR3-expressing subpopulation of breast cancer cells, we established transcriptomic profiles of FACS-sorted CXCR3 + and CXCR3 -SUM-LM1 breast cancer cells ( Fig. 6b and Supplementary Table 2). Intriguingly, GSEA revealed that CXCR3 + cancer cells had increased inflammatory signaling and higher JNK activity, and showed characteristics of basal and stem 13 cells of the mammary gland (Fig. 6c,d and Supplementary Fig. 7b,c), in line with the observed increase of CXCR3 + cells in sphere cultures (Fig. 6a). GO term analysis further indicated an enrichment of genes involved in inflammatory signaling and chemokine production (Supplementary Table 3). We therefore reasoned that CXCR3 + cancer cells may secrete higher levels of IL-1α/β, which could in turn lead to elevated production of CXCL9 and CXCL10 in fibroblasts. Indeed, isolated CXCR3 + cancer cells expressed higher levels of IL1A and IL1B compared to CXCR3counter parts (Fig. 6e,f), and CM from CXCR3 + 4T1 mouse mammary tumor cells, but not from CXCR3cells, induced CXCL10 expression in mouse and human fibroblasts ( Fig. 6g and Supplementary Fig. 7d). Importantly, sorted CXCR3 + 4T1 mammary cancer cells had a higher tumor-initiating ability compared to sorted CXCR3cells when coinjected with lung fibroblasts, subcutaneously in limiting dilutions, and resulting tumors by CXCR3 + cells were significantly larger ( Fig. 6h-j). Furthermore, sorted CXCR3 + MDA-LM2 cells had significantly increased abilities to establish metastases in the lung microenvironment compared to CXCR3cells (Fig. 6k,l). Collectively, these data indicate that CXCR3+ metastasisinitiating cells can induce a fibroblast niche in the lung and underscore the importance of a paracrine interaction with fibroblasts in promoting tumor and metastasis initiation.

Inhibition of CXCR3 blocks lung colonization in xenograft and syngeneic mouse models
In addition to the increased ability of CXCR3 + cancer cells to induce Cxcl9/10 expression in lung fibroblasts, this subpopulation of cancer cells is also likely to benefit from this crosstalk. To test whether CXCR3 is functionally required for the CXCL9/10-mediated increase in metastatic ability in breast cancer cells, we used the CXCR3 antagonist AMG-487 (CXCR3i). Stimulation of MDA-LM2 cells, SUM159-LM1 cells and patient-derived breast cancer cells with recombinant CXCL9 and/or CXCL10 increased sphere formation, and this was reversed by addition of CXCR3i (Fig. 7a,b). Importantly, systemic treatment of NSG mice with CXCR3i significantly diminished lung metastatic outgrowth of MDA-LM2 cells (Fig. 7c). Since CXCL9 and CXCL10 14 are known to function in the regulation of immune responses 25, 26 , we tested the effect of CXCR3i in an immunocompetent mouse model. BALB/c mice were injected intravenously with 4T1 mouse mammary tumor cells and concurrently treated with CXCR3i until the experimental endpoint. Lung metastatic outgrowth was also significantly reduced upon CXCR3i in the syngeneic setting (Fig. 7d,e), indicating that systemic antagonism of CXCR3 may be an effective strategy to disrupt cancer cell-fibroblast crosstalk that fuels metastatic colonization of the lungs. To address the putative link between CXCR3-expressing cancer cells and clinical prognosis, we clustered samples from breast cancer patients according to the expression of CXCR3 signature (CXCR3S) (Supplementary Table 2). Analysis of the TOP trial data set and a collection of independent data sets from basal-like breast cancers, revealed that patients with high CXCR3S associated with poor relapse-free survival, distant metastasis-free survival and overall survival (Fig. 7f,g; Supplementary Fig. 8). Taken together, these evidences suggest that CXCR3 + metastasis-initiating cells not only induce CXCL9 and CXCL10 in MAFs, but also take advantage of these cytokines to promote metastatic colonization.

DISCUSSION
Disseminated cancer cells require a supportive niche to successfully form metastases 5 . Indeed, the vast majority of cancer cells face an unfavorable microenvironment at secondary sites and are eliminated following extravasation 27, 28 . Resting stroma can be highly resistant to the establishment of intruding cells 29 , so cancer cells that arrive at distant organs equipped with their own niche components or niche-promoting ability are likely to have a selective advantage.
Our earlier work provides an example of how cancer cells can benefit from bringing own niche components 15 . In this study, we show how enhanced niche-promoting ability fuels metastatic colonization.
JNK signaling promotes metastasis of breast cancer cells through distinct mechanisms. Our previous study showed that successful breast cancer metastasis requires JNK-induced expression of the extracellular matrix and niche proteins SPP1 and TNC 21 . Here, we reveal that JNK signaling also promotes communication between cancer cells and lung fibroblasts and enables highly metastatic breast cancer cells to rapidly establish a supportive niche in the lung.
Our findings suggest a model (summarized in Fig. 8), in which high JNK activity in metastasisinitiating breast cancer cells induces the expression of IL-1α and IL-1β that interact with IL-1R on stromal fibroblasts in the lung to stimulate an NF-κB-mediated induction of Cxcl9 and Cxcl10.
Once secreted from the fibroblasts, CXCL9 and CXCL10 bind to CXCR3 receptor on the surface of a subpopulation of breast cancer cells to complete a paracrine loop between the two cell types that promotes growth of breast cancer metastases. Our results establish a link between stress signaling, the ability of disseminated cancer cells to modify the microenvironment in secondary organs, and their metastatic potential.
The role of inflammatory signaling in cancer is complex and is likely to be context-dependent. At primary sites, IL-1 signaling has been shown to promote tumor growth 10,30 . However, in secondary organs studies have suggested both a pro-metastatic and anti-metastatic roles for IL-1 signaling in models of breast cancer 31, 32 . The divergent IL-1 responses in metastasis may be explained by different breast cancer subtypes in focus. Our study is focused on basal-like breast cancer that has high propensity to metastasize to lung. Inflammatory signaling such as NF-κB is active in basal-like breast cancer 33, 34 , indicating that this breast cancer subtype can adapt to and take advantage of inflammatory signaling. Moreover, we have previously shown that JNK signaling, that induces IL-1α, IL-1β and other pro-metastatic factors in breast cancer cells, is particularly associated with basal-like breast cancer 21 . Finally, our results based on loss-offunction indicate that indeed IL-1, secreted by basal-like cancer cells, is required for metastatic colonization of the lung.
We find that reactive fibroblasts residing near or within metastatic lesions in the lung acquire an inflammatory phenotype reminiscent of the response of fibroblasts to wound healing and primary tumors 10,35 . For example, collagen expression (particularly fibrillary collagens), extracellular matrix proteins (including fibronectin, TNC, and SPP1), and matrix-modifying enzymes (including serpins and lox-family proteins) are highly induced in MAFs. We find that this inflammatory phenotype is associated with a substantial expansion of fibroblasts, resulting in an approximate 50-fold increase in the number of fibroblasts in macrometastatic nodules compared to micrometastases, which is likely derived from the striking increase in proliferation that we observed within fibroblast populations. However, cancer-and metastasis-associated fibroblasts generally represent a heterogeneous group of mesenchymal cells, including resident tissue fibroblasts, pericytes and bone marrow-derived mesenchymal stromal cells 36 , and therefore this expanded population could include several subtypes of fibroblasts with unique functions and diverse origins. Recent studies suggest that indeed tumors may harbor a number of different fibroblast subtypes 37-39 . Certainly, fibroblasts go through distinct phases that are associated with inflammatory and contractile phenotypes during wound healing, but whether this is caused by changes to existing fibroblasts or by an influx of new fibroblast populations with different phenotypes is not well understood. Additional studies are therefore needed to determine whether the evolution of fibroblasts reacting to the metastatic stroma is mechanistically analogous to the changes that occur in fibroblasts during wound healing and whether it involves distinct subtypes.
Increasing evidence suggests that CXCR3 may play an important role during breast cancer progression [40][41][42] . We show that the ability to promote initiation of tumors and metastases is significantly enriched in CXCR3 + cancer cells compared to CXCR3cells. CXCR3 + breast cancer cells secrete IL-1α and IL-1β to stimulate a paracrine crosstalk with lung fibroblasts from which they benefit through their CXCR3 receptor. We find that only a small subset of metastatic breast cancer cells expresses CXCR3, and this subset is characterized by high JNK activity, which we previously linked to mammary stem cell properties 21 . Thus, CXCR3-expressing cancer cells may be enriched in metastatic stem cells that are equipped to exploit the metastasispromoting paracrine loop between fibroblasts and cancer cells. This conclusion is also supported by previous work showing that cancer stem cells are able to take advantage of microenvironmental cues, such as the extracellular matrix protein periostin expressed by reactive lung fibroblasts, which supports stem cell maintenance and metastatic colonization 43 .
Evidence from studies on colon cancer also indicates that fibroblasts play a major role in the maintenance of cancer stem cells 44,45 . As an important addition to these reports, our findings indicate that metastatic stem cells may not only selectively exploit stromal signals at the distant site, but that they may also be selectively efficient in inducing the required stromal signals.
Thus, CXCR3 expression may mark a unique population of breast cancer cells that strategically communicate with stromal fibroblasts to establish a supportive metastatic niche tailored to their phenotype/need.
Previous studies suggest a conflicting function for the CXCL9/10-CXCR3 axis during cancer progression that includes modulation of the immune microenvironment. Studies have shown that CXCL9/10-CXCR3 can mediate the recruitment and activation of T lymphocytes 25 and in melanoma mouse models, this may lead to inhibition of tumor growth and progression 26 .
However, in breast cancer, elevated CXCL10 levels are associated with increased recruitment of CXCR3-expressing, regulatory T cells and reduced anti-tumor immunity in breast cancer patients 46 . These differences underscore the complexity and context dependency of CXCL9/10-CXCR3-mediated T cell regulation. We show that fibroblast-derived Cxcl9/10 promotes lung colonization by directly stimulating the growth of CXCR3 + cancer cells. Importantly, we detected CXCR3 + subsets of cancer cells also in primary cultures of pleural effusion and ascites samples from patients with metastatic breast cancer where the receptor plays a functional role, indicating a relevance of this direct cellular interaction in human metastasis. Our data showing that CXCL9 and CXCL10 are selectively induced in fibroblasts by highly metastatic cells at early stages of metastasis suggests that they may confer a unique metastatic advantage to cancer cells. Primary pleural effusion and ascites samples from metastatic breast cancer patients were cultured in a 1:1 mix of supplemented M199 medium 47 and modified M87 medium 48 as previously described 21 . To study tumor initiation capacities of CXCR3 + 4T1 mouse mammary tumor cells compared to CXCR3 -, sorted cells were seeded in 10 cm dishes in 10 ml D10f and incubated at 37 °C 21 overnight. The following day, 10,000, 5,000 or 500 CXCR3 + or CXCR3 -4T1 cells were mixed with 100,000 BALB/c mouse lung fibroblasts in PBS. Cell suspensions of fibroblasts and CXCR3 + or CXCR3 -4T1 cells were implanted subcutaneously (s.c.) into either flanks of BALB/c mice in a 1:1 mix of PBS and Matrigel. Mice were sacrificed after 20 days and tumor formation was recorded. Tumor sizes were measured with a digital caliper and tumor volume was calculated as (width x length x height)/2.

RT-qPCR
cDNA was generated from total RNA using the High-Capacity cDNA Reverse Transcription Kit

Production of conditioned medium
For production of conditioned medium (CM), cancer cells were seeded in 10 cm culture dishes in 10 ml D10f medium to 70-80 % confluency. Cells were washed once in PBS and 6 ml serumfree MEMα medium were added. After incubation at 37 °C for 48 h, medium was aspirated and filtered through a 0.45 µm filter. Conditioned medium was used directly or aliquoted and frozen at -80 °C.

COMPETING INTERESTS
The authors declare no competing financial interests.               Figure  1E. d, Quantification of Ki67 and αSMA expression in fibroblasts from Figure 1g. P values were calculated by unpaired two-tailed t-tests. Macromet. Micromet. Macromet.

Micrometastasis
Macrometastasis -1.5 Serpine2 Speg Stat1 Tagln  Tfpi2  Tmem26  Top2a  Tpm2 Figure 6. a-b, CXCL9/10 expression in MRC-5 fibroblasts treated with 1 ng/mL recombinant IL-1α (a) or IL-1β (b) in combination with 20 µg/ml IL1R neutralizing antibody or IgG isotype control for 48 h. Expression was analyzed by RT-qPCR. P values were calculated by ratio-paired one-tailed t-tests; n = 3 independent experiments. c, CXCL10 expression in fibroblasts treated with conditioned media (CM) from MDA-LM2 breast cancer cells alone or in combination with 5 μM NF-κB inhibitor (NF-κBi). P value was determined by ratio-paired one-tailed t-test; n = 3 independent experiments. d, IL1A and IL1B expression determined by RT-qPCR in control and double IL1A/B knockdown MDA-LM2 cancer cells. e, Enrichment of a JNK response signature (Insua-Rodriguez et al., 2018) in cultured MDA-LM2 versus MDA parental breast cancer cells (14). NES, normalized enrichment score. P value was determined by random permutation test. f, CHIP-qPCR analysis of c-Jun binding to IL1A and IL1B promoter chromatin in SUM-LM1 cells. g, Ectopic CXCL9/10 expression in MDA-LM2 breast cancer cells analyzed by RT-qPCR.