Tenascin‐C immobilizes infiltrating T lymphocytes through CXCL12 promoting breast cancer progression

Abstract Immune checkpoint therapy, where CD8 tumor infiltrating T lymphocytes (TIL) are reactivated, is a promising anti‐cancer treatment approach, yet with low response rates. The extracellular matrix, in particular tenascin‐C, may generate barriers for TIL. To investigate this possibility, we used a MMTV‐NeuNT and syngeneic mammary gland grafting model derived thereof with engineered tenascin‐C levels and observed accumulation of CD8 TIL in tenascin‐C‐rich stroma. Inhibition studies revealed that tenascin‐C induced CXCL12 through TLR4. By binding CXCL12, tenascin‐C retained CD8 TIL in the stroma. Blockade of CXCR4, the receptor of CXCL12, enhanced macrophage and CD8 TIL infiltration and reduced tumor growth and subsequent metastasis. Retention of CD8 TIL by tenascin‐C/CXCL12 was also observed in human breast cancer by tissue staining. Moreover, whereas high CD8 TIL numbers correlated with longer metastasis‐free survival, this was not the case when also tenascin‐C and CXCL12 levels were high. Altogether, these results may be useful for improving tumor immunity as diagnostic tool and to formulate a future “TIL‐matrix‐release‐and‐reactivate” strategy.


Introduction
Immune checkpoint therapy (ICT) is a promising approach to activate the body's immune system to fight cancer. Yet, in most breast cancer patients, ICT is only efficient in up to 20% of patients (Duraiswamy et al, 2013). Inactive CD8 TIL can get reactivated with antibodies targeting PD1, PD-L1, or CTLA (Duraiswamy et al, 2013;Huang et al, 2017); however, CD8 TIL are often either scarce or located inside the tumor stroma not reaching the tumor cells (Joyce & Fearon, 2015). The "immune contexture" and tumor microenvironment (TME) have been identified as potential obstacle in ICT (Fridman et al, 2012;Zemek et al, 2020). In particular, the so-called "immunoscore" takes into account the exclusion of CD8 TIL from tumor cell clusters (Galon & Bruni, 2019) which are often organized as nests surrounded by stroma, that is rich in extracellular matrix (ECM) (Pickup et al, 2014;Spenl e et al, 2015). Yet, the roles of the ECM in controlling CD8 TIL infiltration are not well understood (Erdag et al, 2012;Galon et al, 2012;Joyce & Fearon, 2015).
The ECM molecule tenascin-C (TNC) is highly expressed in the tumor stroma in regions called tumor matrix tracks (TMT), that are composed of several matrix molecules and are enriched with fibroblasts and leukocytes (Spenl e et al, 2015;Spenl e et al, 2020). TNC is also expressed in regions with high collagen-rich tumor-associated collagen signatures (TACS) that correlate with poor breast cancer patient survival (Tomko et al, 2018). TNC has many contextdependent functions (Midwood et al, 2016). TNC can impair adhesion, proliferation, and function of cytotoxic T lymphocytes (CTL) and can be degraded by autophagy that is defective in triplenegative breast cancer causing CTL impairment (R€ uegg et al, 1989;Hemesath et al, 1994;Hauzenberger et al, 1999;Parekh et al, 2005;Huang et al, 2010;Li et al, 2020). TNC skewed macrophages toward a M2 phenotype through TLR4 promoting lung metastasis in the syngeneic NT193 breast cancer model, that has been used in the current study (Deligne et al, 2020). Moreover, in an oral squamous cell carcinoma model, TNC promoted an immune suppressive TME through CCR7/CCL21 (Spenl e et al, 2020). Altogether, TNC may contribute to evasion from immune surveillance in breast cancer by regulating CTL function but how is largely unknown.
Here, by investigating two breast cancer progression models with engineered high and low TNC levels, we demonstrated that TNC contributes to the escape from anti-tumor immunity by regulating CD8 TIL. TNC impaired CD8 TIL function through induction of CXCL12. TNC bound CXCL12, thus immobilizing CD8 TIL in the matrix. Subsequently, these lymphocytes were impaired in reaching and killing the tumor cells. Upon inhibition of CXCR4, CD8 TIL were released from the TNC-rich stroma, now accessing and killing the tumor cells causing reduced tumor growth and subsequent metastasis. This mechanism may be relevant in human breast cancer where CD8 TIL accumulation in the TNC-rich stroma in context of high CXCL12 correlated with shortest survival of breast cancer patients.

TNC regulates expression of immunity-associated genes in the MMTV-NeuNT model
To investigate how TNC impacts tumor growth, we used the MMTV-NeuNT (NeuNT) model on a wild-type (WT) and TNC knockout (KO) background. In this model, we have previously shown that TNC increased tumor onset and lung metastasis (Sun et al, 2019). We compared overall gene expression in tumors isolated from WT and TNCKO mice by an Affymetrix chip array. This analysis revealed 2,260 genes to be differently regulated between tumors of both genotypes (P < 0.05) (Fig 1A- Table S1). Gene ontology analysis showed differential expression of several genes including 136 immune-related genes (75 downregulated ( Fig 1B) and 61 upregulated (Fig 1C)) and reduced expression of type I interferon (IFN) responsive genes in WT compared to TNCKO tumors (Figs 1A and EV1A,B,Appendix Fig S1A and B,Appendix Table S1). We also noticed that more than 10 Mhc type II molecules and chemokines such as Cxcl12 were higher and lower in the absence of TNC, respectively, suggesting an impact of TNC on tumor immunity (Fig 1B and C). By flow cytometry, we investigated the immune cell infiltrate and observed that F4/80 + and iNOS + macrophages were more and less abundant in WT than TNCKO tumors, respectively (Fig 1D and E). Moreover, T cell subtype numbers and in particular CD8 T cells (CD44 + and PD-1 + ) were in general very low in these late-stage tumors with no difference between genotypes (Appendix Fig S1C-F). Interestingly, the CD8 T cell activation marker Perforin and several T cell-regulating genes were higher in TNCKO tumors, indicating that TNC impacts CTL function (Figs 1B and EV1B,Appendix Fig S1G).

TNC promotes stromal CD8 TIL enrichment and enhances survival of tumor cells in the syngeneic NT193 breast cancer grafting model
Due to poorly abundant CD8 TIL in late-stage NeuNT tumors (confirmed by immunofluorescence staining (IF), Appendix Fig S1C and S2A), we decided to use our novel syngeneic orthotopic NT193 grafting model with more numerous CD8 TIL (Deligne et al, 2020). In the previous study, the TNC-low condition had some residual TNC that was provided by the host. Here, we engrafted Tnc knockdown (shTNC) NT193 tumor cells into a TNCKO host to obtain a close to TNC-negative condition where indeed TNC is almost undetectable (Sun et al, 2019). For some experiments, it is relevant to use a TNCKO host to mimic the TNC-negative condition as host TNC plays a role in promoting lung metastasis by impacting tumor cell survival and plasticity in the blood vessel invasions (BVI), an important precursor of metastasis (Siegel et al, 2003;Sun et al, 2019). As illustrated in Fig EV1C and D, we compared tumors with high and low TNC levels. In the TNC-high condition control (shC), tumor cells were engrafted into a WT host (WT/shC). For TNC-low tumors, a WT or TNCKO host was engrafted with shTNC cells, giving rise to WT/shTNC or KO/shTNC tumors, respectively. We used different time points (4, 5, 7, and 11 weeks) and grafting methods (iv, orthotopic) as indicated and outlined in Fig EV2. We investigated gene expression in all grafted tumors by RNA-seq analysis and compared gene expression levels to that in NeuNT tumors (Fig EV1B-D). This comparison revealed clear differences between tumors and showed that in tumors with lowered TNC, gene ▸ Figure 1. Loss of the TNC protein in NeuNT tumors impacts gene expression and the immune cell infiltrate.
Source data are available online for this figure. expression was clearly different to tumors with abundant TNC. Moreover, similarities in gene expression were seen in the TNC-high group (NeuNT-WT, WT/shC) and TNC no/low group (NeuNT-KO, WT/shTNC, KO/shTNC), respectively (Fig EV1B and  D, Appendix Table S2). We previously noticed that macrophages were abundant in the stroma of TNC-high (WT/shC) tumors (Deligne et al, 2020). Now we wondered whether TNC influenced the localization of macrophages and other immune subtypes which we addressed by tissue staining in the 11-week model ( Fig EV2) and observed that CD45 + leukocytes were present inside the stroma as well as in the tumor cell nests. As previously seen for macrophages (F4/80) (Deligne et al, 2020), also CD4 T cells and CD11c myeloid/dendritic cells were present inside the tumor nests and the stroma with no apparent difference between genotypes (Fig 2A). This was different for CD8 TIL. Whereas CD8 TIL were similarly abundant in TNC-high and TNC-low (KO/shTNC) tumors (Fig 2B), they resided predominantly inside the stroma of TNC-high tumors, in contrast to TNClow (KO/shTNC) tumors where CD8 TIL more frequently invaded the tumor cell nests (Fig 2A and C). As CD8 TIL were found in the stroma, we asked what contribution tumor and stromal TNC has on this retention. Therefore, we quantified CD8 TIL in WT/shTNC tumors and compared results to KO/shTNC and WT/shC tumors which revealed a tendency of more CD8 TIL infiltration when tumor cell-derived TNC was lowered (WT/shTNC). However, tumor nest infiltration of CD8 TIL was only significantly higher in the absence of host TNC (KO/shTNC). This result clearly shows that host TNC is important in retaining CD8 TIL (Appendix Fig  S2B). To address whether CD8 TIL in the tumor nests had an impact on survival, we stained for cleaved caspase 3 (Cl. casp3) and saw more apoptosis in TNC-low (KO/shTNC) than in TNChigh tumors (Fig 2D and E). Costaining for CD8 and Cl. casp3 revealed no colocalization, but vicinity, suggesting a paracrine mechanism ( Fig 2F). In support, we saw higher expression of Granzyme B (Gzmb) and Perforin (secreted by activated CD8 TIL (Kelso et al, 2002)) in TNC-low (KO/shTNC) than TNC-high tumors (Fig 2G and H). As TNC is expressed in the lungs of NT193 tumor mice (Sun et al, 2019), we addressed whether TNC also impacted CD8 TIL in lung metastasis by determining the abundance of CD8 TIL by tissue staining. We noticed a higher infiltration of CD8 TIL in lung metastases from WT than TNCKO mice (Fig 2I and J) correlating with higher metastasis in WT mice (Sun et al, 2019). Altogether, our results point at CD8 TIL localization and function to be regulated by TNC.

Tenascin-C upregulates and binds CXCL12
To gain insight into the underlying mechanism, we consulted our RNA-seq results which revealed 1,502 genes to be differently expressed, including Cxcl12. Here, we focused on CXCL12, a known regulator of tumor immunity (Okabe et al, 2005;Zhang et al, 2019) that was more expressed in TNC-high in comparison to TNC-low (WT/shTNC, KO/shTNC) tumors and in cultured tumor cells which we confirmed by qRT-PCR and ELISA (Figs 3A-C and EV3A and B, Appendix Table S2). Tumors derived from shC-engrafted tumor cells had the highest CXCL12 levels irrespective of the host, and shTNCengrafted tumor cells expressed less CXCL12 altogether pointing at the tumor cells as major source of CXCL12 (Figs 3B and ). We confirmed higher Cxcl12 levels also in NeuNT-WT than TNCKO tumors (qRT-PCR) and in the supernatant (conditioned medium, CM) of shC in comparison to shTNC cells (qRT-PCR, ELISA) (Figs 3C and D, and EV3D). Next, we addressed how TNC induced Cxcl12 by using inhibitors according to established protocols for tyrosine kinase inhibitors (sunitinib), integrin a4b1 and a9b1 (BOP), TGFb receptor 1 (TGFbRI, GW788388), and Toll-like receptor 4 (TLR4, Cli95) on tumor cells stimulated with TNC. In contrast to BOP, GW, and sunitinib that did not affect Cxcl12 expression, Cli95 reduced TNC-induced Cxcl12 demonstrating that TNC regulates CXCL12 expression through the known TNC receptor TLR4 like other molecules (Fig 3E;Midwood et al, 2009;Spenl e et al, 2020).
Next, we used tissue staining to assess the relative distribution of CXCL12 within the tumor. We noticed higher CXCL12 expression in the tumor nests than in the stroma as also documented in human tumors (Toullec et al, 2010;Lefort et al, 2017). In contrast, TNC was highly expressed in the stroma but not in the tumor nest which we confirmed by using 3 different antibodies (Appendix Fig S2B and  C). We cannot explain how CXCL12 expression is regulated in the tumor nests. TNC may play a role by inducing CXCL12 in the stroma to either bind to TNC and/or diffuse into the tissue. CXCL12 gradients in tumors have been described but how they are established is not well understood (Spinosa et al, 2017).
TNC interacts with many molecules (De Laporte et al, 2013;Midwood et al, 2016;Spenl e et al, 2020). Therefore, we asked whether TNC potentially binds CXCL12. Indeed, by surface plasmon resonance measurement, CXCL12 was found to bind TNC in a range of 790 nM (35-fold lower than binding to CXCR4; Richter et al, 2014; Fig 3F). Moreover, by negative staining electron microscopy, we confirmed CXCL12 binding to TNC. CXCL12 labeled with colloidal gold particles bound to TNC predominantly inside the ▸ Figure 2. Impact of TNC on CD8 TIL in the syngeneic NT193 breast cancer model. fibronectin type III (FNIII) domains, including the 5 th FNIII repeat that was previously shown to bind several other molecules (De Laporte et al, 2013;Spenl e et al, 2020). Additional but much lower binding was also seen in the EGFL repeats and the fibrinogen like globe (FBG) whereas control gold particles (not chemokine coupled), or EGF-or BSA-coupled gold particles (Spenl e et al, 2020), did not bind TNC (Figs 3G and H,and EV3E). Moreover, binding of CXCL12-coupled beads to TNC was competed with free CXCL12 and with increasing concentrations of heparin implying interactions via charged moieties (Fig EV3F and G).

CXCL12 immobilizes CD8 T leukocytes in a CXCR4-dependent manner
We investigated whether TNC influenced expression of the CXCL12 receptors CXCR4 and/or CXCR7 (Bleul et al, 1996;Balabanian et al, 2005) and observed that Cxcr4 levels were higher in TNC-low (WT/shTNC, KO/shTNC) tumors and in cultured tumor cells upon TNC knockdown, revealing an opposite regulation than Cxcl12 by TNC (Appendix Fig S3A). No difference in Cxcr7 mRNA levels was seen in cultured tumor cells in dependence of TNC (Appendix Fig S3B).
To address whether CXCL12 regulates CD8 T cell function, we compared transwell migration (attraction or retention) of CD8 T cells toward defined substrata, collagen I (COL), fibronectin (FN), or TNC together with CXCL12 and CXCL10 or CCL21, respectively, as previously described (Fig 4A, Spenl e et al, 2020). CXCL12 attracted more CD8 T cells than CXCL10 or CCL21. In contrast to FN or COL, CD8 T cell attraction by CXCL12 was surprisingly largely reduced by TNC ( Fig 4B). However, we found more CD8 T cells to be immobilized on TNC than on the other substrata, with a more potent effect of CXCL12 than CXCL10 or CCL21 ( Fig 4C). As TNC binds CXCL12, potentially generating an adhesive substratum, we washed the coating after incubation with CXCL12 and observed that TNC plus CXCL12 indeed retained CD8 T cells whereas the other substrata did not. This effect was CXCR4 dependent as AMD3100 (AMD), a CXCR4 inhibitor, blocked CXCL12-but not CXCL10-dependent CD8 T cell retention (Fig 4D and E).
To address whether CXCL12 is the major chemoattractant by which TNC immobilizes CD8 T cells, we used conditioned medium (CM) from cultured tumor cells in the migration and retention assay, respectively. In contrast to CM from shTNC cells (low CXCL12) that was poorly cell attracting, CM from shC cells (high CXCL12) strongly attracted CD8 T cells toward the FN and COL but less to the TNC coating ( Fig 4F). As for CXCL12, also with the CM from shC cells more CD8 T cells adhered on TNC than on the other substrata which were not seen with CM from shTNC cells ( To rule out other potential effects of AMD on CD8 T cells, we investigated CD8 T cell survival, proliferation, and function upon activation by an anti-CD3 antibody and inhibition of CXCR4. In contrast to Bax and Bcl-w that were unchanged (determined by qRT-PCR), expression of apoptosis inhibitory Bcl-2 was increased by anti-CD3, yet not affected by AMD (Appendix Fig S3E-G). With an MTS assay, proliferation was found to be increased with anti-CD3, which was not changed by AMD (Appendix Fig S3H). Moreover, upon incubation with the anti-CD3 antibody, CD8 T cells upregulated expression of Ifnc, Perforin, and Gzmb which again was not altered by AMD (Appendix Fig S3I-K).
Finally, by higher resolution tissue staining of TNC-high tumors, we noticed a punctate CXCL12 signal in the stroma, importantly colocalizing with TNC and CD8 TIL, altogether suggesting that the described chemoretention by TNC/CXCL12 might be relevant for tethering CD8 TIL in the tumor stroma (Appendix Fig S3L).

Inhibition of CXCR4 restores CD8 TIL tumor nest infiltration triggering tumor cell death
We investigated by flow cytometry whether TNC influenced CD8 TIL function in vivo through CXCL12, by blocking CXCR4 with AMD (for 2 weeks) in tumor-bearing mice in the 4-week model (Figs EV2 and EV4A-H). We observed a tendency toward more CD8 TIL in TNC-high tumors upon AMD treatment (P = 0.07). This was different to TNC-low (WT/shTNC) tumors with no change in CD8 TIL levels ( Fig EV4A). Whereas a tendency toward more CD4 and B cells and less macrophages, DC, NK cells, and neutrophils was seen in the TNC-high tumors upon CXCR4 inhibition, respectively, there was clearly no difference in TNC-low tumors suggesting that CXCR4 inhibition may have an impact on the immune cell infiltrate in the presence but not in the absence of TNC (Fig EV4A-G).
We considered that in the 4-week model, the full effect of AMD may not have developed yet; therefore, we used other AMD treatment protocols (Azab et al, 2009, Domanska et al, 2012Fig EV2). In the 7-week model, we observed a reduced tumor weight in AMDtreated mice, indicating that inhibition of CXCR4 reduced tumor growth ( Fig 5A). In accordance, we observed more apoptosis (quantification of the cl. casp3 staining signal) in the tumors upon CXCR4 inhibition (Fig 5B and C). Next, we addressed CD8 TIL tumor infiltration by tissue staining and observed more CD8 TIL in AMDtreated tumors (Fig 5D and E) that likely were active as Gzmb and IFNc levels were higher upon CXCR4 inhibition (Figs 5F and EV4H).
To rule out other factors, by qRT-PCR, we determined Cxcl12, Cxcr4, and Cxcr7 expression and did not see differences between PBS-and AMD-treated groups (Fig EV4I-K). To gain more insight into the underlying mechanisms, we compared gene expression in PBS with AMD-treated conditions (TNC-high) in the 4-week model, by RNA-seq analysis, and noticed profound differences between the groups. 1,287 genes were deregulated with 269 up-and 1,018 downregulated upon CXCR4 inhibition (Appendix Table S3). The large majority of genes in a signature associated with tumor cell apoptosis were upregulated upon AMD treatment ( Fig 5G). In contrast, signatures associated with angiogenesis, lymphocyte apoptosis, immune suppression, and IL10 production and response were downregulated upon CXCR4 inhibition in comparison to the PBS control (Figs 5H and EV5A-E). Also, CD8 T cell inhibitory Pdcd-1, Cd274, and Ctla4 were reduced upon AMD treatment (Figs 5H and EV5D). As TNC impacted expression of Cd274, we investigated gene expression with inhibitors as described ( Fig 3E) and found that Cd274 was reduced in the tumor cells with the TLR4 inhibitor Cli95 yet not with the other inhibitors ( Fig EV5F). Moreover, upon CXCR4 inhibition, several genes regulating lymphocyte differentiation, T cell activation, and IFNc production were deregulated mimicking gene expression of tumors with low TNC (WT/shTNC) (1,753 deregulated genes, 470 up-and 1,283 downregulated), altogether suggesting that immunity is largely regulated by CXCR4 signaling in these tumors in a TNCdependent manner (Fig EV5D and E, Appendix Table S4).
As we had previously seen that TNC promoted an M2 macrophage phenotype (Deligne et al, 2020), we wondered whether CXCR4 inhibition potentially impacted macrophages. By staining of TNC-high tumor tissue from the 7-week model for F4/80, we observed more activated macrophages upon AMD treatment compared to PBS. More importantly, largely present in control tumors, macrophages were even more abundant upon CXCR4 inhibition, now clearly infiltrating the tumor nests (Figs 5I and J, and EV5G). By tissue staining for CD206, we confirmed M2 macrophage infiltration of TNC-high tumors (Deligne et al, 2020) which we noticed to be abolished with AMD as no M2 macrophages were detectable anymore ( Fig EV5H).
Thus, we conclude that TNC plays an important role in regulating macrophages and CD8 TIL through CXCL12/CXCR4 signaling thereby impairing tumor cell killing and promoting tumor growth.

Impact of CXCR4 inhibition on lung metastasis
As CXCR4 signaling is known to promote cell migration and metastasis (Chatterjee et al, 2014), we investigated whether CXCR4 inhibition by AMD had an effect on lung metastasis in the spontaneous NeuNT and grafting NT193 models, respectively, as detailed in Fig 6A, D, and H. Due to variable latency and multi-focal tumor ◀ Figure 5. Inhibition of CXCR4 induces infiltration of CD8 leukocytes and macrophages and reduces tumor cell death and tumor growth (7-week model).

A
Growth of TNC-high (WT/shC) tumors upon cell engraftment and 5 weeks AMD treatment. N = 20 tumors per condition, ***P = 0.0007, unpaired t-test. The central band represents the median, the ends of the box the first and the third quartiles, and the whiskers reach from each quartile to the minimum or maximum. B, C Imaging and signal quantification upon staining for cl. casp3 and DAPI in PBS-and AMD-treated tumors, N = 5 tumors per condition, 2 sections per tumor, 5 random fields, **P = 0.0079, Mann-Whitney test. ▸ Figure 6. Impact of CXCR4 inhibition on lung metastasis.  To address a potential role of CXCR4 signaling on homing and tumor outgrowth in the NT193 model, we engrafted shC or shTNC cells via tail vein (iv) injection into a WT host followed by a daily AMD treatment for one week as previously described (Yang et al, 2019) (5-week model, Fig 6D). By a stereological approach, we assessed lung metastasis (Saupe et al, 2013) and observed that the majority of shC-engrafted mice got metastases, however with no impact of AMD. In contrast, engrafted shTNC cells caused less metastases which again was not affected by AMD, supporting a potential role of TNC in tumor cell homing to the lung as previously described for another model (Fig 6E; Oskarsson et al, 2011). We further found that the number and size of metastases was not different in the TNC-high condition (Appendix Fig S4C and D). Moreover, we noticed prominent round structures with flat nuclei at the rim (representing EC), reminiscent of blood vessel invasions (BVI) in the lungs of iv tumor cell-engrafted mice (similar to the orthotopic grafting condition (Sun et al, 2019)) (Appendix Fig S4E). Upon CXCR4 inhibition, BVI were more abundant in TNC-high conditions in expense of parenchymal metastasis (PM) (in comparison to PBS controls) (Fig 6F). The PM/BVI ratio was even further reduced upon AMD treatment in the TNC-low (WT/shTNC) condition suggesting a stronger effect of CXCR4 inhibition in the absence of TNC (Fig 6G). These results show that CXCR4 inhibition impacts BVI potentially delaying parenchymal metastasis.
Next, we used the 7-week model where mice with shC orthotopically engrafted tumor cells were treated with AMD for 5 weeks according to established protocols (Azab et al, 2009, Domanska et al, 2012; Fig 6H). We had observed a clear effect of AMD on tumor growth inhibition that may have an effect on lung metastasis; therefore, we examined the abundance of tumor cells in the lungs. As we could not detect tumor cells by tissue staining at this early time point, we used qRT-PCR for Neu which revealed expression in PBS controls that was lower upon CXCR4 inhibition, indicative of less tumor cells in the lung (Fig 6I). We considered that tumor cells invading the lung may elicit an immune response as seen ( Fig 2I); therefore, we used qRT-PCR to determine expression of Cd8, Ifnc, and Gzmb that was indeed seen in PBS conditions, however with lower levels upon CXCR4 inhibition potentially reflecting less tumor cells in the lungs (Fig 6J-L).
Altogether, these results suggest that reduced metastasis is likely due to a major impact of CXCR4 inhibition on tumor growth.
Combined abundance of CD8 TIL and high TNC and CXCL12 expression correlate with worsened breast cancer patient survival To address whether the results from the murine tumor models are relevant in human breast cancer, we stained a human breast cancer tissue microarray (Chen et al, 2010(Chen et al, , 2011 for TNC and CD8. We used a limited dilution approach for TNC that allowed to discriminate between tumors expressing high (TNC + ) and low TNC (TNC À ) levels, respectively ( Fig 7A, Appendix Fig S5A and B). TNC À and TNC + groups almost equally distributed between breast cancer subtypes except for Her2 + breast cancer specimens where the TNC + phenotype was more prominent (41%) than the TNC À phenotype (29%) ( Table 1). Moreover, CD8 TIL were more abundant in TNC À tumors which was pronounced in basal-like cases (Fig 7B, Appendix Fig  S5C). A closer investigation of CD8 TIL distribution showed that in both tumors (TNC À and TNC + ) CD8 TIL were more abundant in the stroma than in the nests, which applied to all three (Her2 + , basal, and luminal) subtypes (Appendix Fig S5D-G, Table 2). However, the ratio of nest-to-stroma CD8 TIL was lower in TNC + than in TNC À tumors ( Fig 7C). Also in TNC + tumors, nest infiltration of CD8 TIL was lower than in TNC À tumors supporting that TNC may play a role in retaining CD8 TIL in the stroma (Appendix Fig S5H and I).
Next, we asked whether TNC expression and CD8 TIL abundance potentially had an impact on patient survival. By Kaplan-Meier analysis of the TMA cohort, we first observed that high TNC levels (above the median) correlated with shorter metastasis-free survival (MFS) as previously seen in other patient cohorts (Fig 7D;Minn et al, 2005;Oskarsson et al, 2011). Importantly, higher CD8 TIL levels were seen in patients with longer MFS (Fig 7E). However, the survival benefit of high CD8 TIL seemed to be abolished when TNC levels were also high ( Fig 7F). In CD8 À tumors, TNC levels did not correlate with survival (Appendix Fig S5J). High or low abundance of CD8 TIL in the tumor nest also showed no difference in MFS (Appendix Fig S5K). In contrast, stromal CD8 TIL abundance and TNC levels, CD8 + /TNC À , correlated with longest MFS whereas CD8 + /TNC + and CD8-conditions correlated with shorter MFS (Fig 7G, Appendix Fig S5J and K). In the well-studied breast cancer patient cohort GSE 19783-GPL6580 representing all subtypes (Enerly et al, 2011), high TNC levels (above the median) and combined high TNC and high CD8 TIL correlated with shorter overall survival (Fig 7H and I,  Appendix Fig S5L). Moreover, a combined high expression of TNC together with either high CD8a (HR 1,56) or CD8b (HR 1,57), GZMB (HR 1,52), IFNc (HR 1,76), and Perforin (HR 1,75), respectively, correlated with a shorter overall survival and stronger power than TNC alone (HR 1,37) suggesting that stromal CD8 TIL tethering by TNC may be more important than CD8 TIL activation which does not seem to be linked to TNC (Appendix Fig S5N-U).
Next we stained the breast cancer TMA for CXCL12 and observed similar staining patterns as reported (Toullec et al, 2010;Lefort et al, 2017). We discriminated 4 staining intensities with no or low CXCL12 and high CXCL12 in tumor nests only, or combined high expression in tumor nest and stroma ( Fig 8A). As in the grafting model, CXCL12 showed a punctate expression in the stroma ( Fig 8A). As previously done, we combined intensity and abundance of CXCL12 expression in the so-called Hscore (Lefort et al, 2017) that we then used to assess whether CXCL12 expression is correlated with TNC levels. Indeed, TNC + tumors had a higher CXCL12 Hscore than TNC À tumors (Fig 8B). The CXCL12 Hscore was similar in the different tumor subtypes (Appendix Fig S6A,  Table 2). More detailed analysis revealed that the CXCL12 Hscore was higher when TNC and CD8 TIL were abundant (Fig 8C). Comparisons of variables in the TMA cohort (Chen et al, 2010(Chen et al, , 2011 were performed by using the chi-square test. * MFS (metastasis-free survival) was analyzed by log-rank test. Numbers of CD8 T leukocytes and expression of CXCL12 in the TMA cohort (Chen et al, 2010(Chen et al, , 2011 was determined on two separate sections per tumor using the median value. Comparison of variables by using the chi-square test. Arrows point at punctate staining in the stroma. B, C Histological score of CXCL12 staining in TNC + and TNC À tumors in combination with CD8 + (above median) and CD8 À (below the median). *P = 0.0206, **P = 0.0040, unpaired t-test. The central band represents the median, the ends of the box the first and the third quartiles, and the whiskers reach from each quartile to the minimum or maximum. Moreover, high TNC correlated with shorter overall survival when CXCL12 levels were high or the CXCL12 Hscore was above the median, whereas the TNC level was not prognostic in patients with tumors that had CXCL12 levels below the median (Fig 8D and E,  Appendix Fig S6B). Altogether, our results suggest that TNC regulates macrophages and CD8 TIL involving their positioning inside the stroma. By upregulating CXCL12 through TLR4 and binding to CXCL12, we showed that TNC generates stromal niches that attract and immobilize CD8 TIL. Subsequently, CD8 TIL may be impaired in reaching and killing the tumor cells, contributing to the well-known immune exclusion phenotype (Fig 9, Galon & Bruni, 2019). Here we showed that high CD8 TIL in the TNC stroma and combined high TNC and CXCL12 expression are parameters that define the immune exclusion phenotype which correlates with shorter survival of breast cancer patients that could have therapeutic value as seen in the preclinical models upon CXCR4 inhibition (Fig 9).

Discussion
Tumors are complex ecosystems where the matrix is an instrumental component not only providing architectural but also signaling cues to cells (Lu et al, 2012;Pickup et al, 2014). In particular, the matrix molecule TNC that is highly abundant in (breast) cancers plays an instructive role in many steps during cancer progression including regulation of tumor immunity (Jachetti et al, 2015;Midwood et al, 2016;Deligne et al, 2020;Spenl e et al, 2020).
We revealed a novel mechanism of CD8 T cell impairment by TNC, retaining CD8 TIL inside the tumor stroma. We showed that Figure 9. "TIL-matrix-release-and-reactivate" strategy through targeting TNC and CXCR4.
TNC contributes to an immune suppressive TME in breast cancer by retaining CD8 TIL in the stroma. TNC induces CXCL12 in a TLR4-dependent manner. The TNC protein binds CXCL12 generating a substratum that tethers CD8 TIL in the stroma thus preventing infiltration of CD8 TIL into the tumor nests and subsequent tumor cell killing. This mechanism could be relevant in human breast cancer as patients with tumors of high TNC expression show high CD8 TIL in the (TNC-rich) stroma which correlates with shorter MFS. (Lower left) Upon removal of TNC anti-tumor immunity is activated and characterized by lower CXCL12 expression, higher CD8 TIL nest infiltration, and a reduced metastasis rate as previously shown (Sun et al, 2019). In human breast cancer, low TNC and high CD8 TIL correlate with longest MFS. (Lower right) TNC impacts CD8 TIL through CXCL12/CXCR4. Inhibition of CXCR4 led to increased CD8 TIL infiltration, higher tumor cell apoptosis, and reduced tumor growth and subsequent lung metastasis. Also macrophages, loosing their M2 phenotype, infiltrate the tumor nests more numerously contributing to the observed activated CD8 TIL phenotype (e.g., less PD-L1, CTLA4). Targeting TNC and CXCL12 could direct a novel "TIL-matrix-release-and-reactivate" strategy to improve ICT in breast cancer.
TNC regulates CXCL12 expression through TLR4. We further showed that as in the murine model also in human breast cancer CXCL12 is predominantly expressed by the tumor cells. However, CXCL12 was also but less expressed by stromal cells presumably fibroblasts as seen in a murine pancreatic cancer model (Feig et al, 2013). Most importantly, we described that CXCL12 binds to TNC and turns the TNC matrix normally poorly adhesive for T cells (Hauzenberger et al, 1999;Huang et al, 2001), into a substratum that retains CD8 TIL in the TNC-rich stroma. This mechanism seems to be relevant in human breast cancer as high TNC and high stromal CD8 TIL correlate with shorter MFS. High CXCL12 levels, potentially acting as protective shield toward CD8 TIL (by an unknown mechanism) (Vianello et al, 2006;Guyon, 2014), may even further potentiate CD8 TIL exclusion from the tumor nest. We have shown that inhibition of CXCR4 releases the matrix-retained, now active CD8 TIL (as Ifnc and Gzmb expression was elevated) to infiltrate the tumor tissue thus increasing tumor cell death and reducing tumor growth and subsequently metastasis. This mechanism could be clinically relevant as combined high TNC and high CXCL12 correlated with shortest MFS.
CXCR4 signaling is active in many cell types (Döring et al, 2014); thus, it is possible that CXCR4 inhibition also affects, for example, other immune subtypes or the vasculature. In support by RNA-seq analysis, we observed a reduced angiogenesis-associated gene signature and an aberrant EC phenotype in BVI upon AMD treatment. CXCR4 inhibition also increased the number and infiltration of activated macrophages whereas tumor-promoting M2 macrophages disappeared. Moreover, CXCR4 inhibition reduced expression of immune suppressive IL10-production/IL-10 response-associated genes and CD8 T cell inhibitory CTLA4, PD-L1, and PD1, altogether presumably boosting CD8 TIL. However, our results propose that tumor cell-directed cytotoxicity may only be accomplished upon release of the CD8 TIL from the matrix where TNC binding CXCL12 could be crucial and exploitable for therapy. It is conceivable that targeting matrix potentiates ICT. This is supported by published work showing that blocking the TLR4/TNC interaction increased the anti-PD-L1 effects on metastasis inhibition (Deligne et al, 2020) and that combined treatment of anti-PD-L1 with AMD increased TIL infiltration reducing tumor growth in a pancreatic tumor model (Feig et al, 2013).
Previously, the so-called immunoscore, reflecting the abundance of CTL and memory T cells in the tumor cell nest or in the stroma, was identified as a better predictive tool for survival of colorectal carcinoma patients than the TNM classification (Galon et al, 2006;Galon et al, 2012). No information about a potential link to TNC or any other matrix was provided. It was also not known whether the immunoscore would also be a predictive tool in breast cancer. Here, we have shown that CD8 TIL immune exclusion in TNC-rich stroma correlates with worsened prognosis and identified novel parameters for stratification of breast cancer patients, irrespective of subtype, likely to benefit from ICT. Our results predict that patients with tumors of high levels of CD8 TIL, TNC, and CXCL12 (detectable at mRNA and protein tissue level), correlating with short MFS, may benefit from ICT in combination with CXCR4 inhibition.
Indeed, clinical trials targeting CXCR4 in cancer patients with different drugs alone or in combination with ICT are ongoing (NCT02104427, NCT01359657). A combination of the CXCR4 antagonist X4P-001 in combination with Pembrolizumab reactivating CD8 TIL (Emancipator, 2020) demonstrated a clinical response in patients with advanced melanoma (phase 1b trial, NCT02823405). Interestingly, the improved response was associated with an increased CD8 TIL density in the tumor center.
Altogether, our results propose that releasing CD8 TIL from the tumor matrix may be an important factor to empower ICT, where targeting matrix-regulated CXCR4 signaling is worth to be further explored. RNA gene expression results from our novel immune competent breast cancer model provides several molecular indicators such as TNC and CXCL12 to direct a future combined "TIL-matrix-release-and-reactivate" strategy for improving ICT in breast cancer.

Material and Methods
The experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. More detailed information can be found in the Appendix Supplementary Information section.

Immunohistochemical analysis of human breast tumors
Paraffin-embedded tissue microarrays (TMA with 219 specimens with invasive breast cancer, Hubei Cancer Hospital, China, Chen et al, 2010;Chen et al, 2011) were used for hematoxylin/eosin (H&E) and antibody staining for TNC (Abcam, ab2074, 1/500), CXCL12 (Abcam, ab9797,1/200) and CD8 (DAKO, C8/144B) (Tables 1 and  2). Informed consent was obtained from each patient and the study was approved by the hospital's Ethics Committee. Slides were independently investigated blinded by two investigators. Duplicates were counted and the average signal count per slide was used for statistical analysis. The expression of TNC was scored according to the staining intensity and the proportion of positive stromal cells in the whole slide per tissue sample (Yang et al, 2017). CXCL12 expression is expressed as histological score (Hscore) (Lefort et al, 2017). CD8stained cells were counted in five representative pictures (20 fields) according to placement in tumor nest or stroma. Numbers were determined as high or low using the median value as cutoff.

Patient survival analysis
The survival data for GSE19783-GPL6480 (Enerly et al, 2011) were obtained from and analyzed by a web application (http://genomics. jefferson.edu/proggene/index.php) as described elsewhere (Goswami & Nakshatri, 2013). Hazard ratio above or below 1 within a confidence interval of 95% (HR, 95% CI), and P values were determined using the log-rank test.

Experimental mice and study design
MMTV-NeuNT female mice in FVB/NCrl (WT or TNCKO background; more than 10 backcrosses into FVB (Sun et al, 2019)) were bred in parallel from heterozygous parents. Mice were housed and used according to the guidelines of INSERM and the ethical committee of Alsace, France (CREMEAS), and agreement number D 67-482-033 at the animal facilities of INSERM U682 and U1109 under pathogen-free conditions in cages providing disposable homes and nesting paper, together with food and water at discretion. Before begin of the experiment, female mice (8-10 weeks of age) were acclimatized in the procedure room for at least two weeks and were daily checked for good health and welfare throughout the whole experiment (including posture and activity). Experimental groups consisted of at least 5 mice per time point and per genotype. Cages with experimental mice were placed in random order, which was then used throughout the experiment. Upon sedation with 4% isoflurane (Isoflurinâ 1,000 mg/g, Axience), female WT or TNCKO FVB mice were grafted with 1 × 10 7 NT193 cells (shC, Tnc knockdown (sh1TNC, sh2TNC;Sun et al, 2019) in the surgically opened left fourth mammary gland that was surgically closed afterward. For most experiments, sh2TNC cells were used and defined as "shTNC" if not further specified. Tumor growth was assessed by measuring the tumor size, and tumor volume was determined using the following calculation V = (width) 2 × length/2. Tumor-bearing mice (or mice with iv engrafted tumor cells) were daily treated with AMD3100 (Sigma, A5602) at 5 mg/kg or PBS as control by peritumoral (pt) injection (4-and 7-week model), intraperitoneal injection (ip) (NeuNT model), or at 7.5 mg/kg ip injection (5-week model). Mice were euthanized by cervical dislocation. Tumors and lungs were processed for freezing in liquid nitrogen (protein and mRNA analysis), or embedding in OCT (Sakura Finetek) or paraffin (Leica, 39601006) for immunostaining. The experimental setup is explained in the text and in Figs 6 and EV1, and EV2.

Immunofluorescence (IF) and H&E staining
OCT-embedded tissue sections were incubated with a blocking solution before incubation with the indicated primary antibody, washed, incubated with the secondary antibodies (Appendix Table S6), and DAPI (Sigma, catalog number D9542) followed by embedding with FluorSave TM Reagent (Calbiochem, 345789). Images were analyzed with a Zeiss Axio Imager Z2 microscope with constant acquisition setting (microscope, magnification, light intensity, exposure time) and quantified by the ImageJ (National Institutes of Health) software using a constant threshold.

Real-time quantitative-PCR analysis
Total RNA was extracted from frozen tumors, lungs, and cultured cells with TRIzol (Invitrogen, 12044977). cDNAs (synthesized using random primers and Moloney murine leukemia virus reverse transcriptase (MultiScribe, Applied Biosystems, 10117254)) were used for qRT-PCR in an Mx3005P Real-Time PCR System (Thermo Fisher Scientific) with Sybr Green Master mix (Thermo Fisher Scientific, 4344463) or Fast Taqman mix (Thermo Fisher Scientific, 4444557).
Expression of mouse Gapdh mRNA (Life Technology, 433764T) was used as endogenous control in the comparative cycle threshold method (2-DDC t ) with the listed primers (Appendix Table S7).

CD8 T leukocyte isolation and attraction and retention assay
CD8 T leukocytes were sorted by using the murine CD8a + T Cell Isolation Kit (Miltenyi Biotec, 130-104-075) from spleens of FVB mice as described (Deligne et al, 2020). The lower surface of polycarbonate membrane transwells (Costar, 5 µm-pore size, 3421) was coated with matrix molecules, washed with PBS, and blocked with 1% BSA overnight at 4°C. The lower chambers of the transwells were filled with TexMACS medium (Miltenyi Biotec, containing CXCL12 (1 µg/ml, R&D Systems, 460-SD-050) or conditioned medium (CM) from shC or shTNC cells. AMD3100 (Sigma, A5602) at 5 µg/ml was added to the CD8 T cells for 1 h at 37°C before seeding cells in the upper chamber for 5 h at 37°C followed by collection of the medium in the lower chamber for cell counting by flow cytometry (Appendix Table S5). Cells on the surface of the lower side of the insert were fixed with 4% PFA (Santa Cruz, sc-281692), stained with DAPI and imaged. Pictures were analyzed with the ImageJ software.
Negative staining, transmission electron microscopy, and CXCL12 binding assay The interaction of murine TNC (Spenl e et al, 2020) with CXCL12 (R&D Systems, 460-SD-050) was visualized by negative staining and transmission electron microscopy as described (Bober et al, 2010;Spenl e et al, 2020). For inhibition experiments, TNC samples were pre-incubated with the indicated concentrations of heparin dp10 (AMS Biotechnology, AMS.HO10). Specimens were examined in a Philips/FEI CM 100 TWIN transmission electron microscope operated at 60 kV accelerating voltage. Images were recorded with a side-mounted Olympus Veleta camera and the ITEM acquisitions software. Binding of CXCL12 to TNC was determined by counting the number of colloidal gold particles along the length of the TNC monomer. Numbers of molecules from 500 randomly picked distinct TNC molecules were determined. Competition was done with increasing concentration of CXCL12 in TBS (50 mM Tris, 150 mM Nacl, pH 7.9).

Gene expression analysis
The sequencing library was prepared from mRNA derived from tumors, shC, and shTNC cells, respectively, with the Ion Total RNA-Seq Kit v2 (Thermo Fisher Scientific, 4475936). Sequencing was performed on an Ion Proton sequencer with the Ion PI TM Hi-Q TM Sequencing 200 Kit (Thermo Fisher Scientific A26433). The transcriptome data were processed by the RNASeqAnalysis plugin from the Torrent Suite Software 5.06 (Thermo Fisher Scientific). RNA from MMTV-NeuNT-WT and TNCKO mammary tumors were used for the Affymetrix Microarray experiments, performed at the IGBMC Sequencing Core Facility (Illkirch, France).

Statistical analysis
Gaussian distribution was tested by the D'Agostino-Pearson normality test. When data followed a Gaussian distribution, statistical differences were analyzed by unpaired t-test (with Welch's correction in case of unequal variance) or ANOVA one-way with Tukey posttest. Otherwise, the Mann-Whitney test or a non-parametric ANOVA followed by Dunn's post-test were used to verify significance of the observed differences. All statistical analyses were performed using the GraphPad Prism software. All results were compared to the respective control by the indicated statistical test, and only P < 0.05 are mentioned in the figure legends. Mean AE SEM. P < 0.05 were considered as statistically significant, *P < 0.05; **P < 0.01; ***P < 0.001.
Expanded View for this article is available online. The paper explained Problem Immune checkpoint therapy (ICT), where CD8 tumor infiltrating T lymphocytes (TIL) get reactivated, is a promising anti-cancer treatment approach, yet with low response rates in breast cancer. Immune exclusion by the matrix, in particular by the highly abundant matrix molecule tenascin-C, forming spatial niches separating tumor cell nests, may generate barriers for TIL.

Results
We combine evidence from clinical material and experimental models to reveal that through induction (involving TLR4) and binding of CXCL12, tenascin-C impairs macrophages and CD8 TIL. We report that a tenascin-C/CXCL12 substratum retains CD8 TIL in the matrix impairing tumor cell killing. This molecular axis is important as inhibition of CXCR4 (CXCL12 receptor) restores anti-tumor immunity by activating and releasing CD8 TIL from the matrix, promoting CD8 TIL tumor infiltration and tumor cell apoptosis, and causing reduced tumor growth and subsequent lung metastasis. Notably, in human breast cancer patients, we document an immune exclusion phenotype correlating with shorter survival characterized by CD8 TIL retention in tenascin-C-rich stroma, together with high CXCL12.

Impact
This study provides novel insight into how matrix retains CD8 TIL contributing to the immune exclusion phenotype, information that could be exploited to formulate a novel "TIL-matrix-release-andreactivate" strategy. Moreover, our results provide rationale for targeting matrix plus CXCR4 for improving ICT in particular in tumors with high tenascin-C, high CXCL12, and high CD8 TIL as this combination correlates with short breast cancer patient survival.