Collagen XVIII promotes breast cancer through EGFR/ErbB signaling and its ablation improves the efficacy of ErbB-targeting inhibitors

The tumor extracellular matrix (ECM) is a critical regulator of cancer progression and metastasis, significantly affecting the treatment response. Expression of collagen XVIII (ColXVIII), a ubiquitous component of basement membranes, is induced in many solid tumors, but its involvement in tumorigenesis has remained elusive. We show here that ColXVIII is markedly upregulated in human breast cancer (BC) cells and is closely associated with a poor prognosis in high-grade BC, especially in human epidermal growth factor receptor 2 (HER2)-positive and basal/triple-negative cases. We identified a novel mechanism of action for ColXVIII as a modulator of epidermal growth factor receptor (EGFR/ErbB) signaling and show that it forms a complex with EGFR, HER2 and α6 integrin to promote cancer cell proliferation in a pathway involving its N-terminal portion and the MAPK/ERK1/2 and PI3K/Akt cascades. In vivo studies with Col18a1 mouse models crossed with the MMTV-PyMT mammary carcinogenesis model showed that the short ColXVIII isoform promotes BC growth and metastasis in a tumor cell-autonomous manner. Moreover, the number of mammary cancer stem cells was significantly reduced in both mouse and human cell models upon ColXVIII inhibition. Finally, ablation of ColXVIII in human BC cells and the MMTV-PyMT model substantially improved the efficacy of certain EGFR/ERbB-targeting therapies, even abolishing resistance to EGFR/ErbB inhibitors in some cell lines. In summary, a new function is revealed for ColXVIII in sustaining the stemness properties of BC cells, and tumor progression and metastasis through EGFR/ErbB signaling, suggesting that targeting ColXVIII in the tumor milieu may have significant therapeutic potential. One Sentence Summary Collagen XVIII is upregulated in breast cancer and promotes mammary carcinogenesis through EGFR/ErbB signaling and by sustaining cancer stem cells, so that its targeting improves the efficacy of ErbB-targeted therapies.


Abstract:
The tumor extracellular matrix (ECM) is a critical regulator of cancer progression and metastasis, significantly affecting the treatment response. Expression of collagen XVIII (ColXVIII), a ubiquitous component of basement membranes, is induced in many solid tumors, but its involvement in tumorigenesis has remained elusive. We show here that ColXVIII is markedly upregulated in human breast cancer (BC) cells and is closely associated with a poor prognosis in high-grade BC, especially in human epidermal growth factor receptor 2 (HER2)-positive and basal/triple-negative cases. We identified a novel mechanism of action for ColXVIII as a modulator of epidermal growth factor receptor (EGFR/ErbB) signaling and show that it forms a complex with EGFR, HER2 and 6 integrin to promote cancer cell proliferation in a pathway involving its N-terminal portion and the MAPK/ERK1/2 and PI3K/Akt cascades. In vivo studies with Col18a1 mouse models crossed with the MMTV-PyMT mammary carcinogenesis model showed that the short ColXVIII isoform promotes BC growth and metastasis in a tumor cellautonomous manner. Moreover, the number of mammary cancer stem cells was significantly reduced in both mouse and human cell models upon ColXVIII inhibition. Finally, ablation of ColXVIII in human BC cells and the MMTV-PyMT model substantially improved the efficacy of certain EGFR/ERbB-targeting therapies, even abolishing resistance to EGFR/ErbB inhibitors in some cell lines. In summary, a new function is revealed for ColXVIII in sustaining the stemness properties of BC cells, and tumor progression and metastasis through EGFR/ErbB signaling, suggesting that targeting ColXVIII in the tumor milieu may have significant therapeutic potential.

INTRODUCTION
Breast cancer (BC) is the most common cancer among women, with over two million new cases diagnosed in 2020, and accounts for 25% of all female cancers (1). Treatment options depend on the type and course of the disease, and on hormone and human growth factor receptor 2 (HER2) status, mutations, proliferation index and differentiation score (2). Hence, BC patients are treated with different combinations of surgery, radiation, chemotherapy and endocrine therapy, as well as with targeted immuno-or small molecule therapies. Despite significant advances in BC care, over 0.6 million women die of BC annually, it accounts for 7% of all female cancer deaths, and the 5-year recurrence rate for all BC cases is around 10% (1,3).
A major challenge in cancer treatment is intrinsic or acquired drug resistance, which is responsible for most of the relapses that occur after an initially favorable response to treatment (4,5). For example, approximately 70% of advanced BCs overexpressing the human epidermal growth factor receptor 2 (HER2) develop resistance to trastuzumab, a monoclonal antibody (mAB) targeting HER2, and progress to metastatic disease. Many patients also become resistant to lapatinib, a small-molecule tyrosine kinase inhibitor of HER2, and to epidermal growth factor receptor 1 (EGFR) (6, 7). In addition, residual disease in the breast or lymph nodes after neoadjuvant chemotherapy may carry a high risk of recurrence in patients who present with earlystage triple-negative BC (TNBC), and no targeted molecular therapies are available for this subtype (8,9).
While genetic alterations in cells predispose to, initiate and drive malignancy, cancer progression is enabled by a dysregulated tumor microenvironment (TME), comprising different types of stromal cells together with the extracellular matrix (ECM) (10,11). Both tumor and stromal cells actively produce ECM proteins and ECM-modifying enzymes to remodel the TME, which then promotes the growth of cancer cells and their invasion into the surrounding tissue and beyond (12,13). Moreover, biological and mechanical cues from the ECM support the acquisition of cancer stem cell (CSC) properties by somatic tumor cells, thus favoring continuous tumor growth and the development of drug resistances and eventually disease relapse (14,15). Our recent analysis has shown that the expression of a variety of ECM components in cancers is precisely regulated by specific oncogenic drivers and downstream transcription factors and correlates with the patient's prognosis (16). This study and a number of others, including our recent works (17)(18)(19), highlight the utility of ECM molecules as diagnostic biomarkers and in disease follow-up and unveil new therapeutic possibilities for inhibiting cancer progression and metastasis and dismantling resistance to cancer therapies by targeting the ECM (12)(13)(14)(15).
Collagen XVIII (ColXVIII) is a ubiquitous component of epithelial and endothelial basement membranes (BM) (20). It is a structurally complex and functionally versatile molecule with roles in the eye, nervous system and adipose tissue, for example. ColXVIII exists in three isoforms, short, medium and long, which differ in their N-terminal non-collagenous (NC) domain structure, tissue specificity and functions. All three isoforms contain an endostatin domain, a widely studied BM-derived anti-angiogenic molecule (20)(21)(22), in their C-terminal NC1 portion and a laminin-G/thrombospondin-1-like (TSP-1) domain in their N-terminal NC11 portion. The long ColXVIII isoform has two additional domains in the N-terminus, a mucin-like domain (MUCL-C18) and a Wnt-binding Frizzled-like domain (FZ-C18) which is spliced out of the medium ColXVIII isoform (20,23). In several neoplasms, including lung, prostate, skin and gastric cancers, both ColXVIII overexpression in tumor tissues and high endostatin levels in the patients' sera have been associated with disease progression and poor prognosis rather than with tumor repression by endostatin (22). However, the mechanism by which ColXVIII promotes tumor growth and progression are still unclear.
We set out here to investigate the role of ColXVIII in BC and its mechanisms of action using genetic mouse tumor models and human BC cell models, and to assess the translational value of ColXVIII by correlating its expression with the clinicopathological features of human BC and by conducting drug tests in ColXVIII-deficient cell and mouse models. Our studies revealed a previously unidentified mechanism for ColXVIII in the regulation of EGFR/ErbB signaling in BC that leads to tumor promotion and demonstrated significant upregulation of ColXVIII expression in high-grade BC, which was associated with a poor clinical outcome. In addition, our preclinical assays showed that ColXVIII targeting has a promising therapeutic potential in the treatment of BC.

ColXVIII expression in human breast cancers
Immunohistochemical (IHC) analysis performed on 116 human BC specimens (Table S1) with a monoclonal custom-made ColXVIII antibody (DB144-N2) (Table S2) showed that the ColXVIII signal is prominent in the BMs of blood vessels, mammary ducts and lobules in normal breast tissue adjacent to tumor regions (Fig. 1A, Fig. S1). In addition, ColXVIII can be detected in the thin BM surrounding the adipocytes. In ductal carcinoma in situ (DCIS) the ducts filled with tumor cells are usually surrounded by an intact ColXVIII-positive BM/myoepithelial cell layer, albeit the ColXVIII signal may be discontinuous or even completely lacking at some tumor borders ( Fig.   1B-C, Fig. S1). Intriguingly, a cytoplasmic ColXVIII staining ranging from weak to moderate can frequently be detected in tumor cells in DCIS ( Fig. 1B-C, Fig. S1). In invasive ductal carcinomas (IDC) of various grades ColXVIII expression is commonly seen in the cytoplasm of tumor cells, though the staining intensity varies considerably from weak to strong between samples and tumor regions, being more intense in high-grade tumors ( Fig. 1D-F, Fig. S1). In IDCs, ColXVIII expression is either fragmented or completely lost from epithelial BM/myoepithelium around the tumors ( Fig. 1D-F). The ColXVIII signal is prominent in the vascular BMs of all DCIS and IDC samples ( Fig. 1A-I, Fig. S1), and occasionally also in other stromal cells, including myofibroblasts ( Fig. S1). The authenticity of the ColXVIII staining patterns was verified here in several tumor samples with a polyclonal custom-made human ColXVIII antibody (QH48.18) (Fig. S1).
Interesting differences in ColXVIII expression were observed when BC samples were classified according to their molecular subtypes. The cytoplasmic ColXVIII signal is usually strong or moderate in HER2-positive and basal/TNBC cases, and the staining intensity in the BM/myoepithelial cell layer around the tumors varies from negative to strong (Fig. 1G, H). In one case we observed that ColXVIII was markedly upregulated in the cytoplasm of invasive tumor cells but absent from the BM/myoepithelium of the invasive tumor area, whereas the cytoplasmic ColXVIII signal at the DCIS site was weak in spite of the fact that both the BM/myoepithelium and the endothelium showed strong ColXVIII staining (Fig. 1G). The ColXVIII signals in the samples of the luminal A subtype were variable both in the cytoplasm and around the tumor nests, ranging from negligible to prominent staining (Fig. 1I).
Analyses of open databases that include survival data on BC patients to estimate the association between ColXVIII mRNA levels and patient survival (24) showed that high ColXVIII expression was significantly associated with poor prognosis in patients with high-grade BCs, where the hazard ratio exceeded 1.5, but not in unclassified patients or in those with low-grade tumors ( Fig. 1J-L, Fig. S2). When the patients with high-risk grade 3 cancers were further categorized into major molecular subtypes, high ColXVIII was more significantly associated with poor survival in the HER2, basal/TNBC and luminal B subgroups than in the luminal A subgroup ( Fig.1 M-O, Fig. S3), indicating that ColXVIII upregulation correlates with recurrence in the case of high-grade BCs.
An in-house indirect ELISA assay was performed to quantify the plasma levels of Nterminal ColXVIII fragments in a small number of healthy controls and BC patients. Most of the patient samples showed higher plasma ColXVIII levels than the healthy controls, and the average plasma ColXVIII concentration in HER2, and particularly in the basal/TNBC subtypes, but not in luminal A cases, was significantly higher than in the controls (Fig. S4A). When the same data were grouped according to metastatic status, plasma ColXVIII levels were significantly higher in lymph node-positive than in node-negative luminal A cases, suggesting that ColXVIII could predict tumor metastasis in this BC subtype (Fig. S4B). On the other hand, ColXVIII levels in the plasma of HER2 and basal/TNBC-type patients were high in both node-negative and node-positive cases, implying that ColXVIII could be used as an early diagnostic marker for these subtypes, even before metastasis (Fig. S4B).

ColXVIII promotes tumor cell proliferation through its N-terminal TSP-1 domain
ColXVIII signals in cultured human BC cell lines were prominent in the HER2-amplified JIMT-1 cells, the triple-negative MDA-MB-231 and HS578T cells and the MCF7 cells representing the luminal A subtype ( Fig. 2A-B). In the other BC cell lines tested, including T47D (luminal A), BT474 (luminal B) and SKBR3 (HER2), and in the non-cancerous breast epithelial cell line MCF10A, ColXVIII was also present but showed some variation ( Fig. 2A-B).
To investigate the role of ColXVIII in breast carcinogenesis, we inhibited its expression in human BC cell lines with RNA interference. A mixture of two small interfering RNAs (siRNA), one targeting the TSP-1 region at the N-terminus of ColXVIII and the other targeting the Cterminal endostatin, was used to achieve an efficient knockdown (KD) of ColXVIII in selected BC cell lines and in MCF10A cells. Typically, a 70-90% inhibition in mRNA synthesis was achieved by this approach as compared with the scrambled vector transfected control cells, leading to a reduced amount of ColXVIII protein ( Fig. 2B-C). Except for the MDA-MB-231 cells, ColXVIII KD significantly reduced the proliferation of human BC cells, ranging from an approximately 20% reduction in the BT474 cells to an almost 70% reduction in the SKBR3 cells during a 96-hour follow-up period (Fig. 2D).
To confirm that the reduction in cell proliferation was caused by ColXVIII KD, and to terminal TSP-1 and NC11 fragments were less impressive, but nevertheless statistically significant, whereas endostatin did not affect the KD cells (Fig. 2F). The non-cancerous MCF10A KD cells also regained their proliferation activity upon the addition of an exogenous NC11 fragment (Fig. 2G). Hence, the results of these in vitro experiments suggest that specifically the N-terminal portion of ColXVIII, and even the TSP-1 domain alone, can constitute an ECM signal that activates BC and mammary epithelial cell proliferation.

ColXVIII supports mammary carcinogenesis in the MMTV-PyMT mouse model
Consistent with the results of human tissue analyses, ColXVIII signals in healthy mouse mammary tissue can be detected around adipocytes, in vascular BMs and in the mammary duct BMs, where it is located next to the alpha smooth muscle actin (αSMA), a marker of myoepithelial cells in mammary ducts and smooth muscle cells in blood vessels (Fig. 3A). As expected, the mammary glands (MG) and adipose tissue of healthy Col18a1 -/mice are not reactive with the anti-ColXVIII antibody, whereas the αSMA signal can be observed in the ducts and vessels (Fig. 3A). In MMTV-PyMT (PyMT) mouse mammary tumor tissues the ColXVIII signal is clearly increased and is located around tumor nests and the vascular BMs (Fig. 3B). In addition, a prominent cytoplasmic ColXVIII staining of tumor cells is evident in the late-stage PyMT tumors (Fig. 3B).

ColXVIII has an autocrine stimulatory function in mammary carcinoma cells
Reciprocal orthotopic allograft transplantation experiments between the WT and Col18a1 -/genotypes were performed to determine whether the tumorigenic functions of ColXVIII are tumor cell-autonomous or microenvironmental. Both the WT and Col18a1 -/females that received WT-PyMT tumor cells developed palpable tumors by week 7 after implantation (Fig. 4A), but these grew faster in the WT-FVB hosts, reaching an average volume of 550 mm 3 , the humane end point size limit, by week 10. In the Col18a1 -/--FVB hosts the tumors were on average half the size of those in the WT-FVB hosts by the same week, and although the difference was not statistically significant, this observation suggested that host-derived ColXVIII can also contribute to the regulation of tumor growth. As surmised from the previous experiments ( Fig. 2 and Fig. 3), cells isolated from the 18 -/--PyMT tumors grew much more slowly and developed palpable tumors 5 weeks later, at week 12, in both hosts. Interestingly, the 18 -/--PyMT tumors also grew faster in the WT mice than in the Col18a1 -/mice, reaching a size of around 400 mm 3 in 17 weeks whereas those in the Col18a1 -/hosts did not exceed 200 mm 3 in volume within the same time frame (Fig.   4A). This again suggests a contribution of stromal ColXVIII to tumor growth. Ki67 immunostaining of the allografts showed that in both hosts the implanted 18 -/--PyMT tumor cells proliferated significantly less than the WT-PyMT cells (Fig. 4B-C). On the other hand, there was no statistical difference in the Ki67 scores for either the 18 -/--PyMT cells or the WT-PyMT cells between the hosts (Fig. 4C).
Immunostaining of the WT-PyMT tumor allografts showed prominent ColXVIII signals at the borders of the tumor nests in the WT hosts but somewhat weaker signals at these sites in the

ColXVIII supports breast cancer stem cells
High ColXVIII expression has been observed in human mammary stem and progenitor cell populations (25) as also in various tissue-specific stem cell niches (22). Using fluorescenceactivated cell sorting (FACS), we found that the frequency of mouse mammary CSCs, defined as CD49f high (integrin 6 high ), CD29 high (integrin 1 high ), hyaluronan receptor CD44 + and heat stable antigen CD24 + cell populations (26), was reduced almost by 90% in the 18 -/--PyMT tumors by comparison with the WT-PyMT tumors (Fig. 5A). In addition, immunostaining of tumor tissues showed a notable reduction in the integrin 1 signal in the 18 -/--PyMT tumors relative to the controls, so that the number of cells that were double positive for integrins 1 and 6 was very low in the knockout tumors (Fig. 5B). Cytokeratin-5 (CK5) is a marker of mature myoepithelial cells when it is co-expressed with αSMA, but discrete CK5 + αSMAcells are regarded as CSCs (27).
Single positive CK5 + cells were abundant inside the WT-PyMT tumor nests, whereas they were approximately 40% less frequent in the 18 -/--PyMT tumors (Fig. 5C-D), further confirming that there are less CSCs in knockout tumors. Moreover, immunostaining of allograft tumors showed more CK5 + and αSMA -CSCs in WT-PyMT tumors grown in both WT and Col18a1 -/hosts, whereas 18 -/--PyMT tumors showed more CK5 + and αSMA + double-positive cells, as is indicative of reduced CSC characteristics and myoepithelial differentiation in these cells (Fig. 5E).
We then analyzed the CD44 + and CD24 low/-CSC populations (28) in the WT and KD MDA-MB-231 human BC cells using FACS. A significant reduction in the frequency of this CSC population was observed in the siRNA-based ColXVIII KD when compared with the scrambledtreated MDA-MB-231 cells (Fig. 5F), resulting in a remarkable decrease in the mean fluorescence intensity levels of CD49f (Fig. 5G). Moreover, the common stem cell-related transcription factors NANOG, SNAI1, SNAI2 (SLUG) and SOX2 (29) were downregulated in the MDA-MB-231 KD cells (Fig. S7A), as also in the MCF7 cells with ColXVIII KD (Fig. S7B). Interestingly, ColXVIII expression was significantly higher in the CSC-enriched (CD44 + CD24 low/-) subpopulation of MCF7 cells than in the non-CSC subpopulation (CD44 + CD24 high ) (Fig. S7C). When these subpopulations were assessed in a colony-forming assay, the CSC-enriched MCF7 population formed irregular, heterogenous colonies whereas the non-CSC population formed well-polarized round colonies of approximately equal size (Fig. S7D). Moreover, ColXVIII KD in the nontumorigenic MCF10A cells reduced the frequency of the CD44 + CD24 low/population by more than half and led to a significant decrease in the stem cell marker mRNA levels ( Fig. S7E-F). Our data thus show that high ColXVIII expression is associated with cellular stemness, and its ablation leads to a significant decrease in the number of tumor-promoting mammary CSC populations.

ColXVIII is co-expressed with EGFR and HER2 in human breast cancer cells
Our open data analyses have indicated that high ColXVIII expression is associated with poor survival in HER2-amplified and basal/TNBC types of human BC (Fig. 1, Fig. S2, Fig. S3), prompting further investigations into the role of ColXVIII in the EGFR/ErbB signaling pathway.
Initial IHC analysis of HER2-type human BC specimens (N=21, Table S1) showed that ColXVIII, EGFR and HER2 are expressed in the same tumor areas that have a high number of Ki67-positive cells (Fig. 6A-B). We then analyzed the expression of ColXVIII and EGFR in a larger BC tissue microarray (TMA) (N=95, Table S1) which had previously been scored for HER2 and Ki67 expression and the nuclear grade. This analysis confirmed that, especially in the HER2 subgroup, strong or moderate cytoplasmic ColXVIII expression in tumor cells correlated with HER2 amplification, and most of these samples also showed strong or moderate EGFR expression (Fig.   6C). In addition, high ColXVIII expression was associated with tumor grade 3 in all the HER2 cases. Correspondingly, a high or moderate cytoplasmic ColXVIII signal was associated with Ki67 expression in almost 60% of the luminal B and TNBC cases and with tumor grades 2 or 3 in 35%, whereas ColXVIII and EGFR were abundantly co-expressed only in a few luminal B specimens in which the tumor cells were also HER2-positive (Fig. 6C). ColXVIII and EGFR signals were found juxtaposed or overlapping in normal human and mouse mammary ducts, with ColXVIII showing a typical BM staining and EGFR signals being localized in the myoepithelial layer ( Fig.   6D-E).

ColXVIII forms a complex with EGFR and integrin 6 and regulates EGFR/ErbB signaling
Immunofluorescence showed ColXVIII and EGFR co-expression in the basal type MDA-MB-231 (Fig. 7A) and in the HER2-amplified JIMT-1 (Fig. S8A) human BC cells. As cooperation between EGFR and ECM receptor integrins is known to promote the progression and aggressiveness of solid tumors (15,30), the expression of 6 and 1 integrins, the key integrins in the mammary epithelium, was also analyzed. These integrins are also determinants of breast CSCs, the incidence of which was found to be reduced upon ColXVIII ablation (Fig. 5, Fig. S7). Immunostainings revealed that both the 6 and 1 subunits are expressed with ColXVIII and EGFR in MDA-MB-231 cells ( Fig. 7B; Fig. S8B). Proximity ligation assays demonstrated potential interactions between ColXVIII and EGFR in MDA-MB-231 and JIMT-1 cells, as well as between ColXVIII and 6 integrin in MDA-MB-231 cells (Fig.   7C, Fig. S8C). Consistently with this, co-immunoprecipitation assays showed that EGFR and integrin 6 antibodies pull down ColXVIII in MDA-MB-231 and JIMT-1 lysates (Fig. 7D, Fig.   S8D-E), and HER2 in JIMT-1 lysates (Fig. 7E). EGFR antibodies also immunoprecipitated HER2 in SKBR3 lysates, and the ColXVIII antibody pulled down both EGFR and HER2 in these cells (Fig. S8F) and EGFR in MCF10A cells (Fig. S8G). Neither EGFR nor ColXVIII antibodies pulled the integrin 1 subunit down in MDA-MB-231 cells (Fig. S8H).
To study further the involvement of ColXVIII in EGFR/ErbB signaling in BC and to better understand its mechanism of action at the cellular level, the effects of ColXVIII KD on EGFR/ErbBs and downstream signaling pathways, including the mitogen-activated protein kinase (Ras/Raf/MEK/ERK1/2) and phosphatidylinositol-3-kinase (PI3K/Akt) pathways, were assessed in several human BC cell lines. Western blot analyses showed that EGFR and HER2 phosphorylations were decreased in the SKBR3 and MCF10A cells upon ColXVIII KD relative to the scrambled cells, and that EGFR phosphorylation was reduced in the HER2-deficient MDA-MB-231 cell line, albeit somewhat less than in the other cell lines tested (Fig. 7F-G). Moreover, pERK and pAKT levels were decreased in SKBR3 cells (Fig. 7F-G

Therapeutic potential of ColXVIII
In view of these results, we finally focused our interest on the potential effects of ColXVIII inhibition on drug responses when combined with the tyrosine kinase inhibitor lapatinib or with humanized mABs against HER2 (trastuzumab) and EGFR (panitumumab). Lapatinib treatment almost completely blocked the proliferation of the HER2-type SKBR3 cells, and thus ColXVIII KD, which by itself resulted in a roughly 30% reduction in cell proliferation in five days, did not yield any additional effect (Fig. 8A, Fig. 2E). SKBR3 cells responded well to HER2-targeting trastuzumab, however, and this alone led to an approximately 25% reduction in cell proliferation in five days of culture. Interestingly, simultaneous administration of ColXVIII-targeting siRNAs and trastuzumab had a synergistic effect on SKBR3 cell proliferation, leading to an over 60% reduction in cell numbers in the course of the experiment as compared with untreated scrambled cells (Fig. S9A). Moreover, EGFR-targeting panitumumab and ColXVIII siRNAs in combination inhibited the proliferation of SKBR3 cells more rapidly and efficiently than did either of these treatments alone (Fig.S9B).
HER2-amplifed JIMT-1 cells are resistant to drugs that directly target ErbB receptors, due to several co-existing drug resistance mechanisms, including mutations in PI3KCA that activate the PI3K/AKT pathway (31). We noticed that whereas neither lapatinib nor ColXVIII KD alone affected the proliferation of this cell line in the early growth phase but led to growth inhibition in the later stages, their combined effect was extremely rapid and efficient and almost completely abolished the proliferation of JIMT-1 cells (Fig. 8B). Lapatinib, panitumumab and trastuzumab treatments did not affect the proliferation of MDA-MB-231 cells because this cell line is HER2negative and has mutations in the downstream effectors KRAS and BRAF that keep the cells in a proliferative state (32) (Fig. 8C, Fig. S9C-D). The EGFR-targeting panitumumab, however, did result in a significant growth restriction in MDA-MB-231 KD cells, although the effect of ColXVIII inhibition was less impressive in this cell line than in the SKBR3 and JIMT-1 cells (Fig.   S9C-D). Besides these three cell lines, the HER2-positive luminal B-type BT474 cell line that has a PIK3CA mutation (33) and is thus resistant to ErbB-targeting drugs was also included in our tests. The proliferation of BT474 cells was not affected at all by trastuzumab, and only marginally by lapatinib. Depletion of ColXVIII KD alone reduced the proliferation of BT474 cells by 25-30% in five days and sensitized these cells to lapatinib (Fig. S9E-F).
Besides reducing cancer cell proliferation, ColXVIII KD slowed down the migration of SKBR3 cells significantly but, as in the proliferation assay, it did not exhibit any additional inhibitory effect on wound closure when combined with lapatinib (Fig. S9G). In MDA-MB-231 cells the inhibitory effect of ColXVIII KD was more evident in cell migration than in cell proliferation (Fig. S9H, Fig. 8C), whereas lapatinib produced only a marginal effect, as was expected due to mutations in signal mediators (Fig. S9H). The combined use of lapatinib and ColXVIII siRNAs, but not single treatments with these reagents, resulted in a significant reduction of JIMT-1 cell migration (Fig. S9I).
A preclinical in vivo experiment with lapatinib confirmed that ColXVIII knockout adds a significant inhibitory effect upon mammary carcinogenesis in the MMTV-PyMT mouse model. In the vehicle-treated 18 -/--PyMT mice the total tumor burden was approximately 30% smaller than in the vehicle-treated WT-PyMT mice (Fig. 8D). The tumor burden was further reduced in the 18 -/--PyMT mice treated with lapatinib, by approximately 35% and 54% compared with the vehicle-  (Fig. 8E-F). Correspondingly, the tumors in the lapatinibtreated 18 -/--PyMT mice were considerably smaller, and those in some MGs of the 18 -/--PyMT mice receiving a high dose of lapatinib had been almost completely eradicated, so that the fat pads contained fairly normal-looking ductal structures (Fig. 8E). The number of intratumoral CK5 + progenitor cells in the 18 -/--PyMT tumors was initially significantly smaller than in the WT-PyMT tumors (Fig. 8E,G; Fig. 5C) and lapatinib treatment did not affect these cell counts in either genotype in the current model (Fig. 8G). In summary, our preclinical experiments demonstrate the importance of ColXVIII for BC cells functions and show that the inhibition of its action in tumor cells has important therapeutic potential.

DISCUSSION
This study shows that ColXVIII expression is high in human and mouse BC and supports tumor cell proliferation in an autocrine manner through a previously unreported mechanism involving EGFR/ErbB signaling. Moreover, it presents evidence that ColXVIII can have significant translational value as a novel biomarker and a potential therapeutic target in BC.
More specifically, our key findings are that 1) ColXVIII expression is induced in human BC cells; 2) high ColXVIII expression is associated with high-grade tumors and reduced survival, especially in the HER2 and basal/TNBC subtypes; 3) ColXVIII is co-expressed with EGFR, Our in vivo studies using the MMTV-PyMT mammary cancer model crossed with our unique total and isoform-specific Col18a1 knockout models convincingly demonstrated for the first time that the short ColXVIII is the key isoform upregulated in mammary tumors. Importantly, the short isoform was found to be responsible for the pro-tumorigenic action of ColXVIII, as the specific deletion of this isoform, but not the medium/long isoforms, significantly inhibited cancer cell proliferation, the primary tumor burden and lung metastasis (Fig. 3, Fig. S6). The short isoform has a TSP-1 domain at the N-terminus of the molecule, and it is shown here that recombinant fragments containing the TSP-1 sequence can at least partially recover the proliferation deficit caused by ColXVIII KD in human BC cells, whereas C-terminal endostatin proved ineffective in mediating this task (Fig. 2E-G). The current understanding of the specific functions of ColXVIII isoforms and their N-terminal NC domains is still limited and is focused on their developmental functions [as summarized in (20)]. Thus our finding that the short ColXVIII and its TSP-1 domain have pro-tumorigenic functions is a pioneering discovery. This is also the first study to demonstrate that ColXVIII is co-expressed and forms a complex with ErbB receptors and integrin 6 in BC cells, thus having the potential to enable downstream signaling through MAPK/ERK and PI3K/AKT pathways, leading to increased tumor cell proliferation and migration. The concept of integrated signaling through growth factor and ECM receptors is well established, and the downstream pathways of the two signaling systems overlap inside the cells (34). At present, however, we have no experimental data to prove whether ColXVIII binds directly to the EGFR/HER2 receptor pair and/or the 6 integrin, but we speculate that ColXVIII might be involved in coordinating the growth factor and ECM receptor signaling events.
We can compare our data on ColXVIII with findings regarding other ECM molecules highly expressed in cancer cells and implicated in EGFR/ErbB signaling. Tenascin-C, for example, resembles ColXVIII in many ways: it is commonly upregulated in cancer cells, especially at the invasive tumor front, it associates with a poor clinical outcome, and binds integrins and EGFR through EGF-like repeats to induce tumor cell proliferation and invasion and to support CSCs (35)(36)(37)(38). Laminin 332 is an example of a BM protein interacting with both EGFR and integrins to sustain tumorigenesis. ECM remodeling in cancer reveals cryptic EGF-like repeats from laminin 332 which is thought to stimulate SCC tumorigenesis in an EGFR-and 64 integrin-dependent mechanism (38)(39)(40). ECM proteins lacking the EGF repeats, such as decorin, can also bind to and activate EGFR and other ErbBs, but unlike tenascin-C and laminin 332, decorin displays antitumor activities (41,42). Our novel data thus place ColXVIII on the list of ECM components with a role in modulating EGFR/ErbB signaling in cancer.
Our analyses of public databases (Fig. 1, Fig. S2, Fig. S3) and human BC tissue and liquid biopsy samples (Fig. 6, Fig. S4) showed that high ColXVIII expression is associated with a bad prognosis, especially in the aggressive HER2 and basal/TNBC subtypes. These observations sustain the hypothesis that ColXVIII is implicated in BC progression and that its targeting could be beneficial in improving treatment outcomes in combination with drugs targeting these GFRs.
This assumption was confirmed by experiments showing that ColXVIII deprivation in HER2positive BC cells can enhance the efficacy of EGFR/ErbB-targeting drugs, both in vitro and in vivo (Fig. 8, Fig. S9). Simultaneous ColXVIII and EGFR/ErbB targeting was furthermore shown to be beneficial in the lapatinib-resistant HER2-positive BT474 cell line, which has an activating PIK3CA mutation downstream of the EGFR/HER2 receptor complex (33) (Fig. S9F), indicating that ColXVIII inhibition can alleviate drug resistance.
Biochemical and biomechanical signals from the three-dimensional ECM are implicated in the response and resistance of cancer drugs (30,43,44). Mechanisms by which the inhibition of ColXVIII can overcome resistance to EGFR/ErbB-targeting drugs in HER2 type BC cells (Fig.   8, Fig. S9) can be speculated upon in the light of our observations and data concerning other ECM molecules. It has been reported that disruption of the interaction between laminin 332 and integrins 64 or 31, and thereby cell adhesion to the BM, can sensitize HER2-positive BC cells to trastuzumab and lapatinib treatments by inhibiting the PI3K/AKT, MAPK/ERK1/2 and focal adhesion kinase (FAK) pathways (45). In other studies, high 1 integrin expression has been shown to predict a poor prognosis for trastuzumab-and lapatinib-treated HER2-positive BC and induce resistance to these drugs through FAK and Src signaling (46,47). The same treatments have been shown to induce expression of several ECM genes through 1 integrin and Src, including the aforementioned decorin and tenascin-C as well as many collagens (48). Interestingly, ECM stiffness per se reduces drug and radiation sensitivity in many cancers, e.g. by forming a physical barrier against drug infiltration and by CSC promotion via various molecular mechanisms, including the regulation of integrin signaling (43,49,50). (22)]. We have shown previously that the N-terminal sequences in the medium/long ColXVIII isoforms in adipose tissue support the differentiation of progenitor cells/committed precursors to form mature adipocytes (51). Studies by Gupta et al. revealed that ColXVIII is overexpressed in therapy-resistant breast CSCs, suggesting that it may have a role in the generation and propagation of these cells (25). Our work provides additional experimental evidence to support this finding, since the numbers of cells with CSC characteristics were reduced both in mouse mammary tumors with Col18a1 deletion and in human BC cells with reduced ColXVIII expression (Fig. 5, Fig. S7). The demonstrated interaction between ColXVIII and integrin 6 in BC cells (Fig. 7), a key integrin subunit in breast CSCs (26), is probably implicated in the maintenance of the stemness properties of BC cells. CSCs are not only responsible for sustaining primary tumors, but are also connected with the metastatic dissemination of neoplastic clones to distant organs (12)(13)(14)(15), and we show that both the primary tumor burden and lung colonization are reduced in mice with full or partial depletion of Col18a1 isoforms (Fig.   3, Fig. S6). Our previous work has shown that deletion of Col18a1 leads to BM loosening and reduced stiffness (52). Thus, it is possible that ColXVIII upregulation in solid cancers may affect both the biomechanical properties of the tumor ECM and the maintenance of CSCs, and thereby regulates tumor promotion and drug responses.

As a ubiquitous niche component, ColXVIII has roles in maintaining various types of tissue stem and progenitor cells [as summarized in
domain of ColXVIII and transmitted through EGFR/ErbBs and/or integrins can potentiate BC cell functions and promote the development of drug resistance, especially in the advanced HER2-type BC. The targeting of ColXVIII in the TME could therefore provide a novel therapeutic approach for achieving BC regression, even in cases where the tumor does not show any response to the clinically tested drugs that inhibit EGFR/ErbB signaling. Our data also show that ColXVIII could be of substantial value as a biomarker of BC progression, either scored in tissues, or observed in liquid biopsies even before metastasis.

Study design
The objective of this study was to examine the expression, roles, mechanisms of action and prognostic and therapeutic relevance of the BM component ColXVIII in BC.

Human BC samples and survival analysis
The

Drug tests
To investigate the medical relevance of ColXVIII targeting in combination with current BC therapies, the EGFR/ErbB inhibitors lapatinib, trastuzumab and panitimumab were used in combination with ColXVIII KD in human BC cell lines cells and their respective controls. In vivo lapatinib treatment was applied to six control and six Col18a1 null PyMT mice using two different doses to test the efficacy of the drug.

Statistical Analysis
Statistical analyses were performed using the unpaired 't' test for experiments with two groups, and a two-way analysis of variance test (Bonferroni's posttest) when comparing data from experiments with multiple groups. A repeated measures one-way analysis of variance was used to analyze the primary tumor growth curves (Dunnett's multiple comparison test and Bartlett's postcorrection test). Mouse survival analysis was performed using the log rank (Mantel-Cox) test.
Differences were considered statistically significant at a p-value less than 0.05. GraphPad Prism software was used for the statistical analyses.   (24) were used for the meta-analyses. Hazard ratios (HR) and log-rank P values were automatically computed using the best-performing threshold as the cutoff. The initial number of patients in each group (N) is indicated in the survival graphs.