ANTXR1, a Stem Cell-Enriched Functional Biomarker, Connects Collagen Signaling to Cancer Stem-like Cells and Metastasis in Breast Cancer

Breast cancer represents a diverse group of tumors with heterogeneous morphology, molecular profiles, therapeutic sensitivity, and outcome (1, 2). A subset of breast tumors is suggested to contain a minor population of cancer cells termed cancer stem cells (CSC). CSCs are believed to contribute to tumor heterogeneity, therapeutic resistance, and metastasis (3). CSCs are typically isolated using cell surface markers such as CD44, CD24, DNER, DLL1, and CD271 (1, 4, 5). Elevated enzymatic activity of aldehyde dehydrogenase or ganglioside synthase GD3S is also observed in cancer cells with CSC phenotype (6, 7). Despite progress in identifying CSC-associated markers, functional relevance of many of these markers remains to be characterized. The majority of signaling molecules that confer CSC phenotype to cancer cells are not cell surface molecules that can be used for therapeutic targeting (8). Among non–cell surface receptors enriched in CSCs, only GD3S is a therapeutic target (7). In this study, we show anthrax toxin receptor 1 (ANTXR1) as a functional CSC marker.

ANTXR1, a transmembrane protein, is encoded by highly conserved tumor endothelial marker 8 gene (9, 10). The cleaved C5A fragment of collagen VI α3 serves as its physiologic ligand (11). ANTXR1 interacts with lipoprotein receptor-related protein 6 (LRP6) and VEGF receptor (VEGFR) 2 and modulates signaling downstream of Wnt and VEGF, respectively (12–15). Furthermore, ANTXR1 is selectively expressed in tumor vasculature and promotes tumor angiogenesis (16, 17). Although ANTXR1 has previously been shown to be expressed in breast cancer cells (18), its functional role in these cells is unknown. This study provides evidence for its role in CSCs by activating Wnt signaling through its natural ligand. As a subgroup of breast cancers contains a “reactive protein group” characterized by elevated levels of collagen VI (19) and missense mutations of collagen VI α3 is observed in 6% of triple-negative breast cancers (TNBC; ref. 20), we propose the existence of a cancer-specific signaling network involving ANTXR1 and collagen VI, which impacts stemness phenotype.

Breast cancer cell lines were purchased from American Type Culture Collection and authenticated using STR Systems for Cell Line Identification (Promega) by a commercial vender (DNAcenter.com) in August 2012. TMD-231 cells have been described previously (21). MCF-10A-ER-SRC cells and the plasmid constructs bearing C5A, C5B, or C5C cDNAs were gifts from Dr. Kevin Struhl (Harvard University, Boston, MA) and Dr. Brad Croix (National Cancer Institute, Frederick, MD), respectively. Supplementary information contains details of short hairpin RNA (shRNA) and siRNAs including catalogue numbers.

MCF-10A cells were incubated with fluorescein isothiocynate (FITC)-conjugated CD44 and phycoerythrin-conjugated CD24 antibody. Primary cells were incubated with FITC-conjugated CD49f, allophycocyanin-conjugated EpCAM, and phycoerythrin-conjugated lineage markers CD31, CD45, and CD140b. Only lineage-negative cells were sorted. Supplementary information has details of antibodies.

One hundred thousand cells were seeded into ultralow adherent 100 mm plates (or 5,000 cells in 6-well plate depending on the experiment) in MammoCult Medium (Stemcell Technologies). After 7 to 10 days of culturing, mammospheres were collected, resuspended in PBS, and large colonies were counted using a hemocytometer. Alternatively, mammospheres were passed through a cell strainer (40 μ) and mammospheres blocked in the strainer were stained with Wright–Giemsa (Fisher Diagnostic). For secondary and tertiary mammospheres, mammospheres were collected, trypsinized, and 5,000 cells were replated in 6-well plates under mammosphere growth conditions. Invasion assay was conducted using Invasion Assay Kit (Millipore).

RNA was prepared using RNeasy Kit (Qiagen) and cDNA from 2 μg of RNA was synthesized using the cDNA Synthesis Kit (Bio-Rad). Quantitative reverse transcription-PCR (qRT-PCR) was conducted using SYBR Green on a TaqMan 7900HT instrument (Applied Biosystems). Microarray with biologic triplicates was conducted using Illumina HumanHT-12 V4 expression beadchip. Genes differentially expressed at P value of less than 0.01 were considered for Ingenuity Pathway and the transcription factor binding site (TFBS) enrichment analysis. Primers unique to ANTXR1 longest isoform were used for qRT-PCR analysis of primary tumor samples (primer sequences are given in the Supplementary Data).

Western blot analyses were done as described previously (22) and details of antibodies and recombinant proteins are provided in the Supplementary Information.

The Institutional Animal Care and Utilization Committee approved all animal experiments. TMD-231 cells (10⁶) with control shRNA vector or ANTXR1 shRNA were implanted into the mammary fat pad of 7-week-old female nude mice. Tumor volume was calculated using the formula: short dimension² × long dimension/2 (23). After 5 weeks, primary tumor and lungs were collected and subjected to hematoxylin and eosin (H&E) staining. A medical pathologist evaluated the slides in a single blinded manner and the metastasis index was calculated as described previously (23).

Microarray analysis of CD44⁺/CD24⁻ and CD44⁻/CD24⁺ subpopulation of immortalized cell line MCF-10A showed significant differences in the expression of more than 2,000 genes (22). Among differentially expressed genes, ANTXR1 drew our attention because this gene was previously considered functional mostly in tumor-associated endothelial cells (16). CD44⁺/CD24⁻ cells expressed 4.2-fold higher levels of ANTXR1 compared with CD44⁻/CD24⁺ cells, which we further confirmed by Western blot analysis (Fig. 1A). ANTXR1 expression in subpopulations of MCF-10A cells correlated positively with N-Cadherin but negatively with E-Cadherin expression (22). To ascertain whether ANTXR1 is expressed at a higher level in specific subpopulation of cells from normal breast, we sorted primary cells into stem cell-enriched basal (CD49f⁺/EpCAM⁻/Lin⁻), bipotent/committed luminal progenitor (CD49f⁺/EpCAM⁺/Lin⁻), and mature luminal (CD49f⁻/EpCAM⁺/Lin⁻) cells (Fig. 1B; ref. 24). Stem cell-enriched basal cells expressed highest levels of ANTXR1 compared with mature luminal cells (Fig. 1C). Analysis of gene expression array datasets from previous publications also showed 3.2-fold and 13.5-fold higher ANTXR1 in stem cell-enriched basal cells compared with luminal progenitor or mature cells of human breast and murine mammary gland, respectively (25, 26).

We next examined ANTXR1 expression in primary breast tumors. ANTXR1 expression in representative tumors sorted on the basis of Lin⁻, CD44 ± EpCAM or Lin⁻, CD49f ± EpCAM is shown in Fig. 1D and E. CD44⁺/EpCAM⁻ or CD49f⁺/EpCAM⁻ cells expressed highest levels of ANTXR1 mRNA. To further show the presence of ANTXR1-positive primary cells, lineage-negative cells from normal breast or breast cancer were stained with an antibody that recognizes the extracellular domain of the longest isoform of ANTXR1. CD44⁺/ANTXR1⁺, CD44⁻/ANTXR1⁺, EpCAM⁺/ANTXR1⁺, and EpCAM⁻/ANTXR1⁺ cells are present in both normal breast and cancer (Fig. 1F). Primary CD49f⁺/ANTXR1⁺ cells expressed lower levels of myoepithelial/basal cell marker keratin 14 (K14) compared with CD49f⁺/ANTXR1⁻ cells. This suggests that ANTXR1-positive cells either have lesser degree of myoepithelial differentiation or have features of epithelial-to-mesenchymal transition (EMT; Fig. 1G).

A subpopulation of CD44⁻/CD24⁺ of MCF-10A cells engineered to overexpress the EMT inducer Slug acquires CD44⁺/CD24⁻ phenotype (22). This phenotypic conversion was associated with elevated expression of ANTXR1 at RNA levels (Fig. 1H) and cell surface protein levels (Fig. 1I). Unlike primary cells, ANTXR1-positive and ANTXR1-negative MCF-10A-Slug cells expressed similar levels of K14 (Fig. 1J), although K14 levels in MCF-10A-Slug cells were 50% less than parental MCF-10A cells (data not shown).

To determine the prognostic relevance of ANTXR1, we evaluated publicly available gene expression array datasets for ANTXR1 expression. In the Loi and colleagues dataset (27), ANTXR1 overexpression correlated with poor recurrence-free survival among patients with estrogen receptor (ER)-negative breast cancer (Fig. 2A). In Minn and colleagues dataset (28), elevated ANTXR1 expression correlated with shorter distant metastasis-, lung metastasis-, and brain metastasis-free survival of patients with ER-negative and progesterone receptor (PR)-negative breast cancer (Fig. 2B–F). Overall, ANTXR1 overexpression is associated with poor outcome in ER-negative breast cancers, which are usually enriched for CD44⁺/CD24⁻ CSCs (29). In addition, in a dataset containing gene expression data of 478 patients with basal breast cancer (30), highest ANTXR1 (top quartile) correlated with early recurrence (Fig. 2G).

We used Oncomine to analyze ANTXR1 copy number and expression changes in breast cancer compared with adjoining normal tissue. In The Cancer Genome Atlas (TCGA) dataset (19), medullary carcinomas displayed ANTXR1 gene amplification (Supplementary Fig. S1A). ANTXR1 expression was elevated in ER-negative breast cancers compared with normal or ER-positive breast cancer (Supplementary Fig. S1B). PR-negative breast cancers also displayed elevated ANTXR1 levels compared with PR-positive tumors (data not shown). ERBB2-positive breast cancers overexpressed ANTXR1 compared with ERBB2-negative breast cancers (Supplementary Fig. S1C). TNBCs also displayed higher ANTXR1 expression than other breast cancer subtypes (Supplementary Fig. S1D). In addition, ANTXR1 is one among the top 10% overexpressed genes in 5 different breast cancer datasets present in the Oncomine database.

The above results prompted us to investigate ANTXR1 activated signals upon ligand stimulation. We selected 5 cell types for these studies; MCF-10A, SRC-transformed MCF-10A (10A-SRC), parental MDA-MB-231 (MD-231), a variant of MD-231 isolated from a mammary fat pad tumor in nude mice (TMD-231), and SUM149 cells (21, 31). At least 2 ligands are known to activate ANTXR1; the Protective antigen (PA) of anthrax and the C5A fragment of collagen VI α3 (11, 12, 32). In addition, anthrax is thought to use the ANTXR1 along with the Wnt receptor LRP6 for entry, although these results remain controversial (12, 33). ANTXR1 has previously been shown to respond to Wnt ligands (15); however, the ability of its natural ligand C5A to engage in Wnt pathway has not been shown. Vector-transfected or C5A-transfected 10A-SRC cells were stimulated with PA antigen, Wnt3a, or both and were examined for the Wnt pathway target phospho-GSK3α/β. As expected, Wnt3a increased GSKα/β phosphorylation, whereas PA antigen marginally increased phospho-GSK3α/β (Fig. 3A). C5A-transfected cells showed elevated basal phosphorylation of GSKα/β with no further increase in phosphorylation upon PA or Wnt3a addition (Fig. 3B). Similar studies were also conducted in MCF-10A cells transiently transfected with C5A, C5B (functionally active variant of C5A), and C5C, an inactive variant that cannot bind to ANTXR1 (11). Transfected C5A and C5C generated secreted proteins (Fig. 3C). C5A and C5B but not C5C increased phospho-GSKα/β (Fig. 3B). C5A and C5B increased AXIN1, a feedback negative regulator of Wnt signaling. Wnt3a and PA antigen did not further increase AXIN1 expression or phospho-GSK3α/β in cells expressing C5A, suggesting that these molecules compete for the same cell surface receptors (Fig. 3B).

AXIN2 and MSX2 are major Wnt target genes active in stem and cancer cells. AXIN2-expressing cells (AXIN2⁺) are described as Wnt-responsive bipotent stem cell subpopulation of adult mammary gland and these AXIN2⁺ cells contribute differently to basal and luminal lineages depending on developmental stages (34). MSX2 is an oncogene (35). Interestingly, C5A but not PA or Wnt3a increased AXIN2, whereas only Wnt3a increased the expression of MSX2 (Fig. 3D). Combined treatment with C5A and Wnt3a caused significantly higher expression of MSX2. To further show the requirement of Wnt pathway receptors for ANTXR1:C5A–activated signals, we treated parental and C5A-overexpressing cells with Dickkopf (DKK), a natural antagonist of Wnt receptors (36). DKK treatment reduced phospho-GSK3α/β levels in C5A-overexpressing cells (Fig. 3E). Collectively, these results show a functional crosstalk between C5A and Wnt pathway through ANTXR1. Mining for coexpression using BIIT (37) or Oncomine revealed several collagens including collagen VI α3 and MMP11, which is required for generating C5A (38), among top 50 genes coexpressed with ANTXR1 in multiple datasets. Thus, tumors are likely equipped with ANTXR1-MMP11-collagen regulatory loop that aids in Wnt pathway activation.

To further extend an association between ANTXR1 and CSC phenotype, we sorted SUM149 cells into ALDEFLUOR-positive and ALDEFLUOR-negative subfractions (Fig. 4A). This cell line also has CD44⁺/CD24⁻ and CD44⁺/CD24⁺ subfractions (Fig. 4B). ANTXR1 expression was higher in ALDEFLOUR-positive cells compared with ALDEFLOUR-negative cells (Fig. 4C). Similarly, CD44⁺/CD24⁻ cells contained higher ANTXR1 compared with CD44⁺/CD24⁺ cells. ANTXR1 levels correlated positively with LRP6 and phospho-GSK3α/β levels. Taken together, these results show an association between ANTXR1 expression and previously established CSC phenotype of breast cancer cells.

To functionally link ANTXR1 with Wnt pathway activation, we generated MD-231 and TMD-231 cells expressing shRNA against ANTXR1 using retrovirus (Fig. 5A) and a lentivirus expression system with three independent siRNAs (Fig. 5B). Another set of ANTXR1-specific siRNA was used for transient knockdown of ANTXR1 in SUM149 and TMD-231 cells (Fig. 5C). The effects of si/shRNA on ANTXR1 mRNA levels are shown in Fig. 5D. ANTXR1 knockdown resulted in reduced expression of Wnt pathway-associated proteins including LRP6, AXIN1, ZEB1, cMyc, phospho-GSK3α/β, and cyclin D1 compared with control cells (Fig. 5A–C). ANTXR1 knockdown also altered phospho-ERK1:ERK2 ratio (Fig. 5A). Note that ANTXR1 knockdown had no effect on survivin expression, suggesting specificity in the ANTXR1 pathway (Fig. 5A). ANTXR1 is likely constitutively active in these cells because microarray results showed Collagen VI α3 expression in TMD-231 cells (data not shown).

The observation of lower LRP6 in ANTXR1 knockdown cells is interesting as LRP6 is one of the Wnt receptor (13). To confirm ANTXR1 directly targets LRP6, we measured LRP6 mRNA levels by qRT-PCR. ANTXR1 knockdown cells displayed lower LRP6 mRNA compared with vector control cells (Fig. 5E). Therefore, ANTXR1 may directly regulate Wnt pathway by controlling the expression of LRP6.

We subjected TMD-231 cells with control and ANTXR1 shRNAs to Wnt pathway-specific array and a global microarray. Wnt pathway-specific array identified BMP4 as an ANTXR1 target gene. qRT-PCR assay showed higher levels of BMP4 in control cells compared with ANTXR1 knockdown cells (Fig. 5F). Microarray analysis showed differential expression of 659 genes (P < 0.01; Supplementary Table S1); 165 genes were downregulated in ANTXR1 shRNA cells compared with control shRNA cells. Ingenuity Pathway Analysis identified three ANTXR1-associated signaling networks (Supplementary Fig. S2A–S2C). β-Catenin and cMyc were integral part of these networks highlighting an association between ANTXR1, Wnt pathway, and the stemness phenotype. TFBS enrichment analysis of the regulatory regions of differentially expressed genes linked ANTXR1 signaling to transcription regulation by transcription factor EB, nuclear factor erythroid-derived 2, SMAD member 4 (SMAD4), GA-binding protein transcription factor, beta subunit 1, androgen receptor, and heat shock transcription factor 2 (P < 4.3E-02).

Gene expression array results indicated additional ANTXR1 targets linked to CSC and metastasis. For example, the levels of HSPB1, which helps to switch dormant tumors to aggressive phenotype (39), were lower in ANTXR1 knockdown cells (Supplementary Table S1). The levels of HSPA1A, which is overexpressed in the metastatic TNBC-CSCs (40), were 3-fold lower in ANTXR1 knockdown cells compared with control cells. Indeed, qRT-PCR assay confirmed the results of microarray results (Fig. 5G). Parental MD-231 cells expressed much lower HSPA1A compared with TMD-231 and its expression was unaffected upon ANTXR1 knockdown. In contrast with lower HSPA1A, TMD-231 cells expressing ANTXR1 shRNA showed elevated expression of luminal epithelial cell-enriched genes such as FOXA1, LRIG1, and TFAP2C (Supplementary Table S1). In fact, Ingenuity Pathway Analysis of ANTXR1 shRNA cell-enriched genes showed ER network (Supplementary Fig. S3A). Although, ANTXR1 knockdown in TMD-231 cells did not result in ER expression (data shown), FOXA1 expression was increased in knockdown cells (Fig. 5A). FOXA1 influences ERα expression, function, ductal morphogenesis, and is a good prognostic marker of breast cancer (41, 42). These results show a link between ANTXR1, Wnt pathway, and the metastasis-associated HSPA1A expression as well as the ability of ANTXR1 signaling to suppress luminal cell-enriched gene network.

To extend a link between ANTXR1 and HSPA1A to primary tumors, we analyzed prognostic significance of ANTXR1 and HSPA1A expression using a public database (30). Patients expressing elevated levels of ANTXR1 and HSPA1A showed worst recurrence-free, distant metastasis-free, and overall survival (Supplementary Fig. S3). Elevated ANTXR1⁺ HSPA1A expression also correlated with poor recurrence-free and metastasis-free survival of patients who did not receive systemic therapy.

To evaluate whether ANTXR1-induced signaling in human breast epithelial cells exert a stem cell-enriched function, MCF-10A with control vector or overexpressing C5A were subjected to mammosphere assay to assess stem cell function and to flow cytometry to measure stem cell-enriched basal markers. Although legitimacy of mammosphere assay for stem cell function is debated, it is at least considered to measure anoikis resistance capabilities. MCF-10A vector control cells formed fewer and smaller mammospheres (Fig. 6A). In contrast, C5A-transfected cells formed much larger and more (>5-fold) mammospheres (Fig. 6A). In addition, C5A-overexpressing cells compared with control cells showed elevated expression of CD44 (Fig. 6B).

We examined TMD-231 cells expressing ANTXR1 shRNA for reversal of stem cell features. Most of the breast CSCs show mesenchymal phenotype as evident from their elongated morphology with very little cell–cell contact (8, 43). TMD-231 cells with control shRNA displayed this morphology (Fig. 6C). In contrast, TMD-231 cells-expressing ANTXR1 shRNA grew in clusters with cell–cell contact and morphologic features of epithelial cells. Furthermore, ANTXR1 shRNA-expressing cells formed fewer and smaller mammospheres compared with cells with control shRNA (Fig. 6D). ANTXR1 knockdown increased cell surface levels of EpCAM, epithelial/luminal cell-enriched molecule (Fig. 6E).

To ensure that observed mammospheres are not just cell aggregates, we plated single cells in an ultra-low adherent, 96-well plates using flow cytometry sorter. Mammospheres developed in few but not all wells (Supplementary Fig. S4A). We also measured the ability of vector control and ANTXR1 knockdown cells to generate secondary and tertiary mammospheres. Size as well as number of secondary and tertiary mammospheres remained low with ANTXR1 knockdown cells compared with parental cells (Supplementary Fig. S4B and S4C).

To further associate ANTXR1 with CSC phenotype, we sorted ANTXR1-positive and -negative cells from TMD-231 cells and examined for their ability to form mammospheres. Cells used in this experiment can be categorized as CD44⁺/CD24⁻/ANTXR1⁺ and CD44⁺/CD24⁻/ANTXR1⁻ cells because approximately 100% of TMD-231 cells are CD44⁺/CD24⁻ (43). ANTXR1-positive cells compared with ANTXR1-negative cells formed larger and increased number of mammospheres (Supplementary Fig. S4D). Thus, CD44⁺/CD24⁻/ANTXR1⁺ cells represent a distinct population of CSCs with higher self-renewal capacity.

To determine whether ANTXR1 expression affects malignant behavior of cancer cells in vivo, we injected TMD-231 cells with control shRNA or ANTXR1 shRNA into the mammary fat pad of nude mice. Unlike parental MD-231 cells, TMD-231 cells injected into the mammary fat pad spontaneously metastasize to lungs. Tumors formed by ANTXR1 shRNA cells were significantly smaller compared with tumors derived from control shRNA cells (Fig. 7A). However, histologically, tumors between 2 groups did not show significant differences (Supplementary Fig. S5A). Furthermore, both TMD231 control shRNA and ANTXR1 shRNA cells grew at similar rate in two-dimensional cultures (Supplementary Fig. S5B). Therefore, differences in growth rate in vivo could be related to differences in stemness phenotype and the ability to establish and/or respond to a proper tumor microenvironment. We do note limitation of our assay, as limited dilution mammary fat-pad implantation is required to fully establish relationship between ANTXR1-dependent stemness phenotype and tumorigenicity.

Lungs after 5 weeks of tumor cell implantation showed extensive metastasis in animals injected with TMD-231 control shRNA compared with ANTXR1 shRNA cells (Fig. 7B). Representative H&E staining of lungs from 2 different groups is shown in Fig. 7C. Reduced lung metastasis of ANTXR1 knockdown cells could be related to lower invasive capacity of these cells or due to the inability to home and grow in lungs. As we had previously shown that CSC (CD44⁺/CD24⁻) phenotype is associated with increased invasive capacity but not homing and growth in lungs (43), and ANTXR1 knockdown reversing some of the EMT features, we examined whether ANTXR1 knockdown reduces invasive capacity using Matrigel invasion assay. Indeed, ANTXR1 knockdown cells showed reduced invasion compared with vector control cells, suggesting that ANTXR1 is required for invasion of cancer cells (Fig. 7D).

In an independent experiment with same clones but higher passage number in culture (lower ANTXR1 knockdown), tumor growth differences persisted even at 8 weeks; however, differences in metastasis rate were not statistically significant at this later time point (Supplementary Fig. S6). These findings indicate that ANTXR1 plays a substantial role in driving breast cancer growth and metastasis and its effective inhibition delays metastasis.

ANTXR1 has been studied extensively in the context of tumor vasculature. Antibodies against its extracellular domain reduced tumor-induced angiogenesis and increased sensitivity of variety of tumors to anticancer agents (16). However, the degree of ANTXR1 dependence for tumor growth and drug response varied among different cancer types and did not correlate with ANTXR1 expression in cancer-associated endothelial cells (16). In light of our discovery, we propose that the tumor-specific differences in CSCs-expressing ANTXR1 determine the sensitivity to ANTXR1-targeting antibodies. Although previous reports have indicated ANTXR1 expression in breast cancer cells and its expression correlating with shorter survival outcome (44), cell types within breast cancer or normal breast that express ANTXR1 are not known. Our study addresses this issue as well as delineates signaling pathways downstream of ANTXR1. ANTXR1 expression was observed predominantly in basal-like cell lines and in few luminal cell lines (data not shown) and had prognostic impact in ERα-negative/basal breast cancers. Iyengar and colleagues recently showed that adipocyte-derived carboxy terminal fragment of collagen VI α3 activates AKT and β-Catenin and promotes breast cancer progression by serving as a ligand for NG2/chondroitin sulphate proteoglycan receptor expressed on cancer cells (45). Our study identified a distinct pathway activated by the carboxy terminus of collagen VI α3 through ANTXR1. As stromal fibroblasts produce MMP11, which is essential for cleavage of collagen VI (38), a three-way communication between cancer cells, adipocytes, and fibroblasts may influence stem cell-enriched gene expression program in cancer cells. Recent publication by The Cancer Genome Atlas Network supports this possibility as a subgroup of breast cancers contains distinct “reactive group”, which is enriched for collagen VI (19). Further relevance of collagen VI α3 in breast cancer is evident from genome-wide sequencing studies as missense mutation affecting this gene is observed in 6% of TNBCs (20). Furthermore, coexpression analysis using Oncomine reinforces the existence of a network comprising ANTXR1, MMP11, and collagens within the tumor microenvironment. This network may be an ideal therapeutic target.

Previous studies showed physical association between LRP6 and ANTXR1 and the influence of this association on Wnt pathway (12, 13, 46). Our results show an additional link as ANTXR1-mediated signaling directly controlled the mRNA levels of LRP6 (Fig. 5). Gene expression array analysis of control and ANTXR1 knockdown cells clearly indicated the effects of ANTXR1 knockdown on transcription factor signaling network including Wnt, JNK/AP1, and E2F (Supplementary Fig. S2A–S2C). LRP6 regulatory regions contain AP1 and E2F-binding sites and ANTXR1 may use these transcription factors to control LRP6 expression.

Our results described in Fig. 5A–C showed ZEB1 as another critical target of ANTXR1. This ANTXR1:ZEB1 axis has implications on sensitivity of cancer cells to targeted therapies. A recent synthetic lethal interaction study showed elevated expression of ANTXR1 correlating with lack of response to combined MEK and BCL-XL inhibitors and sensitivity to this combination could be regained upon ZEB1 knockdown (47). As ZEB1 is a master regulator of mesenchymal differentiation and one of the regulators of CSC phenotype (48), ANTXR1 may control CSC phenotype as well as sensitivity to targeted therapy through ZEB1, which will be the subject of future investigation.

On the basis of the expression pattern, it seems that ANTXR1 controls signaling in stem/progenitor cells of both normal and cancerous breast. In cancer, these pathways include β-catenin, cMyc, TGFβ/SMAD, and JNK (Supplementary Fig. S2). While the first 3 pathways are associated with CSC phenotype, inactivation of JNK pathway is associated with breast cancer initiation/progression (49, 50). Results of these pathway analyses coupled with direct demonstration of the effects of ANTXR1 on the expression of LRP6, ZEB1, mammosphere formation, tumor growth, invasion, and metastasis provide strong evidence for ANTXR1 as a functional marker of CSCs. Moreover, it is believed that CD44⁺/CD24⁻ cells are enriched in basal breast cancers, whereas ALDEFLUOR-positive cells are enriched in luminal breast cancers. Therefore, targeting ANTXR1 would be beneficial for CSC-enriched both basal and luminal breast cancers. However, success of ANTXR1-targeted therapies will depend on which pathways downstream of ANTXR1 the cancer cells use for growth and survival and which are amicable to therapeutic intervention. If cancer cells addicted to ANTXR1 signaling are CSCs as this study indicates, ANTXR1-targeted therapies can potentially limit intratumoral heterogeneity. Furthermore, as ANTXR1 is expressed in hematopoietic and endothelial cells, therapies targeting ANTXR1 would limit the impact of tumor microenvironment on CSC features and heterogeneity.

No potential conflicts of interest were disclosed.

Conception and design: D. Chen, S. Badve, H. Nakshatri

Development of methodology: D. Chen, H. Nakshatri

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Chen, P.B.-Nakshatri, S. Badve

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Chen, C. Goswami, S. Badve, H. Nakshatri

Writing, review, and/or revision of the manuscript: D. Chen, C. Goswami, S. Badve, H. Nakshatri

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.B.-Nakshatri

Study supervision: H. Nakshatri

The authors thank Drs. Kevin Struhl, Brad Croix, and Stephen Leppla for reagents and advice. The authors also thank IUSCC flow cytometry core for flow cytometry and the Breast Cancer Signature Center-IU Simon Cancer Center at Indiana University-Purdue University (Indianapolis, IN) for the database of publicly available microarray data.

The work was supported by NIH grant R21 CA152536-01A1 and Susan Komen for the Cure grant SAC110025 (H. Nakshatri).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.