Identification of the Novel Role of CD24 as an Oncogenesis Regulator and Therapeutic Target for Triple-Negative Breast Cancer

Breast cancer is the most frequently diagnosed cancer among women worldwide with over 1 million cases and nearly 400,000 deaths per year (1, 2). Breast cancer is considered a heterogeneous disease that is classified into four subtypes based on the presence of hormone receptors for estrogen (ER) and progesterone (PR) as well as the expression and gene amplification status of HER2/ErbB2. These four subtypes are luminal A (ER⁺ or PR⁺/HER2⁻), luminal B (ER⁺ or PR⁺/HER2⁺), HER2 (ER⁻ or PR⁻/HER2⁺), and basal-like triple-negative breast cancer (TNBC; ER⁻/PR⁻/HER2⁻; refs. 3, 4). The substantial differences in receptor status in the breast cancer subtypes often serve as a guideline for different therapeutic interventions (3–6). However, the therapeutic choices for patients with TNBC are limited because of the lack of targeted therapeutic approaches (7). Cohort studies have indicated that patients with TNBC are associated with shorter relapse intervals and poorer overall survival within first 5 years after their initial diagnosis than patients with other breast cancer subtypes (3, 8).

EGFR and Met are primary surface receptors for EGF and hepatocyte growth factor (HGF), respectively. Ligand-mediated EGFR/Met activation promotes tumor angiogenesis and progression through a common downstream Stat3/Src/Akt signaling pathway (9–11). In recent years, high frequency of EGFR and Met expression has been observed in patients with TNBC and is strongly associated with poor prognosis and overall survival (12, 13). TNBC patients coexpressing Met and EGFR have shorter disease-free survival than those expressing only EGFR (14). These findings suggest that Met and EGFR play an important role in TNBC progression.

CD24, a heavily glycosylated mucin-type glycosylphosphatidylinostol-anchored cell surface molecule, has been identified as being highly expressed in several human cancers including, breast, lung, and hepatocellular carcinoma. In breast and lung cancers, CD24 has been identified as a key surface receptor in P-selectin binding (15, 16), a cell adhesion molecule expressed on activated endothelial cell and platelets (17), that promotes tumor growth by activating Src kinase through the lipid raft (18–20). Clinically, a high CD24 expression is associated with poor outcomes for cancer patients; however, the role of CD24 in TNBC progression is still poorly understood.

In this study, we used an orthotopic xenograft mouse model to study the metastatic lung and lymph node (LN) colonization of TNBC. Our study revealed a novel mechanism in which CD24 positively regulates lymphangiogenesis and angiogenesis through EGF-induced EGFR/Stat3/Src and HGF-induced Met/Stat3/Src signaling cascades to promote primary tumor growth as well as TNBC LN and lung metastasis. Our clinical analyses identified a new prognostic marker for TNBC in that patients coexpressing CD24 and Met had poorest outcomes. Our study suggested that CD24 could be a treatment target for TNBC, and likely for other breast cancer subtypes with CD24 overexpression.

MDA-MB-231, MCF-7, BT549, Hs578T, and human umbilical vein endothelial cells (HUVEC) were obtained from the ATCC. Human dermal lymphatic epithelial cells (LEC) were obtained from PromoCell. Human breast cancer cell lines were cultured in DMEM supplemented with 10% FBS (Invitrogen). MDA-MB-231–derived sublines LC and IV2 (21) were also cultured in the above condition. The detailed information was provided in the Supplementary File. LECs were cultured in endothelial cell growth medium MV2 (EGM-MV2) according to the manufacturer's instruction. HUVECs were cultured in medium 199 with 25 U/mL heparin (Sigma), 30 μg/mL endothelial cell growth supplement (Millipore), 2 mmol/L l-glutamine, and 10% FBS. All cell lines were cultured in a humidified incubator at 37°C with 5% CO_2. All cell lines were verified as mycoplasma-free on December 20, 2017, by DAPI staining and are routinely checked by DNA short tandem repeat (STR) for verification.

Cells were incubated with 5 mmol/L ETDA for 10 minutes to detach cells from the petri dish. Cells were then washed with 5 mL PBS twice to remove EDTA. Note that 5 × 10⁵ cells were stained with FITC-conjugated mouse anti- human CD44 and Phycoerythrin-conjugated mouse anti- human CD24 (BD Bioscience) at 1/20 dilution at 4°C for 40 minutes and were kept away from light. Calibur CellQuest Pro (BD) software was applied for data acquisition and analysis.

Cells were lysed in RIPA buffer (1% TritonX-100, 50 mmol/L, pH 7.4, Tris-HCl, 150 mmol/L NaCl₂, 0.1% SDS, 1% cholerate), and cell lysates were subjected to SDS-PAGE electrophoresis. Protein was transferred onto PVDF membrane overnight at 35 V. PVDF membrane was blocked in 5% non-fat milk and hybridized with the specific primary antibodies followed by the horseradish peroxidase–conjugated secondary antibody.

The procedure of establishing lung-dormant breast cancer cells was as follows: One million MDA-MB-231 parental cells were injected orthotopically in SCID mouse at second mammary fat pad. After 3 months, mice were sacrificed and tumor nodules in the lung were isolated and subsequently cultured in DMEM supplemented with 10% FBS. The lung-dormant cells isolated from three lung nodules were named as LC-1, LC-2, and LC-3 sublines. The resulting LC sublines were authenticated using STR analysis before carrying out the subsequent functional assays. The highly metastatic IV2 cells were generated from the in vivo tail vein injection protocol, and the detailed procedure was previously described (21). The SCID mice were provided by the National Laboratory Animal Center (Taiwan). All animal studies were approved by Institutional Animal Care and Use Committee of National Health Research Institute.

Detailed procedures of RNA extraction and qRT-PCR are described elsewhere (21). Specific primers used in qRT-PCR are listed in Supplementary Table S4.

Cells were grown on poly-L-lysine–coated coverslips in 6-well plates and were incubated overnight. The control siRNA and CD24 siRNA were transfected into cells respectively using RNAiMAX (Invitrogen) for 8 hours. The transfected cells were fixed and permeabilized with 3.75% formaldehyde and 0.1% saponin, respectively. The permeabilized cells were blocked in 1% BSA for 1 hour and stained with the specific primary antibodies at 4°C overnight followed by staining with fluorescent dye–conjugated secondary antibodies for 1 hour. Cells were washed with PBS and mounted in Prolong Antifade reagents (Invitrogen). Cells were observed and photographed using the Leica SP5 II scanning confocal microscope with 60x objective HCX PLAPO (NA = 1.25, oil immersion; Leica). Colocalization of internalized receptors and lysosomes was analyzed by using the Colocalization module of the Leica LASX software according to the software instruction.

Lentivirus-based CD24 shRNA virus was obtained by transient cotransfection of pLKO vector containing shRNA of CD24 with pCMVDR8.91 and pMD.G, into a human 293T cell line. The supernatant that contains virus particles was concentrated by mixing with 1/4 volume of Lenti-X solution (Clontech) at 4°C overnight and then was centrifuged at 3,500 × g for 50 minutes to obtain the concentrated virus particles. Cells were infected with the control virus or virus containing CD24 shRNA for 48 hours, and the infected cells were selected with 5 μg/mL puromycin for 1 week. The knockdown efficiency of CD24 was examined by qRT-PCR using the specific human CD24 primers. For the In Vivo Imaging Systems (IVIS) study, cells were further labeled with a luciferase reporter by transfection with the pLVX-GFP-2A-Luc plasmid.

Cells were treated with 20 μg/mL cycloheximide (CHX) for the indicated time points and then harvested in RIPA buffer containing protease inhibitors (Roche) for the subsequent Western blot analysis.

Transwell invasion assays were carried out using 8.0 μm Biocoat culture insert (BD Biosciences) as described previously (21).

One million control and CD24-depleted cells were seeded in 6-cm plates respectively 24 hours prior to assay. The conditional medium (CM) was collected from the culture medium of the control and CD24-depleted cells. LECs and HUVECs were seeded in the 0.2 μm Transwell insert sitting on the 6-well plate, and CM was added to the lower chamber. After 24-hour incubation, the migrated LEC and HUVEC cells were stained with 0.5% crystal violet staining solution for 20 minutes. The migrated cells were analyzed using ImageJ software (NIH, USA).

Ten thousand HUVECs cells labeled with CellTracker green (Thermo Fisher Inc.) were mixed with CM collected from the culture medium of the control and CD24-depleted cells and were seeded in the Matrigel-coated 24-well plate for 4 hours followed by fixation with 3.75% formaldehyde. Tube formation was observed and photographed by fluorescent microscope.

Note that 1 × 10⁴ cells were plated in 6-well plate overnight. The plate was then placed on the stage of the live imaging microscope Leica AF 6000 LX (Leica) to perform time-lapse recording. The migration distance and velocity of cells were analyzed using MetaMorph software (Molecular Devices).

Detailed information of antibodies and reagents used in this study is listed in Supplementary Table S3.

Note that 1 × 10⁶ cells with indicated treatments were suspended in 100 μL PBS and injected individually into the tail veins of C.B -17 SCID mice. Mice were sacrificed at the indicated timepoint, and lung tissues were dissected and subjected to histologic examination.

Xenograft tumors and mouse lungs were fixed in 4% formaldehyde overnight and were embedded in paraffin. Sections (6 μm thick) were prepared from tissue blocks and were subjected to hematoxylin–eosin (HE) stain.

Note that 1 × 10⁶ cells were suspended in 100 μL PBS and subsequently injected into the tail veins of SCID mice. After 16 hours, mice were then anesthetized with isoflurane and were subjected to lung perfusion. In brief, mouse chest was opened up, and an incision was made in the left ventricle of the heart with a scalpel. A butterfly needle attached to a 50-mL syringe fill with PBS was inserted to the right ventricle of the heart, and PBS was gently pushed into the pulmonary circulation until the mouse lungs were white and the fluid coming out of the left ventricle became clear.

CD24 monoclonal antibody was purchased from Beckman Coulter, Inc. (IM0118). This antibody was adapted from prior trials in mice (22) and humans (23, 24). Administration of CD24 monoclonal antibody was as the following: 7 days after cancer cells’ i.v. injection, mice were injected with CD24 monoclonal antibody every 2 days (0.01 mg in 100 μL PBS; 0.03 mg/kg) with six injections in total. The control group was similarly injected with mouse IgG1 control antibody. For the survival assay, higher concentration of CD24 mAb was used (0.04 mg/100 μL; 0.12 mg/kg).

Luciferase-labeled cells with different treatments were injected into fourth fat pad of mice. Tumor growth was followed until the endpoint of the experiment where tumors reached 2 cm in diameter.

Cells were transfected with CD24 siRNA purchased from Thermo-Fisher using Lipofectamine RNAiMAX (Invitrogen) for 8 hours. The transfected cells were further incubated for 40 hours in 10% FBS DMEM and harvested at 48 hours posttransfection for subsequent qRT-PCR and Western blot analysis.

IHC staining was carried out as the published procedure (25) using anti-CD31, anti-LYVE1, anti-Ki67, anti-CD24, anti-EGFR, and anti-Met. The results of IHC score of CD24 were determined as follows: 0 score: No observed membrane staining; 1+ score: Incomplete membrane staining that is faint in >10% tumor cells; 2+ score: Circumferential membrane staining that is incomplete in >10% tumor cells or complete circumferential membrane staining in ≦10% tumor cells; 3+ score: Dark, homogenous, chicken wire pattern in >10% tumor cells. The H-score of CD24, EGFR, and Met in 133 TNBC cases was calculated as the staining intensity of cancer cells (0 = none; 1 = weak; 2 = moderate; and 3 = strong) multiplied by the percentage of tumor sections being stained (0%–100%). All IHC samples were independently scored by two investigators in a double-blinded manner and were reviewed by a pathologist.

We applied gene set enrichment analysis (GSEA; ref. 26) to identify the functional gene sets or biological pathways enriched in the differentially expressed genes between the parental 231 cells and LC cells. In the analysis, log₂R was used as ranking metric, where R was the ratio of normalized gene expression level of a gene in LC cells to that in parental 231 cells; C2 (CP and CGP) was the gene set collection used. GSEA Desktop Application (http://software.broadinstitute.org/gsea/downloads.jsp) was used for GSEA implementation.

The mRNA levels of CD24, MET, EGFR, and patients’ survival status in public datasets (27, 28) were fetched from the Oncomine database (https://www.oncomine.org). Breast cancer samples in Curtis dataset were categorized into luminal A, luminal B, HER2, and triple negative for the subsequent Kaplan–Meir survival analysis. The normalized RNA-seq data of CD24 were retrieved from The Cancer Genome Atlas (TCGA) dataset and categorized into LN positive and negative for the subsequent Kaplan–Meier survival analysis. For the association study of CD24 and MET combination with survival status, TNBC samples in Curtis dataset were further grouped as “CD24^high/MET ^high,” “CD24 ^high/MET ^low,” “CD24 ^low/MET ^high,” and “CD24 ^low/MET ^low.” For the association study of CD24 and EGFR combination with survival status, TNBC samples were grouped as “CD24 ^high/EGFR ^high,” “CD24 ^high/EGFR^low,” “CD24 ^low/EGFR ^high,” and “CD24 ^low/EGFR ^low.” The Kaplan–Meir survival analysis was performed based on the median value of mRNA levels of the respective gene analyzed. A gene expression level lower than the median value was grouped as “low expression,” whereas a gene expression higher than the median value was defined as “high expression.” The median values of the analyzed datasets were described in the figure legend of Fig. 6.

Fifteen paired primary and LN metastatic tumors were collected according to National Taiwan University Hospital's Institutional Review Board–approved guideline. One hundred and thirty-three TNBC tumor specimens were obtained through the archives of the Department of Pathology, Kaohsiung Veterans General Hospital (KSVGH), between 1991 and 1999 and were approved by the Institutional Review Board at (KSVGH). Written-informed consents were obtained from the patients. This study was performed in accordance with the Declaration of Helsinki.

Statistical analysis was conducted using the SPSS 18.0 statistical software package (SPSS Inc.). The Student t test was used for the comparison between the control and the experimental group. One-way ANOVA was applied to compare the multiple groups of more than two. Differences were considered significant at P < 0.05 (*, P < 0.05; **, P < 0.01; and ***, P < 0.001). N.S. represented no significance. The Kaplan–Meier survival curve was analyzed with the Cox proportional hazards regression model. Data were presented as the mean ± SEM.

To identify the crucial molecular changes of metastatic TNBC cells that colonize in distant organs and how they react and adapt to a foreign environment, we isolated and analyzed metastatic cancer cells that colonized in mouse lungs. As shown in Supplementary Fig. S1A, mice were sacrificed 3 months after the cancer cells were injected, and metastatic nodules were detected in the mouse lung. Three nodules were dissected individually for in vitro culturing to establish the three sublines, which we referred to as the “lung colonization (LC)” sublines, LC-1, LC-2, and LC-3 (Supplementary Fig. S1B–S1D).

First, we compared the cellular behaviors of the resulting LC cells with the original MDA-MB-231 parental cells and MDA-MB-231-IV2 cells, which were established from the early lung metastases of our previous study (21). The representative data were collected from LC-1, IV2-1, and 231 cells. Time-lapse imaging revealed that the migratory distance and velocity of the LC-1 cells were the shortest and slowest, respectively, among the three compared isogenic lines, whereas the IV2-1 cells outperformed the 231 and LC-1 cells for these aspects (Fig. 1A and B). Transwell invasion assay showed that the LC-1 cells had the poorest invasive ability among three isogenic lines (Fig. 1C). In addition, we found the LC cells, which exhibited decreased migration/invasion ability, could form the highest number of colonies in the soft agar compared with the 231 and IV2 cells, suggesting that the LC cells gained an increased ability for anchorage-independent growth (AIG; Fig. 1D).

Because the LC cells exhibited epithelial-like morphology and poor in vitro migration/invasion (Supplementary Fig. S1D; Fig. 1A–D), we subsequently examined whether the levels of epithelial–mesenchymal transition/mesenchymal–epithelial transition (EMT/MET)- and cancer-initiating cell (CIC)–related genes had been altered. The results of qRT-PCR showed that the EMT/MET-related genes were not significantly changed (>2-fold change) in the LC cells, except that the expression of CLND1, a gene encodes for a tight junction protein claudin-1, exclusively expressed in epithelial cells, was increased by at least 8-fold compared with that in parental cells (Fig. 1E). Screening of the expression of breast cancer–initiating cells (BCIC) makers revealed that CD24 expression was increased approximately 35- and 5-fold in the LC-1 and IV2-1 cells, respectively, compared with that in the parental cells (Fig. 1E). This result was confirmed through flow cytometry analysis; that is, approximately 20% and 7% of the LC-1 cells and IV2-1 cells exhibited CD24 positivity, respectively, compared with the parental cells, of which less than 2% of cells expressed surface CD24 (Fig. 1F). Western blotting analysis showed that the LC cells expressed high levels of CD24, claudin-1, ZO-1, which is another tight junction protein, and fibronectin (Fig. 1G), which were reported to be important for cancer cells to grow in the lung microenvironment (29). Immunofluorescent staining confirmed that claudin-1 was localized to the membrane and the cell–cell junction region in LC-1 cells, and no claudin-1 expression was detected in both 231 and IV2-1 cells (Fig. 1H). To more effectively characterize the intrinsic property of the LC cells, we subsequently examined whether the activity of small GTPase Rac1/Cdc42, which is important for cell motility and invasiveness, was changed in the MET-like LC cells. Using GTPase pull-down assay, we determined that the LC cells exhibited a low level of GTP-bound Rac1 and Cdc42, which could account for their decreased migration and invasion ability (Fig. 1I).

Next, we examined the in vivo behavior of the LC cells. One million luciferase-labeled LC and 231 cells were separately injected into the mammary fat pad no 4 of each SCID mouse to observe the tumor growth and the metastatic spreading. An IVIS imaging system was used to examine the metastatic spreading of tumor cells 7 weeks after injection.

The IVIS results showed that the LC cells grew faster and resulted in bigger tumors that did the parental 231 cells (Supplementary Fig. S2A and S2B) and exhibited increased LN metastasis ability compared with the parental 231 cells (Supplementary Fig. S2B). The IHC staining of cell proliferation marker Ki-67 revealed that LC-derived tumors showed increased Ki-67 expression in the nucleus as compared with 231-derived tumors, suggesting that LC cells were more proliferative than 231 cells in the in vivo setting (Supplementary Fig. S2C). Furthermore, lung and LN metastasis analyses were performed using ex vivo IVIS imaging (Supplementary Fig. S2D). The photon signals (the presence of metastatic cells) in the mouse lung and LNs were normalized to primary tumor size. The quantitative results of ex vivo IVIS imaging showed that more photons were detected in the LNs and lungs of mice injected with the LC cells than in those of mice injected with the parental 231 cells (Supplementary Fig. S2E). Notably, although LC cells had decreased in vitro cell motility and invasiveness (Fig. 1C and D), they could metastasize to the LNs and lung, and efficiently colonize there.

Screening the EMT/BCIC makers revealed that the LC cells expressed a high level of CD24 (Fig. 1E and F); this prompted us to examine whether CD24 is associated with LC cell tumorigenicity and metastasis (Supplementary Fig. S2). Furthermore, the role of CD24 in TNBC LN and lung metastasis has not been addressed. Therefore, we investigated the contribution of CD24 to TNBC metastatic colonization and progression.

The effect of CD24 on a metastatic colonization-related property (i.e., AIG) was assessed. Soft-agar assay showed that CD24 depletion in CD24-expressing LC cells considerably impaired their colony-forming ability (Supplementary Fig. S3A and S3B). This phenomenon was also observed in other CD24-expressing TNBC lines MDA-MB-468 and Hs578T (Supplementary Fig. S3A and S3B). In addition, CD24 inhibition also reduced the cell proliferation of CD24-expressing TNBC cells (Supplementary Fig. S3C). Similar results were observed in the CD24-expressing luminal breast cancer cell line MCF-7 (Supplementary Fig. S4).

To explore the potential role of CD24 in metastasis, we generated stable CD24 knockdown (KD) IV2 cells that also feature CD24 expression. One million luciferase-labeled CD24 KD IV2 cells or the control IV2 cells were injected through the tail vein into SCID mice, respectively, and lung colonization status was examined 3 weeks later using the IVIS imaging system. The results showed that the CD24 KD IV2 cells exhibited substantially weaker lung colonization signals than did the control cells (Supplementary Fig. S5A and S5B). HE staining of mouse lung sections further confirmed this result (Supplementary Fig. S5C and S5D).

Because extravasation is the crucial step for the cancer cells to colonize the distant organs (30), we examined whether the early extravasation event was the defining factor for the decreased lung colonization of the CD24 KD cells. Lung perfusion prior to lung tissue collection was performed to examine the early lung targeting event. Mice i.v. injected with CD24 KD cells or control cells were perfused with 1X PBS 20 hours after injection to remove cancer cells inside the lung blood vessels. Cancer cells in the mouse lung (indicating that the cells had completed extravasation) were detected by qRT-PCR using human-specific HPRT1 gene primers. The result showed no significant difference in the early lung targeting between the group of mice injected with CD24 KD and that injected with control cells (Supplementary Fig. S6). Collectively, these results indicated that CD24 was less important for targeting but was crucial for metastatic lung colonization.

To further assess the role of CD24 in TNBC metastasis, the orthotopic mouse model was used. One million cells each from the two luciferase-labeled CD24 KD IV2 lines (CD24 KD and CD24 KD-2) and the IV2 control cells were injected into the fourth mammary fat pad of SCID mice individually to observe tumor growth and metastatic progression. As the tumor progressed, the CD24 KD tumors not only grew more slowly than those of the control group (Fig. 2A–D) but also failed to metastasize to the neighboring LNs (inguinal and axillary LN; Fig. 2E and F). This was confirmed through ex vivo IVIS imaging which revealed that CD24 depletion led to decreased LN as well as lung metastasis (Fig. 2G). The qRT-PCR detection of metastatic cells in mouse lungs and LNs exhibited the similar results (Fig. 2H). The metastasis signals were normalized to primary tumor size. The effect of CD24 depletion on the in vivo property of the LC line, which had strong LN metastatic ability (Supplementary Fig. S2), was also examined. As expected, the CD24 KD LC cells exhibited decreased tumorigenicity (Fig. 2I–L) and failed to metastasize to LNs and lung compared with the control LC cells (Fig. 2M and N). Conversely, ectopic expression of CD24 in the CD24-low 231 cells considerably promoted primary tumor growth compared with the control cells (Supplementary Fig. S7A–S7D).

Taken together, both tail vein and orthotopic metastasis models indicated that CD24 has a key role in TNBC tumor progression as well as LN and lung colonization.

Because tumor lymphangiogenesis and angiogenesis are crucial in the growth and metastatic progression of tumors, we next examined the effect of CD24 KD on these events in xenograft tumors.

IHC staining of the lymphatic vessel marker LYVE-1 and the blood vessel marker CD31 revealed that IV2 and LC tumors with CD24 KD had considerably decreased lymphatic and blood vessels compared with the control IV2 and LC tumors (Supplementary Fig. S8A and S8B). Conversely, 231 tumors derived from CD24-transfected 231 cells had increased lymphatic and blood vessels compared with the control 231 cells–derived tumors (Supplementary Fig. S7E and S7F). This result motivated us to further examine the association between CD24 and lymphatic vessel and blood vessel formation. Tumor-secreted vascular endothelial growth factors C, D, and A (VEGF-C, VEGF-D, and VEGF-A, respectively) are the well-known growth factors that recruit nearby lymphatic endothelial cells and endothelial cells to the tumor site to form supportive lymphatic and blood vessels for tumor growth (31, 32). First, we examined whether LC or IV2 cells exhibited increased VEGF-A, -C, and -D expression. The qRT-PCR result indicated that the LC subline expressed the highest level of VEGF-A, -C, and -D with a 2- to 6-fold increase compared with the 231 cells (Supplementary Fig. S8C). In addition, the IV2 cells exhibited significantly increased VEGF-A and -C expression (Supplementary Fig. S8C). Knockdown of CD24 in the LC or IV2 cells reduced the VEGF-A and -C mRNA expression level, whereas the expression level of VEGF-D was unaffected (Supplementary Fig. S8D). These results were confirmed through ELISA of these factors in the culture media (Supplementary Fig. S8E).

On the basis of these findings, we subsequently examined whether CD24 depletion could affect endothelial cell migration and tube formation. Coculture Transwell assay showed that knockdown of CD24 in the IV2 and LC cells significantly reduced IV2 or LC cells’ CM-induced migration of LECs and HUVECs (Supplementary Fig. S8F). Treating HUVEC cells with the CM collected from the culture of CD24-depleted IV2 or LC cells reduced the tube formation (Supplementary Fig. S8G).

These results combined with the aforementioned in vivo observations suggested that CD24 plays a central role in regulating tumor-derived VEGF-A and -C expression to establish lymphatic and blood vessel networks at primary tumor site to facilitate growth, metastatic spread, and subsequent metastatic colonization.

Because of the lack of intracellular kinase domain in CD24 (19), we hypothesized that membrane-bound CD24 might affect other membrane receptors that mediate VEGF expression to regulate lymphangiogenesis and angiogenesis. This speculation was supported by our GSEA of differential gene expression profile between CD24^high LC cells and the parental 231 cells, which revealed that the EGFR- and Met-mediated pathways (9, 33, 34), two key receptor tyrosine kinase (RTK) signaling pathways known to activate VEGF expression in cancers, were significantly enriched (Fig. 3A).

These findings prompted us to investigate the role of CD24 in EGFR- and Met-driven downstream signaling as well as VEGF expression in TNBC cells. We first confirmed that treatment of TNBC cells with 50 ng/mL of HGF or 100 ng/mL of EGF could induce VEGF-A and -C expression (Supplementary Fig. S9A and S9B). The effect of CD24 depletion on HGF- and EGF-induced signaling cascade was subsequently examined. CD24 expression in the IV2, LC, and two other TNBC lines, BT-549 and Hs578T, was depleted, and the cells were treated with HGF at a concentration of 50 ng/mL for 20 minutes. The results showed that as compared with the control cells, CD24 depletion significantly reduced HGF-induced Met phosphorylation and downstream Src/Stat3/Akt phosphorylation, which are known to drive VEGF production (Fig. 3B). This phenomenon was observed for all four TNBC cell lines tested. In addition, we noticed that the protein level of Met was significantly decreased in CD24-depleted TNBC cells (Fig. 3B). This might explain in part the inhibition of HGF-induced Met/Stat3/Src/Akt phosphorylation in the CD24-depleted cells.

To analyze EGFR-mediated signaling, TNBC cells were treated with EGF at a concentration of 100 ng/mL. The EGF-induced EGFR/Stat3/Src/Akt phosphorylation cascade was subsequently analyzed through Western blotting. Similarly, the EGF-induced EGFR/Stat3/Src and Akt/MEK1/2 phosphorylation was significantly decreased in CD24-depleted TNBC cells compared with the control cells (Fig. 3B). Notably, we observed EGFR protein was significantly decreased in the absence of CD24 in the TNBC cells (Fig. 3B). Overall, these findings above revealed the new role of CD24 in EGFR^high and Met^high TNBC in that CD24 loss significantly impaired EGF- and HGF-induced signaling.

Because CD24 loss caused the downregulation of EGFR and Met at the protein level, we evaluated whether CD24 could affect EGFR and Met protein stability. CD24-depleted cells were treated with CHX for 24 hours, and EGFR and Met protein levels were analyzed over time. Western blotting showed that EGFR and Met protein levels were significantly decreased at a faster rate over 24 hours in the CD24-depleted cells compared with the control cells (Fig. 3C), indicating that CD24 loss increases the protein instability of EGFR and Met. The protein degradation of membrane EGFR and Met primarily occurs through receptor internalization followed by lysosome-dependent degradation (35). Here, we demonstrated that bafilomycin A1 treatment, a known V-ATPase inhibitor, could not restore the activation of EGFR and Met upon ligand stimulation but the total protein level in CD24-depleted BT549 cells, implying that the receptor internalization may occur upon CD24 depletion (Fig. 3D). We subsequently explored the effect of CD24 depletion on ligand-induced EGFR/Met internalization and degradation through immunofluorescence staining. Eight hours after CD24 siRNA transfection, the control and CD24 siRNA-transfected cells were serum-starved overnight followed by 2 hours of EGF or HGF stimulation, and immunofluorescent staining was performed. The results indicated that at 0 hours before ligand stimulation, the internalized EGFR and Met (indicated by red fluorescence) were observed in CD24-depleted LC cells compared with the control cells, in which EGFR/Met could be found on the membrane (Fig. 4A and C, top plots). We found that some internalized EGFR/Met were colocalized with lysosomes (indicated by green fluorescence; LAMP-1–positive vesicles). These results were supported by Z stack analysis and fluorescence intensity score (Fig. 4A and C). Quantitative colocalization analysis using Mander's coefficient analysis (ImageJ plug-in) suggested that the CD24-depleted cells without EGF/HGF stimulation underwent an approximately 7-fold increase in EGFR/Met and LAMP-1 double-positive vesicles compared with the control (Fig. 4A and C; bottom right). Next, ligand-induced internalization and degradation of EGFR and Met were examined in CD24-depleted cells. After 2-hour EGF/HGF treatment, we determined that the internalized EGFR/Met more frequently colocalized with lysosomes in the CD24-depleted cells compared with the control cells (Fig. 4B and D, top plot). Moreover, the Z stack analysis and fluorescence intensity score provided similar results (Fig. 4A and C, bottom plot). Quantitative colocalization analysis indicated that CD24-depleted cells with EGF/HGF stimulation exhibited an approximately 48% and 31% increase in EGFR/LAMP-1 and Met/LAMP-1 double-positive vesicles, respectively, compared with the control (Fig. 4B and D, bottom right).

Overall, confocal microscopy analysis demonstrated that loss of CD24 expression increased EGFR/Met internalization and subsequent lysosome-mediated degradation in both steady-state and ligand-stimulated conditions.

We subsequently assessed whether CD24, a membrane-bound molecule vital for primary tumor growth and metastasis, could be a therapeutic target for cancer therapy. A previous study reported on the use of the ALB9 mAb against CD24 for bladder cancer therapy in a mouse model (36). In the present study, we evaluated the therapeutic potential of ALB9 mAb on TNBC metastatic colonization. First, we examined the effect of CD24 mAb on AIG of LC cells. The result revealed that incubating of 15,000 LC cells with CD24 mAb ALB9 at a concentration of 2 μg/mL for 30 minutes prior to the soft-agar colony assay significantly impaired the AIG of LC cells as compared with the cells treated with the control antibody (Fig. 5A). Subsequently, we used this mAb for the in vivo experiment to study the effect of CD24 mAb on lung metastatic colonization of TNBC. Mice were i.v. injected with 0.5 million luciferase-labeled LC cells followed by ALB9 mAb treatment (Fig. 5B). The mice were treated with ALB9 mAb or IgG1 control antibody at a dosage of 10 μg in 100 μL 1X PBS once every 2 days over 6 times (Fig. 5B). The lung metastatic burden of the mice was monitored before and after antibody treatment using the IVIS system. The IVIS result showed that the two groups of mice exhibited no significant differences in lung metastatic burden before antibody injection (Fig. 5B). By week four, the mouse group that received ALB9 mAb treatment exhibited a reduced lung metastatic burden as indicated by luminescent signals compared with the mouse group treated with the IgG1 control. We also observed the similar inhibitory effect of CD24 mAb using another CD24-expressing TNBC cell line MDA-MB-468 (Supplementary Fig. S10). Moreover, we assessed whether CD24 mAb treatment could prolong the survival of mice with metastatic lung burden. Ten mice were i.v. injected with 2 × 10⁵ LC cells and randomized into two groups with each group containing 5 mice. One week later, mice were i.v. injected with either the control IgG1 Ab (n = 5) or CD24 mAb (n = 5) every 2 days at a dose of 0.12 mg/kg each injection for a total of six injections (Fig. 5C, left). By 20 days, the body weight of mice without CD24 mAb treatment began to drop significantly as compared with those with CD24 mAb treatment (Fig. 5C, right). The Kaplan–Meier survival analysis showed that mice with CD24 mAb treatment significantly survived longer than the control mice (Fig. 5C, Log-rank: P = 0.02). Our data indicated that CD24 mAb treatment could significantly prolong the survival time of mice bearing lung metastatic burden, suggesting the therapeutic potential of CD24 mAb in CD24-positive breast cancer patients.

We subsequently explored the clinical relevance of CD24 in TNBC patients. The Kaplan–Meier analysis indicated that high expression of CD24 was significantly correlated with poor 5-year survival of breast cancer patients in public Curtis datasets (Fig. 6A). When stratified into different subtypes, a high level of CD24 was significantly associated with poorer 5-year survival in patients with luminal A and TNBC (Fig. 6A; Supplementary Table S1). By contrast, although CD24 expression status did not statistically correlate with 5-year survival of patients with HER2 and luminal B breast cancer (Fig. 6A; Supplementary Table S1), we nevertheless observed a trend of relatively poor survival in the CD24^high population. Moreover, a high level of CD24 was associated with higher 5-year metastatic recurrence, although it did not reach the level of statistical significance (Fig. 6B; Supplementary Table S2). In addition, we found the similar result in TCGA dataset and further showed that high CD24 expression was significantly correlated with poorer overall survival in breast cancer patients with LN metastasis as compared with those without (Fig. 6C). We also collected paired primary tumor and metastatic LN from 15 TNBC patients to examine the expression of CD24. Among those 15 patients, 7 (47%) showed higher CD24 expression in metastatic LN as compared with the corresponding primary tumors. The representative CD24 IHC images from 5 patients were shown in Fig. 6D. Five patients were CD24 negative both in primary and metastatic LN, and CD24 was similarly expressed in primary and metastatic LN in 3 patients (Fig. 6D). Together, these results suggested that breast cancer patients exhibiting high CD24 expression presented poor clinical outcomes, particularly patients with TNBC and luminal A, and were correlated with LN metastasis.

Because the presence of CD24 is important for mediating the EGF- and HGF-driven pathways in TNBC cells (Figs. 3 and 4), we examined the correlation of CD24 expression levels in combination with MET/EGFR levels with the survival status of TNBC patients. As shown in Fig. 6B, we found that TNBC patients coexpressing CD24 and MET (CD24^highMET^high signature) exhibited the poorest 5-year survival among patients stratified by CD24/MET expression status (Fig. 6B). In addition, CD24^highMET^high signature also predicted faster metastatic recurrence in TNBC patients (P = 0.004, Fig. 6C). In terms of the association of coexpression status of CD24 and EGFR in TNBC, we determined that EGFR expression is significantly correlated with the metastatic events in patients with TNBC regardless of CD24 expression (Fig. 6C). By contrast, TNBC patients coexpressing CD24 and EGFR, unlike CD24^high/MET^high signature, did not show poorer 5-year survival (Fig. 6B). In addition, we found CD24 was significantly correlated with EGFR (Spearman correlation: 0.224, P = 0.009) and Met (Spearman correlation: 0.410, P < 0.0001) in terms of protein expression (Supplementary Fig. S11). Taken together, our clinical analyses revealed a novel “CD24^high/MET^high double-positive signature” of TNBC patients, which can predict a poor outcome of patients with TNBC.

In previous studies, enrichment of the CD44⁺/CD24^−/low and CD44⁻/CD24⁺ cell populations has been observed in basal-like and luminal breast cancer cell lines, respectively (37, 38), and CD44⁺/CD24^−/low cells were determined to be more invasive than CD44⁻/CD24^high cells (38). Moreover, several studies have reported that CD24 is a disfavored marker for BCICs, whose commonly known marker signature is CD44⁺/CD24⁻/ALDH1⁺ (39, 40). In this study, we determined that MDA-MB-231–derived LC cells, which were isolated from long-term metastatic lung nodules, exhibited high CD24 level and several epithelial-like phenotypic and molecular features (Fig. 1). Notably, we found that despite their decreased in vitro motility and invasiveness, CD44⁺/CD24⁺ LC cells were more potent in primary tumor growth, as well as LN and lung metastasis than the CD44⁺/CD24⁻ parental MDA-MB-231 cells (Supplementary Fig. S2). Although these findings seemingly contradict the established knowledge that CD24⁺ breast cancer cells exhibit a more differentiated and less tumorigenic phenotype, several reports have associated CD24 expression with tumor progression and metastatic behavior (41, 42). Furthermore, a comprehensive study of molecular and phenotypic analysis of CD24⁺ and CD44⁺ cells from breast carcinomas revealed that the number of CD24⁺ cells was dramatically and consistently increased in distant metastases regardless of the type of the primary tumor and sites of the distant metastasis (43).

In the CD24 KD xenograft mouse model, CD24 depletion greatly reduced the primary tumor growth, as well as LN and lung metastasis in both the IV2- and LC cells–derived tumor model (Fig. 2). We observed that LC/IV2 cells could express high levels of VEGF-A and VEGF-C to attract endothelial cells and lymphatic endothelial cells to the proximity of the primary tumor to form vascular and lymphovascular networks in the tumor microenvironment (Supplementary Fig. S8A–S8C).

Metastasizing process consists of multiple steps including local invasiveness, intravasation into blood or lymphatic vessels and extravasation at target organs, as well as survival and outgrowth at those organs, which are all very important for the metastasis. With respect to the role of CD24 in metastasis, we think it contributes to initial survival and subsequent outgrowth of the disseminated cancer cells in target organs likely via enhancing the angiogenesis and lymphoangiogenesis similar to that in the primary tumor. Our lung targeting assay revealed that CD24KD did not significantly affect the initial lung targeting (Supplementary Fig. S6) but greatly suppressed the later macroscopic lung metastases (Supplementary Fig. S5), suggesting that CD24 plays a more important role in lung colonization and outgrowth than in initial invasiveness and extravasation during dissemination of primary cancer cells. Collectively, out data suggested that CD24 plays an important role in primary tumor growth, as well as in outgrowth of disseminated cancer cells in the metastasizing organ such as lung during TNBC metastasis.

Another important role of CD24 in TNBC cell metastasis is likely its effect on EGF/HGF-induced angiogenesis- and lymphangiogenesis-related signaling pathways. We found that CD24 depletion in CD24-expressing TNBC cells attenuated the ligand-induced EGFR and Met downstream phosphorylation cascades (Fig. 3A and B). We determined that CD24 exerted its functions by stabilizing EGFR and Met at the protein level to sustain the EGF- and HGF-induced Src/Stat3 signaling, which is important for tumor angiogenesis (Fig. 3C and D; refs. 44, 45). Immunostaining analysis supported the results of Western blotting that depletion of CD24 could promote EGFR and Met internalization and lysosome-mediated degradation in the absence of a ligand (i.e., EGF and HGF, respectively), compared with the control cells (Fig. 4A and C). Furthermore, loss of CD24 expression accelerated EGFR and Met degradation upon ligand stimulation, as indicated by the increased colocalization of EGFR/Met and lysosomes (Fig. 4B and D).

Previous studies have reported that tumor-promoting cancer-associated fibroblasts and tumor-associated macrophages express high levels of HGF and EGF to enhance tumorigenesis (46, 47). Thus, the CD24-mediated enhancement of Met and EGFR pathways may play an important role in the context of the tumor-promoting microenvironment, and it may account for a large part of the biological functions affected by CD24. We propose that the presence of CD24 on the cell membrane may positively regulate EGFR/Met stability in TNBC (Fig. 6G).

Clinically, high levels of CD24 are most significantly associated with the shorter survival of TNBC patients compared with the patients with the HER2-positive and luminal subtypes (TNBC: P = 0.02; Luminal A: P = 0.03; Luminal B: P = 0.1212; HER2: P = 0.24, Fig. 6A). Analysis of the clinical relevance of coexpressing CD24 and EGFR/MET in TNBC patients showed that CD24^highMET^high was associated with poor 5-year survival (P = 0.193, Fig. 6E) and significantly increased the risk of recurrence within the first 5 years after initial treatment (P = 0.0043, Fig. 6E). In addition, Deng and colleagues reported that CD24 expression is significantly associated with docetaxel resistance in TNBC patients, adding further clinical relevance of CD24 to TNBC (48). Taken together, our clinical analyses indicate that CD24 alone is a good prognostic factor for TNBC patients, and in combination with MET expression, the CD24^highMET^high signature is an effective prognosis marker for TNBC patients.

Much effort has been devoted to identifying and validating potential markers for developing targeted therapy for TNBC patients. In this study, we explored the possibility of using CD24 mAb to treat the metastatic disease of CD24-positive TNBC in a tail vein xenograft mouse model. We demonstrated that CD24 mAb treatment significantly reduced lung tumor growth (Fig. 5B; Supplementary Fig. S10) and greatly prolonged the survival of mice (Fig. 5C and D). Our finding provides the first evidence to suggest the use of CD24 mAb as a targeted therapy for CD24-positive TNBC patients. In addition, luminal A patients with high CD24 expression also exhibited poor prognoses (Fig. 6A); therefore, the therapeutic efficacy of CD24 mAb could be tested for other CD24-positive breast cancer subtypes.

Overall, our study unveiled the important role of CD24 in TNBC progression and metastasis via a novel mechanism of CD24/Met/EGFR-mediated lymphangiogenesis and angiogenesis in the tumor microenvironment. CD24 may be a potential therapeutic target for CD24-positive TNBC and other breast cancer types.

No potential conflicts of interest were disclosed.

Conception and design: S.-H. Chan, L.-H. Wang

Development of methodology: S.-H. Chan, S.-Y. Chiu, L.-H. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.-H. Chan, S.-Y. Chiu, W.-H. Kuo, H.-Y. Chen, K.-J. Chang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.-H. Chan, K.-W. Tsai, S.-Y. Chiu, S.S. Jiang

Writing, review, and/or revision of the manuscript: S.-H. Chan, L.-H. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.-H. Chan, W.-C. Hung

Study supervision: L.-H. Wang

We thank Taiwan Bioinformatics Institute Core Facility for assistance in using Oncomine and NHRI Optical Biology Core Laboratory (NOBC) for confocal microscopy analysis. This work was financially supported by the "Chinese Medicine Research Center, China Medical University" from the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (to L.-H. Wang) in Taiwan (CMRC-CHM-7). This study was also supported by the grants MOST-105-2320-B-039-067 and MOST-104-2320-B-039-055-MY3 awarded by the Ministry of Science and Technology (to L.-H. Wang).

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.