Glycoprotein nmb Is Exposed on the Surface of Dormant Breast Cancer Cells and Induces Stem Cell–like Properties Free

Breast cancer is the most common cancer among women worldwide, which is a heterogeneous disease and can be classified into five molecular subtypes, luminal A, luminal B, HER2-enriched, basal-like, and normal-like, on the basis of the expression of estrogen receptors (ER), progesterone receptors (PR), and HER2 (1). The basal-like subtype is also referred to as triple-negative breast cancer (TNBC) because it is typically negative for ER, PR, and HER2. TNBC accounts for 10%–30% of all diagnosed breast cancers. In general, TNBC is associated with poor prognosis and high lethality. Moreover, the lack of effective molecularly targeted drugs limits the treatment options for this aggressive disease. Therefore, to develop novel molecularly targeted therapies for patients with TNBC, further molecular characterization of TNBC is required (2, 3).

Glycoprotein nmb (GPNMB) is a type I transmembrane protein. Abundant expression of GPNMB was observed in glioblastoma, melanoma, and breast cancer, especially in TNBC, and was reported as a prognostic factor (4–7). Moreover, the association between GPNMB- and HER2-positive breast cancers has also been reported (8). GPNMB is known to have functions in angiogenesis, tumorigenesis, cell migration, and cell invasion and metastasis, and has received much attention as a target molecule for cancer treatment (9–13).

We previously demonstrated that enhanced expression of GPNMB induces epithelial–mesenchymal transition (EMT) and increases anchorage-independent sphere formation and invasive tumor growth in vivo through the hemi-immunoreceptor tyrosine–based activation motif (hemITAM) in the intracellular domain. Furthermore, knockdown of GPNMB attenuated the tumorigenic abilities of TNBC cell lines (14). EMT is an essential process during embryogenesis, tissue repair, fibrosis, and cancer invasion and metastasis. In 2008, Mani and colleagues reported that the EMT-inducing transcription factors (EMT-TF) SNAIL and TWIST as well as TGFβ signaling are associated with acquisition of cancer stem cell (CSC) properties in breast epithelial cells (15). Evidence of CSC properties induced by EMT-TFs was also shown in breast cancer (16, 17). CSCs harbor the potential of self-renewal, differentiation, tumorigenesis, and resistance to drugs or radiation owing to their ability to enter dormancy and their abundant expression of drug exporters (18–20). For these reasons, CSCs are thought to be the root cause of cancer metastasis and relapse. Therefore, targeting EMT-related molecules might be a promising therapeutic target to eradicate CSCs.

In this study, we investigated whether GPNMB, which induces tumorigenesis and EMT in mammary epithelial cells, affects acquisition of CSC-like properties in breast cancer cells. Our findings propose a novel model in which cell surface expression of GPNMB induces stem cell–like properties through hemITAM in dormant breast cancer cells. In other words, cell surface expression of GPNMB could be a desirable marker and therapeutic target of breast cancer cells with stem cell–like properties.

Breast cancer cell lines BT-474, Hs578T, MDA-MD-468, and 4T1 were obtained from ATCC (14). In the two-dimensional (2D) monolayer cultures, cells were cultured in DMEM (Invitrogen) supplemented with 10% FBS (Gibco), 100 U/mL penicillin G, and 0.1 mg/mL streptomycin sulfate (Wako). In the 3D sphere cultures, cells were cultured in DMEM/F12 (1:1) medium (Invitrogen) with 2% B-27 supplement (Invitrogen), 2 ng/mL bFGF (Wako), 2 ng/mL EGF (Sigma), 100 U/mL penicillin G, and 0.1 mg/mL streptomycin sulfate (Wako) in ultralow attachment culture dishes (Corning) or poly(2-hydroxyethyl methacrylate) (Poly-HEMA; Sigma)-coated dishes. Hs578T cells were cultured in the presence of 10 μg/mL insulin in both the 2D and 3D culture conditions. NMuMG-mock, NMuMG-GPNMB, and NMuMG-GPNMB (YF) cells were established and maintained as described previously (14). Mycoplasma detection was performed using a Mycoplasma Detection Set (Takara) for all the cell lines.

4T1 cells were transfected with 40 nmol/L of siRNA directed against GPNMB or control siRNA (Invitrogen) using Lipofectamine 3000 (Invitrogen) following the manufacturer's recommendations. Stealth siRNAs (Invitrogen) were purchased as follows: siGpnmb #1 (MSS234870) and siGpnmb #2 (MSS294588). Control siRNA was purchased from Invitrogen (Stealth RNAi Negative Universal Control Medium).

Total RNA was isolated using Isogen II (Nippon Gene). The NucleoSpin RNA XS Kit (Macherey-Nagel) was used for extraction of total RNA from small samples after cell sorting. Reverse transcription was performed with High Capacity RNA-to-cDNA Master Mix (Applied Biosystems). mRNA levels were measured by qPCR with gene-specific primers (Supplementary Table S1) using SYBR Green I qPCR Master Mix (Applied Biosystems) on the ABI 7500 Fast Sequence Detection System. All samples were run in triplicate in each experiment.

Cells were suspended in 62.5 mmol/L Tris-HCl (pH 6.8) and solubilized in SDS sample buffer [10% glycerol, 5% 2-mercaptoethanol, 2% SDS, and 62.5 mmol/L Tris-HCl (pH 6.8)]. Western blot analysis was performed as described previously (14).

Immunofluorescence and IHC staining methods used were described previously (14, 21). Anti-GPNMB antibody (AF2550; R&D Systems) was used as the primary antibody, and cells were subsequently incubated with Alexa 488–labeled donkey anti-goat IgG (Invitrogen, Molecular Probes). Cell surface GPNMB proteins were detected without membrane permeabilization. For IHC, bound antibodies were detected using ImmPRESS Reagent Kit Peroxidase Anti-Goat IgG (Vector Laboratories).

Six-week-old female NOD-SCID mice (CLEA) were subcutaneously injected with 5 × 10⁶ Hs578T cells. Four weeks later, the mice were sacrificed and fresh tumors were obtained for RNA isolation or IHC staining as described above.

For tumor formation using 4T1 cells, 1 × 10⁶ 4T1 cells were injected subcutaneously into 6-week-old female Balb/c mice (CLEA). After 3–4 weeks, the mice were sacrificed, and the tumors were minced and dissociated under incubation conditions at 37°C in 1 mg/mL collagenase (Wako) for 2 hours, 0.25% trypsin (Sigma) for 5 minutes, and 0.1 mg/mL DNase I (Roche) and 5 mg/mL Dispase (Gibco) for 5 minutes. After centrifugation, single cells were obtained and the cell number was counted to perform FACS analysis and allograft transplantation for secondary tumor formation. After 33 days, the mice were sacrificed, and the secondary tumors were collected. The tumor weight was measured and the volumes were approximated using the following formula: volume = 0.5 × a × b², where a and b are the lengths of the major and minor axes, respectively.

All the animal experiments were performed with the approval of the Animal Ethics Committee of the University of Tsukuba (Ibaraki, Japan) and in accordance with the university's animal experiment guidelines and the provisions of the Declaration of Helsinki in 1995.

Two-dimensional–cultured cells were treated with trypsin (Sigma), and 3D-cultured spheres were dissociated using Accutase solution (Gibco). Single cells were incubated with anti-GPNMB antibody (AF2550; R&D Systems), and then with Alexa 488–labeled donkey anti-goat IgG (Invitrogen, Molecular Probes) on ice for 30 minutes. The samples were analyzed using BD FACSAria (BD Biosciences) and BD FACSDiva software (100 μm Sorp AriaII 5B 3R 3V 5YG; BD Biosciences). We set the gating to collect cell surface-GPNMB^low and -GPNMB^high cells using the cells incubated with Alexa 488–labeled donkey anti-goat IgG only as the negative control.

Extreme limiting dilution analysis (ELDA) was performed as described previously (22). In brief, the breast cancer cells cultured in 2D or 3D culture conditions and the cell surface-GPNMB^low and -GPNMB^high subpopulations from 3D-cultured spheres were seeded in a series of numbers from 200 cells/well to 1 cell/well in 200-μL sphere culture medium and cultured for 14 days. The number of wells containing spheres for each seeding cell number was counted and then analyzed using online ELDA software (http://bioinf.wehi.edu.au/software/elda).

Quantitative data were expressed as mean ± SEM. Statistical analyses were performed using the t test or one-way ANOVA with the Tukey multiple comparison test using GraphPad Prism 7 software. P < 0.05 was considered significant.

We enriched breast cancer stem cells (BCSC) using the 3D sphere culture method, which is used to examine the anchorage-independent growth potential (23). Three different human breast cancer cell lines, BT-474 (luminal type), Hs578T (basal type), and MDA-MB-468 (basal type), and a mouse breast cancer cell line, 4T1, were cultured in the 2D or 3D culture conditions and compared for the expression levels of GPNMB mRNA together with those of known CSC genes such as SOX2, NANOG, OCT4, CD44, CD133, and FOXO3. All of these mRNA levels were significantly higher in the 3D-cultured cells than in the 2D-cultured cells (Fig. 1A–D; Supplementary Fig. S1A–S1D). We also examined CD24 mRNA expression because BCSCs are often characterized by CD44^high/CD24^low populations in breast cancer cells (24, 25). The expression levels of CD44 mRNA were higher and those of CD24 mRNA were lower in the 3D-cultured cells than in the 2D-cultured cells (Fig. 1A–D, two panels from the right). In addition, we examined the sphere-forming frequencies of the 2D- and 3D-cultured Hs578T cells by ELDA (22). ELDA provides an estimate of sphere-forming frequency using one-sided confidence intervals and statistical analysis. The 3D cultures yielded significantly higher sphere-forming frequency than did the 2D cultures (Supplementary Fig. S1E).

Furthermore, we evaluated the expression of EMT-TF genes such as SNAIL, SLUG, and ZEB1 in BT-474, Hs578T, MDA-MB-468, and 4T1 cells. The expression levels of these EMT-TF mRNAs were also enhanced in the 3D-cultured cells as compared with the levels in the 2D-cultured cells (Fig. 2A–D). We also tested the expression of epithelial and mesenchymal marker genes including CDH1, CDH2, fibronectin, and vimentin in Hs578T and 4T1 as well. An epithelial marker, Cdh1, was detected only in 4T1 cells and downregulated in the 3D cultures. On the other hand, mesenchymal marker genes were enriched in the 3D-cultured both Hs578T and 4T1 cells (Supplementary Fig. S1F and S1G).

These results suggest that the 3D cultures can enrich the cells with CSC-like properties in both human and mouse breast cancer cell lines and that mRNA levels of GPNMB correlate with those of CSC genes, EMT-TF genes, and mesenchymal marker genes in this culture condition.

To examine the importance of GPNMB in the induction of CSC genes and EMT-TF genes, we knocked down Gpnmb and examined the expression levels of these genes in the 2D- and 3D-cultured 4T1 cells. Knockdown of Gpnmb significantly suppressed the induction of CSC genes such as Sox2, Nanog, and CD44, and of EMT-TF genes such as Snail, Slug, and Zeb1, in the 3D-cultured 4T1 cells (Supplementary Fig. S2A and S2B). Moreover, knockdown of Gpnmb reduced the sphere-forming frequency of 4T1 cells (Supplementary Fig. S2C). Therefore, GPNMB is critical for induction of CSC-like properties in the 3D-cultured breast cancer cells.

We next evaluated the proliferative states of the cells in different culture conditions and in in vivo tumorigenic condition because one of the characteristics of CSCs is their slow proliferation or dormancy. MKI67-positive and -negative Hs578T cell numbers were counted in the 2D and 3D cultures by using immunofluorescent staining and in xenograft tumors by using IHC staining. The MKI67-positive and -negative ratios in each condition are shown in Fig. 3A. More than 97% of the 2D-cultured cells were MKI67-positive. However, nearly 80% of the 3D-cultured spheres and nearly 50% of the xenograft tumors were MKI67-negative. In addition, abundance of MKI67 mRNA was significantly higher in the 2D-cultured cells than in the 3D-cultured spheres or xenograft tumors (Fig. 3B). Interestingly, SOX2 and GPNMB mRNA expression was enriched in the spheres and tumors, and highly correlated with the MKI67-negative cell ratios, suggesting that dormant cells may have higher expression levels of SOX2 and GPNMB (Fig. 3C). To confirm whether growth arrest affects the expression of CSC genes and GPNMB, we induced growth arrest by serum starvation for 48 hours under subconfluent or confluent 2D culture conditions. Quantitative data obtained by qPCR showed that the mRNA expression levels of CSC genes SOX2 and NANOG as well as GPNMB were increased in serum-free growth-arrested conditions, especially under confluent cell density (Fig. 3D). On the other hand, GPNMB protein was relatively abundant even in the 2D-cultured Hs578T cells without starvation (Fig. 3E; Supplementary Fig. S3A) when compared with the significantly higher induction of GPNMB mRNA in the 3D cultures than in the 2D cultures (Fig. 1B, left). However, flow cytometry and immunofluorescent staining with/without permeabilization revealed that less than 1% of the 2D-cultured Hs578T cells had only low levels of cell surface GPNMB (Fig. 3F), whereas nearly 10% of the cells were cell surface GPNMB-positive in the 3D-cultured Hs578T cells (Fig. 3G; Supplementary Fig. S3B). Similar regulation of GPNMB mRNA expression and cell surface protein exposure were confirmed with other breast cancer cell lines, BT-474 and MDA-MB-468. These results suggest that growth-arrested conditions such as those after serum starvation for confluent 2D cultures or 3D cultures can activate the sorting of GPNMB protein for its exposure on the cell surface together with enhanced GPNMB transcription and CSC genes' induction.

To characterize the cell surface GPNMB-positive breast cancer cells, we isolated cell surface-GPNMB^low and -GPNMB^high cell populations from the 3D-cultured Hs578T spheres by FACS, and the sphere-forming frequencies were compared in these two populations by use of ELDA together with cell proliferation and CSC genes expression by qPCR (Fig. 4A). ELDA determined that approximately 1 in 76 cell surface-GPNMB^high cells harbored sphere-forming stem cell potential, while the cell surface-GPNMB^low cells had much less frequency (about 1 in 1,955; Fig. 4B). These results indicate significant differences in sphere-forming frequencies between the populations of the cell surface-GPNMB^low and -GPNMB^high cells. We further compared the mRNA abundance of GPNMB and CSC genes SOX2 and NANOG in the 2D-cultured cells and in the cell surface-GPNMB^low and -GPNMB^high cells of 3D-cultured spheres. Both the cell surface-GPNMB^low and cell surface-GPNMB^high cells had higher expression levels of GPNMB, SOX2, and NANOG than did the 2D-cultured cells. Obviously, the cell surface-GPNMB^high population had much higher SOX2 and NANOG than did the cell surface-GPNMB^low population. However, the GPNMB mRNA expression levels did not differ significantly between the cell surface-GPNMB^low and -GPNMB^high cells (Fig. 4C). Moreover, the expression levels of EMT-TF genes SNAIL and SLUG, EMT markers CDH2, fibronectin, and vimentin, and the proliferation marker genes MKI67 and TPX2 (26) were examined subsequently. The expression levels of SNAIL and SLUG (Fig. 4D) as well as of CDH2, fibronectin, and vimentin (Supplementary Fig. S4A) were highly correlated with those of SOX2 and NANOG. In contrast, the 2D-cultured cells had the highest expression levels of proliferation marker genes, while the cell surface-GPNMB^high cells showed no detectable levels of MKI67 or TPX2 (Fig. 4E).

In addition, we confirmed a similar phenomenon in a mouse breast cancer cell line 4T1. ELDA results showed that the cell surface-GPNMB^high cells had higher sphere-forming potential, while the cell surface-GPNMB^low cells had less frequency (Fig. 5A). Next, we compared the mRNA abundance of Gpnmb, Sox2, and Nanog in the 2D-cultured cells and in the cell surface-GPNMB^low and -GPNMB^high cells of 3D-cultured spheres. The expression levels of Gpnmb, Sox2, and Nanog in the cell surface-GPNMB^high cells were significantly higher than those in the 2D-cultured cells or the cell surface-GPNMB^low cells (Fig. 5B). Moreover, the cell surface-GPNMB^high cells showed enhanced expression levels of Snail, Slug (Fig. 5C), fibronectin, and vimentin (Supplementary Fig. S4B, middle and right), whereas they showed the lowest levels of Mki67 (Fig. 5D) and Cdh1 (Supplementary Fig. S4B, left). Taken together, these results suggest that the cell surface-GPNMB^high cells have the properties of dormant CSCs in the 3D-clutured both human and mouse breast cancer cell lines.

We further examined the characteristics of the cell surface-GPNMB^low and -GPNMB^high cells using a 4T1 allograft tumor model. We compared the mRNA abundance of Gpnmb, Sox2, and Mki67 in the 2D or 3D-cultured 4T1 cells and 4T1 tumors. The 4T1 tumors had lower expression levels of Gpnmb and Sox2 than did the 3D-cultured spheres, but those levels were still more enriched than those of the 2D-cultured cells (Supplementary Fig. S5A). Mki67 expression levels were lower in the 3D-cultured cells or tumors than in the 2D-cultured cells, similar to Hs578T tumors (Supplementary Fig. S5B). After harvesting 4T1 tumors, we collected isolated single cells from the tumors and sorted cell surface-GPNMB^low and -GPNMB^high cells by FACS. ELDA showed that the cell surface-GPNMB^high cells harbored higher sphere-forming frequency (about 1 stem cell in 29 cells), while the cell surface-GPNMB^low cells had much less sphere-forming frequency (about 1 in 1,065; Fig. 6A). Moreover, the cell surface-GPNMB^high tumor cells showed the highest expression levels of Gpnmb, Sox2, Nanog (Fig. 6B), Snail, and Slug (Fig. 6C), and the lowest expression level of Mki67 mRNA (Fig. 6D). These results indicated that like 3D-cultured spheres, the tumors grown in vivo are also composed of cell surface-GPNMB^low and -GPNMB^high cells and that the cell surface-GPNMB^high cells have dormant CSC-like properties. Furthermore, we tested the secondary tumor growth of the cell surface-GPNMB^low and -GPNMB^high 4T1 tumor cells. Injected cell surface-GPNMB^high cells generated secondary tumors with a higher incidence rate (10³, 3/5; 10⁴, 5/5) than that of the cell surface-GPNMB^low cells (10³, 1/5; 10⁴, 1/5; Fig. 6E and F; Supplementary Fig. S5C and S5D). These results clearly indicated that the cell surface-GPNMB^high cells have higher secondary tumor-forming frequency. In addition, further analysis of the tumors generated by the cell surface-GPNMB^low cells indicated that these tumors were also composed of a comparable ratio of cell surface-GPNMB^high and -GPNMB^low cells, with the tumor made from the cell surface-GPNMB^high cells, suggesting that the plasticity of cancer cells generate cell surface-GPNMB^high cells from the cell surface-GPNMB^low cells at low frequency and causes the tumorigenic potential (Fig. 6G and H).

To elucidate the molecular mechanism of GPNMB in generation of CSC-like properties, we examined the involvement of the tyrosine residue in hemITAM, because we have previously demonstrated that Tyr⁵²⁹ is essential for GPNMB-inducible EMT and tumorigenesis. We used NMuMG-mock, NMuMG-GPNMB, and NMuMG-GPNMB(YF) cell lines, in which Tyr⁵²⁹ is replaced by a phenylalanine (14). Exogenous Gpnmb mRNA was highly expressed in both NMuMG-GPNMB and NMuMG-GPNMB(YF) cells cultured in the 2D and 3D culture conditions (Fig. 7A, left; Supplementary Table S2, top), but the expression of the CSC genes Sox2 and Nanog was induced in NMuMG-GPNMB cells cultured in the 3D culture condition only. On the other hand, GPNMB(YF) cells failed to induce the expression of Sox2 and Nanog mRNA, even in the 3D culture condition (Fig. 7A, middle and right; Supplementary Table S2, middle and bottom). These results indicate that GPNMB function mediated by the tyrosine residue in hemITAM is essential for the induction of CSC genes in the 3D cultures.

In this article, we have reported a novel function of GPNMB, which is exposed on the surface of growth-arrested breast cancer cells and induces stem cell–like properties through hemITAM (Fig. 7B).

We previously demonstrated that GPNMB can induce EMT in NMuMG cells (14). The correlation between EMT and the acquisition of stem cell-like properties has been proven in breast epithelial cells (15), and EMT-TFs, including TWIST, SLUG, SNAIL, ZEB1, and ZEB2, have been shown to confer the characteristics of CSCs (15, 27, 28). In addition, Chen and colleagues showed that 3D-cultured cells had higher expression of the EMT-TF genes SNAIL and TWIST than did 2D-cultured cells in head and neck squamous cell carcinoma cell lines (29). The 3D culture system is thought to mimic the cellular dynamics in the body better than the 2D culture system (30–32) and is used to enrich cells with stem cell potential in cancers of the breast (33, 34), brain (35), and head and neck (29). In our 3D cultures, the spheres had enhanced expression levels of CSC genes as well as of EMT-TF genes and altered those of EMT marker genes (Figs. 1, 2, 4C and D, 5B and C, and 7A; Supplementary Fig. S1A–S1D, S1F, S1G, and S4). Importantly, similar results were observed in the xenograft and allograft tumors (Figs. 3C, 6B and C; Supplementary Fig. S5A). Taken together, these results indicate that the 3D culture method is a useful in vitro system to examine CSC-like properties.

CSCs are found in acute myeloid leukemia (36) and solid tumors including those of the breast (24), brain (37), prostate (38), colon (39), pancreas (40), and lung (41). Because BCSCs were isolated as cells with the markers of CD44⁺/CD24^−/low/Lin⁻ (24), CD44⁺/CD24⁻ is frequently used as a BCSC marker (25). Other surface markers, such as CD133, CD49f, and CD61, have also been reported (42–45). Whereas GPNMB was located mainly on lysosome or endosome membranes in 2D-cultured cells, it was exposed on the cell surface of dormant BCSCs and induced stem cell-like properties in 3D-cultured spheres and in vivo tumors (Figs. 3F and G, 4B–E, 5, and 6). Our results indicate that surface expression of GPNMB could be a novel indicator of dormant BCSCs. However, the frequency with which stem cells were enriched with cell surface-GPNMB^high cells in the 3D-cultured spheres or tumors was not 100% (Figs. 4B, 5A, and 6A). Therefore, investigation of possible and more suitable combinations of BCSC markers might be required for further characterization of breast cancer stem-like cells.

Although we have not elucidated the mechanism whereby GPNMB is exposed on the surface of dormant cells, we have here shown that growth-arrested conditions, such as serum starvation or 3D culture conditions, induced transcriptional activation of GPNMB and total GPNMB protein (Fig. 3D–G; Supplementary Fig. S3). In addition, GPNMB is exposed on the cell surface only in MKI67-negative growth-arrested cells (Figs. 4E, 5D, and 6D). Moreover, cell growth analyses shown in Fig. 3A indicated that the 2D cultures mostly composed of monotonous MKI67-positive proliferating cells. On the other hand, the 3D cultures and in vivo tumors contain both growth-arrested cells and proliferating cells. This divergence in cell proliferation status is thought to make the critical difference in 2D-cultured cells and 3D-cultured spheres or in vivo tumors. Qian and colleagues have reported that inhibition of the ERK signaling pathway enhances GPNMB protein expression in several melanoma cell lines that have NRAS or BRAF mutations (46). Taken together, these findings suggest that growth inhibition may enhance GPNMB transcriptional activation and protein sorting to the cell surface, and leads to maintenance of CSC-like phenotypes.

GPNMB contains hemITAM (YxxI) and a dileucine motif (D/ExxLL) in its cytoplasmic tail. These motifs are frequently found in transmembrane proteins and function as sorting signals associated with endocytosis or endosomal/lysosomal membrane trafficking (47). Our previous study demonstrated that hemITAM is essential for inducing EMT and tumorigenesis via phosphorylation of the tyrosine residue by SRC kinase (14). We have here further shown that it is involved in the induction of CSC genes in the 3D cultures (Fig. 7A). The loss tumorigenicity made it difficult to analyze the gene expression levels of mutant hemITAM-expressing tumors in vivo. Moreover, Lin and colleagues reported that cell surface GPNMB forms a heterodimer with EGFR and that the tyrosine residue in hemITAM is phosphorylated upon heparin-binding EGF (HB-EGF) stimulation. Subsequently, the BRK– or LRRK2–LINK-A complex binds to GPNMB and stabilizes normoxic HIF1α. They also showed that phosphorylation of GPNMB is detected in TNBC and correlates with the metastatic status of patients with breast cancer (48). We have also shown that EGFR and FGFR1 were enriched in the 3D-cultured Hs578T and 4T1 cells (Supplementary Fig. S6A and S6B). Therefore, it is worthwhile examining the cross-talk between tyrosine receptor kinase signaling and GPNMB in detail. These results together with our current observations suggest that both cell surface expression of GPNMB and phosphorylation of the tyrosine residue in hemITAM are essential for the GPNMB-inducible tumorigenic and metastatic potential, at least partly owing to the generation of CSC-like properties.

GPNMB is known to have a soluble form shed by ADAM10 (49), and it could be detected in patients' blood (8). Therefore, shedding of GPNMB is thought to happen on the cell surface. So far we know of no relationship among GPNMB cell surface localization, hemITAM tyrosine phosphorylation, and shedding of its extracellular domain and it would be interesting to study these points as a serial phenomenon to understand the physiologic and pathologic functions of GPNMB.

CDX-011 (glembatumumab vedotin) is an antibody–drug conjugate that contains CR011, a human mAb against GPNMB, and conjugates with a cytotoxic agent (MMAE). CDX-011 has been developed for the treatment of GPNMB-expressing cancers, and clinical studies of CDX-011 against patients with breast cancer and melanoma were conducted (12, 13, 50). We have shown that the cell surface-GPNMB^high cells have higher stem cell–like properties than do the cell surface-GPNMB^low cells (Fig. 7B). Thus, our novel findings lead us to propose that targeting GPNMB, which is exposed on the surface of dormant cancer cells, has better feasibility to kill CSCs and might result in a highly efficient cancer treatment.

No potential conflicts of interest were disclosed.

Conception and design: C. Chen, Y. Okita, M. Kato

Development of methodology: C. Chen, Y. Okita, F. Abe, M. Kato

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Chen, Y. Okita, Y. Watanabe, F. Abe, M.A. Fikry, Y. Ichikawa, A. Shibuya, M. Kato

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Chen, Y. Okita, F. Abe, M. Kato

Writing, review, and/or revision of the manuscript: C. Chen, Y. Okita, H. Suzuki, M. Kato

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Suzuki

Study supervision: Y. Okita, M. Kato

We thank F. Miyamasu (Medical English Communication Center, University of Tsukuba, Ibaraki, Japan) for proofreading the manuscript. This research was supported by JSPS Grant Numbers JP15K19070, JP17K14981, and JP18H02676 and grants from the Naito Foundation and the Uehara Memorial Foundation.

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.