GD2 enriches for breast CSCs. We recently reported that, following the induction of EMT, human mammary epithelial cells show functional properties similar to those of human bone marrowderived MSCs (13). Therefore, we hypothesized that the cell markers expressed on the surface of MSCs could also be expressed on the surface of breast CSCs. To test this hypothesis, we analyzed for the expression of several known MSC cell surface markers (i.e., CD105, CD90, CD106, CD166, CD73, CD271, MSCA-1, and GD2) on HMECs that had been experimentally transformed to become tumorigenic using oncogenic V12-H-Ras (HMLER cells) (21). Absolute expression of most of the markers analyzed could not divide HMLER cells into two distinct subpopulations (Supplemental Figure 1; supplemental material available online with this article; doi:10.1172/JCI59735DS1), similar to CD44hiCD24lo cells (12). However, ganglioside GD2, one of the cell surface markers for MSCs, was able to separate HMLER cells into GD2+ (4.5% 2.4%) and GD2 (92.7% 3.8%) populations (Figure 1A and Supplemental Figure 1). Strikingly, GD2+ HMLER cells isolated using FACS appeared spindle-shaped, with limited cell-cell contacts; conversely, the GD2 cells displayed cobblestone epithelial morphology (Figure 1B). Moreover, the GD2+ HMLER cells proliferated approximately 5-fold slower than the GD2 HMLER cells (Figure 1C).

GD2 identifies CSCs in breast cancer. (A) HMLER cells were stained with anti-GD2 antibody by indirect staining and analyzed on an LSR II flow cytometer. GD2+/ gates were drawn based on IgG2a isotype control. FSC, forward scatter. (B) GD2+/ HMLER cells were cell sorted and cultured in cell culture dishes for 4 days. Scale bars: 50 m. (C) 2 104 GD2+/ HMLER cells were cultured in 6-well cell culture dishes in triplicate. Total cells were counted on days 2, 4, and 6 using a Vi-CELL (Beckman Coulter) cell counter. (D) HMLER or MDA-MB-231 cells (1 103) were sorted into each well of 24-well ultra-low attachment dishes containing mammosphere growth medium using the FACSAria II cell sorter. Cells were cultured for 12 days, and the photos were taken using a light microscope. Scale bars: 100 m. (E and F) Number of mammospheres formed from GD2+/ HMLER (E) and MDA-MB-231 (F) cells. The experiment was performed in triplicate. P < 0.01 (G) GD2+/ MDA-MB-231 cells were sorted (1 cell or 5 cells/well) into 96-well ultra-low-attachment dishes containing mammosphere growth medium. Cells were cultured for 12 days, and mammospheres were counted using a light microscope. Scale bars: 200 m. (H) Number of mammospheres formed from single GD2+/ MDA-MB-231 cells. *P < 0.002. (I) Size of mammospheres measured using a hemocytometer. *P < 0.0001.

To further investigate the functional properties of GD2+ and GD2 cells, we sorted HMLER and MDA-MB-231 cells based on GD2 expression and examined them by mammosphere assay. Interestingly, the GD2+ cells from HMLER and MDA-MB-231 cells formed 2-fold more mammospheres compared with GD2 cells (Figure 1, DF, P < 0.01). Direct sorting of GD2+ and GD2 MDA-MB-231 cells into low-attachment 96-well plates at either 1 or 5 cells per well also resulted in a 2-fold increase in sphere formation by GD2+ cells regardless of the number of cells per well compared with GD2 cells (Figure 1, G and H). In addition, the mammospheres generated by GD2+ cells were 3 times larger than those generated by GD2 cells (Figure 1, G and I), indicating that the GD2+ cells are capable of growing better in suspension cultures.

CSCs are known to be more migratory and invasive (1, 3). To examine the migration and invasion potential of GD2+/ cells, we fractionated HMLER cells into GD2+ and GD2 cells and analyzed them for migration and invasion using Boyden chamber Matrigel invasion assays. After 24 hours of incubation, GD2+ HMLER cells migrated to a more than 4-fold greater extent compared with GD2 cells, indicating that GD2+ cells are highly migratory (Supplemental Figure 2). The hallmark of CSCs is their ability to initiate tumor better than their bulk tumor counterparts (1, 2). To determine the tumor-initiating potential of GD2+ cells, we sorted GD2+ and GD2 MDA-MB-231 cells and transplanted them subcutaneously into the flank of NOD/SCID mice at limiting dilutions. At lower cell numbers including 100 or 10 cells/site, the GD2+ cells generated 2- and 5-fold more tumors, respectively, compared with the GD2 fraction (Table 1). However, at higher cell numbers (10,000 or 1,000 cells/site), there were no significant differences in tumor initiation between GD2+ and GD2 cells. These data firmly established that GD2 is a marker of cells capable of initiating tumors at a higher frequency than cells without GD2.

Generation of tumors by GD2+/ cells in vivo

Percentage of GD2+ cells is highest in cell lines with a basal molecular signature. On the basis of gene expression profile (22), breast cancer cell lines have been classified into 3 groups: luminal, basal A, and basal B. We randomly selected 12 breast cancer cell lines representing these 3 subgroups and analyzed them for GD2 expression. Interestingly, the majority of these lines, independent of the subgroup, contained a subpopulation of GD2+ cells at variable levels (Table 2). However, basal cell lines contained a much greater number (mean 9%, range 1.2%17%, n = 6) of GD2+ cells compared with luminal cell lines (median 0.2%, range 03%, n = 6, Table 1, P = 0.00237). Since basal-derived cell lines show greater tumor initiation potential and contain more CSCs based on the previously reported CD44hiCD24lo profiles (23), this finding once again confirms GD2 as a stem cell marker.

Expression of GD2 in breast cancer cell lines

GD2 identifies the CD44hiCD24lo population in breast cancer cell lines and patient samples. Since we found that GD2, similar to previously reported CD44 and CD24 cell surface markers, is capable of separating cancer cells into two populations with differing tumor-initiating potential (7), we hypothesized that GD2 would be mostly expressed in the CD44hiCD24lo cancer cell fraction. To test this, we initially analyzed the expression of CD44hiCD24lo cells in GD2+ HMLER cells and found that more than 85% (85% 3.5%) of GD2+ HMLER cells also displayed a CD44hiCD24lo CSC profile, whereas less than 1% (0.7% 0.2%) of GD2 HMLER cells were CD44hiCD24lo (Figure 2A). In addition, through reverse gating analysis of CD44hiCD24lo HMLER cells, we noted that more than 84% (84% 2.5%) of CD44hiCD24lo HMLER cells were also positive for GD2 (Figure 2B), whereas less than 5% of CD44loCD24hi HMLER cells were GD2+ (4.3% 1.2%). To further determine the correlation between the expression of GD2 and the CD44/CD24 profiles, we sequentially gated HMLER cells into GD2hi, GD2lo, and GD2neg cells. This analysis revealed that GD2 expression levels correlated strongly with the CD44hiCD24lo phenotype (Supplemental Figure 3A). Moreover, by determining the MFI, we found that GD2 expression levels correlated positively with CD44 expression (correlation index, r2 = 0.85; P < 0.0003; Supplemental Figure 3B). To validate the coexpression of GD2 on CD44hiCD24lo cells, we used anti-GD2 antibody from a different source (Abcam, clone 2Q549) to stain HMLER cells in a 4-step staining procedure as explained before (24), along with anti-CD44 and anti-CD24 antibodies. Analysis of GD2+ cells revealed that these cells coexpress CD44hiCD24lo, confirming our initial findings with the 14G2a clone (Supplemental Figure 4).

GD2 identifies CD44hiCD24lo stem cell phenotype in breast cancer cells. (A) HMLER cells were stained with anti-GD2 antibody and with CD44-APC and CD24-FITC using the 4-step staining protocol described in Methods. Cells were electrically gated on GD2+/ cells and displayed in a pseudocolor dot plot with CD44 on the y axis and CD24 on the x axis using FlowJo data analysis software. (B) In an identical experiment, CD44hi/loCD24lo/hi cells were displayed on a pseudocolor dot plot with GD2 on the y axis and FSC on the x axis. (C) Primary breast tumor samples were processed as described in Methods, and the single cells in suspension were stained with anti-GD2, CD44-APC, CD24-FITC, CD45-FITC, and DAPI using the 4-step staining protocol. Cells were initially gated on DAPI-negative cells to exclude dead cells, and the cells were then gated on CD45 cells to exclude hematopoietic cells. GD2+CD45 cells were displayed on a dot plot, with CD44 on the y axis and CD24 on the x axis. Analysis was perfumed using an LSR II flow cytometer. Data were analyzed using FlowJo software.

To further investigate the correlation between GD2 expression and the CD44hiCD24lo profile, we also analyzed primary breast tumor samples (n = 12, Table 3). Using multi-parameter flow cytometry, we excluded CD45+ inflammatory and other hematopoietic cells from dissociated tumor samples. The non-hematopoietic CD45 fraction was then analyzed for the expression of GD2, CD44, and CD24. This analysis of the CD45 fraction revealed that GD2 was expressed, at variable levels from 0.5% to 35% (median 4.35%, range 0.5%35.8%), in tumor samples (Table 3). Importantly, similar to what we observed in cell lines, more than 95.5% (95.5% 2.7%) of GD2+CD45 tumor cells also co-segregated with the CD44hiCD24lo phenotype (Figure 2C). In contrast, only 2.4% (2.4% 0.4%) of GD2CD45 cells exhibited the CD44hiCD24lo phenotype (Figure 2C). Together these findings clearly indicated that GD2 is a marker of a subset of cancer cells with stem cell properties. To validate that the identified GD2+ cells are in fact tumor and not MSCs, we stained human breast tumor tissues with anti-GD2 and epithelial-specific antipan-cytokeratin antibodies. We found coexpression of GD2 and cytokeratin in some of the breast cancer cells, suggesting that GD2 identifies breast tumors cells (Supplemental Figure 5, A and B).

Breast cancer patient samples analyzed: patient number, tumor type, percentage of GD2+ cells

GD2+ and CD44hiCD24lo cells have similar gene signatures. Since we found that GD2 is capable of independently enriching for CSCs as a single marker compared with the previously known double marker CD44hiCD24lo, we compared the global gene expression profiles in these two populations isolated from HMLER cells using microarray analysis. We initially compared the GD2+ fraction with the GD2 fraction of cells (GD2 set) and the CD44hiCD24lo with the CD4loCD24hi fraction (CD44 set) and identified gene signatures specific to the GD2+ and CD44hiCD24lo fractions (Figure 3A). Comparison of the top 600 differentially expressed genes in the GD2 set (GEO GSE36643) and the CD44 set (GSE36643) identified 231 genes as being identical in the two sets (Supplemental Table 1). In addition, we applied Pearsons 2 test with a Yates continuity correction to assess the association between these two cell types and found that the identified 231 genes correlated (100%) between the two groups described above (Figure 3B). This gene expression analysis along with the cell surface protein analysis shown in Figure 2 indicated that GD2+ cells share not only functional properties but also a gene signature with CD44hiCD24lo cells.

GD2+ and CD44hiCD24lo cells have a similar gene signature. (A) Heat maps derived from microarray analysis of CD44hi/loCD24lo/hi and GD2+/ populations of HMLER cells. (B) Two hundred thirty-one genes of the top 600 differentially expressed genes were identical in GD2+ versus GD and CD44hiCD24lo versus CD44loCD24hi groups. These genes were cross-classified in a 2-by-2 table by GD2+ up-/downregulation and CD44hiCD24lo up-/downregulation. Pearsons 2 test with a Yates continuity correction was applied to assess the association. Statistical significance was assessed at the 0.05 level. (C) Biosynthesis reaction of GD2. (D and E) To measure the expression of GD2S/GD3S mRNA, CD44hiCD24lo or CD44loCD24hi and GD2+/ cells from HMLER (D) or MDA-MB-231 cells (E) were FACS sorted, and mRNA was analyzed using qRT-PCR. *P < 0.001.

Among the genes differentially expressed between GD2+ and GD2 populations, GD3S, a key enzyme involved in the biosynthesis of GD3 (an intermediate for GD2, Figure 3C), was found to be upregulated approximately 9-fold in GD2+ compared to GD2 cells (Supplemental Table 2). The microarray data were validated by qRT-PCR (Figure 3D). However, expression of the gene encoding GD2S, which is involved in conversion of GD3 to GD2, was not altered (Figure 3D). Expression of a number of genes involved in migration and invasion, including MMPs (MMP2, MMP7, and MMP19), and EMT-associated markers, including N-cadherin and vimentin, were expressed at higher levels, whereas E-cadherin was expressed at low levels in GD2+ cells (Supplemental Table 2). We confirmed these findings by qRT-PCR (Supplemental Figure 6). In addition, CD44 mRNA was upregulated and CD24 mRNA downregulated in GD2+ relative to GD2 cells, which was confirmed by FACS analysis (Figure 2A). In addition, the stem cell marker nestin was also found to be upregulated in GD2+ cells compared with GD2 cells (Supplemental Figure 7). These and other genes that were differentially expressed in GD2+ versus GD2 cells are listed in Supplemental Table 2. Conversely, as in GD2+ versus GD2 cells, GD3S was overexpressed more than 10-fold in CD44hiCD24lo compared with CD4loCD24hi cells (Supplemental Figure 8), but no significant difference was found in the expression of GD2S between CD44hiCD24lo and CD4loCD24hi cells. This again demonstrates that the expression of GD3S and GD2 strongly correlates with the CD44hiCD24lo phenotype. Similar to HMLER cells, GD2+ cells from MDA-MB-231 cells expressed GD3S at a more than 5-fold-higher level than GD cells, and consistent with our earlier finding, no significant differences in GD2S expression were observed (Figure 3E).

GD2 cells can spontaneously generate GD2+ cells. Since we observed only a 2-fold difference in mammosphere formation and a 2- to 5-fold difference in tumor initiation between GD2+ and GD2 populations, we investigated whether this was due to the generation of GD2+ cells from GD2 cells. In fact, GD2+ and GD2 cells were sorted from HMLER (Figure 4A) and MDA-MB-231 cells (Figure 4B) and cultured in vitro for 12 days in their respective growth media. Surprisingly, approximately 10% of GD2+ HMLER cells had become GD2, and 15% of GD2 cells had spontaneously generated GD2+ cells, and this proportion was almost identical to that in the unfractionated original HMLER cells (Figure 4A). Similarly, the GD2+ and GD2 cells from MDA-MB-231 cells also generated 81% (81% 2.5%) of GD2 and 12% of GD2+ cells, respectively, again reflecting the percentage of GD2+ cells within the parental MDA-MB-231 cell composition. To investigate the generation of GD2+ cells from GD2 cells and vice versa in vivo, GFP-labeled MDA-MB-231 cells were sorted into GD2+ and GD2 fractions, and 1 106 GD2+ and GD2 cells (GD2+/ cells) were subcutaneously transplanted into NOD/SCID mice. Four weeks later the tumors were dissected, and single-cell suspensions were prepared as described in Methods. The cells were then stained with anti-GD2 antibody and analyzed by flow cytometry. Tumors generated by GD2+ cells consisted of nearly 91% 4.5% GD2 cells, whereas 2.4% 1.1% of cells in GD2 derived tumors were positive for GD2. These findings indicate that GD2+ cells can spontaneously achieve a GD2 phenotype and vice versa in vivo (Supplemental Figure 9, A and B).

GD2-depleted cells are able to generate GD2+ cells in culture: a possible role of EMT. (A and B) GD2+/ cells from HMLER (A) MDA-MB-231 (B) cells were FACS sorted and cultured in MEGM medium for 10 days. After incubation, the cells were stained with GD2 antibody (BD) and analyzed on an LSR II flow cytometer. Note the regeneration of GD+ cells in a GD2-depleted population. (C) To determine the possible role of EMT, HMLER cells transduced with two known EMT inducers (Twist and Snail) were stained with anti-GD2 antibody and analyzed on an LSR II flow cytometer. Expression of GD2 is shown on the y axis and FSC on the x axis. Lower panels in A and C represent antibody staining controls without primary antibody for each cell line. (D) Graphic representation of percentage of GD2+ vector control or Twist- or Snail-transduced HMLER cells. (E and F) mRNA expression analysis of GD3S (E) and GD2S (F) in vector- or Twist- or Snail-transduced cells was performed by real-time TaqMan qRT-PCR. *P < 0.001.

Induction of EMT generates GD2+ cells. Since we recently reported that the induction of EMT in HMLER cells results in the acquisition of stem cell properties (12), we also examined the expression of GD2 on HMLER cells induced to undergo EMT by the ectopic expression of either Twist or Snail. Strikingly, we found that the induction of EMT by Snail or Twist resulted in a significant increase in the percentage of GD2+ populations from the initial 18% (control) to 40% in HMLER-Snail cells and 100% in HMLER-Twist cells (Figure 4, C and D). Corroborating our previous data suggesting a correlation between GD3S and CSCs, we also found that the expression of GD3S mRNA increased in the EMT-derived HMLER cells following induction of EMT by 2.5-fold in Snail cells and 8-fold in Twist cells (Figure 4E), which correlates with the total percentage of GD2+ cells in their respective population (40% in Snail and 100% in Twist cells). In contrast, we found no significant difference in the expression of GD2S (Figure 4F), supporting the hypothesis that GD3S is the key regulator in the biosynthesis of GD2.

GD3S is necessary for CSC properties. To investigate the functional role of GD2 in CSCs, we suppressed the expression of GD3S, the critical enzyme involved in the biosynthesis of GD2, in MDA-MB-231 cells using a lentiviral-based shRNA expression vector and achieved more than 80% knockdown (Figure 5A). As expected, GD3S knockdown reduced the percentage of GD2+ cells from 12.3% (12.3% 1.7%) to 5.5% (5.5% 0.8%) in MDA-MB-231 cells (Figure 5, B and C). Since GD3S is known to regulate a-series gangliosides including GM3, we tested whether knockdown of GD3S could induce the expression of GM3 in MDA-MB-231 cells. Flow cytometric analysis revealed that expression of GM3 was increased from 0.4% 0.3% (control cells) to 15% 1.4% (in GD3S knockdown [GD3S-KD] cells), suggesting that knockdown of GD3S was efficient in these cells (Supplemental Figure 10, A and B). In addition, functional analysis revealed that GD3S-KD-MDA-MB-231 (GD3S-KD-MDA231) cells migrated approximately 3-fold less in Transwell Matrigel invasion assays (Figure 5D) and formed 3-fold fewer mammospheres compared with controls (Figure 5E). To further investigate the effects of suppression of GD3S on tumor formation, we subcutaneously injected MDA-MB-231cells expressing either control shRNA or the GD3S shRNA into the flank of NOD/SCID mice. Strikingly, even after 8 weeks, 1 106 GD3S shRNA cells had not formed tumors, whereas the control shRNA cells had formed tumors in 4 of 4 mice (Figure 5F). The growth rate (tumor size) was also dramatically altered, as plotted in Figure 5G.

Knockdown of GD3S reduces cell proliferation, mammosphere formation, and tumor initiation in MDA-MB-231 cells. (A) To measure knockdown of GD3S, vector control, or GD3S-KD-MDA231 cells were analyzed for mRNA expression of GD3S by real-time TaqMan RT-PCR. Relative expression of GD3S is shown. *P < 0.0001. (B) To measure levels of GD2 on the cell surface, vector control or GD3S-KD-MDA231 cells were stained with anti-GD2 antibody and analyzed on an LSR II flow cytometer (BD). GD2 expression is shown on the y axis and FSC on the x axis. (C) Percentage of GD2+ cells in vector control and GD3S-KD-MDA231 cells. *P < 0.01. (D) To measure cell migration, vector control and GD3S-KD-MDA231 cells were cultured in the presence or absence of 30% serum in a Transwell migration chamber. The average number of cells per microscopic field is shown. *P < 0.001. (E) Mammosphere formation assay using either vector control or GD3S-KD-MDA231 cells was performed by seeding 1,000 cells per well in 24-well cell culture dishes containing mammosphere growth medium. After 10 days, the mammospheres were counted under a light microscope. Scale bar: 200 m. Numbers of mammospheres formed from either vector control or GD3S-KD-MDA231 cells are shown. *P < 0.0001. (F) To examine tumor initiation potential, 1 106 vector control or GD3S-KD-MDA231 cells were transplanted subcutaneously into flanks of NOD/SCID mice. At the end of the ninth week, mice were shaved to remove excess hair on the tumors, and photographs were taken. (G) The tumors size was measured between 4 and 9 weeks. P < 0.000001.

Triptolide, a small molecule inhibitor, inhibits GD3S expression and CSC properties. Triptolide, a small molecule anti-inflammatory drug, has been shown to inhibit GD3S in a melanoma cancer cell line (25). Therefore, we investigated whether triptolide could inhibit GD3S in breast cancer cell lines as well. MDA-MB-231 and SUM-159 cells were treated with different concentrations of triptolide for 24 hours. Triptolide inhibited GD3S mRNA expression in both cell types in a dose-dependent manner, with greater than 95% inhibition at 125 nM (Figure 6, A and B). To test whether inhibition of GD3S by triptolide also inhibited GD2 expression, we treated MDA-MB-231 cells with different concentrations of triptolide for either 24 or 48 hours. Absolute cell counts were measured using flow cytometry. A dose- and time-dependent decrease in GD2+ cells was observed after triptolide treatment, indicating the successful inhibition of GD3S by triptolide (Figure 6C). Of note, a decrease in GD2+ cells was seen under conditions that induced apoptosis in less than 5% of cells.

Triptolide inhibits the expression of GD3S, induces apoptosis in MDA-MB-231 cells, and blocks tumor growth in NOD/SCID mice. MDA-MB-231 cells (A) or SUM-159 cells (B) (5 105/well) of 6-well cell culture dishes were treated with 25, 50, 75, 100, or 125 nM triptolide for 24 hours. /, no treatment. Total RNA was extracted, and GD3S expression was measured by qRT-PCR. (C and D) To measure GD2+ cell growth inhibition, 5 105 MDA-MB-231 (C) and SUM-159 (D) cells were plated in each well of 6-well cell culture dishes and treated with 25, 50, 75, 100, or 125 nM triptolide for 24 or 48 hours. After incubation, the cells were detached with trypsin and stained with anti-GD2 antibody and Sytox Red (for dead cells; Invitrogen). The stained cells were analyzed on an LSR II flow cytometer. Absolute numbers of live cells were calculated by measuring 1,000 events for Trucount beads as explained in Methods. (E) To determine the inhibition of tumor growth, 1 106 MDA-MB-231 cells were subcutaneously transplanted into NOD/SCID mice (n = 8; 4 mice/group). A group of the mice were treated with 0.15 mg/kg/d triptolide, and the control group was treated with PBS every day by i.p. injection. At the end of 8 weeks, mice were sacrificed, and tumors were dissected out and photographed. (F) Tumor sizes from the mice in experiment in E were measured every week after tumor engraftment, and the measurements are shown. P < 0.001, week 3. (G) The survival analysis was based on Kaplan-Meier estimation, and groups were compared by the log-rank test. Control (n = 4, black line) and triptolide (n = 4, blue line) were analyzed for cumulative survival. Survival was defined as the time (in weeks) from transplantation until death. P = 0.015.

To further examine whether triptolide could also inhibit tumor growth in vivo, we introduced 1 106 MDA-MB-231 cells subcutaneously into NOD/SCID mice (2 injections per mouse and 4 mice per group). After the tumors reached 50 mm3, we randomly divided the mice into two groups and treated half of the mice with triptolide (0.15 mg/kg/d) and the other half with PBS (control mice) every day by i.p. injection. Interestingly, after 4 weeks, triptolide-treated animals showed a dramatic decrease in tumor growth compared with control mice. Fifty percent of triptolide treated breast tumors were completely tumor free, and there was a more than 8-fold reduction in tumor volume in 25% of mice (Figure 6D). In addition, tumors in triptolide-treated mice were 3-fold smaller in size and 4-fold lighter by weight (Figure 6E and Supplemental Figure 11). Moreover, in a repeat, identical experiment, triptolide significantly prolonged survival of the treated mice (log-rank, control vs. triptolide, P = 0.0015) (Figure 6F). These findings indicate that GD3S plays a major role in regulating GD3S expression and the resulting GD2+ population. Specifically, it affects cell proliferation and tumor initiation of GD2+ breast cancer cells and when inhibited, greatly diminishes tumor growth and increases metastasis-free survival of breast cancerbearing mice.

More:
JCI - Ganglioside GD2 identifies breast cancer stem cells and promotes ...