The sialyl-glycolipid stage-specific embryonic antigen 4 marks a subpopulation of chemotherapy-resistant breast cancer cells with mesenchymal features

An antibody screening based on a library of 45 antibodies directed against surface epitopes including published stem cell and/or cancer stem cell markers (Additional file 2: Table S1) was performed to identify novel biomarkers for breast cancer cell subpopulations resistant to chemotherapeutic treatment. For the initial screening, 50–100 xenografted mice were used for each of four independent TNBC PDXs [9]. When tumor volumes reached 150–350 mm³ (pretreatment stage), mice were treated with an AC combination according to the standard of care. After tumor shrinking to volumes of 14–63 mm³, the nodules were removed (residual tumor stage). Tumors from untreated mice were removed (untreated stage) to serve as direct controls. A group of animals with residual tumors were kept until the disease recurred (regrowth stage) (Fig. 1a). All tumors removed at any time point were dissociated and analyzed by flow cytometry. No significant differences concerning marker expression were found between the pretreatment and untreated stages, excluding size-dependent marker regulation, whereas the expression of 10 markers decreased (Fig. 1c) and the expression of 13 markers increased (Fig. 1d) during chemotherapy. Eighty-seven percent (20 of 23) of these markers returned to an expression level similar to the untreated stage upon tumor regrowth (Fig. 1c and d). CD44 and CD133, which have been described to correlate with a cancer stem cell and drug resistance phenotype [17], showed enrichment in only one of the four models. This is possibly due to the tumor heterogeneity that characterizes patients’ tumors, which was here well recapitulated by the use of PDX models (Fig. 1d). However, we identified a distinct subpopulation of tumor cells expressing SSEA4 that was strongly enriched during chemotherapeutic treatment in all tumor models analyzed. Independent replicates of three tumor models confirmed significant enrichment in the number of SSEA4-positive cells in residual tumors upon AC treatment in all analyzed tumors (p < 0.001, n = 8 for each tumor model) (Fig. 1e). This result was further confirmed by immunohistochemistry (Fig. 1b). Similarly to other markers, the amount of SSEA4-positive cells decreased back to pretreatment levels in three of four tumor models after the treatment was stopped.

To further address the correlation between CD24, CD44, and SSEA4 expression, we performed containing of these markers on residual tumor nodules after AC chemotherapy and on untreated tumors of three independent models (Additional file 5: Figure S3). We did not observe a significant enrichment of the CD24^low/CD44^high subpopulation in any of the models. In model HBCx-6, there was a subpopulation of SSEA4⁺/CD44⁺ cells that was enriched by about twofold; however, the SSEA4⁺/CD44⁻ fraction was enriched to the same degree. In models HBCx-10 and HBCx-14, CD44 and SSEA4 expression was observed on separate subpopulations. We concluded that SSEA4 expression is an indicator of treatment-resistant cells that overcomes the heterogeneity observed for the (cancer) stem cell markers in the tumor models used in this study.

In one PDX model, we observed tumors with variable sensitivity to the AC treatment and noticed that the amount of SSEA4-positive cells correlated with these differences. Nineteen tumors showed reduced sensitivity to AC treatment (tumor volume >100 mm³ as maximal regression), and one tumor did not respond to the treatment at all (no tumor shrinkage during treatment). The percentages of SSEA4-positive cells were 36.8 % in sensitive tumors, 42.2 % in tumors with reduced sensitivity, and 98.5 % in the fully resistant tumor (data not shown). Therefore, we compared tumors from PDX models that are sensitive (n = 6) or de novo resistant (n = 4) to AC treatment. Three of four resistant tumor models showed higher percentages of SSEA4-positive cells than the six sensitive tumors (Additional file 6: Figure S4).

To provide further evidence of resistance to drug toxicity, we established primary PDX-derived ex vivo cell lines from tumors containing different amounts of SSEA4-positive cells and treated them with chemotherapeutic drugs in vitro. One of the cell lines derived from model HBCx-17 showed reproducible growth as an adherent culture and as a suspension culture, with cells growing in suspension showing higher SSEA4 expression than the adherent cells (Fig. 2a). In cytotoxicity assays with seven commonly used drugs (Fig. 2b-f and data not shown), the suspension cells showed higher half-maximal inhibitory concentration values than the adherent cells for the DNA synthesis and transcription inhibitors cisplatin, mafosfamide, 5-fluorouracil, and doxorubicin, indicating an increased resistance, but not for the topoisomerase inhibitors etoposide, topotecan, and irinotecan. To provide a dose-escalating reproduction of our in vivo experiment, we treated an adherent cell line derived from model HBCx-17, containing about 15 % SSEA4-positive cells in the steady state, and analyzed the phenotype of the cells surviving the treatment upon administration of increasing drug dosages. In every case, the surviving population showed a significantly higher fraction of SSEA4-positive cells (Fig. 2g-i), while the absolute number of SSEA4-positive cells was not significantly increased.

Next, we evaluated possible differences in the tumorigenic capacity of SSEA4-positive and SSEA4-negative subpopulations in model HBCx-14, which shows about 10 % SSEA4-positive tumor cells when growing without any treatment and a significant upregulation under genotoxic stress. To do so, 10⁵ freshly dissociated SSEA4-positive or SSEA4-negative cells were injected into two groups of eight mice each. Although a trend toward a faster tumor initiation was observed when we grafted the SSEA4-positive tumor subpopulation, both fractions were able to generate tumors, indicating that the positive fraction only has an initial growth advantage (Additional file 7: Figure S5a). Furthermore, we analyzed the expression of SSEA4 in tumors that arose from both subpopulations. Whereas the fraction of SSEA4-expressing cells in tumors that originated from the SSEA4-negative subpopulation recovered to around 5 %, the one from tumors that originated from the SSEA4-positive subpopulation was reduced to about 10 % following the growth phase in vivo (Additional file 7: Figure S5b). The recovery of the original SSEA4-positive versus SSEA4-negative ratio was also reflected by the observation that when freshly dissociated SSEA4-positive and SSEA4-negative cells were placed in culture separately, they restored the original positive versus negative ratio within approximately 6 days (data not shown), which correlated well with the observation of SSEA4 downregulation during disease relapse after chemotherapy.

Local or metastatic breast cancer relapse may occur many years after surgery, despite a short-term response to neoadjuvant chemotherapy [18] or a failure of adjuvant chemotherapy. To investigate SSEA4 expression in metastatic relapse, we conducted a retrospective case history study using PDXs derived from metastatic specimens. These models were derived from confirmed M1-stage patients through collection of liquid biopsies, either peripheral circulating tumor cells (CTCs) or infiltrating CTCs obtained from serous effusions [19]. Robust SSEA4 expression was exclusively found in PDXs whose donors were previously treated with specific neoadjuvant chemotherapy formulations, including the AC combination and other genotoxic drugs with analogous modes of action, such as epirubicin (Fig. 3). Conversely, metastatic patients who were administered taxols (paclitaxel, docetaxel) or antimetabolites (capecitabine, 5-fluorouracil) as adjuvant therapy, but who had no genotoxic treatment before sample collection (Fig. 3 and Additional file 8: Table S2 for patient history), displayed little or no expression of SSEA4 on their matched PDXs (Fig. 3).

We performed microarray-based global mRNA and miRNA expression profiling on SSEA4-positive and SSEA4-negative cells using three independent TNBC models: HBCx-6 (n = 7 tumors), HBCx-10 (n = 5 tumors), and HBCx-14 (n = 9 tumors). To avoid a bias by cells of murine origin [20], all mouse cells were depleted after tumor dissociation (Fig. 4a) and SSEA4-positive cells were labeled and magnetically separated from the negative fraction (Fig. 4b) before expression profiling.

Transcripts with significantly increased (240 genes, p < 0.05, ≥1.5-fold, Additional file 9: Table S3) or decreased (182 genes, p < 0.05, greater than or equal to −1.5-fold, Additional file 9: Table S3) expression in SSEA4-positive compared with SSEA4-negative cells were subjected to a term enrichment analysis based on Gene Ontology categories [21]. In the SSEA4-positive fractions, we found a strong overrepresentation of genes linked to the TGF-β and epidermal growth factor signaling pathways as well as genes involved in cell adhesion and migration and in regulation of apoptosis, proliferation, and differentiation (Additional file 9: Table S3). Among the genes upregulated across all tumor models, seven functional groups involved in cellular import and export, response to toxins and oxidative stress were significantly enriched (Additional file 9: Table S3). In particular, members of the solute carrier (SLC) and multidrug resistance ATP-binding cassette transporter families were significantly upregulated (Additional file 10: Figure S6B).

Also, among the differentially expressed transcripts, we identified a substantial number of genes involved in EMT. Epithelial markers such as cytokeratin 19, CLDN1, CLDN3, and CLDN4 showed lower expression in SSEA4-positive cells, whereas mesenchymal indicators such as fibronectin, vitronectin, ZEB1, and ZEB2 were upregulated (Additional file 10: Figure S6 A). In contrast, published stem cell markers were not consistently regulated among the SSEA4-positive and SSEA4-negative cell fractions (Additional file 10: Figure S6 C). As SSEA4 is a glycolipid epitope, its expression cannot be monitored directly by transcriptome profiling. SSEA3, the direct precursor of SSEA4, showed no increased signal intensity in SSEA4-positive cells when measured with anti-SSEA3 antibody in all analyzed models (data not shown). This indicates that SSEA4 enrichment regulation may be due to increased SSEA3-to-SSEA4 conversion or to increased SSEA4 catabolism in SSEA4-negative cells. However, no difference was observed in the expression of genes coding for enzymes involved in degradation of SSEA4 or of ST3GAL2, the enzyme catalyzing this final step of SSEA4 synthesis [22]. In expression analysis of the miRNA data set, we identified 166 miRNAs more than twofold overrepresented and 68 miRNAs more than twofold underrepresented in SSEA4-positive versus SSEA4-negative cells among all analyzed tumor models (Additional file 9: Table S3). Hierarchical clustering of the Pearson correlation coefficients of our mRNA and miRNA expression datasets showed a higher correlation of SSEA4-positive versus SSEA4-negative phenotype in the miRNA rather than mRNA data set (Fig. 4c). No miRNAs were significantly upregulated, and 18 miRNAs were significantly downregulated (p < 0.05 by paired t test), in SSEA4-positive cells (Fig. 4d and Additional file 9: Table S3). These results were independently validated by using a flow cytometry–based 39-plex miRNA assay (Additional file 11: Figure S7) for direct measurement of miRNAs without prior amplification. To determine putative mRNA targets for these candidates, we used six different computational target prediction tools and considered only targets that were predicted by at least two target prediction algorithms. On the basis of this analysis, 5 of the 18 miRNAs downregulated in SSEA4-positive cells (miR-96-5p, miR-200b-3p, miR-200c-3p, miR-429, and miR-92a-3p) are predicted to target ST3GAL2, suggesting that SSEA4 expression is regulated posttranscriptionally. Furthermore, 12 of the 18 downregulated miRNAs are known to target key mesenchymal regulator and indicator genes such as ZEB1, ZEB2, fibronectin 1, Snail1, Snail2, and Twist (Additional file 12: Figure S8 and Additional file 9: Table S3).

To provide functional confirmation of the role of ST3GAL2 in the regulation of SSEA4 expression in PDX samples, we knocked down ST3GAL2 by small interfering RNA (siRNA) in HBCx-39 cells, which show high levels of SSEA4 expression, and analyzed the expression of SSEA4. A significant decrease of SSEA4 expression was observed upon ST3GAL2 expression inhibition (p < 0.001, n = 6) (Fig. 5a and b). Knockdown of CD133 cell surface expression was used as a positive control for siRNA targeting, and it had no impact on SSEA4 expression. To provide further evidence of the relationship between ST3GAL2 and SSEA4, we compared ST3GAL2 mRNA levels with SSEA4 cell surface levels on nine different tumor models, which showed significant positive correlation (p < 0.002, Fig. 5c).

As EMT has already been correlated with drug resistance [23], we wanted to examine if SSEA4 expression was increased after EMT induction. Upon treatment of the epithelial breast cell line MCF 10A with TGF-β1, almost all cells changed their morphology from an epithelial to an elongated fibroblastic shape (Fig. 6a). As expected [24], epithelial markers such as EpCAM and E-cadherin were downregulated, while mesenchymal markers such as vimentin and fibronectin were upregulated (Fig. 6a and data not shown). Upon EMT induction, the SSEA4-positive fraction increased more than threefold, hence proving a causal correlation between the transition toward a mesenchymal phenotype and SSEA4 expression (Fig. 6b).

Given the functional connection between SSEA4 and ST3GAL2, we evaluated the clinical value of ST3GAL2 in a large, publicly available clinical microarray database of breast tumors from 2977 patients [15]. Highly significant differences (p < 0.01) trending toward a poorer outcome for patients expressing higher levels of ST3GAL2 within the estrogen receptor–negative (ER⁻) or ER⁻/progesterone receptor–negative (PR⁻) subset of patients were found, independent of the treatment (Fig. 7a). When we focused on patients treated with chemotherapy, we observed a highly significant reduction of relapse-free survival independent from the tumor subtype (p < 0.01, HR 1.91) in ER⁻ patients (p < 0.01, HR 2.97) and in ER⁻/PR⁻ patients (p < 0.01, HR 3.08) among patients expressing high levels of ST3GAL2 (Fig. 7b). When we applied distant metastasis-free survival as an endpoint, we found that patients expressing high levels of ST3GAL2 had a worse outcome, confirming the involvement of SSEA4-positive cells in metastasis formation (Fig. 7c), as observed in the case of metastasis-derived PDXs.

To further evaluate SSEA4 expression as a marker of intrinsic tumor cell resistance to chemotherapy, we tested SSEA4 expression in samples from primary clear cell RCC, a tumor entity known to be de novo resistant to chemotherapy in more than 95 % of patients [25] and from late ovarian cancer, an aggressive disease whose treatment relies on chemotherapy as the only therapeutic option [26]. Primary clear cell RCC as well as healthy kidney tissues from the same patients were analyzed for SSEA4 expression. In all of the analyzed patients (n = 3), an elevated number of SSEA4-positive cells was observed in the tumor tissue (Additional file 13: Figure S9). Similarly, when primary cells derived from serous ovarian carcinoma specimens were analyzed for expression of SSEA4, three of three samples from tumors that previously received genotoxic treatment showed SSEA4 expression above 10 %, whereas only one primary cell line matched to treatment-naive patients showed SSEA4 expression above 10 % (Fig. 8a–c). Analysis of mesenchyme-specific (FN1, SNAI1, ZEB2) and epithelial (CLDN3) genes by quantitative real-time polymerase chain reaction revealed that ovarian cancer primary cells with high SSEA4 expression showed higher expression of mesenchyme-specific genes and lower expression of the epithelium-specific ones (Fig. 8a and b).

In light of these results, we evaluated whether ST3GAL2 expression could be associated with patient prognosis in ovarian cancer. Using the whole dataset of ovarian tumors from 1464 patients [16], a significant difference trending toward a worse clinical outcome was observed with respect to progression-free survival (p < 0.05) as well as postprogression survival (p < 0.05) (Fig. 8e). When we assessed patients treated with chemotherapy, the level of significance was also strongly increased for both endpoints (p < 0.01) (Fig. 8f).