|Home | About | Journals | Submit | Contact Us | Français|
Curative eradication of all cells within carcinomas is seldom achievable with chemotherapy alone. This limitation may be partially attributable to tumor cell subpopulations with intrinsic resistance to current drugs. Within squamous cell carcinoma (SCC) cell lines, we previously characterized a subpopulation of mesenchymal-like cells displaying phenotypic plasticity and increased resistance to both cytotoxic and targeted agents. These mesenchymal-like (Ecad-lo) cells are separable from epithelial-like (Ecad-hi) cells based on loss of surface E-cadherin and expression of vimentin. Despite their long-term plasticity, both Ecad-lo and Ecad-hi subsets in short-term culture maintained nearly uniform phenotypes after purification. This stability allowed testing of segregated subpopulations for relative sensitivity to the cytotoxic agent cisplatin in comparison to salinomycin, a compound with reported activity against CD44+CD24− stem-like cells in breast carcinomas. Salinomycin showed comparable efficacy against both Ecad-hi and Ecad-lo cells in contrast to cisplatin, which selectively depleted Ecad-hi cells. An in vivo correlate of these mesenchymal-like Ecad-lo cells was identified by immunohistochemical detection of vimentin-positive malignant subsets across a part of direct tumor xenografts (DTXs) of advanced stage SCC patient samples. Cisplatin treatment of mice with established DTXs caused enrichment of vimentin-positive malignant cells in residual tumors, but salinomycin depleted the same subpopulation. These results demonstrate that mesenchymal-like SCC cells, which resist current chemotherapies, respond to a treatment strategy developed against a stem-like subset in breast carcinoma. Further, they provide evidence of mesenchymal-like subsets being well-represented across advanced stage SCCs, suggesting that intrinsic drug resistance in this subpopulation has high clinical relevance.
Like most other carcinomas, SCCs of the head and neck (HNSCCs) and esophagus respond to treatment with currently available cytotoxic and targeted drugs, and yet they generally are not curable with chemotherapy alone. This observation suggests that individual tumors are heterogeneous, containing subsets of malignant cells with intrinsic resistance to chemotherapy. One basis for defining such heterogeneity within carcinomas derives from epithelial-to-mesenchymal transition (EMT), a phenotypic conversion originally described as a developmental program driving migration and conversion of epithelial cells to a mesenchymal phenotype. Increasing evidence supports the relevance of EMT to the biology of carcinomas, specifically as a process underlying invasion, metastasis and treatment resistance.1–5
Our recent work has provided evidence for discrete subpopulations of malignant cells within SCC cell lines that down-regulate surface E-cadherin and acquire mesenchymal-like gene expression profiles.6 This mesenchymal-like phenotype was shown to be dynamic at a clonal level, with epithelial and mesenchymal-like cells arising from one another on an ongoing basis. The mesenchymal-like subset possessed elevated resistance to two drugs widely used in HNSCC therapy, the conventional chemotherapeutic drug paclitaxel and the epidermal growth factor receptor (EGFR)-targeted agent cetuximab. A potentially related subpopulation with mesenchymal traits has been identified in breast carcinomas, and such features have been attributed to CD44+CD24− stem-like cells within them.7 Using a high-throughput screening approach, a subsequent study demonstrated that this subset within breast carcinomas may respond to distinct pharmacologic targeting strategies.8 This analysis identified the potassium ionophore antibiotic salinomycin as a prototype drug with activity against a CD44+CD24− mesenchymal-like subpopulation within breast carcinomas.
Based on these findings, the use of therapeutics to target mesenchymal-like subpopulations in carcinomas may be a useful strategy for increasing chemotherapeutic efficacy. However, the broad relevance of mesenchymal-like subsets to in vivo biology and therapy for carcinomas remains controversial.9 One basis for this controversy is the difficulty of identifying mesenchymal-like carcinoma cells in human tumors,9,10 where distinguishing malignant cells expressing mesenchymal markers from infiltrating stromal cells is challenging. Here, we use human SCC clinical samples directly xenografted to immunodeficient mice to provide evidence of a mesenchymal-like subpopulation of malignant cells within primary human SCCs of advanced stage. Furthermore, we assess whether the salinomycin sensitivity seen in the CD44+CD24− subset of breast carcinomas can be generalized to mesenchymal-like subpopulations in both SCC cell lines and direct xenografts of human tumors.
Our previous study6 demonstrated that the HNSCC cell lines SCC9 and OCTT2 contain two distinct cellular phenotypes, which are separable based upon surface E-cadherin expression into E-cadherin high (Ecad-hi) and E-cadherin low (Ecad-lo) subsets. Based on these subsets possessing distinct epithelial vs. mesenchymal-like gene expression profiles,6 the terms “Ecad-hi” and “Ecad-lo” are used here interchangeably with “epithelial” and “mesenchymal-like,” respectively. Staining of the SCC9 and OCTT2 cell lines for surface E-cadherin allowed fluorescence-activated cell sorting (FACS)-based purification of these Ecad-hi and Ecad-lo subpopulations (Fig. 1, top parts). When the two purified subsets were grown in short-term culture, they largely maintained mutually exclusive E-cadherin or vimentin-predominant expression patterns (Fig. 1, bottom parts). These results established that, despite the capacity for dynamic interconversion of the Ecad-hi and Ecad-lo phenotypes,6 their relative stability in short-term culture allows direct comparison of drug-resistance properties between them.
To assess the efficacy of salinomycin against the mesenchymal-like subset, drug-induced growth inhibition was compared between FACS-segregated Ecad-hi and Ecad-lo cells. Salinomycin was evaluated against cisplatin, the most widely used cytotoxic agent for HNSCC. With cisplatin, the half-maximal growth inhibitory concentration (IC50) for Ecad-lo cells was higher relative to Ecad-hi cells by 63-fold in OCTT2 and 25-fold in SCC9 (Fig. 2A, top parts). For salinomycin, growth inhibition curves were nearly overlapping between Ecad-lo and Ecad-hi cells, and the IC50 for the Ecad-lo subset treated with salinomycin was lower than that for cisplatin by 50–100-fold (Fig. 2A, bottom parts). These results indicated that the relative in vitro resistance of Ecad-lo cells to cytotoxic therapy, confirmed here using cisplatin, was not present for salinomycin.
This activity of salinomycin against Ecad-lo cells in methylthiazonletetrazolium (MTT) assays comparing sorted subpopulations led us to evaluate whether this agent would show efficacy against the same subset in bulk, unsorted cells. Exposing unfractionated SCC9 and OCTT2 cell lines to salinomycin decreased the percentage of Ecad-lo cells in the surviving population in contrast to cisplatin, which produced the anticipated enrichment (Fig. 2B). Additional efficacy for salinomycin against mesenchymal-like cells was evident upon comparing the growth potentials of Ecad-hi vs. Ecad-lo subsets sorted from residual viable cells after drug treatment (Fig. 2C). Following either cisplatin or salinomycin treatment, residual Ecad-hi cells showed minimal growth when isolated and cultured in the absence of drug. In contrast, surviving Ecad-lo cells retained significantly greater growth capacity after treatment with cisplatin than with salinomycin. In principle, these residual Ecad-lo cells could have been altered by drug exposure to lose their surface E-cadherin while not representing a true mesenchymal-like phenotype. To exclude this possibility, viable Ecad-hi and Ecad-lo cells were isolated post-drug treatment and analyzed by flow cytometry (FC) for vimentin expression (Fig. 2D). Reciprocal staining for E-cadherin and vimentin was retained in residual cells upon either cisplatin or salinomycin treatment. Furthermore, the high proportion of Ecad-lo cells expressing vimentin did not decline after this subset was enriched using cisplatin, confirming that this subpopulation still represented a true mesenchymal-like phenotype following treatment. In aggregate, these results indicated that salinomycin is effective against a drug-resistant mesenchymal-like subpopulation in SCC cell lines of mixed epithelial-mesenchymal phenotype.
The relevance of mesenchymal-like subpopulations to the biology of human SCCs has not been broadly established. Prior to evaluating the efficacy of salinomycin against mesenchymal-like cells in human SCCs, we first sought to better establish a correlating subpopulation in vivo. To confirm that the mixed phenotype cell lines used in this study give rise to comparable cellular heterogeneity in vitro vs. in vivo, they were xenografted to non-obese diabetic/severe combined immunodeficient/interleukin-2 receptor γ-chain-deficient (NSG) mice. Mature tumors were analyzed by dual-label immunohistochemistry (IHC) for E-cadherin and vimentin using humanspecific antibodies (Fig. 3A). Chromogens in IHC micrographs were digitally pseudocolored to highlight the distribution of vimentin-positive cells. This method identified a distinct vimentin-positive tumor cell subset in xenografts of both cell lines, providing evidence that the mesenchymal-like subset in these cell lines is not merely an artifact of in vitro differentiation.
To evaluate primary human SCCs for the presence of a comparable subset, six advanced stage HNSCCs from patient surgical specimens were directly xenografted to NSG mice, and these DTXs were passaged at least twice in vivo to deplete nonmalignant human stromal cells. Following mouse passage, dual-label IHC staining showed DTXs lacking stromal vimentin staining but retaining intratumoral vimentin-positive cells distributed in patterns closely resembling those in their human tumors of origin (Fig. 3B). An abundant human vimentin-positive subset of tumor cells was identified in four of the DTXs (LNT14, OCTT16, LGT17 and, notably, OCTT2, the primary tumor from which the OCTT2 cell line derived). The two remaining DTXs (LST1 and HPPT7) demonstrated infrequent vimentin-positive tumor cells. A high-magnification exam of these two DTXs (Fig. 4) localized some vimentin-positive cells to the invasive front, where they displayed distinctively amoeboid (LST1) or fusiform (HPPT7) morphologies (yellow arrows). These staining patterns, collectively, provided strong evidence for a vimentin-positive, mesenchymal-like subset frequently being present in advanced stage human SCCs.
To determine whether the mesenchymal-like subset could be targeted by salinomycin in vivo, mice xenografted with the OCTT2 DTX (the tumor of origin for the OCTT2 cell line) were treated with either cisplatin or salinomycin. Both drugs arrested growth of established tumors to a comparable degree over 12 d of treatment, during which time control tumors grew rapidly (Fig. 5A). Dual-label IHC staining of residual tumors after cisplatin treatment showed a marked increase in vimentin-expressing areas relative to controls, and visual distinction between the dual subpopulations was enhanced by pseudocoloring of the chromogens (Fig. 5B, second row). These findings suggest that the mesenchymal-like subset resists conventional cytotoxic therapy in vivo, as previously shown for targeted therapy using the anti-EGFR agent cetuximab in the same xenograft system.6 In contrast, residual salinomycin-treated tumors showed a striking, widespread depletion of vimentin-positive cells (Fig. 5C, third row). To quantify these changes, E-cadherin and vimentin-positive areas across wide fields of treated and untreated tumors (Fig. 5C, left column) were digitally mapped in green and red (right column). Comparing relative sizes of E-cadherin vs. vimentin-positive areas demonstrated a significant increase in vimentin staining area with cisplatin treatment but a marked depletion of it with salinomycin (Fig. 5D). By demonstrating selective targeting of a vimentin-positive subpopulation, these results provide in vivo evidence of salinomycin's efficacy against mesenchymal-like SCC cells, which resist other targeted and cytotoxic agents.
Prior studies correlate the presence of mesenchymal traits in carcinoma cell lines with resistance to both conventional cytotoxic11,12 and EGFR inhibitory13–15 agents. Our previous work has identified similar resistance to both classes of compounds in an Ecad-lo, mesenchymal-like subpopulation present within individual SCC cell lines.6 In this study, we demonstrate that salinomycin has efficacy against the mesenchymal-like subpopulation of SCC cells that is resistant to cisplatin, the single most widely used cytotoxic agent for HNSCC.16 Salinomycin was highly toxic in vitro to both Ecad-hi and Ecad-lo subpopulations but only modestly selective for Ecad-lo vs. Ecad-hi cells. This finding contrasts with salinomycin's more striking depletion of vimentin-positive cells from direct xenograft tumors in vivo. This disparity may be accounted for by the role of complex pharmacodynamics in the in vivo setting in determining treatment outcome. For example, the extended multi-dose regimen used in vivo may have allowed differential repopulation of the dual subsets between doses, thus increasing the size disparity between vimentin-positive and -negative areas in residual tumors post-treatment. The potential for such phenomena is supported by the in vitro data, which shows that surviving Ecad-lo cells after salinomycin treatment retain much less growth potential upon reculture than their cisplatin-treated counterparts.
The efficacy of salinomycin against mesenchymal-like SCC cells in this study adds to prior evidence that it can target treatment- resistant malignant subpopulations within other cancers. In particular, it targets CD44+CD24− cells in breast carcinoma7 as well as c-kitlowCD44+CD34+ cells with imatinib resistance in a genetic mouse model of gastrointestinal stromal tumors.17 In principle, its efficacy against any given subpopulation may arise not only from direct toxicity to that phenotype, but by altering the dynamic interconversion between epithelial and mesenchymal-like subpopulations. For instance, salinomycin's depletion of Ecad-lo SCC cells may arise through selective killing, inhibition of epithelial-to-mesenchymal transition and/or promotion of mesenchymal to epithelial differentiation. Although salinomycin's toxicity is readily apparent in vitro, it is difficult to confirm or preclude other possible effects on differentiation, particularly in the in vivo DTX model.
In regard to mechanism of action, salinomycin is a potassium-selective ionophore antibiotic whose basis for anti-cancer activity is presently unknown. Its current therapeutic use is limited to live-stock, chiefly as an anticoccidial drug in poultry and to enhance feed efficiency in ruminants.18,19 Salinomycin induces apoptosis by a caspase-independent mechanism across human cancer cell lines of diverse origin.20 This study also showed its diminished effects against normal cells, indicating that salinomycin's high in vitro toxicity here for both Ecad-hi and Ecad-lo subsets is likely a cancer-specific effect. Other reports suggest mechanisms behind salinomycin's apparent capacity to deplete subpopulations resistant to chemotherapy. It appears to reverse resistance mediated by ABC transporter family members21 and may enhance cytotoxic drug efficacy through inhibiting the P-glycoprotein (ABCB1) member of this family.22 The pertinence of such chemosensitizing mechanisms to salinomycin's action in single-drug use is presently unknown.
Our results suggested that the cytotoxic agents and salinomycin may be additive or synergistic in inducing regression of carcinomas containing mixed epithelial and mesenchymal-like phenotypes. Our failure to observe combined efficacy for the two compounds during in vitro testing may be explained by possible overlap or interference between their killing mechanisms. Because salinomycin does not clearly discriminate between Ecad-hi and Ecad-lo subpopulations in vitro, the benefit from adding cisplatin in the absence of complementary mechanisms for Ecad-hi cells may be minimal. In addition, a high in vivo toxicity and narrow therapeutic window for salinomycin were observed in NSG mice, preventing the daily administrations and higher dosing previously described in reference 8. This narrow therapeutic window hampered efforts to determine usable in vivo doses in combination with cisplatin. Such limitations may be relevant in humans, where major salinomycin toxicity is reported from incidental inhalational exposure.23 Based on these findings, salinomycin seems more likely to remain a prototypic tool compound than to advance into human therapeutic use.
Our detection of vimentin-positive malignant subpopulations across multiple advanced stage human HNSCCs adds new evidence for the physiologic relevance of EMT in carcinoma biology. Whether an EMT-like process commonly occurs in carcinomas in vivo remains controversial,9 with some questioning whether it is primarily an artifact of in vitro cell line differentiation. In this study, the observed maintenance of heterogeneous E-cadherin and vimentin expression within SCC cells after growth as xenografts weighs against this perspective. A barrier to addressing this controversy using clinical carcinoma samples has been the difficulty in distinguishing tumor subpopulations expressing mesenchymal markers from nonmalignant stromal cells of mesenchymal origin.10 After depletion of nonmalignant cells in this study by serial mouse passage of DTXs, xenografts contained human vimentin staining patterns closely recapitulating the staining heterogeneity in their tumors of origin. The maintenance of human vimentin-expressing subpopulations within all six successfully xenografted specimens of advanced metastatic HNSCC increases evidence for a pervasive EMT-like phenomenon in human SCCs. Furthermore, different morphologies seen among vimentin-positive cells in certain xenograft tumors (Fig. 4) may potentially represent two distinct modes of migration previously described within cancer cell lines having undergone EMT.24,25 Specifically, a tentative assignment of mesenchymal or amoeboid migration may be made to the respective fusiform and irregular morphologies of vimentin-positive cells seen at the invasive front. More detailed characterization of these subpopulations across a broader panel of tumors remains to be accomplished; however, our present data show that HNSCCs do not divide into the dichotomous epithelial and mesenchymal-like categories implied by previous studies, with cell lines grouped by predominant phenotype.15 Instead, we show a spectrum of mixed epithelial and mesenchymal-like differentiation within individual HNSCCs, as previously suggested to exist within ovarian carcinomas.26
The therapeutic implications of these mixed phenotypes in vivo are highlighted in this study by salinomycin and cisplatin exerting opposite effects on the size of the vimentin-expressing subpopulation in DTXs. The capacity of salinomycin to target mesenchymal-like carcinoma cells also raises the possibility that such compounds could beneficially alter invasive and metastatic behavior, in which EMT likely participates. Overall, these findings underscore a need to develop novel compounds targeted at specific treatment-resistant phenotypes. Their use in combination with existing drugs offers a promising avenue toward achieving curative outcomes using chemotherapy as single-modality treatment.
Cell lines were maintained in 1:1 Dulbecco's modified Eagle and Ham's F12 media (Gibco) supplemented with 400 ng/ml hydrocortisone (Sigma), 10% fetal calf serum (Atlanta Biologicals) and 50 µg/ml gentamycin (Cellgro). SCC9 cells were obtained from the ATCC (CRL-1629). The OCTT2 cell line was derived from a human HNSCC primary surgical specimen and is previously described in reference 6. Genetic purity of both cell lines was confirmed based on microsatellite DNA markers using an Identify Mapping Kit (MSK, Coriell Institute).
Images were obtained using Nikon TE2000 inverted and E600 upright microscopes and processed with ImagePro-Plusv6.2 and ACT-1 software using uniform settings within each experiment. Immunofluorescence (IF) staining was performed in 24-well plates, as described in reference 27. A FACSAria (BD) sorter was used for FACS. FC was performed using a FACSCalibur instrument (BD) and analyzed using WinMDIv2.9. Dead cells were removed by gating exclusion of propidium-iodide (PI)-positive cells. For intracellular vimentin staining, sorted cells were first permeabilized with acetone, as described in reference 28.
Primary antibodies for IF staining were Alexafluor488-anti-human E-cadherin (560061, BD) and rabbit-anti vimentin (53945, AnaSpec). Alexafluor594-anti-rabbit IgG (A-11012, Invitrogen) was used as a secondary antibody. Antibodies used for FC and FACS were allophycocyanin (APC)-anti-E-cadherin (324108, Biolegend) and phycoerythrin (PE)-anti-vimentin (ab49918, Abcam).
MTT assays were performed by seeding 3,000 cells/well (OCTT2) or 5,000 cells/well (SCC9) on 96-well plates. Drug treatments were started 24 h after seeding. Growth was quantified using a MTT Proliferation Kit (V-13154, Invitrogen). Optical density in each well was measured at 590 nm (OD-590 nm), and values were normalized to OD-590 nm for non-treated controls in drug treatment experiments.
Formalin-fixed, paraffin-embedded specimens were cut in 5 µm sections and processed for dual-label IHC as previously described in reference 6. Briefly, following antigen retrieval, staining was performed sequentially for E-cadherin and vimentin followed by hematoxylin counterstain. Mouse anti-human E-cadherin (IR059, Dako) was used at 1:10 dilution and detected with 3′,3-diaminobenzidine. After requenching peroxidase activity, mouse anti-human vimentin (IR630, Dako) was used without dilution and detected using Vector VIP chromogen (SK-4600, Vector). Specificity was confirmed by substituting mouse IgG1 controls for each antibody.
To generate pseudocolored images, IHC micrographs were processed using ImagePro-Plus. E-cadherin-, vimentin- and hematoxylin-staining areas were separated based on spectral distribution using color cube analysis and converted to solid green, red and blue colors, respectively. For image-based quantitative analysis of E-cadherin vs. vimentin staining areas, respective green and red color masks were created corresponding to each label. Masks were copied to black backgrounds to create the displayed images, and relative areas of each label were calculated using the count/size tool. Batch-processing ensured unbiased comparison among samples within an experiment.
The NSG mice used in all experiments were bred and utilized at the Wistar Institute under protocols approved by the institutional animal care and use committee. To generate xenografts from human SCC specimens, minced 1 mm tumor fragments were placed in subcutaneous pockets through flank incisions in anesthetized mice. Reduced Growth Factor Matrigel (354230, BD) was added, and incisions were closed with suture. Successfully grafted tumors were re-passaged in NSG mice with similar technique, and dual-label IHC staining was performed after 2–3 in vivo passages. Tumor volumes were measured as [length × width2]. Xenografts of SCC9 and OCTT2 cell lines were generated by subcutaneous injection in 100 µl Reduced Growth Factor Matrigel. For drug treatment, salinomycin (S4526, Sigma), cisplatin (P4394, Sigma) or equivalent volume saline control was injected intraperitoneally every 2 d.
Tumors for direct xenografting were derived from surgical specimens of previously untreated HNSCCs from either the University of Pennsylvania (Institutional Review Board-approved protocol #417200) or Philadelphia Veterans Administration Medical Center (Institutional Review Board-approved protocol #01090). Ten stage IV tumors, all with established lymph node metastatic disease, were implanted in NSG mice. Of these ten tumors, the six that successfully xenografted and could be passaged serially in vivo were analyzed further. Five were primary site tumors derived from a diversity of head and neck subsites, including the oral cavity (OCTT2, OCTT16), supraglottic and glottic larynx (LST1, LGT17, respectively) and pyriform sinus (HPPT7). The remaining tumor (LNT14) was derived from a neck lymph node metastasis from an occult primary head and neck tumor.
Data with error bars represent mean ± standard error of mean. Analysis of variance (ANOVA) was performed to evaluate differences between the groups. Where relative growth is measured, two-way ANOVA with an interaction was used. One-way ANOVA was utilized for comparisons of percentage vimentin staining area and percentage Ecad-lo cells. ANOVA for repeated measures was applied to compare tumor growth curves between treated and untreated groups of mice. Analyses were done using SAS/STAT software version 9.2 and p < 0.05 was considered statistically significant.
This work was supported by NIH/NCI Grant P01 CA098101 (M.H., A.R., P.G., J.D., and D.B.) and its morphology, molecular biology and cell culture core facilities. It is also supported by the Wistar Cancer Center Core Grant (P30 CA10815) and made use of its microscopy and flow cytometry facilities. Additionally, resources and facilities of the Philadelphia VA Medical Center were used. We thank Thierry-Thien Nguyen for unpublished pilot data regarding the in vitro use of salinomycin. All authors declare no conflict of interest.