High aldehyde dehydrogenase activity identifies cancer stem cells in human cervical cancer.

High aldehyde dehydrogenase (ALDH) activity characterizes a subpopulation of cells with cancer stem cell (CSC) properties in several malignancies. To clarify whether ALDH can be used as a marker of cervical cancer stem cells (CCSCs), ALDHhigh and ALDHlow cells were sorted from 4 cervical cancer cell lines and 5 primary tumor xenografts and examined for CSC characteristics. Here, we demonstrate that cervical cancer cells with high ALDH activity fulfill the functional criteria for CSCs: (1) ALDHhigh cells, unlike ALDHlow cells, are highly tumorigenic in vivo; (2) ALDHhigh cells can give rise to both ALDHhigh and ALDHlow cells in vitro and in vivo, thereby establishing a cellular hierarchy; and (3) ALDHhigh cells have enhanced self-renewal and differentiation potentials. Additionally, ALDHhigh cervical cancer cells are more resistant to cisplatin treatment than ALDHlow cells. Finally, expression of the stem cell self-renewal-associated transcription factors OCT4, NANOG, KLF4 and BMI1 is elevated in ALDHhigh cervical cancer cells. Taken together, our data indicated that high ALDH activity may represent both a functional marker for CCSCs and a target for novel cervical cancer therapies.


INTRODUCTION
Cervical cancer is the second most commonly diagnosed cancer and ranks second only to breast cancer as the leading cause of cancer death in women in developing countries [1]. Based on the GLOBOCAN estimates, approximately 529,000 women worldwide were diagnosed with invasive cervical cancer, and more than half of these patients died from their disease in 2008 [2]. Cervical carcinoma development begins with the infection of the cervical epithelium by high-risk human papillomaviruses (hr-HPVs) [3][4][5]. Although cervical cancer can be detected in its early stages by HPV testing and Papanicolaou (Pap) smear screening and successfully eradicated through surgery, curative treatments do not yet exist for advanced, recurrent or metastatic disease [6][7][8].
Tumor growth and metastasis are driven by a small population of cancer stem cells (CSCs) [9]. The first extensive documentation of CSCs came from leukemia, in which only a small subset of cancer cells were shown to be capable of transferring the disease to non-obese diabetic/severe combined immunodeficient (NOD/SCID) mice [10]. This concept was then extended to solid tumors. Human breast cancers have been demonstrated to contain a population of cells with stem cell properties that display surface marker expression of CD44 + /CD24 -/ lin - [11]. Subsequently, CSCs have been identified and prospectively isolated using surface markers from a variety of malignancies, including brain tumors [12], melanoma [13], multiple myeloma [14], prostate cancer [15], colon cancer [16,17], head and neck squamous cell carcinoma [18,19] and pancreatic adenocarcinoma [20]. Due to the instability and scarcity of surface markers in solid tumors, other methodological strategies have been widely explored to identify and isolate CSCs, including side population phenotype, sphere formation and aldehyde dehydrogenase (ALDH) activity assay [21].
To verify whether cells with high ALDH activity are CCSCs, fluorescence activated cell sorting (FACS) and standard functional assays were used in the present study to analyze the cellular properties of ALDH high and ALDH low cells isolated from 4 human cervical cancer cell lines and 5 primary cervical cancers. The results indicated that a subpopulation of human cervical cancer cells with high ALDH activity possess enhanced selfrenewal capacities, differentiation potential and increased tumorigenicity, indicating that high ALDH activity may represent a marker of CSCs in cervical cancer.

RESULTS
ALDH expression and activity in human cervical tissue specimens and cervical cancer cell lines ALDH1 expression was evaluated in normal and cancerous cervical tissues ( Figure 1A). Notable differences were observed in the expression patterns of ALDH1 in the basal cells of normal human cervical tissues; these patterns were classified into 4 types: (1) (Figure 1Ad). Because basal cells are known to contain undifferentiated reserve cells of the normal cervix, we speculated that the ALDH1-positive basal cells may represent the stem cells of the normal cervix. A large number of ALDH1-positive cells were found in the stroma in all of the normal cervical tissues (Figure 1Aa-1Ad). Most of these ALDH1-positive cells in the normal cervical stroma are likely CD45-positive leukocytes, which have also been found in the stroma of normal breast tissues by Dr. Deng et al. [45].
ALDH1-positive cells were also found in all 53 cervical cancer tissues. Similar to ALDH1 expression in the normal cervical tissues, the expression patterns of ALDH1 in the cervical cancer tissues could be classified into 3 types: (1) Dot-scattered ALDH1 expression (  Figure 1Ag). In this study, approximately 10% of the cervical cancer cells were ALDH1-positive, which is consistent with the notion that CSCs constitute a minority of the tumor cells. Furthermore, ALDH1-positive cells were present in each of the 19 passages of the serially xenografted tissues in NOD/SCID mice (Figure 1Ah).
The ALDEFLUOR kit was used to test the ALDH enzymatic activity in the cervical cancer cell lines. Cells were labeled with activated ALDEFLUOR reagent in the presence or absence of the ALDH inhibitor, DEAB.  1000×. B, Cells were labeled using the ALDEFLUOR kit, and ALDH high cells (bright green fluorescence) were detected by fluorescent microscopy. C, ALDH enzyme activity in 4 cervical cancer cell lines was analyzed by flow cytometry. As a negative control, cells were treated with the specific ALDH inhibitor DEAB. The gated cells represent the ALDH high cells.
Oncotarget 2013; 4: 2464 www.impactjournals.com/oncotarget A drop of ALDEFLUOR-labeled cells was smeared and examined by fluorescence microscopy. Fluorescent and phase contrast images were acquired and merged. As shown in Figure 1B, each of the 4 cervical cancer cell lines (SiHa, C33A, CaSki and HT-3) contained ALDH-positive cells (indicated by the bright green fluorescence). The remaining ALDEFLUOR-labeled cells were analyzed by flow cytometry. Compared with the DEAB-treated control, high ALDH activity was detected in 13.4% of the SiHa cells, 33.3% of the C33A cells, 24.4% of the CaSki cells and 13.3% of the HT-3 cells ( Figure 1C).
Together, these results suggest that a subpopulation of ALDH high cells exists in normal and cancerous cervical tissues, serially xenografted cervical cancer tissues and cervical cancer cell lines, implying that ALDH may be a marker of stem cells and CSCs in cervical tissues.

ALDH high cervical cancer cells display enhanced self-renewal capacity
Self-renewal is a critical characteristic of stem cells and CSCs. To assess self-renewal in vitro, ALDH high and ALDH low cervical cancer cells were cultured in serumfree medium under conditions optimal for growing tumorspheres. As shown in Figure 2A, ALDH high cells isolated from the 4 cervical cancer cell lines generated classical tumorspheres, while ALDH low cells did not form tumorspheres, but only a few cell aggregates. When plated at a density of 200 cells/well in 24-well plates (low density culture), 6.2%, 8.5%, 6.2% and 9% of the ALDH high cells from SiHa, C33A, CaSki and HT-3 cells, respectively, generated tumorspheres in the 1 st passage, while the ALDH low cells generated no, or very rare, tumorspheres. Upon 3 consecutive passages in culture, the tumorsphere forming efficiency of the ALDH high cells gradually increased ( Figure 2B). To exclude the effects of cell aggregation, which can occur in low density cultures, cells were cultured at a density of a single cell/ well. The ALDH high cells from SiHa, C33A, CaSki and HT-3 cells generated tumorspheres with an efficiency of 32.8%, 24.5%, 26.6% and 38.5%, whereas the ALDH low cells generated tumorspheres with an efficiency of 2.6%, 1.6%, 2.6% and 4.7%, respectively ( Figure 2C). These data indicated that the ALDH high cervical cancer cells have greater self-renewal capacity than the ALDH low cells.

ALDH high cervical cancer cells have enhanced tumorigenic capacity in vivo
One of the most important characteristics of CSCs is their powerful ability to form tumors. To determine whether the ALDH high cells have a greater capacity to form tumors, the ALDH high and ALDH low cell populations were sorted from 4 cervical cancer cell lines, and limiting dilutions of the cells were subcutaneously injected into NOD/SCID mice. The tumor latency, tumor incidence and tumor volume were then monitored.
Firstly, tumor volume was monitored twice a week,   Figure 3A. Inoculation of NOD/SCID mice with 10 4 or 10 3 ALDH high and ALDH low SiHa cells led to tumor formation from both populations. However, the tumors formed by ALDH high SiHa cells were larger and grew faster than those formed by ALDH low SiHa cells. Furthermore, inoculation with 10 2 or 10 1 ALDH high and ALDH low SiHa cells led to tumor formation from only the ALDH high SiHa cells ( Figure 3A, panel 1). In C33A cells, the ALDH high population, but not the ALDH low population, was capable of forming palpable tumors at each cell dose (10 6 , 10 5 , 10 4 and 10 3 ). However, upon sacrifice of the NOD/SCID mice, very small regions of tumor were found in the mice that had been inoculated with 10 6 and 10 5 ALDH low C33A cells ( Figure 3A, panel 2). Upon inoculation with 10 5 or 10 4 CaSki cells, the ALDH high population formed larger palpable tumors more rapidly than the ALDH low population. However, after inoculation with 10 3 or 10 2 cells, the ALDH high CaSki cells, but not the ALDH low CaSki cells, were capable of forming palpable tumors ( Figure 3A, panel 3). In HT-3 cells, 10 5 or 10 4 of the ALDH high and ALDH low cells formed similarly sized palpable tumors almost simultaneously. When the cell dose was decreased to 10 3 or 10 2 , ALDH high HT-3 cells could form palpable tumors, while ALDH low HT-3 Tumor latency was monitored after injection of sorted cells into the NOD/SCID mice and was defined by the period of time during which the mice remained tumor-free ( Figure 3B). ALDH high SiHa cells exhibited a significantly shorter tumor-free period; for instance, the shortest tumor-free period for ALDH high cells was 4 weeks, as compared to the 6 week latent period for ALDH low cells. ALDH high SiHa cells also exhibited a lower tumor-free rate (10% in ALDH high cells versus 47.5% in ALDH low cells) than ALDH low SiHa cells (p<0.001). Similarly, ALDH high C33A cells displayed a significantly shorter tumor-free period (5 weeks in ALDH high cells versus 9 weeks in ALDH low cells) and a lower tumor-free rate (37.5% in ALDH high cells versus 82.5% in ALDH low cells) compared with ALDH low C33A cells (p<0.001). ALDH high CaSki cells exhibited a significantly shorter tumor-free period (4 weeks in ALDH high cells versus 7 weeks in ALDH low cells) and a lower tumor-free rate (27.5% in ALDH high cells versus 60% in ALDH low cells) compared with ALDH low CaSki cells (p<0.001). ALDH high HT-3 cells also showed a significantly shorter tumor-free period (2 weeks in ALDH high cells versus 3 weeks in ALDH low cells) and a lower tumor-free rate (17.5% in ALDH high cells versus 42.5% in ALDH low cells) compared with ALDH low HT-3 cells (p<0.01).
Together, these results from the tumor formation assays in NOD/SCID mice suggest that ALDH high cervical cancer cells have a more rapid tumor growth rate, shorter tumor latency, lower tumor-free rate and higher tumor- initiating frequency than ALDH low cells. Therefore, ALDH high cervical cancer cells have a potent ability to form tumors in vivo.

ALDH high cells, but not ALDH low cells, have the ability to differentiate in vitro and in vivo
One characteristic of CSCs is the capacity to differentiate into non-CSCs and give rise to the heterogeneous tumor cell populations. To determine whether ALDH high cells are capable of differentiation in vitro, ALDH high and ALDH low cells were cultured separately in DMEM medium supplemented with 10% FBS for 2 weeks. After incubation, the cultured populations were analyzed using the ALDEFLUOR assay ( Figure 4A-4D). Approximately 90% of the ALDH high SiHa cells differentiated into ALDH low cells, and only 10% of the cells remained strongly ALDH-positive. However,  greater than 99% of the ALDH low SiHa cells retained the ALDH low phenotype ( Figure 4A). Similarly, 57.2% of the ALDH high C33A cells ( Figure 4B), 72.1% of the ALDH high CaSki cells ( Figure 4C) and 75.9% of the ALDH high HT-3cells ( Figure 4D) generated ALDH low cells. However, 98.52% of the ALDH low C33A cells ( Figure 4B), 99.45% of the ALDH low CaSki cells ( Figure 4C) and 99.86% of the ALDH low HT-3 cells ( Figure 4D) maintained the ALDH low phenotype. Notably, 1.48% of the ALDH low C33A cells generated ALDH high cells, which was a larger fraction than the other cell lines. Because C33A cells contain a larger population of ALDH high cells and higher ALDH activity than the other cell lines, this proportion of ALDH high cells may have resulted from errors during the cell sorting manipulation.
The differentiation capacity of ALDH high and ALDH low cells was also assessed in vivo. In the tumors formed by ALDH high cells, a few ALDH1-positive cells and many ALDH1-negative cells were found, indicating that ALDH high cells were able to generate ALDH high cells through self-renewal and to generate ALDH low cells through differentiation ( Figure 4E, upper panel). However, in the tumors formed by ALDH low cells, no ALDH1positive cells were found, indicating that ALDH low cells did not have the ability to differentiate ( Figure 4E, lower  panel).
Taken together, these data demonstrate that ALDH high cervical cancer cells have the ability to differentiate both in vitro and in vivo. Thus, ALDH high cervical cancer cells establish the cellular hierarchy in tumors through selfrenewal and differentiation.

ALDH high cells are more resistant to cisplatin than ALDHlow cells
The resistance of CSCs to current chemotherapeutics is thought to be responsible for cancer recurrence and metastasis [46]. Because cisplatin is one of the most commonly used chemotherapeutic drugs in the treatment of cervical cancer, we tested the effects of cisplatin on ALDH high and ALDH low cervical cancer cells. After treatment with cisplatin, the population of ALDH high cells expanded from 7.25% to 36.5% in SiHa cells, 48.3% to 56.4% in C33A cells, 24.5% to 70.2% in CaSki cells and 27.7% to 62.4% in HT-3 cells ( Figure 5A). These data suggested that the ALDH high cells, but not the ALDH low cervical cancer cells, are resistant to cisplatin treatment.
Furthermore, ALDH high and ALDH low cells isolated from 4 cervical cancer cell lines were exposed to different concentrations of cisplatin for 24 h, and cell viability was determined using an MTT assay. Cisplatin caused dosedependent decreases in the viability of both the ALDH high and the ALDH low cervical cancer cells ( Figure 5B). ALDH high SiHa cells were significantly more resistant to cisplatin concentrations of ≥6 µg/mL than the ALDH low SiHa cells. ALDH high CaSki cells were significantly more resistant to ≥1 µg/mL cisplatin than the ALDH low CaSki cells. ALDH high HT-3 cells were significantly more resistant to concentrations of ≥0.5 µg/mL cisplatin than the ALDH low HT-3 cells. These results indicated that ALDH high cells are more resistant to cisplatin than ALDH low cells when exposed to the proper concentration for a limited period of time. The viability of the ALDH high C33A cells was significantly greater than that of the ALDH low cells after exposure to any concentration of cisplatin. Cisplatin caused a dose-dependent decrease in the viability of the ALDH low C33A cells. Surprisingly, the viability of the ALDH high C33A cells was significantly enhanced after exposure to cisplatin compared with cells that had not been treated with cisplatin.
Cell viability was also determined by the MTT assay after exposure to constant concentration of cisplatin for 24, 48, or 72 h ( Figure 5C). Cisplatin caused a time-dependent decrease in the viability of both ALDH high and ALDH low cells from the SiHa, CaSki and HT-3 cells. ALDH high SiHa cells were significantly more resistant to ≥48 h of treatment with cisplatin than ALDH low cells. In the CaSki and HT-3 cells, ALDH high cells were significantly more resistant than ALDH low cells to cisplatin treatment for ≥24 h. The results from these 3 cell lines indicate that ALDH high cells are more resistant to constant concentration of cisplatin than ALDH low cells for certain periods of time. The viability of ALDH high C33A cells was significantly greater than that of ALDH low cells after exposure to cisplatin. Cisplatin caused a time-dependent decrease in the viability of ALDH low C33A cells, while the viability of ALDH high C33A cells was significantly enhanced following exposure to cisplatin compared to cells that had not been treated with cisplatin.
In summary, these results suggest that ALDH high cervical cancer cells are more resistant to chemotherapy than ALDH low cells.

ALDH high cells express high levels of stem cellassociated markers
Stem cell-related transcription factors are important for maintaining the self-renewal of embryonic stem cells.
To clarify whether the ALDH high cervical cancer cells express stem cell-related transcription factors, western blot analysis was performed to assess the expression of OCT4, NANOG, KLF4 and BMI1 in ALDH high and ALDH low cells. ALDH high C33A and HT-3 cells were found to express higher levels of OCT4, NANOG, KLF4 and BMI1 than ALDH low cells ( Figure 6A). IHC analysis was also performed on the tumorspheres formed by the ALDH high and ALDH low SiHa cells to evaluate the expression of stem cell-associated markers ( Figure 6B). Similar to the western blot analysis, the stem cell-associated transcription factors OCT4, NANOG, KLF4 and BMI1 were detected in the tumorspheres formed by the ALDH high SiHa cells but not the ALDH low SiHa cells. These data indicate that ALDH high cervical cancer cells display a nuclear stemness signature.

ALDH high cells from primary cervical cancers possess CSC characteristics
Our data suggest that ALDH high cells from cervical cancer cell lines display characteristics of CSCs. However, whether ALDH high cells derived from primary cervical cancers possess the same characteristics remained unknown. To address this question, primary cervical cancer tissues were processed into single cell suspensions and injected subcutaneously into NOD/SCID mice to create tumor xenografts. Only 5 out of 28 primary cervical cancer specimens were successfully serially transplanted for 5 generations; the estimated engraftment rate was 18%. The generated xenografts were resected and dissociated into single cell suspensions. Cells were plated in 24well plates at a density of 10 3 cells/well and cultured for tumorsphere formation. All tumorspheres generated from the same cervical cancer tissue were collected, digested into single cell suspensions, labeled with the ALDEFLUOR kit, analyzed and sorted by FACS.
A subpopulation of ALDH high cells (11-23%) was detected in all 5 of the primary tumors tested ( Figure  7A). The 10% of the cell population with the highest and the lowest ALDH activity were sorted as the ALDH high and the ALDH low cells, respectively, for the subsequent experiments. As shown in Figure 7B, the ALDH high , but not the ALDH low , primary cervical cancer cells were capable of generating tumorspheres in suspension culture. ALDH high cells formed tumorspheres with a frequency of approximately 10% in 3 consecutive passages ( Figure 7C), while ALDH low cells did not generate tumorspheres (but did generate some cell aggregates). These data suggest that the ALDH high primary cervical cancer cells have the ability to self-renew.
To study the tumor formation capacity of the ALDH high and the ALDH low primary cervical cancer cells, 10 6 , 10 5 , 10 4 or 10 3 cells were subcutaneously injected into NOD/SCID mice, and the mice were monitored for tumor growth for 20 weeks. As summarized in Table 2, the ALDH high primary cervical cancer cells exhibited enhanced tumorigenicity compared with the ALDH low cells. Among the 5 primary tumor samples, the tumor-initiating frequency of the ALDH high cells was as high as  times that of the ALDH low cells (Table 2). Furthermore, the tumors generated from ALDH high cells were significantly larger and grew faster than the tumors derived from their ALDH low counterparts. These results indicate that ALDH activity may be a functional marker of CSCs in primary cervical cancer.

DISCUSSION
Several approaches have been utilized to identify CSCs from various human malignancies, including cell surface markers, side population phenotype, spheres formation and ALDH activity assays [21]. ALDH activity (measured using the ALDEFLUOR assay) was first used to isolate leukemia stem cells [28]. Subsequently, ALDH activity has been successfully used as a CSC marker for many solid tumors, including breast [29], colon [30,31], bladder [33], prostate [34], lung [36], head and neck [38], endometrium [39], ovary [40] and thyroid [47]. Therefore, ALDH activity may have potential as a promising universal marker for the identification and isolation of stem cells from various solid tumors. However, previous reports have not indicated whether ALDH activity can also be used as a CSC marker in cervical cancer.
In the present study, the ALDH high cells isolated from 4 cervical cancer cell lines and 5 primary cervical cancer xenografts were demonstrated to fulfill the functional criteria for CSCs. Firstly, using the tumorsphere formation assay, ALDH high cervical cancer cells were found to have self-renewal capacity. ALDH high cells formed significantly more tumorspheres in both single cell culture and low density cell culture than ALDH low cells. Furthermore, ALDH high cells formed more tumorspheres in single cell culture than in low density cell culture ( Figure 2B and 2C). Similar results have been reported for mammospheres, and cell aggregation in the low density cell cultures might contribute to the decreased formation of tumorspheres [29]. Secondly, ALDH high cervical cancer cells could differentiate and reconstitute the cellular hierarchy in vitro and in vivo. After 2 weeks of culture in medium containing FBS, ALDH high cells differentiated into many ALDH low cells, while ALDH low cells generated few ALDH high cells and predominantly maintained the ALDH low phenotype ( Figure 4A-4D). The ALDH1-positive population could be detected in all of the tumor xenograft tissues formed by the ALDH high cells, but no ALDH1-positive cells were found in the tumors formed by the ALDH low cells ( Figure  4E). Therefore, only the ALDH high cells, not the ALDH low cells, could differentiate and re-establish the cellular hierarchy in vitro and in vivo. Thirdly, ALDH high cervical cancer cells had tumor initiating capacity in vivo. The tumors formed by the ALDH high cells were larger and grew faster than those derived from the ALDH low cells ( Figure  3A). A shorter tumor-free period and a lower tumor-free rate were observed in mice injected with the ALDH high cells than in mice injected with ALDH low cells ( Figure 3B). The tumor-initiating frequency of the ALDH high cells was significantly higher than that of the ALDH low cells ( Table  1). Taken together, these data indicate that the ALDH high cells are indeed CSCs in cervical cancer, similar to the results of previous ALDH studies in other solid tumors [29,30,33,34,36,41].
ALDH expression and activity has been reported to be significantly higher in taxane-and platinum-resistant ovarian cancer cell lines [40]. Rahadiani et al. have reported that ALDH high endometrioid adenocarcinoma cells are more resistant to cisplatin treatment than ALDH low cells [39]. In this study, the ALDH high cells were more resistant to cisplatin treatment than the ALDH low cells (Figure 5), suggesting that ALDH high CCSCs exhibit chemoresistance similar to the CSCs found in other solid tumors [39,48]. Surprisingly, the viability of the ALDH high C33A cells was significantly enhanced following exposure to cisplatin ( Figure 5B and C). We attribute this to stimulated proliferation or enhanced activity of succinic acid dehydrogenase, which is the enzyme activity measured by the MTT assay. Further experiments are necessary to clarify the mechanisms that underlie the enhanced viability of the ALDH high C33A cells.
The direct isolation of CSCs from uncultured human primary colon [30] and prostate [34] cancer cells and sorting based on ALDH activity has been reported. ALDH high CSCs in breast cancer [29] and pancreatic cancer [37] have been successfully isolated from primary cancer xenografts established in mice. In the present study, uncultured primary cervical cancer cells were first used to investigate the tumorigenicity of the ALDH high and the ALDH low cells. Unfortunately, neither the ALDH high nor the ALDH low cells from primary cervical cancer resulted in tumors in NOD/SCID mice, similar as the report about ovarian cancer [40]. Three possibilities may contribute to the failure of CSC isolation directly from primary uncultured cervical cancer cells: (1) ALDH-positive noncancerous cells, including white blood cells, stromal cells or normal stem cells, were abundant in the primary cancer tissues and contaminated the cancerous cells during tumor formation; (2) all primary cervical cancer tissues obtained from surgery are likely below the clinical stage of II a, and most of these cancerous cells may be too fragile to be sorted; or (3) the cells may be not resilient enough to form tumors in NOD/SCID mice. In the present study, CSCs were isolated from 5 generations of cervical cancer serially xenografted in NOD/SCID mice but not from uncultured primary cervical cancer cells. The following 3 possibilities may explain this phenomenon: (1) After more than 5 serial xenograft generations in NOD/SCID mice, primary cervical cancer cells acquired the capacity to form tumors; this idea is consistent with the notion that cancer cells can acquire enhanced tumorigenicity through serial transplantation in mice. (2) In this study, only 5 of the 28 cases of primary cervical cancer (approximately 18%) could successfully be serially transplanted for more than 5 generations in mice, indicating that the cancer cells www.impactjournals.com/oncotarget in most primary cancer tissues were too fragile to form tumors in NOD/SCID mice. (3) The cells used for the tumor formation assay were isolated from tumorspheres, which decreases the likelihood of contamination with ALDH-positive non-cancerous cells from the primary cancer tissues.
ALDH1 has been reported to be a marker for normal mammary stem cells [29]. In this study, ALDH1 expression was found in the basal cells of normal cervical tissue ( Figure 1A, upper panel). The basal cells were recognized to contain cervical stem cells. Therefore, ALDH1 may be a marker of normal cervical stem cells. Further experiments should be designed to isolate and test the ALDH1-positive cervical basal cells to verify whether ALDH1 can be used as a marker of normal cervical stem cells.
In summary, this report is the first to describe the use of high ALDH activity to isolate CSCs from cervical cancer cell lines and primary cervical cancer cells. These ALDH high cervical cancer cells possess the ability to selfrenew and differentiate and have enhanced tumorigenicity. Additionally, these cells exhibit chemoresistance and express high levels of stem cell-related transcription factors. Based on this study, ALDH activity may be used as a cytoplasmic marker for CCSCs, and a target to explore novel strategies for diagnosis, prognosis and therapy.

Ethics Statement
Investigation has been conducted in accordance with the ethical standards and according to the Declaration of Helsinki and according to national and international guidelines and has been approved by the review board of the First Affiliated Hospital of Xi'an Jiaotong University.
Human tissue specimens, primary cervical cancer tissue processing and xenograft lines A total of 17 normal cervical tissues and 53 cervical cancer tissues were obtained from the First Affiliated Hospital of Xi'an Jiaotong University. The procedures followed approved medical ethics practices, and the patients provided their informed consent before the specimens were collected. Fresh cervical cancer tissues were obtained from 28 patients after radical hysterectomy and used for xenograft experiments. Single cell suspension was generated by mincing and digesting the tissue with 100 U/mL collagenase IV (GIBCO, Grand Island, NY) in basal medium at 37°C overnight. Xenograft lines were established by subcutaneous implantation of the primary cervical cancer cells in 6-to 8-week old NOD/SCID mice (Charles River Laboratories, Wilmington, MA). Once established, the solid tumor xenografts were serially passaged using the same technique.

Immunohistochemistry (IHC)
Formalin-fixed and paraffin-embedded tissue specimens were sliced into 4 mm sections, which were then deparaffinized and hydrated. An endogenous antigen retrieval procedure was performed using citric acid buffer (10 mmol/L citrate buffer, pH 6.0). The slides were incubated with a mouse monoclonal antibody raised against human ALDH1 (BD Biosciences, Franklin Lakes, NJ) or Ki67 (Santa Cruz, CA) overnight at 4°C, then with secondary antibodies for 30 min at room temperature, followed by diaminobenzidine development. All slides were examined under an Olympus-CX31 microscope (Olympus, Tokyo, Japan).

Flow cytometry analysis and FACS isolation of cells
The ALDH enzymatic activity of the cells was measured using the ALDEFLUOR kit (Stem Cell Technologies, Vancouver, BC, Canada), according to the manufacturer's instructions. The brightly fluorescent ALDH-expressing cells were detected using a FACSCalibur or FACSAria flow cytometer (BD Biosciences). As a negative control, cells were stained under identical conditions after treatment with the specific ALDH inhibitor diethylaminobenzaldehyde (DEAB). The data were analyzed using FlowJo software (Tree Star Inc., Ashland, USA). For FACS, the cells were labeled using the ALDEFLUOR kit and sorted using a FACSAria cell sorter (BD Biosciences). www.impactjournals.com/oncotarget

Tumorsphere culture
Cells were maintained in stem cell media consisting of DMEM/F12 basal media, N2 and B27 supplements (Invitrogen), 20 ng/mL human recombinant epidermal growth factor (EGF) and 20 ng/mL basic fibroblastic growth factor (bFGF; PeproTech Inc., Rocky Hill, NJ). For the tumorsphere formation assay, cells were plated at a density of 200 cells/well in 24-well ultra-low attachment plates or at a density of 1 cell/well in 96-well plates and maintained in stem cell medium. Tumorspheres that arose within 2 weeks were recorded. For serial tumorsphere formation assays, the spheres were harvested, disaggregated with 0.25% trypsin/EDTA, filtered through a 40 µm mesh and re-plated as described above. For each cell type, triplicate samples were done and the spheres were counted by two individuals in a blind fashion.

Drug resistance and MTT assay
For drug resistance assays, cells were plated in 96well plates at a density of 10 4 cells/well and allowed to recover overnight before initiating drug treatments. The cells were exposed to various concentrations of cisplatin (0, 3, 6, 12, 24, or 48 µg/mL for SiHa and C33A cells or 0, 0.5, 1, 2, 4, or 8µg/mL for CaSki and HT-3 cells) for 24 h, and the cell viability was measured. In separate experiments, the cells were exposed to a constant concentration of cisplatin (3 µg/mL for SiHa and C33A cells or 1 µg/mL for CaSki and HT-3 cells) for 24, 48 or 72 h, and the cell viability was measured.

In vivo tumor formation assays
The ALDH high and ALDH low cells were sorted, re-suspended in 200 µL of 1:1 PBS/Matrigel (BD Biosciences) and injected subcutaneously into the flanks of 6-to 8-wk old female NOD/SCID mice; the left flank of the mouse received the ALDH high cells, whereas the right flank received the ALDH low cells. Engrafted mice were inspected twice per week by visual observation and palpation for the appearance of tumors. The tumor volume (V) was determined from the length (a) and the width (b) of the tumor, using the formula V=ab 2 /2 [49]. A portion of each tumor tissue was fixed in 10% formaldehyde and embedded in paraffin for IHC analysis. The frequency of tumorigenic cells (estimated with upper-lower limits) was calculated by limiting-dilution analysis [50].

Statistical analysis
Statistical analyses were performed using GraphPad Prism 5.01 software (La Jolla, CA, USA). In comparisons of 2 groups, Student's t-test was used to determine the statistical significance. To examine differences among 3 groups, an ANOVA analysis was performed. Kaplan-Meier survival analysis was performed and survival curve comparison analyses were performed using the log-rank (Mantel-Cox) test. P values of ≤0.05 were regarded as statistically significant.