Characterization of Biomarkers of Tumorigenic and Chemoresistant Cancer Stem Cells in Human Gastric Carcinoma

Gastric cancer is the fourth most common cancer in frequency and the third leading cause of cancer mortality in the world. Ninety-five percent of gastric cancers are gastric carcinomas, which are divided into two types depending on their localization in the stomach: adenocarcinomas of the cardia whose etiology remains unclear, and noncardia gastric carcinomas for which the main factor is a chronic infection by Helicobacter pylori (H. pylori). Infection with H. pylori, classified as a class 1 carcinogen by the World Health Organization, induces a chronic inflammation evolving over decades from a chronic atrophic gastritis to intestinal metaplasia, dysplasia, and finally adenocarcinoma (1, 2). Some cases also include Lynch syndromes (microsatellite instability, MSI) and Epstein–Barr Virus (EBV) infection. The classification of gastric carcinomas is based essentially on histologic criteria. The Lauren classification distinguishes two main subtypes, the intestinal type, which represents the majority of the cases, and the diffuse type (3). The intestinal type is composed of glands having more or less preserved their organization and differentiation state, or having acquired intestinal characteristics; it is subclassified into tubular, mucinous, or papillary carcinoma in the WHO classification of gastric carcinoma (4). The diffuse type is poorly cohesive, composed of isolated cells (often signet ring cells) producing mucins. These classification systems have little clinical utility, as they cannot orientate patient therapy. With the exception of Her2 positivity which orientates toward a specific treatment, treatment is still based on surgery combined with conventional chemotherapy and/or radiotherapy, and the 5-year survival rates remain under 30% in most countries (5).

Recently, the Cancer Genome Atlas Research Network and Wang and colleagues published a molecular profiling of gastric carcinomas based on two studies with 295 cases and 100 cases, respectively. Both studies led to a classification of gastric carcinomas into four main subtypes according to their molecular profiles: (1) EBV⁺ tumors (frequent PIK3CA mutations, extreme DNA hypermethylation), (2) MSI tumors (elevated mutation rates, hypermethylation), (3) genomically stable tumors (enriched for the diffuse type; driver mutations include CDH1, RHOA, cytoskeleton, and cell junction regulators), and (4) chromosomal instability tumors (marked aneuploidy, focal amplification of tyrosine kinase receptors; refs. 6, 7). These studies were performed without distinguishing between cardia and noncardia gastric carcinomas, whose etiology is different.

Tumors are heterogeneous, composed of cells which are more or less differentiated, and not all proliferative. Over the last decade, extensive research has focused on the discovery and the characterization of cancer stem cells (CSC) at the origin of cancers in numerous organs. Tumors are hierarchically organized with CSCs at the top of this pyramid and at the origin of tumor initiation, heterogeneity, and propagation (8–10). The CSCs correspond to a subpopulation of cells within the tumor defined by self-renewal, asymmetrical division, and differentiation properties, giving rise to the more or less differentiated cells composing the tumor mass. CSCs can stay in quiescence under some conditions, resist conventional therapies, and be at the origin of tumor relapse and metastasis. The definition of CSCs remains largely operational and based on functional assays that register their self-renewal and tumorigenic properties, assessed by the formation of new heterogeneous tumors after xenograft in vivo, and of tumorspheres in particular culture conditions in vitro (8–10).

Indeed CSC may display both genetic and phenotypic heterogeneity, markers allowing their identification have been characterized in tumors of different organs, including CD133, CD44, and CD24 among those studied (8, 10–16). More recently, the activity of aldehyde dehydrogenases (ALDH), intracellular enzymes involved in oxidation of aldehydes and retinoic acid signaling, also led to the identification of CSCs in tumors of the breast (17), lung (18), colon (19), and other organs (20).

In the stomach, the existence of CSCs has been subject to debate. The first study performed by Takaishi and colleagues on gastric carcinoma cell lines proposed CD44 as a gastric CSC marker, but this marker was expressed in three out of six cell lines studied, and confirmation in primary tumors was lacking (21). Then, the study performed by Rocco and colleagues on 12 human primary gastric carcinomas failed to demonstrate tumor-initiating properties of CD133⁺- and CD44⁺-sorted cells after xenograft in both NOD/SCID and nude immunodeficient mice (22).

Another important point concerns the origin of the CSCs. As Houghton and colleagues and our group reported in mouse models of Helicobacter-induced gastric carcinogenesis, gastric dysplasia and carcinoma may originate from the transformation of a local epithelial stem cell or of a bone marrow (BM)–derived stem cell (23–25). In this model, dysplastic lesions were composed of CD44⁺ cells, regardless of their BM or local origin (24). In addition, the heterogeneity of gastric carcinomas suggests that gastric CSC markers, if they indeed exist, may be different according to the origin and/or the histologic type of gastric carcinoma.

In this study, we performed an extensive screening of the expression of putative cell surface markers of CSCs as well as ALDH activity in order to identify biomarkers allowing the detection and isolation of tumorigenic and chemoresistant CSCs in human primary intestinal- and diffuse-type noncardia gastric carcinoma.

Fresh tumors samples were collected from gastric surgical wastes from patients who underwent gastrectomy for noncardia gastric carcinoma and for whose informed consent was obtained. Fresh samples of tumor and paired nontumor tissues were transported in DMEM medium with 20% FCS, 50 IU/mL penicillin, 50 μg/mL streptomycin, 50 μg/mL vancomycin, and 15 μg/mL amphotericin-B. Samples were minced in small pieces of 2 mm × 2 mm size and were subcutaneously transplanted into the right dorsal flank of 7-week-old male NSG mice under 2.5% isoflurane anesthesia (Belamont). Alternatively, after mechanical mincing, cells were dissociated by incubation in a solution of 1 mg/mL collagenase IV and 0.2 mg/mL hyaluronidase in DMEM (Sigma) for 1 hour at 37°C with shaking (15), then suspended in 100 μL of 7 mg/mL ice-cold Matrigel (BD Biosciences) for subcutaneous injection. Xenografts were carried out within 3 to 5 hours following gastrectomy. The tumor size was monitored with callipers once a week, and tumor volume was estimated as (D² x d)/2, where D is the large diameter and d is the small diameter (26). At the end of the experiments (until 10 months after engraftment for the primary xenograft) and when tumor reached approximately 500 mm³, mice were sacrificed by cervical dislocation and tumors were immediately harvested and processed for analyses. Secondary tumors were amplified subcutaneously in mice by serial transplantation of pieces of tumor bulk, or by injection of tumor cells in Matrigel. For xenograft experiments in extreme limiting dilution assay (ELDA), 10,000 to 30 FACS-sorted cells were subcutaneously injected with Matrigel; tumor size was recorded twice a week.

Gastric carcinoma cell lines were cultured in DMEM/F12 media for AGS (ATCC CRL1739) and in RPMI1640 media for NCI87 (ATCC CRL-5822), MKN45, MKN74, MKN7, and MKN28 (all from RIKEN) cells, supplemented with 10% heat-inactivated FCS, 50 IU/mL penicillin and 50 μg/mL streptomycin (all from Invitrogen) at 37°C in a 5% CO₂ atmosphere (26, 27). All cell lines were routinely verified mycoplasma free (by PCR) and were tested and authenticated by short tandem repeat (STR) profiling within 6 months preceding the experiments (last STR profiling report, October 2015; LGC Standards). Cell viability was assessed by the trypan blue exclusion method.

Cells were dissociated by collagenase/hyaluronidase procedure from fresh patient-derived tumor xenografts (PDX), passed through a 70-μm mesh filter (BD), and red blood cells were removed by incubating in a solution containing 2 mmol/L KHCO₃, 0.1 mmol/L EDTA, and 170 mmol/L NH₄Cl for 8 minutes at 4°C. Then, 100,000 cells in 100 μL PBS-0.5% BSA-2 mmol/L EDTA (Sigma) were stained with 3 to 5 μL of fluorescent-labeled primary antibodies including EPCAM-FITC (Stem Cell Technologies) or EPCAM-VioBlue (MACS-Miltenyi Biotec), CD10-PE, CD24-PE, CD73-PE, CD49f-PE, CD105-PE, CD166-PE, CD90-PECy5, CD44-PE, CD44-APC, CD338-APC (all from BD), and CD133-PE (MACS-Miltenyi Biotec) for 20 minutes at 4°C. Cells were rinsed twice with PBS-0.5% BSA-2 mmol/L EDTA containing 50 μg/mL 7-aminoactinomycin-D (7-AAD; BD) before being analyzed using a FACSCanto II instrument and DIVA software (BD; refs. 26, 27). The ALDEFLUOR Kit (Stem Cell Technologies) was used to detect ALDH activity according to the manufacturer's instructions. Dead cells were excluded based on light scatter characteristics and 7-AAD positivity. For SP cells analysis, cells were incubated with 10 μg/mL Hoechst-33342 in HBSS-2% FCS for 60 minutes at 37°C or, when indicated, in ALDEFLUOR buffer for 30 minutes at room temperature, with or without 100 μmol/L verapamil or 50 μmol/L reserpine (Sigma), and then washed with ice-cold HBSS-2% FCS. The Hoechst-33342 dye was excited at 375 nm, and its fluorescence was dual wavelength analyzed (blue, 402–446 nm; red, 650–670 nm; ref. 28). Cell sorting was performed on 5 to 10 million cells stained with primary fluorescent-labeled antibodies or ALDEFLUOR reagent and 7-AAD, on 7-AAD–negative cells using a FACSAria (BD).

A total of 1,000 FACS-sorted cells were plated in nonadherent 24-well plates (or alternatively 200 cells in 96-well plates) previously coated with a 10% poly(2-hydroxyethyl methacrylate) solution in 95% ethanol (v/v; Sigma), in DMEM-F12 media supplemented with 20 ng/mL human-epidermal growth factor, 20 ng/mL basic-fibroblast growth factor, 5 μg/mL insulin, 0.3% glucose, 50 IU/mL penicillin, and 50 μg/mL streptomycin (Sigma; ref. 26). For PDX cells, the media were supplemented with 5% FCS for the first 2 days of culture and were then replaced by serum-free media. After 7 days, the number of spheroids/well was counted under light microscopy using a ×20 objective. For drug treatment experiments, 5-day tumorspheres grown in nonadherent 96-well plates (8 < n < 10 per condition) were treated with 10 to 20 μmol/L of verapamil with 5-fluorouracil, doxorubicin, and cisplatin (all from Sigma). After 48 hours, the number of tumorspheres was recorded. The self-renewal ability of residual viable cells dissociated from tumorsphere by trypsin/EDTA procedure and seeded in new nonadherent 96-well plates (8 < n < 10 per condition) was evaluated after 6 to 9 days.

Ethic statements, histology, immunohistochemistry, immunofluorescence procedures, and statistical analysis are described in Supplementary Materials.

Fresh gastric carcinoma and nontumor tissue samples were collected by pathologists upon surgical resection from consenting patients who underwent gastrectomy for noncardia gastric carcinoma at the University Hospital Center and the Bergonié Institute in Bordeaux. Among the 37 cases studied, the median age was 72 years, 59.5% were males and 54.1% did not receive preoperative chemotherapy (Supplementary Table S1). Approximately half of the tumors were tubular and one third were poorly cohesive according to the WHO and Lauren classifications of gastric carcinomas. More than half were high grade (51.4%), highly penetrant (T4 = 51.3%) with lymph node invasion (N1–4 = 67.6%), and at stages 3 to 4 (51.4%). When tissue size was sufficient, pieces were dissociated by enzymatic procedures, and single cells were analyzed by flow cytometry for the expression of EPCAM, an epithelial marker, in combination with CD24, CD133, and CD44 as putative CSC markers, and 7-AAD to exclude dead cells (Fig. 1A). To overcome the problem of the small size of the human samples which was a limiting factor in the study, we developed PDX models in NSG immunodeficient mice. For this, small pieces of fresh tissue from the patient's biopsy were subcutaneously xenografted in mice (3 < n < 10 mice per case). Nontumor tissues xenografted as controls never led to tumor growth over the 10 months. Among the 37 tumors xenografted, 8 led to the growth of secondary tumors serially transplantable in mice; 7 were intestinal type (GC04, GC07, GC10, GC35, GC40, and GC44), and 1 was diffuse type (GC06; Supplementary Table S1). There was no significant association between PDX growth and the following characteristics: patient's gender, age, preoperative treatment, WHO and Lauren classifications of gastric tumor histologic type, grade, tumor-node-metastasis classification, and stage (Supplementary Table S1). PDXs reached a 500 mm³ size between 2 to 6 months (mean 16.7 ± 3.4 weeks) after the first passage (P) (P1) in mice, and earlier following successive passages (after 10.9 ± 5.9 weeks at P2 and 7.2 ± 0.8 weeks at P5; Fig. 1B). Case GC42, a mucinous type according to the WHO classification, was excluded because tumors developed slowly and were mostly composed of mucus, rendering its study impossible. Histopathologic analyses confirmed that the PDXs obtained between P1 and P5 remained similar to the respective patients' primary tumor for all cases (Fig. 1C) except GC07, which appeared to dedifferentiate after P2 and therefore was excluded (data not shown).

The expression of CD24, CD133, and CD44 was evaluated by flow cytometry in live (7-AAD negative) EPCAM⁺ epithelial cells dissociated from freshly collected paired nontumor and tumor gastric tissue samples from 7 cases. In nontumor gastric tissues, CD24 and CD133 were expressed in about half of the cells, whereas CD44 was expressed only in 10% ± 9% of cells. In paired tumors, CD24, CD133, and CD44 expression was significantly higher compared with nontumor cells, CD24 being expressed in most of the tumor cells (90 ± 8%), CD133 in 71% ± 17%, and CD44 in 27% ± 17% of the tumor cells (Fig. 2A). The percentage of tumor cells expressing these markers were then evaluated on cells dissociated from freshly collected serial xenografts from P1 to P4 of the 6 PDXs cases (Fig. 2B). The percentage of cells expressing the different markers remained stable in serial PDXs from P1 to P4, EPCAM being expressed in most of the cells (>80%) followed by CD24 and CD133 expressed at a relatively high proportion of cells, except in GC35. In all cases, CD44 was expressed in less than a third of total cells. These results were confirmed by immunohistochemistry analyses of the expression of CD44 on patient's nontumor and tumor tissues and on the corresponding P2–3 PDXs for the 6 cases studied (Fig. 2C). In the nontumor area, CD44 was expressed at a low level, preferentially in cells in the isthmus region of gastric glands in the area of gastritis, as previously reported (26). In patients' tumors, CD44 was expressed in some tumor cells but not all, mainly at the periphery of the tumor islets for the intestinal-type tumors. A similar pattern of CD44 expression was observed in PDXs, confirming that the cellular heterogeneity of the primary tumor was reproduced in serial PDXs.

We then analyzed the expression of seven additional cell surface markers as putative markers of gastric CSCs, i.e., CD10, CD49f, and CD166 described in CSCs of other organs (29–31), CD73, CD90, and CD105 as the main markers of mesenchymal stem cells (27, 32), as well as ALDH activity in cells from five PDXs (freshly collected at P2–P4) and five gastric carcinoma cell lines (seven intestinal type: GC04, GC10, GC35, GC44, MKN74, MKN7, NCI87; and three diffuse type: GC06, AGS, MKN45; Fig. 3A). CD10 was negative except for 2 of the 10 cases studied. CD49f was expressed in more than 80% of the cells in all cases, similar to EPCAM. For the other markers, the expression pattern was more homogeneous in PDXs than in cell lines which exhibited more heterogeneity, being either highly positive or negative. CD133 was expressed only in PDXs cells and not in gastric carcinoma cell lines. In PDXs, CD49f and CD24 were highly expressed, followed by CD133 and CD90 expressed in nearly half of the cells, then by CD73 expressed in more than a third of the cells. At a level similar to CD44, CD166 was expressed in 21% ± 13% of the cells. CD105 expression and ALDH activity were detected in only 9% ± 6% and 8% ± 5% of the cells, respectively (Fig. 3A). Results from flow cytometry costaining analyses revealed that CD166 and CD44 were coexpressed and detected the same cell subpopulation (Fig. 3B). The majority of CD44⁺ cells were positive for CD24, CD133, and CD73. Less than 50% of CD44⁺ cells were positive for ALDH, CD105, and CD90. Interestingly, ALDH activity but not CD90 and CD105 expression was recorded mainly in CD44⁺ cells, showing that ALDH⁺ cells representing a core within the CD44⁺ subpopulation of cells (Fig. 3B).

In order to evaluate the tumorigenic properties of the cells expressing or not CD44, CD133, CD73, CD166, CD90, and CD105, P2–P3 tumors of 3 PDX cases, GC04, GC06, and GC10, were freshly dissociated and 7-AAD-EPCAM⁺ cells either positive or negative for these markers were sorted by FACS and submitted to the tumorsphere assay. Similar experiments were performed on 7-AAD–FACS-sorted cells based on ALDH activity. In all of the cases, EPCAM⁺CD133⁺, EPCAM⁺CD44⁺, EPCAM⁺CD73⁺, EPCAM⁺CD166⁺, and ALDH⁺ cells formed significantly more tumorspheres after 10 days of in vitro culture than their respective negative counterparts (Fig. 3C). CSCs forming tumorspheres were essentially present in CD44⁺, CD166⁺, and ALDH⁺ subpopulations, and to a lesser extent in CD133⁺ and CD73⁺ subpopulations; they were essentially CD90⁻ and CD105⁻. The high tumorsphere capacity of ALDH⁺ cells was confirmed on both MKN45 and MKN74 cell lines (Fig. 3C).

To confirm these results in vivo, xenografts were performed in mice with 7-AAD-EPCAM⁺ FACS-sorted cells based on the expression of CD133 and CD44 on GC04, GC06, and GC10 PDXs. The same experiments were performed based on ALDH activity on GC06, GC10, MKN45, and MKN74 cells (Table 1). In all cases, tumors developed at a significant higher frequency in EPCAM⁺CD44⁺ cells (1/29 to 1/1,020) than in their respective EPCAM⁺CD44⁻ cells (1/568 to 1/28,963), and in EPCAM⁺CD133⁺ cells (1/105 to 1/1,911) than in their respective EPCAM⁺CD133⁻ cells (1/781 to 1/66,876; Table 1). CSC frequency was higher in EPCAM⁺CD44⁺ cells than in EPCAM⁺CD133⁺ cells, confirming that CD44 is a better marker of gastric CSCs than CD133, as also determined by the in vitro tumorsphere assays. ALDH⁺ cells led to the development of tumors at a significant higher frequency than the respective ALDH⁻ cells (1/38 to 1/746 for ALDH⁺ cells vs. 1/372 to 1/8,024 for ALDH⁻ cells) for all cases studied, with the exception of MKN45 which was highly tumorigenic at the doses studied (Table 1). For both PDXs cases GC10 (intestinal type) and GC06 (diffuse type), the CSC frequencies were high in ALDH⁺ cells (1/81 and 1/38, respectively) and in EPCAM⁺CD44⁺ cells (1/29 and 1/352, respectively), and both were higher than in EPCAM⁺CD133⁺ cells (1/105 and 1/1,658, respectively), suggesting that ALDH and CD44 are more specific markers of gastric CSCs than CD133. Finally, we showed that the CSCs contained in EPCAM⁺CD44⁺ and ALDH⁺ FACS-sorted cells generated tumors that recapitulated the phenotypic heterogeneity of the initial tumors, giving rise both to CD44⁺ cells containing CSCs and to more differentiated CD44⁻ cells (Fig. 4A). The asymmetric division and differentiation properties of these CSCs was confirmed in vitro with MKN45 cells; ALDH⁺ FACS-sorted cells were able to reproduce heterogeneous tumorspheres in vitro, composed of a similar proportion of cells with ALDH⁺ activity compared with the initial situation and expressing CD44, and of CD44⁺ALDH⁻ and CD44⁻ALDH⁻ cells incorporating the Hoechst-33342 stain (Fig. 4B). In addition, the CD44⁺ALDH⁺ FACS-sorted cells generated more tumorspheres than the CD44⁺ALDH⁻ cells which generated less but still a significant number of tumorspheres and which therefore could correspond to progenitor/transit amplifying cells. The CD44⁻ALDH⁻ cells may correspond to more differentiated cells with very limited proliferation capacities as they formed significantly less or no tumorspheres compared with the CD44⁺ALDH⁻ and CD44⁺ALDH⁺ cells (Fig. 4C).

The combined analysis of CD44 expression, ALDH activity, and Hoechst-33342 incorporation was assessed on live MKN45 tumorsphere (as in Fig. 4B) during their development. In young, small tumorspheres (after 5 days), most cells were positive for both CD44 and ALDH activity and were negative for Hoechst-33342 (Supplementary Fig. S1). When tumorspheres became bigger (after 10–15 days), only a fraction of cells remained CD44⁺ALDH⁺ and Hoechst-33342, which may correspond to CSCs, with the appearance of CD44⁺ALDH⁻ Hoechst-33342⁺ cells which may correspond to progenitor/transit amplifying cells (Supplementary Fig. S1).

An important property of CSCs is to be resistant to conventional therapies, leading to tumor recurrence and metastasis after treatment (8, 9). Verapamil treatment, known to inhibit drug efflux systems, restored Hoechst-33342 incorporation in ALDH⁺ cells in MKN45 and GC10 tumorspheres in vitro (Fig. 5A). This effect was confirmed by flow cytometry analyses on MKN45 cells and confirmed at a lesser extent with reserpine, another inhibitor of drug efflux systems (Fig. 5B). The drug efflux properties of CSCs are usually assessed by the functional analysis of the Side Population (SP), a minor subpopulation of cells defined by Hoechst-33342 stain efflux properties in specific experimental conditions. In these particular experimental conditions, we found a consistent proportion of Hoechst-SP cells in MKN45 cell line (3.9% ± 0.8 % of total cells; Fig. 5C) but not in MKN74 cell line (<0.2%, data not shown) as previously reported by others (28). The percentage of MKN45 SP cells was significantly decreased with verapamil and reserpine compared with untreated cells (0.6% ± 0.1% and 1.2% ± 0.2%, respectively, vs. 3.9% ± 0.8%; Fig. 5C). The determination of ALDH activity within SP cells was then assessed in the experimental conditions of the ALDEFLUOR assay, those of the SP cells assay being incompatible and leading to ALDEFLUOR substrate clearing (Supplementary Fig. S2; Supplementary Methods). Results showed that most of the ALDH⁺ cells were present in the Hoechst-subpopulation of SP-like cells (Fig. 5D). Combined together, these results suggest that ALDH⁺ cells could possess drug efflux properties and may be more resistant to treatments.

To explore this hypothesis, first, the sensitivity of MKN45 and MKN74 cells to drugs commonly used in gastric carcinoma treatment was determined on cell viability in adherent culture conditions. Both cell lines showed dose-dependent antiproliferative responses to 5-fluorouracil, doxorubicin, and cisplatin treatments, which were increased in the presence of verapamil (Supplementary Fig. S3). Second, MKN45 and MKN74 cells were sorted by FACS based on ALDH activity, and the viability of ALDH⁺ and ALDH⁻ cells was evaluated in the presence of 5-fluorouracil, doxorubicin, and cisplatin in combination with verapamil in vitro. This assay cannot be performed on cells from PDXs, given that these cells are not cultivable in adherent culture conditions and that only ALDH⁺ cells and not ALDH⁻ cells can form tumorspheres in vitro (Fig. 3C). In both cell lines, ALDH⁺ cells were more resistant than ALDH⁻ cells to both 5-fluorouracil and doxorubicin treatments but not to cisplatin at the dose studied (Fig. 5E). Verapamil treatment sensitized ALDH⁺ cells to these chemotherapies (Fig. 5E). This effect was confirmed on the formation of tumorsphere by both cell lines, in which verapamil treatment potentiated significantly the reduction of tumorsphere number in response to 5-fluorouracil, doxorubicin, and cisplatin treatments (Fig. 5F). The same result was obtained on tumorspheres of GC10 PDX (Supplementary Fig. S4). Self-renewal assays with residual cells from MKN45- and MKN74-treated tumorspheres confirmed that the combination of verapamil with these conventional chemotherapeutic drugs significantly reduced the number of tumorigenic CSCs (Fig. 5G). Altogether, these results indicate that, within the CD44⁺ subpopulation of tumor cells, the determination of ALDH activity allows the detection and isolation of CSCs with tumorigenic and chemoresistant properties in gastric carcinoma.

In this study, we characterized the expression of cell surface biomarkers and ALDH activity of gastric CSCs in intestinal- and diffuse-type noncardia gastric carcinomas.

Contrary to the situation in colon cancer in which CD133⁺ cells containing colon CSCs represent a small and rare subpopulation within tumors (13, 14), we demonstrated that CD133⁺ cells (detected by similar experimental procedures) were frequent in gastric carcinoma. We also showed that CD133 was a less specific marker for the enrichment of gastric CSCs than CD44 and ALDH, as demonstrated by their lower capacity to form tumorspheres in vitro and a new tumor after xenograft in vivo. CD44 was expressed in all patient-derived primary gastric carcinomas but not in the healthy gastric mucosa or at a very low level in the isthmus of the corpus gastric glands where stem cells reside. We and others previously reported that these CD44⁺ stem/progenitor cells expand from the isthmus toward the base of the unit in metaplastic and dysplastic areas induced in response to chronic H. pylori infection (24, 26, 33). This occurs via an epithelial–mesenchymal-like transition, conferring CSC-like properties to CD44⁺ cells (25, 26). CD44 expression has been reported in gastric carcinoma (34–37). A recent study demonstrated that CD44 inhibition by peptide inhibitors prevented the development of cellular hyperproliferation and chronic atrophic gastritis in animal models of H. pylori–induced gastric carcinogenesis (33, 38).

Interestingly, we showed that CD166 was coexpressed with CD44, and, as a consequence, CD166⁺ cells presented the same tumorigenic properties as CD44⁺ cells in vitro. Similar analyses of in vitro coexpression with CD44 and tumorigenic properties led to the conclusion that the gastric CSC phenotype corresponds to EPCAM⁺, CD24⁺, CD133⁺, CD73⁺, CD90⁻, CD105⁻, CD166⁺, and CD44⁺, associated with ALDH activity. Finally, in all PDXs studied, cells with ALDH activity represented the smallest subpopulation of cells compared with all other markers studied, with high tumorigenic properties both in vitro and in vivo, and with asymmetric division and differentiation properties reproducing the heterogeneity of the initial tumors. As for other cancers, we must consider that the gastric CSC phenotype may be plastic, subjected to regulation by the surrounding tumor microenvironment in vivo. Recent work has demonstrated that breast CSCs coexist between two different phenotypic states: a more quiescent and invasive, mesenchymal-like state characterized by a CD24⁻CD44⁺ phenotype and located mainly at the tumor periphery and invasive front, and a more proliferative epithelial-like state, characterized by ALDH activity and located more centrally (9, 39). However, unlike the situation in breast cancer, we have shown that gastric CSCs express both CD44 and ALDH activity, and that ALDH activity reveals a subpopulation within the CD44⁺ cells (see Fig. 3B) that possess the CSC properties, i.e., to generate a new heterogeneous tumor in vivo and tumorsphere in vitro (Figs. 3C and 4).

The ALDEFLUOR assay used to isolate CSCs in liquid and solid tumors detects the activity of several isoforms of ALDH (40). Among them, the main isoforms expressed in tumors are retinaldehyde dehydrogenases, ALDH1A1 and ALDH1A3, responsible for the oxidation of retinal to retinoic acid and, to a lesser extent, ALDH3A1 (20, 40). They can metabolize and detoxify chemotherapeutic agents such as cyclophosphamide in hematopoietic stem cells, and their level of expression was shown to be predictive of response to treatment in breast cancer (41, 42). In this study, we show that ALDH⁺ cells were more resistant to treatment with conventional chemotherapeutic drugs than ALDH⁻ cells. We adapted the detection of ALDH activity with the ALDEFLUOR reagent to fluorescent microscopy on live cells. Using both methods, we showed that these ALDH⁺ cells did not incorporate the vital DNA dye Hoechst-33342 instead the ALDH⁻ cells incorporated it, confirming that ALDH⁺ cells may correspond to the SP of cells with CSCs properties as previously described by Fukuda and colleagues in gastric carcinoma cell lines (28). The ability of ALDH⁺ cells to efflux Hoechst-33342 and to resist conventional chemotherapy was reversed by verapamil or reserpine treatment, two inhibitors of efflux pumps such as the ATP-binding cassette (ABC) transporters family members, confirming that these cells are associated with chemotherapy resistance as proposed in other cancers (43). In this study, we did not find a noticeable coexpression of BCRP (ABCG2) and MDR-1 (not expressed), the two leaders of the ABC transporters family, in ALDH⁺ gastric CSCs (flow cytometry and qRT-PCR analyses, data not shown), suggesting that the Hoechst-33342 and drug efflux may result from the activity of other members of the ABC transporters family. This family includes at least 49 genes grouped into 7 families, and at least 16 of these proteins have been implicated in cancer drug resistance (44).

A limit to the use of ALDH as a biomarker of chemoresistant gastric CSC is that ALDH activity can be detected only by the ALDEFLUOR assay on live cells by flow cytometry or fluorescent microscopy analyses. So, its detection as a biomarker of CSCs in current practice on patients' specimens may be possible for circulating cancer cells and liquid cancers such as leukemia but remain elusive for the analysis of solid tumors such as gastric carcinoma. These findings also imply that ALDH isozymes can be considered not only as biomarkers of CSCs but also as putative targets to inhibit tumor growth and to overcome resistance to cancer therapy.

It is of importance to note that gastric carcinoma PDXs always remained heterogeneous and composed of tumor cell subpopulations expressing EPCAM, CD24, CD133, and CD44, similar to the patients' situation, whereas gastric carcinoma cell lines were found to be negative for CD133 and either positive or negative for CD44 and others markers including ALDH. These results strengthen the importance and the necessity to study CSC on models as close as possible to the patients' situation, as is the case in this study, and not only on cancer cell lines.

Currently, PDXs represent the most pertinent preclinical model to study the capacity of CSCs to give rise to tumor growth, heterogeneity, and sensitivity/resistance to new treatment strategies. However, PDXs models present some limitations, particularly because the contribution of the patient's tumor microenvironment—including inflammation, CSCs niche within the organ of origin, and cross-talk with immune and stromal cells—cannot be taken into account on CSCs plasticity, tumor progression, and metastasis. These limits are partly illustrated here by the low tumor engraftment success of patient's gastric carcinoma samples, being only approximately 20% of engraftment success, as described by others for other type of cancers, unveiling the contribution of uncontrolled microenvironment parameters for cancer propagation in the patient. Nevertheless, there is an urgent unmet need of new, more efficient and better tolerated therapeutic strategies for gastric carcinomas, which could focus on gastric CSCs.

In this study, the development of original PDXs models allowed us to demonstrate that tumorigenic and chemoresistant gastric CSCs coexpress EPCAM, CD133, CD166, CD44, and ALDH, ALDH activity being the most specific biomarker of CSC enrichment before CD44 in both diffuse- and intestinal-type noncardia gastric carcinomas. This finding led to the hypothesis that treatment strategy for noncardia gastric carcinomas can focus on CD44⁺ALDH⁺ CSCs, independently of the histologic classification of the tumor.

L. Wittkop is a consultant/advisory board member for Bristol-Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Dubus, F. Mazurier, C. Varon

Development of methodology: P.H. Nguyen, P. Dubus, C. Varon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Giraud, P. Dubus, G. Belleannée, D. Collet, I. Soubeyran, S. Evrard, B. Rousseau, F. Mazurier, C. Varon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.H. Nguyen, J. Giraud, P. Dubus, L. Wittkop, C. Varon

Writing, review, and/or revision of the manuscript: J. Giraud, P. Dubus, L. Wittkop, F. Mégraud, C. Varon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Chambonnier, N. Senant-Dugot

Study supervision: S. Evrard, F. Mégraud, C. Varon

We thank Marie-Edith Lafon (CNRS UMR 5234, University of Bordeaux) and technicians from the Department of Tumor Pathology (Haut-Leveque Hospital, University Hospital Center of Bordeaux) for molecular analyses on tumor tissues, Pierre Costet (animal facilities, University of Bordeaux), Vincent Pitard and Santiago Gonzalez (Flow Cytometry and FACS Platform, University of Bordeaux), Philippe Brunet de la Grange (CNRS UMR5164 CIRID, University of Bordeaux) for assistance on SP cells analyses, and Alban Giese (Experimental Pathology Platform of the Canceropole GSO and SIRIC BRIO, University of Bordeaux), Elodie Siffre, and Lucie Benejat (INSERM U853) for technical assistance.

This study was financially supported by the French “Association pour la Recherche contre le Cancer” (grant number 8412), the “Institut National du Cancer” (grant 07/3D1616/IABC-23-12/NC-NG and grant 2014-152), the “Conseil Regional d'Aquitaine” (grant numbers 20071301017 and 20081302203), the French National Society for Gastroenterology, and the Canceropole Grand Sud-Ouest (grant 2010-08-canceropole GSO-Université Bordeaux 2). This project was also supported by SIRIC BRIO (Site de Recherche Intégrée sur le Cancer – Bordeaux Recherche Intégrée Oncologie; grant INCa-DGOS-Inserm 6046).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.