MiR-19b/20a/92a regulates the self-renewal and proliferation of gastric cancer stem cells

In the past several years, a growing body of evidence has shown that cancers are organized in a population of heterogeneous cells with different biological properties that help sustain tumor formation and development. A small proportion of cells that are self-renewing and have the ability to differentiate into multiple lineages are termed cancer stem cells (CSCs) or tumor-initiating cells (TICs) (Dalerba et al., 2007). CSCs have been identified in many cancers (Al-Hajj et al., 2003; Singh et al., 2004). They are not only the source of tumors, but have also been associated with tumor aggressiveness and metastasis (Wicha, 2006). Furthermore, CSCs mediate chemoresistance and subsequent tumor recurrence (Al-Hajj, 2007). CSCs can be enriched by sorting for stemness markers (Han et al., 2011) or by selecting for side-population (SP) cells exhibiting Hoechst dyes efflux (Fukuda et al., 2009). A third method is the isolation of populations of spherical tumorspheres from suspension cultures. For example, sphere-forming cells can be cultured from human glioblastomas using conditions that enrich for neural stem cells and display increasing tumorigenicity and resistance to radiation treatment (Bao et al., 2006). Breast CSCs can also be isolated from spherical cultures on the basis of their stem cell properties (Ponti et al., 2005).

MicroRNAs (miRNAs) are small, non-coding RNAs that can regulate target genes post-transcriptionally through complementary binding to their target mRNAs (Bartel, 2009). In cancers, miRNAs can function either as oncogenes or tumor suppressors to regulate carcinogenesis and cancer development (Croce, 2009), while in stem cells, miRNAs can either promote self-renewal or promote differentiation to determine stem cell fates (Shimono et al., 2009). miRNAs can exert these functions in both cancers and stem cells. One of the best-studied miRNAs, let-7, can act as a tumor suppressor in breast cancer and can impede the self-renewal of breast tumor-initiating cells (Yu et al., 2007). miR-200c not only inhibits the clonal expansion of breast cancer cells, but also strongly suppresses tumor formation driven by human breast cancer stem cells (Shimono et al., 2009).

The existence of CSCs in most organ systems has been demonstrated, and some molecular links between CSCs and cancers have been putatively established. However, it is still not clearly known how CSC activities such as self-renewal are controlled by different genetic molecular factors. In the present study, we used gastric cancer as a research model to explore the roles that miRNAs play in the self-renewal of CSCs and the potential associated mechanisms.

Gastric cancer stem cells (GCSCs) were identified in primary tissues using the cell surface markers EpCAM+ and CD44+ (Han et al., 2011). To examine whether EpCAM+/CD44+ cells from gastric cancer cell lines might also enrich for GCSCs, we compared the tumorspheric ability of two gastric cancer cell lines that exhibit great differences in the levels of EpCAM and CD44 expression. In the gastric cancer cell line, SGC7901, 80% of the cells expressed EpCAM and CD44. Additionally, the SGC7901 cells formed tumorspheres more often than MKN28 cells, which contained only 0.2% EpCAM+/CD44+ cells (Fig. 1A,B). Because tumorspheric generation can be used as an in vitro assay for determining self-renewal potential and a method for culturing CSCs, we isolated EpCAM+/CD44+ cells and non-EpCAM+/CD44+ cells using FACS and cultured them in suspension to generate tumorspheres. After 15 days in culture, 85–95% of the EpCAM+/CD44+ cells formed tumorspheres, compared with 15–20% tumorsphere formation by non-EpCAM+/CD44+ cells (data not shown). Furthermore, the spheres formed by the EpCAM+/CD44+ cells contained more cells than the spheres formed by the non-EpCAM+/CD44+ cells (supplementary material Fig. S1A,B). In BALB/C nude mice, 10⁴ EpCAM+/CD44+ cells generated tumors that showed a more than twofold stronger luciferase signal compared with tumors generated by 10⁵ non-EpCAM+/CD44+ cells, as detected on days 15, 21 and 28 using the IVIS 100 Imaging System (supplementary material Fig. S1C,D). These data suggested that EpCAM and CD44 are also GCSC markers in cultured gastric cancer cell lines and can be used as an in vitro model for the exploration of GCSCs.

An important property of CSCs is multipotency. We took advantage of the above findings to determine whether GCSCs would differentiate when cultured in serum-containing medium. As expected, the cells proliferated more rapidly when attached to flasks than tumorspheric cells, as determined by carboxyfluorescein succinimidyl ester (CFSE) labeling of cells cultured for 8 days (Fig. 1C). Furthermore, the tumorspheric cells expressed lower levels of the myoepithelial marker, CK14, and the luminal epithelial marker, CK18. However, after further attachment to the flasks and differentiation in serum-containing medium, these cells developed into elongated cells with a higher percentage expressing either CK14 or CK18, as determined by real-time PCR analysis (Fig. 1D,E) and immunofluorescence (supplementary material Fig. S1E). The potent differentiation capability of tumorspheric EpCAM+/CD44+ cells in serum-containing medium indicated the multipotency of the GCSCs and suggested that there must be specific mechanisms that regulate this process.

Because miRNAs can regulate the self-renewal and differentiation of stem cells, we exploited our ability to obtain a large number of self-renewing tumorspheric cells to compare miRNA expression in these cells with cells that were further differentiated. Among the cell lines showing increased EpCAM+/CD44+ expression, tumorspheric cells that were freshly dissociated (Fig. 2A, lane 1) or had only briefly attached (Fig. 2A, 8 hours, lane 2) expressed higher levels of some miRNAs compared with cells that were differentiated under serum-containing conditions (Fig. 2A, 12 hours, lane 3; 1 day, lane 4; 8 days, lane 5). We considered the miRNAs that gradually decreased during the attachment of the tumorspheric cells to be miRNAs that might control the self-renewal of GCSCs. Based on ANOVA analysis on normalized chip data, we identified a number of miRNAs whose expression was significantly different in tumorspheric cells compared with differentiated cells. Among these miRNAs, miR-19b, miR-20a and miR-92a emerged as the most significant miRNAs that were gradually reduced during the attachment of the tumorspheric GCSCs (Fig. 2B). A number of other miRNAs, such as miR-106a and miR-30d, showed similar expression patterns as miR-19b, miR-20a and miR-92a (Fig. 2A).

To confirm the decreased expression of miR-17-92 miRNAs during the attachment of tumorspheric cells, we performed real-time PCR using miRNA-specific primers. miR-19b, miR-20a and miR-92a were detected at higher levels in tumorspheric cells; their expression levels were briefly reduced after 8 hours of attachment and began to decrease within 1 day of attachment, then decreased further over 8 days of differentiation (Fig. 2C). The overexpression of two other miRNAs related to tumors, miR-106a and miR-30d, was confirmed using miRNA-specific primers (supplementary material Fig. S2).

To test whether GCSCs from human gastric cancer tissues also have a higher expression of miR-19b/20a/92a, cells were isolated from human gastric cancer tissues and cultured in non-serum low attachment conditions in 96-well plates. After 3 weeks in culture, GCSCs formed tumorspheres, as shown in supplementary material Fig. S6A. The total RNA of GCSCs and non-GCSCs was harvested and miR-19b/20a/92a expression was detected. The expression of miR-19b/20a/92a was also higher in tissue GCSCs than non-GCSCs as indicated in supplementary material Fig. S6B.

To investigate whether miR-19b/20a/92a could sustain the self-renewal of GCSCs, we stably infected EpCAM−/CD44− (separated from SGC7901 cells) and MKN28 cells with lentiviruses containing precursor microRNAs: pre-miR-19b, pre-miR-20a and pre-miR-92a. The successful overexpression of mature miRNAs was confirmed by real-time PCR (supplementary material Fig. S3A,B). The overexpression of the miRNAs lenti-miR-19b, lenti-miR-20a and lenti-miR-92a resulted in markedly greater self-renewal ability in gastric cancer cells, as indicated by the high intensity of tumorspheric cells compared with lenti-NC-infected cells (Fig. 3A; supplementary material Fig. S3C). Furthermore, the number of cells within each sphere of lenti-miR-19b-, lenti-miR-20a- and lenti-miR-92a-infected cells were much higher than observed for the NC-infected cells (Fig. 3B). To avoid experimental artifacts related to cellular changes, miRNA precursors were transiently transfected into EpCAM−/CD44− and MKN28 cells; transcription was confirmed using specific miR-17-92 primers by real-time PCR (supplementary material Fig. S3D–G). Consistent with the findings in stable cell lines, the transient overexpression of the miR-19b, miR-20a and miR-92a precursors significantly increased the number of tumorspheres of EpCAM−/CD44− and MKN28 gastric cancer cells (supplementary material Fig. S3H,I). To detect the effects from knocking down miR-19b/20a/92a, we transient transfected miRNA inhibitors into EpCAM+/CD44+ cells or SGC7901 cells. After testing the tumorspheric abilities of these cells, we found miR-19b/20a/92a inhibitors markedly impeded and delayed the formation of tumorspheres of EpCAM+/CD44+ and SGC7901 cells (Fig. 3C,D). Our results implied that the miRNAs of the miR-17-92 cluster regulate the self-renewal ability of GCSCs under non-adherent conditions by promoting the number of tumorspheres among gastric cancer cells and the number of cells in each sphere.

To test whether these miRNAs can work synergistically or can counteract each other, we co-transfected pre-miR-92a with pre-miR-20a or pre-miR-19b separately in MKN28 cells. Compared with pre-miR-92a transfection alone, the co-transfection of miR-92a with miR-19b resulted in a higher tumorsphere percentage (supplementary material Fig. S7) which indicated that miR-92a and miR-19b might work synergistically in this cluster.

It has been reported that CSCs are resistant to chemotherapy, which leads to multidrug resistance and the subsequent recurrence of cancer. In the present study, using an MTT assay, we found that miR-17-92-infected tumorspheric cells were more resistant to the chemotherapeutic drug 5-FU than negative-control-infected cells (Fig. 3E). Our results indicated that miR-19b/20a/92a not only sustains the self-renewal of GCSCs by increasing the number of tumorspheric cells under non-adherent conditions but also maintains the stemness of GCSCs by inducing resistance to chemotherapeutic drugs.

Because GCSCs present the cell surface markers EpCAM and CD44, we performed FACS to determine whether pre-miR-19b/20a/92a-transfected GCSCs cells exhibit more EpCAM/CD44 than the negative control (NC)-transfected cells. We found that the transient transfection of pre-miR-19b, pre-miR-20a and pre-miR-92a into SGC7901 and MKN28 gastric cancer cells could markedly increase the EpCAM+/CD44+ cell percentage (supplementary material Fig. S4A), whereas the inhibitors of miR-19b, miR-20a and miR-92a decreased the percentage of EpCAM+/CD44+ SGC7901 cells (supplementary material Fig. S4B).

We next assessed the effects of induced miR-17-92 expression on self-renewal in vivo. To facilitate detection, the luciferase-labeled SGC7901 cell line established in our laboratory (SGC7901-luc) was used to create stable transfectants expressing miR-17-92 miRNAs and negative controls. The positive expression of mature miR-17-92 was confirmed using real-time PCR (supplementary material Fig. S4C). Each severe combined immunodeficient (NOD-SCID) mouse was inoculated with 2×10³ tumorspheric miR-19b/20a/92a-infected cells or NC-infected cells. All of the mice injected with lenti-miR-19b/20a/92a-infected cells formed tumors, whereas only one mouse (P<0.001) formed tumors from lenti-NC-infected cells (Table 1). The miR-19b/20a/92a-infected cells formed tumors that exhibited stronger luciferase activity under the IVIS system on the 28th day after injection compared with the NC group cells (Fig. 4A,B). Moreover, the tumors that expressed lenti-miR-19b/20a/92a grew to at least 3 cm in diameter by 28 days after becoming palpable, whereas the lenti-NC-expressing cells did not generate tumors more than 2 cm in diameter by 49 days, which was the end of the experimental period (supplementary material Fig. S4D). Furthermore, mice injected subcutaneously with miR-19b/20a/92a-infected cells presented liver and lung metastasis (Table 1), as detected by Hematoxylin and Eosin (H&E) staining (Fig. 4C), which is consistent with the hypothesis that CSCs are prone to metastasis. The histological analyses of subcutaneous tumors were confirmed using H&E (Fig. 4D) to verify that the miR-17-92 miRNAs sustained the self-renewal ability of GCSCs in vivo. Because the lentivirus contains an EGFP expressing vector, the lenti-miR stable cell lines should express EGFP. We thus performed EGFP staining using immunohistochemistry on livers, lungs and tumors of NOD-SCID mice as indicated in Fig. 4C,D.

SC-regulatory genes generally also have effects on the proliferation of cells. Therefore, we next defined the roles of miR-19b/20a/92a miRNAs in the regulation of gastric cancer cell growth and progression. Using the MTT and colony formation assays, we found stable lenti-miR-19b/20a/92a-expressing gastric cancer cell lines proliferated faster than lenti-NC-expressing cells (Fig. 5A, left, B,C). To test the knockdown function of these miRNAs, we transiently transfected miR-19b/20a/92a inhibitors into gastric cancer cell lines SGC7901 and MKN28 and tested them using proliferation assays. Conversely, we found miR-19b/20a/92a inhibitors made cancer cells proliferate slower than negative control transfection, using the MTT assay (Fig. 5A, right; supplementary material Fig. S5B) and colony formation assay (supplementary material Fig. S5C, lower; Fig. S5D, right). To avoid any cellular changes resulting from the experimental procedures, we also performed sense transient transfection using miRNA precursors and tested the cell proliferative functions. Consistent with stable expressing cells, the transient transfection of miR-19b/20a/92a promoted proliferation as determined by both the MTT and colony formation assays as shown in supplementary material Fig. S5A; Fig. S5C, top and S5D, left.

To validate the results of the cell proliferation assays in vitro, we performed in vivo assays to evaluate the tumorigenic effects of the miR-17-92 miRNAs in BALB/C nude mice using a luciferase-labeled lenti-miR-17-92-infected SGC7901-Luc cell line. The lenti-miR-19b/20a/92a-infected cells showed a proliferative tendency in nude mice, and all of them significantly promoted tumor growth in nude mice compared with the lenti-NC group (Fig. 5D,E). The histological analyses of the tumors were confirmed using H&E and EGFP staining to verify that the miR-17-92 miRNAs promoted tumor growth in vivo (Fig. 5F).

In silico analysis using MiRanda software (http://www.microrna.org/microrna/home.do) showed that the 3′-UTR of human E2F1 contains two conserved putative target sites for miR-20a, and the 3′UTR of human HIPK1 contains one conserved site for miR-19b and one for miR-92a (Fig. 6A). E2F1 was previously reported to be a target of miR-20a and can inversely induce the expression of the miR-17-92 cluster. To further validate these target sites, the 3′UTRs of human E2F1 and HIPK1 were inserted in both orientations downstream of the luciferase gene in the pGL3-Control vector, providing sense (Luc-1S, 2S and S) and antisense (Luc-1AnS, 2AnS and AS) constructs, collectively referred to as Luc-1, Luc-2, Luc-3 and Luc-4 in Fig. 6. Transfecting HEK-293 cells with miR-20a significantly decreased the expression of Luc-1S, while having no effect on Luc-1AnS expression (Fig. 6B). In contrast, miR-150 did not affect the expression of Luc-1S (Fig. 6B), in accordance with the fact that the E2F1 3′UTR contains no miR-150 target sites. The expression of Luc-2S was reduced by either miR-19b or miR-92a, but not by miR-150, as a result of the lack of common target sites (Fig. 6C). Moreover, miR-19b could target Luc-3S, and miR-92a could target Luc-4S (Fig. 6D).

To evaluate the downregulating effect of the miR-17-92 cluster on E2F1 and HIPK1, we performed western blot analysis using anti-E2F1 and anti-HIPK1 antibodies. As would be expected, transfecting SGC7901 cells with pre-miR-20a decreased E2F1 levels (Fig. 6E), whereas transfecting cells with miR-20a inhibitory RNAs had the opposite effect. The cells infected with pre-miR-19b and pre-miR-92a showed reduced HIPK1 levels compared with NC-infected cells, which suggests that the miR-17-92 miRNAs regulate E2F1 and HIPK1 expression in vivo at the post-transcriptional level.

It was previously reported that E2F1 can transactivate CTNNBIP1 (β-catenin-interacting protein 1), also known as ICAT (inhibitor of β-catenin and TCF4), which is required to inhibit β-catenin activity. E2F1 can also induce axin2, which could cause β-catenin degradation. Furthermore, HIPK1, whose expression is frequently altered in cancers, represses Wnt/β-catenin target gene activation. To determine whether the β-catenin pathway was activated by enforced miR-17-92 expression, we performed a western blotting assay using anti-β-catenin antibodies on miR-19b/20a/92a-transfected cells and GCSCs from human GC tissues. The nucleoprotein of miR-19b/20a/92a-transfected cells and GCSCs was used for the detection. We observed an upregulation of β-catenin proteins in miR-19b/20a/92a-transfected cells (Fig. 6F) and GCSCs from human GC tissues (supplementary material Fig. S6C)

To explore the value of the miR-17-92 cluster in a clinical setting, we performed real-time PCR to detect each member of the miR-17-92 cluster in RNA samples from the stomach tissues of 97 gastric cancer patients who underwent gastrectomy and were followed up at Xijing Hospital during the years 2006 to 2008. Furthermore, we ranked the patients on the basis of the relative expression levels, obtained by real-time PCR, and divided them into a high-expression group and a low-expression group using the 50th percentile (median) as the cut-off point. HR from the univariate Cox regression analysis showed that the tumor stage and the expression of miR-20a and miR-92a were correlated with death from any cause (Fig. 7; supplementary material Table S1). A multivariate Cox regression analysis identified the level of miR-92a as an independent factor that was associated with overall survival (supplementary material Table S1).

GCSCs were recently identified by Han et al. and Takaishi et al. using the cell surface markers CD44 and EpCAM (Takaishi et al., 2009; Han et al., 2011). However, the small number of CSCs increases the difficulty of studying how they sustain self-renewal and initiate cancers without differentiating. Through the application of miRNA arrays, we compared tumorspherical GCSCs with more differentiated gastric cancer cells based on CSC properties. We found that the expression of the miR-17-92 cluster, especially of miR-19b/20a/92a, was reduced gradually during GCSC differentiation. These results indicated that miR-19b/20a/92a might have the ability to sustain the self-renewal of GCSCs.

The miRNA miR-17-92 cluster was first identified in the year 2005 as a potential human oncogene by He et al., and it was demonstrated that tumors derived from hematopoietic stem cells expressed a subset of the mir-17-92 cluster (He et al., 2005). miR-17-92 members have been identified as being overexpressed in numerous types of stem cells, such as hematopoietic stem cells (Jin et al., 2008), embryonal carcinoma SCs (Gallagher et al., 2009), mouse embryonic SCs (Gunaratne, 2009) and neuronal SCs (Chen et al., 2010), and are downregulated during neuronal lineage differentiation in unrestricted somatic stem cells (Iwaniuk et al., 2011) and in unrestricted somatic stem cells differentiated into neuronal lineages (Trompeter et al., 2011). Additionally, miR-17-92 was shown to be highly induced during early reprogramming stages in induced pluripotent stem cells (iPSCs) (Li et al., 2011). However, the exact roles of the miR-17-92 cluster in stem cells and CSCs have not been fully elucidated. In the present study, we demonstrated the ability of the miR-17-92 cluster to regulate self-renewal of GCSCs using in vitro and in vivo assays.

The earliest studies (prior to the year 2009) addressing miR-17-92 and stemness maintenance in stem cells were associated with the whole gene function of the miR-17-92 cluster. There were two studies that reported the function of single miRNAs in this cluster. Mu et al. used the whole cluster as a positive control and deleted every member of this cluster and found that miR-19 acted as the most important miRNA in the cluster in B cell lymphoma (Mu et al., 2009). At the same time, through the overexpression of single miRNAs of the miR-17-92 cluster, Olive et al. also demonstrated that miR-19 is the most functional member of the whole cluster (Olive et al., 2009). In the present study, we used gastric cancer as a research model and discovered that in addition to miR-19b, miR-20a and miR-92a also performed similar significant functions in regulating the self-renewal of GCSCs. The difference between previous studies and this study is that that members of the miR-17-92 cluster might perform different roles in different cancers.

Recently, E2F1 was found to suppress Wnt–β-catenin signaling by trans-activating the β-catenin-interacting protein ICAT (Wu et al., 2011) or by upregulating axin2 (Hughes and Brady, 2006), which promotes β-catenin degradation (Fancy et al., 2011). In the present study, exogenous miR-17-92 significantly knocked down E2F1 expression. The Wnt–β-catenin signaling pathway has been reported to regulate the self-renewal and multipotency of stem cells. Mice that lack β-catenin in their hematopoietic cells form HSCs, but are deficient in long-term growth and maintenance (Zhao et al., 2007), indicating an essential role of Wnt/β-catenin in the self-renewal of normal and neoplastic stem cells in the hematopoietic system. Activation of Wnt signaling using a GSK-3-specific inhibitor can maintain the pluripotency of human and mouse embryonic stem cells (Sato et al., 2004). More importantly, activation of Wnt–β-catenin signaling enriches the EpCAM(+) cell population, whereas RNA interference-based blockage of EpCAM, a Wnt–β-catenin signaling target, attenuates the activities of these cells (Yamashita et al., 2009; Terris et al., 2010).

Through a bioinformatics analysis, we first identified another suppressor of Wnt–β-catenin signaling, HIPK1, that can be targeted by the miR-17-92 cluster. A previous report indicated that HIPK1 can repress Wnt/β-catenin target gene activation (Louie et al., 2009), as demonstrated through β-catenin reporter assays in human embryonic kidney cells and by indicators of dorsal specification in Xenopus laevis embryos at the late blastula stage. In addition, a subset of Wnt-responsive genes subsequently requires HIPK1 for activation in the involuting mesoderm during gastrulation. The present study not only demonstrated that miR-19b/20a/92a can target HIPK1 but also showed that transfection of miR-19b/20a/92a decreased the expression of β-catenin protein. Based on our findings, it can be speculated that the miR-17-92 cluster might activate the Wnt–β-catenin signaling pathway through downregulating its direct suppressors E2F1 and HIPK1, which subsequently activates Wnt–β-catenin signaling and increases the abundance of EpCAM+ GCSCs (Yamashita et al., 2007).

Wnt–β-catenin signaling (wnt5a) has previously been reported to be associated with a poor prognosis in gastric cancer (Kurayoshi et al., 2006). miR-17-92 has been reported to contribute to tumorigenesis and a poor prognosis in multiple myeloma (Chen et al., 2011). We performed real-time PCR in tissue samples from patients to examine the prognostic capabilities of the cluster and each individual miRNA. Our results showed that miR-92a could act as an independent factor associated with overall survival.

In conclusion, the results of the present study suggest that members of the miR-17-92 cluster, miR-19b, miR-20a and miR-92a, act as important miRNAs that regulated the self-renewal of GCSCs by targeting E2F1 and HIPK1 and subsequently activating Wnt–β-catenin signaling, increasing the expression of the GCSC marker EpCAM. This study also showed that miR-19b, miR-20a and miR-92a are overexpressed in human GCSCs and that the expression of miR-20a and miR-92a is negatively associated with overall survival in gastric cancer patients. Furthermore, miR-92a acts as an independent factor that might be used as a prognostic marker in gastric cancer. Thus, this study shed light on the potential clinical application of miR-19b/20a/92a. Targeting miR-19b/20a/92a might effectively restrict the growth of gastric cancer.

For the analyzed tissue specimens, all patients gave informed consent to use excess pathological specimens for research purposes. The protocols employed in this study were approved by the hospital's Protection of Human Subjects Committee. The use of human tissues was approved by the institutional review board of the Fourth Military Medical University (Xi'an, Shaanxi Province, P. R. China) and conformed to the Helsinki Declaration, and to local legislation. Patients offering samples for the study signed informed consent forms. In the animal experiments, all procedures for animal experimentation were performed in accordance with the Institutional Animal Care and Use Committee guidelines of the Experimental Animal Center of the Fourth Military Medical University (approval ID no.12039).

All of the cells were grown in RPMI1640 (Invitrogen, Carlsbad, CA) supplemented with 10% heat-inactivated fetal calf serum (FCS) at 37°C under 5% CO₂ in a humidified incubator (Forma Scientific, Marietta, OH). The miRNA precursors, inhibitors and negative controls were obtained from Ambion (Applied Biosystems, USA).

The immunodeficiency lentiviral system with EGFP-expressing miR-19b/20a/92a or lenti-NC (lentivirus expressing the negative control miRNAs) was purchased from Genechem (Shanghai Genechem Co. Ltd). CD44+/EpCAM+ and CD44−/EpCAM− cells were separated from SGC7901 and SGC7901-Luc cells, and used as hosts for lentivirus infection. Gastric cancer cells were infected with a lentiviral system expressing miR-19b/20a/92a or lenti-NC according to the manufacturer's instructions, and stable cells were obtained by sorting GFP-positive cells.

Microarray analysis was performed using a service provider (LC Sciences), as previously reported (Kort et al., 2008), in 2–5 µg of total RNA from tumorspheric non-adherent GCSCs that were attached to the bottoms of flasks for 8 hours, 12 hours, 1 day or 8 days. Total RNA was isolated using TRIzol (Invitrogen). Hybridization was performed overnight on a micro-fluidic chip. The data were analyzed by first subtracting the background and then normalizing the signals using a LOWESS filter (locally weighted regression). ANOVA was conducted on the five samples by LC Science. miRNAs showing P-values <0.01 were considered to be differentially expressed miRNAs.

Tumorsphere culturing of GCSCs was performed as described previously (Takaishi et al., 2009). Cells were suspended and inoculated in serum-free culture medium, RPMI-1640 medium (Hyclone), containing human recombinant epidermal growth factor (EGF; Invitrogen) at a concentration of 20 ng/ml and human recombinant basic fibroblast growth factor (bFGF; Invitrogen) at a concentration of 10 ng/ml. After 3 weeks, each well was examined under a light microscope, and the total number of cells in spheroid colonies and cells within one tumorsphere colony were counted.

BALB/C nude mice and severe combined immunodeficient (NOD-SCID) mice at 4–6 weeks of age were handled using best humane practices and cared for in accordance with NIH Animal Care Institutional Guidelines in the Experimental Animal Center of the Fourth Military Medical University. Nude mice were used to test the proliferation ability using 5×10⁶ cells whereas NOD-SCID mice were used to test the self-renewal ability of gastric cancer stem cells using 5×10³ cells. CD44+/EpCAM+ and CD44+/EpCAM+ cells separated from SGC7901-Luc cells were used to inoculate mice. For the primary experiments, which aimed to test the self-renewal in nude mice, 10⁴ CD44+/EpCAM+ cells and 10⁵ non-CD44+/EpCAM+ cells were injected subcutaneously into the right or left upper back at a single site. Five mice were injected for each group. To investigate self-renewal in NOD-SCID mice, 5×10³ cells infected with either lenti-NC (negative control) or lenti-miR-19b/20a/92a were injected subcutaneously into the back at a single site. Each group contained eight mice. To detect tumorigenecity in nude mice, 5×10⁶ cells containing lenti-miRs were injected subcutaneously into the back of each mouse; each group contained six mice. The mice were injected with 100 mg/kg D-luciferin intraperitoneally 5 minutes before imaging. Bioluminescent signals were detected twice a week using the IVIS 100 Imaging System (Xenogen, Hopkinton, MA). The nude mice were killed after 28 days and NOD-SCID mice were killed after 28 days (miR groups) or 49 days (NC group). Their subcutaneous tumors were harvested and fixed in 10% formalin before embedding in paraffin, sectioning and staining with H&E.

A reporter gene assay was performed as described previously (Tie et al., 2010). Cells were co-transfected with pre-miR-19b/20a/92a (150 nM, Ambion) and 20 ng of the Renilla luciferase control vector (pRL-TK, Promega). Firefly luciferase activities were normalized to Renilla luciferase activities. Pre-NC (Pre-miR™ miRNA precursor negative controls) from Ambion was used as one of the negative control miRNAs.

To conduct drug sensitivity assays, the water-soluble tetrazolium salt WST-8 was employed. At 24 hours after transfection with pre-miR-19b/20a/92a or pre-NC, cells were treated with the serially diluted chemotherapeutic drug 5-fluorouracil (5-FU) at a concentration of 0.25 µg/ml. Optical density was measured at 450 nm and 690 nm using a micro-plate reader.

Cells were adjusted to a density of 10⁶ cells/ml and treated with CFSE at a final concentration of 10 µM. After incubation at 37°C for 10 minutes, labeling was blocked by the addition of RPMI medium with 10% FCS. The fluorescence intensity was determined by flow cytometry analysis.

The in vitro cell proliferation assay was performed as described previously (Wu et al., 2010). A total of 10³ miR-17-92-infected cells or cells transiently transfected with a 150 nM concentration of miR-17-92 precursors or inhibitors were used for the assays in 200 µl of complete medium. The cultures were assayed each day by reading the absorbance at 490 nm with a micro-plate reader (168–1000 Model 680, Bio-Rad, Hercules, USA). Each experiment was performed in triplicate and repeated three times.

The plate colony formation assay was performed as described previously (Wu et al., 2010), and transfection was performed as described above. Cells were plated at a density of 300 cells/well in twelve-well plates and cultured in RPMI 1640 medium with 10% fetal calf serum for 3 weeks. The colonies were stained with Crystal Violet solution, and the number of colonies was counted. Each assay was performed three times.

The attached cells were washed using PBS and then fixed with 4% paraformaldehyde for 20 minutes at room temperature. Cells were then rinsed three times using PBS for 5 minutes and permeabilized with PBS-T solution for 10 minutes. After washing for 5 minutes with PBS, cells were blocked using 1% BSA for 30 minutes. Primary antibodies anti-CK14 and anti-CK18 were added to the cells and they were then incubated at 4°C overnight. After three washes, cells were incubated with secondary antibodies.

Western blotting was performed as described previously (Wu et al., 2010). Log-phase cells were harvested and resolved on SDS-PAGE gels, followed by blotting on nitrocellulose membranes (Amersham, Pittsburgh, PA). The membranes were blocked with 10% non-fat milk and incubated overnight with primary antibodies: anti-E2F1, anti-β-catenin and anti-HIPK1 (1∶1,000; ABcam, USA) or an anti-β-actin antibody (1∶2,000; Sigma-Aldrich, USA).

Total RNA was extracted from FFPE-impeded tissue sections according to the manufacturer's instructions (AM1975, Ambion). Total RNA was extracted from cells as described previously (Wu et al., 2010). Reverse transcription was performed according to the manufacturer's instructions (D350A, TaKaRa Biotechnology Co., Ltd). Quantitative real-time PCR was performed to determine the expression levels of each miRNA using the exact sequences (U to T) of these miRNAs as the forward primer and the unique qPCR primer from the cDNA Synthesis Kit. U6 was used as an internal control, and each plate contained one cDNA sample for each primer as a calibration sample.

Continuous variables were compared with ANOVA tests. If the test of homogeneity of variances between the groups was significant, the Mann–Whitney U-test and Kruskal–Wallis H-test were adopted as appropriate. A Chi-squared or Fisher's test was used for categorical variables. The independent predictors of survival were calculated using the Cox regression model. The covariates incorporated into the multivariate analysis were the variables with P<0.05 in a univariate analysis. Cumulative survival was assessed with Kaplan–Meier curves and compared using the log-rank test. Two-tailed P-values <0.05 were considered statistically significant. All statistical analyses were conducted using SPSS software, 14.0 (Chicago, Illinois, USA).

We thank Zhen Chen and Taidong Qiao for the supply of common lab materials.