Clinical Utility of a STAT3-Regulated miRNA-200 Family Signature with Prognostic Potential in Early Gastric Cancer

The prognosis for patients with gastric cancer the third most lethal cancer worldwide, largely depends on the stage of disease at detection. With advances in diagnostic techniques, the rates of detecting early-stage gastric cancer (EGC), defined by the depth of the neoplastic invasion being restricted to the submucosal layer of the stomach, irrespective of lymph node (LN) metastasis, has increased to almost 20% in some countries (1, 2). Although early diagnosis affords the opportunity for minimally invasive treatment such as endoscopic resection, this procedure remains only potentially curative as a frontline and unimodal therapy in a subset of EGC patients (1–3). Furthermore, the current rudimentary classification system of EGC does not quantitatively phenotype patient tumors at the molecular level, and overlooks any degree of metastatic potential, thus raising concerns that therapeutic endoscopy may still yield incorrect prognoses in some patients. Indeed, such a scenario has been reported in a subset of gastric cancer patients suffering relapse and metastasis despite their tumors being clinically graded as “early” (4, 5). In contrast to the comprehensive molecular characterization of advanced gastric cancer, molecular profiling of EGC to identify tumor-specific gene signatures with prognostic value is limited (6, 7). As such, there is an unmet clinical need to determine a more accurate molecular classification system with prognostic prediction in EGC.

miRNAs are small noncoding RNAs that regulate gene expression through multiple mechanisms including transcript degradation and translational repression (8). Although miRNAs play critical roles in innate immunity and cancer development, there is limited understanding of their functions in the molecular pathways that regulate the initiation of inflammation-associated tumorigenesis (8). In gastric cancer, clinical data suggest that numerous miRNAs involved in cell-cycle progression, apoptosis, and metastasis are differentially regulated, such as miR-21, miR-24, and miR-130b “oncomirs” (9, 10), along with miR-148a and miR-7 “tumor suppressors” (11, 12). In addition, the miR-200 family has been identified as a key regulatory network that defines a mesenchymal gastric cancer subtype associated with poor survival (13, 14). However, such studies are biased towards identifying miRNAs predominately in advanced-stage gastric cancer, and therefore overlook potential key miRNAs which trigger the initiating stages of gastric cancer.

We have previously established the gp130^F/F mouse model of gastric cancer that develops a gastric phenotype with marked similarity to human intestinal-type hyperplasia, a precursor of human intestinal-type gastric cancer (15–17). Gastric tumorigenesis in gp130^F/F mice is driven by hyperactivation of the oncogenic latent transcription factor STAT3 via the IL6 cytokine family member, IL11 (15). The clinical relevance of this model is evidenced by approximately 50% of gastric cancer patients displaying augmented STAT3 activation and upregulated IL11 expression (17, 18). However, the molecular basis governing the stage-specific (i.e., early versus late) role of STAT3 in gastric cancer, and for that matter other cancers, is unclear. Here, by coupling gp130^F/F mice with interrogation of multiple independent clinical gastric cancer data sets, and supported by in vitro studies on human gastric cancer cell lines, we reveal a hitherto unknown mechanism involving crosstalk between STAT3 and the miR-200 family to promote the molecular pathogenesis of EGC. Furthermore, our identification of a miR-200 family-associated gene signature in EGC with prognostic power suggests that the STAT3/miR-200 family axis can be exploited to identify early-stage disease and predict patient survival in gastric cancer, and potentially other related gastrointestinal cancers.

The gp130^F/F, gp130^F/F:Stat3^+/−, gp130^F/F:Il11r^−/−, and Gan mice have been described previously (15, 17, 19). All experiments were endorsed by the Monash University Monash Medical Centre “A” Animal Ethics Committee and, where appropriate, included genetically matched gp130^+/+ (wild-type) controls.

For microarrays, total RNA from 4-week-old mouse antrum samples was subjected to a mouse Affymetrix miRNA 2.0 array comprising 722 mature miRNAs and 690 pre-miRNAs. Raw data were normalized using Cyclic Loess Normalization methodology (20). qPCR profiling of miRNAs was performed using TaqMan assays (Applied Biosystems) normalized to U6 (human) and SnoRNA-202 (mouse) controls.

The miRNA and mRNA sequencing data, along with clinical information (Supplementary Table S1), were obtained from TCGA data portal (https://tcga-data.nci.nih.gov/docs/publications/stad_2014/) using the 2014 version. Alignment of sample identifiers yielded 295 tumor cases, among which 33 had matched nontumor tissues, with 236 reporting survival data. The updated 2017 TCGA data used for signature validation was obtained from http://firebrowse.org/?cohort=STAD&download_dialog=true. We used the reads per kilobase of exon per million reads mapped (RPKM) value to represent expression levels.

Gene microarray data corresponding to miRNA array data for 4-week-old mouse gastric antrum samples have been deposited in the Gene Expression Omnibus database under accession numbers GSE93173 and GSE93169 bundled into the SuperSeries GSE93196. For mRNA expression analyses, qPCR was performed using SYBR Magic, and mouse and human gene expression was normalized to 18S rRNA as described previously (17). Primer sequences are available upon request.

MKN28, MKN1 (Japanese Collection of Research Bioresources Cell Bank), and AGS (ATCC) human gastric cancer cell lines were cultured in RPMI1640 (Gibco) media supplemented with 10% FCS. Cells were authenticated by short tandem repeat profiling (PowerPlex HS16 System kit, Promega) in our laboratory after receipt in 2013, and were passaged during experiments for under 3 months at a time between freeze/thaw cycles. Cells were routinely tested for mycoplasma contamination (MycoAlert PLUS Mycoplasma Detection Kit, Lonza) during the time of experiments. Lenti-X 293T packaging cells for lentivirus production (Clontech) were maintained in DMEM (Invitrogen) containing 10% FCS. Cells were incubated at 37°C and supplemented with 5% CO₂ in a humidified chamber.

For transfections, 50%–60% confluent cells were transiently transfected with 10 nmol/L nontarget control and miR-429 mimetics (Thermo Fisher Scientific) using Lipofectamine 3000 (Thermo Fisher Scientific). Cells were incubated for 48 hours prior to harvesting for extraction of RNA and protein lysates. STAT3 activation in cells was induced by treatment with 200 ng/mL recombinant human IL11 (PeproTech).

Self-complementary oligonucleotides (Sigma-Aldrich) used as single-guided (sg) RNA sequences for targeting human STAT3 (exon 3 and 4) were ligated into the LentiCRISPRv2 construct (Addgene). Lentivirus was produced by transfecting vectors into Lenti-X/H293T cells with LentiCRISPR:psPAX2:pMD2.G at a ratio of 4:3:1. Virus was harvested 48 hours after transfection, filtered, and used to infect AGS cell cultures containing 5 μg/mL polybrene. Infected cells were selected with puromycin, and cells infected with nontarget control sgRNA vector were used as negative controls. Cloning primers are available upon requested.

Tumor epithelial samples were collected from optimal cutting temperature–embedded frozen sections stained with toluidine blue using laser microdissection (Leica). Total RNA was extracted using the miRNeasy microkit (Qiagen), and then reverse transcribed with the PrimeScript RT Kit (Takara). Dissected gastric epithelium from wild-type mice and tumor-bearing gp130^F/F or Gan mice were interrogated for expression of miRNAs and Socs3 using SYBR Premix ExTaqII (Takara), and normalized to mouse U6 or 18S rRNA, respectively. Primer sequences are available upon request.

Total protein lysates were immunoblotted with antibodies against CRTAP (Santa Cruz Biotechnology), EpCAM (Biolegend), Vimentin, pY-STAT3, total STAT3 (Cell Signaling Technology) and Actin (Thermo Fisher Scientific). Protein bands were visualized using the Odyssey Infrared Imaging System (LI-COR Biosciences) as described previously (17).

Four data sets were integrated to optimize the specificity of selecting miR-200 family–related genes. Gene set 1 (G1) comprised downregulated genes in 4-week-old gp130^F/F versus wild-type gastric antrum (cutoff: fold change<0.7) identified by gene microarray expression profiling (GSE93173), thus providing candidate genes paired to the miRNA array data in the context of tumor initiation. Gene set 2 (G2) comprises genes whose expression is inversely correlated with that of miR-200 family members from TCGA data sets (r←0.6; ref. 7). Gene set 3 (G3) contains putative miR-200 family target transcripts derived from interrogation of Ago-HITS-CLIP) data (21), which provides a range of perfect seed matches (6-mer, 7-mer, and 8-mer). Gene set 4 (G4) comprises putative miR-200 family member target genes obtained using Targetscan human v7.0 (22). A combinatorial analysis of G1+G2+G3 or G1+G2+G4 (Fig. 4A and B) was conducted to identify putative miR-200–regulated genes after replicates removal.

We validated the miR-200 family–associated gene signature using the following independent gastric cancer cohorts: Gastric Cancer Project '08 Singapore (n = 200, GSE15459; ref. 23), and ACRG (n = 300, GSE62254; ref. 6). A “single sample” extension of Geneset Enrichment Analysis (ssGSEA) score (24) was defined an enrichment score that represented the degree of absolute enrichment of the 15 genes targeted by the miR-200 family in each sample within a given data set (25). Samples with ssGSEA scores above or below the cohort mean were considered as “Sig_high” or “Sig_low,” respectively.

Cell proliferation was assessed with the ClickIT EdU Microplate Assay (Molecular Probes), and cell viability was measured using the Cell-Glo ATP assay (26). For clonigenicity monolayer assays, 2 × 10³ cells were plated in triplicates in 6-well plates. Colonies (≥50 cells) were counted after staining with Crystal Violet (Sigma-Aldrich). For soft agar colony formation assays, cells were suspended in RPMI1640 containing 0.35% agar/10% FCS, and suspensions layered on RPMI1640 containing 0.4% agar/10% FCS in triplicates in 6-well plates. Colonies were photographed at day 20–22 postplating.

Luciferase reporter plasmids encompassing a −1574 to +120 promoter segment upstream of the putative transcription start site in the human miR-200b-200a-429 gene, along with the 3′UTR reporter constructs have been reported previously (21, 27). Dual luciferase reporter assays (Promega) were conducted following the manufacturer's instructions.

All statistical analyses were performed using GraphPad Prism V6.0 software or R package. Statistical significance (P < 0.05) between the means of two groups was determined using Student t tests or Mann–Whitney U tests. Statistical comparisons of the means of multiple (R3) groups were determined using one-way ANOVA or Kruskal–Wallis nonparametric analyses. A Kaplan–Meier curve was fitted to overall survival data. The log-rank test P value was used to calculate the statistical significance of the difference in survival using the R survival package. ROC curves were constructed by altering the ssGSEA enrichment score cutoff to predict early anatomic stage IA or I. All data are from at least three individual experiments and presented as the mean ± SEM.

The onset of gastric inflammation–associated adenomatous hyperplasia in gp130^F/F mice occurs at approximately 6 weeks of age, with established intestinal-type tumors (12 weeks) progressively growing until a maximum size at 6 months (Fig. 1A; ref. 17). To identify miRNAs involved in the initiating molecular events leading to EGC, we performed miRNA microarrays on mouse gastric antrum tissue from 4-week-old gp130^F/F and sex-matched wild-type (gp130^+/+) control mice; this age was chosen since 4-week-old wild-type and gp130^F/F gastric antrum are histologically comparable, the latter being devoid of any histologic signs of inflammation or hyperplasia (Fig. 1A). As shown in Fig. 1B, 30 miRNAs were significantly (P < 0.05) differentially regulated in pretumorigenic gp130^F/F gastric tissue, with 23 miRNAs upregulated (>1.5-fold) and 7 miRNAs downregulated (<0.4-fold; GSE93169).

Consistent with the microarray data, among several significantly and highly upregulated miRNAs selected for validation by qPCR, the miR-200 family members miR-429 and miR-200b, along with the miR-21 oncomir (28) and miR-146b, were all significantly upregulated in 4-week-old gp130^F/F gastric antrum (Fig. 1C). Similarly, we also validated the significant downregulation of miR-433, miR-138, and miR-124 in gp130^F/F gastric antrum (Fig. 1C). Furthermore, other miR-200 family members miR-200a and miR-200c were upregulated (Fig. 1C), with miR-200b and miR-200c being most abundantly expressed in mouse gastric tissue (Fig. 1D).

As miR-429 was the second most significantly (P = 0.002) altered miRNA in 4-week-old gp130^F/F gastric antrum (Fig. 1B), and shares the same seed sequences with miR-200b and miR-200c, we next evaluated whether expression of miR-429 (as a representative miR-200 family member) was also elevated in established tumors in gp130^F/F mice. Indeed, miR-429 (and as a control, miR-21) was significantly upregulated in 12-week-old and 24-week-old gp130^F/F gastric tumor tissues compared with adjacent nontumor and/or wild-type antrum tissues (Fig. 1E and F). In addition, the increased expression of miR-200 family members was confirmed in laser microdissected gp130^F/F gastric tumor epithelium, along with miR-21 and the STAT3-target gene Socs3 (Fig. 1G and H; ref. 17). Importantly, miR-429 upregulation was also independently observed in the gastric tumor epithelium of the transgenic K19-Wnt/C2mE (Gan) mouse model of gastric cancer that is also characterized by STAT3 hyperactivation (Fig. 1I and J; refs. 17, 19, 29). Collectively, these data suggest the involvement of numerous miRNAs in the initiating stage of gastric inflammation–associated tumorigenesis, in particular miR-429 and other miR-200 family members.

The translational potential of our in vivo findings was explored upon interrogation of TCGA data sets, which revealed that expression of miR-429, miR-200a, miR-200b, miR-200c, and miR-21 was also significantly increased in tumor tissue compared with paired, adjacent nontumor tissue (Fig. 2A; Supplementary Fig. S1A), ranging from 22 to 25/33 (∼70%) of gastric cancer patients. Notably, elevated miR-429 expression also correlated with various clinical covariates of TCGA gastric cancer patients, including earlier age at initial diagnosis (≤65 years; Supplementary Table S1) and exclusively the Lauren classification of intestinal-type gastric cancer (Fig. 2B). The significant correlation between upregulated miR-429 and only intestinal-type gastric cancer was also observed for other miR-200 family members, but not miR-21 (Fig. 2B; Supplementary Fig. S1B). Furthermore, miR-200 family members, but not miR-21, were significantly upregulated in the papillary and/or tubular subtypes compared with the mucinous subtype of WHO intestinal subclass (Supplementary Fig. S1C and S1D). Similarly, significant increases in expression of miR-200 family members were largely restricted to EGC (anatomic stage IA, T1 stage) rather than more advanced disease stages, while miR-21 expression was significantly augmented throughout all progressive stages of gastric cancer (Fig. 2C; Supplementary Fig. S1E and S1G). The increased expression of miR-200 family members and miR-21 did not correlate with other clinical parameters such as lymph node involvement and remote metastasis (Supplementary Fig. S1F, S1H, and S1I). Collectively, these data suggest that elevated expression of miR-429 and other miR-200 family members is associated with EGC, which is consistent with elevated expression of these miRNAs in the gp130^F/F model.

Our demonstration that increased expression of miR-429 (and other miR-200 family members) is associated with EGC suggests that gastric cancer patients characterized by elevated miR-200 family expression would have improved survival outcomes compared with those with low expression levels. Indeed, assessment of the survival of TCGA gastric cancer patients that were stratified into “high” and “low” expressing groups for each individual miR-200 family member revealed that gastric cancer patients with high expression of miR-429 (Fig. 2D), miR-200a, or miR-200b, but not miR-200c (Supplementary Fig. S2A), had a significantly better prognosis than those with low expression of these miRNAs. As miR-429 expression is tightly correlated with that of other miR-200 family members in TCGA gastric cancer patients (Supplementary Fig. S2B), we confirmed that the combination of all miR-200 family members yielded a highly significant (P = 0.0079) predictive power for patient survival (Fig. 2E).

Interestingly, combining miRNA expression data with RNA-Seq data for TCGA cohorts illustrated that patients classified as high expressing for miR-429 (“429 high”) or all miR-200 family members (“all high”), and low expressing for miR-429 (“429 low”) or all miR-200 family members (“all low”), demonstrated a different transcriptomic signature (LogFC > 2, P < 10⁻⁵; Fig. 2F and G). Furthermore, GSEA using hallmark and curated gene sets revealed that the “429 high” patient cohort was enriched for stress-related unfolded protein response, oxidative phosphorylation, mTORC1 and c-MYC signaling gene networks (which are important for cell-cycle progression and metabolism), whereas myogenesis, epithelial–mesenchymal transition (EMT), integrin signaling, and focal adhesion gene networks were enriched in “429 low” patients (Fig. 2H and I; Supplementary Fig. S2C).

The above data suggest that the miR-429 and other miR-200 family members may play a dual role during gastric cancer: upregulation of miR-200 family members during early tumor initiation supports stress-related cellular events within the gastric epithelium, leading to a preoncogenic hyperproliferative state, whereas their downregulation in later stages facilitates EMT-related malignant transformation of the gastric epithelium, resulting in disease progression and an associated poor prognosis. Such a notion is consistent with the observed differential expression patterns of miR-200 family members among histologic subtypes and disease stages, along with their prognostic power (Fig. 2A–E; Supplementary Figs. S1D and S1G and S2A).

To further validate this hypothesis, we assessed the biological function of miR-429 in human gastric cancer cell lines with distinct histologic subtypes that represent early and late stages of disease. The differentiated, intestinal-type cell line MKN28 (30) exhibits a strong epithelial phenotype characterized by high EpCAM expression and little expression of the mesenchymal marker, Vimentin (Fig. 3A and B). Conversely, the nonintestinal-type cell line MKN1 (30) with spindle-shaped mesenchymal morphology displays weak EpCAM expression and high Vimentin expression (Fig. 3A and B). Therefore, the epithelial phenotype of MKN28 cells aligns with cell proliferation and clonal growth, which is reflective of early-stage disease, while the mesenchymal morphology of MKN1 cells favors metastasis during late-stage disease progression (31, 32). In support of our hypothesis, the stable overexpression of miR-429 in epithelial-like MKN28 cells significantly augmented cell viability, proliferation, and anchorage-independent growth in soft agar compared with parental and control empty vector–transduced cells (Fig. 3C–I). Furthermore, increased cellular proliferation and clonogenicity upon miR-429–stable overexpression was confirmed in another intestinal-type gastric cancer cell line, AGS (33) also displaying epithelial traits (Supplementary Fig. S3). In contrast, miR-429 overexpression in mesenchymal-like MKN1 cells significantly suppressed cell viability and proliferation in monolayer clonogenic growth (Fig. 3J–P).

We next sought to identify a signature of target genes regulated by the miR-200 family in human gastric cancer, which initially involved integrating four individual gene sets (see “Materials and Methods”) to optimize the specificity of selecting predicted miR-200 family targets, as follows: genes downregulated in gp130^F/F pretumor gastric tissue (G1) and their negative correlation with miR-200 family members in TCGA gastric cancer patients (G2), plus either direct interaction with miR-200 family members from AGO-HITS-CLIP (G3) or containing perfect seed sites within 3′UTR in silico (G4). Upon undertaking a combinatorial analysis of G1+G2+G3 or G1+G2+G4 sets, 15 putative miR-200 family-regulated genes were identified (Fig. 4A and B).

To experimentally validate our in silico approach of identifying miR-200 family–regulated genes, 6 genes (CRTAP, LHFP, SYDE1, RECK, MSN, ZEB1) representing the 15 identified genes were selected and corresponding 3′UTR reporter constructs were generated. Indeed, the reporter activities of all 6 genes were significantly inhibited when cells were cotransfected with miR-429 mimics (Fig. 4C). As the oncogenic role of CRTAP as a miR-200 family target is unknown, we further investigated its endogenous expression and function in intestinal-type MKN28 and nonintestinal-type MKN1 gastric cancer cell lines. Following transfection of miR-429 mimics, CRTAP expression was reduced by 40%–50% at both the mRNA and protein levels (Fig. 4D–F). Consistent with the cellular function of miR-429, siRNA-mediated knockdown of CRTAP expression (by ∼80%) also significantly increased and reduced proliferation in MKN28 and MKN1 cells, respectively (Fig. 4G and H). Collectively, these data suggest the existence of a miR-200 family-regulated gene signature involved in the pathogenesis of EGC.

Importantly, in TCGA gastric cancer patient data sets, the expression level of the miR-200 family–regulated 15-gene signature significantly and inversely correlated with that of miR-429 (and/or other miR-200 family members; Fig. 5A and B). Also, miR-200 family high-expressing patients, corresponding to low expression of the 15-gene signature (“Sig_low”), demonstrated longer overall survival than miR-200 family low-expressing patients with high expression of the 15-gene signature (“Sig_high”; Fig. 5C and D). In contrast, only 7 of 15 miR-200 family target genes had a significant predictive value for survival when assessed alone (Supplementary Fig. S4), each of which was inferior to that acquired for the full 15-gene signature for predicting survival (Fig. 5D), suggesting that the combination of all 15 genes, rather than each gene in isolation, is optimal for predicting miR-200 family–dependent gastric cancer patient survival. Furthermore, ROC analysis of the expression of the 15-gene signature in TCGA gastric cancer patient data sets for early-stage IA (n = 8) versus nonstage IA (n = 228) disease confirmed the predictive ability of this signature to significantly stratify early-stage IA, as determined by an average area under the curve (AUC) value of 0.82 (Fig. 5E).

We further investigated whether such a gene signature had prognostic power in two other independent gastric cancer cohorts, the “Gastric Cancer Project '08 Singapore” and the “Asian Cancer Research Group” (ACRG) cohorts, for which there was no matching miRNA expression data. On the basis of the approach employed for TCGA gastric cancer patient data sets, patients from these additional cohorts were stratified into ‘and groups for the 15-gene signature (Fig. 5F and I). Consistently, the overall survival of “Sig_low” patients was significantly improved compared with their “Sig_high” counterparts (Fig. 5G and J). In addition, although there was no American Joint Committee on Cancer (AJCC) substage (e.g., stage IA, IB) information available for these latter two data sets, nonetheless the predictive ability (AUC scores = 0.61–0.63) of the gene signature to stratify early-stage I patients was again confirmed in each patient cohort (Fig. 5H and K).

To further evaluate the clinical utility of this 15-gene signature, we next investigated its correlation with key indicators for endoscopic resection, based on current guidelines recommended by the Japan Gastroenterological Endoscopy Society (JGES) and Japanese Gastric Cancer Association (JGCA; ref. 34), using updated TCGA (n = 430) and ACRG (n = 300) cohort data sets. Notably, χ² analyses revealed significant correlations between a high miR-200 family/low 15-gene signature expression profile and tumor grade 1–2 (well-to-moderate differentiated), as well as tubular and papillary histologic subtypes (Supplementary Table S2). ROC modeling on TCGA data sets also confirmed the predictive power of the 15-gene signature to accurately stratify gastric cancer patients with T1 (AUC score = 0.849; n = 22; Fig. 5L) or well-to-moderate differentiated tubular and papillary subtypes (AUC score = 0.748; n = 50; Fig. 5M), and combining these two criteria increased the power of the signature to predict patients satisfying both T1 and well-to-moderate differentiated tubular and papillary histology (AUC score = 0.90; n = 6; Fig. 5N). Importantly, the 15-gene signature also accurately predicted two patients, who were postoperatively diagnosed, as T1a with well-to-moderate differentiated histology, which according to current JGES/JGCA guidelines is considered as an indication for endoscopic resection (AUC score = 0.998; Fig. 5O). Furthermore, a multivariate analysis of overall survival comparing this 15-gene signature against other current clinical parameters in TCGA gastric cancer patients revealed that the 15-gene signature is superior for survival prediction compared with either tumor depth of invasion, histology, or differentiation alone (Supplementary Table S3). Collectively, these data demonstrate the potential clinical utility of the miR-200 family–associated 15-gene signature in EGC as an accurate molecular marker with prognostic power that aligns with current guidelines for indicating endoscopic resection.

As the gastric tumor phenotype of gp130^F/F mice is driven by hyperactivation of STAT3 via IL11 (15, 16), we next investigated whether miR-429 expression was regulated by IL11-induced STAT3 signaling. The augmented expression of miR-429 and miR-200b in 4-week-old gp130^F/F gastric antrum tissue was significantly reduced to wild-type levels in antral tissue of gp130^F/F:Stat3^−/+ and gp130^F/F:Il11r^−/− mice displaying reduced STAT3 activation (Fig. 6A and B). Also, elevated miR-200a and miR-200c expression was reduced in antral tissue of mice with compromised STAT3 signaling, albeit not significantly (Fig. 6B). Furthermore, miR-429, miR-200a, and miR-200b expression significantly correlated with expression of the STAT3 target gene, Socs3 in all genotypes at this age regardless of displaying either elevated or reduced STAT3 activation (Fig. 6C and D). Notably, miR-429 and other miR-200 family members were also significantly upregulated in an IL11/STAT3-dependent manner in 12-week-old gp130^F/F gastric tumor tissue (Supplementary Fig. S5A and S5B). Consistently, a significant and positive correlation also existed between miR-429 and Socs3 among all 12-week-old antrum tissue (Supplementary Fig. S5C and S5D). These findings were also evident for the STAT3-regulated miR-21 (Fig. 6C; Supplementary Fig. S5C; ref. 28). Collectively, these data demonstrate that augmented miR-429 expression is associated with IL11-driven STAT3 hyperactivation during the initiation and early establishment phases of gastric tumorigenesis in the gp130^F/F mouse.

We next examined whether miR-429 expression in human gastric epithelial (cancer) cells is stimulated by IL11, which is the main STAT3-activating cytokine during gastric tumorigenesis (15, 35). In AGS and MKN28 human gastric cancer cells, expression of miR-429, miR-200a, and 200b, along with SOCS3 and miR-21, was induced within 0.5–1 hour of IL11 treatment (Fig. 6E–G; Supplementary Fig. S5E–S5G). In contrast, CRISPR/Cas9-mediated ablation of STAT3 expression (and thus activation) in both AGS and MKN28 cells (Supplementary Fig. S6A–S6E) suppressed the ability of IL11 to upregulate miR-429, as well as SOCS3 and miR-21 (Fig. 6E–G; Supplementary Fig. S5E–S5G). The notion that miR-429 expression is upregulated by IL11-induced STAT3 activation was also supported by the stable overexpression of wild-type STAT3 in MKN28 cells, whereby IL11 stimulation significantly augmented expression of miR-429, miR-21, and SOCS3 above that observed in control cells transduced with empty vector (Fig. 6H and I).

We next investigated whether the STAT3-mediated upregulation of miR-429 was at the transcriptional level. However, expression levels of pri-miR-429, unlike mature miR-429, were unchanged in MKN28 cells following IL11 treatment (Supplementary Fig. S6F and S6G). Overexpression of wild-type STAT3 along with a full-length or truncated miR-200b-200a-429 promoter-reporter construct in HEK293T cells also had no effect on the transcriptional activity of either promoter construct, whereas significantly increased luciferase activity was observed with a control STAT3-regulated TLR2 promoter reporter (17) or upon cotransfection with the transcriptional activator CREB binding protein (CBP; Supplementary Fig. S6H–S6K). Therefore, these data suggest that during gastric tumorigenesis, STAT3 augments the expression of the miR-200 family independent of any direct transcription factor activity on the miR-200b-200a-429 promoter.

It is well established that miR-200 family members play a key role in modulating metastasis-associated processes during advanced stages of cancer progression, which invariably have been attributed to suppression of EMT, and more recently regulation of oxidative stress and the cell secretome (36–38). However, the miR-200 family has been assigned oncogenic and tumor-suppressive roles in diverse cancers, as evidenced by their upregulation in lung, ovarian, and esophageal cancers, and conversely downregulation in hepatocellular and gastric cancers (39, 40). Also, miR-200 family upregulation can have opposing effects on cancer patient prognoses, with high expression correlated with increased (e.g., ovarian) or reduced (e.g., breast) survival rates (36, 37). While these observations imply that the expression and function of miR-200 family members is influenced by cell type and tissue specificity, it is notable that also within a cancer type (e.g., colorectal, lung), family members such as miR-429 are differentially expressed and display contrasting anti- and protumorigenic roles (39, 41), suggesting that at least in some cancers, the role of the miR-200 family is also stage-dependent.

Considering the paucity of information on the role of miR-200 family members during the early initiation stages of tumorigenesis, here we reveal that miR-429 and other miR-200 family members are upregulated in preneoplastic gastric epithelial tissue of gp130^F/F mice, and display a dynamic expression profile in human gastric cancer, which is characterized by the selective upregulation only in EGC, namely TNM stage IA. Indeed, the steady decline in miR-200 family members' expression from stage IB onwards toward the more advanced disease stages is consistent with their tumor suppressor functions (21, 42). In this respect, our current findings suggesting that miR-200 family members display a bimodal functionality in gastric cancer align with their differential and stage-dependent expression, and provide a strong molecular stratification of early versus late-stage gastric cancer. In support of a disease stage-dependent contrasting role for miR-200 family members in gastric cancer, our in vitro data demonstrate that miR-429 overexpression in the MKN28 epithelial-like gastric cancer cell line supports a hyperproliferative response, while conversely, its upregulation in mesenchymal-like MKN1 gastric cancer cells inhibits their proliferative potential.

A key outcome of our current study was the discovery of a stringent miR-200 family–associated 15-gene signature implicated in the initiating stages of EGC, which included several genes controlling cytoskeleton dynamics (e.g., MSN; ref. 21), and identification of CRTAP as a miR-200 family target gene whose expression can influence cell growth dependent on the histologic cellular gastric cancer subtype. Importantly, in three independent gastric cancer patient cohorts, expression of this 15-gene signature strongly aligned (inversely) with the dynamic expression profile of miR-200 family members throughout early to late stages of disease, and demonstrated the independent capacity as a significant predictor of patient survival. Moreover, we demonstrate that the clinical utility of the signature integrates multiple key indicators of EGC, such as tumor invasion depth, differentiation, histology and stage, and the predictive power of this signature for patient survival is superior to either histology, differentiation, or the depth of invasion alone. Therefore, we propose that this signature provides a novel molecular marker, which strongly aligns with the current guidelines for endoscopic resection criteria and the evaluation of prognosis for EGC patients (34), to complement endoscopic-based procedures, especially in cases when an accurate preoperative endoscopic diagnosis for the depth of invasion such as T1b or T2 is challenging. Furthermore, our data showing that low expression of the miR-200 family-associated gene signature (or high miR-200 family expression) exhibited a strong association with tumor grade 1–2 (well-to-moderate differentiated) and tubular/papillary histologic subtypes, which are indicative of favorable outcomes based on current guidelines, suggest that the clinical utility of the gene signature can extend to distinguishing histopathologic characteristics of EGC, which is a key factor influencing tumor recurrence and prognosis following endoscopic treatment.

Although the miR-200 family and associated gene signature displayed no significant correlation with lymph node involvement, a previous study by Gotoda and colleagues suggested that the risk of lymph node metastasis could still be minimized in EGC patients with limited invasion depth (confined to submucosal layer) and with well-differentiated histology (43). Therefore, we propose that EGC patients with high miR-200 family (low gene signature) expression would have minimal risk of lymph node metastasis, although we do acknowledge that a large prospective study is required for further validation of this hypothesis.

It is also worth considering our findings with recent studies showing that impaired expression of miR-200 family members, predominantly miR-200b and miR-200c, is a marker of poor prognosis for gastric cancer patients (13, 14, 44), which concur with our data revealing that high miR-200 family expression is a favorable prognostic factor. In addition, we refer to a recent retrospective study by Sohn and colleagues (45) in advanced gastric cancer patients, which developed a prediction model based on four molecular subtypes of advanced gastric cancer, derived from TCGA data sets, to stratify patients by overall survival and adjuvant chemotherapy outcomes (7). While the clinical potential of such a study to the management of advanced-stage disease patients is clear, unlike our current study, there was no specific analysis of early-stage disease (i.e., TNM I), and therefore such studies have limited scope for EGC.

It is also worth noting that our current study does not assess any effectiveness of postoperative treatment regimens because aggressive clinical treatments such as chemotherapy or other multimodal treatments are only considered in advanced stage patients or in cases with recurrence following minimal invasive treatment. While the survival of TCGA gastric cancer patients can be improved by the addition of chemotherapy, our data reveal that high miR-200 family/low gene signature expressing patients are at a relative earlier disease stage, with less than half of the patients (41%) having pre/postoperative chemotherapy, suggesting that the benefit from chemotherapy treatment in such patients for a favorable prognosis is limited. Importantly, the clinical potential of our current retrospective study in EGC, along with that by Sohn and colleagues in advanced-stage gastric cancer, now set the stage to further verify the clinical implications of molecular-based signatures for early and advanced gastric cancer in larger prospective cohorts.

Another key finding of our study was that miR-429 expression is positively regulated by IL11/STAT3 signaling, a key driver of gastric cancer (15, 35). Intriguingly, however, STAT3 did not appear to directly transcriptionally regulate miR-429 at the transcriptional level, and while the mechanism requires further investigation, this finding is perhaps not surprising as the five miRNAs of the miR-200 family exhibit highly-correlated expression patterns and exist on two separate loci (Fig. 5C; Supplementary Fig. S2B and S6I). In this respect, we cannot exclude the possibility that STAT3 may indirectly regulate miR-200 family members via its transcriptional regulation of miRNA processors, such as DDX1 (46), which in turn regulate miR-200 family maturation. Considering the lack of clinical efficacy of targeting STAT3 in metastatic disease settings (47), our current findings on the importance of the STAT3/miR-200 family axis in early stage of disease, along with our previous reports for IL11/STAT3 signaling playing a critical role in gastric cancer initiation (15, 16), suggest that modulation of the STAT3 pathway in the localized EGC setting should be considered in the clinic.

In summary, we reveal here that upregulation of the IL11/STAT3/miR-200 family axis is associated with EGC (Supplementary Fig. S6L). Furthermore, the discovery of the miR-200 family and its 15-gene signature provides a potential novel molecular marker for “early and less malignant” tumors to accurately stratify EGC patients for endoscopic resection, as well as inform prognosis in gastric cancer. In addition, our current data now also provide the platform to evaluate the potential therapeutic efficacy of miR-429 blockade on arresting tumor initiation (for instance in our gp130^F/F EGC model). An important consideration that also arises from our current study demonstrating the dynamic expression profile of miR-200 family members throughout multiple stages of gastric cancer is the need for future studies on gastric cancer, and other cancers, to evaluate the expression of miRNAs in all stages of disease, the results of which can influence interpretations as to whether a specific miRNA has tumor-suppressive or oncogenic behavior.

No potential conflicts of interest were disclosed.

Conception and design: L. Yu, D. Wu, B.J. Jenkins

Development of methodology: L. Yu, D. Wu, C.L. Kennedy, B.J. Jenkins

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Yu, J.J. Balic, T.-S. Han, Y.D. Liu, C.L. Kennedy, M. Oshima, G.J. Goodall, B.J. Jenkins

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Yu, D. Wu, H. Gao, A. Tsykin, P. Tan, G.J. Goodall, B.J. Jenkins

Writing, review, and/or revision of the manuscript: L. Yu, D. Wu, Y.D. Liu, G.J. Goodall, B.J. Jenkins

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.D. Liu, J.K. Li, J.Q. Mao

Study supervision: J.K. Li, J.Q. Mao, B.J. Jenkins

The authors thank Dr. Michael Gantier for advice and comments, and also Meri Najdovska for technical assistance. The authors are grateful to Dr. Daniel Gough for kindly providing the pLVX mCherry construct. This work was supported by grants awarded to B.J. Jenkins from the Association for International Cancer Research (United Kingdom) (Grant Reference Number 11-0017) and Cancer Council of Victoria (Australia) (Grant Reference Number 1003597), as well as the Operational Infrastructure Support Program by the Victorian Government of Australia. B.J. Jenkins is supported by a Senior Research Fellowship awarded by the National Health and Medical Research Council (Australia).

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.