RETRACTED: KMT2C Mutations in Diffuse-Type Gastric Adenocarcinoma Promote Epithelial-to-Mesenchymal Transition

There are nearly 1 million new gastric cancer cases worldwide per year and nearly 700,000 gastric cancer deaths per year, and this accounts for almost 10% of all cancer deaths (1). Gastric adenocarcinomas (GAs) comprise the vast majority of gastric cancers. The majority of patients with GA present with locally advanced or metastatic disease. Overall survival for patients with metastatic disease is 3 to 5 months with best supportive care (2). The response rate to multiagent chemotherapy can be 50% or greater, but nearly all patients develop chemotherapy resistance, and median survival is extended only to 10 to 12 months (3).

In 1965, Lauren described two distinct histologic types of GAs: intestinal and diffuse (4). The intestinal type exhibits components of glandular or intestinal architecture which is thought to arise through a sequence of multistep carcinogenesis often resulting from chronic mucosal inflammation caused by Helicobacter pylori (H. pylori) infection (5). This type is more common in men and older patients. The diffuse type demonstrates poorly cohesive cells infiltrating the gastric wall, and progressive disease can ultimately lead to linitis plastica (a.k.a. leather bottle stomach; ref. 6). This type is more common in women and in younger patients and more associated with familial occurrence.

The driver mutations leading to GA tumorigenesis, progression, and metastasis are being increasingly defined. The Cancer Genome Atlas (TCGA) proposed a molecular classification of GAs into four subtypes: tumors positive for Epstein–Barr virus, microsatellite-unstable tumors, tumors with chromosomal instability, and genomically stable tumors (7). The vast majority of diffuse-type GAs are genomically stable. CDH1, which encodes for the E-cadherin cell adhesion protein, is frequently inactivated in diffuse-type GAs by mutation or hypermethylation of its promoter (8). The gene encoding the tumor suppressor p53 is mutated in about one half of all GAs, but these mutations occur more commonly in intestinal-type tumors (7). Recently, two studies in Nature Genetics found mutations in RHOA in 14% to 25% of diffuse-type GAs (9, 10). This high rate of RHOA mutation in diffuse GAs was confirmed by TCGA, and TCGA also found additional fusions in GTPase-activating proteins, which regulate RhoA activity (7).

Despite the ever-increasing number of genomic studies of nearly all cancers, no genomic study has specifically examined diffuse-type GA in Western patients with nonmetastatic disease. We performed whole-exome sequencing on extracted DNA from 27 tumors samples along with matched normal mononuclear cell DNA from patients with diffuse GA undergoing potentially curative surgical resection. SNP arrays were also performed on 23 of these samples. Surprisingly, lysine (K)-specific methyltransferase 2C (KMT2C), also known as mixed lineage leukemia 3 (MLL3), was the most commonly mutated gene, with 11 of 27 samples (41%) harboring KMT2C mutations. KMT2C is a histone methyltransferase involved in transcriptional coactivation and often deleted in myeloid leukemias (11). There are a few reports on the KMT2C mutations in solid tumors including gastric cancer (12–14). We thus further explored the role of KMT2C as a prognostic factor by examining KMT2C protein expression in 135 patients with diffuse GA undergoing potentially curative resection and examined the effects of KMT2C loss and gain in gastric epithelial cells and diffuse GA cell lines, respectively.

We examined the prospective Memorial Sloan Kettering Cancer Center (MSKCC) gastric database and identified 271 patients who underwent potentially curative gastric resection between September 2006 and November 2011. Patients with intestinal- or mixed-type GA were excluded, leaving 71 patients with diffuse-type GA. Adequate frozen tumor tissue and blood samples were available in 27 of these patients. All subjects provided written-informed consent for genetic analyses prospectively, and Institutional Review Board approval was obtained for this study. The study was conducted in accordance with the International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS). Blood specimens were collected preoperatively. Tumor specimens were collected at the time of gastrectomy.

Prospectively collected data from the MSKCC gastric database were reviewed retrospectively for each of the patients. Clinicopathologic characteristics, including age, gender, race, H. pylori infection, presence of atrophic gastritis or intestinal metaplasia, tumor invasion depth, number of lymph node metastasis, recurrence, neoadjuvant and adjuvant therapy, and death were recorded. Overall survival was defined as the time from surgery to the date of death or to the last follow-up date. Recurrence-free survival was defined as the time from surgery to the date of recurrence. The pathologic stage was determined using the 7th edition of the American Joint Committee on Cancer classification system (15). Patients were followed until death or January 31, 2014, whichever occurred first. The median follow-up interval was 30 months (range, 0–77 months).

DNA was isolated from tumor specimens, and normal DNA was isolated from leukocytes in peripheral blood. Whole-exome sequencing was performed on DNA from all 27 blood and tumor samples. We extracted the DNA from the paired tumor and peripheral blood leukocytes using QIAamp DNA Mini Kit according to the manufacturer's protocol (Qiagen). Genomic DNA samples were constructed into Illumina paired-end precapture libraries and prepared using protocols recommended by Illumina. Captured DNA libraries were sequenced with the Illumina HiSeq 2000 Genome Analyzer to an average coverage of 144X, yielding 150 (2 × 75) base pairs from the final library fragments. Reads were aligned to the hg19 (GRCh37) build using the Burrows–Wheeler Aligner (BWA; ref. 16), and then Genome Analysis Toolkit (GATK; ref. 17) was used for base quality score recalibration, indel realignment, and duplicate read removal. The MuTect algorithm (18) was used to identify somatic single-nucleotide variants (SNV) in whole-exome sequencing data. MuTect identifies candidate somatic SNVs by Bayesian statistical analysis of bases and their qualities in the tumor and normal BAM files at a given genomic locus.

The data processing pipeline for detecting variants in Illumina HiSeq data is as follows. First, the FASTQ files are processed to remove any adapter sequences at the end of the reads using cutadapt (v1.6). The files are then mapped using the BWA mapper (bwa mem v0.7.12). After mapping, the SAM files are sorted and read group tags are added using the PICARD tools. After sorting in coordinate order, the BAMs are processed with PICARD Mark Duplicates. The marked BAM files are then processed using the GATK toolkit (v3.2) according the best practices for tumor normal pairs. They are first realigned using the InDel realigner, and then the base quality values are recalibrated with the BaseQRecalibrator. Somatic variants are then called in the processed BAMs using MuTect (v1.1.7).

A total of 17 somatic SNVs on KMT2C identified by whole-exome sequencing were verified by Amplicon sequencing or conventional Sanger sequencing. Amplicon sequencing was performed using Ion Torrent AmpliSeq Cancer Panel (19) on tumor and corresponding mononuclear cell DNA. Twelve pairs of forward and reverse primers of each amplicon of the AmpliSeq custom panels (Life technologies) were made according to the manufacturer's instruction (Supplementary Table S1). Sanger sequencing using dye-terminator chemistry was analyzed with an automatic sequencer ABI 3730 (Applied Biosystems). The target regions were amplified by PCR followed by direct sequencing. The sequencing primers are listed in Supplementary Table S2.

Due to limited DNA in a few samples, SNP array analysis was performed in 23 of 27 tumors. DNA samples were assayed on Affymetrix Genome-Wide Human SNP 6.0 Arrays according to the manufacturer's instruction. Data were processed using the Circular Binary Segmentation program to identify statistically significant changes in copy number (20). Data were then analyzed with the RAE algorithm (21) that robustly maps chromosomal alterations. We also analyzed diffuse-type GA tumors (n = 69) out of the TCGA cohort (7) from the TCGA portal (http://cancergenome.nih.gov), which were also analyzed on Affymetrix Genome-Wide Human SNP 6.0 arrays.

We used MKN-45 and SNU-668 diffuse-type GA cell lines obtained from the ATCC and maintained as previously described (22). The ATCC performs cell line characterization using short tandem repeat DNA profiling. Cell lines were actively passaged for less than 6 months from the time that they were received from the ATCC, and United Kingdom Co-ordinating Committee on Cancer Research guidelines were followed (23). The immortalized human normal gastric epithelial cell line HFE145 was a gift from Dr. Hassan Ashktorab and Duane T. Smoot (Howard University, Washington), and it was maintained in RPMI 1640 media as previously described (24). 5-Fluorouracil (5-FU) and cisplatin were purchased from US Biological and Enzo Life Sciences, respectively. To grow cells as spheroids, cells were resuspended in DMEM-F12 containing 20 ng/mL of EGF, bFGF, N-2 (1X), and B27 (1X) and plated on Ultra-Low Attachment culture dishes (Corning Life Sciences) as we have as previously described (22).

Cells were dissociated using Accutase and resuspended in PBS containing 0.5% BSA. The cells were stained with FITC-conjugated CD44 (BD555478) or isotype control antibody (BD555742) from BD Biosciences on ice for 30 minutes. Cells were then washed with PBS and analyzed on a BD FACSCalibur (BD Biosciences) using Cell Quest software.

Silencing of KMT2C was achieved via lentiviral transduction of human KMT2C shRNA (SC-67719; Santa Cruz Biotechnology). A scramble shRNA control (SC-108080; Santa Cruz Biotechnology) was also used. Maximal knockdown of KMT2C occurred 72 to 96 hours after transduction. KMT2C overexpression was created using KMT2C lentiviral activation particles (sc-402052-LAC; Santa Cruz Biotechnology) following the manufacturer's protocol. Control lentiviral activation particles (sc-437282; Santa Cruz Biotechnology) were also used. MLL3 overexpression was created using MLL3 lentiviral activation particles (sc-437348-LAC; Santa Cruz Biotechnology) following the manufacturer's protocol.

Spheroid cells were dissociated with Accutase, and monolayer cells were collected with trypsin. To assay for proliferation, 1 × 10⁴ cells were plated onto 96-well flat bottom plates and maintained in regular media overnight. A colorimetric MTT assay was used to assess cell number by optical density after 3 days as previously described (25). Day 1 represents the time of cell plating. Data reflect the mean of six samples. Soft-agar colony formation, single cell, and colony formation assays were performed as previously described (22). To assay for migration and invasion, cells (2 × 10⁴ cells/well) were suspended in 0.2 mL of serum-free DMEM for invasion and motility assays. For the invasion assay, the cells were loaded in the upper well of the Transwell chamber (8-μm pore size; Corning) that was precoated with 10 mg/mL growth factor reduced Matrigel (BD Matrigel matrix; BD Biosciences) on an upper side of the chamber with the lower well filled with 0.8 mL of DMEM with serum. After incubation for 48 hours at 37°C, noninvaded cells on the upper surface of the filter were removed with a cotton swab, and migrated cells on the lower surface of the filter were fixed and stained with a Diff-Quick kit (Fisher Scientific) and photographed at 20× magnification. Invasiveness was determined by counting cells in five microscopic fields per well, and the extent of invasion was expressed as an average number of cells per microscopic field. Cells were imaged by phase-contrast microscopy. For migration studies, we used invasion chambers with control inserts that contained the same type of membrane but without the Matrigel coating. Note that 2 × 10⁴ cells in 0.2 mL of serum-free DMEM were added to the apical side of each insert, and 0.8 mL of DMEM with serum was added to the basal side of each insert. The inserts were processed as described above for the invasion assay.

To assay for apoptosis, cells were resuspended in 100 μL of 1% FBS in PBS, mixed with 100 μL of Annexin V reagent (Muse Annexin V & Dead cell kit, MCH100105, Millipore), and incubated at room temperature for 20 minutes. Cells were then analyzed on a Muse Cell Analyzer (Millipore Sigma) per the manufacturer's instructions.

Samples were collected, and Western blots were performed as previously described (26). Western blot analysis was performed using the following antibodies: Sox2 (#2748, #3579), Oct-4 (#2750), Nanog (#4893), Slug (#9585), Snail (#3879), and CD44 (#3578) purchased from Cell Signaling Technology; KMT2C (71200) from Abcam; c-Myc (sc-40) and MLL3 (sc-67719P) from Santa Cruz Biotechnology; E-cadherin (BD610182) and N-cadherin (BD610920) from BD Biosciences; Zeb1 (NBP-1-05987) from Novus Biologicals; and β-actin from Sigma.

Formalin-fixed, paraffin-embedded sections were deparaffinized by xylene and rehydrated. Immunohistochemistry was performed with VECTASTAIN Elite ABC kit (Vector Laboratories Inc.) following the manufacturer's protocol. Formalin-fixed, paraffin-embedded sections were processed and stained as previously described (27) using KMT2C antibody (ab71200; 1:100 dilution; Abcam). KMT2C staining was predominantly localized to the cytoplasm. KMT2C scores (0–300) were calculated by multiplying the staining intensity (0, 1, 2, or 3) by the staining extent (0%–100%). We performed immunofluorescence as previously described (26). Antibodies were used as follows: anti-human CD44 (#3570; Cell Signaling Technology), anti-Sox2 (#3579; Cell Signaling Technology), anti–E-cadherin (610182; BD Biosciences), anti-Slug (#9585; Cell Signaling Technology), and anti–phospho-Histone H2AX (#05-636; Millipore). Nuclei were counterstained using DAPI. Stained cells were visualized using an inverted confocal microscope, and image was processed using Imaris 7.6.

Tumor tissue microarrays (TMA) were constructed using tumors from patients with diffuse-type GA who underwent radical gastrectomy or esophagogastrectomy with potentially curative intent (R0 and R1) from May 2006 to March 2012 at Seoul National University Bundang Hospital (SNUBH, South Korea) using a tissue array device (Beecher Instruments Inc.). A representative core biopsy (2 mm in diameter) was obtained from each case of tumor and embedded in a TMA block. KMT2C immunohistochemistry was performed as described above. Prospectively collected data from the SNUBH gastric database were reviewed retrospectively for each of the patients. Clinicopathologic characteristics, including age, gender, tumor invasion depth, number of lymph node metastasis, recurrence, and death, were recorded. The pathologic stage was determined using the 7th edition of the International Union Against Cancer/American Joint Committee on Cancer classification system. The end of the follow-up period was March 5, 2012. The median follow-up time was 60 months (range, 1–96 months). The SNUBH Institutional Review Board approved this study, and informed consent for study of tumor tissue was obtained preoperatively from all patients. The study was conducted in accordance with the International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS).

All mouse protocols were approved by the Institutional Animal Care and Use Committee. Atp4b-Cre; Cdh1^fl/fl; LSL-Kras^G12D; Trp53^fl/fl; Rosa26^LSL-YFP (ACKPY) mice were generated as previously described (28). ACKPY3077 and ACKPY3944 gastric tumors in ACKPY mice were harvested, and organoids were isolated as previously described (29). For in vitro culture, organoids were mixed with 50 μL of Matrigel (Cat. 354248; BD Bioscience) and plated in 24-well plates. After polymerization of Matrigel, cells were overlaid with medium containing AdDMEM)/F12 supplemented with penicillin/streptomycin, 10 mmol/L HEPES, GlutaMAX, 1 × B27, 1 × N2 (Invitrogen), and 1.25 mmol/L N-acetylcysteine (Sigma-Aldrich). Growth factors for organoid culture were added to the following essential components (30): 0.05 μg/mL EGF, 0.1 μg/mL fibroblast growth factor-basic (FGF-basic), 0.01 μmol/L Gastrin I, 10 mmol/L nicotinamide, 10 μmol/L Y-27632, SB202190 (all from Sigma Aldrich), 1 μmol/L prostaglandin E2 (Tocris Bioscience), 0.5 μg/mL recombinant R-spondin1, 0.1 μg/mL mNoggin, 0.1 μg/mL fibroblast growth factor-10 (FGF-10, all from PeproTech), and 0.1 μg/mL Wnt3A, 0.5 μmol/L A83-01 (all from R&D Systems). For passage, organoids were removed from Matrigel with BD Cell Recovery Solution (BD Biosciences) following the manufacturer's instruction and transferred to fresh Matrigel. Passage was performed every week with a 1:5–1:8 split ratio.

All in vitro data are expressed as mean ± SD and analyzed using the Student t test. For human data analyses, the continuous values are expressed as mean ± SDs and analyzed using the Student t test. The categorical variables are analyzed using χ² or Fisher exact test. The correlation between the mutation of KMT2C gene and its protein expression was evaluated using the Phi coefficient. The Kaplan–Meier method was used to construct survival curves, and the log-rank test was used to detect differences between curves, and P < 0.05 was considered significant. Statistical analyses were performed with SPSS Version 12.0 software (SPSS Inc.).

We performed whole-exome sequencing on DNA from fresh-frozen tumors and matched normal blood leukocytes from 27 patients with diffuse-type GA who underwent potentially curative surgical resection. We identified 9,736 somatic point mutations. The number of somatic mutations per tumor varied greatly, with a median of 351 and a range of 120 to 8,274 (Supplementary Fig. S1A). Diffuse-type GAs are generally genomically stable (7), and 25 of 27 samples had less than 2,000 total mutations. Two samples had over 5,000 mutations, and these two outliers were considered hypermutated compared with the rest of the group. In examining the spectrum of base mutations, we observed high C>T transition rates (Supplementary Fig. S1B and S1C).

We generated a list of the commonly mutated genes, and KMT2C was the most frequently mutated gene, with this gene being mutated in 11 of 27, or 41%, of tumors (Fig. 1A). All KMT2C mutations were verified by Amplicon sequencing or conventional Sanger sequencing (Supplementary Tables S1 and S2). KMT2C belongs to the KMT2 family of proteins which methylate lysine 4 on the histone H3 tail at regulatory regions in the genome and thus modulate chromatin structure and DNA accessibility (31). We identified a total of 17 KMT2C mutations, consisting of 14 missense, two nonsense, and one splice site mutations in 11 tumors (Fig. 1B). Three tumors had two KMT2C mutations, and one tumor had three KMT2C mutations. The next most commonly mutated genes were CDH1 (33%) and TP53 (22%), which have previously been reported to be frequently mutated in diffuse-type GA (7).

Figure 1B gives the locations and types of KMT2C mutations found in our diffuse-type GA samples. The majority of mutations were in the N-terminal region of the gene. About 65% of the mutations (11/17) were localized in the plant homeodomain (PHD domain) zinc fingers of the N-terminal region, a highly conserved region (32). We used the functional analysis through Hidden Markov Models (FATHMM) tool to predict the functional outcomes of these single-nucleotide mutations (33, 34). Based on the model, most missense mutations including R284Q, G315S, I323V, D348N, R380L, N393K, and P4310L were pathogenic (score above 0.80). Figure 1C shows the predicted three-dimensional model of the PHD domain zinc fingers (PDB ID: 2YSM and 5ERC) generated using PyMol (DeLano Scientific). Homology modeling was made using the SWISS model (https://swissmodel.expasy.org). Most mutations were located in a residue associated with zinc coordination, which suggests that defects caused by the mutations may affect substrate recognition. SNP array analysis in 24 of the 27 diffuse-type GA tumors did not identify any additional amplifications or deletions (Supplementary Fig. S2; Supplementary Table S3). Figure 1D shows the frequency of KMT2C alterations compared with other genomic analyses of GA. Our study and TCGA study included mutation analysis and somatic copy-number aberration (SCNA) analysis, whereas two other studies just included mutation analysis. In other studies, mutations and SCNAs in KMT2C have ranged from 7% to 17% (7, 9, 10).

Given the KMT2C mutation rate found in this study is higher than in other studies, we designed 12 primer sets to detect the 12 KMT2C mutations identified by whole-exome sequencing in the 27 diffuse-type GAs (Supplementary Table S2). Sanger sequencing using these primer sets was performed on 15 additional diffuse-type GAs from patients undergoing potentially curative resection. This analysis identified KMT2C mutations in five samples (33%). Sanger sequencing is not as sensitive in detecting mutations. Thus, this additional analysis supports the high KMT2C mutation rate found in this study.

We compared clinical and pathologic factors in our 27 patients according to whether or not their tumor had a KMT2C mutation (Table 1). There were no significant differences in gender, age, or race. There was a trend toward more advanced stage in those patients whose tumor had a KMT2C mutation. The incidence of H. pylori infection did not differ between the two groups. Interestingly, the presence of atrophic gastritis or intestinal metaplasia was significantly more common in patients with KMT2C mutations (82% vs. 31%, P = 0.001), suggesting that chronic inflammation may be an etiologic factor for KMT2C mutation.

Patients with KMT2C mutations in their tumors tended to have worse recurrence-free survival than those with KMT2C wild-type tumors (Fig. 2A). Of the 27 tumors examined, additional tumor tissue for immunohistochemical analysis was available in 11 samples. Following immunohistochemical staining for KMT2C protein, all KMT2C wild-type tumors demonstrated protein staining in normal mucosal cells and in tumor cells. In contrast, two of three KMT2C-mutated tumors had little or no protein staining in tumor cells (Fig. 2B). Thus, the KMT2C mutation status correlated well with the expression of KMT2C protein (phi coefficient = −0.770, P = 0.011). The 15 new samples of diffuse-type GA described above were also stained for KMT2C protein expression. Four of five tumors with KMT2C mutation were found to have low expression of KMT2C protein, and eight of ten tumors without KMT2C mutation were found to have high KMT2C protein expression. The combined data from all 26 tumors with mutation and protein expression analysis are shown in Fig. 2B. The phi correlation coefficient between KMT2C mutation and KMT2C protein expression was –0.639 (P = 0.001).

To determine if immunohistochemical expression of KMT2C protein was a prognostic factor for survival, we examined a TMA of 135 diffuse-type GAs from patients undergoing potentially curative surgical resection at Seoul National University Bundang Hospital. Clinical and pathologic characteristics of these combined patients are shown in Supplementary Table S4. Patients with high expression of KMT2C protein in their tumors had less lymphatic, vascular, and neural invasion and earlier T status, N status, and tumor–node–metastasis (TNM) stage. Patients with high KMT2C expression in their diffuse-type tumors also had better recurrence-free survival (Fig. 2C) and overall survival (Fig. 2D) than those with low KMT2C expression. When we examined a separate cohort of patients with intestinal-type tumors (Supplementary Table S4), there was no difference in survival based on high versus low KMT2C expression (Fig. 2E). Thus, KMT2C protein expression inversely correlates with the stage of the tumor and survival in diffuse-type GA but not in intestinal-type GA.

We next examined the functional consequences of KMT2C loss in nontransformed gastric epithelial cells and KMT2C re-expression in diffuse-GA cells. We used HFE-145 cells, which are immortalized human, nonneoplastic gastric epithelial cells, and two diffuse GA cell lines, MKN-45 and SNU-668. By Western blot analysis, HFE-145 cells had significant KMT2C protein expression, whereas MKN-45 and SNU-668 cells had no KMT2C expression (Fig. 3A). Sanger sequencing of the two GA cell lines identified two KMT2C mutations (S3660L and S730Y) in SNU668 cells but no KMT2C mutation in MKN-45 cells. Thus, loss of KMT2C protein expression in MKN-45 cells may be due to an uncommon mutation or promoter methylation.

HFE-145 cells were transduced with KMT2C shRNA or a scrambled control shRNA, and KMT2C knockdown was confirmed by Western blot analysis (Fig. 3B). HFE-145 cells with KMT2C knockdown had moderately increased proliferation in vitro compared with control cells (Fig. 3C). When we examined proteins involved in epithelial-to-mesenchymal transition (EMT), HFE-145 cells with KMT2C knockdown had significantly decreased expression of the epithelial marker E-cadherin and increased expression of the mesenchymal marker N-cadherin along with the EMT transcription factor Slug by immunofluorescence (Fig. 3D) and by Western blot analysis (Fig. 3E). Other EMT transcription factors including Snail and Zeb1 were not significantly changed in HFE-145 cells following KMT2C knockdown. HFE-145 cells also demonstrated a transition from an epithelioid shape to a more spindle cell morphology (Fig. 3F). HFE-145 cells transduced with KMT2C shRNA had 47- to 88-fold increased migration and invasion (Fig. 3G), and had 25-fold more colony formation in soft agar (Fig. 3H) compared with control HFE145 cells. Thus, loss of KMT2C expression in gastric epithelial cells promotes EMT.

Next, KMT2C was overexpressed in MKN-45 and SNU-668 cells using lentiviral transduction. MKN-45 and SNU-668 cells with KMT2C overexpression had mildly decreased proliferation in vitro compared with control cells (Fig. 4A and B). Compared with control cells, MKN-45 and SNU-668 cells with KMT2C overexpression also had significantly increased the expression of E-cadherin and decreased expression of N-cadherin and Slug (Fig. 4C). When cell migration and invasion were assessed, KMT2C overexpression reduced cell migration by 52% to 60% and invasion by 50% to 74% compared with control (Fig. 4D). KMT2C overexpression also reduced the ability of MKN-45 and SNU-668 cells to form colonies in soft agar (Fig. 4E). Thus, overexpression of KMT2C in diffuse-type GA cells lines reverses EMT.

EMT can lead to the acquisition of cancer stem cell (CSC) phenotypes (35). We thus examined spheroid formation and expression of the gastric CSC marker, CD44, in HFE-145 cells following manipulation of KMT2C levels. Compared with control cells, HFE-145 cells with KMT2C knockdown had 2-fold more spheroid formation (Fig. 5A). By fluorescence-activated cell sorting (FACS) analysis, CD44 expression increased from 0.1% in HFE145 monolayers cells to 4.6% in HFE spheroid cells and 7.0% in HFE spheroid cells with KMT2C knockdown (Fig. 5B). By Western blot analysis, HFE-145 cells with KMT2C knockdown not only had increased CD44 expression but also had increased expression of self-renewal proteins including Sox2 by Western blot analysis (Fig. 5C) and by immunofluorescence (Supplementary Fig. S3). Following overexpression of KMT2C in MKN-45 and SNU-668 cells, spheroid formation was dramatically reduced in both cell lines (Fig. 5D) along with the expression of the CD44 and Sox2 by Western blot analysis (Fig. 5E) and by immunofluorescence (Fig. 5F).

Numerous studies have demonstrated that CSCs are more resistant to chemotherapy (27). We examined the effects of KMT2C expression on sensitivity to chemotherapy. Modest doses of 5-FU or cisplatin decreased cell proliferation by only 10% to 14% in control MKN-45 and SNU-668 cells, but KMT2C overexpression combined with 5-FU or cisplatin led to reductions in cell proliferation ranging from 40% to 45% (Supplementary Fig. S4A). We also examined the effects of chemotherapy and KMT2C overexpression on DNA damage using γH2AX staining and on apoptosis using Annexin V staining. KMT2C overexpression combined with chemotherapy led to dramatic increases in both DNA damage (Supplementary Fig. S4B) and apoptosis (Supplementary Fig. S4C) compared with chemotherapy alone. Thus, KMT2C loss promotes CSC phenotypes including spheroid formation and chemotherapy resistance.

We recently described a genetically engineered mouse model (GEMM) in which mice develop diffuse- and intestinal-type GAs which metastasize to lymph nodes, liver, and lung (28). In this GEMM, there is 100% penetrance, and median survival is 76 days. Tumor-derived organoids are an in vitro model that can recapitulate the pathophysiology of the original tumors along with preserving cellular heterogeneity and self-renewal capacity (36). We developed two organoid cell cultures from diffuse-type primary tumors in our GEMM, labeled ACKPY 3077 and AKCPY 3944, as described in the Materials and Methods. KMT2C is known as mixed lineage leukemia 3 or MLL3 in mice. We examined MLL3 expression by Western blot analysis and found loss of MLL3 in both ACKPY3077 and ACKPY3944 organoids (Supplementary Fig. S5A). Organoid cells were subjected to CD44 FACS to isolate for CD44(+) gastric CSCs, and CD44(+) cells were stably transduced with a lentiviral vector for MLL3 expression or a control lentiviral vector. Stably transduced cells were placed back into organoid cell culture (Supplementary Fig. S5B). In both ACKPY3077 and ACKPY3944 organoid cultures, MLL3 transduction led to decreased organoid size (Fig. 6A) and decreased expression of CD44 and EMT-related proteins N-cadherin and Snail (Fig. 6B). Furthermore, MLL3 expression in CD44(+) gastric cancer organoids led to a 73% to 75% decrease in migration and 65% to 67% decrease in invasion (Supplementary Fig. S5C). MLL3 transduction also resulted in decreased expression of CD44 expression and self-renewal proteins (Fig. 6C), decreased ability to form spheroids (Fig. 6D), and decreased ability to form colonies in soft agar (Supplementary Fig. S5D). When ACPKY3077 organoids were grown as flank xenograft in immunodeficient mice, MLL3 transduction decreased tumor growth by 50% (Fig. 6D; Supplementary Fig. S5E). Flank tumors from control ACPKY3077 organoids were highly infiltrative into adjacent normal tissue, whereas flank tumors from ACKPY3077 organoids with MLL3 expression had a well-defined border between the tumor periphery and adjacent normal tissue (Fig. 6E). Thus, MLL3 restoration in gastric cancer organoids reverses EMT and inhibits CSC properties.

In this study, we performed whole-exome sequencing and SNP array analysis on diffuse-type GAs from 27 patients undergoing potentially curative surgical resection at a single U.S. tertiary referral center. Surprisingly, we found in this cohort that KMT2C is the most frequently mutated gene, and that loss of KMT2C protein expression correlates with more advanced tumor stage and worse survival. Knockdown of KMT2C in a nontransformed gastric epithelial cell line resulted in EMT and acquisition of CSC phenotypes, whereas overexpression of KMT2C in two diffuse-type GA cell reversed EMT and blocked CSC phenotypes. Thus, KMT2C mutation appears to be a significant driver of diffuse-type GA progression.

KMT2C is a member of the KMT2 family of proteins which includes KMT2A, MKT2B, KMT2D, KMY2F, and KMT2G (31). This protein family evolved in unicellular eukaryotes and is highly conserved. KMT2 proteins function to methylate histone H3 on lysine 4 to promote genome accessibility and transcription. The different KMT2 proteins have distinct protein-binding partners, and chromatin binding studies demonstrate that KMT2C is highly enriched at enhancer sequences.

KMT2C mutations were known to occur in a subset of leukemias (11), and the targeted inactivation of KMT2C in mice leads to epithelial tumor formation (12). More recent genomic studies have identified KMT2 family mutations as some of the most frequent in human solid tumors including cancers of the lung, colon, and breast (11, 31, 37, 38). For gastric cancer, prior genomic studies found KMT2C mutations in GA in 7% to 17% of tumors (7, 9, 10), whereas our study found mutations in 41%. A genomic study on GA reported that frequent mutations in chromatin remodeling genes including ARID1A, KMT2C, or KMT2D occurred in 47% of all gastric cancers (39). The differences between studies may be explained by differences in the patient cohorts examined and because only diffuse-type GAs were examined in this study.

The majority of previously described mutations in KMT2C were frameshift and nonsense mutations often affecting conserved protein domains including PHD and SET domains (31). In this study, 82% of mutations in KMT2C gene were missense mutations, and about two thirds of mutations were found in the PHD domains. The PHD domain is approximately 50 to 80 amino acids in length, is found in more than 100 human proteins, and is involved in chromatin-mediated gene regulation (40). The PHD domain contains regions coordinating zinc ions, and the residues involved in coordinating the zinc ions are highly conserved among species (41). The zinc ion is essential for the stabilization of the local structure required for DNA binding. The disruption of zinc ion coordination caused by missense mutation may potentially result in deregulation of corresponding proteins. As shown in Fig. 2C, many KMT2C mutations in our study were located around zinc-binding site (R284Q, G315S, D348N, C391, and N393K) of the PHD domain zinc fingers. For the R380L mutation, a hydrophilic amino acid arginine changed to a hydrophobic amino acid leucine which may affect protein structure such as folding. Therefore, many of the mutations at the PHD domain of KMT2C found in this study would likely lead to the defect of substrate recognition, and thus loss of function of KMT2C methyltransferase.

H. pylori can act as an early initiating agent in carcinogenesis (27). Generally at the time of diagnosis of GA, the intragastric environment is inhospitable to H. pylori often resulting in disappearance of infection (42, 43). In our data, the KMT2C mutation rate did not differ according to H. pylori infection status. However, the proportion of atrophic gastritis or intestinal metaplasia (suggesting previous long-standing past infection with H. pylori) was much higher in the KMT2C mutation group than in wild-type group (81.8% vs. 31.3%, P = 0.001). Although diffuse-type GAs are not typically thought to be associated with atrophic gastritis or intestinal metaplasia (44), several studies reported that intestinal metaplasia is more common in the gastric mucosa of patients with diffuse-type GA than that of controls without GA (45–47). Although H. pylori is less related to diffuse-type GA than intestinal-type GA, it is also accepted as a causal factor for diffuse-type GA (48). Taken together, our data suggest that chronic H. pylori infection of gastric mucosa promotes KMT2C mutation, which steers gastric tumorigenesis toward the diffuse type, possibly through a mechanism other than the atrophic gastritis and/or intestinal metaplasia pathway, which traditionally leads to intestinal-type tumors. Further studies can shed light on this interesting topic.

There have been few studies correlating KMT2C mutation or protein expression with survival in patients with GA. In one study in 90 patients with GA from China, low-protein expression of KMT2C correlated worse survival (49). In our study, patients with KMT2C mutation tended to have lower recurrence-free survival than patients without the mutation (Fig. 2A). We found a strong inverse correlation between KMT2C mutation and KMT2C protein expression. After examining KMT2C protein expression in 146 patients with diffuse-type GA undergoing surgical resection, we found that decreased KMT2C protein expression correlated with more advanced tumor stage and worse overall survival. Although KMT2C expression was not an independent prognostic factor, our results support the notion that KMT2C loss increases EMT and promotes tumor progression. Further studies are needed to determine whether KMT2C loss may correlate with chemotherapy resistance.

The EMT program is a naturally occurring transdifferentiation program that regulates changes in cell states between epithelial and mesenchymal (50). This process is vital for cancer cells in order to proliferate and metastasize. The link between EMT and CSCs has been examined in numerous studies, but the recently uncovered link between the passage through EMT and the acquisition of stem-like properties suggests that EMT may be a mechanism for generating CSCs (51). In fact, EMT in cancer cells may be transient, with epithelial tumor cells adopting various mesenchymal states depending on the tumor microenvironment (52). In this study, we found that KMT2C plays a significant role in regulating both EMT and CSC phenotypes.

The ability to grow cancers as organoids allows us to examine heterogeneous tumors in vitro (53). The tumor organoids can be manipulated under controlled conditions in ways that are not possible even in GEMMs. We generated gastric cancer organoids from diffuse-type gastric cancers that developed in a GEMM (28). These organoids at baseline had no KMT2C expression, and transduction of the murine version of KMT2C, MLL3, into the CD44(+) subset of these organoid cells led to EMT and acquisition of CSC phenotypes, thus confirming our findings in gastric epithelial and human gastric cancer cell lines.

In summary, our study demonstrates that KMT2C mutations likely play a key role in the tumorigenesis and progression of diffuse-type GA. Loss-of-function mutations in KMT2C are common in diffuse-type GA, and decreased protein expression of KMT2C is associated with more advanced stage and worse survival. Knockdown of KMT2C in nontransformed gastric epithelial cells promotes EMT and acquisition of CSC phenotypes, and re-expression of KMT2C in type GA cells reverses EMT and attenuates CSC phenotypes. Future studies should focus on how KMT2C mutations alter gene expression and ways to therapeutically target these mutations.

No potential conflicts of interest were disclosed.

Conception and design: C. Yoon, J.-x. Lin, S.S. Yoon

Development of methodology: C. Yoon, S.S. Yoon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.H. Lee, J.-x. Lin, M.-C. Kook, D.J. Park, H. Ashktorab, D.T. Smoot, S.S. Yoon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.-J. Cho, C. Yoon, K.K. Chang, Y.-H. Kim, B.A. Aksoy, D.J. Park, N. Schultz, S.S. Yoon

Writing, review, and/or revision of the manuscript: S.-J. Cho, K.K. Chang, S.S. Yoon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.H. Lee, K.K. Chang, S.S. Yoon

Study supervision: S.S. Yoon

Authors thank Dr. Dongwan Hong for bioinformatics support, Dr. Byung Il Lee for advising on protein structure, and Dr. Hyonchol Jang for critical review of the article.

This study was funded by NIH grants 1R01 CA158301-01 and P30 CA008748, grant 2016R1A2B1010377 from National Research Foundation, Korea, and grant 1610160-2 from National Cancer Center, Korea.

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.