You are here
Sporadic Early-Onset Diffuse Gastric Cancers Have High Frequency of Somatic CDH1 Alterations, but Low Frequency of Somatic RHOA Mutations Compared With Late-Onset Cancers
Gastroenterology, August 2017, Volume 153, Issue 2, Pages 536-549.e26
Editorial comment by Dr Elizabeth Smyth:
Non-familial early onset gastric cancer is often associated with diffuse histology, female gender and poor prognosis. In this Gastroenterology manuscript, Cho and co-authors explore the molecular profile of non-familial young onset diffuse gastric cancer and compare it to later onset diffuse gastric cancer in a large Korean dataset. They find that somatic mutations in CDH1 or TGFBR1 are more common in early onset diffuse gastric cancer whereas somatic mutations in RHOA are more common in patients with later onset diffuse gastric cancer. Both CDH1 alterations and RHOA were associated with increased nuclear β-catenin immunostaining, suggesting WNT pathway activation. Somatic mutations in CDH1 were associated with worse survival; this may account for the poor prognosis of patients with early onset diffuse gastric cancer.
This work is the largest and most comprehensive analysis of early onset and diffuse gastric cancer to date, and suggests a rationale for the aggressive clinical behaviour of early onset diffuse gastric cancer. Although direct targeting of dysregulated CDH1 or RHOA signalling is not yet possible, CDH1 or RHOA dysregulated tumours may have other active pathways e.g. WNT which are suitable for therapeutic intervention. As the detailed genomic and transcriptomic characterisation of gastric cancer continues to demonstrate prognostically important molecularly defined subgroups of patients, clinical trial design incorporating biomarker selection or enrichment for these populations will be essential in order to develop more effective treatments.
Background and Context
Genomic alterations associated with unique clinicopathological characteristics of early-onset gastric cancers are mostly unknown. Previous studies on this population were focused on familial cases and the germline CDH1 mutation.
Sporadic early-onset diffuse gastric cancers harbored less frequent RHOA mutations and more frequent CDH1 and TGFBR1 mutations than late-onset cancers.
This study did not identify the specific mutagens that cause CDH1 and TGFRB1 mutations in young women.
This study elucidates the unique mutation profiles of early-onset gastric cancer and their clinical implications.
Materials and Methods
We conducted WES and RNA sequencing analyses with 80 resected tumors and matched germline samples that were collected from EO (≤45 years) patients with DGC in Korea (EODGC-WES), using HiSeq platforms (Illumina, San Diego, CA). The WES and RNAseq data of EODGC-WES were analyzed by the same centers at which The Cancer Genome Atlas (TCGA) stomach cancer samples were analyzed. 6 To achieve a statistical power adequate for comparison analyses, we expanded the EODGC-WES cohort by recruiting 29 additional EO Korean DGC samples (EODGC, n = 109; [CR] ). EODGC tumors were compared with a control set of diffuse-type stomach cancers that were collected from 115 Korean patients who were ≥46 years at the time of diagnosis (late-onset [LO] DGC). We compared the SNP6.0 array (Affymetrix, Santa Clara, CA) and targeted DNA sequencing (Life Technologies, Carlsbad, CA) data between EODGCs and LODGCs ( Figure 1 A ). The EODGCs and LODGCs were similar with regard to the proportion of metastatic disease at diagnosis and tumor purity as estimated by the histopathologic evaluation of top slides. EODGC-WES tumors were also compared for WES data with 61 LO (LO, ≥46 years) diffuse tumors that originated in TCGA study (LODGC-TCGA). In LODGC-TCGA, 70% of the patients were Caucasians, including 26% females (median age, 63 years; range, 46−82 years). The CDH1 alteration was defined by one of the following somatic alterations in the CDH1 gene encoding E-cadherin: RNA splicing aberrations, homozygous deletions, promoter hypermethylation, single nucleotide variations (SNVs), and indels. All authors had access to the study data, and reviewed and approved the final manuscript. Methods are available in the Supplementary Material .
Whole Exome Sequencing Analysis of Early-Onset Diffuse Gastric Cancers
We performed WES and RNA sequencing analyses of a total of 80 DGCs resected from Korean patients ≤45 years of age (EODGC-WES) ( Figure 1 A ). The EODGC-WES cohort was 58.7% female in composition, with a median age of 38 years. We sequenced the whole exome of EODGC-WES at a mean coverage of 93.6-fold (range, 90.6−95.2).
According to ReCapSeg and ISAR-GISTIC analyses of the WES data, recurrent focal amplification of 10q26.13 at the locus for FGFR2 was the clearly dominant peak within EODGC-WES ( q = 1 × 10 −9 , Supplementary Figure 1 A ). Chromosomal locus 3p21.1 (g.chr3:48,369,660-55,002,466) was the most significant deletion in EODGC-WES ( q = .0003, Figure 1 B ). Among the 161 genes in this locus, we suggest that the deubiquitinase BAP1 may be a novel gastric tumor suppressor gene candidate, because BAP1 was 1 of 6 genes at this locus that were most significantly underexpressed in 3p21.1-deleted EODGC-WES tumors ( Figure 1 C and Supplementary Table 8 ) and BAP1 mutation was the 16th most recurrent mutation in EODGC-WES ( Supplementary Table 9 ). All 4 of the BAP1 somatic mutations found in EODGC-WES were associated with loss of heterozygosity, and BAP1 overexpression suppressed the growth of DGC cells ( Figure 1 D ).
MuTect and Indelocator analyses of the WES data of EODGC-WES identified 6196 nonsynonymous somatic mutations (including 5766 point mutations and 430 indels). Mean and median somatic mutation rates were 1.4 and 1.2 mutations/Mb, respectively (range, 0−7.7: Figure 1 E ). According to MutSig2CV analysis, somatic mutations in CDH1 , TP53, ARID1A, KRAS, PIK3CA, ERBB3, TGFBR1, FBXW7, RHOA, and MAP2K1 were significantly recurrent in EODGC-WES ( Figure 1 E ). Importantly, TGFBR1 mutations were female-specific in EODGC-WES ( P = .014, χ 2 ; Supplementary Figure 3 ). Consistent with their tumor suppressor roles, both TGFBR1 and FBXW7 possessed nonsense mutations ( Supplementary Figure 4 A ). Five of the 7 ERBB3 mutations detected have been reported in other cancers ( Figure 1 F ). While MAP2K1 (encoding MEK1) was mutated in 3 EODGC-WES cases, 2 mutations were the oncogenic C121S mutation ( Figure 1 G ). 11
Somatic CDH1 Mutations Were More Frequent in Early-Onset Diffuse Gastric Cancers Than in Late-Onset Diffuse Gastric Cancers
To compare EODGC and LODGC genomes with a higher statistical power, we expanded our EODGC-WES cohort by recruiting 29 additional EODGC samples from Korean patients ( EODGC ; 54.1% female, median age, 38 years; Figure 1 A ). We then compared SNP6.0 array−based copy number and targeted sequencing data between EODGC and a control set of DGCs that were obtained from 115 Korean patients ≥46 years of age ( LODGC ; 29.6% female, median age, 67 years). In contrast to previous studies suggesting unique copy number profiles of GCs in young patients, 12 our study revealed that copy number profiles were similar between EODGCs and LODGCs ( Supplementary Figure 5 C and 5 D ). Focal amplification of FGFR2 was the most significant in both EODGCs and LODGCs, and focal amplifications in ERBB2, MYC, CCNE1, CCND1, and MET were recurrent in both cohorts ( Figure 2 A ). Consistent with these results, when we classified EODGC tumors according to a molecular classification system proposed by TCGA marker paper (TCGA subgroup), 6 the proportion of the chromosomal instability (CIN) subgroup was not significantly different between EODGCs and LODGCs ( Figure 2 B and Supplementary Figure 5 D ).
Using targeted sequencing analysis, we then compared somatic mutation rates of 10 SNVs, which were significant mutations according to the WES data of EODGC-WES ( Figure 1 E ), between EOGDCs and LODGCs. According to targeted sequencing analysis, EODGC presented significantly higher somatic mutation rates of CDH1 (42.2% vs 17.4% in LODGC) and TGFBR1 (7.3% vs 0.9% in LODGC) than did LODGC ( P < .001 and P = .014 for CDH1 and TGFBR1 , respectively, χ 2 ; Figures 2 C and 3 A−C ). In contrast, somatic RHOA mutation was significantly less frequent in EODGC than in LODGC (9.2% vs 19.1%; P = .033, χ 2 ). Thus, a pattern of frequent CDH1 and TGFBR1 mutations and less common RHOA mutation was identified as the characteristic somatic alteration for EODGC. Somatic TP53 mutations tended to be less frequent in EODGC than in LODGC (33.0% vs 43.5%; P = .11, χ 2 ).
This study evaluates somatic CDH1 mutation rates in relation to patient age in a large-scale setting, and resulted in the identification of a striking difference in CDH1 mutation rates between EODGCs and LODGCs. We also observed that the somatic CDH1 mutation rate was higher in females than in males within EODGC (52.5% [31 of 59] vs 30.0% [15 of 50]; P = .018, χ 2 ). Some CDH1 splicing aberrations were evidenced at the RNA level, but not at the DNA level ( [CR] ). When splicing aberrations at the RNA level, homozygous deletions, promoter hypermethylation, as well as SNVs and indels, were also considered in determining the overall CDH1 alteration rate, EODGC demonstrated a higher CDH1 alteration rate than did LODGC (53.2% [58 of 109] vs 32.2% [37 of 115]; P = .0009, χ 2 ). In addition, EODGC females had higher CDH1 alteration rate than did EODGC males (67.8% [40 of 59] vs 36.0% [18 of 50]; P = .0009, χ 2 ; [CR] ). Thus, we propose that high prevalence of CDH1 alterations may be related to the predominance of females in this population. Notably, patients with pathogenic germline mutations in CDH1, TP53, and ATM were excluded from the onset of this analysis ( [CR] ), given our focus on somatic alterations in sporadic EODGC.
We further investigated CDH1 alterations using transcriptome analyses, which revealed prominent CDH1 splicing aberrations impacting the expression of exons 4, 5, 8, and 9 ( Supplementary Figure 6 A and 6 B ). We identified unexpectedly frequent novel exon truncations and intron retentions in EODGC resulting from CDH1 splicing aberrations impacting the EC1 domain (exons 4 and 5), and this phenomenon was significantly more prominent in EODGC than in LODGC (12.8% [14 of 109] vs 3.5% [4 of 115]; P = .01, Figure 3 A ). In 5.5% of EODGC tumors (n = 6), c.chr16:531+2T>C/A and c.chr16: 531+1G>A mutations in the exon 4-intron 4 junction resulted in intron retentions (with premature termination codons) and 63-bp in-frame deletion transcripts due to the creation of a cryptic splice donor sites ( [CR] and Figure 4 B ). We also identified a 42-bp (exon 5) in-frame deletion transcript in 6.4% (n = 7) of EODGCs using reverse transcription polymerase chain reaction ( Figure 4 B ), and this fragment was also predicted to occur due to the creation of cryptic splice donor sites.
We then performed E-cadherin immunohistochemistry to evaluate the protein expression pattern of CDH1 splicing aberrations ( Figure 4 A ). A majority (63.6%) of EODGC tumors with EC1 splicing aberrations (n = 11) demonstrated abnormal E-cadherin immunostaining (grade 0 or 1: Figure 4 B ). We also evaluated nuclear β-catenin immunostaining, which represents the activation of WNT/β-catenin signaling. Among 10 genes that were significantly mutated in EODGC-WES, CDH1 alterations most significantly correlated with nuclear β-catenin immunostaining in a combined dataset of EODGCs and LODGCs ( P = .005, Wilcoxon; Figure 4 E ).
We further investigated the functional effects of these “hot-spot” CDH1 splicing aberrations impacting the EC1 domain, which have not been characterized previously. According to the 3-dimensional computational modeling, exon truncations in the EC1 domain were predicted to hinder the homodimerization of E-cadherin ( Figure 4 D ). 13 The introduction of CDH1 mutants with truncated exons led to aberrant E-cadherin immunofluorescence ( Figure 4 C ), which is consistent with the immunohistochemistry data of clinical samples ( Figure 4 B ). These results may suggest the instability of mutant E-cadherin proteins, as described previously for CDH1 missense mutations. 14 The introduction of CDH1 mutants with truncated exons was unable to induce the aggregation phenotype that was observed in wild-type CDH1 -transduced GC cells ( Figure 5 B ). Somatic missense mutations in the EC1 domain were also more prevalent in EODGC, suggesting that functional loss of the EC1 domain of E-cadherin shortens the latency of DGC ( Figure 3 A ).
The Effect of RHOA Inactivation on the Metastatic Phenotype of Diffuse Gastric Cancer Cells Was Not as Prominent as That of E-Cadherin Inactivation
As described, somatic mutations in RHOA and TP53 were less frequent in EODGC than in LODGC ( Figure 2 C ). We were not surprised by the relative paucity of TP53 mutations in EODGC, given the association of TP53 mutations with aging, 15 but we were surprised by the relative paucity of RHOA mutations in EODGC (9.2% [10 of 109] vs 19.1% [22 of 115] in LODGC; Figure 3 B ). In EODGC, R5W/Q substitutions were the most frequent RHOA mutation ( Figure 3 B ).
The role of RHOA mutations in diffuse-type gastric carcinogenesis has not been fully elucidated. 6 8 We discovered a clue to the functional effects of the RHOA mutations on DGC from our analysis of nuclear β-catenin immunostaining ( Figure 4 E and Supplementary Figure 6 E ). RHOA -mutated tumors (n = 30) exhibited more prominent nuclear β-catenin immunostaining than did wild-type tumors for RHOA (n = 164) ( P = .011, Wilcoxon; Figure 4 E and [CR] ). Independent of the CDH1 alteration status, the RHOA mutation status was significantly associated with nuclear β-catenin immunostaining (adjusted P = .01, linear regression; Supplementary Table 11 ). This immunohistochemistry data raised the possibility that RHOA mutations might partially overlap with CDH1 mutations in functional effect on diffuse-type gastric carcinogenesis, given that CDH1 mutations could not suppress the β-catenin reporter activity in DGC cells ( Figure 5 A ). We first determined the RHOA activity of an R5W mutation, the most frequent RHOA mutation in EODGC, to evaluate whether it represents a gain- or loss-of-function mutation. Ectopic expression of the RHOA R5W mutation, compared with the ectopic expression of a wild-type RHOA construct in 293FT cells, led to blunted RHOA activation ( Figure 5 C and 5 D ). Thus, biochemical and transcriptional activities of RHOA were compromised by the R5W mutation. Based on this finding, we suppressed the RHOA activity of DGC cells using small interfering RNAs and an exoenzyme C3 transferase, to produce stimuli that are equivalent to the R5W mutation ( Figure 5 E and Supplementary Figure 7 C ). This suppression significantly increased β-catenin reporter activity in DGC cells ( P = .005, paired t test), especially in 2 of the 3 cell lines tested (MKN-45 and NCC-S1 16 17 ; Figure 5 F and Supplementary Figure 7 F ). RHOA knockdown significantly impaired the aggregation of all DGC cell lines evaluated ( Figure 5 G ), which suggests a partial overlap in functional effect between RHOA and CDH1 inactivating mutations ( Figure 5 A and Figure 5 B ). In contrast, whereas Cdh1 knockdown enhanced the migration activity of wild-type DGC cells for Cdh1 , RHOA inactivation did not significantly affect the migration activity of these cells ( Figure 5 H and Supplementary Figure 7 H ). These results may suggest that E-cadherin plays a more central role in suppressing DGC metastasis than does RHOA.
The Aggressive Clinical Course of Early-Onset Gastric Cancer Can Be Attributed to the High Prevalence of CDH1 Mutations and Relative Paucity of RHOA Mutations
To evaluate the clinical relevance of the somatic alterations we identified, we classified EODGCs according to the TCGA subgroup ( Figure 6 A and Supplementary Figure 5 B ). 6 Seven Epstein-Barr virus (EBV)-positive tumors, 5 microsatellite instability (MSI)-high tumors, 69 genomically stable (GS) tumors, and 28 CIN tumors were present in EODGC. The distribution of TCGA subgroup of EODGC did not differ significantly from that of LODGC ( P = .34, χ 2 ; Figure 2 B ). None of the 26 signet ring cell carcinomas in EODGC belonged to the EBV/MSI subgroup ( P = .04, χ 2 ; Figure 6 A ). The EBV subgroup in EODGC was enriched with the male sex (100% vs 42.2% in non-EBV; P = .003) and the proximal tumor location (71.4% vs 24.5% in non-EBV; P = .007). None of the 58 EODGC tumors with CDH1 alterations belonged to the EBV/MSI subgroup ( P < .001, χ 2 ). All 8 TGFBR1 mutations were found in the GS subgroup ( P = .025, χ 2 ).
Then we compared patient survival according to the TCGA subgroup and somatic mutation status. CIN was associated with the highest hazard ratio (HR) and the most adverse prognostic factor in EODGC (HR, 3.7; 95% confidence interval [CI], 1.8−7.7) vs non-CIN; Figure 6 A ). At a median follow-up of 33.7 months, median overall survival times were not reached in EBV/MSI and GS, while the CIN subgroup had the median survival of 28.2 months (95% CI, 14.6−not reached; P = .0002, log-rank). The CDH1 alteration (SNVs, indels, splicing aberrations, homozygous deletions, and promoter hypermethylation) was the second most adverse prognostic factor (HR, 3.4; 95% CI, 1.5−7.7; P = .002, log-rank), and tended to be enriched in patients with distant metastasis (66.7% [16 of 24] vs 49.4% [42 of 85] in those without metastasis; P = .13, χ 2 ; Figure 6 A ). In contrast, RHOA and TGFBR1 mutations were not associated with prognosis (HR, 0.7; 95% CI, 0.2−3.0 and HR, 1.7; 95% CI, 0.5−5.7, respectively; Supplementary Figure 8 C and 8 D ). Mutation frequencies of RHOA and TGFBR1 did not differ between patients with distant metastasis and those without metastasis (8.3% vs 9.4% and 4.1% vs 8.2%, respectively, for RHOA and TGFBR1 ). FGFR2 and ERBB2 amplifications and TP53 mutation were also associated with poor prognosis (HR, 3.8; 95% CI, 1.5−9.4; HR, 2.5; 95% CI, 1.0−6.6 and HR, 2.6; 95% CI, 1.2−5.4, respectively; P = .0018, P = .05, and P = .008, respectively, log-rank). These results indicate that EODGC can be classified into 3 prognostic categories. EBV/MSI was the best prognosis group without CDH1 alterations. The GS subgroup, which was enriched with TGFBR1 mutations ( P = .025, χ 2 ), yielded intermediate prognosis. The CIN subgroup, which was enriched with TP53 mutations ( P = 3.5 × 10 −5 , χ 2 ), was associated with the worst prognosis. Prognostic implications of TCGA subgroups were not documented previously by the TCGA marker paper due to its short-term follow-up duration. 6
EOGC is notable for its advanced stage presentation and for its strong enrichment of diffuse histology and female predominance. Given the equivalent proportion of CIN between EODGCs and LODGCs ( Figure 2 B ), we suggest that EOGC’s aggressive clinical course is primarily due to its high frequency of the CDH1 alteration, which represents the second most significant adverse prognostic factor in EODGC. We again noted trends for the higher prevalence of CDH1 mutations and the lower rate of RHOA mutations in younger patients, when we compared the WES data between EODGC-WES and a TCGA dataset of non-hypermutated, diffuse GCs obtained from patients ≥46 years of age (LODGC-TCGA), 70% of whom were Caucasians. 6 When compared with LODGC-TCGA, EODGC-WES showed trends for relatively high somatic mutation rates of CDH1 (35.0% vs 24.6% in LODGC-TCGA; P = .18) and TGFBR1 (8.8% vs 3.3% in LODGC-TCGA; P = .19) and a relatively low somatic mutation rate of RHOA (5.0% vs 13.1% in LODGC-TCGA; P = .09; Supplementary Figure 4 C ), whereas these 2 cohorts did not differ in tumor purity and ploidy as estimated by ABSOLUTE ( Supplementary Figure 4 B ). Thus, sporadic EOGC was characterized by high prevalence of somatic CDH1 mutations and relative paucity of somatic RHOA mutations not only when compared to an older Korean population, but also when compared to LO populations of other ethnicities.
BRCA and Young Female Signatures Were Identified by Mutation Signature Analyses
Aneuploidy, which was previously represented by the CIN subgroup ( Figure 6 A ), was the most adverse prognostic factor in EODGC. As shown in Figure 2 B , the CIN proportion of EODGC was comparable to that of LODGC, suggesting an equivalent degree of aneuploidy between the 2 cohorts. Mutation signature analysis revealed that germline mutations in homologous recombination pathway were associated with aneuploidy in EODGC. The homologous recombination defect-related BRCA mutation signature 18 was 1 of 6 mutation signatures that were identified in the WES data of EODGC-WES and LODGC-TCGA (DGC, Figure 7 A ), and its signature fraction positively correlated with the degree of aneuploidy ( R = 0.22; P = .01, Pearson, Figure 7 C ). The 3 tumors with the largest BRCA signature fraction, all of which were from aneuploidy EODGC-WES, presented germline mutations in either PALB2 or RAD51D ( Figure 7 B and [CR] and [CR] ). These data suggest that a small group of EODGC may be attributable to inherited DNA repair defects and these tumors manifest a distinct pattern of somatic alterations from other EODGC patients.
We also asked whether other mutation signatures may be preferentially present in EODGC, most especially in female patients, as such a signature may provide insight into the etiology of more frequent DGC diagnoses in this population. We noted that the signature resembling neuroblastoma-associated COSMIC signature 18 with prevalent C>A/T was significantly enriched in EODGC-WES ( P = .0002, vs LODGC-TCGA, Wilcoxon), especially in EODGC-WES females ( P = .007, vs EODGC-WES males, Wilcoxon; Figure 7 D ). This mutation signature was therefore referred to as the young female (YF) signature. Further inquiry into YF signature demonstrated a significant strand bias of C>A/T mutations toward the transcribed strand ( P = .001; Supplementary Figure 9 C ). The fraction of the YF signature was higher in CDH1- and TGFBR1- mutated DGCs than in wild-type DGCs ( P = .016 and P = .018, Wilcoxon; Figure 7 E ). Therefore, the unidentified mutagenic process underlying this signature appears to be enriched in the YFs who develop mutations that are common in EODGC.
The Predominance of Diffuse Histology and Female Sex in Early-Onset Gastric Cancer May Be Related to Frequent Somatic Mutations in CDH1 and TGFBR1
In addition to advanced-stage presentation, EOGC is also notable for its strong enrichment of diffuse histology and female predominance. As described, somatic CDH1 and TGFBR1 mutation rates were higher in EODGCs from women than in EODGC from men ( P = .031 and P = .007, χ 2 ; Figure 6 A ). We also observed that CDH1 and TGFBR1 mutation rates were higher in diffuse tumors than in intestinal tumors within TCGA dataset. When the WES data were compared between LODGC-TCGA (n = 61) and a set of non-hypermutated intestinal tumors from TCGA ( IGC : n = 157), somatic CDH1 and TGFBR1 mutations were more frequent in LODGC-TCGA than in IGC, a finding that had not been recognized previously ( P = 8 × 10 −7 and P = .022, respectively; Figure 6 B and Supplementary Figure 11 A ; [CR] ).
This is a comprehensive study of somatic genomic alterations associated with the unique clinicopathologic features of sporadic EODGC. 8 This study is also the largest WES study of DGC to date, and resulted in the identification of previously unrecognized TGFBR1 and BAP1 alterations in DGC ( Supplementary Figure 11 A and 11 C ) and the characterization of hot-spot splice site mutations of CDH1 involving exon truncations and intron retentions. We documented the prominent difference in somatic CDH1 mutation rates between EODGCs and LODGCs in a large sample set and the clustering of CDH1 mutations at the E-cadherin EC1 domain in EODGC. Somatic CDH1 mutations documented in our sporadic EODGC cases provide a higher level of multi-omic information than did previous CDH1 mutation studies on sporadic DGC cases. 19 Most of the earlier works on EOGC had been on familial cases and focused on the germline CDH1 mutation, which is a known risk factor for hereditary DGC. 20 21 22 In contrast, our analysis from the onset excluded EODGC-WES cases with pathogenic germline mutations in CDH1 , TP53 , and ATM1 ( [CR] ). As a result, there were no pathogenic germline CDH1 mutations among the entire 109 EODGC cases of this study, according to our genomic and functional analyses ( [CR] and [CR] ; Supplementary Figure 6 C ). This study provides the largest somatic alteration data of sporadic EOGC.
Beyond identifying distinct somatic alterations in EODGC, we also observed distinct mutagenic processes. These include the findings of 3 tumors with a marked BRCA signature in patients without family history of GC, yet harboring pathogenic germline mutations in the homologous recombination repair pathway. Among our EODGC-WES cases, the BRCA mutation signature fraction was higher in the CIN subgroup (vs non-CIN, P = .008, Wilcoxon; Supplementary Figure 10 A ). But it was not associated with CDH1 alterations or platinum responsiveness after relapse according to our analysis of a limited sample set ( Supplementary Figure 10 C and Supplementary Table 10 ).
Our study found that the BAP1 locus, 3p21.1, was the most frequent deletion in EODGC-WES. Chromosomal locus 3p21.1 is deleted in mesothelioma, malignant melanoma, and cancers of the lung and breast. 23 We also observed that the BAP1 mutation was the 16th most recurrent somatic mutation in EODGC-WES, and all 4 BAP1 somatic mutations were associated with loss of heterozygosity. Germline mutations in BAP1 predispose to melanocytic tumors and mesothelioma, and somatic mutations in BAP1 have been reported in intrahepatic cholangiocarcinoma, 24 malignant melanoma, mesothelioma, and renal cell carcinoma. However, tumor-suppressive role of BAP1 has not been reported in GC. As in NCI-H226 cells, 25 BAP1 expression suppressed the growth of our DGC cells, suggesting that BAP1 is a novel tumor suppressor candidate in DGC.
While neither the tumor content nor the proportion of metastatic disease differed between the compared age groups, our study identified a consistent pattern of frequent somatic CDH1 mutations and less common somatic RHOA mutations in sporadic EODGC, within the Korean population and within populations of other ethnicities. The unique mutation frequency pattern could not be explained by the rarity of MSI in young patients, given the lack of association between RHOA mutations and MSI (13.6% and 15.0% for MSI-low/microsatellite stable and MSI-high, respectively). We then investigated whether this mutation frequency pattern could be explained by a functional overlap between CDH1 and RHOA mutations, both of which are relatively uncommon in solid tumor types other than diffuse gastric and lobular breast cancers. Wang et al 7 have shown that Y42C and L57V RHOA hotspot mutations cause defective RHOA signaling, promoting resistance to anoikis. Consistent with the Wang et al study, our study demonstrated that an R5W mutation reduced RHOA activity, and that the RHOA inactivation led to β-catenin activation in gastric epithelial cells. RHOA inactivation might possibly activate β-catenin signaling partially through destabilizing the E-cadherin/catenin complex, as was observed in colon cancer cells. 26 Although the exact molecular mechanisms in DGC are not clear, functional consequences of RHOA inactivating mutations in DGC may overlap with those of CDH1 alterations regarding β-catenin activation as well as cell−cell adhesion loss, 7 27 which might, to a degree, explain the relative paucity of RHOA mutations in EODGC harboring highly prevalent CDH1 alterations that might offset the necessity of RHOA mutations for diffuse-type gastric carcinogenesis. More importantly, the migration activity of our DGC cells became more prominent after the knockdown of CDH1 , a central player in epithelial−mesenchymal transition, than after the knockdown of RHOA . Consistent with our genomics data that the CDH1 alteration, but not the RHOA mutation, was associated with poor prognosis of EODGC, these results indicate that a pattern of frequent CDH1 alterations and rare RHOA mutations may contribute to EOGC’s high propensity for metastasis. 8
Intriguingly, Kakiuchi et al 5 reported that G17E and Y42C RHOA mutations were biological gain-of-function mutations conferring a growth advantage on OE19 cells. Whereas these G17E and Y42C RHOA mutations were similar in frequency in our LODGC cohort (10.4%) and Kakiuchi et al’s (mostly LO) cohort (10.3%), they were rare in our EODGC (1.8%). A majority of RHOA mutations found in our EODGC cohort were inactivating 28 R5W/Q mutations. Our experimental data of an R5W mutation, which cannot be directly compared with Kakiuchi et al’s data of G17E and Y42C mutations, 5 demonstrated that biochemical and transcriptional activities of RHOA were significantly reduced by an R5W substitution, and that the impaired cell−cell aggregation was the most prominent phenotype of DGC cells after RHOA inactivation. Continued research may be necessary to evaluate whether R5W/Q substitutions might manifest oncogenic functions beyond the loss-of-function effects we described in this report. As of yet, functional and genomic data do not seem to support the feasibility of a therapeutic approach targeting RHOA mutations, which was proposed by Kakiuchi et al, 5 in cases of EODGC.
The strong enrichment of diffuse histology (94% in our cases) and female predominance in EOGC can be attributed, at least partially, to high rates of somatic CDH1 and TGFBR1 mutations, which were shown to be significantly associated with the characteristics mentioned ( Figure 6 A and B ), and might possibly involve common mutagens ( Figure 7 E ). Our study found that somatic TGFBR1 mutations were enriched in EODGC females ( Figure 6 A ) and diffuse histology-specific ( Figure 6 B ). Previous studies in colon cancer cells have shown that transforming growth factor−β promotes cellular adhesions by extracellular matrix remodeling, 29 but the TGFBR1 mutation itself has not yet been linked to diffuse gastric tumors. Further studies could explore whether and how TGFBR1 mutations lead to the development of DGC. Taken together, our study further elucidates the functional implications of somatic mutations that are recurrent in DGC, and the genomic basis of clinicopathologic characteristics of DGC that differ according to age.
The biospecimens for this study were provided by members of National Biobank of Korea (Asan Bio-Resource Center, Keimyung Human Bio-Resource Bank, Biobank of Pusan National University, Biobank of Chonnam National University Hwasun Hospital, Biobank of Chungnam National University Hospital, and Ajou Human Bio-Resource Bank), which is supported by the Ministry of Health and Welfare, by Resource Banks at Dong-A University Medical Center and Kosin University Gospel Hospital and by the National Cancer Center of Korea. Adam J. Bass was supported by the Schottenstein Fund for Gastric Cancer Research and the DeGregorio Family Foundation.
The authors thank all the patients and their families who contributed to this study. The results published here are in part based on data generated by The Cancer Genome Atlas managed by the National Cancer Institute and National Human Genome Research Institute. Information about TCGA can be found at http://cancergenome.nih.gov .
The raw sequencing and SNP6.0 array data associated with this work were deposited to the European Genome-phenome Archive database under accession numbers EGAS00001001711 and EGAS00001001859.
Soo Young Cho, Jun Won Park and Yang Liu contributed equally to this work.
Supplementary Methods and Results
Patients and Tissue Processing
GC and germline (adjacent normal tissue or blood) samples were collected from young (≤45 years) patients with DGC in Korea. Fresh-frozen tumors were obtained from surgical resection performed between 2003 and 2014. Samples were collected from members of National Biobank of Korea (Asan Bio-Resource Center, Keimyung Human Bio-Resource Bank, Biobank of Pusan National University, Biobank of Chonnam National University Hwasun Hospital, Biobank of Chungnam National University Hospital, and Ajou Human Bio-Resource Bank), which is supported by the Ministry of Health and Welfare; Resource Banks at Dong-A University Medical Center and Kosin University Gospel Hospital; and the National Cancer Center of Korea. All tumor samples were collected from patients who had no prior chemotherapy or radiotherapy. This study was approved by the National Cancer Center Institutional Review Board (NCCNCS-120581) and all patients signed Institutional Review Board−approved consent forms. Tumors were staged according to the 7 th edition of the American Joint Committee on Cancer.
A total of 80 resected DGCs with adequate yield of DNA/RNA for genomic analysis were subjected to WES, RNA sequencing, and SNP6.0 array analysis in Korea (EODGC-WES, [CR] ). WES was performed in these 80 resected tumors and matched germline DNA samples (buffy coats [n = 78] or adjacent normal tissues [n = 2]) from the same patients. The WES and RNA sequencing data were then reprocessed and analyzed by the same centers at which the stomach cancer samples for TCGA study 1 were analyzed, thereby facilitating the cross-comparisons of these tumor sets.
To achieve a statistical power adequate for comparison analyses, we expanded EODGC-WES cohort by recruiting 29 additional EO Korean DGC samples, which included biopsy samples that had been too small for inclusion in the WES analyses (EODGC [n = 109], Figure 1 A ). Assuming proportions of TP53 mutations of 0.44 and 0.28 for LODGCs and EODGCs, respectively, a sample size required to detect the difference of TP53 mutations rates between the 2 cohorts was estimated at 109, with 2-sided α- and β-errors of .05 and .2, respectively. Fresh-frozen EODGC tumors were obtained from either surgical resection or endoscopic biopsy. Targeted sequencing and SNP6.0 array analyses were performed for these cases. [CR] shows the clinicopathologic characteristics of the 109 patients in the EODGC cohort, which includes 80 patients in the EODGC-WES cohort and 29 additional Korean EODGC cases. Fourteen (12.8%) patients had family history of GC in a first- or second-degree relatives ( [CR] ). At a median follow-up of 33.7 months, median overall survival of the 109 patients in the EODGC cohort was 46.6 months (95% CI, 33.3−not reached).
Overall, a total of 216 cases (tumor and matched germline samples) were collected for this study from Korean DGC patients ≤45 years old. During the sample qualification process, a total of 107 cases were disqualified for the inadequate tumor (n = 65) or normal samples (n = 7), inadequate RNA/DNA quality (n = 32) or quantity (n = 3), or intestinal/mixed histology (n = 7). After sequencing analyses, 4 patients with pathogenic germline mutations in CDH1 (n = 1), TP53 (n = 2), or ATM (n = 1) were excluded ( [CR] ).
EODGCs were compared with a control set of diffuse-type stomach cancers were collected from 115 Korean patients who were ≥46 years old at the time of diagnosis in the aforementioned hospitals in Korea between 2002 and 2014 (LODGCs) ( [CR] ). LODGCs and EODGCs were similar with regard to the proportion of metastatic disease (American Joint Committee on Cancer stage IV) and tumor purity as estimated by the H&E evaluation of top slides. These 2 cohorts were different in sex distribution ( P = .0002, χ 2 ) ( [CR] ).
EODGC-WES tumors were also compared with LO (≥46 years) diffuse tumors that originated from TCGA study (LODGCs from TCGA [LODGC-TCGA]). 1 In LODGC-TCGA, 70% of patients were Caucasian, including 26% female, and median age of 63 years (range, 46−82 years). We compared the WES data between LODGC-TCGA and EODGC-WES (n = 80). The proportion of metastatic disease at presentation (American Joint Committee on Cancer stage IV) was 14.8%, and did not differ from 16.3% of EODGC-WES ( P = NS, χ 2 ).
Sample collection and processing were basically performed according to the protocol of a central Biospecimen Core Resource of TCGA. 1 After a gross examination, the non-necrotic portions were excised from the tumor specimens by pathologists, and adjacent normal tissues were collected from the luminal side of the gastrectomy specimen at least 2 cm from the tumor border. To isolate genomic DNA and RNA from frozen tumor tissue samples, light microscopic examinations were performed on top slides by pathologists to analyze the tumor-rich area. Then, genomic DNA was extracted from macrodissected frozen tumor tissues using a DNeasy blood and tissue kit (Qiagen, Hilden, Germany). Genomic DNA was also extracted from buffy coats using the same kit. Total RNA was isolated from the same piece of the tumor that was used for DNA isolation and from adjacent normal tissue samples using a mirVana kit (Thermo Fisher Scientific, Waltham, MA). RNA sequencing was performed in 80 tumors and 65 adjacent normal tissue samples that were collected from patients in the EODGC-WES cohort. SNP6.0 array analyses were performed using 109 tumors that were collected from the EODGC cohort, and 88 tumors that were collected from the LODGC cohort. Targeted sequencing was performed for all tumors collected from the EODGC (n = 109) and LODGC cohorts (n = 115) for 10 mutations that were recurrent among 80 EODGC-WES tumors.
Whole Exome Sequencing and Somatic Mutation Calling
For EODGC-WES tumors (n = 80), a total of 1−3 μg of fragmented DNA was prepared to construct libraries using a SureSelect Human All Exon Kit V4+UTR (Agilent Technologies, Inc, Santa Clara, CA) according to the manufacturer’s protocol. Briefly, the qualified genomic DNA sample was randomly fragmented using an ultrasonicator (Covaris, Woburn, MA) followed by adapter ligation, purification, hybridization, and polymerase chain reaction (PCR). Captured libraries were subjected to an Agilent 2100 Bioanalyzer to estimate the quality and then were paired-end sequenced using HiSeq platforms (Illumina, San Diego, CA). Raw image files were processed using HCS1.4.8 for base-calling with the default parameters, and the sequences for each individual sample were generated as 101-bp paired-end reads. We then re-processed and analyzed the WES data at the Broad Institute.
We used MuTect and Indelocator to identify somatic SNVs and indels, respectively. 2 3 SNVs/indels were annotated using Oncotator. 4 A filter for artificial CC>CA mutations caused by sample oxidation (8-oxoguanine) was applied to remove potential CC>CA artifact. ReCapSeg and AllelicCapSeg ( http://www.broadinstitute.org/cancer/cga/acsbeta ) were used to obtain allele-specific copy number profiles, and germline samples were aggregated as a reference panel for ReCapSeg to calculate normalized copy ratios. Given allele-specific copy ratios, the ABSOLUTE algorithm was used to estimate absolute copy numbers and tumor purity. 5 We used MutSig2CV to evaluate the significance of mutated genes, which combines evidence from background mutation rate, clustering of mutations on hotspots, and conservativeness of genes. 6 Mutations with q value <.1 were declared significant. Hypermutated samples (defined as a mutation rate >11.4/Mb) were removed from MutSig2CV analysis. None of the 80 EODGC-WES tumors were identified as hypermutated. We evaluated the degree of aneuploidy by determining the fraction of aneuploid genome in tumors. The fraction of aneuploid genome was defined as the relative length of the genome (to the length of human genome) that has integer copy number estimates not equal to 2 (or 4 for genome-doubled and 6 for genome-tripled cases), or was identified as loss-of-heterozygosity (LOH) regions. The same procedure was performed for TCGA data.
Comparison of the Whole Exome Sequencing Data Between Intestinal and Diffuse Tumors
We created a WES set from all of the non-hypermutated TCGA diffuse tumors and EODGC-WES (n = 147: diffuse tumors). Then we performed MuTect and Indelocator analyses on diffuse tumors, as described. According to the MutSig2CV analysis, TP53, CDH1, ARID1A, KRAS, PIK3CA, ERBB3, TGFBR1 , and RHOA were significant mutations in 147 diffuse tumors. When the rates of these recurrent mutations were compared between 147 diffuse tumors and a set of non-hypermutated intestinal tumors originating from TCGA (n = 157), TP53, CDH1, and TGFBR1 were significantly different in frequency between diffuse and intestinal tumors ( Supplementary Figure 11 A and [CR] ).
Mutation Signature Analysis
Mutation signatures were analyzed from SNVs identified in the WES data of EODGC-WES and LODGC-TCGA, using a Bayesian version of the non-negative matrix factorization method. 7 The mutation signature analysis is processed using the common classification of SNVs, which is based on 6 base substitutions (C>A, C>G, C>T, T>A, T>C, and T>G) with 16 possible combinations of neighboring bases result in 96 possible mutation types. We constructed input data for the mutation signature discovery, which is given as 96 by M mutation matrix ( M = number of sample). The signature fraction, which is the estimated proportion of mutations that are contributed by each signature, was used as a metric for the intensity of a signature. Transcriptional strand bias of a signature was tested using logistic regression between the transcriptional strand of mutations and the corresponding signature fractions. We used the cosine similarity to compare our 3 signatures with 30 reported COSMIC signatures ( http://cancer.sanger.ac.uk/cosmic/signatures ).
In a combined WES dataset of EODGC-WES and LODGC-TCGA, we found 6 distinct mutation signatures present. As shown in Figure 7 A , these 6 mutation signatures include 2 associated MSI (MSI signature and an alternative type of MSI signature [MSI-2 signature]), a BRCA/homologous recombination deficiency signature (BRCA signature), a signature of C>T transitions at CpG’s sites and signatures similar COSMIC signature 17 (deamination (DA) signature), AA>AC-predominant signature seen previously in gastric and esophageal adenocarcinoma (G/E signature), and 1 resembling COSMIC signature 18 ( http://cancer.sanger.ac.uk/cosmic/signatures ; YF signature; Supplementary Figure 9 A ).
In EODGC-WES, the BRCA mutation signature was stronger in CIN (vs non-CIN, P = .008, Wilcoxon; Supplementary Figure 10 A ). In EODGC-WES, the fraction of BRCA mutation signature was not different between wild-type tumors and tumors with CDH1 alteration ( P = NS, Wilcoxon; Supplementary Figure 10 B ).
We could correlate mutation signatures with clinical response of 15 relapsed EODGC-WES cases to platinum (either cisplatin or oxaliplatin)-containing first-line chemotherapy. This analysis included patients who underwent palliative gastrectomy at metastatic stage and subsequently received chemotherapy without gross residual disease. No mutation signatures were significantly associated with progression-free survival after platinum-containing chemotherapy after relapse ( Supplementary Table 10 and Supplementary Figure 10 C ).
We also asked whether other signatures may be preferentially present in patients with EODGC-WES, most especially in female patients, as such a signature may provide insight into the etiology of more frequent DGC diagnoses in this population. We noted that the signature resembling COSMIC signature 18 with prevalent C>A/T was significantly enriched in EODGC-WES ( P = .0002, compared to LODGC-TCGA, Wilcoxon), especially in EODGC-WES females ( P = .007, compared to EO males, Wilcoxon, Figure 7 D ). This mutation signature was therefore referred to as the YF signature. Although the etiology of signature 18 is not known, it is most strongly associated with the pediatric cancer, neuroblastoma. Further inquiry into YF signature demonstrated a significant strand bias of C>A/T mutations toward the transcribed strand ( P = .001; Supplementary Figure 9 C ). Future studies into the etiology of this signature are highly warranted and may identify factors that contribute to formation of EODGC.
ReCapSeg and GISTIC Analyses
To evaluate SCNA significance, copy numbers were first adjusted for tumor purity and ploidy with the ISAR method, 8 and then GISTIC 2.0 9 was used to evaluate significances. A q value of .25 was set for the significance threshold. To eliminate spurious germline CNVs, we also performed the above ISAR-GISTIC on the matched normal samples and, for the next GISTIC run, we excluded those significant regions obtained from the normal samples. The procedures were repeated until there was no significant CNV in normal samples. Significant deletions and amplifications in EODGC-WES are shown in Figure 1 B and Supplementary Figure 1 A , respectively. Significant deletions and amplifications in EODGC-WES are shown in Figure 1 B and Supplementary Figure 1 A , respectively. ReCapSeg and GISTIC analyses of the WES data of diffuse tumors (n = 147) and non-hypermutated TCGA intestinal tumors (n = 157) are shown in Supplementary Figure 11 B .
Targeted DNA Sequencing
Targeted sequencing was performed on samples from EODGC (n = 109) and LODGC (n = 115) cohorts for 10 mutations that were recurrent in the EODGC-WES cohort ( Figure 1 E ). 6 10 Using the Ion AmplilSeq Designer software, PCR primers were designed for all exons (with 5 bp of padding at the ends of ends) with 98.9% coverage. PCR amplicons ranged from 125 to 175 bp in length. A total of 20 ng of genomic DNA with A260/A280 and A260/A230 ratios >1.6 was used for library generation. Fragment libraries were constructed using DNA fragmentation, barcode and adaptor ligation, and library amplification using an Ion DNA Barcoding kit (Thermo Fisher Scientific, Carlsbad, CA), according to the manufacturer’s instructions. The size distribution of the DNA fragments was analyzed using a 2100 Bioanalyzer High Sensitivity Kit (Agilent). Template preparation, emulsion PCR, and ion sphere particle enrichment were performed using an Ion Xpress Template OT2 200 kit (v3: 4488318; Thermo Fisher Scientific), according to the manufacturer’s instructions. Ion sphere particles were loaded onto a P1 chip (version 2) and sequenced using an Ion P1 sequencing 200 kit (v3: 4488315; Thermo Fisher Scientific). Ion Torrent platform-specific pipeline software (Torrent Suite, version 4.4) was used to separate the barcoded reads, generate sequence alignments with the hg19 human genome reference, perform target-region coverage analysis, and filter and remove poor signal reads. The single-nucleotide variants and small insertions/deletions (indels) were compared with those in the germline genomic DNA. Initial variant calling was generated using Torrent Suite with a plug-in program (Variant Caller, version 4.4). The alignment file from the Torrent Suite was then transferred to Ion Reporter version 4.4 to generate a somatic variant file using default parameters. The somatic calls generated from Ion Reporter version 4.4 were further filtered using the following criteria: (1) >50 reads in tumor samples; (2) >5 somatic variant reads; (3) somatic variant allele frequencies >0.05 and >0.1 for SNV and indels, respectively; and (4) minor allele frequency <0.02 in germline samples. A total of 346 somatic mutations (316 SNVs and 30 indels) were identified using targeted sequencing. EODGC (n = 109) and LODGC (n = 115) tumors were not different in coverage means (960× and 959x, respectively; P = .91, t test). Targeted sequencing and RNA sequencing validated 84.5% of somatic mutations that were identified in the WES data of EODGC-WES ( [CR] ).
Germline Mutation Calling
We searched for germline mutations using the WES data for germline samples obtained from EODGC-WES patients. Germline variations in the normal samples were identified using GATK. 11 Information regarding the allele frequency in the Korean population was obtained from the Korean Reference Genome DB ( http://184.108.40.206/KRGDB/menuPages/firstInfo.jsp ). All pathogenic germline mutations were validated using capillary sequencing.
As listed in [CR] , four germline mutations in ATM, CDH1 , and TP53 were identified as pathogenic germline mutations based on the Clinvar database, 12 the evidence as a somatic alteration, 13 and the data in the literature. One of these pathogenic germline mutations, an R335X CDH1 nonsense mutation, was identified as a somatic mutation in an EOGC patient. These 4 cases with pathogenic germline mutations were excluded from this analysis.
Our targeted DNA sequencing analysis of EODGC and LODGC revealed no pathogenic germline mutations, consistent with the lack of strong familial clustering of GC among our EODGC patients. WES and targeted sequencing revealed several other CDH1 germline mutations of unknown significance among EODGC and LODGC cases ( [CR] ). We performed functional analyses of these CDH1 variants to evaluate their oncogenic potentials. CHO aggregation assays indicated that K182N, T340A, and E880K were likely to be benign germline variants ( Supplementary Figure 6 C ).
During our mutation signature analyses, we additionally searched for pathogenic germline mutations involved in DNA repair, which might have been missed by the methods described. We extracted mutations in a list of candidate genes that are involved in DNA repair according to the MSigDB gene sets. 14 Then, we filtered out mutations that are not likely to have significant functional impact on the genes, using a combination of prediction methods (SIFT, Polyphen2, LRT, MutationTaster, and MutationAssessor) that were included in the dbNSFP. 15 Mutations were filtered out when <3 of these methods predict them as functional. Three pathogenic germline mutations involved in DNA repair were identified from this approach ( [CR] ).
Transcriptome libraries were prepared following Illumina’s TruSeq mRNA kit protocol using 1−2 μg of total RNA obtained from 80 tumors and 65 adjacent normal tissue samples obtained from patients in the EODGC-WES cohort. Poly(A)+ RNA was isolated using AMPure XP beads (Beckman Coulter, Brea, CA) and fragmented using an Ambion Fragmentation Reagents kit (Thermo Fisher Scientific). Complementary DNA (cDNA) synthesis, end repair, A-base addition, and ligation of the Illumina-indexed adapters were performed according to Illumina protocols. Libraries were size-selected for 250−300 bp cDNA fragments on a 3% Nusieve 3:1 (Lonza, Basel, Switzerland) agarose gel, recovered using QIAEX II gel extraction reagents (Qiagen), and PCR-amplified using Phusion DNA polymerase (New England Biolabs, Ipswich, MA) for 14 PCR cycles. The amplified libraries were purified using AMPure XP beads. Library quality was measured on an Agilent 2100 Bioanalyzer to determine the product size and concentration. Paired-end libraries were sequenced using an Illumina HiSeq 2000 instrument (2×100 nucleotide read length). Reads that passed the chastity filter of the Illumina BaseCall software were used for subsequent analyses.
RNAseq data were aligned at the British Columbia Genome Science Center. Using BWA aln & sampe (version 0.5.7), 16 we aligned all reads to a human reference genome consisting of hg19/GRCh37-lite and exon−exon junction sequences that were constructed from known transcript models in EnsEMBL, RefSeq, and UCSC genes. Default BWA parameters were used for alignments, with the exception of the -s sampe option, which was included to disable Smith-Waterman rescue of unmapped mates, as this feature was not designed to handle the insert size distribution that occurs in paired-end RNA sequencing data. After BWA alignment, a post-alignment process was performed to reposition the read alignments that spanned across the exon−exon junctions and transform them into large-gapped genomic alignments. 17 The reads per kilobase of exon per million mapped reads values of exons and genes were calculated. GENCODE V3 was used in the quantification process.
In 80 EODGC-WES tumors and 65 adjacent normal tissue samples, median 5′ to 3′ coverage ratio was 0.78 (interquartile range [IQR], 0.68−0.85). Median coverage in exons was 97 (IQR, 84−123). Median percentage of total coverage in exons was 91.8% (IQR, 90.2%−93.2%). Median number of genes with at least 10× coverage was 15,510 (IQR, 14,757−16,300).
Chromosomal locus 3p21.1 (g.chr3: 48,369,660−55,002,466) was the most significantly deleted in EODGC-WES ( q = .0002, Figure 1 B ). Among the 161 genes in this locus, 6 genes including BAP1 were significantly ( q < .1 and fold-change < −0.5) underexpressed in EODGC-WES tumors with 3p21.1 deletion (n = 6), when compared with EODGC-WES tumors without 3p21.1 deletion (n = 74) ( Supplementary Table 8 and Figure 1 C ).
The deubiquitinase BAP1 may be a novel tumor suppressor gene candidate for DGC, because BAP1 mutation was the 16th most recurrent mutation in EODGC-WES ( Supplementary Table 9 ), and all 4 BAP1 mutations found in EODGC-WES were associated with LOH, suggesting a tumor suppressive role.
SNP6.0 Array Analysis
We performed a SNP6.0 array analysis on all of 109 tumors in the EODGC cohort and 88 of 115 tumors in the LODGC cohort, according to the Affymetrix Genome-Wide Human SNP Nsp/Sty 6.0 User Guide (Santa Clara, CA). There were no significant differences in clinicopathologic factors between LODGC cases with SNP6.0 data (n = 88) and those without SNP6.0 data (n = 27; [CR] ). Contrast QC medians for .cel files were 1.5 (IQR, 1.2−1.9) and 1.5 (IQR, 1.2−1.8), for EODGC and LODGC, respectively ( P = NS, t test). Contrast QC (Nsp) medians were 1.5 (IQR, 1.2−1.9) and 1.5 (IQR, 1.2−1.8), for EODGC and LODGC, respectively ( P = NS, t test). Contrast QC (Sty) medians were 1.3 (IQR, 0.8−1.8) and 1.7 (IQR, 1.4−2.0), for EODGC and LODGC, respectively ( P = NS, t test). Genome-wide copy number estimates were refined using tangent normalization. Recurrent focal copy number alterations were identified using GISTIC 2.0. 9
To determine TCGA subgroup of each sample, CIN status was determined based on Affymetrix SNP6.0 data, using a method described in a TCGA marker paper. 1 We performed the segmentation using the Circular Binary Segmentation method. 18 For copy number-based clustering, EODGC and LODGC tumors were subjected to hierarchical clustering along with TCGA tumors, CIN statuses of which were defined previously. 1 R-based hierarchical clustering was performed for thresholded copy number at reoccurring alteration peaks from GISTIC 2.0 analysis (all_lesions.conf_99.txt), using Euclidean distance and Wards method. The CIN status of new EODGC and LODGC tumors was determined based on the cluster membership ( Supplementary Figure 5 B ).
All tumors were categorized into 1 of 4 subtypes (EBV-positive [EBV], MSI-high [MSI-H], GS, and CIN). CIN-positive tumors without EBV or MSI were assigned to the CIN subgroup, and CIN-negative tumors without EBV or MSI were assigned to the GS subgroup. If a tumor was both EBV-positive and CIN, that sample was assigned to the EBV group ( Figure 6 A ).
Comparison of SNP6.0-Based Copy Number Profiles Between Early-Onset and the Late-Onset Diffuse Gastric Cancers
Copy number profiles based on SNP6.0 data were compared between EODGCs and LODGCs. Cross-validated misclassification rates were determined using arm-level copy number profiles of EODGC and the LODGC by class prediction algorithms built in the BRB-ArrayTools (version 4.4.1; National Cancer Institute, Bethesda, MD). Permutation P values were then determined for cross-validated misclassification rates. Two age groups were considered different for arm-level copy number profiles if permutation P values were <.05. According to prediction analyses built in the BRB-ArrayTools, permutation P values were >.05 ( red dashed line ) with all of the class prediction algorithms at feature selection P values <.05. Permutation P values were .14, .15, .51, .14, .13, and .08 for the compound covariate predictor, the diagonal linear discriminant analysis classifier, the 1-nearest neighbor classifier, the 3-nearest neighbor classifier, the nearest centroid classifier, and the support vector machines classifier, respectively ( Supplementary Figure 5 D ). Thus, EODGC and LODGC did not differ in arm-level copy number profiles.
In EODGC, a total of 21 chromosomal locations with peaks displayed significantly recurrent focal amplifications. The recurrent focal amplification at 10q26.13 at the locus gene encoding FGFR2 was the dominant peak ( q = 1.76 × 10 −10 ). Significant amplifications were observed at 3q26.2, 6q21, 6q23.3, 6p24.1, 7q22.1, 8q24.21, 9p21.2, 10q26.12, 11p13, 12p12.1, 16p13.3, 17q12, 17q25.3, 18q11.2, 19q12, and 21q21.1. CD44 , ERBB2, GATA6, KRAS, MYC , and POU5F1B were included in these loci. In the LODGC cohort, we found 23 chromosomal locations with peaks that were significantly recurrent focal amplifications. Similar to EODGC, 10q26.13 was a clearly dominant peak ( q = 1.75 × 10 −7 ). Significant amplifications were observed at 3q26.31, 7q11.21, 7q31.2, 7p11.2, 8q24.21, 8p23.1, 10q11.21, 11q13.3, 11p13, 12p12.1, 17q12, 18q11.2, 9p24.1, 19q12, and 20q13.2. CD44, EGFR, ERBB2, FGFR2, GATA4 , and GATA6 were present in these loci.
Six peaks, including 8q24.21 ( MYC ), 10q26.13 ( FGFR2 ), and 17q12 ( ERBB2 ), were amplified in both the EODGC and the LODGC cohorts. FGFR gene amplification was identified in 8.2% and 7.9% of EODGC and LODGC, respectively ( P = NS, χ 2 ), suggesting ethnic difference. 3q26.2, 6q21, 6q23.3, 6p24.1, 7q22.1, 9p21.2, 10q26.12, 16p13.3, 17q25.3, 19q12, and 21q21.1 were recurrently amplified in EODGC only. ERBB2 amplifications tended to be more frequent in EODGC than in LODGC (9.1% vs 3.4%, respectively), but the difference was not significant ( P = .1, χ 2 ). The following loci were recurrently amplified only in the LODGC cohort: 3q26.31, 7q11.21, 7q31.2, 7p11.2, 8p23.1, 9p24.1, 10q11.21, 11q13.3, and 20q13.2. EGFR and GATA4 were contained in these loci.
Using a silver in situ hybridization (ISH) analysis, we validated the gene amplification of ERBB2 , which represents the only clinically actionable gene with Food and Drug Administration−approved treatment for GCs, in an independent dataset of DGCs collected from patients who were aged ≤45 years and treated at Asan Medical Center and the National Cancer Center of Korea. The slides were processed using the automated system following the manufacturer’s protocols for INFORM HER2 DNA and chromosome 17 ( CEP17 ) probes (Ventana Medical Systems, Tucson, AZ). 19 Both probes were sequentially hybridized in 1 slide. HER2 gene was visualized as a black dot and CEP17 as a red dot. The specimen was then counterstained with Harris hematoxylin. HER2 gene amplification status was evaluated by counting HER2 and CEP17 signals in nuclei of 20 non-overlapping consecutive tumor cells with the ASCO/CAP guidelines; negative for HER2 gene amplification if the HER2/CEP17 ratio was <1.8, equivocal if the HER2/CEP17 ratio was 1.8 to 2.2, and positive if the HER2/CEP17 ratio was >2.2. 20 21 In equivocal cases, we counted 20 additional tumor cells and considered the case positive for HER2 gene amplification if the HER2/CEP17 ratio was ≥2.0. 20 21 Seventeen of 275 tumors in this validation dataset from patients aged ≤45 years (6.2%) demonstrated silver ISH evidences for ERBB2 amplifications. The silver ISH positivity in this dataset (6.2%) was similar to the ERBB2 amplification rate determined based on WES in the EODGC cohort (5.9%) and slightly lower than the rate determined based on SNP6.0 data in EODGC (8.3%).
Using immunohistochemistry, we validated FGFR2 amplifications. FGFR2 -amplified tumors (n = 9) showed significantly higher FGFR2 immunohistochemistry grades than did tumors without FGFR2 amplification (n = 10) ( P = .036, t test).
Microsatellite Instability Assay
MSI was evaluated using a panel of 2 mononucleotide repeat loci (the polyadenine tracts BAT25 and BAT26) and 3 dinucleotide repeat loci (CA repeats in D2S123, D5S246, and D17S250), as described previously. 22 23 Briefly, fluorescently labeled PCR and capillary electrophoresis were performed using tumor and matched normal tissue DNA samples. The primer sequences for the PCRs were as follows: F (FAM): 5′ -TCG CCT CCA AGA ATG TAA GT-3′ and R: 5′ -TCT GCA TTT TAA CTA TGG CTC-3′ for BAT-25; F (HEX): 5′ -TGA CTA CTT TTG ACT TCA GCC-3′ and R: 5′ -AAC CAT TCA ACA TTT TTA ACC C-3′ for BAT-26; F (FAM): 5′ -AAA CAG GAT GCC TGC CTT TA-3′ and R: 5′ -GGA CTT TCC ACC TAT GGG AC-3′ for D2S123; F (FAM): 5′ -GGA AGA ATC AAA TAG ACA AT-3′ and R: 5′ -GCT GGC CAT ATA TAT ATT TAA ACC-3′ for D17S250; and F (HEX): 5′ -ACT CAC TCT AGT GAT AAA TCG GG-3′ and R: 5′ -AGC AGA TAA GAC AGT ATT ACT AGT T-3′ for D5S246. The touch-up PCR was performed essentially as described previously 23 using HotStarTaq DNA Polymerase (Qiagen). After reaction mixtures with 1 μL of the 1/10 to 1/50-diluted PCR products, 8.5 μL of Hi-Di Formamide, and 0.5 μL of GeneScan 500 ROX Size Standard (Thermo Fisher Scientific) were prepared, denatured at 95°C for 2 minutes, and snap-cooled on ice, the microsatellite patterns were detected after automated capillary electrophoresis using an ABI3730 Genetic Analyzer (Thermo Fisher Scientific) with a POP-7 polymer (Thermo Fisher Scientific). Tumor was classified as high-level MSI (MSI-H) if ≥2 markers demonstrated unequivocal alterations of peak pattern compared with normal DNA.
EODGC (n = 109) and LODGC (n = 115) were significantly different in the frequency of MSI-H (4.6% vs 13.0% for EODGC and LODGC, respectively; P = .027, χ 2 ) ( [CR] ).
Detection of the Epstein-Barr Virus
In EODGC, EBV positivity was determined using ISH and PCR. The presence of EBV in the cancer cells was assessed using EBV chromogenic ISH on an automatic staining device (Benchmark XT; Ventana, Tucson, AZ) according to the manufacturer’s guidelines. 24 Briefly, 4-μm-thick sections were cut from representative blocks obtained from each patient, mounted onto coated slides, and dried at 74°C for 30 minutes. After pretreatment with ISH protease 2 (Ventana) for 8 minutes at 37°C, the slides were denatured at 85°C for 12 minutes and hybridization was conducted at 57°C for 1 hour using EBV-encoded small RNA probes (INFORM, Ventana) that had been labeled with fluorescein. Detection was sequentially performed by applying mouse antifluorescein antibody and biotinylated goat anti-mouse antibody (iView blue; Ventana); counterstaining was performed using Nuclear Fast Red. Obvious, strong staining in the nucleus was considered to indicate positivity, and the proportion of EBV-positive tumor cells was thereby evaluated. A representative sample with EBV-positive nuclei in cancer cells is shown in Supplementary Figure 8 G .
EBV PCR was performed as described previously, 25 26 with some modifications. The PCR mixture (10 μL) contained 0.25 U HotStarTaq DNA Polymerase (Qiagen), 1 μL of 10× PCR Buffer, 200 μM of each dNTP, 0.3 μM of each primer, and 5 ng of genomic DNA from the tumor sample. The primer sequences were as follows 25 : 5′-CCA TGT AAG CCT GCC TCG AG-3′ and 5′-GCC TTA GAT CTG GCT CTT TG-3′. The cycling conditions were as follows: 95°C for 15 minutes followed by 30 cycles of 94°C for 40 seconds, 57°C for 30 seconds, and 72°C for 30 seconds, and a final extension of 72°C for 3 minutes. An unequivocal PCR band was considered to indicate positivity for EBV. In the EODGC cohort, the ISH results and PCR results were 100% concordant. Therefore, in the LODGC cohort, PCR was used to determine EBV status. EBV positivity was almost equivalent between 109 EODGCs and 115 LODGCs (6.2% vs 6.1%, respectively; [CR] ).
Detection of Helicobacter pylori Using Real-Time Polymerase Chain Reaction
Real-time PCR was performed using a LightCycler 480 (Roche Diagnostics) with 8 μL of reaction mixture containing 4 μL of 2× QuantiTect SYBR Green PCR Master Mix (Qiagen), 250 nM of each primer, and 5 ng of genomic DNA that was isolated from adjacent normal tissue samples. The cycling conditions were as follows: 95°C for 15 minutes, followed by 45 cycles of 95°C for 20 seconds, 58°C for 20 seconds, and 72°C for 20 s. The primers used were Hp23S 1835F (5′-GGT CTC AGC AAA GAG TCC CT-3′) and Hp23S 2327R (5′-CCC ACC AAG CAT TGT CCT-3′). Cp value <40 was interpreted as a positive result. In the EODGC cohort, 53 of 109 samples (48.6%) were positive for H pylori . No mutations or mutation signatures significantly correlated with H pylori positivity.
Reverse Transcription Polymerase Chain Reaction Analysis of CDH1 Splice Variants
cDNA was synthesized from 500 ng of total RNA using a SuperScript III First-Strand Synthesis System (Thermo Fisher Scientific), and was PCR-amplified for CDH1 . The PCR mixture (20 μL) contained 0.5 U of HotStarTaq DNA Polymerase (Qiagen), 2 μL of 10× PCR buffer, 200 μM of each dNTP, 0.3 μM of each primer, and 0.3 μL of the synthesized cDNA. The cycling conditions were as follows: 95°C for 15 minutes followed by 40−45 cycles consisting of 94°C for 30 seconds, 55°C for 30 seconds, and 72°C for 30 seconds. All CDH1 splice site mutations were validated using capillary sequencing. The primer sequences were as follows: CDH1 for exons 4 and 5, 5′-CCA TCA GGC CTC CGT TTC TGG-3′ and 5′-TGT CAT TCT GAT CGG TTA CCG TGA TC-3′; CDH1 for exons 8 and 9, 5′-CTC AGC CAA GAT CCT GAG CTC-3′ and 5′-CTT CAT TCA CAT CCA GCA CAT CCA C-3′. All CDH1 splice site mutations were validated using capillary sequencing and are summarized in [CR] .
Bioinformatic Prediction of CDH1 Cryptic Splice Donor Sites and 3-Dimensional Computation Modeling for CDH1 Mutations
According to the Human Splicing Finder ( http://www.umd.be/HSF/ ) public database, 27 chr16:g.68842471G>A and chr16:g.68842472T>C mutations were predicted to inactivate the endogenous splice donor site in the exon 4-intron 4 junction and to generate a novel cryptic splice donor site at c.468 ( [CR] ). Reverse transcription PCR and Sanger sequencing validated the presence of a truncated transcript that lacked 63 bp (21 amino acids) exclusively in tumors with the corresponding mutations. As summarized in [CR] , the same bioinformatics analysis predicted that the endogenous splice donor site in the exon 5−intron 5 junction was inactivated by mutations in chr16:g.68842752G>A, chr16:68842749 ACTG/A, and chr16:68842753del and that a cryptic splice donor site was created at c.645. As predicted, a novel splice variant lacking 42 bp (14 amino acids) was identified in reverse transcription PCR sequencing in tumors with the corresponding mutations. Homology modeling was performed using SWISS-MODEL. 27 28
CDH1 Loss of Heterozygosity Analysis
LOH analyses were performed using 3 microsatellite markers (D16S4025, D16S396, and D16S4067) that flank the CDH1 locus and 3 intragenic CDH1 single-nucleotide polymorphism (SNP) markers (rs16260, rs1801552, and rs1801026), as described previously, 29 30 with modifications.
Microsatellite markers were evaluated as follows: a total of 100 ng of tumor and matched normal tissue DNA were used to perform a fluorescently labeled PCR with the following PCR conditions: an initial activation step of 95°C for 15 minutes, 28 cycles of 94°C for 30 seconds, 58°C for 30 seconds, and 72°C for 30 seconds, and a final extension of 72°C for 3 minutes; the following reaction composition: 0.25 U of HotStarTaq DNA Polymerase (Qiagen), 1 μL of 10× PCR buffer, 200 μM of each dNTP, 0.3 μM of each primer, and 5 ng of genomic DNA in 10 μL of mixture; and the following primer sets: D16S4025, 5′-TCC ATT GGA CTT ATA ACC ATG-3′ and 5′-AGC TGA GAG ACA TCT GGG-3′; D16S396, 5′-GAA AGG CTA CTT CAT AGA TGG CAA T-3′ and 5′-ATA AGC CAC TGC GCC CAT-3′; and D16S4067, 5′-GCC ACC TCA CAC TAG CCT G-3′ and 5′-TTC CAG CTC TCA CTC AAA ATG-3′. The PCR products were visualized using an ABI3730 Genetic Analyzer (Thermo Fisher Scientific). Data were analyzed using Peak Scanner software (version 1.0, Thermo Fisher Scientific), and the samples were determined to display LOH if the LOH index, calculated as follows, was >2 or <0.5.
LOH index = (N1/N2)/(T1/T2),
where N1 is peak areas of germline allele 1; N2 is peak areas of germline allele 2; T1 is peak areas of tumor allele 1; and T2 is peak areas of tumor allele 2.
All experiments involving microsatellite LOH-positive samples were repeated twice for validation.
SNP loci were evaluated as follows: an SNP locus in each tumor sample was considered to be informative when the corresponding germline sequence was heterozygous in capillary sequencing. Targeted sequencing was performed on each tumor genomic DNA sample as described here. The allele frequency of a given SNP locus was determined for each tumor sample. A SNP locus was identified to display LOH if the minor allele frequency was <0.4 in the tumor sample and if the locus was heterozygous in the matched germline sample.
At least 1 LOH marker was informative in 89 and 96 samples in the EODGC and LODGC cohorts, respectively. A total of 31 (34.8%) and 21 (21.9%) samples were determined to display LOH for CDH1 in the EODGC and LODGC cohorts, respectively ( P = .05, χ 2 ; [CR] ).
Real-Time Polymerase Chain Reaction for Genomic Copy of CDH1
A homozygous deletion of CDH1 was determined using real-time PCR. The real-time PCR was performed using a LightCycler 480 (Roche Diagnostics, Mannheim, Germany) with 10 μL of reaction mixture containing 5 μL of 2× QuantiTect Probe PCR Master Mix (Qiagen), 200 nM of each primer, 100 nM of TaqMan probe, and 10 ng of genomic DNA. The cycling conditions were as follows: 95°C for 15 minutes, followed by 99 cycles of 95°C for 20 seconds, 58°C for 20 seconds, and 72°C for 20 seconds. Each sample was run in duplicate. The primer sequences used were as follows: CDH1 , 5′-TTC TAC AGC ATC ACT GGC CAA G-3′ and 5′-AGT GTA TGT GGC AAT GCG TTC TC-3′; and LINE1 , 5′-AAA GCC GCT CAA CTA CAT GG-3′ and 5′-TGC TTT GAA TGC GTC CCA GAG-3′. The probe sequences used were as follows: CDH1 , 5′-FAM-AGC TGA CAC ACC CCC TGT TGG TGT C-BHQ-1-3′; and LINE1 , 5′-FAM-CTG AAC AAC CTG CTC CTG AAT GAC TAC TG-BHQ-1-3′. A germline DNA sample (from buffy coat) was serially diluted with SK-BR3 (a cell line with CDH1 homozygous deletion) DNA to generate a standard curve. A homozygous deletion was defined as a mean LINE -normalized log 2 copy number ratio < −1.0. CDH1 homozygous deletion was identified in 2 EODGC samples ( [CR] ).
CDH1 Promoter DNA Methylation Analysis
CDH1 promoter DNA methylation analysis was performed to analyze 32 CpG islands within the 160-bp upstream and 153-bp downstream of the CDH1 transcription start site. EpiTect Bisulfite Kit (Qiagen) and 100 ng of genomic DNA were used to convert unmethylated cytosines to uracil, whereas methylated residues remained unmodified. The bisulfite conversion reaction was performed as recommended by the manufacturer (Qiagen). Buffy coat DNA that was in vitro methylated using the M.SssI (New England Biolabs) was used as a positive control. The genomic region was PCR-amplified with the following composition: 0.5 U of i-StarTaq GH DNA Polymerase (iNtRON Biotechnology, Seongnam, Korea), 1 μL of 10× PCR buffer, 250 μM of each dNTP, 0.3 μM of each primer, and 1.5 μL of bisulfite-converted genomic DNA from 20 μL of the eluate. The cycling conditions were as follows: 95°C for 5 minutes followed by 60 cycles of 95°C for 30 seconds, 58°C for 30 seconds, and 72°C for 40 seconds, and a final extension of 72°C for 3 minutes. The following primer set was used: 5′-GAT TTT AGT AAT TTT AGG TTA GAG GGT TAT-3′ and 5′-AAA TAC CTA CAA CAA CAA CAA CAA C-3′.
Cloning of the PCR products and transformation of Escherichia coli were performed using TOPO-TA Cloning Kit (Thermo Fisher Scientific) as recommended by the manufacturer. The next day, colony PCRs were performed using the i-StarTaq GH DNA Polymerase with M13 primers. Methylation status of the CpG sites was analyzed using capillary sequencing with the colony PCR products. Tumors that showed methylation at ≥25% of the CpG sites (≥8 of 32 CpG sites) were defined as hypermethylated tumors. 30
A total of 5 and 16 tumors were identified as hypermethylated tumors in EODGC and LODGC, respectively (4.6% vs 13.9%; P = .017, χ 2 ; [CR] ).
The GC tissues were fixed in 10% phosphate-buffered formalin, processed in a routine manner, and embedded in paraffin. IHC was performed on 4-μm-thick serial sections from whole tissue paraffin blocks. Immunohistochemical staining for E-cadherin was performed using an automated Discovery XT (Ventana Medical System, Tucson, AZ) instrument. Briefly, the paraffin sections were deparaffinized, rehydrated using EZ prep (Ventana Medical System), and then washed with reaction buffer (Ventana Medical System). The antigens were retrieved using heat in pH 8.0 Tris−EDTA buffer (CC1; Ventana Medical System) at 90°C for 30 minutes. The slides were incubated for 20 minutes at room temperature with a 4A2C7/monoclonal mouse anti−E-cadherin antibody (1:4000 dilution, 18-0223; Zymed/Thermo Fisher Scientific). 31 For immunohistochemistry of β-catenin, MYC, and FGFR2, an ImmPRESS Peroxidase Polymer kit (Vector Laboratories, Burlingame, CA) was used according to the manufacturer’s protocol. Briefly, dewaxed and rehydrated paraffin sections were subjected to antigen retrieval by heating the sections to 100°C for 20 minutes in 0.01 M citrate buffer (pH 6.0). The slides were incubated with 2.5% horse serum during blocking and then incubated for 30 minutes at room temperature with 14/monoclonal mouse anti−β-catenin antibody 32 33 (1:400 dilution; 610154; BD Biosciences, San Jose, CA), Y69/monoclonal rabbit anti-MYC antibody 20 34 (1:200 dilution; ab32072; Abcam, Cambridge, MA), and polyclonal rabbit anti-FGFR2 antibody 35 (1:200 dilution; ab52246; Abcam). After the sections were washed, the slides were incubated for 30 minutes with the peroxidase polymer-linked secondary antibody that was contained in the kit. The slides were subjected to colorimetric detection using the ImmPact DAB substrate (SK-4105; Vector Laboratories). All of the stained slides were scanned using Aperio ScanScope AT (Aperio Technologies, Inc., Vista, CA). IHC grading was performed under high-power microscopic magnification (×400). The positive rates were depicted as the mean value of at least 6 high power fields (×200). Care was taken to select the fields equally from invasive and mucosal layers so that the grading may not be biased by the proportion of invasive fronts included across tumors. Specimens with <6 high-power fields or 2000 cancer cells, mostly small biopsy samples, were excluded from grading for immunohistochemistry. A total of 197 and 194 tumors in the EODGC and LODGC cohorts were eligible for IHC grading for E-cadherin and β-catenin, respectively.
A team of 5 pathologists (K.S.S., J.L., H.K.C., S.Y.K., and J.W.P.) performed blind reviews of E-cadherin and β-catenin immunohistochemistry. E-cadherin immunohistochemistry was independently reviewed and graded by these 5 pathologists. When the grades differed, the pathologists conferred and came to a consensus. Without prior knowledge of genomic data, a pathologist (J.W.P.) visually scored the percentage of nuclear β-catenin immunostaining, the score was then verified by the other 4 pathologists (K.S.S., J.L., H.K.C., and S.Y.K.). The pathologists conferred for consensus in case they disagreed.
For E-cadherin grading, the presence of strong complete circumferential membrane stain in non-cancerous epithelial cells was defined as positive internal and external controls. Staining was considered aberrant if the cancer cells showed (1) an absence of circumferential membrane staining; (2) reduced-intensity, partial or weak linear membrane staining; or (3) a focal or punctate (dot-like) cytoplasmic or membrane reaction ( Figure 4 A ). 36 E-cadherin was graded into 3 groups according to the percentages of tumor cells with aberrant staining: grade 2, >80% of the cancer cells showed homogenous membranous E-cadherin staining: grade 1, 20%−80% of the cancer cells were positively stained: and grade 0, <20% of the cancer cells showed positive staining. E-cadherin immunostaining was considered abnormal if aberrant or reduced staining was identified in >20% of tumor cells (ie, grade 0 or 1). 10 37 The nuclear expression of β-catenin was scored according to the percentage of cancer cells that exhibited unequivocal and strong nuclear staining.
RHOA -mutated tumors (n = 30) exhibited more prominent nuclear β-catenin immunostaining than did wild-type tumors for RHOA (n = 164) ( P = .011, Wilcoxon). Independent of the CDH1 alteration status, the RHOA mutation status was significantly associated with nuclear β-catenin immunostaining, according to a multiple regression analysis performed using the SAS GLM procedure (adjusted P = .01, linear regression; Supplementary Table 11 ).
To evaluate the correlation of MYC expression with nuclear β-catenin expression, 64 GC cases were randomly selected for MYC immunohistochemistry. MYC expression was scored according to the percentage of cancer cells exhibiting unequivocal and strong nuclear staining ( Supplementary Figure 6 E ).
To validate FGFR2 overexpression in FGFR2 -amplified samples, a total of 19 cases including 9 FGFR2 -amplified and randomly selected 10 tumors without FGRF2 amplification were subjected to FGFR2 immunohistochemistry. Cytoplasmic FGFR2 expression was scored according to the intensity of staining: 0, negative staining; 1, weakly positive staining; 2, moderately positive staining; 3, strongly positive staining. 38 The cytoplasmic immunoreactivity in adjacent normal gastric epithelial cells was defined as internal controls showing negative to weak positivity. FGFR2 -amplified tumors (n = 9) showed significantly higher FGFR2 immunohistochemistry grades than did tumors without FGFR2 amplification (n = 10) ( P = .036, t test).
NUGC-4 (JCRB0834; JCRB) was purchased for this study. The identity of MKN-45 (80103; KCLB, Seoul, Korea) was verified by STR profiling was using AmplFLSTR identifiler PCR Amplification kit (Applied Biosystems, Foster, CA). NCC-S1 cell line was established by our group from a diffuse-type gastric adenocarcinoma formed in a Villin-cre;Smad4 F/F ;Trp53 F/F ;Cdh1 F/wt mouse. 32 39 Pdx1-cre;Smad4 F/F ;Trp53 F/F ;Cdh1 F/+ cells were primary cultured by our group from a diffuse-type gastric adenocarcinoma formed in a Pdx1-cre;Smad4 F/F ;Trp53 F/F ;Cdh1 F/wt mouse. 32 39 The Chinese hamster ovary cell line CHO-K1 (10061; KCLB) was purchased from the Korean Cell Line Bank (Seoul, Korea). All cells were grown in RPMI-1640 with 10% fetal bovine serum (FBS) and 0.5% penicillin-streptomycin solution (WELGENE, Gyeongsan, Korea) at 37°C under 5% CO 2 in a humidified incubator.
BAP1 Expression Lentiviral Vectors
For cDNA cloning of BAP1 , full length CDS of BAP1 was PCR-amplified with the primers 5′-ATG AAT TCG CCA CCA TGA ATA AGG GCT GGC TGG AG-3′/5′-TGT GAT TGT CTA GAA AGG CCG-3′ and 5′-TTT CTA GAC AAT CAC AAT TAT GCC AAG-3′/5′-TAG CGG CCG CTC ACT GGC GCT TGG CCT TG-3′. The amplicons were then digested with EcoRI/XbaI and XbaI/NotI, and were cloned into a pCDH-CMV-MCS-EF1-Puro-copGFP vector.
CDH1 Expression Lentiviral Vectors
Using lentivirus vectors, we ectopically expressed CDH1 using CMV promoters in human GC cells, and expressed CDH1 under CAG promoters CHO-K1 and mouse DGC cells. We first generated vectors by modifying a selection marker (copGFP) in pCDH-CMV-MCS-EF1-copGFP (System Biosciences, Mountain View, CA). To generate an appropriate enzyme site (KpnI) near the downstream promoter of the selection marker to join the vector and the new selection marker together, the EF1 promoter (pEF1) in the original vector was subjected to PCR reaction with the primers 5′-ATG CGG CCG CAA GGA TCT GCG ATC GCT C-3′/5′-ATG GTA CCG GTA GGC GCC GGT CAC AG-3′. For the dual selection marker puromycin resistance gene (PURO) and copGFP, the full-length PURO (without a stop codon) and copGFP genes were amplified from template vectors (pLKO.1-Puro (Sigma, St Louis, MO) and pCDH-CMV-MCS-EF1-copGFP, respectively) using PCR with the primer pairs: 5′-ATG GTA CCG CCA CCA TGA CCG AGT ACA AGC CC-3′/5′-ATA GAT CTG GCA CCG GGC TTG CGG GT-3′ and 5′-ATA GAT CTG CCA CCA TGG AGA GCG ACG AGA GC-3′/5′-TAG TCG ACT TAG CGA GAT CCG GTG GAG-3′. Restriction enzyme digestion of vectors and amplified DNA fragments was performed as follows: NotI/SalI for the vector backbone (pCDH-CMV-MCS-EF1-copGFP), NotI/KpnI for pEF1, KpnI/BglII for PURO, and BglII/SalI for copGFP. All fragments were then ligated altogether to generate a vector (pCDH-CMV-MCS-EF1-Puro-copGFP) containing a PURO-copGFP fusion selection marker. To obtain a selection marker containing a neomycin resistance gene (NEO), the full-length NEO gene was amplified from a pcDNA3.1 (Thermo Fisher Scientific) template vector using PCR with the primers 5′-ATG GTA CCG CCA CCA TGA TTG AAC AAG ATG GAT TG-3′/5′-TAG TCG ACT CAG AAG AAC TCG TCA AGA AG-3′. Digestion was then performed using NotI/SalI for the vector backbone, NotI/KpnI for pEF1, and KpnI/SalI for NEO. All of the fragments were then ligated together to construct pCDH-CMV-MCS-EF1-Neo. To obtain a selection marker containing PURO, pEF1 was first reconstituted using PCR with the primers 5′-ATG CGG CCG CAA GGA TCT GCG ATC GCT C-3′/5′-ATC TCG AGG TAG GCG CCG GTC ACA GC-3′ to tag an appropriate enzyme site (XhoI) that could be used to join pEF1 and PURO together. Then, the full-length PURO gene was amplified from pLKO.1-Puro using PCR with the primers 5′-ATC TCG AGA TGA CCG AGT ACA AGC CC-3′/5′-TAG TCG ACT CAG GCA CCG GGC TTG CG-3′. Digestion was performed using NotI/SalI for the vector backbone, NotI/XhoI for pEF1, and XhoI/SalI for PURO. All fragments were then ligated together to obtain pCDH-CMV-MCS-EF1-Puro. For easy PCR amplification of pCAG, pCAG was divided and amplified into 2 fragments from the Ai9 (Addgene, Cambridge, MA) template vector using the primers 5′-ATT ACT AGT TAT TAA TAG TAA TCA ATT ACG GG-3′/5′-ACA AAG GGC CCT CCC GGA G-3′ and 5′-GGG AGG GCC CTT TGT GCG-3′/5′-ATA GAA TTC GCT AGC TTT GCC AAA ATG ATG AGA CAG-3′. After digestion with SpeI/EcoRI, SpeI/ApaI for upstream pCAG, and ApaI/EcoRI for downstream pCAG all fragments were ligated altogether to pCDH-CAG-MCS-EF1-Puro-copGFP/Neo/Puro. To clone the CDH1 cDNA, the full-length coding sequence (CDS) of CDH1 was divided and PCR-amplified into 2 fragments by tagging a KpnI junction with the primers 5′-ATG CTA GCG CCA CCA TGG GCC CTT GGA GCC GC-3′/5′-AAG GTA CCA CAT TCG TCA CTG C-3′ and 5′-GTG GTA CCT TTT GAG GTC TCT C-3′/5′-ATG CGG CCG CCT AGT CGT CCT CGC CGC C-3′. Wild-type and mutant CDH1 fragments were amplified from the cDNA of the relevant clinical samples. Specifically, for the E880K variant, the downstream fragment of CDH1 was amplified using the primers 5′-GTG GTA CCT TTT GAG GTC TCT C-3′/5′-ATG CGG CCG CCT AGT CGT CCT TGC CGC CTC-3′. Restriction enzyme digestion of the newly made vectors and amplified DNA fragments was performed using NheI/NotI for the vectors, NheI/KpnI for upstream CDH1 , and KpnI/NotI for downstream CDH1 . The cut vector and the upstream/downstream fragments of each CDH1 gene were ligated to wild-type and mutant CDH1 -expressing vectors.
RHOA Expression Vectors
To clone the RHOA cDNA, full-length CDS of RHOA was PCR-amplified from a RHOA-wild-type cDNA sample. Primer pairs (5′-ATG AAT TCG CCA CCA TGG CTG CCA TCC GGA AG-3′/5′-ATG CGG CCG CTC ACA AGA CAA GGC ACC CAG-3′ and 5′-ATG AAT TCG CCA CCA TGG CTG CCA TCT GGA AG-3′/5′-ATG CGG CCG CTC ACA AGA CAA GGC ACC CAG-3′) were used to amplify the wild-type and mutant (R5W) RHOA genes, respectively. The wild-type and mutant RHOA genes were cloned into pcDNA3.1 (+) (Thermo Fisher Scientific) using EcoRI and NotI.
Generation of Stable Cell Lines
All cloned vector sequences were validated using Sanger sequencing. Three days after 293FT cells (Thermo Fisher Scientific) were transfected with one of the cloned gene expression vectors, pMD2.G (Addgene), and psPAX2 (Addgene), the growth medium was harvested and concentrated using ultracentrifugation. Cells were transduced at a multiplicity of infection of 0.1. For antibiotic selection, NUGC-4 cells and MKN-45 cells were treated with 2 μg/mL of puromycin for 3−4 passages. For CHO-K1 cells, 20 μg/mL of puromycin or 500 μg/mL of G418 was used for 3−4 passages. For validation of transgene expression, after cDNA was synthesized from 5 μg of total RNA using amfiRivert cDNA Synthesis Platinum Master Mix (GenDEPOT, Barker, TX), PCR was performed using 0.4 μL of the synthesized cDNA per 20 μL of the ExTaq mixture (Takara, Shiga, Japan). PCR amplifications were performed using 33 cycles for the transgene and 28 cycles for the internal control ( Actb or GAPDH ). The primer sequences used for these experiments were as follows: CDH1 for exons 4 and 5, 5′-CCA TCA GGC CTC CGT TTC TGG-3′ and 5′-TGT CAT TCT GAT CGG TTA CCG TGA TC-3′; GAPDH , 5′-GAG TCA ACG GAT TTG GTC G-3′ and 5′-TGG AAT CAT ATT GGA ACA TGT AAA C-3′; and Actb , 5′-GAA CAT GGC ATT GTT ACC AAC TG-3′ and 5′-GTG TTG AAG GTC TCA AAC ATG ATC-3′.
Small Interfering RNA-Mediated RHOA Knockdown and Exoenzyme C3 Transferase Treatment
Small interfering RNA (siRNA) SMARTpool against RHOA (J-003860-10 and J-003860-13; Thermo Fisher Scientific) and Rhoa (J-042634-05; Thermo Fisher Scientific) were used to knock down RHOA in human and mouse DGC cells, respectively. Cells suspended in OPTI-MEM medium containing 5% FBS were co-transfected with the reporter plasmids along with siRNAs (20 nM). After 48 hours of the co-transfection, luciferase activity was measured as described.
To evaluate the effect of exoenzyme C3 transferase ( C3 ; CT04; Cytoskeleton, Denver, CO) on the reporter activities, GC cells transfected with the reporter plasmids for 24 hours were incubated with 1 μg/mL of C3 in serum-free RPMI-1640 for 24 hours.
BAP1 Proliferation Assays
To evaluate the effect of BAP1 on cell proliferation, we used MTT assays. The 2 × 10 5 cells stably expressing empty vector or BAP1 were seeded to each well of 12-well plate. After the cells were incubated in a mixture of 0.1 mL of MTT solution (Sigma; 5 mg/mL in phosphate-buffered saline [PBS]) and 0.5 mL of growth medium, the mixture was completely removed, and then insoluble reactants of the cells were dissolved with 0.5 mL of dimethyl sulfoxide. Absorbance of 0.1 mL of the solution in a 96-well plate was measured every day for 3−4 days at 570 nm using a microplate reader (Molecular Devices, Sunnyvale, CA). Normalized optical density at 570 nm was calculated by subtracting the optical density of dimethyl sulfoxide blank.
Slow Aggregation Assay
After the cells were detached using trypsin, 2 × 10 4 cells in 0.2 mL of growth medium were seeded into 96-well plates that were pre-coated with 45 μL of 0.8% (w/v) DIFCO Noble Agar (BD Biosciences) in PBS. 40 41 After 24 hours of incubation, images showing cellular aggregation were obtained using a light microscope (50×) (Axio Observer Z1, Carl Zeiss, Jena, Germany).
Measurement of the Serum Responsive Factor and Wnt/β-Catenin Activity
To assess the activity of serum responsive factor and Wnt/β-catenin activity, we conducted luciferase reporter assays using the pGL4.34[ luc2P /serum responsive factor-RE/Hygro] reporter plasmid (Promega, Madison, WI) and Super 8× TOPFlash and FOPFlash reporter plasmids (Addgene). To evaluate whether an R5W RHOA mutation is gain- or loss-of-function, we suspended 2.5 × 10 5 293FT cells in 0.5 mL of OPTI-MEM medium (Thermo Fisher Scientific) containing 5% FBS and cell suspension cells was seeded in 24-well plates. Cells were then transfected with 200 ng of plasmids using Lipofectamine 2000 (Thermo Fisher Scientific), as recommended by the manufacturer. After 24 hours of transfection, cells were transfected with 200 ng of reporter plasmids and 2.5 ng of Renilla luciferase-constitutively expressing vector (Addgene; for internal control) per well using Lipofectamine 2000 (Thermo Fisher Scientific). After 48 hours of the first transfection, luciferase activity was measured using a Dual-Luciferase Reporter Assay System (Promega) and a microplate luminometer (Wallac 1420 Victor 3; PerkinElmer, Waltham, MA).
Immunofluorescence Analysis for Actin Stress Fiber and E-Cadherin
293FT cells suspended in OPTI-MEM media containing 5% FBS were transfected with empty, wild-type, or mutant (R5W) RHOA vectors. GC cells were suspended and transfected with siRHOA (20 nM) . Suspension cells (1 × 10 5 cells) were then seeded on sterile glass coverslips (18 mm in diameter) in 12-well plates. For C3 treatment, 1 × 10 5 GC cells were seed on glass coverslips and grown to 70%−80% confluency, and cells were treated with 1 μg/mL of C3 in serum-free RPMI-1640. After 48 hours after the transfection (24 hours after C3 treatment), cells on glass coverslips were fixed in 4% paraformaldehyde for 10 minutes and permeabilized in 0.5% Triton X-100 (Sigma) for 5 minutes at room temperature. For immunofluorescence (IF) analysis for actin stress fiber, rhodamine-phalloidin (PHDR1, Cytoskeleton) was used. Stained slides were evaluated using LSM510 confocal microscope (Carl Zeiss). For the IF analysis of E-cadherin in wild-type or mutant CDH1 -transduced NCC-S3 cells, cells were seeded on sterile glass coverslips and grown to 70%−80% confluence. Cells on the coverslips were fixed and permeabilized as described, and incubated with 5% goat serum for blocking. The primary 24E10/monoclonal rabbit against E-cadherin antibody (1:1000; 3195; Cell Signaling) was incubated overnight at 4°C. After washing, Alexa Fluor 594 goat anti-rabbit IgG (H+L) secondary antibody, (1:250; R37117; Thermo Fisher Scientific) was incubated for 2 hours at room temperature. Slides were mounted with Vectashield mounting media (H-1200; Vector Laboratories) and evaluated using Zeiss Axio Imager HBO 100 (Carl Zeiss).
RhoA Pull-Down Activation Assay
RhoA pull-down activation assays (RhoTekin assay) were performed using a RHOA Activation Assay Biochem Kit (BK036; Cytoskeleton) as recommended by the manufacturer, with some modifications. Cells were harvested and lysed using T-PER reagent with protease inhibitors. After protein concentration was measured using a BCA reagent, pull-down was performed using 0.3 mg of total protein and 60 μg of rhotekin-RBD beads in 0.5 mL of T-PER Reagent at 4°C for 1 hour. The bead pellets were washed twice with T-PER Reagent that was diluted 1:10 in PBS. After the last wash, buffer was removed leaving 90 μL of buffer and the bead mixture, 30 μL of 4× Laemmli sample buffer was added to the remaining mixture, and the samples were boiled for 5 minutes. Twenty microliters of the sample volume was loaded for Western blot analysis. Band intensity was quantified using ImageJ software and normalized to total RHOA protein.
Cells were plated on 24-well inserts with 8-μm pore (353097; BD) at 5 × 10 4 /well in serum-free RPMI-1640 media. Medium containing 10% FBS was added to 24-well insert plate (354578, BD) and cells were cultured at 37°C under 5% CO 2 . At 24 hours after plating, remaining cells were removed by gently scrapping the upper chamber with a wet cotton swab. Migrated cells were fixed with 10% formalin for 10 minutes and washed with PBS once. The inserts were soaked in hematoxylin for 1 minute and washed. Membranes were cut from inserts and moved to a glass slide. Mean number of migrated cells was determined by counting 3 high-power fields (200×).
Western Blot Analysis
Cells were lysed using T-PER Tissue Protein Extraction Reagent (Thermo Fisher Scientific) supplemented with protease and phosphatase inhibitors. After cellular debris was removed by centrifugation, and protein quantitation was performed using a BCA Protein Assay Kit (Thermo Fisher Scientific), the protein samples were prepared in Laemmli sample buffer by boiling. The same amount of each protein sample was separated on sodium dodecyl sulfate polyacrylamide gels and transferred onto nitrocellulose membranes using electrophoresis and blotting apparatuses (Bio-Rad Laboratories, Hercules, CA). After blocking using 2.5% bovine serum albumin in PBS with Tween20 (PBS-T; Sigma), we probed membranes using primary antibodies (1:1000) in 3% BSA/PBS-T overnight, followed by a horseradish peroxidase−conjugated secondary antibody (anti-mouse IgG or anti-rabbit IgG; GenDEPOT) (1:5000) for 2 hours. SuperSignal West Pico Chemiluminescent Substrate kit (Thermo Fisher Scientific) was used for detection. C4/monoclonal mouse anti-BAP1 (sc-28383; Santa Cruz Biotechnology, Dallas, TX), 4/monoclonal mouse anti−β-catenin antibody (610154; BD Biosciences), 33 39 24E10/monoclonal rabbit against E-cadherin antibody (3195; Cell Signaling, Danvers, MA), 39 monoclonal mouse anti-RHOA (26007; NewEast Biosciences, Malvern, PA), and C4/monoclonal mouse anti−β-actin (sc-47778; Santa Cruz Biotechnology) were used.
|With deletion||Without deletion|
|RPL29||8 × 10 −7||101.7||161.4||−0.7|
|TMEM115||5 × 10 −5||13||19.1||−0.5|
|1||CDH1||28||9 × 10 −13|
|2||TP53||23||9 × 10 −13|
|3||ARID1A||12||1 × 10 −11|
|4||KRAS||5||3 × 10 −6|
|5||PIK3CA||12||9 × 10 −6|
|6||ERBB3||7||2 × 10 −5|
|7||TGFBR1||7||9 × 10 −5|
|8||FBXW7||8||4 × 10 −4|
|Mutation signature||P a||HR (95% CI)|
a Pr > χ 2 .
|Source||DF||Type III SS||Mean square||F value||Pr > F|
Author names in bold designate shared co-first authorship.
Clinicopathologic Characteristics of Gastric Cancer SamplesSupplementary Table 2
Germline MutationsSupplementary Table 3
Somatic Mutations as Identified by Targeted SequencingSupplementary Table 4
Summary of CDH1 AlterationsSupplementary Table 5
ImmunohistochemistrySupplementary Table 6
Mutation Signature FractionsSupplementary Table 7
Single Nucleotide Variations Recurrent in Diffuse and Intestinal Tumors
Conflicts of interest The authors disclose no conflicts.
Funding This work was supported by the Multi-omic Research Program funded by the Ministry of Science, ICT, and Future Planning in Korea, and by National Cancer Center grants 1510111/1710809.
Author names in bold designate shared co-first authorship.
∗ Authors share co-first authorship
- 1 J. Ferlay ,I. Soerjomataram ,R. Dikshit ,et al. Cancer incidence and mortality worldwide: Sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer. 2015;136 :E359-E386 Crossref
- 2 H. Wu ,J.A. Rusiecki ,K. Zhu ,et al. Stomach carcinoma incidence patterns in the United States by histologic type and anatomic site. Cancer Epidemiol Biomarkers Prev. 2009;18 :1945-1952 Crossref
- 3 K. Chen ,D. Yang ,X. Li ,et al. Mutational landscape of gastric adenocarcinoma in Chinese: implications for prognosis and therapy. Proc Natl Acad Sci U S A. 2015;112 :1107-1112 Crossref
- 4 R. Cristescu ,J. Lee ,M. Nebozhyn ,et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med. 2015;21 :449-456 Crossref
- 5 M. Kakiuchi ,T. Nishizawa ,H. Ueda ,et al. Recurrent gain-of-function mutations of RHOA in diffuse-type gastric carcinoma. Nat Genet. 2014;46 :583-587 Crossref
- 6 The Cancer Genome Atlas Research Network. Comprehensive molecular characterization of gastric adenocarcinoma. Nature. 2014;513 :202-209 Crossref
- 7 K. Wang ,S.T. Yuen ,J. Xu ,et al. Whole-genome sequencing and comprehensive molecular profiling identify new driver mutations in gastric cancer. Nat Genet. 2014;46 :573-582 Crossref
- 8 M.C. Bautista ,S.F. Jiang ,M.A. Armstrong ,et al. Impact of age on clinicopathological features and survival of patients with noncardia gastric adenocarcinoma. J Gastric Cancer. 2014;14 :238-245 Crossref
- 9 H.W. Chung ,S.H. Noh ,J.B. Lim. Analysis of demographic characteristics in 3242 young age gastric cancer patients in Korea. World J Gastroenterol. 2010;16 :256-263 Crossref
- 10 W.F. Anderson ,M.C. Camargo ,J.F. Fraumeni ,et al. Age-specific trends in incidence of noncardia gastric cancer in US adults. JAMA. 2010;303 :1723-1728 Crossref
- 11 N. Wagle ,C. Emery ,M.F. Berger ,et al. Dissecting therapeutic resistance to RAF inhibition in melanoma by tumor genomic profiling. J Clin Oncol. 2011;29 :3085-3096 Crossref
- 12 T.E. Buffart ,B. Carvalho ,E. Hopmans ,et al. Gastric cancers in young and elderly patients show different genomic profiles. J Pathol. 2007;211 :45-51 Crossref
- 13 J. Brasch ,O.J. Harrison ,B. Honig ,et al. Thinking outside the cell: How cadherins drive adhesion. Trends Cell Biol. 2012;22 :299-310 Crossref
- 14 J. Simões-Correia ,J. Figueiredo ,R. Lopes ,et al. E-Cadherin destabilization accounts for the pathogenicity of missense mutations in hereditary diffuse gastric cancer. PLoS One. 2012;7 :e33783
- 15 R.B. Richardson. P53 mutations associated with aging-related rise in cancer incidence rates. Cell Cycle. 2013;12 :2468-2478 Crossref
- 16 J.W. Park ,D.M. Park ,B.K. Choi ,et al. Establishment and characterization of metastatic gastric cancer cell lines from murine gastric adenocarcinoma lacking Smad4, p53, and E-cadherin. Mol Carcinog. 2015;54 :1521-1527 Crossref
- 17 J.W. Park ,S.H. Jang ,D.M. Park ,et al. Cooperativity of E-cadherin and Smad4 loss to promote diffuse-type gastric adenocarcinoma and metastasis. Mol Cancer Res. 2014;12 :1088-1099 Crossref
- 18 S. Nik-Zainal ,L.B. Alexandrov ,D.C. Wedge ,et al. Mutational processes molding the genomes of 21 breast cancers. Cell. 2012;149 :979-993 Crossref
- 19 G. Corso ,J. Carvalho ,D. Marrelli ,et al. Somatic mutations and deletions of the E-cadherin gene predict poor survival of patients with gastric cancer. J Clin Oncol. 2013;31 :868-875 Crossref
- 20 X. Yanjun ,C. Wenming ,Y. Lisha ,et al. Detection of CDH1 gene variants in early-onset diffuse gastric cancer in Chinese patients. Clin Lab. 2014;60 :1823-1830
- 21 S. Kim ,J.W. Chung ,T.D. Jeong ,et al. Searching for E-cadherin gene mutations in early onset diffuse gastric cancer and hereditary diffuse gastric cancer in Korean patients. Fam Cancer. 2013;12 :503-507 Crossref
- 22 J.T. Bacani ,M. Soares ,R. Zwingerman ,et al. CDH1/E-cadherin germline mutations in early-onset gastric cancer. J Med Genet. 2006;43 :867-872 Crossref
- 23 T. Wiesner ,I. Fried ,P. Ulz ,et al. Toward an improved definition of the tumor spectrum associated with BAP1 germline mutations. J Clin Oncol. 2012;30 :e337-e340 Crossref
- 24 A. Moeini ,D. Sia ,N. Bardeesy ,et al. Molecular pathogenesis and targeted therapies for intrahepatic cholangiocarcinoma. Clin Cancer Res. 2016;22 :291-300 Crossref
- 25 K.H. Ventii ,N.S. Devi ,K.L. Friedrich ,et al. BRCA1- associated protein-1 is a tumor suppressor that requires deubiquitinating activity and nuclear localization. Cancer Res. 2008;68 :6953-6962 Crossref
- 26 P. Rodrigues ,I. Macaya ,S. Bazzocco ,et al. RHOA inactivation enhances Wnt signalling and promotes colorectal cancer. Nat Commun. 2014;5 :5458-5489 Crossref
- 27 E. Sahai ,C.J. Marshall. ROCK and Dia have opposing effects on adherens junctions downstream of Rho. Nat Cell Biol. 2002;4 :408-415 Crossref
- 28 M. Rohde ,J. Richter ,M. Schlesner ,et al. Recurrent RHOA mutations in pediatric Burkitt lymphoma treated according to the NHL-BFM protocols. Genes Chromosomes Cancer. 2014;53 :911-916 Crossref
- 29 H. Wang ,V. Radjendirane ,K.K. Wary ,et al. Transforming growth factor beta regulates cell-cell adhesion through extracellular matrix remodeling and activation of focal adhesion kinase in human colon carcinoma Moser cells. Oncogene. 2004;23 :5558-5561 Crossref
© 2017 AGA Institute