KRAS status is related to histological phenotype in gastric cancer: results from a large multicentre study

Background Gastric cancer (GC) is histologically a very heterogeneous disease, and the temporal development of different histological phenotypes remains unclear. Recent studies in lung and ovarian cancer suggest that KRAS activation (KRASact) can influence histological phenotype. KRASact likely results from KRAS mutation (KRASmut) or KRAS amplification (KRASamp). The aim of the study was to investigate whether KRASmut and/or KRASamp are related to the histological phenotype in GC. Methods Digitized haematoxylin/eosin-stained slides from 1282 GC resection specimens were classified according to Japanese Gastric Cancer Association (JGCA) and the Lauren classification by at least two observers. The relationship between KRAS status, predominant histological phenotype and clinicopathological variables was assessed. Results KRASmut and KRASamp were found in 68 (5%) and 47 (7%) GCs, respectively. Within the KRASmut and KRASamp cases, the most frequent GC histological phenotype was moderately differentiated tubular 2 (tub2) type (KRASmut: n = 27, 40%; KRASamp: n = 21, 46%) or intestinal type (KRASmut: n = 41, 61%; KRASamp: n = 23, 50%). Comparing individual histological subtypes, mucinous carcinoma displayed the highest frequency of KRASmut (JGCA: n = 6, 12%, p = 0.012; Lauren: n = 6, 12%, p = 0.013), and KRASamp was more frequently found in poorly differentiated solid type (n = 12, 10%, p = 0.267) or indeterminate type (n = 12, 10%, p = 0.480) GC. 724 GCs (57%) had intratumour morphological heterogeneity. Conclusions This is the largest GC study investigating KRAS status and histological phenotype. We identified a relationship between KRASmut and mucinous phenotype. The high level of intratumour morphological heterogeneity could reflect KRASmut heterogeneity, which may explain the failure of anti-EGFR therapy in GC. Electronic supplementary material The online version of this article (10.1007/s10120-019-00972-6) contains supplementary material, which is available to authorized users.


Introduction
Gastric cancer (GC) is histologically a very heterogeneous disease, and this is reflected in the numerous proposed histological classification schemes [1]. The temporal development of different histological phenotypes in GC remains unclear. Recent studies suggest that Kirsten Rat Sarcoma Viral Oncogene Homolog (KRAS) activation and downstream signalling can impact on the properties and functions of the tumour microenvironment [2], and thus may influence histological phenotype. Likely mechanisms of KRAS activation include KRAS mutation (KRASmut) and KRAS amplification (KRASamp) [3].
Mutations in KRAS have been identified in many human cancers and result in the constitutive activation of KRAS and the receptor tyrosine kinase (RTK) pathway [4]. The frequency of KRASmut is variable across different cancer types, with the highest frequency in pancreatic cancer (90%) followed by colon (34.6%), lung (16.5%) and ovarian (11%) cancer and the lowest frequencies in cervical (6.6%), prostate (5%) and oesophageal cancer (2%) [5]. In a review of the literature we identified, on average, only 6.5% of GC have a KRASmut [6]. In colorectal cancer, routine testing for KRASmut is now implemented as a predictor of response to anti-epidermal growth factor receptor (EGFR) therapy [7].
Several studies have demonstrated a relationship between KRASmut status and histological phenotype in lung and ovarian cancer. In the subgroup of invasive mucinous adenocarcinoma of the lung, KRAS is mutated in up to 86% of cases [8]. In ovarian cancer, KRASmut has been identified in almost all cases with a mucinous histological phenotype [9]. The relationship between KRASmut status and histological phenotype in GC remains to be clarified [6].
The reported frequency of KRASamp is 1-9% in GC [10][11][12][13][14][15][16]. There are no reports of a relationship between KRAS DNA copy number and histological phenotype in other cancer types and in GC it has not been investigated in a large study. There is increasing recognition of the clinical importance of KRASamp in GC. KRASamp is also associated with a worse survival [3,10,12], whereas KRASmut do not appear to influence survival of GC patients [17].
Recently, image analysis on lung cancer haematoxylin and eosin (H&E) stained sections using deep learning was predictive of mutation status [18], thus suggesting that morphological phenotype is reflective of molecular phenotype. Investigating the relationship between KRAS activation by KRASmut and/or KRASamp and histological phenotype may provide some insight into gastric adenoma-carcinoma sequence progression and the origin of histological heterogeneity. Based on the studies in lung and ovarian cancer, we hypothesise that KRAS activation influences histological phenotype and is associated with a mucinous phenotype in GC. This would suggest that KRAS activation is an early event in GC, occurring before the phenotype is determined.
The aim of this multicentre GC study was to investigate the relationship of KRAS activation status (KRASmut and/or KRASamp) with the histological phenotype in a large series of GCs from UK, Japan and The Cancer Genome Atlas (TCGA). In addition, the relationship between KRAS status, clinicopathological variables, survival and microsatellite instability status was assessed.

Kanagawa Cancer Center Hospital (KCCH), Yokohama, Japan
This cohort included 250 patients with TNM stage II/III GC who underwent potentially curative surgery at Kanagawa Cancer Center Hospital (Yokohama, Japan) between 2001 and 2010. One hundred and six (43%) patients were treated with surgery alone, 108 (43%), 22 (9%), 14 (6%) patients received S-1, tegafur-uracil or S-1 combined with other cytotoxic drug therapy, respectively. Demographical, clinical and pathological data were retrieved from hospital records. The study was approved by the Local Research Ethics Committee.

Leeds Teaching Hospitals NHS Trust (LTHT), Leeds, UK
This cohort included 277 patients with GC who underwent potentially curative surgery at the Department of Surgery, Leeds General Infirmary (Leeds, UK), between 1970 and 2004. Seven (3%) patients were treated by chemotherapy followed by surgery and the remaining 270 (98%) by surgery alone. Clinical and pathological data were retrieved from histopathology reports, electronic patient hospital records and the Northern and Yorkshire Cancer Registry. The study was approved by the Local Research Ethics Committee (LREC no. CA01/122).

The Cancer Genome Atlas
The TCGA stomach adenocarcinoma (STAD) clinicopathological and molecular dataset of 295 patients was obtained from the publically available TCGA database portal [19].

Tokyo Metropolitan Geriatric Hospital and Institute of Gerontology (TMGH), Tokyo, Japan
This cohort included 420 patients with 460 GC who were treated by surgery in the Tokyo Metropolitan Geriatric Hospital between 2000 and 2008. Three hundred and eighty patients had single carcinoma, and 36 had 2 or more carcinomas. Patients with Lynch syndrome were excluded from the current study. None of the patients underwent neoadjuvant chemotherapy. Histopathological examination and medical research were performed with informed written consent by the patients, and this work was approved by the ethics committee of the Tokyo Metropolitan Geriatric Hospital (#230,225, R16-23).

Histopathological classification
pT and pN stages were reported according to 7th edition of the UICC TNM classification for GC [20].
In all cohorts, H&E stained formalin fixed paraffin embedded (FFPE) tissue sections from the resection specimens were reviewed. In the KCCH and LTHT cohorts, H&E stained slides were scanned at 40 × magnification using an Aperio AT2 scanner for review. In the TCGA cohort, H&E stained slides were viewed online using the cancer digital slide archive (https ://cance r.digit alsli dearc hive.net/). In the TMGH cohort, classification was performed using the glass slides.
Histological classification according to JGCA scheme was performed [21]. Mucinous carcinoma were defined as tumour cells located in mucinous pools comprising an area greater than 50% of the total tumour. GC were classified as signet-ring cell carcinoma when signet-ring cells were present in more than 50% of the tumour volume. In cases where more than one histological phenotype was identified, the most predominant phenotype was recorded, and these GCs were categorised as heterogeneous. JGCA classification was converted to Lauren classification [22] according to Table 1. As there is no Lauren classification for mucinous GC, we retained mucinous carcinomas as a separate category to distinguish them from other histological types.

DNA extraction
The area with the highest tumour cell density was identified on H&E stained sections and the whole tumour area, irrespective of subregions with different histological phenotypes was microdissected after staining with Shandon instant haematoxylin (Thermo Scientific, Cheshire, UK) using a sterile surgical blade. Tumour DNA from FFPE material was extracted from KCCH and LTHT GCs using the QIAmp DNA Micro Kit (Qiagen, Hilden, Germany) as previously described [23]. DNA concentration was measured by ND-100 Spectrophotometer (Labtech International) and samples were diluted using Tris-EDTA buffer. In the TMGH cohort, DNA was extracted using a phenol-chloroform procedure as described previously [24].

KRAS gene copy number and data analyses
KRAS copy number status was investigated in KCCH, LTHT and TCGA cohorts.
In the KCCH and LTHT cohort, KRAS gene copy number was determined by multiplex ligation-dependent probe amplification (MLPA) using the Salsa-FAM-labeled MLPA reagent kit and probemix P458-A1 or the updated version -B1 (MRC Holland, Amsterdam, The Netherlands) as previously described [25]. For further details on the KRAS probes included in this probemix see Supplementary Table 1. Fragment analysis of the MLPA reaction product was performed using capillary electrophoresis ABI-3130 XL (Applied Biosystems, California, USA) as previously described [25]. Failed experiments were repeated at least twice before a case was finally excluded from the analyses.
KRAS DNA copy number data from 237 KCCH GC has been previously published [25], but was re-analysed using a different methodology in the current study. The output files (FSA files) from the sequencer were initially imported into Coffalyser.net for fragment analysis and results were exported as csv files. Subsequent analyses were performed using the MLPAInter method, as previously described [26], implemented in R. Samples were normalised per batch using reference samples processed in each batch. Quality control was performed to exclude samples with low overall intensity, with a large difference in intensity between short probes and long probes, with low intensity of denaturation controls, or high within gene variation, defined as the average of the standard deviation of log-transformed values. Final values were calculated by averaging the peak height of each probe and then averaging the results of replicates. Copy number thresholds were set based on previously published studies [25,27,28], with a DNA copy number > 1.31 categorised as amplification. This analysis was performed separately for KCCH and LTHT cohorts. In TCGA, KRASamp were determined by array-based somatic copy number analysis [29]. Level 3 copy number segmentation data was downloaded from the TCGA data portal [19] and used to estimate copy number for KRAS. Based on previous studies, a LogRatio > 0.4 was categorised as amplification [30].

KRAS mutation status
KRASmut data from a previous study were available for 230 KCCH and 275 LTHT GC patients [17]. KRASmut testing was performed on an additional 12 KCCH GCs as previously described [17]. In TCGA, KRASmut status was determined by whole-exome sequencing [29] and results were downloaded from the TCGA database portal [19] for 289 patients. In the TMGH cohort, KRAS (codon 12 and 13) was examined by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP), using primers and methods previously described [31,32].

Microsatellite instability (MSI) status
Immunohistochemistry of DNA mismatch repair proteins were used as a surrogate marker of MSI status. Results for MLH1, MSH2, MSH6 and PMS2 were available from 230 KCCH GCs, and MLH1 and MSH2 from 253 LTHT GCs from a previous study [17]. MLH1, MSH2, MSH6 and PMS2 immunohistochemistry was performed on additional 13 GCs from the KCCH cohort for this study, as previously described [17].
In TCGA, MSI was determined by a DNA based MSI-Mono-Derived-Dinucleotide Assay using four mononucleotide repeat loci and three dinucleotide repeat loci using a multiplex fluorescent-labeled PCR and capillary electrophoresis [29]. Results were obtained from the TCGA database portal [19] for 295 GC patients. MSI-low GCs were grouped with microsatellite stable (MSS) GCs for further analyses following current guidelines [33].

Statistical analyses
All statistical analyses were performed using SPSS software version 23 (SPSS Inc., Chicago, III). The relationship between KRASmut or KRASamp and clinicopathological variables (age, gender, depth of invasion (pT), lymph node status (pN), TNM stage, Lauren classification [22], JGCA classification [21], MSI status and morphological heterogeneity status) was assessed using Chi-squared or Fisher's exact test. The relationship between KRASmut and survival in LTHT and KCCH cohorts has been published previously [17]. Combining all cohorts, the relationship between KRASmut or KRASamp and 5-year overall survival was analysed using the Kaplan-Meier method and differences were assessed using the log rank test. A p value of < 0.05 was considered significant.

Patient characteristics
The median (range) age of GC patients was as follows; KCCH: 65 years (35- Table 2.

Discussion
This is the largest multicentre study to date to investigate the relationship between KRAS activation by mutation and/ or amplification and histological phenotype in GC. The frequency of KRASamp (7%) was slightly higher than that of KRASmut (5%) which is consistent with other GC studies [10,11,37]. The higher frequency of KRASmut in the TCGA GC cohort compared to the other cohorts could be related to the methodology as TCGA used whole-exome sequencing to test non-hotspot regions, whereas other studies used less-sensitive Sanger sequencing/PCR-RFLP. We found KRASamp and KRASmut were exclusive in > 99% of GC, which is consistent with previous reports [11][12][13]38].
The relationship between KRASmut and histological phenotype has not been investigated in great detail and previous studies were limited by small sample sizes and hence lack of statistical power [6]. In our study, we identified a relationship between KRASmut and mucinous histological phenotype, which is concordant with higher frequencies of KRASmut being reported in mucinous lung [8], ovarian [9] and colorectal cancer [39,40]. However, due to the relatively low frequency of GC with mucinous phenotype and KRASmut (12%), it would not be feasible to use the presence of a mucinous phenotype as a predictor for the presence of a KRASmut in GC. The main component of mucinous GCs is extracellular mucin, which consists of high molecular weight glycoproteins regulated by expression of the MUC2, MUC5AC and MUC6 genes in humans [41]. In mouse models with constitutively activated KRAS in the stomach, irregular MUC4+ cells were found with abnormal mucins confirmed by Alcian-blue staining [42]. Interestingly, our study suggests a relationship between KRASmut and mucinous phenotype, which is characterised by extracellular mucin, but is not related to signet-ring cell type GC, which is characterised by intracellular mucin. Our study confirmed the relationship between KRASmut and the presence of MSI, which our group and others have described previously in a smaller GC cohort [43,44].
The prognostic significance of KRASmut in GC remains controversial [6]. In our study, there was no association with the presence of KRASmut and survival. Interestingly, in lung and colorectal cancer, KRASmut has been associated with a poor prognosis [45,46], whereas in ovarian cancer, KRASmut has been associated with an improved prognosis [47].
The relationship between KRASamp and clinicopathological variables, including histological phenotype in cancer is not well studied. In GC, we found no statistically significant relationship between KRASamp and histological phenotype, or any other clinicopathological variables. In contrast, others found that the presence of KRASamp is associated with a poor prognosis in GC [3,10,12]. This difference might be due to case selection and methodology used. In our study, we used the JGCA scheme for the histological classification of GC and performed a conversion to the Lauren scheme, which is the most widely used histological classification system in Western countries [22]. Previous studies investigating the relationship between KRASmut and histological phenotype performed classification according to the Lauren scheme [6], for which there is no separate category for mucinous GC. The relatively large number of GCs classified as indeterminate according to the Lauren scheme comes from conversion from the JGCA por1 histological phenotype. Direct classification according to the Lauren scheme, would likely result in a higher proportion of GCs classified as either intestinal or diffuse.
In colorectal cancer, KRASmut is known to be an early event in the progression from normal colonic epithelial cell to adenoma, and finally to carcinoma [48]. The evidence of sequential development by accumulation of genetic alterations, including KRASmut, is still controversial in GC [49][50][51]. We were unable to make any comments regarding the role of KRAS activation in gastric carcinogenesis in our cohort as we did not investigate precancerous lesions in the current study. However, evidence from mouse models suggest that KRASmut is one of the key molecular alterations involved in the development of stomach dysplasia [52] and GC [53]. Based on the evidence from other cancer types that KRASmut influence the progression of a mucinous histological phenotype, we therefore speculate based on our results, that KRASmut in GC is an early event in GC development, whereas KRASamp is likely to be a late event occurring after the histological phenotype has been established. This would correspond with experiments in mice expressing oncogenic KRAS in combination with E-cadherin and p53 loss, which resulted in a rapid progression of GC compared to wild type mice [53].
Our study has some limitations. This is a retrospective study. Histological phenotyping was performed on a single slide. Given the high frequency of intra-tumoural morphological heterogeneity in this study and the previously reported intra-tumoural heterogeneity in KRASmut status in GC [54], the sensitivity of some of the techniques used in the current study may not be sufficient to detect KRAS activation in subclones of tumour cells. As we did not perform microdissection of tumour subregions, we cannot comment on KRAS status heterogeneity within the same tumour. Furthermore, we used different techniques for DNA extraction, KRASmut status analysis and MSI analysis in different cohorts included in the current study, each with differing sensitivities [55,56].
In summary, we identified a relationship between KRASmut and mucinous histological phenotype in GC. The high level of intratumour morphological heterogeneity could reflect KRASmut heterogeneity, which may explain the failure of anti-EGFR therapy in GC.