Losses of Chromosome 5q and 14q Are Associated with Favorable Clinical Outcome of Patients with Gastric Cancer

DNA copy number profiling was used to identify subgroups of gastric cancer patients with different clinical outcomes. A subgroup of gastric cancer patients, marked by losses on chromosomes 5q11.2-q31.3 and 14q32.11-q32.33 or low heat shock protein 90 expression, who had an excellent clinical outcome after surgery alone was identified.

Abstract

Purpose.

To improve the clinical outcome of patients with gastric cancer, intensified combination strategies are currently in clinical development, including combinations of more extensive surgery, (neo)adjuvant chemotherapy, and radiotherapy. The present study used DNA copy number profiling to identify subgroups of patients with different clinical outcomes. We hypothesize that, by identification of subgroups, individual treatment strategies can be selected to improve clinical outcome and to reduce unnecessary treatment toxicity for patients with gastric cancer.

Experimental Design.

DNA from 206 gastric cancer patients was isolated and analyzed by genomewide array comparative genomic hybridization. DNA copy number profiles were correlated with lymph node status and patient survival. In addition, heat shock protein 90 (HSP90) expression was analyzed and correlated with survival in 230 gastric cancer patients.

Results.

Frequent (>20%) DNA copy number gains and losses were observed on several chromosomal regions. Losses on 5q11.2-q31.3 and 14q32.11-q32.33 (14% of patients) were correlated with good clinical outcome in univariate and multivariate analyses, with a median disease-free survival interval of 9.2 years. In addition, loss of expression of HSP90, located on chromosome 14q32.2, was correlated with better patient survival.

Conclusion.

Genomewide DNA copy number profiling allowed the identification of a subgroup of gastric cancer patients, marked by losses on chromosomes 5q11.2-q31.3 and 14q32.11-q32.33 or low HSP90 protein expression, with an excellent clinical outcome after surgery alone. We hypothesize that this subgroup of patients most likely will not benefit from (neo)adjuvant systemic treatment and/or radiotherapy, whereas anti-HSP90 therapy may have clinical potential in patients with HSP90-expressing gastric cancer, pending validation in an independent dataset.

Introduction

Despite a declining incidence of gastric cancer, it still is the second most common cause of cancer death worldwide [1]. In The Netherlands, it ranks fifth as a cause of cancer death, with >2,000 new cases each year [2]. Surgery is the only curative treatment option, but the results of gastrectomy depend on the disease stage. To improve the clinical outcome of patients with gastric cancer, (perioperative) treatment strategies, including more extensive surgery, (neo)adjuvant chemotherapy, and chemoradiotherapy, are in clinical development [3]. Previous studies have evaluated surgery with limited lymph node dissection (D1) versus extended lymph node dissection (D2) [4, 5]. Fifteen years of follow-up of the Dutch Gastric Cancer D1/D2 Trial, in which gastric cancer patients were randomly assigned to undergo D1 or D2 resection, revealed that D2 resection was associated with a lower locoregional recurrence rate and fewer cancer-related deaths. However, D2 resection was associated with higher postoperative morbidity and mortality rates in that trial [6].

The development of new molecular markers can be used to identify subgroups of tumors with different biological and also clinical behavior. Altered DNA copy number status or expression of several genes, such as CD44, p27^kip1, Bax, VEGF, nm23, TNFRSF6B and several matrix metalloproteinases (MMPs), is associated with lymph node metastasis in gastric cancer [7–17]. In addition, genomewide mRNA expression analyses have been performed in order to determine genetic signatures correlated with lymph node status in gastric cancer patients [18, 19]. In a small series of gastric cancer patients, specific gene-expression profiles could be identified for lymph node metastasis, and expression of Bik, aurora kinase B, and eIF5A2 were shown to predict lymph node status. Thus far, none of the identified markers are being used in the clinic for treatment guidance and the gold standard to predict the prognosis of a patient is still based on tumor–node–metastasis (TNM) staging.

In routine laboratory practice, DNA-based biomarker tests have advantages over mRNA-based ones because DNA is more stable and therefore easier to work with in routine laboratory practice, whereas mRNA expression of genes can be more easily affected by differences in handling procedures of the tissue samples [20]. Moreover, in many instances, formalin-fixed, paraffin-embedded tissue samples are the only material available, and from these good quality DNA can be more reliably obtained than mRNA.

In a previous pilot study, DNA copy number profiles of gastric cancers were correlated with clinical outcome [21]. In the present study, we assessed whether or not DNA copy number profiles were correlated with clinical outcome in a large series of patients with primary gastric cancer, and we aimed to identify subgroups of patients who might benefit from a differential therapeutic strategy.

Materials and Methods

Material for Genomic Instability Analysis

In total, 206 gastric adenocarcinoma tissue samples were collected for this study for the analysis of genomic instability patterns. Of these, 122 gastric cancers were selected from the Dutch D1/D2 trial [4] and 84 gastric cancers were selected from the archives of the Leeds General Infirmary (Leeds, UK) based on the DNA quality of the tumor samples as previously described [22]. All patients from the Leeds archives had undergone a D2 tumor resection. Patients did not receive (neo) adjuvant chemotherapy or radiotherapy. An overview of the clinicopathological patient data is given in Table 1. Ethical approval for the study was obtained from the Dutch and Leeds research ethics committees.

Table 1.

Clinicopathological data from the 206 gastric adenocarcinoma patients

DNA Isolation

DNA isolation was performed as described previously, using commercially available DNA isolation kits (QIAamp DNA Minikit; Qiagen, Westburg, Leusden, The Netherlands, and QIAamp Microkit; Qiagen, Hilden, Germany) [23, 24]. Briefly, areas containing ≥70% tumor cells were marked on a 4-μm section stained with hematoxylin and eosin. Adjacent serial sections of 10 μm were cut and tumor tissue was macrodissected using a needle. DNA was isolated after an overnight incubation at 37°C with sodium thiocyanate (1 M) and poteinase K treatment. DNA concentrations were measured using a Nanodrop ND-1000 spectrophotometer (Isogen, IJsselstein, The Netherlands). DNA quality was assessed using isothermal amplification as described before [22].

Genomic DNA isolated from peripheral blood from 18 healthy females was pooled and used as control reference DNA.

Array Comparative Genomic Hybridization Procedure

Array comparative genomic hybridization (array CGH) was performed as described previously [23, 25] on a bacterial artificial chromosome (BAC) array comprised of a series of 3,000 BAC clones with ∼1 Mb genomewide resolution and additional clones for areas on chromosomes 6, 8q, 11, 13q, and 20q, resulting in a total of ∼5,000 BAC or PAC clones. Details of the BAC array platform and the labeling and hybridization procedures have been described before [23]. Image analysis and feature extraction were performed using BlueFuse 3.4 software (BlueGnome, Cambridge, U.K.). Spots with a BlueFuse quality flag <1 or with a confidence level <0.1 were excluded from further analyses. For each spot, the tumor to normal fluorescence ratio was calculated and normalized against the mode of the ratios of all autosomes. Clones were mapped according to the positions from the University of California–Santa Cruz May 2004 freeze of the Human Golden Path. Array CGH profiles with a median absolute deviation of chromosome 2 (MAD2) >0.18, as a measure of noise inherent to the measurement and used as indicator of the array CGH quality, were excluded from further analyses [22].

DNA copy number gains and losses were defined using the R package CGH call [26]. CGH region analysis was used to compress the data, using a threshold for average error rate of 0.001 [27]. High-level amplifications were defined when two or more consecutive clones showed log₂ tumor-to-normal ratios >1.

Array data can be accessed using the Gene Expression Omnibus [28].

Microsatellite Instability Analysis

A microsatellite instability (MSI) analysis was performed as previously described [29] using the MSI Analysis System (MSI Multiplex System, Version 1.1, Promega Corp., Madison, WI) consisting of five nearly monomorphic mononucleotide markers (BAT-25, BAT-26, NR-21, NR-24, MONO-27). Separation of polymerase chain reaction (PCR) products was performed by capillary electrophoresis using an ABI 3130 DNA sequencer (Applied Biosystems, Foster City, CA) and analysis was performed using GeneScan 3100 (Applied Biosystems, Foster City, CA). An internal lane size standard was added to the PCR samples for accurate sizing of alleles and to adjust for run-to-run variations. When all markers were stable, the tumor was interpreted as microsatellite stable (MSS). The tumor was interpreted as MSI-low when one marker was instable and as MSI-high when two or more markers showed instability.

Tissue Microarrays and Immunohistochemistry

Tissue microarrays (TMAs) were constructed from a largely independent series of 290 gastric adenocarcinomas obtained from the histopathology archive of the Leeds General Infirmary (Leeds, U.K.). For each tumor, three 0.6-mm cores from different locations within the tumor were included in the array. Sections of 4 μm were used for immunohistochemistry. After deparaffination in xylene and rehydration through graded alcohol to water, the TMA sections were blocked in methanol containing 0.3% hydrogen peroxide with peroxidase. Heat shock protein 90 (HSP90) immunohistochemistry was performed using a rabbit polyclonal primary antibody in a 1:50 dilution (Cell Signaling Technology, Danvers, MA) after antigen retrieval by microwave in citrate buffer (10 mM; pH, 6.0). This was followed by Powervision Plus (Immunologic, Duiven, The Netherlands) incubation, and subsequently staining was visualized with diaminobenzidine (SIGMAFAST™ 3,3′-diaminobenzidine tablets; Sigma-Aldrich, St Louis, MO).

Cytokeratin staining was performed using a mouse monoclonal antibody in a 1:10 dilution (anticytokeratin CAM5.2; Becton and Dickson Biosciences, Franklin Lakes, NJ) after antigen retrieval using pepsin. Detection was performed with a standard avidin–biotinylated peroxidase complex. Incubation without primary antibody was used as a negative control. Tissues shown to be positively stained in previous immunohistochemistry analyses were used as positive controls. Counterstaining was done using Mayer's hematoxylin.

HSP90 immunohistochemistry results were scored by a pathologist (N.C.T.v.G.) according to the overall intensity of the staining in tumor cells as negative, mildly positive, moderately positive, or strongly positive. If the scores were inconsistent among different cores of the same tumor, the highest score of the three individual cores was used as the final score. The immunohistochemistry scores negative and mildly positive were combined for further analyses. TMA sections stained with CAM5.2 were used to confirm the epithelial phenotype of cells.

Statistical Analysis

Box plots and cross tables were used for descriptive statistics of continuous and categorical variables, respectively. For calculating the significance of differences in copy number ratios of chromosomal regions between lymph node–positive and lymph node–negative gastric cancers, CGH test was used [30]. Clones with a false discovery rate (FDR) <0.20 were considered to be significant. The significance of differences for continuous variables between two categories was tested using the Mann-Whitney U-test. Significant differences for categorical variables between two categories were tested using a χ² test. Univariate survival analysis was performed using the Kaplan–Meier method, using the survival length starting from the date of surgery for the primary tumor to the date of death resulting from gastric cancer or recurrence (event) or the last date of clinical follow-up (censored). Differences in survival lengths were analyzed using the log-rank test. For determining hazard ratios (HRs) and for the multivariate survival analysis, Cox regression was used. All standard statistical analyses were conducted with R 2.3.5 (R Development Core Team) and SPSS 18.0 for Windows (SPSS Inc., Chicago, IL). p-values <.05 were considered significant.

Results

DNA Copy Number Aberrations in Gastric Cancers

Analysis of 206 gastric cancers using CGH revealed 23 (11.2%) samples with a MAD2 value ≥0.18. These were excluded from further analyses because of insufficient data quality, leaving 183 gastric cancers (110 from the Dutch D1/D2 trial and 73 from the archives of the Leeds General Infirmary) for the analysis of DNA copy number aberrations (supplemental online Fig. 1). The mean percentage of clones showing a gain or loss was 19.4% (range, 0%–56.8%), with 12.7% (range, 0%–30.3%) showing gains and 6.7% (range, 0%–27.4%) showing losses. Aberrations found in >20% of cases were gains on chromosomes 1p, 6p, 7p, 7q, 8q, 11q, 13q, 16p, 16q, 17q, 19p, 19q, 20p, 20q, 21q, and 22q and losses on chromosomes 4p, 4q, 6p, 6q, 9p, 13q, and 21q (Table 2). An overview of all gains and losses is shown in Figure 1.

Table 2.

Frequent (>20%) gains and losses in the 183 gastric cancers analyzed by array comparative genomic hybridization with a median absolute deviation of chromosome 2 value <0.18

Abbreviations: BAC, bacterial artificial chromosome.

Figure 1.

Frequencies of gains and losses throughout the genome in 183 gastric cancers with a median absolute deviation of chromosome 2 value <0.18. Clones are sorted by position on chromosome (1–22). Vertical lines represent transitions between chromosomes; dashed-vertical lines mark centromere positions.

Of the 183 tumor tissue samples analyzed for DNA copy number aberrations in this study, 118 tumors (64.5%) showed high-level amplifications. The mean number of amplifications per tumor was 2.0 (range, 0–21). No differences in T-stage, N-stage, M-stage, or survival were observed between cancers with and without amplifications (p = not significant [ns]). In total, four patients showed distant metastases at the time of their operation (M1), 23 patients had tumor involvement at microscopic resection margins (R1), and four patients had both M1 and R1 disease. After excluding those patients from the analyses, there was still no significant association observed between TNM stage and survival (p = ns).

Of these 183 gastric cancers, microsatellite status was available for 181. Sixteen (8.8%) cancers were MSI and 165 (91.1%) were MSS. No differences in T-stage, N-stage, or survival were observed between MSS and MSI gastric cancers (p = ns). Data on tumor location were available for 173 gastric cancers. In 26, 40, and 93 patients the tumor was located in the cardia, corpus, and antrum, respectively. In 14 patients, the tumor was located throughout the whole stomach. There was no correlation between tumor location and MSI status. Results did not change after excluding M1 and R1 patients (p = ns).

Lymph Node–Negative Gastric Cancers Show Losses of Chromosomes 5q, 10q, and 14q at Higher Frequencies Than Lymph Node–Positive Gastric Cancers

Of the 183 patients with gastric cancer, significantly better survival outcomes were observed in patients with lymph node–negative gastric cancer than in patients with lymph node–positive gastric cancer, with disease-free survival rates of 75% and 36%, respectively (p < .001; log rank, 25.4; HR, 3.91; 95% confidence interval [CI], 2.21–6.92) (Fig. 2). The median survival times were 8.6 years and 1.7 years, respectively. No differences were observed in gender, age, and histological tumor type between the two tumor groups. Depth of tumor invasion (T-stage) was significantly correlated with lymph node status (N-stage) (p < .001). After excluding M1 and R1 patients, leaving 152 patients (supplemental online Fig. 1), results were similar. Lymph node–negative gastric cancer patients still had a significantly better survival outcome than lymph node–positive gastric cancer patients (p < .001; log rank, 24.0; HR, 4.64; 95% CI, 2.36–9.10), and lymph node status was significantly correlated with T-stage (p < .001). No significant correlation was observed between histological tumor type and survival or between type of operation (D1 versus D2) and survival (p = ns).

Figure 2.

Kaplan–Meier survival plot of 55 patients with lymph node–negative gastric cancer and 128 patients with lymph node–positive gastric cancer. Patients without lymph node metastasis have a significantly better survival outcome than patients with lymph node–positive gastric cancer (p < .001).

0, lymph node–negative gastric cancer; 1, lymph node–positive gastric cancer.

In the 183 gastric cancers, there was no significant difference in the mean percentages of events (gains and losses combined), gains, or losses between lymph node–positive and lymph node–negative gastric cancers. The mean percentage of events was 19.8% (range, 0%–48.3%), with 12.4% (range, 0%–29.5%) showing gains and 7.4% (range, 0%–26.8%) showing losses for the lymph node–negative gastric cancers and 19.3% (range, 0.2%–56.8%), 12.9% (range, 0%–30.3%), and 6.4% (range, 0%–27.4%) for the lymph node–positive gastric cancers, respectively (p = ns). Twenty-one of 55 (38.2%) lymph node–negative gastric cancers did not show any amplification and 44 of 128 (34.4%) lymph node–positive gastric cancers did not show amplifications (p = ns). The mean numbers of amplifications for lymph node–negative and lymph node–positive gastric cancers were 1.9 (range, 0–10) and 2.0 (range, 0–21), respectively (p = ns).

Univariate analysis with correction for multiple comparisons by CGH test identified 18 chromosomal regions that were significantly different between lymph node–positive and lymph node–negative gastric cancers. Of these regions, 13 were located on chromosome 5q11-q35.1, three were location on chromosome 10q11.23-q21.3, and two were located on chromosome 14q32.11-q32.33. All regions showed significantly more frequent losses in lymph node–negative gastric cancers than in lymph node–positive gastric cancers. A detailed overview of the significantly different regions, including FDRs, is shown in Table 3.

Table 3.

Overview of the chromosomal regions in genomic order that show significant differences in DNA copy number between lymph node–positive and lymph node–negative gastric cancers, analyzed by CGH test [29]

Chromosomal regions including false discovery rates (FDRs) and percentages of lymph node–negative (N0) and lymph node–positive (N1) gastric cancers having a loss of each region are shown.

Extensive unsupervised (hierarchical cluster analysis by means of weighted clustering of called array CGH data [31]) and supervised (prediction analysis for microarrays [PAM] and in-house developed dedicated algorithms) multivariate analyses were unable to identify clinically relevant subgroups or generate a multivariate classifier for prediction of survival of patients (data not shown).

Losses on Chromosome 5q and 14q Correlate with Better Survival

After excluding patients with M1 and R1 disease, 152 patients were left to study the association of the chromosomal regions that were significantly correlated with lymph node status and with the disease-free survival rate. Losses of 11 of these 18 chromosomal regions that significantly differed between lymph node–positive and lymph node–negative gastric cancers, that is, 5q11.2, 5q11.2-q12.1, 5q12.1-q13.1, 5q13.2-q13.1, 5q14.1-q14.2, 5q14.3, 15q14.3-q21.1, 5q21.2-q23.2, and 5q31.1-q31.3 on chromosome 5q and 14q32.11-q32.12 and 14q32.13-q32.33 on chromosome 14q, were also found to be related to better patient outcome. An overview of the regions that were correlated with better patient outcome, including their p-values and log-rank values, is shown in supplemental online Table 1. Patients with gastric cancers showing a loss of one of these 11 regions had a significantly better survival outcome than patients without such a loss in their tumor DNA, with disease-free survival rates of 86% and 49%, respectively (p = 0.002; log rank, 9.76; HR, 0.19) (Fig. 3), median survival times of 9.2 years and 2.3 years, respectively, and a mean follow-up duration of 6.0 years (range, 0.5–16.8 years). When comparing the 22 gastric cancers with a loss of one of these regions (14% of the total) with the 130 cancers without a loss of any of these regions, 16 of 22 (73%) were lymph node–negative gastric cancers and six of 22 (27%) were lymph node–positive gastric cancers (χ² p < .001). Of the 130 cancers without a loss of any of these regions, 35 were lymph node negative (27%) and 95 (73%) were lymph node positive.

Figure 3.

Kaplan–Meier survival plot. Patients with gastric cancers harboring loss of one of the 11 chromosomal regions have a significantly better survival outcome than patients with gastric cancers without loss of one of the 11 chromosomal regions (log rank, 9.76; p = .002).

0, no loss; 1, loss.

When analyzing patients with lymph node–positive gastric cancers only, patients with a loss of one of these chromosomal regions showed a trend for a better survival outcome than patients without a loss of one of these chromosomal regions, although this difference was not statistically significant (p = .06).

Because lymph node status was associated with survival outcome in gastric cancer patients in the univariate analysis, we performed a multivariate analysis including loss of one of these 11 chromosomal regions and lymph node status in the model. Type of resection (D1 versus D2) and histological tumor type (intestinal versus diffuse versus mixed type) were not correlated with survival in the univariate analysis and therefore not included in the model for the multivariate analysis. Multivariate analysis revealed that loss of one of these 11 chromosomal regions, in addition to lymph node status, had independent prognostic value (p < .05; HR, 0.31 and p <0.001; HR, 3.84 for loss of one of these regions and lymph node status, respectively).

Expression of HSP90 on Chromosome 14q32.12-q32.33 Correlates with Survival

Immunohistochemistry analysis was performed to analyze expression of HSP90, encoded by one of the candidate genes located on chromosome 14q32.13-q32.33. Again, only patients with M0 and R0 disease were included for correlation of HSP90 expression with survival, leaving 230 patients in the analysis. In total, five gastric cancers showed a negative or mildly positive expression, 63 showed a moderately positive expression, and 162 showed a highly positive expression of HSP90. Patients with gastric cancers with negative or mildly positive expression of HSP90 had better survival outcomes than patients with moderate or high protein expression of HSP90 (p = 0.04; log rank, 6.24; HR, 1.63; 95% CI, 1.09–2.45) (Fig. 4).

Figure 4.

Kaplan–Meier survival plot. Patients with loss of heat shock protein 90 (HSP90) expression have a better survival outcome than patients with strong HSP90 protein expression (log rank, 6.24; p = 0.04).

1, negative or mild positive; 2, moderate positive; 3, strong positive.

Discussion

Gastric cancer is mostly detected at a late disease stage, when the tumor has already spread to the lymph nodes, resulting in a high mortality rate for gastric cancer patients worldwide. One approach to reduce the death rate from gastric cancer is secondary prevention, aiming to detect tumors in a curable stage. In high-prevalence countries like Japan, screening programs are in place and mortality rates are lower than in, for example, western countries [32–34]. In areas with lower incidences, like western Europe, population screening is not a reasonable option. Strategies to reduce death rates from gastric cancer mainly focus on improving therapeutic interventions. In current daily practice, this means more extensive surgery, (neo)adjuvant systemic treatment, and additional chemoradiotherapy. These therapeutic regimens largely follow a “one-size-fits-all” principle. However, more intensive treatment strategies do increase treatment-related morbidity and even mortality, whereas subgroups of patients with gastric cancer may not benefit from these intensified treatment strategies. In the present study, we identified such a potential subgroup of patients based on genomewide DNA copy number profiling.

In a first final series of 183 gastric cancer cases with good quality array CGH data, we found similar gains and losses on chromosomes as in previously published data [23, 35–38]. However, previous studies reported losses of 18q and gains of 20q to be correlated with lymph node status and gastric cancer patient survival outcome [17, 38–40]. In the present study, when only looking at chromosome 20q, indeed, a significantly higher percentage of lymph node–positive gastric cancers showed gain of this chromosome arm than lymph node–negative gastric cancers (88% versus 73%, respectively; p = .009), but in the overall univariate analysis and after correcting for multiple comparisons by CGH test, this significance disappeared. No significant differences were observed in losses of chromosome 18q between lymph node–positive and lymph node–negative gastric cancers (13% versus 20%, respectively; p = ns) (data not shown). One explanation might be the large difference in sample size between the present study and previous array CGH studies on gastric cancer, which were based on ∼35 tumors [38, 39].

A number of specific chromosomal regions located within 5q11.2-q35.1, 10q11.23–21.3, and 14q32.11-q32.33 did occur with significantly different frequencies in lymph node–positive and lymph node–negative gastric cancers, although the frequencies of these events were relatively low. In addition, losses of nine regions located within 5q11.2-q31.3 and two regions located within 14q32.11-q32.33 were also highly correlated with clinical outcome. Patients with losses of at least one of these 11 regions had a much better survival outcome (p = .002; log rank, 9.76; HR, 0.19). Multivariate analysis showed independent prognostic value for loss of one of the 11 chromosomal regions (p < .05; HR, 0.31). We found, using array CGH analysis, a particular subgroup of gastric cancers with excellent clinical behavior. Identification of this subgroup of 14% of the patients may have clinical relevance because they may not need perioperative chemotherapy, radiation therapy, or chemoradiation.

In addition to its clinical relevance, it is important to further analyze the underlying biological mechanism behind this relatively favorable subgroup of tumors. Losses of chromosome 14q32.2 have been shown to correlate with a better survival outcome in patients with non-small cell lung cancer [41]. HSP90AA1 (HSP90), a chaperone HSP located in this region, stabilizes several oncogenes and was shown to be the only gene with lower expression in cancers affected by loss of this region [41, 42]. High HSP90 expression is associated with poor prognosis in breast cancer patients [43]. Zuo et al. [44] showed HSP90α expression in nearly 90% of gastric cancers, with significantly higher expression levels in lymph node–positive than in lymph node–negative gastric cancers. In contrast to our results, Gianinis et al. [45] showed poor survival in gastric cancer patients with low expression levels of HSP90. However, inhibition of HSP90 has been shown to downregulate the oncogenes encoding EGFR and human epidermal growth factor receptor (HER)-2 in gastric cancer cells and to inhibit angiogenesis, tumor growth, and tumor cell invasiveness [46]. In the present study, loss of both 14q32.2 and HSP90 protein expression were correlated with better prognosis in gastric cancer patients. Loss of 14q32.2 in gastric cancer might have resulted in a lower expression level of HSP90, in turn leading to lower tumor growth and invasion, possibly explaining the higher frequencies of loss of 14q32 in lymph node–negative gastric cancers and the significant correlation between this chromosomal region and better survival outcomes of patients. Unfortunately, in the present study, expression of HSP90 and loss of chromosome 14q32.2 could not be investigated side by side because of a lack of sufficient cases with both immunohistochemistry and array CGH data. Therefore, ultimate proof of a gene dose effect of DNA copy number on expression of HSP90 in gastric cancer remains to be determined. In future studies this should be further explored as well as whether or not simple immunohistochemistry staining of HSP90 expression can be used to select patients for surgery only.

Chromosomal losses of regions on 5q are associated with better prognosis in patients with ovarian serous carcinoma [47]. In the present study, several regions within 5q12-q35.1 were correlated with better prognosis. Interestingly, two other HSPs, HSPA4 and HSPA9, are located within this region, on 5q31.1 and 5q31.2, respectively. Both proteins belong to the HSP70 family, which acts as chaperones and plays a role in protection from apoptosis. Inactivation of HSP70 has been shown to activate apoptosis [48, 49]. Loss of chromosome 5q11.2-q35.1 could have resulted in greater apoptosis of the gastric cancer cells by inactivation of normal HSP70 protein function, possibly explaining the lower frequency of lymph node metastasis and better patient outcome.

HSPs are of particular interest because an increasing number of drugs are being developed targeting these proteins. Anti-HSP90 drugs are currently under investigation in several clinical trials for different tumor types, including gastric cancer. The association between HSP90 protein expression and clinical outcome feeds the hypothesis that patients with high expression of HSP90 may benefit from these agents. Because HER-2 is one of the most sensitive target proteins of HSP90, investigating the combination of anti-HSP90 drugs with trastuzumab could be interesting in HER-2⁺ advanced gastric cancer patients. A previous phase I study showed a clinical response in HER-2⁺ breast cancer patients resistant to trastuzumab alone when treated with an HSP90 inhibitor in combination with trastuzumab [50].

Drugs targeting other HSP proteins, including HSP70 that also do seem promising drug targets, have yet to be developed [49, 51–54].

In conclusion, using genomewide DNA copy number profiling, the present study identified a subgroup of gastric cancer patients, marked by losses on chromosomes 5q11.2-q31.3 and 14q32.11-q32.33, who have an excellent clinical outcome after surgery alone, and patients with these tumors may gain little benefit from additional intensified therapies. For implementation in clinical practice, validation of these findings in an independent series including patients who are treated using surgical resection alone is needed [55]. This should ultimately be followed by a clinical trial in which patients with gastric cancers harboring losses on chromosome 5q or 14q are randomized for treatment with surgery alone or surgery with additional treatment with chemoradiation, as the best current practice. However, such a trial will not easily be accomplished because perioperative chemotherapy is the standard of treatment in most western countries.

See www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/Design: Tineke E. Buffart, Gerrit A. Meijer, Cornelis J.H. van de Velde

Provision of study material or patients: Gerrit A. Meijer, Elma Meershoek-Klein Kranenbarg, Cornelis J.H. van de Velde, Heike I. Grabsch

Collection and/or assembly of data: Tineke E. Buffart, Marianne Tijssen, Gerrit A. Meijer, Elma Meershoek-Klein Kranenbarg, Cornelis J.H. van de Velde

Data analysis and interpretation: Tineke E. Buffart, Beatriz Carvalho, Nicole C.T. van Grieken, Wessel N. van Wieringen, Marianne Tijssen, Bauke Ylstra

Manuscript writing: Tineke E. Buffart, Beatriz Carvalho, Nicole C.T. van Grieken, Gerrit A. Meijer, Henk M.W. Verheul

Final approval of manuscript: Tineke E. Buffart, Beatriz Carvalho, Nicole C.T. van Grieken, Wessel N. van Wieringen, Marianne Tijssen, Gerrit A. Meijer, Henk M.W. Verheul, Cornelis J.H. van de Velde, Heike I. Grabsch, Bauke Ylstra