Characteristic gene expression in stromal cells of gastric cancers among atomic-bomb survivors
Article first published online: 9 OCT 2008
Copyright © 2008 Wiley-Liss, Inc.
International Journal of Cancer
Volume 124, Issue 5, pages 1112–1121, 1 March 2009
How to Cite
Oue, N., Sentani, K., Sakamoto, N., Motoshita, J., Nishisaka, T., Fukuhara, T., Matsuura, H., Sasaki, H., Nakachi, K. and Yasui, W. (2009), Characteristic gene expression in stromal cells of gastric cancers among atomic-bomb survivors. Int. J. Cancer, 124: 1112–1121. doi: 10.1002/ijc.24060
- Issue published online: 18 DEC 2008
- Article first published online: 9 OCT 2008
- Accepted manuscript online: 9 OCT 2008 12:00AM EST
- Manuscript Accepted: 29 SEP 2008
- Manuscript Received: 18 JAN 2008
- Ministry of Education, Culture, Science, Sports, and Technology of Japan
- Ministry of Health, Labour and Welfare of Japan
- gastric cancer;
- radiation carcinogenesis;
- atomic bomb;
To elucidate the mechanism of radiation-induced cancers, molecular analysis of cancers in atomic-bomb survivors is important. In our study, we developed a custom oligonucleotide array of 208 genes. We analyzed gene expression profiles of gastric cancers (GCs) from atomic-bomb survivors and identified 9 genes with significantly lower expression in GCs from exposed patients than in GCs from nonexposed patients. Among these 9 genes, expression of versican and osteonectin was investigated in greater detail using immunohistochemistry in 116 GCs from 64 exposed and 52 nonexposed patients who developed GC after the bombing. In the Stage I/II GCs, the clinicopathologic, phenotypic and proliferative characteristics of GCs from exposed and nonexposed patients did not differ significantly; however, versican and osteonectin were expressed at much lower levels in the area of tumor-associated stroma of exposed patients than in nonexposed patients (p = 0.026 and p = 0.024, respectively). These results suggest that the characteristics of tumor-associated stromal cells differ between GCs from exposed and nonexposed patients. © 2008 Wiley-Liss, Inc.
More than 60 years have passed since atomic-bomb (A-bomb) exposure in Hiroshima and Nagasaki. A prospective cohort study [Life Span Study (LSS)] of 120,000 subjects has been conducted by the Radiation Effects Research Foundation (RERF).1 Solid cancers, including breast, colon, lung and stomach, have a long latency period, and the excess relative risks of solid cancers remain high, specifically among those exposed when young.1 Although approximately half of the LSS members have already deceased, cancer mortality in the LSS has continued to increase with the aging of this population, and it is anticipated to peak in 2015. Previous studies conducted to elucidate the mechanism of radiation-associated carcinogenesis mainly used formalin-fixed and paraffin-embedded archival tissues,2, 3 which are not suitable for selected molecular analyses (e.g. quantitative assays of gene expression) because of degradation of RNA.
According to the World Health Organization, gastric cancer (GC) is the fourth most common malignancy world wide, with ∼870,000 new cases occurring yearly. Cancer develops as a result of multiple genetic and epigenetic alterations.4, 5 Better knowledge of changes in gene expression that occur during gastric carcinogenesis may lead to improvements in diagnosis, treatment and prevention of GC. Thus far, the effect of radiation on GC development has been estimated on the basis of the LSS, in which both mortality and incidence were used as end points. The excess relative risks per gray (Gy) were 1.20 for mortality6 and 1.32 for incidence.1 Although several genetic alterations, including mutations in TP53 and BRAF, have been reported in selected cancers of A-bomb survivors,2, 3, 7 changes in gene expression have not been investigated. Furthermore, specific mutations for radiation-associated cancers have not been reported.
In our study, we performed custom array analysis of GCs from A-bomb survivors. We found reduced expression of versican and osteonectin in GCs from exposed patients. Versican, a large chondroitin sulfate proteoglycan, belongs to the aggrecan gene family.8 Versican is a component of the extracellular matrix (ECM) of various soft tissues and is involved in a number of pathologic processes including cancer, atherosclerotic vascular diseases and so on.9 Versican represses cell adhesion and promotes proliferation, migration, and invasion.10, 11 Increased stromal versican deposition correlates with breast cancer relapse and prostate cancer progression.12, 13 Osteonectin, a matricellular glycoprotein, modulates the interaction of cells with the ECM through its regulation of cell adhesion and matrix assembly.14 Increased expression of osteonectin has been reported in several human cancers, and stromal osteonectin expression has been shown to correlates with tumor progression and poor survival.15, 16 Osteonectin enhances the invasive capacity of prostate and breast cancer cells.17, 18 Although overexpression of versican and osteonectin has been reported in GC,19, 20 the relationship with radiation exposure history of patients has not been studied. Therefore, we performed immunohistochemical analysis of versican and osteonectin expression in 116 GCs from A-bomb survivors.
Material and methods
Primary tumor samples from 136 patients with GC were collected. Patients were treated at Hiroshima University Hospital (Hiroshima, Japan) or at an affiliated hospital. All patients underwent curative resection. Only patients who did not undergo preoperative radio- or chemotherapy and did not have clinical evidence of distant metastasis were enrolled in the study.
For use in our oligonucleotide array analysis, 3 freshly frozen GC tissue samples from exposed patients (5, 7 and 18 mGy) were obtained during surgery at the Department of Surgical Oncology, Hiroshima University Hospital, between 2004 and 2005. These patients were A-bomb survivors (3 LSS cohort members, RERF) in Hiroshima, Japan, who developed GC after the bombing. Their corresponding nonneoplastic mucosa samples were also available. In addition, we analyzed 20 freshly frozen GC tissue samples from nonexposed patients, who underwent surgery between 1991 and 1998 at the Department of Surgical Oncology, Hiroshima University Hospital; they were neither A-bomb survivors nor the LSS cohort members. Of these 20 GC samples, 10 corresponding nonneoplastic mucosa samples were available. All 20 GC samples were obtained during surgery at Hiroshima University Hospital. We confirmed microscopically that the tumor specimens were predominantly (>50%, on a nuclear basis) cancer tissue. Samples were frozen immediately in liquid nitrogen and stored at −80°C until use.
For immunohistochemical analysis, we used formalin-fixed and paraffin-embedded archival tissues from 116 patients with GC who underwent surgery between 1975 and 2005 at the Department of Surgical Oncology, Hiroshima University Hospital. All 116 patients were A-bomb survivors (LSS cohort members) in Hiroshima, Japan. Although these patients were survivors who developed GC after the bombing, they were further classified according to the levels of exposed radiation dose (i.e. ≥5 mGy and <5 mGy were defined as “exposed” and “nonexposed,” respectively): 64 exposed (median dose, 51 mGy) and 52 nonexposed patients.
Tumor staging was performed according to the TNM classification system.21 Histologic classification of GC was carried out according to the Lauren classification system.22 The detailed procedures for acquiring informed consent from study patients and collecting tissue specimens were as described previously.23 In accordance with the Ethical Guidelines for Human Genome/Gene Research enacted by the Japanese Government, tissue specimens were collected and used on the basis of the approval from the Ethical Review Committee of the Hiroshima University School of Medicine and from the ethical review committees of collaborating organizations.
A-bomb radiation doses were estimated with the DS02 system.24
Oligonucleotide array construction
Probes, which were all 65 bp in length, were designed to have approximately the same annealing temperature. The oligonucleotide array, Genopal™ (Mitsubishi Rayon, Tokyo, Japan), was made as described previously.25 In brief, plastic hollow fibers were bundled in an orderly arrangement and hardened with resin to form a block. Oligonucleotide-capture probes were chemically bonded inside each hollow fiber with hydrophilic gel. The block was then sliced into thin chips, each of which was set in a holder (www.mrc.co.jp/genome/about/process.html for details). The array contained 208 genes, including GC-related genes identified by our previous SAGE analysis,26 known genes related to development and progression of GC,27, 28 genes related to DNA damage response and repair and genes associated with sensitivity to anticancer drugs.29 A list of the genes on the array is available upon request.
Preparation of labeled probe, hybridization, detection and data analysis
Total RNA was isolated from frozen tissue with Isogene (Nippon Gene, Tokyo, Japan), according to the manufacturer's protocol. Quantification and integrity of RNA were assessed with an Agilent 2100 Bioanalyzer and RNA 6000 LabChip Kit (Agilent Technologies, Palo Alto, CA). One microgram of total RNA was used to prepare antisense biotinylated RNA with MessageAmp™ II-Biotin Enhanced Single Round aRNA Amplification Kit (Ambion, Austin, TX) per the manufacturer's instructions. The biotinylated cRNAs were then cleaned up and fragmented. Hybridization was carried out with the oligonucleotide array in 100 μl of hybridization buffer (0.12 M Tris HCl/0.12 M NaCl/0.05% Tween-20 and 5 μg of fragmented biotinylated aRNA) at 65°C overnight. After hybridization, the oligonucleotide arrays were washed twice in 0.12 M Tris HCl/0.12 M NaCl/0.05% Tween-20 at 65°C for 20 min, followed by washing in 0.12 M Tris HCl/0.12 M NaCl for 10 min, before being cooled slowly to room temperature. After staining with streptavidin–Alexa Fluor 647 (Invitrogen, Carlsbad, CA), the Genopal array was scanned, and the image was captured with a cooled CCD-type Microarray Image Analyzer (Mitsubishi Rayon). Fluorescence intensity was analyzed with software developed by Mitsubishi Rayon. Fluorescence throughout the 3-dimensional structure of each array feature can be efficiently captured because of the long focal depth of the optical system of the image analyzer (www.mrc.co.jp/genome/about/analysis.html). After image acquisition and quantification, spots with signal intensity lower than or equal to that of the background were identified and excluded from the analysis. Next, background-subtracted spot intensities were normalized so that the ACTB gene signal would be 10,000.
Formalin-fixed and paraffin-embedded samples were sectioned, deparaffinized and stained with H&E to ensure that the sectioned block contained tumor cells. Adjacent sections were then stained immunohistochemically with a Dako Envision + Rabbit Peroxidase Detection System (Dako Cytomation, Carpinteria, CA) or Dako Envision + Mouse Peroxidase Detection System (Dako Cytomation). Antigen retrieval was done by microwave heating in citrate buffer (pH 6.0) for 30 min. After peroxidase activity was blocked with 3% H2O2-methanol for 10 min, sections were incubated with normal goat serum (Dako Cytomation) for 20 min to block nonspecific antibody binding sites. Sections were incubated with primary antibodies against transketolase (dilution 1:50; Santa Cruz Biotechnology, Santa Cruz, CA), versican (1:50; Seikagaku Corporation, Tokyo, Japan), THBS-2 (1:50; Santa Cruz Biotechnology), PDGF receptor-β (1:50; Santa Cruz Biotechnology), ribonuclease A (1:50; Abcam, Cambridge, UK), osteonectin (1:50; Novocastra, Newcastle, UK), vimentin (1:50, Dako Cytomation) and Ki67 (1:50, Dako Cytomation) for 1 hr at room temperature, followed by incubations with Envision + antirabbit peroxidase or Envision + antimouse peroxidase for 30 min each. Staining was completed with 10 min incubation with the substrate–chromogen solution. Sections were counterstained with 0.1% hematoxylin. For the Ki67-index, 1,000 nuclei were counted to evaluate the percentage of positive nuclei. The Ki67-index was considered to reflect the proliferative index.
Phenotype analysis of GC
GCs were classified into 4 phenotypes: gastric (G) type, intestinal (I) type, gastric and intestinal mixed (GI) type and unclassified (N) type. For phenotypic expression analysis of GC, we performed immunohistochemical analysis (as described earlier) with 4 antibodies: anti-MUC5AC (Novocastra) as a marker of gastric foveolar epithelial cells, anti-MUC6 (Novocastra) as a marker of pyloric gland cells, anti-MUC2 (Novocastra) as a marker of goblet cells in the small intestine and colorectum, and anti-CD10 (Novocastra) as a marker of microvilli of absorptive cells in the small intestine and colorectum. The criteria30 for the classification of G-type and I-type GCs were as follows. GCs in which more than 10% of cells in the section expressed at least 1 gastric epithelial cell marker (MUC5AC or MUC6) or intestinal epithelial cell marker (MUC2 or CD10) were classified as G-type or I-type cancers, respectively. Sections that showed both gastric and intestinal phenotypes were classified as GI type, and those that lacked both the gastric and intestinal phenotypes were classified as N type.
Correlation between the gene expression profiles from different sample conditions was assessed by Spearman's rank correlation coefficients. Differences in mRNA expression levels between 2 samples were tested by Mann–Whitney U test for individual genes. Univariate analysis for clinicopathologic, phenotypic and proliferative variables in relation to radiation exposure status was done by Mann–Whitney U test for continuous variables and Fisher's exact test for categorical variables. Associations between clinicopathologic variables and immunostaining for versican or osteonectin were analyzed by Fisher's exact test. Multivariate logistic regression analysis was carried out to assess the relationship among clinicopathologic characteristics, expression of versican and osteonectin, and radiation exposure status. A p value of <0.05 was considered statistically significant.
Custom array analysis
Toward identification of potential molecular markers for radiation-associated cancer and also better understanding of its molecular mechanisms, we designed a custom oligonucleotide array comprising 208 genes (Fig. 1). Because this platform has not been characterized, we first validated the performance of this array before analyzing the GCs from exposed patients. To validate the array, total RNA was isolated from the MKN-1 cell line. RNA quality was assessed with a Bioanalyzer (Agilent), and the RNA integrity number (RIN) was confirmed to be 10.0. The isolated RNA was divided into 5 tubes (Samples 1–5), and we subjected the RNA in each tube to a different preparation condition. To analyze the effect of the amount of total RNA, we prepared 3 samples. Sample 1 contained 1.0 μg total RNA, Sample 2 contained 2.0 μg total RNA and Sample 3 contained 0.5 μg total RNA. To analyze the effect of RNA quality, we prepared 2 RNA samples with different RINs. Sample 4 was frozen and thawed 20 times, and Sample 5 was frozen and thawed 40 times. The RNA quality was then assessed with the Bioanalyzer. The RIN of Sample 4 was 8.2 and that of Sample 5 was 5.9. Probes derived from these 5 samples were hybridized simultaneously (Fig. 1). Although the condition of each sample differed, gene expression levels from Sample 1 obtained by oligonucleotide array correlated well with those from Sample 2 (p < 0.0001, r = 0.95), Sample 3 (p < 0.0001, r = 0.96), Sample 4 (p < 0.0001, r = 0.98) and Sample 5 (p < 0.0001, r = 0.97) (Fig. 1). Therefore, 1 μg of total RNA (RIN > 6.0) was used for further array analysis.
We next analyzed the gene expression profiles of GCs from exposed and nonexposed patients by custom oligonucleotide array. Freshly frozen GC tissue samples were obtained from 3 exposed patients (Cases EX01, EX02 and EX03, with radiation dose 7, 5 and 18 mGy, respectively) as well as their corresponding nonneoplastic mucosa samples. Clinicopathologic features of these exposed patients are shown in Table I. Histologically, EX01 was intestinal-type GC of Lauren classification, EX02 was diffuse-type GC of Lauren classification, and EX03 was an α-fetoprotein (AFP)-producing hepatoid adenocarcinoma (Fig. 2a). Immunostaining of EX03 revealed that AFP was present in cancer cells (Fig. 2a). On the other hand, when we analyzed 20 freshly frozen GC tissues from nonexposed patients (Cases NEX01 to NEX20), unsupervised clustering showed that the exposed and nonexposed patients could not be distinguished on the basis of mRNA expression (data not shown). To determine whether there is a gene expression profile characteristic to exposed patients, we compared expression levels of individual genes between exposed and nonexposed patients. Finally, we found 9 genes whose expression was significantly lower in GC from exposed patients than in GC from nonexposed patients (Table II). No gene showed higher expression in GC from exposed patients than in GC from nonexposed patients.
|Sample name||Age (years)||Sex||T grade||N grade||M grade||Stage||Histologic classification||Radiation dose (mGy)|
|Gene symbol||Exposure status||mRNA expression: median (range)||p value1|
Among these 9 genes, antibodies against proteins encoded by TKT (encoding transketolase), VCAN (encoding versican), THBS2 (encoding THBS-2), PDGFRB (encoding PDGF receptor-β), RNASE1 (encoding ribonuclease A) and SPARC (encoding osteonectin) were commercially available. We performed immunohistochemistry of these 6 molecules in 3 formalin-fixed and paraffin-embedded archival GC tissue samples from exposed patients and 10 GC samples from nonexposed patients analyzed by oligonucleotide array to compare oligonucleotide array data and immunostaining results. The immunostaining results for versican (encoded by VCAN) and osteonectin (encoded by SPARC) were consistent with those of the oligonucleotide array. It has been reported that immunoreactivity for versican was present in tumor-stroma associated with malignant areas and in blood vessel walls.31 In Case EX03 (exposed patient), expression of versican (VCAN) mRNA on the oligonucleotide array was low, and immunohistochemistry revealed that tumor-associated stroma did not express versican despite staining of versican in blood vessel walls (Fig. 2b). In Case EX01 and Case EX02, versican staining was not observed. In Case NEX03 (nonexposed patient), strong and extensive staining of versican was observed in tumor-associated stroma (Fig. 2b). Stromal versican staining was also found in several GC cases from nonexposed patients. In addition to versican, osteonectin has also been reported to be stained in tumor-associated stroma. In Case EX01 (exposed patient), expression of osteonectin (SPARC) mRNA on the oligonucleotide array was low, and a small fraction of tumor-associated stromal cells was shown to express osteonectin by immunohistochemistry (Fig. 2c). In Case EX02 and Case EX03, osteonectin staining was not observed. In contrast, in Case NEX15 (nonexposed patient), extensive staining of osteonectin was observed in tumor-associated stroma (Fig. 2c). Stromal osteonectin staining was also found in several GC cases from nonexposed patients. Immunostaining of the remaining 4 molecules (transketolase, THBS-2, PDGF receptor-β and ribonuclease A) did not differ significantly between exposed and nonexposed patients. Therefore, we decided to perform immunostaining of versican and osteonectin in additional GC cases.
Clinicopathologic features, mucin phenotypes and proliferative characteristics of survivor patients
To validate the reduced expression of versican and osteonectin in GC from exposed patients, we collected formalin-fixed and paraffin-embedded archival tissues from 116 patients with GC who underwent surgical resection. All 116 patients were A-bomb survivors (LSS cohort members) in Hiroshima, Japan, and comprised 64 exposed (median dose, 51 mGy; range, 5–2,601 mGy) and 52 nonexposed patients (range, 0–4 mGy) who developed GC after the bombing. Patient characteristics such as latency period (years elapsed from A-bombing to diagnosis, defined only for exposed patients), age at the time of A-bombing, histologic type, sex, age at diagnosis, T grade, N grade and tumor stage are summarized in Table III. Clinicopathologic characteristics of patients did not statistically differ between exposed and nonexposed patients. Diffuse-type GC was found more frequently in exposed patients than in nonexposed patients, but the difference was not statistically significant (p = 0.085).
|Variable||Exposed patients (n = 64)||Nonexposed patients (n = 52)||p value|
|Median radiation dose (mGy, range)||51 (5–2601)||0 (0–4)|
|Median latency period (years, range)||47 (30–60)||46 (30–57)||0.51|
|Median age at the time of atomic-bombing (years, range)||25 (2–47)||28 (2–47)||0.0981|
|Age at diagnosis (years)|
|Median Ki67-index (%, range)||33 (11–69)||39 (10–64)||0.1961|
Despite the usefulness of the Lauren classification, it was previously reported that GC can be subdivided according to mucin expression into 4 phenotypes (G type, I type, GI type and N type).32 Several distinct genetic and epigenetic changes have been reported to be associated with G-type and I-type GCs.33–35 Therefore, we investigated the mucin phenotypes of 116 GCs: Gastric (MUC5AC and MUC6) and intestinal (MUC2 and CD10). MUC5AC was detected in 57 of 116 (49%) cases, MUC6 in 37 (32%) cases, MUC2 in 34 (29%) cases and CD10 in 25 (22%) cases. Expression of these 4 markers did not differ statistically between exposed and nonexposed patients (data not shown). In addition, distribution of the G, I, GI and N phenotypes did not differ significantly between exposed and nonexposed patients (data not shown).
Phenotypic shift from G-type to I-type GC along with tumor progression has been reported.32 Therefore, mucin phenotypes and immunostaining of MUC5AC, MUC6, MUC2 and CD10 were analyzed with respect to tumor stages. In Stage I/II, expression of the 4 markers did not differ significantly between exposed and nonexposed patients (data not shown). In contrast, in Stage III/IV, some GCs from exposed patients showed extensive staining of MUC6 (Fig. 3a). In addition, staining of MUC2 was rare in GCs from exposed patients (Fig. 3a). Of 28 Stage III/IV GCs from exposed patients, MUC6 was expressed in 12 GCs (43%), whereas MUC6 was expressed in 2 (12%) of 17 Stage III/IV GCs from nonexposed patients (p = 0.046). In Stage III/IV cases, the frequency of MUC2 expression in GCs from exposed patients (3/28, 11%) was significantly lower than that in GCs from nonexposed patients (8/17, 47%, p = 0.011). There was no correlation between exposure status and MUC5AC or CD10 (data not shown). Mucin phenotypes with respect to tumor stage are shown in Figure 3b. For Stage I/II GCs, distribution of the G, I, GI and N types did not differ significantly between exposed and nonexposed patients. Among GCs from nonexposed patients, frequency of the G type slightly decreased with advancing tumor stage. In contrast, among GCs from exposed patients, frequency of the G type increased with advancing tumor stage: furthermore, frequency of the I type decreased with advancing tumor stage. However, statistical analysis showed that neither frequency of the G type (Table IV) nor I type (Table V) differed between exposed and nonexposed patients.
|G type||Other type||p value1|
|Exposed patients||14 (50%)||14||0.118|
|Nonexposed patients||4 (24%)||13|
|I type||Other type||p value1|
|Exposed patients||5 (18%)||23||0.5|
|Nonexposed patients||5 (29%)||12|
Proliferative characteristics of GCs from exposed and nonexposed patients were also investigated. The Ki67-index did not differ significantly between exposed and nonexposed patients (Table III). This was also the case in Stage I/II (data not shown). These results indicate that no significant differences in clinicopathologic, phenotypic and proliferative characteristics of GCs were found between exposed and nonexposed patients, at least in Stage I/II GCs.
Decreased expression of versican and osteonectin in exposed patients
We performed immunostaining of versican and osteonectin in 116 GCs. In nonneoplastic gastric mucosa, although epithelial cells and stromal cells exhibited weak or no expression of versican (Figs. 4a and 4b), versican staining was observed in the walls of blood vessels (Fig. 4c). In GC tissue, tumor cells were not stained; however, versican was expressed in tumor-associated stroma. Intracellular staining of versican was detected in fibroblastic cells, because of their morphology and vimentin (a marker of fibroblasts) positivity on serial sections. Staining in fibrous bands within the tumor was also observed. The level of versican immunoreactivity was evaluated in the area of tumor-associated stroma (fibroblastic cell plus ECM). The percentage of versican-stained area of tumor-associated stroma was a continuum from 0 to 80%, and characteristic staining pattern was not observed. To analyze the relationship of versican staining to clinicopathologic characteristics, the GC cases were divided into two groups: diffuse positive group and focal positive or negative group. To maximize the statistical detection power, it is ideal that the number of diffuse positive cases/focal positive or negative cases ratio is less than 2 in nonexposed patients. When we chose 50% (50% of the area of tumor-associated stroma) as a cutoff for diffuse positive group and focal positive or negative group, 32 (62%) were diffuse positive for versican in 52 GC cases from nonexposed patients. Therefore, the immunostaining was considered diffuse positive for versican when more than 50% of the area of tumor-associated stroma was stained by versican. Of 116 GC cases, 55 (47%) were diffuse positive for versican. Figure 4d illustrates the typically diffuse positive immunostaining of versican. In diffuse-type GC, strong and extensive stromal versican staining was frequently observed (Fig. 4d). Staining in fibrous bands within the tumor was also observed. Figure 4e illustrates the typically focal positive immunostaining of versican. In many GCs from exposed patients, although blood vessel wall expressed versican, staining of versican was weak or absent in stromal fibroblasts and stromal matrix (Fig. 4e). Diffuse positive GCs correlated to advanced T grade (p = 0.003), N grade (p = 0.016) and tumor stage (p = 0.001) (Table VI). Diffuse positive GCs were more frequently found in diffuse-type GCs than in intestinal-type GCs (p = 0.008) (Table 6). Focal positive or negative GCs were more frequently found in exposed patients than in nonexposed patients (p = 0.009) (Table 6). Because expression of versican correlated with tumor stage, versican immunostaining was analyzed by tumor stages. As shown in Table 6, in both Stage I/II and Stage III/IV GCs, focal positive or negative GCs were more frequently found in exposed patients than in nonexposed patients (p = 0.026 and p = 0.023, respectively). We then examined whether versican staining in GC from exposed patients was related to radiation dose, latency period or age at the time of A-bombing; however, there was no correlation between versican staining and any of these variables. Although versican staining was correlated with tumor stage in 116 GC cases (p = 0.001), multivariate logistic regression analysis revealed that expression of versican was not significant independent marker for tumor stage in 64 exposed patients (p = 0.6). Among 52 nonexposed patients, multivariate logistic regression analysis showed that expression of versican was not significant independent marker for tumor stage (p = 0.9).
|Diffuse positive (%)||Focal positive or negative||p value1||Diffuse positive (%)||Focal positive or negative||p value1|
|Male||30 (49)||31||0.7||16 (26)||45||0.8|
|Female||25 (45)||30||13 (24)||42|
|Age at diagnosis|
|≤65||15 (50)||15||0.8||9 (30)||21||0.5|
|>65||40 (47)||46||20 (23)||66|
|T1/T2||29 (37)||49||0.003||13 (17)||65||0.006|
|T3/T4||26 (68)||12||16 (42)||22|
|N0||18 (35)||34||0.016||7 (13)||45||0.011|
|N1/N2/N3||37 (58)||27||22 (34)||42|
|I/II||25 (35)||46||0.001||11 (15)||60||0.004|
|III/IV||30 (67)||15||18 (40)||27|
|Intestinal||27 (38)||45||0.008||17 (24)||55||0.7|
|Diffuse||28 (64)||16||12 (27)||32|
|Exposure status (all cases)|
|Exposed||23 (36)||41||0.009||9 (14)||55||0.005|
|Nonexposed||32 (62)||20||20 (38)||32|
|Exposure status (Stage I/II cases)|
|Exposed||8 (22)||28||0.026||2 (6)||34||0.024|
|Nonexposed||17 (49)||18||9 (26)||26|
|Exposure status (Stage III/IV cases)|
|Exposed||15 (54)||13||0.023||7 (25)||21||0.013|
|Nonexposed||15 (88)||2||11 (65)||6|
|Exposure status (intestinal type of Lauren classification)|
|Exposed||9 (26)||26||0.054||5 (25)||30||0.097|
|Nonexposed||18 (49)||19||12 (65)||25|
|Exposure status (diffuse type of Lauren classification)|
|Exposed||14 (48)||15||0.003||4 (14)||25||0.011|
|Nonexposed||14 (93)||1||8 (53)||7|
We next investigated expression of osteonectin. In nonneoplastic mucosa, epithelial cells did not express osteonectin. However, in some cases, stromal cells exhibited osteonectin immunoreactivity (Fig. 4f). In GC tissues, osteonectin was stained in tumor-associated stroma; however, tumor cells were not stained. Intracellular staining of osteonectin was detected in fibroblastic cells because of their morphology and vimentin positivity on serial sections. As in a previous study,20 immunoreactivity was also evident in the stromal matrix. Therefore, the levels of osteonectin immunoreactivity were evaluated in the area of tumor-associated stroma (fibroblastic cell plus ECM). The percentage of osteonectin-stained area of tumor-associated stroma ranged from 0 to 80%. Among 116 GCs, osteonectin-positive stromal cells were observed at the superficial parts of tumors in 44 (38%) cases (Fig. 4g). Osteonectin staining at the superficial parts of tumors was not correlated with T grade, N grade, tumor stage, histological type or exposure status (data not shown). To further analyze the relationship of osteonectin staining to clinicopathologic characteristics, the GC cases were divided into 2 groups: diffuse positive group and focal positive or negative group. When the same cutoff point for osteonectin and versican immunostaining was set (50% of the area of tumor-associated stroma), 20 (38%) were diffuse positive for osteonectin in 52 GC cases from nonexposed patients, and the number of diffuse positive cases/focal positive or negative cases ratio was less than 2 in nonexposed patients. Of 116 GC cases, 29 (25%) were diffuse positive for osteonectin. Figure 4h illustrates the typically diffuse positive immunostaining of osteonectin. Extensive staining of stromal cells was frequently observed in late-stage GC. Immunoreactivity was also evident in the stromal matrix. Figure 4i illustrates the typically focal positive or negative immunostaining of osteonectin. In many GCs from exposed patients, only few stromal cells exhibited osteonectin staining. Immunoreactivity was not found in the stromal matrix. Diffuse positive GCs correlated to advanced T grade (p = 0.006), N grade (p = 0.011) and tumor stage (p = 0.004) (Table VI). Focal positive or negative GCs were more frequently found in exposed patients than in nonexposed patients (p = 0.005) (Table VI). Because expression of osteonectin correlated with tumor stage, we analyzed osteonectin immunostaining by tumor stage. As shown in Table VI, in both Stage I/II and Stage III/IV GCs, focal positive or negative GCs were more frequently found in exposed patients than in nonexposed patients (p = 0.024 and p = 0.013, respectively). We then examined whether osteonectin staining in GCs from exposed patients was related to radiation dose, latency period or age at the time of A-bombing; however, there was no correlation between osteonectin staining and these variables. Although diffuse positive GCs correlated to tumor stage (p = 0.004), multivariate logistic regression analysis revealed that expression of osteonectin was not significant independent marker for tumor stage in 64 exposed patients (p = 0.057). Among 52 nonexposed patients, multivariate logistic regression analysis showed that expression of osteonectin was not significant independent marker for tumor stage (p = 0.9).
Because clinicopathologic characteristics and expression of versican and osteonectin may be interrelated, we performed multivariate logistic analysis to determine which variables are independent markers for radiation exposure status. As shown in Table VII, multivariate logistic regression analysis revealed that both expression of versican (p = 0.002) and osteonectin (p = 0.001) were significant independent markers for GCs from exposed patients.
|Variable||Hazard ratio||95% CI1||χ2||p value|
|Age at diagnosis (years)|
|Focal positive or negative||4.484||1.783–11.364|
|Focal positive or negative||6.452||2.053–20.408|
Our study entails 2-stage strategy to find molecular markers that are specifically involved in radiation-associated gastric carcinogenesis among A-bomb survivors. First, candidate genes were selected by custom oligonucleotide array, developed by us, which was applied to freshly collected cancer tissue specimens from 3 survivor patients exposed to atomic-radiation as well as 20 nonexposed patients for comparison. Second, of these candidate genes, we marked out VCAN (encoding versican) and SPARC (encoding osteonectin) genes, and their protein expression was further investigated by immunohistochemical analysis with formalin-fixed and paraffin-embedded archival cancer tissue specimens from 116 survivor patients comprised of 64 exposed and 52 nonexposed cases. In our study, we found that versican and osteonectin are expressed at significantly lower levels in tumor-associated stromas of exposed patients than in nonexposed patients. Because IR is a carcinogen and can increase an individual's risk of developing cancer, analysis of early-stage GC rather than late-stage GC is important to understand the development of radiation-induced cancer. It is important to note that the clinicopathologic, phenotypic and proliferative characteristics of the early-stage (Stage I/II) GCs analyzed in our study did not differ significantly between exposed and nonexposed patients, indicating that the GC samples analyzed in our study were collected in an unbiased manner. In Stage I/II GCs, versican and osteonectin were expressed at much lower levels in tumor-associated stromas of exposed patients than in nonexposed patients. Similar results were obtained in Stage III/IV GCs. Furthermore, multivariate logistic regression analysis revealed that reduced expression of versican and osteonectin were independent markers for GCs from exposed patients. These findings suggest that tumor–stroma interaction may be altered in the development of GCs in exposed patients, typically as demonstrated by decreased expression of versican and osteonectin in tumor-associated stromas.
In our study, stromal expression of both versican and osteonectin was lower in GCs from exposed patients than in GCs from nonexposed patients. Because transforming growth factor (TGF)-β1 can induce expression of both versican and osteonectin in cultured fibroblasts,36, 37 genes associated with TGF-β signaling pathway may be altered in exposed patients. Mice with a fibroblast-specific knockout of the TGF-β Type-II receptor rapidly develop epithelial tumors,38 suggesting that TGF-β receptor in fibroblasts may be altered in exposed patients.
Stromal expression of both versican and osteonectin correlates with tumor progression. In our study, stromal expression of versican and osteonectin correlated with tumor stage in GC. Versican and osteonectin were expressed less frequently in stroma of exposed patients than in nonexposed patients. Phenotypic analysis revealed that although the frequency of G-type GC decreased with advancing tumor stage in nonexposed patients, the frequency of G-type GC increased with advancing tumor stage in exposed patients. In Stage III/IV GC, a marginally significant difference was observed between the frequency of G-type GC in exposed patients and that in nonexposed patients. These findings suggest the molecular mechanism of GC progression may also differ between exposed and nonexposed patients. It has been reported that tumor stage of GC is quite comparative between exposed and nonexposed patients.39 In the present study, clinicopathologic characteristics of patients including tumor stage did not statistically differ between exposed and nonexposed patients. In addition, multivariate logistic regression analysis revealed that expression of versican and osteonectin was not significant independent marker for tumor stage. Taken together, expression of versican and osteonectin may not be involved deeply in tumor progression in GC. The biologic significance of decreased expression of versican and osteonectin in tumor progression should be investigated in detail.
Versican represses cell adhesion and promotes proliferation, migration and invasion.10, 11 Osteonectin enhances the invasive capacity of prostate and breast cancer cells.17, 18 Therefore, prognosis of exposed patients with GC may be favorable; however, it has been reported that the excess relative risk per Gy is 1.20 for mortality,6 suggesting that stromal expression of versican or osteonectin dose not contribute significantly to aggressiveness of GC. Other molecules may be involved in poor prognosis of exposed patients with GC.
In the present study, strong and extensive staining of both versican and osteonectin were observed in tumor-associated stroma. It has been reported that both versican and osteonectin are component of the ECM. In breast cancer, the induction of versican secretion by fibroblasts isolated from normal and cancer tissues has been reported.12 Expression of SPARC mRNA in liver myofibroblasts has been demonstrated by in situ hybridization.40 In the present study, in addition to extracellular staining, intracellular staining of versican and osteonectin in fibroblasts were observed. Furthermore, GC cells were negative for versican and osteonectin. Taken together, these results indicate that versican and osteonectin observed in tumor-associated stroma are produced mainly by tumor-associated stromal cells.
One weak point of our study is that the A-bomb radiation doses of exposed patients with GC analyzed by oligonucleotide array were low. In addition, only 3 GCs from exposed patients were analyzed by oligonucleotide array in the present study. Analysis of additional GCs is needed. In the present study, diffuse-type GC was found more frequently in exposed patients than in nonexposed patients although the difference was not statistically significant. The diffuse-type GC contains scirrhous-type GC, which is characterized by extensive fibrous stroma, infiltrative and rapid growth and poor prognosis. Because cancer stromal cells are a mixture of fibroblasts, smooth muscle cells, endothelial cells, vascular pericytes, mesenchymal stem cells and so on, difference of cell types composing cancer stroma between A-bomb exposed and nonexposed patients should be investigated.
Although expression of TKT, THBS2, PDGFRB and RNASE1 was significantly lower in GC from exposed patients than in GC from nonexposed patients by oligonucleotide array, immunoreactivities for these proteins were present in GC from exposed patients. Among 6 antibodies used in the present study, the specificity of antibodies against versican and osteonectin has been characterized in detail. In contrast, the specificity of antibodies against transketolase, THBS-2, PDGF receptor-β and ribonuclease A has not been characterized in detail. Therefore, it is possible that inconsistent results between oligonucleotide array and immunostaining represent insufficient specificity of these antibodies. Immunohistochemical analysis of these proteins by specific antibodies should be performed in the near future.
In conclusion, we found significant reduction of stromal expression of versican and osteonectin in GCs from exposed patients. Although it is unclear whether all of the GCs from exposed patients were radiation-induced cancers, versican and osteonectin may be markers for radiation-associated GC. Studies of tumor-associated stromal cells rather than tumor cells may be important to elucidate the precise long-term effects of radiation exposure.
We thank Ms. Emiko Hisamoto for excellent technical assistance and advice. This work was carried out with the kind cooperation of the Research Center for Molecular Medicine, Faculty of Medicine, Hiroshima University. We also thank the Analysis Center of Life Science, Hiroshima University for the use of their facilities.
- 4Recent advances in molecular pathobiology of gastric carcinoma. In: KaminishiM,TakuboK,MafuneK, eds. The diversity of gastric carcinoma pathogenesis: diagnosis, and therapy. Tokyo: Springer, 2005. 51–71., , ,
- 21SobinLH,WittekindCH, eds. TNM classification of malignant tumors,6th edn. New York: Wiley-Liss, 2002. 65–8.
- 23Systematic collection of tissue specimens and molecular pathological analysis of newly diagnosed solid cancers among atomic bomb survivors. Int Congr Ser 2007; 1299: 81–6., .