Gastric cancer (GC) is associated with high mortality rates and an unfavorable prognosis at advanced stages. In addition, there are no effective methods for diagnosing gastric cancer at an early stage or for predicting the outcome for the purpose of selecting patient-specific treatment options. Therefore, it is important to investigate new methods for GC diagnosis.
To facilitate its use in a diagnostic setting, a group of 74 genes with diagnostic and prognostic information was translated into a customized microarray containing a reduced set of 1,042 probes suitable for high throughput processing. In this report, we demonstrate for the first time that the custom mini-array can be used as a reliable diagnostic tool in gastric cancer. With an AUC value of 0.565 (95% CI 0.305-0.825) indicating a perfect test, the sensitivity and specificity of diagnosis from the ROC curve were calculated to be 70% and 80%, respectively.
Citation: Yin Y, Zhuo W, Zhao Y, Chen S, Li J, Wang L, et al. (2013) Converting a Microarray Signature into a Diagnostic Test: A Trial of Custom 74 Gene Array for Clarification and Prediction the Prognosis of Gastric Cancer. PLoS ONE 8(12): e81561. https://doi.org/10.1371/journal.pone.0081561
Editor: Courtney G. Montgomery, Oklahoma Medical Research Foundation, United States of America
Received: June 25, 2013; Accepted: October 14, 2013; Published: December 3, 2013
Copyright: © 2013 Yin et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by the National Natural Science Foundation of China (81302070, 81272678, 81071961) (http://www.nsfc.gov.cn), National Basic Research Program of China (973 Program) (2012CB945004) (www.973.gov.cn). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Gastric cancer (GC) has a high incidence and is the second leading cause of cancer mortality [1,2]. The prognosis of GC is highly dependent on the stage of the disease at diagnosis and the treatment method . The 5-year survival rate in advanced gastric cancer patients is about 20%, whereas in early-stage gastric cancer it is above 60% [4-6]. However, there have been no effective and feasible methods for detecting early stage cancer and for predicting the possible prognosis to provide suitable treatment for each patient. In Japan, though they could more efficiently detect and treat early gastric cancer (EGC) through extensive screening, only endoscopy could be commonly used for the detection EGC. This is why early diagnosis and the ability to distinguish malignant and premalignant lesions are important . Therefore, it is important to investigate new methods for GC diagnostic or prognostic predictions for clinical applications.
Certain gene alterations could be associated with canceration and progression of GC [8-11]. For example, previous data we studied suggested that collagen genes might be a potential biomarker to distinguish malignant from premalignant lesions in the stomach . Because the progression of disease from premalignant conditions to GC is a dynamic process [13-15], the detection of gene alterations could allow identification of disease-associated genes earlier than pathological examinations. In addition, gene expression provides additional information about a patient’s condition . Therefore, microarray analysis may be an important and useful method for diagnosis and risk stratification in gastric cancer.
In addition, using customized mini-arrays for clinical practice may not only be cheaper but may also require less sample RNA input for labeling and hybridization, and the data processing time could be substantially reduced compared with normal microarrays . A custom microarray of 70 genes for the prognosis of breast cancer, which has been approved by the Food and Drug Administration (FDA), verifies the feasibility of custom microarray in clinical use [18,19]. Although there have been several studies on certain groups of genes for GC diagnosis, the microarray technology is presently not used as a diagnostic tool in gastric cancer.
In this paper, we describe for the first time the development of a customized diagnostic GC mini-array and demonstrate that the custom mini-array could be used as a reliable diagnostic and prediction tool in gastric cancer.
A custom mini-array of a group of possible genes related to GC canceration and progress
In a previous study, we used an oligonucleotide microarray of 38,500 genes to systematically examine differential gene expression among 33 samples from normal, premalignant, and malignant lesions in the stomach . A fraction of 696 differentiated-expression genes found in the formal study were designed in the custom mini-array as part of a research base. In addition, for some groups in which genes were found that were possibly closely related to GC, the custom mini-array also included 44 collagen related genes [20-22], 54 genes for sex hormone receptors and pathways, differentiated-expression genes found in other studies, and 915 normalization genes (detailed data in table S1). In this study, a 1042-gene expression profile was established as a powerful diagnosis and predictor of disease outcome in gastric cancer patients.
Comparison of the 1042-Gene Array Performance with That of Original 25k Microarray
To determine whether the customized mini-array test performs similar to the original 25 k microarrays , two samples (LYXT and LYXS) from a same patients used in the original series to develop the 1042-diagnostic classifier were retrieved. The expressions of 696 genes generated using the diagnostic mini-array were highly similar (Pearson correlation of 0.957, p < 0.01) to the original published data (Figure 1).
Identification of genes differentially expressed in malignant and premalignant gastric tissues
All data of hybridizations were background-corrected, normalized, and analyzed to identify the differentiated-expression genes in 40 samples that represent malignant lesions and premalignant lesions (n=20 in each group). A set of 371 genes was found to separate malignant lesions from premalignant lesions using hierarchical clustering and SVM leave-one-out confirmation, whereas a premalignant sample (MGFS) were classified into the malignant group, and five malignant samples (XSHT, GJFT, CXCT, XYT and QLTT) were classified into the premalignant group. The MGFS and sample was collected from the surrounding tissue of a signet-ring carcinoma. The pathological report revealed that the XSHT sample was a Moderately differentiated adenocarcinoma, the GJFT sample was a Moderately to well differentiated adenocarcinoma, whereas CXCT, XYT and QLTT samples were well differentiated adenocarcinoma. These differentially expressed genes included 199 up-regulated genes and 172 down-regulated genes in malignant lesions compared with premalignant lesions (Figure 2).
Each column represents a sample and each row a gene. The character of “T” refers to tumor and the character of “S” refers to premalignant tissue in sample names. The expressions of genes between two groups have significant differences, fold change log2 >=1 or <= -1, P<0.01.
Distinguishing the prognosis of gastric cancer
An unsupervised, hierarchical clustering algorithm allowed us to cluster the 20 GC malignant lesions on the basis of their similarities measured over significant genes of 371 differentially expressed genes between malignant and premalignant lesions. Notably, in the left group, 4 of 10 sporadic patients were at an early stage of GC and the others presented with highly differentiated lesions, whereas in the right group, 2 of 10 sporadic patients were from the group that developed distant metastases within 5 years or with high stage and poorly differentiated lesions. Thus, using unsupervised clustering, we can distinguish between 'good prognosis' and 'poor prognosis' tumors to some extent (Figure 3). In addition, the type and stage of GC of the patients were associated with the sub-groups of poor or good prognosis (Table 1, detailed data in table S2).
Each column represents a tumor and each row a gene. The length and the subdivision of the branches display the relatedness of the GC (bottom) and the expression of the genes (right). The yellow line marks the subdivision into two dominant clusters. The expressions of genes between two groups have significant differences, fold change log2 >=1 or <= -1, P<0.01.
|Group 1 (Good prognosis)||Group 2 (Poor prognosis)|
|Pathological type (Moderately to well differentiated adenocarcinoma / Poorly differentiated adenocarcinoma/ Signet ring cell carcinoma/ Mucinous adenocarcinoma)||7/2/1/0||3/4/1/2|
|Metastasis with 5 years||1||3|
Custom array with minimum number of genes for GC diagnosis and prediction
Unsupervised two-dimensional cluster analysis of gene clustering and GC clustering was performed independently using an agglomerative hierarchical clustering algorithm with 371 genes that could identify malignant GC and premalignant lesions. The correlation coefficient of the expression for each gene with disease outcome was calculated, and 252 genes were found to be significantly associated with disease outcome (correlation coefficient <-0.3 or >0.3) (detailed data in table S1).
These 252 genes were rank-ordered on the basis of the magnitude of the correlation coefficient. The number of genes in the 'prognostic classifier' was optimized by sequentially adding subsets of 5 genes from the top of this rank-ordered list and evaluating its power for correct classification using the 'leave-one-out' method for cross-validation. Classification was made based on the correlation of the expression levels of the remaining samples from the good and poor patients, respectively. The accuracy improved until the optimal number of marker genes was reached. Therefore, 74 genes were determined to be the minimum number of genes that could be classified as two sub groups of different prognosis (Figure 4). With an AUC value of 0.565 (95% CI 0.305-0.825) indicating a perfect test, the sensitivity and specificity of diagnosis from the ROC curve were calculated to be 70% and 80%, respectively (Figure 5).
Each column represents a tumor and each row a gene.
Based on the Gene ontology (GO) function classification, the functional annotation for the genes involved in cell cycle, invasion and metastasis, angiogenesis, and signal transduction are significantly upregulated in the poor prognosis signature (annotation of genes listed in table S1). These genes include several groups for which the function annotation provides insight into the underlying biological mechanism leading to key functions involved in tumorigenesis and progression. The genes involved in the cell cycle, invasion and metastasis, angiogenesis, and signal transduction were significantly differentially expressed between two signatures (for example, GKN1, INHBA, SPP1 and THBS4). Meanwhile, unsupervised cluster analysis distinguishes between different prognostic tumors. By evaluating all 74 prognostic reporter genes, more genes belonging to these functional categories become apparent (for example, GKN1, GKN2, GIF, PSCA and LIPF).
The patients in the two groups classified by the 74 genes and the whole probes are nearly the same, except for the sample LCM, which was classified into the good prognosis group in the whole probes microarray and into the poor prognosis group in the 74-gene classification. The LCM sample was collected from the malignant tissue of a patient suffering from mucinous adenocarcinoma in stage IV with a shorter lifetime than the other patients in the formal groups. Therefore, the 74-gene classification microarray might be more reliable.
Verification of the 74 genes custom array and correlation of the microarray data with the prognostic profile
For the 11 GC patients included in the previous study , we calculated the correlation coefficient of the level of the expression of the 74 genes with the determined average profile of these genes in tumors from patients with good prognosis (CI). A patient with a correlation coefficient of more than -0.007 (the threshold resulted in a 13 percent rate of false negative results) was then assigned to the group with the good-prognosis signature, and all other patients were assigned to the group with the poor-prognosis signature (Figure 6). In addition, the survival curve of the two groups varies markedly (p<0.05) (Figure 7). Thus, the classifier showed a comparable performance on the validation of 11 independent sporadic tumors and confirmed the predictive power and robustness of the prognosis classification of the 74-gene custom array.
The yellow line marks the subdivision into two dominant clusters by two-dimensional cluster analysis. The white line marks the subdivision into two dominant clusters with optimized sensitivity.
Reproducibility of customized mini-array
To validate the data of gene expressions from the microarray data, we chose a differentiated-expression gene, INHBA, and examined its expression with quantitative RT-PCR analysis. Our data showed that the gene significantly changed expression in malignant tissues compared to premalignant tissues in 11 pair-matched samples, consistent with the results obtained from the microarray analysis (Figure 8).
The vertical numbers with log2 transformation are the pair-matched ratio of malignant lesions to premalignant lesions. A. The columns stand for the ratios derived from quantitative QRT- PCR experiment. B. Comparison of the ratios between microarray and quantitative QRT-PCR analyses.
A sub group of reproduction associated genes with gastric cancer
In the 74-gene custom microarray list, we found a group of genes with the GO classification of reproduction (Table 2), in which 5 genes for sex hormone receptors and pathways (ESRRG, DMRT3, DMRTA1, AMHR2 and FOXL2), could not only effectively separate malignant from premalignant samples but also classify poor and good prognosis with hierarchical clustering and SVM (Figure 9). In addition, two sex hormone genes had significant differentiated- expression of good prognosis to poor prognosis of GC (ESRRG 8.83, AR 0.37, p<0.01).
|Symbol||Fold change||Symbol||Fold change|
In the present study, we report for the first time that the custom microarray could be an effective method for diagnosis and prediction of prognosis in GC clinically. A lack of clinical biomarkers for early gastric cancer without any specific early symptoms leads to delayed diagnosis and contributes to the high mortality of gastric cancer . In some cases, changes occur only at the gene level, with no pathological change. Changes in gene expression could aid in early diagnosis, prognosis, and treatment guidance for postoperative radiation and chemotherapy. The development of microarray technologies enables the study of the possibility of pathological reversal and the evaluation and guidance of therapy. In addition, creating targeted microarray equipment might make the technique more useful. Because of the poor specificity and deficiency of mature joint diagnosis, the custom microarray has not been applied in clinical use for gastric cancer yet, although there are studies in gene diagnosis.
Microarray analysis is a widely used technology for studying gene expression on a global scale. Several molecular assays currently employed in the clinical assessment of cancers are derived from microarray-based gene expression profiling. One example of a microarray-based assay is MammaPrint, a custom microarray of 70 genes associated with the risk of the early development of distant metastasis in young patients with lymph-node negative breast cancer. MammaPrint has been ratified by the FDA. The ability to use this profile in a high throughput diagnostic setting could be a great advantage in the prognosis and treatment of breast cancer [17-19]. However, the technology is presently not used as a routine diagnostic tool in gastric cancer, and there has been no study of custom microarrays used in the diagnosis or prediction of prognosis. In this report, we demonstrate for the first time that the custom mini-array can be used as a reliable diagnostic tool in gastric cancer.
In this paper, we describe the development of a customized diagnostic gastric cancer mini-array and describe its reliable use in a diagnostic setting. Many clinical studies have correlated alterations in the expression of individual genes with gastric cancer outcome, often with contradictory results. Examples include CXCL1, HOXA10, and methylation of PCDH10 [24-26]. However, these genes were not included in our mini-array. It is possible that the other studies paid more attention to the functions of these genes, whereas we focused on the expressions of mRNAs. The 74-gene custom array may be a possible predictive tool for gastric cancer. The data clearly demonstrate the reproducibility and robustness of the small custom-made microarray. The use of a custom microarray could provide several advantages, such as accurate information on recurrence risk compared with conventional clinical criteria, and will thus improve the guidance for the requirement of adjuvant therapy.
Meanwhile, because of a 2-fold higher incidence in males than in females with GC , several larger epidemiological surveys suggest that gender was a significant independent prognostic factor for overall survival in GC patients[27,28] and male predominance of gastric cancer correlates with a 10–15 year delay in the onset of intestinal type gastric cancer in women compared with men . In this profile of the 74 gene custom mini-array, 5 genes were differentially expressed between malignant lesions and premalignant lesions of GC (ESRRG, DMRT3, DMRTA1, AMHR2 and FOXL2). This group of sex-associated genes with possible roles in GC was first proposed, and there are few studies on this group of genes associated with cancers. It is important to note that the latest study reported by Matson and colleagues showed a possible association and pathway of DMRT1, FOXL2 and the gender hormone . DMRT1, DMRT3, and DMRTA1 are all included in a cluster of the gene family that have a zinc finger-like DNA-binding motif (DM domain) in common, which is also a key regulator of male development in flies and nematodes. Furthermore, the main genes in this pathway are all included in our data. The data may provide us with a research direction for sex-associated genes in GC and reveal a possible pathway and mechanism of GC canceration.
Therefore, the array might be an excellent tool for classifying and predicting the outcome of disease in gastric cancer patients. However, there are some limitations in our profile. The samples should be expanded to verify the clinical validity and reproducibility.
Materials and Methods
Patients and samples
Forty surgically resected gastric cancer specimens and adjacent non-tumor specimens were obtained from Sir Run Run Shaw Hospital, School of Medicine Zhejiang University and were used during June 2007 to May 2011. We collected malignant and premalignant tissues from different regions of the resected stomach from each patient who underwent surgery. Forty grouped tissue samples from twenty patients with primary gastric cancer who underwent surgery (twenty malignant lesions, twenty premalignant lesions) were chosen for oligonucleotide microarray analysis (Table 1, detailed data in table S2). All of the collected samples were fixed, embedded, stained with H&E, and diagnosed with Lauren's and WHO classification independently by three professional pathologists. Twelve paired samples with malignant and premalignant lesions from patients who underwent surgery were chosen for quantitative reverse transcription-PCR (quantitative RT-PCR). The results from quantitative RT-PCR were compared with the pathological records from the contributing institution. Final pathological analysis was determined by consensus and reviewed if necessary. “Malignant” refers to various types of gastric cancers. “Premalignant” indicates atrophic gastritis, intestinal metaplasia and/or dysplasia. The specimens were immediately frozen in liquid nitrogen and stored at −80°C until further processing. Written informed consent was obtained before sample collection, and the study protocol was approved by the Clinical Research Ethics Committee of Sir Run Run Shaw Hospital.
The original customized 8*15k mini-array contained 1042 canceration and prognosis related genes identical to the probes on the original array . The mini-array included 696 differentially expressed genes between malignant lesions and premalignant lesions of GC patients that we found in previous 38, 500 gene chips, 44 collagen related genes, 54 genes for sex hormone receptors and pathways, and differentiated-expression genes found in other studies that were spotted in triplicates. Each array also includes 1042 probes for hybridization and printing quality control as well as 915 normalization genes (detailed data in table S1). Eight identical mini-arrays are present on a single 1" × 3" slide, allowing for eight individual hybridizations to be performed simultaneously (customized in Shanghai BioChip Co. Ltd., Agilent Technologies).
The total RNA was extracted and purified using the TRIzol Reagent (Invitrogen, USA) and RNeasy Mini Kit (Qiagen, Germany) via standard procedures recommended by the manufacturers. The levels and qualities of cRNA were measured by an Agilent 2100 Bioanalyzer (Agilent, USA), and the quality of RNA was controlled by the standard 2100 RIN>=7.0 and 28S/18S>=0.7. The cRNA was fragmented with the Gene Chip Sample Cleanup Module (Affymetrix, USA) and labeled with a single color using the Agilent enlarge notation method. Hybridization, staining, washing and scanning procedures were carried out as described in the Gene Chip Expression Analysis technical manual (Affymetrix, USA).
Analysis of oligonucleotide microarray data
The results of the microarray were scanned by an Agilent scanner. Data were normalized and conversed after image acquisition and quantification to identify the genes with significant differential expressions using Feature Extraction software. An open source interpreted computer language (R) was used for statistical computation and graphics . The raw data of custom microarray has been uploaded to the ArrayExpress as accession number A-MEXP-2338.
We used a method based on the gene expression profiles to classify gastric cancers into prognostic or diagnostic categories. The method included the following steps: (1) design of a custom mini-array with a group of genes possibly related to GC canceration and progress based on previous studies, (2) selection of differentiated-expression genes between malignant and premalignant lesions (fold change >2 times and p< 0.05) (3), unsupervised two-dimensional cluster analysis of gene clustering and GC clustering performed independently using an agglomerative hierarchical clustering algorithm (4), selection of discriminating candidate genes from genes selected in step 2 according to their correlation with the category (good or poor prognosis) (5), determination of the optimal set of reporter genes using a leave-one-out cross validation procedure (6), prognostic or diagnostic prediction based on the gene expression of the optimal set of reporter genes [18,19], (7) GO annotation of reporter genes, and (8) analysis of correlation of the microarray data with the prognostic profile
Quantitative RT-PCR analysis
The total RNA extracted from samples was reversely transcribed into cDNA using the prime script TM RT reagent kit (Takara, Japan) at 37°C for 15 min and at 85°C for 15 sec. PCR reactions using SYBR® premis EX Taq TM kit were performed at 95°C for 30 sec followed by 40 cycles at 95°C for 15 sec, 60°C for 10 sec, and 72°C for 40 sec with the 7500 Real-time PCR System (Applied Biosystems, USA). The housekeeping gene β-actin served as an internal control. The forward primer sequence of INHBA is GTGATGGCAAGGTCAACATCT, and the reverse one is GCGGTAGTGGTTGATGACTGT.
The information of genes in steps of methods for designing the mini-array and the final 74 mini-array genes.
Conceived and designed the experiments: TZ JS. Performed the experiments: YY. Analyzed the data: YY JL WZ. Contributed reagents/materials/analysis tools: YZ SC LW. Wrote the manuscript: YY TZ.
- 1. Parkin DM, Bray F, Ferlay J, Pisani P (2005) Global cancer statistics, 2002. CA Cancer J Clin 55: 74-108. doi:10.3322/canjclin.55.2.74. PubMed: 15761078.
- 2. Jemal A, Siegel R, Ward E, Hao Y, Xu J et al. (2008) Cancer statistics, 2008. CA Cancer J Clin 58: 71-96. doi:10.3322/CA.2007.0010. PubMed: 18287387.
- 3. Kampschöer GH, Fujii A, Masuda Y (1989) Gastric cancer detected by mass survey. Comparison between mass survey and outpatient detection. Scand J Gastroenterol 24: 813-817. doi:10.3109/00365528909089219. PubMed: 2799284.
- 4. Ohta H, Noguchi Y, Takagi K, Nishi M, Kajitani T et al. (1987) Early gastric carcinoma with special reference to macroscopic classification. Cancer 60: 1099-1106. doi:10.1002/1097-0142(19870901)60:5. PubMed: 3607727.
- 5. Endo M, Habu H (1990) Clinical studies of early gastric cancer. Hepatogastroenterology 37: 408-410. PubMed: 2170256.
- 6. Santoro E, Garofalo A, Scutari F, Zanarini T, Carlini M, et al. (1991) Early gastric cancer: total gastrectomy vs. distal resection. Results of a study of 271 cases. The Italian Stomach Cancer Group (A.C.O.I.-I.S.C.G.). Hepatogastroenterology 38: 427-429.
- 7. Yoshida S, Kozu T, Gotoda T, Saito D (2006) Detection and treatment of early cancer in high-risk populations. Best Pract Res Clin Gastroenterol 20: 745-765. doi:10.1016/j.bpg.2006.01.001. PubMed: 16997158.
- 8. Correa P (1992) Human gastric carcinogenesis: a multistep and multifactorial process--First American Cancer Society Award Lecture on Cancer Epidemiology and Prevention. Cancer Res 52: 6735-6740. PubMed: 1458460.
- 9. Kim JM, Sohn HY, Yoon SY, Oh JH, Yang JO et al. (2005) Identification of gastric cancer-related genes using a cDNA microarray containing novel expressed sequence tags expressed in gastric cancer cells. Clin Cancer Res 11: 473-482. PubMed: 15701830.
- 10. Jinawath N, Furukawa Y, Hasegawa S, Li M, Tsunoda T et al. (2004) Comparison of gene-expression profiles between diffuse- and intestinal-type gastric cancers using a genome-wide cDNA microarray. Oncogene 23: 6830-6844. doi:10.1038/sj.onc.1207886. PubMed: 15273739.
- 11. Lee HJ, Nam KT, Park HS, Kim MA, Lafleur BJ, et al. (2010) Gene expression profiling of metaplastic lineages identifies CDH17 as a prognostic marker in early stage gastric cancer. Gastroenterology 139: 213-225 e213.
- 12. Zhao Y, Zhou T, Li A, Yao H, He F et al. (2009) A potential role of collagens expression in distinguishing between premalignant and malignant lesions in stomach. Anat Rec (Hoboken) 292: 692-700. doi:10.1002/ar.20874.
- 13. Nardone G, Rocco A, Malfertheiner P (2004) Review article: helicobacter pylori and molecular events in precancerous gastric lesions. Aliment Pharmacol Ther 20: 261-270. doi:10.1111/j.1365-2036.2004.02075.x. PubMed: 15274662.
- 14. Tahara E (2004) Genetic pathways of two types of gastric cancer. IARC Sci Publ: 327-349. PubMed: 15055305.
- 15. Yuasa Y (2003) Control of gut differentiation and intestinal-type gastric carcinogenesis. Nat Rev Cancer 3: 592-600. doi:10.1038/nrc1141. PubMed: 12894247.
- 16. Cho WC (2011) Molecular diagnostics for monitoring and predicting therapeutic effect in cancer. Expert Rev Mol Diagn 11: 9-12. doi:10.1586/erm.10.111. PubMed: 21171915.
- 17. Glas AM, Floore A, Delahaye LJ, Witteveen AT, Pover RC et al. (2006) Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics 7: 278. doi:10.1186/1471-2164-7-278. PubMed: 17074082.
- 18. van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA et al. (2002) Gene expression profiling predicts clinical outcome of breast cancer. Nature 415: 530-536. doi:10.1038/415530a. PubMed: 11823860.
- 19. van de Vijver MJ, He YD, van't Veer LJ, Dai H, Hart AA et al. (2002) A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347: 1999-2009. doi:10.1056/NEJMoa021967. PubMed: 12490681.
- 20. Kinoshita J, Fushida S, Harada S, Makino I, Nakamura K et al. (2010) Type IV collagen levels are elevated in the serum of patients with peritoneal dissemination of gastric cancer. Oncol Lett 1: 989-994. PubMed: 22870099.
- 21. Lee KS, Park GS, Hong SH, Byun JH, Woo IS et al. (2010) Prognostic relevance of collagen XVIII expression in metastatic gastric carcinoma. Tumour Biol 31: 165-170. doi:10.1007/s13277-010-0022-z. PubMed: 20361288.
- 22. Yin Y, Zhao Y, Li AQ, Si JM (2009) Collagen: a possible prediction mark for gastric cancer. Med Hypotheses 72: 163-165. doi:10.1016/j.mehy.2008.09.010. PubMed: 18951724.
- 23. Carvajal A, Maciá MA, del Pozo JG, de Abajo F (2003) Small risk ratios may have strong public health impact. BMJ 327: 1050-1051. doi:10.1136/bmj.327.7422.1050-b. PubMed: 14593063.
- 24. Xu J, Zhang C, He Y, Wu H, Wang Z et al. (2012) Lymphatic endothelial cell-secreted CXCL1 stimulates lymphangiogenesis and metastasis of gastric cancer. Int J Cancer 130: 787-797. doi:10.1002/ijc.26035. PubMed: 21387301.
- 25. Sentani K, Oue N, Naito Y, Sakamoto N, Anami K et al. (2012) Upregulation of HOXA10 in gastric cancer with the intestinal mucin phenotype: reduction during tumor progression and favorable prognosis. Carcinogenesis 33: 1081-1088. doi:10.1093/carcin/bgs121. PubMed: 22411957.
- 26. Yu J, Cheng YY, Tao Q, Cheung KF, Lam CN, et al. (2009) Methylation of protocadherin 10, a novel tumor suppressor, is associated with poor prognosis in patients with gastric cancer. Gastroenterology 136: 640-651 e641.
- 27. Yang D, Hendifar A, Lenz C, Togawa K, Lenz F et al. (2011) Survival of metastatic gastric cancer: Significance of age, sex and race/ethnicity. J Gastrointest Oncol 2: 77-84. PubMed: 22811834.
- 28. Roshanaei G, Ghannad MS, Jafarabadi MA, Faradmal J, Sadighi S (2011) Prognostic impact of risk factors in patients with gastric cancer in Iran. Asian Pac J Cancer Prev 12: 3005-3008. PubMed: 22393980.
- 29. Sipponen P, Correa P (2002) Delayed rise in incidence of gastric cancer in females results in unique sex ratio (M/F) pattern: etiologic hypothesis. Gastric Cancer 5: 213-219. doi:10.1007/s101200200037. PubMed: 12491079.
- 30. Matson CK, Murphy MW, Sarver AL, Griswold MD, Bardwell VJ et al. (2011) DMRT1 prevents female reprogramming in the postnatal mammalian testis. Nature 476: 101-104. doi:10.1038/nature10239. PubMed: 21775990.