|Year : 2018 | Volume
| Issue : 12 | Page : 1135-1140
Tracking significant modules and key genes for esophageal squamous cell carcinoma based on differential modules
Meng-Li Zheng1, Nai-Kang Zhou2, Cheng-Hua Luo3
1 Department of Chest Surgery, The 309th Hospital, PLA, Beijing 100091, China
2 Department of Chest Surgery, General Hospital, PLA, Beijing 100853, China
3 Department of General Surgery, Peking University International Hospital, Beijing 100026, China
|Date of Web Publication||11-Dec-2018|
Department of Chest Surgery, The 309th Hospital, PLA, No. 17, Heishanhujia Road, Haidian District, Beijing 100091
Source of Support: None, Conflict of Interest: None
Background: The exact molecular mechanism of esophageal squamous cell carcinoma (ESCC) is still unknown, and the prognosis of ESCC has not been significantly improved.
Objective: To understand the molecular mechanism of ESCC, differential modules (DMs) and key genes were identified through conducting analysis on the differential co-expression network (DCN) based on the gene expression profiles of ESCC and protein–protein interaction (PPI) data.
Materials and Methods: First, gene expression profiles of ESCC and PPI data recruiting and preprocessing were conducted; then, a DCN was constructed based on the gene co-expression and gene differential expression in ESCC; in the following, candidate DMs were mined from DCN through a systemic module searching strategy, and significance analysis was performed on candidate DMs to identify DMs; moreover, significant genes contained in the DMs were analyzed to identify the underlying biomarkers for ESCC. Finally, pathway enrichment analysis was conducted to disclose the function of these DMs.
Results: A total of 10,975 genes were obtained after comprehensively preprocessing on the gene expression profiles and PPI data. Then, a DCN with 915 nodes (1164 interactions) was built, and 45 seed genes were identified. In the following, four DMs that separately enriched in phenylalanine metabolism, nicotine addiction, phenylalanine metabolism, and B-cell receptor signaling pathway were identified, where module 1 and module 3 were all enriched in phenylalanine metabolism pathway. Furthermore, the most significant seed gene myeloperoxidase (MPO) was contained in all of the DMs.
Conclusions: In this study, we successfully identified 4 DMs, three significant pathways, and a key gene MPO in ESCC, which might play key roles during the occurrence and development of ESCC and could be chosen as good indicators and therapeutic schedule for ESCC.
Keywords: Differential co-expression network, differential module, esophageal squamous cell carcinoma
|How to cite this article:|
Zheng ML, Zhou NK, Luo CH. Tracking significant modules and key genes for esophageal squamous cell carcinoma based on differential modules. J Can Res Ther 2018;14, Suppl S5:1135-40
|How to cite this URL:|
Zheng ML, Zhou NK, Luo CH. Tracking significant modules and key genes for esophageal squamous cell carcinoma based on differential modules. J Can Res Ther [serial online] 2018 [cited 2019 Dec 15];14:1135-40. Available from: http://www.cancerjournal.net/text.asp?2018/14/12/1135/189251
| > Introduction|| |
Esophageal squamous cell carcinoma (ESCC) is the predominant type comprising more than 90% of esophageal cancer, which is often deadly cancer diagnosed at the late stage with a 5-year survival rate <10% in advanced cancers. During the past few years, efforts have been made on the ESCC so as to disclose the pathological mechanism of it. However, the prognosis has not been significantly improved, and the prediction of ESCC clinical prognosis still depends on conventional pathologic variables such as tumor size, tumor grade, lymph node, and distal metastasis status. Therefore, the identification of sensitive and specific prognostic biomarkers would be of great clinical value to prevent or control metastatic progression.
Genes are tightly regulated to execute the proper biological functions in a cell for responding internal or external perturbations. Nowadays, systematic biological approaches have been used to reveal the overall physical and functional landscape of molecular changes. Knowledge of the protein–protein interaction (PPI) network provides a number of applications, such as prediction of proteins interaction and protein function, and identification of functional protein modules, disease candidate genes identification, and drug targets identification. Despite all of this exciting prior work in network mapping, at least one point stands out as remarkable. almost all physical and genetic networks, to date, have been examined under a single static (usually standard laboratory) condition. In such networks, two genes are connected and assumed to functionally interact if their expression profiles are correlated across multiple conditions. In this research, based on the PPI network, a differential co-expression network (DCN) was constructed to perform analysis on the ESCC. Furthermore, differential modules (DMs) were mined by integrating a set of genes that were not only differentially expressed under diseased states but also not exhibit correlated expression pattern in the network, so as to identify prognostic biomarkers for ESCC.
To achieve this, DMs and key genes were identified through conducting analysis on the gene expression profiles of ESCC and PPI data based on DCN. First, data recruiting and preprocessing were conducted; then, a DCN was constructed through combining the gene expression profiles and the PPI network; in the following, candidate DMs contained in the DCN were mined and significance analysis was performed to identify DMs and pathway enrichment analysis was conducted to disclose the function of these DMs; moreover, significant genes contained in the DMs were analyzed to identify the prognostic biomarkers for ESCC, so as to further understand the pathological mechanism of ESCC.
| > Materials and Methods|| |
Data recruiting and preprocessing
The gene expression profiles of ESCC, E-GEOD-29001, and E-GEOD-33426 were all downloaded from ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/). Both of these two gene expression profiles existed on A-AFFY-37-Affymetrix GeneChip Human Genome U133A 2.0 [HG-U133A_2] platform. E-GEOD-29001 comprised 45 samples (24 normal control samples and 21 ESCC samples), and E-GEOD-33426 comprised 71 samples (12 normal control samples and 59 ESCC samples). The microarray data and annotation files were all downloaded.
Prior to analysis, data preprocessing was separately conducted on the gene expression profiles to eliminate the influence of nonspecific hybridization. First of all, robust multichip average method was used to carry out background correction and quantiles-based algorithm was performed to conduct normalization. Then, Micro Array Suite 5.0 algorithm was used to revise perfect match and mismatch value, and the value of gene expression values was calculated through the median method. In the following, the gene expression profile on probe level was converted into gene symbol level, and the duplicated gene symbols were wiped off. Finally, inSilicoMerging package was used to combine the two sets of processed data, as well as eliminate the influence of batch processing of these two datasets, and a total of 12,441 genes were obtained.
Protein–protein interaction data
Search Tool for the Retrieval of Interacting Genes/Proteins (STRING, http://string-db.org/) is a database of known and predicted protein interactions. In the present study, the global PPI network was integrated from STRING. There were 787,896 interactions (16,730 genes) of human beings in all. Mapping 12,441 genes from the gene expression profiles on the global PPI network, a specific PPI network contained 10,975 genes was obtained.
Differential co-expression network construction
To further disclose the relationships between the interactions we identified above and ESCC, a DCN was constructed on them. To be specific, the DCN construction contained two steps. First of all, a binary co-expression network was built. The Pearson's correlation coefficients (PCC) were applied to calculate the correlation of each gene pairs in disease condition to identify the co-expression relationship. In the present study, the absolute value of the PCC ≥0.8 was set as the cutoff value and these edges whose absolute value of the PCC ≥0.8 were chosen to construct the binary co-expression network. The second step was to assign a weight value to each edge of the binary co-expression network, which was based on differential gene expression between disease condition and normal condition. First, the one-sided Student's t-test was used to determine the P values of differential gene expression between disease and normal conditions. Then, EdgeR, a Bioconductor package for differential expression analysis of digital gene expression data, was utilized to detect differential expression. The weight wm, n on edge (m, n) in the co-expression network was calculated as following:
Where Pm and Pn were P values of differential expression for genes m and n, respectively. N was the node set of the co-expression network, and cor (m, n) was the absolute value of PCC between genes m, n based on their expression profiles.
In this case, a DCN with each of the edge was assigned a weight value was built. Under this weighting scheme, genes that were co-expressed and significantly differentially expressed were assigned higher weights, which might indicate that these genes exhibited differential activities between two conditions. Therefore, these genes might play important roles during the occurrence and development of ESCC.
Identification of candidate differential modules in differential co-expression network
Having constructed the DCN, candidate DMs were mined to identify disrupted or altered modules between disease and normal conditions. In the present study, the candidate DM search included the following three steps: Seed prioritization to seek seed genes, module search by seed expansion, and refinement of candidate modules.
Seed prioritization to seek seed genes
In the beginning, degree centrality analysis was conducted on the DCN and the genes were ranked in descending order according to their degree values. Then, the importance of each gene in the corresponding network, which we denoted as Z-score was determined based on the following formula:
Where Z-score (m) denoted the importance of vertex m in the corresponding network; Ni(m) denoted the set of neighbors of m in Gi; A'i denoted the degree normalized weighted adjacency matrix. The product A'-g denoted the information propagation on the network through the edges of networks, which meant the importance of a node depends on the number of its neighbors, strength of connection, and importance of its neighbors. Then, all of the Z-score values were ranked in descending order, and the top 5% genes were selected as seed genes.
Candidate differential modules searching
In the step of candidate DMs searching, for each seed gene m, we considered it as a DM T. Then, candidate DMs searching was started from seed gene m, the network gene n that adjacent to m was added to T to form a module T'. The entropy decrease between these two modules was calculated according to the following formula:
ΔH (T', T) >0 indicated that addition of vertex n improved the connectivity of the former module T. The genes that adjacent to m were iteratively added to the module T until there was no increase in the objective function of ΔH. In this case, all of the genes were connected together to form a candidate DM.
Refinement of candidate modules
As we known, the modules with too small amount of genes might not be significant enough and were always ignored by the research. Therefore, in the present study, only these candidate DMs whose sizes ≥5 were thought to be significant in the refinement step, so these candidate DMs with genes sizes <5 were wiped out. Moreover, Jaccard index was applied to measure the ratio of intersection over the union of two sets, and if the Jaccard index ≥0.5, these two sets were merged into a module.
Identifying differential modules
Having identified the candidate DMs, we calculated the significance of candidate DMs to identify DMs. Randomized networks were constructed based on the null score distribution of the candidate DMs. Module search was performed on the randomized networks according to the methods mentioned above. Each network was completely randomized for 100 times. The empirical P value of a DM was calculated as the probability of the module having the observed score or smaller by chance. Benjamini and Hochberg method was performed to conduct multiple testing on the P values. The modules with adjusted P ≤ 0.05 were considered as DMs.
Pathway enrichment analysis of the differential modules
To further disclose the function of these DMs, the module identification algorithm in Genelibs (http://www.genelibs.com/gb/index.jsp), which is based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, was performed to conduct pathway enrichment analysis on the genes contained in each DM, respectively. The Fisher's exact test was utilized to determine the P values of the enrichment condition and Benjamini–Hochberg method was performed to conduct multiple testing on the P values. The pathways of whose adjusted P < 0.05 were considered as the pathways that the certain module enriched in. Moreover, the pathway that with the minimum adjusted P value was considered as the significant pathway that the module enriched in.
| > Results|| |
Constructing differential co-expression network
In the present research, the analysis was conducted on gene expression profiles of E-GEOD-29001 and E-GEOD-33426 to identify DMs in ESCC. After having recruited and preprocessed the gene expression profiles and the PPI data, a specific PPI network was obtained, including 10,975 genes and 579,722 interactions. Under the threshold value of the absolute value of PCC ≥ 0.8, a binary co-expression network contained 915 nodes and 1164 interactions were gained. Then, a weight value was assigned to each edge in the binary co-expression network based on differential gene expression between two conditions, and the DCN was finally built [Figure 1].
|Figure 1: The differential co-expression network that constructed based on these two gene expression profiles of esophageal squamous cell carcinoma and the protein–protein interactions under the cutoff value of the absolute value of Pearson's correlation coefficients ≥0.8. There were 915 (1164 interactions) included in the network, where yellow nodes were on behalf of the seed genes|
Click here to view
Identifying candidate differential modules
As it was indicated in the methods above, the candidate DMs in DCN were identified following the steps of seed prioritization to seek seed genes, module search by seed expansion and refinement of candidate modules. By ranking all of the genes in descending order and setting the top 5% genes as a cutoff value, we gained 45 seed genes in all, and these genes were considered to be significant for ESCC. The details of these seed genes were listed in [Table 1]. Then, candidate DMs were searched by expanding these seed genes. Finally, 14 candidate DMs were identified through removing those modules with genes sizes <5, as well as merging those two sets of whose Jaccard index ≥0.5 into a module.
|Table 1: The details of the seed genes that ranking according to their Z-score values in descending order|
Click here to view
Identifying differential modules
To identify DMs from the candidate DMs, statistically significant analysis was conducted on them. Each randomized network was built through randomly grabbing 1164 interactions from the 579,722 gene interactions, and then module search was performed on the randomized networks according to the methods mentioned above. After having completely randomized each network for 100 times, we obtained 4 DMs in all under the threshold value of P ≤ 0.05: module 1 (P = 0.0360), module 2 (P = 0.0163), module 3 (P = 0.0005), module 4 (P = 0.0190), and the details listed in [Table 2]. We could easily found that there were separately 27 interactions (19 genes), 21 interactions (15 genes), 12 interactions (9 genes), and 9 interactions (8 genes) in module 1, module 2, module 3, and module 4 from [Figure 2].
|Table 2: The details of the seed genes that the four differential modules contained and the pathways these four differential modules enriched in|
Click here to view
|Figure 2: The differential modules for the esophageal squamous cell carcinoma under the threshold value of P ≤ 0.05. (a-d) separately represented module 1, module 2, module 3, and module 4. Yellow nodes represented the seed genes|
Click here to view
Pathway enrichment analysis of the differential modules
As pathway enrichment was separately conducted on these DMs, and under the threshold value of adjusted P < 0.05, we separately obtained 1, 1, 6, and 1 pathway term that module 1, module 2, module 3, and module 4 enriched in. Moreover, as the pathway with the minimum adjusted P value was considered to be the significant pathway, we gained that module 1 was enriched in phenylalanine metabolism (P = 0.00568), module 2 was enriched in Nicotine addiction (P = 0.0116), module 3 was enriched in phenylalanine metabolism (P = 0.0356), and module 4 was enriched in B-cell receptor signaling pathway (P = 0.0412), where module 1 and module 3 were all enriched in phenylalanine metabolism pathway. We predicted that these four DMs mainly affected the three pathways to influence the disease of ESCC.
| > Discussion|| |
ESCC, the most commonly observed histologic subtype of esophageal cancer, is believed to develop through the accumulation of numerous genetic alterations, including inactivation of tumor suppressor genes and activation of oncogenes. The disease is one of the most common cancers and ranks as the 6th leading cause of cancer-related deaths worldwide. Genetic networks, which chart pairs of genetic mutations that in combination cause lethality or other phenotypes, have been widely used to perform analysis on diseases to disclose the pathogenesis of diseases., Moreover, the network properties of complex disease alterations analysis have noted that the complex diseases alterations tend to cluster within closely knit network modules or communities.,
Therefore, in the present study, to further understand the molecular mechanism of ESCC, based on the DCN, DM analysis was conducted through combing the gene expression profile of E-GEOD-29001, E-GEOD-33426, and PPI network. A total of four DMs were obtained under the threshold value of P ≤ 0.05. In addition, pathway enrichment analysis indicated that pathways of phenylalanine metabolism, nicotine addiction, and B-cell receptor signaling pathway were significant pathways for these four DMs. Furthermore, as conducting analysis on the frequencies of the seed enriched in the modules, we found that myeloperoxidase(MPO) was contained in all of the DMs. Meanwhile, the Z-score value of the gene MPO was the highest. Therefore, we predicted that gene MPO might play a very important role during the occurrence and development of ESCC. To further disclose the relationship between ESCC and MPO, the discussion was conducted in the following.
MPO is an enzyme found primarily in the lysosomes of neutrophils, which activates carcinogens in tobacco smoking, including aromatic amines, and catalyzes the endogenous formation of carcinogenic free radicals. The MPO gene is located on chromosome 17q23.1, consists of 11 introns and 12 exons. It covers 11.10 kb, from 56,832,934 to 56,821,840, on the reverse strand. MPO and its reactive by-products have been linked to generation or activation of carcinogens such as benzo (a) pyrene and aromatic amines, DNA strand breakage, and inhibition of DNA repair. It has been reported that gene MPO is associated with several cancers, such as breast cancer, lung cancer, ovarian cancer, cervix cancer, and so on. By conducting case–control analysis, Matsuo et al. indicated that the MPO-463 G/A polymorphism to be linked to esophageal cancer susceptibility, which showed the allele frequency for MPO 463A to be 8.2% for cases and 10.5% for controls. However, there is still no direct report indicates that ESCC is related to MPO. In the present, topological analysis on the DCN indicated that the Z-score value was the highest. Moreover, the DMs analysis indicated that MPO was contained in all of the four DMs. In this case, the roles of MPO played in the ESCC cannot be ignored. Therefore, further analysis should be conducted to disclose the relationship between ESCC and MPO.
However, there were still some drawbacks in our research which must be taken into account. First of all, even though the sample size was large enough to some degree, however, the microarray data were downloaded from the existed database, not obtained by ourselves. Second, the results of the bioinformatics methods were not verified through experimental verification analysis, so the exact conclusion could only be determined after experimental verification analysis on the gene. Although disadvantages existed, we believed that this method and the results offered investigators valuable resources for better understanding the underlying mechanisms ESCC on the gene level.
| > Conclusions|| |
In a word, four DMs were identified via conducting analysis on the gene expression profiles of ESCC and PPI data based on the strategy of DCN. These four DMs might affect the pathways of phenylalanine metabolism, nicotine addiction, and B-cell receptor signaling pathway to perform their functions during the occurrence and development of ESCC. Furthermore, gene MPO was considered to have an important position in ESCC, and we predict that MPO could be chosen as a good indicator and therapeutic schedule for ESCC.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| > References|| |
Ko JM, Wang LD, Lung M. Germline PALB2 genetic variations of familial esophageal squamous cell carcinoma (ESCC) in Henan, a high risk ESCC region. Cancer Res 2014;74:1280.
Hu H, Yang J, Sun Y, Yang Y, Qian J, Jin L, et al.
Putatively functional PLCE1 variants and susceptibility to esophageal squamous cell carcinoma (ESCC): A case-control study in Eastern Chinese populations. Ann Surg Oncol 2012;19:2403-10.
Yu C, Chen K, Zheng H, Guo X, Jia W, Li M, et al.
Overexpression of astrocyte elevated gene-1 (AEG-1) is associated with esophageal squamous cell carcinoma (ESCC) progression and pathogenesis. Carcinogenesis 2009;30:894-901.
Wu B, Li C, Du Z, Yao Q, Wu J, Feng L, et al.
Network based analyses of gene expression profile of LCN2 overexpression in esophageal squamous cell carcinoma. Sci Rep 2014;4:5403.
Kostka D, Spang R. Finding disease specific alterations in the co-expression of genes. Bioinformatics 2004;20 Suppl 1:i194-9.
Cheng CP, Kuo IY, Alakus H, Frazer KA, Harismendy O, Wang YC, et al.
Network-based analysis identifies epigenetic biomarkers of esophageal squamous cell carcinoma progression. Bioinformatics 2014;30:3054-61.
Li Y, Li J. Disease gene identification by random walk on multigraphs merging heterogeneous genomic and phenotype data. BMC Genomics 2012;13 Suppl 7:S27.
Ideker T, Krogan NJ. Differential network biology. Mol Syst Biol 2012;8:565.
Ma X, Gao L, Karamanlidis G, Gao P, Lee CF, Garcia-Menendez L, et al.
Revealing pathway dynamics in heart diseases by Analyzing Multiple Differential Networks. PLoS Comput Biol 2015;11:e1004332.
Yan W, Shih JH, Rodriguez-Canales J, Tangrea MA, Ylaya K, Hipp J, et al.
Identification of unique expression signatures and therapeutic targets in esophageal squamous cell carcinoma. BMC Res Notes 2012;5:73.
Yan W, Shih J, Rodriguez-Canales J, Tangrea MA, Player A, Diao L, et al.
Three-dimensional mRNA measurements reveal minimal regional heterogeneity in esophageal squamous cell carcinoma. Am J Pathol 2013;182:529-39.
Ma L, Robinson LN, Towle HC. ChREBP*Mlx is the principal mediator of glucose-induced gene expression in the liver. J Biol Chem 2006;281:28721-30.
Rifai N, Ridker PM. Proposed cardiovascular risk assessment algorithm using high-sensitivity C-reactive protein and lipid screening. Clin Chem 2001;47:28-30.
Pepper SD, Saunders EK, Edwards LE, Wilson CL, Miller CJ. The utility of MAS5 expression summary and detection call algorithms. BMC Bioinformatics 2007;8:273.
Giorgi FM, Bolger AM, Lohse M, Usadel B. Algorithm-driven artifacts in median polish summarization of microarray data. BMC Bioinformatics 2010;11:553.
Benesty J, Chen J, Huang Y, Cohen I. Pearson correlation coefficient. In Noise Reduction in Speech Processing. Berlin: Springer; 2009. p. 1-4.
Robinson MD, McCarthy DJ, Smyth GK. EdgeR: A Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 2010;26:139-40.
Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Stat Soc Series B Methodol 1995;57: 289-300.
Routledge R. Fisher's exact test. Encyclopedia of Biostatistics. Hoboken: Wiley Online Library; 2005.
Haynes W. Benjamini–Hochberg method. In: Encyclopedia of Systems Biology. Berlin: Springer; 2013. p. 78.
Ma S, Bao JY, Kwan PS, Chan YP, Tong CM, Fu L, et al.
Identification of PTK6, via RNA sequencing analysis, as a suppressor of esophageal squamous cell carcinoma. Gastroenterology 2012;143:675-86.e1-12.
Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer statistics, 2012. CA Cancer J Clin 2015;65:87-108.
Typas A, Nichols RJ, Siegele DA, Shales M, Collins SR, Lim B, et al.
High-throughput, quantitative analyses of genetic interactions in E. coli
. Nat Methods 2008;5:781-7.
Horn T, Sandmann T, Fischer B, Axelsson E, Huber W, Boutros M. Mapping of signaling networks through synthetic genetic interaction analysis by RNAi. Nat Methods 2011;8:341-6.
Cerami E, Demir E, Schultz N, Taylor BS, Sander C. Automated network analysis identifies core pathways in glioblastoma. PLoS One 2010;5:e8918.
Wu G, Feng X, Stein L. A human functional protein interaction network and its application to cancer data analysis. Genome Biol 2010;11:R53.
Li D, Diao Y, Li H, Fang X, Li H. Association of the polymorphisms of MTHFR C677T, VDR C352T, and MPO G463A with risk for esophageal squamous cell dysplasia and carcinoma. Arch Med Res 2008;39:594-600.
Kiyohara C, Yoshimasu K, Takayama K, Nakanishi Y. NQO1, MPO, and the risk of lung cancer: A HuGE review. Genet Med 2005;7:463-78.
Josephy PD. The role of peroxidase-catalyzed activation of aromatic amines in breast cancer. Mutagenesis 1996;11:3-7.
Knaapen AM, Schins RP, Borm PJ, van Schooten FJ. Nitrite enhances neutrophil-induced DNA strand breakage in pulmonary epithelial cells by inhibition of myeloperoxidase. Carcinogenesis 2005;26:1642-8.
Nichols JA, Katiyar SK. Skin photoprotection by natural polyphenols: Anti-inflammatory, antioxidant and DNA repair mechanisms. Arch Dermatol Res 2010;302:71-83.
Pabalan N, Jarjanazi H, Sung L, Li H, Ozcelik H. Menopausal status modifies breast cancer risk associated with the myeloperoxidase (MPO) G463A polymorphism in Caucasian women: A meta-analysis. PLoS One 2012;7:e32389.
Yu PP, Shiao SP, Suarez M. A meta-analysis of lifestyle factors with MPO and GSTM1 human genes in lung cancer prevention. Cancer Res 2014;74:5050.
Papineni RV, Craveiro V, Pizzonia J, Holmberg J, Mor G. Non-invasive determination of in vivo
myeloperoxidase activity in ovarian cancer mouse model. Cancer Res 2012;72:5259.
Castelão C, da Silva AP, Matos A, Inácio Â, Bicho M, Medeiros R, et al.
Association of myeloperoxidase polymorphism (G463A) with cervix cancer. Mol Cell Biochem 2015;404:1-4.
Matsuo K, Hamajima N, Shinoda M, Hatooka S, Inoue M, Takezaki T, et al.
Possible risk reduction in esophageal cancer associated with MPO-463 A allele. J Epidemiol 2001;11:109-14.
[Figure 1], [Figure 2]
[Table 1], [Table 2]