Potential colon cancer biomarker search using more than two performance measures in a multiple criteria optimization approach.
Cancer is the second most relevant cause of death worldwide (1). Microarray experiments aim to measure the change in the genetic expression of tens of thousands of genes simultaneously and have been used to generate many of the genetic pipelines in cancer research (2). When considering normal and cancer tissues, genes with the highest differential expression between these states are potential cancer biomarkers. Many and varied methodologies have been proposed for the identification of these genes (3). Our research group has proposed that the identification of potential cancer biomarkers using microarray data be carried out through Multiple Criteria Optimization (MCO) techniques (4).An MCO problem aims to find the best compromises between two or more conflicting criteria considered. The best compromises are located in the so-called "efficient frontier" of the MCO problem. Results of two or more analyses for a set of genes can be used as conflicting criteria that can be accommodated in an MCO problem. Our hypothesis is that the genes located in the efficient frontier of the related MCO problem are potential cancer biomarkers.
Data Envelopment Analysis (DEA) has been identified as being particularly well suited to the task of identifying the efficient frontiers of MCO problems (5). Here, the ability of the proposed method is explored using more than two performance measures and solved through DEA. Results show that the method identifies a consistent set of genes when increasing the number of performance measures in the MCO problem. It is also observed that convergence of a larger number of potential biomarkers is faster with additional criteria, i.e., more p-values.
Methods
Microarray Data
A colon cancer microarray database was selected for the described exploration. This database was first reported in Alon et al. (6) and is available at www.molbio.princeton.edu/ colondata. It contains the measured expression of 6,500 genes in 22 normal tissues and 40 cancer tissues, all of which were characterized using Affymetrix Hum6000 arrays.
Statistical Analysis
Statistical comparisons between the normal and cancer replicates were performed using the Mann-Whitney nonparametric test. The procedure is illustrated in Figure 1. P-values from different statistical analyses were obtained using partial permutations leaving one, two, or three tissues out of each state; the excluded tissues were selected randomly. From these comparisons, a total of ten different p-values were obtained for each gene. A p-value in the Mann-Whitney test is understood as representing the probability of finding a particular difference of medians between the two states by pure chance. Thus, to favor finding truly significant differences, low p-values are sought.
[FIGURE 1 OMITTED]
Multiple Criteria Optimization and Data Envelopment Analysis
Considering that the aim is to find those genes that change their expressions to the greatest degree between the different states, a p-value can be seen as a criterion to be minimized: smaller p-values show stronger evidence for the rejection of the stated null hypothesis, which relates to not having a significant difference between the two states. Thus, one can build an MCO problem considering the different p-values available for each gene as criteria intended to be minimized simultaneously.
The case of an MCO problem using two different p-values is presented in Figure 2(a). Given the minimization objective for both p-values, the efficient frontier is located in the southwest corner. In order to use DEA to find the efficient frontier, it is necessary to maximize at least one of the conflicting criteria, so a transformation (shown in Figure 2b) should be performed on at least one of the considered p-values. For the instances presented here, half of the p-values were transformed.
Graphical representation becomes complicated when using more than two p-values; however, the use of DEA to find the efficient frontier can be extended to the desired number of dimensions easily without a loss of generality. Banker-Charnes-Cooper (BCC) input- and output-oriented DEA models were used for the frontier search. For this search, genes identified in the previous frontier were removed from the original list and the search process repeated until the tenth frontier was reached. The instances presented here correspond to the use of 2, 4 and 8 p-values for the MCO problem. Results obtained from the different combinations were compared.
In order to express the idea of the method in a simple manner, one can think of a small p-value for a particular gene as being an indicator of its importance. Taking tissues out of the dataset creates a series of somewhat different datasets that allow the computation of multiple p-values for all genes. If all of the p-values for a particular gene are small, then it is likely that that gene will be significantly differentially expressed. Genes with these characteristics tend to cluster along the particular edges of the set of genes under analysis. MCO's objective is to find this specific edge (efficient frontier) and the genes lying on it.
MCO Gene Selection Validation
Since the purpose of this study was to see how useful it would be to model the microarray data analysis for potential biomarker identification as an MCO problem, the results needed to be validated. The validation was performed by undertaking a literature search for the genes identified by this method that changed their expression to the greatest degree between normal and cancer tissues.
[FIGURE 2 OMITTED]
Results
One of the most notable results was that the number of genes identified in the efficient frontiers increased as the number of p-values that were considered in the model also increased. Table 1 shows the genes found in the different combinations of p-values; the information about each identified gene is followed by the frontier where each gene was localized in the corresponding run.
All of the genes selected by the analysis but one, GTF3A, have been previously reported to change their expressions in colorectal cancer and/or other cancer types (Table 2). These reports are based on in vitro and/or in vivo experiments. Even though the role of GTF3A in cancer is still not confirmed, it is quite possible that changes in its expression could be related to cancer development. The GTF3A gene product is a transcription factor that regulates expression of 5 S ribosomal RNA.
Discussion
The MCO method of searching for biomarkers using available microarray data was demonstrated to be robust through the use of a different number of statistical p-values identifying a consistent set of genes. This approach may contribute to the rapid identification of genes by their biological validation as contributors to cancer. The method can also be explored using different DEA models and different types of available data, thus opening several opportunities for the meta-analysis of microarray experiments.
Acknowledgments
M.S.P was supported by a research assistantship from the Industrial Engineering Department at UPRM. Authors acknowledge the support of BioSEI Grant 33 010 3080 301 (awarded to M.C.R.) and that of the PROMEP project (103.5/07/2523, granted to C.I.).
References
(1.) American Cancer Society. Cancer Facts & Figures--2010. Available at: Url: http://www.cancer.org/acs/groups/content/@nho/documents/document/acspc- 024113.pdf.
(2.) Berns A. Gene expression in diagnosis. Nature 2000;403:491-2.
(3.) Tainsky MA. Genomic and proteomic biomarkers for cancer: A multitude of opportunities. Biochimica Biophys Acta 2009;1796:176-93.
(4.) Isaza C, Sanchez-Pena M, Rodriguez C, Cabrera M. Abstract B45: An optimization-based approach to potential biomarker identification with microarray data [Internet]. In: Poster Presentations--Other Biomarkers and Early Detection Topics. Philadelphia PA: Cancer Prevention Research --AACR; p. Supplement 2.
(5.) Castro C, Cabrera-Rios M, Lilly B, Castro JM, Mount-Campbell CA. Identifying the best compromises between multiple performance measures in injection molding (IM) using Data Envelopment Analysis (DEA). J Integr Des Process Sci 2003;7:77-86.
(6.) Alon U, Barkai N, Notterman DA, Gish K, Ybarra S, Mack D, et al. Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc Natl Acad Sci U S A 1999;96:6745-50.
(7.) Ligr M, Patwa RR, Daniels G, Pan L, Wu X, Li Y, et al. Expression and function of androgen receptor coactivator p44/Mep50/WDR77 in ovarian cancer. PLoS One 2011;6:e26250.
(8.) Gu Z, Zhou L, Gao S, Wang Z. Nuclear transport signals control cellular localization and function of androgen receptor cofactor p44/WDR77. PLoS One 2011;6:e22395.
(9.) Ruan W, Wang Y, Ma Y, Xing X, Lin J, Cui J, Lai M. HSP60, a protein downregulated by IGFBP7 in colorectal carcinoma. J Exp Clin Cancer Res 2010;29:41.
(10.) McLean MH, Murray GI, Stewart KN, Norrie G, Mayer C, Hold GL, Thomson J, Fyfe N, Hope M, Mowat NA, Drew JE, El-Omar EM. The inflammatory microenvironment in colorectal neoplasia. PLoS One 2011;6:e15366.
(11.) Ning Y, Manegold PC, Hong YK, Zhang W, Pohl A, Lurje G, Winder T, Yang D, LaBonte MJ, Wilson PM, Ladner RD, Lenz HJ. Interleukin-8 is associated with proliferation, migration, angiogenesis and chemosensitivity in vitro and in vivo in colon cancer cell line models. Int J Cancer 2011;128:2038-49.
(12.) Riener MO, Vogetseder A, Pestalozzi BC, Clavien PA, Probst-Hensch N, Kristiansen G, Jochum W. Cell adhesion molecules P-cadherin and CD24 are markers for carcinoma and dysplasia in the biliary tract. Hum Pathol 2010;41:1558-65.
(13.) Boonstra JJ, van Marion R, Douben HJ, Lanchbury JS, Timms KM, Abkevich V, Tilanus HW, de Klein A, Dinjens WN. Mapping of homozygous deletions in verified esophageal adenocarcinoma cell lines and xenografts. Genes Chromosomes Cancer 2012;51:272-82.
(14.) Chen M, Zhang J, Manley JL. Turning on a fuel switch of cancer: hnRNP proteins regulate alternative splicing of pyruvate kinase mRNA. Cancer Res 2010;70:8977-80.
(15.) Chen SH, Yang W, Fan Y, Stocco G, Crews KR, Yang JJ, Paugh SW, Pui CH, Evans WE, Relling MV. A genome-wide approach identifies that the aspartate metabolism pathway contributes to asparaginase sensitivity. Leukemia 2011;25:66-74.
(16.) Thorsen K, Mansilla F, Schepeler T, 0ster B, Rasmussen MH, Dyrskjot L, Karni R, Akerman M, Krainer AR, Laurberg S, Andersen CL, Orntoft TF. Alternative splicing of SLC39A14 in colorectal cancer is regulated by the Wnt pathway. Mol Cell Proteomics 2011;10:M110.002998.
(17.) Nagaraj SH, Reverter A. A Boolean-based systems biology approach to predict novel genes associated with cancer: Application to colorectal cancer. BMC Syst Biol 2011;5:35.
(18.) Cermak V, Kosla J, Plachy J, Trejbalova K, Hejnar J, Dvorak M. The transcription factor EGR1 regulates metastatic potential of v-src transformed sarcoma cells. Cell Mol Life Sci 2010;67:3557-68.
(19.) Gotte M, Mohr C, Koo CY, Stock C, Vaske AK, Viola M, Ibrahim SA, Peddibhotla S, Teng YH, Low JY, Ebnet K, Kiesel L, Yip GW. miR-145 dependent targeting of junctional adhesion molecule A and modulation of fascin expression are associated with reduced breast cancer cell motility and invasiveness. Oncogene 2010;29:6569-80.
(20.) Dowen SE, Crnogorac-Jurcevic T, Gangeswaran R, Hansen M, Eloranta JJ, Bhakta V, Brentnall TA, Luttges J, Kloppel G, Lemoine NR. Expression of S100P and its novel binding partner S100PBPR in early pancreatic cancer. Am J Pathol 2005;166:81-92.
(21.) Rehbein G, Simm A, Hofmann HS, Silber RE, Bartling B. Molecular regulation of S100P in human lung adenocarcinomas. Int J Mol Med 2008;22:69-77.
(22.) Guerreiro Da Silva ID, Hu YF, Russo IH, Ao X, Salicioni AM, Yang X, Russo J. S100P calcium-binding protein overexpression is associated with immortalization of human breast epithelial cells in vitro and early stages of breast cancer development in vivo. Int J Oncol 2000;16:231-40.
(23.) Jiang L, Lai YK, Zhang J, Wang H, Lin MC, He ML, Kung HF. Targeting S100P inhibits colon cancer growth and metastasis by Lentivirus-mediated RNA interference and proteomic analysis. Mol Med 2011;17:709-16.
(24.) Bertucci F, Salas S, Eysteries S, Nasser V, Finetti P, Ginestier C, Charafe-Jauffret E, Loriod B, Bachelart L, Montfort J, Victorero G, Viret F, Ollendorff V, Fert V, Giovaninni M, Delpero JR, Nguyen C, Viens P, Monges G, Birnbaum D, Houlgatte R. Gene expression profiling of colon cancer by DNA microarrays and correlation with histoclinical parameters. Onco gene 2004;23:1377-91.
(25.) Claerhout S, Lim JY, Choi W, Park YY, Kim K, Kim SB, Lee JS, Mills GB, Cho JY. Gene expression signature analysis identifies vorinostat as a candidate therapy for gastric cancer. PLoS One 2011;6:e24662.
(26.) Ye H, Yu T, Temam S, Ziober BL, Wang J, Schwartz JL, Mao L, Wong DT, Zhou X. Transcriptomic dissection of tongue squamous cell carcinoma. BMC Genomics 2008;9:69.
(27.) Notterman DA, Alon U, Sierk AJ, Levine AJ. Transcriptional Gene Expression Profiles of Colorectal Adenoma, Adenocarcinoma, and Normal Tissue Examined by Oligonucleotide Arrays. Cancer Res 2001;61:3124-30.
(28.) Hong Y, Ho KS, Eu KW, Cheah PY. A susceptibility gene set for early onset colorectal cancer that integrates diverse signaling pathways: implication for tumorigenesis. Clin Cancer Res 2007;13:1107-14.
(29.) Czarnecka AM, Campanella C, Zummo G, Cappello F. Heat shock protein 10 and signal transduction: a "capsula eburnea" of carcinogenesis? Cell Stress Chaperones 2006;11:287-94.
(30.) Chung JH, Lee HJ, Kim BH, Cho NY, Kang GH. DNA methylation profile during multistage progression of pulmonary adenocarcinomas. Virchows Arch 2011;459:201-11.
(31.) Arriaga JM, Levy EM, Bravo AI, Bayo SM, Amat M, Aris M, Hannois A, Bruno L, Roberti MP, Loria FS, Pairola A, Huertas E, Mordoh J, Bi anchini M. Metallothionein expression in colorectal cancer: relevance of different isoforms for tumor progression and patient survival. Hum Pathol 2012;43:197-208.
(32.) Somberg M, Li X, Johansson C, Orru B, Chang R, Rush M, Fay J, Ryan F, Schwartz S. Serine/arginine-rich protein 30c activates human papillomavirus type 16 L1 mRNA expression via a bimodal mechanism. J Gen Virol 2011;92(Pt 10):2411-21.
(33.) Cloutier P, Toutant J, Shkreta L, Goekjian S, Revil T, Chabot B. Antagonistic effects of the SRp30c protein and cryptic 5' splice sites on the alternative splicing of the apoptotic regulator Bcl-x. J Biol Chem 2008;283: 21315-24.
(34.) Wei JJ, Wu X, Peng Y, Shi G, Basturk O, Yang X, Daniels G, Osman I, Ouy ang J, Hernando E, Pellicer A, Rhim JS, Melamed J, Lee P. Regulation of HMGA1 expression by microRNA-296 affects prostate cancer growth and invasion. Clin Cancer Res 2011;17:1297-305.
(35.) Esposito F, Tornincasa M, Chieffi P, De Martino I, Pierantoni GM, Fusco A. High-mobility group A1 proteins regulate p53-mediated transcription of Bcl-2 gene. Cancer Res 2010;70:5379-88.
(36.) Bacher U, Haferlach T, Fehse B, Schnittger S, Kroger N. Minimal residual disease diagnostics and chimerism in the post-transplant period in acute myeloid leukemia. ScientificWorldJournal 2011;11:310-9.
(37.) Bacher U, Schnittger S, Haferlach T. Molecular genetics in acute myeloid leukemia. Curr Opin Oncol 2010;22:646-55.
(38.) Yi Y, Nandana S, Case T, Nelson C, Radmilovic T, Matusik RJ, Tsuchiya KD. Candidate metastasis suppressor genes uncovered by array comparative genomic hybridization in a mouse allograft model of prostate cancer. Mol Cytogenet 2009;2:18.
(39.) Jia D, Wei L, Guo W, Zha R, Bao M, Chen Z, Zhao Y, Ge C, Zhao F, Chen T, Yao M, Li J, Wang H, Gu J, He X. Genome-wide copy number analyses identified novel cancer genes in hepatocellular carcinoma. Hepatology 2011;54:1227-36.
(40.) Arentz G, Chataway T, Price TJ, Izwan Z, Hardi G, Cummins AG, Hardingham JE. Desmin expression in colorectal cancer stroma correlates with advanced stage disease and marks angiogenic microvessels. Clin Proteomics 2011;8:16.
Erika Watts-Oquendo, BS *; Matilde Sanchez-Pena, MS *; Clara E. Isaza, PhD * [[dagger]]; Mauricio Cabrera-Rios, PhD *
* BioIE Lab, Department of Industrial Engineering, University of Puerto Rico Mayaguez Campus, Mayaguez, Puerto Rico; [[dagger]] Immunology Department, School of Biology, Universidad Autonoma de Nuevo Leon, Mexico
The authors have no conflicts of interest to disclose.
Address correspondence to: Mauricio Cabrera-Rios, PhD, Department of Industrial Engineering, University of Puerto Rico Mayaguez Campus, Call Box 9000, Mayaguez, PR 00681-9000. Email: mauricio.cabrera1@upr.edu
Table 1. List of genes identified using multiple p-values in the
Multiple Criteria Optimization (MCO) problem for the biomarker search.
The number describes the frontier where each gene was found according
to the different number of p-values executions used.
Accession Gene Symbol Gene Name
Number
R87126 yq31b10.s1 Soares fetal liver spleen 1NFLS
H08393 WDR77 WD repeat domain 77
R36977 GTF3A General Transcription factor IIIA
M22382 HSPD1 Heat Shock 60kDa protein 1 (chaperonin)
M26383 IL8 Interleukin 8
X63629 CDH3 Cadherin 3, type 1, P-cadherin (placental)
H40095 yn85b03.s1 Soares adult brain N2b5HB55Y
X12671 HNRNPA1 Heterogeneous nuclear ribonucleoprotein A1
J05032 DARS Aspartyl-tRNA synthetase
U09564 SRPK1 SRSF protein kinase 1
Z50753 GUCA2B Guanylate cyclase activator 2B (uroguanylin)
J02854 MYL9 Myosin, light chain 9, regulatory
T47377 S100P Calcium binding protein P
T86473 NME1 Non-metastatic cells 1, protein (NM23A)
H43887 CFD Complement factor D (adipsin)
M36634 VIP Vasoactive intestinal peptide
R08183 HSPE1 Heat Shock 10kDa protein 1 (chaperonin 10)
T71025 MT1G Metallothionein 1G
U30825 SRSF9 Serine/arginine-rich splicing factor 9
X14958 HMGA1 High mobility group AT-hook 1
M26697 NPM1 Nucleophosmin (nucleolar phosphoprotein B23,
numatrin)
R84411 SNRPB Small nuclear ribonucleoprotein polypeptides
B and B1
X12466 SNRPE Small nuclear ribonucleoprotein
polypeptide E
M63391 DES Desmin
p-values used
Accession 2-pv 4-pv 6-pv 8-pv
Number
R87126 1 1 1 1
H08393 2 1 1 1
R36977 3 2 2 1
M22382 4 3 2 1
M26383 5 4 3 2
X63629 5 4 3 2
H40095 5 4 3 2
X12671 5 4 3 2
J05032 6 5 4 2
U09564 6 5 4 2
Z50753 6 4 3 2
J02854 7 6 3 2
T47377 7 5 4 3
T86473 7 6 5 3
H43887 7 5 4 3
M36634 8 7 4 3
R08183 8 6 5 3
T71025 8 7 5 3
U30825 9 7 5 3
X14958 9 7 5 3
M26697 9 7 6 3
R84411 10 8 6 4
X12466 10 8 7 4
M63391 10 8 3 2
Table 2. List of MCO-identified genes with examples of different
types of cancer that have been shown to change their expression
Gene Cancer type involvement (not a References
comprehensive list)
WDRF77 Ovarian, prostate 7, 8
HSPD1 Colorectal 9
IL8 Colorectal 10, 11
CDH3 Biliary tract, esophageal 1, 13
HNRNPA1 Involved in the switch to aerobic
glycolysis, a process common to cancer 14
cells.
DARS Leukemia 15
SRPK1 Colorectal 16
GUCA2B Colorectal 17
MYL9 Chicken sarcoma model for metastasis, 18, 19
breast cancer cell motility
S100P Pancreatic, lung adenocarcinomas, breast, 20, 21, 22, 23
colon
NME1 Colon 24
CFD Gastric, tongue, colon 25, 26, 27
VIP Colorectal 28
HSPE1 Proposed to have a role in cancer etiology 29
MTG1 Lung adenocarcinoma, colorectal 30, 31
SRSF9 Regulation of procancerous proteins 32, 33
HMGA1 Prostate, apoptosis inhibition 34, 35
NPM1 Acute myeloid leukemia 36, 37
SNRPB Proposed metastasis suppressor gene for 38
prostate cancer
SNRPE Hepatocellular carcinoma 39
DES Colorectal 40
MCO: Multiple Criteria Optimization
 Printer friendly
 Cite/link
 Email
 Feedback
| |
| Author: | Watts-Oquendo, Erika; Sanchez-Pena, Matilde; Isaza, Clara E.; Cabrera-Rios, Mauricio |
|---|---|
| Publication: | Puerto Rico Health Sciences Journal |
| Date: | Jun 1, 2012 |
| Words: | 3134 |
| Previous Article: | Rectal Adenocarcinoma: proposal for a model based on pretreatment prognostic factors. |
| Next Article: | Levels of felt stigma among a group of people with HIV in Puerto Rico. |
| Topics: | |