Gene-Expression Profiling Predicts Recurrence in Dukes’ C Colorectal Cancer
2005; Elsevier BV; Volume: 129; Issue: 3 Linguagem: Inglês
10.1053/j.gastro.2005.06.066
ISSN1528-0012
AutoresDiego Arango, Päivi Laiho, Antti Kokko, Pia Alhopuro, Heli Sammalkorpi, Reijo Salovaara, Daniel Nicorici, Sampsa Hautaniemi, Hafid Alazzouzi, Jukka‐Pekka Mecklin, Heikki Järvinen, Akseli Hemminki, Jaakko Astola, Simó Schwartz, Lauri A. Aaltonen,
Tópico(s)Cancer, Lipids, and Metabolism
ResumoBackground & Aims: Although approximately 50% of Dukes' C colorectal cancer patients are surgically cured, it is currently not possible to distinguish these patients from those at high risk of recurrence. The recent advent of routine adjuvant chemotherapy for these patients has greatly complicated the identification of new markers predicting the response to surgery, which is now reliant on archived materials. Microarray analysis allows fine tumor classification but cannot be used with paraffin-embedded archival samples. Methods: We used microarray analysis of a unique set of fresh-frozen tumor samples from Dukes' C patients who had surgery as the only form of treatment to identify molecular signatures that characterize tumors from patients with good and bad prognosis. Results: Unsupervised hierarchical clustering and a K-nearest neighbors–based classifier identified groups of patients with significantly different survival (P = .019 and P = .0001). Expression profiling outperformed previously reported genetic markers of prognosis such as TP53 and K-RAS mutational status and allelic imbalance in chromosome 18q, which were of limited prognostic power in this study. Functional categories significantly enriched in gene-expression differences included protein transport and folding. The prognostic potential of the RAS homologue RHOA, one of the most differentially expressed genes, was further investigated using immunohistochemistry and a tissue microarray containing 137 independent Dukes' C tumor samples. Reduced RHOA expression was associated with significantly shorter survival (P = .01). Conclusions: This study shows that gene-expression profiling of surgical tumor samples can predict recurrence in Dukes' C patients. Therefore, this approach could be used to guide decisions concerning the clinical management of these patients. Background & Aims: Although approximately 50% of Dukes' C colorectal cancer patients are surgically cured, it is currently not possible to distinguish these patients from those at high risk of recurrence. The recent advent of routine adjuvant chemotherapy for these patients has greatly complicated the identification of new markers predicting the response to surgery, which is now reliant on archived materials. Microarray analysis allows fine tumor classification but cannot be used with paraffin-embedded archival samples. Methods: We used microarray analysis of a unique set of fresh-frozen tumor samples from Dukes' C patients who had surgery as the only form of treatment to identify molecular signatures that characterize tumors from patients with good and bad prognosis. Results: Unsupervised hierarchical clustering and a K-nearest neighbors–based classifier identified groups of patients with significantly different survival (P = .019 and P = .0001). Expression profiling outperformed previously reported genetic markers of prognosis such as TP53 and K-RAS mutational status and allelic imbalance in chromosome 18q, which were of limited prognostic power in this study. Functional categories significantly enriched in gene-expression differences included protein transport and folding. The prognostic potential of the RAS homologue RHOA, one of the most differentially expressed genes, was further investigated using immunohistochemistry and a tissue microarray containing 137 independent Dukes' C tumor samples. Reduced RHOA expression was associated with significantly shorter survival (P = .01). Conclusions: This study shows that gene-expression profiling of surgical tumor samples can predict recurrence in Dukes' C patients. Therefore, this approach could be used to guide decisions concerning the clinical management of these patients. Colorectal cancer is one of the leading causes of cancer-related death in the Western world, and a large proportion of these patients are diagnosed with locally advanced disease with regional lymph node metastasis (Dukes' C stage). Despite the high incidence of this disease, the clinical management of Dukes' C colorectal cancer patients is far from optimal. Surgical resection prevents recurrence in approximately 50% of Dukes' C patients. Because of the relatively high risk of recurrence in these patients, several large multi-institutional clinical trials have been performed in the last 2 decades to investigate the benefits of adjuvant chemotherapy. 5-Fluorouracil (5-FU)–based therapy was found to prevent recurrence in 10%–20% of patients.1Moertel C.G. Fleming T.R. Macdonald J.S. Haller D.G. Laurie J.A. Goodman P.J. Ungerleider J.S. Emerson W.A. Tormey D.C. Glick J.H. et al.Levamisole and fluorouracil for adjuvant therapy of resected colon carcinoma.N Engl J Med. 1990; 322: 352-358Crossref PubMed Scopus (2107) Google Scholar, 2Moertel C.G. Fleming T.R. Macdonald J.S. Haller D.G. Laurie J.A. Tangen C.M. Ungerleider J.S. Emerson W.A. Tormey D.C. Glick J.H. et al.Fluorouracil plus levamisole as effective adjuvant therapy after resection of stage III colon carcinoma a final report.Ann Intern Med. 1995; 122: 321-326Crossref PubMed Scopus (882) Google Scholar However, because it is currently not possible to accurately distinguish surgically cured patients from those at high risk of recurrence, the great majority of Dukes' C patients are administered adjuvant chemotherapy; this results in a large number of surgically cured patients undergoing chemotherapeutic treatment without benefit. In addition, routine administration of adjuvant chemotherapy to these patients greatly complicates the identification of surgically cured patients, because it is not possible to distinguish patients cured by surgery from those with a good response to the adjuvant treatment. Therefore, experiments addressing this question heavily rely on archived materials collected before adjuvant treatment was incorporated into routine clinical practice. The potential of several markers, such as TP53 or K-RAS mutational status and loss of heterozygosity in chromosome 18q, to identify patients who vary in their probability of recurrence after surgery has been previously investigated.3Bell S.M. Scott N. Cross D. Sagar P. Lewis F.A. Blair G.E. Taylor G.R. Dixon M.F. Quirke P. Prognostic value of p53 overexpression and c-Ki-ras gene mutations in colorectal cancer.Gastroenterology. 1993; 104: 57-64Abstract PubMed Google Scholar, 4Benhattar J. Losi L. Chaubert P. Givel J.C. Costa J. Prognostic significance of K-ras mutations in colorectal carcinoma.Gastroenterology. 1993; 104: 1044-1048PubMed Google Scholar, 5Goh H.S. Yao J. Smith D.R. p53 point mutation and survival in colorectal cancer patients.Cancer Res. 1995; 55: 5217-5221PubMed Google Scholar, 6Ogunbiyi O.A. Goodfellow P.J. Herfarth K. Gagliardi G. Swanson P.E. Birnbaum E.H. Read T.E. Fleshman J.W. Kodner I.J. Moley J.F. Confirmation that chromosome 18q allelic loss in colon cancer is a prognostic indicator.J Clin Oncol. 1998; 16: 427-433Crossref PubMed Scopus (157) Google Scholar, 7Lanza G. Matteuzzi M. Gafa R. Orvieto E. Maestri I. Santini A. del Senno L. Chromosome 18q allelic loss and prognosis in stage II and III colon cancer.Int J Cancer. 1998; 79: 390-395Crossref PubMed Scopus (134) Google Scholar, 8Watanabe T. Wu T.T. Catalano P.J. Ueki T. Satriano R. Haller D.G. Benson III, A.B. Hamilton S.R. Molecular predictors of survival after adjuvant chemotherapy for colon cancer.N Engl J Med. 2001; 344: 1196-1206Crossref PubMed Scopus (795) Google Scholar, 9Laurent-Puig P. Olschwang S. Delattre O. Remvikos Y. Asselain B. Melot T. Validire P. Muleris M. Girodet J. Salmon R.J. et al.Survival and acquired genetic alterations in colorectal cancer.Gastroenterology. 1992; 102: 1136-1141PubMed Google Scholar, 10Jen J. Kim H. Piantadosi S. Liu Z.F. Levitt R.C. Sistonen P. Kinzler K.W. Vogelstein B. Hamilton S.R. Allelic loss of chromosome 18q and prognosis in colorectal cancer.N Engl J Med. 1994; 331: 213-221Crossref PubMed Scopus (686) Google Scholar, 11Diep C.B. Thorstensen L. Meling G.I. Skovlund E. Rognum T.O. Lothe R.A. Genetic tumor markers with prognostic impact in Dukes' stages B and C colorectal cancer patients.J Clin Oncol. 2003; 21: 820-829Crossref PubMed Scopus (130) Google Scholar, 12Cohn K.H. Ornstein D.L. Wang F. LaPaix F.D. Phipps K. Edelsberg C. Zuna R. Mott L.A. Dunn J.L. The significance of allelic deletions and aneuploidy in colorectal carcinoma. Results of a 5-year follow-up study.Cancer. 1997; 79: 233-244Crossref PubMed Scopus (68) Google Scholar However, the accuracy of these markers is limited, and conflicting reports can be found in the literature.3Bell S.M. Scott N. Cross D. Sagar P. Lewis F.A. Blair G.E. Taylor G.R. Dixon M.F. Quirke P. Prognostic value of p53 overexpression and c-Ki-ras gene mutations in colorectal cancer.Gastroenterology. 1993; 104: 57-64Abstract PubMed Google Scholar, 4Benhattar J. Losi L. Chaubert P. Givel J.C. Costa J. Prognostic significance of K-ras mutations in colorectal carcinoma.Gastroenterology. 1993; 104: 1044-1048PubMed Google Scholar, 5Goh H.S. Yao J. Smith D.R. p53 point mutation and survival in colorectal cancer patients.Cancer Res. 1995; 55: 5217-5221PubMed Google Scholar, 6Ogunbiyi O.A. Goodfellow P.J. Herfarth K. Gagliardi G. Swanson P.E. Birnbaum E.H. Read T.E. Fleshman J.W. Kodner I.J. Moley J.F. Confirmation that chromosome 18q allelic loss in colon cancer is a prognostic indicator.J Clin Oncol. 1998; 16: 427-433Crossref PubMed Scopus (157) Google Scholar, 7Lanza G. Matteuzzi M. Gafa R. Orvieto E. Maestri I. Santini A. del Senno L. Chromosome 18q allelic loss and prognosis in stage II and III colon cancer.Int J Cancer. 1998; 79: 390-395Crossref PubMed Scopus (134) Google Scholar, 8Watanabe T. Wu T.T. Catalano P.J. Ueki T. Satriano R. Haller D.G. Benson III, A.B. Hamilton S.R. Molecular predictors of survival after adjuvant chemotherapy for colon cancer.N Engl J Med. 2001; 344: 1196-1206Crossref PubMed Scopus (795) Google Scholar, 9Laurent-Puig P. Olschwang S. Delattre O. Remvikos Y. Asselain B. Melot T. Validire P. Muleris M. Girodet J. Salmon R.J. et al.Survival and acquired genetic alterations in colorectal cancer.Gastroenterology. 1992; 102: 1136-1141PubMed Google Scholar, 10Jen J. Kim H. Piantadosi S. Liu Z.F. Levitt R.C. Sistonen P. Kinzler K.W. Vogelstein B. Hamilton S.R. Allelic loss of chromosome 18q and prognosis in colorectal cancer.N Engl J Med. 1994; 331: 213-221Crossref PubMed Scopus (686) Google Scholar, 11Diep C.B. Thorstensen L. Meling G.I. Skovlund E. Rognum T.O. Lothe R.A. Genetic tumor markers with prognostic impact in Dukes' stages B and C colorectal cancer patients.J Clin Oncol. 2003; 21: 820-829Crossref PubMed Scopus (130) Google Scholar, 12Cohn K.H. Ornstein D.L. Wang F. LaPaix F.D. Phipps K. Edelsberg C. Zuna R. Mott L.A. Dunn J.L. The significance of allelic deletions and aneuploidy in colorectal carcinoma. Results of a 5-year follow-up study.Cancer. 1997; 79: 233-244Crossref PubMed Scopus (68) Google Scholar As a result, these markers have not become routinely used in the clinical management of these patients. Identification of additional means of distinguishing patients with good and poor prognosis after surgery would allow targeting aggressive chemotherapeutic treatment to patients who can potentially benefit from it and would spare surgically cured patients the side effects, toxicity, and cost associated with such treatment. Microarray analysis allows assessment of the level of expression of thousands of genes simultaneously,13Schena M. Shalon D. Davis R.W. Brown P.O. Quantitative monitoring of gene expression patterns with a complementary DNA microarray.Science. 1995; 270: 467-470Crossref PubMed Scopus (7726) Google Scholar, 14Augenlicht L.H. Kobrin D. Cloning and screening of sequences expressed in a mouse colon tumor.Cancer Res. 1982; 42: 1088-1093PubMed Google Scholar, 15Wodicka L. Dong H. Mittmann M. Ho M.H. Lockhart D.J. Genome-wide expression monitoring in Saccharomyces cerevisiae.Nat Biotechnol. 1997; 15: 1359-1367Crossref PubMed Scopus (869) Google Scholar, 16Mahadevappa M. Warrington J.A. A high-density probe array sample preparation method using 10- to 100-fold fewer cells.Nat Biotechnol. 1999; 17: 1134-1136Crossref PubMed Scopus (133) Google Scholar and the potential of expression profiling as a tool to predict the prognosis of cancer patients has been previously realized for different types of cancer, including lymph node–negative Dukes' B colorectal cancer.17Golub T.R. Slonim D.K. Tamayo P. Huard C. Gaasenbeek M. Mesirov J.P. Coller H. Loh M.L. Downing J.R. Caligiuri M.A. Bloomfield C.D. Lander E.S. Molecular classification of cancer class discovery and class prediction by gene expression monitoring.Science. 1999; 286: 531-537Crossref PubMed Scopus (9432) Google Scholar, 18Alizadeh A.A. Eisen M.B. Davis R.E. Ma C. Lossos I.S. Rosenwald A. Boldrick J.C. Sabet H. Tran T. Yu X. Powell J.I. Yang L. Marti G.E. Moore T. Hudson Jr, J. Lu L. Lewis D.B. Tibshirani R. Sherlock G. Chan W.C. Greiner T.C. Weisenburger D.D. Armitage J.O. Warnke R. Staudt L.M. et al.Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling.Nature. 2000; 403: 503-511Crossref PubMed Scopus (8315) Google Scholar, 19Kihara C. Tsunoda T. Tanaka T. Yamana H. Furukawa Y. Ono K. Kitahara O. Zembutsu H. Yanagawa R. Hirata K. Takagi T. Nakamura Y. Prediction of sensitivity of esophageal tumors to adjuvant chemotherapy by cDNA microarray analysis of gene-expression profiles.Cancer Res. 2001; 61: 6474-6479PubMed Google Scholar, 20Wang Y. Jatkoe T. Zhang Y. Mutch M.G. Talantov D. Jiang J. McLeod H.L. Atkins D. Gene expression profiles and molecular markers to predict recurrence of Dukes' B colon cancer.J Clin Oncol. 2004; 22: 1564-1571Crossref PubMed Scopus (430) Google Scholar In addition, microarray analysis can provide significant insight into biological differences between tumors from patients with good and bad prognosis and can be used as a screening tool to find individual markers that could identify groups of patients who differ in their prognosis. However, it is currently not possible to use microarray analysis with RNA extracted from archived paraffin-embedded tumor samples. In this study, we used high-density oligonucleotide microarray analysis and a unique set of fresh-frozen tumor samples from Dukes' C colorectal cancer patients who had surgery as the only form of treatment, to identify patterns of gene expression that characterize tumors from patients with good and bad prognosis. Moreover, we compared the accuracy of these profiles of expression with that of other genetic markers of prognosis and found that expression profiling outperformed other markers tested. In addition, the level of protein expression of the RAS homologue RHOA, one of the genes with the most significantly different expression at the messenger RNA level in tumors from patients with good and bad prognosis, was identified as a useful immunohistochemical marker capable of predicting the prognosis of an independent set of 137 tumor samples from Dukes' C colorectal cancer patients. Among the 1042-fresh frozen colorectal cancer samples available in our tumor bank (collected from 1994 to 1998), we selected samples from Dukes' C patients (281 patients) who had surgery as the only form of treatment (91 patients). Patients who died of causes unrelated to colorectal cancer were excluded. These samples were collected before adjuvant chemotherapy became standard practice for these patients and constitute, therefore, a unique set of samples to investigate response to surgery. Moreover, complete follow-up was available for all these patients for at least 6 years (mean follow-up, 8.9 years), thus allowing analysis of long-term survival. Informed consent for genetic analysis of the tumor sample was obtained from each patient according to the Human Investigations and Ethical Committee–approved research proposal. Frozen sections from OCT-embedded samples (Tissue-Tek, Zoeterwoude, The Netherlands) of these 91 tumors were cut, and after toluidine blue staining and histological verification, 47 of them were found to contain at least 70% tumor cells. Frozen tissue samples were macrodissected from the selected areas of the frozen OCT blocks and homogenized by using a tissue homogenizer (Ultra-Turrax T8; IKA Labortechnik, Staufen, Germany) in 1 mL of TRIzol reagent (Invitrogen, Carlsbad, CA). Total RNA was extracted with TRIzol as recommended by the manufacturer, followed by an RNeasy cleanup step (Qiagen, Germantown, MD). The RNA quality of these 47 samples was then assessed with a 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA), and 25 of them showed optimal quality for oligonucleotide microarray analysis, as shown the integrity of the observed 18S and 28S ribosomal RNA bands. Of these, 10 had disease recurrence within 5 years of surgery (poor prognosis), and 15 had at least 5 years of disease-free survival after surgical treatment (good prognosis) (Table 1).Table 1Clinical Features of Patients in the Study and Results of the K-Nearest Neighbors ClassifierPatient no.MSI statusGradeaHistological grade.Age (y)Dukes'DFS (y)True prognosisbObserved prognosis.Predicted prognosiscPrognosis predicted by the 5-nearest neighbors (NN) classifier using 17 genes. Bad, recurrence within 5 years of surgery; good, no recurrence within 5 years of surgery.367MSS388C0.5BadBad176MSS263C0.5BadGood146MSS261C0.6BadBad616MSS255C0.7BadBad981MSS256C0.9BadBad707MSS286C1.1BadGood416MSS263C1.2BadBad931MSS263C2.2BadBad986MSS283C2.5BadBad463MSS258C2.6BadBad2000MSS268C5.5GoodGood988MSS255C6.0GoodGood917MSS272C6.2GoodGood939MSS164C6.4GoodGood711MSS272C7.0GoodGood634MSS186C7.0GoodGood613MSS274C7.1GoodGood482MSS184C7.2GoodBad476MSS359C7.5GoodGood466MSS170C7.7GoodGood381MSS283C8.0GoodGood330MSS281C8.1GoodGood280MSS166C8.2GoodGood259MSS270C8.2GoodGood117MSS284C9.1GoodGoodMSI, microsatellite instability; MSS, microsatellite stable; DFS, disease-free survival.a Histological grade.b Observed prognosis.c Prognosis predicted by the 5-nearest neighbors (NN) classifier using 17 genes. Bad, recurrence within 5 years of surgery; good, no recurrence within 5 years of surgery. Open table in a new tab MSI, microsatellite instability; MSS, microsatellite stable; DFS, disease-free survival. Eight micrograms of total RNA was reverse-transcribed, and the resulting complementary DNA template was used for the in vitro transcription reaction. This resulted in biotin-labeled chromosomal RNA as described in the Supplementary Materials (http://research.med.helsinki.fi/cancerbio/gastro). Microarrays were scanned with a GeneChip Scanner GA2500, and MAS 5.0 software (both from Affymetrix, Santa Clara, CA) was used to obtain quantitative expression information for all 22,283 probe sets in the HG-U133A microarrays. Detection call cutoff values used for present and marginal genes were α1 = 0.05 and α2 = 0.065, respectively. The percentage of present and marginal calls ranged from 37% to 52%, and the 3′/5′ ratio of β-actin and glyceraldehyde phosphate dehydrogenase control genes was <2.8. All chips were then scaled to the average gene expression of all 25 chips, and the expression of each probe set was divided by the mean expression of that probe across all the samples so that the resulting mean expression for every probe was 1. Only genes with detectable expression levels in at least 50% of the samples (10,035 genes) were included in this analysis. Gene expression for samples with absent calls was substituted with the gene average expression across all samples to allow class permutation analysis (see below) as previously described.21Tusher V.G. Tibshirani R. Chu G. Significance analysis of microarrays applied to the ionizing radiation response.Proc Natl Acad Sci U S A. 2001; 98: 5116-5121Crossref PubMed Scopus (9905) Google Scholar Genes differentially expressed in patients with good and bad prognosis were identified by using a Student t test. The class labels were then randomly permuted, and a new t test was run with the new groupings for every gene. This process was repeated 1 million times, and the percentage of times was scored in which a P value was lower than the true P value initially calculated. We selected as differentially expressed those genes that were found to have a P value lower than the initial P value in <1% of the permutations. Gene ontology terms (http://www.geneontology.org) were used to annotate all genes with present calls in at least 50% of samples (10,035 genes). A total of 1384 partially overlapping functionally related categories of genes were identified. Categories with 500 genes were removed from this analysis, because they were considered to be too specific or too general to be useful. Using the list of 218 unique genes differentially expressed, we identified categories that were significantly enriched in the number of genes with different expression in tumors from patients with good and bad prognosis (P < .01). These analyses were implemented using GoMiner software,22Zeeberg B.R. Feng W. Wang G. Wang M.D. Fojo A.T. Sunshine M. Narasimhan S. Kane D.W. Reinhold W.C. Lababidi S. Bussey K.J. Riss J. Barrett J.C. Weinstein J.N. GoMiner a resource for biological interpretation of genomic and proteomic data.Genome Biol. 2003; 4: R28Crossref PubMed Google Scholar and a Fisher exact test was used to identify significantly enriched categories. The 5102 genes that were expressed at detectable levels (present and marginal calls) in all 25 samples were selected, and then all data were log2-transformed and the intensity of both the arrays and the genes were centered on the median. These data were used to hierarchically cluster all 25 samples by using Cluster software and were visualized by using TreeView software.23Eisen M.B. Spellman P.T. Brown P.O. Botstein D. Cluster analysis and display of genome-wide expression patterns.Proc Natl Acad Sci U S A. 1998; 95: 14863-14868Crossref PubMed Scopus (13444) Google Scholar The 5102 genes with present or marginal calls in all 25 samples were used for these analyses. Each gene profile was discretized independently to binary values across the samples by applying the Lloyd algorithm.24Lloyd S.P. Least square quantization in PCM.Trans Information Theory. 1982; IT-28: 129-137Crossref Scopus (10197) Google Scholar The Lloyd algorithm minimizes the average discretization error, which represents the distance between the data and the discrete representation. Also, the Lloyd algorithm can be understood as a particular case for 1-dimensional data of the K-means clustering method. These data were then used to generate a K-nearest neighbors classifier capable of using the expression profile of these 25 samples to distinguish patients with good and bad prognosis. These analyses were implemented as previously described by using GeneCluster II software.17Golub T.R. Slonim D.K. Tamayo P. Huard C. Gaasenbeek M. Mesirov J.P. Coller H. Loh M.L. Downing J.R. Caligiuri M.A. Bloomfield C.D. Lander E.S. Molecular classification of cancer class discovery and class prediction by gene expression monitoring.Science. 1999; 286: 531-537Crossref PubMed Scopus (9432) Google Scholar The number of genes used in this classifier varied from 1 to 100, and the number of neighbors varied from 1 to 10. The predictions of the results of every classifier tested were then validated using a leave-one-out cross-validation approach as previously described.17Golub T.R. Slonim D.K. Tamayo P. Huard C. Gaasenbeek M. Mesirov J.P. Coller H. Loh M.L. Downing J.R. Caligiuri M.A. Bloomfield C.D. Lander E.S. Molecular classification of cancer class discovery and class prediction by gene expression monitoring.Science. 1999; 286: 531-537Crossref PubMed Scopus (9432) Google Scholar, 25Arango D. Wilson A.J. Shi Q. Corner G.A. Arañes M.J. Nicholas C. Lesser M. Mariadason J.M. Augenlicht L.H. Molecular mechanisms of action and prediction of response to oxaliplatin in colorectal cancer cells.Br J Cancer. 2004; 91: 1931-1946Crossref PubMed Scopus (215) Google Scholar, 26Mariadason J.M. Arango D. Shi Q. Wilson A.J. Corner G.A. Nicholas C. Aranes M.J. Lesser M. Schwartz E.L. Augenlicht L.H. Gene expression profiling-based prediction of response of colon carcinoma cells to 5-fluorouracil and camptothecin.Cancer Res. 2003; 63: 8791-8812PubMed Google Scholar Briefly, 1 sample is initially removed from the analysis, and the remaining 24 samples are used to build a classifier. The generated classifier is then applied to the sample initially left out, and the class call is recorded. This cross-validation loop is iteratively repeated 25 times each time leaving out a different sample, and the final result is 25 true class/predicted class pairs. The accuracy of the predictor can thus be assessed by the number of correct calls made by the classifier.17Golub T.R. Slonim D.K. Tamayo P. Huard C. Gaasenbeek M. Mesirov J.P. Coller H. Loh M.L. Downing J.R. Caligiuri M.A. Bloomfield C.D. Lander E.S. Molecular classification of cancer class discovery and class prediction by gene expression monitoring.Science. 1999; 286: 531-537Crossref PubMed Scopus (9432) Google Scholar, 25Arango D. Wilson A.J. Shi Q. Corner G.A. Arañes M.J. Nicholas C. Lesser M. Mariadason J.M. Augenlicht L.H. Molecular mechanisms of action and prediction of response to oxaliplatin in colorectal cancer cells.Br J Cancer. 2004; 91: 1931-1946Crossref PubMed Scopus (215) Google Scholar, 26Mariadason J.M. Arango D. Shi Q. Wilson A.J. Corner G.A. Nicholas C. Aranes M.J. Lesser M. Schwartz E.L. Augenlicht L.H. Gene expression profiling-based prediction of response of colon carcinoma cells to 5-fluorouracil and camptothecin.Cancer Res. 2003; 63: 8791-8812PubMed Google Scholar As an additional control, the sample class labels were randomly permuted 200 times, and the average prediction accuracy of the classifier for these random groupings was computed. Using a leave-one-out approach minimizes the bias that could occur when trying to validate a classifier with the sample set used for training of the classifier. Aliquots of RNA (100 ng) were reverse-transcribed using Superscript II according to the manufacturer's recommendations (Invitrogen). Five microliters of the undiluted reverse transcriptase reaction were used to polymerase chain reaction (PCR)-amplify RHOA, CLTC, MTMR1, ARCN1, and IDH3G by using Assays-on-Demand TaqMan primers and probes and TaqMan Universal PCR Master Mix in a GeneAmp 5700 Sequence Detection System (all from Applied Biosystems, Branchburg, NJ). Relative gene-expression levels were quantified with the ΔΔCT method with β-actin as a housekeeping standard control, as previously described.26Mariadason J.M. Arango D. Shi Q. Wilson A.J. Corner G.A. Nicholas C. Aranes M.J. Lesser M. Schwartz E.L. Augenlicht L.H. Gene expression profiling-based prediction of response of colon carcinoma cells to 5-fluorouracil and camptothecin.Cancer Res. 2003; 63: 8791-8812PubMed Google Scholar, 27Arango D. Mariadason J.M. Willson A.J. Yang W. Corner G.A. Arañes M.J. Nicholas C. Augenlicht L.H. c-Myc overexpression sensitizes colon cancer cells to camptothecin-induced apoptosis.Br J Cancer. 2003; 89: 1757-1765Crossref PubMed Scopus (70) Google Scholar, 28Alhopuro P, Alazzouzi H, Sammalkorpi H, Salovaara R, Hemminki A, Järvinen H, Mecklin JP, Aaltonen LA, Arango D. SMAD4 levels and response to 5-fluorouracil in colorectal cancer. Clin Cancer Res (in press).Google Scholar An extended set of 46 fresh-frozen colorectal cancer samples were screened for K-RAS and TP53 mutations and were assessed for allelic imbalance in chromosome 18q. The mutation hotspots of K-RAS (codons 12, 13, and 61) were PCR-amplified as described previously.29Servomaa K. Kiuru A. Kosma V.M. Hirvikoski P. Rytomaa T. p53 and K-ras gene mutations in carcinoma of the rectum among Finnish women.Mol Pathol. 2000; 53: 24-30Crossref PubMed Scopus (41) Google Scholar, 30Laiho P. Launonen V. Lahermo P. Esteller M. Guo M. Herman J.G. Mecklin J.P. Jarvinen H. Sistonen P. Kim K.M. Shibata D. Houlston R.S. Aaltonen L.A. Low-level microsatellite instability in most colorectal carcinomas.Cancer Res. 2002; 62: 1166-1170PubMed Google Scholar Exons 2–11 of TP53 were PCR-amplified as previously described.25Arango D. Wilson A.J. Shi Q. Corner G.A. Arañes M.J. Nicholas C. Lesser M. Mariadason J.M. Augenlicht L.H. Molecular mechanisms of action and prediction of response to oxaliplatin in colorectal cancer cells.Br J Cancer. 2004; 91: 1931-1946Crossref PubMed Scopus (215) Google Scholar, 31Hsu I.C. Metcalf R.A. Sun T. Welsh J.A. Wang N.J. Harris C.C. Mutational hotspot in the p53 gene in human hepatocellular carcinomas.Nature. 1991; 350: 427-428Crossref PubMed Scopus (1476) Google Scholar The complete coding sequences of RHOA (exons 2–5) were sequenced in 5 of the tumor samples that showed the highest RHOA protein levels and in 5 of the tumors with the lowest expression, as described at http://research.med.helsinki.fi/cancerbio/gastro. Two polymorphic microsatellite markers in 18q21 (D18S1110 and D18S1156) were used to assess the allelic imbalance in this region, as previously reported.28Alhopuro P, Alazzouzi H, Sammalkorpi H, Salovaara R, Hemminki A, Järvinen H, Mecklin JP, Aaltonen LA, Arango D. SMAD4 levels and response to 5-fluorouracil in colorectal cancer. Clin Cancer Res (in press).Google Scholar, 32Alazzouzi H. Alhopuro P. Salovaara R. Sammalkorpi H. Järvinen H. Mecklin J.P. Schwartz S.J. Aaltonen L.A. Arango D. SMAD4 as a prognostic marker in colorectal cancer.Clin Cancer Res. 2005; 11: 2606-2611Crossref PubMed Scopus (161) Google Scholar The size of the amplified PCR products was assessed by using an Applied Biosystems ABI3730 Automatic DNA sequencer. Allelic imbalance was scored if there was a difference >40% in the abundance of an allele between normal and tumor samples.32Alazzouzi H. Alhopuro P.
Referência(s)