Clinical Cancer Research CTRC-AACR San Antonio Breast Cancer Symposium Infection and Cancer: Biology, Therapeutics, and Prevention
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Cancer Research Clinical Cancer Research
Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics
Molecular Cancer Research Cancer Prevention Research
Cancer Prevention Journals Portal Cancer Reviews Online
Annual Meeting Education Book Meeting Abstracts Online
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wei, S. H.
Right arrow Articles by Nephew, K. P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wei, S. H.
Right arrow Articles by Nephew, K. P.
Clinical Cancer Research Vol. 12, 2788-2794, May 1, 2006
© 2006 American Association for Cancer Research

Imaging, Diagnosis, Prognosis

Prognostic DNA Methylation Biomarkers in Ovarian Cancer

Susan H. Wei1, Curtis Balch3,8, Henry H. Paik3, Yoo-Sung Kim3,5, Rae Lynn Baldwin6, Sandya Liyanarachchi1, Lang Li7, Zailong Wang2, Joseph C. Wan1, Ramana V. Davuluri1, Beth Y. Karlan6, Gillian Gifford9, Robert Brown9, Sun Kim4, Tim H-M. Huang1 and Kenneth P. Nephew3,8

Authors' Affiliations: 1 Human Cancer Genetics Program, Department of Molecular Virology, Immunology, and Medical Genetics, Comprehensive Cancer Center and 2 Mathematical Bioscience Institute, The Ohio State University, Columbus, Ohio; 3 Medical Sciences, Department of Cellular and Integrative Physiology, Indiana University School of Medicine; 4 School of Informatics, Center for Bioinformatics and Genomics, Indiana University, Bloomington, Indiana; and 5 School of Information and Communication Engineering, Inha University, Incheon, South Korea; 6 Division of Gynecologic Oncology, Cedars-Sinai Medical Center, Department of Obstetrics and Gynecology, University of California at Los Angeles School of Medicine, Los Angeles, California; 7 Division of Biostatistics, Department of Medicine, Indiana University School of Medicine; 8 Indiana University Cancer Center, Indianapolis, Indiana; and 9 Cancer Research UK, Beatson Laboratories, University of Glasgow, Glasgow, Scotland, United Kingdom

Requests for reprints: Kenneth P. Nephew, Indiana University School of Medicine, Jordan Hall 303, 1001 East 3rd Street, Bloomington, IN 47405. Phone: 812-855-9445; Fax: 812-855-4436; E-mail: knephew{at}indiana.edu.


   Abstract
Top
 Abstract
Materials and Methods
 Results
 Discussion
 References
 
Purpose: Aberrant DNA methylation, now recognized as a contributing factor to neoplasia, often shows definitive gene/sequence preferences unique to specific cancer types. Correspondingly, distinct combinations of methylated loci can function as biomarkers for numerous clinical correlates of ovarian and other cancers.

Experimental Design: We used a microarray approach to identify methylated loci prognostic for reduced progression-free survival (PFS) in advanced ovarian cancer patients. Two data set classification algorithms, Significance Analysis of Microarray and Prediction Analysis of Microarray, successfully identified 220 candidate PFS-discriminatory methylated loci. Of those, 112 were found capable of predicting PFS with 95% accuracy, by Prediction Analysis of Microarray, using an independent set of 40 advanced ovarian tumors (from 20 short-PFS and 20 long-PFS patients, respectively). Additionally, we showed the use of these predictive loci using two bioinformatics machine-learning algorithms, Support Vector Machine and Multilayer Perceptron.

Conclusion: In this report, we show that highly prognostic DNA methylation biomarkers can be successfully identified and characterized, using previously unused, rigorous classifying algorithms. Such ovarian cancer biomarkers represent a promising approach for the assessment and management of this devastating disease.

Aberrant DNA methylation is now recognized as a distinguishing characteristic of neoplasia. Specifically, regions normally protected from methylation, known as "CpG islands," frequently gain hypermethylation during tumorigenesis, often exhibiting distinctive tumor-type and stage-related patterns (1). Consequently, identification of specific methylated CpG islands, as biomarkers for distinct cancer types, holds potential for diagnosis and monitoring of disease progression. In rare cases, methylation of single loci is prognostic for specific cancer types. For example, methylation of GSTP1 has been shown as a highly specific marker for prostate cancer (2, 3). In most cases, however, methylation of single genes is not specific enough to be used for screening of large populations, and several groups are now focusing on establishing panels of methylated markers possessing greater disease sensitivity and specificity. To that end, House et al. (4) showed that panels of methylated genes were more predictive than single genes for the prognosis of pancreatic, hepatocellular (5), and gastrointestinal (6) neoplasms. In a pilot feasibility study of ovarian cancer, we recently identified candidate loci that significantly correlated with progression-free survival (PFS, the time period between chemotherapy initiation and tumor relapse) in patients with advanced-stage disease (7). That study and others show that distinct methylated loci may serve as tumor biomarkers.

As many reports now suggest that DNA methylation is not stochastic, as previously proposed (1), but may actually exhibit tumor/stage–specific patterns (8), we endeavored to determine a "methylation signature" for predicting ovarian cancer PFS. In the present study, we subjected methylation microarray results to two recently described biostatistical analyses, Prediction Analysis of Microarray (PAM) and Significance Analysis of Microarray (SAM; refs. 9, 10), to identify a set of potential PFS-discriminatory loci. Those loci were then further examined using a cohort of 40 ovarian tumors, successfully predicting PFS with 95% accuracy. Finally, using two machine-learning algorithms [Support Vector Machine (SVM) and Multilayer Perceptron (MLP)], we further showed the use of those loci. Such clinically correlated methylated loci could serve as valuable biomarkers having broad prognostic and therapeutic applications.


   Materials and Methods
Top
 Abstract
 Materials and Methods
Results
 Discussion
 References
 
Tissues and DNA specimens. Forty advanced-stage epithelial ovarian tumors and seven normal adjacent ovarian tissues were obtained from the Cooperative Human Tissue Network (Columbus, OH), Western Infirmary and Stobhill General Hospital (Glasgow, United Kingdom), and Cedar-Sinai Medical Center (Los Angeles, CA), following pathologist examination and classification. PFS was defined as the time to progression from start of treatment. All patients were treated with carboplatin- or cisplatin-based chemotherapy and most included a taxoid. Clinicopathologic information for the samples is listed in Table 1 . All studies were approved by local institutional review boards.


View this table:
[in this window]
[in a new window]
  		Table 1. Clinical information for the epithelial ovarian specimens

 
Differential methylation hybridization. Differential methylation hybridization was done as we have described previously (7, 11). Briefly, genomic DNA was digested using MseI, before ligation to end-linkers and subsequent restriction with the methylation-sensitive endonucleases BstUI and HpaII, followed by PCR from end-linker primers. PCR-amplified products thus represented pools of methylated (restriction-resistant) species, as unmethylated species were digested. The purified amplicons were then aminoallyl-UTP labeled, coupled with either Cy5 (cancer amplicons) or Cy3 (normal reference amplicons), hybridized to the microarray, washed, and scanned (7, 11). Subpanels of specific loci, derived from extensive analyses of our previous (7,776 loci) microarray results (7), were printed on aminosilane-coated slides, using an Affymetrix/GMS 417 arrayer. For the first subpanel, 220 loci were printed for (single dye) hybridization, using a cohort of nine tumors. For the second subpanel, 112 (significantly identified subset of the 220) sequences were printed and hybridized to DNA from the 40 tumors and 7 normal tissues described above, using lymphocyte DNA as the cohybridization (normal) reference. Thus, for the 112-feature subpanel, both Cy5-labeled (tumor) and Cy3-labeled (normal) DNA were used, with Cy5/Cy3 fluorescence ratios subsequently determined to assess DNA hypermethylation, of specific loci, in tumors.

Data acquisition and normalization. Hybridization signal images were analyzed for spot intensities using a GenePix 4000A (Axon Instruments, Union City, CA) scanner. For each spot, the median Cy5/Cy3 ratio represented the index of methylation difference between tumor and normal amplicons. To correct for variations in hybridization signals, the Cy5/Cy3 intensities were normalized as we have described previously (7, 11).

SAM and PAM analyses. From our previous differential methylation hybridization study, eight patient tumors with high methylation versus 11 samples with low methylation correlated significantly with short versus long PFS, respectively (7). To identify distinct, differentially methylated loci, SAM analysis (10) was used to determine loci displaying different methylation levels between ovarian cancer patient groups with short and long PFS. SAM can select genes with a prespecified false discovery rate (10) and can rank loci based on their scores rather than on changes in methylation. SAM analysis of the log-transformed microarray data set "scored" each locus by statistically determining a methylation ratio difference between the two tumor groups. The SAM score is an extension of a t test statistic, which can penalize unstable, low hybridization signals produced by microarray technologies. SAM analysis requires statistical distribution assumptions of the methylation data for constructing a score (i.e., the first two moments are finite and the distribution is symmetric). However, these assumptions are not necessary for the Wilcoxon test rank sum test, which tends to be more stringent and robust in selecting candidate loci (12). Therefore, to further screen SAM-identified loci having significant (P < 0.05) methylation differences between group 1 and group 2 tumors, a Wilcoxon nonparametric rank sum test (12) in SPlus was applied, excluding missing values. Thus, we implemented both SAM and Wilcoxon tests.

PAM (9) is a statistical analysis that selects an optimal list of loci and their linear combination in such a way that its prediction accuracy of clinical outcome (long or short PFS in this case) is maximized by 10-fold cross-validation. PAM analysis is also a "shrinkage" method, which can shrink coefficients of some loci to zero, when constructing their linear combination. Thus, it is a highly effective way of eliminating less informative loci to arrive at an optimal prediction accuracy. For PAM evaluation, shrunken centroid analysis (9) was used to average methylation values for each locus to derive a "centroid" for the group and "denoise" the group centroids by shrinkage toward zero, the overall centroid for the specimens examined (9). This approach, normalized by within-class SDs for each locus, effectively assigns higher weights to loci with stable (within the same group) methylation ratios. PAM implemented a threshold variable across iterative shrinkage increases, using 10-fold cross-validation to determine class prediction errors. Prediction accuracy was balanced by the smallest set of loci required for the final model at each shrinkage threshold.

Machine learning schemes. Classification algorithms were from Weka (13), a collection of machine learning tools.10 For machine learning, PFS-discriminatory loci were used to build classifiers using SVM (14) and MLP (15) as described in Results. SVM is a tool that optimizes "generalization," referring to the ability to correctly categorize unseen data (14). The SVM algorithm operates by mapping the given training set ("input vectors") into a high-dimensional feature space and attempting to locate in that space a plane that optimally separates the positive from the negative examples, with a minimum classification error for unseen samples (14). MLP is an algorithm that uses simulated neurons ("perceptrons") connected by networks that constitute several "layers." A typical MLP network consists of a set of source neurons (equal in number to the number of variables of the problem) forming the input layer, one or more hidden layers of computation nodes, and an output layer of nodes (15). The input signal disseminates throughout the network from the input to hidden to output layers.

False discovery rate estimation. False discovery refers to the expected proportion of false-positive findings among all rejected null hypotheses (i.e., those features that are statistically significant). In our analyses, the null hypothesis was that loci would display no difference in methylation between long- and short-PFS tumors. The direct approach proposed by Storey (16) was used to estimate false discovery rate for selected significant features.


   Results
Top
 Abstract
 Materials and Methods
 Results
Discussion
 References
 
Identification of methylated biomarkers for predicting PFS. We used two recently developed data set classification methods, SAM (10) and PAM (9), to identify loci significantly correlated with shortened PFS. Previously, we segregated 19 tumors (Table 1, test set 1) from advanced ovarian cancer patients into high- and low-methylation classes (7), with highly methylated tumors correlating significantly with reduced patient PFS (6-month median) and less methylated tumors associating with longer (>12 months) PFS. Specifically, from a total of 7,776 CpG islands, 182 methylated loci were identified to classify reduced PFS using hierarchical clustering and batch sequential forward selection (7). In the current study, we endeavored to validate those predictive loci and perhaps discover additional putative biomarkers, using SAM and PAM. Although both of these algorithms have previously been applied to gene expression microarrays (9, 10), to our knowledge, this is the first use of these methods for high-throughput methylation analyses. For SAM analysis, a "supervised" technique used to correlate microarray results with clinical variables, 811 CpG island loci (CGI; from the total of 7,776) were found methylated in at least 3 of 19 advanced-stage tumors (Fig. 1 ). SAM analysis showed significant (P < 0.05) differences in methylation, between short-PFS (group 1) and long-PFS (group 2) tumors, for 329 of those loci (Fig. 2A ), with an estimated false discovery rate of 0.88% (data not shown). As SAM analysis makes statistical distribution assumptions (i.e., the first two moments are finite and the distribution is symmetric), we used a two-sided Wilcoxon t test, a method that does not require distribution assumptions for data (12), to analyze the 329 loci. The Wilcoxon analysis identified 270 loci possessing significantly greater (P < 0.05) methylation in group 1 versus group 2 tumors. Of the 270 total loci, 107 overlapped with those identified previously by batch sequential forward selection (of the previous total of 182; ref. 7).


Figure 1
View larger version (19K):
[in this window]
[in a new window]
  		Fig. 1. Schematic diagram of this study. Our previous differential methylation hybridization analysis of advanced ovarian carcinomas, using a 7,776 CGI microarray, yielded 811 CGI loci hypermethylated in at least 3 of 19 tumors (7). SAM, applied to these 811 CGI loci, yielded 329 CGI loci displaying significant (P < 0.05) differences. Further refinement (of the 329 CGI), using a two-sided Wilcoxon t test, yielded 270 CGI differentially methylated (P < 0.05) in group 1 versus group 2 (short versus long PFS; see text for details). In addition, PAM was also applied to the 811 candidates. SAM and PAM analyses were combined with loci previously identified by hierarchical clustering and batch sequential forward selection to arrive at 220 CGI for subsequent printing on a subpanel differential methylation hybridization microarray. That panel was then used to analyze nine advanced ovarian tumors, yielding 112 hypermethylated CGI. A second subpanel microarray, constructed from the 112 CGI, was analyzed by differential methylation hybridization using an independent, additional test set consisting of 40 advanced ovarian tumors (20 short and 20 long PFS, groups 1 and 2), yielding 95% classification accuracy between the two groups (bottom right). To further evaluate the 112 loci, we used those markers to construct two separate classifiers, using the algorithms MLP and SVM; both showed 100% classification accuracy. See text for details. BSFS, batch sequential forward selection.

 

Figure 2
View larger version (11K):
[in this window]
[in a new window]
  		Fig. 2. Analysis of 270 CGI showing differential methylation between short- and long-term survival groups. A, scatter plot of observed relative differences versus expected relative differences; solid line indicates where the observed relative difference is identical to the expected relative difference, whereas dashed lines are drawn at a distance {Delta} = 1.2 from the blue line. Red dots indicate CGI loci with significant hypermethylation. B, shrunken centroids of 270 methylated CGI of two PFS groups. C, 10-fold cross-validation of patient groups (as a function of shrinkage). bullet, group 1; {circ}, group 2. D, cross-validated probabilities of individual tumors. E, summary of cross-validation error. See text for details.

 
To further evaluate the SAM/Wilcoxon–validated PFS markers, PAM analysis of the 270 loci (the SAM/Wilcoxon subset) was done. The PAM algorithm involves minimization of a set of genes that characterize a specific classification (in this case, PFS; ref. 9). Here, the average methylation levels ("centroids") of PFS-related genes were reduced ("shrunk"), by increasing threshold values, to decrease the number of discriminative genes. At each shrunken threshold value, misclassification was then determined by 10-fold cross-validation, a standard verification technique that divides the data into 10 approximately equal partitions. Each partition is then, in turn, used for testing, whereas the others are used for training (i.e., 90% of the data used for training and 10% for testing), with the whole procedure repeated 10 times (17). As shown in Fig. 2B, PAM analysis resulted in highly divergent centroids, of the 270 loci, between the two survival groups. These centroids could be reduced at a threshold value of 2.2, with no corresponding increase in misclassification error (Fig. 2C); however, the accuracy of this subset was only 62.5% (Fig. 2C, Y axis), with three of eight group 1 tumors misclassified (Fig. 2D and E). Thus, the total of 270 was not further reduced by this analysis. PAM was also applied to the original test set (811 loci), resulting in the identification of 141 significant features; those were, however, no more discriminatory than the SAM-identified 270 (data not shown).

To improve prediction accuracy, we selected 220 of the most promising loci by combining sequences from (a) our previous study (using batch sequential forward selection; ref. 7), (b) those identified by SAM/Wilcoxon, and (c) those identified by PAM (Fig. 1). Those criteria were balanced by technical issues, including the use of clones that were readily extractable from libraries and those providing PCR products suitable for printing. Following selection, the 220 candidate loci were printed (in triplicate) on arrays. Differential methylation hybridization, an array-based technique capable of examining >7,000 methylated sequences simultaneously (11), was then done using (single-dye) labeled DNA from nine advanced ovarian tumors; of the 220 loci, 112 showed significant hybridization (a list of 40 known genes within this panel is provided in Table 2 ). Subsequently, these 112 loci were printed on slides and hybridized to genomic DNA from 20 advanced ovarian tumors from patients with short PFS (median 6 months, group 1; Fig. 3A, solid line ), 20 tumors from patients with long PFS (median 22 months, group 2; Fig. 3A, dotted line), and 7 normal tissues (Table 1, test set 2). After normalization of the data (Fig. 3B), PAM was used to ascribe loci distinct to group 1 (thus correlated with short PFS). As shown in Fig. 3C, a clear divergence in methylation levels, at specific CpG islands, was observed between the PFS groups. Moreover, further attempts to reduce the number of discriminatory loci (i.e., increase the "shrinkage" threshold) resulted in greatly increased misclassification (Fig. 3D), indicating that these 112 loci were highly class-selective. Overall, the accuracy probabilities of these 40 tumors were estimated at >80% for both groups 1 and 2 (Fig. 3E). As shown, group assignment predictions, using shrunken centroid analysis, were clearly enhanced for the 112-loci panel, yielding reduced misclassification rates (1 of 20, thus 95% accurate) for both PFS groups (Table 3 ).


View this table:
[in this window]
[in a new window]
  		Table 2. SAM-identified CGI genes hypermethylated in group 2 tumors relative to group 1 tumors

 

Figure 3
View larger version (13K):
[in this window]
[in a new window]
  		Fig. 3. Evaluation of the 112 CGI panel for predicting patient relapse. A, Kaplan-Meier curve of represented patient groups. Solid line, short survival (<12 months). Dashed line, long survival (>14 months). B, normalization of experimentally "blinded" methylation data set. C, shrunken centroids of methylated CGI, with corresponding misclassification error. D, data set shrinkage versus classification error rate of 10-fold cross-validation test. bullet, group 1; {circ}, group 2. E, cross-validated probabilities of individual tumors. See text for details.

 

View this table:
[in this window]
[in a new window]
  		Table 3. Ten-fold cross-validated confusion matrix for three separate classification methods

 
For further study of our biomarker panel, we evaluated the 112 loci using various statistical and bioinformatics analyses. Classification models were constructed from the entire 112-loci PFS-predictive panel, using SVM and MLP (Weka, see Materials and Methods). The input variable to these algorithms was a two-dimensional matrix representing methylation values of 112 individual loci within each of the 40 (20 short- and 20 long-PFS patient) tumors previously examined by PAM. Both of these methods showed 100% classification accuracy, as assessed by 10-fold cross-validation (see Table 3). That accuracy (100%), using these alternative algorithms, was similar to the accuracy achieved by PAM (95%); such a similarity would be entirely expected, as PAM had previously optimized the most useful discriminatory features. To determine the false discovery rate (the number of discriminatory loci possibly discovered by chance), we used the direct approach of Storey (16). In total, methylation levels of all 112 loci were compared (based on t test) between the short and long PFS groups, resulting in a determination that only 25% (28 loci) could be due to chance. Therefore, the majority (84 loci or 75% of the total) of the panel would yet possess differential methylation; however, without further validation, we cannot predict which specific loci are hypermethylated. Consequently, we contend that the high-PFS prediction accuracy (95%) from this panel is not merely due to a random set of loci; rather, the majority of the loci indeed possess differential methylation between the two groups.


   Discussion
Top
 Abstract
 Materials and Methods
 Results
 Discussion
References
 
In the present study, we used two recently developed microarray analysis algorithms, SAM (10) and PAM (9), to screen methylated loci associated with PFS in ovarian cancer. By combining SAM and PAM with the results of our previous pilot study (7), we identified 220 PFS-discriminatory candidate loci. Further screening and experimental analysis resulted in the identification of 112 highly discriminatory loci possessing a PFS prediction accuracy of 95% (Fig. 1; Table 3). We believe that this type of approach is highly amenable to the study of other cancers, as methylation patterns have previously been suggested as tumor and stage specific (8).

To supplement traditional cancer prognosis methods, such as that examining tumor size, lymph node presence, and metastasis (i.e., the classic "tumor-node-metastasis" staging system), biomarkers are now emerging as highly informative for monitoring disease status (18). Such markers can be DNA sequences (e.g., single-nucleotide polymorphisms, deletions), mRNA, or proteins/peptides (18). Epigenetically modified DNA, however, offers potential advantages over other markers, in that it is well associated with tumor progression and several sensitive, high-throughput assays are now widely available (19). DNA is also more stable than RNA, and methylation biomarkers have been established as present in patient serum and other body fluids (20).

To date, several methylated genes have been found highly prognostic for specific cancers, including prostate (21), breast (22), and lung (23). Although some methylated markers, such as RASSF1A and GSTP1, hold potential as prognostic indicators individually (2, 3, 24), it seems that panels ("methylation signatures") will be much more informative (25) and accurate for monitoring cancer progression. Indeed, such predictive panels have previously been shown using expression microarrays. In one study of epithelial ovarian cancer, an "expression signature" of 115 genes was described as prognostic for overall survival (26). Similarly, an 11-gene expression panel, corresponding to a pathway involving the putative stem cell renewal protein Bmi-1, was found associated with shortened PFS and distant metastasis in 10 different malignancies (27). However, methylation signatures may offer advantages over expression signatures, for reasons described above. Additionally, methylated DNA has been shown as present in serum (20, 28); it will thus be important to determine whether any of our markers might also be present in body fluids and could thus be used to monitor disease status noninvasively.

To further screen our 112-loci panel for predictive ability, additional testing will be required in an independent prospective study, a validation trial where tumors have not been selected for high- and low-survival classes (i.e., tumors were analyzed and patients were followed over time). Although such a study would require several years, only randomized trials could conclusively determine the predictive value of this panel for response to chemotherapy.

In addition to serving as clinical biomarkers, discovery of specific methylated genes in ovarian cancer could, with further investigation, provide a greater understanding of cellular pathways and/or signals related to disease progression. DNA methylation frequently characterizes a repressive chromatin structure and methylation of specific genes can lead to down-regulation of distinct biochemical pathways. Previous examples of independently discovered methylated genes include PTPRO (29), encoding a tyrosine phosphatase receptor, and a homeobox gene, ALX3 (30). Within our panel, the gene WWOX (Table 2) was previously shown to be involved in breast, lung, and bladder cancer progression (31). In epithelial ovarian cancer, WWOX mRNA is significantly lower than in normal ovaries (32), suggesting its role as a tumor suppressor. Based on our study, WWOX suppression is likely due to promoter methylation, as previously shown for breast, lung, and bladder neoplasms (31). Similarly, ATOX1 (Table 2), a homologue of a yeast antioxidant protein (33), might play a role in protection from reactive oxygen species. This protein has been found to up-regulate an extracellular superoxide dismutase (SOD-3), by chaperoning copper ions required for SOD-3 activity (33). Thus, methylation and transcriptional repression of ATOX1 could potentially reduce reactive oxygen detoxification, which seems critical for prevention of cellular damage that often occurs in tumor progression (34).

In summary, to our knowledge, this is the first study reporting the use of two highly precise classification algorithms, PAM and SAM, for the identification of prognostic panels of DNA methylation cancer biomarkers. We believe that this approach is amenable to classification of clinically relevant methylation patterns in a wide variety of tumors (and other pathologies), potentially allowing improved management of numerous diseases linked with aberrant DNA methylation.


   Footnotes
 
Grant support: NIH, National Cancer Institute grants CA 085289 (K.P. Nephew) and CA 113001 (T.H-M. Huang).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

S.H. Wei and C. Balch contributed equally to this work.

Current address for S.H. Wei: Women's Cancer Research Institute at the Cedars-Sinai Medical Center, 8700 Beverly Boulevard, Los Angeles, CA 90048.

10 www.cs.waikato.ac.nz/ml/weka/. Back

Received 7/19/05; revised 2/ 3/06; accepted 2/24/06.


   References
Top
 Abstract
 Materials and Methods
 Results
 Discussion
 References
 

  1. Egger G, Liang G, Aparicio A, Jones PA. Epigenetics in human disease and prospects for epigenetic therapy. Nature 2004;429:457–63.[CrossRef][Medline]
  2. Bastian PJ, Ellinger J, Schmidt D, et al. GSTP1 hypermethylation as a molecular marker in the diagnosis of prostatic cancer: is there a correlation with clinical stage, Gleason grade, PSA value or age? Eur J Med Res 2004;9:523–7.[Medline]
  3. Henrique R, Jeronimo C. Molecular detection of prostate cancer: a role for GSTP1 hypermethylation. Eur Urol 2004;46:660–9.[CrossRef][Medline]
  4. House MG, Herman JG, Guo MZ, et al. Aberrant hypermethylation of tumor suppressor genes in pancreatic endocrine neoplasms. Ann Surg 2003;238:423–32.[Medline]
  5. Yang B, Guo M, Herman JG, Clark DP. Aberrant promoter methylation profiles of tumor suppressor genes in hepatocellular carcinoma. Am J Pathol 2003;163:1101–7.[Abstract/Free Full Text]
  6. House MG, Guo M, Efron DT, et al. Tumor suppressor gene hypermethylation as a predictor of gastric stromal tumor behavior. J Gastrointest Surg 2003;7:1004–14.[CrossRef][Medline]
  7. Wei SH, Chen CM, Strathdee G, et al. Methylation microarray analysis of late-stage ovarian carcinomas distinguishes progression-free survival in patients and identifies candidate epigenetic markers. Clin Cancer Res 2002;8:2246–52.[Abstract/Free Full Text]
  8. Costello JF, Fruhwald MC, Smiraglia DJ, et al. Aberrant CpG-island methylation has non-random and tumour-type-specific patterns. Nat Genet 2000;24:132–8.[CrossRef][Medline]
  9. Tibshirani R, Hastie T, Narasimhan B, Chu G. Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 2002;99:6567–72.[Abstract/Free Full Text]
  10. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 2001;98:5116–21.[Abstract/Free Full Text]
  11. Yan PS, Chen CM, Shi H, et al. Applications of CpG island microarrays for high-throughput analysis of DNA methylation. J Nutr 2002;132:2430–4S.
  12. Thomas JG, Olson JM, Tapscott SJ, Zhao LP. An efficient and robust statistical modeling approach to discover differentially expressed genes using genomic expression profiles. Genome Res 2001;11:1227–36.[Abstract/Free Full Text]
  13. Witten IH, Frank E. Data mining: practical machine learning tools and techniques with Java implementations. San Francisco (California): Morgan Kaufmann; 1999.
  14. Cristianini N, Shawe-Taylor J. An introduction to Support Vector Machines and other kernel-based learning methods. New York: Cambridge University Press; 2000.
  15. Schölkopf B, Warmuth ME, editors. Computational learning theory and kernel machines: 16th Annual Conference on Computational Learning Theory and 7th Kernel Workshop. COLT/Kernel 2003: Proceedings. Lecture Notes in Computer Science. Washington, DC, USA, 2003 August 24-27. New York Springer; 2003.
  16. Storey JD. A direct approach to false discovery rates. J R Stat Soc Ser B 2002;64:479–98.[CrossRef]
  17. Shao J, Tu D. The jackknife and bootstrap. New York: Springer Verlag; 1995.
  18. Ludwig JA, Weinstein JN. Biomarkers in cancer staging, prognosis and treatment selection. Nat Rev Cancer 2005;11:845–56.
  19. Ushijima T. Detection and interpretation of altered methylation patterns in cancer cells. Nat Rev Cancer 2005;5:223–31.[CrossRef][Medline]
  20. Cottrell SE, Laird PW. Sensitive detection of DNA methylation. Ann N Y Acad Sci 2003;983:120–30.[Abstract/Free Full Text]
  21. Li LC, Carroll PR, Dahiya R. Epigenetic changes in prostate cancer: implication for diagnosis and treatment. J Natl Cancer Inst 2005;97:103–15.[Abstract/Free Full Text]
  22. Bae YK, Brown A, Garrett E, et al. Hypermethylation in histologically distinct classes of breast cancer. Clin Cancer Res 2004;10:5998–6005.[Abstract/Free Full Text]
  23. Zochbauer-Muller S, Minna JD, Gazdar AF. Aberrant DNA methylation in lung cancer: biological and clinical implications. Oncologist 2002;7:451–7.[Abstract/Free Full Text]
  24. Yang Q, Zage P, Kagan D, et al. Association of epigenetic inactivation of RASSF1A with poor outcome in human neuroblastoma. Clin Cancer Res 2004;10:8493–500.[Abstract/Free Full Text]
  25. Esteller M, Corn PG, Baylin SB, Herman JG. A gene hypermethylation profile of human cancer. Cancer Res 2001;61:3225–9.[Abstract/Free Full Text]
  26. Spentzos D, Levine DA, Ramoni MF, et al. Gene expression signature with independent prognostic significance in epithelial ovarian cancer. J Clin Oncol 2004;22:4700–10.[Abstract/Free Full Text]
  27. Glinsky GV, Berezovska O, Glinskii AB. Microarray analysis identifies a death-from-cancer signature predicting therapy failure in patients with multiple types of cancer. J Clin Invest 2005;115:1503–21.[CrossRef][Medline]
  28. Laird PW. The power and the promise of DNA methylation markers. Nat Rev Cancer 2003;3:253–66.[CrossRef][Medline]
  29. Motiwala T, Ghoshal K, Das A, et al. Suppression of the protein tyrosine phosphatase receptor type O gene (PTPRO) by methylation in hepatocellular carcinomas. Oncogene 2003;22:6319–31.[CrossRef][Medline]
  30. Wimmer K, Zhu X-x, Rouillard JM, et al. Combined restriction landmark genomic scanning and virtual genome scans identify a novel human homeobox gene, ALX3, that is hypermethylated in neuroblastoma. Genes Chromosomes Cancer 2002;33:285–94.[CrossRef][Medline]
  31. Iliopoulos D, Guler G, Han SY, et al. Fragile genes as biomarkers: epigenetic control of WWOX and FHIT in lung, breast and bladder cancer. Oncogene 2005;24:1625–33.[CrossRef][Medline]
  32. Gourley C, Paige AJ, Taylor KJ, et al. WWOX mRNA expression profile in epithelial ovarian cancer supports the role of WWOX variant 1 as a tumour suppressor, although the role of variant 4 remains unclear. Int J Oncol 2005;26:1681–9.[Medline]
  33. Jeney V, Itoh S, Wendt M, et al. Role of antioxidant-1 in extracellular superoxide dismutase function and expression. Circ Res 2005;96:723–9.[Abstract/Free Full Text]
  34. Mates JM, Sanchez-Jimenez F. Antioxidant enzymes and their implications in pathophysiologic processes. Front Biosci 1999;4:D339–45.[Medline]



This article has been cited by other articles:


Home page
 BioinformaticsHome page
M. Li, H.-i. H. Paik, C. Balch, Y. Kim, L. Li, T. H-M. Huang, K. P. Nephew, and S. Kim
Enriched transcription factor binding sites in hypermethylated gene promoters in drug resistant cancer cells
Bioinformatics, August 15, 2008; 24(16): 1745 - 1748.
[Abstract] [PDF]

Home page
 Cancer Res.Home page
S. Zhang, C. Balch, M. W. Chan, H.-C. Lai, D. Matei, J. M. Schilder, P. S. Yan, T. H-M. Huang, and K. P. Nephew
Identification and Characterization of Ovarian Cancer-Initiating Cells from Primary Human Tumors
Cancer Res., June 1, 2008; 68(11): 4311 - 4320.
[Abstract] [Full Text] [PDF]

Home page
 JCOHome page
K. Appleton, H. J. Mackay, I. Judson, J. A. Plumb, C. McCormick, G. Strathdee, C. Lee, S. Barrett, S. Reade, D. Jadayel, et al.
Phase I and Pharmacodynamic Trial of the DNA Methyltransferase Inhibitor Decitabine and Carboplatin in Solid Tumors
J. Clin. Oncol., October 10, 2007; 25(29): 4603 - 4609.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
D. R. Yates, I. Rehman, M. F. Abbod, M. Meuth, S. S. Cross, D. A. Linkens, F. C. Hamdy, and J. W.F. Catto
Promoter Hypermethylation Identifies Progression Risk in Bladder Cancer
Clin. Cancer Res., April 1, 2007; 13(7): 2046 - 2053.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wei, S. H.
Right arrow Articles by Nephew, K. P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wei, S. H.
Right arrow Articles by Nephew, K. P.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Cancer Research Clinical Cancer Research
Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics
Molecular Cancer Research Cancer Prevention Research
Cancer Prevention Journals Portal Cancer Reviews Online
Annual Meeting Education Book Meeting Abstracts Online