Printer Friendly

DNA methylation biomarkers for blood-based colorectal cancer screening.

Of the new molecular approaches, the assessment of epigenetic events is one of the most promising means of identifying marker candidates for the early detection of cancer. The regulation of gene expression by aberrant methylation has been well characterized in tumor biology in general (1, 2) and has been extensively described for colorectal cancer (CRC) [6] in particular (3-5). The development of blood-based cancer-detection tests should improve patient compliance and thereby increase the detection of disease at earlier stages. Increased concentrations of circulating methylated DNA have been reported in the blood of cancer patients (6, 7); however, a routine process for validating epigenetic biomarkers has not been implemented. We previously proposed a systematic marker-identification process that identifies aberrantly methylated genes in cancer tissue and validates the results in plasma samples from affected individuals (8). We have used CRC as a model for implementing this process and describe criteria that are necessary to ensure optimal biomarker performance. In particular, sensitivity and specificity must be optimized through the identification of markers that show the highest differences in methylation between the cancer and the background. This goal can be achieved by comparing the degree of DNA methylation in the target cancer with that in a healthy tissue sample from the same organ, healthy blood samples, and organs from which cells could be present in the bloodstream as a result of disease conditions associated with the tested population. Also extremely important is to develop high-performing assays capable of detecting extremely low amounts of methylated DNA in a very high background of unmethylated or partially methylated DNA. In this study, we describe the selection and validation of potential blood-based epigenetic biomarkers for early detection of CRC or for other clinical applications and report the performance of 3 of these markers in the analysis of plasma samples.


Materials and Methods


Written informed consent was obtained from all study participants, and the process adhered to local ethics guidelines. CRC tissue was obtained from surgical resections, fresh-frozen, and stored at -80 [degrees]C. Samples of healthy tissue were obtained at least 6 cm from a tumor and stored as described above. Plasma samples were obtained from healthy individuals and CRC patients aminimum of 1 week after colonoscopy. The clinical diagnosis of CRC was confirmed by histologic analysis, and colonoscopy examinations confirmed that healthy individuals had no colon-related disease (see Table 1 in the Data Supplement that accompanies the online version of this article at


We used methylation-sensitive restriction enzyme-based discovery methods [methylation-specific arbitrarily primed PCR and methylated CpG island amplification (MCA) (9, 10)] to identify sequences that were differentially methylated in pathology-verified tumors by comparing the results with those obtained for healthy tissues and peripheral blood lymphocytes (PBLs) from people in the same age group. We extracted DNA from tissue samples with Genomic-tip 500/G columns (Qiagen) and used the DNA Isolation Kit I with a MagNA Pure LC System (Roche Applied Science) to extract DNA from PBLs (11).


We used microarrays as previously described to study diseased and healthy colon-derived tissues, other cancer tissues, and samples of healthy tissue from the same organs (12). Probes were designed for the region identified in the discovery process (typically a fragment of 250-500 bp), and up to 1000 bp of flanking sequence was included if it was within a CpG island. DNA was extracted from tissues as described above. We designed 2 arrays, each with approximately 250 probes covering [greater than or equal to]54 different amplicons. The first array was interrogated with amplicons from 358 tissues (89 CRC samples, 55 colorectal polyps, 31 samples of inflammatory bowel disease, 116 extracolonic cancers, 29 healthy colon samples, 14 nonpathologic PBL samples, and 24 other noncolonic, nonpathologic samples). The second array was tested with 429 tissue samples (18 CRC samples, 311 extracolonic cancers, 34 nonpathologic PBL samples, and 66 other noncolonic, nonpathologic samples).


MethyLight assays (13) were designed for regions identified in the discovery process and confirmed by microarray analysis. Assays for marker candidates identified from the literature were designed for the promoters or first exons of genes and were within CpG islands. Heavy methyl (HM) quantitative real-time PCR (qPCR) assays were designed as previously described (14). We used CpGenome Universal Methylated DNA (MP Biomedicals) at concentrations between 31.6 [micro]g/L and 31.6 mg/L for calibration curves. We evaluated the 90% limit of detection (LOD) for each assay with a dilution series of CpGenome Universal Methylated DNA in a background of 50 ng human genomic DNA (Roche Applied Science). The 90% LOD was defined as the lowest concentration of spiked methylated DNA in a background of 50 ng human genomic DNA for which the measurement values had an area under the ROC curve (AUC) of [greater than or equal to]0.9 compared with measurements without spiked methylated DNA.


We obtained 114 CRC tissue samples, 51 samples of nonpathologic colon tissues that had been fresh-frozen at surgery, and 51 PBL samples from healthy individuals in the same age range as the patients. Surgical samples were verified for tumor content. We extracted DNA as in the biomarker-discovery process and treated the DNA with bisulfite according to a previously described protocol (15). We quantified bisulfite-modified DNA (bisDNA) with amethylation-nonspecific assay for an intragenic region of the GSTP1 [7] gene (glutathione S-transferase pi) and adjusted the DNA concentration to 10 ng/PCR reaction for subsequent testing. Calibration curves were conducted for each assay as described above with CpGenome Universal Methylated DNA concentrations between 400 [micro]g/L and 5 mg/L. We tested 10 ng DNA per reaction in duplicate for each assay on the LightCycler 2.0 (Roche Diagnostics) or the ABI Prism 7900 instrument (Applied Biosystems).


We chose a sample size of 150 for plasma samples from CRC patients to achieve confidence limits of [+ or -]8% around an assumed sensitivity of 50%. The targeted sample size for plasma samples from healthy individuals was 200 for confidence limits of [+ or -]5% around an assumed specificity of 90%. We obtained blood samples via phlebotomy into EDTA-containing Vacutainer tubes (BD Medical Systems). We separated the plasma by centrifugation (1500g for 10 min) within 4 h of blood collection and then carefully removed the buffy coat. The plasma was placed in a 15-mL tube and centrifuged at 1500g for 10 min. If we obtained more than 1 tube per patient, we pooled the plasma before storage at -80 [degrees]C.


Patient samples were randomized and blinded as to the clinical diagnosis. We prepared DNA by a 3-step process. First, we extracted free-floating DNA with the MagNa Pure LC Total Nucleic Acid Isolation Kit-Large Volume (Roche Applied Science) and the MagNA Pure System. We distributed 6 mL over 6 MagNA Pure wells and concentrated the eluates with Microcon filters (Millipore). We quantified the amount of genomic DNA recovered after extraction by means of a qPCR assay with primers that contained no cytosine residues (cytosine-free fragment, CFF1). Second, we treated the DNA with bisulfite and measured the DNA concentration with the bisDNA assay. Third, we measured the DNA concentrations (in milligrams per liter) of methylated TMEFF2 (transmembrane protein with EGF-like and two follistatin-like domains 2), NGFR [nerve growth factor receptor (TNFR superfamily, member 16)], and SEPT9 (septin 9) DNA for each sample by means of a single qPCR reaction for each HM assay and a calibration curve, as described above. All assays were run on the LightCycler 2.0 instrument. See Table 2 in the online Data Supplement for details regarding the assay components and running conditions for HM marker and control assays.


We analyzed microarray and qPCR tissue data by using [log.sub.10]-transformed percent-methylation rates (PMRs) (12). PCR analyses of plasma samples were based on [log.sub.10]-transformed DNA concentrations (as measured by the respective methylation-marker, bisDNA, or CFF1 assay) normalized to the plasma volume. Plasma samples with no measurable bisDNA were eliminated from analysis. We estimated DNA amounts for all qPCR assays from calibration curves by linear regression on the crossing points by the second-derivative maximum method (16). Plasma DNA estimates from methylation-specific assays were scaled by the maximum fluorescence of the respective amplification curve relative to the maximum fluorescence of the calibrators. For tissue samples, analyses were based on the means of replicate marker measurements.

For univariate analyses, we estimated AUCs by means of the trapezoidal rule. The Wilcoxon-Mann-Whitney test was used for quantitative comparisons of 2 samples, the McNemar test was used to compare the accuracies of qualitative markers, and the Kruskal-Wallis test was used to compare results for disease stages. Logistic regression and the Wald test were used for multivariate panel analysis (17). The predictive performance of each multivariate model is given as an AUC value corrected for overfitting bias and was calculated from the bootstrap-corrected concordance index (C index) (18). All P values are 2-sided. Confidence intervals for proportions of amplified samples were set at 95% and based on binomial distributions.



We assessed >600 marker candidates identified in genome-wide discovery experiments with a previously described scoring system that accounts for the number of independent identifications of that marker during discovery, the location of the given marker within a CpG island or the promoter region of a gene, identification in multiple experiments, and other criteria (12). Only markers with a score of at least 2 were advanced for further analysis.

Although sequences identified by methylation-specific arbitrarily primed PCR and methylated CpG island amplification indicated differentially methylated marker candidates, the statistical significance of differences between sample types cannot be evaluated with these methods. In addition, it is not technically possible to compare more than a few tissue types simultaneously. Therefore, we used DNA-methylation arrays and/or qPCR to quantify the extent of marker candidate methylation in different tissue types. Because of time and resource limitations, a subset of the candidates identified in the genome-wide discovery process (76 total) and 26 candidates identified from the literature were further validated in 2 separate methylation-microarray studies; 25 marker candidates are shown in Table 1 to illustrate the selection process for blood-based screening markers. A list with 31 additional candidates can be found in Table 3 in the Data Supplement that accompanies the online version of this article at


The DNA methylation-array data allowed us to rank marker candidates by their ability to differentiate neoplastic and control samples and by their methylation rate in PBLs. We used 3 criteria to select the best markers among the 45 array candidates for further investigation. First, discrimination between CRC and healthy colon tissue had to be significant with an AUC [greater than or equal to]0.70. Second, the differentiation of CRC and all other tissue samples had to be significant with an AUC [greater than or equal to]0.65. Third, PBL samples were required to have an absolute methylation rate of [less than or equal to]10%.

With these criteria, we increased the probability that our marker candidates obtained their hypermethylation signature during tumorigenesis, that the signature was reasonably specific for CRC, and that candidates would not be methylated in the target analyte.

Twenty-seven of the 45 marker candidates passed the testing with these initial selection criteria for informative DNA methylation-based biomarkers suitable for testing in the screening application (22 markers from the genome-wide discovery process and 5 from the literature). The TMEFF2 marker displayed the best performance in the array study with regard to all of the selection criteria. The SEPT9 and VTN (vitronectin) markers achieved scores comparable with TMEFF2 (Table 1).


Analyses of DNA-methylation microarrays allow the rapid quantification of methylation and ranking of marker candidates, but the analytical sensitivity of microarrays is reduced compared with PCR analysis. Moreover, arrays do not always generate candidates that show the consistent comethylation over several CpG positions that is required for qPCR assays. Therefore, we next developed qPCR assays (MethyLight or HM) (13, 14) for the 27 marker candidates selected from the array studies and assessed methylation with an independent sample set. We added 11 markers [GATA3, GATA binding protein 3; GDNF, glial cell derived neurotrophic factor; OPCML, opioid binding protein/cell adhesion molecule-like; PENK, proenkephalin; TFAP2A, transcription factor AP-2 alpha (activating enhancer binding protein 2 alpha); APP, amyloid beta (A4) precursor protein (peptidase nexin-II, Alzheimer disease); CACNA1G, calcium channel, voltage-dependent, T type, alpha 1G subunit; HOXA1, homeobox A1; NEUROG1, neurogenin 1; APBA2, amyloid beta (A4) precursor protein-binding, family A, member 2; and TRRAP, transformation/transcription domain-associated protein on the basis of a study that indicated increased methylated DNA in CRC tissue (19).

The qPCR assays possessed the following technical characteristics and fulfilled the requisite analytical-performance criteria. First, the maximum length of the amplicon was [less than or equal to]150 bp. Second, the assay consistently detected 50 pg of completely methylated DNA. Third, the assay showed no detectable amplification of 50 ng of unmethylated DNA. Fourth, 50 pg of completely methylated DNA was consistently detected in a background of 50 ng of unmethylated DNA. Assay development was attempted for all 38 marker candidates. Nine candidates failed the initial criteria, such as assay length or unsuitable CpG site location for primer and probe placement [BCOR, BCL6 co-repressor; CAV1, caveolin 1, caveolae protein, 22kDa; CD44, CD44 molecule (Indian blood group); CDH13, cadherin 13, H-cadherin (heart); VCAN (formerly CSPG2), versican; FCGR2A, Fc fragment of IgG, low affinity IIa, receptor (CD32); GSK3B, glycogen synthase kinase 3 beta; PCDH17, protocadherin 17; and TAF11, TAF11 RNA polymerase II, TATA box binding protein (TBP)-associated factor, 28kDa]. Of the remaining 29 assays, 22 passed analytical testing and were tested in studies of tissue samples.

For advancement of a marker candidate to plasma testing, we required the assay to fulfill the following biological criteria. First, differentiation of CRC tissue from healthy colon tissue had to be significant with an AUC [greater than or equal to]0.70. Second, PBLs were required to have an absolute methylation rate of [less than or equal to]0.2%. The requirement for distinguishing CRC tissue from healthy colon tissue is identical to the selection criterion for first microarray and confirms that the markers were successfully transferred to the qPCR platform. The low-methylation requirement for PBLs was made more stringent by approximately 2 orders of magnitude because of the high analytical sensitivity of the qPCR assays. To facilitate faster screening of candidate markers, we applied a simplified qualitative version of the PBL criterion to some assays by requiring no amplification in 5 replicate measurements of 50 ng of PBL DNA.

We tested marker candidate assays in 2 tissue studies. The first study consisted of 28 CRC samples, 46 samples of healthy colon, and 25 PBL samples. The second study consisted of 86 CRC samples, 23 healthy colon samples, and 26 PBL samples. Assays for markers TMEFF2 and NGFR were run in both studies. The results of the first study have previously been reported and are summarized in Table 1 with the results of the second study (12). Methylation-rate distributions for selected candidate markers are shown in Fig. 1A.

On the basis of the results for the tissue samples, 10 of the 22 qPCR marker assay candidates fulfilled the sensitivity and specificity criteria and were advanced to plasma testing. Hypothesizing that a higher methylation rate in cancer tissue implies a higher concentration of methylated target molecules in the bloodstream, we further ranked the candidate assays by median methylation rate in tumor tissue. TMEFF2, SEPT9, EYA4 [eyes absent homolog 4 (Drosophila)], and VTN were clearly superior with [greater than or equal to]20% methylation in >50% of the CRC samples.

We carried out multivariate logistic regression analysis for all possible 2- and 3-marker combinations to try to improve our ability to differentiate control and CRC samples. As we previously have observed (12), markers were hypermethylated in a common subset of CRC samples(88% of CRC samples with a PMR of >10% for any marker) and showed a typical CpG island methylator phenotype (20). TMEFF2 and SEPT9 showed very little complementarity (Fig. 1B). Therefore, the improvement in discrimination performance over the best single marker was small (Table 2). The best-performing marker panel was the combination of SEPT9 and NGFR, with a slight but significant bias-corrected AUC improvement of 2% over SEPT9 alone (P = 0.03).



Final assay optimization. The detection of methylated target in plasma requires assays that amplify extremely low quantities of methylated tumor-marker DNA in a vast excess of unmethylated or partially methylated background DNA. To achieve this combination of technical sensitivity and specificity, we optimized qPCR assays selected from the tissue studies that used the HM technology [TUSC3, tumor suppressor candidate 3; TMEFF2; ZDHHC22, zinc finger, DHHC-type containing 22; VTN; FOXL2, forkhead box L2; NGFR; ALX4, aristaless-like homeobox 4; SIX6, SIX homeobox 6; EYA4; and SEPT9] (14). The HM assays were required to meet several criteria. First, the absolute technical sensitivity of the assay had to be a maximum of 25 pg. Second, the assay could show no amplification of 50 ng of unmethylated DNA. Third, the assay had to detect at least 25 pg of methylated DNA in a background of 50 ng of unmethylated DNA. The correlation between the initial MethyLight assays and the HM versions was required to be >0.8 with a subset of previously tested tissue samples. The LOD in blood was required to be, at minimum, 50 pg of methylated DNA in a background of 50 ng PBL DNA, tested on several batches. Lastly, the positive rate in plasma from healthy individuals, measured as the proportion of healthy plasma samples showing amplification, should not be >20%.

The first 3 of these criteria are stricter versions of the analytical-performance requirements determined for tissue-based studies. The correlation criterion was introduced to ensure that the final assay measures the same biological information as the original candidate assay. The LOD criterion ensures that the assays reliably detect methylated target molecules in blood. The final filter requires reasonable specificity for a small number of plasma samples and ensures the healthy tissue-based and PBL-based model system did not miss any dominant source of methylated target DNA.

We eliminated 1 HM assay (ZDHHC22) during analytical testing because of a lack of concordance with MethyLight results. Two HM assays failed to reliably detect methylated targets in a blood background because of a low level of methylation in some PBL batches (TMEFF2 LOD, 70 pg; TUSC3 LOD, >100 pg). Although the results were negative with human PBLs, the VTN assay was eliminated after it showed high concentrations of methylated target in healthy human plasma.

Study design and results of the control assay. Because of the limited volume of plasma sample per patient, only 3 HM assays could be tested with plasma. We selected SEPT9 because of its superior performance with respect to all of the selection criteria. NGFR was selected despite its low methylation rates in CRC tissue samples because it showed the highest complementarity to SEPT9 in the multivariate tissue analysis. Despite its failure to meet our LOD criterion, we selected TMEFF2 to compare our results with the existing literature (7, 21) and to test our hypothesis that the degree of PBL amplification predicts plasma specificity.

The set of plasma samples included 320 samples from 185 healthy individuals (median age, 56 years; 60% female) and 135 patients with CRC (median age, 65 years; 50% female). Six samples from the healthy category were omitted from analysis (3 because of no measurable bisDNA and 3 because of sample-processing errors), and 2 samples from the CRC category were omitted from analysis (1 because of no measurable bisDNA and 1 because of radiotherapy before blood draw). The results showed a significantly higher median bisDNA amount in plasma samples from CRC patients than in plasma samples from healthy individuals (P<0.001). The distributions in DNA concentration for the 2 classes showed very strong overlap (Fig. 2A), however, and class differentiation was weak (AUC, 0.61) (Table 3).

Clinical performance of methylation markers. A qualitative analysis indicated the most sensitive marker to be SEPT9, which amplified 69% of CRC plasma samples, followed by TMEFF2 (65%) and NGFR (51%) (Table 3). NGFR and SEPT9 had similar specificities and amplified 16% and 14% of plasma samples from healthy individuals, respectively. The least specific marker was TMEFF2, with 31% positivity for plasma samples from healthy individuals. Consequently, SEPT9 predicted the presence of CRC significantly more accurately than TMEFF2 (P <0.001) or NGFR (P <0.01). There was no significant difference in diagnostic accuracy between the TMEFF2 and NGFR markers.

A quantitative analysis demonstrated SEPT9 to have significantly higher concentrations of methylated DNA in plasma samples from CRC patients than NGFR (P <0.001, Fig. 2A). In plasma samples from healthy individuals, SEPT9 had significantly lower concentrations of methylated DNA than TMEFF2 (P <0.001). TMEFF2, SEPT9, and bisDNA showed a significant increase in target DNA to be correlated with the stage of disease (P <0.05, Fig. 2B); NGFR showed no significant difference between disease stages. Consequently, the biomarker with the highest performance was SEPT9 (AUC, 0.80; Fig. 2C). A cutoff of 0.011 [micro]g/L of methylated SEPT9 DNA produces a specificity of 95% and a sensitivity of 52%.

We performed multivariate logistic regression analyses for all combinations of these 3 markers (Table 2). All 3 markers were highly correlated for a subset of plasma samples from CRC patients with high concentrations of tumor DNA(Fig. 3). For cancer samples with amounts of tumor DNA close to or below the LOD, the amplification results were in the stochastic range, and the markers were no longer correlated. As with the analysis of tissue samples, the best performing marker panel was the combination of SEPT9 and NGFR, with a small but significant bias-corrected AUC improvement of 1% over SEPT9 alone (P = 0.04) (Table 2).


A variety of genome-wide methods are currently available for the discovery of differentially methylated markers (22, 23). These methods typically produce large numbers of potential candidates; thus, downstream selection processes are critical for identifying clinically relevant markers that have the properties necessary to perform adequately in future tests. To ensure adequate sensitivity and specificity in plasma samples, we have experimentally established benchmarks for the performance of analytical assays, for minimum methylation levels in CRC tissues, and for maximum tolerable methylation levels in PBLs (see Table 4 in the online Data Supplement for an overview). For example, the qPCR assay for TMEFF2, although fairly specific with PBLs from healthy individuals, does not provide the level of specificity needed for use as a screening test without first establishing a cutoff concentration that consequently reduces the marker's sensitivity. Our results with plasma are somewhat contradictory to those of Sabbioni et al., who found a 100% specificity for the 16 healthy control individuals they tested (7). We suspect that methylation of TMEFF2 occurs randomly in PBLs in the healthy population, although at a very low incidence, making this marker more appropriate for applications that do not have high specificity requirements. On the other hand, the NGFR and SEPT9 assays have equally good PBL specificities and analytical sensitivities, but the SEPT9 assay detects much higher levels of methylation in individual samples of CRC tissue, a phenomenon that appears to manifest as a higher clinical sensitivity in plasma samples. It is this type of marker assay that has the greatest chance for success as a screening marker in blood, and efforts are currently focused on the commercial development of this marker for the early detection of CRC. We found several other markers that passed all stages of marker validation but were not included in the final plasma testing because of insufficient sample volumes. These markers (ALX4, EYA4, FOXL2, and SIX6) will be tested in subsequent studies to evaluate their clinical performance.


Patient compliance and the performance of current noninvasive screening strategies limit the effectiveness of CRC screening tests available on the market today. Because of the relative ease of performing blood-screening evaluations, tests of this type can be administered regularly during routine physical examinations. An easily administered blood-based test for the early detection of CRC followed by colonoscopy for individuals with positive results has the potential to be an effective tool for reducing mortality from this disease. This study has developed a discovery/validation process that produces promising candidates for use in screening tests and other potential diagnostic applications. Further validation is now focused on testing the markers, particularly SEPT9, in larger studies that include plasma samples from CRC patients, individuals with preneoplastic disease, and other healthy and disease controls. Application as a screening biomarker will require a large prospective trial in an asymptomatic, CRC-screening guideline-eligible population.

Grant/funding Support: M.E. is supported by grants from the Else Kroner-Fresenius-Stiftung and the Deutsche Krebshilfe (Germany).

Financial Disclosures: C.L.-D., F.M., T.D., R.T., J.D., M.S., X.S., R.L., V.L., and A.S. are employees of Epigenomics and may be stockholders of the company. M.E. is a consultant for Epigenomics.

Acknowledgments: The authors thank the assay technology group at Epigenomics for support in the development of the qPCR assays and the screening development group for performance of the clinical studies.

Received August 6, 2007; accepted November 21, 2007.

Previously published online at DOI: 10.1373/clinchem.2007.095992


(1.) Jones PA, Laird PW. Cancer epigenetics comes of age. Nat Genet 1999;21:163-7.

(2.) Baylin SB, Esteller M, Rountree MR, Bachman KE, Schuebel K, Herman JG. Aberrant patterns of DNA methylation, chromatin formation and gene expression in cancer. Hum Mol Genet 2001;10: 687-92.

(3.) Toyooka S, Toyooka KO, Harada K, Miyajima K, Makarla P, Sathyanarayana UG, et al. Aberrant methylation of the CDH13 (H-cadherin) promoter region in colorectal cancers and adenomas. Cancer Res 2002;62:3382-6.

(4.) Toyota M, Shen L, Ohe-Toyota M, Hamilton SR, Sinicrope FA, Issa JP. Aberrant methylation of the cyclooxygenase 2 CpG island in colorectal tumors. Cancer Res 2000;60:4044-8.

(5.) Ebert MP, Mooney SH, Tonnes-Priddy L, Lograsso J, Hoffmann J, Chen J, et al. Hypermethylation of the TPEF/HPP1 gene in primary and metastatic colorectal cancers. Neoplasia 2005;7:771-8.

(6.) Herrera LJ, Raja S, Gooding WE, El-Hefnawy T, Kelly L, Luketich JD, Godfrey TE. Quantitative analysis of circulating plasma DNA as a tumor marker in thoracic malignancies. Clin Chem 2005; 51:113-8.

(7.) Sabbioni S, Miotto E, Veronese A, Sattin E, Gramantieri L, Bolondi L, et al. Multigene methylation analysis of gastrointestinal tumors: TPEF emerges as a frequent tumor-specific aberrantly methylated marker that can be detected in peripheral blood. Mol Diagn 2003;7:201-7.

(8.) Lofton-Day C, Lesche R. DNA methylation markers in patients with gastrointestinal cancers. Current understanding, potential applications for disease management and development of diagnostic tools. Dig Dis 2003;21:299-308.

(9.) Liang G, Salem CE, Yu MC, Nguyen HD, Gonzales FA, Nguyen TT, et al. DNA methylation differences associated with tumor tissues identified by genome scanning analysis. Genomics 1998;53: 260-8.

(10.) Toyota M, Ho C, Ahuja N, Jair K-W, Li Q, Ohe-Toyota M, et al. Identification of differentially methylated sequences in colorectal cancer by methylated CpG island amplification. Cancer Res 1999;59:2307-12.

(11.) Baumgartner A, Girgnhuber H, Malmberg W, Nieswandt V, Eberle W, Kirchgesser M. MagNa Pure Compact System: purification of high-quality DNA from PBMCs, WBCs and buffy coat. Biochemica 2005;2:9-11.

(12.) Model F, Osborn N, Ahlquist D, Gruetzmann R, Molnar B, Sipos F, et al. Identification and validation of colorectal neoplasia-specific methylation markers for accurate classification of disease. Mol Cancer Res 2007;5:153-63.

(13.) Eads CA, Danenberg KD, Kawakami K, Saltz LB, Blake C, Shibata D, et al. MethyLight: a high-throughput assay to measure DNA methylation. Nucleic Acids Res 2000;28:e32.

(14.) Cottrell SE, Distler J, Goodman NS, Mooney SH, Kluth A, Olek A, et al. A real-time PCR assay for DNA-methylation using methylation-specific blockers. Nucleic Acids Res 2004;32:e10.

(15.) Tetzner R, Dietrich D, Distler J. Control of carryover contamination for PCR-based DNA methylation quantification using bisulfite treated DNA. Nucleic Acids Res 2007;35:e4.

(16.) Rasmussen R. Quantification on the LightCycler. In: Meuer S, Wittwer C, Nakagawara K, eds. Rapid cycle real-time PCR, methods and applications. Heidelberg: Springer Press, 2001:21-34.

(17.) Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley & Sons, 2000.

(18.) Harrel FE. Regression modeling strategies. 1st ed. New York: Springer-Verlag, 2001:568pp.

(19.) Weisenberger DJ, Siegmund KD, Campan M, Young J, Long TI, Faasse MA, et al. CpG island methylator phenotype underlies sporadic microsatellite instability and is tightly associated with BRAF mutation in colorectal cancer. Nat Genet 2006;38:787-93.

(20.) Issa, JP. CpG-island methylation in aging and cancer. Curr Top Microbiol Immunol 2000;249: 101-18.

(21.) Wallner M, Herbst A, Behrens A, Crispin A, Stieber P, Goke B, et al. Methylation of serum DNA is an independent prognostic marker in colorectal cancer. Clin Cancer Res 2006;12:7347-52.

(22.) Laird PW. The power and the promise of DNA methylation markers. Nat Rev Cancer 2003;3: 253-66.

(23.) Lewin J, Plum A, Hildmann T, Rujan T, Eckhardt F, Liebenberg V, et al. Comparative DNA methylation analysis in normal and tumour tissues and in cancer cell lines using differential methylation hybridisation. Int J Biochem Cell Biol 2007;39: 1539-50.

Catherine Lofton-Day, [1] * [[dagger]] Fabian Model, [2] ([dagger]) Theo DeVos, [1] Reimo Tetzner, [2] Juergen Distler, [2] Matthias Schuster, [2] Xiaoling Song, [1] Ralf Lesche, [2] Volker Liebenberg, [2] Matthias Ebert, [3] Bela Molnar, [4] Robert Grutzmann, [5] Christian Pilarsky, [5] and Andrew Sledziewski [1]

[1] Epigenomics, Inc., Seattle, WA; [2] Epigenomics AG, Berlin, Germany; [3] Technical University, Munich, Germany; [4] Semmelweis University, Budapest, Hungary; [5] Technical University of Dresden, Dresden, Germany.

* Address correspondence to this author at: Epigenomics, Inc., 1000 Seneca St., Suite 300, Seattle, WA 98101. e-mail

[[dagger]] These authors contributed equally to this work.

[6] Nonstandard abbreviations: CRC, colorectal cancer; PBL, peripheral blood lymphocytes; HM, heavy methyl; qPCR, quantitative real-time PCR; LOD, limit of detection; AUC, area under the ROC curve; bisDNA, bisulfite-modified DNA; PMR, percent-methylation rate.

[7] Human genes: TMEFF2, transmembrane protein with EGF-like and two follistatin-like domains 2; NGFR, nerve growth factor receptor (TNFR superfamily, member 16); SEPT9, septin 9; GSTP1, glutathione S-transferase pi; VTN, vitronectin; GATA3, GATA binding protein 3; GDNF, glial cell derived neurotrophic factor; OPCML, opioid binding protein/cell adhesion molecule-like; PENK, proenkephalin; TFAP2A, transcription factor AP-2 alpha (activating enhancer binding protein 2 alpha); APP, amyloid beta (A4) precursor protein (peptidase nexin-II, Alzheimer disease); CACNA1G, calcium channel, voltage-dependent, T type, alpha 1G subunit; HOXA1, homeobox A1; NEUROG1, neurogenin 1; APBA2, amyloid beta (A4) precursor protein-binding, family A, member 2; TRRAP, transformation/transcription domain-associated protein; BCOR, BCL6 co-repressor; CAV1, caveolin 1, caveolae protein, 22kDa; CD44, CD44 molecule (Indian blood group); CDH13, cadherin 13, H-cadherin (heart); VCAN(formerly CSPG2), versican; FCGR2A, Fc fragment of IgG, low affinity IIa, receptor (CD32); GSK3B, glycogen synthase kinase 3 beta; PCDH17, protocadherin 17; TAF11, TAF11 RNA polymerase II, TATA box binding protein (TBP)-associated factor, 28kDa; EYA4, eyes absent homolog 4 (Drosophila); TUSC3, tumor suppressor candidate 3; ZDHHC22, zinc finger, DHHC-type containing 22; FOXL2, forkhead box L2; ALX4, aristaless-like homeobox 4; SIX6, SIX homeobox 6; BCL6, B-cell CLL/ lymphoma 6 (zinc finger protein 51); DNAJC5, DnaJ (Hsp40) homolog, subfamily C, member 5; DUX2, double homeobox, 2; ESTG020896(no HUGO ID available); MSH6, mutS homolog 6 (E. coli); ESTG2308609 (no HUGO ID available); DLX5, distal-less homeobox 5; NPBWR1(formerly GPR7), neuropeptides B/W receptor 1; APBA2, amyloid beta (A4) precursor protein-binding, family A, member 2 (X11-like); SLC32A1, solute carrier family 32 (GABA vesicular transporter), member 1; SLITRK1, SLIT and NTRK-like family, member 1; KCTD12, potassium channel tetramerisation domain containing 12; ONECUT2, one cut homeobox 2; RNF4, ring finger protein 4; SMAD7, SMAD family member 7.
Table 1. Step-wise validation results for a selected subset of 25
marker candidates derived from the discovery process and from the
literature. (a)

 Methylation array

Gene symbol vs all vs colon median
(HUGO) controls controls PMR, %

BCL6 0.73 0.78 25
DNAJC5 0.67 0.73 23
DUX2 0.75 0.70 17
ESTG020896 (b) 0.64 0.63 9
MSH6 (c) 0.58 0.64 0
ESTG2308609 (b) 0.62 0.56 22
DLX5 0.84 0.83 0
NPBWR1 (GPR7) 0.73 0.82 0
APBA2 (c) NT (d) NT NT
SLC32A1 0.85 0.89 0
SLITRK1 0.86 0.85 1
KCTD12 0.67 0.77 0
ONECUT2 0.72 0.77 2
RNF4 0.79 0.84 0
SMAD7 0.68 0.71 0
TUSC3 (c) 0.82 0.74 1
TMEFF2 0.90 0.94 0
ZDHHC22 0.88 0.87 9
VTN 0.93 NT 1
FOXL2 0.67 NT 0
NGFR 0.83 0.87 0
ALX4 0.65 0.83 0
SIX6 0.77 0.85 0
EYA4 0.66 0.81 4
SEPT9 0.84 NT 0

 Real-time PCR

 PBL: Healthy colon: CRC:
Gene symbol median median PMR median PMR,
(HUGO) PMR, % (90% range), % (90% range), %

ESTG020896 (b)
MSH6 (c)
ESTG2308609 (b)
APBA2 (c) PBL +
SLC32A1 1.0
KCTD12 0.0 0 (0-0) 0 (0-28)
ONECUT2 0.0 1 (0-1) 0 (0-2)
RNF4 0.0 0 (0-0) 0 (0-0)
SMAD7 0.0 0 (0-0) 0 (0-1)
TUSC3 (c) 0.2 3 (1-10) 11 (1-32)
TMEFF2 0.1 4 (1-9) 30 (1-81)
ZDHHC22 0.0 0 (0-1) 3 (0-27)
VTN 0.0 2 (0-4) 20 (0-77)
FOXL2 0.0 NT 7 (0-38)
NGFR 0.0 0 (0-2) 3 (0-60)
ALX4 0.1 1 (0-2) 11 (0-58)
SIX6 0.0 0 (0-0) 16 (0-46)
EYA4 0.0 2 (0-4) 26 (0-76)
SEPT9 0.0 2 (1-7) 26 (0-89)

 Real-time PCR

Gene symbol
(HUGO) AUC Exclusion criteria

BCL6 Failed chip-selection
ESTG020896 (b)
MSH6 (c)
ESTG2308609 (b)
DLX5 Failed PBL filter
APBA2 (c)
KCTD12 0.53 Insufficient CRC
ONECUT2 0.24
RNF4 0.52
SMAD7 0.59
TUSC3 (c) 0.71 Failed LOD test
TMEFF2 0.81
ZDHHC22 0.78 HM assay failed
VTN 0.81 Plasma positive
FOXL2 NT Candidates for
 plasma testing
NGFR 0.69
ALX4 0.74
SIX6 0.83
EYA4 0.88
SEPT9 0.87

(a) A list with 31 additional candidates can be found in Table 3 in
the online Data Supplement. Data for some markers have previously been
reported Model et al. (12). Results that led to the exclusion of a
candidate gene are printed in boldface.

(b) No HUGO gene name or symbol available.

(c) Marker candidates derived from the literature.

(d) NT, not tested.

Table 2. Marker panels.

 Odds ratio (a)
Panel Marker (95% CI (b))


TMEFF2 + NGFR TMEFF2 2.3 (1.5-3.6) 0.05 0.81
 NGFR 2.0 (1.4-3.0) 0.07
SEPT9 + NGFR SEPT9 4.6 (2.9-7.1) 0.0005 0.89
 NGFR 2.4 (1.6-3.7) 0.03
SEPT9 + TMEFF2 SEPT9 3.8 (2.5-5.7) 0.001 0.87
 TMEFF2 2.5 (1.6-4.0) 0.05
SEPT9 + NGFR + SEPT9 4.3 (2.7-7.0) 0.003 0.89
TMEFF2 NGFR 2.4 (1.5-3.7) 0.06
 TMEFF2 1.2 (0.7-2.2) 0.7


TMEFF2 + NGFR TMEFF2 1.4 (1.2-1.6) 0.004 0.74
 NGFR 1.7 (1.5-2.0) 0.0003
SEPT9 + NGFR SEPT9 2.6 (2.3-3.1) <0.0001 0.81
 NGFR 1.4 (1.2-1.6) 0.04
SEPT9 + TMEFF2 SEPT9 2.8 (2.4-3.2) <0.0001 0.81
 TMEFF2 1.3 (1.1-1.4) 0.05
SEPT9 + NGFR + SEPT9 2.6 (2.2-3.1) <0.0001 0.79
TMEFF2 + NGFR 1.2 (1.0-1.5) 0.3
bisDNA TMEFF2 1.2 (1.0-1.4) 0.4
 bisDNA 1.0 (0.8-1.3) 0.9

(a) Tissue odds are per 10-fold change in PMR; plasma odds are
per 10-fold change in DNA concentration.

(b) CI, confidence interval.

Table 3. Plasma performance of single markers.


Assay LOD in background of 50 ng
PBL DNA, pg/reaction 70 21
Positives, %
 Normal plasma (a) 31 16
 CRC plasma 65 51
Median DNA (90% range), mg/L
 Normal plasma 0 (0-0.078) 0 (0-0.017)
 CRC plasma 0.010 (0-3.316) 0 (0-0.937)
AUC 0.72 0.70
P 0 0
Cutoff for 95% specificity for 0.098 0.019
 normal plasma (n = 179), mg/L
Sensitivity at 95% specificity by CRC stage, %
 All stages (n = 133)b 30 33
 Stage 0/I (n = 20) 5 20
 Stage II (n = 32) 22 25
 Stage III (n = 47) 34 36
 Stage IV (n = 31) 45 36


Assay LOD in background of 50 ng
PBL DNA, pg/reaction 19
Positives, %
 Normal plasma (a) 14 100
 CRC plasma 69 100
Median DNA (90% range), mg/L
 Normal plasma 0 (0-0.010) 2.7 (0.7-112.3)
 CRC plasma 0.013 (0-4.105) 3.5 (0.6-498.4)
AUC 0.80 0.61
P 0 0.0001
Cutoff for 95% specificity for 0.011 150.3
 normal plasma (n = 179), mg/L
Sensitivity at 95% specificity by CRC stage, %
 All stages (n = 133)b 52 15
 Stage 0/I (n = 20) 30 0
 Stage II (n = 32) 56 16
 Stage III (n = 47) 45 21
 Stage IV (n = 31) 68 16

(a) Normal plasma indicates plasma from healthy individuals
(i.e., non-CRC).

(b) Includes 3 CRC patient samples of unknown stage.
COPYRIGHT 2008 American Association for Clinical Chemistry, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2008 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Molecular Diagnostics and Genetics
Author:Lofton-Day, Catherine; Model, Fabian; DeVos, Theo; Tetzner, Reimo; Distler, Juergen; Schuster, Matth
Publication:Clinical Chemistry
Date:Feb 1, 2008
Previous Article:Description and validation of a novel real-time RT-PCR enterovirus assay.
Next Article:High diagnostic accuracy of antigen microarray for sensitive detection of hepatitis C virus infection.

Related Articles
DNA bar code: screening methods of colorectal cancer.
Rosetta Genomics and Collaborators Unveil Colon Cancer Screening Diagnostic.
Two new blood tests make detection of gastrointestinal cancer simpler.
Detection of epigenetic changes in fecal DNA as a molecular screening test for colorectal cancer: a feasibility study.
15th Congress of the European Cancer Organization and the 34th European Society for Medical Oncology multidisciplinary congress: GI update.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters