[CANCER RESEARCH 64, 4442 4452, July 1, 2004]

Quantitative Multiplex Methylation-Specific PCR Assay for the Detection of Promoter Hypermethylation in Multiple Genes in Breast Cancer
Mary Jo Fackler,1 Megan McVeigh,1 Jyoti Mehrotra,1 Marissa A. Blum,1 Julie Lange,2 Amanda Lapides,3 Elizabeth Garrett,3 Pedram Argani,4 and Saraswati Sukumar1
1 Breast Cancer Program, Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins, and Departments of 2Surgery, 3Clinical Trials and Biometry, and 4Pathology, Johns Hopkins University School of Medicine, Baltimore, Maryland

ABSTRACT
If detected early, breast cancer is eminently curable. To detect breast cancer in samples with little cellularity, a high level of sensitivity is needed. Tumor-specific promoter hypermethylation has provided such a valuable tool for detection of cancer cells in biological samples. To accurately assess promoter hypermethylation for many genes simultaneously in small samples, we developed a novel method, quantitative multiplex-methylationspecific PCR (QM-MSP). QM-MSP is highly sensitive (1 in 104105 copies of DNA) and linear over 5 orders of magnitude. For RASSF1A, TWIST, Cyclin D2, and HIN1, we observed significant differences in both the degree (P < 0.003) and incidence (P < 0.02) of hypermethylation between normal and malignant breast tissues. Evaluation of the cumulative hypermethylation of the four genes within each sample revealed a high level of sensitivity (84%) and specificity (89%) of detection of methylation. We demonstrate the application of this technique for detecting hypermethylated RASSF1A, TWIST, Cyclin D2, HIN1, and RARB in 50 1000 epithelial cells collected from breast ducts during endoscopy or by lavage. Such an approach could be used in a variety of small samples derived from different tissues, with these or different biomarkers to enhance detection of malignancy.

INTRODUCTION Epigenetic alterations including hypermethylation of gene promoters are proving to be consistent and early events in neoplastic progression (1 4). Such alterations are thought to contribute to the neoplastic process by transcriptional silencing of tumor suppressor gene expression and by increasing the rate of genetic mutation (5, 6). DNA methylation is reversible because it does not alter the DNA sequence; however, it is heritable from cell to cell. Methylated genes can serve as biomarkers for early detection of cancer. Widschwendter and Jones (4) reviewed 40 genes whose expression is lost in breast cancer because of promoter hypermethylation, and we and others have studied hypermethylation of genes, including NES-1 (710), APC (1113), Cyclin D2 (14, 15), RARB (16 18), TWIST (19), RASSF1A (15, 20, 21), and HIN1 (22) in tissue (23) and ductal lavage fluids (19). Multigene promoter hypermethylation occurs commonly in tumors (24, 25). Because methylation changes often appear early in disease, detection of hypermethylated genes could identify tissues derived from patients with increased risk. Furthermore, the reversible nature of methylation offers the potential to revert aspects of the cancer phenotype with the appropriate therapy (26). Tumor DNA can be found in various body fluids, and these fluids can potentially serve as diagnostic material (19, 2729). Evaluation of

tumor DNA in these fluids requires methods that are specific as well as sensitive. For example, a PCR-based technique called methylationspecific PCR (MSP) can detect 1 copy of methylated DNA in 1000 unmethylated copies of genomic DNA (30). Palmisano et al. (31) and others (32, 33) have modified this approach to coamplify several genes simultaneously in a nested or multiplex MSP assay. This method has been used to establish the frequency of gene promoter hypermethylation among patients with pulmonary (31) and esophageal (33) carcinoma. However, the method cannot quantitatively measure the levels of gene methylation because the read-out is gelbased and qualitative (all or nothing), based on the visual detection of the presence or absence of a band on a gel. The issue gains importance because benign tissues often show low levels of methylation in several genes. To evaluate the degree of gene methylation within a single sample, quantitative MSP (Q-MSP) methods have been developed (15, 34 39). High and low levels of methylation may help to stratify different types or stages of carcinoma. The Q-MSP method is based on realtime PCR that uses fluorogenic probes to increase the assay specificity and the sensitivity; Q-MSP can detect one copy of the methylated marker gene among 10,000 unmethylated copies (36). The addition of a fluorogenic probe makes the technique more informative, quantitative, and suitable for clinical format. This technique is now becoming widely used (15, 34 39). However, analyses of multiple genes require additional quantities of template DNA. A dilemma is how best to distribute the available DNA to allow quantitative analyses of many different genes from precious small samples. We have developed a technique called quantitative multiplex-MSP (QM-MSP) to coamplify many genes from quantities of sample previously used for just one gene. This technique combines multiplex PCR and Q-MSP in such a way that a panel of five genes can be coamplified in tissues derived from different sources, including those from ductal lavage, endoscopy, and fine-needle aspirates, in which the amount of DNA is limiting, as well as in larger samples, such as formalin-fixed, paraffin-embedded sections of core biopsies. This technique can be used to define the extent of gene promoter hypermethylation in normal tissues on a gene-by-gene basis and provides the ability to discriminate between normal/benign and malignant tissues.

MATERIALS AND METHODS

Tissues and Cells. Paired primary invasive breast carcinomas and adjacent normal tissues (frozen tissue), paraffin-embedded normal breast (routine reduction mammoplasty), and primary breast carcinoma sections were obtained Received 10/23/03; revised 3/18/04; accepted 4/20/04. from the Surgical Pathology archives of The Johns Hopkins Hospital following Grant support: Grants to S. Sukamar from the Specialized Program of Research Excellence in Breast Cancer at Johns Hopkins (Grant P50CA88843), the AVON Founapproval by the Institutional Review Board. The percentage of epithelial cells dation, Susan G. Komen Foundation (Grant BCTR-2000-577), and the Department of in each tissue section ranged from 30 to 80% on each slide. Ductal cell samples Defense (Grant DAMD 17-01-1-0286). obtained by lavage of high-risk women and endoscopy of women with biopsyThe costs of publication of this article were defrayed in part by the payment of page proven breast cancer before surgery were provided after cytological diagnosis. charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Histopathological diagnosis of the resected specimen from patients undergoing Requests for reprints: Saraswati Sukumar, Johns Hopkins University School of endoscopy was obtained from Surgical Pathology. Human sperm was obtained Medicine, Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins, 1650 Orleans from a healthy volunteer. The MDA-MB231 (231) breast cancer cell line was Street, Room 410, Baltimore MD 21231-1000. Phone: (410) 614-2479; Fax: (410) 614obtained from American Type Culture Collection (Manassas, VA) and cultured 4073; E-mail: sukumsa@jhmi.edu. 4442

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Table 1 Sequences of multiplex primers used in round 1 of PCR Primer name Cyclin D2 Ext F Cyclin D2 Ext R RASSF1A Ext F(2) RASSF1A Ext R(2) TWIST Ext R(4) TWIST Ext F(3) RARB Ext F RARB Ext R(2) HIN1 Ext F(2) HIN1 Ext RSEQ ACTB Ext F ACTB Ext R ACTB Ext F Primer sequence (5 3 3 ) tattttttgtaaagatagttttgat tacaactttctaaaaaataaccc gttttatagtttttgtatttagg aactcaataaactcaaactccc cctcccaaaccattcaaaaac gagatgagatattatttattgtg gtaggagggtttattttttgtt aattacattttccaaacttactc gtttgttaagaggaagtttt ccgaaacatacaaaacaaaaccac tatataggttggggaagtttg tataaaaacataaaacctataacc tatataggttggggaagtttg Orientation Forward Reverse Forward Reverse Forward Reverse Forward Reverse Forward Reverse Forward Reverse Forward

as directed. Peripheral blood leukocytes were isolated from blood collected from normal volunteers after receipt of informed consent following review by the Johns Hopkins Institutional Review Board. DNA Extraction from Tissues and Peripheral Blood Cells. For each tissue, the lesion was confirmed on a H&E-stained section. For DNA extraction, one 5- m tissue section from the same block was deparaffinized in xylene (20 min), scraped from the slide, and extracted in 100 l of buffer [10 mM Tris (pH 8.0), 150 mM NaCl, 2 mM EDTA, 0.5% SDS] containing 40 g proteinase K for 16 h at 50C. For extraction of DNA from ductal cells, the number of cells on each Papanicolaou- or Diff-Quick-stained (American Scientific Products, McGraw, IL) cytospin preparation was counted, the coverslip was removed by treatment with xylene, and cells were scraped and transferred to 50 l of buffer [10 mM Tris (pH 8.0), 150 mM NaCl, 2 mM EDTA, 0.5% SDS] containing 40 g/ml proteinase K and 200 ng of salmon sperm carrier DNA. After proteinase K treatment, samples were heat-inactivated at 70C for 10 min and centrifuged at 16000 g for 10 min. Fifty l of the supernatant were used directly as a source of DNA for sodium bisulfite treatment. For leukocytes, frozen tissues, and 231 cells, DNA was extracted with phenol chloroform (40). Human sperm DNA (HSD) was isolated by use of the PUREGENE DNA Purification Kit (Gentra Systems, Minneapolis, MN) and

stored at 4C. One g of purified DNA was modified by sodium bisulfite treatment. Sodium Bisulfite Treatment of DNA. Tissue, control and cell line DNAs were treated with sodium bisulfite and analyzed by MSP as described by Herman et al. (30). This process converts nonmethylated cytosine residues to uracil, whereas methylated cytosines remain unchanged. Bisulfite-modified samples were aliquoted and stored at 80C. Probes and Primers. The sequences of primers used for multiplex and for amplifying unmethylated and methylated CpG islands by Q-MSP are shown in Tables 1 and 2. Gene-specific probes were obtained from Applied Biosystems (Foster City, CA), and primers were obtained from Invitrogen Corporation (Carlsbad, CA). For methylated Cyclin D2 and RASSF1A genes, the Q-MSP primers and probes were as described in Lehmann et al. (15). Methylationindependent Q-MSP primers and probes for -actin (ACTB) were as described by Eads et al. (36). All other sequences for methylation-dependent primers were designed in known regions of promoter hypermethylation in breast carcinoma; each Q-MSP primer set (forward, reverse, and probe) contained 712 CpG dinucleotides of the promoter sequence and numerous independent cytosine residues. QM-MSP. The QM-MSP procedure required two sequential PCR reactions (Fig. 1). In the first PCR reaction (the multiplex step), 1 l of sodium bisulfite-treated DNA was added to 24 l of reaction buffer [1.25 mM deoxynucleotide triphosphates, 16.6 mM (NH4)2SO4, 67 mM Tris (pH 8.8), 6.7 mM MgCl2, 10 mM -mercaptoethanol, 0.1% DMSO, and 2.55 units of Platinum Taq (Invitrogen)] containing 100 ng each of the forward and reverse primers. Conditions were 95C for 5 min, followed by 35 cycles of 95C for 30 s, 56C for 30 s, and 72C for 45 s, with a final extension cycle of 72C for 5 min. The PCR products were diluted up to 125 l with water and stored at 20C. For this reaction, the input DNA ranged from 50 ng (purified DNA) to 40 pg (for some ductal cell samples). For the second round (the Q-MSP step), 1 l of the diluted PCR product from reaction 1 was used directly or after further dilution of up to 1:104 (when 50 ng DNA was used in reaction 1). The diluted DNA was added to the Q-MSP reaction buffer containing 16.6 mM (NH4)2S04, 67.0 mM Tris (pH 8.8), 6.7 mM MgCl2, 10.0 mM -mercaptoethanol, 0.1% DMSO, 200 M deoxynucleotide

Table 2 Sequences of Q-MSPa primers used in round 2 PCR Primer name Cyclin D2 RT-FM Cyclin D2 RT-RM Cyclin D2 M Probe Cyclin D2 RT-FUM Cyclin D2 RT-RUM Cyclin D2 UM Probe RASSF1A RT-FM RASSF1A RT-RM RASSF1A M Probe RASSF1A RT-FUM RASSF1A RT-RUM RASSF1A UM Probe TWIST RT-FM TWIST RT-RM TWIST M-Probe TWIST RT-FUM (3) TWIST-RT-RUM (3) TWIST UM Probe RARB RT-FM RARB RT-RM RARB M Probe RARB RT-FUM RARB RT-RUM RARB UM Probe HIN1 RT-FM HIN1 RT-RM HIN1 M Probe HIN1-RT-FUM (2) HIN1 RT-RUM (2) HIN1UM Probe ACTB F ACTB R ACTB Probe Q-MSP primer sequence (5 3 3 ) tttgatttaaggatgcgttagagtacg actttctccctaaaaaccgactacg 6FAM-aatccgccaacacgatcgacccta-TAMRA ttaaggatgtgttagagtatgtg aaactttctccctaaaaaccaactacaat 6FAM-aatccaccaacacaatcaaccctaac-TAMRA gcgttgaagtcggggttc cccgtacttcgctaactttaaacg 6FAM-acaaacgcgaaccgaacgaaacca-TAMRA ggtgttgaagttggggtttg cccatacttcactaactttaaac 6FAM-ctaacaaacacaaaccaaacaaaacca-TAMRA gttagggttcgggggcgttgtt ccgtcgccttcctccgacgaa 6FAM-aaacgatttccttccccgccgaaa-TAMRA ggtttgggggtgttgtttgtatg cccacctcctaaccaccctcc 6FAM-aaacaatttccttccccaccaaaaca-TAMRA agaacgcgagcgattcgagtag tacaaaaaaccttccgaatacgtt 6FAM-atcctaccccgacgatacccaaac-TAMRA ttgagaatgtgagtgatttgagtag ttacaaaaaaccttccaaatacattc 6FAM-aaatcctaccccaacaatacccaaac-TAMRA tagggaagggggtacgggttt cgctcacgaccgtaccctaa 6FAM-acttcctactacgaccgacgaacc-TAMRA aagtttttgaggt ttgggtaggga accaacctcacccacactccta 6FAM-caacttcctactacaaccaacaaacc-TAMRA tggtgatggaggaggtttagtaagt aaccaataaaacctactcctcccttaa 6FAM-accaccacccaacacacaataacaaacaca-TAMRA Orientation Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Forward Reverse Reverse Status M M M U U U M M M U U U M M M U U U M M M U U U M M M U U U Indep Indep Indep

a Q-MSP, quantitative methylation-specific PCR; M, methylated; 6FAM, 6-carboxyfluorescein; TAMRA, 6-carboxytetramethylrhodamine; U, unmethylated; Indep, methylationindependent.

4443

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Fig. 1. Scheme for the quantitative multiplex methylation-specific PCR. In reaction 1 (RXN 1), a cocktail of gene-specific primer pairs is used to coamplify DNA for multiple genes independent of their DNA methylation status. In reaction 2 (RXN 2) quantitative real-time PCR is performed with gene-specific primers, using the DNA template derived from the end product of RXN 1 (diluted 1:51:104). DNA was analyzed in separate wells, using methylation status-specific [U, unmethylated; M, methylated) primers (forward, reverse, and probe, conjugated with the 6-carboxyfluorescein (FAM) label and 6-carboxytetramethylrhodamine (TAMRA) quencher]. The oligonucleotide probe is progressively degraded with each cycle of PCR by the 5 3 nuclease activity of the DNA polymerase; therefore, the fluorescence signal generated by 6-carboxyfluorescein is directly proportional to the extent of DNA amplification. % M 100 [no. of copies of M/no. of copies of (unmethylated methylated)], as determined by the absolute quantification method computed against a standard curve. NaBi, sodium bisulfite; ext, external primer; diln, dilution.

triphosphates, 1.25 units of Platinum DNA Taq Polymerase (Invitrogen), and 1 ROX (Invitrogen) in a final volume of 25 l. Six hundred nM each of two primers (forward and reverse) and 200 nM labeled probe (Applied Biosystems) were also present. The reaction was carried out in a 96-well reaction plate in an ABI Prism 7900HT Sequence Detector (Applied Biosystems). The reaction conditions were as follows: 95C for 5 min, followed by 35 cycles of 95C for 15 s and 60 65C (depending on the primer set) for 1 min, with a 10-min extension at 72C. For each gene included in the reaction plate, the following were used to create standard curves and to provide controls: (a) serially diluted stock multiplexed HSD/231 DNA (described below, to establish a standard curve); (b) 40,000 copy (40 K) standards; (c) no-template control; and (d) a known DNA (1% M control) to ensure consistency among runs. In addition, 100% methylated DNA (231 cell DNA), 0% methylated DNA (HSD), and a sample lacking template DNA from the first PCR reaction (diluted 1:5) were present as controls. All of the above samples were analyzed with primer sets for both methylated and unmethylated DNA. Because the 1% M control values have remained constant over a period of 1 year (data not shown) we conclude that the assay is stable and reproducible. Preparation of Standards. A stock of multiplexed DNA was prepared as follows: PCR was performed in a reaction that contained all first-round gene primer pairs (Table 1) as well as a mixture of 50 ng each of sodium bisulfitetreated genomic 231 and HSD. Serial dilutions of this stock DNA were used to establish a standard curve in the real-time PCR reaction. To do this, the cycle threshold (CT; the cycle in which the signal exceeds the background) of each dilution was determined during the Q-MSP reaction and then plotted against the dilution to generate a line for the standard curve. For each reaction plate, the standards were diluted from the same stock stored frozen at 80C for all assays, and new dilutions were made each time. All assays had a correlation coefficient of the standard curve of 0.99 or higher and a slope of approximately 3.33, indicating 2-fold increases in PCR product per cycle in the linear phase of the quantitative PCR reaction. Copy Number Standard. For the preparation of this standard, unmethylated or methylated genomic DNAs were amplified for each gene separately by use of a gene-specific pair of external primers and 50 ng of sodium bisulfitetreated genomic DNA derived from either MDA-MB231 (100% methylated) or HSD (100% unmethylated). A single band was observed by gel electro-

phoresis. The reaction products were then purified with the Qiaquick PCR purification kit (Qiagen Inc., Valencia, CA) and eluted in 100 l of water. The eluate was quantitated by use of a NanoDrop spectrophotometer (NanoDrop Technologies, Montchanin, DE), and the DNA concentration ( g/ l) was determined (A260). The molecular weight ( g/ mol) of the PCR product was calculated by use of Biopolymer Calculator v4.1.1 (C. R. Palmer).5 The concentration of each gene template control was adjusted to 3 1010 copies/ l in 1 mg/ml salmon sperm carrier DNA, and then a cocktail of unmethylated and methylated template control was immediately prepared that contained 4 106 copies/ l each of the genes in 1 mg/ml salmon sperm DNA. This stock was stored at 80C. For each reaction plate the stock was diluted 100-fold to 40,000 copies/well. We used this known quantity of standard (40,000 copies/ well, denoted 40K control), prepared as described above, to transform the standard curve to represent copy number. To accomplish this the CT of the 40 K control was determined during the Q-MSP reaction and plotted on the line obtained for the standard curve. The copy number for each dilution was then back calculated, based on where the 40K CT intersected the standard curve. Sample 40K had approximately equal amounts of unmethylated and methylated DNA for each of the five genes (RASSF1A, TWIST, Cyclin D2, HIN1, and RARB) along with carrier salmon sperm DNA (10 g/ml). Calculation of Percentage of Methylation. The relative amount of methylation in each unknown sample was calculated as % M 100 [no. of copies of methylated DNA/(no. of copies of methylated unmethylated DNA)]. The sum of unmethylated plus methylated DNA (U M) was used as an approximation of the total number of copies present of a target gene. To determine the number of copies of methylated and unmethylated DNA, we mixed sample DNA with Q-MSP reaction buffer after the multiplex reaction, assayed the mixture with methylated primers and unmethylated primers (in separate wells) in the Q-MSP reaction, and then determined the CT was for each. Using the ABI Prism SDS 2.0 software supplied by Applied Biosystems (Foster City, CA) with the 7900 HT Sequence Detector, we extrapolated the number of copies of methylated and unmethylated DNA from the respective standard curves, using the sample CT and applying the absolute quantification method according to the manufacturers directions. Only values falling within the range covered by the standard curve (usually 100 10,000,000 copies) were accepted. Direct Q-MSP of Genomic DNA. For direct Q-MSP, standard curves were prepared using 10 pg, 100 pg, 1 ng, and 10 ng total genomic 231 DNA (fully methylated) or HSD (fully unmethylated), according to the absolute quantitation method described by Applied Biosystems in the 7900 HT Sequence Detector manual. The concentrations of methylated and unmethylated DNA were extrapolated from these curves, and the percentage of methylation was calculated as % M 100 [ng methylated gene A/(ng methylated gene A unmethylated gene A)], where total target gene DNA was taken as the sum of U M. For purposes of comparing these results with methods that use -actin (ACTB) as a reference DNA, we also computed the percentage of methylation using two other formulas. We calculated % M 100 (ng methylated gene A in tumor/ng ACTB gene in tumor), or essentially as described by Trinh et al. (37), who calculated the percentage of methylation as % M 100 [(ng methylated gene A in tumor/ng ACTB in tumor)/(ng methylated gene A in 231 controls/ng ACTB in 231 controls)]. Statistical Analysis and Graphical Representation of Data. Statistical analyses and plotting of data were performed with GraphPad Prism (GraphPad Software Inc., San Diego, CA) and Stata 7.0 (Stata Corporation, College Station, TX). P values 0.05 were considered significant, and all tests were two-tailed. The nonparametric MannWhitney test was used to test whether the samples were from identical distributions, indicating that their medians were equal. Sample means were compared by use of the unpaired t test, assuming unequal variances (Welchs correction). For testing of means, data were transformed as a function of Lne(%M 1) where stated to fulfill the assumption of normality. The Fishers exact test was used to test whether the differences between the incidence of positivity for methylation in tumor and nontumor samples were significant.

http://paris.chem.yale.edu/extinct.html.

4444

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

RESULTS Validation of the Two-Step QM-MSP Assay. Multiplex PCR was accomplished by performing two sequential PCR reactions (Fig. 1). In the first PCR reaction, a cocktail of gene-specific primers was used to coamplify RASSF1A, TWIST, Cyclin D2, HIN1, and RARB (Table 1). These external primer pairs were complementary to the sequences flanking the CpG island that was to be assayed in the second PCR reaction. External primers were selected to exclude CG dinucleotides, thereby rendering DNA amplification independent of the methylation status. In the second reaction, quantitative analysis of methylated and unmethylated DNA for each gene was performed separately with the primer pairs and probes shown in Table 2. Unmethylated and Methylated Primers are Equally Efficient in Amplifying Sodium Bisulfite-Converted DNA. Primer sets for specifically amplifying methylated (M) or unmethylated (U) DNA (Table 2) were designed for comparable performance; to confirm this we plotted, the CT (CT M CT U) as a function of sample dilution over a wide range of dilutions (10 310 8) of the standard stock HSD/231 DNA. Analyses were performed as shown for RASSF1A (Fig. 2). The CT was approximately the same for all dilutions, as shown by the horizontal nature of the line, indicating that the primer sets were equally efficient over 5 logs of template quantities (Fig. 2A). In addition, for both unmethylated and methylated DNA, the slopes of the standard curves were approximately 3.33, which is reflective of

Table 3 Determining sensitivity of quantitative multiplexed methylation-specific PCR Methylationa (%) RASSF1A Copies Mb and U templatec,d 1000/100,000 100/100,000 10/100,000 1/100,000 0.1/100,000 HSD (100% U control) 231 (100% M control) Water Copies M and Uc,e 300/29,700 30/2,970 3/297 0.9220 0.0602 0.0078 0.0072 0.0000 0.0000 99.9981 0.0000 0.7823 0.7195 0.2390 TWIST 2.7604 0.2482 0.0272 0.0012 0.0000 0.0003 100.0000 0.0000 1.2190 0.9376 0.1451 HIN1 4.8024 0.6273 0.0024 0.0000 0.0000 0.0000 100.0000 0.0000 Cyclin D2 1.5760 0.1525 0.0025 0.0000 0.0000 0.0000 100.0000 0.0000

a The percentages of methylation of RASSF1A, TWIST, HIN1, and Cyclin D2 from the quantitative multiplexed methylation-specific PCR assay are shown. The assay sensitivity is 110 in 100,000 for methylated DNA, and it detects as few as 13 copies of methylated DNA. b M, methylated; U, unmethylated; HSD, human sperm DNA; 231, MDA-MB231. c Purified methylated and unmethylated stock DNA template were mixed in the proportions indicated. d Unmethylated DNA was kept constant as the amount of methylated DNA was decreased. e A 1% M control was kept constant as the total quantity of DNA decreased.

Fig. 2. Efficiency of quantitative multiplex methylation-specific PCR. For verification of the comparable function and efficient performance of the unmethylated (U) and methylated (M) primers, QM-MSP was performed using RASSF1A U and M primer sets to amplify serially diluted (10 310 8) fragments of multiplexed standard stock human sperm DNA/MDA-MB231 DNA. The cycle threshold (CT) was determined for each dilution of DNA. A, plot of CT ( CT CT M CT U) versus dilution. Nearly identical CT values for each DNA dilution indicate uniform primer performance over 5 logs. B, standard curve plots (CT versus quantity) of serially diluted DNA. The slope of 3.33 reflects a 2-fold amplification of DNA per cycle (high efficiency). The correlation coefficient (R2) of 0.999 shows the high degree of linearity over the entire range.

a 2-fold increase in PCR product per cycle during the linear phase of real-time PCR. Finally, the correlation coefficient (R2) of 0.999 provided evidence of linearity over the entire range of template concentration (Fig. 2B). Similar results were obtained for each of the other genes in this study (data not shown). Specificity and Sensitivity. To assess the sensitivity and specificity of the QM-MSP method, we performed a mixing experiment, using column-purified, PCR-amplified fragments of sodium bisulfitemodified DNA as template (Table 3). The amount of unmethylated DNA was kept constant (100,000 copies/well), and the amount of methylated DNA was decreased (1000 0.1 copy/well). While the general efficiency of real-time PCR is known to fall off at 100 copies, it was expected that some level of methylation would be detected at concentrations 100 copies, although probably not in a linear manner. For RASSF1A and TWIST, methylation was detected at 1 methylated copy in 100,000 unmethylated copies, and for Cyclin D2 and HIN1, methylation was detected at 10 in 100,000 copies. Therefore, the overall sensitivity of the method was 110 in 100,000. The highly specific performance of the methylated primers was demonstrated by the use of HSD control (100% unmethylated DNA) and the 0.1/100,000 sample (diluted to 1 copy of methylated in the presence of 100,000 copies of unmethylated DNA per well), both of which showed 0% M in the Q-MSP reaction (Table 3; Fig. 3). Likewise, no unmethylated signal (0% U) was detected in 231 methylated DNA control (100% methylated DNA). It was expected that the 1000/100,000 sample would be read as 1% methylated (1000 copies of methylated and 100,000 copies of unmethylated). That some samples were slightly higher (e.g., 4.8% for HIN1 and 2.8% for TWIST) probably reflects difficulty in accurately diluting samples from a starting concentration of 3 1010 copies/ l (diluted over 510 logs). To characterize the behavior of the QM-MSP assay below the lower limit of linearity (specifically, whether a 1% M template produces a 1% M assay result as the total quantity of DNA template diminishes), we serially diluted a 1% standard and tested samples beginning at a template quantity estimated to contain 300 copies of methylated DNA to 29,700 unmethylated copies (300/29,700 per well), using column-purified DNA as a template (Table 3). The lowest quantity tested contained 3 copies of methylated DNA to 297 copies of unmethylated DNA (3/297). The results for RASSF1A and TWIST (Table 3) revealed that at ratios of 300/29,700 and 30/2,970 total copies, the

4445

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Fig. 3. Specificity of quantitative multiplex methylation-specific PCR. Shown are amplification plots of multiplexed human sperm DNA (HSD) or MDA-MB231 template DNA tested with unmethylated (U) and methylated (M) primers for RASSF1A. The unmethylated and methylated reactions were 100% specific because U primers did not cross-react with MDA-MB231 DNA (100% methylated) and M primers did not crossreact with human sperm DNA (100% unmethylated).

assay result (% M) was 1% for each. Methylation was still detectable at the lowest ratio template quantity tested, at 3 copies of methylated DNA, consistent with the previous experiment (1 copy of methylated DNA was detected in 100,000 copies unmethylated DNA). We found a bias toward underreporting of the % M below 30 copies of methylated DNA, probably reflecting the relative lack of efficiency of the methylated reaction compared with the unmethylated reaction that contained nearly 100-fold more copies of the gene (Table 3). This result was predicted because linearity is generally known to be lost below 100 copies of DNA template in real-time PCR. Genomic DNA is a more challenging template than PCR-amplified DNA because breakage of genomic DNA is known to occur in the process of sodium bisulfite conversion. To evaluate the sensitivity of the QM-MSP method for detecting methylated alleles when genomic DNA was used, we mixed 40 pg of methylated DNA ( 13 copies derived from 231 cell DNA) with 600 60000 pg of unmethylated genomic DNA ( 200 20,000 copies of HSD), using the conversion estimate of 3 pg/copy of genomic DNA. Our data showed that 40 pg of methylated RASSF1A genomic DNA was easily detected even in the presence of a 1500-fold excess of unmethylated DNA (Fig. 4). To ensure specific amplification of sodium bisulfite-modified DNA (but not nonconverted DNA; Ref. 41), each primer set (forward, reverse, and probe) covered 712 CpG dinucleotides pairs when used together and also covered numerous independent (presumably unmethylated) cytosines. Controls were included during the development of QM-MSP to ensure that no reactivity with nonconverted or sodium bisulfite-treated carrier (salmon sperm) DNA was detectable by QM-MSP (data not shown). Lastly, in each QM-MSP reaction, 100% unmethylated (HSD) and 100% methylated (231) sodium bisulfite-modified DNAs were analyzed as controls, and no background signals were detectable with use of real-time primers. Comparison of QM-MSP with Direct Q-MSP. The QM-MSP method is sensitive and specific (Table 3; Fig. 3). There are two basic differences between most conventional Q-MSP methods and QMMSP: (a) QM-MSP includes an additional PCR step (multiplex), which could lead to greater sensitivity but lower specificity than Q-MSP; and (b) in contrast to the standard use of glyceraldehyde-3-

phosphate dehydrogenase (GAPDH) or ACTB reference DNA, QMMSP uses the sum of methylated and unmethylated DNA (U M) of the same gene for determining total gene DNA present in the sample. The use of the (U M) formula ignores the possible contribution to total DNA from partially methylated DNA and thus could potentially overestimate the percentage of methylation in each sample. We tested the impact of these differences experimentally. QM-MSP versus Direct Q-MSP Using (U M) as Total DNA. It is possible that performing a two-step multiplex PCR method could yield results that differed from those obtained with a direct one-step PCR method because of the addition of the multiplex step. We performed QM-MSP and direct Q-MSP assays on a panel of five tumor DNAs and calculated the percentage of methylation by the (U M) method to estimate total DNA. With few exceptions, there was excellent concordance between the percentage of methylation values obtained for the RASSF1A, TWIST, HIN1, or Cyclin D2 genes (Table 4). The QM-MSP readout was much more robust (as a result of the preamplification of DNA), usually appearing around cycles 1225 compared with the readout by Q-MSP, in which the CT signal appeared around cycles 2737 (data not shown). Comparison of Percentage Methylation Using Actin (ACTB) versus (U M) Target Gene DNA to Approximate Total DNA. We first addressed the question of whether the ACTB reference method and (U M) methods yielded similar data in the setting of Q-MSP, as described in the Materials and Methods. We found that, with few exceptions, there was excellent concordance between the % M values when we used either of the ACTB formulas or the (U M) formula to estimate total DNA in Q-MSP (data not shown). We next tested the same concept in the QM-MSP setting. For the same DNA samples, discordance was observed for % M values between (U M) or ACTB formulas (data not shown). The rate of amplification of ACTB compared with those of the other genes in the mixture was not predictable in the multiplex reaction. For example, values obtained with ACTB as the reference gene (as described in the Materials and Methods) showed concordance for some genes, very high values for others, and lower methylation values in 50% of the samples (data not shown). On the other hand, with the (U M) method, QM-MSP values were reproducible across quadruplicate assays. For example, different tumors gave % M values for RASSF1A of 42.6 3.5, 42 6.9, 71 1.5, 13 6.0, 27 7.5, 81 1.5, and

Fig. 4. Sensitive detection of hypermethylated genomic RASSF1A DNA. Genomic human sperm DNA (HSD) and MDA-MB231 (231) DNA were mixed as indicated before sodium bisulfite treatment (top panel, 600 pg of HSD and 40 pg of 231 DNA; bottom panel, 60 ng of HSD and 40 pg of 231 DNA). Quantitative multiplex methylation-specific PCR was performed. Shown are the amplification plots of each mixture of template. The plots show that 40 pg of genomic DNA is easily detectable even in the presence of a 1500-fold excess of unmethylated DNA.

4446

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Table 4 Comparison between Q-MSPa and QM-MSP The percentage of methylation (% M) was calculated as: (the number of copies of methylated DNA divided by the number of copies of unmethylated methylated DNA) 100, using absolute quantitation. The positive control for unmethylated DNA was HSD, and for methylated DNA was 231. %M Q-MSP RASSF1A 1 2 3 4 5 HSD 231 Water TWIST 1 2 3 4 5 HSD 231 Water HIN1 1 2 3 4 5 HSD 231 Water Cyclin D2 1 2 3 4 5 HSD 231 Water 95 80 39 12 28 0 100 0 96 0 54 1 0 0 100 0 95 81 82 41 47 0 100 0 47 0 1 19 2 0 100 0 QM-MSP 99 81 41 21 34 0 100 0 70 0 26 3 5 0 100 0 82 74 48 16 43 0 100 0 25 0 0 22 0 0 100 0

a Q-MSP, quantitative methylation-specific PCR; QM-MSP, quantitative multiplexed methylation-specific PCR; HSD, human sperm DNA.

41 2.0%, whereas DNA samples that were negative for methylation yielded % M values of 0, 0, 0.4 1.0, and 1.5 0.5%. In QM-MSP, the sum of the unmethylated and methylated alleles of the same gene appear to serve as a reliable internal control for integrity, copy number, and method efficiency. The (U M) formula was therefore used for the rest of the study. Quantitation of Methylation in Invasive Breast Carcinoma and Comparison with Normal Breast Tissue. We analyzed test sets of DNA from 18 normal mammoplasty and 21 tumors specimens by

QM-MSP for gene promoter hypermethylation of RASSF1A, TWIST, Cyclin D2, and HIN1 (Table 5; Fig. 5). For RASSF1A alone, the normal breast test set was further expanded to 28 samples based on a previous report of a higher incidence of hypermethylation in benign breast tissue (15). Occasionally, PCR amplification failed for some genes within a test sample, presumably because of the fragile nature of archival DNA. RASSF1A hypermethylation ranged from 0 to 71% (mean, 18.5%) in carcinoma and from 0 to 56% (mean, 2.6%) in normal tissues (P 0.0001), TWIST hypermethylation ranged from 0 to 72% (mean, 21.1%) in carcinomas and from 0 to 1.6% (mean, 0.11%) in normal tissues (P 0.0001), Cyclin D2 hypermethylation ranged from 0 to 44.5% (mean, 5.0%) in carcinomas and from 0 to 0.2% (mean, 0.02%) in normal tissues (P 0.02), and HIN1 hypermethylation ranged from 0 to 82.2% (mean, 24.5%) in carcinomas and from 0 to 18% (mean, 2.3%) in normal tissues (P 0.003). When we used the MannWhitney test on untransformed data, the differences in the medians were highly significant for all genes tested: RASSF1A (P 0.0001), TWIST (P 0.001), Cyclin D2 (P 0.0009), and HIN1 (P 0.003). We also analyzed normal leukocyte DNA and found that methylation in these samples derived from the buffy coat was extremely low or undetectable (n 25): median leukocyte methylation was 0% for RASSF1A, 0.06% for TWIST, 0% for Cyclin D2, 0.005% for HIN1, and 0.25% for RAR . Because DNA from mammoplasty specimens and leukocytes is largely unmethylated relative to tumor samples, it is likely that the methylation signals observed in the tumors (Fig. 5) are derived largely from the carcinoma cells rather than normal ducts, stroma, and/or infiltrating leukocytes. We chose to establish a cutoff (% M) for each gene at approximately the 10th percentile of the population, such that 90% of normal breast tissues would be at or below the cutoff (we allowed 8590% for HIN1; see above). Using cutoffs of 2% M for RASSF1A and HIN1, 0.5% M for TWIST, and 0.2% M for Cyclin D2 in normal tissues, we considered values above the cutoffs positive for hypermethylation. Among carcinomas, 68% were positive for RASSF1A, 67% for TWIST, 57% for Cyclin D2, and 57% for HIN1. By comparison, 714% of normal mammoplasty samples were positive for RASSF1A, TWIST, Cyclin D2, and HIN1. Some samples had low-level methylation that was below the cutoff. Using these cutoffs, we observed a significant difference in the incidence of positivity between carcinoma and normal tissues (RASSF1A, P 0.00002; TWIST, P 0.0002; Cyclin D2, P 0.002; and HIN1, P 0.02, Fishers exact). Cumulative Gene Promoter Hypermethylation Scores in Primary Breast Cancer. To calculate the total amount of gene promoter hypermethylation as determined by QM-MSP, we used the sum of all

Table 5 Quantitative multiplexed methylation-specific PCR analysis of normal breast and breast cancer DNA Methylation (%) Range Normal breast RASSF1A TWIST HIN1 Cyclin D2 Breast carcinoma RASSF1A TWIST HIN1 Cyclin D2
a b

Median 0c 0d 0e 0f 7.0c 5.0d 9.9e 0.26f

Mean 2.6 0.11 2.3 0.019 18.5 21.1 24.5 5.0

SE 2.0 0.09 1.5 0.014 4.7 5.5 6.1 2.5

Lower 95% confidence limita (0) (0) (0) (0) 8.7 9.6 11.8 (0)

Upper 95% confidence limit 6.7 0.29 5.5 0.48 28.2 32.6 37.3 10.3

Positive for methylation,b n (%) 2/28 (7) 1/18 (6) 2/14 (14) 1/16 (7) 13/19 (68) 14/21 (67) 12/21 (57) 12/21 (57)

056 01.6 018 00.2 071 072 082 044

Values in parentheses indicate that the value is less than zero. Based on percentage of methylation cutoffs of 2% for RASSF1A and HIN1, 0.5% for TWIST, and 0.2% for Cyclin D2. c P 0.0001 for RASSF1A. d P 0.001 for TWIST. e P 0.003 for HIN1. f P 0.0009 for Cyclin D2.

4447

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Fig. 5. Comparison of gene promoter hypermethylation in normal and malignant breast tissues. Quantitative multiplex methylation-specific PCR was performed on DNA from formalin-fixed, paraffin-embedded normal tissues, derived from patients undergoing reduction mammoplasty, for quantitation of RASSF1A (n 28), TWIST (n 18), Cyclin D2 (n 16), and HIN1 (n 14) gene promoter hypermethylation. In parallel, DNA derived from individuals with invasive ductal carcinoma was tested for RASSF1A (n 19), TWIST (n 21), Cyclin D2 (n 21), and HIN1 (n 21). Displayed is the Lne(% M 1). The median of each group is indicated by a horizontal bar. Percentage methylation values for normal and carcinoma DNA were significantly different for all genes. Leukocyte (Leuk; n 25) DNA derived from normal individuals showed low to no levels of methylation.

% M within the panel of genes to provide an overall cumulative score for each sample. In Fig. 6A, this is represented graphically relative to 231 DNA, which is 100% methylated for RASSF1A, TWIST, Cyclin D2, and HIN1; therefore, this control DNA had a relative score of 400. The cumulative methylation profiles of 9 normal mammoplasty samples were compared with those of 19 invasive carcinomas (Table 6; Fig. 6) in a subgroup of our test set in which results for all four markers were available. Normal tissues ranged from 0 to 18 units, and carcinomas ranged from 1 to 248 units. Among the nine normal tissues tested for four genes (36 values) the mean cumulative score was 2.61 2.05 (median 0; Fig. 6B). Among the 19 carcinomas tested for four genes (76 values), the mean cumulative score was 72.8 15.03 units (median 74) out of a possible 400 units (see

above; Table 6; Fig. 6). The difference in log-transformed means between normal and malignant breast tissue was highly significant (P 0.0001, unpaired t test with Welchs correction). We explored the use of 4.7 units as a cumulative score cutoff value. This was based on the individual gene cutoffs (2% each for RASSF1A and HIN1, 0.5% for TWIST, and 0.2% for Cyclin D2 4.7%). When we used this cutoff, 84% (16 of 19) of carcinoma samples were positive. Although 3 of 19 carcinomas fell below the cutoff, all of the negative carcinomas were methylated at low levels for one or more genes in this panel (Fig. 6A). Among normal samples, 89% (8 of 9) were negative. Thus, in this group of 28 samples (9 normal and 19 carcinoma), the sensitivity for detection of carcinoma was 0.84 and specificity was 0.89; the overall accuracy was 0.86 (24 of 28).

4448

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

Fig. 6. Cumulative promoter hypermethylation of RASSF1A, TWIST, Cyclin D2, and HIN1 in normal and malignant breast tissues. A, scores of cumulative methylation in normal mammoplasty and carcinoma. In a subgroup of samples from Fig. 5 for which results were available for all four genes in the panel, normal (Mam; n 9) and carcinoma (Tumor; n 19) samples were scored for cumulative methylation by adding the percentage of methylation for all four genes within each sample. A maximum of 400 relative methylation units was possible [e.g., MDA-MB231 (231 control) DNA is 100% methylated for each of the four genes]. Results are plotted as stacked bar graphs. The bar height reflects total cumulative methylation, whereas the segments corresponding to the relative amounts of methylation of each gene are indicated. B, mean cumulative methylation in normal mammoplasty (Mam) versus carcinoma (Tumor). Plotted is the mean ( SE; bars) amount of cumulative methylation.

Comparison of Paired Carcinoma and Adjacent Normal Breast Epithelium. In an independent experiment, we examined six pairs of carcinoma and adjacent tissue from the surgical margins that were histologically normal to determine the cumulative amount of gene promoter hypermethylation in RASSF1A, TWIST, Cyclin D2, and HIN1 (Table 6; Fig. 7). The cumulative methylation ranged from 2 to 29 units within adjacent normal tissues and from 5 to 258 units within carcinoma tissues, out of a possible 400 units (Fig. 7B). When we used the cutoff established for cumulative normal in mammoplasty samples ( 4.7 units; see above), all six carcinomas were positive. The adjacent normal tissues were also positive in four of six individuals. Although the cumulative methylation levels within carcinoma-adjacent, histopathologically normal tissues were significantly lower than in the nearby carcinoma (P 0.03, MannWhitney), they had a significantly higher levels of methylation than normal mammoplasty samples (P 0.01, Mann-Whitney; Table 6; Fig. 7B). Detection of Multigene Promoter Hypermethylation in Breast Ductal Cells. To test the applicability of this method to investigate small clinical samples, we performed QM-MSP to assess RASSF1A, TWIST, Cyclin D2, HIN1, and RARB gene promoter hypermethylation in ductal cells derived from lavage of high-risk women (n 7) or cells in the irrigation fluid obtained during breast endoscopy in four patients with biopsy-proven cancer, before resection of the lesion. The total number of epithelial cells present in the samples was estimated

to be 50 1000 cells. The level of gene promoter methylation was quantitated for each sample, and the cumulative promoter methylation profile was established. Six of seven samples from women with a high risk of developing breast cancer had no detectable hypermethylation in the five genes tested, whereas one of seven displayed low-level methylation in RASSF1A. Cytological diagnosis of all of the specimens in this group was benign. Of interest, cells obtained during endoscopy from the two women with invasive carcinoma had highlevel multigene promoter hypermethylation (samples 10 and 11 in Table 7 and Fig. 8). The samples from the two women with ductal carcinoma in situ displayed benign cytology and lacked detectable promoter hypermethylation (samples 8 and 9). DISCUSSION Cytological analysis of breast ductal cells could aid in assessing risk and in the early detection of breast cancer, particularly in young women. To provide a robust and objective method to analyze these cells, in this report we describe a technique called QM-MSP. We show that QM-MSP evaluates the level and incidence of promoter hypermethylation of several genes in samples where DNA is limiting. Scoring the cumulative methylation of these genes within a sample provided a high level of sensitivity and specificity of detection of cancer. We tested this method in clinical settings where samples yield

Table 6 Cumulative promoter hypermethylation in normal, adjacent normal and malignant breast tissues Methylation (units)a Tissue Range Median Mean SE 2.05e 15.03e 4.07 39.9 Lower 95% confidence limitb (0) 41.3 1.2 26.5 Upper 95% confidence limit 7.35 104.4 22.1 231.8 Positive for methylation,c n (%) 1/9 (11) 16/19 (84) 4/6 (67) 6/6 (100) n 9 19 6 6

Normal mammoplasty and malignant breast tissues 2.61 Normal 018 0d 72.8 Carcinoma 0248 74d Paired malignant breast tissues and their adjacent normal breast tissues 11.7 Adjacent normal 229 9f 129.2 Carcinoma 5258 133f
a b

Relative units of methylation is the sum of percentage methylation for each of four genes in the panel. Value in parentheses indicates that the value is less than zero. c Based on a cutoff of 4.7 units. d P 0.0001. e P 0.0002. f P 0.03.

4449

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

a study of 103 cases of breast cancer we recently reported that 100% of cases of invasive carcinoma and 95% cases of ductal carcinoma in situ were hypermethylated for one or more gene promoters in a panel of these five genes (23). In fact, the vast majority of carcinomas (80%) were hypermethylated for two or more of these five genes. From our work we developed support for the concept that profiling the cumulative methylation of multiple genes would serve to better distinguish benign from malignant tissues and would provide a more powerful approach than characterizing the status of only one gene marker. Studies in our laboratory have also demonstrated the feasibility of assessing gene promoter hypermethylation in ductal lavage samples (19). We found TWIST, Cyclin D2, or RARB gene promoter hypermethylation in cells derived from patients with ductal carcinoma. Because the ductal cell samples most often contained 50 1000 epithelial

Fig. 7. Cumulative promoter hypermethylation of RASSF1A, TWIST, Cyclin D2, and HIN1 in adjacent normal and malignant breast tissues. A, cumulative methylation in adjacent normal and carcinoma tissues. In an independent experiment, paired samples (n 6) of adjacent normal and carcinoma tissue were quantitated by quantitative multiplex methylation-specific PCR, and the extent of cumulative methylation of the gene panel was determined, as described in the legend for Fig. 6. Shown are results of each sample as well as positive control MDA-MB231 DNA (Control; 400 units). For purposes of comparison, also shown at the far left is the average cumulative methylation of normal mammoplasty (n 9) samples (Mam) from Fig. 6B. Raw scores are shown in the table. B, mean cumulative methylation in adjacent normal (Adj Norm) versus carcinoma (Tumor) compared with normal mammoplasty (Mam). Differences between normal mammoplasty samples (median 0) and adjacent normal tissue samples (median 9 units) were significant (P 0.01, MannWhitney test). Plotted is the mean ( SE; bars) amount of cumulative methylation found above in adjacent normal tissues (mean, 11.7 4.07) and the nearby carcinoma (mean, 129 39.9) compared with normal mammoplasty (mean, 2.61 2.05; as shown in 6B). Actual percentage of methylation values are listed in the table for tumor (T), adjacent normal (N), and MDA-MB231 positive control (Ctrl).

very limited amounts of DNA, such as those from ductal lavage and ductal irrigation fluid collected during endoscopy. In the process, we more clearly defined the extent of gene promoter hypermethylation in normal breast cells and showed the feasibility for adapting this method to clinical testing. Although we applied this technique to breast tissues, it can be used to evaluate gene promoter hypermethylation in a wide variety of tissues. The expression of 40 genes is reportedly lost in breast cancer because of promoter hypermethylation (4). Recent work in our laboratory and by others has shown that some of the genes most frequently hypermethylated (30 90%) in breast carcinomas, but not in normal breast epithelium or circulating blood cells, are Cyclin D2 (14, 15), RARB (16 18), TWIST (19), RASSF1A (15, 20, 21), and HIN1 (22). In

Fig. 8. Cumulative promoter hypermethylation in ductal fluid. RASSF1A, TWIST, Cyclin D2, HIN1, and RARB genes were coamplified, and promoter hypermethylation levels were quantitated by quantitative multiplex methylation-specific PCR in ductal cells derived from lavage of high-risk women (n 7) or from cells present in the irrigation fluid obtained during breast endoscopy in patients with biopsy-proven cancer (n 4), before resection of the lesion. Inadeq, inadequate sample available; Mk Atyp, markedly atypical.

Table 7 Quantitative multiplexed methylation-specific PCR analysis of ductal breast cells Methylation (%) IDa RASSF1A TWIST Cyclin D2 0 0 0 0 0 0 0 0 0 12 4 HIN1 0 0 0 0 0 0 0 0 0 75 65 RARB 0 0 0 0 0 0 0 0 0 0 89 Cytology Benign Benign Benign Benign Benign Benign Benign Benign Benign Inadequate Markedly atypical Mammography Normal Normal Normal Normal Normal Normal Normal DCIS DCIS Invasive carcinoma Invasive carcinoma Histopathology

Ductal lavage cells from high-risk womenb 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0.4 0 Ductal lavage cells from women with carcinomac 8 0 0 9 0 0 10 4 0 11 47 31
a b

ID, identification; DCIS, ductal carcinoma in situ. High-risk women with no mammographic abnormality. c Ductal cells retrieved from irrigation fluid during endoscopy of women with biopsy-proven cancer. Histopathological diagnosis of the resected lesion from these breasts showed ductal carcinoma in situ or invasive ductal carcinoma.

4450

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

cells, assessment of the status of more than three genes in replicate assays could be extremely difficult. There is clearly a need for new strategies to better evaluate methylation in samples where DNA is limited (e.g., ductal lavage, plasma, fine-needle or core biopsy, or nipple aspiration fluid). The QM-MSP method combines the principles of M-MSP (30, 31, 33) with quantitative real-time MSP (15, 34 39). We have shown that this method can detect as few as 110 methylated copies of DNA in a mixture of 100,000 copies of unmethylated DNA (Table 3; Fig. 3) and 40 pg of methylated genomic DNA in up to 1500-fold excess unmethylated DNA (Fig. 4). This compares favorably with Q-MSP, which has a sensitivity of 1:10,000 (36), and conventional MSP, which has a sensitivity of 1:1000 (30). Reactions were specific: no cross-reactivity was observed between methylated and unmethylated primers even in mixtures containing a 105-fold excess of one or the other DNA (Fig. 3). Most often the real-time PCR technology used for absolute quantification of DNA uses -actin (ACTB) or GAPDH as reference DNA (34, 36, 37). We have demonstrated that by assessing the levels of unmethylated and methylated product for each gene, it is possible to quantitate the percentage of methylated gene product. Lo et al. (35) and Wong et al. (42) used a similar approach for Q-MSP, although they also considered the contribution of any unconverted bisulfitetreated DNA. However, two potential pitfalls of the QM-MSP method needed to be addressed. One matter of concern is how much bias is introduced into the estimation of gene methylation by the addition of the multiplex reaction to the Q-MSP procedure, as described here. A second concern is that QM-MSP does not take into account the existence of variable fractions of partially methylated DNA in the tissue samples. Therefore, samples could appear to contain higher levels of methylated DNA in the test genes than are present. By testing the first question experimentally, we showed that, with few exceptions, there was excellent concordance between QM-MSP and Q-MSP for all four genes when we used (U M) as the measure of total gene DNA (Table 4). Because QM-MSP and Q-MSP give essentially the same readout, significant bias is not likely in QM-MSP. To address the second question, using direct Q-MSP we calculated methylation as done with ACTB-based Q-MSP by Trinh et al. (37) and as % M by (U M) (see the Results). There was concordance in percentage of methylation calculated by the two ACTB-based formulas and the (U M) formula. Such concordance would be unlikely if partially methylated DNA formed a substantial component of the DNA. In contrast, in the two-step QM-MSP assay, we observed that ACTB does not perform predictably. The reasons could be as follows: In the first step, that of multiplex PCR amplification, the efficiency of amplification of each of the genes was not identical, however well optimized. In some samples, ACTB did not seem to amplify as well as some of the genes in the mixture. Because the strength of the second Q-PCR reaction depends on the efficiency of the first, differences are magnified in the second reaction. Thus, for QM-MSP we decided to calculate percentage of methylation using the (U M) formula. In addition, in QM-MSP, a gene controls for itself. In this assay, the sum of unmethylated and the methylated alleles of the same gene serve as a reliable internal control for integrity, copy number, and method efficiency. Among these, copy number is an important consideration because allelic losses and amplifications can vary among different areas of the genome and between samples. For example, for RASSF1A two simultaneous methods of gene inactivation have been observed: loss of one allele and methylation of the other (43). This has also been reported for FHIT (44), APC, and CDH1 (13). Wang et al. (45) reported that nearly every breast tumor has an individual pattern of allelic imbalance or loss of heterozygosity at multiple loci, which constitutes its fingerprint.

The QM-MSP technique is applicable to frozen or archival paraffin-embedded clinical tissues (Figs. 57) as well as to ductal lavage material (Fig. 8). In a study of 14 28 tissue samples/group, we observed significant differences in the level of promoter hypermethylation between normal and carcinoma samples for each of four genes, based on comparison of mean and median normal values (Fig. 5; Table 5). QM-MSP enabled definition of the normal range for the percentage of methylation in the genes in normal breast tissue (Table 5; Fig. 5). Techniques that give higher sensitivity usually also give higher background, picking up signals that are missed by other methods. This was also observed in our present study, in which we found a higher incidence of methylation in normal mammoplasty than we did previously (23) using gel-based nonquantitative MSP. Nevertheless, with QM-MSP, the median for normal tissues was 0% M for all genes. By setting an upper threshold for normal, we acknowledge the occasional low-level methylation that occurs in some normal tissues and set criteria that define positive in carcinoma. By determining that peripheral blood cells contain little or no methylation of the genes tested and that normal breast tissue, which is rich in stroma, is for the most part negative, we were able to deuce that the methylation signal is derived largely from the epithelial cells (Fig. 5). The incidence of positivity among carcinomas was 68% for RASSF1A, 67% for TWIST, 57% for Cyclin D2, and 57% for HIN1 (Table 5) according to this stringent criterion. However, all carcinomas showed some degree of methylation of one or more of the genes in our panel. Studies of cumulative multigene promoter hypermethylation revealed striking differences between normal and malignant tissues (Fig. 6; Table 6). There was a highly significant (P 0.0002) difference between levels of cumulative gene promoter hypermethylation in normal tissues compared with malignant tissues. Cumulative methylation profiling of four genes was able to detect 84% of carcinomas, whereas single-gene analyses yielded positive results in only 57 68% cases, depending on the gene analyzed (Figs. 5 and 6). To our knowledge, this is the first study to describe quantitation of cumulative methylation and to show its importance in distinguishing between normal and carcinoma tissues. Molecular alterations in histologically normal-appearing breast tissue adjacent to carcinomas have been reported previously (46 48). Promoter hypermethylation analysis of RASSF1A, TWIST, Cyclin D2, and HIN1 (Table 6; Fig. 7) in six pairs of carcinomas and histologically normal adjacent tissues showed that all six carcinomas were positive. Four of six adjacent normal tissues were also positive, although the levels were considerably lower than in the carcinoma (P 0.01, MannWhitney test). More detailed studies are needed to determine whether methylation in histologically normal tissue adjacent to carcinoma represents a field effect presaging cancer or are normal, age-related changes. That this may represent a field effect is suggested by our observation that in 25 samples of normal breast tissue, including those reported here, we have not observed a correlation between age and methylation in these five genes.6 That it is possible to apply the QM-MSP successfully to samples with little cellularity was demonstrated by our pilot study with ductal cells retrieved by lavage or endoscopy (Fig. 8; Table 7; Ref. 19). In the seven DL samples from high-risk, but mammographically normal breasts, no promoter hypermethylation was detectable. In contrast, ductal lavage samples obtained during endoscopy of two women with invasive carcinoma had high-level multigene promoter hypermethylation, consistent with the histological diagnosis of the resected tissue. Interestingly, samples from the two women with ductal carcinoma in situ demonstrated benign cytology and lacked detectable promoter
6

Our unpublished data.

4451

QM-MSP FOR QUANTITATION OF PROMOTER HYPERMETHYLATION

hypermethylation (samples 8 and 9). The sample size is small, and an ongoing clinical trial collecting ductal cells from both diseased and uninvolved breasts of cancer patients will allow us to address the utility of QM-MSP in greater detail. With the QM-MSP approach, it is possible to put together several gene panels consisting of scores of genes that are designed for early detection or to provide intermediate markers or endpoints for clinical protocols. For example, when retinoids or demethylating agents are being used as chemopreventive agents, a panel can be designed to query pathway-specific genes for their use as intermediate markers in clinical trials (26). Furthermore, the QM-MSP method is applicable to all types of cancer and evaluation of methylated tumor DNA in other small clinical samples, such as prostatic fluid, bile duct washings, and fine-needle aspirates. In summary, we describe a method that assesses the gene promoter hypermethylation status of multiple genes, using only picograms of DNA. We demonstrate the advantages of a cumulative score of promoter hypermethylation among multiple genes and how this approach may better distinguish normal/benign from malignant tissues. With QM-MSP it is possible to objectively define the range of normal/abnormal gene promoter hypermethylation in a manner that could translate to a larger clinical setting. Further studies should examine cumulative hypermethylation in benign conditions and as a predictor of breast cancer risk.

ACKNOWLEDGMENTS
We gratefully acknowledge Christopher Umbricht for helpful discussions and Alan Rein, Scott Kern, and Nancy Davidson for reviewing the manuscript.

REFERENCES
1. Hanahan D, Weinberg RA. The hallmarks of cancer. Cell 2000;100:5770. 2. Warnecke PM, Bestor TH. Cytosine methylation and human cancer. Curr Opin Oncol 2000;12:68 73. 3. Yang X, Yan L, Davidson NE. DNA methylation in breast cancer. Endocr Relat Cancer 2001;8:11527. 4. Widschwendter M, Jones PA. DNA methylation and breast carcinogenesis. Oncogene 2002;21:5462 82. 5. Wajed SA, Laird PW, DeMeester TR. DNA methylation: an alternative pathway to cancer. Ann Surg 2001;234:10 20. 6. Jones PA, Baylin SB. The fundamental role of epigenetic events in cancer. Nat Rev Genet 2002;3:41528. 7. Goyal J, Smith KM, Cowan JM, Wazer DE, Lee SW, Band V. The role for NES1 serine protease as a novel tumor suppressor. Cancer Res 1998;58:4782 6. 8. Dhar S, Bhargava R, Yunes M, et al. Analysis of normal epithelial cell specific-1 (NES1)/kallikrein 10 mRNA expression by in situ hybridization, a novel marker for breast cancer. Clin Cancer Res 2001;7:3393 8. 9. Li B, Goyal J, Dhar S, et al. CpG methylation as a basis for breast tumor-specific loss of NES1/kallikrein 10 expression. Cancer Res 2001;61:8014 21. 10. Yunes MJ, Neuschatz AC, Bornstein LE, Naber SP, Band V, Wazer DE. Loss of expression of the putative tumor suppressor NES1 gene in biopsy-proven ductal carcinoma in situ predicts for invasive carcinoma at definitive surgery. Int J Radiat Oncol Biol Phys 2003;56:6537. 11. Kashiwaba M, Tamura G, Ishida M. Aberrations of the APC gene in primary breast carcinoma. J Cancer Res Clin Oncol 1994;120:72731. 12. Virmani AK, Rathi A, Sathyanarayana UG, et al. Aberrant methylation of the adenomatous polyposis coli (APC) gene promoter 1A in breast and lung carcinomas. Clin Cancer Res 2001;7:1998 2004. 13. Sarrio D, Moreno-Bueno G, Hardisson D, et al. Epigenetic and genetic alterations of APC and CDH1 genes in lobular breast cancer: relationships with abnormal Ecadherin and catenin expression and microsatellite instability. Int J Cancer 2003;106: 208 15. 14. Evron E, Umbricht CB, Korz D, et al. Loss of cyclin D2 expression in the majority of breast cancers is associated with promoter hypermethylation. Cancer Res 2001; 61:27827. 15. Lehmann U, Langer F, Feist H, Glockner S, Hasemeier B, Kreipe H. Quantitative assessment of promoter hypermethylation during breast cancer development. Am J Pathol 2002;160:60512.

16. Widschwendter M, Berger J, Hermann M, et al. Methylation and silencing of the retinoic acid receptor-beta2 gene in breast cancer. J Natl Cancer Inst (Bethesda) 2000;92:826 32. 17. Yan L, Yang X, Davidson NE. Role of DNA methylation and histone acetylation in steroid receptor expression in breast cancer. J Mammary Gland Biol Neoplasia 2001;6:18392. 18. Sirchia SM, Ren M, Pili R, et al. Endogenous reactivation of the RARbeta2 tumor suppressor gene epigenetically silenced in breast cancer. Cancer Res 2002;62:2455 61. 19. Evron E, Dooley WC, Umbricht CB, et al. Detection of breast cancer cells in ductal lavage fluid by methylation-specific PCR. Lancet 2001;357:1335 6. 20. Burbee DG, Forgacs E, Zochbauer-Muller S, et al. Epigenetic inactivation of RASSF1A in lung and breast cancers and malignant phenotype suppression. J Natl Cancer Inst (Bethesda) 2001;93:6919. 21. Dammann R, Yang G, Pfeifer GP. Hypermethylation of the cpG island of Ras association domain family 1A (RASSF1A), a putative tumor suppressor gene from the 3p21.3 locus, occurs in a large percentage of human breast cancers. Cancer Res 2001;61:31059. 22. Krop IE, Sgroi D, Porter DA, et al. HIN-1, a putative cytokine highly expressed in normal but not cancerous mammary epithelial cells. Proc Natl Acad Sci USA 2001;98:9796 801. 23. Fackler MJ, McVeigh M, Evron E, et al. DNA methylation of RASSF1A, HIN-1, RAR-beta, Cyclin D2 and Twist in in situ and invasive lobular breast carcinoma. Int J Cancer 2003;107:970 5. 24. Esteller M, Corn PG, Baylin SB, Herman JG. A gene hypermethylation profile of human cancer. Cancer Res 2001;61:32259. 25. Melki JR, Vincent PC, Clark SJ. Concurrent DNA hypermethylation of multiple genes in acute myeloid leukemia. Cancer Res 1999;59:3730 40. 26. Fackler MJ, Evron E, Khan SA, Sukumar S. Novel agents for chemoprevention, screening methods, and sampling issues. J Mammary Gland Biol Neoplasia 2003;8:75 89. 27. Tsou JA, Hagen JA, Carpenter CL, Laird-Offringa IA. DNA methylation analysis: a powerful new tool for lung cancer diagnosis. Oncogene 2002;21:5450 61. 28. Muller HM, Widschwendter M. Methylated DNA as a possible screening marker for neoplastic disease in several body fluids. Expert Rev Mol Diagn 2003;3:44358. 29. Cottrell SE, Laird PW. Sensitive detection of DNA methylation. Ann N Y Acad Sci 2003;983:120 30. 30. Herman JG, Graff JR, Myohanen S, Nelkin BD, Baylin SB. Methylation-specific PCR: a novel PCR assay for methylation status of CpG islands. Proc Natl Acad Sci USA 1996;93:9821 6. 31. Palmisano WA, Divine KK, Saccomanno G, et al. Predicting lung cancer by detecting aberrant promoter methylation in sputum. Cancer Res 2000;60:5954 8. 32. Buller A, Pandya A, Jackson-Cook C, et al. Validation of a multiplex methylationsensitive PCR assay for the diagnosis of Prader-Willi and Angelmans syndromes. Mol Diagn 2000;5:239 43. 33. Brock MV, Gou M, Akiyama Y, et al. Prognostic importance of promoter hypermethylation of multiple genes in esophageal adenocarcinoma. Clin Cancer Res 2003;9:29129. 34. Eads CA, Danenberg KD, Kawakami K, Saltz LB, Danenberg PV, Laird PW. CpG island hypermethylation in human colorectal tumors is not associated with DNA methyltransferase overexpression. Cancer Res 1999;59:2302 6. 35. Lo YM, Wong IH, Zhang J, Tein MS, Ng MH, Hjelm NM. Quantitative analysis of aberrant p16 methylation using real-time quantitative methylation-specific polymerase chain reaction. Cancer Res 1999;59:3899 903. 36. Eads CA, Danenberg KD, Kawakami K, et al. MethyLight: a high-throughput assay to measure DNA methylation. Nucleic Acids Res 2000;28:E32. 37. Trinh BN, Long TI, Laird PW. DNA Methylation analysis by MethyLight technology. Methods 2001;25:456 62. 38. Lehmann U, Hasemeier B, Lilischkis R, Kreipe H. Quantitative analysis of promoter hypermethylation in laser- microdissected archival specimens. Lab Investig 2001;81: 635 8. 39. Jeronimo C, Usadel H, Henrique R, et al. Quantitative GSTP1 hypermethylation in bodily fluids of patients with prostate cancer. Urology 2002;60:11315. 40. Maniatis T, Fritsch EF, Sambrook J. Molecular cloning: a laboratory manual. New York: Cold Spring Harbor, 1982. 41. Harrison J, Stirzaker C, Clark SJ. Cytosines adjacent to methylated CpG sites can be partially resistant to conversion in genomic bisulfite sequencing leading to methylation artifacts. Anal Biochem 1998;264:129 32. 42. Wong IH, Zhang J, Lai PB, Lau WY, Lo YM. Quantitative analysis of tumor-derived methylated p16INK4a sequences in plasma, serum, and blood cells of hepatocellular carcinoma patients. Clin Cancer Res 2003;9:104752. 43. Hogg RP, Honorio S, Martinez A, et al. Frequent 3p allele loss and epigenetic inactivation of the RASSF1A tumour suppressor gene from region 3p21.3 in head and neck squamous cell carcinoma. Eur J Cancer 2002;38:158592. 44. Yang Q, Nakamura M, Nakamura Y, et al. Two-hit inactivation of FHIT by loss of heterozygosity and hypermethylation in breast cancer. Clin Cancer Res 2002;8:2890 3. 45. Wang Q, Larson PS, Schlechter BL, et al. Loss of heterozygosity in serial plasma DNA samples during follow-up of women with breast cancer. Int J Cancer 2003;106:9239. 46. Deng G, Lu Y, Zlotnikov G, Thor AD, Smith HS. Loss of heterozygosity in normal tissue adjacent to breast carcinomas. Science (Wash DC) 1996;274:20579. 47. Califano J, van der Riet P, Westra W, et al. Genetic progression model for head and neck cancer: implications for field cancerization. Cancer Res 1996;56:2488 2492. 48. Umbricht CB, Evron E, Gabrielson E, Ferguson A, Marks J, Sukumar S. Hypermethylation of 14-3-3 sigma (stratifin) is an early event in breast cancer. Oncogene 2001;20:3348 53.

4452