METHODS FOR PREDICTING ESOPHAGEAL ADENOCARCINOMA (EAC)

This application claims the benefit of the filing dates of provisional patent applications 61/066,281, filed February 19, 2008; 61/131,748, filed June 11, 2008; and 61/132,418, filed June 18, 2008, all of which are incorporated by reference in their entireties herein. This application was made with U.S. government support, including NIH/NCI CA

085069. The U.S. government thus has certain rights in the invention.

BACKGROUND INFORMATION

Barrett's esophagus (BE), a sequela of chronic gastroesophageal reflux disease (GERD), is a highly premalignant condition that increases an individual's chance of developing esophageal adenocarcinoma (EAC) by 30- to 125-fold. Therefore, subjects with BE are usually enrolled in surveillance programs in which they undergo endoscopy at regular intervals for the rest of their lives. However, the incidence of EAC in BE patients under surveillance is only 1/200 patient-years. Conversely, cancers or advanced high-grade dysplasias (HGDs) may develop during the interim and are sometimes missed if surveillance is performed at long intervals. In addition, the current marker of EAC risk in BE, dysplasia, is plagued by high inter- observer variability and limited predictive accuracy. Because neoplastic progression is infrequent in BE, the merits of and appropriate interval for endoscopic surveillance in BE have led to frequent debate. Thus, a means of stratifying patients into groups at high, intermediate, and low risk of neoplastic progression would be highly useful. This process would benefit greatly from effective biomarkers to stratify patients according to their level of neoplastic progression risk.

Methylation constitutes the epigenetic modification of DNA by the addition of methyl groups, usually on cytosines at the sequence 5'-CpG-3'. This event is most relevant when it occurs within CpG islands, which are CpG-rich regions in 5' gene regions of about half of all genes, often involving promoter regions. These islands are normally unmethylated but are vulnerable to de novo methylation, which can silence gene expression. See, e.g., Gardiner- Garden et al. ( 1987) J MoI Biol 1%, 261 -282 or Takai et al. (2002) Proc Natl Acad Sci USA 99, 3740-3745 for discussions of CgG islands. It has been reported that promoter hypermethylation of several tumor suppressor genes is correlated with the incidence of several cancers.

The inventors and their colleagues previously reported that hypermethylation of promoter regions of three genes - cyclin-dependent kinase inhibitor 2a (CDKN2a, or pi 6), runt- related transcription factor 3 (RUNX3), and transmembrane protein with EGF-like and two

follistatin-like domains (HPPl) - occurs early in (BE)-associated neoplastic progression and appears to represent independent risk factors for the progression of Barrett's esophagus (BE) to high-grade dysplasias (HGD) or esophageal adenocarcinoma (EAC). See, e.g., Schulmann et al. (2005) Oncogene ]A, 4138-4148. Later, the inventors and colleagues developed a tiered risk stratification model to predict progression in BE using epigenetic and clinical features, validating the use of a panel of the three markers (see, e.g., Sato et al. (2008) PLoS ONE 3, el890).

Hypermethylation of these and additional promoter regions might serve as useful biomarkers for stratifying subjects according to their risk for development of EAC or HGD.

DESCRIPTION OF THE DRAWINGS

Figure 1 shows the genomic sequence of a promoter region oϊplό, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. IA), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. IB).

Figure 2 shows the genomic sequence of a promoter region of AKAF '12, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 2A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 2B).

Figure 3 shows the genomic sequence of a promoter region of CDHl 3, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 3A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 3B).

Figure 4 shows the genomic sequence of a promoter region oϊHPPl, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 4A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 4B).

Figure 5 shows the genomic sequence of a promoter region of NELLl, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 5A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 5B).

Figure 6 shows the genomic sequence of a promoter region 0ϊRUNX3, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 6A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 6B).

Figure 7 shows the genomic sequence of a promoter region of SST, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 7A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 7B).

Figure 8 shows the genomic sequence of a promoter region of TACl, from 1000 nt upstream of the transcriptional start site into exon 1, indicating the start and end sequences of the CpG Island and the transcriptional start site (TSS) (Fig. 8A), and the bisulfite methyl sequence, indicating the locations of the Cg sequences, the TSS, and the locations of the forward and reverse primers and of the probe used in qMSP (Fig. 8B).

Figure 9 shows receiver-operator characteristic (ROC) curve analysis of normalized methylation value (NMV). ROC curve analysis of CDHlS NMVs in esophageal adenocarcinoma (EAC) vs. normal esophagus (NE) (Fig. 9A), esophageal squamous cell carcinoma (ESCC) vs. NE (Fig. 9B), and EAC vs. ESCC (Fig. 9C). The high area under the ROC curve (AUROC) conveys the accuracy of this biomarker in distinguishing EAC from NE and from ESCC in terms of its sensitivity and specificity.

Figure 10 shows a correlation between Barrett's segment length and CDH13 hypermethylation. Fig. 1OA shows that the normalized methylation value (NMV) of CDHl 3 was significantly higher in long-segment BE (LSBE, mean = 0.4071) than in short-segment BE (SSBE, mean = 0.131; p = 0.0032, Student's t-test). Fig. 1OB shows that positive CDHl 3 hypermethylation status was significantly correlated with BE segment length (p = 0.0005, Student's t-test).

Figure 11 shows CDHlS methylation level and mRNA expression in esophageal cancer cell lines after treatment with the demethylating agent 5-aza-2'-deoxycytidine (5-Aza-dC). KYSE220 and OE33 EAC cells were subjected to 5-Aza-dC treatment. In both cell lines, after 5-Aza-dC treatment, the NMV of CDH13 was diminished, while the normalized mRNA value (NRV) of CDH 13 was increased.

Figure 12 shows receiver-operator characteristic (ROC) curves for the uncorrected 8-marker and 8-marker-plus-age panels, overfitting-corrected ROC curves for the 8-marker and 8-marker- plus-age panels, and ROC curves for age alone in the 2-, 4-year, and combined prediction models. Fig. 12A shows an uncorrected ROC curve (AUC=O.843) and an overfitting-corrected ROC curve (AUC=0.745) for the 8-marker panel in the 2-year prediction model; shrinkage due to overfitting correction minimal, at 0.098. Fig. 12B shows an uncorrected ROC curve (AUC=0.829) and an overfitting-corrected ROC curve (AUC=0.720) for the 8-marker panel in the 4-year prediction model; shrinkage minimal, at 0.109. Fig. 12C shows an uncorrected ROC curve (AUC=0.840) and an overfitting-corrected ROC curve (AUC=O.732) for the 8-marker panel in the combined prediction model; shrinkage minimal, at 0.108. Fig. 12D shows an uncorrected ROC curve for the 8-marker-plus-age panel (AUC=O.858), overfitting-corrected ROC curve (AUC=0.756), and a ROC curve for age alone (0.604) in the 2-year prediction model; increment over age alone substantial, at 0.152. Fig. 12E shows an uncorrected ROC curve for the 8-marker-plus-age panel (AUC=O.850), an overfitting-corrected ROC curve (AUC=O.744), and a ROC curve for age alone (0.630) in the 4-year prediction model; increment over age alone substantial, at 0.114. Fig. 12F shows an uncorrected ROC curve for the 8- marker-plus-age panel (AUC=O.855), an overfitting-corrected ROC curve (AUC=0.753), and a ROC curve for age alone (0.635) in the combined prediction model; increment over age alone substantial, at 0.118. Figure 13 shows risk stratification of BE patients by predictiveness curves of the 8-marker panel, age alone, and the 8-marker-plus-age panel in the combined, 2-year and 4-year prediction models. Fig. 13A shows a predictiveness curve of the 8-marker panel in the combined model. After rigorous overfitting correction, at risk = 0.1 and = 0.5, 45% and 4% of subjects had estimated risks below 0.1 (LR group) and above 0.5 (HR group), respectively; while the remaining 51% had estimated risks between 0.1 and 0.5 (IR group). Fig. 13B shows predictiveness curves of age alone and of the 8-marker-plus-age panel in the combined model. After rigorous overfitting correction, BE patients were stratified into LR (15%) and IR (85%) groups by age alone. BE patients were stratified into LR (51%), IR (44%) and HR (5%) groups

by the 8-marker-plus-age panel. Fig. 13C shows a predictiveness curve of the 8-marker panel in the 2-year model. After rigorous overfitting correction, BE patients were stratified into LR (45%), IR (51%) and HR (4%) groups. Fig. 13D shows predictiveness curves of age alone and of the 8-marker-plus-age panel in the 2-year model. After rigorous overfitting correction, BE patients were stratified into LR (11%) and IR (89%) groups by age alone. BE patients were stratified into LR (52%), IR (44%) and HR (4%) groups by the 8-marker-plus-age panel. Fig. 13E shows a predictiveness curve of the 8-marker panel alone in the 4-year model. After rigorous overfitting correction, BE patients were stratified into LR (44%), IR (51%) and HR (5%) groups. Fig. 13F shows predictiveness curves of age alone and of the 8-marker-plus-age panel in the 4-year model. After rigorous overfitting correction, BE patients were stratified into LR (15%) and IR (85%) groups by age alone. BE patients were stratified into LR (51%), IR (44%) and HR (5%) groups by the 8-marker-plus-age panel. LR: low-risk; IR, intermediate-risk; HR: high-risk; OC: overfitting correction.

Figure 14 shows statistical considerations, including cut-off points for Low and High risk.

DESCRIPTION OF THE INVENTION

The inventors extend herein their previous studies identifying hypermethylation of promoter regions of HPPl, pl6 and RUNX3 as markers for the development of esophageal adenocarcinoma (EAC) or high-grade dysplasia (HGD), by identifying about 55 additional markers that can be used to predict a subject's risk for developing EAC or HGD. These additional markers include, e.g., hypermethylated promoter regions of nel-like 1 (NELLl), tachykinin-1 (TACl), somatostatin (SST), A-kinase anchoring protein 12 (AKAP 12), cadherin 13, H-cadherin (heart) (CDH13), and of the 50 genes shown in Table 11.

This invention relates, e.g., to a method for predicting a subject's risk for developing (e.g., progressing from BE to) EAC or HGD, comprising (a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of at least two or three of the above-mentioned genes, in various combinations that are elucidated elsewhere herein, and (b) calculating a methylation value (e.g., a methylation index or a linear regression score) that takes into account the methylation levels of the measured transcriptional promoter regions, wherein a methylation value that is below a first predetermined threshold value is indicative that the subject has a low risk of developing EAC or HGD, and a methylation value that is above a second predetermined threshold value is indicative that the subject has a high risk of developing EAC or HGD.

As used herein, a subject that is at "high risk"(or at an "increased risk") for developing EAC or HGD has a greater than 90% likelihood of developing EAC or HGD, within 2 years, 4 years, or ever at all, depending on which model as discussed herein is used. "Low risk" (or a "decreased risk") is defined is defined as a greater than 95% chance that the patient will NOT develop HGD or EAC - within 2 years, 4 years, or ever at all, depending on which model as discussed herein is used.

Advantages of a method of the invention include, e.g., that it is rapid, accurate and inexpensive, and that it can be easily adapted to high throughput format, using automated (e.g., robotic) systems, which allow many measurements to be carried out simultaneously. Furthermore, the methods can be miniaturized. A stratification method of the invention can benefit BE patients in two ways: 1) by decreasing the frequency at which low-risk individuals undergo surveillance endoscopy, thus eliminating unnecessary anxiety, expense, and diminishing procedure-related complications; and 2) by identifying the small group of truly high-risk BE patients for more frequent, intensive surveillance, resulting in earlier and more accurate detection of HGDs and EACs.

One aspect of the invention is a first method for predicting a subject's risk for developing esophageal adenocarcinoma (EAC) or high-grade dysplasia (HGD), comprising a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of three or more of the following genes: i) CDHB, ii) TACl, iii) NELLl, W) AKAP 12) or v) SST, and b) calculating a methylation value (e.g., a methylation index or a linear regression score) that takes into account the methylation levels of the measured transcriptional promoter regions, wherein a methylation value that is below a first predetermined threshold is indicative that the subject has a low risk of developing EAC or HGD, and a methylation value that is above a second predetermined threshold is indicative that the subject has a high risk of developing EAC or HGD.

This first method can further comprise

a) determining the methylation levels of promoter regions of one or more of vi) HPP/, mii) RUNX3, and b) calculating a methylation value that takes into account the methylation levels of the measured transcriptional promoter regions.

In embodiments of this first method, the methylation levels can be determined for promoter regions of a total of 3, 4, 5, 6, 7 or all 8 of the genes. In another embodiment of this first method, methylation levels for the promoter regions of one or more of the 50 genes listed in Table 11 can also be determined and included in the calculation of the methylation value.

Another aspect of the invention is a second method for predicting a subject's risk for developing esophageal adenocarcinoma (EAC) or high-grade dysplasia (HGD), comprising a) determining in a sample from the subject the methylation level of a transcriptional promoter region of CDHl 3, and the methylation levels of transcriptional promoter regions of at least two of the following genes: i) HPPl

H) pi 6 iii) RUNXS iv) TACl v) NELLl vi) AKAP '12, or vii) SST, and b) calculating a methylation value (e.g. a linear regression score or a methylation index) that takes into account the methylation levels of the measured transcriptional promoter regions, wherein a methylation value that is below a first predetermined threshold is indicative that the subject has a low risk of developing EAC or HGD, and a methylation value that is above a second predetermined threshold is indicative that the subject has a high risk of developing EAC or HGD. In embodiments of the this second method, the methylation levels can be determined for promoter regions of a total of 3, 4, 5, 6, 7 or all 8 of the genes; in another embodiment, the methylation levels are measured for CDH13, HPPl, pl6 and RUNX3. In another embodiment of this second method, methylation levels for the promoter regions of one or more of the 50 genes

listed in Table 11 can also be determined and included in the calculation of the methylation value.

Another aspect of the invention is a third method for predicting a subject's risk for developing esophageal adenocarcinoma (EAC) or high-grade dysplasia (HGD), comprising a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of at least two of the genes listed in Table 11, and b) calculating a methylation value that takes into account the methylation levels of the measured transcriptional promoter regions, wherein a methylation value (e.g. a linear regression score or a methylation index) that is below a first predetermined threshold is indicative that the subject has a low risk of developing

EAC or HGD, and a methylation value that is above a second predetermined threshold is indicative that the subject has a high risk of developing EAC or HGD. Any combination of two or more of the genes listed in Table 11 can be used.

Another aspect of the invention is a method for predicting a subject's risk for developing EAC or HGD, comprising a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of CDHlS, TACl, NELLl, AKAPl 2, SST, HPPl, pi 6, and RUNXS, and b) calculating a linear regression score for the 8 methylation levels, wherein the methylation levels are determined by qMSP, and wherein a linear regression score that is no more than 0.13 is indicative that the subject has a low risk of developing EAC or HGD within 4 years, and a linear regression score that is equal to or above 0.39 is indicative that the subject has an increased risk (e.g., a high risk) of developing EAC or HGD in 4 years.

Another aspect of the invention is a method for predicting a subject's risk for developing EAC or HGD, comprising a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of CDHlS, TACl, NELLl, AKAPl 2, SST, HPPl, pi 6, and RUNXS, and b) calculating a methylation index for the 8 methylation levels, wherein the methylation levels are determined by qMSP, and wherein a methylation index that is no more than 2 is indicative that the subject has a decreased risk (e.g., a low risk) of developing EAC or HGD in 4 years, and a methylation index that is equal to or above 3 is indicative that the subject has an increased risk (e.g., a high risk) of developing EAC or HGD in 4 years.

In any of the preceding methods, the subject can have, or be suspected of having, BE; the subject can be human; and/or the sample can be a biopsy tissue. Any of a variety of methods can be used to determine the methylation levels, including the quantitative methods: real-time quantitative methylation-specific PCR (qMSP), pyrosequencing, or methylation arrays. A method of the invention can be performed in conjunction with other assays for EAC, including, e.g., performing conventional histological analysis of the sample (wherein the detection of the presence and degree of dysplasia is further indicative that the subject is at increased risk for developing EAC); determining the age of the subject (wherein if a human subject is more than about 60 years old, this is further indicative that the subject is at risk for developing EAC, and wherein the greater the age of the patient, the higher his or her risk); or determining the BE segment length (wherein a length of 3 cm or greater is further indicative that the subject has a higher risk of developing EAC).

Another aspect of the invention is a method for determining a treatment strategy for a subject (or a method for treating a subject), comprising predicting the subject's risk for developing EAC or HGD and, depending on the assessed risk, deciding how to treat or monitor the subject (or treating or monitoring the subject). For example, if the subject is predicted to be at a decreased risk (e.g. at a low risk) for developing EAC or HGD, but has BE, a decision is made to treat the subject by endoscopic monitoring with endoscopy and biopsies every 2-3 years (e.g., every 3 years); whereas if the subject is predicted to be at an increased risk (e.g. at a high risk) for developing EAC or HGD but has BE, a decision is made to treat the subject by monitoring with endoscopy and biopsies every one year or less.

Another aspect of the invention is a method for following the course of development of EAC in a subject, comprising (a) determining in a sample from the subject, at at least two time points, the methylation levels of transcriptional promoter regions of a set of genes as described above, and (b) calculating a methylation value that takes into account the methylation levels of the measured transcriptional promoter regions, wherein an increase in the methylation value between the at least two time points indicates that the EAC has progressed. Any set of time points can be used, e.g., yearly, every other year, etc.

Another aspect of the invention is a method for evaluating a therapeutic method (including, e.g., monitoring the effect of a candidate therapeutic agent), comprising, before and after initiation of the therapy, (a) determining in a sample from the subject the methylation levels of transcriptional promoter regions of a set of genes as described above, and (b) calculating a methylation value that takes into account the methylation levels of the measured transcriptional

promoter regions, wherein an decrease in the methylation value following the therapeutic treatment indicates that the treatment has been effective.

Another aspect of the invention is a kit for predicting a subject's risk for developing esophageal adenocarcinoma (EAC) or high-grade dysplasia (HGD), comprising reagents (e.g., suitable primers and probes for determining in a sample from the subject the methylation level of a transcriptional promoter region from the genes or combinations of genes discussed herein, using qMSP; and, optionally, directions for carrying out a method of the invention, containers or packaging materials, and/or a computer program {e.g., a program that calculates a methylation value in a sample based on the determined methylation levels).

In general, a subject to be evaluated by a method of the invention has, or is suspected of having BE. Human patients suspected of having BE often present with heartburn and are subjected to an endoscopy to definitively determine by endoscopy and biopsies for histology whether they have BE and/or dysplasia. Subjects who evince BE, with or without dysplasia, are generally monitored endoscopically periodically, even if symptoms later disappear.

Subjects that can be evaluated by a method of the invention include any of a variety of vertebrates, including, e.g., laboratory animals (e.g., mouse, rat, rabbit, monkey, or guinea pig, in particular mouse or rat models for EAC), farm animals (e.g., cattle, horses, pigs, sheep, goats, etc.), and domestic animals or pets (e.g., cats or dogs). Non-human primates and, preferably, humans, are included.

Suitable samples (e.g., test samples, or control samples) that can be tested by a method of the invention include, e.g., biopsies of esophageal epithelium. For example, the sample can be from grossly apparent BE epithelium or from mass lesions in patients manifesting these changes at endoscopic examination. Methods for obtaining samples and preparing them for analysis are conventional and well-known in the art.

A transcriptional "promoter region," as used herein, refers to a sequence that is at or near the 5' end of an mRNA transcribed from a gene. The region can be, or can be a portion of, a sequence of the genomic DNA comprising

(a) at least about 500 (and in some cases, as many as about 1,000 or 2,000) contiguous base pairs (bp) extending upstream of a transcription initiation site, wherein the sequence comprises transcriptional regulatory sequences, such as promoter sequences, and at least a portion of a CpG island; or

(b) the transcription initiation site;

(c) at least about 200 bp of the 5' coding sequence of the first exon of a gene, or the entire first exon when the sequence is included in a CpG island; or

(d) overlapping sequences thereof.

Although much of the discussion herein is directed to promoter regions that extend no further downstream into a gene than the first exon, recent studies have suggested that hypermethylation of CpG islands that are located much further downstream in a gene, or in intergenic regions, can also silence gene expression. Furthermore, hypermethylation of CpG "shores" that extend as far as 3,000 bp upstream of a transcriptional start site, have also been reported to silence gene expression. See, e.g., Irizarry et al. (2009) Nat Genet 4L, 178-186. As such more distant CpG islands become identified for the genes identified herein, methylation of those regions can also be assayed by a method of the invention to predict a subject's risk for developing EAC or HGD.

In some embodiments of the invention, a promoter region is selected for analysis which consists of a smaller portion of a larger promoter region as discussed above. For example, if qMSP PCR is used to determine the methylation level, the MSP PCR products are generally designed to be about 80-100 bp long. The actual sequences assayed to determine methylation levels can be the primers (about 18-24 nt) and/or probe (about 18-24 nt). Within each of these about 20-base pieces, there are generally between 2 and 7 CpGs. Annealing of the primers and probes must be "specific," and depends on a complete or near-complete match with the test DNA, with greater fidelity being required at or near the 3' end than at or near the 5' end of the primer or probe. Technically, the smallest region assayable is about 20 nt. A typical CpG island is about 200 bp long and contains at least 60% of its expected quota of CpGs, which would be 7 CpGs for a 200 bp island. A transcriptional promoter region that is assayed by a method of invention should contain one or more sites associated with methylation, such as CpG dinucleotides, and suitable for quantitative measurement, such as by qMSP.

A skilled worker would know how to select a suitable promoter region to analyze for a gene of interest. For example, sequences can be selected from the large promoter regions that are shown in Figures 1 - 8 for genes HPPl, pi 6, RUNX3, NELLl, TACl, SST, AKAPl 2, and CDH13, and from the large promoter regions of SEQ ID NOs: 50-121. Suitable promoter regions can be selected on the basis of sequences found in searchable databases, such as GenBank or Entrez Gene (NCBI), and the annotations in the records therein regarding the positions of, for example, the transcription initiation sites. Such information can also be obtained

from a variety of other sources that will be evident to a skilled worker. See, for example, the discussion in Example IV.

Sequences of promoter regions shown herein and in searchable databases are sometimes presented as just one strand of a DNA double helix. Because the PCR-amplified DNA in a test sample is double-stranded, a probe of the invention can bind to one of the two DNA strands of the double-stranded DNA, and is completely complementary to the other strand. Which of the two strands is being described is not always indicated in the discussion herein; but it will be evident to a skilled worker whether a given probe binds to one of the DNA strands of an amplicon, or to its complete complement. A "methylation level," as used herein, is a function of the method used to measure this value. In the experiments shown herein, the measurement is accomplished with qMSP. Using this method, a methylation level is the quantitative measurement of methylated DNA for a gene promoter region, defined by the percentage of DNA in the sample that is methylated at each MSP primer/promoter location. This can range from 0 to 100%. In general, but not invariably, most of the CpGs in a given CpG island are methylated. If other methods are used to measure the methylation level, such as pyrosequencing or bisulfite sequencing, the measurement can reflect a quantitative measurement of the number of methylated cytosines in CpG islands (or CpG dinucleotides) in a promoter region from a gene of interest.

It is generally desirable to measure a methylation level in relation to a normalization standard or a reference value. In Example II herein, actin is used as an internal control to determine how much DNA is methylated, and the methylation levels are referred to as NMVs (normalized methylation values). The amount of actin is measured by PCR rather than MSP, because the primers/probes for actin contain no methylatable sequences (i.e., there are no CpGs in the sequences). Other suitable normalization controls, including other constitutive genes or genes whose sequences which are methylated to a known degree, will be evident to a skilled worker.

In establishing that the eight markers for which qMSP studies are reported herein are strongly predictive, the inventors compared the methylation levels of the markers to baseline values in normal tissues. ROC curves for normal tissue vs. Barrett's and normal tissue vs. tumor are shown in the Examples herein; these ROC curves had very large areas (AUROCs) under them, indicating values ranging from 0.622 to 0.845.

These eight markers were selected for the panel, at least in part, because they are not methylated in normal tissues of normal subjects. That is, the mean methylation level for one of

these markers in a normal population would be zero. Therefore, an assay using these eight markers does not require a comparison to normal subjects. This is an advantage of using these eight markers. A "normal" subject, as used herein, is one who does not have detectable neoplasia or metaplasia of the esophagus and therefore is not expected to develop BE or EAC. In general, a level associated with a "normal" subject includes a statistically obtained value associated with a population of normal subjects. For example, a methylation level in a normal subject includes the mean or average methylation level in a substantial population of subjects.

For other markers, such as the promoter regions of some of the 50 genes listed in Table 11, there may be some degree of methylation of the markers in normal subjects. In those cases, the methylation level is compared to the level in a comparable region from a normal subject or population of subjects, such as from the same promoter region from a matched tissue. Suitable comparisons can be made to levels in normal white blood cells (WBCs) and/or normal esophagus, from normal and/or diseased subjects.

In some embodiments, it is desirable to express the results of an assay in terms of a statistically significant increase in a value compared to a baseline value. A "significant" increase or decrease in a value, as used herein, can refer to a difference which is reproducible or statistically significant, as determined using statistical methods that are appropriate and well- known in the art, generally with a probability value of less than five percent chance of the change being due to random variation. Some such statistical tests are discussed herein; others will be evident to a skilled worker.

It will be appreciated by those of skill in the art that a baseline or normal level need not be established for each assay as the assay is performed but rather, baseline or normal levels can be established by referring to a form of stored information regarding a previously determined baseline methylation levels for a given gene or panel of genes, such as a baseline level established by any of the above-described methods. Such a form of stored information can include, for example, a reference chart, listing or electronic file of population or individual data regarding "normal levels" (negative control) or polyp positive (including staged tumors) levels; a medical chart for the patient recording data from previous evaluations; a receiver-operator characteristic (ROC) curve; or any other source of data regarding baseline methylation levels that is useful for the patient to be diagnosed.

A "methylation value," as used herein, is a quantitative value that takes into account the methylation levels of all of the markers tested in a particular assay, and weighs the contributions of the levels appropriately. A methylation value can be calculated in several ways, which will be

evident to a skilled worker. These include, e.g., a linear regression score from the linear regression of all of the markers in a panel (e.g., the panel of eight markers described in Example III), or a "methylation index."

In one embodiment of the invention, the methylation value is expressed as a linear regression score, as described, e.g., in Irwin, in Neter, Kutner, Nachtsteim, Wasserman (1996) Applied Linear Statistical Models, 4 ^th edition, page 295. See also the Examples herein. Typical regression scores are indicated in Table 1, for both an 8-marker panel (using all 8 of the markers discussed herein) and a 3-marker panel with the promoter regions oϊHPPl, pi 6 and RUNX3. Table 1 8-marker panel:

3 -marker panel:

When the methylation value is expressed in this manner, the cut-off values to determine that a subject is at low risk or at high risk for developing EAC or HGD can be determined from the linear scores in Table 1. For the low-risk cut-off, a very high sensitivity (e.g., 95%) is desirable, so the cut-off value (threshold value) for the 8-marker panel is about 0.147 for the combined model, about 0.073 for the 2-year model, or about 0.130 for the 4-year model. For the high-risk cut-off, a high specificity (e.g., 90%) is desirable, so that the cutoff (score) is about 0.395 for the combined model, about 0.307 for the 2-year panel, and about 0.391 for the 4-year model. The values for a panel of fewer markers would vary accordingly. See, e.g, the values in Table 1 for the 3 -marker panel. The cut-off values would likely differ if different genes were integrated into the model; suitable cut-off values can be determined without undue experimentation by a skilled worker. The term "about" a number, as used herein, refers to any value within 20% of the number.

In another embodiment of the invention, the methylation value is expressed as a "methylation index." A methylation index (MI) is defined as the number of genes which demonstrated an altered methylation level (i.e., which exceed or fall below a previously determined methylation level cutoff) within a defined set of genes. For example, if there are four genes in a defined gene set and none of these four genes is methylated, the MI equals 0; if any one of the four are methylated, the MI equals 1; if any two of the four are methylated, the MI equals 2; if any three of the four are methylated, the MI equals 3; and if all four of these four genes are methylated, the MI equals 4 (i.e., the maximum possible MI for this gene set).

When the methylation value is expressed in terms of a MI, the cut-off values to determine that a subject is at low risk or at high risk for developing EAC or HGD can be the optimal point (the value closest to the origin) on the ROC curve of normal vs. EAC. In one embodiment of the invention, in which the 8-marker panel is used, a methylation index of at most about two is indicative that the subject has a low risk of developing EAC or HGD, and a

methylation index of equal to or greater than about three is indicative that the subject is at high risk for developing EAC or HGD. The cut-off values will be a function of the number of markers in a panel. For example, for a panel or 50 markers, the cut-off values would be less than

12 or equal to or greater than 13. The actual cut-off values for any given panel of markers can be determined readily by a skilled worker, using routine methods, e.g., as described herein.

The difference between the methylation level of a test subject and normal methylation levels may be a relative or absolute quantity. Thus, "methylation level" is used to denote any measure of the quantity of methylation of the gene or panel of genes. The level of methylation may be either abnormally high, or abnormally low, relative to a defined high or low threshold value determined to be normal for a particular group of subjects. The difference in level of methylation between a subject and the reference methylation level may be equal to zero, indicating that the subject is or may be normal, or that there has been no change in levels of methylation since the previous assay.

The methylation levels and any differences that can be detected may simply be, for example, a measured fluorescent value, radiometric value, densitometric value, mass value etc., without any additional measurements or manipulations. Alternatively, the levels or differences may be expressed as a percentage or ratio of the measured value of the methylation levels to a measured value of another compound including, but not limited to, a standard or internal DNA standard, such as beta-actin. This percentage or ratio may be abnormally low, i.e., falling below a previously defined normal threshold methylation level; or this percentage or ratio may be abnormally high, i.e., exceeding a previously defined normal threshold methylation level. For example, this can be the optimal point on the ROC curve. The difference may be negative, indicating a decrease in the amount of measured levels over normal value or from a previous measurement, and the difference may be positive, indicating an increase in the amount of measured methylation levels over normal values or from a previous measurement. The difference may also be expressed as a difference or ratio of the methylation levels to itself, measured at a different point in time. The difference may also be determined using in an algorithm, wherein the raw data are manipulated.

The following paragraphs describe some of the considerations concerning refining suitable cut-off points, using a proposed, even larger study than the one presented herein:

For the 2- or 4-year models, sensitivity connotes the fraction of subjects testing positive among these who progressed to HGD or EAC within 2 or 4 years of marker measurement; while for the combined model, it is the fraction of subjects testing positive among those who ever

progressed to HGD or EAC within the observation window. Similarly, specificity is the fraction of subjects testing negative among those who did not progress to HGD or EAC within 2 or 4 years (for the 2- or 4-year models). The following null (Ho) and alternative (Hi) hypotheses will be tested in choosing 2 threshold values to define the 3 risk groups {i.e., high-, intermediate-, and low-risk groups) from a single ROC curve. High-risk cutoff:

H ₀ : Sensitivityo <= 0.30 (at Specificity _o =0.90) H ₁ : Sensitivityi >0.30 (=0.45) (at Specificityi = 0.90) (corresponding to PPVo = 0.32 vs. PPVi = 0.41 with progression rate 0.135, or (corresponding to PPV ₀ = 0.20 vs. PPVi = 0.27 with progression rate 0.075)

That is, we maximize specificity at the expense of sensitivity in choosing the high-risk cutoff value, reasoning that the FPC or false-positive cost (unnecessary endoscopies in patients who would not otherwise have been endoscoped) outweighs the FNC or false-negative cost (failure to diagnose/predict 60% of cases in a low-prevalence population [13.5% over the duration of the study]). A sensitivity of 0.30 at a specificity of 0.90 is considered minimally clinically acceptable, and we anticipate that our markers will have sensitivity of 0.45 at this specificity value. Our PPV at alternative of 0.41 (alternative hypothesis) greatly exceeds this population prevalence. Thus, any true positives detected (predicted) will represent gains in early diagnosis and can be considered a significant gain, whereas missed diagnoses (predictions) would have been missed anyway under the current standard endoscopic surveillance interval of 2-3 years.

Low-risk cutoff:

H ₀ : Specificityo <= 0.30 (@ Sensitivity ₀ =0.95) H ₁ : Specificityi > 0.30 (=0.50) (@ Sensitivityi = 0.95) (corresponding to NPV ₀ = 0.98 vs. NPVi = 0.99 with progression rate 0.135)

(corresponding to NPV ₀ = 0.987 vs. NPVi = 0.992 with progression rate 0.075) For the low-risk cutoff, we must minimize FNC (failure to diagnose HGD or cancer) because these cases would have been diagnosed under the current standard surveillance interval of 2-3 years, thus if we lengthen it to 4-6 years, we must be certain that we fail to predict as few progressors as possible. That is, a high FNC is not acceptable. Conversely, a high FPC is acceptable for the low-risk group cutoff value, because in current clinical practice, 100% of this group would have been endoscoped at 2-3-year intervals. Thus, any reduction below 100% in this group can be considered a significant gain because it represents a savings of unnecessary

surveillance endoscopies. In this case we consider a specificity at 0.30 given a sensitivity of 0.95 is minimally clinically acceptable, and we anticipate that our markers will have a specificity of 0.5 at this sensitivity.

Any of a variety of methods can be used to determine (measure) methylation levels. In one embodiment of the invention, quantitative methods are used, such as, e.g., real-time quantitative methylation-specific PCR (qMSP), pyrosequencing, or methylation microarrays (using, e.g., the Human CpG Island Microarrays and/or methylation system sold by Agilent Technologies, Santa Clara, CA and/or methods as described in Beier et al. (2007) Adv Biochem Eng Biotechnol 104, 1-11). A "methylation array," as used herein, refers to an array of probes that can be used to distinguish between methylated and unmethylated DNA (e.g., between cytosines that are, or are not, methylated).

In other embodiments of the invention (e.g., to determine if a subject has BE or EAC, but not necessarily to determine the course of development of the condition or to stratify subjects into different risk groups), non-quantitative measurement methods can be used. These include, e.g., assays based on methylated DNA immunoprecipitation, using a monoclonal antibody against 5-methylcytosine (see, e.g., Weber et al. (2005) Nat Genet 37, 853-862); Southern blotting analysis using a methylation-sensitive restriction enzyme; single nucletotide primer extension (SNuPE - Gonzalgo et al. (1997) Nuc Acids Res 25, 2532-2534); restriction landmark genomic scanning for methylation (RLGS-M); or combined bisulfite restriction analysis (COBRA - see, e.g., Xiong et al (1997) Nuc Acids Res 25, 2532-2534).

Real-time polymerase chain reactions (PCR), such as quantitative real-time PCR, can be employed in many of the assays described herein. Such assays are well-known in the art and can be practiced generally according to the known methods. See for example, Heid et al. (1996) Genome Res. 6, 986-994. Briefly, a sequence of interest (e.g., from a promoter region of the invention) is PCR amplified, using a forward PCR primer and a reverse PCR primer, in the presence of a fluorogenic probe that can distinguish between a methylated and a non-methylated version of the amplified sequence.

For quantitative methylation-specific real-time PCR (qMSP), the target DNA is pretreated before PCR amplification with an agent, such as bisulfite, that converts methylated cytosines (Cs) to uracils (U's). The fluorogenic probe is designed to recognize a sequence in which methylated Cs have been converted to U's by this procedure (or, if a control is required, to recognize a sequence in which Cs are present in those positions). The qMSP assay is designed such that the labeling moieties on the 5' and/or 3' ends of the fluorogenic probe do not

fluoresce unless PCR amplification of the sequence to which the probe binds has occurred, followed by hybridization of the probe to the amplified sequences, in which case fluorescence of the probe can be seen. The labeling moieties on both ends of a probe are fluorescent molecules, which quench one another. For simplicity, the labeling moiety on one end (e.g., the 5' end) is sometimes referred to as a "fluorophore," and the labeling moiety on the other end (e.g., the 3' end) as a "quencher." When a single stranded probe is not hybridized to a target and is free in solution, the probe molecule is flexible and folds back partially on itself, so that the quencher and the fluorophore are close together; the quencher thus prevents the probe from fluorescing. Furthermore, when a probe of the invention is hybridized to a single-stranded target, the two labeling moieties are close enough to one another to quench each other. However, without wishing to be bound by any particular mechanism, it is suggested that when the probe is hybridized to its target to form a perfect double stranded DNA molecule, a 5' to 3' exonuclease which recognizes perfect hybrids, and which is an activity of the enzyme used for PCR, cleaves the duplex, releasing the fluorophore. The fluorophore is thus separated from the quencher, and will fluoresce. The amount of detected fluorescence is proportional to the amount of amplified DNA.

In a real time PCR, the released fluorescent emission is measured continuously during the exponential phase of the PCR amplification reaction. Since the exponential accumulation of the fluorescent signal directly reflects the exponential accumulation of the PCR amplification product, this reaction is monitored in real time ("real time PCR"). Oligonucleotides used as amplification primers (e.g., DNA, RNA, PNA, LNA, or derivatives thereof) preferably do not have self-complementary sequences or have complementary sequences at their 3' end (to prevent primer-dimer formation). Preferably, the primers have a GC content of about 50% and may contain restriction sites to facilitate cloning. Amplification primers can be between about 10 and about 100 nt in length. They are generally at least about 15 nucleotides (e.g., at least about 15, 20, or 25 nt), but may range from about 10 to a full-length sequence, and not longer than 50 nt. In some circumstances and conditions, shorter or longer lengths can be used. Amplification primers can be purchased commercially from a variety of sources, or can be chemically synthesized, using conventional procedures. Suitable primers can be readily designed by conventional methods, such as inspection of known sequences of promoter regions of interest or by use of computer programs such as ClustalW from the European Bioinformatics Institute (world wide web site ebi.ac.uk/clustalw.htm). Some exemplary PCR primers are described in the Examples.

In one embodiment of the invention, rather than, or in addition to, using a fluorogenic probe that can distinguish between methylated and unmethylated cytosines, one of both of the PCR primers are designed so as to distinguish between such sequences.

Probes and conditions can be selected, using routine conventional procedures, to insure that hybridization of a probe to a sequence of interest is specific. A probe that is "specific for" a nucleic acid sequence (e.g., in a DNA molecule) contains sequences that are substantially similar to (e.g., hybridize under conditions of high stringency to) sequences in one of the strands of the nucleic acid. By hybridizing "specifically" is meant herein that the two components (the target

DNA and the probe) bind selectively to each other and not generally to other components unintended for binding to the subject components. The parameters required to achieve specific binding can be determined routinely, using conventional methods in the art. A probe that binds

(hybridizes) specifically to a target of interest does not necessarily have to be completely complementary to it. For example, a probe can be at least about 95% identical to the target, provided that the probe binds specifically to the target under defined hybridization conditions, such a conditions of high stringency.

As used herein, "conditions of high stringency" or "high stringent hybridization conditions" means any conditions in which hybridization will occur when there is at least about 85%, e.g., 90%, 95%, or 97 to 100%, nucleotide complementarity (identity) between a nucleic acid of interest and a probe. Generally, high stringency conditions are selected to be about 5 ⁰ C to 2O ⁰ C lower than the thermal melting point (T _m ) for the specific sequence at a defined ionic strength and pH. Hybridization as used according to the present invention, refers to hybridization under standard conditions used for real-time PCR to achieve amplification.

The size of a probe used to detect the methylation level of a promoter region will vary according to a variety of factors, including, e.g., the ease of preparing (e.g., synthesizing) the probe, the assay method employed to determine the methylation level, etc.

Methods for labeling probes with fluorophores are conventional and well-known. Suitable fluorescer-quencher dye sets will be evident to the skilled worker. Some examples are described, e.g., in Holland et al. (1991) Proc. Natl. Acad. ScL 88, 7276-7280; WO 95/21266; Lee et al. (1993) Nucleic Acids Research 7λ, 3761-3766; Livak et al. (1995), supra; U.S. Pat. No. 4,855,225 (Fung et al); U.S. Pat. No. 5,188,934 (Menchen et al.); PCT/US90/05565 (Bergot et al.), and others.

The fluorogenic probes described in the Examples herein function by means of FRET (fluorescence resonance energy transfer). The FRET technique utilizes molecules having a

combination of fluorescent labels which, when in proximity to one another, allows for the transfer of energy between labels. See, e.g., the Examples herein or "iQ5 Real Time PCR Detection System" Manual (Bio-Rad, Hercules, CA). Other well-known methods for the detection of real-time PCR will be evident to a skilled worker. For example, molecular beacons can be used.

Methods of PCR amplification (including qMSP), and reagents used therein, as well as methods for detecting emission spectra, are conventional. For guidance concerning PCR reactions, see, e.g., PCR Protocols: A Guide to Methods and Applications (Innis et al. eds, Academic Press Inc. San Diego, Calif. (1990)). These and other molecular biology methods used in methods of the invention are well-known in the art and are described, e.g., in Sambrook et al., Molecular Cloning: A Laboratory Manual, current edition, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, and Ausubel et al, Current Protocols in Molecular Biology, John Wiley & sons, New York, NY.

An assay of the invention can be performed in conjunction with one or more other diagnostic (predictive) methods, based, for example, on genomic information, proteomic information, histological analysis of tissue sample, or other methods known in the art. Suitable methods include, e.g., detecting low-grade or high-grade dysplasia (wherein the presence of high-grade dysplasia is further indicative that the subject has an increased risk of developing EAC); taking into consideration the age of the subject (wherein a human subject that is more than 60 years old has an increased risk of developing EAC, and wherein the older the subject, the greater the risk); or determining BE segment length (wherein a length of at least 3 cm in length is further indicative that the subject has a high risk of developing EAC, and wherein the longer the BE segment, the greater the risk).

One aspect of the invention is a kit for predicting a subject's risk for developing EAC or HGD. A skilled worker will recognize components of kits suitable for carrying out a method of the invention. The agents in the kit can encompass reagents for carrying out a method of the invention {e.g., primers and probes for qMSP of sets of promoter regions of the invention). The kit may also include additional agents suitable for detecting, measuring and/or quantitating the amount of PCR amplification. Optionally, a kit of the invention may comprise instructions for performing the method.

Optional elements of a kit of the invention include suitable buffers, control reagents, containers, or packaging materials. The reagents of the kit may be in containers in which the reagents are

stable, e.g., in lyophilized form or stabilized liquids. The reagents may also be in single use form, e.g., for the performance of an assay for a single subject.

Kits of the invention may comprise one or more computer programs that may be used in practicing the methods of the invention. For example, a computer program may be provided that calculates a methylation value in a sample from results of the determining methylation levels.

Such a computer program may be compatible with commercially available equipment, for example, with commercially available microarray or real-time PCR. Programs of the invention may take the output from microplate reader or realtime-PCR gels or readouts and prepare a calibration curve from the optical density observed in the wells, capillaries, or gels and compare these densitometric or other quantitative readings to the optical density or other quantitative readings in wells, capillaries, or gels with test samples.

In addition to the clinical uses discussed herein, kits of the invention can be used for experimental applications.

In the foregoing and in the following examples, all temperatures are set forth in uncorrected degrees Celsius; and, unless otherwise indicated, all parts and percentages are by weight.

EXAMPLES Example I. Studies of methylation levels and frequencies of nel-like 1 (NELLl), tachykinin- 1 (TACl), somatostatin (SST), and λ-kinase anchorine yrotein 12 (AKAP12)

Methylation levels and frequencies of the above genes were studied, using real-time quantitative methylation-specific PCR (qMSP), in 259 endoscopic esophageal biopsy specimens of differing histologies. Prevalences were determined in each of the following histological categories: NE, BE, LGD, HGD or EAC. The methods used in these studies are, in general, similar or identical to those described in Example II. For details of these studies, see, e.g., the following references, all of which are incoiporated by reference herein in their entirety: Jin et al

(2007) Oncogene 26, 6332-6340; Jin et al (2007) Clin Cancer Res 13, 6293-6300; Jin et al.

(2008) Cancer 112, 43-49; Jin et al (2008) Cancer 112, 43-49; and Jin et al (2008) Cancer Epidemiol Biomarkers Prev 17, 111-117. Among 10 genes evaluated, the five genes noted above were methylated early and often in BE-associated neoplastic progression.

Example II. Promoter hypermethylation of CDH13 is a common, early event in human esophageal adenocarcinogenesis and correlates with clinical risk factors

CDH13 (also known as H-cadherin and T-cadheriή), a member of the cadherin gene superfamily, was isolated and has been mapped to 16q24, a locus that frequently undergoes deletion in human cancers, including esophageal carcinoma. In contrast to other known cadherins such as E-cadherin, N-cadherin, and P-cadherin, which are transmembrane proteins, CDH 13 lacks conventional transmembrane and cytoplasmic domains and is attached to the plasma membrane through a glycosyl phosphatidyl inositol anchor. Several studies have suggested that CDH 13 functions as a tumor suppressor gene and possesses potent antitumor activity in several human cancers both in vitro and in vivo. Hypermethylation of CDHl 3 has been described in many human cancers, including ESCC. Prior to the present studies, however, hypermethylation of CDH13 in precancerous lesions such as Barrett's metaplasia (BE), as well as in BE-associated EAC, is an area that had not been explored.

This Example describes studies of hypermethylation of the promoter region of CDH 13 in Barrett' s-associated esophageal adenocarcinogenesis. 259 human esophageal tissues were examined for CDH13 promoter hypermethylation by real-time methylation-specific PCR. CDH13 hypermethylation showed discriminative receiver-operator characteristic curve profiles, sharply demarcating esophageal adenocarcinoma (EAC) from esophageal squamous cell carcinoma (ESCC) and normal esophagus (NE) (p<0.0001). CDHl 3 normalized methylation values (NMV) were significantly higher in Barrett's esophagus (BE), dysplastic BE (D), and EAC than in NE (p<0.0000001). CDH 13 hypermethylation frequency was 0% in NE but increased early during neoplastic progression, rising to 70% in BE, 77.5% in D, and 76.1% in EAC. Both CDH13 hypermethylation frequency and its mean NMV were significantly higher in BE with than without accompanying EAC. In contrast, only five (19.2%) of 26 ESCCs exhibited CDH13 hypermethylation. Furthermore, both CDH13 hypermethylation frequency and its mean NMV were significantly higher in EAC than in ESCC, as well as in BE or D vs. ESCC. Interestingly, mean CDH13 NMV was significantly lower in short-segment than in long- segment BE, a known clinical risk factor for neoplastic progression. Similarly, BE segment length was significantly lower in specimens with unmethylated than with methylated CDH 13 promoters. 5-aza-2'-deoxycytidine treatment of OE33 EAC and KYSE220 ESCC cells reduced CDH13 methylation and increased CDHl 3 mRNA expression. Our results reveal that promoter hypermethylation of CDH13 is a common event in EAC but not in ESCC and occurs early

during BE-associated esophageal neoplastic progression, correlating with clinical criteria associated with neoplastic progression risk.

A. Materials and Methods (1) Tissue samples. The 259 specimens examined in the current study comprised 66 from normal esophagus (NE), 60 of non-dysplastic Barrett's metaplasia {BE, including 36 obtained from patients with BE alone (Ba) and 24 from patients with BE accompanied by EAC (Bt)}, 40 from dysplastic BE {D, including 19 low-grade (LGD) and 21 high-grade (HGD) }, 67 EACs, and 26 ESCCs. All patients provided prior written informed consent under a protocol approved by the Institutional Review Boards at the University of Maryland School of Medicine, the Baltimore Veterans Affairs Medical Center, and the Johns Hopkins University School of Medicine. Biopsies were obtained using a standardized biopsy protocol as previously described (Schulmann et al. (2005) Oncogene 24, 4138-4148). Research tissues were taken from grossly apparent BE epithelium or from mass lesions in patients manifesting these changes at endoscopic examination, and histology was confirmed using parallel aliquots culled from identical locations at endoscopy. All research biopsy specimens were stored in liquid nitrogen prior to DNA extraction. Clinicopathologic characteristics are summarized in Table 2.

Table 2. Clinicopathologic characteristics and methylation status of CDH13 in human esophageal tissues

Number Age (year) NMV ² Methylation Status (cutoff 0.06) ³

Clinical characteristics ¹ of samples mean mean P Frequency UM M P

Histology Normal esophagus 66 64.3 0.0054 0% 66 0

BE 60 63.7 0.3122 ^s < 0.00001 ^* * 70% 18 42 Ba 36 62.5 0.2623 58.3% 15 21 < O.OSí Bt 24 65.5 0.3871 ^s < 0.05 87.5% 3 21

Dysplasia in Barrett's esophagus 40 65.3 0.3383 ^s < 0.00001 ^* * 77.5% 9 31 Low-grade dysplasia 19 65.3 0.2833 ^s < 0.000001 ^* 78.9% 4 15 NS ^f High-grade dysplasia 21 65.2 0.388 ^s < 0.000001 ^* 76.2% 5 16

EAC 67 65.1 0.2392 ^s < 0.000001 ^* " 76.1% 16 51 < 0.0001*

ESCC 26 62.5 0.0458 ^s < 0.001 ^* 19.2% 21 5

Barrett's segment of Ba Short-segment ( <3cm ) 14 62.3 0.131 ^s < 0.01 28.6% 10 4 < 0.01 ^f Long-segment ( >=3cm ) 16 62.8 0.4071 87.5% 2 14

Stage of EAC patients I 7 63 0.3081 ¹ NS 85.7% 1 6 NS ^f II 15 65.2 0.2408 73.3% 4 11 III 25 64.6 0.2111 72% 7 18 IV 7 66.3 0.2921 100% 0 7

Lymph node metastasis in EAC patients Negative 25 64.9 0.2751 ^S NS 75% 5 20 NS* Positive 25 64.6 0.2277 76% 6 19

Smoking status of EAC patients

Never 6 58.5 0.2984 ¹ NS 100% 0 6 NS ^f

Former 24 68.5 0.2143 79.2% 5 19 Current 13 60.8 0.2561 76.9% 3 10

Alcohol drinking status of E4C patients Never 16 65.3 0.2209 ¹ NS 7755%% 4 1122 NS ^' Former 15 63 0.2524 86.7% 2 13 Current 10 65.7 0.2427 80% 2 8

1, BE, Barrett's metaplasia; Ba, BE from patients with Barrett's alone; Bt, BE from patients with Barrett's accompanied by EAC; EAC, esophageal adenocarcinoma; ESCC, esophageal squamous cell carcinoma.

2, NMV: normalized methylation value; $, Mann-Whitney U test; % comparisons made to normal esophagus; #. comparisons made to ESCC; f , Kruskal-Wallis test.

3, UM, unmethylated; M, methylated; f, Fisher's exact test; J, Chi-square for independence test. NS, not significant.

(2) Cell lines. OE33 EAC and KYSE220 ESCC cells were cultured in 47.5% RPMI 1640, 47.5% F-12 supplemented with 5% fetal bovine serum.

(3) DNA and RNA extraction Genomic DNA was extracted from biopsies and cultured cells using a DNeasy Tissue Kit

(Qiagen, Valencia, CA). Total RNA was isolated from cultured cells using TRIzol reagent (Invitrogen, Carlsbad, CA). DNAs and RNAs were stored at -8O ⁰ C prior to analysis.

(4) Bisulfite treatment and real-time methylation-specific PCR

One ug DNA was treated with bisulfite to convert unmethylated cytosines to uracils prior to MSP using an EpiTect Bisulfite Kit (Qiagen, Valencia, CA). Promoter methylation levels of

CDHlS were determined by real-time quantitative MSP with an ABI 7900 Sequence Detection

System (Applied Biosystems, Foster City, CA), using primers and probes as follows: CDHB- forward: 5'-TTTGGGAAGTTGGTTGGTTGGC-S' (SEQ ID NO: 17); CDHl 3 -reverse: 5'-

ACTAAAAACGCCCGACGACG-3' (SEQ ID NO: 18) and probe: 5'- TATGTTTAGTGTAGTCGCGTGTATGAATGAA -3' (SEQ ID NO: 19). β-actin was used for normalization of data. Primers and probe for β-actin were the same as previously reported (Jin et al. (2007) Oncogene 26, 6332-6340). A standard curve was generated using serial dilutions of

CpGenome Universal Methylated DNA (CHEMICON, Temecula, CA). Normalized methylation value (NMV) was defined as follows: NMV = (CDHB-SICDHn-FM)I(ACTB-SZACTB-FM), where CDH13-S and CDHl 3-FM represent CDH13 methylation levels (derived from the standard curve) in sample and fully methylated DNAs, respectively, while ACTB-S and ACTB-

FM correspond to β-actin in sample and fully methylated DNAs, respectively.

(5) Real-time quantitiative RT-PCR

To determine CDH 13 niRNA levels, one-step real-time quantitative RT-PCR was performed using a Qiagen QuantiTect Probe RT-PCR Kit (Qiagen, Hilden, Germany) and an

ABI 7900 Sequence Detection System (Applied Biosystems, Foster City, CA). Primers and probe for CDHB were as follows: CDHl 3-forward: 5'ATGTTGGCAAGGTAGTCGATAGTG-

3' (SEQ ID NO:20); CDHl 3-reverse: 5'- ACGCTCCCTGTGTTCTCATTG -3' (SEQ ID NO:21) and probe: 5'- CCAGAAAGGTCCAAGTTCCGGCTCACT -3 (SEQ ID NO:22). β-actin was used for normalization of data. Primers and probe for β-actin were the same as previously reported (Jin et al. (2007) Oncogene 26, 6332-6340). A standard curve was generated using serial dilutions of qPCR Reference Total RNA (Clontech, Mountainview, CA). Normalized mRNA value (NRV) was calculated according to the following formula for relative expression of target mRNA: NRV = (TarS/TarC)/(ACTB-S/ACTB-C, where TarS and TarC represent levels of target gene mRNA expression (derived from the standard curve) in sample and control mRNAs, respectively, while ACTB-S and ACTB-C correspond to amplified ACTB levels in sample and control mRNAs, respectively.

(6) 5-Aza-dC treatment of esophageal cancer cell lines

To determine whether CDH13 inactivation was due to promoter hypermethylation in esophageal cancer, 2 esophageal cancer cell lines (KYSE220 and OE33) were subjected to 5- Aza-dC (Sigma, St. Louis, MO) treatment as previously described (Bender et al. (1999) MoI Cell Biol 19, 6690-6698; Shibata et al. (2002) Cancer Res 62, 5637-5640). Briefly, 1 x 10 ⁵ cells/ml were seeded onto a 100 mm dish and grown for 24 h. Then, 1 μl of 5mM 5-Aza-dC per ml of cells was added every 24 hours for 4 days. DNAs and RNAs were harvested on day 4.

(7) Data analysis and statistics

Receiver-operator characteristic (ROC) curve analysis (Hanley et al. (1982) Radiology 143, 29-36) was performed using NMVs for the 67 EAC, 26 ESCC and 66 NE specimens by

Analyse-it© software (Version 1.71, Analyse-it Software, Leeds, UK). Using this approach, the area under the ROC curve (AUROC) identified optimal sensitivity and specificity levels at which to distinguish normal from malignant esophageal tissues (NE vs. EAC), yielding a corresponding NMV threshold with which to dichotomize the methylation status of CDHl 3. The threshold NMV value determined from this ROC curve was applied to determine the status of

CDH13 methylation in all tissue types included in the study. For all other statistical tests,

Statistica (version 6.1; StatSoft, Inc., Tulsa, OK) was employed. Differences with p<0.05 were considered significant.

B. Results

(1) CDH13 promoter hvpermethylation in esophageal tissues

Promoter hypermethylation of CDH 13 was analyzed in 66 NE, 60 BE (including 36 Ba and 24 Bt), 40 D (including 19 LGD and 21 HGD), 67 EAC, and 26 ESCC. CDH 13 promoter

hypermethylation showed highly discriminative ROC curve profiles and AUROCs, clearly distinguishing both EAC and ESCC from NE (Figure 9A and 9B), as well as EAC from ESCC (Figure 9C).

The cutoff NMV for CDH 13 (0.06) was identified from the ROC curve (EAC vs. NE) to achieve the highest possible sensitivity while maintaining 100% specificity. Mean NMV and frequency of CDHl 3 hypermethylation for each tissue type are shown in Table 2.

NMVs of CDHl 3 were significantly higher in ESCC, EAC, D, HGD, LGD, BE, Ba and Bt than in NE (p<0.001, Mann-Whitney U test). The frequency of CDH13 hypermethylation was significantly higher in BE (70%), D (77.5%), and EAC (76.1%) than in N (0%; p < 0.0001, p < 0.0001 and p < 0.0001, respectively; Fisher's exact test). Interestingly, both CDH13 hypermethylation frequency and mean NMV were significantly higher in Bt than in Ba (87.5% vs. 58.3%, p = 0.021 and 0.3871 vs. 0.2623, p = 0.045, respectively). The mean CDHl 3 NMV in EAC (0.2722) was significantly higher than that in matching NE (0.0034) for 27 cases in which matching NE and EAC were available (pO.OOOOl, Wilcoxon matched pairs test). In contrast to EAC, only five (19.2%) of 26 ESCCs manifested hypermethylation of CDHIi. There was no significant difference in mean CDHl 3 NMV between tumor and normal tissue in 13 cases for which matching ESCC (0.0337) and NE (0.0131; p = 0.6, Wilcoxon matched pairs test) were available. Both CDH13 hypermethylation frequency and mean NMV were significantly higher in EAC than in ESCC (76.1% vs. 19.2%, p<0.0001 and 0.2392 vs. 0.0458, pO.OOOl, respectively), as well as in D vs. ESCC (77.5% vs. 19.2%, p<0.0001 and 0.3383 vs. 0.0458, pO.OOOl, respectively) and in BE vs. ESCC (70% vs. 19.2%, pO.OOOl and 0.3122 vs. 0.0458, pO.OOOl; Table 2).

According to generally accepted criteria, BE was defined as long-segment (LSBE) if it was equal to or greater than 3 cm in length, or short-segment (SSBE) if less than 3 cm. The mean NMV of CDH 13 was significantly higher in LSBE than in SSBE (0.4071 vs. 0.131; pθ.01, Student's t-test, Table 2 and Figure 10A). Similarly, segment lengths of BEs with methylated CDH 13 promoters (mean = 5.83 cm) were significantly longer than segment lengths of BEs with unmethylated CDH13 promoters (mean = 1.83 cm; p<0.001, Student's t-test; Figure 10B), and the frequency of CDHl 3 hypermethylation was significantly higher in LSBE than in SSBE (87.5% vs. 28.6%; pO.Ol, Fisher's exact test; Table 2).

No significant associations were observed between CDH13 promoter hypermethylation and patient age (data not shown), survival (log-rank test, data not shown), tumor stage, lymph node metastasis, smoking, or alcohol consumption (Table 2).

(2) CDHlS methylation and rtiRNA levels in esophageal cancer cell lines pre- and post- 5-Aza-dC treatment

KYSE220 ESCC and OE33 EAC cells were subjected to 5-Aza-dC treatment. After 5- Aza-dC treatment, the NMV of CDElS was diminished and the mRNA level of CDHlS was increased in both KYSE220 and OE33 cells (Figure 11).

C. Discussion

In this Example, we systematically investigated hypermethylation of the CDHlS gene promoter in cell lines and primary human esophageal lesions of contrasting histological types and grades by qMSP. Our results demonstrate that CDHlS promoter hypermethylation occurs frequently in human EAC, but not in ESCC. In addition, our data show that CDHlS hypermethylation increases early during esophageal adenocarcinogenesis, from 0% in NE to 58.3% in BE, 77.5% in D, and 76, 1% in EAC. These results imply that hypermethylation of CDHlS occurs early in most subjects, that its frequency increases during adenocarcinogenesis, and that it is tissue-specific {i.e., common in EAC but rare in ESCC). Further evidence supporting this tissue specificity is provided by ROC curves, which clearly distinguished EAC from ESCC. Similarly, support for tissue specificity is evident from the finding that both CDHlS hypermethylation frequency and mean CDHlS NMV were significantly higher in EAC than in ESCC. In addition, the low frequency (19.2%) of CDHlS hypermethylation in ESCC, as determined in the current study, is consistent with previous findings by other groups. Thus, CDHlS hypermethylation appears to constitute a critical event unique to human EAC.

Several studies have suggested that methylation of certain genes may occur as a field change and may be associated with an increased risk of malignant progression. CDKN2A, ESRl, and MYODl were methylated only in BE from patients who possessed dysplasia or cancer in other regions of their esophagus, but not in patients with no evidence of progression beyond BE, while CALCA, MGMT, and TIMPS were methylated more frequently in normal stomach, normal esophageal mucosa and intestinal metaplasia from patients with distant dysplasia or esophageal cancer than from patients without dysplasia or cancer. Previously, we demonstrated that hypermethylation oiplό, RUNXS, and HPPl in BE or LGD may represent independent risk factors for the progression of BE to HGD or EAC (Schumann et al. (2005) Oncogene IA, 4138- 4148). Example I herein shows that both hypermethylation frequency and NMV of the nel-like 1, tachykinin-1, somatostatin and AKAP12 genes were higher in BE with accompanying EAC than in BE without accompanying EAC. Interestingly, both CDHlS hypermethylation frequency

and level were significantly higher in BE with than without accompanying EAC in the current study, suggesting that CDHl 3 is a biomarker of more ominous disease lurking nearby.

In the present Example, we also correlated CDH13 methylation with clinicopathologic features. Despite some degree of controversy regarding the length of the BE segment as a predictive factor in BE progression, it is likely that this clinical parameter is an important predictor of neoplastic progression. In the Seattle Barrett's Esophagus Project, BE segment length was not related to cancer risk in a prospective cohort study of 309 Barrett's patients (p> 0.2); however, when patients with HGD at entrance were excluded, a strong trend was observed, with a 5 cm difference in length associated with a 1.7-fold increase in cancer risk (95% CI, 0.8- 3.8-fold). Significant differences in the frequency of both dysplasia and EAC were observed between SSBE and LSBE, at 8.1% vs. 24.4% for dysplasia (p<0.0001) and 0% vs. 15.4% for EAC (p<0.0005). In a comprehensive prospective study of 889 consecutive patients, the prevalence of dysplasia and cancer differed significantly in patients with SSBE vs. LSBE, More recently, a significantly increased risk of progression to HGD or EAC with LSBE after a mean follow-up of 12.7 years was reported. In the studies described in Example I, the nel-like 1, tachykinin- 1, somatostatin and AKAP12 genes were significantly more hypermethylated in LSBE than in SSBE. Notably, in the current study, CDH13 methylation also showed a strong relationship to BE segment length. The mean NMV of CDH 13 was significantly higher in LSBE than in SSBE. Similarly, the length of the BE segment was significantly greater in specimens with methylated than with unmethylated CDH13 promoters. Thus, CDH13 hypermethylation may constitute a molecular correlate of BE segment length, as well as a harbinger of nearby neoplastic disease. These results also suggest that epigenetic alterations, which may account for some of the biologic behavior of BE, clearly differ between LSBE and SSBE, suggesting a need for further large-scale studies. In accordance with previous findings in other primary cancer cell types, we observed that methylation of CDHl 3 in EAC and ESCC cancer cell lines was associated with silenced or reduced expression of CDHl 3 mRNA. Treatment with 5-Aza-dC restored mRNA expression and reversed CDH 13 methylation in these cells. Restoration of CDH 13 mRNA expression by demethylating agent treatment implies that DNA hypermethylation was responsible for silencing of CDHl 3.

In summary, findings of this Example suggest that hypermethylation of the CDH 13 promoter is a common event in human esophageal adenocarcinogenesis, occurs early during Barrett' s-associated esophageal carcinogenesis, and is associated with clinical risk factors of

progression. In addition, CDHl 3 hypermethylation is uncommon in human ESCC, thus making it a potential cell type-specific biomarker for EAC.

Example III. A multicenter, double-blinded validation study of methylation biomarkers for progression prediction in Barrett's esophagus

Esophageal adenocarcinoma risk in Barrett's esophagus (BE) is increased 30- to 125 -fold versus the general population, yet neoplastic progression occurs rarely in BE. Molecular biomarkers would stratify patients for more efficient surveillance endoscopy and improve early detection of progression. We therefore performed a retrospective, multicenter, double-blinded validation study of 8 BE progression prediction methylation biomarkers. Progression or nonprogression were determined at 2 years (tier 1) and 4 years (tier 2). Methylation was assayed in 145 nonprogressors (NPs) and 50 progressors (Ps) using real-time quantitative methylation- specific PCR. Ps were significantly older than NPs (70.6 vs. 62.5 years, p < 0.001). We evaluated a linear combination of the 8 markers, using coefficients from a multivariate logistic regression analysis. Areas under the ROC curve (AUCs) were high in the 2-, 4-year and combined data models (0.843, 0.829 and 0.840; p<0.001, p<0.001 and p<0.001, respectively). In addition, even after rigorous overfitting correction, the incremental AUCs contributed by panels based on the 8 markers plus age vs. age alone were substantial (δ-AUC = 0.152, 0.114 and 0.118, respectively) in all three models. A methylation biomarker-based panel to predict neoplastic progression in BE has clinical value in improving both the efficiency of surveillance endoscopy and the early detection of neoplasia.

A. Materials and Methods

(1) Definition of Barrett's esophagus progressor and nonprogressor patients and sample collection. Progressors (Ps) and nonprogressors (NPs) were defined as described previously (Montgomery et al. (2001) Hum Pathol 32, 379-388). Ps were considered both as a single combined group, and in two tiers: progression within 2 years (tier 1) or 4 years (tier 2). 195 BE biopsies (145 NPs and 50 Ps) were obtained from 5 participating centers: the Mayo Clinic at Rochester/Jacksonville, the University of Arizona, the University of North Carolina, and Johns Hopkins University. All patients provided written informed consent under a protocol approved by Institutional Review Boards at their institutions. Biopsies were taken using a standardized biopsy protocol (Corley et al. (2002) Gastroenterology 122, 633-640; Montgomery et al. (2001, supra). Clinicopathologic features are summarized in Table 3.

Table 3. Clinicopathologic characteristics in 50 progressors and 145 non-progressors

Progressor (N=50) Non-Progressor (N= 145) p value

Age (years, mean±SD)* 70 6±9.1 62.5±12.3 <0.001

Gender

Male 46 (92%) 113 (78%) NS

Female 4 (8%) 32 (22%)

Race

White 50 (100%) 144 (99%) NS

Unknown 0 (0%) 1 (1%)

Index of Histology

BE 26 (52%) 87 (60%) NS

LGD 22 (44%) 58 (40%)

Unknown 2 (4%) 0 (0%)

Barrett's length (cm, mean±SD ) 6.0±3.3 5.4±3.6 NS

Body mass index (kg/m )

Normal weight (18.5-24.9) 10 (20%) 27 (19%) NS

Overweight (25-29.9) 15 (30%) 53 (37%)

Obese (30+) 20 (40%) 53 (37%)

Unknown 5 (10%) 12 (8%)

Smoking status*

Yes 4 (8%) 15 (10%) NS

No 46 (92%) 124 (86%)

Unknown 0 (0%) 6 (4%)

Alcohol Use*

Yes 15 (30%) 43 (30%) NS

No 35 (70%) 94 (65%)

Unknown 0 (0%) 8 (6%)

ASA'NSAID use*

Yes 17 (34%) 61 (42%) 0.039**

No 33 (66%) 81 (56%)

Unknown 0 (0%) 3 (2%)

PPI or H2 Blocker*

Yes 40 (80%) 123 (85%) NS

No 10 (20%) 21 (14%)

Unknown 0 (0%) 1 (1%)

Family History of BE, LGD, HGD orEAC

Yes 3 (6%) 11 (8%) NS

No 47 (94%) 126 (87%)

Unknown 0 (0%) 8 (6%)

SD, standard deviation; BE, Barrett's metaplasia; LGD, low-grade dysplasia; HGD, high-grade dysplasia, EAC, esophageal adenocarcinoma; ASA, acetylsalicylic acid; NSAID, non-steroidal anti-inflammatory drug; PPI, proton pump inhibitor; NS, not significant. *at time of procedure; ** p-value is adjusted for age

(2) Bisulfite Treatment and Real-Time Quantitative Methylation-Specific PCR (qMSP) Bisulfite treatment was performed and promoter methylation levels of 8 genes (pl6, HPPl, RUNX3, CDH13, TACl, NELLl, AKAP12 and SST) were determined by qMSP on an ABI 7900 Sequence Detection (Taqman) System, as described in Montgomery et al. (2001, supra), β-actin was used for normalization. Primers and probes for qMSP are described in Table 4. In this table, the forward primer sequences, reading from top to bottom, are SEQ ID NOs: 23- 31; the reverse primer sequences, reading from top to bottom, are SEQ ID NOs: 32-40; and the TaqMan probe sequences, reading from top to bottom, are SEQ ID NOs: 41-49.

A standard curve was generated using serial dilutions of CpGenome Universal Methylated DNA (CHEMICON, Temecula, CA). A normalized methylation value (NMV) for each gene of interest was defined as described in Montgomery et al. (2001, supra). Wetlab analysts and all SJM laboratory personnel were blinded to specimen P or NP status. (3) Data Analysis and Statistics

Associations between progression status and patient characteristics were tested using Student's t-test or Chi-squared testing. Relationships between biomarkers and patient progression status were examined using Wilcoxon rank-sum testing.

To evaluate the predictive utility of the markers, we constructed receiver operating characteristic (ROC) curves. ROC curve analyses were first conducted on individual markers, then in combination to determine whether a panel performed better than any single marker. Our algorithm rendered a single composite score, using the linear predictor from a binary regression model justified under the linearity assumption (Jin et al. 2007) CHn Cancer Res .13, 6293-6300). The predictive accuracy of composite scores was evaluated based on a resampling algorithm: we randomly split data into a learning set containing 2/3 and a test set including 1/3 of observations. The combination rule derived from the learning set produced two ROC curves, from the learning and test sets, respectively. Vertical differences between these two ROC curves yielded the overestimation of sensitivities at given specificities. This procedure was repeated 200 times, and these 200 differences were averaged to estimate the expected overfitting. We also utilized predictiveness curves (Jin et al. (2008) Cancer \ \2, 43-49) to display risk distribution as a function of the combined marker in the population. This curve represents a plot of risk associated with the v ^th quantile of the marker, P{D=1|Y =F ^~1 (v)} vs. v, with F(-) the cumulative distribution of the marker. These plots display population proportions at different risk levels more clearly than do other metrics (like ROC curves). Since a case-control sample was studied, we used an external progression prevalence rate to calculate risk in the targeted screening population. To calibrate for future samples, a shrinkage coefficient estimated from the logistic regression model was applied to the linear predictors from which risk was calculated (Jin et al. (2008) Cancer Epidemiol Biomarkers Prev JJ, 111-117).

All analyses were performed in R (see the world wide web site www.r-project.org). Statistical data analysts were blinded to the identities of the 8 biomarkers.

B. Results

(1) Clinical characteristics

Ps vs. NPs did not differ significantly by gender, body mass index, BE segment length, LGD patient percentage,family history of BE, LGD, HGD or EAC, cigarette smoking, or alcohol use; however, Ps were significantly older than NPs (70.6 vs. 62.5 years; p< 0.001, Student's t test; Table 3). Samples consisted of one biopsy from each of 50 Ps and 145 NPs (195 patients) in the combined model. In the 2-year model, we redefined progressors whose interval from index to final procedure exceeded 2 years as nonprogressors, yielding 36 Ps and 159 NPs. In the 4-year model, we redefined progressors whose interval from index to final procedure exceeded 4 years as nonprogressors, yielding 47 Ps and 148 NPs.

(2) Univariate analyses

NMVs of HPPl, plό and RUNX3 were significantly higher in Ps vs. NPs by Wilcoxon test (0.456, 0.138, and 0.104 vs. 0.273, 0.069 and 0.063; p = 0.0025, 0.0066 and 0.0002, respectively). The remaining 5 markers did not differ significantly in Ps vs. NPs (Table 5).

We further assessed the classification accuracy of single markers using ROC curve analyses. Areas under ROC curve (AUCs) for HPPl, plό and RUNX3 were all significantly greater than 0.50, (Table 6).

Table 6. Receiver-operator characteristic curve analysis for 8 methylation biomarkers in 50 progressors and 145 non-progressors

Biomarker AUC (95% confidence interval)

HPPl 0.647 (0.556, 0.739) pl6 0.628 (0.534, 0.722)

RUNX3 0.671 (0.586, 0.756)

CDHlS 0.515 (0.409, 0.622)

TACl 0.571 (0.470, 0.673)

NELLl 0.516 (0.420, 0.611)

AKAP12 0.561 (0.451, 0.672)

SST 0.506 (0.401, 0.611)

AUC, area under the receiver-operator characteristic curve

(3) Logistic regression analyses of the 8-marker panel

We then combined all 8 markers by performing logistic regression and treating them as linear predictors (Table 7, Figure 12)

All models exhibited high AUCs (0.843, 0.829 and 0.840, respectively; Table 7, Figures 12A-12C. We performed overfitting correction based on 3-fold cross-validation and 200 bootstraps. The overfitting-corrected AUCs remained high (0.745, 0.720 and 0.732, respectively), while shrinkages from overfitting correction were modest (0.098, 0.109 and 0.108, respectively) in the three models (Table 7, Figures 12A- 12C).

We also explored the incremental AUC value contributed by an 8-marker-plus-age panel to that of age alone (Table 8, Figure 12). The AUCs of the 8-marker-plus-age panels in the three models (0.858, 0.850 and 0.855, respectively) were higher than those of age alone (0.604, 0.630 and 0.635, respectively; Table 8, Figures 12D-12F).

Overfitting-corrected AUCs remained high (0.756, 0.744 and 0.753, respectively), and increments contributed by the age-plus-biomarker panel vs. age were substantial (0.152, 0.114 and 0.118, respectively) in the three models (Table 8, Figures 12D-12F).

(4) Sensitivity and specificity of the 8-marker panel

While maintaining high specificity to minimize false-positive results, our model still predicted a number of new early diagnoses, i.e., diagnoses that would not have occurred earlier without the panel (Table 9). While maintaining specificity at 0.9 or 0.8, sensitivities (0.443 and 0.629 for the combined model, 0.607 and 0.721 for the 2-year model, and 0.465 and 0.606 for the 4-year model, respectively) were above or approached 50% in all three models based on the 8-marker panel alone. Furthermore, at 0.9 or 0.8 specificities, sensitivities (0.457 and 0.757 for the combined model, 0.536 and 0.786 for the 2-year model, and 0.450 and 0.724 for the 4-year model, respectively) exceeded or approached 50% in all models based on the 8-marker-plus-age panel.

From the foregoing description, one skilled in the art can easily ascertain the essential characteristics of this invention, and without departing from the spirit and scope thereof, can make changes and modifications of the invention to adapt it to various usage and conditions and to utilize the present invention to its fullest extent. The preceding preferred specific embodiments are to be construed as merely illustrative, and not limiting of the scope of the invention in any way whatsoever. The entire disclosure of all applications, patents, and publications (including provisional patent applications 61/066,281, filed February 19, 2008; 61/131,748, filed June 11, 2008; and 61/132,418, filed June 18, 2008) cited above and in the figures are hereby incorporated in their entirety by reference.

Atty Doc: 02240-264913; (P10304-02)

Table 4

Supplemental) Table 4. Primer and probe sequences

Gene Forward primer sequence (5'-3') Reverse primer sequence (5'o) TaqMan probe sequence (5'o)

HPPl GTTATCGTCGTCGTTTTTGTTGTC GACTTCCGAAAAACACAAAATCG CCGAACAACGAACTACTAAACATCCCGCG pl6 TGGAATTTTCGGTTGATTGGTT AACAACGTCCGCACCTCCT ACCCGACCCCGAACCGCG

RUNX3 GGGTTTTGGCGAGTAGTGGTC ACGACCGACGCGAACG CGTTTTGAGGTTCGGGTTTCGTCGTT

CDH13 TTTGGGAAGTTGGTTGGTTGGC ACTAAAAACGCCCGACGACG TATGTTTAGTGTAGTCGCGTGTATGAATGAA

Tachykinin-^ 1 GGCGGTTAATTAAATATTGAGTAGAAAGTCGC AAATCCGAACGCGCTCTTTCG AGAATGTTACGTGGGTTTGGAGGTTTAAGGAG

Nel-like 1 GGGTTTTTAGACGAACGTAGTCGTAGC CCACCGCTAACGCGACGA ACGACGTAAAACTCGAAACCCGACCC

AKAP12 GTTTTGTTAGAAACGGGAGGCGTTC GAAACCAAAAACGCTACGACGCG TCGGGTGGGCGGTTGTTTGGATT

Somatostatir i GGGGCGTTTTTTAGTTTGACGT AACAACGATAACTCCGAACCTCG AACGACTTATATACTCTCAACCGTCTCCCTCTA β-actin TGGTGATGGAGGAGGTTTAGTAAGT AACCAATAAAACCTACTCCTCCCTTAA ACCACCACCCAACACACAATAACAAACACA

Atty Doc: 02240-264913; (P10304-02)

Table 7, Logistic regression and overfitting correction for the 2-year, 4-year and combined models AUCi (95% CI) p-value OfAUC ₁ AUC ₂ (95% CI) p-value OfAUC ₂ Shrinkage (AUCi-AUC ₂ )

2-year model 0.843(0.763,0.924) <0.001 0.745(0.685,0.875) 0.001

4-year model 0.829(0.773,0.907) 4001 0.720(0.694,0.856) 0.004 0.109 imbined model 0.840(0.773,0.907) 4001 0.732 (0.697,0.868) 0.002 0.108

AUC, area under the receiver-operator characteristic curve; AUC ₁ , original AUC; AUC ₂ , overling corrected AUC; Cl, confidence interval.

Table 8. Incremental values above age alone for the 2-year, 4-year and combined models AUCi (95% CI) AUC ₂ (95% CI) AUC ₃ (95% CI) Increment over age (AUC ₃ -AUC _I )

2-year model 0.604(0.491,0.718) 0.858(0.783,0.932) 0.756(0.699,0.884) 0.152

4-year model 0.630(0.526,0.735) 0.850(0.784,0.917) 0.744(0.719,0.871) 0.114 combined model 0.635 (0.532, 0.737) 0.855(0.790,0.919) 0.753(0.729,0.878) 0.118

AUC, area under the receiver-operator characteristic curve; AUCi, AUC of age alone;

AUCi, AUC of age plus markers; AUC^, overfitting corrected AUC of age plus markers; CI, confidence interval.

Table 9. Specificity and sensitivity for 2-year, 4-year and combined models

Specificity (95% CI) at sensitivity- Sensitivity (95% CI); at specificity-

0.9 0.8 0.9 0.8

Combined model

Age 0.260(0.162, 0.425) 0.390(0.240,0.508) 0.221 (0.054,0.448) 0.371 (0.204,0.532)

Marker combination 0.567(0.413,0,849) 0.724(0.574,0.914) 0,443 (0.350,0.838) 0.629(0,527,0,941)

Age+marker combination 0.576(0.484,0.867) 0.781 (0.647, 0.944) 0.457(0.372,0.869) 0.757 (0.598, 0.964)

2-year model

Age 0.205(0.138,0.389) 0.351(0.197,0.484) 0.176 (0.021,0.426) 0.354(0.172,0.538)

Marker combination 0.547(0.436,0.873) 0.757(0.595,0.935) 0.607(0.393,0.870) 0.721 (0.593,0.969)

Age+marker combination 0.615(0.474,0.918) 0.786(0.652,0.956) 0.536 (0.400,0.934) 0.786(0.600,0.987)

4-year model

Age 0.249(0.158,0.430) 0.384(0.229,0.506) 0.232 (0.038,0.467) 0.382(0.214,0.541)

Marker combination 0.523 (0.426,0,835) 0.704(0.579,0.909) 0,465 (0.346,0.814) 0.606(0,545,0,941)

Age+marker combination 0.574(0.488,0,864) 0.757(0.649,0.940) 0,450 (0.385,0.885) 0.724(0,600,0,963)

CI, confidence interval

(5) Risk stratification of BE patients

ROC curves derived from these marker-based models were used to establish thresholds to stratify patients into risk categories. This procedure was performed to identify high-risk (HR) individuals for more frequent endoscopic screening. The threshold above which patients were classified as HR was chosen at specificity = 90%, to minimize false-positive, unnecessary endoscopies (type II error). A second threshold was established to identify low-risk (LR) individuals for less frequent endoscopic screening. The threshold below which patients were classified as LR was chosen at sensitivity = 90%, to minimize false-negative, missed HR individuals (type I error). Based on the combined P and NP classification, we classified patients as LR with a threshold that corresponded to 90% true-positives and 43% false-positives; the HR group was defined using a threshold that yielded 43% true-positives and 10% false-positives. Assuming progression to HGD and/or EAC at 13.5% over 5 years (Jin et al. (2008) International Journal of Cancer J_23, 2331-2336), the corresponding negative predictive value relating to our LR threshold was 97% {i.e., progression risk in the LR group was 3%) and the positive predictive value relating to HR was 40% {i.e., progression risk in the HR group was 40%).

(6) Predictiveness curve analyses

We used predictiveness curves (also known as risk plots) to assess the clinical utility of the combined classification rules in stratifying patients according to risk levels in the target population. To create predictiveness curves, we ordered and plotted risks from lowest to highest value. A progression rate to HGD and/or EAC of 13.5% over 5 years was assumed in adjusting estimates from the case-control sample to reflect population risk and its distribution. Results are shown in Table 10, Figure 13.

Table 10. Overfitting-corrected predictiveness curve analyses in 2-year, 4-year and combined models Risk probability < <00..11 ((LLRR)) 0.1-0.5 (IR) >0.5 (HR)

Combined model

8-marker panel 45% 51% 4% age alone 15% 85% 0% age plus 8-marker panel 51% 44% 5% 2-year model

8-marker panel 45% 51% 4% age alone 11% 89% 0% age plus 8-marker panel 52% 44% 4%

4-year model 8-marker panel 44% 51% 5% age alone 15% 85% 0% age plus 8-marker panel 51% 44% 5%

LR, low-risk; IR, intermediate-risk; HR, high-risk.

After overfitting correction, by age alone, nearly 90% of BE patients were classified as intermediate-risk (IR), whereas patients were well-stratified into low-risk (LR), IR, or high-risk (HR) categories by both the 8-marker alone and age plus 8-marker panels in all three models (Table 10, Figure 13).

(7) Discussion

In the current study, with specificity at 0.9, sensitivities of progression prediction approached 50% based on both the 8-marker panel alone and 8-marker-plus-age panel in all three models. These findings indicate that even while performing at high specificity, these biomarker models predicted half of progressors to HGD and EAC that would not have been diagnosed earlier without using these biomarkers.

Based on age alone, with specificity at 90%, only 17.6%, 23.2% and 22.1% of progressors were predicted in the three models. However, with panels based on age plus biomarkers or on biomarkers alone, approximately 60%, 50% and 50% of progressors were accurately predicted in these models. Predicted progressors represent patients in whom we can intercede earlier, resulting in higher cure rates. Finally, our combined risk model outperformed known risk in the general BE population (13.5% progression risk over 5 years), both in negative predictive value (3% progression risk over 5 years for the LR group) and positive predictive value (40% progression risk over 5 years for the HR group). Age is a common risk factor for many cancers, including EAC. In the current study, Ps were significantly older than NPs, and the AUCs of age alone were 0.604, 0.630 and 0.635, respectively in the three models, suggesting that age per se predicts neoplastic progression in BE. However, the incremental prediction accuracy (above age) contributed by the 8-marker panel was substantial in all three models. Thus, the current findings suggest that this 8-marker panel is more objective and quantifiable and possesses higher predictive sensitivity and specificity than do clinical features, including age. Furthermore, although age was a good classifier for disease progression, predictiveness curves revealed that age did not successfully stratify BE patients according to their progression risk. Moreover, age per se is not an accepted risk marker on which to base

clinical decisions regarding surveillance interval or neoplastic progression risk in BE. In contrast, models based on both the 8-marker panel and the age-plus-8-marker panel provided estimated progression risks either close to 0 (i.e., LR) or between 0.1 and 0.5 (i.e., IR) in the majority of individuals, suggesting that these markers exerted a substantial impact on risk category. This finding also suggests that in clinical practice, separate thresholds can be chosen to define high, intermediate, and low risk, based on predictiveness curves.

In conclusion, we have developed a risk stratification strategy to predict neoplastic progression in BE patients based on an 8-marker tissue methylation panel. At high specificity levels, this model accurately predicted approximately half of HGDs and EACs that would not have otherwise been predicted. This model is expected to reduce endoscopic procedures performed in BE surveillance while simultaneously increasing detection at earlier stages. Thus, these findings suggest that a methylation biomarker panel offers a clinically useful tool in the risk stratification of BE patients.

Example IV. Identification of about 50 additional markers for predicting the development of EAC or HGD

Agilent methylation array analyses were performed on 5 BE progressor patients (BE biopsy tissue samples from subjects who later developed HGD or EAC) and 4 BE nonprogressor patients (BE biopsy tissue samples from subjects who did not later develop HGD or EAC). Bioinformatic calculations and gene filtering criteria were applied to the data generated. Based on these procedures, a list of 50 methylation targets strongly associated with progression was derived. The screen included Student's t-testing for significance of associations, selection of loci that were hypermethylated in progressors vs. nonprogressors, and gene ontologic criteria for relevance of loci to cancer. "T-test (unlogged") connotes the t-test performed on methylation values that were not log-transformed. "Fold change" denotes the ratio for each gene or locus of methylation level in progressors to methylation level in nonprogressors.

Table 11 lists about 50 of the most promising markers identified by this study. Column 1 indicates the gene symbol of each gene, column 2 the gene ID number, and column 3 the official name of the gene. Having this information, a skilled worker can readily determine genomic sequences comprising coding sequences for each of these genes (e.g., for part or all of exon 1), and upstream regulatory regions, by accessing, for example, GenBank and/or the Entrez Gene searchable database (NCBI). The position of the transcription start sites can be found on any of these web sites, or elsewhere. Table 12 provides the start position and the end position, relative

to the transcriptional initiation site, for the probes from the Agilent Human CpG Island microarray which was used to identify each of these targets. A skilled worker can readily determine the sequence of each of these probes, by matching the start and end positions relative to the relevant genomic sequence. Sequences of promoter regions of these about 50 genes, extending from -500 to +100 nucleotides relative to the transcriptional start sites, are provided in the Sequence Listing attached hereto, as SEQ ID NOs: 50-121. In some cases, two or more promoter sequences are listed for each gene, so multiple promoter regions are provided.

Table 12 also provides the results of an unlogged T-test, the test of significance of association of progression with methylation level, performed on non-log-transformed methylation values. Values of less than 0.05 indicate statistically significant results. Also included in Table 2 are the fold changes, which further clarifies the significance of the results.

Table 13 provides suitable PCR primers and probes for qMSP analysis for a number of gene promoter regions that can be analyzed by a method of the invention. The forward primers sequences have SEQ ID NOs: 122 to 199, starting at the top of the table (AKAP 12) and ending at the bottom of the table (TDE/TMS/SERTNC3); the reverse primers have SEQ ID NOs: 200- 277, starting at the top of the table (AKAP 12) and ending at the bottom of the table (TDE/TMS/SERTNC3); and the TaqMan probe sequences have SEQ ID NOs: 278 to 295, starting at the top of the table (AKAP 12) and ending at the bottom of the table (TDE/TMS/SERTNC3).

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)

Atty Doc: 02240-264913; (P10304-02)