Differential methylation of cell-free circulating DNA among patients with pancreatic cancer versus chronic pancreatitis

Abstract

BACKGROUND:

Although patients with chronic pancreatitis (CP) have an increased risk of pancreatic cancer (PanCa), the timely detection of PanCa often is difficult, because the symptoms of CP and PanCa are very similar. Moreover, secondary inflammation may be identified in PanCa, further complicating diagnosis. To improve the survival of patients with PanCa, a reliable test to differentiate CP from PanCa is needed. In this article, the authors describe a methylation profile of cell-free plasma DNA that distinguished CP from PanCa with >90% accuracy.

METHODS:

Methylation in cell-free, plasma DNA was compared among 30 samples from patients with CP, 30 samples from patients with PanCa, and 30 samples from healthy controls (N) using a microarray-mediated methylation analysis of 56 fragments in each sample (MethDet56). Statistical analysis was done by using the Fisher exact test, a naive Bayes algorithm, and 25 rounds of 5-fold cross-validation.

RESULTS:

The MethDet56 methylation analysis technique identified 17 gene promoters as informative (8 for distinguishing N from CP and 14 for distinguishing CP from PanCa). It achieved 81.7% sensitivity and 78% specificity (P<.01) in the detection of CP (N vs CP) and 91.2% sensitivity and 90.8% specificity (P<.01) in the differential detection of PanCa (PanCa vs CP).

CONCLUSIONS:

The current data suggested that, among patients with pancreatic disease, the methylation profiles of inflammatory disease and cancer are different and open a new venue for the development of biomarkers for differential diagnosis. Further investigation of diagnostic biomarkers for pancreatic cancer based on methylation in cell-free, circulating DNA appears to be warranted. Cancer 2010. © 2010 American Cancer Society.

In the United States, pancreatic cancer accounts for approximately 2% of all cancers diagnosed each year but is the fourth leading cause of cancer death in the United States, causing 6% of all cancer-related fatalities.1 Approximately 95% of these are adenocarcinomas.2 The National Cancer Institute predicts that 37,680 new cases will have been diagnosed in the United States in 2008 and that 34,290 deaths will be attributable to the disease. With only a 5% total survival rate after 5 years, pancreatic cancer is considered 1 of the deadliest cancers. Only 9% of all patients with pancreatic cancer undergo surgery with curative intent, although resection is necessary for cure.3

The high mortality with pancreatic cancer is primarily because of the relative lack of symptoms early in the disease process, resistance to many treatments, and the concealed location of the pancreas. When a patient does have symptoms, they are often similar to other illnesses and can lead to a further delay in the diagnosis.4 Many of these symptoms, such as pain, diarrhea, weight loss, jaundice, and malabsorption, are similar to chronic pancreatitis.5 In addition, computed tomography, endoscopic ultrasound, and other imaging techniques may not be able to differentiate between chronic pancreatitis and pancreatic cancer.4 For example, van Gulik et al reported that 6% of patients who underwent resection for clinically suspected cancer had pathologically focal pancreatitis lesions instead.6 Although endoscopic retrograde cholangiopancreatography and fine-needle aspiration biopsy have greater specificity and sensitivity than imaging techniques, both are invasive, and their use is associated with some risk.

Noninvasive biomarkers are needed for the screening of patients with suspected pancreatic adenocarcinoma or chronic pancreatitis. Blood-based tests, which essentially are noninvasive and can be used to detect different types of molecules, might be suitable for this purpose. Carbohydrate antigen 19-9 (CA 19-9) levels are elevated in pancreatic cancer but frequently can be detected only in later stages of the disease. Although CA 19-9 is expressed in other cancers, such as colorectal cancer and gastric cancer, its level also is high in chronic pancreatitis (with concurrent biliary obstruction)7, 8 and in inflammatory diseases (such as rheumatoid arthritis9); therefore, the attempted detection of cancer by measuring CA 19-9 probably would have a very high rate of false-positive results.10 Compounding the problem is that a subset of the population lacks expression of Lewis antigen, which is required to produce CA 19-9,11 and that higher levels observed in pancreatic cancer have the highest mortality.12

Changes in DNA methylation have been observed in numerous cancers, including pancreatic cancer.13-19 These changes can be detected in tissue,16, 17 in pancreatic cancer cell lines,17, 18 in pancreatic juice,14, 20 and in cell-free plasma DNA.21 Changes in DNA methylation also have been observed in chronic pancreatitis.14

We demonstrated previously that the MethDet56 test we developed in our laboratory22 was able to detect pancreatic cancer by analyzing cell-free plasma DNA in which a composite biomarker for pancreatic cancer detection had sensitivity of 76% and specificity of 59%.21 In the current study, we analyzed methylation profiles in patients with chronic pancreatitis to determine whether a specific set of genes was methylated in their plasma and whether this set could differentiate between chronic pancreatitis and pancreatic cancer.

MATERIALS AND METHODS

Clinical Samples

This project was approved by the institutional review boards of Rush University Medical Center and Northwestern University. Plasma from patients with ductal adenocarcinoma of the pancreas was provided by the Pathology Core Facility of the Robert H. Lurie Comprehensive Cancer Center, Feinberg School of Medicine, Northwestern University. Tumor assignment was based on the description from final surgical pathology reports. Plasma samples from healthy volunteers of similar age and race were deposited in the same repository and served as controls. Plasma samples from patients with chronic pancreatitis were collected at NorthShore University HealthSystem (previously known as Evanston Northwestern Healthcare [ENH]) using a protocol that was approved by the institutional review board at ENH. All patients with chronic pancreatitis met 1 of the following criteria: 1) an abdominal ultrasound study consistent with chronic pancreatitis according to standard radiologic criteria (ie, echogenic foci in the parenchyma, large or small cavities, calcifications, dilated pancreatic duct); 2) an abdominal computed tomography scan consistent with chronic pancreatitis according to standard radiologic criteria (ie, calcifications, dilated pancreatic duct, irregular contour of the gland, cystic lesions); 3) an endoscopic retrograde cholangiopancreatography examination consistent with chronic pancreatitis according to standard radiologic criteria (dilated, tortuous main pancreatic duct with irregular, secondary branches; intraductal calculi); 4) an endoscopic ultrasound study consistent with chronic pancreatitis according to standard radiologic criteria (≥5 features); and 5) pancreatic calcifications identified on plain film of the abdomen. In addition, all patients had a stable clinical history over the past year with no suspicion for cancer (weight loss, jaundice, or change in abdominal symptoms). A tissue diagnosis was made for patients who underwent surgical exploration and/or resection. Additional samples of chronic pancreatitis (selected by the same criteria) were provided by Analytical Biological Services Inc. (Wilmington, Del). Informed consent was obtained from all participants. A brief description of the participants is presented in Table 1.

Table 1. Clinical Data on Study Participants

Disease Status	Sample Size	Age, y			Race, No
		Median	Range	SD	C	AA	Other
SD indicates standard deviation; C, Caucasian; AA, African American.
Chronic pancreatitis
Men	12	54.8	35-71	11.6	12	0	0
Women	18	61.9	46-81	9.9	16	1	1
Total	30	59.0	35-81	11.0	28	1	1
Pancreatic cancer
Men	12	58.7	48-70	6.1	11	1	0
Women	18	61.9	36-79	11.4	18	0	0
Total	30	60.6	36-79	9.7	29	1	0
Normal pancreas
Men	12	51.1	42-60	7.2	9	3	0
Women	18	62.0	37-79	11.0	18	0	0
Total	30	57.6	39-79	11.0	27	3	0

DNA Isolation

Whole blood was centrifuged at ×2600g to separate plasma from cellular elements. The plasma was removed, recentrifuged to remove any potential contaminating cells, and stored in aliquots at −80°C until it was needed. For DNA isolation, 200 μL of plasma were added to DNAzol BD (Molecular Research Center, Inc., Cincinnati, Ohio) and processed according to the manufacturer's protocol. The DNA was dissolved in 8 mM sodium hydroxide, and the pH was adjusted with 1 M (4-[2-hydroxyethyl]-1-piperazine ethane sulfonic acid) (Molecular Research Center, Inc.) before digestion.

Microarray-Mediated Methylation Assay

After isolation, each DNA sample was separated into equal volumes. One sample was digested by Hin6I, and the other was processed in buffer without enzyme. A nested, multiplexed polymerase chain reaction (PCR) (methyl-sensitive restriction enzyme-dependent PCR) was used to amplify each sample, as described previously.23 After amplification, the digested DNA was labeled with indocarbocyanine (Cy3), and the sham-digested DNA was labeled with indodicarbocyanine (Cy5) (both obtained from GE HealthCare, Piscataway, NJ). The labeled DNA samples were combined and hybridized on custom-designed arrays (MWG Biotech, High Point, NC). Each array was printed in triplicate, according to a previously established protocol.22 Hybridized and washed microarray slides were scanned immediately on a GenePix 4000B Microarray Scanner (Molecular Devices, Union City, Calif) and analyzed with GenePix Pro 6.0 software. The methylation profiles of genomic DNA from MCF-7 and T47 cells were used as control samples throughout the experiment to establish the consistency of the assay.

Statistical Analysis

Each array contained 3 identical subarrays of 64 spots. Three spots had no DNA and were used to determine the influence of background. Another 3 spots were used as controls to quantify the amount of nonspecific annealing. All data for Cy5 and Cy3 were background subtracted using the GenePix Pro 6.0 software. Cy5/Cy3 ratios <4.0 were considered methylated, and ratios >4.0 were considered unmethylated. Methylation calls were made independently for each spot from each subarray, and final gene-specific calls were made according to the majority call from the triplicate spots for that gene. If there was no majority, then the final call was NA for that sample. Nonspecific filtering removed uninformative genes (detectable calls in <66% of samples). Informative genes with P < .1 were selected by using the 2-sided Fisher exact test for differential methylation in gene-specific analyses comparing the methylation status in samples of cancer, normal pancreas, and chronic pancreatitis. A filter was used to eliminate any genes if >25% of the samples were eliminated for any of the above reasons.

Naive Bayes classifiers were constructed with the e1071 R package (R Development Core Team, 2005; R Foundation for Statistical Computing, Vienna, Austria) using an uninformative prior with probabilities of 0.5 for normal or diseased classification.24 The sensitivity and specificity of the naive Bayes classifier were estimated using 25 rounds of 5-fold cross-validation. For each round of cross-validation, the data were partitioned randomly into 5 sets with an equal distribution of diseased and control specimens. Each set then served as a test set based on training of the naive Bayes classifier with the other 4 sets. Sensitivity and specificity were estimated and averaged over rounds and runs. Gene selection and classifier parameter estimation were performed anew with each round of cross-validation.

Sample sizes of 30 for each group were determined based on clinical feasibility and power analysis. With this sample size, we have approximately 80% power to detect a difference ≥30% between groups for methylation proportion in a 2-sided chi-square test with a significance level of 5%. A small, preliminary study indicated that large differences in methylation status for many genes were likely to be observed between groups.

RESULTS

Clinical Samples

We identified 3 groups of individuals with similar age, sex, and ethnic distribution, as shown in Table 1. Each disease status group contained 30 patients, including 18 women and 12 men. There were no significant differences in age between the groups (P > .20 for all comparisons). Plasma samples from patients in each disease group were collected before any treatments. The pathology of ductal adenocarcinoma was determined from final pathology reports.

Classifier Genes

In total, 17 gene promoters were identified as differentially methylated (Tables 2 and 3). In the tables, the data in columns represent the number of methylated calls for each respective group, and the percentage of methylated samples is provided in parentheses. Eight promoters were selected from the comparison between samples of chronic pancreatitis and normal samples >75% of the time in the random training process. Of these, 3 promoters were specific to the analysis for the detection of chronic pancreatitis in normal samples (Table 2). When comparing chronic pancreatitis and pancreatic cancer, 14 gene promoters were selected >75% of the time in the random training process (Table 3). Nine of the 14 gene promoters were specific to chronic pancreatitis.

Table 2. Genes Selected as Classifiers of Each Sample Group in Individuals With Normal Pancreas Versus Chronic Pancreatitis

Gene Promoter	No. of Methylated Samples (%)
	Normal Pancreas	Chronic Pancreatitis
BRCA1 indicates breast cancer 1; CCND2, cyclin D2; HMLH1, human mutL homolog 1; CDKN1C, cycling-dependent kinase inhibitor; PGR-distal, progesterone receptor distal promoter; PGR-proximal, progesterone receptor proximal promoter; SYK, spleen tyrosine kinase; VHL, von Hippel-Lindau tumor suppressor.
BRCA1	8 (26.7)	20 (66.7)
CCND2	4 (17.4)	23 (82.1)
HMLH1	3 (11.5)	18 (78.3)
CDKN1C	18 (60)	27 (90)
PGR-distal	6 (20.7)	19 (70.4)
PGR-proximal	4 (14.3)	22 (75.9)
SYK	17 (56.7)	23 (88.5)
VHL	6 (20.7)	16 (55.2)

Table 3. Genes Selected as Classifiers of Each Sample Group in Individuals With Pancreatitis Versus Pancreatic Cancer

Gene Promoter	No. of Methylated Samples (%)
	Chronic Pancreatitis	Pancreatic Cancer
CCND2 indicates cyclin D2; DAPK1, death-associated protein kinase 1; ESR1 promA, estrogen receptor 1 promoter A; HMLH1, human mutL homolog 1; MGMT, O-6-methylguanine-DNA methyltransferase; MUC2, mucin 2, oligomeric mucus/gel-forming; MYOD1, myogenic differentiation 1; CDKN2B, cyclin-dependent kinase inhibitor 2B; CDKN1C, cyclin-dependent kinase inhibitor 1C; PGK1, phosphoglycerate kinase 1; PGR-proximal, progesterone receptor proximal promoter; RARb, retinoic acid receptor beta; RB1, retinoblastoma 1; SYK, spleen tyrosine kinase.
CCND2	23 (82.1)	6 (24)
DAPK1	2 (7.4)	16 (53.3)
ESR1 promA	22 (95.7)	14 (60.9)
HMLH1	18 (78.3)	8 (26.7)
MGMT	3 (10.3)	14 (46.7)
MUC2	17 (63)	27 (90)
MYOD1	26 (86.7)	10 (33.3)
CDKN2B	15 (50)	26 (86.7)
CDKN1C	27 (90)	8 (26.7)
PGK1	4 (14.8)	18 (60)
PGR-proximal	22 (75.9)	11 (36.7)
RARb	12 (60)	4 (13.8)
RB1	16 (72.7)	11 (36.7)
SYK	23 (88.5)	17 (56.7)

The results of the biomarker selection are presented as 2 × 2 contingency tables (Tables 4 and 5). A sensitivity of 78% for chronic pancreatitis (Table 4) was determined according to the percentage of correctly detected chronic pancreatitis samples in the 25-round, 5-fold cross-validation. In addition, the cross-validation had a specificity of 81.7% for normal samples. When chronic pancreatitis samples were compared with pancreatic cancer samples, the naive Bayes classifier produced a sensitivity of 91.2% and a specificity of 90.8% for cancer (Table 5).

Table 4. Accuracy of Chronic Pancreatitis Detection

	Predicted (95% CI), %
	Normal Pancreas	Chronic Pancreatitis
CI indicates confidence interval. a When calculating the CI, logit transformation was used to achieve a better normal approximation.
Actual
Normal pancreas	81.7 (67.3-90.6)a	18.3
Chronic pancreatitis	22	78 (63.8-87.7)a

Table 5. Accuracy of Differentiation Between Chronic Pancreatitis and Pancreatic Cancer

Actual	Predicted (95% CI), %
	Chronic Pancreatitis	Pancreatic Cancer
CI indicates confidence interval. a When calculating the CI, logit transformation was used to achieve a better normal approximation.
Chronic pancreatitis	91.2 (76.5-97.1)a	8.8
Pancreatic cancer	12.8	90.8 (76.1-96.8)a

Validation of Developed Biomarker

Testing in an independent cohort of patients is considered the gold standard for biomarker validation. This standard was difficult to reach during the development phase for a rare disease like as pancreatic cancer. It is significant that the developed biomarker produced an observed sensitivity of 81.7% with a 95% confidence interval of 67.3% to 90.6% (control vs chronic pancreatitis).

To validate the biomarker and avoid over fitting during model building, we used a statistical technique of cross-validation (25 rounds of 5-fold cross-validation), which reportedly reduced potential bias and produced better accuracy than the individual train-and-test (holdout) method.25, 26 Results of cross-validation (Tables 4 and 5) indicated that our biomarkers were reliable to the extent possible without a full-scale clinical trial.

DISCUSSION

Smoking, diabetes, race, chronic pancreatitis, and heredity all are risk factors for pancreatic cancer. Numerous genes have been implicated in the development of pancreatic cancer. Many of these are cell cycle regulators, tumor suppressors, and oncogenes. Breast cancer 1 (BRCA1), BRCA2, and p16/cyclin-dependent kinase 2A (CDKN2A) are only a few of the genes responsible for hereditary factors, but <10% of pancreatic cancers can be attributable to heredity alone.4, 19, 27

In the past decade, DNA hypermethylation or hypomethylation of CpG islands has played a prominent and well established role in the study of cancer. The aberrant methylation of promoters has been a speculated mechanism by which epigenetics can cause cancer through the up-regulation or down-regulation of cell cycle genes, DNA repair genes, and other signaling pathways. In pancreatic cancer, altered methylation of preproenkephalin (ppENK), forkhead box E1 (FOXE1), CDKN1C, suppressor of cytokine signalling-1 (SOCS1), Reprimo (candidate mediator of p53-dependent G2 arrest), long QT intronic transcript 1 (LIT1), Ras association domain family member 1A (RASSF1A), and many others have been observed in pancreatic juice, cell lines, and tissue biopsies.18, 28-33

Chronic, focal inflammation is a common risk factor in many types of cancer.34-36 Unfortunately, there are limited studies in the area of chronic pancreatitis.14, 37 Matsubayashi et al demonstrated an increase in the methylation pattern in pancreatic juice from patients with chronic pancreatitis compared with controls using an analysis of 17 genes that were highly methylated in pancreatic cancer but reported that no particular gene was useful for the diagnosis of chronic pancreatitis in normal patients.14 In patients with chronic gastritis who had no metaplasia, several hypermethylated CpG islands were observed in death-associated kinase (DAPK), cadherin 1 type 1 (CDH1), thrombospondin 1 (THBS1), human mutL homolog 1 (hMLH1), protease-activated receptor 2 (PAR2), and CDKN2A.38, 39 It is noteworthy that interleukin 6 (IL-6) reportedly affects methylation of the p53 gene promoter and also regulates human DNA methyltransferase (hDNMT).40 Although it is well known that chronic inflammation is characterized by an increase in cytokines, such as IL-1, IL-6, IL-8, transforming growth factor β, and tumor necrosis factor α, these can increase cell proliferation, cause the release of nitric oxide and reactive oxygen species from neutrophils, and inhibit apoptosis.41 Not surprisingly, many of the same cytokines also are present in pancreatic cancer.42 This environment could induce DNA mutations and alter CpG methylation.41, 43

The characteristic fibrosis observed in pancreatic cancer desmoplasia has given rise to the hypothesis that the inflammatory nature of chronic pancreatitis provides a microenvironment milieu in which a progressive, multistep pathogenesis may proceed into pancreatic cancer.44, 45 It has been demonstrated that pancreatic intraepithelial neoplasia progresses to pancreatic adenocarcinoma through the increased mutation and methylation of specific genes.37, 46-48 Chen et al demonstrated that some of the proteins that are expressed differentially in chronic pancreatitis tissue compared with healthy controls also are expressed in pancreatic cancer. Some of these proteins, such as superoxide dismutase 1 (SOD1), SOD3, and macrophage migration inhibitory factor (MIF), are associated with inflammatory response.49

Pancreatic cancer may follow chronic pancreatitis, and inflammation may be a secondary feature of tumor development; therefore, a certain similarity in the biomarkers of tumor and chronic pancreatitis (compared with healthy controls) is expected. Indeed, the biomarker for chronic pancreatitis (Table 2) has 2 genes in common with the biomarker for cancer detection that we reported previously.21 These genes (CCND2 and von Hippel-Lindau syndrome [VHL]) may reflect the inflammatory component of chronic pancreatitis (Table 2) and pancreatic cancer,21 although other explanations (eg, similar involvement of stroma) cannot be excluded. In addition, the origins of cell-free plasma DNA used for biomarker development are only partly understood, and common elements of the composite biomarkers may be derived from different cells. Finally, it is important that, in each case, these elements are only parts of the whole composite biomarker; the biomarker for detection of pancreatic cancer contains 5 genes,21 whereas 8 are required for the detection of chronic pancreatitis (Table 2).

Although hundreds of promoter regions have been identified as methylated in cancers, currently, to our knowledge, there is no single biomarker system for the analysis of multiple methylated gene promoters. By using a platform with multiple genes, a composite biomarker can be generated to distinguish between different cancers. With this 56-gene promoter platform, a methylation pattern already has been established in breast cancer22 and can be a useful test for its detection with specificity of 72.4% and sensitivity of 74.7%. In addition, it can be used to diagnose ovarian cancer with 85% sensitivity and 61% specificity using cell-free plasma DNA in blood.50 Our platform also can be used for blood-based detection of pancreatic cancer.21 In the current report, we establish for the first time that the same approach is useful for the differential diagnosis of pancreatic cancer and chronic pancreatitis. Although the accuracy of the assay does not allow its use as a stand-alone test, it can be applied in combination with other modalities as a first-line screening test. Further improvement of the test—for example, expanding the number of tested CpG regions to select the most differentially methylated—should produce a much more sensitive and specific assay and warrants additional study.

Acknowledgements

We thank Ms. Brenda Robertson and personnel of Analytical Biological Services, Inc. (Wilmington, Del) for their help with the procurement of some of the chronic pancreatitis samples and Dr. Joel Peek and personnel of Microarrays, Inc. (Huntsville, Ala) for their continuous help with custom microarrays.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.