IJC International Journal of Cancer
DNA methylation array analyses identified breast cancer-associated HYAL2 methylation in peripheral blood €tze1,2, Manuela Zucknick3, Christian Sutter4, Barbara Wappenschmidt5, Frederik Marme1, Rongxi Yang1,2, Katrin Pfu 6 1,2 Bin Qu , Katarina Cuk , Christoph Engel7, Sarah Schott1, Andreas Schneeweiss1, Hermann Brenner8, Rainer Claus9, Christoph Plass9, Peter Bugert10, Markus Hoth6, Christof Sohn1, Rita Schmutzler5, Claus R. Bartram4 and Barbara Burwinkel1,2 1
Molecular Biology of Breast Cancer, Department of Gynecology and Obstetrics, University of Heidelberg, Heidelberg, Germany Molecular Epidemiology (C080), German Cancer Research Center (DKFZ), Heidelberg, Germany 3 Division of Biostatistics (C060), German Cancer Research Center (DKFZ), Heidelberg, Germany 4 Institute of Human Genetics, University of Heidelberg, Heidelberg, Germany 5 Department of Gynecology and Obstetrics, Clinical Center of University of Cologne, Cologne, Germany 6 Department of Biophysics, Saarland University, Homburg (Saar), Germany 7 Institute of Medical Informatics, Statistics and Epidemiology, University of Leipzig, Leipzig, Germany 8 Division of Clinical Epidemiology and Aging Research (C070), German Cancer Research Center (DKFZ), Heidelberg, Germany 9 Division of Epigenomics and Cancer Risk Factors (C010), German Cancer Research Center (DKFZ), Heidelberg, Germany 10 Institute of Transfusion Medicine and Immunology, Medical Faculty Mannheim, University of Heidelberg, German Red Cross Blood Service Baden€rttemberg – Hessen, Mannheim, Germany Wu
Breast cancer (BC) is the leading cause of cancer-related mortality in women worldwide. Changes in DNA methylation in peripheral blood could be associated with malignancy at early stage. However, the BC-associated DNA methylation signatures in peripheral blood were largely unknown. Here, we performed a genome-wide methylation screening and identified a BC-associated differentially methylated CpG site cg27091787 in the hyaluronoglucosaminidase 2 gene (HYAL2) (discovery round with 72 BC case and 24 controls: p 5 2.61 3 1029 adjusted for cell-type proportions). The substantially decreased methylation of cg27091787 in BC cases was confirmed in two validation rounds (first validation round with 338 BC case and 507 controls: p < 0.0001; second validation round with 189 BC case and 189 controls: p < 0.0001). In addition to cg27091787, the decreased methylation of a 650-bp CpG island shore of HYAL2 was also associated with increased risk of BC. Moreover, the expression and methylation of HYAL2 were inversely correlated with a p-value of 0.006. To note, the BC-associated decreased HYAL2 methylation was replicated in the T-cell fraction (p 5 0.034). The cg27091787 methylation level enabled a powerful discrimination of earlystage BC cases (stages 0 and I) from healthy controls [area under curve (AUC) 5 0.89], and was robust for the detection of BC in younger women as well (age < 50, AUC 5 0.87). Our study reveals a strong association between decreased HYAL2 methylation in peripheral blood and BC, and provides a promising blood-based marker for the detection of early BC.
Breast cancer (BC) is the most common cancer and the leading cause of cancer-related mortality among women,1 with 1.7 million new cases and 522,000 deaths in 2012.2 In the
United States, one in eight women will develop BC in her lifetime.3 Although therapeutic advances have improved the survival rate, most BC patients still suffer from greatly
Key words: HYAL2, methylation, breast cancer, early detection, marker Abbreviations: AUC: area under curve; BC: breast cancer; FDR: false discovery rate; HYAL2: hyaluronoglucosaminidase 2 gene; IQR: interquartile range; OR: odds ratio; ROC: receiver operating characteristic; SLITRK4: slit and trk like 4 protein gene; SNP: single nucleotide polymorphism Additional Supporting Information may be found in the online version of this article Conflict of interest: Rongxi Yang and Barbara Burwinkel are inventors of a provisional patent application relating to the subject matter of this manuscript and therefore declare a potential conflict of interests Grant sponsor: Deutsche Krebshilfe; Grant number: 107054; Grant sponsors: Dietmar-Hopp Foundation, University Hospital of Heidelberg, Helmholtz Society and the German Cancer Research Center (DKFZ) and Sonderforschungsbereich (SFB) 894 DOI: 10.1002/ijc.29205 History: Received 17 Apr 2014; Accepted 26 Aug 2014; Online 12 Sep 2014 Correspondence to: Rongxi Yang or Barbara Burwinkel, Molecular Biology of Breast Cancer, University Women’s Clinic University Heidelberg, Im Neuenheimer Feld 440, 69120 Heidelberg, Germany, Tel.: 149-6221-568403, Fax: 149-6221-565356, E-mail: [email protected] or [email protected]
C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Early Detection and Diagnosis
2
2
HYAL2 methylation in the peripheral blood
Early Detection and Diagnosis
What’s new? Altered DNA methylation in peripheral blood may be indicative of early-stage breast cancer, but methylation signatures await validation. The present study is among the first to identify and validate decreased methylation at CpG site cg27091787 in the HYAL2 gene as a potential marker of breast cancer. Methylation levels at cg27091787 were inversely associated with HYAL2 expression and were found to be indicative of breast cancer stage, thus enabling discrimination of early-stage disease, as well as facilitating disease detection in women under age 50. Differential methylation at cg27091787 could serve as a bloodbased marker for early breast cancer detection.
reduced quality of life and even metastasis owing to inefficient risk evaluation or delayed diagnosis.4,5 BC is a complex and heterogeneous disease caused by the interactions of both genetic and nongenetic factors. Age, female gender and family history are the major risk factors for BC.4 However, even combining the three major factors with other known BC risk factors, such as exposure to the hormones, life styles, medical and reproductive factors, the predictive accuracy of present model for BC in individuals is only about 58–59%.6,7 The known rare inherited mutations in the BC susceptibility genes, such as BRCA1, BRCA2, P53, PTEN, CHEK2 and ATM, can explain about one-third of the familial BCs, but together only account for 1.5–3% of all BCs.8 Although recent genomewide association studies have successfully detected multiple variants with low-penetrance risk to BC, a panel of ten such SNPs only represents a predictive accuracy for BC of 59.7%.6 Thus, the known environmental and genetic risk factors have limited utility in evaluating the risk of BC. DNA methylation harbors critical relevance for the control of gene transcriptional activities and the architecture of the nucleus of cells.9 The altered methylation in the promoter regions are among the earliest and most common events in the process of cancer.10 Given the accessibility of methodology,11,12 genome-wide methylation analyses have been extended to peripheral blood in case–control studies of ovarian, bladder and head and neck cancers.13–15 A few aberrant methylation patterns in peripheral blood have also been reported in BC by candidate gene approaches.16–19 Recently, Xu et al. reported an extensive blood-based epigenome-wide association study of BC by Illumina 27K Methylation Array in a well-designed prospective sister study cohort, and demonstrated that the methylation profiling of blood are promising markers for BC detection and risk evaluation.20 Hereby, we conducted an Infinium 27K Methylation Array on peripheral blood DNA in case–control study, and performed further independent validations by MassARRAY for BC-associated methylation variability. In addition, the association was replicated in blood cell subtypes and the correlation between gene expression and methylation was analyzed.
ples from BC cases and healthy controls were obtained from centers in Southwest Germany. All the individuals were Caucasian. All the recruited cases and controls gave written informed consent. Genomic DNA was isolated from peripheral whole blood using DNA isolation kits from Qiagen. Detailed information for the samples is shown in Table 1. Peripheral blood samples from familial BC patients: peripheral whole-blood samples from BRCA1/2 mutationnegative index familial BC patients were collected by the centers of the German Consortium for Hereditary Breast and Ovarian Cancer (GC-HBOC) during the years 1997–2007. German index patients were first diagnosed with BC and then referred to a family registry. All the familial BC cases were recruited according to the criteria of family history described previously.21 In the GC-HBOC study, about 97% of patients agreed to use their blood samples for research purposes.22 A total of 410 familial BC samples were selected randomly for our study from the GC-HBOC study in Heidelberg and Cologne. Peripheral blood samples from sporadic BC patients: Peripheral whole-blood samples were consecutively collected at University Hospital of Heidelberg from 2006 to present. In our study, about 90% of patients agreed to use their blood samples for research purposes. A total of 189 sporadic BC samples collected in the years of 2009 and 2010 were randomly selected for our study. The clinical characteristics of BC patients were defined according to the American Joint Committee on Cancer staging manual.23 Peripheral blood samples from controls: Peripheral wholeblood samples were consecutively collected from blood donors by the German Red Cross Blood Service of BadenW€ urttemberg-Hessen (Mannheim, Germany) from 2004 to present. About 95% of blood donors agreed to use their blood samples for research purposes.22 All control individuals were healthy when donating their blood. None of the control individuals had a reported family history of BC.22 A total of 720 healthy and unrelated females were randomly selected during the years 2004–2010 from the blood bank as controls for our study. No further inclusion criteria were applied during recruitment of controls.
Material and Methods Study population
Leukocytes
Our study was approved by the Ethics Committee of the University of Heidelberg (Germany). Peripheral blood sam-
Peripheral blood samples from 36 sporadic BC patients and 40 healthy female controls were collected at the University C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
3
Yang et al.
Table 1. Sample description
Rounds
Sample types
Target N
Assayed N
Call rate (%)
Mean of age (range)
Controls
24
24
100.0
48.5 (32–66) 44.5
Familial BC cases
72
72
100.0
47.0 (26–78) 46
Controls
507
491
96.8
44.7 (30–67) 42
Familial BC cases
338
328
97.0
46.1 (24–77) 46
Controls
189
185
97.9
61.2 (31–69) 64
Sporadic BC cases
189
183
96.8
59.6 (32–87) 60
Controls
96
95
99.0
46.0 (30–67) 43
Familial BC cases
96
96
100.0
48.0 (24–78) 47
Controls
93
92
98.9
60.0 (36–67) 64
Sporadic BC cases
95
95
100.0
59.4 (39–87) 59
40
38
95.0
38.2 (21–63) 39
36
34
94.4
58.2 (29–81) 57
MassARRAY Benign breast tissue 14 real-time PCR
14
100.00 47.7 (32–65) 45.5
Assays
Groups
Median of age
Determine the association between the methylation level of cg27091787 and BC1 Discovery round
Peripheral blood DNA
Illumina 27K Assay
First validation round
Peripheral blood DNA
MassARRAY
Second validation round
Peripheral blood DNA
MassARRAY
Discovery of HYAL2 Peripheral blood DNA MassARRAY CpG island shore2
Validation of HYAL2 Peripheral blood DNA MassARRAY CpG island shore3
HYAL2 methylation and expression in leukocytes
DNA and RNA from peripheral blood leukocytes
MassARRAY, Controls real-time PCR Sporadic BC cases
HYAL2 methylation and expression in tissue
HYAL2 methylation in different blood cell types
1 2 3
DNA and RNA from tissue
Different blood cell types
MassARRAY
Malignant BC tumor
34
32
94.12
56.7 (29–81) 54.6
Controls
14
14
100.0
38.3 (21–51) 42
Sporadic BC cases
7
7
100.0
50.3 (34–76) 49
All the rounds in the sample description are independent from each other except 2 and 3. Samples were randomly selected from the first validation round. Samples were randomly selected from the second validation round.
Hospital of Heidelberg. Leukocytes were isolated freshly from peripheral blood using red blood cell lysis buffer as described previously24 within 2 hr of blood collection. The leukocyte pellets were immediately frozen in liquid nitrogen after isolation and kept at 280 C before use. DNA and RNA were isolated from leukocytes using AllPrep DNA/RNA/Protein Mini Kit from Qiagen. Cases and controls were always processed in parallel.
R CD3 positive isolation kit (Invitrogen, Carlsbad, with a DynalV U.S.). The leftover cells were collected as “B/T-lymphocytesdepleted leukocytes.” The cell pellets were immediately frozen in liquid nitrogen after purification and kept at 280 C before use. DNA was isolated from different blood cell types using AllPrep DNA/RNA/Protein Mini Kit from Qiagen (Germany). All the cases and controls were processed in parallel.
Blood cell fractionations from leukocytes
Tissues
Leukocytes were freshly obtained as described from seven sporadic BC patients and 14 female healthy controls at the University Hospital of Heidelberg. First, B cells were positively R CD19 positive isolation kit (Invitroisolated using a DynalV gen, Carlsbad, U.S.) from fresh leukocytes. Subsequently these B cell-depleted leukocytes were used for T-cell purification
Tissue samples from 14 benign breast tissues and 34 malignant BC tumors were obtained at the University Hospital of Heidelberg before any BC treatment. The tissues were immediately frozen in liquid nitrogen after obtained and kept at 280 C before use. DNA and RNA were isolated from tissue using AllPrep DNA/RNA/Protein Mini Kit from Qiagen
C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Early Detection and Diagnosis
Further exploration of HYAL2 methylation
4
(Germany). Malignant and benign samples were always processed in parallel. Infinium 27K methylation assay
In the discovery round, 500 ng of genomic DNA from each sample was bisulfite converted using the EZ-96 DNA Methylation Kit (Zymo Research, Orange County, U.S.) and subsequently subjected to the genome-wide methylation screening using the Human Methylation27 BeadChip (Illumina, San Diego, California, U.S.) according to the manufacturer’s recommendations.25 All the 96 samples passed the quality control.
Early Detection and Diagnosis
MALDI-TOF mass spectrometry
MALDI-TOF mass spectrometry (Sequenom, San Diego, California, U.S.) described by Breitling et al.11,26 was used in all the validation and further exploring rounds. In brief, DNA was bisulfite converted by EZ-96 DNA Methylation Gold Kit (Zymo Research, Orange County, U.S.) and amplified by bisulfite-specific primers. The bisulfite-specific primers (no single nucleotide polymorphisms (SNPs) in the primers) and polymerase chain reaction (PCR) amplicons are presented in Supporting Information Figure 1. The PCR products were treated according to the standard protocol of SequenomEpiTyper Assay, and further cleaned by Resin and dispensed to a 384 SpectroCHIP by a Nanodispenser. The chips were read by a Sequenom Mass Spectrometer system. Data were collected by SpectroACQUIRE v3.3.1.3 software and visualized with MassArrayEpiTyper v1.2 software. For each batch of MassARRAY analysis, same amount of cases and controls were randomly selected from the cohort. The samples from BC cases and controls were treated and analyzed in parallel in all the processes. Five percent of the samples were randomly chosen for a duplication analysis and achieved a Spearman correlation of 0.84, and a coefficient of variation of 0.07 (95% CI: 0.06–0.09). Quantitative real-time PCR
Hundred nanograms of total RNA from each sample was tranR Reverse Transcription scribed to cDNA using TaqManV Reagents (Applied Biosystems, Germany). Quantitative realtime PCR was performed using a LightCycler 480 (Roche, Germany) in combination with TaqMan gene expression assays (Applied Biosystems) for HYAL2 gene and housekeeping gene HPRT1 (hypoxanthine-guanine phosphoribosyltransferase1) as endogenous control. Crossing point values were calculated using the second-derivative maximum method by the LightCycler 480 basic software (Roche, Germany). Relative expression of HYAL2 for each sample was calculated according to DDCt method27 by normalization to HPRT1. Sequencing
The genomic DNA from the 80 samples (20 familial BC cases and 20 controls with the highest cg27091787 methylation, 20 familial BC cases and 20 controls with the lowest cg27091787 methylation) was used for the detection of germline mutation
HYAL2 methylation in the peripheral blood
in the vicinity of HYAL2-A amplicon. PCR primers were designed to amplify a 568-bp segment (chr3:5033532250335889, 36.1/hg18) covering the whole HYAL2-A amplicon and two database-annotated very rare SNPs in vicinity (rs34465081, no frequency data in dbSNP database; rs4688732, minor allele frequency 5 0.004 in Caucasian). The primer pairs were as follows: forward: 50 -ACCCTGTTTCCCTCC AGAAT-30 and reverse: 50 -CACCCCTCCCAGAGGAGTAT-30 . The PCR products were treated with the ExoSAP-IT purification kit (GE Healthcare, Germany) and sequenced using the forward and reverse primers from both directions via a 3130XL Genetic Analyzer (Applied Biosystems, Germany). Sequencing results were analyzed using Sequencing Analysis 5.2 software (Applied Biosystems, Germany). Statistical analyses
The Illumina 27K Array data were processed by the Illumina BeadStudio software with default settings. Probes with detection p-value > 0.01 were removed and samples were quantile-normalized. Association of probes with case–control status was assessed by beta-regression models with a logistic link and associated Wald tests using R.28 Likelihood ratio tests were used to compare the case–control model with the nested model for chip differences. Multiple testing adjustments were done with the Benjamini-Hochberg method controlling the false discovery rate (FDR) at the level of 0.05. The cell-type proportions were adjusted for the Illumina 27K Array by fitting data linear mixed effects models.29 All the statistical analyses of MassARRAY and gene expression data were conducted by SPSS Statistics 17.0. The correlations were assessed by Spearman’s rank correlation coefficients. Logistic regression models and nonparametric tests were used for comparisons between two and multiple groups. The results of logistic regression were adjusted for possible confounding effects of age and different measurement batches by including additional covariables in the logistic regression models. Receiver operating characteristic (ROC) curve analysis was performed to assess the discriminatory power of methylation levels. All statistical tests were two-sided, and p values 5%, Supporting Information Fig. 2). The CpG loci in the SFRP1, PTGS2 and BRCA1 genes also exhibited lower methylation in familial BC patients compared to controls (all FDR-adjusted p-value: 69%), the three lower quartiles of cg27091787 methylation levels (Q1–Q3) were highly associated with an increased risk of BC, and this association was especially pronounced in the lowest quartile (Q1, methylation level 54%, OR 5 41.5, 95% CI: 22.2–74.4, p < 0.0001 by logistic regression, Supporting Information Table 3). To exclude the possibility that the observed altered HYAL2 methylation was due to the influence of genetic variants, we sequenced the HYAL2-A amplicon and two nearby rare single nucleotide polymorphisms (SNPs) in the peripheral blood DNA from 80 samples with the highest and lowest cg27091787 methylation in BC cases and controls. No mutations or SNPs were found (data not shown, see Material and Methods section). In the second independent validation round, the differential HYAL2 methylation in peripheral blood DNA was investigated in 189 sporadic BC cases and 189 controls (Table 1). Again, cg27091787 site showed lower methylation levels in cases than in controls [median 50.50 (IQR 5 0.46–0.54) and 0.64 (IQR 5 0.58–0.70) for BC cases and controls, respectively, OR per 10% methylation 5 8.14, p < 0.0001 by logistic regression, Fig. 1c, Supporting Information Table 2]. Compared to the highest quartile of cg27091787 methylation (Q4, methylation level > 65%), the other three quartiles (Q1–Q3) were associated with a significantly increased risk of BC, especially the lowest quartile (Q1, methylation level
Early Detection and Diagnosis
The validation of BC-associated blood-based HYAL2 methylation by two independent case–control studies
6
HYAL2 methylation in the peripheral blood
MassARRAY (Supporting Information Fig. 1). However, none of CpG sites in SLITRK4 passed the first validation round (ORs per 10% methylation from 0.92 to 1.04, p > 0.5 for all by logistic regression, Supporting Information Table 4).
Early Detection and Diagnosis
The identification and validation of BC-associated HYAL2 CpG island shore
Figure 2. Definition of BC-associated CpG island shore of HYAL2. (a) Schematic diagram of HYAL2 promoter regions. CpG island-1 and CpG island-2 are located before the translation start site (TSS). The second exon could also be an intron due to variant splicing. The 61,000-bp flanking CpG sites of cg27091787 are presented. The HYAL2-A amplicon covers the CpG cg27091787, HYAL2-B amplicon covers the region between HYAL2-A amplicon and the CpG island-1 and HYAL2-C amplicon covers most region of the CpG island-1. (b) Discovery of the BC-associated significant methylation differences in amplicon HYAL2-A and HYAL2-B in 96 familial BC cases and 96 controls randomly selected from the first validation round. (c) Validation of the BC-associated significant methylation differences in amplicon HYAL2-A and HYAL2-B in 95 sporadic BC cases and 93 controls randomly selected from the second validation round. The p-values of all the 32 measureable CpG loci in the three amplicons were calculated by logistic regression adjusted for age and different batches for the measurements. The lines indicate the thresholds of different p values of 0.05, 0.01 and 0.001.
49%, OR 5 133.0, 95% CI: 40.7–434.5, p < 0.0001 by logistic regression, Supporting Information Table 3). The methylation levels of all the three additional CpG loci in HYAL2-A amplicon were substantially correlated with the methylation level of cg27091787 and were all significantly associated with BC in both validation rounds (all Spearman’s q > 0.55, all OR per 10% methylation > 1.80, p < 0.0001 by logistic regression, Fig. 1, Supporting Information Table 2). An amplicon SLITRK4 covering the cg12842316 site and two additional measurable flanking CpG sites was designed for
Next, we attempted to find the BC-associated differentially methylated region in HYAL2. Two additional amplicons, HYAL2-B and HYAL2-C, were designed to cover the CpGdense region, locating from the downstream of HYAL2-A amplicon to the closest CpG island (Fig. 2a and Supporting Information Fig. 1). Ninety-six familial BC cases and 96 controls randomly selected from the first validation round were analyzed by MassARRAY. Interestingly, the methylation levels of all 11 measureable CpG loci in HYAL2-B amplicon were significantly lower in cases than in controls (ORs per 10% methylation from 1.77 to 3.52, p < 0.017 for all), and were substantially correlated with the methylation levels of cg27091787 (Spearman rho from 0.48 to 0.65). In contrast, all the 17 measureable CpG loci in HYAL2-C amplicon showed no or weak association with BC (ORs per 10% methylation from 0.94 to 2.24, p-values from 0.623 to 0.007) and had weaker correlation with the methylation levels of cg27091787 (Spearman rho from 0.20 to 0.46) (Fig. 2b and Supporting Information Table 5). The association between altered methylation in HYAL2-A and HYAL2-B amplicons and BC was further validated by 95 sporadic BC cases and 93 controls randomly selected from the second validation round (Fig. 2c and Supporting Information Table 6). Together, we have defined a 650-bp BC-associated differentially methylated region or so-called CpG-island shore in HYAL2 from cg27091787 to the beginning of the adjacent CpG island. The inverse correlation between the methylation and expression of HYAL2 in peripheral blood leukocytes
To understand the correlation between methylation and the expression of HYAL2 gene, DNA and RNA were isolated from peripheral blood leukocytes of 36 sporadic BC patients and 40 healthy controls (Table 1). Significantly lower methylation can be observed in all the four CpG sites in the HYAL2-A amplicon in the leukocytes DNA from BC cases compared to that from controls (Fig. 3a). In contrast, the median of relative expression levels of HYAL2 in the leukocytes of sporadic BC cases was 1.3-fold higher than in the controls (Fig. 3b). The methylation levels of cg27091787 were significantly and inversely correlated with the expression levels of HYAL2 with a Spearman rho of 20.323 (p 5 0.006, Fig. 3c). The methylation levels of cg27091787 flanking CpG sites, HYAL2-A_CpG_1 and HYAL2-A_CpG_3, showed even stronger inverse correlation with the expression levels of HYAL2 (Spearman rho 5 20.431 and 20.452, respectively). Replication of HYAL2 methylation differences in T cells
Next, we attempted to replicate the methylation differences in freshly sorted leukocyte fragments. We separated B cells, T C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
7
Yang et al.
without the pattern of changed methylation in all four CpG sites in HYAL2-A amplicon, the weak correlations by individual CpG sites should be taken with caution.
Figure 3. The inverse correlation between the methylation and expression of HYAL2 in leukocytes. (a) The box plots show the methylation levels of cg27091787 and adjacent CpG sites in the HYAL2-A amplicon in leukocytes from 36 sporadic BC cases and 40 healthy controls. The box plot of cg27091787 is framed in box for emphasis. (b) The box plot shows the expression level of HYAL2 in leukocytes from sporadic BC cases and healthy controls. The presented p-values were calculated by Mann–Whitney U-test. The circles indicate outliers. (c) The inverse correlation between the methylation level of cg27091787 and HYAL2 expression in leukocytes.
cells and the B/T-lymphocytes-depleted leukocytes (mainly granulocytes, monocytes and NK cells) from seven sporadic BC cases and 14 healthy controls (Table 1). For the T-cell fraction, all the four CpG loci in HYAL2-A amplicon showed lower methylation in BC cases than in the controls. Of these, cg27091787 was significant at p 5 0.034 and HYAL2-A_CpG_1 was significant at p 5 0.043, even with very small sample number (Fig. 4). To note, the T cells also showed larger HYAL2 methylation difference than that seen in the whole blood. Compared to the controls, the HYAL2 methylation levels in BC cases were also lower in B/T-lymphocytes-depleted leukocytes, but higher in B cells (Fig. 4). HYAL2 methylation and the clinical characteristics of BC
To explore the relationship between blood-based HYAL2 methylation and the clinical characteristics of BC, the sporadic BC patients with available clinical data were interpreted. The methylation levels of all CpG sites in HYAL2-A amplicon showed no correlation with tumor size, grading, menopausal status, the status of Her2 receptor and family history of BC. Weak correlations were observed between individual CpG sites and stage, lymph nodes involvement and ER/PR receptor status (Supporting Information Table 7). However, C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
To estimate the potential clinical utility of HYAL2 methylation as a marker for the detection of early BC, ROC curve analyses were performed adjusted for possible confounding effects by logistic regression. In the discovery round, the methylation level of cg27091787 showed robust discriminatory power for differentiating BC cases from healthy controls when adjusted for age and cell counts [area under curve (AUC) 5 0.95, 95% CI: 0.90–1.00, Fig. 5a]. The discriminatory power of cg27091787 methylation for BC was verified in the validation rounds (adjusted for age and different measurement batches, first validation round, AUC 5 0.83, 95% CI: 0.80–0.86; second validation round, AUC 5 0.89, 95% CI: 0.85–0.92, Figs. 5b and 5c). In addition, the HYAL2 methylation level had outstanding power to distinguish very early BC from controls (adjusted for age and different measurement batches, stages 0 and I cases vs. controls, AUC 5 0.89, 95% CI: 0.85–0.93, Fig. 5d). Although age is a limited factor for the utility of mammography, it had no influence on the discriminatory power of cg27091787 methylation, and was even a benefit for the younger women (adjusted for age and different measurement batches age < 50 years: AUC 5 0.87, 95% CI: 0.84–0.89; age 50 years, AUC 5 0.83, 95% CI: 0.80–0.87, Figs. 5e and 5f). The methylation levels of the other three CpG sites in HYAL2-A amplicon could also efficiently distinguish BC cases from controls (Supporting Information Table 8).
Discussion Although several studies have analyzed BC-related methylation in peripheral blood in epigenome-wide analyses,12,20,32–34 this is the first study that performed independent validation by two different techniques in large sample size. Based on several rounds of independent case–control studies on 1,500 subjects, our study demonstrated a strong association between decreased methylation of HYAL2 in peripheral blood and increased risk of BC. In addition, we have replicated the previously reported BC-associated methylation variability in blood.18,19 The failure of validation for the BC-associated aberrant SLITRK4 methylation also suggested that an independent validation is necessary for further application of array-based data. Differences in methylation profiles might be influenced by the proportions of the leukocyte subpopulations, if cell distribution itself differs by disease status. According to our Illumina 27K data, BC patients had higher granulocytes proportion than controls (68.9 vs. 63.2%), but not in T cells, natural killer cells, B cells or monocytes (Supporting Information Table 1). Given the cg27091787 methylation level of 44% in granulocytes (about 90% of the B/T-lymphocytes-
Early Detection and Diagnosis
HYAL2 methylation as a marker for the detection of early BC
Early Detection and Diagnosis
8
HYAL2 methylation in the peripheral blood
Figure 4. The methylation levels of four CpG sites in HYAL2-A amplicon in sorted leukocyte fractions. The methylation levels were measured in triplicates in the samples (DNA from whole blood and from sorted leukocyte fractions) from seven sporadic BC cases and 14 healthy controls. The methylation difference between cases and controls was calculated by t-test. The methylation levels of cg27091787 are presented by box and whisker plot. The circle indicates an outlier.
depleted leukocytes were granulocytes; Supporting Information Table 1 and Fig. 4), the change of granulocytes proportion maximally contributed to 2.5% of the methylation change in whole blood, in contrast to the observed much larger HYAL2 methylation difference between BC cases and controls. In addition, if the observed HYAL2 methylation change was purely due to the change of leukocyte subpopulation proportion in blood, we should expect similar pattern of altered methylation in the same gene region. However, we just observed BC-associated altered methylation mostly in HYAL2-A amplicon and slightly in HYAL2-B amplicon, but not in HYAL2-C amplicon (Fig. 2 and Supporting Information Table 5). Taken together, we concluded that the change of leukocyte subpopulation proportion is not the only reason for the observed differential HYAL2 methylation in the peripheral blood. The other possibility for the origin of the altered HYAL2 methylation in blood is the circulating tumor DNA. As an enzyme that degrades hyaluronan, HYAL2 is known to be a tumor suppressor gene involved in cell adhesion, cell mobility, chemokinesis, cancer progression, angiogenesis and
metastasis.35–39 Our results also approved the tumor suppressor character of HYAL2 in tissue by observing higher HYAL2 methylation levels in malignant BC tumors than in benign breast tissues (Supporting Information Fig. 3). As the aberrant HYLA2 methylation showed opposite direction in tissue (hypermethylation) and in blood (hypomethylation), it is not possible that the HYAL2 methylation difference was due to circulating tumor DNA. In addition, the proportion of tumor DNA in blood would be miniscule compared with blood cell DNA, and thus, would be unlikely to change the overall blood methylation values. In our study, we discovered BC-associated altered HYAL2 methylation in peripheral blood T cells, and probably also in other leukocyte subpopulations except B cells (Fig. 4). It was notable that T-cell subfraction showed larger BC-associated HYAL2 methylation difference than unfractionated leukocytes, with statistical significance even with only seven cases and 14 controls. Hereby, we suggested that the BC-associated differential HYAL2 methylation in leukocyte subpopulations (T cells and probably other cell types) is one of the main reasons for the origin of BC-associated differential methylation C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Figure 5. Methylation level of cg27091787 in peripheral blood DNA as a marker for the detection of early BC. (a–c) ROC curve analyses for the discriminatory power of cg27091787 methylation to distinguish BC cases from controls in the discovery round and two validation rounds. (d) The discrimination of BC cases with very early-stage tumor from controls by cg27091787 methylation. (e and f) Methylation levels of cg27091787 distinguish BC cases from healthy controls in different age groups. The ROC analyses were calculated by logistic regression: (a) adjusted for age and cell counts, and (b–f) adjusted for age and different batches for the measurements. The gray lines represent the line of no discrimination.
in blood. Alternatively, the methylation pattern might be affected by the lifestyle, environment factors40,41 and other BC-related factors, which are unfortunately not available in our study and warrant further investigation in the future. Moreover, it is important to note that we cannot determine whether the altered HYAL2 methylation is a response of the hematopoietic systems to the presence of tumor or is extant before tumor development and even potentiates the growth of tumor. Inversely correlated with expression (Fig. 3), the decreased HYAL2 methylation in blood cells might be biologically meaningful. However, the function of HYAL2 in hematopoietic systems is still unknown. Future studies for the mechanisms of HYAL2 in blood cells, or even immune cells may provide hits for the initiation and progression of cancer. Despite the prevalence of BC, risk prediction and early detection remains challenging. The FDA-approved bloodbased biomarkers for BC, like CA15-3 and CA 27-29, can only be used for the monitoring of relapse and treatment, but not for the diagnosis of BC.42,43 Although mammogC 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
raphy has been considered as an efficient tool for BC diagnosis, the concern of radiological exposure, the experiencedependent subjective estimation, and the lack of sensitivity to high-dense breasts in young women have limited its utility (mammography for BC detection in women
DNA methylation array analyses identified breast cancer-associated HYAL2 methylation in peripheral blood €tze1,2, Manuela Zucknick3, Christian Sutter4, Barbara Wappenschmidt5, Frederik Marme1, Rongxi Yang1,2, Katrin Pfu 6 1,2 Bin Qu , Katarina Cuk , Christoph Engel7, Sarah Schott1, Andreas Schneeweiss1, Hermann Brenner8, Rainer Claus9, Christoph Plass9, Peter Bugert10, Markus Hoth6, Christof Sohn1, Rita Schmutzler5, Claus R. Bartram4 and Barbara Burwinkel1,2 1
Molecular Biology of Breast Cancer, Department of Gynecology and Obstetrics, University of Heidelberg, Heidelberg, Germany Molecular Epidemiology (C080), German Cancer Research Center (DKFZ), Heidelberg, Germany 3 Division of Biostatistics (C060), German Cancer Research Center (DKFZ), Heidelberg, Germany 4 Institute of Human Genetics, University of Heidelberg, Heidelberg, Germany 5 Department of Gynecology and Obstetrics, Clinical Center of University of Cologne, Cologne, Germany 6 Department of Biophysics, Saarland University, Homburg (Saar), Germany 7 Institute of Medical Informatics, Statistics and Epidemiology, University of Leipzig, Leipzig, Germany 8 Division of Clinical Epidemiology and Aging Research (C070), German Cancer Research Center (DKFZ), Heidelberg, Germany 9 Division of Epigenomics and Cancer Risk Factors (C010), German Cancer Research Center (DKFZ), Heidelberg, Germany 10 Institute of Transfusion Medicine and Immunology, Medical Faculty Mannheim, University of Heidelberg, German Red Cross Blood Service Baden€rttemberg – Hessen, Mannheim, Germany Wu
Breast cancer (BC) is the leading cause of cancer-related mortality in women worldwide. Changes in DNA methylation in peripheral blood could be associated with malignancy at early stage. However, the BC-associated DNA methylation signatures in peripheral blood were largely unknown. Here, we performed a genome-wide methylation screening and identified a BC-associated differentially methylated CpG site cg27091787 in the hyaluronoglucosaminidase 2 gene (HYAL2) (discovery round with 72 BC case and 24 controls: p 5 2.61 3 1029 adjusted for cell-type proportions). The substantially decreased methylation of cg27091787 in BC cases was confirmed in two validation rounds (first validation round with 338 BC case and 507 controls: p < 0.0001; second validation round with 189 BC case and 189 controls: p < 0.0001). In addition to cg27091787, the decreased methylation of a 650-bp CpG island shore of HYAL2 was also associated with increased risk of BC. Moreover, the expression and methylation of HYAL2 were inversely correlated with a p-value of 0.006. To note, the BC-associated decreased HYAL2 methylation was replicated in the T-cell fraction (p 5 0.034). The cg27091787 methylation level enabled a powerful discrimination of earlystage BC cases (stages 0 and I) from healthy controls [area under curve (AUC) 5 0.89], and was robust for the detection of BC in younger women as well (age < 50, AUC 5 0.87). Our study reveals a strong association between decreased HYAL2 methylation in peripheral blood and BC, and provides a promising blood-based marker for the detection of early BC.
Breast cancer (BC) is the most common cancer and the leading cause of cancer-related mortality among women,1 with 1.7 million new cases and 522,000 deaths in 2012.2 In the
United States, one in eight women will develop BC in her lifetime.3 Although therapeutic advances have improved the survival rate, most BC patients still suffer from greatly
Key words: HYAL2, methylation, breast cancer, early detection, marker Abbreviations: AUC: area under curve; BC: breast cancer; FDR: false discovery rate; HYAL2: hyaluronoglucosaminidase 2 gene; IQR: interquartile range; OR: odds ratio; ROC: receiver operating characteristic; SLITRK4: slit and trk like 4 protein gene; SNP: single nucleotide polymorphism Additional Supporting Information may be found in the online version of this article Conflict of interest: Rongxi Yang and Barbara Burwinkel are inventors of a provisional patent application relating to the subject matter of this manuscript and therefore declare a potential conflict of interests Grant sponsor: Deutsche Krebshilfe; Grant number: 107054; Grant sponsors: Dietmar-Hopp Foundation, University Hospital of Heidelberg, Helmholtz Society and the German Cancer Research Center (DKFZ) and Sonderforschungsbereich (SFB) 894 DOI: 10.1002/ijc.29205 History: Received 17 Apr 2014; Accepted 26 Aug 2014; Online 12 Sep 2014 Correspondence to: Rongxi Yang or Barbara Burwinkel, Molecular Biology of Breast Cancer, University Women’s Clinic University Heidelberg, Im Neuenheimer Feld 440, 69120 Heidelberg, Germany, Tel.: 149-6221-568403, Fax: 149-6221-565356, E-mail: [email protected] or [email protected]
C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Early Detection and Diagnosis
2
2
HYAL2 methylation in the peripheral blood
Early Detection and Diagnosis
What’s new? Altered DNA methylation in peripheral blood may be indicative of early-stage breast cancer, but methylation signatures await validation. The present study is among the first to identify and validate decreased methylation at CpG site cg27091787 in the HYAL2 gene as a potential marker of breast cancer. Methylation levels at cg27091787 were inversely associated with HYAL2 expression and were found to be indicative of breast cancer stage, thus enabling discrimination of early-stage disease, as well as facilitating disease detection in women under age 50. Differential methylation at cg27091787 could serve as a bloodbased marker for early breast cancer detection.
reduced quality of life and even metastasis owing to inefficient risk evaluation or delayed diagnosis.4,5 BC is a complex and heterogeneous disease caused by the interactions of both genetic and nongenetic factors. Age, female gender and family history are the major risk factors for BC.4 However, even combining the three major factors with other known BC risk factors, such as exposure to the hormones, life styles, medical and reproductive factors, the predictive accuracy of present model for BC in individuals is only about 58–59%.6,7 The known rare inherited mutations in the BC susceptibility genes, such as BRCA1, BRCA2, P53, PTEN, CHEK2 and ATM, can explain about one-third of the familial BCs, but together only account for 1.5–3% of all BCs.8 Although recent genomewide association studies have successfully detected multiple variants with low-penetrance risk to BC, a panel of ten such SNPs only represents a predictive accuracy for BC of 59.7%.6 Thus, the known environmental and genetic risk factors have limited utility in evaluating the risk of BC. DNA methylation harbors critical relevance for the control of gene transcriptional activities and the architecture of the nucleus of cells.9 The altered methylation in the promoter regions are among the earliest and most common events in the process of cancer.10 Given the accessibility of methodology,11,12 genome-wide methylation analyses have been extended to peripheral blood in case–control studies of ovarian, bladder and head and neck cancers.13–15 A few aberrant methylation patterns in peripheral blood have also been reported in BC by candidate gene approaches.16–19 Recently, Xu et al. reported an extensive blood-based epigenome-wide association study of BC by Illumina 27K Methylation Array in a well-designed prospective sister study cohort, and demonstrated that the methylation profiling of blood are promising markers for BC detection and risk evaluation.20 Hereby, we conducted an Infinium 27K Methylation Array on peripheral blood DNA in case–control study, and performed further independent validations by MassARRAY for BC-associated methylation variability. In addition, the association was replicated in blood cell subtypes and the correlation between gene expression and methylation was analyzed.
ples from BC cases and healthy controls were obtained from centers in Southwest Germany. All the individuals were Caucasian. All the recruited cases and controls gave written informed consent. Genomic DNA was isolated from peripheral whole blood using DNA isolation kits from Qiagen. Detailed information for the samples is shown in Table 1. Peripheral blood samples from familial BC patients: peripheral whole-blood samples from BRCA1/2 mutationnegative index familial BC patients were collected by the centers of the German Consortium for Hereditary Breast and Ovarian Cancer (GC-HBOC) during the years 1997–2007. German index patients were first diagnosed with BC and then referred to a family registry. All the familial BC cases were recruited according to the criteria of family history described previously.21 In the GC-HBOC study, about 97% of patients agreed to use their blood samples for research purposes.22 A total of 410 familial BC samples were selected randomly for our study from the GC-HBOC study in Heidelberg and Cologne. Peripheral blood samples from sporadic BC patients: Peripheral whole-blood samples were consecutively collected at University Hospital of Heidelberg from 2006 to present. In our study, about 90% of patients agreed to use their blood samples for research purposes. A total of 189 sporadic BC samples collected in the years of 2009 and 2010 were randomly selected for our study. The clinical characteristics of BC patients were defined according to the American Joint Committee on Cancer staging manual.23 Peripheral blood samples from controls: Peripheral wholeblood samples were consecutively collected from blood donors by the German Red Cross Blood Service of BadenW€ urttemberg-Hessen (Mannheim, Germany) from 2004 to present. About 95% of blood donors agreed to use their blood samples for research purposes.22 All control individuals were healthy when donating their blood. None of the control individuals had a reported family history of BC.22 A total of 720 healthy and unrelated females were randomly selected during the years 2004–2010 from the blood bank as controls for our study. No further inclusion criteria were applied during recruitment of controls.
Material and Methods Study population
Leukocytes
Our study was approved by the Ethics Committee of the University of Heidelberg (Germany). Peripheral blood sam-
Peripheral blood samples from 36 sporadic BC patients and 40 healthy female controls were collected at the University C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
3
Yang et al.
Table 1. Sample description
Rounds
Sample types
Target N
Assayed N
Call rate (%)
Mean of age (range)
Controls
24
24
100.0
48.5 (32–66) 44.5
Familial BC cases
72
72
100.0
47.0 (26–78) 46
Controls
507
491
96.8
44.7 (30–67) 42
Familial BC cases
338
328
97.0
46.1 (24–77) 46
Controls
189
185
97.9
61.2 (31–69) 64
Sporadic BC cases
189
183
96.8
59.6 (32–87) 60
Controls
96
95
99.0
46.0 (30–67) 43
Familial BC cases
96
96
100.0
48.0 (24–78) 47
Controls
93
92
98.9
60.0 (36–67) 64
Sporadic BC cases
95
95
100.0
59.4 (39–87) 59
40
38
95.0
38.2 (21–63) 39
36
34
94.4
58.2 (29–81) 57
MassARRAY Benign breast tissue 14 real-time PCR
14
100.00 47.7 (32–65) 45.5
Assays
Groups
Median of age
Determine the association between the methylation level of cg27091787 and BC1 Discovery round
Peripheral blood DNA
Illumina 27K Assay
First validation round
Peripheral blood DNA
MassARRAY
Second validation round
Peripheral blood DNA
MassARRAY
Discovery of HYAL2 Peripheral blood DNA MassARRAY CpG island shore2
Validation of HYAL2 Peripheral blood DNA MassARRAY CpG island shore3
HYAL2 methylation and expression in leukocytes
DNA and RNA from peripheral blood leukocytes
MassARRAY, Controls real-time PCR Sporadic BC cases
HYAL2 methylation and expression in tissue
HYAL2 methylation in different blood cell types
1 2 3
DNA and RNA from tissue
Different blood cell types
MassARRAY
Malignant BC tumor
34
32
94.12
56.7 (29–81) 54.6
Controls
14
14
100.0
38.3 (21–51) 42
Sporadic BC cases
7
7
100.0
50.3 (34–76) 49
All the rounds in the sample description are independent from each other except 2 and 3. Samples were randomly selected from the first validation round. Samples were randomly selected from the second validation round.
Hospital of Heidelberg. Leukocytes were isolated freshly from peripheral blood using red blood cell lysis buffer as described previously24 within 2 hr of blood collection. The leukocyte pellets were immediately frozen in liquid nitrogen after isolation and kept at 280 C before use. DNA and RNA were isolated from leukocytes using AllPrep DNA/RNA/Protein Mini Kit from Qiagen. Cases and controls were always processed in parallel.
R CD3 positive isolation kit (Invitrogen, Carlsbad, with a DynalV U.S.). The leftover cells were collected as “B/T-lymphocytesdepleted leukocytes.” The cell pellets were immediately frozen in liquid nitrogen after purification and kept at 280 C before use. DNA was isolated from different blood cell types using AllPrep DNA/RNA/Protein Mini Kit from Qiagen (Germany). All the cases and controls were processed in parallel.
Blood cell fractionations from leukocytes
Tissues
Leukocytes were freshly obtained as described from seven sporadic BC patients and 14 female healthy controls at the University Hospital of Heidelberg. First, B cells were positively R CD19 positive isolation kit (Invitroisolated using a DynalV gen, Carlsbad, U.S.) from fresh leukocytes. Subsequently these B cell-depleted leukocytes were used for T-cell purification
Tissue samples from 14 benign breast tissues and 34 malignant BC tumors were obtained at the University Hospital of Heidelberg before any BC treatment. The tissues were immediately frozen in liquid nitrogen after obtained and kept at 280 C before use. DNA and RNA were isolated from tissue using AllPrep DNA/RNA/Protein Mini Kit from Qiagen
C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Early Detection and Diagnosis
Further exploration of HYAL2 methylation
4
(Germany). Malignant and benign samples were always processed in parallel. Infinium 27K methylation assay
In the discovery round, 500 ng of genomic DNA from each sample was bisulfite converted using the EZ-96 DNA Methylation Kit (Zymo Research, Orange County, U.S.) and subsequently subjected to the genome-wide methylation screening using the Human Methylation27 BeadChip (Illumina, San Diego, California, U.S.) according to the manufacturer’s recommendations.25 All the 96 samples passed the quality control.
Early Detection and Diagnosis
MALDI-TOF mass spectrometry
MALDI-TOF mass spectrometry (Sequenom, San Diego, California, U.S.) described by Breitling et al.11,26 was used in all the validation and further exploring rounds. In brief, DNA was bisulfite converted by EZ-96 DNA Methylation Gold Kit (Zymo Research, Orange County, U.S.) and amplified by bisulfite-specific primers. The bisulfite-specific primers (no single nucleotide polymorphisms (SNPs) in the primers) and polymerase chain reaction (PCR) amplicons are presented in Supporting Information Figure 1. The PCR products were treated according to the standard protocol of SequenomEpiTyper Assay, and further cleaned by Resin and dispensed to a 384 SpectroCHIP by a Nanodispenser. The chips were read by a Sequenom Mass Spectrometer system. Data were collected by SpectroACQUIRE v3.3.1.3 software and visualized with MassArrayEpiTyper v1.2 software. For each batch of MassARRAY analysis, same amount of cases and controls were randomly selected from the cohort. The samples from BC cases and controls were treated and analyzed in parallel in all the processes. Five percent of the samples were randomly chosen for a duplication analysis and achieved a Spearman correlation of 0.84, and a coefficient of variation of 0.07 (95% CI: 0.06–0.09). Quantitative real-time PCR
Hundred nanograms of total RNA from each sample was tranR Reverse Transcription scribed to cDNA using TaqManV Reagents (Applied Biosystems, Germany). Quantitative realtime PCR was performed using a LightCycler 480 (Roche, Germany) in combination with TaqMan gene expression assays (Applied Biosystems) for HYAL2 gene and housekeeping gene HPRT1 (hypoxanthine-guanine phosphoribosyltransferase1) as endogenous control. Crossing point values were calculated using the second-derivative maximum method by the LightCycler 480 basic software (Roche, Germany). Relative expression of HYAL2 for each sample was calculated according to DDCt method27 by normalization to HPRT1. Sequencing
The genomic DNA from the 80 samples (20 familial BC cases and 20 controls with the highest cg27091787 methylation, 20 familial BC cases and 20 controls with the lowest cg27091787 methylation) was used for the detection of germline mutation
HYAL2 methylation in the peripheral blood
in the vicinity of HYAL2-A amplicon. PCR primers were designed to amplify a 568-bp segment (chr3:5033532250335889, 36.1/hg18) covering the whole HYAL2-A amplicon and two database-annotated very rare SNPs in vicinity (rs34465081, no frequency data in dbSNP database; rs4688732, minor allele frequency 5 0.004 in Caucasian). The primer pairs were as follows: forward: 50 -ACCCTGTTTCCCTCC AGAAT-30 and reverse: 50 -CACCCCTCCCAGAGGAGTAT-30 . The PCR products were treated with the ExoSAP-IT purification kit (GE Healthcare, Germany) and sequenced using the forward and reverse primers from both directions via a 3130XL Genetic Analyzer (Applied Biosystems, Germany). Sequencing results were analyzed using Sequencing Analysis 5.2 software (Applied Biosystems, Germany). Statistical analyses
The Illumina 27K Array data were processed by the Illumina BeadStudio software with default settings. Probes with detection p-value > 0.01 were removed and samples were quantile-normalized. Association of probes with case–control status was assessed by beta-regression models with a logistic link and associated Wald tests using R.28 Likelihood ratio tests were used to compare the case–control model with the nested model for chip differences. Multiple testing adjustments were done with the Benjamini-Hochberg method controlling the false discovery rate (FDR) at the level of 0.05. The cell-type proportions were adjusted for the Illumina 27K Array by fitting data linear mixed effects models.29 All the statistical analyses of MassARRAY and gene expression data were conducted by SPSS Statistics 17.0. The correlations were assessed by Spearman’s rank correlation coefficients. Logistic regression models and nonparametric tests were used for comparisons between two and multiple groups. The results of logistic regression were adjusted for possible confounding effects of age and different measurement batches by including additional covariables in the logistic regression models. Receiver operating characteristic (ROC) curve analysis was performed to assess the discriminatory power of methylation levels. All statistical tests were two-sided, and p values 5%, Supporting Information Fig. 2). The CpG loci in the SFRP1, PTGS2 and BRCA1 genes also exhibited lower methylation in familial BC patients compared to controls (all FDR-adjusted p-value: 69%), the three lower quartiles of cg27091787 methylation levels (Q1–Q3) were highly associated with an increased risk of BC, and this association was especially pronounced in the lowest quartile (Q1, methylation level 54%, OR 5 41.5, 95% CI: 22.2–74.4, p < 0.0001 by logistic regression, Supporting Information Table 3). To exclude the possibility that the observed altered HYAL2 methylation was due to the influence of genetic variants, we sequenced the HYAL2-A amplicon and two nearby rare single nucleotide polymorphisms (SNPs) in the peripheral blood DNA from 80 samples with the highest and lowest cg27091787 methylation in BC cases and controls. No mutations or SNPs were found (data not shown, see Material and Methods section). In the second independent validation round, the differential HYAL2 methylation in peripheral blood DNA was investigated in 189 sporadic BC cases and 189 controls (Table 1). Again, cg27091787 site showed lower methylation levels in cases than in controls [median 50.50 (IQR 5 0.46–0.54) and 0.64 (IQR 5 0.58–0.70) for BC cases and controls, respectively, OR per 10% methylation 5 8.14, p < 0.0001 by logistic regression, Fig. 1c, Supporting Information Table 2]. Compared to the highest quartile of cg27091787 methylation (Q4, methylation level > 65%), the other three quartiles (Q1–Q3) were associated with a significantly increased risk of BC, especially the lowest quartile (Q1, methylation level
Early Detection and Diagnosis
The validation of BC-associated blood-based HYAL2 methylation by two independent case–control studies
6
HYAL2 methylation in the peripheral blood
MassARRAY (Supporting Information Fig. 1). However, none of CpG sites in SLITRK4 passed the first validation round (ORs per 10% methylation from 0.92 to 1.04, p > 0.5 for all by logistic regression, Supporting Information Table 4).
Early Detection and Diagnosis
The identification and validation of BC-associated HYAL2 CpG island shore
Figure 2. Definition of BC-associated CpG island shore of HYAL2. (a) Schematic diagram of HYAL2 promoter regions. CpG island-1 and CpG island-2 are located before the translation start site (TSS). The second exon could also be an intron due to variant splicing. The 61,000-bp flanking CpG sites of cg27091787 are presented. The HYAL2-A amplicon covers the CpG cg27091787, HYAL2-B amplicon covers the region between HYAL2-A amplicon and the CpG island-1 and HYAL2-C amplicon covers most region of the CpG island-1. (b) Discovery of the BC-associated significant methylation differences in amplicon HYAL2-A and HYAL2-B in 96 familial BC cases and 96 controls randomly selected from the first validation round. (c) Validation of the BC-associated significant methylation differences in amplicon HYAL2-A and HYAL2-B in 95 sporadic BC cases and 93 controls randomly selected from the second validation round. The p-values of all the 32 measureable CpG loci in the three amplicons were calculated by logistic regression adjusted for age and different batches for the measurements. The lines indicate the thresholds of different p values of 0.05, 0.01 and 0.001.
49%, OR 5 133.0, 95% CI: 40.7–434.5, p < 0.0001 by logistic regression, Supporting Information Table 3). The methylation levels of all the three additional CpG loci in HYAL2-A amplicon were substantially correlated with the methylation level of cg27091787 and were all significantly associated with BC in both validation rounds (all Spearman’s q > 0.55, all OR per 10% methylation > 1.80, p < 0.0001 by logistic regression, Fig. 1, Supporting Information Table 2). An amplicon SLITRK4 covering the cg12842316 site and two additional measurable flanking CpG sites was designed for
Next, we attempted to find the BC-associated differentially methylated region in HYAL2. Two additional amplicons, HYAL2-B and HYAL2-C, were designed to cover the CpGdense region, locating from the downstream of HYAL2-A amplicon to the closest CpG island (Fig. 2a and Supporting Information Fig. 1). Ninety-six familial BC cases and 96 controls randomly selected from the first validation round were analyzed by MassARRAY. Interestingly, the methylation levels of all 11 measureable CpG loci in HYAL2-B amplicon were significantly lower in cases than in controls (ORs per 10% methylation from 1.77 to 3.52, p < 0.017 for all), and were substantially correlated with the methylation levels of cg27091787 (Spearman rho from 0.48 to 0.65). In contrast, all the 17 measureable CpG loci in HYAL2-C amplicon showed no or weak association with BC (ORs per 10% methylation from 0.94 to 2.24, p-values from 0.623 to 0.007) and had weaker correlation with the methylation levels of cg27091787 (Spearman rho from 0.20 to 0.46) (Fig. 2b and Supporting Information Table 5). The association between altered methylation in HYAL2-A and HYAL2-B amplicons and BC was further validated by 95 sporadic BC cases and 93 controls randomly selected from the second validation round (Fig. 2c and Supporting Information Table 6). Together, we have defined a 650-bp BC-associated differentially methylated region or so-called CpG-island shore in HYAL2 from cg27091787 to the beginning of the adjacent CpG island. The inverse correlation between the methylation and expression of HYAL2 in peripheral blood leukocytes
To understand the correlation between methylation and the expression of HYAL2 gene, DNA and RNA were isolated from peripheral blood leukocytes of 36 sporadic BC patients and 40 healthy controls (Table 1). Significantly lower methylation can be observed in all the four CpG sites in the HYAL2-A amplicon in the leukocytes DNA from BC cases compared to that from controls (Fig. 3a). In contrast, the median of relative expression levels of HYAL2 in the leukocytes of sporadic BC cases was 1.3-fold higher than in the controls (Fig. 3b). The methylation levels of cg27091787 were significantly and inversely correlated with the expression levels of HYAL2 with a Spearman rho of 20.323 (p 5 0.006, Fig. 3c). The methylation levels of cg27091787 flanking CpG sites, HYAL2-A_CpG_1 and HYAL2-A_CpG_3, showed even stronger inverse correlation with the expression levels of HYAL2 (Spearman rho 5 20.431 and 20.452, respectively). Replication of HYAL2 methylation differences in T cells
Next, we attempted to replicate the methylation differences in freshly sorted leukocyte fragments. We separated B cells, T C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
7
Yang et al.
without the pattern of changed methylation in all four CpG sites in HYAL2-A amplicon, the weak correlations by individual CpG sites should be taken with caution.
Figure 3. The inverse correlation between the methylation and expression of HYAL2 in leukocytes. (a) The box plots show the methylation levels of cg27091787 and adjacent CpG sites in the HYAL2-A amplicon in leukocytes from 36 sporadic BC cases and 40 healthy controls. The box plot of cg27091787 is framed in box for emphasis. (b) The box plot shows the expression level of HYAL2 in leukocytes from sporadic BC cases and healthy controls. The presented p-values were calculated by Mann–Whitney U-test. The circles indicate outliers. (c) The inverse correlation between the methylation level of cg27091787 and HYAL2 expression in leukocytes.
cells and the B/T-lymphocytes-depleted leukocytes (mainly granulocytes, monocytes and NK cells) from seven sporadic BC cases and 14 healthy controls (Table 1). For the T-cell fraction, all the four CpG loci in HYAL2-A amplicon showed lower methylation in BC cases than in the controls. Of these, cg27091787 was significant at p 5 0.034 and HYAL2-A_CpG_1 was significant at p 5 0.043, even with very small sample number (Fig. 4). To note, the T cells also showed larger HYAL2 methylation difference than that seen in the whole blood. Compared to the controls, the HYAL2 methylation levels in BC cases were also lower in B/T-lymphocytes-depleted leukocytes, but higher in B cells (Fig. 4). HYAL2 methylation and the clinical characteristics of BC
To explore the relationship between blood-based HYAL2 methylation and the clinical characteristics of BC, the sporadic BC patients with available clinical data were interpreted. The methylation levels of all CpG sites in HYAL2-A amplicon showed no correlation with tumor size, grading, menopausal status, the status of Her2 receptor and family history of BC. Weak correlations were observed between individual CpG sites and stage, lymph nodes involvement and ER/PR receptor status (Supporting Information Table 7). However, C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
To estimate the potential clinical utility of HYAL2 methylation as a marker for the detection of early BC, ROC curve analyses were performed adjusted for possible confounding effects by logistic regression. In the discovery round, the methylation level of cg27091787 showed robust discriminatory power for differentiating BC cases from healthy controls when adjusted for age and cell counts [area under curve (AUC) 5 0.95, 95% CI: 0.90–1.00, Fig. 5a]. The discriminatory power of cg27091787 methylation for BC was verified in the validation rounds (adjusted for age and different measurement batches, first validation round, AUC 5 0.83, 95% CI: 0.80–0.86; second validation round, AUC 5 0.89, 95% CI: 0.85–0.92, Figs. 5b and 5c). In addition, the HYAL2 methylation level had outstanding power to distinguish very early BC from controls (adjusted for age and different measurement batches, stages 0 and I cases vs. controls, AUC 5 0.89, 95% CI: 0.85–0.93, Fig. 5d). Although age is a limited factor for the utility of mammography, it had no influence on the discriminatory power of cg27091787 methylation, and was even a benefit for the younger women (adjusted for age and different measurement batches age < 50 years: AUC 5 0.87, 95% CI: 0.84–0.89; age 50 years, AUC 5 0.83, 95% CI: 0.80–0.87, Figs. 5e and 5f). The methylation levels of the other three CpG sites in HYAL2-A amplicon could also efficiently distinguish BC cases from controls (Supporting Information Table 8).
Discussion Although several studies have analyzed BC-related methylation in peripheral blood in epigenome-wide analyses,12,20,32–34 this is the first study that performed independent validation by two different techniques in large sample size. Based on several rounds of independent case–control studies on 1,500 subjects, our study demonstrated a strong association between decreased methylation of HYAL2 in peripheral blood and increased risk of BC. In addition, we have replicated the previously reported BC-associated methylation variability in blood.18,19 The failure of validation for the BC-associated aberrant SLITRK4 methylation also suggested that an independent validation is necessary for further application of array-based data. Differences in methylation profiles might be influenced by the proportions of the leukocyte subpopulations, if cell distribution itself differs by disease status. According to our Illumina 27K data, BC patients had higher granulocytes proportion than controls (68.9 vs. 63.2%), but not in T cells, natural killer cells, B cells or monocytes (Supporting Information Table 1). Given the cg27091787 methylation level of 44% in granulocytes (about 90% of the B/T-lymphocytes-
Early Detection and Diagnosis
HYAL2 methylation as a marker for the detection of early BC
Early Detection and Diagnosis
8
HYAL2 methylation in the peripheral blood
Figure 4. The methylation levels of four CpG sites in HYAL2-A amplicon in sorted leukocyte fractions. The methylation levels were measured in triplicates in the samples (DNA from whole blood and from sorted leukocyte fractions) from seven sporadic BC cases and 14 healthy controls. The methylation difference between cases and controls was calculated by t-test. The methylation levels of cg27091787 are presented by box and whisker plot. The circle indicates an outlier.
depleted leukocytes were granulocytes; Supporting Information Table 1 and Fig. 4), the change of granulocytes proportion maximally contributed to 2.5% of the methylation change in whole blood, in contrast to the observed much larger HYAL2 methylation difference between BC cases and controls. In addition, if the observed HYAL2 methylation change was purely due to the change of leukocyte subpopulation proportion in blood, we should expect similar pattern of altered methylation in the same gene region. However, we just observed BC-associated altered methylation mostly in HYAL2-A amplicon and slightly in HYAL2-B amplicon, but not in HYAL2-C amplicon (Fig. 2 and Supporting Information Table 5). Taken together, we concluded that the change of leukocyte subpopulation proportion is not the only reason for the observed differential HYAL2 methylation in the peripheral blood. The other possibility for the origin of the altered HYAL2 methylation in blood is the circulating tumor DNA. As an enzyme that degrades hyaluronan, HYAL2 is known to be a tumor suppressor gene involved in cell adhesion, cell mobility, chemokinesis, cancer progression, angiogenesis and
metastasis.35–39 Our results also approved the tumor suppressor character of HYAL2 in tissue by observing higher HYAL2 methylation levels in malignant BC tumors than in benign breast tissues (Supporting Information Fig. 3). As the aberrant HYLA2 methylation showed opposite direction in tissue (hypermethylation) and in blood (hypomethylation), it is not possible that the HYAL2 methylation difference was due to circulating tumor DNA. In addition, the proportion of tumor DNA in blood would be miniscule compared with blood cell DNA, and thus, would be unlikely to change the overall blood methylation values. In our study, we discovered BC-associated altered HYAL2 methylation in peripheral blood T cells, and probably also in other leukocyte subpopulations except B cells (Fig. 4). It was notable that T-cell subfraction showed larger BC-associated HYAL2 methylation difference than unfractionated leukocytes, with statistical significance even with only seven cases and 14 controls. Hereby, we suggested that the BC-associated differential HYAL2 methylation in leukocyte subpopulations (T cells and probably other cell types) is one of the main reasons for the origin of BC-associated differential methylation C 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
Figure 5. Methylation level of cg27091787 in peripheral blood DNA as a marker for the detection of early BC. (a–c) ROC curve analyses for the discriminatory power of cg27091787 methylation to distinguish BC cases from controls in the discovery round and two validation rounds. (d) The discrimination of BC cases with very early-stage tumor from controls by cg27091787 methylation. (e and f) Methylation levels of cg27091787 distinguish BC cases from healthy controls in different age groups. The ROC analyses were calculated by logistic regression: (a) adjusted for age and cell counts, and (b–f) adjusted for age and different batches for the measurements. The gray lines represent the line of no discrimination.
in blood. Alternatively, the methylation pattern might be affected by the lifestyle, environment factors40,41 and other BC-related factors, which are unfortunately not available in our study and warrant further investigation in the future. Moreover, it is important to note that we cannot determine whether the altered HYAL2 methylation is a response of the hematopoietic systems to the presence of tumor or is extant before tumor development and even potentiates the growth of tumor. Inversely correlated with expression (Fig. 3), the decreased HYAL2 methylation in blood cells might be biologically meaningful. However, the function of HYAL2 in hematopoietic systems is still unknown. Future studies for the mechanisms of HYAL2 in blood cells, or even immune cells may provide hits for the initiation and progression of cancer. Despite the prevalence of BC, risk prediction and early detection remains challenging. The FDA-approved bloodbased biomarkers for BC, like CA15-3 and CA 27-29, can only be used for the monitoring of relapse and treatment, but not for the diagnosis of BC.42,43 Although mammogC 2014 UICC Int. J. Cancer: 00, 00–00 (2014) V
raphy has been considered as an efficient tool for BC diagnosis, the concern of radiological exposure, the experiencedependent subjective estimation, and the lack of sensitivity to high-dense breasts in young women have limited its utility (mammography for BC detection in women
