SEARCH

SEARCH BY CITATION

Keywords:

  • gene expression;
  • methylation;
  • microarray analysis;
  • prognosis;
  • urinary bladder neoplasms

Abstract

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

DNA methylation patterns are associated with the development and prognosis of cancer. The aim of this study was to identify novel methylation markers for the prediction of patient outcomes using microarray analysis of DNA methylation and RNA expression patterns in samples from long-term follow-up patients with nonmuscle invasive bladder cancer (NMIBC). A total of 187 human bladder specimens were used for microarray array or pyrosequencing (PSQ) analyses: 6 normal controls (NC) and 181 NMIBC. Tumor-specific hypermethylated genes were selected from a data set comprising 24 matched microarray-based DNA methylation and gene expression profiles (6 controls and 18 NMIBC), and their clinical relevance was verified by quantitative PSQ analysis. The methylation status of Homeobox A9 (HOXA9), ISL LIM homeobox 1 (ISL1) and Aldehyde dehydrogenase 1 family, member A3 (ALDH1A3) was significantly associated with decreased gene expression levels and aggressive clinicopathological characteristics. Multivariate regression analyses showed that hypermethylation of these genes was an independent predictor of disease recurrence (HOXA9, ISL1 and ALDH1A3, either alone or in combination) and progression (ISL1 and ALDH1A3, either alone or in combination) (each p < 0.05). The results of this study suggest that these novel methylation markers are independent prognostic indicators in NMIBC patients, which may facilitate the assessment of disease recurrence and progression in NMIBC patients and inform clinical decision making regarding treatment.

Abbreviations
ALDH1A3

Aldehyde dehydrogenase 1 family, member A3

EOMES

Eomesodermin

HOXA9

Homeobox A9

HR

Hazard ratio

ISL1

ISL LIM homeobox 1

M score

methylation score

NC

normal controls

NMIBC

non-muscle invasive bladder cancer

PSQ

pyrosequencing

ROC

receiver operating characteristic

TUR

transurethral resection

Nonmuscle invasive bladder cancer (NMIBC) comprises a heterogeneous cell population, and numerous factors are likely to be involved in disease outcome. Unfortunately, many patients with NMIBC are at high risk of disease recurrence and progression after primary treatment.[1, 2] The challenge for a clinician is to develop both reasonable surveillance protocols that provide cost-effective, noninvasive monitoring for low-risk patients and more aggressive approaches to identify high-risk refractory cancers before they progress.

While several molecular markers used to evaluate the development and prognosis of NMIBC have been studied, the limited value of these established markers has created the need for new molecular indicators of NMIBC.[3] DNA hypermethylation-induced silencing of tumor suppressor and DNA repair genes is a frequent phenomenon in cancer, which has led to new opportunities for the understanding, detection, treatment, and prevention of the disease, including bladder cancer.[4-13] Although many of the different genetic or epigenetic changes that lead to aberrant gene expression in bladder cancer have been identified, the advent of high-throughput microarray technology makes it possible to gain a comprehensive insight into the molecular basis of human disease.[6, 14] Using this technology, genome-wide DNA methylation patterns and RNA expression levels in tumor specimens can be evaluated simultaneously, and tumor-cell specific molecular targets or gene classifiers can be identified.[15]

NMIBC is increasingly regarded as a disease that cannot be treated solely on the basis of clinicopathological parameters. Identification of biomarkers that enable detection or predict the prognosis of NMIBC would be a valuable tool for guiding appropriate management strategies. Therefore, the aim of this study was to identify novel methylation markers that predict patient outcomes using microarray analysis of DNA methylation and RNA expression patterns in long-term follow-up NMIBC samples.

Material and Methods

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

Subjects and sample collection

A total of 187 human bladder specimens were used for methylation array or pyrosequencing (PSQ) analyses: 6 normal controls (NC) and 181 NMIBC (Table 1). NMIBC specimens were obtained from 181 primary NMIBC patients who underwent transurethral resection (TUR) for histologically diagnosed transitional cell carcinomas between 1995 and 2010 at our institute. To exclude the possibility of incomplete resection or confounding factors that might unduly affect the analyses, patients followed-up for less than 6 months or those that experienced disease relapse within 6 months were excluded from the study. Samples of normal bladder urothelium obtained from individuals with benign prostate hyperplasia, bladder injury, or bladder stones were used as controls. Microarray gene expression data for 89 specimens (NC = 6, NMIBC = 83) were available from our previously published study.[16, 17] The study design and validation strategies are outlined in Figure 1.

Table 1. Baseline characteristics of the study subjects
VariablesNC (n = 6)NMIBC (n = 181)
  1. NC: normal control; NMIBC: nonmuscle invasive bladder cancer.

Age, yrs (mean)56.3 ± 25.564.3±13.8
Gender no. of patients (%)  
Male5 (83.3)147 (81.2)
Female1 (16.7)34 (18.8)
No. of tumors (%)  
Single106 (58.6)
Multiple75 (41.4)
Tumor size (%)  
<3 cm99 (54.7)
≥3 cm82 (45.3)
Grade - no. of patients (%)  
G159 (32.6)
G296 (53.0)
G326 (14.4)
T Stage - no. of patients (%)  
Ta71 (39.2)
T1110 (60.8)
Recurrence free survival-months (median)35.8 (6.1–183.3)
Recurrence-no. of patients (%)  
No121 (66.9)
Yes60 (33.1)
Progression free survival-months (median)50.9 (6.6–183.3)
Progression-no. of patients (%) 
No162 (89.5)
Yes19 (10.5)
image

Figure 1. Study design and validation strategies.

Download figure to PowerPoint

All tumors were macrodissected within 15 min of surgical resection. Each NMIBC specimen was confirmed by pathological analysis of a part of the tissue sample (i.e., sections taken from TUR specimens and then snap-frozen in liquid nitrogen and stored at −80°C). The specimens were provided by the Chungbuk National University Hospital, a member of the National Biobank of Korea, which is supported by the Ministry of Health, Welfare and Family Affairs. The collection and analysis of all samples was approved by the Chungbuk National University Hospital Institutional Review Board and informed consent was obtained from each subject.

Tumors were staged according to the 2002 TNM classification and the 1973 WHO grading system.[1] A second TUR was performed 2–4 weeks after the initial resection if a bladder cancer specimen did not include proper muscle, or if a high-grade tumor was detected. Patients with intermediate- or high-risk NMIBC received one cycle of intravesical therapy. Each patient was followed-up and managed according to standard recommendations.[1, 2] Recurrence was defined as the recurrence of primary NMIBC at a lower or equivalent pathologic stage (Ta/T1), and progression was defined as muscular invasion (TNM stage T2 or higher) or metastatic disease.

DNA methylation profiling

Of the 89 bladder specimens for which microarray gene expression data were available, 24 matched DNA samples (NC = 6, NMIBC = 18) were used for DNA methylation profiling. Methylation patterns were assayed using the genome-wide Infinium array (Illumina Inc., San Diego, CA), which enables interrogation of 27,578 CpG dinucleotides covering 14,495 genes, as described previously.[18] The β value represents a quantitative measure of the DNA methylation level of specific CpG islands and ranges from 0 (completely unmethylated) to 1 (completely methylated).

PSQ analysis

The DNA methylation status of candidate NMIBC-specific hypermethylated CpG sites was assessed by PSQ using PyroMark Q96 ID (Qiagen, Valencia, CA) according to the manufacturer's instructions. PSQ primers were designed to encompass the CpG sites assayed on the Illumina Infinium array. The primer sequences and amplification conditions are described in Supporting Information Table 1.

Statistical analysis

The DNA methylation and gene expression profile data were normalized using quantile normalization in the R language environment (version 2.10.0, available at http://www.r-project.org/). The detailed analytical methods have been described previously.[16, 17] Using the 24 matched DNA methylation and expression profiles, three criteria were used to detect those methylation-silenced genes whose methylation levels differed significantly between NMIBC and NC. These criteria were: (i) a difference in DNA methylation levels between NMIBC and NC (Δβ value) > 0.4; (ii) a mean β value for NC < 0.15; and (iii) a > threefold difference in gene expression level between NMIBC and NC. To validate the genes identified in this study, we used microarray methylation data from 26 bladder tumor tissues (Ta = 17, T1 = 5, T2 = 4) and six NC tissues from a Western population.[13]

The differences in continuous variables between groups were assessed using a two sample t-test or ANOVA trend analyses using polynomial contrasts. Pearson's correlation was used to evaluate the relationship between groups with continuous variables. With the exception of 18 discovery samples, clinical relevance of candidate markers was tested with the 163 samples. Receiver operating characteristic (ROC) curves were used to estimate the capability of candidate markers for the prediction of prognosis (recurrence or progression). To minimize the biases against arbitrary cut-off point, median values were applied to divide patient into subgroups (hypomethylation or hypermethylation), and survival function of candidate genes were evaluated. As a mean of incorporating the overall extent of methylation, each patient's methylation score (M score) was calculated as the sum of the methylation levels of selected genes multiplied by the corresponding regression coefficients derived from the Cox regression analysis used to assess the predictive value of each gene for prognosis.[16, 19-21] For the multivariate Cox proportional hazards regression models, the prognostic value of methylation status was evaluated separately and adjusted for well-known clinicopathological (sex, age, tumor size, tumor number, intravesical therapy, grade, and stage) factors. Statistical analysis was performed using SPSS 12.0 software (SPSS, Chicago, IL). p < 0.05 was considered statistically significant.

Results

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

Baseline characteristics

The baseline characteristics of the NC and NMIBC patients are presented in Table 1. Mean recurrence-free survival and progression-free survival in patients with NMIBC was 47.2 ± 40.4 months (median 35.8, range 6.1–183.3) and 61.1 ± 41.7 months (median 50.9, range 6.6–183.3), respectively.

Identification of differentially methylated and expressed genes in NMIBC and NC

Genome-wide methylation and expression profiles from 18 NMIBC patients were compared with those from 6 NC. The complete sets of microarray data derived from the human bladder tissues are available online (http://www.ncbi.nlm.nih.gov/geo/) under the data series accession number GSE37817. Using highly stringent selection criteria (Δβ value > 0.4 and mean β value in NC < 0.15), we identified 42 unique CpG island loci in 39 genes that were hypermethylated in NMIBC compared with NC. To select the putative methylation-silenced genes in NMIBC, the corresponding gene expression levels of 24 human bladder specimens were also compared. Of the 42 unique hypermethylated loci, seven loci in six genes showed at least a threefold decrease in gene expression in NMIBC compared with NC. Putative relation between six candidate genes and NMIBC was provided in Supporting Information Figure S1. The validity of our candidate genes as a methylation marker for NMIBC was evaluated using an independent set of Infinium microarray methylation data derived from a Western population.[13] The Δβ values obtained in this study were almost identical to those of another study (Supporting Information Table 2).

PSQ analysis

To verify the DNA methylation level of the candidate genes using a different method, we performed PSQ analyses using bisulfite-modified genomic DNA obtained from 187 human bladder specimens (NC = 6, NMIBC = 181). PSQ analyses of four out of six candidate genes [Homeobox A9 (HOXA9), ISL LIM homeobox 1 (ISL1), Aldehyde dehydrogenase 1 family, member A3 (ALDH1A3), and Eomesodermin (EOMES)] were technically possible and were analyzed by PSQ in the current study. To test the reliability of the bisulfite PSQ technique, the values for 24 bladder specimens obtained from the Infinium array and PQS were compared. The Pearson correlation coefficient ranged from 0.715 to 0.940, which is an acceptable correlation (Supporting Information Table 3).

Methylation and gene expression patterns in NMIBC and NC

To identify DNA methylation-induced gene silencing, the correlation between DNA methylation and gene expression levels was calculated using microarray expression data and matched PSQ values from 89 subjects. With the exception of EOMES, the methylation and expression levels of HOXA9 (r = −0.453, p < 0.001), ISL1 (r = −0.501, p < 0.001) and ALDH1A3 (r = −0.150, p < 0.049) showed significant inverse correlations. Moreover, NMIBC patients showed significantly higher methylation and lower expression levels than the NC (Fig. 2, p < 0.05).

image

Figure 2. Methylation and expression levels in bladder tissues. Box plot comparing the methylation (A) and expression (B) levels of candidate genes between NC and NMIBC. Significant differences (p-value) between NC and NMIBC patients were identified using a two sample t test. Gene expression data are represented on a log2-transformed scale.

Download figure to PowerPoint

Association between methylation levels and clinicopathological variables

To evaluate the relationship between methylation patterns and clinicopathological factors, we examined methylation levels in terms of well-known prognostic factors such as tumor number, tumor size, and tumor grade and stage. Generally, increased methylation values for HOXA9, ISL1, ALDH1A3, and EOMES were significantly associated with increases in tumor number, size, grade and stage (Supporting Information Table 4).

Methylation status as a predictor of prognosis

The value of candidate methylation markers for the prediction of prognosis (recurrence or progression) was measured by the area under the ROC curve (Supporting Information Table 5). Next, to determine whether our selected methylation markers were relevant to prognosis, the methylation values of each gene were dichotomized (hypomethylation or hypermethylation) with median cut-off points. Univariate and multivariate Cox regression analyses identified methylation markers significantly related to recurrence (HOXA9, ISL1 and ALDH1A3) and progression (ISL1 and ALDH1A3) (Table 2). To integrate the methylation status of the methylation markers with prognostic relevance, the M scores for recurrence (HOXA9, ISL1 and ALDH1A3) and progression (ISL1 and ALDH1A3) were calculated separately, and the patients were classified into two groups with median values. According to multivariate Cox regression analyses, M score classifiers were independent predictors of recurrence (Hazard ratio [HR], 1.69; p = 0.041) and progression (HR, 5.57; p = 0.025, Table 2). Similarly, Kaplan–Meier estimates identified significant differences in time-to-recurrence or progression according to methylation status (Fig. 3, log-rank test: p < 0.05).

Table 2. Univariate and multivariate Cox regression analyses for determining disease outcome based on the methylation status of each gene (n = 163)
 RecurrenceProgression
UnivariateMultivariateUnivariateMultivariate
VariablesHR (95% CI)ap-valueHR (95% CI)bp-valueHR (95% CI)ap-valueHR (95% CI)bp–value
  1. a

    The hazard ratios (HR) for each variable were calculated to predict disease outcome according to methylation status (hypomethylation vs. hypermethylation).

  2. b

    All Hazard ratios (HR) were adjusted for clinicopathological factors [sex (female vs. male), age (continuous variable), tumor size (<3 cm vs. ≥ 3 cm), multiplicity (single vs. multiple), intravesical therapy (no vs. yes), tumor grade (G1 vs. G2 vs. G3), and T stage (Ta vs. T1)]. On multivariate analysis, HR was determined separately according to the methylation status of each gene and the clinicopathological factors.

HOXA92.06 (1.14–3.72)0.0161.87 (1.14–3.47)0.0323.44 (0.96–12.34)0.058-
ISL12.19 (1.21–3.89)0.0091.71 (1.05–3.47)0.0394.56 (1.25–16.59)0.0213.30 (1.05–12.92)0.041
ALDH1A31.81 (1.02–3.22)0.0431.68 (1.02–3.16)0.0446.60 (1.47–29.60)0.0143.55 (1.07–14.22)0.039
EOMES1.28 (0.73–2.26)0.3853.79 (1.06–13.59)0.0412.30 (0.54–9.72)0.257
M score1.93 (1.38–3.46)0.0371.69 (1.03–3.23)0.0418.53 (1.72–37.55)0.0085.57 (1.23–24.75)0.025
image

Figure 3. Kaplan–Meier curves predicting the probability of recurrence and progression according to methylation status (identified by pyrosequencing) in NMIBC patients. HOXA9 (a), ISL1 (b), ALDH1A3 (c), and M score (d) for recurrence, and ISL1 (e), ALDH1A3 (f), and M score (g) for progression.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

Analysis of aberrant DNA methylation is gaining traction in areas of cancer risk assessment, diagnosis, therapy monitoring, prognosis prediction, and novel drug targeting for several different types of cancer.[6-11] In this study, we used microarray analyses to detect novel epigenetic markers that are relevant to NMIBC and identified significant differences in the methylation patterns between NMIBC and NC. These candidate methylation markers showed close associations with unfavorable NMIBC features, including advanced stage and a higher grade. Notably, the methylation status of these candidate genes (either single or combined) was identified as an independent indicator for predicting prognosis.

Microarray-based clinical research has allowed comprehensive insights into the molecular basis of human disease, and new molecular targets at the whole-genome level are rapidly emerging.[15] However, the identification of bona fide candidate methylation markers that have clinical relevance requires rigorous selection criteria, confirmation of the methylation status and gene expression levels in human tissues, and comparison with clinicopathological parameters or disease outcomes.[6] For these reasons, DNA methylation status and gene expression levels were considered simultaneously when screening candidate genes in this study. The correlation between microarray methylation and PSQ analyses was checked, and concordant results were identified. Furthermore, methylation-induced gene silencing and prognostic outcomes were verified in long-term follow-up NMIBC patients. The results suggest that the novel methylation markers identified are specific to NMIBC and are appropriate for predicting prognosis.

There are an increasing number of reports showing that hypermethylation of individual genes can indicate different outcomes for bladder cancer patients.[9-12] However, a normal cell must undergo multiple genetic alterations before adopting a malignant and, ultimately, metastatic phenotype.[22] Therefore, simultaneous assessment of multiple markers might better characterize the biological phenotype of a particular cancer. Additionally, the use of a single gene locus has several drawbacks.[23] First, the maximum sensitivity can only be as high as the frequency of hypermethylation at a specific CpG locus. Second, noncancerous tissues can, in some cases, harbor CpG island hypermethylation at the same gene locus. Third, methylation of a single gene locus may occur in several cancers and, thus, be misleading.[24] Furthermore, the number of hypermethylated genes increases as cancers progress and many methylation markers influence others[12, 13]; thus, cancer detection and prognosis are likely to be best estimated using a combination of markers.

The quantitative integration of methylation levels using a small set of hypermethylated genes is one of the major advantages of PSQ. Quantitative merging allows the quantitative comparison of samples and the accurate segregation of pathologic and prognostic covariates based on methylation levels. Merging the methylation levels of carefully selected genes yields maximum specificity and sensitivity. In this study, such quantitative combination enabled us to predict recurrence in a manner similar to that afforded by single markers; however, the power of multiple markers for predicting progression is much greater than that provided by single markers. These findings imply that the appropriate combination of selected genes will increase the usefulness of methylation-based biomarkers for both detection and prognosis.

A wide array of clinicopathological risk factors need to be considered when making a prognosis for NMIBC, including prior recurrence rates, tumor size, multiplicity, T-category, grade and treatment with intravesical therapy.[1, 2] In NMIBC, tumors with similar morphology may behave differently; thus, it is a major clinical challenge to accurately differentiate those patients whose tumor is likely to recur and/or progress after initial treatment from those in which recurrence and/or progression are unlikely. If candidate prognostic markers are to have considerable clinical relevance, they must provide a predictive capability beyond that offered by conventional clinicopathological parameters. Importantly, the methylation-based biomarkers identified in this study are independent predictors of prognosis, regardless of the clinicopathological characteristics of the NMIBC. However, considering modest value of area under curve and HR for the prediction of recurrence, although our results are promising, further validation study is necessary to reduce false predictive rates.

To the best of our knowledge, the current study is the first to identify HOXA9, ILS1 and ALDH1A3 as methylation-based prognostic markers of NMIBC. Although the relationships between these candidate markers and other types of cancer have been studied previously,[25-31] only one study looked at bladder cancer.[13] HOXA9 and ILS1 are homeobox domain genes. HOXA9 plays an important role during development and hematopoiesis, and abnormalities are associated with leukemia, lung cancer, ovarian cancer, and bladder cancer.[13, 27-29] ILS1 affects various aspects of motor neuron identity, the lineage of cells of pancreatic endocrine origin, and cardiovascular progenitors.[32-34] ALDH1A3 is significantly associated with aldefluor positivity, the specific expression of which is a marker for breast cancer stem cells.[30, 31] A precise causal relationship between these putative genes (HOXA9, ILS1 and ALDH1A3) and NMIBC has not been yet determined. Additionally, even though these genes showed functional significance in bladder cancer, it does not always mean the crucial role in bladder tumor initiation or progression. Without clear demonstration of these kinds of association between these candidate markers and bladder cancer may be possible limitations of our study, but these will be elucidated with further study. Therefore, not only large-scale validation studies of human samples but also functional analysis and gene ontologic approach of these genes will give us more knowledge about their biological mechanisms and clinical relevance.

Because DNA methylation is reversible, it is thought to be a good therapeutic target. Drugs that target epigenetic alterations, such as DNA methylation inhibitors, restore the activity of genes by targeting aberrant heterochromatic regions, ultimately leading to the reactivation of tumor suppressor genes and/or other genes that are crucial for normal cellular function.[8] The hypermethylation markers identified in this study showed significantly reduced expression levels in NMINC compared with NC, and methylation status was associated with prognosis. The identification of these methylation markers is noteworthy because of their potential utility for both detection and prognosis; they may also serve as candidate therapeutic targets.

From a clinical point of view, the most promising applications for epigenetic markers are clearly early detection, prediction of response to treatment and indication of disease prognosis. The results presented herein are promising because the candidate methylation markers were selected from a genome-wide analysis and validated in a relatively large number of human tissue samples obtained from long-term follow-up patients. In addition, the selected methylation markers are independent predictors of disease outcome. An accurate prediction of prognosis made using these candidate methylation markers would aid clinicians in terms of patient counseling, determining the frequency and extent of monitoring, and whether more aggressive therapy is needed. Further research will increase the accuracy of NMIBC detection and enable more realistic predictions regarding outcome. It may also lead to new therapies that target specific molecular defects, thereby significantly lowering the morbidity associated with NMIBC.

In conclusion, our findings suggest that the novel methylation markers HOXA9, ISL1 and ALDH1A3 are independent indicators of prognosis in NMIBC patients. These prognostic markers may constitute a promising tool for assessing the recurrence and progression of NMIBC patients and may facilitate the design of individualized therapeutic modalities.

Acknowledgement

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

All samples derived from the National Biobank of Korea were obtained with informed consent under institutional review board-approved protocols.

References

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgement
  7. References
  8. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
ijc28121-sup-0001-suppinfo01.doc207KSupporting Information Fig S1 (A) Bioinformatic functional classification analyses of 6 candidate genes in non-muscle invasive bladder cancer. Classification enrichment was determined using Ingenuity Pathway Analysis software. The threshold of significance was -log (P = 0.05). (B) Gene networks analysis with 6 candidate genes in non-muscle invasive bladder cancer.
ijc28121-sup-0002-suppinfo02.doc33KSupporting Information Table 1. Primers used for pyrosequencing analysis
ijc28121-sup-0003-suppinfo03.doc39KSupporting Information Table 2. Comparison of β-value differences between tumors and normal controls with Infinium DNA methylation array data
ijc28121-sup-0004-suppinfo04.doc28KSupporting Information Table 3. Correlation between Infinium array and pyrosequencing values for methylation levels (n=24)
ijc28121-sup-0005-suppinfo05.doc49KSupporting Information Table 4. Association between methylation markers and clinicopathological characteristics (n = 163)
ijc28121-sup-0006-suppinfo06.doc36KSupporting Information Table 5. ROC curve analysis of DNA methylation markers to predict disease outcome (n = 163)

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.