Prognostication of patients with clear cell renal cell carcinomas based on quantification of DNA methylation levels of CpG island methylator phenotype marker genes

As a learning cohort, 102 samples of cancerous tissue obtained from specimens surgically resected from 102 patients with primary ccRCCs were subjected to the present analysis. These patients did not receive preoperative treatment and underwent nephrectomy at the National Cancer Center Hospital, Tokyo, Japan. There were 71 men and 31 women with a mean (± standard deviation) age of 62.9 ± 10.4 years (range, 36 to 85 years). Histological diagnosis was made in accordance with the World Health Organization classification [18].

In our previous study, unsupervised hierarchical clustering based on genome-wide DNA methylation analysis using single CpG-resolution Infinium assay divided the 102 ccRCCs in the learning cohort into 88 CIMP-negative ccRCCs and 14 CIMP-positive ccRCCs [7]. In the same study, we showed that the CIMP-positive ccRCCs were clinicopathologically more aggressive and associated with a poorer patient outcome than CIMP-negative ccRCCs [7]: the clinicopathological characteristics [19, 20] of CIMP-negative and CIMP-positive ccRCCs in the learning cohort are summarized in Additional file 1: Table S1.

As a validation cohort, 100 samples of cancerous tissue were obtained from specimens surgically resected from 100 patients with primary ccRCCs. These patients also did not receive preoperative treatment and underwent nephrectomy at the National Cancer Center Hospital, Tokyo, Japan. The patients comprised 68 men and 32 women with a mean (± standard deviation) age of 62.5 ± 11.4 years (range, 33 to 87 years). The clinicopathological characteristics [19, 20] of ccRCCs in the validation cohort are summarized in Additional file 2: Table S2.

Tissue specimens were taken and frozen immediately after surgical removal and have been stored in liquid nitrogen until DNA extraction. ccRCCs are hypervascular tumors with an increased opportunity for infiltration of non-cancerous cells such as lymphocytes [21]: the microscopically examined tumor cell contents (%) of all ccRCC tissue specimens in the learning and validation cohorts are shown in Additional file 3: Table S3. Tissue specimens were provided by the National Cancer Center Biobank, Tokyo, Japan. This study was approved by the Ethics Committee of the National Cancer Center, Tokyo, Japan, and was performed in accordance with the Declaration of Helsinki. All the patients provided written informed consent prior to inclusion in the study.

High-molecular-weight DNA was extracted from fresh-frozen tissue samples using phenol-chloroform followed by dialysis [22]. One microgram of genomic DNA was subjected to bisulfite treatment using an EpiTect Bisulfite Kit (QIAGEN GmbH, Hilden, Germany), in accordance with the manufacturer’s protocol. This process converts non-methylated cytosine to uracil, while methylated cytosine remains unchanged [23].

DNA methylation levels at individual CpG sites were evaluated quantitatively using the MassARRAY platform (Sequenom, San Diego, CA). This method utilizes base-specific cleavage and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) [24]. Specific PCR primers for bisulfite-converted DNA were designed using the EpiDesigner software package (http://​www.​epidesigner.​com, Sequenom), encompassing all promoter CpG islands of the previously identified ccRCC-specific CIMP marker genes [7]. The sequences of the 16 primer sets are given in Additional file 4: Table S4. A T7-promoter tag (5′-CAGTAATACGACTCACTATAGGGAGAAGGCT-3′) was added to each reverse primer for in vitro transcription, and a 10-mer tag (5′-AGGAAGAGAG-3′) was added to each forward primer to balance the PCR.

To overcome PCR bias in DNA methylation analysis, we optimized the annealing temperature and type of DNA polymerase: 0%, 50% and 100% methylated control DNA (Epitect methylated human control DNA; QIAGEN) was used as template to test the linearity of the protocol. Using HotStar Taq DNA polymerase (QIAGEN) or TaKaRa Taq HS DNA polymerase (Takara Bio, Shiga, Japan), the annealing temperature for each of the 16 primer sets was set to give a correlation coefficient (R²) of more than 0.9 and to make the slope of the standard curve close to 1 (Additional file 5: Figure S1 and Additional file 4: Table S4). The PCR products were separated electrophoretically on 2% agarose gel and stained with ethidium bromide to confirm that specific products of the appropriate size and no non-specific products were obtained upon amplification.

Then, the PCR products were used as a template for in vitro transcription and the RNase A-mediated cleavage reaction using an EpiTYPER Reagent Kit (Sequenom). The fragmented samples were dispensed onto a SpectroCHIP array, and then detected on a MassARRAY analyzer compact MALDI-TOF MS instrument. The data were visualized using EpiTYPER Analyzer software v1.0 (Sequenom). The DNA methylation level (%) at each CpG site was determined by comparing the signal intensities of methylated and non-methylated templates. A cluster of consecutive CpG sites, each giving one measured value by the MassARRAY system, is defined as a “CpG unit” in the manufacturer’s protocol. The DNA methylation levels at the 299 examined CpG sites in the CIMP marker genes were then expressed as data for the 193 CpG units. Experiments were performed in triplicate for each sample-CpG unit, and the mean value for the three experiments was used as the DNA methylation level.

Differences in DNA methylation levels at individual CpG units between CIMP-positive ccRCCs and CIMP-negative ccRCCs were analyzed using Mann–Whitney U test. The CpG units having the largest diagnostic impact were identified by receiver operating characteristic (ROC) curve analysis [25]: For 23 CpG units showing area under the curve (AUC) values larger than 0.95, appropriate cutoff values were determined in order to discriminate CIMP-positive from CIMP-negative ccRCCs [26]. For discriminating CIMP-positive from CIMP-negative ccRCCs, the Youden index [26] was used as a cutoff value for each CpG unit. Survival curves for patients with ccRCCs were analyzed by the Kaplan-Meier method and the log-rank test. Correlations between DNA methylation levels and recurrence and disease-related death were analyzed using the Cox proportional hazards model. All statistical analyses were performed using SPSS statistics version 20 (IBM Corp., Armonk, NY). Differences at P values of less than 0.05 were considered statistically significant.

Previously, we had identified 17 ccRCC-specific CIMP marker genes based on genome-wide DNA methylation analysis using the Infinium HumanMethylation27K BeadChip [7]. Six exact Infinium probe CpG sites in ccRCC-specific CIMP marker genes (Probe ID: cg06274159 for the ZFP42 gene, cg03975694 for the ZNF540 gene, cg08668790 for the ZNF154 gene, cg01009664 for the TRH gene, cg22040627 for the SLC13A5 gene, and cg19246110 for the ZNF671 gene) were examined using the MassArray system in the learning cohort (Additional file 6; Figure S2). Significant correlations between DNA methylation levels determined by our previous Infinium assay [7] and those determined by the present MassArray analysis were statistically confirmed (P = 1.25 × 10⁻³⁵, P = 1.98 × 10⁻³², P = 1.31 × 10⁻⁴¹, P = 5.30 × 10⁻³⁴, P = 7.91 × 10⁻²² and P = 7.61 × 10⁻⁴⁴, respectively).

In the present study, our primary intention was to evaluate quantitatively the DNA methylation status of not only the Infinium probe sites but also the entire promoter CpG islands in the ccRCC-specific CIMP marker genes using the MassARRAY system [24]. Since the promoter regions of the CIMP marker genes, KCNQ1, FAM78A and NKX6-2, have a very high GC content, for these three genes we were unable to set optimized PCR conditions. Then, the DNA methylation status of 193 CpG units including 299 CpG sites in the remaining 14 ccRCC-specific CIMP marker genes, i.e. FAM150A, GRM6, ZNF540, ZFP42, ZNF154, RIMS4, PCDHAC1, KHDRBS2, ASCL2, PRAC, WNT3A, TRH, ZNF671 and SLC13A5, was evaluated quantitatively using the MassARRAY system. The average DNA methylation levels of 38 CpG units including 68 CpG sites located within the 1347 bp 5′-region of the representative CIMP marker gene, SLC13A5, in CIMP-negative (n = 88) and CIMP-positive (n = 14) ccRCCs in the learning cohort are shown in Figure 1A. Similarly, the average DNA methylation levels of 21 CpG units including 29 CpG sites located within the 428 bp 5′-region of another representative CIMP marker gene, ZNF671, in CIMP-negative and CIMP-positive ccRCCs in the learning cohort are shown in Figure 1B. The average DNA methylation levels of all the CpG units examined (59 in total) in the SLC13A5 and ZNF671 genes in the CIMP-positive ccRCCs were significantly higher than those in CIMP-negative ccRCCs (the P values for each CpG unit are shown in Additional file 7: Table S5). Similarly, the average DNA methylation levels of 130 CpG units including 195 CpG sites, out of the 134 CpG units examined including 202 CpG sites in the remaining 12 CIMP marker genes, in the CIMP-positive ccRCCs were significantly higher than those in CIMP-negative ccRCCs (Additional file 7: Table S5). These data indicated that almost the entire promoter CpG islands in all the CIMP marker genes examined were methylated in CIMP-positive ccRCCs.

Since quantitative DNA methylation analysis using the MassARRAY system revealed that many CpG sites showed significant differences in DNA methylation levels between CIMP-negative and CIMP-positive ccRCCs among all the promoter CpG islands of CIMP marker genes (Figure 1 and Additional file 7: Table S5), we attempted to identify CpG sites having the largest diagnostic impact, and to establish criteria for discriminating CIMP-positive from CIMP-negative ccRCCs. ROC curves were constructed for all 193 CpG units examined, including 299 CpG sites in the 14 CIMP marker genes examined, and the corresponding AUC values [25] were calculated. Eighty-six CpG units, including 135 CpG sites, showed AUC values larger than 0.9 (Additional file 8: Table S6). Among these 86, the top 23 CpG units including 32 CpG sites showing AUC values larger than 0.95 were used to establish the criteria for discriminating CIMP-positive from CIMP-negative ccRCCs (Table 1). For discriminating CIMP-positive from CIMP-negative ccRCCs, the Youden index [26] was used as a cutoff value for each CpG unit (Table 1).

Figure 2A shows scattergrams of the DNA methylation levels of representative CpG units in CIMP-negative and CIMP-positive ccRCCs in the learning cohort along with cutoff values listed in Table 1. The sensitivity and specificity of such discrimination using the cutoff values derived for each CpG unit are shown in Figure 2A and Table 1. A histogram showing the number of CpG units showing DNA methylation levels higher than the cutoff values listed in Table 1 in the learning cohort is shown in Figure 2B. All 14 ccRCCs showing DNA methylation levels higher than the cutoff values listed in Table 1 at 16 or more CpG units based on the present MassARRAY analysis (red bars in Figure 2B) were CIMP-positive ccRCCs identified by our previous hierarchical clustering based on the Infinium assay. All 88 ccRCCs showing DNA methylation levels higher than the cutoff values listed in Table 1 at less than 16 CpG units based on the present MassARRAY analysis (blue bars in Figure 2B) were CIMP-negative ccRCCs identified by our previous hierarchical clustering based on the Infinium assay.

Based on Figure 2B, we established the following criteria: when ccRCC tissue shows DNA methylation levels higher than the cutoff values listed in Table 1 at 16 or more CpG units (green line in Figure 2B), it is judged to be CIMP-positive. When ccRCC tissue shows DNA methylation levels higher than the cutoff values listed in Table 1 at less than 16 CpG units, it is judged to be CIMP-negative. Using these criteria, CIMP-positive ccRCCs in the learning cohort were discriminated from CIMP-negative ccRCCs with 100% sensitivity and specificity.

It has previously been revealed that patients with CIMP-positive ccRCCs show a poorer outcome [7]. Therefore, we attempted to validate the prognostic impact of CIMP diagnosis using criteria based on the cutoff values listed in Table 1. Using the additional 100 ccRCCs in the validation cohort, DNA methylation levels at the 23 CpG units including the 32 CpG sites in Table 1 were evaluated quantitatively using the MassARRAY system. The DNA methylation statuses of the 100 ccRCCs in the validation cohort were used to construct a histogram showing the number of CpG units with DNA methylation levels higher than the cutoff values listed in Table 1 (Figure 2C). The distribution of DNA methylation status at the 23 CpG units of the ccRCCs in the validation cohort (Figure 2C) was similar to that in the learning cohort (Figure 2B). Based on the criteria for CIMP diagnosis established using the learning cohort, 5 ccRCCs showing DNA methylation levels higher than the cutoff values listed in Table 1 at 16 or more CpG units were diagnosed as CIMP-positive, whereas 95 ccRCCs showing such higher DNA methylation levels at less than 16 CpG units were diagnosed as CIMP-negative.

Survival curves of the 100 patients belonging to the validation cohort were calculated by the Kaplan-Meier method (Figure 3). The period covered ranged from 27 to 5,031 days (mean: 1,860 days). Cancer-free (Figure 3A) and overall (Figure 3B) survival rates of patients with CIMP-positive ccRCCs diagnosed using the criteria based on the cutoff values listed in Table 1 were significantly lower than those of patients with CIMP-negative ccRCCs (P = 1.41 × 10⁻⁵ and 2.43 × 10⁻¹³, respectively, log-rank test).

Among 5 ccRCCs diagnosed as CIMP-positive in the validation cohort, one tumor was grade 2, two were grade 3, and two were grade 4 (Figure 2C); four were stage III and one was stage IV. Even after adjusting the grades, the cancer-free (P = 1.01 × 10⁻³) and overall (P = 7.04 × 10⁻⁴) survival rates of patients with CIMP-positive high-grade (grades 3 and 4) ccRCCs were significantly lower than those of patients with CIMP-negative high-grade (grades 3 and 4) ccRCCs (log-rank test, Additional file 9: Figure S3). The cancer-free (P = 7.76 × 10⁻⁴) and overall (P = 5.48 × 10⁻⁵) survival rates of patients with CIMP-positive high-stage (stages III and IV) ccRCCs were significantly lower than those of patients with CIMP-negative high-stage (stages III and IV) ccRCCs in the validation cohort (log-rank test, Additional file 9: Figure S3).

When compared with CIMP-negative ccRCCs, the CIMP-positive ccRCCs in the validation cohort had a significantly higher likelihood of recurrence (hazard ratio, 10.6; 95 percent confidence interval, 2.81 to 40.2; P = 5.03 × 10⁻⁴), and of disease-related death (hazard ratio, 75.8; 95 percent confidence interval, 7.81 to 735; P = 1.89 × 10⁻⁴) (Cox proportional hazards model). These data indicated that the validation cohort clearly demonstrated the prognostic significance of the criteria for CIMP diagnosis established in the present study.