Quantification of 5-Methylcytosine and 5-Hydroxymethylcytosine in Genomic DNA from Hepatocellular Carcinoma Tissues by Capillary Hydrophilic-Interaction Liquid Chromatography/Quadrupole TOF Mass Spectrometry

Abstract

BACKGROUND

5-Methylcytosine (5-mC) is an important epigenetic modification involved in development and is frequently altered in cancer. 5-mC can be enzymatically converted to 5-hydroxymethylcytosine (5-hmC). 5-hmC modifications are known to be prevalent in DNA of embryonic stem cells and neurons, but the distribution of 5-hmC in human liver tumor and matched control tissues has not been rigorously explored.

METHODS

We developed an online trapping/capillary hydrophilic-interaction liquid chromatography (cHILIC)/in-source fragmentation/tandem mass spectrometry system for quantifying 5-mC and 5-hmC in genomic DNA from hepatocellular carcinoma (HCC) tumor tissues and relevant tumor adjacent tissues. A polymer-based hydrophilic monolithic column was prepared and used for the separation of 12 nucleosides by cHILIC coupled with an online trapping system. Limits of detection and quantification, recovery, and imprecision of the method were determined.

RESULTS

Limits of detection for 5-mC and 5-hmC were 0.06 and 0.19 fmol, respectively. The imprecision and recovery of the method were determined, with the relative SDs and relative errors being <14.9% and 15.8%, respectively. HCC tumor tissues had a 4- to 5-fold lower 5-hmC content compared to tumor-adjacent tissues. In addition, 5-hmC content highly correlated with tumor stage (tumor-nodes-metastasis, P = 0.0002; Barcelona Clinic liver cancer, P = 0.0003).

CONCLUSIONS

The marked depletion of 5-hmC may have profound effects on epigenetic regulation in HCC and could be a potential biomarker for the early detection and prognosis of HCC.

Methylation of DNA at the C5 position of cytosine to give 5-methylcytosine (5-mC)⁴ is one of the best-characterized epigenetic modifications and has been implicated in numerous biological processes, including embryogenesis, X-chromosome inactivation, genetic imprinting, and cellular differentiation (1, 2). Properly established and maintained DNA methylation patterns are vital for mammalian development and normal functions of the adult organism (3). Aberrant promoter methylation leading to inappropriate transcriptional silencing or activation of genes is often found in various types of human cancers (4). Although DNA methylation has been viewed as a stable epigenetic mark, studies have revealed that this modification is not as static as once thought. In fact, active DNA demethylation has been observed and thoroughly studied in plants (5), but the mechanisms for active DNA demethylation in mammalian cells remain elusive (6).

The ten– eleven translocation (TET) proteins are capable of catalyzing the sequential oxidation of 5-mC to 5-hydroxymethylcytosine (5-hmC), 5-formylcytosine (5-foC), and finally 5-carboxylcytosine (5-caC) (7–10). The resulting 5-caC can be recognized and cleaved by thymine-DNA glycosylase, thereby restoring unmethylated cytosine via base-excision repair machinery (11). Thus, active DNA demethylation may be achieved through a multistep oxidation of 5-mC with the generation of various forms of intermediates.

5-hmC has long been noted in bacteriophage DNA, and its presence in mammalian cells was first discovered in embryonic stem cells and adult neural cells (7, 8). 5-hmC is also considered to play a crucial role in cellular differentiation and pluripotency of embryonic stem cells (10). However, the biological significance of 5-hmC in human cancers remains elusive. Mutations and decreased expression of TET genes display lower contents of 5-hmC in tumor tissues compared to healthy controls (12–14). In solid tumors, 5-hmC contents are reduced in the carcinomas of prostate, breast, liver, lung, pancreas, and colon, as revealed by immunohistochemistry (12–14), as well as in lung and brain tumors, asshownby liquid chromatography/tandem mass spectrometry (LC-MS/MS) (15). These findings suggest that decreased 5-hmC in genomic DNA might be associated with tumor development.

5-hmC content in mammalian cells can be as low as 0.009% of cytosine (molar ratio of 5-hmC/cytosine in 293T cells) (9); therefore, a highly sensitive detection method is required for the quantitative analysis of 5-hmC content in mammalian genomes. Methods for detecting 5-hmC in genomic DNA include radioactive labeling followed by thin-layer chromatography detection (8), immunohistochemistry (16), HPLC (17), LC-MS/MS (18), enzymatic glycosylation labeling (19), and single-molecule real-time sequencing (20). The thin-layer chromatography method involves labeling with radioactive isotope and the results are not comparable to those of other available methods. Immunohistochemical staining is tedious and, to some extent, less quantitative. HPLC analysis relies on chromatographic separation to avoid coelution with other components. The glycosylation method is based on enzymatic incorporation of modified glucose into genomic 5-hmC; however, a complete enzymatic reaction may not be achieved, and 5-mC cannot be measured simultaneously. The measurement of 5-hmC by single-molecule real-time sequencing is possible, but the technology still needs improvements. Reversed-phase liquid chromatography (RPLC) coupled with MS/MS has been used for the analysis of 5-hmC (9, 15, 18). However, an inherent weakness of RPLC is that the high aqueous content of mobile phase results in relatively low ionization efficiency during electrospray ionization (ESI), which diminishes the detection capability. Moreover, sodium adducts and in-source collision-activated dissociation (CAD) fragmentation often hamper the determination of target analytes (21).

Hydrophilic-interaction liquid chromatography (HILIC) has emerged as a technique complementary to RPLC (22) owing to its good resolution for polar compounds (23, 24). We previously fabricated a hydrophilic poly(NAHAM-co-PETA) monolith [poly(N-acryloyltris(hydroxymethyl)aminomethane-co-pentaery-thritol triacrylate)] that had excellent column efficiency and separation resolution toward nucleosides (25). Furthermore, the employment of hydrophilic polymer–based monolith can enhance MS response of analytes owing to the high organic solvent–containing mobile phase.

Here we report a system for the simultaneous detection of 5-mC and 5-hmC in genomic DNA by using hydrophilic poly(NAHAM-co-PETA) monolith coupled with high-resolution quadrupole TOF mass spectrometry (qTOF-MS). Additionally, we used an online trapping system that improved the detection capability. We assessed 5-mC and 5-hmC contents in 143 hepatocellular carcinoma (HCC) tissues, which include 75 tumor tissues and 34 matched pairs of tumor and adjacent tissues.

Materials and Methods

REAGENTS

We purchased 2′-deoxycytidine (dC), 2′-deoxyguanosine (dG), 2′-deoxyadenosine (dA), thymidine (T), 2′-deoxyinosine (dI), cytidine (C), guanosine (G), adenosine (A), uridine (U), inosine (I), and 5-methyl-2′-deoxycytidine (5-mdC) from Sigma-Aldrich, and 5-hydroxymethyl-2′-deoxycytidine (5-hmdC) from Berry & Associates. We prepared nucleosides in ACN/H₂O (99/1, vol/vol) at desirable concentration for the construction of calibration curves and method validation. We purchased HPLC-grade isopropanol and acetonitrile (ACN) from Tedia and formic acid (FA) from Shanghai Chemical Reagent. Water used throughout all experiments was purified by use of a Milli-Q water purification apparatus (Millipore). All other reagents were obtained from various commercial sources and were of analytical grade unless otherwise indicated.

HEPATOCELLULAR CARCINOMA TISSUE SAMPLE COLLECTION

This study was approved by the ethics committee of Zhongnan Hospital of Wuhan University. A total of 109 HCC patients [93 males and 16 females, mean (SD) age 48.5 (12.0) years, range 18–80 years] were enrolled from June 2005 to April 2011 at Zhongnan Hospital of Wuhan University with TNM (tumor-nodes-metastasis) stage I (n = 73), stage II (n = 8), stage III (n = 13), and stage IV (n = 15) cancer. Among them, 20 patients did not have hepatitis B virus (HBV) or HCV infection and 89 patients had only HBV infection, including mild hepatitis (n = 23), moderate hepatitis (n = 19), severe hepatitis (n = 23), and liver failure (n = 24); 71 patients did not have cirrhosis, 32 patients had compensated cirrhosis and 6 patients had decompensated cirrhosis; 31 patients were drinkers. All patient diagnoses were confirmed by pathology, and patients underwent liver resection. We used a total of 143 formalin-fixed, paraffin-embedded (FFPE) tissue samples, which included 34 pairs of tumor and matched tumor-adjacent tissues as well as 75 tumor tissues for which matched adjacent tissues were not available (see Supplemental Table 1, which accompanies the online version of this article at http://www.clinchem.org/content/vol59/issue5).

CAPILLARY HILIC-ESI-qTOF-MS SYSTEM

The capillary HILIC (cHILIC) was performed on a Shimadzu Prominence nano-flow liquid chromatography system (Shimadzu) with two LC-20AD nano pumps, two vacuum degassers, a LC-20AB HPLC pump, a SIL-20AC HT autosampler, and a nano-flow control valve (Fig. 1).

Fig. 1. Experimental setup for the analysis of nucleosides (5-hmdC, 5-mdC, dC, dG, dA, dI, T, C, G, A, U, and I) by cHILIC-ESI-Q-TOF-MS.

We used an orthogonal-acceleration TOF mass spectrometer (micrOTOF-Q; Bruker Daltonics) for the cHILIC-MS experiment. The instrument was controlled by Bruker Daltonics Microcontrol software, and Bruker Daltonics Data Analysis 3.4 software was used for data analysis. Spectra were collected with a time resolution of 1 s in the m/z range of 50–600. The hydrophilic poly(MAA-co-EDMA) monolith [poly-(methacrylic acid-co-ethylene glycol dimethacrylate)] (1 cm, 50 μm inner diameter, 360 μm outer diameter) was purchased from Weltech and used as online trapping columns. The poly(NAHAM-co-PETA) monolithic column (50 cm, 100 μm inner diameter, 360 μm outer diameter) was prepared as previously described (25) and used for the separation. The targeted compounds were separated on the poly(NAHAM-co-PETA) monolithic column, which was connected to a PicoTip™ (New Objective) nano-spray tip (360 μm outer diameter, 10 μm inner diameter) with a zero-dead-volume union (Upchurch Scientific) to minimize postcolumn dead volume.

STATISTICAL ANALYSES

We performed all statistical analyses using SPSS 19.0 software (SPSS Inc.). All P values were two-sided, and P values of <0.05 were considered to be statistically significant. We estimated Pearson correlation coefficients for each pair of covariate study and performed ROC analysis to evaluate the ability of 5-mdC and 5-hmdC to discriminate tumor tissues from tumor-adjacent tissues.

Results

ESI-qTOF-MS DETECTION

The full-scan positive-ion ESI-MS of 5-mdC (see online Supplemental Fig. 1A) revealed the formation of [M + H]⁺, [M + Na]⁺, and [M + K]⁺ ions of the analyte at m/z 242.1193, 264.0910, and 280.0710, respectively. The in-source CAD fragment ion at m/z 126.0671 was also observed. The corresponding ions were found for 5-hmdC at m/z 258.1093, 280.0912, 296.0651, and 142.0614, respectively (see online Supplemental Fig. 1B). Previous reports indicated that protonated 5-mdC tends to lose its β-D-2-deoxyribofuranose moiety to give protonated 5-mC (21, 26, 27). Our results showed that, under in-source CAD conditions, protonated 5-mdC and 5-hmdC can also lose the β-D-2-deoxyribofuranose moiety to yield protonated 5-mC and 5-hmC at m/z 126.0671 and 142.0614, respectively (see online Supplemental Fig. 1, A and B). The abundance ratio for the protonated ion (I) of 5-mC (I₁₂₆) over that of 5-mdC (I₂₄₂) was approximately 1/2.5, and the corresponding ratio for 5-hmC at m/z 142.0614 (I₁₄₂) over that of 5-hmdC at m/z 258.1093 (I₂₅₈) was approximately 5/4. These observations demonstrate that the in-source CAD can result in a decrease in the detection of 5-mdC and 5-hmdC. To circumvent this problem, we optimized the in-source ESI-MS/MS conditions to stimulate the in-source CAD occurrence of 5-mdC and 5-hmdC. Under optimized in-source ESI-MS/MS conditions, the I₁₂₆ vs the I₂₄₂ was approximately 10/1 for 5-mdC (see online Supplemental Fig. 1C), and the I₁₄₂ vs the I₂₅₈ was approximately 12/1 for 5-hmdC (see online Supplemental Fig. 1D). Neither the [M + Na]⁺ nor the [M + K]⁺ ion was observed for 5-mC and 5-hmC. Therefore, with the optimized in-source ESI-MS/MS conditions, we used the product ions of 5-mdC (i.e., 5-mC, m/z 126.0671) and 5-hmdC (i.e., 5-hmC, m/z 142.0614) for the identification and quantification of cytosine methylation and hydroxymethylation, respectively. With this strategy, the detection capability for 5-mdC and 5-hmdC was >1 order of magnitude higher than before optimization. The detailed optimized conditions of ESI-MS/MS are shown in the online Supplement.

ONLINE TRAPPING/cHILIC SYSTEM

We first optimized the separation conditions for the above 12 nucleosides by changing the contents of ACN (online Supplemental Fig. 2), FA (online Supplemental Fig. 3), and isopropanol (online Supplemental Fig. 4) in mobile phase. With the optimized mobile phase of ACN/H₂O/isopropanol/FA (90/5/5/0.02, vol/vol/vol/vol), the 12 nucleosides could be baseline resolved within 30 min (online Supplemental Fig. 5).

Next, we investigated the influence of loading flow rate (online Supplemental Fig. 6A), eluent volume (see online Supplemental Fig. 6B), and washing volume (see online Supplemental Fig. 6C) on the signal-to-noise (S/N) ratio of 5-mdC and 5-hmdC. Additionally, we assessed the capacity of the online trapping poly(MAA-co-EDMA) monolith in capturing 5-mdC and 5-hmdC (see online Supplemental Fig. 6D). The optimized conditions consisted of a loading flow rate of 10 μL/min, an eluent volume of 2250 nL, and a washing volume of 400 nL. In combination with the large injection volume (5 μL) to nanoscale separation system and the sample zone compression on the online trapping column, the detection capability for 5-mdC and 5-hmdC by cHILIC-ESI-qTOF-MS/MS was substantially improved without any apparent loss of separation resolution (Fig. 2A).

Fig. 2. Extracted-ion chromatograms of nucleosides.

(A), Nucleoside standards obtained under the optimized conditions. (B), Nucleosides from 2 ng genomic DNA of HCC tissues. Shown in the inset is the expanded chromatogram to reveal better the separation of 5-mdC, dC, C, dG, and 5-hmdC.

METHOD DEVELOPMENT

For the analysis of 5-mC and 5-hmC, extracted ion chromatograms were obtained with 0.01-Da mass width. We investigated the linearity of the method with 1.2 pmol dC standard supplemented with 5-mdC and 5-hmdC at different amounts ranging from 0.6 to 120 fmol (Table 1). With in-source ESI-MS/MS, we used the MS peaks of the 5-mdC and 5-hmdC product ions, 5-mC and 5-hmC, for the identification and quantification of cytosine methylation and hydroxymethylation, respectively. We constructed the calibration curves by plotting the mean peak area ratio of 5-mdC/dC or 5-hmdC/dC vs the mean molar ratio of 5-mdC/dC or 5-hmdC/dC on the basis of data obtained from triplicate measurements. The results showed linearity within the range of 0.05%–10% (molar ratio of 5-mdC/dC or 5-hmdC/dC) with a coefficient value (R²) >0.9979 (Table 1). Limits of detection and quantification (LODs and LOQs) for 5-mdC and 5-hmdC were calculated as the amounts of the analytes at S/N ratios of 3 and 10, respectively. The LODs and LOQs were 0.06 and 0.20 fmol, respectively, for 5-mdC and 0.19 and 0.64 fmol for 5-hmdC (Table 1). The LODs for the 5-mdC and 5-hmdC obtained in this study were, to the best of our knowledge, the lowest compared to other previously reported methods with mass spectrometry (9, 21).

Table 1.

Linearities, LOQs, and LODs for 5-mdC and 5-hmdC obtained by cHILIC-ESI-qTOF-MS/MS.

Analyte	Linear range (vs [dC]), %	Regression line		R2	LOD, fmol	LOQ, fmol
		Slope	Intercept
5-mdC	0.05–10	0.0144 (0.0007)	0.0030 (0.0002)	0.9979	0.06	0.20
5-hmdC	0.05–10	0.0280 (0.0011)	0.0003 (0.0001)	0.9985	0.19	0.64

We validated the method with the synthesized 5-mC- or 5-hmC-containing oligodeoxynucleotide by comparing the measured 5-mdC or 5-hmdC content to the theoretical 5-mdC or 5-hmdC content (online Supplemental Table 2). 5-mdC and 5-hmdC were determined from DNA hydrolysis product with CVs being 2.5%–11.0% and relative errors (REs) being −16.4%–13.0% (Tables 2 and 3), indicating that the cHILIC-ESI-qTOF-MS method was reliable for the simultaneous determination of 5-mdC and 5-hmdC. We evaluated the ion suppression by comparing the MS intensities of 5-mdC and 5-hmdC in ACN/H₂O and in DNA hydrolysis products from synthesized DNA (online Supplemental Fig. 7). The peak areas of 5-mdC and 5-hmdC in ACN/H₂O were 1.8% (0.2%) and 1.9% (0.1%) greater than the peak areas in DNA hydrolysis products from the synthesized DNA, suggesting that ion suppression was negligible. The weak ion suppression for 5-mdC and 5-hmdC may be attributed to the relatively clean DNA hydrolysis product as well as the good chromatographic resolution of the analytes on analytical monolithic column before mass spectrometry analysis. In addition, we evaluated the imprecision and recovery of the cHILIC-ESI-qTOF-MS/MS method (Tables 2 and 3). The relative SDs (CVs) and REs were <14.9% and 15.8%, respectively.

Table 2.

Imprecision and recovery of the method for the detection of 5-mdC (n = 3 for each day).

Nominal [5-mdC]/[dC], %	0.05	0.10	0.20	0.40	0.80	1.00	2.00	4.00	6.00	8.00	10.00	Low QC (1.00)	High QC (5.00)
Day 1
Measured mean [5-mdC]/[dC], %	0.05	0.10	0.18	0.35	0.78	0.90	2.22	3.87	5.57	7.62	11.01	0.94	4.63
RSD, %a	9.8	3.6	4.7	5.3	3.3	4.6	5.2	4.7	6.3	5.4	9.2	3.4	4.9
RE, %	−3.8	−4.8	−7.9	−12.2	−2.9	−8.1	10.6	−2.2	−6.5	−4.7	9.5	−6.0	−7.4
Day 2
Measured mean [5-mdC]/[dC], %	0.05	0.10	0.21	0.38	0.84	1.11	2.18	4.13	5.59	8.22	9.63	0.96	4.18
RSD, %	11.0	5.2	5.9	5.8	5.6	6.1	9.0	7.0	7.3	9.5	12.0	4.3	5.7
RE, %	−1.7	4.2	5.3	−4.9	4.3	10.5	9.0	3.3	−5.9	2.6	−4.2	−4.0	−16.4
Day 3
Measured mean [5-mdC]/[dC], %	0.05	0.11	0.20	0.43	0.84	0.90	2.22	4.38	5.87	8.41	9.23	1.07	5.09
RSD, %	8.2	6.6	8.9	8.2	6.3	7.3	10.6	7.7	8.3	10.6	12.1	6	7.7
RE, %	2.1	5.2	−2.7	6.4	5.1	−4.2	−1.8	11.2	−2.0	5.4	−7.6	7.0	1.8
Day 4
Measured mean [5-mdC]/[dC], %	0.06	0.11	0.21	0.45	0.79	0.91	2.12	3.48	6.61	7.33	11.07	1.09	5.01
RSD, %	10.1	7.2	8.1	6.9	4.9	5.2	7.5	8.7	9.9	13.1	14.9	4.9	4.1
RE, %	14.2	11.1	6.1	12.9	14.0	−11.0	5.3	−11.4	10.8	−8.9	10.5	9.0	0.2
Day 5
Measured mean [5-mdC]/[dC], %	0.06	0.11	0.22	0.43	0.89	1.19	2.02	4.17	5.63	7.19	9.14	1.13	4.95
RSD, %	9.9	6.4	11.8	7.6	5.1	5.7	9.2	11.6	10.8	12.1	13.2	7.3	6.2
RE, %	14.4	13.6	9.4	6.3	10.8	−0.9	−2.4	5.9	−7.3	−9.9	−9.4	13.0	−1.0

^aRSD, relative SD.

Table 3.

Imprecision and recovery of the method for the detection of 5-hmdC (n = 3 for each day).

Nominal [5-hmdC]/[dC], %	0.05	0.10	0.20	0.40	0.80	1.00	2.00	4.00	6.00	8.00	10.00	Low QC (0.50)	High QC (2.00)
Day 1
Measured mean [5-hmdC]/[dC], %	0.05	0.10	0.23	0.40	0.81	1.06	2.25	4.09	6.64	7.91	10.92	0.44	1.99
RSD, %a	5.6	4.2	9.9	7.6	6.8	5.9	3.5	2.6	3.6	4.9	5.9	3.3	4.0
RE, %	4.0	2.0	15.8	0.8	0.8	5.6	12.6	2.5	9.7	−1.1	9.0	−11.4	−0.5
Day 2
Measured mean [5-hmdC]/[dC], %	0.05	0.10	0.20	0.41	0.76	0.98	1.98	3.82	5.77	8.18	9.47	0.52	1.82
RSD, %	11.3	10.7	10.0	6.8	8.0	6.2	11.6	10.5	5.9	8.2	7.2	5.1	2.5
RE, %	−2.0	−3.0	−0.5	2.0	−4.7	−1.7	0.8	−5.0	−3.3	2.5	−5.0	3.2	−8.6
Day 3
Measured mean [5-hmdC]/[dC], %	0.05	0.11	0.20	0.40	0.81	1.04	2.00	4.11	6.24	8.09	11.06	0.47	2.18
RSD, %	6.6	4.6	7.5	8.6	10.2	7.2	10.5	8.3	4.2	7.6	5.2	6.4	7.8
RE, %	6.0	9.0	1.5	0.8	1.3	3.6	1.5	1.3	3.3	1.3	10.6	−6.6	9.4
Day 4
Measured mean [5-hmdC]/[dC], %	0.05	0.09	0.18	0.38	0.81	1.08	2.12	3.80	5.57	7.81	9.79	0.53	2.10
RSD, %	5.6	6.5	6.2	7.2	5.2	6.3	3.2	7.6	3.2	4.6	6.2	7.0	8.4
RE, %	−6.0	−7.0	−12.0	−4.6	1.4	8.4	5.0	−5.0	−6.7	−2.5	−2.0	6.2	5.2
Day 5
Measured mean [5-hmdC]/[dC], %	0.05	0.10	0.19	0.41	0.77	0.96	1.90	4.24	6.11	7.87	10.41	0.49	1.79
RSD, %	6.2	5.2	4.2	6.2	8.7	3.9	6.9	3.9	8.0	7.1	10.9	9.3	11.0
RE, %	4.0	−2.0	−5.5	1.4	−4.4	−3.8	−5.4	5.3	2.0	−1.3	4.0	−1.8	−10.4

^aRSD, relative standard deviation.

MEASUREMENT OF 5-mC AND 5-hmC IN GENOMIC DNA FROM HEPATOCELLULAR CARCINOMA TISSUES

With the developed cHILIC-ESI-qTOF-MS method, we further investigated the minimal sample required for the quantification of 5-mC and 5-hmC. We found that 5-mC could be quantified from 1 ng genomic DNA, whereas 5-hmC could be quantified from 2 ng genomic DNA. Figure 2B displays the extracted-ion chromatogram of 12 nucleosides from the hydrolysis product of 2 ng genomic DNA from HCC tissue (H009 tumor tissue) (see online Supplemental Table 1). The detection of U and G was less sensitive than that of other nucleosides, which may be attributed to the weaker proton affinity of U and low elution efficiency of G from the trapping column. The chromatograms of 5-mdC and 5-hmdC were extracted at m/z 126.0671 (0.01) and 142.0619 (0.01), respectively. The resolution of R_5-mdC/dC and R_5-hmdC/dG were > 1.5; therefore, the presence of high contents of dC or dG does not interfere with the quantification of 5-mdC and 5-hmdC.

5-hmC CORRELATES WITH TUMOR STAGES

A total of 143 HCC tissues derived from 109 patients, including 75 tumor tissues and 34 pairs of matched tumor and tumor-adjacent tissues, were analyzed by cHILIC-ESI-qTOF-MS. The mean contents of 5-mC in genomic DNA of all tumor tissues and all tumor-adjacent tissues were 5.57% (0.83%) and 5.97% (0.84%), respectively (Fig. 3A). The mean contents of 5-mC in genomic DNA from matched-pair tumor tissues and tumor-adjacent tissues were 6.00% (0.67%) and 5.97% (0.84%), respectively (Fig. 3C). The results suggested there was no significant difference of 5-mC between tumor tissues and tumor-adjacent tissues (Fig. 3, A and C). However, the 5-hmC content was markedly lower in genomic DNA of tumor tissues than tumor-adjacent tissues. As shown in Fig. 3, B and D, the mean contents of 5-hmC were 1.72% (0.45%) and 0.37% (0.13%) in tumor-adjacent tissues and tumor tissues, respectively; the mean contents of 5-hmC in genomic DNA from matched-pair tumor-adjacent tissues and tumor tissues were 1.72% (0.45%) and 0.42% (0.19%), respectively.

Fig. 3. Quantification and statistical analysis of 5-mC and 5-hmC in human HCC tumor tissues and tumor-adjacent tissues.

(A), 5-mC content in HCC tumor tissues and tumor-adjacent tissues. (B), 5-hmC content in HCC tumor tissues and tumor-adjacent tissues. (C), 5-mC content in matched-pair HCC tumor tissues and tumor-adjacent tissues. (D), 5-hmC content in matched-pair HCC tumor tissues and tumor-adjacent tissues. (E), Correlation of 5-hmC content with human HCC tumor TNM stages. (F), Correlation of 5-hmC content with human HCC tumor BCLC stages. (G), ROC curve for 5-hmC score for human HCC tumor tissues. (H), ROC curve for 5-mC score for human HCC tumor tissues. AUC [mean (SD)] are shown.

We also compared 5-mC and 5-hmC contents measured by cHILIC-ESI-qTOF-MS/MS and HPLC-MS. The measured 5-mC and 5-hmC contents in tumor adjacent tissues were comparable with these two methods, with REs being −15.9% to 16.0% in all the samples analyzed (online Supplemental Tables 3 and 4), indicating that the cHILIC-ESI-qTOF-MS method is reliable for the determination of 5-mC and 5-hmC in genomic DNA. However, the HPLC-MS cannot detect 5-hmC in tumor tissues (<1.0% vs [dC], as determined by cHILIC-ESI-qTOF-MS/MS) because of its limited analytical sensitivity.

We performed statistical analysis to evaluate the correlation of 5-mC and 5-hmC between tumor tissues and tumor-adjacent tissues with respect to patients’ age, sex, tumor stages [TNM and Barcelona Clinic liver cancer (BCLC)], liver inflammation stages, liver cirrhosis stages, and liver function parameters (alanine transaminase, aspartate aminotransferase, total protein, albumin, globulin, γ-glutamyl transpeptidase, and alkaline phosphatase). On the basis of Pearson correlation coefficient, 5-hmC content was significantly correlated with tumor stages (TNM, r = −0.324, P = 0.0002; BCLC, r = −0.338, P = 0.0003) (online Supplemental Table 5). As shown in Fig. 3, E and F, decreased 5-hmC was associated with tumor progression. However, there was no correlation of 5-hmC with respect to age (P = 0.149), sex (P = 0.190), liver inflammation stages (P = 0.915), liver cirrhosis stages (P = 0.117), alcohol use (P = 0.068) (see online Supplemental Table 5), or liver function parameters (data not shown). Additionally, 5-mC was not associated with tumor stages (TNM, P = 0.765; BCLC, P = 0.681), age (P = 0.979), sex (P = 0.586), liver inflammation stages (P = 0.739), liver cirrhosis stages (P = 0.567), alcohol use (P = 0.056) (see online Supplemental Table 5), or liver function parameters (data not shown).

We further evaluated the possibility of 5-hmC as a biomarker for the early detection and prognosis of human HCC by performing ROC analysis. As shown in Fig. 3G, 5-hmC was highly effective in the detection of HCC, with the area under the curve (AUC) being 0.969; however, 5-mC was not appropriate for the detection of HCC, with AUC being 0.599 (Fig. 3H).

Discussion

HCC is one of the most common human cancers (28). Asians have a high risk of HCC development (29). The importance of epigenetic alterations in human HCC, however, has not been rigorously explored despite a few reports suggesting changes in global cytosine methylation and hydroxymethylation in cancer cells (12, 14, 16, 30, 31).

DNA methylation plays an important role in tumor pathogenesis, and promoter CpG island hypermethylation in tumor-suppressor genes is a common hallmark of human cancers (1, 4, 32). However, it is still unclear why certain regions become hypermethylated and others remain unmethylated. Hypomethylation at promoters can activate the aberrant expression of oncogenes (33). The discovery that 5-mC can be oxidized to 5-hmC by TET enzymes has raised many questions regarding the role of 5-hmC in epigenetic reprogramming. A role for 5-hmC as an intermediate in DNA demethylation has been postulated (7–11). A recent report has shown that the rapid loss of 5-mC from mouse paternal pronuclei was accompanied by an accumulation of genomewide 5-hmC (34). However, the failure to find many of the predicted intermediates of an active oxidative demethylation pathway of normal mouse tissues challenges the existence of such a mechanism (9, 35), which may be attributed to the lack of highly sensitive methods for the detection of intermediates.

We developed a method for simultaneous determination of 5-mC and 5-hmC by cHILIC-ESI-qTOF-MS/MS. The highly sensitive method allowed for the determination of low contents of 5-mC and 5-hmC with a DNA sample of only 2 ng. With this method, we provided evidence of lower content of 5-hmC in HCC by analyzing matched-pair tumor tissues and tumor-adjacent tissues. Because 5-mC is required as a substrate for oxidation to generate 5-hmC, the decrease in 5-hmC could emanate from reduced 5-mC in tumor tissues. To examine this possibility, we also analyzed the 5-mC contents in genomic DNA from tumor tissues and tumor-adjacent tissues. The genomewide 5-mC content was similar between tumor tissues and tumor-adjacent tissues (Fig. 3), revealing that the diminished contents of 5-hmC in HCC tissues is not due to decreased contents of global cytosine methylation.

Our correlation analysis also showed that 5-hmC correlated with tumor stage (Fig. 3, E and F), whereas no such association was found for 5-mC. In addition, ROC analysis suggested that HCC can be characterized by the change of 5-hmC but not 5-mC (Fig. 3, G and H). The discovery that 5-hmC contents were reduced in HCC, together with previous reports of the decreased contents of 5-hmC in other types of cancer tissues (15, 16), suggests that the depletion of 5-hmC could be a general feature of solid tumors. The biological significance of the loss of 5-hmC in tumors remains to be elucidated; nevertheless, loss of 5-hmC could be used as biomarker for the early detection and prognosis of HCC.