Genome-scale detection of hypermethylated CpG islands in circulating cell-free DNA of hepatocellular carcinoma patients

Abstract

Despite advances in DNA methylome analyses of cells and tissues, current techniques for genome-scale profiling of DNA methylation in circulating cell-free DNA (ccfDNA) remain limited. Here we describe a methylated CpG tandems amplification and sequencing (MCTA-Seq) method that can detect thousands of hypermethylated CpG islands simultaneously in ccfDNA. This highly sensitive technique can work with genomic DNA as little as 7.5 pg, which is equivalent to 2.5 copies of the haploid genome. We have analyzed a cohort of tissue and plasma samples (n = 151) of hepatocellular carcinoma (HCC) patients and control subjects, identifying dozens of high-performance markers in blood for detecting small HCC (≤ 3 cm). Among these markers, 4 (RGS10, ST8SIA6, RUNX2 and VIM) are mostly specific for cancer detection, while the other 15, classified as a novel set, are already hypermethylated in the normal liver tissues. Two corresponding classifiers have been established, combination of which achieves a sensitivity of 94% with a specificity of 89% for the plasma samples from HCC patients (n = 36) and control subjects including cirrhosis patients (n = 17) and normal individuals (n = 38). Notably, all 15 alpha-fetoprotein-negative HCC patients were successfully identified. Comparison between matched plasma and tissue samples indicates that both the cancer and noncancerous tissues contribute to elevation of the methylation markers in plasma. MCTA-Seq will facilitate the development of ccfDNA methylation biomarkers and contribute to the improvement of cancer detection in a clinical setting.

Figures

Figure 1

Schematic of MCTA-Seq. First, the CpG tandems in the unmethylated CGIs are converted to UpG tandems via bisulfite treatment, whereas those in the methylated CGIs are unaffected. Second, the MCTA-Seq primer A binds semi-randomly to the converted DNA at the CpG site and is extended by a polymerase with displacement activity. The MCTA-Seq primer A consists of the following three parts: a semi-random sequence (RS) containing one CpG site (see Materials and Methods) is at the 3′-end, a unique molecular identifier (UMI) sequence is in the middle, and an anchor sequence is at the 5′-end. The methylated CGIs are expected to be amplified to a higher degree since they have high density of methylated CpG sites. Third, the MCTA-Seq primer B, which contains the CpG tandem sequence “CGCGCGG” at the 3′-end followed by a 4-bp “DDDD” sequence (D represents A or T or G), selectively amplifies the methylated CpG tandem sites. Last, exponential PCR amplification is performed using primer C and primer D, which is indexed, against the anchor sequences.

Figure 2

Validation of the MCTA-Seq technique. (A) The detection values are shown for CGIs with different number of CGCGCGG sequence. The sequence logos of the targeted genomic sites are shown (black box indicates the CGCGCGG sequence). (B) The throughput and reproducibility of MCTA-Seq. A total of 8 748 CGIs were detected in FMGs (average MePM > 8). The Pearson correlation coefficient (r) is shown. (C) Heat map showing the sensitivity of MCTA-Seq. The 8 748 CGIs are divided into seven groups ranked by their analytic sensitivities. For each group (two technical replicates), the heat map is rank-ordered by the average methylation values of the CGIs in WBCs from the lowest to the highest. The amounts of FMG and WBC of each dilution experiment are shown along with the percentage [M(%)] or the haploid genomic equivalent (GE) of FMG. Representative CGIs that are frequently hypermethylated in human cancers are shown. In the heat map, blue color indicates low, white and yellow intermediate and red high DNA methylation values [Log2(MePM)]. Asterisks indicate that the mass unit is picogram (pg). (D) Genomic view of the promoter CGI of the VIM gene. All aligned reads in technical replicates of four dilution experiments are shown by the Integrative Genomics Viewer in the bisulfite mode. The green box indicates the CGI. The red and green triangles indicate two CGCGCGG sequences positioned at the forward and the reverse strand, respectively. Comparison between the UMI counts and the read counts is shown. The arrow indicates the shortest amplicon (30 bp).

Figure 3

Analysis of HCC and noncancerous liver tissues. (A) Principal component analysis of 27 pairs of HCCs (T(HCC), red) and adjacent tissues (T(Aj), green) and 3 normal liver tissues (T(NL), blue) distinguishes most (23 of 27) HCCs from the noncancerous tissues. PC, principal component. (B) Volcano plot for differentially methylated CGIs between the HCCs and the adjacent noncancerous liver tissues. The x axis shows the fold changes of the average methylation value between the HCCs and the adjacent tissues, and the y axis shows the q-value as the FDR analogue of the P value (−Log10(q-value)) for a two-tailed MWW test of differences between two groups. The dashed line indicates statistical significance (FDR < 0.05). A total of 866 CGIs (tdmCGIs) were differentially hypermethylated in HCC tissues.

Figure 4

Identification of novel HCC-detecting markers in blood. (A) Volcano plot showing CGIs in plasma that are differentially methylated between the HCC patients with small tumors (n = 9, tumor size ≤ 3 cm) and cancer-free individuals (n = 45). The x axis shows the fold changes of the average methylation between the HCC patients and the cancer-free individuals, and the y axis shows the q-value as the FDR analogue of the P value (−Log10(q-value)) for a two-tailed MWW test of differences between two groups. The dashed line indicates statistical significance (FDR < 0.05). A total of 382 differentially methylated plasma CGIs (pdmCGIs) are identified (pink). (B) Venn plot view of the HCC-detecting markers (n = 41) and tdmCGIs (n = 866). (A, B) The type I and type II high-performance HCC-detecting markers are indicated in red and purple, respectively. (C) A heat map showing methylation of the HCC-detecting markers in the tissue samples. Each column represents a tissue sample of HCCs (T(HCC), red), adjacent livers (T(Aj), green), normal livers (T(NL), blue), WBCs (two biological replicates, yellow) or 10% FMG diluted in WBCs (brown), and each row represents a marker. The type I markers (n = 20) are ranked by their methylation levels in the liver calculated by their MePM values in the normal liver relative to the 10% FMG diluted in WBCs from the lowest to the highest. The type II markers (n = 21) are clustered using the hierarchical clustering. In the heat map, blue color indicates low, white and yellow intermediate and red high DNA methylation values, shown by log2(MePM). The markers for the classifiers are boxed. (D) Boxplots of the representative markers of the type I (ST8SIA6) and the type II [SHANK2(i)] HCC-detecting markers showing their methylation in plasma (red) or tissues (blue) samples. **P < 0.01; ND, no statistical difference. Two-tailed MWW test. (E) Scatter plots of the methylation levels of WBCs vs the normal liver. The RRBS results of 3 886 CGI loci are shown. Red circles indicate the type I markers (n = 19); purple circles indicate the type II markers (n = 20), and solid circles indicate the markers for the classifiers.

Figure 5

Comparison between matched tissue and plasma samples. The correlations (Spearman's rho) between the matched plasma and HCC tissue (red), the plasma and the adjacent liver tissue (green), as well as the HCC and the adjacent liver tissues (blue) from nine HCC patients as ordered by the tumor size. **P < 0.01; *P < 0.05.

Figure 6

Development of HCC classifiers. (A, B) For classifier I, the classifier scores of HCC patients, cirrhosis patients and normal individuals (A) and AUC curves (B) are shown. The dash line in A indicates the cutoff at 0.5. (C, D) For classifier II, the number of hypermethylated markers (C) and AUC curve (B) are shown. The dash line in C indicates the cutoff at 5.5. The green arrow in D indicates the point nearest the upper left corner of the ROC curve that gives the cutoff. Black: the training group; red: the testing group.

Figure 7

Performance comparison of ccfDNA methylation classifiers, AFP and ALT. For plasma samples of the HCC patients, cirrhosis patients and normal individuals in the training group (n = 72) and the testing group (n = 19) and post-surgery HCC patients (n = 3), detection results of two ccfDNA methylation classifiers, AFP, gamma-glutamyl transferase (GGT) and ALT are shown as positive (black bars, AFP: > 20 ng/ml, GGT: > 60 U/l, ALT: > 50 U/l) versus negative (blue bars). The MCTA-Seq data of the type I and type II markers in the classifiers are shown in a heat map using a Z-score approach (see Materials and Methods). The tumor sizes are also indicated.