Genome-wide mapping of 5-hydroxymethylcytosines in circulating cell-free DNA as a non-invasive approach for early detection of hepatocellular carcinoma

Abstract

Objective: The lack of highly sensitive and specific diagnostic biomarkers is a major contributor to the poor outcomes of patients with hepatocellular carcinoma (HCC). We sought to develop a non-invasive diagnostic approach using circulating cell-free DNA (cfDNA) for the early detection of HCC.

Design: Applying the 5hmC-Seal technique, we obtained genome-wide 5-hydroxymethylcytosines (5hmC) in cfDNA samples from 2554 Chinese subjects: 1204 patients with HCC, 392 patients with chronic hepatitis B virus infection (CHB) or liver cirrhosis (LC) and 958 healthy individuals and patients with benign liver lesions. A diagnostic model for early HCC was developed through case-control analyses using the elastic net regularisation for feature selection.

Results: The 5hmC-Seal data from patients with HCC showed a genome-wide distribution enriched with liver-derived enhancer marks. We developed a 32-gene diagnostic model that accurately distinguished early HCC (stage 0/A) based on the Barcelona Clinic Liver Cancer staging system from non-HCC (validation set: area under curve (AUC)=88.4%; (95% CI 85.8% to 91.1%)), showing superior performance over α-fetoprotein (AFP). Besides detecting patients with early stage or small tumours (eg, ≤2.0 cm) from non-HCC, the 5hmC model showed high capacity for distinguishing early HCC from high risk subjects with CHB or LC history (validation set: AUC=84.6%; (95% CI 80.6% to 88.7%)), also significantly outperforming AFP. Furthermore, the 5hmC diagnostic model appeared to be independent from potential confounders (eg, smoking/alcohol intake history).

Conclusion: We have developed and validated a non-invasive approach with clinical application potential for the early detection of HCC that are still surgically resectable in high risk individuals.

Keywords: cancer; hepatobiliary cancer; hepatocellular carcinoma.

Figure 1

Study design. The primary aim is to develop a 5hmC-based diagnostic model for early detection of HCC) using the genome-wide 5hmC-Seal profiles derived from plasma cfDNA. A two-step procedure is designed to identify a diagnostic model for early HCC (stage 0/A). The training set and the main validation set (‘validation set 1’) are comprised of HCC samples from Zhongshan Hospital of Fudan University and The Eastern Hepatobiliary Surgery Hospital, Shanghai, China. An independent set of HCC samples from other participating hospitals (‘validation set 2’) are used to evaluate external performance of the 5hmC diagnostic model for HCC. Due to sample size limitation, only controls and patients with HCC are available in the external validation set. *The total number of study subjects does not include the 20 samples that were removed due to technical reasons. Control: healthy individuals and patients with benign liver lesions. CHB, chronic hepatitis B virus infection; cfDNA, cell-free DNA; HCC, hepatocellular carcinoma; 5hmC, 5-hydroxymethylcytosines; LC, liver cirrhosis.

Figure 2

Genomic distribution and regulatory relevance of 5hmC in cfDNA. The 5hmC-Seal data from a random set of 50 patients with HCC and 50 healthy individuals are shown. (A) The profiled 5hmC-Seal data in cfDNA are enriched in gene bodies and depleted in the flanking regions. (B, C) The profiled 5hmC-Seal data in cfDNA are enriched in liver-derived histone modification peaks and depleted in the flanking regions of (B) H3K4me1 and (C) H3K27ac. (D, E) The average fold changes of 5hmC-Seal read counts between HCC and healthy individuals are plotted against histone modification peaks derived from various adult tissues from the Roadmap Epigenomics Project for (D) H3K4me1 and (E) H3K27ac. In (A–C), the shaded area represents the first quantile to the third quantile. In (B–C), each mark at x-axis represents a region relative to the start or end positions of the histone modification peaks. In (D, E), p values of the two-sided t-tests for the ratios of two means were estimated for the fold changes between patients with HCC and healthy individuals, and are shown for the liver-derived peaks. The error bar represents the 95% CI for the fold change. cfDNA, cell-free DNA; HCC, hepatocellular carcinoma; 5hmC, 5-hydroxymethylcytosines; K: kilo base pair; TSS, transcription start site; TES, transcription end site.

Figure 3

Tissue relevance of the 5hmC-Seal data in HCC patient-derived cfDNA. (A) For the most variable genes in cfDNA samples, the number of overlapped genes between cfDNA and tumours (TU: blue line) or adjacent tissues (TI: red line) is significantly higher than that from random sampling (eg, hypergeometric test p<0.0001 for the top 500 modified genes in cfDNA). The green line indicates the shared genes across plasma cfDNA, tumours and adjacent tissues. (B, C) Within-subject correlation is significantly higher (Wilcoxon rank-sum test p<0.0001) between plasma cfDNA and TU/TI genomic DNA (diagonal line, mean of Pearson’s r: 0.88) than that between different individuals (mean of Pearson’s r: 0.73), based on the top 30 most variable genes in cfDNA samples in terms of 5hmC modification. cfDNA, cell-free DNA; HCC, hepatocellular carcinoma; 5hmC, 5-hydroxymethylcytosines; PL, plasma cfDNA; TI, adjacent tissue; TU, tumour.

Figure 4

Development and validation of a 5hmC-based diagnostic model. (A, B) The 32 marker genes used to compute the wd-scores for early HCC (stage 0/A) detection are used to generate the heatmaps for (A) the training set and (B) validation set 1. (C, D) The performance of the wd-scores, AFP, or the combination of wd-scores and AFP in distinguishing early HCC from non-HCC subjects is shown for (C) the training set and (D) validation set 1. (E) The performance of the wd-scores or AFP in distinguishing early HCC from CHB/LC is shown for the training set and validation set 1. (F) The performance of the wd-scores or AFP in distinguishing late HCC (ie, advanced stage B/C) from non-HCC or CHB/LC subjects is shown for validation set 1. Non-HCC:  CHB/LC and controls. AFP, α-fetoprotein; AUC, area under curve; CHB, chronic hepatitis B virus infection; HCC, hepatocellular carcinoma; 5hmC, 5-hydroxymethylcytosines; LC, liver cirrhosis; wd-scores, weighted diagnostic score.

Figure 5

Further evaluation of the 5hmC-based diagnostic model. (A) The performance of the wd-scores or AFP in distinguishing HCC from controls is shown for the independent validation set 2. (B) For those patients with confirmed stages, the performance of the wd-scores or AFP in distinguishing early or late HCC from controls is shown for validation set 2. (C) The boxplots show the relationships between wd-scores and the clinical diagnosis across the training set and both validation sets. Control: healthy individuals and  patients  with benign liver lesions. AFP, α-fetoprotein; AUC, area under curve; CHB, chronic hepatitis B virus infection; HCC, hepatocellular carcinoma; 5hmC, 5-hydroxymethylcytosines; LC, liver cirrhosis; wd-score, weighted diagnostic score.

Figure 6

Read distribution in candidate marker genes co-localised with cis-regulatory elements. The histone modification marks (H3K4me1, H3K27ac) from the ENCODE Project (GM12878) or liver tissue-derived data from the Roadmap Epigenomics Project are shown together with the 5hmC-Seal sequencing reads in a random set of cfDNA samples from patients with HCC and healthy individuals. The boxed regions are examples where patients with HCC and healthy individuals show differences in read distribution overlapped with histone marks or predicted enhancers. The red asterisk represents a predicted enhancer region from the ENCODE Project. Genomic positions are based on the human genome reference (hg19). (A) ESRRG (Chromosome 1q41; boxed region: chr1:216 700 111–216 704 999); and (B) SOX9 (Chromosome 17q24.3; boxed region: chr17:70 121 229–70 122 119). cfDNA, cell-free DNA; HCC, hepatocellular carcinoma; ENCODE, Encyclopaedia of DNA Elements.