Cell-free DNA testing for early hepatocellular carcinoma surveillance.

BACKGROUND: Liver cirrhosis (LC) is the highest risk factor for hepatocellular carcinoma (HCC) development worldwide. The efficacy of the guideline-recommended surveillance methods for patients with LC remains unpromising. METHODS: A total of 4367 LCs not previously known to have HCC and 510 HCCs from 16 hospitals across 11 provinces of China were recruited in this multi-center, large-scale, cross-sectional study. Participants were divided into Stage Ⅰ cohort (510 HCCs and 2074 LCs) and Stage Ⅱ cohort (2293 LCs) according to their enrollment time and underwent Tri-phasic CT/enhanced MRI, US, AFP, and cell-free DNA (cfDNA). A screening model called PreCar Score was established based on five features of cfDNA using Stage Ⅰ cohort. Surveillance performance of PreCar Score alone or in combination with US/AFP was evaluated in Stage Ⅱ cohort. FINDINGS: PreCar Score showed a significantly higher sensitivity for the detection of early/very early HCC (Barcelona stage A/0) in contrast to US (sensitivity of 51.32% [95% CI: 39.66%-62.84%] at 95.53% [95% CI: 94.62%-96.38%] specificity for PreCar Score; sensitivity of 23.68% [95% CI: 14.99%-35.07%] at 99.37% [95% CI: 98.91%-99.64%] specificity for US) (P < 0.01, Fisher's exact test). PreCar Score plus US further achieved a higher sensitivity of 60.53% at 95.08% specificity for early/very early HCC screening. INTERPRETATION: Our study developed and validated a cfDNA-based screening tool (PreCar Score) for HCC in cohorts at high risk. The combination of PreCar Score and US can serve as a promising and practical strategy for routine HCC care. FUNDING: A full list of funding bodies that contributed to this study can be found in Acknowledgments section.

Plasma cell-free DNA 5-hydroxymethylcytosine and whole-genome sequencing signatures for early detection of esophageal cancer.

Esophageal cancer is a highly incidence and deadly disease with a poor prognosis, especially in developing countries. Owing to the lack of specific symptoms and early diagnostic biomarkers, most patients are diagnosed with advanced disease, leading to a 5-year survival rate of less than 15%. Early (n = 50) and middle-advanced (n = 50) esophageal squamous cell carcinoma (ESCC) patients, as well as 71 healthy individuals, underwent 5-hydroxymethylcytosine (5hmC) sequencing on their plasma cell-free DNA (cfDNA). A Northern Chinese cohort of cfDNA 5hmC dataset of 150 ESCC patients and 183 healthy individuals were downloaded for validation. A diagnostic model was developed using cfDNA 5hmC signatures and then improved by low-pass whole genome sequencing (WGS) features of cfDNA. Conserved cfDNA 5hmC modification motifs were observed in the two independent ESCC cohorts. The diagnostic model with 5hmC features achieved an AUC of 0.810 and 0.862 in the Southern and Northern cohorts, respectively, with sensitivities of 69.3-74.3% and specificities of 82.4-90.7%. The performance was well maintained in Stage I to Stage IV, with accuracy of 70-100%, but low in Stage 0, 33.3%. Low-pass WGS of cfDNA improved the AUC to 0.934 with a sensitivity of 82.4%, a specificity of 88.2%, and an accuracy of 84.3%, particularly significantly in Stage 0, with an accuracy up to 80%. 5hmC and WGS could efficiently differentiate very early ESCC from healthy individuals. These findings imply a non-invasive and convenient method for ESCC detection when clinical treatments are available and may eventually prolong survival.

The development of a cSMART-based integrated model for hepatocellular carcinoma diagnosis.

BACKGROUND: Hepatocellular carcinoma (HCC) generally arises from a background of liver cirrhosis (LC). Patients with cirrhosis and suspected HCC are recommended to undergo serum biomarker tests and imaging diagnostic evaluation. However, the performance of routine diagnostic methods in detecting early HCC remains unpromising. METHODS: Here, we conducted a large-scale, multicenter study of 1675 participants including 490 healthy controls, 577 LC patients, and 608 HCC patients from nine clinical centers across nine provinces of China, profiled gene mutation signatures of cell-free DNA (cfDNA) using Circulating Single-Molecule Amplification and Resequencing Technology (cSMART) through detecting 931 mutation sites across 21 genes. RESULTS: An integrated diagnostic model called "Combined method" was developed by combining three mutation sites and three serum biomarkers. Combined method outperformed AFP in the diagnosis of HCC, especially early HCC, with sensitivities of 81.25% for all stages and 66.67% for early HCC, respectively. Importantly, the integrated model exhibited high accuracy in differentiating AFP-negative, AFP-L3-negative, and PIVKA-II-negative HCCs from LCs.

5-Hydroxymethylcytosine signature in circulating cell-free DNA as a potential diagnostic factor for early-stage colorectal cancer and precancerous adenoma.

Approximately 85% colorectal cancers (CRCs) are thought to evolve through the adenoma-to-carcinoma sequence associated with specific molecular alterations, including the 5-hydroxymethylcytosine (5hmC) signature in circulating cell-free DNA (cfDNA). To explore colorectal disease progression and evaluate the use of cfDNA as a potential diagnostic factor for CRC screening, here, we performed genome-wide 5hmC profiling in plasma cfDNA and tissue genomic DNA (gDNA) acquired from 101 samples (63 plasma and 38 tissues), collected from 21 early-stage CRC patients, 21 AD patients, and 21 healthy controls (HC). The gDNA and cfDNA 5hmC signatures identified in gene bodies and promoter regions in CRC and AD groups were compared with those in HC group. All the differential 5hmC-modified regions (DhMRs) were gathered into four clusters: Disease-enriched, AD-enriched, Disease-lost, and AD-lost, with no overlap. AD-related clusters, AD-enriched and AD-lost, displayed the unique 5hmC signals in AD patients. Disease-enriched and Disease-lost clusters indicated the general 5hmC changes when colorectal lesions occurred. Cancer patients with a confirmable adenoma history segmentally gathered in AD-enriched clusters. KEGG functional enrichment and GO analyses determined distinct differential 5hmC-modified profiles in cfDNA of HC individuals, AD, and CRC patients. All patients had comprehensive 5hmC signatures where Disease-enriched and Disease-lost DhMR clusters demonstrated similar epigenetic modifications, while AD-enriched and AD-lost DhMR clusters indicated complicated subpopulations in adenoma. Analysis of CRC patients with adenoma history showed exclusive 5hmC-gain characteristics, consistent with the 'parallel' evolution hypothesis in adenoma, either developed through the adenoma-to-carcinoma sequence or not. These findings deepen our understanding of colorectal disease and suggest that the 5hmC modifications of different pathological subtypes (cancer patients with or without adenoma history) could be used to screen early-stage CRC and assess adenoma malignancy with large-scale follow-up studies in the future.

The Landscape of Cell-Free HBV Integrations and Mutations in Cirrhosis and Hepatocellular Carcinoma Patients.

PURPOSE: Intratumoral hepatitis B virus (HBV) integrations and mutations are related to hepatocellular carcinoma (HCC) progression. Circulating cell-free DNA (cfDNA) has shown itself as a powerful noninvasive biomarker for cancer. However, the HBV integration and mutation landscape on cfDNA remains unclear. EXPERIMENTAL DESIGN: A cSMART (Circulating Single-Molecule Amplification and Resequencing Technology)-based method (SIM) was developed to simultaneously investigate HBV integration and mutation landscapes on cfDNA with HBV-specific primers covering the whole HBV genome. Patients with HCC (n = 481) and liver cirrhosis (LC; n = 517) were recruited in the study. RESULTS: A total of 6,861 integration breakpoints including TERT and KMT2B were discovered in HCC cfDNA, more than in LC. The concentration of circulating tumor DNA (ctDNA) was positively correlated with the detection rate of these integration hotspots and total HBV integration events in cfDNA. To track the origin of HBV integrations in cfDNA, whole-genome sequencing (WGS) was performed on their paired tumor tissues. The paired comparison of WGS data from tumor tissues and SIM data from cfDNA confirmed most recurrent integration events in cfDNA originated from tumor tissue. The mutational landscape across the whole HBV genome was first generated for both HBV genotype C and B. A region from nt1100 to nt1500 containing multiple HCC risk mutation sites (OR > 1) was identified as a potential HCC-related mutational hot zone. CONCLUSIONS: Our study provides an in-depth delineation of HBV integration/mutation landscapes at cfDNA level and did a comparative analysis with their paired tissues. These findings shed light on the possibilities of noninvasive detection of virus insertion/mutation.

Multi-label image categorization with sparse factor representation.

The goal of multilabel classification is to reveal the underlying label correlations to boost the accuracy of classification tasks. Most of the existing multilabel classifiers attempt to exhaustively explore dependency between correlated labels. It increases the risk of involving unnecessary label dependencies, which are detrimental to classification performance. Actually, not all the label correlations are indispensable to multilabel model. Negligible or fragile label correlations cannot be generalized well to the testing data, especially if there exists label correlation discrepancy between training and testing sets. To minimize such negative effect in the multilabel model, we propose to learn a sparse structure of label dependency. The underlying philosophy is that as long as the multilabel dependency cannot be well explained, the principle of parsimony should be applied to the modeling process of the label correlations. The obtained sparse label dependency structure discards the outlying correlations between labels, which makes the learned model more generalizable to future samples. Experiments on real world data sets show the competitive results compared with existing algorithms.