ABSTRACT
-
Background/Aims
Early detection and effective prognosis prediction in patients with hepatocellular carcinoma (HCC) provide an avenue for survival improvement, yet more effective approaches are greatly needed. We sought to develop the detection and prognosis models with ultra-sensitivity and low cost based on fragmentomic features of circulating cell free mtDNA (ccf-mtDNA).
-
Methods
Capture-based mtDNA sequencing was carried out in plasma cell-free DNA samples from 1168 participants, including 571 patients with HCC, 301 patients with chronic hepatitis B or liver cirrhosis (CHB/LC) and 296 healthy controls (HC).
-
Results
The systematic analysis revealed significantly aberrant fragmentomic features of ccf-mtDNA in HCC group when compared with CHB/LC and HC groups. Moreover, we constructed a random forest algorithm-based HCC detection model by utilizing ccf-mtDNA fragmentomic features. Both internal and two external validation cohorts demonstrated the excellent capacity of our model in distinguishing early HCC patients from HC and high-risk population with CHB/LC, with AUC exceeding 0.983 and 0.981, sensitivity over 89.6% and 89.61%, and specificity over 98.20% and 95.00%, respectively, greatly surpassing the performance of alpha-fetoprotein (AFP) and mtDNA copy number. We also developed an HCC prognosis prediction model by LASSO-Cox regression to select 20 fragmentomic features, which exhibited exceptional ability in predicting 1-year, 2-year and 3-year survival (AUC=0.8333, 0.8145 and 0.7958 for validation cohort, respectively).
-
Conclusions
We have developed and validated a high-performing and low-cost approach in a large clinical cohort based on aberrant ccf-mtDNA fragmentomic features with promising clinical translational application for the early detection and prognosis prediction of HCC patients.
-
Keywords: Hepatocellular carcinoma; Circulating cell free mitochondrial DNA; Fragmentomics; Early diagnosis; prognosis
Study Highlights
• The ccf-mtDNA fragmentomic features of patients with HCC are significantly distinct from those of CHB/LC patients and healthy controls.
• The HCC detection model demonstrates superior capability in discriminating patients with early HCC from CHB/LC patients and healthy controls, greatly outperforming AFP and mtDNA copy number.
• The HCC prognosis prediction model exhibits excellent predictive performance as an independent biomarker.
Graphical Abstract
INTRODUCTION
Hepatocellular carcinoma (HCC) ranks second among the leading causes of cancer-related death worldwide (780,000 annual deaths) [
1]. Indeed, more than 50% of new HCC cases and related mortality are present in China, where chronic hepatitis B virus (HBV) infection is a major contributor to the pathophysiological progression of HCC, accounting for approximately 85% of HCC cases [
2]. Additionally, the incidence of HCC caused by chronic hepatitis C virus and other factors such as nonalcoholic fatty liver disease is also on the rise [
3,
4]. Despite the existence of various treatment options, the 5-year survival rate for HCC patients remains below 50%, which is mainly attributed to the usual diagnosis at an advanced stage [
5]. Furthermore, the lack of reliable prognosis prediction tools exacerbates the undesirable situation, making it challenging to achieve optimal therapeutic outcomes for most patients. Currently, alpha-fetoprotein (AFP) is the most commonly used bloodbased tumor biomarker for diagnosis of HCC and surveillance after treatment. However, it has shown serious limitations in terms of sensitivity and specificity [
6]. Consequently, there is an urgent need to develop more reliable HCC detection tools in the clinical setting.
In recent years, significant progress has been made in liquid biopsy for cancer detection through the development of innovative blood-based cell-free DNA (cfDNA) biomarkers [
7]. Among them, somatic mutation-based approaches are usually dependent on the tissue-based mutation identification and significantly impacted by background noise [
8]. The copy number variations-based methods and methylation profiling are limited by potentially low sensitivity, particularly in early-stage cancers where circulating tumor DNA levels may be relatively low [
9-
11]. Recent studies highlight the importance of cfDNA fragmentomic features in cancer detection. Foda et al. [
12] and Zhang et al. [
13] have constructed the high-performing models for early detection of liver cancer based on cfDNA fragmentomic features derived from low-depth whole-genome sequencing (WGS) data. However, it is important to consider the incomplete representation of tumor genomic profiles due to the relatively low sequencing coverage when low-depth WGS method was employed for cfDNA analysis, thus leading to insufficient repeatability of results.
In the cellular context, mitochondria possess their own double-stranded circular genomic DNA (mtDNA), which is approximately 16.6 kb in length and exists in high copy numbers [
14-
16]. Previously, we have demonstrated that somatic mutations and copy number change of mtDNA are closely related to HCC progression and found the existence of tumor-derived mtDNA mutations in the plasma of HCC patients [
17,
18]. Our investigations have also demonstrated the potential diagnostic value of analyzing end motifs and mutation profiles in urine ccf-mtDNA across various cancer types [
19]. Our finding highlights the promising role of ccf-mtDNA in cancer detection. However, it is important to acknowledge that using ccf-mtDNA copy number (CN) alone may not be highly accurate in distinguishing cancer patients from controls [
20]. Therefore, we have developed a cost-effective approach that can simultaneously detect copy number, mutation and fragmentomics of circulating cell free mtDNA (ccf-mtDNA) [
21]. Importantly, we have observed aberrant fragmentomic features of ccf-mtDNA including fragmentation profile, 5’ end base preference and motif diversity in six cancer types and provided a proof-of-principle for the ccf-mtDNA fragmentomics-based cancer detection and tissue-of-origin identification (submitted paper under review). Nevertheless, to date, no study has validated the ccf-mtDNA fragmentomics-based approach for HCC detection and prognosis prediction across different population.
Here, high-efficiency capture-based mtDNA sequencing was employed to analyze ccf-mtDNA fragmentomic features in 1,168 individuals including patients with HCC, chronic hepatitis B (CHB), liver cirrhosis (LC) and healthy control (HC). Then, we focused on the construction and validation of HCC detection and prognosis prediction models by utilizing the ccf-mtDNA fragmentomic features. Furthermore, the performance of these models was comprehensively evaluated in training and validation cohorts.
MATERIALS AND METHODS
Participant enrollment
In this study, a total of 1,168 participants were enrolled from Xijing Hospital (March 2019 to December 2019) and Tangdu Hospital (March 2019 to February 2020) of Fourth Military Medical University (FMMU) and Changhai Hospital of Second Military Medical University (March 2019 to September 2019), including patients with HBV-related HCC (n=541), non-HBV-related HCC (n=30), patients with CHB (n=125), patients with LC (n=176) and healthy controls (n=296). The demographic and clinicopathological characteristics were summarized in
Supplementary Table 1. Patients with HCC, CHB or LC were diagnosed based on the American Association for the Study of Liver Diseases (AASLD) guidelines [
22]. HC was collected from individuals who underwent routine physical examinations. The inclusion criteria and exclusion criteria of patients were described in the
Supplementary information. Tumor stage was determined according to the Barcelona Clinic Liver Cancer (BCLC) staging system. The HCC patients from BCLC stage 0-C were selected and enrolled in this study. All HCC patients included in this study were subjected to surgical removal for confirmation of pathological diagnosis and then followed up every 6 months. The vascular invasion, lymph node metastasis, and bile duct invasion were confirmed by histopathology. Finally, follow-up data was successfully collected from 388 HCC patients, with the final follow-up time in December 2023. The median follow-up duration was 29.9 months, ranging from 3 to 57 months, and 263 patients were dead of HCC during the follow-up. Overall survival (OS) was considered as the time span from diagnosis to death or last follow-up. Recurrence-free survival (RFS) was considered as the time from surgery to disease recurrence, death, or the last follow-up. Diseasefree survival (DFS) was defined as the time from surgery to the occurrence of disease recurrence, the development of a second primary cancer, death or the last follow-up. This study was approved by the three Hospital Review Boards and all participants provided written informed consent.
Sample collection was performed following the acquisition of diagnostic information. Peripheral blood samples (5 mL for each) from all participants were collected 1 day before treatment and plasma was separated within 2 hours using a standard two-step centrifugation protocol and then stored at –80°C.
DNA extraction, library construction, next generation sequencing and sequencing data analysis
The QIAamp Circulating Nucleic Acid kit (Qiagen, USA) was utilized for DNA extraction from the plasma according to the manufacturer’s instructions [
19]. The DNA was quantified using Qubit 4.0 (Thermo Fisher, USA). Subsequently, the whole-genome sequencing (WGS) libraries were constructed using plasma cfDNA (10–20 ng for each sample) with the NEB ultra v2 kit (NEB, US) as previously described [
23]. The number of PCR cycles during library construction was 12. Then, the WGS libraries were hybridized with homemade biotinylated mtDNA capture probes as previously described [
17]. Finally, all captured mtDNA libraries were sequenced by the Illumina Hiseq XTen platform using paired-end runs with 2×150 cycles (PE 150). To minimize bias resulting from systematic errors in the experiment, we ensured that samples of different disease types are included in the same batch of experiments, and standardized experimental procedures were followed. And, all experiments were carried out by two individuals in a centralized laboratory, utilizing the same instruments across all samples.
After sequencing, Illumina adapters and bases with quality scores below 30 were trimmed in raw sequencing data using Fastp software (v.0.20.0) [
24]. Furthermore, to minimize contaminations from nuclear sequences of mitochondrial origin (NUMTs), qualified reads were aligned to both the revised Cambridge reference sequence (rCRS) and the human genome reference (hg19) using BWA software (v.0.7.1521). Subsequently, reads that only mapped to rCRS were retained for further analyses. Picard Tools software (v.1.119) was applied for sorting the reads and the duplicated reads were removed. The Genome Analysis Toolkit 4 software (v.3.2-2) was employed for local realignment. The final mtDNA coverage depths ranged from 2,469 X to 3,383 X for capture-based mtDNA sequencing data (
Supplementary Table 2).
Based on our established method, the average sequencing depth of mtDNA and the average sequencing depth of the nuclear reference gene were used to calculate the relative mtDNA CN. And the following formula was used to calculate the mtDNA copy: 2×average depth of mtDNA/average depth of nuclear reference gene [
21].
Two categories of ccf-mtDNA fragmentomic features including fragment-related and 5’ end-related features were extracted from the mtDNA capture based sequencing (mtDNA Cap-seq) data and detailed methods were described in the
Supplementary information.
A total of 210 patients with HCC and 240 non-HCC individuals (120 CHB/LC and 120 HC) from Xijing Hospital were randomly divided into a training cohort and an internal validation cohort at a ratio of 7:3. A total of 210 patients with HCC and 240 non-HCC individuals (120 CHB/LC and 120 HC) from Tangdu Hospital and 121 patients with HCC and 117 non-HCC individuals (61 CHB/LC, and 56 HC) from Changhai Hospital were used as external validation cohorts 1 and 2, respectively.
The training cohort was used to construct the HCC detection model utilizing fragment-related, 5’ end-related and all fragmentomic features, respectively, using the random forest (RF) algorithm from R package ‘randomForest’ (v.4.6-14) with the number of trees set to 500 and six features randomly selected for each tree. The ten-fold cross validation was used to evaluate the model performance in the training cohort. Consequently, the HD score for each patient was determined from the HCC detection model. A detection model integrating the HCC detection (HD) score and AFP was formulated through logistic regression in the R package ‘stats’ (v.4.1.2). Internal and external validation cohorts were used for evaluating the performance of HCC detection model based on all fragmentomic features. The performance of our model was also evaluated in a non-HBV-related HCC cohort.
Construction and validation of HCC prognosis prediction model
The LASSO-Cox regression analysis, using CoxPHSurvivalAnalysis model in ‘scikit-survival’ package (v.0.22.1), was utilized to develop the HCC prognosis prediction model based on all ccf-mtDNA fragmentomic features in the training HCC cohort, which consisted of 197 patients with HCC from Xijing Hospital. A five-fold cross validation was conducted to identify the minimum lambda value, resulting in 20 features with non-zero coefficients. Subsequently, the HPP score (range: 0 to 1) was calculated as the sum of values weighted by the coefficients from the Cox hazard model. Based on the median HCC prognosis prediction (HPP) score of HCC patients in the training cohort, all HCC patients were divided into HPP score ≤0.65 and HPP score >0.65 groups. Next, a combined HCC prognosis prediction model integrating the HPP score and clinical characteristics (BCLC stage, Child-Pugh class and AFP level) was formulated through multivariate Cox regression. The predictive ability of the HCC prognosis prediction model was evaluated in the validation HCC cohort, which consisted of 191 patients with HCC from Tangdu Hospital.
Totally, based on our sequencing and analysis methods, it will take 5–7 working days to accomplish one processing pipeline (including cfDNA extraction, library construction, next generation sequencing and data analysis) and thus obtain the fragmentomic data of ccf-mtDNA, with cost of 30–40 US dollars for each sample.
Statistical analysis
Statistical analyses were performed using GraphPad Prism v.9.5.1 (GraphPad Software). Detailed methods were described in the
Supplementary information. All
P values reported were two-tailed and a significance level of 0.05 was used.
RESULTS
Characteristics of participants
A total of 1168 participants including 541 patients with HBV-related HCC, 30 patients with non-HBV-related HCC, 125 patients with CHB, 176 patients with HBV-related LC and 296 HC were recruited from three hospitals into the training and validation cohorts (
Fig. 1). As shown in
Supplementary Figure 1, our data indicated that patients with HCC were significantly older than HC and patients with CHB/LC (both
P<0.05). Further analysis revealed a higher male percentage in the HCC group compared to the CHB/LC and HC groups. In addition, no significant difference in the demographic and clinicopathological characteristics was observed among the Xijing, Tangdu and Changhai cohorts (all
P>0.05,
Supplementary Table 1).
A comprehensive analysis of ccf-mtDNA fragmentomic features was conducted based on capture-based NGS data with an average sequencing depth of 2,882 X (
Supplementary Table 2) among 296 HC, 125 CHB, 176 LC and 541 HCC patients from both the training and all validation cohorts. Our results indicated no difference of ccf-mtDNA fragmentomic features between the patients with CHB and LC and thus these two groups were combined for subsequent analyses (
Supplementary Fig. 2). Among four groups consisting of HC, patients with CHB/LC, HCC patients at the early (EHCC) and late BCLC stage (LHCC), we observed a gradual decrease of fragment size from HC to LHCC (
Fig. 2A,
B). Furthermore, to investigate the fragmentation profile of ccf-mtDNA at single-base resolution, the fragment size distribution (FSD) score was determined as the overall coverage ratio of long to short ccf-mtDNA fragments covering the given site, with the median length as a cutoff. Our data revealed that both EHCC and LHCC groups showed highly variable fragmentation profile when compared with other two groups (
Fig. 2C). A gradual decrease of correlation to median HC profile was observed from HC to LHCC group (
Fig. 2D). Subsequently, the peak number of the fragmentation profile was calculated for all groups and new peaks were defined based on the comparison of peak location with the median HC profile. Our data showed a gradual increase in total peak number and new peak number from HC to LHCC group (
Fig. 2E,
F). In addition, as shown in
Figure 2G,
H, our results revealed a significant difference in the 5’ end base preference among four groups. Specifically, we observed an overrepresentation of C-end and an underrepresentation of G-end in the LHCC group compared to other three groups. In comparison, the T-end base preference exhibited an increased trend from CHB/LC to HC and EHCC to LHCC, whereas A-end base preference exhibited a decreased trend from CHB/LC to LHCC. Furthermore, we calculated the 5’ end motif diversity score (MDS) based on the proportion of 4-mer end base motif and observed an increased trend from CHB/LC and EHCC to LHCC to HC (
Fig. 2I). When ccf-mtDNA fragmentomic features among the four groups were stratified by cohort, very similar results were obtained (
Supplementary Fig. 3). Also, some of the 5’ end features such as 5’ C- and G- end base preference and 5’ MDS did not differ between the CHB/LC and EHCC groups. We also observed no obvious association between age or gender and the ccf-mtDNA fragmentomic features in all groups (
Supplementary Figs. 4,
5). Altogether, these findings indicate aberrant ccf-mtDNA fragmentomic features in patients with HCC, strongly supporting the clinical application as promising detection biomarkers.
Based on fragment-related and 5’ end-related ccf-mtDNA fragmentomic features, the widely utilized RF algorithm was used to construct the HD model in the training dataset from Xijing cohort (
Supplementary Fig. 6). A HD score was generated by ten-fold cross-validation for each individual to assess the performance of the HD model in distinguishing HCC from non-HCC individuals (including at-risk subjects with CHB/LC and healthy controls) and at-risk subjects with CHB/LC. As shown in
Supplementary Figure 7, the HD model based on all fragmentomic features exhibited a superior area under the curve (AUC) in distinguishing HCC from non-HCC (0.9965, 95% CI 0.9937–0.9994) and CHB/LC (0.9943, 95% CI 0.9894–0.9992) when compared with models based on fragment-related or 5’ end-related features.
Furthermore, we tested the all fragmentomic features-based HD model in the internal validation dataset (n=135) from Xijing cohort. Our data revealed that both EHCC and LHCC groups exhibited significantly higher HD score than HC and CHB/LC groups (
Fig. 3A). As shown in
Figure 3B and
Supplementary Table 3, the HD score indicated the excellent capacity for distinguishing patients with early HCC from non-HCC subjects (AUC=0.9910, 95% CI 0.9814–1.000); sensitivity=92.06%; specificity=94.44%; score cut-off=0.4968). Considering that CHB and LC in China are at a significantly higher risk of developing HCC and often complicate the diagnosis of HCC, we evaluated the performance of our HD model in distinguishing EHCC from at-risk individuals with CHB/LC. Our results showed a high AUC of 0.9903 (95% CI 0.9778–1.000), with sensitivity of 92.06% and specificity of 94.44%. As expected, our HD model exhibited very good performance for distinguishing EHCC from healthy controls (AUC=0.9916). Moreover, late HCC (BCLC stage B/C) patients were perfectly distinguished by our HD model from non-HCC, CHB/LC and healthy controls with AUC of 0.9934, 0.9921, and 0.9947, respectively.
To verify the applicability of our model in clinical practice, the performance of fixed HD model was then evaluated in two external validation cohorts. Very similar in internal validation cohort, a notable increase in HD score was observed from HC and CHB/LC to LHCC in both external validation cohorts (
Fig. 3D–
I). Furthermore, the results demonstrated remarkable discriminatory capacity of our HD model in distinguishing patients with EHCC from non-HCC in both external val idat ion cohor t 1 and 2 (AUC=0.9861 and 0.9826), as well as distinguishing patients with EHCC from CHB/LC (AUC=0.9840 and 0.9818) and healthy controls (AUC=0.9882 and 0.9835). Very similar performance was obtained in distinguishing patients with LHCC from non-HCC, CHB/LC and HC in both external validation cohorts 1 and 2. In addition, by applying HD model in a set of non-HBV-related HCC (
Supplementary Table 4), including 56 HCV-related HCC and 64 HCC patients without hepatitis, we observed a pretty good performance in distinguishing them from CHB/LC and HC (all AUC>0.9385), although the power was to some extent decreased when compared with those in HBV-related HCC due to the absence of non-HBV-related HCC in the training cohort (
Supplementary Fig. 8). These results clearly indicate the high performance and generalization of ccf-mtDNA fragmentomics-based HD model in HCC early detection.
To evaluate the real-world influence of ccf-mtDNA fragmentomics-based strategy in the context of HCC detection, we compared the performance of our HD model with the current screening measurement of AFP level. As shown in
Figure 4A,
B, in the most challenging clinical scenario of distinguishing EHCC from CHB/LC, our HD model exhibited superior performance to AFP-based method in both external validation cohort 1 (AUC=0.9840 vs. 0.8238) and 2 (AUC=0.9818 vs. 0.8407). Accordingly, HD model had the sensitivity of 89.93% and 89.61% at a score cut-off of 0.4968 in both cohorts, while AFP (cut-off=20 ng/mL) only achieved the sensitivity of 56.83% and 62.34%, which is consistent with previous reports [
12]. Very similar results were observed in distinguishing LHCC from CHB/LC (
Fig. 4C,
D). Additionally, the combination of HD score and AFP only provided a very slight improvement in both EHCC and LHCC detection over the HD model alone in both external validation cohorts. Importantly, our HD model still demonstrated remarkable capacity in distinguishing 79 and 39 HCC patients with AFP level below 20 ng/mL (AFP-negative HCC) who would have been misclassified based on AFP alone from CHB/LC, with AUC of 0.9863 and 0.9859 in external validation cohort 1 and 2, respectively (
Fig. 4E,
F and
Supplementary Table 5). As expected, consistent performance of HD model was achieved in distinguishing AFP-positive HCC from CHB/LC in both cohorts (
Fig. 4G,
H).
Furthermore, considering the previous reports that mtDNA CN in plasma cfDNA may be used as biomarker for HCC detection, we evaluated the performance of mtDNA CN in differentiating HCC from CHB/LC and HC. Similar to previous reports, both EHCC and LHCC groups exhibited a significant increase in mtDNA CN compared to HC and CHB/LC groups (
Supplementary Fig. 9). When mtDNA CN was used as a classifier, an AUC of 0.5809 and 0.6278 was achieved in distinguishing EHCC patients from HC and patients with CHB/LC, respectively, while a slightly increased AUC of 0.6642 and 0.7061 was achieved in distinguishing LHCC patients from HC and patients with CHB/LC. Furthermore, no significant difference in distinguishing HCC from non-HCC was found between the combination of the HD score & mtDNA CN and the HD score alone (
Fig. 4I–
L). Taken together, our results highlight the utility of ccf-mtDNA fragmentomic features as much more effective alternative biomarkers for HCC detection.
As tumor biomarkers may be affected by clinical characteristics, we comprehensively investigated the association of the HD score with demographic variables such as age and gender or measures of liver dysfunction in HC, CHB/LC and HCC groups gathered from two validation cohorts. Our data revealed no obvious association between the HD score and age or gender in all three groups, as well as between the HD score and measures of liver dysfunction in both CHB/LC and HCC groups (all
P>0.05,
Supplementary Fig. 10). Moreover, we further examined the relationship between HD score and variables of HCC progression. Notably, as shown in
Figure 5A, HD score significantly increased as BCLC stage advanced in HCC patients (
P for trend <0.01). In particular, HD score just showed a very slight decrease of detection power for HCC at stage 0 from CHB/LC (AUC=0.9343, sensitivity=77.78%) when compared with those at other three stages (AUC=0.9852, 0.9869 and 1.000, sensitivity=90.40%, 90.67%, and 100.0%, respectively,
Fig. 5B,
C). In addition, a significantly higher HD score was also observed in HCC patients with Child-Pugh class B or increasing tumor size (>2 cm) or more tumor sites (>1) when compared with corresponding control groups (all
P<0.05,
Fig. 5A). And patients with HCC who had the features of vascular invasion, lymph node metastasis, or bile duct invasion exhibited significantly higher HD scores compared to those without these features (all
P<0.001,
Supplementary Fig. 11A). In these subgroups, HD score exhibited almost equivalent detection performance (
Fig. 5B,
C and
Supplementary Fig. 11B,
C). These observations suggest that ccf-mtDNA fragmentomics-based HD model may not be remarkably affected by clinical characteristics, indicating better clinical applicability.
To evaluate the applicability of ccf-mtDNA fragmentomic features in the prognosis prediction of HCC patients undergoing surgical treatment, we assigned 197 HCC patients with complete survival information from Xijing Hospital into a training HCC cohort and 191 HCC patients from Tangdu Hospital into a validation HCC cohort. LASSO-Cox regression was employed to diminish dimensionality and construct an HCC prognosis prediction model which included 20 features (
Supplementary Fig. 12). The patients with HCC were first stratified into high-risk and low-risk groups by the median HPP score of 0.65 generated from the HCC prognostic prediction model constructed based on training HCC cohort. Subsequently, Kaplan–Meier curves revealed a significantly shorter median OS in the high-risk group compared to the low-risk group (log-rank test,
P<0.0001; HR 0.30, 95% CI 0.21–0.44,
Fig. 6A). The AUC of HPP score for predicting 1-year, 2-year, and 3-year survival were 0.8838, 0.8558, and 0.8347, respectively, in the training HCC cohort (
Fig. 6B).
As shown in
Supplementary Table 6, the univariable and multivariable Cox regression analyses indicated that the HPP score was significantly associated with OS and may serve as an independent prognostic factor. Furthermore, the prognosis prediction performance of the HPP score was evaluated in validation HCC cohort. As shown in
Figure 6C, the median OS time in the high-risk group was significantly shorter than that in the low-risk group (log-rank test,
P<0.0001; HR 0.37, 95% CI 0.26–0.53). And the AUC of HPP score for predicting 1-year, 2-year and 3-year survival were 0.8333, 0.8145 and 0.7958, respectively (
Fig. 6D). Stratified analysis indicated that significant association between HPP score and overall survival existed in all subgroups (all
P<0.001,
Fig. 6E), further supporting the independent role of ccf-mtDNA fragmentomic features in HCC prognosis prediction. The predictive performance of the HPP score for RFS and DFS was further evaluated. Kaplan–Meier curves demonstrated a significantly shorter median RFS and DFS in the high-risk group compared to the low-risk group in both the training (HR 0.32 and 0.31, respectively) and validation (HR 0.39 and 0.37, respectively) HCC cohorts (log-rank test,
P<0.0001,
Supplementary Fig. 13). Notably, the combination of HPP score and clinical characteristics significantly improved the ability to predict HCC prognosis in both the training (HR 0.12,
Supplementary Fig. 14A,
B) and validation HCC cohorts (HR: 0.14,
Supplementary Fig. 14C,
D). Altogether, these results demonstrate that ccf-mtDNA fragmentomic features can contribute to death risk stratification and prognosis prediction in patients with HCC.
DISCUSSION
Herein, we observed the aberrant fragmentomic features of ccf-mtDNA in patients with HCC and developed a novel, ultrasensitive, and low-cost blood-based approach (
Supplementary Table 7) for early detection and prognosis prediction of HCC with high performance, which has been independently validated across different cohorts [
25,
26]. To the best of our knowledge, this is the first comprehensive analysis and clinical application of ccf-mtDNA fragmentomic features in HCC. Our findings highlight the significance of ccf-mtDNA as a non-invasive biomarker for HCC patients.
In previous studies [
27], a shorter fragment size of ccf-mtDNA was observed in plasma samples from HCC patients compared to patients with CHB/LC and healthy controls, presumably due to more release of short mtDNA fragment from HCC cells into circulation. Our previous observation and other studies have also demonstrated the 5’ end base preference of ccf-DNA and ccf-mtDNA in plasma and urine samples, indicating that their fragmentation is not random, which is mainly determined by various DNA nucleases with specific cutting preference [
19,
28,
29]. In the present study, we observed the aberrant fragmentomic features of ccf-mtDNA in patients with HCC, including fragment size, fragmentation profile, 5’ end base preference and 5’ end MDS. HCC is a solid tumor characterized by high heterogeneity and malignancy. It is likely that the types and quantities of nucleases present in or released by HCC cells vary significantly, which may at least partially explain the variation in fragmentomic features of ccf-mtDNA in HCC patients. In addition, our recent observation revealed a significant association between ccf-mtDNA fragmentation profile and protein binding of mtDNA (submitted paper under review). Furthermore, a previous study has reported the decreased expression level of mitochondrial transcription factor A (TFAM), an abundant binding protein of mtDNA, in most HCC patients [
30]. These findings provide another plausible explanation for the aberrant ccf-mtDNA fragmentomic features in HCC patients. In the present study, we also observed, to some extent, changes in ccf-mtDNA fragmentomic features of CHB/LC patients compared to those of healthy controls, suggesting another possibility that certain diseases may exert a systemic influence on protein-binding of mitochondrial genome or plasma environments to generate ccf-mtDNA fragmentomics. Moreover, it is necessary to further explore the mechanism underlying the development of aberrant fragmentomic features in future studies.
Considering a large well-defined high-risk population with a 3% to 4% risk of developing HCC annually, routine cancer screening for HCC is recommended every 6 months. Unfortunately, currently available tests such as AFP detection and ultrasonography have limited capacity, especially for early-stage disease. Previously, based on the cfDNA fragmentation profile from low-depth WGS data, Foda et al. [
12] have developed an HCC detection model, achieving an AUC of 0.90 in distinguishing patients with HCC from CHB/LC. In addition, Zhang et al. [
13] have also constructed a model for primary liver cancer detection by applying multiple machine learning algorithms to integrate cfDNA fragmentation profile and achieved an AUC of 0.98. However, the aforementioned studies still need to be confirmed by much more sample size or external validation cohorts. In the present study, we have for the first time constructed an HCC detection model based on ccf-mtDNA fragmentomic features, exhibiting an excellent performance, especially for early HCC. Importantly, our model stands out from previous approaches due to several advantages. By capture-based mtDNA sequencing, our approach can achieve very high mtDNA coverage depth at a low sequencing data quantity and cost, enabling a comprehensive and high-resolution analysis of ccf-mtDNA fragmentomic features, which further ensures the reproducibility of our detection model. Another noticeable advantage of our approach is the insightful exploration of ccf-mtDNA fragmentation profiles by applying both qualitative and quantitative dissection of features, which enables the efficient utilization of tumor-relevant characteristics and thus significantly enhances the overall performance of HCC detection model. Also, further studies on differential diagnosis between low grade degenerative nodule (DN) or high grade DN and HCC are warranted.
Currently, AFP stands as the most extensively utilized biomarker for detecting HCC, yet its application remains fraught with numerous inherent challenges, including low sensitivity in detecting early HCC and the inability to reliably distinguish HCC from CHB/LC [
6]. Our results indicated that ccf-mtDNA fragmentomics-based model greatly outperformed AFP, particularly in distinguishing early HCC from high-risk individuals with CHB/LC. Notably, our model demonstrated a superior capability to accurately identify HCC patients with low AFP levels. Previous studies have also reported that lens culinaris agglutinin-reactive fraction of AFP (AFP-L3) and protein induced by vitamin K absence II (PIVKA-II) are potential biomarkers for diagnosing HCC with certain clinical utility, though their application is less widespread compared to AFP [
31,
32]. In addition, van der Pol et al. [
20] have reported the insufficient ability of mtDNA CN as a diagnostic biomarker for cancer patients with an AUC of 0.65, which is consistent with our results. The reason for this may be that the mtDNA CN represents simple information on mtDNA content in plasma samples, which is affected not only by tumor status, but also by many other factors, such as degradation, excretion, and sample storage. However, the HD score reflects more complete information about the tumor patients. Furthermore, our results demonstrated that the robustness of HD model was independent of various confounding factors such as age and gender. Subgroup analyses further substantiated the ultrasensitivity of our model, particularly in HCC at stage 0 and those with small tumors (≤2 cm), which are often missed by conventional approaches. These observations provide the compelling supporting evidence for the generalizability of our model.
Previously, Xu et al. [
10] and Wang et al. [
11] have developed liquid biopsy-based models for HCC prognosis prediction based on ctDNA methylation markers and cfDNA copy number variations, respectively. In the present study, we have constructed an HCC prognosis prediction (HPP) model, which exhibited superior predictive performance compared to previous models, particularly in estimating 1-year and 3-year survival rates. Concurrently, the combination of our HPP model with clinical characteristics clearly enhanced predictive performance for HCC prognosis. To our knowledge, this is the first investigation to explore the application of ccf-mtDNA fragmentomic features in predicting prognosis. Moreover, it warrants deep exploration of the biological mechanism underlying the association of ccf-mtDNA fragmentomic features with HCC prognosis.
Although our results presented here are highly promising, we acknowledge two limitations which also need to be addressed. First, our present study mainly focuses on HBV-related Chinese patients. Moreover, it is necessary to further validate the HD and HPP models across diverse ethnicities and HCC etiologies for their generalizability through the inclusion of international multi-etiology cohorts. Furthermore, the feasibility of employing ccf-mtDNA fragmentomic features for real-time monitoring of therapeutic responses and minimal residual disease remains an open question, necessitating further investigation.
In summary, we have systematically identified the aberrant fragmentomic features of ccf-mtDNA in HCC patients and developed a novel ultrasensitive and low-cost model for both early diagnosis and prognosis prediction by only relying on capture-based mtDNA sequencing. Independent validation provides compelling evidence for the practical application of our models in assisting the current approaches for HCC patients and their different subgroups, thereby laying a solid foundation for the development of blood-based pan-cancer screening tools.
ACKNOWLEDGMENTS
This work was supported by the National Natural Science Foundation of China (grant no. 82330073), the Science & Technology Co-ordination and Innovation Project of Shaanxi Province, China (grant no. 2023-ZDLSF-46), the clinical research program of Air Force Medical University (grant no. 2021LC2104).
FOOTNOTES
-
Authors’ contribution
J.X. conceived and designed the project. Y.L., K.Z., W.G. and H.Z. performed selection of patients, collection of samples and clinical data and mtDNA sequencing. Y.L., F.P., S.W., H.J. and S.G. analyzed the sequencing data. Y.L., F.P., S.W. and J.X. wrote the manuscript. All co-authors reviewed and approved the final draft of the manuscript.
-
Conflicts of Interest
The authors have no conflicts to disclose.
SUPPLEMENTAL MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (
http://www.e-cmh.org).
The raw sequencing data underlying this article are available in BIG Data Center, Beijing Institute of Genomics (BIG) with access number PRJCA022690.
Supplementary Figure 1.
Comparative analyses of age and gender among HC, CHB/LC and HCC groups. (A) Comparison of age among three groups. (B) The proportion of male and female subjects in the three groups. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma; ns, not significant. ***P<0.001. In (A), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 2.
Similar ccf-mtDNA fragmentomic features between the patients with CHB and LC. Comparison of (A) fragment size distribution, (B) the proportion of short and long fragments, (C) medium ccf-mtDNA fragmentation profiles, (D) the correlation to ccf-mtDNA fragmentation profile of median HC, (E) 5′ end base preference, and (F) 5′ end MDS between CHB group (n=125) and LC group (n=176). CHB, chronic hepatitis B; LC, liver cirrhosis; MDS, motif diversity score; ns, not significant. In (B) and (D–F), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 3.
Comparative analyses of ccf-mtDNA fragmentomic features in three cohorts. Comparison of (A) the proportion of short fragments, (B) the proportion of long fragments, (C) the correlation to ccf-mtDNA fragmentation profile of median HC, (D) the number of total peak, (E) the number of new peak, (F–I) 5′ end base preference, and (J) 5′ end MDS among HC, CHB/LC, EHCC and LHCC groups in Xijing, Tangdu and Changhai cohorts. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; EHCC, early-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 and A); LHCC, late-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage B to D); MDS, motif diversity score; ns, not significant; *P<0.05; **P<0.01; ***P<0.001. Boxes represent the 25th and 75th percentiles, center line indicates the median, and whiskers extend to the maximum and minimum values within 1.5× interquartile range.
Supplementary Figure 4.
The association of age and ccf-mtDNA fragmentomic features in HC, CHB/LC and HCC groups. Comparison of (A) the proportion of short fragments, (B) the proportion of long fragments, (C) medium ccf-mtDNA fragmentation profiles, (D) the correlation to ccf-mtDNA fragmentation profile of median HC, (E) 5′ end base preference and (F) 5′ end MDS between the individuals younger than the first quartile and those older than the third quartile in three groups. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma; Q1, first quartile; Q3, third quartile; MDS, motif diversity score; ns, not significant. In (A, B) and (D–I), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 5.
The ccf-mtDNA fragmentomic features of male and female in HC, CHB/LC and HCC groups. Comparison of (A) the proportion of short fragments, (B) the proportion of long fragments, (C) medium ccf-mtDNA fragmentation profiles, (D) the correlation to ccf-mtDNA fragmentation profile of median HC, (E) 5′ end base preference and (F) 5′ end MDS between male and female subjects in three groups. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma; MDS, motif diversity score; ns, not significant. In (A, B) and (D–I), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 6.
Establishment of HCC detection model based on ccf-mtDNA fragmentomic features. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma; mtDNA Cap-seq, mtDNA capture based sequencing; HD, HCC detection; AUC, area under the curve.
Supplementary Figure 7.
Performance of random forest (RF) models based on different fragmentomic features. Receiver operating characteristic curves evaluating the performance of three models based on fragment-related (blue), 5′ end-related (purple) and all fragmentomic features (orange) using RF algorithm for distinguishing HCC from non-HCC (A) and CHB/LC (B). HCC, hepatocellular carcinoma; CHB, chronic hepatitis B; LC, liver cirrhosis; AUC, area under the curve.
Supplementary Figure 8.
Performance of HCC detection model on detecting non-HBV-related HCC. Receiver operating characteristic curves evaluating the performance of HCC detection model for distinguishing non-HBV-related HCC from CHB/LC (A) and HC (B). (C) Sensitivity of HD model in HCC with different etiologies. HCC, hepatocellular carcinoma; HCV, hepatitis C virus; CHB, chronic hepatitis B; LC, liver cirrhosis; AUC, area under the curve; MASLD, metabolic dysfunction-associated steatotic disease. Factors such as aflatoxin B1, tobacco, and autoimmune hepatitis were integrated into “other HCC”.
Supplementary Figure 9.
Comparative analyses of mtDNA copy number among four groups. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; EHCC, early-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 and A); LHCC, late-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage B to D). ns, not significant; *P<0.05; ***P<0.001. Center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 10.
The association of HD score with age, gender and measures of liver dysfunction in HC (A), CHB/LC (B) and HCC (C) groups. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma. ns, not significant. Center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Supplementary Figure 11.
Distribution and performance of HD score in different HCC subgroups. (A) Comparison of HD score between HCC patients with VI, LNM, or BDI and those without any of these conditions. (B) The ROC curves evaluating the performance of HD model in distinguishing HCC patients with VI, LNM, or BDI and those without any of these conditions from CHB/LC. (C) Sensitivity of HD model in different HCC subgroups. HCC, hepatocellular carcinoma; HD, HCC detection; VI, vascular invasion; LNM, lymph node metastasis; BDI, bile duct invasion; AUC, area under the curve. ***P<0.001. In (A), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively. In (C), the error bars represent the 95% confidence interval.
Supplementary Figure 12.
Establishment of HCC prognosis prediction model based on ccf-mtDNA fragmentomic features. The LASSO-Cox regression model was constructed in the training HCC cohort. The tuning parameter (Lambda) was calculated based on the partial likelihood deviance with five-fold cross validation. An optimal log Lambda value was shown by the vertical black dot-lines in the plots. The 20 features were identified according to the best fit profile in the training HCC cohort and the HPP score of each HCC patient was obtained. HCC, hepatocellular carcinoma; mtDNA Cap-seq, mtDNA capture based sequencing; LASSO-Cox, least absolute shrinkage and selection operator-Cox proportional; HPP, HCC prognosis prediction.
Supplementary Figure 13.
Performance of HPP score in predicting RFS and DFS in patients with HCC. The Kaplan–Meier survival curves were used to compare the RFS in HCC patients with an HPP score ≤0.65 (blue) with those with an HPP score >0.65 (green) based on the HCC prognosis prediction model in both the training HCC cohort (A) and validation HCC cohort (B). The Kaplan–Meier survival curves were used to compare the DFS in HCC patients with an HPP score ≤0.65 (blue) with those with an HPP score >0.65 (green) based on the HCC prognosis prediction model in both the training HCC cohort (C) and validation HCC cohort (D). HCC, hepatocellular carcinoma; HPP, HCC prognosis prediction; RFS, recurrence-free survival; DFS, disease-free survival.
Supplementary Figure 14.
Performance of combination of HPP score and clinical features in HCC prognosis prediction. The Kaplan–Meier survival curves were used to compare the overall survival in HCC patients with low risk of death (blue) versus high risk of death (green) based on the combination of HPP score and clinical features (BCLC stage, Child-Pugh class and AFP level) in both the training HCC cohort (A) and validation HCC cohort (C). The ROC curve evaluating the overall performance for 1-year, 2-year and 3-year survival predicted by combination of HPP score and clinical features in both the training HCC cohort (B) and validation HCC cohort (D). HCC, hepatocellular carcinoma; HPP, HCC prognosis prediction; AUC, area under the curve; ROC, receiver operating characteristic; BCLC, Barcelona Clinic Liver Cancer.
Figure 1.Study design. Participants from Xijing Hospital (n=480) including 210 HBV-related HCC patients, 30 non-HBV-related HCC patients, 120 CHB/LC patients, and 120 healthy controls were divided into training and internal validation cohorts for HCC detection model. Participants from Tangdu Hospital (n=450), which included 210 HBV-related HCC, 120 CHB/LC and 120 HC, and those from Changhai Hospital (n=238), which included 121 HBV-related HCC, 61 CHB/LC and 56 HC, were used as external validation cohorts to evaluate model performance. The cfDNA was extracted from plasma samples and used to construct sequencing library, followed by targeted enrichment of ccf-mtDNA and NGS sequencing. For each sample, ccf-mtDNA fragmentomic features including fragment size, fragmentation profile, 5′ end base preference, 5′ end base motifs and 5′ end MDS were analyzed. Finally, random forest algorithm and LASSO-Cox method were respectively used to establish the HCC detection model and HCC prognosis prediction model. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; HCC, hepatocellular carcinoma; cfDNA, cell-free DNA; ccf-mtDNA, circulating cell-free mtDNA; MDS, motif diversity score; LASSO-Cox, least absolute shrinkage and selection operator-Cox proportional.
Figure 2.Aberrant ccf-mtDNA fragmentomic features in patients with HCC. Comparison of (A) fragment size distribution, (B) the proportion of short and long fragments, (C) ccf-mtDNA fragmentation profiles, (D) the correlation to ccf-mtDNA fragmentation profile of median HC, (E) the number of total peak, (F) the number of new peak, (G, H) 5′ end base preference, (I) 5′ end MDS among HC (n=296), CHB/LC (n=301), EHCC (n=342) and LHCC (n=199) groups. The number of total peaks was calculated based on the fragmentation profile and new peaks were defined based on the comparison of peak location with the median HC profile. 5’ end base preference was calculated based on the proportion of 5′ end base and the base composition of the mitochondria reference genome. 5′ end MDS was calculated based on the proportion of 254 4-mer end motifs. HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; EHCC, early-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 and A); LHCC, late-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage B to D); MDS, motif diversity score; ns, not significant. *P<0.05; **P<0.01; ***P<0.001. In (B) and (D–I), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Figure 3.Performance of HCC detection model in validation cohorts. Comparison of HD score among the four groups in internal validation cohort (A), external validation cohorts 1(D) and 2 (G). The ROC curves evaluating the overall performance of HCC detection model for distinguishing EHCC and LHCC from non-HCC, CHB/LC and HC in internal validation cohort (B, C), external validation cohorts 1 (E, F) and 2 (H, I). HC, healthy control; CHB, chronic hepatitis B; LC, liver cirrhosis; EHCC, early-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 and A); LHCC, late-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage B to D); AUC, area under the curve; ROC, receiver operating characteristic; ns, not significant. *P<0.05; ***P<0.001. In (A, D) and (G), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively.
Figure 4.Performance of HD model, AFP level and mtDNA copy number for HCC detection. The ROC curves evaluating the performance of HD score, AFP or the combination of HD score and AFP in distinguishing EHCC and LHCC patients from CHB/LC in external validation cohorts 1 (A and C) and 2 (B and D). The ROC curves evaluating the performance of HD model in distinguishing HCC patients with a low AFP level (≤20 ng/mL, AFP-negative HCC) from CHB/LC in external validation cohorts 1 (E) and 2 (F). The ROC curves evaluating the performance of HD model in distinguishing HCC patients with a high AFP level (>20 ng/mL, AFP-positive HCC) from CHB/LC in external validation cohorts 1 (G) and 2 (H). The ROC curves evaluating the performance of mtDNA CN and the combination of HD score and mtDNA CN in distinguishing EHCC (I) and LHCC (J) patients from CHB/LC in external validation cohorts 1 and 2. The ROC curves evaluating the performance of mtDNA CN and the combination of HD score and mtDNA CN in distinguishing EHCC (K) and LHCC (L) patients from HC in external validation cohorts 1 and 2. EHCC, early stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 and A); LHCC, late stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage B to D); HD model, HCC detection model; AFP, alpha-fetoprotein; HCC, hepatocellular carcinoma; CHB, chronic hepatitis B; LC, liver cirrhosis; AUC, area under the curve; ROC, receiver operating characteristic; CN, copy number.
Figure 5.Performance of HD model in different HCC subgroups. (A) Comparison of HD score among BCLC 0, A, B, and C stages and between Child-Pugh class A and B, different number of tumor site and tumor size. (B) The ROC curves evaluating the performance of HD model in distinguishing BCLC 0, A, B, and C stages HCC patients, or HCC patients with Child-Pugh class A and B, and HCC patients with different numbers of tumor site or tumor size from CHB/LC. (C) Sensitivity of HD model in different HCC subgroups. BCLC, Barcelona Clinic Liver Cancer; HCC, hepatocellular carcinoma; HD, HCC detection; AUC, area under the curve. *P<0.05; **P<0.01; ***P<0.001. In (A), center line indicates the median, and lower and upper hinges represent the 25th and 75th percentiles, respectively. In (C), the error bars represent the 95% confidence interval.
Figure 6.Performance of HCC prognosis prediction model. The Kaplan–Meier survival curves were used to compare the overall survival in HCC patients with an HPP score ≤0.65 (blue) with those with an HPP score >0.65 (green) based on the HCC prognosis prediction model in both the training HCC cohort (A) and validation HCC cohort (C). The ROC curve evaluating the performance for 1-year, 2-year and 3-year survival predicted by HCC prognosis prediction model in both the training HCC cohort (B) and validation HCC cohort (D). (E) Overall survival analysis for HCC patients with an HPP score ≤0.65 (n=200) and those with an HPP score >0.65 (n=188) in subgroup stratified by the clinical/demographic characteristics. HCC, hepatocellular carcinoma; HPP, HCC prognosis prediction; AUC, area under the curve; ROC, receiver operating characteristic; HR, hazard ratio; BCLC, Barcelona Clinic Liver Cancer; TBIL, total bilirubin; ALB, albumin; AFP, alpha-fetoprotein; CI, confidence interval.
Abbreviations
Barcelona Clinic Liver Cancer
circulating cell free mtDNA
fragment size distribution
revised Cambridge reference sequence
REFERENCES
- 1. Global Burden of Disease Cancer Collaboration, Fitzmaurice C, Allen C, Barber RM, Barregard L, Bhutta ZA, Brenner H, et al. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disabilityadjusted life-years for 32 cancer groups, 1990 to 2015: a systematic analysis for the Global Burden of Disease Study. JAMA Oncol 2017;3:524-548.
- 2. Chen W, Zheng R, Baade PD, Zhang S, Zeng H, Bray F, et al. Cancer statistics in China, 2015. CA Cancer J Clin 2016;66:115-132.
- 3. Pinyopornpanish K, Khoudari G, Saleh MA, Angkurawaranon C, Pinyopornpanish K, Mansoor E, et al. Hepatocellular carcinoma in nonalcoholic fatty liver disease with or without cirrhosis: a population-based study. BMC Gastroenterol 2021;21:394.
- 4. De Battista D, Zamboni F, Gerstein H, Sato S, Markowitz TE, Lack J, Engle RE, et al. Molecular signature and immune landscape of HCV-associated hepatocellular carcinoma (HCC): differences and similarities with HBV-HCC. J Hepatocell Carcinoma 2021;8:1399-1413.
- 5. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin 2021;71:7-33. Erratum in: CA Cancer J Clin 2021;71:359.
- 6. Gupta S, Bent S, Kohlwes J. Test characteristics of alpha-fetoprotein for detecting hepatocellular carcinoma in patients with hepatitis C. A systematic review and critical analysis. Ann Intern Med 2003;139:46-50.
- 7. Ye Q, Ling S, Zheng S, Xu X. Liquid biopsy in hepatocellular carcinoma: circulating tumor cells and circulating tumor DNA. Mol Cancer 2019;18:114.
- 8. Benesova L, Belsanova B, Suchanek S, Kopeckova M, Minarikova P, Lipska L, et al. Mutation-based detection and monitoring of cell-free tumor DNA in peripheral blood of cancer patients. Anal Biochem 2013;433:227-234.
- 9. Cai J, Chen L, Zhang Z, Zhang X, Lu X, Liu W, et al. Genomewide mapping of 5-hydroxymethylcytosines in circulating cell-free DNA as a non-invasive approach for early detection of hepatocellular carcinoma. Gut 2019;68:2195-2205.
- 10. Xu RH, Wei W, Krawczyk M, Wang W, Luo H, Flagg K, et al. Circulating tumour DNA methylation markers for diagnosis and prognosis of hepatocellular carcinoma. Nat Mater 2017;16:1155-1161.
- 11. Wang Y, Zhou K, Wang X, Liu Y, Guo D, Bian Z, et al. Multiple-level copy number variations in cell-free DNA for prognostic prediction of HCC with radical treatments. Cancer Sci 2021;112:4772-4784.
- 12. Foda ZH, Annapragada AV, Boyapati K, Bruhm DC, Vulpescu NA, Medina JE, et al. Detecting liver cancer using cell-free DNA fragmentomes. Cancer Discov 2023;13:616-631.
- 13. Zhang X, Wang Z, Tang W, Wang X, Liu R, Bao H, et al. Ultrasensitive and affordable assay for early detection of primary liver cancer using plasma cell-free DNA fragmentomics. Hepatology 2022;76:317-329.
- 14. Li L, Hann HW, Wan S, Hann RS, Wang C, Lai Y, et al. Cell-free circulating mitochondrial DNA content and risk of hepatocellular carcinoma in patients with chronic HBV infection. Sci Rep 2016;6:23992.
- 15. Weerts MJA, Timmermans EC, van de Stolpe A, Vossen RHAM, Anvar SY, Foekens JA, et al. Tumor-specific mitochondrial DNA variants are rarely detected in cell-free DNA. Neoplasia 2018;20:687-696.
- 16. Johnson P, Zhou Q, Dao DY, Lo YMD. Circulating biomarkers in the diagnosis and management of hepatocellular carcinoma. Nat Rev Gastroenterol Hepatol 2022;19:670-681.
- 17. Yin C, Li DY, Guo X, Cao HY, Chen YB, Zhou F, et al. NGS-based profiling reveals a critical contributing role of somatic D-loop mtDNA mutations in HBV-related hepatocarcinogenesis. Ann Oncol 2019;30:953-962.
- 18. Liu Y, Zhou K, Guo S, Wang Y, Ji X, Yuan Q, et al. NGSbased accurate and efficient detection of circulating cell-free mitochondrial DNA in cancer patients. Mol Ther Nucleic Acids 2021;23:657-666.
- 19. Zhou K, Liu Y, Yuan Q, Lai D, Guo S, Wang Z, et al. Next-generation sequencing-based analysis of urine cell-free mtDNA reveals aberrant fragmentation and mutation profile in cancer patients. Clin Chem 2022;68:561-573.
- 20. van der Pol Y, Moldovan N, Ramaker J, Bootsma S, Lenos KJ, Vermeulen L, et al. The landscape of cell-free mitochondrial DNA in liquid biopsy for cancer detection. Genome Biol 2023;24:229.
- 21. Zhou K, Mo Q, Guo S, Liu Y, Yin C, Ji X, et al. A novel nextgeneration sequencing-based approach for concurrent detection of mitochondrial DNA copy number and mutation. J Mol Diagn 2020;22:1408-1418.
- 22. Marrero JA, Kulik LM, Sirlin CB, Zhu AX, Finn RS, Abecassis MM, et al. Diagnosis, staging, and management of hepatocellular carcinoma: 2018 Practice Guidance by the American Association for the Study of Liver Diseases. Hepatology 2018;68:723-750.
- 23. Tao K, Bian Z, Zhang Q, Guo X, Yin C, Wang Y, et al. Machine learning-based genome-wide interrogation of somatic copy number aberrations in circulating tumor DNA for early detection of hepatocellular carcinoma. EBioMedicine 2020;56:102811.
- 24. Chen S, Zhou Y, Chen Y, Gu J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 2018;34:i884-i890.
- 25. Lima PH, Fan B, Bérubé J, Cerny M, Olivié D, Giard JM, et al. Cost-utility analysis of imaging for surveillance and diagnosis of hepatocellular carcinoma. AJR Am J Roentgenol 2019;213:17-25.
- 26. Hessheimer AJ, Vargas-Martínez AM, Trapero-Bertrán M, Navasa M, Fondevila C. Posttransplant hepatocellular carcinoma surveillance: a cost-effectiveness and cost-utility analysis. Ann Surg 2023;277:e359-e365.
- 27. Mouliere F, Chandrananda D, Piskorz AM, Moore EK, Morris J, Ahlborn LB, et al. Enhanced detection of circulating tumor DNA by fragment size analysis. Sci Transl Med 2018;10:eaat4921.
- 28. Han DSC, Ni M, Chan RWY, Chan VWH, Lui KO, Chiu RWK, et al. The biology of cell-free DNA fragmentation and the roles of DNASE1, DNASE1L3, and DFFB. Am J Hum Genet 2020;106:202-214.
- 29. Zhou Z, Ma ML, Chan RWY, Lam WKJ, Peng W, Gai W, et al. Fragmentation landscape of cell-free DNA revealed by deconvolutional analysis of end motifs. Proc Natl Acad Sci U S A 2023;120:e2220982120.
- 30. Huang Q, Wu D, Zhao J, Yan Z, Chen L, Guo S, et al. TFAM loss induces nuclear actin assembly upon mDia2 malonylation to promote liver cancer metastasis. EMBO J 2022;41:e110324.
- 31. Choi JY, Jung SW, Kim HY, Kim M, Kim Y, Kim DG, et al. Diagnostic value of AFP-L3 and PIVKA-II in hepatocellular carcinoma according to total-AFP. World J Gastroenterol 2013;19:339-346.
- 32. Feng H, Li B, Li Z, Wei Q, Ren L. PIVKA-II serves as a potential biomarker that complements AFP for the diagnosis of hepatocellular carcinoma. BMC Cancer 2021;21:401.
