Cell-free DNA in hepatocellular carcinoma: Biology to treatment response

Abstract

Hepatocellular carcinoma (HCC) is frequently diagnosed without tissue confirmation, and single-site biopsies often fail to capture intratumor heterogeneity, underscoring the need for liquid biopsy approaches. Circulating cell-free DNA (cfDNA) and circulating tumor DNA (ctDNA) provide access to genetic, epigenetic, copy-number, and fragmentomic biomarkers that can support diagnosis, prognostication, and treatment monitoring. Current monitoring modalities, such as alpha-fetoprotein and radiographic assessment, lack sensitivity and typically lag behind molecular changes, whereas cfDNA/ctDNA-based analyses offer the potential for earlier and more accurate prediction of treatment response and disease dynamics. Literature review (PubMed, Google Scholar, 2015-2025) evaluated cfDNA biomarkers for predicting and monitoring treatment response across immune checkpoint inhibitors, tyrosine kinase inhibitors, transarterial chemoembolization, and radiotherapy in HCC. ctDNA profiling detects recurrent mutations in TERT, TP53, CTNNB1, and phosphatidylinositol 3-kinase/mammalian target of rapamycin pathways with variable prognostic associations across therapies. Limited evidence suggests phosphatidylinositol 3-kinase/mammalian target of rapamycin alterations may predict tyrosine kinase inhibitor resistance. Serial ctDNA monitoring shows promise for early response assessment, with variant allele frequency changes correlating with outcomes in small cohorts, though superiority to alpha-fetoprotein remains inconsistently demonstrated. Baseline cfDNA burden associates with survival across multiple treatment modalities. Copy-number variation dynamics correlate with transarterial chemoembolization response. Postoperative ctDNA detection enables recurrence risk stratification. Fragmentomic and methylation signatures demonstrate high diagnostic accuracy but lack treatment-response validation. cfDNA shows potential for treatment monitoring in HCC but faces critical limitations: Low analytical sensitivity, platform heterogeneity, absence of validated thresholds, and limited prospective evidence. Standardized multicenter trials demonstrating clinical utility are essential before routine implementation.

Key Words: Cell-free DNA; Circulating tumor DNA; Hepatocellular carcinoma; Liquid biopsy; Methylation; Fragmentomics; Copy number variation; Chromosomal instability; Minimal residual disease; Treatment response

Core Tip: The study evaluated circulating tumor DNA as a potential indicator of treatment response in hepatocellular carcinoma. Quantitative aspects of cell-free DNA, such as concentration and variant allele frequency, were analyzed in conjunction with qualitative genomic and epigenetic features, including copy-number variation and methylation-associated changes, to assess tumor dynamics during therapy. The findings indicate that cell-free DNA profiling captures both quantitative changes and molecular alterations, including genomic and epigenetic variation, that may reflect treatment-related tumor dynamics in hepatocellular carcinoma.

INTRODUCTION

Hepatocellular carcinoma (HCC) accounts for approximately 90% of primary liver cancers and represents the third leading cause of cancer-related deaths globally[1,2]. A major challenge in HCC management is that the diagnosis is frequently made without biopsy, leaving clinicians with tumor markers and radiology imaging. Even when tissue is available, the biological complexity of HCC presents a problem, such as intratumor and interlesion heterogeneity, which can render a single biopsy insufficient to capture the full mutational genetic landscape. Also, the available biomarkers like alpha-fetoprotein (AFP), AFP-L3 and protein induced by vitamin K absence or antagonist-II are raised in 50%-60% of positive cases, making the rest of the cases difficult to diagnose even with tumor biomarkers, creating a lacuna to search for better biomarkers. This is where liquid biopsy emerges as a promising alternative[1]. Cell-free DNA (cfDNA) and particularly its tumor-derived fraction, circulating tumor DNA (ctDNA), provides a window into tumor biology that is both non-invasive and dynamic. ctDNA offers a non-invasive alternative that can detect clonal and subclonal variants, capture mutations from multiple nodules, and reveal alterations missed by single biopsies. Serial ctDNA sampling can further monitor tumor evolution over time, highlighting its potential as a complementary tool for genetic profiling and treatment monitoring in HCC[3]. cfDNA are fragments of DNA released into the bloodstream from necrotic or apoptotic cells, including tumor cells, and carry tumor-specific genetic and epigenetic alterations that can provide real-time insights into the biology of the tumor (Figure 1).

Figure 1 Circulating cell-free DNA as a tool for cancer diagnostics and monitoring response. The left panel categorises cell-free DNA (cfDNA) assays into qualitative and quantitative domains: Epigenetic markers (DNA methylation, 5-hydroxymethylcytosine), genetic alterations (chromosomal instability, tumor mutation burden, copy number variation), and fragmentomics, with cfDNA level representing a quantitative burden metric. The central wheel illustrates clinical utilities of cfDNA, including noninvasive sampling, real time monitoring, quantitative and prognostic value, minimal residual disease detection, feasibility of repeated measurements, reflection of tumor heterogeneity, early detection of mutations, and identification of resistance mutations. The lower panel depicts sources and forms of tumour-derived DNA: Chromosomal changes, nucleotide level variants, copy number alterations, and structural variants/fusions, converging as circulating cfDNA released into blood; arrows indicate sampling from blood vessels and its application to hepatocellular carcinoma. Icons emphasise longitudinal use across therapy: Serial cfDNA enables dynamic response assessment and resistance surveillance without repeated tissue biopsy, aligning with use in hepatocellular carcinoma management pathways. cfDNA: Cell-free DNA; CNV: Copy number variation; TMB: Tumor mutation burden; HCC: Hepatocellular carcinoma.

Evidence supporting this approach is accumulating. ctDNA has been shown to detect mutations in key oncogenic drivers, including TP53, CTNNB1, and genes in the phosphatidylinositol 3-kinase (PI3K)/mammalian target of rapamycin (mTOR) pathway. Such findings are not merely descriptive; they hold potential implications for both prognosis and therapy selection. In one example, evaluating the CTNNB1 p.T41A mutation in paired tissue and plasma samples increased detection from 8%-9% with either method alone to 13.5% when combined, underscoring the ability of ctDNA to capture heterogeneity and clonal evolution that tissue biopsies may miss[4]. Across different treatments, the clinical need for predictive biomarkers is especially evident in the era of immunotherapy. Although immunotherapy has revolutionized cancer care, its benefit in HCC remains limited: Only about one in five patients achieves an objective radiological response[5]. This uneven response underscores the urgency of identifying reliable biomarkers to guide therapy. Yet, traditional tissue-based biomarkers face familiar obstacles in liver cancers: Invasive sampling procedures, intratumor variability, and restricted accessibility. Liquid biopsy offers a potential way forward[6-11] (Table 1). By analyzing ctDNA, clinicians may be able not only to select the most appropriate treatments but also to monitor response to therapy in real time. This is the central focus of the present review. We aim to evaluate the capacity of ctDNA to detect clinically relevant mutations and epigenetic changes in HCC, to establish their prevalence in advanced disease, and to examine how these alterations may correlate with patient outcomes (Figure 1).

Table 1 Diagnostic performance of cell-free DNA-based biomarkers for hepatocellular carcinoma detection.

Ref.	Patients (HCC cohort)	cfDNA characteristics	Sensitivity	Specificity	AUROC
Xu et al[6], 2017	1098 (training + validation)	HCC bisulphite methylation marker	86%	94%	0.97 (0.96-0.98)
Cai et al[7], 2019	1204 tests	5-hMC signatures	79%	91%	0.92 (0.88-0.92)
Cai et al[8], 2021	135 (training + tests)	5-hMC, AFP and PIVKA2	79%	91%	0.92 (0.88-0.92)
Chen et al[9], 2021	350 (training + validation)	HIFI score = 4 cfDNA genomic features	96%	95%	0.995 (0.99-1.000)
Luo et al[10], 2022	187 (training + validation)	cfDNA methylation	86%	98%	0.98 (0.97-0.99)
Pös et al[11], 2020	159 (training)	cfDNA fragmentomic profiles	97%	99%	0.995
Kim et al[22], 2023	313 tests	Methylation markers of RNF135 and LDHB	57%	94%	0.80 (0.76-0.83)
Guo et al[41], 2024	498 (training + validation)	DMRs HepaAiQ: 20 best DMRs	86%	92%	0.94 (0.93-0.96)
Chen et al[44], 2024	510 (stage1)	PreCar score = 5 cfDNA genomic features	94%	95%	NA

Summarizes index tests, cohort structures (training, validation, and test where applicable), and corresponding sensitivity, specificity, and area under the receiver operating characteristic curve as reported by the original studies, enabling side-by-side appraisal of cell-free DNA methylation, hydroxymethylation, fragmentomics, and integrated multi-feature models for hepatocellular carcinoma detection in at-risk populations and mixed clinical cohorts. HCC: Hepatocellular carcinoma; cfDNA: Cell-free DNA; AUROC: Area under the receiver operating characteristic curve; 5-hmC: 5-hydroxymethylcytosine motif; AFP: Alpha-fetoprotein; PIVKA2: Protein induced by vitamin K absence or antagonist-II; HIFI: Integrated model of 5-hydroxymethylcytosine motif/fragmentation/nucleosome footprint; RNF135: Ring finger protein 135; LDHB: Lactate dehydrogenase B; DMR: Differentially methylated region; HepaAiQ: Hepatic artificial intelligence quantitative; NA: Not available.

LITERATURE SEARCH

A comprehensive literature search was conducted using PubMed and Google Scholar to identify relevant studies on cfDNA ctDNA, methylation markers, fragmentomics, and other liquid biopsy approaches in HCC. The search covered publications from January 2015 to September 2025. Keywords included combinations of “hepatocellular carcinoma or HCC”, “cell-free DNA”, “circulating tumor DNA”, “methylation”, “fragmentomics”, “liquid biopsy”, “biomarkers”, and “treatment response”. Boolean operators were applied to refine search sensitivity and specificity. Studies were included if they investigated cfDNA/ctDNA or related molecular biomarkers in patients with HCC and reported diagnostic, prognostic, predictive, or surveillance outcomes. Relevant peer-reviewed English-language articles, including observational studies, clinical trials, methodological papers, and translational research, were considered based on their relevance and contribution to the current understanding of the field.

CFDNA: BIOLOGY AND ANALYTICAL APPROACHES

cfDNA are short fragments of extracellular DNA that circulate freely in blood plasma and other body fluids. These fragments, typically 120-220 base pairs in length, originate primarily from apoptotic and necrotic cells and reflect the nucleosomal organisation of chromatin. cfDNA is highly dynamic, with a half-life of approximately 15 minutes to 2.5 hours, making it a sensitive indicator of cellular turnover and tissue injury. The release of cfDNA occurs through several biological pathways, including apoptosis, necrosis, and NETosis. During apoptosis, nucleases such as DNase I, DNase1 L3, and caspase-activated DNase fragment chromatin into nucleosomal units of about 147 base pairs, producing the characteristic ladder-like pattern of apoptotic cfDNA. Necrotic and inflammatory cell death also contribute to cfDNA release, while some viable cells actively secrete DNA within extracellular vesicles or lipoprotein complexes[11,12]. In healthy individuals, cfDNA mainly derives from hematopoietic cells such as neutrophils and lymphocytes. In contrast, under pathological conditions such as cancer, infection, trauma, or pregnancy, cfDNA levels increase, and tissue-specific patterns emerge. For example, ctDNA carries genetic and epigenetic features of malignant cells, including mutations, copy-number variations, and methylation profiles[13].

The field of cfDNA fragmentomics explores additional characteristics such as fragment length, end motifs, and methylation landscapes to infer tissue of origin and disease mechanisms. Methodologically, cfDNA analysis involves sample collection, isolation, quantification, and molecular characterization. Plasma is preferred to minimize contamination from lysed blood cells[14]. Extraction is commonly performed using silica- or magnetic bead-based kits optimized for small DNA fragments, followed by fluorometric, quantitative polymerase chain reaction (PCR), or sequencing-based quantification[15]. Advanced approaches such as next-generation sequencing and digital PCR enable high-resolution mutation profiling. Integrating bioinformatics and machine learning tools like cfSort has further enhanced tissue-specific ctDNA analysis[12]. ctDNA is a powerful, non-invasive biomarker in liquid biopsy applications, including cancer detection, prenatal screening, and transplant monitoring, representing a key frontier in precision medicine.

CTDNA MUTATIONAL LANDSCAPE IN PROGNOSIS AND TREATMENT RESPONSE IN HCC

Plasma cfDNA comprises tumor-derived ctDNA alongside abundant wild-type DNA from non-malignant cells, particularly hepatocytes in HCC. Tumor-specific endpoints, such as variant allele frequency (VAF) dynamics, MRD detection, and clonal selection during therapy, align more precisely with ctDNA clearance, as total cfDNA levels are frequently confounded by concurrent liver inflammation, necrosis, or regenerative processes that elevate wild-type fractions independently of tumor burden. Accordingly, ctDNA enables serial genetic monitoring and has emerged as a non-invasive tool to profile clinically relevant mutations in HCC, offering a framework to investigate treatment response and resistance as they evolve. Importantly, ctDNA genetics carries significant prognostic and predictive value. In patients treated with immune checkpoint inhibitors (ICIs) (nivolumab), ctDNA profiling revealed that CTNNB1 mutations were associated with improved progression-free survival (PFS), while alterations in PIK3CA, BRCA1, KIT, and CCND1 predicted poorer outcomes[16]. In a separate cohort of advanced HCC patients, the most frequent ctDNA-detected mutations included TERT promoter, TP53, CTNNB1, PTEN, AXIN1, ARID2, KMT2D, and TSC2, with PI3K/mTOR pathway alterations linked to shorter PFS following tyrosine kinase inhibitor therapy, but not immunotherapy[17].

Extending these insights to the curative setting, a prospective study demonstrated that preoperative ctDNA mutational burden and a defined high-risk gene set (APC, ARID1A, CDKN2A, FAT1, LRP1B, MAP3K1, PREX2, TERT, TP53) predicted early recurrence after liver resection, and when combined with TNM stage, enabled accurate recurrence risk stratification. Moreover, specific ctDNA variants (FAT1 or LRP1B without TP53) were linked to worse outcomes on lenvatinib plus immunotherapy, and ctDNA status reflected differences in immune infiltration[18]. Consistent with these findings, the SORAMIC trial demonstrated that post-operative cfDNA levels correlated with distant metastases and, at later time points during sorafenib therapy, with overall survival (OS), supporting ctDNA as a dynamic biomarker for monitoring treatment response. Next generation sequencing (NGS) analysis revealed 28 variants across 13 patients, with no identical mutational profiles, highlighting HCC’s pronounced heterogeneity. Notably, CYP2B6*6 variants correlated with poorer survival, likely due to altered drug metabolism. BAX mutations were associated with portal vein invasion, which suggests therapy resistance, and HNF1A, the most frequently altered gene, may function as a tumor suppressor and a potential target for personalized interventions. Interestingly, a striking observation was the dynamic shifts in mutation status over time (WT → M, M → WT, or double shifts), which indicates clonal evolution under therapeutic pressure. The appearance or disappearance of mutations in ctDNA reflects the selective advantage or sensitivity of specific clones, underscoring the unique value of liquid biopsy in capturing tumor evolution and informing treatment strategies beyond the limitations of single-site tissue sampling[19].

Similarly, in advanced HCC treated with lenvatinib, monitoring ctDNA via VAF provided predictive insights into treatment response. A reduction in VAF mean after 4 weeks of therapy was associated with longer PFS and correlated with changes in tumor burden, while baseline mutations - including PI3K/mTOR pathway alterations- did not predict response. Discrepancies between AFP kinetics and mean VAF changes in several patients reinforced ctDNA as a more sensitive and dynamic marker. Furthermore, emerging mutations during LEN treatment in genes such as TP53, ESR1, DDR2, ATM, AR, and NFE2 L2 may reflect mechanisms of intrinsic or acquired resistance. Importantly, CTNNB1 mutations did not influence LEN efficacy, suggesting that patients harboring β-catenin pathway alterations might benefit more from LEN than ICIs[20].

Recent evidence suggests that pretreatment cfDNA and ctDNA-based genomic profiling may help predict the efficacy and prognosis of combination immunotherapy [anti-programmed death-ligand 1 (PD-L1) plus anti-vascular endothelial growth factor] in unresectable HCC. Patients with higher cfDNA levels showed lower response rates and shorter PFS and OS, highlighting cfDNA quantification as a simple, noninvasive prognostic biomarker. ctDNA profiling detected recurrent HCC driver mutations, including TERT, TP53, and CTNNB1. Notably, TERT mutations were associated with poorer OS, while CTNNB1 and other mutations did not significantly affect treatment response[21].

CHROMOSOMAL INSTABILITY AND COPY NUMBER VARIATIONS IN HCC TREATMENT RESPONSE

Cancer cells are characterized by genomic copy number instability because of chromosomal instability (CI) and tumor-specific point mutations, a feature commonly found in various solid cancers. Genomic instability and chromosomal instability (CIN) score are proportional, calculated by performing whole-genome sequencing. A few studies have shown association of CI with tumor progression and metastasis; for instance, a recent 2023 study focused on CI for the evaluation of response. They developed an I score, calculated from the plasma ctDNA as a surrogate marker for genomic instability. cfDNA (I-score) measured before radiotherapy (RT) is shown to be clinically useful for predicting tumor burden and treatment outcomes, especially in patients with AFP > 200. However, more studies in HCC are needed[22].

cfDNA can also be used to assess copy number variations (CNVs) as predictive biomarkers. One study evaluated CNVs for changes in tumor fraction between pre- and post-trans arterial chemoembolisation (TACE) plasma samples via low-depth whole-genome sequencing and found they predict PFS, with specific chromosomal alterations (e.g., 16q amplification) linked to poor TACE efficacy and driver genes such as NQO1 mediating resistance mechanisms[23]. Similar to CI, the studies are currently inadequate in this context as well which calls for further explorations.

QUANTITATIVE CFDNA LEVELS AS PREDICTIVE BIOMARKERS: DIAGNOSIS AND RESPONSE TO THERAPY

Quantification of cfDNA has been investigated as a potential biomarker for predicting treatment response in HCC, with recent studies supporting its possible predictive value. Patients with HCC consistently exhibit higher cfDNA concentrations compared to those with chronic hepatitis or healthy individuals, which suggests its utility as a disease-specific marker[24]. Accumulating evidence indicates that baseline cfDNA burden carries prognostic value. In advanced HCC treated with ICIs, elevated cfDNA and ctDNA levels were linked to reduced survival and weaker therapeutic responses, while patients with lower levels achieved more favorable outcomes[25]. Similar observations have been made in patients receiving combination immunotherapy with anti-PD-L1 and anti-vascular endothelial growth factor agents, where higher cfDNA concentrations were associated with lower response rates and shorter PFS and OS[21]. More importantly, serial measurements capture treatment effects in real time: Rising cfDNA levels typically indicated progressive disease, while stable or declining levels were associated with disease control or response[26].

The predictive potential of cfDNA dynamics is also illustrated in immunotherapy-refractory HCC. In the phase 2 LIVERTI trial, patients receiving domvanalimab (anti-T-cell immunoreceptor with Ig and ITIM domains) plus zimberelimab (anti-PD-1) demonstrated a confirmed overall response rate of 17.2% and a median PFS of 4.4 months. Notably, analysis of ctDNA revealed that dynamic changes in ctDNA levels reflected treatment response, suggesting its potential as a pharmacodynamic biomarker. While the trial did not meet its primary endpoint for efficacy, the study highlights that ctDNA profiling can provide real-time insights into therapeutic effects and may guide treatment decisions, particularly in the context of immune checkpoint resistance[27].

More refined analyses further strengthen this paradigm. In a cohort of 28 patients undergoing immune-targeted therapy, reductions in mean VAF after two treatment cycles predicted longer PFS with higher sensitivity and specificity than AFP monitoring, while baseline mutation status was not predictive[28]. Similarly, quantification of telomerase reverse transcriptase promoter-mutant cfDNA using digital PCR revealed that mutant alleles correlated with tumor burden in patients undergoing TACE or tyrosine kinase inhibitor (TKI) treatment. Early peaks in mutant cfDNA within days of therapy initiation reflected tumor necrosis and predicted PFS, whereas wild-type cfDNA levels were more indicative of liver injury. Although short-term declines in mutant cfDNA did not consistently predict OS, likely due to subsequent treatments or resistance, these findings further highlight the value of cfDNA for real-time, non-invasive monitoring[29].

cfDNA levels could also predict treatment response and prognosis in HCC patients receiving radiotherapy. Pre- and post-treatment cfDNA were analyzed in 55 patients, revealing that higher pre-RT cfDNA correlated with larger tumors and advanced disease, while persistently elevated post-RT cfDNA independently predicted poor intrahepatic control. Importantly, post-RT cfDNA, rather than pre-RT levels, was most predictive of treatment outcomes, highlighting its potential as an early, noninvasive biomarker of RT efficacy[30]. Taken together, these studies do not fail to demonstrate that quantitative cfDNA assessment, whether through total cfDNA burden, dynamic changes during therapy, or mutation-specific analysis, indeed offers a sensitive and dynamic biomarker for predicting outcomes and monitoring treatment efficacy in HCC (Table 2)[5,18,23,29,31-33].

Table 2 Summary of cell-free DNA/circulating tumor DNA biomarker studies in hepatocellular carcinoma evaluating prognostic and therapeutic response.

Ref.	Patients	Therapy	Biomarker	Methodology	Prognostic impact/therapy response	Strength
Wang et al[31], 2021	117 HBV-HCC patients	Radical treatments (resection, ablation)	cfDNA CNVs (TFx, P-score, S-score)	Low-coverage WGS; CNV profiling at genome, chromosomal-arm, and 1-Mb bin levels	High scores (TFx ≥ 0.02, P ≥ 0.74, S ≥ 0.04) predicted worse OS and RFS	Multi-level cfDNA CNV profiling; bin-level CNV shown as superior prognostic marker
Fu et al[18], 2022	258 HCC patients undergoing curative liver resection	Surgery (curative resection)	ctDNA mutation profiles (high-risk gene set)	ctDNA mutation detection; RNA sequencing for tumor immune infiltration	Higher number of mutant genes linked to early recurrence (HR = 2.2); high-risk mutations worsened RFS (HR = 13)	Largest prospective study on preoperative ctDNA predicting recurrence; ctDNA-guided risk stratification
Dong et al[23], 2022	64 HCC patients	TACE	Plasma cfDNA; CNV profiling; TFx	Low-depth WGS for CNV detection; TFx quantification; plasma centrifugation protocol	Changes in TFx pre- vs post-TACE predict TACE response; CNVs and NQO1 amplification associated with efficacy; lipiodol deposition correlated with CNVs	cfDNA CNV profiling via low-depth WGS provides noninvasive, cost-effective real-time monitoring of tumor burden and TACE response
Muraoka et al[29], 2021	32 TACE patients, 35 TKI patients with available plasma (advanced HCC)	TACE (epirubicin/cisplatin) and TKI (sorafenib or lenvatinib)	Plasma cfDNA (hTERT promoter mutations)	dPCR quantification of hTERT promoter mutant and wild-type cfDNA	Changes in mutant cfDNA during TACE/TKI reflect therapy response; wild-type cfDNA reflects liver damage	First study to show short-term cfDNA dynamics for monitoring therapy response; noninvasive marker of tumor necrosis
Nakatsuka et al[32], 2021	100 HCC patients	TACE (32), RFA (33), MTAs (35) (lenvatinib, sorafenib, regorafenib)	Plasma cfDNA; TERT promoter mutations; post-treatment cfDNA used for ultra-deep sequencing	ddPCR for TERT; targeted ultra-deep sequencing (22000 × coverage)	cfDNA level post-treatment correlated with therapy response; TERT mutation alone not prognostic	Ultra-deep sequencing of post-treatment cfDNA; analysis of therapy response via liquid biopsy
Li et al[33], 2022	60 patients with primary HCC	Primary liver resection	TP53 mutations in exosomal cfDNA	exoDNA, TP53 mutation	High-frequency TP53 mutations in circulating exoDNA associated with poor prognosis, independent of other clinicopathologic features	80% of patients had detectable TP53 mutations in exoDNA; exoDNA considered more stable and robust than cfDNA
Campani et al[5], 2024	173 HCC patients and 56 chronic liver disease controls	Locoregional (percutaneous ablation, TACE) and systemic (atezolizumab-bevacizumab, others)	ctDNA mutations (TERT, TP53, CTNNB1, PIK3CA, NFE2 L2)	ddPCR and MiSeq sequencing	Persistence of ctDNA mutations under atezolizumab-bevacizumab associated with radiological progression; disappearance linked to response	Longitudinal design; paired tumour-plasma analysis; dynamic ctDNA assessment across therapies

Highlights various assay types, genomic targets, and their clinical correlations across surgical, locoregional, and systemic therapies. Tumor fraction, P-score, and S-score were derived from low-coverage whole-genome sequencing-based copy number variation profiling. cfDNA: Cell-free DNA; ctDNA: Circulating tumor DNA; exoDNA: Exosomal DNA; CNV: Copy number variation; TFx: Tumor fraction; OS: Overall survival; RFS: Recurrence-free survival; WGS: Whole-genome sequencing; ddPCR: Droplet digital polymerase chain reaction; dPCR: Digital polymerase chain reaction; RFA: Radiofrequency ablation; TACE: Transarterial chemoembolization; TKI: Tyrosine kinase inhibitor; MTA: Molecular-targeted agent; HBV: Hepatitis B virus; HCC: Hepatocellular carcinoma; CLD: Chronic liver disease; HR: Hazard ratio; NQO1: NAD(P)H quinone dehydrogenase 1; TERT: Telomerase reverse transcriptase; CTNNB1: Catenin beta 1; PIK3CA: Phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha; NFE2 L2: Nuclear factor erythroid 2-like 2; hTERT: Human telomerase reverse transcriptase.

CFDNA AND FRAGMENTOMICS: DIAGNOSIS IN HCC AND TREATMENT RESPONSE

cfDNA fragmentomics refers to the study of the patterns and properties of cfDNA fragments. Unlike sequence-based analyses such as mutation or methylation profiling, fragmentomics focuses on the physical and structural features of DNA fragments, including fragment size distribution, end motifs, nucleosome footprints, and fragment orientation and topology. Evidence from HCC supports the relevance of these features, for instance, in a study by Foda et al[34], fragmentation profiles were found to be consistent in individuals without cancer but highly variable in HCC patients. Interestingly, profiles from cirrhosis patients more closely resembled noncancer individuals than those with HCC, which suggests that fragmentomics patterns reflect disease state[34]. Given these observations, an important question arises: Whether cfDNA fragmentomics can be used to predict therapy response in HCC.

Chen et al[9] developed an integrated model of 5-hydroxymethylcytosine motif/fragmentation/nucleosome footprint score, which included 4 cfDNA genomic features: Nucleosome footprint, motif, 5-hydroxymethylcytosine motif (5-hmC), and fragmentation profiles. They showed in the training and validation cohort that sensitivity and specificity both reached 95% and the area under the receiver operating characteristic curve was 0.995. Another study in 2024 showed, PreCar score = 5 cfDNA genomic features: Nucleosome footprint, motif, 5-hmC, fragmentation profiles. They showed that PreCar score plus ultrasound improved sensitivity for early/very early HCC, with both sensitivity and specificity close to 95%[35]. Also, recent abstracts offer a window into ongoing developments. Salas et al[36] reported that long cfDNA fragments (> 1650 bp) predicted early progression in pan-cancer patients receiving ICIs, with higher quantities of long fragments associated with reduced likelihood of early progression (odds ratio = 0.37, P < 0.001)[36]. Similarly, Salas et al[36], in their SChISM study of 128 advanced cancer patients treated with ICIs, found that the ratio of long fragments (≥ 1650 bp) normalized by total concentration was significantly associated with non-early progression and longer PFS (areal under curve = 0.74). They also proposed a mechanistic model whereby immunotherapy-induced apoptosis releases short fragments while tumor necrosis produces longer fragments, providing biological insight into cfDNA fragmentome dynamics during treatment[37].

EPIGENETICS IN HCC: TUMOR BIOLOGY AND RESPONSE PREDICTION

cfDNA methylation

While genetic mutations have long been the focus of tumor-profiling, emerging evidence highlights those epigenetic alterations in cfDNA, such as DNA methylation patterns, can provide equally meaningful insights into tumor behavior and therapy response. In a recent pan-cancer phase II study, researchers applied cell-free methylated DNA immunoprecipitation sequencing-seq, a technique that profiles genome-wide methylation and cfDNA fragment length, to monitor patients receiving pembrolizumab. They found that early changes in cancer-specific methylation and fragment-length scores closely tracked ctDNA levels and independently predicted both overall and PFS, regardless of tumor type or traditional biomarkers like PD-L1 expression[38].

Another study demonstrates that cfDNA methylation profiling can serve as a non-invasive biomarker for HCC detection and monitoring. Using targeted massively parallel deep sequencing, methylation differences in VIM and fibulin-1 were consistently detected in cfDNA from HCC patients across two epidemiologically distinct cohorts (France and Thailand), reflecting tumor-associated epigenetic changes. The methylation patterns in cfDNA closely mirrored those in liver tissue, supporting the notion that plasma cfDNA captures tumor-derived epigenetic signals. Functional analyses further suggested that VIM methylation is influenced by TP53 mutation status, with lower methylation observed in tumors harboring non-functional TP53, linking epigenetic alterations to key tumor suppressor pathways. These findings indicate that tracking methylation levels of VIM and fibulin-1 in cfDNA could potentially be used not only for early detection but also for monitoring disease progression, providing a dynamic, non-invasive method to complement traditional biomarkers and inform clinical decision-making. While preliminary, these results highlight the promise of epigenomic cfDNA analysis for real-time tumor surveillance in HCC[39]. In recent years, technological developments have certainly substantially improved the sensitivity and specificity of cfDNA methylation-based HCC detection. For instance, a study developed no end-repair enzymatic methylation sequencing combined with pre-trained neural networks, demonstrating superior performance in detecting aberrant methylation patterns in cfDNA with reduced DNA damage. This advancement addresses key limitations in traditional methylation detection technologies[40]. Also, a recent large multicenter study, investigators developed and validated HepaAiQ, a multilocus qMSP assay based on HCC-specific differentially methylated regions. The assay achieved high sensitivity and specificity for early-stage HCC across multiple cohorts and further demonstrated prognostic value by predicting postoperative recurrence risk[41]. These findings show the promise of epigenomic cfDNA features as non-invasive, real-time indicators of treatment efficacy, offering clinicians a window into tumor dynamics that goes beyond conventional genetic analysis.

cfDNA 5-hydroxymethyl signatures

cfDNA 5-hmC signatures arise when ten–eleven translocation enzymes oxidize 5-methylcytosine in DNA, creating 5-hmC. These modifications are associated with active gene regulation and, when found in cfDNA, they reflect the epigenetic state of the tumor of origin, which provides a non-invasive window into cancer presence, tissue origin, and progression. Several studies have evaluated the diagnostic potential of cfDNA 5-hmC signatures in HCC studies carried out mainly in Chinese populations suggest that these epigenetic patterns can help distinguish patients with HCC from those with cirrhosis or other non-cancer conditions. Overall, the results point to a test that is both sensitive and specific, though performance varies between studies, with some leaning stronger on sensitivity and others on specificity[7-9].

What is clear, however, is that most of the available data come from relatively small or single-population cohorts, which limits how broadly the findings can be applied. Even so, these early studies highlight the potential of 5-hmC signatures as a simple, non-invasive blood test for HCC, offering hope for a tool that could one day complement or even improve upon current diagnostic approaches. When it comes to the question of whether these signatures can predict immunotherapy response in HCC, best of our knowledge, evidence remains very limited. Encouragingly, studies in other cancers, such as lung cancer, suggest that these epigenetic markers can predict treatment response and even outperform established biomarkers like PD-L1 expression[42]. This points to cfDNA 5-hmC signatures as a promising avenue for biomarker development in HCC, though their potential in this setting remains to be explored.

Cell-free mitochondrial DNA

A subset of cfDNA, cell-free mitochondrial DNA (cf-mtDNA), is also being explored for predicting treatment response in HCC. For instance, an interesting study examining cf-mtDNA multiple features (copy number, mutation, CNV and fragmentomics) before or after the first TACE treatment based on mtDNA cap-seq showed that a decrease in these features showed stable disease and PR, and an increase on the contrary showed progressive disease after treatment in patients with HCC. Similarly, there was a significant difference in changes in cf-mtDNA multiple features and CNV burden after treatment in patients with stable disease/PR or those with progressive disease. Furthermore, they found that pre-TACE and post-TACE treatment could predict the OS and PFS in patients with HCC. However, this study lacks external validation and had a small sample size, which limits its generalizability[43].

MULTIDIMENSIONAL CHARACTERISTICS OF CFDNA IN HCC

cfDNA analysis provides a multidimensional view of HCC, capturing tumor mutations, structural changes, fragmentation, and epigenetic signals. Although cfDNA does not fully reflect the full scope of intratumoral heterogeneity, its average tumor-plasma concordance of about 70%, which varies by disease stage, assay depth, and tumor burden, is lower in early-stage disease and indicates that it reliably represents dominant clonal drivers and emerging subclones, particularly as they evolve under therapeutic pressure[5]. Quantitative cfDNA demonstrate robust prognostic value. Elevated baseline cfDNA/ctDNA burdens correlate with inferior OS and reduced responses to ICIs, combination immunotherapy, and TKIs. Serial cfDNA kinetics frequently provide earlier insights into treatment response than radiography or AFP, with declines in variant allele fractions often preceding imaging changes. However, early molecular responses do not consistently translate into durable survival benefit, reflecting treatment-induced clonal reshaping, selective resistance, and confounding effects of subsequent therapies. Although these observations form a compelling mechanistic rationale for cfDNA-guided response assessment, no ctDNA-kinetic threshold has yet been prospectively validated for clinical decision-making in HCC.

GENOMIC, STRUCTURAL, FRAGMENTOMIC, AND EPIGENETIC INSIGHTS

Mutation-based profiling through cfDNA further reinforces the genomic landscape underlying HCC progression. TERT promoter mutations and TP53 alterations, which promote telomere dysfunction and genomic instability, are readily detectable in plasma and consistently associated with early recurrence and shortened OS. Alterations in CTNNB1, ARID1A/1B, AXIN1, and PI3K/mTOR pathway genes provide additional prognostic and therapy-specific insights. Notably, PI3K/mTOR pathway alterations correlate with inferior outcomes on TKIs but appear neutral with respect to ICI response, illustrating mechanistic interplay between pathway activation and therapeutic mechanism. cfDNA-derived structural genomic instability metrics, including CIN scores and genome-wide CNV burdens, offer complementary insight into tumor aggressiveness. CIN is an established determinant of immune evasion and metastatic potential across multiple malignancies. In HCC, early plasma-based studies show that CNV burden correlates with tumor volume, Edmondson grade, and recurrence risk, although these signatures remain investigational due to variable computational pipelines.

Fragmentomic analysis adds an orthogonal layer of information by decoding nucleosomal architecture, fragment-size distributions, and end-motif patterns that reflect chromatin accessibility and the dominant modes of tumor cell death. Fragmentomic profiles in HCC demonstrate substantial interpatient heterogeneity but remain stable in cirrhotic and healthy individuals, suggesting tumor-specific fragmentation phenotypes. Machine-learning frameworks such as DNA evaluation of fragments for early interception (DELFI) and DELFI-tumor fraction have achieved strong diagnostic and predictive performance in other malignancies, highlighting their potential translational relevance in HCC, pending disease-specific optimization. Epigenetic profiling, including CpG methylation and 5-hmC mapping, provides high-resolution insight into tissue-of-origin and transcriptional activity. cfDNA methylation signatures mirror tumor-derived patterns, and assays such as nanopore-enhanced methylation sequencing and multilocus methylation classifiers (HepaAiQ) demonstrate promising sensitivity for early detection and recurrence assessment. 5-hmC profiling yields complementary information about active regulatory elements and transcriptional programs, though evidence for predictive utility in HCC remains emergent.

PROMISING AND EXPLORATORY MODALITIES

Across these domains, several cfDNA modalities demonstrate particularly strong translational promise. Quantitative cfDNA kinetics and postoperative minimal residual disease detection show the most immediate clinical potential, with reproducible associations with recurrence risk and therapeutic response across diverse treatment contexts. Mutation-based profiling of TERT, TP53, and PI3K/mTOR pathway alterations provides biologically coherent prognostic information and may support future treatment stratification. Genome-wide CNV burden represents a promising structural biomarker capable of refining recurrence-risk models once analytic harmonization is achieved. Fragmentomic classifiers and genome-wide fragmentation signatures offer orthogonal insights not captured by genotyping alone, and advanced methylation and 5-hmC assays exhibit high analytical sensitivity, positioning them as compelling candidates for early detection, dynamic monitoring, and characterization of therapy-relevant transcriptional states. At the same time, several cfDNA applications remain distinctly exploratory and require foundational validation. Genome-wide fragmentomic metrics such as end-motif entropy, nucleosome phasing, and machine-learning-derived fragmentation classifiers have shown analytical promise in other cancers but lack HCC-specific training and prospective validation. Chromatin topology–inferred fragmentomic signatures and transcriptionally inferred fragmentation profiles remain conceptual, with limited linkage to clinical outcomes. CIN-based scoring systems and focal CNV amplitude metrics also remain exploratory due to inconsistent analytic workflows and variable correlation with biological aggressiveness across HCC cohorts. Integrated multi-omics cfDNA models that combine mutational, structural, fragmentomic, and epigenetic dimensions are in early developmental stages and have not yet been evaluated in treatment-stratified prospective studies. These areas represent high-potential but early-phase efforts needing rigorous technical and biological validation.

IMPLEMENTATION BARRIERS TO CLINICAL ADOPTION

Clinical integration of cfDNA assays remains restricted by multiple implementation barriers as their mechanistic and translational potential continues to expand. Analytical constraints such as low tumor-shedding and limited sensitivity for detecting low-frequency variants challenge reliable measurement, particularly in early-stage or well-differentiated disease. Methodological heterogeneity across sequencing platforms, error correction approaches, and computational pipelines produces noninterchangeable outputs and limits reproducibility. In addition, no validated clinical thresholds exist for molecular response, CNV or CIN burden, or fragmentomic alterations that could guide therapeutic decisions. Operational and regulatory issues, including lack of assay harmonization, limited multicenter standardization, and the absence of evidence demonstrating survival benefit from cfDNA-guided management, further constrain translation. Accordingly, no quantitative, mutational, structural, fragmentomic, or epigenetic cfDNA modality is approved for routine clinical use in HCC. Among current applications, postoperative MRD detection and treatment monitoring are closest to clinical readiness, although both require prospective validation, harmonized workflows, and consensus thresholds before they can be incorporated into practice.

LIMITATION

Several limitations still constrain the translational readiness of cfDNA-based assays in HCC, and these issues need to be resolved before routine clinical use is realistic. First, analytical sensitivity remains a central challenge. ctDNA shedding in HCC is highly variable and often low, especially in early-stage disease, well-differentiated tumors, and lesions with minimal vascular invasion, making it difficult to reliably detect low-frequency variants or focal CNVs. Additional biological factors, such as fragment length, nuclease activity, and cfDNA clearance dynamics, introduce further variability that complicates quantitative interpretation. Second, substantial methodological heterogeneity persists across platforms. Differences in assay chemistry, sequencing depth, enrichment approaches, error-correction methods, and bioinformatic pipelines lead to non-comparable outputs, including variant calls, CIN/CNV metrics, and fragmentomic features. This lack of standardization may limit reproducibility and prevent the field from defining clinically meaningful thresholds for allele fractions, CNV burden, fragmentation signatures, or methylation-based risk scores.

Third, biological and clinical confounders reduce interpretability. Short-term declines in cfDNA following therapy do not consistently predict durable clinical benefit, partly because of compensatory clonal expansion, treatment-driven evolutionary pressures, and subsequent systemic or locoregional interventions that blur correlations with survival. Although recurrent mutations in TERT, TP53, and the PI3K/mTOR pathway remain prognostically relevant, their predictive utility for treatment response, especially for ICIs and combination regimens, remains insufficiently understood. Fourth, while CIN and CNV signatures appear promising, CIN-derived metrics are still exploratory. Current algorithms vary widely in their segmentation thresholds, normalization schemes, and statistical definitions of aneuploidy, resulting in inconsistent associations with tumor aggressiveness and recurrence across HCC cohorts. Fifth, fragmentomic assays require deeper foundational validation. Features such as size distributions, nucleosome phasing patterns, and machine learning-based classifiers like DELFI-tumor fraction have shown strong performance in other cancers, but HCC-specific training datasets remain limited, and the biological underpinnings of fragmentation differences - ranging from necrosis and apoptosis to chromatin structure and cell-of-origin mixing - are not yet well characterized. Sixth, epigenetic assays, including methylation and 5-hmC profiling, although analytically sensitive, need larger, treatment-stratified cohorts to clarify their predictive value. This is particularly important for immunotherapy, where signal patterns derived from other malignancies may not generalize to HCC due to its distinct etiologies and the influence of underlying liver disease.

Finally, the field is constrained by small cohorts, retrospective designs, and heterogeneous treatment exposures, which collectively limit generalizability. Few studies employ consistent sampling schedules, standardized clinical endpoints, or integrated radiologic-molecular response frameworks. Consequently, no cfDNA modality, whether quantitative, mutational, structural, fragmentomic, or epigenetic, has yet demonstrated reproducible clinical utility or survival benefit in prospective trials. Addressing these limitations will require coordinated efforts toward assay harmonization, standardized analytic pipelines, clinically validated thresholds, and large multicenter prospective studies capable of testing cfDNA-guided decision frameworks in real-world settings.

CONCLUSION

The expanding use of cfDNA analysis in HCC offers an opportunity to redefine disease monitoring and therapeutic decision-making. In the near term, integrating cfDNA profiling with existing biomarkers and radiologic assessment could improve response evaluation and enable real-time adaptation of treatment strategies. To achieve clinical adoption, future work should prioritize large, prospective studies that harmonize pre-analytical handling, sequencing depth, and bioinformatic pipelines. Establishing standardized thresholds for cfDNA dynamics and variant detection will be essential for reproducibility across centers. In addition, HCC-specific validation of CIN and CNV scores, as well as fragmentomic classifiers, is warranted to account for the distinct genomic landscape of liver cancer. Implementing pre-registered or locked analytic algorithms will help mitigate overfitting and improve cross-cohort reproducibility. Orthogonal confirmation of key cfDNA findings using independent assays, such as digital droplet PCR or replicate NGS runs, will further strengthen analytical robustness. Advances in multiomics approaches by combining cfDNA, methylation, and fragmentomic data may refine sensitivity and capture tumor heterogeneity that escapes current assays. Ultimately, close collaboration between oncologists, molecular pathologists, and bioinformaticians will be crucial to transition cfDNA monitoring from exploratory research to a validated, clinically actionable tool in HCC management.