Identification of a G-quadruplex-forming cell-free DNA fragment as a biomarker for the precise diagnosis of hepatocellular carcinoma Authors: Jianzhe Sun, Xucong Teng, Weizhi Zhang, Yicong Dai, Lianxin Liu, Jinghong Li Published: 2026 ABSTRACT Early detection of hepatocellular carcinoma (HCC) remains challenging, as the currently recommended surveillance strategy based on ultrasound combined with alpha-fetoprotein (AFP) is limited by suboptimal sensitivity and accessibility. Cell-free DNA (cfDNA) provides a minimally invasive avenue for cancer detection. However, most existing cfDNA-based approaches either perform unreliably in low-input samples or require analytically complex workflows. Here, we systematically profiled serum cfDNA from individuals with HCC and without HCC and identified a high-abundance tumor-associated single cfDNA fragment at the FAM230F genomic region. Integrative analysis of liver assay for transposase-accessible chromatin with sequencing (ATAC-seq) data revealed consistent tumor-specific chromatin accessibility at this locus, suggesting a tumor-derived origin. Structural characterization further demonstrated enrichment of G-quadruplex (G4) features within the target sequence, which may increase resistance to serum nuclease degradation and promote its preferential retention in circulation. Based on these properties, we established a qPCR-based detection workflow with clinical accessibility. In a validation cohort independent of the discovery cohort, a ΔCt cutoff of 2 was selected by maximizing the Youden index within the same cohort. The assay showed a sensitivity of 94.5% and a specificity of 90.5% for distinguishing HCC from non-HCC. Collectively, our study identifies FAM230F as a structurally stable tumor-associated cfDNA fragment and establishes a simple and scalable qPCR-based assay for HCC detection, providing a practical framework for translating cfDNA fragment analysis into clinical biomarkers. Liver cancer is one of the most common primary hepatic malignancies and ranks as the sixth most frequently diagnosed cancer and the third leading cause of cancer-related death worldwide.1 HCC represents the predominant histological subtype, accounting for more than 90% of all liver cancer cases. Because early-stage HCC often lacks specific clinical symptoms, many patients are diagnosed only at advanced stages. Epidemiological studies estimate that approximately 50–60% of cases present with metastatic disease.2–4 Therefore, developing effective screening strategies for HCC is critical for improving detection and clinical outcomes. The American Association for the Study of Liver Diseases (AASLD) recommends surveillance of high-risk populations using abdominal ultrasound combined with serum AFP testing.5 However, ultrasound requires specialized equipment and experienced operators,6 and its sensitivity for early-stage HCC is only approximately 47%.7 AFP testing alone achieves a sensitivity of around 60%,8 resulting in limited overall diagnostic performance. Consequently, there is an urgent need to develop convenient and highly sensitive molecular diagnostic strategies to improve the accessibility and diagnostic effectiveness of HCC detection. Cell-free DNA (cfDNA) consists of short DNA fragments circulating in peripheral blood, primarily released during cellular apoptosis and necrosis. In patients with cancer, a proportion of cfDNA originates from tumor cells and carries genetic alterations associated with tumor initiation and progression, enabling minimally invasive cancer detection.9,10 In recent years, cfDNA-based liquid biopsy technologies have been widely explored for molecular detection of HCC and have also shown diagnostic potential in multiple solid tumors, including lung cancer, pancreatic cancer, and colorectal cancer.11–13 Multiple biological features of cfDNA are associated with tumor development, including somatic mutations, aberrant DNA methylation, and fragmentomic patterns.13 Mutation-based approaches can be used for tumor detection and therapeutic monitoring. However, these strategies rely on detecting specific mutation sites. Given the high heterogeneity of tumor mutation spectra and the extremely low abundance of circulating tumor DNA in plasma, their sensitivity and coverage are often limited.14,15 Studies investigating cfDNA methylation abnormalities provide additional opportunities for early cancer detection, but these approaches typically require complex bisulfite conversion or signal transformation procedures,16 which are sensitive to sample handling and remain challenging for low-input cfDNA samples.17–19 Fragmentomics-based methods analyze features such as cfDNA fragment length distributions and genome-wide coverage patterns to identify tumor-derived signals.20 However, such approaches often require multi-step preprocessing and next-generation sequencing workflows, which limit their implementation in routine clinical laboratories. In contrast, quantifying the abundance of specific cfDNA fragments with tumor-associated sequences in blood may enable a more practical diagnostic strategy using qPCR platforms, which are widely accessible in hospital settings.21,22 Therefore, identifying tumor-associated cfDNA fragment biomarkers represents a promising approach for improving the simplicity and accuracy of HCC detection. Here, using a discovery cohort of 23 patients with HCC and 33 healthy controls (HC), we performed high-throughput sequencing of serum cfDNA and identified a tumor-associated cfDNA fragment (Fig. 1), FAM230F, with high abundance and strong tumor specificity (fold change = 4.87, P < 0.001). Integrative analysis of chromatin accessibility datasets revealed that this fragment originates from a tumor-specific open chromatin region in liver cancer cells. In addition, the sequence contains a G4-forming motif that may enhance resistance to nuclease degradation, thereby contributing to its preferential enrichment in the circulation of patients with HCC. Based on this target, we established a qPCR-based serum detection workflow. In a validation cohort independent of the discovery cohort, comprising 55 patients with HCC and 63 non-HCC participants, the assay achieved a sensitivity of 94.5%, with a specificity of 90.5% using a cohort-derived ΔCt cutoff of 2. This PCR-based detection strategy targeting the FAM230F cfDNA fragment demonstrates several advantages, including promising diagnostic performance, a simple analytical workflow, and low cost, supporting its further clinical evaluation as an accessible strategy for HCC detection. Identification of an HCC-associated cfDNA fragment and schematic workflow of the qPCR-based diagnostic strategy. Identification: serum samples were collected from patients with HCC and HC participants. cfDNA was extracted and subjected to whole-genome sequencing followed by differential fragment analysis to identify tumor-associated cfDNA fragments. Diagnosis: schematic illustration of the qPCR-based detection workflow. Serum cfDNA was extracted and an exogenous synthetic spike-in DNA control fragment was simultaneously added as an external reference. The diagnostic signal was defined as the Ct difference (ΔCt) between the target cfDNA fragment and the spike-in control. Experimental section Clinical sample collection Serum samples and matched clinical information were collected from patients with HCC and non-HCC participants. In the first batch of samples, serum specimens from 23 patients with HCC and 33 HC participants were collected and used for cfDNA sequencing and candidate target identification. In the second batch, serum samples from 55 patients with HCC, 47 HC participants and 16 patients with non-malignant liver disease (NMLD) were collected and used to evaluate the diagnostic performance of the selected cfDNA targets. Clinical specimens included both serum samples and liver tissue samples. Among HCC patients with available clinical staging information, 29 cases were classified as early-stage disease and 34 as advanced-stage disease. The NMLD group included 4 patients with hepatic hemangioma and 12 hepatitis B virus-infected patients without radiologically detectable liver lesions. The age distribution of participants was relatively balanced. Detailed clinical characteristics of all participants are provided in Table S1. Throughout this study, the term “non-HCC” collectively refers to the HC and NMLD groups. All participants were recruited at The First Affiliated Hospital of the University of Science and Technology of China. The study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Institutional Ethics Committee of The First Affiliated Hospital of the University of Science and Technology of China (Approval No. 2023KY-373). Written informed consent was obtained from all participants prior to sample collection. Oligonucleotides All oligonucleotides used in this study were synthesized by Shanghai Sangon Biological Engineering Technology & Services Co., Ltd. The detailed sequences are listed in Table S2. Serum processing Peripheral blood samples were collected in vacuum tubes without anticoagulant. Blood samples were allowed to stand at room temperature for 30–60 min prior to processing. Samples were then centrifuged at 3000 rpm for 10 min at 4 °C, after which the supernatant serum was carefully transferred to new tubes and immediately stored at −80 °C until further analysis. All serum samples were stored for no longer than one year before cfDNA extraction and downstream analyses. Serum cfDNA extraction Serum cfDNA was extracted from 1 mL of serum using a MolPure® Magnetic Circulating Cell-Free DNA Kit (Yeasen; Cat. No. 18821ES50) according to the manufacturer's instructions. To reduce technical variation caused by differences in cfDNA extraction efficiency and low-input recovery across samples, a spike-in DNA control was added to each serum sample before extraction at a final concentration of 0.1 nM. The purified cfDNA was finally eluted in 20 µL of elution buffer. For sequencing-based screening, purified cfDNA was subjected to library construction. For qPCR-based validation, extracted cfDNA was used directly as an amplification template. Library construction and cfDNA sequencing For the discovery cohort, cfDNA libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, E7645) according to the manufacturer's instructions. The resulting libraries were subjected to paired-end sequencing on an Illumina platform. Processing of cfDNA sequencing data Raw paired-end sequencing reads were trimmed using Trim Galore (v0.6.6). Trimmed reads were first aligned to a human repetitive-sequence reference using Bowtie2 (v2.5.4), and reads mapped to repetitive regions were removed. The remaining reads were aligned to the human reference genome (hg19) using Bowtie2 (v2.5.4). Aligned reads were processed using SAMtools (v1.14) and duplicate reads were removed using Picard MarkDuplicates (v2.25.5). Genome-wide normalized coverage tracks were generated using bamCoverage from deepTools (v3.5.5) with reads-per-genomic-content (RPGC) normalization. Enriched cfDNA regions were identified using SEACR (v1.3). Quality control of serum processing, cfDNA extraction, and sequencing data To assess pre-analytical and technical variations, all serum samples were processed using the same standardized procedure and underwent only one freeze–thaw cycle before cfDNA extraction. Following serum separation, samples were visually inspected for residual clots or fibrin precipitates. Extraction consistency was monitored using the spike-in control added before extraction, and the resulting spike-in Ct values were compared between the HCC and non-HCC groups and examined for outlying values. For samples with available sequencing data, sequencing-derived cfDNA fragment-size distributions were assessed using properly paired alignments. For fragment-size quality-control analysis only, read pairs with a mapping quality of <30 and reads marked as duplicates, secondary alignments, supplementary alignments, or quality-control failures were excluded. Fragment lengths were inferred from the genomic span of each read pair, with each fragment counted only once. The overall fragment-size distribution was evaluated for each sequenced sample as part of routine quality control to assess the presence of long DNA fragments and potential contamination by high-molecular-weight genomic DNA. ATAC-seq ATAC-seq data were generated in this study from clinical liver tissue specimens, including HCC tissues (n = 3) and normal liver tissues (n = 2). ATAC-seq was performed as described in a previous study23 by Spatio-Temporal Molecule Biotechnology Co., Ltd (Hefei, Anhui, China). Briefly, approximately 50 mg of frozen liver tissue was mechanically dissociated and homogenized to release nuclei. The nuclei were isolated using NE buffer (20 mM HEPES, pH 7.5, 50 mM NaCl, 2 mM MgCl2, 0.1% Triton X-100, 20% glycerol), washed twice with Tn5 reaction buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 2 mM MgCl2, and 1× protease inhibitor), and resuspended in the same buffer. 2 µL of 10 µM Tn5 transposome was added to 20 µL of Tn5 reaction buffer containing isolated nuclei, and the mixture was incubated at 37 °C for 45 minutes. The resulting DNA fragments were purified and used for library construction and subsequent Illumina sequencing. After adapter trimming, sequencing reads were aligned to the human reference genome (hg19) using Bowtie2. Low-quality, mitochondrial, and duplicate reads were removed. ATAC-seq peaks were identified using MACS2, and RPGC-normalized coverage tracks were generated using bamCoverage in deepTools. The normalized ATAC-seq signals at the FAM230E and FAM230F regions were used for subsequent visualization and quantitative comparison. G-quadruplex prediction Potential G-quadruplex-forming sequences within the candidate cfDNA fragments were predicted using the QGRS Mapper web server24 with default parameters. G4-overlap and random-region analysis Genome-wide overlaps between cfDNA peaks and reported G4 motifs were analyzed using GenomicRanges. An overlap was defined as a shared genomic interval of at least 1 bp. Each random set contained 3178 regions, corresponding to the median peak count across the observed cfDNA datasets. The width of the random regions was fixed at 327 bp, corresponding to the mean of the dataset-specific median peak widths. Random regions were generated across the standard hg19 chromosomes (chr1–chr22, chrX, and chrY). Chromosomes were sampled in proportion to the number of valid start positions, and start positions were sampled uniformly within each selected chromosome. Regions extending beyond chromosome boundaries or overlapping another random region were resampled. A total of 1000 independent random sets were generated using different random seeds, and the G4-overlap proportion was calculated for each set. Enrichment was assessed using a one-sided empirical randomization test. Circular dichroism spectroscopy Synthetic oligonucleotides corresponding to the candidate cfDNA fragments were dissolved in folding buffer (10 mM Tris–HCl, pH 7.4, 50 mM KCl) to a final concentration of 5 µM. Unless otherwise specified for K+ titration experiments, all experiments were performed with 50 mM KCl. DNA samples were heated to 95 °C for 5 min and slowly cooled to room temperature to allow secondary structure formation. Circular dichroism (CD) spectra were recorded using a Chirascan Plus spectropolarimeter (Applied Photophysics, UK) at 26 °C using a 10-mm pathlength quartz cuvette. Spectra were collected from 220 to 320 nm with a 1-nm step size, and each spectrum represents the average of two scans. For K+ titration experiments, KCl was added to final concentrations of 0, 5, 10, 25, 50, and 100 mM, while all other buffer components and DNA concentrations remained unchanged. For CD melting experiments, the CD signal was monitored at the characteristic wavelength identified from the corresponding spectrum while the temperature was increased from 20 to 90 °C at a rate of 0.4 °C min−1. Data were recorded at 2 °C intervals after equilibration for 5 s at each designated temperature. The apparent melting temperature (Tm) of FAM230F was determined from the midpoint of a Boltzmann sigmoidal fit to the normalized CD melting curve. Serum and DNase I stability For serum digestion assays, synthetic DNA fragments (1.5 µL of a 100 µM stock solution) were incubated in a 20 µL reaction mixture containing 50% human serum and 50 mM KCl. The samples were incubated at 37 °C for 30 min. Reactions were stopped by addition of EDTA (final concentration 10 mM), followed by agarose gel electrophoresis to assess DNA degradation. For DNase I digestion assays, synthetic DNA fragments (1.5 µL of a 100 µM stock solution) were incubated in a 20 µL reaction mixture containing 1× DNase buffer and 50 mM KCl. DNase I was added (0.01 U µL−1), and samples were incubated at 37 °C for 10 min. Reactions were terminated by addition of EDTA (final concentration 10 mM), followed by agarose gel electrophoresis to assess DNA degradation. qPCR Quantitative PCR (qPCR) was performed using an MA-9600P real-time PCR system (Molarray, China) with a SYBR Green-based detection system (Hieff® qPCR SYBR Green Master Mix, Yeasen, China). Each 20 µL reaction mixture contained 10 µL of 2× SYBR Green qPCR Master Mix, 0.5 µL of forward primer and 0.5 µL of reverse primer (final concentration 50 nM each), 1 µL of DNA template, and nuclease-free water to make up a final volume of 20 µL. A spike-in control added during cfDNA extraction was amplified in parallel in separate reactions to control for extraction and amplification variability across samples. PCR amplification was performed under the following conditions: 98 °C for 30 s, followed by 44 cycles of 98 °C for 15 s, 60 °C for 20 s, and 72 °C for 30 s. Each sample was analyzed with four technical replicates. Ct values of the target fragments were normalized to the spike-in control using the ΔCt method, calculated as:ΔCt = Ct (target) − Ct (spike-in control) The resulting ΔCt values were used for downstream diagnostic analyses. Analytical performance was evaluated using FAM230F DNA standards ranging from 103 to 107 copies per reaction, together with melting-curve analysis and a no-template control. Results were retained only when the no-template control showed a Ct value greater than 40. Diagnostic performance analysis Receiver operating characteristic (ROC) curves were constructed to evaluate the diagnostic performance of the cfDNA marker. The ΔCt cutoff was selected by maximizing the Youden index in the HCC-versus-non-HCC comparison within the validation cohort. The same HCC, HC, and NMLD samples were used to estimate the corresponding sensitivity and specificity. The area under the curve (AUC), sensitivity, and specificity were calculated using the pROC package in R. Statistical analysis Differences between the two groups were assessed using the two-sided Mann–Whitney U test unless otherwise specified. For comparisons of qPCR ΔCt values, P values from the two prespecified comparisons (HCC versus HC and HCC versus NMLD) were adjusted using the Holm method. RPGC-normalized ATAC-seq signals in Fig. 3b and c were compared using a two-sided Welch's t-test. G4 enrichment in Fig. 4b was assessed using a one-sided empirical randomization test based on 1000 independently generated random sets. P < 0.05 was considered statistically significant. Figure preparation Schematic illustrations were created using https://BioRender.com. Genome visualization figures were generated using Figeno. Results and discussion Identification of HCC-associated cfDNA fragments To systematically identify serum cfDNA fragments associated with HCC, we performed high-throughput sequencing of cfDNA from the discovery cohort (Fig. 1) and conducted genome-wide differential analysis of cfDNA coverage signals. A set of cfDNA fragments showing significant differences between HCC and HC was identified using a false discovery rate (FDR) threshold of <0.05 and fold change (FC) >1. The top 100 fragments with the highest increase in FC in HCC samples were selected for heatmap visualization. After Z-score normalization, HCC samples and control samples showed clearly separated clustering patterns in fragment abundance profiles (Fig. 2a). Genomic annotation revealed that differential cfDNA peaks were markedly enriched in distal intergenic regions, whereas the proportions of peaks located in promoter and proximal transcription-related regions were correspondingly reduced (Fig. 2d). This distribution suggests that tumor-associated cfDNA fragments are preferentially associated with distal regulatory regions rather than promoter–proximal regions. Such a pattern may be consistent with chromatin structural remodeling associated with HCC and altered regulatory element activity, rather than enrichment driven solely by increased transcription initiation accessibility.25,26 Together, these results indicate that serum cfDNA from patients with HCC exhibits characteristic genome-wide enrichment patterns. Identification of HCC-associated cfDNA fragments. (a) Heatmap showing the top 100 HCC-enriched cfDNA fragments ranked by HCC/HC fold change among differentially enriched cfDNA regions (FDR <0.05, FC >1) in the discovery cohort. Differential analysis was based on genome-wide cfDNA coverage signals. Signals were Z-score normalized. The top annotation indicates sample groups. (b) Intensity signal of candidate fragments in HCC samples, calculated as RPGC-normalized coverage. (c) Fold change of candidate fragments in HCC samples relative to HC samples, calculated based on group medians. (d) Genomic functional annotation of cfDNA peaks. Left panel: total cfDNA peak set (n =5128). Right panel: differential cfDNA peaks. Colors indicate genomic annotation categories including promoter, untranslated region (UTR), exon, intron, and distal intergenic regions. The x-axis indicates the proportion of each category. (e) Distribution of RPGC-normalized coverage for the FAM230E fragment in HC samples (n = 33) and HCC samples (n = 23). (f) Distribution of RPGC-normalized coverage for the FAM230F fragment in HC samples (n = 33) and HCC samples (n = 23). Boxplots indicate the median and interquartile range, with individual points representing samples. Fold change (based on group medians) and statistical significance (***, P < 0.001) are indicated. (g and h) Representative genome browser views of the FAM230E and FAM230F regions (hg19). The region of the target fragment is shaded. Tracks display RPGC-normalized cfDNA coverage signals. Tracks show four representative samples from each group (HCC and HC). CB, cancer blood, indicates serum samples from patients with HCC. NB, normal blood, indicates serum samples from healthy controls. Based on the genome-wide differential analysis, we further prioritized candidate fragments for diagnostic applications. Because cfDNA fragments generally exhibit low abundance in serum, candidate fragments were prioritized based on both fold change and intensity signal (IS) in HCC samples (Fig. 2b and c), representing differential magnitude and absolute signal intensity, respectively. Fold change was calculated using the group median as a robust statistic to reduce the influence of extreme values. The results showed that cfDNA fragments located in the genomic regions associated with FAM230E and FAM230F exhibited the highest intensity signal in serum samples from patients with HCC (IS (FAM230E) = 135.98; IS (FAM230F) = 69.00) and showed the most significant differences between disease and control groups (FC (FAM230E) = 9.10; FC (FAM230F) = 4.87). cfDNA signal distributions of the FAM230E and FAM230F fragments were clearly separated between HCC and control samples (Fig. 2e and f). Genome browser tracks further showed consistent enrichment peaks in HCC samples, whereas signals in HC samples were minimal (Fig. 2g and h). Therefore, cfDNA fragments associated with FAM230E and FAM230F were selected as priority candidate targets. Chromatin origin of HCC-associated cfDNA fragments To investigate the chromatin origin of tumor-associated cfDNA fragments, we analyzed the chromatin accessibility profiles of the identified candidate fragments. Previous studies have shown that tumor-derived cfDNA fragments are closely associated with chromatin fragmentation patterns during cell death.26 To examine whether the candidate fragments originate from tumor-specific open chromatin regions, we profiled chromatin accessibility in HCC and HC tissues using ATAC-seq and compared these profiles with serum cfDNA landscapes (Fig. 3a). We observed higher chromatin accessibility at the FAM230E and FAM230F genomic regions in HCC tissues. The average ATAC-seq signal in tumor tissues was 7.14-fold higher at the FAM230E locus (two-sided Welch's t-test, P = 0.0786) and 5.85-fold higher at the FAM230F locus (two-sided Welch's t-test, P = 0.0210) compared with normal liver tissues (Fig. 3b and c). Notably, these regions also showed concordant enrichment in serum cfDNA profiles from HCC patients (Fig. 3d and e). These observations suggest that enrichment of FAM230E/F cfDNA fragments in circulation may originate from selective fragmentation of tumor-associated open chromatin regions. Chromatin origin analysis of FAM230E/F cfDNA fragments. (a) Schematic overview of the analytical workflow used to trace the chromatin origin of target cfDNA fragments. Chromatin accessibility profiles from HCC tissues (ATAC-seq) were compared with serum cfDNA sequencing signals. (b) ATAC-seq signal intensity (RPGC) at the FAM230E region in normal liver tissues (n = 2) and HCC tissues (n = 3). The difference between groups was assessed using a two-sided Welch's t-test (P = 0.0786). (c) ATAC-seq signal intensity (RPGC) at the FAM230F region in normal liver tissues (n = 2) and HCC tissues (n = 3). The difference between groups was assessed using a two-sided Welch's t-test (P = 0.0210). Bar plots represent mean ± s.d., with dots indicating independent biological samples. (d and e) Representative genome browser visualization at the FAM230E and FAM230F regions in normal liver and HCC samples. Genome browser tracks display ATAC-seq and cfDNA sequencing signals. ATAC-seq data were derived from cancer tissues and normal tissues, whereas cfDNA data were derived from serum of HCC and HC. Shaded regions indicate the candidate cfDNA fragment location. CT, cancer tissue, indicates liver tissue samples from patients with HCC. NT, normal tissue, indicates liver tissue samples from healthy controls. CB, cancer blood, indicates serum samples from patients with HCC. NB, normal blood, indicates serum samples from healthy controls. G-quadruplex mediated the maintenance of HCC-associated cfDNA fragments in serum To further explain the high abundance of these fragments in serum, we examined the structural characteristics of the target sequences. G4 structures are non-canonical DNA secondary structures formed by guanine-rich sequences (Fig. 4a). These structures participate in gene regulation and tumor biology,27,28 and can increase DNA stability and reduce nuclease susceptibility.29 These properties suggest that G4 structures may contribute to the preferential retention of tumor-derived cfDNA fragments in circulation. Genome-wide analysis revealed a higher overlap between HCC-enriched cfDNA fragments and reported G4 motifs30 than that observed for median-width-matched random genomic regions (P = 0.000999, Fig. 4b). Prediction using QGRS Mapper24 further indicated that both FAM230E and FAM230F sequences contain regions with high G4-forming potential (Fig. S1). CD spectroscopy indicated that both sequences can form G4-compatible folded structures under the tested conditions (Fig. 4c). The FAM230E sequence exhibited a pronounced positive peak at 285–290 nm, a shoulder near 255–260 nm, and a negative peak around 240 nm. This composite spectral profile is consistent with G4-related folding and may reflect a more complex conformational state involving mixed topological contributions. In addition, the FAM230F sequence showed a strong positive peak at 265 nm and a negative peak near 240 nm, suggesting a parallel-type G4 topology. Mutations of the G4 motifs abolished these characteristic CD spectral features. To further validate the formation of the predicted G4 structures and assess their stability, we performed K+ titration (Fig. 4d) to examine cation-driven folding and conformational changes, followed by CD melting analysis (Fig. 4e) to evaluate their thermal stability. With increasing K+ concentrations, the major positive peak of FAM230F shifted from approximately 270–275 nm to 265 nm, suggesting that K+ promoted a conformational transition toward a more pronounced parallel-type G4 structure. For FAM230E, the positive peak at 285–290 nm progressively increased, while the shoulder near 255–260 nm remained evident. The potassium-dependent enhancement of the G4-associated positive peak suggested that K+ promoted and stabilized its G4-related folded state. CD-melting analysis of FAM230F showed a progressive decrease in the characteristic CD signal at 265 nm with increasing temperature, followed by stabilization at higher temperatures, indicating thermal unfolding of the folded structure. The apparent Tm of FAM230F was approximately 51 °C. FAM230E also showed a temperature-dependent decrease in the CD signal at 285 nm, with a noncanonical melting profile lacking a single well-defined transition. This complex thermal response may reflect multiple structural transitions during heating, consistent with the composite CD spectral profile described above. To directly assess fragment stability under physiological conditions, we evaluated the degradation of the target fragments in human serum and DNase I digestion assays (Fig. 4f). The full gel images are provided in Fig. S2. The remaining DNA fragments were visualized by agarose gel electrophoresis. Fragments containing intact G4 motifs exhibited greater residual band intensity after DNase I or serum treatment compared with the G4-disrupted mutant controls. Collectively, these findings indicate that the enrichment of FAM230E/F fragments in serum is likely driven by G4-mediated structural stabilization, providing a mechanistic basis for their utility as HCC-associated cfDNA detection targets. Contribution of G-quadruplex structures to cfDNA fragment stability. (a) Schematic representation of a G4 structure. Guanine-rich DNA sequences form G-tetrads via Hoogsteen hydrogen bonding, which stack and are stabilized by monovalent cations such as potassium ions (K+). (b) Overlap between cfDNA-enriched regions and genomic G4 motifs. The cfDNA peak set represents enriched regions identified from cfDNA sequencing. Each random peak set contained 3178 mutually non-overlapping regions with a fixed width of 327 bp, corresponding to the median number of peaks across the observed cfDNA datasets and the mean of their dataset-specific median peak widths, respectively. Random regions were generated across the standard hg19 chromosomes, and the procedure was independently repeated 1000 times. An overlap was defined as sharing at least 1 bp with a G4 motif. Bars represent mean ± s.d. Statistical significance was assessed using a one-sided empirical randomization test (P = 0.000999). (c) CD spectra of the FAM230E and FAM230F target sequences and their mutants. Synthetic DNA oligonucleotides containing the predicted G4 motifs were analyzed in vitro. Mutant sequences were generated by substituting key guanine residues with adenine to disrupt G4 formation. (d) Potassium-dependent CD analysis of FAM230E and FAM230F sequences. CD spectra were recorded in the presence of increasing KCl concentrations (0, 5, 10, 25, 50, and 100 mM). (e) CD melting analysis of FAM230E and FAM230F sequences. Temperature-dependent CD changes were monitored at 285 nm for FAM230E and 265 nm for FAM230F during heating from 20 to 90 °C. The apparent Tm of FAM230F was approximately 51 °C. (f) Degradation assays of FAM230E and FAM230F DNA fragments in DNase I and human serum treatments. WT indicates sequences containing intact G4 motifs, whereas Mut indicates G4-disrupted mutants. Samples were incubated for the indicated time points (10 min and 30 min), and the remaining DNA fragments were visualized to assess stability. A qPCR-based diagnostic assay targeting the FAM230F cfDNA fragment Based on the tumor-associated features of the FAM230E/F fragments, we further established a qPCR-based diagnostic workflow (Fig. 1). Because the input amount and extraction efficiency of serum cfDNA may vary among samples, a spike-in control was added to each serum sample at a final concentration of 0.1 nM prior to cfDNA extraction. The amplification signal of the target fragment was normalized using this spike-in control. The difference between the Ct values of the target fragment and the spike-in control (ΔCt) was used as the diagnostic readout to reduce technical variability and improve cross-sample comparability. We focused on a reproducible 124-bp cfDNA peak region that was consistently observed in HCC samples (Fig. 5a). The qPCR amplicon was restricted to a 60-bp core region within this window (Fig. 5a) to minimize the influence of cfDNA fragment-end heterogeneity and improve amplification specificity. We first evaluated the feasibility of detecting the FAM230E and FAM230F fragments (n = 6). The results showed that the Ct values for FAM230E were generally close to the lower detection range of the qPCR assay and did not display a stable separation between groups (Fig. S3). The FAM230E sequence contains AT-rich and continuous poly(A) tracts, which may introduce non-specific primer binding and reduce amplification efficiency. Therefore, subsequent assay development focused on the FAM230F locus, which showed robust amplification performance. An optimized amplification system was established for this fragment (Fig. S4), including optimization of annealing temperature and primer concentration, and was used for subsequent diagnostic performance evaluation. Establishment and validation of the cfDNA diagnostic assay. (a) Schematic illustration of cfDNA fragment detection at the FAM230F target region. The blue region indicates the target detection window (124 bp). Black arrows represent the positions of qPCR primers, producing a 60-bp amplicon. Dashed lines mark the boundaries of the amplification region (chr22: 18 879 943–18 880 067). Using the center of the target fragment as position 0, the length and relative position distribution of cfDNA fragments within ±200 bp are displayed. Each horizontal line represents a cfDNA fragment observed in an individual sample. Genomic coordinates corresponding to the displayed region are indicated below (chr22: 18 879 800–18 880 200). (b) qPCR detection of the FAM230F fragment in serum cfDNA from HC (n = 47), NMLD (n = 16) and HCC (n = 55). ΔCt represents the relative abundance of the FAM230F fragment in serum cfDNA. Each point represents an individual sample. Box plots indicate the median and interquartile range. The dashed line indicates the diagnostic threshold (ΔCt = 2). (c) Receiver operating characteristic (ROC) curve based on qPCR ΔCt values for evaluating the diagnostic performance of the FAM230F fragment in distinguishing HCC from non-HCC. The optimal cutoff value and corresponding sensitivity and specificity are indicated. The area under the curve (AUC) quantifies overall discriminative performance. ΔCt is defined as the difference between the Ct value of the target fragment and that of the spike-in control. ROC analysis was performed using samples from the validation cohort. To evaluate the diagnostic performance of the FAM230F target in HCC detection, we conducted a systematic assessment in a validation cohort independent of the discovery cohort. A total of 55 patients with HCC and 63 non-HCC participants, including 47 HC participants and 16 patients with NMLD were included. The ΔCt value was used as the readout for target abundance. A ΔCt cutoff of 2 was determined by maximizing the Youden index in the HCC-versus-non-HCC comparison. ΔCt values were lower in the HCC group than in both the HC and NMLD groups (Fig. 5b, adjusted P < 0.0001 for both comparisons). Receiver operating characteristic (ROC) analysis showed that the assay achieved an area under the curve (AUC) of 0.971 for distinguishing HCC patients from non-HCC participants (Fig. 5c). The FAM230F cfDNA marker achieved a sensitivity of 94.5% (95% CI, 85.1–98.1%) and a specificity of 90.5% (95% CI, 80.7–95.6%) for distinguishing HCC from non-HCC. When the non-HCC participants were evaluated separately, the specificity was 91.5% (95% CI, 80.1–96.6%) in the HC group and 87.5% (95% CI, 64.0–96.5%) in the NMLD group. We further performed a direct comparison between the FAM230F cfDNA marker and the conventional serum biomarker AFP in the same HCC validation cohort. Among the 49 HCC patients with available AFP measurements, the sensitivity of AFP detection was only 77.6%, lower than the 95.9% sensitivity achieved by the FAM230F cfDNA marker. Collectively, these results indicate that the FAM230F cfDNA fragment demonstrates promising diagnostic accuracy in the validation cohort and maintains robust detection performance in AFP-negative cases, supporting the potential clinical utility of this single-fragment cfDNA detection assay for HCC diagnosis (Table S4). Conclusions HCC surveillance remains challenging in clinical practice, as current strategies based on ultrasound and AFP show limited sensitivity and are influenced by operator expertise and clinical conditions, contributing to delayed diagnosis and highlighting the need for more sensitive, reliable, and accessible diagnostic approaches. In the present study, we established a simple and scalable qPCR-based diagnostic assay for HCC by targeting a tumor-associated cfDNA fragment. We identified a fragment mapped to the FAM230F genomic region that is preferentially enriched in patients with HCC. Mechanistically, we demonstrated that its enrichment is associated with tumor-specific chromatin accessibility and enhanced structural stability potentially mediated by a G4-forming sequence. This approach exhibited reliable detectability and achieved an AUC of 0.97 with a sensitivity of 94.5% and a specificity of 90.5%, outperforming AFP-based detection (77.6%). These findings support a practical and clinically translatable strategy for HCC detection. In previous studies, multiple liquid biopsy strategies have been explored for HCC detection. In addition to cfDNA-derived molecular features, including somatic mutations, DNA methylation alterations, and fragmentomic characteristics, serum protein biomarkers (such as AFP, lens culinaris agglutinin-reactive AFP (AFP-L3), and des-gamma-carboxy prothrombin (DCP)), circulating RNAs, exosomal RNAs and proteins, tumor-associated metabolomic features, and tumor-associated autoantibodies have also been reported to possess diagnostic potential for HCC. These detection strategies vary considerably in both diagnostic performance and analytical complexity (Table S5). The single-fragment cfDNA detection strategy proposed in this study uses a tumor-associated cfDNA fragment as a quantitative target, demonstrating competitive diagnostic performance relative to existing liquid biopsy approaches for HCC. Importantly, because this method relies on a qPCR-based platform, it reduces analytical complexity while enabling a simplified workflow, stable quantitative readout, and adaptability to low-input serum samples, thereby improving accessibility and facilitating clinical implementation. The use of FAM230E/F as diagnostic targets is supported by a meaningful biological background. Both genes belong to the human FAM230 family of long non-coding RNA (lncRNA) genes. Under an annotation framework based on the nearest transcription start site (TSS), these loci are categorized as distal intergenic regions, consistent with the chromosomal enrichment pattern observed for the tumor-enriched peaks identified in this study. Previous studies have reported that FAM230F expression correlates with the expression of the cancer-associated protein USP41 and is associated with poor prognosis in multiple tumor types.31,32 In addition, other members of the FAM230 family, including FAM230B, FAM230C, and FAM230D, have been implicated in tumorigenesis and cancer progression.33,34 Furthermore, the abundance of the FAM230E/F cfDNA fragment increased with tumor progression across different cancer stages (Fig. S7). Together, these findings provide biological support for the potential use of the FAM230F genomic region as a serum molecular biomarker. In summary, we identified a tumor-associated cfDNA fragment mapped to the FAM230F genomic region and established a simple, accurate, and clinically accessible strategy for HCC detection. Nevertheless, further optimization is still required, particularly in simplifying sample preprocessing and standardizing the detection workflow. Because the ΔCt cutoff was selected by maximizing the Youden index in the same cohort used to estimate sensitivity and specificity, the reported diagnostic performance may represent an optimized estimate. Therefore, the cutoff and its associated performance require confirmation in independent prospective cohorts. Although the spike-in control was added before cfDNA extraction and provides internal normalization for variations associated with sample input, cfDNA extraction, and amplification, it cannot fully eliminate platform-dependent differences in fluorescence detection, threshold-setting algorithms, reagents, and thermal cycling conditions. Consequently, transferability of the ΔCt cutoff across laboratories and qPCR platforms will require standardized reference materials, harmonized pre-analytical and analytical procedures, platform-specific calibration, and multicenter inter-laboratory validation. Moreover, the molecular pathways involving FAM230F during tumor development and progression, as well as its potential relationship with nucleosome stability in serum, remain to be further elucidated through mechanistic studies. In addition, the tumor specificity and cross-cancer performance of a single-fragment cfDNA detection strategy at the serum level require systematic evaluation in larger and multi-cancer cohorts. Overall, this study proposes an HCC detection strategy based on a tumor-associated cfDNA single-fragment target, providing a feasible framework for translating cfDNA biomarkers into practical molecular detection assays. Author contributions J. S. and X. T. conceived the study, designed the experimental strategy, performed the experiments, analyzed the data, and contributed to manuscript writing. W. Z. collected and curated the clinical samples. Y. D., L. L., and J. L. supervised the study and provided critical revisions of the manuscript. All authors reviewed and approved the manuscript. Conflicts of interest The authors declare no competing financial interests. Supplementary Material