Urinary cell-free DNA as a noninvasive liquid biopsy for hepatocellular carcinoma

Abstract

BACKGROUND

Cell-free DNA (cfDNA) has advanced cancer genetic profiling through liquid biopsy. Recent evidence suggests that urinary cfDNA may serve as a noninvasive alternative to plasma-derived cfDNA.

AIM

To comprehensively characterize transrenal DNA as a source for genetic profiling of hepatocellular carcinoma (HCC), using hepatitis B virus (HBV) DNA as a representative marker.

METHODS

HBV- and HCC-targeted next-generation sequencing (NGS) was performed to compare HBV fragment size distributions, 5’ 4-mer end motifs, and HCC mutation detection rates between urine and plasma cfDNA.

RESULTS

HBV transrenal DNA fragments from HCC patients were significantly shorter than those from patients with hepatitis or cirrhosis and displayed distinct 5’ 4-mer end motifs. Compared with plasma, urine samples contained higher levels of HCC-distinctive 5’ 4-mer end motifs. Mutation detection rates were comparable between urine and plasma, with urine higher mutation for TP53 in NGS. Among 101 HCC patients, detection of TP53, CTNNB1, and TERT mutations showed a 78.3% overall concordance between urine and plasma cfDNA. In a subset of 15 patients analyzed by HCC-targeted NGS, positional-level concordance between urine and plasma cfDNA reached 97%.

CONCLUSION

Urine cfDNA demonstrates both overlapping and unique features compared with plasma cfDNA, supporting its potential as a noninvasive and accessible source for genetic profiling of liver cancer.

Key Words: Cell-free DNA; Hepatocellular carcinoma; Urine; Hepatocellular carcinoma mutation; Liquid biopsy

Core Tip: Urinary cell-free DNA may serve as a simple, noninvasive tool for detecting liver cancer (hepatocellular carcinoma). In this study, we found that urine contains shorter, tumor-specific DNA fragments and unique genetic markers that help distinguish liver cancer patients from those with other chronic liver conditions. With similar accuracy to blood-based tests and the advantage of easier sample collection, urine cell-free DNA has the potential to improve liver cancer diagnosis and monitoring.

INTRODUCTION

Cell-free tumor DNA (ctDNA) in biofluids has created new opportunities for cancer screening and management. While most applications have focused on ctDNA in plasma, urine offers a fully noninvasive alternative for ctDNA that overcomes limitations associated with blood collection, such as restricted specimen volume and repeated blood draws[1]. Urine has been shown to contain ctDNA originating from both urological and non-urological cancers, including hepatocellular carcinoma (HCC), where tumor-derived DNA is filtered through the kidneys into urine as transrenal DNA (trDNA)[2-8]. Our previous work has shown that HCC ctDNA can be detected in trDNA for early detection of primary and recurrent HCC[6,7], as well as for disease monitoring and management.

Despite the promise of transrenal ctDNA, the inconsistent reports regarding its performance for detection compared to blood ctDNA have posed challenges to the clinical application of urine DNA[2,5,9-14]. These discrepancies may reflect differences in specimen handling and processing, as well as the biological complexity of urinary DNA. Urinary DNA represents a complex mixture derived from prerenal, renal, and postrenal sources, it contains high-molecular-weight DNA (> 1000 bp) from exfoliating urothelial cells, along with cell-free DNA (cfDNA) originating from urinary tract cells and filtered transrenally from non-urinary tract cell[8]. This complexity highlights the need for a further understanding of urine DNA biology and its impact on ctDNA detection.

Unlike plasma, urine is not under stringent homeostatic regulation, and ctDNA detection can be affected by preanalytical factors and assay methodologies[15-17]. For instance, reported trDNA fragment sizes vary considerably, from ~50 bp to 150-300 bp, likely due to differences in DNA isolation methods and assay platforms[18,19]. If the majority of trDNA fragments are ~50 bp, polymerase chain reaction (PCR) assays must employ ultra-short amplicon designs, and NGS library preparation must be optimized to accommodate such short substrates to ensure effective target detection. Understanding the size distribution of liver-derived DNA is therefore critical for reducing the variability and improving the reliability of urine-based liquid biopsy in liver-related diseases.

Building on this foundation, we evaluated urine-based liquid biopsy in comparison with plasma-based liquid biopsy for HCC to address existing gaps in its biological characterization and clinical utility. As outlined in Figure 1, we first examined the fragment sizes of urine cfDNA (human DNA) and trDNA [hepatitis B virus (HBV) DNA] from patients with chronic HBV infection, cirrhosis, and HCC. Consistent with previous studies[18,19], urinary HBV DNA fragment sizes showed substantial variability. Analysis of HBV DNA 5’ 4-mer end motifs, HCC-associated mutation detection, and HCC-targeted variant allele frequencies (VAFs) indicates that urine cfDNA distinguishes HCC from non-HCC more effectively than plasma cfDNA. Collectively, these findings support urine liquid biopsy as a promising noninvasive complementary or alternative assay for genetic profiling, enabling frequent sampling for early detection and disease monitoring in clinical practice.

Figure 1 The study outline. trDNA: Transrenal DNA; HBV: Hepatitis B virus; cfDNA: Cell-free DNA; HCC: Hepatocellular carcinoma; qPCR: Quantitative polymerase chain reaction; NGS: Next-generation sequencing.

MATERIALS AND METHODS

Patients and body fluid collection

Specimen collections from participants were approved by each study center’s Institutional Review Board. This study was conducted in accordance with both the Declarations of Helsinki and Istanbul. A written informed consent was received prior to participation. Detailed patient information is summarized in Supplementary Tables 1 and 2. Patients with glomerular filtration rates < 30 ng/mL were excluded from this study. Matched urine and plasma were collected from each patient. Urine was mixed with EDTA (10-50 mmol/L final concentration) immediately upon collection and stored at -80 °C until isolation. Blood (10 mL) was collected in K₂EDTA BD Vacutainer tubes (BD, Franklin Lakes, NJ, United States). Plasma was separated from whole blood by centrifugation at 1300 × g for 20 minutes at 4 °C followed by transfer of plasma to a new tube for a second centrifugation at 3000 × g for 10 minutes at 4 °C. Double-spun plasma was aliquoted in 1-mL portions and stored at -80 °C until downstream processing.

DNA isolation and quantitation

DNA was extracted from urine samples as previously described[20]. The low molecular weight fraction (cfDNA; < 1 kb) was size-selected with HighPrep PCR magnetic beads (MagBio Genomics, Gaithersburg, MD, United States) at 0.55X bead/sample ratio. Plasma cfDNA used in Table 1 was extracted using the Quick-cfDNA Serum & Plasma kit (Zymo Research, Irvine, CA, United States). Plasma cfDNA used for the remainder of the study was isolated using the JBS plasma cell-free DNA isolation kit (JBS Science Inc., Doylestown, PA, United States). DNA concentrations were determined using the Qubit dsDNA HS assay kit (Thermo Fisher Scientific, Waltham, MA, United States).

Table 1 Detection of cell-free tumor DNA markers by quantitative polymerase chain reaction in matched urine and plasma samples from hepatocellular carcinoma patients (n = 101).

	TP53	CTNNB1	TERT	Total	Cohen’s
Overall concordance % (95%CI), concordant/total	87.8 (78.7-94.0), 72/82	83.3 (69.8-92.5), 40/48	64.9 (53.2-75.5), 50/77	78.3 (72.0-83.7), 162/207	0.565
Positive concordance with at least one positive test (positive agreement) % (95%CI), concordant/total	0.0 (0.0-30.8), 0/10	20.0 (2.5-55.6), 2/10	30.8 (17.0-47.6), 12/39	23.7 (13.6-36.6), 14/59	-0.525

CI: Confidence interval.

Mutation detection by quantitative PCR

Archived matched plasma-urine cfDNA from 101 HCC patients[6] was used for mutation analysis. First, the total copy number of each target gene in each DNA sample was determined using quantitative PCR (qPCR) assays (JBS Science Inc., Doylestown, PA, United States) following the manufacturer’s instructions. Only urine DNA samples with > 100 copies/mL were included in the subsequent analysis. Mutation detection by qPCR was performed for TP53 codon 249T mutation, CTNNB1 codon 32-37 mutations, and TERT c.-124C>T promoter mutation in duplicate using JBS mutation assay (JBS Science Inc., Doylestown, PA, United States).

NGS library preparation, sequencing, and data analysis

Targeted sequencing was performed using the HBV and HCC panels from JBS Science Inc. Libraries were prepared from 10 ng of urine or plasma cfDNA using the NEBNext Ultra II DNA Kit (NEB, Ipswich, MA, United States) and xGen UDI-UMI adapters (IDT, Cornwall, IA, United States), followed by target enrichment according to the manufacturer’s protocols. Paired-end sequencing was performed on a NovaSeq S1 flow cell (Psomagen, Rockville, MD, United States). The HCC panel targets 24 genes, spanning a total region of 64 kb (Supplementary Table 3). The sequencing data are available at NCBI SRA under BioProject number PRJNA1365867.

For HBV-targeted NGS analysis, sequencing reads were processed with an in-house pipeline. Briefly, raw reads were first trimmed for adapter contaminations and low quality sequences, followed by an initial alignment to 42 HBV reference genomes[21] using BWA MEM[22]. Aligned sequences were consolidated using fgbio CallMolecularConsensusReads with information from unique molecular identifier (UMI) at family size of one. The consensus sequences were then realigned to the same set of HBV references to identify HBV genotype and subgenotype. HBV sequences were finally extracted from an additional alignment to the reference of each sample’s subgenotype for downstream analysis. Similar process was applied to obtain human cfDNA sequences from the same samples using hg38 as reference. The analysis workflow is summarized in Supplementary Figure 1A. A total of 48 urine samples and 4 plasma samples were processed, generating an average of 25 million paired-end reads per sample (range: 0.3 million to 110 million), with 93.2% (range: 88.7%-96.5%) of bases achieving a quality score above Q30.

For HCC-targeted NGS analysis, raw sequencing data were first demultiplexed, and each paired-end read was tagged with a UMI. Unmapped BAM files were converted to FASTQ format for initial alignment using BWA MEM[22] against hg19. The resulting BAM files were downsampled within each plasma-urine pair. UMI processing was conducted using fgbio, Connor, and Picard suites, with a minimum UMI family size of three applied. Consensus reads were then realigned, quality-filtered, and used to generate variant call format files using bcftools[23]. Variants with quality score ≥ 30 were further filtered to exclude common germline variants (dbSNP138). VAFs were calculated for the variants that passed filtering. The analysis workflow is summarized in Supplementary Figure 1B. A total of 15 matched urine-plasma pairs (30 samples) were processed, generating an average of 42 million paired-end reads per sample (range: 22 million to 65 million), with 92.1% (range: 89.0%-93.7%) of bases achieving a quality score above Q30.

RESULTS

Comparison of HBV trDNA fragment sizes with human cfDNA in urine

To investigate the fragment size profile of liver-derived trDNA, we analyzed urinary HBV DNA as a liver-specific biomarker. Urinary HBV DNA was examined in patients with HBV infection using HBV-targeted NGS and compared it with human cfDNA in urine. Although HBV trDNA reads were fewer than human cfDNA reads in urine, HBV trDNA fragments were significantly shorter (orange, Figure 2A) than human cfDNA fragments (blue, Figure 2A; P < 0.001, Student’s t-test).

Figure 2 Hepatitis B virus transrenal DNA fragment size distributions compared to human cell-free DNA in urine and liver-derived hepatitis B virus cell-free DNA in plasma. A: Comparisons of DNA fragment sizes of transrenal DNA (hepatitis B virus DNA; orange) and urine cell-free DNA (human DNA; blue) from two hepatitis B patients, Pt 1 and Pt 2, with serum viral load undetectable at the time of specimen collection; B: Comparisons of hepatitis B virus DNA fragment sizes between matched urine (transrenal DNA) and plasma (cell-free DNA) from two hepatitis B patients, hepatitis I and hepatitis II, and two hepatocellular carcinoma (HCC) patients, HCC I and HCC II. Hepatitis II had a 6.9 Log IU/mL serum viral load at the time of specimen collection. The other three patients had undetectable viral load at the time of specimen collection. HBV: Hepatitis B virus; HCC: Hepatocellular carcinoma.

We next compared the HBV DNA size distribution in urine (trDNA) and plasma, collected on the same day from individuals with HBV and HBV-HCC to control for temporal variation and technical variables (Figure 2B). To minimize potential contamination from circulating viral particles, we selected three patients with undetectable serum viral load, hepatitis I, HCC I, and HCC II. We also included one patient with a high viral load (hepatitis II, 6.9 Log IU/mL) for comparison. Plasma HBV DNA exhibited typical nucleosomal sizes, whereas HBV trDNA was significantly shorter, peaking within a broader range of 80-150 bp, suggesting greater inter-individual variability (Figure 2B). Size distributions of trDNA were similar between undetectable and high viral load samples. However, the size distribution of plasma HBV DNA from the high-viral-load patient (hepatitis II) was more dispersed, possibly due to the presence of circulating viral particles. These findings confirm that liver-derived trDNA is shorter than both urine cfDNA (human) and plasma HBV DNA, highlighting the need for assays optimized for short, fragmented templates to improve urine-based detection of HCC.

Comparison of HBV trDNA fragment sizes among hepatitis, cirrhosis, and HCC

Having demonstrated that liver-derived trDNA, identified using HBV DNA as a biomarker, is shorter than human urine cfDNA, we next investigated whether these shorter fragments reflect underlying tumor biology in HCC. To address this, we analyzed urinary trDNA size distributions across patients with different liver disease states. In plasma, ctDNA is known to be shorter than total cfDNA[24] and represents only a small fraction of circulating DNA. Given that trDNA originates from the bloodstream[8,15], we hypothesized that shorter urinary HBV trDNA fragments in HCC patients reflect tumor-derived HBV DNA through clonal expansion or increased tumor cell death.

We compared transrenal HBV DNA size distributions among patients with HCC (n = 17), hepatitis B (n = 14), and cirrhosis (n = 13) (Figure 3A-C). Consistent with our hypothesis, urinary HBV trDNA differed significantly among the three disease groups (P < 0.001, one-way ANOVA). HBV trDNA fragments from HCC patients had the shortest median insert size (100 bp), significantly shorter than those with hepatitis (145 bp, P < 0.001) and cirrhosis (113 bp, P = 0.048) as shown in Figure 3D. This distinction was not correlated with HBV viral load. This finding is likely driven by increased tumor cell death or clonally expanded HBV DNA within the HBV-infected HCC microenvironment, which dominates over DNA fragments from non-tumor hepatocytes.

Figure 3 Comparison of hepatitis B virus DNA fragment sizes in urine from 14 hepatitis, 13 cirrhosis, and 17 hepatocellular carcinoma patients. A-C: Distribution of hepatitis B virus DNA insert size from three disease groups, hepatitis (A), cirrhosis (B), and hepatocellular carcinoma (C), as listed in Supplementary Table 1, are plotted for each group; D The insert size statistics. HCC: Hepatocellular carcinoma.

Diagnostic characteristics of HBV trDNA fragmentation patterns compared to plasma

cfDNA fragmentation is known to be nonrandom, influenced by DNA nucleases activity, generating sequence-specific cleavage patterns or unique 5’ end motifs. These specific end motifs have been reported to distinguish plasma samples from cancer and non-cancer patients, including those with HCC[25-28]. Hence, we analyzed the 5’ 4-mer end motifs of HBV DNA in HCC and non-HCC (HBV) patients to determine whether trDNA exhibits distinct fragmentation signatures and how this compared to plasma-derived HBV DNA.

First, we compared the frequencies of the 10 most abundant HBV DNA 5’ 4-mer end motifs in plasma (Figure 4A, top panel) with their corresponding frequencies in urine (Figure 4A, bottom panel) from HBV-infected patients without HCC (non-HCC, n = 14, orange) and with HCC (HCC, n = 17, blue). We then performed a reciprocal analysis by identifying the top 10 most frequent 5’ 4-mer end motifs in urine (Figure 4B, top panel) and assessing their frequencies in plasma (Figure 4B, bottom panel) for the same patient cohorts. Interestingly, CGTT was the only one shared between the top 10 motifs in both plasma and urine. CGTT ranked as the 10^th most frequent motif in plasma and 3^rd in urine, and its frequency was consistently lower in HCC patients compared to non-HCC patients in both sample types.

Figure 4 Comparison of the 10 most frequent 5’ 4-mer end motifs of hepatitis B virus DNA fragments in plasma and in urine between hepatocellular carcinoma and non-hepatocellular carcinoma hepatitis B patients. A and B: The 10 most frequent 5’ 4-mer end motifs identified in plasma (A) and in urine (B) (top panel) and in the corresponding matched urine and plasma (bottom panel) from hepatocellular carcinoma (blue) and non-hepatocellular carcinoma (orange) patients are compared. Motifs are sorted by motif frequencies in the top panel. HCC: Hepatocellular carcinoma.

Notably, the two most frequent motifs in plasma, CCCA and CCAA, were more abundant in urine from HCC patients than in non-HCC patients, despite showing no appreciable difference in plasma. Similarly, the top two most frequent 5’ 4-mer end motifs in urine, CCAA and GAGA, were 3-fold and 6-fold more abundant, respectively, in HCC compared to non-HCC patients, while no detectable difference was observed in plasma. Collectively, these comparisons suggest that differences in 5’ 4-mer end motifs of HBV DNA between HCC and non-HCC are more pronounced in urinary trDNA than in plasma cfDNA.

Comparison of mutation detection between urine and plasma ctDNA

We evaluated concordance in mutation detection between urine and plasma ctDNA by qPCR using archived DNA and HCC-targeted NGS (patient clinical information as detailed in Supplementary Table 2). Across the three HCC-associated mutations (TP53, TERT, CTNNB1) analyzed by qPCR, the mutation detection rates were comparable between the two specimen types, with urine and plasma showing 78.3% overall agreement across all analyzed pairs (n = 101) by qPCR (Table 1 and Supplementary Table 4). The global variant concordance was 96.8% overall agreement across all analyzed pairs (n = 15) across the HCC panel by NGS (Supplementary Table 5). However, for mutation-positive specimens, concordance reduced to 23.7% for qPCR and 17.7% for NGS (Table 1 and Supplementary Table 5). Despite lower concordance for positive detection, VAFs demonstrated high correlations between matched urine and plasma for concordant variants (Supplementary Figure 2; Pearson’s r = 0.966; 95% confidence interval: 0.966-0.966), supporting the reliability of variant detection in both fluids.

Among targeted genes analyzed by NGS, the TERT promoter showed the highest somatic variant call rate in both fluids (21.9% in urine vs 20.2% in plasma), followed by KMT2D, TP53, and ARID1A. Importantly, six target genes (CDKN2A, HNF1A, ARID1B, BAP1, IL6ST, PIK3CA) had more than 2-fold higher mutation call rates in urine compared to plasma suggesting that specific HCC-related mutations may be selectively enriched in urine cfDNA (Supplementary Table 6). Taken together, these findings demonstrate that mutational profiles of HCC tumors are reflected in both urine and plasma cfDNA. Differences in detection rates highlight the complementary roles of urine and plasma in identifying tumor-associated mutations, supporting the diagnostic potential of urine cfDNA as a reliable and noninvasive biomarker for HCC.

DISCUSSION

This study demonstrated that urinary cfDNA provides a viable and noninvasive source for genetic profiling, capable of detecting HCC-associated mutations. By analyzing HBV trDNA fragment size distributions and 5’ 4-mer end motifs alongside HCC-associated mutations, we show that urine cfDNA offers comparable, if not superior, diagnostic utility relative to plasma cfDNA. Importantly, our findings reveal that urine cfDNA harbors shorter, tumor-specific DNA fragments and detectable HCC-associated mutations (e.g., TP53, TERT), supporting urine’s role as a complementary or alternative liquid biopsy platform to blood-based approaches.

We previously demonstrated that HBV DNA detected in urine originates from chronically infected hepatocytes[29]. Using HBV DNA as a trDNA marker, we discovered important features that distinguish urinary cfDNA from plasma cfDNA. First, HBV trDNA displayed a shorter size distribution than plasma HBV DNA, with median sizes ranging from 100-145 bp. This is consistent with previous findings that trDNA is significantly shorter than cfDNA in bloodstream[30,31]. Importantly, our analysis was performed on DNA isolated directly from urine without pre-centrifugation, thereby avoiding potential loss of cell debris associated with trDNA[20,32]. This methodological difference may explain discrepancies in the reported trDNA sizes in the literature[33], which range from predominantly 50 bp or less, with median sizes of 80-100 bp[18,19]. Additionally, HBV trDNA from HCC patients was shorter than that from non-HCC patients, likely reflecting tumor-derived clonal shedding from HBV-infected tumorous cells. These cells undergo increased turnover and DNA fragmentation which dominates over other HBV-infected non-tumorous hepatocyte-derived HBV-DNA, resulting in shorter fragment size[24,34]. In addition to fragment size, our analysis revealed distinct 5’ 4-mer end motif profiles in urinary cfDNA compared to plasma. While CCCA was frequently observed in both plasma and urine, CCAA was found in higher abundance in HCC compared to non-HCC samples, with particular enrichment in HCC urine. These motif profiles may reflect distinct DNA nuclease populations in urine and plasma (e.g., DNASE1, DNASE1 L3, and DFFB) that are capable of nonrandom cfDNA fragmentation. Although the comparison of 4-mer motifs across studies is not always consistent, our HBV-based motif analysis highlights the diagnostic and biological insights that urinary cfDNA can offer and may better distinguish HCC from non-HCC compared to plasma. Validation in larger patient cohorts will be required to confirm this interesting discovery. Beyond simple size metrics, further studies are underway to explore additional fragmentomic features and to determine whether specific genomic regions within these HBV sequences may provide deeper biological insights and enhance the clinical utility of this approach.

Our analysis also demonstrated that urinary cfDNA reliably detects key somatic mutations, with qPCR showing comparable detection rates for TP53, TERT, and CTNNB1 between urine and plasma. Across 101 HCC patients, overall mutation concordance between fluid types was high (78.3%), though positive agreement for mutation-positive samples was lower (23.7%), reflecting biological variability between fluids as observed in patients with non-genitourinary malignancies[12]. In HCC-targeted NGS of 15 patients, urine cfDNA demonstrated > 2-fold higher mutation call rates for six genes, suggesting urine cfDNA may enrich specific tumor-associated mutations and provide complementary insights to plasma cfDNA. Notably, TP53 mutations were detected more frequently in urine than plasma, consistent with observations in other cancers, including non-small cell lung cancers[14] and urothelial cancers[35,36].

Our study has several limitations. First, most patients in our cohort had early-stage disease, and low DNA input (10 ng) may have substantially reduced the detection sensitivity. Second, the urinary cfDNA isolation method used may not efficiently recover fragments < 100 bp, which are often enriched for tumor-derived variants[37]. Third, tumor mutational profiles from tissue were unavailable for most patients, precluding direct comparisons with plasma and urine cfDNA. Differences between tissue-derived and liquid biopsy-derived mutations are well-documented, with tissue biopsies reflecting localized tumor regions while cfDNA represents tumor DNA shed systemically[38]. Lastly, the lack of standardized protocols for DNA isolation and data acquisition technologies across liquid biopsy studies poses challenges for cross-study comparisons. Even minor technical differences (e.g., DNA input for NGS or sequencing depth) can drastically affect study outcomes and conclusions. It is therefore critical to validate our findings in larger-scale studies with standardized methodologies.

CONCLUSION

In conclusion, urinary cfDNA offers a promising, noninvasive tool for liver cancer diagnostics, monitoring, and personalized treatment. Its ability to reliably capture tumor-specific fragments, enriched motifs, and key mutations underscores its clinical utility as a valuable complement or an alternative to plasma-based methods. The ease of urine collection makes urine liquid biopsy an attractive tool to improve patient compliance and gain access to genomic profiling in resource-limited settings. This is a pilot study to demonstrate the potential utility of urine as a body fluid for HCC genetics. A larger controlled cohort study will be needed to determine the specific utilities that urine liquid biopsy can be applicable and further optimize this approach for widespread clinical adoption.

ACKNOWLEDGEMENTS

The authors thank Max Chao from JBS Science Inc. for his technical and bioinformatics assistance.