Skip to main content
Download PDF

Discovery and validation of a novel dual-target blood test for the detection of hepatocellular carcinoma across stages from cirrhosis

BMC Medicine volume 23, Article number: 278 (2025) Cite this article

Abstract

Background

Hepatocellular carcinoma (HCC) is one of the most common cancers. Early detection of HCC helps improve the patients’ 5-year survival rate. Our goal was to identify superior methylation biomarkers to develop a methylation-specific quantitative PCR (MS‒qPCR) assay.

Methods

A five-phase case–control study identified HCC methylation biomarkers via capture sequencing, TCGA/RNA-seq filtering, technical (MS-qPCR/Sanger) and biological (quadruplex MS-qPCR) validation. Methylated biomarkers were selected based on differential methylation expression using a tissue discovery cohort (43 HCC, 32 normal) and validated in plasma validation cohorts (Phase 1: 53 HCC, 52 cirrhosis, 20 benign, 50 healthy; Phase 2: 67 HCC, 81 cirrhosis). Then, the final assay's HCC detection performance was compared with existing blood-based surveillance methods.

Results

Two methylated genes, OSR2 and TSPYL5, and a novel internal reference gene, SDF4, were identified and developed into an MS‒qPCR assay named Qliver. Qliver had an AUC of 0.955 (95% CI: 0.924–0.987) for distinguishing HCC patients from non-HCC patients in the Phase 1 plasma cohort, with a sensitivity of 88.68% (95% CI: 76.97%-95.73%) and a specificity of 89.34% (95% CI: 82.47%-94.20%), and 0.958 (95% CI: 0.927–0.989) for distinguishing HCC patients from cirrhosis patients in the Phase 2 plasma cohort, with a sensitivity of 88.06% (95% CI: 77.82%–94.70%) and a specificity of 92.59% (95% CI: 84.57%–97.23%). For the Phase 1 plus Plasma 2 cohort, Qliver had an AUC of at least 0.958 for detecting HCC in healthy individuals, cirrhosis patients and patients with benign liver diseases, which was superior to that of the GALAD score (AUC: 0.777 to 0.849). For BCLC stage 0 and A HCC patients, the sensitivity of Qliver ranged from 62.50% (95% CI: 24.49%–91.48%) to 72.73% (39.03%–93.98%), with a specificity of 90%. Overall, Qliver was superior to the AFP, AFP-L3, DCP and the GALAD score in terms of cirrhosis history, tumor stage, tumor size and tumor count.

Conclusions

Qliver demonstrated superior performance in detecting HCC compared with currently widely used blood biomarkers, suggesting its potential clinical benefit in HCC surveillance in high-risk populations.

Peer Review reports

Background

Primary liver cancer is a common malignant tumor worldwide, ranking sixth in incidence and third in mortality among cancers [1]. Hepatocellular carcinoma (HCC) accounts for 75% to 85% of primary liver cancers [1]. HCC patients diagnosed at early stages can achieve a 70% 5-year survival rate through transplant or resection, whereas those with advanced HCC who are only eligible for palliative treatments have a medium survival rate of less than one year [2, 3]. Therefore, early detection of HCC can significantly improve patient survival rates. Liver ultrasound (US) is the recommended strategy for HCC surveillance in high-risk populations and is inexpensive but less effective in detecting early-stage HCC, with a sensitivity of 84% (95% confidence interval [CI]: 76%–92%) for any stage HCC detection, but only 47% (95% CI: 33%–61%) for early-stage HCC [4]. US performance depends on the examiner’s experience, and obesity may further reduce its sensitivity [5]. Serum alpha-fetoprotein (AFP) is insufficient for screening for HCC, with a sensitivity of 25%−65% and a specificity of 80%−94% at a cutoff of 20 ng/mL, and only approximately 60%−80% of HCC patients have elevated AFP levels, resulting in a large margin for false negatives [6]. The sensitivity of combined US and AFP for detecting early HCC reached 63% (95% CI: 48%–75%) [4], but there is still room for improvement. The GALAD score, a blood biomarker-based model that combines age, sex, α-fetoprotein (AFP), the lens culinaris agglutinin-reactive fraction of AFP (AFP-L3) and des-gamma-carboxyprothrombin (DCP), outperformed US in detecting early-stage HCC, with an AUC of 0.92 (95% CI: 0.88–0.96; cutoff: 1.18, sensitivity 92%, specificity 79%) [7]. Although the GALAD score performs well in the early detection of HCC, its performance still needs to be fully validated in the Chinese population, as chronic HBV infection is the main cause of HCC. Therefore, there is an urgent need for a noninvasive HCC detection method suitable for the Chinese population with high sensitivity and specificity.

DNA methylation is an epigenetic mechanism that regulates gene expression [8]. Hypermethylation of tumor suppressor genes is an early event in the carcinogenesis of many cancers [9, 10]. Circulating cell-free DNA (cfDNA) is an extracellular nucleic acid fragment released from necrotic, apoptotic or viable cells [11]. Circulating tumor DNA (ctDNA) originates from tumor cells and accounts for a small fraction of cfDNA. Studies have shown that the methylation level of ctDNA in plasma is positively correlated with the number of primary tumor cells [12]. cfDNA methylation patterns have great potential as biomarkers for noninvasive cancer screening and monitoring [10, 13]. Guo et al. [14] developed a targeted methylation capture sequencing panel based on 283 CpG sites that has high accuracy in detecting HCC, with an AUC of 0.957 (sensitivity 90%, specificity 97%), but its workflow is cumbersome and costly. The use of a methylation-specific quantitative PCR (MS‒qPCR) assay that combines two genes, RNF135 and LDHB, is less expensive, but its performance in detecting HCC in high-risk groups (AUC = 0.7306; 95% CI: 0.6955–0.7658) needs to be improved [15].

This study aimed to screen methylated genes with excellent performance, develop a cost-effective MS‒qPCR assay for HCC detection, and validate its performance by comparing it with currently used biomarkers, including AFP, AFP-L3, DCP and the GALAD score.

Methods

Patient samples and characteristics

Patients aged 18 years or older with clinically diagnosed HCC in TNM stages I to III without treatment and individuals with liver cirrhosis or benign liver tumors such as hepatic adenomas, liver focal nodules or hepatic hemangiomas who were negative for HCC following disease surveillance were enrolled in the study. Twelve HCC patients with unknown TNM stage and 100 healthy volunteers without liver cirrhosis were also included. Frozen tissue and whole-blood samples from HCC patients and all non-HCC subjects and formalin-fixed paraffin-embedded (FFPE) slides from breast cancer and lung cancer patients were collected from Fujian Cancer Hospital. Whole blood samples of 8 to 10 mL were collected from each participant using Cell-Free DNA BCT® tubes (Streck, USA) and shipped to the laboratory at ambient temperature before plasma separation. Patient characteristics and demographic information are provided in Additional file 1: Table S1. All procedures were approved by Medical Ethics Committee of the Fujian Cancer Hospital (K2022-103–01). All research was conducted in accordance with the Declaration of Helsinki and the Declaration of Istanbul. Written consent was given by all the subjects.

Study design

This study was conducted through five sequential case‒control experiments (Fig. 1). A capture sequencing panel was employed to identify differentially methylated genes in tissue and validated in plasma samples from HCC patients and control subjects. These genes were then filtered through the TCGA 450 K and TCGA RNA-seq databases to select genes that exhibited high methylation in HCC samples and showed significant changes in RNA expression level compared with healthy. The selected genes were subsequently validated using MS-qPCR and Sanger sequencing with HCC tissues and adjacent normal tissues (ADJ) (tissue technical validation), and the top genes were chosen based on their differential methylation levels and haplotypes. Furthermore, the diagnostic performance of individual genes was assessed through biological validation using two quadruplex MS‒qPCR assays in plasma from HCC patients, subjects with liver cirrhosis or benign liver diseases, and healthy individuals (Phase 1 plasma cohort), with the two genes exhibiting the best discriminative performance selected for the construction of the final assay. The diagnostic performance of this assay was subsequently validated using plasma obtained from HCC patients and subjects with liver cirrhosis (Phase 2 plasma cohort). Its diagnostic performance was then compared with that of existing blood biomarkers, including AFP, AFP-L3, DCP, and the GALAD score, across both the Phase 1 and Phase 2 plasma cohorts (see the Additional file 2 for details).

Fig. 1
figure 1

Flow diagram to discovery and validation of a novel dual-targets blood test for detection of hepatocellular carcinoma across stages from cirrhosis patients. In tissue discovery cohort, we utilized a human methylome bisulfite panel which targets 123 Mb of genomic content to identify DNA methylation-based biomarkers starting with 43 primary HCC tissue and 32 adjacent normal tissue samples. 21 candidate markers were validated in 55 HCC, 55 cirrhosis and 50 healthy plasma samples by bisulfite sequencing in plasma discovery cohort. Technically validated methylated candidates were then biologically validated on 50 primary HCC tissue and 50 adjacent normal tissue samples (including tissue samples for tissue discovery) using methylation-specific quantitative PCR (MS-qPCR), yielding six candidate DNA markers for further validation. In phase 1 plasma cohort, candidate markers were validated in 53 HCC, 52 cirrhosis, 20 benign liver diseases and 50 healthy plasma samples. Two methylation genes, OSR2 and TSPYL5, and a novel internal reference gene (SDF4) were identified, and an MS-qPCR assay named Qliver was constructed. An HCC detection model was trained using the relative methylation levels of OSR2 and TSPYL5. Finally, Qliver model was further validated in 67 HCC and 81 cirrhosis plasma samples in phase 2 plasma cohort

Full size image

Data analysis

Methylation profiling via targeted methylation sequencing

The FASTQ files were processed via the Cutadapt package (https://github.com/marcelm/cutadapt/) to obtain clean data by finding and removing adapter sequences and poly-tailed low-quality sequences and discarding reads shorter than 50 bp. Clean bisulfite reads were aligned to the hg19 human reference genome from the 1000 Genomes Phase 3 resources with decoy and patch sequences using BSMAP software (https://code.google.com/archive/p/bsmap/). The mapped reads were split into top/bottom strands using bamtools software (https://github.com/pezmaster31/bamtools) according to the ZS tag in the BAM file generated by the BSMAP aligner, which indicated the top/bottom strands and the forward/reverse read status. Duplicates were removed separately by MarkDuplicates (Picard) (https://github.com/broadinstitute/picard), and the split files were rejoined and sorted by coordinates. ClipOverlap (bamUtil) (https://github.com/statgen/bamUtil) was used to prevent the converted/unconverted C bases in the overlapping regions from being double-counted. The percentage of methylated C bases was determined using the methratio.py script provided by BSMAP. BisMark (https://github.com/FelixKrueger/Bismark) was used to determine read-level cytosine methylation states (also known as the methylation haplotype).

Identification of informative CpG markers for HCC

HCC-specific informative CpG markers were identified using the following criteria: CpGs with more than 70% missing values in tissue or plasma samples were excluded during the tissue discovery and plasma validation phase. CpG retention with a P value < 0.01 (two-sided Wilcoxon rank-sum test) was used to compare HCC tissues and adjacent normal tissues (ADJ). CpGs with a P value < 0.05 (two-sided Wilcoxon rank-sum test) between plasma samples of patients with HCC and patients with liver cirrhosis and between plasma samples of patients with HCC and healthy controls were preserved. To identify differentially methylated CpGs, the methylation difference for each CpG was calculated as the mean methylation value of the HCC samples minus the mean methylation value of the control samples. CpGs with methylation differences ≥ 0.3, ≥ 0.2 and ≥ 0.1 in HCC tissues vs. ADJ tissues, HCC plasma vs. healthy subject plasma, and HCC plasma vs. liver cirrhosis plasma were filtered to retain CpGs with an FDR < 0.05 (Student’s two-sided t test and the Benjamini–Hochberg false discovery rate for P value correction). CpGs with median methylation levels ≥ 40% in HCC tissues and ≥ 20% in HCC plasma were retained. The removal of genes with stable expression levels, defined as a fold change between 0.95 and 1.05 between HCC patients and normal controls, was performed. This evaluation was based on RNA-Seq data obtained from the UCSC Xena Hub (https://gdc.xenahubs.net). Moreover, for dimensionality reduction and the selection of highly associated methylation features associated with hepatocellular carcinoma, the Boruta package (https://github.com/scikit-learn-contrib/boruta_py) was utilized for feature selection to improve the performance of the model and resistant to overfitting and noise in the data.

Identification of the methylated haplotype in HCC

In the aggregation step, neighboring informative CpGs with a predefined window size range (80–300 bp) were merged into candidate methylation haplotypes using a sliding window-based segmentation method. Three or more informative CpGs in each haplotype were required. Additionally, each identified haplotype had to be observed in sequencing reads from no fewer than 20 primary HCC tissues and 20 HCC plasma samples. The genes containing candidate methylation haplotypes were filtered using the TCGA 450 K dataset and TCGA RNA-Seq data. The genes with significantly higher methylation levels in HCC patients than in non-HCC individuals in the 450 K dataset but lower RNA expression levels in RNA-Seq data were retained because dysregulated DNA methylation can lead to the silencing of tumor suppressor genes or the expression of oncogenes, thus contributing to the development of cancer.

Identification of a novel reference gene for MS‒qPCR

Genes containing CpGs showing stable methylation levels were identified via the following criterion: exclusion of CpGs with more than 80% missing values in tissue samples. CpG methylation differences ≤ 0.2 were detected between HCC tissues and adjacent tissues, between HCC plasma samples and healthy plasma samples, and between HCC plasma samples and cirrhosis plasma samples. CpGs with median methylation levels ≥ 60% in HCC tissues and ≥ 50% in HCC plasma samples were retained. CpGs with a tau (τ) index ≥ 0.05 were removed from the analysis.

The tau (Ï„) index, which indicates whether a gene is tissue specific or ubiquitously methylated across tissues, produces a single specificity score for each gene or CpG site, indicating the ability to distinguish different cancer types from each other. The beta values of the DNA methylation 450 K array from the TCGA PanCan Atlas Cohort (9,639 samples across 32 tumor types) were downloaded and compiled by combining available data from all TCGA cohorts. The tau (Ï„) index was calculated as follows:

$$\uptau =\frac{{\sum }_{i=1}^{n}1-\widehat{{x}_{i}}}{n-1};\widehat{{x}_{i}}=\frac{{x}_{i}}{\underset{1\le i\le n}{\text{max}}({x}_{i})}$$

where xi is the beta value of the gene in cancer type i and n is the number of cancer types in the TCGA. Ï„ varies from 0 to 1, where 0 indicates consistent methylation levels across different tissues and 1 indicates tissue specificity.

Several steps were taken to refine the selection of candidate reference genes: the exclusion of genes containing CpG-SNP sites or with a low level of CpG density; the elimination of genes annotated as pseudogenes; and the removal of DEGs with expression fold changes > 1.05 or < 0.95 between patients with HCC and normal controls was performed. This evaluation was based on RNA-Seq data obtained from the UCSC Xena Hub (https://gdc.xenahubs.net).

Statistical analysis

Candidate biomarker genes were selected based on the area under the receiver operating characteristic (ROC) curve (AUC) estimated using the R package pROC (version 1.18.0). The 95% confidence intervals (CIs) were calculated using 2000 stratified bootstrap replicates. A generalized linear model (GLM) was generated to analyze the correlations between the DNA methylation levels of selected genes based on the ΔCt values determined via MS‒qPCR using the GLM function of R (4.1.1). All participants in phase 2 plasma cohort were random split into training and validate set with enough number of repeats (t = 10) and a reasonable balance between training and validate set (30% for training, 70% for validate). Comparing the difference (ΔAUC) between the training and validation cohorts allows us to assess whether the difference is statistically significant, thereby evaluating the risk of overfitting.

Sample size calculation

According to the literature and previous related studies, the expected sensitivity of the Qliver assay is 90% when the specificity is 90%. At a tolerance of 0.05 (α = 0.05), at least 138 patients with HCC and 138 non-HCC participants were included in the pooled plasma cohort (Phase 1 plus Phase 2) according to the sample size (n) calculation formula for diagnostic performance validation as follows [16]:

$$n=\frac{{\left[{Z}_{1-\alpha /2}\right]}^{2}\times P\times (1-P)}{{\Delta }^{2}}$$

where P is the predetermined value of the sensitivity (or specificity) that was ascertained from previously published data or clinician experience and for α = 0.05 (meaning a 95% confidence level), and Z1-α/2 is the standard normal variate (1.96 at 5% error). Δ is the maximum marginal error of the estimate.

Performance calculation for an intended-use 10 K real-world population

The positive predictive value (PPV) of the Qliver assay and GALAD score (defined as the proportion of patients with HCC among participants with positive test results) were computed via Bayes’ theorem as follows:

$$PPV= \frac{sensitivity \times prevalence}{sensitivity \times prevalence + (1 - specificity) \times (1 - prevalence)}$$

Similarly, the negative predictive value (NPV) of the Qliver assay and GALAD score (defined as the proportion of non-HCC subjects among participants with negative test results) were computed via Bayes’ Theorem as follows:

$$NPV = \frac{specificity \times (1 - prevalence)}{(1 - sensitivity) \times prevalence + specificity \times (1 - prevalence)}$$

Similarly, positive predictive agreement (PPA) and negative predictive agreement (NPA) were computed as follows:

$$PPA = \frac{sensitivity \times prevalence}{sensitivity \times prevalence + (1 - specificity) \times (1 - prevalence) + (1 - sensitivity) \times prevalence}$$
$$NPA = \frac{specificity \times (1 - prevalence)}{(1 - sensitivity) \times prevalence + specificity \times (1 - prevalence) + (1 - specificity) \times (1 - prevalence)}$$

The benefit of using Qliver model to the intended-use population was evaluated as follows:

$$\frac{sensitivity}{1-specificity}\ge \frac{1-prevalence}{prevalence}\cdot \frac{harm}{benefit}$$

In particular, the incidence of HCC in the cirrhosis population is 2.1% per year according to R Fan et al. [17], which is an intended-use population.

Results

Identification of methylated genes and a novel internal reference gene for hepatocellular carcinoma detection

Through the analysis of methylation differences at individual CpG sites in tissue and plasma samples, feature selection using machine learning, methylated haplotype identification and filtering through the TCGA database, we identified 21 candidate genes that exhibit highly methylation and linkage in hepatocellular carcinoma, including C1QL4, CR1L, CYP26C1, FOXG1, GHSR, HIST1H1D, IRX5, KCNG3, LHX2, MEX3A, NEFM, OSR2, OTX1, OXTR, PCDHGB6, PCDHGB7, PITX1, PRLHR, PRRX1, TSPYL5, and ZIC4. The heatmap visually demonstrated that the methylation levels of these 21 genes were higher in HCC tissues than in adjacent normal tissues (Fig. 2A). Detailed information on the methylation blocks of the 21 candidate genes is summarized in Additional file 1: Table S2.

Fig. 2
figure 2

Identification of methylated markers for hepatocellular carcinoma. A Unclustered heatmap of the 21 most differentially methylated markers between 43 primary HCC and 32 adjacent normal tissue samples (p-value were computed with Wilcoxon rank sum test.). B The distribution of methylation differential levels of 21 candidate markers in 55 HCC, 55 cirrhosis and 50 healthy plasma samples (p-value were computed with Wilcoxon rank sum test.). Plasma_Ctrl: cirrhosis and healthy plasma samples. Plasma_HCC: HCC plasma samples (C) The distribution of ΔCT value differences of 20 candidate markers (OTX1 was excluded from the analysis due to nonspecific PCR amplification and primer-dimers) in 50 primary HCC and 50 adjacent normal tissue samples (including tissues for tissue discovery). HCC: hepatocellular carcinoma. ADJ: adjacent normal tissue. D ROC curves and associated AUC values with 95% confidential interval for six candidate DNA markers in 53 HCC, 52 cirrhosis, 20 benign liver disease and 50 healthy plasma samples. E Heatmap of methylation levels of CpGs in OSR2 gene for discriminating Primary HCC tumor (n = 377), Recurrent HCC tumor (n = 2) and Solid Tissue Normal (n = 50) in the GDC TCGA Liver Cancer (LIHC) 450 K dataset. F Heatmap of methylation levels of CpGs in TSPYL5 gene for discriminating Primary HCC tumor (n = 377), Recurrent HCC tumor (n = 2) and Solid Tissue Normal (n = 50) in the GDC TCGA Liver Cancer (LIHC) 450 K dataset. G The distribution of RNA expression values of OSR2 and TSPYL5 gene in Primary HCC tumor (n = 377), Recurrent HCC tumor (n = 2) and Solid Tissue Normal (n = 59) in the GDC TCGA Liver Cancer (LIHC) gene expression RNAseq dataset

Full size image

Among the 21 candidate genes, six genes, namely, KCNG3 (p = 4.4e-11), OSR2 (p = 5.9e-09), IRX5 (p = 8.8e-08), PITX1 (p = 2.3e-08), OTX1 (p = 4.7e-08), and TSPYL5 (p = 1.1e-08), presented the most significant differences in methylation levels in plasma samples from HCC patients, subjects with liver cirrhosis and healthy individuals (Fig. 2B). Moreover, the methylation profiles of 43 primary HCC tissues and 32 adjacent normal tissues revealed that these six genes exhibited HCC-specific methylation haplotypes (Additional file 3: Fig. S1). These results suggest that these 21 genes have great potential in HCC detection, especially these 6 genes.

Several steps were undertaken to refine the selection of candidate reference genes, as mentioned in the Methods section. Ultimately, UBE2K, SDF4, PIGG and KIAA0562 were selected as candidate reference genes. Homo sapiens stromal cell-derived factor 4 (SDF4), which encodes a stromal cell-derived factor that is a member of the CREC protein family, was ultimately identified in an independent pilot experiment as the most stably conserved reference gene in the tissue and plasma of patients with HCC and control subjects. SDF4 gene was hypermethylated in both HCC tissues and adjacent normal tissues and that the methylation levels of its hypermethylated chromosomal regions were not significantly different between HCC tissues and adjacent normal tissues (Additional file 3: Fig. S2). We also observed from the GDC TCGA Liver Cancer (LIHC) 450 K database that most CpG sites in the SDF4 gene are hypermethylated in HCC primary tumor tissues, normal solid tissues, and recurrent HCC tissues (Additional file 3: Fig. S3A). According to the GDC PanCancer (PANCAN) 450 K dataset, the methylation levels of most CpG sites in the SDF4 gene did not differ significantly between different cancer types (Additional file 3: Fig. S3G). The results of capture sequencing and public database analysis revealed that SDF4 could be used as an internal reference gene for MS-qPCR because of its stable high methylation level.

Technical validation of methylated genes and internal reference gene

The 21 candidate genes were technically validated in 50 HCC tissues and 50 adjacent normal tissues via MS‒qPCR to screen out several more reliable genes. OTX1 was excluded from the MS‒qPCR validation because of nonspecific amplification and a high proportion of primer dimers. The differences in the methylation of OSR2 (p = 5.2e-11), C1QL4 (p = 2.2e-11), PITX1 (p = 1.4e-10), KCNG3 (p = 1.2e-10), IRX5 (p = 5.4e-09), TSPYL5 (p = 8.2e-08), ZIC4 (p = 3.4e-08) and FOXG1 (p = 1.8e-08) were more significant than those of the other 12 genes (Fig. 2C). Sanger sequencing was performed on tissue samples to confirm the methylation linkage status of the candidate genes. The methylation levels of the CpG sites of the OSR2 and TSPYL5 genes in HCC tissues were significantly greater than those in adjacent normal tissues, and most of these CpGs were comethylated (Additional file 3: Fig. S4). On the basis of the p values from low to high, we selected the top six genes, namely, OSR2, C1QL4, PITX1, KCNG3, IRX5 and TSPYL5, for biological validation in independent plasma samples by two quadruplex MS-qPCR.

In addition, the potential of the SDF4 gene as an internal reference was further verified using genomic DNA from leukocytes and tissues, as well as plasma cfDNA, in comparison with the commonly used internal reference gene ACTB. The Ct values of MS-qPCR for SDF4 and ACTB were moderately correlated with the input amount of leukocyte genomic DNA (Additional file 3: Fig. S3E, F). There was no significant difference in the Ct values of SDF4 between 38 primary HCC tissues and 38 adjacent normal tissues (Additional file 3: Fig. S3C) or between 12 breast cancer tissues and 20 lung cancer FFPE tissues (Additional file 3: Fig. S3D), and the same was true for ACTB. However, SDF4 was amplified more efficiently than ACTB. Importantly, when SDF4 was used as an internal reference gene, the hypermethylated gene ZIC4 performed better in distinguishing HCC patients, cirrhosis patients, and healthy controls then when ACTB was used as an internal reference gene in 2 mL plasma samples (AUC 0.826, 95% CI: 0.664–0.988 vs. AUC 0.786, 95% CI: 0.608–0.964) (Additional file 3: Fig. S3B).

Establishment of Qliver score in 2 mL plasma cohort

The six candidate genes were validated by two quadruplex MS–qPCR using SDF4 as an internal reference gene in an independent set of plasma samples (Phase 1 plasma cohort), which included 53 HCC patients, 52 cirrhosis patients, 20 individuals with benign liver diseases and 50 healthy volunteers. Subsequently, the six candidate genes were combined in pairs, and their diagnostic performance for HCC detection was evaluated. The combination of OSR2 and TSPYL5 exhibited the highest diagnostic performance (AUC = 0.955, 95% CI: 0.924–0.987) (Additional file 1: Table S3), outperforming either gene individually (OSR2: AUC = 0.927, 95% CI: 0.884–0.971; TSPYL5: AUC = 0.927, 95% CI: 0.879–0.975), highlighting their potential as optimal methylation markers for HCC detection (Fig. 2D).

Subsequently, an MS‒qPCR assay named Qliver, which includes OSR2, TSPYL5 and SDF4, was developed for HCC detection. The Qliver score, which represents the probability of a subject having HCC, was calculated according to the following formula generated by the GLM:

$$Qliver score=1- \frac{1}{1+ {e}^{11.608-0.882* {\Delta }_{OSR2}- 0.835* {\Delta }_{TSPYL5}}}$$

where e is Euler’s number, a mathematical constant approximately equal to 2.71828;

ΔOSR2 refers to the ΔCt value of OSR2, which is obtained by subtracting the Ct value of SDF4 from the Ct value of OSR2;

ΔTSPYL5 refers to the ΔCt value of TSPYL5, which is obtained by subtracting the Ct value of SDF4 from the Ct value of TSPYL5.

The â–³Ct values of OSR2 and TSPYL5 in the Phase 1 plasma cohort were trained to construct a model (Qliver model) for predicting HCC using the generalized linear model (GLM), and the performance of Qliver was compared with that of protein biomarkers, such as AFP, AFP-L3, DCP and the GALAD score. The coefficients and intercept of the GLM of the Qliver model are shown in Additional file 1: Table S4. The AUC of Qliver for distinguishing HCC patients from non-HCC patients was 0.955 (95% CI: 0.924–0.987), which was significantly greater than that of AFP (0.751, 95% CI: 0.665–0.838), AFP-L3 (0.685, 95% CI: 0.597–0.773), DCP (0.728, 95% CI: 0.622–0.834), and the GALAD score (0.837, 95% CI: 0.764–0.911; P value = 0.003, DeLong’s test). Even the AUCs of the single genes OSR2 (0.927, 95% CI: 0.884–0.971) and TSPYL5 (0.927, 95% CI: 0.879–0.975) were greater than those of the GALAD score when multiple protein markers were combined (Fig. 3B). When the specificity was set at almost the same level (85.25%–89.34%), the sensitivity of Qliver for distinguishing HCC patients from non-HCC patients was 88.68% (95% CI: 76.97%–95.73%), which was better than that of AFP (47.17, 95% CI: 33.30%–61.36%), AFP-L3 (41.51%, 95% CI: 28.14%–55.87%), DCP (62.26%, 95% CI: 47.89%–75.21%) and the GALAD score (58.49%, 95% CI: 44.13%–71.86%) (Table 1). The PPV of Qliver for detecting HCC in non-HCC patients was 78.33% (95% CI: 68.19%−85.91%), which was significantly greater than that of AFP (65.79%, 95% CI: 51.66%–77.58%), AFP-L3 (57.89%, 95% CI: 44.05%–70.60%), DCP (64.71%, 95% CI: 53.27%–74.68%) and the GALAD score (70.45%, 95% CI: 57.62%–80.70%). The NPV of Qliver for detecting HCC in non-HCC patients was 94.78% (95% CI: 89.51%–97.48%), which was significantly greater than that of AFP (79.56%, 95% CI: 74.98%–83.49%), AFP-L3 (77.37%, 95% CI: 72.96%–81.25%), DCP (83.87%, 95% CI: 78.50%–88.10%) and the GALAD score (83.21%, 95% CI: 78.16%–87.28%) (Table 1).

Fig. 3
figure 3

Construction and validation the diagnostic performance of Qliver model to detect HCC in plasma cohort. A 2D kernel density estimation of ΔCT value of OSR2 and TSPYL5 gene in combined plasma cohort (phase 1 plus phase 2), which consists with 120 HCC, 133 cirrhosis, 20 benign liver disease and 50 healthy plasma samples. B ROC curves and associated AUC values for OSR2, TSPYL5, AFP, AFP-L3, DCP, GALAD and Qliver in phase 1 plasma cohort consisting of 53 HCC, 52 cirrhosis, 20 benign liver disease and 50 healthy plasma samples. C ROC curves and associated AUC values for OSR2, TSPYL5, AFP, AFP-L3, DCP, GALAD and Qliver in phase 2 plasma cohort consisting of 67 HCC and 81 cirrhosis plasma samples. D The comparison of ΔCT values of OSR2 (D), TSPYL5 (E), Qliver score (F) and GALAD score (G) in different sample groups in combined plasma cohort (phase 1 plus phase 2), consisting of 120 HCC, 133 cirrhosis, 20 benign liver disease and 50 healthy plasma samples (p-value were computed with Wilcoxon rank sum test). *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001; p > 0.05 was considered not significant (ns). Healthy, healthy volunteers; HCA, Hepatocellular adenoma; FNH, Focal Nodular Hyperplasia; HH, Hepatic hemangioma; Cir, liver cirrhosis; HCC, hepatocellular carcinoma. H Comparison of HCC prediction results between Qliver and other classifiers in combined plasma cohort (phase 1 plus phase 2) consisting of 120 HCC, 133 cirrhosis, 20 benign liver diseases and 50 healthy plasma samples. A Z-score normalization is performed on the normalized value across samples for each marker. Heatmaps and dendrograms were also created to depict the Qliver characteristics based on the Euclidean distance and ward. D2 clustering methods. Since OSR2 and TSPYL5 are highly methylated in HCC patients, the Δct value of their MS-qPCR amplification is lower than that of the control subjects

Full size image
Table 1 HCC detection metrics of Qliver,AFP,AFP-L3,DCP, and GALAD score
Full size table

With a specificity of 90%, Qliver was more sensitive than AFP, AFP-L3, DCP and the GALAD score in detecting all stages (BCLC stage) of HCC. In particular, the sensitivity of Qliver in detecting stage 0 + A HCC was 62.50% (95% CI: 24.49%–91.48%), which was significantly greater than that of AFP (37.50%, 95% CI: 8.52%–75.51%), AFP-L3 (50.00%, 95% CI: 15.70%–84.30%), DCP (25.00%, 95% CI: 3.19%–65.09%) and the GALAD score (25.00%, 3.19%–65.09%) (Table 2). Moreover, Qliver was more specific than these protein biomarkers in detecting hepatic adenomas and hemangiomas (Additional file 1: Table S5). A more detailed comparison of Qliver with the protein biomarkers is shown in Additional file 1: Table S5.

Table 2 Sensitivity for the HCC detection of Qliver, AFP, AFP-L3, DCP, and GALAD score across BCLC stages
Full size table

Stable performance of Qliver score in independent 1 mL plasma cohort

The Qliver model was further validated in an additional independent cohort of plasma samples (Phase 2 plasma cohort), with the volume of each plasma sample reduced from 2 to 1 mL. The AUC of Qliver (0.958, 95% CI: 0.927–0.989) in distinguishing 67 HCC patients from 81 patients with liver cirrhosis was significantly greater than that of AFP (0.749, 95% CI: 0.667–0.832), AFP-L3 (0.702, 95% CI: 0.618–0.785), DCP (0.790, 95% CI: 0.704–0.877), and the GALAD score (0.827, 95% CI: 0.753–0.901; P value = 0.001, DeLong’s test). Even the AUCs of the single genes OSR2 (0.938, 95% CI: 0.898–0.978) and TSPYL5 (0.911, 95% CI: 0.860–0.962) were greater than those of the GALAD score (Fig. 3C). When the specificity was set at 90.12–92.59%, the sensitivity of Qliver (88.06%, 95% CI: 77.82%–94.70%) for detecting HCC was greater than that of AFP (50.75%, 95% CI: 38.24%–63.18%) and the GALAD score (59.70%, 95% CI: 47.00%–71.51%). The sensitivity and specificity of Qliver were both greater than those of AFP-L3 (47.76%, 95% CI: 35.40%–60.33%; 83.95%, 95% CI: 74.12%–91.17%) and DCP (68.66%, 95% CI: 56.16%–79.44%; 76.54%, 95% CI: 65.82%–85.25%) (Table 1). The PPV (90.77%, 95% CI: 81.92%–95.52%) and NPV (90.36%, 95% CI: 82.99%–94.74%) of Qliver were both greater than those of the GALAD score (85.11%, 95% CI: 73.26%–92.26%; 73.27%, 95% CI: 67.02%–78.71%) and other protein biomarkers (Table 1).

When the plasma volume was reduced to 1 mL, Qliver still performed better than the other biomarkers in detecting early-stage HCC. For stage 0 + A stage HCC patients, the sensitivity of Qliver was 72.73% (95% CI: 39.03%–93.98%), whereas those of AFP, AFP-L3, DCP and the GALAD score were 9.09% (95% CI: 0.23%–41.28%), 18.18% (95% CI: 2.28%–51.78%), 54.55% (95% CI: 23.38%–83.25%) and 18.18% (95% CI: 2.28%–51.78%), respectively. For stage B and C HCC, the sensitivity of Qliver was still greater than those of these protein markers (Table 2). A comparison of biomarker performance in the Phase 2 plasma cohort is detailed in Additional file 1: Table S5.

Qliver outperforms protein biomarkers in the combined cohort

The MS‒qPCR data of the OSR2 and TSPYL5 genes from the Phase 1 plasma cohort were combined with those from the Phase 2 plasma cohort to further evaluate the performance of Qliver in detecting HCC in a larger sample size. The â–³Ct values of OSR2 and TSPYL5 were clearly clustered into two large groups between the HCC patients (n = 120) and the non-HCC controls (n = 203) (Fig. 3A). In the combined plasma cohort, the â–³Ct values of OSR2 and TSPYL5 in the HCC patients were significantly lower than those in the non-HCC patients (p < 0.05) (Fig. 3D, E). The Qliver score and the GALAD score of the HCC patients were both significantly greater than those of the healthy volunteers, patients with focal nodular hyperplasia (FNH), patients with hepatic hemangioma (HH) and cirrhosis patients (Cir) (p < 0.05), but Qliver performed better overall (Fig. 3F, G). There was a significant difference in the Qliver score, but not the GALAD score, between HCC patients and patients with hepatocellular adenoma (HCA) (Fig. 3F, G).

The sensitivity of Qliver in detecting HCC at BCLC stages 0 and A was 68.42% (95% CI: 43.45%–87.42%), which was superior to that of the GALAD score at 21.05% (95% CI: 6.05%–45.57%) (Table 2). Similar results were observed in other BCLC stages.

The AUC of Qliver was greater than that of the GALAD score in distinguishing HCC patients from healthy volunteers, cirrhosis patients, and individuals with benign liver disease (0.958, 95% CI: 0.934–0.983 vs. 0.849, 95% CI: 0.797–0.901; P < 0.001; 0.958, 95% CI: 0.938–0.978 vs. 0.810, 95% CI: 0.761–0.858, P < 0.001; 0.959, 95% CI: 0.930–0.988 vs. 0.777, 95% CI: 0.679–0.874, P < 0.001; DeLong’s test) (Fig. 4A‒C). Similar results were observed for AFP, AFP-L3 and DCP. For single-gene comparisons, OSR2 performed better than TSPYL5.

Fig. 4
figure 4

Comparison of the performance of Qliver and GALAD score in different subgroups and the possibility of using Qliver for HCC prognosis. A ROC curves and associated AUC values with 95% confidential interval for OSR2, TSPYL5, AFP, AFP-L3, DCP, GALAD and Qliver in combined plasma cohort (phase 1 plus phase 2) consisting of 120 HCC and 50 healthy plasma samples. B ROC curves and associated AUC values with 95% confidential interval for OSR2, TSPYL5, AFP, AFP-L3, DCP, GALAD and Qliver in combined plasma cohort consisting of 120 HCC and 50 cirrhosis plasma samples. C ROC curves and associated AUC values with 95% confidential interval for OSR2, TSPYL5, AFP, AFP-L3, DCP, GALAD and Qliver in combined plasma cohort, consisting of 120 HCC and 20 benign liver disease samples. D Comparison of positive and negative proportions of HCC detection by Qliver and GALAD score in HCC patients (n = 120) at different AFP concentrations. (< 20 Î¼g/L, 20 Î¼g/L ≤ AFP ≤ 400 Î¼g/L, > 400 Î¼g/L), E at different DCP concentrations (D < 40mAU/mL, and ≥ 40mAU/mL), F at BCLC 0/A/B/C stages, G at different cirrhosis history, H at different tumor size and (I) at different tumor count. Kaplan–Meier estimates of disease-free survival in HCC patients (n = 365) in the GDC TCGA Liver Cancer (LIHC) 450 K dataset, stratified by methylation level of OSR2 gene (J), TSPYL5 gene (K) and dual-target (L). The DFS outcomes between OSR2high /TSPYL5high and OSR2low /TSPYL5low groups were compared using the log-rank test

Full size image

The sensitivity and specificity of Qliver for detecting HCC were greater than those of AFP, AFP-L3, DCP and the GALAD score in patients with different liver cirrhosis histories, tumor sizes, tumor counts and tumor stages (Fig. 3H). The positive rate of Qliver in detecting HCC was higher than that of the GALAD score under different AFP and DCP concentrations, BCLC stages, liver cirrhosis histories, maximum tumor sizes and tumor counts (Fig. 4D-I). A detailed performance comparison of Qliver with these biomarkers in the combined plasma cohort is shown in Additional file 1: Table S5.

Evaluating the benefit of Qliver score for a 10 K intended-use population

To evaluate the performance of Qliver in the real world, a population of 10,000 cirrhosis patients was simulated. When the annual incidence of HCC was 2.1%, the PPV and NPV of Qliver were 20.33% (95% CI: 18.97%–21.76%) and 99.72% (95% CI: 99.60%–99.81%), respectively, which were higher than those of the GALAD score (12.87%, 95% CI: 11.50%–14.39%; 99.06%, 95% CI: 98.99%–99.20%) (Table 1). In addition, the PPV of Qliver was significantly greater than those of AFP, AFP-L3 and DCP.

The benefit of using the Qliver model for the intended user population was evaluated. When the prevalence of HCC in the cirrhotic population was 2.1%, the sensitivity of Qliver for detecting HCC was 88.33%, and the specificity was 90.64%, the harm/benefit ratio was ≤ 0.2024273, which means that in order to benefit one Qliver-positive case subject, 5 Qliver-positive control subjects should be tolerated to undergo unnecessary clinical measures such as ultrasound examination.

Potential applicatioin of Qliver score in HCC prognosis

The potential of Qliver in HCC prognosis was also explored in this study. By analyzing the GDC TCGA Liver Cancer (LIHC) 450 K dataset, we found that the methylation levels of some CpG sites in OSR2 and most CpG sites in TSPYL5 were greater in primary HCC tumors (n = 377) and recurrent tumors (n = 2) than in normal solid tissues (n = 59) (Fig. 2E, F). According to the GDC TCGA Liver Cancer (LIHC) gene expression RNA-seq dataset, the mRNA expression level of OSR2 in primary HCC tumors (n = 377) and recurrent HCC tumors (n = 2) was significantly greater than that in normal solid tissues (n = 59). However, the opposite results were observed for TSPYL5 (Fig. 2G).

The disease-free survival (DFS) of 365 HCC patients in the GDC TCGA Liver Cancer (LIHC) 450 K dataset was evaluated on the basis of the methylation levels of the OSR2 and TSPYL5 genes. Patients with higher OSR2 methylation levels had significantly worse DFS than those with lower methylation levels, with a hazard ratio (HR) of 1.393 (95% CI: 1.028–1.888) (p = 0.029; log-rank test) (Fig. 4J). However, there was no significant difference in DFS between HCC patients with high and low TSPYL5 methylation levels, with an HR of 1.317 (95% CI: 0.974–1.781) (p = 0.072; log-rank test) (Fig. 4K). Nevertheless, the DFS of HCC patients with high dual-gene methylation levels was significantly lower than that of patients with low dual-gene methylation levels. Moreover, double gene methylation was better than single gene methylation in predicting prognosis (p = 0.008; log-rank test, HR = 1.627, 95% CI: 1.127–2.350) (Fig. 4L). These findings indicate that hypermethylated OSR2 and TSPYL5 are promising markers for HCC prognosis.

Patients with higher methylation levels of OSR2 or TSPYL5 gene had a significantly lower DFS rate than those with low methylation levels across different age or AJCC stage. Moreover, combining the TSPYL5 and OSR2 genes, the DFS rate of HCC patients with high methylation levels of both genes was significantly lower than that of patients with low methylation levels (Additional file 3: Fig. S5).

Potential application of Qliver score in multi-cancer detection

The methylation levels of OSR2 and TSPYL5 were validated in multiple tumor cell lines and white blood cells (WBCs). The methylation levels of the OSR2 gene were significantly greater in most tumor cell lines than in leukocytes, especially in CAL-62 (thyroid), HCC-2279 (lung), MKN74 (stomach), MKN28 (stomach), EFM-19 (breast) and MCF7 (breast) cells (Additional file 3: Fig. S6). This finding suggests that the OSR2 gene may be an oncogene, so Qliver can be used for multicancer detection.

Discussion

Early detection is crucial for improving the 5-year survival rate of patients with various cancers, including liver cancer [18]. In this study, we identified two methylated genes, OSR2 and TSPYL5, for HCC detection and developed an MS‒qPCR assay named Qliver. We compared the performance of Qliver for detecting HCC with that of existing surveillance methods with respect to cirrhosis history, tumor size, tumor count, and tumor stage. Qliver performed best in HCC detection, followed by the GALAD score. The performance of Qliver far exceeded that of AFP, AFP-L3 and DCP. In cirrhosis patients, subjects with benign liver diseases and healthy individuals as controls, the pooled sensitivity of Qliver for detecting HCC using 2 mL of plasma was 88.68% (95% CI: 76.97–95.73%), with a specificity of 89.34% (95% CI: 82.47–94.20%), which was greater than the sensitivity of the GALAD score of 58.49% (95% CI: 44.13–71.86%) at the same specificity. When cirrhosis patients were used as controls, the pooled sensitivity of Qliver for HCC detection with 1 mL of plasma was still greater than that of the GALAD score, with almost the same specificity (88.06%, 95% CI: 77.82%–94.70% vs. 59.70%, 95% CI: 47.00%–71.51%). For early HCC detection in cirrhosis patients, the sensitivity of Qliver for detecting BCLC 0 and A HCC using 1 mL of plasma was significantly greater than that of the GALAD score (72.73%, 95% CI: 39.03%–93.98% vs. 18.18%, 95% CI: 2.28%–51.78%). In previous studies, the GALAD score performed well for detecting any stage of HCC, with a pooled sensitivity, specificity, and AUC of 0.82 (95% CI: 0.78–0.85), 0.89 (95% CI: 0.85–0.91), and 0.92 (95% CI: 0.89–0.94), respectively [2]. The GALAD score for identifying BCLC 0/A HCC has a moderate sensitivity of 0.73 (95% CI: 0.66–0.79) and a high specificity of 0.87 (95% CI: 0.81–0.91) [2]. The performance of the GALAD score in this study appeared to be suboptimal, possibly due to the unofficial configuration of the assay reagents and equipment for AFP, AFP-L3, and DCP compared with that of Roche's Elecsys GALAD assay and the use of plasma instead of serum. Another possible reason may be that most HCC patients in this study had chronic HBV infection, but the GALAD score has a higher sensitivity in the HCC subgroups with HCV or nonviral liver diseases [2]. Therefore, compared with existing surveillance methods, Qliver holds great promise for the early detection of HCC caused by HBV infection in the Chinese population, and can supplement the deficiencies of US and AFP in early HCC detection. However, this potential of Qliver needs to be further fully validated in a larger population.

In this study, the performance of single methylated genes, OSR2 and TSPYL5, for HCC detection was better than that of single tumor proteins including AFP, AFP-L3 and DCP. Even the performance of the GALAD score when these three proteins were integrated was lower than that of the two single methylated genes. The AUCs of OSR2 and TSPYL5 for distinguishing HCC patients from healthy individuals, cirrhosis patients and patients with benign liver diseases were greater than 0.94 and 0.90, respectively, while those of the GALAD score were less than 0.85. Gene methylation should be more suitable for the early detection of cancer than other biomarkers, such as tumor proteins, ctDNA mutations, copy number variations, and ctDNA fragmentomic features. Hypermethylation of tumor suppressor genes is likely to be the earliest event in carcinogenesis [10], and methylated CpG sites in gene promoter regions are clustered [13]. Therefore, DNA methylation has increased sensitivity for detecting early cancers. The average detection rates of early-stage HCC were only 56.38%, 58.84%, and 33.55%, respectively, using gene methylation, ctDNA mutation and genome-wide cfDNA fragmentation profiles obtained via low-coverage WGS [19]. Although the average detection rate of DNA methylation is not better than that of ctDNA mutations, the detection rate of single gene methylation in early-stage HCC can reach 100%, whereas that of single gene mutations is only 59% [19]. Therefore, identifying methylated genes with excellent performance is very important for early HCC detection.

A meta-analysis of 33 eligible articles from 4113 patients suggested that RASSF1A methylation in ctDNA could be used as a potential biomarker for HCC screening, with a sensitivity of 0.644 (95% CI: 0.608–0.678) and a specificity of 0.875 (95% CI: 0.847–0.900) [20]. Oussalah et al. [21] reported that SEPT9 methylation exhibited high diagnostic accuracy for HCC, with an AUC of 0.944 (95% CI: 0.900–0.970), but the sensitivity was 87.88% (95% CI: 71.8%–96.6%) and the specificity was 67.54% (95% CI: 60.4%–74.1%) when ≥ 1 of the triplicate samples was positive. Owing to the heterogeneity of hepatocellular carcinoma, the diagnostic sensitivity of single gene methylation is low. Multitarget panels are expected to improve the sensitivity of HCC detection. Chalasani et al. [22] developed a multitarget panel consisting of four methylated DNAs (HOXA1, EMX1, TSPYL5 and B3GALT6) and two protein markers (AFP and AFP-L3), which had a high sensitivity of 71% (95% CI: 60%–81%) at 90% specificity for early-stage HCC detection, which was higher than those of the GALAD score (41%, 95% CI: 30–53%) and AFP concentration ≥ 7.32 ng/mL (45%, 95% CI: 33–57%). Luo et al. [23] proposed an HCC screening model based on 2321 methylation markers, which achieved 84% sensitivity and 96% specificity in an independent validation cohort with an AUC of 0.934 (95% CI: 0.905–0.963) for distinguishing early-stage HCC patients from high-risk individuals. Xu et al. [12] constructed a model combining 10 CpG sites with a sensitivity of 85.7% and a specificity of 94.3% for detecting HCC in the training dataset and a sensitivity of 83.3% and a specificity of 90.5% in the validation dataset. In this study, only two methylation genes using 1 mL of plasma were required to achieve an AUC of 0.958 (95% CI: 0.927–0.989) to distinguish HCC patients from cirrhosis patients, with a sensitivity of 72.73% (95% CI: 39.03%–93.98%) and a specificity of 90% for detecting BCLC 0 and A stage HCC. Therefore, Qliver, with its low cost, convenient experimental procedures, and superior performance, is more effective for HCC screening compared to targeted sequencing and WGS methods involving hundreds of genes.

The methylation levels of OSR2 and TSPYL5 were validated in a variety of tumor cell lines and white blood cells, and the methylation level of the OSR2 gene in most tumor cell lines was significantly greater than that in white blood cells, especially CAL-62 (thyroid), HCC-2279 (lung), MKN74 (stomach), MKN28 (stomach), EFM-19 (breast) and MCF7 (breast) cells, suggesting that OSR2 may be a multiple-cancer biomarker. OSR2 is a mammalian homolog of the Drosophila odd-skipped family of transcription factors [24]. OSR2 methylation has been used to detect other cancers, such as gastric cancer [25] and oropharyngeal squamous cell carcinomas [26]. The methylation level of OSR2 in HCC primary tumors and recurrent tumors was greater than that in normal solid tissues, and its mRNA expression level was also significantly greater in HCC primary tumors and recurrent HCC tumors than in normal solid tissues, suggesting that OSR2 may be an oncogene. OSR2 has been shown to play an important role in cell proliferation and development [27]. Wen et al. [28] provided evidence that OSR2 promotes prostate cancer tumorigenesis. Han et al. [29] knocked down OSR2 in human adenocarcinoma (H838) cells and found that cell proliferation was significantly inhibited compared with the non-targeting siRNA group, suggesting that OSR2 may have an oncogenic role in lung cancer. TSPYL5, a member of the nucleosome assembly protein (NAP) superfamily, is likely a tumor suppressor gene in ovarian, lung, and colorectal cancers according to several studies [30, 31]. The higher methylation level but lower mRNA expression in HCC primary tumors and recurrent tumors than in normal solid tissues also supports the role of TSPYL5 as a tumor suppressor gene. The combination of OSR2 as an oncogene and TSPYL5 as a tumor suppressor gene has the potential to detect multiple cancers but requires rigorous and adequate validation.

Qliver’s potential for HCC prognosis was also explored in this study. By mining the GDC TCGA Liver Cancer (LIHC) 450 K dataset and the GDC TCGA Liver Cancer (LIHC) gene expression RNA-seq dataset, hypermethylated OSR2 was negatively correlated with the disease-free survival (DFS) of HCC patients (p = 0.029), and when combined with hypermethylated TSPYL5, this correlation was further increased (p = 0.008). These findings suggest that Qliver has good potential for predicting HCC prognosis, but adequate validation in clinical samples is needed, especially in the Chinese population.

The ideal internal reference genes are stably expressed under any experimental conditions, but many studies have shown that genes stably expressed in different species or under different conditions will change [32]. Therefore, screening appropriate reference genes under specific conditions is highly important. Genes such as ACTB, B2M, GAPDH, 18S rRNA and 28S rRNA are commonly used as internal reference genes in qPCR. However, the stability of these classical reference genes has been questioned in recent years [33, 34]. We found that the MS‒qPCR amplification efficiency of ACTB was lower than that of highly methylated target genes during technical validation in tissues. This may be because the DNA sequence of ACTB is hypomethylated and the bisulfite-converted sequence is AT-rich, which is not conducive to PCR amplification. In this study, we identified for the first time a novel reference gene, SDF4, which has a stable high methylation level and high MS‒qPCR amplification efficiency compared with ACTB, which can improve the diagnostic performance of methylated target genes. We believe that these findings will improve the analytical performance of bisulfite conversion-based MS‒qPCR for cancer detection and other applications.

In addition, deconvolution of ctDNA from highly heterogeneous and noisy backgrounds is essential for translating ctDNA methylation data into accurate and effective noninvasive cancer markers. This is especially true for early and/or less aggressive cancers [35]. Lehmann-Werman et al. [36] demonstrated the superior sensitivity of multiple CpG haplotypes in detecting tissue-specific features in cfDNA. Guo et al. [37] used methylated haplotypes for quantitative estimations of tumor load and tissue origin profiles in circulating cell-free DNA from 59 patients with cancer, suggesting that plasma methylated haplotyping is an important tool for the early detection of tumors and their major growth sites and that it is a promising strategy for the early detection of tumors.

This study has several limitations. First, the sample sizes of the Phase 1 plasma cohort and Phase 2 plasma cohort were relatively small, and the inclusion of more patients with early-stage HCC (BCLC stage 0‒A) will improve the robustness of the Qliver model. Second, although our results were encouraging, they were based on a single-center retrospective case; a prospective multicenter study should be organized to independently validate our findings.

Conclusions

In this study, we identified two methylated genes, OSR2 and TSPYL5, for HCC detection using a pipeline that included tissue discovery and plasma validation, tissue technical validation and plasma biological validation. A novel reference gene, SDF4, was also identified that outperformed ACTB in improving the diagnostic performance of the bisulfite-converted MS‒qPCR assay of the target methylated genes for cancer detection. An MS‒qPCR assay named Qliver containing OSR2, TSPYL5 and SDF4 was subsequently developed and validated for the detection of HCC in an independent plasma cohort. Qliver outperformed existing surveillance methods, such as AFP, AFP-L3, DCP and the GALAD score. The potential of Qliver in HCC prognosis was also explored in this study, and the analysis of the methylation and mRNA expression databases revealed that it has a strong ability to predict HCC prognosis; however, this finding needs to be fully validated in clinical samples. OSR2 is highly methylated in many cancer cell lines compared with leukocytes, suggesting that it may be a multicancer biomarker. Qliver, which combines a possible oncogene, OSR2, and a possible tumor suppressor gene, TSPYL5, may have great potential in detecting multiple cancers.

Data availability

The original contributions data presented in the current study are included in the manuscript/supplementary material; further inquiries are available from the corresponding author upon reasonable request. The 450K methylation, mRNA expression and survival data were downloaded from UCSC Xena Hub repository, https://xenabrowser.net/. Targeted bisulfite sequencing data from the tissue discovery cohorts were deposited into the Sequence Read Archive (SRA) under accession number SRP577656, https://www.ncbi.nlm.nih.gov/sra/SRP577656.

Abbreviations

HCC:

Hepatocellular carcinoma

AFP:

Alpha-fetoprotein

AFP-L3:

Lens culinaris agglutinin-reactive fraction of AFP

DCP:

Des-gamma-carboxyprothrombin

cfDNA:

Circulating cell-free DNA

ctDNA:

Circulating tumor DNA

MS-qPCR:

Methylation-specific quantitative PCR

FFPE:

Formalin-fixed paraffin-embedded

gDNA:

Genomic DNA

ROC:

Receiver operating characteristic curve

AUC:

Area under the ROC curve

CI:

Confidence interval

GLM:

Generalized linear model

TCGA-LIHC:

The Cancer Genome Atlas of Liver Hepatocellular Carcinoma

CNGBdb:

China National GeneBank DataBase

PPV:

Positive predictive value

NPV:

Negative predictive value

WBCs:

White blood cells

FNH:

Focal nodular hyperplasia

HH:

Hepatic hemangioma

Cir:

Cirrhosis

HCA:

Hepatocellular adenoma

HR:

Hazard ratio

References

  1. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–63.

    Article  PubMed  Google Scholar 

  2. Guan MC, Zhang SY, Ding Q, Li N, Fu TT, Zhang GX, et al. The performance of GALAD score for diagnosing hepatocellular carcinoma in patients with chronic liver diseases: a systematic review and meta-analysis. J Clin Med. 2023;12(3): 949.

    Article  PubMed  PubMed Central  Google Scholar 

  3. Singal AG, Pillai A, Tiro J. Early detection, curative treatment, and survival rates for hepatocellular carcinoma surveillance in patients with cirrhosis: a meta-analysis. PLoS Med. 2014;11(4): e1001624.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Tzartzeva K, Obi J, Rich NE, Parikh ND, Marrero JA, Yopp A, et al. Surveillance imaging and alpha fetoprotein for early detection of hepatocellular carcinoma in patients with cirrhosis: a meta-analysis. Gastroenterology. 2018;154(6):1706-1718.e1.

    Article  PubMed  CAS  Google Scholar 

  5. Esfeh JM, Hajifathalian K, Ansari-Gilani K. Sensitivity of ultrasound in detecting hepatocellular carcinoma in obese patients compared to explant pathology as the gold standard. Clin Mol Hepatol. 2020;26(1):54–9.

    Article  PubMed  Google Scholar 

  6. Ayoub WS, Steggerda J, Yang JD, Kuo A, Sundaram V, Lu SC. Current status of hepatocellular carcinoma detection: screening strategies and novel biomarkers. Ther Adv Med Oncol. 2019;11: 1758835919869120.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  7. Yang JD, Addissie BD, Mara KC, Harmsen WS, Dai J, Zhang N, et al. GALAD score for hepatocellular carcinoma detection in comparison with liver ultrasound and proposal of GALADUS score. Cancer Epidemiol Biomarkers Prev. 2019;28(3):531–8.

    Article  PubMed  Google Scholar 

  8. Jones PA. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat Rev Genet. 2012;13(7):484–92.

    Article  PubMed  CAS  Google Scholar 

  9. Hata T, Dal Molin M, Hong SM, Tamura K, Suenaga M, Yu J, et al. Predicting the grade of dysplasia of pancreatic cystic neoplasms using cyst fluid DNA methylation markers. Clin Cancer Res. 2017;23(14):3935–44.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  10. Newsham I, Sendera M, Jammula SG, Samarajiwa SA. Early detection and diagnosis of cancer with interpretable machine learning to uncover cancer-specific DNA methylation patterns. Biol Methods Protoc. 2024;9(1):bpae028.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  11. Aucamp J, Bronkhorst AJ, Badenhorst CPS, Pretorius PJ. The diverse origins of circulating cell-free DNA in the human body: a critical re-evaluation of the literature. Biol Rev Camb Philos Soc. 2018;93(3):1649–83.

    Article  PubMed  Google Scholar 

  12. Xu RH, Wei W, Krawczyk M, Wang W, Luo H, Flagg K, et al. Circulating tumour DNA methylation markers for diagnosis and prognosis of hepatocellular carcinoma. Nat Mater. 2017;16(11):1155–61.

    Article  PubMed  CAS  Google Scholar 

  13. Lakshminarasimhan R, Liang G. The role of DNA methylation in cancer. Adv Exp Med Biol. 2016;945:151–72.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  14. Guo P, Zheng H, Li Y, Li Y, Xiao Y, Zheng J, et al. Hepatocellular carcinoma detection via targeted enzymatic methyl sequencing of plasma cell-free DNA. Clin Epigenetics. 2023;15(1):2.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  15. Kim SC, Kim DW, Cho EJ, Lee JY, Kim J, Kwon C, et al. A circulating cell-free DNA methylation signature for the detection of hepatocellular carcinoma. Mol Cancer. 2023;22(1):164.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  16. Hajian-Tilaki K. Sample size estimation in diagnostic test studies of biomedical informatics. J Biomed Inform. 2014;48:193–204.

    Article  PubMed  Google Scholar 

  17. Fan R, Chen L, Zhao S, Yang H, Li Z, Qian Y, et al. Novel, high accuracy models for hepatocellular carcinoma prediction based on longitudinal data and cell-free DNA signatures. J Hepatol. 2023;79(4):933–44.

    Article  PubMed  CAS  Google Scholar 

  18. Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin. 2023;73(1):17–48.

    Article  PubMed  Google Scholar 

  19. Manea I, Iacob R, Iacob S, Cerban R, Dima S, Oniscu G, et al. Liquid biopsy for early detection of hepatocellular carcinoma. Front Med (Lausanne). 2023;10:1218705.

    Article  PubMed  Google Scholar 

  20. Zhang Z, Chen P, Xie H, Cao P. Using circulating tumor DNA as a novel biomarker to screen and diagnose hepatocellular carcinoma: a systematic review and meta-analysis. Cancer Med. 2020;9(4):1349–64.

    Article  PubMed  CAS  Google Scholar 

  21. Oussalah A, Rischer S, Bensenane M, Conroy G, Filhine-Tresarrieu P, Debard R, et al. Plasma mSEPT9: a novel circulating cell-free DNA-based epigenetic biomarker to diagnose hepatocellular carcinoma. EBioMedicine. 2018;30:138–47.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  22. Chalasani NP, Ramasubramanian TS, Bhattacharya A, Olson MC, Edwards VDK, Roberts LR, et al. A novel blood-based panel of methylated DNA and protein markers for detection of early-stage hepatocellular carcinoma. Clin Gastroenterol Hepatol. 2021;19(12):2597-2605.e4.

    Article  PubMed  CAS  Google Scholar 

  23. Luo B, Ma F, Liu H, Hu J, Rao L, Liu C, et al. Cell-free DNA methylation markers for differential diagnosis of hepatocellular carcinoma. BMC Med. 2022;20(1):8.

    Article  PubMed  PubMed Central  Google Scholar 

  24. Lan Y, Ovitt CE, Cho ES, Maltby KM, Wang Q, Jiang R. Odd-skipped related 2 (Osr2) encodes a key intrinsic regulator of secondary palate growth and morphogenesis. Development. 2004;131(13):3207–16.

    Article  PubMed  CAS  Google Scholar 

  25. Li WH, Zhou ZJ, Huang TH, Guo K, Chen W, Wang Y, et al. Detection of OSR2, VAV3, and PPFIA3 methylation in the serum of patients with gastric cancer. Dis Markers. 2016;2016:5780538.

    Article  PubMed  PubMed Central  Google Scholar 

  26. Kostareli E, Holzinger D, Bogatyrova O, Hielscher T, Wichmann G, Keck M, et al. HPV-related methylation signature predicts survival in oropharyngeal squamous cell carcinomas. J Clin Invest. 2013;123(6):2488–501.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  27. Kawai S, Abiko Y, Amano A. Odd-skipped related 2 regulates genes related to proliferation and development. Biochem Biophys Res Commun. 2010;398(2):184–90.

    Article  PubMed  CAS  Google Scholar 

  28. Wen P, Lei H, Deng H, Deng S, Rodriguez Tirado C, Wang M, et al. Hyd/UBR5 defines a tumor suppressor pathway that links Polycomb repressive complex to regulated protein degradation in tissue growth control and tumorigenesis. Genes Dev. 2024;38(13–14):675–91.

    PubMed  PubMed Central  CAS  Google Scholar 

  29. Han Q, Fernandez J, Rajczewski AT, Kono TJY, Weirath NA, Rahim A, et al. A multi-omics study of epigenetic changes in type II alveolar cells of A/J mice exposed to environmental tobacco smoke. Int J Mol Sci. 2024;25(17): 9365.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  30. Huang C, He C, Ruan P, Zhou R. TSPYL5 activates endoplasmic reticulum stress to inhibit cell proliferation, migration and invasion in colorectal cancer. Oncol Rep. 2020;44(2):449–56.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  31. Huang C, Ruan P, He C, Zhou R. TSPYL5 inhibits the tumorigenesis of colorectal cancer cells in vivo by triggering DNA damage. J Cancer Res Ther. 2023;19(4):898–903.

    Article  PubMed  CAS  Google Scholar 

  32. Ferreira MJ, Silva J, Pinto SC, Coimbra S. I choose you: selecting accurate reference genes for qPCR expression analysis in reproductive tissues in Arabidopsis thaliana. Biomolecules. 2023;13(3): 463.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  33. Ahn HR, Baek GO, Yoon MG, Son JA, You D, Yoon JH, et al. HMBS is the most suitable reference gene for RT-qPCR in human HCC tissues and blood samples. Oncol Lett. 2021;22(5):791.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  34. Gao Q, Wang XY, Fan J, Qiu SJ, Zhou J, Shi YH, et al. Selection of reference genes for real-time PCR in human hepatocellular carcinoma tissues. J Cancer Res Clin Oncol. 2008;134(9):979–86.

    Article  PubMed  CAS  Google Scholar 

  35. Hong S, Lin B, Xu M, Zhang Q, Huo Z, Su M, et al. Cell-free DNA methylation biomarker for the diagnosis of papillary thyroid carcinoma. EBioMedicine. 2023;90: 104497.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  36. Lehmann-Werman R, Neiman D, Zemmour H, Moss J, Magenheim J, Vaknin-Dembinsky A, et al. Identification of tissue-specific cell death using methylation patterns of circulating DNA. Proc Natl Acad Sci U S A. 2016;113(13):E1826–34.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  37. Guo S, Diep D, Plongthongkum N, Fung HL, Zhang K, Zhang K. Identification of methylation haplotype blocks aids in deconvolution of heterogeneous tissue samples and tumor tissue-of-origin mapping from plasma DNA. Nat Genet. 2017;49(4):635–42.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

This study was sponsored by the Natural Science Foundation of Fujian Province (2023J05234, 2023J011297), Joint Funds for the Innovation of Science and Technology, Fujian province (2021Y9232, 2021Y9227, 2024Y9620), high level talents training project of Fujian Cancer Hospital (2022YNG01), and Young and Middle-aged Scientific Research Major Project of Fujian Provincial Health Commission (2022ZQNZD009).

Author information

Author notes

  1. Wenhao Teng, Hui Li and Hao Yang contributed equally to this work.

Authors and Affiliations

  1. Department of Hepatopancreatobiliary Surgery, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, China

    Wenhao Teng, Fuli Xin, Lihui Yu, Ming Wang, Fayong Ke, Hui Zhang & Jingfeng Liu

  2. Fujian Provincial Key Laboratory of Tumor Biotherapy, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, China

    Wenhao Teng

  3. Berry Oncology Corporation, Beijing, China

    Hui Li, Liying Xi, Aiyuan Zhang, Lu Zheng, Jian Bai, Yin Wang & Lin Wu

  4. Fujian Key Laboratory of Advanced Technology for Cancer Screening and Early Diagnosis, Fuzhou, China

    Hui Li, Hao Yang, Liying Xi, Aiyuan Zhang, Lu Zheng, Jian Bai, Yin Wang, Fuming Sun, Lin Wu & Jingfeng Liu

  5. Genetrix Biotech Corporation, Beijing, China

    Hao Yang & Fuming Sun

  6. Department of Medical Oncology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, China

    Yu Chen

Authors
  1. Wenhao Teng
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Hui Li
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Hao Yang
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Yu Chen
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Liying Xi
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Fuli Xin
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Aiyuan Zhang
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Lihui Yu
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Lu Zheng
    View author publications

    You can also search for this author inPubMed Google Scholar

  10. Ming Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  11. Jian Bai
    View author publications

    You can also search for this author inPubMed Google Scholar

  12. Fayong Ke
    View author publications

    You can also search for this author inPubMed Google Scholar

  13. Yin Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  14. Fuming Sun
    View author publications

    You can also search for this author inPubMed Google Scholar

  15. Hui Zhang
    View author publications

    You can also search for this author inPubMed Google Scholar

  16. Lin Wu
    View author publications

    You can also search for this author inPubMed Google Scholar

  17. Jingfeng Liu
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

J.L., L.W., H.Z., W.T., H.L. and H.Y. contributed to the conception and design of the work. W.T., Y.C., F.X., L.Y., M.W. and F.K. collected samples and clinical information. H.L., H.Y., L.X., A.Z., L.Z., J.B., Y.W. and F.S. conducted data analysis and interpretation; W.T., H.L. and H.Y. drafted the manuscript. J.L., L.W. and H.Z. reviewed the data and revised the manuscript. All authors read and approved the final manuscript.

Corresponding authors

Correspondence to Hui Zhang, Lin Wu or Jingfeng Liu.

Ethics declarations

Ethics approval and consent to participate

All procedures were approved by Medical Ethics Committee of the Fujian Cancer Hospital (K2022-103–01). All research was conducted in accordance with the Declarations of Helsinki. Written consent was obtained from all subjects.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

12916_2025_4115_MOESM1_ESM.xlsx

Additional file 1. Table S1-S6. Table S1- [Patient demographics and clinical information included in the development of the Qliver model]. Table S2- [Characteristics of the 21 candidate methylation blocks in discovery stage]. Table S3- [Comparison of the performance of all HCC Patients with non-HCC subjects]. Table S4- [Characteristics of the dual methylation markers and their coefficients in HCC diagnosis]. Table S5- [Performances of the Qliver detection model on different subgroups based on clinical characteristics]. Table S6- [Primer and Probe Sequences used to amplify of OSR2, TSPYL5 and SDF4].

12916_2025_4115_MOESM2_ESM.docx

Additional file 2. Detailed experimental workflow for HCC methylation biomarker development: from sample processing to gene validation.

12916_2025_4115_MOESM3_ESM.zip

Additional file 3. Fig. S1-S6. Fig. S1- [Fig. S1 Visual representation of HCC-specific methylation haplotypes from Targeted Methylation Sequencing. Visual representation of methylation profiles between 43 primary HCC and 32 adjacent normal tissue samples focus on the OSR2(A), KCNG3(B), PITX1(C), C1QL4(D), IRX5(E) and TSPYL5(F). Each row in the representation signifies the methylation status of a single sample, while each column represents a different CpG locus. The color intensity depicted in the figure corresponds to the different degrees of methylation at these loci]. Fig. S2- [Fig. S2 Visual depictions illustrating the consistent hypermethylation status of SDF4 across various samples. Visual representation of methylation profiles between 43 primary HCC and 32 adjacent normal tissue samples focus on the Homo sapiens stromal cell-derived factor 4 (SDF4). SDF4 encodes a stromal cell-derived factor belonging to the CREC protein family, identified as the most consistently conserved reference genes in tissues samples from both HCC patients and non-HCC subjects. Each row within the representation signifies the methylation status of an individual sample, while each column represents a distinct CpG locus. The color intensity depicted in the plot corresponds to the varying degrees of methylation at these loci]. Fig. S3- [Fig. S3 Identification of a Novel Reference Gene for MS-qPCR. (A) Heatmap of methylation levels of CpGs in SDF4 gene for discriminating Primary HCC tumor (n = 377), Recurrent HCC tumor (n = 2) and Solid Tissue Normal (n = 50) in the GDC TCGA Liver Cancer (LIHC) 450 K dataset. (B) ROC curves and associated AUC values with 95% confidential interval for ZIC4 as target gene which normalized by ACTB and SDF4 reference gene in archived plasma cohort, which consists with 22 HCC, 23 cirrhosis and 23 healthy plasma samples by MS-qPCR. The receiver operating characteristic curve analysis indicated that SDF4 (AUC = 0.826) might be an optimal reference gene for normalization of MS-qPCR data in liver cancer, which show higher HCC detection rate than ACTB (AUC = 0.786) with p-value 0.06822 use DeLong's test. (C) The distribution of ΔCT value of ACTB and SDF4 in archived FFPE samples, which consists with 38 primary HCC tissue and 38 adjacent normal tissue samples. ADJ, adjacent normal tissue; HCC, hepatocellular carcinoma. (D) The distribution of ΔCT value of ACTB and SDF4 in archived FFPE samples, which consists with 12 breast cancer and 20 lung cancer tissue samples (p-value were computed with Wilcoxon rank sum test. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001; p > 0.05 was considered not significant (ns).). (E) Correlation between the CT value of SDF4 and ACTB in WBC DNA (n = 7, Spearman’s rank correlation rho). (F) Correlation between the CT value of SDF4 and ACTB with DNA input amount in WBC DNA (n = 7, Spearman’s rank correlation rho). (G) The distribution of methylation levels of CpGs in SDF4 gene for pancancer tissues in the GDC PanCancer (PANCAN) 450 K dataset.]. Fig. S4- [Fig. S4 Visualization of HCC-specific methylation haplotype of OSR2 and TSPYL5 determined by Sanger sequencing. Visualization of methylation haplotype determined by Sanger sequencing within OSR2 gene in 8 primary HCC and 9 adjacent normal tissue samples (A) and TSPYL5 (B) gene in 6 primary HCC and 6 adjacent normal tissue samples. These tissues were also used for tissue discovery and tissue technical validation. Allele-specific methylation status from the Sanger data clearly showed that methylated alleles in OSR2 and TSPYL5 gene were associated with HCC, highlighting that DNA methylation status in these regions might be a novel biomarker for HCC detection.]. Fig. S5- [Fig. S5 Methylated OSR2 and TSPYL5 as Prognostic Biomarkers. The DFS rate of HCC patients from the TCGA 450 K dataset was assessed based on the methylation levels of the OSR2(A) and TSPYL5(B) genes. Patients with higher methylation levels of OSR2 or TSPYL5 gene had a significantly lower DFS rate than those with low methylation levels across different age or AJCC stage. Moreover, combining the TSPYL5 and OSR2 genes(C), the DFS rate of HCC patients with high methylation levels of both genes was significantly lower than that of patients with low methylation levels. The findings indicate that the presence of methylated OSR2 and TSPYL5 holds promise as prognostic indicators for individuals diagnosed with hepatocellular carcinoma]. Fig. S6- [Fig. S6 Methylation status of OSR2 and TSPYL5 gene in cancer cell lines and WBCs. For an initial assessment of the methylation status of the OSR2 and TSPYL5 genes, MS-qPCR was performed on cancer cell lines from uterine cancer, leukemia, thyroid cancer, liver cancer, stomach cancer, neuroblastoma, lung cancer, glioma, pancreatic cancer, esophageal squamous cell, colon cancer, ovarian cancer, submandibular, bladder cancer, breast cancer, lymphoma, prostate cancer, glioma, duodenal adenocarcinoma, melanoma, myeloma, teratoma and WBCs. Most of cancer cell lines were either completely or partially methylated in the region of OSR2 gene, highlighting that DNA methylation status in these regions might be a novel biomarker for multiple cancer or pan-cancer detection].

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Teng, W., Li, H., Yang, H. et al. Discovery and validation of a novel dual-target blood test for the detection of hepatocellular carcinoma across stages from cirrhosis. BMC Med 23, 278 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12916-025-04115-w

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12916-025-04115-w

Keywords

BMC Medicine

ISSN: 1741-7015

Contact us