Integrated diagnostic model based on serum exosomal microRNAs and functional mechanism of miR-200b-3p in hepatocellular carcinoma

Abstract

BACKGROUND

The early detection of hepatocellular carcinoma (HCC) remains a major clinical challenge due to the limited sensitivity and specificity of current biomarkers. This study sought to develop a diagnostic model based on serum exosomal microRNAs (miRNAs) and to elucidate the functional mechanism of a key miRNA, miR-200b-3p, in HCC pathogenesis.

AIM

To develop and validate a serum exosomal miRNA-based diagnostic model for HCC.

METHODS

Serum exosomes were isolated from a discovery cohort (68 HCC, 131 non-HCC) and a validation cohort (66 HCC, 135 non-HCC). Candidate miRNAs were identified via next-generation sequencing and machine learning, and validated by quantitative real-time polymerase chain reaction. A diagnostic model was constructed using logistic regression. The role of miR-200b-3p was examined using in vitro functional assays.

RESULTS

A diagnostic model incorporating three exosomal miRNAs (miR-200b-3p, miR-215-5p, miR-452-5p) and age was established. It showed strong performance in the training set (area under the curve = 0.863, sensitivity = 86.4%, specificity = 76.5%; all P < 0.001). Mechanistically, miR-200b-3p acted as a tumor suppressor by directly targeting phosphoserine aminotransferase 1. Its overexpression inhibited HCC cell proliferation, migration, and invasion (all P < 0.05), while knockdown promoted these phenotypes (all P < 0.05).

CONCLUSION

This study establishes and validates a non-invasive exosomal miRNA-based diagnostic model for HCC, reveals that miR-200b-3p directly targets phosphoserine aminotransferase 1, and provides mechanistic support for early detection.

Key Words: Hepatocellular carcinoma; Exosomal microRNAs; Diagnostic model; MicroRNA-200b-3p; Phosphoserine aminotransferase 1; Liquid biopsy

Core Tip: This study developed and validated a novel three-exosomal-miRNA (miR-200b-3p, miR-215-5p, miR-452-5p) and age-based diagnostic model for hepatocellular carcinoma (HCC) using a rigorous multi-cohort design. The model demonstrated high accuracy with an area under the curve of 0.863, offering a robust non-invasive tool for early detection. Furthermore, we experimentally identified phosphoserine aminotransferase 1 as a direct functional target of miR-200b-3p, establishing its tumor-suppressive role in HCC. Our work provides both a clinically promising liquid biopsy biomarker panel and a key mechanistic insight, advancing the field of precision diagnosis for HCC.

INTRODUCTION

As the sixth most frequently diagnosed malignancy and the third leading cause of cancer-associated mortality globally, hepatocellular carcinoma (HCC) poses a significant burden on public health worldwide. The epidemiological burden is disproportionately high in East Asia, particularly in China; indeed, China alone accounts for approximately half of all global cases and fatalities[1,2]. The poor prognosis of HCC patients is primarily attributable to late-stage diagnosis, as the disease is often asymptomatic in its early phases, limiting curative treatment options[3]. This underscores the critical need for highly sensitive and specific non-invasive biomarkers for early detection.

The current diagnostic algorithm for HCC integrates radiological imaging with serological biomarker testing. Although diagnostic mainstays including ultrasound, computed tomography, and magnetic resonance imaging play a central role in clinical assessment, their ability to characterize small hepatic nodules (< 2 cm) can be ambiguous, often requiring invasive biopsy for confirmation[4]. For decades, serum alpha-fetoprotein (AFP) has been the most widely used biomarker; however, its clinical utility is limited by suboptimal diagnostic accuracy (sensitivity: 40%-65%) and poor specificity, since AFP levels can also be elevated in benign chronic liver diseases[5,6]. Although other biomarkers like AFP-L3 and protein induced by vitamin K absence or antagonist-II (PIVKA-II) have been investigated and show improved performance in combination with AFP[7,8], even multi-marker panels lack the desired accuracy for reliable early-stage detection across diverse patient populations[9].

Liquid biopsy, particularly the analysis of extracellular vesicles such as exosomes, offers a promising non-invasive approach for cancer diagnosis[10]. Exosomes are secreted by virtually all cells and carry a rich cargo of proteins and nucleic acids reflective of their cell of origin[11]. Critically, their lipid bilayer membrane confers exceptional stability upon these molecules in the circulation, making exosomes ideal vessels for biomarker discovery[12]. Among the cargo, microRNAs (miRNAs) are of particular interest due to their stability and known associations with cancer pathogenesis[13,14]. Previous studies have indeed demonstrated the potential of serum exosomal miRNAs in distinguishing HCC from healthy controls and patients with non-malignant liver diseases[15-17]. However, the specific miRNA signatures identified often show limited overlap across different studies, and few have been successfully translated into robust, clinically applicable diagnostic models. Among the myriad of miRNAs implicated in cancer, miR-200b-3p is a notable member of the miR-200 family, which is widely recognized for its role in regulating epithelial-to-mesenchymal transition[18], it has been reported in various cancers, including HCC, suggesting its potential importance in hepatocarcinogenesis[19].

Therefore, this study was designed to identify a novel, stable serum exosomal miRNA signature for HCC through comprehensive screening and validation. We placed particular focus on miR-200b-3p due to its preliminary significance, aiming to assess its diagnostic value and elucidate its functional role. The ultimate goal was to develop and rigorously validate a non-invasive diagnostic model, integrating this miRNA signature with clinical characteristics, to accurately differentiate HCC patients from non-HCC individuals, thereby addressing the urgent need for more effective early detection strategies.

MATERIALS AND METHODS

Study population and sample collection

A total of 403 adult participants (age ≥ 18 years) were recruited from the Hebei Medical University Third Hospital between November 2020 and May 2022. This cohort included 135 patients with HCC, 171 individuals with non-HCC liver diseases (including hepatitis, fibrosis, or cirrhosis), and 97 healthy volunteers as normal controls (NC). All diagnoses were comprehensively assessed and confirmed according to established Chinese clinical practice guidelines, including the “Guidelines for the Prevention and Treatment of Chronic Hepatitis B (2019)”, “Standard for Diagnosis and Treatment of Primary Liver Cancer (2019)”, “Guidelines for the Prevention and Treatment of Hepatitis C (2019)”, and “Guidelines for the Prevention and Treatment of Non-alcoholic Fatty Liver Disease (2018 Update)”, based on clinical manifestations, imaging examinations, laboratory tests, and pathological evaluation when applicable.

Participants were excluded if they met any of the following criteria: (1) HCC patients who had undergone any prior chemotherapy or radiotherapy; (2) Presence of severe comorbidities in other organs; (3) Pregnancy or lactation; or (4) Co-existing other malignancies.

Clinical data, including age and sex, were collected from all participants. All diagnoses were confirmed according to established clinical practice guidelines. The study was approved by the Ethics Committee of the Hebei Medical University Third Hospital, and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.

Peripheral venous blood samples (5 mL) were collected from each participant. Serum was separated by centrifugation at 1800 × g for 10 minutes at room temperature and stored in aliquots at -80 °C for subsequent analysis. All sample processing steps were completed within 6 hours of collection to prevent degradation.

Exosome isolation and total RNA extraction

Serum exosomes were isolated using the GlyExo-Capture technology on a REXO-32 Extracellular Vesicle Separation and Extraction Instrument (Beijing Hotgen Biotech Co., Ltd., China) according to the manufacturer’s protocol. Briefly, 500 μL of serum was incubated with magnetic beads to capture exosomes. After washing, the exosomes were eluted for subsequent RNA extraction. Total RNA, including miRNAs, was extracted from the isolated exosomes using the miRNeasy^® Mini Kit (QIAGEN, Germany) following the manufacturer’s instructions. The concentration and quality of the extracted RNA were assessed using a Qsep100 automated nucleic acid analysis system (BiOptic Inc., Taiwan).

Human tissue samples

Fresh-frozen HCC tissues and paired adjacent non-tumor liver tissues were collected from patients who underwent surgical resection at the Hebei Medical University Third Hospital. A total of 5 HCC tissues and 5 non-HCC liver tissues were used for next-generation sequencing (NGS) analysis. Additional paired tissues were used for quantitative real-time polymerase chain reaction (qRT-PCR) and immunohistochemical validation. All tissue samples were immediately snap-frozen and stored at -80 °C until use.

High-throughput sequencing and biomarker discovery

A discovery cohort, consisting of serum samples from 68 HCC patients and 131 non-HCC controls (100 with liver disease and 31 healthy volunteers), was used for NGS. Small RNA libraries were prepared and sequenced to profile the exosomal miRNA expression. The raw sequencing data were processed using FastQC (v0.11.9) for quality control, and adapter sequences were trimmed using Cutadapt (v3.7). The cleaned reads were aligned to the human reference genome (hg38) using Bowtie2. miRNA annotation was performed using HT-seq with the miRBase v23 database, and expression levels were normalized to transcripts per million. Differential expression analysis between HCC and non-HCC groups was conducted using the DESeq2 package in R (v4.2.3). A fold change > 2 and an adjusted P value < 0.05 were set as the thresholds for significantly differentially expressed miRNAs. Recursive feature elimination (RFE) with five-fold cross-validation and a random forest algorithm was employed to select the most significant miRNA candidates for building the diagnostic model.

Cell culture

HepG2 and MHCC97H, two human HCC cell lines, were obtained from Shanghai Fuheng Biotechnology Co., Ltd. (Shanghai, China). HepG2 cells were maintained in Minimum Essential Medium supplemented with 10% fetal bovine serum, both sourced from Thermo Fisher Scientific, Inc. (Waltham, MA, United States). MHCC97H cells were cultured in Dulbecco’s Modified Eagle Medium containing 10% fetal bovine serum, as well as penicillin G sodium (100 U/mL) and streptomycin sulfate (100 μg/mL), all provided by Gibco (Thermo Fisher Scientific, Inc., MA, United States). All cultures were incubated at 37 °C in a humidified atmosphere containing 5% CO₂. Lentiviral constructs for NC, miR-200b-3p sponge, and miR-200b-3p overexpression were acquired from Tianjin Sheweisi Biotechnology Co., Ltd. (Tianjin, China). Authentication of all cell lines was carried out via short tandem repeat profiling, and they were confirmed to be free of mycoplasma contamination.

Dual-luciferase reporter assay

To validate the direct targeting of miR-200b-3p to phosphoserine aminotransferase 1 (PSAT1), a dual-luciferase reporter assay was conducted in HEK-293T cells. This cell line was selected for its high transfection efficiency and widespread use in miRNA-target verification studies. Briefly, HEK-293T cells were seeded in 24-well plates at a density of 1 × 10⁵ cells per well one day before transfection. Cells were then co-transfected with 0.4 μg of either the pmirGLO-PSAT1-WT (wild-type, WT) or pmirGLO-PSAT1-MT (mutant, MT) plasmid, along with 50 nM of has-miR-200b-3p mimics or NC mimics, using Lipofectamine 2000 (Invitrogen; Thermo Fisher Scientific, Inc., MA, United States) according to the manufacturer’s instructions. After 48 hours of transfection, cells were lysed, and firefly and Renilla luciferase activities were measured sequentially using the Double-Luciferase Reporter Assay Kit (JinKaiRui Biotech, Wuhan, China). The Renilla luciferase activity was used as an internal control for normalization. The relative luciferase activity was expressed as the ratio of Firefly to Renilla luminescence. All experiments were performed in five replicates, and data are presented as mean ± SD from three independent experiments.

qRT-PCR assays

The expression levels of the selected candidate miRNAs were validated in a separate cohort from our institution comprising 67 patients with HCC and 137 non-HCC controls, which included 71 individuals with liver disease and 66 healthy volunteers. qRT-PCR was performed using the TB Green^® Premix Ex Taq™ II kit (Takara Bio Inc., Japan) on an ABI 7500 Real-Time PCR System (Applied Biosystems, United States). For the initial validation of candidate miRNAs, the relative expression levels were calculated using the -ΔCT method. In the subsequent mechanistic investigations, the relative expression levels of the target genes were determined using the comparative Ct (2^-ΔΔCt) method[20], with the primer sequences used for qRT-PCR detailed in the Supplementary Table 1.

Cell proliferation assay

A CCK-8 assay (Super-Enhanced Cell Counting Kit-8, Dalian Meilun Biotechnology Co., Ltd, China) was employed to evaluate cell proliferation. Briefly, cells were plated in 96-well plates at a density of 3000 cells per well and cultured for 24, 48, or 72 hours. Following each time point, 10 μL of CCK-8 solution was introduced into each well, and the plates were incubated for an additional 2 hours. Absorbance was measured at 450 nm to quantify viable cell numbers. Each experiment was performed in triplicate. For the colony formation assay, cells were seeded into 6-well plates at a low density (500 cells per well) and cultured for 10-14 days to allow colony development. The resulting colonies were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and counted manually to assess clonogenic survival.

Wound healing assay

Cell migration was evaluated using a wound scratch assay. HepG2 and MHCC97H cells were cultured in 6-well plates until forming a confluent monolayer. A uniform scratch was introduced using a sterile 20 μL pipette tip. After rinsing twice with phosphate-buffered saline to remove detached cells, wound closure was monitored and photographed at 0, 24, and 48 hours under a Nikon microscope. The migration distance was quantified by measuring scratch width changes using ImageJ software. All experiments were performed in triplicate.

Transwell assay

Cell invasive ability was further examined using Transwell chambers pre-coated with Matrigel. In total, 1 × 10⁵ cells suspended in 200 μL serum-free medium were seeded into the upper compartment, while the lower chamber was filled with complete medium. Following 48 hours of incubation, cells that had invaded through the membrane were fixed and stained with 0.5% crystal violet. The number of invaded cells was imaged with a Nikon microscope and counted using ImageJ. Each assay was independently repeated three times.

Western blotting

Cells were rinsed twice with cold phosphate-buffered saline and subsequently lysed using radioimmunoprecipitation assay buffer (Epizyme Biomedical Technology, #PC101, Shanghai, China) supplemented with 1% protease inhibitor cocktail (Epizyme, #GRF101, Shanghai, China). The lysates were collected and subjected to immunoblotting with the following primary antibodies: Anti-PSAT1 (1:1000, Proteintech, #1B2C9, Wuhan, China) and anti-β-actin (1:10000, Affinity, #AF7018, Jiangsu Province, China).

Statistical analysis

Data from functional mechanism experiments investigating miR-200b-3p were derived from at least three independent assays and are presented as the mean ± SD. Statistical analyses for these experiments were performed with SPSS standard version 22.0 software (SPSS, Inc., Chicago, IL, United States). The differences between the two groups were compared using Student’s t-test. A P value of less than 0.05 was considered statistically significant.

For the clinical validation and diagnostic model construction, statistical analyses were performed using R software. After excluding three samples with missing clinical information, the qRT-PCR validation cohort was randomly divided into a training set (n = 142, comprising 44 HCC and 98 non-HCC subjects) and a validation set (n = 59, comprising 22 HCC and 37 non-HCC subjects) at a 7:3 ratio. Differences in continuous variables (e.g., age, miRNA expression) between groups were assessed using the independent samples t-test. The χ² test was used to compare categorical variables (e.g., sex). The association between variables and HCC risk was evaluated using Spearman correlation analysis.

A binary logistic regression analysis was performed on the training set to construct a diagnostic model. Variables that were significant in the univariable analysis were included as candidates. Multicollinearity among these variables was assessed using the variance inflation factor, and variables with a variance inflation factor > 5 were excluded. The final diagnostic model was built by incorporating the selected miRNAs and clinical variables. The diagnostic performance of the model was evaluated by calculating the area under the receiver operating characteristic curve (AUC), along with its sensitivity and specificity. The model’s performance was assessed internally in the training set and then validated in the independent validation set and the total cohort.

RESULTS

Characteristics of the study participants

This study was structured into three sequential phases: Discovery, technical verification, and model construction and evaluation. The overall design is summarized in Figure 1. In the discovery phase, small RNA sequencing was performed on serum exosomes from a retrospective cohort (TH-HebMU cohort-1), comprising 68 HCC patients and 131 non-HCC controls, to conduct a genome-wide screening for differentially expressed miRNAs. In the technical verification phase, the expression levels of the candidate miRNAs identified from sequencing were independently verified using qRT-PCR in a second cohort (TH-HebMU cohort-2), which included 66 HCC patients and 135 non-HCC controls. In the model construction and evaluation phase, the verified candidates were used to develop a multivariate diagnostic model (HCC-3miRNA) in a training set (n = 142) derived from TH-HebMU cohort-2. The model’s performance was then rigorously evaluated in an internal validation set (n = 59) from the same cohort, as well as in the total combined cohort (n = 201). Additionally, the expression of key miRNAs was further confirmed in a subset of tissue samples (TH-HebMU cohort-3, n = 5 HCC and 5 non-HCC) to support mechanistic investigation. The baseline characteristics of all participants are summarized in Table 1.

Figure 1 Study flowchart. miRNA: MicroRNA; HCC: Hepatocellular carcinoma; qRT-PCR: Quantitative real-time polymerase chain reaction; RFE: Recursive feature elimination; NGS: Next-generation sequencing.

Table 1 Baseline status of enrolled cases.

	TH-HebMU cohort-1			TH-HebMU cohort-2			TH-HebMU cohort-3
	Non-HCC (n = 131)	HCC (n = 68)	P value	Non-HCC (n = 135)	HCC (n = 66)	P value	Non-HCC (n = 5)	HCC (n = 5)	P value
Sex			0.002b			0.142			1.000
Male	85	58		93	52		4	4
Female	46	10		42	14		1	1
Age	43.13 ± 13.671	55.79 ± 10.74	< 0.001c	50.81 ± 11.31	59.98 ± 12.65	< 0.001c	47.00 ± 11.55	58.20 ± 9.07	> 0.05
miR-200b-3p	28.42 ± 23.21	61.75 ± 63.99	< 0.001c	-9.24 ± 1.03	-8.51 ± 0.78	< 0.001c	33369 ± 12333	15753 ± 7165	< 0.05a
miR-215-5p	942.51 ± 1422.19	3927.90 ± 6286.45	< 0.001c	-8.46 ± 0.83	-7.53 ± 0.78	< 0.001c	156.1 ± 103.5	754.8 ± 247.7	< 0.01b
miR-335-5p	35.96 ± 22.04	53.92 ± 33.97	< 0.001c	-9.41 ± 0.72	-8.78 ± 0.81	< 0.001c	35 ± 23.7	238.3 ± 52.6	< 0.01b
miR-378a-3p	818.66 ± 305.82	1063.99 ± 376.92	< 0.001c	-7.15 ± 0.84	-6.27 ± 0.95	< 0.001c	79340 ± 15325	66567 ± 27158	> 0.05
miR-452-5p	9.55 ± 10.45	24.62 ± 48.77	< 0.001c	-9.98 ± 1.07	-8.90 ± 1.27	< 0.001c	268.7 ± 184.4	1087 ± 1108	> 0.05
miR-501-3p	263.94 ± 123.31	410.85 ± 240.83	< 0.001c	-8.64 ± 0.79	-7.79 ± 0.82	< 0.001c	3226 ± 1680	1196 ± 199.7	< 0.05a

¹Age and microRNA expression values are presented as mean ± SD. For cohort-1, values are transcripts per million from next-generation sequencing; for cohort-2, values are -ΔCT from quantitative real-time polymerase chain reaction; for cohort-3, values are raw quantification cycle values or normalized counts.

^aP < 0.05.

^bP < 0.01.

^cP < 0.001.

TH-HebMU: Hebei Medical University Third Hospital; HCC: Hepatocellular carcinoma.

Identification of a candidate serum exosomal miRNA signature via NGS

To identify candidate exosomal miRNAs for HCC diagnosis, small RNA sequencing was performed on serum exosomes from 68 HCC patients and 131 non-HCC controls (including individuals with chronic liver diseases and healthy volunteers). A total of 2278 mature miRNAs were detected across all samples. Using DESeq2 for differential expression analysis, 112 miRNAs were identified as significantly differentially expressed between HCC and non-HCC groups, based on criteria of false discovery rate < 0.05 (Supplementary Table 2).

Subsequent feature selection was performed using RFE and random forest algorithms, resulting in the selection of six miRNAs most strongly associated with HCC status: MiR-200b-3p, miR-215-5p, miR-335-5p, miR-378a-3p, miR-452-5p, and miR-501-3p. All six candidate miRNAs were significantly upregulated in the HCC group compared to the non-HCC group (all P < 0.01), with consistent expression trends observed across patients (Figure 2). These six miRNAs were selected for further validation and model development.

Figure 2 Expression levels of six key differentially expressed microRNAs in serum exosomes between hepatocellular carcinoma and non-hepatocellular carcinoma groups. A: MiR-200b-3p; B: MiR-215-5p; C: MiR-335-5p; D: MiR-378a-3p; E: MiR-452-5p; F: MiR-501-3p. ^dP < 0.0001. HCC: Hepatocellular carcinoma.

To assess whether age confounded the initial screening, we performed an age-matched sensitivity analysis. After 1:1 matching (60 pairs), age was balanced between groups (HCC: 55.7 ± 10.7 years; non-HCC: 52.4 ± 10.4 years; P = 0.085). In this age-balanced subset, all six miRNAs remained significantly upregulated in HCC patients: MiR-200b-3p (P = 0.0024), miR-215-5p (P = 0.0034), miR-335-5p (P = 0.0018), miR-378a-3p (P = 0.0014), miR-452-5p (P = 0.0232), and miR-501-3p (P < 0.0001) (Supplementary Table 3). These results confirm that the differential expression of these miRNAs is independent of age.

Independent validation of the 6-miRNA signature by qRT-PCR

To validate the diagnostic relevance of the six candidate miRNAs, qRT-PCR was performed in an independent cohort (TH-HebMU cohort-2). After the exclusion of one HCC patient and two non-HCC controls due to incomplete clinical information, this cohort ultimately comprised 66 HCC patients and 135 non-HCC controls. All six miRNAs exhibited expression patterns consistent with the sequencing results: Significantly elevated levels in the HCC group compared to non-HCC controls (P < 0.05 for each miRNA; Table 1). These findings confirmed the reproducibility and robustness of the six miRNAs as potential diagnostic biomarkers for HCC.

Diagnostic performance of individual miRNAs

Univariate logistic regression analyses were conducted for each of the six miRNAs to assess their individual diagnostic performance (Supplementary Table 4). Receiver operating characteristic analysis showed that each miRNA possessed moderate diagnostic power in distinguishing HCC from non-HCC (AUCs ranging from 0.687 to 0.798; Figure 3). Among them, miR-215-5p achieved the highest AUC (0.798), followed by miR-452-5p (AUC = 0.741). However, the diagnostic utility of individual miRNAs was insufficient for clinical application, motivating the development of a multivariate model combining multiple biomarkers.

Figure 3 Area under the receiver operating characteristic curve values of six candidate microRNAs in diagnostic models. A: MiR-200b-3p; B: MiR-215-5p; C: MiR-335-5p; D: MiR-378a-3p; E: MiR-452-5p; F: MiR-501-3p. ^cP < 0.001, ^dP < 0.0001. AUC: Area under the curve.

Construction of a multivariate diagnostic model: HCC-3miRNA panel

To improve diagnostic accuracy, a multivariate logistic regression model was constructed using the six validated miRNAs along with patient age as an additional covariate. The final model incorporated three miRNAs, miR-200b-3p, miR-215-5p, and miR-452-5p, and age, based on statistical significance and multicollinearity assessment (Figure 4). The resulting diagnostic model, hereafter referred to as the HCC-3miRNA model, was defined by the following logistic equation: Logit(P) = -0.685 × miR-200b-3p + 1.794 × miR-215-5p + 0.516 × miR-452-5p + 0.067 × age + 8.74.

Figure 4 Construction and variable assessment of the hepatocellular carcinoma-3miRNA diagnostic model. A: Graphical abstract illustrating the model construction workflow; B: Forest plot displaying the odds ratios and confidence intervals of each variable included in the final multivariate logistic regression model. RT-qPCR: Real-time quantitative polymerase chain reaction; HCC: Hepatocellular carcinoma; CI: Confidence interval.

Diagnostic performance of the HCC-3miRNA model

The model was developed and validated using a split-sample design within the qPCR cohort. Training cohort (n = 142): The model achieved an AUC of 0.863, with sensitivity = 86.4% and specificity = 76.5% (Figure 5A). Validation cohort (n = 59): Performance remained consistent with an AUC of 0.822, sensitivity = 81.8%, and specificity = 78.4% (Figure 5B). Combined cohort (total dataset) (n = 201): The model maintained robust performance with an AUC of 0.852, sensitivity = 80.3%, and specificity = 80.0% (Figure 5C). These findings demonstrate the reliability, generalizability, and clinical applicability of the HCC-3miRNA model for distinguishing HCC from non-HCC individuals.

Figure 5 Diagnostic performance of the 3-miRNA model. A: The receiver operating characteristic curve and area under the curve of the training set; B: The validation set; C: The total development set. AUC: Area under the curve.

miR-200b-3p directly targets and suppresses PSAT1 by binding to its 3’ untranslated region

To elucidate the functional mechanism of miR-200b-3p in HCC, we first sought to identify its direct downstream targets. We integrated our NGS data from serum exosomes and HCC tissues, in which the serum exosomal miRNAs were derived from the six feature miRNAs screened by RFE, revealing four differentially expressed miRNAs common to both sources (Figure 6A). Bioinformatic prediction using TargetScan was then employed to identify potential mRNA targets of these miRNAs (Supplementary Figure 1). By cross-referencing with mRNA expression data from HCC tissues, we identified five mRNAs that were also differentially expressed. Among these, a compelling inverse correlation was observed specifically between miR-200b-3p and PSAT1: MiR-200b-3p was significantly downregulated in HCC tissues, while PSAT1 was markedly upregulated (Figure 6B-G). This strong negative association prompted us to hypothesize that PSAT1 is a direct functional target of miR-200b-3p.

Figure 6 MiR-200b-3p directly targets and suppresses phosphoserine aminotransferase 1 by binding to its 3’ untranslated region. A: Venn diagram showing the four microRNAs that were significantly differentially expressed in both serum exosomes and liver tissues of hepatocellular carcinoma (HCC) and non-HCC cases; B-F: Expression levels of the five predicted target mRNAs that are differentially expressed in both serum exosomes and liver tissues of HCC and non-HCC cases; G: Analysis of phosphoserine aminotransferase 1 expression levels by immunohistochemical staining; H and I: Quantitative real-time polymerase chain reaction validation of miR-200b-3p overexpression efficiency in HepG2 and MHCC97H cells; J and K: Western blot analysis validating the efficiency of miR-200b-3p knockdown and overexpression in HepG2 and MHCC97H cells; L: Dual-luciferase reporter assay validates phosphoserine aminotransferase 1 as a direct target of miR-200b-3p. ^aP < 0.05, ^bP < 0.01, ^cP < 0.001, and ^dP < 0.0001. PSAT1: Phosphoserine aminotransferase 1; NC: Normal control; HCC: Hepatocellular carcinoma; OE: Overexpression; SP: Sponge; NS: Not significant.

To investigate the functional role of miR-200b-3p in HCC, we selected two cell lines with distinct genetic backgrounds and malignant phenotypes: HepG2 (a less aggressive, well-differentiated cell line) and MHCC97H (a highly invasive and metastatic cell line). As shown in Supplementary Figure 2, the basal expression of miR-200b-3p was relatively high in HepG2 cells and low in MHCC97H cells, which aligns with its potential tumor-suppressive role suggested by our tissue data. To rigorously test this hypothesis, both loss- and gain-of-function approaches were performed in each cell line. We next asked whether miR-200b-3p could regulate PSAT1 expression. Using lentiviral vectors, we established stable HepG2 and MHCC97H cell lines with either miR-200b-3p overexpression or sponge-mediated knockdown. qRT-PCR analysis confirmed the efficient modulation of miR-200b-3p levels in these cells (Figure 6H and I). Crucially, western blot analysis demonstrated that overexpression of miR-200b-3p led to a decrease in PSAT1 protein levels, whereas its knockdown increased PSAT1 (Figure 6J and K). This inverse regulatory relationship indicated that miR-200b-3p negatively regulates PSAT1.

To determine whether this regulation was direct, we performed a dual-luciferase reporter assay. We cloned the WT 3’ untranslated region of PSAT1, containing the predicted miR-200b-3p binding site, or a MT version downstream of a luciferase gene. When co-transfected with miR-200b-3p mimics into HEK-293T cells, the luciferase activity of the WT reporter was significantly suppressed compared to the control, while the activity of the MT reporter remained unchanged (Figure 6L). This result confirms that miR-200b-3p directly binds to the specific site within the 3’ untranslated region of PSAT1 to suppress its expression.

miR-200b-3p suppresses the malignant phenotypes of HCC cells

Functional analyses using miR-200b-3p knockdown and overexpression models defined its critical role in constraining the malignant properties of HCC cells. Notably, miR-200b-3p overexpression functioned as a potent inhibitor of proliferation, migration, and invasion in both HepG2 and MHCC97H cells (P < 0.05, Figure 7). In contrast, its knockdown consistently promoted these oncogenic phenotypes (P < 0.005).

Figure 7 MiR-200b-3p suppresses the malignant phenotypes of hepatocellular carcinoma cells. A-D: CCK-8 assays and cell colony formation assay (n = 3) indicated that overexpression of miR-200b-3p suppressed hepatocellular carcinoma cell proliferation, whereas its knockdown enhanced proliferative capacity; E and F: Wound healing assays (n = 3) demonstrated that miR-200b-3p knockdown accelerated wound closure, while its overexpression impeded cell migration; G: In Matrigel invasion assays (n = 3), silencing of miR-200b-3p promoted cellular invasion, in contrast to the suppressive effect observed upon miR-200b-3p overexpression. ^aP < 0.05, ^bP < 0.01, and ^cP < 0.001. NC: Normal control; OE: Overexpression; SP: Sponge.

DISCUSSION

The pursuit of a reliable, non-invasive biomarker for the early detection of HCC remains a paramount objective in clinical oncology, driven by the profound impact of early diagnosis on patient survival. In this study, we addressed this critical need by systematically identifying a novel serum exosomal miRNA signature and developing a robust diagnostic model. Our final, optimized model, which integrates scores from three key miRNAs (miR-200b-3p, miR-215-5p, and miR-452-5p) with patient age, demonstrated strong discriminatory power, achieving an AUC of 0.863, a sensitivity of 86.4%, and a specificity of 76.5%. This performance not only underscores the model’s potential clinical utility but also contributes a rigorously validated tool to the rapidly evolving field of liquid biopsy for HCC.

Beyond its statistical performance, a key strength of our study lies in the elucidation of the functional role of miR-200b-3p, a central component of our diagnostic signature. Our functional experiments revealed that miR-200b-3p acts as a tumor suppressor in HCC cells. We identified PSAT1 as a direct and functional target of miR-200b-3p, confirmed by a dual-luciferase reporter assay. Furthermore, overexpression of miR-200b-3p led to the downregulation of PSAT1 and significantly suppressed the proliferation, migration, and invasion of HepG2 and MHCC97H cells. Conversely, inhibiting miR-200b-3p had the opposite effect, enhancing these malignant phenotypes. This miR-200b-3p/PSAT1 axis provides a mechanistic foundation for its inclusion in our diagnostic model, suggesting that the elevated levels of exosomal miR-200b-3p in HCC patient serum may reflect a compensatory tumor-suppressive response. This biological plausibility significantly strengthens the clinical relevance of our biomarker signature, positioning it not merely as a correlation but as a reflection of underlying tumor biology.

A cornerstone of our study was the deliberate and transparent model-building process. We began with a data-driven discovery phase, employing high-throughput sequencing and advanced machine learning algorithms, a combination of RFE and random forest analysis, to distill an initial list of 112 differentially expressed miRNAs down to a potent 3-miRNA panel. This initial panel represented the most statistically significant candidates. However, a central tenet of developing clinical prediction models is the principle of parsimony. A model with fewer variables is often more desirable as it is less susceptible to overfitting, more likely to be generalizable to new populations, more cost-effective, and easier to implement and interpret in a busy clinical setting. An overly complex model, while potentially fitting the training data perfectly, risks capturing noise rather than a true biological signal, leading to poor performance on external datasets.

Guided by this principle, we undertook a crucial second step of model optimization. An initial multivariable model incorporating all six miRNAs and age was constructed. We then systematically evaluated the contribution of each variable and proceeded to remove those with lower statistical significance. This iterative refinement process is a well-established strategy in biostatistics aimed at enhancing model stability and ensuring that the final model contains only the most powerful and independent predictors[21]. This led to our final, more parsimonious model. The robust performance of this final model validates our optimization strategy, demonstrating that we successfully retained the core diagnostic information while creating a more streamlined and clinically practical tool.

The performance of our model is highly competitive when contextualized within the landscape of existing miRNA-based biomarkers for HCC. The field is replete with promising signatures, yet direct comparisons are often confounded by variations in study design, patient cohorts, and analytical platforms. For example, a pioneering study by Zhou et al[22] reported a plasma-based seven-miRNA panel with an AUC of 0.864 for discriminating hepatitis B virus-related HCC from healthy, chronic hepatitis B and cirrhosis. While impressive, our study focused on exosomal miRNAs, which offer a distinct advantage due to the protective lipid bilayer that shields them from degradation by circulating RNases, ensuring greater pre-analytical stability[12,13]. Among studies on exosomal miRNAs, Sohn et al[23] developed a four-miRNA signature (miR-18a, miR-221, miR-222, and miR-224) that was significantly higher in patients with HCC than those with chronic hepatitis B or liver cirrhosis (P < 0.05). However, their cohort was restricted to patients with hepatitis B virus etiology. In contrast, our study included a more heterogeneous control group, including healthy individuals and patients with various liver diseases, making our diagnostic task more challenging but arguably more reflective of a real-world screening scenario. More recently, Li et al[16] reported a five-exosomal miRNA signature with an AUC of 0.95 in their validation set, a performance superior to ours. However, it is noteworthy that our model achieved this competitive accuracy with a more concise panel of three miRNAs, further reinforcing the value of our parsimonious approach. When compared against traditional protein markers, the combination of AFP, AFP-L3 (lens culinaris agglutinin-reactive AFP), and PIVKA-II typically yields AUCs in the range of 0.751-0.765[9], placing our model at the upper end of performance for existing non-invasive tests.

The methodological foundation of our study is another significant strength. The isolation of high-purity exosomes is a critical yet often inconsistent step in many studies. We employed the GlyExo-Capture technology, a state-of-the-art affinity-based method that leverages specific glycan patterns on the exosome surface[24]. This technique circumvents many of the well-documented limitations of traditional methods like ultracentrifugation (e.g., protein contamination, exosome damage, low yield) and precipitation kits (e.g., co-precipitation of non-exosomal proteins and lipoproteins)[25,26]. By ensuring the isolation of a purer and more homogenous population of intact exosomes, GlyExo-Capture minimizes pre-analytical variability and enhances the accuracy and reproducibility of the downstream miRNA quantification[27]. This methodological rigor lends higher confidence to our findings and provides a more solid basis for potential clinical translation.

Furthermore, the miRNAs constituting our final diagnostic signature are not merely statistical correlates but are deeply embedded in the biology of cancer. In HCC tissues, the expression level of miR-215-5p is negatively associated with the formation of vasculogenic mimicry, whereas ZEB2 expression exhibits a positive correlation. Through its regulation of ZEB2, miR-215-5p may suppress vasculogenic mimicry, thereby potentially reducing vascular invasion and lowering the risk of HCC recurrence[28]. Meanwhile, miR-501-3p has been characterized as a potent player in suppressing metastasis and progression of HCC through targeting LIN7A. The function of miR-335-5p is more context-specific but has been established as a powerful regulator of progression in multiple cancer types[29,30]. Similarly, as our data shows, miR-200b-3p exerts tumor-suppressive effects by targeting PSAT1, a key enzyme in serine biosynthesis that has been implicated in cancer progression. The fact that our data-driven approach converged on a set of miRNAs with established, antagonistic roles in tumor progression provides strong biological plausibility for our model. It suggests that our signature captures a snapshot of the complex interplay between oncogenic and tumor-suppressive forces that defines the malignant state.

An intriguing observation from our study is the elevated levels of miR-200b-3p in serum exosomes but reduced expression in HCC tissues. Rather than being contradictory, we propose that this reflects an “active miRNA sorting” or “exosome escape” mechanism, whereby cancer cells actively package tumor-suppressive miRNAs into exosomes for disposal to maintain low intracellular levels conducive to proliferation. This hypothesis is supported by a recent study demonstrating that lung cancer cells selectively sort tumor-suppressive miR-4732-3p into exosomes via hnRNPK to sustain malignant phenotypes[31]. Thus, the high serum exosomal levels of miR-200b-3p can be viewed as a “sink” reflecting active tumor disposal, reinforcing rather than undermining its diagnostic value.

Despite the robust design and promising results, this study is not without limitations. First, its retrospective and single-center nature introduces a potential risk of selection bias and may limit the immediate generalizability of our findings to different ethnic populations or patient cohorts with varying HCC etiologies (e.g., hepatitis C virus-related, alcoholic, or non-alcoholic fatty liver disease). Although we validated the signature in separate patient cohorts, all cohorts were from the same center and therefore do not constitute true external validation. Therefore, the foremost future direction is the validation of our model in large-scale, multi-center prospective trials. Second, while our model demonstrates excellent performance, we did not perform a direct head-to-head comparison against the combination of AFP, PIVKA-II, and AFP-L3 within the same patient cohort. Such a study would be invaluable for definitively positioning the clinical utility of our miRNA panel relative to the existing standard of care. Furthermore, investigating the potential synergistic value of combining our miRNA panel with these markers in a multi-biomarker model could pave the way for more accurate diagnostic strategies. Third, our “non-HCC” control group, while reflecting a real-world population, could be further stratified in future analyses to assess the model’s performance specifically for distinguishing HCC from high-risk cirrhotic patients, which represents the most pressing clinical challenge. Finally, while we elucidated the tumor-suppressive role of miR-200b-3p via the PSAT1 pathway, the precise mechanisms governing its selective packaging into exosomes and its functional impact on the tumor microenvironment upon delivery to recipient cells remain to be fully explored.

CONCLUSION

In conclusion, this study successfully developed and rigorously validated a parsimonious and powerful diagnostic model for HCC, based on a novel signature of three serum exosomal miRNAs and patient age. By integrating an advanced exosome isolation technology with a meticulous, multi-stage bioinformatics approach, our model achieves a diagnostic accuracy that is highly competitive with other reported biomarkers. This non-invasive tool holds significant promise for improving the early detection of HCC, addressing a clear and urgent clinical need. The findings strongly support its advancement into larger prospective validation studies, a critical step towards translating this biomarker signature into a clinically impactful test that can ultimately improve outcomes for patients at risk for this deadly disease. By combining a robust diagnostic signature with mechanistic insights into miR-200b-3p’s function, our work provides a more comprehensive foundation for the future development of exosomal miRNA-based liquid biopsies for HCC.

ACKNOWLEDGEMENTS

The authors thank Beijing Hotgen Medical Laboratory for their collaborative technical support in exosome isolation and sequencing. This individual and institution has read and endorsed the data and conclusions presented in this manuscript.