Development and Validation of a Lectin-independent Liquid Chromatography–Tandem Mass Spectrometry Method for Serum Glycosylated Alpha-fetoprotein Analysis and Comparison with a Liquid-phase Binding Assay

Abstract

Background

Alpha-fetoprotein (AFP) and its isoform AFP-L3 are well-established serum biomarkers for hepatocellular carcinoma (HCC), a common malignancy and a leading cause of cancer-related mortality worldwide. Current methods for measuring these biomarkers are primarily lectin-based assays including the liquid-phase binding assay (LiBA) and liquid chromatography–tandem mass spectrometry (LC-MS/MS), both of which have limitations in diagnostic sensitivity and clinical utility for samples with low AFP concentrations. We aimed to develop a lectin-independent LC-MS/MS method for quantifying fucosylated AFP proteins (AFP-Fuc%).

Methods

We conducted analytical validation, including method comparisons, over 2 months. The analytical sensitivity and diagnostic performance of this method were evaluated using 525 human serum samples—235 from HCC patients and 290 from non-HCC individuals—and compared with those of LiBA, which measured AFP-L3 levels.

Results

The LC-MS/MS method demonstrated acceptable within-laboratory imprecision (CVs<17.1%) without detectable bias, carryover, or matrix effects. Our method exhibited a broader linear dynamic range (spanning five orders of magnitude) and 10-fold higher analytical sensitivity than LiBA. The diagnostic performance of our method was significantly superior to that of LiBA, particularly in patients with low AFP concentrations (<7 ng/mL, P<0.001), with improved accuracy, sensitivity, and precision at a specificity of 96.2%.

Conclusions

The validated LC-MS/MS method demonstrated robust analytical performance and superior diagnostic accuracy over LiBA for HCC diagnosis while avoiding the inherent limitations of lectin-based assays. Our LC-MS/MS assay shows promise for early HCC detection and may contribute to enhanced patient care.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a common malignancy and a leading cause of cancer-related mortality worldwide [1]. Chronic hepatitis B virus (HBV) infection remains the most prominent risk factor for HCC development, and liver cirrhosis (LC) continues to be a major underlying condition [1–3]. More than 90% of HCC cases occur in the setting of chronic liver disease. Although HBV can promote HCC development even in the absence of cirrhosis, most patients with HBV-induced HCC present with cirrhosis. Given that HCC arises in a well-defined high-risk population, namely those with chronic HBV infection or cirrhosis, many cases are detected from regular surveillance. Therefore, effective surveillance and early diagnosis based on suitable biomarkers are critical for improving clinical outcomes and reducing mortality [4].

Biomarkers are measurable indicators of biological processes or disease states, and identifying reliable biomarkers is crucial for early diagnosis. Despite ongoing efforts, current biomarkers for HCC perform suboptimally, particularly for early-stage detection. Although numerous metabolomics- and genomics-based biomarkers have been proposed, most remain unvalidated and lack sufficient clinical evidence [5]. Alpha-fetoprotein (AFP), a glycoprotein produced by the liver, and its Lens culinaris agglutinin (LCA)–reactive glycoform, AFP-L3, are the most extensively studied and clinically adopted biomarkers for HCC [6, 7]. AFP is broadly used in clinical screening and diagnosis, whereas AFP-L3 is specifically associated with early-stage HCC and is considered helpful when total AFP concentrations are inconclusive [8]. However, AFP-L3 demonstrates limited sensitivity at low AFP concentrations, particularly in AFP-negative HCC or for serum AFP concentrations ≤20 ng/mL [5, 9], compromising its clinical utility in a large subset of patients. AFP and AFP-L3 are typically measured using immunoassay-based platforms such as the electrochemiluminescence immunoassay (ECLIA) and liquid-phase binding assay (LiBA) [10]. A percentage of fucosylated AFP (AFP-Fuc%), which corresponds to AFP-L3, is a serological marker for HCC. It can be measured using liquid chromatography–tandem mass spectrometry (LC-MS/MS) without the need for lectins such as LCA [11, 12]. A comparison of these biomarkers is summarized in Supplemental Data Table S1.

LC-MS/MS offers a high-specificity alternative for glycopeptide quantification, addressing key limitations of immunoassays [13–16]. Although MS-based core-fucosylated AFP detection has been explored, lectin-based methods are more relied on to identify glycan structures rather than to directly quantify proteins [17, 18]. Lectin-affinity techniques, particularly those using LCA, have been employed to enrich core-fucosylated AFP forms [19, 20]. These lectin-based methods exhibit poor sensitivity for low-abundance targets, resulting in missing values or reduced diagnostic performance [11]. Immunoprecipitation coupled with LC-MS/MS enables the direct and sensitive quantification of protein isoforms and glycosylation states, offering a more robust clinical tool [12].

In this study, we developed an automated LC-MS/MS method for core-fucosylated AFP glycopeptide (AFP-Fuc%) measurement without lectins that provides enhanced performance, particularly at low AFP concentrations. We validated the assay following guidelines from the Clinical and Laboratory Standards Institute (CLSI), U.S. Food and Drug Administration (FDA), European Medicines Agency (EMA), and Korean FDA [21–32]. By enabling sensitive and glycoform-specific quantification even at low AFP concentrations, this method offers a clinically meaningful advancement for improving early HCC detection in high-risk populations.

MATERIALS AND METHODS

Sample collection

All experiments were conducted in accordance with the ethical guidelines of the 1975 Declaration of Helsinki. Ethical approval for the use of residual samples was obtained from the Institutional Review Board of Seegene Medical Foundation (Seoul) (SMF-IRB-2023-002) and Ajou University Hospital (Suwon) (AJOUIRB-KSP-2016-365), Korea. Informed consent was waived because of prior donor consent and de-identification. Residual human sera obtained after routine blood testing at the Seegene Medical Foundation were used for method development, optimization, automation, and validation. For clinical performance evaluation, serum samples were retrospectively collected from Ajou University Hospital between January 2015 and December 2019. This study was conducted over 4 yrs, from 2021 to 2024. Samples included sera from 50 healthy individuals, 240 patients with liver diseases (120 HBV and 120 LC), and 235 HCC patients (80 stage 1, 80 stage 2, and 75 stage 3) (Supplemental Data Table S2). All patients with chronic liver disease (HBV, LC, or HCC) had underlying HBV infection. Healthy controls were individuals aged 18–50 yrs who visited the Ajou Health Promotion Center for routine health checkups and had no history of liver disease, chronic illness, or regular medication use. Chronic HBV infection was defined as persistent hepatitis B surface antigen positivity for ≥6 months. LC was diagnosed based on ultrasonographic evidence of cirrhotic morphology, liver stiffness >14 kPa on elastography, and laboratory abnormalities such as thrombocytopenia and/or hypoalbuminemia [33]. HCC was diagnosed according to the Korean Liver Cancer Association guidelines based on characteristic imaging findings, without requiring histological confirmation in high-risk patients [34]. The inclusion criteria were as follows: (i) adults aged ≥18 yrs; (ii) confirmed chronic HBV infection; and (iii) verified HBV as the etiology in LC and HCC groups. All HCC patients were treatment-naïve at the time of blood sampling. Exclusion criteria included coinfection with hepatitis C virus (HCV) or human immunodeficiency virus, a history of autoimmune hepatitis, alcoholic or genetic/metabolic liver diseases, and the presence of any extrahepatic malignancy. Peripheral blood samples were collected once at baseline, processed immediately, and stored at –80 °C until analysis.

Calibrator, QC material, and matrix serum

Calibrators and QC samples were selected from residual sera with the highest AFP concentrations and prepared by dilution in AFP-depleted pooled human serum. AFP concentrations and AFP-Fuc% in calibrators and QC samples were determined using ECLIA, LiBA, LC-MS/MS, and fluorescence-based glycan analysis. The absence of AFP in depleted serum was confirmed using LC-MS/MS, ECLIA, and LiBA. A calibrator with a low-fucosylated AFP level (8.7% fucosylation, 1–533 ng/mL) was employed for analytical validation over a 2-month period. A calibrator with a medium-fucosylated AFP level (48.3% fucosylation, 0.07–8,600 ng/mL) was utilized to assess the analytical measurement range including the lower limit of quantification (LLOQ) and upper limit of quantification (ULOQ). QC samples were prepared with low- and high-fucosylated AFPs. Low-fucosylated AFPs (<5%, 100 μg/mL) were purchased from Fitzgerald Industries International (Acton, MA, USA); high-fucosylated AFPs (>95%, 100 μg/mL) were isolated and purified in house. Detailed information is provided in the Supplemental Methods.

LC-MS/MS

Evotip-loaded samples were directly eluted and separated in an Evosep One LC (Evosep Biosystems, Odense, Denmark) and analyzed using a Triple Quad 6500+ MS (Sciex, Framingham, MA, USA) system equipped with an OptiFlow Turbo V ion. The Evosep One method was designed for processing 60 samples per day, with a 21-min gradient, 24-min cycle time, and 1.0 μL/min flow rate. Chromatographic separation was achieved using an analytical column (8 cm×150 μm, 1.5 μm, EV1109; Evosep) with the column heater set at 40 °C. Mobile phases A (0.1% formic acid in water) and B (0.1% formic acid in acetonitrile) were used. The mass spectrometer was operated in positive-ion, multiple reaction monitoring mode. The dwell time for mass transitions was 40 msecs, and the cycle time was 0.7651 secs. The Q1 and Q3 quadrupoles were set to unit resolution (0.7 Da). The LC-MS/MS system was controlled with the Evosep plugin (v.2.3.58.0, Evosep Biosystems, Odense, Denmark) in Chronos (v.5.1.8.0, Evosep Biosystems, Odense, Denmark) and Analyst software (v.1.7.2, Sciex, Framingham, MA, USA) for method establishment and data acquisition.

Method validation

To validate the LC-MS/MS method, we evaluated its linearity, imprecision, accuracy, analytical sensitivity, analytical specificity, carryover, matrix effect, dilution integrity, stability, assay recovery, immunoprecipitation recovery, and comparability with immunoassays over 2 months according to CLSI guidelines and US FDA, EMA, and Korean FDA standards (Supplemental Data Table S3) [21–32]. Detailed information is provided in the Supplemental Methods.

Clinical performance evaluation

To evaluate the clinical performance of the LC-MS/MS method for AFP-Fuc% measurement, serum samples from healthy individuals and patients with HBV, LC, or HCC were retrospectively analyzed. The diagnostic performance of AFP-Fuc% to distinguish HCC from non-HCC patients was assessed by calculating the area under the precision-recall curve (AUPRC) and area under the receiver operating characteristic curve (AUROC). Sensitivity and specificity were determined based on optimal cut-off values derived from Youden’s index. Comparative analyses with conventional biomarkers (AFP and AFP-L3) were conducted to assess the relative diagnostic performance. Subgroup analyses were performed using different AFP cut-offs (<7, <10, and <20 ng/mL) to assess the diagnostic utility of AFP-Fuc%, particularly in patients with low AFP concentrations. All clinical samples were collected before therapeutic intervention and evaluated in a blinded manner with respect to the clinical diagnosis.

Statistical analysis

Raw LC-MS/MS data were processed using quantitation tools within Analyst v.1.7.2 (Sciex, Framingham, MA, USA) and analyzed (AUPRC, AUROC, ANOVA, Bland–Altman plot, Deming fit regression, and correlation analysis) and visualized using MedCalc v.19.7 (Ostend, Belgium). The sample size was calculated using G*Power (v.3.1.9.7; Heinrich-Heine-Universität, Düsseldorf, Germany) based on a two-tailed z-test for the correlation between two independent Pearson correlation coefficients, with an effect size of q=0.6, alpha error probability of 0.05, and power of 0.95 (>76 in each group). The assumption of q=0.6 was chosen a priori because the true effect size was unknown at the planning stage. Following Cohen’s guideline (q≈0.5 as a medium effect), we set a slightly higher threshold to ensure practical significance and to account for limited preliminary data.

RESULTS

Assay development

The assay development and evaluation process is illustrated in Fig. 1. We optimized immunoprecipitation and enzyme treatment (reduction/alkylation/trypsin digestion and desialylation) parameters within the automated sample preparation system. Additionally, we optimized LC-MS/MS parameters (e.g., transition selection and collision energy) for a surrogate AFP peptide and AFP glycopeptides (Supplemental Data Table S4 and Fig. S1). Next, we validated the LC-MS/MS assay over 2 months following integrated guidelines and compared the assay’s performance with that of conventional immunoassays (e.g., LiBA). Finally, we evaluated the diagnostic performance in discriminating non-HCC and HCC patients.

Fig. 1. Overview of clinical assay development, validation, and diagnostic performance evaluation.

Clinical assay validation

Analytical validation is summarized in Table 1. The analytical measurement range for AFP, including the LLOQ and ULOQ, was 0.07–8,600 ng/mL, with a linear fit (R²>0.99), meeting the signal-to-noise, imprecision, and accuracy criteria (Supplemental Data Tables S5–S7 and Figs. S2 and S3). The range for glycosylated AFP without or with fucose was 0.03–2,150 ng/mL, spanning five orders of magnitude (Supplemental Data Fig. S2). Analytical sensitivity and specificity for the LLOQ met guideline criteria (Supplemental Data Tables S8 and S9). Imprecision and accuracy were analyzed across five QC samples over 2 months, with repeatability, between-day, and within-laboratory precision <17.1% (Supplemental Data Table S10, A: Imprecision). Accuracy, focusing on imprecision, was acceptable (Supplemental Data Table S10, B: Accuracy). Assay recovery ranged from 92.8% to 114.4% for AFP, from 88.9% to 93.9% for glycosylated AFP without fucose, and from 84.8% to 104.0% for glycosylated AFP with fucose (Supplemental Data Table S11). Accuracy of the assay was maintained even after 20,000-fold dilution (Supplemental Data Table S12). No carryover was observed (Supplemental Data Table S13).

Table 1. Summary of the analytical performance of AFP and glycosylated AFP.

Validation factors	AFP	Glycosylated AFP
Linearity (R2)	0.9989	0.9978 (fucosylated AFP)
LLOQ (ng/mL)	0.07	0.07
ULOQ (ng/mL)	8,600	2,150
Repeatability (CV%)	<8.2	<16.7
Between-day precision (CV%)	<13.3	<6.3
Within-lab precision (CV%)	<15.6	<17.1
Accuracy (%Bias)	–7.2 to 14.4	0.1–21.3 (AFP-Fuc%)
Recovery (%)	92.8–114.4	100.1–108.6 (AFP-Fuc%)
Matrix effect (accuracy, %)	100.5–109.9	89.7–94.0 (AFP-Fuc%)
Carryover (%)	13.3	0
Immunoprecipitation recovery (mean, %)	105	Not determined
Stability (accuracy <20%)	Maintained at RT, 4°C for 7 days, at –20°C for 4 weeks, up to three freeze–thaw cycles
Dilution integrity (accuracy <20%)	Maintained up to 20,000-fold

Abbreviations: LLOQ, lower limit of quantification; ULOQ, upper limit of quantification; AFP, alpha-fetoprotein; RT, room temperature.

Method comparison

LC-MS/MS assay performance was compared with that of LiBA. In total, 126 samples were selected, with AFP concentrations of 0.7–682 ng/mL and fucosylated AFP ratios of 0%–98.2% based on LiBA. The correlation between the two assays was high for both AFP and AFP-Fuc%. For AFP, the Pearson correlation coefficient between LC-MS/MS and LiBA was R=0.93, with a Deming regression slope of 1.19 [95% confidence interval (CI): 1.06–1.32] and intercept of –5.46 (95% CI: –20.67 to 9.75). The 95% CI for the slope did not include 1, indicating a proportional difference, whereas that for the intercept included 0. The mean bias, including the 95% CI, was –12.15 (Fig. 2A and 2B and Supplemental Data Table S15). For the fucosylated AFP ratio, the Pearson correlation coefficient between LC-MS/MS and LiBA was R=0.99, with a Deming slope of 1.02 (95% CI: 1.01–1.04) and intercept of 1.70 (95% CI: 0.65–2.75). The 95% CI for the slope and intercept did not include 1 and 0, respectively, suggesting a significant proportional and constant bias between the methods. The mean bias, including the 95% CI, was –7.06 (Fig. 2C and 2D and Supplemental Data Table S15).

Fig. 2. Method comparison for AFP and AFP-Fuc%. (A, B) Comparison of LiBA and LC-MS/MS for AFP. (C, D) Comparison of LiBA and LC-MS/MS for AFP-Fuc%. (A, C) Correlation based on Deming regression. (B, D) Bias estimation based on Bland–Altman plots.

Abbreviations: AFP, alpha-fetoprotein; AFP-Fuc%, fucosylated AFP percentage; LiBA, liquid-phase binding assay; LC-MS/MS, liquid chromatography–tandem mass spectrometry.

Sensitivity comparison with LiBA

LC-MS/MS showed superior sensitivity for glycosylated AFP with fucose over LiBA. To compare sensitivity, we first tested calibrators with low AFP concentrations and then clinical sera (N=525) (Fig. 3). LC-MS/MS detected all diluted samples, whereas LiBA missed glycosylated AFP with fucose in samples with AFP concentrations <0.7 ng/mL (Fig. 3A). In clinical sera, LiBA missed glycosylated AFP with fucose in 309 out of 525 samples, whereas LC-MS/MS detected it in all samples (Fig. 3B). The median values filtered by the 10% fucosylation cut-off were zero for all clinical groups, indicating high undetectable rates for LiBA (Fig. 3C). In particular, the undetectable rates were 98%, 81%, and 64% in the healthy, HBV, and liver LC groups, respectively. Even in the HCC group, undetectable values accounted for 51%, 35%, and 25% of stage-1, stage-2, and of stage-3 HCC samples, respectively. However, the LC-MS/MS assay for AFP-Fuc% reported no missing values for any of the clinical samples (magnification in Fig. 3C).

Fig. 3. Sensitivity comparison of LC-MS/MS and LiBA for the quantification of fucosylated AFP. (A) Calibrators with low AFP concentration and median-level AFP-Fuc% (10 points, 5 days, fucosylation: 48.2%, AFP range: 0.1–1.3 ng/mL, mean concentrations with standard errors shown as error bars). (B) Clinical sera (N=525, fucosylation range: 0.3%–98.9%, AFP range: 0.5–14,189 ng/mL) and clinical sera filtered based on <10% fucosylation (dashed line box, N=409). (C) Clinical serum groups classified based on clinical characteristics, and clinical serum groups filtered based on <10% fucosylation (dashed line box) (1: healthy, 2: chronic hepatitis, 3: liver cirrhosis, 4: HCC stage 1, 5: HCC stage 2, 6: HCC stage 3).

Abbreviations: AFP, alpha-fetoprotein; AFP-Fuc%, fucosylated AFP percentage; LiBA, liquid-phase binding assay; LC-MS/MS, liquid chromatography–tandem mass spectrometry; HCC, hepatocellular carcinoma; N, number of samples.

Diagnostic power comparison with LiBA

For all comparisons between non-cancer and cancer groups, AFP-Fuc% consistently outperformed AFP-L3 when analyses were limited to the lower 30% of AFP concentrations or when various AFP cut-offs for low AFP concentrations were applied (Fig. 4 and Supplemental Data Table S16). Specifically, at AFP concentrations <2.3 ng/mL, AFP-Fuc% had an AUPRC of 0.478, which was significantly higher than the 0.222 recorded for AFP-L3 (ΔAUC=0.2564, 95% CI: 0.1186–0.4055; Fig. 4A and Supplemental Data Table S16). The diagnostic performance measured by AUROC significantly improved from 0.511 for LiBA to 0.710 for LC-MS/MS (N=165, P<0.001; Fig. 4E and Supplemental Data Table S16). Similar improvements in diagnostic accuracy were consistently observed with other AFP cut-offs (7, 10, and 20 ng/mL; Fig. 4B–4D and 4F–4H).

Fig. 4. Evaluation of diagnostic performance based on PRC (A–D) and ROC (E–H) analysis comparing LC-MS/MS and LiBA across various AFP concentration cut-offs. (A, E) AFP <2.3 ng/mL, (B, F) AFP <7 ng/mL, (C, G) AFP <10 ng/mL, (D, H) AFP <20 ng/mL.

Abbreviations: PRC, precision-recall curve; ROC, receiver operating characteristic curve; AFP, alpha-fetoprotein; LiBA, liquid-phase binding assay; LC-MS/MS, liquid chromatography–tandem mass spectrometry; N, number of samples.

In differentiating early-stage HCC (stages 1 and 2) from chronic hepatitis or LC groups (AFP concentrations <7 ng/mL), LC-MS/MS demonstrated significantly improved diagnostic precision over LiBA, as indicated by the higher AUPRC values (Fig. 5A and 5B and Supplemental Data Table S16). AUROC values were significantly higher for LC-MS/MS than for LiBA when distinguishing HCC stages 1 and 2 from chronic hepatitis (P=0.0078), whereas the improvement in differentiating HCC from LC was not significant (P=0.2349; Fig. 5C and 5D, Supplemental Data Table S16).

Fig. 5. Evaluation of diagnostic performance based on PRC (A, B) and ROC (C, D) analysis comparing LC-MS/MS and LiBA at the AFP cut-off (7 ng/mL) differentiating between chronic hepatitis and HCC stage 1, 2 (A, C) and between liver cirrhosis and HCC stage 1, 2 (B, D).

Abbreviations: PRC, precision-recall curve; ROC, receiver operating characteristic curve; LC-MS/MS, liquid chromatography–tandem mass spectrometry; LiBA, liquid-phase binding assay; AFP, alpha-fetoprotein; HCC, hepatocellular carcinoma; N, number of samples.

AFP-Fuc% demonstrated superior diagnostic performance over AFP-L3, with improved accuracy (1.9%), sensitivity (4.3%), and precision (0.8%) at a specificity of 96.2% (Supplemental Data Table S17). The cut-off value for LC-MS/MS (9.2%) was slightly lower than that for LiBA (10%) at the same specificity. Accordingly, AFP-Fuc% measured by LC-MS/MS reclassified 10 false negatives in AFP-L3 as true positives, increasing the true positives from 99 to 109 and reducing the false negatives from 136 to 126.

DISCUSSION

Serological cancer biomarkers, such as AFP-L3, are often glycoproteins, and their fucosylation is considered a critical disease-related post-translational modification (PTM). Current diagnostic assays primarily rely on lectin-based methods targeting core-fucosylation of AFPs in serum and face significant limitations, including cross-reactivity with highly abundant other glycoproteins and reduced sensitivity, particularly when the level of PTM or the target protein concentration is low [12, 35, 36]. These issues compromise the diagnostic accuracy of assays, particularly in early-stage disease or low-AFP cases.

MS-based methods offer enhanced diagnostic sensitivity for HCC and overcome the limitations of lectin-based assays. However, existing MS methods still rely on lectins and require deglycosylation that may introduce further inaccuracies as the calculation of fucosylation ratios is complicated because of peptide conversion (Asn → Asp, ~1-Da increase) and overlapping isotopic spectra with non-glycosylated peptides, leading to intensity miscalculations and inflated values [37, 38]. Furthermore, the calculation often involves using different peptide species for the numerator and denominator, which can result in intensity miscalculations and, consequently, inflated values surpassing 100% (Supplemental Data Table S18).

Our MS method, based on direct glycosylated AFP peptide quantification without the use of lectin, instead of relying on AFP glycopeptide deglycosylation and lectin-based enrichment, demonstrated superior sensitivity than LiBA, achieving the lowest LLOQ (0.07 ng/mL) value. Lectin-based assays, including those incorporating MS, previously exhibited low detection rates for AFP-L3, identifying only 46% of 400 clinical samples [20]. This limitation may lead to significant differences in diagnostic performance. Furthermore, methods utilizing LCA enrichment showed insufficient diagnostic accuracy in samples with low AFP concentrations (<7 ng/mL). For samples with low AFP concentrations (<2.3 or <7 ng/mL), our method showed a significant improvement in diagnostic performance over a previous method relying on LCA enrichment (Supplemental Data Fig. S6). Similarly, techniques based on LC in parallel reaction monitoring mode employing high-resolution MS without lectin enrichment have successfully distinguished HCC patients from others with high diagnostic accuracy [11, 12].

We developed an unbiased clinical LC-MS/MS assay that eliminates the disadvantages of lectin-based enrichment while taking advantage of the high sensitivity and accuracy of MS. Our method has several significant advantages. It exhibited substantially higher analytical sensitivity than LiBA, with 10 times lower LLOQ for AFP. It had a wider linear dynamic range, enhancing its clinical applicability. Most importantly, our method detected fucosylated AFP (AFP-Fuc%) even at very low AFP concentrations across disease groups. In contrast, LiBA failed to detect fucosylated AFP in 59% of clinical samples, including a significant portion of HCC cases, particularly at AFP concentrations <0.7 ng/mL, despite its reported limit of detection of 0.3 ng/mL. Similarly, previous lectin-based MS approaches showed a low detection rate, missing over half of clinical samples [20]. In our LC-MS/MS analysis, no missing values were reported. This detection gap is largely attributable to the limitations inherent to lectin-based methods, which often require additional lectin treatment to enrich specific glycoforms but struggle with low-abundant targets and suffer from insufficient sensitivity, particularly in early-stage disease or low-AFP cases [9, 19]. Because of inconsistent detection in early-stage HCC or in LC patients with low AFP concentrations, the clinical utility of AFP-L3 has been debated.

Our clinical data revealed an inverse relationship between AFP-L3 non-detection rates and AFP concentrations. As disease severity and AFP concentrations increased, AFP-L3 became more detectable, indicating that the analytical sensitivity of LiBA may be insufficient, particularly for early-stage HCC detection. In contrast, our LC-MS/MS method consistently quantified AFP-Fuc% across all disease stages and AFP concentrations, minimizing diagnostic blind spots. AUPRC and AUROC analyses confirmed its superior diagnostic performance over AFP-L3, especially at AFP concentrations <20 ng/mL (Fig. 4 and Supplemental Data Table S16), with the largest AUPRC and AUROC differences observed at 2.3 ng/mL, where LiBA showed minimal diagnostic power. Given that a substantial proportion of HCC patients, including those awaiting liver transplantation, have AFP concentrations <20 ng/mL [39, 40], accurate detection of fucosylation at low concentrations is clinically critical. In our dataset, AFP-Fuc% demonstrated higher diagnostic precision (90%–91%) than AFP alone (72%–74%), supporting its role as a promising biomarker for early HCC. These and previous findings [11, 12] reinforce the clinical value of glycoform-specific AFP measurement in low-AFP populations.

The correlation between LC-MS/MS and LiBA was high when missing data were excluded, whereas the bias was within acceptable limits (Fig. 2, Supplemental Data Table S10). When using LiBA as the reference method, both LC-MS/MS and ECLIA showed positive biases in AFP quantification, with deviations of +12.2% and +12.4%, respectively. No significant bias was observed between LC-MS/MS and ECLIA, as the 95% CIs of their bias included zero (Supplemental Data Fig. S7 and Table S15). Thus, LC-MS/MS aligns more closely with the clinical standard, ECLIA, and the observed biases with LiBA are likely caused by systematic underestimation rather than random variability.

Our automated system (Bravo) showed several advantages over manual processing, including an eight-fold increase in sample throughput, a two-fold reduction in material costs, and a five-fold decrease in personnel workload. It also showed higher quantitation performance, with four-fold lower CVs and 3.5-fold lower bias. Additionally, recovery was significantly enhanced, with peak areas for the GYQ peptide increasing by 236%. The automated system was validated according to multiple guidelines.

Our study had some limitations. We only used samples from patients with HCC and HBV or LC, limiting the generalizability of our findings to other liver diseases such as HCV and non-alcoholic steatohepatitis. Although LC-MS/MS demonstrated statistically better performance than LiBA, the clinical significance of reclassifying 10 samples remains to be confirmed, and a prospective study is planned. Additionally, we used an IS only for AFP surrogate peptides, as heavy isotope-coded glycopeptides for internal standards exhibit low purity and high interference levels. Lastly, the cost of the assay is high because of the use of two enzymes and antibodies, which may hinder its widespread adoption in clinical settings.

In conclusion, we developed a robust clinical assay for glycosylated AFP with fucose that demonstrated higher diagnostic performance than LiBA in early HCC diagnosis, particularly at low serum AFP concentrations. Our LC-MS/MS method is a promising tool for improving early HCC diagnosis and has the potential to contribute to better patient care.

SUPPLEMENTARY MATERIALS

Supplementary materials can be found via https://doi.org/10.3343/alm.2025.0003.