A total of 55 HCC patients admitted to the Second Affiliated Hospital of Anhui Medical University between August 2024 and July 2025 were enrolled. All patients were diagnosed on the basis of the imaging findings and/or histopathological confirmation. The inclusion criteria were as follows (1): age ≥18 years (2); newly diagnosed HCC confirmed by imaging, pathology, or postoperative recurrent HCC (3); receiving PD-1/PD-L1-based immunotherapy (including monotherapy or combination with TKI or VEGF inhibitors) (4); availability of complete clinical and immunological data before and during treatment; and (5) adequate residual peripheral blood samples remaining after routine clinical laboratory testing. The exclusion criteria included the following (1): the presence of other malignancies (2); autoimmune diseases (3); concurrent treatment with TACE; or (4) incomplete medical data.

Treatment response was assessed by contrast-enhanced CT/MRI according to the modified Response Evaluation Criteria in Solid Tumors (mRECIST) for HCC (14). Complete response (CR) was defined as disappearance of any intratumoral arterial enhancement in all target lesions; partial response (PR) as at least 30% decrease in the sum of diameters of viable (arterial enhancing) target lesions relative to baseline; progressive disease (PD) as an increase of at least 20% in the sum of diameters of viable (enhancing) target lesions recorded since treatment started; and stable disease (SD) as neither PR nor PD. ORR was defined as CR plus PR. Retrospective clinical information, including sex, age, BCLC stage, Child-Pugh classification, hepatitis B/C virus (HBV/HCV) infection status, presence of cirrhosis and history of postoperative recurrence, was collected from electronic medical records. Only discarded peripheral blood samples were used in this study. Serum AFP and PIVKA-II levels were measured before initiation of PD-1/PD-L1-based therapy and at an on-treatment time point. In contrast, white blood cell counts (including lymphocytes, monocytes, and neutrophils), cytotoxic lymphocyte immunophenotyping, and functional analysis data were collected at a single on-treatment time point after therapy initiation, with the on-treatment sampling cycle varying across patients.

Serum AFP and PIVKA-II levels were measured via the ARCHITECT immunoassay kits (chemiluminescent microparticle immunoassay, Abbott, Germany; REF: 07P9077 for AFP and 01R1774 for PIVKA-II) according to the manufacturer’s instructions. All assays were performed on the Abbott ARCHITECT i System.

White blood cell (WBC) counts, including those of lymphocytes, neutrophils, and monocytes, were measured via an automated hematology analyzer (Mindray BC-7500, Shenzhen, China) with EDTA-anticoagulated whole blood.

Whole blood samples were subjected to red blood cell lysis and subsequently stained for surface and intracellular markers. CD16 and CD56 were stained using antibodies conjugated to the same fluorochrome and acquired as a single combined CD16/CD56 channel. The detailed antibody information was provided in Table 1. Data were acquired on a BD FACSLyric flow cytometer (3-laser, 12-color, BD Biosciences, USA) and analyzed via BD FACSUITE software (v1.5, BD Biosciences).

After red blood cell lysis, washed cells were resuspended in RPMI-1640 and incubated with a PMA/ionomycin/monensin stimulation cocktail (final concentrations: PMA, 81 nM; ionomycin, 1.34 µM; monensin, 2 µM; REF: TB106008, TOIMMY BIOTECH, China) together with CD16 and CD56 antibodies conjugated to the same fluorochrome. Cells were incubated at 37 °C in a 5% CO₂ incubator for 3 h. After stimulation, cells were washed with phosphate-buffered saline (PBS) and stained with antibodies against CD45, CD3, and CD8, along with a fixable viability dye (REF: 655-0866-14, Invitrogen, USA), for 20 min at room temperature in the dark. Cells were then washed, fixed for 20 min at room temperature in the dark, and permeabilized using permeabilization buffer. Intracellular staining was performed with fluorochrome-conjugated antibodies against GZMB and IFN-γ for 40 min at room temperature in the dark. Finally, cells were washed with PBS and acquired on the flow cytometer.

Circulating IFN-γ was quantified using a multiplex cytokine detection kit (REF: 281501HN, Wellgrow, China) based on flow fluorescent immunoassay, according to the manufacturer’s instructions. Data acquisition was performed on a BD FACSLyric flow cytometer.

Correlations between immune cell functional markers and clinical parameters (AFP, PIVKA-II, and treatment response) were assessed via Spearman’s rank correlation.

Statistical analyses were performed using R (v4.5.1) and GraphPad Prism (v9). Two-group comparisons were performed using Student’s/Welch’s t tests or the Mann-Whitney U test, and comparisons involving more than two groups were performed using the Kruskal-Wallis test. Categorical variables were compared using the chi-square or Fisher’s exact test, as applicable. Data are presented as mean ± standard errors of the means (SEMs) or median (interquartile range), as appropriate. Given the right-skewed distributions of serum AFP and PIVKA-II, log10 transformation was applied in model-based analyses (multivariable logistic regression and the derived receiver operating characteristic (ROC) analyses) and in correlation analyses involving tumor marker changes to reduce skewness and stabilize variance. ROC curves for individual parameters were generated in GraphPad Prism. Model-based ROC curves were generated in R using the pROC package, and area under the curves (AUCs) were compared using DeLong’s test. Multivariable logistic regression was performed in R to assess independent associations with treatment response, reporting odds ratios (95% CIs) and Wald P-values. Covariates in multivariable logistic regression were predefined based on clinical relevance and included log10-transformed AFP, BCLC stage, Child-Pugh class, and treatment regimen. A two-sided P < 0.05 was considered statistically significant.

The clinical and laboratory characteristics of patients receiving PD-1/PD-L1-based ICT alone, ICT plus TKI, or ICT plus bevacizumab were summarized in Table 2. There were no statistically significant differences in age, sex, etiology (HBV/HCV/nonviral), BCLC stage, cirrhosis status, Child-Pugh classification, post-resection recurrence, treatment cycles, or peripheral immune cell counts (lymphocytes, neutrophils, and monocytes) among the three groups. The serum levels of AFP and PIVKA-II were also comparable across the groups.

Assessment of treatment outcomes revealed that the ICT plus TKI group achieved a higher ORR (54.5%, 12/22) compared to the ICT monotherapy group (29.4%, 5/17) and ICT plus bevacizumab group (37.5%, 6/16), suggesting that the addition of TKIs may enhance therapeutic efficacy. To identify potential immune correlates of treatment response, we compared the frequency and functional characteristics of peripheral cytotoxic lymphocyte subsets between patients treated with ICT alone (lower response rate) and those treated with ICT plus TKI (higher response rate). The analyzed subsets included CTLs, NK cells, and NKT cells.

Detailed gating strategies for surface and intracellular cytokine staining were shown in Supplementary Figures S1, S2. Flow cytometric analysis demonstrated that ICT plus TKI treatment increased the proportion of CD8⁺ T cells compared with ICT monotherapy (Figure 1A). Functional profiling revealed that only NKT cells (CD3⁺CD16/CD56⁺) exhibited a higher percentage of GZMB⁺ cells (Figure 1B). In contrast, the percentage of IFN-γ⁺ cells was elevated in ICT plus TKI group across all cytotoxic lymphocyte subsets, including CTLs, NK cells, and NKT cells (Figure 1C). No differences were observed between the ICT monotherapy and ICT plus bevacizumab groups in either proportions or functional status of cytotoxic lymphocytes, which is consistent with their comparable ORR (Figures 2A–C).

To further investigate the association between peripheral immune status and treatment response to PD-1/PD-L1-based therapy in HCC patients, patients were stratified into responders and non-responders on the basis of ORR. No significant differences were observed in the proportions of these subsets between responders and non-responders (Figure 3A).

We next assessed the functional status of cytotoxic lymphocytes by quantifying the percentage of GZMB⁺ and IFN-γ⁺ cells. Among the three cytotoxic lymphocyte subsets, only NK cells showed a significantly higher percentage of GZMB⁺ cells in responders compared with non-responders (Figure 3B). In contrast, the percentage of IFN-γ⁺ cells was markedly higher in responders across all cytotoxic lymphocyte subsets, including CD8⁺ T cells, NK cells, and NKT cells (Figure 3C).

To evaluate whether the clinical and laboratory characteristics of these patients are associated with treatment response in HCC patients, we compared baseline clinical characteristics and peripheral blood parameters between responders and non-responders. As shown in Table 3, the serum AFP and PIVKA-II levels were significantly lower in responders than in non-responders. Although there was a modest difference in age, no significant differences were observed in sex, etiology (HBV/HCV/nonviral), BCLC stage, cirrhosis status, Child-Pugh classification, post-resection recurrence, treatment cycles, or peripheral immune cell counts (lymphocytes, neutrophils, and monocytes) between the two groups.

To evaluate the discriminative performance of peripheral cytotoxic lymphocyte measurements for treatment response, we performed ROC analysis. In single-parameter ROC curves, the proportions of CTLs, NK cells, and NKT cells (Figure 4A), as well as GZMB⁺ (Figure 4B) and IFN-γ⁺ functional subsets (Figure 4C), showed varying degrees of discrimination between responders and non-responders, with IFN-γ⁺ subsets exhibiting the most favorable curve patterns. The specific AUC values and P values were summarized in Table 4. For model-based ROC analyses, the clinical model yielded an AUC of 0.626. Incorporating the frequencies of IFN-γ⁺ CTL, IFN-γ⁺ NK, or IFN-γ⁺ NKT to the clinical model increased the AUC to 0.829, 0.769, and 0.810, respectively (Figures 4D–F). AUC comparisons using DeLong’s test indicated significant improvements for all three augmented models (P = 0.0122, 0.0392, and 0.0168, respectively).

In multivariable logistic regression adjusting for AFP (log10-transformed), Child-Pugh class, BCLC stage, and treatment regimen, higher percentage of IFN-γ⁺ cytotoxic lymphocytes remained independently associated with treatment response: IFN-γ⁺ CTL% (adjusted OR = 1.083, 95% CI 1.035-1.134; P = 0.0006), IFN-γ⁺ NK% (OR = 1.049, 95% CI 1.012-1.086; P = 0.0089), and IFN-γ⁺ NKT% (OR = 1.086, 95% CI 1.033-1.143; P = 0.0013) (Table 5).

We then examined the association between IFN-γ⁺ cytotoxic lymphocyte subsets (CTL, NK, and NKT) and tumor marker dynamics. The results showed that higher percentages of IFN-γ⁺ cytotoxic lymphocyte subsets were significantly associated with declines in AFP and PIVKA-II during treatment, as indicated by more negative Δlog10(AFP + 1) (Figures 5A–C) and Δlog10(PIVKA-II+1) (Figures 5D–F) values.

Given the strong association between the percentage of IFN-γ⁺ cytotoxic lymphocytes and the treatment response, we next evaluated whether circulating IFN-γ levels might similarly serve as a potential biomarker. Unexpectedly, baseline serum IFN-γ concentrations were lower in responders than in non-responders (P < 0.05), in contrast to the elevated percentage of IFN-γ⁺ cytotoxic lymphocytes from the same patients (Figure 6A). ROC analysis of serum IFN-γ yielded an AUC of 0.67 with a P value of 0.07 for discriminating responders from non-responders (Figure 6B), indicating limited discriminatory power.