Some antibodies were labeled with commercially available metal-conjugated antibodies, while others were labeled using custom-conjugated metal-labeled antibodies according to the metal antibody labeling method provided by Standard Biotools. All antibody information is presented in Supplementary Table 1 . The Maxpar X8 Multi-Metal Labeling Kit, EQ Four Element Calibration Beads, and FixI were purchased from Standard Biotools. DPBS was purchased from Corning; EDTA anticoagulant tubes from BD; TCEP from Thermo; antibody storage solution from Candor Bioscience; and 3 k and 50 k filters from Amicon Ultra.

This study included 23 patients with hepatocellular carcinoma undergoing anti-angiogenic agent therapy in combination with immune checkpoint inhibitors at Beijing You’an Hospital. All patient information is presented in Supplementary Table 2 . These patients were treated with Sindilizumab as an immune checkpoint inhibitor, administered at a fixed dose of 200 mg via intravenous injection every three weeks. Lenvatinib or Bevacizumab was used as an anti-angiogenic agent, with Lenvatinib administered orally at 8 mg to 12 mg per day, or Bevacizumab administered intravenously at a dose of 15 mg/kg every three weeks. The treatment duration was more than 12 weeks. Based on treatment efficacy, patients were categorized into PD, SD, and PR groups. The PD group comprised patients with a lesion diameter increase of more than 20% or the emergence of new lesions. The SD group included patients with a lesion diameter reduction of less than 30%. The PR group consisted of patients with a lesion diameter reduction exceeding 30%, sustained for at least four weeks. Tumor size before and after treatment, along with patient grouping details, is presented in Supplementary Table 3 . Inclusion criteria included age over 18, clinically or pathologically confirmed primary liver cancer (Guidelines for the Management of Primary Liver Cancer v2022), stage II or III according to the diagnostic staging of hepatocellular carcinoma in China, and ineligibility for surgical treatment. Child-Pugh score ≤ 10, with no serious cardiac, pulmonary, or renal disease, and a life expectancy of at least 3 months. Eastern Cooperative Oncology Group performance status of 0 or 1, with no prior systemic therapy. All patients provided written informed consent to participate based on the principles of the Declaration of Helsinki.

First, 95 μl of L-Buffer was added to the X8-polymer tube and mixed until the polymer was completely dissolved. Then, 5 μl of a 50 mM lanthanide metal element solution was added, mixed, and incubated in a water bath at 37°C for 30 minutes. A total of 100 μg of antibody was added to a 50k filter, and R-buffer was added to reach a total volume of 400 μl. The sample was centrifuged at 12,000 g at room temperature for 10 minutes. TCEP was diluted in R-buffer to a final concentration of 4 mM (1:100). A volume of 200 μl of L-Buffer was added to a 3k filter, and the metal-polymer mixture was transferred to the 3k filter. The sample was then centrifuged at 12,000 g at room temperature for 30 minutes. Next, 100 μl of the 4 mM TCEP solution was added to the 50k filter containing the antibody, mixed, and immediately incubated at 37°C for 30 minutes. The waste in the 3k filter collection tube was discarded, 400 μl of C-buffer was added, and the sample was centrifuged at 12,000 g at room temperature for 30 minutes. Then, 300 μl of C-buffer was added to the 50k filter, and any remaining antibody was gently aspirated from the tube walls. The sample was centrifuged at 12,000 g at room temperature for 10 minutes, the waste in the 50k filter collection tube was discarded, and 400 μl of C-buffer was added for a second wash. The sample was centrifuged again at 12,000 g at room temperature for 10 minutes. After centrifugation, the 3k filter was removed, followed by the 50k filter. The correspondence between each antibody and metal was confirmed. A volume of 60 μl of C-buffer was added, and all liquid was transferred to the 50k filter, then gently mixed by pipetting. The sample was incubated in a water bath at 37°C for 90 minutes. After incubation, the 50k filter was removed from the water bath, and 300 μl of W-buffer was added. The sample was centrifuged at 12,000 g at room temperature for 10 minutes, the waste in the collection tube was discarded, and the filter was washed three times with W-buffer. After the final wash, 100 μl of antibody storage solution was added, and the filter was inverted into another collection tube. The sample was centrifuged at 1,000 g at room temperature for 2 minutes to collect the antibody. Finally, the antibody was stored at 4°C for later use.

Blood samples were collected before treatment and a few hours afterward, while samples from 10 healthy donors served as a control group. After centrifugation, the plasma layer was transferred to 0.5 ml polypropylene tubes and frozen at −80°C. The donors had not used antiplatelet drugs within the past two weeks. Blood was collected from the elbow vein using 5% EDTA anticoagulant tubes and centrifuged at 200 g for 10 minutes. The upper layer containing platelet-rich plasma was then collected.

Metal-conjugated antibodies were incubated with 200 μl of platelet-rich plasma at room temperature for 30 minutes. Subsequently, 1 ml of DPBS was added, followed by centrifugation. Next, 1 ml of Fix I fixative solution was added, mixed, and centrifuged at 800 g for 5 minutes. After three washes with double-distilled water, the platelets were mixed with 10% EQ beads and filtered through a 40 μm filter before analysis using Helios mass cytometry. For each sample, 100,000 events were collected.

The CyTOF data analysis process began with sample preprocessing using CyTOF software version 7.0. Raw FCS files were normalized using EQ Four Element Calibration Beads, resulting in standardized FCS files. These standardized data were then uploaded to the Cytobank platform (https://www.Cytobank.org) for further analysis and processing. CD41a+ platelets were isolated by applying gating, while simultaneously removing beads and debris. The gated data were subsequently exported for more comprehensive analysis using R (https://bioconductor.org/packages/cytofkit/). To enhance data visualization and understanding, Phenograph clustering and T-SNE (t-distributed Stochastic Neighbor Embedding) were used for dimensionality reduction. Additionally, statistical analysis was performed to assess clustering results and protein expression for each sample. Furthermore, Sangbox (http://www.sangerbox.com/login.html) was utilized for correlation analysis, evaluating the impact of different cell clusters and protein expression on survival outcomes.

First, we conducted a clustering analysis of platelet samples from the plasma of 23 liver cancer patients before treatment and 10 healthy donors. We identified 11 distinct platelet subpopulations ( Figures 1A, B ). Independent t-SNE heatmaps reflected the relative relationships between proteins and platelet subpopulations ( Figure 1C ). Comparing platelet subpopulations between liver cancer patients and healthy donors, we found that the proportions of Cluster 7 and Cluster 11 were lower in liver cancer patients than in healthy donors ( Figure 1D ). The heatmap displaying protein expression across platelet subpopulations revealed that Cluster 7 had the highest expression of CD107a among all subpopulations, and its expression was almost exclusively restricted to Cluster 7, with little to no expression in other subpopulations. Similarly, Cluster 11 exhibited the highest expression of CD62P, although other subpopulations also expressed CD62P, but at lower levels than Cluster 11. Notably, compared to other CD62P⁺ subpopulations, Cluster 11 showed lower CD29 expression, which can serve as a distinguishing feature for differentiating Cluster 11 from other CD62P⁺ subpopulations ( Figures 1E ). The distribution of platelet subpopulations in different cases was described using stacked bar charts ( Figures 1F ). Thus, we concluded that the state and function of platelets in the plasma of liver cancer patients differ from those in healthy donors. To further explore these differences, we grouped the pre-treatment samples of liver cancer patients according to treatment efficacy to investigate the variations in platelets.

Next, we classified the pre-treatment platelet samples from liver cancer patients into three groups based on clinical outcomes: PD (progressive disease), SD (stable disease), and PR (partial response). These samples, along with those from healthy donors, were subjected to clustering analysis, identifying a total of 12 platelet subpopulations ( Figure 2A ). Using an unpaired Wilcox test to calculate the p-values for cluster percentages, we found significant differences in cluster 6 when comparing the PD group to both the SD and PR groups ( Figure 2B ). Platelets in cluster 6 exhibited higher CD29 expression compared to other clusters, indicating that CD29+ platelet subpopulations might be related to future treatment outcomes ( Figure 2C ).

Additionally, independent t-SNE heatmaps reflected the relative relationships between proteins and platelet subpopulations ( Figure 2D ). The previously identified CD107a+ and CD62P+CD29- platelet subpopulations corresponded to clusters 11 and 3 in this clustering. Distinct differences were observed between liver cancer patient groups and healthy donors in the CD62P+CD29- platelet cluster 3. A similar trend was seen in the CD107a+ platelet cluster 11, although the distinction was only significant between the PD and HD groups ( Figure 2E ). The proportion of each platelet subpopulation was visually represented using bar charts ( Figure 2F ).

In conclusion, grouping liver cancer patients’ pre-treatment plasma samples according to treatment efficacy revealed significant differences between the PD group and other groups, primarily in the CD29+ platelet subpopulations. Therefore, this subpopulation may be associated with poor prognosis in liver cancer patients.

To determine whether combined immunotherapy and anti-angiogenic therapy affect platelet subpopulations in patients’ plasma, we analyzed platelet samples from 23 liver cancer patients before and after treatment. A direct comparison of platelet subpopulations in pre- and post-treatment plasma showed no significant differences ( Figure 3A ).

Subsequently, we separately analyzed platelet samples before and after treatment in the PD, PR, and SD groups, comparing changes in platelet subpopulations. Clustering analysis of platelets before and after treatment in liver cancer patients identified 11 subpopulations ( Figure 3B ). Given the previously observed differences in CD29+ platelet subpopulations among different patient groups, we focused on these subpopulations. t-SNE heatmaps reflected the relative relationships between CD29 and platelet subpopulations ( Figure 3C ). The expression of each subpopulation in individual samples was displayed in heatmaps ( Figure 3D ).

Comparing platelet changes before and after treatment in each group, we found that cluster 6 increased in the PD group, while cluster 11 decreased in the PR group ( Figure 3E ). Paired analysis of pre- and post-treatment platelet samples showed statistically significant differences ( Figure 3F ). Notably, both clusters 6 and 11 were CD29+ ( Figure 3G ). Analyzing the differences in protein expression levels within these two subpopulations, we found that the primary difference between the two CD29+ platelet subpopulations was the expression level of CD29, with cluster 11 showing higher CD29 expression ( Figure 3H ). Therefore, we suggest that CD29+ platelet subpopulations play a significant role in the progression and treatment of liver cancer.