Research Schematic Diagram

Figure 1 displayed the overall flow sketch map of this work.

Fig. 1

Schematic diagram of this work

Differential Expression Definition and Functional Annotation of PRGs

DEPRGs in the TCGA-LIHC cohort were extracted complying with the filtering criterion (|log2FC| > 1, p < 0.01), the results indicated that 90 PRGs were up or down-regulated in tumor tissues compared to the normal specimens in the training set. More specifically, 15 PRGs were up-regulated in normal specimens, while the remaining 75 PRGs were raised in HCC tissues relatively (Fig. 2A). PPI analysis with high confidence (0.7) was performed to describe the functional association network of these DEPRGs (Fig. 2B). Additionally, the correlation coefficients among DEPRGs were calculated that satisfied the correlation threshold (cutoff > 0.3), and as Fig. 2C exhibited, most PRGs possessed a positive regulatory relationship, except for a negative association among PLG, ORM1, SPP2, HRAS, PPIA, and ALB. Subsequently, to make a better recognition of the potential biological functions of these DEPRGs, GO enrichment analysis was exerted and the results implied that the GO terms observably enriched by PRGs were “platelet degranulation and activation” in BP, “platelet alpha granule” in CC, as well as “growth factor activity” in MF, respectively, implying that these differential genes were closely related to platelets (Fig. 2D).

Fig. 2

Expression and functional annotation of platelet-related genes (PRGs). A PRGs with differential expression (DEPRGs) in the TCGA-LIHC cohort that satisfied the screening threshold (|log2FC| > 1, p < 0.01). B Protein-Protein interaction (PPI) network of DEPRGs. C Regulatory relationship map of DEPRGs. D Functional annotation of these DEPRGs based on GO enrichment analysis

Identification of Prognostic PRGs

Based on 90 DEPRGs, we further conducted the uni-cox regression analysis (p < 0.01) to obtain prognosis-associated PRGs. As the forest plot displayed in Fig. 3A, 29 PRGs with prognostic values were identified, and except for SPP2 (HR: 0.876, 95%CI: 0.817–0.939) and GNA14 (HR: 0.558, 95%CI: 0.406–0.767), the remaining PRGs were regarded as risk factors for HCC patients in the training set. And their correlation network was also displayed in Fig. 3B. Furthermore, the influence of prognostic gene expression on OS of HCC patients was displayed with KM survival curves more intuitively (Fig. 3C).

Fig. 3

Recognition of prognosis-related PRGs. A 29 prognostic PRGs with respective hazard ratios were displayed with the forest plot via univariate cox analysis (p < 0.01). B Interaction network of prognostic PRGs. C The overall survival (OS) curves of HCC cases distinguished by the expression of these prognostic PRGs

Clustering Analysis of HCC Subtypes Based on 29 Prognosis-Related PRGs

Based on 29 prognostic PRGs, we further utilized the unsupervised clustering analysis to recognize different HCC subtypes. And two distinct clusters (cluster C1: 239 cases, cluster C2: 131 cases) were determined in the TCGA-LIHC cohort (Fig. 4A). Both OS and PFS time demonstrated that there was an obvious survival discrepancy between the two HCC subtypes, a poorer prognosis was observed in patients belonging to cluster C2 compared with those in cluster C1 (Fig. 4B). Whereafter, the intrinsic connection between HCC clusters and clinicopathological parameters was analyzed, and Fig. 4C displayed that cases in cluster C2 were closely associated with the higher expression of most PRGs and worse clinical features including T stage, tumor stage, and pathological grade, which also confirmed the poor prognosis of this HCC subtype. To further explore the discrepancies in functional pathways and immune features between different clusters, we performed GSVA and ssGSEA analyses in R software. As Fig. 4D displayed, compared to cluster C1, the C2 subtype mainly enriched in KEGG pathways like Cell cycle, Homologous recombination, DNA replication, cancer-associated (like Bladder, Pancreatic, and Renal cell carcinomas) pathways, and signaling transduction axis including NOD-like receptor and mTOR signaling pathways, implying that cluster C2 might own a close association with the occurrence and evolution of HCC. Furthermore, employing GO and KEGG analyses to make a functional annotation of the DEGs identified between two different subtypes (|log2FC| > 2, p < 0.01) and the top 10 enriched terms of both were displayed respectively (Fig. 4E-F). Moreover, according to the immune infiltrating scores calculated by the ssGSEA algorithm, several immunocytes (including aDCs, iDCs, pDCs, Macrophages, Th1 and Th2 cells, as well as Tregs) were observed that had a higher infiltrating level in cluster C2, while Mast cells had a higher infiltrating score in C1 subtype (Fig. 4G). Similarly, most immunological functions like APC co-inhibition and co-stimulation, CCR, Check-point, HLA, MHC class I, Para-inflammation, as well as T cell co-inhibition and co-stimulation, showed a higher score in cluster C2, whereas Type I/II IFN Response were mainly enriched in cluster C1 (Fig. 4H). As well known, the tumor immune microenvironment is tightly associated with tumor development and immunotherapeutic response, therefore, the differences in infiltrating immunocytes and immunological functions between two distinct HCC subtypes might be a key element determining the effectiveness of immunotherapy and prognosis of patients with HCC.

Fig. 4

Clustering analysis of prognostic PRGs in the TCGA-LIHC cohort. A The whole patients were classified into two discrepant clusters (k = 2) according to the unsupervised clustering analysis. B The OS as well as progression-free survival (PFS) curves of two distinct clusters. C Heatmap of connection between clinicopathological parameters and HCC clusters. D Gene set variation analysis (GSVA) disclosed the respective enriched functional accesses in two different clusters. E The chordal plot of the GO terms and (F) the KEGG pathways enriched by differentially expressed genes (|log2FC| > 2, p < 0.01) between two distinct clusters. G The discrepancies in the composition of infiltrating immunocytes between distinct clusters. H The boxplot displayed the discrepancies in immune-related functions between distinct clusters. (*p < 0.05; **p < 0.01; ***p < 0.001)

Development and Corroboration of Platelet-Associated Risk Signature

In this section, 29 candidate prognosis-related PRGs obtained previously were first applied to establish a risk model via the lasso cox regression analysis (Fig. S1). Then a 12 PRGs-contained risk signature was constructed according to their respective gene expression and regression coefficients as the formula indicated below: Risk Score = (0.029 * PRKCD) + (0.066 * HRAS) + (− 0.021 * SPP2) + (0.050 * TUBA4A) + (− 0.370 * GNA14) + (0.035 * EGF) + (0.066 * GNG4) + (0.109 * CFL1) + (0.028 * PPIA) + (0.231 * GNA12) + (0.150 * OLA1) + (0.080 * ANXA5) (Supplementary table: Table S3). HCC patients in the training set were subsequently classified into two distinct risk groups hinging on the median risk score (185 cases at high risk and 185 cases at low risk), KM survival curve illustrated that high-risk patients suffered a worse OS compared with low-risk ones (Fig. 5A). Moreover, similar results were further corroborated in the external datasets that the OS of high-risk patients both in the GSE14520 and ICGC-LIRI cohorts was significantly inferior to those at low risk (Fig. 5B-C). And the differences in risk scoring distribution, survival status, and risk gene expression schema between distinct risk groups within the respective cohorts were also described in Fig. 5D-F. By the way, the representative IHC staining results of these risk genes both in the normal and HCC tissues obtained from the online HPA database also proved their expression patterns from the protein level (Fig. S2). Additionally, t-SNE as well as PCA analyses also indicated that our prognostic model could effectively distinguish patients from distinct risk groups both in the training and external validation cohorts (Fig. 5G-I). To estimate the efficiency of the risk signature in prognosis prediction, we counted the AUCs in the 1-, 2-, and 3-year ROC curves respectively. The results showed that the year-standardized AUCs in the TCGA-LIHC cohort were 0.772, 0.731, and 0.717, which were 0.705, 0.672, and 0.664 in the GSE14520 cohort, as well as 0.720, 0.708, and 0.706 in the ICGC-LIRI cohort, respectively (Fig. 6A), implying that our risk signature had good accuracy and stability in prognosis assessment. Meanwhile, we also made a comparison between our risk signature with other published prediction models, including Yi’s signature [12], Wang’s signature [13], Su’s signature [14], Lin’s signature [15], and Zhang’s signature [16]. The concordance index (C-index) was used to appraise the predictive capability of distinct risk signatures, and the restricted mean survival (RMS) time was applied to compare survival differences among groups distinguished by different risk models. It was observed that our risk signature had a higher C-index and a more noteworthy survival difference compared to other signatures, indicating that our prediction model also had some advantages in horizontal comparison (Fig. 6B-C). Besides, both the 1-, 3-, and 5-year ROC curves as well as survival curves of these contrastive risk signatures were also displayed respectively (Fig. 6D-E).

Fig. 5

Establishment and verification of platelet-related risk model. A Twelve-gene included prognostic model developed via the lasso cox analysis. B-D The OS curves of patients in distinct risk groups that divided according to the model in the TCGA-LIHC (left), GSE14520 (middle), and ICGC-LIRI cohorts (right). E The risk score delamination, survival state, as well as the expression pattern of model genes in the TCGA-LIHC cohort. F The risk score delamination, survival state, as well as the expression pattern of model genes in the GSE14520 cohort. G The risk score delamination, survival state, as well as the expression pattern of model genes in the ICGC cohort. H-J PCA and t-SNE analyses between different risk groups in the TCGA-LIHC, GSE14520, and ICGC-LIRI cohorts

Fig. 6

Cross-validation of predictive efficiency of the prognostic signature. A ROC curves of HCC patients at 1, 2, and 3 years in the TCGA-LIHC (left), GSE14520 (middle), and ICGC-LIRI (right) cohorts. B C-index of each risk signature: 0.705 for our platelet signature, 0.69 for Lin signature, 0.691 for Yi signature, 0.653 for Zhang signature, 0.675 for Su signature, and 0.657 for Wang signature. C Crosswise comparison of survival differences among distinct risk signatures based on the restricted mean survival (RMS) time: Platelet signature (HR: 3.583, 95%CI: 2.508–5.118, p < 0.001), Lin’s signature (HR: 1.147, 95%CI: 1.014–1.192, p < 0.001), Yi’s signature (HR: 1.188, 95%CI: 1.132–1.247, p < 0.001), Zhang’s signature (HR: 1.595, 95%CI: 1.346–1.891, p < 0.001), Su’s signature (HR: 1.306, 95%CI: 1.175–1.452, p < 0.001), and Wang’s signature (HR: 1.492, 95%CI: 1.213–1.836, p < 0.001). D ROC curves of HCC patients at 1, 3, and 5 years in several other published prediction models. E OS curves of HCC patients stratified by other published risk signatures

Clinicopathological Parameters Correlation Analysis

Subsequently, to explore the prognostic value of our signature in patients with different clinical characteristics, we further analyzed whether there were significant connections between risk groups and clinical features. The results illustrated that the high-risk score was closely correlated with worse tumor stage (Stage II-III), T stage (T2–3), and pathological grade (G3) (Fig. 7A). The levels of pathological grade and stage increased with the elevation of risk scores, meanwhile, a conspicuous connection was also observed between risk scores and HCC clusters that cluster C2 with poor clinical outcome was markedly associated with the high-risk score (Fig. 7B), and the linear relationship among clusters, risk groups, and patients’ statuses was described with Sankey plot in Fig. 7C. In addition, the survival curves of patients with different clinical features in both risk groups were shown in Fig. S3. And the association between risk scores and clinicopathological parameters in the GSE14520 and ICGC-LIRI cohorts was also analyzed and displayed in Fig. S4. Additionally, the correlation among these risk genes both in the training and the other two external cohorts was displayed in Fig. 7D and Fig. S5. Subsequently, both uni- and multi-cox analyses were performed to assess the independent prognostic efficacy of risk scores, uni-cox analysis in the training set disclosed that both risk score and several other variables (tumor stage, T /M stage) were hazard factors for the poor prognosis, and multi-cox analysis further proved the ability of the risk score as an independent hazard factor (p < 0.001, HR = 4.331, 95% CI: 2.571–7.293). Meanwhile, in the GSE14520 cohort, both BCLC stage (p = 0.032, HR = 1.516, 95% CI: 1.037–2.216), as well as the risk score (p = 0.048, HR = 1.665, 95% CI: 1.005–2.759) were demonstrated to be independent hazard factors via the multi-cox analysis, the analogous results were also discovered in the ICGC-LIRI cohort that the risk score (p = 0.046, HR = 2.007, 95% CI: 1.012–3.984), tumor stage (p = 0.002, HR = 2.074, 95% CI: 1.316–3.270), and pathological grade (p = 0.028, HR = 2.216, 95% CI: 1.088–4.513) were verified to be independent hazard factors, while gender (p = 0.003, HR = 0.304, 95% CI: 0.140–0.660) was thought as a protective factor for patients’ OS (Fig. 7E). In a word, these findings indicated that our 12-gene-included risk signature was closely associated with clinical characteristics, which also had a fine predictive capacity and was expected to act as a potential prognostic indicator for HCC patients.

Fig. 7

Clinical significance of the risk signature. A Association between clinicopathological parameters and risk scores in the TCGA-LIHC cohort (upper), percentage of tumor stage, pathological grade, and T stage between different risk groups (below). B Boxplots of distribution of risk scores in patients with distinct clusters, tumor stage, pathological grade, as well as T stage in the TCGA-LIHC cohort. C Sankey diagram displayed the potential liner link among the cluster, risk status, and patients’ survival state. D Correlation matrix among risk genes in the prognostic model. E The risk score calculated by the regression coefficient of each model gene was demonstrated to be an independent hazard factor for patients’ prognosis both in the TCGA-LIHC, GSE14520, and ICGC-LIRI cohorts. (*p < 0.05; ***p < 0.001)

Functional Analysis Based on Platelet-Related Risk Signature

Considering the potential discrepancies in biological processes and functional pathways between distinct risk groups in the TCGA-LIHC cohort, we exerted both GSVA and GSEA analyses to make an individualized investigation. The GSVA (GO part) results illustrated that the conspicuously enriched GO terms in patients at high risk were “regulation of protein localization and folding”, “regulation of cell cycle G2-M phase transition”, and “mitotic process” in BP, “DNA polymerase binding” and “ubiquitin-like protein conjugating enzyme activity” in MF, and “anaphase-promoting complex” in CC, implying that high-risk score was tightly connected with cell cycle and mitotic process, which might be implicated in the mediation of tumorigenesis and progression. While the primary GO terms enriched in patients at low risk included “metabolic-related process”, “coagulation and fibrinolysis process” in BP, “enzymatic activity” in MF, and “platelet dense granule lumen” in CC, indicating that low-risk score was tightly connected with platelet and metabolic related processes. Moreover, several noteworthy KEGG pathways like “cell cycle”, “DNA replication”, “RNA degradation”, as well as “ubiquitin-mediated proteolysis” were significantly concentrated in the high-risk group, by contrast, multiple metabolic-associated processes like “nitrogen metabolism”, “amino acid and lipid-related metabolism”, and “drug metabolism cytochrome P450” were primarily concentrated in the low-risk one (Fig. 8A). Subsequently, GSEA analysis was used to validate the biological annotation obtained above. It was observed that the major KEGG pathways concentrated in patients at high risk were “cytokine-cytokine receptor interaction”, “ECM receptor interaction”, and “neuroactive ligand-receptor interaction”, whereas “complement and coagulation cascades”, “drug metabolism cytochrome P450”, as well as “fatty acid metabolism” were mainly concentrated in those at low risk (Fig. 8B). GO analysis indicated that the top terms correlated with high-risk scores were “humoral immune response”, “phagocytosis”, and “immunoglobulin complex”, while “alpha amino acid catabolic process”, “drug metabolic process”, and “epoxygenase P450 pathway” were the main terms correlated with low-risk scores (Fig. 8C). Additionally, 328 DEGs were screened out between two risk groups with the threshold |log2FC| > 2 and p < 0.01, and GO and KEGG analyses were also employed according to these DEGs and the results were displayed in Fig. 8D. Similarly, the results of both GSEA and GO/KEGG analyses in the GSE14520 and ICGC-LIRI cohorts were also shown in Fig. S6.

Fig. 8

Functional enrichment analysis between distinct risk groups in the TCGA-LIHC cohort. A GSVA revealed the discrepancies in GO terms and KEGG pathways between different risk groups based on the platelet-related risk signature. B The respective enriched KEGG pathways in both risk groups according to the GSEA method. C The respective enriched GO terms in both risk groups according to the GSEA method. D Bubble plots of KEGG pathways as well as GO terms enriched by differentially expressed genes (|log2FC| > 2 and p < 0.01) between two distinct risk groups based on the platelet-related risk signature

Investigation of Immune Characteristics

According to the immunological landscape depicted by multiple immune algorithms, an observable discrepancy in infiltrating immunocytes between distinct risk groups was discovered that the infiltrating level of immunocytes was positively associated with the risk score (Fig. 9A). And ssGSEA method-based immune analysis further confirmed that most immunocytes (including aDCs, iDCs, Macrophages, Tfh and Th2 cells, as well as Tregs) were calculated with a higher infiltrating score in patients at high risk, inversely, other immunocytes like Mast cells and NK cells displayed a higher infiltration level in those at low risk. With regard to immunological functions, such as “APC co-stimulation”, “CCR”, as well as “MHC class I” were linked to high-risk scores, while functions like “cytolytic activity”, and “Type I/II IFN Response” were linked to low-risk scores (Fig. 9B). These findings were generally consistent with the consequences obtained in the previous clustering analysis, implicating that these observations acquired in both subgroup analyses were convincing to an extent. Additionally, the correlation coefficients between risk PRGs and immunocytes were calculated respectively and described in Fig. 9C. And the discrepancies in infiltrating immunocytes or immunological functions in the other two validation cohorts were also investigated and exhibited in Fig. S7. Furthermore, the discrepancies in the expression level of immune checkpoint genes, as well as HLA genes which were connected with antigen presentation and immune response, were also displayed in Fig. 9D. Nearly all of these genes showed a higher expression level in the high-risk group, indicating that there might be a potential difference in immune status of two different groups distinguished by individual risk scores.

Fig. 9

Immune landscape of patients in the TCGA-LIHC cohort classified by platelet-related prognostic signature. A Heatmap (left) and lollipop plot (right) of differences in infiltrating immunocytes between distinct risk groups. B Comparison of scores of immunocytes as well as immunological functions between two groups based on the ssGSEA algorithm. C Correlation matrix between model genes and infiltrating immunocytes. D Boxplots of the discrepancies in the expression of immune checkpoint and human leukocyte antigen (HLA) genes between distinct groups. (*p < 0.05; **p < 0.01; ***p < 0.001)

Genetic Mutation and Drug-Sensitive Analysis

Considering the potential value of somatic mutations in tumor progression and individual clinical treatment, we analyzed and depicted the somatic mutation landscape of patients in the TCGA-LIHC cohort in whole and different risk groups, separately. Waterfall plots respectively displayed the somatic mutation feature of patients in the entire cohort (Fig. 10A), as well as those in two distinct risk groups (Fig. 10B). Briefly speaking, patients at high risk were demonstrated to be accompanied by a higher mutation frequency compared with low-risk ones, and missense mutation was the most frequent mutation pattern in two risk groups. Specifically, TP53 with missense mutation was markedly frequent in patients at high risk, whereas AXIN1 with frameshift insertion was relatively common in patients at low risk, respectively.

Fig. 10

Tumor mutation and drug sensitivity analyses of patients at different risk statuses. A Genetic mutation landscape in the entire TCGA-LIHC cohort, with the top 30 genes displayed in the waterfall plot. B Waterfall plots showed the genetic mutation features both in the high- (left) and low-risk (right) groups. C Violin plot of the comparison of the TIDE scores between distinct risk groups (left) and their liner dependence (right). D Sensitivity prediction of chemotherapeutic regents in different groups. (***p < 0.001)

TIDE, represents tumor immune dysfunction and exclusion, is an algorithm usually employed to evaluate the possibility of immune escape and then make a prediction of immunotherapy response [17]. Here, we explored the relationship between the risk score and TIDE score to investigate the potential significance of the risk model in immunotherapy prediction. The results indicated that there was an obvious negative association between two variables (R = − 0.31, p = 7.1e− 10), patients at low risk had a higher TIDE score, implying that our risk signature possessed the capacity of immunotherapy prediction and patients at low risk might be more susceptive to immunotherapeutic compared with high-risk patients (Fig. 10C). Subsequently, the correlation between risk scores and IC50 values of frequently used chemotherapeutic medicaments was analyzed to explore the feasible sensitive agents. As Fig. 10D showed, a lower IC50 value of gemcitabine, doxorubicin, bortezomib, bleomycin, bicalutamide, bexarotene, mitomycin C, and obatoclax mesylate was observed in the high-risk group, implicating that patients at high risk might benefit from these chemotherapeutic agents, while patients at low risk might profit from the following drugs, including erlotinib, docetaxel, metformin, cyclopamine, bosutinib, axitinib, nilotinib, and gefitinib.

In Vitro Functional Verification

To further investigate the interaction between tumor cells and platelets in vitro, the platelets from human peripheral blood were first extracted and isolated and then co-cultured with HCC cells (HepG2 and MHCC97H) for 48 h. As Fig. 11A exhibited, the consequences of RT-qPCR demonstrated that the mRNA expression levels of PRKCD, HRAS, TUBA4A, EGF, GNG4, CFL1, PPIA, GNA12, OLA1, and ANXA5 were distinctly elevated after platelet stimulation, whereas the expression of SPP2 as well as GNA14 displayed a significant decreasing trend, which was precisely consistent with the expression patterns of these risk genes in our prognostic signature. Meanwhile,

Fig. 11

A Effects of in vitro platelet stimulation on the mRNA expression levels of risk genes in HepG2 and MHCC97H cells. B Effects of in vitro platelet stimulation on the mRNA expression levels of EMT-related biomarkers in HepG2 and MHCC97H cells. (**p < 0.01; ***p < 0.001)

the expression of N-cadherin (N-ca), Vimentin (Vim), and Snail was also increased obviously, whereas the expression of E-cadherin (E-ca) displayed an opposite trend both in HepG2 and MHCC97H cells, implying that platelet stimulation could induce epithelial-mesenchymal transition (EMT) in HCC cells (Fig. 11B). Subsequently, considering the research-supported close relationship between PRKCD and platelets, we chose PRKCD as the target molecule for further exploration [18,19,20,21]. As Fig. 12A-B displayed, both the mRNA and protein expression levels of PRKCD were first knocked down by si-PRKCD transfection, and after 24 h transfection, platelets were added and co-cultured with tumor cells for another 48 h. Subsequently, we investigated the impacts of distinct transfection treatments (including si-NC (Control), si-PRKCD-1, si-PRKCD-2, si-PRKCD-1 + PLT, si-PRKCD-2 + PLT) on HCC cells’ vitality and motility. Both the CCK-8 assay and colony formation test illustrated that the proliferative vitality of two HCC cells was significantly restrained by PRKCD silencing, and analogously, this inhibitory influence could be corrected by platelet stimulation (Fig. 12C-D). Moreover, the transwell assay illustrated that the invasion and migration capabilities of HepG2 as well as MHCC97H cells were markedly suppressed by PRKCD silencing, which could be reversed by direct co-culture with platelets (Fig. 12E). And the wound healing test further demonstrated that the reduction of PRKCD could obviously attenuate the migration capacity of HepG2 and MHCC97H cells and this inhibitory influence was apparently reversed by the supplement of platelets (Fig. 12F). These findings indicated that platelet stimulation could up-regulate the expression of PRKCD, which was further confirmed to be involved in mediating tumor malignant phenotypes through in vitro experiments, implicating that PRKCD was involved in platelet-induced HCC progression. While as a positive feedback loop, whether PRKCD could affect platelet activation remained unclear. Here, we hypothesized that the expression level of PRKCD in HCC cells could affect the activation of platelets in vitro. To confirm our assumption, the expression level of cellular PRKCD was first suppressed by si-PRKCD transfection (si-PRKCD-1 with better inhibition performance was selected in this section), and the PRKCD-silencing HCC cells were then utilized to stimulate the isolated platelets by direct contact, eventually, the activation level of platelets was analyzed by flow cytometric assay. The results demonstrated that PRKCD-silencing HCC cells could effectively inhibit the activation level of platelets compared to the negative control (Fig. 13), which was a confirmation of our conjecture. In summary, these results preliminarily proved that our risk gene PRKCD could either participate in the modulation of malignant biological behaviors of cancer cells as well as regulate the activation of platelets in vitro, implicating that PRKCD might act as a key molecular bridge in the cross-talk between HCC cells and platelets and also take an essential part in the platelet-induced HCC progression. Meanwhile, to explore whether platelet stimulation can mediate HCC progression by regulating the expression levels of other risk genes, based on the previous literature reports [22, 23], we selected another two risk genes (ANXA5 and CFL1) that might be related to platelets for subsequent in vitro functional verification. And based on previous studies of these two genes in HCC [24, 25], we synthesized two siRNA sequences respectively to knockdown the expression level of ANXA5 and CFL1 to investigate their potential function. In vitro functional experiments indicated that ANXA5 was involved in the regulation of HCC cells proliferation, migration, and invasion, and knockdown of its expression could inhibit the malignant phenotype of HCC (Fig. S8), and similar results were also observed after CFL1 knockdown treatment (Fig. S9). These findings are consistent with the hypothesis that platelets may mediate the malignant progression of HCC by regulating the expression of the above risk genes. However, the results of flow cytometry showed that intervention with ANXA5 or CFL1 had no significant effect on platelet activation, implying that these two genes may not be involved in the regulation of HCC-mediated platelet activation.

Fig. 12

In vitro validation of PRKCD on HCC cells proliferation, invasion and migration. A-B Measurement of both mRNA and protein expression levels of PRKCD in HepG2 and MHCC97H cells transfected with two PRKCD siRNA sequences, si-NC was utilized as the negative control. C Effects of si-PRKCD and platelet stimulation on proliferation abilities of HepG2 and MHCC97H cells measured by the CCK-8 assay. D Effects of si-PRKCD and platelet stimulation on proliferation abilities of HepG2 and MHCC97H cells measured by the colony formation test. E The transwell assay was performed to assess the impacts of si-PRKCD and platelet stimulation on HCC cells migration (upper) and invasion (below) capacities. Scale bar: 200 μm (200 ×). F The wound healing test displayed the migration ability of HCC cells undergone different treatments. Scale bar: 100 μm (40 ×). The results were presented with representative images from three times independent replicate experiments, and all data were shown as Means ± SD. (**p < 0.01; ***p < 0.001)

Fig. 13

Flow cytometry detection of platelets after co-cultured with cancer cells. HepG2 and MHCC97H cells were transfected with si-NC and si-PRKCD respectively, platelets were added and co-incubated with them, and the collected platelets were stained using CD62P and PAC-1. (**p < 0.01)