A Novel PI3K/AKT Signaling Pathway- Associated Prognostic Signature Correlated with Immune Inltration in Hepatocellular Carcinoma

Background: Hepatocellular carcinoma (HCC) has become the third leading cause of death from cancer worldwide, and PI3K/AKT signaling pathway acts as the most common carcinogenic pathway in HCC. But few studies have reported the prognostic value of PI3K/AKT associated genes (PAGs) and their association with immune inltration in HCC. Methods: The mRNA sequencing data and clinical information were gained from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO) and International Cancer Genome Consortium (ICGC) database. The 105 PAGs gene sets were from the Gene set enrichment analysis (GSEA) website. The biological processes of differently expressed PAGs were explored by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. A four-gene prognostic signature (SFN, PRKAA2, PITX2 and CDK1) was built based on the multivariate analyses results. Its prognostic value was trained in TCGA database and tested in GEO database and ICGC database. The association between risk score and immune inltration were explored by CIBERSORT method. Results: a total of nine differently expressed PAGs were identied. GO and KEGG revealed these genes were correlated with different carcinogenic pathways. The patients were divided into high-risk group and low-risk group according to the median risk score. The high-risk group had shorter survival time compared with the low-risk group in Kaplan-Meier (KM) curves. Receiver operating characteristic (ROC) curves showed the better prognostic value of risk score (ROC=0.736) compared with other clinicopathological characteristics (AUC ≤ 0.511). The consistent results were obtained in the GEO database and ICGC database. The nomogram predicted the 1-year, 3-year, and 5-year overall survival rates in HCC. The risk score was correlated with immune inltration of monocytes and M0 macrophages, and the expression of immune checkpoints (PDCD1, CTLA4, TIM3 and TIGIT) were related to risk score. Conclusion: Our study established a novel four-gene prognostic signature by integrated bioinformatic analysis. It provided new insight into the immunotherapy targets and may direct the individualized treatment in HCC. (HRAS, MAPK8, PTEN, NGF, GRB2, RAC1, EGFR, MAPK1, AKT1 and GSK3B) were screened out in the PPI network. The univariate and multivariate Cox regression analyses were employed to identify the differently expressed genes associated with the prognosis of patients. We developed the prognostic signature based on the 4 PAGs. The independent prognostic value of 4-gene-signature was identied by univariate and multivariate regression analyses. The KM curve showed the low-risk cohort had the higher survival probability in comparison to the high-risk cohort. The area under the ROC curve indicated the 4-gene-signature had a larger prognostic value compared with other clinicopathologic characteristics. The consistent results were obtained in the GSE14520 dataset and ICGC database. Ultimately the nomogram was created to accurately predict the 1-year, 3-year, and 5-year overall survival rates in HCC. The risk score was correlated with the inltration of monocytes and macrophages M0 in HCC, and the expression levels of common immune checkpoints (PDCD1, CTLA4, TIM3 and TIGIT) was associated with the risk score.


Introduction
Hepatocellular carcinoma (HCC) has become the rank-third malignant related to cancer deaths all over the world [1]. During the early stage, it is hard to obtain a clear HCC diagnosis, and effective treatments are not reliable to the majority of the patients. Some research has pointed out the great recurrence risk outcomes in the poor prognosis in patients with HCC [2]. Hence, searching for effective biomarkers for prognosis prediction as well as target spots for the patients' treatment utilizing HCC is essential.
Phosphatidylinositol 3-kinases (PI3Ks)/protein kinase B (AKT) signaling pathway was signi cant intracellular signaling react to extracellular stimulators, and it provided the novel insight into the new strategy for cancer therapy [3]. Recently some research had highlighted that the components of the PI3K/AKT signaling pathway were altered in different human tumors [4]. In particular, the PI3K/AKT/mTOR signaling pathway was often activated in the development and progression of HCC [5].
NCAPG can promote the proliferation of HCC through PI3K/AKT/FOXO4 pathway [6]. THBS4 also functioned as an oncogene in order to regulate the FAK/PI3K/AKT pathway during the development of HCC [7]. APLN participated in the PI3K/Akt pathway promoted the progression of HCC [8]. Interleukin-8 (IL-8) induced the invasion of HCC cells by participating in PI3K/Akt pathway [9]. Some research reported that apatinib can even enhance the sensitivity of radiotherapy by affecting the PI3K/AKT pathway [10]. It was also reported that VersicanV1 activated the EGFR-PI3K-AKT pathway to accelerate the proliferation and metastasis of HCC [11]. However, the prognostic value of PI3K/AKT signaling pathway associated genes (PAGs) had not been elucidated yet.
Recent research had suggested that tumor-in ltrating lymphocytes (TILs) can inhibit tumor growth through protective immunity [12]. In some human cancers, particularly the levels of TILs were related to the patient's survival time [13][14][15]. The immune checkpoints (PD-1/PD-L1) blockade can reverse the exhausted CD8 + T cells through PI3K/Akt/mTOR pathway in gastrointestinal stromal tumors [16]. It was also reported that in breast cancer, the alteration of the PI3K/AKT signaling pathway was associated with the components of the tumor microenvironment, speci cally, the in ltration of CD8 + lymphocytes had a signi cant correlation with the mutation of the PI3K/AKT signaling pathway [17]. However, the relationship between the immune in ltration of lymphocytes and PI3K/AKT signaling pathway alterations in HCC was rarely reported.
In our study, the mRNA sequencing data and corresponding clinical information were acquired from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO) and International Cancer Genome Consortium (ICGC) database. The PAGs were obtained from Gene set enrichment analysis (GSEA) online website. Then a four-gene prognostic signature was constructed based on the results of multivariate Cox regression analysis. The prognostic value of the four-gene prognostic signature was veri ed in three independent databases involving TCGA database, GEO database (GSE14520 dataset) and ICGC database. The association among the prognostic signature and the in ltrating immune cells and common immune checkpoints expression were investigated utilizing using CIBERSOFT and ESTIMATE algorithm. We concluded that the tumor immune microenvironment of HCC can be re ected by the prognostic signature. The four-gene prognostic signature provided novel insight in predicting the therapeutic e ciency of immune checkpoints blockade in HCC, and it may provide new immunotherapy targets for HCC patients.

The differentially expressed PAGs
The "limma" package was utilized to extract the differentially expressed PAGs by assessing the mRNA sequencing data of 105 PAGs in R software 4.0.2 (version 4.0.2, https://www.r-project.org/). Cut-off criteria set as |log2FC| >2, and p < 0.05 was signi cantly. The results were visualized through the volcano, the heatmap and the boxplot. The employed R-packages included the "BiocManager", the "ggpubr" and the "pheatmap".

Function enrichment analysis of differently expressed PAGs
The underlying biological processes and signaling pathways correlated with differently expressed PAGs were analyzed by the Gene Ontology enrichment analysis (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis. The results were eventually visualized by bar plots and the bubbles by using R 4.0.2 software. The employed R-packages involved the "ggpolt2", the "Cluster Pro ler", the "enrich plot" and the "DOSE".

Constructing the PPI network
The STRING website (STRING: http://www.string-db.org/) was utilized to carry out the Protein-Protein interaction (PPI) network. The PPI network revealed the underlying interactions between the encoded proteins of differently expresses PAGs. It was eventually visualized by Cytoscape software, whose website is https://cytoscape.org/.
The establishment of four-gene prognostic signature The univariate and multivariate Cox regression analyses were employed to identify the differently expressed PAGs with prognostic value. The prognostic signature was built in accordance with the multivariate Cox regression analysis results. The formula was illustrated as follows, Risk score = the expression level of gene 1×G1 + the expression level of gene 2×G2+...+the expression level of gene n×Gn (Gn refers to regression coe cient of prognostic gene in multivariate cox regression analysis). The fourgene prognostic signature formed by PAGs was carried out by using the "glmnet" package in R 4.0.2 software. Kaplan-Meier (K-M) curve and receiver operating characteristic (ROC) curve were employed to assess the predictive prognostic value of the constructed prognostic signature. The R-package ("rms") was utilized to plot the nomogram for predicting the survival rate of HCC patients.

ONCOMINE database
The ONCOMINE database (https://www.oncomine.org/resource/login.html) [19] performs the tumorrelated analyses as an online website. The differential mRNA expression of 4 PAGs (SFN, PRKAA2, PITX2 and CDK1) were analyzed in 44 types of human cancers. The threshold parameters set as follows: p = 0.0001, fold change = 2, and gene rank of the top 10% genes. The p value was calculated utilizing the Student's t-test.

CBioPortal analysis
Cancer Genomics related cBioPortal (https://www.cbioportal.org/) [20] had been an online site capable of identifying, analyzing, and visualizing several human cancer's genomics data through a multidimensional perspective. The changes of genomic pro les involving the putative copy number alterations (CNAs), mutations, deep deletion as well as ampli cation were all analyzed by the website. The genetic alterations of differently expressed prognostic PAGs (SFN, PRKAA2, PITX2 and CDK1) were calculated by the cBioPortal website.

Human Protein Atlas online website
The Human Protein Atlas online website (https://www.proteinatlas.org/) [21] was a human protein program that originated from Swedish. It was an online platform for analyzing multiple human proteins at the aspect of cells, tissues and organs by utilizing proteomics technology. The proteomics technology was comprised of systems biology, transcriptomics, antibody-based imaging and mass spectrometrybased proteomics. But the HPA database does not include protein expression data of PITX2. Fortunately, the expression of SFN, PRKAA2 and CDK1 at the protein levels were obtained from the Human Protein Atlas online website.

Statistical analysis
The Wilcoxon signed-rank test was used to analyze the differentially expressed genes between the HCC tumor tissues and normal tissues. The mRNA sequencing data were standardized through log2 transformation. The mRNA sequencing data and clinical information were both extracted by utilizing R 4.0.2 software (https://www.r-project.org/), and the Perl languages (https://www.perl.org/) was meanwhile employed to participate in the process. The abundance of immune cells in ltration in each sample was assessed through CIBERSORT method [22], the R packages involved "parallel", "e1071" and "BiocManager". The association between the immune in ltrating cells and risk score was examined through several packages ("vioplot", "BiocManager", "ggplot2", "ggpubr", "ggExtra", "scales", "ggtext" and "limma") in R software 4.0.2.

Acquiring the differently expressed PAGs
The TCGA online website was employed to download the mRNA sequencing data and corresponding clinical information regarding 374 HCC tissue samples and 50 normal tissue samples. The clinicopathological information included age, gender, grade, TNM stage, T stage, N stage and M stage.
The 105 PAGs were obtained from the HALLMARK_PI3K_AKT_MTOR_SIGNALING gene set of GSEA (Table S1). The 105 genes had been proved to associated with the PI3K/AKT/MTOR signaling pathway. The R 4.0.2 software "limma" package and the Wilcoxon signed-rank test method were employed to identify the differently expressed PAGs between tumor tissues and normal tissues in HCC (FDR < 0.05, |log FC| > 2). The results showed 9 upregulated PAGs were identi ed and it was presented by volcano plot (Fig. 1a). The boxplot (Fig. 1c) and heatmap (Fig. 1b) were also used to prove the differential mRNA expression of the above 9 upregulated genes.

GO and KEGG function enrichment analyses
Through GO and KEGG analyses, the biological processes and underlying pathways associated with 9 upregulated genes were identi ed. The top two signi cant biological processes included the regulation of G1/S transition of mitotic cell cycle and the regulation of cell cycle G1/S phase transition (Fig. 2a). The top two signi cant signaling pathways were Rap1 signaling pathway and calcium signaling pathway (Fig. 2b).

The establishment of PPI network
The correlation among the encoded proteins was analyzed by through STRING website and the results were visualized by Cytoscape software. The detail information of PPI network was also described by Cytoscape software (Fig. 2c). We identi ed the proteins whose degree of protein interaction was more than 30 by Cytoscape software, 10 hub genes were selected on the basis of the degree of protein interaction in PPI network, which included HRAS, MAPK8, PTEN, NGF, GRB2, RAC1, EGFR, MAPK1, AKT1 and GSK3B (Fig. 2d).

Searching for the hub PAGs associated with prognosis
The univariate cox regression analysis and multivariate cox regression analysis were employed to identify the differently expressed PAGs with prognostic value. The univariate Cox regression analysis obtained six prognostic PAGs including PLCB1, SFN, PRKAA2, PITX2, CDK1 and E2F1 (Fig. 3a). The nal four prognostic PAGs (SFN, PRKAA2, PITX2, CDK1) for constructing the prognostic signature were screened out through multivariate cox regression analysis (Fig. 3b). The regression coe cients were exhibited in Table 1. Constructing the PAGs-associated prognostic signature In accordance with the four selected PAGs (SFN, PRKAA2, PITX2, CDK1) and their regression coe cients, the prognostic signature was constructed and the formula utilized was PAGs-associated prognostic signature = (0.443949 * the expression level of PITX2) + (0.204509 * the expression level of PRKAA2) + (0.222894 * the expression level of CDK1) + (0.116124 * the expression level of SFN). The risk score of patients were calculated by the formula. The four-gene prognostic signature split patients into the highrisk cohort and the low-risk cohort on the basis of risk sscore (Fig. 3d). The signi cant differential expression of PITX2, PRKAA2, CDK1 and SFN were identi ed between the high-risk cohort and the lowrisk cohort, and it was visualized by heatmap (Fig. 3c). The survival time of HCC patients was considerably reduced along with the increasing risk score in the scatterplot, and the proportion of dead samples meanwhile increased (Fig. 3e). The KM curve showed that the survival probability of HCC patients was poorer in the high-risk cohort compared with low-risk cohort (p = 4.817 E-05; Fig. 4a). The samples of the high-risk cohort and the low-risk cohort were up to 185.
The get set enrichment analysis of the prognostic signature The underlying mechanism correlated with the prognostic signature was analyzed in KEGG enrichment analysis. The results revealed that the risk score was associated with 5 upregulated biological signaling pathway and 2 downregulated biological signaling pathway. The upregulated signaling pathway involved the cell cycle, the insulin signaling pathway, the MTOR signaling pathway, the NOTCH signaling pathway and the p53 signaling pathway. While the downregulated signaling pathway included the fatty acid metabolism signaling pathway and PPAR signaling pathway (Fig. 4b).
Comparing the independent prognostic value of the prognostic signature with clinicopathological characteristics The ROC curve was performed to compare the independent prognostic value between the constructed prognostic signature and various clinicopathological characteristic. The results showed prognostic value of constructed signature (AUC = 0.736) was greater than other clinicopathological characteristics (AUC ≤ 0.511) (Fig. 4c). The ROC curve evaluated the accuracy of prognostic signature and the area below the ROC curve were all statistically signi cant (1-year AUC, 0.737; 2-year AUC, 0.704; 3-year AUC, 0.694; Fig. 4d). The univariate and multivariate cox regression analyses were carried out to investigate the independent prognostic value of the constructed signature adjusted by several clinicopathological characteristics. The clinicopathologic characteristics were consist of age, gender, grade, T stage, M stage and N stage. All results suggested that the risk score of the prognostic signature had potential to be the independent prognostic factor in hepatocellular carcinoma (hazard ratio > 1; p < 0.001; Fig. 4e-f).
The differential expression of 4 PAGs and their prognostic value in HCC The mRNA expression of SFN, CDK1, PRKAA2 and PITX2 were all elevated in HCC tissue samples compared with normal tissue samples ( Figure S1a). The consistent results were gained in ONCOMINE database ( Figure S2a), but the expression data of PITX2 wasn't involved in ONCOMINE database. The four elevated PAGs expression were all correlated with the poor survival of HCC patients ( Figure S1b). We further employed cBioPortal online website to determine the genetic alternations of four PAGs in HCC.
The results showed that the deep deletion, the ampli cation, the truncating mutation and the missense mutation were all observed among the changes, and the deep deletion was the most common ( Figure  S2c). The human protein atlas (HPA) database was also utilized to verify the elevated expression of the four PAGs at the protein levels ( Figure S2b). The expression data of PITX2 wasn't involved in HPA. We explored the expression of 3 PAGs (SFN, CDK1 and PRKAA2) at the protein levels in this website. The results highlighted that SFN, CDK1 and PRKAA2 were all elevated in tumor tissues compared with the normal tissues.
Relationship between the prognostic risk model and clinicopathological characteristics The relationship among the prognostic signature and clinicopathological characteristics was further investigated. The data composed of mRNA sequencing data and corresponding clinical information was originated from TCGA database. The results showed that the risk score was associated with the stage (p = 7.9E-05; Figure S3a), the grade (p = 1.8E-07; Figure S3b) and the T classi cation (p = 0.00015; Figure  S3c), the expression of CDK1 was also related to the stage (p = 0.0034; Figure S3d), the grade (p = 4.5E-08; Figure S3e) and the T classi cation (p = 0.004; Figure S3f), the PRKAA2 expression was associated with the stage (p = 0.022; Figure S3g), the grade (p = 0.00035; Figure S3h) and the T classi cation (p = 0.049; Figure S3i). The PITX2 expression was related with stage (p = 0.00011; Figure S3j) and T classi cation (p = 0.00094; Figure S3k). The expression of SFN was associated with the grade (p = 0.032; Figure S3l).
To validate the prognostic value of constructed prognostic signature in GSE14520 dataset and ICGC database To validate the potent prognostic value of constructed prognostic signature and verify its accuracy, we further downloaded the ICGC data and GSE14520 dataset. In the testing group of GSE14520 dataset. The HCC patients were divided into the high-risk cohort and low-risk cohort based on the value of risk score (Fig. 5c). The survival time of patients was signi cantly reduced accompanied by the increasing risk score of patients in the scatterplot (Fig. 5e), The signi cant differential expression of PITX2, PRKAA2, CDK1 and SFN were identi ed and visualized by heatmap (Fig. 5a). The KM curves indicated that high risk score was correlated with the poor prognosis of HCC patients (p = 1.812E-04; Fig. 5g). To further identify the prognostic value of constructed prognostic signature. We likewise used the data originated from ICGC database to verify it. The HCC patients were also categorized into the high-risk cohort and lowrisk cohort based on the value of risk score (Fig. 5d). The results showed the high-risk group has shorter survival time than the low-risk group (Fig. 5f), The signi cant differential expression of 4 PAGs (SFN, PRKAA2, PITX2 and CDK1) in ICGC database (Fig. 5b) were also identi ed and visualized by heatmap. In the KM curve, the high-risk group had a lower survival probability in comparison to the low-risk group (p = 6.493E-05; Fig. 5h). The ROC curve was constructed to assess the accuracy of prognostic signature by analyzing the data from both GSE14520 dataset (Fig. 5i) and ICGC database (Fig. 5j) The nomogram was constructed for predicting prognosis The nomogram was constructed on the basis of the different clinicopathology characteristics and risk scores. The clinical prognostic value of conducted prognostic signature was developed by the nomogram. The 1-year survival, 2-year survival, and 3-year survival of HCC patients were predicted by the total point (Fig. 6a). The calibration plots revealed good coherence between the realistic outcome and the prediction of nomogram ( Fig. 6c-d).
The risk score was associated the immune in ltration in HCC The concrete composition of in ltrating immune cells was calculated by utilizing CIBERSORT method, the results suggested the percentage of in ltrating immune cells in each TCGA sample (Fig. 7a). The different in ltrating immune cells had a slight correlation with each other (Fig. 7b). The samples were divided into two groups based on the median value of risk score. We also employed a heatmap to determine the difference between the high-risk group and the low-risk group (Fig. 8a). The results demonstrated that the in ltrating M0 macrophages and Monocytes were differently enriched between the two groups, and the in ltrating levels of Monocytes was higher in low-risk group. The in ltrating levels of M0 macrophages was greater in high-risk group compared with the low-risk group (Fig. 8b). It is worth mentioning that the risk score was positively correlated with B cells, CD4 + T cells, neutrophil, macrophages and Myeloid dendritic cells in TIMER database ( Figure S2a). Particularly the macrophages M0 and neutrophils were strongly associated with the risk score ( Figure S2b and S2c).
The risk score was correlated with the expression levels of immune checkpoints Recent studies indicated the application of immune checkpoints inhibitors was not e cient in HCC patients [23]. Hence, we investigated the association between the prognostic signature and the several immune checkpoints (PD-1, PD-L1, PD-L2, CTLA4, TIM3, LAG3, TIGIT, CD96). The results showed the risk score was correlated with the expression levels of PD-1, CTLA4, TIM3, and TIGIT, and the expression of above immune checkpoints was higher in high-risk group compared with low-risk group (Fig. 8c). It indicated that the risk score can re ect the expression levels of immune checkpoints and it had great predictive value for the sensitivity of immune checkpoints inhibitors in HCC.

Discussion
Although the diagnosis, prognosis prediction, and treatment strategies of HCC have made great progress, hepatocellular carcinoma remained to be the major cause of cancer death worldwide [23][24][25]. At the same time, the PI3K/AKT/mTOR signaling pathway acted as a key cellular signaling pathway associated with the appearance of several human cancers [26]. The inhibitors of the PI3K/AKT/mTOR signaling pathway were thoroughly studied and evaluated in clinical trials in human cancers [27]. It had been suggested that PI3K/AKT signaling pathway associated genes can serve as the potential biomarkers for the development of numerous human cancer [28]. However, few studies mentioned the prognostic value of PAGs in HCC. Some research indicated constructing the early predictive model for prognosis in human cancer is necessary [29], and the different therapeutic strategy should be formulated according to different prognosis of patients [30,31]. Currently, an opinion had been recommended that clinicopathological characteristics can be employed to predict prognosis of patients, but some studies highlighted that clinicopathological characteristics can't provide adequate evidence for accurate prognostic evaluation [32,33]. Thus, we constructed a novel prognostic signature based on the PI3K/AKT signaling pathway associated genes in HCC. Recent studies have shown that the immune escape and tolerance can be mediated through various signaling pathways in tumor microenvironment [34]. Hence, the association between constructed prognostic signature and immune in ltration was also determined.
In this study, we developed a 4-gene prognostic signature based on TCGA database. The prognostic value and accuracy of constructed prognostic signature were further validated in GSE14520 dataset and ICGC database. First of all, the mRNA sequencing data and corresponding clinical information were downloaded from the TCGA database, and 105 PAGs were obtained from GSEA online website. Subsequently, 9 upregulated PAGs were screened out from 374 HCC tissue samples and 50 normal tissue samples. The GO and KEGG enrichment analysis were further employed to investigate the biological processes correlated with the upregulated PAGs. The PPI network was utilized to exhibit the connection between encoded proteins, and the hub genes (HRAS, MAPK8, PTEN, NGF, GRB2, RAC1, EGFR, MAPK1, AKT1 and GSK3B) were screened out in the PPI network. The univariate and multivariate Cox regression analyses were employed to identify the differently expressed genes associated with the prognosis of patients. We developed the prognostic signature based on the 4 PAGs. The independent prognostic value of 4-gene-signature was identi ed by univariate and multivariate regression analyses. The KM curve showed the low-risk cohort had the higher survival probability in comparison to the high-risk cohort. The area under the ROC curve indicated the 4-gene-signature had a larger prognostic value compared with other clinicopathologic characteristics. The consistent results were obtained in the GSE14520 dataset and ICGC database. Ultimately the nomogram was created to accurately predict the 1-year, 3-year, and 5- year overall survival rates in HCC. The risk score was correlated with the in ltration of monocytes and macrophages M0 in HCC, and the expression levels of common immune checkpoints (PDCD1, CTLA4, TIM3 and TIGIT) was associated with the risk score.
It was proved that immunosuppressive microenvironment enhanced the evasion and immune tolerance in HCC [34]. HCC is a type of malignant tumor accompanied by chronic liver in ammation and liver cirrhosis, some HCC cases expressed gene markers indicating the existence of immune response, which provided a novel insight into immunotherapy in HCC [35]. Our results showed the immune in ltration of monocyte and macrophages M0 were higher in high-risk group, which had poorer survival in HCC. Tumor associated macrophages (TAMs) are essential elements in HCC microenvironment and lead to the poor survival in HCC patients [36]. TAMs obtained the immunosuppressive function to regulate the tumor microenvironment as they grow from monocytes to macrophages [37]. Current studies even reported that macrophages were the most common immune cells in tumor tissues and it induced the invasion, proliferation and metastasis in tumor cells, it also mediated the immune tolerance in tumor microenvironment [38][39][40]. We further calculated the correlation between the risk score and several common immune checkpoints including PDCD1(PD-1), PDL1, PDL2, CTLA4, TIM3, LAG3, TIGIT and CD96. And the results showed the expression levels of PDCD1, CTLA4, TIM3 and TIGIT were associated with risk score. Some research indicated that anti-PD-1 and anti-CTLA4 antibodies tended to be the effective clinical therapeutic targets in HCC [41], and PD-1 inhibitor (nivolumab ) can stimulate the speci c immune response with controllable side effects in tumor [42]. However, the response to immune checkpoints in HCC was different among various organs [43]. Our results hinted that our risk score can revealed the expression of immune checkpoints and explained why the e cacy of anti-immune checkpoints was low in some HCC patients. The four-gene-signature has potential to predict the immunotherapeutic e cacy of PD-1, CTLA4, TIM3 and TIGIT, and directs immunotherapeutic strategy in HCC.
Previous research had pointed out the crucial role of PRKAA2 in human cancers. PRKAA2 functioned as a member of heterotrimeric AMP-activated protein kinase (AMPK) complex, and was reported to be involved in the p53 and mTOR signaling pathways in carcinogenesis of human cancers [44]. In our study, PRKAA2 was identi ed as the PAG and determined its prognostic value. Current studies indicated that MiR-4999-5p can promote the cell growth and glycolysis by targeting PRKAA2 in colorectal cancer [45], and PRKAA2 was also dysregulated in cervical cancer cells and had potential to be the prognostic biomarker in cervical cancer [46]. Some research had likewise suggested that metformin suppressed the proliferation of HCC cells by targeting miR-378/CDK1 axis [47]. The inhibition of CDK1/PDK1/β-Catenin pathway can improve the therapeutic effect of sorafenib and suppress the proliferation of HCC cells [48]. SNHG16 activated the let-7b-5p/CDC25B/CDK1 axis to in uence the G2/M transition resulting in the accelerating progression of EMT and cell metastasis [49]. Circular RNA (circ-ADD3) suppressed the metastasis of HCC cells through the ubiquitination mediated by CDK1 [50]. We therefore can summarize that CDK1 was an important regulatory factor in the carcinogenesis of HCC. PITX2 also plays a carcinogenic role in several human cancers. The overexpression of PITX2 can induce the letrozole-resistance in breast cancer cells [51], and it also participated in the progression of ovarian cancer by promoting the proliferation and invasion of ovarian cancer cells [52]. The methylation of PITX2 also served as a prognostic biomarker in colorectal cancer [53], and it was reported to mediate proliferation and invasion of colorectal cancer cells in vitro experiment [54]. In human gliomas, SFN was reported to be the potential prognostic biomarker [55], and it provided great prognostic value in the progression of esophageal squamous cell carcinoma [56]. The overexpression of SFN was correlated with the poor prognosis in ovarian cancer and pancreatic ductal adenocarcinoma [57,58].

Conclusions
To determine the prognostic value of PAGs and their association with immune in ltration in HCC, we established a four-gene prognostic signature for predicting HCC patients' survival. The strong relationship among the risk score, immune in ltrating cells and immune checkpoints was also identi ed. In order to predict the prognosis of HCC patients, a nomogram was established based on the risk score. The great prognostic value of the four-gene prognostic signature was trained in TCGA database, and it was further tested in both GSE14520 dataset and ICGC database. It was concluded that the four-gene-signature had better prognostic value in comparison to other clinicopathological characteristics. Our study offered a novel insight into new biomarkers and therapeutic targets for immunotherapy in HCC. The association between PI3K/AKT signaling pathway alterations and immune in ltration in HCC was also identi ed. However, the practicability of 4-gene prognostic signature and its prognostic value should be validated in clinical samples.