Comprehensive Analysis of the Mechanism of Anoikis in Hepatocellular Carcinoma

Background Hepatocellular carcinoma (HCC), ranking as the second-leading cause of global mortality among malignancies, poses a substantial burden on public health worldwide. Anoikis, a type of programmed cell death, serves as a barrier against the dissemination of cancer cells to distant organs, thereby constraining the progression of cancer. Nevertheless, the mechanism of genes related to anoikis in HCC is yet to be elucidated. Methods This paper's data (TCGA-HCC) were retrieved from the database of the Cancer Genome Atlas (TCGA). Differential gene expression with prognostic implications for anoikis was identified by performing both the univariate Cox and differential expression analyses. Through unsupervised cluster analysis, we clustered the samples according to these DEGs. By employing the least absolute shrinkage and selection operator Cox regression analysis (CRA), a clinical predictive gene signature was generated from the DEGs. The Cell-Type Identification by Estimating Relative Subsets of RNA Transcripts (CIBERSORT) algorithm was used to determine the proportions of immune cell types. The external validation data (GSE76427) were procured from Gene Expression Omnibus (GEO) to verify the performance of the clinical prognosis gene signature. Western blotting and immunohistochemistry (IHC) analysis confirmed the expression of risk genes. Results In total, 23 prognostic DEGs were identified. Based on these 23 DEGs, the samples were categorized into four distinct subgroups (clusters 1, 2, 3, and 4). In addition, a clinical predictive gene signature was constructed utilizing ETV4, PBK, and SLC2A1. The gene signature efficiently distinguished individuals into two risk groups, specifically low and high, demonstrating markedly higher survival rates in the former group. Significant correlations were observed between the expression of these risk genes and a variety of immune cells. Moreover, the outcomes from the validation cohort analysis aligned consistently with those obtained from the training cohort analysis. The results of Western blotting and IHC showed that ETV4, PBK, and SLC2A1 were upregulated in HCC samples. Conclusion The outcomes of this paper underscore the effectiveness of the clinical prognostic gene signature, established utilizing anoikis-related genes, in accurately stratifying patients. This signature holds promise in advancing the development of personalized therapy for HCC.


Introduction
In the context of liver cancer, the most common malignant tumor is hepatocellular carcinoma (HCC), which is its predominant subtype.It ranks second in global mortality rates and represents the fourth major cause of mortality worldwide [1,2].Tis underscores the substantial burden HCC places on global public health [3].Recent advances in therapeutic strategies have contributed to a gradual reduction in both the incidence and mortality rates associated with HCC.However, addressing the high postoperative recurrence rate and the relatively low fve-year survival rate of HCC remains imperative [4,5].HCC progression is characterized by distinct early and late stages.Te accurate risk stratifcation of patients with HCC is crucial for determining the clinical prognosis and treatment recommendations for individuals with HCC.Despite the development of multiple staging and prognostic systems, achieving consensus on a unifed survival-prediction system has proven elusive [6][7][8].Terefore, elucidating the molecular mechanisms underlying HCC development, screening potential tumor-specifc biomarkers, and establishing novel prognostic models are critical for risk stratifcation, early diagnosis, treatment, and personalized therapy development for patients with HCC.
Anoikis, a type of programmed cell death, is initiated by the disruption of the interaction between cells and extracellular matrix (ECM) [9,10].Prior research has reported that anoikis is critically involved in cancer cell invasion and metastasis [11,12].Anoikis can prevent cancer cell dissemination to distant organs, limiting cancer progression [13].In the course of tumor progression, a protective barrier shields tumor cells, exerting protective efects against cell death by inhibiting the activation of molecules that initiate anoikis.Tis confers resistance against anoikis and promotes tumor cell survival [14,15].Terefore, multiple types of invasive metastatic cancer may acquire and maintain anoikis resistance through diferent mechanisms.Previous studies have reported that the activated NF-κB regulates the PI3K/ Akt pathway, a survival-promoting pathway that signifcantly contributes to tumor growth and metastasis.Tus, targeting the PI3K/AKT pathway emerges as a viable therapeutic strategy for cancer [16,17].In addition, the PI3K/ AKT pathway can suppress anoikis through the inhibition of BCL2L11, BAD, and TMPRSS9 by promoting their phophorylation [12,16].NTRK2, a type of NF-κB, is an antiapoptotic molecule that confers anoikis resistance.Previous research has reported the involvement of NTRK2 in the development and metastasis of multiple cancer types, encompassing cervical, colorectal, and gastric cancers [18][19][20].Furthermore, Mak et al. suggested that miR-141 can upregulate surviving in ovarian cancer by downregulating KLF12, contributing to tumor metastasis and anoikis resistance [21].Nevertheless, the underlying mechanisms and pathways regulating anoikis in the advancement and invasion of HCC remain unclear.It is imperative to unravel the mechanisms conferring resistance to anoikis in HCC to enhance treatment outcomes and prognosis for patients with this disease.
Tis investigation employed bioinformatics analysis to identify pivotal regulatory factors associated with anoikis in HCC.Anoikis-associated genes linked to the prognosis were clustered, and diferent anoikis-related subgroups were identifed using unsupervised clustering algorithms [22].Diferent subgroups underwent Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis to elucidate potential biological pathways associated with their distinct characteristics.Finally, prognostic genes were screened, and a prognostic risk model was constructed utilizing the least absolute shrinkage and selection operator (LASSO) regression, as well as both univariate and multivariate Cox regression analyses.Tis research highlighted that the predictive model, utilizing genes related to anoikis, could stratify the risk among patients with HCC, accurately predicting their prognosis.Tis model contributes signifcantly to the clinical management and personalized treatment strategies for individuals with HCC.

Acquisition and Processing of Data.
Te integration of comprehensive data resources was fundamental to this research, wherein RNA sequencing (RNA-seq) data, along with clinical details of individuals diagnosed with HCC, were retrieved from the Cancer Genome Atlas (TCGA).TCGA-HCC cohort comprised the data of 50 noncancerous samples and 374 HCC samples.Moreover, the GSE76427 dataset, comprising 115 HCC and 52 noncancerous samples, was acquired from the Gene Expression Omnibus (GEO).TCGA-HCC and GEO datasets were merged to obtain the HCC dataset.Te batch efects in various datasets were eliminated using the ComBat function with the R "SVA."Te HCC cohort was randomly split into validation and training groups by the "caret."Te TCGA database provided data on somatic mutation and copy number variation (CNV) for individuals with HCC.Genes related to anoikis were sourced from GeneCards [23].

Unsupervised Clustering Analysis.
Te isolation of expression levels for genes related to anoikis within the TCGA-HCC cohort preceded the determination of diferentially expressed genes (DEGs) linked to anoikis in the comparative analysis between HCC and noncancerous samples.Subsequently, unsupervised clustering analysis, facilitated by the R "ConsensusClusterPlus," was utilized to unveil expression patterns of these DEGs within the HCC cohort.Diferential expression analysis between distinct subtypes was then conducted utilizing the R "limma."

Functional Enrichment Analysis and Gene Set Variation
Analysis (GSVA).Te gene sets labeled as "c2.cp.kegg.v6.2.symbols" were retrieved from MSigDB and used in GSVA via the R "GSVA" to discern the diferential biological pathways between diferent subgroups of anoikis [24].Te R "clusterProfler" was utilized with the signifcance criterion P < 0.05 to screen for potential biological processes and KEGG pathways enriched with the DEGs [25].Te Gene Set Enrichment Analysis (GSEA) software was utilized to conduct a diferential KEGG pathway analysis, comparing the pathways between high-risk (HR) and low-risk (LR) groups.Te nominal P value and normalized enrichment score (NES) were determined to analyze enrichment levels and signifcance, respectively.

Single-Sample GSEA (ssGSEA).
Te ssGSEA algorithm was applied to determine the infltration level of 23 immune cells in HCC and their correlation with the NES.

Establishment and Verifcation of a Prognostic Gene
Signature.DEGs underwent univariate CRA to determine anoikis-associated genes with prognostic value as per the criterion of P < 0.05.Te training and validation sets have a median risk score.In this case, the patients diagnosed with HCC were classifed into risk groups, specifcally LR and HR.
A comparison of the survival rates of individuals in both risk groups was carried out utilizing Kaplan-Meier (KM) curve analysis.Te visualization of risk gene expression levels in these groups was accomplished through the application of heatmaps.Te model's predictive performance was assessed through the (a) ROC (receiver operating characteristic) curve analysis and (b) the AUC (area under the ROC curve).By performing multivariate and univariate CRA, we established the risk score's prognostic signifcance, which is independent.

Nomogram Development.
We devised a nomogram model through the amalgamation of the risk score alongside various clinical characteristics.Moreover, in view of the HCC patients, we used the nomogram in conjunction with the "rms" and "survival" tools to forecast their 5-, 3-, and 1year survival rates.Te utilization of the calibration curve allowed for the assessment of the precision and distinguishing ability of the nomogram.

Tumor Microenvironment (TME) Infltration and Somatic
Mutations in Distinct Risk Groups.Concerning the RNA (CIBERSORT) algorithm, the approximation of its relative subsets took place.Alternatively, it is also called cell-type identifcation.Accordingly, we identifed the risk score's link to infltrating immune cells.Te R "maftools" was utilized to assess somatic cell variation.Te waterfall plot was employed to present the mutation profle of individuals with HCC in both risk groups.
2.8.Drug Sensitivity Analysis.Te drug response was predicted using the Cancer Drug Sensitivity Genomics database with the R "pRRophic".Ridge regression was employed for estimating the half-maximal inhibitory concentration (IC50) values of drugs for all individuals.Estimation of the predictive accuracy was achieved through 10-fold cross-validation [26].
2.9.Single-Cell RNA-Seq Data Analysis.Te single-cell RNA-seq dataset GSE166635, encompassing the data of two HCC samples, was retrieved from GEO. Te R "Seurat" was used to preprocess the original dataset, ensuring data quality.Subsequently, cells were clustered utilizing fltered principal components.For visual classifcation purposes, the dimensionality reduction techniques of the Unifed Manifold Approximation and Projection (UMAP) were used.With a threshold of corrected P < 0.05, with regard to the immune cells, the screening of marker genes was carried out.Following that, to ascertain the category group of immune cells, we extracted marker genes associated with immune cells from PanglaoDB and cross-referenced them with the genes of every category group.Finally, this paper examined the risk genes' correlation with the single-cell subpopulation.
2.10.Western Blotting.Cells in the culture fask were subjected to a phosphate-bufered saline (PBS) rinse, followed by lysis using ice-cold lysis bufer (Solarbio, China).Te resulting lysates underwent centrifugation at 12,500 rpm and 4 °C.Te supernatant was boiled in the supersampling bufer for 10 minutes.Following that, the protein samples underwent the process of sodium dodecyl sulfatepolyacrylamide gel electrophoresis, utilizing a gel with a concentration of 12.5%.Subsequently, the separated proteins were transferred onto a polyvinylidene difuoride membrane, characterized by a specifc pore size of 0.2 µm.To block nonspecifc binding sites, the membrane was exposed to 5% skim milk.Following the aforementioned steps, the membrane underwent an overnight incubation on a shaker at 4 °C.During this incubation, primary antibodies, including anti-ETV4, anti-PBK, anti-SLC2A1 (all 1 : 1000, CST), and antiaction (1 : 5000, CST) were applied.Te appropriate dilutions were made following the provided guidelines to ensure accurate and efective antibody-antigen interactions.Subsequently, the membrane underwent a subsequent phase involving a 2-hour incubation at room temperature.During this interval, secondary antibodies were employed, and the incubation was carried out with a slow shaking technique.After washing the membrane thrice with PBS, immunoreactive signals were developed using the ECL Plus kit.Normalization of the expression levels of target proteins to the corresponding levels of action was conducted.

Clinical Specimen Collection and Immunohistochemical
(IHC) Analysis.Tis study obtained eight paired HCC tumor tissues and their corresponding noncancerous lung tissues from the Afliated Hospital of Nantong University.Approval of this investigation was provided by the Ethics Committee of Nantong University Hospital.Te provision of informed consent was considered necessary for inclusion in the study and for the use of the samples.
Te sections were dewaxed and subjected to antigen retrieval.Following blocking with bovine serum albumin, the sections were exposed to anti-PDXK primary antibodies (1 : 200) at 4 °C overnight, followed by three rinses with PBS.Subsequently, the sections underwent incubation with secondary antibodies for 30 minutes at 37 °C.Ten, the sections underwent staining with 3,3′-diaminobenzidine for 5-10 minutes and hematoxylin for 10 seconds.IHC staining scores were calculated as per the intensity and amount of staining.Te staining intensity was assessed utilizing the following distinct scores: score 0 signifying a negative result, score 1 signifying a weak intensity, score 2 signifying a moderate intensity, and score 3 representing a strong intensity.In addition, the scoring system for immunopositive cell proportions ranged from 1 to 4, with score 1 signifying 0-25% immunopositive cells, score 2 signifying 26-50% immunopositive cells, score 3 signifying 51-75% immunopositive cells, and score 4 signifying 76-100% immunopositive cells.Tis comprehensive scoring methodology aimed to capture both the intensity and extent of immunopositivity, providing a nuanced assessment of the staining results.
2.12.Statistical Analysis.R v 4.01 was used for all statistical analyses.Te comparison of means between two groups was conducted through the Wilcoxon test, while the comparison of means among multiple groups was performed utilizing the Kruskal-Wallis test.In addition, the survival rates among various groups were compared using KM analysis and the logarithmic rank test.Pearson correlation analysis was utilized to verify the correlation.Te "limma" package was utilized to construct principal component analysis (PCA) plots for diferent subgroups and risk groups.For ROC curve analysis, the R "survival" and "timeROC" were utilized.Te statistical signifcance was set at P < 0.05.

Identifcation of Anoikis-Related Subgroups in HCC.
Unsupervised cluster analysis based on 23 anoikis-related genes revealed four subgroups (clusters A, B, C, and D) (Figures 2(a), 2(b), and 2(c)).Te KM survival curve shows the survival rates of patients in the four subgroups.Cluster A was associated with the worst survival outcome (Figure 2(d)).PCA demonstrated that the four subgroups exhibited distinct clustering (Figure 2(e)).Te heatmap revealed the diferential gene expression levels among clusters A, B, C, and D (Figure 2(f )).

Correlation between Diferent Subgroups and Immune Cell Infltration and the Biological Pathways between
Distinct Subgroups.Te tumor-infltrating immune cells in the four subgroups were analyzed.Te immune cell infltration levels signifcantly varied between the four subgroups (Figure 3(a)).GSVA was executed to examine the diferential biological pathways among the four subgroups.Cluster A was remarkably correlated with various pathways, including PPAR signaling and peroxisome pathways.Cluster B was signifcantly associated with pathways, including homologous recombination and cell cycle.Cluster C was signifcantly correlated with metabolic pathways, including tyrosine, histidine, and tryptophan metabolism.Cluster D was signifcantly associated with energy metabolism pathways, including fatty acid metabolism and nitrogen metabolism (Figures 3(b), 3(c), 3(d), 3(e), 3(f ), and 3(g)).
Subjects were grouped according to their risk, specifcally low or high.As observed in the training cohort, this approach follows the median risk score.KM analysis revealed that the overall survival (OS) of LR individuals was elevated relative to the HR individuals (P � 0.007) (Figure 4(c)).Scores of individuals under the latter displayed a reduced probability of survival and poorer survival status relative to those under the former.Te heatmap illustrated remarkable variations in the expression of risk genes between two risk groups (Figure 4(d)).Te corresponding AUC values predicting 5-, 3-, and 1-year OS rates across were 0.716, 0.677, and 0.726, respectively (Figure 4(e)).
Based on the risk scores within the validation cohort, individuals were sorted according to their risks, specifcally HR or LR.Tose in the former exhibited lower OS rates in comparison to individuals in the latter (P � 0.001) (Figure 4(f )).Individuals with HR scores exhibited a lower probability of survival and a worse survival status relative to those exhibiting LR scores.Te heatmap indicated notable variations in risk gene expression between the two risk groups (Figure 4(g)).Te corresponding AUC values to predict the 5-, 3-, and 1-year OS were 0.641, 0.628, and 0.753, respectively (Figure 4(h)).

Establishment of the Nomogram.
A nomogram, combining risk scores and other clinical pathological parameters, was constructed to predict the 5-, 3-, and 1-year OS of individuals with HCC, aiming to enhance the predictive accuracy of the gene signature (Figure 5(a)).Calibration curves for 5-, 3-, and 1-year survival were constructed (Figure 5(b)).Te cumulative risk for HR individuals was substantially higher than those with LR (Figure 5(c)).

Association between Gene Signature and Immune Cell and
Somatic Cell Mutations.Subsequently, we delved into the correlation of risk genes with immune cells.Te ETV4 expression showed a negative link to the proportion of resting natural killer cells.PBK expression displayed a considerable negative association with the proportions of immune cells, encompassing resting memory CD4+ T cells.Meanwhile, PBK expression exhibited a remarkable positive association with the proportions of immune cells, such as resting memory CD4+ T cells.SLC2A1 expression exhibited

Drug Sensitivity Analysis.
To evaluate the therapeutic efects of chemotherapy drugs and targeted drugs in individuals with HCC in both risk groups, the IC50 values of various commonly used drugs were quantifed using R "pRRophic."Te response to erlotinib in the HR individuals was higher than the LR individuals (Figure 7

Validation of Risk Genes.
Te protein expression levels of risk genes were validated using IHC analyses and western blotting.IHC analysis demonstrated that the ETV4, PBK, and SLC2A1 levels in HCC samples were upregulated when compared with those in paracancerous samples (Figures 9(a), 9(b), and 9(c)).Consistently, Western blotting revealed that the ETV4, PBK, and SLC2A1 levels were upregulated in HCC samples (Figures 9(d), 9(e), and 9(f )).

Discussion
Tis research delved into the mechanism of genes related to anoikis in HCC and established a clinical predictive model.Utilizing the prognostic genes linked to anoikis, the HCC study cohort was stratifed into four subtypes (clusters A, B,         GSVA demonstrated that cluster A exhibited a remarkable association with various pathways, encompassing the PPAR signaling and peroxisome pathways.Cluster B was signifcantly associated with diferent pathways, including homologous recombination and cell cycle pathways.Cluster C exhibited a remarkable association with metabolic pathways, including tyrosine, histidine, and tryptophan metabolism.Te PPAR signaling pathway has been reported to be critically involved in multiple cellular processes, encompassing cell diferentiation, infammatory responses, modulation of glucose and lipid metabolism, immune modulation, and the intricate processes of tumorigenesis [27,28].PPAR comprises three main isoforms (PPARα, PPARβ, and PPARc).Prior research has revealed that the  [36,37].ETV4, a member of a subfamily of ETS transcription factors, is upregulated in HCC.In addition, ETV4 expression exhibits a negative association with the survival outcomes of individuals with HCC [38].Previous research has demonstrated that ETV4 can facilitate the development of HCC by activating several key oncogenes and related signaling pathways, including the MMP1, uPAR, and Wnt/β-cyclin pathways [38][39][40][41].Su et al. reported that ETV4 can enhance the angiogenic capacity of endothelial cells in the HCC microenvironment by modulating the transcriptional level of MMP14, promoting the development of HCC [42].PBK, which functions as a mitotic kinase, regulates cell survival, proliferation, growth, apoptosis, and infammation [43][44][45].PBK upregulation exhibits diagnostic and prognostic relevance in cancers.PBK upregulation is linked to the development of ovarian plasma membrane and unfavorable prognosis in patients with ovarian plasma cystic adenocarcinoma, esophageal cancer, and gastric adenocarcinoma [46][47][48].Previous studies have reported that PBK knockdown or treatment with PBK inhibitors suppresses tumorigenicity and exerts growth-inhibitory efects [49,50].Yang et al. demonstrated that PBK promotes HCC cell invasion and migration by the ETV4-uPAR signaling pathway [39].Te expression of SLC2A1 is upregulated in human tumors, such as colon and gastric cancers [51,52].SLC2A1 upregulation was signifcantly associated with poor prognosis in patients with HCC [53].Peng et al. demonstrated that YAP1 inhibition decreased the SLC2A1-mediated Warburg efect, suppressing the development of HCC [54].
Somatic mutation analysis highlighted that the mutation frequency of TP53 in HR individuals was elevated relative to LR individuals.Wild-type TP53 protein initiates and promotes apoptosis in cells with DNA damage to prevent the   Genetics Research aberrant proliferation of damaged cells [55].However, TP53 mutations inhibit apoptosis mechanisms, transforming damaged cells into cancer cells owing to apoptosis evasion.Mutant TP53 proteins lose their wild-type function and are unable to mediate apoptosis and cell cycle regulatory processes.Additionally, the nucleus may witness an accumulation of mutant TP53 proteins, a characteristic often regarded as a highly specifc marker of malignancy [56].Te most common mutations in HCC are TP53 mutations [57].TP53 assumes a crucial role in preserving genomic stability.
A defciency in the functional aspects of TP53 can result in centromere expansion, proliferation of aneuploid cells, and chromosomal instability (CIN) [34].When TP53 mutations coexist with functional defects in the oncogene pRb or defects in the spindle checkpoint, it may contribute to the upregulation of CIN and genomic instability [58].Previous studies have demonstrated that mutant TP53 proteins lose their tumor suppressor functions while acquiring functions that promote tumorigenesis [59].Tese mutations may lead to enhanced tumor cell proliferation, invasion, metastasis, and treatment resistance.Tis suggests that the outcomes of the current research are consistent with those of earlier research, demonstrating the prognostic relevance of TP53 mutations in HCC.Terefore, the detection and evaluation of TP53 mutations are important for treatment decisions and prognostic assessment in HR patients.Several limitations are associated with this study.Te analysis relied on data from public databases, and the available datasets provided limited information on the clinicopathological characteristics of patients.Consequently, it is essential to incorporate practical and critical factors for accurate predictions of survival outcomes in patients with HCC.Furthermore, the validation of risk gene expression levels was conducted through only IHC and western blotting analyses.To enhance the robustness of the fndings, increasing the sample size is imperative, and the clinical value of this prognostic model should be verifed through prospective studies.

Conclusions
In this research, four distinct anoikis-associated subgroups were identifed in HCC, and a clinical prognostic model comprising three risk genes was established.Tis model demonstrated a high degree of accuracy in stratifying HCC patients, exhibiting excellent performance in predicting their survival outcomes.Te insights gained from this research contribute valuable knowledge to the understanding of anoikis mechanisms in HCC, ofering potential guidance for the development of personalized treatments for individuals with this disease.

Figure 1 :
Figure 1: Expression and genetic changes of anoikis-related genes in hepatocellular carcinoma (HCC).(a-b) Te diferentially expressed genes (DEGs) between HCC and noncancerous samples in the cancer genome atlas (TCGA)-HCC cohort were visualized using heatmap and volcanic maps.(c) Anoikis-related prognostic genes in the HCC cohort.(d) Te changes in the copy number variation of anoikis-related regulatory factors on chromosomes.(e) CNV frequency of anoikis-related regulatory factors.Te height of the column represents the frequency of change.Missing frequency, green dot; amplifcation frequency, red dot.(f ) Tis network shows the interactions between anoikis-related genes.

Figure 3 :
Figure 3: Te diferential immune landscape characteristics and biological pathways between hepatocellular carcinoma (HCC) subtypes.(a) Te diferential abundances of infltrating immune cells in the tumor microenvironment between HCC subgroups.(b-g) Gene set variation analysis revealed the activation status of biological pathways in diferent subgroups.Heatmaps are used to visualize these biological processes (red and blue represent activated and inhibited pathways, respectively).B: cluster A vs. cluster B; C: cluster A vs. cluster C; D: cluster A vs. cluster D. E: cluster B vs. cluster C; F: cluster B vs. cluster D; G: cluster C vs. cluster D.

Figure 4 :Figure 5 :
Figure 4: Construction of a clinical prognosis model.(a) Te coefcient distribution of anoikis-related prognostic genes from least absolute shrinkage and selector operator (LASSO) regression analysis.(b) Ten-fold cross-validation of tuning parameter selection in LASSO analysis.(c)Diferential survival rates between the high-risk and low-risk groups.(d) Te distribution of risk scores and survival status among diferent risk groups in the training group, as well as the expression levels of risk genes in the high-risk and low-risk groups.(e) Receiver operating characteristic (ROC) curve analysis for predicting the 1-year, 3-year, and 5-year survival rates in the training group.(f ) Differential survival rates between the high-risk and low-risk groups in the validation group.(g) Te distribution of risk scores and survival status among diferent risk groups in the validation group, as well as the expression levels of risk genes in the high-risk and low-risk groups.(h) ROC curve analysis for predicting the 1-year, 3-year, and 5-year survival rates in the training group.

Figure 8 :
Figure 8: Single-cell RNA sequencing data analysis of risk genes for constructing gene signatures.(a) Quality control of single-cell RNA sequencing data.(b) Correlation analysis between sequencing depth and mitochondrial genes.(c) Red represents 2000 highly variable genes and highlights the top 10 highly variable genes.(d) Principal component analysis (PCA).(e) Te single-cell gene expression profle was used to determine the elbow plot of the optimal principal component.(f ) Te appropriate resolution was selected based on the clustering tree.(g) Te uniform manifold approximation and projection (UMAP) algorithm was used to display 25 cell subpopulations.(h) Twenty-fve cell subpopulations were annotated as 10 subpopulations.(i) Te expression of risk genes in various cell subpopulations.(j) Te correlation between risk genes and cell subpopulations.

Figure 9 :
Figure 9: Validation of the protein expression levels of risk genes.(a-c) Immunohistochemical analysis of the ETV4, PBK, and SLC2A1 levels in tumor and noncancerous samples.(d-f ) Western blotting analysis of ETV4, PBK, and SLC2A1 expression levels.A, D: ETV4; B,E: PBK; C,F: SLC2A1.

16
Trough the LASSO Cox regression model, the candidate genes were further narrowed down using the LASSO Cox regression model to develop 2 Genetics Research prognostic gene signatures.Te risk score was derived as mentioned in the following.Risk score � (gene 1 expression level × β1) + (gene 2 expression level × β2) + (gene 3 expression level × β3) + . . .+ (gene N expression level × βN), where β depicts the regression coefcient.