Multi-omics analyses based on genes associated with oxidative stress and phospholipid metabolism revealed the intrinsic molecular characteristics of pancreatic cancer

Oxidative stress (OS), which impacts lipid metabolic reprogramming, can affect the biological activities of cancer cells. How oxidative stress and phospholipid metabolism (OSPM) influence the prognosis of pancreatic cancer (PC) needs to be elucidated. The metabolic data of 35 pancreatic tumor samples, 34 para-carcinoma samples, and 31 normal pancreatic tissues were obtained from the previously published literature. Pan-cancer samples were obtained from The Cancer Genome Atlas (TCGA). And the Gene Expression Omnibus (GEO), International Cancer Genome Consortium (ICGC), ArrayExpress, and the Genotype-Tissue Expression (GTEx) databases were searched for more PC and normal pancreatic samples. The metabolites in PC were compared with normal and para-carcinoma tissues. The characteristics of the key OSPM genes were summarized in pan-cancer. The random survival forest analysis and multivariate Cox regression analysis were utilized to construct an OSPM-related signature. Based on this signature, PC samples were divided into high- and low-risk subgroups. The dysregulations of the tumor immune microenvironment were further investigated. Quantitative reverse transcription polymerase chain reaction (qRT-PCR) was conducted to investigate the expression of genes in the signature in PC and normal tissues. The protein levels of these genes were further demonstrated. In PC, metabolomic studies revealed the alteration of PM, while transcriptomic studies showed different expressions of OSPM-related genes. Then 930 PC samples were divided into three subtypes with different prognoses, and an OSPM-related signature including eight OSPM-related genes (i.e., SLC2A1, MMP14, TOP2A, MBOAT2, ANLN, ECT2, SLC22A3, and FGD6) was developed. High- and low-risk subgroups divided by the signature showed different prognoses, expression levels of immune checkpoint genes, immune cell infiltration, and tumor microenvironment. The risk score was negatively correlated with the proportion of TIL, pDC, Mast cell, and T cell co-stimulation. The expression levels of genes in the signature were verified in PC and normal samples. The protein levels of SLC2A1, MMP14, TOP2A, MBOAT2, ANLN, and SLC22A3 showed up-regulation in PC samples compared with normal tissues. After integrating metabolomics and transcriptomics data, the alterations in OSPM in PC were investigated, and an OSPM-related signature was developed, which was helpful for the prognostic assessment and individualized treatment for PC.


Methods
Data acquisition.The metabolic data of 35 pancreatic tumor samples, 34 para-carcinoma samples, and 31 normal pancreatic tissues were obtained from the previously published literature, completed by Liu et al. 28 .In total, 10,028 OS-related genes and 4,479 PM-related genes were searched in the Molecular Signatures Database (MSigDB) 29 and GeneCard websites 30 .Specifically, the GeneCard platform provided 10,022 OS-related genes and 4477 PM-related genes.MSigDB platform provided two OS-related pathways (i.e.GOBP_RESPONSE_TO_OXI-DATIVE_STRESS, WP_OXIDATIVE_STRESS_RESPONSE) and two PM-related pathways (i.e.GOBP_REGU-LATION_OF_PHOSPHOLIPID_METABOLIC_PROCESS, REACTOME_PHOSPHOLIPID_METABOLISM)."GOBP_RESPONSE_TO_OXIDATIVE_STRESS" pathway involves 437 OS-related genes, and "WP_OXIDA-TIVE_STRESS_RESPONSE" pathway involves 33 OS-related genes."GOBP_REGULATION_OF_PHOSPHO-LIPID_METABOLIC_PROCESS" pathway involves 37 PM-related genes, and "REACTOME_PHOSPHO-LIPID_METABOLISM" pathway involves 211 PM-related genes.After removing duplicate genes, a final set of 10,028 OS-related genes and 4479 PM-related genes was obtained.
The gene expression data and clinical information were obtained from public databases.In total, PC datasets contained 930 PC samples made up of GSE28735, GSE57495, and GSE62452 from the Gene Expression Omnibus (GEO), MTAB-6134 from ArrayExpress, TCGA-PC samples from The Cancer Genome Atlas (TCGA), and ICGC-AU and ICGC-CA datasets from International Cancer Genome Consortium (ICGC).The Genotype-Tissue Expression (GTEx) database was searched to obtain normal pancreatic samples.
For pan-cancer summarization, mRNA expression, clinical information, single-nucleotide variation (SNV), copy number variation (CNV), and methylation data of common cancers were downloaded from TCGA database.
Identification of metabolic reprogramming in the occurrence of PC.Liu et al. detected and analyzed the metabolites in PC and presented the metabolic reprogramming in PC.We downloaded the published literature compiled by Liu et al., which contained the metabolomics data of normal and tumor tissues as supplementary materials 28 .This study involves the secondary mining of metabolomics data based on the supplementary materials provided in the article by Liu et al.After collecting and analyzing the metabolomics data, the alterations in metabolites caused by the occurrence of PC were further investigated.The detailed approaches are described as follows.The metabolic changes were investigated, and each fold change (FC) value was calculated after comparing tumor tissues with para-carcinoma tissues.Meanwhile, the differences between tumor tissues and normal tissues were also explored.On the basis of the p.adjust < 0.05 and FC > 2 or FC < 0.5, the metabolites with different contents in PC were obtained.

Identification of key OSPM-related genes in PC.
Considering the prominent roles of OS and PM in the development of PC, we integrated the gene sets corresponding to OS and PM, resulting in the formation of OSPM genes.Then the OSPM-related differentially expressed genes (DEGs) were explored in three OSPM-based cluster analysis.With the mRNA expression of the OSPM-related genes in PC, the function states of OSPM were estimated quantitatively with ssGSEA algorithm.Whether the discrepancies of OSPM could influence the survival of the sample was investigated after performing cluster analysis.During the process, "gplots" and "pheatmap" packages in R were employed to visualize the results of cluster analysis and relative expression levels of each OSPM-related genes 34 .Based on the ssGSEA scores obtained in the normal samples, PC samples were grouped into cluster1 with OSPM-active function, cluster2 with OSPM-inactive function, and cluster3 with OSPM-normal function.
Then the discrepancies in the different clusters were searched by performing the following analyses 35,36  The estimation of the tumor immune microenvironment was conducted as follows 37 : (1) ssGSEA was performed to quantify the immune-associated gene sets; (2) Spearman's correlation analysis was conducted determine the link between prognostic OSPM-related DEGs and immune cells; (3) The correlation between the OSPM function score and immune components was found by utilizing "ggstatsplot, " "data.table," "dplyr, " "tidyr, " and "ggplot2" packages in R; and (4) MCPCOUNTER, XCELL, CIBERSORT, EPIC, CIBERSORT-ABS, and TIMER were performed to assess immune responses, and the statistically significant discrepancies among different clusters were exhibited.

Group division and OSPM-related signature construction.
To develop a quantitative signature for guiding the clinical diagnosis and treatment of PC, an OSPM-related signature was constructed and validated.The training cohort (635 PC samples) included GSE57495, GSE28735, GSE62452, MTAB-6134, and TCGA-PC datasets, whereas the validation cohort (295 PC samples) included the ICGC-CA and ICGC-AU datasets.To make the signature concise and convenient, random survival forest analysis was conducted using the "random-ForestSRC" package in R to select the most appropriate genes from all the prognostic OSPM-related DEGs to gain an optimal prognostic signature 38 .Multivariate Cox regression analysis was then conducted, and the coefficients helped to obtain the risk score of each sample 39 .
The risk score corresponding to each sample was determined, and the median score helped to distinguish high-and low-risk subgroups in the training and test cohorts, respectively.Then the discrepancies of the highand low-risk subgroups were investigated in the two cohort as follows: (1) Sample division was visualized through principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE); (2) Survival probability was compared through KM analysis 40,41 ; (3) The expression levels of the genes in the signature were compared between high-and low-risk subgroups and exhibited in heatmap; (4) The diagnostic accuracy was demonstrated by developing time-dependent receiver operating characteristic (ROC) curves and obtaining the AUC values 42 ; (5) The expression levels of ICGs were compared, and the immune response was estimated through the methods mentioned above 43,44 .In addition, the GEPIA website was utilized to compare the expression levels of the model genes in PC samples with different clinical stages 45 .
Cell culture and qRT-PCR.Human pancreatic cell line (HPDE6-C7) and PC cell lines (BxPC-3, CFPAC-1, Panc-1, and Mia-Paca-2) were obtained for PCR experiments of model genes.HPDE6-C7, Panc-1, and Mia-Paca-2 were cultured in DMEM medium.BxPC-3 was maintained in 1640 medium.CFPAC-1 was maintained in IMDM medium.All these mediums were supplemented with 10% fetal bovine serum at 37 °C and 5% -CO 2 .Total RNA was extracted utilizing TRIzol reagent and synthesized into cDNA utilizing M-MLV reverse transcriptase under the instruction of the manufacturer.qRT-PCR was performed using SYBR Green assay and β-actin (Forward: CCT GGG CAT GGA GTC CTG TG, Reverse: TCT TCA TTG TGC TGG GTG CC) was selected as an endogenous reference.www.prote inatl as.org/) was utilized for further exploration about the protein levels of model genes 47,48 .Additionally, the HPA was also utilized to exhibit the cellular localization on the basis of immunofluorescence.

Identification of metabolic alterations and OSPM-related genes in PC.
The workflow is depicted in Fig. 1.The alteration of metabolites plays a vital role in tumor initiation and progression.The discrepancies of the content of metabolite between pancreatic tumor samples and para-carcinoma samples were investigated, and 99 metabolites with significant changes were exhibited in the heatmap (Fig. 2A).Similarly, 60 metabolites altered significantly between tumor samples and normal samples, as shown in another heatmap (Fig. 2B).After taking an intersection, 35 shared metabolites were obtained (Fig. 2C).Of note, the shared metabolites primarily belonged to diacylglycerol, glycerophosphocholine, and glycerophosphoethanolamine.It is well-known that PM mainly includes the metabolism of glycerophospholipids and sphingophospholipids.And diacylglycerol is the major substrates for the synthesis of phosphatidylglycerol in glycerophospholipid metabolism.Conversely,  www.nature.com/scientificreports/ of PC.Due to the influence of OS on PM, the following investigations were mainly focused on the two biological processes.
The role of OSPM in pan-cancer.Due to the significant role of the 21 prognostic OSPM-related DEGs, they were viewed as the functional genes in the OSPM progress.From a pan-cancer perspective, the general www.nature.com/scientificreports/summarization of the 21 OSPM function genes was made.First, the mRNA levels of these genes in tumor samples were compared with those in normal samples.MELK, ANLN, and TOP2A were markedly up-regulated in almost all types of tumors (Fig. 4A).Then the prognostic performances of the 21 genes were explored and exhibited in the heatmap.Genes with hazard ratio (HR) > 1 and p < 0.05 were risky genes, while those with HR < 1 and p < 0.05 were protective genes (Fig. 4B).The CNV and SNV characteristics of 21 genes in pancancer were displayed in Fig. 4C-E.Among all types of tumor samples, LUAD, LUSC, SKCM, STAD, and UCEC patients exhibited the most prominent SNV phenomena (Fig. 4D).Additionally, among the 21 genes, CTNND2 www.nature.com/scientificreports/showed the highest mutation rate in these cancers, and missense mutation was the most common mutation type (Fig. 4E).Promoter methylation levels revealed that CTNND2 exhibited hypermethylation in CESE, COAD, and HNSC.Conversely, SLPI showed hypomethylation in nearly all cancers, with the exception of PRAD (Fig. 4F).

OSPM-based cluster analysis.
To further illustrate the role of OSPM in PC, the ssGSEA was performed using a score derived from the expression of 21 OSPM function genes.Based on the enrichment scores in normal samples, 930 PC samples were divided into three clusters (C1, C2, and C3).Subsequently, the enrichment scores of the OSPM function were compared.C1, C2, and C3 were the OSPM-active, OSPM-inactive, and OSPMnormal cluster (scores: C1 > C3 > C2; Fig. 5A).The heatmap presented herein showcases the expression levels of the 21 OSPM function genes (Fig. 5B).For example, SLCO1B3 exhibited a clear upregulation in C1 and downregulation in C2 and C3 (Fig. 5B).The survival probabilities in the three clusters demonstrated statistically significant differences, with C2 showing the highest survival probability, followed by C3, and then C1 (Fig. 5C).
Identification of the OSPM-related signature.For correct establishment and complete validation of the OSPM-related signature, the training and test cohorts were grouped.In the training cohort, random survival forest analysis helped identify the most appropriate genes (i.e., SLC2A1, MMP14, TOP2A, MBOAT2, ANLN, ECT2, SLC22A3, and FGD6) related to survival of PC patients, and an OSPM-related signature was identified (Fig. 7A).The following multivariate Cox analysis helped identify the risk score of each sample.The computational formula of risk score is as follows.The risk score = 0.229994697234363*TOP2A + 0.12478779571962 3*SLC2A1 + 0.220093179394101*MMP14 + 0.0671974510904209*MBOAT2 + 0.0484008588233876*ANLN − 0 .17194413420108*ECT2+ 0.0362333570976207*SLC22A3 + 0.0192255505103104*FGD6.The median risk score of training cohort is 0.002583997.Based on the median risk score, samples were grouped into high-and lowrisk subgroups (Fig. 7B and C).The PCA and t-SNE demonstrated the distinction between the two subgroups (Fig. 7D and E).The survival analysis indicated that samples with high-risk scores had poor prognoses (Fig. 7F).The eight genes in the OSPM-related signature showed different expression in the two subgroups.All these genes had increased expression in the high-risk samples (Fig. 7G).Interestingly, the expression of ECT2, FGD6, MMP14, and SLC2A1 were statistically different in the patients with different tumor stage, which implied that ECT2, FGD6, MMP14, and SLC2A1 might be associated with the development of PC (Supplementary Fig. 5).
The AUC value of the ROC curve was 0.682 for 5-year survival in the training cohort, which implied its accurate diagnosis (Fig. 7H).The immune responses in high-and low-risk subgroups also exhibited discrepancies.As is depicted in the heatmap, the infiltration of various immune cells in the two subgroups differed from each other, which is one of the root causes of the prognosis (Supplementary Fig. 6A).Furthermore, the expression level of the ICG might be another root cause of the prognosis of PC.As depicted in Supplementary Fig. 6B, SIGLEC15, HAVCR2, PDCD1LG2, YTHDF1, CD274, CD80, TNFRSF14, TNFSF4, TNFRSF9, and CD70 showed increased expression in the high-risk subgroup.
To validate the performance of the OSPM-related signature, samples in the test cohort were grouped into high-and low-risk subgroups according to the median risk score of 0.002583997 (Fig. 8A and B).The analyses described above were performed again in the test cohort.Samples in high-and low-risk subgroups also showed distinct separation through PCA and t-SNE (Fig. 8C and D).High-risk subpopulations also had a poor prognosis (Fig. 8E).All the genes in the signature showed increased expression in the high-risk subgroup (Fig. 8F).The AUC value of the ROC curve in the test cohort were 0.753 for 5-year survival (Fig. 8G).Also, the immune responses in the test cohort are depicted in Supplementary Fig. 6C.The expression of ICGs showed a similar trend in the test cohort.TIGIT, HAVCR2, PDCD1LG2, TNFRSF4, TNFSF4, TNFRSF9, and CD70 were up-regulated in the high-risk subgroup in the test cohort (Supplementary Fig. 6D).All these results in the test cohort were similar to those in the training cohort, which indicated the satisfactory performance of the OSPM-related signature.
The expression of the eight genes in the OSPM-related signature.For further investigation of the eight genes (i.e., SLC2A1, MMP14, TOP2A, MBOAT2, ANLN, ECT2, SLC22A3, and FGD6) in the OSPMrelated signature, the qRT-PCR was conducted.And the results indicated that the expression levels of these genes in PC were different with that in normal samples.For ANLN, its expression was higher in CFPAC-1 while lower in BxPC-3 and Mia-Paca-2.For ECT2, its expression was lower in CFPAC-1, BxPC-3 and Mia-Paca-2.For FDG6 and MBOAT2, the expression levels were higher in CFPAC-1 while lower in Panc-1 and Mia-Paca-2.For MMP14 and SLC2A1, the expression levels were higher in BxPC-3 and CFPAC-1 while lower in Panc-1 and

Single-cell expression level, immunohistochemistry, and immunofluorescence of model genes in PC.
To investigate the tumor microenvironment, single-cell expression levels of the eight genes in the signature were explored.The expression of MBOAT2 primarily existed in the epithelium and the expression of MMP14 mainly existed in fibroblasts, while the expression of FGD6 primarily existed in the progenitor (Supplementary Fig. 7A).Regarding the protein levels of these genes, SLC2A1, MMP14, TOP2A, ANLN, and SLC22A3 exhibited higher expression in tumor samples compared to normal samples.Additionally, immunofluorescence helped determine the cell location of these proteins (Supplementary Fig. 7B).

Discussion
PC, as a pancreatic malignancy, is projected to be the second leading cause of cancer-related deaths by 2030 49 .Owing to the lack of a reliable diagnostic method, most PC patients are diagnosed at an advanced stage 50 .The incidence of pancreatic cancer was reported to be around 57,600 with 47,050 deaths in the United States in 2020 51 .Surgery is now the only curative treatment method for PC; therefore, the underlying cause and precise mechanisms of PC needed to be explored.Searching for more effective biomarkers for the diagnosis and treatment of PC patients is urgent.OS is the imbalance of the production and clearance of ROS.High levels of ROS lead to severe oxidative damage in proteins, lipids, and DNA [52][53][54] .Phospholipids, are especially susceptible to ROS attack 55,56 .It has been discovered that OS promotes lipid peroxidation and can cause cell damage 57 .Regulating the dysfunction of OS which is related to the dysfunction of phospholipids is supposed to be a novel approach for cancer treatment 20,58,59 .
Given the occurrence of the metabolism disorder in tumor cells, the metabolite alteration in PC samples was explored.After taking an intersection of the different metabolites between tumor samples and normal samples and different metabolites between tumor samples and para-carcinoma samples, the key metabolites differentially generated in PC were identified.Among all the metabolites, diacylglycerol, glycerophosphocholine, and glycerophosphoethanolamine were the main differentially-produced metabolites in PC samples.As a result, the disorder of PM was regarded as essential in the development of PC.
Due to the influence of OS on PM, the potential mechanisms of OSPM-related genes in PC were investigated.After differential expression analysis, univariate Cox regression analysis, and KM survival analysis, 21 prognostic OSPM-related DEGs of 3,197 OSPM-related genes emerged as OSPM function genes.Subsequently, ssGSEA was performed to assess the OSPM function of all samples, and then PC samples were divided into three clusters based on the ssGSEA function score of normal samples.Samples in cluster 1 (OSPM-active cluster), with relatively high ssGSEA function score, had a poor prognosis.In the in-depth exploration, the discrepancies of the expression of tumor suppressor genes and oncogenes (i.e., COL24A1, FAT2, CENPJ, ECT2, PXDN, CCND1, ADAMTS12, and KMT2C) and ICGs (HAVCR2, TNFSF4, TNFRSF9, PDCD1LG2, CD80, TNFSF9, ICOS, CD70, IDO1, PDCD1, and HHLA2) may result in different prognoses.
In addition, the 21 genes were summarized in pan-cancer.Compared with their corresponding normal samples, these genes were differentially expressed in tumor samples and were related with their prognosis.The high frequencies of CNV and SNV and the methylation level of promoters might be related to the alteration of the gene expression levels.The above findings shed new light on the in-depth research concerning OSPM in cancer.
In view of the impact of 21 prognostic OSPM-related DEGs on PC patients, an OSPM-related prognostic signature was developed utilizing the most appropriate variables selected from the 21 genes.Eight genes (i.e., SLC2A1, MMP14, TOP2A, MBOAT2, ANLN, ECT2, SLC22A3, and FGD6) were included in the signature.Interestingly, all the eight genes had increased expression levels in the high-risk samples, and the protein levels of SLC2A1, MMP14, TOP2A, ANLN, and SLC22A3 also increased in tumor samples compared with normal samples.Additionally, the expressions of ECT2, FGD6, MMP14, and SLC2A1 were related to the staging of the tumor.The potential roles of these eight genes in PC have been identified in other studies, and all these genes were demonstrated to combine with unfavorable functions in pancreatic cancer.Solute carrier family 2 member 1 (SLC2A1), known as glucose transporter 1 (GLUT1), is reported to express highly in many cancers and promote cancer growth [60][61][62] .It is reported to be highly enriched in active glucose metabolism subtype, which is associated with lower survival rate 63 .Matrix metalloproteinase 14 (MMP14) was also identified as a prognostic gene in PC 64 .Certainly, knock-down of MMP14 was proven to prevent the invasion of pancreatic cancer 65 .DNA topoisomerase 2-alpha (TOP2A), as the downstream molecule of DGCR5, could influence the sensitivity of pancreatic cancer cells to gemcitabine 66 .Membrane-bound O-acyltransferase domain-containing 2 (MBOAT2), correlated with decreasing infiltration of CD8 + T cells and KRAS activation, has been demonstrated as an unfavorable biomarker in pancreatic cancer 67 .Anillin (ANLN) promoted the progression of pancreatic cancer through inducing EZH2 up-regulation by mediating the miR-218-5p/LASP1 signaling 68 .Epithelial cell transforming 2 (ECT2) was regulated by Yes-associated protein 1 and mediates pancreatic cancer progression and metastasis 69 .Ephrin-2 was found to promote the migration and invasion of PC through targeting of miR-557 70 .Solute carrier family 22 member 3 (SLC22A3) showed up-regulation in pancreatic cancer when compared with non-neoplastic pancreatic tissues, and patients with high expression of SLC22A3 may benefit from nucleoside analogs 71 .Faciogenital dysplasia 6 (FGD6) was found to promote the proliferation, macropinocytosis, and tumor growth both in vitro and in vivo in pancreatic cancer 72 .Our results suggested that the expression of MBOAT2 mainly existed in the epithelium, and the expression of MMP14 primarily existed in fibroblasts, while the expression of FGD6 mainly existed in progenitor.
Based on the OSPM-related signature, PC samples were divided into high-and low-risk subgroups.Further investigation of the discrepancies between these subgroups indicated that the high-risk subgroup showed statistically different immune response compared with the low-risk subgroup.Immune system disorders in PC may result in the poor prognosis.Also, the expressions of ICGs (i.e., SLC2A1, FGD6, MMP14, MBOAT2, TOP2A, ANLN, SLC22A3, and CD70) were up-regulated in the high-risk subgroup in both the training cohort and the test cohort.All these discrepancies related the risk of PC samples were potential targets for the treatment of PC.
This study has a few limitations that need to be addressed.Our signature was developed according to the analyses of retrospective data.Although the expression levels of genes in the signature were explored at both transcriptional level and translational level according to the online databases, additional basic exploration is required for further demonstration.Additionally, large clinical cohorts are needed to confirm the accuracy of the signature.

Conclusion
In this study, metabolomics analyses uncovered alterations in PM, while transcriptomic studies revealed differential expression of OSPM-related genes in PC.Then three OSPM-based PC subtypes were identified, and an OSPM-related signature was developed, which helped to unmask the OSPM-based discrepancies of prognosis, tumor microenvironment, immune response, and the expression of immune checkpoint genes in PC.Finally, the expression levels of genes in the signature were further demonstrated.All these findings are valuable for estimating prognosis and facilitating individualized treatment in pancreatic cancer.

Figure 1 .
Figure 1.Flow chart of the study.

Figure 3 .
Figure 3. Identification of key OSPM-related genes in PC. (A) The intersection of OS-related genes and PM-related genes; (B) The common OSPM-related differentially expressed genes (DEGs); (C,D) Univariate Cox regression analysis and Kaplan-Meier survival analysis for demonstrating the prognostic value of OSPM-related DEGs.

Figure 4 .
Figure 4. OSPM-related genes in pan-cancer.The mRNA alteration (A), the prognostic value (B), the CNV frequency (C), the SNV frequency (D), the mutation types (E), and status of promoter methylation (F) of OSPM-related genes.The heatmaps were created with the help of R language (version 4.0.3,https:// www.r-proje ct.org/).

Figure 5 .
Figure 5. OSPM-based cluster analysis.The discrepancies of OSPM function scores (A), OSPM function genes' expressions (B), survival probabilities (C), and the expression of tumor suppressor genes and oncogenes in three clusters (D).The heatmaps were created with the help of R language (version 4.0.3,https:// www.r-proje ct.org/).

Figure 6 .
Figure 6.The OSPM-related alteration of immune status in PC. (A) Comparing the expression of ICGs in the three PC subtypes; (B) The link between OSPM-related genes and immune cells; (C) The correlation between the OSPM function score and immune components; The correlation between OSPM function score and the proportion of TIL (D), T cell co-stimulation (E), pDC (F), and Mast cell (G).

Figure 7 .
Figure 7. Identification of the OSPM-related signature in the training cohort.(A) Random survival forest analysis for the selection of the most appropriate genes combination; (B) Group division of PC samples; (C) The link between risk score and survival; (D) PCA analysis; (E) t-SNE analysis; (F) The survival analysis in the highand low-risk subpopulations; (G) Heatmap exhibiting the expression of the signature genes; (H) ROC curves for demonstrating the diagnostic accuracy of the signature.The heatmaps were created with the help of R language (version 4.0.3,https:// www.r-proje ct.org/).

Figure 8 .
Figure 8. Validation of the OSPM-related signature in the test cohort.(A) Group division; (B) The link between risk score and survival; (C) PCA analysis; (D) t-SNE analysis; (E) The survival analysis in the high-and low-risk subpopulations; (F) Heatmap exhibiting the expression of the signature genes; (G) ROC curves for demonstrating the diagnostic accuracy of the signature.The heatmaps were created with the help of R language (version 4.0.3,https:// www.r-proje ct.org/).

Figure 9 .
Figure 9. qRT-PCR for the investigation about the expression of the eight genes in the OSPM-related signature.
: (1) Comparing the OSPM function scores by the Kruskal-Wallis test; (2) Comparing the survival probability by KM analysis; (3) Comparing expression levels of tumor suppressor and promoter genes through the Kruskal-Wallis test; and (4) Comparing the expression levels of immunological checkpoint genes (ICGs) by the Kruskal-Wallis test.