LncRNA RP3-525N10.2-NFKB1-PROS1 triplet-mediated low PROS1 expression is an onco-immunological biomarker in low-grade gliomas: a pan-cancer analysis with experimental verification

PROS1 gene expression data were downloaded from the GTEx portal (https://​gtexport.​org/​home/​), the Cancer Cell Line Encyclopedia database (https://​portals.​broadinstitute.​org/​ccle/​about), the TCGA database (https://​genome-cancer.​ucsc.​edu/​), and the CGGA database (http://​www.​cgga.​org.​cn/​). These data were normalized, and differential expression analyses were performed for PROS1 using the R package “limma” [28].

The free data analysis platform Sangerbox (http://​www.​sangerbox.​com/​tool) was used to validate the pan-cancer expression of PROS1 and explore the correlation of PROS1 expression with the expression of DNA methyltransferases, RNA modification genes, and DNA mismatch repair system genes, as well as copy number variation, single nucleotide variation frequency, genomic heterogeneity, and cancer stemness using Spearman’s or Pearson’s method.

GEPIA (http://​gepia.​cancer-pku.​cn/​index.​html) is a free web tool based on TCGA and GTEx data. In the current study, PROS1 expression, survival analysis, and possible involvements of lncRNAs and TFs were evaluated using the GEPIA modules “Expression DIY” and “Survival”. In addition, the relationships between PROS1 and Gene markers were determined using Spearman’s correlation coefficient in the module “Correlation analysis”.

TIMER (https://​cistrome.​shinyapps.​io/​timer/​) is a web server for the comprehensive analysis of tumour-infiltrating immune cells. In our study, PROS1 expression and survival were evaluated using the “Gene” and “Survival” modules. TIMER was also applied to investigate the relationships between PROS1 expression and different gene marker sets of immune cells using the “Correlation” module [29].

Fresh glioma tissues from histologically confirmed cases were obtained from the Union Hospital, Tongji Medical College, Huazhong University of Science and Technology. The study was approved by the Ethics Committee of the Union Hospital, Tongji Medical College, Huazhong University of Science and Technology.

Since there are no specific LGG cell lines, common glioma cell lines (U87, U251, and T98G) and the human astrocyte cell line NHA were used. Cells were cultured in Dulbecco’s modified Eagle’s medium (Gibco) containing 10% heat-inactivated foetal bovine serum and 1% penicillin/streptomycin. Real-time PCR was conducted to compare gene expression in 30 tumour samples with that in adjacent normal tissue. Real-time PCR was performed in triplicate using samples derived from three independent experiments. Primers for PROS1 (forward, 5’-GTGCCTTCCCTTGAACCRRG-3’, reverse, 5’-CCACGCTGAGTGATCGATAGA-3’) and GAPDH (forward, 5’-AAAAGCATCACCCGGAGGAGAA-3’, reverse, 5’-AAGGAAATGAATGGGCAGCCG-3’) were used for qPCRs. Ten formalin-fixed, paraffin-embedded LGG tissues and normal brain tissues were used for immunohistochemistry stainings.

Short hairpin RNA (shRNA) against PROS1 (shPROS1) and a negative control shRNA (sh NC) were designed and synthesised by genomeditech (Shanghai, China). In addition, the pcDNA3.1 vector (Vigene Biology) containing the full-length cDNA sequence of PROS1 was used to overexpress PROS1. The empty pcDNA3.1 vector was used as a negative control. The lentivirus pLent-shPROS1-GFP-Puro or its negative control (NC) pLent-GFP-Puro (genomeditech) was used to infect GBM cells with enhanced infection solution (genomeditech) according to the manufacturer’s protocol. Similarly, pLent-PROS1-GFP-Puro lentivirus or empty vector (vector) pLent-GFP-Puro lentivirus (genomeditech) was used to overexpress genes. Seventy two hours after the cells were infected with lentivirus, 2 μg/ml puromycin was added to kill the cells that had not been transfected.

U87 and U251 cells was assessed with the Cell Counting Kit-8 (Dojindo Molecular Technologies, Kyushu, Japan) reagent according to the manufacturer’s instructions. Cells were inoculated on 96- well plates at a density of 1000 cells per well with 100 μl of medium. Every 24 h for a total of 96 h, CCK8 solution (10 μl) was added to each well, and the cells were further incubated at 37 °C for 3 h. The absorbance of each well was measured at 450 nm with a spectrophotometer.

U87 and U251 Cells were prepared into a single-cell suspension, respectively, and seeded into a six-well plate (200 cells/well) for two-week incubation to form colonies. After staining with 0.01% crystal violet (Sigma), the colonies were subjected to microscopic examination. The rate of colony formation and survival fraction were calculated.

2 × 10⁴ cells were added into Matrigel-coated upper Transwell chambers for the invasion assay. The lower chambers were filled with DMEM containing 10% FBS. After incubation at 37 °C for 24 h, cells on the lower surface of the membrane were fixed in 100% methanol and stained with 0.1% crystal violet dye for 20 min at room temperature. Finally, after washing with PBS, cells were imaged in five randomly selected fields under a light microscope (Olympus Corporation) at × 100 magnification.

According to the manufacturer’s instructions, EdU Kit (Roche, Mannheim, Germany) was utilized to monitor the proliferation of transfected cells. Zeiss Axiophot Photomicroscope (Carl Zeiss, Oberkochen, Germany) was used to capture representative images.

Genomic heterogeneity includes the parameters TMB, MATH, tumour ploidy, HRD, LOH, MSI, NEO, and tumour purity [21, 30]. Cancer stemness, including DNAss, EREG-METHss, DMPss, ENHss, RNAss, and EREG.EXPss [31], was calculated using one-class logistic regression algorithms with mRNA expression and methylation signature.

MethSurv (https://​biit.​cs.​ut.​ee/​methsurv/​) was used to evaluate the effects of methylation levels and PROS1 expression on prognosis in LGG [32], and we used the OncoSplicing website (http://​www.​oncosplicing.​com/​) to explore differential alternative splicing events of PROS1 in LGG [33].

The PrognoScan database (http://​www.​prognoscan.​org/​) is a large collection of publicly available cancer microarray datasets [34]. The OncoLnc database (www.​oncolnc.​org), a tool for interactive explorations of survival correlations, contains survival data of 8,647 patients from 21 cancer studies supported by the TCGA. The relationships between PROS1 expression and patient prognosis (overall survival, disease-specific survival, disease-free interval, and progression-free interval) were visualised with forest plots and Kaplan–Meier curves.

The GeneMANIA database (http://​www.​genemania.​org) was used to construct the PROS1 interaction network [35]. The STRING database (https://​string-db.​org/​) was used to construct the protein–protein interaction network of PROS1 [36].

CancerSEA (http://​biocc.​hrbmu.​edu.​cn/​CancerSEA/​home.​jsp), a database that aims to comprehensively decode functional states of cancer cells at single-cell resolution, was used to explore PROS1 functions [37].

GO and KEGG analyses were applied to explore the biological functions of PROS1 in LGG. GSEA was used to investigate potential PROS1 mechanisms. GO, KEGG, and GSEA analyses were analysed using the R package “ClusterProfiler”.

TIDE (http://​tide.​dfci.​harvard.​edu/​), based on tumour pretreatment expression profiles, can estimate multiple published transcriptomic biomarkers to predict patient responses. It was used to investigate the association between PROS1 expression and therapy outcomes in clinical studies of immune checkpoint blockade [38].

Most analyses were conducted using R software, and the rest were GraphPad Prism 8.0. Logistic regression, univariate, and multivariate analyses were used to assess the influence of clinical variables on patient survival. Two-tailed P-values less than 0.05 were considered statistically significant.

To explore possible anti- and pro-carcinogenic roles of PROS1, its mRNA expression was first analysed in human cancer and validated using the Sangerbox tools. As shown in Fig. 1A, PROS1 expression was markedly increased in 11 tumour types (DLBC, GBM, KIRC, KIRP, LGG, PAAD, PRAD, STAD, THCA, THYM, and UCEC) and significantly decreased in 17 tumour types (ACC, BLCA, BRCA, CESC, CHOL, COAD, ESCA, HNSC, KICH, LAML, LIHC, LUAD, LUSC, OV, TGCT, UCEC, and UCS) compared to that in normal samples. However, PROS1 expression was not significantly different among MESO, PCPG, READ, SKCM, and UVM. Moreover, Sangerbox tools were used to validate mRNA expression results. As shown in Fig. 1B, PROS1 expression in COAD, GBM, KIRC, KIRP, LGG, PAAD, PRAD, SKCM, STAD, TGCT, and THCA was significantly increased when compared to that in control samples of the genotype-tissue expression (GTEx) project, whereas in BLCA, BRCA, CESC, CHOL, HNSC, KICH, LAML, LUAD, LUSC, OV, READ, UCEC, and USC, PROS1 expression was substantially decreased. Taken together, PROS1 was upregulated in GBM, LGG, KIRC, KIRP, PAAD, PRAD, STAD, and THCA, and downregulated in BLCA, BRCA, CESC, CHOL, HNSC, KICH, LAML, LUAD, LUSC, OV, UCEC, and USC, demonstrating that PROS1 may function as a vital regulator of carcinogenesis in 20 different types of cancer. We additionally analysed PROS1 expression in 31 types of tissues using the GTEx dataset (Additional file 1: Fig. S1), determined its expression in 21 tumour cell lines using the Cancer Cell Line Encyclopedia database (Additional file 1: Fig. S2), and compared the expression between tumour and normal tissues using The Cancer Genome Atlas (TCGA) database (Additional file 1: Fig. S3). A separate analysis of PROS1 mRNA levels showed a significant difference between LGG and normal GTEx samples (Fig. 1C). Likewise, in the Gene Expression Profiling Interactive Analysis (GEPIA) database, higher PROS1 mRNA expression was found in LGG tissues than in normal brain tissues (Fig. 1D). Receiver operating characteristic curves were used to analyse the efficacy of PROS1 levels to distinguish between LGG samples and normal brain samples. The area under the curve (AUC) of 0.934 (0.922–0.946) suggests that the PROS1 gene may have the potential to identify LGG tissues. At the protein level, immunohistochemistry stainings were used to investigate the PROS1 expression in 15 paired tumour samples compared with adjacent normal tissues. The analysis demonstrated that PROS1 levels were substantially increased in LGG tissues (Fig. 2A). Moreover, 30 paired samples and 4 cell lines were investigated by qRT-PCR. PROS1 mRNA expression was significantly upregulated in all 30 tumour samples compared to that in adjacent normal samples (Fig. 2B), as well as in 3 glioma cell lines (T98G, U87, and U251) compared to that in the non-malignant human astrocyte cell line NHA (Fig. 2C).

PROS1-high samples and PROS1-low samples from HTSeq-Counts of TCGA-LGG were explored using the R package “DESeq2”. In total, 2,050 DEGs showed statistically significant group differences, including 1,023 upregulated genes and 1,027 downregulated genes (Fig. 1F). The relative expression values of the top five upregulated and downregulated DEGs in these two groups are depicted as a heatmap in Fig. 1G.

At the epigenomic level, various epigenetic modifications have the possibility of functional gene modification. DNA methylations such as N1-methyladenosine (m1A), cytosine-5-methylation (m5C), and N6-methyladenosine (m6A) are common epigenetic modifications, so we explored their correlations with PROS1 expression levels. As shown in Fig. 3A, B PROS1 expression has a close relationship with the expression of 4 DNA methyltransferase genes (DNMT1, DNMT2, DNMT3A, and DNMT3B) and 44 RNA modifications (m1A [n = 10], m5C [n = 13], and m6A [n = 21]) across most human cancers, but especially in patients with LGG, which indicates that PROS1 may mediate tumorigenesis by epigenetic modification. DNA mismatches are potentially mutagenic and, thus, must be corrected by the mismatch repair system to maintain the integrity of the genetic information. As shown in Fig. 3C, we evaluated the correlation between the mutation levels of five mismatch repair system genes (MLH1, MSH2, MSH6, PMS2, and EPCAM) and PROS1, which suggested that the mismatch repair system may play a critical role in regulating tumorigenesis of PROS1. We analysed SNP data to detect variants and their frequencies in LGG. As shown in Fig. 3D, F missense mutations (0.2%) were the main type of SNPs. A frequency analysis of single nucleotide variations comparing groups with high and low PROS1 expression levels revealed IDH1, TP53, ATRX, CIC, TTN, FUBP1, MUC16, NOTCH1, PIK3CA, and EGFR as the top 10 mutated genes, with mutation percentages of 76%, 45%, 33%, 20%, 12%, 9%, 7%, 7%, 6%, and 6%, respectively (Fig. 3F).Moreover, our analysis results showed that IDH1 and TP53 were the top2 mutation gene, and mutation frequency of IDH1 and TP53 in low PROS1 expression group was more than high PROS1 expression group, which also identified PROS1 may promote tumour progression and lead to poor prognosis in LGG patients. To identify alterations in copy number variation, data of PROS1 copy number variation extracted from the TCGA were explored using the R package “copynumber”, and significant sample differences among the three types (491 neutral, 14 loss, and 3 gain) were detected by the Kruskal–Wallis test (Fig. 3E).

As shown above, PROS1 expression was significantly correlated with the expression of four DNA methyltransferases. To explore the potential mechanism between PROS1 expression and DNA methylation, MethSurv was utilised to investigate the impact of different methylation levels and PROS1 expression on the prognosis in LGG. As shown in a heatmap (Fig. 4A), we found high levels of methylation in some of the probe regions (cg03680898, cg10993409, cg10959048, cg03451959, and cg14753809) and low methylation levels in other probe regions (cg05897638, cg24305970, cg01408194, cg03168026, and cg09400966). The survival analysis indicated that cg14753809, cg05897638, and cg24305970 were associated with poor prognosis (Fig. 4B–D). Alternative splicing has been identified as a vital mechanism regulating phenotypic diversity, gene expression, and proteomics; it can be used in biomarker and drug resistance research, as well as a therapeutic target. We used the OncoSplicing website to explore differential alternative splicing events of the PROS1 gene in LGG. Two alternative promoters (PROS1_AP_65674 and PROS1_AP_65675; Fig. 4E, F) found in the TCGA SpliceSeq database were associated with poor prognosis (Fig. 4G–J).

Genomic heterogeneity and cancer stemness were closely correlated with tumour treatment selection and overall survival. We analysed the association of PROS1 expression with genomic heterogeneity and cancer stemness in patients with LGG. The results indicated that tumour mutational burden (TMB; R = 0.131, P = 0.00094), mutant-allele tumour heterogeneity (MATH; R =  − 0.117, P = 0.0082), tumour ploidy (R =  − 0.100, P = 0.024), homologous recombination deficiency (HRD; R = 0.140, P = 0.0021), loss of heterozygosity (LOH; R = 0.227, P = 3.31e − 7), DNA methylation-based stemness (DNAss; R = 0.092, P = 0.037), enhancer elements/DNA methylation-based stemness (ENHss; R = 0.097, P = 0.0288), epigenetically regulated DNA methylation-based stemness (EREG-METHss; R = 0.109, P = 0.0138), and RNA expression-based stemness (RNAss; R =  − 0.341, P = 2.933e − 15) were significantly different from those of PROS1 (Additional file 1: Fig. S4). However, there was no significant difference in microsatellite instability (MSI), neoantigen (NEO), tumour purity, differentially methylated probes-based stemness (DMPss), and epigenetically regulated RNA expression-based stemness (EREG.EXPss; Additional file 1: Fig. S5). These data indicate that PROS1 expression may influence cancer treatment and prognosis by affecting genomic heterogeneity and cancer stemness.

To evaluate PROS1 expression among groups of patients with different clinicopathological characteristics, we analysed 510 LGG samples from the TCGA database with their clinical data. As shown in Additional file 1: Fig. S6 and Table S1, PROS1 expression was significantly correlated with the WHO grade, IDH status, primary therapy outcome, histological type, and outcome measures (overall survival, disease-specific survival, and progression-free interval). All other examined clinical parameters were not significantly correlated with PROS1 expression (Additional file 1: Fig. S6). The results of the logistic regression analysis indicated that the categorical dependent variable PROS1 expression was correlated with poor prognostic clinical parameters, including WHO grade, IDH status, age, and histological type (Additional file 1: Table S2). Taken together, LGG with high PROS1 expression may be more likely to progress to a more advanced grade and less susceptible to IDH1 mutations than LGG with low PROS1 levels.

Since PROS1 expression levels were closely correlated with LGG tumour progression, we subsequently explored their prognostic significance. As shown in the forest plot (Fig. 5A), we first analysed the correlation between PROS1 expression and overall survival in 33 tumours from the TCGA database. The results indicated that high PROS1 expression was related to poor survival in several cancers (LGG, BCLA, STES, STAD, and UVM), and the pan-cancer analysis demonstrated similar results for disease-specific survival, disease-free interval, and progression-free interval (Additional file 1: Figs. S7–S9). Next, we analysed the correlations of PROS1 expression with overall survival and disease-specific survival (Fig. 5B, C respectively). These results were validated by data from various databases and websites, including the Chinese Glioma Genome Atlas (CGGA; Fig. 5D), GEPIA database (Fig. 5E), OncoLnc database (Fig. 5F), and PrognoScan database (Fig. 5G, H). Based on receiver operating characteristic curves of TCGA (Fig. 5I) and CGGA (Fig. 5G) data, the associations of PROS1 expression with 1-, 2-, and 3 year were explored. The results suggested that PROS1 expression had a moderate prognostic value. Next, we conducted subgroup analyses of overall survival based on clinicopathological characteristics. The prognosis of LGG patients with high PROS1 levels was poor in women, age > 40 years, 1p/19q non-codeletion, SD&PD, and astrocytoma subgroups (Fig. 6A–E). Univariate (Fig. 6F and Additional file 1: Table S3) and multivariate (Fig. 6G and Additional file 1: Table S3) Cox regression analyses showed that PROS1 expression, WHO grade, and age were independent prognostic factors for overall survival of patients with LGG. Risk factor association (Fig. 6H) and Sankey (Fig. 6I) diagrams were used to visualise the overall prognostic trends of the groups with high and low PROS1 expression, as well as the relationships among the three independent factors in living status. Finally, a nomogram was constructed based on PROS1 expression, age, and WHO grade as a quantitative tool for predicting the prognosis of patients with LGG (Fig. 6J). The prediction efficiency of the nomogram was evaluated using the C-index (0.754, confidence interval: 0.727–0.782), which implied that this model had moderate prediction accuracy. Consistent with the former results, the agreement between predictions and observations was good (Fig. 6K).

The GeneMANIA database was used to analyse the gene–gene interaction network of PROS1 and identify altered neighbouring genes. The results showed that 20 genes were most closely related to PROS1, including MERTK, C4BPB, F5, and F8 (Fig. 7A). Similarly, the STRING database was used to assess the PROS1 protein–protein interaction network which contained 29 edges and 11 nodes, including TYRO3, PROC, and GGCX (Fig. 7B). Taking the above results into account, the relationships between coagulation-related genes and PROS1 was explored in LGG data from the TCGA. The significant results of this analysis demonstrated that in LGG, PROS1 was positively correlated with F2R, FGG, and PROC but negatively correlated with F5 (Fig. 7C).

To better understand the potential regulatory mechanisms of PROS1 in LGG, we examined the regulation of PROS1 expression by lncRNA-TF-gene triplets. We used the LncMAP database to explore possible PROS1-containing triplet formations in LGG. The results showed that seven TFs may modulate PROS1 expression in LGG (Fig. 8A). To identify the TFs most likely regulating the PROS1 gene, we conducted an expression, prognosis, and correlation analysis of these seven TFs (Fig. 8B). Among the four TFs (STAT1, SPI1, NFKB1, and CEBPA) with high expression and prognostic significance in LGG (Fig. 8C–I), the correlation coefficient was the highest (R = 0.514) between NFKB1 and PROS1. Subsequently, we explored lncRNAs possibly acting on NFKB1 and identified 31 lncRNAs that can interact with NFKB1 (Fig. 8J). Differential expression and prognostic analyses were used to determine the target lncRNAs. Ultimately, the two lncRNAs RP3-525N10.2 and MIR497HG were significantly differentially expressed in LGG (Fig. 8K–L), but only RP3-525N10.2 was significantly downregulated, suggesting a good prognosis. In summary, we identified the triplet RP3-525N10.2-NFKB1-PROS1 and, considering the downregulated expression of RP3-525N10.2 in LGG, suggest its possible role as a tumour suppressor gene affecting the prognosis in LGG. We propose that the lncRNA RP3-525N10.2 may act as a “decoy” that binds to the TF NFKB1 or guide the TF, and prevents its association with the target PROS1, which can reduce PROS1 expression, thereby improving the prognosis in patients with LGG.

GO and KEGG enrichment analyses were used to investigate PROS1-related biological pathways and functions. Circular plots depict the top 10 significant terms of BP, MF, CC, and KEGG enrichment analyses (Fig. 9A–D). Notably, in all analysis results, PROS1 was enriched in immune response-related processes or pathways, such as complement activation, B cell-mediated immunity, immunoglobulin-mediated immune response, and lymphocyte-mediated immunity in BP term, immunoglobulin complex, T-cell receptor complex, and MHC class II protein complex in CC term, and immunoglobulin receptor binding in MF term. In the KEGG pathway analysis, several immune-related pathways were highly correlated with PROS1 expression, including Th17 cell differentiation, Th1 and Th2 cell differentiation, primary immunodeficiency, T-cell receptor signalling pathway, and natural killer (NK) cell-mediated cytotoxicity. Consistent with the results of the GO and KEGG analyses, GSEA, including KEGG (Fig. 9E) and Reactome (Fig. 9F) pathways, showed that the top eight signalling pathways affected by PROS1 were enriched mainly in immune-related activities, including NK cell-mediated cytotoxicity, T-cell receptor signalling pathways, immunoregulatory interactions between lymphoid and non-lymphoid cells, signalling by the B-cell receptor BCR, and programmed death 1 (PD1) signalling. These results strongly imply that PROS1 participates in the regulation of immune responses in LGG.

To explore the functions of PROS1 in LGG tissues at the single-cell level, CancerSEA was used to explore phenotypes possibly regulated by PROS1. As shown in t-SNE and box plots (Additional file 1: Fig. S10), PROS1 was upregulated at the single-cell level in EXP0059 glioma cell groups. The results of the functional analyses (Additional file 1: Fig. S10) demonstrated that PROS1 positively influenced cell invasion in two glioma single-cell sequencing datasets (BCH1126 and BCH836). As shown in Fig. 10, CCK8 assays depicted that knocking down PROS1 expression significantly reduced the proliferation ability of U87 cells and U251 cells; conversely, overexpressing PROS1 increased cell proliferation (Fig. 10B). Transwell assays were used to assess the invasion ability of cells. PROS1 knockdown resulted in a lower U87 cell and U251 cell invasion rate, while PROS1 over-expression showed the opposite results (Fig. 2D). In addition, colony formation assay (Fig. 10C) and 5-Ethynyl-2’-deoxyuridine (EdU) incorporation assay (Fig. 11) also suggested that knockdown of PROS1 inhibited cell proliferation. In contrast, over-expression of PROS1 increased cell proliferation rate. In summary, PROS1 increased the proliferation and invasion of GBM cells.

The functional enrichment results indicated that PROS1 may play a critical role in the immune system. Therefore, we investigated the relationships between PROS1 expression and immune cell infiltration. According to analyses of the Tumor Immune Estimation Resource (TIMER) database, six types of immune cells (B cells, CD8 + T cells, CD4 + T cells, macrophages, neutrophils, and dendritic cells) were associated with PROS1 expression and prognosis in patients with LGG (Fig. 12A, B). Further analyses using the ssGSEA (Fig. 12C, D) showed that PROS1 expression levels were significantly positively associated with the infiltration of B cells, Tem, DC, Tgd, CD8 + T cells, NK cells, Th2 cells, NK CD56dim cells, eosinophils, T cells, Th17 cells, neutrophils, cytotoxic cells, iDC, aDC, T helper cells, and macrophages, but significantly negatively correlated with infiltration of CD56bright cells, pDC, and TReg. Moreover, various copy numbers of PROS1 were not significantly correlated with immune cell infiltration levels in LGG (Fig. 12E). To analyse the possible influence of PROS1 expression on the tumour microenvironment during tumour development, we analysed immune and stromal scores of LGG samples using the R package “Estimate”. As shown in Fig. 10F, H PROS1 was positively correlated with immune score (R = 0.440, P < 0.001), stromal score (R = 0.430, P < 0.001), and estimated immune score (R = 0.440, P < 0.001).

To better understand the interaction between PROS1 expression and immune responses, the TIMER and GEPIA databases were utilised to validate the association between PROS1 expression and immune cell markers in LGG. The genes listed in Table 1 represent various immune cells, including B cells, T cells, CD8 + T cells, monocytes, tumour-associated macrophages (TAMs), M1 macrophages, M2 macrophages, neutrophils, NK cells, and dendritic cells. In LGG, expression of PROS1 was significantly correlated with that of most immune cell markers (Table 1). Some studies have shown that PROS1 can influence the activation of TAMs [8, 17‐20]; therefore, we explored the relationships between PROS1 expression and TAMs, including microglia and monocyte-derived macrophages (Table 1). According to the analyses of the TIMER and GEPIA databases, PROS1 expression levels had a significant positive correlation with 10 of the 13 immune cell markers in LGG (Additional file 1: Fig. S11).

The correlation analysis between immunomodulators and PROS1 expression may reveal the types of cancers that benefit from immunotherapy targeting PROS1. In LGG, PROS1 was positively correlated with most of the 150 analysed immunomodulators (Fig. 13A). Based on the improved understanding of the functions of novel human immune checkpoint inhibitors [21], 60 immune checkpoint genes (24 inhibitory, 36 stimulatory) were explored regarding their correlations with PROS1 in different cancer types (Fig. 13B). Notably, more than 50 immune checkpoint markers were significantly associated with PROS1 expression in LGG. The scatter plots of five selected common immune checkpoint genes (CD274, CTLA4, HAVCR2, IDO1, and PDCD1) in the TIMER and GEPIA databases are shown in Fig. 11C–G. Collectively, these results strongly imply that PROS1 may play a critical role in LGG immunity and may be a potential immunotherapy target.

Based on the predictive power of response outcomes and overall survival among ICB subcohorts, we evaluated the relevance of PROS1 as a biomarker by comparing its performance with those of standardised biomarkers. We found that PROS1 alone had an AUC value > 0.5 in 8 of the 20 examined ICB subcohorts (Fig. 14A). PROS1 exhibited a higher predictive value than B.Clonality, which gave AUC values > 0.5 in 6 ICB subcohorts. However, as a predictive biomarker, PROS1 was comparable to the combination of TMB score and B.Clonality (AUC > 0.5 in 8 ICB subcohorts) but inferior to MSI, CD27A, TIDE, IFNG, Merck18, and CD8. Our results also indicated that high PROS1 expression was correlated with worse PD1 outcomes in kidney renal clear cell carcinoma (Braun2020_PD1 Clear), melanoma (Liu2019_PD1 Ipi_Naive, Gide2019_PD1, and Riaz2017_PD1 Ipi_Naive), PD-ligand 1 (LI) (BladderMariathasan2018_PDL1 mUC), ACT melanoma (Lauss2017_ACT), and CTLA4 (Nathanson2017_CTLA4 Post) but achieved good CTLA4 therapeutic outcomes in melanoma (Nathanson2017_CTLA4 Pre), and kidney renal clear cell carcinoma (Miao2018_ICB Clear) cancer cohorts (Fig. 14B). Screening results of knockout phenotypes also implied that PROS1 knockout may positively impact lymphocyte-mediated tumour killing in MC38 colon cancer (Kearney2018_T_PD1) models (Fig. 14C). Moreover, the MXD3 expression levels were correlated with overall survival (Fig. 14D–H) in 3/4 of the glioma cohorts (Nutt_Glioma@PRECOG, TCGA, and GSE16011@PRECOG) and likewise with CTL levels (Fig. 14D–H) in 3/4 of the glioma cohorts (ca00037@PRECOG, TCGA, and GSE16011@PRECOG). CTL expression levels in the GSE16011@PRECOG cohort were also correlated with overall survival in PROS1 mRNA expression and copy number alteration data (Fig. 14I, J). Finally, the Enrichr platform was used to identify drug molecules, collected from the DSigDB database, targeting PROS1. The results showed that spiperone PC3, thapsigargin PC3, quercetin CTD 00006679, and pentabromodiphenyl ether CTD 00003077 are the four drug molecules with which most DEGs interact (Fig. 14K).