The GDC database’s (https://portal.gdc.cancer.gov/) TCGA-LGG cohort, which contains 509 LGG samples, was downloaded. Clinical information, FPKM values for gene expression, and RNA sequencing information were received from GDC. For further investigation, the FPKM values were subsequently transformed to transcripts per kilobase million (TPM) values (29). Table 1 displays the TCGA-LGG cohort’s starting data. The mRNA expression profiles of normal brain tissue were acquired from the Genotype-Tissue Expression Project (GTEx, https://www.gtexportal.org). Data of a total of 527 simple nucleotide variation from the GDC database were downloaded. Tumor mutation burden (TMB) (mut/mb) = total number of mutations (including synonymous and nonsynonymous point mutations, substitutions, insertions, and deletions mutations)/size of the target region coding area. For each sample, TMB values were computed taking into consideration the definition of TMB. From UCSC Xena (https://xenabrowser.net/datapages/), data for 533 gene copy number variants (CNVs) were retrieved. The CGGA database (http://www.cgga.org.cn) was used to download the gene expression profiles and clinical information for the CGGA cohort. The IMvigor210 dataset was obtained in order to assess the anoikis riskscore’s ability to forecast the immunotherapy response. Using previously released studies (14, 22–27) and datasets(GSE145806, GSE106592, GSE155457, GSE40690, GSE55958, GSE39220, and GSE40171), we chose 19 ANOIRGs.

Mutation annotation format (MAF) files of TCGA mutation data were subjected to analysis utilizing the “maftools” R package, and waterfall plots were drawn to visualize the mutations of the 19 ANOIRGs in the TCGA-LGG cohort. A CNV landscape of 19 ANOIRGs was developed based on CNV data from the TCGA-LGG cohort. To examine the differences in mRNA expression between normal and LGG samples, a differential analysis of the 19 ANOIRGs based on the LGG integrated expression profiles of GTEx and TCGA was carried out. The samples were separated into low and high expression groups by employing an optimal cutoff value of the gene expression profile, and comparison of the difference in overall survival (OS) between the low and high expression groups was made using the log-rank test and univariate Cox regression. A co-expression prognostic network of 19 ANOIRGs was built employing univariate cox regression analysis and Pearson’s correlation analysis.

Unsupervised consistency clustering and classification based on 19 ANOIRGs was attempted via the “ConsensusClusterPlus” R package (30), re-sampling 80% of the samples 50 times using coalescent pam clustering with Euclidean distance. After that, the differences in OS between various subtypes were compared by employing the Kaplan-Meier (K-M) survival analysis, the expression of the 19 ANOIRGs between subtypes was compared using box line plots, and comparison of the distribution of the 19 ANOIRGs’ expression across subtypes was made using the t-Distributed Stochastic Neighbor Embedding (t-SNE). In an attempt to display the distribution of the 19 ANOIRG expressions and clinicopathological characteristics across the various subtypes, heat maps were produced using the “pheatmap” R package.

Molecular Signature Database (MsigDB, http://software.broadinstitute.org/gsea/msigdb/) was employed for obtaining data for the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway, HALLMARK pathway and Reactome pathway and “c2.cp.kegg.v7.5.1. symbols.gmt”, “h.all.v7.5.1.symbols.gmt” and “c2.cp.reactome.v7.5.1.symbols.gmt” was obtained as the reference gene set. We subsequently performed Gene Set Variation Analysis (GSVA) (31) on various subtypes using the “GSVA” R package to examine the variation in the biological processes of different subtypes and visualize it using a heatmap.

The StromalScore, ImmuneScore, and ESTIMATEScore were computed using the “ESTIMATE” R program. The quantitative metrics known as the ImmuneScore and the StromalScore, which measure the quantity of stromal and immune components, respectively, are obtained from gene expression profiling data. Also, the two scores are added to create the ESTIMATEScore, which has a negative correlation with tumor purity (32). Using the ssGSEA method based on the “GSVA” R package, the enrichment score of 23 immune cells in the TIME was then calculated, which is a depiction of the relative infiltration abundance of individual immune cells (33).

Using |logFoldChange|>1 and FDR<0.05, differentially expressed genes (DEG) were screened between various subtypes. Gene Ontology (GO) and KEGG functional enrichment analyses were performed by employing the “clusterProfiler” R package (34), and statistically significant results can be represented by adjusted p-value <0.05.

DEGs with p <0.05 were screened using univariate Cox regression analysis, and genes with differentially expressed prognostic genes with strong prognostic significance (SDEPGs) were subsequently selected based on a threshold of |1-HR| >0.5. Forest plots of the results from the Cox regression analysis, were drawn using the “forestplot” R package. A secondary unsupervised cluster analysis was then performed based on SDEPGs, classifying the samples into different subtypes with identical specific clustering parameters. Subsequently, we compared the OS differences between subtypes using survival analysis, compared the differential expression of 19 ANOIRGs between subtypes using box-line plots, and plotted heatmaps to show the distribution of SDEPGs expression and clinicopathological features between the subtypes.

The “glmnet” R package was used to execute the Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis, which reduces the dimensionality of high-dimensional data by capping the sum of the absolute values of coefficients at less than a set threshold. Only genes with non-zero coefficients in the LASSO regression analysis were chosen for additional investigation since the coefficients of the relatively tiny contributing variables were zero. We increased the stability and reproducibility of the LASSO model by adding a random seed. Then, the “randomForest” R package was used to screen genes for anoikis characteristics. The default iteration number of random forest algorithm is 100. When 500 trees are constructed, the model is considered to be robust enough. Based on Gini coefficient method, the “important” function was used to score for genes screened by LASSO model, and genes with a score above two were proceeded for further analysis. Finally, the genes obtained were screened using the multivariate Cox proportional risk regression analysis to obtain 12 potentially relevant anoikis genes that were identified as the best predictive traits and named as APRGs. These genes were selected to further calculate the anoiS for each patient using a multivariate Cox regression model: anoiS = ho(t) * exp (β 1X1+ β ₂X2+…. + β nXn). In the equation mentioned, β denotes the regression coefficient, and ho(t) refers to the baseline risk function. Multivariate Cox regression model was constructed from the “predict” function of the “rms” R package. Using the median anoiS, patients from the TCGA database were split into high- and low-anoiS groups. To ascertain the clinical independence of the anoiS for prognostic prediction, univariate and multivariate Cox analyses were utilized. The K-M survival analysis was then used to assess the differences in OS and progression-free survival (PFS) between LGG patients with high and low anoiS. Receiver operating characteristic (ROC) analysis and area under the curve (AUC) values were utilized for evaluating the prediction accuracy of the anoiS for 1-year, 3-year, and 5-year OS and PFS. The calibrate function of the “rms” R package plots the calibration curve, with a maximum resampling sample size of 1000. PFS and OS calibration curves for 1, 3, and 5 years were drawn. The forecast output of the model matches the actual one more closely the closer the calibration curve is near the line “y=x”. Last but not least, Sankey plots were created with the “ggalluvial” R package to show the relationship between various subtypes, an anoiS group, and prognosis. Comparison of the variations in anoiS between subtypes was achieved via box plots.

Through the correlation matrix, the ssGSEA algorithm was employed for quantifying the infiltration abundance of 23 different kinds of immune cells in the TIME, to show the correlation among the anoiS and immune cell infiltration levels, as well as investigate any potential relationships between the APRGs and the anoiS and 46 immune checkpoints. Finally, the prognostic predictive effect of the anoiS in the immunotherapy population was investigated using a survival analysis based on the IMvigor210 cohort.

First, we created distinct mutation landscapes for the high and low anoiS groups by making use of the “Maftools” R package. We analyzed variance to examine the variations in TMB levels between the high and low anoiS groups based on the TMB values of each TCGA-LGG cohort sample and Spearman’s correlation analysis to investigate the relation between anoiS and TMB. The optimal cutoff value was then determined using the “surv cutpoint” and “surv categorize” functions of the “survminer” R package and was taken into consideration as the boundary. Survival analysis was then conducted for the comparison of the differences in OS between patients with different anoiS and TMB statuses. The sample was then divided into high and low TMB groups.

“Age,” “gender,” “grade,” and “histological type” were selected as the clinical subgroup characteristics of patients with LGG, the distribution ratio of different clinical subgroup characteristics were counted in the high and low anoiS groups, and explored the differences of the anoiS among patients with different clinical subgroup characteristics.

First, the prognostic differences between the whole population, chemotherapy alone, radiotherapy alone, and radiochemotherapy population were compared using the K-M survival analysis based on OS data. Next, the potential association of anoiS with three classical genetic statuses, 1p19q co-deletion, IDH mutation, and MGMT promoter methylation, was determined using differential analysis and correlation analysis, and the AUC values of ROC curves were employed for comparing the predictive efficacy of the three gene statuses, anoiS, and grade. For 1-, 3-, and 5-year OS survival, the corresponding calibration curves were plotted. Finally, we compared the differences of OS among people with different treatment modalities and genetic status in the high and low anoiS groups, respectively.

The sensitivity to TMZ in patients with LGG was predicted using the “pRRophetic” R package (35) and the “oncopredict” R package (36) by predicting the IC50 value of temozolomide, ridge regression model was constructed to predict the AUC value of TCGA cohort by the “oncopredict” R package based on the expression profile data and AUC data of GDSC2 database. the lower the value of IC50 or AUC, the greater the sensitivity of the patient to TMZ. Differences in drug sensitivity to TMZ among the low and high anoiS groups were subjected to comparison, and correlations between the IC50 values of the anoiS and TMZ were demonstrated using Spearman’s correlation analysis. Correlation matrices were constructed to visualize the correlation between the 19 ANOIRGs with 12 APRGs and the IC50 values of TMZ.

Based on the correlation matrix of genes with predicted IC50 values of TMZ, the genes with the largest negative correlation coefficients, when correlated with the IC50 values of TMZ obtained from the combination of the two algorithms, were screened from the 19 ANOIRGs. The results of the combined differential analysis and prognostic analysis were excluded for obtaining the anoikis key genes (AKGs), and the gene co-expression network was constructed by employing AKG and the 12 APRGs with a correlation coefficient |r|>0.5 as the threshold to screen APRGs closely related to AKG as their potential reciprocal genes by using Pearson’s correlation coefficient. Additionally, the Human Protein Atlas (HPA; https://www.proteinatlas.org/) database was used to acquire the protein level immunohistochemical (IHC) staining results for chaperonin containing TCP1 subunit 5 (CCT5) between normal and LGG tissues.

The Cell Bank of Type Culture Collection of the Chinese Academy of Science (Shanghai, China), supplied the human Hs683 low-grade glioma cell line employed in the current work. American Type Culture Collection (ATCC, Manassas, VA, USA) provided human astrocytes (NHA) and human SW1088 low-grade glioma cell lines. Hs683 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; HyClone, Logan, USA) supplemented with 10% fetal bovine serum (FBS; Gibco, NY, USA) and 1% penicillin-streptomycin (HyClone, Logan, USA) at 37°C with 5% CO₂. 10% Fetal bovine serum (FBS; Gibco, NY, USA) was supplemented to Leibovitz’s (L)-15 media to develop a culture media for SW1088 cells. The fetal bovine serum (FBS; Gibco, NY, USA), 15%, and 1% penicillin-streptomycin (HyClone, Logan, USA) were added as supplements to the Dulbecco’s modified Eagle’s medium (DMEM; HyClone, Logan, USA) for the NHAs’ culture.

A faster reagent (Invitrogen) was used to extract total RNA from cultured cells. PrimeScript RT Reagent Kit (TaKaRa, Shiga, Japan) was employed to create cDNA from isolated RNA, and SYBR Green PCR Master Mix was utilized for quantitative reverse transcription polymerase chain reaction (qRT-PCR). GAPDH was employed as an internal loading control, and expression levels were quantified by employing the 2^–ΔΔCt method. GraphPad Prism version 9.0.1 (GraphPad Software, San Diego, California USA, www.graphpad.com) was used for the visualization of qRT-PCR results and a two-sample unpaired t-test. Tsingke Biotech (Tsingke, China) synthesized the complete set of primers used for qRT–PCR. Supplementary Table 2 lists all the primer sequences used in this work.

R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) was used to perform analyses, and the Perl language, which was mainly used for batch cleaning and collation of the data. The “limma” R package (37) was employed for DEGs screening. The Wilcoxon test was applied for differential analysis to compare the two groups in the bioinformatics analysis part. For the Analysis of Variance, the Kruskal-Wallis test was employed for comparisons involving more than two groups. The Spearman’s correlation coefficient was employed in this study’s correlation analysis unless otherwise noted. The survival of the various patient groups was compared using the K-M survival analysis and the log-rank test. A two-tailed p value < 0.05 was regarded as statistically significant for all statistical analyses.

Figure 1 illustrates the workflow of this study. The differential expression and prognostic significance of 19 ANOIRGs in LGGs were analyzed using the combined expression profiles of the TCGA and GTEx cohorts and survival prognostic data of the TCGA cohort. The findings demonstrated that PLAG1 and SNAI1 were expressed at low levels in LGGs compared to normal tissues, and the remaining 17 ANOIRGs were highly expressed in LGGs ( Supplementary Figure 1A ). The co-expression prognostic network of the 19 ANOIRGs showed that GLUD1 and NTRK2 were protective factors based on Univariate Cox regression analysis, whereas the remaining 17 ANOIRGs were LGG risk factors. Broadly speaking, 17 risk ANOIRGs positively correlated with each other, whereas two protective ANOIRGs correlated negatively with each other ( Supplementary Figure 1B ). The K-M survival analysis showed that the remaining 18 ANOIRGs had a significant effect on the prognosis of patients with LGG except for WNT2 (p < 0.05; Supplementary Figure 1C ). Mutation analysis showed that ANOIRGs-related mutations occurred in 43 of 523 samples, with an incidence of 8.22% ( Supplementary Figure 2A ). Among the 43 mutations, EGFR had the highest mutation frequency of 6%, mainly missense mutations ( Supplementary Figure 2A ). The outcome of CNV analysis showed that SPIB had the highest copy number deletion frequency, while EGFR had the highest increase in copy number frequency ( Supplementary Figure 2B ). Finally, we constructed a CNV landscape of 19 ANOIRGs at chromosomal loci ( Supplementary Figure 2C ).

TCGA-LGG cohort samples were clustered according to the consistent expression of the 19 ANOIRGs into four different modification patterns representing four different molecular subtypes ( Figure 2A ). The consensus clustering results are shown in Supplementary Figures 3A–C . The K-M survival analysis revealed significant differences in OS among the four anoikis related subtypes, with patients in anoirgclusterA and anoirgclusterC having a better prognosis than those in anoirgclusterB and anoirgclusterD (p < 0.001, Figure 2B ). In addition, we plotted box plots ( Figure 2C ) and heatmaps ( Figure 2D ) to visualize the differential expression of 19 ANOIRGs in different molecular subtypes and found that the expression of HGF, KIF11, ECT2, CCT5, ERBB2, POU3F2, SPIB and SNAI2 was higher in anoirgclusterB and anoirgclusterD than in anoirgclusterA and anoirgclusterC (p < 0.01). Compared with the histological type, there were mostly astrocytomas in anoirgclusterB and anoirgclusterD and oligoastrocytomas and oligodendeogliomas in anoirgclusterA and anoirgclusterC. Grade 3 was also the most common grade in anoirgcluster B and anoirgcluster D, while grade 2 was mostly observed in anoirgcluster A and anoirgcluster C ( Figure 2D ). Furthermore, GSVA was employed to compare the enriched pathway of both macrosubtypes of anoirgclusterA and anoirgclusterC, and anoirgclusterB and anoirgclusterD in the KEGG pathway ( Supplementary Figure 4A ), HALLMARK pathway ( Supplementary Figure 4B ), and Reactome pathway ( Supplementary Figure 4C ), and detected significant pathway variations between them (primarily enriched in cell cycle-related pathways).

Patients with different anoikis related subtypes exhibited t-SNE-based ANOIRGs’ expression distinguishability feature ( Supplementary Figure 5A ). Then, we applied stromal, immune, and ESTIMATE scores to all LGG samples to gauge the degree of TIME cell infiltration in various subtypes. According to Supplementary Figure 5B , anoirgclusterB and anoirgclusterD outperformed anoirgclusterA and anoirgclusterC in terms of StromalScore, ImmuneScore, and ESTIMATEScore (p < 0.001). Additionally, we employed ssGSEA for quantifying the infiltrating abundance of 23 immune cells and investigate the distinct patterns of the immune-infiltrating landscape of the four subtypes in order to characterize the immune cell infiltration in various anoikis-related subtypes. anoirgclusterB and anoirgclusterD manifested a higher infiltration level of activated B, dendritic, CD4 T, CD8 T, CD56 bright natural killer, macrophages, gamma delta T, immature B, mast, dendritic, natural killer, natural killer T, regulatory T, plasmacytoid dendritic, type 1 T helper, type 2 T helper, T follicular helper, and myeloid-derived suppressor cells (MDSCs) than anoirgclusterA and anoirgclusterC (p <0.05, Supplementary Figure 5C ). The outcome suggests that anoirgclusterB and anoirgclusterD have an elevated degree of stromal and immune cell infiltration than anoirgclusterA and anoirgclusterC. Based on these results, we found that anoirgclusterAC and anoirgclusterBD have similar ANOIRG expression profiles and prognostic and immune infiltration characteristics, which supports our classification of samples into two macrosubtypes: anoirgclusterAC and anoirgclusterBD.

For an additional investigation on the potential biological behavior of various anoikis subtypes, we carried out a differential analysis of the two anoikis macrosubtypes. DEGs with |logFoldChange| >1 and p < 0.05 were then screened, and 1251 DEGs were identified. Volcano plots showed that DEGs mainly expressed in the large anoirgclusterBD subtypes were highly expressed (p < 0.05, Figure 3A ). KEGG ( Figure 3C ) and GO ( Figure 3D ) enrichment analysis was conducted for these DEGs, showing that the top five pathways were based on adjusted p-values in KEGG and the corresponding network of relationships with associated genes (p < 0.05, Figure 3B ). A number of pathways are associated with the cell cycle.

A univariate Cox regression analysis on 1251 DEGs associated with anoikis was performed for the identification of DEGs with prognostic significance for LGG, yielding 1218 differentially expressed prognostic genes. A further 669 genes (SDEPGs) were selected based on the criterion of |1-HR| >0.5. Based on the 669 SDEPGs, secondary clustering was performed two clusters were identified: genecluster C1 and C2 ( Figure 4B ). Supplementary Figures 3D–F presents the clustering results. As depicted by the K-M survival curves, a significantly better prognosis was predicted for genecluster C1 in comparison to genecluster C2 (p < 0.001, Figure 4A ). Eighteen ANOIRGs had significantly different expressions between genecluster C1 and C2 (p < 0.001, Figure 4C ). In addition, a heatmap was employed to show the distribution of expression of 669 SDEPGs and clinicopathological characteristics for both geneclusters ( Figure 4D ).

First, we performed LASSO regression analysis for downscaling screening based on 669 SDEPGs ( Supplementary Figures 6A, B ) and obtained a total of 34 genes. We then calculated the gene importance scores based on the Gini coefficient method in random forest, selected genes with scores ≥2 as disease signature genes ( Supplementary Figures 6C, D ), obtained 20 genes to enter the multivariate cox regression analysis, and finally screened 12 APRGs to construct the anoikis scoring system anoiS. The results of the multivariate Cox regression analysis for the 12 APRGs are shown in Supplementary Table 1 . Among them, KDELR2, SMC4, IQGAP2, WEE1, HOXD13, SLC43A3, CYP27B1, MAP3K1, PIM1, and APOBEC3C were highly expressed in LGGs ( Supplementary Figure 7A ), and their high expression was associated with poor prognosis ( Supplementary Figure 7B ). Next, we divided the cohort samples into high- and low-anioS groups based on the median anioS. The K-M survival analysis showed that patients in the high anoiS group had poorer OS and PFS than those in the low anoiS group (p < 0.001, Figures 5A, D ). The ROC curves also showed strong predictive accuracy of anioS, with AUCs for OS at 1, 3, and 5 years of 0.922, 0.947, and 0.870 for OS, respectively ( Figure 5B ), and AUCs of 0.749, 0.705 and 0.726 for PFS at 1, 3 and 5 years, respectively ( Figure 5E ). It was discovered that the anoiS has good accuracy in predicting OS and PFS at 1, 3, and 5 years in patients with LGG by using the calibration curve to test the predictive utility of the model ( Figures 5C, F, I ). We also used the CGGA dataset to confirm the reliability and stability of the 12-gene signature prediction model. The prognosis of the high anoiS group was considerably poorer than that of the low anoiS group, according to a survival analysis (p < 0.001, Figure 5G ). With AUCs of 0.741, 0.735, and 0.724 at 1, 3, and 5 years, respectively, the ROC curve data showed that anoiS had a significant prognostic prediction potential ( Figure 5H ). The anoiS appeared to be a clinically independent risk prognostic factor for patients with LGG in the TCGA-LGG cohorts, according to univariate and multivariate Cox regression analyses paired with other clinical subgroup features (p < 0.001, Figures 5J, K ). Sankey plots revealed a correlation between the prognosis, anioS, and subtype ( Supplementary Figure 6E ). AnoirgclusterB and anoirgclusterD had a higher chance of matching geneclusterC2, higher anioS, and a worse prognosis in LGG patients. Boxplots were also used to confirm this conclusion. (p < 2.22e-16, Supplementary Figures 6F, G ).

Stacked histograms were employed to demonstrate the percentage of individual clinical characteristics in the high and low anoiS groups and box plots were plotted to show the variations in the anoiS between various clinical subgroup features, in order to further investigate the relationship between the anoiS and clinical subgroup characteristics. Patients older than 45 years demonstrated a higher risk in comparison to those younger than 45 years (p = 6.1e-06, Figure 6A ), the anoiS did not show significant differences between gender ( Figure 6B ), anoiS was higher in patients graded G3 than in those with G2 (p < 2.22e-16, Figure 6C ), and patients with astrocytomas had higher anoiS than those with oligoastrocytomas and oligodendeogliomas (p < 0.05, Figure 6D ).

We further investigated TMB’s connection to the anoiS because it is intimately tied to tumor immunity. The findings revealed a positive correlation between TMB and anoiS ( Figure 6F ) and that TMB in the high anoiS group was higher in comparison to that in the low anoiS group ( Figure 6E ). Additionally, the samples were split into high and low TMB groups using the optimal cutoff value as the boundary, and it was discovered that the high TMB group’s prognosis was poorer than the low TMB group’s ( Figure 6G ). Patients with high TMB and anoiS demonstrated the poorest prognosis, whereas those with low TMB and anoiS had the best prognosis, according to a combined study of the effect of TMB and anoiS status on patient prognosis ( Figure 6H ).

To elucidate the potential genomic mutational mechanisms associated with anoikis, we used waterfall plots to depict the mutational landscape in the high and low anoiS populations; IDH1, TP53 and ATRX had a higher mutation frequency in the cohort, with mutation rates of 64%, 50%, and 35% in the high anoiS group ( Figure 6I ) and 90%, 41%, and 33% in the low-anoiS group, respectively ( Figure 6J ). The mutation rates of IDH1 and IDH2 were lower in the high anoiS group than in the low anoiS group, while the mutation rate of TP53 was higher than that in the low anoiS group.

A correlation matrix ( Figure 7A ) was constructed for investigating the association between anoiS and immune cell infiltration, and we found that there was a substantial positive correlation between the two (p < 0.05, with the exception of CD56 dim natural killer cells and Eosinophils). In Supplementary Figure 8A , findings for particular individual cells have been displayed. With the utilization of ESTIMATE to assess the infiltration characteristics of the tumor microenvironment in LGG patients, it was found that the StromalScore, ImmuneScore, and ESTIMATEScore were substantially higher in the high anoiS group than in the low anoiS group (p < 0.05, Figure 7B ). Thereafter, the association between anoiS and immunotherapy efficacy was verified using the IMvigor210 immunotherapy cohort. The outcome implies that individuals with a high anoiS had a good prognosis ( Figure 7C ). We additionally evaluated the correlation between the level of expression of 12 APRGs and the infiltration level of 23 immune cells. As shown in Supplementary Figure 8B , among the 12 APRGs, FAM133A negatively correlated with most immune cells (except CD56 dim natural killer cells and Eosinophils), CMYA5 negatively correlated with Eosinophils, whereas positively correlated with activated CD56 dim natural killers, CD4+T cells, CD8+T cells, type 1 T helper cells, and type 2 T helper cells, plasmacytoid dendritic cells. CYP27B1 negatively correlated with CD56 dim natural killer cells, type 1 T helper cells, Eosinophils, and Monocytes, whereas positively correlated with CD56 bright natural killers, activated CD4+T cells, and type 2 T helper cells. The remaining nine APRGs positively correlated with most immune cells (except CD56 dim natural killer cells, type 1 T helper cells, Eosinophils, and Monocytes). Next, we evaluated the correlation between the 12 APRGs and 46 immune checkpoint genes (ICGs). Most of the 12 APRGs positively correlated with a number of immune checkpoints, such as PD-1, PD-L1, CTLA-4, TIM-3, B7-H3, IDO1, and LAG3, while FAM133A negatively correlated with CTLA-4, TIM -3, PD-1, PD-L1, B7-H3, IDO1, and LAG3, and anoiS positively associated with most immune checkpoints, such as CTLA-4, TIM-3, PD-1, PD-L1, B7-H3, IDO1, and LAG3 ( Figure 7D ).

The baseline data for the CGGA-LGG cohort is presented in Table 2 . The proportion of individual gene status in the high and low anoiS groups was shown using stacked histograms and boxplots were plotted to show the difference in the anoiS between gene statuses. The findings depict that individuals with the IDH mutation (p<0.001, Figure 8A ) and 1p19q co-deletion (p<0.001, Figure 8B ) had lower anoiS, and patients placed in the high anoiS group manifested a lower proportion of IDH mutations (p<0.001, Figure 8A) and 1p19q co-deletion (p<0.001, Figure 8B ). Patients with MGMT non-methylation had higher anoiS than MGMT-methylated patients, while MGMT methylation status did not vary substantially between patients in the low and high anoiS groups (p=0.024, Figure 8C ). The comparison of clinical characteristics showed that the anoiS was the highest and MGMT promoter methylation was the lowest in AUC values at 1, 3, and 5 years of OS ( Figures 8D–F )

In the high anoiS group, the prognosis of patients in the IDH mutant group was better in comparison to that of patients in the IDH wildtype group, as shown by the survival analysis (p < 0.001, Figure 8G ); in the low anoiS group, IDH mutant group and the patients placed in the IDH wildtype group, did not exhibit a significant difference in terms of the prognosis (Figure 8H). For 1p19q co-deletion, survival analysis demonstrated that the prognosis of patients in the 1p19q co-deletion group was improved in comparison to that of patients in the 1p19q co-deletion negative group, regardless of high or low anoiS ( Figures 8I, J ). For MGMT methylation status, regardless of the high and low anoiS, no significant difference existed in terms of patients’ prognosis in the MGMT methylation group and MGMT non-methylation group as depicted by the survival analysis ( Figures 8K, L ).

In addition, the prognosis was predicted separately for the different treatment modalities using the anoikis scoring system, and survival analysis depicted that patients placed within the high anoiS group manifested a significantly worse prognosis in comparison to the ones in the low anoiS group in the chemotherapy alone (p = 0.001, Figure 9A ), radiotherapy alone (p < 0.001, Figure 9B ), and radiochemotherapy (p < 0.001, Figure 9C ) groups. Additionally, we found that for patients treated with radiotherapy alone or chemotherapy alone, the prognosis of tended to be better than that of the no-treatment and radiotherapy populations in both the high anoiS group (p = 0.031, Figure 9D ) and the low anoiS group (p = 0.024, Figure 9E ).

We first predicted the IC50 values of TMZ in LGG samples using both “pRRophetic” and “oncoPredict” algorithms to reflect the sensitivity of patients to TMZ. Patients in the high anoiS group demonstrated lower IC50 values of TMZ ( Figures 10A, B ) and lower AUC values of TMZ ( Figure 10C ), The IC50 values ( Figures 10D, F ) and AUC values ( Figure 10G ) of TMZ were negatively correlated with the anoiS. We then constructed a correlation matrix of 19 ANOIRGs and 12 APRGs using TMZ IC50 values. CCT5 showed the strongest negative correlation with TMZ IC50 values among the 19 ANOIRGs ( Figure 10E ) and was a differential prognostic gene ( Figures 11A, B ), which plays a potential cancer-promoting role in LGGs. Therefore, we regarded CCT5 as the key gene for the study. We selected CCT5 and 12 APRGs for constructing a gene co-expression network and found that CCT5 had a strong positive correlation with KDELR2, WEE1, SMC4 and MAP3K1 (r > 0.5; Figure 11C ). These four anoikis potential genes also showed differential prognosis, exerted potential pro-cancer effects ( Figure 11D ), and negatively correlated with the TMZ IC50 values. Therefore, we considered CCT5 and these four anoikis genes as potential gene teams. The key gene CCT5 showed the strongest negative correlation with the IC50 value of TMZ ( Figures 10H, I ). Additionally, qRT-PCR was used to detect the expression differences of CCT5 ( Figure 11E ), KDELR2 ( Figure 11F ), SMC4 ( Figure 11G ), WEE1 ( Figure 11H ), and MAP3K1 ( Figure 11I ) between NHA and LGGs cells (SW1088 and Hs683). In LGGs cells, the expression of these genes demonstrated a notable up-regulation in comparison to that in human astrocytes (NHA) (p <0.05).