Abstract
There is mounting evidence that ischemic cerebral infarction
contributes to vascular cognitive impairment and dementia in elderly.
Ischemic stroke and glioma are two majorly fatal diseases worldwide,
which promote each other's development based on some common underlying
mechanisms. As a post-transcriptional regulatory protein, RNA-binding
protein is important in the development of a tumor and ischemic stroke
(IS). The purpose of this study was to search for a group of
RNA-binding protein (RBP) gene markers related to the prognosis of
glioma and the occurrence of IS, and elucidate their underlying
mechanisms in glioma and IS. First, a 6-RBP (POLR2F, DYNC1H1, SMAD9,
TRIM21, BRCA1, and ERI1) gene signature (RBPS) showing an independent
overall survival prognostic prediction was identified using the
transcriptome data from TCGA-glioma cohort (n = 677); following which,
it was independently verified in the CGGA-glioma cohort (n = 970). A
nomogram, including RBPS, 1p19q codeletion, radiotherapy, chemotherapy,
grade, and age, was established to predict the overall survival of
patients with glioma, convenient for further clinical transformation.
In addition, an automatic machine learning classification model based
on radiomics features from MRI was developed to stratify according to
the RBPS risk. The RBPS was associated with immunosuppression, energy
metabolism, and tumor growth of gliomas. Subsequently, the six RBP
genes from blood samples showed good classification performance for IS
diagnosis (AUC = 0.95, 95% CI: 0.902–0.997). The RBPS was associated
with hypoxic responses, angiogenesis, and increased coagulation in IS.
Upregulation of SMAD9 was associated with dementia, while
downregulation of POLR2F was associated with aging-related hypoxic
stress. Irf5/Trim21 in microglia and Taf7/Trim21 in pericytes from the
mouse cerebral cortex were identified as RBPS-related molecules in each
cell type under hypoxic conditions. The RBPS is expected to serve as a
novel biomarker for studying the common mechanisms underlying glioma
and IS.
Keywords: elderly, glioma, ischemic stroke, RNA binding protein (RBP),
dementia
Introduction
There is mounting evidence that ischemic cerebral infarction
contributes to vascular cognitive impairment and dementia in elderly,
but the origins of ischemic cerebral infarction are unclear.
Understanding the vascular pathologies that cause ischemic cerebral
infarction may yield strategies to prevent their occurrence and reduce
their deleterious effects on brain function (Hartmann et al.,
[37]2018). Worldwide, ischemic stroke (IS) accounts for about 70% of
all stroke events, with a proportion as high as 87% in the United
States, and is also the second leading cause of death (Musuka et al.,
[38]2015; Benjamin et al., [39]2019; Phipps and Cronin, [40]2020).
Glioma is a common type of invasive brain tumor and the leading cause
of primary brain tumor-related deaths (Musuka et al., [41]2015; Velasco
et al., [42]2019; Wang E. et al., [43]2019). Among them, glioblastoma
multiforme (GBM; WHO IV) accounts for 45.2% of all primary malignant
tumors of the central nervous system (CNS) and is one of the fatal
tumor types, with a median survival time of fewer than 15 months
(Ostrom et al., [44]2013; Bi et al., [45]2020; Wang et al., [46]2020).
Clinical studies show that glioma and cerebral ischemia can promote
each other's occurrence during disease development and treatment (Fraum
et al., [47]2011; Seidel et al., [48]2013; Wojtasiewicz et al.,
[49]2013; Farkas et al., [50]2018; Noda et al., [51]2022). Previous
studies have reported that the incidence of brain tumors is higher in a
cohort of patients with IS than in those without a history of IS
(Qureshi et al., [52]2015; Chen et al., [53]2017; Tanislav et al.,
[54]2019). Similarly, stroke is a common complication among patients
with tumor. A postmortem-based study reported that about 14.6% of
non-CNS cancer cases showed cerebrovascular disease (CVD) (Graus et
al., [55]1985). Moreover, embolic strokes are the most common cause of
strokes in patients with cancer, possibly due to intravascular
coagulopathy (Cestari et al., [56]2004); patients with active cancer
show multiple infarcts (Kikuno et al., [57]2021). Gliomas account for
60% of ischemic strokes secondary to primary brain tumors, whereby
complications due to surgery and radiotherapy form the majority (Kreisl
et al., [58]2008). In these coexisting diseases, stroke is usually
missed, often leading to increased neurological disabilities and
injuries in susceptible individuals. Therefore, the pathogenesis of
glioma could provide potential mechanisms for cerebral ischemia.
RNA-binding protein (RBP) is a large protein family, which plays a
vital role in regulating gene expression through interactions with RNA.
These proteins participate in many biological processes, such as
splicing, lysis, and polyadenylation, as well as mRNA editing,
localization, stabilization, and translation (Kedde et al., [59]2007;
Liao et al., [60]2020; Van Nostrand et al., [61]2020). In addition,
some studies suggest that the interaction between RBP and RNA plays a
vital role in the occurrence and development of cancers (including
renal cell carcinoma, triple-negative breast cancer, and lung squamous
cell carcinoma) (Mohibi et al., [62]2019; Duan and Zhang, [63]2020; Kim
et al., [64]2020; Li et al., [65]2020; Qin et al., [66]2020). In this
context, many RBPs are reportedly associated with a poor prognosis of
gliomas (Shao et al., [67]2013; Barbagallo et al., [68]2018; Lan et
al., [69]2020). In IS, several RBPs participate in the development and
influence the prognoses of these patients by promoting inflammatory
reactions (Zhou et al., [70]2014; Sharma et al., [71]2021), increasing
cerebrovascular permeability, promoting vasogenic cerebral edema
(Ardelt et al., [72]2017), regulating apoptosis (Si et al., [73]2020;
Zhang et al., [74]2020), and protecting neurons (Fang et al.,
[75]2021).
In the development of glioma and ischemic stroke events, some common
pathways, such as hypoxia, brain inflammation, angiogenesis, and
hypercoagulability, have been identified (Ghosh et al., [76]2019).
Among them, hypoxia is the most widely accepted basis for building
research models for studying the common mechanisms underlying glioma
and IS (Søndergaard et al., [77]2002; Kasivisvanathan et al.,
[78]2011). However, the specific mechanism of co-occurrence or mutual
promotion of glioma and ischemic stroke remains unclear. Many clinical
studies have described this problem from a clinical perspective and
expounded the possible common pathway underlying the pathogenesis from
the perspective of each disease. Several RBP molecular or molecular
combination markers are used to identify specific subgroups of patients
with glioma, showing poor survival. Similarly, several RBPs are
involved in the development of IS. However, there is a lack of a
comprehensive analysis of the RBP family of genes in the common pathway
underlying glioma and IS. Through an in-depth study on the role of
RBPs, we hypothesized that RBP signature could not only provide an
effective identification for molecular subtypes of patients with glioma
with a poor prognosis but also yield certain reference values for the
diagnosis of IS. Such biomarkers will also provide more reliable risk
stratification and treatment targets for the clinicians to customize
more accurate personalized treatment plans and ultimately improve the
treatment efficacy.
Bioinformatics based on medical big data has solid advantages for
analyzing the common molecular mechanisms and pathways for such
coexisting diseases. In addition, the combination of radiomics and
machine learning shows a good performance in image-based diagnosis or
molecular subtype prediction and is more convenient for clinical
application (Acs et al., [79]2020). The primary purpose of this study
was to identify a group of RBP genes related to the prognosis of glioma
and the occurrence of IS, and elucidate their mechanism in glioma,
dementia, and IS. First, we identified a panel of RBP genes related to
the prognosis and analyzed the pathogenesis of these genes in glioma.
Next, using the radiomics features from MRI images, an automatic
machine learning classifier was used to predict risk stratification
based on this RBP gene signature in glioma. Finally, using bulk RNA
sequencing (RNA-seq) and single-cell RNA sequencing (scRNA-seq) data,
the classificatory performance and the potential mechanism of these RBP
genes in IS were analyzed.
Materials and methods
Research design and data extraction
According to the research purpose, the study design was divided into
two stages. The first stage involved the identification of a 6-RBP gene
signature (RBPS) and functional analysis of the RBPS in glioma. The
second stage was evaluating the expression and functions of the RBPS in
IS and mouse cerebral cortex cells under hypoxic conditions.
The first stage could be subdivided into three steps as follows: the
discovery and verification of biological gene markers and automatic
machine learning prediction based on radiomics features. First, the
standardized RNA expression profile data of 677 patients with glioma
(including 698 tumor tissues and 5 adjacent normal tissue samples) were
downloaded from TCGA ([80]https://portal.gdc.cancer.gov/), and a 6-RBP
gene signature related to the prognosis of glioma was identified. Next,
the identified biomarkers were verified in independent clinical data
sets using the transcriptome RNA expression profile data and clinical
characteristics of patients with glioma (N = 970; Verification Cohort
1) in CGGA ([81]https://www.cgga.org.cn/). Moreover, the clinical data
of patients with Grade IV glioma in the [82]GSE72951 data set
(Erdem-Eraslan et al., [83]2016) (N = 110; Validation Cohort 2) and the
gene expression profile data from gene chip analysis were obtained from
the Gene Expression Omnibus (GEO) database for verification. Finally,
using MRI-based radiomics features, an automatic machine learning
classifier was constructed to predict the RBPS. MRI-based radiomics
feature data from 132 patients with glioma were downloaded from TCIA
(Clark et al., [84]2013) and used to train classifiers for predicting
RBPS risk stratification (Bakas et al., [85]2017; Beers et al.,
[86]2018).
In the second stage, the possible mechanism underlying the six RBP
genes in stroke and dementia was evaluated, and the gene regulatory
network related to hypoxia was analyzed in mouse cerebral cortex cells.
First, the [87]GSE16561 dataset was retrieved from the GEO database to
examine differentially expressed genes (DEGs) related to ischemic
stroke. RNA-seq data in this dataset were derived from peripheral blood
samples of 39 patients with ischemic stroke and 24 healthy controls
(Barr et al., [88]2010; O'Connell et al., [89]2016, [90]2017). The
[91]GSE36980 dataset was used to explore DEGs associated with
Alzheimer's disease, which included 33 patients with AD and 47 non-AD
controls (Hokama et al., [92]2014). In addition, to study the
expressions of the related genes at a single-cell level, RNA-seq data
from 7,925 isolated mouse cerebral cortex cells were obtained from the
[93]GSE125708 dataset. In this data set, mice were divided into two
groups: one group living in indoor air for 7 days was the normal oxygen
concentration group, and the other group living in 7.5% oxygen
concentration for 7 days was the hypoxia concentration group. Using
this dataset, we examined the regulatory changes for the RBPS-related
genes with changes in the oxygen concentration (Heng et al., [94]2019).
Analysis of differentially expressed RBP genes
A total of 1,542 RNA-binding protein genes were collected from a
published dataset (Gerstberger et al., [95]2014). Differentially
expressed RBP genes were analyzed between tumor samples and normal
samples adjacent to cancer in the TCGA-glioma dataset. An adjusted p
value < 0.05 using the Benjamini-Hochberg false discovery rate (FDR)
method (FDR < 0.05) and a logarithmic value of fold change >1
(|log[2]FC| > 1) were used as the cut-off criteria to screen
differentially expressed RBP genes. Differentially expressed genes
(DEGs) were used to perform Gene Ontology (GO) and Kyoto Encyclopedia
of Genes and Genomes (KEGG) pathway enrichment analysis using the
online DAVID database. The protein-protein interaction (PPI) analysis
of DE-RBP genes was performed using STRING software
([96]https://string-db.org/). Cytoscape software was used to build a
sub-network to identify the PPI network's core DEGs. The “limma”
(Ritchie et al., [97]2015) and “sva” (Leek et al., [98]2012) R packages
were used to remove the batch effect for the gene expression data of
the shared RBP genes in TCGA, CGGA, and [99]GSE72951 datasets.
Construction of a 6-RBP gene signature
To identify a clinically translatable RBP gene signature, the
univariate Cox proportional hazard regression model and the Lasso
penalty Cox regression model were used for evaluating the association
of RBP genes in predicting overall survival (OS) in patients with
glioma. Next, RBPS was constructed, and its value in predicting OS was
evaluated. The risk score (RS) of the RBPS was calculated based on the
linear combination of the gene expression (EXP[i]) multiplied by the
corresponding coefficient (Coef[i]).
[MATH: RS=∑i=1nCo<
/mi>efi×EXP
i :MATH]
(1)
The median value of the gene signature risk score was used as a cut-off
threshold to divide the entire patients with glioma into high- and
low-risk groups. The Kaplan-Meier (K-M) method was used to plot
survivor curves. The log-rank test was used to calculate the
statistical difference between the two groups to evaluate the
correlation of the RBPS with the OS outcome. Receiver operator
characteristic (ROC) curve analysis of the RBPS with prognosis was
performed using the “survivalROC” package
([100]https://CRAN.R-project.org/package=survivalROC), and 95%
confidence intervals (CI) of the area under the curve (AUC) were
calculated by the “timeROC” (Blanche et al., [101]2013) package.
Risk stratification of the RBPS
The expressions of the RBPS genes in samples were analyzed using the
“pheatmap” package. The risk scores of RBPS were sorted from low to
high to evaluate the relationship between the risk scores and patients'
living status and overall survival time. Circosplot was drawn using the
“RCircos” (Zhang et al., [102]2013) package to show the copy number
variant status distribution of the RBPS genes and their position on
chromosomes. To explore the relationship between the expression and
copy number variant status of the RBPS genes, the differential
expression analyses of RBPS genes among different copy number variants
were performed to explore the role of a gene copy number variant in
RBPS genes expression.
Gene set enrichment analysis
Gene set enrichment analysis (GSEA) is a bioinformatics algorithm used
to identify the differential expression of biological pathways between
two biological states (Subramanian et al., [103]2005). GSEA was used to
identify the pathway related to the RBPS. The
“c2.cp.kegg.v7.1.symbols.gmt[Curated]” gene set collection from the
Molecular Signatures Database (MSigDB) was used as a reference for
enrichment analysis (Subramanian et al., [104]2005; Liberzon et al.,
[105]2011, [106]2015). The false discovery rate (FDR) and the
normalized enrichment score (NES) were used to sort the KEGG pathways.
Association between the RBPS and glioma stemness
The tumor stemness index refers to the gradual loss of cell
differentiation phenotype and acquisition of progenitor cell and
stem-cell-like characteristics during tumor progression (Malta et al.,
[107]2018). Two types of glioma stemness indices, namely, the RNA
expression-based stemness score (RNAss) and the DNA methylation-based
stemness score (DNAss) were downloaded from UCSC Xena (Goldman et al.,
[108]2020) to evaluate the correlation between the RBPS and glioma
stemness indices. The stemness indices range from 0 to 1, where 0
indicates a high degree of differentiation, and 1 indicates
undifferentiated.
Immune-related tumor microenvironment and potential compounds
First, the “ESTIMATE” algorithm (Yoshihara et al., [109]2013) was used
to calculate the immune-related tumor microenvironment features from
gene expression data, including stromal, immune, and ESTIMATE scores.
The profiles of six immune subtype categories representing TME features
and potential therapeutic and prognostic implications were downloaded
from UCSC Xena (Thorsson et al., [110]2018). In addition, the abundance
of 22 infiltrating immune cell types was calculated and inferred from
RNA expression profiles using CIBERSORTx (Newman et al., [111]2019).
Moreover, a list of important immune checkpoint molecules, including
PD-1, PDL1, and CTLA-4, was obtained. In the TCGA-glioma cohort, the
correlations between these immune-related features and RBPS were
analyzed. Finally, to identify the potential drugs targeting these RBPS
genes, drug concentration and gene expression profiles were downloaded
from CellMiner (Reinhold et al., [112]2012) to perform correlation
analysis. Drugs were filtered according to FDA's approval results for
clinical trials.
Radiomics-based TPOT analysis
Radiomics features data were downloaded from TCIA to establish an
Automatic Machine Learning (AutoML) prediction model. Radiomics
features were extracted from T1WI, T2WI, Flair, and T1Gd images (Bakas
et al., [113]2017; Beers et al., [114]2018), including 483 usable
features. Univariate logistic regression analysis evaluated the
association between each Radiomics feature and RBPS in patients with
glioma, and RBPS-related radiomics features were selected for autoML
model training. The steps of autoML model construction include features
selection, parameters selection, and final model selection, which were
fully automated using the Tree-based Pipeline Optimization Tool (TPOT)
(Le et al., [115]2020). TPOT is an automated machine learning tool
based on Python, which uses genetic algorithm programming
([116]https://github.com/rhiever/tpot) to optimize the machine learning
pipeline. Before TPOT analysis, the dataset was randomly divided into a
training set (99 patients) and a test set (33 patients) according to
3:1, and the random number state was fixed at 42. The training process
was set as follows: generations = 100, population size = 100, and
10-fold cross-validation on the training set. Finally, TPOT will return
a model with the best classification performance and parameters. The
TPOT was repeated 20 times, and the models were sorted by the area
under the curve (AUC). After that, the top 10 models with the best
performance were screened out. In addition, ROC curves and
precision-recall curves were also used to compare the performance of
these ten models. By comparing the sensitivity, specificity, accuracy,
AUC, and average precision (AP) of these ten models, the best model was
finally determined based on the accuracy metrics (Su et al.,
[117]2019).
Single-cell analysis
RNA-seq data of 7,925 single cells from the mouse cerebral cortex under
normoxia and hypoxia conditions were analyzed using the “Seurat”
package (Stuart et al., [118]2019). Based on pre-set filter conditions
(at least 200 expressed genes but no more than 6,000 expressed genes,
RNA counts >1,000, mitochondrial gene expression <20%, and
hemoglobin-related gene expression <1%), a total of 7,789 cells and
14,271 gene features were filtered for further single-cell analysis.
The scRNA-seq data were integrated with the “SCTransform” function and
then processed using “RunPCA” and “RunUMAP” functions, including noise
removal, information extraction, and cell dimension reduction. The
results of cell dimensionality reduction were visualized with uniform
manifold propagation and projection (UMAP) (Becht et al., [119]2019) to
observe the effect of batch effect removal between groups. The
“FindNeighbors” and “FindClusters” functions were used to detect cell
clusters. Finally, each cell cluster was annotated according to the
commonly used marker genes of cell types. After cell annotation,
microglia, astrocytes, and pericytes were extracted as cell subsets,
and “FindMarker” was used to calculate differentially expressed genes
in those cell types between hypoxia and normoxia conditions.
Single-cell regulatory network inference and clustering
Single-cell regulatory network inference and clustering (SCENIC) was
used to identify the main gene regulatory networks in each cell type
between different groups from single-cell transcripts (Aibar et al.,
[120]2017). First, pySCENIC (version 0.11) was used to identify the
major transcription factors and their corresponding gene regulatory
networks in mouse cerebral cortex cells. Transcription factors and
their gene regulatory networks constitute a regulatory module called
regulon. Based on the expression of transcription factors and
downstream-regulated molecules in the regulon, the regulon activity
score (RAS) is used to measure the regulatory ability of each regulon
in each cell. Finally, based on the RAS, the regulon activity score
(RSS) is calculated to describe the regulatory power of each regulon in
each cell subtype, and the regulons in each cell type are ranked
according to RSS so as to infer the influence of each regulon on cell
transcription regulation in a specific cell type.
Pseudotime analysis and cell trajectory inference
Monoclec3 (version 1.0) and Monoclec2 (version 2.4) (Trapnell et al.,
[121]2014; Qiu et al., [122]2017a,[123]b; Cao et al., [124]2019) were
used to calculate the pseudotime and analyze cell trajectory based on
scRNA-seq transcripts from the mouse cerebral cortex for further
identifying transcriptional differences among these cells and examining
changes in RBPS and its transcription factors during cell fate
transition. First, differentially expressed genes were determined for
each cell type between normoxia and hypoxia groups. Then, the “DDRTree”
method was used to calculate the cell state for each cell type. The
velocyto.py (version 11.2) was used to calculate the RNA velocity in
each cell. The workflow, annotation files, and visual tools can be
obtained following the methods described in the previous studies (Vidal
et al., [125]2019; Lin et al., [126]2021).
Statistical analysis
All statistical analyses were performed using the R software (version
4.0.2, R Foundation for Statistical Computing, Vienna, Austria;
[127]http://www.r-project.org/) and Python (version 3.8). The “rms” R
package was used to draw the nomogram. Spearman correlation coefficient
and the Benjamini-Hochberg method adjusted-p value (FDR) were used for
correlation analysis. All p-values were two-sided, and p < 0.05 was
considered statistically significant.
Results
Differentially expressed RBP genes
First, a panel of 1,542 RBP genes was collected. Among them, 1,472 were
selected to analyze the differentially expressed RBP genes between
tumor and normal samples in TCGA. A total of 170 DEGs were identified
according to the preset filter conditions, and the results are shown in
the heat map ([128]Supplementary Figure 1A). Subsequently, the GO and
KEGG pathways for DEGs were analyzed, and the results showed that the
differentially expressed RBP genes were mainly enriched in RNA
processing-related pathways ([129]Table 1). Furthermore, the
protein-protein interactions (PPI) of DEGs were predicted and analyzed
using the STRING website, following which a PPI sub-network analysis of
DEGs was performed using the Cytoscape software ([130]Supplementary
Figures 1B,[131]C). The core genes and molecular interaction networks
related to the differential RBP genes were obtained through PPI
analysis.
Table 1.
GO function and KEGG pathway enrichment result.
ID Term P-value
hsa03010 Ribosome <0.001
hsa03015 mRNA surveillance pathway <0.001
hsa03018 RNA degradation <0.001
hsa03013 RNA transport 0.011
hsa03040 Spliceosome 0.022
GO:0000956 Nuclear-transcribed mRNA catabolic process <0.001
GO:0006401 RNA catabolic process <0.001
GO:0006402 mRNA catabolic process <0.001
GO:0022626 Cytosolic ribosome <0.001
GO:0000184 Nuclear-transcribed mRNA catabolic process,
nonsense-mediated decay <0.001
hsa03013 RNA transport <0.001
hsa03018 RNA degradation 0.020
hsa03015 mRNA surveillance pathway 0.020
hsa04914 Progesterone-mediated oocyte maturation 0.020
hsa03008 Ribosome biogenesis in eukaryotes 0.022
hsa04114 Oocyte meiosis 0.027
hsa04962 Vasopressin-regulated water reabsorption 0.028
hsa05134 Legionellosis 0.040
GO:0008380 RNA splicing <0.001
GO:0043484 Regulation of RNA splicing <0.001
GO:0050684 Regulation of mRNA processing <0.001
GO:0048024 Regulation of mRNA splicing, via spliceosome <0.001
GO:0000377 RNA splicing, via transesterification reactions with bulged
adenosine as nucleophile <0.001
[132]Open in a new tab
Identification of the 6-RBP gene signature
First, 170 differentially expressed RBP genes were screened and the
commonly shared intersecting genes in RNA expression profiles of
patients with glioma in TCGA, CGGA, and [133]GSE72951 datasets were
obtained. The filtered expression profiles from these three datasets
were further processed to remove batch effects. Next, by univariate Cox
analysis for TCGA glioma expression profile data, a total of 100 RBP
genes were analyzed along with the total survival time data, and the
top 17 RBP genes significantly related to survival were screened out
([134]Figure 1A). Finally, in the TCGA training set, the Lasso penalty
Cox regression analysis was performed to screen gene variables, and a
prognosis model was constructed according to the multivariate Cox
regression model. Using the lambda.min threshold ([135]Figure 1G), a
6-RBP gene signature (RBPS) was identified, comprising 6 RBP genes
(TRIM21, BRCA1, ERI1, POLR2F, DYNC1H1, and SMAD9). The RBPS was
associated with the adverse OS in glioma. The volcanic plot showed the
differential analysis results of these six RBP genes, which showed that
POLR2F and DYNC1H1 were downregulated in glioma, while TRIM21, BRCA1,
ERI1, and SMAD9 were upregulated in glioma ([136]Figure 2A).
[137]Figure 2B showed the copy number variation of these six RBP genes
and their positions on 24 chromosomes. The RBPS risk score (RS) was
calculated based on the linear combination of the expression values of
the six RBP genes multiplied by their corresponding coefficients. The
formula for calculating the RBPS RS was as follows:
Figure 1.
[138]Figure 1
[139]Open in a new tab
Construction of a prognostic signature based on OS events in glioma.
(A) Univariate Cox analysis for the top 17 RBP genes. (B) Multivariate
Cox analysis for six RBP genes. Survival analysis for patients with
glioma between the high- and low-risk groups in (C) TCGA and (D) CGGA
datasets. Yellow indicates high risk and blue indicates low risk for
glioma. Bioinformatics analyses for the 6-gene risk stratification
signature; receiver operator characteristic curve analysis for the
6-gene signature in (E) TCGA and (F) CGGA datasets. (G) Selection of
the tuning parameter (lambda) in Cox-penalized regression analysis via
10-fold cross-validation in the TCGA cohort. The vertical dotted lines
on the left and the right indicate “lambda.min” and “lambda.1se”
criteria, respectively. The red dots represent partial likelihood
deviation values, while the gray lines are the corresponding standard
errors. AUC, the area under the curve.
Figure 2.
[140]Figure 2
[141]Open in a new tab
Characteristics of the six RBP genes in the RBPS. (A) Differently
expressed genes between the normal and tumor samples are shown in the
volcano plot. The dots in red represent upregulated genes (Yang et al.,
[142]2020), while those in green are signed downregulated genes (Boucas
et al., [143]2015) in tumor samples. Significant differences were
determined using the thresholds of |log[2] FC|> 1 and FDR < 0.05. (B)
The location of the six RBP genes on the 24 chromosomes, as well as the
copy number variation events. The expression of the six RBP genes,
distribution of the RBPS risk scores, survivor status, and survival
time of the patients with glioma ranked by their risk scores in (C)
TCGA and (D) the CGGA datasets.
[MATH: RSRBPS=
0.294×EX<
mi>PTRIM21+0.525×EXPBRCA1
+0.400×EXPERI1-0.313×
EXPPO
LR2F
-0.303×E
XPDYN
C1H1-0.432×EX
PSMAD9 :MATH]
(2)
Among the six RBP genes constituting the RBPS, higher expression levels
of POLR2F, DYNC1H1, and SMAD9 were associated with a lower risk of
death (HR < 1). In contrast, higher expressions of TRIM21, BRCA1, and
ERI1 were associated with poorer overall survival (HR > 1; [144]Figure
1B; [145]Supplementary Table 1). Patients with glioma were stratified
according to the median value of the RBPS risk score in the TCGA cohort
and were divided into high-risk and low-risk groups. The 5-year OS
rates for RBPS-derived high- and low-risk patients were 19 and 75%,
respectively; WHO II-IV (HR: 6.76, 95% CI: 4.84–9.44; p < 0.001), WHO
II (HR: 3.47, 95% CI: 1.68–7.18, p < 0.001), WHO III (HR: 2.85, 95% CI:
1.75–4.64, p < 0.001), WHO IV (HR: 1.54, 95% CI: 1.04–2.26, p = 0.028;
[146]Figure 1C; [147]Supplementary Figures 2A–[148]C).
Subsequent validation in the CGGA dataset showed the outcomes were
consistent with the findings in the TCGA cohort; WHO II-IV (HR: 4.11,
95% CI: 3.40–4.95; p < 0.001), WHO II (HR: 2.02, 95% CI: 1.32–3.10; p =
0.001), WHO III (HR: 3.12, 95% CI: 2.32–4.19; p < 0.001), and WHO IV
(HR: 1.28, 95% CI: 1.03–1.60, p = 0.027; [149]Figure 1D;
[150]Supplementary Figures 2D–[151]F). These findings indicated that
Subsequent validation in tRBPS could predict adverse prognosis for
patients with glioma as well as the glioma subgroups based on the WHO
grades. In addition, the gene expression differences for these six RBP
genes with respect to copy number variation events were analyzed
([152]Supplementary Figures 3A–[153]F). The results showed that copy
number variants were significantly associated with mRNA expressions of
POLR2F (p < 0.001), DYNC1H1 (p < 0.001), TRIM21 (p < 0.001), SMAD9 (p <
0.001), and ERI1 (p = 0.005). This suggests that copy number variants
may be an important factor in the poor prognosis of RBPS.
RBPS is associated with a poor OS for glioma
In the TCGA discovery cohort, the RBPS showed robustness for
identifying the poor survival of gliomas, as evidenced by the good AUC
values for WHO grades: WHO II-IV (AUC = 0.887, 95% CI: 0.854–0.937),
WHO II (0.736, 95% CI: 0.577–0.919), WHO III (0.775, 95% CI: 0.7–0.893)
and WHO IV (0.665, 95% CI: 0.417–0.855; [154]Figure 1E). Similarly, in
the CGGA validation cohort, the AUC value of RBPS for identifying poor
OS prognoses in all patients with glioma was 0.819 (95% CI:
0.794–0.851): WHO II (0.705, 95% CI: 0.622–0.799), WHO III (0.769, 95%
CI: 0.721–0.829), and WHO IV (0.603, 95% CI: 0.507–0.687; [155]Figure
1F). These results indicated that the RBPS had a potential clinical
value, and the gene signature comprised of the six RBP genes could be
used to identify the adverse OS in patients with glioma with various
WHO grades. Additionally, in the CGGA validation cohort, the expression
of the six RBP genes, survival status, and survival time distribution
for patients according to their RBPS risk scores are shown in
[156]Figures 2C,[157]D.
RBPS is an independent predictor of glioma risk and survival outcome
To further evaluate the performance of RBPS as a clinical marker for
risk stratification, its utility was analyzed along with clinical
features for predicting survival and prognosis. First, in the TCGA
cohort, univariate and multivariate Cox regression analyses were
performed for various clinical features, including age, sex, WHO grade,
and histopathology, along with the RBPS. In univariate analysis, age (p
< 0.001), WHO grade (p < 0.001), histopathology (p < 0.001), and RBPS
(p < 0.001) were important predictors for adverse OS
([158]Supplementary Figure 3G). Subsequently, multivariate Cox
regression analysis showed that age (p < 0.001), grade (p < 0.001), and
RBPS (p < 0.001) were independent risk factors in predicting adverse OS
in patients with glioma ([159]Supplementary Figure 3H). In the CGGA
cohort, univariate and multivariate cox regression analyses were
conducted. Apart from age, WHO grade, and histopathology, the clinical
features included radiotherapy, chemotherapy, IDH mutation, 1p19q
codeletion, and methylation status of the MGMT gene promoter region
(MGMTp). The results showed that WHO classification (p < 0.001), age (p
= 0.012), and RBPS (p < 0.001) remained independent risk factors in
predicting adverse OS ([160]Supplementary Figures 3I,[161]J). These
results verified that RBPS based on these six RBP genes was reliable in
predicting OS and could be used as an independent predictor of survival
outcomes in patients with glioma.
The [162]GSE72951 dataset included patients with recurrent glioblastoma
only. In this dataset, K-M analysis showed that the median survival
time in the high-RBPS-risk group was longer than that in the
low-RBPS-risk group (p = 0.010, [163]Supplementary Figures 4A,[164]B),
while univariate and multivariate Cox analyses suggested no statistical
correlation between RBPS and survival outcomes ([165]Supplementary
Figures 4C,[166]D). According to statistical significance and
comparison of RBPS risk scores of WHO IV glioma in the three data sets,
it was speculated that the RBPS risk scores of patients with WHO IV
glioma in the [167]GSE72951 dataset were relatively close to each
other, thereby resulting in no statistically significant correlation
between RBPS and survival outcomes ([168]Supplementary Figures
4E–[169]L). In addition, the expressions of protective genes (POLR2F,
DYNC1H1, and SMAD9) for glioma in the [170]GSE72951 dataset increased,
while those of the risk genes (TRIM21, BRCA1, and ERI1) decreased so
that the risk scores of patients in [171]GSE72951 were the lowest among
the three groups, but the median overall survival time was the shortest
among the three datasets. The survival time of patients with WHO IV in
the [172]GSE72951 data set was the shortest, which could be attributed
to the fact that the total survival time in this data set was
calculated from the first recurrence and could be related to the
inclusion of patients with recurrent glioblastoma. In addition, these
patients received CCNU and/or bevacizumab treatment, which may be the
reason why gliomas in the [173]GSE72951 data set have lower RBPS risk
scores. These findings suggested that the RBPS risk score may show
dynamic changes with chemotherapy, which may, in turn, reflect the
therapeutic efficacy.
Construction of a nomogram for predicting the OS for patients with glioma
In order to further improve the predictive ability and applicability of
RBPS in clinical practice, RBPS, and other critical clinical features
(WHO grade, age, radiotherapy, chemotherapy, and 1p19q codeletion) were
used to construct a multivariate Cox regression model and a risk
nomogram for ease of use in clinical settings for predicting survival
probabilities of patients with glioma. The parameters of this model are
listed in [174]Table 2. As shown in [175]Figure 3A, the total score was
calculated based on the sum of scores for each factor. The higher the
total score, the lower the OS rate for 1 year, 3 years, and 5 years. As
shown in the example (the red dot) in the figure, a patient with WHO
grade III and an RBPS risk score of 1 (wherein no radiotherapy, no
chemotherapy, and no 1p19q codeletion all corresponded to 34 points,
WHO grade III corresponded to 70 points, and the RBPS risk score of 1
corresponded to 33 points in the nomogram), the total score
corresponding to all characteristics was 233 points, and the predicted
survival probabilities for 3 years and 5 years based on this total
score were 0.344 and 0.219, respectively. [176]Figure 3B shows the AUC
of the model between 0.74 and 0.85 for predicting the overall survival
rate for 1–5 years. The calibration curve showed that the predicted
values using the model were in good agreement with the actual values
([177]Figure 3C), suggesting a good prediction performance.
Table 2.
Prediction factors for survival in glioma.
Variables Prediction model
β Hazard ratio (95% CI) P value
Grade (III vs. II) 1.074 2.928 (2.506–3.421) <0.001
Grade (IV vs. II) 1.753 5.773 (4.905–6.795) <0.001
Age 0.011 1.011 (1.007–1.014) 0.006
Radiotherapy (yes vs. no) −0.252 0.777 (0.689–0.876) 0.035
Chemotherapy (yes vs. no) −0.357 0.7 (0.623–0.785) 0.002
1p19q codeletion (yes vs. no) −1.043 0.352 (0.3–0.413) <0.001
Risk score 0.062 1.064 (1.048–1.079) <0.001
[178]Open in a new tab
β is the Cox regression coefficient. For grade, radiotherapy, and 1p19q
codeletion, HR represents the average HR over the entire time period.
Figure 3.
[179]Figure 3
[180]Open in a new tab
Construction of a CGGA-based clinical prediction model. (A) The
nomogram for predicting the 3- and 5-year overall survival of patients
with glioma based on the six independent prognostic factors from the
CGGA dataset. (B) Relationship between the AUC values for the
prognostic prediction model and the correspondingly predicted survival
times. (C) The calibration plot shows that the prediction using the
model is in good agreement with the actual situation. (D) Glutathione
metabolism, an amino sugar, and nucleotide sugar metabolism, lysosome,
pyrimidine metabolism, viral myocarditis, base exception repair, and
cytosolic DNA-sensing path were significantly differentially enriched
between the high- and low-risk-score groups in the TCGA dataset. (E)
JAK-STAT signaling, ECM-receptor interaction, cytokine-cytokine
receptor interaction, systematic lupus erythematosus, intestinal immune
network for IgA production, focal adhesion, and small cell lung cancer
pathways were differentially enriched between the high- and low-risk
groups in the CGGA database.
Gene set enrichment analysis for RBPS
GSEA was performed using MSigDB Collection [c2. cp.kegg. v7.1. symbols
(curated)] to identify differentially expressed signaling pathways in
gliomas between high- and low-risk groups of patients with glioma. All
genes were ranked according to their fold changes between the high- and
low-risk groups, following which a GSEA was performed. FDR < 0.05 was
used to filter and select significant enrichment signaling pathways.
The results showed that a high RBPS risk score was related to the
carcinogenesis of glioma, including multiple pathways related to
cellular metabolism, immunity, and proliferation ([181]Figures
3D,[182]E). Furthermore, based on the sharing signaling pathways in
TCGA and CGGA datasets, GSEA showed that the RBPS was associated with
cytokine-cytokine receptor interaction (TCGA: NES = 1.72, size = 264,
FDR = 0.046; CGGA: NES = 1.91; size = 209; FDR = 0.037) and intestinal
immune network for IgA production (TCGA: NES = 1.75, size = 46, FDR =
0.048; CGGA: NES = 1.86; size = 42; FDR = 0.040). Taken together, the
activity of immune, metabolic, and proliferative pathways may be
enhanced, which may be related to the enhanced carcinogenic phenotype
in patients with a high RBPS risk score.
Relationship between RBPS and glioma stemness
To evaluate the relationship between RBPS and tumor stemness of glioma,
the correlation of the RBPS score with DNAss and RNAss was calculated
([183]Figure 4A). In all WHO grade II-IV gliomas, DNAss was positively
correlated with the RBPS score, ERI1, BRCA1, and TRIM21, while
negatively correlated with POLR2F, DYNC1H1, and SMAD9 [Spearman
correlation, Benjamini-Hochberg (BH)-adjusted p < 0.05]. However, RNAss
was negatively correlated with the RBPS score, ERI1, BRCA1, and TRIM21,
while positively correlated with POLR2F, DYNC1H1, and SMAD9 (Spearman,
BH-adjusted p < 0.05). In WHO grade II and III gliomas, the correlation
of RBPS with DNAss and RNAss also showed a similar pattern in the
overall glioma cohort. However, no significant correlation between RBPS
and stemness index was observed in WHO grade IV gliomas, which may be
attributed to their high malignancy and stemness.
Figure 4.
[184]Figure 4
[185]Open in a new tab
(A) Correlation analysis for the expression of the six RBPs and the
RBPS with stemness (RNAss and DNAss), TME (the stromal score, the
immune score, the ESTIMATE score, and tumor purity), and immune
checkpoints (CD274, CD276, CD80, CD86, CTLA4, PDCD1, PDCD1LG2, and
VTCN1). Correlation analysis of RBPS for Grades II, III, and IV glioma,
respectively; red: positive correlation and blue: negative correlation.
Relationship of the expressions of the six RBP genes (POLR2F, DYNC1H1,
SMAD9, TRIM21, BRCA1, and ERI1) and RBPS with (B) infiltration of eight
types of immune cells (B cells, CD8^+ T cells, CD4^+ T cells, NK cells,
monocytes, macrophages, dendritic cells, neutrophils), and (C) immune
subtypes in TCGA.
Correlation between RBPS and tumor microenvironment
GSEA showed that RBPS was associated with immune-related pathways. In
order to evaluate the relationship between RBPS and the immune
microenvironment of glioma, the correlation between RBPS and
immune-related characteristics was analyzed. [186]Figure 4A shows that
RBPS and these six RBP genes were significantly correlated with the
stromal score (Spearman, BH-adjusted p < 0.05), the immune score
(Spearman, BH-adjusted p < 0.05), the ESTIMATE score (Spearman,
BH-adjusted p < 0.05), and tumor purity (Spearman, BH-adjusted p <
0.05), as, also, tumors of all WHO subtypes.
As shown in [187]Figure 4B, significant correlations between RBPS and
individual immune cell types were observed. Specifically, RBPS was
positively correlated with CD8^+ T cells, M1 and M0 macrophages,
activated memory CD4^+ T cells, regulatory T cells, γδ T cells, and
neutrophils (Spearman, BH-adjusted p < 0.05), and negatively correlated
with naive B cells, naive CD4^+ T cells, eosinophils, activated mast
cells, activated NK cells, monocytes, and dendritic cells (Spearman,
BH-adjusted p < 0.05). In addition, the RBPS scores and the expressions
of the six RBP genes were significantly different among the immune
subtypes C1, C3, C4, C5, and C6 (the Kruskal-Wallis test, p < 0.05;
[188]Figure 4C). Among the WHO subtypes of glioma, the expression
differences for RBPS and the six RBP genes among different immune
subtypes were analyzed, and the results are illustrated in
[189]Supplementary Figure 5.
To further elucidate the potential role of RBPS in immunotherapy, the
correlations of RBPS and six RBP genes with common immune checkpoint
molecules were analyzed. The results showed that, for glioma, the RBPS
scores were positively correlated with the expression of immune
checkpoint molecules, PDCD1, CD274, PDCD1LG2, CTLA4, CD86, CD80, CD276,
and FAS (Spearman, BH-adjusted p < 0.05) but negatively correlated with
VTCN1 (Spearman, BH-adjusted p < 0.05; [190]Figure 4C). In WHO grades
II, III, and IV gliomas, RBPS was positively correlated with CD274,
CD276, CD80, CD86, CTLA4, FAS, and PDCD1LG2. Finally, to identify
potential drugs that targeted RBPS, the potential drugs related to the
expression of these six RBP genes were queried in the database, and a
correlation analysis was performed. The top 16 compounds with the
highest correlation with the six RBP genes are shown
([191]Supplementary Figure 5D). As shown, the top 16 predicted
compounds were mainly related to DYNC1H1 and POLR2F.
Radiomics features for RBPS and automatic machine learning prediction model
First, by univariate logistic regression analysis, 180 radiomics
features were selected according to p < 0.05 and included in the
automatic machine learning model ([192]Figure 5A). When splitting the
training and test sets from the whole dataset to reduce the randomness
in selecting patients for high- and low-RBPS risk between training
different models and comparing their performances, the samples of the
two sets were fixed (the random state was set at 42) and standard TPOT
was performed. TPOT was used to calculate the average cross-validation
score (AC) for each model in the training set (each model was trained
100 times/generation) and return the model with the best accuracy in
the test set. Finally, by repeating the TPOT process ten times, ten
independent classificatory models were obtained to predict the risk
stratification according to RBPS. Overall, these 10 models showed good
classification performances in training and test sets, along with high
accuracy (Accuracy, ACC) ([193]Supplementary Tables 2, [194]3). During
the training process, each model showed the following performance in
training and test sets: Model 1 (AC = 0.829, ACC = 0.727), Model 2 (AC
= 0.868, ACC = 0.758), Model 3 (AC = 0.868, ACC = 0.667), Model 4 (AC =
0.829, ACC = 0.697), Model 5 (AC = 0.858, ACC = 0.818), Model 6 (AC =
0.859, ACC = 0.697), Model 7 (AC = 0.858, ACC = 0.697), Model 8 (AC =
0.839, ACC = 0.727), Model 9 (AC = 0.848, ACC = 0.788), Model 10 (AC =
0.829, ACC = 0.727), and 10 Average of models generated based on TPOT
(AC = 0.848, ACC = 0.736). Among them, according to the accuracy in the
test set, Model 5 was selected as it showed the best classification
performance. [195]Figures 5B,[196]C show the average accuracy (AP) and
the area under the curve (AUC) for the 10 models in the test set. The
parameters of Model 5 are as follows: Model [5] = make_pipeline
[binarizer (threshold = 0.3), OneHotEncoder (minimum_fraction = 0.15,
sparse = false, threshold = 10], GradientBoostingClassifier [the
learning_rate = 0.5, max_depth = 8, max_features = 0.3,
min_samples_leaf = 1, min_samples_split = 3, n_estimators = 1t00,
subsample = 0.95)]. In this model, Binarizer and OneHotEncoder were
used to process the radiomics features (see [197]Supplementary Table 4
for detailed parameter descriptions of the other nine models).
Figure 5.
[198]Figure 5
[199]Open in a new tab
(A) The heatmap of the 180 radiomics features between the high- and
low-RBPS-risk-score samples. (B) Receiver operating characteristic
curves and (C) precision-recall curves for 10 models based on the
testing set. AP, average precision; AUC, area under the curve.
The six RBP genes are associated with ischemic stroke, dementia, and aging
In order to examine the potential role of RBPS genes in ischemic
stroke, RNA transcripts from peripheral blood samples of 39 patients
with ischemic stroke and 24 healthy controls were analyzed. The six RBP
genes included in the RBPS could distinguish IS from the healthy
control group ([200]Figure 6A, AUC = 0.950). Differentially expressed
analysis showed that POLR2F, BRCA1, and TRIM21 in this RBPS were
associated with ischemic stroke. Among them, TRIM21 and BRCA1 were
upregulated, while POLR2F was downregulated in IS ([201]Figure 6B).
GSEA showed that the upregulation of TRIM21 was significantly related
to upregulated pathways, including (REACTOME) response to elevated
platelet cytosolic Ca^2+, (REACTOME) cellular response to hypoxia,
(KEGG) complex and coagulation cascades, and (KEGG) focal adhesion
([202]Figure 6C). In BRCA1-upregulated samples, (REACTOME) oncogenic
MAPK signaling, (REACTOME) platelet activation signaling and
aggregation, (WP) angiogenesis, and the (PID) VEGFR1 and VEGFR2 pathway
were upregulated, while the (REACTOME) respiratory electron transport
pathway was downregulated significantly ([203]Figure 6D). In
POLR2F-upregulated samples, (REACTOME) response to elevated platelet
cytosolic Ca^2+, (KEGG) complement and coagulation cascades, and (KEGG)
focal adhesion pathways were downregulated, while (REACTOME) cellular
response to hypoxia was upregulated ([204]Figure 6E). In IS, the
upregulation of TRIM21 was related to platelet function activation,
increased coagulation, and response to hypoxia. Upregulation of BRCA1
was related to tumor progression, platelet activation, and
angiogenesis. The downregulation of POLR2F was accompanied by an
upregulation of platelet reaction and coagulation, and downregulation
of hypoxia-related response.
Figure 6.
[205]Figure 6
[206]Open in a new tab
Diagnostic efficacy of the six RBP genes for IS. (A) Diagnostic
efficacy of the six RBP genes for IS using blood samples (AUC = 0.950,
95% CI: 0.902–0.994). (B) The volcano plot shows the DERBPs associated
with IS. GSEA for (C) TRIM21, (D) BRCA1, and (E) POLR2F in IS. (F) The
volcano plot shows SMAD9 is associated with dementia. (G) The
correlation heatmap shows that aging correlates with POLR2F and ERI1
expression.
Further analyses revealed that SMAD9 in the RBPS was associated with
the Alzheimer's disease onset ([207]Figure 6F). In addition, aging was
positively associated with ERI1 expression and negatively with POLR2F
expression ([208]Figure 6G).
Cell clustering shows the highest proportion of microglia and astrocytes
First, quality control for single-cell RNA-seq (scRNA-seq) data
([209]Supplementary Figures 6A–[210]G) was performed according to cell
characteristic distributions and preset quality filtering conditions.
UMAP showed no significant batch effects for cells between the two
groups after integration analysis ([211]Supplementary Figure 6H). By
clustering, 16 cell clusters were finally identified, and 11 cell types
were annotated ([212]Figures 7A–[213]C; [214]Supplementary Figures 6I,
[215]7). Among them, microglia and astrocytes had the highest
proportion ([216]Figure 7D).
Figure 7.
[217]Figure 7
[218]Open in a new tab
Cell clustering and annotation for the mouse cerebral cortex. (A)
Single-cell analysis shows 16 cell clusters in hypoxic and normoxic
conditions. (B) Expression of marker genes for cell annotation between
the cell clusters. (C) Cell annotation results show 11 cell types. (D)
Comparison of proportions of cells between the normoxia and the hypoxia
groups.
RBPS-related genes associated with pseudotime in microglia
Pseudotime analysis showed three main cell stages of microglia at
normal- and low-oxygen concentrations ([219]Figures 8A,[220]B,
[221]Supplementary Figure 8). In microglia, Sox4 and Tcf7l2, which
regulated Brca1, and Irf5, which regulated Trim21, were significantly
related to the pseudotime of these cells ([222]Figures 8C,[223]D). The
expression of transcription factors and RBPs in microglia during the
transition from hypoxia to normal oxygen concentrations ([224]Figures
8A,[225]E) were observed using the pseudotime distribution plot. Sox4,
Irf5, and Tcf7l2 were downregulated at the early stages of pseudotime
but upregulated at normal-oxygen concentrations. These results
suggested that Sox4 and Tcf7l2, which regulated Brca1, and Irf5, which
regulated Trim21, may change under hypoxia, thus participating in
cellular phenotypic changes.
Figure 8.
[226]Figure 8
[227]Open in a new tab
Results of pseudotime analysis for microglia. (A) The pseudotime
distribution plot of microglia. (B) The RNA velocity plot; the longer
is the arrow, the stronger is the transcriptional activity. (C) The
Sankey diagram shows the RBPS-related transcription factors (E2f4,
Elk3, Irf1, Irf5, Nr3c1, Sox11, Sox4, Sox8, Taf7, Tcf712, and Zbtb7a),
which are associated with the pseudotime. (D,E) Changes in the
expression of these RBPS-related transcription factors with changes in
pseudotime.
RBPS-related genes associated with pseudotime in astrocytes
Pseudotime analysis showed that astrocytes went through eight major
cell stages in normal-oxygen concentration and hypoxia conditions
([228]Figures 9A,[229]B; [230]Supplementary Figure 9). According to
SCENIC analysis, a regulatory relationship between Tcf7l2 and Brca1 was
observed ([231]Figure 9C). In astrocytes, Tcf7l2 was an important
pseudotime-related gene ([232]Figure 9D). In addition, from the
pseudotime distribution map, astrocytes were found in the early,
middle, and late pseudotime stages in the normal oxygen concentration
group, while, in the hypoxia group, astrocytes were dominant in the
middle stage and lesser in the early and late stages; Tcf7l2 increased
in the early stages and decreased toward the later stage ([233]Figures
9A,[234]E). These results suggested that (Brca1-related) Tcf712 may
play a role in the transition to a hypoxic environment.
Figure 9.
[235]Figure 9
[236]Open in a new tab
Results of pseudotime analysis for astrocytes. (A) The pseudotime
distribution plot of astrocytes. (B) The RNA velocity plot, wherein the
longer the arrow, the stronger the transcriptional activity. (C) The
Sankey diagram shows the RBPS-related transcription factors (Bclaf1,
E2f4, Rad21, Sap30, Six1, Sox8, Tcf712, and Zbtb7a), which are
associated with the pseudotime. (D,E) Changes in the expression of
these RBPS-related transcription factors with changes in pseudotime.
RBPS-related genes associated with pseudotime in pericytes
Pseudotime analysis showed that pericytes went through six cell stages
([237]Figures 10A,[238]B; [239]Supplementary Figure 10) in normal
oxygen concentration and hypoxia conditions. According to the
prediction of the gene regulatory network by SCENIC analysis, a
regulatory relationship between Taf7 and Trim21 was obtained
([240]Figure 10C). In astrocytes, Taf7 was an important
pseudotime-related gene ([241]Figure 10D). In addition, as shown in the
pseudotime distribution plot, pericytes were obviously stagnating in
the early pseudotime stages under hypoxia ([242]Figure 10A). Taf7
increased at an early stage of pseudotime but decreased toward the end
stage; pericytes under hypoxia were mostly dominant in the early stage
of pseudotime ([243]Figure 10E). These results suggested that Taf7 may
play an important role in cell-state transition between hypoxia and
normal oxygen conditions.
Figure 10.
[244]Figure 10
[245]Open in a new tab
Results of pseudotime analysis for pericytes. (A) The pseudotime
distribution plot of pericytes. (B) The RNA velocity plot, wherein the
longer the arrow, the stronger is the transcriptional activity. (C) The
Sankey diagram shows the RBPS-related transcription factors (Bclaf1,
Nr3c1, Taf7), which are associated with the pseudotime. (D,E) Changes
in the expression of these RBPS-related transcription factors with
changes in pseudotime.
Pseudotime-related regulons
SCENIC analysis was performed for single-cell data to identify
important regulons of each cell subtype. In the SCENIC analysis flow,
UMAP and tSNE showed single-cell dimension reduction results and the
distributions for each cell type ([246]Supplementary Figure 11).
[247]Figures 11A–[248]C show the distribution of microglia, astrocytes,
and pericytes under normal- and low-oxygen conditions. Irf5 was an
essential and specific regulon of microglia in the normal oxygen and
hypoxia concentration groups ([249]Figures 11D,[250]G,[251]J). The RSS
and rank of Tcf7l2 were related to oxygen concentration. The rank of
Tcf7l2 in the normoxia group was higher than that in the hypoxic group
([252]Figures 11E,[253]H,[254]K). The Taf7 regulon played a regulatory
role in many other cells apart from pericytes ([255]Figures
11F,[256]I,[257]L).
Figure 11.
[258]Figure 11
[259]Open in a new tab
Major regulons in microglia, astrocytes, and pericytes. (A–C) tSNE
shows the distribution of microglia, astrocytes, and pericytes. (D–F)
Major gene regulatory networks in the three types of cells under
normoxia condition, wherein red dots represent the gene regulatory
networks regulated by corresponding transcription factor related to the
RBPS. (G–I) Major gene regulatory networks in the three types of cells
under hypoxia condition. (J–L) Distribution of regulons associated with
the RBPS in cells.
Discussion
The RBP family of proteins plays an important regulatory role in glioma
and IS (Shao et al., [260]2013; Zhou et al., [261]2014; Barbagallo et
al., [262]2018; Lan et al., [263]2020; Si et al., [264]2020; Zhang et
al., [265]2020; Sharma et al., [266]2021); however, there is a lack of
a systematic analysis of the role of RBP in both these diseases.
Herein, we describe a set of previously unreported six RBP genes that
can be used to predict the prognosis of glioma and diagnostic
classification for IS. In particular, we found that the RBPS was
associated with tumor immunosuppression in glioma and hypoxia and
coagulation in IS. In addition, automatic machine learning was used to
predict the risk stratification based on RBPS in glioma. In this RBPS,
SMAD9 was found to be associated with dementia; POLR2F and ERI1 were
identified to be associated with aging. In view of hypoxia as the basis
of common models for studying glioma and IS, the expressions of these
six RBP genes in microglia, astrocytes, and pericytes, along with their
gene regulatory networks, were analyzed using single-cell data from the
mouse cerebral cortex. The six RBP genes and the transcription factors
in their gene regulatory networks were analyzed using pseudotime
analyses between normal oxygen and hypoxia conditions. Irf5/Trim21 and
Tcf712/Brca1 in microglia, Tcf712/Brca1 in astrocytes, and Taf7/Trim21
in pericytes were identified as RBPS-related genes that were regulated
in response to hypoxia. These new findings indicated that RBPs,
post-transcriptional regulators, are essential regulatory molecules
involved in the underlying common pathways in the development of glioma
and IS.
Significance of identification of molecular markers for glioma
Glioma is the most common primary intracranial tumor with high
mortality, among which glioblastoma is the most malignant type (Liu et
al., [267]2020). According to molecular genetic characteristics, some
important glioma subtypes, including IDH mutation, TERT promoter, and
1p/19q codeletion, improve the therapeutic efficacy for glioma (Wang et
al., [268]2020). It is worth trying to identify biomarkers that are
robust and can guide the treatment and predict a prognosis so as to
stratify the patients according to the risk and help choose appropriate
treatment methods.
Role of RBP in tumors
Previous studies have shown that RBP plays a vital role in tumor
progression. For example, in the progression of HCC, global changes in
RBP are more evident than those of transcription factors (Dang et al.,
[269]2017). In immunity, RBP CAPRIN1 promotes innate immunity mediated
by IFN-γ-STAT1 by stabilizing the Stat1 mRNA (Xu H. et al., [270]2019).
In addition, some studies suggest that the genetic system of RBP
dysfunction can provide methods for describing different immunological
conditions (Kafasla et al., [271]2014). In AML, the effects of RBM39
deletion on splicing further lead to preferential lethality for AML
with spliceosome mutations, which provides a strategy for the treatment
of those carrying RBP-splicing mutations (Wang E. et al., [272]2019;
Villanueva et al., [273]2020). In glioma, although some studies report
several RBPs related to a poor prognosis of these patients (Boucas et
al., [274]2015; Bhargava et al., [275]2017; Barbagallo et al.,
[276]2018; Velasco et al., [277]2019; Wang J. et al., [278]2019; Lan et
al., [279]2020; Wang et al., [280]2020), their potential clinical
application, including for an individual prognostic risk assessment,
lacks systematic evaluation. In the research on RBPs, some new
technologies have been developed to enrich and extract RBPs and their
homologous RNAs, such as the orthogonal organic phase separation (OOPS)
(Villanueva et al., [281]2020), which is a fast, efficient, and
reproducible method to purify cross-linked RNA-protein complexes in an
unbiased manner, thus making it more efficient for identifying and
studying new RBPs. Taken together, we first developed a risk
stratification gene signature based on RBP gene expression profiles and
an automatic machine learning prediction model based on radiomics for
individualized risk assessment of patients with glioma. Below, we
discuss the roles of these core RBPs in glioma genesis.
Identification of RBPS and the role of the six RBP genes in glioma
First, six prognostic-related RBP genes (POLR2F, DYNC1H1, SMAD9,
TRIM21, BRCA1, and ERI1) were obtained from the TCGA-glioma dataset.
Based on these six RBP genes, a 6-RBP gene signature (RBPS) with risk
stratification characteristics was constructed. Among them, POLR2F,
DYNC1H1, and SMAD9 in tumor tissues of patients with glioma were
downregulated as compared to normal tissues adjacent to cancer, while
TRIM21, BRCA1, and ERI1 were upregulated. In literature, only BRCA1,
TRIM21, and POLR2F have been implicated in the progression of glioma
(Rasmussen et al., [282]2016; Yang et al., [283]2020; Zhao et al.,
[284]2020). Breast cancer susceptibility gene (BRCA) mutations,
including BRCA1, are found in several tumors (Sun et al., [285]2020).
Umphlett et al. ([286]2020) reported a case of a patient with GBM with
extensive metastases, whereby BRCA1 (p.I571T) was considered the
possible driving mutation. Through bioinformatics analyses based on the
[287]GSE53733 dataset, Yang et al. ([288]2020) report that POLR2F is
one of the four potential key genes that affect the OS in GBM. Higher
levels of TRIM21 expression are associated with a poor prognosis of
glioma and promote proliferation, drug resistance, and migration of
glioma cells (Zhao et al., [289]2020). SMAD9 mutations have been
reported in the progression of gastrointestinal ganglioma. In addition,
a low expression of SMAD9 is related to a poor OS in lung
adenocarcinoma (Ngeow et al., [290]2015; Zhai et al., [291]2021). The
microtubule motor protein encoded by DYNC1H1 is involved in many
cellular processes, such as mitosis and intracellular transport.
DYNC1H1 mutations have been implicated in nervous system diseases
(Hoang et al., [292]2017) and pancreatic cancer (Furukawa et al.,
[293]2011), and these mutations are consistent with a high immune
activity of tumor mutation load in various cancer types (Bai et al.,
[294]2020). In addition, DYC1H1 is upregulated in gastric cancer (Gong
et al., [295]2019) and downregulated in primary gallbladder carcinoma
(Huang et al., [296]2014). In mice, Eri1 is a histone mRNA-related
protein involved in RNA metabolism pathways and various cellular
processes regulated by RNA (Thomas et al., [297]2014). Declercq et al.
([298]2020) show that the exogenous nuclease, ERI1, interacts with PB2,
PB1, and NP components of the viral ribonucleoprotein, thus promoting
viral transcription. Previous studies have reported that gene copy
number variations in glioma may lead to changes in RBP gene expression
(Bhargava et al., [299]2017), which was also observed in this study. In
addition, in tumor stemness, RBPS was positively correlated with DNAss
but negatively correlated with RNAss. For results of RBP genes and
tumor stemness, we speculated that, due to the characteristics of
post-transcriptional regulation of RBPs, the correlations of DNAss and
RNAss with RBP would be different, and the underlying mechanism needs
to be elucidated in the future.
A high risk of RBPS in glioma is related to immunosuppression
GSEA showed differences in immune-related functional pathways between
high- and low-RBPS-risk-score groups. By evaluating the relationship
between RBPS and immune-related characteristics, it was possible to
improve the understanding of the anti-tumor immune intervention and
highlight feasible immunotherapeutic strategies. Therefore, the
associations of RBPS with the tumor microenvironment, immune subtypes,
immune cell types, and immune checkpoint molecules were further
analyzed. Xu et al. ([300]2021) report that higher stromal and immune
scores predict a poor prognosis in patients with LGG. In LGG and GBM of
this study, gliomas with higher RBPS risk are related to higher immune
and stromal scores, thus indicating that the RBPS index was related to
immune responses in gliomas. A previous study reports that macrophage
infiltration indicates a worse OS in GBM (Iglesia et al., [301]2016).
Differentiated GBM cells promote GSC-dependent tumor progression by
enhancing macrophage infiltration into tumor tissues (Uneda et al.,
[302]2021). RBPS also showed a positive correlation with the proportion
of infiltrated macrophages in the tumor, which indicated that RBPS may
play a potential role in the involvement of macrophage infiltration in
the development of glioma. Conventional type-1 dendritic cells (cDC1)
play an important role in immunotherapy-mediated reactivation of
tumor-specific CD8^+ T cells to promote tumor regression (Liang et al.,
[303]2021). In this study, RBPS was negatively correlated with
dendritic cell infiltration and positively correlated with CD8^+ T
cells, which implied that, in gliomas with a high RBPS risk, a complete
CD8^+ T cell reactivation for immunotherapy may be difficult due to the
lack of dendritic cells, thus making the anti-tumor effects difficult
to be achieved. Due to several reasons, including inherent challenges
in drug application, a unique immune environment of the brain, and
heterogeneity between and within tumors, immune checkpoint blockade
therapy has not been effective for GBM (Khasraw et al., [304]2020). A
comprehensive understanding of the unique tumor microenvironment of the
brain is important for glioma immunotherapy with immune checkpoint
blockade (Qi et al., [305]2020). In this study, RBPS was positively
correlated with the expressions of CD274, CD276, CD80, CD86, CTLA4,
FAS, PD1, and PDL1 in gliomas, indicating that the expressions of
immune checkpoint-related genes increased with a high RBPS risk, thus
leading to a worse prognosis. The relationship between RBPS and immune
checkpoint molecules needs further studies.
Automatic machine learning model predicts RBPS
Generally, the RBPS showed reliable prognostic value for predicting the
OS and immune-related characteristics of glioma and comprised only six
RBP genes, making its clinical translation convenient. Recently, with
the development of computing power, researchers have tried to replace
some expensive molecular detection techniques using MR image-based
artificial intelligence so as to stratify the risk of tumor phenotypes,
screen patients with cancer, and predict their responsiveness to
treatment (Acs et al., [306]2020). Therefore, using MRI-based radiomics
features, we developed an automatic machine learning classification
model to predict the risk of RBPS in glioma, thus making the molecular
signature more convenient and attractive for preoperative evaluation.
Diagnostic performance and the roles of the six RBP genes in IS
The prediction model based on the six RBP genes from blood samples
could also predict the occurrence of IS, suggesting their association
with IS. Among the six RBP genes, TRIM21 and BRCA1 were upregulated in
IS, while POLR2F was downregulated in IS. Functional pathway enrichment
analysis showed that TRIM21 upregulation was related to platelet
activation, enhanced coagulation, and response to hypoxia. Previous
studies have shown that TRIM21 is mainly expressed in hematopoietic
cells, wherein it is induced by IFNs in case of infections and
autoimmune diseases (Sjöstrand et al., [307]2013). Pan et al.
([308]2016) show that TRIM21 modulates redox homeostasis through the
ubiquitination of p62, and TRIM21-deficient cells exhibit enhanced
antioxidant responses and reduced cell death under oxidative stress. In
addition, TRIM21 deficiency induces naive T cells to differentiate into
Th17 and promotes IL-17 expression, along with a stable atherosclerotic
plaques phenotype formation (Brauner et al., [309]2018). In cerebral
ischemia/reperfusion (I/R), BRCA1 overexpression can alleviate or
prevent nerve injury caused by I/R due to reduced production of
reactive oxygen species (ROS) and lipid peroxidation (Xu et al.,
[310]2018). Overexpression of BRCA1 in neural stem cells (NSCs) reduces
apoptosis and oxidative stress after the oxygen-glucose
deprivation/reoxygenation (OGD/R) insert, stimulating their
proliferation, thus improving the therapeutic effects of NSC
transplantation in cases of ischemic stroke (Xu P. et al., [311]2019).
Genome-wide association analysis shows that POLR2F (22q13.1) is
associated with periventricular white matter hyperintensions (PVWMH),
and PVWMH are associated with ischemic stroke (Armstrong et al.,
[312]2020).
Dual action of ROS under hypoxia
In the cerebrovascular unit, hypoxia can induce astrocytes, microglia,
pericytes, and neuronal cells to produce ROS and reactive nitrogen
species (RNS) (Sumbayev and Yasinska, [313]2007; Chen et al.,
[314]2013). ROS and RNS play dual roles in the neurovascular unit,
destroying tissues and macromolecules upon injury (global cerebral
ischemia and reperfusion injury) while promoting cellular
proliferation, tissue repair and regeneration, and angiogenesis in the
recovery stage (acute ischemic stroke and hypoxic tumor core)
(Kalogeris et al., [315]2014).
The role of hypoxic stress in tumor immunity and angiogenesis
In tumors, hypoxic stress plays an important role in tumor progression
and immune escape by controlling angiogenesis, promoting
immunosuppression, and tumor resistance (Noman et al., [316]2015).
Several hypoxia-induced immunosuppressive cells in the hypoxic zones of
solid tumors, such as myeloid-derived suppressor cells,
tumor-associated macrophages (MDSCs), and T-regulatory (Treg) cells,
have been reported (Mantovani et al., [317]2002; Ohta et al.,
[318]2011). Hypoxia increases MDSC-mediated T cell tolerance by
upregulating the tumoral MDSC expression of PD-L1 (Noman et al.,
[319]2014); hypoxia-inducible factor-1 (HIF-1) is the primary regulator
of PD-L1 (Barsoum et al., [320]2014). Hypoxia decreases the expression
of several molecular markers of differentiation and maturation of DCs
in response to lipopolysaccharide and inhibits the stimulating ability
of DCs to activate T cell functions (Mancino et al., [321]2008). In
addition, VEGF produced by human tumors can inhibit the functional
maturation of DCs and promote the escape of tumor cells (Gabrilovich et
al., [322]1996). Hypoxic stress increases the lytic functions of CD8^+
T cells and decreases their proliferation and differentiation (Noman et
al., [323]2015). Hypoxia attracts Treg cells to the tumor bed by
affecting the distribution of cytokines in the tumor microenvironment
and enhancing the immunosuppressive functions of Treg cells (Noman et
al., [324]2015). For cancer stem cells (CSCs), hypoxia and HIFs are
considered to induce tumor cells to dedifferentiate into immature
phenotypes and maintain their stemness (Kallergi et al., [325]2009;
Semenza, [326]2012). In this study, RBPS in glioma was related to tumor
immunosuppression. Glucose and amino acid metabolism increased in
gliomas with a high RBPS risk score. This may serve the increased
demand for energy and oxygen of highly proliferating tumor masses. In
addition, among these six RBP genes, the upregulation of TRIM21 and
BRCA1 in IS was related to angiogenesis and responses to hypoxia. The
upregulation of TRIM21 and BRCA1 and the downregulation of POLR2F were
related to platelet activation and increased coagulation, thus
suggesting that an imbalance among these genes may result in a state of
hypercoagulation, which could easily lead to an ischemic cerebral
infarction. Along with aging, POLR2F was downregulated, while ERI1 was
upregulated. In IS, downregulated POLR2F was associated with
downregulation of pathways in response to hypoxic responses, implying
that POLR2F may be associated with aging-related hypoxic stress.
Astrocytes, microglia, and pericytes are important cells that maintain
brain homeostasis. In order to observe the regulation and potential
mechanism underlying the six RBP genes in the hypoxic environment at
the cellular level, the gene regulatory network in various cell types
was investigated.
Gene regulatory networks in astrocytes
Astrocytes are the most abundant cell types in the central nervous
system. As an integral part of the neuron-glial system, astrocytes
serve as housekeeping functions, including the formation of the
blood-brain barrier (BBB), regulation of cerebral blood flow, repair of
blood vessels (Williamson et al., [327]2021), and the resistance to
oxidative stress (Blanc et al., [328]1998; Ransom and Ransom,
[329]2012). After an ischemic stroke, reactive astrogliosis involving
astrocytes exerts harmful and beneficial effects on neuronal survival
and nerve recovery (Liu and Chopp, [330]2016; Xu et al., [331]2020).
The upregulation of BRCA1 in IS was related to tumor promotion,
platelet activation, and angiogenesis. On examining the gene regulatory
network in astrocytes, Brca1 was identified in Tcf712 regulon that was
upregulated under hypoxia. These results suggested that Tcf712/Brca1
may play an important role in the response of astrocytes to hypoxic
stress.
Gene regulatory networks and immune responses of microglia
As resident macrophages in the central nervous system, microglia are
the first immune cells that perceive and respond immediately during
cerebral ischemia (Lambertsen et al., [332]2019). During a stroke, with
the dynamic changes in pathology, microglia undergo polarization
(Tsuyama et al., [333]2018). According to the phenotypic changes in
microglia, they can be roughly classified into pro-inflammatory (M1) or
anti-inflammatory (M2) types (Ransohoff, [334]2016). The interferon
regulatory factor (IRF) family of proteins has an important
relationship with microglial polarization after stroke (Zhao et al.,
[335]2017; Al Mamun et al., [336]2018). For instance, IRF4 negatively
regulates inflammation and promotes M2 polarization of macrophages
(Eguchi et al., [337]2013), while IRF5 induces M1 polarization (Paun et
al., [338]2008). Recent studies have shown that the IRF5-IRF4
regulatory axis in microglia regulates neuroinflammation after ischemic
stroke and affects stroke outcomes (Al Mamun et al., [339]2020). IRF5
mediates pro-inflammatory activation of microglia and affects
anti-inflammatory responses (Fan et al., [340]2020; Al Mamun et al.,
[341]2021). We found that Irf5 regulon was an important regulator in
microglia, and Trim21 was a downstream molecule in this gene regulatory
network. The expression of Irf5 was downregulated under hypoxia, which
may be related to the time of experimental conditions, suggesting that
the main microglia types may be changing toward the anti-inflammatory
phenotype after living in 7.5% oxygen concentration for 7 days. The
Tcf712 regulon was another important gene regulatory network identified
in microglia, and Brca1 was a member of this network. The expression of
Tcf712 was also downregulated under hypoxia.
Multiple microvascular regulatory functions of pericytes
Pericytes play an important role in regulating various microvascular
functions, such as angiogenesis (Winkler et al., [342]2011), the
formation and maintenance of the BBB (Armulik et al., [343]2010;
Daneman et al., [344]2010), capillary blood flow regulation (Hall et
al., [345]2014; Korte et al., [346]2022), neuroinflammatory regulation
(Stark et al., [347]2013; Korte et al., [348]2022), glial plaque
formation (Göritz et al., [349]2011), and stem cell characterization
(Özen et al., [350]2014; Nakagomi et al., [351]2015). Pericytes are
important therapeutic targets in stroke, glioma, Alzheimer's disease,
spinal cord injury, and other diseases due to their vital role in the
nervous system diseases (Cheng et al., [352]2018). Variable
permeability of BBB can be observed in the high cell proliferation
regions, which may be related to an increase in the NG2-expressing
pericytes herein (Jackson et al., [353]2017). In addition, hypoxic
regions of tumors recruit activated pericytes through the regulation of
hypoxia-inducible factors (Svensson et al., [354]2015). In acute
ischemic stroke, pericyte HIF-1 can destroy BBB and affect the
prognosis of stroke (Tsao et al., [355]2021). In addition, glioma stem
cells can also differentiate into pericytes, thus supporting BBB
integrity and angiogenesis (Cheng et al., [356]2013; Segura-Collar et
al., [357]2021). Under hypoxia, in vitro, pericytes derived from the
human brain acquire a microglial phenotype and are a new source of
inflammatory cells during cerebral ischemia (Özen et al., [358]2014).
Interestingly, Trim21 (Irf5 and Taf7 regulon) is present in different
gene regulatory networks of microglia and pericytes in response to
hypoxia. This suggests that RBPs, as post-transcriptional regulators,
participate in different regulatory pathways, thus performing various
cellular functions.
Owing to the heterogeneity of pericytes (Armulik et al., [359]2011),
the selection of specific cell markers and the correct identification
of pericytes in “-omics” studies pose a challenge (Cheng et al.,
[360]2018). Using transcriptomics and proteomics, mRNA and protein
expressions in pericytes at different positions of the capillary bed
would be accurately defined.
This study is the first attempt to comprehensively evaluate the role of
RBPs in glioma and IS using computational biology, thus providing a
panoramic map of a panel of genes between the two diseases and a
research paradigm for the study of such scientific issues. Using bulk
RNA-seq and scRNA-seq data, we examined the important roles of a panel
of RBP genes in glioma and IS and identified the relationship between
the two diseases. In this study, a prognostic RBPS consisting of six
RBP genes was identified for glioma. These six RBP genes obtained from
blood samples had a high classificatory performance for diagnosing IS.
RBPS was associated with immunosuppression, enhanced energy metabolism,
and tumor growth in glioma, and hypoxia response, angiogenesis, and
enhanced coagulation in IS. In this RBPS, SMAD9 was found to be
associated with dementia; POLR2F and ERI1 were identified to be
associated with aging. Under hypoxia, Irf5/Trim21 in microglia and
Taf7/Trim21 in pericytes were identified as RBPS-related networks.
There are some limitations to this study. The gene signature was
developed based on large publicly available databases and retrospective
cohorts. However, no independent clinical cohort in local hospitals for
validation was evaluated. In addition, the properties of these RBP
genes need to be verified at cellular levels and using animal models.
With the identification of new RBP molecules, computational biological
analyses need to be updated to identify important molecules in the
occurrence and development of glioma and IS.
Conclusion
In conclusion, we developed a 6-RBP gene signature associated with a
glioma prognosis and an IS diagnosis. In addition, an automatic machine
learning classification model based on radiomics features from MRI was
developed to stratify the RBPS risk. The RBPS was associated with
immunosuppression, energy metabolism, and enhanced tumor growth in
glioma, and hypoxia response, angiogenesis, and increased coagulation
in IS. Upregulation of SMAD9 was associated with dementia, while
downregulation of POLR2F was associated with aging-related hypoxic
stress. The RBPS is expected to serve as a biomarker to study the
common mechanism between glioma and IS. These six RBP gene markers play
a critical role in the association of IS with glioma, as revealed by
our study.
Data availability statement
The original contributions presented in the study are included in the
article/[361]Supplementary material, further inquiries can be directed
to the corresponding author/s.
Author contributions
WL conceptualized the data, involved in formal analysis and
methodology, investigated the study, and wrote the original draft. QW,
YC, NW, QN, CQ, and QW participated in literature review, data
collection, and statistical analysis. YZ conceptualized the study,
supervised the data, and reviewed and edited the manuscript. All the
authors read and approved the final manuscript.
Funding
This study was funded by Provincial Key R&D Program, Science and
Technology Department of Zhejiang Province (Grant No. 2017C03018), Key
Program of Administration of Traditional Chinese Medicine, Zhejiang
Province (Grant No. 2018ZZ015), Nursery Project of the Affiliated
Tai'an City Central Hospital of Qingdao University (Grant No. 2022M
PM06), Shandong Medical and Health Technology Development Fund (Grant
No.202103070325), and 2021 Zhejiang Normal University Interdisciplinary
Advance Research Fund.
Conflict of interest
The authors declare that the research was conducted in the absence of
any commercial or financial relationships that could be construed as a
potential conflict of interest.
Publisher's note
All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
reviewers. Any product that may be evaluated in this article, or claim
that may be made by its manufacturer, is not guaranteed or endorsed by
the publisher.
Acknowledgments