Abstract
Background
Not only glycolysis but also lncRNAs play a significant role in the
growth, proliferation, invasion and metastasis of of ovarian cancer
(OC). However, researches about glycolysis -related lncRNAs (GRLs)
remain unclear in OC. Herein, we first constructed a GRL-based risk
model for patients with OC.
Methods
The processed RNA sequencing (RNA-seq) profiles with
clinicopathological data were downloaded from TCGA and
glycolysis-related genes (GRGs) were obtained from MSigDB. Pearson
correlation coefficient between glycolysis-related genes (GRGs) and
annotated lncRNAs (|r| > 0.4 and p < 0.05) were calculated to identify
GRLs. After screening prognostic GRLs, a risk model based on five GRLs
was constructed using Univariate and Cox regression. The identified
risk model was validated by two validation sets. Further, the
differences in clinicopathology, biological function, hypoxia score,
immune microenvironment, immune checkpoint, immune checkpoint blockade,
chemotherapy drug sensitivity, N6-methyladenosine (m6A) regulators, and
ferroptosis-related genes between risk groups were explored by abundant
algorithms. Finally, we established networks based on co-expression,
ceRNA, cis and trans interaction.
Results
A total of 535 GRLs were gained and 35 GRLs with significant prognostic
value were identified. The prognostic signature containing five GRLs
was constructed and validated and can predict prognosis. The nomogram
proved the accuracy of the model for predicting prognosis. After
computing hypoxia score of each sample by ssGSEA, we found patients
with higher risk scores exhibited higher hypoxia score and high hypoxia
score was a risk factor. It was revealed that a total of 21
microenvironment cells (such as Central memory CD4 T cell, Neutrophil,
Regulatory T cell and so on) and Stromal score had significant
differences between the two groups. Four immune checkpoint genes
(CD274, LAG3, VTCN1, and CD47) showed disparate expression levels in
the two groups. Besides, 16 m6A regulators and 126 ferroptosis-related
genes were expressed higher in the low-risk group. GSEA revealed that
the risk groups were associated with tumor-related pathways. The two
risk groups were confirmed to be sensitive to several chemotherapeutic
agents and patients in the low-risk group were more sensitive to ICB
therapy. The networks based on co-expression, ceRNA, cis and trans
interaction provided insights into the regulatory mechanisms of GRLs.
Conclusions
Our identified and validated risk model based on five GRLs is an
independent prognostic factor for OC patients. Through comprehensive
analyses, findings of our study uncovered potential biomarker and
therapeutic target for the risk model based on the GRLs.
Supplementary Information
The online version contains supplementary material available at
10.1186/s13048-021-00881-2.
Keywords: Ovarian cancer, Glycolysis, lncRNA, Risk model, Immune
Introduction
Ovarian cancer (OC) is a gynecological tumor with high morbidity and
mortality and about 150,000 women die of OC each year [[33]1]. The
occurrence and development of OC is a multi-system, multi-step cellular
biochemical process, which is regulated by a variety of cytokines and
signaling pathways [[34]2]. Due to the lack of typical clinical
symptoms in the early stages of OC, 75% of OC patients are diagnosed at
an advanced stage, and more than 70% of patients relapse after
treatment [[35]3]. Therefore, how to diagnose early, effectively treat
and improve the prognosis of OC patients is an urgent problem to be
solved.
Tumor cells are mainly metabolized by glycolysis regardless of the
presence of oxygen. Large amounts of glucose are consumed with the
production of lactic acid. This phenomenon is called aerobic glycolysis
or Warburg effect [[36]4]. Long non-coding RNA (lncRNA) is defined as a
large class of non- protein-coding, regulatory RNAs with molecules
longer than 200 nucleotides, which play key roles in tumorigenesis and
progression [[37]5, [38]6]. In recent years, more and more studies have
shown that lncRNA plays a key regulatory role in tumor metabolism and
is involved in glucose metabolism pathway [[39]7, [40]8]. For instance,
lncRNA ANRIL up-regulates the expression of glucose transporter
1(GLUT1) and LDHA, thereby increasing glucose uptake and promoting the
progression of nasopharyngeal carcinoma [[41]9]. LINC00092 directly
binds to PFKFB2 to enhance glycolysis and ultimately promote tumor
genesis and development [[42]10]. In bladder cancer, lncRNA UCA1 is
overexpressed and promotes glycolysis by upregulation of hexokinase 2
(HK2), and also promotes aerobic glycolysis [[43]11]. However, lncRNAs
involved in the glycolysis reprogramming of OC remain unclear.
Therefore, in our study, we found five glycolysis-related lncRNAs
(GRLs) with significant prognostic value from TCGA dataset. A
GRL-signature with prognostic value was developed. In addition, we
identified differences in enrichment pathways, immune microenvironment,
immune checkpoints, m6A regulatory factors, and ferroptosis-related
genes between risk groups. The networks based on co-expression, ceRNA,
cis and trans interaction provided insights into the regulatory
mechanisms of GRLs.
Material and methods
Data downloading and pretreatment
We downloaded the clinical data with RNA sequencing profiles of OC
patients from TCGA dataset [[44]12]. The Ensemble expression matrix was
transformed into Gene Symbol expression matrix and compared it with the
position of lncRNA chromosome in GENCODE database to identify lncRNAs
[[45]13]. A total of 274 glycolysis-related genes (GRGs) were obtained
from MSigDB database [[46]14]. We screened differentially expressed
GRGs and annotated lncRNAs using limma package (P < 0.05, |logFC| > 1)
[[47]15]. Pearson correlation coefficients between differential GRGs
and lncRNAs were computed to filtrate GRLs (|r| > 0.4, P < 0.05) using
cor function of R.
Development of the signature
After screening prognostic GRLs through Univariate Cox regression
(P < 0.05) [[48]16], the LASSO Cox regression [[49]17] from glmnet
package of R [[50]18] and 20 times cross-validation analysis was
employed to filtrate optimal combination of GRL markers. A risk score
model for OC patients was constructed based on following formula:
[MATH: Riskscore=∑βlncRNA×ExplncRNA :MATH]
In the risk score (RS) formula, β[lncRNA] meant the regression
coefficient of each lncRNA calculated in the multivariate Cox
regression analysis and Exp[lncRNA] represented the expression value of
each lncRNA in the sample. Whereafter, the RS of each OC patient was
calculated and the calculated median RS was used as the critical value
to further divided the OC patients into high-risk and low-risk groups
(low-risk group 0.4,
P < 0.05) network was constructed and visualized by Cytoscape [[62]30].
The targeted glycolysis-related mRNAs by corresponding miRNAs were
speculated by miRWalk [[63]31]. Further, we synthesized the results of
six commonly used databases (miRWalk, Microt4, miRanda, miRDB, RNA22
and Targetscan) to obtain the miRNA-GRG relationship pair if the
predicted miRNA-GRG relationship pair appeared in ≥5 databases. The
miRNAs targeted by corresponding GRLs of the risk model were speculated
by miranda (v3.3a, Score > =140, Energy<= − 20) [[64]32]. GRLs and GRGs
regulated by the same miRNA with positive co-expression relationship
were defined as ceRNAs mutually. Based on previous literature, we
predicted cis [[65]33] and trans [[66]34] interaction between GRLs and
GRGs.
Moreover, the potential response to immune checkpoint blockade (ICB)
was predicted with TIDE algorithm [[67]35]. We extracted chemotherapy
drugs from GDSC database [[68]36] and evaluated the IC50 level by using
pRRophetic [[69]37].
Statistical analysis
R packages (v4.0.2) and GraphPad Prism (v8.0) were used for statistical
analysis. T test was used for inter-group comparison. Pearson
correlation analysis was conducted to analyze the correlation between
GRGs and lncRNAs. Univariate and multivariate Cox regression analysis
was conducted to analyze the related factors affecting the overall
survival of OC patients. P < 0.05 was considered statistically
significant.
Results
Figure [70]1 exhibited the flowchart we created for our entire study.
Fig. 1.
[71]Fig. 1
[72]Open in a new tab
Flow diagram of our study
Differential and enrichment analysis
A total of 116 differential GRGs (Fig. [73]2A) and 1145 exegetical
lncRNAs (Fig. [74]2B) were identified. In addition, 62 GO BP and 33
KEGG pathways were enriched based on the 116 GRGs
(Additional file [75]1: Table S1). The enrichment pathways were ranked
according to p value, and the top 20 was selected for display (Fig.
[76]2C-D). The result showed that most of these differential GRGs were
enriched in metabolic pathways. In addition, the identified GRGs were
associated with several important biological processes in tumor genesis
and development observably, such as, response to hypoxia, AMPK
signaling pathway, HIF-1 signaling pathway, and so on. This further
proves that glycolysis is closely related to tumor hypoxic
microenvironment.
Fig. 2.
[77]Fig. 2
[78]Open in a new tab
Differential and enrichment analysis. A, B Volcano map of
differentially expressed mRNAs (A) and lncRNAs (B). Red triangle: up
regulated; blue square: down regulated. C GO analysis. The color scale
represented p value and the histogram size indicated count. D KEGG
analysis. The color scale represented p value and the histogram size
indicated count
Construction and validation of the risk model based on GRLs
Univariate Cox regression and K–M survival analysis was performed on
535 GRLs acquired from Pearson correlation analysis (|r| > 0.4,
p < 0.05) to excavate GRLs with significant prognosis (P < 0.05). A
prognostic GRL-signature was constructed according to the LASSO Cox
analysis of 35 prognostic GRLs obtained and a total of five GRLs were
selected to build the risk model (Table [79]1). The results showed that
all the five GRLs were protective factors with HR < 1 (Fig. [80]3A). A
heatmap of the associations between the expression levels of the five
GRLs and clinical features illustrated that the expression of the
expressions of the five GRLs decreased with increasing risk scores
(Fig. [81]3B). The K–M survival curves confirmed that higher expression
of them were associated with better OS of OC patients (Fig. [82]3C-G).
Table 1.
The coefficients of the five GRLs
GRL coef logFC
[83]AC133644.2 −0.149267499 4.214519614
CTD-2396E7.11 −0.250907132 6.5602714
CTD-3065 J16.9 −0.235361863 1.732399016
LINC00240 −0.227486983 1.732645895
TMEM254-AS1 −0.2997883 −1.568258405
[84]Open in a new tab
Fig. 3.
[85]Fig. 3
[86]Open in a new tab
Features of the five GRLs. A Forest plot of the prognostic ability of
the nine optimal GRLs. All the five GRLs were protective factors with
HR < 1. B Heatmap of the associations between the expression levels of
the five GRLs and clinical features. C-G The K-M survival curves of the
five optimal GRLs. TMEM254 − AS1 (C), CTD − 2396E7.11 (D), LINC00240
(E), CTD − 3065 J16.9 (F), and [87]AC133644.2 (G)
The λ selection diagram was shown in Fig. [88]4A-B. The OC patients
were divided into two risk subgroups based on the mean of RSs. The K–M
survival curves revealed that OS of the high-risk group was markedly
lower than that of the low-risk group in TS (Fig. [89]4C), VS1 (Fig.
[90]4D), VS2 (Fig. [91]4E), which indicated the accuracy of the risk
model in predicting survival status. The time-dependent ROC curve
proved that the risk assessment model was relatively stable in
predicting 1-year, 2-year, 3-year and 5-year survival for OC patients
(The AUC for survival was over 0.6, Fig. [92]4F-H).
Fig. 4.
[93]Fig. 4
[94]Open in a new tab
Construction and validation of the risk model. A LASSO Cox analysis. B
λ selection diagram. The two dotted lines indicated two particular
values of λ. The left side was λ[min] and the right side was λ[1se].
The λ[min] was selected to build the model for accuracy in our study.
C-E The K-M survival curves of total (C) and validation sets (D, E).
F-H Time-dependent ROC curve analysis of total (F) and validation sets
(G, H)
The univariate and multivariate Cox regression analysis of clinical
features and the risk model demonstrated that “tumor residual disease”
and risk model was an independent prognostic factor for OC patients
(Fig. [95]5A-B). A nomogram was further constructed based on “tumor
residual disease” and risk model (Fig. [96]5C). The calibration curve
(the closer it was to 45 degrees or the gray lines in the graph, the
better the fitting effect) was drawn to prove the accuracy of the model
(Fig. [97]5D).
Fig. 5.
[98]Fig. 5
[99]Open in a new tab
Nomogram construction. A The Univariate analysis of risk model and
clinical features. B The Multivariate analysis of risk model and
clinical features. C The Nomogram model based on risk model and
clinical features. D The calibration plots of the nomogram. The closer
it was to 45 degrees or the gray lines in the graph, the better the
fitting effect
In conclusion, our risk model was a stable, independent prognostic
factor for OC.
Functional pathways of the risk groups
Functional pathway enrichment analysis based on GSVA algorithm showed
that a total of 66 pathways exhibited significant differences between
the two risk subgroups (Additional file [100]2: Table S2). The KEGG
pathways were ranked according to the p value, and the top ten were
selected for display (Fig. [101]6A). According to GSEA enrichment
analysis, four pathways were enriched in the high-risk group (Fig.
[102]6B), and six pathways were enriched in the low-risk group (Fig.
[103]6C).
Fig. 6.
[104]Fig. 6
[105]Open in a new tab
Differences in functional pathway between the risk groups. A Top10 KEGG
Pathway GSVA enrichment score heat map. B-C The GSEA of KEGG pathway in
the two risk groups. Significant enrichment in the high-risk group (B);
Significant enrichment in the low-risk group (C)
Hypoxia score analysis
Considering that the aerobic glycolysis of tumor is closely related to
its hypoxic microenvironment, the hypoxia enrichment score of each
sample was calculated. Interestingly, we found that patients in the
high-risk group had a higher hypoxia score (Fig. [106]7A). According to
the median of hypoxia scores, the OC patients were divided into two
subgroups. The K–M survival curves revealed that OS of the patients
with high hypoxia score was markedly lower (Fig. [107]7B), indicating
that high hypoxia score and high-risk score were both risk factors
(Fig. [108]7C). Between the two risk groups, a total of 76 differential
hypoxia-related genes were received (Additional file [109]3: Table S3)
and the top 20 were displayed in Fig. [110]7D.
Fig. 7.
[111]Fig. 7
[112]Open in a new tab
Hypoxia score analysis. A Hypoxia scores between high and low-risk
groups. B The K–M survival curves of OC patients with high or low
hypoxia score. C The K–M survival curves of four subgroups based on
risk score and hypoxia score. D Top20 hypoxia-related gene expression
distribution box diagram of the difference multiple between the high
and low risk groups. *p < 0.05, **p < 0.01, ***p < 0.001,
****p < 0.0001
Immunity microenvironment and checkpoint analyses
We assessed immune status by applying five algorithms mentioned in the
Methods section, which was shown after merging the five algorithms in
the heat map (Additional file [113]4: Figure S1). Further, Wilcoxon was
utilized to compare the significance of each cell between the two
groups. Results showed that a total of 21 microenvironment cells and
Stromal score emerged remarkable differences (Fig. [114]8A).
Fig. 8.
[115]Fig. 8
[116]Open in a new tab
Immune and gene analysis. A Heatmap of immune microenvironment revealed
that a total of 21 immune cells and stromal score had significant
differences between the two risk groups. B Expression of seven immune
checkpoint genes between high and low-risk group. CD274, LAG3, VTCN1,
and CD47 had a lower expression in the high-risk group. C TIDE scores
in the low-risk group were lower than those in the high-risk group. D
The expression of 20 m6A regulators between high and low-risk groups. E
The expression of top20 ferroptosis-related genes between high and
low-risk groups. Data are shown as means ± S.D. ns: not significant,
*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001
Immunosuppressive checkpoint inhibitors play a biological role by
inhibiting the immunosuppressive signal pathway in the immune system.
In view of this, in order to further explore the clinical application
of the risk model, we compared the difference in seven checkpoint genes
between the two risk groups. The expression distribution box diagram of
the seven immune checkpoint genes (CD274, CD47, HAVCR2, LAG3, SIRPA,
TNFRSF4, and VTCN1) between the two risk groups was shown in Fig.
[117]8B. The results showed that CD274, LAG3, VTCN1, and CD47 had a
lower expression in the high-risk group. The TIDE score was correlated
closely with response to immune checkpoint blockade (ICB). In Fig.
[118]8C, OC patients in low-risk group exhibited lower TIDE scores than
those in high-risk group, indicating that OC patients with lower RSs
were more sensitive to ICB therapy.
m6A and ferroptosis analyses
The expression levels of m6A regulators and FRGs between the two risk
groups were also compared. A total of 20 m6A regulators were matched,
and as can be seen, except for FTO, IGF2BP2, WTAP, and ZC3H13, the
expression levels of the remaining 16 m6A regulators were significantly
higher in the low-risk group (Fig. [119]8D). A total of 126 FRGs were
matched and showed significant differences between the high and
low-risk groups (Additional file [120]5: Table S4). The top20 genes
ranked by the difference multiple were shown in Fig. [121]8E. It can be
seen that all the genes had low expression in the high-risk group.
Sensitivity of chemotherapy drug
In light of the significance of chemotherapy in the treatment of OC, we
quantified the response ability of OC patients with different risk
scores to 137 chemotherapeutic drugs. We compared IC50 values for nine
commonly used chemotherapeutic agents in two risk groups (Fig. [122]9).
A lower IC50 value indicated that this group of patients was more
sensitive to the drug. Our data showed that the IC50 levels of
Rucaparib (Fig. [123]9A) were significantly higher in the low-risk
group than that in high-risk group. Inversely, the IC50 levels of
Paclitaxel (Fig. [124]9B), Gemcitabine (Fig. [125]9C), Veliparib (Fig.
[126]9D), Vinblastine (Fig. [127]9E), and Vinorelbine (Fig. [128]9F)
were significantly lower in low-risk group than that in high-risk
group, indicating that the OC patients in the low-risk group were more
sensitive to these drugs. However, the sensitivity of the two risk
groups to Bleomycin (Fig. [129]9G), Cisplatin (Fig. [130]9H), and
Docetaxel (Fig. [131]9I) did not reach significant difference.
Fig. 9.
[132]Fig. 9
[133]Open in a new tab
Sensitivity of chemotherapy drugs. A-I Difference in the estimated IC50
levels of Rucaparib (A), Paclitaxel (B), Gemcitabine (C), Veliparib
(D), Vinblastine (E), Vinorelbine (F), Bleomycin (G), Cisplatin (H),
and Docetaxel (I). Data are shown as means ± S.D. ns: not significant,
*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001
Network analyses
In order to explore the relevant regulatory mechanism of this risk
prediction model, we established networks based on co-expression,
ceRNA, cis and trans interaction. A total of 3524 GRG-GRL co-expression
relation pairs were obtained ((|r| > 0.4, p < 0.05)
(Additional file [134]6: Table S5). Only 35 GRLs with significant
prognosis and the corresponding co-expression pair network were
selected to fabricate the GRG-GRL co-expression network
(Fig. [135]10A). A total of 59 GRL-miRNA-GRG relationship pairs were
obtained, including four GRLs, 48 miRNAs, and nine GRGs
(Additional file [136]7: Table S6, Fig. [137]10B). The nine GRGs
included: COL5A1, ELF3, ENO3, NT5E, PGP, PHKA2, SLC25A10, TGFBI, and
VCAN. The networks based on the cis and trans interaction were
displayed in Fig. [138]10C-D. Interestingly, we found that p53 may
regulate GRLs and GRGs through trans interactions. The regulatory
relationships revealed by these networks may provide a direction for
exploring the molecular mechanism of the GRLs.
Fig. 10.
[139]Fig. 10
[140]Open in a new tab
Network analyses. A GRG-GRL co-expression network. Ellipses: GRGs;
rhombus: GRLs. B ceRNA network. Ellipses: GRGs; rhombus: GRLs;
arrowhead: miRNAs. C lncRNA-nearby mRNA interaction networks. Rhombus:
GRLs; ellipses: nearby mRNAs. D Trans interaction for lncRNA-TF-mRNA
relationship pairs. Rhombus: GRLs; ellipses: mRNAs; triangle: TFs
Discussion
The fate of tumor cells is directly related to their energy metabolism
[[141]38]. Tumor cells prefer glycolysis as an inefficient metabolic
mode, which provides new ideas and methods for clinical treatment of
tumors [[142]39]. The reasons are as follows [[143]40]: glycolysis can
provide the energy needed for tumor cell proliferation; It can maintain
a low pH tumor microenvironment, which is conducive to tumor cell
proliferation, drug resistance, invasion and metastasis; A large number
of nucleic acid precursors can be produced in preparation for
proliferation. For ovarian cancer and other tumors with high
proliferation, invasion, metastasis and chemotherapy resistance, it is
of great significance to explore the regulation of its glycolytic
pathway. The aim of studying the glycolysis pathway of OC is to develop
ideal targeted drugs. However, the mechanism of action of some drugs
targeting the glycolytic pathway of OC is still not clear, so the
in-depth study of their molecular mechanism is still of great
significance.
As is known to all that lncRNAs have been proved to play an important
role in the occurrence and development of tumors. In recent years,
lncRNAs have been reported to regulate the energy metabolism of tumors
and thus affect the malignant behavior of tumors, which also partially
reveals the molecular mechanism of glycolysis reprogramming. For
example, lncRNA AGPG increases the stability of PFKFB3 by inhibiting
ubiquitination at Lys302 and subsequent proteasomal-dependent
degradation of PFKFB3 and activates glycolytic flux, causing metabolic
reprogramming in esophageal cancer cells [[144]41]. In addition, PFKFB3
can also be phosphorylated by lncRNA YIYA, increasing the conversion of
fructuce-6-phosphate to fructuce-2, 6-Bisphosphate, and promoting the
reprogramming and growth of glucose metabolism in breast cancer
[[145]42]. However, researches of GRLs are still scarce in OC.
In order to verify the importance of glycolysis-related lncRNAs (GRLs)
in ovarian cancer progress, GRL-related prognostic and diagnostic model
were developed. The gene expression level of 535 GRLs were in
investigated in OC and normal tissues. The significance of these GRLs
related to survival rates were then studied and 35 GRLs were discovered
significantly prognostic. In our study, we identified and validated a
signature containing five GRLs with prognostic value. A total of 21
microenvironment cells, four immune checkpoint genes (CD274, LAG3,
VTCN1, and CD47), 16 m6A regulators, and 126 FRGs showed different
levels between the two groups. It has been reported that the hypoxic
and acidic microenvironment induced by tumor glycolysis can cause
metabolization-mediated T cell dysfunction, which may be one of the
mechanisms of tumor cell metabolic reprogramming mediated immune escape
[[146]43, [147]44]. It has also been found glycolysis of tumors can
induce tumor immunosuppression and immune escape [[148]45]. Therefore,
tumor immunotherapy strategies based on metabolic regulation can
improve the effectiveness of immunotherapy [[149]45]. Many studies have
found that m6A regulatory factors can regulate the expression of
enzymes related to the glucose metabolism pathway, thus affecting the
glycolysis of tumors [[150]46–[151]48]. All these provide a reference
for us to study the specific mechanism of glycolysis in tumor. The
occurrence and development of malignant tumors is sophisticated and we
hope to explore the molecular mechanism of glycolysis (via m6A
modification, ferroptosis or immune) to promote the efficacy of
immunotherapy with further research.
Our study still has some limitations. Firstly, due to the limited
number of OC samples that can annotate lncRNA expression, more patients
with homologous information were needed to incorporate into study and
prove the credibility of our study. Secondly, we explored the functions
of these five lncRNAs only through bioinformatics analysis, and
experimental data were needed to support these conclusions. Despite
these limitations, our study used two validation sets, ROC, and
nomogram to demonstrate the effectiveness of the risk model for
prognostic prediction.
Conclusions
In summary, our identified and validated risk model based on five
glycolysis -related lncRNAs is an independent prognostic factor for OC
patients. Through comprehensive analyses, the GRL-model provides
insights into clinical applications for OC.
Supplementary Information
[152]13048_2021_881_MOESM1_ESM.xlsx^ (21.7KB, xlsx)
Additional file 1: Table S1. The enriched 62 GO BP and 33 KEGG pathways
based on 116 GRGs.
[153]13048_2021_881_MOESM2_ESM.xlsx^ (15.8KB, xlsx)
Additional file 2: Table S2. A total of 66 pathways exhibited
significant differences between the two risk subgroups.
[154]13048_2021_881_MOESM3_ESM.xlsx^ (16KB, xlsx)
Additional file 3: Table S3. A total of 76 differential hypoxia-related
genes were received.
[155]13048_2021_881_MOESM4_ESM.pdf^ (317.4KB, pdf)
Additional file 4: Figure S1. Immunity microenvironment analysis.
[156]13048_2021_881_MOESM5_ESM.xlsx^ (20.1KB, xlsx)
Additional file 5: Table S4. A total of 126 FRGs were matched and
showed significant differences between the high and low-risk groups.
[157]13048_2021_881_MOESM6_ESM.xlsx^ (166.3KB, xlsx)
Additional file 6: Table S5. 3524 GRG-GRL co-expression relation pairs.
[158]Additional file 7: Table S6. ceRNA.^ (10.9KB, xlsx)
Acknowledgments