Abstract
Purpose
Clear cell renal cell carcinoma (ccRCC) is one of the most common
cancers with high mortality worldwide. However, biomarkers for
predicting prognosis in ccRCC are limited. In this study, we attempted
to identify potential prognostic biomarkers of ccRCC.
Methods
Clinical information and the preprocessed ccRCC mature miRNA expression
profiles in The Cancer Genome Atlas database were downloaded from UCSC
Xena. The miRNAs differentially expressed between ccRCCs and matched
normal tissues were analyzed using the “limma” package. A miRNA-based
signature was constructed using the multivariate Cox regression model
with prognosis index (PI) formula. Patients with ccRCC were divided
into low-risk and high-risk subgroups according to median PI. The
survival times were compared between the two groups using Kaplan–Meier
analysis with log-rank test. The training set was used to construct a
miRNA-based signature for predicting prognosis. The test set was used
to verify the signature. Target gene prediction and functional
enrichment analysis of the four miRNAs were performed using miRNet.
Results
We identified four miRNAs, miRNA-21-5p, miRNA-9-5p, miR-149-5p, and
miRNA-30b-5p, as independent prognostic indicators. Next, we used these
four miRNAs to construct a four-miRNA PI for each patient. Results
revealed that patients in the high-risk group (n=119) had significantly
shorter survival time than those in the low-risk group (n=118)
(high-risk/low-risk group log-rank P=0.000). This four-miRNA signature
is an independent prognostic factor compared with routine
clinicopathological features in the test set. These miRNAs targeted
1,634 genes, and a miRNA-target gene network was constructed using
miRNet. The target genes of these four miRNAs were involved in various
pathways related to cancer.
Conclusion
Our observations suggest that the four-miRNA signature correlated with
the survival of patients with ccRCC and can be used as a prognostic
biomarker of ccRCC.
Keywords: ccRCC, miRNA signature, overall survival, prognostic
biomarkers
Introduction
Renal cell carcinoma (RCC) is a common malignant tumor of the urinary
system, accounting for 2%–3% of adult malignancies,[41]^1 and more than
100,000 people die of kidney cancer every year worldwide.[42]^2 The
most common subtype of RCC is clear cell renal cell carcinoma (ccRCC),
which is associated with high morbidity and poor prognosis.[43]^3
miRNAs were first identified by Lee et al[44]^4 in Caenorhabditis
elegans in 1993 as 19- to 24-nucleotide-long ncRNAs. It is estimated
that miRNAs regulate the expression of >60% protein-coding genes.
miRNAs are involved in various biological processes, such as cell
growth, proliferation, differentiation, and apoptosis.[45]^5 Owing to
the tissue-specific expression of miRNAs, their expression profile has
been associated with various diseases. Currently, abnormally expressed
miRNAs have been detected in many human tumors, such as bladder
cancer,[46]^6 lung cancer,[47]^7 prostate cancer,[48]^8 pancreatic
cancer,[49]^9 gastric cancer,[50]^10 liver cancer,[51]^11 and other
malignancies. Several recent studies have suggested that miRNA
expression profiling can be used to predict the clinical outcome of
patients with malignant tumors.[52]^12^,[53]^13 Specific miRNAs have
been used as potential diagnostic tools to distinguish the four
subtypes of RCC (clear cell RCC, papillary RCC, chromophobe RCC, and
benign oncocytomas).[54]^14 However, studies on the association of
miRNAs with ccRCC prognosis are limited. Currently, The Cancer Genome
Atlas (TCGA) database ([55]https://cancergenome.nih.gov/) can be used
to analyze complicated clinical characteristics and cancer genomics. In
this study, we screened the differentially expressed mature miRNAs
between ccRCC tissues and matched normal tissues, and determined the
association between these miRNAs and overall survival (OS). We
constructed a four-miRNA signature that may be used as a potential
prognostic biomarker of ccRCC.
Materials and methods
Data processing
The preprocessed ccRCC mature miRNA expression profiles in TCGA
database, displayed as log[2] converted reads per million (log[2] (RPM
+ 1)), and clinical information, were downloaded from the UCSC Xena
([56]https://xenabrowser.net/datapages/, version 09-08-2017). It
contains miRNA expression data from two different platforms, including
311 samples (241 ccRCC tissues and 70 matched normal kidney tissues)
based on the IlluminaHiSeq_miRNASeq platform (Illumina Inc., San Diego,
CA, USA) and 259 ccRCC tissues based on the IlluminaGA_miRNASeq
platform. The samples based on the IlluminaHiSeq_miRNASeq platform were
used as the training set to identify differentially expressed miRNAs
and to construct a miRNA-based signature for predicting prognosis. The
samples based on IlluminaGA_miRNASeq platform were used as the test set
to verify the signature. The mature miRNA sequencing data were
processed using R language.
Screening of differentially expressed miRNAs
In the training set, miRNAs that were not expressed in >10% samples
were removed. The differentially expressed miRNAs between ccRCCs and
matched normal tissues were analyzed using the “limma” package[57]^15
in R. The fold changes (FCs) in the expression of individual miRNAs
were calculated, and differentially expressed miRNAs with |log2FC|>1.0
and P<0.05 were considered to be significant. We applied bidirectional
hierarchical clustering to the differentially expressed miR-NAs based
on Euclidean distance and displayed the results as a heat map.
Construction and validation of the miRNA-based prognostic signature for ccRCC
In the training set, the patients were separated into high- and
low-level groups based on the median value of the differential
expression of miRNAs, followed by univariate and multivariate Cox
proportional hazards analyses. Finally, a miRNA signature-based
prognosis index (PI) score was constructed on the basis of a linear
combination of the expression level multiplied by a regression
coefficient derived from the multivariate Cox regression model (β)
using the following formula.
[MATH: PI=M1*β<
mn>1+M2*β<
mn>2+M3*β<
mn>3+… :MATH]
The “β” value is the estimated regression coefficient of miRNAs and is
derived from the multivariate Cox regression analysis, and “M”
indicates the expression profiles of the miRNAs. Patients with ccRCC
were divided into low- and high-risk groups based on median PI. The
survival times were compared between the two groups using Kaplan– Meier
analysis with log-rank test at P-value<0.05. The test set was used to
confirm the robustness and transferability of the miRNA-based
prognostic signature. We conducted the univariate and multivariate Cox
proportional hazards analyses in the training and test sets to compare
the relative prognostic value of this four-miRNA signature with that of
routine clinicopathological features.
Target gene prediction and functional enrichment analysis
Target gene prediction and functional enrichment analysis of the four
miRNAs were performed using miRNet ([58]http://www.mirnet.ca/).[59]^16
miRNet is an easy-to-use web-based tool that offers statistical,
visual, and network-based approaches to assist researchers understand
miRNA function and regulatory mechanisms and construct a miRNA-target
gene network. The Kyoto Encyclopedia of Genes and Genomes (KEGG)
pathway enrichment analysis was subsequently performed for the target
genes. P-value <0.05 was set as the cutoff criteria.
Statistical analysis
The chi-squared test was used for categorical data, and the unpaired
Student’s t-test was used to screen differentially expressed miRNAs.
Univariate/multivariate Cox proportional hazards analyses and
Kaplan–Meier survival analysis were used to compare the two groups of
patients. The chi-squared test and survival analysis were performed
using IBM SPSS statistics software program version 22.0 (IBM, Armonk,
NY, USA). All tests were two-sided, and P<0.05 was considered
statistically significant.
Results
Differential expression of miRNAs between ccRCC and matched normal kidney
tissues
The detailed clinical characteristics of patients with ccRCC, including
gender, age at diagnosis, histological grade, and TNM stage, are shown
in [60]Table 1. The training set contained more patients with
metastasis than the test set (16.88% vs 14.67%, chi-squared test,
P=0.000). This also partially assisted us to test the prognostic value
of this miRNA-based signature in different patients. According to the
cutoff criteria (P<0.05 and |log2FC|>1), 138 miRNAs were differentially
expressed between ccRCC and matched normal kidney tissues in the
training set. These included 54 upregulated miRNAs and 84 downregulated
miRNAs in ccRCC tissues. miRNA-21-5p was upregulated, whereas
miRNA-9-5p, miR-149-5p, and miRNA-30b-5p were downregulated in ccRCC
tissues. The results of the expression analysis are presented as a heat
map ([61]Figure 1), and the results of hierarchical clustering showed
that the expression patterns of these differentially expressed miRNAs
can correctly distinguish ccRCC from normal kidney tissues.
Table 1.
Summary of patient cohort information
Factors Training set
__________________________________________________________________
Test set
__________________________________________________________________
P
N=237 (%) N=259 (%)
__________________________________________________________________
Gender
Male 158 (66.67) 162 (62.55) 0.338
Female 79 (33.33) 97 (37.45)
Age
<65 153 (64.56) 162 (62.58) 0.711
≥65 84 (35.44) 97 (62.59)
Histologic
G1–2 99 (41.77) 119 (45.95) 0.227
G3–4 136 (57.38) 134 (51.74)
Gx 2 (0.84) 6 (2.32)
T
T1–2 151 (63.71) 156 (60.23) 0.481
T3–4 86 (36.29) 103 (39.77)
M
M0 166 (70.04) 221 (85.33) 0.000
M1 40 (16.88) 38 (14.67)
Mx 31 (13.08) 0 (0.00)
N
N0 97 (40.93) 123 (47.49) 0.323
N1 8 (3.38) 9 (3.47)
Nx 132 (55.70) 127 (49.03)
Stage
I–II 143 (60.34) 146 (56.37) 0.507
III–IV 92 (38.82) 112 (43.24)
Not reported 2 (0.84) 1 (0.39)
[62]Open in a new tab
Notes: Bold figure indicates statistically significant, P<0.05.
Figure 1.
[63]Figure 1
[64]Open in a new tab
Hierarchical clustering dendrograms of expression patterns of
differentially expressed miRNAs that can distinguish between normal
kidney tissue and ccRCC tissue.
Abbreviation: ccRCC, clear cell renal cell carcinoma.
Construction of miRNA-based signature with differentially expressed miRNAs
Four samples were removed because of lack of survival record in the
training set. For each of the 138 differentially expressed miRNAs, we
used the median expression level as a cutoff to stratify the remaining
237 patients into high-level and a low-level groups. The univariate Cox
proportional hazards regression analysis revealed that eight miRNAs
possessed prognostic value ([65]Table 2). Next, we performed a
multivariate Cox proportional hazards regression analysis and
identified four miRNAs, namely, miRNA-21-5p ([66]Figure 2A), miRNA-9-5p
([67]Figure 2B), miR-149-5p ([68]Figure 2C), and miRNA-30b-5p
([69]Figure 2D), as independent prognostic indicators. Thus, we used
these four miRNAs to construct a four-miRNA PI as follows:
[MATH:
PI=miRNA−21−5p*0.788+miR
−9−5p*0.536+miR
−149−5p*0.566+miR
−30b−5p*0.683. :MATH]
Table 2.
Univariate and multivariate analyses in ccRCC patients
miRNA Univariate analysis
__________________________________________________________________
Multivariate analysis
__________________________________________________________________
P 95% CI b P 95% CI
__________________________________________________________________
miR-21-5p 0.000 1.884–5.149 0.788 0.031 1.073–4.502
miR-9-5p 0.004 1.270–3.465 0.536 0.044 1.014–2.883
miR-149-5p 0.003 1.303–3.473 0.566 0.027 1.066–2.908
miR-204-5p 0.003 0.293–0.775 –0.118 0.692 0.195–1.594
miR-146b-5p 0.001 1.436–3.800 –0.014 0.976 0.388–2.505
miR-223-3p 0.007 1.193–3.086 0.130 0.615 0.685–1.894
miR-30b-5p 0.009 1.173–3.069 0.683 0.006 1.211–3.236
miR-146b-3p 0.004 1.247–3.254 0.143 0.727 0.516–2.580
[70]Open in a new tab
Notes: Bold figure indicates statistically significant, P<0.05.
Abbreviation: ccRCC, clear cell renal cell carcinoma.
Figure 2.
[71]Figure 2
[72]Open in a new tab
Four miRNAs were associated with overall survival in ccRCC patients
using Kaplan–Meier curves and log-rank tests.
Note: (A) miRNA-21-5p; (B) miRNA-9-5p; (C) miRNA-149-5p; (D)
miRNA-30b-5p.
Abbreviation: ccRCC, clear cell renal cell carcinoma.
A PI was calculated for each patient in the training set. Then, 237
patients were separated into low- and high-risk groups according to
median PI. Survival analysis was performed using the Kaplan–Meier
method with log-rank test. Results revealed that patients in the
high-risk group (n=119) had significantly shorter survival time than
those in the low-risk group (n=118) (high-risk/low-risk group log-rank
P=0.000; [73]Figure 3A). This four-miRNA signature is an independent
prognostic factor compared with routine clinicopathological features
([74]Table 3).
Figure 3.
[75]Figure 3
[76]Open in a new tab
The Kaplan–Meier curves obtained using the four-miRNA signature to
separate patients into high- and low-risk groups.
Notes: (A) Kaplan-Meier curve for training set; (B) Kaplan-Meier curve
for testing set.
Table 3.
Univariate and multivariate analyses of routine clinicopathological
features and four-miRNA prognostic signature PI in the training set
Factors Univariate analysis
__________________________________________________________________
Multivariate analysis
__________________________________________________________________
P 95% CI P 95% CI
__________________________________________________________________
Gender (female/male) 0.862 0.579–1.580
Age (<65 years/≥65 years) 0.149 0.885–2.245
Histologic (G3–4/G1–2) 0.000 1.819–6.294 0.641 0.490–3.189
M (M0/M1) 0.000 2.933–7.587 0.702 0.467–3.103
N (N1/N0) 0.000 2.518–18.038 0.048 1.008–8.724
T (T1–2/T3–4) 0.000 2.018–5.283 0.408 0.146–2.187
Stage (I–II/III–IV) 0.000 2.677–7.515 0.116 0.729–17.584
PI (high/low risk) 0.000 1.692–4.759 0.008 1.363–7.740
[77]Open in a new tab
Notes: Bold figure indicates statistically significant, P<0.05.
Abbreviation: PI, prognosis index.
Verification of the four-miRNA signature in the test set
Similar to that observed in the training set, patients in the test set
were divided into low- and high-risk groups according to median PI, and
Kaplan–Meier analysis was used to compare the patient’s OS. The results
of survival analysis revealed that patients in the high-risk group
(n=130) had significantly shorter survival time than those in the
low-risk group (n=129) (high-risk/low-risk group log-rank P=0.000;
[78]Figure 3B). The four-miRNA signature is an independent prognostic
factor compared with routine clinicopathological features in the test
set ([79]Table 4). Overall, these results are consistent with that in
the training set.
Table 4.
Univariate and multivariate analyses of routine clinicopathological
features and four-miRNA prognostic signature PI in the test set
Factors Univariate analysis
__________________________________________________________________
Multivariate analysis
__________________________________________________________________
P 95% CI P 95% CI
__________________________________________________________________
Gender (female/male) 0.490 0.767–1.740
Age (<65 years/≥65 years) 0.001 1.360–3.056 0.011 1.193–2.808
Histologic (G3–4/G1–2) 0.000 1.713–4.202 0.407 0.665–2.735
M (M1/M0) 0.000 2.057–4.719 0.108 0.820–7.363
N (N1/N0) 0.026 1.118–5.520 0.412 0.588–3.659
T (T3–4/T1–2) 0.000 2.688–6.388 0.000 2.189–9.396
Stage (I–II/III–IV) 0.000 2.314–5.529 0.349 0.155–1.931
PI (high/low risk) 0.000 2.481–6.265 0.000 1.847–8.377
[80]Open in a new tab
Notes: Bold figure indicates statistically significant, P<0.05.
Abbreviation: PI, prognosis index.
Target gene prediction and functional enrichment analysis
To investigate the potential biological functions of these four miRNAs,
we predicted their target genes using miRNet. We observed that 1,634
genes were targeted by these four miRNAs and a miRNA-target gene
network was constructed using miRNet ([81]Figure 4). KEGG pathway
enrichment analyses of the target genes revealed that they were
involved in various pathways related to cancer, such as the MAPK, p53,
and Wnt signaling pathways, cell cycle, and RNA transport ([82]Figure
5).
Figure 4.
[83]Figure 4
[84]Open in a new tab
The miRNA-target genes network constructed using the miRNet and the
corresponding target genes involved in KEGG Pathways.
Note: (A) The total network of four miRNAs and their targets; (B)
pathways in cancer, (C) p53 signaling pathway; (D) cell cycle pathway;
(E) renal cell carcinoma; (F) Wnt signaling pathway.
Abbreviation: KEGG, Kyoto Encyclopedia of Genes and Genomes.
Figure 5.
[85]Figure 5
[86]Open in a new tab
Kyoto Encyclopedia of Genes and Genomes pathway analysis of the
predicted targets of the four miRNAs.
Abbreviation: HTLV-1, human T lymphotropic virus type 1.
Discussion
ccRCC is one of the most common renal malignancies associated with high
mortality and morbidity.[87]^17 However, clinical tools for predicting
patient outcome utilize traditional clinical parameters. Therefore,
accurate identification of predictive factors from data obtained from
analysis of ccRCC specimens is clinically challenging. Identification
and validation of novel biomarkers form an important part of practical
studies on ccRCC. During tumorigenesis, miRNAs act as oncogenes or
tumor suppressors; hence, the biological behavior of tumors can be
inhibited by regulating miRNA levels for therapeutic purposes.
Identification of RNA profiles and selective targets is the basis for
individualized treatment of different tumors. Previous studies have
shown that some specific miRNAs were aberrantly expressed in RCC and
participate in its development.[88]^18^–[89]^21 However, detailed
analyses of the associations between miRNA expression and prognosis of
patients with ccRCC remain limited.
In this study, we identified 138 miRNAs that were differentially
expressed between ccRCC and normal kidney tissues. The univariate Cox
proportional hazards regression analysis revealed that eight miRNAs
possessed prognostic value. We confirmed that the four-miRNA
(miRNA-21-5p, miRNA-9-5p, miRNA-149-5p, and miRNA-30b-5p) signature can
be regarded as an independent predictor of prognostic OS after
considering the various variables, including gender, age, histology,
and stages. Previous studies have also identified these four miRNAs,
and several studies have investigated the relationship between miRNA
expression patterns and cancer. Kowalczyk et al[90]^22 reported that
special AT-rich sequence binding protein 1 (SATB1) may be a potential
prognostic marker for ccRCC, as low SATB1 expression in ccRCC may
result from overexpression of miR-21-5p. SATB1 downregulation and
miR-21-5p upregulation were associated with shorter patient survival.
At present, the role of miRNA-9-5p in tumors has not been clarified.
Certain studies show that downregulation of miR-9-5p expression can
reverse the effect of miRNA-9-5p on proliferation, colony formation,
cell cycle arrest, and apoptosis in osteosarcoma cells.[91]^23 Okato et
al[92]^24 demonstrated that dual strands of pre-miR-149 (miRNA-149-5p
and miRNA-149-3p) acted as antitumor miRNAs by targeting FOXM1, which
was shown to be associated with survival of patients with ccRCC. Liu et
al[93]^25 suggested that miR-30b-5p acts as a novel tumor suppressor to
regulate RCC cell proliferation, metastasis, and epithelial to
mesenchymal transition by downregulating GNA13 expression. In other
words, miR-30b-5p may be considered a potential biomarker for RCC
diagnosis. However, studies demonstrating that the four differentially
expressed miRNAs were predictors of ccRCC were lacking.
In this study, we constructed a four-miRNA signature, and the PI of
this signature was calculated for each patient, which successfully
separated patients into low- and high-risk groups. Specifically,
patients considered high-risk by our four-miRNA signature had
significantly poor prognosis than those in the low-risk group
(P<0.001). We confirmed that the four-miRNA signature is an independent
predictor of OS in patients with ccRCC. Similar to that observed in the
training set, patients in the test set were divided into low- and
high-risk groups based on the risk score of individual patients, and
Kaplan–Meier analysis was used to compare patient survival differences.
Statistically significant differences (P<0.0001) were observed between
high- and low-risk groups. This confirmed that our four-miRNA signature
is an independent and universal predictor of ccRCC.
It is well known that miRNAs modulate gene expression. Hence, we
screened the target genes of these four miRNAs and used bioinformatics
to predict the pathways and biological functions associated with their
targets. The target genes were significantly enriched in multiple
cancer-associated pathways, such as MAPK, p53, and Wnt signaling
pathways. Abnormal regulation of these signaling pathways is involved
in the development of various human cancers, such as breast cancer,
hepatocellular carcinoma, hematological cancer, and lung
cancer.[94]^26^–[95]^28 Thus, abnormal regulation of signaling pathways
may play a crucial role in the pathogenesis and progression of ccRCC.
Further molecular investigations may provide new therapeutic targets
for ccRCC.
Limitations
However, our study has few limitations. First, the ccRCC tissues were
more than normal kidney tissues. Second, the miRNA expression profiles
in the test set were not based on the same platform. Therefore, this
four-miRNA signature has to be verified in a larger independent cohort
of patients.
Conclusion
A comprehensive analysis of differentially expressed miRNA profiles and
corresponding clinical information suggested that a four-miRNA
signature was an independent and universal prognostic factor in
patients with ccRCC. These miRNAs modulated genes associated with
multiple cancer-associated pathways. However, further studies are
required to verify our observations and establish the molecular
mechanism underlying the interplay of miRNAs, their target genes, and
ccRCC progression.
Acknowledgments