Abstract

   To explore and understand the competitive mechanism of ceRNAs in
   intrahepatic cholangiocarcinoma (ICC), we used bioinformatics analysis
   methods to construct an ICC-related ceRNA regulatory network (ceRNET),
   which contained 340 lncRNA-miRNA-mRNA regulatory relationships based on
   the RNA expression datasets in the NCBI GEO database. We identified the
   core regulatory pathway RP11-328K4.1-hsa-miR-27a-3p-PROS1, which is
   related to ICC, for further validation by molecular biology assays. GO
   analysis of 44 differentially expressed mRNAs in ceRNET revealed that
   they were mainly enriched in biological processes including “negative
   regulation of epithelial cell proliferation” and "positive regulation
   of activated T lymphocyte proliferation.” KEGG analysis showed that
   they were mainly enriched in the “complement and coagulation cascade”
   pathway. The molecular biology assay showed that lncRNA RP11-328K4.1
   expression was significantly lower in the cancerous tissues and
   peripheral plasma of ICC patients than in normal controls (p<0.05). In
   addition, hsa-miR-27a-3p was found to be significantly upregulated in
   the cancer tissues and peripheral plasma of ICC patients (p<0.05).
   Compared to normal controls, the expression of PROS1 mRNA was
   significantly downregulated in ICC patient cancer tissues (p<0.05) but
   not in peripheral plasma (p>0.05). Furthermore, ROC analysis revealed
   that RP11-328K4.1, hsa-miR-27a-3p, and PROS1 had significant diagnostic
   value in ICC. We concluded that the upregulation of lncRNA
   RP11-328K4.1, which might act as a miRNA sponge, exerts an antitumor
   effect in ICC by eliminating the inhibition of PROS1 mRNA expression by
   oncogenic miRNA hsa-miR-27a.

   Keywords: intrahepatic cholangiocarcinoma, ceRNA, ceRNA regulatory
   network, biomarkers, prognosis

INTRODUCTION

   Intrahepatic cholangiocarcinoma (ICC) is an adenocarcinoma that
   originates from the intrahepatic secondary bile duct and its branch
   epithelium. ICC is anatomically different from the other two types of
   cholangiocarcinoma (CCA): perihilar cholangiocarcinoma (pCCA) and
   distal cholangiocarcinoma (dCCA) [[33]1, [34]2]. The incidence of ICC
   accounts for 10%-15% of primary hepatic malignant tumors, only second
   to the incidence of primary hepatocellular carcinoma [[35]1].
   Statistics from recent years indicate that the morbidity and mortality
   rates of ICC continue to show an increasing trend globally [[36]1,
   [37]3]. Due to a lack of obvious clinical symptoms and limited
   effective screening methods in the early stage of the disease, most ICC
   patients do not have the option of surgery at diagnosis; only 30%-40%
   of patients have the opportunity to get surgery after diagnosis [[38]4,
   [39]5]. Patients who have not undergone surgery have an extremely poor
   prognosis, and few patients survive for more than three years. The
   3-year survival rate of patients undergoing surgery is only 40%-50%
   [[40]6]. Therefore, it is particularly urgent and necessary to better
   understand the molecular mechanisms underlying the pathogenesis and
   progression of ICC and to find potential biomarkers for diagnosis and
   prognosis as well as therapeutic targets for ICC.

   Early research on the molecular mechanism of carcinogenesis has mainly
   focused on different protein-coding genes. With the development and
   popularization of high-throughput whole-genome sequencing technology, a
   variety of noncoding ribonucleic acids (ncRNA) of different lengths
   have been clearly found to play key regulatory roles in human
   carcinogenesis, including noncoding RNAs (ncRNAs), such as long
   noncoding RNA (lncRNA) and microRNA (miRNA) [[41]7]. lncRNAs are a
   subset of noncoding transcripts of over 200 nucleotides in length, with
   little or no protein-coding ability, and they play key roles in a
   series of biological processes by regulating gene expression through
   mechanisms including transcription, splicing, and translation [[42]8,
   [43]9]. Due to their greater tissue specificity, lncRNAs are more
   effective as biomarkers for the early diagnosis and screening of cancer
   patients [[44]10]. Recent studies [[45]11, [46]12] have found that
   lncRNAs may be potential diagnostic and prognostic biomarkers for CCA
   and that they may be related to the pathogenesis and progression of
   CCA. miRNAs are a class of endogenous small RNAs approximately 20-24
   nucleotides in length that play various important regulatory functions
   within cells [[47]13]. Each miRNA can regulate multiple target genes,
   and several miRNAs can regulate a single gene. Therefore, this complex
   regulatory network can regulate the expression of multiple genes
   through a single miRNA or specifically regulate the expression of a
   single gene through the combination of multiple miRNAs [[48]14].
   Similarly, some studies [[49]15–[50]19] also suggest that aberrantly
   expressed miRNAs can be used as diagnostic and prognostic markers for
   CCA that are closely associated with the pathogenesis, progression and
   metastasis of CCA. However, the underlying mechanisms of lncRNAs and
   miRNAs in CCA, especially in ICC, are not fully understood.

   Recent studies have shown that lncRNAs, as competitive endogenous RNA
   (ceRNA) with miRNA response elements (MRE), can compete with mRNAs for
   binding with miRNAs, thus affecting gene expression [[51]20–[52]22].
   Abnormal regulation of ceRNA is involved in multiple types of cancers,
   such as breast cancer, lung cancer, gastric cancer, colorectal cancer,
   hepatic cancer and CCA [[53]23–[54]29]. Mathematical modeling,
   informatics-based analysis and experimental validation have been used
   to describe the structure of ceRNA regulatory networks (ceRNETs) and
   their role in regulating cellular physiology under normal conditions
   and pathological conditions such as cancer [[55]22]. Previous studies
   have thoroughly discussed the diagnostic and prognostic value of
   lncRNA-related ceRNETs and their pivotal role in the pathogenesis and
   progression of HCC. Gao M et al. [[56]30] constructed ceRNETs of
   HCC-related lncRNAs (HOTAIR and MALAT1) by using bioinformatics
   methods. These networks predicted that MALAT1 and HOTAIR can act as
   miRNA sponges to inhibit hsa-miR-1 and hsa-miR-20a-5p, thereby removing
   the inhibition of the expression of cyclin D1, E2F1, EGFR, MYC, MET,
   NOS2A and VEGFA. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and
   Genomes (KEGG) enrichment analyses of these seven HCC-related miRNA
   target genes indicates that MALAT1 and HOTAIR could promote cell
   growth, cell cycle progression and mitosis by involving in cell cycle,
   focal adhesion and disease progression pathways. Yan Y et al. [[57]31]
   identified nine key lncRNAs in the overall ceRNET by constructing
   lncRNA-related ceRNETs (HCG18, [58]AC021078.1, ENT-PD1-AS1, MCM3AP-AS1,
   GMDS-AS1, [59]AC019080.1, [60]AC245452.1, LINC00630 and
   [61]AP000766.1). Functional enrichment analysis of coexpressed adjacent
   mRNAs revealed a close association with liver function and the
   pathogenesis of HCC. Further construction of a ceRNA subnet associated
   with HCC prognosis was done by screening 16 lncRNAs associated with HCC
   prognosis, which were further used for constructing a risk scoring
   model. According to the median risk score, the overall survival (OS)
   was significantly higher in the low-risk group than in the high-risk
   group (P = 8.31e-05). At present, studies on the pathogenesis of CCA by
   constructing ceRNETs and examining their role in the diagnosis,
   prognosis and treatment of CCA are relatively rare and lack depth. The
   limited CCA-related ceRNETs constructed by Song W et al. [[62]29]
   contain 116 lncRNAs, 14 miRNAs and 59 mRNAs. Functional enrichment
   analysis revealed that these ncRNAs promote CCA progression mainly
   through the estrogen signaling pathway and MAPK. Meanwhile, seven
   lncRNAs with negative correlations with CCA prognosis and four lncRNAs
   with positive correlations with CCA prognosis were identified. However,
   CCA includes three types, ICC, pCCA and dCCA, which are significantly
   different in anatomical location, clinical manifestation, morphology
   and epidemiology. At present, the role of ceRNETs in ICC remains
   unclarified.

   In this study, to determine the diagnostic, therapeutic, and prognostic
   value of lncRNA-related ceRNETs and their key role in the pathogenesis
   and progression of ICC, we performed integrated analysis of expression
   profile data on ICC-related lncRNA, miRNA and mRNA from National Center
   for Biotechnology Information Gene Expression Omnibus (NCBI GEO).
   Afterwards, we integrated these identified differential expression (DE)
   RNAs and constructed a lncRNA-miRNA-mRNA regulatory network. Moreover,
   functional enrichment analysis was performed on mRNA involved in ceRNET
   construction. Then, based on the ranking of each ceRNA in ceRNETs
   (including connectivity and log-fold change, logFC), their
   relationships (whether their direction of expression changes is
   consistent or opposite), and their mention in the results of previous
   studies in other tumors, we identified the most related core regulatory
   pathways of ICC. Finally, we experimentally validated the mRNA and
   corresponding protein in the core ceRNET regulatory pathway with ICC
   fresh tissues, blood, and paraffin sections. We also clinically
   validated the RNAs in the core ceRNET regulatory pathway with datasets
   from the NCBI GEO and TCGA. In combination with experimental results,
   clinical outcomes, and previous research, the mechanism of these
   noncoding RNAs and their constituent pathways in ICC was further
   discussed.

RESULTS

ICC-related differentially expressed lncRNAs, miRNAs and mRNAs based on GEO
microarray chips

   From the [63]GSE61850 dataset, a total of 3533 differentially expressed
   mRNAs (1650 upregulated and 1883 downregulated mRNAs) and 692
   differentially expressed lncRNAs (286 upregulated and 406 downregulated
   lncRNAs) were obtained. The heat map and volcano map are shown in
   [64]Figures 1 and [65]2, respectively (partial data are shown in
   [66]Supplementary Tables 1 and [67]2):

Figure 1.

   [68]Figure 1
   [69]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed mRNAs in
   the [70]GSE61850 dataset.

Figure 2.

   [71]Figure 2
   [72]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed lncRNAs in
   the [73]GSE61850 dataset.

   From the [74]GSE103909 dataset, a total of 948 differentially expressed
   mRNAs (447 upregulated and 501 downregulated mRNAs) and 283
   differentially expressed lncRNAs (56 upregulated and 227 downregulated
   lncRNAs) were obtained. The heat map and volcano map are shown in
   [75]Figures 3 and [76]4, respectively (partial data are shown in
   [77]Supplementary Tables 3 and [78]4):

Figure 3.

   [79]Figure 3
   [80]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed mRNAs in
   the [81]GSE103909 dataset.

Figure 4.

   [82]Figure 4
   [83]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed lncRNAs in
   the [84]GSE103909 dataset.

   From the [85]GSE57555 dataset, a total of 2047 differentially expressed
   mRNAs were obtained (942 upregulated and 1105 downregulated mRNAs), and
   106 differentially expressed miRNAs (64 upregulated and 42
   downregulated miRNAs) were obtained. The heat map and volcano map are
   shown in [86]Figures 5, [87]6, respectively (partial data are shown in
   [88]Supplementary Tables 5 and [89]6):

Figure 5.

   [90]Figure 5
   [91]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed mRNAs in
   the [92]GSE57555 dataset.

Figure 6.

   [93]Figure 6
   [94]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed miRNAs in
   the [95]GSE57555 dataset.

   From the [96]GSE53992 dataset, a total of 155 differentially expressed
   miRNAs (71 upregulated and 84 downregulated miRNAs) were obtained. The
   heat map and volcano map are shown in [97]Figure 7 (partial data are
   shown in [98]Supplementary Table 7):

Figure 7.

   [99]Figure 7
   [100]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed miRNAs in
   the [101]GSE53992 dataset

   From the [102]GSE53870 dataset, a total of 207 differentially expressed
   miRNAs were obtained (104 upregulated and 103 downregulated). The heat
   map and volcano map are shown in [103]Figure 8 (partial data are shown
   in [104]Supplementary Table 8):

Figure 8.

   [105]Figure 8
   [106]Open in a new tab

   Heat map (A) and volcano map (B) of differentially expressed miRNAs in
   the [107]GSE53870 dataset

   The intersections of differentially expressed mRNAs, differentially
   expressed lncRNAs, and differentially expressed miRNAs are shown in the
   Venn map ([108]Figure 9), with a total of 236 consensus differentially
   expressed mRNAs, 71 consensus differentially expressed lncRNAs, and 16
   consensus differentially expressed miRNAs. Some of the results are
   shown in [109]Supplementary Tables 9–[110]11.

Figure 9.

   [111]Figure 9
   [112]Open in a new tab

   Venn diagram of differentially expressed mRNAs, lncRNAs, miRNAs in all
   datasets (from A to B to C: mRNA, lncRNA, miRNA).

Coexpression analysis and miRNA target gene prediction analysis for
constructing lncRNA-miRNA and miRNA-mRNA relationship pairs

   Coexpression analysis and the intersection of [113]GSE61850 and
   [114]GSE103909 datasets revealed 1403 lncRNA-mRNA relationship pairs
   with synergistic expression, including 194 mRNAs and 54 lncRNAs (some
   results are shown in [115]Supplementary Table 12). Coexpression
   analysis of the [116]GSE57555 dataset revealed 1166 miRNA-mRNA
   inversely correlated relationship pairs, including 16 miRNAs and 220
   mRNAs (some results are shown in [117]Supplementary Table 13).

   The online tool mirwalk2.0 was used to perform target gene prediction
   on the above 16 miRNAs. A total of 29,426 miRNA-mRNA relationship pairs
   were obtained (some results are shown in [118]Supplementary Table 14).
   By intersecting with the miRNA-mRNA obtained in the first step, a total
   of 113 miRNA-mRNA relationship pairs were obtained, including 12 miRNAs
   and 58 mRNAs (detailed results are shown in [119]Supplementary Table
   15). A total of 54 lncRNAs and 16 miRNAs from the coexpression analysis
   were extracted for the prediction of miRNA-lncRNA binding sites by
   using the local software Miranda (v3.3a), which revealed 362
   miRNA-lncRNA relationship pairs, including 16 miRNAs and 53 lncRNAs
   (some results are shown in [120]Supplementary Table 16).

Construction of a ceRNA regulation network (ceRNET) based on GEO chip
databases

   To further explore how lncRNAs regulates mRNA expression by binding to
   miRNAs in ICC, based on the miRNA-lncRNA and miRNA-mRNA relationship
   pairs obtained from the previous step and according to the premise that
   mutual ceRNAs with the same miRNA binding sites exist in the ceRNA
   network, we first screened the mRNAs and lncRNAs regulated by the same
   miRNA. Then, according to the consistent expression trend among ceRNAs
   and the synergistic expression relationships between mRNAs and lncRNAs,
   we finally obtained 340 pairs of lncRNA-miRNA-mRNA regulatory
   relationships, containing 44 mRNAs, 12 miRNAs and 24 lncRNAs (some of
   the results are shown in [121]Supplementary Table 17).

   Cytoscape 3.4.0 was used to construct a ceRNA network for the 340
   lncRNA-miRNA-mRNA regulatory relationships obtained above. We also
   determined the upregulation and downregulation of these nodes (some
   results are shown in [122]Supplementary Table 18). The ceRNA network is
   shown in [123]Figure 10.

Figure 10.

   [124]Figure 10
   [125]Open in a new tab

   The ICC-related ceRNA network map (green circles indicate the
   downregulated mRNAs, red circles indicate the upregulated mRNA, dark
   blue diamonds indicate the downregulated lncRNA, light purple diamonds
   indicate the upregulated lncRNA, pink triangles indicate the
   upregulated miRNA, brown triangles indicate the downregulated miRNA,
   and gray triangles indicate that upregulation or downregulation of
   miRNAs cannot be determined. The gray arrow indicates the regulatory
   relationship between miRNA and lncRNA, the yellow arrow indicates the
   regulatory relationship between miRNA and mRNA, and the green dotted
   line indicates the synergistic expression relationship between lncRNA
   and mRNA).

GO and KEGG pathway enrichment analyses of differentially expressed mRNAs in
the constructed ICC-related ceRNETs

   Functional enrichment analysis was performed on differentially
   expressed mRNA in the constructed ICC-related ceRNETs (shown in
   [126]Figure 11 and [127]Table 1). GO analysis revealed that the
   biological processes associated with differentially expressed mRNAs and
   tumors were mainly involved in the negative regulation of epithelial
   cell proliferation and the positive regulation of activated T cell
   proliferation. KEGG analysis revealed that differentially expressed
   mRNAs were mainly involved in the complement and coagulation cascade
   pathways.

Figure 11.

   [128]Figure 11
   [129]Open in a new tab

   GO terms (including biological process (BP), cellular component (CC)
   and molecular function (MF)) and KEGG pathways involved in the
   construction of ICC-related ceRNETs of 44 DE mRNAs.

Table 1. Specific DE mRNAs enriched in each GO term and KEGG pathway in
ICC-related ceRNA networks.

   Category Term Count Genes PValue
   GOTERM_BP muscle cell cellular homeostasis 2 CFL2, SOD1 0.033571
   GOTERM_BP positive regulation of activated T cell proliferation 2 IGF2,
   IGFBP2 0.035927
   GOTERM_BP regulation of vasoconstriction 2 AGTR1, ADRA2B 0.03121
   GOTERM_BP single fertilization 2 AR, CECR2 0.052263
   GOTERM_BP negative regulation of epithelial cell proliferation 2 AR,
   MCC 0.09083
   GOTERM_BP spermatogenesis 3 AR, SERPINA5, SOD1 0.058154
   GOTERM_CC extracellular space 10 SERPINA11, CFL2, SERPINA5, CCBE1,
   C1RL, RELN, IGF2, IGFBP2, SOD1, GREM2 1.36E-04
   GOTERM_CC extracellular exosome 15 THRB, C6, IGF2, ISOC1, METTL7A,
   SOD1, ALDH7A1, GPM6A, SERPINA5, CFL2, C1RL, GFRA1, PEBP1, IGFBP2, PROS1
   2.97E-04
   GOTERM_MF steroid binding 2 AR, PAQR9 0.024354
   GOTERM_MF steroid hormone receptor activity 2 THRB, PAQR9 0.094041
   GOTERM_MF serine-type endopeptidase inhibitor activity 3 SERPINA11,
   SERPINA5, PEBP1 0.01002
   KEGG_PATHWAY Prion diseases 2 C6, SOD1 0.090786
   KEGG_PATHWAY Complement and coagulation cascades 3 SERPINA5, C6, PROS1
   0.017377
   [130]Open in a new tab

   Meanwhile, the Cytoscape plugin CytoNCA was used to analyze the node
   connectivity of the network, with unweighted parameters. The node size
   in the figure indicates the degree of connectivity in the network: the
   larger the node, the higher the degree of connectivity (some of the
   data are shown in [131]Supplementary Table 19 for details). Based on
   the ranking of network nodes, the reproducibility of expression trends
   in several GEO datasets, the logFC and ceRNA relationship, and previous
   outcomes in other types of tumors, the following ceRNA regulatory
   relationships were finally chosen for further validation:
   RP11-328K4.1-hsa-miR-27a-3p-PROS1; RP11-328K4.1-hsa-miR-27a-3p-METTL7A;
   RP11-328K4.1-hsa-miR-200a-3p-METTL7A;
   RP11-328K4.1-hsa-miR-200a-3p-CECR2; ADORA2A-AS1-hsa-miR-200b-3p-CECR2;
   ADORA2A-AS1-hsa-miR-27a-3p-PROS1; ADORA2A-AS1-hsa-miR-200c-3p-CECR2;
   LINC01485-hsa-miR-200c-3p-CECR2; LINC01485-hsa-miR-200b-3p-METTL7A; and
   LINC01485-hsa-miR-200b-3p-CECR2.

   After comprehensive analysis, we finally obtained the ICC-related core
   regulatory pathway, RP11-328K4.1-hsa-miR-27a-3p-PROS1, which was
   further validated in relevant fresh tissue, blood samples and paraffin
   sections. The results are shown as follows: The expression of lncRNA
   RP11-328K4.1 in 10 pairs of fresh ICC cancer and adjacent paracancerous
   tissues is shown in [132]Figure 12. The expression of lncRNA
   RP11-328K4.1 was significantly decreased in the ICC experimental group
   (cancer tissue) compared to that in the control group (paracancerous
   tissue) (P = 0.000007).

Figure 12.

   [133]Figure 12
   [134]Open in a new tab

   qRT-PCR analysis showed the difference in lncRNA RP11-328K4.1
   expression between the experimental group and control group in ICC
   fresh tissue samples after normalization to internal controls.
   RP11-328K4.1 was normalized to β-actin.

   The expression level of lncRNA RP11-328K4.1 in the peripheral plasma of
   10 ICC patients and 10 healthy subjects is shown in [135]Figure 13. The
   expression of lncRNA RP11-328K4.1 was significantly decreased in the
   experimental group (peripheral plasma of ICC patients) compared to that
   in the control group (peripheral plasma of healthy controls)
   (P=0.036093).

Figure 13.

   [136]Figure 13
   [137]Open in a new tab

   qRT-PCR analysis showed the difference in lncRNA RP11-328K4.1
   expression between the experimental group and control group in
   peripheral plasma samples after normalization to internal controls.
   RP11-328K4.1 was normalized to β-actin.

   The expression levels of hsa-miR-27a-3p in 10 pairs of fresh ICC cancer
   and adjacent tissues are shown in [138]Supplementary Figure 1. The
   expression level of hsa-miR-27a-3p was significantly higher in the
   experimental group (ICC cancer tissue) than that in the control group
   (paracancerous tissues) (P = 0.00016).

   The expression levels of hsa-miR-27a-3p in the peripheral plasma of 10
   ICC patients and 10 healthy subjects are shown in [139]Supplementary
   Figure 2 below. The expression of hsa-miR-27a-3p was significantly
   higher in the experimental group (peripheral plasma of ICC patients)
   than that in the control group (peripheral plasma of healthy subjects)
   (P=0.04942034).

   The expression levels of PROS1 mRNA in 10 pairs of fresh ICC cancer and
   paracancerous tissues are shown in [140]Supplementary Figure 3. The
   expression of PROS1 mRNA was significantly decreased in the
   experimental group (ICC cancer tissues) compared with that in the
   control group (paracancerous tissues) (P = 0.006611).

   The expression levels of PROS1 mRNA in the peripheral plasma of 10 ICC
   patients and 10 healthy subjects are shown in [141]Supplementary Figure
   4. The expression of PROS1 mRNA was decreased in the experimental group
   (peripheral plasma of ICC patients) compared to that in the control
   group (peripheral plasma of healthy subjects), however, the difference
   was not statistically significant (P=0.171259).

Western Blot (WB) assay to detect protein corresponding to PROS1 mRNA in
tissues

   The expression levels of the protein corresponding to PROS1 mRNA in 10
   pairs of fresh ICC cancer and adjacent noncancer tissues are shown in
   [142]Supplementary Figures 5, [143]6. The expression of protein
   corresponding to PROS1 mRNA was lower in the experimental group (ICC
   cancer tissue) than that in the control group (paracancerous tissues),
   however, the difference was not statistically significant (P =
   0.668353048).

WB assay to detect protein corresponding to PROS1 mRNA in plasma

   The expression levels of the protein corresponding to PROS1 mRNA in the
   peripheral plasma of 10 ICC patients and the peripheral plasma of 10
   healthy subjects are shown in [144]Supplementary Figures 7 and [145]8.
   The expression of protein corresponding to PROS1 mRNA was increased in
   the experimental group (peripheral plasma of ICC patients) compared to
   that in the control group (peripheral plasma of healthy subjects),
   however, the difference was not statistically significant
   (P=0.597799476).

Immunohistochemistry (IHC) results of PROS1 expression in paraffin sections
from ICC patients

   In this study, we performed IHC on paraffin sections from 88 ICC
   patients. The median age of patients was 62 years (range: 30-83 years).
   There were 52 males and 36 females, with a male:female ratio of 1.75:1.
   A total of 56 patients underwent radical surgery, accounting for 63.6%
   of the total number of patients. The detailed data are shown in
   [146]Table 2. The IHC results of PROS1 staining in cancer tissues and
   adjacent normal tissues were analyzed and are shown in
   [147]Supplementary Figure 9A–[148]9D: PROS1 showed positive staining in
   the cytoplasm of ICC cancer tissues, and the staining intensity could
   be divided into high, medium and low degrees. However, the expression
   of PROS1 in the cytoplasm of normal adjacent tissue was nearly
   negative.

Table 2. The baseline characteristics and IHC of 88 ICC patients receiving
surgery.

   Characteristics Number of patients (n=154) Characteristics Number of
   patients (n=154)
   Age (year) 62 (30-83) yes 1(1.1)
   ≤ 60 34 (38.6) jaundice
   > 60 54 (61.4) no 71 (80.7)
   Gender yes 17 (19.3)
   male 52 (59.1) Blood type
   female 36 (40.9) A 27 (30.7)
   smoking B 33 (37.5)
   no 59(67.1) AB 6 (6.8)
   yes 29(32.9) O 22 (25.0)
   alcohol GGT 270.1(12-2769)
   no 71(80.7) ≤50 24 (27.3)
   yes 17(19.3) >50 64 (62.7)
   BMI 24.10 (16.9-32.6) differentiation
   <18.5 1(1.1) Poorly differentiated 29 (33.0)
   ≥18.5 and <24 42(47.8) Moderately-well 59 (67.0)
   24 45(51.1) differentiated
   gallstone Margin status
   no 74 (84.1) negative 56 (63.6)
   yes 14 (15.9) positive 32 (36.4)
   cholangiolithiasis Largest tumor diameter (cm) 4.83(1.0-14)
   no 80(91) ≤ 5 53 (60.2)
   yes 8(9) > 5 35 (39.8)
   cholecystitis T stage
   no 70(79.5) Tis-T1a 21 (23.9)
   yes 18(20.5) T1b 18 (20.5)
   diabetes T2 16(18.2)
   no 69 (78.4) T3 15 (17.1)
   yes 19 (21.6) T4 18 (20.5)
   hypertension N stage
   no 61(69.3) 0 stage 62 (70.5)
   yes 27(30.7) 1 stage 26 (29.5)
   Fatty liver M stage
   no 86(97.7) no 73 (83.0)
   yes 2(2.3) yes 15 (17.0)
   cirrhosis TNM stage
   no 82(93.2) 1A-IB 31 (35.3)
   yes 6(6.8) II 10 (11.4)
   HBV IIIA-IIIB 33 (37.5)
   no 77(87.5) IV 14 (15.9)
   yes 11(12.5) CA199 (U/ml) 1829.2 (0.5-28411)
   HCV ≤ 39 29 (40.0)
   no 87(98.9) > 39 59 (60.0)
   total bilirubin (umol/L) 39.9 (5.9-420) 28-35 9 (10.2)
   ≤17.1 58 (65.9) > 35 79 (89.8)
   > 17.1 30 (34.1) AFP(ug/L) 20.3 (0.6-1091)
   albumin level (g/L) 41.5 (29.0-51.0) ≤25 83 (94.3)
   <28 0 (0.0) >25 5 (5.7)
   [149]Open in a new tab

Receiver Operating Characteristic (ROC) analysis of RP11-328K4.1,
hsa-miR-27a-3p and PROS1

   ROC curves indicated that RP11-328K4.1, hsa-miR-27a-3p and PROS1
   exhibited great diagnostic efficiency in ICC tumor tissues and nontumor
   tissues ([150]Figure 14A–[151]14C). The areas under the ROC curve
   (AUCs) of RP11-328K4.1 were 1.000, 0.802, and 1.000 in [152]GSE61850,
   [153]GSE103909, and TCGA, respectively. The AUCs of hsa-miR-27a-3p were
   0.965, 0.814, 0.748 and 1.000 in [154]GSE53870, [155]GSE53992,
   [156]GSE57555, and TCGA, respectively. The AUCs of PROS1 were 0.967,
   1.000, 0.852, and 1.000 in [157]GSE57555, [158]GSE61850, [159]GSE103909
   and TCGA, respectively

Figure 14.

   [160]Figure 14
   [161]Open in a new tab

   (A–C) ROC analysis of RP11-328K4.1, hsa-miR-27a-3p, and PROS1 in ICC
   tumor tissues and matched adjacent nontumor tissues. (D) ROC analysis
   of RP11-328K4.1, hsa-miR-27a-3p, and PROS1 in early ICC and advanced
   ICC.

   When comparing different clinical stages of ICC, hsa-miR-27a-3p was a
   promising biomarker with an AUC of 0.821. However, the AUC of
   RP11-328K4.1 was 0.670, and the AUC of PROS1 was 0.554, which suggested
   that they have limited diagnostic utility ([162]Figure 14D).

DISCUSSION

   ICC, the second most common malignant hepatic tumor, second only to
   hepatocellular carcinoma (HCC). Although ICC is far less common than
   extra-cholangiocarcinoma (ECC), the morbidity and mortality rates of
   ICC have been increasing for the last 10 to 20 years [[163]3, [164]32,
   [165]33]. Therefore, increasing attention has been paid to the
   pathogenesis and prognosis of ICC [[166]34]. In recent years, the roles
   of ncRNAs in the pathogenesis and progression of tumors have become
   increasingly important. The ceRNA hypothesis, which was proposed in
   2011, is considered to be a landmark in understanding the mutual
   regulatory relationship and interactions of RNA-RNA in their entirety.
   Accumulated evidence suggests that the dysregulation of ceRNA
   interactions and ceRNETs is involved in the pathogenesis, progression
   and prognosis of a variety of cancers, including HCC [[167]31] and CCA
   [[168]29, [169]35]. Of note, ICC is not only significantly different
   from HCC in terms of etiology, pathogenesis and invasion and metastasis
   mode but is also significantly distinct from pCCA and dCCA (another two
   types of CCA) in terms of anatomical location, pathogenesis and
   prognosis. Therefore, it is of great importance to further use the
   ceRNET theory to study the pathogenesis of ICC, to explore the key
   regulatory pathways causing ICC and to screen potential molecular
   biomarkers for optimizing individualized and precise therapy of ICC.

   Based on the ceRNET theory and the GEO microarray database, for the
   first time, we constructed ICC-related ceRNETs by using a
   bioinformatics method, subsequently screened the core regulatory
   pathway related to the pathogenesis of
   ICC:RP11-328K4.1-hsa-miR-27a-3p-PROS1, and finally conducted
   preliminary experimental validation of the expression levels,
   expression trends and regulatory relationships of the screened ceRNAs
   of this core regulatory pathway by using molecular experiments. In our
   previous bioinformatic analysis, the expression of lncRNA RP11-328K4.1
   and PROS1 mRNA was downregulated in cancer tissues compared to that in
   adjacent normal tissues in ICC, while the expression of miRNA
   hsa-miR-27a-3p was upregulated in ICC cancer tissues, which was
   consistent with the mechanism of action and expression trends between
   ceRNA and miRNA in the ceRNA hypothesis. Our subsequent qRT-PCR
   validation in tissue and plasma also revealed low expression of lncRNA
   RP11-328K4.1 and PROS1 mRNA but high expression of miRNA hsa-miR-27a-3p
   in cancer tissue and peripheral plasma compared to the levels observed
   in adjacent normal tissue and healthy human peripheral plasma. These
   results suggest that the upregulation of lncRNA RP11-328K4.1 could
   eliminate the inhibited expression of PROS1 mRNA by oncogenic miRNA
   hsa-miR-27a through sponge adsorption. Therefore, lncRNA RP11-328K4.1
   could exert its role as a tumor suppressor gene in ICC. Meanwhile, the
   expression of the protein corresponding to PROS1 mRNA was preliminarily
   validated in the fresh tissue specimens, peripheral plasma and paraffin
   specimens of ICC patients. In addition, to clinically validate the
   differential diagnostic ability of the RNAs in this core regulatory
   pathway of the ICC-related ceRNET, not only in distinguishing ICC tumor
   tissues and matched adjacent nontumor tissues but also in identifying
   ICC of different stages, ROC analysis was performed by utilizing data
   from the NCBI GEO database and the TCGA database. ROC analysis revealed
   that elements of the core regulatory pathway, including RP11-328K4.1,
   hsa-miR-27a-3p, and PROS1, might play important roles in ICC diagnosis.
   Specifically, hsa-miR-27a-3p might have significant diagnostic value in
   identifying ICC of different clinical stages, while RP11-328K4.1 and
   PROS1 does not.

   The ceRNET constructed in this study contains 340 lncRNA-miRNA-mRNA
   regulatory relationships. Functional enrichment analysis of 40 DE mRNAs
   in this regulatory network revealed that this regulatory network is
   primarily associated with the regulation of proliferation of epithelial
   cells and activated T cells, which might play a role via the complement
   and coagulation cascade pathways in ICC. ICC is a malignant tumor
   derived from the bile duct epithelium at the proximal end of the
   secondary branch of the intrahepatic bile duct [[170]36]. The
   proliferation of biliary epithelial cells promotes CCA progression.
   Studies have shown [[171]37] that the proliferation capacity of
   activated T and T lymphocyte-mediated killing activity are
   significantly decreased in patients with malignant tumors. It is
   conceived that the occurrence of tumors can activate the antitumor
   immune response of activated T lymphocytes. However, tumor cells and
   their metabolites will block host T lymphocytes in response to tumor
   progression, thereby leading to dysregulated metabolism of T
   lymphocytes, weakened immune response against tumor antigens, hindered
   differentiation of effector cells and suppressed tumor immunity. In
   addition, their study indicates that the active proliferative response
   of T lymphocytes stimulated by antigens is an important part of
   mediating the effects and roles of cellular immunity in vivo.
   Meanwhile, bioinformatics analysis of proteomics and genomics regarding
   differentially expressed genes in prostate cancer also shows [[172]38]
   that the complement and coagulation cascades are involved in the
   pathogenesis of prostate cancer. However, the specific mechanism of its
   action in cancer has not yet been reported, but deserves continuous
   attention and in-depth discussion in the future.

   At present, there are few studies on the expression level, diagnostic
   and prognostic value of lncRNA RP11-328K4.1 in malignant tumors. In a
   Chinese study on the expression of lncRNA in gastric cancer and its
   prognostic value by the Center for Gastric Cancer Diagnosis and
   Treatment of Sun Yat-sen University, Chen W et al. [[173]39]
   investigated the expression and prognostic value of lncRNAs in gastric
   cancer tissues. In this study, the expression of lncRNA RP11-328K4.1
   was downregulated in gastric cancer tissues compared to normal tissues,
   which was consistent with the bioinformatic analysis and experimental
   validation in our study. In addition, their subsequent survival
   analysis showed that lncRNA RP11-328K4.1 was a protective factor for
   the prognosis of gastric cancer patients; gastric cancer patients with
   high expression of lncRNA RP11-328K4.1 had a better prognosis.
   Therefore, they suggested that lncRNA RP11-328K4.1 is expected to
   become a new targeted therapeutic target and prognostic molecular
   marker for gastric cancer. Similarly, in the present study, we found
   that the expression of lncRNA RP11-328K4.1 was decreased in the cancer
   tissue and peripheral plasma of ICC patients, suggesting that it may be
   a protective factor for the prognosis of ICC patients that plays a role
   as a tumor-suppressor gene. However, the mechanism of how lncRNA
   RP11-328K4.1 acts as a tumor suppressor gene in gastric cancer is not
   mentioned in the study by Chen W et al. [[174]39]. In our study, the
   RP11-328K4.1-hsa-miR-27a-3p-PROS1 regulatory pathway in ICC was
   detected and can provide ideas and references for the mechanism of