Abstract

Purpose

   Ulcerative colitis (UC) patients have an increased risk of colorectal
   cancer (CRC), and compared with sporadic CRC, ulcerative
   colitis-associated colorectal cancer (CAC) is more aggressive with a
   worse prognosis. This study aimed to identify a gene signature to
   predict the risk of CAC for patients with UC in remission.

Patients and Methods

   Series of quiescent UC-related transcriptome data obtained from the
   Gene Expression Omnibus (GEO) data set were divided into a training set
   and a validation set. Gene Set Variation Analysis (GSVA), Gene Set
   Enrichment Analysis (GSEA), and \Weighted Correlation Network Analysis
   (WGCNA) combined with protein–protein interaction (PPI) analysis were
   used to identify the pathways and gene signatures related to
   tumorigenesis among quiescent UC patients. A generalized linear model
   (GLM) of Poisson regression based on the training set was applied to
   estimate the diagnostic power of the gene signature in our validation
   set.

Results

   The tumor necrosis factor (TNF) signaling via NF-κB pathway was
   significantly augmented with the highest normalized enrichment score
   (NES). The genes in the brown module from WGCNA have shown a
   significant correlation with CAC (Pearson coefficient = 0.83, p =
   6e-06). A subset of NF-κB related genes (FOS, CCL4, CXCL1, MYC, CEBPB,
   ATF3, and JUNB) were identified with a relatively higher expression
   level in CAC samples. The diagnostic value of this 7-gene biomarker was
   estimated by the receiver operating characteristic (ROC) curve with an
   area under the ROC curve (AUC) at 0.82 (p<0.0001, 95% CI:
   0.7098–0.9400) in the validation cohort.

Conclusion

   In summary, the increased expression of this seven-NF-κB-related gene
   signature may act as a powerful index for tumorigenesis prediction
   among patients with UC in remission.

   Keywords: ulcerative colitis; UC, remission, colitis-associated
   colorectal cancer; CAC, diagnostic model

Introduction

   Ulcerative colitis (UC) is a kind of chronic idiopathic intestinal
   inflammatory disease with poorly understood etiology, mainly affecting
   the epithelial mucosa from the anus to the ileocecal area.[32]^1 It has
   been demonstrated that the extent of colonic involvement, severity of
   the inflammation, and duration of UC could increase the risk of
   colorectal cancer (CRC).[33]^2^,[34]^3 Patients with inflammatory bowel
   disease (IBD) often suffer from CRC within the first 7 years after
   initial diagnosis.[35]^4 Moreover, compared with sporadic CRC,
   ulcerative colitis-associated colorectal cancer (CAC) is more
   aggressive and has a worse prognosis as evidenced by the multiple
   involved sites,[36]^5 and hence surveillance of CAC by colonoscopy is
   recommended among the UC patients during the remission state.[37]^5
   Whereas, a proportion of dysplastic lesions cannot be detected exactly
   by surveillance colonoscopy.[38]^6 The researches on the assessment of
   the risk of CRC occurring between each two scheduled surveillance
   procedures are still limited.[39]^7 Therefore, this calls for further
   development of methods to predict and detect UC patients with high a
   risk of CAC at an earlier stage to improve the clinical outcomes during
   the scheduled surveillance procedures.

   The progression of CAC from UC including the development from low-grade
   dysplasia, high-grade dysplasia, to invasive adenocarcinoma, is
   reported to have a dynamic, heterogeneous, and complex communication
   between the immune system and cytokines.[40]^8 Various immunological
   and inflammatory pathways including PI3K-Akt signaling, tumor necrosis
   factor (TNF) signaling, cytokine-cytokine receptor interaction, and
   extracellular matrix (ECM)-receptor interaction pathways have been
   confirmed to orchestrate the fate of tumorigenesis and
   progression.[41]^9–11 Moreover, some of the candidate biomarkers were
   demonstrated to be able to promote colonic tumorigenesis by regulating
   the immune system, such as CXCL10 and CXCL9,[42]^12 IDO1,[43]^13
   CCR7,[44]^14 VCAM1[45]^15 as well as ICAM1.[46]^16

   Considering the comprehensive biological process, we have compared the
   differentially expressed genes in intestinal epithelium biopsy tissues
   among patients with UC in remission or UC with remote neoplasm and
   normal individuals from the Gene Expression Omnibus (GEO) database. A
   group of 7 nuclear factor-kappa B (NF-κB) related genes were generated
   with up-regulated expression among UC with remote neoplasm tissues and
   then was validated to be an effective signature for discrimination of
   UC patients with a high risk of CAC.

Patients and Methods

Patients and Samples

   In our study, both the discovery cohort and the validation cohort were
   based on patients with UC in remission exclusively. The patients with
   active UC and Crohn’s disease were not included. The discovery cohort
   consisted of 20 patients from Chicago including 5 normal controls, 4 UC
   patients in remission, and 11 patients with remote neoplasia.[47]^17
   The total RNA of the 20 samples extracted from the colonic mucosae was
   analyzed via the Affymetrix Human U133p2 platform (GLP570) and the
   normalized microarray data was obtained from GEO
   ([48]https://www.ncbi.nlm.nih.gov/geo/) repository via the accession
   number of [49]GSE37283. The validation cohort used in our study was
   composed of 41 normal controls, 26 UC patients in remission, as well as
   15 patients with CRC derived from the other 3 GEO data sets
   ([50]GSE13367,[51]^18 [52]GSE38713,[53]^19 and [54]GSE4183[55]^20), in
   which the normalized microarray data generated from colonic mucosae of
   these 82 samples was also analyzed through the GLP570 platform. All of
   the samples in both discovery cohort and validation cohort were
   obtained with the ethical approval from their original
   institutions.[56]^17–20

Profiling of RNA Differential Expression

   Both the normalized microarray data and the corresponding clinical
   features were downloaded from the GEO database and statistical analyses
   were performed in the R platform (version 3.6.2). Differentially
   expressed genes were identified by the Limma package[57]^21 from the
   comparison among the normal control, UC in remission, CAC, and CRC
   groups. To merge the microarray data from these 3 different GEO data
   sets for our validation cohort, the existed bias and variation due to
   the batch effect of the different high-throughput data were removed by
   the Combat function of the sva package[58]^22 in the R platform.

Gene Set Enrichment Analyses

   To investigate the variation of the Hallmark gene set enrichment among
   each sample in our discovery cohort (downloaded from Molecular
   Signatures Database (MSigDB);
   [59]https://www.gsea-msigdb.org/gsea/msigdb/index.jsp), Gene Set
   variation Analysis (GSVA) was conducted by the GSVA package in R
   platform to calculate the sample-gene set enrichment scores,[60]^23
   which were then visualized in a heatmap by the pheatmap package.
   Furthermore, Gene Set Enrichment Analysis (GSEA) was carried out by the
   ClusterProfiler package[61]^24 to identify the core genes in some
   significantly enriched pathways based on the Hallmark gene set
   associated with CAC. Significantly enriched biological pathways were
   identified if their normalized enrichment scores (NES) were ≥2 or ≤-2
   as well as false discovery rates (FDR) <0.05 after 1000-time
   permutations.

Immune Cell Infiltration

   The assessment of immune infiltration of 28 immune cell types among the
   CAC, UC in remission, and normal control in our discovery cohort was
   implemented via single-sample gene set enrichment analysis (ssGSEA) by
   using the GSVA package in R.[62]^23 The identified genes matched for
   each immune cell type were downloaded from the recent published
   work,[63]^25 and the ssGSEA scores for each immune cell type were then
   visualized in a heatmap. Considering the potential effect of these 28
   immune cells on tumorigenesis, they were divided into 2 groups: an
   anti-tumor group and a pro-tumor group, by suppressing the immune
   system in the microenvironment.[64]^26

Weighted Correlation Network Analysis for the Key Module

   The undirected, weighted gene co-expression network was constructed to
   detect the most correlated cluster of genes with CAC by using the
   Weighted Correlation Network Analysis (WGCNA) in our discovery cohort.
   The top 5,000 genes according to their median absolute deviation were
   extracted from the normalized microarray expression data and used for
   WGCNA. The selection of soft threshold, construction of correlation
   networks based on gene expression, detection and selection of highly
   correlated hub genes in the significantly related modules to CAC,
   calculation of softConnectivity, intramodularConnectivity, and
   topological overlap measure (TOM) similarity, as well as visualization
   of the module structure and network connections were all implemented in
   the WGCNA package.[65]^27

Identification of Diagnostic Gene Signature for CAC Patients

   The overlapped genes including those enriched by GSEA and involved in
   the most significant module of WGCNA were illustrated in a Venn diagram
   from an online tool
   ([66]http://bioinformatics.psb.ugent.be/webtools/Venn). The
   protein–protein interaction (PPI) information of these common genes was
   then integrated based on the STRING (11.0) online database
   ([67]https://string-db.org/)[68]^28 with the confidence score > 0.7
   between each 2 nodes, and 3 clusters were identified by K-means
   clustering algorithm. After removing the isolated and lower connected
   nodes, the achieved PPI network was further analyzed for hub genes by
   CytoHubba plugin[69]^29 and visualized in the Cytoscape
   software.[70]^30 The normalized expression of the selected hub genes in
   both training set and validation set was illustrated in boxplots by
   ggplot2 package in R platform.

Validation of the Diagnostic Signature for UC Patients with a High Risk of
Colorectal Carcinoma

   A generalized linear model (GLM) with Poisson regression based on our
   training set was employed to evaluate the diagnostic power of the
   identified signature. The predicted diagnostic scores were generated
   from the weighted linear diagnostic model combined with the gene
   expression in our validation cohort by the predict function in R
   software. The predicted risk score = expression of
   gene[1]*β1+expression of gene[2]*β2+expression of gene[n] *βn. The
   receiver operating characteristic (ROC) curve plotted by ROCR
   package[71]^31 in R platform combined with the corresponding area under
   the ROC curve (AUC) was applied to quantify the accuracy of the gene
   signature in the predicted model. A two-tailed P value < 0.05 with 95%
   confidence intervals (CI) was supposed to be significant.

Results

Profiling of Microarray Data for CAC in the Training Set

   Samples of UC patients in remission (quiescent UC), CAC patients, and
   normal controls were collected from [72]GSE37283 as our discovery data
   set. Considering the fact that the extent, duration, as well as the
   severity of inflammation on the intestinal epithelial cells may
   increase the risk for CAC,[73]^17 we first investigated the inherent
   heterogeneity of the microarray-based transcriptomic profiles of these
   samples based on 50 pathways of the Hallmark gene set by GSVA. The
   hierarchical clustering diagram of the 50 gene sets variation analysis
   showed that the pathways on the regulation of immune system such as
   interferon alpha/gamma response, allograft rejection, IL6-JAK-STAT3
   signaling, IL2-STAT5 signaling, TNF signaling via NF-κB displayed a
   strong similarity with higher expression in CAC patients compared with
   those in other 2 groups ([74]Figure 1A). Then, the ssGSEA was carried
   out to quantify the relative abundance of immune cell infiltration
   among these patients. Visualized in the heatmap, we observed that UC
   patients with remote neoplasia could be distinguished from the other 2
   groups of patients remarkably ([75]Figure 1B). According to the
   potential function on tumorigenesis, these 28 immune cell populations
   were divided into 2 clusters: anti-tumor immunity (activated CD4 T cell
   (ActCD4), ActCD8, central memory CD4 T cell (TcmCD4), TcmCD8, effector
   memory CD4 T cell (TemCD4), TemCD8, Th1, Th17, activated dendritic cell
   (ActDC), CD56 bright natural kill cell (CD56briNK), natural kill T cell
   (NKT), NK) and pro-tumor reactivity (Treg, Th2, CD56dimNK, immature
   dendritic cell (imDC), tumor-associated macrophage (TAM),
   myeloid-derived suppressor cells (MDSCs), Neutrophil, and plasmacytoid
   dendritic cell (pDC)) by suppressing the immune system.[76]^26 As
   depicted in the correlation diagram ([77]Figure 1C), the relative
   abundance of anti-tumor immune cell types of these samples was
   positively related to the pro-tumor ones, especially for CAC samples (R
   = 0.9676, p = 3.2966e-12). This highly positive correlation may suggest
   that the enhancement of immune suppression could be fed back by the
   facilitation of anti-tumor immunity.[78]^26 The presence of
   inflammation caused by UC may have an effect on tumorigenesis for CAC.

Figure 1.

   [79]Figure 1
   [80]Open in a new tab

   Heterogeneity of transcriptomic profiles of samples from patients with
   quiescent ulcerative colitis, ulcerative colitis with neoplasia, and
   normal control. (A) Heatmap with clusters displaying the results of
   GSVA on the Hallmark gene set among the three groups. (B) The
   heterogeneity of immune cell types for the three groups calculated by
   ssGSEA. (C) The correlation of immune cell populations of the three
   groups between anti-tumor clusters (ActCD4, ActCD8, TcmCD4, TcmCD8,
   TemCD4, TemCD8, Th1, Th17, ActDC, CD56briNK, NK, NKT) and pro-tumor
   clusters by suppressing the immune system (Treg, Th2, CD56dimNK, imDC,
   TAM, MDSC, Neutrophil, and pDC). R and the gray shaded area represent
   the coefficient of correlation and 95% CI, respectively.

   Abbreviations: GSVA, Gene set variation analysis; ssGSEA, single-sample
   gene set enrichment analysis; ActCD4, activated CD4 T cell; TcmCD4,
   central memory CD4 T cell; TemCD4, effector memory CD4 T cell; DC,
   dendritic cell; CD56birNK, CD56 bright natural kill cell; NK, natural
   kill cell; NKT, natural kill T cell; TAM, tumor-associated macrophage;
   MDSC, myeloid-derived suppressor cells; pDC, plasmacytoid dendritic
   cell; CI, confident interval.

Functional Pathways Enrichment Analysis

   To further explore the special mechanism of tumorigenesis among the
   quiescent UC patients, GSEA was used to compare the difference between
   CAC and UC in remission samples, between CAC and normal control
   samples, and between UC in remission and normal control samples based
   on the Hallmark gene set. Consistent with the result of GSVA, the TNF
   signaling via NF-κB pathway was significantly augmented with the
   highest NES in both CAC patients compared with UC (NES: 2.64, p =
   0.0016, p.adjust = 0.0064) as well as normal controls (NES:2.47, p =
   0.001, p.adjust = 0.0027) and quiescent UC patients compared with
   normal controls (NES:2.17, p = 0.001, p.adjust = 0.0018) ([81]Figure 2
   and [82]Supplementary Tables 1–3).

Figure 2.

   [83]Figure 2
   [84]Open in a new tab

   GSEA plots depicting the enrichment of signal pathways based on the
   Hallmark gene set. Ridge plots (left) of the most enriched 20 pathways
   of CAC patients relative to those with UC in remission (A) or normal
   patients (B), and patients with UC in remission compared with normal
   controls (C). The TNF signaling via NF-κB pathway is positively
   enriched with the highest NES (right).

   Abbreviations: GSEA, gene set enrichment analysis; CAC, ulcerative
   colitis-associated colorectal cancer; UC, ulcerative colitis; NES,
   normalized enrichment score.

Construction of Co-Expression Networks by WGCNA

   Co-expression networks among these 20 samples were established to
   figure out those genes with a high relationship to CAC. The top 5,000
   genes of these samples were clustered into 8 modules (MEblack, MEblue,
   MEgreen, MEred, MEyellow, MEbrown, MEturquoise, and MEgrey), of which
   genes in the MEbrown module were most positively related to CAC with
   significance ([85]Figure 3A and [86]B, Pearson coefficient = 0.83, p =
   6e-06). There were a total of 1,234 genes involved in the brown module
   ([87]Supplementary Table 4), and the correlational analysis of these
   genes between gene significance to CAC and the module membership was
   conducted. As shown in [88]Figure 3C, these 1,234 genes presented a
   significant contribution to both CAC and the brown module membership
   (correlation coefficient = 0.85, p<1e-200). Besides, the hierarchical
   clustering analysis indicated that these genes in the CAC group also
   expressed differently from the other 2 groups ([89]Figure 3D).
   Therefore, these genes co-expressed with CAC should be paid attention
   to.

Figure 3.

   [90]Figure 3
   [91]Open in a new tab

   Correlated genes in the interested modules of CAC identified by WGCNA.
   (A) Heatmap depicting the strength of relationships between each module
   and the clinical features. The ρ coefficients as well as corresponding
   p values were exhibited in various shade of colors. The positive
   correlation was exhibited in red and the negative was in blue. (B) The
   dendrogram (top) and heatmap (bottom) displaying the strength of
   correlations between CAC and other modules in which the red represented
   a higher positive adjacency and the blue represented a lower one. (C)
   The scatterplot of the correlation between gene significance (y-axis)
   and the selected module (brown) membership for CAC. In this brown
   module, genes were highly correlated with both the selected module and
   the clinical feature of CAC. (D) Heatmap with clusters of
   differentially expressed genes in brown module among the 3 different
   groups.

   Abbreviations: CAC, ulcerative colitis-associated colorectal cancer;
   WGCNA, weighted correlation network analysis.

Identification of Hub Genes Involved in CAC in the Discovery Cohort

   Having found that the TNF signaling via NF-κB pathway was positively
   enriched both CAC and UC patients, 81 NF-κB related genes were commonly
   upregulated in CAC patients compared with quiescent UC patients and
   normal controls ([92]Figure 4A, the common part of the red and green
   channels). Additionally, considering the NF-κB related genes also
   enriched in quiescent UC patients compared with normal controls, there
   were 56 NF-κB related genes commonly increased in both the CAC group
   and the quiescent UC group ([93]Figure 4A). To further discriminate the
   NF-κB related genes with a closer correlation with CAC, genes from
   WGCNA were taken into account and a total of 32 NF-κB related hub genes
   were found ([94]Figure 4A). The identified genes were then submitted to
   the STRING online database to map a PPI network and divided into 3
   clusters by K-means clustering algorithm ([95]Figure 4B). 23 nodes with
   high interaction score (confidence > 0.7) were discriminated and
   further analyzed by 12 topological algorithms including Degree, Edge
   Percolated Component (EPC), Maximum Neighborhood Component (MNC),
   Density of Maximum Neighborhood Component (DMNC), Maximal Clique
   Centrality (MCC) and centralities based on shortest paths, such as
   Bottleneck (BN), EcCentricity, Closeness, Radiality, Betweenness, and
   Stress in CytoHubba plugin[96]^29 of Cytoscape software ([97]Table 1).
   Finally, 7 NF-κB related hub genes were identified as follows with a
   relatively higher expression level in CAC patients: Fos proto-oncogene
   (FOS), C-C motif chemokine ligand 4 (CCL4), C-X-C motif chemokine
   ligand 1 (CXCL1), MYC proto-oncogene (MYC), CCAAT enhancer binding
   protein beta (CEBPB), activating transcription factor 3 (ATF3), and
   JunB proto-oncogene (JUNB) ([98]Figure 4C and [99]Figure 5A).

Figure 4.

   [100]Figure 4
   [101]Open in a new tab

   Identification of hub genes that may be involved in the tumorigenesis
   among the ulcerative colitis patients. (A) The Venn diagram exhibiting
   the overlaps of genes from the results of GSEA and WGCNA. Core genes
   enriched in the TNF signaling via NF-κB pathway by GSEA were picked out
   (the green part: CAC compared with UC in remission; the red part: CAC
   compared with normal control; and the purple part: UC in remission
   compared with normal control). The yellow part represented the genes in
   the brown module most associated with CAC analyzed by WGCNA. (B) PPI
   network of the common 32 NF-κB signaling genes analyzed by STRING
   database was shown in 3 clusters via K-means clustering algorithm. The
   different colors of spots represent the different clusters. (C) PPI
   network of the hub genes with high confidence (interaction score >0.7)
   extracted from (B), calculated by CytoHubba, and visualized in
   Cytoscape software. The various sizes as well as the colors (from blue
   to orange) of dots represented the degree between two genes and the
   coefficients were exhibited by the different sizes and colors of the
   edges.

   Abbreviations: UC, ulcerative colitis; CAC, ulcerative
   colitis-associated colorectal cancer; WGCNA, weighted correlation
   network analysis; PPI, protein–protein interaction.

Table 1.

   The Top 8 Genes Calculated by CytoHubba
   Category Several Topological Algorithms in CytoHubba
   Betweenness Closeness Clustering Coefficient Degree DMNC EcCentricity
   EPC MCC MNC Radiality Stress BN
   Gene symbols FOS FOS ZFP36 FOS ATF3 CCL4 FOS FOS FOS FOS FOS FOS
   CCL4 ATF3 GPR183 ATF3 CEBPB CXCL1 ATF3 ATF3 ATF3 CCL4 CCL4 CCL4
   CXCL1 CCL4 CEBPB MYC JUNB GPR183 CEBPB CEBPB CEBPB ATF3 CXCL1 CXCL1
   ATF3 MYC JUNB CEBPB MYC FOS JUNB JUNB JUNB CEBPB ATF3 ATF3
   MYC CEBPB CCL4 JUNB CCL4 ATF3 MYC MYC MYC MYC MYC MYC
   ICAM1 JUNB CXCL1 CCL4 CXCL1 CEBPB CCL4 CCL4 CCL4 JUNB ICAM1 ICAM1
   CEBPB CXCL1 ATF3 CXCL1 ZFP36 JUNB ZFP36 CXCL1 CXCL1 CXCL1 CEBPB CEBPB
   JUNB ZFP36 FOS ICAM1 GPR183 MYC CXCL1 ICAM1 ZFP36 ZFP36 JUNB JUNB
   [102]Open in a new tab

   Abbreviations: EPC, Edge Percolated Component; MNC, Maximum
   Neighborhood Component; DMNC, Density of Maximum Neighborhood
   Component; MCC, Maximal Clique Centrality; BN, Bottleneck.

Figure 5.

   [103]Figure 5
   [104]Open in a new tab

   The box plots depicting the 7 hub genes differentially expressed in
   CAC, UC in remission, and normal control groups in our training data
   set ((A) [105]GSE37283) and validation data set ((B) [106]GSE4183 &
   [107]GSE13367 & [108]GSE38713). ^#P≥0.1, ^●P<0.1, * P<0.05; ** P <0.01;
   *** P <0.001.

   Abbreviations: UC, ulcerative colitis; CAC, ulcerative
   colitis-associated colorectal cancer.

Performance of the Diagnostic Seven-Gene Signature in the Validation Cohort

   In order to assess the strength of the seven-NF-κB-related gene
   signature in predicting tumorigenesis among quiescent UC patients, the
   transcriptomic data of 82 patients were integrated as our validation
   cohort, including 41 normal controls (20 cases in [109]GSE13367; 13
   cases in [110]GSE38713; 8 cases in [111]GSE4183), 15 patients with CRC
   ([112]GSE4183), and 26 quiescent UC patients (18 patients in
   [113]GSE13367; 8 patients in [114]GSE38713). Consistent with the
   expression levels observed in our training set, these 7 NF-κB related
   genes were also expressed at a higher level in CRC samples, suggestive
   of the potential role in tumorigenesis of colorectal carcinoma
   ([115]Figure 5B). Moreover, gene functional analysis also indicated
   that the TNF signaling via NF-κB pathway was highly augmented in CRC
   samples compared with quiescent UC patients or normal intestinal
   mucosae significantly ([116]Figure 6A; [117]Supplementary Table 5 and
   6).

Figure 6.

   [118]Figure 6
   [119]Open in a new tab

   (A) GSEA plots depicting the TNF signaling via NF-κB pathway which was
   positively enriched in CRC patients in our validation cohort compared
   with patients with UC in remission (top) and normal patients (bottom).
   (B) The ROC curve with AUC at 0.82 for the diagnosis value of the
   7-NF-κB-signaling-gene signature in detecting UC patients with high
   risk for CRC (p<0.0001, 95% CI: 0.7098–0.9400).

   Abbreviations: GSEA, gene set enrichment analysis; UC, ulcerative
   colitis; CRC, colorectal cancer; ROC, receiver operating
   characteristic; AUC, area under curve.

   According to the clinical feature of our discovery cohort, a GLM with
   the Poisson regression was then applied to our validation cohort. The
   ROC curve with an AUC at 0.82 (p<0.0001, 95% CI: 0.7098–0.9400)
   demonstrated the potential ability of diagnosis for quiescent UC
   patients with a high risk of colorectal carcinoma ([120]Figure 6B).

Discussion

   The elucidation of pathogenesis of CAC remains an unmet need to develop
   new surveillance tools or therapeutic targets for patients with both UC
   and CAC. The consistent action of inflammatory on intestinal mucosa has
   been elucidated to predispose UC to CAC by multiple tumor-related
   genetic changes in intestinal epithelial cells.[121]^32^,[122]^33
   Watanabe et al generated 20 discriminator genes to predict the
   development of CAC among the UC patients, including the cancer-related
   genes CYP27B1, RUNX3, SAMSN1, EDIL3, NOL3, CXCL9, ITGB2, and LYN.
   Nevertheless, they did not distinguish the active UC patients from
   those in remission.[123]^34 Besides, the previous study based on the
   same population as our discovery cohort has identified nine highly
   expressed genes in UC patients with neoplasia compared to quiescent UC
   patients and healthy controls, which might regulate the immune
   function, inflammation, proliferation, and apoptosis of the intestinal
   mucosa cells and contribute to neoplastic transformation,[124]^17 but
   without validation in a large clinical cohort. Furthermore, Shi’s group
   integrated 8 gene profiles and re-analyzed the differentially expressed
   genes between patients with quiescent UC and active UC, UC and adenoma,
   as well as UC and CRC. A group of hub genes (CXCL10, VCAM1, CXCL9,
   MMP9, IDO1, and CCR7) were found to be significantly associated with
   tumorigenic processes via pathways such as platelet activation,
   ligand-receptor interaction, immune dysregulation, and
   inflammation.[125]^35 Among the involved pathways, the aberrant
   stimulation of NF-κB have been detected in intestinal epithelial cells
   from both IBD and CRC,[126]^36 and can contribute to
   inflammation-associated tissue injury[127]^37 as well as tumor cell
   proliferation via the overexpression of angiogenesis-related
   genes.[128]^33 In our present study, we have identified a
   7-NF-κB-related gene signature (FOS, CCL4, CXCL1, MYC, CEBPB, ATF3, and
   JUNB) which was potentially a tool for the selection of patients UC in
   remission stage with a high risk of cancerization during the scheduled
   surveillance procedures.

   Considering the crucial role in UC and CAC, there is no doubt that the
   TNF signaling via NF-κB pathway could be highly enriched in CAC samples
   with significance in our study. The activation of the NF-κB pathway
   upregulates the expression of pro-inflammatory cytokines and adhesion
   factors, inhibits the production of apoptotic factors, and promotes
   tumor angiogenesis and metastasis.[129]^33^,[130]^38 In azoxymethane
   (AOM) followed dextran sulfate sodium (DSS) induced CAC mice models,
   the deletion of IκB kinase β (IKKβ) led to inactivation of the NF-κB
   pathway and thereby reduced the incidence of tumors, which proved the
   stimulation on tumor progression.[131]^39 On the other hand, the
   activation of the NF-κB pathway promotes the production of TNF-α, which
   also plays an important role in tumorigenesis. The expression of TNF-α
   was observed to be significantly increased in AOM/DSS CAC models, and
   blocking TNF-α obviously reduced tumors amount and lesion as well as
   reduced colonic infiltration by neutrophils and macrophages.[132]^40
   The higher expression level of these identified genes in the CAC group
   in our discovery and validation cohort might point out the potential
   importance in tumorigenesis during the development from quiescent UC to
   CAC. Indeed, interrogation of extra-data which was composed of the
   normal intestinal mucosa, UC in remission stage, and CRC also revealed
   an outstanding diagnostic power of this identified gene signature.
   Nevertheless, given the limited size of the training set and the
   non-coherent sample features of our validation set, the
   seven-NF-κB-related gene signature needs to be further validated in
   large scale cohorts of UC and CAC patients, and it may serve as a
   useful tool for predicting and identifying UC patients in remission
   stage with a high risk of CAC after the routine therapy.

   Among the 7 NF-κB related genes, CCL4 and CXCL1 encode members of
   chemotactic cytokines families which are small secreted proteins and
   are known to control immune cell migration. Besides, they also
   participate in pathological processes such as tumor occurrence,
   development, and metastasis.[133]^41 MYC is a type of proto-oncogenes
   encoding nuclear transcriptions which regulates cell cycles. The
   inhibition of TNF α-induced c-Myc upregulation can be mediated by
   attenuating NF-κB signaling,[134]^42 which can then induce uncontrolled
   cell proliferation and inhibit apoptosis of colorectal
   carcinoma.[135]^43 ATF3 is a stress-responsive factor that plays a
   vital role in controlling the expression of cell-cycle regulators and
   tumor suppressor.[136]^44 Intriguingly, it has been reported that ATF3
   played a protective role in regulating the gut follicular helper T
   cells among IBD patients[137]^45 and inhibiting the invasion of
   CRC,[138]^46 which may need further investigation. CEBPB encodes a
   member of the CCAAT/enhancer-binding protein family, which can be
   activated by the NF-κB and STATS pathways in the inflammatory
   microenvironment to regulate gene transcription in response to IL-1 and
   IL-6.[139]^47 The deletion of CEBPB would impair the function of
   regular T cells and thus aggravate the T cell-mediated colitis.[140]^48
   In contrast, the overexpression of CEBPB may promote tumor cell
   invasion in CRC individuals,[141]^49 and regulate the immunosuppressive
   environment via MDSCs in CAC.[142]^50 The JUNB and FOS encode proteins
   which belongs to the dimeric transcription factor activator protein
   1(AP-1), play roles in regulating tumor cell proliferation, survival,
   differentiation, invasiveness, or angiogenesis.[143]^51 Regardless of
   the controversial perspective over these NF-κB related genes, our
   results from discovery and validation cohorts both indicated the
   potential effect on tumorigenesis from UC in the remission stage.

   Nevertheless, there are also some limitations in our study. In terms of
   the patients in our discovery cohort, the detailed clinical features
   including the patients’ demographics, the treatment of these patients,
   and the grade of CAC were not assessable. Therefore, we cannot exclude
   the confounding factors and make a further investigation on the process
   of tumor transformation from quiescent UC to CAC. Moreover, unlike the
   discovery cohort, patients in our validation cohort were collected from
   3 different institutions, which could lead to bias and variation due to
   the batch effect inevitably. Considering the heterogeneity of
   intestinal mucosa derived from these different institutions, further
   researches based on the comparison between quiescent UC patients
   harboring or not CAC after receiving the unified regimen will be more
   instrumental to delineate the tumorigenesis function of the seven-gene
   signature in NF-κB pathway.

Conclusion

   In summary, we have identified a 7-NF-κB-related gene signature (CCL4,
   CXCL1, MYC, JUNB, FOS, ATF3, and CEBPB) with high expression in CAC,
   which may facilitate the tumorigenesis by regulating the immunological
   and inflammatory reactions via the NF-κB pathway. High levels of this
   signature may act as a powerful predictor for CAC among quiescent UC
   patients during their scheduled surveillance.

Abbreviations

   UC, Ulcerative colitis; CRC, colorectal cancer; CAC, ulcerative
   colitis-associated colorectal cancer; TNF, tumor necrosis factor;
   NF-κB, nuclear factor-kappa B; GSEA, gene set enrichment analysis;
   ssGSEA, single-sample gene set enrichment analysis; GSVA, gene set
   variation analysis; DC, dendritic cell; NK, natural kill cell; NKT,
   natural kill T cell; TAM, tumor-associated macrophage; MDSC,
   myeloid-derived suppressor cells; pDC, plasmacytoid dendritic cell;
   PPI, protein–protein interaction; WGCNA, weighted correlation network
   analysis; NES, normalized enrichment score; GEO, Gene Expression
   Omnibus; IBD, inflammatory bowel disease; GLM, generalized linear
   model; ROC, receiver operating characteristic curve; AUC, area under
   curve; CI, confidence intervals.

Disclosure

   The authors report no conflicts of interest in this work.

References