Introduction

Multiple clinical observations as well as epidemiologic studies show the differences in the extent and severity of disease between pediatric-onset ulcerative colitis (UC) and its adult counterpart. Most studies on UC have demonstrated that up to 60–70% of pediatric-onset UC presents with a "panenteric" phenotype compared to 20–30% of adult-onset UC patients [1,2,3]. Within 10 years of diagnosis, more than 40% of pediatric-onset UC patients had undergone colectomy compared to less than 20% in adult-onset UC patients [2]. In addition to having a more severe phenotype with rapid progression of inflammatory disorders of the gastrointestinal tract, pediatric-onset UC patients are at higher risk of developing extra-intestinal manifestations (EIM) than adult-onset UC patients, which is predictive of more a severe disease outcome [4]. Cortisol is widely used for the treatment of adult-onset UC; however, cortisol is not recommended for use in pediatric-onset UC patients because of the high rate of refractory disease and drug dependence and detrimental effects on growth and bone density accrual [5, 6]. According to the first pediatric specific UC guidelines jointly produced by ECCO and ESPGHAN, indefinite maintenance therapy is necessary for the treatment of adult-onset UC patients, which combines high doses of 5-ASA with thiopurines or the anti-TNF biologic infliximab [7, 8]. Although several studies have focused on the differences in clinical characteristics and administration strategies between pediatric and adult UC, transcriptomic comparisons at the molecular level are few reported and the differences in gene expression, biological processes and pathways between pediatric and adult UC are still unknown [9,10,11].

The pathological tissue damage in UC is partly driven by exaggerated and inappropriate immune responses involving both the innate and adaptive immune systems. Activation or inactivation of various types of immune cells in UC contributes to disease initiation, augmentation, and perpetuation by regulating the suppression, maintenance or promotion of immune responses [12,13,14]. However, to our knowledge, no studies focusing on a comparison between the characteristics of immune cell infiltration in pediatric and adult UC have yet been reported.

A comparison of the gene expression profile between two subtypes of a disease is useful to identify reliable biomarkers and novel treatment targets for the improvement of therapeutic efficacy and personalized therapy [15,16,17,18,19,20]. Bioinformatics analysis of gene expression profiles is a promising tool for the comprehensive identification of the differentially expressed genes (DEGs) and biological pathways involved in the development of UC in pediatric and adult patients at the whole genome level. In our current study, gene microarray dataset GSE87473 was obtained from Gene Expression Omnibus (GEO), which contains 19 pediatric and 87 adult UC samples. We conducted differential gene expression analysis, followed by Gene Ontology (GO) and pathway enrichment analysis of DEGs. The protein–protein interaction (PPI) network of DEGs was then constructed and key modules were mined and functionally annotated. Furthermore, we investigated and compared the immune cell infiltration characteristics between pediatric and adult UC using a novel gene-based approach CIBERSORT to deconvolute gene expression data and determine 22 immune cell types in UC samples [21].

Materials and Methods

Microarray Data

The procedures used in our study are illustrated in the flow chart shown in Fig. 1. The mRNA microarray dataset GSE87473 deposited by Li et al. was downloaded from Gene Expression Omnibus (GEO: https://www.ncbi.nlm.nih.gov/geo/). The inclusion criteria for the gene expression profile are set as follows: (1) The samples detected are colon tissues from homo sapiens, (2) tissues are diagnosed with pediatric and adult UC tissues, (3) gene expression profiling of mRNA, (4) the number of pediatric or adult UC samples was more than 15. The database contained colon biopsies from 19 pediatric and 87 adult UC patients, and the GPL13158 [HT_HG-U133_Plus_PM] Affymetrix HT HG-U133+PM Array Plate was utilized to obtain gene expression profiles. As shown in Figure S1, the box plot of the distribution of value data is median-centered, indicating that the data are cross-comparable and appropriate for further investigations.

Fig. 1
figure 1

Flowchart of the study procedures

Full size image

Identification of Differentially Expressed Genes

The DEGs between pediatric and adult UC colon samples were screened using the interactive web tool GEO2R (https://www.ncbi.nlm.nih.gov/geo/geo2r/), which is capable of comparing and analyzing two sets of GEO array data [22]. P values < 0.05 and |log FC|> 2.5 were set as cut-off standards and considered to indicate statistical significance. The volcano plot of DEGs from GSE87473 were constructed using ImageGP (https://www.ehbio.com/ImageGP/index.php/Home/Index/Volcanoplot.html).

Gene Ontology (GO) and Pathway Enrichment Analysis

Gene Ontology (GO) analysis is a major tool for annotating genes, gene products and sequences. The Kyoto Encyclopedia of Genes and Genomes (KEGG) is an integrated database resource for biological interpretation of genome sequences and other high-throughput data to identify functional and metabolic pathways. Metascape (https://metascape.org) is an online program for gene functional enrichment analysis by analyzing gene/protein lists and for intuitive and visually representation of the results [23]. We used the "Custom Analysis" mode of Metascape to perform GO [including Biological process (BP), cellular component (CC) and molecular function (MF)] and KEGG pathway analysis. P < 0.01 was considered to indicate significantly enrichment of the DEGs.

PPI Networks and Hub Gene Definition

The PPI networks of DEGs were generated using the Search Tool for the Retrieval of Interacting Genes (STRING 11.0; https://string.embl.de/) database, with a confidence score > 0.9 (highest confidence) set as the cut-off criterion [24]. The PPI network was then visualized using Cytoscape 3.6.1 software and hub modules of the network were screened using the Molecular Complex Detection (MCODE) plug-in with MCODE scores > 10, with the degree of connectivity cut-off, node score cut-off, k-core, and maximum depth set as 2, 0.2, 2 and 100, respectively [25, 26]. In addition, using the NetworkAnalyzer, hub genes with a degree > 60 in the PPI network were defined as hub genes [27].

Analysis of Drug–Target Interactions

The drug–gene interaction database (DGIdb, v3.0.2, https://www.dgidb.org/) was applied to predict the drug–gene interactions using the list of hub DEGs with the filter condition ‘FDA Approved’ [28]. Drug–gene interaction was constructed and visualized by Cytoscape.

Immune cell Infiltration Evaluation and Analysis

CIBERSORT is an analytical tool used for sensitive and specific quantification of the relative levels of 22 phenotypes of human hematopoietic cells within a complex gene expression mixture (https://cibersort.stanford.edu) [21]. The CIBERSORT P value reflects the statistical significance of the deconvolution results across all cell subsets, which is used to filter out deconvolution with less significant fitting accuracy. Normalized gene expression profiles from 87 adult UC and 19 pediatric UC patients were uploaded to the CIBERSORT web portal (https://cibersort.stanford.edu), and the number of permutations was set to 500. Immune cell profiles of samples that met the CIBERSORT P < 0.05 requirements were then obtained.

Statistical Analysis

Two-tailed Student's t-test was used to analyze the differences between immune cell fractions of eligible adult and pediatric UC samples using GraphPad Prism 7.0 software (GraphPad Software, Inc.). Correlation analysis of immune cell proportions in adult and pediatric UC samples was performed using the OmicShare tools, which is a free online platform for data analysis (https://www.omicshare.com/tools). Correlation matrixes of adult and pediatric UC were then constructed.

Results

Identification of DEGs

A total of 19,779 genes were detected in colon biopsies from 19 pediatric and 87 adult UC patients, of which 1700 [694 (40.8%) upregulated and 1006 (59.2%) downregulated] were identified as DEGs in the pediatric UC samples compared to adult UC samples (Fig. 2). The most significantly upregulated and downregulated genes were folate receptor 1 (FOLR1) (logFC = 7.2786365) and transcription factor 4 (TCF4) (logFC = − 4.5852488), respectively (Table S1).

Fig. 2
figure 2

Volcano plot of DEGs selection. Red dots represent upregulated genes and green dots represent downregulated genes

Full size image

Enrichment Analysis of DEGs

GO and KEGG pathway enrichment analyses of overlapping DEGs showed a high level of enrichment in numerous functional groups and pathways. The 694 upregulated DEGs were mainly enriched in ‘anchoring junction’ (GO: 0070161), ‘cell junction organization’ (GO: 0034330), ‘regulation of actin cytoskeleton’ (hsa04810), ‘supramolecular fiber organization’ (GO: 0097435), ‘chromatin’ (GO: 0000785), and ‘cell adhesion molecule binding’ (GO: 0050839) (Fig. 3a, b). The 1006 commonly downregulated DEGs were mainly enriched in ‘regulation of mRNA metabolic process’ (GO: 1903311), ‘RNA localization’ (GO: 0006403), ‘ribonucleoprotein complex biogenesis’ (GO: 0022613), ‘RNA splicing via transesterification reactions’ (GO: 0000375), ‘regulation of cell morphogenesis’ (GO: 0022604), and ‘regulation of response to endoplasmic reticulum’ (GO: 1905897) (Fig. 3c, d).

Fig. 3
figure 3

Functional enrichment analysis of DEGs. a, c Top 20 clusters from Metascape pathway enrichment analysis of up- and downregulated DEGs colored according to P values. b, d Network of the top 20 enriched terms colored according to cluster ID. Nodes of the same color are typically close to each other

Full size image

PPI Network Analysis and Hub Gene Identification

The PPI network of DEGs was constructed and visualized using Cytoscape based on the STRING output with highest confidence and consisted of 893 nodes (genes) and 5143 edges (interactions) (Fig. 4). After using the MCODE plug-in to identify modules from the PPI network, the top three central modules with MCODE scores > 10 were selected. Module 1 with MCODE scores of 32.431 consisted of 66 nodes and 1054 edges. Module 2 with MCODE scores of 17.417 consisted of 49 nodes and 418 edges. Module 3 with MCODE scores of 13.111 consisted of 46 nodes and 295 edges (Fig. 5) (Table 1). Through analysis of GO-BP and KEGG pathway enrichment, Module 1 was found to be mainly enriched in ‘RNA splicing via transesterification reactions with bulged adenosine as the nucleophile’ (GO: 0000377) and ‘spliceosome’ (hsa03040). Module 2 was mainly enriched in ‘cell division’ (GO: 0051301) and ‘organelle localization’ (GO: 0051640), and Module 3 was mainly enriched in ‘regulation of megakaryocyte differentiation’ (GO: 0045652) and ‘post-translational protein modification’ (GO: 0043687) (Table 2).

Fig. 4
figure 4

The protein–protein interaction network of the common nodes. Up- and downregulated DEGs are marked in red and green, respectively, and the size of the node represents the degree of connectivity

Full size image
Fig. 5
figure 5

Modules that contain hub genes of the PPI network. Up- and downregulated DEGs are marked in red and green, respectively

Full size image
Table 1 Modules that contain hub genes of the PPI network
Full size table
Table 2 Top enriched GO-BP and KEGG pathways of DEGs in the modules that contain hub genes in the PPI network
Full size table

DEGs with a high degree of centrality (> 60) were then identified as hub nodes in this PPI network. These hub nodes were identified as cell division control protein 42 (CDC42; degree = 97), DNA-directed RNA polymerase II subunit RPB1 (POLR2A; degree = 84), Ras-related C3 botulinum toxin substrate 1 (RAC1; degree = 84), phosphatidylinositol 3-kinase regulatory subunit alpha (PIK3R1; degree = 71), mitogen-activated protein kinase 1 (MAPK1; degree = 69) and proto-oncogene tyrosine-protein kinase Src (SRC; degree = 66) (Table 3). Specifically, CDC42 was in Module 2, POLR2A was in Module 1, and MAPK1 and SRC were in Module 3. Besides, Module 8 that contains RAC1 and PIK3R1 and its GO-BP and KEGG pathway enrichment were also provided in Fig. 5, Tables 1 and 2.

Table 3 Top 6 hub genes with higher degree of connectivity
Full size table

Analysis of Drug–Target Interactions

Study of drug–target interaction is great important for drug discovery and design. Based on the hub DEGs in the PPI network, the drug–target interactions were predicted by the DGIdb, which provided a theoretical therapeutic mechanism of UC for reference (Fig. 6). Four hub DEGs including SRC, MAPK1, RAC1 and CDC42 were found theoretically important in the UC therapy and they were targeted by 15, 6, 2 and 1 FDA-approved drugs, respectively.

Fig. 6
figure 6

The drug–target interaction network of DEGs. Circle represents DEGs and square represents drug

Full size image

Immune Cell Infiltration Characterization

The 22 immune cell types infiltrating all pediatric and adult UC samples were investigated according to the CIBERSORT algorithm. Nine pediatric and 75 adult UC samples that matched the requirements of CIBERSORT P < 0.05 were filtered out (Table S2). As a result, 22 immune cell types were detected in adult UC samples, while 20 immune cell types were detected in pediatric UC samples in addition to naive CD4 T cells and eosinophils. The immunocyte composition of the entire cohort is summarized in Fig. 7a and the subpopulations of immune cells identified by unsupervised hierarchical clustering are shown in Fig. 7b. Correlation analysis of pediatric and adult UC samples furtherly revealed a low to moderate correlation between various immunocyte subpopulation fractions (Fig. 7c, d). In addition, relative to adult UC samples, higher proportions of naive B cells, resting memory T helper cells, regulatory (Tregs) T cells, monocytes, and M0 macrophages as well as activated mast cells were detected in pediatric UC samples, along with lower proportions of memory B cells, follicular helper T cells, γδ T cells, M2 macrophages, and activated dendritic cells (Fig. 7e).

Fig. 7
figure 7

Immune infiltrate landscape of pediatric and adult ulcerative colitis samples. a Stacked bar chart representing deviations in immune infiltration in each sample. b Hierarchical clustering of adult and pediatric UC samples based on immune cell proportions. Red, white and blue indicate high, moderate and low proportions of immune cells, respectively. c, d Correlation matrix of the immunocyte proportions in pediatric (c) and adult (d) UC samples. Red and blue color represent positive and negative correlations, respectively. e Differences in proportions of each immune cell type in pediatric and adult UC samples. *P < 0.05 **P < 0.01 ***P < 0.001 ****P < 0.0001

Full size image

Discussion

Pediatric-onset UC is more severe and extensive with a devastating prognosis compared with its adult counterpart; however, the genetic differences and characteristics of these two forms of the disease are still unknown. In the present study, we used genomic analysis to screen a total of 1700 DEGs from GSE87473 microarray dataset in the GEO. As was suggested by the GO and KEGG enrichment analysis, 694 upregulated DEGs were mainly manifested in inter-cellular items that relate to cell junctions, cell adhesion, actin cytoskeleton and transmembrane receptor signaling pathways. The 1006 downregulated DEGs were mainly manifested in intra-cellular items that relate to the splicing, metabolism and localization of RNA.

According to the PPI network of DEGs, six hub genes with high degrees of connectivity were identified: CDC42, POLR2A, RAC1, PIK3R1, MAPK1 and SRC. MAPK1 (also known as ERK2) is a member of the MAPK family, playing a major role in MAPK/ERK signaling, which promotes inflammatory responses and inflammatory cytokine production. The expression of (p)-ERK1/2 was increased in DSS-induced UC rats, and its inhibition reduced inflammation and promoted restoration of epithelial barrier function [29,30,31,32,33,34,35]. CDC42 and RAC1 are small GTPases of the Rho family that control the switch between the active GTP­bound and inactive GDP-bound forms. These molecules have similar biological characteristics that stabilize the intestinal epithelial barrier by promoting various cellular processes of intestinal epithelial cells (IECs) such as proliferation, adhesion, motility and actin cytoskeletal organization [36,37,38]. In UC, high expression of miR-15a stimulated by TNF­α was found to directly target CDC42, causing low levels of expression compared to that in the healthy group [39]. Downregulated CDC42 in UC leads to decreased expression of zona occludens (ZO)-1 and E-cadherin, which negatively regulate intestinal epithelial junctions and disrupt the epithelial barrier [39]. Through activating the NADPH oxidase Nox1, Rac1 promotes the generation of reactive oxygen species (ROS) and recruitment of neutrophils in injured intestinal tissues, which promote intestinal epithelial wound repair by positively regulating adhesion of IECs to the matrix [40]. However, persistent neutrophil infiltration and the release of proinflammatory cytokines induced by Rac1 were found to cause destruction of the intestinal epithelium, which contributes to the development of UC [41,42,43]. In addition, two single nucleotide polymorphisms (SNP) of Rac1, rs10951982 and rs4720672, were identified as risk alleles associated with a higher risk of UC [43].

Two pathological molecular mechanisms of SRC were revealed in DSS-induced UC rats. SRC can be upregulated by Calcium/Ask1/MKK7/JNK2 signaling and induces Tyr-phosphorylation of occludin, ZO-1, E-cadherin and β-catenin, which disrupts tight junctions in the intestinal epithelium and leads to the epithelial barrier dysfunction [44]. Furthermore, SRC binds to and activates PIK3R1 (p85), a driver of PI3K signaling, resulting in activation of the NF-κB signaling cascade, which contributes to persistent inflammatory responses in injured intestinal tissues [45].

POLR2A is the largest catalytic subunit of RNA polymerase II (pol II) and is contained in Module 1. RNA pol II catalyzes various aspects of transcriptional events, including pre-mRNA splicing, which is a major step of mRNA processing [46, 47]. Enrichment analysis of Module 2 showed that DEGs in this module were mainly associated with RNA splicing and the spliceosome. Inflammation has been reported to perturb intra-cellular processes and lead to alternative pre-mRNA splicing in IECs, which alters the function, location or expression of a variety of proteins and promotes UC progression [48, 49].

A total of 24 drugs approved by FDA were predicted to have potential effects on four hub DEGs SRC, MAPK1, RAC1 and CDC42 and some of these drugs were confirmed to inhibit UC progression. MAPK1 is predicted to be targeted by Arsenic trioxide (As2O3) and ursodeoxycholic acid (ursodiol). In UC model rats induced by acetic acid or 2,4,6-trinitrobenzene sulfonic acid (TNBS), As2O3 treatment ameliorates colon inflammation through the reduction of inflammatory cells and oxidative stress [50, 51]. Ursodiol is found to protect epithelial barrier function by preventing the release of proinflammatory cytokines from IECs in DSS and TNBS-induced UC rats [52, 53]. Besides, the use of ursodiol is associated with a low risk of neoplasia in UC patients with primary sclerosing cholangitis [54, 55]. As an important tyrosine kinase, SRC is predicted to be targeted by several kind of tyrosine kinase inhibitors (TKIs). An SRC-specific TKI dasatinib shows significant anti-inflammatory effects in acetic acid-induced UC rats [56]. In addition to inhibiting the immune response, dasatinib can also protect the intestinal epithelial barrier of acetic acid-induced UC rats by decreasing the colonic vascular permeability [57]. Besides, the combination of dasatinib with TOP1210, a narrow spectrum kinase inhibitor that can inhibit SRC activity, exhibit improved potent inhibitory effects on pro-inflammatory cytokine release and immune response in UC [58].

The CIBERSORT algorithm revealed differences in the immune cell infiltration characteristics between pediatric and adult UC. Significant differences in the proportions of naive and memory B cells were revealed. In a recent study, CD23 was identified as a biomarker of mature naive B cells in pediatric UC, and the percentage of CD23+ B cells in pediatric UC patients was increased compared to that in non-IBD (inflammatory bowel disease) control children [59]. In contrast, the numbers of CD27+ memory B cells were decreased in UC compared to healthy controls, indicating that this population of cells participates in the pathogenesis of UC [60]. Compared with adult UC samples, we identified higher percentages of naive B cells and lower percentages of memory B cells in pediatric UC samples. Thus, it can be speculated that this profile of B cell types contributes to a more severe phenotype compared to adult UC.

Changes in the proportion of different subtypes of T cells are closely related to the development of UC [59, 61, 62]. Here, we found higher proportions of resting CD4 memory T cells and regulatory T cells (Tregs) in pediatric UC samples, while the proportions of follicular helper T cells and γδ T cells were lower. Tregs are essential for maintaining T cell homeostasis and immunological tolerance. In UC, the Treg cell frequency is decreased, which contributes to intestinal immunodysregulation and inflammation [63, 64]. Follicular helper T cells and γδ T cells were found to be activated in UC, leading to the positive regulation of inflammatory processes and dysimmunity typical of UC [61, 62, 64,65,66,67]. In addition, we found a higher percentage of dendritic cells (DCs) in adult UC, which is consistent with reports that alterations in DC phenotypes in UC drive aberrant T-cell responses [68, 69]. Thus, dysregulation of T cells in adult UC seems to be more severe than that in pediatric UC; this phenomenon requires further exploration.

Inflammatory processes in UC can be sustained by inappropriate activation or inactivation of monocytes and macrophages [70]. Compared with adult UC samples, we found higher proportions of monocytes and M0 macrophages in pediatric UC samples, while the proportions of M2 macrophages were lower. M2 macrophages are decreased in UC compared to healthy controls, and recovery of this population can antagonize M1 macrophage responses to inhibit inflammation [71,72,73,74,75]. Lower proportions of M2 macrophages in pediatric UC may lead to more severe inflammatory processes compared to those in adult UC. Furthermore, higher proportions of mast cells were detected in pediatric, representing another dysregulated immune cell population in UC that are activated and involved in disease development [76, 77].

In conclusion, a total of 1700 DEGs were identified in our bioinformatics analysis of colon samples from pediatric and adult UC patients based on gene dataset GSE87473. CDC42, POLR2A, RAC1, PIK3R1, MAPK1 and SRC were identified as the hub DEGs and may play important roles in the differences between pediatric and adult UC. Furthermore, pathway enrichment analysis of DEGs revealed that inter-cellular items that relate to cell junctions, cell adhesion, actin cytoskeleton and transmembrane receptor signaling pathways and intra-cellular items that relate to the splicing, metabolism and localization of RNA were involved in the differences between pediatric and adult UC. Finally, we investigated immune cell infiltration characteristics in pediatric and adult UC and identified the significant differences.

Inevitably, there were some innate limitations in the present study. The sample size was relatively small, and further experiments were lacked to verify the results due to the lack of relevant samples. All in all, these DEGs, pathways and immune cell infiltration characteristics identified in this study may provide insights into the molecular mechanisms of the differences between pediatric and adult UC, which require further confirmation in future studies.