What Is Known/What Is New
What Is Known
- The introduction of new treatments for pediatric inflammatory bowel disease generally relies on extrapolation from small pediatric studies based on the results of larger adult trials. This extrapolation rests on an assumption of similarity of natural history and pathogenic processes in adult and pediatric populations.
- Differences in mucosal gene expression between adult and pediatric ulcerative colitis are not well characterized, whereas genetic studies have not detected pathways that distinguish adult and pediatric populations.
- What Is New
- The ulcerative colitis gene expression landscape is shared by adults and children, independent of disease extent.
- Fundamentally similar processes at the level of genetic pathways and predicted regulators support extrapolation of efficacy from adults to children for development of new therapies for ulcerative colitis.
There is a need for new medical therapies for pediatric ulcerative colitis (UC), but approval of drugs for children with UC typically lags several years behind those of adult therapies. To address this need, the Food and Drug Administration (FDA) has instituted requirements for pediatric studies to accompany drug development in adult populations (1). In light of the difficulties inherent in conducting large controlled studies in children, the FDA permits extrapolation of efficacy results from adult populations to children. A foundation of extrapolation is similarity of natural history and pathogenic processes in the 2 disease populations. Several studies have shown that children with pediatric UC tend to have a greater extent of colonic involvement and a more aggressive course than adults (2–6), raising the possibility that the molecular mechanisms driving pediatric UC differ from those of adult UC. The greater disease extent at presentation seen in children with UC also suggests that extensive UC (E4 in the Paris classification) reflects a distinct underlying process than limited UC. In contrast, genome-wide association studies have indicated that the genetic associations with UC are similar in adult and pediatric-onset UC (excluding extremely early-onset inflammatory bowel disease [IBD]) (7,8). Likewise, the medical therapies for UC are generally similar in adults and children (9).
Gene expression profiling provides an opportunity to understand disease pathogenesis on a molecular level and can provide another means of evaluating whether extrapolation from adult populations to children is appropriate. To our knowledge, transcriptomic comparisons between pediatric and adult intestinal tissue in UC have not been described. We have therefore taken advantage of samples obtained in the course of clinical studies, including a phase 3 study of golimumab in adults with UC, a phase 1b study of golimumab in children with UC, and a public pediatric dataset to compare the 2 populations at the levels of gene expression, pathway enrichment, and predicted upstream regulators.
Because pediatric UC is generally more extensive than in adults, it is also important to determine if the molecular features of extensive UC differ from those of limited disease. Therefore, for each of the adult and pediatric populations, we compared gene expression profiles of extensive UC versus those of limited UC.
Study Populations and Biopsy Collection
The molecular profile of UC in adults was derived using data from the PURSUIT study (10), a phase 3 clinical trial of golimumab in adults with moderate-to-severe UC (biopsy substudy, n = 87). Additional healthy control biopsies (n = 21) were obtained from the Department of Gastroenterology, Perelman School of Medicine at the University of Pennsylvania (Philadelphia, PA) and the Department of Gastroenterology, University Hospital Gasthuisberg (Leuven, Belgium). The adult control biopsies were obtained from the colon without specification of colonic segment. Pediatric data were obtained from a phase 1b clinical trial of golimumab in pediatric patients between 6 and 17 years of age, with moderate-to-severe UC (CNTO148UCO1001; NCT01900574; n = 19), and a publicly available gene expression dataset (GSE10616; 10 UC, 11 healthy) of pediatric subjects (age range not available) (11,12). Biopsies for the GSE10616 dataset were obtained from the ascending colon. In the clinical trials, endoscopists were instructed to obtain biopsies 15 to 20 cm from the anal verge from locations representative of the degree of inflammation seen in the region. Biopsies were performed during the study screening periods. To assess colonic gene expression by anatomic location, we analyzed the publicly available GSE48634 dataset in adults with pan (total) UC, in which paired biopsies were collected from 4 anatomical locations of each patient, including terminal ileum (n = 8), descending colon (n = 11), ascending colon (n = 9), and rectum (n = 11) (13,14). Institutional review boards approved the PURSUIT, CNTO148UCO1001, University of Pennsylvania, and University Hospital Gasthuisberg healthy subject protocols and all patients provided written informed consent, including the analysis of gene expression.
Gene Expression Analysis
Biopsies collected from the PURSUIT (including the healthy adult controls) and CNTO148UCO1001 studies were stored in RNAlater at −80°C (Ambion Inc, Austin, TX) until RNA isolation. Total RNA was isolated using the RNAeasy mini kit according to the manufacturer's instructions (Qiagen Inc, Valencia, CA). Microarray analysis was conducted using Affymetrix HT HG U133+ PM array according to the manufacturer's protocol (Affymetrix, Santa Clara, CA). Gene expression measurements were obtained by the Robust Multi-array Average (RMA) algorithm (15). Raw data are accessible through accession number GSE87473 (SubSeries: PURSUIT GSE87466 and CNTO148UCO1001 GSE87465) in the Gene Expression Omnibus at the NCBI.
Identical analysis methods and stringency were applied to the adult and pediatric datasets. The microarray data were analyzed using ArrayStudio v9 (OmicSoft Corp, St Morrisville, NC). The General Linear Model was performed to identify significant expression modulations between UC and healthy and between extensive (pancolitis) and limited (left-sided) UC. Differential gene expression was considered significant if |fold change| was >2-fold and the false discovery rate (FDR) adjusted P value was <0.05 (16). Colonic UC disease profiles were derived by comparing the biopsy mRNA expression of UC patients to that of age-matched healthy subjects. The adult UC disease profile consisted of probe sets with significant differential expression between 87 adult patients in the PURSUIT study and 21 healthy adults. The pediatric UC disease profile comprised differentially expressed genes between pediatric UC patients (n = 10) and healthy children (n = 11) in GSE10616. Gene expression data from CNTO148UCO1001 was not combined with that from GSE10616 to increase the N number of the pediatric group because of batch effects in the data generation.
Functional enrichment analysis was conducted to identify canonical pathways that are significantly enriched in adult and pediatric UC pathogenesis using Ingenuity Pathway Analysis (IPA, www.qiagen.com/ingenuity). Statistical significance of pathway enrichment was determined by Fisher's exact test. Canonical pathways with a P value <0.05 were considered statistically significant. IPA was also used to identify upstream transcriptional regulators that may regulate the adult and pediatric UC disease profiles. The activation z score was calculated to infer the activation states of predicted transcriptional regulators, with positive and negative z scores indicating activation and inhibition, respectively. Results from adult and pediatric disease profiles were then compared to identify common and distinct canonical pathways and predicted upstream regulators between these different age groups.
To compare gene expression in patients with extensive versus limited disease distribution, the PURSUIT dataset and colonic biopsy gene expression from CNTO148UCO1001 were used. The disease profiles of intestinal biopsies of distal colon from adult patients with extensive (n = 27) or limited (n = 60) UC, respectively, and pediatric patients with extensive (n = 13) and limited (n = 6) UC, respectively, populations were compared. The Spearman correlation between extensive and limited UC profiles was assessed using ArrayStudio v9 (OmicSoft Corp). Principal component analysis and pair-wise correlation were conducted to assess colonic gene expression by anatomic location.
Comparability Between the Adult and Pediatric Ulcerative Colitis Datasets
Table 1 summarizes the baseline demographic characteristics of the 87 and 19 patients who participated in the biopsy substudies in PURSUIT and CNTO148UCO1001, respectively, and the full cohorts (PURSUIT; n = 1065 and CNTO148UCO1001; n = 35). Overall the demographic characteristics of the patients who provided biopsies were not significantly different from those of the full corresponding cohort. The pediatric dataset (GSE10616) was selected because it implemented biopsy locations and collection methods similar to those of the adult study and included healthy controls (Table 2). While different platforms were used for the PURSUIT (Affymetrix HT HG U133+ PM array) and GSE10616 (Affymetrix HG U133 Plus 2) datasets, they share the same content. To ensure comparability, the same analysis methods and stringency were applied to both datasets (Table 2). In addition, to assess the potential effect of different colonic locations for the biopsies in the studies, we analyzed gene expression in the public GSE48634 dataset. While gene expression of colonic biopsies was distinct from that of terminal ileum, we found that all the colonic locations (ie, descending colon, ascending colon, and rectum) clustered together by principal component analysis and were not separated from each other in unsupervised clustering on pairwise correlation coefficients (Supplemental Figure 1, Supplemental Digital Content, http://links.lww.com/MPG/B271).
Similarity Between the Adult and Pediatric Ulcerative Colitis Disease Profiles
The adult and pediatric UC disease profiles overlapped substantially. A total of 2253 and 2129 probe sets were assigned to the UC disease profile (ie, differentially expressed genes in UC vs healthy controls) in adult (Supplemental Table 1, Supplemental Digital Content, http://links.lww.com/MPG/B396) and pediatric (Supplemental Table 2, Supplemental Digital Content, http://links.lww.com/MPG/B397) patients, respectively, using a cutoff of |fold change| >2 and adjusted P value <0.05 (Fig. 1). There was 49.3%, (1049/2129 genes) overlap of the pediatric UC disease profile with that of the adult UC profile. All of the 1049 overlapping genes were positively correlated (ie, genes modulated in the same direction in adult and pediatric UC). If the fold-change criterion in the adult UC dataset is reduced to 1.5-fold, then the proportion of the pediatric profile that overlaps with that of adult increases from 49.3% to 75.4%. Genes associated with inflammatory response (IL8, S100A8/9, SAA1, TWIST1), tissue remodeling (MMPs), host defense functions (DEFB4A, DUOX2), and intestinal barrier and absorption (CLDN1/2/8, and solute carrier family members) were common to both pediatric and adult UC disease profiles, with modulation in the same direction (Table 3).
It can also be informative to determine the proportion of each disease profile that is truly unique to that population, that is, genes differentially expressed in 1 patient population (|fold change| >2 and adjusted P value <0.05) that fail to meet even low stringency inclusion criteria in the other population (|fold change| >1.2 or adjusted P value <0.05). By this measure, <10% of the disease profile was unique to pediatric (9.4%, 201 of 2129) or adult UC (8.8%, 198 of 2253) population, and in most cases the direction of change was the same in both age groups. The profile genes with the greatest difference in differential expression between adults and children included 2 antimicrobial peptide genes, DEF4BA and REG3A. Compared to healthy controls, DEF4BA expression was increased over 24-fold in adults with UC, but was increased only 10.6-fold in children with UC; REG3A was increased 14.7-fold in adults with UC, compared with 8.6-fold in children.
Comparison of differentially expressed genes at the level of canonical pathways across data sets may provide more biologically relevant results than single gene comparisons (17). Pathway enrichment was therefore used to identify canonical pathways altered in pediatric or adult UC. The top 10 canonical pathways enriched in pediatric or adult UC disease profiles are shown in Figure 2 (all Fisher's exact test P values <0.05). The top canonical pathways enriched in pediatric UC are also significantly enriched and highly ranked in adult UC. Upstream regulators that may explain the differential mRNA expression modulations observed in adult or pediatric disease profile were also identified (Table 4), suggesting that the pediatric and adult UC disease profiles (relative to healthy controls) are controlled by the same set of factors, including tumor necrosis factor alpha (TNFα), interleukin 1A (IL1A), and interferonγ. To further seek differences between adult and pediatric UC, pathway analysis was performed on the set of genes differentially expressed in the adult versus the pediatric UC profile. No pathways were identified that were specific to either population.
Similarity of Disease Profiles Between Patients With Extensive Versus Limited Disease
The extent of disease in pediatric UC is greater at the time of diagnosis than in the adult UC population (2,18,19). Thus, extrapolation of clinical efficacy for new agents from adult to pediatric UC may not be justified if extensive UC differs from limited disease at the level of gene expression. To address this, the gene expression profiles of extensive disease were compared to those with limited disease in both age groups. General Linear Model analysis did not reveal any significant expression modulation between extensive and limited UC in pediatric or adult populations. Unsupervised clustering was conducted for pediatric (Fig. 3A) and adult (Fig. 3B) UC disease profiles separately, and no segregation of patients with extensive versus limited disease was found. Average disease profiles of patients with extensive and limited UC were calculated to evaluate the correlation between the 2. Extensive and limited disease profiles were found to be significantly correlated in both pediatric (r = 0.9674, P value <0.0001; Fig. 4A) and adult populations (r = 0.9915, P value <0.0001; Fig. 4B), indicating strong similarity among UC patients regardless of disease extent.
In this study, gene expression data from colonic biopsies obtained in recent clinical trials and a publicly available dataset were used to compare gene expression profiles in adult and pediatric UC patients. At the level of individual genes, the overlap of differential expression between the 2 populations was high, with 50% to 75% of their profiles overlapping with each other, depending on the fold-change cutoff that was used. While this degree of overlap may seem limited, prior studies comparing disease gene profiles have shown that the use of rigid fold-change cutoffs tends to falsely exclude shared gene expression changes (20,21). For example, in a study of gene expression in psoriasis, technical variations and arbitrary thresholds used to call “significant” expression modulations resulted in overlaps of 50% to 60%, even in similarly aged populations (20). Many of the nonoverlapping genes in this case were near the cutoffs for fold change or significance, and so were not truly unique to 1 population. In the current study, we found that very few (∼10%) of the differentially expressed genes were truly unique to the adult or pediatric populations. Two genes, however, encoding antimicrobial peptides (REG3A and DEF4BA), while upregulated in both populations, were more strongly induced in adult UC than in children, suggesting that host-microbe interactions may differ in the 2 age groups.
The shared differentially expressed genes encompass functions considered to be central to UC pathogenesis or the development of complications, such as inflammatory signals (eg, IL8), epithelial barrier function (claudin genes), tissue remodeling/fibrosis (MMPs), and host factors controlling microbial growth (defensins). In light of these similarities, it is not surprising that clustering algorithms were not able to distinguish limited from extensive disease in both the adult and pediatric datasets.
The gene expression similarity across UC age groups is also marked at the level of gene sets representing biological pathways, with 100% of the top enriched canonical pathways being shared in adult and pediatric UC. These pathways include dendritic cell maturation, which has been associated with UC in genome-wide association studies (CD40, ICOSLG(22)), and cytokine production by epithelial cells, an area that has received relatively little attention. Enrichment of the granulocyte recruitment pathway was also observed and is consistent with the abundance of these cells in the UC mucosa. The enrichment in fibrosis-related genes may reflect the collagen deposition that occurs in long-standing UC.
From the standpoint of therapeutics, it is meaningful to evaluate the regulatory factors predicted to drive the observed differential gene and pathway levels, as these regulators are candidate drug targets. The regulators predicted to control the differentially expressed genes are nearly 100% shared between adult and pediatric UC, and these include TNFα, the target of biologic therapeutic agents approved for use in adults and children with UC. In addition, granulocyte-macrophage colony stimulating factor (GM-CSF) is a predicted regulator for adult and pediatric UC and has shown promise in clinical studies in adults with Crohn disease (23–25).
Some limitations of the study should be noted. Comparisons across separate studies are subject to confounding by batch effects and other uncontrolled variables, although the transcriptomic platforms and analysis pipelines were very similar. The pediatric sample sizes were much smaller than those of adults, resulting in reduced power to detect gene expression differences in children. The duration of disease in the pediatric CNTO148UCO1001 samples (1.2 years) was shorter than in adult PURSUIT samples (4–5 years). In addition, over twice as many children in the CNTO148UCO1001 biopsy substudy had extensive UC (68%) in comparison to the adult study (31%); this difference is mitigated by our finding that extensive and limited disease have highly similar gene expression profiles. The adult and pediatric biopsies also differed in anatomic location, with the pediatric biopsies coming mostly from the ascending colon and the adult biopsies from the rectum. The similarity in disease profiles despite this difference further supports the similarity of the 2 UC disease gene expression profiles. Finally, the ages of the children in GSE10616 were not available; it is likely that very early-onset IBD (VEO-IBD; disease manifesting at age <6 years) was not represented. VEO-IBD occurring <2 years of age is often of monogenic etiology (eg, IL10 pathway or XIAP mutations) and is likely to represent distinct gene expression patterns from pediatric-onset IBD, making it less amenable to extrapolation from clinical studies in adults. It will be important to validate these findings in independent cohorts using uniform specimen collection, handling, and analysis.
Overall, this analysis supports the concept that pediatric and adult UC reflect fundamentally similar processes at the level of genetic pathways and predicted regulators. This finding has implications for the development of new UC treatments for children. One criterion for the use of extrapolation is a similarity in disease progression, typically based on natural history studies. These generally show a greater extent of disease and higher rates of colectomy in children than adults. Our study indicates that these clinical differences unlikely to be due to the activity of distinct molecular disease pathways in adults and children or to intrinsic differences in extensive versus limited disease. These findings thus provide support at the level of intestinal gene expression for the use of extrapolation to accelerate the development of new treatments for children with UC.
Editorial and submission support was provided by Kirsten Schuck Gross of Janssen Scientific Affairs, LLC
1. Requirements on Content and Format of Labeling for Human Prescription Drug and Biological Products, Food and Drug Administration Final Rule, 2006. https://www.fda.gov/OHRMS/DOCKETS/98fr/06-545.pdf
2. Levine A, de Bie CI, Turner D, et al. EUROKIDS Porto IBD Working Group of ESPGHAN. Atypical disease phenotypes in pediatric ulcerative colitis
: 5-year analyses of the EUROKIDS Registry. Inflamm Bowel Dis
3. Lev-Tzion R, Turner D. Is pediatric IBD treatment different than in adults? Minerva Gastroenterol Dietol
4. Van Limbergen J, Russell RK, Drummond HE, et al. Definition of phenotypic characteristics of childhood-onset inflammatory bowel disease. Gastroenterology
5. Turner D, Walsh CM, Benchimol EI, et al. Severe paediatric ulcerative colitis
: incidence, outcomes and optimal timing for second-line therapy. Gut
6. Dinesen LC, Walsh AJ, Protic MN, et al. The pattern and outcome of acute severe colitis. J Crohns Colitis
7. Cutler DJ, Zwick ME, Okou DT, et al. Dissecting allele architecture of early onset IBD using high-density genotyping. PLoS One
8. Essers JB, Lee JJ, Kugathasan S, et al. Established genetic risk factors do not distinguish early and later onset Crohn's disease. Inflamm Bowel Dis
9. Ruemmele FM, Turner D. Differences in the management of pediatric and adult onset ulcerative colitis
—lessons from the joint ECCO and ESPGHAN consensus guidelines for the management of pediatric ulcerative colitis
. J Crohns Colitis
10. Lichtenstein GR, Feagan BG, Cohen RD, et al. Drug therapies and the risk of malignancy in Crohn's disease: results from the TREAT Registry. Am J Gastroenterol
12. Kugathasan S, Baldassano RN, Bradfield JP, et al. Loci on 20q13 and 21q22 are associated with pediatric-onset inflammatory bowel disease. Nat Genet
14. Smith PJ, Levine AP, Dunne J, et al. Mucosal transcriptomics implicates under expression of BRINP3 in the pathogenesis of ulcerative colitis
. Inflamm Bowel Dis
15. Irizarry RA, Hobbs B, Collin F, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics
16. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol
17. Tian S, Krueger JG, Li K, et al. Meta-analysis derived (MAD) transcriptome of psoriasis defines the “core” pathogenesis of disease. PLoS One
18. Kalkan IH, Dagli U, Oztas E, et al. Comparison of demographic and clinical characteristics of patients with early vs. adult vs. late onset ulcerative colitis
. Eur J Intern Med
19. Lee JH, Cheon JH, Moon CM, et al. Do patients with ulcerative colitis
diagnosed at a young age have more severe disease activity than patients diagnosed when older? Digestion
20. Suarez-Farinas M, Li K, Fuentes-Duculan J, et al. Expanding the psoriasis disease profile: interrogation of the skin and serum of patients with moderate-to-severe psoriasis. J Invest Dermatol
21. Suarez-Farinas M, Lowes MA, Zaba LC, et al. Evaluation of the psoriasis transcriptome across different studies by gene set enrichment analysis (GSEA). PLoS One
22. Jostins L, Ripke S, Weersma RK, et al. Host-microbe interactions have shaped the genetic architecture of inflammatory bowel disease. Nature
23. Dieckgraefe BK, Korzenik JR. Treatment of active Crohn's disease with recombinant human granulocyte-macrophage colony-stimulating factor. Lancet
24. Korzenik JR, Dieckgraefe BK, Valentine JF, et al. Sargramostim for active Crohn's disease. N Engl J Med
25. Dejaco C, Lichtenberger C, Miehsler W, et al. An open-label pilot study of granulocyte colony-stimulating factor for the treatment of severe endoscopic postoperative recurrence in Crohn's disease. Digestion