Correlation of Biomarker Expression in Colonic Mucosa with Disease Phenotype in Crohn’s Disease and Ulcerative Colitis

Study Individuals

Colonic mucosal biopsies were obtained from subjects undergoing colonoscopy at the University of Kentucky Medical Center, following written informed consent approved by the Institutional Review Board. The diagnosis of UC was based on clinical, endoscopic, radiological and histopathological criteria. The presence of active disease was based on endoscopic evidence of inflammation. Biopsies were collected from regions of the colon that were visibly inflamed and non-inflamed, as subsequently confirmed by histological examination. Control subjects underwent colonoscopies for evaluation of constipation or chronic abdominal pain or for routine screening for colon cancer. Patients were classified as normal when endoscopic, radiologic and/or histological evaluation detected no colonic disease.

Analysis of mRNA Levels in Colonic Mucosal Biopsies

Biopsied tissue samples were collected into an RNA-stabilizing solution (RNAlater, Qiagen, Germantown, MD) and stored at −80°C until analyzed. Total RNA was purified using the RNeasy Protect mini kit (Qiagen) and reverse-transcribed to generate cDNA templates using the TaqMan Gold RT-PCR kit (Applied Biosystems, Foster City, CA). Levels of individual mRNA transcripts were analyzed by quantitative real-time polymerase chain reaction (qRT-PCR), using the ABI Prism 7700 Sequence Detection System (Applied Biosystems) as previously described [12]. The threshold cycles (C_T) for test mRNAs were normalized to the C_T for β2-microglobulin according to the formula: 2^-(C_T^test-C_T^{β2-microglobulin)} × 100%.

Statistical Analyses

Human gene expression data were analyzed by non-parametric Mann-Whitney and Spearman correlation tests (http://www.winstat.com). Comparisons between gene expression in paired biopsies from non-inflamed and inflamed colonic mucosa were made using the Wilcoxon signed-rank test (http://vassarstats.net/wilcoxon.html). Multifactorial principal component analyses (PCA) of gene expression data were conducted using Multibase statistical analysis software (http://www.numericaldynamics.com).

Biomarker expression in colonic mucosa of IBD patients and healthy controls

To determine whether mRNA levels for 5 candidate genes could be used to classify IBD patients into clinically relevant subsets, biopsies of colonic mucosa were collected during colonoscopy from 113 IBD patients (60 CD, 53 UC) and 44 healthy controls (Table 1). Diagnosis of CD or UC was based on clinical, radiological, and endoscopic criteria according to the Montreal classification [35], supported by histopathological findings. IBD patients were sub-classified into those with or without acute inflammation, as evidenced by areas of macroscopically visible tissue inflammation on colonoscopy. We hypothesized that changes in gene expression throughout the colonic epithelium of IBD patients could promote development of an inflammatory response, and would not necessarily be a consequence of localized inflammation. Accordingly, comparisons in gene expression between IBD patients and healthy controls were restricted to those regions of the colon that were macroscopically and microscopically non-inflamed. Although the median age of the cohort of healthy controls was higher than that of IBD patients (Table 1), there was no significant correlation between age at the time of biopsy and expression of RELA, TNFAIP3, PIGR or TNF (Table 2). However, younger subjects in general had significantly higher levels of IL8 expression in non-inflamed colonic mucosa, as evidenced by the negative correlation with age. Within individuals, levels of RELA, TNFAIP3 and PIGR mRNA were highly correlated, whereas expression of these genes was not correlated with IL8 (Table 2). Interestingly, expression of TNF was correlated with expression of all the other biomarkers within individuals. This finding is consistent with the hypothesis that elevated expression of RelA enhances transcription of the TNFAIP3, PIGR and TNF genes through the classical NF-κB signaling pathway. Other signaling pathways may also enhance transcription of the TNF and IL8 genes. While expression levels of individual biomarkers varied considerably among individuals, some differences were noted between clinical groups (Fig. 1). Median levels of TNFAIP3 and TNF mRNA were significantly lower in CD patients but not UC patients, compared to healthy controls. Interestingly, median levels of PIGR mRNA were reduced in both CD and UC patients compared to healthy controls. Although levels of RELA mRNA were correlated with levels of TNFAIP3, PIGR and TNF mRNA within all individuals (Table 2), the wide variation in RELA expression among individuals precluded the finding of significant differences among clinical groups. Although some individual CD and UC patients had elevated expression of IL8 mRNA, median levels for each category did not differ significantly from that of healthy controls. The finding that median IL8 mRNA levels were significantly elevated in UC patients with acute inflammation compared to UC patients in remission suggests that increased expression of IL8 may be associated with the inflammatory process in UC. This concept was supported by the finding that IL8 levels in UC patients, but not CD patients, were significantly elevated in biopsies from visibly inflamed colonic mucosa compared to paired samples from non-inflamed colonic mucosa from the same individuals (Table 3). By contrast, no significant differences were observed in the levels of RELA, TNFAIP3, PIGR or TNF between inflamed and non-inflamed colonic mucosa in either CD or UC patients.

Multivariate analysis of biomarker expression in colonic mucosa

Given the wide variability in expression of individual biomarkers within groups in our cohort, we hypothesized that a multivariate approach would be more useful for predicting clinical outcome in IBD patients. Accordingly, the composite data for expression of the 5 biomarkers were reduced to 2 “supervariables,” or principal components (PCs) for each individual (Table 4). The results of the PCA indicated that PC1 accounted for 40.7% of the overall variance in expression of the 5 biomarkers in this population of individuals, while PC2 accounted for 27.8%. For each individual, the value of PC1 and PC2 is calculated as the sum of normalized mRNA level for each gene, multiplied by the weighting coefficient (on a scale of −1 to +1) for that gene. As noted in Table 4, PC1 had strongly negative coefficients for RELA, TNFAIP3 and PIGR, and moderately negative coefficients for TNF and IL8. In other words, high expression levels for RELA, TNFAIP3 and/or PIGR (and to a lesser extent TNF and IL8) in an individual subject would result in a low score for PC1. By contrast, PC2 had strongly positive coefficients for TNF and IL8, a moderately negative coefficient for TNFAIP3, and weakly negative coefficients for RELA and PIGR. Therefore, high expression of TNF and/or IL8 in any individual subject would result in a high score for PC2, which could be lowered somewhat by high expression of TNFAIP3 and to a lesser extent RELA or PIGR.

To examine the distribution of PC scores among clinical groups, a 2-dimensional scatter plot was constructed showing PC1 and PC2 scores for each individual (Fig. 2). This plot was divided into 4 quadrants based on positive or negative values for PC1 and PC2. By examining the distribution of individuals within each clinical group among the 4 quadrants, it can be determined whether levels of the composite variables PC1 and PC2 are useful for predicting clinical outcome. Chi-square analysis was used to determine whether individuals in different clinical groups were evenly distributed among the 4 quadrants, or whether there was a bias toward one or more quadrants (Table 5A). Healthy control subjects were distributed roughly equally between quadrant A (41%) and quadrant B (48%) and only minimally represented in quadrant C (7%) and quadrant D (5%). This pattern indicates that the majority of healthy controls had low values for PC2, but evenly distributed values for PC1. By contrast, CD patients were more heavily distributed in quadrant B (52%), followed by quadrant A (24%), quadrant C (18%) and quadrant D (8%). However, the subgroup of CD patients with clinical evidence of acute inflammation was more evenly distributed among the 4 quadrants (as noted by the non-significant chi-square value), with the greatest numbers found in quadrant B (44%) and quadrant C (33%). The distribution of all UC patients was similar to that of all CD patients in that there was a bias toward quadrant B (45%) followed by quadrant A (28%), but UC patients had a more even distribution between quadrants C (13%) and D (15%). As was the case with CD, UC patients with clinical evidence of acute inflammation were more evenly distributed among the 4 quadrants, as noted by the non-significant chi-square value.

Another approach for evaluation of the 2-dimensional PCA data was to compare the distribution of individuals in different clinical groups within each quadrant (Table 5B). Within quadrant A (PC1 < 0, PC2 < 0), there was a significantly higher proportion of healthy controls compared to CD patients (p = 0.017). However, this did not hold true for UC patients compared to healthy controls, nor were there significant differences within quadrant A between CD or UC patients with or without acute inflammation. Within quadrant B (PC1 > 0, PC2 < 0), there were no significant differences among any of the groups. Interestingly, a bias toward active CD was seen in quadrant C (PC1 > 0, PC2 > 0), as demonstrated by a significantly higher proportion of CD patients compared to healthy controls (p = 0.028), and an even greater proportion of CD patients with acute inflammation compared to CD patients in remission (p = 0.002). By contrast, a bias toward UC was seen in quadrant D (PC1 < 0, PC2 > 0), as demonstrated by a significantly higher proportion of UC patients compared to healthy controls (p = 0.025), as well as a significantly higher proportion of UC patients with acute inflammation compared to UC patients in remission (p = 0.04). In summary, a multivariate approach combining 5 biomarkers was more predictive of clinical outcome than was any single biomarker. A characteristic of both CD and UC was a shift toward elevated levels of PC2 (quadrants C and D), which could result from elevated expression of TNF and/or IL8, and could be mitigated somewhat by increased expression of TNFAIP3. The finding that healthy controls were more highly represented than were CD patients in quadrant A (PC1 < 0, PC2 < 0) suggested that lower levels of PC1, resulting from elevated expression of RELA, TNFAIP3 and/or PIGR, could be protective against inflammation in this disease. This conclusion is consistent with the greater representation of CD patients with acute inflammation than those in remission in quadrant C (PC1 > 0, PC2 > 0). However, low PC1 values were not associated with improved outcome in UC patients, as evidenced by no significant difference from healthy controls in quadrant A (PC1 < 0, PC2 < 0), and the greater representation of UC patients with acute inflammation compared to those in remission in quadrant D (PC1 < 0, PC2 > 0). These differences between CD and UC patients in overall patterns of biomarker expression, which were not revealed by analysis of single biomarkers, may represent inherent differences in the underlying pathology of the two forms of IBD.