EP1667669A2 - Inflammatory bowel diseases - Google Patents

Abstract

Biomarkers and combinations of biomarkers are defined useful for diagnosis and monitoring of inflammatory bowel diseases.

Description

INFLAMMATORY BOWEL DISEASES

[01] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone ofthe patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

FIELD OF THE INVENTION

[02] The invention relates in some of its aspects to diagnosis of inflammatory bowel disease. More particularly, it relates in some of its aspects to Ulcerative Colitis and Crohn's Disease.

BACKGROUND OF THE INVENTION

[03] Inflammatory bowel disease (IBD) is a chronic inflammation at various sites in the gastrointestinal (GI) tract and includes 2 conditions: Crohn's disease (CD) and ulcerative colitis (UC). CD represents a nonspecific chronic transmural inflammatory disease that most commonly affects the distal ileum and colon but may occur in any part ofthe GI tract. UC is defined as a chronic, inflammatory, and ulcerative disease arising in the colonic mucosa, characterized most often by bloody diarrhea.

[04] The etiology of inflammatory bowel disease is unknown. Most likely, multiple factors are involved. Evidence suggests that a genetic predisposition leads to an unregulated intestinal immune response to an environmental, dietary, or infectious agent. Epidemiological observations may be helpful in identifying the true causative factors of this evasive disease. Geographically, the prevalence of the disease has a slope from North to South and, to a lesser degree, from West to East (Karlinger et al. 2000). In the Western world, a sharp increase in the incidence of both Crohn's disease and ulcerative colitis was observed during the sixties and seventies, after which it reached a plateau at around 7 and 12 new cases per 100,000 inhabitants per year, respectively (Russel and Stockbrugger 2001). The Western-Eastern discrepancy can be attributed to a difference in life styles. The incidence of the disease has been increasing worldwide, but its spread has been slowing down in highly affected countries. Racial and ethnic relations in different populations and immigration studies offer interesting data, which can reflect genetic, environmental and behavioral factors (Karlinger et al. 2000). The disease seems to have a characteristic racial-ethnic distribution: the Jewish population is highly susceptible everywhere, but its prevalence in that population nears that of the domestic society in which they live. In Hungary, the Roma (Gypsies) have a considerably lower prevalence than the average population. This can be attributed to a genetic or environmental influence. The onset of the disease occurs more often in the second or the third decade of life, but there also is another peak in the 60s. As young people are especially prone to develop IBD, most of those affected will have their disease for many years. Regarding sexual distribution, there is a slight preponderance of UC in men and of CD in women. It may correspond to the stronger autoimmune affection in the process of Crohn's disease.

[05] Numerous environmental and genetic factors, as well as behavioral influences have been implicated in the etiology of IBD. Of these exogenic factors smoking is the most consistent, being of negative influence in CD and protective in UC. Diet and oral contraceptives may influence disease expression, and perinatal events such as viral infections may alter adult susceptibility (Russel and Stockbrugger 1996). The influence of hormonal status and drugs are viewed as contributing factors in the manifestation ofthe disease (Karlinger et al. 2000). Genetic studies show that one-fourth of IBD patients have an affected family member. HLAB27 histocombatibility also plays an important, but not determining role in the development ofthe disease. Genetic factors seem to have a stronger influence in CD than UC. The existence of multiple sclerosis-IBD families may reflect a common genetic background or similar microbial effect. Numerous bacteria and viruses have been suspected of being infectious factors in IBD, mostly in CD, e.g., Mycobacteria, Yersinia, Campylobacter, Clostridium, Clamidia, herpesvirus, rotavirus and the primary measles virus (Karlinger et al. 2000). None of them has been proven as a real and exclusively pathogenic factor. Irnmunological background has an important function in the manifestation of the disease. If an individual has a genetic susceptibility to infections, the down regulation of an inflammation in the bowel wall may not occur in a proper way and may initiate an autoimmune process. Considering the epidemiological, genetic and irnmunological data, it can be concluded that ulcerative colitis and Crohn's disease are heterogeneous disorders of mutifactorial etiology in which hereditary (genetic) and environmental (microbial, behavior) factors interact to produce the disease.

[06] The role of inflammatory mediators and especially cytokines in the pathogenesis of IBD has been investigated extensively but still remains unclear. Different studies have identified aberrant levels of numerous cytokines in IBD. Human IL- 10 serum levels, measured by ELISA, were found to be significantly increased in patients with active UC (144 +/- 34 pg/mL (mean +/- s.e.m.), P < 0.001) and in active CD (132 +/- 32 pg/mL, P < 0.001) compared with healthy controls (44 +/- 9.5 pg/mL) (Kucharzik et al. 1995). Only patients with active CD and active UC presented with significantly increased IL-10 serum levels, while patients with inactive disease did not show any significant increase. There was no statistically significant difference between IL-10 serum levels in patients with CD or UC. Compared with clinical disease activity indices there was a significant correlation between IL-10 serum concentration and CDAI in patients with CD (r = 0.45, P < 0.01) and CAI in UC patients (r = 0.39, P < 0.05). Comparing IL-10 serum levels with serum concentrations of other proinflammatory cytokines there was a significant correlation to serum levels of sIL-2R (r = 0.417, P < 0.05) and IL-6 (r = 0.387, P < 0.05) in patients with CD. Serum cytokine levels in patients with UC did not show any significant correlation to IL-10 serum concentration. IL-10 inhibits the production of proinflammatory cytokines such as IL-1, tumor necrosis factor-alpha (TNF-a), interferon-gamma (EFN-gamma) and IL-6 through inhibitory action on Thl cells and macrophages, and it is thought to be a suppressor type cytokine. IL-10 is elevated in serum of patients with active CD and UC, suggesting that IL-10 acts as a naturally occurring damper in the acute inflammatory process of IBD.

[07] Mucosal macrophages have been implicated in the pathogenesis of IBD. Macrophage-colony stimulating factor (M-CSF) influences monocyte/macrophage proliferation, differentiation, and activation. Serum M- CSF levels were found to be increased in active IBD, and compared to normal intestine, in active IBD the frequency of M-CSF-expressing cells was significantly increased and their distribution markedly altered, although no increase in mucosal M-CSF mRNA levels in intestinal tissue was observed (Klebl et al. 2001). However, the changes were not specific to IBD, as there were similar findings in intestinal tuberculosis and ischaemic colitis. Other studies have suggested that the down-regulation of IL-3 in mast cells derived from steroid- treated IBD patients occurs in vivo and could be an important mechanism for immunomodulation in IBD (Ligumsky et al. 1997). The mucosal expression of the chemokines IL-8, RANTES and MCP-1 and the pro-inflammatory cytokines TNF-alpha and IL-6 have been measured by ELISA in intestinal mucosa homogenates from patients with inflammatory bowel disease (McCormack et al. 2001). IL-8 was significantly increased in both disease groups compared to controls. Similarly, RANTES levels were also significantly increased. While MCP-1 levels were increased in both disease groups, this increase was statistically significant in the UC group only. TNF-alpha and IL-6 were significantly increased in the CD group only. Chemokines, together with key cytokines that promote their release are elevated in mucosal tissues from patients with IBD. It is likely that these chemokines play an important role in the perpetuation of tissue destructive inflammatory processes.

[08] Animal models have also been used to study the role of cytokines in IBD. Scid mice develop a severe, chronic, and lethal IBD 3-6 months after engraftment of gut wall from immunocompetent congenic donors, induced by donor-derived CD4+ T cells migrating from the graft. Intracellular T-helper type 1 (Thl) cytokines in the spleens of gut wall-transplanted scid mice with EBD have been investigated (Bregenholt and Claesson 1998). Increased fractions of IFN-gamma, TNF-alpha and IL-2-positive CD4+ T cells were found in the spleens of diseased mice compared with control mice. Moreover, a small but significant population of CD4+ T cells which stained positive for GM-CSF was found in scid mice with IBD but was virtually absent in congenic non-scid control mice. These observations point towards a dominant role for Thl-type CD4+ T cells in the immunopathogenesis of IBD.

[09] Preliminary results have shown that antisense ohgonucleotides (anti-ICAM), anti- cytokine antibodies (anti-TNF) or recombinant human cytokines (IL-10 or IL-11) are effective in some patients with Crohn's disease refractory to steroids (Heresbach et al. 1999). However, these data need to be confirmed and the potential side effects of these treatments must be further considered.

[10] Analysis of cytokine production and function is an indispensable tool to study Crohn's disease and ulcerative colitis. This analysis has generated a tremendous amount of data, but a clear interpretation of results has been hampered by limited attention paid to several patient- and sample-related pitfalls (Fiocchi et al. 1996). These include: clinical parameters; the relative value of circulating vs. intestinal cytokine levels; the selection of tissue specimens and their processing method; the specific cellular source of each cytokine; the effect of cytokine inducers; and the technique utilized to obtain qualitative or quantitative data on cytokine protein or mRNA. A more systematic approach utilizing a single sample type and a uniform assay for simultaneous measurement of multiple analytes may result in a better understanding of the true role of cytokines in the pathogenesis of inflammatory bowel disease.

[11] The question remains open whether UC and CD are one diseases entity (Russel and Stockbrugger 1996). Similarities in the epidemiologic features of UC and CD support the idea of IBD being one disease. Other findings suggest dividing UC and CD into further subgroups: in CD it has been suggested that fibrostenotic, penetrating, and inflammatory behavior should be considered different disease entities; in UC some groups consider ulcerative proctitis a disease entity on its own, separating it from the proximalry extending colitis. In therapeutic trials this approach has proved to be of importance, and it is not inconceivable that in subgroups, with regard to etiopathogenetic mechanisms, different factors have to be looked for. Therefore analysis of multiple cytokine levels maybe necessary to identify biomarkers for IBD and its variations.

[12] Currently, there is no specific in vitro diagnostic test available for IBD and diagnosis is made based on clinical criteria, nonspecific laboratory findings (e.g., anemia, leukocytosis, hypoalbuminemia), x-ray and endoscopic examinations. Furthermore, the current indices used to calculate the disease activity score are based on similar criteria, including subjective patient historical information, physical examination findings, laboratory assessment, weight and height data (Hyams et al. 1991; Seo et al. 1992).

[13] There is a need in the art for identification of biomarkers that can be used to develop an in vitro test for molecular diagnosis of IBD and which correlate with disease activity.

BRIEF SUMMARY OF THE INVENTION

[14] Methods for monitoring and diagnosing ulcerative colitis and Crohn's Disease are provided.

BRIEF DESCRff TION OF THE DRAWINGS

[15] Figures 1-15 provide data which were generated in identifying the biomarkers of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

[16] Using a protein microarray and RCA technology we determined the serum cytokine profiles of IBD patients and normal controls. Based on LDA, we were able to build cytokine classifiers for most of the 11 comparisons and to correctly categorize both original and cross-validated cases with a high success rate. The cytokines, identified as classifiers, represent biomarkers that can be used to diagnose IBD, and biomarkers that correlate with disease activity. The ability to diagnose IBD early, based on biomarkers, will result in timely therapeutic intervention and improved outcome. The ability to evaluate and predict the disease activity will enable the design of more appropriate treatments, adapted to the precise molecular mechanisms that are dysregulated in EBD. Anti-TNF therapy has been shown to be effective against rheumatoid arthritis and Crohn's disease (Raza 2000; Emery et al. 2001). The identification of additional cytokines, involved in the pathogenesis of IBD, may allow the design of combination therapies v/ith several anti-cytokine antibodies. The results from our study demonstrate that the combination of an RCA-based protein chip with bioinformatics tools is a powerful and novel approach for studying diseases involving complex pathologies and for identifying critical biomarkers and therapeutic targets.While the invention has been described with respect to specific examples including presently preferred modes of carrying out the invention, those skilled in the art will appreciate that there are numerous variations and permutations of the above described systems and techniques that fall within the spirit and scope ofthe invention as set forth in the appended claims.

EXAMPLES

Example 1

[17] Samples

[18] Serum samples from children with EBD were provided by Dr. Dhavalkumar Patel, Duke University Medical Center. Control serum samples from normal children were purchased from Diagnostic Support Services, West Barnstable, MA. A total of 124 samples were used in the analysis, including 41 CD patient samples (22 from patients with clinical remission and 19 from patients with active CD; only for Comparisons 6 and 7, two additional CD patients were used since their age was known, increasing the group of CD patients < 9 years of age from 6 to 8), 20 UC patient samples and 63 normal control samples. The experimental and the control populations were age-matched as a group.

[19] Microarray manufacture

[20] Antibody microarrays were printed using a Packard Biosciences (Downers Grove, IL) BCA-II piezoelectric microarray dispenser on cyanosilane-coated glass slides divided by Teflon boundaries into sixteen 0.5 cm diameter circular subarrays. Monoclonal antibodies for 78 cytokines (see Supplementary Material for listing of antibodies and vendors) were dispensed in quadruplicate at a concentration of 0.5 mg/mL. Printed slides were blocked as described (Schweitzer et al. 2002) and stored at 4oC until use. Batches of slides were subjected to a quality control consisting of incubation with a fluorescently-labeled anti-mouse antibody, followed by washing, scanning and quantitation. Typically, the coefficient of variability (CV) of antibody deposition in printing was <5%.

[21] RCA Immunoassay

[22] The assay was performed by a liquid-handhng robot (Biomek 2000, Beckman Instruments, Fullerton, CA), which was enclosed in an 80% humidified, HEPA- filtered, plexiglass chamber. 15 mL sample was applied to each subarray, and immunoassays with RCA signal amplification were performed as described (Schweitzer et al., 2002). Slides were scanned (GenePix, Axon Instruments Inc., Foster City, CA) at 10-mm resolution with laser setting of 100 and PMT setting of 550. Mean pixel fluorescence were quantified using the fixed circle method in GenePix Pro 3.0 (Axon instruments, Foster City, CA). The fluorescence intensity of quadruplicate microarray features was averaged for each feature and sample, and the resulting cytokine values were determined. For every slide, a set of blanks was run and the intensity values were used to correct for background signal.

[23] Data Quality Control [24] Samples were excluded from statistical analysis if fluorescent intensities were generally weak (indicating sample degradation), if there were visible defects in the array (such as scratches), or if there was high background signal. Analyses were performed using complete set of data containing the levels of all 78 cytokines from 124 individuals (41 CD patient samples, 20 UC patient samples and 63 normal control samples). Untransformed fluorescent intensities were used as data values in all ofthe analyses.

[25] Univariate analysis of individual cytokine levels

[26] Using the Kruskal-Wallis test (analysis of variance by rank), the levels of each of the 78 cytokines were analyzed for each ofthe 11 comparisons (Tables 1-11).

[27] For the ANCA/ASCA analysis, ANOVA was done (GLM procedure of SAS v.8.2).

[28] Results:

[29] Results from the univariate analysis of active vs. remission are shown in Table I. Cytokine levels (intensity units) in serum samples from CD patients (clinical remission) vs. CD patients (active disease), and corresponding P-values based on Kruskal-Wallis test

[30] Table I.

Cytokine levels (intensity units) in serum samples from UC patients (clinical remission) vs. UC (active disease), and corresponding P-values based on Kruskal-Wallis test (Significance > 10%)

UC remission UC active

Analyte N Mean SD N Mean SD P-value (Kruskal-Wallis test)

IL12P40 5 1154 436 7 358 308 0.0185

IP10 5 9962 6774 7 3116 1740 0.0618

MCP1 5 24843 11017 7 13911 7127 0.0618

MIP1D 5 33734 16344 7 50372 16344 0.0882

NT4 5 674 536 7 303 214 0.0882

[31] Results from the ANCA/ASCA analysis (all possible combinations of ABS values 0, 1, 2 and 3 were compared).

Analytes showed significant (p Value < 0.05) correlation with ABS levels

ANALYTE Comparison pValue

BLC neg vs ASCA 0.0489

MIG neg vs pANCA/ASCA 0.0372

MIG pANCA vs pANCA/ASCA 0.0233

MIPIA neg vs pANCA/ASCA 0.0196

MEP1A pANCA vs pANCA/ASCA 0.0471

MIPIA ASCA vs pANCA ASCA 0.0347

MJPID negvs pANCA/ASCA 0.0251 MPIF1 negvspANCA/ASCA 0.0120

MSP neg vs p ANCA 0.0141

MSP negvspANCA/ASCA 0.0064

Patients ID used in ABS Study

N ED

1 4

2 9

3 10

4 15

5 18

6 22

7 27

8 37

9 44

10 45

11 48

12 50

13 51

14 62

15 67

16 69

17 71

18 91 Example 2

[32] We measured the levels of 78 different cytokines and growth factors in serum samples using an antibody microarray to determine and analyze the cytokine profiles of patient and control populations.

Materials and Methods Samples

[33] Serum samples from children with IBD were provided by Dr. Dhavalkumar Patel, Duke University Medical Center. Control serum samples from normal children were purchased from Diagnostic Support Services, West Barnstable, MA. A total of 124 samples were used in the analysis, including 41 CD patient samples (22 from patients with clinical remission and 19 from patients with active CD; only for Comparisons 6 and 7 (see below), two additional CD patients were used since their age was known, increasing the group of CD patients < 9 years of age from 6 to 8), 20 UC patient samples and 63 normal control samples. The experimental and the control populations were age-matched as a group.

Microarray manufacture

[34] Antibody microarrays were printed using a Packard Biosciences (Downers Grove, EL) BCA-II piezoelectric microarray dispenser on cyanosilane-coated glass slides divided by Teflon boundaries into sixteen 0.5 cm diameter circular subarrays. Monoclonal antibodies for 78 cytokines were dispensed in quadruplicate at a concentration of 0.5 mg/mL. Printed slides were blocked as described (Schweitzer et al. 2002) and stored at 4°C until use. Batches of slides were subjected to a quality control consisting of incubation with a fluorescently-labeled anti-mouse antibody, followed by washing, scanning and quantitation. Typically, the coefficient of variability (CV) of antibody deposition in printing was <5%. RCA Immunoassay

[35] The assay was performed by a liquid-handling robot (Biomek 2000, Beckman Instruments, Fullerton, CA), which was enclosed in an 80% humidified, HEPA- filtered, plexiglass chamber. 15 μL sample was applied to each subarray, and immunoassays with RCA signal amplification were performed as described (Schweitzer et al., 2002). Slides were scanned (GenePix, Axon instruments Inc., Foster City, CA) at 10-μm resolution with laser setting of 100 and PMT setting of 550. Mean pixel fluorescence were quantified using the fixed circle method in GenePix Pro 3.0 (Axon Instruments, Foster City, CA). The fluorescence intensity of quadruplicate microarray features was averaged for each feature and sample, and the resulting cytokine values were determined. For every slide, a set of blanks was run and the intensity values were used to correct for background signal.

Data Quality Control

[36] Samples were excluded from statistical analysis if fluorescent intensities were generally weak (indicating sample degradation), if there were visible defects in the array (such as scratches), or if there was high background signal. Analyses were performed using complete set of data containing the levels of all 78 cytokines from 124 individuals (41 CD patient samples, 20 UC patient samples and 63 normal control samples). Untransformed fluorescent intensities were used as data values in all ofthe analyses.

Linear Discriminant Analysis

[37] In order to determine the cytokines that contribute the most to discrimination between classes, a stepwise method of LDA from SPSS 8.0 for Windows was used with following settings: Wilks' lambda (Λ) method was used to select cytokines that maximize the cluster separation and cytokine entrance into the model was controlled by its F-value. A large F-value indicates that the level of the particular cytokine is different between the two groups, and a small F-value (F < 1) indicates that there is no difference. In this method, the null hypothesis is rejected for small values of Λ. Thus, we aimed to minimize Λ..

[38] To construct a list of cytokine predictors, we first calculated the F- values for each of the 78 cytokines. Cytokines with F-values <1 were rejected. We then started with the cytokine with the largest F-value (the cytokine that differs the most between the two groups) and determined the value of A. The cytokine with the next largest F-value was then added to the list and Λ recalculated. If the addition of the second cytokine lowered the value of A, it was kept in the list of cytokine predictors. The process of adding cytokines one at a time was repeated until the reduction of Λ no longer occurred. Cross-validation, a method for testing the robustness of a prediction model, was then carried out. To cross-validate a prediction model, one removes and sets aside one sample, uses the remaining samples to build a prediction model based on the preselected cytokine predictors, and determines whether the new model is able to predict the one sample not used in building the new model correctly. This process is repeated for all samples one at a time, and a cumulative cross-validation rate can then be calculated. The final list of cytokine predictors was determined by manually entering and removing cytokines to maximize the cross-validation rate, using information obtained from the univariate analyses and cross-validations. The final cytokine classifier is then defined as the set of cytokine predictors that gives the highest cross-validation rate.

Results

Strategy

[39] The aim of our study is to find biomarkers for IBD and for disease activity. We hypothesize that significant differences exist between the serum cytokine profiles of IBD patients and control individuals, as well as between patients with active disease and patients in clinical remission. We determined the serum cytokine profiles of IBD patients and that of normal controls. We utilized an ordered antibody microarray designed to detect the levels of 78 different analytes including cytokines, cytokine receptors, and growth factors found in blood and other bodily fluids. For simplicity, we will refer to the 78 different analytes as cytokines and the ordered antibody microarray as the cytokine chip. Serum samples from IBD patients and from age-matched normal controls were analyzed using the cytokine chip (see Materials and Methods). After QC, data from 61 IBD patients and 63 normal controls were used in statistical analyses. Univariate analysis (Kruskal-Wallis test) was carried out to detect differences in the serum cytokine levels between the groups and to identify potential predictors. Multivariate analysis (linear discriminant analysis) was then performed to determine differences between the cytokine profiles of the groups. Prediction models were built and found to be capable of predicting IBD with high success rate in both original and cross-validation cases (see below).

[40] The following comparisons were initially defined in order to identify biomarkers that can be used to diagnose IBD, and biomarkers that correlate with disease activity:

1. CD patients (clinical remission) vs. CD patients (active disease) - to identify biomarkers that correlate with Crohn's disease activity 2. All CD patients vs. all normal controls - to identify biomarkers that can be used to diagnose CD 3. CD patients (active disease) vs. all normal controls - to identify biomarkers that can be used to diagnose active CD 4. All UC patients vs. all CD patients - to identify biomarkers that can be used to differentiate UC from CD 5. All UC patients vs. all normal controls - to identify biomarkers that can be used to diagnose UC [41] Subsequent to the analysis of the results (see below) from the above 5 comparisons, the following additional comparisons were defined to further refine the identified biomarkers based on the observed age-dependent differences:

6. CD patients (<9 years of age) vs. normal controls (<9 years of age) - to identify age-dependent biomarkers that can be used to diagnose CD 7. CD patients (>9 years of age) vs. normal controls (>9 years of age) - to identify age-dependent biomarkers that can be used to diagnose CD 8. All CD patients vs. normal controls (<9 years of age) - to identify age-dependent biomarkers that can be used to diagnose CD 9. All CD patients vs. normal controls (>9 years of age) - to identify age-dependent biomarkers that can be used to diagnose CD 10. CD patients (active disease) vs. normal controls (<9 years of age) - to identify age-dependent biomarkers that can be used to diagnose active CD 11. CD patients (active disease) vs. normal controls (>9 years of age) - to identify age-dependent biomarkers that can be used to diagnose active CD

Univariate analysis of individual cytokine levels

[42] Using the Kruskal-Wallis test (analysis of variance by rank), the levels of each of the 78 cytokines were analyzed for all of the 11 comparisons (Tables 1-11). We set the level of significance at ≤O.Ol .

[43] As an example, Figure 1 shows the serum PLGF (placenta growth factor) levels in CD patients in clinical remission and in CD patients with active disease. The mean intensity of PLGF in the remission group (556+233) was higher than the active group (336±250). The Kruskal-Wallis test identified PLGF as the marker with highest statistical significance in Comparison 1 (CD remission vs. CD active) (Table 1), suggesting that PLGF is a potential class predictor. However, the ranges of PLGF levels for the two groups overlap significantly (Figure 1). Thus, serum PLGF level by itself is not a good predictor of Crohn's disease activity. Other analytes, such as MSP, showed some age-dependent changes in their levels in the control group (Figure 2). MSP levels were considerably higher in controls >9 years of age. In an effort to identify such age-specific biomarkers, we defined Comparisons 6-11 (see above). No single analyte exhibited a clear separation with non-overlapping levels for any of the comparisons (data not shown). Therefore, single cytokines cannot be used as a class predictor for IBD.

Mulύvariate analysis ofthe cytokine profiles

[44] We were interested to distinguish between the two groups in each comparison based on the levels of multiple cytokines simultaneously by performing linear discriminant analysis (LDA). The goal of LDA is to build a prediction model based on known cases and use it to classify unknown cases. Multiple variables (in our case, multiple cytokines) termed predictors are selected as described in Materials and Methods and used to build a prediction model. Based on a different number of cytokine classifiers, we constructed models for most comparisons (Figures 3-11). A list of the cytokine predictors used in building the cytokine classifiers and their canonical function values (CFVs) are given in Table 12. The absolute magnitude of the CFV represents the contribution or the weight of that cytokine in the classifier. A large I CFNI means that the cytokine makes a large contribution to predictions, and a small I CFNI means that the cytokine makes a small contribution.

[45] As shown in Figures 3-11, increasing the number of cytokine classifiers enabled us to build a better model for group separation. However, there is a practical limit on the number of predictors one can use to build a prediction model. As shown in Figure 12, confidence level (inversely related to the prediction rate) for correct prediction for a given model is dependent on the number of cytokine predictors used. As a general rule, when we build a classifier to distinguish two groups with a total of n subjects using d predictors, the resulting classifier would be able to classify completely random sets of data if 2*d ≥ n (Miller et al. 1979; Gardner 1987; Napnik 1995). However, we would have little or no confidence in the ability of such a model to correctly predict new cases. In other words, prediction models based on excess number of predictors frequently make incorrect predictions. The exact shape of the curves shown in Figure 12 depends on the number of classes we are trying to distinguish and the number of cases in each class. Therefore, there is a different optimal number of cytokine predictors for building a prediction model for each comparison.

[46] The ability of the prediction models to correctly classify the samples was tested. The results of these tests for each comparison are shown in Tables 13-1 to 13-9. In general, both original grouped cases and cross-validated grouped cases were predicted with a high success rate, usually >90%.

[47] There have been a number of recent reports of measuring gene expression on microarrays to classify various disease states, including, for example, the distinction of acute lymphoblastic leukemia from acute myeloid leukemia cases (Golub et al., 1999; Hakak et al. 2001; Ramaswamy et al. 2001). In this study, we employed an antibody microarray designed to detect 78 different proteins (Schweitzer et al., 2002) to determine the cytokine profiles of IBD (CD and UC) patients and normal controls. We defined 11 comparisons to identify biomarkers that can be used to diagnose IBD, and biomarkers that correlate with disease activity. Based on differences detected between the groups, we were able to build prediction models for most of the comparisons, using different number of cytokine predictors (Tables 13-1 tol3-9, Table 12, and Figures 3-11).

[48] For Comparison 1, CD patients (clinical remission) vs. CD patients (active disease), we were able to build a 10-cytokine classifier that correctly categorized 92.7% of original grouped cases and 85.4% of cross- validated grouped cases. The number of cytokines used in this classifier is within the range of maximum confidence level, as shown in Figure 12. This 10-cytokine classifier was considerably more successful in correctly categorizing patients in remission than those with active CD (Table 13-1). Using LDA, we were able to identify the following 10 biomarkers that correlate with Crohn's disease activity: BDNF, I- 309, IL-17, MCP-1, MPIF-1, PLGF, TARC, TRAIL, SCD23, UPAR. One of them, placenta growth factor (PLGF), which had one of the highest |CFV|, was identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 1). PLGF and vascular endothelial growth factor (NEGF) represent two closely related angiogenic growth factors active as homodimers or heterodimers (Niglietto et al. 1997). PLGF strongly potentiates both the proiiferative and the permeabilization effects exerted by NEGF on the vascular endothelium (Niglietto et al. 1995). The serum levels of NEGF have been found to be increased and to correlate with disease activity in patients with inflammatory bowel diseases, indicating a role for this cytokine in promoting inflammation in these chronic inflammatory diseases. The mechanism of action may be through increasing the vascular permeability and/or wound healing via its proangiogenic effects (Bousvaros et al. 1999). Interestingly, monocytes express the VEGF receptor Flt-1, and this receptor specifically binds also the NEGF homolog PLGF. Both NEGF and PLGF stimulate tissue factor production and chemotaxis in monocytes at equivalent doses. It has been suggested that Flt-1 is a functional receptor for VEGF and PLGF in monocytes and endothelial cells and a mediator of monocyte recruitment and procoagulant activity (Clauss et al. 1996). Blood monocytes are recruited to the inflammatory bowel disease mucosa, and the phenotype of the recently-arrived monocytes indicates their susceptibility to stimulation by lipopolysaccharide, suggesting a mechanism for the continuing inflammation in the bacterial product-rich milieu of IBD (Grimm et al. 1995). If used individually, PLGF will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 10- cytokine classifier, the correct classification rate is considerably improved. PLGF was also identified as a classifier cytokine in Comparison 8, all CD patients vs. normal controls (<9 years of age). [49] For Comparison 2, all CD patients vs. all normal controls, we were able to build a 22-cytokine classifier that correctly categorized 96.2% of original grouped cases and 93.3% of cross-validated grouped cases. The number of cytokines used in this classifier is within the range of maximum confidence level, as shown in Figure 12. This 22-cytokine classifier was almost equally successful in correctly categorizing individuals from both groups (Table 13-2). Using LDA, we were able to identify the following 22 biomarkers that could be used to categorize CD and normal controls: SCD23, TGF-B1, BDNF, CNTF, ENA-78, EOT2, FGF-6, G-CSF, GM-CSF, IL-11, IL-12P70, IL-15, IL-1A, IL-5, MCP-3, MIP-1D, MSP, NT3, PARC, RANTES, SCF, TNF-R1. Four of them (G-CSF, ENA-78, PARC, TNF-R1) were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 2), with G-CSF being most significant. The amount of specific messenger RNA for G-CSF and IL-1 in lamina propria mononuclear cells from mucosa of patients with Crohn's disease has been found to be increased compared with normal controls (Pullman et al. 1992). IL-1 A was another analyte in the 22-cytokine classifier. Additional evidence for a potential role of G-CSF in CD has been suggested by (Ina et al. 1999). Mucosal specimens obtained from patients with active IBD have been found to exhibit higher levels of G-CSF and GM-CSF activity than controls. Notably, the levels of G-CSF activity were approximately 1000-fold higher than those of GM-CSF activity. GM-CSF was also one of the cytokines in the classifier. If used individually, G-CSF or other cytokines will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 22-cytokine classifier, the correct classification rate is considerably improved. G-CSF was also identified as a classifier cytokine in Comparisons 4 (all UC patients vs. all CD patients), 5 (all UC patients vs. all normal controls ) and 6 (CD patients <9 years of age vs. normal controls <9 years of age).

[50] For Comparison 3, CD patients (active disease) vs. all normal controls, we were able to build a 12-cytokine classifier that correctly categorized 93.9% of original grouped cases and 93.9% of cross-validated grouped cases. The number of cytokines used in this classifier is within the range of maximum confidence level, as shown in Figure 12. This 12-cytokine classifier was considerably more successful in correctly categorizing normal individuals than those with active CD (Table 13-3). Using LDA, we were able to identify the following 12 biomarkers that could be used to categorize active CD and normal controls: BLC, ENA-78, 1- 309, IL-10, IL-1B, IL-4, IL-7, MIG, MIP-1B, MSP, NT3, PARC. Two of them (MIG, ENA-78) were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 3). In murine models of IBD, mRNA levels encoding MIG and other chemokines (some of which, e.g. MIP-1B and TCA-3, the murine homolog of human 1-309, were also part ofthe 12-cytokine classifier) were significantly increased in chronically inflamed colons when compared with wild type mice. Interestingly, reversal of colitis by anti-IL-12 mAb was accompanied by the inhibition in the expression of MIG and other chemokines (Scheerens et al. 2001). Quantitative analysis of mRNA expression showed an up-regulation of MIG and other molecules in the colons of IL-4-treated mice with significantly accelerated development of colitis (Fort et al. 2001). Immunofluorescence staining has shown presence of epithelial neutrophil- activating peptide 78 (ENA-78) protein in > 90% of preserved epithelial cells in IBD, while in control tissues ENA-78 mRNA was not detectable, and ENA-78 protein was detectable in 0%-30% of epithelial cells, suggesting a role ofthe C-X- C chemokine ENA-78 in the pathogenesis of IBD (Z'Graggen et al. 1997). Other studies have demonstrated that ENA-78 and other chemokines are highly expressed in the intestinal mucosa in areas of active Crohn's disease and ulcerative colitis (MacDermott 1999). Chemokines, including ENA-78, have been proposed to play a central role of in the immunopathogenesis of IBD (MacDermott et al. 1998). If used individually, MIG and ENA-78 will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 12-cytokine classifier, the correct classification rate is considerably improved. ENA-78 was also identified as a classifier cytokine in Comparisons 2 (all CD patients vs. all normal controls) and 9 (all CD patients vs. normal controls >9 years of age). [51] A Venn diagram representing the overlap of the best cytokine classifiers from Comparisons 1, 2 and 3 is shown in Figure 13. Although there are no cytokines common for all 3 classifiers, there are 2 common cytokines (BDNF and sCD23) between the classifiers from Comparisons 1 and 2, one cytokine (1-309) is shared between the classifiers from Comparisons 1 and 3, and 4 cytokines (ENA-78, MSP, NT3 and PARC) are common between the classifiers from Comparisons 2 and 3.

[52] For Comparison 4, all UC patients vs. all CD patients, we were able to build a 6- cytokine classifier that correctly categorized 91.8% of original grouped cases and 90.2%o of cross-validated grouped cases. The number of cytokines used in this classifier is within the range of maximum confidence level, as shown in Figure 12. This 6-cytokine classifier was considerably more successful in correctly categorizing CD patients than UC patients (Table 13-4). Using LDA, we were able to identify the following 6 biomarkers that can be used to categorize UC and CD: BDNF, FAS, G-CSF, BLC, EOT, MEP-1A. Four of them (FAS, G-CSF, BDNF, EOT), which had the highest |CFV|, were identified also by Kruskal- Wallis test to be significantly different between the 2 groups (Table 4). It has been shown that the majority of lamina propria CD4(+) T-cells from a murine model of chronic colitis undergo Fas-mediated apoptosis in vivo (Bregenholt et al. 2001), pointing towards the Fas-FasL system as the primary apoptosis-inducing mechanism of these cells. Abnormal gut-associated lymphoid tissue system was found in UC probably due to dysregulation of T cells through Fas-Fas ligand system (Asakura et al. 1999). Fas mediated apoptosis of CD45RO+CD8+ T cells was reported to be higher in UC patients than the controls, while the number of apoptotic CD45RO+CD4+ T cells from UC mucosa was not (Suzuki et al. 2000). In UC patients, CD45RO+CD4+ T cells are less sensitive to apoptotic signals mediated by Fas, which may contribute to the pathogenesis of UC. If used individually, FAS or other cytokines will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 6-cytokine classifier, the correct classification rate is considerably improved. [53] For Comparison 5, all UC patients vs. all normal controls, we were able to build a 7-cytokine classifier that correctly categorized 97.6% of original grouped cases and 97.6% of cross-validated grouped cases. The number of cytokines used in this classifier is within the range of maximum confidence level, as shown in Figure 12. This 7-cytokine classifier was slightly more successful in correctly categorizing UC patients than normal individuals (Table 13-5). Using LDA, we were able to identify the following 7 biomarkers that could be used to categorize UC and normal controls: HCC-4, IL-7, MCP-1, MSP, IL-11, G-CSF, AR. Three of them (HCC-4, IL-7, G-CSF) were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 5). It has been shown that the serum IL-7 concentration was significantly increased in UC patients (Watanabe et al. 1997). IL-7 mRNA expression is increased in the thymus tissues from patients but decreased in the colonic mucosa. Since IL-7 is a crucial cytokine for proliferation and differentiation of T cells in the thymus, these results indicate that TL-7 may contribute to the disturbance of immune regulatory T cells in ulcerative colitis. It has also been demonstrated that IL-7 transgenic mice develop acute and chronic colitis with histopathological similarity to ulcerative colitis in humans (Watanabe et al. 1999). IL-7 stimulates the proliferation of inactivated mucosal lymphocytes but eliminates activated lymphocytes in the inflamed mucosa of human ulcerative colitis. These findings suggest that chronic inflammation in the colonic mucosa is mediated by dysregulation of epithelial cell-derived IL-7 system. If used individually, IL-7 or other cytokines will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 7-cytokine classifier, the correct classification rate is considerably improved.

[54] A Venn diagram representing the overlap of the best cytokine classifiers from Comparisons 4 and 5 is shown in Figure 14. There is 1 common cytokine (G- CSF) between the classifiers from Comparisons 4 and 5. The relevance of G-CSF to IBD was discussed above. [55] For Comparison 6, CD patients (<9 years of age) vs. normal controls (<9 years of age), we were able to build a 6-cytokine classifier that correctly categorized 100.0% of original grouped cases and 100.0% of cross- validated grouped cases. This 6-cytokine classifier was equally successful (100%) in correctly categorizing both groups (Table 13-6). Using LDA, we were able to identify the following 6 age-dependent biomarkers that could be used to categorize age-specific CD and normal controls: BDNF, EGF, G-CSF, IL-18, IL-1SR1, IL-6SR. Three of them (EGF, G-CSF, IL-18) were identified also by Kruskal-Wallis test to be 1 significantly different between the 2 groups (Table 6). The pathogenesis of both ulcerative colitis and Crohn's disease may be associated with an inability of the intestinal mucosa to protect itself from luminal challenges and/or inappropriate repair following intestinal injury (Beck and Podolsky 1999). Growth factors can be distinguished by their actions regulating cell proliferation. These factors also mediate processes such as extracellular matrix formation, cell migration and differentiation, immune regulation, and tissue remodeling. Several families of growth factors may play an important role in IBD including: epidermal growth factor family (EGF) [transforming growth factor alpha (TGF alpha), EGF itself, and others], the transforming growth factor beta (TGF beta) super family, insulinlike growth factors (IGF), fibroblast growth factors (FGF), hepatocyte growth factor (HGF), trefoil factors, platelet-derived growth factor (PDGF), vascular endothelial growth factor (VEGF) and others. Collectively these families may determine susceptibility of EBD mucosa to injury and facilitate tissue repair (Beck and Podolsky 1999). Intestinal barrier dysfunction concomitant with high levels of reactive oxygen metabolites (ROM) in the inflamed mucosa have been observed in IBD (Banan et al. 2000b). The cytoskeletal network has been suggested to be involved in the regulation of barrier function. Growth factors (epidermal growth factor (EGF) and transforming growth factor alpha (TGF-^' alpha)) protect gastrointestinal barrier integrity against a variety of noxious agents. Caco-2 monolayers were preincubated with EGF, TGF-alpha, or vehicle before incubation with ROM (H(2)O(2) or HOC1). Growth factor pretreatment decreased actin oxidation and enhanced the stable F-actin, while in concert prevented actin disruption and restored normal barrier function of monolayers exposed to ROM. Cytochalasin-D, an inhibitor of actin assembly, not only caused actin disassembly and barrier dysfunction but also abolished the protective action of EGF and TGF-alpha. Moreover, an actin stabilising agent, phalloidin, mimicked the protective actions ofthe EGF and TGF-alpha (Banan et al. 2000b). Organization and stability ofthe microtubule cytoskeleton appears to be critical to both oxidant-induced mucosal barrier dysfunction and protection of intestinal barrier mediated by growth factors. Therefore, microtubules may be useful targets for development of drugs for the treatment of IBD (Banan et al. 2000a). Epidermal growth factor, and its human homologue urogastrone (EGF/URO), are secreted by the gut-associated salivary and Brunner's glands. Recombinant EGF/URO is a powerful stimulator of cell proliferation and differentiation in the rodent and neonatal human intestine (Wright et al. 1990). Ulceration of the epithelium anywhere in the human gastrointestinal tract induces the development of a novel cell lineage from gastrointestinal stem cells. This lineage initially appears as a bud from the base of intestinal crypts, adjacent to the ulcer, and grows locally as a tubule, ramifying to form a new small gland, and ultimately emerges onto the mucosal surface. The lineage produces neutral mucin, shows a unique lectin-binding profile and immunophenotype, is nonproliferative, and contains and secretes abundant immunoreactive EGF/URO. It has been proposed that all gastrointestinal stem cells can produce this cell lineage after mucosal ulceration, secreting EGF/URO to stimulate cell proliferation, regeneration and ulcer healing. This cell lineage is very commonly associated with gastrointestinal mucosal ulceration, and a principal in vivo role for EGF/URO is to stimulate ulcer healing throughout the gut through induction of this cell lineage in the adjacent mucosa (Wright et al. 1990). The relevance of G-CSF to IBD was discussed above (see Comparison 2). Elevated expression of IL-18 mRNA and protein in intestinal mucosa, attributable to activated monocytes and macrophages in that site, has been reported in patients with IBD (Furuya et al. 2002). Bioactive IL-18 was measured in serum from patients with IBD, using an enzyme-linked immunosorbent assay (ELISA). Mean serum IL-18 concentrations in 5 patients with Crohn disease (CD) were 400 pg/mL, approximately 1.7 times higher than concentrations in 21 control subjects (p < 0.01). However, serum IL-18 was not increased in patients with ulcerative colitis (UC). These results suggest that like other T-helper type 1 (Thl) cytokines IL-18 may play a key pathogenetic role in Thl -mediated disorders, such as CD. Regulation and expression of IL-18 appears to differ between CD and UC, and serum IL-18 may be a useful clinical marker for CD (Furuya et al. 2002). Macrophages, and the macrophage-derived IL-18, play a pivotal role in the establishment of TNBS-induced colitis in mice (Kanai et al. 2001), highlighting the potential use of therapy directed against IL-18 in the treatment of patients with CD. If used individually, EGF, G-CSF, IL-18 or other cytokines will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 6-cytokine classifier, the correct classification rate is considerably improved.

[56] For Comparison 7, CD patients (>9 years of age) vs. normal controls (>9 years of age), we were able to build a 4-cytokine classifier that correctly categorized 97.2% of original grouped cases and 97.2% of cross- validated grouped cases. This 4-cytokine classifier was equally successful in correctly categorizing both groups (Table 13-7). Using LDA, we were able to identify the following 4 age-dependent biomarkers that could be used to categorize age-specific CD and normal controls: MCP-1, MSP, PARC, RANTES. Two of them (MSP and PARC) were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 7). When colitis was induced in MSP+/+ and MSP-/- mice with administration of 5% dextran sulfate sodium in drinking water for 7 d, there was no appreciable difference in clinical parameters or in colonic histology; however, MSP-/- mice tended to be more affected by dextran sulfate sodium (diarrhea, visible blood in stools, weight loss), when compared with MSP +/+ mice (Bezerra et al. 1998). If used individually, MSP or other cytokines will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 4-cytokine classifier, the correct classification rate is considerably improved. [57] For Comparison 8, all CD patients vs. normal controls (<9 years of age), we were able to build a 3-cytokine classifier that correctly categorized 86.8% of original grouped cases and 86.8% of cross-validated grouped cases. This 3-cytokine classifier was approximately equally successful in correctly categorizing both groups (Table 13-8). Using LDA, we were able to identify the following 3 age- dependent biomarkers that could be used to categorize CD and age-specific normal controls: EGF, HCC-4, PLGF. All three analytes were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 8). The potential significance of EGF and PLGF in IBD were discussed above (see Comparisons 1 and 6, respectively). If used individually, EGF and PLGF will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 3-cytokine classifier, the correct classification rate is considerably improved.

[58] For Comparison 9, all CD patients vs. normal controls (>9 years of age), we were able to build a 3-cytokine classifier that correctly categorized 89.6% of original grouped cases and 89.6% of cross-validated grouped cases. This 3-cytokine classifier was approximately equally successful in correctly categorizing both groups (Table 13-9). Using LDA, we were able to identify the following 3 age- dependent biomarkers that could be used to categorize CD and age-specific normal controls: ENA-78, MSP, PARC. All three analytes were identified also by Kruskal-Wallis test to be significantly different between the 2 groups (Table 9). The potential significance of ENA-78 and MSP in IBD were discussed above (see Comparisons 3 and 7, respectively). If used individually, ENA-78 and MSP will not be able to correctly classify samples with high success rate, but when used in combination with other cytokines in the 3-cytokine classifier, the correct classification rate is considerably improved.

[59] For Comparison 10, CD patients (active disease) vs. normal controls (<9 years of age), we used Kruskal-Wallis test to identify significantly different analytes between the 2 groups (Table 10). A number of these analytes (EGF, ENA-78, MIG, IL-18) and their potential significance in IBD were discussed above. [60] For Comparison 11, CD patients (active disease) vs. normal controls (>9 years of age), we used Kruskal-Wallis test to identify significantly different analytes between the 2 groups (Table 11). A number of these analytes (M-CSF, IL-8, MCP-1, PLGF, MIG, ENA-78, Fas, MSP) and their potential significance in IBD were discussed above.

[61] In an effort to obtain a global visual representation of the expression differences between patients with ulcerative colitis and normal subjects, a heat map of cytokine profiles of UC patients vs. normal controls was generated (Figure 15). Colors signify the different expression levels relative to the median level of the normal controls. For each cytokine, the median and the standard deviation (SD) of the normal control group were determined. The analyte level for that cytokine in each UC patient and normal control samples was then colored according to the following scheme: samples expressing median level are shown in black, those with > 2 SD above the median are shown in bright red, those with < 2 SD below the median are shown in bright green, and shades of red and green denote the values in between. The SDs for each analyte for all UC patients were summed, and analytes were sorted according to the sums, with the lowest sums shown on the left and the highest sums shown on the right. The sample IDs are shown on the left, with the top 20 rows representing UC patients, and the rest of the rows representing normal controls. The analyte abbreviations are shown at the bottom of the figure. Upon visual inspection, considerable differences in the expression patterns in UC patients and normal controls were observed for a number of analytes, for example G-CSF, EOT. The relevance of some of these cytokines in IBD was discussed above.

[62] Table 1. Comparison 1 : cytokine levels (intensity units) in serum samples from CD patients (clinical remission) vs. CD patients (active disease), and corresponding P-values based on Kruskal-Wallis test

Table 6. Comparison 6: cytokine levels (intensity units) in serum samples from CD patients (<9 years of age) vs. normal confrols (<9 years of age), and corresponding P-

Table 7. Comparison 7: cytokine levels (intensity units) in serum samples from CD patients (>9 years of age) vs. normal controls (>9 years of age), and corresponding P- values based on Kruskal-Wallis test (p≤O.01)

Table 8. Comparison 8: cytokine levels (intensity units) in serum samples from all CD patients vs. normal controls (<9 years of age), and corresponding P-values based on Kruskal-Wallis test

Table 9. Comparison 9: cytokine levels (intensity units) in serum samples from all CD patients vs. normal controls (>9 years of age), and corresponding P-values based on Kruskal-Wallis test

Table 13-1. Prediction rates of LDA for Comparison 1. All numbers are in %. Original Cases Cross-Validation Remission Active Remission Active 4-cytokine classifier Remission 82 18 82 18 Active 32 68 37 63 10-cytokine classifier Remission 100 0 95 5 Active 16 84 26 74

Table 13-2. Prediction rates of LDA for Comparison 2. All numbers are in %. Original Cases Cross-Validation All CD All normal All CD All normal 11 -cytokine classifier All CD 90 10 88 12 All normal 11 89 11 89 22-cytokine classifier All CD 98 2 93 7 All normal 5 95 6 94

Table 13-3. Prediction rates of LDA for Comparison 3. All numbers are in %. Original Cases Cross-Validation CD active All normal CD active All normal 10-cytokine classifier CD active 84 16 74 26 All normal 0 100 3 97 12-cytokine classifier CD active 84 16 84 16 All normal 3 97 3 97

Table 13-4. Prediction rates of LDA for Comparison 4. All numbers are in %. Original Cases Cross-Validation All CD AII UC All CD AII UC 3-cytokine classifier All CD 93 7 90 10 AII UC 15 85 20 80 6-cytokine classifier All CD 98 2 95 5 AII UC 20 80 20 80 Table 13-5. Prediction rates of LDA for Comparison 5. All numbers are in % Original Cases Cross-Validation All normal AII UC All normal All UC 3-cytokine classifier All normal 98 2 98 2 AII UC 25 75 25 75 7-cytokine classifier All norm a 1 97 3 97 3 AII UC 0 100 0 100

Table 13-6. Prediction rates of LDA for Comparison 6. All numbers are in %. Original Cases Cross-Validation CD < 9 years Normal < 9 yrs CD < 9 years Normal < 9 yrs 3-cytokine classifier CD < 9 years 88 12 88 12 Normal < 9 yrs 0 100 0 100 6-cytokine classifier CD < 9 years 100 0 100 0 Normal < 9 yrs 0 100 0 100

Table 13-7. Prediction rates of LDA for Comparison 7. All numbers are in %. Original Cases Cross-Validation CD > 9 years Normal > 9 yrs CD > 9 years Normal > 9 yrs 2-cytokine classifier CD > 9 years 86 14 86 14 Normal > 9 yrs 11 89 14 86 4-cytokine classifier CD > 9 years 97 3 i 97 3 Normal > 9 yrs 3 97 3 97

Table 13-8. Prediction rates of LDA for Comparison 8. All numbers are in %. Original Cases Cross-Validation All CD Normal < 9 yrs All CD Normal < 9 yrs 3-cytokine classifier All CD 88 12 88 12 Normal < 9 yrs 15 85 15 85

Table 13-9. Prediction rates of LDA for Comparison 9. All numbers are in %. Original Cases Cross-Validation All CD Normal > 9 yrs All CD Normal > 9 yrs 3-cytokine classifier All CD 90 10 90 10 Normal > 9 yrs 11 89 11 89 References:

Asakura H, Suzuki A, Ohtsuka K, Hasegawa K, Sugimura K (1999) Gut-associated lymphoid tissues in ulcerative colitis. JPEN J Parenter Enteral Nufr 23:S25-8 Banan A, Choudhary S, Zhang Y, Fields JZ, Keshavarzian A (2000a) Oxidant-induced intestinal barrier disruption and its prevention by growth factors in a human colonic cell line: role ofthe microtubule cytoskeleton. Free Radic Biol Med 28:727-38 Banan A, Zhang Y, Losurdo J, Keshavarzian A (2000b) Carbonylation and disassembly ofthe F-actin cytoskeleton in oxidant induced barrier dysfunction and its prevention by epidermal growth factor and transforming growth factor alpha in a human colonic cell line. Gut 46:830-7 Beck PL, Podolsky DK (1999) Growth factors in inflammatory bowel disease. Inflamm Bowel Dis 5:44-60 Bezerra JA, Carrick TL, Degen JL, Witte D, Degen SJ (1998) Biological effects of targeted inactivation of hepatocyte growth factor-like protein in mice. J Clin Invest 101:1175-83 Bousvaros A, Leichtner A, Zurakowski D, Kwon J, Law T, Keough K, Fishman S (1999) Elevated serum vascular endothelial growth factor in children and young adults with Crohn's disease. Dig Dis Sci 44:424-30 Bregenholt S, Claesson MH (1998) Splenic T helper cell type 1 cytokine profile and extramedullary haematopoiesis in severe combined immunodeficient (scid) mice with inflammatory bowel disease (IBD). Clin Exp Immunol 111 :166-72 Bregenholt S, Petersen TR, Claesson MH (2001) The majority of lamina propria CD4(+) T-cells from scid mice with colitis undergo Fas-mediated apoptosis in vivo. Immunol Lett 78:7-12 Clauss M, Weich H, Breier G, Knies U, Rockl W, Waltenberger J, Risau W (1996) The vascular endothelial growth factor receptor Flt-1 mediates biological activities. Implications for a functional role of placenta growth factor in monocyte activation and chemotaxis. J Biol Chem 271:17629-34 Emery P, Reginster JY, Appelboom T, Breedveld FC, Edelmann E, Kekow J, Malaise M, Mola EM, Montecucco C, Sanda M, Sany J, Scott DL, Serni U, Seydoux G (2001) WHO Collaborating Centre consensus meeting on anti-cytokine therapy in rheumatoid arthritis. Rheumatology (Oxford) 40:699-702

Fiocchi C, Fukushima K, Strong SA, Ina K (1996) Pitfalls in cytokine analysis in inflammatory bowel disease. Aliment Pharmacol Ther 10 Suppl 2:63-9; discussion 70-1 Fort M, Lesley R, Davidson N, Menon S, Brombacher F, Leach M, Rennick D (2001) IL- 4 exacerbates disease in a Thl cell transfer model of colitis. J immunol 166:2793- 800 Furuya D, Yagihashi A, Komatsu M, Masashi N, Tsuji N, Kobayashi D, Watanabe N (2002) Serum interleukin-18 concentrations in patients with inflammatory bowel disease. J l munother 25 Suppl LS65-7

Gardner E (1987) Maximum Storage Capacity in Neural Networks. Euro-physics letters 4:481-485 Grimm MC, Pullman WE, Bennett GM, Sullivan PJ, Pavli P, Doe WF (1995) Direct evidence of monocyte recruitment to inflammatory bowel disease mucosa. J Gastroenterol Hepatol 10:387-95 Hakak Y, Walker JR, Li C, Wong WH, Davis KL, Buxbaum JD, Haroutunian V, Fienberg AA (2001) Genome-wide expression analysis reveals dysregulation of myelination-related genes in chronic schizophrenia. Proc Natl Acad Sci U S A 98:4746-51 Heresbach D, Semana G, Gosselin M, Bretagne MG (1999) An immunomodulation strategy targeted towards immunocompetent cells or cytokines in inflammatory bowel diseases (IBD). Eur Cytokine Netw 10:7-15 Hyams JS, Ferry GD, Mandel FS, Gryboski JD, Kibort PM, Kirschner BS, Griffiths AM, Katz AJ, Grand RJ, Boyle JT, et al. (1991) Development and validation of a pediatric Crohn's disease activity index. J Pediafr Gastroenterol Nutr 12:439-47 Ina K, Kusugami K, Hosokawa T, Imada A, Shimizu T, Yamaguchi T, Ohsuga M, Kyokane K, Sakai T, Nishio Y, Yokoyama Y, Ando T (1999) Increased mucosal production of granulocyte colony-stimulating factor is related to a delay in neutrophil apoptosis in Inflammatory Bowel disease. J Gastroenterol Hepatol 14:46-53 Kanai T, Watanabe M, Okazawa A, Sato T, Yamazaki M, Okamoto S, Ishii H, Totsuka T, Iiyama R, Okamoto R, Dceda M, Kurimoto M, Takeda K, Akira S, Hibi T (2001) Macrophage-derived IL-18-mediated intestinal inflammation in the murine model of Crohn's disease. Gasfroenterology 121:875-88 Karlinger K, Gyorke T, Mako E, Mester A, Tarjan Z (2000) The epidemiology and the pathogenesis of inflammatory bowel disease. Eur J Radiol 35:154-67 Klebl FH, Olsen JE, Jain S, Doe WF (2001) Expression of macrophage-colony stimulating factor in normal and inflammatory bowel disease intestine. J Pathol 195:609-15 Kucharzik T, Stoll R, Lugering N, Domschke W (1995) Circulating antiinflammatory cytokine IL-10 in patients with inflammatory bowel disease (IBD). Clin Exp Immunol 100:452-6 Ligumsky M, Kuperstein V, Nechushtan H, Zhang Z, Razin E (1997) Analysis of cytokine profile in human colonic mucosal Fc epsilonRI-positive cells by single cell PCR: inhibition of IL-3 expression in steroid-treated IBD patients. FEBS Lett 413:436-40 MacDermott RP (1999) Chemokines in the inflammatory bowel diseases. J Clin Immunol 19:266-72 MacDermott RP, Sanderson ER, Reinecker HC (1998) The central role of chemokines (chemotactic cytokines) in the immunopathogenesis of ulcerative colitis and Crohn's disease. Inflamm Bowel Dis 4:54-67 McCormack G, Moriarty D, O'Donoghue DP, McCormick PA, Sheahan K, Baird AW (2001) Tissue cytokine and chemokine expression in inflammatory bowel disease. Inflamm Res 50:491-5 Miller CH, Graham JB, Goldin LR, Elston RC (1979) Genetics of classic von Willebrand's disease. II. Optimal assignment ofthe heterozygous genotype (diagnosis) by discriminant analysis. Blood 54:137-45 Pullman WE, Elsbury S, Kobayashi M, Hapel AJ, Doe WF (1992) Enhanced mucosal cytokine production in inflammatory bowel disease. Gastroenterology 102:529-37

Ramaswamy S, Tamayo P, Rifldn R, Mukherjee S, Yeang CH, Angelo M, Ladd C, Reich M, Latulippe E, Mesirov JP, Poggio T, Gerald W, Loda M, Lander ES, Golub TR (2001) Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci U S A 98:15149-54

Raza A (2000) Anti-TNF therapies in rheumatoid arthritis, Crohn's disease, sepsis, and myelodysplastic syndromes. Microsc Res Tech 50:229-35

Russel MG, Stockbrugger RW (1996) Epidemiology of inflammatory bowel disease: an update. Scand J Gastroenterol 31:417-27

Russel MG, Stockbrugger RW (2001) [Epidemiological developments and insights in chronic inflammatory bowel diseases]. Ned Tijdschr Geneeskd 145:1448-52

Scheerens H, Hessel E, de Waal-Malefyt R, Leach MW, Rennick D (2001) Characterization of chemokines and chemokine receptors in two murine models of inflammatory bowel disease: IL-10-/- mice and Rag-2-/- mice reconstituted with CD4+CD45RBhigh T cells. Eur J Immunol 31:1465-74

Schweitzer B, Roberts S, Grimwade B, Shao W, Wang M, Fu Q, Shu Q, Laroche I, Zhou Z, Tchernev NT, Christiansen J, Nelleca M, Kingsmore SF (2002) Multiplexed protein profiling on microarrays by rolling-circle amplification. Nat Biotechnol 20:359-65

Seo M, Okada M, Yao T, Ueki M, Arima S, Okumura M (1992) An index of disease activity in patients with ulcerative colitis. Am J Gastroenterol 87:971-6

Suzuki A, Sugimura K, Ohtsuka K, Hasegawa K, Suzuki K, Ishizuka K, Mochizuki T, Honma T, Narisawa R, Asakura H (2000) Fas/Fas ligand expression and characteristics of primed CD45RO+ T cells in the inflamed mucosa of ulcerative colitis. Scand J Gastroenterol 35:1278-83

Vapnik V (1995) Estimation of Dependences Based on Empirical Data. Springer-Nerlag, New York

Viglietto G, Maglione D, Rambaldi M, Cerutti J, Romano A, Trapasso F, Fedele M, Ippolito P, Chiappetta G, Botti G, et al. (1995) Upregulation of vascular endothelial growth factor (VEGF) and downregulation of placenta growth factor (P1GF) associated with malignancy in human thyroid tumors and cell lines. Oncogene 11:1569-79

Viglietto G, Romano A, Manzo G, Chiappetta G, Paoletti I, Califano D, Galati MG, Mauriello V, Bruni P, Lago CT, Fusco A, Persico MG (1997) Upregulation ofthe angiogenic factors P1GF, NEGF and their receptors (Flt-1, Flk-1/KDR) by TSH in cultured thyrocytes and in the thyroid gland of thiouracil-fed rats suggest a TSH- dependent paracrine mechanism for goiter hypervascularization. Oncogene 15:2687-98

Watanabe M, Ueno Y, Yamazaki M, Hibi T (1999) Mucosal IL-7-mediated immune responses in chronic colitis-IL-7 transgenic mouse model. Immunol Res 20:251-9

Watanabe M, Watanabe Ν, Iwao Y, Ogata H, Kanai T, Ueno Y, Tsuchiya M, Ishii H, Aiso S, Habu S, Hibi T (1997) The serum factor from patients with ulcerative colitis that induces T cell proliferation in the mouse thymus is interleukin-7. J Clin Immunol 17:282-92 Wright NA, Pike C, Elia G (1990) Induction of a novel epidermal growth factor-secreting cell lineage by mucosal ulceration in human gasfrointestinal stem cells. Nature 343:82-5 Z'Graggen K, Walz A, MazzuccheUi L, Strieter RM, Mueller C (1997) The C-X-C chemokine ENA-78 is preferentially expressed in intestinal epithelium in inflammatory bowel disease. Gastroenterology 113:808-16

Claims

We Claim:

1. A method of determining whether a Crohn's Disease patient is in clinical remission or has active disease, comprising: determining levels of at least four of a set of proteins in a body fluid sample of said patient, wherein said set of proteins consists of BDNF, I- 309, 11-17, MCP-1, MPIF-1, PLGF, TARC, TRAIL, SCD23, and UPAR; comparing levels determined for said Crohn's Disease patient with levels determined for a population of known Crohn's Disease patients in clinical remission or for a population of known Crohn's Disease patients with active disease; identifying said patient as having active disease if levels ofthe at least four proteins determined in the body fluid sample is significantly lower than the levels ofthe population of Crohn's Disease patients in clinical remission or if levels ofthe at least four proteins determined in the body fluid sample statistically matches the levels ofthe population of Crohn's Disease patients with active disease; and identifying said patient as being in clinical remission if levels ofthe at least four proteins determined in the body fluid sample is significantly higher than the levels ofthe population of Crohn's Disease patient in active disease or if levels ofthe set of proteins determined in the body fluid sample statistically matches the levels ofthe population of Crohn's Disease patients in clinical remission.

2. The method of claim 1 wherein the patient is identified as being in clinical remission.

3. The method of claim 1 wherein the at least four proteins comprise PLGF.

4. The method of claim 1 wherein levels of all proteins ofthe set are determined.

5. A method of diagnosing a patient as having Crohn's Disease, comprising: determining levels of a subset comprising at least four proteins in a body fluid sample of a patient, wherein said subset is selected from the set of proteins consisting of : SCD23, TGF-B1, BDNF, CNTF, ENA-78, EOT2, FGF-6, G-CSF, GM-CSF, IL-11, IL-12P70, IL-15, IL-1A, IL-5, MCP-3, MIP-ID, MSP, NT3, PARC, RANTES, SCF, and TNF-R1; comparing levels determined for the patient with levels determined for a population of normal confrols or for a population of known Crohn's Disease patients; identifying the patient as having Crohn's Disease if levels ofthe subset of proteins in the patient is significantly different than levels determined for the population of normal controls or if the levels ofthe subset of proteins in the patient statistically matches levels determined for the population of known Crohn's Disease patients.

6. The method of claim 5 wherein the at least four proteins comprise G-CSF, ENA- 78, PARC, and TNF-R1.

7. The method of claim 5 wherein the at least four proteins comprise G-CSF.

8. The method of claim 5 wherein levels of all proteins of the set are determined.

9. A method of diagnosing a Crohn's Disease patient with active disease, comprising: determining levels of a subset comprising at least two proteins in a body fluid sample of a patient, wherein said subset is selected from the set of proteins consisting of BLC, ENA-78, 1-309, IL-10, IL-1B, IL-4, IL-7, MIG, MEP-1B, MSP, NT3, PARC; comparing levels determined for the patient with levels determined for a population of normal confrols or for a population of known Crohn's Disease patients with active disease; identifying the patient as having Crohn's Disease with active disease if levels ofthe subset of proteins in the patient is significantly different than levels determined for the population of normal controls or if the levels of the subset of proteins in the patient statistically matches levels determined for the population of known Crohn's Disease patients with active disease.

10. The method of claim 9 wherein levels of all proteins ofthe set are determined.

11. The method of claim 9 wherein the at least two proteins comprise MIG and ENA78.

12. A method of determining whether a patient has Ulcerative Colitis or Crohn's Disease, comprising: determining levels of at least four of a set of proteins in a body fluid sample of said patient, wherein said at least four proteins are selected from the set of proteins consisting of: BDNF, FAS, G-CSF, BLC, EOT, and MIP-1A; comparing levels determined for said patient with levels determined for a population of known Crohn's Disease patients or for a population of known Ulcerative Colitis patients; identifying said patient as having Crohn's Disease if levels ofthe at least four proteins determined in the body fluid sample is significantly different than the levels ofthe population of known Ulcerative Colitis patients or if levels ofthe at least four proteins determined in the body fluid sample statistically matches the levels ofthe population of known Crohn's Disease patients; and identifying said patient as having Ulcerative Colitis if levels ofthe at least four proteins determined in the body fluid sample is significantly different than the levels ofthe population of known Crohn's Disease patients or if levels ofthe at least four proteins determined in the body fluid sample statistically matches the levels ofthe population of known Ulcerative Colitis patients.

13. The method of claim 12 wherein the at least four proteins comprise FAS, G-CSF, BDNF, and EOT.

14. The method of claim 12 wherein levels of all proteins ofthe set are determined.

15. A method of diagnosing a patient as having Ulcerative Colitis, comprising: determining levels of at least three of a set of proteins in a body fluid sample of a patient, wherein said set of proteins consists of HCC-4, IL-7, MCP-1, MSP, IL-11, G-CSF, and AR; comparing levels determined for the patient with levels determined for the at least three proteins in a population of known Ulcerative Colitis patients or a population of normal control patients; identifying the patient as having Ulcerative Colitis if levels ofthe at least three proteins in the patient is significantly different than the levels in the population of normal control patients or if the levels statistically match those in the population of known Ulcerative Colitis patients.

16. The method of claim 15 wherein levels of all proteins ofthe set are determined.

17. The method of claim 15 wherein the at least three proteins comprises HCC-4, IL- 7, and G-CSF.

18. A method of diagnosing a Crohn's Disease patient younger than 9 years of age, comprising: determining levels of at least three proteins in a body fluid sample of a patient younger than 9 years of age, wherein said at least three proteins are selected from a set of proteins consisting of: BDNF, EGF, G-CSF, IL-18, IL-lSRl, and IL-6SR; comparing levels determined for the patient with levels determined for the at least three proteins in a population of known Crohn's Disease patients younger than 9 years of age or in a population of normal control patients younger than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe at least three proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients.

19. The method of claim 18 wherein levels of all proteins ofthe set are determined.

20. The method of claim 18 wherein the at least three proteins comprise EGF, G-CSF, and IL-18.

21. A method of diagnosing a Crohn's Disease patient older than 9 years of age, comprising: determining levels of at least two proteins in a body fluid sample of a patient older than 9 years of age, wherein said at least two proteins are selected from a set of proteins consisting of: MCP-1, MSP, PARC, and RANTES; comparing levels determined for the patient with levels determined for the at least two proteins in a population of known Crohn's Disease patients older than 9 years of age or in a population of normal control patients older than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe at least two proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients.

22. The method of claim 21 wherein levels of all proteins ofthe set are determined.

23. The method of claim 21 wherein the at least two proteins comprise MSP and PARC.

24. A method of diagnosing a Crohn's Disease patient, comprising: determining levels of EGF, HCC-4, and PLGF; comparing levels determined for the patient with levels determined for the three proteins in a population of known Crohn's Disease patients or in a population of normal control patients younger than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe three proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients.

25. A method of diagnosing a Crohn's Disease patient, comprising: determining levels of at least two proteins in a body fluid sample of a patient, wherein said at least two proteins are selected from a set of proteins consisting of: ENA-78, MSP, and PARC; comparing levels determined for the patient with levels determined for the at least three proteins in a population of known Crohn's Disease patients or in a population of normal control patients older than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe at least three proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients.

26. The method of claim 25 wherein the at least two proteins comprise ENA-78.

27. The method of claim 25 wherein the at least two proteins comprise MSP.

28. The method of claim 25 wherein levels of all proteins ofthe set are determined.

29. A method of diagnosing a Crohn's Disease patient with active disease, comprising: determining levels of one or more proteins in a body fluid sample of a patient, wherein said proteins are selected from a set of proteins consisting of: EGF, ENA-78, MIG, and IL-18; comparing levels determined for the patient with levels determined for the set of proteins in a population of known Crohn's Disease patients with active disease or in a population of normal control patients younger than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients with active disease.

30. The method of claim 29 wherein levels of at least two proteins ofthe set are determined.

31. The method of claim 29 wherein levels of at least three proteins of the set are determined.

32. The method of claim 29 wherein levels of all proteins ofthe set are determined.

33. A method of diagnosing a Crohn's Disease patient with active disease, comprising: determining levels of one or more proteins in a body fluid sample of a patient, wherein said proteins are selected from a set of proteins consisting of: M-CSF, IL-8, MCP-1, PLGF, MIG, ENA-78, Fas, and MSP; comparing levels determined for the patient with levels determined for the set of proteins in a population of known Crohn's Disease patients with active disease or in a population of normal control patients older than 9 years of age; identifying the patient as having Crohn's Disease if levels ofthe proteins in the patient are statistically different than the levels in the population of normal control patients or if the levels statistically match the levels in the population of known Crohn's Disease patients with active disease.

34. The method of claim 33 wherein levels of at least 2 proteins ofthe set are determined.

35. The method of claim 33 wherein levels of at least 4 proteins ofthe set are determined.

36. The method of claim 33 wherein levels of at least 6 proteins ofthe set are determined.