Discovery of proteic biomarkers for Crohn Disease and Ulcerative Colitis by SELDI-TOF-MS M-A. Meuwis 1, D de Seny 1, M Fillet 1, E Louis 2, J Belaiche 2, P Geurts 3,L Wehenkel 3, M Malaise 4, M-P. Merville 1. INTRODUCTION 1 Laboratory of Medical Chemistry and Human Genetics, CTCM, CHU Sart-Tilman, 4000 Liege 2 Stochastic Methods, Sart Tilman 4000 LG. 3 Hepato-Gastroenterology. CHU Sart Tilman Lg. 4 Clinical Sciences- Rhumatology, CHU, Sart Tilman, Lg. CBIG Université de Liège. Different types of surfaces interactions : Ion exchange, hydrophobe, normal phase, IMAC, (affinity antibody capture…) Crohn Disease ( CD) and Ulcerative Colitis (UC) both generally known as Inflammatory Bowel Disease (IBD) are chronic autoimmune inflammatory pathologies affecting the gastro intestinal tract. Their ethiopathogenesis has not been fully elucidated and involve a complex interplay among genetic, environmental, pathogenic and immune factors. The still growing knowledge in the ethiology of these disorders gave rise to new promising treatments. Nevertheless, the success of those drugs are cases dependent: CD or UC. Therefore, accurate and early diagnosis is a real important step in circumventing these pathologies. Today, clinical diagnosis are made on many biological data such as CRP, ASCA, ANCA determinations, according to severity of symptoms ( recorded in Harvey-Bradshow test) or with invasive techniques as gastro-endoscopies. No simple, rapid, unique and efficient technique is able to discriminate IBD from other inflammatory diseases (as infectious colitis) or among IBD itself: CD versus UC. Here, we present a strategy of sera protein profiling on SELDI-TOF ( Surface Enhanced Laser Desorption-Ionization, Time of Flight Mass Spectrometry ) and the downstream statistical analysis. Methods : Protein profiling by SELDI-TOF-MS (Surface Enhanced Lazer Desorption Ionisation - Time Of Flight - Mass Spectrometry) Detector Laser TOF-MS Gel View Sera sample loadded on chip in specific conditions ( pH, salinity, detergent…) Washing steps discarding unbonded material Addition of energy absorbing molecule, EAM PROFILESPROFILES Technic: SELDI -TOF-MS « reading » of chip: Comparaisons of profiles : on « ion exchange » surfaces : CM10 – Q10 CROHN (C) - 30 samples UHRC (UC) - 30 samples healthy controls (HC) - 30 samples non IBD inflammatory controls (IC) - 30 samples 120 samples, in quadruplicate, on 2 different surfaces = 960 spectra !!!!!!!!!!!!!!! Many comparisons : CD vs UC CD vs (HC, IC, UC) UC vs (HC, IC, UC) IBD (UC, CD) vs ( IC, HC) To discover powerful biomarkers able to discriminate sera of patients suffering from CROHN disease : CD, from patients UC (Ulcero Hemoragic Recto Colitis) or sera from IBD patients (CD and UC) from other inflammatory diseases (IC) and healthy controls (HC). IBD « pValue » calculation : by Biomarker Wizards Ciphergen Models of Classification based on calculations of Decision Trees (unique or multiple) by different statistical approches : (Geurts et al., 2004) Note : Calculations are made on 2 sets of data : integrated peaks and « raw data » = every m/z increments of the spectra. Probability of missclassification - good if <10 -3 Sensitivity (true positives) and Specificity ( true negatives) of the models Multiple decision trees : by boosting and extra tree are the best methods and are selfs validated statistical analysis comparison : Q10 CM10 specificity sensitivity specificity sensitivity CD vs UC % 85.00% 90.83% CD vs (HC, IC, UC)95.83% 70.83% 92.76% 77.50% UC vs (HC, IC, UC) 92.22% 37.50% 96.38% 43.33% IBD (UC, CD) vs ( IC, HC)88.75% 82.92% 90.42% 86.25% Q10 : 1469 CD vs UC = – CD vs IC <10 –12 et UC vs IC <10 – CD vs UC <10 –12 et IC = 6.10 – CD vs UC = 8.10 –6 et IBD vs IC = –5 CM CD vs UC <10 – CD vs UC <10 – CD vs UC <10 – IBD vs IC <10 –12 et CD vs UC= –6 m/z Values = peaks = proteins are not so powerful in discriminating taken independently. All together combined in « models », they are a lot more efficient and discrimination of samples is obtained with good scores of specificity and sensitivity. Combination of the 2 models corresponding to the 2 surfaces : Q10 and CM10 would may be increase specificity and sensitivity !?! * Validation of our « models » on a new batch of samples (10 of each categories) * Correlation with existing tests (ANCA - ASCA – CRP) * Purification of the « more powerful » proteins in terms of discrimination and their identification by MS-MS. AIMS STRATEGY : RESULTS : RESULTS : STATISTICS : PERSPECTIVES : 8 7 CM10: Intensity mean of peaks per categories: Q 10 : Intensity mean of peaks per categories: Classification of m/z values used to build this model with « boosting » and associated « pValue » illustrating their power of discrimination : Importance of each value of m/z on each surface and for every comparisons: CD vs UC CD vs UC, IC, HC IBD vs HC, IC UC vs CD, IC, HC CD vs UCIBD vs HC, IC CD vs UC, IC, HCUC vs CD, IC, HC Specificity and sensitivity of the proposed models by « boosting » : Trace View IBD HC