Proteomics Using Surface Enhanced Laser Desorption Ionization-Time Of Flight-Mass Spectrometry for the Diagnosis of Crohn's Disease and Ulcerative Colitis

INTRODUCTION: The diagnosis of Crohn s disease (CD) and ulcerative colitis (UC), the two main inflammatory bowel diseases (IBD), still relies on invasive diagnostic test including endoscopy and histology. The best blood markers, pANCA and ASCA have relatively good specificity (around 80%) but rather low sensitivity (around 50%), making them weak diagnostic tools. The aim of our work was to use sera protein profiling on SELDI-TOF (Surface Enhanced Laser Desorption-Ionization, Time of Flight Mass Spectrometry ) to obtain relevant markers for the diagnosis of CD and UC.

AIMS & METHODS: We studied protein profiles obtained from four categories of patients: CD, UC, HC (healthy controls) and ID (other inflammatory diseases), with 30 patients in each groups. These protein profiles were assessed on two types of surfaces: CM10 and Q10. We analyzed results by means of two different statistical approaches : p Value determination on integrated peaks (each corresponding to some form of protein) and calculation of multiple decision trees ( extra-trees or boosting methods) built on total data set as well as on integrated peaks.

RESULTS: comparing IBD as a whole, to HC and ID, decision trees gave a specificity of 88.8% on Q10 chips and 90.4% on CM10 chips and a sensitivity of 82.9% and 86.3%, respectively. When CD was compared to the 3 other groups, we reached 95.8% and 92.8% specificity and 70.8% and 77.7 % of sensitivity on Q10 and CM10 respectively .When UC was Compared to the 3 other groups, sensitivity was lower, around 40%. Nevertheless, the specificity is still very good with 92.2% on Q10 and 96.4% on CM10 chips. When UC was compared to CD we obtained models with 80% and 85% specificity, while the sensitivities were 81.6% and 90.8% on Q10 and CM10, respectively. The peaks occupying important places in the decision trees had also very significant p Value in individual analysis.

CONCLUSION: Using two different surfaces to assess protein profiles, we obtained two predictive models of classification which are around 80% specific and sensitive. These models could lead, after cross validation and blind tests on new samples batches, to a very robust and easy-to-use diagnosis tool. Moreover, these models revealed many discriminating peaks and the downstream purification and identification of their associated protein could help us to better understand such diseases and to highlight new therapeutic targets.