FRI0009 MOLECULAR PROFILING OF CIRCULATING B-LYMPHOCYTES REVEALS THE SUPERIOR PERFORMANCE OF METHYLOME OVER TRANSCRIPTOME DATA FOR DISCRIMINATING RHEUMATOID ARTHRITIS PATIENTS IN AN EARLY ARTHRITIS CLINIC: IMPLICATIONS FOR TRANSLATING "BIG DATA" INTO CLINICALLY USEFUL TOOLS. (June 2019)

• Record Type: Journal Article
• Authors: Naamane, Najib; Thalayasingam, Nishanthi; Nair, Nisha; Clar, Alexander; Anderson, Amy; Lendrem, Dennis; Reynard, Louise; Eyre, Stephen; Barton, Anne; Isaacs, John; Pratt, Arthur
• Is Part Of: Annals of the rheumatic diseases. Volume 78(2019)Supplement 2
• Page Start: 665
• Page End: 666
• Publication Date: 2019-06
• Subjects: Rheumatism -- Periodicals; 616.723005
• DOI: 10.1136/annrheumdis-2019-eular.7074
• Languages: English
• ISSNs: 0003-4967

Abstract

Abstract : Background: Defining optimal strategies for translating "big datasets" of transcriptome and methylome data into clinically valuable tools in RA will benefit from comparisons where potential confounders – including therapeutic background, disease phase and cell substrate heterogeneity – are controlled as far as possible. Objectives: We have obtained paired B- and CD4+ T-lymphocyte whole genome expression and methylation data from drug-naïve early arthritis clinic patients at a single centre. Focussing on B-lymphocyte data we here ask which of these datasets has most value in discriminating early RA – and the additive value of combining them. Methods: CD19+ B-lymphocytes were isolated by positive selection from fresh peripheral blood of 90 drug-naïve patients attending the Newcastle Early Arthritis Clinic (NEAC), comprising 36 RA patients and 54 disease controls matched, so far as possible, for age, sex and acute phase response. Paired RNA and DNA extracted. Gene expression was profiled using the Human HT12 v4 BeadChip, and DNA methylation at >850, 000 CpG sites quantified with the MethylationEPIC array (both Illumina). Gene expression and/or DNA methylation classifiers for RA prediction were developed based on a combined approach of classification algorithm selection and hyperparameter optimisation. The Sequential Model-Based Optimization procedure was used to automatically select the best model configuration amongst 10 feature selection filters, 2 class imbalance correction techniques, 3 data cleaning methods and 4 classification algorithms along with their respective hyperparameters. Model optimisation was performed using a 10-fold Cross Validation (CV) while a 10 times repeated 10-fold CV was used for the predictive performance evaluation. Receiver operating characteristic (ROC) and precision recall (PR) curves were used to directly compare the classification potential of molecular approaches. Results: 12, 035 gene expression features and 714, 486 methylation features were available for classifier building after quality control in our early arthritis B-lymphocyte dataset. Performance estimation showed the expression profile of B-lymphocytes alone to have no significant discriminatory utility for RA in this setting (average area under [AU] ROC 0.48). Methylation profiles were significantly superior for discriminating RA (average AU ROC 0.71). Interestingly, when a combined trancriptome/methylome model was generated, further enhancement of discriminatory function was observed, although this was modest (AU ROC 0.74). Conclusion: Methylation profiling of B-lymphocytes affords better discrimination for RA than does transcriptional profiling. This may differ in other cell types. Combining data from molecular platforms may be an optimal approach. Functional analyses of the resultant B-lymphocyte signatures may yield insight into molecular pathobiology. Acknowledgement: Versus Arthritis Centres of Excellence for Rheumatoid Arthritis Pathogeneisis and Genetics Disclosure of Interests: Najib Naamane: None declared, Nishanthi Thalayasingam: None declared, Nisha Nair: None declared, Alexander Clar: None declared, Amy Anderson: None declared, Dennis Lendrem: None declared, Louise Reynard: None declared, Stephen Eyre: None declared, Anne Barton: None declared, John Isaacs Grant/research support from: Pfizer, Grant/research support from: Pfizer, Consultant for: Abbvie, Pfizer, Roche, Galvani, Merck, Gilead, Eli Lilly, Amgen, Janssen, Celltrion, NAPP, Consultant for: Abbvie, Pfizer, Roche, Galvani, Merck, Gilead, Eli Lilly, Amgen, Janssen, Celltrion, NAPP, Speakers bureau: Abbvie, Pfizer, Eli Lilly, Speakers bureau: Abbvie, Pfizer, Eli Lilly, Arthur Pratt Grant/research support from: Dr. Pratt is in receipt of an externally peer-reviewed Investigator Initiated Research grant from Pfizer (£66, 000)., Grant/research support from: Dr. Pratt is in receipt of an externally peer-reviewed Investigator Initiated Research grant from Pfizer (£66, 000)., Speakers bureau: Dr Pratt has received honoraria from Eli Lilly and Janssen-Cilag Ltd. for his time in preparing presentations for non-promotional meetings that have been paid directly to Newcastle University., Speakers bureau: Dr Pratt has received honoraria from Eli Lilly and Janssen-Cilag Ltd. for his time in preparing presentations for non-promotional meetings that have been paid directly to Newcastle University.