The diagnostic evaluation of Diamond Blackfan Anaemia (DBA), an inherited bone marrow failure syndrome characterised by erythroid hypoplasia, is challenging because of a broad phenotypic variability and the lack of functional screening tests. In this study, we explored the potential of untargeted metabolomics to diagnose DBA. In dried blood spot samples from 18 DBA patients and 40 healthy controls, a total of 1752 unique metabolite features were identified. This metabolic fingerprint was incorporated into a machine-learning algorithm, and a binary classification model was constructed using a training set. The model showed high performance characteristics (average accuracy 91·9%), and correct prediction of class was observed for all controls (n = 12) and all but one patient (n = 4/5) from the validation or 'test' set (accuracy 94%). Importantly, in patients with congenital dyserythropoietic anaemia (CDA) - an erythroid disorder with overlapping features - we observed a distinct metabolic profile, indicating the disease specificity of the DBA fingerprint and underlining its diagnostic potential. Furthermore, when exploring phenotypic heterogeneity, DBA treatment subgroups yielded discrete differences in metabolic profiles, which could hold future potential in understanding therapy responses. Our data demonstrate that untargeted metabolomics in dried blood spots is a promising new diagnostic tool for DBA.
Keywords: Diamond Blackfan Anaemia; disease fingerprint; dried blood spots; machine-learning algorithm; untargeted metabolomics.
© 2021 The Authors. British Journal of Haematology published by British Society for Haematology and John Wiley & Sons Ltd.