Untargeted Metabolomics on Dried Blood Spots for Diagnosing and Investigating Rare Hereditary Anemia
Rare hereditary anemias (RHA) include a large variety of intrinsic defects of the erythrocyte and erythropoiesis. Our knowledge of the pathophysiology of RHAs has improved dramatically recently. However, in a substantial number of patients, the clinical phenotype does not fit classical criteria of disease, response to therapy is unexpectely poor, or a molecular defect cannot be identified. In addition, in patients with well-described genetic defects, there is often no clear genotype-phenotype correlation. We here report on untargeted metabolomics on dried blood spots to identify metabolic disease fingerprints for various RHAs.
Untargeted metabolomics profiling was applied on dried blood spots (DBS) from patients and controls. We used direct infusion high resolution mass spectrometry (DI-HRMS) following previously established protocols. This technique detects thousands of metabolites simultaneously, thereby providing a comprehensive overview of the metabolic status of samples. Statistical analysis was performed in MetaboAnalyst and Graphpad-Prism. Machine learning algorithms and classifications were designed and executed within R-software.
For all investigated anemias (Pyruvate Kinase Defiency (PKD), Hereditary Spherocytosis (HS), and Diamond Blackfan anemia (DBA)) patients and controls yield distinct metabolic profiles. Different approaches, using both supervised and unsupervised multivariate analyses confirmed unique DBS disease fingerprints, with generally evident clustering for controls and more heterogeneity for patients. Discriminative metabolites contributing to this fingerprint were identified, which could serve as potential biomarkers. For PKD, glycolytic intermediates contribute to the fingerprint, as expected, but also numerous acyl-carnitines and polyamines were identified. Interestingly, none of the discriminative parameters correlate with biochemical parameters like reticulocyte count or hemoglobin levels, indicating we are observing disease-specific pathophysiological phenomena instead of only a metabolic signature of anemia.
For both PKD and DBA, where the diagnostic workflow and clinical practice still present challenges, binary classification models were constructed using support vector machine learning. In both disorders, models were validated with new patient samples, and classification proved highly accurate with excellent performance characteristics (Accuracy 94-95%, AUC 0.979-0.990). In addition, we demonstrated that a small subgroup of congenital dyserythropoietic anemia patients, another hypoplastic non-regenerative anemia, is characterized by a completely distinct DBS-profile compared to DBA patients, further underlining the differential diagnostic potential of our approach.
Apart from diagnostic applications, discriminating metabolites in the respective fingerprints can provide novel leads for investigating pathophysiological mechanisms and clinical phenomena, as illustrated for HS, where we observed deregulated polyamine metabolism that significantly correlates with clinical severity and RBC deformability parameters.
In general, the application of untargeted metabolomics in DBS is a novel functional tool that holds promise for diagnostic stratification and studies on disease pathophysiology in RHA. This non-invasive and minimally laborious method provides a comprehensive overview of metabolic disturbances in RHA, that might eventually reveal potential therapeutic targets.