MSACL 2019 EU Abstract

Computational Analysis of Mass Spectrometry Data for Standardised Diagnosis of Inborn Disorders of Steroidogenesis

Elizabeth S. Baranowski (1,2), Sreejita Ghosh (3), Cedric HL. Shackleton (1,4), Angela E. Taylor (1,2), Beverly A. Hughes (1), Lorna C. Gilligan (1,2), Agustini Utari (5,6), Sultana MH. Faradz (6), Antonius E. van Herwaarden (7), Hedi L. Claahsen - van der Grinten (8), Michael Biehl (3), Tulay Guran (9), Kerstin Bunte (3), Peter Tino (10), Wiebke Arlt (1,2)
(1)Institute of Metabolism and Systems Research, University of Birmingham, Birmingham, UK (2)Centre for Endocrinology, Diabetes and Metabolism, Birmingham Health Partners, Birmingham, UK (3)Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, Groningen, The Netherlands (4)UCSF Benioff Children's Hospital Oakland, Oakland, California, USA (5)Division of Pediatric Endocrinology, Department of Pediatrics, Faculty of Medicine, Diponegoro University, Semarang, Indonesia (6)Center for Biomedical Research, Faculty of Medicine, Diponegoro University, Semarang, Indonesia (7)Radboud University Medical Center, Department of Laboratory Medicine, Nijmegen, The Netherlands (8)Radboud University Medical Center, Department of Pediatric Endocrinology, Nijmegen, The Netherlands (9)Marmara University, School of Medicine, Department of Paediatric Endocrinology and Diabetes, Istanbul, Turkey (10)School of Computer Science, University of Birmingham, Birmingham, UK

Warning: Undefined variable $headshot in /var/www/html/view_abstract/view_abstract_in_program.php on line 704

Elizabeth Baranowski (Presenter) IMSR, University of Birmingham

Presenter Bio: I completed my medical training at the University of Birmingham and am currently a paediatric trainee in the West Midlands Deanery. My clinical training has now been paused while I complete a clinical research training fellowship, awarded by the Medical Research Council, which will culminate in a PhD. I am completing my fellowship within the Institute of Metabolism and Systems Research at the University of Birmingham. My project combines Endocrinology with statistics, computer science, steroid hormone chemistry and mass spectrometry as we characterise key changes in steroidogenesis over life and in relevant disease states, with the intention of creating improved diagnosis and monitoring tests. I intend to continue academic work following this fellowship, alongside completing subspecialty training in Paediatric Diabetes and Endocrinology, to become a future Clinician Scientist.

Relevant Financial Disclosures (within past 24 months) No relevant financial relationship(s) to disclose. Discussing Products? NO, presenter will NOT present educational content related to products/services of the ineligible company(ies) listed above. Conflict mitigation NOT REQUIRED.

Introduction: The adrenal cortex and gonads produce steroid hormones involved in salt and glucose homeostasis, blood pressure regulation, stress response and sex differentiation. These hormones are produced via a series of enzymatic steps and metabolites of steroids from each step are excreted and measurable in urine. Inborn disorders of steroidogenesis result from genetic mutations in distinct enzymes, causing a block in hormone production and lead to several forms of Congenital Adrenal Hyperplasia (CAH) and differences in sex development (DSD). Each enzyme deficiency is characterised by a distinct pattern of altered excretion of individual steroid metabolites relating to the specific enzymatic block. Ratios of urine steroid metabolites measured by gas chromatography-mass spectrometry (GC-MS) can be employed as surrogates of distinct steroidogenic enzyme activities. Widespread use of GC-MS multi-steroid profiling for rapid diagnosis of these disorders in the acute setting is often hampered by lack of specialist expertise. Here, we developed a novel steroid metabolomics approach for the detection and differentiation of inborn steroidogenic disorders, comparing its performance to that of conventional biochemical analysis by established steroid metabolite ratios.
Methods: We performed multi-steroid profiling by GC-MS in urine samples from 829 healthy controls and 178 patients with inborn steroidogenic disorders. This included the following enzyme deficiencies: CYP21A2(n=26), CYP11B1(n=12), CYP17A1(n=30), POR(n=37), HSD3B2(n=22), SRD5A2(n=51). We assessed the diagnostic performance of conventional biochemical diagnosis based on 14 previously published steroid metabolite ratios, indicative of distinct enzyme reactions. We compared this to the performance of a novel, machine learning-based steroid metabolomics algorithm applied to the GC-MS steroid profiling data, Angle Learning Vector Quantization (ALVQ). ALVQ uses 496 steroid metabolite ratios (all pairwise combinations of 32 measured metabolites) to classify samples by comparing similarity (quantified as cosine of the angle) of their steroid metabolome to learned representative steroid metabolome prototypes for each inborn disorder.
Results: ALVQ showed excellent sensitivity and specificity in the training set (100% and 99.7%, respectively), and good generalisability with sensitivity and specificity on testing random subsets of 89.6% and 99.0%, respectively. In comparison, conventional biochemical diagnosis, using 97.5% age and gender matched normative limits, demonstrated comparable sensitivity of 90.9%, but inferior specificity of 73.9%.
Discussion: We present a steroid metabolomics approach with acceptable performance for non-invasive, rapid and automated differentiation of inborn steroidogenic disorders. This approach is suited to facilitate much more widespread availability of steroid metabolome analysis for routine diagnosis but also to provide novel insights into steroid pathway systems.