Background: AHT including shaken baby syndrome (SBS), is the most common form of traumatic brain injury (TBI) in infants^1,2. Nearly 40 infants per 100,000 live births are affected each year, and one third of cases are misdiagnosed at the first medical visit^1,2. In fact, in cases of mild to moderate AHT (Glasgow Coma Scale (GCS): 9-15), which we have termed silent AHT, clinical symptoms may overlap non-specifically with those of infants with acute malaise (such as non-febrile convulsions or Brief Resolved Unexplained Event (BRUE)). Currently, there is no screening test to support standard clinical practice in identifying children with brain injury who are at risk of silent AHT. Our study in the infants with severe AHT (GCS: 3-4) compared to atraumatic controls highlighted the serum increase of proteins physiologically enriched in the brainstem, eye and spinal cord². Based on our most recent analysis of the large pediatric AHT blood cohort (UPMC Children's Hospital of Pittsburgh, USA)¹, we sought to test the hypothesis that serum protein expression profiles provide the basic knowledge for computational machine learning to obtain signatures that discriminate silent AHT. Methods: We used data-independent acquisition mass spectrometry (DIA-MS) proteomics and analysed the cohort of 246 pediatric sera, including the cases of silent and severe AHT, as well as atraumatic control and accidental pediatric TBI cases, approved by the University of Pittsburgh IRB. We analysed 50 pediatric serum samples (i.e. mainly AHT cases) and two samples of adult cerebrospinal fluid by data-dependent acquisition (DDA-MS), generating a high-quality project-specific spectral library. The library contains information for 1725 Protein Groups (Qvalue ≤ 0.01) of which 232 brain-associated proteins according to the Human Protein Atlas. We performed targeted data analysis of raw DIA-MS files using commercial software Spectronaut (Biognosys) Results: We quantify 1539 Protein Groups in the unique pediatric AHT dataset, comprising nearly 300 MS injections including technical and biological replicates, and observe excellent technical reproducibility (replicates Spearman's rho>0.97; p < 2.2e-16). To investigate the biological data quality, we divided the samples into i) low GCS cases (3-8), moderate GCS cases (9-15) and other cases (non-communicating GCS or atraumatic). We performed a supervised PLS-DA analysis and the first two variants distinguish atraumatic samples (Fig.1, red circles) from low and moderate GCS cases (green/black circles). Among the top selected markers contributing to the serum variability based on the GCS, we detected proteins elevated in the anatomical regions of the brain(e.g. Alpha-synuclein, Fig1.B). Conclusions: This project aims to improve the diagnosis of silent AHT in infants through the proteomic identification of a panel of less-invasive serum markers.

¹Berger, R. P. et al. JAMA Pediatr 171, e170429 (2017).

²Wiskott, K. et al. Proteomics 23, e2200078 (2023).