Acute rheumatic fever (ARF), a condition which precedes the development of rheumatic heart disease (RHD), causes 275,000 deaths per year worldwide and has a global prevalence of 3.9 million. In Australia, ARF and RHD incidence remains high in Indigenous communities, where Aboriginal people are up to eight times more likely than other Australians to be hospitalised, and nearly 20 times more likely to die from RHD.

ARF is an autoimmune response that follows infection with group A streptococcus (GAS). RHD develops after repeated episodes of GAS infection and causes progressive heart valve damage. Accurate and timely diagnosis of an initial ARF episode and prophylaxis with antibiotics are the only current approaches for preventing RHD. At present, the diagnosis of ARF is made using the Jones criteria, which relies mainly on clinical features and has changed little in over 50 years. Misdiagnosis of ARF is common and a major contributor to the high rates of RHD seen in Aboriginal communities. Despite a clear need, there is no definitive method for diagnosing ARF.

In this study, we developed a novel peptide-based classification strategy using Random-forest based machine-learning algorithms in the R statistical computing environment. Data was generated using a label-free MS-based proteomics approach using Bruker’s new Impact II UHR-QTOF and MaxQuant software. To detect novel diagnostic and prognostic markers of ARF and RHD, we compared affected individuals with healthy individuals and those with alternative diagnoses. A total of 53 plasma samples from patients in the Northern Territory were collected including those with ARF (21), RHD (8), healthy controls (7) and other diagnoses (17). In order to improve the dynamic range of low-abundance proteins, samples were immunodepleted with Agilent’s Multi-Affinity Removal System (MARS-14) LC column. Utilising this approach we have identified statistically-relevant peptide signatures that distinguish patients with ARF from all other patient groups.