Development and validation of a serum microRNA biomarker panel for bovine Johne's disease

Johne's disease (JD) is a chronic wasting disease of ruminants caused by Mycobacterium avium subspecies paratuberculosis (MAP). For decades, JD management has been hampered by the lack of a sensitive diagnostic test. Here we have developed a machine learning (ML)-driven pipeline that identifies host-encoded microRNA (miRNA) biomarkers and deploys a classification model to provide a diagnostic output. This optimized model is integrated with a molecular assay to form the diagnostic test. Serum miRNA were profiled using RNA-sequencing from MAP-uninfected and MAP-infected cattle across Australia and New Zealand. The MAP-infected cattle were selected based on ante-mortem fecal culture assays or a combination of ELISA (sample-to-positive ratio >200%) and fecal PCR (>10³ genomes/mL). The ML pipeline identified a biomarker consisting of 3 miRNA that categorized MAP infection (receiver operating characteristic, area under the receiver operating curve = 0.97). A JD biomarker quantitative reverse-transcription PCR (RT-qPCR) assay underwent validation in accordance with World Organisation for Animal Health guidelines of the Manual of Diagnostic Tests and Vaccines for Terrestrial Animals. Using Bayesian latent class analysis, the RT-qPCR assay combined with an ML model displayed improved diagnostic sensitivity (median 0.928 vs. 0.755) and comparable diagnostic specificity (median 0.948 vs. 0.960) to existing reference tests (combined performance of fecal culture, ELISA, and fecal PCR). The positive/negative predictive values of this assay were 1.0/47.1 (combined reference tests 1.0/18.8). In summary, we have developed a novel JD detection technology with the potential to improve the detection of MAP infection and support further investigation in longitudinal studies.