Discovery, verification and validation of urinary protein signatures in HIV-associated nephropathy in South Africa

Introduction: Approximately one in five South Africans is HIV positive, accounting for ~10% of the global burden of HIV/AIDS. South Africa administers the largest antiretroviral therapy (ART) programme worldwide. However, 8-10% of first-line ART patients experience HIV-associated nephropathy (HIVAN), possibly due to both ART and virus insult. Current kidney injury tests are unreliable and reflect following significant kidney damage. Therefore, it is necessary to identify sensitive and accurate markers for early detection of kidney injury. This study employed mass spectrometry-based urinary proteomics for HIVAN biomarker discovery in distinct South African cohorts.

Methods: Two cohorts matched in age, race, and gender (case-control) were analysed in the first two phases. A simple sample collection procedure was coupled to a robust, in-house developed, magnetic microparticle-based method for protein capture, clean-up, and digestion.

Proteome profiles were generated using an Evosep One coupled to a timsTOF HT in diaPASEF mode. The raw data were processed using DirectDIA+ in Spectronaut 18. Proteins common to both cohorts were filtered as HIVAN reporters and processed using Perseus (v1.6.12.0) and Cytoscape (v3.8.2). Machine learning (ML) was applied to protein abundance data.

Results: Several common proteins (q-value ≤ 0.01, unique peptides ≥ 2) showed the same directional change in abundance between the two cohorts. These indicated that damaged kidneys may undergo renal hepatization showing increased levels of liver-associated proteins. Immune system, complement cascade and serine protease inhibitors, were the dominant protein families showing increased abundance in HIVAN patients. The XGBoost algorithm showed ≥ 95% sensitivity and specificity in classifying HIVAN and severity.

Conclusions: The workflow applied here showed clinical robustness and can be used as a routine workflow for urinary proteomics studies. ML offers promise for use with complex urinary proteomics datasets for clinically applicable outcomes. Validation of identified protein biosignatures is ongoing in a large cohort (n = 1000) using short gradient DIA-MS, which is the first of its kind study on the African continent.