Integration of proteomics and artificial intelligence-driven OCT biomarker analysis in central retinal vein occlusion

Abstract

Retinal OCT biomarker analysis by artificial intelligence (AI) has not previously been integrated with proteomics. Here, we combined the two techniques to elucidate novel molecular mechanisms in central retinal vein occlusion (CRVO). Proteomic data on aqueous humor samples from patients with treatment naïve CRVO complicated by macular edema (n = 21) and an age-matched, healthy control group (n = 20) was obtained from an existing cohort. Swept-source OCT macular scans (Topcon) from CRVO patients were analysed by AI-driven OCT software (Discovery platform, RetinAI AG, Bern Switzerland) to quantify retinal thicknesses and OCT biomarkers, including intraretinal fluid (IRF), subretinal fluid (SRF), and the outer nuclear layer thickness (ONL). Correlation analysis between protein expression and OCT biomarkers was performed. Proteins involved in complement activation and immune responses exhibited positive correlations with IRF, notably complement component C8 beta chain (r = 0.63, p = 0.0020) and Ig lambda-6 chain C region (r = 0.59, p = 0.0050). Negative correlations with IRF were identified for phosphatidylethanolamine-binding protein 1 (r = −0.66, p = 0.0010) and chordin-like protein 1 (r = −0.61, p = 0.0030). SRF was linked with insulin-like growth factor-binding protein complex acid labile subunit (r = 0.55, p = 0.010) and with extracellular superoxide dismutase (r = −0.63, p = 0.0022). Retinal layer thickness, particularly ONL and total retinal thickness, was positively associated with fibrinogen beta chain (r = 0.58, p = 0.0060) and fibronectin (r = 0.51, p = 0.019). The integration of AI and proteomics allowed for linking biological processes with specific OCT biomarkers. The combined analysis identified complement activation, immune response, oxidative stress, and extracellular matrix remodelling as likely driving forces of macular edema secondary to CRVO.