Advanced Flow Cytometry and Machine Learning for Profiling Ocular Leukocytes in Human Inflammatory Eye Disease

Abstract

Purpose : Inflammation contributes to the pathophysiology of eye diseases, including in uveitides, infections, traumas, and nonimmune entities such as in inherited retinal dystrophies. Ocular leukocyte profiling by flow cytometry (FC) of fluid specimens such as tear washes, aqueous humor taps, or vitreous samples (as appropriate) may therefore help better characterize, discover markers, and find therapeutic targets in treating inflammation. Conventional FC, however, is limited to measuring defined bands in wavelengths, whereas spectral FC measures bands across the entire spectrum, allowing us to to increase our parameters by nearly 4-fold. Here, we established a 36-parameter panel to examine eye samples from human ocular immune disease.

Methods : We utilized a 36-parameter panel provided by Cytek Biosciences that was devised with the Cytek panel design tool. Single color, or unstained controls, were carried out with human PBMCs from consented healthy human subjects, or compensation beads. Using SpectroFlow software, we spectrally unmixed the data from each flourochrome. Once completed, we ran our panel on samples from consented healthy subject PBMCs, tear washes from consented ocular surface immune disease patients, and vitreous samples from consented posterior uveitis patients.

Results : Clustering and dimensionality reduction were used for annotating PBMCs based on lineage-defining and supporting markers. Each cluster-defined population was validated by traditional gating schemes for adaptive and innate lymphocytes, plasma cells, granulocytes, and mononuclear phagocyte populations. As we previously showed the utility of FC in characterizing immune cells in ocular surface disease (Reyes NJ. et al., 2018), we tested our panel on tear washes from ocular surface disease patients. We also tested vitreous samples from posterior uveitis patients. We were able to successfully integrate PBMC dimensionality reduction maps (UMAP) with ocular sample data, although not all circulating populations were represented.

Conclusions : We demonstrate the utility of 36 parameter spectral FC. Anchored with PBMC cluster mapping, even low input ocular samples of 100 cells or less could be integrated and reliably annoted for B and T cells, natural killer cells, mononuclear phagocytes, and granulocytes. Hence, this approach may be a powerful tool to help characterize inflammation in various eye diseases.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.