Revolutionary approach links protein analysis and single-cell study to uncover disease mechanisms

By Tarun Sai LomteReviewed by Lily Ramsey, LLMOct 24 2023

In a recent study published in Cell, researchers integrated liquid-biopsy proteomics with single-cell transcriptomics from ocular cell types to trace the cellular origin of nearly 6,000 proteins in the aqueous humor (AH).

Study: Liquid-biopsy proteomics combined with AI identifies cellular drivers of eye aging and disease in vivo. Image Credit: K illustrator Photo/Shutterstock.com

Background

Cell-level analyses are severely limited in non-regenerative tissues and organs, such as the retina and brain, as direct biopsies would cause irreversible damage. Liquid biopsies of locally enriched fluid could be the only way to obtain proteomics.

Nevertheless, proteins cannot be resolved at cellular resolution. Although single-cell ribonucleic acid (RNA) sequencing (scRNA-seq) and transcriptomics provide cell-level gene expression, they require tissue biopsy or are limited to autopsy tissues.

In addition, it cannot be inferred whether a given messenger RNA (mRNA) is translated; thus, studying proteome-wide relevance to human disease is challenging.

Fifty-seven types of cells have been identified in the human eye, which secrete proteins into the two fluid-filled chambers containing AH or vitreous fluid. Moreover, the eye fluid is separated from plasma by the blood-ocular barrier, which results in a distinct protein composition within the eye.

The study and findings

In the present study, researchers applied high-resolution proteomics to trace the cellular origins of nearly 6,000 proteins in the vitreous or AH. They termed this approach “Tracing Expression of Multiple Protein Origins” (TEMPO).

They obtained 120 liquid biopsies from the vitreous or AH of patients undergoing surgery. scRNA-seq data of over 82,000 single cells from 57 ocular cell types and 15 extra-ocular cell types were integrated with data from liquid biopsy proteomics.

The team confirmed gene expression of 5,923 (99.5%) proteins within RNA-seq data. There were 5,887 locally expressed genes in the eye. Unsupervised cluster analysis of 5,923 genes revealed cell-type-dependent clustering.

Besides, 1,920 cell-type protein markers were identified, which showed highly specific expression within their respective cell type. Next, the team established cell-specific markers for individual retinal, immune, and other ocular cell types.

Proteins from eye-specific cell types were highly enriched in AH. Next, the team examined which markers were altered in retinitis pigmentosa (RP), a rare genetic disease. Proteins specific to rod photoreceptors were the most affected.

Moreover, marker proteins from retinal bipolar cells, amacrine cells, cone photoreceptors, vascular endothelial cells, and erythrocytes were decreased in RP. In contrast, other retinal cell types, like ganglion and horizontal cells, were unaffected.

Further, the team also tested uveitis patients and observed a significant increase in markers from several immune cells (mast cells, B cells, T cells, macrophages, neutrophils, and hyalocytes). Besides, they observed molecular damage from uveitis in other retinal cell types, such as amacrine cells, bipolar cells, and cone photoreceptors.

Next, the researchers analyzed liquid biopsies from diabetic retinopathy (DR) patients and found changes in the expression of markers from immune cells, retinal glial cells, pericytes, and vascular endothelial cells. Some liver proteins were also significantly elevated, suggesting increased vascular permeability due to the disruption of the blood-retina barrier in DR.

Besides, 58 angiogenic proteins were differentially enriched in DR; 29 proteins were significantly increased in both non-proliferative (NPDR) and proliferative (PDR) stages of DR. Another 29 proteins were only increased in PDR. Proteins enriched in both stages were primarily derivatives of pericytes and endothelial cells.

However, PDR-specific proteins were derived from a completely distinct class of cells, i.e., immune cells (retinal microglia, neutrophils, and macrophages), suggesting that PDR differs from NPDR at cellular and molecular levels.

Further, the authors analyzed AH from patients with Parkinson’s disease (PD). Among 48 PD-relevant proteins, 18 were present in the AH, including nine significantly increased PD.

There was evidence of molecular dysfunction in several retinal cell types, consistent with reported anatomical changes in the inner retinal layers of PD patients. However, molecular changes in cone and rod photoreceptors that do not exhibit anatomical changes were also noted, suggesting that molecular markers may be more sensitive than conventional structural or imaging biomarkers.

Next, the team developed an artificial intelligence (AI) model for eye age prediction. They trained an XGBoost model to predict the chronological age.

The final model was based on 26 proteins; of these, 20 proteins were known to be associated with aging but were not identified (as biomarkers) in analyzed diseases, except for one protein. The final model demonstrated strong age prediction upon validation.

The team applied this model to patients with eye diseases unrelated to birth age. The model-predicted molecular age of NPDR patients was 12 years older than healthy patients. It increased by 31 years in PDR patients, 16 years in RP patients, and 29 years in uveitis patients.

This suggested the accelerated aging of diseased eyes. Age-associated proteins in normal eyes were primarily derived from pericytes, immune cells, and endothelial cells.

Next, additional AI models were built to predict retinal, immune, and vascular age based on proteins specific to these cell types. The models revealed accelerated aging of cell types specific to disease, viz., retinal cells in RP, immune cells in uveitis, and vascular cells in PDR.

Interestingly, accelerated aging of retinal and immune cells was observed in diabetes mellitus patients who did not have visible retinopathy signs.

Conclusions

Taken together, the team demonstrated the potential of the TEMPO approach to examine aging and disease mechanisms in vivo at the cellular level. Nearly 6,000 proteins from a limited volume (50 µl) of the biological fluid were traced back to their cellular origins.

The AI aging models specific to cell types were robust in assessing molecular age, offering novel insights into aging and disease. Overall, TEMPO will prove vital in identifying cellular mechanisms and determining the interplay between disease and aging.