Conjunctivitis and uveitis are well-known inflammatory ocular diseases. Recently, it has been speculated that inflammation may be involved in multiple other ocular diseases such as dry eye, glaucoma, age-related macular degeneration, diabetic retinopathy, retinitis pigmentosa, retinopathy of prematurity, Grave's ophthalmopathy, and orbital tumors. It is not uncommon to face challenges when diagnosing inflammatory ocular diseases such as uveitis and IgG4-related ophthalmic disease and related diseases. Thus, going forward, it is anticipated that diagnostic biomarkers will be established. Multi-omics analysis comprehensively analyzes information of diverse biomolecules such as proteins (proteome), nucleic acids (genome, transcriptome and microRNAs) and metabolites (metabolome). Advances in research equipment and omics analysis enabled researchers to attempt to create biomarkers free of researcher bias using reduced amounts of clinical specimens. This review reports our attempts to create biomarkers for inflammatory ocular diseases using bioinformatics and artificial intelligence to comprehensively analyze a wide range of information ranging from the results of clinical tests that have been conducted under medical insurance coverage to massive data obtained from recent multi-omics analysis.
I. Differentiation of inflammatory ocular diseases based on immune mediators in aqueous humor using machine learning
When narrowly defined, uveitis is considered to be a common inflammatory ocular disease. In recent years, however, the involvement of inflammation has been speculated in ocular diseases with invisible immune mediators (such as cytokines, chemokines, and growth factors) and inflammatory cell infiltrating such lesions, even in the absence of ophthalmoscopic"inflammation"or"inflammatory cells"detected by slit lamp biomicroscopy and fundoscopy. Therefore, ocular diseases such as glaucoma, age-related macular degeneration, and diabetic retinopathy, which were previously considered to be unrelated to inflammation, are now coming to be classified as inflammatory ocular diseases. Using the cytometric beads array flex set, we comprehensively measured 28 types of immune mediators such as cytokines, chemokines, and growth factors using small volumes of aqueous humor samples from eyes with 17 types of ocular diseases, as well as attempted to create models for predicting the diagnoses of 17 diseases based on the 28 immune mediator data using various types of machine learning (ML). First, we applied a training dataset to a ML algorithm to construct a disease classification model, and then randomly applied a test dataset to the model to predict the diagnoses of intraocular diseases. Among ML algorithms, random forest showed the best performance for prediction. Among immune mediators, interleukin (IL) -10 was useful to predict vitreoretinal lymphoma, monokine induced by gamma interferon (Mig) for acute retinal necrosis, IL-6 for endophthalmitis, and monocyte chemoattractant protein 1 (MCP-1) for rhegmatogenous retinal detachment and primary open-angle glaucoma. As per the cumulative correct prediction rates for the 17 types of inflammatory ocular diseases, the prediction models were able to discriminate vitreoretinal lymphoma, acute retinal necrosis, and endophthalmitis with high accuracy. Thus, we were able to identify certain inflammatory ocular diseases (vitreoretinal lymphoma, acute retinal necrosis, and endophthalmitis) by analyzing 28 types of immune mediators in aqueous humor samples.
II. Creation of biomarkers for uveitis and vitreoretinal lymphoma
There are currently no disease-specific biomarkers for uveitis, including Behcet's disease and Vogt-Koyanagi-Harada disease. Therefore, the identification of novel biomarkers will both contribute to early diagnosis and provide important markers for evaluating the therapeutic effect and recurrence. For vitreoretinal lymphoma that is often difficult to differentiate from uveitis, accurate diagnosis of the disease is causally linked to the prognosis of survival; hence, the availability of a universal biomarker will be highly desirable. We therefore performed comprehensive microRNA analysis and metabolome analysis using peripheral blood and intraocular fluid samples and separately created biomarkers for identifying uveitis and vitreoretinal lymphoma. Furthermore, unsupervised cluster analysis using numerical values of general blood test data and concentrations of humoral factors in intraocular fluid revealed the possibility of predicting the clinical course of vitreoretinal lymphoma using serum IgA level, which is a standard clinical test item, but which had not been reported.
III. Creation of biomarkers for orbital lymphoproliferative disorders
Lymphoproliferative disorders, which are the most frequent orbital tumors encountered clinically, can be broadly classified into IgG4-related ophthalmic disease and malignant lymphomas. Conventionally, this disease is diagnosed based on histopathological examination; however, the diagnosis is often difficult because the pathological features of these two diseases are highly similar and there are no established biomarkers. We performed omics analysis on these diseases and revealed the presence of microRNAs and metabolites that characteristically changed in IgG4-related ophthalmic disease and orbital MALT lymphoma. Unsupervised cluster analysis of routine blood test data suggested the possibility of predicting the presence or absence of lesions in extraocular organs as well as the visual function outcomes in IgG4-related ophthalmic disease using serum IgG4 and serum IgE levels.
Nippon Ganka Gakkai Zasshi (J Jpn Ophthalmol Soc) 125: 230-265, 2021.