Ocular Microbiology and Immunology Group
Back to OMIG Main Page

< Previous | 2026 Agenda and Abstracts | Next >

2026 OMIG Abstract

Real-Time Adaptive Long-Read DNA Sequencing for Rapid Diagnosis of Microbial Keratitis

Vridhi Vinaykiya^1,2, Sudeep Mehrotra^1*, Ananya Mukundan¹, Lucia Sobrin¹, and Paulo J. M. Bispo¹

¹Department of Ophthalmology, Massachusetts Eye and Ear, Harvard Medical School, Boston, Massachusetts; and ²The Broad Institute of MIT and Harvard, Cambridge, Massachusetts
[*Currently not at MEE]

Purpose: Unbiased metagenomic next-generation sequencing (mNGS) holds promise for pathogen detection in infectious uveitis, where low microbial biomass and broad pathogen diversity challenge conventional diagnostics. Rigorous validation of the diagnostic approach and bioinformatics workflow is essential before clinical deployment. Here, we report on the design, optimization and analytical validation of an mNGS workflow for pathogen detection from vitreous samples for the diagnosis of infectious uveitis.

Methods: Simulated uveitis vitreous samples were created using a six-point serial dilution series raging (log2) from 5,000 to 156 copies/mL or CFU/mL (n =3 per concentration). Pathogens known to cause intraocular infections were spiked in control uninfected human vitreous including Cutibacterium acnes, Mycobacterium tuberculosis, Herpes Simplex Virus 1 (HSV-1), Herpes Simplex Virus 2 (HSV-2), Varicella Zoster Virus (VZV), Candida albicans, Cryptococcus neoformans, and Toxoplasma gondii. An in-house curated NCBI Reference Sequence Database was used for pathogen identification. Read quality trimming and filtering were performed, followed by host read decontamination using a new dual-stage depletion method. The remaining non-human reads were classified using Kraken2 with default settings. Read abundances were quantified as reads per million (RPM), with background read normalization evaluated using RPM ratio (RPM-r) against blank extraction and uninfected vitreous controls. Organism-specific limits of blank (LoB-Poisson) were derived from four controls using a Poisson model. The limit of detection at 95% detection probability (LoD95) was estimated by probit regression.

Results: Post-filtering, a mean of 124.5 million raw read pairs was reduced to a mean of 4.95 million high-quality read pairs, with host DNA comprising a mean of 98.7%. Kraken2 classified 77% of the remaining microbial reads. Four of the eight spiked organisms (C. albicans, C. neoformans, T. gondii, and HSV-1) achieved 100% detection across all six tested concentrations (3/3 replicates at 156 copies/mL), placing their analytical sensitivity below the lowest tested concentration. The remaining four pathogens exhibited dose-dependent detection variation at lower concentrations. Organism-specific LoB-Poisson thresholds effectively controlled false-positive rates by achieving 100% specificity across all negative control samples without generating false-positive calls above background.

Conclusions: The validated mNGS pipeline demonstrated high analytical sensitivity and specificity for detection of various intraocular pathogens. Organism-specific LoB-Poisson thresholds effectively controlled false-positive rates in this low-biomass context while retaining raw count sensitivity for blank-absent organisms. These results support the analytical readiness of the pipeline for prospective clinical evaluation of mNGS for the diagnosis of infectious uveitis.