This prospective, multicenter study included 22 patients with endophthalmitis who presented to Shenzhen Eye Hospital and the Ophthalmology Department of the Affiliated Hospital of Guizhou Medical University in China between January 2023 and March 2024. This study was approved by The Ethics Committee of Shenzhen Eye Hospital (2024KYPJ090) and conducted in adherence to the Declaration of Helsinki. All participants provided informed consent, and all diagnosis and treatment procedures were in accordance with clinical guidelines.

Inclusion criteria: (1) with endophthalmitis or suspected endophthalmitis; (2) accepted PPV due to endophthalmitis; (3) with written informed consent form.

Exclusion criteria: (1) exclusion of endophthalmitis diagnosis; (2) no need for PPV; (3) unstable overall condition; (4) refusal to participate in; (5) deemed unsuitable for the study.

The study collected demographic data, including gender, age, address, etc. Clinical information includes but is not limited to best-corrected visual acuity (BCVA), intraocular pressure (IOP), clinical manifestations, signs, underlying diseases, and predisposing factors (history of trauma, surgery, other systemic infections, etc.). All data were initially organized, processed, and stored by Excel software (Microsoft Corp., Redmond, WA, USA).

Depending on different patient conditions, intraocular fluid specimens were obtained through intraocular fluid biopsy (anterior chamber tap and/or vitreous tap), PPV, or a combination of both methods. For the intraocular fluid biopsy, a 1 ml syringe with a 30G needle was inserted into the anterior chamber or vitreous cavity to extract the specimens. For those who underwent PPV, specimens were obtained directly by vitrectomy probe or from the drain bag. During this process, the use of fluid/air exchange mode ensured the concentration of intraocular pathogens and avoided the impact of antibiotics in the infusion fluid. Specimens were stored at − 80 ℃ until use.

The intraocular fluid specimens were sent to the biological laboratory for routine microbiology culture processing. Firstly, the specimens were inoculated onto Columbia agar (for aerobic bacteria), CDC anaerobic blood agar (for anaerobic bacteria), and Sabouraud dextrose agar (for fungi) (Detgerm Microbiological Science Ltd., Guangzhou, Guangdong, China). Columbia agar and CDC anaerobic blood agar was incubated at 35 to 37 ℃ for 48 h, while Sabouraud dextrose agar was incubated at 28 to 31 ℃ for 7 days. All culture plates should be observed at least once a day. In cases of positive cultures, bacterial and/or fungal isolates were further identified using smear microscopic examination and VITEK-2 Compact system (bioMérieux, Marcy l’Etoile, France). Lactophenol cotton blue was used in the identification of filamentous fungi. According to clinical needs and with informed consent from the patients, intraocular specimens from a subset of patients (n = 7) were sent to an external testing laboratory (Giantmed Medical Diagnostics Lab, Beijing, China) for the mNGS. This sequencing technology utilizes NGS to compare nucleic acid sequences of pathogens with those in the databases, thereby identifying pathogens’ species [39, 40].

The chips have a double-layer structure, with silicon base layers and glass layers, which are shown in Fig. 1 (a). The silicon layers are processed by photolithography and deep silicon etching technologies to develop microgrooves, and the reaction chambers are formed by bonding glass layers on the silicon layers after the surface oxidation process. Special plastic caps are designed for the addition of reagents and a fully enclosed reaction environment, which is shown in Fig. 1 (b) [41, 42]. The self-build microfluidic system is based on an equipment called SWME-03 (Shineway Technology Corp., Shenzhen, Guangdong, China) (Fig. 2) and a self-developed microfluidic chip (Fig. 1), utilizes a silicon-based platinum-patterned heater to achieve the functions of Joule heating and temperature sensing, which is manufactured by the lift-off process on the silicon substrate. A fan is utilized as a rapid cool-down device, and a proportional–integral–derivative controller is utilized to achieve precise temperature control [41, 43]. As shown in Fig. 3, SWME-03 utilizes a motor-driven rotary optical path module to achieve the detection of multicolor fluorescence intensity. A total of five kinds of fluorescence are capable of detecting, including 6-carboxyfluorescein (FAM), hexachlorofluorescein (HEX), carboxy-X-rhodamine (ROX), cyanine 5 (CY5), and cyanine 5.5 (CY5.5). In this research, FAM, HEX, and CY5 are used, which means that three sets of primer-probe are placed in one reaction chamber and a total of 18 targets (6 chambers ×3 fluorescence channels) could be tested in on go. Fluorescent images of the chip will be captured by the camera (IMX392 sensor, Sony Group Corporation, Tokyo, Japan), which will acquire an image at each fluorescent channel during every temperature cycle. The self-developed software will then process these images to extract real-time fluorescence intensity values from the reaction chamber, thereby enabling the determination of the cycle threshold (Ct) value for PCR.

(a) Double-layer structure of the chip. The chip is composed of an upper layer made of glass and a lower layer constructed from silicon, with the reaction chamber and liquid flow channel integrated into the silicon substrate. (b) The photo of the microfluidic chip. The plastic cap consists of a joint and a cover, in the joint there are 6 channels connected to the 6 channels of the chip for liquid sampling and venting, and a cover for sealing the chip after sampling

The overall structure of the microfluidic equipment: Heating module (yellow circle); Multicolor fluorescence detection module (green circle); Control board (red circle)

Structure of the heating module and multicolor fluorescence detection module. The heating module consists of a microheater and a fan, the air channel is used to quickly exhaust hot air during cooling. In the multicolor fluorescence detection module, an optical rotary module is fitted with different colored filters for multi-color fluorescence detection

The intraocular fluid specimen was first added to a nucleic acid extraction kit, and a 50 µL nucleic acid sample was obtained utilizing GenePure Pro fully automatic nucleic acid purification system (BIOER TECHNOLOGY, Hangzhou, Zhejiang, China). The sample is mixed with the reaction solution, enzyme mixture, template, and double-distilled water and then added into the chips. The reaction programs are then set up after the chip placed into the SWME-03, which included: (1) 1 cycle at 95 ℃ for 2–3 min; (2) 40 cycles at 95 ℃ for 5–10 s and 60 ℃ for 15–30 s, fluorescent intensity collection. We use 12.5 µL reagent with sample for one channel. The whole detection of 17 pathogens will use 6 channels (2 chips), and the total volume is 75 µL with 15 µL sample in it. There are two different reaction systems in the program, involving different reagent ratios and reaction times, which are presented together with the template and primer-probe sequences in Supplementary Table S1 to S3.

Based on the patient’s medical history and clinical presentations, treatment plans were preliminarily selected, such as broad-spectrum antibiotic therapy. Treatment plans were then adjusted based on the results of microbiology culture and mNGS. The follow-up visits after treatment were at 1 week, 1 month, 3 months, and 6 months. BCVA, IOP, essential ophthalmic examination, and general condition were recorded at follow-up visits.

IBM SPSS Statistics 25.0 software (IBM Corp., Armonk, NY, USA), STATA 15.0 software (Stata Corp., College Station, Texas, USA), Origin 2024 software (Origin Lab Corp., Northampton, USA) and GraphPad Prism 10 software (GraphPad Corp., San Diego, USA) were used for statistical analysis and graphical representation. Detailed descriptive statistics were performed. For the data of microbiology culture, and the microfluidic RT-PCR system, the McNemar-Bowker test was utilized to test the statistical difference, and Cohen’s kappa was used to test the agreement. For the data with mNGS results, Cochran’s Q test was utilized to test the statistical difference, and Fleiss’ Kappa to test the agreement. A correlation plot was made based on Kendall’s tau correlation analysis. P < 0.05 was considered statistically significant.