The present study is part of the "All Eyes on PCS" study, which is a prospective, observational single-center study examining the retinal microvasculature of PCS patients and providing an in-depth clinical characterization of the patients. The study protocol was approved by the local ethics committee (Ethics Committee of the Technical University of Munich, School of Medicine, Klinikum rechts der Isar; Approval number: 2022-317-S-SR) and was previously registered (https://clinicaltrials.gov/ct2/show/NCT05635552). All participants in this study provided written informed consent.

We recruited 43 patients from the PCS outpatient department (67.4%; 29/43) and through social media (32.6%, 14/43). Out of those, 41 patients (95.3%) were included in our study. One patient was excluded because they no longer exhibited PCS-typical symptoms during measurement, and one patient was excluded because there was no temporal association between SARS-CoV-2 infection and the onset of PCS symptoms. For patients recruited through social media, an initial survey was sent out with a questionnaire exploring acute SARS-CoV-2 infection and ongoing PCS-typical symptoms. Patients had to provide proof of a positive SARS-CoV-2 reverse transcription-polymerase chain reaction (RT-PCR) or rapid antibody test conducted at least 3 months ago, and they had to have a PCS-typical complaint complex ongoing for at least 2 months. Additionally, the temporal relationship between SARS-CoV-2 infection and the onset of PCS-typical symptoms and alternative diagnoses were reviewed. Exclusion criteria included missing or incomplete consent forms, age under 18 years, pregnancy, malignancy, diseases associated with a significant change in life expectancy, autoimmune diseases of the rheumatological type, cataract, epilepsy, and glaucoma. In cases of uncertainty, eligibility was discussed in a weekly meeting with the Chief Investigator (CI), Principal Investigator (PI), and the study base team, and decisions were made by majority vote. The healthy control (HC) group consisted of 204 participants recruited before the COVID-19 pandemic [31].

RVA was performed using the Dynamic Vessel Analyzer (DVA, commercial product DVAlight; IMEDOS Systems, Jena, Germany), and SVA measurements were performed using the Static Vessel Analyzer (IMEDOS Systems, Jena, Germany, based on TRC-NW8 non-mydriatic retinal camera; Topcon, Tokyo, Japan), as previously published [30]. Before the examination, pupils were dilated using topical tropicamide (0.5% Mydriaticum Stulln; Pharma Stulln, Germany), and patients were seated in a quiet dark room for a ten-minute rest period. Static analysis was performed before DVA.

For DVA, arteriole and venule segments of a length between 0.5 to 1 mm were analysed roughly two-disc diameters away from the optic nerve rim in the upper-temporal or lower temporal direction. Patients were instructed to focus on a needle attached inside the camera, and diameters of one arterial and one venous segment were automatically and continuously recorded for 350 s with the DVAlight device. The baseline recording was 50 s, followed by a flickering phase of 20 s, and then a recovery period of 80 s. Three cycles of these phases were performed in total. To ensure the highest quality standards and exclude substandard data, we compared the quality of the vessel response curves using a cumulative scoring method ranging from 0 to 5, as described earlier [30] Retinal images with a total score of < 2.5 were re-evaluated by a second experienced observer, and potentially excluded after reaching a consensus. In two patients, we were not able to obtain quality DVA data (score value < 2.5) due to a lack of information in the measured region and in one patient, DVA measurement had to be stopped due to excessive fatigue. For each individual patient and participant with quality data, the median arteriolar and venular vessel diameter of the three measurement cycles was calculated and plotted on a diameter-time curve (Graphical Abstract). We calculated the percentage of maximum dilation in relation to the baseline diameter (aFID and vFID), as described before [30].

For SVA analysis, at least three quality images from one eye were taken with a focus on the optic disc in the middle and at an angle of 50°. Two pictures with the highest quality were analysed using Vesselmap 2® (IMEDOS Systems GmbH, Jena, Germany). Segments of retinal arterioles and veins were semi-automatically marked using a mask, within a ring 1 disc diameter away from the optic disc rim, and parameters of SVA were assessed using the Paar-Hubbard formula [32]. Diameters of arteries (CRAE) and veins (CRVE) were averaged with this formula, and the arteriolar-to-venular ratio (AVR) was calculated as the ratio between CRAE and CRVE. One measured unit of the imaging device relates to 1 µm in the model of Gullstrand's normal eye. If two independent analyses were performed by independent examiners, the mean value of CRAE, CRVE, and AVR was calculated. Previous studies have shown high reproducibility of the procedure [33]. Inter- and intra-observer inter-class correlation coefficients for CRAE and CRVE ranged from 0.75 to 0.87 [32, 34]. Thirty retinal images were re-analysed, and the correlation coefficients were 0.98 for CRAE, 0.97 for CRVE, and 0.97 for AVR, indicating a high reproducibility of static retinal vessel parameters [35].

A standardized questionnaire assessed PCS symptoms, essential demographic characteristics, pre-existing illnesses, and medication on the recruitment day. Ongoing post-acute symptoms of the last two weeks were asked, with a focus on 12 symptom complexes, including chemosensory deficits, fatigue, exercise intolerance, joint or muscle pain, ear-nose-throat (ENT) ailments, coughing/wheezing, chest pain, gastrointestinal, neurological, and dermatological ailments, acute infection, and sleep disturbances. Reported symptoms were then encoded as 1 (persistent) and 0 (not constant), and the PCS severity score was calculated [36]. Patient-reported outcome measures (PROMs) focused on the assessment of fatigue (FSS) and depression (PHQ-9) [37, 38]. For patients with myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS), the Canadian Consensus Criteria for ME/CFS was used [39].

Blood sampling was performed as previously described [40]. Standard laboratory measurements were performed in an ISO-certified routine laboratory. IL-6, IL-8, CXCL10, MCP-1 and ICAM-1 in the patient`s serum were quantified using the Cytometric Bead Array Flex system (BD Biosciences, San Diego, US), according to the instructions of the manufacturer. For the determination of von Willebrand Factor (vWF) antigen (Ag) on human samples, the LIAPHEN™ vWF: Ag kit (Hyphen Biomed, Neuville-sur-Oise, France), an immunoturbidimetric assay, was used. The measurements were performed according to the manufacturer's instructions. Briefly, calibrator and controls were reconstituted as indicated in the specific instructions. The calibration concentrations were programmed from 0 to 150% vWF:Ag in Imidazole buffer. The specimens and controls were diluted 4:15 in the same buffer. A calibration curve was established and tested with the quality controls. For high concentrations, between 150 and 1600%, samples were pre-diluted in Imidazole buffer. The tests were performed at 37 °C, and the turbidity was measured at 575 nm.

All statistical analyses were performed using R (Version: 4.2.1). The “All Eyes on PCS” study was designed following the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines [41]. All statistical analysis were performed using R (Version: 4.2.1) and R Studio (Version: RStudio 2023.03.0 + 386). Normally distributed values are shown as mean ± standard deviation (SD), non-normally distributed as median and their inter-quartile-range (IQR) and categorical data are shown as counts and their percentages if not otherwise stated. To analyse distribution, data was visualized as boxplot and histogram and the Shapiro–Wilk test on normality was performed. To compare baseline characteristics Welch two-sample test was used for normally distributed values, Wilcoxon rank sum test for non-normal and χ2 – test for categorical value. We compared means of the DVA parameters aFID and vFID and SVA parameters CRAE, CRVE and AVR between PCS and matched healthy controls. For age- and gender matching we used the Matching package in R with exact matching for gender and a caliper for age. Success of matching was than controlled using the MatchBalance function. We fitted linear multivariate regression models to adjust for potential confounders [42,43,44]. Normality of residuals was assessed using the olsrr package. Histograms of residuals were inspected and the Shapiro–Wilk test on normality was used. All graphs were generated with ggplot2, for interaction blots we used the interaction package and for correlation plots we used the ggpubr package. The ggpubr package was also used to calculate Spearman’s correlation coefficient. To combine the two variables AVR and vFID we fitted a logistic regression model (Table E1: Online Supplement). The binary response variable was cohort dependency (PCS or healthy cohort) and two predictor variables were vFID and AVR. By fitting a regression model, we then estimated the relationship between the predictor variables and the probability of belonging to a particular cohort. The AUC was calculated for the respective regression model and the confidence interval was calculated using DeLong´s method. For assessment of discriminatory ability of biomarkers receiver operating characteristics analysis were performed with the plotROC and pROC package and areas under the curve (AUCs) are presented. To calculate pinteract we fitted a multivariate linear regression model with SVA parameters as the depended variables and the interaction terms PCS severity and level of inflammatory markers in PCS patients as two predictor variables (Table E2: Online Supplement). All values used for data analysis were typed in by two independent researchers in separate sheets and then checked for discrepancies (double-data verification). The graphical abstract was created with BioRender.com.