Clinical data analysis
An independent data board reviewed clinical, laboratory, and imaging data with access to unblinded data. Clinical data were extracted from the patients’ medical records, including medical history and comorbidities inquired by the medical team to patients and relatives at hospital admission anamnesis, clinical characteristics at hospital admission, in-hospital symptoms, complications, and medication used, according to an approved clinical research form (CRF) (International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) and World Health Organization (WHO), 2020a) and clinical characterization protocol (CCP) (International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) and World Health Organization (WHO), 2020b) created by members of the International Severe Acute Respiratory and Emerging Infection Consortium (ISARIC) in collaboration with the World Health Organization. Multimorbidity was defined by the presence of two or more chronic illnesses. Detailed laboratory results were also collected, and qualitative and quantitative estimations of lung disease caused by COVID-19 were determined by computerized tomography (CT) scan examination.
COVID-19 severity classification followed the “Ordinal Scale for Clinical Improvement” proposed by a special World Health Organization (WHO) committee (World Health Organization, 2020b). Mild disease included hospitalized patients who did not receive oxygen therapy or received oxygen by masks or nasal cannula. Severe disease included hospitalized patients who required at least one of the following treatments: oxygen by non-invasive ventilation; high-flow oxygen; intubation; and mechanical ventilation with or without additional organ support.
An experienced neurologist reviewed patients’ clinical data and defined a major neurological complaint. Encephalitis was defined as presenting altered mental status (altered level of consciousness, lethargy, or personality change) for at least 24h and two or more of the following criteria: a) seizures not attributable to a pre-existing condition; b) new-onset focal neurologic finding; c) elevated CSF white blood cell count (above 5 cells/mm3); d) acute neuroimaging finding consistent with encephalitis; e) abnormal electroencephalography consistent with encephalitis, or f) fever (above 38°C) within 3-days of symptom onset (Venkatesan et al., 2013).
COVID-19 hyperinflammatory syndrome score (cHIS)
Blood test data most proximal to CSF sampling were retrieved to compute cHIS as previously described (Webb et al., 2020). The score was calculated using reported fever at admission or hospitalization and blood indicators of cytokinaemia (C-reactive protein (CRP) above 15 mg/dL or triglyceride concentration above 150 mg/dL or blood IL6 concentration above 15 pg/mL); hematological dysfunction (neutrophil to lymphocyte ratio (NLR) above 10, less than 110 billion/L platelets, or hemoglobin bellow 9.2 g/dL); coagulopathy (D-dimer concentration above 1.5 µg/mL); hepatic injury (lactate dehydrogenase concentration above 400 U/L or aspartate aminotransferase above 100 U/L); macrophage activation (ferritin concentration above 700 µg/L).
Blood cytokine evaluation.
We retrieved 16 serum samples from COVID-19 patients collected during hospitalization (median 0 days, interquartile range 0-5.5, maximum 44 days apart CSF collection date), stored in polypropylene tubes, and frozen at -80ºC until molecular analysis. Control serum came from the above specified five healthy donors (negative RT-qPCR for SARS-CoV-2). Before assays, samples were thawed and kept on ice. Cytokines were measured using the Human Cytokine/Chemokine Magnetic Bead Panel kit (Millipore, #HCYTOMAG-60K) following manufacturers’ instructions. Results were read at a MagPix Luminex xMAP instrument and analyzed with XPonent software. Data is represented as a fold change computed by subtracting the control mean value (X) from COVID-19 individual values (Y) and dividing by the control mean, [(Y-X)/X]. Raw data (pg/mL) was used for statistical analysis. Multiple comparisons are corrected by FDR, Q = 5%. IL3 was only detected in one COVID-19 sample and was further excluded from the analysis.
CSF collection and analysis.
Irrespective of patients' grouping, after lumbar puncture, CSF was immediately processed for routine laboratory analysis consisting of cell counts, total protein, glucose, lactate, microbiological analysis, and the opening pressure estimation. For the COVID-19 group, 14 CSF samples were also investigated for the presence of SARS-CoV-2 RNA and other neuropathogens using the Biomanguinhos (E + P1) RT-qPCR kit (FIOCRUZ, Brazil), XGEN Master COVID-19 (Mobius Brazil), XGEN Viral Meningitis Panel (Mobius, Brazil) or FilmArray Meningitis/Encephalitis Panel (bioMérieux, Brazil). Oligoclonal bands IgG investigation results were available for four COVID-19 patients (HYDRASYS FOCUSING, Sebia, France). For all patients from all groups, the remaining cell-free CSF supernatants were stored in polypropylene tubes and immediately frozen at -80ºC until used for molecular analysis described next.
NPH pre-pandemic uninfected controls. CSF samples from ten donors diagnosed with NPH and subject to lumbar puncture to drain fluid excess in 2019 were used as uninfected control.
CSF IL6 and TNFα quantification.
Before assays, CSF samples were thawed and kept on ice. TNFα concentration was measured using Cayman TNFα (human) ELISA kit (#589201). Calibrators were diluted in a blocking solution (5% bovine serum albumin, 0.05% Tween-20 in phosphate-buffered saline). Samples were tested undiluted. IL6 levels were measured using a Human IL6 Quantikine ELISA kit (R&D Systems, #D6050). Before analysis, samples were diluted (1:2) in a solution provided by the kit (RD6F). Samples, quality controls, and calibrators were run in duplicates for all targets, following manufacturers’ indications. Standard curves were calculated using a 4-parameter logistic regression model. Low undetermined levels were expressed as 0 pg/mL.
LC-MS/MS Shotgun Proteomics Analysis.
COVID-19 and control CSF samples first received a protease inhibitor cocktail (Halt Protease Inhibitor Cocktail, Thermo Scientific, #78430). The CSF samples’ protein amount was quantified using the BCA and diluted 1:1 v/v in buffer (Tris-HCL 100mM and 2M Urea). Aiming to obtain a higher quality of buffer exchange and protein digestion, we performed the FASP protocol for tryptic digestion (Distler et al., 2016) in 20 µg of protein, briefly described in washing steps to buffer exchange in a microcolumn tip (10kDa MW cut off), and tryptic digestion performed in the column. Samples were reduced, alkylated, and later digested using trypsin. Digested peptides from each sample were resuspended in 0.1% FA. The separation of tryptic peptides was performed on an ACQUITY MClass System (Waters Corporation). 1 µg of each digested sample was loaded onto a Symmetry C18 5 µm, 180 µm × 20 mm precolumn (Waters Corp.) used as a trapping column and subsequently separated by a 120-min reversed-phase gradient at 300 nL/min (linear-gradient, 3–55% ACN over 90 min) using an HSS T3 C18 1.8 µm, 75 µm × 150 mm nanoscale and LC column (Waters Corp.) maintained at 30°C. The gradient elution Water-Formic Acid (99.9/0.1, v/v) was used as eluent A and Acetonitrile Formic Acid (99.9/0.1, v/v) as B. The Separated peptides were analyzed by a High Definition Synapt G2-Si Mass spectrometer directly coupled to the chromatographic system. Differential protein expression was evaluated with a data-independent acquisition (DIA) of shotgun proteomics analysis by Expression configuration mode (Mse). The mass spectrometer operated in “Expression Mode”, switching between low (4 eV) and high (25–60 eV) collision energies on the gas cell, using a 1.0s scan time per function over 50–2000 m/z. All spectra have been acquired in Ion Mobility Mode by applying a 1.000m/s wave velocity for the ion separation and a 175m/s transfer wave velocity. The processing of low and elevated energy added to the data of the reference lock mass ([Glu1]-Fibrinopeptide B Standard, Waters Corp.) provides a time-aligned inventory of accurate mass retention time components for both the low and elevated-energy (EMRT, exact mass retention time). Each sample was run in three technical replicates.
Continuum LC-MS data from three replicate experiments for each sample was processed for qualitative and quantitative analysis using the software Progenesis (Waters Corp.). The qualitative identification of proteins was obtained by searching the Homo sapiens database (UniProt KB/Swiss-Prot Protein reviewed). Search parameters were set as automatic tolerance for precursor ions and product ions, a minimum of one fragment ions matched per peptide, a minimum of three fragment ions matched per protein, a minimum of one unique peptide matched per protein, 2 missed cleavage, carbamidomethylation of cysteines as fixed modification and oxidation of methionines as variable modifications, FDR of the identification algorithm < 1%.
Label-free quantitative analysis was obtained using the relative abundance intensity integrated with Progenesis software, using all peptides identified for normalization. The expression analysis was performed considering technical replicates available for each experimental condition, following the hypothesis that each group is an independent variable. It was considered only proteins present in two out of three technical replicates, and a statistical cut-off of ANOVA > 0.05 was adopted. The dataset containing statistically significant deregulated proteins was used for Protein Interaction and pathway-enrichment analysis performed in Cytoscape, and in silico analyses were performed in an R environment. For Heatmap, volcano plot and PCA analysis, it was used a homemade Python language open source software OmicScope (Reis-de-Oliveira et al., 2023). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD033979 and 10.6019/PXD033979.
Multiple Reaction Monitoring (MRM) for SARS-CoV-2’s spike protein.
Approximately 500 fmols of digested peptides from a recombinant spike protein produced by Cell Culture Engineering Lab (COPPER/UFRJ)(Alvim et al., 2020) were spiked in tryptic peptides from an E. Coli total protein extract and further loaded onto a Symmetry C18 5 µm, 180 µm × 20 mm precolumn (Waters Corp.) used as trapping column and subsequently separated by a 90 min reversed-phase gradient at 300 nL/min (linear-gradient, 3–55% ACN over 90 min) using an HSS T3 C18 1.8 µm, 75 µm × 150 mm nanoscale and LC column (Waters Corp.) at 40°C. For the gradient elution, Water-Formic Acid (99.9/0.1, v/v) was used as eluent A and Acetonitrile Formic Acid (99.9/0.1, v/v) as B. The Separated peptides were analyzed by a High Definition Synapt G2-Si Mass spectrometer directly coupled to the chromatographic system.
The generated raw file was imported into the Skyline software. The Spike protein was chosen as a FASTA file as a reference for the theoretical tryptic peptides, isolating the mass of the peptides of interest (parent ions) used for the MS/MS selection (fragment ions) and their retention time to create the MRM method. Four peptides’ transitions (543.2774++; 570.3035++; 609.7987++, and 679.8386++) were chosen, and their respective fragments of the recombinant spike protein considering only peptides with at least three fragment ions and 8–25 amino acids. The Skyline software created the collision energy to be applied for each peptide. The MRM method for searching for the Spike protein was applied to all CSF samples.
SARS-CoV-2’s genome sequencing, assembly, and phylogeny.
Five SARS-CoV-2 nasopharyngeal swabs positive samples (collected during hospitalization) with viral genes N1 or N2 amplification Ct < 30 were sequenced. Sequencing libraries were prepared using the QIAseq FX DNA Library Prep kit (QIAGEN, Germany), and reactions were sequenced on the Illumina MiSeq platform (Illumina, USA) with a v3 (600 cycles) cartridge. A custom pipeline for data quality control and consensus genome assembly was used (Moreira et al., 2021). The viral genomes were classified into Pango lineages using the Pangolin tool v2.4.2. The assembly and classification are available in Suppl. Table 6. A dataset (n = 58) containing only lineages identified in Rio de Janeiro (Brazil) between April and September 2020 was created using genomes publicly available on the GISAID EpiCoV database to corroborate the classification (See Suppl. Table 6). The dataset was aligned with minimap18, and a maximum likelihood phylogeny was inferred with IQ-tree v2.0.319 under the GTR + F + I + G4 model (Li, 2018).
Statistical Analysis
Statistical analyses were performed using Prism 9 Software, v 9.0.2 or Matlab R2019b (Mathworks, USA), or IBM SPSS Statistics version 29.0.2.0 (20). Statistical analysis for proteomic and MRM analyses are described in the above sessions. Categorical variables are expressed as frequency and proportion (No, %) and analyzed using Fisher’s exact or Chi-squared tests, followed by post-hoc Chi-Squared pairwise comparison. Continuous variables were checked for normal distribution using the D’Agostino & Pearson and Shapiro-Wilk normality tests. As indicated, normally distributed data were expressed as the mean and standard error of the mean (SEM) or standard deviation (SD). Variances were compared using the F test. Groups were compared using a t-test, Welch’s t-test (correction for different variances). To identify possible confounders in analyses, we considered previously reported associations between sex, aging, multimorbidity, cardiovascular, diabetes, neurological, and neuropsychiatric comorbidities with the risk for severe COVID-19 and chronic systemic and CNS inflammation (De Felice and Ferreira, 2014; Friedman et al., 2019; Grande et al., 2020; Guan et al., 2020; Williamson et al., 2020; Zeng et al., 2020; Evans et al., 2021; Jun et al., 2021; Lau et al., 2021; Tahira et al., 2021; Taquet et al., 2021c; ten-Caten et al., 2021; De Felice et al., 2022). Thus, group comparisons for cHIS and CSF neuroinflammatory biomarkers (IL6 and TNFα) utilized ANCOVA using sex, age, diabetes, and the numbers of cardiovascular, neurological/psychiatric, and all other comorbidities as covariates. Heteroskedasticity was checked using the White and F tests. The equality of variances was checked using Levene’s test. Covariates correlation was checked using Pearson’s correlation tests and no high correlation (r > 0.8) was found. Residuals’ normal distribution was checked using the Kolmogorov-Smirnov and Shapiro-Wilk tests. Because residuals for ANCOVA analyses using CSF IL6 levels as a dependent variable did not follow a normal distribution, statistical analyses were performed after log transformation (LogIL6). Due to the small sample size, CSF IL6 and TNFα comparisons between neurological diagnostic groups was performed using Kruskal-Wallis followed by Dunn’s test. Non-parametric data were expressed as the median and interquartile range (IQR) and analyzed using the Mann-Whitney test. Multiple comparisons were corrected by the FDR method when indicated. The level of statistical significance was set at 5%. Missing or unavailable data were not included in the statistical analysis.