Class III patients have an impaired blood-brain barrier and a polyclonal B cell response
In class III patients, CSF protein and lactate levels were significantly increased compared to class I and II (Fig. 2a). In contrast, CSF leukocytes were not elevated in COVID-19 patients, which signified inflammatory controls. Importantly, CSF glucose was increased even in class I and II patients versus (vs) healthy controls and was a significant discriminator between class III and inflammatory controls (Fig. 2a). Routine CSF parameters are described in Table 1 and Supplementary Table 1. In tendency, the CSF/plasma albumin ratio (Fig. 2a), as well as the total CSF IgG levels (Supplementary Fig. 1b), were higher in class III patients.
In line with recent research 2, SARS-CoV-2 RNA was not detectable in the CSF. However, we were able to detect SARS-CoV-2 Spike (S) protein antibodies in 12 plasma and 3 CSF samples (Fig. 2c), yet the antibody index (AI) pointed to a peripheral synthesis of these intrathecal antibodies (Supplementary Table 1).
We could not detect reactivities against known CNS myelin antigens (Supplementary Fig. 1, CSF data not shown), but found elevated anti-dsDNA-IgG/IgA and anti-gut microbiota IgA responses in the CSF of class III compared to class I patients and also inflammatory controls (Fig. 2c, Supplementary Fig. 1). This was paralleled by an elevated anti-BSA reactivity. The AI pointed towards a peripheral production of these polyclonal antibodies.
Targeted proteomic analysis of CSF and plasma reveals a robust peripheral immune response in Neuro-COVID and a class III-specific signature
We identified predominant plasma secretion of a large number of soluble proteins in Neuro-COVID class III patients compared to controls (Fig. 3a, Supplementary Fig. 2a), suggesting a Neuro-COVID class-dependent plasma signature. Class I and II patients had an increased plasma secretome compared to controls, in line with the previously described peripheral cytokine storm in COVID-19 7.
The CSF soluble protein pattern was different: while class I and II patients had relatively similar profiles as healthy control patients, a Neuro-COVID class III-specific signature with differences to inflammatory controls emerged (Fig. 3a, Supplementary Fig. 2b). Notably, CSF total protein levels progressively increased from class I to III, indicating a correlation between CSF proteomics and neurological symptoms (Fig. 3a).
Principal component analysis (PCA) of soluble proteins distinctly separated controls from COVID-19 patients. Protein patterns largely overlapped between healthy and inflammatory controls, whereas COVID-19 classes clustered apart from the controls, with class III patients displaying the strongest separation (Fig. 3b, Supplementary Table 2).
By investigation of the relative concentration of each molecule between CSF and plasma (log2 normalized CSF/plasma index), we found that most proteins were intrathecally (CSF/plasma ratio > 0) secreted in inflammatory controls, whereas proteins in COVID-19 patients were mainly peripherally synthesized (CSF/plasma ratio < 0). Of note, TRANCE/RANKL was the only intrathecally synthesized protein in class III compared to inflammatory controls (Supplementary Fig. 3, Supplementary Table 3, Supplementary Table 4).
Neuro-COVID class III features are manifestations of microglia regulation, neurodegeneration and blood-brain barrier disruption
As reported previously, plasma IL-6, IL-8, EN-RAGE, HGF, VEGFA, PD-L1 and TNFRSF12A levels were associated with COVID-19 severity 8 and distinct from inflammatory controls (Supplementary Fig. 4a), which lacked peripheral inflammation. Further, plasma TNFRSF11B, EZR and CCL23 were increased in class III vs inflammatory controls (Supplementary Fig. 4a). In contrast, plasma levels of neurotrophic and neuroprotective factors such as BMP-4, CLEC10A, CNTN5, GDF-8, NTRK2, ROBO2 and GDNFRα3 were lower in class III compared to both inflammatory controls and class I patients (Supplementary Fig. 4b) 9–16. Compared to inflammatory controls, class III patients displayed higher plasma 4E-BP1 levels (Supplementary Fig. 4c). HAGH was the only protein displaying higher plasma levels in class I vs class III (Supplementary Fig. 4d).
Several CSF soluble protein levels, particularly some deemed involved in microglia regulation, neurodegeneration and blood-brain barrier (BBB) disruption, including IL-8, MSR1, 4E-BP1, CD200R1, TNFRSF12A and EZR were increased in class III compared to class I (Supplementary Fig. 4e) 9,10. However, only TNFRSF11B levels were both discriminating class III from inflammatory controls and gradually increasing among Neuro-COVID classes (Supplementary Fig. 4f).
CSF-plasma correlations identify a neuronal damage signature in class III, encompassing predictive markers for severe Neuro-COVID
Assuming a cut-off of > 0.45 in the Kendall-Tau correlation matrix, class-specific CSF-plasma correlations were noted and ranked (Fig. 3c, Venn diagram). Inflammatory controls and healthy controls were characterized only by few strong CSF-plasma correlations compared to the Neuro-COVID groups.
We observed a gradual change in correlations from class I to class III. Only a few overlapping soluble proteins with strong correlations were detected, whereas 10–12 individual class-defining proteins were identified (Fig. 3c, Venn diagram and UpSet plot). In class I, the strongest correlations (value > 0.55) were characterized by a myeloid/eosinophil proinflammatory signature exemplified by SIGLEC1, MCP2, IL-8 and CLM1 11–14 (Fig. 3c, heatmap). In class II, a T cell-mediated signature prevailed, defined by CCL25, CD8A, GZMA, TNFRSF9 and IL2-RB, while some myeloid correlations overlapped between class I and II 15–17. In class III, the pattern with the strongest CSF-plasma correlations shifted to a chronic inflammatory and neuronal damage signature encompassing CTSC, KYNU, TNFRSF12A, and CXCL9 9,10,18,19.
To forecast severe Neuro-COVID, we attempted to identify biomarkers displaying strong correlations. Nine analytes (4 CSF and 5 plasma proteins) displayed an AUC-ROC score of > 0.85, suggesting a high predictive power for class III development (Fig. 3d, Supplementary Table 5). Among these, TNFRSF12A had a strong correlation (class III: 0.56; class II: -0.4; class I: 0.2), validating it as a predictive biomarker for severe Neuro-COVID.
Neuro-COVID class III patients feature striking findings on brain imaging while most class I and II patients lack evidence of neuroinflammation
Exemplary brain images of each class are depicted in Fig. 4a-c. Detailed imaging findings are presented in Supplementary Table 6. The most frequent MRI findings were bilateral, multifocal hyperintense signal abnormalities on fluid-attenuated inversion recovery (FLAIR)/T2-weighted (T2w) imaging (n = 18, 56.3%; class I: n = 5, 33.4%; class II: n = 5, 83.4%; class III: n = 11, 72.7%). These signal abnormalities were predominantly located in the periventricular region (13 patients, 40.6%) and the semioval center (16 patients, 50%). Additional (FLAIR)/T2w signal abnormalities were observable in the corpus callosum (9 patients, 28.1%) and in the brain stem (7 patients, 21.9%). In one class III patient, bilateral thalamic signal hyperintensities were present on T2w imaging (Fig. 4c). Further, diffusion-weighted imaging (DWI) changes were present in 4 patients (12.5%): 1 class I/II, and 2 class III patients. Black blood and/or time of flight (TOF) imaging was acquired in 4 patients, in 2 of which (both from class III) focal vessel wall enhancement was visible, indicative of cerebral vasculitis (Fig. 4c). No signal changes were detected in the olfactory bulb. In 3 CT-scanned class III patients, we found 1 infratentorial or supratentorial infarction, 1 thrombosis of the sigmoid sinus with intracerebral hemorrhage and 1 bifrontal subarachnoid hemorrhage (Fig. 4c).
Gray matter volumes in olfactory pathway structures decrease in Neuro-COVID patients and negatively correlate with inflammatory CSF parameters
Twenty brain regions displayed a smaller volume in the Neuro-COVID group compared to age and sex-matched healthy controls (Supplementary Table 7). Thereof, 81% corresponded to the olfactory and gustatory cortex’s telencephalic connections, including the amygdala, entorhinal cortex, basal ganglia, cingulate gyrus and orbitofrontal areas. However, this finding was not significant after FDR correction. Further, we found negative correlations between regional GMVs and the CSF leukocyte count, protein levels and the CSF/plasma albumin ratio (Supplementary Table 8) in the Neuro-COVID group within 16 specific brain regions (Supplementary Fig. 5). Mean CPV to TIV did not differ significantly between COVID-19 patients and controls (7.9 (IQR 2.1) vs. 9.4 (IQR 3.9) x10E-4; p = 0.39, Mann-Whitney-U Test), but we found marginally higher mean and median CPVs relative to TIV in class III compared to class I and II (median choroid plexus volume relative to TIV 9.6 (IQR 4.0) vs. 7.9 (IQR 2.0) x10E-4). However, these differences were not significant (p = 0.53; Mann-Whitney U Test).
High PD-L1 and HGF plasma levels are associated with decreased regional gray matter volumes, while GDF-8 and BMP-4 are neuroprotective in Neuro-COVID patients
Additionally, we investigated the correlation of class-defining soluble CSF and plasma proteins and regional GMVs in regions with significantly lower volumes in COVID-19 patients compared to healthy controls (Supplementary Table 7).
GDF-8 and BMP-4, implicated in neuroprotection and tissue reparatory responses 20,21, were associated with several preserved GMVs in class I (Fig. 4d, Supplementary Fig. 6). Also, GDF-8 and BMP-4 plasma levels were significantly lower in class III and II compared to class I (Fig. 4e). Conversely, PD-L1 and HGF were associated with decreased GMVs in Neuro-COVID patients (Fig. 4f, Supplementary Fig. 6). Accordingly, PD-L1 and HGF plasma levels were higher in class III and II than in class I patients (Fig. 4g, Supplementary Fig. 6). Additional CSF and plasma proteins associated with particularly decreased and preserved GMVs underline the potential importance of COVID-19 dysregulated plasma and CSF proteins on brain structural changes (Supplementary Fig. 7). However, none of the p-values were significant after BH-procedure.