Severe Neuro-COVID is associated with peripheral immune signatures, autoimmunity and neurodegeneration: a prospective cross-sectional study

Abstract

Growing evidence links COVID-19 with acute and long-term neurological dysfunction. However, the pathophysiological mechanisms resulting in central nervous system involvement remain unclear, posing both diagnostic and therapeutic challenges. Here we show outcomes of a cross-sectional clinical study (NCT04472013) including clinical and imaging data and corresponding multidimensional characterization of immune mediators in the cerebrospinal fluid (CSF) and plasma of patients belonging to different Neuro-COVID severity classes. The most prominent signs of severe Neuro-COVID are blood-brain barrier (BBB) impairment, elevated microglia activation markers and a polyclonal B cell response targeting self-antigens and non-self-antigens. COVID-19 patients show decreased regional brain volumes associating with specific CSF parameters, however, COVID-19 patients characterized by plasma cytokine storm are presenting with a non-inflammatory CSF profile. Post-acute COVID-19 syndrome strongly associates with a distinctive set of CSF and plasma mediators. Collectively, we identify several potentially actionable targets to prevent or intervene with the neurological consequences of SARS-CoV-2 infection.

Conflict of interest statement

I.J. has received speaker honoraria or unrestricted grants from Biogen Idec and Novartis and has served as an advisor for Alexion, Biogen, Bristol Myers Squibb, Celgene, Janssen-Cilag, Neuway, Merck, Novartis, Roche and Sanofi Genzyme; none of these are related to this study. G.H. has equity in and is a co-founder of Incephalo Inc. A-K.P. has received speaker honoraria or research/travel support from Roche and Biogen all used for research. The remaining authors declare no competing interests.

© 2022. The Author(s).

Figures

Fig. 1. CONSORT diagram and schemes illustrating the project design.

a Consort flow diagram. Patients who tested positive for SARS-CoV-2 were assessed for eligibility (n = 310), of which 269 declined to participate and 1 failed to meet inclusion criteria. Enrolled patients (n = 40) were allocated to different severity classes of Neuro-COVID according to Fotuhi et. al. with 18 in class I, 7 in class II and 15 in class III. Schemes illustrating the study design:b Paired cerebrospinal fluid (CSF) and plasma samples were collected from 40 COVID-19 patients. Paired CSF and plasma samples from healthy (n = 25) and non-MS inflammatory neurological disease controls (n = 25) were retrospectively obtained. c In 37 of the COVID-19 patients, a contrast-enhanced MRI or CT scan was conducted and evaluated by a board-certified neuroradiologist. d Brain volumetric analysis was performed in 35 COVID-19 patients. This cohort included 22 patients of the main study cohort from whom Magnetization prepared—rapid gradient echo (MPRAGE) pulse sequences and paired CSF and plasma samples were obtained (light blue) and an additional 13 patients who underwent brain MRI during COVID-19 infection (dark blue). A cohort of 36 healthy age and sex-matched individuals served as the control group. b–d Created with Biorender.com.

Fig. 2. Routine inflammatory CSF parameters and B-cell response in Neuro-COVID patients.

a Box plot representation of routine cerebrospinal fluid (CSF) parameters including glucose (mmol/L), glucose CSF/plasma ratio (log2 scale), lactate (mmol/L), albumin CSF/plasma ratio (log2 scale), total protein (mg/L, log10 scale) and leukocytes (cell count × 106/L − log10 scale) (center line at the median, upper bound at 75th percentile, lower bound at 25th percentile) with whiskers at minimum and maximum values. Patients that were excluded from the analysis are indicated in violet. Statistics: The data for each parameter, except the leukocyte count, was marginalized on sex and age. Statistics: The data for each parameter was marginalized on sex and age. Two-sided Mann–Whitney-U test was applied, p value correction was performed with Benjamin-Hochberg (BH)-procedure (adjusted p: *<0.05; **<0.01; ***<0.001; if not otherwise indicated: not significant). b Non-metric multidimensional scaling (NMDS) plots of merged anti-BSA-, anti-dsDNA- and anti-gut bacteria antibodies in the CSF (left plot) and plasma (right plot) of each class. Data points are colored by category. Each point represents one patient. Patients with a more similar antibody composition are closer together and, conversely, those that were more dissimilar depict a greater distance. The ellipses represent the 95% confidence interval within subgroups. c Box plot representation of CSF levels (OD450; optical density at 450 nm) of anti-BSA, anti-dsDNA-and anti-gut bacteria (RePOOPulate)-IgG/IgA per patient and control group. Patients with anti-SARS-CoV-2 Spike protein antibodies in the CSF indicated in red, those with intrathecal IgG or IgA production in orange, respectively. Statistics: The data for each parameter except the IgG data was marginalized on sex and age. Two-sided Mann–Whitney-U test was applied, p value correction was performed with Benjamin-Hochberg (BH)-procedure (adjusted p: *<0.05; **<0.01; ***<0.001; if not otherwise indicated: not significant). Source data of (a, c) are provided as a Source Data file.

Fig. 3. B-cell receptor sequencing reveals a more specific antibody response and a higher B-cell clone number in class I patients, whereas class III patients depicted a more diverse response due to polyclonality.

a Box plot representation of B-cell clone number in class I (n = 4) and class III patients (n = 4) respectively. B-cell clone number was significantly higher in the blood of class I patients compared to class III (Mann–Whitney-U test: p: =0.02857, spearman’s rank correlation test: R = 0.738, p = 0.04583). Class I patients (n = 4) had higher plasma IgG levels which correlated with the number of B-cell clones in the blood. b Box plot representation of Shannon Diversity and Evenness in class I (n = 4) and class III (n = 4) patients. Diversity analysis showed that there was no significant difference between class I and II. c Representation of all B-cell clones from each class clustered together and ranked according to their total frequency (relative abundance) in the immune repertoire of a patient. Then, top 10 highest frequency clones from each class selected, grouped together and ranked again according to the total frequency. As indicated in the dot plot, the clones making up more than 5% of BCR immune repertoire belonged to class I patients (n = 4). Only two B-cell clones of class III patients were in the top 10. d B-cell immune repertoire of both classes (class I and III patients: n = 8) showed a similar Gaussian distribution in CDR3 nucleotide length and same median CDR3 nucleotide length. Each bar represents the number of B cells having a specific CDR3 nucleotide length. Boxplots indicate median and lower (25th)–upper (75th) quartile and whiskers show the minimum-maximum values. Each dot represents individual samples. Statistics: Two-sided Wilcoxon rank-sum test and spearman’s rank correlation were applied (adjusted p: *<0.05; if not otherwise indicated: not significant, R Spearman correlation coefficient). BCR B-cell receptor.

Fig. 4. Neuro-COVID patients display a vigorous peripheral immune response and specific CSF alterations including analytes with high predictive value for class III development and strong CSF-plasma correlation.

a Rose plots representing Z-scores of marginalized normalized protein expression (NPX) of 192 soluble proteins in CSF and plasma. For better visualization, analytes have been grouped into ‘inflammatory’ (left panels) and ‘neurological’ (right panels) proteins. b Non-metric multidimensional scaling (NMDS) plots of 192 examined soluble proteins in CSF and plasma. Each patient is presented by one dot, and colored according to healthy controls, CNS inflammatory controls or Neuro-COVID class I-III. The ellipses represent the 95% confidence interval within subgroups. Source data are provided as a Source Data file.

Fig. 5. Individual CSF and plasma analytes discriminating different groups.

Box plot representations (center line at the median, upper bound at 75th percentile, lower bound at 25th percentile) of marginalized normal protein expression (NPX) of individual analytes significantly discriminating selected groups with whiskers at minimum and maximum values. Each dot represents one participant a Plasma, increasing NPX from class I (n = 18) to III (n = 15), and higher than in controls (n = 50): IL-6 (class III vs. I: p = 0.007, class III vs. infl. ctrl: p = 0.001), IL-8 (class III vs. I: p = 0.003, class III vs. infl. ctrl: p = 0.0002), HGF (class III vs. I: p = 0.04, class III vs. infl. ctrl: p = 0.0007), VEGFA (class III vs. I: p = 0.01, class III vs. infl. ctrl: p = 0.0005), EN-RAGE (class III vs. I: p = 0.003, class III vs. infl. ctrl: p = 0.0002), TNFRSF12A (class III vs. I: p = 0.006, class III vs. infl. ctrl: p = 0.002), PD-L1 (class III vs. I: p = 0.04, class III vs. infl. ctrl: p = 0.002), CCL23 (class III vs. infl. ctrl: p = 0.01), EZR (class III vs. infl. ctrl: p = 0.002), TNFRSF11B (class III vs. infl. ctrl: p = 0.049). b Plasma, decreasing NPX from class I (n = 18) to III (n = 15), higher in controls (n = 50) than in COVID-19 patients (n = 40): BMP-4, CLEC10A, CNTN5, GDF-8, NTRK2, GDNFRalpha, ROBO2. c Plasma, class-independent, higher NPX in COVID-19 (n = 35) than in controls (n = 50): 4E-BP1 (class III vs infl. ctrl: p = 0.007). d Plasma, higher NPX in COVID-19 (n = 40) than in controls (n = 50), decreasing from class I (n = 18) to III (n = 15): HAGH (class III vs. I: p = 0.008). e Cerebrospinal fluid (CSF), increasing NPX from class I (n = 16) to III (n = 14): IL-8 (class III vs. I: p = 0.012), MSR1 (class III vs. I: p = 0.016), 4E-BP1 (class III vs. I: p = 0.02), CD200R1 (class III vs. I: p = 0.04), TNFRSF12A (class III vs. I: p = 0.008), EZR (class III vs. I: p = 0.01). f CSF, increasing NPX from class I (n = 16) to III (n = 14), and higher in class III than in inflammatory controls (n = 25): TNFRSF11B (class III vs. I: p = 0.04, class III vs. infl. ctrl: p = 0.02). Source data are provided as a Source Data file. Statistics (a–f): statistical significance was calculated using two-sided Mann–Whitney-U test and p values were adjusted using Benjamin-Hochberg (BH)-procedure (p: *<0.05; **<0.01; ***<0.001, if not otherwise indicated: not significant).

Fig. 6. Specific CSF and plasma analytes correlate with COVID-19 severity and have predictive value for class III development.

a Venn diagram representing cerebrospinal fluid (CSF) and plasma mediators associated with COVID-19 severity assessed with the WHO progression scale. The association of protein sets and COVID-19 severity was assessed using a complement of four models (ordinal backward, ordinal-forward, best linear, most-regularized ordinal). Results of each model were finally cross-referenced to provide robust data sets of mediators associated with severe COVID-19. Plasma MSR1 and CSF TNFRSF12A and IL-8 represent the most robust set of proteins associated with COVID-19 severity (high association in each model used). Plasma IL-8 and IL-6, TNFRSF11B and EZR CSF levels depicted strong association using two models, whereas plasma 4E-BP1 and CSF PD-L1, BMP-4, CLEC10A and ROBO2 were strongly associated with COVID-19 severity in one model only. b ROC-AUC analysis of class I and II vs. class III. Five predictive plasma markers, including IL-8, EN-RAGE, TNFRSF12A, MCP-3, and 4 CSF markers, including 4E-BP1, EZR, TNFRSF12A, MSR1, emerged for the prediction of class III development. The Y-axis represents the sensitivity, the X-axis represents the 1-specificity (represented for IL-8, plasma). The names of relevant proteins in the study are compiled in Supplementary Data 4. c Plot representing the ROC-AUC values on the Y-axis vs. the random forest importance score on the X-axis. Relative importance of each single protein is represented by a high random forest importance score. Red points: plasma proteins, light blue points: CSF proteinsd Venn diagram representing CSF and plasma mediators associated with COVID-19 severity assessed with the WHO progression scale (COVID-19 severity) and mediators with high predictive value for class III development (Neuro-COVID severity). Plasma IL-8, CSF TNFRSF12A and CSF EZR depicted a high predictive value for severe COVID-19 and also class III development. lm linear model, RMSE root mean squared error.

Fig. 7. Routine brain imaging, regional GMVs and association with inflammatory CSF parameters.

a–c Conventional brain MRI and CT scans depicting exemplary imaging findings. Scale bar 15 mm (MRI) and 15.4 mm (CT). a Class I: Axial FLAIR images of the same class I patient show multifocal hyperintensities in the right precentral gyrus (top left), semioval center, left frontal cortex (top right), deep white matter, periventricular region (bottom left), right temporal lobe, left parietal white matter (bottom right). b Class II: Axial FLAIR images of a class II patient depict multifocal hyperintensities in the left frontal superior gyrus (top left), white matter of left frontal lobe (top right), left parahippocampal white matter (bottom left), right mesial temporal region (bottom right). c Class III: Axial FLAIR image shows bilateral thalamic hyperintensities (top left). Axial T1-weighted image depicts left middle cerebral artery (M2-segment) enhancement in the insular cistern (top right). Coronal CT scans demonstrate right cerebellar infarction (bottom left) and right temporo-occipital intracerebral hemorrhage (bottom right) (secondary to thrombosis of the right sigmoid sinus). d–f Conventional pre-COVID-19 MRIs (left) and MRIs during COVID-19 (right). d, e Axial DWI of the same class I patient depicts hyperintense signal alterations during COVID-19 (right sided MRI scans, orange arrows) in the left cerebellar hemisphere (d) and frontal juxtacortical region (e). f Axial FLAIR imaging demonstrates bilateral hyperintense signal alterations in the cerebellar peduncles (right sided MRI scan, orange arrows) of a class III patient. g–i Map (g) and 3D view (h, i) of the 16 brain regions with significant correlation values of gray matter volume (GMV) and clinical variables in the Neuro-COVID group after multiple comparison correction (FDR). These regions are represented in different colors on a T1-weighted template. j shows a matrix representing the association significance (significant p-corrected <0.05 in red squares). Associations between regional volumes and clinical measures were assessed using partial correlation, allowing to calculate the linear partial correlation between variables of interest adjusting for different covariates (age, sex, age*sex interaction, MRI magnetic field strength, total intracranial volume (TIV)). Statistical analysis was performed using the JASP software (https://jasp-stats.org/). MRIcroGL software was used to generate this figure (https://www.nitrc.org/projects/mricrogl). Source data of (g–j) are provided as a Source Data file. L left, R right, leuk leukocytes, prot protein, albR Albumin CSF-plasma ratio.

Fig. 8. Specific CSF and plasma mediators have high predictive value to forecast long-COVID.

ROC-AUC analysis of CSF and plasma parameters and long-COVID. The Y-axis represents the sensitivity, the X-axis represents the 1-Specificity (represented for plasma MCP-3). a Assuming an AUC cut-off of >0.75 for single mediators, high MCP-3 and CLM-6, as well as low RGMA plasma levels were predictive for long-COVID. Confusion matrix analysis of predictive plasma proteins revealed high predictive value of a plasma mediator signature consisting of RGMA, EZR, FcRL2 and ST1A1. b Assuming an AUC cut-off of >0.75, three single CSF proteins emerged for the prediction of long-COVID development. Low CSF levels of TRANCE, as well as high TNFRSF9 and IFN-γ levels were the best predictors. Confusion matrix of CSF proteins revealed high predictive power of a mediator pattern composed of TNFRSF9, IFN-γand TRAIL.

Fig. 9. Overview of proposed pathomechanisms leading to Neuro-COVID.

The proposed main determinants of severe Neuro-COVID are: (1) peripherally induced cytokine derangements, followed by (2) impaired BBB with ingressing polyreactive autoantibodies, resulting in (3) microglia reactivity and neuronal damage. Created with Biorender.com.