Para-infectious brain injury in COVID-19 persists at follow-up despite attenuated cytokine and autoantibody responses

Abstract

To understand neurological complications of COVID-19 better both acutely and for recovery, we measured markers of brain injury, inflammatory mediators, and autoantibodies in 203 hospitalised participants; 111 with acute sera (1-11 days post-admission) and 92 convalescent sera (56 with COVID-19-associated neurological diagnoses). Here we show that compared to 60 uninfected controls, tTau, GFAP, NfL, and UCH-L1 are increased with COVID-19 infection at acute timepoints and NfL and GFAP are significantly higher in participants with neurological complications. Inflammatory mediators (IL-6, IL-12p40, HGF, M-CSF, CCL2, and IL-1RA) are associated with both altered consciousness and markers of brain injury. Autoantibodies are more common in COVID-19 than controls and some (including against MYL7, UCH-L1, and GRIN3B) are more frequent with altered consciousness. Additionally, convalescent participants with neurological complications show elevated GFAP and NfL, unrelated to attenuated systemic inflammatory mediators and to autoantibody responses. Overall, neurological complications of COVID-19 are associated with evidence of neuroglial injury in both acute and late disease and these correlate with dysregulated innate and adaptive immune responses acutely.

Fig. 1. Brain injury markers are elevated acutely in COVID-19 participants with an abnormal Glasgow coma scale score (GCS) and in participants who experienced a neurological complication associated with COVID-19.

a The acute ISARIC cohort included Day 1–11 hospital admission timepoints. b–e Acute serum brain injury markers were assessed by Simoa: b NfL, c UCH-L1, d GFAP, and e tTau. All four were elevated in COVID-19 cases with normal Glasgow coma scale scores (GCS) relative to controls overall. Dotted lines show lower limit of quantification (LLOQ). f–j The Simoa analyses were performed for the sera from the COVID-CNS COVID and neuro-COVID groups at early and late convalescent timepoints (g, j) showing persistence of NfL, GFAP, and tTau in COVID participants, with NfL higher in neuro-COVID than COVID participants. k Within the combined early and late convalescent COVID-19 neurological cases, the highest levels of NfL were observed in participants who had suffered a cerebrovascular event at the time of SARS-CoV-2 infection. l tTau levels were raised in the cerebrovascular, CNS inflammatory and seizure conditions. m, n Serum NfL remained elevated in both the early (<6 weeks from positive SARS-CoV-2 test) and late convalescent phases (>6 weeks) in neuro-COVID compared to COVID non-neurological cases. o, p GFAP was elevated in neurological cases in the early and late convalescent phase. Box and whisker plots show all data points with median as centre line with 25th and 75th percentiles. Sample sizes shown in (a) and (f). Group comparisons are by Kruskal–Wallis test with Dunn’s post-hoc multiple comparison test, no statistical comparison made for panel (h) as medians were at LLOQ.

Fig. 2. Immune mediators are elevated acutely and correlate with different brain injury markers at different timepoints.

Serum mediators from the ISARIC and COVID-CNS cohorts were assessed by Luminex. a A volcano plot was generated to identify mediators which were elevated in participants with an abnormal GCS (GCS ≤ 14) compared to normal GCS (GCS = 15) and b, c a network analysis identified the highest correlations between the mediators (***significantly different from ISARIC GCS = 15 by Steiger test p < 0.001). d Unbiased Euclidean hierarchical cluster and correlation analyses identified two clusters of up-regulation of several pro-inflammatory mediators in concert. The first group included interleukin (IL)-6, IL-12p40, CCL2, CXCL9 and hepatocyte growth factor (HGF) and the second group included the IL-1 family, interferons, and macrophage colony stimulating factor (M-CSF); to the right is shown the correlations between each cytokine with the four brain injury biomarkers (significance indicated by asterisks). e Network analysis and heatmaps of correlations between mediators did not demonstrate the tight interconnectedness that had been identified in acute samples and there were differences between neuro-COVID (e) and ISARIC GCS = 15 and GCS ≤ 14 by Steiger test (***p < 0.001). f At this later stage several mediators correlated with tTau. Volcano plot used multiple two-tailed Mann–Whitney U tests with a false discovery rate set to 5%. Correlations are Pearson’s coefficients (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Fig. 3. There is an IgM antibody response in participants with COVID-19 directed at SARS-CoV-2 spike protein and against several self-antigens.

a Acute samples were tested for IgM antibodies by protein microarray with normalized fluorescence Z-scores shown. b COVID-19 participants showed considerably more binding ‘hits’ than healthy controls (fluorescence with a Z-score of 3 or above compared to controls), although overall there was no difference in the acute samples between participants with normal (GCS = 15) or abnormal GCS (GCS ≤ 14)). Nevertheless, c COVID-19 participants with abnormal GCS (GCS ≤ 14) more frequently had raised IgM antibodies than COVID-19 participants with a normal GCS (GCS = 15), including those directed at SARS-CoV-2 spike protein (Fisher’s exact tests *p < 0.05). d A chord diagram shows the associations between antibodies, including those against Spike. e IgM antibodies were also analysed in the convalescent participants. f A largr proportion of COVID and Neuro-COVID participants had positive antibody ‘hits’ for IgM (defined by Z-score 3 and above compared to controls). g Of those antibodies against self-antigens identified, they were only two with different frequencies between the groups (Fisher’s exact tests *p < 0.05). At this timepoint there was no significant difference in the proportion of individuals with IgM against SARS-CoV-2 spike or nucleocapsid epitopes. Violin plots show all data points with median at centre line and 25th and 75th quartile lines. Group comparisons are by Kruskal–Wallis test with post-hoc Dunn’s multiple comparison test, pairwise comparisons by two-tailed Mann–Whitney U test, and correlations are Pearson’s coefficients.

Fig. 4. There is an IgG antibody response in participants with COVID-19 directed at SARS-CoV-2 nucleocapsid and spike proteins and against several self-antigens.

a Acute samples were tested for IgG antibodies by protein microarray with normalized fluorescence Z-scores shown. b COVID-19 participants showed considerably more binding ‘hits’ than healthy controls (fluorescence with a Z-score of 3 or above compared to controls) but, overall, there was no difference in the acute samples between participants with normal (GCS = 15) or abnormal GCS (GCS ≤ 14). c COVID-19 participants with abnormal GCS more frequently had several raised IgG antibodies than COVID-19 participants with a normal GCS, including those directed at SARS-CoV-2 spike protein and several CNS proteins ((DRD2, GRIN3B, and UCH-L1) Fisher’s exact tests * p < 0.05)). d A chord diagram shows the association of antibodies with differences in frequency, including those against Spike. e IgG antibodies in early and late convalescent sera were also analysed. f A larger proportion of Neuro-COVID participants had positive antibody ‘hits’ for IgG (defined by Z-score 3 and above compared to controls). g Of those antibodies against self-antigens identified, only five showed a difference in frequency between groups (Fisher’s exact tests *p < 0.05). At this timepoint there was no significant difference in the proportion of individuals with IgG against SARS-CoV-2 spike or nucleocapsid epitopes. h A chord diagram shows the association of antibodies with differences in frequency plus anti-Spike antibody. i Representative images of rat brains incubated with participants’ sera and screened for IHC binding of anti-human IgG to detect CNS reactivity. Scale bars = 1 mm. j Percentage of participant serum IgG reactivity to rat brainstems, detected by Fisher’s exact test with Benjamini and Hochberg correction. Violin plots show all data points with median at centre line and 25th and 75th quartile lines. Group comparisons are by Kruskal–Wallis test with post-hoc Dunn’s multiple comparison test, pairwise comparisons by two-tailed Mann–Whitney U test, and correlations are Pearson’s coefficients.