key: cord-0694677-t3xd28ly
authors: Song, Eric; Bartley, Christopher M.; Chow, Ryan D.; Ngo, Thomas T.; Jiang, Ruoyi; Zamecnik, Colin R.; Dandekar, Ravi; Loudermilk, Rita P.; Dai, Yile; Liu, Feimei; Sunshine, Sara; Liu, Jamin; Wu, Wesley; Hawes, Isobel A.; Alvarenga, Bonny D.; Huynh, Trung; McAlpine, Lindsay; Rahman, Nur-Taz; Geng, Bertie; Chiarella, Jennifer; Goldman-Israelow, Benjamin; Vogels, Chantal B.F.; Grubaugh, Nathan D.; Casanovas-Massana, Arnau; Phinney, Brett S.; Salemi, Michelle; Alexander, Jessa R.; Gallego, Juan A.; Lencz, Todd; Walsh, Hannah; Wapniarski, Anne E.; Mohanty, Subhasis; Lucas, Carolina; Klein, Jon; Mao, Tianyang; Oh, Jieun; Ring, Aaron; Spudich, Serena; Ko, Albert I.; Kleinstein, Steven H.; Pak, John; DeRisi, Joseph L.; Iwasaki, Akiko; Pleasure, Samuel J.; Wilson, Michael R.; Farhadian, Shelli F.
title: Divergent and self-reactive immune responses in the CNS of COVID-19 patients with neurological symptoms
date: 2021-05-03
journal: Cell Rep Med
DOI: 10.1016/j.xcrm.2021.100288
sha: 4d203a0b039cc641c71bd1cf6530f5617fc77bab
doc_id: 694677
cord_uid: t3xd28ly

COVID-19 patients frequently develop neurological symptoms, but the biological underpinnings of these phenomena are unknown. Through single cell RNA-seq and cytokine analyses of CSF and blood from COVID-19 patients with neurological symptoms, we find compartmentalized, CNS specific T cell activation and B cell responses. All COVID-19 cases had CSF anti-SARS-CoV-2 antibodies whose target epitopes diverged from serum antibodies. In an animal model, we find that intrathecal SARS-CoV-2 antibodies are found only during brain infection, and are not elicited by pulmonary infection. We produced CSF-derived monoclonal antibodies from a COVID-19 patient, and find that these mAbs target both anti-viral and anti-neural antigens—including one mAb that reacted to both spike protein and neural tissue. Overall, CSF IgG from 5/7 patients contains anti-neural reactivity. This immune survey reveals evidence of a compartmentalized immune response in the CNS of COVID-19 patients and suggests a role for autoimmunity in neurologic sequelae of COVID-19.

The causative pathogen of pandemic coronavirus disease 2019 (COVID- 19) , SARS-CoV-2, primarily causes a respiratory illness. Yet, in some patients, SARS-CoV-2 infection is associated with severe and debilitating neurological symptoms 1 . About a third of patients with moderate to severe COVID-19 experience neurological sequelae including anosmia, dysgeusia, headache, impaired consciousness, and seizure-only some of which is explained by systemic complications including hypercoagulability 2 . Rarely, SARS-CoV-2 RNA is detected in the CSF of COVID-19 patients, and some studies have found evidence of SARS-CoV-2 protein in brain parenchyma. Yet, in either case, there is little evidence that SARS-CoV-2 directly damages neural tissue 3-7 . Collectively, these observations suggest that mechanisms other than direct cytopathic effects of SARS-CoV-2 contribute to neurological symptoms. Therefore, a broad characterization of CNS immunity may provide further insight into the causes of neurologic impairment in COVID- 19 . In this exploratory study, we profiled intrathecal and peripheral immune responses in patients with COVID-19 complicated by diverse neurological symptoms.

Hospitalized COVID-19 patients with varied neurological symptoms who underwent clinicallyindicated lumbar puncture were consented for collection of surplus CSF to be used for research.

Six participants with acute COVID-19, based on positive SARS-CoV-2 RT-qPCR of nasopharyngeal swab, were enrolled (Supplementary Table 1 , Supplementary Case File).

Neurological symptoms included encephalopathy, intractable headache, and seizure. All participants donated paired blood and CSF, except for one patient who did not donate blood.

Lumbar punctures were performed on median hospital day 12.5 (range 2 -43 days). Prepandemic CSF from age and gender-matched healthy controls (n=3) was obtained from a neuroinfectious disease biorepository at Yale. Because fresh CSF is required for single cell transcriptomics, we recruited additional uninfected control participants during the COVID-19 pandemic (n=3): 2 were healthy community-dwelling adults, and one was hospitalized for the work-up of frequent falls. Additional blood and CSF single cell sequencing data was included from healthy controls derived publicly available data (n=8) 8 were negative for SARS-CoV-2 RNA by RT-qPCR using the CDC primer-probe sets 9 .

Transcriptional analysis reveals a coordinated innate immune cell response to In the CSF of COVID-19 patients, dendritic cells had an activated transcriptional profile, as 57% and 47% of their upregulated genes were classified as type 1 and type 2 interferon stimulated genes, respectively ( Figure 1c ). Genes associated with NK cell activation also were upregulated in the CSF of COVID-19 patients (Figure 1d ). While NK cells in the CSF and the peripheral blood demonstrated comparable changes in the number of differentially expressed genes in COVID-19 cases compared to controls, the genes affected were mostly unique to each compartment (Supplementary Figure 1c) . Using CellphoneDB signaling network analysis 10, 11 , in COVID-19 patients, CSF dendritic and NK cells were predicted to have significantly increased interactions with CSF CD8 and CD4 T cells relative to healthy controls, while interactions between CD4 T cells and monocytes were diminished, suggesting a dysregulated innate to adaptive immune interface (Figure 1e ).

Because signaling network analysis predicted that T cells were the main recipients of altered innate-adaptive cross talk, we isolated and re-clustered CSF and peripheral T cells for 

To validate the transcriptional enrichment in IL-12 and IL-1 signaling in the CSF of COVID-19 patients with neurologic symptoms, we measured inflammatory cytokine levels in the CSF and plasma using a Luminex cytokine panel (Figure 2g -h). Consistent with the single-cell RNA sequencing results, IL-1b and IL-12 were elevated in the CSF of COVID-19 cases compared to healthy controls, but were not elevated in the plasma of COVID-19 cases.

J o u r n a l P r e -p r o o f 6 Conversely, CCL2, CXCL9, and IL-8 were significantly increased in the plasma of COVID-19 cases compared to controls, but not in their CSF. Because IL-12 is thought to be produced by activated antigen presenting cells to orchestrate Th1 responses through T and NK cell activation, we examined the cellular source of IL-12 in COVID-19 cases. The innate immune cells with the highest IL12A expression were CSF NK and dendritic cells (Supplementary Figure1e). Taken together, these data support the single cell RNA sequencing analyses that identified IL-12 as differentially expressed in CSF but not blood innate immune cells of COVID-19 cases.

Moreover, they suggest a distinct effect of COVID-19 in the CNS on cytokines important for innate immunity and for the induction of cell-mediated immunity, including IL-1 and IL-12.

We found a significant enrichment of B cells in the CSF of COVID-19 cases when distinct CSF plasma cell clusters. We therefore asked whether antibody-secreting B cells in the CSF exhibit a different anti-SARS-CoV-2 antibody profile than the those in the periphery. To do so, we utilized a recently developed SARS-CoV-2 epitope Luminex panel 12 to screen for anti-SARS-CoV-2 antibodies in the CSF and plasma of COVID-19 cases and controls. As expected, anti-SARS-CoV-2 antibodies were not detected in any controls. In contrast, all COVID-19 cases had anti-SARS-CoV-2 antibodies in both CSF and plasma. However, while all COVID-19 cases developed antibodies to SARS-CoV-2 spike and nucleocapsid in both the plasma and CSF, antireceptor binding domain (RBD) antibodies were rare in CSF but uniformly present in the plasma ( Figure 3c ). In addition, we found that in all COVID-19 cases, both the relative prevalence (rank score: 12 being most frequent, 1 being least frequent; Figure 3d ) and levels of antibody (Supplementary Figure 5d ) diverged between the CSF and plasma, indicating a different anti-SARS-CoV-2 antibody profile between the CSF and plasma of the same patient.

Direct detection of SARS-CoV-2 in the CSF is extremely rare in reported cases of neurological complications of COVID-19 13 , and SARS-CoV-2 RNA was not detected in the CSF J o u r n a l P r e -p r o o f 7 our cohort. However, we detected intrathecal anti-viral antibodies in all cases. In some other encephalitis-causing viruses, including West Nile virus, Japanese encephalitis virus and measles virus [14] [15] [16] , the presence of anti-viral antibodies is consistent with viral neuroinvasion, even in the absence of viral nucleic acid. To determine whether CNS infection is sufficient to stimulate a CNS humoral response during COVID-19, we used a recently developed mouse model that reliably dissociates pulmonary and neurological infection of SARS-CoV-2 17 .

We used an adeno-associated virus to express the human ACE2 (hACE2) receptor in the lung, brain, or the lung and brain, allowing us to target SARS-CoV-2 infection to specific tissues. First, we used mice that express hACE2 in both the lung and the brain and administered SARS-CoV-2 intranasally (Figure 4a ). This permits SARS-CoV-2 to infect the lung and the brain. In these mice, we detected increased titers of SARS-CoV-2 RNA in both lung and brain tissue following inoculation. However, despite robust brain infection, we did not detect SARS-CoV-2 RNA in the CSF of these mice (Figure 4b ). This suggests that direct detection of SARS-CoV-2 RNA in CSF at a single time point may be insensitive to parenchymal or short lived SARS-CoV-2 neuroinvasion.

We next used the mouse model to evaluate whether the detection of intrathecal anti-SARS-CoV-2 antibodies in our COVID-19 patients was more likely triggered by a local antigen (i.e., as a consequence of SARS-CoV-2 neuroinvasion) or reflected passive transfer of antibody from the systemic circulation 13 . When SARS-CoV-2 was administered intranasally to mice expressing hACE2 only in the lung (thus generating mice with pulmonary but not brain infection), we detected significantly elevated anti-spike SARS-CoV-2 IgG in the lungs and serum, but not in the brain or CSF (Figure 4b , red). When SARS-CoV-2 was administered intranasally to mice expressing hACE2 in both the brain and lung (thus generating mice with pulmonary and brain infection), we detected increased anti-spike antibodies in all four compartments: lung, serum, brain, and CSF (Figure 4b , orange). Finally, when hACE2 was expressed in brain only and SARS-CoV-2 was administered intracranially (causing infection in the brain but not the lung), we detected increased anti-spike antibodies in the brain and CSF, but not in the lungs or serum (Figure 4b , green). These data support the hypothesis that CSF antibodies do not solely reflect the passive transfer of antibodies from the systemic circulation.

Indeed, in these mice, anti-spike antibodies in the CSF and brain were only observed in the anti-spike monoclonal antibody for neutralizing activity against wild type SARS-CoV-2 18 . None of the monoclonal antibodies exhibited neutralizing activity at concentrations ranging from 2.5 -25µg /mL (Supplemental Figure 6 ).

Given reports of new-onset humoral autoimmunity in COVID-19, we wondered whether any of the monoclonal antibodies from CSF-expanded B cells were autoreactive to neural tissue. Therefore, we tested all monoclonal antibodies using a standard and validated screening method for anti-neural autoreactivity, anatomic mouse brain tissue staining. 19 Monoclonal antibodies were used as a primary antibody to immunostain mouse brain tissue and labeled with an antihuman IgG secondary antibody. An anti-influenza antibody targeting the hemagglutinin antigen (anti-HA) was used as a negative control. Similar to anti-HA, none of the PBMC-derived J o u r n a l P r e -p r o o f 9 monoclonal antibodies recognized mouse brain tissue. In contrast, four of five CSF-derived monoclonal antibodies exhibited some degree of anti-neural immunoreactivity-including the anti-spike monoclonal antibody (mAb C2) ( Figures 5D and E) . Notably, mAb C2 produced a neuropil-predominant immunostaining pattern suggesting that the antigen may be enriched in neuronal process or harbor an extracellular epitope.

The emergence of inflammatory and humoral autoimmune disorders of the nervous system during the para-or post-infectious period in COVID-19 is increasingly recognized and includes: acute disseminated encephalomyelitis (ADEM), autoimmune encephalitis associated with known autoantibodies, transverse myelitis, Guillain-Barré syndrome and one of its variants, Miller Fisher syndrome [20] [21] [22] [23] [24] . Given this literature, and the autoreactivity of CSF-derived monocloncal antibodies from case 1, we hypothesized that our other COVID-19 cases might harbor intrathecal autoantibodies. To test this, we screened our cohort of COVID-19 cases for intrathecal anti-neural antibodies using a suite of complementary autoantigen detection platforms: anatomic mouse brain tissue immunostaining, immunoprecipitation mass spectrometry (IP-MS), and pan-human proteome phage display immunoprecipitation sequencing (PhIP-Seq) [25] [26] [27] . In these screens, we included one additional patient with post-COVID-19 seizures and cognitive impairment who had been recruited after the completion of the transcriptomic and cytokine analyses were completed (Case 7, Supplementary Figure 7) .

More COVID-19 CSF (5 of 7) were immunoreactive to mouse brain tissue at a 1:10 dilution than control CSF (2 of 6) ( Figure 5A and Fig s8A) . Control CSF staining was not specific to any anatomic region, weakly pan-nuclear, or primarily subpial (Fig s8B) . None of the control CSF samples were immunoreactive beyond a 1:10 dilution, indicating the absence of high titer or high affinity anti-neural autoantibodies. In contrast, at a 1:10 dilution, COVID-19 CSF produced immunoreactive staining of specific anatomic regions including: cortical neurons 

To screen for the neural protein targets of intrathecal autoantibodies, we immunoprecipitated whole mouse brain lysate using CSF and plasma, trypsinized precipitated proteins, and analyzed the resulting peptides by mass spectrometry. Immunoprecipitations (IPs) were performed in technical replicate, by different individuals, using different mice as input.

First, we searched resulting spectra against the SARS-CoV-2 proteome (Uniprot SARS-CoV-2 reference proteome, 06/2020). SARS-CoV-2 proteins were not detected. We then searched for human proteins. Consistent with circulating protein expression patterns 28, 29 and circulating prothrombotic autoantibodies 30 in COVID-19, complement, coagulation, and platelet degranulation pathway proteins of human origin were significantly overrepresented in the IgGbound protein fraction of plasma from our cases (Supplementary Figure 9 ).

To specifically identify candidate anti-neural autoantibodies in the CSF, we searched for mouse proteins that were observed in both technical replicates and significantly enriched by spectral counting and/or 1.5x enriched by MS1 peak area in COVID-19 CSF immunoprecipitations relative to controls 31 . Between 5 and 56 (median = 20) proteins were enriched by the CSF, but not by the plasma of the same case or control CSF, indicating they were unique to the CSF compartment of COVID-19 patients. Gene ontology pathway analysis indicated that both COVID-19 CSF and plasma immunoprecipitation-mass spectrometry fractions were enriched for brain-enriched and synaptic proteins. In some cases, neural antigens were statistically enriched by CSF but not plasma (e.g., NEFM and NEFH by Cases 3 and 4, and APP by Case 4).

J o u r n a l P r e -p r o o f 11 COVID-19 CSF was also screened for autoantibodies using a previously described PhIP-Seq platform (T7 bacteriophage display) displaying ~730,000 overlapping 49 amino acid peptides spanning all human proteins 25 , including all known and predicted isoforms 26, 27 . To identify peptides that were significantly enriched by COVID-19 CSF compared to controls, we first established an empirical enrichment threshold using a validated commercial antibody targeting the protein GFAP (Supplemental Data File 4). For COVID-19 CSF samples, peptides with suprathreshold enrichment in both technical replicates were considered candidate autoantigens.

Data File 4). By gene ontology, synaptic proteins were also enriched by COVID-19 CSF (e.g., NRG3, SYNJ2, and DPYSL2) (Bonferroni-corrected p = 1.6x10 -1 ). COVID-19 plasma PhIP-Seq candidates were enriched for transcriptional activators of catabolism (Bonferroni-corrected p = 2.6x10 -2 ), but not synaptic or brain-enriched proteins. COVID-19 CSF that was immunoreactive to mouse brain tissue at a 1:50 dilution was associated with greater enrichment of candidate autoantigens by PhIP-Seq (19 -40, median = 37) than CSF that immunostained at a lower dilution or was not immunoreactive (1 -16, median = 6) suggesting a correlation between immunostaining status and the burden of CSF autoantibodies. In some instances, the same candidate autoantigen was detected in COVID-19 cases by IP-MS and PhIP-seq: UHRF1BP1 (Case 2), NUAK1 (Case 3), and DBN1 (Case 7).

To validate autoantibodies identified in CSF, we selected two COVID-19 candidate autoantigens that were enriched more so by CSF than plasma on PhIP-Seq: intraflagellar transport protein 88 homolog (IFT88) and THAP domain containing 3 (THAP3). To validate IFT88, HEK 293T cells were transfected with a plasmid encoding RFP-IFT88, methanol fixed, and immunostained with CSF from Case 1. Case 1 CSF IgG, but not control CSF IgG, colocalized with an anti-RFP and an anti-IFT88 commercial antibody ( Figure 6E and Figure   S10 ). To validate THAP3, lysate from HEK 293T cells overexpressing FLAG-tagged THAP3 was separated by SDS-PAGE, transferred to a PVDF membrane, and sequentially probed with 

Neurotropic viruses, such as herpes simplex virus, West Nile virus and HIV, are common pathogens of the CNS, but it is increasingly recognized that respiratory viruses, including influenza virus and human respiratory syncytial virus, also lead to neurological complications in a minority of patients [32] [33] [34] [35] [36] [37] [38] . The biological underpinnings of neurological complications of respiratory viruses are diverse, and include neuronal damage due to direct viral neuroinvasion, as well as parainfectious processes, including elevations of pro-inflammatory cytokines, and postviral autoimmune reactions. In this exploratory study, we performed an extensive set of immunologic investigations to assess for CNS-specific immune responses in a series of COVID-19 patients with neurologic symptoms. While systemic multi-organ dysfunction is almost certainly a driver of neurologic complications in a proportion of COVID-19 patients, here we identified both innate and adaptive anti-viral immune responses, as well as humoral autoimmunity that appears to be unique to the CNS, and may therefore contribute to COVID-19 neuropathology.

CSF, while not identical to brain, is produced by the choroid plexus and bathes the CNS.

It is the only CNS tissue surrogate readily sampled in living humans. Analysis of CSF immune cells has shed light on immune mechanisms of neuronal injury during other infections, including HIV, neurosyphilis, and neuroborreliosis [39] [40] [41] . By assessing CSF and blood in patients with acute COVID-19 and neurological symptoms, we find evidence for a compartmentalized CNS immune response to SARS-CoV-2. Through transcriptional and cytokine analyses, we find an increase in CSF but not plasma IL-12 and IL-1b, factors that are central for coordinating innate and adaptive immune responses to invading pathogens. Notably, neuroinvasion of mouse hepatitis virus, a coronavirus of laboratory mice, also leads to IL-12 production by astrocytes and microglia 42 .

Our data identified increased and divergent humoral responses within the CNS. This Hippocampal immunostaining from three cases was primarily restricted to the CA3 region similar to a recent report of SARS-CoV-2 associated encephalitis 43 . Other immunoreactive anatomic regions included the olfactory bulb in three cases and cerebrovasculature in twoanatomic regions with prima facie relevance to common neurologic sequelae of COVID-19 (i.e., anosmia and stroke). Subsequent unbiased protein and peptide screens of CSF identified a diversity of candidate autoantibodies, and two of these autoantigens, THAP3 and IFT88, were subsequently validated. IFT88 is a ciliary protein whose mutation causes a ciliopathy in humans and anosmia in mice 44 . THAP3 is expressed in brain, among other organs, and may be implicated in genetic causes of dystonia 45 .

However, the mere presence of an intrathecal autoantibody does not mean that a patient has an autoimmune encephalitis. Indeed, the patients in our exploratory cohort lacked evidence for active inflammation on neuroimaging and/or did not have elevated conventional CSF markers of neuroinflammation (i.e., white blood cell count, IgG index, and CSF-restricted oligoclonal bands) that are typically, but not always, found in patients with autoimmune encephalitis.

Notably, COVID-19 patients with neurological symptoms appear to have immune responses to multiple autoantigens, implying that the increased compartmental humoral immune response may reflect a broader immune activation syndrome. This is particularly true given that humoral autoimmunity has been observed to target other organ systems in COVID-19, and may also contribute to neuropathology during COVID-19 46-49 .

J o u r n a l P r e -p r o o f 14 Taken together, our exploratory data suggest that even in COVID-19 patients with neurologic symptoms who lack overt evidence for neuroinflammation on MRI or conventional CSF testing, there is a compartmentalized immune response involving the innate and adaptive arms of the immune system. Future research, involving careful clinical phenotyping and timely investigations of the CSF, will help place these findings into a broader clinical context and inform whether anti-viral and/or immunomodulatory therapies might be indicated for carefully selected neurologically impaired patients with COVID-19.

Limitations of the Study: 

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Shelli F. Farhadian (Shelli.Farhadian@yale.edu).

Human monoclonal antibodies Immunoglobulin sequencing data for patient-derived monoclonal antibodies are available in Supplementary Data File 5.

Gene expression and repertoire data in the study are available in the NCBI repository SRA 

Six to twelve-week-old mixed sex C57Bl/6 (B6J) purchased from Jackson laboratories were subsequently bred and housed at Yale University. All procedures used in this study (sex- Cells were maintained at 37°C in 5% CO 2 and split 1:5 every 3 -5 days based on having reached 80 -100% confluence by visual inspection. Cells were not passaged more than 50 times. For overexpression cell-based assays, 0.5mL of 293 cells at 200,000 cells / mL were added to each well of a 24-well plate the day prior to transfection.

RNA was extracted from nasopharyngeal swabs, CSF, and plasma using the MagMax Viral/Pathogen Nucleic Acid Isolation kit. A modified CDC RT-qPCR assay was used to detect SARS-CoV-2 with the N1, N2, and human RNase P (RP) primer-probe sets and the NEB Luna Universal Probe One-Step RT-qPCR kit on the Bio-Rad CFX96 Touch Real-Time PCR Detection System 9 .

Peripheral blood mononuclear cells (PBMCs) were isolated from heparinized whole blood after 1:1 PBS dilution. The blood was layered over a Histopaque (Sigma-Aldrich, #10771-500ML) gradient in a SepMate tube (Stemcell Technologies; #85460) and isolated according to manufacturer's instructions. The PBMCs were then aliquoted and stored at -80 for subsequent analysis. CSF was centrifuged at 400G for 8 minutes, with cell-free supernatant removed for cytokine and antibody assays, and cell pellet processed for single cell RNA sequencing, as below.

Approximately 8,000 single cells from CSF and from PBMC from each participants were loaded into each channel of a Chromium single-cell 5′ Chip (V3 technology). 5' 10X libraries were sequenced on Illumina Novaseq at approximately 50,000 reads per cell. Raw sequencing reads were aligned to the human GRCh38 genome and gene counts were quantified as UMIs using Cell Ranger count v3.0 (10x Genomics). We removed cells with >10% mitochondrial RNA content, and included cells with > 500 and <2000 genes expressed per cell. Dimensionality reduction, clustering, and visualization was performed using Seurat. Clusters were identified based on expression of canonical immune cell markers (heatmap Supplementary Figure 1a) .

Interferon regulated genes were identified using Interferome 50 To identify potential intercellular interactions between different cell types in the scRNAseq data, we utilized CellPhoneDB v2 10 . Normalized count matrices and associated cell type labels were provided to CellPhoneDB and analyzed under both the statistical mode and the thresholding mode. Of note, since the statistical mode of CellPhoneDB seeks to assess the specificity of a given interaction, a lack of statistical significance does not necessarily mean a given interaction is not present. Therefore, when comparing the number of potential intercellular interactions in COVID-19 patients vs healthy controls, the simpler threshold-based analysis mode was used. In contrast, for pinpointing the top candidate cell-cell interactions in each dataset, the statistical analysis mode was used, with a significance threshold of p < 0.05.

We initially combined all 76,473 CSF and blood cells and generated clusters using Seurat. For each cluster we assigned a cell-type label using statistical enrichment for sets of marker genes, and manual evaluation of gene expression for small sets of known marker genes. We then created a separate Seurat object consisting only of T cells clusters from the original analysis, and a separate Seurat object consisting only of plasma and B cells. We then re-clustered these T and B cells and annotated sub-clusters using previously annotated marker genes.

Single cell V(D)J sequences were generated using CellRanger vdj function. Assignments of V(D)J sequences were performed using IgBLAST v.1.6.1 with the September 12, 2018 version of the IMGT gene database (as described previously) 53 

Soluble chemokines and cytokines were assessed in CSF supernatant and paired plasma using the HD71 Human Cytokine Array/Chemokine Array (Eve Technologies, Calgary, AB).

Statistical analysis was carried out using Qlucore Omics Software, version 3.6 (Lund, Sweden).

Cytokines that were absent from CSF or plasma under both COVID and controls conditions were excluded from the respective analyses. Heatmaps were plotted using Z-scores, with the color scale set to range from -2 to +2. Hierarchical clustering was applied to samples.

Adeno-associated virus 9 encoding hACE2 were purchased from Vector biolabs (AAV-CMV-hACE2). Intratracheal injection. Animals were anaesthetized using a mixture of ketamine (50 mg kg −1 ) and xylazine (5 mg kg −1 ), injected intraperitoneally. The rostral neck was shaved and disinfected. A 5mm incision was made and the salivary glands were retracted, and trachea was visualized. Using a 500µL insulin syringe a 50µL bolus injection of 10 11 GC of AAV-CMV-hACE2 was injected into the trachea. The incision was closed with VetBond skin glue.

Following intramuscular administration of analgesic (Meloxicam and buprenorphine, 1 mg kg −1 ), animals were placed in a heated cage until full recovery. Intracisternal magna injection. Mice were anesthetized using ketamine and xylazine, and the dorsal neck was shaved and sterilized. A 2 cm incision was made at the base of the skull, and the dorsal neck muscles were separated using forceps. After visualization of the cisterna magna, a Hamilton syringe with a 15 degree 33 gauge needle was used to puncture the dura. 3µL of AAV 9 (3.10 12 viral particles/mouse) or mRNA (4-5 µg) was administered per mouse at a rate of 1µL min -1 . Upon completion of the injection, needle was left in to prevent backflow for an additional 3 minutes. The skin was stapled, disinfected and same post-operative procedures as intratracheal injections were performed.

To generate SARS-CoV-2 viral stocks, Huh7. intraperitoneally. After sterilization of the scalp with alcohol and betadine, a midline scalp incision was made to expose the coronal and sagittal sutures, and a burr holes were drilled 1 mm lateral to the sagittal suture and 0.5 mm posterior to the bregma. A 10 µl Hamilton syringe loaded with virus, and was inserted into the burr hole at a depth of 2 mm from the surface of the brain and left to equilibrate for 1 min before infusion. Once the infusion was finished, the syringe was left in place for another minute before removal of the syringe. Bone wax was used to fill the burr hole and skin was stapled and cleaned. Following intramuscular administration of analgesic (Meloxicam and buprenorphine, 1 mg kg −1 ), animals were placed in a heated cage until full recovery. For high condition, 5µL of SARS-CoV-2 (3x10 7 PFU/ml) and for low condition 5µL of SARS-CoV-2 (3x10 6 PFU/ml) was used.

ELISAs were performed as previously reported 56 . In short, Triton X-100 and RNase A were added to serum samples at final concentrations of 0.5% and 0.5mg/ml respectively and incubated at room temperature (RT) for 3 hours before use to reduce risk from any potential virus in serum. 

Statistical analyses were performed using commercially available software (Prism or Excel) except as noted. Differences in means between two groups were analysed using unpaired twosided t-tests, unless otherwise noted. For scRNA-seq analyses, we corrected for multiple comparisons and report adjusted P values using Benjamini-Hochberg correction. For pathway analyses, Fisher's exact test was used with Bonferroni correction for multiple testing.

Highly immunogenic linear regions of the SARS-CoV-2 proteome were isolated by ReScan and conjugated to Luminex beads as previously described 57 

The binding affinities of purified antibodies to recombinant SARS-CoV-2 spike protein 59 were measured on an Octet QK system using Anti-Human Fc-Capture (AHC) biosensor tips (Sartorius). Purified monoclonal antibodies were diluted to 5 µg/mL in phosphate buffered saline containing 0.1% bovine serum albumin and 0.02% Tween-20 at pH 7.4 (PBSTB).

Antibodies were loaded and analyzed according to the following protocol: (0) Tips preequilibrated in PBSTB for 10 min; (1) Equilibration in PBSTB for 60 s; (2) Antibody loaded on tips for 300 s; (3) Tips washed in PBSTB to reach baseline for 300 s; (4) Tips dipped in spike-containing wells to allow for spike association for 300 s; (5) Tips dipped in PBSTB to allow for dissociation for 1800 s. All steps were carried out at 30 °C with shaking at 1000 rpm. Binding curves for each antibody were recorded at spike concentrations ranging from 0-316 nM. A single affinity value for each antibody was calculated using a 1:1 global fit binding model (Octet Data Analysis HT software) with all R 2 values > 0.95.

SARS-CoV-2/human/USA/CA-UCSF-0001H/2020 from a UCSF clinical specimen was isolated and titered by standard plaque assay as described in 60 To assess neutralization capacity, monoclonal antibodies were incubated with SARS-CoV-2 for one hour at 37°C and virus/antibody dilutions were used to infect Vero E6 cells and Huh7.5.1 cells overexpressing ACE2-TMPRSS2 at an MOI of 1. 61 

Postnatal day 40 -60 mice (F1 generation of FVB x C57BL/6J cross) were transcardially perfused with 4% paraformaldehyde (PFA) and brains post-fixed in PFA overnight. After sucrose equilibration, brains were blocked in OCT and sectioned at 12 µm on a standard cryostat.

For screening and determination of anti-neural autoreactivity, sections were permeabilized and blocked in PBS containing 10% lamb serum and 0.1% triton x-100. Sections were then incubated with patient-derived monoclonal antibodies (18 µg/mL) or CSF at 1:10, 1:25, and 1:50 overnight at 4C. In some cases, CSF that was immunoreactive at 1:10 was repeated at 1:4 for additional confocal imaging. Sections were rinsed at least 5x with PBS and counterstained with anti-human IgG (Alexafluor 488). Nuclei were stained with DAPI at 1:2000 and stained sections were coversliped with ProLong Gold. Studies were approved the UCSF IACUC committee.

J o u r n a l P r e -p r o o f 24 Panoramic images of immunostained sagittal mouse brain sections were captured at a 20x on a Zeiss Axio Scan.Z1. Confocal images of sagittal mouse brain sections and HEK 293T cell-based assays were captured at 60X at the UCSF Nikon Imaging Center using a Nikon CSU-W1 spinning disk confocal microscope, equipped with an Andor Zyla sCMOS camera. Images of the SARS-CoV-2 neutralizing assay were collected on a Nikon Ti inverted fluorescence microscope. Postnatal day 40 -60 mice (F1 generation of FVB x C57BL/6J cross) were used as the source of antigen. Mice anesthetized in isofluorane and sacrificed by cervical dislocation. For each set, 3 brains (two males and one female for one replicate and two females and one male for the other replicate) were rapidly dissected in ice cold PBS. For each replicate, 3 brains were homogenized in ice cold tissue lysis buffer (7 mL) using a dounce homogenizer (approximately 20 strokes). Homogenized brain lysate was transferred to 1.5 mL microcentrifuge tubes and centrifuged at 4°C for 10 minutes at 10,000 rcf. The supernatant from each set of brains was pooled yielding two separately prepared stocks of brain lysate. After BCA protein concentration determination, brain lysate stocks were diluted to 5 ug / uL in lysis buffer.

After IgG conjugation, the IP plate was placed on a magnetic plate, and the supernatant aspirated and discarded into 10% bleach. To each well, 200 µL of brain lysate (5 µg / µL) was added. Plates were sealed with adhesive aluminumized plate covers (Bio-Rad, Microseal® 'F' J o u r n a l P r e -p r o o f 25 PCR Plate Seal, foil, pierceable Cat. No. #MSF1001). Antibody-bead-lysate complexes were incubated for 1 hour at room temperature under constant gentle agitation.

After 1 hour, IP plates were placed on magnetic plates and the lysate was aspirated and discarded into 10% bleach. Beads and their respective immune complexes were washed twice with 180 µL of detergent wash buffer, then once in high salt wash buffer, then once in nondetergent wash buffer, and finally once with ammonium bicarbonate buffer. For each well, washed beads were then resuspended in 35 µL of ammonium bicarbonate buffer to which 1 µL sequencing grade porcine trypsin was added (Promega, Cat. No. V5111). Immune complexes were digested on-bead for 1 hr at 37°C. After digestion, IP plates were placed on magnetic plates and the digestion reaction containing trypsinized peptides was transferred to a protein LoBind Eppendorf tube (Eppendorf, Cat. No. 022431081) and stored at -80°C until liquid chromatography (LC) was performed.

Mass Spectrometry: LC separation was done on a Dionex nano Ultimate 3000 (Thermo Scientific) with a Thermo Easy-Spray source . The digested peptides were reconstituted in 2% acetonitrile /0.1% trifluoroacetic acid and 1 µg in 5 µl of each sample was loaded onto a PepMap 100Å 3U 75 um x 20 mm reverse phase trap where they were desalted online before being separated on a 100 Å 2U 50 micron x 150 mm PepMap EasySpray reverse phase column.

Peptides were eluted using a 60 minute gradient of 0.1% formic acid and 80% acetonitrile with a flow rate of 200nL/min. The separation gradient was ran with 2% to 5% acetonitrile over 1 minutes, 5% to 10% acetonitrile over 7 minutes, 10% to 55% acetonitrile over for 43 minutes, 55% acetonitrile to 99% acetonitrile over 1 minutes, a 4 minute hold at 99% acetonitrile, and finally 99% acetonitrile to 2% acetonitrile held at 2% acetonitrile for 10 minutes. Mass spectra were collected on a Fusion Lumos mass spectrometer (Thermo Fisher Scientific) in a datadependent top speed mode with one MS precursor scan followed by MS/MS spectra for 3 seconds. A dynamic exclusion of 60 seconds was used. MS spectra were acquired with an isolation window of 1. 

The design of our human proteome phage display library 25 containing 731,724 peptides of 49 amino acids has been previously described 27 . Each peptide overlaps with its N-terminal by 25 amino acids. Preparation and titering phage libraries from stocks was as previously described 62 .

For our studies, phage libraries were incubated with 1 µL of patient biofluid (CSF or serum, both diluted 1:1 in 40% glycerol storage buffer) or 0.06µg of anti-GFAP antibody overnight at 4ºC.

IgG and bound phage were then isolated with protein A/G beads and phage were expanded by inoculating E. coli BLT5403 with bead-bound phage as previously described 62 . E. coli lysate from this inoculation was used as the input phage library for a second round of immunoprecipitation using patient biofluids or anti-GFAP. Enrichment of phage DNA, and barcoding of individual IP reactions was performed using a single nested PCR reaction using panOME and multiplexing primers as previously described 62 . PCR products were pooled and bead cleaned (SPRISelect, Beckman Coulter). Resultant libraries were sequenced on the NovaSeq 6000 or the iSeq (Illumina) using 150nt paired-end reads with a 20% PhiX spike in.

HEK293T cells were plated onto 10mm poly-d-lysine coated (50µg/mL) coverslips in 24-well plates. 293 cells were transfected overnight with pEF5-FRT-TagRFP-T-IFT88 using Lipofectamine 3000 (ThermoFisher, Cat# L3000001). The following day, after two rinses with ice cold 1X PBS, RFP-IFT88 transfected cells were fixed with ice-cold methanol for 10 minutes.

The fixed cells were rinsed with PBS, blocked with 5% lamb serum in PBS (blocking buffer), and permeabilized for 30 minutes using with blocking buffer containing 0.5% Triton. RFP-IFT88

HEK293T overexpressing cells were stained overnight using mouse anti-RFP at 1:100, rabbit anti-IFT88 at 1:100, and undiluted CSF. The cells were rinsed with PBS four times, and stained with anti-human 488, anti-mouse 594, and anti-rabbit Cy5 each at a 1:1,000 dilution in 5% blocking buffer. Nuclei were stained with DAPI at 1:2,000 in PBS for 5 minutes. Stained slides were then mounted onto microscope slides with Prolong Gold antifade. IRDye 800CW at 1:5,000 at room temperature for 1 hour. Membrane was rinsed five times in TBST and imaged on a LI-COR Odyssey. The same membrane was then probed with rabbit anti-FLAG antibody at 1:1,000 (Cell Signal, #2368) for 1 hour at room temperature, rinsed five times in TBST, probed with goat-anti-rabbit IgG at for 1 hour at room temperature, rinsed five times in TBST and reimaged on a LI-COR Odyssey.

Tandem mass spectra were extracted by Proteome Explorer v1.4 (Thermo Scientific). Charge state deconvolution and deisotoping were not performed. All MS/MS samples were analyzed identifications were accepted if they could be established at greater than 95.0% probability by the Scaffold Local FDR algorithm. Peptide identifications were also required to exceed specific database search engine thresholds. X! Protein identifications were accepted if they could be established at greater than 5.0% probability to achieve an FDR less than 5.0% and contained at least 2 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm 63 . Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. Proteins sharing significant peptide evidence were grouped into clusters. For identification of significant peptides, Scaffold settings were as such: protein threshold = FDR < 5%, minimum peptides = 1, peptide threshold = MRS_otIt. Mouse proteins that were significantly enriched after Benjamini-Hochberg corrected t-test (alpha = 0.05) and that were observed in both replicates were considered candidate autoantigens.

The Quandenser pipeline was used with its default settings on the raw mass spectrometry files.

The Quandenser pipeline consists of Quandenser, Crux, and Triqler. 31

Raw FASTQ reads generated from the PhIP-Seq peptidome assay were aligned to our reference PhIP-Seq database using RAPSearch (v2.2). Peptide counts outputted from this workflow were normalized to reads per 100 thousand (RPK) for every sample by dividing each peptide count by the sum and multiplying by 100,000. The resulting peptide RPK count matrices were analyzed in R as described below.

For the analysis, these data were divided into disease and reference groups for both CSF and plasma samples. The disease group contained COVID-19 patient samples. The reference group contained healthy control (HC) and A/G bead samples. Peptide fold change (FC) was calculated for each sample. Peptide counts for COVID-19 patient samples were divided by the mean RPK of the reference group, and healthy control samples were divided by the mean RPK of the combined set of COVID-19 samples and A/G beads. In addition, the FC for GFAP samples were calculated in the same way using the mean of all A/G bead samples.

J o u r n a l P r e -p r o o f 29 To identify enriched peptides, results from each sample were filtered using a set of thresholds that, when using a commercial antibody to GFAP, consistently identified GFAP peptides while minimizing nonspecific off-target peptide identification. Each peptide was required to have a minimum of 1 RPK as well as a FC > 10. In addition, thresholds were applied at the gene level. Genes were kept if at least one peptide had a FC > 100 and a total (summed) RPK > 20 across all peptides in the gene. A Kmer analysis was applied to amino acid sequences of all peptides that passed the previous filters. Using a sliding window algorithm, with a window size of 7 amino acids and a step size of 1, all 7-mers were compared across COVID-19 and HC samples. Proteins for which peptides containing at least one 7-mer overlap with another peptide whose total rpK was ≥ 20 were carried forward in the analysis. Additionally, proteins with nonoverlapping peptides with an individual rpK ≥ 20 an FC ≥ 100 were also carried forward.

Proteins that passed these thresholds in both technical replicates but were not enriched by reference samples were considered candidate autoantibodies. This workflow was repeated on a per sample level, and the results for each sample were stored separately.

ToppGene.org was used for gene ontology analyses of IP-MS and PhIP-Seq Data. 

Extrapulmonary manifestations of COVID-19

Neurologic Manifestations of Hospitalized Patients With Coronavirus Disease

Cerebrospinal fluid findings in COVID-19 patients with neurological symptoms

Cerebrospinal fluid features in SARS-CoV-2 RT-PCR positive patients

Patients with COVID-19 and neurological manifestations show undetectable SARS-CoV-2 RNA levels in the cerebrospinal fluid

Systematic SARS-CoV-2 screening in cerebrospinal fluid during the COVID-19 pandemic

Neuropathology of patients with COVID-19 in Germany: a post-mortem case series

Clonally expanded CD8 T cells patrol the cerebrospinal fluid in Alzheimer's disease

Analytical sensitivity and efficiency comparisons of SARS-CoV-2 RT-qPCR primer-probe sets

CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes

Anti-SARS-CoV-2 antibodies in the CSF, blood-brain barrier dysfunction, and neurological outcome: Studies in 8 stuporous and comatose patients

West Nile encephalitis: an emerging disease in the United States

Cerebrospinal fluid findings in adults with acute Lyme neuroborreliosis

The value of detecting anti-VZV IgG antibody in CSF to diagnose VZV vasculopathy

Neuroinvasive potential of SARS-CoV-2 revealed in a human brain organoid model

SARS-CoV-2 neutralizing antibody structures inform therapeutic strategies

Detection Methods for Autoantibodies in Suspected Autoimmune Encephalitis

COVID-19-associated acute necrotising encephalopathy successfully treated with steroids and polyvalent immunoglobulin with unusual IgG targeting the cerebral fibre network

Miller Fisher syndrome and polyneuritis cranialis in COVID-19

Guillain-Barre Syndrome Associated with SARS-CoV-2

Neurological and neuropsychiatric complications of COVID-19 in 153 patients: a UK-wide surveillance study

Autoantigen discovery with a synthetic human peptidome

High-resolution epitope mapping of anti-Hu and anti-Yo autoimmunity by programmable phage display

Emerging evidence of a COVID-19 thrombotic syndrome has treatment implications

Proteomic and Metabolomic Characterization of COVID-19 Patient Sera

Prothrombotic autoantibodies in serum from patients hospitalized with COVID-19

Focus on the spectra that matter by clustering of quantification data in shotgun proteomics

Acute encephalopathy associated with respiratory syncytial virus infections in childhood. A literature review

Clinical manifestations of respiratory syncytial virus-associated encephalopathy in Fukushima

Respiratory syncytial virus-associated encephalopathy complicated by congenital myopathy

Encephalitis associated with influenza B virus infection in 2 children and a review of the literature

Fatal influenza A(H1N1)pdm09 encephalopathy in immunocompetent man

A case of respiratory syncytial virus-associated encephalopathy in which the virus was detected in cerebrospinal fluid and intratracheal aspiration despite negative rapid test results

High Number of Activated CD8+ T Cells Targeting HIV Antigens are Present in Cerebrospinal Fluid in Acute HIV Infection

Cerebrospinal Fluid Treponema pallidum Particle Agglutination Assay for Neurosyphilis Diagnosis

Antigen-specific proliferation of CSF lymphocytes in Lyme disease

Coronavirus Neurovirulence Correlates with the Ability of the Virus To Induce Proinflammatory Cytokine Signals from Astrocytes and Microglia

Autoimmune Encephalitis Presenting with Acute Excited Catatonia in a 40-Year-Old Male Patient with Covid-19

IFT88 mutations identified in individuals with non-syndromic recessive retinal degeneration result in abnormal ciliogenesis

Genotype-phenotype correlations in THAP1 dystonia: Molecular foundations and description of new cases

Autoantibodies against type I IFNs in patients with life-threatening COVID-19

The Immunology of Multisystem Inflammatory Syndrome in Children with COVID-19

Clinically identifiable autoreactivity is common in severe SARS-CoV-2 Infection

Interferome v2.0: an updated database of annotated interferon-regulated genes

Integrated analysis of single cell transcriptomic data across conditions, technologies, and species

Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists

IMGT, the international ImMunoGeneTics database

Single-cell repertoire tracing identifies rituximab-resistant B cells during myasthenia gravis relapses

Migrant memory B cells secrete luminal antibody in the vagina

Mouse model of SARS-CoV-2 reveals inflammatory role of type I interferon signaling

ReScan, a Multiplex Diagnostic Pipeline, Pans Human Sera for SARS-CoV-2 Antigens

Efficient generation of monoclonal antibodies from single human B cells by single cell RT-PCR and expression vector cloning

A serological assay to detect SARS-CoV-2 seroconversion in humans

Performance characteristics of a rapid SARS-CoV-2 antigen detection assay at a public plaza testing site in San Francisco

Genetic Screens Identify Host Factors for SARS-CoV-2 and Common Cold Coronaviruses

Pan-viral serology implicates enteroviruses in acute flaccid myelitis

A statistical model for identifying proteins by tandem mass spectrometry

We are grateful to the participants who volunteered to be a part of this study, Santos Bermejo and Allison Nelson for study assistance, Patrick Wong and Orr-el Weizman for helpful discussion

K08NS096117 (MRW), R01MH122471 (JLD, SJP, MRW), R21MH118109 (SS )

Chan Zuckerberg Biohub (JLD), Brain Research Foundation (SJP), Program for Breakthroughs in Biomedical Research (SJP, MRW)

CMB), the Beatrice Kleinberg Neuwirth Fund (AIK), and Fast Grant funding support from the Emergent Ventures at the Mercatus Center. LC-MS was supported by the NIH shared instrumentation grant S10OD021801. Author contributions: ES devised and executed single cell RNA sequencing and mouse SARS

CoV-2 experiments, analysis and interpretation of data resulting from said assays, and drafted the manuscript. CMB designed and assisted with anatomic immunostaining, immunoprecipitation mass spectrometry (IP-MS), phage display immunoprecipitation sequencing (PhIP-Seq), analysis and interpretation of data resulting from said assays, and wrote and edited the manuscript. RDC, RJ, and SHK assisted in analysis of single cell RNA sequencing. CRZ designed and performed experiments with SARS-CoV-2 Luminex assay. immunostained with mAbs 1 -9 and a representative whole brain sagittal image is shown for PBMC-derived monoclonals (mAb 7) and CSF-derived monoclonals

Subpanels: i) mAb 1 immunostaining of cerebellar Purkinje cells (arrow) and the overlying molecular layer, ii) mAb C2 immunostaining of cortical neuropil and occasional staining of neuron-like soma (arrow), iii) mAb C3 immunostaining of large cells within the hilus of the hippocampus, iv) mAb C4 immunostains of mitral-like cells of the olfactory bulb (arrow), v) mAb C4 immunostaining of pyramidal neurons (arrow

Scale bars = 100 µm. (B) Binary matrix indicating anatomic immunoreactivity of COVID-19 CSF at a 1:10 dilution. (C) Select examples of COVID-19 CSF anatomic immunostaining: of the hippocampus (n = 3, arrows = CA3, left column, scale bar = 100 µm), cerebrovasculature (top panel second column arrow shows endothelial staining, scale bar = 50 µm; bottom panel arrow shows a perivascular cell, scale bar = 10 µm), olfactory bulb (n = 3, two shown, third column, top panel neuron-like cells

(D) Heatmaps of sequence-sharing peptides mapping to IFT88 (Case 1, top) and THAP3

For COVID-19 and CONTROL columns, cell values represent the mean of log 10 (fold change enrichment) of technical replicates. For both Case 1 and Case 3, candidate IFT88 and THAP3 peptides, respectively, were enriched moreso by CSF than plasma. (E) HEK 293 overexpression cell-based assay were performed in technical replicate. A representative example demonstrates that Case 1 CSF is immunoreactive to overexpressed RFP-IFT88 (CSF = green, anti-RFP = red, anti-IFT88 antibody = magenta). Scale bars = 10 µM (F) Western blot anti-THAP3 autoantibodies in CSF of Case 3. Both the CSF IgG (green) and anti-FLAG (red) recognize the same ~25kDa

Immune cell scRNA-seq showed divergent T cell activation in the CNS during COVID-19 • COVID-19 patients had a compartmentalized cytokine response in the CNS • All COVID-19 patients had anti-SARS-CoV-2 antibodies in their CSF • Five out of seven COVID-19 patients had anti-neural autoantibodies in their CSF ETOC blurb: Neurological symptoms are frequent in hospitalized patients with acute COVID-19. Song et al. find that compared to controls, COVID-19 patients with neurologic symptoms have divergent immune responses between the CNS and periphery