Cerebrospinal fluid findings in COVID-19: a multicenter study of 150 lumbar punctures in 127 patients

Background Comprehensive data on the cerebrospinal fluid (CSF) profile in patients with COVID-19 and neurological involvement from large-scale multicenter studies are missing so far. Objective To analyze systematically the CSF profile in COVID-19. Methods Retrospective analysis of 150 lumbar punctures in 127 patients with PCR-proven COVID-19 and neurological symptoms seen at 17 European university centers Results The most frequent pathological finding was blood-CSF barrier (BCB) dysfunction (median QAlb 11.4 [6.72–50.8]), which was present in 58/116 (50%) samples from patients without pre-/coexisting CNS diseases (group I). QAlb remained elevated > 14d (47.6%) and even > 30d (55.6%) after neurological onset. CSF total protein was elevated in 54/118 (45.8%) samples (median 65.35 mg/dl [45.3–240.4]) and strongly correlated with QAlb. The CSF white cell count (WCC) was increased in 14/128 (11%) samples (mostly lympho-monocytic; median 10 cells/µl, > 100 in only 4). An albuminocytological dissociation (ACD) was found in 43/115 (37.4%) samples. CSF l-lactate was increased in 26/109 (24%; median 3.04 mmol/l [2.2–4]). CSF-IgG was elevated in 50/100 (50%), but was of peripheral origin, since QIgG was normal in almost all cases, as were QIgA and QIgM. In 58/103 samples (56%) pattern 4 oligoclonal bands (OCB) compatible with systemic inflammation were present, while CSF-restricted OCB were found in only 2/103 (1.9%). SARS-CoV-2-CSF-PCR was negative in 76/76 samples. Routine CSF findings were normal in 35%. Cytokine levels were frequently elevated in the CSF (often associated with BCB dysfunction) and serum, partly remaining positive at high levels for weeks/months (939 tests). Of note, a positive SARS-CoV-2-IgG-antibody index (AI) was found in 2/19 (10.5%) patients which was associated with unusually high WCC in both of them and a strongly increased interleukin-6 (IL-6) index in one (not tested in the other). Anti-neuronal/anti-glial autoantibodies were mostly absent in the CSF and serum (1509 tests). In samples from patients with pre-/coexisting CNS disorders (group II [N = 19]; including multiple sclerosis, JC-virus-associated immune reconstitution inflammatory syndrome, HSV/VZV encephalitis/meningitis, CNS lymphoma, anti-Yo syndrome, subarachnoid hemorrhage), CSF findings were mostly representative of the respective disease. Conclusions The CSF profile in COVID-19 with neurological symptoms is mainly characterized by BCB disruption in the absence of intrathecal inflammation, compatible with cerebrospinal endotheliopathy. Persistent BCB dysfunction and elevated cytokine levels may contribute to both acute symptoms and ‘long COVID’. Direct infection of the CNS with SARS-CoV-2, if occurring at all, seems to be rare. Broad differential diagnostic considerations are recommended to avoid misinterpretation of treatable coexisting neurological disorders as complications of COVID-19. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-021-02339-0.

Eleven patients (19 CSF samples) were excluded from primary analysis due to pre-/coexisting inflammatory (multiple sclerosis [MS] in 3, acute herpes simplex virus [HSV] type 1 encephalitis [HSVE] in 1, acute varicella zoster virus [VZV] encephalitis in 1, anti-Yo-associated autoimmune encephalitis in 1, chronic viral meningitis of unknown etiology in 1), neoplastic (primary CNS lymphoma in 1, meningeal carcinomatosis in 1), or cerebrovascular (severe subarachnoid hemorrhage [SAH] resulting in severe blood contamination of the CSF sample in 2) CNS conditions known to cause CSF alterations and considered by the treating physicians to be most likely not related to COVID-19; these patients were analyzed separately as 'group II' (f:m ratio = 1:0.8; median age 54 years, range . Of the remaining patients (N = 116; 'group I'), 90 were male (f:m ratio = 1:3.5). Eighty-eight percent were of Caucasian, 5% of African, 4% of Turkish or Middle-Eastern, and 3% of Asian origin. Overall, 131 LPs were performed in these 116 patients following onset of COVID-19 (one sample available from 102 patients, two samples available from 13, and three samples from 1). The median age at LP was 65 years (range . The median time between first SARS-CoV-2-PCR-positive swab and first LP was 12 days (percentile range [0.1-0.9] 0-44), and the median time between LP and onset of the first neurological symptoms that prompted it was 5 days (percentile range . The median time between the first SARS-CoV-2-PCR-positive swab and the onset of neurological symptoms was 7 days. Neurological manifestations at the time of LP included encephalopathy, disturbed consciousness, or delayed wake-up reaction (63 LPs), seizures or epilepsy-like EEG changes (27 LPs), cerebral ischemia or bleeding (14 LPs), myelitis (3 LPs), other CNS manifestations (25 LPs; including cerebellar ataxia, sensorimotor symptoms of unknown cause, and cognitive impairment), peripheral neuropathy (22 LPs; including Guillain-Barré syndrome (GBS) [at least 6 LPs]), cranial nerve symptoms (at least 10 LPs; including anosmia and dysgeusia), and headache (at least 13 LPs; including 3 × isolated severe headache and 1 × headache and nausea) (multiple manifestations per patient possible). The presence of neurological disease was supported by paraclinical evidence from magnetic resonance imaging (MRI) or computed tomography (CT) at the time of LP in 94 cases, by EEG in 35, and by ENG/EMG in 18. At the time of LP, most patients had symptoms attributable to brain or spinal cord involvement (N = 108 [82%] samples; 'brain/spinal cord [B/ SC] subgroup'), while some had exclusively hyposmia, dysgeusia, isolated cranial nerve involvement, peripheral nerve damage, and/or headache (N = 22 samples [17%]; 'peripheral nerve/cranial nerve/headache [PN/ CN/H] subgroup') (no exact classification possible in 1 case). The neurological symptoms present at the time of LP were coarsely classified by the treating neurologists as 'severe' for 57.9% of LPs, 'moderate' for 24.8%, and 'mild' for 17.4%, and samples were stratified accordingly for further analysis; at least 57% of the samples were taken during or within 1 week before or after ICU treatment and 57% during or within 1 week before or after ventilation. Serum IgG antibodies against SARS-CoV-2 were determined by enzyme-linked immunosorbent assay (ELISA) in 58 samples from 56 patients in group I (as part of routine clinical workup) and were positive in 51 (88%) samples from 49 (87.5%) patients (median 28 days since non-neurological disease onset; percentile range [0.1-0.9] 12-45) and negative in 7 samples from 6 patients (median only 5 days, up to 12 days; N = 7); in the remaining patients, either serum antibody testing was not done because of the short time interval between COVID-19 onset and LP or data were not available retrospectively.
For the purpose of this study, samples obtained within 14 days of the onset of a patient's neurological symptoms were classified as 'acute' (75% of all samples in the 'B/ SC subgroup' and 89% in the 'PN/CN/H subgroup'). To avoid bias due to differences in acuity, some subgroup analyses were restricted to 'acute' samples ('acute B/SC subgroup' , 'acute PN/CN/H subgroup'). The study was approved by the review boards of the participating centers and patients or their legal representatives gave written informed consent. LPs were performed for diagnostic purposes in all cases; no samples were obtained for use in this study.
Due to the retrospective nature of this study, not all data of interest were available for all patients and samples. Accordingly, absolute patient numbers and/or sample numbers differed among subanalyses (e.g., CSF white cell counts were determined in more patients and samples than anti-neural autoantibodies). No data were excluded from analysis unless explicitly stated (e.g., for not meeting subgroup criteria).

Methods
Methods were adopted from our previous studies on CSF findings in inflammatory CNS disorders [13][14][15][16][17][18] adhering to the German Guidelines on CSF diagnostics of the German Society for CSF Diagnostics and Clinical Neurochemistry and the German Society of Neurology [19][20][21].

Antibody indices
The intrathecal synthesis of antibodies was detected by calculation of the corresponding anti-microbial A CSF WCC > 5/µl was classified as 'increased' [20]. An age-dependent reference range for CSF l-lactate was employed (16-50: 2.1 mmol/l, > 50: 2.6 mmol/l) [20]. The upper reference limit for CSF total protein was set at 0.45 mg/ l [20].

Statistics
Samples were analyzed in total as well as after stratification according to disease status and treatment status. Fisher's exact test, Mann-Whitney U test and Kruskal-Wallis test were used to assess the statistical significance of differences between groups. Spearman's rho was used to assess correlations. Due to the exploratory nature of this study, no correction for multiple testing was applied other than Dunn's post-test. P values < 0.05 were considered statistically significant. Reiber diagrams ('reibergrams') [21,24] were generated using CSF Research Tool v3.0 (CoMed GmbH, Soest, Germany) and FLC-K Statistics v1.02 (Albaum IT Solutions, Möhnesee, Germany).

Blood-CSF barrier function
An elevated CSF/serum ratio for albumin (QAlb), indicating dysfunction of the BCB, was found in 58/116 (50%) samples with available data in group I and was present at least once in 54/106 (50.9%) patients tested for this marker in group I. In those patients with BCB dysfunction, QAlb ranged from 6.72-50.8 × 10 -3 (median 11.4) ( Table 1 and Fig. 1A, B).
The frequency of BCB dysfunction was slightly higher in the 'acute B/SC subgroup' (56.9% [37/65] 10.9 [N = 9] if only those with elevated QAlb were considered), but the differences did not reach statistical significance (Table 1 and Fig. 1A, B).
Although QAlb values decreased over time, both in the total cohort (r = − 0.201, p < 0.05; r 2 = 0.041, p < 0.05; N = 103 samples with available data, from 95 patients) and in the B/SC subgroup (r = − 0.278, p < 0.03; r 2 = 0.059, p < 0.03; N = 86 samples with available data, from 78 patients) ( Fig. 2A) Table 1), including in 5/9 (55.6%) samples from 9 patients in whom neurological symptoms had started > 30 days before the respective LP (delirium, disturbed consciousness or delay in recovery of consciousness after ventilation in 4, with signs of epileptic activity in 2 and evidence of brain infarction in 1; ataxia, paresis, dysesthesia and urinary retention of unknown cause in 1).
Median serum albumin concentrations did not differ significantly between samples from patients with or without elevated QAlb values (28.35 (Table 1), which is in line with the fact that QAlb remained elevated > 14 days in several cases as well. The frequency of CSF TP elevation did not differ between the acute 'PN/CN/H subgroup' and the acute 'B/SC subgroup' , and median CSF TP levels did not differ between the two subgroups either; Table 1). As QAlb, CSF TP levels were negatively correlated with the time since onset of the neurological symptoms both in the total cohort (r = − 0.216, p < 0.03; r 2 = 0.047, p < 0.03; N = 101 samples with available data, from 90 patients) and in the 'B/ SC subgroup' (r = − 0.257, p < 0.02; r 2 = 0.066, p < 0.02; N = 88 samples with available data, from 77 patients) (Fig. 2B).

Cellular immune response
Only 14/128 (10.9%) samples from 12 different patients exhibited an increased CSF WCC, with a median of 10 cells/µl (Table 2 and Fig. 1D) (WCC not determined in 3 samples). Only 6 LPs from 5 patients yielded WCC counts > 10 cells/µl. In three of the 14 samples with Table 1 Blood-CSF barrier function, CSF and serum albumin, CSF total protein, and CSF L-lactate in group I Results are given as medians (with ranges and sample or patient numbers in brackets) and frequencies (with percentages in brackets), respectively. Note that columns 6 and 7 in Tables 1, 2  Although some parameters were more markedly or more frequently altered in the 'B/SC subgroup' than in patients from the 'PN/CN/H subgroup' , the differences were not statistically significant. N indicates the number of samples tested. Note that the figure shows all samples with available data; data stratified according to disease duration at the time of LP can be found in Tables 1, 2 pleocytosis, blood contamination was noted (1500, 5000, and 15,500 erythrocytes/µl, respectively). Although WCC are routinely corrected for blood contamination, the rules applied (e.g., reduction by 1 leukocyte/1000 erythrocytes) are rather crude, which renders it possible that the WCC were nevertheless falsely elevated in these few cases, all the more as the reported WCCs were low. If these samples are excluded, pleocytosis was present in only 11/128 (8.6%) samples with available data. The median WCC was higher in the 'B/SC subgroup' than in the 'PN/CN/H subgroup' , but the difference did not reach statistical significance (Table 2 and Fig. 1D).
Lymphocytes were the predominant cell type in patients with pleocytosis and were present in 10/10 samples with available cytological data (accounting for up to 99% of all cells), monocytes in 9/10, activated lymphocytes in 3/10, neutrophils in 3/10 and plasma cells in 1/10; eosinophils and basophils were absent in all. Blood contamination was excluded in 9/10 (no data available in 1), including in the few with neutrophil pleocytosis.
The WCC exceeded 100 cells/µl in only 4 samples from 3 patients. The diagnoses in these cases included "fever, acute aphasia, apathy" (651 cells/µl at first LP and 373 cells/µl 7 days later; lymphomonocytic; CSF l-lactate 2.8 and 2.6 mmol/l; glucose CSF/serum ratio 63%; massive Correlation analyses for QAlb (A) and CSF total protein (B), respectively, and days since onset of the neurological symptoms in the total cohort and in the B/SC subgroup. Although a mildly significant correlation was found for both parameters, it should be noted that both parameters were still pathologically altered > 14 and even > 30 days after onset of the neurological symptoms in a subset of cases (see Results section and Table 1 for details). B/SC brain/spinal cord, N number of samples; QAlb albumin CSF/serum ratio, TP total protein BCB dysfunction; LP performed 5 and 12 days after onset of non-neurological symptoms and 1 and 8 days after onset of neurological symptoms; almost complete recovery); "meningitis, fever, and moderate headache" (510 cells/µl; 68% neutrophils; CSF l-lactate 3.3 mmol/l; glucose CSF serum ratio 53%; BCB dysfunction; LP performed around 26 days after onset of non-neurological and 5 days after onset of neurological symptoms; full recovery); and "headache, meningism, cognitive impairment, later motor aphasia, flexion synergy" (9 cells/µl 3 days and 247 cells/µl 5 days after onset of the neurological symptoms; 85 and 92% lymphocytes, respectively, 12 and 4% neutrophils, 2 and 3% monocytes, and 3% plasma cells and 1% activated lymphocytes at second LP; full recovery).
While WCC > 100 occurred only in the 'B/SC subgroup' , median WCCs did not differ between the acute 'B/ SC subgroup' (2 cells/µl) and the acute 'PN/CN/H subgroup' (2 cells/µl), reflecting the low overall frequency of pleocytosis in both subgroups. The pleocytosis rate was slightly higher in the former group but the difference did not reach statistical significance. If ACD is defined with reference to albumin instead of TP, 47 samples with elevated QAlb (from 47 different patients) of 58 tested (81%) showed an ACD (including 9/10 samples tested from the 'PN/CN/H subgroup' and 38/48 (79.2%) samples from the 'B/SC subgroup'). ACD was associated with stroke in only five of these patients and in 2 with (micro)hemorrhages. If not only samples with BCB dysfunction but all samples with available data are considered, ACD was present in 47/113 (41.6%) samples.

CSF l-lactate
Of note, CSF l-lactate levels were increased in 26/109 (23.9%) CSF samples tested (and at least once in 25/97 [25.8%] patients tested), with a median concentration of 3.04 mmol/l (range 2.2-4) (Fig. 1F). Elevated l-lactate levels were found more frequently in samples taken during the first 14 days after onset of neurological symptoms than in samples taken later (27%, N = 74 vs. 14.3%, N = 21); however, the difference did not reach statistical significance ( Table 1).
Of those samples with elevated CSF l-lactate, the WCC was elevated in only 5 (and neutrophils in none), and the frequency of CSF l-lactate elevation did not differ significantly between samples with and without neutrophil granulocytes (40% [4/10] vs. 35% [13/37]), widely ruling out neutrophil pleocytosis as a major cause of CSF l-lactate elevation in our patients. However, a weak yet significant positive correlation of l-lactate with the CSF WCC was found (r = 0.199, p < 0.04; r 2 = 0.04, p < 0.04; N = 109 samples with available data, from 97 patients). l-Lactate levels were also significantly higher in patients who required ventilation at the time of LP or within 1 week before or after LP (median 2.2 vs. 1.75 mmol/l; p < 0.00002). Finally, CSF l-lactate levels correlated with QAlb (r = 0.307, p < 0.003; r 2 = 0.094, p < 0.003; N = 96 samples with available data, from 87 patients) and, accordingly, also with CSF total protein (r = 0.237, p < 0.02; r 2 = 0.056, p < 0.02; N = 106 samples with available data, from 95 patients). The correlation of L-lactate with QAlb was much stronger after exclusion of samples with unusually high QAlb values ≥ 30 × 10 -3 (r 2 = 0.24; p < 0.0001; N = 91 samples with available data, from 85 patients) (Fig. 3B). CSF l-lactate levels were also significantly higher (p < 0.002) and CSF l-lactate elevation significantly more frequent (76.5 vs. 17%; p < 0.00001) in patients with elevated QAlb than in those with normal QAlb.

Intrathecal IgG synthesis
In 58/103 (56.3%) samples of group I tested, identical oligoclonal IgG bands in serum and CSF with no additional CSF-restricted bands (the so-called 'mirror pattern' or 'OCB pattern 4' [21,22,35]) were present, a pattern thought to reflect extrathecal immune activation and passive diffusion of peripheral oligoclonal IgG from the serum into the CSF. By contrast, CSF-restricted OCB, indicative of intrathecal IgG synthesis [21,22,35], were positive in only 2/103 (1.9%) samples (pattern 3 in both, pattern 2 in none) or 2/96 (2.1%) patients. QIgG, another, albeit less sensitive, marker of intrathecal IgG synthesis, was elevated (8.3; Q lim [IgG] 5.6) in just 1/115 (1%) samples or 1/106 (0.9%) patients tested, namely in a patient with CSF-restricted OCB and multiple brain infarctions (Table 3 and Fig. 1G). In this patient, the intrathecal IgG fraction was 32%, corroborating the positive OCB result, which corresponded to an absolute amount of intrathecally produced IgG of 38 mg/l. In the second patient with CSF-restricted OCB, QIgG was particularly low (1.86) and did not exceed Q lim (IgG). However, low amounts of intrathecal IgG may in fact be detectable only by isoelectric focusing but not by QIgG determination; of note this patient was one of only two with elevated QIgM (see below). In the first patient, no pre-or coexisting condition known to cause OCB positivity has been detected as of last follow-up; the second patient had been previously diagnosed with mild dementia of unknown cause. Pattern 5, indicating monoclonal gammopathy, was present in 1/103 samples (1%), obtained from a patient with punctate multiple ischemia on brain MRI, bilateral visual impairment, decreased vigilance, and capillary leak syndrome. Altogether, intrathecal IgG synthesis was very rare and even below the rate reported in healthy central European volunteers [36]. Unfortunately, CSF SARS-CoV-2-IgG was determined in none of the 2 patients with CSF-restricted OCB; a SARS-CoV-2 CSF PCR was performed in one and was negative. In the only patient with pattern 5 OCB, no SARS-CoV-2-IgG was found in the CSF and SARS-CoV-2 CSF PCR was negative, suggesting that the monoclonal band observed in this case was likely unrelated to SARS-CoV-2 infection.
The frequency of pattern 4 ('mirror pattern' , i.e. identical OCB in serum and CSF) [21,22,35] was higher in samples obtained > 14 days after onset of neurological symptoms (71.4%) than in samples obtained earlier, possibly reflecting the increase in systemic SARS-CoV-2 antibody production as well as general immune activation during the first few weeks after infection. In contrast, the proportion of patients with pattern 1 (no OCB), declined over time (≤14 days: 46.5%, N = 71; > 14 days: 28.6%, N = 21), although the difference did not reach statistical significance.
Of note, CSF IgG concentrations exceeded the upper reference limit of 40 mg/l in 50/100 (50%) OCB-negative samples tested (Table 3). However, this must not be mistaken for evidence of an intrathecal immune response in COVID-19. In none of these cases, QIgG was elevated, indicating that the increased CSF concentrations were caused by passive transfer of IgG rather and not by intrathecal synthesis. Indeed, QAlb, indicating a leaky BCB, was increased in 80% (40/50) of these patients, but in only 20% (10/50) of those with normal CSF IgG values and available data. Moreover, CSF IgG values were strongly dependent on QAlb, both among patients with elevated QAlb (r = 0.863, p < 0.00001; r 2 = 0.745, p < 0.00001; N = 57 samples with available data, from 54 patients) and in the total cohort (r = 0.871, p < 0.00001; r 2 = 0.759, p < 0.00001; N = 115 samples with available data, from 106 patients) (Fig. 3C, D). QAlb was predictive of CSF IgG in those exceeding the reference limit of 40 mg/l (r = 0.83, p < 0.00001; r 2 = 0.689, p < 0.00001; N = 58 samples with available data, from 53 patients) but serum IgG levels were not, suggesting that elevated CSF IgG levels were mainly driven by BCB dysfunction.
Neither CSF IgG levels ( Serum IgG levels were elevated (> 16 g/l) in 12 samples (10%; from 11 patients) out of 120 samples tested. Six of these patients were tested for serum SARS-CoV-2-IgG and 5/6 were positive. Median serum IgG concentrations did not differ significantly between acute samples (≤ 14 days after onset of neurological symptoms) and samples obtained later (Table 3) nor between the 'B/ SC subgroup' and the 'PN/CN/H subgroup' (Table 3 and Fig. 4A). Median serum IgG levels were higher in samples from patients with positive serum SARS-CoV-2-IgG than in the few samples from SARS-CoV-2-IgGseronegative patients (11.25 g/l [N = 46] vs. 8.30 g/l [N = 7]); however, the difference did not reach statistical significance.

Intrathecal IgA synthesis
QIgA was increased in just 1/101 (1%) samples tested (Table 4 and Fig. 1H), obtained from a patient with disturbed consciousness (QIgA = 6.6). However, the intrathecal fraction (3.2%) and the absolute amount of intrathecally produced IgA (0.8 mg/l) were very low in this case, leaving the possibility of a false-positive result.
Notably, IgA serum levels were elevated (> 4 g/l) in 31 samples (29%; from 29 patients) out of 106 samples tested in group I, ranging between 4.1 and 7.3 g/l (median 5). Of these samples, 8 were tested for SARS-CoV-2-IgA and 6/8 were positive. Median CSF and serum IgA concentrations and CSF/serum ratios did not differ significantly between samples obtained ≤ 14 days since onset of the neurological symptoms and samples obtained later (Table 4) nor between the 'B/SC subgroup' and the 'PN/CN/H subgroup' (Table 4 and Fig. 4C), although median values were all slightly higher in the former subgroup.

Immunoglobulin (Ig) class patterns
None out of 99 (0%) samples from 90 SARS-CoV-2-positive patients tested in group I exhibited the three-classreaction (as defined by elevation of QIgG, QIgM and QIgA) or two-class-reaction (defined by either positive QIgG and QIgM, positive QIgM and QIgA, or positive QIgG and QIgA), based on QIg > Q lim (Ig) ( Table 5), often seen in viral and bacterial infections of the CNS. Intrathecal Ig synthesis was restricted to one immunoglobulin class in 6/99 (6.1%) samples with available data (IgG in 1; IgM in 4; IgA in 1) from 5 patients based on Ig CSF/serum ratios and in 5/99 (5.1%) samples based on Ig-IF > 10% (Table 5).
In one (IgM-IF 32%) out of 5 patients with intrathecal IgM and/or IgA synthesis but no quantitative evidence of intrathecal IgG synthesis, at least qualitative evidence for intrathecal IgG synthesis (i.e., CSF-restricted IgG OCB) was detectable.

SARS-CoV-2 antibody indices
In total, 70 individual SARS-CoV-2-IgG, -IgM or -IgA AI tests were performed in group I. This included 46 individual SARS-CoV-2-IgG AI determinations (13 × AIs for nucleocapsid [N]; 18 × spike protein subunit 1 [S1]; 13 × spike protein subunit 2 [S2]; 1 × mixture of N, S1 and S2; 1 × antigens not documented) in 20 samples from 19 patients. Remarkably, only 1 sample yielded an unequivocally positive result (AI > 1.5), indicating that COVID-19 with neurological involvement is not associated with substantial intrathecal IgG synthesis against SARS-CoV-2 in the majority of cases, at least not in the acute stage shortly after onset of the neurological symptoms (the median interval for all patients tested was 5 days; range 1-49), i.e., at the time when LP is mostly performed.

Table 5 Immunoglobulin class response patterns (ICRPs) in group I
ICRPs have been shown to be of differential diagnostic relevance in various neurological disorders [13,14,19,20,35]  In a further patient-one of only two other patients in this cohort with a WCC > 50 (510 cells/µl) -, the SARS-CoV-2-N-IgG AI was 1.35. According to current guidelines [35], AI values > 1.3 should be considered positive if substantially higher than other AIs in the same sample, based on the assumption that immunoglobulin molecules of the same isotype should pass the BBB at the same rate, irrespective of their epitope specificity. Thus, based on the marked difference between the observed IgG-AI of 1.35 for SARS-CoV-2 and the very low VZV-IgG-AI (< 0.5) found in the same paired CSF/serum samples, intrathecal synthesis to SARS-CoV-2 likely occurred also in this case [35]. The SARS-CoV-2-IgA AI and the SARS-CoV-2-IgM AI were not tested in this patient.
SARS-CoV-2-IgM AI and SARS-CoV-2-IgA AI were tested in three additional patients (all with a normal SARS-CoV-2-IgG AI) but were elevated in none (Table 6).
Importantly, the median interval between onset of the neurological symptoms and AI testing was not longer in the AI-positive subgroup (4 days; range 4-4) than in the AI-negative subgroup (5.5 days; range 1-49).
As a limitation, different assays were used to determine SARS-CoV-2 AI at different centers. While in-house assays were used by one center (University Hospital Zurich, Switzerland) [38], commercial assays manufactured by Euroimmun (Lübeck, Germany) and Generic Assays (Berlin, Germany) were adapted and used for CSF analysis by others. No officially approved SARS-CoV-2-AI assays are currently available in Europe.

CSF SARS-CoV-2-IgG
Absolute SARS-CoV-2-IgG concentrations in the CSF exceeded the assay-specific upper reference levels in 20/29 (69%) samples tested in group I (Table 6), including the two SARS-CoV-2-IgG-AI-positive samples. 13 samples were tested separately for CSF IgG against N, S1 and S2; the 8 positives among these samples all reacted against all three antigens (and the 5fff negative samples were negative for SARS-CoV-2-IgG against all three antigens).

SARS-CoV-2 CSF PCR
Of particular note, CSF PCR for SARS-CoV-2 was negative in 76/76 (100%) samples from 75 patients with neurological symptoms tested in group I, including those with an increased SARS-CoV-2-IgG (and -IgM) AI, all of those with a WCC > 50, and 63/63 from patients with B/ SC involvement (with neurological symptoms classified as "severe" in 41 of them). The median time between neurological onset and PCR was just 5 days (percentile range [0.1-0.9] 1-25). However, 26 samples were taken 7 or more days and 15 of these 14 or more days after neurological onset. PCR was negative in 69 initial CSF samples and 6 CSF samples obtained at follow-up LP (in 6 different patients). CNS tissue PCR from biopsy or autopsy samples was performed in none of the patients.

CSF PCR for other viruses
Results from 113 further PCR tests in 32 samples from 30 patients were available for analysis (59 × HSV, 20 × CMV, 6 × EBV, 5 × HHV6, and 2 × panel PCR neurotropic viruses) ( Table 6). In one patient, viral metagenome analyses were weakly positive for pegivirus. In another one, CSF VZV PCR was low positive with 150 copies/ml, but no diagnosis of VZV encephalitis was made (6 cells/µl; final diagnosis: GBS). All other tests were negative.

MRZ reaction
Measles , which is detectable in around 63% of cases in MS [39], was absent in the only patient with available data (Table 6).

Other antibody indices
A total of 76 further AI tests were conducted in 40 samples in group I. A positive IgG AI (1.7; cut-off 1.5) for herpes simplex virus (HSV) was found in 1/22 (4.5%) samples tested. While symptoms were generally compatible with HSVE in this patient (acute disorientation, personality changes, encephalopathy), HSV-PCR was negative and no CSF pleocytosis present; in consequence, no diagnosis of HSVE was made by the then treating physicians. SARS-CoV-2-AI testing was not done in this case. None of the patients tested had a positive IgG-AI for Borrelia burgdorferi (BB), CMV, or EBV; one patient exhibited a borderline positive (1.6; cut-off 1.5) IgM-AIfor BB (Table 6).

Anti-neuronal and anti-glial antibodies
Over the past few decades, a multitude of anti-neural autoantibodies have been identified in patients with autoimmune encephalitis, myelitis, or polyneuropathy, some of which are considered to be directly pathogenic, and a paraor postinfectious etiology has been suggested for some of these reactivities [40,41]. Not all centers routinely test for anti-neural antibodies in patients with neurological symptoms of uncertain cause, and antibody panels differ among centers. In this cohort, 54  However, in the vast majority of cases, no anti-neuronal, anti-glial or anti-skeletal antibodies were found (Table 7), except for not further characterized "anti-myelin antibodies" in 6/40 serum samples, as detected using peripheral nerve tissue sections. Anti-myelin antibodies, as detected by IIF, are a relatively frequent finding, and may occur even in control subjects; the clinical significance in the patients studied here is unclear; in one case, a follow-up sample was available and was negative. 0/34 CSF samples were positive for myelin antibodies. MOG antibodies were negative in the only 3 serum samples and the only CSF sample tested.
One patient, who had presented with encephalopathy and seizure, was positive for GAD65 serum antibodies, but the evidence was considered insufficient to make a formal diagnosis of autoimmune encephalitis by the treating physicians at last follow-up (no pleocytosis, no intrathecal IgG synthesis, CSF GAD65-IgG only weakly positive). A further patient tested positive for serum NMDAR-IgG antibodies; the patient's CSF was not tested for NMDAR-IgG. While a diagnosis of NMDAR encephalitis cannot be excluded based on the symptoms documented in this case (downbeat nystagmus, orofacial myoclonus, impaired consciousness and delirium, in addition to hyposmia and dysgeusia), no such diagnosis was made by the treating physicians who favored a diagnosis of hypoxic brain damage. Finally, antibodies to skeletal muscle cells of unknown specificity were found in 1 further patient. All remaining 1499 antibody tests were negative.
The median time between onset of the neurological or non-neurological symptoms and the date of LP in those tested for autoantibodies was 6 days (percentile range [0.1-0.9] 1-27.4) and 18 days (percentile range 4-46), respectively.

Free kappa and lambda light chains
Free kappa light chains (FLC-kappa) and free lambda light chains (FLC-lambda) (i.e., immunoglobulin light chains not bound to heavy chains), which are produced by B cells during antibody synthesis in excess and subsequently secreted, are determined by some laboratories as a supportive quantitative marker of (especially intrathecal) IgG synthesis. FLC-kappa, and FLC-lambda [21,44,45] were determined in 6 patients of group 1 in the serum and also in a matched CSF sample in 4 of them.

FLC-lambda/serum FLC-lambda]/[CSF albumin/serum albumin]) was normal (cut-off 5.51) in three patients.
In the fourth patient, serum and CSF FLC-lambda concentrations did not exceed the upper reference limit, but the L index was elevated (10.61; cut-off 5.51). However, this patient had developed peripheral neuropathy during COVID-19 with signs neither of CNS disease nor of intrathecal total IgG synthesis (normal QIgG, no CSFrestricted OCB) and QAlb was unusually low, rendering a false-positive result at least conceivable.
To take into account a possible effect of BCB dysfunction, we also calculated the IL-6 index (= IL-6 ratio/ albumin ratio). Here, the marked difference between the SARS-CoV-2-IgG-AI-positive sample with pleocytosis (247 cells/µl) and the SARS-CoV-2-IgG-AI-negative patients with a normal or almost normal WCC was still present (IL-6 index 0.5 vs. median index of 0.03 in the remainder) (Fig. 6).
In most patients, blood IL-6 levels were repeatedly determined. Overall, 748 samples from 63 patients were tested within a median period of 26.5 days (percentile range [0.1-0.9] 8.3-64.8) around the date of the LP. In total, 723 (96.7%) of these samples exhibited elevated IL-6 levels (median IL-6 concentration 54.45 pg/ml; percentile range 14.99-364) (Fig. 7). In 59/63 (93.7%) patients, IL-6 concentrations were elevated at least once. Of particular note, blood IL-6 levels remained continuously elevated over weeks and months (see Fig. 8 for exemplary data). Serum IL-6 was still elevated at last follow-up (median 28 pg/ml) in 18/19 patients with repeat IL-6 measurements and a follow-up of > 30 days (range 33-80) since first IL-6 measurement and in 12/12 of those with a follow-up of > 45 days (median 33 pg/ml; range 7.2-463) since first measurement, indicating persisting systemic inflammation. From 6 patients follow-up samples obtained over a period of > 60 days were available; serum IL-6 was still elevated in all of these at last follow-up (median 33 pg/ml; range 7.2-138).
Notably, IL-6 was also detectable at high levels in the aqueous humor in the only patient tested because of suspected viral retinitis (1183 pg/ml).

Interleukin-8
CSF interleukin-8 (IL-8) was elevated in 7/7 (100%; 6 × B/ SC subgroup) CSF samples from 7 patients in group I, based on a cut-off of 40 pg/ml derived from a German cohort of patients with normal pressure hydrocephalus (21.40 ± 7.96 pg/ml) [26] (which is in good accordance with the upper reference range found in a recent   Relevant pleocytosis was present in none of the samples (median 1 cell/µl; range 0-6). Regrettably, none of the 4 samples with high WCC > 100/µl (including all patients with a positive SARS-CoV-2-IgG AI) was tested for CSF IL-8.
In 6/7 (86%) samples with elevated CSF IL-8 levels and available data also increased CSF IL-6 levels were found.
The CSF sample with the highest TNF-alpha level (10.37 pg/ml) came from the abovementioned patient with an unusually high WCC (247 cells/µl), an unusually high IL-6 ratio and index, an elevated SARS-CoV-2-IgG AI, compatible with possible direct CNS infection with SARS-CoV-2, and severe clinical CNS involvement. The corresponding serum sample was positive for TNF-alpha at a concentration of 9.4 pg/ml, resulting in a CSF/serum ratio of 1.14. By contrast, the TNF-alpha CSF/serum ratio was markedly below 1 and thus at least 10 times lower in all (N = 7) further cases with available data (median 0.05; range 0.02-0.29), 6 of them without pleocytosis, and also not beyond the mean TNF-alpha CSF/serum ratio (0.29) reported in patients with low backpain [30].
A marked difference was also found for the TNF-alpha index (= TNF-alpha ratio/albumin ratio), which was 0.032 in the SARS-CoV-2-IgG-AI-positive sample with marked pleocytosis and a positive IL-6 index and 0.0038 (median; range 0.0015-0.008) in the remainder with normal or almost normal WCC (Fig. 6).
Overall, serum or plasma TNF-alpha was determined in 63 samples from 26

Interferon-gamma
Interferon-gamma (IFN-gamma) was detectable in only 1 CSF sample (from the B/SC subgroup) out of 7 samples from 7 patients tested (6 × B/SC subgroup), at a concentration of 2.6 pg/ml. While a generally accepted reference range for CSF IFN-gamma has not been defined, that level is lower than the mean CSF IFN-gamma concentrations in a recent cohort of control patients with noninflammatory neurological diseases (mean 7.99 ± 2.75 pg/ ml) [27], in patients with vertebrogenic low backpain in another cohort (49.93 ± 30.28 pg/ml) [30], and in a cohort of "symptomatic controls" in a further study (3.09 ± 10.64 pg/ml) [34]. Serum IFN-gamma was normal in 7/7 (100%; 6 × B/SC subgroup) serum samples from 7 patients (median 0 pg/ml; range 0-8.9; cut-off 15).

Other soluble markers and combined results
CSF concentration of the chemokine CXCL-13 (C-X-C motif chemokine ligand 13) was tested in a single patient in group I (with cranial nerve and peripheral nerve involvement) and was below the detection limit (< 4 pg/ ml). By contrast, CXCL-13 CSF levels were extremely high (38,776 pg/ml) in a patient from group II; however, this patient suffered from primary CNS lymphoma at the time he acquired COVID-19. Serum concentrations of the soluble interleukin-2 receptor (sIL2R), a marker of active cellular immunity, was determined in 4 patients in group I and was elevated in all of them; in total, 13/13 (100%) samples were positive (median 3130 IU/ml; percentile range 1466.6-8894; cut-off 710 IU/ml). Relatively high levels were also found in the aqueous humor in one patient (1068 IU/ml), in whom CSF was not tested. Serum neopterin was tested in two patients in group I and was elevated in both (5.04 and 6.48 ng/ml, respectively; normal < 2.5 ng/ml). Interleukin-17 (IL-17), an important mediator of the mucosal immune response, was not directly assessed in CSF or serum, but IL-17 secretion after whole blood stimulation with Staphylococcus aureus enterotoxin B (SEB) for 24 h, a marker of lymphocyte function/immunodeficiency, was assessed in 10 samples from 10 patients in group I. Interestingly, decreased IL-17 secretion was found in 4 samples from 4 patients (normal values in the remainder), none of whom was treated with immunosuppressants or steroids at the time of testing. In 9 patients, blood levels of 3 cytokines and cytokine receptors (from the following panel: IL-6, IL-8, TNFalpha, IL-1b, sIL2R, IL-10, IL1RA) were assessed; in 8 of these, increased levels of all three were observed at least once and in 1 patient increased levels of two markers. In another 11 patients, blood levels of 5 markers from the same panel were determined; in 4 of these, levels of all 5 markers were elevated; in 5 samples those of four markers, and in 2 samples those of at least two markers. In a further patient, 6 of the beforementioned cytokine markers were assessed in the blood and all were elevated.

First vs. follow-up LP
The frequency of OCBs, pleocytosis, BCB dysfunction, CSF TP elevation, and CSF l-lactate elevation did not differ significantly between the first LP and follow-up LP (median time interval between punctures was 13 days) in group I (Table 8). All but one of the follow-up samples were obtained from patients in the B/SC subgroup. The proportions of patients with severe, moderate, or mild disease, respectively, among those with repeat LP did not differ significantly from those among the remaining patients (severe: 54% vs. 60%, moderate: 31% vs. 22%, severe or moderate: 85 vs. 82%; mild: 15 vs. 18%).
A total of 13 repeat tests for OCB were performed in 10 patients. No change in OCB status or OCB pattern (6 × pattern 4, 2 × pattern 1) over time was noted. Similarly, 8/8 (100%) patients who were tested more than once had a normal IgG CSF/serum ratio both at first LP and at follow-up. An intrathecal 'IgM to IgG switch' was observed in 0/8 patients in whom QIgM and QIgG were determined more than once (median period since first LP: 14 days, range 2-33).

Disease severity
In patients classified as having 'severe' or 'moderate' neurological disease at the time of LP by the treating physicians, many CSF parameters were more markedly or more frequently altered than in patients classified as having mild disease, including the frequency of OCB pattern 4 (as opposed to OCB pattern 1, i.e., a lack of OCB, which was more common in 'mild' disease), median QIgG and QAlb levels, the frequency of BCB dysfunction, median TP levels in those with elevated TP, the frequency of TP values > 100 mg/dl, median l-lactate levels in those with and those without elevated l-lactate, and the frequency of samples with elevated l-lactate (Table 9). As pleocytosis was generally rare and, if present at all, mild in the vast majority of cases, no significant difference in WCC was found between the two groups.

Impact of immunotherapy
Several CSF parameters were either pathologically altered only in samples from patients untreated at the time of LP (N = 89) or more severely pathologically altered in these samples than in samples from patients treated with steroids, immunosuppressants or immunomodulatory drugs at the time of LP (N = 29) in group I (e.g., pleocytosis: 13/89 [ ; higher CSF and serum IgG, IgM, IgA concentrations; higher QAlb values in those with BCB dysfunction; higher TP and l-lactate concentrations in those with elevated values). However, the differences did not reach statistical significance. The steroids used included methylprednisolone, prednisone and dexamethasone (N = 26); the immunosuppressive or immunomodulatory treatments used included intravenous immunoglobulins, sulfasalazine, tacrolimus, mycophenolate mofetil, and ciclosporin A and were given either alone (N = 3) or in combination with steroids (N = 3).
Of note, BCB dysfunction, the most frequent pathological finding in this cohort, as well as related alterations (ACD rate, rate of samples with elevated TP, samples with TP concentrations > 100 mg/dl) were found with equal or even higher frequency in samples from treated patients versus untreated patients. This could partly reflect the fact that the subgroup of treated patients included more patients with severe disease and/or a delayed or insufficient effect of the treatments applied on these parameters. Of note, WCC was rarely elevated in general in this cohort, including in untreated patients, which would in any case render it highly difficult to detect treatment effects on the rate of samples with ACD. To explain the apparent lack of effect on BCB function and related findings, detailed analysis would be required of all individual cases taking account of steroid timing, dosage, and the severity of BCB dysfunction, which was not part of the study protocol.

'Normal' CSF
A substantial number of CSF samples in group I exhibited no pathological changes. If CSF WCC, OCB, QIgG, Link index, QIgM, QIgA, QAlb, CSF TP, and CSF L-lactate are taken into account, 28 (35%) of 80 samples in group I in which all of these 9 parameters were assessed showed exclusively normal values. Of these 28 samples, 17 were obtained ≤14 d and 7 > 14 days after onset of neurological symptoms (no data in 4). If only a basic panel consisting of CSF WCC, CSF TP, and CSF L-lactate is considered (reflecting clinical practice in some non-tertiary centers and in emergency

Quotient diagrams ('reibergrams')
Plots of QIgG, QIgA, and QIgM [21,23], respectively, against QAlb (as a measure of BCB function), visually indicating absence of intrathecal IgG, IgM, and IgA synthesis in most cases (as evident from the fact that most quotients are below the upper hyperbolic discrimination line, Q lim , of the respective immunoglobulin class), are shown for group I in Fig. 9 (see figure legend for details). Note that reibergrams, as a limitation, permit visualizing the frequency of BCB dysfunction only for individual patients or for groups of patients of the same age, but not at population level, since the upper reference limit of QAlb is age-dependent; in the plots shown in Fig. 9, the gray vertical lines indicate the area above the upper limit of QAlb for the median age in group I, which was 65 years.

CSF findings in group II
CSF findings in patients with pre-or co-existing CNS disorders did not substantially differ from those typically present in patients with the respective disorders and no COVID-19.
In group II, positive pattern 2 or 3 OCB, which were extremely rare in group I (~ 1%), pointed to the presence of coexisting MS (3 patients; 1 × with complicating John Cunningham [JC] virus-positive progressive multifocal leukoencephalopathy [PML] and a subsequent immune reconstitution inflammatory response [IRIS] after withdrawal of natalizumab) or viral meningoencephalitis (1 × HSV, 1 × VZV, 1 × HSV and CMV in an immunosuppressed patient). A bispecific MRZ reaction was noted in none of the patients in group I but was present in the only MS patient in group II tested for this marker. However, QAlb values were higher than expected (15 × 10 -3 ; typically < 12 × 10 -3 in MS) in one case, probably due to additional JC virus-associated (and possibly SARS-CoV-2-associated) BCB dysfunction.
HSV and VZV meningoencephalitis was associated with pleocytosis in 3/3, higher WCC at first LP (73, 177, 32 cells/µl) than observed in 97% of all patients in group I, CSF-restricted OCB in 2/3 patients, and BCB dysfunction in 3/3 with QAlb values (higher than those in 90% of patients in group II). In an immunosuppressed patient (after kidney transplantation), CSF PCR for CMV and HSV1/2 as well as serum PCR for CMV, HSV1/2 and EBV were all positive, further underlining the need for extensive PCR diagnostics in patients presenting with COVID-19 and neurological symptoms. Neutrophilic (70%) pleocytosis (32 cells/µl) was also noted in this patient, probably because septic brain infarction has occurred as a result of acute bacterial endocarditis rather than reflecting early viral infection. Neutrophils were found in only 3% of patients with available cytology data in group I and should thus be considered a red flag. In a further patient with a history of chronic viral meningitis of unknown cause In patients with additional SAH, high CSF erythrocyte counts (6050-> 100,000), siderophages, and xanthochrome CSF were noted as well as high TP level (up to 300 mg/dl).
Primary CNS lymphoma, present in one patient, was associated with a higher WCC (21 and 88 cells/µl) than observed in most patients in group I, elevated l-lactate levels and marked BCB dysfunction (QAlb 19 and 17, respectively; compared with a median QAlb of 7.9 in group I). In a patient with non-small cell-lung cancer, meningeal carcinomatosis and cerebral metastases, mild pleocytosis (10 cells/µl), and, elevated CSF l-lactate and TP levels were found.  19:19 Finally, in a patient with delayed wake-up reaction after sedation, encephalopathy, and myoclonus, a PET-CT-supported diagnosis of anti-Yo-associated autoimmune encephalitis (with positive antibodies in both CSF and serum) was made. CSF analysis revealed severe BCB damage (QAlb 26.7) and, of note, intrathecal synthesis of IgA and IgM (IgM-IF 51.83%, IgA-IF 35.24%), a feature not seen in any of the patients in group I; WCC could be determined only at follow-up and was then mildly elevated (17 cells/µl).

Discussion
A substantial proportion of patients with COVID-19 develop neurological symptoms. However, the pathophysiology of such complications is not fully understood. CSF analysis may aid in addressing this question. Furthermore, CSF analysis may be clinically important by virtue of guiding diagnostic and differential diagnostic decisions.
This study, which is, to the best of our knowledge, the largest and most comprehensive on CSF findings in COVID-19 conducted to date, demonstrates that in the majority of cases neurological involvement in COVID-19 is not associated with signs of intrathecal inflammation, as indicated by a normal WCC and a lack of quantitative (immunoglobulin CSF/serum ratios, reibergrams, antibody indices, Link index, FLC-kappa index) or qualitative (CSF-restricted OCB) evidence of total or SARS-CoV-2-specific IgG, IgA, or IgM synthesis within the CNS. Importantly, the remarkable absence of intrathecal inflammation was not restricted to patients with mild disease or those with peripheral symptoms but was found also in almost all patients with severe disease and in the B/SC subgroup.
The median interval between onset of the non-neurological COVID-19 symptoms or the first positive nasopharyngeal SARS-CoV-2 PCR and LP was just 19 and 12 days, respectively. Moreover, and, most importantly, the neurological symptoms had started only 5 days (median) before LP. It is theoretically possible that intrathecal antibody synthesis began later and was thus not detected. In fact, 10.9% of patients tested had not yet developed SARS-CoV-2 serum antibodies at the time of LP. On the other hand, OCB were missing also in all 21 samples obtained > 14 days after neurological onset and  , suggesting that the observed lack of intrathecal immunoglobulin synthesis was not simply due to timing. Rather, our findings favor the notion that the neurological symptoms observed in patients with COVID-19 are not caused by direct CNS infection with SARS-CoV-2 in most cases.
This said, it should not go unmentioned that in two patients in group I direct infection of the CNS with SARS-CoV-2 cannot be ruled out. These were the almost only patients with a strongly elevated WCC (247 and 510 cells/µl, compared to a median of 2 cells/µl in group I; neutrophils present in both) and both exhibited at the same time a positive SARS-CoV-2-IgG-AI. Further support in favor of infection of the CNS comes from the finding of a strongly elevated IL-6 CSF/serum ratio in one of them (AI > 17; not tested in the other patient)no IL-6 ratio (and index) as high was present in any of the SARS-CoV-2-IgG-AI-negative samples tested (which showed a median ratio of just 0.35) -, of a positive TNFalpha ratio (and high index), and of a positive SARS-CoV-2-IgM AI and SARS-CoV-2-IgA AI in the same patient (not tested in the second patient). In both cases, severe BCB dysfunction was noted. The capability of SARS-CoV-2 to infect neurons has been suggested by a recent study, and neuropathological autopsy findings suggesting CNS infection in some patients with COVID-19 have indeed been published [5,6].
As a limitation, SARS-CoV-2-CSF-PCR was negative in both cases. PCR was also negative in 82 further samples tested in this study. These findings from a large cohort confirm results from small case series and case reports also describing negative SARS-CoV-2-CSF-PCR results in patients with COVID-19 and neurological symptoms (summarized in [47]) and from a smaller French singlecenter study that found no unequivocally PCR-positive samples among 23 CSF samples from patients with COVID-19 (two post-mortem CSF samples were borderline positive in that study, but blood contamination was suspected in at least one of these, which was blood-PCR positive) [48].
A negative PCR result does not per se rule out infection of the nervous system. In a recent study, 96/96 Danish CSF samples from patients with encephalitis due to infection with tick-borne encephalitis virus, all of whom met the European Centre for Disease Prevention and Control (ECDC) criteria, were found to be reverse transcriptase (RT)-PCR-negative. In HSVE, a positive CSF PCR result is sometimes obtained only in sequential LP [49][50][51], probably reflecting both a relatively low viral load in the CSF at onset and the rostro-caudal gradient (lower concentrations in lumbar than in ventricular or cisternal CSF samples) that affects all CSF parameters in patients with CNS disease. In accordance with this notion, SARS-CoV-2-CSF-PCR was initially negative in a patient reported by Huang et al. (2020) but found positive by RT-PCR upon repeat LP [52]. This said, PCR was negative both in the initial and in a follow-up CSF sample in 6 patients of the present cohort and both in samples taken shortly after neurological onset and in samples taken 1-2 weeks later after neurological onset. No ventricular CSF or brain tissue were available for PCR testing in this study.
Similarly, the lack of CSF-restricted total IgG OCB and of quantitative evidence of intrathecal total IgG synthesis in those two patients with a positive SARS-CoV-2-AI and pleocytosis does not formally rule out CNS infection with SARS-CoV-2 either. Total IgG OCB (and also QIgG) are rather markers of chronic inflammation and may be negative at onset. Higher sensitivity of antigen-specific IgG AI than of markers of total IgG synthesis such as OCBs and immunoglobulin ratios has been previously shown [53][54][55][56][57][58][59][60][61][62]. For example, it has long been known that OCB are detectable only in a subset of VZV-AI-positive patients with (PCR-proven) varicella zoster meningitis or facial paresis [56]. Similarly, positive virus-specific AI and positive antigen-specific OCB, being part of the so-called polyspecific, intrathecal humoral immune response ('MRZ reaction'), have been found in patients with negative total IgG OCB and normal QIgG in multiple sclerosis [59][60][61][62].
We recommend that future studies on patients with COVID-19 and neurological symptoms should generally consider not only WCC, PCR, and markers of total IgG synthesis such as OCB and QIgG but also the SARS-CoV-2-IgG-AI (and, ideally, the SARS-CoV-2-IgA and -IgM AI), which might have diagnostic and prognostic implications, and, provided sequential LP are performed for clinical purposes, should examine the temporal dynamics of the intrathecal antiviral immune response in more detail. As the median interval between onset of neurological symptoms and AI determination was relatively short in the present cohort (median 5, range 1-49 days), results from follow-up LPs may be of particular interest. Based on the finding of high IL-6 and TNF-alpha CSF/ serum ratios and indices in the only SARS-CoV-2-IgG-AI-positive patient tested but not in any other patients, such studies should evaluate also IL-6 and TNF-alpha as supplementary markers of acute intrathecal inflammation in COVID-19. Finally, also potential cross-reactivity between anti-SARS-CoV-2 antibodies and neuronal antigens (as recently discussed [63]) could explain the rare finding of a positive SARS-CoV-2 AI in the absence of a positive SARS-CoV-2 PCR and deserves to be investigated in more detail.
A striking feature in this cohort was the high frequency of BCB dysfunction. Disturbed BCB integrity, indicated by an increased albumin CSF/serum ratio, was present not only at onset but also in samples taken more than 30 days later. The presence of BCB dysfunction sufficiently explains the elevated CSF immunoglobulin and free kappa and lambda light chain concentrations and the high total TP concentrations found in some patients, as indicated by a highly significant correlation of both CSF IgG and CSF TP with QAlb and the lack of QIg elevation.
Most importantly, in most cases BCB dysfunction was not associated with signs of cellular or humoral inflammation. Accordingly, a high rate of ACD was found (around 40% of all samples and around 80% in those with elevated QAlb). It is thus likely that BCB dysfunction was not primarily a result of intrathecal inflammation in our patients. The generally high rate of ACD needs also to be considered as a caveat when diagnosing patients presenting with COVID-19 and signs and symptoms of peripheral nerve disease. While ACD is an important marker of GBS, its predictive value for that disease may be lower in patients with COVID-19.
Given the existing reports on severe peripheral inflammation in COVID-19 and a cytokine storm in the blood of severely affected patients, inflammatory bystander damage of peripheral origin to the BCB seems at least conceivable. Blood leukocytosis is present in many patients with acute COVID-19, and a 'mirror OCB pattern' (also termed pattern 4), which may indicate peripheral, polyclonal acute inflammation, was present in 56% of our patients. Moreover, inflammatory cytokines, including markers indicating cellular inflammation, were frequently and persistently elevated in many patients in this cohort at the time of LP, with very high values in some. The spectrum of increased soluble markers of inflammation observed in our patients was broad, including both proinflammatory (such as IL-6, IL-8, IL1-beta, TNF-alpha and IFN-gamma) and anti-inflammatory (IL-10 and IL-1RA) cytokines and cytokine receptors. In some patients extremely high serum IL-6 levels were noted, suggesting severe systemic inflammation, and IL-6 remained persistently positive over time in > 95% of more than 700 samples taken over a median period of around 1 month during COVID-19-related hospitalization. This broad and strong activation of inflammatory pathways renders it likely that other mediators of inflammation implicated in BBB disruption not investigated in this cohort such as ROS (reactive oxygen species), NETs (neutrophil extracellular traps), MMPs (matrix metalloproteinases) and complement may have been elevated as well. In addition, hypoxia-or acidosis-related damage to the BCB in ICU-treated patients [74], who account for a high percentage of those included in this study, cannot be ruled out; in fact, a slightly (albeit not statistically significant) higher rate of BCB dysfunction was found in ICU-treated patients (58 vs. 41.5%). Finally, QAlb elevation and ACD were associated with GBS (N = 3) or brain infarction/bleeding (N = 7), conditions known to cause BCB dysfunction, in a relatively small subset of patients in this cohort.
An alternative (or additional) explanation for QAlb elevation is a change in CSF flow rate [25], which could, hypothetically, be caused by decreased CSF production or CSF resorption (e.g. due to infection [75][76][77] or inflammatory bystander damage affecting the function of the choroid plexus, the arachnoid granulations, or the cranial and/or spinal nerve sheaths).
CSF l-lactate levels were elevated in a quarter of our patients, based on age-dependent reference ranges, with a trend towards higher levels in acute samples. Physiologically, CSF l-lactate levels are relatively independent from serum l-lactate levels due to saturated lactate transporters. CSF l-lactate correlated significantly with QAlb, however, and the median l-lactate CSF level was significantly higher and CSF-lactate elevation significantly more common in patients with BCB dysfunction, suggesting a possible contribution of serum l-lactate to CSF level. Regrettably, l-lactate levels were not determined in the serum. However, CSF levels are usually higher than serum levels, rendering it more likely that CSF l-lactate and QAlb independently reflect the extent of CNS damage.
CSF l-lactate levels were also weakly correlated with the CSF WCC. Among CSF white cells, granulocytes are a known source of CSF l-lactate [78][79][80][81][82]. However, the frequency of pleocytosis was low and only few patients had neutrophils in the CSF. Accordingly, the frequency of l-lactate elevation did not differ between samples with and without granulocytes and no significant correlation between CSF granulocyte counts and CSF l-lactate levels was found in the present cohort. CSF l-lactate is thought to be produced also by astrocytes, which form an essential part of the BBB, following glutamate stimulation [83,84].
Finally, neurons may switch to glycolysis, in particular when their capacity to metabolize anaerobically the lactate of astrocytic origin is exhausted [84]. The elevated l-lactate levels may thus reflect cerebral hypoxia due to COVID-19-induced pneumonia or cerebral ischemia [85,86]. CSF l-lactate levels were indeed significantly higher in the subgroup of patients requiring ventilation at the time of LP or within 1 week before or after LP. Further studies are needed to better characterize the sources of intrathecal l-lactate in COVID-19.
The relatively high frequency of normal CSF findings means that altered CSF parameters in patients presenting with COVID-19 and neurological symptoms, in particular pleocytosis, intrathecal immunoglobulin synthesis and unusually high QAlb values, should always prompt clinicians to consider the presence of pre-or coexisting conditions, which may require specific treatment and must thus not be overlooked or mistaken for direct effects of COVID-19. In fact, a broad panel of coexisting neurological diseases was present in this cohort, which is hardly astonishing given the currently high prevalence of COVID-19. Importantly, these patients were analyzed separately (group II). The CSF findings in this subgroup were mainly representative of the respective comorbidity.
From a clinical point of view, it is also important to keep in mind that restricting CSF analysis to a basic panel (CSF WCC, TP, and l-lactate), as is often practiced in non-tertiary care or emergency settings, was associated with a relatively high rate of false-negative reports in our cohort. We strongly recommend applying a more complete panel including QAlb and, to rule out differential diagnoses, also QIgG, QIgA and QIgM, whenever possible.
While the absence of signs of intrathecal inflammation also in almost all samples obtained ≥14 days and even ≥30 days since neurological onset (including 89% and 81% in the B/SC subgroup) suggests that timing issues were not a general confounder, it cannot be fully ruled out that the humoral immune response had simply not yet developed at the time of LP at least in some cases. In HSVE, CSF is normal in 10-20% of cases during the early phase of disease, with a delay of around 1 week in the onset of the antibody response [35,87,88]. Moreover, a normal cell count or normal immunoglobulin ratios do not exclude indirect effects on the CNS due to systemic SARS-CoV-2 infection (e.g. induced by hypoxia, thromboembolic events triggered by COVID-19, or cytokines entering the CNS from the periphery, which are known to mediate not only inflammatory but also non-inflammatory effects in the nervous system). Therefore, it is important to underline that the absence of CSF signs of intrathecal inflammation should not lead one to prematurely dismiss patient-reported neurological symptoms as not being related to COVID-19. Of note, a recent study of autopsy specimens from patients with COVID-19 detected SARS-CoV-2 in cortical neurons as well as pathological features suggestive of infection, which were associated with only minimal immune cell infiltrates [6].
A lack of inflammatory pleocytosis and intrathecal total IgG synthesis does also per se not exclude CNS inflammation of autoimmune etiology. For example, in myelin oligodendrocyte glycoprotein-IgG-associated encephalomyelitis (MOG-EM) [89], a normal CSF has been observed in ~ 10% of all samples, both in children and adults, strongly depending on lesion sites (more commonly normal in optic neuritis than in brain disease, and least frequently in spinal cord disease) and OCB are missing in around 90% in this condition [13,14]. Similarly, a lack of intrathecal IgG synthesis has been observed in the vast majority of patients with AQP4-IgG-positive NMOSD [15,90,91], in LGI1-IgGassociated encephalitis [92,93], and in IgLON5-IgGpositive encephalomyelitis [92]), in a subset of patients with other CNS disorders of supposed autoimmune etiology such as Balo's concentric sclerosis [17], Schilder's disease [18], histopathologically defined "pattern II" and "pattern III" MS [16].
It is of interest in this context, that antibodies to neural antigens have been reported in some patients with COVID-19 [94,95], rendering it theoretically possible that virus-induced autoimmunity plays a role in a subset of patients with COVID-19 and neurological symptoms. However, independent confirmation of specific autoantibodies from larger and independent cohorts is widely lacking, as is information about target antigens in patients with supposed post-COVID-19 autoimmunity. In the present cohort, 1413 individual tests for established anti-neuronal antibodies were performed in the CSF and serum but were almost all negative. Moreover, 41 samples were in addition tested on nervous tissue sections and did not reveal evidence of unknown autoantibodies. Timing may be critical, however, and follow-up studies, especially in those with long-term sequelae ('long COVID-19') are certainly warranted. This is the largest number of tests for antineural antibodies reportedso far in studies on COVID-19.
The fact that BCB dysfunction and elevated levels of inflammatory cytokine levels (especially IL-6) persisted over weeks and months, respectively, is of interest in light of the many reports of patients with 'non-specific' longterm sequelae of COVID-19 such as malaise, fatigue, or cognitive impairment. Studies addressing the relevance of cytokines and BCB disruption in 'long COVID' are currently underway.
Of note, HSVE and VZV meningoencephalitis, respectively, were present in two patients. Whether infection with SARS-CoV-2 may promote reactivation of latent herpes virus infection remains to be investigated. While a number of reports on COVID-19-associated VZV (and, in one case, HSV-1 [96]) reactivation exist [97][98][99], including a case of suspected VZV encephalitis [100], coincidence cannot be ruled out given the high current prevalence of COVID-19. A further patient in the present series suffered from JC-virus induced PML. These findings underline the importance of considering the presence of additional, coexisting viral disease in all patients diagnosed with COVID-19. The detection of SARS-CoV-2 infection should not prompt one to dispense additional microbiological laboratory diagnostics. Rather, the diagnostic work-up in patients with proven SARS-CoV-2 infection and neurological symptoms should always be supplemented by PCR and AI analyses for coexisting viral infections.

Strengths and limitations
We consider the high number of samples included and the availability of follow-up samples as particular strengths of this study, together with the high number of parameters assessed. A further strength is the stratification according to the presence of pre-or coexisting disorders, which helped to identify CSF alterations attributable to conditions other than SARS-CoV-2 infection.
The retrospective and multicenter design of this study is a potential limitation. However, LP is an invasive and potentially harmful procedure that widely precludes prospective studies. The multicenter design can also be viewed as a strength, since it helped to reduce potential center-specific selection bias. Moreover, CSF diagnosis is highly standardized in central European university CSF laboratories, with centers participating in regular roundrobin tests required for obtaining quality certification. Finally, given the relatively low prevalence of neurological involvement severe enough to justify LP in COVID-19, a multicenter approach was necessary for inclusion of a sufficiently large number of patients in a short time, which was important given the current pandemic situation.
It is a limitation that some parameters, such as antineural autoantibodies, interleukins, interleukin receptors, chemokines, and kappa and lambda light chains, are not routinely assessed at some centers and, accordingly, data were available only for a subset of patients. It is therefore unknown whether the results reported here are representative for the total cohort. While we provide a large dataset for some of these parameters (1509 individual cytokine measurements; 939 individual autoantibody test results), more studies investigating these parameters are certainly warranted, for which our results may provide a rationale. Similarly, no predefined criteria for performing LP were used. Instead, the decision for LP was made in a clinical 'real-life' context based on the patient's clinical and paraclinical presentation. While this could theoretically have introduced referral bias (e.g., if thresholds for performing LP were higher at some centers), this is unlikely given that all LPs were performed in the setting of academic tertiary care hospitals with similar standards of care. CSF analysis is part of the general routine workup of patients with neurological symptoms of unknown cause in Germany, Austria, Switzerland, and Italy.
A trend towards more frequent and more severe CSF alterations in patients classified as having 'severe' or 'moderate' neurological deficits at the time of LP by the treating centers than in those with 'mild' deficits was noted. However, no standardized criteria for disease severity could be applied due to the retrospective nature of this study and the plethora of manifestations present. Further studies are needed to better define the relationship of disease severity and CSF findings in patients with COVID-19 and neurological involvement.
Finally, as this study was performed at European centers, most patients were of 'Caucasian' origin. It is unknown if our results are valid also for other ethnicities.

Conclusions
In summary, our study demonstrates (i.) that BCB dysfunction, as indicated by elevated QAlb, is the most common pathological CSF feature in patients with COVID-19 and neurological symptoms and may remain present for several weeks; (ii.) that BCB dysfunction in COVID-19 is not typically caused by intrathecal inflammation, as indicated by a lack of pleocytosis (and, correspondingly, a high rate of ACD) and intrathecal IgG synthesis in most patients with elevated QAlb, favoring a peripheral origin (e.g., due to systemic hypoxia, ischemia, elevated serum levels of cytokines and other mediators of inflammation, endothelial cell/choroid plexus infection with SARS-CoV-2, possibly antiendothelial autoimmunity); (iii.) that classical signs of intrathecal CSF inflammation (pleocytosis, CSF-restricted OCB, positive QIgG, QIgA or QIgM) are generally rare in patients with COVID-19 and neurological symptoms, both in samples obtained in the acute phase and in samples obtained later and irrespective of clinical disease severity; (iv.) that the SARS-CoV-2-IgG antibody index, which was elevated in 2 of the 3 only patients with marked pleocytosis (and in 2/20 patients tested), may be a marker of potential interest (alongside the SARS-CoV-2-IgM and -IgA AI) that may indicate either rare SARS-CoV-2 infection of the CNS in COVID-19 or cross reactivity of SARS-CoV-2-IgG with CNS antigens and which may be more sensitive than markers of total IgG synthesis, although more studies