The study was conducted in two phases using three different cohorts ( Table 1 ). In the first phase, we used a cohort (Multiplex discovery cohort) consisting of 6 patients with GBS, 6 amyotrophic lateral sclerosis (ALS) and 20 age-matched healthy controls (HC). In the second phase, we used two additional cohorts: the Multiplex replication and ELISA cohorts. The Multiplex replication cohort consisted of 6 patient with GBS, 6 CIDP, 4 non-inflammatory polyneuropathies (NIP), and 6 age-matched HC. The ELISA cohort consisted of 28 patients with GBS, 23 CIDP, 6 paraproteinemia-related demyelinating polyneuropathy (PDN), 6 multifocal motor neuropathy (MMN), and controls. The control groups in ELISA cohort consisted of 44 HC, 18 NIP, 33 pre-treatment multiple sclerosis (MS) sampled at diagnosis prior to treatment, and 31 ALS also sampled at diagnosis. There was no overlap between patients in the multiplex discovery cohort and the other two cohorts, i.e., different samples were analyzed. The NIP group in the Multiplex replication cohort consisted of hereditary demyelinating and axonal neuropathies, and in the ELISA cohort of chronic idiopathic axonal polyneuropathies and hereditary demyelinating and axonal neuropathies.

In the ELISA cohort, 7 patients with GBS and 7 patients with CIDP underwent a second lumbar punction (LP) after initiation of immunomodulatory treatment. These samples were classified as post-treatment.

The diagnosis of GBS was based on Asbury criteria (4) and/or Brighton criteria (level 1 or level 2 of diagnostic certainty) (5), whereas the CIDP, PDN and MMN diagnoses were based on the EFNS criteria from 2010 (16–18). The diagnosis of MS and ALS was based on well-recognized international criteria (19, 20).

Demographics, clinical features, electrodiagnostic and laboratory findings of all cohorts are summarized in Tables 1 , 2 .

All patients and controls (except HC) were diagnosed at the Department of Neurology at Karolinska University Hospital, Stockholm. The time of sampling in all three cohorts was same as the time of diagnosis, i.e., the samples were collected during the diagnostic workup. The only exception were samples collected post-treatment in the ELISA cohort in some patients with GSB and CIDP as presented in Table 1 . CSF and EDTA plasma samples from patients and controls were handled in a similar manner and were stored at the biobank of Department of Neurology between 2005 and 2017. Samples from HC used in Multiplex and ELISA cohort were collected at the Department of Neurology, University Hospital of Umeå between 2012 and 2013. CSF and plasma were centrifuged at 2000 g for 10 min and 400 μl of each was aliquoted into glass tubes and stored at −80°C. All samples were thawed once and aliquoted before shipment to the Olink laboratory, Uppsala, and the laboratory of the Department of Clinical Chemistry, Karolinska University Hospital.

The CSF and plasma proteins were measured using all available (thirteen) Olink panels in 2018. Each assay panel consisted of ninety-two protein targets (92 + 4 controls), resulting in a total of >1100 protein measures per sample and per patient. Protein quantification was performed using PEA technology, which utilizes paired oligonucleotide antibody probes and quantitative polymerase chain reaction (qPCR) for detection and quantification (15, 21). A mixture of antibody probes was incubated with each sample, i.e., two matched antibodies, labeled with DNA oligonucleotides bound to a target protein in solution, This brought the two antibodies into proximity, which allowed DNA oligonucleotides to hybridize and extend by a DNA polymerase. After hybridization and extension this newly created piece of DNA, which is unique for the specific antigen, was amplified by PCR. Following binding, hybridization, extension, and amplification protein levels were quantified using qPCR and expressed on a relative logarithmic scale with two as a base (log2) and with arbitrary units presented as normalized protein expression (NPX). Proteins with call rate <60%, i.e., proteins detected in less than 60% of the samples were excluded from further analyses. The call rates for proteins that passed quality control, their main function and alternative names (synonyms), as well as the list of all panels that were analyzed are provided in Supplementary Table 1 in the Supplementary Material .

Levels of IL8 in CSF and plasma were analyzed at the laboratory of the Department of Clinical Chemistry, Karolinska University Hospital using a Chemiluminescence assay on an Immulite 1000 instrument from Siemens (Siemens Healthcare Diagnostics Products GmbH, Marburg, Germany). Measuring range 2-7500 ng/L. Imprecision expressed as CV 7,2% at 109 ng/L, and 6,4% at 484 ng/L.

Clinical data were retrieved from the medical chart e-health system TakeCare and the national Swedish Neuro Registries (https://www.neuroreg.se). Patients with GBS were assessed with the GBS-disability scale (GBS-ds) (22) on the day of CSF/blood sampling, at nadir, 3 months ( ± 1 week), 6 months ( ± 2 weeks), and 12 months ( ± 4 weeks) later. The Erasmus GBS outcome score (EGOS) (23) was calculated for patients with GBS. Patients with CIDP, PDN, and MMN were assessed with the Inflammatory Neuropathy Cause and Treatment (INCAT) (24) scale on the day of CSF/blood sampling, at the above mentioned time points, and additionally at 18 months, ( ± 4 weeks), 24 months ( ± 4 weeks), 5 years ( ± 2 months), and 8–10 years post-onset.This study was approved by Regional Ethical Review Board of Stockholm (EPN 2014/1815-31/4 and 2017/952–31/1) and Umeå Ethical Review Board (EPN 2011–39-31 M).

The multiplex proteomics data and associations between protein levels were determined using a multi-variable linear regression model adjusted for sex and age at sampling, as previously described (25). Protein measures were analyzed with the default log base-two transformed protein levels and illustrated through volcano plots, negative 10log based transformed significance over NPX-based beta-coefficient. Associations were additionally corrected for multiple testing using a false discovery rate (FDR) approach using a FDR-corrected significance of p_FDR<0.05. Considering sample handling issues, the model was also corrected with markers used for assessing sample handling (CCL19 for CSF and CD40L for plasma) (25, 26).

The normality of ELISA IL8 data distribution was assessed with the Anderson-Darling test, D`Agostino & Pearson test, Shapiro-Wilk test, and Kolmogorov-Smirnov test. If the distribution was non-Gaussian, a log-based transformation was used before performing multiple comparisons. If the distribution was non-Gaussian or if the sample size was small, multiple comparisons were performed with the Kruskal-Wallis test. Otherwise, multiple comparisons were performed using one-way ANOVA. Multiple comparisons were corrected with Holm-Sidak or Dunn’s test. Likewise, depending on data distribution, correlation analysis was done by calculating Pearson or Spearman correlation coefficient. To further evaluate the diagnostic value of IL8 receiver operating characteristic (ROC) curve analysis was performed, and the area under the curve (AUC), sensitivity, specificity, and likelihood ratio (LR) were calculated. To evaluate the association between levels of IL8 and clinical outcome, Fisher´s exact test and logistic regression analysis were performed.

All statistical analyses were performed, and figures computed in R-4.1.3 and GraphPad Prism 8.0.

After correcting for multiple comparisons and using the model corrected with the additional sample handling markers, we found that pre-treatment levels of IL8 in CSF, but not plasma, in both Multiplex cohorts were upregulated in patients with GBS compared to all other groups, i.e., HC, CIDP, NIP and ALS (p_FDR < 0.05) ( Figures 1A–C ; Table 3 ). Furthermore, we also found that pre-treatment levels of IL8 in CSF, but not plasma, were upregulated in patients with CIDP compared to HC (p_FDR < 0.05) ( Figure 1D ; Table 3 ).

In this study, we focused particulary on those protein markers pertaining to IL8 and its pathway.

Considering results from both the discovery and replication multiplex cohorts showing increased levels of IL8 in CSF, but not plasma, we then analyzed CSF from a larger group of patients with acute and chronic inflammatory neuropathies, and controls using the ELISA test. With regards to plasma, we examined only five patients with GBS and five HC and found the same result as in the Multiplex cohorts, i.e., no statistical difference between these two groups (p > 0.05).

In patients with GBS, pre-treatment median levels of IL8 in CSF were significantly higher compared to all groups of patients with chronic inflammatory neuropathies, HC, and all other control groups ( Figure 2 ). Post-treatment levels were lower than pre-treatment levels (p < 0.0001), but higher compared to HC (p<0.03).

Comparing the levels of IL8 in patients with subacute inflammatory demyelinating polyneuropathy (SIDP; only 3 cases) to those with CIDP, we found that patients with SIDP had higher levels compared to patients with CIDP (Mann-Whitney p = 0.0008).

In patients with CIDP, pre-treatment median levels of IL8 in CSF were significantly higher compared to HC, NIP, and MS, but lower compared to GBS. Post-treatment levels were as high as pre-treatment levels and significantly higher than HC (p < 0.0001).

In patients with PDN, pre-treatment median levels of IL8 in CSF showed a similar pattern to CIDP. In patients with MMN, pre-treatment levels of IL8 in CSF did neither differ from HC, nor other disease controls ( Figure 2 ).

In patients with GBS, the time period between symptom onset and sampling was 9 days (median, IQR 4-14; mean 9.3, SD 5.1). In patients with CIDP, the time period between symptom onset and sampling was 9 months (median, IQR 6-12; mean 8.4, SD 3.6).

We did not find any correlation between time of symptom onset and levels of IL8 in CSF (GBS Spearman coefficient r = 0.18, p = 0.43; CIDP Spearman coefficient r = 0.23, p = 0.39).

ROC analysis comparing levels of IL8 in CSF in GBS versus HC, GBS vs. CIDP, CIDP vs. HC, CIDP vs. NIP, PDN vs. HC, and PDN vs. NIP, yielded AUC values of 1.0, 0.96, 0.90, 0.89, 0.94, and 0.94, respectively. Ninety-five percent confidential interval, p-values, cut-offs, sensitivity, specificity, and likelihood ratios (LR) are summarized in Figure 3 . An IL8 concentration of 73 ng/L in CSF was identified as cut-off to differentiate patients with GBS from those with CIDP, and 51 ng/L to differentiate CIDP and PDN from non-inflammatory polyneuropathies ( Figure 3 ).

We found a correlation between pre-treatment levels of IL8, and disease severity measured with GBS-ds at the time of sampling (Spearman coefficient r = 0.78, p = 0.0002), and nadir (r = 0.86, p < 0.0001), as well as outcome at 3-months follow-up (r = 0.62, p = 0.009), 6-months (r = 0.53, p = 0.03), and EGOS prognostic score (r = 0.66, p = 0.003) ( Figure 4 ). Taking GBS-ds of 3 points (inability to walk unaided) as a cut-off for severe impairment in the acute phase of the disease, we found that the group with a more severe impairment had significantly higher pre-treatment levels of IL8 compared to the group with a milder impairment at the time of sampling (p = 0.01), and at nadir (p = 0.002). The same result was found after excluding patients with MFS (p = 0.002, p = 0.0001 respectively) ( Figure 5 ).

To investigate if the higher pre-treatment levels of IL8 is a factor predictive of more severe impairment at the nadir, a Fisher´s test and logistic regression analysis were performed. The GBS group median (135 ng/L) was defined as the cut-off between low versus high levels of IL8 and GBS-ds of 3 points as the cut-off for more severe impairment. Using the Fisher´s exact test a statistically significant association was found between high pre-treatment levels of IL8 and more severe impairment at nadir (p=0.0019) ( Figure 6 ).

With the logistic regression model ( Table 7 ) the same outcome as a cut-off for more severe impairment was used and levels of IL8 as both continuous and binary variables were investigated. Taking pre-treatment levels of IL8 as a binary variable, the same cut-off as above for high levels of IL8 (135 ng/L) was used. In the simple logistic regression model, IL8 as both a continuous and binary variable was associated with more severe impairment at nadir. In the multiple logistic regression model with age taken into consideration, a similar result was found, i.e., higher levels of IL8 were associated with more severe impairment at nadir.

Thus, the logistic regression analysis showed the same association between pre-treatment levels of IL8 and impairment at the nadir as Fisher´s exact test.

We found a positive correlation between pre-treatment levels of IL8 in patients with CIDP and disease severity measured with the INCAT scale only at the 18-month follow-up (Spearman r = 0.65, p = 0.008) ( Supplementary Figure 1 in the Supplementary Material ). There was no correlation between IL8 levels and INCAT in patients with PDN and MMN.

There was no correlation between levels of IL8 and age, neither in all groups combined (r = 0.3, p < 0.0001), nor analyzed only in patients with inflammatory neuropathy (r = -0.01, p = 0.9) nor HC (r = 0.1, p = 0.5).

In patients with inflammatory neuropathies, we did not find any correlation between pre-treatment levels of IL8 and CSF cell counts (r = 0.01, p = 0.9), nor IgG-index (r = 0.3, p = 0.06). However, we found a positive correlation between pre-treatment levels of IL8 and albumin quotient (Q.alb) (CSF/plasma quotient of albumin) (r = 0.6, p < 0.0001), and with neurofilament light (NfL) (CSF r = 0.42, p = 0.01; plasma r = 0.5, p = 0.002). The correlation between IL8 and NfL was strongest in patients with CIDP (CSF r = 0.68, p = 0.003; plasma r = 0.62, p = 0.01). Of importance, five of the 21 patients with GBS with normal pre-treatment Q.alb had elevated levels of IL8 in CSF.