Autoreactive T cells target peripheral nerves in Guillain–Barré syndrome – Nature

Study participants

The study included 20 patients with GBS and 5 patients with CMT1 recruited from University Hospital Zurich and the Cantonal Hospital of Lugano (EOC), and 21 healthy donors obtained from the Swiss Blood Donation Center of Lugano (n = 15) and from the CoV-ETH study (n = 6). All participants provided written informed consent for participation in the study. The study was approved by the ethical committees of Zurich (NeuroMyoCyTOF study, BASEC-Nr: 2016-00929; CoV-ETH study, BASEC-Nr: 2020-00949) and Lugano (IGOS study, BASEC-Nr: 2018-01860). We included patients who were diagnosed with AIDP (n = 16), as well as patients with AMAN (n = 4, all associated with preceding gastroenteritis) or CMT1 (n = 5) on the basis of the criteria for GBS of the National Institute of Neurological Disorders and Stroke (NINDS)⁶¹ (Extended Data Table 1). Specifically, we included ten patients with AIDP who were sampled before the outbreak of the COVID-19 pandemic (non-COVID-19 GBS), for whom the potential trigger was either unknown (n = 6) or associated with VZV (n = 2) or CMV (n = 2) in the two to three weeks before disease onset. We also included five patients with AIDP with SARS-CoV-2 infection as a preceding trigger (post-COVID-19 GBS, range 6–20 days after infection, 11 ± 5.4 (mean ± s.d.)) (Extended Data Table 1). All patients with AIDP included in the study were HLA-typed by high-resolution next-generation-sequencing-based typing at University Hospital Zurich (Supplementary Table 4). One patient (PT16) also suffered from Waldenström’s macroglobulinaemia in 2013, for which he received allogeneic bone marrow transplantation owing to progression in 2021. The patient had herpes zoster before developing severe GBS disease in 2022 (more than one year after transplantation), and mild pre-existing axonal polyneuropathy due to chemotherapy was documented. Peripheral blood samples from patients with AIDP were collected both at acute phase (range 7–36 days from disease onset, 12.7 ± 9.9 (mean ± s.d.)) and/or at follow-up visits during the recovery stage (range 135–509 days from disease onset, 244.6 ± 115.7 (mean ± s.d.)) (Extended Data Table 1). When available, we also obtained CSF (n = 3) at disease onset. Moreover, a nerve biopsy was obtained from the left sural nerve from one patient (PT16).

Peptides and antigens

Peptides were synthesized as crude material on a small scale (1 mg) by Pepscan. Peptides used in the study included 15-mers overlapping by 10 covering the entire sequence of P0 (UniProtKB: P25189-1, n = 48), P2 (UniProtKB: P02689, n = 25) and PMP22 (UniProtKB: Q6FH25, n = 30) as well as human CMV and EBV HLA class I peptides (122 peptides, 46 EBV and 76 CMV). In some experiments, we used the heat-inactivated human CMV strain VR 1814 (ref. ⁶²) or peptide pools covering the entire sequence of SARS-CoV-2 proteins; namely, spike-domain S1 (UniProtKB: QHD43416.1, S325 and S536-S685 amino acid, pool S1(ΔRBD), 91 peptides), spike-domain RBD (UniProtKB: QHD43416.1, S316-S545 amino acid, 44 peptides), spike-domain S2 (UniProtKB: QHD43416.1, S676-S1273 amino acid, 118 peptides), nucleocapsid (UniProtKB: QHD43423.2, 82 peptides), membrane (UniProtKB: QHD43419.1, 43 peptides) and envelope (ENV; UniProtKB:QHD43418.1, 13 peptides). Seasonal influenza virus vaccine Influvac 2019/2020 was obtained from Mylan.

Cell purification and sorting

PBMCs were isolated with Ficoll-Paque Plus (GE Healthcare). Monocytes were enriched by positive selection using CD14-coated microbeads (Miltenyi Biotec). From the CD14^– cell fraction, memory CD4⁺ and CD8⁺ total cells were sorted to over 98% purity on a FACSAria Fusion (BD) excluding CCR7⁺CD45RA⁺, CD25^bright, CD14⁺ and CD56⁺ cells as well as either CD8⁺ cells (for memory CD4⁺ T cell enrichment) or CD4⁺ cells (for memory CD8⁺ T cell enrichment), according to the gating strategy shown in Extended Data Fig. 1b. The following fluorochrome-labelled mouse monoclonal antibodies were used for staining: CD4–PE/Dazzle 594 (1:500, clone RPA-T4), CD45RA–BV650 (1:500, clone HI100), CD8–APC Fire750 (1:80, clone RPA-T8) and CCR7–BV421 (1:80, clone G043H7) from BioLegend; CD14–PE–Cy5 (1:30, clone RMO52), CD25–PE–Cy5 (1:30, clone B1.49.9) and CD56–PE–Cy5 (1:30, clone N901) from Beckman Coulter; and CD19–FITC (1:20, clone HIB19) and CD25–PE (1:20, clone M-A251) from BD Biosciences. Cells were stained on ice for 15–20 min and sorted on a FACSAria Fusion (BD Biosciences). Within a few hours of sampling, the nerve biopsy sample was minced and then filtered through a 40-μm cell strainer to obtain a single-cell suspension. CSF samples (1–2 ml) were collected by lumbar puncture. Cells from the nerve biopsy or the CSF were stimulated polyclonally with 1 μg ml⁻¹ PHA (Remel) in the presence of irradiated (45 Gy) allogeneic feeder cells (1 × 10⁵ per well) and IL-2 (500 IU ml⁻¹) in a 96-well plate format, as previously described²⁹. On day 15, expanded T cells were stained with CD3–BV785 (1:100, clone UCHT1) and CD4–PE/Dazzle 594 (1:500, clone RPA-T4) antibodies from BioLegend, and CD8–FITC (1:30, clone B9.11) and CD56–PE–Cy5 (1:30, clone N901) antibodies from Beckman Coulter, and CD3⁺CD4⁺CD8⁻CD56⁻ or CD3⁺CD8⁺CD4⁻CD56⁻ T cells were sorted on a FACSAria Fusion (BD Biosciences).

In vitro stimulation of T cells

T cells were cultured in RPMI 1640 medium supplemented with 2 mM glutamine, 1% (v/v) non-essential amino acids, 1% (v/v) sodium pyruvate, penicillin (50 U ml⁻¹), streptomycin (50 μg ml⁻¹) (all from Invitrogen) and 5% heat-inactivated human serum (Swiss Red Cross). Ex vivo sorted memory CD4⁺ and CD8⁺ T cells or EM and CM memory CD4⁺ T cell subsets (PT16) from the blood as well as in vitro expanded and sorted CD4⁺ T cells from CSF or nerve biopsy were labelled with CFSE and cultured at a ratio of 2:1 with irradiated autologous monocytes untreated or pulsed for 1 h with selected peptide pools from P0, P2 and PMP22 (3 μg ml⁻¹ per peptide) or with control antigens Inflexal V (5 μg ml⁻¹) or EBV or CMV (1 μg ml⁻¹). After six days, cells were stained with antibodies to CD25–PE (1:20, clone M-A251) and ICOS–Pacific Blue (1:100, clone H4A3) from BioLegend. The T cell response was scored positive on the basis of a cut-off value of (i) a stimulation index ≥ 2 (% of CFSE^low cells with antigen and APC/% of CFSE^low cells with APC only and (ii) a Δ value ≥ 1.5% (% of CFSE^low cells with antigen and APC – % of CFSE^low cells with APC only). This threshold was chosen on the basis of previous observations made across multiple negative and positive samples assessed by ex vitro T cell stimulation techniques in a variety of donors with self-antigens²⁹. The list of samples analysed ex vivo is reported in Extended Data Table 2.

Isolation of autoreactive T cell clones

To isolate autoreactive T cell clones, CFSE^lowCD25⁺ICOS⁺ T cells from ex vivo cultures were sorted and cloned by limiting dilution, as previously described²⁹. T cell clones were analysed by stimulation with irradiated autologous B cells that were untreated or pulsed for 1 h with P0, P2 or PMP22 peptide pools (3 μg ml⁻¹ per peptide). To determine MHC restriction, the assay was performed in the absence or presence of blocking anti-MHC class II monoclonal antibody (10 μg ml⁻¹; anti-HLA-DR, clone L243; anti-HLA-DQ, clone SPVL3; anti-HLA-DP, clone B7/21). In the cross-reactivity experiments with SARS-CoV-2 or CMV antigens, T cell clones were stimulated with irradiated autologous B cells after 2–3 h of pulsing with P0, P2 or PMP22 peptide pools (3 μg ml⁻¹ per peptide) or SARS-CoV-2 peptide pools (2 μg ml⁻¹ per peptide) or the heat-inactivated human CMV strain VR 1814 (2.5 μg ml⁻¹) (ref. ⁶²). Epitope mapping experiments were performed by stimulating of autoreactive T cell clones with irradiated autologous B cells after one hour of pulsing with single 15-mer overlapping peptides (3 μg ml⁻¹ per peptide) covering the whole P0, P2 or PMP22 protein lengths. In all experiments, proliferation was measured on day 3 after 16-h incubation with 1 μCi ml⁻¹ [methyl-³H]-thymidine (Perkin Elmer). Cell lines were routinely tested to exclude mycoplasma contamination.

Cytokine analysis

For the quantification of cytokine release by autoreactive T cell clones, cells were stimulated with irradiated autologous B cells, either untreated or exposed for 1 h to P0, P2 or PMP22 peptide pools (3 μg ml⁻¹ per peptide). Cytokines released in the 48-h culture supernatants were quantified by the LEGENDplex multiplex bead-based immunoassay, using the predefined Human T Helper Cytokine Panels Version 2 (BioLegend) according to the manufacturer’s instructions. Data were acquired using the FACS LSR Fortessa (BD Biosciences) and analysed with the Data Analysis Software Suite for LEGENDplex (BioLegend).

For intracellular cytokine staining, autoreactive T cell clones were restimulated with phorbol-12-myristat-13-acetat (PMA) and ionomycin in the presence of brefeldin A (all from Sigma-Aldrich) for the last 2.5 h of culture. Cells were stained with LIVE/DEAD Fixable Aqua dye (Thermo Fisher Scientific) and then fixed and permeabilized with Cytofix/Cytoperm (BD Biosciences) according to the manufacturer’s instructions. After fixation, cells were stained with anti-granzyme A (1:50, clone CB9), anti-granzyme B (1:50, clone QA18A28), anti-perforin (1:50, clone dG9), anti-TNF (1:160, clone MAb11), anti-IL-10 (1:50, clone JES3-9D7) and anti-IL-17A (1:400, clone BL168) all from Biolegend; anti-IFNγ (1:160, clone B27) and anti-IL-4 (1:100, clone MP4-25D2) from BD Biosciences; and anti-IL-22 (1:50, clone 22URTI, Thermo Fisher Scientific), conjugated with different fluorochromes. Cells were acquired on a FACS LSR Fortessa (BD Biosciences) using BD FACS Diva (v.9.0) and flow cytometry data were analysed with FlowJo v.10.8.1 software (FlowJo).

scRNA-seq analysis

scRNA-seq analysis was performed on memory CD4⁺ T cells from two patients with AIDP (PT2 and PT16) at day 6 after in vitro stimulation either with a mixture of P0, P2 and PMP22 antigens (PNS-myelin antigens) or with influenza vaccine (Flu). Cells were incubated with a unique oligonucleotide barcode conjugated to a human universal antibody (Sample Tag, BD Single-Cell Multiplexing Kits) for backtracking both the condition and the patient of origin of each cell. For each condition, antigen-reactive CFSE^low and non-reactive-CFSE^high T cells were FACS-sorted to retrieve the total cell numbers and later combined at a 1:1 ratio in single tubes for further processing using the BD Rhapsody Express Single-Cell analysis system. In brief, cells were labelled with viability dies following the manufacturer’s instructions and loaded onto BD Rhapsody Cartridges. The cartridges were subsequently analysed in the BD Rhapsody Scanner to obtain an estimate of the total cells and to verify their viability. After single-cell capture with the gravity-based, beads-assisted microwell technology we amplified the whole transcriptome, the TCR library and the Sample Tag library according to the manufacturer’s protocols. We sequenced the library at the Functional Genomic Center Zurich (FGCZ) using the Illumina NextSeq 500 System. In detail, we sequenced 20,000 reads per cell for the WTA libraries, 5,000 reads per cell for the TCR libraries and 1,000 reads per cell for the Sample tag libraries. We used the SevenBridges online platform to perform read alignment on the reference genome ‘Homo_sapiens_GENCODE_GRCh38-p13_Release_37-2021-05-04’ and to generate feature-barcoded matrices for downstream analysis. The computational analysis allowed us to assign patient, condition, TCR and whole transcriptome information to each single cell analysed. After quality control, which involved the filtering of low-quality cells and cell doublets or multiples, and cells with mitochondrial counts higher than 5%, we normalized the data and performed scaling, dimensionality reduction and clustering on the top 2,000 highly variable features in the dataset (Seurat v.4.9.9.9059). In total, we obtained 1,980 cells (PT2 acute, n = 608; PT2 recovery, n = 262; PT16 acute, n = 1,110) from cultures stimulated with PNS-myelin antigens and 2,232 cells (PT2 acute, n = 287; PT2 recovery, n = 224; PT16 acute, n = 1,721) from cultures stimulated with influenza vaccine. We later allocated the cluster on the basis of the expression levels of activation and proliferation genes^25,26 to define antigen-specific and non-specific clusters. Antigen-reactive T cell clusters comprised 413 PNS-myelin-reactive (PT2 acute, n = 181; PT2 recovery, n = 40; PT16 acute, n = 192) and 414 Flu-reactive single T cells (PT2 acute, n = 148; PT2 recovery, n = 149; PT16 acute, n = 117) encompassing, respectively, 209 and 242 single TCRα/β clonotypes (data not shown). We combined the acute and recovery datasets and compared the expression levels of gene signatures previously associated with different T helper subsets and cellular cytotoxicity^{63,64,65,66,67,68,69} (all with Seurat v.4.9.9.9059) and then performed gene set enrichment analysis using the software package ‘escape’ (v.1.10.0, https://github.com/ncborcherding/escape)⁷⁰ using R v.4.2.1.

TCR Vβ sequencing

To determine the TCR Vβ sequences of autoreactive T cell clones, total cDNA was obtained from 10³–10⁴ cells and TCR sequencing was performed following an established protocol²⁹. In brief, the reaction was carried out using HPLC-purified oligo dT(25) primers (Microsynth) and Maxima H Minus reverse transcriptase (Thermo Fisher Scientific), in a reaction mix containing 0.2% Triton, dNTPs and RNase inhibitor. Reactions were run with the following conditions: 50 °C × 60 min; 55 °C × 5 min. Five microlitres of cDNA was added to a PCR mix (final volume 25 μl) containing Q5 Hot Start High-Fidelity DNA Polymerase (New England Biolabs). TCR Vβ sequences were amplified using TCR Vβ-specific forward primer pools and reverse primers pairing to constant regions, as previously described²⁹. Sequence amplifications were assessed through agarose gel electrophoresis. Successfully amplified fragments were sequenced by the Sanger method, and TCR Vβ sequence annotation was performed using the IMGT/V-QUEST algorithm²⁹.

Deep sequencing of TCR was performed on CD4⁺ memory T cells sorted ex vivo from PMBCs or in vitro expanded and sorted CD4⁺ T cells from CSF or nerve biopsy as well as on PNS-myelin reactive T cells enriched as CFSE^low fractions from in vitro stimulation (2.5 × 10⁵–5 × 10⁵ cells). In brief, cells were washed in PBS and genomic DNA was extracted from the pellet using the QIAamp DNA Micro Kit (Qiagen), according to the manufacturer’s instructions. Sequencing of TCR Vβ was performed by Adaptive Biotechnologies using the ImmunoSEQ assay, as described previously²⁹. In brief, after a multiplex PCR reaction designed to target any CDR3 Vβ fragments, amplicons were sequenced using the Illumina HiSeq platform. Raw data consisting of all retrieved sequences of 87 nucleotides or corresponding amino acid sequences, and containing the CDR3 region, were exported and further processed. Each TCRβ clonotype was defined as the unique combination of nucleotide sequence; data processing was done using the productive frequency of templates provided by ImmunoSEQ Analyzer v.3.0 (http://www.immunoseq.com) and by R package immunarch V.0.9.0 (https://github.com/immunomind/immunarch).

Antigen-specific TCRβ clonotypes in each donor’s repertoire were identified through bioidentity overlap, defined as identical identified V gene, amino acid sequence of the CDR3 β region and identified J gene. The samples analysed are listed in Extended Data Table 2. Cumulative frequencies of shared TCRβ clonotypes were calculated as the sum of frequencies of each TCRβ clonotype in the respective patient’s TCR Vβ repertoire. CDR3β length was calculated on the total productive rearrangements from the ImmunoSEQ Analyzer v.3.0 or the IMGT/V-QUEST algorithm.

GLIPH2 analysis

The GLIPH2 algorithm^32,71 from the HetzDra/turboGliph v.0.99.2 R package (https://github.com/HetzDra/turboGliph/) was used to identify lymphocyte interaction by paratope hotspots and predict specificity groups, herein referred as clusters, on the basis of global or local similarity (convergence). TCR global convergence relies on the CDR3 hamming distance between TCRs; namely, the number of different amino acid residues within the CDR3 region amongst two TCRs with identical length and sharing the same Vβ segment. TCR local convergence relies instead on similarity based on shared CDR3 amino acid motifs (2mers, 3mers, 4mers and 5mers) within any given set of T cell receptors (>10× fold enrichment, probability < 0.001). Notably, TCRs are allowed to be assigned to multiple clusters if computed similar to one another. GLIPH2 scores result from a combination of probabilities of a set of features, which are then combined into a single score by conflation. Such features include global similarity probability, local motif probability, network size; enrichment of V gene in the cluster, enrichment of CDR3 length in the cluster, enrichment of clonal expansion in the cluster and enrichment of common HLA alleles among TCRs from donors contributing to the cluster. The GLIPH2 algorithm is trained by a reference dataset of 162,165 CDR3β sequences and the query sample size should be comparable to the size of the training set^32,71. Therefore, we run multiple rounds of analyses on different groups of patients and cohorts. We performed the GLIPH2 analysis on the TCR Vβ repertoire of total memory CD4⁺ T cells from the blood of patients with GBS (n = 10), grouped by disease phase. In detail, we analysed seven samples from the acute phase (PT1, PT2, PT5, PT7, PT10, PT11, PT12; total: 82,826 TCR Vβ sequences) and eight samples from the recovery phase (PT1, PT2, PT4, PT5, PT7, PT9, PT12, PT13; total: 239,501 TCR Vβ sequences, two rounds). We also performed the analysis on CFSE^low enriched fractions of PNS-myelin specific CD4⁺ T cells from PT16 (EM and CM T cell populations from the acute phase; total: 568 TCR Vβ sequences) and on a published TCR Vβ dataset of memory CD4⁺ T cells from healthy donors (n = 9; C5, C6, C7, C8, C9, C10, C12, C13, C15; total: 149,939 TCR Vβ sequences, two rounds)^27,28,29. In addition, we applied the GLIPH2 analysis to the TCR Vβ repertoire of total CD4⁺ T cells expanded from the CSF of patients with GBS (n = 3; PT10, PT11, PT12; total: 2,525 TCR Vβ sequences) and from the nerve tissue of one patient with GBS (n = 1, PT16; total: 99 TCR Vβ sequences). The TCR Vβ sequences from PNS-myelin specific memory CD4⁺ T cells isolated from the blood after in vitro stimulation (PT16), and total CD4⁺ T cells from the CSF and nerve biopsy (PT16) were analysed together in one round of GLIPH2 computations. Finally, we conducted one further round of GLIPH2 analyses on the TCR Vβ repertoire retrieved from scRNA-seq experiments on two patients with GBS (PT2 and PT16; total: 1,733 unique TCR Vβ sequences). In detail, the analysis was conducted on 526 TCR Vβ sequences from antigen-specific and 1,207 TCR Vβ sequences from non-specific CD4⁺ memory T cells after six days of stimulation with PNS-myelin antigens or influenza vaccine. In each round of analysis, we included the autoreactive TCRβ clonotypes isolated from the blood of patients with GBS (n = 167, of which n = 18 were shared across the memory CD4⁺ T cells TCR Vβ repertoires of several patients). Clusters were considered of relevance if they included one autoreactive TCRβ clonotype of known specificity and were shared by multiple patients with GBS. If the same cluster could be identified in different rounds of GLIPH2 analysis amongst different groups (for example, GBS acute, GBS recovery or healthy donors), that cluster would be considered as one, but the identifier code would be maintained to preserve positional information (Supplementary Table 5).

Prediction of binding affinity of self-epitopes to HLA class II alleles

Binding-affinity predictions between HLA alleles carried by patients with GBS and the PNS-myelin peptide of interest identified through epitope mapping were performed using the NetMHCIIpan-4.0 server provided by the DTU Health Tech Department of Health Technology³⁷. In brief, the artificial neural networks are trained over half a million experimental measurements of binding affinity and eluted ligand mass spectrometry covering the human HLA-DR, HLA-DQ and HLA-DP. When instructed with information regarding the HLA subtype of interest and a peptide of choice (15-amino-acid peptides), it can forecast the likelihood of a peptide being naturally presented, its predicted affinity and, the likelihood of that peptide being presented as compared with a group of random peptides. From NetMHCIIpan-4.0, we extrapolated the binding affinity of the HLA alleles of each patient known to be carrying public TCRβ clonotypes versus the specific epitope recognized by those public TCRβ clonotypes.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.