Hemocyanin from keyhole limpet (KLH) was purchased from Sigma-Aldrich (Saint-Quentin-Fallavier, France). rFVIII Kogenate was a kind gift from Dr. Pedro Paz and Dr. Jeanette Lo (Bayer, San Francisco, CA, USA). Potential T cell epitopes were selected from the full-length FVIII (UniProt ID: P00451) using NetMHCIIpan 4.0 as described in Supplementary Table S1. The corresponding peptides were synthesized by Proteogenix Science (France). Ovalbumin was purchased from Sigma-Aldrich (France).

PBMCs were purified from the whole blood of anonymous healthy donors, who gave informed consent (Etablissement Français du Sang, Rungis, France), using Ficoll-Paque®. Monocyte-derived dendritic cells (DCs) were produced from cells isolated using Percoll gradient by 5 days of culture in AIM-V medium (Invitrogen, Villebon-sur-Yvette, France) supplemented with 1,000 U/ml recombinant human (rh) IL-4 and rhGM-CSF (both from R&D Systems, Lille, France). Mature DCs were produced by two supplementary days of culture in the presence of LPS (1 µg/ml) and R848 (10µg/ml). CD4 T cells were isolated from autologous PBMCs by positive selection using magnetic labeling with anti-CD4 mAbs conjugated to magnetic microbeads followed by magnetic cell sorting, as recommended by the manufacturer (Miltenyi Biotec, Bergisch Gladbach, Germany). Immature DCs were loaded overnight at 37˚C with either 0.2 µM of rFVIII, 0.2 µM of ovalbumin, or 0.25 µM of KLH in 96-well U-bottom plates, while mature DCs were used for FVIII peptides (10 ug/ml). CD4 T cell lines were generated in vitro in IMDM (Invitrogen) supplemented with 10% human serum, 0.24 mM glutamine, 0.55 mM asparagine, 1.5 mM arginine (all amino acids from Sigma-Aldrich), 50 U/mL penicillin/50 μg/mL of streptomycin (Invitrogen), 20 IU/mL human IL-2, and 17 IU/ml human IL-7 (premium-grade; Miltenyi Biotec) by three once-per-week rounds of stimulation of CD4 T cells (200,000 per well) with autologous DCs (20,000 per well). 10,000 lymphocytes were collected at day 19 from each T cell line (expanded T cells contained in one well) and their specificity was assessed by using IFN-g ELISPOT.

T cells were incubated with autologous DCs alone (control) or with DCs incubated with rFVIII (0.2 µM), ovalbumin (3 µM), and KLH (1 µM) or PBMCs for FVIII peptides (10 µg/ml) in MultiScreen 96-well plates (Merck Millipore) previously coated with 2.5 µg/mL of anti-human IFN-γ monoclonal antibody (mAb; 1-D1K; Mabtech). The T cell lines and APCs were plated at equal number for both stimulated and unstimulated control conditions. After overnight incubation, spots were revealed using 0.25 µg/mL of biotinylated anti-human IFN-γ mAb (7-B6-1; Mabtech) in phosphate-buffered saline/bovine serum albumin 1%, extravidin-phosphatase (dilution 1:3,000 in phosphate-buffered saline/Tween20 0.05%/bovine serum albumin 1%; Sigma-Aldrich), and nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate-toluidine salt (Sigma-Aldrich). The spot number was determined by using the ELISPOT Reader System (Autoimmun Diagnostika, Ebinger, Germany). HLA restriction was assessed using monoclonal antibodies (10 μg/ml) specific to HLA-DP (B7/21), HLA-DQ (SPVL3), or HLA-DR (L243).

Positive antigen-specific T cell lines determined by using the ELISPOT assay were pooled together at day 21, after 24 h of resting in IMDM medium and with the absence of cytokine stimuli. The pooled T cell lines were re-stimulated overnight with autologous DCs alone (control) or with DCs previously loaded with either 0.2 µM of rFVIII, 0.2 µM of ovalbumin, or 0.25 µM of KLH. Following 24 h of stimulation, cells were harvested in cell staining buffer (phosphate-buffered saline/EDTA 2 mM/bovine serum albumin 0.5%) and then stained for CD4 (Vio-Blue, clone REA623, Miltenyi Biotec) and activation-induced markers CD134 (OX-40) (PE anti-human, clone Ber-ACT35, Biolegend) and CD137 (4-1BB) (APC anti-human, clone 4B4-1, Biolegend). The CD4/CD134/CD137-positive population for each condition was sorted by flow cytometry (FACSARIA III; Becton Dickinson, San Jose, CA, USA).

RNA was extracted from antigen-specific T cells using the RNeasy Plus Mini Kit (Qiagen), and sample RNA was reverse-transcribed using a TCRB chain constant region reverse primer with Superscript IV First-Strand Synthesis System (Invitrogen). cDNA was column-purified with the Oligo Clean and Concentrator Kit (Zymo Research). cDNA was used as a template for PCR with KAPA2G Fast Multiplex Mix (Roche). Forward TCRB FR3 primers were used with a single nested TCRB constant region reverse primer. A total of 20 cycles for NGS library preparation for the PCR1 and 20 cycles of PCR2 were performed on the PCR1 product to incorporate sample barcodes and Illumina sequencing adaptors using Herculase II Fusion DNA Polymerase (Agilent). Specific principles and methodology have been previously reported (20). The PCR2 product underwent a run of agarose gel electrophoresis. Bands at 300 pb were purified using a GeneJET purification kit (Thermo Fisher) and quantified using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific).

The samples were sequenced on an Illumina NextSeq 500 with 15% of PhiX control v3 (Illumina). All data were aligned using the MiXCR v4.2.0 tool (https://github.com/milaboratory/mixcr). The high-quality reads were further assembled into clonotypes, correcting for PCR and sequencing errors. Shared clonotypes between the unstimulated and antigen-stimulated samples were removed. The downstream analysis was performed with MIXCR, “Immunarch” R package, and in-house R https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE253521script. All data are accessible on the NCBI GEO database (accession number: GSE253521; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE253521).

Graphical illustrations were made within FlowJo 10, GraphPad Prism 8.3.0, the “Immunarch” R package, and BioRender (biorender.com). VJ chord diagram was created using “circlize” CRAN package.

Firstly, we aimed to characterize the T cell response to rFVIII in vitro using chicken ovalbumin (Ova) as a non-self-protein benchmark. Given the low frequency of FVIII-specific T cells in healthy donors, we enriched the CD4 population in vitro before analyzing their specificity and sorting them for TCR CDR3β sequencing (Figure 1A). CD4 T cells from four healthy donors were stimulated by three rounds of stimulation with DCs loaded with FVIII, Ova, or KLH as a positive control. Antigen specificity was assessed using the IFN-γ ELISPOT assay. All four donors responded to KLH, while only a portion of the T cell lines raised against rFVIII and Ova showed a specific response, indicating the limited dilution conditions of specific T cells at the start of the culture (19) (Figure 1B). FVIII and ovalbumin generated 69 and 58 specific T cell lines, respectively, among 96 T cell lines tested for each antigen. As demonstrated previously (21), the distribution of the antigen-specific CD4 T cells at the initiation of our assay follows a Poisson distribution. This allows us to calculate the mean frequency of antigen-specific CD4 T cells. Interestingly, the number of precursor frequencies able to recognize specifically FVIII was comparable to the response obtained for the non-self-protein ovalbumin, namely, 6 and 5 cells/M (Figure 1C). The identified positive T cell lines were re-stimulated overnight with the respective antigen of specificity or unloaded DCs as a negative control. The cells were then stained for T cell receptor (TCR)-dependent activation-induced markers (AIMs) CD137 (4-1BB) and CD134 (OX-40). The activation percentages obtained were comparable, with a mean of 1.85% for FVIII and 1.25% for ovalbumin, and the variability across the four tested donors ranged from 0.8% to 4.0% of activation for FVIII and 0.9% to 1.6% for Ova (Figure 1D). Antigen-specific CD4 T cells (CD134/CD137 double-positive cells) (Figure 2) and the corresponding unstimulated control samples were thus sorted and their TCR CDR3β sequenced (Figure 3A). Within individual donors, the total repertoire distribution among FVIII and Ova samples was generally similar. The average number per donor of detected clonotypes was 535 and 918, for a total of 2,143 and 3,672 clonotypes detected for FVIII and Ova. Using the Gini-Simpson index as a measure of T cell repertoire diversity and clonotype abundance, Ova and FVIII show a very similar polyclonal distribution among the donors tested, 0.95 and 0.98, respectively. Interestingly, for FVIII, on average 13 clonotypes are covering 50% of the total repertoire and 33 clonotypes instead for Ova. To investigate the CDR3β characteristics of these highly expanded clonotypes, the following analysis will refer to the clonotypes covering the top 50% of the total repertoire for both antigens. β-CDR3 sequences specific for FVIII had lengths ranging from 10 to 19 aa, with a preferential length between 12 and 13 aa (Figure 3B). While those for Ova had lengths ranging from 11 to 21 aa, with a preferential length of 13 aa (Figure 3C). CDR3 sequences of different lengths were aligned using “MUSCLE” (22). The CDR3β motifs showed no strong over-representation of specific aa, except for a slightly higher prevalence of glycine residues. In the FVIII CDR3β motif, we observed an enrichment in hydrophobic, polar, and neutral aa, where alanine, leucine, threonine, proline, and valine were the most present in the core sequence (Figure 3D). On the other hand, Ova shows a prevalence of polar, basic, and neutral aa composition, with arginine, serine, asparagine, glutamine, and tyrosine highly present (Figure 3E). Concerning the V motif, FVIII T cell clones show a narrow gene usage signature, predominantly represented by TRBV7-4 (60%) and TRBV7-3 (25%). Similarly, TRBJ1-2 and TRBJ2-7 are the genes encoding 50% of the total FVIII repertoire, followed by TRBJ2-5 (16%) and TRBJ1-1 (9%) (Figure 3F). On the other hand, ovalbumin showed a slightly wider VJ usage and less representative gene pairing. Among the TRBV, TRBV7-4 accounts for 35% of expression frequency among the total repertoire, followed by TRBV7-3 (34%), TRBV4-1 (7%), and TRBV12-3 (8%). The TRBJ genes showed a similar usage distribution, with TRBJ2-7 being expressed on 25% of the total repertoire and TRBJ2-1 (18%), TRBJ1-1 (11%), TRBJ2-5 (9%), TRBJ1-2 (8%), and TRBJ2-2 (8%) (Figure 3G).

Based on the T cell epitope prediction software NetMHCpan4.0 (23), we selected 63 20-mer peptides that are predicted for their ability to bind promiscuously, with a percentile score below 10%, at least four HLA class II molecules from a selection of 15 preponderant HLA-DR alleles in our panel (as shown in ST1). The predicted peptides spread all along the full-length FVIII sequence and account for more than 50% of its amino acid coverage. A total of 16 healthy donors with unrelated HLA typing (ST2) were used as sources of T cells to identify FVIII-specific CD4 T cell epitopes. Specific T cell lines were generated by in vitro T cell culture using FVIII and KLH. The peptide specificity of the T cell lines raised against FVIII was assessed by two independent IFN-γ ELISPOT using peptide pools and individual peptides. All donors responded to rFVIII with a total of 409 identified rFVIII-specific T cell lines (Figure 4A). The mean of CD4 T cell precursors was 6.5 cells/M (Figure 4B). Each of the 63 FVIII peptides induced a T cell response in at least one of the donors tested, among the 16 healthy donors tested, with 872 peptide-specific T cell lines identified (Figure 4C). The CD4 T cell epitopes identified were ranked based on their donor coverage capacity (Figure 4D) and response intensity (Figure 4E). Furthermore, 18 out of 63 FVIII T cell epitopes were responsible for 50% of the total T cell response and 100% of the donor coverage. These immunoprevalent epitopes are located in the A1, A3, C1, and C2 domains of the FVIII protein (ST3). Furthermore, four independent healthy donors were used as sources of T cells to characterize the HLA restriction and to assess the reactivity of T cell lines generated with the 18 peptides toward rFVIII. T cell lines were amplified in vitro for 28 days through stimulation with the pool of 18 peptides and then tested through IFN-γ ELISPOT. All four healthy donors tested responded to the pool of 18 FVIII peptides selected, with a total of 87 specific T cell lines identified. Moreover, 16 out of 18 FVIII peptides in the pool generated a specific T cell response among the four donors. A total of 14 of the T cell peptides were tested for recognition of rFVIII-loaded on autologous DCs, and it was confirmed that they were part of the naturally processed FVIII peptides. HLA restriction was assessed by introducing anti-HLA-DR, anti-HLA-DP, and anti-HLA-DQ antibodies. None of the peptides tested was HLA-DQ restricted. Eight peptides were exclusively HLA-DR-restricted, one peptide was HLA-DP, and seven were promiscuously presented on both HLA-DR and HLA-DP molecules (ST4). These data confirmed the in silico binding affinity prediction for both HLA-DR and DP molecules (ST6). Furthermore, the 18 FVIII peptides selected were ulteriorly tested in an in vitro binding assay to six HLA-DRB1 molecules (as described in ST5).

Ultimately, we aimed to characterize the T cell response against the 18 FVIII epitopes selected and compare it to the response to the full-length protein. Five healthy donors were used as sources of T cells. The T cells were amplified in vitro for 28 days as previously described. For each donor, 24 T cell lines were stimulated with a pool of the 18 FVIII epitopes, 24 T cell lines with rFVIII, and eight T cell lines with KLH simultaneously. Specificity was then assessed by using IFN-γ ELISPOT. All donors tested responded to the pool of 18 FVIII epitopes and rFVIII. A total of 76 T cell lines specific for the pool and 85 specific for rFVIII were identified (Figure 5A). Calculation of the antigen-specific precursor frequency in this experiment confirmed again 6.4 cells/M for rFVIII and 6 cells/M for the pool (Figure 5B). To characterize TCR sequences responding to the pool and their distribution within the total rFVIII repertoire, specific T cell lines were sorted based on AIM, and TCR CDR3β bulk sequencing was performed. We found 2,990 shared expanded sequences between the FVIII and pool repertoire among the five donors (Figure 5C). Surprisingly, 178 of these sequences are public clonotypes shared among multiple HLA-unrelated donors, as shown in Figure 5D, accounting individually for up to 32% of the total FVIII repertoire.