After informed consent according to the Declaration of Helsinki, healthy individuals (homozygously) expressing HLA-A*01:01 and HLA-B*08:01 or HLA-A*02:01 and HLA-B*07:02 were selected from the Sanquin database and the biobank of the department of Hematology, Leiden University Medical Center (LUMC). Peripheral blood mononuclear cells (PBMCs) were isolated by standard Ficoll-Isopaque separation and used directly or thawed after cryopreservation in the vapor phase of liquid nitrogen. Donor characteristics (HLA typing, CMV and EBV serostatus) are provided in Table 1. For isolation of donor-derived virus-specific T cells using fluorescence-activated cell sorting (FACS, gating strategy see Supplementary Figure 1) with pMHC-tetramers (Table 2) and expansion of donor-derived virus-specific T cells, see Supplementary Material and Methods.

TCRβ-sequences were identified using ARTISAN PCR adapted for TCR PCR (29, 30). Total mRNA was extracted from 190 pMHC-tetramer^pos purified (Supplementary Figure 2A) virus-specific T-cell populations (31) using magnetic beads (Dynabead mRNA DIRECT kit; Invitrogen, Thermo Fisher Scientific). Ten µl (~1µg) of mRNA per sample was mixed with TCRβ constant region-specific primers (1μM final concentration) and SmartSeq2modified template-switching oligonucleotide (SS2m_TSO; 1μM final concentration) and denatured for 3 minutes at 72°C. After cooling, cDNA was synthesized for 90 minutes at 42°C with 170 U SMARTscribe reverse transcriptase (Takara, Clontech) in a total volume of 20µl containing 1.7U/μl RNasin (Promega), 1.7mM DTT (Invitrogen, Thermo Fisher Scientific), 0.8mM each of high-purity RNAse-free dNTPs (Invitrogen, Thermo Fisher Scientific) and 4µl of 5x first-strand buffer. During cDNA synthesis, a non-templated 3’polycytosine terminus was added (Supplementary Figure 2B), which created a template for extension of the cDNA with the TSO (32). PCR (2min at 98°C followed by 40 cycles of [1s at 98°C, 15s at 67°C, 15s at 72°C], 2 min at 72°C) of 5µl of cDNA was then performed using Phusion Flash (Thermo Fisher Scientific) with anchor-specific primer (SS2m_For; 1μM final concentration) and each (1μM final concentration) of the nested primers specific for the constant regions of TCRβ constant 1 and TCRβ constant 2. Both forward and reverse PCR primers contained overhanging sequences suitable for barcoding. Amplicons were purified and underwent a second PCR (2min at 98°C followed by 10 cycles of [1s at 98°C, 15s at 65°C, 30s at 72°C], 2 min at 72°C) using forward and reverse primers (1μM final concentration) with overhanging sequences containing identifiers (sequences of 6 base-pairs) and adapter sequences appropriate for Illumina HiSeq platforms (or PacBio; Pacific Biosciences). Unique identifiers were used for each T-cell population targeting one epitope. Forward or reverse identifiers were shared between T-cell populations targeting different epitopes. For all primer sequences see Supplementary Table 1. For identifier sequences see Supplementary Table 2. Amplicons with identifiers were purified, quantified and pooled into one library for paired-end sequencing of 150bp on an Illumina HiSeq4000. Deep sequencing was performed at GenomeScan (Leiden, The Netherlands). Raw data were de-multiplexed and aligned to the matching TRBV, TRBD, TRBJ and constant (TRBC) genes. CDR3β-sequences were built using MIXCR software using a bi-directional approach (5’-3’ and 3’-5’ read) (Supplementary Figure 2C) (33). CDR3β-sequences with a stop-codon were removed from the library. Bi-directional readings using MIXCR could result in out-of-frame CDR3β AA-sequences due to the even number of nucleotides. These sequences (n=392) were manually aligned with the germline TRBV and TRBJ-sequence. CDR3β-sequences were further processed using custom scripts in R to compare specificities and sharing of CDR3β-sequences.

The following R-packages were used in R-software to generate a nodal plot of CDR3β AA-sequences with the levenshtein distance as parameter for similarity: “igraph” to create network objects, obtain the degree of a node and its betweenness (34), “data.table” to organize CDR3β-sequences; “stringdist” to calculate Levenshtein distances (35), “Biostrings” for fast manipulation of large biological sequences or sets of sequences (36), “dplyr” to arrange and filter data (37), “tibble” for providing opinionated data frames, “ggplot2” for generating figures (38) and “RColorBrewer” to create graphics (39). A levenshtein distance of 0.25 was added to visualize multiple identical sequences. Nodes with identical sequences (levenshtein distance of 0.25) were manually replaced by pie-charts using Adobe Illustrator CC 2018.

To identify which positions of PUB-I and PUB-HS CDR3β AA-sequences were conserved and which were variable, all CDR3β AA-sequences with the most frequent CDR3 length were included and the AAs were stacked for each position in the sequence. The overall height of the stacks indicates the sequence conservation at that position, while the height of symbols within the stacks indicates the relative frequency of each AA at that position. AAs have colors according to their chemical properties; polar AAs (G, S, T, Y, C, Q, N) show as green, basic (K, R, H) blue, acidic (D, E) red, and hydrophobic (A, V, L, I, P, W, F, M) AAs as black (40).

As an independent TCR-database containing TCR-sequences from virus-specific T cells, we used the information obtained from the virus-specific T-cell products generated in the context of the phase I/II safety and feasibility study T Control (EudraCT-number 2014-003171-39) using the MHC-I-Streptamer isolation technology (Juno Therapeutics, Munich, Germany) (20, 41). Sequencing was performed as described above for all virus-specific T-cell populations per donor, resulting in unique identifiers for all virus-specific T-cell populations in the TCRβ-library.

TCR-Sequencing data is deposited to the Sequence Read Archive (SRA); submission: SUB10993301 (access to bioproject: https://www.ncbi.nlm.nih.gov/bioproject/PRJNA803981).

To examine the composition of the virus-specific TCR-repertoires in different individuals, the CDR3β-regions of purified expanded pMHC-tetramer-binding virus-specific T-cell populations were sequenced (Supplementary Figure 2). We analyzed the TCR-repertoires of CMV, EBV and AdV-specific T cells, restricted to four prevalent HLA alleles (HLA-A*01:01, HLA-A*02:01, HLA-B*07:02 and HLA-B*08:01) and specific for 21 different peptides (Table 2). Purified CMV, EBV and AdV-specific T-cell populations targeting CMV (n=8), EBV (n=10) or AdV (n=3)-derived peptides were isolated from 17 HLA-A*01:01/B*08:01^pos individuals and 12 HLA-A*02:01/B*07:02^pos individuals (Tables 1, 2). In total, 190 virus-specific T-cell populations, each targeting a single viral epitope, were successfully isolated and showed high purity (>97% pMHC-tetramer positive). The mean precursor frequencies of the different T-cell specificities in the starting PBMC materials are shown in Supplementary Figure 3. Sequencing of the CDR3β-regions of these virus-specific T-cell populations resulted in 3346 CDR3β nucleotide-sequences that occurred at frequencies of more than 0.1%. 135 of these nucleotide-sequences were present at high frequencies (>5%) within one specificity, but were also found at low frequencies (around 0.1%) in another specificity, indicating contamination due to FACSorting impurities. These low frequency nucleotide-sequences were discarded from further analysis. 41 nucleotide-sequences were present at low frequencies in two unrelated specificities and could not be correctly annotated and these 41 duplicates (n=82) were also discarded from the analysis. Therefore, a total of 3129 nucleotide-sequences could be annotated. In total, 2224 (71%) of these nucleotide-sequences represented unique CDR3β AA-sequences that were found in only one individual and 905 nucleotide-sequences (29%) resulted in 131 different PUB-I CDR3β AA-sequences that were found in two or more unrelated individuals (Flowchart; Figure 1).

To investigate the relationship between the numbers of CDR3β nucleotide-sequences and the translated number of CDR3β AA-sequences, we compared the nucleotide-sequences of the 131 PUB-I CDR3β AA-sequences found in different individuals. Different nucleotide-sequences can result in the same CDR3β AA-sequence, a phenomenon known as convergent recombination. We found that PUB-I CDR3β AA-sequences present at high (representative example; Figure 2A) or low frequencies (range 0.1-1%) (representative example; Figure 2B) could be encoded by different CDR3β nucleotide-sequences in the junctional regions of the CDR3β-regions in TCRs of T cells isolated from different individuals (Figures 2C, D and Supplementary Figure 4). Because the majority of nucleotide-sequences encoding the same CDR3β AA-sequences were different between individuals, these data exclude contamination as an explanation for the finding of PUB-I CDR3β AA-sequences.

We then investigated the distribution of the 131 PUB-I CDR3β AA-sequences within the 21 different specificities and the prevalence among individuals for each of the PUB-I CDR3β AA-sequences per viral-epitope. T cells with PUB-I CDR3β AA-sequences were found for 19 out of the 21 specificities (Supplementary Table 3). PUB-I CDR3β AA-sequences were not observed in AdV-IE1^LLD and EBV-LMP2^ESE-specific T-cell populations. Some T-cell populations (e.g. EBV-LMP2^FLY) contained many different PUB-I CDR3β AA-sequences (n=24) that were all highly-similar. For this reason, we investigated the distribution of PUB-I CDR3β AA-sequences with unique TRBV and TRBJ-gene usage. This resulted in 29 different PUB-I CDR3β AA-sequences, distributed over 19 specificities (Figure 3A; grey bars). Six specificities contained two or three different (expressing different TRBV and/or TRBJ-genes) PUB-I CDR3β AA-sequences that were highly prevalent among individuals. To investigate how often these PUB-I CDR3β AA-sequences could be found in our cohort of healthy donors, we quantified the prevalence of each of these 29 PUB-I CDR3β AA-sequence (Figure 3A; grey bars). Because we classified a PUB-I CDR3β AA-sequence as being present in at least 2 individuals, the prevalence among donors could not be less than 2 out of 17 (12%; 17 is maximum number of T-cell populations for 1 specificity). Only 4 out of 29 PUB-I CDR3β AA-sequences were found in only 2 individuals (Figure 3A; grey bars). Overall, these 29 PUB-I CDR3β AA-sequences had a prevalence of 33% among healthy individuals (median; range 12%-82%). Importantly, most PUB-I CDR3β AA-sequences were found in at least 25% of individuals and 5 were even present in more than half the donors.

We and others hypothesized that the binding/docking of TCRs to HLA-peptide complexes might allow for small changes/flexibility in the CDR3 AA-sequences without significantly changing the conformation or interaction (42). Therefore, we investigated if there were CDR3β AA-sequences present in our data set that were highly-similar (PUB-HS) to PUB-I CDR3β AA-sequences and differed by 1, 2 or 3 AAs. The 2224 unique TCR nucleotide-sequences identified in our previous analysis may contain PUB-HS CDR3β AA-sequences that are in fact part of the same public response as the respective PUB-I TCRs. In total, 379 PUB-HS CDR3β nucleotide-sequences were present that also resulted in 379 PUB-HS CDR3β AA-sequences that differed by 1, 2 or 3 AAs from one of the 131 PUB-I CDR3β AA-sequences. This shows that 41% of the total virus-specific TCR-repertoire contained PUB-I and PUB-HS CDR3β nucleotide-sequences. We investigated if these PUB-HS CDR3β AA-sequences were also present in individuals that did not contain the respective PUB-I CDR3β AA-sequences. PUB-HS CDR3β AA-sequences were present for 21 out of 29 PUB-I CDR3β AA-sequences (Figure 3A; shaded orange bars). When we include the PUB-HS CDR3β AA-sequences and quantified the 29 PUB-I and PUB-HS CDR3β AA-sequences, these had a median prevalence of 50% among healthy individuals (range 23%-100%). The AdV-IE1^LLD and EBV-LMP2^ESE-specific T-cell populations, where PUB-I CDR3β AA-sequences were not found, did contain PUB-HS CDR3β AA-sequences in multiple individuals at high frequencies (43) (Supplementary Figure 5). The frequencies of PUB-I combined with PUB-HS CDR3β AA-sequences were relatively high within each virus-specific T-cell population of each individual (Figure 3B). The frequencies of all PUB-I and PUB-HS CDR3β AA-sequences ranged from 0.1%-99.4% within the 19 different virus-specific T-cell populations with a median of 13.1%. When combined, all but one PUB-I plus PUB-HS CDR3β AA-sequences were found in at least 25% of individuals and 3 were even found in over 75% of individuals. These data show that for many PUB-I CDR3β AA-sequences we found sequences that were similar (1, 2 or 3 AA-differences), making up more than 40% of the total virus-specific TCR-repertoire and together these sequences were found in the majority of individuals at high frequencies.

To investigate how the PUB-HS CDR3β AA-sequences related to the PUB-I CDR3β AA-sequences, we analyzed if the PUB-HS (1, 2 or 3 AA-differences) CDR3β AA-sequences showed variations at random positions or at specific positions compared to the PUB-I CDR3β AA-sequences. We hypothesized that if the binding/docking of PUB-HS TCRs was not significantly different, conserved regions and regions that allow for some variation could be identified in the CDR3β AA-sequences. As Illustrated in Figure 4A for the EBV-LMP2^FLY-specific PUB-I CDR3β AA-sequence CASSYQGGNYGYTF, two motifs were identified with AA-differences predominantly located at positions 5 and 9/10 of the CDR3β-region. The AAs [QGG] at positions 6-8 were conserved for both motifs. In total, the two motifs consisted of 86 PUB-HS CDR3β AA-sequences with 1 or 2 AA-differences. The majority (n=71) had the same CDR3 length of 14 AAs as the PUB-I CDR3β AA-sequence, implying that variations were caused by AA-substitutions. The remaining 15 PUB-HS CDR3β AA-sequences had a CDR3 length of 15 AAs, due to AA-inserts, compared to the PUB-I CDR3β AA-sequence (Figure 4B). Similar rules were found for the other 20 PUB-I CDR3β AA-sequences. Also here, some AA-positions were highly conserved, whereas others were variable. However, the precise locations of the variable AAs differed between specificities (Representative examples; Figure 4C). Interestingly, the corresponding CDR3-alpha sequences of a few highly-frequent PUB-I and PUB-HS CDR3β-sequences were also identical or highly-similar between different individuals (Supplementary Table 4).

As a control, we assessed if these conserved motifs were predictive for the specificity when searching in our database of 2355 unique CDR3β AA-sequences. The requirement was that each motif should not be present in another specificity. We observed that some specificities contained motifs of only 3 or 4 AAs that were exclusive for that specificity and were not observed in any other specificity (Table 3). Altogether, these data show that the variations in the CDR3β AA-sequences were not random, but occurred at specific positions that resulted in conserved regions that were predictive for the specificities.

We hypothesized that if the conserved junctional region is a crucial part of the peptide-HLA binding, virus-specific TCR-repertoires could also contain CDR3β AA-sequences with the same conserved region, while allowing different TRBJ-gene usage, as long as the 3-dimensional conformation would allow this. Since the TRBJ-regions often differ by more than 3 AAs, we were not able to include these as PUB-HS CDR3β AA-sequences. Such PUB-HS CDR3β AA-sequences that use different TRBJ-genes might even further increase the prevalence of PUB-I and PUB-HS CDR3β AA-sequences in the virus-specific TCR-repertoire. To investigate this, we performed a computational analysis using the levenshtein-distances (AA-differences) between all different CDR3β AA-sequences. For four different specificities (EBV-LMP2^FLY, EBV-EBNA3A^RPP, AdV-E1A^LLD, and AdV-HEXON^TDL) we observed clustering of CDR3β AA-sequences that expressed the same TRBV-genes while using different TRBJ-genes. For example (Figure 5A), the HLA-A*02:01-restricted EBV-LMP2^FLY-specific CD8^pos T-cell repertoire contained 2 clusters within the cluster of TRBV6-5-expressing T cells (TRBV6-5/TRBJ1-2 and TRBV6-5/TRBJ2-1). The majority of CDR3β AA-sequences within the TRBJ1-2 cluster had a length of 14 AAs, while CDR3β AA-sequences from the TRBJ2-1 cluster had a length of 13 AAs (Figure 5B). Analysis of the junctional regions of the TRBV6-5/TRBJ1-2 and TRBV6-5/TRBJ2-1-encoded CDR3β AA-sequences revealed strong conservation of AAs [QGG] on positions 6-8, despite different TRBJ-usage and CDR3 lengths (Figure 5C). Similarly, the HLA-A*01:01-restricted AdV-HEXON^TDL-specific CD8^pos T-cell repertoire contained two large clusters of CDR3β AA-sequences, using TRBV20-1 or TRBV5-1 (Figure 5D), all with a CDR3 length of 13 AAs. The first cluster (TRBV20-1) contained sub-clusters of CDR3β AA-sequences using TRBJ1-1, TRBJ2-3 or TRBJ2-7 and the second cluster (TRBV5-1) contained CDR3β AA-sequences using TRBJ2-1 or TRBJ2-7. AdV-HEXON^TDL-specific CDR3β AA-sequences expressing TRBV20-1 revealed strong conservation of AAs [PGQG] on positions 4-7, which fell outside the region encoded by TRBJ (Figure 5E). Additionally, AdV-HEXON^TDL-specific CDR3β AA-sequences expressing TRBV5-1 revealed strong conservation of AAs [N:D] on positions 4 and 7, despite different TRBJ-usage. These examples illustrate that virus-specific TCR-repertoires can have conserved CDR3β-regions, while using different TRBJ-genes, allowing substantial variability at specific positions encoded by the TRBJ-region. This will further increase the prevalence of PUB-I and PUB-HS CDR3β AA-sequences in the total virus-specific TCR-repertoire.

To determine whether the magnitude of PUB-I and PUB-HS CDR3β AA-sequences was particular for our cohort of individuals with a homozygous HLA background, we investigated if the same phenomenon was also present in individuals with a heterogeneous HLA background. We performed the same analyses on virus-specific CD8^pos T-cell populations targeting 11 different viral epitopes that were generated and used in the context of a clinical study (20). A total of 1157 CDR3β nucleotide-sequences could be correctly annotated. In total, 695 (61%) nucleotide-sequences resulted in unique CDR3β AA-sequences, that were only found in one individual, and 462 nucleotide-sequences (39%) resulted in 89 different PUB-I CDR3β AA-sequences. From the 695 unique CDR3β AA-sequences, 134 PUB-HS CDR3β nucleotide-sequences were present that differed by 1, 2 or 3 AAs from one of the 89 PUB-I CDR3β AA-sequences. This shows again that also in this cohort a large part (51%) of the total virus-specific TCR-repertoire contained PUB-I and PUB-HS CDR3β nucleotide-sequences. Because the targeted viral epitopes were not fully identical in both cohorts, we could investigate the prevalence of 20 out of 29 PUB-I CDR3β AA-sequences in this cohort. In total, 17 out of 20 CDR3β AA-sequences that were previously identified, could also be identified in this independent cohort. When we included the PUB-HS CDR3β AA-sequences and quantified the 17 PUB-I and PUB-HS CDR3β AA-sequences, these sequences had a similar high prevalence of a median of 89% among healthy individuals (range 26-100%). (Figure 6A). These CDR3β AA-sequences were also present at high frequencies within each virus-specific T-cell population (Figure 6B). These data show that the same PUB-I or PUB-HS CDR3β AA-sequences are also present in virus-specific T cells isolated from an independent cohort of individuals with a heterogeneous HLA background with a similar prevalence among donors and frequency within donors.