Peripheral Blood Mononuclear Cells (PBMCs) and plasma samples of consenting healthy donors were purchased from Cellular Technology Limited (CTL). Initial assay development and expansion experiments were performed on a cohort of 20 healthy donors, comprising 10 AAV9 seropositive and 10 AAV9 seronegative individuals ( Supplementary Table 1 ). Deconvolution studies to identify immunodominant CD8 T cell epitopes were conducted on a cohort of 24 healthy donors, which included 17 donors from the initial cohort (3 donors were no longer available due to vendor stock exhaustion) as well as 7 additional donors. Additional donors were selected to ensure representation of predicted peptide binding alleles that were absent in the initial cohort. In total, PBMCs and/or plasma from 27 healthy donors were utilized in the study. Donor IDs and HLA allele information for all the donors can be found in Supplementary Table 2 .

Pre-existing anti-AAV9 antibodies were measured in human plasma for a cohort of 76 healthy donors using the affinity capture and elution (ACE) TAb assay as previously published (15). Briefly, Nunc Maxisorp 96-well plates were incubated for 1 hour with 100 µL per well of AAV9-GFP (8E10 DRP/mL) and washed. Human plasma samples and controls were diluted 1:10 in Tris-buffered saline and added to the washed plates, followed by overnight incubation on a shaker at 4 degrees Celsius. The following day, plates were washed and treated with 65 µL per well of 300 mM acetic acid. After a 5-minute incubation, 50 µL of the acid-eluted antibodies were transferred to MesoScale Discovery (MSD) plates containing 50 µL of 1 M Tris-HCl neutralization buffer (pH 9.5). The MSD plates were incubated for 2 hours at room temperature to allow the immobilization of anti-AAV9 antibodies. Following this, the MSD plates were washed, treated with 100 µL per well of ruthenium labeled AAV9-GFP (9.8E10 DRP/mL), and incubated for 1 hour at room temperature. After a final wash, 2X MSD read buffer was added and signals were collected in an MSD Quick Plex SQ 120 reader. Donors with antibodies above the assay Tier 1 cut point were defined as seropositive and those below the assay Tier 1 cut point were defined as seronegative ( Supplementary Table 1 ).

The overlapping (O/L) peptide pool contained peptides spanning the VP1 capsid protein of AAV9, organized as 15-mers with a 11 amino acid overlap (182 peptides). The MAPPs-derived peptide pools comprised naturally presented peptides of the VP1 capsid protein of AAV9 identified through MAPPs. Three MAPPs-derived peptide pools were utilized for the study- Class I MAPPs pool (containing 41 MAPPs-derived capsid peptides presented on HLA Class I), Class II MAPPs pool (containing 12 MAPPs-derived capsid peptides presented on HLA Class II), and the total MAPPs pool (containing 53 MAPPs-derived peptides, combining HLA Class I and Class II peptides) ( Supplementary Table 3 ). Peptide pools were dissolved in 100% DMSO and tested at 1 µg per peptide/mL. All peptide pools were synthesized by GenScript.

For the identification of immunodominant CD8 T cell epitopes, the 41 HLA Class I MAPPs peptides were arranged in a matrix of 13 pools, each containing 6–7 peptides ( Supplementary Table 4 ). Per the matrix, each matrix pool was constructed by combining the peptides listed in each column or each row (16). Peptides present at the intersection of responding matrix pools were shortlisted and tested individually. All matrix pools were dissolved in 100% DMSO and tested at 1 µg per peptide/mL. For individual testing, peptides were dissolved in 100% DMSO and tested at 2 µg/mL. Matrix pools and individual peptides were synthesized by GenScript.

The manufacturer’s instructions were followed for FluoroSpot plate set up and processing. AIM V containing 5% Serum Replacement (SR) (Gibco) was used as the culture media. Frozen PBMCs were thawed in the water bath and rested overnight in the CO2 incubator at 37 degrees Celsius. The following day, rested PBMCs were washed and plated in triplicates (250,000 cells/well) on precoated IFN-γ/TNF-α FluoroSpot plates (MabTech) along with test peptide pools (1 µg per peptide/mL). Anti-CD3 antibody (MabTech, 100 ng/mL) and DMSO were used as positive and negative controls respectively. Plates were incubated at 37 degrees Celsius for ~ 24 hours, washed, and treated with primary and secondary antibodies. Processed plates were dried and read on the MabTech IRIS instrument to record the number of spots per well.

Frozen PBMCs were thawed and resuspended in AIM V+ 5% SR at 1X10^6 cells/mL. Cells were stimulated with test peptide pools (1 µg per peptide/mL) and expanded in vitro for 10 days as described previously (17). On Days 4 and 7, cells were supplemented with a cytokine cocktail containing IL-2 (R&D Systems, 10 units/mL), IL-7 (R&D Systems, 10 ng/mL), and IL-15 (PeproTech, 10 ng/mL). On Day 10, cells were harvested, washed, and restimulated on precoated IFN-γ FluoroSpot plates (MabTech) in triplicates (125,000 cells/well). Peptide pools (1 µg per peptide/mL) or individual peptides (2 µg/mL, for deconvolution studies) were used for restimulation. Anti-CD3 antibody (MabTech, 100 ng/mL) and DMSO were used as positive and negative controls respectively. Plates were incubated at 37 degrees Celsius for ~ 24 hours and processed as described in the previous section.

For experiments involving T cell depletion, CD4 T cells were depleted via magnetic separation using human CD4 Microbeads (Miltenyi Biotec) and an AutoMACS separator (Miltenyi Biotec) according to manufacturer’s instructions. Depleted cells were washed with MACS buffer prior to seeding on culture plates. An average of 85-95% T cell depletion was achieved in all donors as confirmed through flow cytometry.

Urea-induced denaturation was used to evaluate the stability of peptide-HLA complexes as described before (18). Briefly, peptides were dissolved in DMSO and 1mM β-mercaptoethanol and dispensed in 96-well plates at a final concentration of 2 µM. The HLA allele of interest was diluted in assay buffer containing β-2 microglobulin and added to the plate in the range of 2–10 nM to ensure excess of peptides during peptide-HLA complex formation. Each plate also included reference peptides with known stable binding to each allele. Following complex formation, samples were transferred to a 384-well plate and stressed with 8 different urea concentrations ranging from 0–7 M. Plates were then developed using ELISA as previously described (19, 20). For each peptide, binding stability to a given allele was calculated relative to the 100% binding stability of the reference peptide. Peptides with a relative binding stability of 50% or higher were annotated as stable binders.

The affinity of peptide/HLA binding was determined as described before (21, 22). Briefly, biotinylated HLA Class I heavy chains were denatured and diluted into refolding buffer containing β-mercaptoethanol and varying concentrations of the test/control peptides. The mixtures were incubated with agitation at room temperature for 20–24 hours to enable peptide/HLA complex formation. The solutions containing the complexes were transferred to streptavidin plates and incubated for 30 minutes. Plates were then developed using ELISA. Absorbance measured at 450 nm was graphed against the varying concentrations of the test peptide used. The affinity of peptide/HLA binding was determined by calculating the EC50, defined as the peptide concentration resulting in half saturation. Peptides with an EC50 under 100 nM were classified as high affinity binders.

For FluoroSpot assays, results were expressed as Spot Forming Units (SFU), calculated as the spot count per million PBMCs. The Stimulation Index (SI) was calculated as the average SFU in test wells/average SFU in DMSO control wells. For assays with fresh PBMCs, an SI of 2 or higher with a minimum of 50 SFU was counted as a positive IFN-γ/TNF-α response. For assays with expanded PBMCs, an SI of 1.4 or higher was set as the cutoff for a positive IFN-γ response. Predicted HLA binding alleles for each peptide were determined using MHCMotifDecon 1.0 as previously described (13, 14, 23). Predicted binding affinities for identified epitopes were determined using NetMHC 4.0 (24). Data tables and heatmaps were created using Excel. Data analysis and graphical representations were performed on Graph Pad Prism 10.

A previous study reported the detection of AAV capsid-specific T cells in PBMCs isolated from AAV2 seropositive healthy donors (25). To determine if we could detect AAV9 capsid-specific T cells in seropositive healthy donors, we first analyzed plasma samples from a cohort of 76 healthy donors for anti-AAV9 antibodies using a total antibody assay (15). We identified 19 seropositive and 57 seronegative donors, of which we selected 10 seropositive and 10 seronegative donors for further analysis ( Supplementary Table 1 ). Frozen PBMCs from the 20 selected donors were stimulated for 24 hours on a FluoroSpot assay with either an overlapping peptide pool derived from the AAV9 capsid (15-mers with 11 amino acid overlap, 182 peptides) or a total MAPPs peptide pool containing both HLA Class I and Class II MAPPs-derived peptides of the AAV9 capsid (53 peptides). Circulating AAV9 capsid-specific T cells were detected by measuring IFN-γ and TNF-α release in response to peptide stimulation. We noted a low overall T cell frequency in all tested donors, with no significant detection of capsid-specific T cells in response to either peptide pool ( Figures 1A, B ). Importantly, there was no differentiation in the cytokine response to AAV9 capsid peptides between seropositive and seronegative donors ( Figure 1B , Supplementary Figure 1 ). This lack of connectivity between anti-AAV antibody and T cell response is best exemplified by Donor # 526 which had the highest antibody signal among all analyzed donors yet lacked T cell response to either overlapping or MAPPs-derived peptide pools ( Figure 1A , Supplementary Table 1 ). Overall, irrespective of donor serostatus, healthy donor PBMCs appeared to have low numbers of circulating capsid-specific T cells, which resulted in poor sensitivity on the FluoroSpot assays.

To increase the frequency of capsid-specific T cells in healthy donors, we adopted a PBMC expansion protocol, wherein PBMCs from 20 healthy donors were expanded with the overlapping or the total MAPPs capsid peptide pools and a cocktail of cytokines for 10 days prior to restimulation on FluoroSpot assays. Expanded PBMCs were then restimulated with either the overlapping peptide pool, the total MAPPs peptide pool, the Class I MAPPs peptide pool (41 HLA Class I MAPPs capsid peptides), or the Class II MAPPs pool (12 HLA Class II MAPPs capsid peptides) ( Figure 2A ). Through PBMC expansion, we sufficiently improved the detection of capsid-specific T cells in healthy donors and were also able to compare the T cell populations expanded by the overlapping and MAPPs-derived capsid peptides.

PBMCs expanded with the overlapping peptide pool displayed a strong IFN-γ response when restimulated with the same pool of peptides. A positive response was seen in 80% of tested donors, with a number of donors showing a hypersaturated response on the FluoroSpot assay indicating strong T cell activation ( Figure 2B , first column, 16 in 20 positive responders). Conversely, little to no IFN-γ response was observed in these PBMCs upon restimulation with any of the MAPPs-derived peptide pools ( Figure 2B , remaining columns). In PBMCs expanded with the total MAPPs peptide pool, only 25% of tested donors displayed a positive IFN-γ response upon restimulation with the overlapping peptide pool ( Figure 2C , first column, 5 in 20 positive responders). In contrast, 55% of tested donors displayed a positive IFN-γ response when restimulated with the total MAPPs pool ( Figure 2C , second column, 11 in 20 positive responders). Interestingly, the restimulation response to the total MAPPs pool did not correlate with the serostatus of the tested donors ( Supplementary Figure 2 ). Furthermore, we observed a distinction in the restimulation response to HLA Class I vs. Class II MAPPs peptides. 40% of tested donors displayed a positive IFN-γ response to the Class I MAPPs pool ( Figure 2C , third column, 8 in 20 positive responders) compared to 20% of tested donors that responded to the Class II MAPPs pool ( Figure 2C , fourth column, 4 in 20 positive responders). Overall, our results suggest that overlapping and MAPPs-derived capsid peptides expand distinct T cell populations in healthy donors and that the response to the total MAPPs pool is dominated by the Class I MAPPs pool, suggesting a preferential expansion of capsid-specific CD8 T cells.

To confirm the prevalence of CD8 T cells in the response to the HLA Class I MAPPs peptides, CD4 T cells were depleted from PBMCs of a responding donor prior to expansion with the Class I MAPPs pool and restimulation with either MAPPs-derived or overlapping peptides. Consistent with our previous observations, no response was observed to the overlapping or the Class II MAPPs pools, irrespective of the presence of CD4 T cells. In contrast, restimulation with Class I MAPPs pool elicited robust IFN-γ release in PBMCs even in the absence of CD4 T cells ( Figure 2D ). Overall, our findings confirm that the response to HLA Class I MAPPs peptides is dominated by CD8 T cells.

To identify the immunodominant CD8 T cell epitopes within the HLA Class I MAPPs peptides, we utilized a system of matrix pools which facilitates rapid mapping of immunodominant epitopes when investigating a large number of peptides. The 41 HLA Class I MAPPs peptides were arranged into 13 matrix pools (MPs), each containing 6–7 peptides ( Supplementary Table 4 ). The matrix arrangement allows each Class I peptide to feature in two MPs, and a restimulation response to both MPs containing a specific peptide identifies putative immunodominant peptides that can be verified on an individual basis. Peptide deconvolution was performed on a cohort of 24 healthy donors. PBMCs were depleted of CD4 T cells and expanded as described before with the Class I MAPPs pool (41 peptides). Expanded PBMCs were restimulated on FluoroSpot assays with the 13 MPs. The putative immunodominant peptides identified through responding MPs were individually analyzed in a separate experiment to confirm their ability to generate a CD8 T cell response.

Figure 3 describes the peptide deconvolution process for one donor. CD4-depleted PBMCs from Donor # 474 were expanded with the Class I MAPPs pool and restimulated on a FluoroSpot assay with 13 MPs. This donor displayed an IFN-γ response to four MPs- MP 3, MP 4, MP 9, and MP 10 ( Figure 3A ). Based on the matrix, we identified the common peptides between the four responding MPs and shortlisted four candidate immunodominant peptides for this donor ( Figure 3B ). In a subsequent experiment, CD4-depleted PBMCs from Donor # 474 were expanded with the Class I MAPPs pool and restimulated on a FluoroSpot assay with the four candidate immunodominant peptides individually. IFN-γ response was observed to Peptide 16 and Peptide 21, thereby verifying their immunodominance in this donor ( Figure 3C ). Supplementary Tables 5, 6 contain the summary of the deconvolution data for all peptides and donors.

Our analysis revealed 10 Class I MAPPs peptides that displayed an IFN-γ response in at least one donor. These included 1 previously reported and 9 novel capsid T cell epitopes ( Figure 4A , Supplementary Table 5 ). Among the novel epitopes, responses were most prevalent to Peptide 21 (7 responders), followed by Peptide 26 and Peptide 41 (2 responders each), with the remaining 6 peptides having one responding donor each ( Figure 4B , third column). The novel peptides display significant diversity in their lengths, varying between 9 and 13 amino acids ( Figure 4B , fourth column). Furthermore, two of the novel peptides- Peptide 21 and Peptide 26- displayed a high percentage of responses in donors expressing their predicted HLA Class I binding alleles (13, 23). Peptide 21 elicited a response in 62.5% of donors with its predicted binding allele (5 responders among 8 donors with the A*02:01 allele) ( Figure 4B , columns 5, 6). It also elicited responses in two donors with the closely related A*02:02 and A*02:05 alleles (Donor # 399, Donor # 440, Supplementary Table 6 ). Similarly, Peptide 26 elicited a response in 100% of donors with its predicted binding allele (2 responders among 2 donors with the A*68:01 allele). The remaining novel peptides, on the other hand, did not display similar trends and had lower response rates in donors with their predicted binding alleles ( Figure 4B , columns 5, 6). We also observed T cell responses to the known epitope Peptide 16 in 4 donors. Given its reported immunodominance, this peptide was individually tested in all donors with its predicted binding allele irrespective of whether or not a response was observed to the corresponding matrix pools. Peptide 16 elicited a response in only 50% of the donors with its predicted binding allele (3 responders among 6 with the B*07:02 allele) ( Figure 4B , row 10).

To further validate the identified CD8 T cell epitopes, we measured the binding stability and affinity of the epitopes to their predicted HLA binding alleles ( Figure 4B , columns 7, 8, Figure 4C ). The majority of the identified AAV9 capsid CD8 T cell epitopes were strong binders (high stability and affinity to their respective alleles). Two novel epitopes- Peptide 10 and Peptide 20- have weak binding to their predicted alleles, suggesting that their presenting alleles were not correctly predicted. The measured binding affinities of most epitopes also correlated strongly with the predicted affinities generated using NetMHC ( Supplementary Table 7 ). The novel 13-mer epitope Peptide 26 elicited a weak NetMHC rank score despite a high measured binding affinity, which is unsurprising given that the NetMHC model is curated for peptides ranging between 8–11 amino acids (24).

In addition to Peptide 16, the HLA Class I MAPPs peptides included 3 other peptides that have been previously reported as immunodominant AAV9 capsid epitopes - Peptide 24, Peptide 7, and Peptide 17 (26, 27). Surprisingly, these peptides did not elicit a response in any tested donors, including the donors with the predicted HLA binding allele ( Table 1 ). Notably, Peptide 24 demonstrated strong binding stability and affinity to its predicted A*02:01 allele ( Table 1 , columns 7, 8) but remained a non-responder despite individual testing in seven different donors with the allele (Donors # 403, # 406, # 474, # 505, # 513, # 519, # 593, Supplementary Table 6 ). This presents a sharp contrast to the novel epitope Peptide 21, which also binds strongly to the A*02:01 allele and elicits responses in multiple donors with the allele as previously described.

Overall, despite comparable binding parameters, many of the identified novel epitopes display stronger abilities to induce the capsid-specific CD8 T cell response compared to previously described epitopes.

Many of the identified HLA Class I MAPPs peptides were organized as peptide clusters, each comprising 2–3 peptides of varying lengths and the same peptide core ( Supplementary Table 8 ). Of these, T cell responses were observed to peptides belonging to 2 clusters. The first cluster contained 9-mer (Peptide 15), 10-mer (Peptide 14), and 11-mer (Peptide 13) peptides that were presented by the same donor on the MAPPs assay ( Supplementary Table 8 , Cluster 3). All three peptides were predicted to bind to the B*40:02 allele, but the rank score threshold for accurate prediction was met only for the 9-mer Peptide 15 (13). Interestingly, despite having a non-significant prediction score, only the 11-mer Peptide 13 elicited a T cell response in a donor expressing the B*40:02 allele ( Figure 5A ). The second cluster contained 9-mer (Peptide 27) and 13-mer (Peptide 26) peptides that were presented by two different donors on the MAPPs assay and had two different predicted binding alleles ( Supplementary Table 8 , Cluster 4). Here, the 13-mer Peptide 26 elicited a T cell response in donors containing its predicted binding allele (A*68:01). Peptide 27, on the other hand, did not elicit a response in any donors with its predicted binding allele C*03:04 ( Figure 5B , Supplementary Figure 3 ). Binding assays indicate stable and high affinity binding of both peptides to their respective predicted or closely related alleles ( Figure 5C ). This indicates that, even with comparable binding stabilities, longer epitopes are more likely to induce T cell activation.

Taken together, our peptide deconvolution experiments identified several immunodominant CD8 T cell epitopes, of which only one was previously reported. A majority of the identified epitopes are novel peptides of varying lengths that display high response rates in donors with the predicted binding alleles. Binding assays further confirmed stable binding and strong affinity of novel epitopes to their predicted binding alleles. Finally, we observed that longer epitopes within the same peptide cluster are preferred for CD8 T cell activation.