The major histocompatibility complex class I immunopeptidome of extracellular vesicles

Extracellular vesicles (EVs) are released by most cell types and have been associated with multiple immunomodulatory functions. MHC class I molecules have crucial roles in antigen presentation and in eliciting immune responses and are known to be incorporated into EVs. However, the MHC class I immunopeptidome of EVs has not been established. Here, using a small-scale immunoisolation of the antigen serotypes HLA-A*02:01 and HLA-B*27:05 expressed on the Epstein-Barr virus-transformed B cell line Jesthom and MS of the eluted peptides from both cells and EVs, we identified 516 peptides that bind either HLA-A*02:01 or HLA-B*27:05. Of importance, the predicted serotype-binding affinities and peptide-anchor motifs did not significantly differ between the peptide pools isolated from cells or EVs, indicating that during EV biogenesis, no obvious editing of the MHC class I immunopeptidome occurs. These results, for the first time, establish EVs as a source of MHC class I peptides that can be used for the study of the immunopeptidome and in the discovery of potential neoantigens for immunotherapies.

Extracellular vesicles (EVs), 2 including those known as exosomes, are released by many types of cells, and have been ascribed numerous functions in cellular communication, including the propagation of niche sites for tumor metastases (1), facilitating viral infection (2), and in angiogenesis (3), among many others. EVs are thought to originate from multivesicular bodies and are secreted following fusion of multivesicular bodies (MVB) with the plasma membrane (4,5). Within the immune system EVs have been reported to have mostly immunoinhibitory functions (6), although dendritic cell-and tumor cell-derived EVs have shown some promise in the stimulation of anti-tumor responses (7). EVs have also been demonstrated to have a role in transplantation responses, leading to a reassessment of the passenger leukocyte hypothesis (8) and have been demonstrated to influence thymic selection (9). These physiological and therapeutic roles in the immune sys-tem imply that peptide presentation by major histocompatibility complex (MHC) molecules on EVs have a profound effect on their role in eliciting immune responses. To date, no information exists on the peptide repertoire that is presented by EVs. MHC expression and peptide presentation by EVs could be an important determining factor for their general effectiveness, as they could modulate EV-immune cell interactions, which potentially may affect their therapeutic value.
In relationship to the MHC class I system proteins and their role as antigen presenting molecules to CD8 ϩ T cells, we have previously shown that MHC class I molecules are incorporated into EVs from a number of cell lines, including the Jesthom line used in the current study (10 -12). However, their ability to modulate CD8 ϩ activity has not been studied in detail, although EVs loaded exogenously with common immunogenic viral peptides can stimulate IFN␥ and TNF␣ release from purified human CD8 ϩ T cells, indicating EVs can directly impact on the behavior of these cells (13).
To further understand the role of MHC class I on EVs we have undertaken here to determine their immunopeptidome, i.e. the repertoire of the peptides being presented. The null hypothesis would be that the EV pool of MHC class I-bound peptides would be an exact copy of those present on the cell surface. However, because EVs can be derived from an endocytic route involving the formation of MVB, editing of the peptide repertoire could occur, by dissociation of weak affinity peptides and/or loading of peptides from within the MVB compartment, thus generating potential EV-associated neoantigens. This could be of significant importance in the search for novel MHC class I epitopes and development of patient-tailored immunotherapies.
In this study we have performed small-scale MHC class I immunoisolation and peptide extraction to determine the repertoire of peptides from EVs of the EBV-transformed B cell line Jesthom, providing the first data that EVs can be used as a source for immunopeptidomic studies of MHC class I epitopes in health and disease.

cro ARTICLE
were processed concomitantly to prevent temporal sample variations due to dissociation of low affinity peptides. EVs were isolated by standard procedures of filtration and ultracentrifugation. Immunoblot analysis of a sample of the cell and EV lysates indicated enrichment of MHC class I, and the prototypical EV markers CD9 and CD81 in the EV isolates, whereas the non-EV marker GRP78 (BiP) was present at very low levels in EVs (Fig. 1A). From a series of cultures with ϳ10 million Jesthom cells, the enrichment of HLA-A molecules and HLA-B molecules in EV was between 1.8-and 2-fold (supplemental Fig.  S1). Nanoparticle tracking analysis (NTA), which utilizes light scattering of particles in suspension undergoing Brownian motion, of culture supernatant after 0.2-m filtration, but prior to ultracentrifugation indicated three typical peaks of around 61, 86, and 142 nm (Fig. 1B), with a mean of 151 nm and mode of 115 nm from three preparations. The samples therefore represent typical EVs. From a separate test run of 400 million Jesthom cells, an estimated concentration of 400 ϫ 10 9 particles were detected by NTA, indicating the release of ϳ1,000 EV per cell over the course of 48 h.
The cell and EV lysates were immunoisolated with the anti-HLA-A, -B, and -C mAb W6/32, pre-absorbed (but not crosslinked) to Protein G-agarose beads. After extensive washing, 2% of the beads were removed and analyzed by SDS-PAGE and Coomassie Blue staining or immunoblotting with the anti-HLA-B and -C specific mAb HC10. Compared with the control W6/32 (no lysate) sample additional bands at ϳ43 kDa were detected in both cell and EV samples (Fig. 1C), which were HC10 reactive (Fig. 1D). Additional bands were detected migrating at the dye front, which were presumed to be ␤2-microglobulin (Fig. 1C). The remaining 98% of the immunoisolated cell and EV samples were acidified in 0.5% TFA to denature the MHC class I molecules, and the low molecular mass peptide pool fraction isolated with Centricon 3 centrifugal filters. The isolated peptide fraction was then processed and analyzed by mass spectrometry.
Identified peptides from the resulting MASCOT files between 8 and 13 amino acids long were analyzed for HLA-A*02:01 and HLA-B*27:05 predicted binding affinity using the NetMHCcons 1.1 server, which combines three algorithms (NetMHC, NetMHCpan, and PickPocket). From the combined data of the three biological replicates 145 and 94 peptides were identified from HLA-B*27:05 cell and EV preparations, respectively, and 172 and 105 peptides from HLA-A*02:01 cell and EV preparations, respectively (Table 1 and supplemental Table S1). In addition 11 HLA-C*01:02 (also expressed by Jesthom cells) peptides were also identified but are not reported here due to low numbers. The mean predicted binding affinity of the HLA-A*02:01-binding peptide pool was 33.3 nM for cells and 26.7 nM for EVs ( Fig. 2A), and for the HLA-B*27:05-binding peptide pool the mean predicted affinity was 225.9 nM for cells and 199.5 nM for EVs (Fig. 2C). Although this might suggest that there is a loss of some lower affinity peptides during the biogenesis of EVs, two-tailed Mann-Whitney tests indicated no significant differences between the cell and EV pools for either MHC class I allele (HLA-A*02:01, p ϭ 0.8329, and HLA-B*27:05, p ϭ 0.3199, respectively). Peptide lengths did not alter, with a predominance of 9-mer peptides in both cells and EV for both HLA-A*02:01 and HLA-B*27:05 (Fig. 2, B and D). The anchor binding motifs were also analyzed by Seq2Logo for 9-, 10-, and 11-mer peptide pools from each source, with no significant alterations between the typical P2 Leu and C-terminal Val/Leu anchors for HLA-A*02:01 (Fig. 2E) and the P2 Arg and C-terminal Phe/Tyr anchors for HLA-B*27:05 (Fig. 2F). 26 of the 94 HLA-B*27:05 peptides in the EV pool were not detected in the cell-derived pool, and 34 of the 105 HLA-A*02:01 peptides from EVs were not detected in the cell-derived pool (highlighted in red in Table 1), however, the cellular origin of these peptides was mostly from the cytoplasm and nucleus, suggesting they would also appear in the cell-derived pool in a larger sample size. One HLA-A*02:01-binding peptide (LLLD-VPTAAV) was identified from the endosome-located thiol-reductase GILT, but this has previously been reported in cells and therefore unlikely to be EV-specific (14). Taken together the data indicate that the MHC class I immunopeptidome of EVs is a replica of that found on the cell surface.

Discussion
Our data has several important implications. It demonstrates that the EV immunopeptidome is essentially identical to that of the cell of origin. As such, important antigenic peptides, such as viral or tumor-associated antigens and tumor-specific antigens  Table 1 Peptide

identification and predicted binding affinities
The sequence of the identified peptides for cell and EV derived peptides eluted from HLA-A*02:01 and HLA-B*27:05 are shown in alphabetical order. Peptides in red are those detected in the EV pool only. Predicted binding affinities were generated using NetMHCcons 1.1.

The immunopeptidome of extracellular vesicles
are likely to be released in EVs, potentially subverting antigenspecific CD8 ϩ cytotoxic T cells at a distance from the infected cell or main tumor, thus potentially reducing effective CTL responses. The same observation would, however, also imply that EVs can be used as an effective source to isolate and detect tumor-specific antigens and tumor-associated antigens from

The immunopeptidome of extracellular vesicles
readily available biological samples such as blood. EVs are known to be raised in pathological conditions (15,16), suggesting a relatively non-invasive technique for screening. As such, the EV-derived peptidome can now be studied as a source for neoantigens for personalized immunotherapeutic approaches, as recently demonstrated in principle for melanoma solid tissue biopsies (17). Such identified peptides could then be utilized in dendritic cell exosome-based therapies (18), for which phase I and II trials have already been conducted. Furthermore, this MS technique could be used to monitor the efficacy of target-peptide loading onto dendritic cell exosomes.
Our study does have some limitations. Our current smallscale study yielded a few hundred peptides, but larger samples and improved detection could yield thousands of identified peptides. The small, but consistent alteration in predicted binding affinities in the EV peptide pool would be worth studying in greater detail with such larger sample sizes. The residency time of an MHC class I complex during its incorporation into and secretion via an EV would be expected to promote the loss of low-affinity peptides. Larger sample sizes would help resolve this issue. Furthermore, variations in the biogenesis of MVBs in different cell types could also have a significant impact upon the EV immunopeptidome. An extensive study of multiple cell types is now required. Of technical interest, the antibody-based immunoisolation of MHC molecules for peptide isolation utilized here is just one of several possible techniques (19). We have also performed preliminary studies on EV samples using mild acid elution (MAE), which in theory would not disrupt the cell or EV samples. The MAE technique removes the detergent lysis, immunoisolation, and extensive washing steps that could lead to loss of low-affinity peptides. However, MAE is known to produce increased peptide signals of non-MHC origin (20), but intriguingly we were able to detect some peptides with known binding motifs for the MHC class II molecules expressed on Jesthom cells (data not shown), thus further enhancing the capacity of EVs to produce useful immunopeptidomic information. Taken together our study opens a new avenue for the characterization of the immunopeptidome from highly biologically relevant vesicles.

Cell and EV isolation
The EBV-transformed B cell line (obtained from the European Collection of Authenticated cell Lines no. 88052004) was grown in RPMI 1640 (Invitrogen, UK) supplemented with 5% FBS (Invitrogen, UK). Once the required number of cells was obtained, the medium was replaced with serum-free medium to prevent contamination from FBS-derived exosomes (EX-Cell 610-HSF serum-free, Sigma-Aldrich, UK) for 48 h. Cells were then isolated by centrifugation (300 ϫ g, 10 min). The cells were washed once in PBS, then immediately resuspended in 5 ml of lysis buffer (1% Nonidet P-40, 150 mM NaCl, 10 mM Tris, pH 7.6, with 1 mM PMSF). After 10 min on ice the lysates were centrifuged at 20,000 ϫ g for 5 min and the supernatant stored on ice. 10 -20 l was removed for immunoblotting.
The EV containing supernatant was processed immediately by 0.2-m filtration and ultracentifugation at 100,000 ϫ g for 2 h. The pellets were resuspended in 500 l of PBS and 10 -20 l was removed for BCA protein estimation or EV characterization by immunoblotting. The remaining bulk EV suspension was immediately lysed in 5 ml of lysis buffer, as above.

Immunoisolation of MHC class I peptides
0.5 ml of Protein G-agarose (code 20399, binding capacity 11-15 mg/ml of IgG, Thermo Scientific UK) were pre-loaded with 30 ml of W6/32 containing tissue culture supernatant for 20 min at room temperature, then washed twice in lysis buffer. The beads were then added to the cell and EV lysates and mixed for 1 h at 4°C. Control W6/32-loaded beads received lysis buffer alone. The beads were then washed with 60 volumes (3 ϫ 10 ml) of lysis buffer without Nonidet P-40. 10 l of beads was then removed for reducing SDS-PAGE and Coomassie Blue staining (Gelcode Blue, Thermo Scientific, UK). The remaining beads were resuspended in 1 ml of 0.5% TFA for 10 min at room temperature. The supernatant was then spun at 12,000 ϫ g through pre-washed (in 0.1% TFA) Centricon 3 filtration units (MerckMillipore, UK). The peptide containing flow-through was then stored at Ϫ20°C until analysis by mass spectrometry.

Mass spectrometry
Peptides were concentrated using a C18 column (NEST, Thermo Scientific UK), eluted in 70% acetonitrile, 0.5% TFA and dried down by SpeedVac. Peptides were then analyzed on an AB Sciex TripleTOF 5600ϩ system mass spectrometer (Sciex, Framingham, MA) coupled to an Eksigent nanoLC AS-2/2Dplus system. The samples were loaded in loading buffer (2% acetonitrile and 0.05% trifluoroacetic acid) and bound to an Aclaim pepmap 100 m ϫ 2-cm trap (Thermo Fisher Scientific), and washed for 10 min to waste after which the trap was turned in-line with the analytical column (Aclaim pepmap RSLC 75 m ϫ 15 cm). The analytical solvent system consisted of buffer A (2% acetonitrile and 0.1% formic acid in water) and buffer B (2% water with 0.1% formic acid in acetonitrile) at a flow rate of 300 nl/min with the following gradient: linear 1-20% of buffer B over 90 min, linear 20 -40% of buffer B for 30 min, linear 40 -99% of buffer B for 10 min, isocratic 99% of buffer B for 5 min, linear 99 -1% of buffer B for 2.5 min, and The immunopeptidome of extracellular vesicles isocratic 1% solvent buffer B for 12.5 min. The mass spectrometer was operated in the DDA top 20 positive ion mode, with 120 and 80 ms acquisition time for the MS1 (m/z 400 -1250) and MS2 (m/z 95-1800) scans, respectively, and 15-s dynamic exclusion. Rolling collision energy was used for fragmentation. Peak lists were generated within PeakView by using the "create mgf file" script. The MASCOT search engine with the following search parameters was used to identify peptides: no enzyme specificity, maximum of 4 miscleavages, oxidation as variable modification, peptide tolerance was set to 20 ppm, and the MS-MS tolerance to 0.1 Da. Data were searched against Swiss Prot database downloaded November 2016, restricted to proteins from humans only.