Immunopeptidomics-Guided Warehouse Design for Peptide-Based Immunotherapy in Chronic Lymphocytic Leukemia

Antigen-specific immunotherapies, in particular peptide vaccines, depend on the recognition of naturally presented antigens derived from mutated and unmutated gene products on human leukocyte antigens, and represent a promising low-side-effect concept for cancer treatment. So far, the broad application of peptide vaccines in cancer patients is hampered by challenges of time- and cost-intensive personalized vaccine design, and the lack of neoepitopes from tumor-specific mutations, especially in low-mutational burden malignancies. In this study, we developed an immunopeptidome-guided workflow for the design of tumor-associated off-the-shelf peptide warehouses for broadly applicable personalized therapeutics. Comparative mass spectrometry-based immunopeptidome analyses of primary chronic lymphocytic leukemia (CLL) samples, as representative example of low-mutational burden tumor entities, and a dataset of benign tissue samples enabled the identification of high-frequent non-mutated CLL-associated antigens. These antigens were further shown to be recognized by pre-existing and de novo induced T cells in CLL patients and healthy volunteers, and were evaluated as pre-manufactured warehouse for the construction of personalized multi-peptide vaccines in a first clinical trial for CLL (NCT04688385). This workflow for the design of peptide warehouses is easily transferable to other tumor entities and can provide the foundation for the development of broad personalized T cell-based immunotherapy approaches.


INTRODUCTION
The breakthrough clinical success of T cell-based immunotherapy approaches, including immune checkpoint inhibitors, CAR T cells, bispecific antibodies, and adoptive T cell transfer, has recently revolutionized the treatment of solid tumors and hematological malignancies. However, some patients do not respond to available therapies at all, others only temporarily calling for further efficacy improvement of T cell-centered immunotherapies. Peptide-based approaches, which rely on the specific immune recognition of tumor-associated human leukocyte antigen (HLA)-presented peptides, represent promising alternatives with low side effects. Targeting mutation-derived neoepitopes by vaccination in melanomaa high-mutational burden tumor entityhas demonstrated immunogenicity and first clinical efficacy (1). However, only a minority of mutations at the DNA level is translated and naturally processed to HLA-presented neoepitopes targetable for T cells (2,3). Non-mutated tumor-associated antigens, arising through differential gene expression or protein processing in tumor cells, might supplement or even substitute neoepitope targeting in low-mutational burden malignancies. Whereas a huge number of clinical studies have shown immune responses, including strong CD8 + T cell responses, to vaccines targeting non-mutated tumor antigens, so far, all performed trials failed to show meaningful clinical results (4)(5)(6)(7)(8)(9)(10). Nonetheless, several studies showed a pathophysiological relevance of these tumor antigens, correlating spontaneous, pre-existing as well as immune checkpoint inhibitor-induced T cell responses, targeting non-mutated tumor antigens with improved clinical outcome of malignancies (11)(12)(13)(14). Therefore, the usage of novel technologies and methods, to unravel and resolve the underlying issues and limitations of non-mutated antigen-based vaccines, might enable that T cell responses, induced or boosted by these vaccines, may result in clinical effectiveness in the future (15).
One issue of former peptide-vaccine trials was the selection of non-mutated tumor antigens that were not proven to be naturally presented on the tumor cell surface of the individual patients. For multiple of the applied "classical" tumor antigens novel analyses showed a distorted correlation of tumorassociated presentation on mRNA level, and limited or even lacking presentation on cell surface HLA (16)(17)(18)(19)(20)(21)(22).
In recent years, characterization of naturally presented mutated and non-mutated tumor antigens was refined using direct, mass spectrometry-based analyses of the entirety of HLA ligands termed the immunopeptidome (13,(20)(21)(22)(23)(24)(25). While mass spectrometric identification of neoepitopes remains rare with inter-individual heterogeneity and requires a time-and costintensive fully personalized vaccine design (1), distinct panels of high-frequent non-mutated tumor-associated antigens have been previously identified (13,(20)(21)(22)(23)(24). Such panels could enable the construction of off-the-shelf pre-manufactured peptide warehouses for the design of broadly applicable personalized therapeutics, such as multi-peptide vaccines or products for adoptive T cell transfer, that could be assembled based on patient-individual characteristics. To evaluate this concept in chronic lymphocytic leukemia (CLL) as representative example of an immunogenic, low-mutational burden malignancy (26,27), we here developed a widely applicable workflow for the immunopeptidomics-guided design of such a warehouse, that provides the basis for the selection of personalized multi-peptide vaccine cocktails for a first-in-man clinical trial (NCT04688385).

Patients and Blood Samples
Peripheral blood mononuclear cells (PBMCs) from CLL patients as well as PBMCs from healthy volunteers (HVs) were isolated by density gradient centrifugation and stored at -80°C until further use for subsequent HLA immunoprecipitation or T cell-based assays. For immunopeptidome analysis, PBMCs from CLL patients with white blood cell counts ≥ 20 000 per µl were used. Informed consent was obtained in accordance with the Declaration of Helsinki protocol.

Isolation of HLA Ligands
HLA class I and HLA class II molecules were isolated by standard immunoaffinity purification (28) using the pan-HLA class Ispecific monoclonal antibody W6/32, the pan-HLA class IIspecific monoclonal antibody Tü-39, and the HLA-DR-specific monoclonal antibody L243 (all produced in-house) to extract HLA ligands.

Analysis of HLA Ligands by Liquid Chromatography-Coupled Tandem Mass Spectrometry (LC-MS/MS)
Isolated peptide samples were analyzed as described previously (13,29). Peptides were separated by nanoflow high-performance liquid chromatography. Eluted peptides were analyzed in an online-coupled LTQ Orbitrap XL or Orbitrap Fusion Lumos mass spectrometer (both Thermo Fisher, Waltham, Massachusetts, USA).

Data Processing
Data processing was performed as described previously (13,29 Table 3) as described in the literature (31,32) and assigned in the COSMIC database (http://cancer.sanger.ac. uk) (33) without enzymatic restriction. Precursor mass tolerance was set to 5 ppm. Fragment mass tolerance was set to 0.5 Da for ion trap spectra (LTQ Orbitrap XL) and 0.02 Da for orbitrap spectra (Orbitrap Fusion Lumos). Oxidized methionine was allowed as dynamic modification. The false discovery rate (FDR) estimated by the Percolator algorithm 2.04 (34) was limited to 5% for HLA class I-and 1% for HLA class II-presented peptides. HLA class I annotation was performed using SYFPEITHI 1.0 (35) and NetMHCpan 4.0 (36,37).

Peptide Synthesis
Peptides were produced by the peptide synthesizer Liberty Blue (CEM, Kamp-Lintfort, Germany) using the 9-fluorenylmethyloxycarbonyl/tert-butyl strategy (38). Peptides for the peptide warehouse were selected with regard to a subsequent good manufacturing practice (GMP) and manufacturing license conform production process, including peptide length restriction of 20 amino acids, avoiding cysteine-containing peptides as well as peptides with histidine or proline at the C-terminus.

Spectrum Validation
Spectrum validation of the experimentally eluted peptides was performed by computing the similarity of the spectra with corresponding synthetic peptides measured in a complex matrix. The spectral correlation was calculated between the MS/MS spectra of the eluted and the synthetic peptide (39).

Cytokine and Tetramer Staining
The functionality of peptide-specific CD8 + T cells was analyzed by intracellular cytokine staining (ICS) as described previously (40,43). Cells were pulsed with 10 µg/ml of individual peptide and incubated with FITC anti-human CD107a (BioLegend, San Diego, CA, USA) for one hour before incubating with 10 mg/ml Brefeldin A (Sigma-Aldrich, Saint Louis, MO, USA) and 10 mg/ ml GolgiStop (BD, Franklin Lakes, NJ, USA) for 12-16 h. Staining was performed using aqua fluorescent reactive dye (Invitrogen), Cytofix/Cytoperm (BD), PE-Cy7 anti-human CD8 (Beckman Coulter, Brea, CA, USA), Pacific Blue antihuman tumor necrosis factor (TNF), and PE anti-human IFN-g (Biolegend) monoclonal antibodies. PMA and ionomycin (Sigma-Aldrich) served as positive control. Negative control peptides were used as described for the ELISpot assays. The frequency of peptide-specific CD8 + T cells after aAPC-based priming was determined by PE-Cy7 anti-human CD8 monoclonal antibody and HLA:peptide tetramer-PE staining. Cells of the same donor primed with an irrelevant control peptide and stained with the tetramer containing the test peptide were used as negative control. The priming was considered successful if the frequency of peptide-specific CD8 + T cells was ≥ 0.1% of CD8 + T cells within the viable single cell population and at least three-fold higher than the frequency of peptide-specific CD8 + T cells in the negative control. The same evaluation criteria were applied for ICS results. Samples were analyzed on a FACS Canto II cytometer (BD).

Software and Statistical Analysis
An in-house Python script was used for the calculation of FDRs of CLL-associated peptides at different presentation frequencies (13). Overlap analysis was performed using BioVenn (44). The population coverage of HLA allotypes was calculated by the IEDB population coverage tool (www.iedb.org) (45,46). Flow cytometric data was analyzed using FlowJo 10.0.8 (Treestar, Ashland, Oregon, USA). All figures and statistical analyses were generated using GraphPad Prism 9.0.2 (GraphPad Software, San Diego, CA, USA).

Mass Spectrometry-Based Identification of Naturally Presented CLL-Associated Antigens
In the first step of our workflow for the definition of an off-theshelf peptide warehouse ( Figure 1A), we comprehensively mapped the immunopeptidome of 61 primary CLL samples (n = 52 for HLA class I, n = 49 for HLA class II, Supplementary Table 1). We identified a total of 58 554 unique HLA class I ligands representing 10 854 source proteins, obtaining 96% of the estimated maximum attainable source protein coverage ( Figure 1B). The number of identified unique peptides per patient ranged from 527 to 9 530 (mean 3 035, Supplementary  Figure 1A).

The Role of Neoantigens in the Immunopeptidome of CLL
In addition to the identification of high-frequent non-mutated tumorassociated peptides, we screened our CLL cohort for naturally presented neoepitopes derived from common CLL-associated point and frameshift mutations (95 point and 3 frameshift mutations within 61 different genes representing 85 different mutation sites, Supplementary Table 3). Even though these mutations potentially provide HLA binding motifs for several HLA allotypes, no naturally HLA-presented neoepitopes were identified in our immunopeptidome analyses. Of note, we were able to identify wild-type peptides derived from 57/61 (93%) mutation-bearing proteins in benign and/or CLL immunopeptidomes with an average of 15 and 11 HLA class I-and HLA class II-presented peptides per protein, respectively. However, only 17/85 (20%) mutation sites are covered directly by wild-type peptides as most of the recurrent CLL-associated mutations are located in "dark spots" of the immunopeptidome, defined as protein regions without any detectable HLA-presented peptides (Figures 2A, B).
To enable the peptide warehouse production for clinical application, further selection steps, including approval of producibility, solubility, and stability of target antigens under GMP conditions, delineated a 14-peptide panel ( Table 1) comprising 9 HLA class I-(3 for each allotype) and 5 HLA class II-restricted CLL-exclusive high-frequent target antigens. Experimentally acquired spectra of the selected peptide identifications were validated by comparison of mass spectrometric fragment spectra using isotope-labeled synthetic peptides ( Figures 2C, D and Supplementary Figure 3).

Warehouse Peptides Show Pre-Existing and De Novo Inducible Immune Responses in CLL Patients and HVs
In the final step, selected peptide targets were analyzed for their immunogenicity, i.e., their potential to induce antigen-specific T cell responses. Using aAPC-based in vitro priming of naive CD8 + T cells of HLA-matched HVs, we confirmed induction of peptidespecific CD8 + T cells for all 9 HLA class I peptides in at least 2/3 HVs ( Figure 3A, Supplementary Figures 4A, B and Table 1). Intracellular cytokine and degranulation marker staining revealed induction of multifunctional peptide-specific T cells for 7/9 (78%) peptides ( Figure 3B, Supplementary Figures 4C, D and Table 1). Moreover, IFN-g ELISpot assays, using PBMCs from HLAmatched CLL patients, revealed preexisting peptide-specific memory T cells targeting 2/9 (22%) and 4/5 (80%) HLA class Iand HLA class II-restricted warehouse peptides in up to 29% and 50% of CLL patient samples, respectively (Figures 3C, D, Supplementary Figure 5 and Table 1). In total, we validated 13/14 (93%) of the preselected naturally presented CLL-associated HLA class I-and HLA class II-restricted peptides as immunogenic, with either pre-existing or de novo inducible immune responses, unveiling these as ideal targets for peptidebased immunotherapy approaches.

DISCUSSION
We present a novel workflow for the immunopeptidomics-guided design of off-the-shelf peptide warehouses, here applied to CLL as a representative of low-mutational burden tumor entities. The mass spectrometry-based characterization of high-frequent, nonmutated, tumor-associated, and immunogenic antigens naturally presented on primary CLL samples enables the development of a personalized peptide-based immunotherapy in a time-and cost-effective manner, and can be easily transferred to other tumor entities.
The treatment landscape for CLL has faced profound changes with the development and clinical success of targeted therapies in the past years (47,48), including the Bruton's tyrosine kinase (BTK) inhibitor ibrutinib (49). However, beside the combination of anti-CD20 antibodies and venetoclax, all novel substances  require continuous treatment bearing the risk of therapy resistance and accumulation of side effects. Current efforts are now focusing on the further and earlier elimination of minimal residual disease (MRD) to allow for reduced treatment duration and the therewith associated side effects, as well as the achievement of long-lasting remission and potential cure in the future. The favorable immune effector-to-target cell ratios in the MRD setting and the immunogenicity of CLL (26,27) suggest that this malignancy might be effectively targeted by peptide-based immunotherapy.
For the clinical application of such approaches three different strategies have been proposed. (i) Stratification, applying an invariant drug product to every patient, seems unsuitable due to the patient-individual HLA allotypes. Therefore, such an approach must focus on very common HLA allotypes thereby excluding a substantial proportion of patients. In addition, as shown by us and others, even in allotype-matched patients of the same entity, presentation of tumor-associated peptides is not given in 100% of tumors, calling for more personalized approaches of target selection (21,50). However, the broad applicability of (ii) completely individualized peptide vaccine concepts is hampered, since these are based on time-and cost-intensive patient-specific identification and selection processes and on-demand de novo drug production (1,5,51). The approach of (iii) peptide warehouse design, comprising a collection of pre-defined and premanufactured high-frequent tumor-associated peptides, enables the subsequent composition of patient-specific vaccine cocktails based on individual characteristics (5,10,52). In recent years, multiple peptide vaccination approaches were focused on the targeting of neoepitopes from tumor mutations as prime tumor-specific targets. For tumor entities with high mutational burden, such as melanoma, high immunogenicity and first signs of clinical efficacy have been demonstrated (1,53). However, only a minority of mutations at the DNA sequence level is translated and processed to naturally presented HLArestricted neoepitopes that can be targeted by T cells (2,16,54,55). This is in line with our data demonstrating the lack of naturally presented neoantigens in the immunopeptidome of CLL. Together with recent immunopeptidomic studies (16,21,56) we showed that HLA-presented peptides are not randomly distributed across protein sequences but rather show "hotspot" locations of presentation. Most of the recurrent CLL-associated mutations are located in "dark spots" of the immunopeptidome, defined as protein regions without any detectable HLA-presented peptides, explaining the rare detection of neoepitopes especially in low-mutational burden entities. The underlying reasons for the occurrence of such hotspots and corresponding dark spots still remain ambiguous but might include differential proteasomal cleavage, peptide processing, and HLA-binding (16,56,57). Thus, the role of neoantigen-based T cell responses in tumor entities with lowmutational burden remains obscure, calling for the application of alternative targets in peptide-based immunotherapy approaches (1)(2)(3). Non-mutated tumor peptides, arising through altered gene expression or protein processing in the tumor cells have been suggested as vaccine targets for many years (4-10, 13, 58, 59). However, although numerous clinical trials reported peptidespecific T cell induction upon vaccination with non-mutated tumor antigens, no correlation with clinical activity was shown and no meaningful clinical results were achieved (4-10). Nevertheless, there are two main points that prompt us and others, despite former disappointing clinical data, to use novel technologies and methods to unravel and resolve the underlying issues and limitations of non-mutated antigen-based vaccines: (i) Several studies report on the existence of spontaneous preexisting T cell responses targeting non-mutated tumor antigens and their correlation with beneficial clinical outcome, suggesting a pathophysiological relevance of these immune responses in vivo (11)(12)(13)(14) and (ii) recent data show that immune checkpoint inhibitor-mediated T cell responses are not only targeting neoepitopes but also non-mutated tumor antigens (60)(61)(62).
Several unmet issues that hamper the development of effective peptide vaccines have been identified in recent years and need to be considered and addressed during future peptide vaccine design. These include target antigen selection, time points of application, and selection of combinatorial drugs (15,63). Within this study we aimed to address one obvious and often discussed issue of former vaccination trials, where peptide selection included non-mutated tumor antigens that were never proven to be naturally presented on the tumor cell surface of the individual patients. For multiple of the applied "classical" tumor antigens novel analyses showed a distorted correlation of tumor-associated presentation on mRNA level, and limited or even lacking presentation on the immunopeptidome level (16)(17)(18)(19)(20)(21)(22), highlighting that the immunopeptidome is an independent complex layer formed by the antigen presentation machinery, and does not necessarily mirror the transcriptome or proteome. Therefore, it is essential to use direct methods of peptide target identification (64). This can be realized by mass spectrometry-based analysis of the entirety of naturally presented HLA ligands, termed the HLA ligandome or immunopeptidome of cancer cells (65). In recent years, we and others worked intensively on the characterization of such naturally presented tumor-associated peptides based on the direct isolation of HLA class I-and class IIpresented ligands from tumor cells and the subsequent identification by mass spectrometry (21-24, 66, 67). Our approach allowed the identification of distinct panels of highfrequent non-mutated tumor peptides across multiple donors.
In addition to cytotoxic CD8 + T cells, CD4 + T cells play important direct and indirect roles in anti-cancer immunity (68) and therefore are indispensable for vaccination approaches. Consequently, the here described workflow comprises target selection of multiple HLA class I-and HLA class II-presented peptides to prevent antigen loss and reduce the risk of immune escape, which often occurs under therapeutic pressure (69,70). Furthermore, the promiscuous binding motifs of HLA class II molecules enable a broad allotype-independent application of these peptides.
Moreover, the detection of preexisting memory T cell responses targeting our warehouse peptides underscores the pathophysiological relevance of our selected antigen targets for immune surveillance in CLL.
Further limitations of peptide vaccines in general, and in particular of non-mutated antigen vaccine peptides, comprise tumor evasion, immune-editing and immune cell exhaustion (71). Increased numbers of regulatory T cells (T reg ) have been associated with a worse outcome of vaccination, both in mice and patients (72,73). In addition, peripheral tolerance limits the available T cell repertoire capable of recognizing cancer cells with high affinity. T cell recruitment is often impaired by an immunosuppressive tumor environment (74) and the aberrant tumor vasculature actively suppresses the access of cancer-specific T cells (75), which limits therapeutic vaccine efficacy. Therefore, combinatorial approaches to further improve vaccine-induced effects are currently being investigated. These include strategies to deplete or modulate T regs (7,(76)(77)(78)(79)(80) as well as the development of modified, so-called heteroclitic, vaccine peptides to enhance low-affinity T cells (81,82). Furthermore, to overcome limited T cell function and recruitment, combinatorial approaches with immune checkpoint inhibitors (83,84) as well as with direct or indirect microenvironment modifiers, such as MEK or PARP inhibitors as well as VEGF-targeting antibody-or inhibitorbased therapies, are already evaluated (85). The importance of a rational selection of combinatorial drugs was recently demonstrated in a phase III peptide vaccination study (86), which failed to confirm the vaccine-induced immune responses and clinical outcome reported in the preceding phase II trial (7). Here, the combination of the peptide vaccine with the tyrosine kinase inhibitor sunitinib, for which a negative impact on T cell responses was described (87), was suggested as an underlying reason. In contrast, a supporting and positive effect on T cell functionality was proven for other tyrosine kinase inhibitors, such as ibrutinib (88)(89)(90), suggesting this BTK inhibitor as a potential combination drug for peptide vaccines in CLL patients.
Together, this study presents a mass spectrometry-based workflow for the design of an immunopeptidome-derived offthe-shelf CLL-associated peptide warehouse, which is currently being evaluated within a first personalized multi-peptide vaccine trial in combination with the novel adjuvant XS15 (91) in CLL patients under ibrutinib treatment (iVAC-XS15-CLL01, NCT04688385). Integrating next generation developments and insights, in terms of antigen selection, interaction of tumor cells with the immune system and rational selection of combination therapies, we aim to contribute a further step on the way to clinically effective peptide vaccinations with this study. Our warehouse design concept is further easily transferable to other tumor entities, enabling the construction of broadly applicable peptide warehouses, which provide the foundation for the development of time-and cost-effective personalized T cellbased immunotherapy approaches.

DATA AVAILABILITY STATEMENT
The mass spectrometry data have been deposited to the ProteomeXchange Consortium (http://proteomecentral. proteomexchange.org) via the PRIDE (92) partner repository with the dataset identifier PXD024871.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ethics committee of the University Hospital Tübingen. The patients/participants provided their written informed consent to participate in this study.