Validation of a high-performance liquid chromatography-tandem mass spectrometry immunopeptidomics assay for the identification of HLA class I ligands suitable for pharmaceutical therapies

For more than two decades naturally presented, human leukocyte antigen (HLA)-restricted peptides (immunopeptidome) have been eluted and sequenced using liquid chromatography-tandem mass spectrometry (LC-MS/MS). Since, identified disease-associated HLA ligands have been characterized and evaluated as potential active substances. Treatments based on HLA-presented peptides have shown promising results in clinical application as personalized T cell-based immunotherapy. Peptide vaccination cocktails are produced as investigational medicinal products under GMP conditions. In order to support clinical trials based on HLA-presented tumor-associated antigens, in this study the sensitive LC-MS/MS HLA class I antigen identification pipeline was fully validated for our technical equipment according to the current US Food and Drug Administration (FDA) and European Medicines Agency (EMA) guidelines. The immunopeptidomes of JY cells with or without spiked-in, isotope labeled peptides, of peripheral blood mononuclear cells of healthy volunteers as well as a chronic lymphocytic leukemia and a bladder cancer sample were reliably identified using a data-dependent acquisition method. As the LC-MS/MS pipeline is used for identification purposes, the validation parameters include accuracy, precision, specificity, limit of detection and robustness.


Introduction
The immunopeptidome is a vast and diverse compilation of HLA-presented peptides (HLA ligands), which serve as a showcase of inter-and intracellular processes. T cells recognize presented peptides in the immunopeptidome, which is constantly modulated by gene expression, transcription, translation, posttranslational modification, and antigen processing and presentation [1][2][3][4] . Especially in tumor immunology, HLA ligands are used in many ways. They are suited as biomarkers, presenting intracellular abnormalities like malignant transformation and as active pharmaceuticals, activating cancer specific T cells 5 .
Natural HLA ligands have been isolated and sequenced using LC-MS/MS for almost three decades [6][7][8][9][10][11][12] . So far, the LC-MS/MS analysis is the only method to investigate the entirety of HLA presented peptides. However, based on these peptide data in silico prediction tools have been developed, which allow the prediction of possibly presented peptides from exome, RNA or whole genome sequencing data and have extended the toolbox even further [13][14][15][16][17][18] .
Developing from such identifications, peptide vaccination cocktails have been produced as active pharmaceuticals under GMP conditions 19 . The acceptance, safety and efficacy of peptide vaccinations has been investigated 19 and several clinical studies testing peptide vaccinations have been performed with our contribution (GAPVAC 5 , NCT02149225 and NOA-16, NCT02454634) or are ongoing (iVAC-CLL01, NCT02802943). The procedures of active substance production, analysis, and batch release have been validated and reliably lead to reproducible products of desired quality. However, the initial antigen identification procedure using mass spectrometrybased immunopeptidomics has not been validated yet. In this study, the LC-MS/MS antigen identification procedure was fully validated for our technical equipment according to current FDA and EMA guidelines to support further clinical trials based on HLA-presented tumor-associated antigens. This validation should serve as a guidance that can be adapted to other LC-MS/MS platforms and samples.
Protocols for large-scale immunopeptidomics using LC-MS/MS and the identification of HLA ligands are established and have been published [20][21][22] . To our knowledge, no validation of an omics method using LC-MS/MS-has been published so far 23,24 . This article presents an immunopeptidomics assay using LC-MS/MS, which is fully validated according to the latest US Food and Drug Administration (FDA) and European Medicines Agency (EMA) guidelines [23][24][25][26][27] . We have to emphasize that such validations are specific only for dedicated equipment of one distinct laboratory. We provide a first protocol and template to enhance the validation of other laboratories with similar equipment and other omics fields using LC-MS/MS such as proteomics, metabolomics, and lipidomics.

Peptide synthesis
The automated peptide synthesizer Liberty Blue (CEM) was used to synthesize peptides following the 9-fluorenylmethyl-oxycarbonyl/tert-butyl (Fmoc/tBu) strategy. The identity and purity of the peptides were confirmed using a reversed-phase liquid chromatography (Alliance e2965, Waters) and an uHPLC system (nanoUHPLC, UltiMate 3000 RSLCnano, Dionex) on-line coupled LTQ Orbitrap XL hybrid mass spectrometer (ThermoFisher) system. Synthesized peptides were employed in the validation of LC-MS/MS identifications. Peptide sequences used for the validation are listed in supplemental Table S1.

Tissue samples
The EBV-transformed human B-cell line JY (ECACC 94022533) was cultured in RPMI1640 with 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin/streptomycin to a total number of 1*10 11 cells, centrifuged at 1,500 rpm for 15 min at 4 C, washed two times with cold PBS and aliquots containing 75*10 6 cells were frozen and stored at -80 C until use. The cells were tested negative for mycoplasma contamination via PCR.
The peripheral blood mononuclear cells (PBMC), chronic lymphocytic leukemia (CLL) and bladder cancer (BC) tissue samples were collected at the University Hospital of Tübingen with the informed consent of patients according to the principles of the Declaration of Helsinki. The local institutional review board (Ethics Committee at the Medical Faculty and the University Hospital of Tübingen) has approved the use of the patient samples.

Immunoaffinity purification of HLA ligands
HLA class I molecules were isolated using standard immunoaffinity purification as described 9,20,28,29 using the HLA class I-specific monoclonal antibody W6/32 30 . First, the cell pellets were lysed in 10 mM CHAPS (Applichem)/PBS (Lonza) containing protease inhibitors (Complete, Roche) and subsequently HLA molecules were purified using the pan-HLA class Ispecific monoclonal W6/32 Ab covalently linked to CNBr-activated Sepharose (GE Healthcare).

Analysis of HLA ligands by LC-MS/MS
Peptides were separated by nanoflow high-performance liquid chromatography (nanoUHPLC, UltiMate 3000 RSLCnano, Dionex) and subsequently analyzed in an on-line coupled Orbitrap Fusion Lumos or LTQ Orbitrap XL mass spectrometer (Thermo Fisher Scientific). Volumes of 5 μl peptide solution were injected onto a 75 μm × 2 cm trapping column (Acclaim PepMap RSLC, Dionex) at 4 μl/min for 5.75 min in five technical replicates. Subsequently, peptide separation was performed at 50°C at a flow rate of 300 nl/min on a 50 μm × 25 cm separation column (Acclaim PepMap RSLC, Dionex) applying a gradient ranging from 2.4 to 32.0% of AcN over the course of 90 min. Eluted peptides were ionized by nanospray ionization and analyzed in the Orbitrap Fusion Lumos implementing top speed collision-induced dissociation (CID) fragmentation. Survey scans were performed at 120,000 resolution and fragment detection at 60,000 resolution in the Orbitrap.
To demonstrate a method transfer, the immunopeptidomics pipeline was transferred from the Orbitrap Fusion Lumos to a LTQ Orbitrap XL. In the LTQ Orbitrap XL peptides were analyzed using a top five CID method with survey scans at 60,000 resolution and fragment ion detection in the ion trap operated at normal scan speed. On both instruments, the mass range was limited to 400-650 m/z with precursors of charge states 2+ and 3+ eligible for fragmentation.
Maintenance and OQ of the LC-MS/MS system are performed annually (Thermo Fisher Scientific).
A positive ion calibration using a Pierce™ LTQ velos or LTQ ESI positive ion calibration solution (Thermo Fisher Scientific) and a system suitability test using natural HLA class I-presented peptides of JY cells is performed weekly.

Database search and spectral annotation
Data was processed against the human proteome included in the Swiss-Prot database Precursor mass tolerance was set to 5 ppm, product ions mass tolerance was set to 0.02 Da for Orbitrap Fusion Lumos data and 0.5 Da for LTQ Orbitrap XL Data and oxidized methionine was allowed as the only dynamic modification with no restriction by enzymatic specificity.
Percolator 32 -assisted false discovery rate (FDR) calculation was set at a target value of q ≤ 0.05 (5% FDR). Peptide-spectrum matches with q ≤ 0.05 were filtered according to additional orthogonal parameters to ensure spectral quality and validity. Peptide lengths were limited to 8-12 amino acids.

Validation procedures
The validation of the immunopeptidomics procedure was done according to the OECD principles of Good Laboratory Practice (GLP) 33 and accuracy, precision, specificity, limit of detection and robustness were validated according to the FDA and EMA guidelines 25,26 . Clear definitions can be found in 25,26,33 .
As the immunopeptidome LC-MS/MS system separates the peptides using the LC and subsequently identifies the peptide ions in MS mode and product ions in the MS/MS mode, we tried to consider these three parts for every validation parameter summarized in Table 1 and

Experimental design and statistical rationale
A summary of the performed experiments, samples, technical replicates and MS RAW files is given in supplemental Table S2. Results were analyzed using GraphPad Prism (GraphPad software Inc).
The recovery rate was obtained by taking the average of the percentual overlapping peptides between the technical replicates normalized to the total number of peptides. The LC peptide retention times (RTs) were compared calculating the average of the Pearson correlations of the technical replicates.

Accuracy
To investigate the accuracy and specificity, the purified HLA-eluted peptides from one JY batch was spiked with 100 fmol isotope labeled synthetic peptides (Table S1) and analyzed in three separate analytical replicates (for identified peptides, see supplemental Table S3). The accuracy of the mass spectrometer did fulfill the acceptance criteria (Table 1) with a deviation below 1 ppm between the median mass deviation from the theoretical mass of all identified natural (median M: 0.02 ppm) and synthetic peptides (median M: 0.32 ppm) in Figure 2A. The peptide RTs between the replicates of all natural and all synthetic peptides do have a mean Pearson correlation above 95%, verifying the accuracy of the LC ( Figure 2B).

Specificity
Based on our experience from the first series of analyses, the six peptides AIVDKVPSV, SPQGRVMTI, RPSGPGPEL, YLLPAIVHI, KVLEYVIKV, and SPSSILSTL are expected as natural HLA class I-presented peptides of JY cells. In order to prove the specificity, the mass spectrometer must fulfill the MS mode acceptance criteria for precursor ions and the MS/MS mode acceptance criteria for five selected top product ions of the expected six peptides (Table 1). Here, we use two ways to select the five top product ions, we simply choose the top five most intensive fragments (last paragraph of specificity) or the most intensive fragments such as b and y fragments with the highest intensity and relevance (penultimate paragraph of specificity). Furthermore, in the LC separation, the correlation of the retention times of the natural and synthetic counterparts of the six peptides have to fulfill the acceptance criteria.
The difference of the median of the mass deviation from the theoretical mass (M ppm) of the six selected peptides AIVDKVPSV, SPQGRVMTI, RPSGPGPEL, YLLPAIVHI, KVLEYVIKV, and SPSSILSTL, which were identified as natural (median M: 0.17 ppm) and synthetic peptides (median M: 0.21 ppm), is below 1 ppm ( Figure 3A) (for identified peptides and product ions, see supplemental Table S4).
The peptide RTs between the replicates of the six selected peptides AIVDKVPSV, SPQGRVMTI, RPSGPGPEL, YLLPAIVHI, KVLEYVIKV, and SPSSILSTL, which were identified as natural and synthetic peptides, do have a Pearson correlation above 95% ( Figure 3B).
The standard deviation (SD) of the mass accuracy of the precursor masses in MS mode ( Figure   3C) and of the five selected top product ions, selected based on intensity and relevance, in MS/MS mode ( Figure 3D, for MS/MS spectra, see supplemental Figure S1) of the six selected peptides AIVDKVPSV, SPQGRVMTI, RPSGPGPEL, YLLPAIVHI, KVLEYVIKV and SPSSILSTL is below 0.001 Da for both the natural and synthetic peptides.
As an additional step to prove the specificity of the LC-MS/MS system, we validated our manual quality control method, to distinguish all peptides based on the mass of a precursor ion combined with the masses of the top five most intensive product ions. As the SD of the mass accuracy deviates at four decimals, our peptide identity criteria of the quality control should enable specificity at a four-decimal level (Table 1). Based on the precursor masses in MS mode at fourdigit level there was an overlap of 21 to 23 natural peptide masses in the tree replicates (Table   2A). These peptide masses could be separated using the top five masses with the highest intensity in MS/MS mode. In MS/MS mode the highest number of overlapping product ion masses was three (Table 2B). All 62 synthetic peptides could be separated based on the precursor masses at four decimals (Table 2C).

Limit of Detection
To determine the limit of detection (LOD), four aliquots of purified HLA-eluted peptides from JY cells were spiked with 0.1 fmol, 1 fmol, 10 fmol, or 100 fmol isotope labeled synthetic peptides (Table S1) and analyzed in three replicates leading to 12 separate analytical replicates (for identified peptides, see supplemental Table S3). Based on our experience with JY cells, in HLA ligandomic experiments an optimal setting enables a peptide recovery rate of 80 ± 20% between two replicates. Thus, there is no LOD where 100% of the peptides will be discovered, especially in data-dependent acquisition. Here, we set the LOD to the peptide concentration that enables an identification of at least 50% (n = 31) of the peptides per replicate and with a recovery rate of 80% ± 20% and a Pearson correlation of the peptide retention times above 95% between three replicates.
The JY sample spiked with 1 fmol synthetic peptides had the lowest peptide content enabling a reproducible identification of 50% of the 62 added isotope labeled peptides ( Figure 4A). At the LOD the recovery rate of peptides in a replicate mass spectrometric measurement is in the range of 80% ± 20% ( Figure 4B) and the mean Pearson correlation of the retention times of the synthetic peptides in the three technical replicates analyzed in the LC is above 95% ( Figure 4C).

Precision
The precision (also referred to as imprecision) was determined by assaying three aliquots of HLAeluted peptides from JY samples in three technical replicates leading to nine separate analytical replicates. To prove the intermediate precision, the measurement series was repeated after seven days (for identified peptides, see supplemental Table S5). The number of identified peptides fulfilled the acceptance criteria of ± 10% SD of the repeatability on the initial day and after one week ( Figure 5A, B). Furthermore, the acceptance criteria of the recovery rate with a recovery of 80 ± 20% of identified peptides in a repeated replicate were fulfilled on both measuring days ( Figure 5C). A closer look at the LC demonstrates that the mean Pearson correlation of the peptide retention times between all nine replicates was above 95% and fulfilled the criteria of the repeatability and intermediate precision ( Figure 5D).

Robustness of the precision, accuracy and specificity
The robustness was investigated performing a retrospective analysis of the purified HLA-eluted peptides of three primary samples: peripheral blood mononuclear cells from a healthy donor, a chronic lymphocytic leukemia sample, as well as a bladder cancer sample. The immunopeptidomes were isolated and analyzed by three different persons. To validate the robustness the specifications indicated in Table 1 should be fulfilled with regard to accuracy, precision, and specificity.
The identified peptides of the three primary samples fulfill the acceptance criteria of the accuracy for the mass spectrometer and for the LC. The difference of the median of the mass deviation from the theoretical mass of all identified natural peptides (median M: PBMC -0.09 ppm, CLL 0.04 ppm, BC 0.12 ppm) ( Figure 6A) and synthetic peptides (median M: 0.32 ppm) is below 1 ppm ( Figure 6A). The peptide RTs between the replicates of all natural and all synthetic peptides have a Pearson correlation above 95% ( Figure 6B).
Regarding the precision, the three technical replicates of the three primary samples fulfill the acceptance criteria of the repeatability for both the mass spectrometer and the LC. The replicates have a percentage SD of identified peptide numbers below 10% ( Figure 7A, B) and the recovery rate is in the range of 80% ± 20% ( Figure 7C). The peptide RTs between the replicates have a Pearson correlation above 95% ( Figure 6B).
The specificity can be investigated, as all analyzed primary samples are expected to contain at least one of the previously used 62 synthetic peptides as natural HLA class I-presented peptide.
The natural peptides should fulfill the acceptance criteria of the accuracy and specificity indicated in Table 1 for the mass spectrometer and the LC, respectively (for identified peptides and product ions, see supplemental Table S6).
The three replicates fulfill the acceptance criteria of the specificity for the mass spectrometer and for the LC. The difference of the median of the mass deviation from the theoretical mass (M ppm) of the selected peptides AIVDKVPSV and YLLPAIVHI in PBMC and CLL and SLLQHLIGL in BC, which were identified as natural and 100 fmol spiked synthetic peptides, is below 1 ppm ( Figure   8A).
The SD of the mass accuracy of the precursor masses in MS mode and of the selected five top product ions in MS/MS mode of the selected peptides AIVDKVPSV and YLLPAIVHI in PBMC and CLL and SLLQHLIGL in BC is below 0.001 Da for both the natural and synthetic peptides ( Figure   8B and C, for MS/MS spectra, see supplemental Figure S1).

Transfer of the method to other LC-MS/MS systems
In addition to the previous robustness analyses, the method was transferred to an LC-MS/MS system with a less sensitive LTQ Orbitrap XL and the HLA-eluted peptides from JY cells were analyzed. To demonstrate the method transfer to another LC-MS/MS system, the robustness measurements of the method were investigated with regard to accuracy, precision, and specificity on the LTQ Orbitrap XL containing LC-MS/MS system regardless of the specifications indicated in Table 1  The precision was determined by assaying the previously mentioned three technical replicates.
The number of identified peptides was similar to the validated LC-MS/MS system with 3% SD of the repeatability on the initial day ( Figure 9C, D). However, the recovery rate with a recovery of 62 ± 5% of identified peptides in a repeated replicate is different ( Figure 9E).
The specificity was again investigated using the mass deviation from the theoretical mass of the six selected peptides AIVDKVPSV, SPQGRVMTI, RPSGPGPEL, YLLPAIVHI, KVLEYVIKV, and SPSSILSTL. The median of the identified natural peptides is M: -0.50 ppm ( Figure 9A) (for identified peptides and product ions, see supplemental Table S7).
The peptide RTs between the replicates of the six selected peptides identified as natural and synthetic peptides do have a Pearson correlation above 95% ( Figure 9B).
The standard deviation (SD) of the mass accuracy of the precursor masses in MS mode ( Figure   9F

Conclusion/Discussion
To provide reliable biomarker and patient-individual tumor-associated target antigen identification for clinical studies, the fast and sensitive LC-MS/MS assay for the identification of natural and synthetic HLA-restricted peptides was validated for the technical equipment of our laboratory, consisting of a nanoUHPLC, UltiMate 3000 RSLCnano on-line coupled to an Orbitrap Fusion Lumos mass spectrometer. The immunopeptidomics pipeline is used for identification and impurity detection, thus according to FDA and EMA guidelines a validation of the specificity and LOD is required. Additionally, we validated the accuracy, precision, and robustness to demonstrate the reliability of the pipeline.
The results of the JY samples spiked with isotope labeled synthetic peptides enabled verification of the accuracy of the LC-MS/MS system in terms of similarity of the analyzed natural peptides to the theoretical and synthetic peptide masses. With the same dataset we were able to identify six specific peptides, expected as natural and synthetic peptides, which fulfil the acceptance criteria of the accuracy and could prove the specificity. Furthermore, we could show that with five selected MS/MS product ions all identified peptides within one replicate can be distinguished.
Instead of picking simply the five most intensive product ions for therapeutic peptide candidates, our quality control selects the top five product ions also according to meaningfulness (expert review: b-or y-ions are preferred), thus further increasing specificity. The validation of the precision showed a reliable identification of peptides with a uniform recovery rate proving the repeatability and intermediate precision after one week.
A major limitation of MS-based data-dependent acquisition (DDA) discovery approaches is the low recovery rate. In our immunopeptidomics experiments a recovery rate of 80% ± 20% was achieved for cell lines and tissue samples, owing to the tissue heterogeneity and high dynamic range. Due to the recovery rate, in our lab routinely triplicate measurements are performed. At the LOD a reasonable reliability should be provided with a recovery rate of 50%, when triplicate measurements are performed. A peptide content of 1 fmol synthetic peptides in JY matrix enabled a reliable identification of 50% of the peptides. An improvement of the recovery rate might be obtained with a replacement of the DDA analysis with data-independent acquisition, which has demonstrated a superior reproducibility [34][35][36][37][38] .
In order to prove the robustness of the immunopeptidomics assay, we synthesized a large variety of known HLA ligands, with different length, mass, grand average of hydropathicity (GRAVY), theoretical isoelectric point (pI), and HLA allotype restriction. In addition, we employed several primary, clinically relevant samples in addition to the JY cell line and further analyzed soluble peripheral blood mononuclear cells from a healthy donor, a soluble chronic lymphocytic leukemia and a solid bladder cancer sample. Lastly, for the three primary samples the HLA immunoaffinity chromatography and immunopeptidomics analysis was performed from three different persons.
We could successfully verify the specifications of the accuracy, precision, and specificity for both the mass spectrometer and LC, respectively.
Besides the robustness of the method, we exemplarily further investigated for precision, accuracy and specificity after the method transfer to a LC-MS/MS system utilizing an LTQ Orbitrap XL.
However, in order to obtain a high number of identified peptides, the MS/MS analysis is performed in the ion trap, instead of the Orbitrap, to enable a faster scanning throughput.
Consequently, the mass accuracy of the product ions in the MS/MS analysis varies already at the second decimal and for the previously selected top five ions, two new ions had to be defined for SPSSILSTL and one new ion for the other peptides, except of AIVDKVPSV. Furthermore, the recovery-rate is much lower using the LTQ Orbitrap XL.