Since the discovery of the first human leukocyte antigen (HLA) restricted peptide, a plethora of studies have sought to elucidate the peptide antigens that are presented during the development of adaptive immunity (1). The series of peptides presented on the cell surface by HLA molecules to be surveilled by CD4 and CD8+ T cells, is collectively termed the immunopeptidome (2) ( Figure 1 ). The in-depth interrogation of these HLA peptides offers unique insights into the interplay between cells (whether healthy or perturbed) and the adaptive immune system. Characterising and understanding the immunopeptidome provides researchers with an excellent opportunity to develop precision therapeutics against cancer. Immunopeptidome discovery and analysis is directly dependent on the advancement of mass spectrometry, which in recent years has substantially progressed the field. In particular, the increased sensitivity, resolution and speed of mass spectrometers have allowed for tremendous leaps in the number of identifications of HLA-bound peptides in complex samples. This has facilitated the discovery of over 80,000 peptides in a single cell line, highlighting the substantial depth possible in comparison to 10 years ago, where only a handful of peptides were identified (3–6).

To this day, there is still an urgent need to circumvent the current drawbacks associated with immunopeptidomic analysis, these include the quantity of input material needed and the laborious nature required to isolate and analyse HLA bound peptides which become a major bottleneck when we want to analyse a big cohort of clinical samples (7–9). The majority of protocols currently in use either require manual processing for each sample or are designed for high throughput studies that necessitate the use of several 96-well plates to complete the entire process (7, 10–14)

We have developed a semi-automated immunopeptidomics method with mid-throughput capabilities, suitable for mid-sized cohorts and comparative immunopeptidomics studies. This protocol utilises a single 96-well plate and can process up to 12 samples per run, covering all the steps from loading anti-HLA antibody cocktails on magnetic microparticles to eluting purified HLA class complexes.

To evaluate the effectiveness of this protocol, we processed 12 samples in parallel (at cell counts of 5e5, 5e6, 1e7, and 5e7) with each sample processed in three biological replicates. We incorporated hyper-porous magnetic protein A beads to improve reproducibility, a KingFisherDuo liquid handling machine to reduce the potential for human error and decrease sample preparation time, and a data-independent acquisition (DIA) based approach to quantify HLA-bound peptides with high confidence and improve sensitivity and reproducibility.

In order to validate our method, we profiled the peptides liberated from HLA molecules on the MDA-MB-231 cell line to determine their characteristics. Using this semi-automated approach, we identify 13,312 unique HLA-I peptides across all samples at 1% FDR. When we compare the average number of peptides across each replicate, we observe 13,111 HLA bound peptides from eluates isolated from 5e7 cell pellet samples, 3,486 peptides from 1e7 samples, 2,814 from 5e6 and 397 from 5e5 cell pellets ( Figure 4 ). In a separate experiment, we studied the immunopeptidome of three 1e5 cell pellets to determine our limit of detection with only a handful of peptides identified in each sample ( Figure 4 ). As expected for HLA class I, all samples contain a high proportion of 9mers in comparison to other peptide lengths ( Figure 5 ). In order to ascertain whether these HLA peptides are indicative of the allotypes expressed on the MDA-MB-231 cell line, we examined the predicted binding of peptides between 8 and 13 amino acids in length to the HLA alleles expressed on this cell line. This showed ~96% of peptides were binders using NetMHCpan (16) ( Figure 6 ).

As anticipated, increasing the cell number resulted in an expansion of the peptide repertoire. We compared the peptides identified across each cell count and observed that the majority of peptides identified in smaller pellets were also identified in larger pellets. We observed that the 5e7 cell samples encompassed a very high proportion of the identified peptides in smaller pellets with less than 1% of the peptides been missed. ( Figure 7 ).

Utilising our optimised DIA variable window approach, we compared the relative intensities of commonly identified peptides in all four conditions predicted (n=521). Our observations indicate an increasing trend corresponding to log2 mean intensities, with a ~13-fold difference between 5e5 and 5e6, a ~97-fold difference between 5e5 and 5e7, and a ~7-fold change between 5e6 and 5e7. However, we did not observe a significant difference in the intensity of common ions between 5e6 and 1e7 pellets ( Figure 8 ).

The identification of HLA-bound peptides is crucial in the development of T cell-based immunotherapy, where potential epitopes are recognised by the adaptive immune system and used to design peptide-based vaccines for targeted elimination of diseased cells. Interest in this field has increased exponentially, leading to the development of novel protocols, the expansion of search algorithms that are not limited to reference proteomes and the development of different initiatives that have driven data sharing and bolstered data repositories (17–26). However, a constant “Achilles Heel’’ in this space has always been the difficult and laborious nature of immunopeptidomic workflows, with large sample material needed, high resolution mass spectrometers required and the lack of standardised protocols to ascertain HLA peptides. These limitations have collectively hindered the growth of this field and served as a constant challenge to overcome (17, 18, 25, 26).

Here, we have developed and showcased SAPrIm, A Semi-Automated Protocol for Mid-Throughput Immunopeptidomics and address some of the current limitations in this space.

Using MDA-MB-231 cell line as a model, we have demonstrated that even with limited starting material (5e5 cells), we can concordantly identify and quantify HLA bound peptides peptides that recapitulate the expected HLA peptide characteristics observed at both 10 and 100 times the starting material. Our results demonstrate that significant depth of coverage can be achieved using this protocol, as evidenced by the identification of approximately 13,000 peptides concordantly from 5e7 cells. Notably, this was accomplished without the need for offline fractionation methods and with a sample preparation time of less than 4 hours, with around 2 hours of the preparation time being hands-off experiments. We highlight that SAPrIm not only reduces manual handling time but exceeds current identification rates observed with other protocols (7, 11, 12).

When we perform comparative analysis, we see a high degree of overlap between each input amount highlighting the reproducibility of our approach. Furthermore, using this workflow we identify 697 peptides derived from known 195 cancer antigens highlighting that even with limited input material we are able to identify actionable targets for T cell mediated immunotherapy. Of note, in 5e5 cells we identify 36 peptides derived from 28 cancer antigens ( Supplementary Data 1 ).

Although DIA was initially introduced as a quantitative method in proteomics, recent advancements in data acquisition and analysis have made it a suitable method for both discovery and quantitative immunopeptidomics (14). This is due to the development of techniques such as spectral library-free searches, the generation of pan-spectral libraries from publicly available data, and the use of MS2 prediction algorithms for generating spectral libraries (14, 27). These improvements have expanded the capabilities of DIA, making it a valuable tool for researchers in the field of immunopeptidomics.

Although not performed in this study, it is possible to label the 12 samples resulting from the SAPrIm protocol with TMT tags using more than 12 channels and analyse them in a single LC-MS/MS run (28, 29). Additionally, the SAPriM protocol has the potential to be used for analysing HLA-II peptides and could be scaled up to analyse up to 96 samples using the KingFisher Apex system (Thermo Fisher).

Taken together, the use of this semi-automated approach facilitates many of the current limitations when it comes to the immunopeptidomic space. SAPrIm automates this approach, cuts sample preparation time and reduces the complexity of sample preparation. In addition to this, we support the feasibility of using relatively low quantities of antibody and starting material in comparison to traditional workflows, without the need for crosslinking (5, 7). We have also integrated this with the use of DIA, allowing for the accurate quantification and mapping of all the ions/peptides for future re-inspections of the data. A reliable and efficient workflow with a short turnover time is crucial in research and clinical settings to confidently screen biological and clinical materials for tumour-specific epitopes. These capabilities make SAPrIm technology a promising approach for translational immunopeptidomics.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: ProteomeXchange Consortium via the PRIDE (30) partner repository with the dataset identifier PXD041046.

PF, TLKS, SS and GG conceived and designed the method. GG, TLKS, TS, JS, LB, DJ, MS and TLKS performed experiments and collected data. GG and TLKS performed the data analysis. GG, AWP, SR, PF and TLKS wrote the manuscript. All authors contributed to the article and approved the submitted version.