Sensitive neoantigen discovery by real-time mutanome-guided immunopeptidomics

Abstract

Targeting cancer-specific HLA-peptide complexes is a promising approach in immunotherapy. Mutated neoantigens are excellent targets due to their immunogenicity and cancer-specificity. Mass spectrometry (MS)-based immunopeptidomics guides the selection of naturally presented immunogenic targets within the immunopeptidome, refining immunogenicity predictions. Implementation in clinical settings, however, must achieve global depth, capturing the entirety of the immunopeptidome, maintain high target sensitivity, and cater to scarce sample inputs and short turnaround time. Here, we present NeoDiscMS, an extension of NeoDisc that enables the acquisition of personalized immunopeptidomics data. Leveraging next-generation sequencing-guided real-time spectral acquisitions, NeoDiscMS maximizes sensitivity with minimal loss of global depth. Designed for effectiveness and ease of use, with minimal effort required for implementation, NeoDiscMS enhances the detection of peptides derived from tumor-associated antigens by up to 20% and improves confidence in neoantigen identification compared to the gold standard method. NeoDiscMS advances personalization in clinical antigen discovery with more confident neoantigen detection and easy implementation.

Fig. 1. Schematic overview of NeoDiscMS.

a Matched tumor and germline genome data, along with tumor transcriptome data, are processed by NeoDisc to create a tumor sample-specific personalized proteome reference, annotated with single-nucleotide polymorphisms, somatic mutations, which are then used for HLA binding and immunogenicity prediction of antigenic targets. From this prioritized list of peptides, NeoDisc generates a personalized inclusion list for NeoDiscMS acquisition of the matched tumor immunopeptidome sample. The MS data is then searched by NeoDisc against the personalized proteome to identify naturally presented antigenic peptides, refining the list to prioritize the most clinically relevant and likely immunogenic targets. b Comparison of DDA, DDA with an inclusion list (ilDDA) and NeoDiscMS methods. In NeoDiscMS, the MS acquisition of three second cycles was divided into three levels, in the following order of priority: the MS1 scan, the targeted branch MS2 scans, and the discovery branch (DDA) MS2 scans. c The targeted branch of NeoDiscMS consists of scouting scans (sMS2) that get triggered by an MS1 precursor mass match and a retention-time-restricted scheduled inclusion list. sMS2 are searched against a database in real-time to assess if a high-sensitivity scan (hMS2) should be triggered. The metrics that constitute the real-time search are the cross-correlation between predicted and measured fragments (Xcorr) and the mass deviation of the predicted and measured precursor mass (ΔPPM). Schematics were created in Biorender (https://BioRender.com/s85z530).

Fig. 2. Deconvolution of chimeric spectra.

a The upper panel depicts the number of identified unique stripped sequences in cell lines JY and RA957 immunopeptidome of injections of 5- or 10 million cells-equivalent, measured in DDA and with MS2 isolation windows of 1.2Th or 3.2Th (n = 3 per condition). The lower panel shows overlap coefficients between raw files and within or between processing modes. MSFragger can be operated with (MSFragger-DDA+) or without (MSFragger-DDA) taking chimeric spectra into account for finding PSMs. Wider isolation windows of 3.2Th and use of MSFragger-DDA+ increase the number of identified unique sequences, as demonstrated by searching the same raw files with either MSFragger-DDA or MSFragger-DDA+ (upper panel), resulting in an overlap coefficient of at least 88% (bottom panel; the top 3 gray dots for each sub-panel). b The fraction of MS2 scans with at least one assigned PSM. Each raw file was processed with MSFragger-DDA or MSFragger-DDA+. For all raw files and processing modes, most scans with one PSM or more were matched by both processing modes. Percentages were averaged across technical replicates (TRs). c The fraction of immunopeptides that were identified by MSFragger-DDA+ exclusively as co-isolated PSMs. d The fraction of binders (%-rank <2; MixMHCpred) and e sequence length distribution among immunopeptides that were uniquely identified as chimeric PSMs or not f The co-isolated PSMs derive from precursors that deviate in mass from the isolated precursor with a clear pattern: mass deviation distribution reflects differences in multiples of 1 Da.

Fig. 3. Benchmarking the global and target sensitivity of NeoDiscMS against DDA and ilDDA using an immunopeptidome dilution series.

a 9-mer motifs of immunopeptides that bind to the depicted HLA alleles. The key for diluting the class I immunopeptidome of RA957 in the class I immunopeptidome of JY is the distinctive A*68:01 binding restriction. We selected A*68:01-restricted peptides as targets because they can be clearly assigned to RA957 when the RA957 immunopeptidome is diluted in the JY immunopeptidome. b Histogram of the binding affinity of A*68:01-specific targets to JY alleles as predicted by MixMHCpred. c Summary of the dilution series employed to evaluate the scalability and sensitivity of NeoDiscMS. d Count of identified A*68:01-binding target peptides and global depth for each step in the dilution series (n = 3 per condition). e Number of target peptides identified in either zero, one, two, or all three DDA (x-axis) or NeoDiscMS replicates (y-axis). Bold numbers show the sum of all targets in the respective row/column. Large bold numbers show the sum of all targets identified with either acquisition method. f Linear regression of all 1/16 dilution PSM hyperscores (left) and spectral angles (right), made with either dMS2, sMS2, or hMS2 scans. g Correlation of target and non-target precursor MS1 intensities between technical replicates (TRs) for 1/1024 dilution injections.

Fig. 4. Accurate and sensitive detection of neoantigens and TAA-derived immunopeptides with NeoDiscMS compared to DDA in three melanoma derived cell-line.

a The individual identification count (upper panel) of target peptides (x-axis) among DDA and NeoDiscMS replicates (n = 3 per condition). Gene mappings (y-axis) order the identified neoantigens at the top. b The number of target identifications (TIDs; identification of precursor that triggered the targeted branch and was identified in either sMS2 and hMS2) made with either sMS2, hMS2 or both within an RTS event. Three adjacent blocks separated by thin black lines represent replicates 1,2, and 3. c The best TID hyperscore (left) or spectral angle (right) of each identified target peptide compared to the same peptide’s best hyperscore or spectral angle in the discovery branch.

Fig. 5. NeoDiscMS improves the confidence of neoantigen identifications.

a Each dot represents a PSM for the corresponding neoantigen (TR = replicate). b The retention time and identification rate of example neoantigen YPAAVNTIVAI in all three DDA and NeoDiscMS replicates of Ml2. c Zoom-in of the retention time window in NeoDiscMS technical replicate 1 of (b). Real-time search events where the real-time search filters are passed (RTS hit events) connect the respective sMS2 and hMS2 with a linker (gray arches). The real-time search parameters of each RTS hit event hovers above the respective linker. d The top YPAAVNTIVAI PSMs, based on spectral angle, for each scan type in DDA and NeoDiscMS demonstrate that hMS2 scans offer important improvements, including an increased number of identified fragments.

Fig. 6. NeoDiscMS can be integrated into DIA library search strategies for tumor biopsies.

a The analysis of three low-input uveal melanoma tissues (10 mg/injection) of the same patient (Ti1, Ti2, and Ti3, respectively), reveals a global depth of 12,316, 13,490, and 10,232 immunopeptides (binders only) with a DIA-NeoDiscMS hybrid-library search, with more than 95% of the identified peptides predicted as binders by MixMHCpred (%-rank <2). b Venn diagram describing the intersection of presented immunopeptides between Ti1, Ti2, and Ti3 with either a spectrum centric NeoDiscMS search (left) or a DIA-NeoDiscMS hybrid-library search (right). c Identified TAAs with either NeoDiscMS or spectrum-centric searches of DIA data, or a peptide-centric search of DIA with a DIA-NeoDiscMS hybrid-library.