Proteogenomics in Neoantigen Discovery

Like fire, immunity is a “handy servant but a fearful master” capable of both harming and curing. While the benefits of natural immunity (e.g., to smallpox) have been known for millennia it was not until the recent convergence of immunology as a hard science with associated analytic methodologies that immunity could be harnessed to treat cancer.

Like all drug-based strategies, immunotherapy requires knowledge of disease pathways and one or more identifiable targets. Where conventional oncology drugs show modest success against targets shared by most somatic cells but which are “overexpressed” in tumors, immunotherapies instead focus on biomarkers unique to cancer known as neoantigens or tumor-specific antigens.

The National Cancer Institute defines neoantigen as “a new protein that forms on cancer cells when certain mutations occur in tumor DNA,” and thus distinct from tumor-associated antigens and cancer-germline antigens, which are often expressed (usually at lower levels) in non-cancerous cells.

Since they are found exclusively in tumors, neoantigens offer the potential for therapies that kill cancer cells while sparing healthy cells.

Self and non-self

Unlike infectious viruses or bacteria, whose genes and proteins are distinctly non-human (and treatable with antimicrobials or antivirals), cancer cells share many unmutated “self” genes with healthy cells. To recognize neoantigens as distinct from “self,” T cells must distinguish from among subtle genomic changes at the level of either gene expression or antigen presentation.

This is no easy task, given that in cancer both coding and noncoding genomes generate, through mutations, thousands of potential antigens (neo- and otherwise). Just ten percent of nonsynonymous tumor cell mutations (i.e., those that change the protein sequence) generate peptides with high MHC affinity, and patients’ T cells recognize only 1% of those that bind to MHC.

“Mutation and RNA dysregulation may drive the generation of dysfunctional proteins or protein complexes, but many genetic mutations are silent and therefore, from the perspective of neoantigen discovery, inconsequential,” says Dr. Oliver Rinner, CEO and Co-Founder of Biognosys, which specializes in proteomic methods and workflows.

And among the ‘omics methods, proteogenomics offers the greatest potential for identifying these antigens, as well as elucidating relevant T-cell responses.

“Detecting neoantigens is challenging because tumors produce a large number of possible neoantigens—that is, altered or mutated proteins—but only some of them may be sufficiently immunogenic to elicit a robust antitumor immune response,” says Nebojsa Janjic, Chief Science Officer at Somalogic. “Sorting this out is not trivial.”

One starting point for neoantigen discovery is the analysis or characterization of dysregulated genes, RNA, or proteins/peptides. Investigators choose proteins because, as the genome’s “executive branch,” they represent all active coding proteins. “Unlike the genome, which is mostly static during a person’s life, proteins change in response to a variety of perturbations and stimuli, according to the health status,” Janjic tells Biocompare. “That most drugs target proteins is evidence that protein changes can cause disease. RNA also changes in response to stimuli, but these changes cover a smaller dynamic range, change more rapidly over time, and most importantly, they are not always reflective of the changes in proteins.”

But what makes the proteome the main area for target discovery also complicates the search for neoantigens. Among the research priorities are identifying TSAs that are unique to individual patients (e.g., for personalized medicine) or common to many patients (for off-the-shelf therapies). Proteogenomics also opens the way to unraveling the significance of neoantigens from noncoding regions of the genome for which very few analytical tools exist, particularly for cancers with a low tumor mutation burden (TMB).

For example, Chinese researchers have developed a method, PGNneo, specifically to address the noncoding issue. PGNneo consists of four modules that execute: noncoding variant calling (identifying single-nucleotide polymorphisms), human leukocyte antigen typing, extraction and library creation of mutated genes/proteins, identification of peptide variants, and neoantigen selection. Investigators have validated PGNneo for tumors of the liver and colon, and have made the software available online, presumably at no charge.

Low tumor burden

While low TMB is not an issue for epithelial and other tumors that can recruit neoantigen-active immune cells, the low neoantigen density in many other low-TMB tumors requires deep analysis to uncover immunogenic neoantigens. In these instances, researchers turn to high-throughput methods, for example proteogenomics integrating both next-generation sequencing and mass spectral analysis.

Acute myeloid leukemia is an example of a low TMB cancer for which the identification of suitable neoantigens has been difficult. In such cases investigators often turn to “atypical transcripts” which often arise from noncoding genes.

Unlike neoantigens derived from somatic mutations, which are patient-specific, mutations arising from atypical transcripts could potentially arise in multiple patients, thereby providing the possibility of “semi-personalized” therapies that work for groups or classes of patients.

The value of proteogenomics in these instances was established in 2020 by a group at the University of Montreal, which used an MS approach to identify 103 tumor-specific antigens in a collection of ovarian cancer samples. Conventional discovery focusing only on mutated exonic sequences would have uncovered only three of those neoantigens. Ninety-one of the tumor-specific antigens arose from unmutated, aberrantly expressed, non-exonic sequences that were not expressed in normal tissues. Investigators concluded that given their quantity and presence in many tumors, aberrantly expressed tumor-specific antigens “may be the most attractive targets for HGSC [high-grade serous ovarian cancer] immunotherapy.”

Proteogenomics easily integrates with other analytical modalities. In a 2023 paper, researchers at the Harvard/MIT Broad Institute described a novel approach, MONTE (Multi-Omic Native Tissue Enrichment), which expands the utility of proteogenomics through deep, efficient isolation of HLA-relevant peptides from clinical specimens. MONTE allows simultaneous analysis of proteome subsets, including the ubiquitylome, proteome, phosphoproteome, and acetylome, from the same tissue sample.

A small fraction of a small sample

Not surprisingly, proteogenomics-based neoantigen discovery is challenged by the scarcity of starting materials derived from human patients, only a tiny fraction of which is biologically relevant.

“Neoantigens represent a small fraction of peptides present on MHC complexes, which are in turn a tiny percentage of peptides present in biological samples,” says Rinner. The first step in characterizing these molecules is therefore enrichment, which typically involves antibodies targeting complexes composed of MHC and putative neoantigens.

Enrichment is followed by analysis through standard proteomics methods. “There are complications, however, with immunopeptides that make their analysis much harder compared with ‘normal’ peptides,” Rinner adds.

The first hurdle: the peptides cannot be further broken up proteolytically, which makes their ionization (and therefore their detection in the MS) more difficult.

So with limited sample quantities the mass spectrometer must be very sensitive. Specialized mass spectrometers like the SCP or timsTOF Ultra instrument from Bruker, which can handle very small sample amounts, can help with this challenge.

But the larger challenge, according to Rinner, is bioinformatics. “It is not the canonical protein sequences but the non-canonical, or mutated sequences that are the most interesting components of the immunopeptidome. Since MS proteomics requires DNA sequences as templates for peptide detection, one solution is to sequence the RNA and then build a specific sequence database for every sample. The other approach involves de novo sequencing, which is difficult but becomes more feasible with the progress in artificial intelligence-based methods that predict fragmentation spectra.”

MS is a cost-effective, highly scalable physical method, particularly when run in automated fashion. “MS uses standard reagents and provide easily-comparable results. What held MS-based proteomics back for so long was the high complexity and the way mass spectrometers generate data,” Rinner notes.

MS-based proteomics originally employed stochastic sampling (shotgun proteomics), which led to irreproducible data of limited value for biologists. “This changed when, together with Prof. Ruedi Aebersold (ETH Zurich and a Biognosys cofounder), and AB Sciex, we pioneered a new way to analyze mass spectrometric data. Thanks to powerful algorithms we could now analyze data very reproducibly. This approach, called data independent acquisition, is now widely used and is a major contributor to the rapid growth in MS proteomics,” Rinner explains.