A paper in Nature Biotechnology in October 2024 presents an “end-to-end clinical proteogenomic pipeline” to address the challenges associated with identification and prioritisation of antigenic peptides. NeoDisc combines “state-of-the-art publicly available and in-house software” with in silico tools to identify, predict, and prioritise tumour-specific and immunogenic antigens from multiple sources. The authors demonstrate the pipeline’s superiority over recent prioritisation pipelines and showcase the features that enable “both rule-based and machine-learning approaches”. They also reveal how NeoDisc’s multiomics integration identifies “defects in the cellular antigen presentation machinery”.
Antigen discovery
The researchers comment on the importance of personalised antigen discovery for the development of cancer vaccines. Common approaches for translational research and clinical trials include whole-genome sequencing (WGS) or whole-exome sequencing (WED) and RNA sequencing (RNAseq). However, the recent application of mass spectrometry (MS) to identify HLA-bound peptides and the use of proteogenomics have “facilitated the exploration of novel targets from a variety of antigens naturally processed and presented in cancer”.
“Their identification is laborious and current clinical pipelines do not support immunopeptidomics and are restricted to predicted neoantigens.”
Although immunotherapies are “remarkably effective” against some indications, “robust immune pressure” can force immune editing. Therefore, it is “essential” to understand the heterogenous antigenic landscape and the tumour’s capacity to present antigens.
The study
In the study, the authors introduce an “end-to-end” clinical antigen discovery proteogenomic pipeline: NeoDisc. It compiles publicly available and in-house software for the identification of immunogenic tumour-specific HLA-I and HLA-II antigens from genomics and transcriptomics and MS-based immunopeptidomics and enables their prediction and prioritisation with rule-based and machine-learning (ML) tools. It allows assessment of tumour heterogeneity and the functionality of the antigen processing and presentation machinery (APPM). The study compares NeoDisc’s performance with other tools, demonstrating its application for personalised antigen discovery and clinical implementation.
NeoDisc is a “dedicated computational framework” combining genomic, transcriptomic, and immunopeptidomic data and integrating curated public databases of known immunogenic TSAs, TAAs, oncoviral elements, and noncanonical transcripts. It uses matched tumour and germline genomic data for sample-specific variant characterisation, tumour content estimation, and copy number variation (CNV) and somatic mutation (SM) identification.
Four variant-calling algorithms are applied to WES and WGS data; variants that are detected by two or more callers are considered to have “high identification confidence”. Although highly mutated tumours usually respond better to immunotherapy, the selection of immunogenic neoantigens among “numerous possibilities” is “challenging”. Recent screening of large datasets of neoantigens in tumours from 112 participants has allowed the training of ML for prioritisation. ML classifiers trained on a fraction of this dataset have been integrated into NeoDisc to ensure “effective prioritisation”.
When NeoDisc’s rule-based and ML ranking approaches were compared with existing tools, the ML prioritisation algorithm “surpassed all the evaluated tools”. The researchers demonstrated NeoDisc’s “efficient prioritisation” on a cervical adenocarcinoma (CESC-1) characterised by an “exceptionally high mutational burden” (25 SMs per Mb).
Of the 393 identified actionable mutations, representing a pool of 19,051 peptides with a predicted binding rank ≤2%, 66 HLA-I neoantigenic short peptides (minimal epitopes) were selected through rule-based prioritisation for T cell screening of autologous tumour-infiltrating lymphocytes (TILs) by IFNγ ELISpot. 11 of the 66 peptides were immunogenic, including two that “ranked among the top ten candidates”. The NeoDisc ML model reordered the tested neoantigens, resulting in an “impressive” ranking of six immunogenic peptides in the top ten. NeoDisc also successfully identified two confirmed immunogenic neoantigens in the CESC-1 tumour MS immunopeptidomic data.
Personalised vaccines
While the default NeoDisc settings “exhibit good performance”, biopsies with low tumour content and low mutation burden could result in the detection of an “insufficient number of actionable high-confidence expressed mutations”. This would lead to a “suboptimal vaccine”. Thus, NeoDisc offers two additional modes:
- “Sensitive mode” considers the union of mutations called by all four variant-calling tools, to be used when an insufficient number of mutations are detected.
- “Panel mode” uses mutations listed in the available diagnostic clinical gene panel (GP) as input, allowing the design of vaccines for persons lacking dedicated biopsies. Note that GPs “often provide insufficient number of mutations leading to suboptimal lists of neoantigens or potentially none”.
The paper states that, in cancer vaccines, long sequences are favoured over minimal short peptides, motivated by the “efficient uptake and processing by APCs”. The NeoDisc ML tool ranks mutations according to their potential immunogenicity. Long sequences are “optimally” designed through maximised coverage of high-quality predicted HLA-I and HLA-II neoantigens.
In use in the field
NeoDisc is already being used in Phase I clinical trials for personalised cancer vaccines and adoptive T cell therapies in Switzerland; the authors hope that these demonstrate its “practical utility and potential for clinical translation”. Dr Michal Bassani-Sternberg of the Lausanne Branch of the Ludwig Institute for Cancer Research comments that NeoDisc provides “invaluable” insights into the immunobiology of tumours and the “mechanisms by which they evade targeting by cytotoxic T cells”.
“Notably, NeoDisc can also detect potential defects in the machinery of antigen presentation, alerting vaccine designers and clinicians to a key mechanism of immune evasion in tumours that can compromise the efficacy of immunotherapy. This can help them select patients for clinical studies who are likely to benefit from personalised immunotherapy, a capability that is also of great importance to optimising patient care.”
Florian Huber, first author, also reflected on the significance of this research.
“NeoDisc can detect all these distinct types of tumour-specific antigens along with neoantigens, apply machine learning and rule-based algorithms to prioritise those most likely to elicit a T cell response, and then use that information to design a personalised cancer vaccine for the relevant patient.”
Join us at the Congress in Barcelona this month to hear from experts in the Cancer and Therapeutic Vaccines track on their approaches to personalised vaccine development, and don’t forget to subscribe to our weekly newsletters here.
