Major histocompatibility complex (MHC)-bound peptides that originate from tumor-specific genetic alterations, known as neoantigens, are an important class of anti-cancer therapeutic targets. Accurately predicting peptide presentation by MHC complexes is a key aspect of discovering therapeutically relevant neoantigens. Technological improvements in mass-spectrometry-based immunopeptidomics and advanced modeling techniques have vastly improved MHC presentation prediction over the past two decades. However, improvement in the sensitivity and specificity of prediction algorithms is needed for clinical applications such as the development of personalized cancer vaccines, the discovery of biomarkers for response to checkpoint blockade and the quantification of autoimmune risk in gene therapies. Toward this end, we generated allele-specific immunopeptidomics data using 25 mono-allelic cell lines and created Systematic HLA Epitope Ranking Pan Algorithm (SHERPA TM), a pan-allelic MHC-peptide algorithm for predicting MHC-peptide binding and presentation. In contrast to previously published large-scale mono-allelic data, we used an HLA-null K562 parental cell line and a stable transfection of HLA alleles to better emulate native presentation. Our dataset includes five previously unprofiled alleles that expand MHC binding pocket diversity in the training data and extend allelic coverage in underprofiled populations. To improve generalizability, SHERPA systematically integrates 128 mono-allelic and 384 multi-allelic samples with publicly available immunoproteomics data and binding assay data. Using this dataset, we developed two features that empirically estimate the propensities of genes and specific regions within gene bodies to engender immunopeptides to represent antigen processing. Using a composite model constructed with gradient boosting decision trees, multi-allelic deconvolution and 2.15 million peptides encompassing 167 alleles, we achieved a 1.44 fold improvement of positive predictive value compared to existing tools when evaluated on independent mono-allelic datasets and a 1.15 fold improvement when evaluating on tumor samples. With a high degree of accuracy, SHERPA has the potential to enable precision neoantigen discovery for future clinical applications.
Background: Neoantigens are increasingly critical in immuno-oncology as therapeutic targets for neoantigen-based personalized cancer vaccines (PCVs) and as potential biomarkers for immunotherapy response. However, identifying which neoepitopes are more likely to provoke an immune response remains an important challenge for improving the effectiveness of PCVs and enabling neoantigens as a biomarker in immunotherapy. In recent years, Immuno-peptidomics has greatly improved in sensitivity and specificity, providing large number of peptides bound to MHC class I alleles in vivo. These advances make it possible to identify processed cell surface MHC bound peptides in an in vivo setting, providing accurate and representative presented peptide data for development of an improved neoantigen prediction pipeline. Methods: We generated high quality allele-specific training data for development of an accurate predictive algorithm. Mono-allelic HLA class I cell lines were generated by transfecting individual class I HLA alleles into the HLA class I null cell line K562, prioritizing alleles which will ultimately allow for development of a pan-class-I-allele prediction algorithm. Cell surface bound MHC class I peptides were identified for each transfected allele using immuno-peptidomics. We then developed and trained neural networks to predict MHC class I presentation for each assayed HLA allele. The predictive accuracy for each allele was comprehensively validated using immuno-peptidomic results derived from three sources: mono-allelic cell lines, deconvoluted cell lines, and patient derived tumor samples. Results: We applied immuno-peptidomics to develop a large and highly representative profile of MHC class I peptidomics across 30 HLA class I alleles. We then utilized this dataset to develop a highly accurate HLA class I presentation neural network. Through our work, we have identified thousands of HLA class I peptides bound to each of 30 unique HLA class I alleles, greatly expanding the known mono-allelic space. Our neoantigen prediction algorithm has been extensively validated, consistently achieving a higher overall accuracy across alleles (precision 0.88) than other publicly available tools (precision less than 0.7) based on both in vitro binding data and immuno-peptidomics, when tested on a broad set of peptide sources: mono-allelic cell lines, deconvoluted cell lines, and patient-derived tumor samples. Conclusions: Effective neoantigen identification can be greatly improved through application of immuno-peptidomics. We have generated extensive mono-allelic HLA class I cell lines and extensively characterized their class I ligandomes. We have used this data to develop and train a novel presentation neural network. Finally, we have extensively validated this tool using multiple newly-derived in vitro and in vivo sources, demonstrating very strong accuracy. Citation Format: Datta Mellacheruvu, Nick Phillips, Gabor Bartha, Jason Harris, Robert Power, Rena McClory, John West, Richard Chen, Sean Michael Boyle. Applying immunopeptidomics and machine learning to improve neoantigen prediction for therapeutic and diagnostic use [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4536.
Loss of human leukocyte antigen (HLA) is of increasing interest as a mechanism of cancer immune evasion and biomarker for cancer immunotherapy response. Each cancer patient has six class I HLA alleles that are capable of presenting a set of tumor-specific neoantigens. However, HLA alleles are often deleted in tumors, resulting in a loss of heterozygosity (LOH). When LOH occurs, neoantigen presentation is significantly impaired, potentially facilitating tumor immune evasion. Given the biologic impact of HLA LOH, there is a need for robust algorithms that can detect allele-specific HLA LOH in tumor samples. Here, we describe a novel computational approach to detect HLA LOH from exome sequencing, demonstrate the robustness of the method, and apply the method to calculate the frequencies of HLA LOH in key cancer types. We performed exome sequencing with augmented HLA region capture on the ImmunoID NeXT platform for tumor and normal samples of 184 patients across several cancer types and identified 430 nonhomozygous HLA genes. Next, we extracted the reads mapping to a custom HLA database and mapped them on to the patient-specific HLA alleles. For each allele, we calculated two key features: 1) the tumor b-allele frequency normalized by the native b-allele frequency and 2) the allele-specific tumor to normal coverage ratio. Using these two features, along with tumor purity and ploidy values, we trained a random forest model on a subset of the HLA genes (n=300). While standard copy-number variant (CNV) tools are unable to detect LOH in the polymorphic HLA genes, they can accurately measure deletions in their flanking regions, which we used to validate the accuracy of our allele-specific HLA LOH algorithm. In our test set (n=130), we found a high concordance between our allele-specific deletion calls and the generic deletion calls (94% accuracy, 0.85 F1 Score). When we constrained our test set to samples with high tumor content (>50%, n=40), we saw even stronger concordance (98% accuracy, 0.95 F1 Score). The only discordant call was a focal deletion within an HLA gene that was detected by our algorithm but missed by the CNV tool. Next, we ran our algorithm on patient samples of different cancer indications. For non-small cell lung cancer, we found a high frequency of patients affected by LOH (35%, 9 of 26), which is similar to frequencies previously reported in the literature. Furthermore, we found a lower frequency of melanoma tumors with LOH (15%, 7 of 48). In conclusion, we developed a novel algorithm to call allele-specific HLA LOH on the ImmunoID NeXT exome sequencing platform that augments coverage in the polymorphic HLA locus and demonstrated overall robust performance. The relatively high frequency of LOH events we detected in the melanoma and lung cancer samples suggests the importance of LOH analysis to inform cancer immunotherapy biomarker studies and personalized cancer therapies that depend on neoantigen presentation. Citation Format: Rachel Marty Pyke, Charles Abbott, Simo V. Zhang, Datta Mellacheruvu, John West, Richard Chen, Sean Michael Boyle. HLA allele-specific loss of heterozygosity detection using augmented exome capture approach [abstract]. In: Proceedings of the AACR Special Conference on Tumor Immunology and Immunotherapy; 2019 Nov 17-20; Boston, MA. Philadelphia (PA): AACR; Cancer Immunol Res 2020;8(3 Suppl):Abstract nr A19.
e18030 Background: Human leukocyte antigen loss of heterozygosity (HLA LOH) restricts immune recognition of tumors by limiting the major histocompatibility complex (MHC) presentation of neoantigens to T cells and correlates with reduced response to immune checkpoint blockade therapy (ICB) in non-small cell lung cancer. To explore the mechanism behind the impairment of HLA LOH on ICB, we analyzed the relationship between the antigen presentation pathway, neoantigen presentation and response to ICB in a head and neck squamous cell carcinoma (HNSCC) cohort. Methods: Following baseline sample collection, a cohort of 14 HNSCC patients recieved a single dose of PD-1 inhibitor. The primary tumor mass was definitively resected approximately one month later. If resection was impractical, a second biopsy was taken. Response to therapy was evaluated using RECIST criteria. Each pre- and post-intervention tumor sample and normal PBMC sample were profiled using Personalis’ ImmunoID NeXT Platform, an HLA-enhanced exome/transcriptome platform. HLA LOH was detected using a digital PCR validated machine learning algorithm (DASH). Neoantigen presentation was computationally predicted using a machine learning algorithm (SHERPATM) trained on mono-allelic immunopeptidomics data. Results: We found that 50% of the HNSCC cohort had HLA LOH, a larger percentage than in a large pan-cancer cohort (23%, n=611) and a distinct HNSCC cohort (40%, n=20). Further, two patients had B2M LOH and one patient had a deleterious mutation in an HLA allele. Despite the high frequency of somatic alteration in the antigen presentation pathway, we did not find an association between HLA LOH and ICB response. However, if HLA LOH was still shaping tumor evolution in response to ICB, we would expect to see immune pressure against subclonal tumor populations with neoantigens presentable by the retained HLA alleles but not the deleted HLA alleles. Indeed, we found that significantly more novel post-treatment neoantigens were predicted to bind to deleted HLA alleles compared to their homologous alleles (p=0.045). Conclusions: Given the high prevalence of HLA LOH across tumor types, a greater understanding is needed regarding the impact of HLA LOH on tumor evolution during ICB treatment. Though HLA LOH does not correlate with response to ICB, the consistent shift in neoantigen composition suggests that it acts as an evolutionary force in resistance to response during immunotherapy.
HLA loss of heterozygosity (LOH) is increasingly being recognized as an important immune escape mechanism in response to checkpoint inhibitor therapy. HLA LOH reduces the repertoire of neoantigens displayed on the cell surface of cancer cells, limiting the efficacy of the immune system to detect and eliminate them. Though highly accurate HLA LOH detection algorithms are needed to allow clinical utility, the field lacks robust, allele-specific validation approaches. Moreover, algorithms of unknown sensitivity and specificity have led to significant discrepancies in the estimated occurrence of HLA LOH as an immune escape mechanism across tumor types. To address these challenges, we have developed a machine learning algorithm to detect HLA LOH (DASH - Deletion of Allele-Specific HLAs), established the accuracy of the algorithm with an allele-specific PCR validation strategy, investigated the frequencies of HLA LOH across 14 tumor types in a cohort of over 800 patients and observed allele-specific neoantigen expansion in response to immunotherapy. To build DASH, we profiled 279 patients on the ImmunoID NeXT Platform to create a training dataset. Our novel features, which account for allele-specific differences in exome probe capture and capitalize on our whole exome platform by including information about copy number alterations in the regions flanking the HLA genes, were used to train an XGBoost model. Orthogonal, allele-specific validation was required to accurately assess sensitivity and specificity for clinical utility. Thus, we profiled over 30 paired tumor-normal cell lines on the ImmunoID NeXT Platform® and identified cell lines with HLA LOH. Using in silico mixtures, we found 100% sensitivity and specificity for tumors with at least 36% tumor purity. Next, we designed a digital PCR (dPCR) assay using patient-specific, allele-specific primers that target a single HLA allele while avoiding all other HLA alleles and tested the limit of detection of the assay in the same cell lines. Then, we performed dPCR with patient-specific primers on 20 tumor and normal sample pairs and found 94% sensitivity. After establishing the high sensitivity and specificity of DASH, we profiled over 800 patients spanning 14 tumor types on the ImmunoID NeXT Platform. We found that over 25% of patients in the majority of tumor types had at least one HLA LOH event. Further, we observed that novel neoantigens that arose during checkpoint treatment were significantly more likely to bind to deleted HLA alleles as compared to the remaining HLA alleles in a head and neck carcinoma cohort treated with anti-PD-1 therapy, shedding light on the mechanism of immune escape in response to checkpoint inhibitors. In summary, we introduced an HLA LOH detection method, performed allele-specific validation, exposed widespread HLA across tumor types and observed the mechanism of immune escape in response to immunotherapy. Citation Format: Rachel Marty Pyke, Datta Mellacheruvu, Charles Abbott, Steven Dea, Eric Levy, Simo V. Zhang, Nikita Bedi, A. Dimitrios Colevas, Devayani Bhave, Manju Chinnappa, Gabor Bartha, John Lyle, John West, Michael Snyder, John Sunwoo, Richard Chen, Sean Michael Boyle. Pan-cancer survey of HLA loss of heterozygosity using a robustly validated NGS-based machine learning algorithm [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 399.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.