HLA-peptide Binding Prediction Research Articles

HLAB: learning the BiLSTM features from the ProtBert-encoded proteins for the class I HLA-peptide binding prediction.

Human Leukocyte Antigen (HLA) is a type of molecule residing on the surfaces of most human cells and exerts an essential role in the immune system responding to the invasive items. The T cell antigen receptors may recognize the HLA-peptide complexes on the surfaces of cancer cells and destroy these cancer cells through toxic T lymphocytes. The computational determination of HLA-binding peptides will facilitate the rapid development of cancer immunotherapies. This study hypothesized that the natural language processing-encoded peptide features may be further enriched by another deep neural network. The hypothesis was tested with the Bi-directional Long Short-Term Memory-extracted features from the pretrained Protein Bidirectional Encoder Representations from Transformers-encoded features of the class I HLA (HLA-I)-binding peptides. The experimental data showed that our proposed HLAB feature engineering algorithm outperformed the existing ones in detecting the HLA-I-binding peptides. The extensive evaluation data show that the proposed HLAB algorithm outperforms all the seven existing studies on predicting the peptides binding to the HLA-A*01:01 allele in AUC and achieves the best average AUC values on the six out of the seven k-mers (k=8,9,...,14, respectively represent the prediction task of a polypeptide consisting of k amino acids) except for the 9-mer prediction tasks. The source code and the fine-tuned feature extraction models are available at http://www.healthinformaticslab.org/supp/resources.php.

DeepNetBim: deep learning model for predicting HLA-epitope interactions based on network analysis by harnessing binding and immunogenicity information

BackgroundEpitope prediction is a useful approach in cancer immunology and immunotherapy. Many computational methods, including machine learning and network analysis, have been developed quickly for such purposes. However, regarding clinical applications, the existing tools are insufficient because few of the predicted binding molecules are immunogenic. Hence, to develop more potent and effective vaccines, it is important to understand binding and immunogenic potential. Here, we observed that the interactive association constituted by human leukocyte antigen (HLA)-peptide pairs can be regarded as a network in which each HLA and peptide is taken as a node. We speculated whether this network could detect the essential interactive propensities embedded in HLA-peptide pairs. Thus, we developed a network-based deep learning method called DeepNetBim by harnessing binding and immunogenic information to predict HLA-peptide interactions.ResultsQuantitative class I HLA-peptide binding data and qualitative immunogenic data (including data generated from T cell activation assays, major histocompatibility complex (MHC) binding assays and MHC ligand elution assays) were retrieved from the Immune Epitope Database database. The weighted HLA-peptide binding network and immunogenic network were integrated into a network-based deep learning algorithm constituted by a convolutional neural network and an attention mechanism. The results showed that the integration of network centrality metrics increased the power of both binding and immunogenicity predictions, while the new model significantly outperformed those that did not include network features and those with shuffled networks. Applied on benchmark and independent datasets, DeepNetBim achieved an AUC score of 93.74% in HLA-peptide binding prediction, outperforming 11 state-of-the-art relevant models. Furthermore, the performance enhancement of the combined model, which filtered out negative immunogenic predictions, was confirmed on neoantigen identification by an increase in both positive predictive value (PPV) and the proportion of neoantigen recognition.ConclusionsWe developed a network-based deep learning method called DeepNetBim as a pan-specific epitope prediction tool. It extracted the attributes of the network as new features from HLA-peptide binding and immunogenic models. We observed that not only did DeepNetBim binding model outperform other updated methods but the combination of our two models showed better performance. This indicates further applications in clinical practice.

MATHLA: a robust framework for HLA-peptide binding prediction integrating bidirectional LSTM and multiple head attention mechanism

BackgroundAccurate prediction of binding between class I human leukocyte antigen (HLA) and neoepitope is critical for target identification within personalized T-cell based immunotherapy. Many recent prediction tools developed upon the deep learning algorithms and mass spectrometry data have indeed showed improvement on the average predicting power for class I HLA-peptide interaction. However, their prediction performances show great variability over individual HLA alleles and peptides with different lengths, which is particularly the case for HLA-C alleles due to the limited amount of experimental data. To meet the increasing demand for attaining the most accurate HLA-peptide binding prediction for individual patient in the real-world clinical studies, more advanced deep learning framework with higher prediction accuracy for HLA-C alleles and longer peptides is highly desirable.ResultsWe present a pan-allele HLA-peptide binding prediction framework—MATHLA which integrates bi-directional long short-term memory network and multiple head attention mechanism. This model achieves better prediction accuracy in both fivefold cross-validation test and independent test dataset. In addition, this model is superior over existing tools regarding to the prediction accuracy for longer ligand ranging from 11 to 15 amino acids. Moreover, our model also shows a significant improvement for HLA-C-peptide-binding prediction. By investigating multiple-head attention weight scores, we depicted possible interaction patterns between three HLA I supergroups and their cognate peptides.ConclusionOur method demonstrates the necessity of further development of deep learning algorithm in improving and interpreting HLA-peptide binding prediction in parallel to increasing the amount of high-quality HLA ligandome data.

DeepSeqPan, a novel deep convolutional neural network model for pan-specific class I HLA-peptide binding affinity prediction

Interactions between human leukocyte antigens (HLAs) and peptides play a critical role in the human immune system. Accurate computational prediction of HLA-binding peptides can be used for peptide drug discovery. Currently, the best prediction algorithms are neural network-based pan-specific models, which take advantage of the large amount of data across HLA alleles. However, current pan-specific models are all based on the pseudo sequence encoding for modeling the binding context, which is based on 34 positions identified from the HLA protein-peptide bound structures in early works. In this work, we proposed a novel deep convolutional neural network model (DCNN) for HLA-peptide binding prediction, in which the encoding of the HLA sequence and the binding context are both learned by the network itself without requiring the HLA-peptide bound structure information. Our DCNN model is also characterized by its binding context extraction layer and dual outputs with both binding affinity output and binding probability outputs. Evaluation on public benchmark datasets shows that our DeepSeqPan model without HLA structural information in training achieves state-of-the-art performance on a large number of HLA alleles with good generalization capability. Since our model only needs raw sequences from the HLA-peptide binding pairs, it can be applied to binding predictions of HLAs without structure information and can also be applied to other protein binding problems such as protein-DNA and protein-RNA bindings. The implementation code and trained models are freely available at https://github.com/pcpLiu/DeepSeqPan.

Anti-LGI1 encephalitis is associated with unique HLA subtypes.

Autoimmune encephalitis (AE), represented by anti-leucine-rich glioma-inactivated 1 (anti-LGI1) and anti-N-methyl-D-aspartate receptor (anti-NMDAR) encephalitis, has increasing clinical significance based on recent discoveries of neuronal autoantibodies. However, its immunopathogenesis is not fully understood. Here, we investigated whether AE is associated with the human leukocyte antigen (HLA) subtypes. We compared the HLA genotypes of 11 anti-LGI1 and 17 anti-NMDAR encephalitis patients to the control groups, which consisted of 210 epilepsy patients and 485 healthy Koreans. Anti-LGI1 encephalitis was associated with the DRB1*07:01-DQB1*02:02 haplotype (10 patients; 91%) in HLA class II genes, as well as with B*44:03 (8 patients; 73%) and C*07:06 (7 patients; 64%) in the HLA class I region. The prevalence of these alleles in anti-LGI1 encephalitis was significantly higher than that in the epilepsy controls or healthy controls. By contrast, anti-NMDAR encephalitis was not associated with HLA genotypes. Additional analysis using HLA-peptide binding prediction algorithms and computational docking underpinned the close relationship. This finding suggests that most anti-LGI1 encephalitis develops in a population with specific HLA subtypes, providing insight into a novel disease mechanism. Ann Neurol 2017;81:183-192.

Machine Learning Methods for Predicting HLA-Peptide Binding Activity.

As major histocompatibility complexes in humans, the human leukocyte antigens (HLAs) have important functions to present antigen peptides onto T-cell receptors for immunological recognition and responses. Interpreting and predicting HLA–peptide binding are important to study T-cell epitopes, immune reactions, and the mechanisms of adverse drug reactions. We review different types of machine learning methods and tools that have been used for HLA–peptide binding prediction. We also summarize the descriptors based on which the HLA–peptide binding prediction models have been constructed and discuss the limitation and challenges of the current methods. Lastly, we give a future perspective on the HLA–peptide binding prediction method based on network analysis.

Tumor Neoantigens Are Abundant Across Cancers

Tumor neoantigens are a promising class of vaccine immunogens as they arise from gene alterations in tumor cells and are hence exquisitely tumor-specific. We recently reported the development of a pipeline that leverages massively parallel sequencing data with HLA-peptide binding predictions to identify candidate neoantigens. By applying this pipeline to cases of chronic lymphocytic leukemia (CLL) with known HLA typing, we described the prediction of personal tumor neoantigens against which long-lived memory T cell responses developed following remission-inducing therapy. Our pipeline thus provides a method for selecting neoantigens for developing future personalized tumor vaccines. In order to extend this approach beyond CLL, we sought to gain estimates of tumor neoantigen loads across cancers. We hypothesized that the numbers of neoantigens within cancers would be proportional to their mutation frequency.To examine this hypothesis, we turned to the extensive collections of whole-exome sequencing (WES) data that have been generated through recent large-scale cancer sequencing projects. In order to generate accurate estimates of personal tumor neoantigen loads, HLA typing information is required. While in theory this information should be directly extractable from WES, direct inference of HLA type from standard WES reads has not been previously possible due to suboptimal alignments against a standard reference genome arising from the highly polymorphic nature of the HLA region. We therefore developed a strategy to optimize alignment. Based on the IMGT database, we constructed a reference library of all known HLA alleles (6597 unique entries) and aligned WES reads containing one or more short sequence segments corresponding to any HLA allele against this reference using the Novoalign software. HLA alleles were then inferred through a model that enabled calculation of allele probabilities by taking into account the number and quality of reads aligned to each allele. Alleles with the highest probabilities were then identified as winners. We trained the algorithm on 8 CLL cases for which WES data and HLA typing (based on conventional molecular typing) were available, and established a performance accuracy of ∼94% (45 of 48 alleles). This was further validated using a set of 133 Hap Map samples with known HLA typing, in which 94.61% (755 of 798) alleles were identified correctly at protein level resolution.We applied the HLA typing algorithm together with the neoantigen discovery pipeline across WES from 2488 cases collected from publicly available datasets of 13 diverse cancers. Mutation rates in solid tumor malignancies were consistently higher, in some cases by more than an order of magnitude, than the blood malignancies. For example, the high mutation rate tumor melanoma displayed a median of 300 (range, 34-4276) missense mutations per case, while renal cell carcinoma (RCC) had 41 (range, 10-101) and CLL had 16 (range, 0-75). The number of frame-shifting events (indels and termination read-throughs) was generally 10-fold or more lower in each tumor type than missense mutations and did not correlate with the number of missense mutations. As expected, the rate of predicted HLA binding peptides mirrored the somatic mutation rate per tumor type. The median number of predicted class I HLA-binding neopeptides (with IC50 < 500 nM) per sample generated from missense and frameshift events for melanoma was 488 (range: 18-5811), for RCC, 80 (range: 6-407), and for CLL 24 (range 2-124). Overall, we found an average of 1.5 HLA-binding peptides (i.e. with IC50<500nM) was generated per missense mutation and 4 binding peptides per frameshift mutation.By predicting tumor neoantigens in a variety of low and high mutation rate cancers, we established that dozens to hundreds of potential neoantigens are present in most tumors. In the process, we developed a highly accurate analytic approach that provides a solution for extracting HLA typing information from WES data but which could, in principle, be applied to other highly polymorphic regions of the genome. Ongoing studies focus on integrating estimates of tumor neoantigen load with understanding of HLA expression in order to optimize selection of antigen targets to build future personalized tumor vaccines. Disclosures:No relevant conflicts of interest to declare.

Analysis of HLA‐A24‐restricted peptides of carcinoembryonic antigen using a novel structure‐based peptide‐HLA docking algorithm

Carcinoembryonic antigen (CEA) is a very common tumor marker because many types of solid cancer usually produce a variety of CEA and a highly sensitive measuring kit has been developed. However, immunological responses associated with CEA have not been fully characterized, and specifically a weak immunogenicity of CEA protein as a tumor antigen is reported in human leukocyte antigen (HLA)-A24-restricted CEA peptide-based cancer immunotherapy. These observations demonstrated that immunogenic and potent HLA-A24-restricted CTL epitope peptides derived from CEA protein are seemingly difficult to predict using a conventional bioinformatics approach based on primary amino acid sequence. In the present study, we developed an in silico docking simulation assay system of binding affinity between HLA-A24 protein and A24-restricted peptides using two software packages, AutoDock and MODELLER, and a crystal structure of HLA-A24 protein obtained from the Protein Data Bank. We compared the current assay system with HLA-peptide binding predictions of the bioinformatics and molecular analysis section (BIMAS) in terms of the prediction capability using MHC stabilization and peptide-stimulated CTL induction assays for CEA and other HLA-A24 peptides. The MHC stabilization score was inversely correlated with the affinity calculated in the docking simulation alone (r = -0.589, P = 0.015), not with BIMAS score or the IFN-γ production index. On the other hand, BIMAS was not significantly correlated with any other parameters. These results suggested that our in silico assay system has potential advantages in efficiency of epitope prediction over BIMAS and ease of use for bioinformaticians.

Identification of CD4+ and CD8+ T cell epitopes of woodchuck hepatitis virus core and surface antigens in BALB/c mice

A therapeutic vaccine against chronic hepatitis B virus (HBV) infection requires the development of a strong and multispecific Th1 cell immune response. Woodchucks chronically infected with the woodchuck hepatitis virus (WHV) closely resemble HBV infection and represent the best animal model for this hepadnavirus-induced disease. Using the BIMAS “HLA Peptide Binding Predictions” program, we have identified and further characterized novel H-2d-restricted CD8+ epitopes within the WHV core (peptides C#12–21, C#18–32, C#19–27, C#61–69) and surface antigens (peptides preS2#10–18, preS2#27–35, S#76–84, S#133–140 and S#257–265), respectively. These peptides bind to H-2d with high efficiency and upon immunization of mice with peptide and Freund's adjuvant they induce the development of IFN-γ producing T cells. More importantly, WHV core peptides C#19–27 and C#61–69 and WHV surface peptides S#133–140 and S#257–265 were also recognized by CD8+ T cells after immunization of mice with DNA/PEI nanoparticles. Direct stimulation of splenocytes obtained from such DNA-immunized mice with peptides C#18–32, S#76–84, and S#257–265 resulted in significant production of IFN-γ. Thus, we have identified T cell determinants in mice from WHV core and surface antigens that have important value for designing and evaluating an effective vaccine against hepadnavirus infection.

Abstract 4783: Identification of a novel TAA, SPARC, as a target for immunotherapy of various cancers

Abstract It is important to identify Tumor-associated antigens (TAAs) to direct the immune system to attack cancer in order to establish efficient anti-cancer-immunotherapy. We analyzed the gene expression profiles of gastric cancers by using the genome wide cDNA microarray consisting of probes for 20,340 genes, and found that SPARC (Secreted Protein Acidic and Rich in Cysteine) gene was overexpressed in diffuse-type gastric cancer tissues at higher levels in comparison to adjacent normal gastric tissues. The levels of expression of SPARC in pancreatic cancer and colorectal cancer tissues were also higher than those of normal counterparts. On the other hand, the expression of SPARC among the normal tissue specimens was detected only in limited tissue specimens at very low levels. Therefore, we evaluated the possibility of SPARC as a target antigen for anticancer immunotherapy. This study attempted to identify HLA-A24 (A*2402)-restricted and SPARC derived CTL epitopes. To identify the SPARC-derived and HLA-A24-restricted CTL epitopes, we selected a total of 4 different candidate 9- or 10- amino acid peptides that were expected to have a higher binding affinity to HLA-A24 (A*2402) by the HLA Peptide Binding Predictions in the BIMAS software program. The H-2 Kd-binding peptide motif is comparable to that of HLA-A24 binding peptide. Thus, we examined CTL response of BALB/c mice to these SPARC-derived epitopes. We found that two SPARC derived peptides elicited high magnitude of CTL response in BALB/c mice in an H-2Kd-restricted manner. Therefore, we attempted to generate SPARC-specific CTLs from the PBMCs of healthy donors and various cancer patients positive for HLA-A24 by stimulation of PBMCs with these two peptides. SPARC peptides-reactive CTLs were successfully induced from PBMCs by in vitro stimulation with these two peptides in HLA-A24 (A*2402) positive healthy donors and cancer patients. Moreover the SPARC-reactive CTLs established from PBMCs derived from cancer patients by stimulation with these two peptides, exhibited cytotoxicity to the cancer cells expressing both HLA-A24 and endogenous or transgene-derived SPARC. On the other hand, the CTLs did not kill the cells which were negative for either HLA-A24 or SPARC. Furthermore, the adoptive transfer of the SPARC-specific CTLs could inhibit the tumor growth in NOD/SCID mice bearing human cancer cells expressing both HLA-A24 (A*2402) and SPARC. These findings suggest that SPARC is a potentially useful target candidate for cancer immunotherapy. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 4783.

Prediction of CTL epitopesof Tat exon 1 region in HIV-1 CRF07_BC strains in China

：Objective To predict theCTL epitopes of Tat exon 1 region in HIV-1 CRF07_BC strains, which were prevailing inChina. Methods Total of 236 plasma samples were from the 3rd National HIV MolecularEpidemic Survey (NMES3). All the subjects were infected with HIV-1 CRF07_BC viruses. Thetat exon 1 region was amplified by reverse transcription reaction and nested polymerasechain reaction (nested-PCR), then the PCR products were sequenced. The distribution of CTLepitopes of this region were predicted by on-line software BIMAS HLA Peptide BindingPredictions and statistics software. Results To-tal of 236 CRF07_BC strains were from 16provinces, mainly in intravenous drug asers(58.9%)and then sex(25.0%). It was showed thatthere were 12 CTL epitopes of 236 Tat exon 1 region of CRF07_BC strains mainly located inproline-rich region, cysteine-rich region and core-region. Those epitopes were banded by 5HLA presenting molecules in genotype(A * 2501 ,A * 2902, B * 15,B * 5301 and Cw * 1203)and 6 HLA presenting molecules in serotype (B53, B58 ,B57 ,A3 ,A68 and Cw12). Thefrequency of single amino acid substitution was more than 50% in 7 CTL epitopes.Conclusion The CTL epitopes in Tat exon 1 of CRF07 _BC strains were located in differentfunctional regions, and there were some amino acid variations in them.

Book review on Bioinformation Discovery: Data to knowledge in Biology

Researchers in almost all disciplines of Biology use mathematical, statistical models and computer programs to analyze and validate biological data. Systematic data mining is frequently completed in recent years using Bioinformatics soft-wares. The available Bioinformatics techniques and tools help annotate functions for newly generated data in biological investigation. The book (ISBN 978-1-4419-0518-5 e-ISBN 978-1-4419-0519-2) entitled “Bioinformation Discovery: Data to knowledge in Biology” by Dr. Pandjassarame Kangueane is published by Springer USA. This book describes the use of Biological data and Bioinformatics techniques and tools in Biological knowledge discovery. The book contains ten chapters. The first chapter describes concepts, principles and components of Bioinformatics with its various applications in agriculture, healthcare and industrial biotechnology sectors. The author precisely highlights that a sound knowledge in the basic concepts of bioinformatics is the real foundation for Bioinformation discovery from data. In the same chapter he also discussed the importance of Bioinformatics in Drug Discovery. The second chapter describes biological datasets and methods to create data subsets from primary databases like GenBank, EMBL, DDBJ and Protein databank (PDB). The provided dataset examples illustrate the importance of specialized datasets in Biological knowledge discovery. There are number of Bioinformatics tools and techniques freely available online for data mining and knowledge discovery in specific areas of Biology. In this context, Chapter 3 (Tools and Techniques) gives an account of some of the commonly used tools frequently cited in this book. The ongoing trends in biological sciences are to understand the protein-protein interaction within and across cells. The chapter 4 (Protein Subunits Interaction) highlights the use of 3D structure datasets to derive information and geometrical (interface size, planarity, sphericity, and complementarity) and chemical (the types of amino acid chemical groups, hydrophobicity, electrostatic interactions, and H-bonds) properties for protein subunit interactions. The formation and mechanism of folding for homodimers (dimers of identical subunits) is intriguing. The fifth chapter relates structural features to homodimer folding and binding using structural datasets. The provided schematic illustrations help in understanding the homodimer folding and binding mechanism. In Chapter 6 (Fusion Proteins), the author discusses on fusion proteins in one species mimic operons, protein-protein interaction, alternative splicing and multiple functions in another species. This chapter help relate gene fusion in establishing optimal dynamics and kinetics. The study of genetic variation among ethnic groups (American Indian, Australian aboriginal, Black, Caucasoid, Oriental, Hispanic, Mixed race, and Pacific Islander) is fascinating. Chapter 7 (Major Histocompatibility Complex (MHC) and Peptide Binding) discusses major histocompatibility complex (MHC) genes and their sequence polymorphism (sequence level variation). The discussion on issues in human MHC genes and methods for MHC peptide binding prediction helps to get insights into gene polymorphism and its impact in population based medicine. Chapter 8 (HLA Supertypes) gives the precise account of the human leukocyte antigen (HLA) alleles that demonstrates functional overall lap among HLA alleles for population coverage. To take readers further, in Chapter 9 (TEpitope Designer) the author described a web server named ’TEpitope Designer‘ which can be used to facilitate HLA-peptide binding prediction for peptide vaccine design. The last Chapter (Chapter 10: Eukaryotic Genes, Functions, Genomes, Design, and Evolution) has highlighted the main issues associated with the study of eukaryotic genes, genomes, function, design, and their evolution. Database such as U-Genome (database of relevant genome information in unicellular eukaryotes) which plays an important role in the understanding of unicellular genomes, their design and evolution is also discussed along with some other databases like ExInt (database of exon­intron gene structure in eukaryotic genes). This chapter gives a summary of several issues (exons, introns, single exon genes, intron organization and chromosome content) in comparative and functional genomics. In summary, this is a very useful book. It is written meticulously by an expert in the emerging field of Bioinformation discovery. This book provides plenty of interesting tips to use data in knowledge discovery. Each chapter concludes with a set of exercise problems for mastery of the subject. The language used in the book is extremely lucid, and hence book is very useful for undergraduate. This book is highly recommended for post-graduate students in Biology and Life Sciences. It should be noted that the book is available in both hard bound and electronic versions.

A modular concept of HLA for comprehensive peptide binding prediction

A variety of algorithms have been successful in predicting human leukocyte antigen (HLA)-peptide binding for HLA variants for which plentiful experimental binding data exist. Although predicting binding for only the most common HLA variants may provide sufficient population coverage for vaccine design, successful prediction for as many HLA variants as possible is necessary to understand the immune response in transplantation and immunotherapy. However, the high cost of obtaining peptide binding data limits the acquisition of binding data. Therefore, a prediction algorithm, which applies the binding information from well-studied HLA variants to HLA variants, for which no peptide data exist, is necessary. To this end, a modular concept of class I HLA-peptide binding prediction was developed. Accurate predictions were made for several alleles without using experimental peptide binding data specific to those alleles. We include a comparison of module-based prediction and supertype-based prediction. The modular concept increased the number of predictable alleles from 15 to 75 of HLA-A and 12 to 36 of HLA-B proteins. Under the modular concept, binding data of certain HLA alleles can make prediction possible for numerous additional alleles. We report here a ranking of HLA alleles, which have been identified to be the most informative. Modular peptide binding prediction is freely available to researchers on the web at http://www.peptidecheck.org .

T-Epitope Designer: A HLA-peptide binding prediction server

The current challenge in synthetic vaccine design is the development of a methodology to identify and test short antigen peptides as potential T-cell epitopes. Recently, we described a HLA-peptide binding model (using structural properties) capable of predicting peptides binding to any HLA allele. Consequently, we have developed a web server named T-EPITOPE DESIGNER to facilitate HLA-peptide binding prediction. The prediction server is based on a model that defines peptide binding pockets using information gleaned from X-ray crystal structures of HLA-peptide complexes, followed by the estimation of peptide binding to binding pockets. Thus, the prediction server enables the calculation of peptide binding to HLA alleles. This model is superior to many existing methods because of its potential application to any given HLA allele whose sequence is clearly defined. The web server finds potential application in T cell epitope vaccine design. http://www.bioinformation.net/ted/

Identification of HLA-A24-restricted CTL epitope encoded by the matrix protein pp65 of human cytomegalovirus

The activation of a specific cellular immune response against human cytomegalovirus (CMV) is an important key factor to solving CMV infection after bone marrow transplantation (BMT). In the present study, our purpose was to identify the HLA-A24-restricted cytotoxic T cell (CTL) epitope from the CMV immunogenic matrix protein pp65. We selected five CMV pp65 peptides with HLA-A24 binding motif from the HLA peptide binding predictions web site. Peptide binding assay was performed using biotinylated HLA-A24-restricted MAGE-1 peptide as a reference peptide and transporter associated with antigen processing (TAP)-deficient T2-A24 cells expressing high level of HLA-A24 protein as target cells. After co-incubation of biotinylated MAGE-1 and titrated amounts of competitor peptides with T2-A24 cells, the binding of each peptide was analyzed on a flow cytometer. Peptide binding assay showed that three out of five peptides derived from CMV pp65 bound to T2-A24 cells with various affinity levels. CTL induction assay using peptide-pulsed DC derived from eight HLA-A24 + donors revealed that the peptide (QYDPVAALF) with the highest affinity was able to elicit potent CTLs which killed peptide-pulsed TISI cells. These CTLs were found to show the killing activity against human fibroblast cells transduced with both HLA-A*2402 and CMV pp65 cDNAs, and CMV-infected HLA-A24 + fibroblast cells. These results suggested that the peptide (QYDPVAALF) is one of HLA-A24-restricted CTL epitope derived from CMV pp65 protein and may be of therapeutic value in peptide-based vaccines against CMV infection in BMT patients.