Skip to main content
NCBI home page
Immunology logoLink to Immunology
. 2014 Sep 8;143(2):193–201. doi: 10.1111/imm.12301

Comparison of experimental fine-mapping to in silico prediction results of HIV-1 epitopes reveals ongoing need for mapping experiments

Julia Roider 1, Tim Meissner 1, Franziska Kraut 1, Thomas Vollbrecht 2, Renate Stirner 1, Johannes R Bogner 1, Rika Draenert 1
PMCID: PMC4172136  PMID: 24724694

Abstract

Methods for identifying physiologically relevant CD8 T-cell epitopes are critically important not only for the development of T-cell-based vaccines but also for understanding host–pathogen interactions. As experimentally mapping an optimal CD8 T-cell epitope is a tedious procedure, many bioinformatic tools have been developed that predict which peptides bind to a given MHC molecule. We assessed the ability of the CD8 T-cell epitope prediction tools syfpeithi, ctlpred and iedb to foretell nine experimentally mapped optimal HIV-specific epitopes. Randomly – for any of the subjects' HLA type and with any matching score – the optimal epitope was predicted in seven of nine epitopes using syfpeithi, in three of nine epitopes using ctlpred and in all nine of nine epitopes using iedb. The optimal epitope within the three highest ranks was given in four of nine epitopes applying syfpeithi, in two of nine epitopes applying ctlpred and in seven of nine epitopes applying iedb when screening for all of the subjects' HLA types. Knowing the HLA restriction of the peptide of interest improved the ranking of the optimal epitope within the predicted results. Epitopes restricted by common HLA alleles were more likely to be predicted than those restricted by uncommon HLA alleles. Epitopes with aberrant lengths compared with the usual HLA-class I nonamers were most likely not predicted. Application of epitope prediction tools together with literature searches for already described optimal epitopes narrows down the possibilities of optimal epitopes within a screening peptide of interest. However, in our opinion, the actual fine-mapping of a CD8 T-cell epitope cannot yet be replaced.

Keywords: CD8 T cell, epitope mapping, epitope prediction, MHC class I, optimal epitope

Introduction

MHC class I molecules present epitopes obtained from proteolytic digestion of endogenously synthesized proteins, for example, of viruses. The CD8 T-cell recognizes its cognate antigen, which is usually an 8–11 amino acid epitope bound to the MHC class I molecule, with its specific T-cell receptor. Methods to identify optimal CD8 T-cell epitopes are critically important not only for the development of T-cell-based vaccines but also to understand host–pathogen interactions. As pathogens can evade CD8 T-cell immune pressure by escape mutations within the respective epitope, the exact length of the epitope and the position of certain amino acids within the epitope are crucial for the development of escape mutations and as a consequence for the quality of the CD8 T-cell epitope.1,2 For example the amino acid at position two of an epitope is essential for MHC class I receptor binding and therefore leads to escape when mutated.3,4 To understand the complex interaction between epitope, MHC class I molecule and T-cell receptor, it is indispensable to know the exact length of the epitope and the exact sequence of amino acids within the epitope.5

CD8 T-cell epitopes are unlimited in number and differ for every disease. Mapping an optimal CD8 T-cell epitope is a tedious multi-step procedure that requires the evaluation of the actual length of the epitope by peptide truncations and assessing the HLA class I restriction of the epitope. Therefore, it is often replaced by bioinformatic tools.6–10 A variety of computational algorithms – many of which are freely available online – have been created to predict which peptides contained in a pathogen are likely T-cell epitopes11–15 (reviewed in refs 16,17). We hypothesized that freely available prediction tools do not reliably predict the actually mapped CD8 T-cell epitope in all cases.

To address this question, we compared nine optimal HIV-1 CD8 T-cell epitopes that we had already experimentally mapped – partly for previous projects – with the results from three prediction programmes that are listed on the Los Alamos database (http://www.hiv.lanl.gov/content/immunology/tools-links.html). We chose the sypeithi program12 because of its longevity and popularity within the research community as well as two computational programs that combine different popular machine learning algorithms (ctlpred and iedb).18,19 Randomly, the optimal epitope was predicted (for any of the subjects' HLA allele and with any matching score) in seven of nine epitopes using syfpeithi, in three of nine epitopes using ctlpred and in all nine of nine epitopes using iedb. However, the matching score was sometimes low. The optimal epitope within the top three results was given in four of nine epitopes applying syfpeithi, in two of nine epitopes applying ctlpred and in seven of nine epitopes applying iedb when searching for all of the subjects' HLA alleles.

Materials and methods

Study subjects

The study was approved by the Institutional Review Board of the university hospital of the Ludwig-Maximilians-Universität, Munich. All eight HIV-1-infected individuals participated after signing informed consent. For all study subjects, high-resolution HLA typing from extracted genomic DNA was performed by the Department for Transfusion Medicine of the University of Munich (HLA types of study subjects: C02: HLA-A*02:01, HLA-A*32:01, HLA-B*08:01, HLA-B*18:01, HLA-C*07:01; P02: HLA-A*03:01, HLA-A*25:01, HLA-B*35:01, HLA-B*39:01, HLA-C*04:01, HLA-C*12:03; P05: HLA-A*24:02, HLA-A*31:01, HLA-B*35:08, HLA-B*35:01, HLA-C*04:01; P07: HLA-A*24:02, HLA-A*68:01, HLA-B*18:01, HLA-B*35:01, HLA-C*07:01, HLA-C*07:04; P08: HLA-A*01:01, HLA-B*37:01, HLA-B*44:02, HLA-C*05:01, HLA-C*06:02; T06: HLA-A*01:01, HLA-A*30:04, HLA-B*08:01, HLA-B*49:01, HLA-C*07:01; T31: HLA-A*02:01,HLA-A*02:02, HLA-B*35:01, HLA-B*49:01, HLA-C*04:01, HLA-C*07:01; T32: HLA-A*02:01, HLA-A*25:01, HLA-B*44:02, HLA-C*05:01, HLA-C*07:04).

Peptides

Overlapping synthetic peptides corresponding to the HIV proteins Gag, Pol and Nef were used for screening (15–20 amino acids long, overlap of 5–10 amino acids; Gag: HIV-1 SF-2, Nef: HIV-1 Bru, NIBSC, England; Pol clade B consensus sequence of 2001 or according to patients' autologous viral sequence: EZBiolab, Carmel, IN). Truncations of longer peptides for epitope mapping were ordered as needed (EZBiolab).20 Peptides had a purity of ≥ 70%.

Cell lines

Epstein–Barr virus-transformed B lymphoblastoid cell lines (BLCL) were established in R20 medium (RPMI-1640; PAA, Pasching, Austria) supplemented with 2 mm l-glutamine, 50 U/ml penicillin, 50 mg/ml streptomycin, 10 mm HEPES and 20% heat-inactivated fetal calf serum (PAA).21 Antigen-specific CD8 T-cell lines were generated from peripheral blood mononuclear cells (PBMC) stimulated with synthetic peptide pulsed HLA-matched BLCL in the presence of 20 million irradiated feeder cells in R10 medium (RPMI-1640 supplemented with 2 mm l-glutamine, 50 U/ml penicillin, 50 mg/ml streptomycin, 10 mm HEPES and 10% heat-inactivated fetal calf serum) supplemented with 100 IU/ml of recombinant interleukin-2 (ImmunoTools GmbH, Friesoythe, Germany).22

Interferon-γ ELISPOT

HIV-1-specific CD8 T-cell responses and CD8 T-cell epitope mapping were evaluated partly within different projects.23,24 HIV-specific CD8 T-cell responses were quantified by ELISPOT assay using fresh or frozen PBMC (0·5 × 105 to 1 × 105 per well) and peptides (final concentration 12·5 μg/ml) as described previously.25,26 Interferon-γ-producing cells were counted by direct visualization on an AID Elispot Reader (Autoimmun Diagnostika GmbH, Strassberg, Germany) and were expressed as spot-forming cells (SFC) per 106 PBMC. Negative controls had to have no more than five spots. Wells were counted as positive if there were ≥ 50 SFC/106 PBMC. As the upper cut-off limit, 2000 SFC per 106 PBMC was chosen.

Truncations of a peptide were designed by shortening the best recognized screening peptide. After the optimal epitope was detected, it was verified by creating peptides where one amino acid was either added to or deleted from the N-terminal and C-terminal ends of the sequence of interest. The total of five peptides was then directly compared within the same ELISPOT assay (Fig. 1).25 For peptide titration assays, peptides were used at concentrations of 12·5 μg/ml to 1·25 × 10−4 μg/ml using 10-fold dilutions. Peptide comparisons were performed in two or more independent experiments and each peptide dilution was carried out in duplicate.

Figure 1.

Figure 1

Open in a new tab

Truncated peptide titrations reveal two HLA-A*02:01 restricted optimal epitopes of different length but contained within each other. Recognition of truncations (octamers to dodecamers) of the screening peptide RKQNPDIVIYQYMDDLYV (HIV pol 327–344) as measured by titration ELISPOT assays. Recognition of VIYQYMDDLYV (VV11) is depicted in black squares, recognition of YQYMDDLYV (YV9) is depicted in white squares and recognition of further truncated peptides is shown in grey (different shapes). In subject T31 the optimal epitope is the undecamer VIYQYMDDLYV (VV11) with highest avidity and highest magnitude of CD8 T-cell responses. In subject T32, the optimal epitope is (as predicted) the nonamer YQYMDDLYV (YV9) with highest avidity and highest magnitude of CD8 T-cell responses. The x-axis indicates the peptide concentration in increasing dilution; the y-axis indicates the magnitude of CD8 T-cell response expressed in spot-forming cells (SFC) per million peripheral blood mononuclear cells.

Intracellular cytokine staining for HLA restriction of CD8 T-cell epitopes

For determination of HLA class I restriction of CD8 T-cell responses, intracellular cytokine staining assays for interferon-γ using autologous or partly matched BLCL were performed as described elsewhere.27 The cells were analysed using a FACS Calibur Flow Cytometer (BD Biosciences, Heidelberg, Germany) and flowjo (Tree Star Inc., Ashland, OR) software. For negative controls, cells were incubated with BLCL without peptide, but were otherwise treated identically. Assays were performed in at least two independent experiments.

Prediction programs

The sypeithi program (http://www.syfpeithi.de/) is based on published motifs (pool sequencing, natural ligands) and takes into consideration the amino acids in the anchor and auxiliary anchor positions, as well as other frequent amino acids. The amino acids of a certain peptide are given a specific value depending on whether they are an anchor, auxiliary anchor or preferred residue. A score is calculated, based on this information.12 It is among the oldest, most renowned and cited (1830 times) prediction programs.

The ctlpred tool by the Institute of Microbial Technology (IMTECH; http://www.imtech.res.in/raghava/ctlpred/) combines two popular machine learning algorithms: artificial neuronal network (ANN)28–30 and support vector machine (SVM).31 T-cell epitopes are indirectly predicted by identifying MHC binding motifs.18 With the consensus and combined prediction approach of the two algorithms, an improvement of specificity and sensitivity compared with the individual methods is intended (cited 129 times).

The iedb tool (http://www.iedb.org/) is a combination of various machine learning algorithms: ANN,28 stabilized matrix method (SMM),19 SMMPMBEC (SMM with a Peptide: MHC Binding Energy Covariance matrix),32 NetMHCpan33 and Comblib (scoring matrices derived from combinatorial peptide libraries).34 These can be applied separately, as consensus35 or as default, which means that the program chooses the best method for a given HLA allele (cited 75 times).

Screening results were generated in November 2013 so the given scores are of this date.

Each program provides a binding score to rank the prediction results. Higher scores in syfpeithi represent better binding affinity. Similarly, a higher fraction with the ctlpred tool implies a better score whereas in the iedb tool a lower score indicates better binding affinity.

Results

Discrepancy between epitope prediction and experimental mapping results

Computational tools can predict which amino acid sequence within a given protein is likely to be an epitope for a T-cell receptor. We found a discrepancy between epitope prediction results for nine long screening HIV-specific peptides made by the programs syfpeithi,12 ctlpred18 and iedb19 and the experimentally mapped epitopes. Table 1 gives an overview of the prediction results of all three programs as well as the results of the epitope mapping experiments. As HLA restriction is usually unknown when starting to map an epitope, the displayed results are for all HLA alleles of the respective subject. If the correct amino acid sequence was predicted for one of the patient's HLA alleles, we counted it as a match. In the case of some subjects (T06, P08 – LVSAGIRKVL) ctlpred predicted the right optimal epitope, but not for the respective patients' HLA alleles. For this reason displayed results are indicated as ‘none’. The optimal epitope was predicted (randomly for any of the subjects' HLA allele and with any matching score) in seven of nine epitopes using syfpeithi, in three of nine epitopes using ctlpred and in all nine of nine epitopes using iedb. The optimal epitope within the best three results was given in four of nine (44%) epitopes applying syfpeithi, in two of nine (22%) epitopes applying ctlpred and in seven of nine (78%) epitopes applying iedb when searching for HLA alleles of all patients. Interestingly, in many subjects the different prediction programs predicted the same amino acid sequence, but with a different HLA restriction (P08, P02, P07, C02, T06) (Table 1).

Table 1.

Overview of all nine HIV-1-specific CD8 T-cell epitopes.

Inline graphic
Open in a new tab

Therefore entering a screening peptide in popular prediction tools gives the actual optimal epitope within the top three matches in 78% at best.

Known HLA restriction leads to better ranking of the optimal epitope within the prediction results, which highly depend on frequency of HLA type

Computational algorithms often predict the peptide binding affinity to a given HLA allele. We next assessed if knowing the restricting HLA allele improved prediction accuracy. Interestingly, overall prediction accuracy did not increase significantly if the HLA restriction of the peptide of interest was known. However, it improved the ranking of the optimal epitope within the prediction results in four of nine epitopes (Table 2).

Table 2.

Known HLA restriction leads to better ranking of the optimal epitope within the prediction results

ID Screening peptide Mapped epitope HLA restriction All HLA types HLA restriction known
P08 QVDKLVSAGIRKVLFL LVSAGIRKVL C*06:02 3rd rank 1st rank
P02 EWRFDSRLAFNHMARELHPE SRLAFNHMA B*39:01 7th rank 2nd rank
T31 RKQNPDIVIYQYMDDLYV VIYQYMDDLYV A*02:01 2nd rank 1st rank
T06 LVEICTEMEKEGKISKI EKEGKISKI B*49:01 7th rank 4th rank
Open in a new tab

The second last column shows the ranking of the optimal epitope by searching with all of the respective patient's HLA alleles and the last column shows the ranking of the optimal epitope when searching with the known restricting HLA allele. Results from iedb only.

Computational tools are regularly updated by training data sets, which also depend on the frequency of HLA types. We based the estimated allele frequency on calculations of Frahm et al.,36 who also studied HIV populations. As most of our subjects were Caucasians, HLA allele frequencies are referred to this group. Common HLA alleles (e.g. HLA-A*02:01, HLA-A*24:02, HLA-B*35:01, HLA-C*06:02) led to more accurate prediction results (four of four epitopes within the best three prediction results) than uncommon HLA alleles (e.g. HLA-B*39:01, HLA-B*49:01) (none of two epitopes within the best three prediction results) (Table 3). In Tables 2 and 3, we display results from the iedb tool only due to the search modes of the other two programs.

Table 3.

Prediction results highly depend on the frequency of the restricting HLA allele

Inline graphic
Open in a new tab

Therefore, knowing the HLA restriction as well as the epitope being restricted by common HLA alleles improves prediction results.

Aberrant polymers are probably not predicted

Usually, the length of a CD8 T-cell epitope is between 8 and 11 amino acids, the most common length is a nonamer. When the optimal epitope had a length other than nine amino acids it was probably not predicted. In the case of patients T31 and T32 the optimal HLA-A*02:01 epitope for the screening peptide RKQNPDIVIYQYMDDLYV (HIV pol 327–344) was evaluated. Epitope prediction revealed the nonamer YQYMDDLYV (YV9) with good scores in all three prediction programs (iedb: first rank, syfpeithi fourth rank, ctlpred: second rank). YV9 was found to be the optimal epitope for subject T32. However, in patient T31 epitope fine-mapping by using peptide truncations revealed the undecamer VIYQYMDDLYV (VV11) as the optimal epitope (Fig. 1).

The well-known HLA-A*24:02 epitope RYPLTFGW (RW8)37,38 was predicted by syfpeithi to be restricted by HLA-B*18 with a low binding score (score 2). The HLA-B*35:01 restricted octamer VPLRPMTY (VY8) was not predicted by syfpeithi at all. ctlpred calculates nonamers only. Interestingly, for both octamer epitopes the iedb tool predicted the optimal epitopes when we screened for octamers.

Our study reveals that unusual epitope length hinders epitope prediction.

Discussion

Experimentally mapping an optimal CD8 T-cell epitope is both a time-consuming and a funds-consuming procedure. Therefore, it has been replaced in many recent publications by epitope prediction programs.7–10 In this study we compared epitope prediction results from the programs syfpeithi, ctlpred and iedb with the results of epitope mapping experiments for nine HIV-specific CD8 T-cell epitopes. The optimal epitope was predicted in syfpeithi seven of nine, ctlpred three of nine, iedb nine of nine cases when screening for all of the respective patients' HLA alleles and including all possible epitopes (also those with low internal scores). When focusing on the three top hits, prediction accuracy decreased to four of nine for syfpeithi, two of nine for ctlpred and seven of nine for iedb.

Interestingly, knowing the HLA restriction of the peptide of interest did not improve overall epitope prediction accuracy. This can be explained by the fact that for some of the peptides the optimal epitope was predicted correctly but with a different restricting HLA allele. For example in P08, the epitope LVSAGIRKVL is restricted by HLA-C*06:02. syfpeithi predicted this epitope to be restricted by HLA-B*44:02, HLA-C*06:02 was not given as a possible restricting allele. In addition the predicted HLA restrictions for the same epitopes differed between the programs, which in our case could not be explained by common HLA supertypes.13 For example, the epitope DVKDTKEAL, which iedb predicted to be restricted by B*18:01 or B*08:01 (score 13·5/16 versus score 51 for A*02:01) whereas syfpheithi predicted it to be restricted by A*02:01 (score 14 versus score 0 for B*18:01/B*08:01) (Table 1).

The syfpeithi tool includes solely MHC class I alleles for which a large amount of data is available.12 The more commonly known the restricting MHC I allele is, the more reliable the predictions are. A similar observation can be made for the ctlpred tool because uncommon HLA alleles are not given for calculation. This indicates the need for more HLA-binding data and its incorporation into existing prediction tools. Aberrant polymers are more likely to be predicted by the iedb tool than by syfpeithi (or ctlpred). Even when search conditions are favourable (common HLA type) as in the case of the HLA-A*02:01 restricted RKQNPDIVIYQYMDDLYV (HIV pol 327–344), the optimal epitope in subject T31 VIYQYMDDLYV (VV11) would be missed out, solely relying on the mathematical prediction. This is surprising because the ‘HLA anchor position’ (position two of the epitope),3,4 which is supposed to be crucial for HLA binding of the respective epitope, differs between the nonamer (YQYMDDLYV) and the undecamer (VIYQYMDDLYV).

Experimental evaluation of in silico prediction results has been performed in mice.15,39 One of them finds good prediction accuracy for CD8 T-cell epitopes derived from vaccinia virus WR strain.15 However, they only included two MHC molecules. The second study – examining an artificial malaria antigen – also concludes that the available prediction tools need improvement and actual fine-mapping cannot yet be replaced.39 To our knowledge, a direct comparison between experimental fine-mapping of epitopes and in silico prediction in humans has not yet been performed. There is one study testing predicted T-cell epitopes of human respiratory syncytial virus and human metapneumovirus for HLA binding, which depended highly on the frequency of the HLA allele (HLA-A*02:01: 95%).40 This is in accordance with our findings.

All nine studied epitopes are HIV-specific CD8 T-cell epitopes. HIV is a rapidly evolving virus that escapes immune response (e.g. exerted by CD8 T cells) by viral escape mutations in the respective epitopes.41–44 One of the hallmarks of HIV infection is the rapid development of a genetically complex population (quasispecies) from an initially limited number of founder viruses. The sequences we used for screening with the prediction programs correspond to the consensus sequence, which does not necessarily look like the autologous virus in the respective study subject. Therefore, results of prediction programs may be biased. There is some evidence in the literature that prediction results for more conserved viruses (e.g. vaccinia virus) are more reliable.15 As a consequence, we can only state our conclusions for HIV.

Our findings with a prediction accuracy of 78% at best in HLA class I restricted HIV epitopes is comparable with a previously published study of HLA class II restricted HIV epitopes. Here, 73% of the predicted epitopes generated a CD4 T-cell response.45 In a different study of 184 predicted epitopes 114 (62%) were recognized by at least one of 31 study subjects.46 These above-mentioned studies focused, however, on identifying immunogenic epitopes in a broad population of HIV-1-infected individuals as opposed to mapping ‘the’ optimal epitope in one specific study subject as we did in our study.

There are plenty of epitope prediction programs that differ from each other by the mathematical algorithms they apply as well as by their search and output criteria.47,48 In the past years new prediction tools have been developed, some of them more accurate regarding, for example, variant epitope length than others.49–52 For the individual researcher, however, it is difficult to choose the most suitable method. For this reason we picked for our analyses three of ten listed epitope prediction programs published on the Los Alamos database (http://www.hiv.lanl.gov/content/immunology/tools-links.html). We chose syfpheithi because of its 14-year long experience and its popularity within the research community (cited 1830 times). Furthermore, we wanted to include popular mathematical algorithms like the artificial neural network method (ANN) as well as matrix-based prediction methods (SMM, SVM) and chose two frequently cited programs (ctlpred cited 129 times and iedb cited 75 times) that combine different approaches (reviewed in ref. 17). It has to be stated that this work is not meant and cannot serve as a systematical comparison between syfpeithi, ctlpred and iedb. Due to different application modes these three programs are not fully comparable. For example, searching for polymers of different lengths or different HLA types is laborious with syfpeithi. With ctlpred it is not possible at all as the program calculates nonamers only and provides solely HLA types with high binding affinities. However, an additional difficulty for researchers is the abundance of prediction programs that are available. In this respect, iedb gave the best results for our studies.

Epitope prediction tools applied together with literature search for already described optimal epitopes can narrow down the possibilities of optimal epitopes within a screening peptide of interest and therefore the synthesis of huge sets of peptides for experiments can be avoided. Knowledge of the HLA restriction, common restricting HLA alleles and usual polymer lengths (namely nonamers) improve prediction results. However, not knowing any of these variables beforehand, the actual experimental fine-mapping of an epitope cannot yet be replaced by prediction tools.

Acknowledgments

We thank all study participants and the dedicated clinical staff at the hospital. In addition, we thank Paige C. Cooper for critical review of the manuscript. This work was supported by the Deutsche Forschungsgemeinschaft (DR 424/3-1 to R.D.); the Friedrich-Baur-Stiftung (grant number 36/09 to R.D.) and the BayImmuNet (F2-F5121.7.1.1/8/1 to R.D.).

Disclosures

The authors state no conflicts of interest.

References

  • 1.Schneidewind A, Brumme ZL, Brumme CJ, Power KA, Reyor LL, O'Sullivan K, et al. Transmission and long-term stability of compensated CD8 escape mutations. J Virol. 2009;83:3993–7. doi: 10.1128/JVI.01108-08. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Brockman MA, Schneidewind A, Lahaie M, Schmidt A, Miura T, Desouza I, et al. Escape and compensation from early HLA-B57-mediated cytotoxic T-lymphocyte pressure on human immunodeficiency virus type 1 Gag alter capsid interactions with cyclophilin A. J Virol. 2007;81:12608–18. doi: 10.1128/JVI.01369-07. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Klein J, Sato A. The HLA system. Second of two parts. N Engl J Med. 2000;343:782–6. doi: 10.1056/NEJM200009143431106. [DOI] [PubMed] [Google Scholar]
  • 4.Klein J, Sato A. The HLA system. First of two parts. N Engl J Med. 2000;343:702–9. doi: 10.1056/NEJM200009073431006. [DOI] [PubMed] [Google Scholar]
  • 5.Rudolph MG, Stanfield RL, Wilson IA. How TCRs bind MHCs, peptides, and coreceptors. Annu Rev Immunol. 2006;24:419–66. doi: 10.1146/annurev.immunol.23.021704.115658. [DOI] [PubMed] [Google Scholar]
  • 6.Martin W, Sbai H, De Groot AS. Bioinformatics tools for identifying class I-restricted epitopes. Methods. 2003;29:289–98. doi: 10.1016/s1046-2023(02)00351-1. [DOI] [PubMed] [Google Scholar]
  • 7.Bronke C, Almeida CA, McKinnon E, et al. HIV escape mutations occur preferentially at HLA-binding sites of CD8 T-cell epitopes. AIDS. 2013;27:899–905. doi: 10.1097/QAD.0b013e32835e1616. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Brockman MA, Chopera DR, Olvera A, et al. Uncommon pathways of immune escape attenuate HIV-1 integrase replication capacity. J Virol. 2012;86:6913–23. doi: 10.1128/JVI.07133-11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Hoof I, Perez CL, Buggert M, Gustafsson RK, Nielsen M, Lund O, Karlsson AC. Interdisciplinary analysis of HIV-specific CD8+ T cell responses against variant epitopes reveals restricted TCR promiscuity. J Immunol. 2010;184:5383–91. doi: 10.4049/jimmunol.0903516. [DOI] [PubMed] [Google Scholar]
  • 10.Almeida CA, Bronke C, Roberts SG, et al. Translation of HLA-HIV associations to the cellular level: HIV adapts to inflate CD8 T cell responses against Nef and HLA-adapted variant epitopes. J Immunol. 2011;187:2502–13. doi: 10.4049/jimmunol.1100691. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Peters B, Sidney J, Bourne P, et al. The design and implementation of the immune epitope database and analysis resource. Immunogenetics. 2005;57:326–36. doi: 10.1007/s00251-005-0803-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Rammensee H, Bachmann J, Emmerich NP, Bachor OA, Stevanovic S. SYFPEITHI: database for MHC ligands and peptide motifs. Immunogenetics. 1999;50:213–9. doi: 10.1007/s002510050595. [DOI] [PubMed] [Google Scholar]
  • 13.Heckerman D, Kadie C, Listgarten J. Leveraging information across HLA alleles/supertypes improves epitope prediction. J Comput Biol. 2007;14:736–46. doi: 10.1089/cmb.2007.R013. [DOI] [PubMed] [Google Scholar]
  • 14.Lundegaard C, Nielsen M, Lund O. The validity of predicted T-cell epitopes. Trends Biotechnol. 2006;24:537–8. doi: 10.1016/j.tibtech.2006.10.001. [DOI] [PubMed] [Google Scholar]
  • 15.Moutaftsi M, Peters B, Pasquetto V, Tscharke DC, Sidney J, Bui HH, Grey H, Sette A. A consensus epitope prediction approach identifies the breadth of murine T(CD8+)-cell responses to vaccinia virus. Nat Biotechnol. 2006;24:817–9. doi: 10.1038/nbt1215. [DOI] [PubMed] [Google Scholar]
  • 16.Korber B, LaBute M, Yusim K. Immunoinformatics comes of age. PLoS Comput Biol. 2006;2:e71. doi: 10.1371/journal.pcbi.0020071. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Castelli M, Cappelletti F, Diotti RA, Sautto G, Criscuolo E, Dal Peraro M, Clementi N. Peptide-based vaccinology: experimental and computational approaches to target hypervariable viruses through the fine characterization of protective epitopes recognized by monoclonal antibodies and the identification of T-cell-activating peptides. Clin Dev Immunol. 2013;2013:521231. doi: 10.1155/2013/521231. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Bhasin M, Raghava GP. Prediction of CTL epitopes using QM. SVM and ANN techniques. Vaccine. 2004;22:3195–204. doi: 10.1016/j.vaccine.2004.02.005. [DOI] [PubMed] [Google Scholar]
  • 19.Peters B, Bui HH, Frankild S, et al. A community resource benchmarking predictions of peptide binding to MHC-I molecules. PLoS Comput Biol. 2006;2:e65. doi: 10.1371/journal.pcbi.0020065. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Draenert R, Brander C, Yu XG, Altfeld M, Verrill CL, Feeney ME, Walker BD, Goulder PJ. Impact of intrapeptide epitope location on CD8 T cell recognition: implications for design of overlapping peptide panels. AIDS. 2004;9:18. doi: 10.1097/00002030-200404090-00004. [DOI] [PubMed] [Google Scholar]
  • 21.Walker BD, Chakrabarti S, Moss B, et al. HIV-specific cytotoxic T lymphocytes in seropositive individuals. Nature. 1987;328:345–8. doi: 10.1038/328345a0. [DOI] [PubMed] [Google Scholar]
  • 22.Walker BD, Flexner C, Birch-Limberger K, et al. Long-term culture and fine specificity of human cytotoxic T-lymphocyte clones reactive with human immunodeficiency virus type 1. Proc Natl Acad Sci USA. 1989;86:9514–8. doi: 10.1073/pnas.86.23.9514. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Vollbrecht T, Eberle J, Roider J, et al. Control of M184V HIV-1 mutants by CD8 T-cell responses. Med Microbiol Immunol. 2012;201:201–11. doi: 10.1007/s00430-011-0222-1. [DOI] [PubMed] [Google Scholar]
  • 24.Roider J, Kalteis AL, Vollbrecht T, Gloning L, Stirner R, Henrich N, Bogner JR, Draenert R. Adaptation of CD8 T cell responses to changing HIV-1 sequences in a cohort of HIV-1 infected individuals not selected for a certain HLA allele. PLoS ONE. 2013;8:e80045. doi: 10.1371/journal.pone.0080045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Altfeld MA, Trocha A, Eldridge RL, et al. Identification of dominant optimal HLA-B60- and HLA-B61-restricted cytotoxic T-lymphocyte (CTL) epitopes: rapid characterization of CTL responses by enzyme-linked immunospot assay. J Virol. 2000;74:8541–9. doi: 10.1128/jvi.74.18.8541-8549.2000. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Draenert R, Altfeld M, Brander C, et al. Comparison of overlapping peptide sets for detection of antiviral CD8 and CD4 T cell responses. J Immunol Methods. 2003;275:19–29. doi: 10.1016/s0022-1759(02)00541-0. [DOI] [PubMed] [Google Scholar]
  • 27.Draenert R, Allen TM, Liu Y, et al. Constraints on HIV-1 evolution and immunodominance revealed in monozygotic adult twins infected with the same virus. J Exp Med. 2006;203:529–39. doi: 10.1084/jem.20052116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Buus S, Lauemoller SL, Worning P, et al. Sensitive quantitative predictions of peptide-MHC binding by a ‘Query by Committee’ artificial neural network approach. Tissue Antigens. 2003;62:378–84. doi: 10.1034/j.1399-0039.2003.00112.x. [DOI] [PubMed] [Google Scholar]
  • 29.Nielsen M, Lundegaard C, Worning P, Lauemoller SL, Lamberth K, Buus S, Brunak S, Lund O. Reliable prediction of T-cell epitopes using neural networks with novel sequence representations. Protein Sci. 2003;12:1007–17. doi: 10.1110/ps.0239403. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Lundegaard C, Lamberth K, Harndahl M, Buus S, Lund O, Nielsen M. NetMHC-3.0: accurate web accessible predictions of human, mouse and monkey MHC class I affinities for peptides of length 8-11. Nucleic Acids Res. 2008;36:W509–12. doi: 10.1093/nar/gkn202. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Furey TS, Cristianini N, Duffy N, Bednarski DW, Schummer M, Haussler D. Support vector machine classification and validation of cancer tissue samples using microarray expression data. Bioinformatics. 2000;16:906–14. doi: 10.1093/bioinformatics/16.10.906. [DOI] [PubMed] [Google Scholar]
  • 32.Kim Y, Sidney J, Pinilla C, Sette A, Peters B. Derivation of an amino acid similarity matrix for peptide: MHC binding and its application as a Bayesian prior. BMC Bioinformatics. 2009;10:394. doi: 10.1186/1471-2105-10-394. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Hoof I, Peters B, Sidney J, Pedersen LE, Sette A, Lund O, Buus S, Nielsen M. NetMHCpan, a method for MHC class I binding prediction beyond humans. Immunogenetics. 2009;61:1–13. doi: 10.1007/s00251-008-0341-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Sidney J, Assarsson E, Moore C, Ngo S, Pinilla C, Sette A, Peters B. Quantitative peptide binding motifs for 19 human and mouse MHC class I molecules derived using positional scanning combinatorial peptide libraries. Immunome Res. 2008;4:2. doi: 10.1186/1745-7580-4-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Kim Y, Ponomarenko J, Zhu Z, et al. Immune epitope database analysis resource. Nucleic Acids Res. 2012;40:W525–30. doi: 10.1093/nar/gks438. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Frahm N, Korber BT, Adams CM, et al. Consistent cytotoxic-T-lymphocyte targeting of immunodominant regions in human immunodeficiency virus across multiple ethnicities. J Virol. 2004;78:2187–200. doi: 10.1128/JVI.78.5.2187-2200.2004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Goulder PJ, Edwards A, Phillips RE, McMichael AJ. Identification of a novel HLA-A24-restricted cytotoxic T-lymphocyte epitope within HIV-1 Nef. AIDS. 1997;11:1883–4. doi: 10.1097/00002030-199715000-00015. [DOI] [PubMed] [Google Scholar]
  • 38.Ikeda-Moore Y, Tomiyama H, Ibe M, Oka S, Miwa K, Kaneko Y, Takiguchi M. Identification of a novel HLA-A24-restricted cytotoxic T-lymphocyte epitope derived from HIV-1 Gag protein. AIDS. 1998;12:2073–4. doi: 10.1097/00002030-199815000-00020. [DOI] [PubMed] [Google Scholar]
  • 39.Bergmann-Leitner ES, Chaudhury S, Steers NJ, Sabato M, Delvecchio V, Wallqvist AS, et al. Computational and experimental validation of B and T-cell epitopes of the in vivo immune response to a novel malarial antigen. PLoS ONE. 2013;8:e71610. doi: 10.1371/journal.pone.0071610. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Rock MT, McKinney BA, Yoder SM, Prudom CE, Wright DW, Crowe JE., Jr Identification of potential human respiratory syncytial virus and metapneumovirus T cell epitopes using computational prediction and MHC binding assays. J Immunol Methods. 2011;374:13–7. doi: 10.1016/j.jim.2011.08.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Allen TM, O'Connor DH, Jing P, et al. Tat-specific cytotoxic T lymphocytes select for SIV escape variants during resolution of primary viraemia. Nature. 2000;407:386–90. doi: 10.1038/35030124. [DOI] [PubMed] [Google Scholar]
  • 42.Friedrich TC, Dodds EJ, Yant LJ, et al. Reversion of CTL escape-variant immunodeficiency viruses in vivo. Nat Med. 2004;10:275–81. doi: 10.1038/nm998. [DOI] [PubMed] [Google Scholar]
  • 43.Peyerl FW, Barouch DH, Yeh WW, Bazick HS, Kunstman J, Kunstman KJ, Wolinsky SM, Letvin NL. Simian-human immunodeficiency virus escape from cytotoxic T-lymphocyte recognition at a structurally constrained epitope. J Virol. 2003;77:12572–8. doi: 10.1128/JVI.77.23.12572-12578.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.O'Connor DH, Allen TM, Vogel TU, et al. Acute phase cytotoxic T lymphocyte escape is a hallmark of simian immunodeficiency virus infection. Nat Med. 2002;8:493–9. doi: 10.1038/nm0502-493. [DOI] [PubMed] [Google Scholar]
  • 45.Buggert M, Norstrom MM, Czarnecki C, et al. Characterization of HIV-specific CD4+ T cell responses against peptides selected with broad population and pathogen coverage. PLoS ONE. 2012;7:e39874. doi: 10.1371/journal.pone.0039874. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Perez CL, Larsen MV, Gustafsson R, et al. Broadly immunogenic HLA class I supertype-restricted elite CTL epitopes recognized in a diverse population infected with different HIV-1 subtypes. J Immunol. 2008;180:5092–100. doi: 10.4049/jimmunol.180.7.5092. [DOI] [PubMed] [Google Scholar]
  • 47.Yang X, Yu X. An introduction to epitope prediction methods and software. Rev Med Virol. 2009;19:77–96. doi: 10.1002/rmv.602. [DOI] [PubMed] [Google Scholar]
  • 48.Zhang GL, Ansari HR, Bradley P, et al. Machine learning competition in immunology – prediction of HLA class I binding peptides. J Immunol Methods. 2011;374:1–4. doi: 10.1016/j.jim.2011.09.010. [DOI] [PubMed] [Google Scholar]
  • 49.Karosiene E, Lundegaard C, Lund O, Nielsen M. NetMHCcons: a consensus method for the major histocompatibility complex class I predictions. Immunogenetics. 2012;64:177–86. doi: 10.1007/s00251-011-0579-8. [DOI] [PubMed] [Google Scholar]
  • 50.Lundegaard C, Lund O, Nielsen M. Prediction of epitopes using neural network based methods. J Immunol Methods. 2011;374:26–34. doi: 10.1016/j.jim.2010.10.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Zhang GL, DeLuca DS, Keskin DB, et al. MULTIPRED2: a computational system for large-scale identification of peptides predicted to bind to HLA supertypes and alleles. J Immunol Methods. 2011;374:53–61. doi: 10.1016/j.jim.2010.11.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Vider-Shalit T, Louzoun Y. MHC-I prediction using a combination of T cell epitopes and MHC-I binding peptides. J Immunol Methods. 2011;374:43–6. doi: 10.1016/j.jim.2010.09.037. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Immunology are provided here courtesy of British Society for Immunology

RESOURCES