HIV Databases HIV Databases home HIV Databases home
HIV sequence database

Prediction of T cell epitopes for HIV vaccine development
by computer-driven algorithm

Anne S. De Groot
Gabriel E. Meister
Bill M. Jesdale
Neal Muni
Caroline G.P. Roberts

TB/HIV Research Laboratory, International Health Institute, Brown University School of Medicine, Providence RI, 02912

NOTE: If you are interested in computer algorithms for predicting epitopes in protein sequences, also see Anne S. DeGroot's web site

1. Overview of T cell epitope prediction algorithms

2. Applications of T cell epitope algorithms to HIV research

1. Overview of T cell epitope prediction algorithms

In the past 10 years, several computer-driven algorithms have been devised to take advantage of the alphabetic representation of protein sequence information to search for T cell epitopes. These algorithms search the amino acid sequence of a given protein for characteristics believed to be common to immunogenic peptides, locating regions that are likely to induce cellular immune response in vitro. Given the rapid expansion of sequence data on geographic subtypes (clades) of HIV and individual HIV quasi-species, the application of these algorithms to HIV proteins may significantly reduce the number of regions which would require in vitro testing for immunogenicity, directing research to more promising segments of HIV proteins and thus potentially reducing the time and effort needed to develop HIV vaccines.

Computer-driven algorithms can identify regions of HIV proteins that contain epitopes and are less variable among geographic isolates; alternatively, computer-driven algorithms can rapidly identify regions of each geographic isolate's more variable proteins that should be included in a multi-clade vaccine. Furthermore, computer-driven searches can be weighted to reflect selected HLA alleles that are most representative of geographic populations or subgroups within one geographic area. Computer-driven searches can also be used as a preliminary tool to evaluate the evolution of immune response to an individual's own quasi species. This text will review the development of computer algorithms for T cell epitope prediction, with a particular focus on the novel algorithm EpiMer, and will describe new directions for the application of computer algorithms such as EpiMer to HIV vaccine research.

A. Characteristics of T cell epitopes

Peptides presented in conjunction with class I MHC molecules are derived from foreign or self protein antigens that have been synthesized in the cytoplasm.1, 2, 3 Peptides presented in the context of class II MHC molecules are usually derived from exogenous protein antigens.4, 5, 6 Peptides binding to class I molecules are generally shorter (8-10 amino acid residues) than those that bind to class II molecules (8 to greater than 20 residues). An interpretation of peptides positioned in the binding cleft of class I and class II MHC molecules is shown in Figure 1.

The identification of T cell epitopes within protein antigens has traditionally been accomplished using a variety of methods, including the use of whole and fragmented native or recombinant antigenic protein, as well as the more commonly employed "overlapping peptide" method. The latter method for the identification of T cell epitopes within protein antigens involves the synthesis of overlapping peptides which span the entire sequence of a given protein antigen. These peptides are then tested for their capacity to stimulate T cell cytotoxic or proliferative responses in vitro.

As one might imagine, implementation of the overlapping peptide method is both cost- and labor-intensive. For example, to perform an assay using 15 amino acid long peptides overlapping by 5 amino acids spanning a given antigen of length n (a small subset of all possible 15-mers spanning the protein), one would need to construct and assay(n /5)-1 peptides. Even still, this method does not ensure the identification of all possible T cell epitopes, as potential sites can be "missed" between overlapping fragments.

The first research groups to suggest that computer algorithms based on patterns of amino acids might be used as a tool for discovering T cell epitopes were DeLisi and Berzofsky7 and Rothbard and Taylor8. DeLisi and Berzofsky originally proposed the hypothesis that T cell antigenic peptides are amphipathic structures bound in the MHC groove, with a hydrophobic side facing the MHC molecule and a hydrophilic side interacting with the T cell receptor9. Rothbard and Taylor's algorithm describes a similar periodicity for a smaller number of amino acid residues. The AMPHI algorithm, based on the DeLisi and Berzofsky observations and developed by Margalit et al.10, has been widely used for the prediction of T cell antigenic sites from sequence information alone.

Algorithms such as AMPHI, which are based on the periodicity of T cell epitopes, have been re-evaluated due to recent crystallographic determination of MHC structures with bound peptides. These peptides were demonstrated to be lying extended in the MHC groove, in non alpha-helical conformations11, 12. An explanation of the predictive strength of AMPHI has been provided by Cornette et al.13, based on the periodicity analysis of a table of motifs compiled by Meister et al.14. Essentially, AMPHI describes a common structural pattern of MHC binding motifs, since MHC binding motifs appear to exhibit the same periodicity as an alpha helix. More recently, the rapid expansion of information on the nature of peptides that bind to MHC molecules has led to the evolution of a new class of computer-driven algorithms for vaccine development.

B. Algorithms based on MHC binding motifs

MHC binding motifs are patterns of amino acids that appear to be common to most of the peptides that bind to a specific MHC molecule. For example, a lysine might be required in position N+1 (one amino acid from the amino terminus), and a valine in position N+8, while any amino acid may occur at any of the other positions. In theory, this would explain why MHC molecules are able to present many different peptides from different proteins, yet MHC specificity can still occur. The peptide motif-MHC specificity appears to be due to the interaction of the amino acid side chains of certain conserved "anchor" residues with pockets in the MHC peptide binding cleft (as diagrammed in Figure 1).

Identification of T cell epitopes by locating MHC binding motifs in the sequence of a given protein has been shown to be effective when used to identify immunogenic epitopes for malaria15 and for Listeria monocytogenes16, however the number of regions of any given protein that contain single MHC motifs is usually much too large to be of any use for vaccine development. Furthermore, MHC binding motifs appear to be relatively imprecise: only about one-third of peptides containing one of the current motifs that is said to predict binding to a given class I MHC allele have been shown to be bound by that MHC molecule, and in some cases, epitopes that do not contain known MHC binding motifs have been described17, 18, 19. This may be due to missing information about the requirements for peptide-MHC interactions, or to errors in the descriptions of MHC binding motifs in the literature. In addition, MHC binding is necessary but not sufficient for a peptide to be antigenic; the peptide-MHC complex must still interact with the TCR of a neighboring cell, allowing the induction of a cellular immune response (reviewed in 20).

Since 1992, members of the TB/HIV laboratory at Brown University have been working on the development of a computer algorithm that locates MHC binding motifs in amino acid sequences of HIV proteins. In the process of developing this algorithm, we demonstrated that MHC binding motifs tend to cluster within proteins21. Some of the clustering may be due to the similarity of certain MHC binding motifs to one another, however, dissimilar motifs are also found to cluster. These motif-dense regions appear to correspond with peptides that may have the capacity to bind to a variety of MHC molecules (promiscuous or multi-determinant binders) and to stimulate an immune response in these various MHC contexts as well (promiscuous or multi-determinant epitopes).

The algorithm we developed (EpiMer) uses a library of MHC binding motifs for multiple class I and class II HLA alleles to predict antigenic sites within a protein that have the potential to induce an immune response in subjects with a variety of genetic backgrounds. EpiMer locates matches to each MHC-binding motif within the primary sequence of a given protein antigen. The relative density of these motif matches is determined along the length of the antigen, resulting in the generation of a motif-density histogram. Finally, the algorithm identifies protein regions in this histogram with a motif match density above an algorithm-defined cutoff density value, and produces a list of subsequences representing these clustered, or motif-rich regions (Figure 3). The regions selected by EpiMer may be more likely to act as multi-determinant binding peptides than randomly chosen peptides from the same antigen, due to their concentration of MHC-binding motif matches. An example of a multi-determinant epitope is shown in Figure 4.

The MHC binding motif library used by EpiMer for its searches is updated regularly from the literature. This list can be tailored for a number of different types of searches. For example, one can use the entire MHC binding motif library to identify peptides that contain both MHC Class I and Class II binding motifs; one can restrict the list of binding motifs used in the searches to Class I or Class II, and one can tailor the search to the set of MHC alleles of geographic subpopulation or even those of a single individual.

The utility of computer-algorithm driven predictions for in vitro and in vivo research was recently demonstrated in an analysis of peptides predicted by the EpiMer algorithm from Mycobacterium tuberculosis (Mtb) protein sequences. Twenty-seven of 28 EpiMer peptides derived from Mtb proteins stimulated immune responses in peripheral blood cells from Mtb immune subjects21. There was a good correlation between the number of motifs per peptide and the number of responders to the peptide in a population of Mtb-infected individuals (p < 0.001), and 40 percent of the variation in the relationship between the motifs and the responses could be explained by the presence or absence of MHC binding motifs22. As only about a third of peptides that are predicted using single MHC binding motifs are shown to bind and to stimulate immune responses, the relationship between the EpiMer predictions and the number of responders to the peptides was much better than might have been expected. We believe that the selection of regions that are MHC binding motif-dense increases the likelihood that the predicted peptide contains a "valid" motif, and furthermore, that the reiteration of identical motifs may contribute to peptide binding23.

Additional MHC binding motif-based algorithms have been described by Parker et al.24 and Altuvia et al.25. In these algorithms, binding to a given MHC molecule is predicted by a linear function of the residues at each position, based on empirically defined parameters, and in the case of Altuvia et al., known crystallographic structures are also taken into consideration25, 26. DeLisi et al. have proposed an alternative method of determining MHC binding peptides, based on the free energy relationships of each amino acid in the predicted peptide, and analyzing whether the tertiary structure of the peptide conforms to a predetermined MHC binding peptide configuration27, 28. Finally, Brusic and colleagues are using artificial neural networks to determine the "rules" for binding to MHC molecules from the array of binding peptides that have been described for each of the human HLA alleles29. None of these algorithms have been tested in vivo. Should any of these variations on "motif matching" prove to be accurate predictors of peptides that bind to individual MHC alleles, they may be easily incorporated as subprograms into a clustering algorithm such as EpiMer, and might improve the algorithm's overall predictive capacity.

Several new developments in the prediction of allele-specific binding peptides could be of singular importance to the MHC motif-based algorithms. Hammer et al.30 describe a technique known as "peptide side chain scanning," which they used to predict binding peptides for the MHC allele DRB1*0401. This allowed the construction of a matrix of all possible amino acid side chain effects for a single MHC binding motif, which was later converted into an algorithm able to run through a protein's primary structure and predict, within reasonable error, the binding capacities of all possible peptides of a fixed length to a single MHC molecule. Such matrices can be easily incorporated into computer-driven algorithms: a matrix-based algorithm based on MHC binding motifs that takes into consideration the relative prevalence of amino acids at each of the positions in the peptide is currently being developed in the TB/HIV Research laboratory.

Most of the novel computer-driven algorithms depend on published information on MHC binding motifs. One methodological concern when designing a multiple binding motif-based predictive algorithm is the accuracy of the MHC binding motifs used to predict putative epitopes, and thus the overall validity of the motif database. Previously reported motifs are often redefined in the literature, after peptide truncation and alanine substitution experiments are performed; likewise, new emphasis has been placed on the role of protein processing and on the identification of specific amino acid residues at non anchor sites, which interfere with the relative capacities of peptides to bind to the MHC cleft31, 32. In addition, several MHC binding motif databases have been constructed. Rammensee et al.33 have published a motif database, aided by the alignment of actual MHC binding peptides and known T cell epitopes. A new prediction algorithm based on the Rammensee motifs is has been developed in the TB/HIV Research Laboratory (Bill M. Jesdale and Gabriel E. Meister, unpublished data). Brusic et al.34 have taken this MHC motif library concept further by providing an Internet-accessible database of binding motifs and peptides known to bind with affinity to MHC molecules.

An important consideration when comparing the different computer-driven models described above is that these methods for epitope prediction are not mutually exclusive. As the contributions of side chains and tertiary peptide structure to peptide-MHC binding are better quantified, the development of a computer algorithm that predicts T cell epitopes based on a matrix of side chain information such as the one described by Hammer35 will only be a matter of time. The identification of novel structural features which are able to independently predict peptide binding or immunogenicity, and their subsequent synthesis into a combined algorithm with statistically verifiable predictive capacity, may allow a dramatic reduction in the time and effort required to synthesize and test potential T cell antigenic sites for HIV proteins, by allowing the prediction of sites with a high concentration of antigenic features.

2. Applications of T cell epitope algorithms to HIV research

A. Searching for T cell epitopes

Identification of T cell epitopes that stimulate cell-mediated immunity is essential to HIV vaccine development. The identification of HIV peptide epitopes that contain clusters of MHC binding motifs representing multiple HLA alleles from HIV protein sequences may be useful for HIV vaccine development.

There appear to be more stringent binding criteria for class I-restricted binding peptides, and few multi-determinant class I epitopes have been identified for any pathogen. However, several HIV protein regions that contain multiple overlapping class-II restricted epitopes, also known as "multi-determinant" or multi-determinant peptides, have been identified in mice and humans. Such regions might be important to include in the synthesis of multiple antigenic peptides (MAPS) for HIV vaccine development, particularly if a multi-determinant T cell epitope is required for boosting immune response to B cell epitopes.

The EpiMer algorithm is readily applied to HIV protein sequences. In a recent comparison of EpiMer predictions to published HIV protein T cell epitopes, the EpiMer algorithm was shown to be 2.4 fold more sensitive than the overlapping method for detecting published T cell epitopes for four HIV proteins, gp160, nef, tat, and gag (CGP Roberts et al., manuscript submitted36). A summary of these comparisons of the overlapping method to the EpiMer prediction method is shown in Table 1. A complete list of the MHC binding motifs contained within the most widely used laboratory strains of HIV-1 proteins, based on the MHC binding motif list compiled by the TB/HIV Research Laboratory, is being developed for the TB/HIV Research Laboratory Web Site; a partial list for amino acids 628 to 678 of the HIV-1 BH 10 protein gp160 is shown in Table 2.

The EpiMer algorithm predicted putative T cell epitopes from protein sequences for HIV-1 nef, gp160, gag p55, and tat that required fewer peptides and therefore fewer amino acid residues to be synthesized than either AMPHI-predicted peptides or overlapping peptides. For the four HIV-1 proteins, EpiMer predicted 43 peptide epitopes, AMPHI predicted 68 peptides , and the overlapping peptide method (20 amino acid long peptides overlapping by 10 amino acids) would have required 161 peptides. Details (amino acid start and stop, number of MHC binding motifs) of the predicted proteins are available36. Regions of HIV proteins that contain as many as 20 to 30 MHC binding motifs can be identified using EpiMer. Such regions should be good candidates for inclusion in a subunit HIV vaccine.

Application of MHC binding motifs to HIV vaccine development may be restricted by the amount of sequence variation in individual quasi-species, HIV strains, and HIV clades, as well as by the MHC background of the target populations. One might consider evaluating regions of MHC clustering that occur in sites of low HIV sequence variability, as shown in Figure 5. The region 130 to 160, which has a great deal of inter-strain variation described by the variability plot, might best be avoided for subunit HIV vaccine development. HIV peptide epitopes which contain multiple MHC binding motifs, either conserved across HIV strains or derived from several different HIV strains, may be ideal candidates for inclusion in a multi-subunit vaccine.

An alternative to searching for conserved regions of HIV proteins would be to identify regions of the sequences that predominate in the clades that are most likely to be presented in the context of the MHC molecules of the geographic sub populations of interest. If MHC binding motif-based peptides are to be used in subunit vaccines, the best strategy may be to custom design the peptides, using the sequences of the HIV clades that are prevalent in that populations and the set of MHC alleles that are also prevalent in that population. We have proposed one method of weighting predicted peptides by the prevalence of MHC binding motifs37; the DeLisi laboratory has proposed yet another method29.

B. Evaluating the effect of immune response on the evolution of HIV

An additional application of EpiMer might be to evaluate the effect of pressure from the immune system of the individual on the HIV quasi-species of that individual. Ongoing research has been suggested that rapid progression might be related to the capability of the virus to avoid immune detection through variation at the MHC binding anchors of a given T cell epitope, or through variation at the TCR binding site. To date, several laboratories have described in vitro evidence for escape mutations in the epitope of a given individual38, 39. We have examined the evolution of class I MHC-binding peptides in HIV-1 quasi-species in the contexts of clinically quiescent HIV-1 infection and rapid progression to advanced disease, by implementing EpiMer to predict MHC binding peptides from primary protein sequences40. Using each patient's own MHC allele subset to tailor the EpiMer searches, regions of the patient's own quasi-species can be searched for putative MHC-binding peptides. This search can be repeated for quasi-species isolated from the patient at each of several timepoints, and analyzed for patterns of MHC-binding motif escape, or replacement by an alternate binding region. This novel approach identifies regions of HIV quasi-species that should be the focus of binding assays and epitope mapping which may improve our comprehension of host immune response to HIV.


Identification of T cell epitopes that stimulate cell-mediated immunity is essential to HIV vaccine development. Computer driven algorithms for T cell epitope prediction appear to provide rapid and relatively inexpensive means of T cell identification for in vitro investigations. The EpiMer algorithm, described in this text and in more detail in reference 14, identifies peptide epitopes from HIV proteins by identifying clustering of MHC binding motifs within the protein sequences. Peptide epitopes containing multiple MHC binding motifs may be immunogenic in individuals from a variety of genetic backgrounds. Identification of such clusters may improve the immunogenicity of a given peptide, and permit the development of a subunit vaccine that can induce immunity to multiple strains and clades of HIV.

Identification of T cell epitopes within the sequences from quasi-species of HIV-infected individuals may also permit the investigations of the evolution of HIV in response to host immune pressure. While the relationship between MHC binding motifs and immunogenicity is less than absolute, the utilization of computer driven algorithms such as EpiMer may permit the identification of regions of increased interest for in vitro confirmation of HIV evolution within an individual or within a given geographic subpopulation.

Figure 1: Illustration of MHC I and II complexes with bound peptide.
Key amino acid residues within the MHC molecules interact with "anchor residues" on the bound peptide, conferring a peptide binding specificity on the specific MHC allele. The arrangement and characteristics of the anchor residues within the peptide are collectively termed the "MHC binding motif." A library of such motifs has been generated for both Class I and Class II MHC alleles.
Figure 2.The overlapping peptide method (hypothetical protein)
The sequences required for a typical set of overlapping peptides for a region of the gp160 protein are shown in light gray in (a). The primary protein sequence is shown in black (b), with known fine-mapped epitopes indicated in medium gray. Sequences for a hypothetical prediction method are shown in (c), and where these peptides fully contain the epitopes, this is indicated with medium gray. In this example, the original sequence is 140 amino acids long. The overlapping method requires nine peptides of 20 amino acids (180 amino acids total), and should reveal both epitopes. The hypothetical prediction method requires 4 peptides of varying lengths (53 amino acids total), but should reveal only one epitope.
Figure 3. MHC binding motif clustering for gp160 of strain BH 10 Figure 3
A histogram of the density of MHC-binding motif matches along the sequence of the gp160 protein of HIV BH 10 is shown here, to illustrate the EpiMer method of putative epitope identification. For this analysis, both class I and II MHC-binding motifs were used in our search. Peptides that include peaks of motif density, such as the 10- to 25-mers including amino acids 19 to 34 (14 motifs), 36 to 54 (14 motifs), 84 to 95 (6 motifs), 115 to 127 (7 motifs) and168 to 185 (22 motifs) shown in this example, are predicted as putative T cell epitopes by the EpiMer algorithm. The EpiMer peptides are shown in bold, and are slightly shorter than the stated predictions because the midpoint of the amino terminal 11-mer reading frame of the predicted peptide to the midpoint of the carboxy terminal 11-mer reading frame are designated in this picture, rather than the full length of the predicted peptide.
Figure 4. Multi-determinant peptide. Figure 4
An example of a multi-determinant peptide is shown. The MHC binding motifs for the predicted peptide are shown at the right. Note that HLA DRB1*0101 motif occurs 6 times, that 5 unique motifs can be identified, and that a total of 12 potential MHC binding regions are contained within this protein sequence.
Figure 5. gp160 (variability plot) Figure 5
The mean variability of the 11-residue segments of known gp160 sequences (
Los Alamos HIV Sequence Database) are shown as well. Variability = (number of different amino acids at a given position)/(frequency of the most common amino acid at that position).

Table 1. Efficiency and sensitivity of the Overlapping method, compared to EpiMer, for the HIV proteins nef, gag, gp160, and tat.

Overlapping EpiMer
Per Cent Efficiency 60% 62%
Range 43%-100% 61%-64%
Per Cent Sensitivity 100% 59%
Range 100%-100% 22%-86%
Average Sensitivity per amino acid 2.7 4.9
Range 0.6-6.4 1.3-8.1
Average Ć sensitivity per AA (ref) 2.4

Efficiency, = (total length, in amino acid residues, of the peptides that overlap by at least eight or eleven amino acid residues with Class I or Class II published epitopes, respectively)/(total length, in amino acid residues, of all putative epitopes identified by the algorithm in question).
Sensitivity, S, (number of published epitopes that were predicted by the algorithm in question)/(total number of published epitopes for the protein).
Sensitivity per amino acid (SAA), = 1000 x (Sensitivity)/(total length, in amino acids, of peptides to be synthesized). Ć Sensitivity/AA, (ĆSAA) = (Sensitivity per amino acid residue for a given method)/(Sensitivity per amino acid residue of the overlapping peptide method).

Table 2. A partial list of putative MHC binding motifs contained amino acids 628-678 of the HB 10 strain of HIV-1, showing regions of clustering
Amino Acids Amino Acid Sequence Motif Match
629 - 638 MEWDREINNY HLA-B*44
629 - 638 MEWDREINNY HLA-DRB1*0301
629 - 638 MEWDREINNY HLA-DRB1*0801
630 - 638 EWDREINNY HLA-A1
631 - 639 WDREINNYT HLA-DRB1*0401(DR4Dw4)
633 - 641 REINNYTSL HLA-B*40012
633 - 641 REINNYTSL HLA-B40
633 - 641 REINNYTSL HLA-B44
633 - 641 REINNYTSL HLA-Cw*0301
635 - 644 INNYTSLIHS HLA-DRB1*1501
637 - 645 NYTSLIHSL HLA-A24
637 - 645 NYTSLIHSL HLA-Cw*0401
637 - 645 NYTSLIHSL HLA-Cw*0602
637 - 645 NYTSLIHSL HLA-Cw*0702
637 - 645 NYTSLIHSL HLA-DQ3.1
638 - 646 YTSLIHSLI HLA-DQ3.1

679 - 688 LWYIKLFIMI HLA-DRB1*0801
679 - 688 LWYIKLFIMI HLA-DRB1*1501
680 - 688 WYIKLFIMI HLA-A24
680 - 688 WYIKLFIMI HLA-Cw*0301
680 - 688 WYIKLFIMI HLA-DPA1*0102/DPB1*0201
681 - 689 YIKLFIMIV HLA-Cw*0602
681 - 689 YIKLFIMIV HLA-DPA1*0102/DPB1*0201
681 - 689 YIKLFIMIV HLA-DRB1*0401(DR4Dw4)
682 - 690 IKLFIMIVG HLA-DRB1*0401(DR4Dw4)
682 - 691 IKLFIMIVGG HLA-DRB1*1501
684 - 692 LFIMIVGGL HLA-Cw*0401
684 - 692 LFIMIVGGL HLA-Cw*0602
685 - 693 FIMIVGGLV HLA-DRB1*0101
686 - 695 IMIVGGLVGL HLA-DRB1*1501
687 - 695 MIVGGLVGL HLA-A*0205
688 - 696 IVGGLVGLR HLA-A*3302
688 - 696 IVGGLVGLR HLA-DQ3.1
688 - 697 IVGGLVGLRI HLA-A68
689 - 697 VGGLVGLRI HLA-B*5101
689 - 697 VGGLVGLRI HLA-B*5102
689 - 697 VGGLVGLRI HLA-B*5103
689 - 697 VGGLVGLRI HLA-DQ3.1
689 - 697 VGGLVGLRI HLA-DRB1*0101
690 - 698 GGLVGLRIV HLA-B*5102
690 - 698 GGLVGLRIV HLA-B*5103
691 - 699 GLVGLRIVF HLA-A3
691 - 699 GLVGLRIVF HLA-B*1501
692 - 700 LVGLRIVFA HLA-DRB1*0401(DR4Dw4)
692 - 701 LVGLRIVFAV HLA-DRB1*0801
693 - 700 VGLRIVFA HLA-B*7801
693 - 701 VGLRIVFAV HLA-B*5102
693 - 701 VGLRIVFAV HLA-B*5103

Table 2. Amino acid sequences of the EpiMer-predicted epitopes for the amino acid residues 628 to 678 of the gp160 protein are listed (in boldface), as are the individual MHC-binding motif matches found within each peptide. These two predicted epitopes overlap with published epitopes for this same HIV-1 strain.


1 Matsumura, M, Fremont, DH, Peterson, PA and Wilson, IA. Emerging principles for the recognition of peptide antigens by MHC class I molecules. Science 1992; 257: 927-934.
2 Germain, RN and Margulies, DH. The biochemistry and cell biology of antigen processing and presentation. Annu Rev Immunol 1993; 11: 403-450.
3 Falk, K, Rštzschke, O, Stevanovic, S, Jung, G and Rammensee, H-G. Allele-specific motifs revealed by sequencing of self-peptides eluted from MHC molecules. Nature 1991; 351: 290-296.
4 Chicz, RM, Urban, RG, Gorga, JC, Vignali, DAA, Lane, WS and Strominger, JL. Specificity and promiscuity among naturally processed peptides bound to HLA-DR alleles. J Exp Med 1993; 178:27-47.
5 Srinivasan, M, Domanico, SZ, Kaumaya, PTP and Pierce, SK. Peptides of 23 residues or greater are required to stimulate a high affinity class II-restricted T cell response. Eur J Immunol 1993; 23:1011-1016.
6 Rštzschke, O and Falk, K. Origin, structure and motifs of naturally processed MHC class II ligands. Curr Opin Immunol 1994; 6:45-51.
7 DeLisi, C and Berzofsky, JA. T-cell antigenic sites tend to be amphipathic structures. Proc Natl Acad Sci USA 1985; 82:7048-7052.
8 Rothbard, JB and Taylor, WR. A sequence pattern common to T cell epitopes. EMBO J 1988; 7:93-100.
9 Spouge, JL, Guy, HR, Cornette, JL, Margalit, H, Cease, K, Berzofsky, JA, and DeLisi, C. Strong conformational propensities enhance T cell antigenicity. J Immunol 1987; 138:204-212.
10 Margalit, H, Spouge, JL, Cornette, JL, Cease, KB, DeLisi, C and Berzofsky, JA. Prediction of immunodominant helper T cell antigenic sites from the primary sequence. J Immunol 1987; 138:2213-2229.
11 Cornette, J. L., Margalit, H., DeLisi, C., & Berzofsky, J. A. in The Amphipathic Helix, R. M. Epand, Ed. CRC Press, Boca Raton, FL, 1993, 333-345.
12 Stern, LJ, Brown, JH, Jardetzky, TS, Gorga, JC, Urban, RG, Strominger, JL, ,Wiley, DC. Nature 1994; 368: 215-221.
13 Cornette, JL, Margalit, H, DeLisi, C, & Berzofsky, JA. Proc Nat Acad Sci 1995, 92: 8368-3872.
14 Meister, GE, Roberts, CGP, Berzofsky, JA, De Groot, AS. Two novel T cell epitope prediction algorithms based on MHC-binding motifs; comparison of predicted and published epitopes from Mycobacterium tuberculosis and HIV protein sequences. Vaccine 1995; 13: 581-591.
15 Hill, AVS, Elvin, J, Willis, AC, Aidoo, M, Allsopp, FM, Gotch, XM, Gao, T, Takiguchi, M, Greenwood, BM, Townsend, ARM, et al., Molecular analysis of the association of HLA-B53 and resistance to severe malaria Nature 1992; 340:434-439.
16 Pamer, EG, Harty, JT, Bevan, MJ, Precise prediction of a dominant class I MHC-restricted epitope of Listeria monocytogenes Nature 1991; 353: 852- .
17 Lipford, GB, Hoffman, M, Wagner, H, and Heeg, K. Primary in vivo responses to ovalbumin. Probing the predictive value of the Kb binding motif. J Immunol. 1993; 150:1212-1222.
18 Nijman, HW, Houbiers, JGA, Vierboom, MPM, van der Burg, SH, Drijfhout, JW, D'Amaro, J et al. Identification of peptide sequences that potentially trigger HLA-A2.1-restricted cytotoxic T lymphocytes. Eur J Immunol 1993; 23:1215-1219.
19 Calin-Laurens V, Trescol-BiŽmont M-C, Gerlier D and Rabourdin-Combe, C. Can one predict antigenic peptides for MHC class I-restricted cytotoxic T lymphocytes useful for vaccination? Vaccine 1993; 11:974-978.
20 Sercarz, E, Lehmann, P, Ametani, A, Benichou, G, Miller, A, and Moudgil, K. Dominance and crypticity of T cell antigenic determinants. Annu Rev Immunol 1993; 11:729-766.
21 Gabriel E. Meister, Caroline G.P. Roberts, Jay A. Berzofsky, Anne S. De Groot, Mycobacterium tuberculosis peptide epitopes predicted by two novel epitope identification algorithms, Vaccines 95, Cold Spring Harbor Laboratory, Cold Spring Harbor NY, 1995.
22 De Groot AS, Carter EJ, Roberts CGP, Edelson BT, Jesdale BM, Meister GE, Houghten RA, Montoya, J, Romulo RC, Berzofsky JA, Ramirez BDLL A novel algorithm for the efficient identification of T cell epitopes: prediction and testing of candidate TB vaccine peptides in genetically diverse populations, to be published inVaccines 96 Cold Spring Harbor Laboratory, Cold Spring Harbor NY 1995.
23 Sette, A., Sidney, J., Albertson, M., Miles, C., Colon, S.M., Pedrazzini, T., Lamont, A.G., and Grey, H.M. A novel approach to the generation of high affinity class II-binding peptides. J. Immunol 1990, 145, 1809-1813.
24 Parker, KC, Bednarek, MA, and Coligan, JE. Scheme for ranking potential HLA-A2 binding peptides based on independent binding of individual peptide side chains. J. Immunol 1994, 152:163-175.
25 Altuvia, Yael, Schueler, Ora, Margalit, H, Ranking potential binding peptides to MHC molecules by a computational threading approach, J Mol Biol 1995; 249:244-250.
26 Altuvia, Y, Berzofsky, JA, Rosenfeld, R, and Margalit, H. Sequence features that correlate with MHC restriction. Molecular Immunology 1994, 31:1-19.
27 S. Vajda and C. DeLisi, Determining Minimum Energy Conformations of Polypeptides by Discrete Dynamic Programming. Biopolymers1990; 29:1755-1772.
28 Gulukota, K, Vajda, S, and DeLisi, C. Peptide Docking using dynamic programming, J Comp Biol, submitted.
29 Brusik, V, Rudy, G, Harrison LC, Prediction of MHC binding peptides using artificial neural networks, In: Stonier RJ, Yu XS, eds., Complex systems, mechanisms of adaption. Amsterdam; IOS Press 1994:253-260.
30 Hammer J, Bono E, Gallazzi F, Belunis C, Nagy Z, and Sinigaglia F. Precise prediction of major histocompatibility complex class II-peptide interaction based on peptide side chain scanning. J Exp Med 180; 1994:2353-2358.
31 Boehncke, W-H, Takeshita, T, Pendleton, CD, Houghten, RA, Sadegh-Nasseri, S et al. The importance of dominant negative effects of amino acid side chain substitution in peptide-MHC molecule interactions and T cell recognition. J Immunol 1993; 150:331-341.
32 Dicke, LR, Aldrich, C, Jameson, CR, Moomaq, BC, Praminic, CK, Coyle, GN, DeMartino, MJ, Bevan, JM. Proteolytic processing of ovalbumin and B-galactosidase by the proteasome to yield antigenic peptides, J Immunol 1994, 152:3884-3894.
33 Rammensee, H-G, Friede, T & Stevanovic, MHC ligands and peptide motifs: first listing. S. Immunogenetics 1995; 41: 178-228.
34 Brusic, V, Rudy, G, Harrison, LC. MHCPEP: a database of MHC-binding peptides. Nuc Acids Res 1994, 22, 3663-3665.
35 Hammer J., New methods to predict MHC Binding sequences within protein antigens, Current Opinion in Immunology 1995, 7:263-269.
36 Roberts, CGP, Meister, GE, Jesdale, BM, Lieberman, J, Berzofsky, J, and AS De Groot, Identification of HIV peptide epitopes by a novel algorithm, submitted 1995.
37 De Groot et al., Predicting Mtb epitopes for geographic subpopulations: progress on the development of a novel TB vaccine for West Africans. manuscript.
38 Philips RE, McMichael, AJ, How does the HIV escape cytotoxic T cell immunity, Chem Immunol 1993. 56: 150-164.
39 Safrit, JT, Lee, AY, Andrews, CA, Koup RA. A region of the third variable loop of HIV -1 gp120 is recognized by HLA b7 restricted CTLs from two acute seroconversion patients, J Immunol. 1994, 153: 3822-3830.
40 Jesdale, BM, Santiago, MLO, Wolinsky, SM, Korber, BT, Koup, RA, and De Groot, AS, Evolution of HIV-1 Quasi-species: sequence diversity in predicted class I MHC binding regions over time (env C2-V5), manuscript.

last modified: Wed Jan 13 15:15 2010

Questions or comments? Contact us at

Operated by Triad National Security, LLC for the U.S. Department of Energy's National Nuclear Security Administration
© Copyright Triad National Security, LLC. All Rights Reserved | Disclaimer/Privacy

Dept of Health & Human Services Los Alamos National Institutes of Health