Allocation of helper T-cell epitope immunodominance according to three-dimensional structure in the human immunodeficiency virus type I envelope glycoprotein gp120.

The specificity and intensity of CD4(+) helper T-cell responses determine the effectiveness of immune effector functions. Promiscuously immunodominant helper T-cell epitopes in the human immunodeficiency virus (HIV) envelope glycoprotein gp120 could be important in the development of broadly protective immunity, but the underlying mechanisms of immunodominance and promiscuity remain poorly defined. In this study, gp120 helper T-cell epitopes were systematically mapped in CBA/J and BALB/c mice by restimulation assays using a set of overlapping peptides spanning the entire sequence of the gp120 encoded by HIV strain 89.6. The results were analyzed in the context of the HIV gp120 structure determined by x-ray crystallography. One major finding was that all of the promiscuously immunodominant gp120 sequences are located in the outer domain. Further analyses indicated that epitope immunogenicity in the outer domain correlates with structural disorder in adjacent N-terminal segments, as indicated by crystallographic B-factors or sequence divergence. In contrast, the correlation was poor when the analysis encompassed the entire gp120 sequence or was restricted to only the inner domain. These findings suggest that local disorder promotes the processing and presentation of adjacent epitopes in the outer domain of gp120 and therefore reveal how three-dimensional structure shapes the profile of helper T-cell epitope immunogenicity.

The HIV/AIDS 1 epidemic continues to be a serious health threat worldwide. Although much progress has been made in understanding the HIV virus since it was first isolated in 1983, no effective vaccines against HIV are yet available. It is well documented that cytotoxic T lymphocytes (CTLs) constitute an important effector mechanism for clearance and control of HIV infections and that a specific infectivity-neutralizing antibody can prevent HIV infections (1). CD4 ϩ helper T lymphocytes are required for the generation of these effector mechanisms. Therefore, an effective vaccine must be able to elicit potent helper T-cell responses. Although immunodominant epitopes are consistently exposed to immune surveillance, they may not promote development of protective immunity. Thus, it is of interest to characterize immunodominant epitopes and identify the mechanisms that are responsible for immunodominance.
Immunodominant epitopes have been identified by strength and frequency in the lymphocyte recall response (2). An immunodominant epitope that is frequently observed in sensitized humans may also be described as promiscuous or universal if the responding humans have different alleles of MHC class II protein (3)(4)(5). In many mapping studies, lymphocytes from a group of inbred mice were pooled, and thus the frequency of response was not available (6 -8). In these studies, epitope immunodominance was based strictly on stimulation index (SI), the level of peptide-stimulated proliferation divided by the level of unstimulated proliferation. In studies on outbred populations, strength and frequency have been treated independently (9 -11). In human subjects sensitized to tetanus and diphtheria toxoids, the SI for the most common epitopes correlated with SI for the whole antigen, suggesting that the dominant epitope-specific responses were responsible for most of the response to the antigen (11). Because these subjects were heterogeneous with respect to MHC alleles, peptide affinity for the MHC protein probably was not the controlling factor.
Available methods of predicting T-cell epitopes score antigen sequences by preference for binding to MHC antigen-presenting proteins (12)(13)(14)(15)(16). These methods have been more successful for MHC class I-restricted CTL epitopes than for MHC class II-restricted helper T-cell epitopes (14). CTL epitopes typically are few and consistent among individuals with a given MHC haplotype. In contrast, a large fraction of antigen sequence is able to stimulate helper T-cell proliferation, and there is a great variation in the breadth and intensity of the response, even among inbred animals. Moreover, CTL epitopes seem to be largely context-independent (14), whereas helper T-cell epitopes often become cryptic in an unnatural context (17)(18)(19)(20)(21)(22). At least two differences in the processing and presentation machinery contribute to the different behaviors of the two types of epitopes. First, the ends of the peptide-binding groove in MHC class I proteins are closed, whereas they are open in MHC class II proteins (23). Second, the cytoplasmic proteasome generates peptides for presentation to CTL (24,25), whereas lysosomal proteases generate peptides for presentation to helper T-cells (26,27).
Antigen primary sequence influences helper T-cell epitope immunodominance at several levels of processing and presentation. First, landmark studies have shown that a suitable arrangement of two or three residues in the antigen primary sequence plays an important part in determining binding affinity for the MHC class II protein (28 -31). The various MHC alleles exhibit distinct anchor residue preferences, and thus epitope patterns can be quite different between individuals with different alleles. Second, residues beyond the MHC-binding sequence influence binding to the MHC protein (32,33). Third, the primary sequence has the capacity to influence epitope immunodominance at the level of tertiary structure in the antigen. For example, the overall immunogenicity of tetanus toxoid antigen is determined by susceptibility to proteolysis by a single protease at a single cleavage site (34). Thus, presentation of any and all epitopes of this antigen depends on this first cleavage step, which may be necessary to destabilize the protein structure and facilitate downstream processing. Nevertheless, almost all epitope mapping studies have focused solely on how primary sequence affects immunodominance at the level of binding to the MHC protein.
The limited number of systematic mapping studies and the recent crystallographic determination of the HIV gp120 structure (35) warrant a systematic examination of helper T-cell epitope immunodominance in gp120. Several studies identified helper T-cell epitopes in gp120 and related molecules, gp140 and gp160, in experimental animals and infected humans. However, the early studies in mice did not systematically sample the entire gp120 sequence (36,37). Studies in humans were even less systematic in that they sampled individuals who were heterogeneous in MHC haplotype, had been exposed to undefined HIV strains, and were probably at various levels of disease progression (9, 38). One recent study systematically characterized the specificity of T-cell clones derived from immunized C57BL/6 mice and then interpreted the results in light of the crystal structure of gp120 (39). These authors noted the frequent occurrence of helper T-cell epitopes near exposed strands of gp120 and concluded that the pattern could be related to antigen processing. This conclusion is consistent with our previous work suggesting that structurally disordered regions within protein antigens direct presentation of adjacent sequences by providing preferred sites of proteolytic cleavage (40,41). We propose this mechanism as a theory to complement the well established influence of peptide affinity for MHC class II proteins on helper T-cell epitope immunodominance.
To obtain a complete picture of HIV gp120 helper T-cell epitopes, we have mapped immune epitopes in CBA/J and BALB/c mice using splenocyte proliferation in response to overlapping synthetic peptides. The resulting gp120 helper T-cell epitope patterns were then correlated with domain structure and segmental disorder in gp120.

MATERIALS AND METHODS
Experimental Animals-Pathogen-free CBA/J (H-2 k ) and BALB/c (H-2 d ) female mice 8 -12 weeks of age were purchased from the Jackson Laboratory (Bar Harbor, ME). Each animal was randomly assigned to an experimental group.
Expression of Recombinant gp120 in Insect Cells-The cDNA encoding gp120 was obtained by polymerase chain reaction amplification using the p89.6 clone provided by Ronald Collman as a template (42). Primer design incorporated restriction enzyme cleavage sites and encoded a C-terminal hexahistidine fusion. The amplification product was cleaved with restriction enzymes BamHI and SphI and ligated into pFastBac-1 (Life Technologies, Inc.). The nucleotide sequence of clone pFBgp120-2 contained a number of mutations, all of which were silent by comparison with the published sequence (42). Escherichia coli DH10Bac cells were transformed with pFBgp120-2 for transposition of the insert into the Bacmid DNA according to the manufacturer's recommendations (Life Technologies, Inc. Bac-to-Bac Baculovirus Expression System). Spodoptera frugiperda (Sf9) cells were transfected with Bacmid DNA using Life Technologies, Inc. Cellfectin. The passage three viral stock was optimized with respect to virus titer and time of infection in High Five cells (Invitrogen) by analysis of gp120 expression using Western blots with IgG from pooled HIV patient sera (James Robinson, Tulane Pediatrics). For protein production, infection was initiated with 0.1 ml of passage three stock/ml of 2 ϫ 10 6 High Five cells and allowed to proceed at 27°C for 72 h. The cells were sedimented at 6,000 ϫ g, and the supernatant was filtered through a 0.4-m filter and stored at Ϫ80°C.
Purification of Recombinant gp120 -The gp120-containing supernatant from insect cells was made 1% in Triton X-100, and gp120 was purified by affinity for human monoclonal gp120 antibodies A32 and 17B (43) coupled to an Affi-gel Matrix (Bio-Rad) and equilibrated with phosphate-buffered saline. Column-bound proteins were washed with phosphate-buffered saline to remove the detergent and then eluted with 3 M MgCl 2 . The eluted material was concentrated using a Centricon-30 (Amicon), dialyzed into phosphate-buffered saline, and analyzed by SDS-polyacrylamide gel electrophoresis, enzyme-linked immunosorbent assay, and Western blot.
Peptides-Forty-four overlapping 20-mer and three 19-mer peptides spanning the entire sequence of HIV gp120 (strain 89.6) were synthesized by Core Laboratory, Louisiana State University Health Sciences Center (New Orleans, LA). Because it is unclear whether a given cysteine-containing epitope is recognized primarily as the thiol or cysteinylated derivative (44), all cysteine residues were incorporated as the acetamide derivative. The acetamide derivative represents a compromise in terms of size and polarity, and its inert chemistry favors reproducibility.
Immunization and Necropsy Procedure-Mice were intranasally immunized with gp120 in combination with mutant (R192G) heat-labile toxin from enterotoxigenic E. coli (mLT) as an adjuvant (45). Each mouse received 10 l of protein (1 mg/ml) and 10 l of mLT (0.5 mg/ml) in phosphate-buffered saline separately. Control mice received the same amount of mLT only. Mice were boosted twice using the same protocol at intervals of 7 days. One week after the second boosting, mice were euthananized with drops of Metofane into noses. A blood sample was obtained immediately via cardiac puncture. The sera were separated by centrifugation and stored at Ϫ20°C. The abdominal cavity of each mouse was opened aseptically, and the spleen was removed for lymphocyte isolation by gently teasing against a stainless steel mesh.

T-cell Epitope Immunodominance Governed by Structure in gp120
The cells were treated with red blood cell lysing buffer for 3 min at room temperature and washed three times in RPMI 1640 medium, viability was determined by trypan blue exclusion, and the number of cells was adjusted to 4 ϫ 10 6 cells/ml in RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum, 100 units/ml of penicillin, 100 g/ml of streptomycin, and 2 M L-glutamine (working medium). Lymphocyte Blastogenesis-The splenocytes were distributed into 96-well flat-bottom tissue culture plates at 4 ϫ 10 5 cells/well in RPMI 1640 working medium. Duplicate cultures were stimulated with a single peptide or protein (gp120 or mLT). The cultures were incubated for 3 days at 37°C in a 5% CO 2 atmosphere, labeled with 1.0 Ci of tritiated thymidine/well for another 18 h, and harvested onto glass wool fiber filters using a cell harvester. Tritiated thymidine incorporation into cellular DNA was measured in a liquid scintillation counter, and the values for duplicate cultures were averaged. The result for each peptide in each mouse was expressed as the SI, which is the quotient of counts/min in stimulated wells divided by counts/min in medium control wells.

RESULTS AND DISCUSSION
HIV gp120 Helper T-cell Epitope Patterns in CBA/J and BALB/c Mice-Groups of 10 mice were intranasally immunized with gp120, and the splenocytes from each mouse were stimulated in vitro with individual overlapping peptides spanning the sequence of gp120 (Table I). Mice whose splenocytes were stimulated with SI Ͼ 4 were regarded as having a positive response. No more than one of four CBA/J control mice or one of six BALB/c control mice responded to any peptide with SI Ͼ 4. The helper T-cell epitope patterns for gp120 in CBA/J and BALB/c mice were similar but not identical (Fig. 1). The differ-ence in epitope pattern between the mouse strains can be explained by the fact that the two strains have different MHC haplotypes and genetic background.
A traditional definition of immunodominance based on the number of individuals responding is not adequate for the ranking of 47 peptides using data from only 10 mice. For this work, the immunodominant sequences were assigned as the 10 peptides having the highest average SI within each group of CBA/J or BALB/c mice (Table II). These include all but one peptide (peptide 39 in BALB/c mice) that stimulated a majority of mice in each strain ( Fig. 1 and Table II). Therefore, essentially the same peptides were identified whether immunodominance was analyzed by strength or frequency in these two groups of inbred mice.
Several gp120 sequences were promiscuously immunodominant. Three immunodominant sequences (corresponding to peptides 30, 32, and 38) were identified in both CBA/J and BALB/c mice, and several additional sequences that were immunodominant in a single strain overlap each other (corresponding to peptides 27-28, 28 -29, and 46 -47). Some of these sequences substantially overlap immunogenic sequences identified in other studies, which further emphasizes their promiscuity. Sequences corresponding to peptides 27-30 and 38 -41 also were identified as epitope "hot spots" in C57BL/6 mice immunized with gp140 (39), and sequences corresponding to peptides 28 and 30 were identified as immunogenic in HIVinfected humans (9,38).
What structural feature is shared by the promiscuously imunodominant sequences? One possibility is that they contain multiple clustered epitopes that each satisfy different MHC II specificities (36). In this scenario, we assume that the primary sequence dictates presentation and conclude that the epitope cluster contains two or more sequences that bind well to distinct MHC II. Alternatively, the promiscuously dominant epitopes are more available for presentation because antigen processing preferentially exposes them. In this case, we assume that many sequences bind adequately, but only some sequences are preferentially exposed. Allocation of promiscuously dominant epitopes in gp120 suggests a shared structural feature rather than epitope clustering. All of the promiscuously dominant sequences are in the outer domain identified in the crystal structure by Kwong et al. (Ref. 35; Figs. 2, A and B, and 3). The uniquely dominant sequences (corresponding to peptides 6, 17,   Correlation of Epitope Immunogenicity with Local Disorder in the Outer Domain of gp120 -Inspection of the profiles of epitope frequency and crystallographic B-factors suggested that the promiscuous epitopes in the outer domain correlated with segments of local structural disorder. In both BALB/c and CBA/J mice, each of four peaks of epitope frequency in the outer domain overlaps a peak in B-factor (Fig. 2, A and B).
Promiscuously immunogenic sequences corresponding to peptides 28 -30, 38, 46 -47, and 32-38 overlap the disordered segments associated with each of the hypervariable loops, V3, V4, and V5, and a small disordered segment in C3, respectively. Similar patterns of epitopes and corresponding disordered segments have been reported for hen egg lysozyme (40), staphylococcal nuclease (40), cytochrome c (40), diphtheria toxin (11), tetanus toxin (11), and Hsp10s from mycobacteria (40) and bacteriophage T4. 2 The disordered segments are likely to be preferentially cleaved by endoproteolytic enzymes, and thus they could provide entry points for antigen processing. Adjacent epitopes might then become more accessible to binding by MHC proteins, resulting in their preferential presentation to T-cells. Numerous reports suggest that MHC proteins bind to antigens before proteolytic processing is complete (46 -50); thus, the structural context has an opportunity to modulate presentation.
Correlation between epitope immunogenicity and local structural disorder in gp120 was analyzed by determining the Pearson correlation coefficient for profiles of epitope frequency and crystallographic B-factor. This strategy would quantify the influence of structure for a given spatial relationship between the site of proteolytic cleavage and the epitope. Epitope mapping data for CBA/J and BALB/c mice were combined to reduce the influence of MHC II specificity (Fig. 2). Previously, we found a correlation between local structural disorder and C-terminally flanking epitopes in hen egg lysozyme (41). Development of the correlation depended on introducing an offset in epitope mapping data, such that local structural disorder is correlated with epitopes 8 residues C-terminal. In gp120, only a weak correlation between epitope frequency and B-factors was found when the analysis was applied to the complete sequence. The value of r never exceeded 0.35 (r max ) over the range of offset from Ϫ20

FIG. 2. Profiles of epitope frequency and structural parameters in HIV gp120.
Four peaks of epitope frequency in both CBA/J and BALB/c mice correspond to promiscuously immunodominant sequences of peptides 28-30, 32-33, 38, and 46-47 (A and B). Epitope frequencies for the two mouse strains were combined to reduce the influence of MHC alleles (C and D). Peaks of epitope frequency overlap peaks of disorder in the outer domain, whereas a relationship is much less evident in the inner domain (C). Likewise, peaks of epitope frequency overlap dips in sequence conservation in the outer domain, whereas a relationship is much less evident in the inner domain (D). Visual impressions of correlation were confirmed by calculation of Pearson correlation coefficients (see Fig. 4, Table III, and text). Epitope frequency indicates the number of mice for which the residue occurred in a peptide that obtained an SI Ͼ 4. Because the peptides overlap by 10 residues, each residue occurs in two peptides, and therefore some residues have epitope frequencies greater than the number of mice/group. Crystallographic B-factors for backbone amide nitrogen atoms were from the Protein Data Bank (1GC1) or assigned a value of 100 Å at positions not included in the crystal structure. Sequence conservation indicates the fractional identity to a consensus sequence generated from 23 gp120 sequences (41). The profile of conservation was smoothed by a 17-residue moving window average. to 20 (Fig. 4 and Table III). However, as noted above, all of the promiscuously dominant epitopes are in the outer domain, suggesting that antigen structure exerts greater influence in the outer domain. Thus, the correlation coefficient was reevaluated using data only from the outer domain, resulting in an r max of 0.64 for an offset of Ϫ8.
Similar values of offset at r max for hen egg lysozyme and the outer domain of gp120 suggest that the N-terminal end of the C-terminal cleavage product most often contains the associated immunodominant epitope. As in gp120, strongly immunogenic sequences in hen egg lysozyme tend to be located 8 amino acids C-terminal from structurally disordered segments (r max ϭ 0.51 for offset of Ϫ8) (41). The requirement for an 8-residue offset has several possible explanations. MHC proteins could bind at sites of intermediate disorder because these sites have a smaller loss of conformational entropy. Alternatively, intermediate disorder could coincide with a sequence composition that maximizes the probability of an MHC-binding sequence motif. Yet another possibility is that sites of maximum disorder favor endoproteolytic cleavage, and cleavage increases the probability that an MHC protein will bind the adjacent sequence. The latter possibility can explain the specific value of the offset. Eight residues correspond to the length of peptide that is protected from proteolysis by MHC II protein I-A k , measured from the N terminus to the center of a 10-residue determinant core (49). If the typical endoproteolytic cleavage event generates the N-terminal end of the peptide that is to be bound to the MHC protein, then the middle of the average epitope would be 8 residues C-terminal (Fig. 5).
The overall probability that a sequence will be become an epitope is determined by a combination of factors including the T-cell repertoire, MHC-binding motif, details of proteolytic processing, and antigen three-dimensional structure. The influence of structure cannot be discounted on the basis of an assumption that lysosomal acidity destroys all antigen structure. The crystals used to obtain the structure of gp120 were grown at pH 5.6 (51), and many proteins retain native-like structure at considerably lower pH levels (52). For the outer domain of gp120, the coefficient of determination (r 2 ) indicates that local disorder predicts adjacent helper T-cell epitopes to the extent of at least 41%. This must be an underestimate because the correlation employs a single value of offset for all epitopes, when it is likely that MHC selectivity and additional proteolytic processing influence the position of the epitope relative to the cleavage site. Thus, three-dimensional structure exerts a substantial influence on epitope immunodominance in the outer domain of HIV gp120.
Although the correlation coefficient detects the matching patterns of epitopes and adjacent disordered sites in the outer domain, it was formally possible that the correlation simply detects a recurring pattern of sequence and structure. It is evident that the four peaks of B-factor in the outer domain appear at regular intervals of ϳ50 amino acid residues. In testing larger values of offset, we should find additional maxima in the correlation, in effect, associating epitopes with disordered segments more distant in the sequence. For example, an offset of Ϫ64 aligns V4 with the promiscuous epitopes in peptides 46 -47. (To compare equally sized data sets, the unmatched B-factors at the C terminus were wrapped to the N terminus and aligned with epitope frequency for N-terminal sequences.) However, all such alternate alignments yield lower correlation coefficients, r 2 max Ͻ 0.15, as compared with r 2 max ϭ 0.41 for an offset of Ϫ8 (data not shown). Thus, the pattern of epitopes optimally matches the pattern of B-factors at an offset of Ϫ8.
A one-to-one relationship of immunogenic sequences and disordered sites could be a general feature of epitope immunodominance. To examine this possibility we identified seven proteins for which comprehensive mapping data and an x-ray crystal structure were both available. The sample included data from studies that utilized overlapping peptides spanning the complete protein sequence and that were carried out with inbred animals or a large number of sensitized humans. Each stimulatory peptide or cluster of overlapping stimulatory peptides was scored as an immunogenic sequence. The disordered FIG. 4. Plots of correlation coefficient versus offset for correlations of epitope frequency (mouse strains combined) with structural parameters (A and B) and for the correlation of epitope frequencies in the two mouse strains with each other (C). Correlations of epitope frequency with structural parameters are much better when the analysis is restricted to the outer domain of gp120. In the outer domain, correlations of epitope frequency with B-factor and sequence conservation reach maxima at offsets of Ϫ8 and Ϫ12, respectively, suggesting that epitopes tend to occur on the Cterminal flank of disordered, poorly conserved segments. Epitope frequencies from the two mouse strains correlate well in the outer domain and poorly in the inner domain. A poor correlation is expected if selectivity by MHC alleles controls the allocation of epitopes and if the respective MHC-binding motifs in gp120 are randomly distributed with respect to each other. The strong correlation at zero offset in the outer domain indicates that epitopes in the two mouse strains tend to be in the same sequence or narrowly distributed to both sides of each other. Pearson correlation coefficients were determined using the facility implemented in Excel (Microsoft). For offsetting prior to correlation, the window of one data set was shifted relative to the other data set by the specified number of residues, e.g. epitope frequencies in residues 258 -488 were correlated with B-factors in residues 250 -480 to yield the correlation coefficient for offset ϭ Ϫ8. Offsets were sampled in fourresidue increments over the range Ϫ20 to 20 residues. segments were identified by three or more consecutive backbone amide nitrogen atoms with above average B-factors in the crystal structure. Similar criteria were used to identify disordered segments associated with promiscuous epitopes in diphtheria and tetanus toxins (11) as well as in bacteriophage T4 Hsp10. 2 A striking correlation was found between the number of immunogenic sequences and the number of disordered segments (Fig. 6A). A similar result might have been expected for the correlation of either variable with protein size because the number of immunogenic sequences and disordered segments are each expected to increase with protein size. However, the number of epitopes correlates poorly with protein size (Fig. 6B). Clearly, proteins of equal size can have disparate numbers of epitopes (Fig. 6B, compare points for T4 Hsp10 and lysozyme). The prevalence of a one-to-one relationship between immunogenic regions and disordered sites further justifies the residueby-residue correlation of immunogenic regions with adjacent disordered segments in the outer domain of HIV gp120.
Modulation of Helper T-cell Epitope Immunodominance by Domain Structure in gp120 -We considered alternative possibilities for the poor correlation of epitopes with B-factors in the inner domain of HIV gp120. The inner and outer domains were initially distinguished by the fact that the promiscuously dominant epitopes were only in the outer domain and the uniquely dominant epitopes were only in the inner domain. This differ-ence in behavior could be due to influence by antigen processing on presentation of epitopes from the outer domain but not the inner domain. A striking illustration of the distinct behaviors of immunodominance in the two domains is given by the correlation of epitope frequencies in CBA/J and BALB/c mice with each other (Fig. 4 and Table III). In the inner domain the correlation is very low, as expected if MHC alleles determine epitope frequencies, whereas in the outer domain the correlation is high (r max ϭ 0.70, offset ϭ 0).
Nevertheless, we considered the possibility that the poor correlation of epitope frequencies with B-factors was due to the lack of structural data for large segments that were not included in the crystallized gp120 protein. To fill-in the gaps, sequence variability was used as a surrogate for structural disorder. Low sequence conservation coincides with high Bfactors (either measured or assigned) in each of the variable loops of gp120 (Fig. 2). However, assignment of high B-factors to the N-terminal 82 residues, which were truncated for the FIG. 5. Schematic illustration of antigen processing and peptide loading. Waves indicate local structural disorder in the native or acid-denatured antigen. The star indicates the center of the disordered segment and the position of the N-terminal residue in the MHC-bound peptide. Conformational distortion of the disordered segment accompanies binding of the protease. The lower energetic cost of distorting disordered segments probably explains their preferential cleavage (59). The MHC II protein binds to the C-terminal product, protecting the peptide against further proteolysis on the N-terminal side of the epitope. The MHC protein protects approximately four residues extending N-terminally from a 10-residue determinant core of hen egg lysozyme (49). This mechanism can account for the 8-residue offset required to obtain maximum correlation between local disorder and epitope frequency in hen egg lysozyme (41) and in the outer domain of HIV gp120 (this work). aa, amino acids.
FIG. 6. Correlation of the number of immunogenic sequences with the number of disordered segments in seven proteins. Immunogenic sequences were identified by peptides or clusters of overlapping peptides that stimulated T-cells from sensitized humans or animals. Disordered segments were identified by three consecutive backbone amide nitrogens with above average crystallographic B-factors. T-cell epitope data were from the following sources: Hsp65, Ref. 60  crystal structure, is not consistent with the high sequence conservation in that segment. Sequence conservation may provide a more accurate picture of disorder in the inner domain than the assemblage of experimental and assigned B-factors. Thus, epitope frequencies were tested for an inverse correlation with sequence conservation in gp120. The correlation of epitope frequencies with sequence conservation was similar to that for B-factors in three aspects. First, epitope frequencies only weakly correlated with sequence conservation when the analysis encompassed the entire gp120 sequence but correlated well (inversely) when the analysis was restricted to the outer domain ( Fig. 4 and Table III). Second, the value of the offset at r max was similar whether using Bfactors (Ϫ8) or sequence conservation (Ϫ12) in the outer domain. The optimum correlation with conservation was specific for this offset because alternative alignments at large offsets yielded relatively poor correlations (data not shown). Third, as observed with B-factors, the correlation with sequence conservation was poor when the analysis was restricted to the inner domain (͉r͉ Ͻ 0.3 for Ϫ20 Յ offset Յ 20). The lack of correlation in the inner domain suggests that variation in local disorder is insufficient to influence antigen processing, and thus immunodominance is controlled almost entirely by other factors such as the sequence preference of the MHC protein.
A plausible explanation for the poor correlation of epitopes and disordered segments in the inner domain is that the entire domain is structurally unstable. Independent observations support this conclusion. Substantial deletions in the inner domain can be introduced without significant consequences to the structure of the remainder of gp120 (53). The inner domain of gp120 provides all of the contacts to gp41 (53), which undergoes a substantial conformational change during membrane fusion. Conformational shifts in gp120 associated with CD4 and chemokine receptor binding were proposed to destabilize the gp120/gp41 interface and thereby trigger the conformational change in gp41 (35). Thus, the inner domain could be metastable to facilitate the conformational transition of gp41. During antigen processing, the inner domain may expose all of its epitopes equally well.
Implications of the Relationship of Helper T-cell Epitope Immunodominance to Antigen Structure-The influence of antigen structure on immunodominance has several implications for design of vaccines and immunotherapeutics. First, the association of immunogenicity with disordered segments increases the probability that the epitope sequences will be hypervariable. Vaccines that trigger the natural immunodominance pattern may not stimulate broadly protective immunity because they fail to stimulate T-cells against conserved epitopes. Second, the association of immunogenicity with disordered segments potentially aggravates the problem of escape variants. Pathogen variants that fail to provoke the recall of T-cell responses can be selected by immune pressure in the course of a single infection (54,55). In the battle between host and virus, the advantage is to the virus if the immune system is directed by antigen structure to survey only hypervariable segments. Lastly, immunodominant helper T-cell epitopes may influence the specificity of the dominant antibodies through T-B collaboration (56). Others have discussed the possibility that structural features in gp120 sequester broadly neutralizing antibody epitopes (57). However, structural features may also sequester T-cell epitopes that are necessary for induction of broadly neutralizing antibodies.