In reviewing approaches to epitope characterization described in the literature, both experimentally and through the use of computer-based analysis, three broad shortcomings become apparent.

First, literature reports of experimental approaches to epitope characterization have often been narrow in scope, based on the response of individual patients, cells from a few individual donors or single strains of mice, or focused on isolated peptides. This has generated valid data, but which is specific to the narrow set of circumstances and not reflective of the broader host or organism population. Discovering binding affinity for an MHC molecule of a single HLA haplotype will not necessarily be predictive for a population of diverse heterozygotic individuals. Many literature reports claim T-cell epitope characterization but fail to report the MHC restriction (mouse) or HLA of cells used. By limiting consideration to isolated peptides, an important feature of cell biology is overlooked. Binding to MHC-I and MHC-II molecules is a competitive and dynamic process 1 2 . MHC molecules bind to peptides selected from among all those competitors which result from the proteolysis of the whole organism. Predictive determinations of preferential epitope binding can thus only be made when considered in the context of the whole proteome, or, at very least, the whole protein, but not for isolated peptides.

Second, from an epidemiologic perspective the outcome of infection is dependent on the interface between a population of heterozygous hosts and a diverse array of microbial strains. Many possible interactions of individual and strain are possible. Depending on the context, the challenge in vaccine design may be to choose the best combination of epitopes conserved across multiple strains of an organism to protect an entire immunogenetically diverse community (for infectious diseases), or to select the immunostimulant optimal for a specific patient (in cancer immunotherapeutics).

Third, while there is broad recognition that strong T-cell responses are essential to good memory, and in many cases to effective immunity, efforts to characterize B-cell and T-cell responses have not always been well integrated.

B-cell and T-cell cooperative interaction in antigen presentation has been the subject of many landmark papers 3 4 5 6 . More recently, Sette et al demonstrated that, at least for vaccinia, T-cell stimulation is specific to a B-cell epitope located within the same protein 7 , pointing to a close determinant association between B-cell and T-cell epitopes. Cross reactivity, or polyspecificity, is a necessary feature of the T-cell recognition of epitopes comprised of MHC-peptide complexes 8 9 .

There has been increasing recognition that, both for anti-infective immunity, and for cancer immunity, distinctions between the role of MHC-I and MHC-II in responding to intra or extra-cellular organisms are not clear cut 10 11 12 13 . MHC-II molecules bind longer peptides (15-20 amino acids) whereas MHC-I molecules bind shorter peptides of ~9 amino acids or less 1 . Binding of MHC molecules to peptides is characterized by a large degree of degeneracy and it is now recognized that a particular MHC molecule may bind peptides that vary widely in composition and origin 9 .

B-cell epitopes may be continuous or discontinuous peptides, in some cases requiring multiple linear peptides to be configured together to make up a complete epitope 14 . Location of B-cell epitope motifs in loops external to the cell membrane may allow for grouping into a multi-component epitope. Multiple peptides may need to act together to provide an immunostimulant adequate to initiate a B-cell response. Batista has described the need for B-cells to have sufficient stimulation to form immune synapses, initiating and enabling the uptake of surface proteins 15 . In other cases B-cell responses occur independent of T cell stimulation 16 .

Most successful antimicrobial vaccines target surface exposed B-cell epitopes and vaccines have been evaluated by their ability to stimulate an antibody response. Peptide epitopes are a major component of the overall epitope complex, or immunome, and are genetically specified. In many cases antibodies to bacterial proteins are indeed protective, and complement fixing antibodies have been used as an index of vaccinal efficacy 17 .

Immunization protocols for laboratory production of antibodies have long recognized the utility to linkage to a known T-cell epitope 18 19 . T-cell responses to epitopes arrayed in an organism of interest are harder to evaluate 20 . Those working in reverse vaccinology 21 have been frustrated by the difficulty of reliably characterizing T-cell epitopes 17 . Proteins with multiple transmembrane domains have proven challenging to express as sub-unit vaccines 21 . In understanding the interaction of B-cell and T-cell responses, it is therefore useful to readily understand the topology of epitopes relative to the cell membrane. In the case of immunotherapeutic cancer vaccines, the ability to stimulate a multifaceted T-cell response may be even more necessary 22 23 .

Various bioinformatic programs for B-cell epitope and T-cell epitope analysis are available on the Internet (Additional File 1) and have contributed significantly to our understanding. However, a number of limitations are evident. Limits on the sequence size which can be submitted to website servers generally only allows single protein analysis and thus preclude contextual understanding of competitive binding affinity for a whole proteome.

We sought appropriate benchmark datasets to test the system developed. None were available which provided comparable levels of information on all the features we sought to integrate. While a useful repository, the IEDB tabulation of individual epitopes is less useful for proteomic-scale work. We selected three datasets for comparison. The "AntiJen" database provided a benchmark for evaluating a diverse repository of epitopes within the context of entire protein molecules. Two well studied infectious organisms, Staphylococcus aureus and vaccinia virus, enabled retrospective comparisons with published data.

A system for integrated analysis of proteome scale epitope information was designed which comprises a number of sub-processes. All computations were done and graphics generated using JMP^® version 8 http://www.jmp.com. Figure 1 provides an overview of the system; the sub-processes are described briefly below and in detail in Additional File 4.

Process A consists of developing a set of Neural Network (NN) binding predictions for 14 MHC-II and 35 MHC-I molecules. Once developed these equations are stored for further use. Briefly, principal component amino acid analysis was carried out on the physical properties of amino acids measured in a total of thirty-one different published studies. In the NN each amino acid is assigned 3 numerical values based its principal components rather than the standard alphabetical representation commonly used in bioinformatics. This type of descriptor is commonly used in QSAR analysis where it is known as the "z"-scale 43 44 . The principal component descriptors are uncorrelated, mutually orthogonal metrics, and embody about 90% of the variance in all physical properties of the 20 amino acids commonly found in proteins. The z-scales are not in themselves physical properties, but rather uncorrelated dimensionless proxies for amino acid physical properties that can be used predictively: z₁ is a hydrophobicity or polarity correlate, z₂ a size correlate and z₃ an electronic correlate. A characteristic of principal component analysis is that it also produces a set of descriptors that are appropriately weighted for regression analysis. This process is described in detail in a companion paper where it is benchmarked against several other prediction schemes 29 .

Process B, also described in the companion paper, consists of replacing the alphabetic notation of amino acids by z-scales so that each 9-mer in the proteome is represented as a vector of 27 numbers and each 15-mer as a vector of 45 numbers. These numerical values are then used to compute predicted binding affinities for peptides in the proteome using the NN prediction equations from Process A.

Process C involves the use of one of several publicly available programs for protein topology predictions. We have variously used PHOBIUS 45 , PHILIUS 46 , MEMSAT 47 , and TMH 48 . The output is a probability prediction for each amino acid in the protein as being intracellular, extracellular, within a membrane or a signal peptide. A determination of B-cell epitope predictions is also made. Unlike the MHC predictions which provide a predicted affinity, in the case of B-cell epitopes we are making a binary "yes-no" probability prediction that a specific amino acid lies in a B-cell epitope. The B-cell epitope probability may be achieved by submission to one of several publicly available programs for B-cell epitope predictions 28 , however we generated a B-cell epitope prediction based on principal components, enabling us to achieve this step as an integral part of the process.

Process D integrates the output from the first three processes and involves the use of self-organizing mapping algorithms to identify Coincident Epitope Groups (CEG) for protein segments likely to be accessible to the immune system. CEGs are peptides in which high affinity MHC binding peptides and B cell epitopes are found to overlap or whose borders lie within a user-specifiable distance of each other. As described herein the distance was set at 3 amino acids.

Process E is a database task to assemble nearly identical protein sets (NIPS) from different strains of organisms to arrive at a minimalist set of conserved or near-conserved peptide epitopes for further consideration.

To facilitate further statistical procedures, the MHC binding affinities (as natural logarithms) were standardized. Standardization is a common statistical process where the data points are transformed to a mean of zero and unit variance (and standard deviation as the standard deviation is the square root of the variance). Thus all binding affinities of all different supertypes, and paired supertype combinations, were put on the same basis for further computations. This process is reversible so that a more experimentally meaningful ic₅₀ can be obtained at any point if desired. Secondly, the Bayesian probabilities for each individual amino acid being in a B-cell epitope were subjected to global standardization like that for the MHC binding affinities. Thus, all the peptides and other metrics subject to statistical screening are standardized, so that thresholding or other selections are made on single or joint normal probability distributions.

Following the standardization processes, the tables of binding affinities contained columns of the original predicted binding affinity data for the different MHC supertypes (as natural logarithms) and the original B-cell epitope probabilities, as well as corresponding columns of standardized (zero mean, unit standard deviation) data of the immunologically relevant endpoints.

Visualization of all epitope components in relation to topology facilitates understanding of function. A graphical scheme (Step 13 in Figure 1) was developed that made it possible to readily visualize the topology of proteins at the surface of the organism as well as three standardized probabilities for high affinity binding petides for MHC-I and MHC-II, and B-epitopes. Predictions for MHC-I and MHC-II binding were done routinely for all organisms, although it is recognized that MHC-I is generally considered most relevant for intracellular infectious organisms and MHC-II for extracellular organisms. Simultaneous visualization of both provides a method of conceptualization of potential cross-presentation of epitopes.

We adopted a convention for graphical display in which the amino acids positions are aligned along the X axis from N to C. The Y axis is in standardized units (zero mean, unit std dev) to show MHC binding affinity. Topological information is displayed in the background shading. The permuted minima ln(ic50) representing the mean population phenotype are plotted for MHC-I and/or MHC-II at each peptide position the number representing a mean of 105 (MHC-II) or 630 (MHC-I) allelic combinations at that position. This is windowed average minima ± 4 amino acids from the plotted point. A smooth line fit through the points is produced using the polynomial filter of Savistsky and Golay 49 . Another line is overlaid to show the standardized probability of B-cell epitope binding. Across the base of the graphic we use ribbons of various colors and intensities to designate regions of high binding affinity and coincidence of B-cell and MHC binding. Different thresholding stringencies can be applied to the ribbons (see Additional File 5); we have mostly found that the 25^th percentile of the permuted minimum distribution of MHC binding along with the 25^th percentile of B-cell epitope probability to be useful thresholds and these are used throughout the graphics below. The 25^th percentile should be clearly understood to be a threshold from within the permuted minima distributions that have their own mean and standard deviations, and not the 25^th percentile of all peptides. Colored vertical lines are used to show the behavior of any particular HLA allele as compared to the permuted population phenotype. The line extends from the permuted population value to the standardized value of the indicated HLA at the N-terminus of the peptide 9-mer or 15-mer. By creating a further overlay experimentally defined epitopes can be compared with predictions. The display can easily be rescaled to visualize the individual amino acids in the peptides.

Table 1 is summary of the binding affinities for MHC-II supertypes for the surfome (surface proteome) and secretome (secreted proteins) of Staphylococcus aureus COL (Genbank genome accession number = NC_002951). The surfome consists of all proteins coded for in the genome that have a molecular signature(s) predicting their insertion in cell membranes. Some proteins in the surfome also have signal peptides that control topology but do not lead to secretion.

Prediction of B-cell epitopes, MHC-II binding, and topology for 15 strains of Staph. aureus (listed in Additional File 3, Table S3b) have been done. Predicted B-cell epitopes were found to be located inside and outside the bacterial cell membrane, but virtually none in the transmembrane domains, perhaps due to alpha helical structure of the transmembrane peptides. In the very few instances where extension into membranes did occur (<2%), the predicted B-cell epitope only penetrated a few amino acids. This may represent an error in the prediction of the edge of the transmembrane domain.

A summary of the topology of the proteins and the predicted MHC-I and MHC-II binding affinity of peptides in Staph. aureus COL is shown in Table 2. It is recognized that the immunological response to Staph. aureus should be mediated primarily through MHC-II; MHC-I is included for completeness. The table shows the relatively higher binding affinity of peptides found in membrane spanning segments of the proteome. The difference in the ln(ic50) of the MHC-II affinities of the intracellular and extracellular segments is small but is statistically significant.

The array of genetic variants (supertypes) of HLA molecules in the human population vastly exceeds that for which there are peptide training sets. Additionally, and yet further increasing the combinatorial possibilities, is the fact that each individual has both parental genotypes of MHC on their cell membranes. Despite the combinatorial complexity, examination of the statistics of the predicted binding affinities to a number of different proteins in the proteome of Staph. aureus gave rise to several observations which suggested that it would be possible to derive a system for determining the probability of binding not only for single supertypes, but for a population of combinatorial supertypes for which a trained NN was available. The processes outlined above make it possible to put entire proteomes (or multiple proteomes) consisting of millions of binding affinities into a single data table, in a familiar spreadsheet interface on a standard workstation computer.

Table 3 shows the predicted binding affinities for each of the DRB supertypes in combination with each of the other DRB molecules (105 permutations) simulating heterozygous individuals (detail in Additional File 6). Inside an antigen presenting cell where peptides from a digested organism (e.g. Staph. aureus COL) are coming into contact with MHC-II molecules, those molecules with higher affinity (smaller of the two ln affinity numbers) would be expected to dominate in the binding process. One of the striking features that emerges from this table (bottom rows Table 3) is the overall advantage of heterozygosity. Individuals randomly inheriting combinational pairs of the 14 supertypes stand to have a higher binding affinity than if they had only one type. A second observation is the segregation of different alleles within the sorted list. The heterozygosity advantage and the 10 percentile threshold, being in a range considered a useful biological range of affinity, suggested the possibility of averaging over all genotypes as a means of predicting binding in a population of individuals carrying MHC-II molecules of unknown genotype on their cells (as would be the case in a randomly selected vaccinee population). These results suggest that combinatorial pairs of supertypes need to be considered in statistical selection and screening processes, for example in clinical trials.

The different alleles and allelic combinations shown in Table 3 have significantly different means and variance and this complicates statistical analysis and thresholding. All of the proteins in the Staph. aureus surfome, comprising about 210,000 15-mers, were used in a global standardization process (summarized in Additional File 5). By using all the 15-mers in the proteome for standardization, the statistical processes are brought into line with the biological process where an engulfed foreign organism would be digested and the peptides presented would be the entire repertoire of the organism. Furthermore, the construction of normally distributed populations provides a means of rigorous and meaningful statistical screening and selection processes from normal Gaussian distributions (Figure 2).

A second distribution anomaly is shown in Figure 2. Not only does the binding affinity vary across different MHC alleles as described above, it also varies between proteins in a proteome, giving rise to distributions like that seen in Figure 2. This clearly demonstrates why any organism-level conclusions about binding affinity cannot be made based on measurements made on peptides in isolation and should preferably not be made on individual proteins. The global standardization process based on standardized ln(ic50) ranks all peptides and produces a ranking of affinities of all peptides in a proteome against all other peptides in that proteome. This is the situation that would arise as an infectious organism is digested in an antigen presenting cell.

By examining the plots of many different proteins with different types of data portrayal we observed that, despite individual 15-mer peptides showing widely different predicted binding affinities for the different MHC supertypes, there was a tendency for high binding for all supertypes to locate in certain regions of molecules and low binding in other regions. This can be seen by undulations in the averaged mean affinities across a protein sequence. Not only was this the case among MHC-II supertypes, but was also seen with the overall means of all MHC-I and MHC-II supertypes (Figure 3; Table 4). After examining many different proteins individually it emerged that each protein has a characteristic undulation pattern regardless of the supertype or MHC. Computed on an affinity basis the variations are very large with the mean affinity varying over a thousand-fold for peptides from different regions within a protein molecule. This long-range variation is superimposed on a large peptide to peptide variation within the protein.

It also became apparent that, despite the large differences in affinities between peptides, for a particular peptide some of the ic50 values were highly correlated across MHC alleles. The Pearson correlation coefficients for the top ten and bottom ten pairwise comparisons are shown in Table 4. For MHC-II all of the pairwise correlations are statistically significant and positive, though of varying magnitude. For MHC-I there is a subset that is positively correlated, another that is negatively correlated and a third group of non-correlated alleles.

The positive correlations among MHC alleles and the other statistical characteristics led us to experiment with methods of computing binding metrics that encompassed a population of heterozygotic combinations; effectively a population phenotype.

From an immunological perspective the low affinity peptides are irrelevant; it is more useful to calculate a running average of high affinities in a way that captures the undulations in binding affinity across a protein sequence, while also capturing the population phenotype. Based on these concepts we developed a system to compute an average of standardized affinities for the permuted pairs for all supertypes within an adjustable (filtering) window. The window is defined as a stretch of contiguous amino acids positions over which averaging was carried out. Various windows (filtering stringencies) were tested, but the most useful smoothing was achieved with a window of ± half the size of the binding pocket (± 4 amino acids) as shown in Figure 4.

For MHC-II this is reasonably simple to envisage, as the ends of the pocket are open and peptides longer than 15 amino acids could undergo rapid association:dissociation "jiggling" until the highest binding configuration is found. In practice, the range of affinity constants in a pool of peptides may be as much as 1000 fold so that higher affinity peptides will very rapidly occupy the MHC binding site and remain bound in place. For MHC-I, with closed ends on the binding pocket, the possibilities are more limited.

Another factor, which has not been included in the predictions at this point, is the effect of the differential proteolysis that will contribute to the variable lengths of peptide with a possibility to interact with a binding pocket. Several tests of the potential impact of proteasomal cleavage were carried out with the webserver NetChop 3.1 at CBS on sample protein sets (not shown). From those experiments it appears that peptides are very likely to be cut into pieces shorter than 9 amino acids, so that the MHC-I presentation of a peptide is the result of interception and capture by the MHC binding reaction during proteolytic cleavage. The patterns of proteolytic cleavage of proteins by lysozomal enzymes suggest that they would be equally aggressive and that peptide processing for MHC-II presentation would be expected to be comparable.

The permuted minima within the window described above (and shown in Figure 4) are averaged to arrive at a single number for all MHC-I allelic combinations and another for all MHC-II combinations for each particular amino acid position. Through experimentation we found that this process produced metrics whose undulations tracked the visually obvious patterns of MHC binding in proteins (as seen in Figure 3). As the numbers were drawn from a standardized dataset the resulting sample were also normally distributed albeit at a distance (negative) from the population mean as a whole. Thus, statistical thresholds could be applied to these metrics that were based on normal populations. Each allelic combination was given an equal weight. This is a generalizable concept; it is also possible to compute the predicted population phenotype for various subpopulations using appropriate weighting for genetic frequencies. The windowing operation is not dependent on standardized populations and the actual ln(ic50) can be used as well to compute a running average of binding affinity for any single allele.

The output of these computational processes was tabulated in a master database for the organism (Figure 1, Step 13). Selected coincident epitope groups comprising regions of proteins where peptides met three criteria were determined. In these both binding threshold for MHC and the B-cell epitope probability threshold were in the 10 percentile range and the run of amino acids in the predicted BEPI peptide was ≥4 amino acids. Selection of the 10th percentile in two characteristics in normally distributed variables on a probability basis should be a product of two probabilities or about a 1% coincidence where MHC binding regions overlapped either partially or completely with predicted B-epitope regions.

Figure 5 shows an annotated example of the graphical output from the system we have described above.

We describe an integrated epitope analysis system which is based on multi-dimensional and orthogonal physicochemical properties of sequences of amino acids using a multilayer perceptron neural net to conduct QSAR regression predictions for peptide affinities to 35 MHC-I and 14 MHC-II alleles. The system allows rapid processing of single proteins, entire proteomes or subsets, as well as multiple strains of the same organism. It allows consideration of diversity of both microorganisms and of host immunogenetics.

The program can be used to predict B-cell epitope peptides and MHC-I and MHC-II binding peptides, across strains or unique to one strain of organism, plus spatial and topological correlation of membrane proximity. It predicts peptide affinities for HLA supertypes, heterozygous pairs and population-permuted heterozygous pairs.

The system is built on JMP^® (and JMP^® Genomics) data visualization and statistical platform framework and configured to run on a desktop computer and generate graphical and tabular outputs. The predictions can be expanded to other MHC molecules as more MHC training sets become available.

We have tested the system retrospectively against proteins from two organisms for which epitopes have been documented independently by many labs and have used the 'AntiJen' benchmark data set. The approach we describe has performed well for a wide variety of prokaryotic and eukaryotic proteins, including mammalian cellular surface and secreted proteins. The graphical visualization output allows perception of patterns among predicted epitopes heretofore not recognized. Just as GIS have permitted a more integrated view of landscapes and have provided insights which have aided land use and public policy decisions, the layering and integration of all available immunological and topological information offers new insights into organization in the immune response.

Having applied the integrated analysis system to the test data sets, as well as to other proteomes, a number of patterns emerged. Furthermore, the ability to visualize coincident features within proteins brings a new level of insight into possible functional interactions.

B-cell recognition gives rise to an antibody response. MHC-peptide binding is an intermediate step to presentation of MHC-peptide complexes for T cell recognition. We can speculate on whether a CEG represents a functional as well as a physical association. Given the consistency of the pattern, a functional association appears likely in which B-cell binding leads to uptake of the adjacent, overlapping or identical peptide to yield high affinity MHC binding peptides (via MHC-I or MHC-II), and, once bound, the peptide-MHC complex can lead to a productive T-cell response.

When viewed in the light of the observations of Batista of B-cell attachment and "pinching off" of surface segments during the formation of an immune synapse 15 , one can envision how this physical proximity might facilitate the internalization of a peptide with a competitive advantage as a high affinity MHC binder. The dual presentation of B and T-cell epitopes would be consistent with the B-cell-T-cell interaction proposed by Lanzavecchia 6 . Similarly, preservation of B-cell epitopes by dendritic cells 57 58 59 would lead to the preservation and delivery to B-cells, not only of B-cell epitopes but also of overlapping MHC binding peptides.

The concept of specificity runs throughout the immunology literature. Generally specificity implies an absolute lock and key relationship. Rather, we see a pattern consistent with the dynamic competition for higher affinity MHC binding positions within the constraints of each set of unique host-pathogen interactions. Confronted with peptides from a different protein or different organism, the competition among peptides for binding to the same MHC molecules would simply be expanded. However, as affinity may range over binding constants of 1000 fold or more, there is a marked difference between high binders and weak binders. This is not inconsistent with a degeneracy in the binding specificity of peptides to MHC molecules. What matters is simply the ability of an MHC molecule to bind with high affinity a peptide representing the microorganism (or other immunogen) presented at that moment. The same MHC could just as well bind a peptide from another unrelated immunogen, not present in the same cell at that time. Given this multi step process, the effective "specificity" of the adaptive immune response is the product of multiple sequential binding affinities - {B-cell epitope binding }x {MHC binding} x{T-cell receptor binding to MHC-peptide complex}. No one of these steps needs to confer complete "specificity" but the combination increases the uniqueness or specificity of the antigen-immune response, essentially as a combination lock.

We do not speculate on the downstream implications of T-cell epitope binding, but note that T-cell stimulation may have both up and down-regulation effects and both positive and negative cytokine mediated feedback loops. Whether MHC binding regions located in CEGs have different roles from those MHC binding regions which are unrelated to B-cell epitopes is unknown. High affinity MHC binding regions located in membranes may be hidden until internalized and released by proteolysis, as proposed by Benacerraf 3 . This could occur through internalization of whole organisms or when such peptide fragments are "towed along" by internalization of adjacent tethered peptides.

It is of interest that we predict that a large percentage of predicted MHC-I high affinity binding peptides are coincident with those which bind MHC-II well. This implies that the precise intracellular pathway, and hence which proteolytic machinery the peptide encounters, may allow such peptides to stimulate T_CD8 or T_CD4, simultaneously or sequentially. Others have noted the need to understand the degree of overlap in the peptidome leading to stimulation of each pathway 60 .

Given the apparent frequent overlap of B-cell epitopes with MHC binding regions, we suspect the literature contains observations on many peptide epitopes where, depending on what the experiment was designed to observe, function as either B-cell or T-cell epitopes has been described. Vaughan 61 , in reviewing the literature on epitope characterization for Plasmodium, noted that 14% epitopes characterized by some workers as B-cell epitopes are reported by others to be T-cell epitopes, a percentage not dissimilar from that seen in the surfome of Staph. aureus and the AntiJen dataset.

At higher resolution, the fine structure of the diverse binding patterns of different HLAs shown (Figure 7) in vaccinia (and observed in other proteins; unpublished data) may shed light on the concept of immunodominance, in which immune responses by an individual are directed to a few peptides 60 . At one level it underscores the utility of HLA transgenic mice as indicators of immunodominance (binding affinity) for humans 54 , while also calling into question extrapolation of peptide level MHC binding evaluations conducted in murine inbred strains and raising questions about the need to reflect immunogenetic diversity in clinical trials. Taking a broader perspective, if each HLA shows highest binding to a very narrow peptide sequence in given region of a protein (immunodominant peptide) it may represent a risk of microbial escape mutants 60 . However, because other HLAs bind to different adjacent peptides with high affinity, heterozygosity provides a backstop. Viewed in an immunogenetically diverse population, it provides a possible survival strategy in which any one escape mutant does not threaten an entire host population.

Overall, the predicted patterns which emerge from mapping imply yet greater coordination and organization of B-cell and T-cell responses that have heretofore been recorded. We do not underestimate the gap between prediction and experimental testing; however the ability to envision hypothetical functional interactions is a necessary precursor to design of experimentation.