There is considerable interest in developing immunotherapeutic approaches to elicit tumor-specific CTL responses in cancer patients. Epitope-based approaches aim to deliver the antigenic peptides or epitopes recognized by CTLs rather than the intact tumor antigen. Many tumor-associated proteins are nonmutated self proteins for which the dominant peptide epitopes are usually poorly immunogenic. The subdominant epitopes, however, can elicit robust T cell responses if optimized for their ability to bind to class I MHC molecules. Only the epitopes for a few tumor antigens expressed in human cancers have been defined to this level, mainly for technical reasons. The means to rapidly screen and characterize the binding of epitopes derived from complex tumor-associated antigens is an important enabling technology. Here, we have used the high-throughput technology iTopia™ to identify those peptides derived from the tumor-associated antigen survivin that bind 8 class I alleles. A library of overlapping nonamers spanning the length of the survivin protein was initially screened for peptides capable of binding each allele. Nineteen HLA-A*0201, zero HLA-A*0101, seven HLA-A*0301, twelve HLA-A*1101, twenty-four HLA-A*2402, six HLA-B*0702, six HLA-B*0801, and eight HLA-B*1501 binding peptides were identified based on an arbitrary cutoff. Peptides capable of binding a given allele were further characterized by their affinity for MHC class I molecules and by the rate of dissociation of the complex. This information should help guide functional studies and future epitope-based immunotherapies.
This article was published in Cancer Immunity, a Cancer Research Institute journal that ceased publication in 2013 and is now provided online in association with Cancer Immunology Research.
Since it was first demonstrated that tumors could be immunogenic (1, 2), over 60 tumor-associated antigens have been identified. This has made tumor-specific immunotherapy of cancers possible. A large body of work has shown that cellular, rather than humoral, immune responses play the central effector role in antitumor immunity. Therefore, many immunotherapeutic strategies are aimed at stimulating antitumor T cell responses and, in particular, these strategies are aimed at stimulating CD8+ CTL responses.
The antigen receptors of CTLs recognize peptide fragments of proteins, or epitopes, which are generated during the intracellular degradation of proteins. MHC class I molecules subsequently bind these peptides and display them on the cell surface to circulating T cells. TCRs recognize the complex of the MHC product and peptide antigen, and T cell antigen recognition is thus said to be MHC-restricted. The elucidation of the crystal structures of MHC/peptide complexes (3) and the discovery of anchor-residue sequence motifs that account for the binding specificity of peptides to MHC molecules (4) have provided the mechanistic answer as to how T cells recognize antigens in the form of short peptides. This detailed understanding of the molecular requirements for MHC class I/peptide binding and subsequent recognition by CTLs has permitted the development of several epitope-based immunotherapeutic approaches to cancer, all of which use T cell epitopes rather than the intact tumor-associated antigen to stimulate tumor-specific T cells. Such approaches are advantageous in that they allow for specific manipulation of the CTL response. However, they are currently limited by the number of defined, highly immunogenic T cell epitopes, particularly for restricting alleles other than HLA-A*0201.
Mutated protein antigens that are unique to tumors occur, but the majority of tumor-associated antigens are nonmutated self proteins that are preferentially or aberrantly expressed in tumor cells. Tumor-associated antigens that are shared among cancer patients are especially attractive candidates for use in vaccines. Survivin is one of a limited number of such shared tumor-associated antigens that is also overexpressed in the majority of human cancers. Of the solid malignancies, survivin has been reported to be overexpressed in lung, colon, breast, ovarian, pancreatic, prostate, and gastric carcinoma (4, 5, 6, 7, 8, 9, 10, 11, 12). It is also overexpressed in hematopoietic malignancies (13, 14). A whole genome analysis of human transcriptomes revealed that survivin is the fourth most prominent tumor-associated transcript (15).
The broad overexpression of survivin in human cancers has led to an intense interest in it as a target for therapeutic CTL responses. However, as with the majority of tumor-associated antigens, survivin is expressed not only in tumors, but also in fetal tissues and, to some extent, in adult tissues (14, 16, 17). As such, survivin would be expected to be weakly immunogenic, with most of the antigen-specific T cells being either clonally deleted or tolerized in the periphery. Several groups of researchers, however, have demonstrated the immunogenicity of survivin either in vitro or in vivo. Antisurvivin antibodies have been detected in the sera of lung cancer patients and of colon cancer patients (18, 19). Spontaneous HLA-A2-restricted CTL responses against the survivin nonamer beginning at position 96 and the overlapping decamer beginning at position 95 have been detected in melanoma patients and in leukemia patients (20, 21). Additionally, several groups have been able to induce CTL responses in vitro using DCs pulsed with survivin peptides (22, 23, 24, 25). Thus, while survivin can be immunogenic, the number of defined T cell epitopes that are immunogenic in vivo is still limited.
Tolerance to self antigens primarily involves the dominant epitopes (26, 27, 28). These are often the peptides from a protein that bind class I molecules with the highest affinity (29) and the lowest dissociation rates (30, 31, 32), as they have a tendency to outcompete the subdominant epitopes for presentation at the cell surface. Several groups have shown that high-affinity tumor epitopes do not necessarily induce efficient CTL responses, and repertoire limitations related to T cell tolerance are thought to be at least part of the basis for this (33, 34, 35, 36). Low- or moderate-affinity epitopes, then, could circumvent these problems provided they are able to bind well enough to prime a response. Indeed, it is now clear that agonist analogs of subdominant epitopes, once optimized for binding class I molecules, can be used to effectively recruit a nontolerized CTL repertoire (37, 38, 39). For this reason, many epitope-based strategies for cancer aim to use optimized analogs of low- or poor affinity epitopes.
As a first step toward such a strategy with survivin, we have used the high-throughput technology iTopia™ to map all peptides capable of binding 8 different HLA alleles. The alleles were chosen to provide broad population coverage. Nineteen HLA-A*0201, zero HLA-A*0101, seven HLA-A*0301, twelve HLA-A*1101, twenty-four HLA-A*2402, six HLA-B*0702, six HLA-B*0801, and eight HLA-B*1501 binding peptides were identified and characterized in terms of their affinity for MHC class I molecules and the rate of dissociation of the complex. The identification and characterization of these epitopes should make possible additional functional studies and future epitope-based immunotherapies.
Identification of peptides that bind HLA alleles
A library of 134 survivin nonamers, overlapping by 8 aa and spanning the length of survivin, was commercially obtained (Synpep Corp., Dublin, CA, USA). Peptide binding assays were performed to identify peptides capable of binding each HLA allele. In the assay, peptides were incubated with HLA-coated wells at the set concentration of 11 µM. Binding to each allele is reported as a percentage relative to a positive control peptide for that allele. An arbitrary cutoff of 30% of the control is suggested by the manufacturer as a positive cutoff for binders. When binding for all 134 peptides is plotted against the 8 alleles (Figure 1), regional clusters of binding are observed, such that a series of adjacent peptides in one region binds one or more allele, and this is interspersed with a region of adjacent peptides that fail to bind any of the alleles. In most instances where a peptide binds more than one allele, the alleles involved are either part of the same HLA motif superfamily (for example, A*0301 and *A1101) (40) or they have fairly broad and related binding motifs (for example, A*0201, A*2402, B*1501). Three of the peptides (37, 130, and 133), however, demonstrated promiscuous binding to four or more of the eight alleles, albeit with a range of relative binding capacities. None of the peptides was found to bind HLA-A*0101. Table 1 lists the peptides for which relative binding was >18% of the control for any of the alleles tested. Based on the 30% cutoff, 0 HLA-A*0101, 19 HLA-A*0201, 7 HLA-A*0301, 12 HLA-A*1101, 23 HLA-A*2402, 6 HLA-B*0702, 6 HLA-B*0801, and 8 HLA-B*1501 binding peptides were identified. Three of the HLA-A*0201 binding peptides had previously been reported to bind HLA-A*0201 (20).
Relative affinities and dissociation rates
For each allele, those peptides binding to a level >30% of the positive control were characterized further by their affinity for HLA, as well as by the relative stability of the complex formed. For the affinity measures, the ability of each peptide to bind was tested over a range of peptide concentrations (10-4 to 10-9 M). Curves were fitted using Graphpads nonlinear regression, single-site binding program. As for the peptide binding assay, the results are reported relative to the positive control for each allele. Based on repeat testing of the positive control for all of the alleles, either with replicate plates in a single experiment or with repeat experiments, the assay-to-assay variation was found to be in the range of 10-15% (data not shown). Representative peptide titration curves are shown for HLA-A*0201 in Figure 2a. Curves for the previously characterized nonamers (96, 101, 130, 133) are shown. Curves for peptide 5, which overlaps with a decamer that had previously been reported to bind A*0201 (20, 22), and for peptide 1, which bound A*0201 with relatively high affinity in the iTopia™ binding assay, are also shown. The peptide titration curves shown are representative of the curves for survivin peptides binding to A*0201 in general. Half-maximal binding spans approximately 3 logs (10-5 to 10-7 M) and is relatively low compared to the positive control, a known viral CTL epitope. The estimated dose at which 50% maximal binding occurs (ED50) was calculated for each peptide/class I allele and the results are listed in Tables 2 (A alleles) and 3 (B alleles). The ED50s for survivin peptides binding all of the HLA-A alleles ranged from 10-5 to 10-7 M. Most survivin peptides binding HLA-A*0301 bound poorly relative to the other HLA-A alleles, but were within 1 log of the A*0301 control. Interestingly, the ED50s for survivin peptides binding to all of the HLA-B alleles also spanned 3 logs, but with a range that was 1 log lower than for the HLA-A alleles. Binding of survivin peptides was lowest, and of a more limited range, for HLA-B*0801.
Off-rate assays were used to evaluate the relative stability of each complex. Peptides were added to HLA-coated wells to a final concentration of 11 µM and incubated overnight at 21°C. Plates were shifted to 37°C, and strips containing the appropriate wells were read at 8 time points over an 8-h period. The binding of the peptides was analyzed using the nonlinear regression, first-order decay equation within Prisms® Graphpad software. Decay curves for HLA-A*0201 and the same survivin peptides for which the peptide titration curves were provided in Figure 2a are shown in Figure 2b. As for the titration curves, the decay curves for these A*0201 binding peptides are representative of those for the A*0201/survivin peptide complexes in general. Half-lives of the complexes (T½) were determined using the GraphPad software and are listed in Tables 2 and 3. Seventeen of the nineteen A*0201 binding peptides were evaluated. Of these, 11 had half-lives below 1.5 h, and only 1 peptide, 101, had a T½ over 5 h. Of note is the very rapid off-rate of peptides 20 and 133, which were detected as binders in the iTopia™ peptide binding assay but failed to bind in the T2 lysate binding assay (20). The A*0201 binding and optimized MART peptides (ELAGIGILTV, EAAGIGILTV, and AAGIGILTV) have previously been defined in terms of complex stability (41) and were included in the assay as additional controls. Consistent with earlier reports, these peptides were found to have T½ values of 22.7 h, 2.9 h, and 1.6 h, respectively (data not shown). Thus, relative to the positive control viral peptide and a well-characterized, optimized MART peptide, the A*0201 binding survivin peptides have notably faster rates of dissociation. For A*1101, five of the twelve binding peptides form complexes that are comparable to the control in terms of stability; for A*2402, nine of the twenty-three binding peptides have T½ values comparable to the control; however, for the A*0301 binding peptides, only one of six has a half-life comparable to the control. All survivin peptides binding the B alleles had T½ values of 1 h or less. B*0801 survivin complexes were particularly short-lived, with most of complexes having half-lives in the range of 7 to 10 min. The control peptides for the B alleles formed complexes with T½ values that ranged from 2 to 4 h, suggesting that whereas the B alleles, in general, seem to form complexes that turn over more rapidly (relative to the A allele complexes), the survivin peptides form especially short-lived complexes with these alleles.
iTopia™ and predictive algorithms
In Table 4, survivin peptides are ranked according to their relative (to each other) ability to bind A*0201 in the peptide binding assay. To allow for a broad comparison with predictive algorithms, this ranking includes peptides that bound to as little as 20% of the positive control and is compared to the predicted rankings obtained using both the BIMAS and SYFPEITHI algorithms (42, 43, 44). Both algorithms are based on known epitopes published in the literature. SYFPEITHI is based largely on allele-specific binding motifs and scores each peptide with regards to its likelihood of being processed and bound by a given class I molecule, whereas the BIMAS algorithm ranks peptides based on a predicted rate of dissociation. The predicted ranking, for both algorithms, is usually used to guide epitope selection. In Table 4, the peptides are also ranked on the basis of an iTopia™-based iScore. The iScore is intended to define the overall level of binding and is calculated using a multiparametric analysis that integrates the peptide binding score, the ED50, and the half-life of the MHC/peptide complex for each peptide. With this analysis, the software generates an index and then assigns each peptide an index or iScore. The relative rankings for the peptide binding and the iScore were compared to the relative rankings in both algorithms, and the Spearman rank analysis in Excel was used to evaluate the correlations. Poor to moderate relationships were found. Rank correlations are only shown for selected comparisons. Weak to moderate correlations which were statistically significant at the 5% level were found between the top 19 binders in the iTopia™ binding assay and the predicted rankings for these peptides using BIMAS or SYFPEITHI. A moderately strong positive correlation (P value = 0.01) was found for the iScore ranking of the 17 tested peptides and the SYFPEITHI ranking of these peptides, but none of the other correlations was significant. Similar results were obtained for the other alleles (data not shown).
Although the predictive algorithms did not accurately predict the relative rankings observed in either the relative binding or the iScores, they were reasonably successful in predicting binding per se. Seventy percent of the top ten peptides predicted in BIMAS, and 50% of the top ten predicted in SYFPEITHI, bound well (>60% of the control) in the iTopia™ binding assay. Sixty-three percent of the top nineteen ranked peptides predicted in BIMAS, and 75% of the top nineteen ranked in SYFPEITHI, were scored as positive binding peptides (>30% of the control) by the iTopia™ binding assay. Thus, although the predictive algorithms successfully identified many of the survivin peptides that bound to A*0201, they were unable to predict the relative ability of these peptides to bind.
Survivin is a unique member of the inhibitor of apoptosis protein (IAP) gene family. As do other members of this family, survivin is thought to mediate suppression of apoptosis by direct inhibition of effector caspases. Uniquely, though, it is also thought to play a role in the maintenance of mitotic progression, as it physically associates with different components of the spindle apparatus during cell division (45). Survivins role in cytoprotection, its physical association with the spindle, and its sharp cell cycle-dependent expression at mitosis have led to the suggestion that the main function of survivin is to prevent apoptosis of the cell during division, a function that could clearly be exploited by transformed cells. Indeed, a whole genome analysis of human transcriptomes showed survivin to be the fourth most prominent tumor-associated transcript (15). This broad overexpression of survivin in human cancers is responsible for the intense interest in targeting the survivin pathway for cancer therapy. One strategy has been to interfere with the survivin pathway through the use of molecular antagonists such as ribozymes, antisense RNA, or even dominant negative mutants (45). More recently, survivin has been investigated as an immunotherapeutic target.
Ambrosini et al. (14) originally reported that survivin was expressed during fetal development but not in terminally differentiated adult tissues. Several groups have now shown that survivin is overexpressed at the mRNA and/or protein level in a range of cancers including colon (6), breast (7), pancreatic (10), ovarian (9), gastric (11), and prostate (12). Many of these were immunohistochemistry (IHC) studies that assessed normal adjacent tissues and did not detect survivin expression (6, 7, 11), but a few recent studies have reported some survivin expression in normal tissues. Konno et al. reported that survivin is expressed in normal endometrium (46); 36% of normal prostate samples obtained from prostates that contained carcinoma were found to express survivin (12); and, finally, Hirohashi et al. report low levels of survivin mRNA in many normal tissues (24). Our in-house Taqman™ studies indicate that survivin is undetectable or low in most tissues, but our IHC studies indicate that certain cell types within a tissue may express survivin (data not shown). Thus, as for many tumor-associated antigens, survivin is expressed not only in tumors, but also during fetal development and, at least to some extent, in normal adult tissues. Accordingly, self tolerance is likely to be a barrier to the generation of effective T cell responses. Of note, then, is that spontaneous HLA-A2-restricted CTL responses against the survivin nonamer beginning at position 96 and the overlapping decamer beginning at position 95 have been detected in melanoma, leukemia, and breast cancer patients (20, 21). The in vivo functionality of these CTLs is not known, however, and there is good evidence that the T cell repertoire against the dominant (27, 28) and the higher affinity (26, 38) epitopes from a tumor-associated antigen may be partially tolerized. For this reason, many epitope-based strategies for cancer aim to use optimized analogs of subdominant epitopes (38, 47, 48, 49). Still, there is no definitive approach for determining which epitopes within a given tumor antigen should be selected to circumvent tolerance and hence serve as the best target in antitumor vaccination. Herein lies the value of an enabling technology such as iTopia™, which facilitates the full-scale mapping of tumor antigens because it allows for the evaluation of multiple candidate peptides.
Several peptides from the survivin protein had previously been tested for their ability to bind MHC class I molecules and, for those that bound, to elicit CTLs (20, 22, 24). The class I binding assays in these studies primarily utilized the mutant cell lines RMA-S or T2. Both cell lines are defective in the translocation of class I peptide ligands into the lumen of the ER, and, as a consequence, unstable class I molecules devoid of peptide accumulate in, and on, the cell surface. The empty class I molecules may be stabilized either in an assembly assay in which peptide is added to cell lysates metabolically labeled with 35S and the properly assembled class I molecules are then immunoprecipitated, or in a stabilization assay in which peptide is added exogenously to cells and unstable molecules at the cell surface are then rescued and measured by cell surface staining with antibody. Neither approach is particularly high-throughput, and for this reason, in all epitope identification studies conducted with survivin thus far, as well as with other antigens, predictive algorithms have been used to narrow the screen to those peptides predicted to bind well.
In two such studies with survivin, seven survivin-derived nonamers (beginning at amino acids 20, 88, 96, 101, 127, 130, and 133) were tested for their ability to either stabilize HLA-A*0201 following exogenous addition of peptide to T2 cells (22) or to assemble HLA-A*0201 molecules following addition of peptide to T2 cell lysates (20). Peptides 96, 130, and 133 were tested in the stabilization assay and were not found to stabilize HLA-A*0201 at the cell surface, although the decamer beginning at position 95 did bind in this study (22). In contrast to the results from the stabilization assay, peptides 96 and 130, in addition to peptide 101, were found to bind HLA-A*0201 in the assembly assay, albeit with relatively low affinities. Peptides 20, 88, 127, and 133 were reported not to bind in this study (20). Our results extend these earlier studies by confirming that peptides 96, 101, and 130 can bind HLA-A*0201 and, in contrast to earlier findings, we find that the survivin nonamers 20 and 133 do bind HLA-A*0201. Peptide 88 bound to levels that were detectable but below the arbitrary cutoff of 30% of the positive control. In a separate study, two survivin nonamers, 85 and 92, were identified by motif scanning and evaluated for their ability to stabilize A*2402 following exogenous addition to RMA-S A24/Kb cells. Neither peptide was found to bind, although a nonamer from the survivin 2B variant did bind (24). In support of these earlier results, we also found that peptide 92 did not bind, but in contrast to the previous results, peptide 85 bound A*2402 with an affinity higher than the control. The three assays, therefore, clearly have differing limits of detection. The stabilization assay appears to be the least sensitive of the three assays, as peptides that bound in both the assembly and the iTopia™ assays were not detected in the stabilization assay.
Of the five nonamers tested in the assembly assay, three of them (96, 101, and 130) did bind. The ED50s for peptides 101 and 130, as measured in the ITopia™ assay, were 1 log lower than the C50s measured in the assembly assay, and the ED50 for peptide 96 was 2 logs lower than the C50, results that suggest that the iTopia™ assay is able to achieve a higher "effective" concentration of peptide. An extension of this would be if, when comparable concentrations of peptide are used in the two assays, the lower affinity peptides could then fall below the limits of detection in the assembly assay. But differences in the binding affinities alone cannot explain the failure to detect binding of these peptides in the stabilization and assembly systems. For the five A*0201 binding peptides, the ED50s, as measured in the iTopia™ assay, spanned 2 logs so that there were relatively high- and low-affinity peptides in this group; but it was peptide 130 that had the lowest affinity, and this peptide was detected as a binder in the assembly assay. It is interesting to note, then, that of the five peptides tested in the assembly assay, the two that did not bind are ones for which the half-lives were below 1 h (at 21°C). The failure to detect binding of peptides 20 and 133 in the assembly assay may therefore relate to the off-rate rather than to the binding affinity. It is notable that the A*2402 binding peptide bound with a high affinity, similar to that of the positive control, but it also had a relatively rapid off-rate.
In summary, we have used the iTopia™ epitope discovery technology to identify survivin peptides capable of binding 8 different HLA alleles. Peptides that bound were further characterized in terms of their relative binding affinity and their rate of dissociation from the complex. The study then provides a ranking of the peptides for each allele, with regards to their binding affinity and their off-rates. Such information should help guide future epitope-based therapeutic approaches for survivin, including a strategy of subdominant epitope selection and modification.
Materials and methods
The test peptides used in this assay were synthesized by Synpep Corp. (Dublin, CA, USA). A total of 134 nonamers overlapping by 8 aa comprised the 142-aa sequence of the main isoform of survivin. Peptides were tested across 8 alleles (A*0101, A*0201, A*0301, A*1101, A*2402, B*0702, B*0801, and B*1501). Avidin-coated microtiter plates containing HLA class I MHC monomer loaded with a so-called placeholder peptide were used to evaluate peptide binding, affinity, and off-rate. The monomer-coated plates were part of the iTopia™ Epitope Discovery System Kit (for research only; not for use in diagnostic procedures) from Beckman Coulter, Inc. (BCI, San Diego, CA, USA). In addition to the monomer plates, the kits also contained assay buffers, anti-HLA-ABC-FITC (anti-MHC) mAb, and beta2-microglobulin, as well as the allele-matched positive peptide control.
The 134 test peptides were first evaluated for their ability to bind to each MHC molecule in the peptide binding assay. Monomer-coated plates were first stripped, releasing the placeholder peptide and leaving only the MHC heavy chain bound to the plate. Test peptides were introduced under optimal folding conditions, along with the anti-HLA-ABC-FITC monoclonal tracer antibody. Plates were incubated for a specified period of time at 21°C. The anti-HLA-ABC-FITC mAb binds preferentially to a refolded MHC complex. Test peptides that do not bind to the MHC result in a well with no relative fluorescence, since the tracer antibody does not bind to a nonfolded MHC. The relative fluorescent intensity of each peptide was read on a fluorometer (Molecular Devices, SpectraMax Gemini, Sunnyvale, CA, USA) equipped with software that produces a data file in tab-delimited text file format for export to the iTopia™ System Software. Each test peptides binding was evaluated relative to the specific peptide control provided in the kit, with the result expressed as a percentage of the control. An arbitrary cutoff of >30% of the binding by the positive control was used to identify peptides as binders. Peptides identified as binders were subsequently evaluated for their affinity (ED50) and off-rate (T½) values, again relative to the peptide controls.
For the affinity assay, after the initial stripping, increasing concentrations (range 10-4 to 10-9 M) of each binder for a given allele were added to a series of wells and incubated under the conditions described previously. Plates were read on the fluorometer, and relative fluorescence was determined by assessing the amount of peptide required to support 50% refolding of the plate-bound monomer (ED50 value).
For the off-rate assay, the plates were stripped again and the binders previously identified in the peptide binding assay were incubated on the allele-specific monomer plates at 37°C. The wells were tested at multiple time-points for relative fluorescent intensity as described previously and the data analyzed. The time, expressed in hours, required for 50% of the peptide to dissociate from the MHC monomer is defined as the T½ value.
Peptide binding, affinity, and off-rate values were all calculated using the iTopia™ System Software. Sigmoidal dose response curves were generated using Prism® software by GraphPad. A final iScore was assigned to each test peptide assay to rank potential immunogenicity. The iScore is a multiparametric calculation within the iTopia™ software. The calculation utilizes the data obtained in the peptide binding, affinity, and off-rate assays.
The authors thank Claes Ohlen for his critical reading of the manuscript.
- Received September 6, 2004.
- Accepted January 18, 2005.
- Copyright © 2005 by Lynda G. Tussey