Activating BRAF somatic missense mutations within the kinase domain are present in 60–66% of melanomas. The vast majority of these represent a single substitution of glutamate for valine (V599E). Here, we demonstrate spontaneous HLA-B*2705-restricted cytotoxic T-cell responses against an epitope derived from V599EBRaf. These T-cell responses were mutation specific as the corresponding epitope derived from wild-type BRaf was not recognized. The loss of the V599EBRAF genotype during progression from primary to metastatic melanoma in patients with V599EBRaf specific T-cell responses suggests an active immune selection of nonmutated melanoma clones by the tumor-bearing host.
Targeted cancer immune therapies critically depend on the choice of antigen. Antigens derived from mutated oncogenic proteins would be ideally suited as targets. Alas, only a few mutation specific epitopes derived from such oncogenic proteins have been identified (1) . The recent discovery that activating mutations in the BRAF gene are present in a large percentage of human melanomas together with the observation that some of the patients for whom the BRAF mutational status was established in a longitudinal study lost the activating mutation during progression from primary to metastatic disease prompted us to search for mutation specific HLA-restricted T-cell epitopes derived from mutated BRaf (2, 3, 4, 5) . The vast majority of BRAF mutations represent a single missense mutation of T-A at nucleotide 1796, resulting in a valine to glutamic acid change at residue 599 (V599E) within the activation segment of BRaf. This exchange is believed to mimic the phosphorylation of S598/T601 in wtBRaf, leading to constitutive activation. Consequently, we analyzed peripheral blood and tumor-infiltrating lymphocytes obtained from melanoma patients for the presence of spontaneous CD8 T-cell responses against V599EBRaf.
MATERIALS AND METHODS
For peptide prediction, the web-based MHC-I Antigenic Peptide Processing Prediction (MAPPP) 7 algorithm developed at the Max Planck Institute for infection biology in Berlin was used (6) . MAPPP combines proteasome cleavage predictions with MHC class I binding algorithms. For proteasome cleavage prediction, two different algorithms (PAProC and FRAGPREDICT) can be chosen. The principle of the algorithms has been described elsewhere (7) . In brief, PAProC uses a stochastic hill-climbing algorithm assessing the influence of amino acids at a given position based on experimentally deduced cleavage data (8) . FRAGPREDICT uses two modules (9) . The first one is a statistical prediction of putative cleavage sites, whereas the second step, which is fed by the results of the first module, analyzes the probability for a given peptide to reach the MHC class I processing pathway, taking into account its proteasomal degradation kinetics. In both cases, a probability is computed expressing the likelihood for a given peptide to be generated. The default values were set arbitrarily as 0.5 and were changed to a less sensitive value of 0.4 for both algorithms. Before feeding the peptides into the binding algorithms, additional NH2- and/or COOH-terminal truncation was allowed. However, COOH-terminal trimming and trimming of both ends is fairly unlikely and NH2-terminal trimming seems to be most efficient in antigen-presenting cells (10) , and therefore, the output file contains the information whether trimming was necessary to generate the final peptide.
For prediction of peptide binding to MHC molecules, both the multiplicative BioInformatics and Molecular Analysis Section (BIMAS) algorithm and the additive SYFPEITHI algorithm are used by the MAPPP package using precomputed MHC class I binding matrices comprising 31 different HLA-A and HLA-B alleles (11) . In this case, the score for a given peptide is presented as a percentage in relation to the maximally possible score for a given MHC allele and peptide length. Because of the additive nature, the relative SYFPEITHI scores are usually higher compared with multiplicative BIMAS scores. Also, in this case, the default values were changed toward a lower stringency (>0.5 for SYFPEITHI and >0.015 for BIMAS). With these settings, all four possible combinations for a 33 aa BRaf fragment comprising the mutation and 31 different HLA-A and HLA-B alleles were computed and compared.
Assembly Assay for Peptide Binding to Class I MHC Molecules.
All peptides were purchased from KJ Ross-Petersen ApS (RJ Ross-Petersen ApS, Holte, Denmark) and provided at >80% purity as verified by HPLC and MS analysis. The assembly assay was carried out as described previously (12) . It uses a HLA-B*2705-transfected T2 cell line (T2-B27) (a kind gift from P. Cresswell) that are defective in their ability to load endogenous peptides into MHC class I because of a defect in the transport of peptides within the cell. Such cell lines accumulate empty, unstable class I MHC molecules in the ER. These dissociate rapidly in cell lysates unless they are stabilized by the addition of an appropriate peptide during lysis. Stably folded HLA-molecules can be immune-precipitated using the HLA class I-specific, conformation-dependent mAb W6/32. Subsequently, these HLA-molecules are separated by isoelectric focusing gel electrophoresis. MHC heavy chain bands are then quantified using the Imagequant Phosphorimager program (Molecular Dynamics, Sunnyvale, CA). The intensity of the band is directly related to the amount of peptide-bound class I MHC complex recovered during the assay. The extent of stabilization of HLA-B27 molecules is directly related to the binding affinity of the added peptide.
Flow Cytometric Analysis.
FITC- or phycoerythrin-conjugated monoclonal antibodies against CD4 and CD8 were purchased from Becton Dickinson (Mountain View, CA). Flow cytometric analyses were carried out in a FACSCalibur Flow cytometer (Becton Dickinson, Immunocytometry Systems, San Jose, CA)
Enzyme-Linked Immunospot (ELISPOT) Assay.
To extend the sensitivity of the ELISPOT assay, peripheral blood lymphocytes (PBLs) were stimulated once in vitro before analysis. At day 0, PBLs were thawed and plated in 2 ml/well at a concentration of 2 × 106 cells in 24-well plates (Nunc, Roskilde, Denmark) in X-vivo medium (BioWhittaker, Walkersville, MD), 5% heat-inactivated human serum, and 2 mm l-glutamine in the presence of 10 μm peptide. Two days later, 20 IU/ml recombinant interleukin-2 (Chiron, Ratingen, Germany) were added to the cultures. The cultured cells were tested for reactivity in the ELISPOT on day 12. The ELISPOT assay was used to quantify peptide epitope-specific IFN-γ-releasing effector cells as described previously (13) . Briefly, nitrocellulose bottomed 96-well plates (MultiScreen MAIP N45; Millipore, Hedehusene, Denmark) were coated with anti-IFN-γ antibody (1-D1K; Mabtech, Nacka, Sweden). The wells were washed, blocked by X-vivo medium, and the cells were added in duplicates at different cell concentrations. Peptides were then added to each well, and the plates were incubated overnight. The following day, media were discarded, and the wells were washed before addition of biotinylated secondary antibody (7-B6-1-Biotin; Mabtech). The plates were incubated for 2 h, washed, and Avidin-enzyme conjugate (AP-Avidin; Calbiochem, Life Technologies, Inc.) was added to each well. Plates were incubated at room temperature for 1 h, and the enzyme substrate nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (Invitrogen-Life Technologies, Inc.) was added to each well and incubated at room temperature for 5–10 min. The reaction was terminated by washing with tap-water upon the emergency of dark purple spots. The spots were counted using the ImmunoSpot Series 2.0 Analyzer (CTL Analyzers, LLC, Cleveland, OH), and the peptide-specific CTL frequency could be calculated from the numbers of spot-forming cells/105 input CD8+ cells after subtraction of the number of background spots formed without added peptide.
Multimerized peptide/HLA complexes were used to identify antigen-specific T cells in situ in tumor lesions of cancer patients, as described previously (14) . Biotinylated B*2705 HLA molecules complexed with the BRaf27R2E8 or the BRaf27R2V8 monomer were supplied by Proimmune limited (Oxford, United Kingdom). The biotinylated monomers were multimerized with streptavidin-FITC-conjugated dextrane molecules (kindly provided by Lars Winther; DAKO, Glostrup, Denmark) to generate multi-valent HLA-dextrane compounds for immunohistochemistry. Tissue sections were dried overnight and subsequently fixed in cold acetone for 5 min. All incubation steps were performed in the dark at room temperature: (a) 45 min of the primary antibody directed against CD8 (clone 33291A, 1:200 diluted; PharMingen, Heidelberg, Germany) or granzyme B (clone ab4059, 1:100 diluted; Abcam, Cambridge, United Kingdom); (b) Cy 3-conjugated goat antimouse (1:500 diluted; code 115-165-100; Jackson ImmunoResearch, obtained from Dianova, Hamburg, Germany) for 45 min; and finally (c) the multivalent HLA-dextrane compounds for 75 min. Between each step, the slides were washed twice for 10 min in PBS/BSA 0.1%. The slides were mounted in Vectashield and kept in the refrigerator until observed under the confocal microscope (Leica).
Isolation of Peptide-Specific T Cells.
Antigen-specific cells were isolated by means of BRaf27R2E8/HLA*B2705-coated magnetic beads as described previously (15) . Biotinylated peptide/HLA monomers were coupled to streptavidin-coated magnetic beads (Dynabeads M-280; Dynal A/S, Oslo, Norway) by incubating 2.5 μg of monomers with 5 × 106 beads in 40 μl of PBS for 20 min at room temperature. The magnetic complexes were washed three times in PBS in a magnetic field (Dynal A/S) and subsequently mixed with PBLs, at a ratio of 1:10 in PBS with 5% BSA, and rotated very gently for 1 h. Antigen-specific CD8+ T cells associating with magnetic complexes were gently washed three times. Isolated cells were resuspended numerous times in X-vivo with 5% human serum and incubated for 2 h before the magnetic beads were released and removed from the cell suspension. The isolated cells were cultured in a 96-well plate in X-vivo with 5% human serum. One day after isolation, 20 units/ml interleukin-2 were added, and on day 5, the capacity of these cells to kill target cells was tested either by ELISPOT or in standard 51Cr release assays.
Conventional [51Cr]release assays for CTL-mediated cytotoxicity was carried out as described elsewhere (13) . Target cells were T2 cells with or without the relevant peptide, as well as a HLA-B*2705 melanoma cell line harboring the V599E mutation (FM56).
RESULTS AND DISCUSSION
To identify putative MHC class I epitopes, peptide prediction was performed using the MAPPP algorithm 7 with the V599EBRaf aa 579–611 fragment comprising the mutation. To circumvent a bias due to the prediction algorithm, two algorithms predicting proteasome cleavage (PAProC and FRAGPREDICT) are used in combination with two MHC class I binding algorithms (SYFPEITHI and BIMAS) using HLA-binding matrices covering 10 different HLA-A alleles, including subtypes, and 21 different HLA-B alleles, including subtypes. The used HLA matrices include all HLA-A and HLA-B alleles with carrier frequencies >6% in stage IV melanoma patients with the exception of HLA-B18 for which no matrix was available. 8 The MAPPP prediction was performed using rather low cleavage probabilities (>0.4 for both algorithms) and low relative binding affinities (>0.5 for SYFPEITHI and >0.015 for BIMAS) to get a broad panel of MHC class I peptides for an initial screening. Despite the low stringency, only two candidate epitopes and MHC class I molecules were identified (Table 1) ⇓ : HLA-A*3302 (FGLATEKSR) and HLA-B*2705 (GDFGLATEK). Albeit the high overall score attributed by the MAPPP algorithm, the HLA-A*3302 epitope was predicted by only one protein cleavage algorithm with the need for simultaneous NH2 and COOH-terminal trimming; hence, a much lower probability was predicted for the natural occurrence of this epitope (10) . Furthermore, HLA-A*3302 itself has a low frequency in stage IV melanoma patients, 8 and therefore, this putative epitope was no longer evaluated. In contrast, the putative HLA-B*2705 epitope was predicted by both processing algorithms and ranked first among peptides comprising the mutation for binding with the HLA-B*2705 molecule with both matrices. Therefore, despite the fact that the multiplicative BIMAS algorithm predicted a rather low relative binding score, we focused on this epitope.
In vitro binding assays confirmed that the peptide bound only very poorly to HLA-B*2705 (Fig. 1, A and B) ⇓ . However, many of the established T-cell epitopes presented by melanoma cells such as gp100 and MART-1 have relatively low binding affinities to the respective HLA class I molecules (16) . Hence, it is common practice to generate heteroclitic peptides from such low-affinity epitopes by substitution of amino acids at specific positions, i.e., the anchor positions, which are crucial for the binding of the peptide to the HLA molecule (17, 18, 19) . Consequently, we synthesized a peptide analogue, BRaf27R2E8 (G R F G L A TE K), in which a more suitable anchor residue for HLA-B*2705 (arginine) replaced the natural aspartic acid at position 2. Indeed, this heteroclitic peptide bound with similar affinity to HLA-B*2705 as the high-affinity binding peptide HIV gag263–272 (Fig. 1, A and B) ⇓ . For control purposes, we also synthesized a similarly modified peptide deduced from the nonmutated BRaf amino acid sequence, BRaf27R2V8 (G R F G L A TV K), which bound with a comparable affinity to HLA-B*2705 as BRaf27R2E8 (data not shown).
Subsequently, we scrutinized PBLs from HLA-B*2705 melanoma patients for the presence of specific T-cell responses against the modified, mutated BRaf27R2E8 (G R F G L A TE K) peptide by means of the ELISPOT IFN-γ secretion assay (20) . It is important to note that we and others have previously been able to measure spontaneous T-cell responses against modified peptides in cancer patients by the ELISPOT without detectable responses against the native analogs. This is largely due to technical reasons, e.g., stabilization of sufficient amounts of class I molecules on the surface of the T2 cells (19, 20, 21) ; hence, because of limitations of available samples from HLA-B*2705 positive patients, we initially restricted our analysis to the modified peptides. As depicted in Fig. 2A ⇓ , BRaf27R2E8 specific T-cell responses were present among PBL of 7 of 28 patients. In contrast, no spontaneous immune responses against BRaf27R2V8, (G R F G L A TV K) deduced from the nonmutated BRaf could be detected in any of the analyzed patients (n = 18, including the BRaf27R2E8-reactive patients). Correlation of the BRAF mutational status and the presence of BRaf27R2E8 specific T-cell responses revealed that such responses were not detectable in any of the patients with a homozygote wtBRAF genotype (0 of 9 patients, 0% reactive; Fig. 2A ⇓ ), whereas four of the BRaf27R2E8 responses were present in the patients (4 of 10 patients, 40% reactive), which had been confirmed to harbor the V599EBRaf mutation in their tumor lesion; for the remaining 3 patients with BRaf27R2E8 responses, the BRAF mutational status remains unknown because no tumor material was available for analysis (3 of 9 patients, 33% reactive; Fig. 2A ⇓ ). Most noteworthy, in two of the patients displaying a V599EBRaf genotype in the primary tumor, we could not detect this genotype in any of the subsequently evolving metastases suggesting an immune selection of melanoma cells with a wild-type BRAF genotype (3) .
For a limited number of patients enough cells were available to characterize the ability of the mutated nonaltered HLA-B*2705-restricted peptide (BRaf27D2E8) to react with peripheral blood T cells of melanoma patients. Hence, ELISPOT assays were performed comparing the reactivity of PBL from melanoma patients toward the mutated nonmodified BRaf27D2E8 with modified BRaf27R2E8 presented by HLA-B2705 T2 cells. This analysis demonstrated that although the response to the BRaf27D2E8 was not as strong as the response against BRaf27R2E8, we were still were able to detect a response (Fig. 2B) ⇓ . Because of limited patient material, we were only able to examine material from nine patients, of which, only four reacted against the BRaf27R2E8 peptide. The anti-BRaf27D2E8 responses were detected in patients hosting stronger responses against BRaf27R2E8. This observation is in concordance with our in vitro binding assays.
We were also able to analyze a small number of tumor infiltrating lymphocytes obtained from primary and metastatic melanoma lesions of HLA-B*2705-positive patients for reactivity against BRaf27R2E8. This analysis showed the presence of reactive T cells in two of three samples (data not shown), indicating that T-cell responses to V599EBRaf may indeed be responsible for observed loss of melanoma cell harboring this mutations. To additionally investigate this notion, i.e., whether mutated BRaf-specific CTLs actually reside in the tumor microenvironment, BRaf27R2E8/HLA-B*2705 complexes were multimerized using FITC-conjugated dextrane molecules, and these complexes were used to stain fresh-frozen samples as described previously (13 , 14) . Fig. 3 ⇓ depicts representative immune fluorescence results from sections of two representative melanoma lesions, i.e., a primary tumor and a lymph node metastases, from a total of 5 HLA-B*2705-positive patients. BRaf27R2E8/HLA-B*2705-reactive cells were counterstained with Cy3-conjugated anti-CD8 (Fig. 3, A and B) ⇓ or anti-granzyme B antibodies (Fig. 3, C and D) ⇓ . The costaining of BRaf27R2E8/HLA-B*2705 with the anti-granzyme B antibody suggests that these cells are able to execute cytotoxic activity. For control, BRaf27R2V8/HLA-B*2705 multimers were used under identical staining and scanning conditions, which showed no reactive cells (Fig. 3, E and F) ⇓ .
To confirm the cytotoxic capacity of BRaf27E8/HLA-B*2705-reactive T cells, we isolated V599EBRaf-reactive T cells by means of magnetic beads coupled with BRaf27R2E8/HLA-B*2705 complexes as described previously (Fig. 4A ⇓ ; Refs. 13 , 15 ). Cells were stimulated once with peptide in vitro before isolation. One day after isolation interleukin-2 was added, and on day 5, the capacity of these cells to kill peptide loaded T2-B*2705 cells was tested in standard 51Cr release assays. To this end, T2-B*2705 cells loaded with either the modified mutated BRaf27R2E8 or the nonmutated BRaf27R2V8 peptide analogues served as targets. This assay revealed that only T2-B*2705 cells pulsed with BRaf27R2E8 were killed (Fig. 4B) ⇓ . In a second series of experiments, we used these BRaf27E8/HLA-B*2705-reactive T cells to test their capacity to kill melanoma cells harboring the V599EBRaf mutation. For this purpose, the well-characterized HLA-B*2705-positive melanoma cell line FM56, which is heterozygous for the V599EBRAF, served as target cells. This cell line was killed with high efficacy (Fig. 4C) ⇓ , indicating both the processing and presentation of BRaf27D2E8 by melanoma cells, as well as the recognition of this nonmodified peptide epitope in the context of HLA-B*2705 by BRaf27R2E8/HLA-B*2705-reactive T cells.
Interestingly, both cleavage algorithms FRAGPREDICT and PAProC attribute a cleavage P < 0.1 to each of the amino acids LATEK of the mutated epitope using the described 33 aa fragment for prediction. In contrast, FRAGPREDICT predicts a >0.99 cleavage probability after the valine residue in the wild-type LATVK motive, which is paralleled by a relatively high probability in the PAProC algorithm. Although we have no experimental evidence for these predictions, it is tempting to speculate that presentation occurs only in the case of the mutated peptide and, to a certain degree, also in the case of the V599K mutation, whereas it is very likely that the wild-type epitope is destroyed because of the very high cleavage probability after the valin residue. In this respect, it is important to note that recently a 29-mer BRaf peptide incorporating the V599E mutation was used for in vitro stimulation of lymphocytes derived from melanoma patients, generating MHC class II-restricted CD4 T cells specific for this peptide or melanoma cells expressing V599EBRaf (22) .
In summary, the present article provides direct evidence for spontaneous immune responses to the constitutively active V599EBRaf in melanoma patients. During the past decade, new insights have been gained into the role of T lymphocytes in the host’s immune response to cancer in general and to melanoma in particular, suggesting that the immune system is essential in the control of tumor growth (23) . Specific immune therapies have accomplished objective clinical responses in a small but significant number of patients (24 , 25) . However, it has also become evident in animal cancer models that targeting to mutated tumor peptides produces more favorable results than using nonmutated self-antigens such as differentiation antigens (26 , 27) . In human patients, a correlation between the presence of T cells specific for unique tumor mutations and a better prognosis has also been reported (28 , 29) . This suggests that HLA class I-restricted mutated epitopes might be excellent antigens to target by immunotherapy. In this respect, oncogenic mutations may be especially good target antigens, as far as CTL killing of tumor cells, expressing the oncogene would lead to a direct tumor growth disadvantage, even if antigen-loss variant tumor cells were not eradicated by the therapy.
We thank Merete Jonassen, Claudia Siedel, and Katrin Müller-Blech for excellent technical assistance and the Danish Melanoma Group for their continued support.
Grant support: The Danish Cancer Society, the John and Birthe Meyer Foundation, Christian og Ottilia Brorsons rejselegat, the Bundesmisterium für Forschung und Bildung (Interdiszipliäres Zentrum für klinische Forschung Würzburg, project B17), the Deutsche Krebshilfe Grant 10-1845-Be, and the Deutsche Forschungsgemeinschaft Grant KFO 124/1-1.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Requests for reprints: Jürgen C. Becker, Department of Dermatology and Dermato-Oncology, University of Würzburg, Josef Schneider Str. 2, D-97078 Würzburg, Germany. Phone: 49-931-201-26396; Fax: 49-931-201-26700; E-mail:
↵7 Internet address: http://www.mpiib-berlin.mpg.de/MAPPP/.
↵8 J. Fensterle and J. C. Becker. Are HLA-B8 and HLA-B35 key MHC class I restriction elements for protective T-cell epitopes in melanoma? Manuscript in preparation.
- Received March 17, 2004.
- Revision received May 20, 2004.
- Accepted June 1, 2004.
- ©2004 American Association for Cancer Research.