A Conserved E7-derived Cytotoxic T Lymphocyte Epitope Expressed on Human Papillomavirus 16-transformed HLA-A2+ Epithelial Cancers

Human Papillomavirus 16 (HPV-16) has been identified as the causative agent of 50% of cervical cancers and many other HPV-associated tumors. The transforming potential/tumor maintenance capacity of this high risk HPV is mediated by two viral oncoproteins, E6 and E7, making them attractive targets for therapeutic vaccines. Of 21 E6 and E7 peptides computed to bind HLA-A*0201, 10 were confirmed through TAP-deficient T2 cell HLA stabilization assay. Those scoring positive were investigated to ascertain which were naturally processed and presented by surface HLA molecules for CTL recognition. Because IFNγ ELISpot frequencies from healthy HPV-exposed blood donors against HLA-A*0201-binding peptides were unable to identify specificities for tumor targeting, their physical presence among peptides eluted from HPV-16-transformed epithelial tumor HLA-A*0201 immunoprecipitates was analyzed by MS3 Poisson detection mass spectrometry. Only one epitope (E711–19) highly conserved among HPV-16 strains was detected. This 9-mer serves to direct cytolysis by T cell lines, whereas a related 10-mer (E711–20), previously used as a vaccine candidate, was neither detected by MS3 on HPV-transformed tumor cells nor effectively recognized by 9-mer specific CTL. These data underscore the importance of precisely defining CTL epitopes on tumor cells and offer a paradigm for T cell-based vaccine design.

The transforming potential of human Papillomavirus (HPV), 4 first suspected in the 1970s, has now been firmly established both biologically and epidemiologically (1)(2)(3). The single most important variable linked to malignant transformation is persistent infection with one of the high-risk HPV types. The E6 and E7 proteins encoded by high-risk HPVs have transforming activities and functionally inactivate the p53 and retinoblastoma (Rb) tumor suppressor proteins, respectively (3,4). HPV-16 is the most abundant high risk HPV and has been detected in Ͼ50% of cervical cancer cases and in most other HPV-induced tumors, such as carcinomas of the vagina, anus, vulva, penis, and oropharynx (3,5,6). Worldwide, high risk HPVs are thought to be responsible for Ͼ500,000 malignancies per year, representing more than 5% of human cancers (7).
A major breakthrough in combating HPV-induced disease was the development of prophylactic vaccines to prevent HPV infection in previously unexposed individuals. These vaccines are based on virus-like particles consisting of the L1 capsid protein (8,9). Virus-like particles resemble natural virions and are able to induce high titers of L1-neutralizing antibodies. Two vaccines, one against HPV-16, -18, -6, and -11 and another against HPV-16 and -18, were approved for clinical use in 2006 (10 -12). Although the impact of prophylactic HPV vaccination on the incidence of vaccine type HPV-associated disease and cancer is unquestionable over time, these vaccines have no therapeutic efficacy for established HPV infections. Antibodies neutralize virus particles only before infection. Moreover, as HPV capsid proteins are exclusively expressed late in the viral replication cycle within the upper layers of the epithelium, immune responses against capsid proteins do not affect persistently infected basal cells and, thus, fail to clear the infection (13). Moreover, high risk HPV-associated cancers generally represent nonproductive infections, and the capsid proteins are not expressed (5,6). For these reasons, viral capsid based strategies are not useful in the development of therapeutic HPV vaccines.
HPV-16 infection is widespread in the sexually active population, but Ͼ95% of infections are either transient and/or cleared by the immune system (13). Regression of lesions has been shown to be dependent on strong localized cell-mediated immune responses. In particular, antigen-specific T cell-mediated immunity is required for the clearance of persistent highrisk HPV infections (14). Hence, the immune system is capable of terminating high risk HPV-associated lesions and tumors. Therapeutic vaccines aimed to induce targeted T cell-mediated immune responses against dysplastic and neoplastic cells, therefore, seem a logical extension for achieving beneficial clinical results. Given that E6 and E7 are consistently expressed in HPV-associated cancers, these proteins themselves represent promising targets for vaccine design. Although most tumor-specific antigens are derived from normal or mutated endogenous self-proteins (15) (TANTIGEN: Tumor T cell Antigen Database), E6 and E7 are foreign viral antigens. These two proteins are required for the induction and maintenance of the malignant phenotype of high-risk HPV-associated cancer cells (5,6), and because HPV uses the cellular DNA replication machinery for genome synthesis, the mutation rate of HPV proteins is low. Thus, it is unlikely that HPV will evade immune attack through loss or mutation of the E6 and/or E7 gene products (16).
Studies on therapeutic vaccines, therefore, have mostly focused on E6 and E7 as target antigens. To date these targets have been delivered as naked DNA vaccines, with recombinant viral or bacterial vectors, as protein or peptide vaccines, and as fusion constructs with toll-like receptor agonists or proteins that enhance antigen delivery or presentation (for review, see Refs. 13, 17, and 18). Most clinical studies have been performed using DNA vaccines. These include a DNA HPV-16 E7 vaccine that has been tested with various fusion partners to enhance antigen presentation and with another DNA vaccine encoding E6 and E7 peptides from HPV-16 and -18 (for review, see Ref. 19). Live viral vectors also have been tested in the clinic, with vaccinia virus constructs coding for either bovine Papillomavirus E2 (20,21) or HPV-16 and -18 E6 and E7 (22). As for protein vaccines, a fusion protein of heat shock protein 65 with HPV-16 E7 has been tested in three phase II trials (23)(24)(25). In addition, a fusion protein of HPV-16 E6, E7, and L2 was also in a phase II trial (26), and a HPV-16 E7 fusion protein with a Haemophilus influenza protein or HPV-16 E6 and E7 were applied in phase I trials in various adjuvants (13,27). Such studies, however, have yielded disappointing clinical responses.
For induction of HPV-specific T lymphocytes in a focused manner, vaccination against defined epitopes is an attractive option. Indeed, various MHC class I-restricted CTL epitopes of HPV-16 E6 and E7 have been tested in early phase clinical studies (28 -34). Nonetheless, little or no benefit over historic controls has been observed. Recently, multiple long synthetic peptide fragments of E6 and E7 have been used to create a polyepitope vaccine, which when tested in patients with HPV-16-positive vulvar intraepithelial neoplasia, exhibited promising clinical efficacy (35). This type of vaccination-induced clinical response has been the most efficacious to date and argues that a robust outcome can be engendered by peptides in conventional adjuvants.
One of the key practical challenges to specific epitope-based vaccines to stimulate cytotoxic T lymphocytes stems from the fundamental nature of T cell receptor (TCR)-based recognition. TCR recognition is referred to as MHC-restricted as, unlike antibody-based recognition, a TCR physiologically interacts with a peptide in complex with an MHC molecule (pMHC) (for review, see Refs. 36 and 37). Further complexity to TCR-based recognition is that a given peptide binds to some but not all MHC molecules. Each human being expresses 3-6 MHC class I molecules (so-called HLA molecules) and at least as many MHC class II molecules. More than 3000 variants of human MHC class I and 1000 variants of MHC class II have been characterized throughout the world to date (38). Cytotoxic T cell recognition of foreign protein antigens occurs via short (generally 9 -10 amino acids long) peptides produced through proteolytic cleavage in the cytoplasmic proteasome complex. These are subsequently transported into the endoplasmic reticulum, bound to MHC class I molecules and ultimately displayed on the cell surface as a pMHC. The viral pMHC serves as a flag to target an infected or transformed cell for destruction by a CTL.
Bioinformatic approaches are important tools for peptidebased vaccines and immunotherapy. Computational methods now offer accuracies that are useful in reducing the number of potential candidate peptides that must be tested experimentally for binding to a given MHC allele (39 -41). In silico methods cannot predict, however, which MHC class I-binding peptides are actually processed and displayed on a cell surface. We have developed an MS 3 Poisson detection mass spectrometry approach to directly assess the physical presence of predicted CTL target epitopes on tumors and infected cells. Our "predict/ detect" method achieves sensitivities comparable with that of a T cell with a dynamic range of one peptide among 100,000 pMHCs displayed per cell.
Here for the first time we have interrogated the MHC class I peptide array of several HLA-A*0201 HPV-16-transformed epithelial tumor cells for the presence of any and all predicted HLA-A*0201-binding E6-and E7-derived peptides. Among E6 and E7 proteins, only a single 9-mer epitope was found on all HPV-16 transformants tested. This conserved peptide, termed E7 [11][12][13][14][15][16][17][18][19] , is predicted to have the capacity to bind to the vast majority of globally distributed A2 alleles (100 of 116 HLA-A2 alleles). We suggest that the lack of prior clinical effectiveness of targeted CTL epitope vaccination (32) is a consequence of misidentification of peptides displayed on tumor cells because of the use of indirect immunological surrogates (killing, proliferation, cytokine production, etc.) to judge T cell epitope expression. Our results offer a direct path to select allele-specific targets that should afford tumor protection to a broad population of patients.
In Silico Prediction of Potential T Cell Epitopes of HPV-16 Proteins E6 and E7-Predictions of HLA-A*0201-binding peptides (both 9-mers and 10-mers) were calculated by the three best predictive servers as described previously (43), namely the Immune Epitope Data base and Analysis Resource server, the NetMHC 3.0 server, and the NetMHCpan 2.2 server. The average predicted IC 50 was calculated, and peptides were ranked accordingly. The 21 peptides considered in this study were synthesized by SYNBIOSCI (Livermore, CA). HPLC analysis showed that the purity of the synthesized peptides was Ͼ95%. All peptides had expected masses as confirmed by mass spectrometry. Peptides were reconstituted in DMSO at 100 M each.
Interferon ␥ (IFN␥) ELISpot Assay-CD8 ϩ T-cell responses to the 10 HPV-16 peptides that were found to be binders in the HLA-A*0201 binding assay were quantified by IFN␥ ELISpot assay. Peripheral blood mononuclear cells (PBMC) isolated from six HLA-A*0201-positive healthy donors under Institutional Review Board approval were plated at 200,000 per well with peptides at a final concentration of 10 M in anti-IFN␥ mAb 1-D1K (Mabtech, Cincinnati, OH)-coated polyvinylidene 96-well plates (Millipore, Billerica, MA). For each individual peptide, the assay was run in duplicate. A HLA-A*0201-restricted HIV-1 peptide (LTFGWCFKL-HIV/Nef 137-145 ) was used as a negative control, and a CMV/EBV/influenza peptide mix (CEF Peptide Pool Classic, Cellular Technology Ltd., Cleveland, OH) and phytohemagglutinin as positive controls. Secreted IFN␥ was detected by biotin-labeled anti-IFN␥ mAb 7B6 -1, and the reaction was developed with streptavidin-ALP and the color reagent nitro blue tetrazolium/5-bromo-4chloro-3-indolyl phosphate (Sigma). The number of specific IFN␥-secreting T cells was determined with an automated ELISpot reader, calculated by subtracting the average negative control value, and expressed as the number of spot-forming units (SFU) per 10 6 input cells. A response was considered positive if the activity was at least three times as great as the mean background activity. Of note, three of these six donors tested for HPV-16 antibody scored positive (data not shown).
Nanoscale Immune-affinity Purification by Immunoprecipitation-For each immunoprecipitation, 10 g of anti-HLA-A02 BB7.2 mAb (BD Biosciences) was non-covalently coupled to 20 l of Gamma Bind beads (GE Biosciences) for 1 h at room temperature. Tumor cells were harvested during the log growth phase and washed with PBS. Cells were pelleted, and the washed and dried cell pellet was lysed using 1.5 ml of lysis buffer consisting of 20 mM Tris, pH 8.0, 1 mM EDTA, 100 mM NaCl, 1% Triton X-100, and 60 mM n-octylglucoside (protease inhibitor tablet, Roche Applied Science, and phenylmethylsulfonyl fluoride) for 10 min on ice. Cell debris was removed using centrifugation for 30 min at maximum speed (13,000 rpm) at 4°C. Cleared supernatant was incubated with 20 l of antibody coupled Gamma Bind Plus beads for 2-3 h at 4°C. Beads were washed 4 times using lysis buffer without Triton X-100 and protease inhibitors. Beads were further washed 4 more times with 10 mM Tris pH v8.0. Dried bead-antibody-HLA pellets were stored at Ϫ80°C for a brief period before MS analysis. Peptides were recovered with 10% acetic acid followed by C18 reverse phase extraction and analyzed on MS.
MS 3 Poisson Detection Mass Spectrometry-Mathematical details, Poisson scoring, confidence estimation, numerical sampling, and other general principles of MS 3 detection are described in a separate manuscript with an analytical focus. 5 Nanospray MS 3 detection uses a hybrid quadrupole filter, collision cell, and a linear ion trap mass spectrometer (MDS Sciex QTrap 4000). In nanospray MS 3 detection a complex mixture is analyzed for a limited number of molecular targets. In place of chromatographic separations, a combination of selective isolations and dissociations filters out a fraction of the ion current highly enriched in fragments specific to a target molecule. This fraction is identified against a background of other ion fragments by using a probabilistic measure. MS 3 X/Y spectra are generated by first selectively transmitting a narrow m/z window centered at X into a cell where it dissociates by collision activation. The fragments collect in a linear ion trap downstream of the collision cell. After a collection period, an m/z window centered at Y is isolated in the linear ion trap, and the ion fragments at m/z Y are again dissociated by collision activation. These fragments of a fragment are scanned out and measured to create an MS 3 spectrum. For targeted detection, MS 3 spectra of synthetic versions are first studied for optimal conditions and MS 2 fragment choices, and then reference MS 3 spectra are acquired for each of the chosen MS 2 fragments. Sample MS 3 spectra with corresponding m/z windows and dissociation conditions are acquired, and these spectra are compared against the set of reference spectra using a Poisson probability metric to quantify the likelihood that the experimental spectra contain fragment intensities consistent with the relative arrival rates given by the reference spectra (44). 5 In contrast to chromatographic separations, the different ionizable components are simultaneously present in the ion beam. This means molecular abundance of a target can be measured by use of an added calibrant molecule at known concentration. This is done in two steps. In the first step one measures a solution with known target and calibrant concentrations to relate the MS 3 ion flux of the target to the calibrant in the detection spectra. For example, to quantitate E7 [11][12][13][14][15][16][17][18][19] , known amounts of this and a control peptide P (KSPWFTTK) are added to a mock MHC I workup using 1.2 pmol of ␤-galactosidase digest as a carrier. After C18 trapping and elution into the electrospray buffer, the detection MS 3 spectra of both peptides is taken in an alternating series to compensate for time variation in the nanospray ion flux. The 5 B. Reinhold, D. B. Keskin, and E. L. Reinherz, submitted for publication. MS 3 signal amplitudes of P (base peak at m/z 597.4) and E7 [11][12][13][14][15][16][17][18][19] (base peak at m/z 634.4), corrected for relative amounts, are recorded. As noted in Fig. 8A, after a 2-min MS 3 collection for E7 [11][12][13][14][15][16][17][18][19] at 9 fmol/l in the ␤-galactosidase sample, one measures a signal amplitude of 10 7 (arbitrary units) for m/z 634, whereas P at 10 fmol/l gives 7.5 ϫ 10 6 for m/z 597. This gives the relative molar MS 3 signal response of P to E7 [11][12][13][14][15][16][17][18][19] at about 0.68. In the second step a known amount of the calibrant peptide is added to the sample being quantitatively analyzed for target. For E7 [11][12][13][14][15][16][17][18][19] quantitation in 10 million CaSki cells, 40 fmol of KSPWFTTK is added to the sample at the beginning of the acid elution step. The detection MS 3 spectra of peptide P and E7 [11][12][13][14][15][16][17][18][19] is again taken in alternating sequence. The measured ratio of MS 3 ion flux for E7 [11][12][13][14][15][16][17][18][19] and P in the CaSki sample is 2 ϫ 10 5 /1.3 ϫ 10 7 (m/z 634.4/m/z 597.4). Corrected for the molar response (0.68) as determined in the first step, one has the relative molar amounts of E7 [11][12][13][14][15][16][17][18][19] to P in the sample as 0.0105. As 40 fmol of P was added, this gives the amount of E7 [11][12][13][14][15][16][17][18][19] as 422 amol or 254 million molecules from 10 million CaSki cells. As the nanospray ion source often shows significant intensity variations with time and MS 3 spectra may be collected for long periods, quantitation data are always collected in a series where a single scan of the target alternates with a single scan of the calibrant and the respective scans are then summed.
T Cell Proliferation Assay-5 ϫ 10 5 T cells from the E7 11-19specific T cell line were stimulated with 1 ϫ 10 5 donor B cells loaded with 10 g/ml E7 11-19 on 96-well plates. Donor B cells were obtained through stimulation of donor PBMC with 3T3-CD40L cells as described previously (45). CD40-activated B cells were irradiated at 3200 rads before plating. Cells were plated in DMEM medium supplemented with 10% human serum, 1% L-glutamine, 1% penicillin/streptomycin, and 0.05 mM 2-mercaptoethanol. Plates were incubated in a 37°C tissue culture incubator for 3 days and pulsed with 1 Ci of [ 3 H]thymidine for 16 h. Plates were harvested, and [ 3 H]thymidine incorporation was detected with a liquid scintillation mixture (PerkinElmer Life Sciences Beta Plate Scint) in a luminescence counter (PerkinElmer Life Sciences 1450 LSC).

Analysis of IFN␥ Secretion Associated with HPV-16-specific
Proliferative Responses-IFN␥ quantitation was performed using cytometric bead arrays (BD Biosciences) according to the manufacturer's instructions. Cut-off values were based on the standard curve for IFN␥ (100 pg/ml). Antigen-specific cytokine production was defined as a cytokine concentration above cutoff level and Ͼ2ϫ the concentration of the medium control.
Characterization of T Cell Lines by IFN␥ ELISpot-5 ϫ 10 4 T cells from the E7 11-19 -specific T cell lines were incubated with 1 ϫ 10 4 T2 cells loaded with 10 g E7 11-19 peptide overnight on precoated IFN␥ ELISpot plates. The assay was processed and developed as described above.
Cytotoxicity Assay-Donor B cells from an EBV-immortalized B cell line (Laz 509) were pulsed with 10 g or 10 ng of the respective HLA-A*0201-restricted HPV-16 peptides at 37°C overnight. The cells were washed twice with DMEM and pulsed with 100 Ci/ml 51 Cr for 90 min at 37°C. Target cells were washed three times with serum-free Opti-MEM media (Invitrogen) to remove excess 51 Cr and plated with peptidespecific CD8 ϩ T cells at 30:1, 10:1, 3:1, and 1:1 ratios. After 4 h of incubation, 50 l of culture supernatant were mixed with liquid scintillation mixture (PerkinElmer Life Sciences Optiphase Supermix) and analyzed for 51 Cr release using a luminescence counter (PerkinElmer Life Sciences 1450 LSC). Percent specific chromium release was calculated using the formula (experimental release Ϫ spontaneous release)/(maximum release in 5% Triton X-100 Ϫ spontaneous release) ϫ 100.

In Silico Prediction of Potential T Cell Epitopes of HPV-16
Proteins E6 and E7-HPV is a small non-encapsulated DNA virus containing ϳ8000 bp encoding two major sets of genes (E, early region genes; L, late region genes) that infect stratified squamous epithelium (Fig. 1). As the E6 and E7 proteins bind host regulators of keratinocyte cell division and thereby degrade and/or perturb the cell cycle inhibitors p53 and Rb, respectively, those viral proteins are of keen target interest for immunotherapeutic purposes. Human cell lines transformed by HPV-16 (CaSki, C66-3, and C66-7) or transduced with HPV-16 E6 and E7 containing retroviruses (N/E6E7 and OKF6/ E6E7) are listed in Table 1.
HLA-A*0201-binding Assay-To measure the binding capabilities of predicted peptides, an HLA-A*0201 binding assay was performed using the T2 cell line (47,48). This cell line is TAP1/2-deficient, displaying low levels of HLA-A*0201 on its surface. After exogenous addition of peptides capable of binding to HLA-A*0201, this HLA complex is stabilized on the surface with a concomitant increase in the number of HLA-A*0201 molecules, as determined by mean fluorescence intensity staining using a fluorochrome-labeled anti-HLA-A2 antibody and flow cytometry. Binding was calculated relative to a known strong HLA-A*0201 binder, the TAX 11-19 peptide from the human T cell leukemia virus-1 (HTLV-1). As shown in Fig. 2, peptide E7 86 -93 was found to be the best binder in this assay followed by E7 [11][12][13][14][15][16][17][18][19] . Altogether, the top-ranked predicted peptides were the strongest binders in the T2 assay so that with the exception of E7 12-20 , the peptides with a predicted average IC 50 of Ͼ500 nM were not found to bind to HLA-A*0201 in the T2 assay. The T2 assay results correlated well with the previously published data (46). Those 10 peptides that were determined to bind experimentally were included in further assays. IFN␥ ELISpot Assay-Current epidemiological data suggest that virtually all individuals among the sexually active population have

TABLE 1 HPV-16 transformed and E6/E7 transduced cell lines
The HPV status, names, and origins of cell lines used in the current study are provided. CaSki was obtained from ATCC (CRL-1550 TM ) and described in Pattillo (94). C66-3 and C66-7 were gifts of J. H. Lee (42), whereas N/E6E7 and OKF6/E6E7 were gifts of J. G. Rheinwald (unpublished data).

Cell lines HPV status Origin
CaSki  been exposed to HPV infection (for review, see Refs. 13 and 49 and references therein). We, therefore, examined the ability of T cells from fresh peripheral blood mononuclear cells of six HLA-A*0201-positive healthy donors to recognize these HPV candidate peptides in IFN␥ ELIspot assay. As shown in Fig. 3A, SFUs per 10 6 cells were low or undetected in these individuals. However, when SFUs were observed, their size was substantial (Fig. 3B). The low numbers of IFN␥-producing cells reflect the paucity of HPV-specific memory T cells in periph-eral blood. These findings are also consistent with previous studies (50) reporting low HPV-specific SFUs and distinct from the robust memory recall SFU response to CEF (a mix of cytomegalovirus, Epstein-Barr virus, and influenza A virus) peptides or the phytohemagglutinin (PHA) assay control (Fig. 3B). The only HPV peptides eliciting an SFU number 4 -5-fold over background in one donor each were E7 [11][12][13][14][15][16][17][18][19] and E6 29 -38 (Fig.  3A). As a consequence of these equivocal responses, we pursued mass spectrometry analysis to identify which viral peptides are physically displayed on HPV-16-transformed, HLA-A*0201-positive cells.
HLA-A*0201 Immunoprecipitation and MS 3 Analysis of Eluted Peptides-For the investigation of HPV-16 antigen presentation by MS analysis, HLA-A*0201 ϩ HPV-16-transformed tumor cell lines as well as E6/E7 expressing human epithelial cell lines enumerated in Table 1 were used. The analytic approach employed is schematically shown in Fig. 4. In brief, tumor cells (ϳ20 -60 ϫ 10 6 ) were solubilized using detergent buffers, and then HLA-A*0201 molecules were immunoprecipitated with the BB7.2 anti-HLA-A2-specific mAb coupled to Gamma Bind Plus beads. Peptides were recovered from pMHC complexes by acid elution and analyzed on a hybrid quadrupole-linear ion trap mass spectrometer using MS 3 and Poisson statistics as described under "Experimental Procedures." Peptides from HPV-16 E6 and E7 oncoproteins that were shown by T2 assay to increase surface HLA-A2 expression (Fig. 2) were targeted for detection by mass spectrometry. MS 3 spectra from immunoaffinity-purified HLA-A*0201 complexes isolated from HPV-16-transformed cell lines (MS 3 -HLA-A2) were compared against the MS 3 patterns of synthetic peptides (MS 3 -reference) using a probabilistic metric (44 5 as shown in Fig. 5. Unexpectedly, only one peptide, E7 [11][12][13][14][15][16][17][18][19] , was easily identified, whereas E6 29 -38 , the sole other peptide detected, was near the limit of sensitivity. For E7 [11][12][13][14][15][16][17][18][19] , the double-charged molecular ion was selected at m/z 555.3 and dissociated. The proline in E7 [11][12][13][14][15][16][17][18][19] (YMLDLQPET) tends to direct fragmentation to the amide bond on the amino side of the proline residue. This generates a strong b 6 (YMLDLQ-) fragment at m/z 764.4, making an optimal candidate for MS 3 detection but suppressing the detection by other fragments (supplemental Fig.  S1). The molecular ion is abundant in the peptides recovered from the HPV-16-positive CaSki cervical carcinoma line so that the dissociation pattern of the b 6 fragment from the synthetic peptide is immediately recognized in the MS 3 555.3/764.3 spectrum (Fig. 5, A and B, and supplemental Fig. S3). The Poisson signature of detection (peak at 0 m/z shift, Fig. 5C) is clear cut but, in this case, largely superfluous. The b 8 fragment (YMLDLQPE-) was also detected in the MS 3 555.3/990.4 spectrum of recovered peptides (supplemental Fig. S2). In contrast to detecting the b 6 fragment of The TAX [11][12][13][14][15][16][17][18][19] binding was set at 1.0. The order of peptides is from predicted strongest to weakest binders, top to bottom, respectively. Shaded entries are E6-derived peptides, whereas unshaded entries are E7-derived.   Fig. 5, G-I) is not so evident by inspection of the MS 3 spectra (Fig. 5, G and H) but does produce the detection signature using the Poisson metric (Fig. 5I). E6 29 -38 detection was further supported by a similar Poisson analysis for the signature of the b 7 ion (TIHDIIL-) at m/z 806.5 (data not shown).
Unexpectedly, the peptide recovery as characterized by mass spectrometry from the C66-7, N/E6E7, and OKF6/E6E7 lines compared with the recovery from the C66-3 and CaSki lines was substantially lower than the amount expected from comparing HLA-A2 expression by flow cytometry. The reason for this is currently under study, but the large variation in overall recovery is best addressed by a direct method of comparing the relative fraction of E7 [11][12][13][14][15][16][17][18][19] to total peptide among the different cell lines. Such a measure can be obtained from the MS 2 555.3  (Fig. 7). These spectra show a number of intense shared peaks at the high m/z end that arise from common terminal losses. Different peptides with molecular masses near 1108. 6 Da are co-selected as double-charged ions in the m/z 555.3 window, and if these ions lose a common amino acid residue at the amino or carboxyl terminus, their product ions will appear as a single m/z peak. For example, m/z 992.6, 978.6, 949.5, and 859.5 could be b-type ions that arise from the carboxyl-terminal loss of Val, Leu or Ile, Ala-Ala, and (Leu or Ile)-Glu, respectively. The peak at m/z 978.6 could be a y-type ion arising from the amino-terminal loss of methionine and so on. Because these high m/z peaks reflect the contribution of many peptides, their collective intensity is a qualitative measure of the peptide background. In contrast the peak at m/z 764.4 that appears above the background in the C66-3-, CaSki-, and E7 11-19 -loaded Laz 509 B cell line spectra is shown by MS 3 to be predominantly a fragment of the E7 [11][12][13][14][15][16][17][18][19] peptide. The ratios of m/z 764.4 to the peaks at m/z 992.6, 978.6, 949.5, and 859.5 (Fig. 7) are a relative measure of the E7 [11][12][13][14][15][16][17][18][19] fraction and show that the low MS 3 ion flux for the E7 [11][12][13][14][15][16][17][18][19] peptide in the C66-7, N/E6E7, and OKF6/ E6E7 lines (Fig. 6) is not just a reflection of low peptide recovery overall but that the fraction of E7 [11][12][13][14][15][16][17][18][19] to total peptide is reduced compared with that fraction in C66-3-, CaSki-, or 10 ng/ml E7 11-19 -loaded Laz 509.

DISCUSSION
HPV-induced dysplasia and cancer cause significant morbidity worldwide (7). Although prophylactic vaccines are now available, immunization does not reach everyone at risk. Given that HPV-associated cancers develop years and often decades after initial infection, it was predicted that no measurable decline of HPV-associated cancers in women may occur before 2040. This prediction was based upon higher acceptance rates for the vaccines than is currently achieved in the United States (for review, see Ref. 13). Furthermore, the approved prophylactic vaccines have no therapeutic effects (54), leaving HPV-infected individuals in need of treatment options. Fortunately, most HPV infections are cleared naturally by the immune system (14). If the lesions do not regress, surgical treatments are necessary. These procedures are associated with significant morbidity ranging from dysfunction to infertility depending on the site and stage of the lesion. A noninvasive treatment such as a therapeutic vaccine fostering an effective anti-HPV state would be an attractive alternative. A vaccine could be offered to patients who do not clear HPV infection spontaneously during a finite observation period as well as to patients with established lesions.
The lack of clinical impact was thought to be a consequence of an advanced stage of disease in the patient groups. However, as there is evidence that HPV infection influences antigen presentation, this lack of success might be caused by a paucity (or even absence) of epitopes presented on HPV-16-transformed cells. In this regard several HPV immune evasion mechanisms have been described (for review, see Refs. 70 -72) including down-regulation of components of the antigen-processing machinery and MHC class I molecules (73,74), resulting in decreased presentation of antigenic peptides. Furthermore, precise and direct identification of T cell epitopes expressed on HPV-transformed cells has been lacking. Instead, determination of relevant epitopes has been inferred by bioinformatic prediction, synthetic peptide HLA binding studies, and peripheral T cell functional activation readouts employing various immunologic assays. However, because the success of a therapeutic vaccine is dependent on accurate identification of HPV epitopes displayed as pMHC on HPV-infected target cells or HPV-transformed tumor cells, it is essential to define HPV-16 E6 and E7 T cell epitopes that are naturally processed and presented on the surface of virally altered cells. Only those HPV peptide/MHC class I complexes are capable of being recognized by cytolytic T lymphocytes to target destruction of transformed cells.
To this end, we have developed a new methodology, nanospray MS 3 Poisson detection mass spectrometry. This methodology works by filtering the ion beam through two stages of mass selection and fragmentation (generating MS 3 spectra) and detecting a target molecule by a probabilistic measure of the  [11][12][13][14][15][16][17][18][19] . The common peaks at the high m/z end are fragments from different peptides sharing amino or carboxyl terminal amino acids and a molecular mass near 1108.6 Da (hence, co-selected in the m/z 555.3 window). Because the intensity of these high m/z peaks is an average of many peptides, their intensity serves as an approximate measure of the peptide background. Their amplitude relative to m/z 764.4 provides in a single spectrum a characterization of the fraction of A2-bound peptide that is E7 [11][12][13][14][15][16][17][18][19] (see "Results"). The C66-7, N/E6E7, and OKF6/E6E7 samples show not just lower absolute amounts of E7 [11][12][13][14][15][16][17][18][19] (Fig. 6) but also that E7 11-19 is a smaller relative fraction of the total peptide population.
target's known dissociation patterns in the MS 3 spectra. The methodology combines instrumental and ionization optimizations in a detection mode format to provide a high dynamic range from limited sample amounts. An instrumental geometry in which a quadrupole filter is placed in front of an ion trap (QTrap 4000) achieves a high duty cycle for MS 3 spectra. Static nanospray avoids losses from surface exposure associated with chromatography and in its low (a few nanoliters per minute) flow, an optimal conversion of molecules in the condensed phase into gas phase ions.
Our findings are that none of the clinically targeted A2 peptides employed in epitope-based T cell vaccines to date could be detected on HPV-16-transformed cell lines tested herein. Two examples of this discordance are of particular note. The E7 86 -93 peptide has been previously reported to be by far the best HLA-A*0201-binding peptide derived from HPV-16 (46). It is among the top predicted binders in the present study and the strongest binder in the HLA-A*0201 T2 binding assay. Nonetheless, E7 86 -93 could not be detected by mass spectrometry on any of the HPV-16-transformed tumor cell lines. Furthermore, MS and fragmentation analyses of the peptides recovered from E7 86 -93 -loaded T2 cells showed that more than 90% of the E7 86 -93 peptide complexed with HLA-A*0201 was modified with an additional cysteine that could be localized to the backbone cysteine residue and is most likely linked via a disulfide bridge (data not shown). This modification was also checked by MS 3 analysis but was not detected on an HPV-16-transformed cell line. Likewise, E7 [11][12][13][14][15][16][17][18][19][20] was not present on HPV-16 human tumor cells, although E7 [11][12][13][14][15][16][17][18][19] was rather abundant. Based on our current observations, we suggest that one reason for the lack of clinical efficacy of epitope-based therapeutic T cell vaccines for HPV-16 and, by extension, other infectious diseases and cancers may be that vaccine-elicited responses were misdirected.
The striking specificity of human CTL for E7 [11][12][13][14][15][16][17][18][19] is worth underscoring. E7 11-19 -specific T cells recognize and lysed E7 11-19 -loaded but not E7 11-20 -loaded target cells even though both peptides bind well to HLA-A*0201. Alloreactivity as a basis for cytotoxicity in the assay was excluded by using a peptide pulsed autologous B cell line (Laz 509) as a target. Although not shown, E7 11-20 10-mer-specific T cells lysed autologous E7 11-20 but not E7 11-19 -pulsed B cells as well. Why should this recognition be so discrete in view of the fact that the two peptides differ from one another by only a single carboxylterminal threonine residue? The answer lies in the general nature of peptide binding to MHC class I molecules. One pocket exists for the amino terminus and a second for the carboxylate of the peptide, thereby docking the peptide ends in the HLA groove between ␣1 and ␣2 helices. The fixed "ends" mandate that the 10-mer peptide bulges further out of the HLA binding groove than the 9-mer, with greater surface area exposed to the TCR. Both the position of the main chain and side chain positions are altered. Hence, these two peptides will differ in their TCR recognition features (Ref. 76 and references therein). Failure to identify the precise epitope for use in the FIGURE 11. E7 [11][12][13][14][15][16][17][18][19] binds to the vast majority of HLA-A2 alleles. HLA binding predictions of E7 [11][12][13][14][15][16][17][18][19] were performed using NetMHCpan on the 116 known HLA-A2 alleles. Panel A shows the predicted IC 50 values for each allele (in nM) with strong binding (Ͻ50 nM) shaded magenta, weak binding (50 -500 nM) shaded green, and no binding (Ͼ500 nM) unshaded. Panel B gives a bar graph representation of E7 11-19 IC 50 for each allele along the x axis corresponding to those going from the left to right columns in the order defined in panel A.  SEPTEMBER 17, 2010 • VOLUME 285 • NUMBER 38 vaccine elicitation strategy will likely engender a misdirected response even for such similar sequences.

Tumor Antigen and MS 3 Poisson Detection
A recent study of 19 women with grade 3 vulvar intraepithelial neoplasia vaccinated with a non-epitope targeted mix of long peptides from HPV-16 viral E6 and E7 oncoproteins in incomplete Freund's adjuvant showed promising clinical responses (35). At 3 months post-vaccination, 5 women had complete regression, and at 12 months, 79% of subjects appeared to show significant clinical responses. These results demonstrate that therapeutic vaccination harnessing cellular immunity can be effective. But vaccinating with any set of long peptides that span the target antigen and incorporate the expected T cell epitopes is unlikely to be optimal. Broad and nonselective immune responses arising from an uncharacterized processing of vaccine components in secondary lymphoid organs coupled with restricted presentation by primary tumor cells would limit the population of responding CTLs at the tumor and therein directly reduce TCR-mediated cytolysis and secondarily dilute requisite inflammatory signals in the microenvironment. Dendritic cell presentation of multiple T cell epitopes, few of which are relevant, combined with well known mechanisms of immunodominance may further result in misguided responses (76). Targeting a response in a precise way in future immunotherapeutic efforts offering appropriate adjuvant and delivery to dendritic cells will focus T cells on relevant protective/therapeutic epitopes. In addition, by precisely selecting T cell epitopes in vaccine formulation, bioinformatics can be used to calculate population protection coverage, ensuring that there is an adequate breadth of epitopes incorporated.
Our results strongly imply that the presence or absence of memory responses in the peripheral blood is not a useful surrogate to guide immunity to relevant tumor target antigens. Precursor frequencies may be too low, especially if cells have already trafficked to target organs as effector memory populations. Moreover, as antigen is transported to lymph node and presented to T cells on dendritic cells, a focusing of immune response through cross-presentation or other means may select for immunodominant epitopes not necessarily reflective of relevant peptide array displayed on tumor cells. Immune responses against the latter are what will lead to protection. The MS technology described here offers an approach toward making identification of such a display tractable using a limited number of cells.