Residue analysis of a CTL epitope of SARS-CoV spike protein by IFN-gamma production and bioinformatics prediction

Background Severe acute respiratory syndrome (SARS) is an emerging infectious disease caused by the novel coronavirus SARS-CoV. The T cell epitopes of the SARS CoV spike protein are well known, but no systematic evaluation of the functional and structural roles of each residue has been reported for these antigenic epitopes. Analysis of the functional importance of side-chains by mutational study may exaggerate the effect by imposing a structural disturbance or an unusual steric, electrostatic or hydrophobic interaction. Results We demonstrated that N50 could induce significant IFN-gamma response from SARS-CoV S DNA immunized mice splenocytes by the means of ELISA, ELISPOT and FACS. Moreover, S366-374 was predicted to be an optimal epitope by bioinformatics tools: ANN, SMM, ARB and BIMAS, and confirmed by IFN-gamma response induced by a series of S358-374-derived peptides. Furthermore, each of S366-374 was replaced by alanine (A), lysine (K) or aspartic acid (D), respectively. ANN was used to estimate the binding affinity of single S366-374 mutants to H-2 Kd. Y367 and L374 were predicated to possess the most important role in peptide binding. Additionally, these one residue mutated peptides were synthesized, and IFN-gamma production induced by G368, V369, A371, T372 and K373 mutated S366-374 were decreased obviously. Conclusions We demonstrated that S366-374 is an optimal H-2 Kd CTL epitope in the SARS CoV S protein. Moreover, Y367, S370, and L374 are anchors in the epitope, while C366, G368, V369, A371, T372, and K373 may directly interact with TCR on the surface of CD8-T cells.


Background
Severe acute respiratory syndrome (SARS) is an emerging infectious disease caused by the novel coronavirus SARS-CoV [1,2]. The fatality rate is as high as 15% for patients younger than 60 years old and can be higher than 50% for patients 60 years or older. Nearly 40% of infected patients develop respiratory failure that requires assistant ventilation [3].
Coronaviruses (CoVs) are positive-strand RNA viruses. The virion consists of a nucleocapsid (N) core surrounded by an envelope containing three membrane proteins, spike (S), membrane (M) and envelope (E), which are common to all members of the genus [4,5]. The M and E proteins are important for viral particle assembly and N is important for viral RNA packaging. The S protein, which provides the virion with a corona-like appearance, binds to host receptors and mediates membrane fusion [6]. The successful development of effective treatments and vaccines against SARS-CoV depends on understanding the roles of various immune effectors in protective immunity and on identifying protective viral antigens recognized by these effector cells. In a preliminary study, the S protein fragment S358-374 was shown to stimulate the production of IFN-γ by CD8-T cells from immunized BALB/c mice [7].
The capacity of a diverse array of peptides to bind to an individual class I molecule is due to anchor residues in the peptides [8]. The surface features of the binding cleft of the class I MHC molecule are complementary to side chains of the anchor residues in the displayed peptide. The amino acid residues lining the binding sites may vary among different class I allelic variants [9,10]. Here, an SARS CoV S protein CTL epitope, S366-374, was identified and the functions of individual residues were evaluated by bioinformatics tool prediction and by IFN-γ responses induced by a series of modified S366-374 peptides.

Mice
Female BALB/c mice, 6-8 weeks old, were purchased from Zhongshan University Animal Center (Guangzhou, China) and maintained in our animal care facility under pathogen free conditions. For experimental purposes, six to eight week-old female mice were used. All experiments were performed according to the guidelines in the Institutional Animal Committee of Zhongshan School of Medicine, China.

SARS-CoV S DNA vaccine
Plasmids encoding SARS-CoV spike (S) protein was constructed as described [11], and kindly provided by Dr. Gary J. Nabel from Vaccine Research Center, NIAID, National Institutes of Health, MD, USA. Plasmid DNA was purified by plasmid-purified kit (QIAGEN, USA). The 260/280 ratios ranged from 1.8 to 2.0. The endotoxin content from purified plasmid DNA was found below 20 U/ml. The endotoxin level within this range had no effect on the immune response.
Fmoc deprotection was performed with a solution of piperidine in dimethylformamide in a 2/8 (v/v) ratio. Final deprotection of the peptides from the resin was performed in a mixture containing trifluoroacetic acid, phenol, water, triisprpylsilane in a 88/5/5/2 (v/v) ratio for 3.5 h. Peptides were then precipitated by addition of cold diethyl ether and dissolved in a mixture of 0.1% trifluoroacetic acid in water/acetonitrile and lyophilized. Purity of all peptides was checked by analytical high performance liquid chromatography on a Waters instrument using a C18 column (Novarpack, 5 μm, 300 Å, 10.0 × 200 mm) and all were at least 70% pure (UV detection at 214 and 254 nm). They were characterized by electrospray mass spectrometry.

Immunization of mice
Female BALB/c mice were injected (i.m.) with 50 μg/ mouse of SARS-CoV S plasmid DNA in 100 ul of sterile PBS. Mice were boosted twice at 2-3 weeks interval.
Cell culture and IFN-γ ELISA Mice were sacrificed. Spleen from individual mouse was harvested one to two weeks after the final boost vaccination. Single cell suspensions were prepared and plated in a 96-well micro-titer plate at 4 × 10 5 cells/200 μl per well. Pooled SARS CoV S peptides (1 μg/ml for each) or single peptide (1 μg/ml) with anti-mouse CD28 mAb (1 μg/ml) were added to cultures. Supernatants of cell cultures were collected 72 h later, and levels of IFN-γ were assessed by specific ELISA kit (BD PharMingen) according to the manufacturer's protocol. The detection limit of the IFN-γ assay kit was 3.13 pg/ml.

IFN-γ ELISPOT
Assessment of SARS-CoV S-specific IFN-γ producing cells after vaccination was determined by ELISPOT (Diaclone, France) according to the manufacturer's protocol. In brief, single cell suspensions were prepared from spleens of mice after vaccination, and plated in 96well microplate precoated with anti-IFN-γ antibody specific for ELISPOT. Cells were incubated overnight in the presence or absence of peptide (1 μg/ml) and anti-CD28 (1 μg/ml). The plates were then washed and alkaline phosphatase conjugated anti-mouse IFN-γ antibody was added, developed with ready-to-use BCIP/NBT, and read by Champ Spot II ELISPOT reader (Sage Creation, China).

Cell surface and intracellular cytokine staining
Single-cell suspensions from spleens of mice after vaccination were stimulated with or without SARS CoV S peptides plus anti-CD28 (1 μg/ml) for 5 h at 37°C and 5% CO 2 . Brefeldin A (10 μg/ml, Sigma) was added in the last 4 h incubation. Cells were washed, fixed with 4% paraformaldehyde and permeabilized in PBS buffer containing 0.1% saponin (Sigma), 0.1% BSA and 0.05% NaN 3 overnight at 4°C. Cells were then stained with conjugated mAbs specific for CD4, CD8 and intracellular cytokine IFN-γ for 20-3 0 min at 4°C in dark. Cells (300,000) were acquired on flow cytometer (BD Calibur) and data were analyzed with FlowJo program, version 6.0 (Tree Star, Inc., USA). Isotype matched controls for cytokines were included in each staining.
Artificial Neural Network (ANN) is a connectionist models that consist of a number of interconnected units that can be activated by transmitting signals [13,14]. ANN can tolerate a degree of erroneous data, and can classify nonlinear data, which makes them highly suitable for processing noisy biological information. ANN applications have been described for predictions of MHC class I binding peptides and for MHC class II peptides. The prediction accuracy of ANN-based methods was reported to be close to 80% sensitivity and 80% specificity [15].
Stabilized Matrix Method (SMM) is a T cell epitope predictive tool based on quantitative matrices. It has been successfully applied to predicting peptide binding to MHC molecules, peptide transport by the transporter associated with antigen presentation (TAP) and proteasomal cleavage of protein sequences [16,17]. The sensitivity of SMM-based methods was reported to be close to 60% [18].
Average Relative Binding (ARB) is a matrix method, which allows combination of searches involving different peptide sizes and alleles into a single global prediction [19]. ARB has achieved a favorable performance in predicting MHC I and MHC II molecules [20,21].
Application of these tools was according to prompt dialog box showed in the website. In brief: a) Choose MHC class I model. b) Enter sequence of various peptides. c) Choose prediction method as ANN, SMM and ARB, respectively. d) Specify what to make binding predictions for. Select MHC source species as mouse, allele as H-2K d , length as 9. e) Specify output. These tools predict IC 50 values for peptide binding to specific MHC molecules. IC 50 value means binding ability. Note that binding to MHC is necessary but not sufficient for recognition by T cells.
HLA Peptide Binding Predictions tool provided by BioInformatics and Molecular Analysis Section (BIMAS) (http://www-bimas.cit.nih.gov/cgi-bin/molbio/ken_parker_ comboform) was used to compare the binding kinetics of peptides. The analysis is based on coefficient tables deduced from the published literature by Dr. Kenneth Parker [22]. Application of this tool was according to prompt dialog box provided by the website. Higher half lime value means better binding ability.

Statistics
Statistical evaluation of differences between means of experimental groups was performed by analysis of variance and a non-parametric two-tailed t test. P value <0.05 was considered to be significant.

N50 is a MHC-I restricted peptide in SARS-CoV S protein
To identify SARS CoV S epitopes, the potential SARS-CoV S epitopes were tested repeatedly by splenocytes from DNA vaccine immunized BALB/c mice. ELISA and ELISPOT results indicated that the adjacent peptides P50 and P51 possessed the same ability to induce IFN-γ production [9]. The overlapping sequence between P50 and P51 (N50, KCYGVSATKL) was synthesized. ELISA ( Figure 1A Amino acid residue L 374 is essential for stimulation of IFNγ production in response to S365-374 To identify the optimal epitope in S365-374, a series of S358-374-derived peptides were synthesized and used to stimulate splenocytes from SARS-CoV S DNA vaccine immunized BALB/c mice. The fraction of IFNγ-producing T cells was determined by ELISPOT (Figure 2A), and the level of IFN-γ in supernatants was measured by ELISA ( Figure 2B). Both results indicated that IFN-γ was produced only in response to peptides preserving residue L 374 . Thus, S367-374 (YGVSATKL), S365-374 (KCYGVSATKL), and S364-374 (FKCYGV-SATKL) could elicit robust IFN-γ production. Only S370-374 (SATKL) was inactive, likely due to weak affinity to MHC-I (data not shown). In contrast, L 374 deleted peptides, including S369-373 (VSATK), S366-373 (CYGVSATK), and S363-373 (FKCYGVSATK) could not induce IFN-γ production. The IFN-γ response induced by S365-374 was much stronger than that induced by S367-374 (P < 0.05).

S366-374 is the optimal epitope
To identify the optimal epitope, we analyzed the binding affinity of S365-374 peptides to H-2 K d , H-2 D d , and L d by several bioinformatics tools. The MHC-binding scores were determined by three peptide-binding prediction methods: artificial neural network (ANN) [23], stabilized matrix method (SMM) [16], and average relative binding (ARB) [19]. Predicted binding scores were expressed as IC 50 values, which represented the equilibrium dissociation constant (KD) of the peptide in relation to a particular MHC molecule. The binding affinities of all 9 and 10 amino acid peptide stretches in S358-381 were predicted. The data indicated that the binding of 9 aa peptides was stronger than all 10 aa peptides and that these 9 aa peptides binded with higher affinity to H-2 K d than to H-2 D d or H-2 L d (data not shown). Therefore, we concluded that the optimal epitope should be an H-2 K d restricted 9 amino acids peptide. In addition, the results demonstrated that S366-374 (CYGVSATKL) was the highest affinity peptide to H-2 K d ( Table 1).
Predicted binding affinity of the S366-374 series with one mutated residue To elicit an effective IFN-γ response, an epitope must bind to an MHC molecule first. Artificial neural network (ANN) was used to estimate the binding affinity of single S366-374 mutants to H-2 K d ( Table 3). The IC 50 value of wild type S366-374 (CYGVSATKL) was 59.2 nM. When Y 367 and L 374 were replaced by alanine (A), lysine (K), or aspartic acid (D), the IC 50 value increased dramatically (to more than 28,000 nM), indicating that Y 367 and L 374 were important for peptide binding and might serve as the main anchors in this epitope.
Although G 368 A (G 368 replaced by A) and G 368 K isoforms of S366-374 possessed higher affinity than wild type S366-374, they could not elicit IFN-γ responses (Figure 3), indicating that G 368 might directly contact the T-cell receptor. The residues V 369 , A 371 , and K 373 had functions similar to G 368 , while the role of S 370 was distinct. The IC 50 value of S 370 A was 216.5 nM, about 4 times higher than wild type S366-374. The IC 50 values of S 370 K and S 370 D were nearly 1000 nM, indicating that S 370 might be a weak anchor to H-2 K d . The role of T 372 was similar to S 370 , as the IC 50 values of T 372 A, T 372 K, and T 372 D ranged from nearly 1000 nM to 3000 nM.

The production of IFN-γ induced by one residue-mutated S366-374
To confirm the key residues in S366-374, 27 analogs were synthesized in which each of the nine residues was replaced by A, K, or D. These peptides were used to stimulate splenocytes from SARS CoV S DNA immunized mice. ELISPOT ( Figure 3A) and ELISA ( Figure 3B) were performed to detect IFN-γ production.
The frequency of IFN-γ producing cells induced by K 365 A or K 365 D was similar to that induced by wild type S365-374 (KCYGVSATKL), and there was no obvious difference in the IFN-γ levels in the culture cell supernatants following peptide treatment. When C 366 was replaced by A, the frequency of IFN-γ producing cells and the IFN-γ levels in the supernatant were also similar to wild type S365-374. When C 366 was replaced by K or D, the frequency of IFN-γ producing cells was decreased by nearly 50%, and the level of IFN-γ in the supernatant was much lower than that induced by wild type S365-374. The S 370 residue appeared to be more important than C 366 for induction of IFN-γ as IFN-γ responses induced by S 370 A were similar to that induced by wild type S365-374, while treatment with S 370 K and S 370 D peptides resulted in a 75% lower frequency of IFN-γ producing cells and a significantly reduced supernatant IFN-γ concentration compared to wild type S365-374.   The L 374 was also an essential residue in the epitope. No IFN-γ response was detected in mutated peptides without L 374 ( Figure 2). However, when L 374 was replaced by A, IFN-γ responses could still be detected. The frequency of IFN-γ producing cells following L 374 A treatment was about half that induced by wild type S365-374. When L 374 was replaced by K, some IFN-γ spot forming units (SFUs) could also be detected. In addition, Y 367 was important for IFN-γ induction; although about 200 SPU per 10 6 cells could be induced by Y 367 A or Y 367 K, the level of IFN-γ in the supernatant was significantly reduced compared to that induced by wild type S365-374. The K 373 residue had a role similar to Y 367 . When K 373 was replaced by A or D, the IFN-γ responses were similar to those induced by Y 367 mutated peptides.
The most important residues in the epitope were G 368 , V 369 , A 371 , and T 372 . Almost no IFN-γ response could be detected when these residues were replaced by A, K, or D.

C 366 A and S 370 A are H-2 K d restricted epitopes
As shown in Figure 4, IFN-γ responses induced by C 366 A and S 370 A were similar to that elicited by wild type S365-374, indicating that C 366 and S 370 could be replaced by A without affecting peptide function. To

Figure 3
The production of IFN-γ induced by mutated S366-374. BALB/c mice were immunized by SARS CoV S DNA vaccine and splenocytes were prepared as described previously [7]. Each amino acid residue in S366-374 was replaced by alanine (A), lysine (K), or aspartic acid (D). The 27 distinct S366-374 mutants were used to stimulate splenocytes. (A) After incubation for 14-18 h, the frequency of IFN-γ spot forming unit (SFU) was detected by ELISPOT. (B) After incubation for 72 h, supernatants were collected and levels of IFN-γ were detected by ELISA. "0" represent unstimulated control. KYGVSATKL was a scrambled peptide control. Experiments were performed in duplicate and representative results are shown.
further confirm this result, fluorescence activated cell sorting (FACS) was performed. The result showed that C 366 A and S 370 A could only stimulate CD8 + T cells to produce IFN-γ, indicating that both were H-2 K d restricted epitopes.

Discussion
The T cell epitopes of the SARS CoV spike protein are well known, but systematic evaluation of the functional and structural roles of each residue has not been reported for these antigenic epitopes [24][25][26]. In the preliminary study [7], by using a synthesized peptide pool of SARS CoV S protein to stimulate the splenocytes from SARS CoV S DNA vaccine immunized mice, we identified that the peptides of P50 and P51 could induce IFN-γ responses. P50 and P51 contained a same animo sequence, N50 (S365-374, KCYGVSATKL). In present study, N50 was synthesized and used to stimulate the splenocytes from immunized mice, ELISA, ELISPOT and FACS results all indicated that N50 contained a main CD8 + T cell epitope ( Figure 1). Moreover, S366-374 (CYGVSATKL) was shown to be an optimal H-2 K d restricted epitope of the SARS CoV S protein by both bioinformatics prediction (Tables 1 and 2) and a functional INF-γ release assay (Figure 2, Figure 3).
To induce effective T cell responses, the T cell epitope must bind to an MHC molecule. All peptides that bind to class I molecules contain a carboxyl-terminal anchor [27,28]. The anchor residues at both ends of the peptide are buried within the binding cleft, holding the peptide firmly in place. A previous study demonstrated that nonameric peptides bound preferentially and that the main contacts between class I MHC molecules were residue 2 at the amino-terminal end and residue 9 at the carboxyl terminus of the nonameric peptide. These anchors were generally hydrophobic residues (e.g., leucine and isoleucine) [29].
Site-directed mutagenesis is a powerful tool for probing protein or peptide structure and function. Alaninescanning, lysine-scanning, or aspartic acid-scanning by Figure 4 KAYGVSATKL and KCYGVAATKL could induce CD8 + T cells produce IFN-γ. BALB/c mice were immunized as described previously [7]. Splenocytes were prepared and stimulated with KAYGVSATKL and KCYGVAATKL. FACS was performed to determine the expression of IFN-γ in CD4 + and CD8 + T cells. Numbers at the corner represent the percentage of positive (expressing) cells. Representative results from three independent experiments are shown.
systematic replacement of side-chains with alanine, lysine, or aspartic acid have been used widely to study binding sites on proteins [22,30]. Analysis of the functional importance of side-chains by mutational study may exaggerate the effect by imposing a structural disturbance or an unusual steric, electrostatic, or hydrophobic interaction. Alanine-scanning is the least disruptive to the peptide structure because alanine is uncharged and has the smallest amino acid side group next to glycine, and thus has been particularly useful for mapping protein functional domains. In contrast, glycine can change the main chain conformation of the protein [31], so individual alanine mutations are preferred to infer the roles of individual amino acid residues. Charged residues such as lysine and aspartic acid are capable of forming ion pairs and hydrogen bonds, so they often play an important role in protein binding as well as in the recognition of interacting proteins. Thus, lysine and aspartic acid are often used as substitutes to study electrostatic effects between proteins [32].
Computational prediction and modeling of MHC/peptide binding can greatly facilitate peptide screening, with tremendous savings in time and experimental effort. Using these methods, T cell epitopes in many vaccine candidates were identified [33,34]. A number of prediction servers are available for identification of peptides that bind MHC molecules. Lin et al. have reported a comparative evaluation of thirty prediction servers for seven human MHC-I molecules. It showed that the best prediction server across all HLA molecules in this study is NETM_ANN, closely followed by IEDB ANN and IEDB SMM [35,36].
In this study, both peptide-binding prediction methods and functional experiments were used to evaluate the roles of the different residues within the optimal epitope S365-374. Though K 365 was excluded from this optimal epitope (Tables 1 and 2), the wild type S365-374 containing K 365 induced a strong IFN-γ response ( Figure 2). Thus, K 365 was included in these synthesized mutant peptides. A comparison of IFN-γ responses showed that mutants K 365 A or K 365 D were as potent as wild type S365-374 (Figure 3), demonstrating that K 365 is not in this optimal epitope and that residue in this position do not influence the function of S366-374.
The L 374 at position P9 was predicted to be the carboxyl-terminal anchor in the epitope by ANN (Table 3) and the functional experiment confirmed that analogs without L 374 could not induce IFN-γ secretion from immunized mouse splenocytes ( Figure 2). However, L 374 A could still induce significant IFN-γ secretion, while L 374 K and L 374 D could not, suggesting that electrostatic effects and hydrophobic interaction may play an important role at this position.
Residue 2 at the amino-terminal end is another important anchor in many epitopes. In S366-374, Y 367 at position P2 was predicted to be another important anchor by ANN (Table 3). The functional assay showed that Y 367 mutated peptides (Y 367 A and Y 367 K) could induce IFN-γ responses, but that IFN-γ production was lower than that induced by S366-374. That might indicate an important role for the Y side chain in determining the binding affinity to H-2 K d [10].
The S 370 at P5 is also a functionally significant residue in this epitope. Though not a traditional anchor, bioinformatics tools indicated that it might act as a weak anchor for H-2 K d binding. The functional assay indicated that IFN-γ responses induced by S 370 mutant epitopes were stronger than L 374 mutant epitopes.
X-ray crystal lographic analyses of peptide-class I MHC complexes has revealed how the peptide-binding cleft in a given MHC molecule interacts stably with peptides [37]. Vesselin Mitaksov described the crystal structure of the MHC class I protein H2-K d in complex with the antigenic peptide TYQRTRALV (Flu) derived from an influenza nucleoprotein. They found that Flu residues Tyr P2 , Thr P5 , and Val P9 were sequestered into the B, C, and F pockets of the K d groove, respectively [32]. In the sequence of S366-374, the positions Tyr 367 , Ser 370 and Leu 374 were at P2, P5, and P9 as well, implying that Tyr 367 , Ser 370 and Leu 374 are anchors in this optimal CTL epitope.
All three bioinformatics tools indicated that C 366 was included in this optimal epitope (Table 1). In its absences, however, S367-374 could still induce an IFN-γ response, albeit weaker than that induced by wild type S365-374 ( Figure 2). Thus, C 366 could influence the function of this epitope. The ANN tool predicated that C 366 was not an anchor to H-2K d and C 366 mutant peptides could still induce strong IFN-γ responses, especially C 366 A, so our results indicate that C 366 is not an important residue for the peptide-TCR interaction. Conversely, mutations of G 368 , V 369 , A 371 , T 372 , and K 373 induced only modest IFN-γ production, demonstrating that these residues made greater contributions in presenting to TCR in this optimal CTL epitope. As predicated by ANN, T 372 contributed to H-2 K d binding (Table 3). Indeed, IFN-γ responses induced by T 372 mutant peptides were significantly weaker than peptides containing T 372 , indicating that T 372 might contribute to the TCR interaction.

Conclusions
In the present study, we demonstrate that S366-374 is an optimal H-2 K d CTL epitope in the SARS CoV S protein. Moreover, Y 367 , S 370 , and L 374 are anchors in the epitope, while C 366 , G 368 , V 369 , A 371 , T 372 , and K 373 may directly interact with TCR on the surface of CD8-T cells.