Sterile Protective Immunity to Malaria is Associated with a Panel of Novel P. falciparum Antigens*

The development of an effective malaria vaccine remains a global public health priority. Less than 0.5% of the Plasmodium falciparum genome has been assessed as potential vaccine targets and candidate vaccines have been based almost exclusively on single antigens. It is possible that the failure to develop a malaria vaccine despite decades of effort might be attributed to this historic focus. To advance malaria vaccine development, we have fabricated protein microarrays representing 23% of the entire P. falciparum proteome and have probed these arrays with plasma from subjects with sterile protection or no protection after experimental immunization with radiation attenuated P. falciparum sporozoites. A panel of 19 pre-erythrocytic stage antigens was identified as strongly associated with sporozoite-induced protective immunity; 16 of these antigens were novel and 85% have been independently identified in sporozoite and/or liver stage proteomic or transcriptomic data sets. Reactivity to any individual antigen did not correlate with protection but there was a highly significant difference in the cumulative signal intensity between protected and not protected individuals. Functional annotation indicates that most of these signature proteins are involved in cell cycle/DNA processing and protein synthesis. In addition, 21 novel blood-stage specific antigens were identified. Our data provide the first evidence that sterile protective immunity against malaria is directed against a panel of novel P. falciparum antigens rather than one antigen in isolation. These results have important implications for vaccine development, suggesting that an efficacious malaria vaccine should be multivalent and targeted at a select panel of key antigens, many of which have not been previously characterized.

The Plasmodium spp. parasite causes significant global mortality and morbidity. Via the bite of an infected female Anopheline mosquito, sporozoites are inoculated into the human host and migrate to the liver, traversing through Kupffer cells and several hepatocytes before finally infecting a hepatocyte (1). After a period of liver stage development, during which there are no clinical symptoms of disease, merozoites are released from liver schizonts into the blood stream to invade erythrocytes. This initiates the blood stage of the parasite life cycle, which is responsible for the clinical manifestation of malaria.
No licensed malaria vaccine exists (2) and the development of an efficacious vaccine has been hindered by the complexity of the parasite and by our poor understanding of the antigenic targets of protective immunity. To date, only a very small fraction of the ϳ5300 proteins expressed during the multistage parasite life cycle has been evaluated as vaccine candidates (http://www.who.int/vaccine_research/links/Rainbow/ en/index.html). Candidate subunit vaccines based on a single or a few of these antigens have failed to induce optimal protection, or protection on genetically diverse backgrounds. Characterized P. falciparum (Pf) 1 sporozoite and liver stage antigens including the circumsporozoite protein (CSP), sporozoite surface protein (SSP2/TRAP), liver stage antigen 1 (LSA1), and exported protein 1 (EXP1) are recognized by CD4 ϩ and CD8 ϩ T cells and antibodies but are weakly reactive (3). The CSP is the core component of the RTS,S malaria vaccine currently in phase 3 clinical development; this vaccine is the most advanced malaria vaccine candidate but has conferred only short-lived sterile protection against infection after experimental sporozoite challenge in about 40% of malaria-naïve Caucasian subjects and delayed time to new infection and reduced episodes of severe malaria after field challenge in ϳ30 -50% of malaria-exposed African individuals (4). Prime-boost regimens (for example, adenovirus AdCh63 plus modified vaccinia virus Ankara) of recombinant virus vectors expressing sporozoite surface protein 2 (SSP2/TRAP) has been repeatedly protective but only in a small number of human volunteers (5,6). Most recently, a mixture of two Ad5 adenovectors expressing either CSP or apical membrane antigen 1 (AMA1) has protected some malaria-naïve adults against sporozoite challenge (C. Tamminga, unpublished). Each of these antigens, CSP, SSP2, and AMA1, has been identified as strongly immunoreactive in protein microarray studies (7). To date, clinical trials of blood stage malaria vaccine candidates, including AMA1 and merozoite surface protein 1 (MSP1), have been disappointing (8 -11). It is possible that the failure to develop an effective malaria vaccine might be because of the limited and arbitrary list of potential antigens thus far evaluated.
Immunization with radiation-attenuated Plasmodium spp. sporozoites can induce sterile protection against sporozoite challenge in rodent, primate, and human models (12)(13)(14), establishing that an effective malaria vaccine should be achievable. The irradiated sporozoite (IrrSpz) can invade hepatocytes but parasite development is arrested at the liver stage, before the blood stage of the life cycle (reviewed in (15)). This model suggests that antigens expressed by the liver stage parasite are effective vaccine targets, and that antigens associated with sterile infection-blocking immunity might be identified using specimens from IrrSpz immunized humans (reviewed in (16)).
Recently, technological advances have facilitated the largescale production of recombinant proteins and the generation of protein microarrays. These arrays can be applied to elucidate the profile of antibodies that develop after natural or experimental infection or after vaccination with attenuated organisms and to identify the immunoreactive antigens of interest for vaccine development or diagnostics (reviewed in (16)).
We have fabricated protein microarrays representing 23% of the P. falciparum genome and have screened these arrays with plasma from clinically divergent groups of individuals immunized with IrrSpz-infected mosquitoes to identify antigens strongly associated with sterile protective immunity.

EXPERIMENTAL PROCEDURES
Ethics Statement-The study protocol for clinical specimens used in this research was conducted in compliance with all applicable Federal Regulations governing protection of human subjects. The protocol was approved by the Naval Medical Research Institutional Review Board, the Office of the Special Assistant for Human Subject Protections at the Naval Bureau of Medicine and Surgery, and the Human Subjects Research Review Board of the Army Surgeon General. All study subjects gave written informed consent. The protein microarray studies were approved by the Naval Medical Research Center Institutional Review Board, the Queensland Institute of Medical Research Human Research Ethics Committee, and the University of California Irvine Institutional Review Board.
Open Reading Frame Selection-A subset of 1200 Pf proteins consisting of 22.6% of the entire genome and represented by 2320 whole or partial protein fragments (because open reading frames (ORFs) Ͼ3000 base pairs were cloned as overlapping segments) were evaluated. Putative proteins were derived from the Pf genomic sequence database (www.plasmodb.org) and selected based on stagespecific transcription or protein expression, subcellular localization, secondary protein structure, and documented immunogenicity in humans or animal models; this list included all putative Pf proteins with evidence of expression at some point during the parasite life cycle as indicated by multidimensional protein identification technology ( (17), www.plasmoDB.org) at the time of antigen selection (n ϭ 1049). Because of restrictions in producing long PCR products, proteins with exons longer than 3000 bp were divided into multiple overlapping sections, with 50 nucleotide overlaps. In total, the protein microarray comprised 2320 protein fragments. This array has been described previously (18).The life cycle stages of the 1200 Pf proteins as determined by mass spectrometry are depicted in supplementary Fig. S1.
PCR Amplification-PCR amplification of selected genes used Pf gDNA (3D7 strain) and custom PCR primers that included homologous cloning sites to the pXT7 plasmid (19). Primer and sequence information for the 2320 fragments used to fabricate the protein microarrays can be found on the web-interface http://contact14. ics.uci.edu/virus/mal_index.php. PCRs were carried out in a 50-l reaction volume containing 1-10 ng of Pf genomic DNA (gDNA) (3D7 strain), 0.04 U/l proofreading Taq polymerase (Triplemaster, Eppendorf), and 0.4 mM each dNTPs, with the following cycling conditions: 95°C for 3 min; 35 cycles of 95°C for 15 s, 40°C for 30 s and 50°C for 60 s/kb; and a final extension of 50°C for 10 min. For some proteins that proved difficult to amplify, 50 ng of gDNA was used. Products were visualized by agarose gel electrophoresis, and quantified by fluorometry (Picogreen; Molecular Probes, Eugene, OR).
Recombination Cloning-Plasmids were created from the PCR amplified fragments using in vitro recombination cloning and the pXT7 cloning vector, which encodes a N-terminal 10ϫ Histidine (His) and C-terminal Hemagglutinin (HA) tag (3.2 kB, KanR) (19). Briefly, 1 ng of BamHI digested linearized pXT7 template and custom primers 5Ј-CTACCCATACGATGTTCCGGATTAC and 5Ј-CTCGAGCATATGCT-TGTCGTCGTCG were used to generate a linear acceptor vector containing the target gene by PCR (50 l reaction) with 0.02 U/l Taq polymerase (Fisher Scientific), 0.1 mg/ml gelatin (Porcine, Bloom 300; Sigma), 0.2 mM each dNTPs. The following cycling conditions were used: 95°C for 5 min; 30 cycles of 95°C for 0.5 min, 50°C for 0.5 min and 72°C for 3.5 min; and a final extension of 72°C for 10 min. After purification (Qiagen, Valencia, CA), PCR products were visualized by gel electrophoresis, and quantified by fluorometry (Picogreen, Molecular Probes).
ORFs were cloned into the linearized pXT7 plasmid by a recombination reaction as previously described (19). Briefly, a 20 l mixture of linear vector and PCR-generated ORF fragment at a 1:1 molar ratio between vector and insert was transformed into DH5␣ competent cells without further purification and incubated for 1 h at 37°C, before dilution into an overnight culture of 3 ml LB broth containing Kanamycin 50 g/ml. Plasmids were isolated and purified using the QIAprep spin miniprep, (Qiagen, Valencia, CA) without further selection. A subset of these plasmids were sequence confirmed (supplementary Fig. S2).
Protein Expression and Detection-Escherichia coli in vitro cell-free transcription and translation reactions (rapid translation system (RTS) 100 E. coli HY kits; Roche, Indianapolis, IN) were carried out in 25 l volumes with a 5 h incubation at 30°C, according to manufacturer's instructions. For quality control purposes, relative protein expression efficiency for ϳ31% of all ORFs was assessed by immunodot blots by spotting 0.3 l of the RTS reaction on nitrocellulose and air drying before blocking in 5% nonfat milk powder in Tris-buffered saline (TBS) containing 0.05% Tween-20. Dot blots were stained with mouse anti-polyHIS mAb (clone HIS-1; Sigma) and rat anti-HA mAb (clone 3F10; Roche) and detected with alkaline phosphatase-conjugated goat anti-mouse IgG (HϩL) (Bio-Rad, Hercules, CA) or goat anti-rat IgG (HϩL) (Jackson ImmunoResearch, West Grove, PA) secondary antibodies respectively or with human hyperimmune plasma (diluted 1:1000 in blocking buffer with 10% E. coli lysate) followed by alkaline phosphatase-conjugated goat anti-human IgG secondary antibody (HϩL) (Jackson ImmunoResearch). Blots were visualized with nitroblue tetrazolium developer according to manufacturer's instructions (Thermo Fisher Scientific).
Microarray Chip Printing-To prepare the recombinant proteins for microarray printing, 15 l of RTS reaction was mixed with 10 l of 0.125% Tween 20/phosphate-buffered saline (PBS), and then 15 l volumes were transferred to 384-well plates. Plates were centrifuged (1600 ϫ g, 5 min) to pellet any precipitates and proteins in the supernatant were immediately printed without further purification onto 3-pad nitrocellulose coated FAST glass slides (Schleicher and Schuell, Keene, NH) using an OmniGrid 100 microarray printer (Genomic Solutions, Ann Arbor, MI). Arrays were allowed to dry and stored away from light at room temperature in a desiccator.
Array Controls-RTS reactions carried out in the absence of DNA plasmids were printed on each array as negative or nondifferentially recognized control spots. Purified human total IgG and Epstein-Barr nuclear antigen 1 (EBNA1) protein were also printed in serially diluted concentrations on each array, as probing and plasma controls respectively.
Serological Screening-Because high titers of anti-E. coli antibodies present in the human plasma could mask protein-specific reactivity in the arrays, plasma were pre-absorbed against E. coli lysate in protein array blocking buffer (Schleicher and Schuell) (1:100 dilution) for 30 min at room temperature (19). Slides were rehydrated and blocked in blocking buffer for 30 min at room temperature. Then 500 l plasma diluted 1:100 in blocking buffer was added to each pad and slides were incubated overnight at 4°C with a gentle constant speed on a platform rocker (Ratek, InVitro Technologies, Noble Park North, VIC, Australia). Serum antibodies were detected with biotin-conjugated goat anti-human IgG secondary antibody (1:1000 dilution, 1 h at room temperature with gentle constant speed, Fc fragment, Jackson ImmunoResearch) and visualized with a streptavidin P-3-conjugated antibody (1:200 dilution, 1 h at room temperature with gentle constant speed, Columbia Biosciences, Columbia, MD). Air dried slides were scanned on an Axon GenePix 4300A array scanner (Molecular Devices, Sunnyvale, CA) and fluorescence intensities quantified using the Axon GenePix Pro 7 software (Molecular Devices). Using the above software, all signal intensities were corrected for spot-specific background, where the background value for a spot was calculated from a region surrounding the spot.
Validation and Reproducibility of Array Data-To validate the Pf protein microarrays, antibody reactivity to three well-characterized malaria vaccine candidates, CSP, AMA1 and merozoite surface protein 2 (MSP2) were expressed via the RTS system or using traditional methods and were printed on the same protein microarray chip. There was a high degree of correlation between reactivity to the proteins produced by the two methods, (CSP (r ϭ 0.77; p Ͻ 0.001), AMA1 (r ϭ 0.78; p Ͻ 0.001), and MSP2 (r ϭ 0.96; p Ͻ 0.001)). To examine the reproducibility of the array probing, the reactivity against all HA spots from two independent chips was assessed and was highly correlated (r ϭ 0.92, p Ͻ 0.001). In addition, a smaller microarray chip was fabricated with a subset of 49 Pf proteins and probed with the same plasma used to probe the larger 2320 fragment Pf protein microarray. There was a strong correlation between plasma reactivity on the two microarrays (r ϭ 0.91, p Ͻ 0.001) (18).
Sporozoite-immunized Volunteers-Subjects were experimentally immunized with radiation-attenuated Pf (3D7) sporozoites and challenged with Pf-infected Anopheline mosquitoes (D. Freilich, unpublished) as described previously (14,20). Subjects were monitored daily post-challenge by thin blood smears to determine if they developed blood-stage malaria. A complete absence of blood-stage parasitemia during the 28 day follow-up was considered sterile protection. Six sporozoite-immunized volunteers were protected against sporozoite challenge and five were not protected (i.e. developed clinical malaria). One individual is represented in both groups because he was not protected in the initial challenge but was after a second series of immunizations. Plasma was collected from each individual before immunization (pre-immunization), post-third immunization, at the completion of the immunization series (fifth, sixth, or seventh immunization), immediately before challenge (pre-challenge) and after challenge (post-challenge). An infectivity control group (n ϭ 3) was simultaneously infected with the same Pf-infected mosquitoes used for challenge to demonstrate parasite infectivity; plasma was collected from these individuals at corresponding time points before (pre-challenge) and after (post-challenge) challenge. An additional group (n ϭ 5) were mock immunized by the bite of noninfected mosquitoes and plasma was collected at the time points corresponding to pre-immunization, post-third immunization, and post-last immunization time points of the IrrSpz immunized subjects. It should be emphasized that the specimens collected from volunteers experimentally immunized with radiation attenuated Pf sporozoites, or mock immunized controls, are a unique and valuable reagent and the sample sizes are limited by nature of the protocol. Plasma collected from volunteers with no known history of malaria exposure (n ϭ 10) was also evaluated.
Analysis of Array Data-To analyze low and high signal intensities (differential recognition) using standard statistical methods, the heteroskedastic nature of microarray platforms (21,22) and inherent variance-mean dependence in the data (23,24) needed to be considered and stabilized. Raw signal intensities (SI) were therefore variant log-transformed by using either asinh (Excel 2007, Microsoft; (23)) or variance stabilizing normalization (vsn) (Bioconductor software (www.bioconductor.org) (25)) transformation, to reduce the variance and experimental effects in the raw signal among differentially recognized antigens.
Before ranking and selection of antigens, data for each plasma sample was scaled by multiplication to a normalization ratio (average SI of all negative controls for that individual and average SI of all negative controls for all individuals) to apply the same baseline for all plasma. The post-immunization and pre-challenge time points for each individual in each group were combined for sample size purposes, because data showed no significant differences among those time points, and data was biologically and statistically analyzed. Area under receiver operating characteristic curves (ROC) (AUC) and Bayesian regularized t-tests (R statistical environment software, www.r-project.org) were used to identify differentially recognized an-tigens between the protected and not-protected groups (26,27) (Fig.  2). Note that the statistical testing (both AUC and p values) have not undergone multiple testing correction; rather these statistics were used as a method to compare the two groups and rank the data accordingly. The identification of antigens of interest was made on the combination of the rankings in addition to other relevant criteria i.e. recognition in different groups.
The criterion for positive immunoreactivity to a given protein for both transformation methods was determined as an average signal intensity two standard deviations above the negative control (all individuals). Data were visually presented as a heatmap where colors represent the SI range for that antigen. Red represents high SI (reactivity), black represents intermediate reactivity, and green represents low reactivity. Gene ontology (GO) annotation analysis was performed using the GOstats and org.Pf.plasmo.db R packages (Bioconductor software (www.bioconductor.org)). Significance of annotation enrichment was assessed using Fisher's exact test. Proper use of background distributions, i.e. using the antigens on the array rather than the whole genome, was used in enrichment statistic calculations.
Predicted functional protein interactions among antigens in the Top 20 list were analyzed using PlasmoMap (28). The PlasmoMap is a computationally constructed model of the Pf interactome in which protein-protein interactions are inferred by integrating in-silico and experimental functional genomics data using a Bayesian framework. Furthermore, the PlasmoMap assigns each predicted interaction a likelihood score. In our study, small functionally connected subgraphs of proteins in the Top 20 list were constructed by looking at all predicted interactions in the 3D7 strain with a likelihood threshold Ͼ3.3. On the basis of figures in Date and Stoeckert (28) this threshold corresponds to a false discovery rate of 0.25.

Antibody Profiles of Clinically Distinct Cohorts-Volunteers
experimentally immunized with radiation attenuated Pf sporozoites were clinically divergent after challenge with infectious Pf sporozoites; six individuals were sterilely protected and were classified as sporozoite-immune (IrrSpz protected) and five individuals developed blood stage parasitemia and were classified as sporozoite-exposed but non-immune (IrrSpz notprotected). To identify antigens associated with IrrSpz-induced protection, plasma collected from protected, not protected, infectivity, mock immunized, and naïve individuals at different stages of the immunization process (or corresponding time points for the non-immunized individuals) were probed on Pf microarrays against 2320 fragments representing 1200 Pf proteins. Antibody recognition of each fragment was assessed and depicted as a heatmap for each individual (Fig. 1A) and for each cohort (Fig. 1B). Antibody recognition for Epstein-Barr nuclear antigen 1 (EBNA-1), a methodological control for the arrays, was detected in all individuals and there were no apparent changes in the level of recognition through the time course (data not presented). There was minimal reactivity to the RTS reaction with no plasmid vector control (negative control) (data not presented).
There were distinct antibody profiles for each immunization group, and variability in responses among individuals within all groups. Overall, a markedly different pattern in antibody recognition was apparent between the protected and notprotected groups (Fig. 1), consistent with previous data (7).
There was no difference in antibody reactivity at pre-challenge and post-challenge time points for protected individuals who did not develop blood stage parasitemia or clinical disease (p Ͻ 0.6) (Fig. 1) and no change in the number of antigens recognized (340 pre-challenge versus 380 post-challenge). However, a marked change in seroreactivity was noted at pre-challenge versus post-challenge time points for the notprotected individuals where there was a significant increase in overall SI (p Ͻ 2e-91) (Fig. 1) and number of antigens recognized (292 pre-challenge versus 739 post-challenge), consistent with exposure to blood stage antigens on development of patent parasitemia.
The profile for the infectivity control individuals also showed a significant increase in SI pre-challenge versus post-challenge (p Ͻ 2e-30) to a subset of antigens (289 pre-challenge versus 546 post-challenge) (Fig. 1) and 89% (546) of the antigens recognized post-challenge were also recognized by the not protected group post-challenge, and therefore appear to be expressed by the blood stage parasite. Minimal reactivity to Pf antigens was noted with plasma from malaria naïve individuals or mock immunized individuals, consistent with a lack of exposure to Pf (Fig. 1).
Identification of Antigens Associated with IrrSpz Induced Protection-To identify Pf antigens putatively correlated with protection against sporozoite challenge, normalization and variance correction were performed using either the vsn and asinh methods. Because there were no significant changes in SI or number of antigens recognized at post-third immunization or post-last immunization (pre-challenge) time points for both protected and not-protected groups (p Ͻ 0.25 (notprotected), p Ͻ 0.2 (protected)), data from these biologically similar time points were combined for subsequent analyses. CyberT (a Bayesian regularized t test) p values and a ROC curve of the mean SI were used to measure the power of an antigen to discriminate between protected and not-protected cohorts, thus identifying antigens putatively associated with sporozoite-induced protection (Fig. 2B). An area under the ROC curve (AUC) value of 1.0 is a perfect prediction, 0.5 is a random chance, and Ͻ0.5 is a prediction in the wrong direction.
A subset of 20 fragments representing 19 proteins was common to both vsn and asinh approaches, designated the "Top 20" list (Table I, Fig. 3A). Three of these antigens (PFB0285c, PFE1085w, and PF08_0054) were not recognized by not-protected individuals and 10 of the antigens had a higher frequency of recognition in the protected cohort as compared with the not-protected cohort. Seven antigens (including CSP, SSP2/TRAP, and AMA1) were recognized at a similar frequency by the two cohorts but the protected individuals had a significantly higher magnitude of response (0.0034 Ͻ p Ͼ 0.0077, two-to four-fold fold higher average SI) (Fig. 3C).
The 10 antigens on the global list with the highest AUC values (0.86 -0.93) also had very low p values (protected versus not-protected; 0.0015-0.033). Unexpectedly, all of these antigens had an average SI below 3000 (protected), suggesting that high antibody recognition and reactivity (serodominance) to a single antigen does not correlate with sporozoite-induced protection (supplementary Fig. S3). Indeed, the average SI for 75 of the 86 antigens was below 5000, and 80 of the 86 were below 10000 (supplementary Table S1A).
Notably, protected individuals had a significantly greater cumulative antibody response to the Top 20 antigens as compared with not-protected individuals (p Ͻ 0.0055, Fig.  3B). This remarkable difference in cumulative response between the protected and not-protected individuals, suggests that sterile immunity is associated with a panel of antigens, not with individual antigens.
Genomic and Proteomic Features of Antigens Associated with IrrSpz Protection-To identify specific genomic or proteomic features of antigens putatively associated with IrrSpz-induced protection, selected characteristics of the global IrrSpz list and Top 20 list were compared with the Pf genome and to the known blood stage antigens (BSA) and sporozoite/ liver stage antigens (SLA) (supplementary Table S3). The antigens from the global IrrSpz list and Top 20 list average respectively three-and 2.5-fold larger for their transcription length, protein length and molecular weight as compared with the Pf genome, BSA, or SLA (Table II); this is consistent with the fact that large proteins potentially present a greater number of B-cell epitopes for antibody recognition. There was a moderate size difference between the proteins represented on the Pf protein microarray and the IrrSpz lists (two-fold and 1.7-fold respectively). There was no difference in average isoelectric points (pI) or the number of exons compared with the Pf genome, although the BSA and SLA had a lower average pI value than the other groups (Table II).
Using the "Protein Motif" search tool on PlasmoDB (www. PlasmoDB.org), we also determined the number of Plasmodium export element motifs (sequence R.L. [EQD]) in the P. falciparum genome, known sporozoite/liver stage antigens, known blood stage antigens, our global IrrSpz list, and our FIG. 2. Summary of selection criteria and AUC profile. A, Analyses were carried out to identify antigens that were putatively associated with sporozoite-induced protection. After spot quantification, the signal intensity was asinh/vsn log transformed to control variance, scaled by the negative control, post-immunization (post-I) and pre-challenge (pre-C) time points combined, and then analyzed according to defined statistical (Bayes-regularized t-tests, area under the receiver operating characteristics curve (AUC) analysis) and biological criteria. A total of 86 fragments (77 Pf proteins) were identified by either approach (global irradiated sporozoite list) and 20 were in common to both approaches (Top 20). B, AUC values of all antigens for the not-protected and protected cohorts were determined by R statistical environment software (www.r-project.org). Antigen rank is plotted relative to their AUC value. An AUC value approaching 1.0 suggests a very strong association of an antigen to protection induced by irradiated sporozoite immunization; an AUC value of 0.5 indicates pure chance. An AUC value of 0.7 was chosen as a threshold for positivity.

Characteristics of the Top 20 antigens highly associated with sporozoite-induced protection
Twenty fragments (representing 19 proteins) are common to both analysis approaches (asinh and vsn). Antigens are ranked by AUC value. Genomic and proteomic features of these fragments (www.plasmodb.org) are detailed with array data for protected and not-protected cohorts: average signal intensity, fold change, and frequency of recognition for protected and not-protected cohorts, fragment details, gene details (exons, size (bp)), protein details (size (aa), molecular weight (Mw), isoelectric point (pI)), and other proteomic features (presence of transmembrane domains, PEXEL motifs, exported protein prediction). Also indicated is whether or not the protein has been reported in any of eight independent sporozoite-specific proteomic and genomic data sets (see supplementary Table S1 for details). Antigens with multiple fragments present on this list ( §) and antigens with other fragments from the same gene present on the global irradiated sporozoite list (supplementary   in the transport of liver stage parasite proteins to hepatocytes. Although signal peptides have been identified in all known SLA and 11 of 17 known BSA, consistent with a role for secretion in inducing antibody responses, antigens containing a signal peptide were not enriched in our Top 20 list (35%, 7/20, p value ϭ 0.08) or global list (24%, 21/86, p value ϭ 0.31). These data are consistent with the concept that IrrSpz protection is mediated by T cells rather than antibodies. Also consistent with this is the observation that only 30% (p value ϭ 0.64) and 34% (p value ϭ 0.9) of Top 20 or Global IrrSpz list antigens, respectively, had defined transmembrane domains, which are associated with exposure of surfaces for antibody recognition (Table II). Identified proteins were assigned functional categories based on annotation information available from PlasmoDB (www.PlasmoDB.org). Fifty-six percent of the global list (supplementary Fig. S4) and 40% of the Top 20 list were hypothetical proteins (Fig. 4A), compared with 56% of proteins present in the Pf protein microarray (Fig. 4B); with the remainder assigned to multiple functional categories (Fig. 4). Interestingly, the functional categories for the Top 20 showed a notable concentration of genes and proteins involved in or regulating DNA processing and cell cycle (20%) and protein synthesis (20%); cellular communication and transport facilitation were not represented in this subset (Fig. 4A). Gene ontology analysis also indicates an overrepresentation of biological processes involving DNA, invasion, and metabolic processes (supplementary Table S4A), DNA related activity (supplementary Table S4B) and perhaps a localization of these proteins for invasion purposes (supplementary Table S4C).
Identification of Novel Blood Stage Specific Pf Antigens-A strong response to all well-characterized blood stage antigens was seen post-challenge in the not-protected cohort. As expected, negligible responses to antigens known to be expressed at the blood stage were observed in the protected cohort and no changes in their heatmap profile occurred throughout the study time course (supplementary Fig. S5C).
Analysis of the responses post-challenge for the not-protected cohort allowed us to identify novel blood stage specific Pf antigens, using the following criteria: (1) antigenic; i.e. average SI greater than two SDs above negative control in at least two not-protected individuals; (2) statistically significant recognition (P) after challenge as compared with pre-challenge Ͻ 0.05 (not-protected); and (3) not recognized by protected subjects. A total of 23 antigens met these criteria, of which two have been previously characterized (MSP1 (PFI1475w) and liver stage antigen 3 (LSA3, PFB0915w)) and 21 are novel. For 22 of the 23 antigens, evidence of their expression at the blood stage has been independently reported from data sets of Pf infected red blood cell MS/MS (36), Pf merozoite MS/MS (17) (Leiden Malaria Group, unpublished (PlasmoDB)), P. yoelii 17XNL asexual blood stage EST data sets (Pf orthologs) (37) and previous Pf protein microarrays, which analyzed plasma from individuals with blood stage malaria (7, 18) (Fig. 5). Four antigens expressed in both the blood stage and the sporozoite/liver stage of the parasite life cycle were also identified (Fig. 5). DISCUSSION Using Pf protein microarrays, we aimed to identify from genomic sequence data a subset of antigens associated with sterile protective immunity against malaria. Plasma from clinically defined individuals either protected or not protected against Pf sporozoite challenge was screened against a protein microarray consisting of 2320 fragments representing 1200 proteins or ϳ23% of the Pf proteome. We identified 86 fragments representing 77 Pf proteins putatively associated with sporozoite-induced protection ("global IrrSpz list"; supplementary Table S1). Of particular interest are 19 proteins that form a potential signature associated with sporozoiteinduced protection (Top 20 list; Table I and supplementary Table S2). Three proteins on this list (CSP, SSP2/ TRAP, and AMA1) have been already assessed as vaccine candidates and shown to be partially protective in animal models and in humans; the other 16 proteins have not yet been evaluated. In addition, 23 antigens specific to the blood stage (21 novel) as well as four antigens expressed in both the blood stage and the sporozoite/liver stage were identified. The

Novel Antigens Associated with Sterile Immunity
FIG. 5. Antigens putatively associated with blood stage malaria. Twenty-one novel antigens and two current vaccine candidates (MSP1 and LSA3) were recognized by individuals who developed blood stage infection postsporozoite challenge. A, Average signal intensity for protected or not-protected individuals. Group (1) putative blood stage specific antigens; Group (2) antigens highly expressed at blood stage that might also be expressed in the liver stage. B, Presence in independent blood stage specific proteomic or transcriptomic data sets: blood stage mass spectrometry, published (4, 7) and unpublished (Leiden Malaria group, www.PlasmoDB.org); blood stage EST (13); and experimentally infected or naturally exposed protein microarray data sets (3,14). and/or liver stage of the Pf parasite lifecycle by multidimensional protein identification technology (17), MS/MS (29,30), or EST expression analysis (31)(32)(33)(34)(35), validating our identification as putative sporozoite and/or liver stage antigens (supplementary Table S1). Likewise, almost all of the bloodstage specific antigens (91%, 21/23 proteins) have been previously identified in MS ( (17,36), Leiden group, unpublished), transcript (37) or protein array data sets (7,18).
Some of our global IrrSpz or Top 20 proteins have been previously identified in other protein microarray studies. In the proof-of-concept study for protein microarrays, using a microarray fabricated with 250 Pf proteins, 72 serodominant antigens were shown to be recognized by naturally exposed or experimentally immunized (IrrSpz) individuals (7). Forty-two of those 72 antigens were present in the Pf microarray fabricated for this study and four were identified in our Top 20 list (PF11_344/AMA1, PFE0060w, PFC0210c/CSP, and PF13_ 0201/SSP2) and two others on our global IrrSpz list (PF11_0226, PF10_0350) (supplementary Table S1B). Another study analyzed the antibody reactivity in children and adults naturally exposed to malaria in Mali using the same version of Pf protein microarrays as this study and identified 49 antigens associated with protection from uncomplicated malaria (18). Four of these are present in our Top 20 list (PFI0925w, PFL1620w, PFE0060w, and MAL13P1.22) and six others in the global IrrSpz list (PF13_0190, PF11_0008, PF14_0419, PF13_0179, PF13_0350, and MAL7P1.138) (supplementary Table S1B). Four of the 23 antigens identified in our study as putative blood-stage specific (PFI0855w, PF11_0270 (18), PFI1475w, PFB0915w (7)) were identified in the Mali study and a previous Pf protein microarray study (Fig. 5).
Our data have important implications for malaria vaccine development. First, these studies have identified a panel of 19 antigens putatively associated with sterile protection against sporozoite challenge and suggest that immune responses against any one of these antigens in isolation are insufficient to protect. Second, data show that the magnitude of the immune response against these key antigens does not correlate with protection status (supplementary Fig. S3), consistent with the concept that IrrSpz-induced protection is mediated by T cells (38 -41) and that antigens are recognized by T cells are also recognized by antibodies. Antibody responses to sporozoite proteins might also have a role in preventing sporozoite invasion of the hepatocyte (42,43).
It is now generally accepted that an effective malaria vaccine will likely need to target multiple antigens (multivalent) probably expressed at more than one stage of the parasite life cycle (multistage) (44,45). In the murine model, immunization with a combination of plasmid DNA vaccines encoding two pre-erythrocytic stage antigens was able to circumvent the genetic restriction of protection with either vaccine alone and could protect mice that were not protected by the individual vaccines (46). Genetic restriction of host immune responses has been also documented in humans (47), highlighting the need to induce protective immune responses in diverse genetic backgrounds. Indeed, Saul et al. suggested that a multiantigen vaccine targeting a single organism increased the efficacy of a vaccine by reducing the number of poor responders (48). More recently, in the field, the breath (number of targets recognized), magnitude (antibody levels) and combination of characterized malaria antigens (AMA1, MSP2, and MSP3) were shown to be necessary components for protective efficacy in Kenyan children (49). In other studies, high levels of antibodies to multiple pre-erythrocytic stage antigens (CSP, LSA1, and SSP2) were more strongly associated with protection than antibodies to a single antigen (50), and associated with a lower risk of developing clinical malaria in Kenyan children (51).
Genome-based annotations suggest that our Top 20 antigens fall into seven functional groups, excluding proteins with "unknown or hypothetical" function. Unexpectedly, the two major functional groups are cell cycle (PF14_0051, PF08_0054, PFE1085w, and MAL13P1.22) and protein synthesis (PFI0925w, PFL1620w, PF10_0211, and PF13_0222) (Fig. 4B, Table I), perhaps reflecting the need for expression and production of proteins that facilitate parasite survival at the liver stage and drive development to the erythrocytic stage in the relatively short time frame of 5-10 days. Indeed, gene ontology analysis strongly suggests that the Top 20 proteins are involved in DNA processes, particularly to damage and stress responses, and might also play a role in invasion (supplementary Table S4). In contrast, transcription, cell cycle (DNA processing), metabolism, protein synthesis, cellular communication, and transport facilitation are more evenly represented in the global IrrSpz list (supplementary Fig. S4).
Six antigens putatively associated with sporozoite-induced protection are implicated in the response to DNA damage and/or directly repair damaged DNA (PFE1085w (52), PF08_0034 (53), MAL13P1.22 (54), PF14_0051 (55,56), PF11_0404, and PFI0925w). The temperature change during irradiation or during the transmission from vector to host (25°C to 37°C) and the host immune response itself, would elicit a stress response in the parasites. Consistent with this, one of the antigens on our list, HSP70 (PF08_0054), is known to be an abundant "stress protein" and acts as a molecular chaperone of immunogenic peptides to antigen presenting cells for induction of T-cell responses (57). Antibody and T-cell recognition of HSP70 has been observed in malaria exposed individuals (58) but passive transfer of anti-HSP70 mAbs from mice infected with P. berghei sporozoites could not protect against sporozoite challenge (59).
Plasmodium parasite infection from the skin to the liver leaves a trail of damaged Kupffer cells and hepatocytes (60,61) and exposes the parasite to stress induced by exposure to reactive oxygen that causes structural damage to DNA, disrupting nuclear and cytoplasmic processes and triggering apoptosis (reviewed in (62)). Elevated ⅐OH levels in P. yoelii-infected mice caused mitochondrial oxidative stress and hepatocyte apoptosis (63) and it is known that in vivo ⅐OH can directly damage nuclear DNA (64). It is also known that the glutathione (GSH) pathway via the thioredoxin redox system acts as a Plasmodium defense against oxidative damage such as ROS from the host immune system (65). Notably, gammaglutamylcysteine synthase (␥GCS, PFI0925w), an antigen on our Top 20 list, is the rate-limiting enzyme in the synthesis of GSH. The disruption of this biosynthetic pathway caused a minor decrease in parasite growth at the blood stage (possibly because of availability of host GSH) but had severe effects for the parasite at the mosquito stage with arrested development of oocysts and no sporozoite production (66). Pf ␥GCS is expressed in P. berghei sporozoites (31) and host ␥GCS is highly induced in P. yoelii-infected livers (67).
Intriguingly, PF11_0404, a putative transcription factor with AP2 domains identified in our Top 20 list, might have a pivotal role in the stress response to hepatocyte invasion by transcriptionally regulating sporozoite proteins (68,69). In fact, PF11_0404 is predicted to have a direct functional interaction with four other proteins on the Top 20 list (Fig. 6, PlasmoDB, http://www.cbil.upenn.edu/plasmoMAP/index-v1.html, (28)) and AMA1 is predicted to have a PF11_0404 binding site (70). Hence, PF11_0404 could be a key transcription factor for liver stage development. Pf ApiAP2 transcription factor has no human homolog, making this protein a strong vaccine candidate (71).
The mechanism behind repair and survival or cell death is not well understood but this mechanism might be critical for the survival of the parasite in the liver. Furthermore, the release of parasite nuclear material could be an important source of pathogen-specific T-cell targets. Products of translational errors, such as mistakes in RNA synthesis, splicing, truncation, amino acid substitution, and post-translational abnormalities (errant covalent modification, protein-protein interactions) are collectively known as "defective ribosomal products" or DRiPs (reviewed in (72)). DRiPs are rapidly moved to the proteosome for degradation, and a subset of these degradation products become associated to MHC class I molecules in the endoplasmic reticulum (ER) (73). The fast turnover of DRiP products could be an effective mechanism for the host to eradicate intracellular pathogens before they have time to replicate (72). Failure of DNA repair proteins to mend or control the DNA damage at the liver stage potentially provides a source of parasite DRiP products to antigen presenting cells for presentation to CD8 ϩ T cells. Presentation of MHC class I molecules to CD8 ϩ T cells and consequent IFN-␥ secretion are critical mediators of infective, radiation, and genetically attenuated parasite-induced protection (38, 39, 41, 74 -80).
Several proteins in the Top 20 list are involved in infectivity, including current vaccine candidates, AMA1, CSP, and SSP2. Protein kinases are also known to be involved in sporozoite infectivity (reviewed in (81)); the conditional disruption of P. berghei cGMP-dependent protein kinase in sporozoites arrested development at the late liver stage and conferred protection against wild-type sporozoites (82). In addition, P. berghei calcium dependent PK-6 mutants were less infective for hepatocytes and had a delayed onset of blood stage malaria (83). Therefore, the function of PFB150c in parasite infectivity at the liver stage might be worth exploring. Another antigen in our Top 20 list, Rhoptry neck protein 3 (RON3, PFL2505c), is expressed in sporozoites (17) and this family of proteins is reported to be important for parasite invasion and the formation of the parasitophorous vacuole (84 -86). A complex of RON2, RON4, and AMA1 localizes to the moving junction formed at invasion (87,88). AMA1 is a critical sporozoite protein for hepatocyte invasion (89) and is also in our Top 20 list. Although RON2 is expressed in sporozoites (34,90), in our study it was not differentially recognized by protected versus not-protected individuals; RON4 was not present on our Pf protein microarray. The potential role of RON3 in sporozoite invasion should be evaluated. In addition, it should be noted that antibodies to Plasmodium rhoptry proteins might inhibit merozoite invasion (91)(92)(93).
Parasite-encoded protein on the surface of infected erythrocytes 2 (PIESP2, PFE0060w), also identified in our putative IrrSpz signature, might have a role in parasite pathogenesis because this family of proteins facilitates the transport of proteins in and out of the parasite and enables the expression of variant proteins and endothelial adherence proteins to evade detection (94). This protein has been identified in previous studies as highly reactive (7) and associated with protection from uncomplicated malaria in children (18). Although previous data indicated this is a blood stage protein (multidimensional protein identification technology analysis of erythrocyte ghosts (36)), our data suggests that PIESP2 (PFE0060w) is also expressed in the liver stage and associated with sterile immunity directed at the pre-erythrocytic stage. The export of this parasite protein to the host makes it a promising vaccine candidate.
In summary, in this study, we have identified a panel of antigens that are associated with the sterile protection induced in humans by immunization with IrrSpz. The literature provides compelling support that this list of antigens contains potentially interesting targets worthy of further investigation. All but one of these proteins have P. vivax or rodent/simian orthologs, which would facilitate validation in rodent or simian FIG. 6. Network interactions of PF11_0404 to other Top 20 antigens. Map of predicted protein interactions of PF11_0404 and associated score threshold for each interaction link is shown (PlasmoDB, http://www.cbil.upenn.edu/plasmoMAP/index-v1.html, (62)). models or cross-species vaccine development (supplementary Table S2). Our data also suggest that immune responses directed against a single or a few Pf antigens will be insufficient to confer protection; rather, protection is likely because of the recognition of an appropriate magnitude or threshold of response against a select panel of signature antigens, eliciting a robust, broad, and protective response. Our data further suggest that serodominance does not correlate with sporozoite-induced immunity and that antigens identified historically on the basis of immunodominant antibody recognition cannot be good targets for a malaria vaccine.
Acknowledgments-We thank the members of the Felgner laboratory and Antigen Discovery Inc. for assistance with this work. We also thank Lolita Bebris and Mara Berzins for the collection of irradiated sporozoite specimens. We are particularly indebted to the volunteers, without whom this work could not have been done. DAF and TLR were active duty military personnel at the time they contributed to this work. The work of these individuals was prepared as part of official government duties. Title 17 U.S.C. §105 provides that "Copyright protection under this title is not available for any work of the United States Government.