Protection against malaria at 1 year and immune correlates following PfSPZ vaccination

Fifty-five percent of individuals vaccinated with an attenuated Plasmodium falciparum sporozoite vaccine remained without parasitemia after controlled human malaria infection one year later; immune correlate analysis in humans and non-human primates suggest a role for liver-resident T cells. An attenuated Plasmodium falciparum (Pf) sporozoite (SPZ) vaccine, PfSPZ Vaccine, is highly protective against controlled human malaria infection (CHMI) 3 weeks after immunization, but the durability of protection is unknown. We assessed how vaccine dosage, regimen, and route of administration affected durable protection in malaria-naive adults. After four intravenous immunizations with 2.7 × 105 PfSPZ, 6/11 (55%) vaccinated subjects remained without parasitemia following CHMI 21 weeks after immunization. Five non-parasitemic subjects from this dosage group underwent repeat CHMI at 59 weeks, and none developed parasitemia. Although Pf-specific serum antibody levels correlated with protection up to 21–25 weeks after immunization, antibody levels waned substantially by 59 weeks. Pf-specific T cell responses also declined in blood by 59 weeks. To determine whether T cell responses in blood reflected responses in liver, we vaccinated nonhuman primates with PfSPZ Vaccine. Pf-specific interferon-γ-producing CD8 T cells were present at ∼100-fold higher frequencies in liver than in blood. Our findings suggest that PfSPZ Vaccine conferred durable protection to malaria through long-lived tissue-resident T cells and that administration of higher doses may further enhance protection.

In 2015 there were an estimated 214 million clinical cases and 438,000 deaths due to malaria 1 , primarily caused by Pf in children in sub-Saharan Africa. A highly effective vaccine is urgently needed to prevent malaria in individuals and to facilitate elimination of malaria from defined geographic areas. To achieve these goals, we established an interim target of >85% sterile protection against Pf infection for >6 months 2 .
There is currently no malaria subunit vaccine that approaches this level of protection. The most extensively studied candidate malaria vaccine, RTS,S (a subunit vaccine based on the Pf circumsporozoite protein (PfCSP)), confers sterilizing protection against controlled human malaria infection (CHMI) in about 22% of healthy malarianaive adults 5 months after vaccination 3 . In a phase 3 field study, the efficacy of RTS,S against clinical malaria was 26% and 36% in young infants and children between the ages of 5 and 17 months, respectively, through 38-48 months of follow-up following a fourdose regimen on a 0-, 1-, 2-, and 20-month schedule 4 . Therefore, it is necessary to investigate alternative vaccination strategies that confer long-lived sterilizing protection 5,6 .
Sustained sterilizing immunity against the pre-erythrocytic stages of Pf has been observed in humans immunized by wholeparasite approaches using mosquitoes for vaccination 7,8 . In a study of malaria-naive adults, 5/6 subjects exposed to >1000 irradiated mosquitoes carrying attenuated PfSPZ were protected when CHMI occurred 23-42 weeks after immunization 8 . To advance from using mosquitoes for inoculation of attenuated PfSPZ toward a clinical product, we previously reported that immunization by intravenous (i.v.) injection of radiation-attenuated, aseptic, purified, cryopreserved PfSPZ, a product called Sanaria PfSPZ Vaccine 9 (hereafter referred to as PfSPZ Vaccine), was well tolerated and immunogenic (the VRC 312 study) 10,11 . PfSPZ Vaccine induced a dose-dependent increase in PfSPZ-specific antibodies and frequencies of multifunctional T H 1 cytokine-producing CD4 T cells and γδ T cells in the blood. For vaccine recipients that underwent CHMI 3 weeks after final immunization, Pf parasitemia was observed in 3/9 and 0/6 subjects who received four or five doses of 1.35 × 10 5 PfSPZ, respectively, whereas parasitemia was observed in 5/6 unvaccinated controls, demonstrating that PfSPZ Vaccine confers high-level, short-term protection. The next critical milestones for PfSPZ Vaccine were to assess the durability of vaccine efficacy and to investigate the immune correlates and mechanisms of protection.

PfSPZ Vaccine efficacy at 21 weeks
To assess the durability of protection, subjects from the VRC 312 study were re-enrolled for repeat CHMI with the homologous Pf clone 3D7. Because vaccine efficacy (VE) is assessed at multiple time points in the same vaccinated subjects, we defined vaccine efficacy as first VE (VE at first CHMI), subgroup VE (VE among the subgroup of subjects who were not parasitemic after the first CHMI and returned for repeat CHMI), and cumulative VE (first VE × subgroup VE). VE is calculated as '1 − relative risk' , where relative risk is the ratio of the infection rate among the vaccinated subjects divided by the infection rate among the controls. Thus, VE is always adjusted to account for those cases in which the infection rate in the controls is not 100%.
Six subjects who had received four or five doses of 1.35 × 10 5 PfSPZ by i.v. injection and had not developed parasitemia following CHMI at 3 weeks 14 underwent repeat CHMI 21 weeks after the final vaccination. Four of six subjects developed parasitemia, as compared to 6/6 unvaccinated control subjects (Supplementary Fig. 1a,b), for a subgroup VE of 33% (P = 0.23). For this dosage group, the first VE was 76% (ref. 11), and therefore the cumulative VE was 25%. Thus, 1.35 × 10 5 PfSPZ administered four or five times did not confer adequate protection at 21 weeks.
Additionally, eight subjects in the VRC 312 study who were parasitemic at a prior CHMI (six immunized subjects and two unvaccinated controls) underwent repeat CHMI, and all of them developed parasitemia (Supplementary Fig. 1c,d). Thus, a single previous episode of Pf parasitemia followed by drug treatment did not confer protection to a subsequent CHMI. Therefore, only vaccinated subjects who did not develop parasitemia following their first CHMI underwent repeat CHMI in the following studies.

Study design
We assessed in a new study (VRC 314) whether increasing the dosage of PfSPZ Vaccine from 1.35 × 10 5 to 2.7 × 10 5 PfSPZ per dose affected VE. With this higher dose, we tested three-dose (group 1) and fourdose (groups 4 and 5) regimens. A fourth group of subjects received 1.35 × 10 5 PfSPZ four times, followed by a fifth dose of 4.5 × 10 5 PfSPZ (group 3), to determine whether a higher final dose improved VE, as compared to that for subjects who received five doses of 1.35 × 10 5 PfSPZ (Fig. 1a). For the subjects in each of these groups, PfSPZ Vaccine was administered by rapid i.v. injection.
The trial also assessed route of administration. Studies in nonhuman primates (NHPs) and humans show that i.v. administration of PfSPZ Vaccine is substantially more immunogenic and protective than subcutaneous (s.c.) or intradermal (i.d.) administration at a dose of 1.35 × 10 5 PfSPZ 10,11 . However, studies in rodents show that intramuscular (i.m.) administration of radiation-attenuated SPZs is more protective than s.c. or i.d. administration, although it is considerably less protective than by i.v. administration 12,13 . Because there may be instances in which administration by i.v. injection is logistically complex, we compared the protective efficacy of administering 2.7 × 10 5 PfSPZ by i.v. injection versus 2.2 × 10 6 PfSPZ by the i.m. route (an 8.1-fold higher dose) with the same four-dose regimen (group 2) (Fig. 1a).
For the primary efficacy analysis, the results are presented for the first CHMI for each vaccine group as compared to those for the controls. To account for multiple comparisons, P < 0.01 was taken as evidence of an effect, and P values between 0.01-0.05 were considered to be suggestive of an effect. This threshold for significance of the primary analysis was specified in the protocol to balance the potential for type 1 error with the conservative Bonferroni approach (Online Methods). Secondary analyses compared the results from the different groups, as well as describe the results of repeated challenges among those subjects who remained parasite free in each group. No formal multiplecomparison adjustment was used for secondary end points.
Daily monitoring by PCR analysis, rather than thick blood smear, was used for diagnosis of parasitemia and treatment. In subjects of the VRC 312 cohort, PCR analysis allowed detection of parasitemia 1-2 d earlier than by blood smear (Supplementary Table 1), which enabled clinical monitoring criteria to be changed from overnight stays after day 7 to once-daily outpatient visits.

Adverse events
Vaccinations were well tolerated. Of 57 vaccine recipients, 41 (72%) had no solicited adverse events at the injection site after any vaccination, 15 (26%) had mild symptoms, and one (2%) had moderate symptoms (Supplementary Table 4). There were no solicited systemic symptoms for 32 (56%) vaccine recipients, mild ones for 19 (33%) recipients, and moderate ones for six (11%) recipients (Supplementary Table 5). There were no serious adverse events attributed to vaccination. Alanine aminotransferase (ALT) levels measured 14 d after each vaccination were not elevated in any dosage group, as compared to those for the same subjects before starting the vaccinations (Supplementary Fig. 3).
VE at 21-25 weeks, by i.v. administration. Only vaccine recipients who did not develop parasitemia after their first CHMI underwent repeat CHMI. Seven subjects from group 3 (four doses of 1.35 × 10 5 PfSPZ and a fifth dose of 4.5 × 10 5 PfSPZ) underwent repeat CHMI at 25 weeks after final vaccination, and 3/7, as compared to 6/6 controls, developed parasitemia. Subgroup VE was 57%, and cumulative VE was 35% (Fig. 1d). These results were similar to the results from the VRC 312 study for CHMI that was performed at 21 weeks (Supplementary Fig. 1). Thus, increasing the final dose by 3.3-fold did not improve short-or long-term protection, as compared to that with the five-dose regimen of 1.35 × 10 5 PfSPZ 11 . Three subjects in group 1 (three doses of 2.7 × 10 5 PfSPZ) underwent repeat CHMI at 25 weeks, and 1/3 developed parasitemia (Fig. 1d); subgroup VE was estimated at 67%, and cumulative VE was 16%.

Immunogenicity
Antibody responses. Antibodies induced by vaccination with attenuated SPZs can limit parasites from infecting hepatocytes in animals [14][15][16] . Therefore, antibodies to PfCSP and whole PfSPZ were assessed by   npg enzyme-linked immunosorbent assays (ELISAs) and automated immunofluorescence assays (aIFAs), respectively, 2 weeks after the final vaccination. There were no differences in antibody levels among the groups that received the different i.v. regimens. However, subjects immunized by the i.m. route had lower antibody responses than did subjects immunized by i.v. injection ( Fig. 2a,b). Cellular responses. Cellular immunity is critical for protective efficacy by vaccination with live-attenuated SPZs in rodent and NHP models 10,[17][18][19] . Although interferon (IFN)-γ-producing CD8 T cells are necessary and sufficient for protection in the majority of mouse and NHP studies 20 , NK, γδ, and CD4 T cells also influence protection 19 Table 6). There is considerable overlap in the proteomes of PfSPZ and PfRBC. Although there was an increase in PfSPZ-specific CD4 T cell cytokine responses after the final immunization in subjects from groups 1, 2, 4, and 5, there were no differences between the vaccine groups (including the i.m. group) in the magnitude (Fig. 2c) or quality of PfSPZ-specific CD4 cytokine responses ( Supplementary  Fig. 6). The PfRBC-specific CD8 T cells in blood 2 weeks after the final immunization were no different from those observed before administration of the vaccine (background) in most vaccine groups (Fig. 2d). Finally, there was a vaccine-induced increase in the frequency of total γδ T cells, but no differences between the groups ( Fig. 2e,f). The expansion was restricted to the Vγ9 + Vδ2 + family (hereafter referred to as Vδ2 + ), which comprises ~75% of γδ T cells in blood (Supplementary Fig. 7a-f). A higher percentage of γδ T cells expressed the activation markers CD38, the cytotoxic molecule perforin, and the liver-homing chemokine receptor CXCR6 (ref. 22) after vaccination (Supplementary Fig. 7g-l). Of note, γδ T cells from PBMCs analyzed before vaccination expressed IFN-γ when stimulated ex vivo with live-attenuated PfSPZ (Supplementary Fig. 7m).

Immune correlates
To identify potential immune correlates of protection 23 , we assessed whether there were associations between the immune responses measured 2 weeks after the last immunization and the outcome after each CHMI (parasitemia versus no parasitemia).
Antibody correlates. PfCSP-and PfSPZ-specific antibody levels correlated with outcome for CHMIs that were done 3 weeks and 21-25 weeks after the final immunization ( Fig. 3a,b). For this analysis, we assumed that subjects who developed parasitemia after the 3-week CHMI would also develop parasitemia after the 21-to 25-week CHMI, on the basis of our earlier findings (Supplementary Fig. 1). Cellular correlates. The percentage of Pf-specific CD4 and CD8 T cells in the blood that produced IFN-γ, IL-2 and/or TNF-α did not correlate with outcome at the 3-week or the 21-to 25-week CHMI (Supplementary Fig. 8). However, the absolute frequency of unstimulated Vδ2 + T cells as a percentage of total lymphocytes correlated with outcome at the 21-to 25-week CHMI (Fig. 3c). Notably, the frequency For c-f, P values were corrected by the Bonferroni method. *P < 0.05, **P < 0.01, ***P < 0.001. Dagger ( †) indicates to see Figure 1 for doses.
npg of the Vδ2 + T cell subset measured before the first vaccination also correlated with outcome at the 3-week CHMI and the 21-to 25-week CHMI (Fig. 3d).

Role of PfSPZ antibodies in protection
The finding that PfCSP-and PfSPZ-specific antibody levels 2 weeks after the final immunization correlated with outcome of the 3-week and 21-to 25-week CHMIs suggested that antibodies were a biomarker of a successful vaccine response; however, antibodies could also have a functional role in mediating protection [14][15][16] . To investigate this we assessed the levels of antibodies at the times of the 21-to 25-week CHMIs and the 59-week CHMIs. At 59 weeks, PfCSP-specific antibody levels in the five subjects who did not develop parasitemia were 8-fold lower (PfCSP geometric mean (GM) = 1,640) than the responses among subjects who did not develop parasitemia at 3 weeks (PfCSP GM = 13,200; P = 0.0051; Fig. 4a). Of note, the levels of antibodies to PfCSP in the five subjects at 59 weeks (PfCSP GM = 1,640) were no different than those in subjects immunized with the same regimen and who developed parasitemia following CHMI at 21-24 weeks (PfCSP GM = 1,860; P = 0.87; Fig. 4a). Assessment of antibody responses to whole PfSPZ (Fig. 4b) also revealed a decline in Pf-specific antibody levels over time.
To assess the in vivo functional activity of Pf-specific antibodies, we used Fah -/-Rag2 -/-Il2rg -/-(FRG) mice (which are deficient in T cells, B cell, and NK cells) reconstituted with human hepatocytes (which we refer to as FRG-huHep mice), which are capable of supporting liver-stage development of Pf (ref. 24). Total IgG from the five vaccine recipients that did not develop parasitemia following CHMI at 59 weeks was purified from plasma and serum. The IgG was passively transferred to FRG-huHep mice, and each mouse was challenged by exposure to 50 mosquitoes infected with luciferase-expressing PfSPZ. Liver-stage burden was quantified by whole-animal imaging.
The negative control consisted of pooled IgG obtained from the blood of the five subjects before vaccination. Passive transfer of 150 µg of the PfCSP-specific monoclonal antibody 3C1, which was used as the positive control, reduced liver-stage burden by ~90% as compared to that by transfer of the negative control. Purified IgG from the five vaccinated subjects taken 2-3 weeks after vaccination reduced liver-stage burden by 88% (median), and IgG taken at 59 weeks after the final immunization reduced liver-stage burden by 65% (median) (Fig. 4c). There was higher inhibitory activity in IgG taken at 2-3 weeks from 4/5 subjects as compared to that taken at 59 weeks, but the difference did not reach the level of statistical significance (P > 0.05 by Wilcoxon signed-rank test). However, liver-stage burden did correlate inversely with anti-PfCSP levels (Fig. 4d).
PfSPZ-specific T cell activation is dependent on the timing of CHMI To provide additional insight into the potential mechanisms of protection, we assessed how CHMI affected PfSPZ-specific T cell responses in vivo after each CHMI. For this analysis, we identified PfSPZ-specific T cells by the expression of IFN-γ, interleukin (IL)-2, and tumor necrosis factor (TNF)-α that was induced by ex vivo stimulation with PfSPZ antigens, and we simultaneously measured the expression of Ki-67 (a marker used to detect recent lymphocyte cell division 25 ;  Fig. 5a). One week after the subjects in the VRC 312 (Fig. 5b) and VRC 314 (Fig. 5e) clinical trials received their final immunizations, ~20-30% of PfSPZ-specific CD4 T cells were Ki-67 + , demonstrating that vaccination induced T cell division. Ki-67 was only detected in PfSPZ-specific CD4 T cells and not in total memory CD4 T cells (Fig. 5b-h), providing evidence for PfSPZ specificity. One to two weeks following the 3-week CHMI, Ki-67 was detected in ~40-60% of PfSPZ-specific CD4 T cells from subjects with parasitemia (both vaccinated and unvaccinated subjects), showing that malaria infection was associated with activation of a high proportion of PfSPZ-specific npg CD4 T cells (Fig. 5c,d). In contrast, there were low to undetectable numbers of Ki-67 + PfSPZ-specific CD4 T cells in PBMCs from the 25 subjects in the VRC 312 (n = 12; Fig. 5c) and VRC 314 (n = 13; Fig. 5f) studies who did not develop parasitemia after the 3-week CHMI. One to two weeks after the 21-to 25-week and 59-week CHMIs, ~10-20% of PfSPZ-specific CD4 T cells from the subjects who did not develop parasitemia were Ki-67 + (Fig. 5d,g,h), as compared to the 0-1% of PfSPZ-specific CD4 T cells that were present on the day of the CHMI. These data suggest that there was differential antigen exposure in vivo when CHMI was done at 3 weeks versus 21-25 or 59 weeks among subjects who remained without parasitemia. The percentage of Pf-specific CD4, CD8, or γδ T cells in blood did not change after CHMI at any time in subjects who did not have parasitemia (Supplementary Fig. 9a-f), nor did PfCSP-specific antibody levels (Supplementary Fig. 9g).

Tissue distribution of Pf-specific T cells
The low amount of Pf-specific antibody at 59 weeks for all five vaccinated subjects who were not parasitemic after CHMI (Fig. 4a,b) suggested that antibodies did not probably have a major role in mediating protection at this time point. Therefore, PfSPZ-specific T cell responses were assessed in blood throughout the course of vaccination and following each CHMI.

Longitudinal cellular responses in humans.
In PBMCs, cytokineproducing PfSPZ-specific CD4 T cell responses declined significantly over the course of 59 weeks in vaccinated subjects who remained without parasitemia (P = 0.047; Fig. 6a), and cytokine-producing PfRBC-specific CD8 T cell responses were induced by vaccination; however, they returned to pre-vaccine levels shortly after the final vaccination (Fig. 6b). Thus, neither antibodies nor Pf-specific T cells were readily detected in the blood of the five vaccinated subjects who were not parasitemic after CHMI at 59 weeks. We hypothesized that the T cells resident in the liver might have a critical role in protection 10,26 .

Tissue-resident T cell responses in NHPs.
Because liver-resident cellular immunity could not be directly assessed in the human subjects, such responses were assessed in NHPs following immunization with PfSPZ (Supplementary Fig. 10 and Supplementary Table 6).
Immunogenicity studies with PfSPZ Vaccine in NHPs previously provided immune data that guided successful translation to human studies 10,11 . We had previously shown that 1.35 × 10 5 PfSPZ administered by the s.c. route did not induce detectable T cell responses in the blood or liver of NHPs 10 , so we administered tenfold more PfSPZ in each of five doses that were administered by the i.m. or i.d. route and compared the T cell responses to those resulting from five i.v. doses The experiment was performed once. Data are mean ± s.e.m. y axis on the right denotes the raw luciferase signal. y axis on the left denotes the percentage of signal in mice that received the IgG from each of the five subjects (identified by the identification numbers on the x axis) at each of the two time points relative to the signal in the mice that received pre-vaccination IgG, which was set to 100%. (d) Correlation between PfCSP antibody abundance (as determined by ELISA) and liver-stage burden (as measured by luciferase expression) in the FRG-huHep mice. In a,b, n = 7 (−) and n = 2 (+) for left-hand graphs; n = 9 (−) and n = 6 (+) for middle graphs; n = 5 (−) for right-hand graphs. In c, n = 7 for pre-vaccine, n = 2 for PfCSP mAb, and n = 4 for each IgG sample, except 511 (n = 3). In d, n = 10 for IgG samples. In a,b, bar denotes geometric mean, and comparison (P values) between groups was assessed by the Mann-Whitney U-test.
In d, Pearson correlation coefficient was used to determine the relationship between Pf liver-stage burden and PfCSP level.
npg of 1.35 × 10 5 PfSPZ and four i.v. doses of 2.7 × 10 5 PfSPZ (Fig. 6c). PfSPZ-specific CD4 T cell responses in PBMCs were modestly higher in NHPs vaccinated by the peripheral (i.m. or i.d.) routes than by the systemic (i.v.) route (P = 0.046), and there were no differences in PfRBC-specific CD8 T cell responses in PBMCs (Fig. 6d). In contrast, i.v. administration induced twofold higher frequencies of PfRBCspecific CD8 T cells in the liver, as compared to those induced by the 6-to 10-fold higher total doses of PfSPZ that were administered by the i.m. or i.d. route (Fig. 6e). Notably, the frequency of PfRBCspecific CD8 T cell responses in the liver were ~100-fold higher than in PBMCs. Moreover, whereas multiple populations of IFN-γ-producing lymphocytes-including NK cells, γδ T cells, and CD4 T cells-were identified in the livers of NHPs, the majority (~60%) were CD8 T cells (Fig. 6f and Supplementary Fig. 11). The composition and differentiation state (i.e., memory phenotype) of lymphocytes in NHP livers largely reflect that of unvaccinated human liver samples, underscoring the biological relevance of the NHP model (Supplementary Fig. 11).

DISCUSSION
This was the first clinical trial of a malaria vaccine in which the first CHMI in one of the vaccine groups was done more than 4 weeks after the final immunization, thus representing a shift beyond a search for short-term protection toward trials that focus on dose and regimen optimization for durable sterilizing protection. At 59 weeks after four doses of 2.7 × 10 5 PfSPZ each, the cumulative VE against CHMI was estimated to be 55%-the most durable efficacy against CHMI reported to date with an injectable malaria vaccine. In the vaccination regimens tested here, the efficacy of PfSPZ Vaccine depended on both the dose per vaccination and the number of vaccinations. The estimated VE against CHMI done 3 weeks after immunization with three doses of 2.7 × 10 5 PfSPZ was 24%, as compared to 73% with four doses of 2.7 × 10 5 PfSPZ. Similarly, four or five doses of 1.35 × 10 5 PfSPZ conferred an estimated 25% efficacy against CHMI at 21 weeks, as compared to 55% with four doses of 2.7 × 10 5 PfSPZ. On the basis of these data, we hypothesize that additional increases in the dosage of PfSPZ Vaccine will further increase the magnitude and durability of protective efficacy. Ongoing studies using 4.5 × 10 5 to 2.7 × 10 6 PfSPZ per dose are assessing this for homologous CHMI, heterologous CHMI, and natural exposure in all age groups. The route of administration influenced VE, further underscoring the importance of PfSPZ administration by the i.v. route in achieving protection with this vaccine. Completed studies in Mali, Tanzania, and Equatorial Guinea have provided the first data for the feasibility of direct venous inoculation (DVI) of PfSPZ Vaccine to adults in Africa (M. Sissoko (International Center for Excellence in Research, Mali), S. Healy (NIH), S. Shekalaghe & A. Olotu (both from Ifakara Health Institute, Tanzania), personal communication), and field studies in Africa have begun to determine the safety, efficacy, and feasibility of DVI administration in young children and infants.
An unexpected immunological finding was that the pre-vaccination frequency of circulating Vδ2 + T cells correlated with outcome of CHMI. Vδ2 + T cells recognize intracellular lipid metabolites (such as hydroxyl-methyl-butenyl-pyrophosphate (HMBPP)) from Plasmodium 27,28 . It is possible that these cells have a role in the initial priming of adaptive immune responses through rapid Pf recognition, npg IFN-γ production (Supplementary Fig. 7m), and/or by serving as antigen-presenting cells 29 at the time of immunization, leading to enhanced CD8 T cell priming 30 . γδ T cells have been shown to provide help to dendritic cells to facilitate protective responses to blood-stage malaria 31 , and IFN-γ-secreting γδ T cells are a correlate of protection against blood-stage malaria 32 . It is possible Vδ2 + T cells also mediate direct effector functions in the liver. A high frequency of circulating γδ T cells express CXCR6, a chemokine receptor that facilitates surveillance of the liver sinusoids 22 , and produce IFN-γ and perforin (Supplementary Fig. 7i-m), two potent effector molecules shown to mediate killing of intracellular Pf. γδ T cells have been shown to contribute to protection against pre-erythrocytic malaria in mouse models 21 , and these T cells expand in humans following whole-SPZ immunization 33 . The correlation between γδ T cell frequency and CHMI outcome described herein suggests that PfSPZ Vaccine induces protective immunity not only through induction of conventional CD8 and CD4 T cell responses but also through induction of γδ T cells. Such broad-based T cell responses that are induced by live-attenuated vaccines may not occur after immunization with the most commonly used protein-or virus-based subunit vaccines, highlighting one of many important differences between these approaches. We previously showed a dose-dependent increase in Pf-specific antibodies and multifunctional cytokine-producing CD4 T cells with doses between 7.5 × 10 3 and 1.35 × 10 5 PfSPZ that were administered by i.v. injection, but we could not assess correlates of protection because of the high-level of protection at the 3-week CHMI 11 . The clear dose response in this prior study 11 probably reflects the wider range (18-fold) of doses given. In the present study, there was a smaller range of doses that were given by the i.v. route, which probably accounts for the narrower range of antibody and T cell responses. The correlates of protection described here are hypothesis-generating and require validation in large prospective studies. Moreover, correlates of protection may differ depending on the dose per vaccination, the number of vaccinations, the time of sampling, and history of malaria exposure.
The data presented here are consistent with the following role for antibodies: at the time of early CHMIs (3 weeks), antibody responses were highest, and subjects may have reduced the numbers of PfSPZ that successfully invaded hepatocytes (Fig. 4c,d) [14][15][16] , allowing for rapid elimination of the remaining parasites in the liver by cellular responses 34,35 , thereby resulting in low Ki-67 expression by PfSPZ-specific CD4 T cells in the protected subjects. At the time of the later CHMIs (21-25 weeks and 59 weeks), antibody levels were considerably lower, resulting in more parasites reaching the liver after CHMI, thus leading to greater activation (based on Ki-67 expression) of PfSPZ-specific CD4 T cells during the clearance of liver-stage parasites (Fig. 5d,g,h).
Circulating Pf-specific T cell responses were low in blood at 59 weeks in all of the subjects who were not parasitemic after CHMI  npg (Fig. 6a,b). Thus, if T cells are contributing to protection, then it may be due to an unmeasured aspect of the response in blood or to differential responses in the liver. Indeed, tissue-resident Pf-specific CD8 T cell responses in the liver of PfSPZ-vaccinated NHPs were highest after i.v. administration despite using 6-to 10-fold higher doses with the i.m. or i.d. route. Infectivity studies in mice show that >90% of SPZs that are administered by the i.m. or i.d. route fail to reach the liver 12 , and the parasite antigens are limited to muscle-or skin-draining lymph nodes 36 . Intravenous administration, in contrast, results in the efficient delivery of attenuated PfSPZ to the liver, where they presumably mediate priming of T cells that are programmed to remain as non-recirculating, liver-resident cells 26 . For PfSPZ Vaccine, efficacy is critically dependent on the route of vaccination. The data from the NHP studies indicate this is likely because i.v. administration induces robust tissue-resident responses in the liver, whereas peripheral (i.m. or i.d.) administration results in T cell responses in blood but more limited responses in the liver. Thus, for vaccines seeking to induce protective T cells, vaccine development efforts may be misled by focusing solely on maximizing T cell responses in the blood without considering how the vaccination strategy affects T cell responses at the infection site. In summary, the immune data presented here support the conclusion that whereas antibodies may have some contribution to protection early after final immunization, tissue-resident CD8 T cells are probably necessary for durable sterile protection 10,[17][18][19][20]26,37,38 .

METHODS
Methods and any associated references are available in the online version of the paper. stimulation antigens, and protocol as the intracellular cytokine staining assay used for human PBMC samples (described above). The staining panel is shown in Supplementary Table 6; the gating tree is shown in Supplementary Figure 10; and the antibody clones and manufacturers are shown in Supplementary Table 6.
Investigators were blinded to the treatment group during lymphocyte stimulations and flow cytometry gating. All antigen-specific cytokine frequencies are reported after background subtraction of identical gates from the same sample incubated with the control antigen stimulation (HSA or uRBC).

Statistical analyses.
Flow cytometry data was analyzed using FlowJo v9.8.5 (Tree Star). Statistical analyses were performed with Pestle v1.7 and SPICE v5.3 (M. Roederer) 49 , JMP 11 (SAS), Prism 6 (GraphPad), and R v3.2.2 with RStudio v0.99.483. For vaccine immunogenicity, comparisons between groups were performed using Kruskal-Wallis with Dunn's post-test correction for multiple comparisons. If no differences between vaccines groups were identified by Kruskal-Wallis, then differences from pre-vaccination were assessed by Wilcoxon matched-pairs signed rank test with Bonferroni correction for multiple comparisons, as specified in the figure legends.
Immune responses were assessed 2 weeks after the final immunization or prevaccination (as specified in the figure legends) and were compared to outcome at either 3-week CHMI or 21-to 25-week CHMI. Assessment of immune responses that correlate with outcome at CHMI (parasitemia or no parasitemia) was made using a stratified Wilcoxon test controlling for vaccine regimen as a covariate.