Background

Plasmodium falciparum malaria remains a pressing global health emergency. Encouraging progress in its control has been made in some areas of Africa [1], but elimination does not appear realistic in many areas. The current lead vaccine candidates are based on the circumsporozoite protein (CSP) and have been shown to be protect against clinical manifestations of P. falciparum disease in children [2, 3]. Higher vaccine efficacy against clinical manifestations might be achievable through inducing immune responses against antigens from the asexual blood-stages [4]. The clinical development pathway for any one candidate vaccine is expensive and lengthy. None of the blood-stage candidate vaccines subjected to field trials have progressed to Phase III trials [5, 6]. The need to understand and interrogate naturally acquired immunity to malaria is fundamental to antigen selection and vaccine design. A common approach is to use immuno-epidemiological studies in malaria endemic regions, where immunological responses from cross-sectional surveys of children are linked with the risk of subsequent malaria episodes [7,8,9,10,11]. A limitation of this approach has been the reliance on uncontrolled natural but heterogenous exposure to malaria [8, 9, 12] as well as exposure to genetically diverse parasites [13] in the field.

Controlled Human Malaria Infection (CHMI) studies have the potential to accelerate the selection of antigens for vaccine development by controlling for malaria exposure, including parasite strain, as well as level of infectious dose. For ethical reasons, CHMI requires adult volunteers rather than children. In endemic areas immunity is acquired with age and adults usually have high levels of immunity to the consequences of infection [14]. Nevertheless, even among adults’ levels of immunity may be variable. We have recently described clinical outcomes and safety of CHMI in Kenyan adults after infection with cryopreserved viable, aseptic, and purified Plasmodium falciparum sporozoites (PfSPZ Challenge) at a dose of 3200 injected by syringe [15]. We showed that using CHMI in this population of 142 pre-exposed adults, 26 (18.3%) had febrile symptoms and were treated; 30 (21.1%) reached ≥ 500 parasites/µl and were treated; 53 (37.3%) had parasitaemia without meeting thresholds for treatment and; whilst 33 (23.2%) remained qPCR negative (in a subset of volunteers, some of those qPCR negative between days 8 and 10 post-infection had low parasitaemia in comparison to two other qPCR methods) [15]. These findings are consistent with other CHMI studies in volunteers from endemic areas [16, 17]. However, the outcomes of CHMI that are most strongly associated with naturally acquired immunity have not yet been determined. Furthermore, categorizations into multi-level descriptive outcomes do not maximize analytical power for correlates of immunity, and either a binary classification or a continuous variable would be analytically optimal.

We therefore in this study conducted an analysis using anti-schizont antibody responses and location of residence as surrogates of immunity. We examined various parameters from the patterns of parasite growth during CHMI. This was to determine which parameters most closely associated with these two surrogates of immunity, and to identify whether any were more discriminatory of host immunity than the standard immuno-epidemiological studies conducted in the field.

Methods

Study design and population

The full protocol [18] and description of safety and outcome [15] has been published. Briefly, data from the CHMI-SIKA study, which was an open, un-blinded and non-randomised with all volunteers receiving an intravenous injection [direct venous inoculation (DVI)] dose of 3.2 × 103 PfSPZ Challenge PfNF54 strain (i.e. cryopreserved, infectious sporozoites). The volunteers were monitored for blood parasitaemia by quantitative polymerase chain reaction (qPCR) to determine parasite growth. The 3.2 × 103 PfSPZ dose was selected because this has infected 100% malaria-naïve volunteers undergoing CHMI in studies in the US and EU [19, 20]. PfNF54 is African in origin and it is therefore expected that < 100% of African volunteers with well-developed naturally acquired immunity will become infected [21].

Anti-malarial drug concentration

We retrospectively measured the concentrations of anti-malarial drugs (artemether, dihydroartemisinin, sulfadoxine, pyrimethamine, chloroquine, lumefantrine, and desbutyl-lumefantrine), retrospectively at a day before challenge (C − 1) and after challenge (at C + 8) [15]. We excluded those with drug levels above the minimum inhibitory concentration (MIC) for lumefantrine but retained those with levels below the MIC for sulfadoxine (in absence of pyrimethamine) and with trace levels of chloroquine as described previously [15].

Anti-schizont antibody levels

Plasma samples were tested by ELISA for the presence of human IgG against schizont extract as described previously [22, 23]. P. falciparum 3D7 strain parasites were cultured to schizont stage to make a preparation of schizont extract. To run the ELISAs, the extract was used to coat high absorbance plates at an established concentration shown to have saturation of responses using plasma from hyper immune individuals. The assay was repeated if duplicate optical density (OD) values for an individual plasma sample varied by more than a factor of 1.5. A pool of serum samples from an area in Africa where malaria is highly endemic was titrated on each plate and acted both as a positive control and provided values for a standard curve for converting optical density (OD) readings into concentrations (Antibody Units, AU).

Location of residence

Volunteers were recruited from differing malaria endemic regions in Kenya: Ahero in Western Kenya (moderate to high transmission region); Kilifi North on the Kenyan Coast (low to no malaria transmission); and Kilifi South (moderate transmission region) [1, 24]. In this analysis volunteers from Ahero and Kilifi South were combined as resident at “high transmission” intensity and Kilifi North taken as resident at “low transmission” intensity.

Parasite detection by qPCR

For parasite detection, venous blood samples were collected twice every day from days 8 to 15 of CHMI and then once every day from days 16 to 22 of CHMI for qPCR analysis by detection of the 18 S ribosomal RNA P. falciparum gene [15] in triplicates in a TaqMan assay using primers and probes previously described [25]. Non-template control was used as a negative control (in triplicate wells) with parasite quantification against known cultured parasite standards comprising of 6 serial dilutions of extracted DNA also run in triplicates. The cultured parasite standards were produced in 3 different batches. Selected samples were re-ran from each CHMI cohort against a final set of standards, including the WHO external quantified quality control sample [26].

Statistical analysis

PCR results are presented as the geometric mean of three replicate assays at each timepoint. Time to treatment was the number of days between challenge and the treatment decision taken either because: (a) the volunteer reached the pre-assigned threshold of 500 falciparum parasites/ml by qPCR; (b) they had developed febrile symptoms and clinicians had treated them at lower parasite density; or (c) they reached the end of the study without reaching parasite density threshold or having symptoms. Other parameters to describe the outcomes were derived from qPCR results as follows: (i) the time to particular parasite density thresholds, where volunteers not reaching those thresholds were described as missing data; (ii) the “proportion of days growing” where any consecutive increase in parasite density is considered a “day growing” (this was calculated from raw data, then also from smoothed data taking the moving average over 2 days); (iii) the mean parasite density as a geometric mean, excluding timepoints after treatment; (iv) the maximum number of days of continuous consecutive growth; (v) the gradient of growth from a best-linear-fit of the period defined in (iv); (vi) the median number of days since challenge for the days of parasite growth as defined in (ii); (vii) the converse of (ii), (iv), and (v); (viii) for days of decline in parasite density rather than growth; (ix) the “inoculum” defined as the peak parasite density observed between days 8.5 and 10 after challenge and; (x) the “variability” calculated as the summed day-to-day variation in parasite density. Kruskal–Wallis tests with multiple comparisons were used to compare anti-schizont antibodies by location and qPCR outcome and Spearman’s rank correlation was used to explore correlations between anti-schizont antibody, location, and qPCR parameters.

Survival models were developed using Cox regression in three stages; (a) univariable analysis of all potential independent predictors; and (b) multivariable analysis including significant predictors from (a); second multivariable analysis retaining only significant predictors from (b). The variability in outcome explained by anti-schizont antibodies was calculated using pseudo r2. To compare the CHMI cohort with a previous observational field study of children [9, 11, 23], the antibody levels were divided into two groups (above and below the median), and analysis of the child cohort was restricted to the asymptomatically infected group where the protective effect of anti-schizont antibodies had been shown to be most evident [8, 11, 23].

Results

Anti-schizont antibody responses for the volunteers enrolled in the study

Data from 142 volunteers were included in the analysis as previously described [15]. The median age of the volunteers was 28 years old (range 18–45) and 30% were female. Antibody responses to schizont extract were measured for all the volunteers at screening (Additional file 1: Fig. S1). Volunteers from Kilifi North had significantly lower anti-schizont antibodies (median of 896 Antibody units (AU), 95% CI 566 to 1473) compared with volunteers from Kilifi South (median of 9238 AU, 95% CI 6399 to 12,324, p < 0.00001) and Ahero (median of 4666 AU, 95% CI 966 to 28,702, p < 0.00054) but volunteers from Kilifi South and Ahero had similar antibody levels (p = 0.085). For further analysis, volunteers from Kilifi North (N = 34) were considered to be residents of an area of “low transmission” whilst volunteers from Kilifi South (N = 93) were combined with Ahero volunteers (N = 15) and considered to be residents of an area of “high transmission”.

qPCR categorized outcomes in relation to location and antibody response

We had previously observed four distinct outcomes based on parasite growth as measured by qPCR following CHMI [15] being parasite growth by qPCR meeting the threshold criteria for malaria diagnosis (≥ 500 parasites/ml) either: (a) with fever (i.e. “treated febrile”); (b) without fever but reaching a parasite density requiring treatment (i.e. “treated non-febrile”); (c) with parasites detected by qPCR but not at a parasite density meeting the threshold criteria for treatment (i.e. “PCR positive untreated”); or (d) parasites not identified by qPCR throughout monitoring (i.e. “PCR negative”) (Additional file 2: Fig. S2). Volunteers who were “treated febrile” had lowest anti-schizont antibodies and were least likely to be residents of the high transmission areas (Table 1, Additional file 3: Fig. S3). The “treated non-febrile” group had intermediate levels of anti-schizont antibodies and intermediate likelihood of being residents of the high transmission area. The “untreated PCR positive” group and then “PCR negative” group had high levels of anti-schizont antibodies and were both very likely to be residents of the high transmission areas. Those who were qPCR positive untreated could further be examined in sub-groups by dividing them into those who were positive either early, late, or throughout the period of qPCR monitoring. We did not identify any significant differences in anti-schizont antibodies or location of residence for these additional sub-groups (Additional file 5: Table S1).

Table 1 qPCR outcome in relation to anti-schizont antibody responses
Full size table

qPCR parameter associations with location and antibody response

We examined various parameters that described the qPCR results per individual volunteer (Table 2). The strongest non-parametric correlates of location of residence (i.e. residence at high vs. low transmission intensity) or anti-schizont antibodies were time to reaching a threshold of 250 parasites/ml; time to treatment; and the categorization of treatment versus no treatment (Table 2). Other parameters that were strong correlates of location of residence or of anti-schizont antibodies were highly cross correlated with each other (Additional file 4: Fig. S4) and we did not identify a second independent predictor using parametric analyses after adjusting for time to treatment (Additional file 5: Table S2). The parameter time to treatment was used for further analysis over the use of time to a threshold of 250 parasites/ml since some volunteers were treated at lower parasite densities leading to missing data for time to 250 parasites/ml.

Table 2 Non-parametric analysis of qPCR parameters with anti-schizont antibody responses and location
Full size table

Survival analysis

We developed a multivariable Cox regression model of time to treatment, finding both anti-schizont antibodies and location of residence to be strong independent predictors of outcome (Table 3). The presence of parasites at screening and plasma lumefantrine drug concentrations were weak predictors of outcome in univariable analysis (Table 3), but not in multivariable analysis (Multivariable 1, Table 3). Parasites at screening and lumefantrine drug concentrations were both confounded by location of residence (r = 0.20, p = 0.016 and r = 0.30, p = 0.0003 for associations with location of residence, respectively). The year of enrolment in the trial (cohort year), anti-malarial drug concentration, age, and gender were not significant predictors of outcome. In the final model (Multivariable 2), the two independent predictors were residence (i.e. at high vs. low transmission) and anti-schizont antibody concentration, explaining 35% of the variability in outcome on logistic regression (Table 3, and Fig. 1).

Table 3 Cox regression analysis of time to treatment
Full size table
Fig. 1
figure 1

Time to treatment survival analysis. Kaplan–Meier curves comparing time to treatment with location of residence (left panel) and anti-schizont antibody response (right panel). Shown are survival curves for location of residence is low transmission (blue) vs. high transmission (red). For anti-schizont antibody responses shown is low antibodies (blue) vs. high antibodies (red)

Full size image

We compared the predictive strength of anti-schizont antibodies in the CHMI model with previous cohort studies based on natural exposure in the field [9, 11, 23], to determine whether the CHMI model would advance the field in examining for correlates of infection. In order to make comparisons across models, we used dichotomized anti-schizont antibody levels above and below the median for each study to have a bi-level comparison in each setting that was not dependent on the different range of antibody levels. We compared the pseudo R2 in logistic regression to determine the variability in outcome explained by antibody levels in each setting. In CHMI, the odds ratio (OR) of requiring treatment based on anti-schizont antibody levels above the median was OR = 0.12 (95% CI 0.06 to 0.27, p = 2 × 10−7) and explained 17% of the variability. In the previously reported cohort of 121 children between the ages of 1 and 8 years found to be parasite positive at baseline with the inclusion criteria for analysis being residence in the study area, having anti-schizont antibodies above the median level was associated with OR = 0.64 (95% CI 0.29 to 1.4, p = 0.26) for febrile malaria, explaining 0.8% of the variability in outcome. Survival plots from the CHMI study showed a clear distinction in time to treatment by anti-schizont antibody responses (Fig. 2, left panel), in contrast to the less clear distinction seen in field studies based on natural exposure (Fig. 2, right panel).

Fig. 2
figure 2

Survival analysis of CHMI vs. field-based observational study. Kaplan–Meier curves comparing of the CHMI cohort (left panel) and field-based cohort (right panel) in relation to requirement for treatment and anti-schizont antibody responses. Antibody responses are shown as low (red) vs. high (blue)

Full size image

Discussion

We used serial qPCR to determine the outcomes most strongly associated with anti-schizont antibodies and location of residence (low vs. high transmission), in order to define the outcomes for CHMI in exposed adults that are most strongly associated with surrogates of immunity. We used anti-schizont antibody levels and location of residence at varying prior exposure to malaria as surrogates for immunity to malaria. We examined several potential parameters based on the qPCR monitoring done for CHMI for their association with anti-schizont antibodies and location. Time to treatment and time to 250 parasites/µl were strongly associated with anti-schizont antibodies and with location. We preferred time to treatment rather than time to 250 parasites/µl, as the former included the full set of volunteer data and avoids the potential bias of missing data from volunteers who were treated before reaching 250 parasites/µl.

After adjusting for time to treatment, there were no other independent predictors of anti-schizont antibodies or of location of residence. We therefore developed a survival analysis based on time to treatment. The combination of location of residence and anti-schizont antibodies as a continuous variable explained 35% of the variability in time to treatment in CHMI. Since prior residence and anti-schizont antibodies only offer limited information on the true extent of host immunity, this implies that a very significant proportion of the variability in outcome in CHMI is due to host immunity.

We examined whether the analysis of CHMI for naturally acquired immunity was a significant advance over previous studies conducted in the field based on natural exposure to malaria. Adults have higher levels of immunity than children, different endpoints are used for adults participating in CHMI compared with children in field observational studies, but nevertheless these study designs both share the aim to define potential correlates of immunity. In order to make comparisons, we used logistic regression with febrile malaria as the outcome in the field studies, and with treatment criteria as the outcome in CHMI. We used anti-schizont antibodies as the predictor variable. Levels of anti-schizont antibodies higher among adults than children, so we divided antibodies into high or low categories based on the median antibody level in each study. In the field-based observational study analysed here in a cohort of children, anti-schizont antibody responses explained less than 1% of the observed variability, but anti-schizont antibodies explained 17% of the variability in CHMI outcomes. This is not surprising given the variability in exposure to malaria seen in the field [8, 9], whereas in CHMI exposure is controlled and does not vary between participants.

This analysis, here, shows how, adjusting and accounting for heterogeneity of exposure and infection, and given that anti-schizont antibodies in field-based studies account for a small fraction of the variability, CHMI in an adult pre-exposed population has a larger discriminatory power to study immunity in relation to past exposure. Furthermore, in non-immune CHMI studies, a large proportion of the volunteers develop illness and require treatment at relatively low parasitaemia thresholds (between 5 and 50 parasites/ml) whilst in our study, individuals were often asymptomatic and parasite-free and this could largely be as a result of differences in responses in non-immunes with semi-immunes [27, 28]. With the exception of innate factors such as sickle cell trait [17], this resistance in previously malaria exposed individuals thus might be as a result of acquired adaptive immunity which is confirmed by anti-schizont antibody responses.

Thus, these findings presented here, provide a unique opportunity to advance the field of vaccine antigen discovery utilizing the CHMI platform with characterization and better understanding of the development of immunity to infection in the context of past malaria exposure. A comprehensive analysis of signatures or correlates of immunity as has been recent detailed utilizing systems serology approaches (both qualitative and quantitative antibody-based approaches) [29] will significantly advance the field.

This analysis, not relying on one or two parameters of the outcome measure (PCR) especially in the context of undertaking these studies in populations with varying past exposures to malaria is warranted. Traditionally, studies have relied on the parasite growth kinetics/rate as an important measure of outcome in CHMI studies—including as an assessment of vaccine or drug efficacy [30]. CHMI studies enrolling volunteers with a range of parasite exposures, to date in Africa, have taken the approach of endpoint measurement largely based on thick-blood microscopy at a particular threshold for diagnosis [17, 22, 31] to explain parasite growth rates. Achan et al. [16] despite using PCR as a criteria for endpoint, did not have the same breadth of past exposures as presented here. Hence, it is important for studies particularly utilizing PCR to undertake a detailed analysis of the most reliable parameter that would account for diversity in parasite growth.

Conclusions

We thus conclude that CHMI studies in malaria endemic areas, using a standardized inoculum, are an effective platform and powerful tool with which to study host immunity. Variability in outcomes are more closely attributable to host immunity than for field-based studies based on natural exposure. Anti-schizont antibodies are generally not considered to be mechanistically related to immunity but rather a marker of past exposure. Anti-schizont antibodies are thus likely to be cross correlated with multiple other potential mechanisms of immunity [22, 23, 32, 33]. Hence in further studies of host immunity, we would expect antigen-specific responses and functional antibodies to explain remaining variability in outcome. It is possible that a few mechanistic markers will be independently associated with outcome and that, on adjusting, the associations with other markers of exposure will be attenuated, or that outcome will be explained by a range of parameters independently. In either case, the experimentally controlled conditions are expected to leave less unexplained variation than occurs in the field due to variable exposure to mosquito bites. In follow up analyses, immunological parameters, including but not limited to for example functional assays for blood-stage immunity [34, 35] and protein microarray analyses to identify blood-stage antigens [36], will need to be undertaken in relation to the PCR parameters described here. This will aid in the further identification of signatures or correlates of immunity. Previous immuno-epidemiological studies using observational cohorts have identified several immunological markers much more strongly associated with immunity than anti-schizont antibody [11], and these findings can now be tested using the CHMI approach in malaria endemic areas.