Background
Despite the existence of effective anti-parasitic drugs, malaria remains a critical global health problem, estimated at causing 409,000 deaths and 229 million cases in 2019. Some 94% of cases were in Africa, where almost all infections were caused by
Plasmodium falciparum [
1]. Currently, the most advanced malaria vaccine, RTS,S, exhibits 28–36% efficacy in infants and children observed over an average time period of 4 years [
2]. A more effective malaria vaccine would be a valuable tool for curbing malaria, especially given the emergence of resistance to frontline artemisinin combination therapy and development of insecticide-resistant mosquito vectors [
3,
4]. Repeated natural malaria infections can result in acquisition of semi-protective immunity with persistent low level parasitaemia and primarily asymptomatic cases [
5]. Serious malaria-related complications and death occur primarily in infants and children, prior to the development of partially protective immune responses [
6]. However, acquisition of sterilizing immunity targeting the pre-erythrocytic stage of the parasite, resulting from immunization with radiation-attenuated malaria sporozoites, has been experimentally demonstrated in animal models and in humans [
7‐
10].
Malaria sporozoites develop in the mosquito and are injected into the skin during a female mosquito blood meal from where they make their way to the liver and infect hepatocytes. There they multiply and over the course of 5–9 days, asymptomatically develop into thousands of merozoites which emerge from the liver and serially infect erythrocytes, resulting in blood-stage infection and disease. The pre-erythrocytic stage initiated by sporozoites is a population bottleneck in the parasite lifecycle, and is an attractive target for vaccine development strategies. It was first demonstrated in a mouse model in 1967 that immunization with radiation-attenuated sporozoites (RAS) results in effective protective immunity against challenge with infectious sporozoites [
10], and demonstrated subsequently for
P. falciparum RAS (PfRAS) in human cohorts in multiple clinical trials [
7,
11].
The immune mechanisms of human protection resulting from immunization with whole-sporozoite vaccines remain poorly understood but available evidence indicates that the development of this immunity requires liver infection. Work in animal models shows important roles associated with protection for antibodies, liver resident CD8+ memory T cells (Trm) and type I interferon responses [
12‐
14]. However, results in animal models may not directly translate to humans, and the ability to directly monitor responses in human liver during vaccination and after controlled human malaria infection (CHMI) is very limited. Blood represents an accessible and immunologically important tissue which is reflective of systemic immune responses and its analysis can aid investigation of immune protection against malaria.
Two human RAS vaccination trials that resulted in a portion of the trial participants being protected from infection following CHMI have been performed, allowing comparisons between protected and non-protected subjects. Immunization by mosquito bite with radiation attenuated sporozoites (IMRAS) [
15], [NCT01994525] and Bagamoyo sporozoite vaccine 1 study; immunizations with Sanaria
® PfSPZ Vaccine (BSPZV1) [
16] trials, both included immunization of volunteers with five consecutive PfRAS vaccinations followed by homologous CHMI using
P. falciparum strain NF54. Whole blood was sampled repeatedly from participants and analysed by RNAseq to provide longitudinal data on the immune responses. These trials differed with respect to malaria experience of the volunteers and route of administration for both immunization and CHMI, as described in “
Methods”. This study analysed comparable time points in both trials up to 1 week after the first RAS vaccination. Shared transcriptional responses were identified in volunteers participating in both studies after initial RAS vaccination, indicating the existence of an initial core transcriptional response programme to RAS vaccination across diverse populations. Deviations from this transcriptional programme were associated with lack of protection.
Methods
Challenge trials
The IMRAS trial was performed in Bethesda, MD, USA with malaria-naive participants immunized by four rounds of ~ 190 bites of PfRAS-infected mosquitoes at 4-week intervals followed by a fifth immunization a further 5 weeks later. Protection was assessed 3 weeks after last vaccination by homologous CHMI via bites of five mosquitoes carrying non-irradiated P. falciparum sporozoites. This intentionally suboptimal immunization schedule resulted in 55% protection among the IMRAS cohort examined in this study.
The BSPZV1 trial was carried out in Bagamoyo, Tanzania with volunteers who potentially had prior malaria infections, but were free of symptomatic malaria for the previous 2 years. They were immunized 5 times by direct venous injection (DVI) with either 1.35 × 105 or 2.7 × 105 PfSPZ of Sanaria® PfSPZ Vaccine (aseptic purified cryopreserved PfRAS) given at 4-week intervals. This was followed by CHMI at 3 weeks after the final immunization via DVI with PfSPZ Challenge (NF54) (purified cryopreserved non-irradiated SPZ), at a dose of 3200 PfSPZ. The resultant protective efficacy was 20%. Because not all non-protected individuals were included in RNAseq analysis, there was 36% protection among analysed individuals.
Total RNA was extracted from whole blood in PAXgene Blood RNA tubes that had been stored at – 80 °C, using the PAXgene Blood Kit (PreAnalytiX) following the manufacturer’s protocols. RNA was quantified by spectrophotometry and ~ 2.3 μg of total RNA per sample was processed using the GLOBINclear Human kit (Ambion) in order to remove globin mRNA. RNAseq was performed by Beijing Genomics Institute using either the Illumina Hiseq2000 (75 bp read length, paired-end), or the BGI500 platform (100 bp read length, paired-end) to a depth of at least 30 million reads per sample. Reads were aligned to the human hg19 genome using STAR [
17], and htseq-count [
18] using the intersection-strict option was used to convert mapped reads into a gene count table. Genes were filtered to retain only genes with > 10 read counts in at least 5% of samples. Counts were then normalized using the R limma [
19] and voom [
20] packages to account for sequencing depth and log2 transformed for all downstream analysis.
Linear mixed-model analysis
The R lme4 [
21] package was used to fit nested mixed models to assess differential gene expression in both cohorts, using the lmer function. Random intercepts were fit per-subject to account for multiple samples being drawn from the same study subjects. Five mixed models were fit to each gene, separately for differential expression at each sample day relative to the previous sample day in each study. For IMRAS these time intervals were days 1 vs 0, 3 vs 1, 7 vs 3, and for BSPZV1 days 2 vs 0 and 7 vs 2. Time and protection status were encoded as binary variables.
Formulae were fit as follows:
$$geneExpr\sim 1 + Time*Protection + \left( {1|Subject} \right)$$
(1)
$$geneExpr\sim 1 + Time + \left( {1|Subject} \right)$$
(2)
$$geneExpr\sim 1 + Protection + \left( {1|Subject} \right)$$
(3)
This model structure assumes that individual transcriptional responses are a function of (vaccine induced) responses over a time interval, and individual protection status. In the full model (1), both time and protection status are included. Reduced models comprising only time (2) or protection status (3) were also fit and, to assess the statistical significance of these parameters of interest, p-values associated with time- or protection-associated gene responses were determined by contrasting full models with reduced models. Specifically, p-values for time-associated genes were calculated by ANOVA to determine whether Eq. (
1) significantly improved fit to the data compared to Eq. (
3) using the anova.merMod() function. Similarly, Protection p-values were obtained by comparing Eq. (
1) to Eq. (
2). Response gene p-values were then false-discovery rate adjusted for multiple testing.
Directionality (UP, DN or NC (no change)) of each gene response was assessed using confidence intervals (CIs). The 90% CIs were estimated for the coefficient of Eq. (
2), cases where the lower CI > 0 were considered UP genes, upper CI < 0 were considered DN genes.
Significant response genes were identified as those that met an false discovery rate (FDR) threshold < 0.2 a nominal p-value < 0.05, and were classified as UP or DN in direction. Candidate protection-associated genes were filtered to only include significant response genes, then false discovery rate adjusted. Protection genes were selected that met an FDR threshold < 0.33 and nominal p-value < 0.05. A more permissive FDR threshold was selected for response genes vs protection genes to reflect reduced statistical power comparing protection status within a timepoint vs comparing gene expression for all samples over a time interval.
Gene-set enrichment analysis
Previously published coherent blood transcriptional modules (BTMs) published by Li et al. [
22], Chaussabel et al. [
23], and the MsigDB Hallmark collection [
24] were used for gene set enrichment analysis (GSEA) [
25] of whole blood RNAseq profiles. A total of 656 BTMs were tested. Samples were grouped by sample time interval (as above), study, and protection status. Genes were ranked by calculating the average change in normalized expression for each gene over each time interval. These rankings were used as input to calculate GSEA normalized enrichment scores (NES) with accompanying p-values, nominal and FDR adjusted. GSEA was performed using the R fgsea package [
26], with 10,000 random permutations. GSEA NES that did not pass the significance cut-off FDR < 0.05 were set to 0 for all further analysis. Correlations between sub-group module NES profiles were assessed using Spearman’s rho.
Spline curve fitting
Gene-averaged responses for each BTM were calculated using the 25% trimmed mean of BTM gene expression per sample. Responses were made relative to day 0 for each subject by subtracting the subject day 0 BTM expression from subsequent days. A spline curve was fit to all IMRAS and BSPZV1 samples at all time points for each BTM, using the R smooth.spline function with 3 degrees of freedom. The 99% CIs around each spline were calculated using 500 bootstrap replicates, taking approximately 2/3 of samples for each replicate. To calculate CI deviation from the 0 response line, lower and upper 99% CIs were assessed at each time point measured, i.e., days 1, 2, 3, and 7. The BTM response was considered significant if both lower and upper CIs were either above or below the 0 response line and the magnitude of the difference was defined as: if upCI < 0, upCI else lowCI.
Discussion
Human malaria correlates of protection/response to PfRAS vaccination have been difficult to identify. This may be due to the mechanisms of PfRAS-mediated protection because PfRAS vaccination results in a truncated liver stage parasite development, with no subsequent blood stage. Presumably, key protective immune processes happen at the site of infection, in the liver, where direct measures of responses are limited. This study represents the first time, to the authors’ knowledge, that common transcriptional correlates of RAS vaccination have been identified in multiple cohorts. This was done via analysis of whole blood RNAseq after vaccination, which reflects systemic immune responses.
The time-points examined here, up to 1 week post vaccination, were too early to capture the adaptive immune response to vaccination. Therefore, it is likely that the shared response identified here represents an effective innate immune response capable of presenting antigen and kickstarting the adaptive response. Whole blood RNAseq can detect systemic responses to PfRAS vaccination but is limited by the fact that sporozoite antigen specific cells represent a small minority of circulating blood cells. Additionally, whole blood RNAseq reflects the transcriptional state of every circulating leukocyte, and it is not possible to definitively link changes in expression to immune cell populations. Neither can it be determined whether changes in transcription reflect expansions or decreases in numbers of immune cells, or changes in transcriptional state of specific cell types. Despite these limitations, modules identified suggest an initial innate immune response followed by upregulation of cell cycle BTMs which may reflect rapid innate detection of PfRAS, followed by immune cell activation and proliferation. That the majority of the 14 shared protection associated genes identified were also associated with innate immunity or cell cycle suggests that deviations from the coordinated cross-study response programme are associated with non-protection. Recent work by Tran et al
. [
27] on whole sporozoite vaccination by administration of infectious sporozoites under chloroquine prophylaxis observed changes in T-cell associated BTMs in protected individuals 3 weeks after initial vaccination. This suggests that an extended series of cross-study comparable time-points may reveal adaptive immune associated responses. Unlike this study, they did not see significant changes in BTM expression within 1 week, however their study design did not include any sampling within 1–5 days post vaccination. They did observe increase of inflammatory/interferon and DC activation processes specifically in non-protected individuals 9 days after the third CPS immunization. Despite the difference in timescale, this shares some similarity with the IMRAS study, where over-induction of interferon is associated with non-protection. Recent work in a mouse model also supports the hypothesis that excessive type-I interferon inhibits the production of malaria-specific IFNγ producing CD8+ T cells [
13]. In contrast, transcriptional profiling of RTS,S vaccinees revealed multiple BTMs correlated with immunogenicity and protection within 1 week of vaccination, including cell cycle and inflammatory response BTMs [
28]. In contrast to the results seen after whole-sporozoite immunization, expression of interferon-associated and inflammatory-associated genes was positively associated with protection after RTS,S vaccination, and this is especially evident during the week after the second immunization. This may reflect differing pathways to protection induced by whole-sporozoite and subunit vaccines.
Stark differences in protective efficacy were observed between IMRAS, where 55% of study subjects were protected after intentional suboptimal immunization, and BSPZV1 where only 20% of the subjects were protected from CHMI. Although 20% of BSPZV1 subjects overall were protected, the RNAseq analysis was performed on a subset of BSPZV1 participants in which 36% were protected (Table
1). Unlike vaccination with PfRAS, it has long been observed that repeated natural malaria infection does not lead to sterilizing immunity [
29‐
34]. Reasons for this are unclear; however, it may be associated with low doses of sporozoites delivered in natural infections or active immune evasion strategies mounted by the parasite. Indeed, it has been observed that blood stage malaria inhibits or actively dysregulates the development of effective CD8+ T-cell and antibody-mediated liver stage immunity [
35,
36]. Previous immune exposure may be a factor in the observed reduction in protective efficacy of PfRAS vaccination in BSPZV1 vs IMRAS. Thus, it is plausible that the functional pathways leading to antigen presentation and adaptive immune priming may operate using different mechanisms in both trials. In other words, the different delivery method, dose, and previous malaria experience of BSPZV1 participants may have led to a qualitatively different, or non-naive, immune response to PfRAS in BSPZV1 compared to that observed in IMRAS. Therefore, true correlates of protection may exist that are distinct to both studies and would not be captured by this joint analysis.
Host intrinsic factors, e.g., genetic differences, and extrinsic factors, e.g., co-morbidities, microbiome, and general immune status, could also have contributed to the differences between IMRAS and BSPZV1. However, reduced levels of PfRAS-induced protection have been previously observed in malaria-experienced vs malaria-naive adults [
9] suggesting that this issue is not specific to these two study cohorts. Intriguingly, IMRAS showed increased expression of inflammation associated BTMs pre-immunization, compared with BSPZV1. This may influence the induction of inflammatory responses post RAS vaccination, and a more effective initial innate response may partially explain better protection in IMRAS.
However, other differences between the two trials are likely to be associated with protection status. IMRAS participants received PfRAS vaccination and PfSPZ infection via mosquito bite, while BSPZV1 used cryopreserved parasites administered intravenously. A comparison of protection vs non-protection does not capture any potential immune response to mosquito bites, independent from PfRAS, in the IMRAS cohort. Previous work suggest that mosquito saliva can affect T cell and NK cell populations potentially up to 7 days post-bite [
37]. The doses of PfRAS are very difficult to compare between trials, as IMRAS PfRAS doses were measured in terms of numbers of bites from infected irradiated mosquitos (~ 200 per vaccination), while BSPZV1 injected precise numbers of cryopreserved PfRAS. Another potential consequence of cryopreservation may be reduced PfRAS viability. Additionally, cryopreserved parasites were injected directly into the circulation while mosquito bites deliver PfRAS into the skin. Altogether, these differences would have resulted in differences in the number of hepatocytes that were infected by PfRAS during vaccination, and the number of PfRAS cleared by the innate immune system without reaching the liver.
While this study revealed common and distinct differences in whole blood gene expression patterns that correlate with protection and non-protection between these two studies, there are limitations to this analysis. This work relied on two study cohorts, with small numbers of study participants, comprising a total of 33 individuals to draw our conclusions. In addition, this analysis is limited to systemic responses the time period shortly following prime RAS vaccination. While consistent cross-study responses were identified, differing mechanisms of non-protection in each study were observed that would ideally be validated by further RAS cohorts. To validate these observations, it would be expected that malaria-naive RAS cohorts behave similarly to IMRAS and malaria experienced cohorts behave similarly to BSPZV1.
Acknowledgements
We would like to acknowledge all the study participants who made this work possible. We thank the NMRC Clinical Trials Center that conducted the IMRAS clinical trial and the NMRC Malaria Department Clinical Immunology Laboratory staff, led by Dr. Martha Sedegah, that processed/cryopreserved clinical samples. The study protocol for the IMRAS Trial was approved by the NMRC and WRAIR Institutional Review Boards, in compliance with all applicable Federal Regulations governing protection of human subjects. All study subjects gave written informed consent. The views expressed in this article are those of the authors and do not necessarily reflect the official policy or position of the Department of the Navy, the Department of the Army, the Department of Defense, or the U.S. Government.
JEE was an active duty military personnel at the time she contributed to this work. The work was prepared as part of official government duties. Title 17 U.S.C. 1105 provides that ‘Copyright protection under this title is not available for any work of the United States Government.’ Title 17 U.S.C. 1101 defines a U.S. Government work as a work prepared by a military service member or employee of the U.S. Government as part of that person’s official duties.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.