Distinct transcriptomic signatures define febrile malaria depending on initial infective states, asymptomatic or uninfected

This study was conducted in Junju, a moderate to high malaria transmission region in Kilifi County, Kenya. This region experiences two rainy seasons annually, the long rains in April to July and the short rains in October to November, which are characterised by increased malaria transmission [28]. The samples were collected from a cohort of 425 children living in Junju who were recruited at birth and followed up weekly by active surveillance until the age of 15, where P. falciparum malaria was diagnosed (using a rapid diagnostic test (RDT) and confirmed with microscopy) and treated [29]. Annual cross-sectional blood surveys were conducted in this cohort before the long rains from 2007 to 2018 and the children were defined as uninfected, asymptomatic, febrile malaria and non-malarial fever.

Uninfected children were defined as being parasite negative by RDT and microscopy. Asymptomatic infections were parasite positive by RDT, and confirmed by microscopy, with an axillary temperature < 37.5 °C and no symptoms of fever, and with no febrile malaria episode within the month up to the survey or the subsequent 7 days after the date of the survey [13, 16]. Sub-microscopic asymptomatic infections were excluded as they yielded insufficient RNA material for RNAseq during optimization assays. For the first-febrile malaria infections from individuals who were initially uninfected their parasitemia was defined as ≥ 2500 parasites/µl by microscopy and a tympanic temperature ≥ 37.5 °C based on definitions previously described for this cohort [27] while any parasitaemia by microscopy and a tympanic temperature ≥ 37.5 °C was considered for the first-febrile infections from those who were initially asymptomatic. Only microscopy-positive samples were included in this study. A total of 2801 uninfected, 510 asymptomatic and 3205 febrile samples collected between 2009 and 2017 were selected for subsequent analysis. About 154 paired samples from children who were asymptomatically infected on the survey and subsequently experienced febrile infections were identified for parasite and PBMC transcriptomic and proteomic analysis but only 21 pairs were available in the biobank. Additionally, 22 paired uninfected and corresponding febrile infections were available for PBMC transcriptomic and proteomic analysis out of 1500 pairs identified in the database (Fig. 1).

Blood samples were drawn by venipuncture from all the participants during the cross-sectional bleed and at their follow-up febrile malaria episode. The samples were collected in sodium citrate-containing cell preparation tubes (Vacutainer CPT Tubes, BD) and transported to the laboratory where they were separated into peripheral blood mononuclear cells (PBMCs) and red blood cell (RBC) pellets, harvested and processed before storage in liquid nitrogen (LN₂). PBMCs were isolated using the Lymphoprep™ density gradient solution, carefully collected using a wide mouth Pasteur pipette and then washed twice with RPMI media before storage in LN₂. Each sample was then separated into two components, one for transcriptomics and the other for proteomics.

At the time of the study, infected erythrocytes from the 21 asymptomatic-febrile sample pairs were retrieved and thawed using decreasing concentrations of NaCl (16% (w/v), 1.2% (w/v) and 0.9% (w/v)) to remove the glycerolyte storage medium. The samples were digested using 0.02% (w/v) saponin, and the parasites were separated from RBC components by centrifugation. RNA was extracted using the ISOLATE II RNA Mini Kit (Bioline) and eluted into 40 µl RNAse-free water. The RNA quality, concentration and purity were assessed using NanoDrop 2000c (Thermo Scientific). mRNA was enriched using the NEBNext™ Poly(A) mRNA magnetic isolation module, and first strand cDNA was synthesized using Superscript III reverse transcriptase (Invitrogen), while the second strand was synthesized using the NEBNext™ RNA Second Strand Synthesis Module (New England Biolabs). Next, 13.5 µl of cDNA was prepared into libraries using the NEBNext™ Ultra II FS DNA library preparation kit (New England Biolabs) and amplified in 17 cycles using the KAPA Library Amp Primer mix (KAPA Biosystems) to increase yield. The Agilent DNA 1000 chips on the Agilent TapeStation 2200 system were used to assess the quality and concentration of the libraries, which were then quantified using the KAPA Library Quantification Kit – Complete Kit (ABI Prism) and pooled into equimolar amounts. Paired-end sequencing was performed using the Illumina HiSeq 4000 platform at the Wellcome Sanger Institute (WSI), UK.

Raw sequence reads in fastq file format were assessed for quality and mapped to the reference P. falciparum (3D7 strain, PlasmoDB, release 55) genome using Kallisto’s v0.46.1 [28] default settings. Only genes with more than 2 counts per million in 10 or more samples were retained. Parasite age hours post-invasion (hpi) was determined using the maximum likelihood estimation method described in [29] with reference data from [30]. Proportions of parasite stages per sample were determined using a mixture model adapted from [31] using reference data from López-Barragán et al., (2011) [32]. Differences between stage proportions of paired samples were analysed using the Wilcoxon rank sum test. Pearson correlation was used to associate the proportion of rings and nonrings with parasitemia. Normalization was performed using the trimmed mean of M values (TMM) normalization in edgeR v3.38.1 [33] and unwanted variation factors were estimated using RUVSeq v1.30.0 [34].

RNA was extracted from host PBMCs from 21 asymptomatic-febrile pairs and 22 uninfected-febrile pairs of host PBMCs and libraries prepared using the same protocol as the parasite samples described above. Paired-end sequencing was performed on a NovaSeq 6000 (Illumina) system at the Advanced Sequencing Facility at The Francis Crick Institute, UK. The raw fastq files generated were processed using the RNA-Seq pipeline from the NF-core framework (v21.10.6) [35] using STAR (v2.7.10a) [36] aligner and RSEM (v1.3.1) [37] quantification and all other parameters as default. Samples were mapped to the human reference genome version GRCh38 release 95 obtained from Ensembl. Only samples with greater than 10 million aligned reads were considered. Hemoglobin genes described in [38] were removed from the analysis. Genes having 5 counts per million or higher in 13 (number of samples in the least popular group) or more samples were retained and normalized using TMM normalization. To determine differentially expressed genes among the uninfected, asymptomatic and febrile sample groups, gene analysis was performed by fitting a negative binomial generalized log-linear model in edgeR v3.38.1 [33]. The following matrix formula was applied to the edgeR model: ~ 0 + treatment + pair + batch. Here, ‘treatment’ aligned to either uninfected, asymptomatic or febrile, ‘pair’ indicated the pairing of samples per individual; and ‘batch’ represented the RNA processing batch. Differentially expressed genes were partitioned into four clusters with the least intracluster variation using K-means clustering and plotted using a heatmap. Additional pairwise analyses were conducted to identify differences between two sample groups. Genes in each cluster were used as input during functional overrepresentation analysis to identify enriched pathways using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets [39] in clusterProfiler v4.4.4 [40]. Pathways below a Benjamin-Hochberg adjusted P value cut of 0.05 were retained. Deconvolution of log-transformed and normalized counts was performed in CIBERSORT v1.04 [41] using the LM22 signature gene set as a reference to determine the proportion of each immune cell per sample [42]. The differences in cellular proportions among the four conditions were then examined.

Proteins were extracted from PBMCs by resuspending the pellet with 5 µl of 6 M UREA (Thermo Scientific). The protein samples were then adjusted with 50 mM triethylamonium bicarbonate (TEAB, Sigma-Aldrich) to 100 µl and the protein concentration was determined using the bicinchoninic acid (BCA) protein assay (Thermo Scientific). The protein samples were then reduced with 40 mM dithiothretol, alkylated with 80 mM iodoacetamide in the dark, and quenched with 80 mM iodoacetamide at room temperature, followed by digestion with1 µg/µl of trypsin [43]. Nine pools, each containing 9 samples and 1 control for batch correction, were prepared by combining 1 µl aliquots from each sample. The samples were pooled using a custom randomization R script. The pooled samples were then individually labelled using the Tandem Mass Tag (TMT) 10-plex kit (Thermo Scientific) according to the manufacturer’s instructions. One isobaric tag was used solely for the pooled samples and combined with peptide samples labelled with the remaining 9 tags. The labelled peptide pools were then desalted using P10 C18 pipette ZipTips (Millipore) according to the manufacturer’s instructions. Eluted peptides were dried in a Speedvac concentrator (Thermo Scientific) and resuspended in 15 μl loading solvent (98% H₂O, 2% acetonitrile, 0.05% formic acid). The peptides were then quantified using a Qubit Protein Assay Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. A standardized protein concentration of 5 µg was finally injected into the LC–MS/MS for analysis. The peptides were then loaded onto the liquid chromatograph, separated on reverse-phase analytical column of 75 µm x 50 cm C18 (Thermo Scientific) and measured using a Q Exactive Orbitrap mass spectrometer as described by [43]. To identify and quantify proteins, mass spectrometer output files were analyzed using MaxQuant software version 2.0.3.0 [44] by searching against the UniProt human proteome (downloaded on 10/06/2021) using the Andromeda search engine [45]. N-terminal acetylation and methionine oxidations were set as variable modifications while cysteine carbamidomethylation and TMT-10plex labelled N-terminus and lysine were set as fixed modifications. The false discovery rate (FDR) cut-off was set as 0.01 for both proteins and peptides with a minimum length of seven amino acids and was determined by searching a reverse database. Enzyme specificity was set as C-terminal to arginine and lysine with trypsin as the protease. Only up to two missed cleavages were allowed in the database search. Peptide identification was performed with an allowed fragment mass deviation maximum of 20 ppm (parts per million) and an initial precursor mass deviation maximum of 7 ppm. Default parameters for Orbitrap-type data were used. The pooled sample channels were used for batch correction. The 10-plex corrected reported ion intensity matrix was extracted from the protein group output file and used for downstream analysis. Proteins matching the reversed part of a decoy database, potential contaminants and proteins only identified by a modification site were excluded. Differential protein abundance analysis of the labelled sample intensities generated by MaxQuant was performed using PERSEUS v2.05.0 MaxQuant software (MaxPlanck Institute of Biochemistry, Martinsried, Germany) as described in [46]. The differentially expressed proteins were loaded in STRING version 11.5 (https://​string-db.​org/​) databases for protein–protein interaction and Gene Ontology (GO) functional analyses. Statistical significance was set at FDR < 0.05.

Between 2009 and 2017, the cohort’s distribution of febrile malaria cases throughout the year showed perennial malaria transmission, with at least 2 peak months per year. The overall trend in malaria cases was consistent, but there was a drastic drop in the prevalence in malaria cases in 2012 (Fig. 2). The proportion of P. falciparum positive asymptomatic infections during the cross-sectional bleed from 2009 to 2017 was highest in 2010 at 25.2% and lowest in 2017 at 4.2%.

Parasite RNA was successfully isolated from 16 asymptomatic (A) and febrile (F), AF paired samples from 12 females and 4 males. The participants had a mean age of 7.5 years (range: 0.8 – 13.4) during the asymptomatic P. falciparum infections and were followed for an average of 114.6 days (range: 22—235) before developing malaria symptoms. In contrast, though not significant, individuals who went from uninfected to febrile did so over an average of three months (96 days). The median parasitaemia was significantly lower during asymptomatic infections (1,720 parasites/ µl: range: 120 – 220,000) than during febrile infections (19,229 parasites/µl; range: 480 – 1,280,000) (P = 0.0204) (Additional file 1: Table S1). On average, approximately 15.4 million paired-end reads were generated per sample (range = 0.6 – 111 million reads). The median parasite age, as expressed in hours post-infection (hpi) and determined using the maximum likelihood estimation method [29], was statistically higher (P = 0.0004) in asymptomatic infections (median = 11.7 hpi, (inter-quartile range) IQR[10.48—11.9]) compared to febrile infections (median = 8.7 hpi, IQR(7.6—9.55) (Fig. 3A). The estimated proportions of each intraerythrocytic parasite stage were determined using a mixture model [31], revealing a significantly lower (P = 0.012) proportion of ring stage parasites in asymptomatic infections than in febrile infections (Fig. 3B). Similarly, a statistically significant (P = 0.0025) bias towards the early trophozoites was revealed in asymptomatic infections (Fig. 3B).

An all-gene expression principal component analysis (PCA) showed a clear distinction between febrile malaria and asymptomatic infections after correcting for library size, parasite blood stage development and unwanted variations (Fig. 4A). The proportion of the rings and non-ring stages versus parasitemia across all samples was determined whereby the proportion of non-ring parasites was significantly negatively correlated with parasitemia (r =—0.45; P = 0.01) (Fig. 4B). This indicates that RBCs infected with mature parasite stages circulate longer in low parasitemia.

A total of 49 PBMC samples from uninfected (n = 14), asymptomatic (n = 13) and febrile (n = 22) individuals were successfully sequenced. Of these, 20 were paired samples, i.e., 8 UF and 12 AF paired samples. The other nine samples were unpaired and comprised 6 uninfected, 1 asymptomatic and 2 febrile samples. The mean age of uninfected individuals (8.15 years) was not significantly different from that of asymptomatic individuals (7.36 years) (P = 0.56). There was no significant difference in parasitemia observed between the febrile infections ensuing from uninfected or asymptomatic infections (P = 0.504) ((Additional file 1: Table S1). On average, approximately 60 million reads mapped to each sample with a range of 11.5 – 374.4 million paired-end reads.

Differential expression analysis revealed 4311 differentially expressed genes (DEGs) (Additional file 1: Table S3). Principal component analysis (PCA) using DEGs separated the samples into febrile and non-febrile groups on PC1, which was associated with 21.64% of the variation, while PC2 (7.52%) partially separated U from A and the febrile infections into those who were initially asymptomatic (FA) and uninfected (FU), suggesting a transcriptional distinction in the febrile infections (Fig. 5A). Gene clustering using DEGs identified four gene clusters 1 to 4 (c1-c4). Overall, c1 (1506 genes) was upregulated in non-febrile children (A and U groups), while c4 (1325 genes) was upregulated in febrile children (FU and FA). There was a marked upregulation of c2 (552 genes) in the FA group and a contrasting pattern of expression in c3, with an upregulation of 928 genes in FU samples (Fig. 5B).

Gene ontology (GO, Fig. 5C-F) and KEGG analysis (Additional file 1: Table S4) produced similar results and were used to determine the predominant functions enriched in each cluster. GO terms for cytoplasmic translation, ribosomal activity, T cell activation and cell adhesion (Fig. 5C) predominated in c1, including genes associated with protein translation factors (EIF3D, EIF3E, EIF3G and EIF3K) and ribosomal proteins (RPL18A, RPL19, RPS12, RPS16, RPL35, RPL8, RPS11, RPS27A) and genes associated with T cell activation and adhesion (CD3D, CD3E, CD3G, CD4, CD96, CD247, CD5, ZAP70, LCK, CXCR4, CXCR5, CD74 and TBX21, GATA3), Figs. 6B and C. These data suggest that protein synthesis and T cell activation were downregulated in the febrile cases.

In contrast, c4 genes were enriched for GO terms involving immune effector processes, adaptive and innate immunity and cytokine pathways (TLR2, STAT3, IL6, IL6R, ANXA3, S100A11, CXCR1, C3AR1, LRG1, S100A8, S100A12, S100A9, JAK3) and regulation of immunity (CD274, CD276, IL1RN, IL10, CR1, CD55, IDO1, IL2RA, BST2) (Fig. 5F). These data show, as expected, that markers of immune activation, particularly neutrophil are upregulated in febrile infections (FA and FU).

The deconvolution analysis revealed some significant differences in the estimated proportions of nine cell types in PBMCs among the four clinical groups. A lower proportion of naïve B cells was observed in FA, while a higher proportion of plasma cells was observed in the FU group. The proportion of CD8⁺ T cells was greatest in U and A; on the other hand, that of naïve and activated memory CD4⁺ T cells was greatest in FA and FU. A higher proportion of activated mast cells was also observed in the FA group (Fig. 6).

Clusters 2 and 3 drive the separation of febrile infections based on whether they were initially from asymptomatic or uninfected individuals. Cluster 2 was enriched for GO terms involving immune effector processes, leucocyte migration, host defense and TNF production (Fig. 5D). The greatest differences in expression were in the FA group, particularly in genes associated with innate immune effector processes (TLR4, CD14, CD68 TGFB1, EOMES, CCR7, CCL3R1, CD84, TNFRSF1B, IL1B, IFNGR1). There were also some differences between the A and U groups, and a pairwise analysis of the two groups revealed an upregulation of several genes related to inflammatory responses (CCR7, CCL2, NOD2, SIRPA, TNFSF14, NFKB1) in the U group (S5 Table). The innate and inflammatory responses were strongly upregulated in FA compared to the FU group.

The GO terms enriched in c3 were nuclear division and chromosomal segregation, humoral responses, genes for pathways associated with cell cycle regulation (E2F5, CCNA2, CDC6, CDC14B, CCNB1, CDKN2C), and immunoglobulin, phagocytosis and complement (IGHM, IKC, IGHG4, IGHG2, C5) (Fig. 5E). There was a clear distinction in the gene expression of c3 genes between the FA and FU samples, where many genes were upregulated in the FU samples (Fig. 5B), indicating a relatively stronger humoral response. Pairwise analysis of the A and U groups revealed an upregulation of humoral responses (C1QB, IGHG4, IGKC, IGLC3, IGHV5-51, IGHV3-15, IGHV3-21) in the A group ((Additional file 1: Table S5).

A summary of the 10 most highly differentially expressed genes for the gene ontology terms, humoral responses, cytokines, TNF and T-cell activation in each group is shown in Fig. 7A, with a comparison of the normalized expression of a selection of relevant genes (Fig. 7B). Both the A and U samples show upregulated T-cell activation compared to the respective febrile samples. The FU samples had the strongest expression of immunoglobulin-related genes and humoral response whereas the FA samples had the greatest gene expression for cytokines and TNF production.

A paired analysis of PBMC samples from 16 FU pairs and 22 FA pairs was performed using TMT proteomics technology to examine differentially expressed proteins. A total of 318 proteins were identified (q-value < 0.1) after filtering out contaminant proteins. Differential protein expression analyses identified 35 proteins (Fig. 8A and Additional file 1: Table S6). Protein–protein interactions (PPI) analysis in the STRING database revealed significant enrichment of proteins in either the febrile or non-febrile groups (PPI enrichment P value < 0.001) (Fig. 8B). Consequently, 16 proteins were upregulated in the non-febrile samples and were enriched for protein-DNA complex and nucleosome (HIST1H1E, HIST1H2BC, H2AFJ, HIST1H4A) cellular components (Fig. 8C). The remainder were upregulated in febrile infections and were enriched for chemokine production (MPO, HP, S100A8, MMP9, ANXA1, DEFA1), neutrophil aggregation and chemotaxis (S100A8, S100A9, S100A12) and chronic inflammatory response (CAMP, S100A8) biological processes (Fig. 8C). Although a low number of proteins were detected, it indicates chromatin changes in asymptomatic and uninfected individuals, and innate immune responses in febrile infections.