Differential gene expression in cerebral malaria
Twelve matched pairs met the inclusion criteria for comparing cases of CM to uncomplicated malaria controls without a history of CM. CM cases experienced greater parasitemia (126,603 versus 33,408 parasites/μL, respectively, p-value = 0.02; Table 1). There was no statistically significant difference in hemoglobin levels for the comparison.
Transcripts of IL1R2, FKBP5, and MMP8 had the most significant association to CM in comparison to uncomplicated controls (logFC 4.41, 2.58, and 4.20, with FDR-adjusted p-values of 1.75E-11, 2.00E-09, and 2.73E-08, respectively; Fig. 1, Supplementary Table 1). 405 genes met the threshold for statistical significance in the comparison (Supplementary Data 1).
A Principal component analysis showing the separation of CM from UM along the first two principal components. B Volcano plot showing the significance of association from a quasi-likelihood F test based on the negative binomial distribution (red and blue FDR-adjusted p-value < 0.05) and effect size (green and red >1 absolute log2-fold change). C Heatmap of the top 20 differentially expressed transcripts, showing the clustering of associated genes from a quasi-likelihood F test (y-axis) and CM cases versus UM controls (x-axis). D Bar plot of the top 10 significantly associated gene ontology biological processes. Enrichment analysis was performed using the clusterProfiler package in R. Enriched terms were determined using a hypergeometric test, with p-values < 0.05 considered significant. Pathway names are represented on the Y-axis, and gene counts on the X-axis. The color spectrum represents the extent of statistical association, from red, p-value 1E-05, to blue, p-value 1E-04.
Gene Ontology (GO) pathway analyses revealed that T cell activation, positive regulation of cytokine production, and regulation of immune effector process were the top three biological processes for differentially expressed transcripts between CM and uncomplicated controls (Fig. 1D, Supplementary Table 2). Regulation of actin cytoskeleton and focal adhesion were among the eight significantly upregulated KEGG pathways for this comparison (Supplementary Table 3). No KEGG pathways were significantly downregulated in this comparison.
Differential gene expression in severe malarial anemia
Eight matched pairs met the inclusion criteria in comparing cases of SMA to controls with uncomplicated disease. Hemoglobin levels were significantly lower in SMA compared to controls (2.87 versus 8.41 g/dL, respectively, p-value = 1.05E-04; Table 1). There was no statistically significant difference in parasitemia for the comparison.
Transcripts of DAAM2, LTF, and ARG1 had the most significant associations with SMA in comparison to uncomplicated controls (logFC 4.83, 3.87, and 2.82, with FDR-adjusted p-values of 1.54E-03, 2.36E-03, and 4.97E-03, respectively; Fig. 2, Supplementary Table 4). 19 genes met the threshold for statistical significance in this comparison.
A Principal component analysis showing the separation of SMA from UM along the first two principal components. B Volcano plot showing the significance of association from a quasi-likelihood F test based on the negative binomial distribution (red FDR-adjusted p-value < 0.05) and effect size (green and red >1 absolute log2-fold change). C Heatmap of the differentially expressed transcripts, showing the clustering of associated genes from a quasi-likelihood F test (y-axis) and CM cases versus UM controls (x-axis). D Bar plot of the top 10 significantly associated gene ontology biological processes. Enrichment analysis was performed using the clusterProfiler package in R. Enriched terms were determined using a hypergeometric test, with p-values < 0.05 considered significant. Pathway names are represented on the Y-axis, and gene counts on the X-axis. Color spectrum represents the extent of statistical association, from red, p-value 6.5E-03, to blue, p-value 8E-03.
Gene Ontology (GO) pathway analyses revealed that negative regulation of cytokine production, tumor necrosis factor production, and killing of cells of other organisms were among the top biological processes for differentially expressed transcripts between SMA and controls (Fig. 2D, Supplementary Table 5). KEGG analyses showed that vitamin B6 metabolism, arachidonic acid metabolism, and metabolic pathways were among the 12 significantly upregulated pathways in comparing SMA to controls (Supplementary Table 6). No downregulated pathways met the threshold for significance in this comparison.
Differential gene expression in concurrent cerebral malaria and severe malarial anemia
Eight matched pairs met the inclusion criteria for comparing cases of concurrent CM and SMA to uncomplicated controls without a history of CM. Concurrent CM and SMA cases had significantly lower hemoglobin levels compared to controls (3.30 versus 8.11 g/dL, respectively, p-value = 2.15E-03; Table 1). There was no statistically significant difference in parasitemia in this comparison.
Transcripts of OLMF4, XAF1, and IL1R2 had the most significant association with concurrent CM and SMA in comparison to uncomplicated controls (logFC 6.90, −2.39, and 4.04, with FDR-adjusted p-values of 1.75E-05, 2.77E-03, and 2.77E-03, respectively; Fig. 3, Supplementary Table 7). 67 genes met the threshold for statistical significance in the comparison.
A Principal component analysis showing the separation of concurrent CM and SMA from UM along the first two principal components. B Volcano plot showing the significance of association from a quasi-likelihood F test based on the negative binomial distribution (red FDR-adjusted p-value < 0.05) and effect size (green and red >1 absolute log2-fold change). C Heatmap of the top 20 differentially expressed transcripts, showing the clustering of associated genes from a quasi-likelihood F test (y-axis) and CM cases versus UM controls (x-axis). D Bar plot of the top 10 significantly associated gene ontology biological processes. Enrichment analysis was performed using the clusterProfiler package in R. Enriched terms were determined using a hypergeometric test, with p-values < 0.05 considered significant. Pathway names are represented on the Y-axis, and gene counts on the X-axis. Color spectrum represents the extent of statistical association, from red, p-value 9E-06, to blue, p-value 3E-05.
Gene Ontology (GO) pathway analyses revealed that defense response to a symbiont, response to type 1 interferon, and cytokine-mediated signaling pathway were among the top ten biological processes for the comparison of cases with concurrent CM and SMA to uncomplicated controls (Fig. 3D, Supplementary Table 8). KEGG analysis revealed that ubiquitin-mediated proteolysis was among the three statistically significant upregulated pathways, and necroptosis was significantly downregulated when comparing concurrent CM and SMA to controls (Supplementary Table 9).
Differential gene expression between controls with and without a history of cerebral malaria
Sixteen matched pairs met the inclusion criteria in comparing controls with and without a history of CM. There was no statistically significant difference in age, hemoglobin levels, or parasitemia between controls with and without a history of CM (Supplementary Table 10). There were no significantly associated transcripts at the FDR corrected level for differential gene expression analysis in comparing controls with and without a history of CM.
Differential gene expression between severe malaria subtypes
21 cases of CM to 20 cases of SMA met the inclusion criteria for an unpaired comparison of gene expression. CM cases had significantly higher hemoglobin (7.62 versus 3.56 g/dL, p-value 1.8E-12) and parasitemia levels (145,634 versus 61,859 parasites/μL, p-value 0.022; Table 2). There was no statistically significant difference in gender or age. There were 165 differentially expressed genes between the two groups, 137 of which were expressed at higher levels in SMA. H1-2, H1-0, and SCL2A1 were the transcripts with the most significant association for genes expressed at a higher level in SMA (logFC 1.86, 2.37, and 2.96; p-values 4.15E-04, 1.55E-03, and 1.55E-03, respectively; Fig. 4, Supplementary Table 11). GBP3, H2BC21, and SELENOT were the transcripts with the most significant association for genes expressed at a higher level in CM (logFC 1.26, 1.31, and 0.79; p-values 8.06E-03, 8.21E-03, and 9.47E-03, respectively; Supplementary Table 12). Gene Ontology (GO) pathway analyses revealed that cellular response to toxic substance and erythrocyte differentiation were among the top biological processes for the comparison of CM to SMA (Fig. 4D, Supplementary Table 13). KEGG analyses revealed that GABAergic synapse and cortisol synthesis and secretion were among the significantly associated pathways upregulated in SMA, while taurine and hypotaurine metabolism were the only significantly associated pathways upregulated in CM (Supplementary Table 14).
A Principal component analysis showing the separation of CM from SMA along the first two principal components. B Volcano plot showing the significance of association from a quasi-likelihood F test based on the negative binomial distribution (red and blue FDR-adjusted p-value < 0.05) and effect size (green and red >1 absolute log2-fold change). C Heatmap of the top 20 differentially expressed transcripts showing the clustering of associated genes from a quasi-likelihood F test (y-axis). D Bar plot of the top 10 significantly associated gene ontology biological processes. Enrichment analysis was performed using the clusterProfiler package in R. Enriched terms were determined using a hypergeometric test, with p-values < 0.05 considered significant. Pathway names are represented on the Y-axis, and gene counts on the X-axis. The color spectrum represents the extent of statistical association, from red, p-value 2E-03, to blue, p-value 0.018.
21 cases of CM and 10 cases of concurrent CM and SMA met the inclusion criteria for an unpaired comparison of gene expression. CM cases had significantly higher hemoglobin levels (7.62 versus 3.21 g/dL, p-value 1.3E-08; Table 2). There were no statistically significant differences in parasitemia, gender, or age. Unpaired analysis comparing cases of CM to cases of concurrent CM and SMA yielded 19 differentially expressed genes between the two groups, four of which were higher in CM. MPO, TBC1D14, and ELANE were among the transcripts with the most significant association for genes expressed at a higher level in concurrent CM and SMA (logFC 3.75, 2.77, and 3.16; p-values 2.03E-03, 6.58E-03, and 0.020, respectively; Supplementary Fig. 1, Supplementary Table 15). SUB1, SSR3, TMED2, and JCHAIN were the transcripts with the most significant association for genes expressed at a higher level in CM (logFC 0.89, 1.27, 0.95, and 1.82; p-values 0.020, 0.028, 0.034, and 0.038, respectively; Supplementary Table 15). KEGG analyses revealed that TGF-beta signaling pathway, apelin signaling pathway, and steroid biosynthesis were among the significantly associated pathways upregulated in concurrent CM and SMA, while TNF signaling pathway and GnRH signaling pathway were among the significantly associated pathways upregulated in CM (Supplementary Table 16).
20 cases of SMA and 10 cases of concurrent CM and SMA met the inclusion criteria for an unpaired comparison of gene expression. There was no statistically significant difference in hemoglobin levels, parasitemia, sex, or age (Table 2). Unpaired analysis comparing cases of SMA to cases of concurrent CM and SMA yielded four differentially expressed genes between the two groups, all of which were higher in the concurrent CM and SMA group; GNA13, CREBBP, and WDR74 were among the transcripts with the most significant association for genes expressed at a higher level in concurrent CM and SMA (logFC 3.82, 3.02, and 3.67; p-values 0.025, 0.031, and 0.049, respectively; Supplementary Table 17). KEGG analyses revealed that TGF-beta signaling pathway and steroid biosynthesis were among the significantly associated pathways upregulated in concurrent CM and SMA (Supplementary Table 18).
Proteomics in cerebral malaria
Serum proteins were assayed for 14 age-, sex-, and ethnicity-matched pairs of CM and uncomplicated malaria controls without history of CM, 11 of which overlapped with the transcriptomic analyses. CM cases had higher average parasitemia (107,330 versus 28,670 parasites/μL, p-value 0.039), and there was no statistically significant difference in hemoglobin (Supplementary Table 19). After removing data with ambiguous assignments, 180 of 212 proteins remained for comparison, 43 of which were statistically significant (Fig. 5A, Supplementary Tables 20 and 21). Of the 28 proteins that were higher in CM, TIMP1 was the most significant (logFC 1.80, p-value 1.71E-03). Of note, CD14, VCAM-1, and ICAM-1 levels were also significantly higher in CM cases (Supplementary Table 20). SERPIND1 had the most significant association for proteins with lower levels in CM than controls (logFC −0.55, p-value 1.65E-03). APOM, C3, and C1QB were also among the proteins expressed at significantly lower levels in CM (Supplementary Table 21). Response to hypoxia and response to endogenous stimulus were among the most significantly upregulated GO biological processes in CM (Supplementary Table 22). Antimicrobial humoral immune response mediated by antimicrobial peptide, granulocyte chemotaxis, positive regulation of TGFβ1 production, and positive regulation of epithelial cell apoptotic process were significantly downregulated in CM (Supplementary Table 23). Kegg pathway analysis revealed that the complement and coagulation cascades were the only statistically significant pathways, and it was downregulated in CM.
Volcano plot showing significance of association assessed with a two-tailed paired t-test (red and blue FDR-adjusted p-value < 0.05) and effect size (green and red >1 absolute log2-fold change) in comparisons of cerebral malaria to uncomplicated malaria without a history of cerebral malaria in A proteomic and B metabolomics analyses.
Of the 180 assayed proteins, 91 had sequenced transcripts that passed quality controls in the CM comparison to uncomplicated controls. Six of the 405 transcripts that were significantly associated with CM to uncomplicated controls had protein level data. CD14 was the only significantly upregulated transcript in CM (logFC 0.85, p-value 0.044) that was also significantly upregulated in the proteomic analysis (logFC 0.64, p-value 2.00E-03).
Metabolomics in cerebral malaria
Metabolomic profiles were ascertained for the same set of 14 CM cases and 14 age-, sex-, and ethnicity-matched uncomplicated malaria controls without a history of CM as in the proteomic analysis. After removing ambiguous assignments, 147 lipids and metabolites met the multiple testing-corrected level of significance. Nervonic acid (logFC 0.89, p-value 0.026), pipecolic acid (logFC 1.76, p-value 0.044), cortisol (logFC 1.59, p-value 3.61E-03), and mannitol (logFC 7.43, p-value 0.014) were among the metabolites with significantly higher levels in CM (Fig. 5B, Supplementary Table 24, Supplementary Data 3). Paracetamol was also higher in CM cases (logFC 7.37, p-value 0.035). Arginine and linoleic acid were among the metabolites with lower levels in cases of CM compared to uncomplicated controls (logFC −1.04 and −0.79, p-value 0.038 and 0.017, respectively; Supplementary Table 25). Pathway analysis of serum metabolites in CM compared to uncomplicated malaria controls revealed that phenylalanine, tyrosine, and tryptophan biosynthesis, and alanine, aspartate, and glutamate metabolism were among significantly upregulated pathways in cerebral disease (Supplementary Table 26).
Differential gene expression in cerebral malaria compared to controls with a history of cerebral malaria
Eight matched pairs met the inclusion criteria for comparing CM cases to uncomplicated malaria controls with a history of CM. Parasitemia was not significantly higher in CM cases than controls with a history of CM (189,211 versus 37,837 parasites/μL, respectively, p-value = 0.10; Supplementary Table 27A). There was no statistically significant difference in hemoglobin levels for the comparison.
Transcripts of IL1R2, CD163, and FKBP5 had the most significant association in the comparison between CM and controls with a history of CM (logFC 4.13, 3.36, and 2.18 with FDR-adjusted p-values of 1.51E-05, 2.77E-04, and 8.34E-04, respectively; Supplementary Fig. 3A). 72 genes met the threshold for statistical significance in this comparison (Top 20 are listed in Supplementary Table 28). 57 statistically significant transcripts overlapped in comparisons of CM to controls with and without a history of CM. IL10 was the most significantly upregulated gene in the comparison of CM to uncomplicated controls with a history of CM (logFC 3.49 and FDR-adjusted p-value 6.12E-03) that did not associate with uncomplicated controls without a history of CM (logFC 1.16 and FDR-adjusted p-value 0.12).
Response to hydrogen peroxide, cellular response to reactive oxygen species, and cellular response to chemical stress were among the top ten biological pathways in GO analysis when comparing CM to controls with a history of CM (Supplementary Fig. 4A).
Differential gene expression in severe malarial anemia compared to controls with a history of cerebral malaria
Six matched pairs met the inclusion criteria in comparing cases of SMA to controls with a history of CM. Hemoglobin levels were significantly lower in SMA compared to controls with a history of CM (3.35 versus 9.45 g/dL, respectively, p-value 5.27E-05; Supplementary Table 27B). There was no statistically significant difference in parasitemia for the comparison.
No genes reached the threshold for statistical significance when comparing SMA to controls with a history of CM.
Differential gene expression in concurrent cerebral malaria and severe malarial anemia versus controls with a history of cerebral malaria
Four matched pairs met the inclusion criteria for comparing cases of concurrent CM and SMA to controls with a history of CM. Concurrent CM and SMA cases had significantly lower hemoglobin levels compared to controls without and with a history of CM (2.97 versus 8.77 g/dL, respectively, p-value 0.023; Supplementary Table 27C). There was no statistically significant difference in parasitemia in this comparison.
Transcripts of MPO, IFI44L, and ELANE had the most significant association in the comparison between concurrent CM and SMA to controls with a history of CM (logFC 8.77, −6.49, and 9.93 with FDR-adjusted p-values of 3.52E-05, 1.57E-04, and 1.57E-04, respectively; Supplementary Fig. 3B). 67 genes met the threshold for statistical significance in this comparison (Top 20 transcripts are listed in Supplementary Table 29). 26 statistically significant transcripts overlapped in comparisons of concurrent CM and SMA to controls with and without a history of CM.
Gene Ontology (GO) pathway analyses revealed that defense response to a symbiont, response to type 1 interferon, and regulation of cytokine-mediated signaling pathway were among the top ten biological processes for the comparison of cases with concurrent CM and SMA to controls with a history of CM (Supplementary Fig. 4B).
Deconvolution
Deconvolution fractions of T cells, B cells, and neutrophils did not significantly associate with any of the severe malaria outcomes when compared to controls without a history of CM, and only SMA compared to T cell proportion was statistically significant in comparisons to controls with a history of CM (p-value 5.50E-03, Supplementary Table 30). Principal component analyses for comparisons of severe malaria subtypes to controls without a history of CM with and without adjustment for deconvolution are presented in Supplemental Fig. 5A–F. Adjustment for deconvolution lowered the number of genes passing the FDR threshold from 405 to 145 in the CM versus controls without a history comparison, and the top three associated genes without adjustment were significant after adjusting for B cell, T cell, and neutrophil proportions (Supplementary Table 31). Adjustment for deconvolution lowered the number of significantly associated genes from 19 to 9 for the SMA versus controls without a history comparison, and the top three genes associated without adjustment for cell proportions remained significant (Supplementary Table 32). Adjustment for deconvolution lowered the number of genes passing the FDR threshold from 67 to 20 for the concurrent CM and SMA versus controls without a history comparison, and two of the top three associating genes without the adjustment for cell proportions (OLFM4 and XAF1) remained significant after (Supplementary Table 33).
Adjustment for deconvolution lowered the number of genes passing the FDR threshold from 72 to 1 in the CM versus controls with a history of CM comparison, and 67 to 43 for the concurrent CM and SMA versus controls with a history of CM comparison (Supplementary Table 34A, B, respectively). No genes were significant in comparing SMA to controls with a history of CM, with and without adjustment for deconvolution.
Deconvolution analyses adjusting for T cell, B cell, and neutrophil proportions did not significantly alter the clustering of samples, although it did lower the number of significant genes in each comparison. However, we believe that the robustness of our findings to different cell proportions further validates our findings.
