Genome-Scale Identification of Essential Metabolic Processes for Targeting the Plasmodium Liver Stage

Abstract

Plasmodium gene functions in mosquito and liver stages remain poorly characterized due to limitations in the throughput of phenotyping at these stages. To fill this gap, we followed more than 1,300 barcoded P. berghei mutants through the life cycle. We discover 461 genes required for efficient parasite transmission to mosquitoes through the liver stage and back into the bloodstream of mice. We analyze the screen in the context of genomic, transcriptomic, and metabolomic data by building a thermodynamic model of P. berghei liver-stage metabolism, which shows a major reprogramming of parasite metabolism to achieve rapid growth in the liver. We identify seven metabolic subsystems that become essential at the liver stages compared with asexual blood stages: type II fatty acid synthesis and elongation (FAE), tricarboxylic acid, amino sugar, heme, lipoate, and shikimate metabolism. Selected predictions from the model are individually validated in single mutants to provide future targets for drug development.

Keywords: Plasmodium berghei; Plasmodium liver stage; amino sugar biosynthesis; fatty acid biosynthesis; fatty acid elongation; genome-scale knockout screen; genome-scale metabolic model; malaria; metabolic model; metabolic network.

Graphical abstract

Figure 1

Barseq Identifies Pre-erythrocytic Phenotypes (A) Schematic showing the gene knockout (KO) screen to identify mosquito-liver stage (M-L) phenotypes. P. berghei schizonts were transfected with pools of barcoded PlasmoGEM knockout vectors and parasites selected by drug-treatment in mice. Infected mice, from which blood was sampled for barcode sequencing (B1), were used to infect mosquitoes. Midguts (MG) and salivary glands (SG) of infected mosquitoes were then sampled, and salivary gland sporozoites were collected from mosquitoes to infect mice by i.v. injection. Blood from sporozoite-infected mice (B2) was also collected for barcode sequencing. Barcode counts determined by sequencing PCR amplicons were used to determine the relative abundance of each gene knockout parasite at the life-cycle transitions shown. (B) Abundance of gene KOs at different life-cycle stages in a pilot screen, shown relative to their initial abundance at B1. Genes included in the pilot screen are shown in Table S1. Error bars represent standard deviations.

Figure 2

Impact of Sexual Reproduction and Ploidy on the Transmission of KO Alleles (A) Schematic illustrating ploidy changes during sexual and mosquito stages (adapted with permission from Lee et al., 2014). (B) Illustration of inheritance where KO of gene a leads to a strong reduction in fertility in both sexes. Reduced transmission (red) of a^- from less fertile gametes is not rescued (dotted arrows) by cross-fertilization with a⁺ parasites (solid arrows), leading to much reduced inheritance of the a^- allele. The line graph displays screen data from known fertility genes showing strong reductions of the corresponding barcode (strongly negative log₂FC) among midgut (MG) oocysts. (C) As in (B), but assuming a sex specific fertility phenotype for gene b, allowing the b^- alleles to be transmitted effectively by the fertile sex. The line graph shows real data for genes with known functions, illustrating how the expected log₂FC or −1 is barely noticeable. (D) Similar illustration for a hypothetical gene c with known function in ookinete or oocyst development. Inheritance of c^- allele may be almost unhindered due to heterozygous rescue. Real data are plotted for genes whose homozygous disruption is known to block ookinete development or infectivity. Error bars in the line graphs shown in (B), (C), and (D) show standard deviations from three replicate transmissions of the same mutant pool.

Figure 3

M-L Screen Gene Knockout Phenotypes at the Liver Stage Show a Bias toward Genes with Predicted Metabolic Function (A) Ranked, normalized log₂FC values at each stage transition and for all genes with data. SG-B2 data were corrected for blood stage fitness, and an apparent increase in some mutants (right end of distribution) results from some overcompensation for slow growth. Error bars show standard deviations. (B) Pie charts show the distribution of phenotypes (see C for legend). (C) SG-B2 phenotypes shown separately for genes that at the blood stage (Bushell et al., 2017) grow slowly (yellow) or are indistinguishable from wild type (green). (D) Plots showing the blood-stage (inner ring) and liver-stage (SG-B2, outer ring) phenotypes for genes pertaining to specific metabolic subsystems (upper row) or GO biological processes (lower row). Liver-stage phenotypes for genes are clustered according to their corresponding blood-stage phenotype. The association of genes to metabolic subsystems is based on iPbe. GO biological process data, where available, are displayed in Table S2.

Figure S1

Phenotype Assignment and Statistical Analysis, Related to Figure 1 (A) A scheme showing how phenotypes of not reduced, no power or reduced were assigned to each gene knockout at each stage transition. This was based on the normalized (and blood stage fitness-corrected for SG-B2) relative change in abundance within the pool (Log2-FC) and associated standard deviation. A Log2-FC (−2XSD) value was calculated by the subtraction of 2xSD from the diff value. A Log2-FC (+2XSD) value was calculated by the addition of 2xSD to the diff value. Genes with a “not reduced” phenotype have a Log2-FC (−2XSD) value > -1. Genes with a no power phenotype have a Log2-FC (+2XSD) > -1. Genes with a reduced phenotype have a Log2-FC (+2XSD) value of < -1. The two “not reduced” bars serve to illustrate how both a small effect size or a high variance can lead to a conservative phenotype call of “not reduced” where the mean remains close to 1. While the first “reduced” bar shows a clear reduced phenotype, the second “reduced” bar and its comparison with the “no power” bar shows how at a given log₂FC, variance can determine whether a low stage transition rate is called “reduced” or “no power.” (B) Mean relative abundance of all mutants in B1 and MG samples. A high level of correlation shows representative sampling by the mosquito for all but the least abundant mutants. Color-coding of phenotypes showing that underrepresented mutants lack statistical power to make a phenotype call. (C) Violin plots showing that phenotypes at the B1-MG transition were assigned preferentially for the well-represented mutants in the B1 sample, which were sampled accurately by the mosquito.

Figure S2

Combined Experimental and Computational Workflow to Study Blood and Liver Phenotypes and Their Mechanistic Origin using iPbe. Description of the iPbe Model and Essentiality Predictions, Related to Figure 4 (A) Workflow diagram showing how data form the experimental screening platform are integrated to study blood and liver phenotypes and their underlying mechanisms and to develop a more comprehensive metabolic model following the cycle of systems biology. (B) Distribution of metabolic enzymes in iPbe. (C) Metabolic subsystems in iPbe. (D) Relation between genes predicted as essential in iPbe, iPbe-blood and iPbe-liver. (E) Contingency matrix for gene essentiality predictions and the liver stage M-L screen phenotypes compared with iPbe liver. (F) Contingency matrix as for (E) but compared with the general iPbe model.

Figure 4

The PhenoMapping Workflow and Degree of Agreement for Metabolic Subsystems in iPbe-Liver with the M-L Screen (A) Illustration of the PhenoMapping workflow for the integration of organism- and context-specific information into the genome-scale iPbe metabolic models. Context-specific information denotes life-cycle stage-specific processes, such as gene expression, as well as environmentally specific factors, such as substrate availability. Metabolic tasks are at the interface between organism- and context-specific information. The production of molecules, such as amino acids, fatty acids, nucleotides, etc., is required for growth independent of the context, but the ratios in which they are required might change with the growing conditions or life stage. See STAR Methods and Table S4 for a detailed description of iPbe. (B) Degree of agreement (DoA) between the gene essentiality predictions in iPbe-liver and the experimental phenotypes at the SG-B2 transition. Metabolic subsystems are ranked by level of agreement. Numbers show genes with screen data per subsystem (needs to be >1 for inclusion).

Figure 5

Mutations FASII, Lipoate Synthesis, and Biotin Metabolic Pathways Affect Liver-Stage Development (A) Pathway maps for FAS II, lipoate metabolism, and biotin metabolism in the Plasmodium apicoplast. See Table S4 for gene IDs, enzyme functions, and reactions. Pep, phosphoenolpyruvate; Pyr, pyruvate; Ac-CoA, acetyl-CoA; Mal-CoA, malonyl-CoA; Mal-ACP, malonyl-[acp]; Acetoac-ACP, acetoacetyl-[acp]; Octanoyl-ACP, octanoyl-[acp]; Octanoyl-E2, protein N6-(octanoyl)lysine; Lipoyl-E2, protein N6-(lipoyl)lysine. (B) Schematic representation of phenotypes of single knockout (KO) mutants. Green, phenotype not significantly different from wild type (WT) parasites. Red, phenotype significantly different (>2-day delay in pre-patent period). (C) Size of 250 cultured EEFs (48 hpi) per mutant; median and interquartile ranges are shown in red. ^∗ = p < 0.05 by Kruskal-Wallis test. (D) Relative maturation of EEFs measured as conversion of infected host cells to detached cells at 48 hpi. Error bars show standard deviations from 8 biological replicates (for PDH-E2) or 3 biological replicates (all other mutants). The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (^∗∗p ≤ 0.01; ^∗∗∗p ≤ 0.001). (E) The number of mice that developed blood-stage infections after injection of 5,000 mutant sporozoites and the mean delay (range) in pre-patency compared to mice infected with WT sporozoites. ^∗, gene KO mutants with a significantly “slow fitness” blood stage phenotype (Bushell et al., 2017). See Figure S4B for plots showing the course of blood stage infections after sporozoite injection.

Figure S3

Genotyping of Single Gene Knockout Parasite Lines, Related to Figures 5, 6, and 7 (A) Schematic representation of the endogenous gene of interest (GOI) locus, (B) the gene deletion targeting construct (PlasmoGEM) containing the 3xHA-hdhfr-yFCU cassette and the GOI locus after disruption following double homologous recombination. Such a replacement strategy was used in cases where no sexual stage crossing was performed to bypass lethal mosquito stage phenotypes. The positions of primers used in PCRs to confirm GOI deletion is indicated by arrows labeled QCR1 or QCR2 and QCR2 or QCR1. (C) Schematic representation of the endogenous gene of interest (GOI) locus, the gene deletion construct (PlasmoGEM) containing the GOMO-GFP-mCherry-FACS cassette and the GOI locus after disruption following double homologous recombination. PCR primers used to confirm successful integration of the construct are indicated by arrows GT and GW1 or GW2 and PCR primers used to confirm deletion of the GOI are indicated by arrows QCR1 or QCR2 and QCR2 or QCR1 respectively. All primer sequences are shown in Table S6. The gene replacement strategy with the GOMO-GFP-mCherry-FACS cassette was used to generate mutant parasites that were crossed with a WT parasite line for cases in which the sKO was blocked at the mosquito stage. (D) Pulse field gel electrophoresis (PFGE) for the following gene knockout parasite lines (chromosome location of replaced gene shown in brackets, knockouts generated via transfection with constructs containing the 3xHA-hdhfr-yFCU cassette): ΔPDH-E2 (5), ΔHCS1 (5), ΔFabD (14), ΔFabG (8), ΔFabH (3), ΔLipA (13), ΔELO-A (8), ΔKCR_v1 (5), ΔCBR (11), ΔGFPT (5), ΔUSP (12), ΔPGM3 (9) and ΔPMM_v1 (5). A probe was used that recognized the 3′UTR of the pbdhfr hybridized to the knockout cassette that replaced the above genes on the chromosomes listed above. A probe of ≈800 bp fragment of the 5′UTR of the PBANKA_0508000 gene located on the chromosome 5 was also used for the ΔFabH mutant parasite and a probe of ≈800 bp fragment of the 5′UTR of the PBANKA_0508000 gene located on the chromosome 5 was additionally used for the ΔCBR mutant parasite. All images have been cropped from PFGE images showing other parasite lines. (E) Diagnostic PCR of the single gene knockout parasite lines ΔKCRv2, ΔPMMv2 and ΔUAP using primers to test for the presence of the WT locus and successful integration at the kcr, pmm and uap loci, respectively (parasites generated by transfection with a construct containing the GOMO-GFP-mCherry-FACS cassette). GT and GW2 primers were used to show integration for ΔKCRv2 and ΔPMMv2 and GT and GW1 for ΔUAP. Parasites were generated with the aim of crossing with WT parasites to rescue the lethal mosquito stage phenotype. Lanes showing markers have been removed and also other PCR products from other clones. (F) Control PCR to show successful amplification from gDNA samples taken from all single gene knockout parasite lines generated in this study Dotted lines between lanes indicate the reordering of lane images from the same gel photo. A space between lanes indicates lane images taken from separate gels. (G) PCRs showing the presence of genes in WT parasites and absence in mutant parasites for all single gene knockout lines generated in this study. All primers used for genotyping PCRs are listed in Table S6. Lanes showing markers have been removed and also other PCR products from other clones.

Figure S4

Phenotypic Analysis of Gene Knockout Mutant Parasite Lines Associated with the FASII Pathway, Related to Figure 5 (A) Graph showing oocyst numbers (relative to control; set to 100%) at day 7 post infection for ΔPDH-E2, ΔHCS1, ΔFabD, ΔFabG, ΔFabH and ΔLipA parasites. Error bars indicate standard deviation. The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (all results non-significant at 95% confidence). Boxes shown below the graph indicate the presence (green) of sporozoites from the salivary glands for each knockout strain between days 18 and 21 post-infection. (B) Graphs to show parasitemia for mice injected with ΔPDH-E2 sporozoites (based on FACS) or relative blood stage parasitemia for mice injected with ΔHCS1, ΔFabD, ΔFabG, ΔFabH and ΔLipA parasites (based on relative light units by luciferase assay). Data from all mice (control and KOs) are shown for two independent experiments and lines are drawn for the relative light unit level or parasitemia considered as being the point at which a mouse has become positive.

Figure 6

Mutations in Fatty Acid Elongation Disrupt Mosquito- and Liver-Stage Development (A) Pathway map for elongation of fatty acids (FAE) in the Plasmodium endoplasmic reticulum. See Table S4 for gene IDs, enzyme functions, and reactions. Blood-stage screen data suggested KCR to be essential, but we here correct the phenotype to dispensable, since a genotyped ΔKCR parasite shows comparable blood-stage growth to control parasites (Figure S5A). (B) Schematic representation of developmental blocks for single KOs and ΔKCR sporozoites derived from a ΔKCR × WT genetic cross. Green, phenotype not significantly different from WT. Red, block in life-cycle progression, except for liver stage, where red indicates phenotype significantly different from WT (>2-day delay in pre-patent period). (C) Size of cultured liver stages (48 hpi) of 250 EEFs. Median and interquartile ranges in red. ^∗p < 0.05 by Kruskal-Wallis test. (D) Relative maturation of EEFs measured as conversion of infected host cells to detached cells at 48 hpi. Error bars show standard deviations from 3 biological replicates (for ELO-A) or 8 biological replicates (all other mutants). The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (^∗∗∗p ≤ 0.001). (E) The number of mice with blood infection after injection of 5,000 sporozoites and the mean delay (range) in pre-patency compared to mice infected with WT sporozoites. ^∗, gene KO parasites with a significantly “slow” blood-stage phenotype (Bushell et al., 2017). See Figure S5E for plots showing the course of blood-stage infections after sporozoite injection.

Figure S5

Phenotypic Analysis of Gene Knockout Mutant Parasite Lines Associated with the FAE Pathway, Related to Figure 6 (A) Graph showing growth rate of ΔKCRv1 blood stage parasites in relation to control parasites, as shown as progression of parasitemia at successive days after injection of blood stage parasites. (B) Graph showing oocyst numbers (relative to control; set to 100%) at day 7 post infection for ΔELO-A, ΔKCRv1, ΔKCRv2, ΔKCRv2 (xWT) and ΔCBR parasites. Error bars indicate standard deviation. The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (^∗∗p < 0.01). Boxes shown below the graph indicate the presence (green) or absence (red) of sporozoites from the salivary glands for each knockout strain between days 18 and 21 post-infection. (C) Graph showing oocyst numbers in the ΔKCRv1 parasite strain relative to the control on days 4, 6, 8, 10 and 13 post-infection. Error bars display standard deviation. (D) Table displaying sexual and mosquito stage phenotypic data for ΔKCRv1 parasites; gametocyte conversion rate, exflagellation rate, female: male ratio, zygote to ookinete conversion rate and percentage of ookinetes showing abnormal morphology. Data is shown in relation to normal ranges for such phenotypic assessments. (E) Graphs showing relative blood stage parasitemia for mice injected with ΔELO-A parasites (relative light units by luciferase assay) and parasitemia for mice injected with ΔKCRv2 (x WT) and ΔCBR parasites (based on FACS). Data from all mice (control and KOs) are shown for two independent experiments and lines are drawn for the relative light unit level or parasitemia considered as being the point at which a mouse has become positive.

Figure 7

Mutations in Amino Sugar Metabolism Disrupt Liver-Stage Development (A) Activation of sugars in the Plasmodium cytosol based on iPbe. See Table S4 for gene IDs, enzyme functions, and reactions. Gal, D-galactose; Gal1P, D-galactose 1-phosphate; UDP-Gal, UDP-d-galactose; Glc, D-glucose; Glc6P, D-glucose 6-phosphate; Glc1P, D-glucose 1-phosphate; UPD-Glc, UDP-glucose; Fru, D-fructose; Fru6P, D-fructose 6-phosphate; Man, D-mannose; Man6P, D-mannose 6-phosphate; Man1P, D-mannose 1-phosphate; GDP-Man, GDP-mannose; GDP-4-keto-6-deoxy-Man, GDP-4-dehydro-6-deoxy-d-mannose; GDP-Fuc, GDP-L-fucose; Fuc, 6-deoxy-L-galactose/fucose; Fuc1P, L-fucose 1-phosphate; GlcN, D-glucosamine; GlcN6P, D-glucosamine 6-phosphate; GlcNAc6P, N-acetyl-d-glucosamine 6-phosphate; GlcNAc1P, N-acetyl-alpha-d-glucosamine 1-phosphate; UDP-GlcNAc, UDP-N-acetyl-d-glucosamine. (B) Schematic representation of developmental phenotypes of single KOs and mutants from ΔPMM × WT and ΔUAP × WT genetic crosses. Green, phenotype not significantly different from WT. Yellow, significantly reduced. Red, developmental block, except for liver stage, where red indicates phenotype significantly different from WT (>2-day delay in pre-patent period). (C) Size of 250 cultured EEFs 48 hpi; median and interquartile ranges in red. ^∗∗∗p < 0.001; ^∗p < 0.05 by Kruskal-Wallis test. (D) Relative maturation of EEFs measured as conversion of infected host cells to detached cells at 48 hpi. Error bars show standard deviations from 8 biological replicates. The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (^∗∗∗p ≤ 0.001). (E) Overall transmission success given as the number of mice that became blood stage positive after injection of 5,000 sporozoites and the mean delay (range) in pre-patency compared to mice infected with WT. ^∗, gene KO parasites with a significantly “slow” blood stage phenotype (Bushell et al., 2017). See Figure S6C for plots showing the course of blood stage infections after sporozoite injection. Supplemental Information

Figure S6

Phenotypic Analysis of Gene Knockout Mutant Parasite Lines Associated with the Amino Sugar Biosynthesis Pathway, Related to Figure 7 (A) Graph showing oocyst numbers (relative to control; set to 100%) at day 7 post infection for ΔUSP, ΔPMMv1, ΔPMMv2, ΔPMMv2 (xWT), ΔUAP, ΔUAP (xWT) and ΔPGM3 parasites. Error bars indicate standard deviation. The results were statistically evaluated by a one-way analysis of variance (ANOVA) test with Dunnet’s multiple comparisons (^∗∗p < 0.01, ^∗∗∗p < 0.001, ^∗∗∗∗p < 0.0001). Boxes shown below the graph indicate the presence (green) or absence (red) of sporozoites from the salivary glands for each knockout strain between days 18 and 21 post-infection. (B) Table displaying sexual and mosquito stage phenotypic data for ΔGFPT, ΔPMMv1 and ΔPGM3 parasites; gametocyte conversion rate, exflagellation rate, female: male ratio, zygote to ookinete conversion rate and percentage of ookinetes showing abnormal morphology. Data is shown in relation to normal ranges for such phenotypic assessments. (C) Graphs to show parasitemia for mice injected with ΔUSP parasites and ΔPMMv2 (xWT) and ΔUAP (xWT) parasites (based on FACS). Data from all mice (control and KOs) are shown for two independent experiments and lines are drawn for the relative light unit level or parasitemia considered as being the point at which a mouse has become positive.

Figure S7

A Graph Showing Correlation between the SG-B2 Transition Phenotype Seen in the M-L Screen and In Vivo Data for Single Gene Knockout Parasite Lines, Related to Figures 5, 6, and 7 The prepatent delay in the appearance of blood stage infection following injection of sporozoites is shown for each mouse injected with single gene knockout parasite lines ΔFabD, ΔFabH, ΔFabG, ΔPDH-E2, ΔHCS1, ΔLipA, ΔELO-A, ΔKCR (xWT), ΔCBR, ΔUAP (xWT) and ΔPMM (xWT) in relation to the log2-fold reductions at the SG-B2 transition in the M-L screen. The sizes of data point indicate the number of mice (total of 6) that show each pre-patent period. The ∞ symbol indicates mice that never became positive after sporozoite injection.

Figure S8

The PhenoMapping Workflow to Curate Metabolic Models and PhenoMapping Analyses to Identify Context-Specific Conditions Underlying Phenotypes, Related to Figure 4 (A) The PhenoMapping workflow involves three sets of analyses: (i) essentiality studies using the metabolic model and comparison with phenotypic data (blue); (ii) correction of False Negatives (FN, orange); and correction of False Positives (FP, blue). The curation of the metabolic network provides hypotheses on the metabolic function, here essential genes and underlying processes responsible for the essentiality. The classification of genes as fully, partially, or not agreeing with observed phenotypes might change when one modifies the metabolic network (at each iteration through the workflow). GPR denotes the gene-protein-reaction associations in the metabolic model. (B) PhenoMapping analysis to study four context-specific cellular processes or conditions underlying phenotypes: the uptake of substrates, thermodynamic directionality of reactions and transports for a set of context-specific metabolite concentrations, gene expression, and regulation of gene expression. The PhenoMapping analysis is modular: linear arrows from the center of the figure to the boxes of cellular processes indicate an independent PhenoMapping analysis of cellular processes; and curved arrows between cellular processes indicate that the PhenoMapping analysis can be applied cumulatively following the order number suggested.