Novel Plasmodium falciparum metabolic network reconstruction identifies shifts associated with clinical antimalarial resistance

Background Malaria remains a major public health burden and resistance has emerged to every antimalarial on the market, including the frontline drug, artemisinin. Our limited understanding of Plasmodium biology hinders the elucidation of resistance mechanisms. In this regard, systems biology approaches can facilitate the integration of existing experimental knowledge and further understanding of these mechanisms. Results Here, we developed a novel genome-scale metabolic network reconstruction, iPfal17, of the asexual blood-stage P. falciparum parasite to expand our understanding of metabolic changes that support resistance. We identified 11 metabolic tasks to evaluate iPfal17 performance. Flux balance analysis and simulation of gene knockouts and enzyme inhibition predict candidate drug targets unique to resistant parasites. Moreover, integration of clinical parasite transcriptomes into the iPfal17 reconstruction reveals patterns associated with antimalarial resistance. These results predict that artemisinin sensitive and resistant parasites differentially utilize scavenging and biosynthetic pathways for multiple essential metabolites, including folate and polyamines. Our findings are consistent with experimental literature, while generating novel hypotheses about artemisinin resistance and parasite biology. We detect evidence that resistant parasites maintain greater metabolic flexibility, perhaps representing an incomplete transition to the metabolic state most appropriate for nutrient-rich blood. Conclusion Using this systems biology approach, we identify metabolic shifts that arise with or in support of the resistant phenotype. This perspective allows us to more productively analyze and interpret clinical expression data for the identification of candidate drug targets for the treatment of resistant parasites. Electronic supplementary material The online version of this article (doi:10.1186/s12864-017-3905-1) contains supplementary material, which is available to authorized users.


BACKGROUND:
Three billion people are at risk for malaria infection globally and treatment approaches are failing. Malaria is caused by Plasmodium parasites, and most deaths are associated with humaninfective P. falciparum. Without an efficacious vaccine, antimalarials are essential to combat the severity and spread of disease. Combination therapies are implemented to preserve antimalarial efficacy and slow resistance development [1][2][3]; despite this approach, this eukaryotic pathogen has developed resistance to every antimalarial on the market [4][5][6].
Typically, resistance is conferred by genomic changes that lead to drug export or impaired drug binding (for example [7]); however, non-genetic mechanisms have also been implicated in Plasmodium resistance development [8][9][10] and other pathogenic organisms, such as Pseudomonas aeruginosa [11] (reviewed in [12]). These laboratory-based studies provide insight into metabolic flexibility but the presence of relatively few examples limit our understanding of this method of adaptation, especially in malaria. Here, we aim to look beyond genetic mechanisms of resistance to identify resistance-associated metabolic adaptation. We hypothesize that metabolic changes must occur to support the resistance phenotype and resistance-conferring mutations. Ultimately, these changes, or 'shifts,' are required to increase the fitness of resistant parasites, or support the development of additional genetic changes that affect fitness. Metabolic or phenotypic 'background' could be as important as genetic background in the development of resistance.
Here, we use a systems biology approach to analyze the metabolic profile associated with resistant and sensitive parasites. First, to maximize the accuracy of our predictions, we curated an existing genome-scale network reconstruction of asexual blood-stage P. falciparum metabolism. Using constraint-based metabolic modeling, we integrated transcriptomic data from over 300 clinical isolates from Cambodia and Vietnam with varying levels of artemisinin sensitivity. This approach identified innate metabolic differences that arise with or in support of the resistant phenotype, despite large clinical variability, over multiple genetic backgrounds.
Additionally, we were able to explore the functional consequences of expression changes by predicting essential enzymes within these distinct metabolic contexts; these enzymes are candidate drug targets for the prevention of drug resistance.

RESULTS:
Analysis of Artemisinin Sensitive and Resistant Transcriptomes. In order to investigate the presence of a distinct metabolic phenotype in artemisinin resistant parasites, we analysed a previously published expression dataset of clinical isolates from Southeast Asia (NCBI Gene Expression Omnibus accession: GSE59097). Patient blood samples were collected immediately prior to beginning artemisinin combination therapy and relative expression was evaluated using microarrays [49]. We confined our analysis of this previously published expression data to ringstage parasites from Cambodia and Vietnam, two countries that had clear resistant and sensitive parasite populations as defined by parasite clearance half-life, an in vivo phenotypic measure of resistance, and PfKelch13 mutations, a commonly-used genetic marker of resistance ( Figure 1A & 1B). There were 97 and 24 ring-stage resistant parasite expression profiles from Cambodia and Vietnam, respectively; resistant parasites are defined by both the presence of PfKelch13 mutations and a parasite clearance half-life of more than 5 hours. There were 141 and 43 ringstage sensitive parasite expression profiles from Cambodia and Vietnam, respectively, as defined by wild-type PfKelch13 alleles and clearance half-life of less than 5 hours. Despite obvious genotypic and phenotypic separation ( Figure 1A & 1B), artemisinin sensitive and resistant parasites do not separate well by hierarchical clustering of expression data ( Figure 1C).
Additionally, fold change of the majority of transcripts is moderate; no genes exhibited notable differential expression across both analyses (fold change > 2 or < 0.5 for both Cambodia and Vietnam sample sets, data not shown). Among metabolic genes specifically, expression differences are small (maximum fold change 0.6 and 1.6) and few are both significant and conserved between data sets (11 in common from 174 in Cambodia and 37 in Vietnam; Suppl.

Figure 1A & 1B).
Large amounts of transcriptional variation (due to stage-dependent expression, genotypic variability, and host-pathogen interactions) across the population of clinical parasites may hide differences in the data sets. Moreover, we built a Random Forest classifier with expression data to predict resistance outcomes; the classifier predicted resistance poorly, with only 30.77% sensitivity (indicating only 30.77% of resistant samples were correctly identified) and 97.96% specificity (indicating 97.96% of sensitive samples were correctly identified) (Suppl.

Figure 2A).
Although the expression data classifier performed poorly, a similar classifier built from metadata associated with each sample (patient and parasite characteristics) was highly predictive of resistance status with 85.71% sensitivity and 88.91% specificity (Suppl. Figure 2B). In our analysis, two specific mutations were the most predictive of resistance status, with sample collection site as the third most important variable; removing any of these three variables decreased classifier accuracy by over 20%. If Kelch13 mutations are used to predict resistance (rather than used to define resistance), Kelch13 mutations are most predicative of resistance (data not shown). Thus, metadata better predicts resistance than expression data. In order to deconvolve this innate variability and identify functional cellular changes associated with varying levels of artemisinin sensitivity, we integrated metabolic expression data into a genomescale metabolic model of blood-stage P. falciparum.
Manual metabolic network curation. To maximize the predictive ability of the metabolic network model, we curated an existing, well-validated reconstruction of asexual blood-stage P.
falciparum [50] to improve its scope, and species-and stage-specificities. Our curated reconstruction, iPfal17, includes all metabolic reactions encoded by characterized genes in the parasite's genome, summarizing metabolic behavior during the asexual blood-stage parasite. It is larger in scope from the previously published version due to the addition of 268 reactions ( Table   1, Suppl. Tables 1 & 2), with 9.6% more enzymatic reactions and 2.3% more reactions with gene annotations. We also added 124 genes to the network (  Following curation, the species-and stage-specificity of the model was also improved. Gene annotations were evaluated against PlasmoDB resources [43], resulting in 124 additional gene annotations (Suppl . Table 1). Importantly, we removed cellular import of pyrimidines from the host erythrocyte, as P. falciparum relies on de novo synthesis (Suppl . Table 2) [47,51].
Blood-stage specificity was improved by removing genes only used in other life stages (specifically the gene encoding chitinase [52]). Additionally, 77 functionally unnecessary reactions were removed due to a lack of genetic and biochemical support (Suppl. Table 2).
Reactions necessary for growth were added manually (Suppl. Table 1). Reactions were individually curated, changing metabolite utilization and stoichiometry (Suppl. Table 1).
The iPfal17 reconstruction contains five compartments: extracellular space and four intracellular compartments (cytoplasmic, mitochondrial, apicoplast, and food vacuole, Suppl. We also included annotations that will accelerate future curation efforts. First, we did not remove blocked reactions (those that do not carry flux due to their lack of connectivity to other components of the network) because further research may add connectivity to these network components. iPfal17 contains 303 blocked reactions and 78 dead-end metabolites (specifically, 32 metabolites are not consumed and 46 are not produced). For example, 4-pyridoxate (a byproduct of vitamin B6 biosynthesis) is included; production is supported by bioinformatic analyses of the parasite genome, but the metabolite function or excretion pathway is not known.
Second, citations are included within iPfal17 to identify the date of discovery and degree of literature support for each reaction (Suppl. degradation is essential for the blood-stage parasite to produce free amino acids. Parasites can also import and synthesize some amino acids, but the breakdown of hemoglobin (and subsequent production of its byproduct, hemozoin) is necessary for growth [23,63,64]. Thus, by requiring hemozoin export, we force the in silico parasite to degrade hemoglobin as the primary pathway for amino acid production.
• SO 4 f-thf = formyl tetrahydrofolate; mthf = methyltetrahydrofolate; thf = tetrahydrofolate iPfal17 validation and functional requirements. To validate the model against experimental results, essential metabolic tasks of blood-stage growth were identified and evaluated (Table 3).
These tasks describe experimental and clinical observations, such as the parasite's ability to grow on glucose and hypoxanthine as a sole carbon source and purine source, respectively, and the parasite's induction of blood acidosis via lactate [65][66][67]. Additional tasks include the parasite's failure to grow in the presence of antimetabolites for riboflavin, nicotinamide, thiamine, and pyridoxine [68]. We defined this set of tasks to provide a framework for curation and validation efforts of future network reconstructions. Although iPfal17 fails to pass all metabolic tasks, we believe this is the most comprehensive and accurate model to date due to the curation efforts and results from tests of the metabolic tasks. Failures generally exist in pathways that currently contain many reversible reactions (i.e. tasks 5a-b for glycolysis) or if the experimental evidence is not mechanistic (i.e. tasks 1a-d) or fully characterized (i.e. task 4; Table 3).  Table 4 and Suppl. Table 7 11 Accuracy of P. berghei essentiality predictions -61.4% See Suppl. Table 6 * Accuracy calculated as the sum of true positives and true negatives, divided by total observations We also evaluated predictions of the effects of gene knockouts and enzyme inhibitors using previously published experimental results (  MADE integration of sensitive and resistant expression data from both countries generated four condition-specific models (Figure 3). Genes (and associated reactions) are removed from condition-specific models if their transcripts are down-regulated and not functionally necessary for metabolism in that condition. Those remaining in the constrained model are a subset of genes annotated in our original curated reconstruction; these genes are either more highly expressed in the corresponding condition, not differentially expressed across conditions, or necessary for network functionality, and, thus, remain in the condition-specific model. By comparing these models, we identified differences in gene and pathway utilization between resistant and sensitive parasites that are consistent between the isolates from the two countries (Suppl. Figure 3).
First, we conducted an enrichment analysis on genes that remain in (i.e. can be utilized by) each constrained model by comparing to the unconstrained curated model. As expected, all four models were enriched with genes involved in pathways with many essential reactions or little redundancy, such as transport reactions, tRNA synthesis, purine metabolism, and others (Suppl. Figure 4, see model). Sensitive, wild type, models corresponding to isolates from both Cambodia and Vietnam are uniquely enriched with the utilization of genes involved in the metabolism of nicotinate/nicotinamide (p-value = 1.47*10 -2 ), glutamate (p-value = 1.28*10 -13 ), and selenocysteine (p-value = 5.85*10 -4 ). Thus, sensitive models contain more reactions in these pathways than the unconstrained model, resulting from increased expression of these pathways in sensitive parasites (Suppl. Figure 4). Resistant models from both countries are uniquely enriched with the utilization of genes involved in pyrimidine (p-value = 2.18*10 -7 ), polyamine (p-value = 4.39*10 -4 ), redox reactions (p-value = 5.13*10 -5 ), and central carbon metabolism (glycolysis [p-value = 4.39*10 -4 ] and the pentose phosphate pathway [p-value = 6.06*10 -3 ]).
Thus, resistant models have a larger proportion of their total reactions associated with these pathways than the original unconstrained model, whereas sensitive models do not have this enrichment. This indicates that these pathways are upregulated in resistant parasites and may remain important for metabolism in the resistant state (Suppl. Figure 4).
Identification of conserved and uniquely essential pathways. Beyond general differences in pathway utilization, which encompasses both essentiality and pathway-level differences in expression, artemisinin sensitive and resistant parasites have unique essential genes and reactions.
To identify these essential reactions and provide insight on targetable metabolic enzymes in the clinical isolates, we performed in silico single gene and reaction deletions with each of the four condition-specific models. Datasets from the parasites from each country were initially analyzed separately and then lists were compared to ensure resistance-associated trends are reproducible and observed in independent analyses. As expected, we identified many essential functions conserved in all models (Suppl . Table 5), which is consistent with an active core metabolism required for basic parasite survival. Importantly, 21 reactions were essential in only resistant models, but not sensitive models (Table 5). Theoretically, drugs targeting these reactions would kill resistant parasites and have no effect on sensitive parasites; thus, there would be no selective pressure within the sensitive parasite population to develop resistance to these drugs. This list included serine hydroxymethyltransferase (PFL1720w in folate metabolism), the glycine cleavage system (PFL1550w and others in folate metabolism), thiamine diphosphokinase (PFI1195c in cofactor metabolism, specifically thiamine diphosphate), fumarate hydratase and malate dehydrogenase (PFI1340W and PFF0895w, respectively, in the mitochondrial electron transport chain and TCA cycle), and fructose hexokinase (PFF1155W in glycolysis; Table 5).
We also identified 12 reactions that were essential only in artemisinin sensitive parasites ( Table   6). Drugs targeting these reactions should not be combined with artemisinin, as they would not kill (and may select for) resistant parasites. Fortunately, no existing drug targets were found in this list of essential genes and reactions ( Table 6). Among those identified were sphingomyelin synthase 2 (PFF1215w) and several transport reactions, which furthers our understanding of condition-specific intra-organellar function ( Table 5 & 6). Overall, our systems biology-based approach reveals unique metabolic phenotypes associated with artemisinin sensitivity; these differences were not detected in the original analysis of the expression dataset or by separately analyzing Cambodian or Vietnamese isolates ([49] and data not shown).

DISCUSSION:
Systems biology approaches enable unbiased analyses of antimalarial resistance phenotypes.
Here, we describe a newly curated metabolic network reconstruction of the malaria parasite that can serve as a platform for the analysis of gene expression and other 'omics data, and as a tool to generate testable hypotheses regarding essential genes and metabolic phenotypes. In particular, we used this network reconstruction to characterize key metabolic dependencies in resistant and sensitive parasites. We revealed emergent patterns in pathway activity, differential utilization of organelles, metabolic flexibility, and targetable weakness of resistant parasites. We generated gene and reaction essentiality predictions with our curated network model, prior to integration of expression data, and found results largely consistent with previous models

Table 3 & 4)
; 24 of these have been empirically tested in cultured P. falciparum parasites ( Table 4, and in P. berghei-Suppl. Table 6). iPfal17 better predicts experimentally determined essential reactions than previous models, across a broad set of metabolic pathways ( iPfal17 also fails to predict the lethal nature of adenosine deaminase in purine-free conditions [78]. Adenosine deaminase converts adenosine to hypoxanthine; as 38 reactions produce AMP, which then generate hypoxanthine products, we propose adenosine deaminase may be essential for nonmetabolic functions or the inhibitor of adenosine deaminase has off target effects.
Furthermore, these results generate hypotheses about the differential metabolic capabilities of P. falciparum and P. berghei, as experimental results in the rodent parasite conflict with some P.

Data integration reveals distinct metabolic patterns. The integration of expression data from
clinical parasites into our network reconstruction highlights the differential utilization of metabolic genes and reveals metabolic shifts associated with variation in innate artemisinin sensitivity (Suppl. Figure 3 & 4). Enriched metabolic pathways detected in sensitive and resistant models are consistent with previous experimental observations. For example, resistant models are uniquely enriched with genes involved in pyrimidine biosynthesis and mitochondrial redox reactions . This finding is consistent with the importance of mitochondrial function in surviving artemisinin stress [26,29]. Additionally, the metabolic disruption of the redox reactions in the electron transport chain upon artemisinin treatment (via decreased production of orotate and fumarate, presumably via dihydroorotate dehydrogenase and succinate dehydrogenase enzymes [22,28,116]) suggests that changes in these pathways may be important for survival in the presence of the drug. Thus, this metabolic network analysis approach allows us to filter out noise from diverse clinical isolates to identify alternative utilization of pathways associated with artemisinin resistance. However, these enrichment results do not implicate specific reactions that are uniquely active in artemisinin sensitive or resistant parasites.
Condition-specific models have unique metabolic requirements. Upon integration of expression data and the identification of differentially utilized pathways above, we next used these models to predict targetable differences in sensitive and resistant parasites by identifying reactions that are essential within the context of the metabolic network (Suppl . Tables 5, Tables   5 & 6). We identified (1) differences in intra-organellar function, (2) metabolic flexibility of scavenging and biosynthesis pathways, and (3) targetable weakness of resistant parasites. These metabolic shifts primarily reside in mitochondrial metabolism, as well as folate and polyamine metabolism. Together, these results highlight the overall plasticity of P. falciparum metabolism and opportunities for further development of potential drug targets.
Interestingly, several transport reactions are found to be differentially essential in our constrained models (Tables 5 & 6). Many transport reactions (79.5%) have no associated gene due to the incomplete characterization of the P. falciparum genome (Figure 2). They are included in the model due to biochemical evidence or functional necessity (i.e. a metabolite is produced in one compartment but it is a substrate for an enzyme in another). Transcriptomic data integration does not constrain their behavior explicitly: expression integration reduces the total number of reactions in a model, forcing transport of metabolites among organelles if withincompartment biosynthesis is non-functional. Function within organelles requires transport and loss of function reduces transport needs. Specifically, several mitochondrial and apicoplast transport reactions are uniquely essential in the sensitive and resistant parasite populations ( Figure 4). In resistant models, this includes the mitochondrial transport of metabolites associated with the TCA cycle and electron transport chain (fumarate, oxaloacetate, and NADPH) and those involved in generation of folates (tetrahydrofolate, glycine, CO 2 , and NH 4 +) ( Figure 4A). In sensitive models, apicoplast transport of ADP, ATP, and phosphate is essential ( Figure 4A). Overall, these results indicate that sensitive and resistant parasites are differentially utilizing pathways within these organelles, and have unique requirements for transport of essential substrates. This observation is consistent with previous studies and our enrichment results highlighting the influence of mitochondrial metabolism on survival in the presence of artemisinin [26,29]. Moreover, oxygen transport into the cell and then into the mitochondria is only essential in sensitive parasites, further predicting differential use of the mitochondria in these parasites as oxygen serves as the terminal step in the electron transport chain. Resistant parasites are predicted to generate oxygen within the mitochondria via superoxide dismutase as opposed to transport ( Figure 4A).
We also identify differential utilization of transport pathways from the extracellular environment into the parasite. Plasmodium metabolism contains redundancies; for many essential metabolites, the parasite's genome encodes one or more biosynthetic pathways, while there is also evidence for a parallel host-scavenging pathway [108] (e.g. lipid [56] and amino acid [56, 63] scavenging). Upon model integration, we find that artemisinin resistant and sensitive parasites utilize some of these metabolic pathways in alternative ways ( Figure 4A).
Plasmodium can either scavenge or synthesize putrescine and adenosyl-methionine (essential polyamines and precursors to spermidine [50,117]), as well as p-aminobenzoate, a folate precursor generated by branch of glycolysis necessary for nucleotide synthesis ([118]; Figure 4 A&B, Tables 5 & 6). These metabolites are measurable via blood sample metabolomics [118,119]; therefore, host scavenging is a viable option for blood-stage parasites. We predict that 1 sensitive parasites rely on the import of putrescine, adenosyl methionine, and p-aminobenzoate.
Resistant parasite expression supports either this host scavenging or direct biosynthesis due to parasite survival upon reaction knockout in silico. Thus, we expect that resistant parasites are more metabolically flexible for these metabolites; perhaps resistant parasites have failed to appropriately modulate their transition to the nutrient-rich blood-stage environment, and this unexpected flexibility is evolutionarily beneficial once confronted with artemisinin treatment.
Interestingly, recent metabolomics studies demonstrate that intra-parasitic putrescine Our systems biology approach also identifies metabolic weaknesses of resistant parasites; these weaknesses can be used to identify drug targets for combination therapies ( Figure 5). For example, we identified the mitochondrial import of fumarate and subsequent conversion to oxaloacetate (via fumarate hydratase, PFI1340W, and malate dehydrogenase, PFF0895W) to be uniquely essential in resistant parasites ( Figure 5A, Additionally, we identified serine hydroxymethyltransferase (SHMT) and thiamine diphosphokinase as potential drug targets of resistant parasites (Table 5, Figure 5B); see below for discussion of SHMT. Both the import of thiamine and thiamine diphosphokinase are essential only in resistant parasites (Figure 5C), and we predict inhibition of import or enzyme activity would specifically target resistant parasites.  (Figure 5B left). In sensitive parasites, the essential metabolites can be generated by SHMT or the mitochondrial glycine cleavage system, given the reversible nature of these enzymes [127,128]. Therefore, in our sensitive models, neither SHMT nor the glycine cleavage system is essential when knocked out individually. This observation conflicts with the literature, as SHMT is essential in cultured parasites [127,129,130]. Thus, iPfal17 is unable to predict this intricacy of parasite metabolism, revealing interesting regulatory effects, an uncharacterized location dependency for metabolite generation, or in vivo/in vitro differences in enzyme reversibility.
Similarly, model integration reveals that protein localization influences essentiality predictions. We predicted that the cyclical oxidization and reduction of glutathione, a key regulator of oxidative stress [131][132][133][134], and supporting reactions were essential only in resistant parasites when the glutathione redox system was located within the mitochondria (data not shown). This is consistent with artemisinin's induction of reactive oxygen species, the parasite's obvious need to survive this stress [24,[33][34][35], and data showing artemisinin sensitivity is correlated with glutathione levels in rodent Plasmodium [27]. However, upon moving these reactions to the cytosolic and apicoplast compartments (as supported by [135]), these reactions were no longer essential. Thus, model analysis challenges the integration of previously incomparable datasets by demonstrating that this localization and role of glutathione yield different predictions. Future studies will be required to clarify these findings.
Here, we have presented a novel blood-stage-specific P. falciparum metabolic network reconstruction, iPfal17, and investigation of the metabolic differences between artemisinin sensitive and resistant parasites. Antimalarial resistance is a major public health problem and we demonstrate that constraint-based modeling can be used to reveal metabolic shifts that arise with or in support of the resistant phenotype and discrepancies between otherwise incomparable datasets. We find inherent differences in artemisinin resistant and sensitive parasite metabolism, even before artemisinin treatment. Artemisinin resistant parasites have major metabolic shifts in the mitochondria and in the synthesis of folates and polyamines, indicating incomplete transition to the metabolic state most appropriate for the blood-stage environment. These findings generate areas of future research to elucidate Plasmodium biochemistry, understand the evolution of artemisinin resistant parasites, and tackle antimalarial resistance.
Probes on the microarray platform GPL18893 were annotated using NCBI's stand-alone BLAST correcting the gene labels for 647 probes. Only top hits were used; specifically, hits with greater than 95% identity, no gaps, and a score of over 100 were used (Suppl.  We simulated in vitro growth requirements by modifying media components or access to particular metabolites. Metabolite import or production was eliminated from the reconstruction, and subsequent biomass production was observed. Lethal modifications were defined as changes that resulted in no production of biomass; growth-reducing modifications were defined as producing less than 90% of unconstrained flux value [98,143]. Curation. Manual curation of an existing P. falciparum metabolic network reconstruction [50] was conducted by a literature review and reference to generic and Plasmodium-specific databases (KEGG, Expasy, and PlasmoDB, MPMP) [43,[145][146][147]. Data obtained from these sources were used to evaluate the inclusion of reactions as well as their stoichiometry, reversibility, localization, and gene annotations. Genetically and biochemically supported reactions were kept and new reactions were added. Reactions were removed if (1) explicitly determined to be false or (2) were nonfunctional and not supported biochemically or genetically.
Spontaneous reactions (reactions that occur without enzymes) are noted to differentiate from orphan reactions (reactions with unknown enzyme catalysts).
In order to assess gene essentiality, we used a biomass reaction as the modeling objective function. Thus, flux through this reaction, simulating cellular growth, was maximized for all in silico experimental procedures. We used the biomass reaction from a previous study [50] with modifications. Curation of the biomass reaction was informed by metabolites detected in metabolomics studies [28,[56][57][58]; if possible, metabolite ratios were predicted from metabolomics data. We curated the biomass reaction with in consideration of published essentiality data; metabolites detected in metabolomics experiments with no known catabolism or import pathways were excluded from the biomass reaction.
Essentiality studies. We predicted essentiality by performing single deletion studies with both genes and reactions and double gene deletion studies in our curated model and each expressionconstrained sensitive and resistant models. Gene deletions were simulated by removing the gene of interest from the model. This change results in the inhibition of flux through all reactions that require that gene to function. If the model could not produce biomass with these constraints, the gene was deemed essential. Growth reducing phenotypes were also observed and noted. For reaction deletion studies, we removed reactions sequentially. Subsequent growth effects were used to determine reaction essentially. Consensus results for resistant or sensitive models are discussed.

LIST OF ABBREVIATIONS
Metabolic Adjustment for Differential Expression algorithm (

Acknowledgements
We acknowledge the members of the Guler, Papin, and Petri labs for their thoughtful conversations and insight. We would also like to thank Dr. Paul Jensen, Dr. Edik Blais, and Gregory Medlock for project feedback.

Funding
The study was financed by institutional funding from the University of Virginia (JLG) and by the National Institute of Allergy and Infectious Disease R21AI119881 (JLG and JP). MAC is supported by an institutional training grant (T32GM008136).

Availability of data and materials
The supporting data and materials of this article are included within the article and additional files; additionally, our model and code is available on https://github.com/maureencarey/iPfal17.

Authors' contributions
MAC, JG, and JP designed the study. MAC curated the model and performed metabolic network, statistical, and machine learning analyses. All the authors participated in data interpretation, and read and approved the manuscript.

Additional files
Additional information is available upon request, in Additional file 1 (Supplemental Figures 1-4), Additional file 2 (Supplemental Tables 1-8), and Additional file 3 (model).      A  b  u  n  d  a  n  t  p  r  o  t  o  n  p  u  m  p  i  n  g  i  n  P  l  a  s  m  o  d  i  u  m  ,  b  u  t  w  h  y  ?   T  r  e  n  d  s  i  n  p  a  r  a  s  i  t  o  l  o  g  y  ,  2  0  0  2  .   1  8   (  1  1  )  :  p  .  4  8  3  -4  8  6  .  5  6  .  G  u  l  a  t  i  ,  S  .  ,  e  t  a  l  .  ,   P  r  o  f  i  l  i  n  g  t  h  e  e  s  s  e  n  t  i  a  l  n  a  t  u  r  e  o  f  l  i  p  i  d  m  e  t  a  b  o  l  i  s  m  i  n  a  s  e  x  u  a  l  b  l  o  o  d  a  n  d  g  a  m  e  t  o  c  y  t  e  s  t  a  g  e  s  o  f  P  l  a  s  m  o  d  i  u  m  f  a  l  c  i  p  a  r  u  m  .   C  e  l  l  h  o  s  t  &  m  i  c  r  o  b  e  ,  2  0  1 S  T  U  D  I  E  S  O  N  M  A  L  A  R  I  A  L  P  A  R  A  S  I  T  E  S  :  V  I  I  I  .  F  A  C  T  O  R  S  A  F  F  E  C  T  I  N  G  T  H  E  G  R  O  W  T  H  O  F  P  L  A  S  M  O  D  I  U  M  K  N  O  W  L  E  S  I  I  N  V  I  T  R O .  P  o  t  e  n  t  a  n  d  s  e  l  e  c  t  i  v  e  a  c  t  i  v  i  t  y  o  f  a  c  o  m  b  i  n  a  t  i  o  n  o  f  t  h  y  m  i  d  i  n  e  a  n  d  1  8  4  3  U E  v  i  d  e  n  c  e  f  o  r  t  h  e  i  n  v  o  l  v  e  m  e  n  t  o  f  P  l  a  s  m  o  d  i  u  m  f  a  l  c  i  p  a  r  u  m  p  r  o  t  e  i  n  s  i  n  t  h  e  f  o  r  m  a  t  i  o  n  o  f  n  e  w  p  e  r  m  e  a  b  i  l  i  t  y  p  a  t  h  w  a  y  s  i  n  t  h  e  e  r  y  t  h  r  o  c  y  t  e  m  e  m  b  r  a  n  e  .   M  o  l  e  c  u  l  a  r  m  i  c  r  o  b  i  o  l  o  g  y  ,  2  0 P  h  e  n  o  t  y  p  i  c  C  h  a  n  g  e  s  i  n  A  r  t  e  m  i  s  i  n  i  n  -R  e  s  i  s  t  a  n  t  P  l  a  s  m  o  d  i  u  m  f  a  l  c  i  p  a  r  u  m  L  i  n  e  s  I  n  V  i  t  r  o  :  E  v  i  d  e  n  c  e  f  o  r  D  e  c  r  e  a  s  e  d  S  e  n  s  i  t  i  v  i  t  y  t  o  D  o  r  m  a  n  c  y  a  n  d  G  r  o  w  t  h  I  n  h  i  b  i  t  i  o  n  .   A  n  t  i  m  i  c  r  o  b  i  a  l  A  g  e  n  t  s  a  n  d  C  h  e  m  o  t  h  e  r  a  p  y  ,  2  0  1 2 .