Metabolomic Analysis of Diverse Mice Reveals Hepatic Arginase-1 as Source of Plasma Arginase in Plasmodium chabaudi Infection

ABSTRACT Infections disrupt host metabolism, but the factors that dictate the nature and magnitude of metabolic change are incompletely characterized. To determine how host metabolism changes in relation to disease severity in murine malaria, we performed plasma metabolomics on eight Plasmodium chabaudi-infected mouse strains with diverse disease phenotypes. We identified plasma metabolic biomarkers for both the nature and severity of different malarial pathologies. A subset of metabolic changes, including plasma arginine depletion, match the plasma metabolomes of human malaria patients, suggesting new connections between pathology and metabolism in human malaria. In our malarial mice, liver damage, which releases hepatic arginase-1 (Arg1) into circulation, correlated with plasma arginine depletion. We confirmed that hepatic Arg1 was the primary source of increased plasma arginase activity in our model, which motivates further investigation of liver damage in human malaria patients. More broadly, our approach shows how leveraging phenotypic diversity can identify and validate relationships between metabolism and the pathophysiology of infectious disease.

elevated plasma levels of the metabolic enzymes aspartate and alanine transaminase (AST and ALT) (6,7). Kidney dysfunction also leads to plasma metabolic changes, the most characteristic being elevated plasma urea and creatinine (8,9). Dysfunction in either organ can lead to organ failure and other systemic complications (7). Hypoglycemia (10), metabolic acidosis (5), and other metabolic changes also accompany severe disease. However, it remains largely unclear how metabolic changes relate to the nature and degree of malaria pathophysiology.
Metabolites have recently emerged as important controllers of infection pathophysiology. Manipulation of glycolysis, for example, alters disease severity in malaria (11,12) and controls tissue damage in other infections (13). Other metabolic targets, including iron (14), immunomodulatory metabolic enzymes like arginase (15)(16)(17), and metabolic hormones (18) have high potential to reduce pathology during infection. Undoubtedly, many metabolic therapies have yet to be discovered.
To uncover metabolic processes that are important during Plasmodium infection, we recently identified hundreds of plasma metabolites that change significantly when C57BL/6 mice are infected with Plasmodium chabaudi (19). Many of these metabolites also change in human malaria (4,5,(19)(20)(21)(22), but the causes and consequences of many of these metabolic changes remain unknown.
In a typical experiment, the field uses inbred strains to limit signal to noise to increase our probability of finding significant results. Here we used a collection of diverse inbred strains to deliberately introduce variation and then measure that variation. We selected eight inbred mouse lines (C57BL/6, WSB/EiJ, NZO/HILtJ, 129S1/SvImJ, A/J, CAST/EiJ, PWK/PhJ, and NOD/ShiLtJ) that represent 90% of laboratory mouse genetic diversity (23) and vary widely in P. chabaudi infection severity as measured by survival, anemia, parasite load, temperature loss, and weight loss ( Fig. 1; see also Fig. S1 in the supplemental material). These founder strains were used to establish the Collaborative Cross (CC) and Diversity Outbred (DO) mouse populations (24), which more closely approximate human genetic and phenotypic diversity than individual inbred lines. Collectively, these groups of diverse mice have improved our understanding of genetic and immune factors that alter the severity of infections like influenza and tuberculosis (25)(26)(27)(28)(29)(30). Our goal was to use diverse mice to identify metabolic factors that alter infection severity in malaria.
To link malaria pathophysiology with host metabolic changes, we measured plasma metabolites and markers of pathology and immune responses in the eight P. chabaudiinfected founder strains. Using principal component analysis, we found that the magnitude of metabolic response to P. chabaudi infection was directly related to disease severity. Using correlation analysis, we found that arginine depletion was a specific marker for malaria-induced liver damage. We identified additional candidate molecules for future study. We showed that P. chabaudi-induced liver damage releases arginineconsuming hepatic arginase-1 (Arg1) into circulation, which explains increased plasma arginase activity in our model and may explain elevated plasma arginase in human malaria. Our results support the use of diverse mice to understand the links between metabolism and disease severity during infection and motivate further investigation of the metabolic consequences of liver damage in human malaria.

RESULTS
Infection alters metabolomic profiles in diverse mice. We previously identified 370 metabolites that change significantly during P. chabaudi infection in C57BL/6 mice (19). However, wild-type C57BL/6 mice are largely resilient to infection, which limited our ability to understand how host metabolism changes in severe disease. To measure metabolism in mice with a broader range of disease severities, we used the eight founder strains (C57BL/6, WSB/EiJ, NZO/HILtJ, 129S1/SvImJ, A/J, CAST/EiJ, PWK/PhJ, and NOD/ShiLtJ), which vary widely in survival following P. chabaudi infection (Fig. 1A). For each strain, we monitored disease severity, plasma metabolites, plasma cytokines, and circulating immune cells during acute infection (days 5 to 12 postinfection or days 5 to 17 for WSB/EiJ mice, n = 5 infected mice per strain per day), when metabolic perturbations in C57BL/6 mice are greatest (19). The founder strains varied widely across nine metrics of malaria severity that largely tracked with survival and parasite load: anemia (red blood cells [RBCs] per microliter of blood), hypothermia, weight loss, hypoglycemia, liver injury (plasma ALT), and kidney injury (plasma urea, creatinine) ( Fig. 1B and Fig. S1). Plasma cytokines tended to be higher in resilient strains than in nonresilient strains, with two exceptions: cytokines in resilient WSB/ EiJ mice remained low throughout infection ( Fig. 2A), and two nonresilient strains (PWK/PhJ and NOD/ShiLtJ) mounted hyperinflammatory responses (e.g., high gamma interferon [IFN-g], tumor necrosis factor alpha [TNF-a], and interleukin 6 [IL-6] and low IL-10 and transforming growth factor b [TGF-b]) on day 8 or 9. Immune cell responses were largely consistent across strains, with elevated T cells, B cells, NK cells, and monocytes between days 7 to 9 postinfection (Fig. 2B). NK cells were particularly high in resilient NZO/HILtJ mice, as were T and B cells in resilient 129S1/SvImJ and nonresilient CAST/EiJ mice, respectively. Collectively, our data highlight the diversity of disease and immune phenotypes in the eight founder strains. The eight founder strains (left to right): CAST/EiJ, 129S1/SvImJ, WSB/EiJ, C57BL/6J, NZO/HILtJ, NOD/ShiLtJ, PWK/PhJ, and A/J. For survival, strain colors reflect average survival and parasite load, with dark blue for the most resilient/ fewest parasites and dark red for the least resilient (n = 10 for 100% resilient WSB/EiJ, NZO/HILtJ, and 129S1/ SvImJ; n = 30 for 73% resilient C57BL/6; n = 32 for 72% resilient CAST/EiJ; n = 15 for 40% resilient NOD/ShiLtJ; n = 15 for 20% resilient A/J; n = 22 for 4.5% resilient PWK/PhJ). DPI, days postinfection. Photo reproduced with permission from Jackson Laboratories. (B) Nine metrics of disease severity across the eight founder strains over the course of acute P. chabaudi infection. Parasitemia = percent parasitized RBCs, parasite density = 10 6 RBCs per microliter blood, anemia = 10 6 RBCs per microliter blood). Lines indicate the mean (6 SE) of $5 mice per strain per day. Z-scores = standard deviations from the means for uninfected C57BL/6 mice.
Interstrain variation in malarial disease severity allowed us to ask several questions about the relationship between disease severity and metabolism: is overall metabolic disruption greater in sicker animals? How do individual metabolites relate to different facets of disease? Can these metabolic markers of pathology provide new mechanistic insights into the pathophysiology of malaria or other infectious diseases?
To answer how disease severity impacted the overall composition of plasma metabolomes, we performed dimensionality reduction and visualization of 635 metabolites (see Materials and Methods) in the same samples used to determine disease severity using principal component analysis (PCA). Variance in principal component 2 (PC2) (11.2%) was associated with baseline differences among mouse strains, with NZO/HILtJ samples having the highest PC2 values and wild-derived strains (CAST/EiJ, WSB/EiJ, and PWK/PhJ) having the lowest PC2 values (Fig. 3A). Variance in PC1 (31.2%) was associated with infection severity. Highly resilient mice (WSB/EiJ, NZO/HILtJ, and 129S1/SvImJ) had high PC1 values throughout infection, while nonresilient strains had low PC1 values during periods of acute illness (days 7 to 10). Phase curves of PC1 versus parasite density show larger loops in disease space for less resilient strains, as is expected for variables that influence disease severity (Fig. 3B) (31). We next used canonical correspondence analysis (CCA), a technique similar to PCA, to identify molecules that drove the most variation among samples (see Table S1 in the supplemental material). Molecules that differentiated sick from healthy samples were associated with altered feeding behavior (food components, ketosis markers, indoles, serotonin, and acylcarnitines), oxidative stress (glutathione and 2-hydroxyisobutyrate), and liver and kidney dysfunction (formiminoglutamate, bilirubin, homocitrulline, urea, p-cresol glucuronide, p-cresol sulfate, and N-acetyltyrosine). We also performed Student's t tests to identify molecules that differentiate resilient and nonresilient mice before infection (Table S2). Plasma levels of a number of metabolites differed significantly between uninfected resilient and nonresilient mice, including the food component and osmolyte betaine (32), sphingomyelins, lysine, and pantothenic acid.
Correlations identify metabolic markers for disease severity in P. chabaudi malaria. Having determined that the sickest mice displayed the greatest changes in plasma metabolome composition, we next obtained a more granular understanding of pathology-metabolism relationships by correlating individual metabolites with our nine metrics of disease (Fig. 4A). The kidney dysfunction marker urea had the highest number of strong correlations (R 2 . 0.4), followed by ALT, creatinine, and temperature loss. This is consistent with results from our PCA, which identified metabolic markers for organ dysfunction as strong drivers of variation in PC1. Anemia and hypoglycemia also correlated well with some plasma metabolites, while few metabolites correlated well with weight loss, parasitemia, or parasite density. Altogether, our correlations suggest the host is a stronger driver of infection-induced plasma metabolic changes than the parasite.
Some metabolites correlated well with multiple disease metrics. For example, a-hydroxyisovalerate, a hallmark of defective leucine metabolism (33) and the best indicator of hypoglycemia in our malarial mice (Fig. 4B), correlated well with all disease metrics except parasite load (see Data Set S2 in the supplemental material). Similarly, elevated long-chain acylcarnitines were the best predictors of high plasma urea levels and temperature loss (Fig. 4B), but many also positively correlated with ALT. Long-chain acylcarnitines are also elevated in other pathological states like lipopolysaccharide (LPS)-induced inflammation (34) and acute kidney injury (35) (Fig. S2). Levels of a-hydroxyisovalerate and long-chain acylcarnitines also increase during starvation (36,37) (Fig. S2), which is consistent with our previous observations of P. chabaudi-induced anorexia (11). Collectively, these multicorrelative metabolites may point to shared underlying mechanisms of metabolic disruption across disease states.
In contrast to metabolites that were broadly indicative of severe disease, some metabolites specifically correlated with just one pathology. Some of these associations have been identified previously, including pipecolate as a marker for parasitemia (38) and N,N,N-trimethyl-alanylproline betaine, which positively correlates with creatinine in our data and was recently identified as a sensitive marker for kidney injury (39). Other metabolites suggested directions for future study. For example, 3-methylhistidine, a proposed marker for muscle protein breakdown (40), correlated with weight loss in our malarial mice. Muscle protein breakdown has not been reported in P. chabaudi-infected mice, but cachexia occurs in other inflammatory conditions (18,41,42). We also noted a strong association between arginine depletion and elevated ALT (Fig. 4B), a molecular proxy for liver damage (6,7). While both abnormalities are prevalent in human malaria and pose health concerns (7,20,(43)(44)(45)(46), the connection between liver damage and arginine in malaria is understudied.
Arginine depletion is a specific indicator of liver damage in P. chabaudi malaria. To better determine how specifically plasma arginine depletion indicates liver damage, we applied sparse partial least-squares discriminant analysis (sPLS-DA) to identify the best metabolic markers for high ALT in our data and in a second data set (19). sPLS-DA is a supervised learning method that selects variables that best differentiate samples based on user-assigned groups and is well-suited to sparse and heterogeneous multi-omics data (47-49) (Materials and Methods). We limited analysis in our data set to two strains of mice-CAST/EiJ and C57BL/6-that had low and high levels of ALT, respectively (n = 5 mice per strain per day on days 7 to 9) but similar levels of other disease severity metrics. By comparing samples from only these two strains, we reasoned we could identify metabolites that are specific to liver damage rather than other pathologies. Our sPLS-DA model identified arginine, adenosine, and methionine along with other metabolites in component 1 as best able to discriminate among samples with "high," "medium," or "low" ALT values (Fig. 4C, top, and Table S3). We performed a similar analysis (Fig. 4C, bottom) on plasma metabolomes from P. chabaudi-infected wild-type and Ahr 2/2 C57BL/6 mice (n = 6 mice per genotype per day for days 6 to 8 postinfection), which have high and low levels of ALT during infection, respectively. Upon comparing metabolic candidates from sPLS-DA and our correlation analysis (n = 40 to 50 metabolites for each of three analyses), we found that overlap between the three analyses was low-just one to three metabolites were identified by any two of three analyses-but arginine was selected by all three (Fig. 4C, Data Set S2, and Table S3).
Malarial liver damage releases hepatic Arg1 into circulation in conjunction with arginine depletion. In healthy, ureotelic animals, hepatic arginase-1 (Arg1) converts arginine to ornithine and urea (50) (Fig. 5A) in a process that is spatially restricted to hepatocytes. However, drug-induced hepatocellular injury releases hepatic arginase-1 into circulation where it depletes plasma arginine (6). Only associative evidence has linked arginine depletion with hepatic arginase and liver injury in malaria (44). Given the strong and specific association between arginine depletion and ALT in our data, we hypothesized that P. chabaudi-induced liver damage depletes plasma arginine by releasing hepatic Arg1 into circulation. In support of our hypothesis, we found that the arginase product ornithine is elevated in mice with high ALT and low arginine (Fig. 5B, top row, C57BL/6, A/J, NOD/ ShiLtJ, and PWK/PhJ, n = 3 to 5 mice per strain per day). In contrast, the urea cycle metabolite and nitric oxide synthase product citrulline varied independently of arginine depletion and ALT. Further analysis of founder strain samples revealed a positive correlation between plasma arginase activity and plasma ALT ( Plasma arginase activity is increased in malaria patients, but it has been attributed to other sources of Arg1 such as red blood cells or monocytes (44,45). The other mammalian arginase arginase-2 (Arg2), which is highly expressed in immune cells as well as the gut and kidney (51)(52)(53), is rarely considered (45). To determine the isoform and tissue source of plasma arginase during P. chabaudi infection, we utilized a globally Arg2-deficient and two tissue-specific Arg1-deficient C57BL/6J mouse strains. Because murine Arg1 deletion results in lethality 14 days after birth (54), we crossed Arg1 floxed mice (Arg1 fl/fl mice [16]), with Tek-Cre mice, generating Arg1 fl/fl ; Tek-Cre mice that lack Arg1 specifically in blood cells and the endothelium. To overcome lethality caused by liver-specific Arg1 deficiency (55), we injected Arg1 fl/fl mice with adeno-associated viral (AAV) vectors that express Cre (or green fluorescent protein [GFP] for control animals) under the control of the liver-specific TBG promoter. This yielded Arg1 fl/fl ; TBG-Cre mice that display hyperargininemia and reduced plasma arginase activity at 2 weeks postinjection and that die from liver Arg1 deficiency at 3 weeks postinjection ( Fig. S3A and S3B).
We infected these knockout mice and their respective controls (n = 3 mice per genotype per experiment with two experimental replicates) with P. chabaudi and measured ALT and plasma arginase activity at peak infection intensity. As expected, uninfected mice of all genotypes exhibited low ALT and arginase activity. In infected, nonknockout mice, ALT and arginase activity positively correlated (Fig. 5D). ALT and arginase activity also correlated in Arg2 2/2 and Arg1 fl/fl ; Tek-Cre mice, suggesting neither Arg1 in the blood and endothelium nor Arg2 contribute to plasma arginase activity during P. chabaudi infection. In contrast, Arg1 fl/fl ; TBG-Cre mice exhibited high levels of ALT but significantly reduced plasma arginase activity relative to infected Arg1 fl/fl ; TBG-GFP mice (Fig. 5). This suggests hepatic Arg1 is the source of plasma arginase activity in P. chabaudi-infected mice.
Low plasma arginine is associated with increased vascular stress (44) and negative pregnancy outcomes in malaria patients (46). However, arginine infusion in malaria patients sometimes fails to restore arginine levels and mitigate vascular stress (56), motivating further exploration of the causes of hypoargininemia. Plasma arginase is associated with arginine depletion in malaria (44,45), which is consistent with our hypothesis that hepatic Arg1 depletes arginine. Infection-induced anorexia and altered flux of arginine through the urea cycle are other explanations that accounted for some but not all arginine depletion in mice with experimental cerebral malaria (57). To assess the importance of these factors on plasma arginine dynamics, we restricted food intake in uninfected Arg1 fl/fl ; TBG-Cre and Arg1 fl/fl ; TBG-GFP mice to the level of age-and sex-matched P. chabaudi-infected mice (Fig. S3C, n = 4 infected and 3 uninfected mice per sex, sampled longitudinally). Despite comparable weight loss in infected and uninfected mice (Fig. S3C), decreases in their plasma arginine during dietary restriction were relatively modest; only infected mice displayed dramatic hypoargininemia (Fig. 5E). Arginine depletion correlated strongly with plasma arginase activity in both genotypes, even in Arg1 fl/fl ; TBG-Cre mice with significantly reduced arginase activity (Fig. 5F), suggesting that even residual amounts of circulating arginase are sufficient to reduce plasma arginine by 100-fold or more. Finally, (C) Plasma arginase activity (nanomoles of H 2 O 2 /minute/microliter of blood) and scaled imputed ion counts of arginine (log 10 -transformed Zscores) in seven of eight P. chabaudi-infected founder strains (n = 6 per strain). A Loess curve with standard errors (gray shaded area) is shown. (D) Plasma arginase activity (nanomoles of H 2 O 2 /minute/microliter of blood) and ALT in arginase knockout C57BL/6 mice before and during P. chabaudi infection (n $ 3 uninfected mice per genotype, n $ 6 infected mice per genotype in two experimental replicates). Statistical significance by Welch's t test is indicated as follows: *, P , 0.05; ns, not significant. given the dramatic elevation of plasma arginine following knockdown of hepatic Arg1 ( Fig. S3A and Fig. 5E) (58), it is unlikely that this type of urea cycle disruption contributes to arginine depletion. Collectively, our data suggest that hepatic Arg1 maintains plasma arginine homeostasis and that circulating hepatic Arg1 depletes plasma arginine during P. chabaudi infection.
Metabolic changes in murine malaria recapitulate facets of human malaria. Serum metabolites change dramatically in human malaria patients, which led us to ask if P. chabaudi infection shares metabolic features with P. falciparum infection. However, prior to a direct comparison between human and mouse metabolic changes, we must acknowledge differences between human malaria and our murine model. First, human and rodent infections are caused by different parasites, P. falciparum and P. chabaudi, respectively. P. falciparum can cause cerebral malaria and death in humans (2), while lethal cerebral malaria in mice is restricted primarily to Plasmodium berghei infection models; in contrast, P. chabaudi-induced death is thought to be due to severe anemia. Furthermore, P. falciparum infections are self-resolving while mice continue to carry P. chabaudi, albeit at dramatically reduced levels. Of particular relevance for vascular health, P. falciparum protein PfEMP1 mediates cytoadherence of parasites to the vascular endothelium (3), which is associated with inflammation and vascular stress. Less is known about cytoadherence in P. chabaudi, though it is thought to occur (59). Finally, natural Plasmodium infection begins when parasite sporozoites infect the host liver, while our murine model bypasses the sporozoite stage. Nonetheless, the sporozoite stage of infection is largely symptomatically silent. Liver damage in human malaria occurs later in infection, during the acute blood stage and concurrently with parasite clearance (7). This is consistent with our findings in P. chabaudi-infected mice (Fig. 1). Similarly, most metabolic studies of humans, including the data presented below, sample the blood stage of infection when severe-disease-like liver dysfunction and acute kidney injury are most evident.
To metabolically compare murine and human malaria, we started with publicly available metabolic data from two populations of P. falciparum-infected individuals: Thai adults with uncomplicated malaria (60) (n = 10) and Malawian pediatric patients with severe cerebral malaria (20) (n = 10). Using metabolites that were measured in both human studies and our mouse study (n = 42, predominantly amino acids and lipids), we created a network using the correlation matrix of mouse samples at peak disease severity (using median metabolite values from five mice per day) and all human samples (Fig. 6A). In the network, mice with mild disease were better connected to Thai adults with uncomplicated malaria, while mice with severe disease correlated better with Malawian pediatric patients with severe disease. A subset of mouse samples (C57BL/6 at peak infection severity) was highly connected to samples from both human data sets, suggesting they shared similarity with both human populations.
To understand which metabolic changes drove interspecies similarity in the correlation network, we compared the levels of individual metabolites across our mice and the human samples ( Fig. 6B and Fig. S4). In most mice and humans, infection significantly increased levels of phenylalanine and kynurenine and decreased levels of arginine, glycine, serine, and long-chain glycerophosphatidylcholines (GPCs). Notably, some metabolites like valine and 1-linoleoyl-GPC changed more in severe disease (Malawian patients and less resilient mouse strains [red]) than in mild disease (Thai patients and resilient mouse strains [blue]). For many other metabolites, the two human data sets qualitatively differed, while mice tended to either match one population or display intermediate responses (Fig. 6B). For example, ornithine was elevated in Thai patients and our mice but depleted in Malawian patients. Conversely, valine was elevated in Malawian patients and mice but relatively unchanged in Thai patients.

DISCUSSION
Host metabolism changes in response to infection, but we still know little about how host metabolism responds to severe versus mild disease. To better understand how disease severity affects host metabolism in murine malaria, we monitored disease severity metrics (Fig. 1) and plasma metabolites in eight genetically diverse mouse strains that vary widely in survival following acute Plasmodium chabaudi infection. While P. chabaudi primarily causes nonfatal severe anemia due to high parasitemia, we observed a wide range of other disease severity phenotypes in which parasitemia, anemia, temperature loss, weight loss, liver and kidney injury, and hypoglycemia were generally more severe in mice that succumb to infection. We also observed a wide range of metabolic responses and used PCA to show that the magnitude of metabolic response scaled with overall disease severity (Fig. 3).
Previous human studies (e.g., references 4, 5, and 21) also reported malaria severitydependent changes in host metabolism. To our knowledge, Leopold et al. provide the most comprehensive picture of these changes in Bangladeshi patients, both in terms of the number of metabolites assessed and in number of facets of disease severity examined, including acute kidney injury, metabolic acidosis, and coma (4,5). Our work complements these studies by including additional metrics of disease severity with quantitative data for the degree of disease severity. Using our approach, we showed that malarial mice and humans share some plasma metabolic changes during infection (Fig. 6). These conserved metabolic changes fell into one of two broad categories: responses consistent in all three populations, and those shared by mice and just one of the two human populations.
All hosts shared some malaria-induced metabolic changes. Kynurenine and phenylalanine were elevated in all individuals, mouse and human. Kynurenine is a tryptophan breakdown product and ligand of the aryl hydrocarbon receptor (AhR), which protects the host in multiple models of malaria and sepsis (19,61,62). Phenylalanine is also elevated during infection, as in other malaria studies (4, 63, 64). Phenylalanine elevation may be caused by increased protein catabolism (57) and malaria-induced depletion of tetrahydrobiopterin (63), a cofactor required for enzymatic conversion of phenylalanine to tyrosine (65). Arginine was depleted in most individuals, which is consistent with all mouse and human studies we examined (4,5,20,21,43,44,57,66). Finally, most infected mice and humans had depleted long-chain GPCs (also observed in references 21 and 22), which may reflect changes in cell membranes, cell membrane lysis, or altered lipid metabolism. The nonessential amino acids glycine and serine, which P. falciparum uses in folate derivative-dependent DNA synthesis (67), were also depleted. Given the similarities we identified between humans and mice, our work supports the use of P. chabaudi-induced murine malaria as a model to understand host metabolic responses to malaria.
For a number of metabolites, mice and one human population differed from the second human population. For instance, ornithine was generally higher in our malarial mice and in Thai adults with uncomplicated malaria (60) but low in Malawian children with cerebral malaria (20). We anticipate elevated plasma arginase explains ornithine elevation in malarial humans as it does in our murine data. Like ornithine, long-chain acylcarnitines (ACs) (e.g., C16 and C18) were also elevated in malarial mice and Thai patients but depleted in Malawian patients (data not shown). We found that longchain ACs correlated with many pathologies in our model, including liver and kidney dysfunction and temperature loss. Others have also noted elevation of long-chain ACs following acute kidney injury, fasting, and LPS injection in mice (see Fig. S2 in the supplemental material) (34)(35)(36). Fasting elevates long-chain ACs and is usually accompanied by an increase in fatty acid b-oxidation gene expression (36). However, in infection (34) and long-chain fatty acid oxidation disorders (68), b-oxidation is suppressed. Measuring fatty acid b-oxidation gene expression in P. chabaudi-infected mice would help determine whether elevated long-chain ACs in our model indicate a normal fasting response or impaired b-oxidation. We also found that the branched-chain amino acids (BCAAs) leucine, isoleucine, and valine were elevated in mice with severe disease. Leopold et al. (4) also noted BCAA elevation in patients with severe but not uncomplicated malaria, which is consistent with our mice and the two human data sets examined here. BCAAs are altered in fasting states in both mice and humans (36). Though not present in either human data set, a-hydroxyisovalerate was a sensitive marker for hypoglycemia and weight loss in our murine data, and it is also indicative of defective leucine metabolism (33). Collectively, these data suggest infection-induced anorexia or changes in energy metabolism produce a strong molecular signature in severe malaria.
Our data suggest that diverse mice capture more of the diversity of human malaria responses than one human population alone can. However, the small number of metabolites (42 metabolites), patients (n = 10 for each of two human data sets), and data sets (3 data sets) included in this study limit our ability to make broad generalizations. There is significant variability in metabolic responses in humans (Fig. S4), which may be due to any number of confounding factors, including differences in age, geography, malarial syndrome, disease severity, study design, and instrumentation. We selected these data sets because they had the highest number of metabolites available for comparison with our murine data, but we recognize that we cannot definitively establish causal links between disease severity and metabolic responses in these data sets. To increase robustness in linking metabolic change with disease severity in humans, future studies would benefit from measuring samples from multiple mouse and human populations in tandem, alongside careful measurements of disease severity in both species like we present in Fig. 1 and like those presented in the works of Leopold et al. (4,5).
In addition to metabolites that generally indicate severe disease, several molecules from our study associated with individual pathologies (Fig. 4). Two of these correlations corroborated known associations: pipecolate as a biomarker for parasitemia (38), and N,N,N-trimethyl-alanylproline betaine (TMAP), which correlated with the kidney injury marker creatinine and was recently identified as a sensitive marker of kidney dysfunction (39). Other correlations like the association of 3-methylhistidine with weight loss and the correlation between a-hydroxyisovalerate and hypoglycemia are consistent with previous reports of muscle protein catabolism elevating plasma 3-methylhistidine (40) and of defective leucine metabolism causing hypoglycemia and a-hydroxyisovalerate elevation (33). Together, these findings support the validity of our correlation method and motivate further study of associations like dimethyllysine and red blood cells (Fig. 4B), N 6 -threonylcarbamoyladenosine (t 6 A) and creatinine (Fig. 4B), acylcarnitines (Fig. 4B) as discussed above, and others.
We demonstrated that our bioinformatic approach yielded precise and testable mechanistic hypotheses. Specifically, our correlation analysis pointed to the liver, rather than red blood cells, immune cells, or the parasite as the source of arginine-depleting arginase ( Fig. 4B and C). We tested this hypothesis by measuring plasma arginase and arginine in liver-specific Arg1 knockout mice, blood-and endothelium-specific Arg1 knockout mice, and Arg2 knockout mice on a C57BL/6 background. In support of our bioinformatic data, we found that plasma arginase activity in malaria was reduced only in liver-specific Arg1 knockout mice. These results challenge previous hypotheses about red blood cell, monocyte, or endothelial sources of Arg1 in malaria. Plasmodium parasites also carry genes that encode an arginase that significantly depletes arginine in vitro (69) but not plasma arginine in vivo (57). Consistent with these studies, our results suggest that the host, rather than the parasite, is the source of plasma arginase and arginine depletion in malaria. This, along with the relatively few strong parasite-metabolite correlations we observed, suggests the host is more directly responsible than the parasite for driving plasma metabolic changes in acute malaria.
Many studies observe decreased plasma arginine during infection (2, 4, 5, 22, 43-46, 56, 57, 70, 71) and even attempt to restore levels of this vasoprotective amino acid (43,56,66), but the cause(s) of arginine depletion in malaria is still debated. Anorexia and urea cycle changes accounted for some but not all infection-induced plasma arginine depletion in mice with experimental cerebral malaria (57). Our data support a minor role for anorexia in depleting plasma arginine and suggest that disrupting the urea cycle by hepatic Arg1 knockout does not deplete arginine but rather increases it. Instead, our data suggest that even small amounts of circulating hepatic Arg1 can dramatically deplete plasma arginine in diverse mice. Thus, arginase may be relevant even at the low concentrations seen in the P. berghei model (72,73).
We did not identify any single factor that is capable of maintaining normal arginine levels in malarial mice; even Arg1 fl/fl ; TBG-Cre mice (with significantly reduced plasma arginase activity) eventually suffered from hypoargininemia. Thus, despite our strong correlative evidence for small amounts of circulating hepatic Arg1 depleting arginine, this raises the possibility that there remains an unidentified factor that is the direct cause of arginine depletion. To better delineate the arginine-depleting role of hepatic Arg1 relative to other factors, future studies may benefit from careful kinetic analysis of arginine depletion following arginine infusion in P. chabaudi-infected Arg1 fl/fl ; TBG-Cre and Arg1 fl/fl ; TBG-GFP mice, which theoretically differ only in their levels of circulating plasma arginase.
Like arginine depletion (Fig. 6), liver damage is prevalent but not universal in malaria patients (7). Failure to account for differences in liver damage and plasma arginase activity may explain why arginine infusion does not always restore plasma arginine in malaria patients (43,56). While liver damage and plasma arginase activity did not predict survival across the eight mouse strains used in our study, increased plasma arginase and arginine depletion are associated with pathologies we did not measure, including vascular stress (44,74), negative birth outcomes in pregnant women (46), and gut vascular permeability in Plasmodium-Salmonella coinfection (71). Collectively, our data motivate further study of the relationships between liver damage, arginase, and vascular health in infectious disease. C57BL/6 and CAST/EiJ mice, which had high and low ALT, respectively, may provide a useful model for those studies.
Future directions. Should liver damage emerge as an important player in hypoargininemia in human malaria, NETosis (75) and IL-1-mediated inflammation (76) may be therapeutic targets for limiting liver damage. Given the dramatic differences in plasma immune cell and cytokine responses between C57BL/6 and CAST/EiJ mice (Fig. 2), these two strains may provide a natural study system for further dissection of immunopathology in the liver. Future efforts might also determine why some nonresilient mouse strains had weak overall cytokine responses, while others had hyperinflammatory responses. These data point to a trade-off between effective immunity and immunopathology in malaria that should be explored. We characterized other strain-specific responses that suggest non-C57BL/6 mouse strains will be the best model for some malarial host responses. The P. chabaudi model does not recapitulate some aspects of P. falciparum infection, perhaps most notably cerebral symptoms. Future metabolic studies would benefit from using the P. berghei murine model of cerebral malaria. We used genetically diverse mice because they displayed high phenotypic diversity, but we did not determine how genetics contributed to phenotypic diversity. Quantitative trait locus mapping in malarial DO mice would achieve that goal. Finally, the insights we gained in this study relied heavily on correlative data that do not link processes across time. To identify metabolic processes that connect early, acute, and late parts of the host response (19), longitudinal metabolic sampling of diverse mice would enable use of predictive tools like hidden Markov models.
Implications and conclusion. We saw that humans and diverse mice exhibit variation in metabolic responses to malaria that depends on disease severity. This implies that future metabolic studies of infection should account for both the nature and severity of disease. We used diverse mice to highlight the variability of metabolic responses and disease severity following infection and used this variability to yield new insights about liver damage and arginine metabolism. P. chabaudi is not a perfect model of human malaria, but it recapitulates some important facets of human disease. We view our work with diverse mice as a first step toward determining which human malarial metabolic changes can be studied in the P. chabaudi model. Collectively, our data provide new bioinformatic and mechanistic insights into infection metabolism and motivate the use of genetically and phenotypically diverse mice in future studies.

MATERIALS AND METHODS
Ethics statement. Mouse studies were conducted at Stanford University in accordance with NIH guidelines, the Animal Welfare Act, and federal law. They were approved by the Animal Care and Use Committee. Stanford University is accredited by the International Association for Assessment and Accreditation of Laboratory Animal Care (AALAC). All animal work was done according to protocols approved by Stanford University's Administrative Panel on Laboratory Animal Care (APLAC) and overseen by the Institutional Animal Care and Use Committee (IACUC) under protocol identifier (ID) 30923 (D. S. Schneider).
Parasites. Plasmodium chabaudi chabaudi AJ (Malaria Research and Reference Resource Center [MR4]) was tested for contaminating pathogens prior to use.
Mice. Wild-type mice were purchased from the Jackson Laboratory (Bar Harbor, ME, USA) (WSB/EiJ stock no. 001145, 129S1/SvImJ stock no. 002448, NZO/HILtJ stock no. 002105, CAST/EiJ stock no. 000928, A/J stock no. 000646, NOD/ShiLtJ stock no. 001976, and PWK/PhJ stock no. 003715), and Diversity Outbred [stock no. 009376] [24f] Charles River [C57BL/6]). A subset of expensive mice (WSB/EiJ, NZO/HILtJ, and PWK/PhJ) were also bred in-house. Animals were maintained specific pathogen free (SPF) and housed in the Stanford Research Animal Facility according to Stanford University guidelines, accredited by the Association of Assessment and Accreditation of Laboratory Animal Care (AAALAC) International. All mouse experiments were approved by the Stanford Administrative Panel on Laboratory Care (APLAC).
P. chabaudi infection. Two female C57BL/6 mice were given intraperitoneal (i.p.) injections of 100 ml frozen stock of P. chabaudi-infected red blood cells (iRBCs). When parasitemia reached 10 to 20% at 8 to 10 days postinfection, mice were euthanized, and blood was obtained via cardiac puncture. Blood was diluted to 10 5 iRBCs/100 ml in Kreb's saline with glucose (KSG) and administered i.p. to experimental animals at a dose of 10 5 iRBCs (78). Control animals received 100 ml of vehicle i.p. Only female mice 8 to 12 weeks of age were used for P. chabaudi experiments. Experiments were performed in multiple cohorts. Parasitemia was quantified via thin blood smear, methanol fixation, KaryoMAX Giemsa (Gibco) staining, and manual microscope counting at Â100 magnification. RBCs were quantified using a BD Accuri C6 Plus cytometer (see "Longitudinal P. chabaudi monitoring" below).
Longitudinal P. chabaudi monitoring. Longitudinal monitoring was performed as described previously (31). For each mouse, baseline RBC, weight, body temperature, and blood glucose measurements were collected between 1 and 5 days prior to infection. In some cases, blood glucose was collected only at baseline sampling and on the day of sacrifice. Mice were restrained during sample collection using tail-access rodent restrainers (Stoelting Co.). Blood was collected from the tail vein by nicking the end of the tail with disinfected surgical scissors, and depositing the blood into EDTA-coated capillary tubes to prevent clotting. For total RBC quantitation, 2 ml of blood was diluted in 1 ml of cold 1Â Hanks' balanced salt solution (HBSS) and kept on ice until absolute RBC counts were obtained using forward and side scatter gates on a BD Accuri C6 Plus flow cytometer. To record body temperature, mice in the metabolic screen experiments were implanted with subcutaneous electronic temperature and ID transponders (IPTT-300 transponders, Bio Medic Data System, Inc.) 1 week prior to infection. Mice were locally anesthetized using a 2% lidocaine solution (100 mg delivered per dose) prior to implantation. Temperature data were recorded using a DAS-7006/7 s reader (Bio Medic Data System, Inc.). Subsequent to metabolic screen experiments, body temperatures were measured using a thermocouple thermometer and mouse rectal probe (World Precision Instruments, RET-3). Blood glucose measurements were obtained with 2 ml of tail vein blood analyzed with a Bayer Contour Blood Glucose Monitor and Test Strips. Postinfection sampling began on day 4 or 5 postinfection. Parasitemia values were obtained as detailed above. Parasite density is the number of iRBCs per microliter of blood and is calculated by multiplying parasitemia by the number of total RBCs.
Flow cytometry. White blood cells, platelets, and reticulocytes were quantified in a subset of mice at baseline sampling and on odd-numbered sampling days from day 3 through day 11. Cells were first quantitated on an Accuri cytometer as detailed in "Longitudinal P. chabaudi monitoring." RBC counts were used as an approximation for total blood cell counts. All stains were performed for 12 to 15 min at 4°C. Five milliliters of CountBright counting beads (Invitrogen) were added to each sample such that absolute counts per milliliter of blood could be back calculated. Data were acquired on an LSR Fortessa (BD Biosciences) and analyzed using FlowJo (Tree Star). Prior to sample acquisition, splenocytes were obtained from a healthy mouse spleen, stained with anti-CD4 antibodies, and used for instrument compensation.
Plasma collection for cross-sectional analyses. Between three and five infected mice of each strain were euthanized each day from days 3 to 12 postinfection for cross-sectional analysis. Because the WSB/ EiJ strain experiences delayed peak infection severity relative to the other mouse strains in this study, three or four WSB/EiJ mice were euthanized each day from days 3 to 17 postinfection for cross-sectional analysis. For each mouse strain, two uninfected control animals were euthanized at baseline and generally on odd-numbered days between days 3 to 12 or days 3 to 17 for WSB/EiJ mice. Euthanasia was performed using carbon dioxide asphyxiation in accordance with Stanford University and APLAC guidelines for humane euthanasia. Following euthanasia, blood was collected via cardiac puncture using 25-gauge (25G) 5/8-in. tuberculin syringes (Fisher Scientific catalog no. 14-841-34). Syringes were primed by filling the syringe barrel with 0.5 M EDTA (pH 8.0) anticoagulant and dispensing all but 50 ml. Collected blood was stored on ice in 1.5-ml Eppendorf tubes for 15 to 45 min before spinning at 1,000 Â g at 4°C for 5 min in a tabletop centrifuge. Plasma was frozen at 280°C immediately and thawed/refrozen once to aliquot for downstream cytokine, metabolite, and liver enzyme analyses.
Cytokine analysis. Seventy-five microliters of plasma was sent to the Human Immune Monitoring Center at Stanford University. Mouse 38-plex kits were purchased from eBiosciences/Affymetrix and used according to the manufacturer's recommendations with modifications as described below. Briefly, beads were added to a 96-well plate and washed in a Biotek ELx405 washer. Samples were added to the plate containing the mixed antibody-linked beads and incubated at room temperature for 1 h followed by overnight incubation at 4°C with shaking. Cold and room temperature incubation steps were performed on an orbital shaker at 500 to 600 rpm. Following the overnight incubation, plates were washed as described above, and then a biotinylated detection antibody was added for 75 min at room temperature with shaking. The plates were washed as described above, and streptavidin-PE was added. After incubation for 30 min at room temperature, a wash was performed as described above, and reading buffer was added to the wells. Each sample was measured as singletons. The plates were read using a Luminex 200 instrument with a lower bound of 50 beads per sample per cytokine. Custom assay control beads by Radix Biosolutions were added to each well. Samples in which .50% of cytokines returned low bead counts (#25) were filtered from the data set. Individual data points with a bead count of #25 were also removed. For each cytokine, raw median fluorescence intensities are reported as the number of standard deviations from the mean of all samples (Z-scores).
Metabolite analysis. One hundred microliters of plasma was shipped to Metabolon (Durham, NC, USA), which performed a combination of gas and liquid chromatography with mass spectrometry (GC/ LC-MS). Compounds were identified by comparing sample peaks to an internal Metabolon library of known and unknown compounds. Raw peak values were obtained using area-the-curve. Additional data normalizations were performed by Metabolon to account for sample dilutions and day-to-day variation in instrument performance. The two murine metabolomics experiments in this study (the eight-strain P. chabaudi experiment and the Ahr 2/2 P. chabaudi experiment) were measured in two separate Metabolon sample batches. PHE), glutamine (GLUT; IS = d5 GLUT), monosodium glutamate (MSG; IS = d5 GLUT), proline (PRO; IS = d8,15N PHE). Standards were obtained from Cambridge Isotope Laboratories (Tewksbury, MA, USA).
(ii) Internal standard preparation. All reference standards were commercially available in powder form. Stock solutions for each standard (10 to 200 mM) were prepared by solubilization in water and diluted further with water-acetonitrile to 200 mM working solutions.
(iii) Sample preparation. For plasma amino acid quantitation, 25 ml of frozen plasma was sent for analysis. Five microliters of internal standard solution was added to 20-ml plasma aliquot followed by vortexing. Seventy-five microliters of an ice-cold solution of acetonitrile20.1% formic acid was added to the sample, followed by vortexing and then centrifugation. Supernatant was transferred to a new vial and analyzed by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS).
(iv) Calibration curves. Individual analyte primary stock solutions were prepared in water (10 to 200 mM). Intermediate stock solution (200 mM) consisting of all unlabeled analytes (ARG, ORN, CIT, ADMA, SPD, PHE, MSG, GLUT, and PRO) was prepared from individual primary stock solutions. This intermediate stock solution was serially diluted with 50Â diluted charcoal stripped plasma (CSP) to obtain a series of standard working solutions which were used to generate the calibration curve. Standard working solutions were prepared freshly for sample analysis. Calibration curves were prepared by spiking internal standard solution (10 mM) consisting of six labeled compounds (13C6, 15N4 ARG; d7 ORN; d7 ADMA-HCl; d8 SPD; d8, 15N PHE; d5 GLUT). Because of interference due to endogenous metabolites, calibration curves were prepared in diluted CSP to closely match the study samples. A calibration curve was prepared fresh with each set of samples, and it ranged from 0.2 nM to 4,000 nM.
(v) Instrumentation. All analyses were carried out by high-resolution LC-MS/MS using an Agilent Poroshell 120 hydrophilic interaction liquid chromatography (HILIC)-Z column (2.7-mm particle size, 3.0 Â 100 mm) on a Quattro Premier triple quadrupole mass spectrometer (Waters) coupled with a 1100 high-performance liquid chromatography (HPLC) system (Agilent). The HPLC conditions were as follows. The column was operated at 30°C at a flow rate of 0.45 ml/min. Mobile phases consisted of solution A (40 mM ammonium formic acid [pH 3.2] in water) and solution B (40 mM ammonium formate [pH 3.2] in 90% acetonitrile). The elution profile was as follows: 90% solution B for 1.5 min, followed by a gradient from 90% to 30% in 8.5 min, then 30 to 95% in 1 min, and held at 90% solution B for a total run time of 16 min. The injection volume was 5 ml.
Arginase activity assay. Arginase activity assays were performed using a colorimetric assay (Abcam catalog no. 180877) according to the manufacturer's instructions with the following modifications: hydrogen peroxide standards were not incubated at 37°C; they were prepared immediately prior to addition of reaction mix. Prior to assay, plasma samples (2.5 to 10 ml) were loaded into Amicon 10-kDa spin filters (Fisher Scientific catalog no. UFC501096). Four hundred microliters of water was added, and filters were spun at 15,500 Â g at 4°C for 10 min in a tabletop centrifuge. After discarding flowthrough, this step was repeated for a total of two spins. Retentate was resuspended in arginase activity buffer (provided by Abcam [catalog no. 180877]) to a final ratio of 2.5-ml original sample to 100-ml total volume. Spin-filtered samples were kept on ice and assayed the same day. To control for interassay variability, arginase activity values were adjusted so that the negative-control sample for each batch (an aliquot from the same uninfected plasma sample) had arginase activity of zero.
Data availability. Data for founder strain experiments (survival, disease severity, metabolomics, cytokines, flow cytometry) are provided in Data Set S1 in the supplemental material. Metabolomics data for the founder strains are also available at Metabolomics Workbench, doi:10.21228/M8RX1D, under title "Plasma metabolomics of diverse mouse strains infected with Plasmodium chabaudi." Data for Fig. 4, C57BL/6 wild type (WT) versus Ahr 2/2 data, are from reference 19. Other metabolic data, human and animal, were obtained from existing publications as noted in the text. Data for the remaining figures, and scripts for all figures/tables, are maintained at Github (https://github.com/dave1618/strainsMetabspaper/).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. DATA SET S1, XLSX file, 9.