Our experiment to examine trade-offs between infection, immunity, and within-host resources (i.e., diet quality) included four treatment groups: high-protein infected, high-protein uninfected, low-protein infected, and low-protein uninfected. Eighty-eight female C57Bl/6 mice aged 5–6 weeks were obtained from Jackson Laboratories and individually identified with both ear tags and RFID tags (see below). All animal care was conducted in accordance with protocols approved by the Princeton University Animal Care and Use Committee (Protocol no. 1982-14). Mice were housed in groups of five in the laboratory and randomly assigned to the two diet treatments as well as two cohorts that were staggered by 2 days for logistical purposes (Figure 2). The high-protein (HP; 20%; Envigo Teklad Custom Diet TD.91352) and low-protein (LP; 6%; Envigo Teklad Custom Diet TD.90016) diets had the same energy density (3.8 kcal/g) and micronutrient composition. The typical chow fed to lab mice (e.g., PicoLab^® Rodent Diet 20) has very similar composition to the HP diet. For 10 days mice were fed the assigned diets in the lab while temperature and light cycles were gradually altered to mimic outdoor conditions (June/July in New Jersey, USA: 26 ± 1°C with a 15-h light–9-h dark cycle; Figure 2).

Next, the 22 mice in each diet–cohort combination were transported to four outdoor enclosures (Figure 2), two of which contained the HP chow and the other two the LP chow (Figure 2 inset). The enclosures are replicate pens of approximately 180 m², fenced in by zinced iron walls extending 1.5-m high and 80-cm deep, and topped with electric fencing and reflective aluminum pans to deter ground and aerial predators, respectively (See text footnote 1). Diets were provided ad libitum at feeding stations monitored by two RFID readers to track each individual’s time spent feeding. The natural environment could serve as an additional source of food (e.g., berries, seeds, insects). Each enclosure provided two watering stations inside a small (180 cm × 140 cm × 70 cm) straw-filled shed (See text footnote 1). Mice were weighed at the start of the experiment, the day of release, and approximately weekly thereafter as they were trapped overnight in the outdoor enclosures using chow-baited Longworth traps. At each weekly trapping, fecal samples were collected and blood samples were taken via shallow cutaneous tail snips into heparinized capillary tubes.

Mice were acclimated to the enclosures for 2 weeks (14–17 days) prior to T. muris infection (Figure 2). A dose of 200 embryonated T. muris eggs (strain E) was then given via oral gavage to the first 16 mice trapped per enclosure. If more than 16 mice were trapped on a given day, individuals were randomly assigned to infection treatment. In total, 29 mice on the HP diet and 31 mice on the LP diet were infected. Thus, infected and uninfected animals were cohoused in the same enclosures. The remaining mice (15 HP, 13 LP) served as uninfected controls. Nematode infection could not be transmitted between mice assigned to different treatments in the shared enclosures due to the long life cycle of T. muris [>28 days to maturity (40)], and the relatively short duration of the experiment [<26 days postinfection (dpi)].

Approximately 3 weeks postinfection (mean ± SE: 21.5 ± 1.7 dpi), range: 19–26 (dpi), all mice were trapped, weighed, and transported back to the laboratory (Figure 2). Mice were anesthetized via isoflurane inhalation followed by terminal cardiac puncture. Cardiac blood was drawn from each mouse, spun in a heparinized tube, and plasma was separated and stored at −80°C. During necropsy, mesenteric lymph nodes (MLNs) were excised for cell culture (see below). Finally, to determine carcass weight, all major organs (except the brain) were removed and weighed. The spleen, large intestine (emptied of contents), and cecum were separated and weighed. Ceca were frozen for later dissection and parasite enumeration.

Individual-level feeding behavior was assessed using a custom-built monitoring system (Datasheet S1 in Supplementary Material). To track feeding behavior, Radio Frequency IDentification tags (RFID; 8 mm × 1.4 mm FDX-B “Skinny” PIT Tag, Oregon RFID, Portland, OR, USA) were injected subcutaneously. When in the vicinity of an antenna, the RFID tag emits a unique string of numbers that serves as an individual’s identifier (i.e., RFID number). In each enclosure, Feeding Event Tracking Apparatuses (FETA) with RFID antennas were deployed. The FETA transmit a signal to an Event Acquisition and Reporting System (EARS), which automatically compiles the data produced by the FETA (FETA number, RFID number, and timestamp). Two FETA were placed in sequence and connected on one end to the sole entrance to the chow hopper.

The directionality of mouse movement through the sequential FETA boxes was used to determine when mice were in the feeder. Specifically, mice were inferred to be in the feeder for the duration of time between two EARS readings on the FETA nearest the chow hopper. Intervals <5 s in duration (7.5% of visits) were removed because they allowed insufficient time to feed. Intervals greater than 2 h were also removed because mice were likely not eating for such a long duration and this small fraction (<0.1%) of visits skewed feeding time distributions. Intervals when chow was not available in the feeder (i.e., during trapping sessions) were also dropped. Next, total time feeding was summed per mouse, then divided by the number of days each mouse was in the enclosure because some mice were in the enclosures for longer than others (35–40 days depending on when they were trapped). This produced a comparable measurement of time spent feeding (min) per experiment day. Mice that lost their RFID tags (n = 6) and two individuals that did not eat at the feeder (working RFIDs but stopped visiting feeder) were excluded from feeding behavior analyses. Chow consumption was monitored by weighing each chow hopper every time it was removed for trapping as well as before and after refilling (i.e., every 1–3 days).

Following established protocols (See text footnote 1, 41), MLN tissue was excised during necropsy, prepared into single-cell suspensions at a density of 5 × 10⁶ cells/mL, stimulated with T. muris antigen at 5 µg/mL, and cultured for 48 h. Supernatants were collected and analyzed in duplicate for interleukin 13 (IL13), interleukin 10 (IL10), interleukin 17 (IL17), and interferon gamma (IFNg) using half-reactions of Beckton Dickinson Cyometric Bead Array Mouse/Rat Soluble Protein Flex Set system (BD Biosciences, Oxford, UK) and an LSRII flow cytometer (BD Biosciences) (See text footnote 1). Concentrations were analyzed with the FCAP Array software (version 3.0.1, BD Biosciences).

Worm-specific IgG1 titers were measured using sandwich enzyme-linked immunoassays (ELISA). Ninety-six-well plates were coated with T. muris excretory-secretory (ES) antigen at a concentration of 5 µg/mL in carbonate buffer (0.06 M, pH = 9.6) with 2% bovine serum albumin (BSA). Plates were then blocked overnight with 2% powdered milk in carbonate buffer. After a 2-h incubation at 37°C, plates were washed three times with tris-buffered saline-Tween (TBST). Plasma samples were added and serially diluted from 1:50 to 1:6,400 with TBST and incubated for 2 h at 37°C. After washing five times with TBST, horseradish peroxidase (HRP) conjugated IgG1 antibody (1:4,000 in TBST) was added. Following a 1-h incubation at 37°C, plates were washed five times and Substrate ABTS (Sigma-Aldrich) was added. Plates were developed for 20 min at 37°C, then stopped with a 1% sodium dodecyl sulfate solution. Absorbance was measured at 405 nm using a Multiscan™ GO spectrophotometer (Thermo Scientific). Titers were determined by the dilution at which absorbance exceeded 4 SDs above background, defined as the plate-wide average absorbance of the two most dilute concentrations of each sample.

To quantify total IgG, a mouse antibody IgG pair was purchased from StemCell Tech (Catalog no. 01998C). Plates were coated at 1 µg/mL 50 μL per well with carbonate buffer (pH 9.6) overnight at 4°C. Plates were blocked for an hour with TBST 20 with 0.1% BSA at 37°C, then washed. Preliminary assays revealed that IgG levels were quite high, so to fall in the range of the standards, samples were diluted 1:81,920 in TBST containing 0.1% BSA. After incubation for 2 hat 37°C and washing, a secondary antibody was applied at a concentration of 1:1,000 in TBST with 0.1% BSA. Plates were incubated at 37°C for 1 h then washed. Next, we added 100 µL/well of p-nitrophenyl phosphate (pNPP) substrate to all wells. Finally, plates were incubated for 30 min and read at 405 nm. Concentrations were calculated in comparison to a plate-specific standard curve (all R² > 0.999).

Caeca were cut open longitudinally and examined under a dissecting microscope by an observer blind to infection status. Worms were isolated by scraping the gut mucosa into a series of clean petri dishes of water. After enumeration, worms were stored in 70% ethanol for subsequent length measurements. Each worm was photographed under a dissecting scope (2–4× power) and ImageJ (version 1.49, NIH, USA) was used to measure total length.

Plasma albumin concentrations were measured colorimetrically. QuantiChrom BCG albumin assay kits (BioAssays) were used following the manufacturer’s instructions. Briefly, 5-µL duplicates of each standard and sample (diluted 1:2 with ultrapure water) were mixed with 200 µL of BCG reagent. Plates were incubated at 23°C for 5 min before being read at 620 nm. Concentrations were determined in comparison to a standard curve run in duplicate (R² = 0.997).

Plasma leptin concentrations were analyzed using a RayBio^® Mouse Leptin ELISA kit following manufacturer’s instructions (RayBiotech, Norcross, GA, USA). Briefly, 10 µL of plasma samples were diluted 1:10, incubated on a 96-well plate coated with mouse leptin antibody. After washing, a biotinylated anti-mouse leptin antibody was added. Next, the plate was washed, horseradish peroxidase-conjugated streptavidin was added. Following another washing step, a buffered 3,3,5,5′-tetramethylbenzidine solution was used to produce a color change reaction stopped after 30 min with 0.2-M sulfuric acid. Color intensity was read at 450 nm. Concentrations were calculated using a standard curve (standards run in duplicate, R² = 0.995).

To compare the wild plant species diversity and composition with which lab mice supplemented their diets, we employed a DNA metabarcoding method that involves sequencing and identifying undigested plant DNA obtained from fecal pellets (42, 43). The plant DNA in fecal pellets is likely to reflect very recent foraging activity because the half-life of ingesta in lab mice is approximately 74 min (44), meaning that <1% of contents are retained after 8 h. Thus, the resulting dietary profiles represent plants eaten over the ~2.5- to 5-h period prior to sample collection (44, 45).

Fecal samples from 26 mice (n = 7 HP inf, 9 LP inf, 6 HP uninf, 4 LP uninf) from the end of the study were obtained for dietary analysis. Samples were frozen on dry ice and stored at −80°C to preserve DNA prior to extraction. Total DNA from 1 to 2 pellets (~15–30 mg) per mouse was extracted using a Zymo Xpedition Soil/Fecal DNA mini kit with an extraction blank to monitor for potential cross-contamination. Using PCR, the P6 loop of the chloroplast trnL(UAA) marker was amplified with primers g and h (42). The PCRs were run with unique combinations of the g and h primers that had been modified with 8-nt multiplex identification (MID) tags in order to enable pooling of amplicons for sequencing using established protocols (46). These PCR products were cleaned and normalized using SequelPrep normalization plates, then pooled for sequencing in a 170-nt single-end run of the Illumina HiSeq2500 at Princeton University’s Lewis Sigler Institute following Kartzinel et al. (46).

The resulting DNA sequence data were assigned to samples of origin (i.e., demultiplexed), screened to reduce potential sequencing errors, and identified by comparison to a plant DNA reference library. The sequences were demultiplexed and primers were removed using the ngsfilter command in the Obitools software (47). We discarded sequences that contained ambiguous base calls (i.e., non-A, T, C, or G characters) that were <9-nt long or that had mean Illumina quality scores of <32. Unique sequences were merged and tabulated within samples to permit quantification of DNA sequence relative read abundance (RRA), a measure of the proportion of each dietary plant species in each dietary sample. Putative errors were screened by using the obiclean command to identify sequences that differed by 1 nt from another sequence in the same sample, but that occurred at <5% of the abundance. These sequences were removed from further analyses. Plant DNA was identified by comparison to a reference library developed using the European Molecular Biology Laboratories (EMBL) (Database release no. 130). From this archive, we extracted 35,776 unique sequences (229,430 entries) with ≤3 mismatches to the trnL-P6 primers g and h. Unique dietary sequences were identified by comparison to this reference database using the ecoTag command. Operational taxonomic units (OTUs) were identified by clustering at the >97% level using the uclust algorithm (48). To focus on abundant and well-identified plant taxa, we considered only OTUs with >80% identity to the EMBL database and those representing >5% of reads within each sample. Samples were rarefied to even sequencing depth using phyloseq (49) in R (50).

We inspected taxonomic identifications to identify imprecision that could arise from gaps or misidentified DNA sequences in the reference library. Any OTUs that exactly matched a taxon not known to occur in the region were revised to higher taxonomic levels, and the set of references matching any OTUs that were poorly identified (family or higher taxonomic levels) were scrutinized for taxonomic outliers (Table S2 in Supplementary Material). Only two plant taxa were included in the manufacturing process of the chow provisioned to lab mice in this experiment were corn (Zea mays) and beets (Beta vulgaris), and DNA from these plant products was expected to be destroyed by irradiation during the chow production process; indeed, no DNA sequences that match either of these plant species were identified in the final set of OTUs from our analysis. We calculated RRA by converting the rarefied number of reads into a proportion of reads per sample (i.e., ranging 0–1). Although RRA is not always a reliable measure of proportional dietary utilization in DNA metabarcoding studies (51), the analysis of RRA based on the trnL-P6 protocol employed in this study has been supported at least to the level of plant family and functional group in independent studies from different systems (46, 52, 53).

Nuclear magnetic resonance spectroscopy (NMR) was used to examine dietary metabolites from a random subsample of uninfected mice (12 HP and 8 LP) with sufficient frozen fecal samples from the end of the experiment. Fecal pellets from the mice were weighed, crushed, and diluted 1:3 with phosphate buffered saline (PBS). Samples were vortexed, allowed to dissolve overnight, and re-vortexed. After centrifugation, the supernatant was decanted and brought to ca. 30 µL with PBS in a 1.7-mm OD capillary tube (New Era Enterprises, Vineland, NJ, USA). This capillary was inserted into a 5/2.5-mm OD NMR tube containing small amount of D₂O, which served as an external lock material. Samples were analyzed on an 800-MHz Bruker Avance III HD NMR spectrometer equipped with a custom-made ¹H/¹⁹F/¹³C/¹⁵N//²H cryoprobe using Topspin v.3.2. Water suppression was done using the first increment of the NOESY pulse sequence (delay–G1–90^o–t₁–90^o–t_m–G2–90^o–acquisition) applying weak presaturation during the relaxation delay and the 100-ms mixing time period. Data processing was done offline using MNova v.11.0 (Mestrelab Research, Santiago de Compostela, Spain). After zero filling, apodization with 1-Hz Gaussian broadening, careful phase and baseline correction were applied manually. In order to compensate for variance of concentrations the spectral intensities were normalized to total integral, excluding the small region of the residual water signal. Local peak alignment was applied where necessary and possible using the inherent function in MNova. A few key metabolites were identified based on literature data (54). Out of the 20 samples, two were discarded because of technical problems (poor shimming and water suppression) to maintain statistical integrity of the residual data. The collection of 18 spectra was then exported to a spreadsheet.

The final sample size of this study was 80 (LP infected = 31, LP uninfected = 10, HP infected = 24, HP uninfected = 15). Several mice eluded capture for over 20 days beyond when the rest were trapped and sampled; these were excluded from statistical analyses. An additional mouse that had a severe congenital uterine defect was also excluded.

We assessed effects of diet and infection on immunity and condition. No uninfected mice had detectable IL13 concentrations, so a Kruskal–Wallis rank sum test was first used to compare infected and uninfected mice. The other cytokines (IFNγ, IL10, IL17) also had highly skewed distributions that could not be normalized. Thus, effects of diet and infection were analyzed using Kruskal–Wallis tests. Next, a general linear model (GLM) was used to test the effect of diet on log-transformed IL13 concentration among infected mice. GLMs were also used to test effects of diet and infection on log-transformed worm-specific IgG1, total IgG, and spleen weight relative to carcass weight. Using GLMs, we next tested for effects of diet and infection on mouse condition, including weight gain per day, plasma albumin concentration, plasma leptin concentration, and carcass weight. Plasma albumin and leptin concentrations were log transformed for the analysis. Diet–infection interaction terms were dropped when non-significant. To account for potential correlation structure among plasma components, we conducted a principal component analysis (PCA) of worm-specific IgG1, total IgG, albumin, and leptin. All components were log transformed and scaled prior to analysis. Next, the relationship between diet and infection status and each principal component (PC) was tested using GLMs. This PCA yielded no new insights beyond the univariate analyses described above, and results can be found in Table S1 in Supplementary Material.

To assess differences in fecal metabolites between the diet treatments, the SIMCA-P v.12.1 software package was used (Umetrics, Umea, Sweden). To determine if there was spectrum-wide variation between the diets, partial least squares-discriminant analysis (PLS-DA) was performed and orthogonal atrial least squares-discriminant analysis (OPLS-DA) were performed. Widely used in metabolic phenotyping studies, the PLS-DA analyses are better able to detect clusters than traditional PCA, and OPLS-DA provides even stronger discrimination based on known groupings (e.g., diet) (55). The DA methods were validated by calculating R² and Q². Prior to PLS-DA and OPLS-DA analysis, both UV and Pareto scaling were applied. UV scaling is a better choice for maintaining uniform contribution from all peaks regardless of their absolute intensity, while Pareto scaling provides access to “spectrum-like” loadings plot.

To examine morphological and behavioral changes, we first tested for effects of diet and infection on large intestine size (tissue weight/carcass weight) using a GLM. Similarly, we used a GLM to determine if time spent feeding per day in the enclosures varied with diet or infection status. Third, we examined supplemental foraging on natural plants in the enclosures using fecal DNA metabarcoding data. We tested for significant differences in the composition of plant species eaten according to diet, infection status, and leptin levels (log-transformed) using permutational MANOVA (perMANOVA) in vegan (56) in R. For the perMANOVA, we calculated pairwise Bray–Curtis dissimilarity metrics for each pair of samples, which ranges from 0 to 1 (from completely identical to mutually exclusive dietary compositions, respectively). For the six wild-plant families with the greatest overall RRA, we compared mean RRA between treatment groups (infected vs. uninfected and LP vs. HP diets) and across levels of leptin. Exploratory trend lines were fit to the data using generalized linear modeling. We investigated differences in RRA between genera of the legume family (Fabaceae) in more detail because this family comprised greatest overall RRA.

Lastly, we tested for differences in parasite load and weight gain, corrected for the amount of time each individual spent feeding. Among infected individuals, a non-parametric Kruskal–Wallis test was used to determine if diet affected worm count. In infected individuals, we also tested for an effect of diet on worm length (log transformed for normality) with a GLM. The amount of weight each mouse gained while in the enclosures was divided by the amount of time each spent feeding. Weight gain per minute feeding was compared between diets and infection status using a GLM. All analyses were run in R version 3.1.2.

Measures of immune function were affected by both diet and infection. MLN cells of the uninfected mice produced no detectable IL13 [a cytokine strongly induced by nematodes (57)] in response to stimulation with nematode antigen, making them significantly lower than the infected mice (Kruskal–Wallis χ² = 7.52, df = 1, p = 0.0061; Figure 3A). Among infected mice, those on the high-protein diet had higher IL13 concentrations in culture supernatants than those on the low-protein diet [Est (±SE): 0.56 ± 0.25, t = 2.20, p = 0.033; Figure 3A]. The cytokines IFNγ and IL17 mediate pro-inflammatory responses, primarily to intracellular pathogens and extracellular bacteria and fungi, respectively (58, 59). IL10 mediates anti-inflammatory, regulatory responses (58, 60). Surprisingly, IFNγ, IL17, and IL10 were also higher in infected mice but did not differ between diets (Figure S1 in Supplementary Material). In mice and many other mammals, IgG1 is an important antibody class for fighting GI nematode infection (57). In addition to being higher in infected mice (p = 0.00006), T. muris specific IgG1 titers were elevated in mice eating the high-protein diet (p = 0.032; Table 2; Figure 3B). Total IgG concentrations did not vary with diet or infection status (all p > 0.33; Table 2; Figure 3C), so we cannot account for the elevated IgG1 titers observed in uninfected mice eating the high-protein diet as a correlate of high total IgG concentrations. Spleen size also did not differ significantly with infection status or diet (all p > 0.10; Table 2; Figure 3D).

Dietary protein affected multiple aspects of mouse nutrition and condition, whereas parasite infection did not. Although neither diet (p = 0.19) nor infection (p = 0.052) significantly affected rates of weight change (Figure 4A; Table 2), other condition measures were more sensitive. Mice on the LP diet had significantly lower albumin concentrations (p = 0.006; Figure 4B; Table 2) than those on the high-protein chow. The low-protein diet also led to elevated leptin concentrations, a metabolic and immune-regulatory hormone released in proportion to body fat, as well as to higher carcass weights (p = 0.044; Figures 4C,D; Table 2). Infection did not affect any of these other condition measures (all p > 0.15; Table 2).

Nuclear magnetic resonance spectroscopy provides data and information on metabolites at the molecular level. The average spectrum of the 18 fecal samples, after normalization, peak alignment, and identification of some components (54), is shown in Figure 5A. PLS-DA (UV scaled) revealed distinct clustering by diet (Figure 5B), with a high proportion of variance explained by PC1 and PC2 (R²Y = 0.993) and decent predictability (Q² = 0.533). The OPLS-DA (Pareto scaled) coefficient plot (Figure 5C) shows many positive and negative intensities, corresponding to metabolites in higher or lower relative concentrations in LP and HP diets, respectively. A great variety of sugars are present in the fecal extract, and aliphatic amino acids are clearly in higher abundance in the LP diet group (Figure 5C). The differences in metabolite relative abundances between the diet treatment indicates the LP diet produces clear alterations in the metabolism of the mice and/or their gut microbiota.

The morphology and behavior of the mice was affected by diet and infection. First, infected mice had significantly heavier large intestines (emptied large intestinal tissue, weight relative to carcass weight) than uninfected mice (p = 0.022; Figure 6A; Table 2). Diet and infection status affected cecum size, with larger ceca in infected mice on the HP diet (diet: p = 0.017, infection: p = 0.050; Figure 6B; Table 2). Additionally, mice on the LP diet spent more time feeding than those on the HP diet (p = 0.0004; Figure 6C; Table 2).

Our DNA metabarcoding strategy yielded a total of 1,433,421 demultiplexed sequence reads of high quality, 14,521 of which were unique. After removing putative sequencing errors and picking OTUs, our cleaned-up database comprised 3,017 OTUs representing 1,296,521 sequence reads (>90% of the original). Sequence counts per sample were uneven (range = 1,648–162,180, mean = 49,702), and all were much higher than the extraction blank (N = 229). After removing sequences that poorly matched the reference database and that were never >5% of reads in any sample, we were left with 82% of the raw DNA sequence reads (1,167,514; of raw DNA sequence reads). These raw DNA sequence reads represented 26 plant OTUs in subsequent analyses (Table S3 in Supplementary Material). We rarefied samples to an even depth of 1,301 sequences/sample. The six most heavily utilized wild plant families (RRA > 0.05; Table S3 in Supplementary Material) included legumes (Fabaceae, cumulative RRA across all samples = 0.32), grasses (Poaceae = 0.18), wood sorrel (Oxalidaceae = 0.12), roses (Rosaceae = 0.08), violets (Violaceae = 0.07), and asters (Asteraceae = 0.07).

Individuals varied considerably in the composition of wild plants eaten, but the overall composition of wild foods eaten did not differ between treatment groups. The composition of consumed wild plants did not differ by diet quality (perMANOVA: pseudo-F_1,22 = 0.83, R² = 0.03, p > 0.05) or infection status (pseudo-F_1,22 = 0.96, R² = 0.04, p > 0.05; Figure 7A) of the mice, at least in part reflecting the high inter-individual variation in diet composition (mean pairwise Bray–Curtis dissimilarity = 0.89). Diet compositions were more closely associated with leptin levels (pseudo-F_1,22 = 1.61, R² = 0.06, p = 0.06; Figure 7B), although this trend was marginally non-significant.

Nevertheless, some variation among groups in their proportional utilization of different plant types was apparent. Most strikingly, legumes (family Fabaceae) were virtually unutilized by LP-uninfected lab mice even though the mean RRA of legumes eaten by mice in other treatments ranged from ~0.3 to 0.4 (Figure 7A). Within Fabaceae, clover (Trifolium sp.) was proportionally more utilized by individuals assigned to all groups except the LP-uninfected group, while beggar’s lice (Desmodium sp.) was proportionally more utilized by infected individuals (Figure 7C). Legumes tended to have higher RRA in the diets of individuals with high leptin levels, and there was a trend of decreasing Oxalidaceae and Violaceae RRA with leptin (Figure 6B). Lab mice from the LP-uninfected treatment utilized proportionally more Oxalidaceae (mean RRA ~0.3 vs. <0.1) and Violaceae (mean RRA ~0.2 vs. <0.1; Figure 7A). These trends did not, however, reflect a significant relationship (p > 0.05).

Perhaps due to the greater chow consumption of mice on the LP diet and increased large intestine size of infected mice, the net effects of diet and infection reveal the dynamic complexity of scaling-up this host–parasite interaction to the individual level. For example, worm counts did not differ by diet treatment (Kruskal–Wallis χ² = 0.27, df = 1, p = 0.60; Figure 8A). Worm length also did not vary with dietary quality (t = 0.57, p = 0.57).

Costs of infection were nonetheless detectable if individual-level feeding behavior was accounted for. Although overall weight change did not vary among treatments (Figure 4A), mice on the LP diet spent more time feeding (p = 0.0004; Figure 6C). Examining weight gain per time feeding revealed that infected mice gained weight at a slower rate than the uninfected mice (p = 0.042; Figure 8B; Table 2), suggesting that they had to invest that food energy into something other than growth (e.g., the immune response) or that parasites usurped it.

The 6% protein (low protein) chow reduced investment in both IgG1 and IL13. The direction of the effect of dietary protein on IgG1 responses to T. muris E/S antigen, the predominant antibody response to primary infection (61), was difficult to predict since both higher and lower responses are reasonable given previous studies. In the lab, higher IgG1 antibody concentrations during protein restriction were documented in mice infected with T. muris (14). However, mice infected with the nematode Heligmosomoides polygyrus experience lower levels of total IgG1 (15) when fed a 3% protein diet. Interestingly, a 7% protein diet permitted total IgG1 levels indistinguishable from those in mice fed a 24% protein diet (15). Similarly, the reduction in IL13 on the LP diet was not a certainty; in a previous laboratory experiment, IL13 concentrations did not vary with dietary protein level in mice infected with H. polygyrus (27). Spleen size was not sensitive to dietary protein, a somewhat unexpected result given that low-protein diets reduce spleen size during H. polygyrus infection (15, 34). However, differences between these nematodes in their infection sites (small intestine vs. cecum) and the immune responses they typically induce in C57BL/6 mice (Treg vs. Th2) (62), could account for their different effects in protein-limited hosts.

The reductions in IgG1 and IL13 could be due to direct effects of protein limitation on T- and B-cell function, which are known to be regulated by host metabolic activity (63, 64). Indeed, trade-offs between markers of protein nutrition and worm-specific antibodies to the nematode Teladorsagia circumcinta were visible in a wild Soay sheep population. Moreover, these trade-offs predicted overwinter survival with nutrition increasing survival for older individuals while investment in immunity led to greater survival odds for young sheep (37). Alternatively, the elevated carcass weights and leptin concentrations of the LP mice, due to increased chow consumption, suggest they had a higher body fat content than HP mice. Food intake in uninfected mice may be regulated by dietary protein content (65). Increased chow consumption, weight, and leptin levels have also been detected in mice fed a low-protein diet in previous nematode infection-diet manipulation experiments (27, 66). Carcass weight has been posited as an indicator of tolerance during helminth infection (67), but our data suggest that compensatory feeding may disrupt that association. Obesity is associated with pro-inflammatory responses (68), which may be mediated in part by leptin (6, 69, 70). Thus, the reduced IgG1 and IL13 levels could be a consequence of the fatter LP mice shifting from Th2 responses to pro-inflammatory Th1 responses. Although it is a common finding across multiple nematodes (27, 66), it remains unknown whether protein limitation itself or increased body fat due to limitation-induced overeating drive the pro-inflammatory shift during protein limitation.

Given how integral IgG1 and IL13 are to the development of an effective Th2 response to T. muris infection (61, 71), it is surprising—and contrary to our initial hypotheses regarding potential relationships between diet and immunoparasitological outcome—that parasite load did not differ between the dietary treatments. IgG1 levels in uninfected HP mice were much higher than those in uninfected LP mice, which contributed to the overall effect of diet on IgG1 levels. Among infected mice, IgG1 levels were more similar across diets, offering a potential explanation for their indistinguishable worm counts. Yet, among infected mice, IL13 concentrations were over 70 times higher in mice eating the HP diet, so it is unclear why that did not translate to differences in parasite load. Across both diets, infected mice had cleared most of their parasites by the time they were sampled at the end of the study. Due to this clearance rate and the high numbers of individuals with below-detection immune responses, there was insufficient statistical power to examine individual-level variation in immunity and its relation to parasite load. The treatment-level patterns suggest that perhaps the lower concentrations of IL13 in the mice given the LP diet were also sufficient to reduce T. muris survival by that time point, while those on the HP protein treatment had excess expression. Alternatively, the higher IL13 concentration (generated by stimulating MLN cells with T. muris antigen), might indicate the strength of the memory response. Thus, mice on the HP diet might be better protected during reinfection, a pattern also seen during protein limitation and infection with H. polygyrus (27, 34).

The low worm counts are surprising, given the 5.5-fold higher loads observed at a similar time point in a prior experiment (mean ± SE; Leung et al.: 34.2 ± 7.7 worms, this experiment: 6.1 ± 1.9 worms) in these same enclosures with similar mouse age and weight at infection, enclosure acclimation time, experiment duration, and parasite dosing performed by the same technician (See text footnote 1). The number of dpi was a strong predictor of worm counts, which declined quickly over time. If sampled sooner, differences in burden between diet treatments may have been more detectable. However, this rate of decline was indistinguishable from the prior experiment (See text footnote 1), and thus cannot explain the overall lower worm counts.

Instead, differences in hatching ability between batches of T. muris eggs, differential availability or composition of supplementary forage, slightly different sampling time points and/or effects of the different brand and composition of chow could contribute to the disparity in worm counts between experiments. For example, although the chow used by Leung et al. (See text footnote 1) had a similar macronutrient composition to the HP diet in this study, the highly refined ingredients in the specialty diet might have altered the gut microbial flora, reduced T. muris hatching, made the intestines a less hospitable environment for T. muris, and/or increased the efficacy of mouse immune defenses. This is an important area of future inquiry. In any case, IgG1 and IL13 data confirm that the mice in the current study were infected for long enough to stimulate an immune defense to T. muris.

In the uninfected mice, the dietary quality significantly altered the nutrient environment within the mouse; in infected hosts, T. muris would likely experience similar nutrient alterations. Estimating cecum nutrient content from those in feces using NMR revealed that sugar metabolites varied greatly between diets, and amino acids like phenylalanine and alanine were higher in the HP group. This is not surprising given that the difference in protein between the treatments was compensated with a higher percentage of carbohydrates to achieve equal calorie density. Relative abundances of tyrosine, which is involved in the regulation of immune signaling pathways (72), and the aliphatic amino acids tended to be higher among mice on the LP diet. This preliminary exploration demonstrates that NMR metabolite analysis is a powerful, non-species-specific tool for examining host nutrition. Further analysis can include 2D NMR and spiking to identify additional metabolites, quantification of absolute concentrations using reference samples [PULCON method (73)], and metabolic pathway analysis [STOCSY (74)]. Despite differences in the nutrient environment revealed by NMR, T. muris counts and length did not differ between the dietary treatments. T. muris may feed on intestinal tissues and secretions (75, 76), rather than host blood or ingesta. T. muris also strongly affects, and is affected by, the host intestinal microbiome (76–78), so any changes in metabolic profile that alter microbiome composition could potentially have stronger effects on T. muris hatching, development, and survival. Future work could explore how host gut microbiome and metabolites effect T. muris infection success, and in turn, how they are affected by helminth infection.

Our data support the hypothesis that mice attempted to compensate for differences in the protein composition of the chow by altering their physiology and behavior. Mice spent 30% more time eating the LP diet per day, a significant time investment that could also come with increased predation risk in fully natural settings (79). While this investment could partially close the gap in protein acquisition between treatments, consumption would need to be 400% higher in the LP treatment to achieve similar protein levels to individuals on the HP diet. However, dietary protein does not affect host metabolic rate (16), so maintenance costs (Figure 1) are likely similar between treatments. Insects were also present in the enclosures and diet metabarcoding tools could be used in future studies to examine if, and to what degree, lab mice are able to utilize them as a food source. Albumin was reduced on the LP diet, revealing protein limitation within those hosts. Albumin’s primary role is maintaining osmotic pressure in the blood (80), and, in wildlife, lower levels can reflect costs of reproduction (81) and indicate reduced survival probability in wildlife (37).

Interestingly, the choice of supplemental wild forage was marginally associated with differences in leptin concentrations. Leptin concentrations tended to be higher in animals that ate proportionally more plants in the legume family, Fabaceae, which includes clover (Trifolium sp.) and beggar’s lice (Desmodium sp.). These plants have a higher protein content and are widely known to be good forage for livestock and wildlife (82), even increasing sheep weight gain by an average of 40% compared with feeding on ryegrass alone (83). Thus, behavioral compensation for immunological or infection costs may explain the trend toward higher consumption of these plants in infected mice. Conversely, lab mice that ate proportionally more plants in the families Oxalidaceae (Oxalis sp.; wood sorrel) and Violaceae (Viola sp.) tended to have low leptin levels. Mice consumed similar amounts of chow in the enclosures (g chow/g mouse) as they did in the laboratory setting, so wild plants probably do not represent a replacement food source for most individuals. We cannot quantify the amount of wild plant matter eaten by mice in the different treatments using DNA metabarcoding, but these emerging trends are suggestive of compensatory foraging behaviors worth further investigation.

Infected mice found physiological ways to compensate for the costs of T. muris infection, rather than following our alternate hypotheses of increased foraging or infection-induced inappetence. The increased cecum size of infected individuals could be a consequence of parasite manipulation to create more habitat space, but the increase in large intestine size with infection is more difficult to explain. Large intestines, emptied of contents and relative to body size, were over 10% heavier in infected mice. This additional weight was not due to the worms themselves, which were located in the cecum. In the average size mouse, this difference translates to a 15-mg increase in colon weight. Hosts could increase colonic tissues to enhance water balance, nutrient reabsorption, or house more commensal microbes to aid in digestion. Similarly, H. polygyrus, which resides in the small intestine, is associated with increased investment in small intestine tissues (67). An influx of immune cells or an increase in gut microbe communities could also contribute to these differences; this hypothesis could be examined histologically in future studies. However, given estimates of approximately 1 pg per E. coli cell (84) and 2.2 pg per lymphocyte (85), they cannot fully account for the increase in colon size in T. muris-infected mice.

To some degree, infected mice may also supplement their food intake with clovers. This trend was driven by a lack of beggar’s lice in feces of uninfected mice and an absence of clover in LP-uninfected mice. Plants in the Fabaceae family, including clovers and beggar’s lice, tend to be high in protein (82), and that could help hosts either resist infection by providing more resources for immune defense or tolerate infection by compensating for costs of infection and immune defense. A larger sample size, particularly of LP-uninfected mice, would help elucidate the efficacy and generality of this potential behavioral compensation mechanism. At this stage, however, we can conclude that infected mice gained less weight per minute of chow feeding than uninfected mice, despite increased large intestine size and use of potentially high-protein wild forage. This in turn suggests that their compensation for the costs of infection (e.g., parasite resource theft, elevation of immune defenses, tissue repair, etc.) was incomplete.