A novel mouse model of Campylobacter jejuni enteropathy and diarrhea

Campylobacter infections are among the leading bacterial causes of diarrhea and of ‘environmental enteropathy’ (EE) and growth failure worldwide. However, the lack of an inexpensive small animal model of enteric disease with Campylobacter has been a major limitation for understanding its pathogenesis, interventions or vaccine development. We describe a robust standard mouse model that can exhibit reproducible bloody diarrhea or growth failure, depending on the zinc or protein deficient diet and on antibiotic alteration of normal microbiota prior to infection. Zinc deficiency and the use of antibiotics create a niche for Campylobacter infection to establish by narrowing the metabolic flexibility of these mice for pathogen clearance and by promoting intestinal and systemic inflammation. Several biomarkers and intestinal pathology in this model also mimic those seen in human disease. This model provides a novel tool to test specific hypotheses regarding disease pathogenesis as well as vaccine development that is currently in progress.


Antibiotics are required for Campylobacter jejuni colonization
Non-antibiotic treated C57BL/6 mice fed a standard house chow (HC) diet grew slightly better than antibiotic (gentamicin (35mg/L), vancomycin (45mg/L), metronidazole (215mg/L), and colistin (850U/ml) in drinking water days -4 through -1 prior to infection as described in Methods) treated mice and showed similar weight loss following a single dose of Campylobacter (10 6 via gavage) infection ( Fig 1A). Both infected groups recovered weight quickly, 3 days post infection. However, there was a striking increase in Campylobacter shed in stool in antibiotic pretreated animals ( Fig 1B). We were able to re-isolate and confirm the viable C. jejuni strain from feces of infected animals to confirm qPCR results (S1 Fig). Diarrhea was defined as unformed stool upon daily examination and sampling. As shown in the photographs in Fig 2, this was often watery and even bloody as noted in the legend. Diarrhea was observed only in antibiotic pretreated infected animals. Additionally, the fecal biomarker myeloperoxidase (MPO) was significantly elevated only in the antibiotic-pretreated Campylobacter infected group ( Fig 1C).

Feeding diets deficient in protein or zinc alter Campylobacter outcomes
Having established the benefit of antibiotic pretreatment in this model, we then probed the effect of zinc or protein deficiency in Campylobacter infection in antibiotic-pretreated mice. We have recently found that protein or zinc deficiencies can substantially increase susceptibility to Cryptosporidium and EAEC (enteroaggregative E. coli), respectively [33][34][35], and because these deficiencies are common in children living in impoverished areas where these infections are common, we examined the effects of these diets on Campylobacter infections in our murine models. Mice (n = 8 / group) were infected with 10 7 C. jejuni on day 14 following feeding of specific diet and antibiotic disruption of resident microbiota as described in methods. As shown in Fig 2A, infected mice fed an isocaloric protein source diet without zinc (dZD) diet had significantly greater weight loss than infected mice receiving the HC diet or a diet deficient in protein (dPD). While all HC or dZD fed mice developed diarrhea by day 3 post infection, all dZD fed mice had persistent bloody diarrhea for greater than 7 days post infection ( Fig 2C). No diarrhea was seen with mice fed the dPD diet. Only mice that were fed dZD diet were colonized with C. jejuni for the duration of the experiment as measured by shedding of organism in stool ( Fig 2B). As with HC-fed mice, non-antibiotic treated dPD or dZD fed mice infected with C. jejuni did not develop diarrhea and cleared the organism by day 3 post infection.

Biomarkers of inflammation are increased following Campylobacter infection
MPO has been shown to increase with Campylobacter infections in humans across MAL-ED study sites [5]. In the MAL-ED studies, we demonstrated the correlation of MPO, lipocalin-2 (LCN-2), and calprotectin [36]. The fecal biomarkers of neutrophilic inflammation (MPO) and epithelial damage (LCN-2) were measured on day 2 and day 14 post C. jejuni infection. Infected mice receiving the dZD diet had significantly greater LCN-2 and MPO at both time points compared to infected control or protein deficient mice (Fig 3).

Intestinal morphology is altered by Campylobacter in zinc deficient mice
Ileal histology of dZD-fed mice infected with C. jejuni showed increased mucus secretion and flattened villi ( Fig 4A). As has been suggested in the cecum by Stahl et al [29], we noted colonic submucosal edema and mucosal thinning, cell infiltration, crypt hyperplasia and striking goblet cell discharge (Fig 4B). These observations were confirmed by careful histopathologic assessment with a pathologist (KT). The predominant changes were seen in the colon, with increased lamina propria cellular infiltrate and striking luminal mucus discharge (S2A Fig), with only moderate mucus discharge seen in the ileum (S2B Fig). No notable changes in histology were observed in HC or dPD-fed mice.

Campylobacter capsule gene kpsM is required for colonization
Having seen robust shedding, weight loss and overt diarrhea with a zinc deficient model of Campylobacter infection, we then tested the mutant of the same strain lacking kps, a postulated key virulence trait being proposed for potential vaccine development. [37]. All mice were fed the dZD diet for 14 days and pretreated with antibiotics to disrupt resident microbiota as above. Mice infected with the kpsM C. jejuni gained weight significantly better than mice infected with the wildtype strain ( Fig 5A). Additionally, shedding of C. jejuni kpsM was undetectable following day 1 suggesting an inability to colonize as the wildtype strain ( Fig 5B). Diarrhea was observed only in wildtype, infected mice.

Metabolic perturbations induced by C. jejuni infection
Orthogonal projection to latent structures-discriminant analysis (OPLS-DA) models were constructed to assess the urinary metabolic perturbations induced by the Campylobacter jejuni challenge on days 5, 9, 11 and 14 post-infection. No significant differences in the urinary metabolic profiles of infected and uninfected mice fed the protein deficient diet were observed, consistent with the rapid clearance of the pathogen and absence of diarrhea. Infection resulted in biochemical alterations in the nourished mice 2 days post-infection causing increases in the excretion of taurine (Q 2 Y = 0.42, p = 0.04).
In agreement with results obtained from the growth, inflammatory and shedding measures, the most notable metabolic changes in response to infection were seen in dZD-fed mice. The metabolic changes induced by infection in the zinc deficient uninfected and infected mice are summarized in   (4-PY)) were excreted in greater amounts at all sampling points. This is consistent with a sustained attempt to dampen the inflammatory response to infection through the induction of indoleamine 2,3-dioxygenase.
Additionally, shifts in the excretion of gut microbial-host co-metabolites were observed on days 5, 11 (Q 2 Y = 0.82, p = 0.023) and (Q 2 Y = 0.56, p = 0.028) 14 post-infection. Phenylacetylglutamine (PAG), a co-metabolite derived from phenylalanine, and 4-hydroxyphenylacetate (4-HPA) and 4-HPA sulfate, derivatives of tyrosine, were excreted in lower amounts in the C. jejuni infected mice. 4-HPA can be further processed by intestinal bacteria to 4-cresol, which once absorbed is either glucuronidated or sulfated in the liver to 4-cresyl glucuronide (4CG) and 4-cresyl sulfate (4CS), respectively prior to excretion. Both 4-CG and 4-CS were excreted in greater amounts in the infected dZD mice compared to their uninfected equivalents suggesting greater metabolism of 4-HPA to 4-cresol in the dZD infected mice.
Consistent with histopathological findings, an increase in the excretion of diet-derived sucrose and tartrate was observed at days 5 and 14 post-infection indicative of impaired barrier function. Changes in the excretion of metabolites associated with oxidative stress and intestinal damage such as allantoin, pseudouridine and fucose were also noted in the infected animals. These findings are in agreement with the elevated biomarkers of intestinal inflammation and disturbed gut morphology in infected mice fed the dZD diet. Correlation analysis between biomarkers of inflammation and urinary metabolites identified significant positive correlations between LCN-2 and fucose and between MPO and hexanoylglycine and the two gut microbial derived metabolites 4-CS and hippurate in zinc deficient infected mice (Fig 7).

Perturbations to the microbiota induced by C. jejuni infection
16S rRNA gene sequencing was performed on fecal samples collected from infected and uninfected mice on all diets at 0, 1, 5, and 9 days post-infection/ antibiotic exposure. Richness (defined here as the number of unique amplicon strain variants (ASVs) detected in each sample) and evenness (the inverse of the Simpson index) of the microbiota was calculated for all samples (Fig 8A and 8B). Infected dPD-fed mice had higher richness at day 5 post-infection than at all other time points for both infected and uninfected dPD-fed mice. Similarly, infected dPD-fed mice at day 5 post-infection had higher evenness than any other time point for both infected and uninfected dPD-fed mice. Uninfected dPD mice had higher evenness at day 9 post-infection than all dPD samples other than those from infected mice at days 5 and 9 post-infection. dZD-fed mice had no significant differences in richness or evenness over time within infection status or across infection status. In infected HC-fed mice, richness was lower at 0 days post infection than at all other time points for infected and uninfected HC-fed mice. The same was true for uninfected HC-fed mice, except that no significant difference was detected in the comparison with uninfected HC-fed mice at 1 day post-infection. No significant difference was detected in richness between uninfected and infected HC-fed mice at 0 days post-infection. In infected HC-fed mice, richness increased with each successive time-point. The same was true for uninfected HC-fed mice, except that richness plateaued at 5 days post-infection. Evenness was higher in infected HC-fed mice at 9 days post-infection than at all other time points for both infected and uninfected HC-fed mice. Evenness of infected HC-fed mice at day 0 was lower than at all other time points.
We calculated beta diversity to describe the global dissimilarity of all samples in the experiment ( Fig 8C). Within each diet, uninfected and infected samples from the same time point tended to cluster together, except for dPD-and dZD-fed mice at 5 days post-infection and HC-fed mice at 9 days post-infection. For dPD-fed mice, this time point is also when the Given that mice were on antibiotics prior to infection with C. jejuni, the increases in richness and evenness over time in HC-fed mice are expected. Surprisingly, infected HC-fed mice had higher evenness and richness than uninfected mice at 9 days post infection. Antibiotic cessation may not increase evenness and richness in dPD-and dZD-fed mice given the incomplete diets they are receiving, from which dietary fiber is also absent. Substantial increases in either richness or evenness over time were not observed for either defined diet with or without infection. However, infected dPD-fed mice had a transient increase in evenness and richness at 5 days post infection, which is also the last time point for which shedding is detected from dPD-fed mice. Infected dPD-fed mice at this time point had higher abundance of amplicon sequence variants (ASVs) assigned to Bacteroidales S24-7 group, several from the order Clostridiales, multiple Enterobacteriaceae, as well as several other ASVs (Fig 9). Many of these increases persisted to day 9 post-infection. Overall, infection led to a greater number of differentially abundant ASVs in dPD-fed mice at days 5 and 9 post-infection than any other diet or time point.
Infection had varied effects on the microbiota of dZD-fed mice over time. At day one postinfection, there was an increase of ASVs assigned to Enterobacteriaceae, Defluviitaleaceae UCG-011, and decreases in ASVs assigned to Ruminococcaceae UCG-014 and Streptococcus (Fig 9). This change is consistent with blooms of Enterobacteriaceae observed immediately after infection with other pathogens that may be induced by the host inflammatory response [38]. However, none of these changes persisted through day 5 post-infection, where the same Enterobacteriaceae ASV was substantially less abundant in the infected group rather than more abundant as at day 1 post-infection. Two ASVs assigned to the S24-7 group of the Bacteroidales order and an ASV assigned to the genus Erysipelatoclostridium were increased in infected mice, while Lachnospiraceae NK4A135 decreased. By day 9 post-infection, no differences were observed between infected and uninfected dZD-fed mice other than in increase in Enterobacteriaceae in infected mice. Given that dZD-fed mice had persistent bloody diarrhea, the relatively low number of differentially-abundant ASVs may be due to their extremely limited diversity, with fewer than 25 unique ASVs in all but two samples at all time points.
HC-fed mice also had few ASVs that were consistently differentially-abundant over time. Two ASVs assigned to the S24-7 group of the Bacteroidales order were decreased in abundance at day 5 post-infection, one of which was also decreased at day 1 post-infection. At day 1 postinfection, one Ruminococcaceae UCG-014 ASV was also decreased in abundance, while Lachnospiraceae NK4A135 increased. At day 5 post-infection, the change in Lachnospiraceae NK4A135 persisted and two additional Lachnospiraceae ASVs had increased abundance. At day 9 post-infection, Lachnospiraceae NK4A135 and one additional Lachnospiraceae ASV were less abundant in infected mice than uninfected mice, and 2 Ruminococcaceae ASVs and 2 Erysipelotrichaceae ASVs were more abundant in infected mice.

Discussion
Given the increasingly recognized importance of Campylobacter infections (as causes of severe bloody diarrhea or growth failure in children living in developing areas) and of food and milkborne outbreaks in developed areas, not to mention the troubling sequelae like Guillain-Barré Syndrome (GBS), the need for improved understanding of its pathogenesis and effective interventions or vaccines is paramount. A major limitation of work in this area is the lack of an inexpensive animal model that mimics human disease. Here, we demonstrate that nutritional and antimicrobial manipulations can enable us to model growth failure or overt bloody diarrhea. With manipulation of the antibiotics or diet, we may be able to model weight loss without overt diarrhea (as suggested by the transient effects of Campylobacter without antibiotic pretreatment shown in Fig 1A). This model opens tremendous opportunities to help enable vaccine development or other potential interventions.
The biomarkers that are common to both humans and our murine models also provide relevant markers to assess these interventions and mechanisms as well as the directly relevant clinical disease outcomes [5,36]. Interestingly, a non-encapsulated mutant of 81-176 was severely attenuated in colonization and disease in animals fed the zinc deficient diet. This same mutant was attenuated in a ferret model of diarrheal disease and was slightly reduced in its ability to colonize BALB/c mice compared to the 81-176 parent [37,39]. Although not shown in the current study, older mice (>10 weeks of age) were susceptible to C. jejuni infection; hence this model should enable vaccine studies. Whether we can model post-infectious sequelae such as GBS [40] or the roles of the microbiome in susceptibility to disease or sequelae remains to be seen.
There are several limitations on any non-human model of infectious diseases that must always be kept in mind. Coprophagic rodents clearly have quite different microbiota, diets and host determinants of disease. Nevertheless, recognizing these limitations, the extent to which we can model key major disease outcomes, including shared biomarker and even metabolomic pathways [35,36,41,42] in an inexpensive, relative to existing models, defined murine model (without major genetic alterations or artificial pharmacologic immune manipulation) at least enables hypothesis generation and testing, as well as assessment of potential new vaccines that are currently in development [20].
C.jejuni infection in the zinc deficient mice elicited a robust inflammatory response evident by bloody diarrhea, intestinal and systemic inflammation parallel to growth faltering, highlighting the direct relevance of this model in clinical Campylobacteriosis research. This was accompanied by an activation of the kynurenine pathway as a result of systemic inflammation in response to infection. Indoleamine 2,3-dioxygenase (IDO) metabolizes tryptophan to kynurenine and it is increased in response to infection and inflammation resulting in a greater excretion of NMND and its downstream metabolites 2-PY and 4-PY.
NMND is generated from nicotinamide adenine dinucleotide (NAD) by the action of the enzyme N-methyl transferase (NMNT), which is considered a master regulator of energy expenditure [43]. Increased excretion of NMND by zinc deficient infected mice reflects lower energy expenditure. The energy acquisition and utilization by the host is compromised by zinc deficiency [44] (S3 Fig) and further exacerbated following C.jejuni infection. Metabolic exhaustion suggests reductions in the levels of available NAD and here is reflected in the compromised activity of the energy generating TCA cycle (decreased citrate and succinate excretion) that happens concurrently to the loss of β-oxidation activity (hexanoylglycine and butyrylglycine), branchedchain amino acid catabolism (2-MOV, 2-OIC, 2-OIV) and muscle catabolism (increased creatine excretion on day 5 post infection). Zinc is a necessary cofactor for numerous enzymes involved in energy regulation [45], zinc deficiency therefore narrows the metabolic flexibility of the infected mice hindering pathogen clearance. We invite parallels to be drawn between the metabolic phenotype of the C.jejuni infected mice described here and that of our murine model of Shigella flexneri (manuscript currently under peer-review) also achieved with zinc deprivation and the use of antimicrobials. Zinc deficient S.flexneri infected mice displayed a similar metabolic profile characterized by energy exhaustion and catabolic stress combined with delayed pathogen clearance in these mice, compared to the HC-fed mice.
The presence of C. jejuni in the gut of zinc deficient mice perturbed metabolites derived from the gut microbiota. While PAG, 4-HPPA, 4-HPA and 4-HPA sulfate excretion was decreased with infection; the glucuronidated and sulfated forms of 4-cresol (microbial metabolites of 4-HPA) were increased. 4-cresol is a known bacteriostatic, negatively influencing the growth of gut bacteria [46] contributing to the lack of microbial diversity in the zinc deficient infected mice. Interestingly, increased urinary excretion of 4-CS was positively correlated with the fecal inflammatory biomarker, MPO. Increased excretion of 4-CS implies greater production and exposure of 4-cresol in the gut. This suggests that the release of 4-cresol in the intestine may contribute to the acute inflammatory response and subsequent damage to the intestinal epithelium. However, further work is necessary to establish causality. The lack of persistent changes in microbiota composition in dZD-fed mice make any signature difficult to resolve. Metagenomic studies are needed to identify the key functions disrupted by infection, as well as biomarkers of infection. The dynamics of differentially-abundant ASVs over the course of infection suggest that microbiota composition changes more rapidly as a result of infection than the host metabolome. In the other diets (dPD and HC), the ASVs that become differentially-abundant upon infection have little overlap with those from dZD-fed mice. This implies that the effect of infection on the microbiota is diet-dependent, which could be due to the difference in microbiota background induced by the diet, or that the intestinal environment changes in unique ways with each phenotype resulting from infection while being fed each diet. It seems likely that a combination of these factors (e.g. pathogen-microbiome, diet-pathogen, host-diet, host-microbiome, and microbiome-diet interactions) influence the virulence of the pathogen. It will be interesting to compare metabolic responses with C. jejuni infection in mammals with those in the avian reservoir(s), in which intriguing effects of microbial metabolites associated with reduced colonization after potential vaccine and probiotics has recently been reported [47].
Intriguingly, the glycoprotein marker of epithelial damage LCN-2 positively correlated with the excretion of fucose. Fucose is a sugar, found in high quantities in the mammalian gut either as part of glycan structures or as a free monosaccharide and current research suggests it can have pleiotropic roles in protection against infections and inflammation, predominantly as a nutrient substrate for commensal microbial communities supporting their activity and improving the health of the host. It can however, provide a competitive advantage for certain pathogens. Fucose utilization appears to be important for the survival of C.jejuni NCTC11168 in the gut [14] and the pathogen exhibits chemotaxis and increased biofilm formation in the presence of fucose [16]. It is unclear if the correlation between LCN-2 and fucose is a protective host immune response or against C.jejuni 81-176 used in this study, since this strain lacks the fuc locus that regulates chemotaxis towards fucose and therefore the pathogen cannot use fucose to its benefit. Elevated urinary fucose excretion accompanied by increased LCN-2 levels were also observed in our murine model of Cryptosporidium infection [34].
We did not identify any ASVs belonging to the Campylobacter genus within our study. Interestingly, a previous study on a murine model of C. jejuni infection also did not report the presence of sequences assigned to Campylobacter [48]. In that study, the authors amplified the V35 region of the 16s rRNA gene, while the V4 region was chosen for our study. The lack of detection of Campylobacter even when qPCR indicates abundant shedding suggests that existing universal primers for amplification of 16S rRNA genes may have bias that prevents amplification of 16S rRNA genes from Campylobacter species.
The specific host nutrient and microbiome alterations that alter relevant disease outcomes offer another potentially fruitful area to better dissect and understand the mechanisms (and thus potentially effective approaches to intervene) for these clinical effects of Campylobacter infections. Thus we can now explore whether zinc or key repair nutrients might ameliorate these disease outcomes. For example, zinc can suppress expression of not only such host genes as the CFTR chloride channel that is responsible for secretory diarrhea with cholera and the muc-2 gene, but also microbial genes such as the aggR-regulated virulence gene expression in enteroaggregative E. coli (EAEC) in vitro or in vivo [33], and homologous genes have been reported in several other enteric bacterial pathogens like Vibrio cholerae, Shigella and Citrobacter, and possibly Campylobacter [49,50]. Effects of zinc on enteric infections are complex. For example, in addition to our findings herein regarding C. jejuni infection, we have also recently reported enhanced enteroaggregative E. coli infection in zinc deficient mice [33,51]. However, the opposite is seen with C. difficile infection which is worse with added zinc, potentially via altering gut microbiota, toxin production and even the zinc binding host antimicrobial protein, calprotectin [52].
We conclude that zinc deficiency and antimicrobial treatment enable a robust weaned murine model of symptomatic enteric Campylobacter infection that exhibits its major clinical manifestation in humans of bloody diarrhea and growth failure, inflammatory histopathology and biomarker expression and fecal shedding that is sustained over at least 2 weeks. Others have explored the role of zinc and innate immune response in the pathogenesis of Campylobacter infections. Interestingly, Davis et al. describes the requirement of a zinc transporter (znuA) for growth of Campylobacter in zinc-limiting media and in chickens. The same wildtype strain was used in this study (81-176) and was able to colonize in zinc deficient mice [53,54]. Whether or not a znuA mutant could colonize in this mouse model is unknown. Whether zinc deficiency benefits Campylobacter (similar to our findings with Enteroaggregative Escherichia coli [33]) or alters normal microbiota in combination with antibiotics requires further investigation. Studies are currently underway to address these possible effects of zinc deficiency for both host and pathogen mechanisms. This may also help explain the requirement for antibiotic disruption of native microbiota prior to Campylobacter challenge in our model. In contrast, with either standard or protein deficient diets (instead of zinc deficiency), we see modest enteropathy (as demonstrated by shedding and biomarkers), without overt diarrhea, thus potentially modeling 'asymptomatic' Campylobacter infections in children in developing areas worldwide.
In conclusion, these models now enable more rapid and clinically relevant vaccine testing and dissection of mechanisms that may help in development of novel approaches to effective interventions for this common global threat.

Ethics statement
This study included the use of mice. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of Virginia (Protocol Number: 3315). All efforts were made to minimize suffering. This protocol was approved and is in accordance with the Institutional Animal Care and Use Committee policies of the University of Virginia. The University of Virginia is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International (AAALAC).

Animal husbandry
Mice used in this study were male, 22 days old, C57BL/6 strain, and ordered from Jackson Laboratories (Bar Harbor, ME). Mice weighed approximately 11 grams on arrival and were cohoused in groups up to five animals per cage. The vivarium was kept at a temperature of between 68-74˚F (20-23˚C) with a 14 hour light and 10 hour dark cycle.

Rodent diet
Weaned mice (22 days old) were acclimated, fed a regular diet for 2-5 days, and then were fed either standard rodent 'House Chow' (HC), a protein source diet without zinc (dZD), or protein (2%) deficient (dPD) diet (Research Diets, Inc.). All diets were isocaloric and calories from fat, protein, and carbohydrates are as previously reported [35].

C. jejuni infection
Weaned, 3 week old male Jackson mice were acclimated for 3 days on standard rodent house chow (HC). Mice were then given HC, dZD or dPD described above for 14 days before challenge. Prior to infection, mice were given gentamicin (35mg/L), vancomycin (45mg/L), metronidazole (215mg/L), and colistin (850U/ml) in drinking water to disrupt resident microbiota as previously published [33,55]. After three days on antibiotic water, mice were given untreated water without antibiotics for 1 day prior to challenge. C. jejuni 81-176 and an isogenic kpsM mutant were used [37]. Cultures were grown from glycerol stocks in Mueller Hinton broth in airtight canisters with 'campy packs' (Fisher Scientific). Each infected mouse received an inoculum~1x10 7 C. jejuni in 100 µL of freshly prepared broth via oral gavage; controls received 100 µL of broth alone. Mice were maintained on specific diet and untreated filtered drinking water following infection.
Each experiment used N = 8 mice per experimental group and mice were euthanized at either day 3 or day 14 post infection. We have previously published that we obtain both serum and tissue zinc deficiency following 14 days on dZD diet [33].

Protein extraction from stool and tissue
After rapid dissection of the mouse intestines, cecal contents and stool were flash frozen in liquid nitrogen and stored in a -80C freezer. At time of assay, samples were lysed in RIPA buffer (20 mM Tris, pH 7.5, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 1 mM EDTA, 0.1% SDS) containing protease inhibitors cocktail (Roche) and phosphatase inhibitors (1 mM sodium orthovanadate, 5 mM sodium fluoride, 1 mM microcystin LR, and 5 mM beta-glycerophosphate). Tissue lysates were cleared by centrifugation, and the supernatant was used for total protein measurement (Pierce BCA), cytokine measurement by Luminex assay (BioRad), and specific ELISAs for lipocalin-2 (LCN2) and myeloperoxidase (MPO) as previously described [35].

DNA extraction
DNA was isolated from cecal contents and fecal pellets using the QIAamp DNA stool mini kit (Qiagen) as previously described [56]. For enhancing the pathogen's DNA extraction we made an improvement in the original protocol: a vigorous homogenization of the samples with 300 mg of 1.0 mm zirconia beads (BioSpec, Bartlesville, OK, USA) using a MiniBeadBeater (BioSpec, Bartlesville, OK, USA). After extraction, DNA was eluted in 200μl Elution Buffer and stored at −20˚C.

Real-time PCR for C. jejuni quantification
Stool and tissue DNA were analyzed for the C. jejuni/coli cadF gene to determine shedding of organism in stool or burden in the tissue. Primers for cadF were F, 'CTGCTAAACCATAGA AATAAAATTTCTCAC'; R, 'CTTTGAAGGTAATTTAGATATGGATAATCG'; Probe, 'FAM-CATTTTGACGATTTTTGGCTTGA-MGB' as published by Liu et al. [57].
Quantification of the infection was performed in a Bio-Rad CFX PCR Detection System by interpolating Ct values of each run with a standard curve of known amounts of C. jejuni DNA and transformed into number of organisms per milligram of sample. The master mix solution and primers were used as described elsewhere (3). Amplification consisted of 5 minutes at 95˚C, followed by 40 cycles of 15 seconds at 95˚C, 60 seconds at 58˚C. PCR efficiency was >90% with reliable detection from 10 2 −10 10 cfu as determined with a standard curve verified by plate counts.

Intestinal morphometry and mucus staining
Ileal segments were fixed in 4% paraformaldehyde, embedded in paraffin, and stained with hematoxylin-eosin or PAS at the University of Virginia Histology Core.

Biomarkers
Myeloperoxidase (MPO) and lipocalin-2 (LCN-2) were measured in stool and cecal contents according to manufacturer's instructions (R&D Systems). Values were normalized to total protein as measured by BCA kit (Invitrogen) and presented as pg/mg protein.

Statistical analysis
Data analyses were performed with GraphPad Prism 6 software (GraphPad Software). All statistical analyses were done from raw data with the use of analysis of variance, Student t tests, and Bonferroni post hoc analysis where applicable. Differences were considered significant at P < 0.05. Data are represented as means ± standard errors of the mean.

H NMR spectroscopy based metabolic profiling
Urine samples were analysed by 1 H nuclear magnetic resonance (NMR) spectroscopy. Each sample was prepared by combining 30 μl of urine with 30 μl of phosphate buffer (pH 7.4, 100% D 2 O) containing 1mM of the internal standard, 3-trimethylsilyl-1-[2,2,3,3-2 H4] propionate (TSP). Samples were then vortexed and spun (10,000 g) for 10 minutes at 4˚C before transfer to 1.7mm NMR tube. Spectroscopic analysis was performed at 300K on a 600 MHz Bruker NMR spectrometer equipped with a BBI probe. Standard one-dimensional spectra of the urine samples were acquired with saturation of the water resonance, using a standard pulse sequence. For each sample, 4 dummy scans were followed by 64 scans collected in 64K time domain points and with a spectral window set to 20 ppm. A relaxation delay of 4 s, a mixing time of 10 ms, an acquisition time of 2.73 s and 0.3 Hz line broadening was used. Spectra were referenced to the TSP resonance at δ 0.0. 1 H NMR spectra (δ -0.5-10) were digitized into consecutive integrated spectral regions (~20,000) of equal width (0.00055 ppm). Spectral regions corresponding to TSP (δ -0.5-0.5), water (δ 4.5-4.8) and urea (δ 5.6-6.1), were removed. The resulting spectra were then normalized to unit area. Multivariate statistical modeling was performed using inhouse scripts, In MATLAB (R2016a). This included principal components analysis (PCA) using Pareto scaling and orthogonal projections to latent structures-discriminant analysis (OPLS-DA) using data mean centering. OPLS-DA models were built to facilitate data interpretation. 1 H NMR spectroscopic profiles were used as the descriptor matrix and class membership (e.g., uninfected zinc deficient mice and Campylobacter infected zinc deficient mice) was used as the response variable. The predictive performance (Q 2 Y) of the models was calculated with the use of a 7-fod cross-validation method, and model validity was defined upon permutation testing (1000 permutations). Significant metabolites were identified and extracted from valid pair-wise OPLS-DA models and summarized in heat maps.

16S rRNA gene sequencing and analysis
16S libraries were pooled and sequenced according to the protocol developed by Kozich et al., and the detailed version of the specific protol used is provided in the supplemental text. Briefly, the V4 region of the 16S rRNA gene was amplified and sequenced using the MiSeq Reagent Kit v2 that produces up to 25 million reads of 2 x 250 bp per run at the Genomics Core Facility at UVA. Reads were assigned to samples using Illumina BaseSpace demultiplexing. From these reads, analysis was performed using the DADA2 package (benjjneb.github.io/dada2, version 1.4.0) [58]. The first 10 bases and last 10 bases were trimmed from each 250 base forward read due to low quality scores, resulting in 230 base sequences. Similarly, the first 10 bases and the last 90 bases were trimmed from reverse reads, resulting in 150 base sequences. Reads were then filtered according to 3 criteria: presence of no ambiguous 'N' bases in the read, no base with quality scores below 2 in the read and having a maximum of 2 expected errors in the read. Filtered reads were then dereplicated with the derepFastq function, resulting in unique sequences each with a corresponding abundance and consensus quality profile. The DADA2 sample inference function was applied to dereplicated sequences with joint inference of sample composition and error-rate parameter estimation (e.g. selfConsist = TRUE). Chimeras were then removed using the removeBimeraDenovo function, then taxonomy was assigned to each sequence using the Ribosomal Database Project's naïve bayes classifier [59] with RDP training set 14.
Amplicon sequence variant (ASV) counts for each sample, with accompanying taxonomy assignments, were further analysed using the phyloseq [60] package (version 1.20) in the R programming language (version 3.4.0). Differentially abundant ASVs were determined using the DESeq2 [61] package (version 1.16.1). Prior to all analyses, sample counts were rarified to control for uneven sequencing depth across samples. For each sample, 5000 reads were randomly chosen using the rarefy_even_depth function in phyloseq. These rarefied counts were used for all downstream analyses. Alpha diversity metrics were calculated using the plot_richness function in phyloseq. A three-way ANOVA was used to compare richness over time, between diets, and between infection statuses, with multiple testing correction via Tukey's Honestly Significant Difference (HSD) method as implemented in the TukeyHSD function. Comparisons with a multiple testing corrected p value of less than 0.05 were considered significantly different. The same procedure was used to compare evenness across sample groups. Bray-Curtis dissimilarity calculations and non-metric multidimensional scaling (k = 2) were performed using the plot_ordination function in phyloseq. Differential abundance testing was performed between time-matched infected and uninfected mice within each diet. Prior to differential abundance testing, ASVs with fewer than 10 counts in any sample or fewer than 20 counts across the entire dataset were removed. Within DESeq2, Wald significance tests were used with parametric dispersion fitting. ASVs with a p value of less than 0.05 were considered significantly differentially abundant. The DESeq2 design was chosen to compare groups across all three-group categories (infection status, diet, and time), and the results were extracted specifically for comparisons between infected and uninfected groups with matched diets at the same time points. ASVs that were differentially abundant in any of these infected vs. uninfected comparisons are included in the heatmap in Fig 9. DADA2 processing scripts, processed data as a phyloseq object, and scripts to recreate all microbiome analyses can be found at https://github.com/gregmedlock/campy_murine_diets. Sequencing data are available under the National Center for Biotechnology Information BioProject PRJNA470950.