Effect of chronic and acute enterotoxigenic E. coli challenge on growth performance, intestinal inflammation, microbiome, and metabolome of weaned piglets

Post-weaning enteropathies in swine caused by pathogenic E. coli, such as post-weaning diarrhea (PWD) or edema disease (ED), remain a significant problem for the swine industry. Reduction in the use of antibiotics over concerns of antibiotic resistance and public health concerns, necessitate the evaluation of effective antibiotic alternatives to prevent significant loss of livestock and/or reductions in swine growth performance. For this purpose, an appropriate piglet model of pathogenic E. coli enteropathy is required. In this study, we attempted to induce clinical signs of post-weaning disease in a piglet model using a one-time acute or lower daily chronic dose of a pathogenic E. coli strain containing genes for both heat stable and labile toxins, as well as Shiga toxin. The induced disease state was monitored by determining fecal shedding and colonization of the challenge strain, animal growth performance, cytokine levels, fecal calprotectin, histology, fecal metabolomics, and fecal microbiome shifts. The most informative analyses were colonization and shedding of the pathogen, serum cytokines, metabolomics, and targeted metagenomics to determine dysbiosis. Histopathological changes of the gastrointestinal (GI) tract and tight junction leakage as measured by fecal calprotectin concentrations were not observed. Chronic dosing was similar to the acute regimen suggesting that a high dose of pathogen, as used in many studies, may not be necessary. The piglet disease model presented here can be used to evaluate alternative PWD treatment options.


Results and discussion
Growth performance. Pigs used as an experimental model for PWD were challenged with a spontaneous nalidixic acid-resistant mutant of Escherichia coli strain NCDC 62-57 (ATCC 23,545) referred to hereafter simply as E. coli 62-57nal in either a single acute high-titer dose (1.5 × 10 10 CFU), or in a series of daily lower-dose challenges (average daily dose 5.4 × 10 8 CFU). This strain was chosen for its ability to cause illness during preliminary trials amongst F18 + and K88 (F4) ETEC strains. All pigs were held for two days prior to the start of the trial and were asymptomatic for gastroenteritis. Additionally, pigs were not colonized by organisms capable of forming colonies on MacConkey amended with 50 µg/ml nalidixic acid (MacConkey + nal), and no endogenous phage infecting E. coli 62-57nal were identified. Thirty-six presumptive coliform colonies from pooled fecal samples plated on MacConkey agar (0 µg/ml Nal; three colonies per pen) were also tested by PCR for the presence of Shiga toxin type 1 (Stx1), Shiga toxin type 2 (Stx2), heat-stable enterotoxin I (ST1), heat-stable enterotoxin II (ST2) and heat-labile toxin (LTI). All colonies were negative for Stx1, Stx2, ST2 and LTI, but three colonies were positive for ST1. Presence of ST1 gene alone is not a strong predictor of ability to cause disease 8,10,11 and pigs were asymptotic, so all animals were retained in the study.
In general, pigs administered E. coli 62-57nal via both the acute and chronic dosing regimens presented similar clinical signs with the majority of pens developing scours by day 2 and continuing through day 6. Control pens had visibly soft feces on day 5 and 6 with a single incident of scours on day 6, however the animals in control pens remained visibly healthy throughout the trial period. Additionally, the control pen with the incidence of scours was culture negative for the inoculated E. coli 62-57nal throughout the trial, so scours may have been induced by stress or other native microbiota. There was no evidence of difference for overall average daily gain (ADG), average daily feed intake (ADFI), and gain:feed (G:F) of the different treatment groups (P > 0.184, Table 1). However, there were numerical differences between pigs fed the treatments, suggesting that the modest number of replicates and the inherently high post weaning variability in performance were responsible for the failure to detect significant differences in growth performance. This lack of evidence for significant growth differences is similar to previously reported results 12 ; however, significant changes to growth performance in response to ETEC challenge have also been reported 13 . Pigs administered the acute and chronic dose of E. coli 62-57nal had a 54.7% and 14.9% reduction in ADG compared to the control pigs, respectively ( Table 1). The control group had the lowest ADFI among the three treatments with acute and chronic dosing regimens increasing feed intake by Table 1. Effects of E. coli 62-57nal treatment on nursery pig performance. Table Growth performance: A total of 24 weaned barrows (initial BW approximately 14 lb; Landrace × Large White) were used in a 7 d study. Pigs were housed 2 pigs per pen with a total of 4 pens per treatment. Treatments refer to the following: Control = Pigs administered a sham challenge of PBS each day, Acute = pigs administered approximately 10 10 CFU of E. coli 62-57nal on d 1 then sham challenged for the remaining time, Chronic = pigs administered approximately 10 8 CFU of E. coli 62-57nal from d 1 to d 6.  14 had similar results with a decrease in weight of weaned piglets inoculated with pathogenic E. coli expressing K88 fimbriae from day 0 to day 2 which then recovered by day 9 of the trial. In this study, the acute challenge group had the poorest mean G:F conversion with the control group having the highest mean feed efficiency. Piglets experiencing PWD have been reported to exhibit reduced weight gains 3,15 , however statistically significant reductions in weight performance were not observed, perhaps due to the relatively brief duration of the trial or small sample sizes.
Bacterial colonization and fecal shedding. The ability of E. coli 62-57nal to colonize the gastrointestinal tract of inoculated piglets was determined by measuring colony-forming units recovered from intestinal mucosa, intestinal luminal contents, and in feces. Inoculated strain counts adherent to the mucosal lining were found to be variable, with ~ 50% of samples, ranging from 0.21 to 1.71 g of intestinal scraping, yielding counts above the detection limit (5000 CFU/ml of tissue homogenate). Of the samples yielding enumerable colonies, bacterial counts ranged from ~ 10 4 to ~ 10 7 CFU/g in the duodenum, jejunum, ileum, cecum and colon (Fig. 1A). Bacterial counts in the cecal and colonic luminal contents, ranging from 0.14 to 11.07 g of digesta, were more reliably above the detection limit and ranged from ~ 10 3 to ~10 9 CFU/g, suggesting bacterial proliferation in the unattached population. Acute and chronic treatments had higher levels of E. coli 62-57-nal in feces (~ 10 5 to 10 7 CFU/g) than control pigs on all sampling days (Fig. 1B). Despite the preventative efforts described in the Materials and Methods, control pens sporadically shed E. coli 62-57-nal in the feces at levels near the lower detection limit (500 CFU/g), likely reflecting low levels of pen cross-contamination. Pigs administered the acute pathogenic E. coli dose exhibited significantly higher fecal shedding on day 1 (~ 10 7 CFU/g, P = 0.001) compared to the chronic dose, however there was no statistically significant difference in fecal E. coli 62-57nal counts between the acute and chronic treatments after d 1. This result is consistent with other piglet studies which observed peak shedding between 24 and 48 h post-inoculation 16,17 . Pathogen shedding in the acute group remained high through d 6, indicating that the E. coli 62-57nal was able to persistently colonize the gastrointestinal tract of swine. Although, the possibility of low levels of reinfection from the contaminated pens cannot be completely ruled out. High inter-animal variability in intestinal bacterial loads and fecal shedding has been observed in other studies examining F18 E. coli challenge in immature pigs, commonly spanning 3-4 orders of magnitude in both genetically susceptible and resistant populations 18,19 .

Markers of inflammation and intestinal leakage. Infection-induced inflammation is mediated by
increased levels of pro-inflammatory cytokines 20 . To evaluate the level of E. coli induced inflammation, serum interleukins 6 and 8 (IL-6 and IL-8) were measured. IL-6 and IL-8 are useful biomarkers since they have been linked to intestinal inflammation [21][22][23][24] and have been employed as markers of porcine immune activation in response to E. coli previously [25][26][27][28] . On d 6 of the study, pigs challenged with E. coli 62-57nal had increased (P < 0.05) concentrations of serum IL-6 compared to control pigs (Fig. 2). However, there was no difference in Increased concentrations of fecal calprotectin have been positively correlated with the histological activity of inflammatory bowel disease in humans 30 , and serve as a marker of neutrophilic intestinal inflammation 31 . Fecal calprotectin has also been suggested as a noninvasive marker of intestinal inflammation in swine 32 , however additional study is necessary. Past studies have investigated calprotectin in swine plasma, intestinal lumen, and jejunal mucosa, all of which were found to be correlated with bacterial infection 33,34 . In the present study, there were no significant treatment effects on fecal calprotectin concentration. This is the first study to our knowledge to test calprotectin levels in the feces of pigs inoculated with E. coli, and this indicates fecal calprotectin may not be an informative biomarker in this model. Villi length and histopathology. Previous studies involving swine challenged with ETEC strains have reported villous atrophy and reductions in crypt depth 35 ; similar symptoms have also been reported in chronic intestinal enteropathies in humans 36 . At time of necropsy, sections were collected to evaluate villi length in the piglet model, but no morphologic changes were observed between treatment groups. General bacterial rod attachment was evaluated by an anatomic pathologist and observed sporadically in all samples with no apparent correlation between rod attachment and direct bacterial plating as only 38% (11/29) of samples with rod attachment tested positive for E. coli 62-57nal by direct plating. Villus length in E. coli challenged animals did not differ from the controls in the duodenum (P = 0.7125), jejunum (P = 0.3719), and ileum (P = 0.778). Lack of villus blunting may be due to the limited duration of this study. A prior longer-term study (21 d), with a murine model of chronic intestinal inflammation obtained villus blunting through a combination of bacterial challenge and malnutrition 29 . Similarly, post-weaning anorexia in piglets has been shown to be associated with reduced villus heights 37 . Therefore, given a longer trial period and/or malnourishment, blunting may have been eventually observed in our present model. Histology is also only able to evaluate a tiny fraction of the intestinal tract, so lesions must be broadly distributed throughout the tissue to be detectable by this method. Based on this data, histologic analysis does not appear to be a useful method for evaluating this model.

Effects on the microbiome by 16S qPCR analysis.
To observe any changes of the gut microbiota caused by our acute or chronic dosing treatments, targeted 16S qPCR was performed for select bacterial groups on fecal samples collected from pens at day −1, day 1, day 3, and day 6. Relative abundances obtained were consistent with previous examinations of the piglet microbiome, showing a microbiome dominated by Bacteroidetes and Firmicutes 38,39 . Overall, the bacterial groups tended to increase relative to control and pre-treatment samples, likely due to natural microbiome succession. A summary of these significant (P < 0.05) or marginally significant (P < 0.10) bacterial group changes at each time point is shown in Table 2. Both acute and chronic  showing increased levels from pre-to post-treatment. The control group showed only two altered bacterial groups, Enterococcus and E. coli, from pre-to post-treatment. The observed increase for Enterococcus and E. coli within the control treatment is consistent with the previously reported natural post-weaning piglet microbiome maturation which shows an increase in levels of Enterococcus and Enterobacteriaceae at 8 days postweaning 40 . The acute dose of E. coli had a slightly more pronounced impact on the gut microbiome maturation than the chronic dose, however both acute and chronic treatments were sufficient to cause a detectable dysbiosis. This dysbiosis indicates incidences of ETEC amongst weaned piglets can significantly disrupt the normal microbiome progression during this transitionary period. Principal component analysis (PCA) of 16S qPCR results also provides clear evidence of dysbiosis in E. colitreated groups (Fig. 3A). PCA of pre-vs post-treatment samples indicates that by day 6 the acute treatment clearly clustered away from its pretreatment sample, while the chronic day 6 sample showed an intermediate clustering from its pre-treatment sample. In contrast, the control group remained tightly clustered throughout the trial period. This contrast in clustering suggests the microbiome perturbations are induced by the E. coli challenge and are not merely normal microbiota progression. The observed relative stability of the control microbiome is consistent with other studies, which reported a microbiome shift immediately after weaning and reached relative stability within 10 days after weaning 39 . These findings indicate that both the single acute dose and the chronic lower dose of E. coli 62-57nal caused varying degrees of a similar dysbiosis. While in this present study the acute dose of pathogenic E. coli provided a more pronounced microbiome defect, the slight alteration caused by the chronic dose may still be more reflective of chronic exposure to pathogens, as piglets may experience.

Alterations in the fecal metabolome.
To further characterize the differences in disease state caused by acute and chronic E. coli 62-57nal challenge, untargeted metabolomics was performed on fecal samples collected pretreatment (d −1) and days 1, 3 and 6 post-treatment. Metabolite profiles of fecal samples were analyzed by Metaboanalyst 41 . Analysis of the identifiable metabolites by PCA clearly distinguished between challenge and control groups (Fig. 3B). Similar to the results of microbiome analysis (Fig. 3A), the acute and chronic day 6 samples clearly cluster separately from their pre-treatment samples, while the control samples did not separate. The stability of the control group indicates the natural enzymatic, microbial, and structural maturation of the weaned piglet gut 40 was not responsible for the observed shifts in the acute or chronic treatment groups.
As with the 16S qPCR approach, metabolomic comparison of post-treatment samples with their respective pre-treatment samples was more informative when identifying significant changes in individual metabolites. Volcano plots (P-value < 0.10 and > twofold change) were used to identify metabolites that significantly changed following treatment 41 (Fig. 4). A full list of metabolites identified by volcano plot is provided in S1 Table. Metabolomics data related to swine, while growing, is fairly limited [42][43][44][45][46] , particularly metabolomics pertaining to piglets in a diseased stated, most of which are serum or tissue based [47][48][49][50] . The Human Metabolome Database 51 is much and LOGSQ treatment mean at day 6 compared to its corresponding pretreatment sample at Day-1 (Right). Non-parametric Wilcoxon exact test performed on LOGSQ. Only those significant or marginally significant listed here, significant defined at P < 0.05 and marginally significant at P < 0.10. (▲) or (▼) indicate mean LOGSQ are greater than or less than relative to control or Pretreatment. Significant** Marginally Significant*.   52 . Given these limitations in existing swine metabolomics data, the metabolites identified in this manner were categorized based on the Human Metabolome Database chemical taxonomy and compared with relevant human diseased states in which fecal metabolomics are more prevalent. Only four identified metabolites were shared between the chronic and control groups, and ten were common to both the chronic and acute treatment groups (Fig. 4). Changes in metabolites in the control group were presumed to be associated with the normal development of the weaned piglet gastrointestinal tract. Within the chronic and acute treatments, increased levels of amino acid metabolites were identified in the post-treatment samples, including lysine, ornithine, homoserine, histidine, tyramine, beta-alanine, (Fig. 4, S1 Table). The increased levels of amino acids and amino acid metabolites in post-treatment fecal samples suggests the E. coli 62-57nal treatment led to amino acid malabsorption and/or secretion, likely due to disruption of chemical gradients, inflammation, and microbiome perturbations within the gut caused by pathogenic E. coli treatment. Additionally, metabolites associated with bacterial amino acid degradation, 5-aminovaleric acid and putrescine, were found at increased levels within acute and chronic treatment samples. These have been previously associated with ulcerative colitis 53,54 . The presence of these metabolites is consistent with the model that suggests inflammation caused by E. coli treatment induced amino acid malabsorption and subsequent degradation by the resident microbiota. Increased presence of fecal amino acids is supported by Wu et al. who reported decreased amino acid metabolites in jejunal tissue from ETEC challenged piglets 49 , supporting malabsorption in small intestine leading to increased presence of amino acids in the feces. This observation of increased fecal amino acid and amino acid metabolite levels also agrees with studies examining fecal metabolite profiles of humans with inflammatory bowel diseases like ulcerative colitis and Crohn's disease of which chronic diarrhea is a symptom 55,56 .
Fecal metabolites that were significantly reduced in post-E. coli treatment samples were primarily fatty acid metabolites (Fig. 4), including stearic acid, myristic acid, and arachidic acid. The levels of lipid-soluble alphatocopherol (vitamin E) was also reduced in both treatment groups. The observed depletion of fatty acids within the feces may be indicative of immune system activation, which is supported by our observation of increased www.nature.com/scientificreports/ levels of serum IL-6 and IL-8. Growing evidence on the importance of "immunometabolism" suggests activated macrophage subtypes and chronically activated T-cells demonstrate increased uptake of fatty acids as they rely more on fatty acid oxidation in order to maintain the high energy levels required to mount an immune response 57,58 . Prior studies examining metabolomic profiles of human inflammatory bowel diseases found pronounced decreases in the levels of short chain fatty acids (SCFA), which are the end products of bacterial fermentation that are absorbed by the large intestine; this presumably signaled a dysbiosis of gut flora 55,56 . In our current study a reduction in the SCFA metabolites butyrate, alpha-ketoglutarate and fumaric acid were observed. Our metabolomic findings indicate both the chronic and acute E. coli 62-57nal treatments caused sufficient dysbiosis to statistically distinguish pre-and post-treatment samples (Fig. 3B) in large part due to amino acid www.nature.com/scientificreports/ malabsorption and reduction in fatty acid metabolites (Fig. 4), generating metabolomic profiles resembling those of human inflammatory gastrointestinal disease associated with diarrhea.
Whole genome sequencing. To better understand the gene content of E. coli 62-57nal that may contribute to its pathogenicity, its genome was sequenced. The genome of E. coli 62-57nal was assembled into 378 contigs of greater than 200 bp totaling in 5.6 Mbp length and at an average 45-fold coverage. The entire set of 378 contigs was submitted for sequence typing using SerotypeFinder v1.1 59 and confirmed to be O138 and H14 as reported previously 60 . Analysis of the assembled contigs by BLASTx (> 40% identity, E value < 10 -5 ) against a database of known E. coli virulence genes identified a number of genes in E. coli 62-57nal associated with pathogenesis, shown in S2 Table. Major identified virulence factors in E. coli 62-57nal include hemolysin (hlyABCD), Shiga toxin (stx2e), an intimin-like adhesin (fdeC), heat-stable enterotoxin II (stiI), heat-labile enterotoxin IIA (eltAB), a heme uptake system for acquisition of iron (chuUAVYTWSX), and the F18ac + fimbrial adhesin. Hemolysin (Hly) is an exotoxin that is associated with many pathogenic strains of E. coli 61 . Hemolysin is primarily thought to have a role in pathogenesis in extra-intestinal infections, such as those of the urinary tract or septicemia and studies have shown that Hly plays little to no role in clinical signs of diarrhea 62 . However recent data, using both in-vivo murine and in-vitro models, show secretion of Hly can disrupt tight-gap junctions and increase colonic mucosal inflammation 63,64 . This inflammation from Hly has been proposed as a contributing factor for ulcerative colitis in humans, a chronic inflammation 63 . Consistent with E. coli 62-57's original edema isolation source, an Stx2e Shiga toxin was identified on a putative prophage element. The Stx2e subtype is known to be associated with edema disease of swine 65 , however in our present study we did not observe signs of edema in challenged piglets. The primary contributors to the observed scouring phenotype were most likely the identified heat-stable enterotoxin II (STIIB) and heat-labile enterotoxin IIA (LT-IIA). While acting by different modes, both STIIB and LT-IIA have been shown to cause release of water and electrolytes from host membranes, thereby causing diarrhea or scours 11,66 . Heat stable and heat labile enterotoxins are the most common exotoxins that are associated with diarrhea in piglets, present in 72% and 57% of ETEC isolates from piglet scours, respectively 67 . E. coli 62-57nal also contains a number of virulence factors associated with colonization and survival within the host, including the iron scavenging chu genes and various adhesin genes coding for the proteins AIDA-I autotransporter, F18ac + fimbrial adhesin, and the intimin-like FdeC 61,67,68 . Taken together, the presence of these virulence factors explains the observed diarrhea/scouring and colonization phenotype. Furthermore, the enterotoxin mode of action that induces this diarrhea also disrupts the Na + gradient 66 , which amino acid absorption is dependent upon in the gut 69 , likely explaining the increased amino acid levels observed in the fecal metabolomes of ETEC-challenged animals.

Conclusion
Suitable piglet models exploring the pathogenesis of the E. coli pathotypes responsible for common post-weaning diseases like PWD and ED are necessary for evaluating alternative treatment options. The model presented here may serve as a detailed model for the evaluation of novel PWD treatments. In this study, weaned piglets were challenged with either a single bolus of 1.5 × 10 10 CFU of a pathogenic STIIB + , Stx2e + , and LT-IIA + E. coli strain or average daily doses of 5.4 × 10 8 CFU of the same strain. Both chronic and acute treatment groups exhibited significant increases in fecal E. coli 62-57nal shedding, intestinal E. coli 62-57nal colonization and serum IL-6 levels compared to controls (Figs. 1, 2). Furthermore, both treatments induced similar levels of dysbiosis as measured by targeted 16S qPCR and untargeted metabolomics (Fig. 3). These alterations of the microbiome and metabolome consisted of significant increases in Bacteroidetes and Enterococcus, amongst others, and increased levels of amino acid metabolites within the feces. These findings imply that high acute doses of inoculum, as are often utilized in studies of E. coli gastrointestinal disease in pigs, are not necessarily required to establish a disease state, and lower levels of inoculum may represent a comparable disease state important for the study of chronic inflammation or chronic exposure to pathogens. Although, as presented here, it is unclear if a single low dose of E. coli would induce similar clinical signs to the high acute dose without repeated dosing. Future work may benefit from an additional treatment group consisting of a single low dose. In this study, fecal calprotectin measurements and histological examination of intestinal sections from challenged animals did not indicate any significant differences between control and treatment groups. Given the lack of blunted villi in this work which was previously associated with PWD, and likely a contributing factor of decreased nutrient uptake, future studies employing the model described here may benefit from an extended trial period, as well as controlled changes in animal nutrition, to achieve blunted villi.
In addition to a model for PWD, the relatively mild disease state presented here may also be adapted and appropriate for modeling of chronic human enteropathies in which an underlying dysbiosis is a presumed cause like inflammatory bowel disease 70 or the understudied environmental enteric disfunction 36 . Rodent models have often been used as disease surrogates, however these models are at a disadvantage when it comes to accurately representing human diseases and syndromes. Swine more accurately resemble humans in anatomy, genetics, and physiology, making them a more appropriate model for human biomedical research 71 . For example, many of the differential metabolites identified in this study (Fig. 4) and the increases in inflammatory cytokines (Fig. 2) reflect those found in human chronic inflammatory disorders 70,72 . However, if modeling of the villous blunting associated with disorders like environmental enteric disfunction is desired, increase in trial length and additional malnutrition may be required, as mentioned previously. Bacterial strains and culture conditions. Escherichia coli strain NCDC 62-57 (O138:K81(B):H14) was originally isolated from swine showing clinical signs of porcine edema disease 60 . This strain was obtained from the ATCC (ATCC 23,545). A spontaneous nalidixic acid-resistant mutant of this strain was isolated and used throughout this work, and will be referred to as strain 62-57nal. This mutation was stable through multiple consecutive transfers. The bacterium was routinely cultured in LB broth (Bacto tryptone 10 g/L, Bacto yeast extract 5 g/L, NaCl 10 g/L) or on LB agar plates (LB broth plus 15 g/L Bacto agar) aerobically at 37 °C. Samples obtained from animals challenged with E. coli 62-57nal were plated on MacConkey agar (Becton-Dickinson) amended with 50 µg/ml nalidixic acid (MacConkey + nal). Putative virulence factors were identified based on homology using BLASTx of WGS contigs with a custom database of E. coli virulence factors containing E. coli-associated proteins contained in mVirDB 75 and characterized E. coli virulence factors 61 . As a control, the genome of the non-pathogenic lab strain of E. coli MG1655 (Accession: GCF_000005845.2) was analyzed against the same database. Hits in common from MG1655 and 62-57nal were excluded based on the presumption that they were part of the non-pathogenic E. coli gene repertoire. Protein sequences identified in this initial screening were extracted from the 62-57nal genome and manually investigated using BLASTp and InterProScan to confirm conserved domains were intact and putative gene products were approximately full length 76,77 . The supplied O-and H-antigen serotype provided by ATCC were also confirmed bioinformatically using the SerotypeFinder v1.1 tool located at the Center for Genomic Epidemiology website 59 . Weaned piglet challenge model of E. coli 62-57nal. Animals and facilities. A group of 24 weaned barrows approximately 21 days of age (Landrace × Large White, initial mean BW 6.35 kg) were collected on the day of their weaning and housed at SPARC in College Station, TX. Pigs were randomly assigned to pens (4.634 m 2 ) with 2 barrows per pen that had solid concrete flooring and was equipped with a nipple waterer, rubber mat, and feeder. Genetic susceptibility to F18ac + ETEC infection was not determined in the present study. Pigs were provided ad libitum access to water and feed; the diet was a standard phase 1 nursery pig pelleted diet (S3 Table) formulated to meet or exceed the National Research Council (2012) recommended requirements of nutrients.

Whole genome sequencing of
E. coli 62-57nal challenge trial. The pigs were held 2 d prior to the start of treatment in order to be pre-screened for endogenous enterotoxigenic E. coli (ETEC). Pigs were randomly allotted to one of three treatments: Nonchallenged control, acute challenge (a single dose of 1.5 × 10 10 CFU), and chronic challenge (an average daily dose of 5.4 × 10 8 CFU). Chronic daily doses ranged from 1.9 × 10 8 to 1.6 × 10 9 CFU with a median of 3.4 × 10 8 CFU. After random assignment of piglets to pens and treatments, researchers were unblinded in administration of the different treatments, and samples from each treatment were handled and collected uniformly. Each treatment had a total of 4 pens, 2 pigs per pen, for n = 8 pigs total. All pigs were housed in the same barn, with pens separated by empty pens to prevent cross-contamination between treatments. Additionally, disinfectant foot baths, routine cleaning of pens, disinfection of sampling equipment between treatment groups were used to prevent cross-contamination, and pig/pen sampling and dosing was performed in an order to reduce chance of contamination. Pens were routinely cleaned to further reduce sources of potential cross-contamination. E. coli 62-57nal was cultured overnight (16-18 h) in Tryptic Soy Broth (TSB; B-D Bacto) at 37 ºC with aeration. The acute treatment received a single dose of 6 ml of overnight E. coli culture on d 1 and the chronic treatment received a daily dosage of 6 ml of a tenfold dilution in phosphate buffered saline (PBS, Corning Cellgro) starting on d 1 until the termination of the trial. Pigs and feeders were weighed on d 0, 1, 3, and 6 for calculation of average daily gain (ADG), average daily feed intake (ADFI), and gain to feed (G:F). The d 1 collection of weight data, feces, and blood was approximately 12 h after the initial E. coli dose was administered. All animals were humanely euthanized and necropsied on d 7 for collection of intestinal scrapings and sections which were used for analysis of E. coli colonization and determination of villi blunting, respectively.
Pre-screening of animals for endogenous phage and ETEC. Prior  www.nature.com/scientificreports/ presence of 50 µg/ml nalidixic acid. Briefly, approximately 1 g of feces from each pen was mixed with 9 ml PBS, vortexed until homogenous, serially diluted (tenfold increments), and plated onto both MacConkey agar plates and MacConkey + nal. A chloroformed aliquot of the first sample dilution was plated onto Tryptic Soy Agar (TSA) plates overlaid with 0.5% top agar (5 g Tryptone, 5 g NaCl, 500 ml dH2O, 0.5% w/v Agar) inoculated with 100 μl from an overnight culture of E. coli 62-57nal. MacConkey + nal plates were screened for breakthrough colonies and TSA plates were screened for plaque formation or zones of clearing to determine phage presence. After overnight incubation, 3 colonies were selected from MacConkey plates from each pen and mixed with 150 µl Tris EDTA (TE) buffer. Each sample was boiled for 10 min then centrifuged at 8,000×g for 2 min. These colonies were screened via multiplex PCR for: Universal stress protein A (uspA), heat-labile toxin (LTI), heat-stable enterotoxin I (STI), heat-stable enterotoxin II (STII), Shiga toxin type 1 (Stx1) and Shiga toxin type 2 (Stx2) using previously established and validated primers 78,79 . Positive bands of appropriate size were confirmed using individual primer sets and resultant PCR products were visualized on a 1.5% agarose gel with gel red (Biotium).
Fecal collection and determination of E. coli 62-57nal and fecal calprotectin within feces. A representative fecal sample was collected from each pen on d 1, 3, and 6 to determine fecal E. coli populations. The samples were collected in individual, sterile 50 ml conical tubes and transported on wet ice to the laboratory. These fecal samples were processed and diluted in the same manner as described above, and were also spot plated (20 µl) on MacConkey agar containing 50 µg/ml nalidixic acid which were incubated for 10-12 h at 37 °C to avoid colonies merging within the spots. Fecal calprotectin was determined by a commercially available porcine ELISA kit (MyBioSource, San Diego, CA) with a minimum detection limit of 6.25 ng/ml with an intra-assay CV of less than 15%.
Blood sampling and serum analysis. Blood samples were collected from 2 pigs per pen on d 6. Blood was collected from the cranial vena cava via a 20 gauge needle and a 10 mL serum vacutainer tube (BD, with clot activator and gel for serum separation). Tubes were inverted and allowed a minimum of 30 min to clot. Samples were centrifuged at 1,600×g for 10 min at 2 °C, and the separated serum samples were stored at −80 °C until analysis was performed. Serum concentrations of interleukin 6 (IL-6) and interleukin 8 (IL-8) were determined via porcine ELISA kits (R&D Systems, Minneapolis, MN). The minimum detection for IL-6 was 18.8 pg/ml and 62.5 pg/ml for IL-8. Assays were conducted as outlined by the manufacturer.
Intestinal sampling and histology. Pigs were humanely euthanized, necropsied, and had samples collected for intestinal histology and E. coli populations from the duodenum, jejunum, ileum, cecum, and colon.
Segments of the small intestine (duodenum, jejunum, and ileum) and large (cecum and colon) intestine were tied off with plastic zip ties to prevent cross contamination. Adherent bacterial samples were collected by rinsing the intestinal mucosal surface with sterile PBS, scraping a 2-3 cm section of the surface with a glass microscope slide, and then placing the sample into sterile 15 ml conical tubes containing 4.25 g of 2 mm glass beads (Walter Stern Inc.) and 8 ml of sterile PBS. Tissue scraping samples were vortexed for 5 min at 3000 rpm on a platform vortex mixer to homogenize samples. Cecum and distal colon contents were also collected and homogenized by thoroughly vortexing 0.5-5 g sample in 25 ml of PBS. Sample homogenates were serially diluted (tenfold increments) in PBS and spot-plated (20 ml) to MacConkey agar with 50 µg/ml nalidixic acid. All bacterial population estimates were normalized to the initial sample wet weight. At the time of the necropsy, histopathologists were blinded for both tissue sample collection and examination of fixed tissue for adherent bacteria as well as evidence of tissue inflammation. For histology, the distal end of each intestinal segment (duodenum, jejunum, ileum, cecum and colon) directly adjacent to the section used for bacterial sampling was collected. Samples were positioned onto a 5.08 cm × 5.08 cm cardboard sections and secured with small clips to prevent tissue curling. Consecutive tissue samples were fixed in Carnoy's solution (60% ethanol, 30% chloroform, 10% glacial acetic acid) at a 20:1 ratio for 30-45 min and 10% neutral buffered formalin (VWR scientific) at a 10:1 ratio for 24 h, followed by storage in 70% Ethanol until further processing. Tissues were trimmed into longitudinal sections of 5 mm width and embedded into paraffin using standard procedures 80 . After processing, 4 µm sections were placed on slides and stained with hematoxylin and eosin and evaluated histologically in a treatment blinded fashion. The slides were analyzed by a board-certified pathologist for rod attachment, presence of inflammation, and morphological changes (i.e., villous blunting, epithelial erosion). A total of 12 pigs, 4 from each treatment, were randomly selected for the evaluation of villus length, and well-oriented and intact villi were measured from the lamina muscularis mucosae layer to villus tip.
Analysis of fecal metabolites. Fecal samples collected from the floors of pens on d −1, 1, 3 and 6 were lyophilized and sent to the West Coast Metabolomics Center at the University of California Davis for untargeted metabolomic analysis. Untargeted GC-TOF profiling was performed following previously published parameters 81 . The resultant dataset is available on Metabolomics Workbench 82 , under study number ST001041.
16S qPCR. For the purposes of quantifying select bacterial populations, microbial DNA was extracted from 50 mg of lyophilized feces using the Zymo Quick-DNA Fecal/Soil Microbe Kits following the manufacturer's instructions. Five ng of DNA was used to amplify 16S regions of select bacterial groups using Biorad SsoFast EvaGreen Supermix using reaction conditions described previously 83 and family/genus/species specific primers described previously 83,84 . qPCR data is reported as log 10  www.nature.com/scientificreports/ Statistical analysis. Growth performance along with cytokine and intestinal bacterial population data were analyzed using the PROC MIXED procedure in SAS 9.3 (SAS Inst. Inc., Cary, NC). The model fixed effect was treatment with pen set as a random effect for growth performance, cytokine and intestinal bacterial population data. Fecal samples were collected on a pen basis, therefore pen was not included as a random effect. Calprotectin levels and fecal colony counts were analyzed as repeated measures using the PROC GLIMMIX procedure. Treatment, day, and treatment × day served as fixed effects. Day of collection also served as the repeated measure with pen as the subject. Metabolite data was normalized to sum, mean-centered and divided by the standard deviation of each variable, and analyzed for significant or trending metabolites between treatments using Meta-boAnalyst version 4.5 41 . Comparison of qPCR LOGSQ values was carried out using JMP Version 13 (SAS Inst. Inc., Cary, NC.). qPCR treatment means were compared pairwise on a per time point basis; many of the datasets did not pass the Shapiro-Wilk test for normality and were of small sample size, therefore treatments were compared using the nonparametric Wilcoxon Exact Test. qPCR data was also considered using multivariate methods on a pre/post treatment basis using principal component analysis (PCA). Results were considered significant at P ≤ 0.05 and marginally significant between P > 0.05 and P ≤ 0.10.

Data availability
Data supporting the findings of this study are available within this article and/or supplementary materials, additional supplements can be made available from corresponding author upon reasonable request. Data pertaining to resultant metabolomic analysis is available at Metabolomics Workbench, under study number ST001041. Genomic sequencing data available from Genbank under Bioproject (PRDF00000000).