The effects of signalment, diet, geographic location, season, and colitis associated with antimicrobial use or Salmonella infection on the fecal microbiome of horses

Abstract Background The fecal microbiome of healthy horses may be influenced by signalment, diet, environmental factors, and disease. Objectives To assess the effects of age, breed, sex, geographic location, season, diet, and colitis caused by antibiotic use (antimicrobial‐associated diarrhea [AAD]) and Salmonella infection on fecal microbiota. Animals Healthy horses (n = 80) were sampled from nonhospital environments across multiple geographical locations in the United States. Horses with AAD (n = 14) were defined as those that developed diarrhea secondary to antimicrobial use. Horses with Salmonella infection (n = 12) were presented with spontaneous onset of colitis and subsequently tested positive on Salmonella quantitative polymerase chain reaction. All horses were >1 year of age and stratified by a dietary scale that included forages (pasture and hay) and concentrates grouped by percentage of fiber and amount. Methods Illumina sequencing of 16S rRNA genes was performed on fecal DNA. Results Healthy horses fed higher amounts of grain clustered separately from those fed lower amounts of grain (analysis of similarities [ANOSIM], R = 0.356‐0.385, Q = 0.002). Horses with AAD and Salmonella had decreased richness and evenness compared to healthy horses (P < .05). Univariable analysis of the 3 groups identified increases in Bacteroidetes (Q = 0.002) and Protebacteria (Q = 0.001) and decreases in Verrucomicrobia (Q = 0.001) in AAD horses whereas Salmonella horses had less Firmicutes (Q = 0.001) when compared to healthy horses. Conclusions and Clinical Importance Although the amount of grain in the diet had some impact on the fecal microbiome, colitis had a significantly larger influence. Horses with ADD have a more severe dysbiosis than do horses with Salmonella.

in Verrucomicrobia (Q = 0.001) in AAD horses whereas Salmonella horses had less Firmicutes (Q = 0.001) when compared to healthy horses.
Conclusions and Clinical Importance: Although the amount of grain in the diet had some impact on the fecal microbiome, colitis had a significantly larger influence.
Horses with ADD have a more severe dysbiosis than do horses with Salmonella. Changes in the bacterial community of the gastrointestinal tract that occur in association with disease states, termed dysbiosis, affect these physiologically important metabolic processes. In humans, dysbiosis has been associated with inflammatory bowel disease, ulcerative colitis, Crohn's disease, obesity, and metabolic disease. 2 In companion animals such as dogs and cats, dysbiosis has been associated with acute and chronic enteropathies. [3][4][5] Researchers studying horses have utilized similar molecular techniques in hopes of decreasing the substantial morbidity and mortality attributed to gastrointestinal disease in the horse. 6 To date, studies have indicated that feces represent an adequate proxy for the hindgut of the horse, 7 with 17 to 19 phyla identified and present in variable relative abundance. [8][9][10][11] Inherent factors such breed, 8,11,12 age, 13 and pregnancy status 14 and external factors such as geographic location, 8 transport, 15 exercise intensity, [16][17][18] fasting, 19,20 and season 21 appear to have some influence on the fecal microbiome. Dietary variables such as exposure to pasture, 8,22 abrupt feed change, 22 and feeding concentrate vs forage 23 alter the microbiome to some extent, whereas the feeding of high starch concentrates 13,24 can induce gastrointestinal disease. To date, the factors with the greatest impact on the fecal microbiome include diet, antibiotic use, 25,26 and the presence of gastrointestinal disease such as colic or colitis. 27,28 Initial efforts to characterize the microbiome have been incomplete because of small sample size and use of university herds or horses at single locations, which may not adequately represent the equine population. [29][30][31] Furthermore, because these factors have been described independently across studies, the size effects of these individual factors remain unclear as does whether the dietary effect is similar to that of disease of antibiotics.
Our purpose was to define the fecal microbiota in a large population of healthy horses to assess the influence of age, breed, sex, geographic location, season, dietary factors, and health status. We hypothesized that these individual or external factors would have relatively minor impact on the fecal microbiome when compared to alterations induced by gastrointestinal disease. We chose 2 common variants of colitis seen in our hospital population, horses with primary colitis attributable to Salmonella and those with colitis induced by antimicrobial use.

| MATERIALS AND METHODS
No official animal care and use protocol was sought from the Texas A&M University Institutional Animal Care and Use Committee because all fecal samples were collected after natural elimination and clinical data were gathered historically.

| Study population and sample collection
The study consisted of 2 populations of horses, healthy and those with colitis. All fecal samples were collected during the period of in an effort to bank fecal samples from horses with gastrointestinal disease. Horses with antimicrobial-associated diarrhea (AAD) were defined as those that had received antimicrobial prophylaxis before elective surgery or to treat a suspected or known infection before the development of diarrhea. Horses in the AAD group had no history of gastrointestinal disease before antibiotic administration, and the clinician of record classified colitis as being associated with antibiotic use. Horses in the AAD group received the following antimicrobial agents: ceftiofur crystalline (n = 4); metronidazole (n = 1); doxycycline (n = 1); penicillin and gentamicin (n = 3); penicillin, gentamicin, and doxycycline (n = 2); penicillin, gentamicin, and metronidazole (n = 2); and chloramphenicol (n = 1). Antimicrobials were used to treat known infections in 11 horses (respiratory, n = 6; lacerations, n = 2; cellulitis, n = 1; navicular bursoscopy, n = 1; sequestrum, n = 1) and as surgical prophylaxis in 3 horses (check ligament desmotomy, fasciotomy and neurectomy for proximal suspensory disease, and resection of a skin tumor under general anesthesia).
Information regarding antimicrobial treatment of AAD horses is included in Supplementary Table 1.
Horses with Salmonella colitis were defined as those admit- Fecal samples from horses with colitis were defined as the first natural defecation upon admission and stored at 4 C until transported to the Gastrointestinal Laboratory at Texas A&M University for processing. Samples were stored at 4 C for ≤12 hours, and kept frozen at À80 C until DNA extraction.
The following information was collected for all horses: age, breed, sex, weight (estimated by the veterinarian), season of fecal collection, geographic location, and diet. Diet was analyzed by individual factors such as hay type, pasture type (warm season grasses such as Bermuda grass, Bahia or Buffalo pasture; transition zone grasses such as fescues or zoysia; cool season grasses such as Kentucky bluegrass and ryegrass), time spent in pasture (none, some or continuous), percentage maximum crude fiber in concentrate (low; 5%-8%, medium; 10%-15%, high; 18%-33%), and amount of concentrate (none, 0.5% and 1%-2% of body weight in kilograms per day).
Finally, a dietary scale was created that included for all aspects of diet. Horses were categorized as 1 of the following: A, forage only (hay, pasture or both); B, forage plus low fiber concentrate (5%-7%) fed at ≤0.5% of body weight in kg/day; C, forage plus medium fiber concentrate (10%-15%) fed at ≤0.5% of body weight in kg/day; D, forage plus high fiber concentrate (18%-33%) fed at ≤0.5% of body weight in kg/day; and, E, forage plus medium fiber concentrate (10%-15%) fed at 1%-2% of body weight in kg/day.

| Analysis of sequences
A total of 106 samples were analyzed, which generated 4 386 598 quality sequences. Sequences were analyzed using a QIIME 2 (Quantitative Insights into Microbial Ecology) 35 v.2019.7 pipeline as described elsewhere. 36,37 Briefly, barcodes and primers were removed and short (<150 bp), ambiguous, homopolymeric sequences were depleted from the dataset. The program divisive amplicon denoising algorithm (DADA2) was used to identify and remove chimeric sequences. 38 The amplicon sequence variant (ASV) table was created using DADA2, 39  Alpha diversity metrics Chao1 (richness), observed ASVs (species richness), and Shannon diversity (evenness) were generated in QIIME2.
Beta diversity was evaluated with the phylogeny-based weighted and unweighted UniFrac distance metric, and Principal Coordinate Analysis (PCoA) plots for visualization were generated in QIIME2.

| Statistical analysis
Before analysis, data was tested for normality using the Shapiro-Wilk test (JMP Pro 14, SAS, Marlow, Buckinghamshire, UK). Because data were not normally distributed, nonparametric measures were used throughout the study.
Beta diversity (bacterial community composition) was measured using weighted and unweighted UniFrac metrics and visualized for clustering using Principle Coordinate Analysis (PCoA) plots. An analysis of similarity test (ANOSIM) within the PRIMER 6 (PRIMER-E Ltd, Luton, UK) software package was performed on the distance matrices to assess the significance of the differences in the bacterial commu- Linear discriminant analysis effect size (LEfSe) using the webbased program Calypso v8.62 (http://cgenome.net/wiki/index.php/ Calypso) was performed to analyze the relative abundance of bacterial taxa and their associations with any of the 5 diet categories. A cutoff threshold of 3.5 was set for significance.

| Study participants
Breed, age, sex, and diet of the healthy horses and those with colitis are summarized in Supplementary Figures 1-4. Supplemental Table 1A contains information regarding signalment, diet, season of collection, geographic location, individual dietary components, and estimated weight of all healthy horses.

| Beta-diversity measures
A principal coordinate analysis plot of unweighted UniFrac distances in normal horses by diet category is shown in Figure 1. Figure  F I G U R E 1 Principal coordinates analysis (PCoA) plots based on unweighted UniFrac distances showing mild differences between healthy horses on diet E compared to diets A-D (ANOSIM, R = 0.156, P = .05) of healthy horses by diet. A, Horses fed forage only (hay and/or pasture), purple; B, horses fed forage plus low fiber concentrate (5%-7%) at ≤0.5% of body weight in kg/day, green; C, forage plus medium fiber concentrate (10%-15%) fed at ≤0.5% of body weight in kg/day, orange; D, forage plus high fiber concentrate (18%-33%) fed at ≤0.5% of body weight in kg/day blue; E, forage plus medium fiber concentrate (10%-15%) fed at 1% to 2% of body weight in kg/day, red. B, Horses on diets A-D combined (blue) show mild clustering from those on diet E (red) (ANOSIM, R = 0.15, P = .03)

| Alpha diversity measures
The alpha diversity metrics of healthy horses as stratified by diet cate-     Table 6.

| The effect of colitis associated with AAD and
Salmonella on the fecal microbiome A population of clinical patients with colitis associated with AAD (n = 14) and Salmonella (n = 12; Supplementary Table 1B) was compared to healthy horses previously described.

| Diversity between samples (beta diversity)
A principal coordinate analysis plot of unweighted UniFrac distances is presented in Figure 4, and indicates significant clustering between healthy horses and those with colitis (overall ANOSIM, R = 0.565, P = .0001).

| Diversity within samples (alpha diversity)
Alpha diversity indices were significantly lower in horses with colitis compared to healthy horses ( Figure 5), but horses with AAD and Salmonella were not different from each other. Observed ASVs were decreased in healthy horses compared to horses with AAD (diets A-D, P < .0001; diet E, P = .001) and Salmonella (diets A-D, P = .001; diet E, P = .02), but no difference was found between horses with each type of colitis. Chao 1 was significantly different between healthy horses and those with AAD (diets A-D, P < .0001; diet E, P = .02) and T A B L E 2 Pairwise ANOSIM analysis of unweighted and weighted UniFrac distances in healthy horses (diets A-D and diet E) and those with antimicrobial-associated diarrhea (AAD) and Salmonella colitis

| Enteric pathogens
Based on the inclusion criteria, none of the horses in the AAD or  horses. This can be seen on PCoA plots with AAD and Salmonella horses clustering distinctly from healthy horses, and with the AAD group having greater distance (ie, a more different microbiome composition) from healthy horses than those with Salmonella.
Horses with colitis had significant changes in the bacterial com- To date, the fecal microbiome of horses with colitis only has been described in a population of horses with undifferentiated colitis.
Although these horses also had significant decreases in richness and evenness, they experienced changes in the percentages of Bacteroidetes and Firmicutes simultaneously. These results suggest that colitis associated with antimicrobial use or Salmonella may have different effects on the fecal microbiome composition. Additional investigation utilizing a population of healthy horses and those with colitis better characterized by enteric pathogen testing, and potentially stratified by antimicrobial agent for AAD horses, is warranted.
The presence of colitis caused by AAD and Salmonella produced marked dysbiosis of the fecal microbiome with different effects on major phyla. Although both colitis groups clustered apart from healthy horses, horses with AAD had a larger shift in microbiome composition than did horses with Salmonella, compared to healthy controls. The effect of gastrointestinal disease was larger than that produced by diet.

This study was funded by The Donley Family and The Paula and
Michael Gaughan Fund.