Gut Dysbiosis and Its Associations with Gut Microbiota-Derived Metabolites in Dogs with Myxomatous Mitral Valve Disease

ABSTRACT Gut dysbiosis and gut microbiota-derived metabolites, including bile acid (BA), short-chain fatty acid, and trimethylamine N-oxide (TMAO), are associated with cardiovascular disease. Canine myxomatous mitral valve disease (MMVD) is a model for human MMVD. The aim of the study is to evaluate gut microbial dysbiosis and its relationship with gut-produced metabolites in dogs with MMVD. Fecal samples from 92 privately owned dogs, including 17 healthy, 23 and 27 asymptomatic MMVD dogs without (stage B1) and with (stage B2) secondary cardiac enlargement, respectively, and 25 MMVD dogs with history of congestive heart failure (stage C or D), were analyzed by 16S rRNA sequencing. Alpha and beta diversities were different between healthy and MMVD dogs (adjusted P < 0.05). The average dysbiosis indexes were −1.48, −0.6, 0.01, and 1.47 for healthy, B1, B2, and C/D dogs, respectively (P = 0.07). Dysbiosis index was negatively correlated with Clostridium hiranonis (P < 0.0001, r = −0.79). Escherichia coli, capable of trimethylamine production in the gut, had an increased abundance (adjusted P < 0.05) and may be responsible for the increased circulating TMAO levels in stage B2 and C/D MMVD dogs. Primary and secondary BAs showed opposite associations with C. hiranonis, a key BA converter (P < 0.0001 for both, r = −0.94 and 0.95, respectively). Secondary BAs appeared to promote the growth of Fusobacterium and Faecalibacterium but inhibit that of E. coli. Multivariate analysis revealed significant but weak associations between gut microbiota and several circulating metabolites, including short-chain acylcarnitines and TMAO. IMPORTANCE Our study expands the current “gut hypothesis” to include gut dysbiosis at the preclinical stage, prior to the onset of heart failure. Gut dysbiosis index increases in proportion to the severity of myxomatous mitral valve disease (MMVD) and is inversely associated with Clostridium hiranonis, a key bile acid (BA) converter in the gut. Secondary BAs appear to promote the growth of beneficial bacteria but inhibit that of harmful ones. An intricate interplay between gut microbiota, gut microbiota-produced metabolites, and MMVD pathophysiological progression is implicated.

B1 and 27 stage B2 dogs with asymptomatic MMVD (groups B1 and B2, respectively), and 25 dogs with MMVD and history of CHF (group C/D). No age difference was found except between group A and group C/D (P , 0.05). No differences in body weight, body condition score (BCS), or sex was observed. A total of 12.6 million paired-end sequences were obtained. The median sequence length after trimming and filtering was 443 nucleotides, with the interquartile range (IQR) between 440 and 461 nucleotides.
A total of 528 operational taxonomic units (OTUs) with known taxonomic lineages were identified, and their abundances were calculated (see Table S1, tab 1, in the supplemental material). The abundances of taxa in the ranks of phylum, class, order, and family were also calculated based on the OTU abundances (Table S1, tab 2).
Alpha and beta diversities. The alpha diversity indexes based on the Faith's phylogenetic diversity (PD) metric and the number of distinct species were calculated ( Fig. 1A and B and Table S1, tab 3). Significant differences in both indexes were observed between group A and the three MMVD groups (adjusted P , 0.05 in all cases).
Significant changes on the Bray-Curtis (BC) distances were observed among the groups using the permutational multivariate analysis of variance (PERMANOVA) test (P = 0.008) ( Fig. 2A; Table S1, tab 4). Differences were observed between group A versus group B2 and versus group C/D (PERMANOVA, P = 0.013 and 0.005, respectively). But the difference between group A and group B1 did not reach statistical significance (P = 0.06). On principal coordinate 1 (PC1), which accounted for 22.1% of the data variance, a difference between group A and group C/D was observed (adjusted P = 0.038) (Fig. 2B). No difference was found on PC2.
No difference in alpha diversity or beta diversity was observed between groups B1, B2, and C/D.
Age effect on alpha diversity. To test the hypothesis that the observed diversity differences were independent of age difference, bootstrap subsampling was performed. In one simulation, 125 bootstrapped subsamples with no age difference were identified (P age . 0.05). Of those, significant changes in Faith's PD were observed in 85.6% (107/125) of the bootstrapped data sets (P , 0.05 in all cases) (Table S1, tab 5), with the median P value of 0.018 (IQR, 0.007 to 0.031) (Fig. 1C).
Taxonomical differences. First, we examined the overall changes in taxa. No significant change in abundance was observed in any taxonomical rank (Table S1, tab 6, and Fig. S1). The five predominant phyla, Firmicutes, Bacteroidetes, Fusobacteria, Proteobacteria, and Actinobacteria, accounted for more than 99% of the total bacteria ( Fig. 1D; Fig. S1A). Fusobacteria was increased in groups B1 and B2 but decreased in group C/D compared to that in group A. However, these changes did not reach statistical significance. At the family level, Paraprevotellaceae were decreased while Actinomycetaceae were increased (false-discovery rate [FDR] = 0.07 in both cases) (Fig. S1B).
Second, we examined individual OTUs. Fifteen significant OTUs were identified (P , 0.05, q , 0.20 in all cases) (Table S1, tab 7). Among them are six species, five genera, and three families. The abundances of Megamonas, Blautia, Bacteroides, and Turicibacter were decreased in MMVD dogs versus those in healthy dogs, while Oscillospira was increased ( Fig. 3A to E). At the species level, Eubacterium dolichum, Faecalibacterium prausnitzii, Blautia producta, and Butyricicoccus pullicaecorum had reduced abundances, while Escherichia coli and Bacteroides uniformis had increases in MMVD dogs compared to that in healthy dogs ( Fig. 3F to K). The abundance of Ruminococcaceae was increased while those of Bacteroidaceae and Erysipelotrichaceae were decreased (Fig. 3L to O).
Fecal DI, BA, and SCFA. The fecal samples from 121 dogs, including 85 (85/92) from the 16S sequencing study, were included for the quantitative PCR (qPCR)-based dysbiosis (DI) analysis ( Fig. 4; Table S1, tab 8). The average DIs for groups A, B1, B2, and C/D were 21.48, 20.6, 0.01, and 1.47, respectively (Kruskal-Wallis test, P K-W = 0.07) (Fig. 4A). The difference between group A and group C/D was significant (adjusted P = 0.034). Similar to the 16S rRNA sequencing data, the abundances of Turicibacter and E. coli were different (P K-W = 0.007 and 0.025, respectively). Turicibacter was more abundant in group A than in groups B1 or C/D (adjusted P = 0.035 and 0.002, respectively) ( Fig. 4D), while E. coli abundance was greater in group C/D than in group A or group B1 and greater in group B2 than in group B1 (adjusted P = 0.032, 0.026, and 0.043, respectively) (Fig. 4F). The abundances of total bacteria appeared to be equal between the groups (Fig. 4B).
No difference was observed in fecal BAs except glyco-CA (P K-W = 0.008) (Table S1, tab 9). Group B2 had a greater amount of glycol-CA than group A or group B1 (adjusted P = 0.002 and 0.024, respectively). No difference was found in fecal SCFAs (Table S1, tab 10).  Faith's PD index showed a negative correlation with DI but a positive one with Clostridium hiranonis (P = 0.0006 and 6e 26 ; r = 20.37 and 0.47, respectively) ( Fig. 5A and B). The BC distance-based plots showed visible clustering of samples along PC1 by DI or C. hiranonis abundance (P = 1.9e 214 and 8.5e 214 , respectively) ( Fig. 5C and D). No difference was observed on PC2 (P . 0.7).
Correlations between DI, BAs, and bacterial abundances. DI was positively associated with Streptococcus and E. coli but negatively associated with Faecalibacterium, Fusobacterium, and C. hiranonis (jrj . 0.6, P , 1e 26 in all cases) (Fig. S2A to E and Table S1, tab 11).
Faecalibacterium, E. coli, and Fusobacterium showed modest to moderate associations with BAs (P , 0.0001 in all cases) ( Fig. 7A to F, green for DI # 0, red for DI $ 2, gray for 0 , DI , 2; Table S1, tab 12). While E. coli was positively associated with primary BA but negatively associated with secondary BA, the opposite was observed for Faecalibacterium and Fusobacterium. Strong associations between C. hiranonis and BAs were observed. While positive associations with the secondary bile acids DCA and LCA were observed (P = 2.2e 216 in both cases) ( Fig. 7G and H), C. hiranonis was inversely correlated with the primary bile acids CA and CDCA, the secondary bile acid UDCA, and 1°/2°(P , 5.4e 210 in all cases) ( Fig. 7I to L). Notably, when the log 10 abundance of C. hiranonis was below the threshold of 4.5, little conversion from primary BAs to secondary BAs took place ( Fig. 7G to L).
Correlations between OTUs, serum metabolites, and echo variables. Fifty dogs had matching serum samples in the previously published metabolomics study (Table S1, tab 13) (7). Generalized linear models identified several significant but weak correlations between the OTUs and metabolites (jrj $ 0.2, FDR # 0.05) ( Table 2; Table S1, tab 14). Remarkably, Megamonas showed positive associations with 7 short-chain acylcarnitines and carnitine, while Lactobacillus was positively correlated with 6 short-chain acylcarnitines. A negative correlation was observed between Erysipelotrichaceae and TMAO.
No correlation was found between OTUs and the three key echo variables after adjusting for multiple testing errors (FDR . 0.05 in all cases) (Table S1, tab 15).
Serum BAs. No BA met the stringent selection criteria in the serum metabolomics profiling study (7). The mean concentrations (in relative quantification) of CDCA were 1.04, 0.77, 0.95, and 1.39 for groups A, B1, B2, and C/D, respectively (P = 0.07) (Table S1, tab 16), and was higher in group C/D than in group B1 (adjusted P = 0.046).

DISCUSSION
A limited number of studies have examined the roles of gut microbiota in heart failure (HF) in humans and animal models (20)(21)(22)(23)(24)(25). Emerging evidence causally links gut microbiota to atherosclerotic disease, but observations between gut bacteria and HF remain associative (26). To our knowledge, no study has documented gut microbiota changes in the preclinical phases leading to HF. In this study, we compared gut microbiome profiles in MMVD dogs at preclinical B1 and B2 stages and those with histories of CHF with that of healthy dogs. Alpha diversity, which describes the number (richness) or distribution (evenness) of different bacterial species in the gut, was greater in the healthy dogs than in dogs with MMVD, but no difference was found among the three MMVD groups. So far, two studies reported reduced alpha diversities in human HF patients and in a rodent model with induced HF (24,25), but other studies showed no change in alpha diversity between HF subjects and healthy controls in humans or dogs (20)(21)(22)(23). We also observed nominal differences in beta diversity, which measures similarity of different microbial groups. Although it is difficult to fully understand the changes and assess the discrepancies with only a small number of microbiome studies on HF, our results clearly showed that shifts in gut microbiota began at the very early preclinical stages when dogs had little or no evidence of cardiac remodeling. Initially, we expected to see differences in gut microbial diversity among the three MMVD groups. Although both stage B1 and B2 dogs are asymptomatic of CHF, stage B2 dogs exhibit evidence of more advanced disease with hemodynamically significant mitral valve regurgitation, as evidenced by radiographic and echocardiographic findings of cardiomegaly. The stage C dogs are characterized as having exhibited past or current clinical signs of CHF secondary to MMVDs. The current "leaky gut" hypothesis postulated bowel wall edema and impaired intestinal barrier function due to heart failure (9). To date, no report on gut microbiota in preclinical patients before the onset of HF is available. It was recently reported that levels of circulating uremic toxins, including TMAO and other nitrogenous wastes, were increased in dogs with preclinical MMVD versus that in healthy dogs (7). We thus expand the current hypothesis that the changes in gut microbiota in the preclinical stage may have already compromised the integrity of intestinal barrier function to a degree that the resultant "leaked" gut-derived metabolites trigger an initial inflammatory response, leading to progressive worsening of MMVD. There is a trade-off between sensitivity and specificity in high-dimensional data analysis. While it is important to adjust for multiple testing, a stringent P value threshold may also increase false-negative calls. In a recent untargeted metabolomics study that linked gut flora metabolism and cardiovascular disease, L-carnitine was not on the top tier of metabolites that met the stringent P value cutoff but was later identified using less stringent criteria (8,12). As a hypothesis-driven study, we decided to relax the adjusted P value (q value) threshold to 0.20. We identified changes in five genera and six species, including E. coli and Turicibacter (P , 0.05, q = 0.16 in both cases), both of which were confirmed by qPCR (P = 0.025 and 0.007, respectively). Turicibacter, a genus in the phylum Firmicutes, is commonly found in the gastrointestinal (GI) tracts of animals. This bacterium influences the host's physiological processes by modulating certain gut microbe-dependent hormones (27)(28)(29). Specifically, Turicibacter sanguinis, one of the spore-forming gut microbes that were shown to signal gut cells to increase serotonin (5-HT) production, expresses a protein homologous to mammalian 5-HT transporter and is able to import 5-HT into the cell (28,29). Genetic interactions between Turicibacter sp. and bile acids were also reported (27). The 5-HT signaling  (30). Previous studies reveal an association between serum 5-HT concentration and disease state in which dogs at high risk for MMVD as well as dogs in early stages of MMVD exhibit increased serum 5-HT, whereas dogs with end-stage MMVD exhibit decreased serum 5-HT (31,32). The potential association between Turicibacter, 5-HT signaling, and MMVD pathogenesis warrants further investigation. Bacteroides, Blautia, and Megamonas all express enzymes for propionate production pathways and are normally found in the guts of healthy carnivores (33)(34)(35)(36)(37). Oscillospira, a genus of commensal bacteria that produces butyrate and is commonly found in healthy guts, appeared to confer protection against atherosclerosis and reduce plaque size in a mouse study (38)(39)(40)(41). F. prausnitzii and B. pullicaecorum, both members of the Ruminococcaceae family and clostridial cluster IV, produce a high level of butyrate with promising probiotic potentials (35,36). E. dolichum of the Erysipelotrichaceae family also produces butyrate and acetate (40). Butyrate and propionate possess well-known anti-inflammatory effects (42). The reduced abundances of these bacterial genera and species in dogs with MMVD suggested a net decrease in SCFA production in the gut and a reduced protection against inflammation. In addition, SCFAs are important signaling molecules which regulate diverse biological processes, including cardiovascular disease (9,11,19). However, no difference was observed in fecal SCFAs in our study. It was reported that 90% to 95% SCFAs produced in the colon were reabsorbed by the gut mucosa (43). Thus, the concentrations of fecal SCFAs may not accurately reflect those produced by gut microbiota. More studies are needed to explore whether the changes in these SCFA-producing bacteria have any effect on the levels of SCFAs in circulation.
A search for TMA-producing genes in microbial metagenomes showed that the E. coli genome displayed ;99% identities to the gene for carnitine oxygenase (cntA), the key gene in the main TMA synthesis pathway (44). Prior studies also reported increased E. coli abundances in human and canine HF patients compared to that in healthy controls (20,24). The observations that circulating TMAO concentrations were increased in dogs with B2 stage MMVD and CHF (6,7), and that E. coli abundance was higher in B2 and CHF dogs than in B1 or healthy dogs, underscored a potential involvement of E. coli in TMAO production. The dysbiosis index is a PCR-based tool that allows quantifications of gut dysbiosis using a panel of eight selected bacterial groups (45). Recent studies showed that DI was increased in dogs with inflammatory bowel disease (IBD) (46)(47)(48). We tested the hypothesis that DI increased with the severity of MMVD. Indeed, our data showed that the average DI progressively increased from 21.48 in group A, to 20.6, 0.01, and 1.47 in groups B1, B2, and C/D, respectively, and that the difference between groups A and FIG 7 Pearson's correlations between fecal bile acids (BAs) and gut microbes. CA, cholic acid; LCA, lithocholic acid; DCA, deoxycholic acid; CDCA, chenodeoxycholic acid; UDCA, ursodeoxycholic acid; 1°BA/2°BA, the ratio of primary to secondary BAs; r, correlation coefficient. (A to L) The percentage of each BA was calculated as the ratio of BA to the sum of primary and secondary BAs (y axis). Bacterial abundance was expressed as log 10 DNA abundance (x axis). Only unconjugated BAs were considered. Samples were colored by dysbiosis index (DI): green for DI # 0, red for DI $ 2, gray for others. P , 1e 25 in all cases.
Gut Dysbiosis in Canine Mitral Valve Disease C/D reached statistical significance (adjusted P , 0.05). Notably, there were considerable within-group variations. It is possible that a larger sample size may further improve statistical power. Nevertheless, this PCR-based DI tool showed promises to monitor MMVD onset and progression and offered an opportunity to improve its sensitivity and specificity with the selection of a cardiac-specific bacterial panel. The qPCR results validated the significant differences observed for Turicibacter and E. coli, and similar trend for F. prausnitzii but not for Blautia, from the 16S rRNA gene sequencing. Four bacteria showed moderate to strong correlations with DI: Streptococcus and E. coli had positive associations, while Fusobacterium and C. hiranonis had negative ones (P , 1e 210 , jrj $ 0.6 in all cases). Our results demonstrated a progressive increase in gut microbiota dysbiosis in MMVD dogs.
Gut microbiota influences host metabolism and physiology by producing numerous metabolites. Conversely, these metabolites reshape the structure and composition of gut microbiota (49). We further investigated the relationships among BAs, DI, and gut microbes. Bile acids have received considerable attention recently due to their ability to regulate many physiological processes as signaling molecules. Primary BAs are converted to secondary BAs by gut bacteria whose genomes contain the BA-inducible operons (bai) with 7a/b-dehydroxylase activities (50). Secondary BAs are thought to render protections against the growth of several pathogens, including Clostridium difficile, E. coli, and Clostridium perfringens (51)(52)(53). Deoxycholic acid, a secondary BA, reduces accumulations of inflammatory mediators in ileum, and supplementation of DCA in diet diminished C. perfringens-induced severe inflammatory effect and body weight loss in chickens (54). Interestingly, one recent study reported increased ratio of plasma secondary/primary BA ratio in 142 chronic HF patients compared with that in 20 healthy control subjects (55), but it is difficult to draw any meaningful conclusion with the sole observation. C. hiranonis, one of the Clostridium spp. that possess 7ahydroxylation ability (56), was one of the bacteria in the DI panel. Remarkably, when C. hiranonis abundance (in log 10 ) was .4.5, essentially all primary BAs were converted to secondary BAs (1°BA/2°BA % 0) (Fig. 7L), suggesting that, when abundant, C. hiranonis is sufficient to convert the majority, if not all, of primary BAs to secondary BAs. Furthermore, our data suggested that secondary BAs inhibited the growth of E. coli but promoted that of Fusobacterium and Faecalibacterium (Fig. 7A to F).
Positive correlations were found between circulating short-chain acylcarnitines and gut bacteria, Lactobacillus and Megamonas. The abundance of Megamonas was reduced in MMVD dogs compared to that in healthy dogs. Acylcarnitines, key intermediates of long-chain-FA transport and oxidation, accumulate in circulation as a result of incomplete or inefficient fatty acid oxidation and have been used as diagnostic markers for disorders in peroxisomal or mitochondrial oxidation processes (57)(58)(59). TABLE 2 Correlations between gut microbes and circulating carnitine and short-chain acylcarnitines (C 2 to C 6 ) a a Green color indicates a positive correlation, while orange color indicates a negative correlation between the corresponding microbes and metabolites. All correlations had adjusted P values of #0.05 and absolute values of correlation coefficient of $0.2. The analysis was performed using the multivariate statistical framework implemented in the R package MaAsLin2.
Elevated long-chain acylcarnitines in circulation were documented in human HF patients compared to that in normal ones (60,61). Accumulation of long-chain acylcarnitines was thought to contribute to HF by stimulating reactive oxygen species (ROS) production and releasing circulating inflammatory mediators (61). In dogs, the severity of MMVD was correlated with the concentrations of both short-chain and long-chain acylcarnitines (7), some of which were reduced in response to diet intervention with demonstrated clinical benefits (62,63). In addition, a negative association with TMAO was observed with the Erysipelotrichaceae family, whose abundance was reduced in MMVD dogs. Despite the significance (FDR , 0.05 in all cases), the associations between these gut bacteria and circulating metabolites were weak. It is important to note that only 50 samples had both fecal rRNA gene sequencing and serum metabolomics data and that there were only 6 dogs (6/50) in the healthy control group. Nevertheless, this pilot experiment may offer an opportunity for future studies.
We explored, for the first time, the relationships between gut microbiome and gutderived metabolites in dogs with all stages of MMVD. Because these dogs were privately owned, we were unable to rule out potential cofounding effects from diet and breed. In addition, all C/D dogs and the majority of the B2 dogs were on one or more common cardiac medications at the time of the study. Although the bootstrapping study supported the hypothesis that the changes were not due to age difference, we were unable to completely rule out the possibility of a small confounding effect from age. Significantly, our study expands the current "gut hypothesis" to include gut dysbiosis at the preclinical stages when there is little or no evidence of cardiac remodeling and lays a foundation for future microbiome research in canine and human MMVD.

MATERIALS AND METHODS
Animals and study approval. The study protocol was reviewed and approved by the University of Pennsylvania Institutional Animal Care and Use Committee, and informed owner consent was obtained. Clinically healthy dogs 7 years of age or older without a heart murmur and without concurrent systemic disease were prospectively enrolled as the control group (group A). This group of dogs primarily consisted of systemically healthy dogs owned by students and staff of the hospital. A cohort of dogs 7 years of age or older with a left apical systolic murmur and echocardiographic (echo) diagnosis of thickened and prolapsing mitral valve leaflet(s) and mitral regurgitation as well as clinical history and physical exam consistent with stage B1, B2, C, or D MMVD were considered for groups B1, B2, and C/D, respectively (3). Any dog with severe concurrent systemic disease, including diabetes mellitus, cancer, or renal failure, or those with any congenital heart disease were excluded. Dogs with signs of gastrointestinal illness such as vomiting or diarrhea and those that had received antibiotics within 30 days were also excluded.
Echocardiography. Echo studies (iE33; Philips Healthcare, Andover, MA) were performed without sedation. Left ventricular internal dimensions in end-diastole (nLVIDd) and end-systole (nLVIDs), left atrial diameter (nLAD), and aortic root diameter (nAoD) were measured from right parasternal short axis 2-dimensional images and normalized to body weight (64). The ratio of the left atrial diameter to the aortic root diameter (LA/Ao) was calculated.
Fecal 16S rRNA gene sequencing. Dog owners were instructed to collect a fresh fecal sample on the day of their pet's appointment. The sample was divided in aliquots and frozen at 280°C until analysis was performed.
Fecal genomic DNA extractions were performed using PowerFecal DNA isolation kit (Mo Bio Laboratories) and quantified by Quant-It Pico Green (Thermo Fisher Scientific) according to manufacturers' protocols. The 16S rRNA gene library was constructed according to Illumina's 16S metagenomic sequencing library preparation guide. Sequencing on the V3-V4 region of the 16S gene was performed in an Illumina MiSeq machine with 2 Â 250 cycles as previously described (65). The sequences for the 16S amplicon PCR forward and reverse primers were 59-TCGTCGGCAGCGTCAGATGTGTATAAG AGACAGCCTACGGGNGGCWGCAG and 59-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVG GGTATCTAATCC, respectively.
The bioinformatics pipeline for sequence analysis, alpha diversity, and beta diversity were described previously (65) and can also found in Text S1 in the supplemental material. Alpha diversity indexes of observed species, and Faith's phylogenetic diversity (PD) and beta diversity indexes based on the Bray-Curtis (BC) metric were calculated at the rarefaction depth of 7,000 sequences. Sequences that shared a minimum of 97% of identity were clustered into an operational taxonomic unit (OTU), and a taxonomical lineage was assigned to each OTU by searching the Greengenes database (August 2013 release).
DI, BA, and SCFA. Fecal samples from additional dogs were collected and included in these analyses. Fecal genomic DNA (100 mg) was shipped to Texas A&M University GI Lab (College Station, TX) for the quantitative PCR analysis to measure the DNA abundances of eight bacterial groups: total bacteria, Faecalibacterium, Turicibacter, Streptococcus, Escherichia coli, Blautia, Fusobacterium, and Clostridium hiranonis. The PCR primers and protocol and method for dysbiosis index (DI) calculation were described in detail previously (45). Bacterial DNA abundances were transformed using the logarithm with base 10.
Each fecal sample was split into two aliquots, one of which was shipped to Metabolon, Inc. (Morrisville, NC) for the targeted BA analysis, including 2 primary BAs (CA and CDCA) and 3 secondary BAs (DCA, LCA, and UDCA). The panel also included 10 glycine-and taurine-conjugated primary and secondary BAs. The other set of aliquots was shipped Texas A&M GI Lab (College Station, TX) for 4 straightchain and 2 branched-chain short-chain fatty acid (SCFA) assays. The quantitation of BAs was performed using an Agilent 1290 Infinity/Sciex QTRAP 6500 liquid chromatography-tandem mass spectrometry (LC-MS/MS) system equipped with a C 18 reverse-phase ultrahigh-performance liquid chromatography (UHPLC) column (Text S1; Table S2). The concentrations of SCFAs were measured using a stable isotope dilution gas chromatography (GC)/MS assay using the Agilent 6890N/Agilent 5975C system with a C 18 solid-phase extraction column (47,66).
The total amount of primary BAs was calculated by summing the concentrations of CA and CDCA, while that of secondary BAs was the sum of DCA, LCA, and UDCA. The percentage of each BA was calculated by dividing the concentration of the BA by the sum of all unconjugated BAs. The ratio of primary BAs to secondary BAs (1°BA/2°BA) was also calculated.
Alpha and beta diversity analyses were performed from the 85 samples that also had 16S sequencing data. Pearson's correlation analysis was performed between Faith's phylogenetic diversity (PD) index, DI, and C. hiranonis abundance. In addition, samples were trichotomized based on DI: L (DI # 0), M (0 , DI , 2), and H (DI $ 2), or dichotomized based on the abundance of C. hiranonis: L (log 10 DNA , 4.5) or H (log 10 DNA $ 4.5). Analysis of variance (ANOVA) was performed using the first two principal coordinates, PC1 or PC2, as the dependent variable on the categorical DI or C. hiranonis.
Statistical analysis. Total taxa abundances at the taxonomical ranks of phylum, class, order, and family were calculated using the OTU table. The relative abundance for each OTU or taxon was calculated using the total sum scaling method and was further transformed using square root. OTUs or taxa with 0% relative abundance in more than 50% of the samples in every group were removed. Two group comparisons were performed using the Mann-Whitney test. For multiple groups, the nonparametric Kruskal-Wallis (K-W) analysis of variance was performed, and significant OTUs or taxa were subject to post hoc Dunn's tests. The P values from both K-W tests and Dunn's tests were adjusted for multiple testing error using the Benjamini-Hochberg (BH) method. Taxa or OTUs with adjusted P values of less than 0.10 were considered significant. Data normality was evaluated using the Shapiro-Wilk test. To compare alpha diversity, ANOVA and Tukey's post hoc tests were performed. To compare beta diversity distances, permutational multivariate analysis of variance (PERMANOVA) was performed on the distance matrices with 1,000,000 permutations using the R package vegan (67). Principal-coordinate analysis (PCoA), an eigen-analysis of a dissimilarity matrix, was performed using the BC distance matrix. The first two principal coordinates (PCs), PC1 and PC2, were examined for their abilities to separate the groups using ANOVA and Tukey's post hoc tests in which each PC was the dependent variable and group was the predictor variable. To compare the differences in DI, SCFA, and BA, K-W analysis and Dunn's multiple comparisons were used.
Correlation analyses. Pearson's correlation analysis was performed between DI, PCR-based bacterial abundances, and BAs. Dogs with the matching serum samples were identified from the previously published metabolomics study, which reported 173 differential metabolites (7). The adonis function in the R package vegan was used to evaluate how many variations in the BC distance can be explained by each of the 173 metabolites. P values were determined by 1,000 permutations and adjusted for multiple testing using the BH method. The Spearman correlation was performed to assess the association between the alpha diversity indexes and the metabolites and between OTUs and echo variables. The association between individual OTUs and metabolites was performed using general linear models implemented in the R package MaAsLin2, a multivariate statistical framework that finds associations between microbial community abundances and clinical metadata (68). The MaAsLin2 function was used with the following settings: low abundance OTUs with a minimal relative abundance of 0.01% in less than 10% of the samples were filtered. The proportional value of each OTU was transformed by taking the logarithm to the base 10. Age, body weight, and MMVD group were included as random effects. The BH method was applied for multiple-testing corrections. Correlations with adjusted P values of less than or equal to 0.05 and correlation coefficients greater than 0.2 or less than 20.2 were considered significant.
Analysis of age effect. To assess the potential age contribution to gut microbial diversity, 1,000 iterations of bootstrap resampling experiments without replacement were performed, where 12 samples from each group were randomly selected. The Faith's PD indexes from each bootstrapped data set with no age difference were subject to ANOVA. The distributions of the P values on age, body weight, body condition score (BCS), sex, and Faith's PD were analyzed.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.1 MB.