Exercise Prevents Weight Gain and Alters the Gut Microbiota in a Mouse Model of High Fat Diet-Induced Obesity

Background Diet-induced obesity (DIO) is a significant health concern which has been linked to structural and functional changes in the gut microbiota. Exercise (Ex) is effective in preventing obesity, but whether Ex alters the gut microbiota during development with high fat (HF) feeding is unknown. Objective Determine the effects of voluntary Ex on the gastrointestinal microbiota in LF-fed mice and in HF-DIO. Methods Male C57BL/6 littermates (5 weeks) were distributed equally into 4 groups: low fat (LF) sedentary (Sed) LF/Sed, LF/Ex, HF/Sed and HF/Ex. Mice were individually housed and LF/Ex and HF/Ex cages were equipped with a wheel and odometer to record Ex. Fecal samples were collected at baseline, 6 weeks and 12 weeks and used for bacterial DNA isolation. DNA was subjected both to quantitative PCR using primers specific to the 16S rRNA encoding genes for Bacteroidetes and Firmicutes and to sequencing for lower taxonomic identification using the Illumina MiSeq platform. Data were analyzed using a one or two-way ANOVA or Pearson correlation. Results HF diet resulted in significantly greater body weight and adiposity as well as decreased glucose tolerance that were prevented by voluntary Ex (p<0.05). Visualization of Unifrac distance data with principal coordinates analysis indicated clustering by both diet and Ex at week 12. Sequencing demonstrated Ex-induced changes in the percentage of major bacterial phyla at 12 weeks. A correlation between total Ex distance and the ΔCt Bacteroidetes: ΔCt Firmicutes ratio from qPCR demonstrated a significant inverse correlation (r2 = 0.35, p = 0.043). Conclusion Ex induces a unique shift in the gut microbiota that is different from dietary effects. Microbiota changes may play a role in Ex prevention of HF-DIO.


Introduction
Approximately 35% of adults [1] and 17% of adolescents and children in the United States are obese [2] and this epidemic is fueling a rise in type-2 diabetes, metabolic syndrome and heart disease [3]. The financial burden on an already stressed economy has been estimated to be $147 billion/year in medical care alone in the US (2008 estimate) [4]. Lifestyle changes or other treatments that can prevent or limit obesity would significantly impact the health of the nation as well as reduce the socioeconomic burden on society. On average, Americans consume approximately 34% of their total dietary calories from fat [5], more than recommended by the American Heart Association [6]. High fat (HF) diets are known to induce metabolic stress on the body, leading to obesity and low grade systemic inflammation [7].
Recent research indicates that the gut microbiota induces the development of obesity in both humans [8] and mouse models [9,10]. Studies have shown that obese individuals have an imbalance in the primary bacterial phyla comprising the gastrointestinal (GI) microbiota: Bacteroidetes and Firmicutes [8,10,11]. Phyla level shifts are reported with HF, Western diets that appear to directly impact host metabolism and contribute to the development of obesity [10,12]. In addition, sub-phylum level changes with a HF diet have also been reported, including an increase in the Mollicutes [12] and Erysipelotrichi classes [13].
While the relationship between diet-induced changes in gut microbiota may be model dependent, ''microbiota transplant'' models using germ-free mice demonstrated that gut microbes from obese mice have a direct effect in increasing fat mass and body weight compared to introduction of microbes from lean mice [10,12]. Moreover, the obese-associated gut microbiota have increased capacity for energy harvest from ingested food [10] and when weight gain is controlled in the design, it is the HF diet, not body fatness, that determines whether a gut microbiota shift occurs [14].
Exercise (Ex) is known to confer numerous health-related benefits. Recent studies indicate that Ex can normalize body weight, body fat, hyperglycemia and markers of inflammation in mouse models of diet-induced obesity (DIO) [15,16]. In fact, Ex has been reported to share many of the same anti-inflammatory effects as calorie restriction in terms of reversing obesity and insulin resistance in a model of HF-DIO [16]. Exercise also reduces the incidence of a number of GI-related disorders such as colon cancer [17,18] and diverticular disease [19]. A study by Oliveira et al. suggested that physical Ex attenuates the inflammatory response of HF feeding [20], but whether this effect is mediated by activity-induced changes in the gut microbiota was not explored.
Few studies have specifically examined the effect of Ex on the gut microbiota. An early study by Matsumoto et al. examined the effect of chronic, load-bearing Ex on cecal microbiota in a nonobese rat model [21]. In the Matsumoto study, forced running altered cecal n-butyrate concentration as a consequence of a corresponding change in the n-butyrate producing bacteria compared to sedentary rats. More recently, Choi et al. examined the effect of Ex on polychlorinated biphenyl metabolism and the gut microbiota. They found that Ex altered the gut microbiota in normal chow-fed mice, predominantly affecting the families Erysipelotrichaceae and Enterococcaceae [22]. Whether Ex directly alters the gut microbiota, preventing HF diet-induced changes during development is not known.
In this study, we tested the hypothesis that voluntary Ex would reshape the gut microbiota in both LF and HF-fed mice and that Ex would normalize the gut microbiota in HF mice at the phylum level, in addition to mitigating the development of HF-DIO. Our results indicate that HF diets do indeed promote increased fatness, hyperglycemia and an altered gut microbiota profile compared to LF fed mice. Voluntary Ex prevented these changes in HF mice and resulted in restoration of the phylotypic profile observed with LF feeding. There was a significant correlation between the distance run and alterations in the major gut bacterial phyla for the Ex mice. In addition, differences were also observed at the subphylum level between the LF and HF groups with Ex. These findings may offer new insights into the beneficial effect of regular Ex in preventing DIO by modifying the bacterial balance of the gut microbiota.

Animals
All experimental protocols were approved by the Midwestern University Institutional Animal Care and Use Committee (IACUC) and all procedures were performed in accordance with Midwestern University's guidelines for the care and use of laboratory animals. Five week old, male C57BL/6 mice (n = 48) (Jackson Laboratories, Bar Harbor, ME, USA) were received as littermates (15-19 g) and allowed to acclimate in Midwestern University's animal facility for 1 week prior to their random assignment to one of 4 diet and activity treatment groups. During the acclimation period, mice were kept in their respective litters and provided standard chow (composition: 18 kcal% fat, 58 kcal% carbohydrate, 24 kcal% protein; 3.1 kcal/g total energy content, Teklad Diets, Madison WI) and reverse osmosis water ad libitum.

Diet and activity protocol
Following the acclimation phase, mice were randomly assigned to diet and activity groups based on equal distribution of littermates (i.e. mice in each litter were distributed as equally as possible among the 4 groups). Dietary intake included either high fat diet (HF; composition 60 kcal% fat, 20 kcal% carbohydrates, 20 kcal% protein; 5.24 kcal/g total energy content) or low fat diet (LF; composition 10 kcal% fat, 70 kcal% carbohydrates, 20 kcal% protein; 3.85 kcal/g of food, Research Diets, Inc.; New Brunswick, NJ). Mice were individually housed in rat cages in the absence (sedentary; Sed) or presence (exercise; Ex) of a non-load bearing 14.6 cm hamster wheel (PetSmart, Phoenix, AZ) suspended from a metal rod based on a design reported by Konhilas et al. [23]. The cages were modified to record wheel movement 24 h/day, 7 days/ week, using Schwinn 17-function odometers (Schwinn, Madison, WI.). The odometers were hand-tested for reliability of time of movement and distance covered. Exercise was recorded continuously (kilometers) and the cages were visually checked at least 3 times/week to ensure the odometers were working correctly and the wheels were functioning properly. Mice were observed exercising during the 12 hour light cycle, but were not disturbed during the dark cycle.
Littermates were separated into groups as follows: LF/Sed, n = 12, LF/Ex, n = 10, HF/Sed, n = 14, and HF/Ex, n = 12. Two separate batches of mice were used: an initial batch of mice were studied for preliminary indications of a global microbiota shift (terminal-restriction fragment length polymorphism (T-RFLP)) and measurement of gut transit (N = 24: LF/Sed, n = 6, LF/Ex, n = 4, HF/Sed, n = 8, and HF/Ex, n = 6) and a second batch for quantitative PCR (qPCR) and sequencing studies (N = 24; n = 6 all groups). All mice were followed on the protocol for 12 weeks for collection of fecal samples and then an additional two weeks for collection of blood samples and harvesting of tissues.

Weight and food intake measurements
Initial, weekly and final body weights were determined for all mice. Twenty-four hour food intake was measured at week 12 of the experimental period. For these measurements, mice were placed in cages without bedding and each mouse was given one pre-weighed pellet of their respective diet, either LF or HF food, in a ceramic dish designed to minimize food loss or contamination by feces. After 24 hours, the pellet and its remnants were weighed to determine food consumption. Food intake was expressed as kcal/ 24 hours.

Blood glucose measurements
Fasting blood glucose (FBG) and oral glucose tolerance (OGT) were measured at 13 weeks on the diet and activity program. Mice were fasted overnight in cages without bedding prior to measuring tail vein blood glucose levels using a OneTouch UltraMiniH monitor (LifeScan, Inc. Milpitas, CA). Oral glucose tolerance was measured immediately following the FBG test. Mice were gavaged with 25% glucose in saline (1 g/kg body weight) followed by blood glucose measurements at 15, 30, 60 and 120 minutes post glucose challenge [24] using the tail vein with a lancet stick according to the method of Golde et al. [25].

Tissue harvest and measurements
At the end of week 13 of the protocol, mice were anesthetized with CO 2 and then subjected to pneumothorax. Heart and soleus muscles were removed, rinsed several times, and blotted dry to determine their respective wet weights. Epididymal fat pads were excised, rinsed, blotted dry, weighed, and used as indicators of adipose tissue accumulation.

Intestinal transit
Measurement of intestinal transit was performed as a terminal experiment. Mice were fasted for 16 h with water available ad libitum. At 30 minutes prior to euthanasia, 0.1 ml of 0.5 mmol/L 70 kDa fluorescein isothiocyanate (FITC)-dextran (Sigma-Aldrich, St. Louis, MO) was administered by oral gavage [26,27]. After 30 minutes, mice were euthanized and the entire gastrointestinal tract, from stomach to anus, was removed. The small intestines were cut into 8 segments of equal length from the stomach to the cecum. The stomach and each segment were flushed with 3 ml of 50 mmol/l Tris buffered saline, the effluent was collected and centrifuged at 1200 rpm for 5 minutes to pellet any suspended debris. Fluorescent intensity in each sample was measured using a fluorescence spectrophotometer (emission wavelength = 518 nm; excitation wavelength = 492 nm; F-2000, Hitachi, Inc., Japan). Intestinal geometric center (IGC) was calculated as: (fraction of amount of FITC in each segment) 6 (segment number). The dextran concentration in each segment was expressed as a fraction of total dextran recovery.

Fecal pellet collection and DNA extraction
Extraction of DNA from fecal bacteria followed the protocol of Wang et al. [28]. Fecal pellets were obtained from all mice at the start (prior to beginning the protocol, week 0), mid-point (week 6), and at week 12 of the diet and Ex protocol. Fecal pellets were collected directly from the mice and stored at 280uC prior to lysis in 1 ml extraction buffer [50 mM Tris (pH 7.4), 100 mM EDTA (pH 8.0), 400 mM NaCl, 0.5% SDS] containing 20 ml proteinase K (20 mg/ml). Bacterial disruption was achieved using 0.1-mmdiameter zirconia/silica beads (BioSpec Products, Bartlesville, OK) and a Mini-Beadbeater-8 k Cell Disrupter (BioSpec Products). The homogenized samples were incubated overnight at 55uC with shaking. The beads were then spun down by centrifugation and 200 ml of room temperature saturated NaCl (175.3 g/500 ml) was added to 500 ml of the supernatant. After 2 minutes of vigorous shaking, the samples were spun for 15 minutes to pellet the protein and debris. Five hundred ml of phenol:chloroform:isoamylalcohol (25:24:1;4uC; Ambion) was added to 500 ml of the supernatant. Following shaking and centrifugation, 500 ml of the top phase was mixed with 500 ml of 100% ethanol (EtOH) and placed in 220uC for 30 minutes to precipitate all DNA. Tubes were spun at 4uC for 5 minutes to pellet the DNA, the pellet was washed with 70% EtOH, allowed to air dry at room temperature, and the DNA was dissolved in 50-100 ml of water and stored at 280uC. DNA concentrations and purity were then assessed by spectrophotometry using the NanoDropH (NanoDrop ND-2000 spectrophotometer; Thermo Scientific, Wilmington, DE). A minimum ratio of 1.9-2.0 was accepted for the 260:280 ratio.

Terminal Restriction Fragment Length Polymorphism
The T-RFLP analysis was performed as previously described [28]. Briefly, the 16S rRNA encoding gene was amplified from fecal sample DNA using a broad-range universal primer set: 8F (59-AGAGTT TGATCCTGGCTCAG-39) with a 69 carboxyfluorescein (6-FAM) label and 1492R (59-GGTTACCTTGT-TACGACTT-39). Each 50 ml PCR reaction contained 50 ng of DNA, 5.0 ml 106 Taq Buffer, 4.0 ml dNTP mixture (2.5 mM each), 0.25 ml TaKaRa Ex Taq (High fidelity Taq Enzyme, 5 units/ml; TaKaRa Mirus Bio, Madison, WI, USA), 1 ml forward primer and 1 ml reverse primer (10 pmol/ml each). PCR was run for 30 cycles using the thermal profiles of: 1 min at 94uC; 30 cycles of (94uC for 30 sec, 58uC for 1 min, 72uC for 1.5 minutes); and a final extension period of 6 min at 72uC. PCR products were verified by electrophoresis and visualization of a band at ,1400 bp. Upon confirmation, the remaining PCR products were purified and stored at 280uC.
Purified 16S rDNA was digested using 20 U MspI enzyme (New England Biolabs Inc.) in a total volume of 15 ml (including 1.5 ml NE Buffer 2, 1.5 ml enzyme, and 12 ml PCR product) at 37uC for 3 hours. To enhance the fluorescent signal, each digested T-RFLP sample was desalted by single drop analysis. Digested samples were desalted, mixed with deionized formamide and 0.5 ml GeneScan-500 size standard (Applied Biosystems) followed by capillary electrophoresis (Applied Biosystems DNA Sequencer 3130, Life Technologies, Grand Island, NY). Fragment length and abundance were determined using GeneMapper software (Applied Biosystems, Life Technologies, Grand Island, NY). Raw electropherograms were collected and analyzed for artifacts. To visualize relationships between the samples, sample distance values were analyzed by T-RFLP Analysis EXpedited (T-REX) and plotted using principal coordinates analysis (PCA) [29].

Quantitative PCR
The two gut bacterial phyla of interest, Bacteroidetes and Firmicutes, were targeted by modifying a quantitative PCR assay developed by Armougom et al. using primer and probe sets that recognized a conserved region of the 16S rRNA encoding gene unique to these taxonomic groups [30]. This assay was tested by Armougom et al. on a variety of Bacteroidetes and Firmicutes genera and found to have high sensitivity (.90% for all but one Bacteroidetes subgroup and .80% for all but one Firmicutes subgroup) and a low rate of false positives (0.01% for Bacteroidetes and 0.83% for Firmicutes genera). The efficiency of the assays was determined in the original publication [30]. A broad-range universal primer and probe set that recognize the conserved region of the 16S rRNA encoding gene for a wide variety of bacteria was used to normalize the assay to total bacterial DNA [31]. This primer-probe set was found to have high sensitivity to a broad range of lower taxonomic bacterial groups including members of both the Bacteroidetes and Firmicutes phyla by Nadkarni et al. [31].
The assay was performed on the ABI-PRISM 7700 Sequence Detection System (Applied Biosystems, Life Technologies, Grand Island, NY). The methods on which this assay was based have been published previously [30,31], but briefly the conditions for quantitative amplification for all qPCR runs were as follows: preamplification stage 95uC for 10 min and (40 cycles at 95uC615 seconds, 60uC61 minute) for each cycle. The reaction was performed in a 96 well optical grade plate (LifeTechnologies, Inc.) in a total volume of 25 mL using Environmental PCR Master Mix (TaqManH, LifeTechnologies). Wells contained 400 nM of the probe and 200 nM of the forward and reverse primers for each of the 3 assays and all runs were performed in triplicate (see Table S1 for primer and probe sequences). The amplicon size was 184 base pairs for Bacteroidetes, 179 for Firmicutes, and 466 for the universal set (see Figure S1 for gel showing PCR product size). Prior to qPCR, amplicon size for each target was verified by standard PCR followed by gel electrophoresis and visualization of bands consistent with the appropriate size (see Figure S1). The raw fluorescent signal was analyzed on Sequence Detection Software version 1.4 (Applied Biosystems, LifeTechnologies) and reported as Ct value. The universal Ct value for each respective sample was subtracted from the Ct values for Bacteroidetes and Firmicutes to calculate a DCt value for each sample, respectively.

Sequencing
Microbial sequencing was performed on the MiSeq Illumina platform (Argonne National Laboratory, Institute for Genomics and Systems Biology, Next Generation Sequencing Core) as previously described by Caporaso et al. [32,33]. Briefly, the V4 region of 16S bacterial rDNA was amplified using the custom degenerate primer pair 515F (GTGCCAGCMGCCGCGGTAA) and 806R (GGACTACHVGGGTWTCTAAT) to generate an amplicon size of 253 base pairs. Reverse primers included a 12 base pair nucleotide barcode to facilitate multiplex sequencing of all 72 samples (detailed in File S1) on a single 150 bp62 pairedend MiSeq run. The resulting 5,179,000 paired reads were joined using EA-Util's fastq-join script with default parameters, then screened to exclude sequences which contained one or more base calls with a Phred score less than 20.

Bioinformatics
Bioinformatic analysis of sequencing data was conducted using the QIIME 1.5.0 software suite [33]. Reads from all samples were clustered at 97% sequence identity into operational taxonomic units (OTUs) then aligned to the October 12th, 2012 Greengenes bacterial reference tree [34]. Following removal of OTUs composed of only one sequence, sequence sets for each of the 72 samples were rarefied to an equal sampling depth of 1,600 reads per sample. Unifrac distances between samples were calculated using both weighted and unweighted algorithms and visualized with PCA plots and dendrograms generated with Archeopteryx [35]. Preliminary screening of data for treatment effects was calculated using ANOVA, G Test, ADONIS, ANOSIM, MRPP, PermDISP, RDA, and Student's T-Test algorithms. Commands used in the analysis of this dataset are available in the (File S1). Treatment effects relative to the primary outcome were calculated for specific weeks and between weeks with a two-way and a repeated measure ANOVA, respectively.

Statistical Analysis
Quantitative data is represented as the mean 6 SEM, unless otherwise indicated. Differences between treatment groups were compared using an independent t-test (for running distance), a Pearson's r for correlation, a one-way ANOVA for qPCR data (on the DCt values) and a two-way ANOVA for all other data comparisons on Prism 5 (GraphPad, Inc.) and SPSS 18 (IBM, Inc.) software. Two-way ANOVA results were reported as the Fstatistics and p values and a Sidak post-hoc test was used for within group comparisons. To determine the significance of microbial community partitioning due to experimental treatments, permutational multivariate analysis of variance (PERMANOVA) was employed to generate p-values based on the observed Unifrac distance matrix relative to 10,000 randomly rearranged distance matrices. For display purposes only, the effect of exercise on the qPCR fold change in Bacteroidetes and Firmicutes in LF/Ex and HF/Ex mice is shown by the 2 2DDCt with the LF/Sed and HF/ Sed as the respective comparison groups. A value of p,0.05 was considered statistically significant and p,0.10 was considered to indicate a trend.

Characterization of Mouse Model of Voluntary Exercise and Diet-induced Obesity
Voluntary Exercise. Mice in the Ex groups (LF/Ex and HF/Ex) spontaneously ran on the cage-mounted wheel and averaged 8263.9 km/week and 8967.9 km/week for the LF/Ex and HF/Ex groups, respectively (p.0.05; Figure 1A). By the end of the protocol, mice accumulated an average total of 985646.5 km and 1076695.5 km for the LF/Ex and HF/Ex groups, respectively. Bilateral soleus muscle:body weight ratios and heart:body weight ratios were determined for all groups as indicators of an Ex training effect. Soleus muscle:body weight ratios were increased by Ex  Table 1).
Diet-induced Obesity. Figure 1B demonstrates the weekly body weight for each group. Mice were 5 weeks old at the start of the diet and Ex protocol (Time 0; Figure 1B), with comparable starting body weights (Table 1) Figure 1B). Consumption of a LF diet and increased physical activity both limited weight gain and showed a tendency for an interaction (F(1,20) = 3.43, p = 0.08).
Metabolically, the HF/Sed mice demonstrated elevated body fat and altered glucose metabolism. Body fat, as measured by epididymal fat pad weight (Table 1) Table 1).
Diet and Ex did not significantly influence gastrointestinal transit under fasting conditions. Based on the geometric center measurements reported in Table 1, gut transit time was not

Diet and Ex-induced Changes in Fecal Bacterial Content
Characterization of changes in the gut bacterial community as a consequence of diet and Ex was examined in three ways using bacterial 16S rRNA encoding genes present in the fecal pellets: terminal restriction fragment length polymorphism (T-RFLP), sequencing, and quantitative PCR (qPCR).

Terminal
Restriction Fragment Length Polymorphism. T-RFLP was used to determine whether there was a global shift in the gut microbial community after the 12 week diet and activity protocol using the 16S rRNA encoding genes present in the fecal pellets. Subjective analysis of raw electropherograms from T-RFLP analysis demonstrated visible changes in the abundance and distribution of community groups as a result of diet and activity ( Figure S2A). Principal coordinates analysis indicated some degree of clustering by litter at the start of the study ( Figure S2B), with obvious clustering by diet and activity by week 12 ( Figure S2C). Bacterial 16S rRNA Amplicon Sequencing. Closer analysis of bacterial differences induced by diet and Ex were determined by sequencing the 16S rRNA encoding genes present in the fecal pellets. Principal coordinates analysis of Illumina MiSeq amplicon data demonstrated significant clustering at week 0 based on litter (Figure 2A), which was no longer evident by week 12 ( Figure 2B). In contrast, there was no clustering of samples by diet and activity in week 0 ( Figure 2C), but robust clustering based on treatment group by week 12 ( Figure 2D). PERMANOVA analysis indicated a significant litter effect at week 0 (litter pseudo-F = 4.13, p,0.001), but no diet/activity effect (diet/activity pseudo-F = 0.53, p = 1.0) and only a trend toward a litter effect at week 12 (litter pseudo-F = 1.18, p = 0.064), but a significant diet/activity effect (diet/activity pseudo-F = 2.17, p,0.001). Dendrogram analysis ( Figure S3) corroborated the PCA findings, demonstrating clustering by intervention after 6 and 12 weeks on the diet and Ex protocol rather than by litter. The effect of Ex appears to be more distinct in the HF group based on a greater degree of separation of the HF/Ex from the HF/Sed group. The Shannon diversity index (beta diversity) for the samples at the start of the protocol were similar ( Figure 3A), but at week 12, both diet (F(1,20) = 39.22, p,0.001) and activity (F(1,20) = 7.8, p = 0.011) increased diversity of the community ( Figure 3B), with Ex exerting a greater effect in the HF group (interaction F(1,20 = 6.12, p = 0.022; p,0.001).
Six different bacterial phyla were identified in the MiSeq 16S rRNA based amplicon libraries at the start and completion of the diet and Ex protocol (Bacteroidetes, Firmicutes, Actinobacteria, Proteobacteria, Verrucomicrobia and Tenericutes). While there were no differences at baseline (data not shown), there were measureable differences in the percentage of these bacterial phyla among the 4 diet and activity groups by week 12 ( Figure S4). For the phylum Bacteroidetes ( Figure 4A Figure 4D). Tenericutes and Verrucomicrobia demonstrated no significant changes based on diet or Ex ( Figure S4). Comparison of the Bacteroidetes to Firmicutes content in fecal samples was also calculated using sequencing data ( Figure S5B). Comparable to the qPCR analysis, there was no demonstrable difference in the Bacteroidetes:Firmicutes ratio for any of the treatment groups at the start of the protocol. However, at 6 weeks, there was a significant reduction in the Bacteroidetes:Firmicutes ratio with HF diet (LF/Sed = 1. As a test of the criterion validity of qPCR, a comparison was made between the qPCR and sequencing data. The DCt Bacteroidetes:DCt Firmicutes ratio determined by q-PCR and %Bacteroidetes:%Firmicutes ratio from sequencing was compared using a Pearson test on all data points from weeks 6 and 12. As shown in Figure 5B, there was a significant inverse correlation between the Bacteroidetes:Firmicutes ratios using delta Ct values and sequencing (r 2 = 0.33, p,0.001). As expected, there was no relationship between the data points determined for week 0 (r 2 = 0.005, p = 0.81).
To validate the activity effect on bacterial balance, total Ex distance was correlated to the resultant DCt ratio of Bacteroidetes:Firmicutes determined using qPCR ( Figure 6). The ratio of delta Ct values for Bacteroidetes:Firmicutes correlated inversely with the Ex distance for the combined LF/Ex and HF/Ex groups    A third class of Firmicutes, Erysipelotrichi, was also altered by diet and activity ( Figure 7C). Within the class Erysipelotrichi, the percentage of the family Erysipelotrichaceae was reduced by both HF diet and Ex (diet F(1,20) = 17.15, p,0.001; activity F(1,20) = 6.17, p = 0.022), but Ex only exerted an effect in the LF fed mice (interaction F(1,20) = 5.64, p = 0.03; p = 0.016 LF/Sed greater than LF/Ex).
Differences in the percentage of the phylum Bacteroidetes based on the diet and activity protocol could be explained primarily by differences in one class (Bacteroidia) and one family (S24-7, Figure 8A). High fat diet lowered the percentage of S24-7 (F(1,20) = 5.89, p,0.025), but Ex increased it (F(1,20) = 5.63, p = 0.028) equally in LF and HF mice (F(1,20) = 0.204, p = 0.66). Differences in the percentage of the phylum Actinobacteria based on the diet and activity protocol could be explained by differences in only one class (Actinobacteria), one order (Bifidobacteriales) and one family (Bifidobacteriaceae, Figure 8B). Both HF diet and Ex lowered the percentage of Bifidobacteriaceae (diet F(1,20) = 6.45, p = 0.02; activity F(1,20) = 4.86, p = 0.039), with Ex affecting only the LF condition (interaction F(1,20) = 4.86, p = 0.039). Within the phylum Proteobacteria, a family in the Deltaproteobacteria class, Desulfovibrionaceae, demonstrated a trend toward a diet effect ( Figure 8C) (diet F(1,20) = 3.81, P = 0.065), but there was no effect of activity or interaction. Other changes resulting from the diet and Ex protocol occurring at the class, order and family levels that did not reach statistical significance are provided in the Table S2.

Discussion
This study examined the effect of voluntary Ex on the gut microbiota and the potential role of Ex-induced microbial changes in preventing HF-DIO in mice. Voluntary Ex, in and of itself, had significant effects on the relative balance of the major bacterial phyla, Bacteroidetes and Firmicutes, concurrent with prevention of DIO and normalization in glucose tolerance. In addition, Ex increased the Bacteroidetes:Firmicutes ratio in a manner that was proportional to the distance run. These results may suggest a possible microbial basis for Ex-mediated prevention of HF-DIO and may offer new insights regarding avenues to prevent DIO that could be useful to physically-challenged individuals who are not able to voluntarily Ex to manage their weight.
Balance between Bacteroidetes and Firmicutes has been reported to change with obesity as well as with the proportion of lean body mass [11,13]. Bacteroidetes in the gut microbiome is directly related to lean body mass: levels are reduced with obesity [8,13,36] and elevated when obese individuals lose weight [8]. While many rodent and human studies demonstrate an inverse relationship between obesity and the proportion of Bacteroidetes, there are reports in rats and humans that demonstrate an increase in the relative proportion of Bacteroidetes in response to a HF diet [37]. Our study supports that prevention of weight gain, in this case by Ex, is associated with a relative increase in Bacteroidetes. Alternately, fiber consumption [38] and the weight loss-promoting effects of mono-unsaturated fatty acids (oleic acid) and polyphenols are also associated with a rise in Bacteroidetes [39,40]. Two recent studies suggest that the richness and diversity of the gut microbiota inversely correlates with obesity [41,42]. Our results, in part, support this observation since Ex elevated the diversity of the gut microbiota, but only in the HF group.
The exact effect of HF consumption on the microbiome is likely dependent on the relative proportion of saturated, mono-and poly-unsaturated fat in the diet. In the current study, lard was the fat source with 32% saturated, 35.9% mono-and 32% polyunsaturated fats. Huang, et al. reported that diets rich in polyunsaturated fat (PUFA;74%) elevated levels of Firmicutes, Tenericutes, Proteobacteria, and Actinobacteria relative to the LF controls [43]. The PUFA diet also increased inflammatory gene activation in mesenteric fat compared to lard or milk fat containing diets. However, in IL-10 knockout mice, a line genetically susceptible to intestinal inflammation, milk fat (58% saturated fat) produced a bloom in the Proteobacteria, Bilophila wadsworthia, which resulted in a doubling of the incidence rate for spontaneous colitis [44]. Interestingly, the HF lard diet used in this study also induced an increase in Proteobacteria primarily resulting from an increase in the order Desulfovibrionales, to which B. wadworthia belongs. In the current study, Ex significantly increased the relative proportion of butyrate-producing bacteria such as Bacteroidales S24-7. High-fat sedentary mice had the lowest level of S24-7 family compared to the other treatment groups and the level was elevated by Ex. Exercise also led to an increase in butyrateproducing Firmicutes in fecal samples: Clostridiaceae, Lachnospiraceae and Ruminococcaceae. These bacteria are all members of Clostridiales order and are also linked to butyrate production [45]. Matsumoto et al. demonstrated that voluntary running significantly increased butyrate producing cecal bacteria (members of Clostridiales order) as well as elevated the level of cecal n-butyrate in non-obese rats [21]. Butyrate production in the large intestines is associated with intestinal epithelial cell health and the production of heat shock protein 70 (Hsp70) [46,47]. Hsp70 has been reported to maintain the functional and structural properties of intestinal epithelial cells in response to bacterial challenge, ischemia, and other environmental stresses, including Ex [47][48][49][50]. Previous reports indicated that epithelial cell integrity can be compromised by HF induced changes in barrier function, thus producing low grade systemic inflammation [51,52]. Butyrate stimulates epithelial cell Hsp70 production, which can provide structural and functional stability to cells under the stress of HF dietary intake.
Exercise also reduced the percentage of the family Lactobacillaceae in the HF mice, dropping it from greater than 20% in the HF/ Sed mice to 0.35% in the HF/Ex mice. This dramatic reduction with Ex suggests an important role for Lactobacillaceae in sedentary lifestyle-associated weight gain. Lactobacilli are often used as probiotic agents. Depending on the Lactobacillus species and experimental model, the reported relationship between detectable Lactobacillus spp and the effects of Lactobacillus on weight gain and obesity can be quite different. Lactobacillus spp are reported to be elevated in obese children [53]. In obese adults, Lactobacillus reuteri is elevated, but L. paracasei and L. plantarum are reported to be reduced [54]. Lactobacillus spp given as probiotic therapy have dose-dependent effects with regards to weight changes [55,56] whereas L. plantarum has been reported to improve glucose sensitivity in response to a HF diet [57,58].
The central questions raised by this study are, how does Ex alter the gut microbiota and does this contribute to the prevention of HF-DIO? Whether Ex operates to prevent HF-DIO through direct changes in the gut microbiota is an intriguing question because the answer could lead to new probiotics or prebiotics that provide for some of the benefits of Ex. The presence of a gut microbiota is known to help break down ingested food and likely allows for additional nutrient extraction not available to a germ free host [10,59]. Turnbaugh et al. [8] showed that a gut microbiota associated with obesity in ob/ob mice that was depleted in Bacteroidetes and enriched in Firmicutes was capable of extracting more energy from the diet compared to the microbiota from their lean counterparts. In particular, the presence of a gut microbiota has been shown to increase uptake of SCFAs in the colon [60,61].
Three factors determine the bacterial production of SCFA in the colon: the gut microbiota, availability of substrates, and intestinal transit time [60]. Based on our findings, changes in the gut microbiota with Ex could result in altered SCFA production and/or utilization. While gut transit time for substrates into and within the colon is also important [60], this study and that of Choi, et al. did not detect an effect of Ex on gut transit [22]. Several studies have shown that Ex is associated with increased gut motility [18,62,63]. We examined, primarily, small intestinal transit, and detected no change between the sedentary and Ex mice. If Ex had changed small intestinal transit, it could significantly enhance nutrient delivery to the cecum and colon, thereby changing microbial balance simply due to nutrient availability.
The amount of nutrients ingested could also impact bacterial content. In this study, Ex increased 24-hour caloric intake compared to Sed mice. Thus, microbial changes detected in the HF/Ex group beyond that of the HF/Sed could be related to even greater fat intake compared to HF/Sed or to an independent effect of Ex. A recent study by Simoes et al. demonstrated a change in the proportion of Bacteroidetes in monozygotic twins who were discordant with respect to diet [38]. They reported an inverse relationship between energy intake and the proportion of Bacteroidetes. This inverse relationship is interesting in light of the current results since in this study Ex increased both energy consumption and the proportion of Bacteroidetes. Acute bouts of exercise have also been shown to cause suppression of appetite during and immediately afterward with short-term negative energy balance, followed by increased appetite and positive energy balance [64,65]. These changes could lead to altered feeding patterns in Ex mice that could affect the gut microbiota.
An Ex-induced gut microbiota may increase fatty acid oxidation and utilization or decrease fat storage in the intestine. Exercise upregulates AMP-activated protein kinase (AMPK) [66], a molecular sensor of both cellular and systemic energy states. This pathway is also up-regulated in germ free (GF) mice and is believed to play a critical role in resistance to HF-DIO in this model [61]. Whether Ex only directly activates the AMPK system or also works through changes in the microbiota is still not clear, but activation of the AMPK system has been reported to be protective from hyperglycemia and type-2 diabetes and our findings related to improved OGT in HF/Ex mice compared to HF/Sed support this. We must also consider that Ex might act independently of the gut microbiota, by increasing energy utilization despite consumption of a HF diet. Another way in which Ex could exert a potentially beneficial effect through the gut microbiota is by suppressing inflammation. Exercise might prevent HF-DIO by blocking HF feedingassociated inflammation. Obesity and HF diets are linked to increased markers of endotoxemia and inflammation [67]. IL-6, a key myokine that is up-regulated by muscle contraction, suppresses TNF-a and the inflammatory response [68,69]. Exercise could regulate the gut microbiota through IL-6-mediated immunomodulation.
One of the limitations of this study is that bacterial changes were only measured in fecal samples. Differences in the composition of the gut microbiota have been shown as one moves from the epithelial surface out into the lumen of the intestine [70]. Different bacterial profiles may have been obtained if luminal scrapings or cecal contents had been examined. Another possible limitation is that while the correlation between the Bacteroidetes:Firmicutes ratios calculated using qPCR and sequencing is significant (p,0.001), it is a weak relationship (r 2 = 0.33). This may be due to differences in the two techniques, including the methods of quantification (cycle number versus database-matching), universal primer sets used, and the specificity and inclusiveness of sequencing compared to qPCR. However, it can be seen in the Figure S4A and S4B that qPCR and sequencing detected similar changes with development as well as dietary and activity interventions. This study did not identify bacteria below the taxonomic level of family. Changes at the genus and species level would be interesting in terms of comparisons to species with known metabolic or pathological functions. While food intake and gut transit were explored, energy extraction from the diet and changes in indicators of endotoxemia and inflammation were not included. These mechanisms are important to examine in terms of understanding whether the changes in the gut microbiota demonstrated with Ex are part of the anti-inflammatory, beneficial effect of Ex in preventing HF-DIO or whether they are coincidental with only a direct effect of Ex on energy balance and inflammation. Given the complex nature of the gut microbiota's interactions with the host's immune and metabolic systems, teasing out the contribution of Ex to these interactions will be difficult.

Summary/Conclusions
Exercise alters the gut microbiota in mice on both a LF and HF diet and normalizes major phylum-level changes for mice on the HF diet; furthermore, the volume of exercise (total distance run) inversely correlates with the Bacteroidetes:Firmicutes ratio. The Ex-induced gut microbiota in both LF and HF conditions are different than their sedentary counterparts and yet, also different from each other. At the taxonomic level of bacterial family, HF diet induced blooms in some of the major constituent groups within Firmicutes such as Lactobacillaceae, Lachnospiraceae, and Ruminococcaceae and a decrease in the major constituent group of Bacteroidetes, S24-7. Exercise in the HF condition prevented some of the dietary changes, such as with Lactobacillaceae and S24-7, but intensified the blooms seen with HF feeding in others such as with Lachnospiraceae, and Ruminococcaceae. Further exploration of the gut microbiota changes induced by Ex may allow for exploitation of this effect and the development of treatments for obesity and dysbiosis associated with high fat intake.  File S1 Commands used in the analysis of the sequencing dataset. (PDF)