Uncultured fecal gut microbiota from an underweight donor confers weight loss on gnotobiotic mice
We used anthropometric data collected from members of a birth cohort study (14) of 100 children living in Mirpur thana in Dhaka, Bangladesh, to define whether they were healthy or undernourished (table S1). Those with height-for-age z scores (HAZ) greater than or equal to −2 were classified as “healthy,” whereas those with scores less than or equal to −3 were deemed severely stunted. At 18 months, 30 and 25 children satisfied these criteria for healthy and severely stunted, respectively, whereas at 24 months, 27 and 20 children received these designations; the remaining children were classified as moderately stunted (HAZ between −2 and −3). A PCR-based screen for ETBF targeting all three fragilysin gene subtypes (14) was performed using DNA isolated from fecal samples that had been collected from these children at 18 and 24 months of age. The results revealed that ETBF was variably present between individuals and within a given individual over time, with a total of 25% of 18-month-old and 14% of 24-month-old children having a positive test (table S1). In this small cohort, ETBF carriage was not significantly correlated with indices of linear or ponderal growth [HAZ, weight-for-age z score (WAZ), and weight-for-height z score (WHZ) measured at 12 and 24 months of age (P = 0.8 and P = 0.4, P = 0.7 and P = 0.2, and P = 0.5 and P = 0.2, respectively; two-tailed Student’s t test)]. We combined anthropometric and PCR data to select fecal samples collected at 24 months from two children: (i) a healthy individual (child ID 7114 in table S1) with a HAZ score of −0.71, a WAZ score of −1.49, and a WHZ score of −1.62 who was ETBF-negative at the two time points tested, and (ii) a severely stunted and moderately underweight individual (child ID 7004) with a HAZ score of −3.02, a WAZ score of −2.51, and a WHZ score of −1.34 who was ETBF-positive at both time points. Of the 35 individuals with a positive ETBF test at either time point, only this stunted/underweight child was positive at both 18 and 24 months of age. Fecal samples obtained from members of this singleton birth cohort were screened for parasites using microscopic methods (5); neither of the two donors tested positive (see Materials and Methods for details).
To define the effects of diet and these two childrens’ gut microbiota on host biology, we generated three representative versions (embodiments) of the diets consumed by the population represented by the donors. To do so, we determined the relative daily caloric contributions of various selected ingredient types, based on a study by Arsenault and coworkers (16). Selection of specific food items as representative of each ingredient type was based on consumption incidence surveys tabulated by Islam et al. (17), and the results were incorporated into a database consisting of 54 food ingredients. We filtered this database to remove items consumed by <20% of households and categorized each of the remaining 39 items (see Materials and Methods for additional details). From the resulting diet ingredient matrix, we randomly sampled (without replacement) one item each from cereals, pulse vegetables, roots/tubers, leafy vegetables, fruits, and fish, plus three nonleafy vegetables, to populate three separate diet lists. Using the U.S. Department of Agriculture National Nutrient Database for Standard References (18), we determined the caloric information for each ingredient and subsequently calculated proportions required to match the predetermined contributions of each ingredient type. Food items were cooked in a manner intended to simulate Bangladeshi practices, and the resulting three embodiments of a Bangladeshi diet were sterilized by irradiation. This approach allowed us to generate several representative Bangladeshi diets that were not dominated by the idiosyncrasies of a single individual’s diet or by our own biases. The composition and results of nutritional analysis of the three diet embodiments are described in table S2 (A and B). The nutritional requirements of mice and children are compared in table S2C.
The results of a 12-year survey of demographic variations in the nutritional status of 16,278 Bangladeshi children found no significant sex differences in WHZ, WAZ, or HAZ (19). Therefore, in these and subsequent experiments, we eliminated gender as an experimental variable and only studied male mice. We gavaged separate groups of 8- to 9-week-old germfree C57BL/6 mice with the intact uncultured fecal microbiota samples obtained from the healthy or stunted/underweight Bangladeshi donors (two independent experiments; n = 4 singly caged mice per donor microbiota per experiment; see fig. S1A for study design). Fecal microbiota transplantation occurred 2 days after mice had been switched from an irradiated, nutritionally complete, low-fat/high-plant polysaccharide (LF/HPP) mouse chow that they had received since weaning to the first of the three embodiments of the Bangladeshi diet. Animals were subsequently fed, ad libitum, embodiment 1 for 1 week, followed by embodiment 2 for 1 week, and finally embodiment 3 for 1 week, with frequent sampling of their fecal microbiota during the course of each diet. Sequencing PCR amplicons generated from variable region 4 (V4) of bacterial 16S ribosomal RNA (rRNA) genes present in the donor’s fecal sample and in fecal samples collected over time from recipient gnotobiotic mice (table S3) provided an in vivo assay of colonization efficiency for each human donor sample. 16S rRNA sequencing reads were grouped into operational taxonomic units (OTUs) on the basis of a threshold of ≥97% nucleotide sequence identity (97% ID). The results revealed that at the conclusion of the experiment, 65.8 ± 2.5% (mean ± SEM) of OTUs in the stunted/underweight donor’s fecal microbiota sample and 68.4 ± 8.8% (mean ± SEM) of the OTUs in the healthy donor’s microbiota were detectable in recipient mice (that is, each OTU had a relative abundance of ≥0.1% in ≥1% of fecal samples obtained from the animals).
Although gnotobiotic animals colonized with the healthy donor’s intact uncultured fecal microbiota maintained weight, recipients of the severely stunted/underweight donor’s intact uncultured fecal microbiota exhibited progressive and significant weight loss (P < 0.005, paired two-tailed Student’s t test, comparison of final versus initial weights between the two treatment groups; Fig. 1A). In contrast to mice colonized with the healthy donor’s microbiota, those that received the stunted donor’s microbiota exhibited statistically significant weight loss at 10 days postgavage (dpg), during consumption of diet embodiment 2. Weight loss in this group worsened progressively, reaching 31 ± 6% (mean ± SEM) of original starting weight by 21 dpg (P < 0.001, two-tailed Student’s t test, comparison of final weights; Fig. 1A); in a linear mixed-effects model, both dpg and the interaction between microbiota and dpg were significant factors affecting weight throughout the experiment (P < 1 × 10−7 for each). Food consumption was not different between the two treatment groups as their weight phenotypes diverged. The relative abundance of B. fragilis, defined by V4-16S rRNA analysis of fecal samples obtained at the time of killing, was significantly greater in mice colonized with the stunted/underweight donor’s microbiota than in mice colonized with the healthy donor’s microbiota (P = 1.9 × 10−6, two-tailed Student’s t test; Fig. 1B).
Bacterial culture collections from donor fecal microbiota transmit contrasting weight phenotypes
We next cultured bacterial strains from the healthy and stunted/underweight donors’ fecal samples (20, 21). Each collection of cultured strains was clonally arrayed in multiwell plates so each well contained a monoculture of a given bacterial isolate (20). Each culture collection consisted of organisms that had coexisted in the donor’s gut and thus were the products of the donor’s history of environmental exposures to various microbial reservoirs (including those of family members and various enteropathogens endemic to the Mirpur thana), as well as the selective pressures and evolutionary events placed on and operating within their microbiota (for example, immune, antibiotic, dietary, and horizontal gene transfer). Individual isolates in the clonally arrayed culture collection were grouped into “strains” if they shared an overall level of nucleotide sequence identity of >96% across their assembled draft genomes (21). On the basis of this criterion and the results of sequencing amplicons generated from the isolates’ 16S rRNA genes, we determined that the healthy and stunted donors’ culture collections contained 53 and 37 strains, respectively. Only one strain was shared between the two culture collections: Bifidobacterium breve hVEW9 [see table S4 for a list of all isolates in the culture collection derived from the stunted/underweight child and (21) for details of the healthy donor’s culture collection]. The two B. fragilis strains present in the healthy donor’s culture collection (hVEW46 and hVEW47) lacked a BfPAI and were therefore classified as NTBF. The stunted donor’s collection contained a single B. fragilis strain (mVEW4) with a bft-3 allele. ETBF strains of this type are globally distributed but most common in Southeast Asia (22) (see table S5 for a comparison of the functions encoded by genes in the genomes of these ETBF and NTBF strains and the reference B. fragilis type strain ATCC 25285).
To ascertain whether the contrasting weight phenotypes conferred by the two intact uncultured fecal microbiota samples could be transmitted by the strains captured in their derivative culture collections, we colonized 8-week-old adult male germfree C57BL/6 mice with all members of either of these two culture collections (n = 6 singly caged mice per collection; all mice receiving a given culture collection were maintained in a single gnotobiotic isolator). As a reference control for this experiment, and to compare results between this and the previous experiment, we colonized mice with the corresponding intact uncultured fecal microbiota samples, housing these mice in separate isolators from those used for the culture collection transplants. All mice were fed three embodiments of the Bangladeshi diet (1 week per diet) in the same order described for the previous experiment. As with the intact uncultured microbiota, the corresponding culture collections transmitted discordant weight phenotypes to recipient animals (P < 0.002, two-tailed Student’s t test, comparison of final weights; Fig. 1C). Moreover, the weight phenotypes (change in body weight over time as a percentage of initial weight before gavage) observed with each intact uncultured fecal microbiota and the corresponding derivative culture collection were not significantly different (P > 0.05 for both microbiota donors, two-tailed Student’s t test; Fig. 1C). The difference in weight phenotypes first became statistically significant between the two groups of mice midway through consumption of diet embodiment 2, continued to increase with diet embodiment 3 (Fig. 1C), and again were not attributable to differences in food consumption.
Effect of diet.
To test whether the weight loss phenotype was sensitive or robust to diet embodiment type, we gavaged the two clonally arrayed bacterial culture collections into separate groups of 8-week-old adult male germfree C57BL/6 mice who were monotonously fed Bangladeshi diet embodiment 1, 2, or 3 for 3 weeks (n = 6 singly caged recipient mice per culture collection per diet embodiment; fig. S1B). The discordant weight phenotype observed previously was preserved irrespective of the Bangladeshi diet embodiment consumed (P < 0.01, two-tailed Student’s t test, comparison of final weights of mice regardless of diet embodiment consumed; n = 18 mice per culture collection; Fig. 1D). Moreover, no significant differences in weights were noted between groups of mice colonized with the same culture collection but fed different diet embodiments (P > 0.05 for embodiments 1 versus 2, 1 versus 3, and 2 versus 3; two-tailed Student’s t test, comparison of final weights; Fig. 1D).
Transmission of strains was assessed by short-read shotgun sequencing of DNA isolated from fecal samples collected at the end of the experiment. This method, known as community profiling by sequencing (COPRO-Seq) (21), maps reads onto the draft genome assemblies of community members. At the depth of sequencing used [354,352 ± 23,216 (mean ± SEM), 50-nucleotide (nt) unidirectional reads/fecal DNA sample], we could reliably detect strains whose relative abundance is ≥0.1%. COPRO-Seq demonstrated that transplantation of the culture collections was efficient and reproducible, with 98.1 ± 0.6% and 94.5 ± 1.6% (mean ± SEM) of strains in the collections derived from the healthy and stunted donors, respectively, appearing in recipient animals. The relative abundance of ETBF in the fecal microbiota of mice containing the stunted/underweight donor’s culture collection was significantly greater than the cumulative relative abundance of the two NTBF strains in recipients of the healthy culture collection irrespective of the diet embodiment consumed (75.5 ± 4.1% versus 17.0 ± 4.4%; P = 2.8 × 10−9, two-tailed Student’s t test; Fig. 2A). The relative abundances of the ETBF strain in recipients of the stunted/underweight donor’s culture collection, the two NTBF strains in the healthy donor’s collection, and all other Bacteroides species did not differ significantly between diet embodiments [P > 0.2 for all Bacteroides, one-way analysis of variance (ANOVA); Fig. 2B].
Intergenerational transmission of weight phenotypes.
To assess whether this weight loss phenotype was transmissible across generations of mice, two C57BL/6 male mice from the transplant experiment, one containing the stunted/underweight donor’s culture collection and the other containing the healthy donor’s collection, were switched to and subsequently maintained on an irradiated nutritionally enhanced mouse breeder chow from 21 to 48 dpg, at which time they were each cohoused with two germfree 6-week-old female mice that had received breeder chow since weaning. Seven days after cohousing, each male mouse was withdrawn from each mating trio, and the female mice were subsequently maintained on breeder chow throughout their pregnancy and as their pups completed the suckling period (fig. S1C). Male pups (n = 3 to 4 per litter) were then weaned onto an irradiated, nutritionally sufficient, LF/HPP chow, until they were 9 weeks old, at which time they were switched to the Bangladeshi diets (10 days per diet; same order of sequential presentation of the embodiments as before). Mice born to mothers colonized with either of these arrayed culture collections experienced identical weight gain profiles while consuming the LF/HPP diet (P = 0.9, two-tailed Student’s t test; table S6). However, once they were transitioned to the sequence of three Bangladeshi diet embodiments (consumed from postnatal days 56 to 86), mice born to mothers harboring a stunted/underweight donor’s microbial community exhibited significantly greater weight loss (P = 0.03, two-tailed Student’s t test comparing weights at killing). The total relative abundance of the two NTBF strains in fecal samples obtained from recipients of the healthy donor’s culture collection was 4.2 ± 0.7% at the conclusion of the LF/HPP diet period and 4.6 ± 0.9% at the conclusion of the Bangladeshi diet embodiment sequence, whereas the relative abundance of ETBF at these two time points was 34.3 ± 4.2% and 50.0 ± 0.7%, respectively, in mice colonized with the stunted/underweight donor’s culture collection.
An independent intergenerational transfer experiment was performed, in this case using the donors’ intact uncultured fecal microbiota. The efficiency of ETBF and NTBF transmission from mothers to pups was 100%. As with the culture collections, there was diet-dependent transmission of the discordant weight loss phenotype (Fig. 1E; compare with Fig. 1A).
Microbial community context determines the effects of ETBF on community members and host
To establish whether ETBF is necessary and sufficient to cause marked weight loss in multiple community contexts, we performed a series of manipulations that involved removing the ETBF strain from the stunted/underweight donor’s culture collection and adding it to the healthy donor’s culture collection, with or without subtraction of its two NTBF strains (Fig. 2C). These manipulations allowed us to characterize (i) the role of community context in determining ETBF pathogenicity, (ii) the community/host responses to ETBF, (iii) the ability of NTBF to modulate ETBF effects, and (iv) the effects of ETBF on NTBF. Recipient C57BL/6 male mice in each of the different treatment groups were 8 to 9 weeks old at the time of colonization; all were placed on diet embodiment 2 for 2 days before gavage and subsequently maintained on this diet for 14 days until they were killed (n = 5 singly caged animals per treatment group, maintained in separate gnotobiotic isolators). Fecal samples were collected at the time points described in fig. S1D.
Removal of the ETBF strain from the stunted/underweight donor’s culture collection prevented the transmissible weight loss phenotype (Fig. 2D; P = 5.9 × 10−8, two-tailed Student’s t test, comparison of weights at killing). However, addition of the ETBF strain to the healthy donor’s culture collection did not produce significant weight loss, regardless of whether the NTBF strains were present or absent (P = 0.3 and P = 0.2, respectively, two-tailed Student’s t test, comparison of weights at killing; Fig. 2D). On the basis of these findings, we concluded that whether ETBF produces weight loss (cachexia) is dependent on microbial community context.
COPRO-Seq analysis of the fecal microbiota of recipients of the unmanipulated ETBF(−) NTBF(+) healthy donor’s culture collection revealed that it contained the two NTBF strains [total relative abundance of 14.5 ± 3.0% (mean ± SEM), with B. fragilis hVEW46 and B. fragilis hVEW47 comprising 1.1 and 13.5%, respectively], two other Bacteroides (B. thetaiotaomicron and B. caccae), plus Bifidobacterium breve and Enterococcus. The relative abundance of B. fragilis was not significantly different between mice harboring the transplanted unmanipulated healthy donor’s culture collection and its two manipulated ETBF(+) NTBF(−) and ETBF(+) NTBF(+) versions (P > 0.5, two-tailed Student’s t test; Fig. 2A). (The term “unmanipulated” indicates that all bacterial isolates that comprise a culture collection were pooled before transplantation, whereas “manipulated” refers to the inclusion and/or exclusion of B. fragilis strains as part of the gavaged consortium.) The fecal microbiota of recipients of the unmanipulated stunted donor’s culture collection was dominated by ETBF (relative abundance, 62.3 ± 4.0%). Removal of ETBF led to significant increases in the relative abundances of B. breve, another Bifidobacterium strain, Enterococcus lactis, and Enterococcus gallinarum (P < 0.02, two-tailed Student’s t test; Fig. 2A).
To determine whether NTBF alone is sufficient to protect mice from ETBF’s cachectic effects, we colonized three groups of C57BL/6 male gnotobiotic mice, each with a different version of the stunted donor’s culture collection: the unmanipulated culture collection containing ETBF alone or one of two manipulated versions, one with NTBF alone, and the other with both ETBF and NTBF strains. Mice were placed on diet embodiment 2 for 2 days before gavage and maintained on this diet for 2 weeks until killed (n = 6 animals per treatment group, all singly caged; one treatment group per gnotobiotic isolator; fig. S1D). We observed a significant difference in weight phenotypes between mice colonized with the unmanipulated undernourished donor’s ETBF(+) NTBF(−) culture collection compared to the manipulated ETBF(−) NTBF(+) version (P = 0.01, one-tailed Student’s t test; Fig. 2E). Addition of NTBF [yielding the ETBF(+) NTBF(+) community] markedly ameliorated the weight loss phenotype (P = 0.0004 for weights at killing compared to mice with the unmanipulated community, one-tailed Student’s t test; Fig. 2E). Follow-up COPRO-Seq analysis revealed that the relative abundances of ETBF at the conclusion of the experiment were 38.9 ± 3.9% and 39.0 ± 3.5% when animals were colonized with and without NTBF, respectively. Thus, NTBF does not appear to mediate its effects by reducing the fractional representation of ETBF in the community. However, ETBF appears to reduce the relative abundance of NTBF, which constituted 41.8 ± 3.2% of the total community when ETBF was absent but only 19.2 ± 2.6% when ETBF was present (P = 0.04, one-tailed Student’s t test).
The effects of intraspecific interactions on microbial gene expression.
We performed microbial RNA sequencing (RNA-seq) of cecal contents harvested at killing to characterize the transcriptomes of members of the unmanipulated and manipulated versions of the healthy and stunted communities. Our goal was to assess (i) the effects of intraspecific competition (NTBF on ETBF and vice versa) in the healthy and stunted community contexts, (ii) the effects of the cultured stunted/underweight versus healthy donor community on ETBF, and (iii) the effects of cocolonization with ETBF on other bacterial members (including other Bacteroides). ETBF genes with significant differential expression attributable to the presence or absence of NTBF, in both healthy and stunted community contexts, are listed in table S8 (B and F). Conversely, NTBF genes with significant differential expression attributable to the presence or absence of ETBF, in both healthy and stunted community contexts, are highlighted in table S8 (C and E).
Fragipain is a cysteine protease that activates fragilysin by removing its autoinhibitory prodomain. In mouse models of colitis, host proteases can also serve this function, but fragipain is required for sepsis to occur (23, 24). In the presence of NTBF, ETBF expression of fragilysin (bft-3) in the cecal metatranscriptome of mice harboring the manipulated ETBF(+) NTBF(+) healthy donor’s community was significantly decreased compared to the manipulated version of the community where ETBF, but not NTBF, was present (39-fold, based on normalized transcript counts; P = 0.002, one-tailed Student’s t test). Fragipain expression was also significantly reduced (14.2-fold; P = 0.0005, one-tailed Student’s t test) (table S8B). In the context of the stunted community, the reduction in bft-3 expression associated with introducing NTBF was considerably more modest (5.9-fold; P = 0.09, one-tailed Student’s t test), whereas fragipain expression was not significantly different between the two treatment groups (P > 0.5, one-tailed Student’s t test; table S8).
When we abrogated fragilysin (bft-3) expression through insertional mutagenesis (fig. S2), the mutant Δbft-3 strain grew robustly in vitro. However, when germfree mice were gavaged with a manipulated version of the stunted donor’s culture collection containing this isogenic strain with a disrupted bft-3 locus substituted for the wild-type ETBF strain, we observed no detectable colonization of the mutant (n = 5 mice fed diet embodiment 2 for 14 days); the number of COPRO-Seq reads mapping to the mutant Δbft-3 strain was no greater than background, and a PCR assay that used B. fragilis–specific bft primers was negative. However, these results led us to conclude that this locus functions as an important colonization factor for this particular ETBF strain in this community context. However, these experiments did not allow us to directly address the hypothesis that attenuation of bft-3 expression produced by inclusion of NTBF in the stunted community contributed to the observed mitigation of weight loss.
Looking beyond the effects of intraspecific interactions on btf-3 expression, we compared the cecal metatranscriptomes of gnotobiotic mice colonized with the unmanipulated NTBF(+) ETBF(−) healthy donor’s culture collection versus mice harboring the two manipulated versions where ETBF was added, with or without removal of the two NTBF strains. The results revealed that ETBF in the absence of NTBF produced significant alterations in the expression of a number of transcripts related to various features of stress responses in several community members [Enterococcus faecalis, E. gallinarum, B. breve, and two members of Enterobacteriaceae; differentially expressed genes identified using the Robinson and Smyth exact negative binomial test (25), with Bonferroni correction for multiple hypotheses] (Fig. 3). Both rpoS, which is a key general stress response sigma factor that positively controls expression of genes involved in transport of carbon sources and iron acquisition, and recD, which is involved in DNA repair, exhibited significant increases in their expression in the setting of ETBF without NTBF (P < 0.05). Several genes involved in the acquisition and metabolism of iron were either up-regulated in the presence of ETBF (for example, ferric aerobactin receptor, ferric uptake regulation protein, and aerobactin synthase) or repressed (for example, an Enterobacteriaceae strain hVEW34 homolog of the Escherichia coli BasSR system component BasS, which is normally induced under high-iron conditions) (26). ETBF’s effect on expression of these latter genes was mitigated when NTBF was present (Fig. 3), highlighting the importance of iron in intraspecific and interspecific interactions in the healthy donor’s consortium of transplanted cultured bacterial strains. In contrast, the presence or absence of ETBF or NTBF did not evoke significant changes in the expression of these or other genes involved in iron metabolism in the context of the stunted/underweight donor’s community. Numerous genes related to prophage and mobile DNA element biology were also expressed at significantly higher levels by healthy community members when ETBF was present in the absence of NTBF (P < 0.05; Fig. 3). Prophage activation occurs in response to stress. Some studies have postulated that phage induction can “shuffle” community structure to favor an increased proportion of pathobionts (27).
Studies in gnotobiotic mice have shown that signaling by members of the human gut microbiota involving the quorum sensing molecule, autoinducer-2 (AI-2), can alter virulence factor expression in enteropathogens (28) and have linked AI-2 signaling to modulation of the levels of Bacteroidetes in the gut (29). LuxQ is involved in the detection of AI-2. In the context of the healthy community, expression of three of the four luxQ homologs in the ETBF genome was decreased when NTBF was present [log2(fold change) of −2.8, −4.4, and −9.5, P < 0.005, exact negative binomial test; table S8B]. Comparing the mice colonized with the unmanipulated ETBF(−) NTBF(+) and manipulated ETBF(+) NTBF(+) versions of the healthy donor’s culture collection revealed differential regulation of five other luxQ transcripts encoded by Bacteroides members (three in B. thetaiotaomicron hVEW3, and two in B. caccae hVEW51; table S8D). In the context of the stunted donor’s community, the presence of ETBF had no significant effects on lux gene expression in NTBF or any other community members, nor did the presence of NTBF have any effect on lux expression in ETBF (table S8, E and F). Together, these results illustrate the importance of community context in determining the transcriptional effects of intraspecific (and interspecific) interactions involving ETBF.
The metabolic effects of manipulating the representation of ETBF and NTBF in the healthy and stunted donor’s communities were studied by targeted mass spectrometry (MS) of tissue samples obtained from mice in the fed state (table S7). Quantifying amino acids, organic acids, acylcarnitines, and acyl-CoAs in livers obtained from animals colonized with either of the two unmanipulated culture collections disclosed that compared to mice harboring the healthy donor’s ETBF(−) NTBF(+) culture collection, those colonized with the stunted donor’s ETBF(+) NTBF(−) culture collection had higher concentrations of propionyl-CoA and isovaleryl-CoA [by-products of oxidation of branched-chain and other amino acids; P < 0.05, false discovery rate (FDR)–adjusted two-tailed Student’s t test; Fig. 4A], and lower concentrations of acetyl-CoA (P = 0.07) and its cognate metabolite acetyl carnitine (that is, C2 acylcarnitine; P = 0.001; Fig. 4B). Mirroring these trends, cecal contents harvested at the time of killing from mice harboring the stunted donor’s unmanipulated culture collection contained higher concentrations of branched-chain amino acids (P = 0.066 for isoleucine/leucine and P = 0.1 for valine) and lower concentrations of acetyl-carnitine (P = 0.067). However, these trends were not observed in skeletal muscle.