The nonfermentable dietary fiber hydroxypropyl methylcellulose modulates intestinal microbiota

Abstract

Diet influences host metabolism and intestinal microbiota; however, detailed understanding of this tripartite interaction is limited. To determine whether the nonfermentable fiber hydroxypropyl methylcellulose (HPMC) could alter the intestinal microbiota and whether such changes correlated with metabolic improvements, C57B/L6 mice were normalized to a high-fat diet (HFD), then either maintained on HFD (control), or switched to HFD supplemented with 10% HPMC, or a low-fat diet (LFD). Compared to control treatment, both LFD and HPMC reduced weight gain (11.8 and 5.7 g, respectively), plasma cholesterol (23.1 and 19.6%), and liver triglycerides (73.1 and 44.6%), and, as revealed by 454-pyrosequencing of the microbial 16S rRNA gene, decreased microbial α-diversity and differentially altered intestinal microbiota. Both LFD and HPMC increased intestinal Erysipelotrichaceae (7.3- and 12.4-fold) and decreased Lachnospiraceae (2.0- and 2.7-fold), while only HPMC increased Peptostreptococcaceae (3.4-fold) and decreased Ruminococcaceae (2.7-fold). Specific microorganisms were directly linked with weight change and metabolic parameters in HPMC and HFD mice, but not in LFD mice, indicating that the intestinal microbiota may play differing roles during the two dietary modulations. This work indicates that HPMC is a potential prebiotic fiber that influences intestinal microbiota and improves host metabolism.—Cox, L. M., Cho, I., Young, S. A., Kerr Anderson, W. H., Waters, B. J., Hung, S.-C., Gao, Z., Mahana, D., Bihan, M., Alekseyenko, A. V., Methé, B. A., Blaser, M. J. The nonfermentable dietary fiber hydroxypropyl methylcellulose modulates intestinal microbiota.

The gastrointestinal microbiome is composed of trillions of organisms (1, 2) in any mammalian host, and is largely affected by a combination of host genetic factors and diet (3, 4). The intestinal microbiota is involved in many physiological processes (5), highlighting the need for methods to modulate the composition of the organisms present. Strategies to shape the microbiome include antibiotics to reduce specific populations, transfer of live organisms in either foods containing probiotics or via fecal transplant, or by delivering prebiotic compounds that stimulate the growth of specific microorganisms (6). Prebiotics have the advantage of being easy to transport and administer, and have few safety concerns. However, to date; the focus has been on compounds that are actively metabolized by members of the microbiota.

Obesity and the metabolic syndrome are increasing problems in the United States and worldwide (7, 8). The roots of these disorders often are multifactorial (9), and many potential treatments have emerged, including dietary and lifestyle modifications. Reducing dietary calorie and fat intake leads to weight loss and improved metabolic parameters by directly altering energy balance (10). Adding indigestible fiber to diets has long been associated with health benefits in both humans and animals, including effects on weight control (11, 12), metabolic syndrome (13, 14), blood cholesterol (15), and glucose tolerance (16–19); however, the mechanisms of action are not well understood.

Recently, much attention has focused on the role of intestinal microbiota in the regulation of host metabolism as a significant contributing factor to obesity (20–27). The microbiota perform a diverse range of metabolic functions, including short-chain fatty acid production from carbohydrate fermentation (28, 29), regulation of energy extraction from food (30, 31), and bile salt metabolism (32). Microbial interactions with dietary fiber, and consequently the mechanisms by which fiber-induced microbiome changes affect host metabolism, partially depends on the extent that the fiber can be fermented. However, fibers ranging from completely fermentable to nonfermentable (33), have been reported to improve host metabolism (34). Health improvements are usually attributed to an increase in short-chain fatty acid production for fermentable fibers (35, 36), but a similar conclusion cannot be drawn for nonfermentable fibers.

Hydroxypropyl methylcellulose (HPMC) is a nonfermentable dietary fiber with many beneficial health effects, making it an ideal substrate to probe the interaction between fiber, metabolism, and microbiota in a system in which the treatment was previously believed to be inert with respect to the microbiota (37). It is a semisynthetic cellulose derivative with hydroxypropyl and methyl side chains, used in the manufacturing of many foods (38). HPMC has a long safety record and is classified as having “generally recognized as safe (GRAS)” status, up to 20 g/d in the United States (39). HPMC supplementation lowers cholesterol and postprandial insulin levels, in both rodents (15, 40, 41) and humans (16, 42, 43), and decreases weight gain (44) and fat mass (45) in rodents fed a high-fat diet (HFD). HPMC is not absorbed by the host (46); however, it can modulate the intestinal nutrient environment by selectively increasing the excretion of fecal bile acids and fats (44) and increasing fecal water content (47), and thus have downstream effects on the intestinal microbiota.

In this study, we used high-throughput 16S rRNA taxonomic profiling to examine the effects of HPMC on the murine microbiome. We sought to determine whether HPMC and a low-fat diet (LFD), which yielded parallel metabolic effects, would impose similar changes on the intestinal microbiome. Our goal was to characterize overall changes in microbial community structure in response to dietary perturbation, and to identify specific taxa associated with particular diets or metabolic phenotypes.

MATERIALS AND METHODS

Animals and diets

Eighteen-week-old adult C57B6/L6J mice from Jackson Laboratories (Bar Harbor, ME, USA) were acclimated to an HFD (5.24 kcal/g, 60% kcal from fat; D12492; Research Diets, New Brunswick, NJ, USA) for 2 mo, then weighed and randomized into 3 groups of 10 mice: LFD (3.85 kcal/g, 10% kcal from fat; D12450B; Research Diets), continuing HFD, and HFD (60% kcal from fat) supplemented with 10% (weight percentage of the diet) of HPMC (4.71 kcal/g, K250M lot no. VJ241907R1; Dow Chemical, Midland, MI, USA). The mice were housed individually in polycarbonate cages maintained with a 12-h alternating light-dark cycle at normal temperature (22±4°C) with relative humidity 50 ± 15%. Total food intake (g) was measured for the duration of the 4-wk experiment, and total energy intake (kcal) was calculated by multiplying total grams of food consumed by the energy (kcal/g) in each diet. This study was performed according to an Institutional Animal Care and Use Committee-approved protocol (48) and conducted by In Vivo Services at the Jackson Laboratory West (Sacramento, CA, USA), an Office of Laboratory Animal Welfare–assured and Association for Assessment and Accreditation of Laboratory Animal Care–accredited organization.

Hepatic lipid analysis

Lyophilized liver samples were extracted using an accelerated solvent extractor (Dionex ASE; Dionex, Sunnyvale, CA, USA) at 100°C, ∼13.8 MPa with 75/25 hexane/2-propanol, dried, and weighed to determine the percentage of total hepatic lipids, and hepatic total cholesterol, free cholesterol, and triglyceride levels using colorimetric assays and a clinical analyzer (Roche Diagnostics/Hitachi 914; Roche Diagnostics, Indianapolis, IN, USA).

Fecal lipid analysis

Fecal lipids were extracted on a Dionex ASE system using a mixture of hexane and 2-propanol (3:2, v/v, 2% acetic acid) at 15 MPa and 60°C for 30 min. One aliquot was analyzed for saturated and unsaturated fatty acid composition by gas chromatography as described previously (44). The second aliquot was analyzed for total bile acids and sterols using a modified chromatographic method (49).

Bomb calorimetry

C57BL/6J mice were normalized to an HFD (60% kcal from fat) for 2 mo, then either maintained on an HFD or changed to an HFD supplemented with 8% HPMC. Fecal pellets were collected at baseline, prior to the change in diets, and 25 d after the dietary intervention. Samples were oven-dried at 55°C for 24 h. Total energy in the fecal pellets per gram was measured by bomb calorimetry using a Parr 6725 Semimicro calorimeter, 1109A semimicro oxygen bomb, and 6772 thermometer (Parr Instrument Co., Moline, IL, USA). The calorimetry energy equivalent factor was determined using benzoic acid standards, which showed 99.36–99.92% reproducibility.

Plasma biomarker analysis

Total cholesterol, free cholesterol, and triglycerides in plasma were determined by enzymatic colorimetric assays using a Roche Diagnostics/Hitachi 914 clinical analyzer as described previously (44). The VLDL-cholesterol levels were calculated by subtracting HDL-cholesterol and LDL-cholesterol from total cholesterol levels. Fasting plasma concentrations of adiponectin, leptin, and insulin of mice were determined after 12 h of food withdrawal using mouse adiponectin (B-Bridge International, Sunnyvale, CA, USA), leptin (Assay Designs, Ann Arbor, MI, USA), and insulin (Mercodia, Winston Salem, NC, USA) immunoassay kits, as described previously (41). Fasting glucose levels were measured from a drop of blood collected by tail vein puncture and analyzed using a OneTouch Ultra meter with FastDraw test strips (Johnson & Johnson, Milpitas, CA, USA).

Microbiome sample collection and processing

Fecal pellets were collected at baseline, 2 wk, and 4 wk, and cecal and ileal samples at euthanasia were frozen at −80°C until DNA extraction. DNA extraction was performed using the MoBio PowerSoil DNA Isolation Kit (MoBio Laboratories, Carlsbad, CA, USA).

Quantitative PCR

Quantitative PCR assays were performed on a Rotor Gene 3000 quantitative PCR cycler using the Power SYBR Green kit (Applied Biosystems, Carlsbad, CA, USA), according to the manufacturer's instructions, with the addition of 1 ng/μl of BSA. Total bacterial levels were determined by standardized quantitative PCR (qPCR) targeting universal regions of the 16S rRNA gene using primers Eub519F and Eub785R (50). Firmicutes and Bacteroidetes levels were determined using the primer pairs Firm934F/Firm1060R and Bact934F/Bact1060R, respectively (51), and were normalized on the basis of total bacteria 16S copies.

PCR amplification for 454-pyrosequencing

Samples were prepared for amplification and sequencing at the J. Craig Venter Institute (JCVI) Joint Technology Center (JTC). Genomic DNA sample concentrations were normalized to ∼2-6 ng/μl. The V3–V5 region of the 16S rRNA gene was amplified using forward primer 341F (5′-CCTACGGGAGGCAGCAG-3′) attached to the Roche B adapter for 454-library construction and reverse primer 926R (5′-CCGTCAATTCMTTTRAGT-3′) attached to the Roche A adapter and a 10-nt barcode [5′-A-adapter-N (10)+16S primer-3′]. A barcoded primer design was completed using a set of algorithms developed at the JCVI. Every effort was made to prevent contamination of PCR reactions with exogenous DNA, including a set of reactions in a laminar flow hood. PCR reactions were completed as follows (per reaction): 2 μl of gDNA, 1× final concentration of Accuprime PCR Buffer II (Invitrogen, Carlsbad, CA, USA), 200 nM forward and reverse primers, 0.75 U of Accuprime TaqDNA polymerase high fidelity (Invitrogen), and nuclease-free water to bring the final volume to 20 μl. PCR cycling conditions consisted of an initial denaturation of 2 min at 95°C, 30 cycles of 20 s at 95°C, 30 s at 50°C, and 5 min at 72°C. A negative control (water blank) reaction was examined after 35 cycles. Samples were then quantified, cleaned, and sequenced on the Roche 454-FLX (454 Life Sciences, Branford, CT, USA) as described previously (27), and a read processing pipeline consisting of a set of modular scripts designed at the JCVI were employed for deconvolution, trimming, and quality filtering, as described previously (27). The number of reads that passed the quality filter totaled 676,440 sequences (mean 4602±1625 reads/sample). The sequencing data were denoised with shhh.seqs, a denoising algorithm developed by Chris Quince (SeqNoise; http://dev.man-online.org/man1/SeqNoise/) as implemented in Mothur v 1.25.0 (52), with the following settings: minimum flow = 360, primer difference = 6, and barcode difference = 1.

Statistical and bioinformatics analysis

Quality-filtered sequences were further preprocessed through the Qiime pipeline (53) as described previously (27). The rarefactions for richness and Shannon diversity indices were calculated in the R statistical programming environment (54, 55) using Community Ecology Package vegan, on the denoised reads. Principal coordinates analysis (PCoA) plots from Qiime output were produced based on unweighted UniFrac (56) distances by Kinemage, Next Generation (KiNG 2.16; http://kinemage.biochem.duke.edu/software/king.php). Comparison of unweighted and weighted UniFrac distances was performed using 2-sided t tests. The operational taxonomic unit (OTU) relative abundances were calculated by dividing the absolute abundances by total sequence count per sample analyzed. The resulting relative abundance matrix was used to produce heat maps for major (relative abundance≥1%) taxa. Hierarchical clustering of major taxa from the resulting dense abundance matrices was performed using euclidean distances. Association of major clusters and treatment group was measured using Fisher's exact test at 0.05 significance threshold. Differences between dietary groups were evaluated using the Mann-Whitney U test. When multiple testing was performed, the P values were adjusted for false discovery using the Benjamini-Hochberg procedure (57).

RESULTS

Dietary interventions improve metabolic phenotypes

All C57BL/6J mice were fed an HFD for 2 mo prior to the study and then randomized to either continue the HFD, to receive the HFD with 10% HPMC supplementation (HPMC diet), or to receive an LFD. Over the course of the 4-wk study, the mice fed the HFD continued to gain weight, while HPMC mice showed reduced weight gain (Fig. 1A), despite isocaloric intake of food (Fig. 1B). Mice receiving the LFD lost weight and had significantly lower energy intake. Weight change and energy intake were correlated in both groups of HFD and LFD mice (Pearson correlation coefficient, r=0.84 and 0.65, P=0.003 and 0.042, respectively); when combined, the correlation strengthened (Pearson correlation coefficient, r=0.98, P<0.0001; Fig. 1C). No relationship was detected between energy intake and weight change in HPMC mice (P=0.39, for a nonzero slope).

Figure 1.

Effect of diet on host metabolism. Adult C57BL/6 mice were fed an HFD (60% kcal from fat) for 2 mo prior to the study, then continued on HFD, or switched to either an LFD (10% kcal from fat) or to HFD with 10% HPMC supplementation (HPMC group) (10 mice/group). A) Weight change over the 4-wk study. B) Total 4-wk energy intake (kcal; bar at median). C) Correlation by linear regression between 4-wk weight change and total energy intake. D–H) Biomarkers at 4 wk: fasting plasma cholesterol (D), liver triglycerides (E), fecal saturated fat (F), fecal unsaturated fat (G), fecal bile acids (H). I) Total energy content per gram in fecal pellets in C57BL/6J mice normalized to an HFD (60% kcal from fat) for 2 mo, then either maintained on HFD (n=5) or changed to HFD supplemented with 8% HPMC (n=5). Values are shown at baseline (d 0) before randomization into the two treatment groups and at 25 d after dietary intervention. *P < 0.05, **P < 0.01, ***P < 0.001; FDR-adjusted Mann-Whitney U (B, E–I); nonzero slope (C); Mann-Whitney U (D).

After 4 wk, mice fed the LFD and HPMC diets had significantly reduced total cholesterol, HDL, LDL, and VLDL cholesterol, leptin, liver triglycerides, and liver percentage adiposity, compared to the HFD mice (Fig. 1D, E and Supplemental Fig. S1A–C, H, J). The LFD mice had decreased fasting blood glucose, and free fatty acids (Supplemental Fig. S1D, F), while plasma insulin was decreased in the HPMC mice (Supplemental Fig. S1G). No changes were seen in serum triglycerides or adiponectin levels (Supplemental Fig. S1E, I). Compared to both the LFD and HFD, HPMC supplementation increased saturated, unsaturated, and trans-unsaturated fecal fat, as well as fecal monoacylglycerides/free fatty acids, and bile acids (Fig. 1F–H and Supplemental Fig. S1K, M). The levels of sterol excretion decreased in the HPMC group compared to the HFD group (Supplemental Fig. S1L), while diacylglycerides or triacylglycerides were unchanged (Supplemental Fig. S1M–O). HPMC-group mice had increased fecal energy, as measured by bomb calorimetry (Fig. 1I).

Microbial diversity and community structure

To assess changes in murine intestinal microbial community structure induced by the dietary changes, measures of richness and evenness were calculated for microbial 16S rRNA sequences from variable regions V3–V5 in fecal and cecal samples. At the OTU level, the LFD and HPMC mice had decreased fecal community richness compared to the HFD mice, and the HPMC mice had reduced cecal microbial richness (Fig. 2A). At the class level, richness in the fecal and cecal communities from the three groups was similarly maintained (Supplemental Fig. S2A). Community evenness, measured by the Shannon evenness metric, decreased at the OTU level in the fecal samples from LFD and HPMC mice, and similarly in cecal samples for the HPMC mice (Fig. 2B). However, Shannon evenness indices increased in the same groups at the class level (Supplemental Fig. S2B).

Figure 2.

Assessment of microbial diversity in relation to treatments. Graphs depict α diversity at the OTU level, calculated on denoised sequences of C57B/L6J mouse fecal microbiota at wk 0, 2, and 4, and cecal microbiota at wk 4. Mice were normalized on HFD, then maintained on HFD, or switched to either LFD or HPMC; 10 mice/group. Rarefaction curves display the α diversity when subsampled at a lower depth. A) Population evenness calculated by the Shannon evenness metric. B) Taxonomic richness as measured by number of observed OTUs.

The phylogenetic differences within the intestinal microbial ecosystem between the three treatment groups also were assessed by PCoA of the unweighted UniFrac (58) distances, and 24.5% of the total variation was explained on the first 3 PCoA axes (Fig. 3). As expected, all samples cluster together at baseline (wk 0; Fig. 3A). At wk 2 (Fig. 3B) and wk 4 (Fig. 3C), the HFD samples were little different from baseline, whereas the LFD and HPMC communities had shifted in separate directions; HPMC samples formed a distinct cluster, while LFD samples were more variable. Pairwise unweighted UniFrac distances were calculated for all 10 mice within each dietary group, comparing baseline, wk 2, and wk 4 fecal microbiota to either baseline (Fig. 3D) or wk 2 (Fig. 3E) microbiota. There were no significant changes in the mean UniFrac distance over time for the HFD fecal microbiota, as might have been anticipated. However, for the LFD samples, pairwise distances significantly increased from wk 0 to 2, but not from wk 2 to 4. For the HPMC fecal microbiota, pairwise distances significantly increased from wk 0 to 2, and from wk 2 to 4. Analysis of the weighted UniFrac distances showed similar trends (Supplemental Fig. S2D, E). The community structure of the cecal samples was similar to the wk 4 fecal samples for HFD and HPMC samples; however, LFD cecal samples differed from the LFD fecal samples and were intermixed with HFD samples (Fig. 3F). The ileal specimens showed no distinctive clustering by treatment group (Fig. 3G).

Figure 3.

Effect of diet and fiber on microbial community structure. PCoA of the unweighted UniFrac distances of microbial 16S rRNA sequences from the V3–V5 region in fecal samples at wk 0 (baseline; A), wk 2 (B), and wk 4 (C), cecal samples at euthanasia (F), and ileal samples at euthanasia (G) in mice maintained on HFD, switched to LFD, or switched to HPMC, 10 mice/group. Panels D and E show change in β diversity over time. Average unweighted UniFrac distances from baseline microbiota (wk 0) to wk 0, 2, and 4 microbiota, within a dietary group (D) and distances from wk 2 microbiota to 2- or 4 wk microbiota (E). Each sample is represented as a colored circle. *P < 0.001.

Hierarchical clustering of the most abundant (≥1%) families in fecal specimens was visualized on heat map plots (Supplemental Fig. S2C). The samples did not cluster according to treatment at baseline, as expected. As the study progressed, the fecal samples gradually shifted toward 3 clusters based on diet. The cecal specimens showed a clear separation (P<0.001, Fisher's exact test; Supplemental Fig. S2) of the HPMC mice from the HFD and LFD groups, which were not distinguishable, consistent with the PCoA results (Fig. 3). In the ileal specimens, the LFD group clustered distinctly from the HFD and HPMC groups (P<0.05, Fisher's exact test; Supplemental Fig. S2).

Microbial population changes induced by dietary intervention

Quantitative PCR of the 16S rRNA gene showed that prior to randomization to the three treatment groups, there were no significant differences in total fecal bacteria or Bacteroidetes/Firmicutes (B/F) ratio between the mice (Fig. 4A). Over time, total bacteria levels were significantly decreased in fecal, cecal, and ileal samples from HPMC mice compared to the LFD and HFD groups. In addition, the LFD mice showed lower ileal bacteria compared to the HFD mice. As expected, the B/F ratio in fecal samples was unchanged in the HFD mice but increased in both the LFD and HPMC mice and in cecal samples from HPMC mice. In contrast, in ileal samples, the B/F ratios were decreased in LFD and HPMC mice. These data indicate that the dietary changes, especially HPMC supplementation, affected the overall size and composition of the intestinal microbiota.

Figure 4.

Changes in intestinal microbiota composition with dietary treatments. A) Total bacteria per gram of sample, and Bacteroidetes and Firmicutes, normalized by total bacteria, as determined by quantitative PCR using primers that target 16S rRNA (see Materials and Methods), and calculated B/F ratio for the following samples: wk 0 fecal (0F), wk 2 fecal (2F), wk 4 fecal (4F), wk 4 cecal (4C), wk 4 ileal (4I). *P < 0.05, **P < 0.01, ***P < 0.001; Mann-Whitney U test, 10 mice/group. B, C, D) Relative abundance of the predominant intestinal microbiota (≥1%) at the phylum (B), family (C), and the genus (D) level, determined by 454-pyrosequencing of the V3–V5 region of the 16S rRNA gene.

To assess specific changes in intestinal microbiota, we compared the relative abundance of predominant (≥1%) taxa identified from 454-pyrosequencing in HFD vs. LFD mice (to assess for the effects of dietary fat) and HFD vs. HPMC mice (to assess for the effects of fiber supplementation) (Supplemental Table S1 and Fig. 4B–D). No significant differences were seen in wk 0 fecal samples, as expected. Over the course of the experiment, changes at the phylum level were not significant after false discovery rate (FDR) adjustment; however, the trends observed from the pyrosequencing results were similar to those detected by qPCR (Fig. 4A). All significant differences at lower taxonomic levels were within the phylum Firmicutes, including members of class Bacilli, Clostridia, and Erysipelotrichi, and of order Lactobacillales, Clostridiales, and Erysipelotrichales.

At the family level, populations of the predominant microbiota in fecal samples from mice fed HFD were consistent over the course of the 4-wk study (Fig. 4C), with major stable populations of Lachnospiraceae, Peptostreptococcaceae, Ruminococcaceae, Erysipelotrichaceae, and Porphyromonadaceae, while a major Lactobacillaceae population is transiently present only in baseline and 4-wk fecal samples. The cecal microbiota of HFD mice is similar to fecal samples, but with near absence of Erysipelotrichaceae and Lactobacillaceae. The ileal microbiota of HFD mice is dominated by Peptostreptococcaceae and contains minor populations of families seen in the fecal specimens. In both treatment groups, changes in family level microbiota was observed in the 2-wk fecal sample and maintained in the 4-wk fecal sample, including a significant decrease in Lachnospiraceae and an increase in Erysipelotrichaceae. HPMC mice also had significantly increased Peptostreptococcaceae and decreased Ruminococcaceae in fecal and cecal samples, while LFD mice had reduced ileal Peptostreptococcaceae. While both dietary treatments altered fecal microbiota, differences in cecal and ileal populations were limited to HPMC and LFD mice, respectively. At the genus level, there were significant changes in the community representation (Fig. 4D and Supplemental Table S1); HPMC-supplementation led to significant decreases in Johnsonella and Lactobacillus, and significant increases in Erysipelotrichaceae incertae sedis (inc. sed.) and Peptostreptococcus inc. sed. LFD led to decreases in Johnsonella and Erysipelotrichaceae inc. sed., and increases in Clostridium. Thus, the dietary changes from HFD or to LFD and HPMC induced significantly different compositions within Firmicutes, with reproducible compositional effects at the family and genus levels.

Specific associations of weight change and metabolic variables via partial correlation analysis

Since dietary intervention affects both host metabolism and intestinal microbiome, we attempted to factor out treatment-mediated changes and directly capture interactions of the host and microbiome. In standard correlation analysis, significant correlations can result from large treatment-mediated changes, rather than a direct interaction of the two variables. We performed partial correlation analysis (59–61) to remove the effect of dietary variables by conditioning on either presence of fiber or percentage of fat in the diet. Significant FDR-adjusted P values for each pairwise comparison of taxa, in particular, samples with each metabolic variable are indicated (Supplemental Table S2); regression analysis was used to avoid reporting correlations influenced by outlier samples (Supplemental Fig. S3). Weight changes in HFD mice were positively correlated with cecal Firmicutes and cecal Erysipelotrichaceae inc. sed. and negatively associated with cecal Bacteroidetes (Supplemental Fig. S3A–C). For the HPMC mice, weight changes and fecal saturated fat were positively associated with cecal Erysipelotrichaceae inc. sed. and cecal Erysipelotrichales, respectively (Supplemental Fig. S3C, SF). The 4-wk fecal abundances of Lachnospiraceae were negatively correlated with energy intake in the HFD and HPMC mice (Supplemental Fig. S3D), and cecal Porphyromonadaceae correlated positively with liver free cholesterol in HFD and HPMC mice (Supplemental Fig. S3E). However, the only significant correlation verified by linear regression analysis in the LFD group was between baseline fecal Clostridiales and insulin (Supplemental Fig. S3O) and was not observed in the other diet groups.

DISCUSSION

The current study confirms and extends prior work, indicating that reducing dietary fat or adding dietary fiber improves markers of metabolic health in mice receiving a high-calorie HFD (12, 62–65). In particular, the addition of HPMC disrupts the relationship between energy intake and weight change (66), consistent with a prior study (44), and is accompanied with increased excretion of bile salts and fats in the feces, resulting in an increased loss in calories. LFD mice consumed 36.6% fewer calories and weighed 29.2% less than the HFD mice, whereas HPMC mice did not differ in caloric intake, but excreted 49.1% more energy and weighed 12.6% less than HFD mice. Thus, in mice with diet-induced obesity, HPMC supplementation improved metabolic biomarkers to an extent comparable to that of caloric reduction, which was associated with increased fecal loss of specific metabolites and energy content.

We have shown that changing from HFD to LFD or 10% HPMC supplementation resulted in marked shifts in the intestinal microbiota over a 4-wk period by PCoA representations of the UniFrac analyses (67) and by heat map analysis. The induction of intestinal microbial community shifts by dietary changes is no longer surprising (3, 68–70); however, we provide the first evidence that HPMC alters the intestinal microbiota, challenging the previously held notion that HPMC is inert with respect to the microbiota. The results are highly consistent within the experimental groups and show progressive changes in the microbiota sampled in the feces, with similar shifts in the cecal and ileal samples for HPMC and LFD, respectively. Although ecological richness showed no changes at higher taxonomic orders, richness of the LFD- or HPMC-affected intestinal community and evenness at the OTU (species) level declined. Such dynamics are consistent with HPMC selection for a small group of specialist organisms that become over-represented at the expense of the preexisting community members.

Because all of the major HPMC-induced shifts involve families within the phylum Firmicutes, differences in their functional or anatomic niches merit further exploration. Members of the microbial communities interact cooperatively by mutualistic cross-feeding (71), or compete for nutrients and physical space, or by secreting antimicrobial peptides (72). The populations of Lachnospiraceae, Ruminococcaceae (both members of the order Clostridiales), and unclassified Clostridiales diminished after both dietary changes, suggesting that the HFD diet specifically provides a competitive advantage to Clostridiales. This observation provides a confirmation and further characterization of the altered Bacteroidetes/Firmicutes ratio described by Ley et al. (20).

HPMC and LFD both improve metabolic health and shape the microbiome; however, correlations between specific microbial taxa and host metabolic phenotypes using partial correlation analysis, which minimizes the effect of the diet, could only be identified in HPMC mice and not in LFD mice. This dichotomy suggests that specific microbes may play a lesser role in the beneficial effects of LFD, than in the health improvements induced by HPMC. We now identify specific taxa that are associated with weight gain (e.g., Erysipelotrichaceae inc. sed.), and with other metabolic phenotypes. Although these associations do not indicate the direction of causality between microbiome and host metabolic change, they provide candidate taxa to be explored in future studies involving dietary manipulations. Therefore, HPMC supplementation may be used as a probe for better understanding of intestinal microbiome population structure due to dietary fiber intervention.

Although both LFD and HPMC increase levels of Erysipelotrichaceae, HPMC increases organisms within Erysipelotrichaceae inc. sed., while LFD increases members of the Erysipelotrichaceae that cannot be classified at the genus level. The taxonomy of bacteria is constantly evolving as our knowledge expands (73). Incertae sedis means “uncertain seat” in Latin, and is typically used when a new genus is justified, but a name has not yet been assigned. The genus Erysipelotrichaceae inc. sed. contains species previously classified in Clostridiaceae, Eubacteriaceae, and Lactobacillaceae (Table 1), and several members, including Eubacterium bioforme, E. cylindroides, E. dolichum, and Clostridium spiroforme increase in response to an HFD (68). The high levels of fecal fat in HPMC mice could be driving this specific increase. In contrast, LFD is likely increasing other genera within Erysipelotrichaceae, potentially Allobaculum, which has been reported to be higher in animals fed LFD compared to HFD and increases in response to weight loss (74). LFD also shows the greatest decrease in Lachnospiraceae, which have been observed to be lower in animals on LFD compared to HFD, and lowest in animals on LFD after undergoing weight loss (74).

Table 1.

Classification within inc. sed. genera and related families in RDP

Erysipelotrichaceae		Peptostreptococcaceae
Genera	Inc. sed.	Genera	Inc. sed.
Erysipelothrix	Clostridium catenaformis	Peptostreptococcus	Eubacterium yurii
Allobaculum	Clostridium cocleatum	Filifactor
Bulleidia	Clostridium innocuum	Sporacetigenium
Catenibacterium	Clostridium ramosum	Tepidobacter
Coprobacillus	Clostridium spiroforme	Anaerobium
Holdemania	Eubacterium bioforme	Proteocatella
Solobacterium Erysipelotrichaceae inc. sed. Turicibacter	Eubacterium cylindroides Eubacterium dolichum Eubacterium tortuosum	Peptostreptococcaceae inc. sed. Clostridium XI
	Lactobacillus catenaformis
	Lactobacillus vitulinus
	Streptococcus pleomorphus

RDP, Ribosomal Database Project (Michigan State University, Ann Arbor, MI, USA; http://rdp.cme.msu.edu/).

Fermentable dietary fibers, such as inulin (35), arabinoxylan (12, 64), and chitin-glucan (65), improve metabolic health and alter the microbiota, partially by increasing short-chain fatty acid production (33, 75). Similar to HPMC, arabinoxylan and chitin-glucan lowered total bacteria in the cecum. Arabinoxylan increased Bacteroides/Prevotella, which is consistent with increased Bacteroidetes in HPMC-fed mice. However, unlike HPMC, members of the family Lachnospiraceae as classified by the Ribosomal Database Project (RDP; Michigan State University, Ann Arbor, MI, USA; Roseburia and Eubacterium rectale/Clostridium coccoides group) were increased by arabinoxylan and chitin-glucan supplementation, whereas Lachnospiraceae was decreased by HPMC. In a study of adult men, polydextrose supplementation decreased populations of Lachnospiraceae, similarly to HPMC, and specifically decreased Roseburia (76). Bifidobacterium was increased by arabinoxylan as measured by qPCR, but not by chitin-glucan, and not detected as a predominant phylotype (>1%) in the HPMC-fed mice. These fermentable dietary fibers and nonfermentable HPMC similarly improve metabolism; however, the subsequent shifts in microbiota differ by type of fiber, highlighting the need for further characterization of the effects of specific fibers on the microbiota and host metabolism. The ability of HPMC to modulate the intestinal microbiota and maintain stable changes, whether due to a primary effect or to a secondary increase in fecal fats, indicates that it may be used as a prebiotic agent, able to reduce populations of specific microbes, including members of Lachnospiraceae (which contains common intestinal genera Blautia, Butyrivibrio, Coprococcus, Dorea, Johnsonella and Roseburia) and Ruminococcaceae (containing Faecalibacterium, Oscillospira, Ruminococcus, and Subdoligranulum), and induce the growth of specific microbes, such as members of Erysipelotrichaceae (containing Allobaculum, Coprobacillus, Holdemania, and Turicibacter) and Peptostreptococcaceae (containing Peptostreptococcus, Sporacetigenum, and clostridial cluster XI; ref. 77).

HPMC is considered to be not fermentable by gut microbiota in vitro (78) and consequently, is not absorbed. In a study administering ¹⁴C-radiolabeled HPMC to rats, 99% of the radioactivity was detected in the feces, 1% detected in the urine, and none detected in the tissues of the animal, demonstrating that HPMC does not undergo substantial degradation in the intestinal tract (46); thus, the way in which it alters the intestinal microbiota and improves health must differ from other commonly used prebiotic dietary fibers. The increases in fecal fats may be a primary source of microbiome perturbation; examination of similarly structured nonfermentable fibers may provide further insights. The similar fiber methylcellulose (MC) is not absorbed on passage through the rat digestive tract (79) or in humans (80) but is subject to partial glycosidic hydrolysis (81); in vitro, MC is subject to weak fermentation by fecal bacteria (78, 80, 82). Fermentation of hydroxyethylcellulose by human fecal microbiota in vitro yields 10-fold higher short-chain fatty acid concentrations than MC (82), providing evidence that substitution of modifying groups on cellulose affects fermentative potential, and raising the question whether side groups of HPMC could be acted on by the intestinal microbiota.

Although the results obtained in the current study do not provide evidence for direct interaction of HPMC with fecal microbiota, specific experiments for delineation of the mechanisms by which HPMC mediates changes in the microbial ecosystem can be designed in future investigations. Nevertheless, including baseline samples has allowed the assessment of early microbiologic changes and the identification of taxa that respond to HPMC treatment and that may interact with host metabolism. In addition, individual animals differed in phenotypic responses to dietary constituents; thus, such variation was used to identify candidate taxa associated with particular stressors. The current study provides the first observations of variation in genus-level taxa, in broad analyses without a priori candidates.