Dietary Fatty Acids Sustain the Growth of the Human Gut Microbiota

We built a three-vessel simulator system of the human gut by extending the previously published models (21, 22). The design of the system is shown in Fig. 1A. Each vessel of the HGS was seeded with an identical aliquot of fecal microbiota obtained from healthy adult donors. The HGS was validated to maintain anaerobicity, closed state (no contamination of the sterile medium in the assembled system with any environmental microbes), and precise temperature, pH, and transfer rate over a test period of 4 weeks. Figure 1B shows that the community in each vessel stabilized within the first 8 to 12 days of cultivation and community composition and then remained stable over several weeks of observation, similar to the findings from previous reports (23). The established communities resembled the original fecal inocula in comparison with unrelated human fecal sample (statistically significant similarity observed [Fig. 1C]). Not surprisingly, the similarity to fecal microbiota was highest for the distal vessel community. The microbes stratified among the vessels in agreement with the expected gradients of microbial concentrations along the colon (24). To illustrate this point, Fig. 1D shows that the levels of polysaccharide- and protein-degrading Bacteroides spp. decreased from the proximal to distal vessel, in concordance with the gradient of dietary glycan and protein availability. Mucin-degrading Akkermansia spp. displayed the opposite trend due to its ability to degrade host mucins that were added in equal amounts to each vessel (Fig. 1D).

To assess the ability of the human gut microbiota to utilize dietary fatty acids as the only readily available source of carbon and energy, we first established a stable community in each HGS vessel on a balanced medium that resembled colon influent on a typical Western diet. This Western medium (WM) contained complex and simple saccharides, proteins, and fats in proportions matching previously reported amounts in the human colon (Table 1) (13, 22). After 14 days of stabilization, the WM was switched to a fats-only variant, where all carbohydrates and proteins were removed, with the exception of a small amount of yeast extract that was a necessary source of vitamins and cofactors but which also contains carbohydrates and amino acids. Mucins were also removed from all three vessels to prevent microbes from relying on these glycoproteins and thus confounding the determination of which community members can be sustained specifically by dietary fats. The microbiota of the proximal vessel responded to this change in the supplied medium within 24 h, as was evident from the sharp drop in the community cell density on day 15 (Fig. 2A). A similar initial decline in the community cell density was also observed in the transverse and distal vessels, with an expected delay of response accounting for the transit time required for the new medium to reach the second and third vessels. Interestingly, the cell densities in all three vessels partially recovered after the initial drop. This recovery in cell density coincided with a decrease in the average size of cells and the lowering of cell metabolic activity in proximal vessel, as displayed in Fig. 2B and 3A, respectively. Note that while 146.6 ± 4.3 kcal · liter⁻¹ of macronutrient based energy was available in the Western medium, fats-only medium (FOM) only contained 57.0 ± 0.6 kcal · liter⁻¹ (39% of that of the WM), as determined by bomb calorimetry. The final stabilized cell density of communities grown in FOM was 56 to 57% of that observed on the Western medium (Table 2). Because FOM also contained yeast extract, an additional run was performed with the medium containing yeast extract and salts only (Table 1). Community density was drastically lower in all three vessels when supplied such yeast extract medium (Fig. 2A) and constituted only 5 to 7% of that on the WM (Table 2). Taken together, these data indicate a preference of the community to decrease cell size in order to maintain comparable cell density and division rate upon the switch to environments with lower nutrient availability. This possibly represents a solution to avoid cell washout from the colon when the supply of foods is reduced (25).

Since oxygen is displaced passively in the HGS system by N₂/CO₂ sparging, it is possible that residual oxygen can serve as the primary electron acceptor during fatty acid oxidation. To rule out that possibility, we performed batch-culture growth tests of HGS inocula in the anaerobic chamber system, where oxygen is completely removed (below 1 ppm) via a palladium-catalyzed reaction with H₂ gas. As displayed in Table 2, community densities in three different media in these anaerobic tests were similar to those measured in the HGS experiments. These experiments provided further evidence that human gut microbiota can utilize dietary fatty acids anaerobically for growth and community maintenance.

We used high-throughput 16S rRNA gene amplicon sequencing to assess the structure of microbial communities at each time point. To reduce biases in community composition estimates associated with the use of any single region of the 16S rRNA gene (26), we profiled both the V1–V2 and V4 16S rRNA gene regions. To reduce the PCR drift-associated bias of 16S rRNA gene amplification (27), 10 cycles of linear PCR were incorporated prior to exponential PCR phase. The distribution of diversity and evenness measures of community organization among vessels and between different media are shown in Fig. 2C. There was a noticeable gradient of decreasing community evenness and richness from the proximal to the transverse to the distal vessels (P ≤ 0.01). This effect was largely attributed to a very high abundance of mucin-degrading Akkermansia spp. in the transverse and especially distal vessels. The high prevalence of this single genus is explained by the transverse and distal vessels containing an ample amount of mucins but not receiving any other nutrients directly (the medium is moved from proximal vessel after it has been subjected to the fermentation there). Removing Akkermansia members from the data set largely evened out the measures of richness (Shannon H′ index) and evenness (Simpson E index) among the vessels (data not shown). Upon the switch to the fats-only medium, the measures of the richness and evenness remained unchanged for the proximal vessel community, but they increased for the transverse and distal vessels (see Fig. 2C). This observed increase in community richness in the transverse and distal vessels was due to the more even distribution of cell counts among different phylogenetic clades in FOM, in large part because the lack of mucin previously supplied to these vessels led to a disappearance of highly abundant Akkermansia species.

The community composition on the balanced Western medium was predictably dominated by members of the classes Clostridia and Bacteroidia, with a significant presence of Verrucomicrobiae (mostly Akkermansia spp.) in transverse and distal vessels (Fig. 2D). The switch to FOM led to a significant reduction in the overall abundance of Clostridia, with a smaller decrease in Bacteroidia and almost complete disappearance of Verrucomicrobiae. In contrast, the numbers of different clades of Proteobacteria increased (see Fig. 2D).

These changes in community composition were significant enough to be the main source of variability in the genus abundance data set. Phylogenetic UniFrac distance-based principal-coordinate analysis (PCoA) separated samples largely according to the medium supplied at a particular time point (Fig. 4A). Interestingly, while the cell counts in the proximal vessel dropped noticeably 24 h after the medium switch (day 15 of the run; see Fig. 2A), the community composition had not had sufficient time to change, as was evident by the clustering of day 15 samples from the proximal vessel with day 12 and day 14 samples in the PCoA ordination plot (Fig. 4A). Constrained canonical ordination analysis (28) similarly revealed that the type of supplied medium was the dominant predictor of community composition, accounting for 47% of the overall data set variance (see Fig. 4B and C).

To quantitatively determine which specific genera drove the observed alterations in community structure, we performed a weighted UniFrac distance-based principal response curves (dbPRC) analysis (see reference 29) on our genus abundance data set. PRC is a partial redundancy analysis which can isolate the variance explained by a chosen explanatory variable (medium switch/collection time point, in our case) (28). As shown in Fig. 4D, microbial communities in all three vessels responded to the medium switch by changing community structure, with good concordance in community alterations between replicate runs. Genus weights incorporated into the dbPRC model identified the genera that contributed the most to the community alterations. The genera that could least adapt to the sudden disappearance of carbohydrates, proteins, and mucins included well-known carbohydrate-degrading members of Clostridium XIVa, Roseburia, Bacteroides, and Dorea, as well as mucin-degrading Akkermansia spp. (Fig. 4E). This void in the community was quickly filled by representatives of Alistipes and Bilophila and by members of Gammaproteobacteria (e.g., Escherichia/Shigella, Enterobacter, and Citrobacter spp.). Figure 4F shows that the abundances of these taxa increased rapidly upon the switch to the fats-only medium, and they remained abundant until the end of the experimental run.

To determine how the switch to fats-only medium impacted the production of short-chain fatty acids by gut microbiota, we measured the levels of acetate, butyrate, and propionate with high-performance liquid chromatography (HPLC). In Western medium, acetate and propionate were the primary end products of fermentation (Fig. 3B). The large amount of propionate produced by the gut microbiota in WM was likely due to the fact that this fatty acid is formed during mucin degradation, consistent with the finding that mucin-degrading Akkermansia spp. release more propionate than acetate during fermentation (30, 31). Longer transit time also favors propionate formation (32). As shown in Fig. 3B, the overall production of SCFAs was reduced dramatically upon the removal of carbohydrates and proteins from the medium (86.2 ± 7.3 mM total SCFAs in proximal vessel supplied with WM and 25.5 ± 10.3 mM with FOM; P < 0.01). This constituted a 70% decrease in the overall SCFA production, compared with the 61% decrease in the total energy supplied by the FOM in comparison to WM, making SCFA production less efficient in FOM. The decrease in SCFA production in the proximal vessel was already evident 24 h after the medium switch, in that the total SCFA concentration on day 15 was 71% of that on day 12 and 14 (see Fig. 3B). By the end of the observation period, the production of acetate diminished the least (concentration on days 40 and 42 was on average 46% of that on days 12 and 14) but was more striking for butyrate and propionate (33% and 12% of the Western medium levels, respectively). The differential reduction in SCFA levels is likely explained by the fact that acetate is a direct end product of the β-oxidation of fatty acids.

Similarly, the total antioxidant capacity was significantly higher in the WM-cultured communities, even after adjusting for the lower cell densities of communities grown on FOM (Fig. 3C). This reduction also occurred within 24 h after the medium switch.

To determine the functional basis for the observed changes in gut microbiota structure upon medium swap to FOM, we sought to determine the predicted repertoire of functional metabolic genes in the communities before and after the medium switch. By comparing PICRUSt-based predictions of the functional capacities of each community, we found that the overall abundance of the lipid degradation-related genes was higher in the communities maintained on the fats-only medium (see Fig. 5A). Looking at the fatty acid degradation pathway in-depth, we uncovered that the genes encoding the enzymes of the β-oxidation pathway were predicted to be substantially (between 1.6- and 18.4-fold) more prevalent in the FOM-adapted communities than in the microbiomes grown on the Western medium (Fig. 5B). At the same time, genes participating in the biosynthesis of fatty acids were less prevalent in FOM microbiomes, highlighting the presence of microbes accustomed to the availability of lipids in the environment. In contrast, the overall abundance of carbohydrate digestive enzymes was lower in these specialized communities, likely because many gut microbiota members with large repertoires of the carbohydrate-active enzymes (CAZymes) were not able to adapt to fat-exclusive nutrient supply and were outcompeted by fat-utilizing specialists. At the same time, the predicted abundance of glycan biosynthesis genes was higher on FOM, possibly because the members of that community are adapted to the low availability of carbohydrates in the environment and thus possess pathways to synthesize necessary glycans (e.g., cell wall components) from other compounds.

The design of a three-vessel in vitro human gut simulator (HGS) system was based on previously published models (21, 22) and is shown in Fig. 1. The HGS system consisted of three continuously linked fermentation vessels, each mimicking environmental conditions of the specific sections of the colon which house the majority of human gut microbes. Vessel 1 simulated the proximal colon, vessel 2 simulated the transverse colon, and vessel 3 modeled the distal colon (21, 47). The medium (Table 2), which closely matched food bolus contents that reach the colon (11, 21, 48), was supplied to vessel 1, and vessel contents were moved “along the colon” in 42-ml pulses every 2 h. The medium transfer rate was set to allow an overall system retention time of 72 h to mimic the experimentally derived upper estimate of the transit time within the human colon (49, 50). The vessel volumes were set to 500 ml, 600 ml, and 600 ml, respectively. Medium was transferred between vessels via FlexFlo peristaltic pumps (Cole Parmer, Inc.). The environmental conditions (temperature, pH, relative volume, and movement of contents) in each vessel matched those experimentally measured in the corresponding section of the human gut (4, 50, 51). Temperature (37°C) and agitation (60 rpm) were controlled via Isotemp hot stir plates (Thermo Scientific) equipped with in-vessel temperature probes (Omega Engineering). The pH was controlled automatically using Etatron pH pumps (Cole Parmer, Inc.) and reservoir of 0.5 M NaOH. An anaerobic atmosphere was maintained through periodic daily sparging of each vessel headspace with filtered O₂-free 90%–10% mix of N₂ and CO₂ gases. Fermentation vessels were equipped with syringe adaptors, which enabled direct sampling from each vessel. A concentrated mucin solution (made from porcine gastric mucin) was supplied via syringe injection equally to all three vessels at a rate of 2 g per day.

Fecal material was collected from three healthy male volunteers (27 to 31 years old) who had no history of antibiotic or probiotic use, and no gastrointestinal illness for 6 months immediately preceding the fecal collection. All fecal samples were mixed together and thoroughly homogenized to ensure that all inocula contained equivalent microbial populations. All fecal processing was performed on ice under N₂ atmosphere. Fecal slurries were prepared at 10% (wt/vol) in chilled anaerobic phosphate-buffered saline (PBS) and frozen.

Two media were used to supply nutrients to microbial communities in the proximal vessel. The balanced Western medium was a rich medium containing carbohydrates, fats, and proteins in proportions matching the expected macronutrient distribution in colon influent on an average Western diet (55%, 20%, and 25%, respectively; see Table 2). The second medium (denoted fats-only medium [FOM]) was depleted in carbohydrates and proteins while maintaining the same level of dietary fatty acids (11%, 71%, and 18% of carbs, fats, and proteins, respectively). Resazurin was added to both media as an anaerobic indicator. All HGS vessels were seeded with a 50-ml aliquot of 10% fecal slurry and were incubated for 12 h to allow community establishment. The proximal vessel of the HGS system was then supplied semicontinuously (pulse every 2 h) with the Western medium (WM) for 14 days to allow community stabilization. At the end of day 14, the Western medium was replaced with the fats-only medium and the HGS system was operated for a further 28 days. Two completely independent 42-day-long runs were performed. In addition, a single control run was carried out where after 2 weeks on Western medium, the system was supplied with a medium containing only yeast extract and salts (denoted yeast extract medium [YEM]). To obtain cell density counts (in cells per milliliter) of each community, HGS samples collected from each vessel were diluted 100-fold, and cells were enumerated on Spencer hemacytometer via phase-contrast microscopy.

Ten-milliliter aliquots of WM, FOM, and YEM were first equilibrated for 24 h in an anaerobic atmosphere (85% N₂, 10% CO₂, and 5% H₂) inside the Coy anaerobic chamber. Cultures were then inoculated with fecal microbial inoculum identical to those used for HGS runs. All batch cultures were incubated inside the anaerobic chamber for 72 h to match the HGS transit time. Cell counting was performed as described above. Two replicate batch cultures were carried out for each type of medium.

A Parr 6200 isoperibol calorimeter (Parr Instrument Co.) was used to carry out isoperibol bomb calorimetry in order to determine the gross energy of growth media. Prepared aliquots of each medium were lyophilized, packed into capsules, weighed, and then combusted in the isoperibol bomb. Measurements were done in triplicate and calibrated to the energy of glycolic acid, and the capsule energy was subtracted from the total energy values obtained.

Total nucleic acids were extracted following the hot phenol-chloroform method and incorporated a bead-beating step to break down microbial cells (27). Total RNA and DNA were quantified on Qubit 2.0 fluorometer using Qubit RNA BR and Promega Qubit double-stranded DNA (dsDNA) HS assay kits, respectively, according to the manufacturer's protocols. Community metabolic activity was defined as the ratio of total RNA to DNA based on previous studies showing an association between this measure and population metabolic activity (52, 53).

Bacterial genomic DNA was isolated from each HGS sample using the ZR fungal/bacterial DNA MiniPrep kit (Zymo Research), as we did previously (27). Genomic DNA (gDNA) was amplified using two pairs of primers, one targeting 16S rRNA gene V1–V2 region (forward primer 16S gene complementary sequence, AGRGTTYGATYMTGGCTCAG; reverse primer 16S gene complementary sequence, GCWGCCWCCCGTAGGWGT), and another targeting V4 region (forward [GCCAGCMGCCGCGG] and reverse [GGACTACHVGGGTWTCTAAT] complementary sequences, respectively). Two different regions were interrogated to reduce biases in community composition estimates associated with the use of any one region of the 16S rRNA gene (26). Forward primers contained an Ion Torrent P1 adapter sequence, 6- to 7-nucleotide barcode sequence, and variable region-specific sequence. Reverse primer sequences included the adapter A and variable region-specific sequence. PCR amplification was performed with 25 ng of starting gDNA material and included 10 cycles of linear elongation with only the forward primers used, followed by 25 cycles of traditional exponential PCR (54). The inclusion of a linear PCR step decreased the stochasticity of the first few PCR steps (27) and allowed the use of a single PCR amplification reaction per sample. Purified amplicons were pooled equimolarly, and sequencing libraries were prepared using the Ion PGM template OT2 400 kit (Life Technologies, Inc.), according to the manufacturer's protocol. High-throughput sequencing was performed on an Ion Torrent PGM using the Ion PGM sequencing 400 kit and Ion 316 Chip. We obtained an average of 40,129 sequence reads per sample. Sequence reads were processed in QIIME (55). For each sample, sequence reads for different 16S rRNA gene variable regions were subsampled to the lowest value and then merged. Annotation of operational taxonomic units (OTUs) was performed via an open reference method against the Ribosome Database Project reference database version 11 of 16S rRNA sequences. Any sequences with below 60% annotation confidence were labeled “unassigned” at that taxonomical level. To obtain cell counts, sequence read counts for each OTU were first adjusted by dividing them by a known or predicted number of 16S rRNA gene copies in that organism's genome, following a previously described approach (27). Incorporating 16S rRNA gene copy number information was shown previously to improve estimates of microbial diversity and abundance (56). For the OTUs annotated above the species level, an average of 16S rRNA gene copies in all organisms of the next taxonomical level (e.g., genus, and if no data for that genus, a family) was used to calculate an estimated 16S rRNA gene copy number. These 16S rRNA gene copy-adjusted read counts were then converted to sample-specific cell counts based on the previously determined cell density measurements for each sample. As a result, the sum of cells of all taxa within each sample equaled the cell density in that sample, so that a sample with higher cell density had a proportionally higher overall number of cell counts. This approach allowed us to estimate the actual cell counts for each taxon in every sample and not rely on the relative abundance values (28).

Terminal restriction fragment length polymorphism (TRFLP) analysis was used to assess the stability of microbial communities during HGS validation experiments. Full-length 16S rRNA gene was amplified in a PCR, as described previously (27). Bact-27F and Univ-1492R primers were fluorescently labeled with 6-carboxyfluorescein (6-FAM) and 5-HEX fluorophores, respectively. Following amplification, three reaction mixtures were pooled and subjected to restriction endonuclease digestion by HaeIII and RsaI enzymes (New England BioLabs) at 37°C for 4 h. Restriction digests were prepared for genotyping by mixing 40 ng of digested sample with 8 μl of formamide. Samples were run on a 3730x capillary sequencer employing ROX500 internal standard for fragment size determination. Raw TRFLP profiles were annotated against the internal standards using the PeakScanner version 1.0 software and were further processed in Microsoft Excel utilizing custom-designed Visual Basic scripts. The scripts removed the ladder fragments and nonessential labels, reformatted character strings, and calculated sum of the peak areas/heights. Fragment peaks contributing less than 0.4% of the overall profile area were discarded.

Collected samples were centrifuged and supernatants filtered to remove particles. Twenty microliters of each supernatant were injected onto an Aquasil C₁₈ reverse-phase column attached to the HPLC system (Thermo Scientific). The mobile phase was 50 mM phosphate buffer and acetonitrile (99:1 ratio) at pH 2.8. Short-chain fatty acids were eluted at a flow rate of 1.25 ml · min⁻¹ over a 30-min period and detected at a 210 nm wavelength. Standard curves were constructed for each acid over a range of 100 mM to 10 μM concentrations.

Antioxidant capacity in samples was estimated by measuring supernatant ability to neutralize and counteract oxidation by ABTS [2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid)] radical. The supernatant was diluted 20-fold and mixed with a fresh 1:1 solution of 7 mM ABTS and 2.45 mM potassium persulfate. The reaction was measured colorimetrically at a 730 nm wavelength in triplicate. Trolox (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid) was used as a standard to create daily calibration curves for the ABTS-potassium persulfate solution. Antioxidant capacity was expressed as micromoles per milliliter of Trolox.

Bacterial cells were pelleted by centrifugation at 14,000 × g for 5 min. The supernatant was discarded, and the pellet was washed twice with 1 ml of isotonic Tris-buffered saline (TBS). The final pellet was diluted with TBS to an approximate cell concentration of 1 × 10⁶ cells · ml⁻¹ based on the cell density measurements of the cultures. Cells were then labeled with SYTO9 fluorescent dye and analyzed on an Accuri C6 flow cytometer, with 66 μl · min⁻¹ flow rate, detection threshold of 1,000 relative fluorescence units, and 22-μm core size. The distribution of cell sizes was estimated from the forward-scatter values of all SYTO9-positive events.

Repeated-measures analysis of variance (ANOVA) was used to test the statistical significance of the differences in measured values between WM and FOM time points. Multivariate statistical analyses (principal-coordinate analysis, canonical correspondence analysis, and principal response curves) were performed on the genus-level microbial abundance data set generally following the approaches we described previously (28). Matlab- and R-based scripts were used to run all algorithms. Statistical significance of group separation in principal-coordinate analysis space was calculated based on the permutation analysis of the Davies-Bouldin index, as we did previously (57). Weighted UniFrac distance-based principal response curves (dbPRC) analysis was performed as described previously (29). This method used phylogenetic UniFrac distance as a measure of sample dissimilarity and carried out a partial redundancy analysis to isolate the part of the total data set variance attributable to sample collection time as a variable (see reference 28) for a more detailed explanation. dbPRC is especially suited to analyze time-series data sets (29).

The sequence data set is available in the Sequence Read Archive repository under BioProject PRJNA487598.