Total Lipopolysaccharide from the Human Gut Microbiome Silences Toll-Like Receptor Signaling

Stool samples were collected from nine healthy adults and analyzed as illustrated in Fig. 1A. The microbial composition of each sample was analyzed by metagenomic whole-genome sequencing (WGS), which revealed an overall bacterial composition that was similar to the general microbial landscape depicted in Human Microbiome Project 1 (HMP1) (14) (Fig. 1B), suggesting that these samples are representative of the general population.

LPS was purified from each donor as previously described (15) and further processed as previously described by Hirschfeld et al. to remove contaminating lipoproteins and yield purified LPS (16). The immunostimulatory potency of each sample of fecal LPS was then assessed by stimulating human primary PBMCs. The production of six inflammatory cytokines, interleukin-10 (IL-10), IL-6, IL-8, TNF-α, IL-12p70, and IL-1b, was measured in parallel. All cytokines that reached detectable levels (IL-10, IL-6, IL-8, IL-1b, and TNF-α) showed similar patterns of production, in accordance with their common regulation by NF-κB activation downstream of LPS. IL-6 and IL-1b are shown as representative results. All LPS samples showed ≥2 orders of magnitude lower stimulatory potency than purified Escherichia coli LPS (no. 1, 2, 5, 8, and 10), and many (no. 3, 4, 6, and 9) were immunologically silent up to the largest dose tested on primary peripheral blood mononuclear cells (PBMCs; 1 mg/ml) (Fig. 1C). Similar results were obtained when stimulating human TLR2 (hTLR2) and hTLR4--NF-κB reporter cell lines, confirming the absence of direct signaling through these LPS-sensing signaling pathways (Fig. 1D and E). Hence, total fecal LPS has a very limited capacity to activate the TLR4--NF-κB pathway and elicit the production of inflammatory cytokines. This result is particularly striking given that the gut microbiome is thought to elicit robust TLR4 signals and that total fecal LPS is likely derived from diverse microbial origins, including species known to produce potent forms of LPS (e.g., gammaproteobacteria). We further examined the immunomodulating properties of fecal LPS by cotreating PBMCs with fecal LPS prior to stimulation with E. coli LPS to assess their ability to interfere with immune stimulation. Fecal LPS from each donor potently inhibited the cytokine (IL-6, IL-1b) production elicited by E. coli LPS. Most fecal samples reached significant inhibition at a 1:10 E. coli-to-fecal LPS ratio, while some reached significant IL-1b inhibition at a 1:1 ratio (Fig. 1F). Hence, total fecal LPS is a potent inhibitor of TLR4 stimulation.

Previous studies have demonstrated that a few bacterial species known to be present in the human gut microbiome produce nonstimulatory or immunoinhibitory LPS (10, 17), which prompted us to examine which bacterial species were present in these samples that contributed to the total pool of fecal LPS. Metagenomic WGS of the fecal samples revealed that the distribution of the main phyla, as determined by using MetaPhlAn2 (18), was similar in the nine volunteers and not distinguishable from the 140 healthy-donor samples in HMP1 (Fig. 1B; see Fig. S1A in the supplemental material), suggesting that these samples are representative of the general population. We also analyzed both sample cohorts by using the HUMAnN2 algorithm (unpublished; http://huttenhower.sph.harvard.edu/HUMAnN2 ) to determine encoded functional pathways and inferred gene ontology (GO) functions with UniRef50 gene families. As previously described for the HMP1 cohort, the distribution of the encoded functions is similar across all samples and was also comparable within the newly collected samples (Fig. S1B). To identify which strains are contributing to LPS production in the human gut, we focused on the GO terms relevant to LPS biosynthesis (Fig. 2A and B). Both in fresh samples and in the HMP1 data set, Bacteroidetes species dominate the three main GO terms related to LPS biosynthesis. Overall, Bacteroidetes species contribute 79% of the LPS biosynthesis in healthy volunteers and 92.4% of that in HMP1 samples (Fig. 2B). In contrast, proteobacteria are minor contributors to LPS biosynthesis, with an average of 14% of the total produced LPS being of E. coli origin in volunteers; in HMP1 donors, the percentage is 5.2% (Fig. 2B; Fig. S2C). The resulting estimated average Bacteroidetes-to-E. coli LPS ratio in the gut would be between 6:1 in volunteers and 18:1 in the HMP1 cohort. The abundances of the Bacteroidales species, and therefore their likely contribution to the LPS pool, are shown in Fig. 2C. Bacteroides species contributing to LPS are diverse but follow similar trends. Notably, Bacteroides ovatus, B. uniformis, and B. vulgatus dominate LPS production in both cohorts (Fig. 2C).

We next examined whether the bacterial composition of the samples would correlate with the potency of stimulation or inhibition in these samples (Fig. 2D and E). We could not identify a strong correlation between function and composition at the phylum level and could only identify weak-to-moderate correlations between a few individual species and the stimulatory potency of individual fecal LPS samples (Fig. 2D). However, we found that the abundance of several Bacteroidales species (Alistipes putredinis, Bacteroides caccae, Alistipes finegoldii) show a moderate-to-strong correlation with the inhibition of IL-6 cytokine production (Fig. 2E), as well as TNF-α and IL-1β production (data not shown). Cytokine production and trends were similar for all of the cytokines measured. Interestingly, we also identified several Gram-positive species whose abundance correlates with functionality, notably, Lachnospiraceae bacterium 5_1_63FAA (Fig. 2E). However, these strains are present at very low abundance (<0.5% of the genomic pool), Gram-positive strains do not contribute to the LPS pool, and it remains unclear what indirect impact they may have on LPS signaling. Our results show that Bacteroidetes species are by far the most abundant contributors to LPS biosynthesis functions in the healthy human intestinal microbiota, consistent with their high abundance relative to other Gram-negative species in the gut (12, 14).

We had previously shown that LPS from several members of the order Bacteroidales were unable to activate TLR4, consistent with literature reports (10, 17, 19), and that B. dorei produces an immunoinhibitory form of LPS (10). Thus, we examined whether immunosilent LPS is produced not only by Bacteroides and Prevotella spp. but also by additional gut-associated Bacteroidales bacteria (i.e., Alistipes spp.) not examined in our previous studies and whether immunoinhibitory LPS is a common, phylum-wide feature of the order Bacteroidales. We isolated LPS from 11 Bacteroidales strains and tested its capacity to elicit inflammatory responses from primary human PBMCs. All three Alistipes spp. tested produced an LPS with low immunostimulatory capacity, similar to other members of the order Bacteroidales (Fig. 3A; Fig. S2A and B). We then tested its capacity to modulate the response of primary human PBMCs to stimulation by E. coli LPS. Across the order Bacteroidales, LPS induces a potent reduction of TNF-α or IL-6 production. Production and trends were similar for all of the cytokines measured. This included a significant suppression at a 1:10 Bacteroidales-to-E. coli LPS ratio for all of the species tested, with the notable exception of Alistipes finegoldii, which only suppresses at higher ratios (100:1 or 1,000:1) (Fig. 3B and C). Comparable suppression was seen with IL-1β (Fig. 3B and C), and unexpectedly, similar results were obtained for inhibition of zymosan-elicited cytokine production (Fig. S2C).

Inhibitory forms of LPS have been previously described whose activity is derived from variations in the structure of the lipid A moiety (20, 21). The lipid A domain of LPS is responsible for the endotoxic properties associated with LPS due to recognition by the TLR4 complex and downstream activation of the NF-κB pathway (22). A reduction in the number of lipid A acyl chains by bacteria has been shown to modulate the recognition of LPS by TLR4 (23). Structural analysis of the lipid A domain from all of the members of the order Bacteroidales studied by matrix-assisted laser desorption ionization--time of flight mass spectrometry (MALDI-TOF MS) analysis revealed masses with a mass/charge ratio (m/z) of <1,700, consistent with the presence of underacylated lipid A structures (Fig. 3D). Specifically, all Bacteroides spp. and Prevotella copri showed a dominant peak having an m/z of 1,677.5 to 1,696.5, consistent with the [M − H]⁻ ions of a penta-acylated form of lipid A. Also common but with variable abundance is a peak at an m/z of approximately 1,435, consistent with the [M − H]⁻ ions of a tetra-acylated form of lipid A. Variation in determined mass is consistent with variable length in acyl chains. The spectrum of P. copri shows an additional minor peak with an m/z of 1,757.6, consistent with the addition of a phosphate group but otherwise identical to other members of the order Bacteroidales. All Alistipes spp. showed masses consistent with underacylated lipid A structures in addition to larger masses in A. shahii and A. finegoldii of unknown significance (Fig. 3D). Our analysis indicate that immunosilent and immunoinhibitory underacylated lipid A structures are conserved across the order Bacteroidales.

In a parallel effort to identify novel microbe-derived immunomodulatory molecules, we screened extracellular polymeric substances (EPS), a secreted cell-associated matrix often containing biologically active molecules (24). Given the intimate association of the commensal gut microbes with mucosal surfaces, EPS represent an unexplored niche for immunologically active factors. We extracted the EPS of 29 bacterial strains selected to comprise representatives of the four main phyla of the human gut microbiome (7) (Fig. S3A). We then assessed the capacity of these EPS extracts to modulate inflammation in primary human immune cells and showed that TLR2 (zymosan)- and TLR4 (LPS)-driven cytokine production is reduced 4- to 20-fold by extracts from several strains, mainly members of the order Bacteroidales (Fig. S3B to E). Using an in-gel LPS staining procedure (25), we determined that the EPS extracts from Bacteroides strains exhibit high LPS content (Fig. S3F). These results further confirm that LPS from members of the order Bacteroidales potently antagonizes TLR signaling and suggest that the EPS matrices of commensal organisms are a likely vehicle for immunomodulatory factors.

The bacterial strains used in this study are summarized in tables at https://figshare.com/s/5e56cc1a347ef4f1de49 . All strains were started from 20% glycerol stocks stored at −80°C, plated onto brain heart infusion (BHI) agar supplemented with hemin and vitamin K (Teknova; B1093), and grown anaerobically at 37°C. Liquid cultures of all strains were started from a single colony inoculated into 1,000 ml of BHI liquid medium supplemented with 10 ml of vitamin K-hemin solution (BD; 212354), 10 ml of trace minerals (ATCC; MD-TMS), 10 ml of trace vitamins (ATCC; MD-VS), and 50 ml of fetal bovine serum (HyClone; SH30071) and grown anaerobically for at least 48 h at 37°C. A flexible anaerobic chamber (Coy Laboratory Products) containing 20% CO₂, 10% H₂, and 70% N₂ was used for all anaerobic microbiology steps.

EPS matrix was extracted from all strains as previously described (7). Briefly, 250-ml volumes of 24-h cultures were recovered by centrifugation at 18,400 × g for 10 min (4°C). The prewashed cell pellet was suspended in 8 ml of phosphate-buffered saline by vortexing for 5 min, allowing the cell-bound EPS to dissolve. Planktonic cells were subsequently pelleted by centrifugation at 18,400 × g for 10 min (4°C). The supernatant was then carefully removed, filter sterilized with a 0.2-µm-pore-size filter, and added to 4 volumes of ice-cold absolute ethanol to precipitate the EPS. After centrifugation at 3,300 × g for 30 min, the precipitated-EPS pellet was washed with 70% ethanol, lyophilized, and then stored at −20°C. For further experiments, lyophilized EPS fractions were normalized by being dissolved in ultrapure water at the desired concentration. An average of 5 mg of lyophilized extract was recovered for each strain by using this protocol.

Blood buffy coats were obtained from healthy volunteers after informed consent was obtained. The study protocol and any amendments were reviewed and approved by an independent review board (New England IRB, Newton, MA) before the start of the study. This study was conducted in accordance with the ethical principles of the Declaration of Helsinki.

PBMCs were freshly isolated from blood by Ficoll-Hypaque gradient centrifugation as previously described (39). Monocyte-derived dendritic cells were differentiated in vitro from freshly isolated human monocytes as previously described (39). Briefly, CD14⁺ monocytes were isolated from freshly purified PBMCs by negative selection and magnetic bead sorting (Miltenyi). Cells were then incubated in complete RPMI 1640 in the presence of 50 ng/ml recombinant human granulocyte-macrophage colony-stimulating factor and 20 ng/ml recombinant human IL-4 (R&D Systems) for 7 days. For all experiments using human donor cells, data were generated independently with at least two donors. A representative data set was selected for incorporation into the figure.

To isolate the total LPS from a fecal sample, approximately 5 g of fecal material was homogenized into 10 ml of endotoxin-free water with a gentleMACS Dissociator. The resulting fecal slurry was allowed to settle for 5 min, allowing large particles to settle, and the supernatant was lyophilized for LPS purification. LPS purification from fecal material was performed with 500 mg of lyophilized material but otherwise performed as described for bacterial strains below.

For LPS isolation from bacterial strains, LPS from all strains was isolated from a 1,000-ml liquid culture grown under standard conditions for ~48 h by the hot water-phenol method as previously described (15). To remove trace amounts of endotoxin protein, phenol-purified LPS was further treated as previously described (16), with the modifications described below. Following the final ethanol precipitation, LPS was lyophilized to determine the yield with a Mettler Toledo XS105 Dual Range analytical balance (sensitivity, ≥0.1 ng) and resuspended in HyPure cell culture grade endotoxin-free water (HyClone) to a final concentration of 1 mg/ml without the addition of triethanolamine. To confirm the purity and normalization of feces-derived LPS, the final product was visualized with the Pro-Q Emerald 488 in-gel staining kit (Thermo Fisher Scientific) in accordance with the manufacturer’s instructions. In all cases, the Pro-Q Emerald 488 in-gel staining kit indicated a purity identical to that of LPS purified from bacterial isolates. However, given the complex molecular nature of human fecal material, our analysis does not exclude the possibility of contaminating substances in feces-derived LPS and it is not considered ultrapure.

PBMCs (10⁵) or monocyte-derived dendritic cells (5 × 10⁴) were incubated in the presence of LPS purified from the bacterial isolates indicated at doses ranging from 10⁻³ to 10⁴ ng/ml for 18 to 20 h. For inhibition assays, cells were plated in medium. LPS purified from the strain indicated was then added, followed immediately by 100 pg/ml LPS purified from E. coli. Supernatants were collected after 18 to 20 h of culture and analyzed with the cytokine bead array human inflammation kit (BD Biosciences) in accordance with the manufacturer’s instructions. This kit analyzes the levels of IL-10, IL-6, IL-8, TNF-α, IL-12p70, and IL-1b in the same samples. Groups were compared by using a two-tailed nonhomoscedastic t test corrected for multiple testing by the Sidak-Bonferroni method with GraphPad Prism software.

Stool samples were collected from healthy volunteers after informed consent was obtained. The study protocol and any amendments were reviewed and approved by an independent review board (Western IRB, Puyallup, WA) before the start of the study. Stool samples were collected by the participants in the morning and transported to the Novartis Institute for Biomedical Research in Cambridge, MA, on the same day. Samples were then stored at −80°C until shipping to the Broad Institute for DNA extraction. DNA extractions from stool samples were carried out with the QIAamp DNA Stool minikit (QIAGEN).

Metagenomic whole-genome shotgun sequencing libraries were prepared as follows. Metagenomic DNA samples were quantified by Quant-iT PicoGreen dsDNA Assay (Life Technologies, Inc.) and normalized to a concentration of 50 pg/μl. Illumina sequencing libraries were prepared from 100 to 250 pg of DNA with the Nextera XT DNA Library Preparation kit (Illumina) in accordance with the manufacturer’s recommended protocol, with reaction volumes scaled accordingly. Batches of 24, 48, or 96 libraries were pooled by transferring equal volumes of each library with a Labcyte Echo 550 liquid handler. Insert sizes and concentrations for each pooled library were determined with an Agilent Bioanalyzer DNA 1000 kit (Agilent Technologies).

Metagenomic whole-genome shotgun sequencing was performed essentially as previously described (10). Metagenomic libraries were sequenced on the Illumina HiSeq 2500 platform, targeting ~2.5 Gb of sequence per sample with 101-bp paired-end reads. Reads were quality controlled by trimming low-quality bases and removing reads of <60 nucleotides. Reads aligning with the human genome were identified with bowtie (40) and filtered out. Samples were profiled taxonomically with MetaPhlAn 2.0 (41) (http://huttenhower.sph.harvard.edu/MetaPhlAn2 ) and profiled functionally with HUMAnN2 (42) (http://huttenhower.sph.harvard.edu/HUMAnN2 ). HUMAnN2 maps metagenomic reads to UniRef50 (41) gene families of species identified in the MetaPhlAn2 taxonomic profiling step. Protein-coding sequences in these pangenomes have been preannotated to their respective UniRef50 families, which serve as a comprehensive, nonredundant protein sequence database. Reads that do not align with a known pangenome are separately mapped to the entirety of UniRef50 by translated search with DIAMOND (42). All hits are weighted on the basis of alignment quality and sequence length, with per-species and unclassified hits combined to produce community totals for each protein family (in addition to species-stratified totals) in numbers of reads per kilobase (RPK). RPK units were further normalized to numbers of RPK per million sample reads to account for variation in sequence depth across samples. Principal-component analysis plots were generated with the scikit-learn python package by using species abundance from MetaPhlAn2. Data are available at https://figshare.com/s/5e56cc1a347ef4f1de49 and https://www.ncbi.nlm.nih.gov/bioproject/PRJNA414479 .

We used HUMAnN2 to map UniRef50 gene families to GO terms, which were then aggregated into larger metabolic clusters with the CateGOrizer tool (43). This procedure yielded a comprehensive but manageable set of 13 nonredundant GO biological process terms for comparison of HMP1 and new samples (https://figshare.com/s/5e56cc1a347ef4f1de49 ).

For analysis of crude biomass, pellets from 10-ml overnight broth bacterial cultures were washed three times in water, methanol, and chloroform (0.8:1:2). Alternatively, purified LPS (200 µg) was used directly. The material was subjected to mild acid hydrolysis at 100°C for 30 min in 12.5 mM sodium acetate buffer, pH 4.5, in the presence of 1% SDS to break the 3-deoxy-d-manno-octulosonic acid linkage, and free lipid A was recovered by two-phase Bligh-Dyer extraction. The lipid A species were analyzed with a MALDI-TOF mass spectrometer (Bruker Ultraflex) equipped with a smartbeam laser at a 2-kHz firing rate. Spectra were acquired in negative-ion linear mode. The matrix used was a saturated solution of 6-aza-2-thiothymine in 50% acetonitrile and 10% tribasic ammonium citrate (9:1, vol/vol). Samples were dissolved in chloroform-methanol (4:1, vol/vol) and deposited on the sample plate, followed by an equal portion of matrix solution (0.5 µl).

HEK-293 cells (5 × 10⁴) stably expressing the NF-κB-inducible Lucia luciferase reporter gene and the genes for either hTLR4, CD14, and MD2 (TLR4-HEK) or hTLR2 and CD14 (TLR2-HEK) (5 × 10⁴) were seeded into the wells of 96-well plates and stimulated with doses of LPS purified from the strains indicated at doses ranging from 10⁻³ to 10⁴ ng/ml for 6 to 8 h. For inhibition assays, cells were stimulated simultaneously with 1 ng/ml LPS purified from E. coli. Luciferase activity was measured by BrightGlo (Promega) in accordance with the manufacturer’s instructions. HEK-293 reporter cells were purchased from InvivoGen. All cell lines were tested for mycoplasma contamination with a PCR-based assay by an independent service provider.

Human PBMCs were stimulated in the presence of increasing amounts of fecal LPS or cotreated with 1 ng/ml E. coli LPS and increasing doses of fecal LPS from the donors indicated. IL-6, TNF-α, and IL-1β concentrations in supernatants were measured after 20 h of culture and compared to those obtained with LPS treatment alone (0:1). Cytokine concentration upon stimulation at 1 or 10 μg/ml was obtained for LPS from individual stool samples and normalized. Inhibition potency at a ratio of 10:1 or 100:1 was obtained for individual stool samples and normalized. Five independent experiments were performed. Spearman rank correlation between the abundance of individual species and either stimulation or inhibition of IL-6, TNF-α, or IL-1β production was calculated by using pooled data from all experiments. Species that significantly correlate with function for all three cytokines measured after Benjamini-Hochberg correction are shown. Data are available at https://figshare.com/s/5e56cc1a347ef4f1de49 .