Microbiota-Derived Compounds Drive Steady-State Granulopoiesis via MyD88/TICAM Signaling

On detection of wounding or infection, neutrophils are recruited and swarm within minutes (1–3). The extreme sensitivity of neutropenic patients to infection demonstrates that this rapid response is essential for survival (4). However, production of sufficient neutrophils is a costly procedure. An adult human produces thousands of neutrophils per second, which during health predominantly senesce (5). The possibility to “tune” the levels of neutrophil production to the level of threat faced by the organism would therefore appear to be beneficial.

It is well documented that during severe bacterial infection, a process referred to as “emergency myelopoiesis” is induced (5, 6). In this situation, a much greater proportion of multipotent progenitor cells are diverted into the myeloid lineage at the expense of lymphopoiesis. This process appears to require sensing of pathogen-associated molecular patterns in a TLR-dependent manner (7). It is further known that massive production of IFN-γ can enhance the production of monocytes over granulocytes and can drive erythropoiesis to start at sites outside the bone marrow (BM) (8). However, increased production of mature granulocytes requires hours to days in the human system, and although this increased production is central to the later control and eradication of infection, the initial encounter with pathogenic or apathogenic bacteria requires the immediate action of the homeostatically maintained circulating and marginated granulocyte pools (1, 6).

We and others have previously noticed that germ-free (GF) mice show delayed kinetics of clearance of bacteria given i.v. (9, 10). GF mice are known to have an immature mucosal immune system, including reduced secondary lymphoid tissues, lower levels of secretory IgA, and fewer intestinal plasma cells (11). In addition, GF mice have lower levels of serum Abs and increased susceptibility to infection with a number of bacterial pathogens, both in the intestinal tract and systemically (11). It has been observed previously that treatment of specific pathogen-free (SPF), but not GF mice, with the lipid A–binding antibiotic polymyxin B reduces the number of granulocyte-monocyte colonies formed during in vitro culture of BM (12). Decreased numbers of granulocytes have also been observed in the BM of kanamycin-treated mice (13). More recent studies have identified increased bone mass and decreased osteoclast numbers in GF mice (14). Thus, we hypothesized that cross talk between the microbiota and the BM is important to regulate the size of the steady-state myeloid cell pool. This regulatory network may be implicated in the increased susceptibility to infections observed in GF mice and patients with dysbiosis.

C57BL/6, MyD88^−/−TICAM1^−/−, and NMRI mice were rederived GF as previously described (9, 11), and maintained GF in flexible film isolators at the Clean Animal Facility, University of Bern, Switzerland. C57BL/6 mouse colonies associated with a low-complexity microbiota (LCM) consisting of 10–100 different bacterial species (15) were originally generated by selective colonization of GF mice. Conventionally raised laboratory SPF mice were housed in individually ventilated cages (IVCs) at the Central Animal Facilities, University of Bern. All animals were housed in Bern, Switzerland, except for LCM mice shown in Figs. 2 and 4, which were maintained in a full-barrier facility in IVC cages at the Swiss Federal Institute of Technology (ETH) Zurich. In all experiments, mice were fed the standard mouse chow Kilba Nafag 3437.P.M.L15 M/R autoclaved at either 121°C (SPF and LCM mice in IVC housing) or 132°C (GF and gnotobiotic mice). All animal experiments were approved by the local Animal Care Committee.

Stable Escherichia coli K-12 JM83 monocolonizations, Enterococcus faecalis (mouse isolate) monocolonizations, and triple colonizations with E. coli K-12 JM83, S. xylosus (mouse isolate), and Enterococcus faecalis were done by intragastric gavage of GF animals with 10⁹ CFUs of pure-cultured bacteria in 500 μl. Inocula were aseptically prepared and imported into flexible film isolators. This mixture of bacteria was chosen based on our previously published work demonstrating absence of pathology in MyD88^−/−TICAM1^−/− animals (9), ease of culture and quantification, and to examine two relevant isolates commonly found in SPF mouse colonies alongside an apathogenic Gammaproteobacteria strain. Antibiotic-treated mice were administered 1 mg/ml metronidazole, 1 mg/ml ampicillin, 0.5 mg/ml neomycin sulfate, and 1 mg/ml vancomycin in the drinking water (refreshed every 4 d) for 4 wk before analysis. GF, E. faecalis monocolonized, and triple-colonized mice received 10⁷ live E. coli K-12 i.v. into the tail vein. Blood was collected retro-orbitally using sterile glass Pasteur pipettes under isoflurane anesthesia (Halocarbon Laboratories) 30 min, 3 h, or 6 h after injection of live bacteria. Animals were euthanized and spleen, liver, lungs, and BM were removed aseptically. Fecal pellets were collected, the cecum was opened, and an aliquot of cecal content was taken. Organs were homogenized in 0.5% Tergitol/PBS using a Tissuelyser (Qiagen) and sterile stainless-steel ball bearings. Cecal contents, blood, and organ suspensions were then plated on Luria–Bertani (LB) containing appropriate antibiotics for overnight culture at 37°C and CFU counting.

Eight-week-old GF C57BL/6 mice were exported into sterile IVCs and injected into the tail vein with heat-killed (20 min at 121°C) E. coli 10⁴ or 10⁶ diluted in PBS to a final volume of 500 μl. Pooled serum from SPF or GF J_H^−/− or RAG^−/− mice was sterile filtered and 400 μl injected per mouse. PBS was autoclaved, sterile filtered, and 500 μl injected per mouse. All mice were analyzed after 24 h, and GF status was confirmed by plating of feces on blood-agar plates and anaerobic incubation for 7 d at 37°C, as well as liquid cultures in thioglycollate medium (#CM0173; Oxoid) for 2 wk.

Reversible colonization was performed as described previously (16). In brief, D-Ala (200 μg/ml)/m-diaminopimelic acid (50 μg/ml)–supplemented LB cultures were aseptically inoculated from single colonies of E. coli HA107 and incubated with shaking at 160 rpm at 37°C for 18 h. Bacteria were harvested by centrifugation (15 min, 3500 × g, 4°C) in a sterile aerosol-proof assembly, washed in sterile PBS, and concentrated to a density of 2 × 10¹⁰ CFU/ml in PBS, all performed aseptically under a sterile laminar flow hood. The bacterial suspensions were sealed in sterile tubes, with the outside surface kept sterile, and imported into flexible film isolators, where 500 μl (10¹⁰ CFU) was gavaged into the stomachs of GF C57BL/6 mice. Inoculum samples were then re-exported from the isolators for bacterial quantification by plating on supplemented agar plates. Fecal samples exported from the isolator were bacteriologically analyzed to monitor HA107 shedding and bacteriological status of the inoculated mice.

GF or separately housed SPF C57BL/6 mice were i.p. injected with 1 mg BrdU (BD Biosciences) 36 and 24 h before analysis and retro-orbitally bled 8, 24, or 72 h after the second injection into sterile EDTA tubes. Femurs and tibias were collected and BM recovered by flushing in 2 ml RPMI 1640. Cells were surface-stained using Pacific blue–anti-Ly6G (Biolegend), PerCP–Cy5.5–anti-Ly6C (eBioscience), allophycocyanin–Cy7–anti-CD11b (Biolegend), and FITC–anti-B220 (Biolegend) Abs in 50 μl PBS/2% BSA, fixed, permeabilized using Cytofix/Cytoperm (BD Biosciences) and Cytofix/Cytoperm Plus (BD Biosciences) solutions, and intracellularly stained with PE-conjugated anti-BrdU Ab or PE-isotype control (BD Biosciences) following the manufacturer’s instructions. Samples were then acquired on a BD LSR II flow cytometer, and data were analyzed using FlowJo Software (Tree Star).

Fifty microliters whole blood or BM were incubated with 50 μl PBS/BSA2% containing Pacific blue–anti-Ly6G (Biolegend), PerCP–Cy5.5–anti-Ly6C (eBioscience), allophycocyanin–Cy7–anti-CD11b (Biolegend), FITC–anti-B220 (Biolegend), and PE–anti-CD54 (Biolegend) Abs, incubated 20 min on ice, and RBC lysis was performed using FACSLysis solution (BD Biosciences). Cells were washed in PBS/BSA 2%, and all samples were acquired on a BD LSR II Flow Cytometer. Data analysis was performed using FlowJo Software (Tree Star), and frequencies were normalized to the total number of leukocytes as determined using a VetABC animal blood counter (Medical Solution) or a hemocytometer (BM). Myeloid cells were identified by expression of CD11b and absence of B220, and then further subdivided into Ly6G⁺ granulocytes and Ly6C⁺ monocytes.

Bacterial FACS analysis was performed as described previously (11, 17). In brief, single colonies of plated bacteria were inoculated into 5 ml LB medium and cultured overnight at 37°C. Bacteria were then gently pelleted for 3 min at 7000 rpm in an Eppendorf minifuge and washed three times with sterile-filtered PBS/2% BSA/azide before determining the OD₆₀₀ and resuspending at ∼10⁷ bacteria/ml. Mouse serum was diluted 1:10 in PBS/2% BSA/azide and heat-inactivated at 60°C for 30 min. The serum solution was then spun at 13,000 rpm in an Eppendorf minifuge for 10 min to remove any bacteria-sized contaminants, and the supernatant was used to perform serial dilutions. Serum solution and bacterial suspension were then mixed at a ratio of 1:1 and incubated at 4°C for 1 h. Bacteria were washed twice before resuspending in monoclonal PE anti-mouse IgG1 and allophycocyanin anti-mouse IgM (BD Pharmingen). After a further hour of incubation, the bacteria were washed and then resuspended in 2% paraformaldehyde/PBS for acquisition by FACSArray using forward scatter and side scatter parameters in logarithmic mode.

Femurs and tibias were flushed into 20 ml RPMI 1640, and whole BM cells were counted in a hemocytometer using Trypan blue staining to exclude dead cells. BM lineage depletion was performed using biotinylated Abs against red cell precursors (anti-Ter¹¹⁹), B cells (anti-CD19, clone 6D5), T cells (anti-CD3ε, clone 145-2C11), and myeloid cells (anti-Gr1, clone RB6-8C5; all Biolegend), MACS anti-biotin beads, and LS columns (Miltenyi Biotec). Cells were then stained with anti-CD127–FITC (SB/199), anti-CD135–PE (A2F10), anti-CD16/32–PE–Cy7, anti-Thy1.1–allophycocyanin (OX-7), anti–c-kit–allophycocyanin–Alexa750 (all Biolegend); anti–Sca-1–PerCP–Cy5.5 (D7) and anti-CD34–eFluor 450 (RAM34; eBioscience). The hematopoietic stem/progenitor cell (HSPC)–enriched fraction was identified by the lack of lineage markers (lin⁻) and by the expression of Sca-1 and c-kit (lineage⁻, Sca1⁺, c-kit “high” cells [LSK]). Granulocyte-monocyte progenitors (GMPs) were identified from the myeloid lineage-restricted progenitors (c-kit⁺, Sca-1⁻, CD127⁻) according to FcγR and CD34 expression.

For Figs. 2 and 3, total BM cells (3.0 × 10³ cells/well) were cultured for 12–14 d in methylcellulose media (MethoCult M3234; Stemcell Technologies) supplemented with mouse IL-3 (10 ng/ml), human IL-6 (10 ng/ml), mouse stem cell factor (10 ng/ml), mouse GM-CSF (10 ng/ml), mouse thrombopoietin (10 ng/ml), and human erythropoietin (10 U/ml; all Peprotech). The colonies were counted on the basis of their morphological characteristics (18).

For Fig. 4, a total of 3.3 × 10³ MACS-purified lin⁻ BM cells were plated into MethoCult M3134 medium (Stemcell Technologies) supplemented with 15% FCS, 20% BIT (50 mg/ml BSA in IMDM, 1.44 U/ml human insulin [Actrapid; Novo Nordisk], and 250 ng/ml human holo transferrin [Prospec]), 100 μM 2-ME, 100 U/ml penicillin, 100 μg/ml streptomycin, 2 mM l-glutamine, and 50 ng/ml mouse SCF, 10 ng/ml mouse IL-3, 10 ng/ml human IL-6, and 50 ng/ml mouse Flt3-ligand (all from Prospec). Colonies were enumerated after 7 d (≥30 cells/colony) on a DMIL inverted microscope (Leica) equipped with an Intensilight C-HGFI unit (Nikon) (17).

Lipocalin 2 ELISAs were performed according to the manufacturer’s instructions with a few modifications (murine; R&D Systems). Nunc ELISA plates were coated with 50 μl capture Ab (1:200 in PBS) overnight at 4°C in a humidified chamber. After washing in PBS/Tween 0.05% (Sigma-Aldrich) and blocking in 150 μl PBS/BSA 2% for 15 min at room temperature, samples and standards were added in 3-fold dose titrations starting at 1:10 (serum and standard) and incubated overnight at 4°C in a humidified chamber. After washing in PBS/Tween 0.05%, 50 μl detection Ab (1:200 in PBS/BSA 2%) was added and plates were incubated for 1 h at room temperature. Plates were then washed in PBS/Tween 0.05% and 100 μl HRP-streptavidin (1:1000 in PBS; Biolegend) was added for 1 h. Plates were then washed and developed with 100 μl of substrate solution (10 ml substrate buffer, 1 mg ABTS, 10 μl H₂O₂ [both from Sigma-Aldrich]). OD was measured at 415 nm and four-parameter curves were generated to compare EC₅₀ values of samples and standards.

Serum cytokines were measured using cytometric bead arrays (BD Biosciences). Serum was 1:4 diluted and incubated for 1 h with 0.5 μl capture beads per sample (Mouse IL-6 flex set B4 #558301, Mouse TNF flex set C8 #558299, Mouse MCP-1 flex set B7 #558342; BD Biosciences). After washing in PBS, 0.5 μl PE-detection beads was added per sample and incubated 1 h at room temperature. Samples were then washed in PBS and acquired on a FACS Array (BD Biosciences). Four-parameter standard curves were fitted for each cytokine standard, and concentrations of cytokines in the serum samples were calculated.

Serum LPS concentrations were determined using the Limulus Assay kit according to the manufacturer’s protocol (50-647U, 50-648U; Lonza).

Differences were analyzed for statistical significance using Prism 4 for Macintosh (GraphPad Software). The details of the test carried out are indicated in the figure legends. Where data were approximately normally distributed, values were compared using either a Student t test for single variable or two-way ANOVA for two variables. Approximate p values were computed for two-way ANOVA. Where data were not normally distributed (e.g., bacterial CFU counts close to or equal to zero), nonparametric two-tailed Mann–Whitney U tests were applied. In all cases, p < 0.05 was considered significant.

We and others have previously reported that GF mice show impaired clearance of systemically administered bacteria, even when the bacterial strain given is fully avirulent (9, 10). To confirm these observations, we compared clearance of an i.v. dose of ampicillin-resistant E. coli K-12 from the blood of GF C57BL/6 mice, with clearance from littermates that had been colonized for 21 d with E. faecalis, S. xylosus, and ampicillin-susceptible E. coli K-12. At 6 h after injection, GF mice had higher counts of ampicillin-resistant E. coli K-12 in their blood (Fig. 1A). Because E. coli K-12 was present in the intestines of the colonized mice, we first sought to exclude a role for Ag-specific adaptive immunity in the improved bacterial clearance. Although intestinal secretory IgA specific for E. coli K-12 had been induced in the colonized animals (data not shown), this isotype at mucosal surfaces is not thought to augment bacterial clearance from the peripheral blood (19), and colonization of clean mice induces mucosal IgA independently of serum IgG induction (20). In the serum, there was no measurable specific IgG or IgM (Fig. 1B, 1C) directed against E. coli K-12 in either GF or colonized animals. As a positive control, we compared the titer of E. coli–specific IgG present in the serum of E. coli K-12 monocolonized MyD88^−/−TICAM1^−/− mice, which are known to mount strong systemic Ab responses to their intestinal bacteria because of failed mucosal containment (9) (Fig. 1B, inset). Supporting the interpretation that the differences in innate immunity likely underlie the delay in systemic vascular bacterial clearance in the GF state, our previous data demonstrated an identical phenomenon when we compared i.v. E. coli K-12 challenge of GF wild-type mice and mice colonized with the evolutionarily distant Firmicute E. faecalis (9). No Ab cross-reactivity could be observed between these two bacterial strains (9). We also confirmed experimentally that innate immune responses were enhanced during peripheral blood bacteremia in mice carrying intestinal microbes. After i.v. challenge there were enhanced responses of serum lipocalin 2 (Fig. 1D) and the myeloid-associated cytokines and chemokines IL-6 (Fig. 1E), MCP1 (Fig. 1F), and TNF (Fig. 1G) in colonized compared with GF mice, despite equal steady-state levels of serum lipocalin 2 in unmanipulated animals, regardless of their colonization status (Fig. 1H).

Initial clearance of bacteria from the blood involves both phagocytes and bactericidal serum factors such as the complement cascade. As murine complement activity is difficult to measure because of its labile nature (21) and complement factor production has previously been described to be normal in GF and Ag-free mice (22, 23), we decided to determine whether the microbiota was linked to steady-state myelopoiesis. We first quantified blood and BM HSPCs (which are found in the LSK fraction), granulocyte precursor populations, and mature granulocytes (Fig. 2A, 2B) from mice housed under four different hygiene conditions with increasing microbiota complexity: 1) GF, that is, mice housed stringently in isolators that have never encountered live microbes; 2) “triple-colonized mice,” which were generated by experimentally colonizing GF mice from pure cultures of E. coli K-12, E. faecalis, and S. xylosus; 3) LCM mice harboring between 20 and 100 bacterial strains but lacking any Gammaproteobacteria (4, 15) SPF mice, which were bred and maintained with a full microbiota of >100 bacterial species, but screened to be free of known mouse pathogens. In agreement with the literature on fully GF and antibiotic-treated mice (7, 12, 13), the number of myeloid cells, mature granulocytes, mature monocytes, and GMPs in BM all increased consistently with increasing microbiota complexity (Fig. 2C–F), suggesting the existence of a microbiota-dependent signal that increases myeloid cell production in healthy animals. The LSK cell population was only increased in the presence of a complete “SPF” microbiota, suggesting that amplification of this population requires either a stronger microbiota-associated signal or the presence of microbial species absent from gnotobiotic mice (Fig. 2G). The size of the common lymphoid progenitor (CLP) population did not differ significantly with hygiene status (Fig. 2H). These changes were also observed in classical CFU assays of total BM from GF and SPF mice, confirming a difference in the HSPC frequencies on a functional basis (Fig. 2I). In addition, SPF mice treated with broad-spectrum antibiotics showed a corresponding decrease in BM granulocyte and in the LSK cell numbers (Fig. 3A, 3B). The antibiotic-induced decrease in myelopoiesis could also be observed at the functional level in total BM CFU assays (Fig. 3C). Thus, the presence of an intestinal microbiota appeared to specifically amplify myelopoiesis in the BM, and with simple microbiota, this amplification occurred mainly between the production of multipotent progenitors and their differentiation into GMPs.

We next examined whether changes in BM populations induced by the microbiota were dynamic or represented an irreversible developmental shift in response to colonization. Thus, GF mice were treated with the strongly auxotrophic E. coli K-12 mutant HA107 (16). This bacterium colonizes the intestine for 12–24 h postgavage but cannot replicate in vivo, so GF status is regained between 24 and 48 h postgavage (16). At 12 h postgavage, an expansion of the BM myeloid cell pool was observed (Fig. 4A, 4B). However, 14 d postgavage, that is, 13 d after regaining GF status, HA107-treated mice were equivalent to GF mice in their BM myeloid cell status (Fig. 4A, 4B). Although the changes in myeloid progenitor cells did not reach significance in these experiments (Fig. 4C), we could observe an increase in CFU formation 12 h after HA107 treatment that returned to baseline 14 d after regaining GF status (Fig. 4D). Together, these data imply that changes in myeloid cell production are regulated dynamically depending on the current level of microbiota exposure.

In contrast with the BM, no significant differences were observed in peripheral blood granulocyte or monocyte numbers between any of the colonization statuses (Fig. 5A–C). We therefore used a 24-h pulse of BrdU labeling to address the kinetics of granulocyte production and release in GF and SPF mice. Although the total number of HSPCs, GMPs, and granulocytes in BM was consistently higher in SPF mice, the fractions of these populations incorporating label over time was identical between GF and SPF animals (Fig. 6A, 6B). This is consistent with published observations (7) and may indicate that different population levels are established early after colonization, but no steady-state kinetic differences in myelopoiesis between GF and SPF mice. However, the fraction of BrdU⁺ cells within any FACS marker–defined cell population is a complex function of influx from precursor populations, the rate of cell division within the defined population, the rate of death within the population, and the rate of differentiation into subsequent populations (24): any of these components may differ in GF and colonized animals. Alternatively, because the absolute numerical difference between GMPs in GF and SPF BM was ∼2-fold, either transient differences at the time of initial colonization or very small steady-state differences in the proliferation, immigration, and emigration rates that are not detectable within the resolution of our BrdU analysis could also sufficiently explain the observed data.

Blood granulocytes are terminally differentiated cells that are thought to be incapable of further cell division (postmitotic) and are directly derived from the nonreplicating BM granulocyte pool. Thus, the fraction of BrdU⁺ granulocytes in blood depends only on the rate of cell influx from the BM pool and the rate of granulocyte removal by death or tissue emigration. Because the steady-state number of granulocytes in blood is similar in GF and SPF mice (Fig. 5B), these two rates must be approximately equal. Thus, the significantly decreased fraction of BrdU⁺ blood granulocytes that is measured 48 h postlabeling in GF mice (Fig. 6C) represents a decreased rate of influx of granulocytes into the blood and a correspondingly decreased rate of clearance of granulocytes from the blood in GF mice.

Extending this logic, the rate of influx into the blood granulocyte population must be equal to the rate of efflux from the BM granulocyte population. However, no difference in the labeling kinetics between GF and SPF BM granulocytes was observed (Fig. 6D). This may be because of matched decreased differentiation and proliferation rates throughout the whole myeloid lineage, or may simply be too small an effect to see above biological noise, as only 1–2% of the total granulocyte population is in the circulation at any one time (25).

In summary, in GF mice, the BM granulocyte pool and myeloid precursors are present at half to one quarter of the normal numbers observed in SPF animals. This corresponds to slower kinetics of granulocyte turnover in the peripheral blood. Using our current methods, no change in the kinetics of BM myelopoiesis could be observed between GF and SPF animals. Taken together with the rapid dynamic responses seen during reversible colonization with HA107, this suggests that the increased output in steady-state is rather related to expansion of the BM that has occurred soon after colonization (shortly after birth) in SPF mice.

To define how the intestinal microbiota might be communicating with the BM, we attempted to replicate the colonization phenotype by sterile transfer of serum from SPF mice or GF mice into otherwise unmanipulated GF mice. This transfer of SPF but not GF serum was sufficient to greatly expand the BM myeloid cell pools (Fig. 7A, 7B) without affecting the peripheral blood granulocyte numbers (data not shown). Intriguingly, transfer of an equivalent volume of serum in which the majority of protein had been fully denatured (data not shown) and precipitated by boiling at 90°C for 30 min, followed by high-speed centrifugation and sterile filtration, showed an enhanced effect on BM myelopoiesis when compared with untreated serum, demonstrating that the active component is highly heat stable (Fig. 7A, 7B). Of interest, this phenomenon could be observed already at 24 h after serum transfer, indicating a very rapid effect on the myelopoietic lineage. Correspondingly, these treatments were also associated with an increase in the fraction of immature granulocytes in the blood, as revealed by a mild “left shift” of the granulocyte population (18) (Fig. 7C, 7D).

The heat stability of the active serum compound driving myelopoiesis suggested that microbial products in the serum might be necessary to maintain a sufficient myeloid cell pool. This was a surprising suggestion given the known firewall effect of the mesenteric lymph nodes and liver in preventing systemic spread of live microbes (26, 27), and the absence of microbiota-specific systemic adaptive immunity in healthy mice (9, 28). We therefore hypothesized that only concentrations of microbial products well below the threshold required to activate systemic adaptive immunity would be sufficient to recapitulate the myeloid cell expansion observed on colonization. Wild-type GF mice received injections of either 10⁴ or 10⁶ heat-killed E. coli i.v. to test this. The lowest dose of heat-killed bacteria accurately recapitulated the effects of colonization 24 h postinjection, whereas higher doses resulted in complex phenotypes, most likely associated with high levels of cytokine production and rapid release and consumption of BM granulocytes, as would be expected in sepsis-associated responses (1, 6, 29) (Fig. 7E, 7F).

Strong candidates for recognition and signaling downstream of systemic microbial compounds are TLRs (30). Indeed, lower rates of emergency myelopoiesis have been previously reported in MyD88^−/− animals (5, 18, 29), although continuously elevated systemic bacterial counts in SPF MyD88^−/− mice (9) could be associated with exhaustion of the myeloid system, complicating interpretation of these results. To determine whether TLR signaling is necessary for microbiota-induced increases in the myeloid cell pool, we carried out “triple” recolonizations of GF MyD88^−/− TICAM1^−/− mice and examined BM myeloid cell populations 21 d postgavage. In agreement with observations in SPF mice, no amplification of the BM myeloid cell pool was observed in MyD88^−/−TICAM1^−/− mice (Fig. 8A–C), despite elevated systemic bacterial exposure (9). In addition, transfer of sterile-filtered, boiled serum from wild-type SPF mice into GF MyD88^−/−TICAM1^−/− mice failed to induce any expansion of the BM myeloid compartment (Fig. 8D, 8E). MyD88 is also required for signaling via IL-1 family cytokine receptors, but because these cytokines should be destroyed by the vigorous heat-inactivation process used and were undetectable in the heat-treated sera (data not shown), these data strongly suggested that heat-stable microbial compounds present in SPF mouse serum are necessary to drive sufficient steady-state myelopoiesis. Correspondingly, MyD88^−/−TICAM1^−/− mice monocolonized with E. faecalis did not display improved clearance of E. coli (Fig. 8F).

It is now well established that the innate immune system is required to maintain homeostasis with the microbiota (9, 31–33). However, the details of cross talk between the intestinal microbiota and the host are only starting to be realized. In this article, we reveal that systemic recognition of microbiota-derived products by TLRs is necessary to maintain a sufficient pool of BM myeloid cells. This suggests that coevolution of the host with its resident microbes has led to a reliance on microbiota-derived signals to maintain vigilance to infection. At first, it seems an odd evolutionary trajectory to have been selected, as surely it would be safer to genetically encode this level of myeloid cell production within the host genome. In contrast, changes in microbiota composition can be strong indicators of increased infectious challenge, for example, overgrowth of Gammaproteobacteria can be associated with increased susceptibility to intestinal colonization with pathogens of the same phylum (15). It further seems likely that “microbiota complexity” alone is not sufficient to explain the stimulus for myeloid cell expansion. Microbial species display highly variable relationships to the host, both with respect to the niche that they colonize (adhesion to epithelial cells, mucus, luminal) and to the microbial products that they produce (34). It remains highly possible that particular species and their constituent molecular products are much stronger stimuli to the systemic innate immune system than others, and this will be an important area for future systems-biochemistry research.

In this study, animals were not perfused before analysis, and we therefore do not directly distinguish between “marginated” and “circulating” granulocytes in the BM. Nevertheless, because the concentration of granulocytes in peripheral blood does not differ significantly between GF and colonized animals, the increase in mature BM granulocytes would be likely indicative of an increase in the marginated pool of granulocytes, as well as an increased rate of release and consumption, as revealed by BrdU labeling. The importance of the marginated pool size is illustrated by human individuals with benign ethnic neutropenia, in whom expansion of this pool in the BM is thought to fully compensate the low circulating granulocyte numbers in the peripheral blood (35). Interestingly, although we saw no differences in the blood neutrophil numbers between GF and colonized mice, as the complexity of the microbiota increased, there was a 2- to 3-fold expansion of the total BM granulocyte population, which may represent the marginated pool, although this compartmentalization was not directly addressed. Steady-state granulopoiesis contrasts markedly with the situation in emergency myelopoiesis, where the marginated pool is rapidly mobilized into the circulation, reducing the total BM granulocyte numbers, at the same time as massively increasing the rate of granulocyte production at the expense of other hematopoietic lineages (5, 6). In keeping with the homeostatic nature of colonization-induced myelopoiesis, CLPs are unaffected by colonization. A further possible complication in interpreting the data is the existence of reverse migration of neutrophils to BM, either in steady-state (36) or as a clearance mechanism for aged cells (37). Nevertheless, the reverse migrating population, where observed, makes up ∼0.25% of circulating neutrophils in a healthy animal, so this phenomenon is unlikely to contribute dramatically to the doubling in the BM granulocyte pool observed upon colonization of GF mice.

Within the scope of this study we cannot exclude the role of other mechanisms in controlling the size of the myeloid cell pool in GF and colonized animals. Indeed, the decreased blood granulocyte turnover in GF mice (Fig. 6C) is indicative that there is a lower rate of granulocyte consumption in GF mice. Numerous previous studies have indicated that the diurnal pattern of granulocyte release and removal can strongly influence the rate of granulocyte production (38). However, Ab-induced neutropenia was shown to increase myelopoiesis identically in GF and SPF mice (7), suggesting that this mechanism is microbiota independent. Other studies have suggested that IL-17 (39, 40) and IFN-γ (8) play important roles in steady-state granulopoiesis, and that the production of these cytokines is known to be lower in the intestines of GF mice than of SPF animals (41, 42). Additional molecules originating from the microbiota, such as bile acid metabolites or short-chain fatty acids, are also known to shape the host immune system, for example, in the differentiation of regulatory T cells (39) or the activation status of intestinal macrophages (43). Such components are likely to be present in SPF mouse serum and be absent from GF mouse serum, and may not be fully inactivated by heat treatment; thus, we cannot exclude their role in modulation of granulocyte production. However, the complete absence of response of MyD88^−/−TICAM1^−/− mice to either recolonization or serum transfer suggests that TLR stimulation is both necessary and epistatic to any other regulatory mechanisms.

Intriguingly, TLR signaling is also essential for induction of emergency hematopoiesis (6, 29), suggesting the existence of a quantitative, rather than a “black and white,” switch from the steady-state. It is currently unclear whether there are differences in TLR sensitivity of different cell types, or differential signaling induced by low and high concentrations of TLR ligands within the same cells. Because of technical and ethical limitations in making axenic BM chimeras and maintaining them in strict GF conditions, it was not feasible to use this technique to determine whether the stromal or hematopoietic compartment was required to be MyD88 sufficient. Future generation of GF MyD88-Flox and cell-type–specific Cre-expressing lines (29) would permit accurate discrimination of these effects.

Despite the easily measurable differences in the population sizes of BM myeloid cells, we did not find a difference in the kinetics of replication and differentiation by BrdU labeling. This may be simply because of a larger BM niche in colonized mice, established at the time of colonization, as suggested by the observed lower bone density (14). It remains likely that a transient increase of precursor proliferation and/or differentiation rates would be observed at the moment of colonization, accompanied by generation of osteoclasts and expansion of the BM niche, that subsequently settles into an overall larger myeloid precursor pool that divides and replicates with indiscernibly different rates (24).

The presence of microbial products in serum that are capable of stimulating the production of a sufficient BM myeloid cell pool is an intriguing observation with potential clinical relevance in the management of patients with microbial dysbiosis, or patients on long-term antibiotic therapy. In combination with the observation that steady-state levels of granulocyte production are dynamically regulated based on the tonic stimulus received when using reversibly colonizing auxotrophic E. coli mutants, this suggests a system amenable to medical manipulation. A previous report observed the NOD1 ligand in both serum and BM of colonized mice (44), and although a potential effect on myeloid cell numbers was not addressed, a positive effect of the NOD1 ligand on granulocyte function was shown. Modulation of the intestinal pool of microbial molecules or their systemic derivatives may thus provide a novel way to adjuvant antibiotic therapy for improved patient outcomes or to dampen inappropriate inflammatory responses.

We thank J. Kirundi, A. Huguenin, and B. Flogerzi for technical support and C. Benarafa for helpful comments.

The authors have no financial conflicts of interest.