Despite the fact that many hypoxia-inducible genes are important in hematopoiesis, the spatial distribution of oxygen in the bone marrow (BM) has not previously been explored in vivo. Using the hypoxia bioprobe pimonidazole, we showed by confocal laser scanning microscopy that the endosteum at the bone-BM interface is hypoxic, with constitutive expression of hypoxia-inducible transcription factor-1α (HIF-1α) protein in steady-state mice. Interestingly, at the peak of hematopoietic stem and progenitor cell (HSPC) mobilization induced by either granulocyte colony-stimulating factor or cyclophosphamide, hypoxic areas expand through the central BM. Furthermore, we found that HSPC mobilization leads to increased levels of HIF-1α protein and increased expression of vascular endothelial growth factor A (VEGF-A) mRNA throughout the BM, with an accumulation of VEGF-A protein in BM endothelial sinuses. VEGF-A is a cytokine known to induce stem cell mobilization, vasodilatation, and vascular permeability in vivo. We therefore propose that the expansion in myeloid progenitors that occurs during mobilization depletes the BM hematopoietic microenvironment of O2, leading to local hypoxia, stabilization of HIF-1α transcription factor in BM cells, increased transcription of VEGF-A, and accumulation of VEGF-A protein on BM sinuses that increases vascular permeability.
Disclosure of potential conflicts of interest is found at the end of this article.
In adult mammals, most hematopoietic stem and progenitor cells (HSPC) reside in the bone marrow (BM). However, following stress, challenge, or stimulation of the BM compartment, a proportion of these bone marrow HSPC egress from the BM to circulate into the blood, a phenomenon called “mobilization” . Mobilized HSPC can then be harvested from the peripheral blood by apheresis, enriched, and stored for transplantation. Mobilization was first discovered in human patients in the 1980s, and there are now over 45,000 patients a year worldwide who receive cellular support using mobilized HSPC. The main reasons that mobilization is preferred to BM aspiration for collection of HSPC are increased safety and comfort for the donor, faster reconstitution, and greater disease-free survival in the recipient [1, 2].
Although significant advances have recently been made in understanding the molecular mechanisms responsible for HSPC mobilization [3, 4], additional mechanisms remain to be discovered. Granulocyte colony-stimulating factor (G-CSF), alone or in combination with chemotherapy, is the most commonly used agent in the clinic to elicit mobilization of transplantable HSPC. Both G-CSF and chemotherapeutic agents such as cyclophosphamide (CY) induce the expansion of neutrophils and their progenitors within the BM . Neutrophils are essential to HSPC mobilization, as mice made neutropenic by homozygous-targeted deletion of the G-CSF receptor gene (G-CSFR−/− knockout mice), or by administration of the specific anti-neutrophil antibody Ly-6G/Gr1, do not mobilize in response to G-CSF, CY, or neutrophil-activating chemokine interleukin-8 [6, –8]. This neutrophil accumulation results in the release of active neutrophil proteases, such as neutrophil elastase, cathepsin G, and matrix metalloproteinase-9 [5, 9], within the extravascular compartment of the BM. These proteases in turn degrade and inactivate the adhesive and chemotactic interactions necessary to retain HSPC within the BM, particularly VCAM-1  and CXCR4 and its ligand CXCL12/SDF-1 [10, 11], as well as the tyrosine-kinase receptor c-KIT  and transmembrane c-KIT ligand [13, 14], whose roles in BM retention and mobilization of HSPC have also been reported [15, 16].
The main function of neutrophils is to destroy pathogens and necrotic tissues during inflammation. Damaged tissues release proinflammatory cytokines, chemokines, and other biologically active products derived from invading microorganisms, which activate the subjacent endothelium. Circulating neutrophils roll on, adhere to, and extravasate through this activated endothelium and then migrate to the site of inflammation or injury, guided by the released chemokines and bacterial products. Inflamed tissues are generally very hypoxic because of the local disruption of the blood supply. Thus, neutrophils and macrophages must migrate against an oxygen gradient to reach the site of inflammation. Although most cells are stressed in hypoxic conditions, neutrophils and macrophages are attracted to and activated in hypoxic areas, a property dependent on the stabilization of a family of hypoxia-inducible transcription factors (HIFs) [17, , –20].
HIFs are heterodimers composed of one stable β subunit, also called arylhydrocarbon receptor nuclear translocator (ARNT), and one α subunit (HIF-1α, HIF-2α, or HIF-3α, collectively called HIF-α), which, in normoxic conditions, is rapidly hydroxylated by prolyl hydroxylases on proline residues, leading to polyubiquitination and rapid degradation by the proteasome [21, 22]. In hypoxic conditions, HIF-α subunits are not hydroxylated and are consequently stable, persisting as functional heterodimeric HIFs in the nucleus. Dimeric HIFs bind to hypoxia-response elements (HRE) to activate the transcription of a variety of genes, such as glucose transporters, anaerobic glycolytic enzymes, vascular endothelial growth factor A (VEGF-A), and erythropoietin . Thus, HIFs are the built-in cell oxygen sensors whose expression is regulated by protein stabilization rather than the mRNA level.
The essential role of HIF-1α in neutrophil and macrophage function is illustrated by the fact the homozygous deletion of the HIF-1α gene specifically in myeloid cells results in a complete ablation of the inflammatory response elicited by macrophages and neutrophils in a variety of in vivo models of chronic and acute inflammation . Neutrophils and macrophages lacking HIF-1α display dramatically impaired motility, invasiveness, and bacterial killing . Conversely, mice with a myeloid-specific homozygous deletion of the von Hippel-Lindau tumor suppressor gene, which controls the proteolytic degradation of HIF-1α protein in normoxic conditions , have supraphysiological concentrations of HIF-1α in neutrophil and macrophage nuclei and an exacerbated inflammatory response in vivo .
In addition to the critical role of hypoxia and HIFs in neutrophil function and inflammatory responses, it is well known that HSPC themselves expand more extensively and differentiate less rapidly in hypoxic than normoxic conditions in vitro [25, –27]. Furthermore, functional HIFs are essential to the development and survival of the hematopoietic system, as mouse embryonic stem cells lacking the ARNT gene have impaired ability to form hematopoietic cells in vitro , whereas mice lacking the HIF-2α/EPAS-1 gene exhibit a pancytopenia .
Considering that (a) HSPC mobilization involves a dramatic alteration of the BM hematopoietic microenvironment and of the hematopoietic stem cell (HSC) niche, (b) HSPC mobilization absolutely requires functional neutrophils, and (c) hypoxia and HIFs play a critical role in neutrophil function, we explored the possibility that mobilizing agents could cause local hypoxia within the BM. By tracking hypoxia in vivo with the hypoxia-specific probe pimonidazole hydrochloride, we demonstrate that mouse BM becomes extremely hypoxic following administration of mobilizing doses of G-CSF or CY. Furthermore, mobilization of HSPC and increased medullar hypoxia coincide with a marked increase in HIF-1α protein levels in the BM and enhanced expression and levels of VEGF-A, a cytokine that is known to induce HSPC mobilization and whose transcription is regulated by HIF-1α.
Materials and Methods
Eight-week-old BALB/c and 129SvJ male mice were injected twice daily with 125 μg/kg recombinant human G-CSF (Filgastrim/Neupogen; Amgen, Thousand Oaks, CA, http://www.amgen.com) on 4 consecutive days during the period from day 0 to day 4. A second group of mice received a single injection of 200 mg/kg CY (Endoxan; Baxter, Deerfield, IL, http://www.baxter.com) on day 0. At specified time points, mice were euthanized by cervical dislocation, and BM and bones were harvested as described below.
Mice were injected with G-CSF, CY, or saline as described above. At day 4 of G-CSF or saline injection, or at day 6 following CY injection, mice were injected intraperitoneally 3 hours prior to sampling with 60 mg/kg pimonidazole hydrochloride (Hypoxyprobe; Chemicon, Temecula, CA, http://www.chemicon.com) solubilized in sterile saline. Following anesthesia with isoflurane, hind limbs were fixed by perfusing 2% paraformaldehyde/0.05% glutaraldehyde into the descending femoral aorta prior to sacrifice. Dissected femurs were decalcified in 10% EDTA for 10 days at 4°C prior to paraffin embedding and sectioning.
Following dewaxing and rehydration, femur sections were washed in phosphate-buffered saline (PBS) prior to antigen retrieval in 10 mM citrate buffer, pH 6.0, at 90°C for 20 minutes. Sections were cooled to room temperature in the buffer prior to being washed twice in PBS. Sections were incubated in 50 mM glycine in PBS (pH 3.5) for 5 minutes, washed in PBS with 0.3% Triton X-100 for 15 minutes and had two changes of PBS for 5 minutes, and treated with 3% H2O2 in methanol for 15 minutes. Sections were washed in three changes of PBS 0.05% Tween-20 (PBST) for 5 minutes prior to being blocked in 5% skim milk, 5% bovine serum albumin, 0.05% Triton X-100 in 4× standard saline citrate buffer (0.6 M NaCl and 0.06 M sodium citrate, pH 6.4) containing 10 μg/ml donkey IgG in a humidified chamber for 60 minutes at room temperature. Then, sections were incubated with 20 μg/ml mouse IgG1 fluorescein isothiocyanate (FITC)-labeled anti-pimonidazole monoclonal antibody (mAb) (Chemicon) or 20 μg/ml FITC-labeled mouse IgG1 control. Sections were washed three times in PBST and incubated with 2 μg/ml mouse anti-FITC IgG1 conjugated to horseradish peroxidase (Chemicon). Following three washes in PBST, the signal was amplified for 6 minutes using the PerkinElmer (PerkinElmer Life and Analytical Sciences, Wellesley, MA, http://www.perkinelmer.com) TSA biotin system as described in the manufacturer's instructions. Sections were incubated with 1 μg/ml Alexa488-conjugated streptavidin (Molecular Probes Inc., Eugene, OR, http://www.probes.invitrogen.com) for 30 minutes at room temperature and washed in PBST three times prior to mounting in Vectashield Mounting Medium (Vector Laboratories, Burlingame, CA, http://www.vectorlabs.com) containing 5 μg/ml 4′,6-diamidino-2-phenylindole (DAPI) (Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com).
As a positive control for hypoxia, 100,000 metastatic breast tumor 4T1.2 cells  were mixed with an equal volume of Matrigel (BD Biosciences, San Diego, http://www.bdbiosciences.com) and injected into the mammary fat pad of a female BALB/c mouse. Twenty-three days later, pimonidazole was injected as described above; 3 hours later, the mouse was sacrificed, and the tumor and healthy mammary gland were dissected, fixed overnight in PBS containing 4% paraformaldehyde, and embedded in paraffin for pimonidazole staining as described above. Staining of the hypoxic tumor is shown in supplemental online Figure 1.
Sections of marrow prepared and processed as described above were also labeled with rabbit anti-mouse HIF-1α (Chemicon) or rabbit IgG at 10 μg/ml as a control at room temperature overnight, followed by biotin-conjugated goat anti-rabbit IgG at 1.5 μg/ml for 60 minutes at room temperature and then 1/100 streptavidin-horseradish peroxidase (TSA kit; PerkinElmer) for 30 minutes at room temperature. Sections were finally incubated with 1 μg/ml Alexa488-conjugated streptavidin as described above.
Sections of marrow prepared and processed as described above were also labeled with either a rabbit anti-human VEGF-A 165 cross-reacting with mouse VEGF (Abcam, Cambridge, MA, http://www.abcam.com) or rabbit control IgG at 1 μg/ml overnight at room temperature, except that antigen retrieval was done using 10 mM citrate buffer at pH 10.0.
Image Acquisition and Quantification of Fluorescence
Images were acquired on a Zeiss LSM 510 Meta confocal laser scanning microscope with either an EC-Pan Neofluar ×10/0.30 M27 or a Plan Apochromat ×20/0.75 objective (Carl Zeiss, Jena, Germany, http://www.zeiss.com). DAPI fluorescence was excited at 405 nm with a diode 405-30 laser and detected through a 420–480-nm band-pass filter. Alexa488 was excited at 488 nm with an argon laser, and emission was detected through a 505–530-nm band-pass filter. Each fluorescence was excited, scanned, and acquired separately. Images are average of 16 consecutive scans. Images were analyzed and fluorescence was quantified using LSM 510 software from Carl Zeiss.
To quantify fluorescence on each acquired image, a parallelogram with a 200-μm width and 1-mm length (approximately 200,000 μm2) was drawn from the endosteum into the BM and defined as the endosteal region. A second parallelogram of equivalent size was drawn in the middle of the BM and defined as the central BM region. The fluorescence intensity of each pixel within these regions was measured and counted, and fluorescence histograms were plotted. Following comparison with consecutive serial sections stained with control antibodies, all pixels with fluorescence intensities from 1 to 1,471 were scored as negative, whereas pixels with fluorescence intensities from 1,472 to 4,095 were scored as positive. The percentage of fluorescence-positive areas within endosteal and central BM regions was calculated for each section.
HIF-1α Western Blots
For HIF-1α protein quantification by Western blotting, BM cells were flushed from one femur with 200 μl of ice-cold PBS containing 200 μM phenylmethylsulfonyl fluoride, 1 μg/ml pepstatin-A, 5 μM leupeptin, 20 μg/ml aprotinin, 10 μM elastase inhibitor III (Calbiochem, San Diego, http://www.emdbiosciences.com), 10 μM cathepsin G inhibitor N-methoxysuccinyl-alanine-alanine-phenylalanine-PO(O-phenyl)2 (MP Biomedicals, Solon, OH, http://www.mpbio.com). The remaining bone cavity was then immediately flushed with 800 μl of urea cell lysis buffer (7 M urea, 10% glycerol, 1% SDS, 5 mM EDTA, 20 mM Tris-HCl, pH 6.8) containing the same protease inhibitors to extract proteins from cells of the endosteal region. These cell lysates were then immediately frozen on dry ice and stored at −80°C.
As controls for HIF-1α, mouse myeloid FDCP-1 cells and human myeloid MO7e cells were cultured for 6–24 hours under normoxic or hypoxic conditions in degassed medium in a sealed container containing an Anaerogen anaerobic pouch (Oxoid, Basingstoke, U.K., http://www.oxoid.com). Following centrifugation at 400g for 4 minutes, cell pellets were washed once with ice-cold PBS before cell lysis in 500 μl of ice-cold urea lysis buffer as described above.
To determine whether mobilizing cytokines can stabilize HIF-1α protein expression, BM cells were flushed with 2 ml of PBS containing 2% fetal calf serum (FCS) from the femurs and tibias of C57BL/6 mice, washed twice in Iscove's modified Dulbecco's medium supplemented with 10% FCS and resuspended at 5 × 106 nucleated cells per milliliter. Duplicates of 1-ml cultures were established in two 24-well plates. Recombinant cytokines (human G-CSF, rat KIT ligand, and human thrombopoietin [TPO]) were added at 100 ng/ml. Plates were then incubated for 24 hours in normoxic or hypoxic conditions as described above. Duplicates were pooled, and cells were washed and lysed in 150 μl of ice-cold urea lysis buffer as described above.
Protein content in cell lysates was determined using the micro BCA kit (Pierce, Rockford, IL, http://www.piercenet.com) according to the manufacturer's directions. For each time point, equal protein concentrations of cell lysates were mixed with 5× loading buffer containing 10 mM dithiothreitol and boiled for 3 minutes before electrophoresis on 7% SDS-polyacrylamide gel electrophoresis gels. Following transfer onto nitrocellulose membranes and blocking overnight in Odyssey blocking buffer (Li-COR Biosciences, Lincoln, NE, http://www.licor.com), membranes were incubated overnight at 4°C with mouse anti-human HIF-1α monoclonal antibody H1α67 cross-reacting with mouse HIF-1α (Novus Biologicals, Littleton, CO, http://www.novusbio.com) diluted 1/1,000 in equal volumes of Odyssey blocking buffer and PBST. Following several washes in PBST, membranes were subsequently incubated for 1 hour with AlexaFluor 680-conjugated goat anti-mouse IgG antibody (Molecular Probes) diluted 1/5,000 in equal volumes of Odyssey blocking buffer and PBST.
In later experiments, BM cell lysates from 129SvJ and C57BL/6 mice were analyzed using a rabbit anti-mouse HIF-1α polyclonal antibody (Novus Biologicals) diluted 1/1,000 and subsequent incubation with an IRD800-conjugated donkey F(ab)′2 fragment anti-rabbit IgG (Rockland Immunochemicals, Gilbertsville, PA) at a 1/10,000 dilution. For normalization of the results, blots were then stripped by a 20-minute incubation with 25 mM glycine-HCl, 2% SDS, pH 2.0, buffer and reprobed with a rabbit anti-β actin loading control polyclonal antibody (Novus Biologicals).
Visualization and quantification were performed on the Odyssey Infra-Red Imaging System (Li-COR Bioscience) equipped with two solid-phase lasers at 700 and 800 nm with a resolution of 169 μm.
Femurs were flushed with ice-cold PBS to isolate “central” BM cells. Total RNA was extracted from 5 × 106 central BM cells using 1 ml of Trizol (Invitrogen, Carlsbad, CA, http://www.invitrogen.com). Empty femurs flushed twice with PBS were then reflushed with 1 ml of Trizol to extract RNA from cells of the endosteal region adjacent to the bone.
Following DNase treatment and reverse transcription using random hexamers, quantitative real-time reverse transcription-polymerase chain reaction (qRT-PCR) with SYBR green (ABI Systems, Foster City, CA) was performed according to the manufacturer's instructions using the following oligonucleotide combinations: osteocalcin (forward, 5′-TTCTGCTCACTCTGCTGACCCT-3′; reverse, 5′-CCCTCCTGCTTGGACATGAA-3′), cathepsin K (forward, 5′-GGCTGTGGAGGCGGCTAT-3′; reverse, 5′-AGAGTCAATGCCTCCGTTCTG-3′), CXCL12 (forward, 5′-ATGCCCCTGCCGGTTCT-3′; reverse, 5′-TGTTGAGGATTTTCAGATGCTTGA-3′). VEGF-A primers (forward, 5′-ACATCTTCAAGCCGTCCTGTGT-3′; reverse, 5′-CGCATGATCTGCATGGTGAT-3′) were designed across exons 3 and 4 to amplify all known transcripts of the mouse VEGF-A gene. RNA levels were standardized by parallel qRT-PCRs using primers to two different housekeeping genes, vimentin (a cytoskeletal protein; forward, 5′-CACCCTGCAGTCATTCAGACA; reverse, 5′-GATTCCACTTTCCGTTCAAGGT) and β2-microglobulin (forward, 5′-TTCACCCCCACTGAGACTGAT; reverse, 5′-GTCTTGGGCTCGGCCATA). A PCR from each sample prior to reverse transcription was also performed to confirm the absence of contaminating genomic DNA.
All significance levels were calculated using the Student's t test.
Expansion of the Granulocyte Pool Predicts Increased Oxygen Consumption in Mobilized BM
Oxygen tension within a solid tissue can be mathematically predicted by the Krogh's model . This model and has recently been modified to predict O2 tension distribution within the hematopoietic compartment of the BM as a function of the distance to BM endothelial sinuses (which provide O2) [32, 33]. Oxygen diffuses from the blood, which is the source of O2 for all solid tissues except the lung. Because of its slow diffusion, O2 forms a decreasing gradient from the blood vessel that decreases rapidly with the distance from the blood vessel. Other than the distance to the BM sinus, the most important parameters that influence O2 tension within any solid tissue are (a) the number of cell layers between the blood vessel and the point of prediction and (b) the O2 consumption rate of the cells comprised between the point of measurement/prediction and the blood vessel [32, 33]. The more O2 is consumed by cells of the solid tissue, the more rapidly O2 will be depleted. Interestingly, in vitro measurements have revealed that cells that consume O2 most actively are megakaryocytes (14 × 10−13 mol of O2 per cell per hour), granulocyte/monocyte progenitors (6.53 × 10−13 mol of O2 per cell per hour), and adipocytes (3.24 × 10−13 mol of O2 per cell per hour), whereas mature myeloid cells, lymphocytes, and erythrocytes have much lower O2 consumption rates (Table 1). Consequently, this model predicts that three layers of monocyte/granulocyte progenitors are sufficient to completely deplete the subjacent tissue of O2 [32, 33] (Table 1). Megakaryocytes and adipocytes, although functionally important, are minor populations within the BM. We have previously shown that mobilization induced by G-CSF or CY leads to a dramatic expansion of myeloid progenitors [5, 34], which have high O2 consumption rates. Thus, we hypothesized that O2 depletion may be increased in mobilized BM. Taking published O2 consumptions rates, we calculated the total O2 consumption rate in the femoral BM from a typical 8–10-week-old BALB/c mouse prior to and at the peak of G-CSF-induced mobilization (Table 2). We measured the frequency of the most abundant cell types encountered in mobilized and nonmobilized BM, that is, mature CD11b+ Gr-1+ granulocytes, immature CD11b+ Gr-1− granulocyte/monocyte progenitors, and CD11b− Gr-1− lymphoid cells and nonmyeloid progenitor cells (Table 2). With an average number of BM cells of 2 × 107 per femur for 8–10-week-old mice, we calculated that mobilized BM should consume twice as much O2 as steady-state BM (Table 2), which, according to the Krogh's model, should result in enhanced levels of hypoxia, particularly in areas distal from the endothelial sinuses that supply oxygen to the BM hematopoietic tissue.
Table Table 1.. O2 consumption rates of bone marrow cells
Table Table 2.. Estimation of O2 consumption in steady-state versus mobilized BM
Increased Incorporation of Pimonidazole in Mobilized BM
Pimonidazole hydrochloride is a nontoxic, water-soluble hypoxia marker that covalently binds to protein thiol groups when O2 tension is below 10 mmHg or less than 1.3% (O2 tension is 159 mmHg in air at sea level, corresponding to 21% of air). BALB/c mice were mobilized by injecting G-CSF twice daily for 4 consecutive days, and pimonidazole was injected 3 hours prior to sacrifice. Steady-state mice were injected with saline instead of G-CSF and given pimonidazole prior to sacrifice. Negative controls for immunostaining were not injected with pimonidazole. Mice were then anesthetized, and hind limbs were fixed by perfusion. Following demineralization with EDTA, paraffin-embedded femur sections were stained with a monoclonal antibody specific for pimonidazole and analyzed by confocal laser scanning microscopy (Fig. 1). In BM from saline-injected mice, pimonidazole was preferentially incorporated immediately adjacent to the endosteum and rapidly decreased within the first 50 μm from the bone-BM interface (Fig. 1A). The pimonidazole staining was specific, as staining with the anti-pimonidazole mAb on mice that were not injected with pimonidazole was negative (Fig. 1B). Thus, these stainings clearly demonstrate that the HSC niche at the endosteum is hypoxic, with less than 10 mmHg or 1.3% O2 in steady-state conditions. This is in agreement with a recent report showing that most HSC in the Hoechst 33342 side population incorporate pimonidazole at higher levels than non-side populations in vivo .
To document the distribution of hypoxia at the peak of mobilization induced by G-CSF or CY (supplemental online Fig. 2), mice were injected with pimonidazole at day 4 of G-CSF treatment or on day 7 following a single injection of CY. At day 4 of mobilization with G-CSF, the hypoxic area was considerably increased, occupying a large fraction of the BM cavity and forming a gradient from the endosteum to the central BM (Fig. 1C, 1D). Quantification of fluorescence-positive areas revealed that pimonidazole staining was significantly increased in both endosteal (8-fold) and central BM (5-fold) regions (Fig. 2F).
In mice mobilized with CY, a fivefold increase in pimonidazole staining was also measured in the central BM region, although the staining was not as marked in the endosteal region as in saline-treated and G-CSF-treated mice (Fig. 2E, 2F). Taken together, these data demonstrate that G-CSF and CY, which both cause a dramatic expansion of the granulocytic pool in the BM, also render the BM hematopoietic microenvironment more hypoxic, particularly in the central region of the BM.
HIF-1α Protein Is Increased in Mobilized BM
To further document that hypoxia increases in the BM during mobilization, femurs from BALB/c male mice were flushed at different time points of mobilization induced by G-CSF or CY with urea/SDS cell lysis buffer to extract nuclei with whole-cell content. BM protein extracts from three separate mice per time point were combined, and equal amounts of proteins were loaded in each lane to be Western blotted with a mouse anti-human HIF-1α monoclonal antibody cross-reacting with mouse HIF-1α and analyzed quantitatively using the Odyssey Infra-Red system  (Fig. 2A). Total protein extracts from mouse myeloid FDCP-1 cells and human myeloid MO7e cells cultured for 6 hours in normoxic or hypoxic conditions were used as controls for HIF-1α expression. The specificity of the antibody was demonstrated by the induction of a 120-kDa band in FDCP-1 cells and a 125-kDa band in MO7e cells in hypoxic conditions, corresponding to the apparent molecular masses of mouse and human HIF-1α, respectively. Quantitative analysis of pooled mouse BM cell extracts clearly showed an increase in HIF-1α protein levels between day 2 and day 6 of G-CSF administration and at days 6 and 8 following CY administration, corresponding to the peak of HSPC mobilization into the peripheral blood, reaching levels 20–30-fold higher than those observed in the BM of saline-injected mice. To estimate the variability within each group, the experiment was repeated by loading BM cell extracts from three separate mice for each time point following G-CSF administration (Fig. 2C). Quantification of the three Western blots required to run all the cell extracts confirmed that HIF-1α protein is increased in BM cell extracts as soon as day 2 of G-CSF administration.
To confirm these data, experiments were repeated in male 129SvJ mice using a rabbit anti-mouse HIF-1α polyclonal antibody (Fig. 3). The polyclonal antibody was more sensitive than the monoclonal and confirmed in a different mouse genetic background the significant increase in HIF-1α protein in BM cell extracts at day 4 of mobilization with G-CSF (p = .0001).
Western blotting results were confirmed by immunohistofluorescence on femur sections from mobilized and steady-state BALB/c mice. In saline-injected mice, HIF-1α protein was expressed at the endosteum similarly to pimonidazole staining (Figs. 1A, 4A). At day 4 of G-CSF-induced mobilization, HIF-1α expression was markedly increased throughout the BM (Fig. 4A, 4B, 4F). At day 7 of CY-induced mobilization, HIF-1α protein level was also increased throughout the BM (Fig. 4C, 4F), including in the endosteal region, in which hypoxia was not increased (Fig. 1C, 1F).
Mobilizing Cytokines Have Little Effect on HIF-1α Stabilization in Cultured BM Cells
TPO has recently been described to stabilize and enhance HIF-1α protein in the TPO-dependant human myeloid cell line UT-7/TPO and in mouse Gr1− Sca-1+ c-KIT+ HSPC cultured in normoxic conditions . To determine whether mobilizing cytokines could stabilize HIF-1α in whole BM cells independently of hypoxia, total BM cells isolated from C57BL/6 mice were cultured for 24 hours in the presence of recombinant human G-CSF, rat KIT ligand, or human TPO in normoxic or hypoxic conditions. Importantly, BM cells cultured for 24 hours were 100% viable in all conditions, including in the absence of cytokines. Total cell lysates were then analyzed by Western blot with the rabbit anti-mouse HIF-1α antibody (Fig. 5). G-CSF and KIT ligand in normoxic conditions were both able to induce accumulation of low but significant amounts of HIF-1α protein compared with controls without cytokines, whereas TPO had a barely detectable effect. This experiment also clearly showed that (a) hypoxia induces HIF-1α to a much higher extent than cytokines in normoxic conditions, and (b) in hypoxic conditions, the effect of additional cytokines on HIF-1α protein expression was marginal. Thus, the increased HIF-1α protein levels observed in mobilized BM in vivo are mostly due to increased hypoxia.
VEGF-A Expression in the BM Is Increased During HSPC Mobilization
VEGF-A expression is strongly induced by hypoxia and HIF-1α , as the VEGF-A gene contains several HRE in its 5′-untranslated flanking region. VEGF-A mRNA was analyzed by qRT-PCR. Total RNA was isolated from both the central marrow and the endosteal region. For this purpose, femurs were extensively flushed with PBS to remove central BM cells and then flushed with 1 ml of Trizol to extract RNA from the remaining endosteal cells adherent to the bone surface in the BM cavity. This technique enables a 30–100 times enrichment in mRNAs specific for osteoblasts (e.g., osteocalcin) and osteoclasts (e.g., cathepsin K) (Fig. 6A). Similarly, the chemokine CXCL12/SDF-1, which is mostly produced by osteoblasts in the BM , was 10 times more abundant in the endosteal extracts than in central BM extracts. Importantly, not all mRNAs tested were enriched in the endosteal extracts, as mRNA for the cytoskeleton protein vimentin remained similar in both endosteal and central BM extracts. Taken together, these results show that our extraction techniques successfully enriched for RNA from endosteal cells.
Using this technique, we isolated RNA from the central BM and endosteal region at different time points of G-CSF-induced mobilization. We found that levels of VEGF-A transcripts in the endosteal region gradually increased, peaking at four times the basal levels on day 6 of G-CSF administration (Fig. 6B) and persisting until at least 4 days after cessation of G-CSF administration (day 10) before returning to basal levels. In the central BM, VEGF-A mRNA levels doubled on day 4 of G-CSF administration and then rapidly returned to basal levels. Taken together, these data show that VEGF-A expression is induced in the BM during HSPC mobilization, with a more marked and persistent increase in the endosteal region compared with the central BM.
VEGF-A was then followed at the protein level by immunohistofluorescence. Mobilization with either G-CSF or CY resulted in a marked increased of VEGF-A staining on BM endothelial sinuses, whereas BM sinuses stained for VEGF-A were undetectable in nonmobilized mice injected with saline (Fig. 7A, 7C, 7E). The VEGF-A staining was specific, as shown by the absence of staining with a rabbit control antibody on successive serial sections (Fig. 7B, 7D, 7F).
Hypoxia plays a key role in the regulation of specific immunity , inflammation [17, 19, 20], and hematopoiesis. In vivo, prolonged hypoxia increases erythropoiesis but inhibits megakaryocytopoiesis , an effect due to increased erythropoietin production in response to hypoxia. In vitro, culturing HSC in hypoxic conditions (1% O2) maintains their reconstituting potential, which is rapidly lost in normoxic conditions [25, –27]. In addition, the conditional deletion of genes whose products are components of HIF transcription factors, such as ARNT or HIF-2α, compromises HSC potential [28, 29]. Furthermore, HIFs regulate a number of genes important for the survival, proliferation, and trafficking of HSPC, such as CXCR4 , CXCL12 , β2 integrin chain , and VEGF-A . Thus, a hypoxic microenvironment is critical for HSC maintenance and function. Despite the important role of hypoxia-responsive genes in hematopoiesis, O2 distribution in the BM has been poorly investigated. Oxygen concentration is highest in lung alveoli and arterial blood, where it is approximately 30% lower than in air (90–100 mmHg). In tissue capillaries, where O2 diffuses through tissues and is exchanged with CO2, O2 drops to 20 mmHg, whereas efferent venous blood returning to the lung contains 40 mmHg O2 . Direct measurements of O2 concentration show that O2 tension can decrease to 5 mmHg in ischemic tissues in the mouse , whereas the mean O2 tension in BM aspirates from healthy humans is 54 mmHg . However, these studies provide a global value for a relatively large volume of tissue and do not take into account the fact that O2 spatial distribution in solid tissues is heterogeneous, with higher levels in close proximity to blood vessels decreasing rapidly with the distance from the blood vessel and the number of cell layers that O2 has to cross [31, –33]. Our study is the first to reveal O2 spatial distribution in the BM in vivo. In steady-state conditions, incorporation of pimonidazole, which cross-links to protein adducts at O2 tension below 10 mmHg (corresponding to less than 1.3% O2), was high along the endosteum and decreased rapidly within the first 50 μm of the BM stroma toward the center of the BM, a staining pattern similar to that of HIF-1α protein. This clearly demonstrates that the endosteal HSC niche [46, , , , –51] is hypoxic, with oxygen tension below 10 mmHg. This is consistent with the recent findings that (a) cells located in BM areas away from BM endothelial sinuses are enriched in cobblestone area-forming cells compared with BM cells located in proximity of sinuses, (b) HSC comprised in the high Hoechst 33342 dye efflux side population bind pimonidazole at much higher levels in vivo than other BM cells, and (c) HSC from the side population are selectively depleted in vivo following administration of tirapazamine, a compound that, once reduced in hypoxic conditions, breaks double-stranded DNA .
In contrast to the endosteum, the central BM containing putative HSC endothelial niches  was not hypoxic in steady-state conditions, with low levels of HIF-1α protein, probably because of the proximity to endothelial sinuses, where oxygenated blood circulates. Thus, our observation supports the notion that the endosteal and endothelial HSC niches may have distinct functions, with the hypoxic endosteal niche maintaining HSC in an immature and relatively quiescent state requiring low oxygen tension, whereas the more oxygenated endothelial niche could be a more proliferative niche where maintenance of stem cell-ness is less critical .
Our study is also the first to reveal that at the peak of HSPC mobilization induced by either G-CSF or CY, the proportion of the BM tissue becoming hypoxic dramatically increases to expand through most of the femoral BM cavity. There was, however, a noticeable difference in the distribution of pimonidazole staining following G-CSF and CY treatments. In G-CSF-mobilized mice, hypoxia was increased in both endosteal and central regions of the BM cavity, whereas in CY-mobilized mice, hypoxia was increased in the central BM but not in the endosteal region. The reason for this different distribution of hypoxia in response to G-CSF and CY treatments remains unclear. However, it must be noted that during the myeloablative phase that follows CY injection, the continuity of BM sinuses is disrupted, with the formation of large gaps between adjacent endothelial cells  enabling the diffusion of large macromolecules, such as α2-macroglobulin, from the blood into the BM stroma . Thus, CY treatment may lead to a rapid loss of hypoxic areas in the first days following injection and hence to a loss of hypoxia in the endosteal region. Hypoxia could later be re-established by proliferating myeloid progenitors once the integrity of the endothelial barrier is recovered during the rebound phase. If myeloid progenitors expand preferentially in the central BM, then hypoxia would be more pronounced in this region.
In G-CSF-mobilized mice, HIF-1α protein levels increased throughout the BM in a pattern similar to that of pimonidazole staining (Figs. 1, 4), suggesting that in G-CSF-mobilized mice, most of HIF-1α stabilization is due to increased hypoxia. However, in CY-mobilized mice, HIF-1α protein levels were increased in both endosteal and central BM regions, suggesting that, at least in the endosteal region, HIF-1α stabilization occurs independently of hypoxia. The facts that the BM stroma overexpresses cytokines such as KIT ligand in response to myeloablation induced by CY [55, 56] and that KIT ligand also stabilizes HIF-1α independent of hypoxia (albeit at much lower levels than hypoxia [Fig. 5]), indicate that increased HIF-1α protein levels in nonhypoxic areas of CY-mobilized BM may also involve supraphysiological release of cytokines by the myeloablated BM.
We also report that VEGF-A transcripts also increase in the BM during mobilization, with a greater increase in the endosteal region. However, at the protein level, we could detect a marked increase in VEGF-A protein only on BM endothelial sinuses. This disparity in the spatial distribution between transcript expression and protein accumulation may be due to the accumulation of VEGF-A protein at the surface of endothelial cells or to its internalization by them. Indeed, several cytokines, such as CXCL12 [57, 58] or basic fibroblast growth factor , are known to be captured by ECM proteoglycans and accumulate at the periphery of endothelial cells . Similarly, VEGF-A can also be internalized by cells expressing high levels of VEGF receptors, particularly endothelial cells [61, 62]. Therefore, it is possible that the relatively low sensitivity of immunofluorescence detected only that VEGF-A that had accumulated in or at the surface of endothelial cells.
Since VEGF-A gene transcription is known to be induced by hypoxia and HIF-1α , it is tempting to speculate that the hypoxia and HIF-1α protein accumulation we observed in mobilized BM is responsible for increased VEGF-A transcription in mobilized BM. In addition, VEGF-A transcription has recently been described to be directly induced by G-CSF in Gr-1+ granulocytes in vitro , suggesting that VEGF-A transcription could be directly activated by G-CSF in granulocytes in normoxic areas of the BM.
The fact that VEGF-A transcript increases in mobilized BM with marked accumulation of the protein on BM endothelial sinuses suggests that VEGF-A could play an important role in HSPC mobilization induced by G-CSF and CY. Sustained levels of VEGF-A protein in the circulation mobilize both HSPC and endothelial progenitor cells . VEGF-A induces rapid vasodilatation and vascular hyperpermeability in vivo [65, 66]. Increased vascular permeability and blood flow in the BM could be important contributors to HSPC egress from the BM stroma into the blood. Interestingly, in mice deficient for endothelial nitric oxide synthase (Nos3), mobilization and increased vascular permeability in response to VEGF-A are both compromised . This effect of Nos3 deletion is due to nonhematopoietic cells, as lethally irradiated Nos3−/− recipients reconstituted with wild-type BM cells have impaired mobilization, whereas wild-type recipients reconstituted with Nos3−/− BM cells mobilize normally in response to VEGF-A . Therefore, Nos3 and cells that express it, such as endothelial cells, play a critical role in VEGF-induced HSPC mobilization. However the effect of Nos3 gene deletion on G-CSF-induced HSPC mobilization remains to be determined. Therefore, to definitively ascertain the role of VEGF-A in G-CSF-induced and CY-induced mobilization, experiments with inducible deletion of the VEGF-A gene or using specific antagonists of VEGF receptors are now required.
In conclusion, we have found that the BM hematopoietic microenvironment becomes highly hypoxic during mobilization of HSPC induced by G-CSF or CY, with accumulation of transcription factor HIF-1α probably caused by the stabilization of the protein in response to increased hypoxia. In turn, higher levels of HIF-1α lead to increased VEGF-A gene transcription and accumulation of VEGF-A protein around BM endothelial sinuses, a cytokine known to induce vasodilatation, vascular permeability, and HSPC mobilization, when administered in vivo. Therefore, oxygen depletion by proliferating myeloid progenitors in the BM may play an important role in HSPC mobilization.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.
This work was supported by Grant 2888701 from the National Health and Medical Research Council of Australia (J.-P.L. and I.G.W.), Senior Research Fellowship from the Queensland Cancer Fund (J.-P.L.), and Grant PO15 from the Australian Stem Cell Centre (J.-P.L. and S.K.N.). J.P.L. is a Senior Research Fellow of the Queensland Cancer Fund.