Glucocorticoid-induced eosinopenia results from CXCR4-dependent bone marrow migration

Glucocorticoids are considered first-line therapy in a variety of eosinophilic disorders.¹  Although the therapy is often effective, prolonged glucocorticoid use is associated with substantial toxicity.²  Our understanding of the mechanism of action of glucocorticoids in eosinophilic diseases remains very limited and has prevented the development of therapeutic alternatives that are equally efficacious but less toxic.

The first evidence that glucocorticoids could influence human eosinophil biology predated their now widespread clinical use as anti-inflammatory and immunosuppressive agents. In the year preceding the first therapeutic use of glucocorticoids, it was noted that the administration of either exogenous adrenocorticotropic hormone or hydrocortisone (then known as Kendall’s compound F) led to a profound, transient drop in circulating human eosinophils.³  The rapid eosinopenic response to exogenous adrenocorticotropic hormone was found to be so reproducible that the hormone was historically used in a clinical test of adrenocortical function, the Thorn test.⁴  The mechanism by which this phenomenon occurs has intrigued clinicians since that time and remains unexplained.

In principle, the rapid clearance of eosinophils from the peripheral circulation after glucocorticoid administration could result from changes in the expression of surface molecules on eosinophils, endothelial cells, or cells of the mononuclear phagocyte system; from an increase in the secretion of eosinophil chemoattractants in tissues; or from a combination of these factors. The kinetics of the response, with onset between 60 and 120 minutes after glucocorticoid administration,⁵  is not compatible with a decreased rate of myelopoiesis or eosinophil release from the bone marrow as a likely mechanism. In the earliest studies, intravascular lysis of eosinophils was considered but not supported by analyses of human blood exposed ex vivo to a glucocorticoid, and a normal eosinopenic response to glucocorticoids in splenectomized humans⁶  suggested that splenic sequestration is not the only mechanism involved.

Rodent models have not offered a convincing explanation for glucocorticoid-induced eosinopenia. The phenomenon has been documented in rats and guinea pigs, but histologic studies in those species have yielded inconclusive results.^6-8  The recent findings of only modest levels of overlap in the glucocorticoid receptor binding sites identified by chromatin immunoprecipitation sequencing in different mouse or human cell lines,^9,10  and evidence that the transcriptional response to glucocorticoids is highly cell type dependent,¹¹  highlight the need for careful species model selection and a focus on specific cell types when trying to understand the effects of glucocorticoids in humans.

Studies of the effects of glucocorticoids on molecules that promote eosinophil adhesion^12-14  or chemoattraction^15-18  provide important clues about the mechanisms by which glucocorticoids could help decrease eosinophil recruitment to sites of inflammation, but they do not provide an explanation for the rapid clearance of eosinophils from the peripheral circulation.

Our recent study of the genome-wide transcriptional response of human eosinophils to glucocorticoids after in vivo treatment suggested that glucocorticoids induce a proapoptotic transcriptional program, although eosinophil apoptosis was not observed before migration out of the circulation.⁵  The same study revealed a consistent increase in expression of the chemokine receptor CXCR4, at the RNA and protein levels, between 60 and 120 minutes after glucocorticoid administration, which coincided with the timing of eosinophil egress from the circulation. This finding led us to hypothesize that the mechanism of glucocorticoid-induced eosinopenia involves CXCR4 upregulation, followed by migration of eosinophils to tissues rich in the CXCR4 ligand CXCL12, such as the bone marrow. To test this hypothesis, we first studied the response of circulating eosinophils to glucocorticoid administration in humans and 2 animal models and identified the rhesus macaque (Macaca mulatta) as an appropriate model for glucocorticoid-induced eosinopenia. We then developed a method of purification of rhesus macaque eosinophils from peripheral blood by immunomagnetic negative selection and performed live tracking of radiolabeled eosinophils before and after in vivo glucocorticoid administration. For this, we used an ex vivo cell-labeling technique with zirconium-89 (⁸⁹Zr)-oxine for positron emission tomography (PET) imaging that was recently developed by our group.^19,20  Taking advantage of the high sensitivity and quantitative capability of PET and the long half-life of ⁸⁹Zr (3.3 days), this method enables longitudinal tracking and quantitative measurement of cell distributions in vivo, with negligible levels of radioactivity.^19,20  Finally, we tested whether selective CXCR4 blockade would be sufficient to reduce the rapid egress of eosinophils from the peripheral circulation after glucocorticoid administration.

Twenty healthy volunteers (10 men, 10 women; mean age, 37 years) were enrolled under National Institutes of Health (NIH) protocol 16-I-0126, which was approved by the Institutional Review Board of the National Institute of Allergy and Infectious Diseases (NIAID). Written informed consent was obtained. Each participant received a single IV dose of 250 mg methylprednisolone (Solu-Medrol; Pfizer) over 30 minutes. A complete blood count (CBC) was performed before and 4 hours after the start of infusion.

B6.SJL mice were obtained from Taconic Biosciences [B6.SJL Cd45a(Ly5a)/Nai] and were handled in accordance with protocols approved by the NIAID Institutional Animal Care and Use Committee. Mice were randomly assigned to 1 of 2 groups. One group (n = 5) received methylprednisolone (5 µg/g, IV). Another group (n = 5) received normal saline (0.9% sodium chloride, IV). A CBC was performed 4 hours after injection.

Rhesus macaques were housed, handled, and underwent procedures conforming to a protocol approved by the Institutional Animal Care and Use Committee of the National Heart, Lung, and Blood Institute. Supplemental Table 1 (available on the Blood Web site) summarizes the information on each rhesus macaque. Apheresis was performed on a Fenwal CS-3000 cell separator (Baxter Healthcare).²¹  After apheresis, blood (40-44 mL) was diluted 1:2 with phosphate-buffered saline (PBS), layered over Ficoll-Paque Plus (GE Healthcare Life Sciences), and centrifuged (700g for 30 minutes at 25°C without brake). The granulocyte layer was transferred to new tubes and red blood cells (RBCs) were hypotonically lysed with 30 mL ice-cold double-distilled water. Lysis was stopped with 10 mL 4× PBS. After centrifugation (170g for 10 minutes at 4°C), hypotonic lysis was repeated. Pooled granulocytes were washed and resuspended to a concentration of 10 to 12 million cells/mL in wash buffer (PBS, 1% bovine serum albumin, and 1 mM EDTA) with 1.1% human FcR Blocking Reagent (Miltenyi Biotec). After 10 minutes at 4°C, the cells were labeled with anti-human CD64-PE antibody followed by anti-PE MicroBeads UltraPure (Miltenyi Biotec) for negative selection on an autoMACS Pro Separator (Miltenyi Biotec). Eosinophil purity was measured by counting ≥300 cells on cytospin preparations after staining with Kwik-Diff (Thermo Fisher Scientific).

Membrane phenotyping of blood eosinophils from humans and rhesus macaques was performed on fresh whole blood. Aliquots of blood (100 µL) were incubated with antibodies, and the RBCs were lysed with 1 mL lysis buffer (BD Biosciences). Samples were then washed with 1 mL PBS with 1% bovine serum albumin (fluorescence-activated cell sorting [FACS] buffer), centrifuged (300g for 5 minutes at 4°C), and resuspended in FACS buffer. When indicated, 7-amino actinomycin D (BD Biosciences) was added 10 minutes before flow cytometry.

Intracellular staining for eosinophil peroxidase (EPX) was performed on rhesus granulocytes obtained by Ficoll separation of blood and RBC lysis. The cells were incubated in FACS buffer with 10% mouse serum for 10 minutes on ice, then with anti-human VLA4-PE or anti-human CD64-PE. Samples were washed twice in PBS, then fixed with PBS with 1% paraformaldehyde (EM Sciences), for 10 minutes. After they were washed in FACS buffer with centrifugation (400g for 7 minutes at 4°C), the samples were permeabilized for 10 minutes with 100 μL FACS buffer with 0.1% saponin (IC FACS buffer; Sigma-Aldrich). The samples were washed with IC FACS buffer, blocked again for 10 minutes with 10% mouse serum, and stained with an anti-human EPX-Alexa700 antibody in IC FACS buffer.

To assess viability with or without ⁸⁹Zr-oxine labeling, we incubated the cells with annexin V-FITC and propidium iodide in 1× binding buffer for 15 minutes at room temperature. Cell activation was assessed by staining with an anti-human CD69 antibody.

Acquisition was performed on an LSRII flow cytometer (BD Biosciences). Data were analyzed with FlowJo, v9.9.6 or v10.4.2.

Chemotaxis was assessed in Costar 24-well Transwell plates with 5.0-µm pore inserts (Corning). Cell culture medium (500 µL), with or without 100 ng/mL recombinant rhesus CCL11 (eotaxin; MyBiosource), was added to the lower chamber. Eosinophils (3 × 10⁵ cells/100 µL culture medium) were added to the upper chamber and incubated for 2 hours at 37°C with 5% CO₂. Cells collected separately from the upper or lower chambers were quantified on an LSRII flow cytometer with a 180-second acquisition interval at high speed and constant flow. The proportion of migrating eosinophils was defined as the number of cells in the lower chamber over the sum of the cells in the upper and lower chambers. Data were analyzed with FlowJo, v9.9.6.

⁸⁹Zr-oxalate solution was produced at the institutional cyclotron facility and converted to ⁸⁹ZrCl₄.²²  The ⁸⁹Zr-oxine complex was generated as previously described.¹⁹  Purified eosinophils in PBS were incubated with ⁸⁹Zr-oxine at 61.7 to 98.7 kBq/10⁶ cells in a 30:1 volume ratio for 15 minutes at room temperature and washed 3 times by centrifugation at 170g for 10 minutes at 4°C in PBS with 1% autologous rhesus serum. The activity used for labeling was determined based on previous studies,^19,20  to achieve doses that would not be radiotoxic to the cells (∼37 kBq/10⁶ cells), with a labeling efficiency of ∼30% of the incubated activity.

Deferoxamine was infused through a central catheter at an initial dose of 500 mg for 2 to 4 hours, followed by 2 to 3 mg/kg per hour, for the duration of the imaging study. Autologous ⁸⁹Zr-oxine-labeled eosinophils were transferred to rhesus macaques IV (10.5 × 10⁶ to 97.3 × 10⁶ cells; 35.5 ± 8.0 kBq/10⁶ cells, at 4.5 × 10⁶ ± 3.1 × 10⁶ cells/146.0 ± 81.5 kBq/kg) and PET/computed tomography (CT) imaging was serially acquired on a GE Discovery MI DR PET/CT system. On day 0, imaging was performed immediately after cell transfer (10-minute time point, corresponding to the midpoint of a 20-minute imaging period). To evaluate the kinetics of eosinophil passage through the lungs after reinjection, day 0 imaging was performed again 1 hour after cell transfer in 4 of the animals (supplemental Table 1). On day 1, imaging was performed at baseline (BL), then hourly for 4 hours. In the glucocorticoid group, 4 mg/kg methylprednisolone was injected 30 minutes after BL imaging. In the control group, no injection was performed. On day 2, a single image was obtained. PET images were reconstructed using GE’s Q CLEAR algorithm. VivoQuant (build 3.5; InVicro) was used to merge maximum intensity projection (MIP) PET images and CT. For quantification of ⁸⁹Zr activity, volumes of interest were drawn on the acquired images, delineating the major organs with MIM Software v6.7.13; activity concentrations were then extracted and corrected for decay. For bone marrow, ⁸⁹Zr activity in the trunk, humeri, and proximal half of femurs was quantified to keep the measured range consistent among PET scans and animals. The percentage of injected dose (% ID) was calculated by dividing the organ activity by the injected dose.

For quantification of ⁸⁹Zr activity in blood, it is not possible to draw volumes of interest by delineating the entire vascular compartment, because most of the blood is expected to be in small vessels and capillaries. Therefore, we drew volumes of interest on the proximal aorta and inferior vena cava, to extract activity concentrations, and used the estimated blood volume of 60 mL/kg²³  to calculate the blood % ID.

Each animal received plerixafor (1 mg/kg, subcutaneously; Sigma), glucocorticoid (methylprednisolone, 4 mg/kg, IV), or plerixafor and glucocorticoid. In the plerixafor experiment, a CBC was performed at 0, 2, 4, 6, and 8 hours after treatment. In the glucocorticoid experiment, a CBC was performed at 0, 2, 4, and 6 hours after treatment. At each time point, CXCR4 expression on blood eosinophils was quantified by flow cytometry. In the combination experiment, glucocorticoid was given 6 hours after plerixafor injection, when the absolute eosinophil counts (AECs) had plateaued (supplemental Figure 6). CBCs were performed before (at BL) and 2 hours after glucocorticoid treatment. When the same animal was studied in more than 1 regimen, at least 2 weeks of washout time was given between experiments.

Statistical comparisons of sample means in 2 independent groups of subjects were performed by 2-sample Student t tests. Statistical comparisons of sample means within the same group of subjects were performed by paired Student t tests.

To study the longitudinal trend of eosinophil localization in the different organs in glucocorticoid-treated or control rhesus macaques, linear regression was performed over the 3 macaques in the glucocorticoid arm and over the 3 macaques in the control arm, separately. In both regressions, the response was the organ ⁸⁹Zr activity (expressed as a decay-corrected percentage of the injected dose) observed from BL to 4 hours; the regressors included the type of organ, the observation time, and their interaction. P values were calculated by the Wald test, which indicates whether a slope is significantly different from 0.

Additional details and product catalog numbers may be found in the supplemental Methods and in supplemental Table 2.

We began by comparing the kinetics of the response of eosinophils and other major blood cell types to the administration of glucocorticoids in healthy human volunteers and in 2 animal models: rhesus macaque and mouse (Mus musculus). In the human volunteers (n = 20), who received an IV dose of 250 mg of methylprednisolone, we observed the expected change in cell composition: the proportion of neutrophils increased, whereas the proportions of lymphocytes, monocytes, eosinophils, and basophils decreased (Figure 1A, left; supplemental Figure 1). The decline in the relative and absolute counts of circulating eosinophils was reproducible across subjects and was statistically significant (Figure 1B, left). These findings were very similar to those observed in the adult rhesus macaques (n = 4), which received an IV dose of 4 mg/kg methylprednisolone. The similarities between the rhesus model and humans included the overall cell composition of peripheral blood (Figure 1A, center), as well as the relative and absolute counts of circulating eosinophils (Figure 1B, center), at BL and in response to the glucocorticoid. The results observed in rhesus and humans were different from those observed in the mice, which received an IV dose of 5 µg/g of methylprednisolone (n = 5) or vehicle (n = 5). In the mouse model, the relative increase in neutrophils in response to glucocorticoids was much less pronounced (Figure 1A, right) and the decline in blood eosinophils was not observed (Figure 1B, right). We concluded that the rhesus macaque would be a more appropriate model for the study of glucocorticoid-induced eosinopenia.

As an initial step toward live cell tracking, we needed a method of purification of circulating rhesus eosinophils. Circulating rhesus neutrophils do not express CD16,²⁴  which is the major surface marker used to discriminate human neutrophils from eosinophils. Siglec-8, a surface receptor that serves as a marker for human eosinophils, is not expressed on rhesus eosinophils.²⁵  We found that the integrin VLA4 was expressed on human and rhesus blood eosinophils and that CD64 was expressed on rhesus neutrophils but not on eosinophils (Figure 2A). To confirm that CD64-negative rhesus granulocytes were indeed eosinophils, we studied granulocyte pellets after Ficoll centrifugation. VLA4⁺ rhesus granulocytes were also positive for the eosinophil-specific intracellular marker EPX, and the same was true of the CD64⁻ fraction (Figure 2B).

Immunomagnetic negative selection of rhesus granulocytes with anti-CD64 antibodies yielded a cell population with the classic histologic features of eosinophils on light microscopy (Figure 2C). No signs of eosinophil degranulation were observed on cytospin preparations. The eosinophil purity of the CD64⁻ fraction after negative selection, as measured by manual counting on cytospin slides, had a mean of 89.7% and a range of 82.6% to 96.0%. The main contaminating cell types were lymphocytes (mean = 4.8%) and neutrophils (mean = 2.1%) (Figure 2D).

Purified rhesus eosinophils were then subjected to ⁸⁹Zr-oxine labeling at 35.5 ± 8.0 kBq/10⁶ cells. Radiolabeling with ⁸⁹Zr-oxine did not induce significant changes in the proportion of cells that underwent apoptosis (Figure 2E), the proportion of nonviable cells (Figure 2F), the proportion of activated cells (Figure 2G), or the functional response of the cells to a chemotactic stimulus (Figure 2H).

To directly address the question of where eosinophils migrate when they leave the circulation after glucocorticoid administration, we transferred autologous ⁸⁹Zr-oxine-labeled rhesus eosinophils and performed serial PET/CT imaging. Because ⁸⁹Zr is released from dead cells, deferoxamine was infused throughout the imaging period to chelate free ⁸⁹Zr for renal clearance and thus allow for serial imaging of live cells. Figure 3A summarizes the experimental design.

We first assessed the pattern of eosinophil distribution after reinjection, in the absence of glucocorticoid administration. Shortly after IV transfer, eosinophils migrated primarily to the lungs. By 1 hour, the cells had transited through the lungs, and ∼22% had migrated to the bone marrow, 19% to the liver, and 5% to the spleen, where they remained in similar proportions 1 and 2 days after reinjection (Figure 3B; supplemental Figure 2).

To test the effect of glucocorticoid administration, we chose day 1 after cell transfer, when the cells had reached a steady state of distribution. We administered methylprednisolone (4 mg/kg, IV) to 3 unrelated adult rhesus macaques and performed serial PET imaging hourly for 4 hours after BL imaging. As a negative control, we followed the same imaging regimen on 3 adult rhesus macaques without methylprednisolone administration (Figure 3A). Within 4 hours of glucocorticoid administration, we observed the expected drop in the relative counts of circulating eosinophils, as well as a decrease in cell-associated ⁸⁹Zr activity in peripheral blood (supplemental Figure 3). At the same time, ⁸⁹Zr signals in the bone marrow had increased, with the signals often filling the bone marrow area, suggesting that glucocorticoids had induced rapid bone marrow homing of eosinophils (Figure 3C-D; supplemental Video 1). We then quantified the ⁸⁹Zr activity over time in the 4 organs in which ⁸⁹Zr signal had been detected: bone marrow, liver, lungs, and spleen. As expected, in untreated animals, the distribution of ⁸⁹Zr remained stable in the 4 organs (Figure 3E, left). In contrast, in glucocorticoid-treated animals, the distribution of ⁸⁹Zr in the bone marrow increased significantly after glucocorticoid administration (slope = 1.58; P = .038; mean increase = 5.6% of injected dose), whereas the distribution in the other organs showed no significant change (Figure 3E, right). In the same time interval, the estimated reduction in ⁸⁹Zr activity in the blood (mean reduction = 6.1% of injected dose; supplemental Figure 4) corresponded approximately to the observed increase in signal in the bone marrow, indicating that the observed change in the bone marrow was sufficient to explain the depletion of radiolabeled eosinophils from the blood.

We had previously found that glucocorticoids induce expression of the chemokine receptor CXCR4 in human eosinophils between 60 and 120 minutes after medication administration, which coincides with the timing of eosinophil egress from the circulation.⁵  The human and rhesus macaque genomes have a fully conserved glucocorticoid response element (GRE) near the CXCR4 transcription start site (Figure 4A). This specific recognition motif is known to be absent from the mouse genome at the Cxcr4 locus, and in vitro treatment of mouse hematopoietic stem or progenitor cells with a glucocorticoid has been shown not to upregulate Cxcr4 at the transcript or protein levels.²⁶  A sequence homology search revealed that the human and rhesus GRE immediately upstream of CXCR4 is partially conserved in guinea pigs and absent in the mouse or rat genomes, although a search for matches of the GR position weight matrix (JASPAR²⁷  Core 2018: MA0113.3) with the program PWMScan²⁸  revealed other predicted GRE sequences upstream of Cxcr4 in the mouse and rat genomes.

Given the high level of conservation between the rhesus and human genomes at the CXCR4 locus, including the presence of a fully conserved GRE, we then tested whether glucocorticoids lead to an increase in CXCR4 expression on the surface of rhesus eosinophils and whether the kinetics of this effect are consistent with those of glucocorticoid-induced eosinopenia, as had been observed in humans. We administered methylprednisolone (4 mg/kg, IV) to 3 unrelated adult rhesus macaques and measured circulating eosinophil counts and CXCR4 expression on eosinophils serially, before and after each administration. We found that the kinetics of glucocorticoid-induced eosinopenia was similar between rhesus and human, with the greatest effect occurring in the first 2 hours after administration of the medication. Surface expression of CXCR4 in circulating rhesus eosinophils increased as the eosinophil count dropped (Figure 4B; supplemental Figure 5), consistent with our prior observations in humans.⁵ 

To test whether glucocorticoid-induced eosinopenia was CXCR4-dependent, we then administered methylprednisolone (4 mg/kg, IV) to 3 unrelated adult rhesus macaques, in the presence or absence of the CXCR4 blocker plerixafor, which inhibits the interaction between CXCR4 and its canonical ligand, CXCL12 (Figure 4C). Because plerixafor is known to mobilize cells from the bone marrow, we first studied the effect of plerixafor alone on circulating eosinophil counts over time. Plerixafor led to an increase in circulating rhesus eosinophils, which stabilized after 6 hours (supplemental Figure 6). Based on this observation, we chose 6 hours as the most appropriate time interval between plerixafor and glucocorticoid administration (Figure 4C). We then measured circulating eosinophil counts before and at the critical 2-hour interval after glucocorticoid administration. In the absence of CXCR4 blockade, we observed the expected decrease in circulating eosinophils within 2 hours of glucocorticoid administration. In the presence of CXCR4 blockade, glucocorticoid-induced eosinopenia was decreased or eliminated (Figure 4D).

Although plerixafor is usually considered to be a specific CXCR4 blocker, there is biochemical evidence that it also binds the atypical chemokine receptor ACKR3 (previously called CXCR7). To assess whether ACKR3 could also play a role in glucocorticoid-induced eosinopenia or be a potential confounder in our CXCR4 blockade experiments, we reviewed our existing data set from the human eosinophil transcriptome analysis before and at 3 time points after in vivo administration of a glucocorticoid (GEO accession number GSE111789). We found no evidence of ACKR3 expression in human eosinophils at BL or in response to glucocorticoids (supplemental Figure 7).

To explore the possibility that, in addition to induction of CXCR4 expression, glucocorticoids could induce CXCL12 expression in target tissues, we reviewed our existing data set of transcriptome analysis in 9 hematopoietic or nonhematopoietic primary human cell types treated in vitro with a glucocorticoid (GEO accession number GSE112101). We found no evidence of induction of CXCL12 expression in any of the human cell types studied, and a statistically significant reduction in CXCL12 expression in fibroblasts, myoblasts, and osteoblasts (supplemental Figure 8).

The phenomenon of glucocorticoid-induced eosinopenia has intrigued clinicians for 7 decades and is the basis for the now widespread use of glucocorticoids to treat diseases in which eosinophils play an important pathophysiologic role. Our findings indicate that the mechanism of glucocorticoid-induced eosinopenia involves CXCR4-dependent migration of circulating eosinophils to the bone marrow.

We first studied the eosinophil response to in vivo glucocorticoid administration in 3 species: human, mouse, and rhesus macaque. Our results highlight the importance of relying on experimental data rather than predictions or assumptions about interspecies differences or similarities when studying the response to glucocorticoids, in keeping with the growing evidence of substantial species and cell type-specificity in the patterns of GR binding and transcriptional response to glucocorticoids. The extent of this specificity and the fact that there are thousands of GREs in the genome of each species, with different GR binding patterns in each cell type, mean that, even when a similar phenotypic response to glucocorticoids is observed among species, the underlying mechanism could be different. We identified the rhesus macaque as an appropriate model for the study of glucocorticoid-induced eosinopenia, because the kinetics of the eosinopenic response, the GRE of interest, and the timing of CXCR4 upregulation by eosinophils in response to in vivo glucocorticoid treatment mirrored those in humans. We then developed a method for purification of rhesus eosinophils by immunomagnetic negative selection, which resulted in high purity, viability, and normal function. To study the migration pattern of eosinophils after glucocorticoid administration in vivo, we relied on an ⁸⁹Zr-oxine ex vivo cell labeling method for PET imaging that had recently been developed by our team. This method offers the advantages of high sensitivity and minimal toxicity to hematopoietic cells when optimal radiolabeling doses are used.^19,20,29  In the present study, ⁸⁹Zr-oxine labeling did not induce measurable changes in apoptosis, activation, or in vitro migration in purified eosinophils before reinjection. Deferoxamine was infused continuously throughout the imaging period, to bind any free ⁸⁹Zr released from eosinophils and facilitate its rapid urinary clearance. In the absence of deferoxamine, free ⁸⁹Zr can bind hydroxyapatite within bones³⁰  or be transferred from dead cells to phagocytes.

Shortly after reinjection, eosinophils were distributed primarily to the lungs. Their transit time through the lungs was shorter than that of most cells that have been tracked in vivo to date,^31-33  but consistent with a previous study of human eosinophils and neutrophils labeled with ^99mTc–hexamethylpropyleneamine oxime.³⁴  Within 1 hour of reinjection, the cells reached a steady state of distribution, which was stable for at least 48 hours, providing a time window for evaluating the effect of glucocorticoid administration on eosinophil migration in vivo.

This method enabled us to document that the primary site of eosinophil migration after glucocorticoid administration is the bone marrow. In human and rhesus circulating eosinophils, glucocorticoids induced expression of the chemokine receptor CXCR4 between 60 and 120 minutes after the medication was administered, coinciding with the timing of eosinophil egress from the circulation. The bone marrow is rich in the CXCR4 ligand CXCL12, which likely serves as the main chemoattractant. We therefore tested the effect of selective pharmacologic blockade of the CXCR4/CXCL12 interaction by the small molecule plerixafor on glucocorticoid-induced eosinopenia. Because CXCR4 blockade is known to mobilize eosinophils from the bone marrow, we first had to identify a kinetic window for investigating the effect of CXCR4 blockade on glucocorticoid administration. To accomplish this, we measured eosinophil counts over time, before and after CXCR4 blockade with plerixafor alone. We found that plerixafor led to the expected early rise in eosinophil counts and that this increase stabilized at ∼6 hours. This time frame offered a kinetic window for studying the effect of CXCR4 blockade on glucocorticoid administration at a time when the eosinophil count was stable, but when we would still expect there to be sufficient plerixafor (a drug with a plasma half-life of 3-5 hours) in the circulation to effectively block the newly expressed CXCR4. Knowing from the preceding imaging studies that glucocorticoids induce migration of eosinophils to the bone marrow, if the mechanism of glucocorticoid-induced eosinopenia was unrelated to CXCR4, we would have expected glucocorticoids to disrupt the balance and lead to a decrease in circulating eosinophil counts, similar to that observed in the absence of CXCR4 blockade. Instead, glucocorticoid-induced eosinopenia was blunted or eliminated in the presence of the CXCR4 blocker, indicating that bone marrow migration of eosinophils induced by glucocorticoids is indeed CXCR4-dependent and likely is the major mechanism behind glucocorticoid-induced eosinopenia, although the contribution of other mechanisms cannot be ruled out.

It is possible that glucocorticoid-induced upregulation of CXCR4 on eosinophils already in the bone marrow, resulting in increased retention, contributes to the decrease seen in the blood, but data supporting this hypothesis are lacking to date. Moreover, pharmacokinetic data on an anti-IL-5 antibody, which causes maturational arrest of cells in the bone marrow, suggest that the appreciable impact of decreased bone marrow release of eosinophils is not reflected in blood levels until 12 to 24 hours,³⁵  which is inconsistent with the rapid kinetics of glucocorticoid-induced eosinopenia. Also, although we cannot completely exclude a role for other mechanisms, our review of existing data sets of primary human cells treated in vivo or in vitro with glucocorticoids suggests that upregulation of ACKR3 in eosinophils, or of CXCL12 in target tissues, is unlikely to be involved in glucocorticoid-induced eosinopenia.

Our results are most consistent with a model in which the threshold for surface CXCR4 expression for eosinophils to migrate to the bone marrow is considerably lower than the peak level of expression induced by glucocorticoids. Such a model would explain why migration occurs as surface CXCR4 levels increase and is nearly complete before the peak level of surface CXCR4 expression is reached. In this model, the levels of CXCR4 expression observed at 6 hours, which are still substantially higher than those observed at BL, could be sufficient to keep the cells out of the circulation. However, the fate of eosinophils after glucocorticoid-induced bone marrow migration is not yet clear, and it is possible that eosinophils are retained in the bone marrow by other mechanisms after the peak of CXCR4 expression. We know from previous studies that, in addition to CXCR4, glucocorticoids induce the expression of a proapoptotic transcriptional program in circulating eosinophils, which includes increased expression of the proapoptotic genes BTG1, PRUNE2, TXNIP, and BCL2L11.⁵  However, apoptosis of human eosinophils does not occur before their egress from the peripheral circulation.⁵  Because the CXCR4-dependent migration of eosinophils to the bone marrow occurs within the first 2 hours of glucocorticoid administration, if the glucocorticoid-induced transcriptional reprogramming of the cells is sufficient for them to undergo apoptosis, it seems likely that this occurs in the bone marrow, where they could be cleared by cells of the mononuclear phagocyte system.

Beyond the interest in explaining a decades-old medical mystery, dissecting the mechanism of glucocorticoid-induced eosinopenia is important for at least 2 reasons. First, glucocorticoids are highly effective in most patients, but their use results in substantial toxicity related to effects that are outside of the target tissue. A better understanding of their effects on eosinophils opens the possibility of identifying more targeted therapies that, alone or in combination, could target the specific effects that result in clinical benefit for patients with eosinophilic disorders, without the many side effects that are associated with glucocorticoid therapy. Second, patients with eosinophilic disorders vary in the magnitude of their clinical response to glucocorticoids, and the identification and management of glucocorticoid-resistant patients therefore remains a challenge.³⁶  It is important to establish the extent to which glucocorticoid resistance in patients with eosinophilic disorders could be related to the level of induction of CXCR4 expression, to the BL level of CXCL12 expression, or to genetic variation at these 2 loci.

The authors thank Kate Stringaris, Robert Reger, and Robert Donahue for constructive discussion and technical assistance; Karen Keeran, Kenneth Jeffries, Victoria Haley, Arthur Zetts, and the Division of Veterinary Resources (DVR) staff for animal care; Lawrence Szajek for generating ⁸⁹ZrCl₄; Philip Eclarinal, Alicia Forest, and Hoa Tieu for assisting with the PET/CT scans; and Erina He (National Institutes of Health [NIH] Medical Arts) for creating the video.

This work was supported by the Intramural Research Programs of the National Institutes of Health, National Institute of Allergy and Infectious Diseases, the National Heart, Lung and Blood Institute, the National Cancer Institute, and the National Institute of Arthritis and Musculoskeletal and Skin Diseases.

Contribution: S.G.H., N.S., F.L., and L.M.F. drafted the manuscript; N.S., S.G.H., F.L., M.G., M.M., K.S., K.N.H., S.J.Y., N.S.L., R.R.C., T.H., and L.M.F. performed the experimental work; L.M.F., N.S., A.D.K., C.E.D., S.G,H., F.L., and P.L.C. designed the study; L.M.F., N.S., S.G.H., F.L., and Z.H. analyzed the data; and all authors reviewed and approved the final version of the manuscript.

Conflict-of-interest disclosure: P.L.C. and N.S. possess a US patent for the ⁸⁹Zr-oxine complex cell-labeling method and have filed US and international patent applications for the generation and use of the ⁸⁹Zr-oxine complex employed in this study. The remaining authors declare no competing financial interests.

The current affiliation of L.M.F. and M.G. is Systemic Autoimmunity Branch, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health, Bethesda, MD.

Correspondence: Luis M. Franco, Systemic Autoimmunity Branch, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health, Bethesda, MD 20892; e-mail: luis.franco@nih.gov; and Noriko Sato, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892; e-mail: saton@mail.nih.gov.