Bone Marrow Is a Reservoir for CD4⁺CD25⁺ Regulatory T Cells that Traffic through CXCL12/CXCR4 Signals

CD4⁺CD25⁺ T cells (Tregs) mediate peripheral T-cell homeostasis (1, 2, 3). Studies in murine models show that CD4⁺CD25⁺ T cells are essential for the induction of tolerance to alloantigens and inhibit graft-versus-host disease (GVHD; refs. 4, 5). Human blood Tregs express CCR4 and CCR8 and migrate in response to their ligands in vitro (6). We recently showed that pathologic Tregs in human ovarian cancers migrate into tumor in response to CCL22 (7). Homeostatic Tregs may originate in the thymus, although their differentiation in the periphery also has been suggested. Despite these recent advances in our knowledge, relatively little is known regarding the natural reservoirs of migrating Tregs or signals that induce their mobilization of trafficking.

Bone marrow is vascularized by blood but not by lymphatic vessels. Bone marrow is a part of the lymphocyte recirculation network (8), with billions of lymphocytes recirculating through it each day. We hypothesized that bone marrow might harbor CD4⁺CD25⁺ Tregs and function as a reservoir for them. In this capacity, it could be an important organ to fine tune T-cell immunity by modulating Treg trafficking. We further hypothesized that granulocyte colony-stimulating factor (G-CSF) would mobilize bone marrow CD4⁺CD25⁺ Tregs. We tested these hypotheses in defined models and human subjects and now show a high prevalence of functional Tregs in human bone marrow. We provide evidence that CXCR4/CXCL12 signals play an important role in regulating Treg egress from bone marrow and in maintaining homeostatic levels of Tregs in the periphery. G-CSF mobilizes Tregs from bone marrow into the periphery by decreasing marrow CXCL12 expression.

These data may be used to manipulate Tregs for therapeutic purposes and may help to explain the low prevalence of acute GVHD in recipients of G-CSF–mobilized bone marrow transplants (9, 10) and the improvement in autoimmune diseases following G-CSF treatment (11, 12).

Written, informed consent was obtained for all of the subjects. The Institutional Review Board of Tulane Medical School approved the study. Healthy adults were studied untreated or immediately following subcutaneous injection of 5 μg/kg recombinant human G-CSF (Filgastrim; Amgen Inc., Thousand Oaks, CA) once daily for four consecutive days. Peripheral blood cells, bone marrow cells, and bone marrow fluid (cell-free bone marrow) were collected and frozen for later use. Human thymus and tonsils were collected from young children undergoing cardiac surgery or other treatments and mechanically disrupted into single cell suspensions. Cells were stained with monoclonal antibodies and analyzed on a FACSCalibur (Becton-Dickinson, Franklin Lakes, NJ) using mouse antihuman CD4-FITC (SK3, IgG1) and mouse antihuman CD25-phycoerythrin (MA251, IgG1; all from BD PharMingen, San Diego, CA).

The study was approved by the Institutional Animal Care and Use Committee of the Tulane Medical School. Six- to 8-week-old female C57/black 6 mice were used (Jackson Labs, Bar Harbor, ME). Peripheral blood was collected by cardiac aspiration into heparinized glass tubes. Spleen and inguinal lymph nodes were mechanically disrupted into single cell suspensions. Bone marrow was collected from the femur, tibia, and humerus by flushing with 200 μL sterile PBS. After centrifugation, the supernatant (bone marrow fluid) was frozen for later detection of chemokines and for migration assays. CD3⁺CD4⁺CD25⁺ T cells in blood, spleen, and lymph nodes were identified with fluorescence-activated cell sorting (FACS) using antimouse CD4 (clone RM 4–5; eBioscience, San Diego, CA), CD25 (PC61; eBioscience), and CD3 (145–2c11; BD PharMingen). At least 5000 gated events per condition were analyzed using CellQuest software (Becton-Dickinson).

Real-time reverse transcription-PCR was carried out for FOXP3 using primers for upstream 5′-cagctgcccacactgcccctag-3′, downstream 5′-catttgccagcagtgggtag-3′, and the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH). cDNA was against normalized GAPDH and expressed as fold difference relative to GAPDH (13).

Human bone marrow CD4⁺ T cells were purified with Untouched kits (Miltenyi, Auburn, CA). CD25⁺ cells were sorted with anti–CD25-phycoerythrin (BD PharMingen). Cell populations were >90% pure by flow cytometry. Monocyte-derived dendritic cells (MDCs) were differentiated as we described previously (14). Dendritic cells were incubated with HLA-A2–binding influenza virus peptide GILGFVFTL (Multiple Peptide System, Seattle, WA), and 4 to 20 × 10³/mL MDCs were used to activate autologous CD3⁺ T cells (10⁵/mL). The Treg to CD3⁺ T-cell ratio was 0:1, 0.5:1, or 1:1 as indicated. On day 6, antigen-specific T-cell proliferation and cytokines were detected as we described previously (15, 16).

Migration and transmigration were assessed as we described previously (15) using human CD4⁺CD25⁺ Tregs (5 to 20 × 10⁴). Tregs were induced to migrate with recombinant human CXCL12 (100 ng/mL; R&D Systems, Minneapolis, MN) or human bone marrow fluid or mouse bone marrow fluid. Tregs were incubated with mouse antihuman CXCR4 (44717, IgG2b; 500 ng/mL) for 2 hours as indicated. Identity of migrating Tregs was further confirmed using FACS for CD3, CD4, and CD25.

CXCL12 protein in bone marrow fluids was detected by ELISA (R&D Systems).

Human blood CD4⁺CD25⁺ T cells (2.5 × 10⁶ in 200 μL PBS) were injected in 100 μL volume into tail veins in female NOD.CB17-SCID mice (6 to 8 weeks old; Jackson Labs; ref. 16). Mice were injected intraperitoneally two times with mouse antihuman CXCR4 (500 ng/200 μL; 44717, IgG2b; R&D Systems) or control antibody (IgG; BD PharMingen) 12 hours before and 12 hours after Treg injection. Twenty to 60 hours later, animals were killed, and bone marrow, blood, and spleens were collected. Human CD3⁺ T cells were identified using FACS with antihuman-CD3 antibody, expressed as human CD3⁺ T cells per 10⁶ mononuclear cells.

Differences in cell surface molecule expression were determined by χ² test and in other variables by unpaired t test with P < 0.05 considered significant.

Homeostatic CD4⁺ Tregs (identified as CD4⁺CD25⁺ T cells) are found in peripheral blood and lymphoid organs (2, 3, 17, 18). Recent reports suggest that bone marrow is a site for important T-cell immune events (19, 20, 21). Thus, we examined bone marrow for the presence of CD4⁺CD25⁺ T cells. Flow cytometric analysis (FACS) showed the presence of CD3⁺CD4⁺CD25⁺ T cells in the bone marrow of healthy human donors (Fig. 1,A). The fraction of CD3⁺CD4⁺CD25⁺ T cells among all of the CD4⁺CD3⁺ T cells (Fig. 1,A) or all of the nucleated cells (Fig. 1 B) was significantly higher in human bone marrow than in blood, lymph nodes, or thymus (P < 0.01 for each).

The expression of FOXP3 is correlated with regulatory activity of human blood Tregs (22). Reverse transcription-PCR showed that the expression of FOXP3 was higher in bone marrow CD4⁺CD25⁺ T cells than in blood CD4⁺CD25⁺ T cells (P < 0.05; Fig. 1,C). CD4⁺CD25^high Tregs were reportedly to be more suppressive than CD4⁺CD25^low Tregs (23). Multicolor staining revealed that the fraction of CD4⁺CD25^high T cells was higher in bone marrow (5.5 ± 3.5%; n = 7) than in thymus (1.5 ± 1.2%; n = 7), lymph nodes (0.5 ± 0.3%; n = 9), or blood (0.5 ± 0.4%; n = 12; P < 0.01 for each; Fig. 1 A). Thus, bone marrow harbored a significant population of CD4⁺CD25⁺FOXP3⁺ T cells. We hypothesized that these bone marrow CD4⁺CD25⁺FOXP3⁺ T cells were functional regulatory T cells as described for such cells in blood (3) and other compartments (7).

To test whether human bone marrow CD4⁺CD25⁺ T cells are functional regulatory T cells, we used our antigen-specific T-cell culture system (15, 16). Myeloid dendritic cells pulsed with influenza peptides induced antigen-specific T-cell activation as expected (Fig. 2,A–C). Inclusion of Tregs purified from autologous bone marrow significantly inhibited antigen-specific T-cell proliferation (Fig. 2,A), γ-interferon, and interleukin 2 production (Fig. 2,B and C) in a dose-dependent manner (n = 5; P < 0.05). Strikingly, bone marrow CD4⁺CD25⁺ T cells were significantly more potent than blood CD4⁺CD25⁺ T cells in inhibiting T-cell activation (n = 5; P < 0.05 for all; Fig. 2,A–C). Thus, CD3⁺CD4⁺CD25⁺ T cells in bone marrow are functional regulatory T cells (Tregs). Human bone marrow Tregs were superior to blood counterparts in suppressing T-cell activation (Fig. 2,A–C) and expressed more FOXP3 than their blood counterparts (Fig. 1 C). The data suggest that bone marrow Tregs may be activated and/or memory regulatory T cells.

We next examined whether Tregs trafficked into bone marrow. In preliminary in vitro chemotaxis assays, freshly isolated blood CD3⁺CD4⁺CD25⁺ Tregs did not efficiently migrate in response to bone marrow fluid (not shown). Because Tregs in bone marrow expressed an apparently activated phenotype (Fig. 1,C, Fig. 2), we activated human blood CD4⁺CD25⁺ T cells with allogeneic MDCS for 48 hours to replicate this activated state. Now MDC-activated blood Tregs efficiently migrated in response to human bone marrow fluid in the in vitro chemotaxis assay and in the in vitro adhesion and transmigration assay (P < 0.01) in a dose-dependent manner (Fig. 3,A and B). Antihuman CXCR4 significantly (P < 0.05) decreased bone marrow fluid–mediated Treg migration (Fig. 3,A and B). In support, we showed that human bone marrow fluid contained a high level of CXCL12 (the sole CXCR4 ligand) but not CCL22 or CCL17 (Fig. 3,C), chemokines shown to mediate blood Treg migration in vitro (6). In further support, MDC activation increased CXCR4 expression on CD4⁺CD25⁺ T cells (Fig. 3 D). Therefore, human bone marrow mediates Treg migration and local accumulation by CXCR4/CXCL12 signals in vitro.

Having shown a role for CXCL12/CXCR4 signals in vitro, we next addressed the role for this axis in vivo in our human–nonobese diabetic/severe combined immunodeficient (NOD/SCID) chimeric mouse model (16). Forty to 60 hours after intravenous human Treg transfusion, human Tregs were primarily found in bone marrow. Far fewer Tregs were found in spleen (Fig. 4,A), and <100 cells/mL were detected in peripheral blood. Strikingly, in vivo administration of a specific antihuman CXCR4 monoclonal antibody significantly (P < 0.05) decreased Treg migration into bone marrow but not into spleen (Fig. 4,B). We further showed that mouse bone marrow produced high level of CXCL12 but not CCL22 or CCL17 (Fig. 4,C). In further support, mouse bone marrow efficiently mediated human MDC-activated Treg migration in vitro in a dose-dependent manner. This in vitro migration was significantly blocked by mouse antihuman CXCR4 monoclonal antibody (Fig. 4 D). Collectively, these data indicate that CXCL12/CXCR4 signals are critical for Treg trafficking to bone marrow in vivo.

After determining that Tregs would migrate to and remain in bone marrow, we next examined whether bone marrow Tregs would have the potential to traffic back into the periphery. G-CSF mobilizes mouse hematopoietic stem cells and neutrophils by disrupting the CXCR4/CXCL12 chemotactic interaction (24, 25). Human bone marrow also expresses significant quantities of the chemokine CXCL12 (Fig. 3,C). We hypothesized that G-CSF would mobilize human bone marrow Tregs. Consistent with our hypothesis, G-CSF treatment of healthy human volunteers significantly (P < 0.01) decreased Tregs in their bone marrow and significantly (P < 0.05) increased Treg in their peripheral blood (Fig. 5,A and B). In support, we showed that G-CSF treatment significantly reduced bone marrow CXCL12 expression in healthy human volunteers (Fig. 5,C). Bone marrow from G-CSF–treated volunteers was significantly less efficient in mediating Treg migration than bone marrow from control subjects (Fig. 3 A; P < 0.001, compared with normal bone marrow). These data suggested that G-CSF mobilizes Tregs from bone marrow through disruption of CXCL12/CXCR4 signals in vivo and that bone marrow Tregs can traffic into the periphery.

Homeostatic Tregs mediate peripheral tolerance to self-antigens by suppressing autoreactive immune cells. Emerging evidence implicates regulatory T cells, particularly CD4⁺CD25⁺ regulatory T cells (Tregs) in the pathogenesis of autoimmune diseases, tumors, and organ transplantation (2, 3, 17, 18, 26). Classic, natural CD4⁺CD25⁺ regulatory T cells are thought to reside primarily in lymphoid organs (2, 3, 17, 18). On the basis of these observations, it is postulated that Tregs mediate their suppressive effects by inhibiting T-cell priming, which occurs in lymph nodes. Bone marrow is vascularized by blood but not by lymphatic vessels. Bone marrow is a part of the lymphocyte recirculation network. Billions of lymphocytes circulate through bone marrow each day (8). We found large numbers of functional Tregs in bone marrow. This finding provides evidence for a previously unidentified role of bone marrow in T-cell homeostasis: Bone marrow is a preferential site for migration or selective retention and function of Tregs. The study significantly complements recent reports that bone marrow harbors antigen-specific memory T cells (20) and is an important site for T-cell priming (19).

Conventional lymphocyte trafficking has been extensively investigated (27). However, little is known regarding the natural reservoirs for migrating Tregs and the trafficking signals for human Tregs. Our study is the first to show a mechanism for human Treg homeostatic trafficking in vivo.

We also suggest that bone marrow is a significant reservoir for human Tregs and that CXCL12/CXCR4 signals are critical for Treg trafficking between bone marrow and periphery. Several lines of evidence support this notion. First, CXCL12 is expressed in marrow and induces Treg chemotaxis and adhesion/transmigration. Second, G-CSF mobilizes human bone marrow Tregs through reducing bone marrow–derived CXCL12. Third, human Treg preferentially home to bone marrow but not to spleen under homeostatic conditions in NOD/SCID mice. Finally, blocking CXCL12/CXCR4 signals significantly reduces Treg trafficking to bone marrow. There is one amino acid difference between mouse and human CXCL12 (28, 29). Thus, our work also suggests that this human–NOD/SCID chimeric model is relevant to study the role of CXCL12/CXCR4 signals in human Treg trafficking.

Expression of FOXP3 can be induced by activation and is associated with the suppressive capacity of Tregs (22). Activated Tregs are more efficient in blocking T-cell activation than nonactivated Tregs (30). Here we show that bone marrow Tregs express more FOXP3 and CD25 than blood counterparts and are functionally superior to their blood counterparts in suppression. We also show that MDC activation of blood Tregs augments their CXCR4 expression, which significantly enhances their migration toward bone marrow–derived CXCL12. Thus, we suggest that bone marrow Tregs contain “memory” and/or activated Tregs.

We showed that human ovarian tumors produce high-level CXCL12 that mediates tumor-associated plasmacytoid dendritic cell trafficking (15). Although we recently showed that there are significant numbers of functional, tumor-infiltrating Tregs in human ovarian tumors, tumor environmental CCL22 but not CXCL12 is crucial for this Treg tumor trafficking in vivo (7). Thus, local tumor microenvironmental factors may account for these migratory differences. Alternatively, there may be distinct Treg subsets specifically recruited in response to selected stimuli. In either regard, our data suggest that Treg distribution and trafficking may be regulated in a tissue- and/or organ-specific manner that is further subject to modification by local environmental factors.

Agents that mobilize Tregs from bone marrow could be therapeutically beneficial in some clinical settings. We showed here that G-CSF decreases bone marrow CXCL12 and in turn mobilizes bone marrow Tregs. These findings may help explain how G-CSF administration reduces the severity and mortality in acute GVHD (9, 10). In support, mouse Tregs reduce the severity of GVHD (5, 31). G-CSF also ameliorates the autoimmune diseases systemic lupus erythematosus and experimental autoimmune allergic encephalomyelitis in mouse models (11, 12).

Administration of anti-CXCL12 counteracts B1a-lymphocyte expansion and T-lymphocyte activation and decreases autoantibody production and nephritis in murine lupus (32). The CXCR4 antagonist AMD3100 also inhibits autoimmune collagen-induced arthritis in mice (33). These findings may result from Treg mobilization through interrupting CXCR4/CXCL12 signals in bone marrow and suggest additional means for Treg mobilization.

Our data show that recruitment of Tregs into bone marrow through CXCL12 represents a novel and important mechanism of Treg homeostatic traffic. Mobilizing bone marrow Tregs may be a novel strategy to manipulate systemic immunity therapeutically.

We thank Dominique Emilie for critical reading of the manuscript, Roy Weiner and Jules Puschett for their support, and Sherry Price for her technical assistance.