|Home | About | Journals | Submit | Contact Us | Français|
Multiple cell types have been proposed to create niches for haematopoietic stem cells (HSCs). However, the expression patterns of HSC maintenance factors have not been systematically studied and no such factor has been conditionally deleted from any candidate niche cell. Thus, the cellular sources of these factors are undetermined. Stem Cell Factor (SCF) is a key niche component that maintains HSCs. Using Scfgfp knock-in mice we found Scf was primarily expressed by perivascular cells throughout bone marrow. HSC frequency and function were not affected when Scf was conditionally deleted from haematopoietic cells, osteoblasts, Nestin-Cre, or Nestin-CreER-expressing cells. However, HSCs were depleted from bone marrow when Scf was deleted from endothelial cells or Leptin receptor (Lepr)-expressing perivascular stromal cells. Most HSCs were lost when Scf was deleted from both endothelial and Lepr-expressing perivascular cells. HSCs reside in a perivascular niche in which multiple cell types express factors that promote HSC maintenance.
Stem cells are maintained in specialized microenvironments in tissues, termed niches, in which supporting cells secrete factors that promote stem cell maintenance1. In most mammalian tissues, including the haematopoietic system, the identities of the cells that promote stem cell maintenance remain uncertain1,2. One popular model of the HSC niche proposed that osteoblasts express many factors that promote HSC maintenance3, including SCF, CXCL12, Angiopoietin-1, and Thrombopoietin4-7. However, none of these factors have been conditionally deleted from osteoblasts, so there is no direct evidence that osteoblasts are a functionally important source of these factors.
We found that most HSCs localize adjacent to sinusoidal blood vessels throughout the bone marrow8,9. This led us to hypothesize that the HSC niche is perivascular2,9. Others found that perivascular stromal cells secrete high levels of CXCL12 and other factors proposed to regulate HSC maintenance10,11. Nestin-expressing mesenchymal stem cells also localize adjacent to blood vessels in the bone marrow and express factors thought to promote HSC maintenance12. Ablation of the Cxcl12-expressing cells or the Nestin-expressing cells reduced HSC frequency12,13. Administration of antibodies against endothelial cells in vivo impairs HSC engraftment and transformed endothelial cells promote HSC expansion in culture14,15. Nonetheless, no factor that promotes HSC maintenance has been conditionally deleted from any perivascular cell so there is no direct evidence that endothelial or perivascular cells are functionally important sources of factors for HSC maintenance.
SCF is non-cell-autonomously required for HSC maintenance in vivo 16-19. Differential splicing and proteolytic cleavage lead to the expression of a membrane-bound form and a soluble form of SCF. HSCs are depleted in Sl/Sld mutant mice20, which express soluble SCF but lack the membrane-bound form, indicating that membrane-bound SCF is particularly important for HSC maintenance21. Mice with a mixture of wild-type and Sl/Sld stromal cells only exhibit normal haematopoiesis in the immediate vicinity of the wild-type cells, demonstrating that SCF acts locally in creating the niche22.
Scf has been suggested to be expressed by endothelial cells, bone marrow fibroblasts, osteoblasts, Cxcl12-expressing perivascular stromal cells, and Nestin-expressing mesenchymal stem cells5,12,13,23-25 but Scf has not been conditionally deleted to test which source(s) are functionally important for HSC maintenance. We generated Scfgfp and Scffl mice to systematically examine Scf expression and to conditionally delete Scf from subpopulations of bone marrow cells.
We generated Scfgfp knock-in mice by inserting Egfp into the endogenous Scf locus (Supplementary Fig. 1a-c). Scfgfp/gfp mice died perinatally (Fig. 1a; Supplementary Fig. 1f, g) with severe anemia (Fig.1b; Supplementary Fig. 2c) as observed in mice with a strong loss of SCF/c-Kit function17. By quantitative reverse transcription PCR (qRT-PCR) Scf transcripts were nearly undetectable in Scfgfp/gfp newborn liver (Fig. 1c).
The overall cellularity of the newborn liver was reduced about 2-fold in Scfgfp/+ and about 5-fold in Scfgfp/gfp mutant mice compared to Scf+/+ controls (Fig. 1d). The frequency of HSCs (CD150+CD48-CD41-Sca1+cKit+ cells9,26) in the newborn liver was reduced about 8-fold in Scfgfp/gfp mutant mice compared to littermate Scfgfp/+ or Scf+/+ controls (Fig. 1e). Consistent with this, newborn Scfgfp/gfp liver cells gave significantly lower levels of donor cell reconstitution in irradiated mice compared to Scfgfp/+ or Scf+/+ controls (Fig. 1f; Supplementary Fig. 2d). Scfgfp/gfp mice therefore have a severe loss of SCF function.
By flow cytometry, only rare (0.027±0.0099%, mean±s.d.) enzymatically dissociated bone marrow cells were positive for GFP. The actual frequency of GFP+ cells in the bone marrow may be somewhat higher as our dissociation conditions may not recover all of the GFP+ stromal cells. These GFP+ cells were negative for CD45 and Ter119, indicating a non-haematopoietic source of SCF (Fig. 1g). Endogenous Scf transcripts were highly enriched in GFP+ stromal cells and highly depleted in GFP negative stromal cells (Suppl. Fig. 2f, g), suggesting GFP expression faithfully reflected endogenous Scf expression.
GFP was mainly expressed by cells surrounding sinusoids throughout the Scfgfp/+ bone marrow, with some expression by cells surrounding venuoles and arterioles (Fig. 1h-m; Supplementary Fig. 2h, i). GFP partially overlapped with endothelial marker staining (Fig. 1h-j; o-q; Supplementary Fig. 2i), suggesting that both endothelial and perivascular stromal cells express Scf. In contrast, GFP was not concentrated near the endosteum (Supplementary Fig. 2h) and we did not detect any GFP expression by bone-lining cells that expressed osteoblast markers in either the diaphysis (Fig. 1k-m) or trabecular bone (Fig. 1n). Perivascular stromal and endothelial cells therefore appeared to represent the major sources of SCF in bone marrow.
We isolated Scf-GFP+ cells by flow cytometry and performed gene expression profiling. Several mesenchymal stem/stromal cell markers, including Cxcl12, alkaline phosphatase, Vcam1, Pdgfrα and Pdgfrβ were highly elevated in Scf-GFP+ cells relative to whole bone marrow cells (Supplementary Table 1). This suggests that Scf-GFP+ cells included mesenchymal stem/stromal cells27 and Cxcl12-expressing perivascular stromal cells10. Nestin was not detected in Scf-GFP+ perivascular cells (Supplementary Table 1).
As we observed previously8,9, CD150+CD48-Lineage- candidate HSCs were mainly found adjacent to sinusoidal blood vessels throughout the bone marrow. 65% (47/73) of all CD150+CD48-Lineage- cells were immediately adjacent to GFP-expressing stromal cells (Fig. 1r-u). Almost all of the remaining cells (30%; 22/73) were within 5 cell diameters of GFP-expressing cells. This suggests Scf-GFP-expressing cells form a perivascular niche for HSCs.
We generated a floxed allele of Scf (Scffl) to conditionally delete Scf from candidate niche cells (Supplementary Fig. 3a-c). Mice homozygous for the germline recombined allele of Scf, Scf-/-, were perinatal lethal and anemic (Fig. 2a) similar to other Scf-deficient mice (Fig. 1a and ref17). Recombination of the Scffl allele therefore gave a strong loss of SCF function. We were unable to amplify Scf transcripts by PCR from the liver of Scf-/- newborns (Fig. 2b).
We generated Ubc-CreER; Scffl/fl mice to ubiquitously delete Scf upon tamoxifen administration. We administered tamoxifen-containing chow to Ubc-CreER; Scffl/fl mice and littermate controls for 1-2 months beginning at 8 weeks of age then sacrificed them for analysis. Some of the mice became anemic and ill during tamoxifen administration. The Ubc-CreER; Scffl/fl mice had significantly lower red blood cell counts than controls (Fig. 2c) and a trend toward lower white blood cell and platelet counts (Supplementary Fig. 3d). Ubc-CreER; Scffl/fl mice exhibited approximately 2-fold reductions in the overall cellularity of bone marrow and spleen compared to controls (Fig. 2d).
CD150+CD48-Lin-Sca1+c-Kit+ HSCs were also depleted in the bone marrow and spleen of Ubc-CreER; Scffl/fl mice compared to controls treated concurrently with tamoxifen (Fig. 2e). Limit dilution analysis demonstrated that long-term multilineage reconstituting cells were 3.5-fold less frequent in the bone marrow of Ubc-CreER; Scffl/fl mice compared to controls upon transplantation into irradiated mice (Fig. 2f). Bone marrow cells from Ubc-CreER; Scffl/fl mice gave significantly lower levels of donor cell reconstitution in irradiated mice (Fig. 2g). These data confirmed that SCF is required for HSC maintenance in adult mice.
CD150+CD48-Lin-Sca1+cKit+ HSCs from Scfgfp/+ mice did not express GFP by flow-cytometry (Fig. 2h). This is consistent with prior studies17,21,22 in suggesting that Scf non-cell-autonomously regulates HSC maintenance. To test the role of other haematopoietic cells we conditionally deleted Scf using Vav1-Cre. As expected28, Vav1-Cre recombined a conditional loxpEYFP reporter29 in virtually all HSCs, CD45+ and Ter119+ haematopoietic cells (Fig. 3a; Supplementary Fig. 4a). Eight week-old Vav1-Cre; Scffl/- mice exhibited normal blood cell counts, bone marrow composition (Supplementary Fig. 4b, c), and bone marrow and spleen cellularity (Fig. 3b). Scf+/- heterozygous mice exhibited a 2-fold decline in the frequency of CD150+CD48-Lin-Sca1+c-Kit+ HSCs relative to wild-type littermates. However, deletion of the second allele of Scf from haematopoietic cells in Vav1-Cre; Scffl/- mice did not further reduce HSC frequency in the bone marrow or spleen (Fig. 3c). Bone marrow cells from adult Vav1-Cre; Scffl/- mice had a normal capacity to reconstitute irradiated mice (Fig. 3d; Supplementary Fig. 4d) and to form colonies in methylcellulose (Supplementary Fig. 4e, f). Therefore, Scf expression by haematopoietic cells is not required for HSC maintenance in adult bone marrow.
Col2.3-Cre recombines genes in fetal and postnatal osteoblasts30. Consistent with this we found strong EYFP expression among bone-lining cells in Col2.3-Cre; loxpEYFP mice (Fig. 3e). To test whether osteoblasts produce SCF for HSC maintenance we analyzed 8 week-old Col2.3-Cre; Scffl/- mice. Col2.3-Cre; Scffl/- mice had normal blood counts (Supplementary Fig. 5a), normal lineage composition in the bone marrow and spleen (Supplementary Fig. 5b) and normal bone marrow and spleen cellularity (Fig. 3g). Although Scf+/- germline heterozygous mice exhibited a 2-fold decline in the frequency of CD150+CD48-Lin-Sca1+c-Kit+ HSCs relative to wild-type littermates, conditional deletion of the second allele of Scf from osteoblasts in Col2.3-Cre; Scffl/- mice did not further reduce HSC frequency in the bone marrow or spleen (Fig. 3h). Bone marrow cells from Col2.3-Cre; Scffl/- mice had a normal capacity to reconstitute irradiated mice (Fig. 3i; Supplementary Fig. 5c) and to form colonies in methylcellulose (Supplementary Fig. 5d, e). Therefore, Scf expression by osteoblasts is not required for HSC maintenance in adult bone marrow.
In Nestin-Cre; loxpEYFP mice we found rare EYFP-expressing perivascular stromal cells around larger blood vessels, not sinusoids, in the bone marrow (Fig. 3f). These cells exhibited a very different distribution than Scf-expressing cells (compare Fig. 3f to Fig. 1h-m and Supplementary Fig. 2h, i). Eight week-old Nestin-Cre; Scffl/- mice had normal blood cell counts (Supplementary Fig. 6b), normal lineage composition and cellularity in the bone marrow and spleen (Supplementary Fig. 6c; Fig. 3j). Comparing Nestin-Cre; Scffl/- mutants with Scf+/-controls, deletion of Scf from Nestin-Cre-expressing cells did not reduce HSC frequency in the bone marrow (Fig. 3k). Nestin-Cre; Scffl/- mice did exhibit a significant decline in HSC frequency in the spleen (Fig. 3k), raising the possibility that Nestin-Cre-expressing cells are a component of the HSC niche in the spleen. Bone marrow cells from adult Nestin-Cre; Scffl/- mice had a normal capacity to reconstitute irradiated mice (Fig. 3l; Supplementary Fig. 6d). Conditional deletion of Scf by administering tamoxifen for 2-5 months to adult Nestin-CreER; Scffl/fl mice also did not affect hematopoiesis, HSC frequency, or reconstituting capacity in irradiated mice (Supplementary Fig. 7). Therefore, Scf expression by Nestin-Cre-expressing or Nestin-CreER-expressing perivascular cells is not required for the maintenance of HSCs in adult bone marrow.
Since Nestin-GFP-expressing bone marrow cells express Scf12, we independently characterized Nestin-GFP expression. Consistent with the prior report12, we observed strong Nestin-GFP staining along larger vessels in the bone marrow (Supplementary Fig. 8; see Supplementary Fig. 1 from ref12). Nestin-GFP was also observed in perisinusoidal stromal cells in a pattern that resembled Scf-GFP+ expression (Supplementary Fig. 8a). This appeared to be different from the Nestin-Cre expression pattern, which we detected only around larger blood vessels in the bone marrow (Fig. 3f). In Nestin-Cherry and Nestin-GFP double transgenic mice we detected Nestin-Cherry expression around larger vessels but not around sinusoids while Nestin-GFP was detected around both (Supplementary Fig. 8). Thus, different Nestin transgenes appear to be expressed by different subpopulations of perivascular stromal cells. Nestin-GFP appears to exhibit more expression in perisinusoidal stromal cells than other Nestin transgenes. Our data are therefore consistent with the possibility that Nestin-GFP-expressing stromal cells contribute to the HSC niche as suggested12, even though conditional deletion of Scf with Nestin-Cre and Nestin-CreER did not affect HSC frequency.
We conditionally deleted Scf from endothelial cells using Tie2-Cre31. Tie2-Cre recombined in endothelial (Fig. 4a) and haematopoietic cells (Fig. 4b) but not in mesenchymal stem/stromal cells from the bone marrow (Supplementary Fig. 9d, e). Since haematopoietic cells do not express Scf (Fig. 1g; Fig. 2i) and conditional deletion of Scf from haematopoietic cells did not affect HSC frequency (Fig. 3a-d), the use of Tie2-Cre allowed us to test whether SCF expression by endothelial cells regulates HSC frequency.
Eight week-old Tie2-Cre; Scffl/- mice exhibit normal blood cell counts (data not shown), bone marrow and spleen cellularity (Fig. 4c). However, the frequency of CD150+CD48-Lin-Sca1+c-Kit+ HSCs in the bone marrow was significantly reduced in Tie2-Cre; Scffl/- mice relative to controls (Fig. 4d). Consistent with this, 300,000 bone marrow cells from Tie2-Cre; Scffl/- mice gave significantly lower levels of donor reconstitution upon transplantation into irradiated mice (Fig. 4g). In five independent experiments, 24 of 25 recipients of Scf+/+ cells, 15 of 15 recipients of Scf+/- cells, and only 7 of 21 recipients of Tie2-Cre; Scffl/- cells were long-term multilineage reconstituted. By poisson statistics this corresponds to an HSC frequency in control bone marrow of at least 1/93,200 cells but only 1/739,900 in Tie2-Cre; Scffl/- mice. Endothelial cells are therefore an important source of SCF for HSC maintenance.
The HSC depletion in Tie2-Cre; Scffl/- mice likely reflects an ongoing need for SCF expression by endothelial cells in adult bone marrow because when HSCs are depleted as a consequence of reduced SCF/c-Kit signaling, HSC frequencies return to normal levels upon restoration of normal SCF/c-Kit signaling19,21. Nonetheless, we also examined whether SCF expression by endothelial cells during development is required by HSCs. We found a 1.7 to 2.1-fold reduction in HSC frequency in the liver of newborn Tie2-Cre; Scffl/- mice (Fig. 4e) and a 2-fold reduction in HSC frequency in the bone marrow of one month-old Tie2-Cre; Scffl/- mice (Fig. 4f) relative to Scf+/- and Scf+/+ controls. The magnitude of HSC depletion in adult bone marrow appeared to increase as we found a 2.7-fold and 5.2-fold reduction in HSC frequency in the bone marrow of 8 week-old Tie2-Cre; Scffl/- mice relative to Scf+/- and Scf+/+ controls, respectively (Fig. 4d). These data suggest that ongoing SCF expression by endothelial cells in adult bone marrow contributes to HSC maintenance; however, HSC depletion in adult bone marrow may reflects a loss of SCF expression by endothelial cells during development.
We found that Leptin receptor (Lepr) is highly restricted in its expression within the bone marrow to Scf-GFP-expressing perivascular stromal cells (Supplementary Table 1). Consistent with this, Lepr-Cre; loxpEYFP mice exhibited EYFP expression in perivascular stromal cells (Fig. 5b, e) but not in haematopoietic cells (Fig. 5b, c, e), bone-lining cells (Fig. 5c), or endothelial cells (Fig. 5d).
Consistent with the gene expression profile of Scf-GFP+ cells (Supplementary Table 1), EYFP+ cells from Lepr-Cre; loxpEYFP mice did not detectably express Nestin but did express mesenchymal stem/stromal cell markers including Cxcl12, alkaline phosphatase, PDGFRα, and PDGFRβ (Supplementary Fig. 9a-c). These data suggest a mesenchymal identity for the Lepr-expressing stromal cells; however, the lack of EYFP expression in bone-lining cells from Lepr-Cre; loxpEYFP mice suggests that the Lepr-Cre-expressing perivascular cells did not give rise to osteoblasts during normal development. Future studies will be required to assess the relationship between Lepr-Cre-expressing perivascular cells, mesenchymal stem cells, and other perivascular stromal cells.
Bone marrow cellularity was significantly reduced in Lepr-Cre; Scffl/gfp mice compared to Scf+/+ controls, but not compared to Scf+/gfp controls (Fig. 5g). Spleen size (Fig. 5f) and cellularity were significantly increased in Lepr-Cre; Scffl/gfp mice (Fig. 5g). Sections through the spleen revealed increased extramedullary haematopoiesis in Lepr-Cre; Scffl/gfp mice (data not shown). The frequency of CD150+CD48-Lin-Sca1+c-Kit+ HSCs was significantly reduced in the bone marrow of Lepr-Cre; Scffl/gfp mice, but significantly increased in the spleen (Fig. 5h). The total number of bone marrow and spleen HSCs per mouse was significantly reduced in Lepr-Cre; Scffl/gfp mice (Fig. 5i). Lepr-Cre-expressing cells thus promote HSC maintenance in the bone marrow, but not in the spleen, by producing SCF.
In limit dilution transplantation studies the frequency of long-term multilineage reconstituting cells in Scf+/+ and Scf+/gfp control cells was 1/38,311 and 1/38,352, respectively (Fig. 5j). In Lepr-Cre; Scffl/gfp bone marrow cells the frequency of long-term multilineage reconstituting cells was 1/78,136, significantly lower than in Scf+/+ and Scf+/gfp controls (Fig. 5j). Thus conditional deletion of Scf from Lepr-Cre-expressing perivascular stromal cells depletes HSCs from adult bone marrow. The frequency of GFP+ cells in the bone marrow of Lepr-Cre; Scffl/gfp mice did not significantly differ from Scfgfp/+ controls (Supplementary Fig. 10), suggesting that Scf deletion did not lead to the death of Lepr-expressing cells.
The HSC depletion observed in Lepr-Cre; Scffl/gfp mice did not reflect a developmental effect of SCF expression by Lepr-Cre-expressing cells as no HSC depletion was detected in the liver of newborn Lepr-Cre; Scffl/gfp mice (Fig. 5k). Furthermore, the magnitude of the HSC depletion increased with time in the adult bone marrow (Fig. 5h, l).
To test whether deletion of Scf from endothelial and Lepr-expressing perivascular cells has additive effects on HSC depletion we analyzed 8 week-old Tie2-Cre; Lepr-Cre; Scffl/- mice. Bone marrow cellularity was significantly reduced in Tie2-Cre; Lepr-Cre; Scffl/- mice compared to Tie2-Cre; Scffl/- and Lepr-Cre; Scffl/- mice (Fig. 5m). Spleen cellularity was significantly increased in Tie2-Cre; Lepr-Cre; Scffl/- mice compared to Scf+/- or Tie2-Cre; Scffl/- mice (Fig. 5m). HSC frequency was significantly reduced in the bone marrow of Tie2-Cre; Lepr-Cre; Scffl/- mice compared to Tie2-Cre; Scffl/- or Lepr-Cre; Scffl/- mice (Fig. 5n). The frequency and absolute number of HSCs in the bone marrow of Tie2-Cre; Lepr-Cre; Scffl/- mice was less than 5% of wild-type levels (Fig. 5m-o). This suggests endothelial and Lepr-expressing perivascular stromal cells are the major sources of SCF for HSC maintenance in normal adult bone marrow and that deletion of Scf from each cell population has additive effects on HSC depletion.
qRT-PCR revealed that endothelial and Lepr-Cre-expressing perivascular cells expressed both long and short splice isoforms of Scf rendering both cell types capable of expressing membrane-bound and soluble SCF (Supplementary Fig. 11). The levels of both isoforms of Scf in the two cell populations were significantly higher than in whole bone marrow cells, though Lepr-Cre-expressing cells expressed much higher levels of both isoforms compared to endothelial cells (Supplementary Fig. 11).
Our data demonstrate that HSCs reside in a perivascular niche in which endothelial and Lepr-expressing perivascular stromal cells are two functionally important components of the niche (Supplementary Fig. 12). The simplest interpretation of our data is that both cell types produce SCF for the maintenance of HSCs in adult bone marrow; however, endothelial cells also produce SCF for HSC maintenance/expansion during development so it is formally possible that the depletion of bone marrow HSCs in adult Tie2-Cre; Scffl/- mice reflects a developmental effect of endothelial SCF. Endothelial cells and perivascular stromal cells are probably not the only components of the HSC niche as other cell types likely contribute through mechanisms other than SCF production (e.g. refs32,33).
Lepr-Cre-expressing stromal cells did not express endogenous Nestin (Supplementary Fig. 9c). Nestin-Cre, or Nestin-CreER mediated deletion of Scf did not deplete HSCs (Fig. 3j-l; Supplementary Fig. 6-7). However, Lepr-Cre-expressing perisinusoidal cells do partially overlap with Nestin-GFP expressing perivascular cells (Supplementary Fig. 8; Supplementary Fig. 9a-c). The Lepr-Cre-expressing stromal cells therefore include stromal cells that express certain Nestin transgenes, consistent with Mendez-Ferrer et al12 and may also include Cxcl12-abundant reticular (CAR) cells10. Perivascular stromal cells are likely heterogeneous and may include multiple cell types that contribute to HSC maintenance through different mechanisms.
While we have partially characterized the bone marrow niche for HSCs in adult mice under homeostatic conditions, other studies will be required to functionally characterize HSC niches in other haematopoietic tissues and after haematopoietic stress.
Targeting vectors for making Scfgfp and Scffl mice were constructed by recombineering34. The Frt flanked Neo cassette was removed by mating with Flpe mice35. Scfgfp and Scffl mice were backcrossed onto a C57BL background before analysis. Mice used in this study included Ubc-CreER 36, CMV-Cre37, Vav1-Cre28, Nestin-Cre38, Tie2-Cre31, Lepr-Cre39, and LoxpEYFP29 (all from the Jackson Laboratory), Col2.3-Cre30, Nestin-CreER40 and Nestin-GFP41. All were maintained on a C57BL background. Unless otherwise indicated, data always reflect mean±s.d. and two-tailed student’s t-tests were used to assess statistical significance (*p<0.05, **p<0.01, ***p<0.001). Detailed methods are available online.
This work was supported by the Howard Hughes Medical Institute (HHMI) and the National Heart, Lung and Blood Institute (5R01HL097760). L.D. was supported by a Helen Hay Whitney Foundation Fellowship and by HHMI. G.E. was supported by the National Institute of Aging (R01AG040209) and NYSTEM. We thank M. White and D. Adams for flow cytometry, E. Hughes at the UM transgenic core for helping to generate Scfgfp and Scffl mice, M. Purkey for technical assistance, J. Peyer and M. Lim for discussion, and C. Mountford, S. Grove and R. Coolon for managing the mouse colony. This work was initiated in the Life Sciences Institute at the University of Michigan then completed at Children’s Research Institute at UT Southwestern.
Author Contributions L.D. performed all of the experiments. T.L.S. helped to design and generate the Scffl and Scfgfp mice. G.E. generated the Nestin-Cherry transgenic mice. L.D. and S.J.M. designed the experiments, interpreted the results, and wrote the manuscript.
The microarray data were deposited in the GEO repository (www.ncbi.nlm.nih.gov/geo/) under accession number GSE33158.
The authors declare no competing financial interests.
Supplementary Information is linked to the online version of the paper at www.nature.com/nature.