The liver maintains hematopoietic stem cells (HSCs) during development. However, it is not clear what cells are the components of the developing liver niche in vivo. Here, we genetically dissected the developing liver niche by systematically determining the cellular source of a key HSC niche factor, stem cell factor (SCF). Most HSCs were closely associated with sinusoidal vasculature. Using Scfgfp knockin mice, we found that Scf was primarily expressed by endothelial and perisinusoidal hepatic stellate cells. Conditional deletion of Scf from hepatocytes, hematopoietic cells, Ng2+ cells, or endothelial cells did not affect HSC number or function. Deletion of Scf from hepatic stellate cells depleted HSCs. Nearly all HSCs were lost when Scf was deleted from both endothelial and hepatic stellate cells. The expression of several niche factors was down-regulated in stellate cells around birth, when HSCs egress the developing liver. Thus, hepatic stellate and endothelial cells create perisinusoidal vascular HSC niche in the developing liver by producing SCF.
Hematopoietic stem cells (HSCs) rely on niche for their maintenance. In adults, HSCs are sustained by bone marrow perivascular niche composed of multiple cellular components, including endothelial cells (Ding et al., 2012), and mesenchymal stromal cells (Ding and Morrison, 2013; Ding et al., 2012; Greenbaum et al., 2013; Sugiyama et al., 2006), along with other cell types (Morrison and Scadden, 2014). But the bone marrow is not the only organ that supports HSCs. During development, the liver is the major hematopoietic organ in which HSCs expand and mature (Mikkola and Orkin, 2006). It is only after birth that HSCs complete their migration to the bone marrow and reside there throughout adult life (Lee et al., 2017).
In contrast to the adult bone marrow niche, the cellular components of the developing liver niche remain largely undefined in vivo. Stromal cell lines derived from murine fetal liver and fibroblasts derived from human fetal liver can support hematopoiesis in vitro (Hackney et al., 2002; Moore et al., 1997; Tsai et al., 1986). Cells undergoing epithelial-to-mesenchymal transition can also maintain hematopoietic cells in vitro (Chagraoui et al., 2003; Zhang et al., 2005). But the in vivo identities of these cells are not clear. DLK+ hepatic progenitors have been shown to express key HSC niche factors and to possess the ability to support HSCs in vitro (Chou and Lodish, 2010). In addition, fetal liver Lyve+ endothelial cells can maintain HSCs short-term in vitro (Iwasaki et al., 2010). However, it is not clear whether any of these cells maintain HSCs in vivo. Recently, Nestin+NG2+ periportal stromal cells in the fetal liver have been suggested to be an important component of the niche (Khan et al., 2016), as genetic ablation of NG2+ cells in a Ng2-cre/iDTA mouse model leads to an estimated 30% reduction of phenotypic HSCs. However, fetal liver cells from Ng2-cre/iDTA mice have higher reconstituting activity in T and B cell lineages, but not myeloid lineage, compared with littermate controls (Khan et al., 2016), suggesting that functional HSCs are not compromised in these mice. Therefore, there must be other cellular components with critical roles in maintaining HSCs in the developing liver. Overall, it is not clear what cells create a niche for HSCs in the developing liver in vivo.
A niche is defined as the cellular microenvironment that generates necessary cytokines and growth factors for HSC maintenance in vivo. To uncover the HSC niche components in the developing liver, we focused on identification of the cellular sources of key factors required for HSC maintenance in vivo. Stem cell factor (SCF) is one of the very few cytokines known to be genetically required by HSCs. Signaling through the cKIT receptor on HSCs, SCF is essential for HSC maintenance in vivo (Barker, 1994; Ding et al., 2012). Scf transcript is alternatively spliced into a short and a long form, with the exclusion or inclusion of an exon containing a proteolytic cleavage site. This leads to two forms of SCF protein, a membrane-bound form and a soluble form with protease-mediated cleavage (Broudy, 1997). The membrane-bound form of SCF is particularly important for HSC maintenance and implies that its engagement with HSCs is within a local niche (Barker, 1997; Wolf, 1978). Scf knockout mice have a profound loss of fetal liver HSCs and are perinatal lethal due to severe hematopoietic defects (Ding et al., 2012; Ikuta and Weissman, 1992). Furthermore, administration of a cKIT functionally blocking antibody during development at times after but not before the liver hematopoietic stages abolishes hematopoietic progenitors, suggesting that the SCF–cKIT pathway is required for HSC maintenance starting from the fetal liver stage (Ogawa et al., 1993). Based on antibody staining, PCR, and a bacterial artificial chromosome transgenic Scf reporter allele with likely incomplete promoter, Scf has been suggested to be expressed by hepatic progenitors (Azzoni et al., 2018; Chou and Lodish, 2010), Nestin+NG2+ periportal cells (Khan et al., 2016), and endothelial cells (Azzoni et al., 2018). Yet no study has systematically examined which sources of SCF are functionally important for HSCs in the developing liver in vivo. Here, we used Scfgfp knockin and Scffl conditional knockout mice to systemically characterize the expression of Scf and to uncover functionally important cellular sources of SCF for HSC maintenance in the developing liver in vivo. We found that most HSCs localized close to the sinusoidal vasculature, and endothelial and stellate cells are the major functional source of SCF in the perinatal liver. Interestingly, stellate cells down-regulated SCF and several other putative HSC maintenance factors around birth, suggesting that the developing liver is uniquely empowered with HSC-supporting function.
HSCs reside in a perisinusoidal vascular niche in the developing liver
HSCs mature and expand in the fetal liver (Ema and Nakauchi, 2000; Morrison et al., 1995). Around birth, HSCs migrate to and take up residence in the bone marrow (Lee et al., 2017). However, the perinatal time window during which the liver maintains HSCs has not been assessed. We found that liver CD150+CD48−Lin−Sca1+cKit+ HSC frequency and number increased from embryonic day 12.5 (E12.5), peaked at about E15.5, and started to decline from postnatal day 0 (P0; Fig. 1, A and B). By P5, there was a fourfold decrease in fetal liver HSC frequency compared with E15.5. HSCs were essentially undetectable in the liver by P10 (Fig. 1, A and B). To test liver HSC function at these developmental stages, we transplanted 500,000 whole liver cells from E15.5, P0, P5, and P10 mice into lethally irradiated recipient mice along with 500,000 recipient-type bone marrow cells. P0 liver cells had a similar capacity to reconstitute all major hematopoietic lineages (myeloid, B, and T cells) in irradiated mice compared with E15.5 liver cells (Fig. 1 C and Fig. S1 A). The reconstitution potential of P5 liver cells was significantly reduced, although substantial functional HSCs were still present in the liver (Fig. 1 C and Fig. S1 A). P10 liver cells showed a further decline with minimal if any multilineage reconstitution potential (Fig. 1 C and Fig. S1 A). These data suggest that the newborn liver still harbors a functional niche that maintains HSCs. We thus analyzed P0 developing livers throughout our study unless otherwise noted.
We performed SLAM staining on P0 liver sections to localize HSCs in situ and found that CD150+CD48−Lin− candidate HSCs were closely associated with Laminin+ sinusoidal endothelial cells throughout the liver (Fig. 1, D–H; and Fig. S1, B–E). Similarly, E15.5 fetal liver HSCs were also closely associated with sinusoidal endothelial cells (Fig. S1, F–M). Compared with random mononuclear cells or Lin+CD48+ mature bone marrow cells, HSCs localized significantly closer to sinusoidal endothelial cells (Fig. 1 I). Similar results were obtained when vascular endothelial (VE)–cadherin was used as an endothelial cell marker (Fig. S1, N–S). These data suggest that HSCs localize within a perisinusoidal vascular niche in the developing liver.
Scf is primarily expressed by endothelial cells and hepatic stellate cells
The close association of HSCs to sinusoidal vasculature prompted us to investigate what cells are the source of a key cytokine, SCF, within the perivascular niche. We systematically analyzed Scf expression pattern in the developing liver using Scfgfp knockin mice (Ding et al., 2012). Flow-cytometric analyses of enzymatically dissociated P0 Scfgfp/+ livers showed that GFP+ cells were rare (∼1%) and negative for CD45 and Ter119, markers of hematopoietic cells, indicating that Scf is expressed by nonhematopoietic cells (Fig. 2 A and Fig. S2 A). About 83% of CD45−Ter119−CD31+ endothelial cells were positive for GFP, demonstrating that endothelial cells are a source of SCF (Fig. 2 B).
At P0, ∼20% of GFP+ stromal cells were CD31−, suggesting that Scf is also expressed by stromal cells other than endothelial cells in the developing liver (Fig. 2 C). To elucidate the identity of these cells, we sorted CD45−Ter119−CD31−Scf-GFP+ cells by flow cytometry and analyzed their gene expression by quantitative RT-PCR (qRT-PCR). Compared with whole liver, hematopoietic, or CD31+ endothelial cells, CD45−Ter119−CD31−Scf-GFP+ cells expressed significantly higher levels of Desmin, Lrat, Lhx2, and Pdgfrα, markers of liver-specific perisinusoidal stellate cells (Mederacke et al., 2013; Tsutsumi et al., 1987; Wandzioch et al., 2004; Wright et al., 2014; Yin et al., 2013; Fig. 2 D), suggesting that these Scf-GFP+ cells are hepatic stellate cells. Cxcl12, a chemokine required for HSC maintenance and homing (Ara et al., 2003; Ding and Morrison, 2013; Greenbaum et al., 2013; Sugiyama et al., 2006), was also highly expressed by the nonendothelial CD45−Ter119−CD31−Scf-GFP+ cells (Fig. 2 D). It has been reported that fetal liver stellate cells express Cxcl12 (Kubota et al., 2007). Consistently, using Cxcl12DsRed/+ knockin reporter mice (Ding and Morrison, 2013), we found that virtually all (95%) Cxcl12-DsRed+ cells were Desmin+ in P0 livers (Fig. 2, E–J). CD45−Ter119−Cxcl12-DsRed+ stromal cells from P0 livers of Cxcl12DsRed/+ mice consistently expressed significantly higher levels of Desmin, Lrat, Lhx2, and Pdgfrα compared with whole liver cells (Fig. S2 B). We systematically examined the expression of Cxcl12 in P0 livers and found that Cxcl12 was not expressed by hematopoietic, hepatic, or CD31+ endothelial cells (Fig. S2, C and D). Mature hepatic stellate cells have the unique characteristic of storing vitamin A in their cytoplasm (Mederacke et al., 2013). Indeed, the majority of CD45−Ter119−Cxcl12-DsRed+ stromal cells were vitamin A+ by flow cytometry analysis (Fig. 2 K). Taken together, our data established Cxcl12-DsRed as a specific marker for hepatic stellate cells in the developing liver.
We generated Scfgfp/+; Cxcl12DsRed/+ mice to directly test whether hepatic stellate cells are indeed a cellular source of SCF in the developing liver. By flow cytometry, ∼78% of CD45−Ter119−Cxcl12-DsRed+ stellate cells were Scf-GFP+ (Fig. 2 L). Cxcl12+ stellate cells accounted for the majority (81%) of nonendothelial CD45−Ter119−CD31−Scf-GFP+ cells (Fig. 2 M). Thus, besides endothelial cells, stellate cells are a major source of SCF in the developing liver.
qRT-PCR analyses revealed that endothelial cells and hepatic stellate cells expressed Scf at ∼155-fold and ∼50-fold the levels found in whole P0 liver cells, respectively (Fig. 2 N). Consistent with our flow cytometry data (Fig. 2 A and Fig. S2 A), hematopoietic cells and hepatocytes expressed very little if any Scf (Fig. 2 N). Ng2+ cells expressed an ∼12-fold higher level of Scf compared with whole liver cells (Fig. 2 N). The short (membrane-bound) and long (with the potential to become soluble) forms of Scf showed similar expression patterns in endothelial and stellate cells (Fig. S2 E). Endothelial and stellate cells were also the major source of Scf in E15.5 fetal livers, although at this stage, stellate cells expressed Scf at a higher level than endothelial cells (Fig. S2, F–O). Throughout development, stellate cells gradually mature with up-regulated expression of Desmin and Lrat, and accumulation of vitamin A (Fig. S2, P and Q). Taken together, these data suggest that endothelial cells and hepatic stellate cells are the major sources of SCF in the developing liver.
CD150+CD48−Lin− HSCs were closely associated with Desmin+ perisinusoidal stellate cells (Fig. S3, A–F). This raised the possibility that HSCs reside in a perisinusoidal vascular niche composed of endothelial cells and hepatic stellate cells in the developing liver.
HSCs do not require SCF from hepatic cells in the developing liver
To test which cells are the functionally important sources of SCF for HSC maintenance in the developing liver in vivo, we systematically deleted Scf from candidate niche cells. Hepatic lineage cells have been proposed to be a candidate niche cell type for fetal liver HSCs (Chou and Lodish, 2010). But it is not clear whether hepatocytes are a source of SCF for developing liver HSC maintenance in vivo. Albumin-cre effectively recombines genes in fetal and postnatal hepatocytes (Weisend et al., 2009). Consistent with this, we found that Albumin-cre recombined a loxp-tdTomato+ reporter in HNF4α+ hepatic cells of P0 livers efficiently (Fig. 3, A–E). To test whether hepatocytes are an important source of SCF for HSC maintenance in the developing liver, we generated and analyzed newborn Albumin-cre; Scffl/− mice. P0 livers from Scf+/− germline heterozygous mice exhibited a normal HSC frequency but an approximately twofold decline in cellularity compared with WT controls (Fig. 3, F and G). No discernable morphological abnormalities were observed in the P0 livers from Scf+/− mice, suggesting that the decrease in cellularity is largely due to loss of hematopoietic cells (Fig. S3, G–V). Deletion of Scf from hepatic cells in Albumin-cre; Scffl/− mice did not significantly reduce HSC frequency in the developing liver compared with either WT or germline Scf+/− heterozygous mice (Fig. 3 F). Compared with Scf+/− heterozygous mice, deletion of the second allele of Scf from hepatic cells in Albumin-cre; Scffl/− mice did not lead to a further loss of liver cellularity or HSC number (Fig. 3, G and H). To determine whether functional HSCs were impaired in the developing livers of Albumin-cre; Scffl/− mice, we transplanted 500,000 P0 liver cells from Albumin-cre; Scffl/−, littermate Scf+/+, or Scf+/− mice along with 500,000 recipient-type bone marrow cells into lethally irradiated recipient mice. Consistent with the normal HSC frequency in Scf+/− livers (Fig. 3 F), 500,000 P0 liver cells from Scf+/+ and Scf+/− mice gave similar reconstitution (Fig. S3 W). Thus, hereafter throughout this study, reconstitution data from liver cells of Scf+/+ and/or Scf+/− mice were combined as controls unless otherwise noted. Importantly, P0 liver cells from Albumin-cre; Scffl/− mice had normal reconstitution activity compared with controls (Fig. 3 I). These data demonstrate that Scf from hepatocytes is not required for HSC maintenance in the developing liver in vivo.
HSCs do not require SCF from hematopoietic cells in the developing liver
Although hematopoietic cells express very little if any Scf (Fig. 2, A and N; and Fig. S2 A), it was still possible that hematopoietic cells are an important functional source of SCF for HSC maintenance in the developing liver. We thus conditionally deleted Scf from hematopoietic cells using Vav1-cre (de Boer et al., 2003). Consistent with prior reports (de Boer et al., 2003; Ding et al., 2012), recombination was highly efficient in HSCs and other hematopoietic cells from Vav1-cre; Scffl/− P0 livers (Fig. 4, A and B). Vav1-cre; Scffl/− mice had normal HSC frequency in the P0 livers relative to WT or germline Scf+/− heterozygous littermate control mice (Fig. 4 C). Compared with Scf+/− mice, Vav1-cre; Scffl/− mice had similar cellularity and HSC numbers in the P0 livers (Fig. 4, D and E). P0 liver cells from Vav1-cre; Scffl/− mice had normal reconstitution capacity in all major blood lineages (Fig. 4 F). Therefore, Scf expressed by hematopoietic cells is not required for HSC maintenance in the developing liver.
HSCs do not require SCF from Ng2+ cells in the developing liver
Ng2+ periportal stromal cells have been proposed to be a component of the fetal liver niche (Khan et al., 2016). Consistent with the prior report, we found that Ng2-cre recombined in α-SMA+ periportal stromal cells in the developing liver (Fig. 5, A–E). Deletion of Scf from periportal stromal cells in Ng2-cre; Scffl/− mice did not significantly reduce HSC frequency in the developing liver (Fig. 5 F). P0 liver cellularity and HSC number did not significantly differ in Ng2-cre; Scffl/− mutants compared with littermate Scf+/− controls (Fig. 5, G and H). Moreover, P0 liver cells from Ng2-cre; Scffl/− mice had normal long-term myeloid and T cell reconstituting activity and a slight increase in B cell lineage reconstitution compared with controls (Fig. 5 I), consistent with a lack of hematopoietic reconstitution defects when Ng2+ cells are genetically ablated (Khan et al., 2016). These data demonstrate that Scf from Ng2-cre+ cells do not contribute to HSC maintenance in the developing liver.
Deletion of Scf from endothelial cells does not cause significant HSC depletion
Tie2-cre recombines in endothelial and hematopoietic cells (Ding et al., 2012; Kisanuki et al., 2001). We generated Tie2-cre; Scffl/− mice to conditionally delete Scf from these cells. Since hematopoietic cells do not express Scf (Fig. 2, A and N; and Fig. S2 A) and Scf from hematopoietic cells is not required for HSC maintenance in the developing liver (Fig. 4), this model allowed us to investigate whether endothelial cells are a critical source of SCF for HSC maintenance. As expected, Tie2-cre efficiently recombined in Scf-expressing CD31+ endothelial cells in P0 livers (Fig. 6, A and B). Tie2-cre; Scffl/− mutants had normal HSC frequency compared with littermate controls in the developing liver (Fig. 6 C). Liver cellularity and the number of HSCs in the developing livers of Tie2-cre; Scffl/− mutants did not differ significantly from littermate Scf+/− controls (Fig. 6, D and E). Competitive reconstitution assays revealed that P0 liver cells from Tie2-cre; Scffl/− mice had normal reconstitution capacity compared with controls (Fig. 6 F). Thus, deletion of Scf from endothelial cells alone does not significantly affect either the number or the function of HSCs in the developing liver.
Deletion of Scf from hepatic stellate cells leads to loss of HSCs in the developing liver
We found that Pdgfrα, a mesenchymal cell marker (Andrae et al., 2008; Li et al., 2018), was highly expressed by CD45−Ter119−Cxcl12-DsRed+ hepatic stellate cells in the developing liver (Fig. 2 D and Fig. S2 B), raising the possibility that we could conditionally delete Scf from stellate cells using a Pdgfrα-cre transgenic line (Roesch et al., 2008). We generated Pdgfrα-cre; loxp-tdTomato mice to examine its recombination pattern in the developing liver. About 90% of tdTomato+ cells were Desmin+ stellate cells (Fig. 7, A–E), with minimal if any recombination in endothelial cells (0%), hematopoietic cells (0.3%), or hepatocytes (0.6%; Fig. S4, A–J). A minimal number of α-SMA+ cells (5%) were targeted by Pdgfrα-cre in P0 livers (Fig. S4, K–O). Thus, the Pdgfrα-cre transgenic line recombines specifically and efficiently in perisinusoidal stellate cells in the developing liver.
To test whether Pdgfrα-cre–expressing stellate cells are an important source of SCF for HSC maintenance, we generated Pdgfrα-cre; Scffl/fl and Pdgfrα-cre; Scffl/− mice. These mice had normal HSC frequency in the developing liver compared with controls (Fig. 7 F). However, liver cellularity as well as HSC number in the P0 livers from Pdgfrα-cre; Scffl/fl or Pdgfrα-cre; Scffl/− mice were significantly reduced compared with either Scf+/+ or Scf+/− littermate controls (Fig. 7, G and H). Importantly, 500,000 P0 liver cells from Pdgfrα-cre; Scffl/fl or Pdgfrα-cre; Scffl/− mice gave modest but significantly lower levels of reconstitution compared with controls when transplanted into lethally irradiated recipient mice (Fig. 7 I). We also analyzed these mice at E15.5 and did not observe significant decreases in liver cellularity or HSC number compared with Scf+/− littermate controls (Fig. S5, A–C). E15.5 liver cells from these mutant mice had normal long-term multilineage reconstitution activity (Fig. S5 D), suggesting that the depletion of HSCs in P0 livers (Fig. 7, F–I) was not due to developmental defects before E15.5. These data suggest that HSC niche in the mutant mice was functionally compromised in the developing liver. We also analyzed these mice at P5, when the liver still contains a substantial number of HSCs (Fig. 1, A–C). Compared with controls, Pdgfrα-cre; Scffl/fl or Pdgfrα-cre; Scffl/− mice had a significant decrease of HSC frequency (Fig. 7 J). Liver HSC number was markedly depleted at P5 (Fig. 7, K and L). These data suggest that HSCs are progressively depleted in the developing livers of Pdgfrα-cre; Scffl/fl or Pdgfrα-cre; Scffl/− mice.
Compared with littermate controls, P5 Pdgfrα-cre; Scffl/fl mice were smaller in size, with smaller and pale spleens and livers (Fig. 7, M–O) and significant decreases in white blood cell, red blood cell, and platelet counts in the peripheral blood (Fig. S5, E–G). By P10, Pdgfrα-cre; Scffl/fl mice had severe growth retardation with small spleens and livers (Fig. S5, H–J). The vast majority of the Pdgfrα-cre; Scffl/fl mice (14 out of 15 mutants from 11 independent litters) died before the weaning age (P21), while none of the controls did so (Fig. S5 K). The preweaning death of Pdgfrα-cre; Scffl/fl mice could be rescued by transplantation of WT bone marrow cells (four out of four mutants survived to adulthood after the bone marrow transplantation; Fig. S5 L), suggesting that these mice died of hematopoietic failure. Taken together, our data demonstrate that loss of Scf from hepatic stellate cells leads to loss of HSCs in the developing liver and early postnatal lethality due to hematopoietic defects.
HSCs require SCF from both endothelial and hepatic stellate cells
Both endothelial and hepatic stellate cells are the major source of SCF in the developing liver (Fig. 2 and Fig. S2). However, deletion of Scf from endothelial cells did not result in significant HSC depletion (Fig. 6, C–F), while deletion of Scf from stellate cells resulted in only modest reconstitution defects (Fig. 7 I). This raised the question of whether SCF from endothelial and hepatic stellate cells synergistically maintains HSCs in the developing liver. To directly test this, we generated Pdgfrα-cre; Tie2-cre; Scffl/− mice.
Pdgfrα-cre; Tie2-cre; Scffl/− mice were born smaller with pale livers, suggesting a hematopoietic defect in the liver (Fig. 8, A and B). P0 livers from Pdgfrα-cre; Tie2-cre; Scffl/− mice had significant decreases in HSC frequency, cellularity, and HSC number compared with controls (Fig. 8, C–E). P0 liver cells from Pdgfrα-cre; Tie2-cre; Scffl/− mice had a significantly reduced capacity to reconstitute lethally irradiated recipient mice in competitive reconstitution assays (Fig. 8 F), with significantly lower contribution to bone marrow HSCs and Lin−Sca1+cKit+ (LSK) progenitors in the recipient mice compared with controls (Fig. 8 G). The hematopoietic defects of the mutant mice became more prominent at P5 (Fig. 8, H and I). Strikingly, ∼90% of HSCs were depleted from the P5 livers of Pdgfrα-cre; Tie2-cre; Scffl/− mice (Fig. 8, J–L). Importantly, multilineage reconstitution activity was largely depleted from P5 liver cells of Pdgfrα-cre; Tie2-cre; Scffl/− mice upon transplantation, with mutant liver cells contributing significantly lower levels of HSCs and LSK progenitors in the recipients (Fig. 8, M and N). These data suggest that endothelial and stellate cells are the major source of SCF in the developing liver.
At E14.5, livers from Pdgfrα-cre; Tie2-cre; Scffl/− mice had normal HSC frequency, cellularity, and HSC numbers compared with Scf+/− heterozygous mice (Fig. S5, M–O). Notably, E14.5 liver cells from Pdgfrα-cre; Tie2-cre; Scffl/− mice had normal reconstitution capacity (Fig. S5 P). These data suggest that the HSC depletion observed in the P0 developing livers from Pdgfrα-cre; Tie2-cre; Scffl/− mice (Fig. 8) does not reflect developmental defects before E14.5. Taken together, these data strongly suggest that endothelial cells cooperate with hepatic stellate cells to promote HSC maintenance in the developing liver by producing SCF.
Hepatic stellate cells progressively lose HSC niche gene expression signature in the neonatal liver
From E15.5 to P5, stellate cells significantly down-regulated Scf-GFP expression (10.8-fold), while endothelial cells did so to a much lesser extent (2.7-fold; Fig. 9 A). These data suggest that gene expression changes in stellate cells may deplete HSC-supporting activity and drive HSC egress from the liver around birth. To study these gene expression changes in detail, we conducted a genome-wide expression profiling of hepatic stellate cells from E15.5, P5, and P10, when HSCs gradually migrate out of the liver (Fig. 1 C and Fig. S1 A). 2,441 genes were significantly up-regulated and 2,253 genes were significantly down-regulated (P ≤ 0.05, fold-change two or greater) in hepatic stellate cells comparing E15.5 to either P5 or P10 (Fig. 9 B). Gene ontology analysis using the Database for Annotation, Visualization and Integrated Discovery comparing stellate cells at E15.5 and P10 revealed several significantly enriched biological processes, including oxidation reduction, metabolic process, and liver development (Fig. 9 C).
Consistent with the loss of HSC-supporting activity from the liver by P10, gene set enrichment analysis (GSEA) showed that P10 hepatic stellate cells have expression profiles significantly depleted of hematopoietic stem and progenitor cell–supporting genes (Charbord et al., 2014) compared with those from E15.5 livers (Fig. 9 D). The expression levels of HSC-supporting factors, such as Scf (Barker, 1994), Vcam1 (Dutta et al., 2015), Pleiotrophin (Himburg et al., 2012), and Jagged1 (Poulos et al., 2013), in hepatic stellate cells decreased significantly over time from E15.5 to P10 (Fig. 9 E), correlating with the gradual depletion of HSC-supporting activity in the livers from E15.5 to P10. Interestingly, principal component analysis (PCA) showed that E15.5 hepatic stellate cells are more similar to bone marrow Scf-GFP+ mesenchymal stromal cells, the key niche cells in the adult bone marrow (Ding and Morrison, 2013; Ding et al., 2012; Lee et al., 2017), than adult hepatic stellate cells at the transcriptome level (Fig. 9 F). Collectively, these data suggest that fetal stellate cells are endowed with HSC-supporting activity, similar to adult bone marrow mesenchymal cells, and this activity is lost around birth.
Here, we undertook a systematic genetic approach to functionally dissect the HSC niche in the developing liver in vivo. We found that HSCs localized in a perisinusoidal vascular niche where endothelial cells and hepatic stellate cells are two major cellular sources of SCF. Both cell types are functionally important as combined deletion of Scf from endothelial cells and hepatic stellate cells led to profound HSC depletion (Fig. 8). Hepatic stellate cells play important roles in storage of vitamin A in adult livers (Yin et al., 2013). But their role during hematopoietic development was previously unknown. Our results suggest that fetal hepatic stellate cells are a critical component of the HSC niche. Whole body deletion of stellate cell–specific transcriptional factors, Lhx2 and Hlx, resulted in fetal liver hematopoietic defects in a non–cell-autonomous manner (Hentsch et al., 1996; Porter et al., 1997), supporting the notion that stellate cells are a critical component of the liver hematopoietic niche, although HSCs were not analyzed in these studies. In line with these data, fetal hepatic stellate cells are associated with hematopoietic foci and have a greater capacity in maintaining hematopoiesis in vitro than other stromal cells in the fetal liver (Kordes et al., 2013).
Our genetic approach does not distinguish the function of membrane-bound or soluble SCF, although it likely acts locally (Wolf, 1978). Consistent with this, we found that most HSCs are close to endothelial and stellate cells, the major source of SCF, in the developing liver. Through systematic conditional deletion in candidate cell types, we found that endothelial and stellate cells are the major, if not exclusive, source of SCF, as ∼90% HSCs are lost when Scf is deleted from both endothelial and hepatic stellate cells (Fig. 8). It is not clear why endothelial and stellate cells synergistically promote HSC maintenance. The redundancy between these cells as sources of SCF may offer extra robustness to the system, and thus an evolutional advantage. Khan et al. (2016) identified Nestin+NG2+ periportal stromal cells as an important cellular component of the fetal liver niche. However, ablation of these cells resulted in enhanced reconstitution activity, mainly in the T and B cell lineages (Khan et al., 2016), suggesting that NG2+ periportal stromal cells are not required for the maintenance of functional HSCs in the fetal liver. Consistent with this, we did not observe depletion of HSC number or function when Scf was deleted from Ng2-cre–expressing cells (Fig. 5, F–I). Hepatocytes are a critical source of thrombopoietin for bone marrow HSC maintenance (Decker et al., 2018). But it appears that thrombopoietin is not required for E14.5 fetal liver HSC maintenance, suggesting distinct fetal and adult HSC maintenance mechanisms (Qian et al., 2007). Additional cell types may also contribute to the HSC niche in the developing liver by generating other HSC-supporting factors.
Murine fetal liver hematopoiesis starts from around E12.5 (Mikkola and Orkin, 2006). Blockade of the SCF-cKIT pathway by administration of cKIT functional blocking antibody after E12.5 quickly eliminates hematopoietic progenitors (within 2 d), while applying the same blocking antibody before E12.5 does not affect those progenitors, suggesting that hematopoiesis before the fetal liver stage is SCF independent (Ogawa et al., 1993). Alternatively, using a whole-body knockout approach, a recent report suggests that SCF is required for preHSCs before the fetal liver stage (Azzoni et al., 2018). Although further investigation is clearly needed to address the role of SCF in early hematopoiesis, it is well established that SCF is required for fetal liver HSCs (Broudy, 1997; Ding et al., 2012; Ikuta and Weissman, 1992). We found that conditional deletion of Scf from Pdgfrα-cre and/or Tie2-cre–expressing cells does not lead to HSC depletion in early (E14.5 or E15.5) livers (Fig. S5). In contrast, we observed significant HSC depletion in later (P0 and P5) developing livers (Fig. 7 and Fig. 8). Although antibody blockade has faster kinetics (Ogawa et al., 1993) than our conditional genetic models, our data suggest that the observed HSC depletion in later developmental stages in our study reflects disruption of niche function and a critical role of SCF for HSCs in the developing liver.
Materials and methods
Scfgfp, Scffl, Scf −, and Cxcl12DsRed mice were described previously (Ding and Morrison, 2013; Ding et al., 2012). Vav1-cre (de Boer et al., 2003), Albumin-cre (Weisend et al., 2009), Tie2-cre (Kisanuki et al., 2001), Pdgfrα-cre (Roesch et al., 2008), Ng2-cre (Zhu et al., 2008), and loxp-tdTomato (Madisen et al., 2010) mice were obtained from The Jackson Laboratory and maintained on C57BL/6 background. All mice were housed in a specific pathogen–free, Association for the Assessment and Accreditation of Laboratory Animal Care–approved unit at Columbia University Medical Center. All protocols were approved by Columbia University Committee on the Institute Animal Care and Use. Unless otherwise noted, data are mean ± SD, and two-tailed Student’s t tests were used to evaluate statistical significance (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).
Long-term competitive reconstitution assay
A cesium-137 irradiator (JL Shepherd and Associates) was used to lethally irradiate (total 1,080 rads in a split dose) adult recipient mice. After anesthetization, 5 × 105 developing liver cells along with 5 × 105 recipient bone marrow cells were retro-orbitally injected into the recipients. Mice were maintained on antibiotic water (Baytril 0.17 gl−1) for 14 d after transplantation, then changed to regular water. Recipient mice were bled every 4 wk up to at least 16 wk after transplantation to evaluate the level of donor-derived hematopoietic lineages including myeloid, B, and T cells. Blood was incubated with ammonium chloride potassium red cell lysis before proceeding to antibody staining. Cells were stained with anti-CD45.2 (104), anti-CD45.1 (A20), anti-Gr1 (8C5), anti-Mac-1 (M1/70), anti-B220 (6B2), and anti-CD3 (KT31.1; Biolegend) and analyzed by flow cytometry. For transplantation of neonatal mice, 1 million freshly prepared bone marrow cells from CD45.1 mice were injected into the liver or temporal vein of neonatal mice (P0–P3). The mice were then followed for up to 4 mo.
Liver cells were isolated by crushing the tissue between two glass slides. The cells were passed through 25-G syringes with Ca2+ and Mg2+ free HBSS with 2% heat-inactivated bovine serum for single cell suspension. Then cells were filtered through 70-µm nylon mesh. For staining HSCs, the following antibodies were used: lineage markers (anti-Ter119, anti-B220 [6B2], anti-Gr1 [8C5], anti-CD2 [RM2-5], anti-CD3 [17A2], anti-CD5 [53–7.3], and anti-CD8 [53–6.7]), anti-CD150 (TC15-12F12.2), anti-CD48 (HM48-1), anti-Sca-1 (E13-161.7), and anti-cKit (2B8; Biolegend). DAPI was used to preclude dead cells. For flow-cytometric analysis of Scf-GFP+, Cxcl12-DsRed+, and endothelial cells, the tissue was digested with Collagenase IV (200 U ml−1) and DNase I (200 U ml−1) at 37°C for 20 min. Digested tissue was further dissociated by repetitive pipetting and passed through 70-μm nylon mesh. Samples were then stained with antibodies and assessed by flow cytometry. Anti-CD45 (30F-11) and anti-Ter119 antibodies (Biolegend) were used to label hematopoietic cells. For staining endothelial cells, samples were incubated with anti-CD31 (MEC13.3; Biolegend) antibody. To detect autofluorescence of vitamin A, cells were excited with a 405-nm laser, and the emission signal was detected with a 450/50-nm bandpass filter. To exclude dead cells, 7-AAD (BD Pharmingen) was used. Samples were run on FACSAria II, BD LSR II, FACSCelesta, or FACSCanto flow cytometers. FACSDiva (BD) or FlowJo (Tree Star) software was used for data analysis.
Immunostaining of liver sections
Freshly dissected livers were fixed in 4% paraformaldehyde for 1–3 h followed by overnight incubation in 30% sucrose in PBS. The tissues were embedded in optimal cutting temperature compound and snap-frozen on dry ice. Liver sections were cut at 10 µm using a CryoJane system (Instrumedics) and air-dried overnight at room temperature. Sections were rehydrated in PBS for 5 min and blocked using either 5% goat serum or 5% donkey serum in PBS for 30 min. Primary antibodies were applied overnight at 4°C. Then slides were incubated in secondary antibodies at room temperature for 2 h with washes in between. Primary antibodies were rabbit-anti-Laminin antibody (L9393; Sigma-Aldrich), rabbit-anti-Desmin antibody (GTX103557; GeneTex), goat-anti-Desmin antibody (AF3844; R&D Systems) rabbit-anti-HNF4α antibody (ab201460; Abcam), goat-anti-VE-cadherin antibody (BAF1002; R&D Systems), goat-anti-Albumin-FITC antibody (A90-234F; Bethyl Laboratories), and mouse-anti-α-SMA-FITC antibody (F3777; Sigma-Aldrich). Secondary antibodies were anti-rabbit Alexa Fluor 488 (Thermo Fisher Scientific), anti-rabbit Alexa Fluor 555 (Thermo Fisher Scientific), anti-goat Alexa Fluor 488 (Thermo Fisher Scientific), anti-goat Alexa Fluor 555 (Thermo Fisher Scientific), Alexa Fluor 647–conjugated streptavidin (Jackson ImmunoResearch), and anti-FITC Alexa Fluor 488 (Jackson ImmunoResearch). For localizing CD150+CD48−lineage− candidate HSCs, slides were incubated in rat-anti-CD150 antibody (TC15-12F12.2; Biolegend) overnight at 4°C. CD150 was visualized by incubation in anti-rat Alexa Fluor 555 antibody (Thermo Fisher Scientific) for 2 h at room temperature with three washes in between. Rat IgG (Sigma-Aldrich) was then applied to the slides for 10 min and washed with PBS. FITC-conjugated anti-B220, anti-Gr-1, anti-Mac1, anti-CD5, anti-CD8a, anti-CD2, anti-CD3, anti-CD41, anti-Ter119, and anti-CD48 antibodies (Biolegend) along with rabbit-anti-Laminin antibody (Sigma-Aldrich) or rabbit-anti-Desmin antibody (GeneTex) were used for subsequent primary staining at 4°C overnight. Anti-FITC Alexa Fluor 488 (Jackson ImmunoResearch) and anti-rabbit Alexa Fluor 647 (Thermo Fisher Scientific) were used for secondary staining at room temperature for 2 h. Slides were mounted with Prolong gold antifade (Invitrogen), and images were acquired on Nikon Ti Eclipse confocal microscopes. For visualizing CD150+CD48−lineage− candidate HSCs with VE-cadherin+ endothelial cells, after staining for CD150 as described above, the slides were stained with VE-cadherin antibody at 4°C overnight. Anti-goat Alexa Fluor 488 (Thermo Fisher Scientific) was used for secondary staining followed by blocking with goat serum for 1 h at room temperature. Slides were then washed with PBS and were incubated with biotin-conjugated anti-B220, anti-Gr-1, anti-Mac1, anti-CD5, anti-CD8a, anti-CD2, anti-CD3, anti-CD41, anti-Ter119, and anti-CD48 antibodies (Biolegend) at 4°C overnight. Alexa Fluor 647–conjugated streptavidin (Jackson ImmunoResearch) was applied for 2 h at room temperature.
Quantitative PCR (qPCR)
Enzymatically digested cells were sorted directly into Trizol. Total RNA was purified according to the manufacturer’s instructions and subjected to reverse transcription using SuperScript III Reverse Transcriptase (Invitrogen). The levels of gene transcripts were quantified by quantitative real-time PCR using GoTaq qPCR Master Mix (Promega) on a CFX Connect Real-Time PCR Detection System (Bio-Rad). β-Actin was used to normalize the expression of genes. Primers used in this study were as follows: Scf: 5′-GTCACAGGATTCCCGCAG-3′ and 5′-AGCGCTGCCTTTCCTTATG-3′; shortScf: 5′-GAGGCCAGAAACTAGATCCTTT-3′ and 5′-TAAGGCTCCAAAAGCAAAGC-3′; longScf: 5′-GCCAGAAACTAGATCCTTTACTCCTGA-3′ and 5′-ACATAAATGGTTTTGTGACACTGACTCTG-3′; Cxcl12: 5′-TGCATCAGTGACGGTAAACCA-3′ and 5′-GTTGTTCTTCAGCCGTGCAA-3′; Desmin: 5′-CCTGGAGCGCAGAATCGAAT-3′ and 5′-TGAGTCAAGTCTGAAACCTTGGA-3′; Lrat: 5′-GCAGTTGGGACTGACTCCAT-3′ and 5′-CAGATTGCAGGAAGGGTCAT-3′; Lhx2: 5′-AGTGACCGGGCAGCGTTGTGT-3′; and 5′-GAGCGCGCATCACCATCTCTGA-3′; Pdgfrα: 5′-TTGACCCTGTTCCAGAGGAG-3′ and 5′-CACCAGGTCCGAGGAATCTA-3′; β-actin: 5′-GCTCTTTTCCAGCCTTCCTT-3′ and 5′-CTTCTGCATCCTGTCAGCAA-3′; VE-cadherin: 5′-TCCTCTGCATCCTCACCATCACA-3′ and 5′-GTAAGTGACCAACTGCTCGTGAAT-3′; and Tie2: 5′-ATGTGGAAGTCGAGAGGCGAT-3′ and 5′-CGAATAGCCATCCACTATTGTCC-3′. For quantifying deletion efficiency of Vav1-cre, CD45/Ter119+ hematopoietic cells were sorted into tail DNA digestion buffer (Viagen Biotech) and digested. Genomic DNA was used as template for qPCR. The Cxcl12 genomic region was used to normalize the input concentrations. The following primers were used: Cxcl12 genomic region: 5′-GAGCCCAGAACTCTGCCACC-3′ and 5′-TCTTGCAAAGACCATCCCCTC-3′; and Scf genomic region: 5′-GGAAAAGAACCAAGTGAAGTC-3′ and 5′-GTCCGCAGCAAGCTCACCAGC-3′. Scffl/fl genomic DNA was used as control.
Gene expression profiling and analysis
50,000 CD45−Ter119−Cxcl12-DsRed+ stromal cells from freshly prepared livers of E15.5, P5, and P10 Cxcl12DsRed/+ mice were sorted into Trizol by flow cytometry. Total RNA was isolated and amplified with the WT-Ovation Pico RNA Amplification system (Nugen) according to the manufacturer’s instructions. Sense strand cDNA was generated using the WT-Ovation Exon Module (Nugen). Then cDNA was fragmented and labeled using FL-Ovation DNA Biotin Module V2 (Nugen). The labeled cDNA was hybridized to Affymetrix Mouse Gene ST 1.0 chips following the manufacturer's instructions. Expression values for all probes were normalized and determined using the robust multi-array average method with an Affymetrix Expression Console (Irizarry et al., 2003). Normalized data were analyzed using Transcriptome Analysis Console 2.0. Significantly up- or down-regulated genes (P ≤ 0.05 and fold-change two or greater) were used to carry out gene ontology analysis using the Database for Annotation, Visualization and Integrated Discovery online tools (Huang et al., 2009). GSEA analysis was conducted as previously described (Subramanian et al., 2005). A gene set of 297 genes up-regulated in the HSC supportive stromal cell lines compared with nonsupportive stromal lines (Charbord et al., 2014) was downloaded from Broad Institute website. Heatmap analysis and PCA were performed using MATLAB2016. The microarray data were deposited to the Gene Expression Omnibus under accession no. GSE159308.
Online supplemental material
Fig. S1 shows multilineage reconstitution of perinatal liver cells and perivascular localization of HSCs in the developing liver. Fig. S2 shows hepatic stellate cells are a major source of SCF in the developing liver, in addition to endothelial cells. Fig. S3 shows that HSCs are close to hepatic stellate cells in the developing liver, and Scf+/− developing liver structure and reconstitution activity are normal. Fig. S4 shows that the recombination pattern of Pagfrα-cre is specific to stellate cells in the developing liver. Fig. S5 shows that deletion of Scf from stellate cells does not lead to HSC depletion before E15.5 but leads to severe hematopoietic defects at P5. It also shows that deletion of Scf from both stellate and endothelial cells does not lead to HSC depletion before E14.5.
We thank R. Schwabe at Columbia University Irving Medical Center, New York, NY, for providing the adult hepatic stellate cell gene expression profile. We thank S. Ho at the Columbia Center for Translational Immunology, A. Figueroa at the Department of Microbiology and Immunology, and M. Kissner at the Columbia Stem Cell Initiative for flow cytometry, and T. Swayne and L. Munteanu at the Columbia Confocal and Specialized Microscopy core for confocal microscopy.
This work was supported by the Rita Allen Foundation, the Schaefer Scholar program, and the National Heart, Lung, and Blood Institute (R01HL132074 and R01HL153487). Y. Lee was supported by the Korea Foundation for Advanced Studies and the NYSTEM Columbia training program in stem cell research. This research was funded in part through the National Institutes of Health/National Cancer Institute Cancer Center Support Grant P30CA013696.
Author contributions: Y. Lee, J. Leslie, and L. Ding performed all of the experiments. Y. Yang provided technical help and performed some HSC localization experiments. Y. Lee and L. Ding designed the experiments, interpreted the results, and wrote the manuscript.
Disclosures: The authors declare no competing interests exist.