|Home | About | Journals | Submit | Contact Us | Français|
To provide a lifelong supply of blood cells, hematopoietic stem cells (HSCs) need to carefully balance both self-renewing cell divisions and quiescence. Although several regulators that control this mechanism have been identified, we demonstrate that the transcription factor PU.1 acts upstream of these regulators. So far, attempts to uncover PU.1’s role in HSC biology have failed because of the technical limitations of complete loss-of-function models. With the use of hypomorphic mice with decreased PU.1 levels specifically in phenotypic HSCs, we found reduced HSC long-term repopulation potential that could be rescued completely by restoring PU.1 levels. PU.1 prevented excessive HSC division and exhaustion by controlling the transcription of multiple cell-cycle regulators. Levels of PU.1 were sustained through autoregulatory PU.1 binding to an upstream enhancer that formed an active looped chromosome architecture in HSCs. These results establish that PU.1 mediates chromosome looping and functions as a master regulator of HSC proliferation.
Hematopoietic stem cells (HSCs) guarantee the continuous supply of all mature blood lineages throughout adult life. In response to stress, HSCs are capable of extensive proliferative expansion, whereas in the steady state, HSCs largely remain in a quiescent state to prevent their exhaustion (Cheng et al., 2000; Hock et al., 2004; Matsumoto et al., 2011; Miyamoto et al., 2007; Zhang et al., 2006).
Transcription factor PU.1 is crucial for the development of almost all blood cells, and it is now recognized that PU.1 exerts its various functions in a dose-dependent manner (Carotta et al., 2010b). Recent examples of dose-dependent PU.1 functions are the differentiation choices of dendritic cells versus macrophages, neutrophils versus macrophages, and B2 versus B1 B cells (Bakri et al., 2005; Carotta et al., 2010a; Dahl et al., 2003; Rosenbauer et al., 2006; Ye et al., 2005). PU.1 gene expression is strictly regulated through the proximal promoter (PrPr) (Chen et al., 1995) and an upstream regulatory element (URE) located −14 kb or −17 kb upstream of the transcription start site in mice and humans, respectively (Li et al., 2001; Rosenbauer et al., 2004). Removal of this URE results in an 80% reduction of PU.1 expression in bone marrow in comparison to wild-type (WT) mice and leads to the development of leukemias or lymphomas (Rosenbauer et al., 2006; Rosenbauer et al., 2004). These results emphasize that tight regulation of PU.1 levels is critical for specifying cell fate and tumor suppression and establish that PU.1 mediates its functions via gradual expression level changes rather than via binary on/off states.
So far, the dose dependency of PU.1 functions has not been considered in any study of HSCs. Previous studies with fetal liver HSCs reported a lack of homing-related integrins in PU.1 complete knockout cells, which resulted in defects in colonizing bone marrow in transplantation assays, preventing further functional testing (Fisher et al., 1999; Iwasaki et al., 2005; Kim et al., 2004). Therefore, besides its importance for HSC homing after transplantation, no further functional role of PU.1 in HSCs could be retrieved from these models. Interestingly, when the homing defect was bypassed in adult mice (through PU.1 deletion after engraftment of transplanted HSCs had occurred), erythromyeloid repopulation capacity persisted, suggesting that PU.1 might not have a role in adult HSC maintenance (Dakic et al., 2005). However, we have now developed a mouse model with decreased PU.1 levels specifically in phenotypic HSCs, which preserves normal bone marrow homing capabilities. HSCs with decreased PU.1 levels are functionally compromised in competitive repopulation and serial transplantation assays and are insufficient for the regeneration of bone marrow after injuries. Mechanistically, we found that, in HSCs, PU.1 acts as a master regulator of multiple cell-cycle genes, restricting disproportionate HSC proliferation and sustaining HSC functional integrity. Moreover, we present direct evidence that positive autoregulation is necessary for the establishment and maintenance of normal PU.1 levels in the HSCs of adult mice. Furthermore, our study provides experimental proof to connect the binding of a single transcription factor, PU.1, to changes in chromosome structure and gene expression.
Previously, we identified a potential autoregulatory site within the −14 kb URE of murine PU.1, which we characterized in vitro (Okuno et al., 2005). To genetically dissect a functional role for the autoregulation of PU.1 in vivo, we generated knockin mice (PU.1ki/ki) with targeted disruption of this particular binding site by homologous recombination (Figure 1A, and Figure S1A available online). Chromatin immunoprecipitation (ChIP) analyses of total bone marrow cells confirmed the successful abolishment of PU.1 binding to the −14 kb URE in PU.1ki/ki mice, whereas URE binding of RUNX1 to binding sites in close proximity to the PU.1 site remained largely preserved (Figure S1B). PU.1 levels of PU.1ki/ki mice were not changed in unselected total bone marrow cells (data not shown). However, in phenotypic HSCs (defined in this study as Lin−Sca1+c-kit+CD150+CD48− [Kiel et al., 2005]), PU.1 messenger RNA (mRNA) levels of PU.1ki/ki mice were reduced by 66% in comparison to controls, similar to the levels of PU.1 heterozygous knockout (PU.1+/−) mice in which exon 4 and exon 5 were deleted (Iwasaki et al., 2005) (Figure 1B). Interestingly, both PU.1ki/ki and PU.1+/− mice displayed increased numbers of total bone marrow cells (Figure 1C) and phenotypic HSCs (Figure 1D) in comparison to control HSCs.
Previous reports, including one from our group, indicated that the complete loss of PU.1 resulted in decreased numbers of HSCs because of defective homing (Fisher et al., 1999; Iwasaki et al., 2005; Kim et al., 2004). In spite of this, PU.1ki/ki hypomorphs with decreased but not absent PU.1 levels demonstrated normal HSC homing in short-term bone marrow transplantation assays (Figure S2A). With the use of laser scanning cytometry on femur sections of WT and PU.1ki/ki mice, we further established that HSCs of PU.1ki/ki localized normally within the bone marrow niche (Figure 1E). In addition, data from microarray gene expression analysis of adult HSCs of PU.1ki/ki and WT mice were compared with data from analyses of fetal liver PU.1 knockout mice, which have been reported to exhibit severe homing defects (Fisher et al., 1999; Kim et al., 2004). In contrast to PU.1ki/ki, PU.1 knockout HSCs showed profound changes in the expression of genes involved in mediating interactions with the microenvironment (Figures S2B and S2C), thus indicating that the homing defects described for PU.1 fetal liver knockouts are not present in HSCs of adult PU.1ki/ki mice.
To test whether the observed increase of phenotypic HSC numbers in PU.1ki/ki mice reflected the quantitative differences of functional long-term HSCs (LT-HSCs), we performed competitive repopulation transplantations with limiting dilutions of purified phenotypic HSCs. After 6 months, long-term lymphomyeloid reconstitution in peripheral blood and HSCs in bone marrow of donor cells was assessed and plotted as the percentage of non-responding mice (Figures 2A and S3A). Intriguingly, phenotypic HSCs of PU.1ki/ki mice demonstrated a dramatic (13.6-fold) decrease of functional LT-HSCs.
To evaluate the potential of PU.1ki/ki HSCs to regenerate bone marrow after repetitive injuries, we analyzed mice after weekly administration of the antimetabolite 5′-fluorouracil (5-FU) (Figure 2B). We observed a decreased survival rate in PU.1ki/ki mice, indicating that HSCs of PU.1ki/ki exhaust prematurely, have increased cell-cycle activity, or both exhaust prematurely and have increased cell-cycle activity. To test for HSC exhaustion, we further analyzed long-term reconstitution capacity in serial transplantation assays. Interestingly, after three rounds of transplantation, HSCs of PU.1ki/ki mice failed to repopulate the bone marrow of lethally irradiated recipients, thus showing that HSCs of PU.1ki/ki hypomorphs did exhaust prematurely (Figure 2C). Moreover, competitive long-term reconstitution transplantation assays utilizing total bone marrow cells also revealed a reduced HSC function of PU.1+/− similar to what was observed with PU.1ki/ki. Restoration of PU.1 levels by crossing to a human PU.1 transgenic strain (Leddin et al., 2011) (Figure S3B) rescued HSC function of both PU.1ki/ki and PU.1+/− bone marrow cells (Figure 2D), demonstrating that HSC function was strictly related to PU.1 levels.
To directly assess the role of PU.1 in HSC proliferation, we measured 5-bromodeoxyuridine (BrdU) incorporation in vivo. Indeed, the proliferative fraction of PU.1ki/ki and PU.1+/− (Back et al., 2004) HSCs was doubled compared to WT HSCs (Figure 3A). In concordance with BrdU incorporation assays, cell-cycle analysis with PyroninY and Hoechst staining revealed that purified HSCs of PU.1+/− (Iwasaki et al., 2005) and PU.1ki/ki mice had a significant increase in dividing cells, as evidenced by an increased fraction of the S, G2, and M phases of the cell cycle. Restoration of PU.1 levels by crossing PU.1ki/ki mice to a human PU.1 transgenic (Leddin et al., 2011) (Figure S3B) reversed the increased S/G2/M fraction to normal levels (Figure 3B). These results demonstrate that PU.1 regulates proliferation in HSCs and that this effect was directly related to PU.1 levels.
To reveal potential mechanisms through which PU.1 levels might control HSC proliferation, we performed microarray gene expression analysis of purified HSCs of PU.1ki/ki and WT mice. Differentially expressed genes were mapped to known pathways with Gene Ontology (GO) pathway analysis. Strikingly, cell-cycle genes and genes of pathways directly affecting proliferation (such as canonical Wnt, MAPK, and p53 signaling) were significantly overrepresented (Figure 3C). Similarly, gene set enrichment analysis (Mootha et al., 2003) revealed significant enrichment of genes representing the activated G2 cell-cycle phase in PU.1ki/ki HSCs (Figure 3D). From the list of cell-cycle genes differentially expressed in microarray analysis (Table S1), we selected eight genes to validate differences in mRNA expression by quantitative real-time PCR (quantitative real-time PCR) (Figure 3E). Low PU.1 levels resulted in decreased expression of cell-cycle inhibitors, such as Gfi1, Cdkn1a (p21), and Cdkn1c (p57), and increased levels of cell-cycle activators such as Cdk1, Cyclin D1, E2f1, and Cdc25a.
To functionally study whether Cdk1, E2f1, and Cdc25a mediate the hyperproliferation of PU.1ki/ki HSCs, we designed five small hairpin RNA (shRNA) constructs for each gene and cloned them into the lentiviral vector pGhU6 (containing eGFP). We analyzed the knockdown efficiency of individual shRNA constructs at both protein and RNA levels and selected the most efficient ones for subsequent experiments (Figure S4A). HSC-enriched Lin−Scal+c-kit− (LSK) cells of individual WT and PU.1ki/ki mice (n = 4) were transduced with lentivirus expressing either a shRNA against specific cell-cycle activators (shCdk1, shE2f1, shCdc25a) or a nonsilencing control shRNA (NSC). Knockdown efficacy of all three genes was analyzed in GFP+ sorted LSKs, respectively (Figure S4B). We performed BrdU incorporation assays to assess the proliferation of transduced GFP+ LSKs 12 hr after BrdU application (Figure 3F). Similar to our previous results with SLAM+ LSKs, the proliferative (BrdU+) fraction of PU.1ki/ki LSKs was doubled in comparison to WT LSKs. Surprisingly, knockdown of either E2f1 or Cdc25a alone was sufficient to restore normal proliferation. This highlighted the critical role of either factor for the hyperproliferative phenotype induced by lower PU.1 levels. Interestingly, knockdown of either factor in WT cells did not perturb LSK proliferation, indicating that LSKs might not be essential for the maintenance of basic proliferation in healthy conditions and that their reduction could be compensated. Knockdown of Cdk1 alone in either WT or PU.1ki/ki LSKs was not sufficient to change proliferation, pointing to a compensatory mechanism by other factors in play. However, given that many cell-cycle regulators are changed in PU.1ki/ki mice, it is still possible that CDK1 might contribute to the hyperproliferative phenotype as part of a composite effect.
To test whether the binding of PU.1 to promoters of putative target genes might be correlated to “active” or “repressed” histone marks, we chose a whole-genome approach. By obtaining H3K4me3 and H3K27me3 ChIP sequencing (ChIP-seq) data for LSK cells and PU.1 ChIP-seq data for HPC-7 cells from Adli et al. (2010) and Wilson et al. (2010a), respectively, we mapped sequencing reads to the mouse reference genome. Figure 4A shows the patterns of PU.1 binding to ±5 kb either side of all annotated mouse promoters in relation to the histone marks H3K4me3 (active) and H3K27me3 (repressed) as stacked heatmaps. Three types of promoters, ones with the active histone marks, ones with the repressed histone marks, or ones with no histone marks at all, could be distinguished. Interestingly, PU.1 binding was largely restricted to promoters with the active H3K4me3 mark. As shown in the right panel, a substantial proportion of H3K4me3-bound promoters were also bound by PU.1 (Figure 4A).
Next, we combined PU.1 ChIP-seq data (Wilson et al., 2010a) with the differentially expressed genes in SLAM+ LSK cells of PU.1ki/ki and WT mice. Among all genes that were dysregulated in PU.1ki/ki mice, the binding of PU.1 to promoters of cell-cycle genes was significantly enriched, pointing to the predominant role for PU.1 in regulating the cell cycle in HSCs (Figure 4B).
In line with differential expression of cell-cycle genes and concurrence of PU.1 association with the active histone mark H3K4me3 at promoters genome-wide, we expected PU.1 to bind to promoters of genes that were downregulated in HSCs of PU.1ki/ki mice. Indeed, PU.1 bound to promoters and enhancers of cell-cycle inhibitors such as Gfi1 and Cdkn1a (Figure 4C). Surprisingly, we also observed distinct PU.1 binding to promoters and enhancers of cell-cycle activators such as Cdk1, E2f1, and Cdc25a (Figure 4C and Table S1). To evaluate the transcriptional regulation through PU.1 on a subset of promoters and enhancers of its putative target genes, we applied functional transcriptional reporter assays. On the basis of the ChIP-seq data, we selected promoter and enhancer regions of Gfi1, Cdkn1a, Cdk1, E2f1, and Cdc25a and cloned them into pXP2, pGL2, and pGL4 reporter vectors, respectively (see Table S2 for detailed sequence information). Each sequence contained between one and three putative PU.1 binding sites, according to calculated transcription factor binding performed with MatBase software (Genomatix) (indicated with red or green marks on the promoter- or enhancer-indicating black bar in Figure 4). Cotransfection of the indicated reporter constructs with increasing amounts of PU.1 expression plasmid resulted in a dose-dependent increase of reporter activity for the Gfi1 promoter and a Cdkn1a (p21) intron 2 enhancer (Figures 4D and 4E, red bars). In contrast, reporter activity of an E2f1 intron 1 enhancer (Figure 4D) and of the Cdk1 and Cdc25a promoters demonstrated a PU.1 dose-dependent decrease (Figure 4E, green bars). We went on to mutate the putative PU.1 binding sites of the Gfi1 promoter, the Cdk1 promoter, and the Cdc25a promoter. Mutation of PU.1 binding sites in any of the promoters significantly reduced the PU.1’s activating function on Gfi1 and its repressive function on Cdk1 and Cdc25a transcriptional activity (Figure 4E). In summary, these reporter assays demonstrated that PU.1 positively regulated the transcription of the cell-cycle inhibitors Gfi1 and Cdkn1a and negatively regulated the transcription of the cell-cycle activators Cdk1, E2f1, and Cdc25a through direct binding to their promoters and enhancers. These results demonstrate that PU.1 directly controls multiple regulators of cell division in HSCs.
To confirm the positive PU.1 autoregulation of HSCs in different in vivo models, we utilized conditional PU.1 knockout mice. In these mice, excision of PU.1 exon 4 and exon 5 could be induced by polyinosinic-polycytidylic acid administration. Resulting truncated transcripts (PU.1ex1–3) demonstrated the expected loss of the RNA corresponding to the DNA-binding ETS domain. Truncated transcripts were equally stable compared to WT (full-length) PU.1 mRNA (Figures 5A and 5B) (Iwasaki et al., 2005). We quantified murine truncated PU.1 transcripts with a murine-specific TaqMan quantitative real-time PCR (exon 1 and exon 2) (Figure S5A) and found an average reduction of 61% after excision in phenotypic HSCs (Figure 5C). Importantly, the introduction of a human PU.1 transgene into the PU.1ex1–3 background (Leddin et al., 2011) rescued the repression of PU.1ex1–3 transcripts, demonstrating that PU.1 is indeed autoregulated (Figure 5C). However, this autoregulatory loop may also involve other transcription factors, especially other Ets factors. TaqMan quantitative real-time PCR analysis of the Ets factors Fli1, Elf1, Erg, and Etv6 in HSCs and LSKs revealed that at least Erg and Etv6 were expressed at detectable levels in HSCs (Figure S5B).
To provide direct experimental proof that PU.1 binding to the −14 kb URE is essential for autoregulatory PU.1 transcription in HSCs and for ruling out secondary effects related to the consequences of decreased transcription-factor concentration, we designed an additional mouse model (Figure 5D). In this model PU.1 levels were maintained through the balanced expression of a human PU.1 bacterial artificial chromosome (BAC) (Figure S3B) (Leddin et al., 2011). Transcription of a murine allele with the mutated PU.1 binding site could be quantified with murine-specific PU.1 TaqMan quantitative real-time PCR (exon 3 and exon 4) that detected only full-length transcripts. The second murine allele was truncated (exon1–exon 3) and therefore undetectable with this assay. In macrophages, mutation of the PU.1 site at the −14 kb URE had no impact on PU.1 transcription, even though PU.1 was bound at that site (Figures 5D and S5C) (Heinz et al., 2010; Wilson et al., 2010a). However, in HSCs, we found that transcription levels of the mutated alleles were decreased by 60%, which was similar to those observed in HSCs of PU.1ki/ki mice. These results proved (1) that direct PU.1 binding to the −14 kb URE is essential for PU.1 transcription in HSCs; (2) that after excluding the secondary effects of reduced PU.1 levels, the decrease of PU.1 transcription in HSCs was comparable to that in PU.1ki/ki mice; (3) that secondary effects on PU.1 regulation in HSCs of the PU.1ki/ki model appeared to have no impact; and (4) that the PU.1ki/ki model is thus an unbiased model for the study of direct involvement of the −14 kb PU.1 binding site in gene regulation of HSCs.
In conclusion, positive autoregulation of PU.1 could be demonstrated in three independent mouse models and accounted for more than 60% of PU.1 levels in HSCs.
Recently, we reported that, in cells with a high expression of PU.1, the −14 kb URE physically interacts with PrPr, thereby forming a chromosomal conformation poised for active transcription (Ebralidze et al., 2008). These studies employed chromosome conformation capturing (3C) and were performed on cell lines so that sufficient material was attained (Dostie and Dekker, 2007; Ebralidze et al., 2008). We modified the 3C protocol to reduce the amount of cellular material needed and to quantify the degree of interaction between regulatory elements. We verified the linear (i.e., quantitative) range of this assay from 1 × 106 down to 5 × 104 cells, which allowed us to assay the LSK population, which includes HSCs (Figures S6A and S6B). With the use of 2 × 105 purified primary cells of pooled WT and PU.1ki/ki animals, we found that the mutation of the −14 kb PU.1 site led to a loss of the physical interaction between −14 kb URE and PrPr (Figure 6), specifically in HSC and progenitors but not in macrophages.
Previously, we have described an autoregulatory PU.1 site in a −12 kb cis element and reported its activity exclusively in mature myeloid cells as compared to other cell types, especially lymphoid cells (Leddin et al., 2011). To test whether the −12 kb element compensated for the −14 kb URE, we quantified the strength of interaction of the −12 kb element with the proximal promoter in macrophages of WT mice compared to PU.1ki/ki mice (Figure S6C). Interestingly, we found that the crosslinking frequency was significantly higher in macrophages of mice with a PU.1 binding site mutation in the −14 kb URE. This indicated that the −12 kb cis-regulatory element, which also harbors a PU.1 autoregulatory site, was more active in mature myeloid cells and might at least partially contribute to the observed normal PU.1 levels in macrophages.
In summary, we propose a model in which PU.1 binding is necessary in HSCs to establish an active chromosomal conformation for proper PU.1 transcription to balance cell-cycle activators and inhibitors (Figure 7).
Our results demonstrate that PU.1acts as a key regulator of HSC proliferation by restraining the cell cycle through multiple downstream targets. Limiting cell-cycle activity is critical in order to maintain life-long HSC function and to prevent HSC exhaustion (Cheng et al., 2000; Hock et al., 2004; Matsumoto et al., 2011; Miyamoto et al., 2007; Zhang et al., 2006). It has been shown that PU.1 induces the proliferation of erythroid progenitors (Back et al., 2004; Fisher et al., 2004). However, a previous study using inducible PU.1 overexpression suggested that PU.1 could reduce proliferation in HSCs (Fukuchi et al., 2008). As indicated by the gene expression and GO analysis presented in this study, PU.1 significantly regulates pathways that have been associated with HSC maintenance and self-renewal, such as canonical Wnt, MAPK, and p53 signaling (Kirstetter et al., 2006; Liu et al., 2009; Luis et al., 2009; Scheller et al., 2006; Wang et al., 2011). Surprisingly, in HSCs, PU.1 directly regulates various components of cell-cycle machinery by inhibiting cell-cycle-promoting factors such as Cdk1, E2f1, and Cdc25a and inducing the expression of inhibitors such as Gfi1, Cdkn1a (p21), and Cdkn1c (p57). Similar to PU.1ki/ki hypomorphs, young Gfi1−/− mice demonstrate increased phenotypic HSCs, increased cell-cycle activity, and decreased HSC maintenance and function (Hock et al., 2004; Zeng et al., 2004). PU.1 binds the promoter and a −35 kb upstream regulatory element of Gfi1, which induces its expression in HSC and progenitor cells (Wilson et al., 2010a; Wilson et al., 2010b). Also, in concordance with the PU.1ki/ki model, mice deficient for the G1 checkpoint regulator Cdkn1a (p21) have increased HSC proliferation and susceptibility to exhaustion in stress conditions such as serial bone marrow transplantation and repetitive 5-FU injections (Cheng et al., 2000). Interestingly, HSCs of the β-catenin gain-of-function mouse model also demonstrated decreased p21 levels and a similar increase in HSC proliferation (S/G2/M: 3.5-fold in β-catenin gain-of-function mice; 2.5-fold in PU.1ki/ki mice) and consecutive exhaustion after serial transplantation (see Figure S7A for a comparison in gene expression) (Scheller et al., 2006). Moreover, it has been reported recently that Mx1-Cre conditional deletion of Cdkn1c (p57) resulted in increased HSC-cycle activity and exhaustion in adult mice (Matsumoto et al., 2011). Altogether, our data demonstrate that PU.1 levels control multiple components of the complex regulatory network that balances HSC quiescence and proliferation. Several of these components, which are under the direct transcriptional control of PU.1, have been demonstrated to modify HSC proliferation and to impact HSC exhaustion.
Our experiments functionally connect the autoregulatory binding of PU.1 to changes in chromosome structure and gene activity. Among other reports using 3C assays to study enhancer-promoter interactions (for review, see Bulger and Groudine, 2011), enhancer occupancy by the transcription factors Oct4, Nanog, and Sox2 and the cofactors mediator and cohesin was associated with enhancer-promoter colocalization of actively transcribed genes in embryonic stem cells (Kagey et al., 2010). Although it can be suggested that the binding of these factors actively mediates enhancer-promoter interaction, it is just as likely that they are consequences of independent events, thus leaving the mechanisms of action uncertain (Bulger and Groudine, 2011). A recent study identified a novel Igh V(D)J recombination control region harboring CTCF binding sites and reported a loss of interaction with a long-distance element in knockout lines for that region (Guo et al., 2011). However, these data did not answer the question of whether CTCF binding or other mechanisms would mediate loop formation. Importantly, we recently described that targeted mutations of RUNX binding sites in a downstream regulatory element (DRE) of a human CD34 transgene caused the perturbation of the DRE-promoter interaction in transgenic mice (Levantini et al., 2011). Along with the functional models presented here, specific disruption of PU.1 binding in the URE of the endogenous PU.1 locus can be used to distinguish between correlation and causation of the transcription factor binding and chromosome looping necessary for gene activation.
Previously, we reported the generation of a mouse model in which the entire −14 kb URE was deleted (and in which the −12 kb regulator element was intact). PU.1 levels in whole bone marrow were reduced to 20% of the normal level. In contrast, PU.1ki/ki mice had unchanged PU.1 levels in the whole bone marrow compartment. In HSCs, PU.1ki/ki mice demonstrated a reduction to 40%, which led to HSC exhaustion. PU.1 levels of URE knockout mice in HSCs were even further reduced (4%); however, the major phenotypic difference between the strains appeared after the HSC stage in myeloid progenitors (Figure S7B). HSCs of URE knockout mice progress to leukemia, and it is likely that dysregulated cell-cycle regulators might be involved. Given that these cells proliferate but do not exhaust, additional changes related to the difference in PU.1 levels might be responsible.
Furthermore, our data revealed that another cis-regulatory element, the −12 kb element (Leddin et al., 2011), which was demonstrated to be active in macrophages and which contains an active PU.1 site, at least partially compensated for the lack of PU.1 binding at the −14 kb URE. A more than 50% increase in crosslinking efficiency of the −12 kb region with the proximal promoter as observed in macrophages of PU.1ki/ki mice might be sufficient to compensate for the lack of autoregulatory binding via the −14 kb URE.
Our data demonstrate that positive autoregulation of a transcription factor occurs in vivo and that it has an essential function in an adult mammalian organism. Distinct cellular states are established and maintained by transcription factor networks, which can create cellular memory and stability (Acar et al., 2005; Shykind et al., 2004; Xiong and Ferrell, 2003). Autoregulation is suggested to play a critical role in fine-tuning the steady state of transcription factor concentrations in these networks. In particular, positive transcriptional autoregulation is implicated in preserving stability and enhancing cellular memory by increasing factor concentration and response time to fluctuations of stimuli (Acar et al., 2005; Murugan and Kreiman, 2011). Synthetic circuits and in vitro culture-based studies have been utilized to identify mechanisms of autoregulatory gene networks. Recently, genome-wide transcription factor occupancy studies have also emphasized a potentially fundamental role of autoregulation in maintaining distinct cellular phenotypes such as embryonic stem cell state (Young, 2011). However, these studies are inherently descriptive and are also limited in that they have not accounted for changes secondary to decreased transcription factor expression. Therefore, the natural occurrence and in vivo functional relevance of autoregulation in complex organisms such as mammals have remained unclear. With the use of a series of in vivo genetic models, including knockouts, transgenics, targeted deletions, and, in particular, models which maintain normal PU.1 expression, we can now demonstrate conclusively that PU.1 autoregulates its expression through a binding site in an upstream regulatory region. Surprisingly, these in vivo data revealed that, in mature cells such as macrophages with high PU.1expression, PU.1 binding to the −14 kb URE is functionally irrelevant, whereas it is important for the establishment and maintenance of critical levels exclusively at the earliest stage of the hematopoietic hierarchy in HSCs.
All mice were kept in a sterile barrier facility approved by the Beth Israel Deaconess Medical Center Institutional Animal Care and Use Committee. To generate mutant PU.1ki/ki mice, we used the pPNT (−14 kb URE)-targeting vector (Rosenbauer et al., 2004) and introduced a GGAA-to-TCGC mutation by site-directed mutagenesis (Okuno et al., 2005). Figure S1A indicates the exact -position of the binding-site mutation. For the transfection of R1 embryonic stem cells, 20 μg of the linearized (Not1) construct was used. The generation, identification, and conformation of positive embryonic stem cell clones, chimera mice, and heterozygous mice were performed as previously reported (Rosenbauer et al., 2004). Nonconditional and Mx1Cre conditional PU.1 knockout mice, as well as transgenic mice with human PU.1 BAC, used in this study have been described previously (Back et al., 2004; Iwasaki et al., 2005; Leddin et al., 2011). All mouse strains were crossed into C57B6 mice for at least six generations. Primer sequences for genotyping PCRs are listed in Table S3. All mice used in this study were 3–4 months of age.
For limiting dilution long-term competitive reconstitution assays, HSCs were collected from CD45.2+ mice (WT and PU.1ki/ki). The indicated number of cells were resorted into individual wells of a 96-well plate containing 2 × 105 CD45.1+ whole bone marrow cells in PBS and transferred intravenously into lethally irradiated CD45.1+ recipients (650 rads twice with a 4 hr interval). Peripheral blood and bone marrow were obtained from each mouse after 6 months and analyzed by fluorescence-activated cell sorting. A recipient mouse was considered positive if CD45.2+CD45.1− cells were present in myeloid and B and T cells, if CD45.2+CD45.1− cells comprised more than 0.3% of cells in peripheral blood, and if CD45.2+CD45.1− HSCs were detectable in the bone marrow. Serial and competitive long-term total bone marrow transplantations were performed as described in the figure legends.
BrdU incorporation was measured by flow cytometry with an APC BrdU Flow Kit (BD Pharmingen) according to the manufacturer’s protocol on gated Lin−, Sca1+, c-kit+, CD150+, or CD48−cells or cultured LSKs (shRNA knockdown studies). For Hoechst and Pyronin Y staining, sorted HSCs (Lin−Sca1+c-kit+ CD150+CD48−) were suspended in a phosphate-citrate buffer solution with 0.02% saponin for permeabilization and then incubated with Hoechst 33342 (Invitrogen) and Pyronin Y (Sigma-Aldrich).
Bone marrow of four individual animals (either PU.1ki/ki or WT) was pooled and 2 × 105 sorted cells were used (Macrophages [Mac], Lin−, Sca1+, and c-kit+ [HSCs]). We used a modified protocol from Dostie and Dekker (2007) to reduce the amount of cellular material needed and to quantify interactions of the −14 kb URE (H2 region) or PrPr with other genetic elements of the PU.1 locus (detailed protocol is shown in Table S4). Figure 6A demonstrates the localization of Bgl2 restriction sites. A TaqMan probe was designed for the H2 fragment of the −14 kb URE containing the PU.1 autoregulatory site. We used an amplicon within two Bgl2 restriction sites in intron 3 of the PU.1 gene as a “housekeeping gene.” exact sequence information of all primers and probes used are shown in Table S3. All 3C experiments were performed as two independent biological and two technical replicates.
CRU frequencies were calculated with the L-Cal software (STEMCELL Technologies). The statistical differences in frequencies between sets of limiting dilution analyses were assessed on the basis of the asymptotic normality of the maximum likelihood estimates and calculated with a chi-square test. A log-rank nonparametric test (Mantel-Cox) was used for survival studies upon 5-FU treatment. For laser scanning cytometry analysis, or data, which did not meet the criteria of a normal distribution, the nonparametric Mann Whitney U test was used (real-time PCR for Ccna2). In all other experiments, statistical significance was assessed by Student’s unpaired t test. Statistical significance was indicated as *, p < 0.05, and **, p < 0.01. For normalized data, Gaussian error propagation was applied.
We thank Christopher Hetherington and Alexander Ebralidze for expert assistance with quantitative 3C analysis, the Dartmouth Transgenic Facility directed by Steven Fiering, and the BIDMC Flow Cytometry Facility. We thank Dr. Wenyi Wei and Dr. Peter Sicinski for providing reagents (the anti-Cdc25a antibody and the Cdk1 expression construct, respectively). We are grateful to the members of the Tenen laboratory for helpful discussions, especially Annalisa Di Ruscio, Elena Levantini, and Deepak Bararia. This work was supported by NIH grant R01HL112719 to D.G.T. and by fellowships from the Austrian Research Foundation (Erwin Schrödinger Stipendium: J2876-B12) and the Austrian Academy of Science (APART Stipendium: 11379) to P.B.S. Starting in January, 2012, P.B.S. has received a Marie Curie International Outgoing Fellowship from the European Union (PIOF-254486). C.B. was supported by the German Research Foundation (DFG fellowship BA 4186/1-1). F.N. is the recipient of a Yousef Jameel scholarship administered by the Cambridge Commonwealth and Overseas Trust.
Original raw data (CEL files) of Mouse430_2 arrays (Affymetrix) were stored at the gene expression repository at the Harvard Stem Cell Institutes (http://bloodprogram.hsci.harvard.edu) and in the NIH GEO database at accession code GSE33031.