Architecture of a lymphomyeloid developmental switch controlled by PU.1, Notch and Gata3

T-cell development depends on the correct expression of an intricate transcription factor network and on signaling from the environment. T cells develop from multipotent progenitors that migrate from the bone marrow to the thymus, where they become dependent upon Notch signaling for their development and survival (Yang et al., 2010). At the early CD4^- CD8^- double negative (DN) stages, pro-T cells retain lineage plasticity until the DN2b stage where they become committed pre-T cells. The ETS family transcription factor PU.1 (Sfpi1 - Mouse Genome Informatics) is important during early T-cell development (Back et al., 2005; Nutt et al., 2005), and is highly expressed initially but repressed during commitment (Fig. 1A). This pattern must be maintained for development to succeed. In early T-cell stages, PU.1 drives expression of cytokine receptors such as Il7r and Flt3, and of genes that are important for cell communication (Turkistany and DeKoter, 2011). However, it is also required for the development and function of other cell types, including hematopoietic stem cells (Iwasaki et al., 2005), multipotent progenitors (Wontakal et al., 2011), myeloid cells (Ghani et al., 2011) and B cells (Houston et al., 2007). Forced overexpression of PU.1 can divert early T cells to a myeloid lineage (Anderson et al., 2002; Dionne et al., 2005; Lefebvre et al., 2005; Laiosa et al., 2006b). However, in the context of T-cell development the progenitors are normally protected from diversion, even while expressing high levels of PU.1, by their exposure to Notch signaling from the environment (Franco et al., 2006; Laiosa et al., 2006b) (Fig. 1A).

Tracking the effects on several dozen genes has shown that the interaction between PU.1 and Notch can have dichotomous effects on early T-cell progenitors, with cells partitioning between those that maintain a T-cell gene expression pattern and those that shift towards a myeloid pattern (Dionne et al., 2005; Franco et al., 2006). This suggests competition between two self-reinforcing network states. However, the actual gene network underlying this choice has been obscure.

Here, we explore the mechanisms that mediate the regulatory competition between PU.1 and Notch signals, using primary mouse fetal thymocytes and a clonal pro-T-cell line system to dissect the regulatory impacts of PU.1 and Notch signaling. We show that Notch signaling does not inactivate PU.1 protein but re-channels its transcriptional effects. However, PU.1 and Notch signaling are involved in a mutually inhibitory network, as PU.1 can repress Notch targets. Our results further reveal two branches of the T-cell gene network that collaborate against the PU.1-mediated diversion: one involving basic helix-loop-helix E proteins in a tight positive-feedback loop with Notch; and a separate branch for Gata3 and the Gata3-activating factor Myb. We show that PU.1 undermines Gata3 expression, foreshadowing diversion in individual cells. The two T-cell lineage protective pathways converge as Myb and Notch signaling each enable Gata3 expression to be maintained in the face of high levels of PU.1.

C57BL/6-Bcl2tg mice [B6.Cg-Tg(BCL2)25Wehi/J] were housed under specific pathogen-free conditions, bred and cared for by Caltech Animal Facility staff. Embryonic day (E) 14.5 or 15.5 fetal thymocytes were used. All animal work followed protocols approved by the Institutional Animal Care and Use Committee.

Scid.adh.2C2 cells were cultured in RPMI1640 with 10% fetal bovine serum (Sigma-Aldrich), sodium pyruvate, non-essential amino acids, penicillin/streptomycin/glutamine (Gibco/Life Technologies/Invitrogen) and 50 μM β-mercaptoethanol. Cells were incubated at 5% CO₂ and 37°C.

For Notch signaling inhibition, InSolution γ-Secretase Inhibitor X (EMD Millipore) were added at 0.5 μM.

Fetal thymocytes were cultured on OP9-Delta-like1 (OP9-DL1) or OP9-control stroma in α-MEM with 20% fetal bovine serum, penicillin/streptomycine/glutamine, 50 μM β-mercaptoethanol, 5 ng/ml Il7, 5 ng/ml Flt3 ligand (cytokines from Peprotech).

FITC, PE, APC, APCe780, Pacific Blue and PerCPCy5.5-conjugated antibodies from eBioscience or Cell Signaling (against CD25, CD44, CD45, Mac1/CD11b, CD11c, Thy1, NGFR and human CD8) were used for cell surface staining. Fc receptors were first blocked with 2.4G2. PU.1 intracellular staining using the BD cytofix/cytoperm kit (Becton Dickinson Immunocytometry Systems) was carried out using PU.1 (9G7) rabbit mAb-AlexaFluor 647 and rabbit mAb IgG (#2985, Cell Signaling) as an isotype control. Gata3 intracellular staining using Foxp3 Staining Buffer Kit (eBioscience) was carried out with mouse anti-Gata3-AlexaFluor 647 and mouse IgG1 κ as an isotype control (BD Biosciences Pharmingen). Data presented are representative of multiple independent experiments with n=2 (Fig. 1C; Fig. 7C,D) and n=3-5 (Fig. 1B; Fig. 3A,B; Fig. 4A; Fig. 5A; Fig. 6B,C; Fig. 7B).

Cells were sorted using BDIS FACS Aria IIu or iCyt Mission Technology Reflections cell sorters and analyzed using FACSCalibur (BDIS) or MACSQuant (Miltenyi Biotec) analyzers and FlowJo software (Tree Star).

cDNA was prepared from total RNA using RNeasy extraction kits (Qiagen) and reverse transcribed using random primers and SuperscriptIII (Invitrogen).

Specific gene expression in cDNA samples was measured by qRT-PCR (ABI Prism 7900HT) using SyberGreenER mix (Invitrogen). Results were calculated (ΔCt method) and normalized to Actinb levels. For actual values see supplementary material Table S1A,B (Fig. 2) and Table S2A-E (Figs 3, 4, 5, 6, 7). Primers used for qRT-PCR were described previously (David-Fung et al., 2009; Li et al., 2010; Yui et al., 2010), or are listed in supplementary material Table S3.

Heatmaps were generated using a Matlab (MathWorks) script written by Dr Hao Yuan Kueh (California Institute of Technology, Pasadena, CA, USA). Briefly, values are log₁₀-transformed averages of expression levels determined by qRT-PCR from 2-4 independent experiments: n=2 or 3 (Fig. 2C,D), n=4 (Fig. 3C; Fig. 4B), n=2 (Fig. 5B; Fig. 6D; Fig. 7E). Levels for each gene in different samples are presented relative to the level in the control sample (empty vector transduced=1.0). The color scale ranges from ∼10^-2 to 10² times this reference value, as indicated. Ordering of genes was by hierarchical clustering (median method, Matlab).

Gfi1-, Id2-, Bambi- and Cebpa-coding sequences were purchased from Genscript and subcloned into retroviral vectors: LZRS or MIGR1 with a GFP marker, or derivatives with an NGFR marker (New England Biolabs reagents). For retroviral packaging, Phoenix-Eco cells were transfected with long-term puromycin selection for LZRS-based vectors, whereas 293T cells were transiently co-transfected with pCL-Eco plasmid for MIGR1-based vectors. Tcf7 in a retroviral vector with a Vex reporter and ICN1 and dnMAML in MIGR1 were kind gifts from Avinash Bhandoola and Warren Pear, respectively (University of Pennsylvania, Philadelphia, USA). Gata3shRNA in the Banshee retroviral vector was made by Gabriela Hernandez-Hoyos (California Institute of Technology, Pasadena, CA, USA).

Non-tissue culture treated plates (Corning) were incubated with Retronectin (Takara) at 40-50 ng/ml overnight at 4°C. Retronectin was removed and viral supernatant added and spun at 2000 g for 2 hours at 32°C. Unbound virus was removed and cells added in their preferred medium at 1×10⁶ cells/ml, then incubated for 4 hours or overnight.

Cell extracts in Laemmli sample buffer were boiled for SDS-PAGE. Proteins were transferred to PVDF Immobilin (Millipore) and blots were blocked with 5% milk in TBS-T (Tris-buffered saline, 0.5% Tween-20), incubated with SP1 (sc-59) or PU.1 (sc-352) antibody (Santa Cruz Biotechnology, 1:1000 dilution) and then with secondary antibody (1:2000). Samples were then incubated with substrate (SuperSignal, Pierce) for film detection.

In the early T-cell stages when PU.1 is active, it provides cells with access to developmental alternatives and is therefore a risk to T-lineage fidelity. We have shown previously that thymocytes can be protected from PU.1-mediated lineage diversion if they receive Notch signals (Franco et al., 2006), as they would in the normal thymus in vivo. However, the mechanism through which Notch signaling counteracts the activity of PU.1 has been obscure.

To investigate the critical time interval in which Notch signaling affects thymocyte responses to PU.1, we forced fetal thymocytes to express PU.1 by retroviral transduction and exposed them to differing Notch signaling conditions for 3 days, using switch cultures based on co-culture with OP9-DL1 or OP9-control stromal cells. BCL2-transgenic (Bcl2tg) thymocytes were used to enhance recovery of cells after regulatory perturbation (supplementary material Fig. S1) (Franco et al., 2006; Taghon et al., 2007). OP9-control stroma supports B cell, natural killer cell and myeloid development, but when transfected to express the Notch ligand Delta-like1 (DL1), OP9-DL1 cells support T-cell development (Schmitt and Zúñiga-Pflücker, 2002). Thus, thymocytes were infected with empty vector or with PU.1-expressing retrovirus during a 4-hour incubation, cultured with OP9-DL1 or OP9-control stroma for 1 day, and then either returned to the same Notch signaling condition or switched to the opposite condition for 2 more days (Fig. 1B). Thy1, which is itself Notch insensitive, was used to identify cells that had entered the T-cell pathway (Taghon et al., 2007), as Mac1 (CD11b; encoded by Itgam) marked entrance to the myeloid pathway (Dionne et al., 2005). As these markers are normally mutually exclusive, activation of Mac1 on Thy1⁺ cells identifies T-lineage cells beginning myeloid diversion. Later, these become Mac1⁺ Thy1^-.

On OP9-control stroma in the absence of Notch signals, the empty vector-transduced thymocyte population was revealed to include some cells with natural myeloid potential, but few were Thy1⁺, i.e. derived from T-lineage precursors (Fig. 1Bb: 13% total Mac1⁺ versus 2% Thy1⁺ Mac1⁺). Fetal thymocytes transduced with PU.1 generated far more Mac1⁺ cells than thymocytes transduced with empty vector under all conditions (Fig. 1B): if cultured in the absence of Notch signals, most became Mac1⁺, over 50% derived from Thy1⁺ cells. As expected, the samples cultured in the presence of Notch signaling on OP9-DLl cells throughout the 3-day culture contained a far smaller percentage of Thy1⁺Mac1⁺ diverted cells as well as fewer Mac1⁺ cells overall than those cultured on OP9-control. Notch signals restored for the last 2 days after an initial day of deprivation also reduced diversion, as these samples mimicked the conditions we had used previously (Franco et al., 2006). Notably, however, samples that were initially cultured on OP9-DL1 for only 1 day and then shifted to OP9-control were also protected, almost as strongly as in continuous presence of DL1 (Fig. 1Bc). Thus, Notch signaling through the onset of PU.1 overexpression could establish a regulatory state making Thy1⁺ fetal thymocytes relatively resistant to diversion.

A possible mechanism for protection of pro-T cells from PU.1-mediated diversion could be to inactivate PU.1 protein. PU.1 phosphorylation can affect its DNA binding (Seshire et al., 2011) and transactivation domain engagement (Hamdorf et al., 2011). Notch signaling can regulate protein phosphorylation (Vo et al., 2011) and trigger protein degradation by promoting ubiquitylation (Lim et al., 2011). To test directly whether Notch-Delta signaling resulted in changes in PU.1 protein levels, fetal thymocytes were infected with PU.1 or empty vector and cultured on either OP9-DL1 or OP9-control for 2 days, and then stained for both intracellular PU.1 and cell-surface Mac1. The intermediate and high levels of PU.1 protein in transduced cells matched the levels of endogenous PU.1 in those control thymocytes that revealed natural myeloid potential when Notch signals were removed (Fig. 1Ca, empty vector, OP9-control).

The distributions of PU.1 protein in PU.1-transduced cells were not globally altered by the presence or absence of Notch signals. However, the response to a given level of intracellular PU.1 depended strongly on Notch signaling, as cells made all-or-none choices between remaining Mac1^- and diverting to high Mac1⁺ states (Fig. 1Ca,c). High, intermediate and low levels of intracellular PU.1 protein all drove over 90% of cells to become Mac1+ in the absence of Notch ligand (Fig. 1Cb). In the presence of Notch ligand, Mac1 could still be induced at the highest levels of PU.1 protein, and Notch signaling did not affect the levels of Mac1 expressed (Fig. 1Cc). However, the high, intermediate and low PU.1 level cells each generated substantially lower Mac1+ percentages in the presence of Notch signaling. Importantly, cells that now resisted Mac1 upregulation (Fig. 1Cb,c) expressed the same levels of PU.1 protein that promoted Mac1 expression when Notch signals were absent. Thus, Notch signaling can sharply raise the dose-dependent threshold for PU.1 to induce expression of Mac1, without affecting accumulation of PU.1 protein itself.

Similarly, when PU.1 was introduced into a pro-T-cell-like cell line, Scid.adh.2C2, western blotting measurements showed that Notch signaling affected the PU.1 dose threshold for the cells to divert to a Mac1⁺ state (supplementary material Fig. S2). However, qualitative PU.1 electrophoretic mobility patterns were the same in diverted and diversion-resistant cells, whether Notch signaling was active or inhibited, offering no evidence for differential phosphorylation or ubiquitylation. This suggests that PU.1 itself remains biochemically competent in the presence of Notch signaling.

Because even transient exposure to Notch signaling could protect PU.1-overexpressing thymocytes from diversion, Notch signaling might alter the earliest responses to PU.1. Previous studies had shown that Notch signals protect important T-cell genes from repression 40-48 hours after PU.1 transduction (Franco et al., 2006). Those analyses were potentially skewed toward diversion, however, because the cells were initially deprived of Notch signals during the >16-hour transduction. In addition, survival effects could obscure gene-specific regulation, for thymocytes naturally increase Notch-dependence as they progress from DN2 to DN3, when many T-cell genes are induced (Yui et al., 2010). Therefore, we infected fetal thymocytes with PU.1 or empty vector for only 4 hours, then cultured them with or without Notch signaling for 16 hours before sorting for RNA analysis, separately isolating transduced DN2 and DN3 cells (Fig. 2; supplementary material Table S1A,B).

The genes analyzed showed four different patterns of response to PU.1 and Notch-DL1 interaction, as illustrated by representative bar graphs of expression for individual genes (Fig. 2A,B) and summary heatmaps of PU.1 effects on DN2 and DN3 cell gene expression (Fig. 2C,D). The normal developmental expression patterns of key genes are also shown (Fig. 2E). One group of genes was upregulated efficiently by PU.1 overexpression, whether Notch signaling was present or absent (e.g. Fig. 2A,C,D). This group included the stem and progenitor cell-associated genes Lyl1, Bcl11a and Hhex, and the myeloid gene Fes. The effectiveness of PU.1 was partly constrained by the natural limits of the expression of these genes from DN2 to DN3 (Zhang et al., 2012) (Fig. 2E): e.g. effects on Hhex and Lmo2 were seen in DN2 cells but not significantly in DN3 cells (supplementary material Table S1C). Only select genes, e.g. Lmo2 (Fig. 2A) and Mac1 (Itgam), were inhibited from responding to PU.1 by Notch-DL1 interaction. Thus, PU.1 can indeed act positively on many target genes, even in the presence of Notch signaling.

Genes specifically expressed in the T lineage showed three patterns of response (Fig. 2B-D). Some were downregulated by PU.1 whether or not Notch signaling was present. These included Ets1 and the crucial T-cell regulatory gene Tcf7, a gene that is initially induced by Notch (Germar et al., 2011; Weber et al., 2011) but is not acutely dependent on Notch signaling for its maintenance. Another pattern was defined by Notch target genes [e.g. Deltex1, Hes1, HEBalt (Tcf12) and Nrarp], which depended on Notch signals even in control cells: e.g. Deltex1. A third group consisted of genes that were downregulated by PU.1, but much more severely if Notch signaling was absent. These included genes important for T-cell development, such as Myb, Fog1 (Zfpm1) and Gfi1. However, in general, the Notch target genes were also PU.1 inhibited, and additively affected by Notch deprivation and PU.1 (Fig. 2B-D: Nrarp, HEBalt and Hes1).

Though we anticipated Notch to influence PU.1 effects, these results suggest that PU.1 in early T cells also antagonizes responses to Notch. Thus, PU.1+ cells may demand higher-intensity Notch signaling to maintain expression of directly and indirectly Notch-regulated genes.

The lasting protective effects of Notch signaling in early pro-T cells and its impact on early responses to PU.1 overexpression imply that these early affected genes may be involved in deciding between protection and diversion in cells with high PU.1 expression. Testing these genes for epistatic or synergistic effects by co-transfection would be difficult in fetal thymocytes. Therefore, we used a previously described early T-cell line (Dionne et al., 2005) devoid of intrinsic myeloid potential, which is much more permissive for co-transduction experiments.

Scid.adh.2C2 cells, DN3-like cells that do not express endogenous PU.1, have previously been used to demonstrate the all-or-none diversion response of early T-cells after PU.1 overexpression (Dionne et al., 2005). Scid.adh.2C2 cells were cloned from a cell line, Scid.adh, which is derived from a spontaneous pro-T cell tumor (Carleton et al., 1999), and show spontaneous, ligand-independent Notch pathway activation. We tested whether the Scid.adh.2C2 response to PU.1 was also subject to Notch-dependent protection. Notch signaling in these cells was inhibited by γ-secretase inhibitor (GSI), as shown by the downregulation of the Notch-dependent marker, CD25 (Fig. 3A, ‘empty vector’, 0.5 μM GSI). Cells survived well with or without Notch signaling. Scid.adh.2C2 cells transduced with PU.1 upregulated Mac1 in a fraction of the population, and the percentage of cells becoming Mac1⁺ increased with the addition of GSI (Fig. 3A). Interestingly, another Scid.adh subclone that was unable to divert in response to PU.1 alone (6D4) (Dionne et al., 2005) also showed strong diversion when Notch signaling was inhibited (supplementary material Fig. S3). Thus, Notch signaling limits the response to PU.1 in these Scid.adh-derived clonal cell lines as in primary thymocytes.

Although many PU.1-overexpressing Scid.adh.2C2 cells upregulated Mac1, a population of Mac1-CD25+ cells still remained. CD25 is encoded by a Notch target gene, Il2ra (Maillard et al., 2006), and expression levels of other Notch target genes correlate with CD25 levels (M.M.D.R., unpublished). Individual Scid.adh.2C2 cells that remain Mac1 negative might simply express insufficient PU.1 to divert, or they might resist because of higher Notch signaling, suggested by their high CD25 expression. To distinguish these possibilities, we transduced Scid.adh.2C2 cells with PU.1 for 2 days, sorted the apparently diversion-resistant PU.1⁺Mac1^-CD25⁺ cells, then cultured them for 2 more days with or without GSI and assessed whether they remained Mac1 negative (Fig. 3B). Some cells in the vehicle control samples did upregulate Mac1 after 2 days, but the cells cultured in GSI generated a much higher percentage of Mac1⁺ cells (Fig. 3B). Thus, Scid.adh.2C2 cells expressing levels of PU.1 that are barely adequate for diversion can be efficiently diverted when endogenous Notch signaling is blocked.

Although inhibition of Notch signaling facilitated diversion, the final molecular phenotype of the diverted cells was the same with or without Notch inhibition, and the features of this response largely matched those of fetal thymocytes. Fig. 3C and Table 1 (values in supplementary material Table S2A; Fig. S4A) summarizes gene expression patterns in cells that were transduced with PU.1 or empty vector and cultured for 2 days with GSI or control vehicle, then sorted to separate Mac1+ diverted cells from cells remaining Mac1^-. A set of Notch-dependent target genes was detectably inhibited by GSI, both in the absence of PU.1 and in PU.1-transduced cells (Table 1). In addition, PU.1 turned on one set of genes that was neither dependent on Notch signaling nor on Notch inhibition (Table 1). These were activated in Mac1⁺ and Mac1^- PU.1-expressing cells alike, showing that PU.1 is active in all these cellular contexts. However, the induction of Mac1 by PU.1 heralded a global gene expression shift. Macrophage-associated genes such as Csf1r and Mac1 (Itgam) were upregulated by PU.1 selectively in the cells becoming Mac1⁺ (Table 1). As in fetal thymocytes, PU.1 also inhibited T-cell genes (Table 1). Unlike activation, repression primarily occurred in Mac1⁺ cells, not in cells remaining Mac1^- (Table 1), implying that these genes are repressed only when the regulatory threshold for diversion has been crossed. Notably, cells becoming Mac1⁺ in response to PU.1 alone maximally downregulated the Notch target genes, with or without GSI (Table 1). Thus, forced PU.1 expression can initiate a mechanism that leads to severe Notch pathway inhibition in Scid.adh.2C2 cells, and this event is tightly correlated with diversion.

To dissect the mechanism of Notch pathway interaction with PU.1, we used Scid.adh.2C2 cells for co-transduction experiments to combine PU.1 with constitutively active Notch1 (ICN1) or the dominant-negative inhibitor of Notch-dependent transcription, dnMAML (Maillard et al., 2004). Doubly transduced cells were sorted based on their co-expression of both viral vectors after 2 days. When ICN1 was co-expressed with PU.1, most of the cells remained CD25⁺ and did not upregulate Mac1. This protection depended on Notch-dependent transcription, as the addition of dnMAML with PU.1 not only extinguished CD25 expression but also caused most of the cells to upregulate Mac1 (Fig. 4A). However, PU.1 could still induce gene expression changes in Scid.adh.2C2 cells, including expression of the dendritic-cell marker CD11c, even in the presence of ICN1 (Fig. 4A).

The ability to manipulate Notch signaling independently of PU.1, while maintaining viability, enabled us to ask how much of the ‘PU.1’ effect on T-cell gene expression depended on its Notch inhibition effects (Fig. 4B). Doubly transduced cells (Fig. 4A) were sorted for RNA analysis (fewer than 12% of PU.1⁺ ICN1⁺ cells were Mac1⁺; over 70% of PU.1+ dnMAML+ cells were Mac1⁺). As expected, PU.1 with dnMAML mimicked the full range of the diverted phenotype. However, separate regulatory components were distinguished with dnMAML alone, and when PU.1 expression was combined with ICN1 (Fig. 4B; supplementary material Table S2B; Fig. S4B). Forced expression of ICN1 could protect classic Notch target genes, even in the presence of PU.1 (supplementary material Table S2B, ‘response group’ 5), and these genes could be upregulated by ICN1 alone (supplementary material Table S2B, group 2), implying an additive effect. However, three additional relationships emerged.

First, ICN1 could not protect all T-cell genes from PU.1 (supplementary material Table S2B, group 6). Thus, PU.1 represses these genes through a mechanism that depends on something besides Notch inhibition. Second, expression of some PU.1-dependent genes were actually enhanced by ICN1, implying distinct gene-specific rules for interaction (supplementary material Table S2B, group 4). Third, importantly, T-cell regulatory genes, including Myb, Tcf7 and, to a lesser extent, Gata3 were only downregulated by PU.1 when combined with loss of Notch signaling (supplementary material Table S2B, group 7): they were minimally affected by ICN1, dnMAML or PU.1 alone. dnMAML alone was highly effective at blocking Notch target gene expression (supplementary material Table S2B, group 1), and yet had absolutely no effect on Myb, Tcf7, Gata3 or Fog1. However, in a CD11c⁺ Mac1^- intermediate stage leading to diversion, Myb, Tcf7 and Gata3 also remained less affected (supplementary material Fig. S5). Thus, to complete diversion (Fig. 3C), PU.1 must shut off these genes by another mechanism, beyond antagonism of Notch, even though Notch signaling maintains the inputs that protect their expression.

The data thus far indicate that diversion to a Mac1⁺ state is linked with PU.1-dependent repression of at least two distinct groups of T-cell genes. Of these, Notch-dependent target genes such as those inhibited by dnMAML (supplementary material Table S2B, group 1) represent one component, but others such as Myb, Gata3, Gfi1 and Tcf7 represent a separate, possibly rate-limiting, component. These genes encode among the most important transcription factors known for T-cell development (Rothenberg et al., 2008) and may themselves play a role in maintaining T-cell identity.

We reasoned that extinction of the T-cell program must occur only when the T-cell gene(s) that resist(s) diversion was finally turned off or neutralized. This resistance factor might be TCF1 (encoded by Tcf7), Myb, Gfi1 or Gata3, but it might also be basic helix-loop-helix E protein (E2A, HEB, TCF12) activity, which reportedly controls both T-cell differentiation genes such as Rag1 and other T-cell regulatory genes (Ikawa et al., 2006; Schwartz et al., 2006). Indeed, PU.1 could neutralize E proteins: in Mac1⁺ diverted cells, the E protein antagonist Id2 is upregulated, and this upregulation is blocked by Notch signaling. Although this response is weak on its own, PU.1 overexpression also reduces expression of the E proteins E2A, HEB (canonical) and HEBalt in Scid.adh.2C2 cells and fetal thymocytes alike (Franco et al., 2006).

To test whether E protein activity could set the threshold for diversion in response to PU.1, we co-expressed Id2 with PU.1 in Scid.adh.2C2 cells. In fact, Id2 and PU.1 together reproducibly increased the percentage of cells becoming Mac1⁺ when compared with PU.1 alone (Fig. 5A). This distinguished Id2 from two other regulators that we tested as alternative candidates for collaborators with PU.1. Both the well-known myeloid factor C/EBPα and the PU.1-induced factor BAMBI failed to increase the percentage of PU.1-transduced Scid.adh.2C2 cells becoming Mac1⁺, although C/EBPα did reduce CD25 expression (supplementary material Fig. S6A and data not shown). Id2 overexpression alone also decreased CD25 levels, although it did not upregulate Mac1. This suggested that the Id2 effect might involve inhibition of Notch signaling. E proteins have been shown to be rate-limiting positive regulators of Notch1 (Yashiro-Ohtani et al., 2009), as well as positive contributors to the expression of some Notch target genes (Ikawa et al., 2006) such as Ptcra.

Gene expression analysis confirmed that Notch target genes are downregulated maximally in samples with Id2 alone, as well as in samples co-expressing PU.1 and Id2 (Fig. 5B; supplementary material Table S2C, group 1). If Id2 overexpression affects the same pathway as Notch inhibition, then forced Notch signaling in PU.1 and Id2 co-expressing samples might be epistatic to Id2. In a triple-transduction experiment, Scid.adh.2C2 cells were infected with PU.1, Id2 and ICN1. Cell surface staining of these cells after 2 days showed that the effect of Id2 to enhance diversion to Mac1⁺ cells was indeed canceled out when Notch signaling was enforced by the addition of ICN1 (Fig. 5A, bottom). Thus, E protein antagonism does play a role in diversion, and induction of Id2 and repression of E2A and HEB probably provide one part of the mechanism through which PU.1 inhibits Notch activation in a positive feedback to promote a myeloid fate.

However, Id2 alone had minimal effect on Gfi1, Myb or Tcf7 expression (supplementary material Table S2C, groups 4 and 5). Furthermore, as reported in earlier E2A knockdown studies (Xu and Kee, 2007), we detected an upregulation of Gata3 with Id2 alone (supplementary material Table S2C, group 2), an effect reversed when PU.1 was present and quite different from the phenotype of diverted cells. Therefore, the mechanism through which these T-lineage regulatory target genes are inhibited by PU.1 to complete diversion is not simply by blocking E protein activity, any more than it is simply by blocking Notch activity.

Myb and Tcf7 were consistently downregulated in response to PU.1 during diversion, and were prominent candidates as diversion ‘barriers’ because the cells do not turn on Mac1 until these two genes are downregulated (Fig. 3C; supplementary material Fig. S3). Myb is already expressed strongly during the first stage of T-cell development (DN1), increasing slightly in the DN2 and DN3 stages (Tydell et al., 2007) (Fig. 2E). To test whether forced expression of Myb could block the ability of PU.1 to upregulate Mac1, we infected Scid.adh.2C2 cells with retroviral Myb for 24 hours, then superinfected them with PU.1 and cultured the cells for an additional 48 hours (Fig. 6A). Despite increasing Myb less than threefold over the level normally expressed in Scid.adh.2C2 cells, co-transduction of Myb with PU.1 modestly but reproducibly decreased the percentage of Mac1⁺ cells (Fig. 6B,C).

Gene expression analysis (Fig. 6D; supplementary material Table S2D) showed that Myb did not inhibit PU.1 from upregulating targets such as Bcl11a. Importantly, protection by Myb did not seem to be mediated primarily through Notch signaling either, as Myb did not prevent PU.1 repression of Notch target genes (supplementary material Table S2D3). However, Gfi1, Tcf7, Gata3 and HEBalt were expressed at higher levels in cells with Myb and PU.1 compared with those with PU.1 alone (supplementary material Table S2D, groups 2 and 4). This group of protected genes was tested in turn for protection against PU.1-mediated diversion, but they did not perform as well as Myb. TCF1 (Tcf7) was a high priority candidate; however, the percentage of cells co-expressing TCF1 and PU.1 that were Mac1⁺ was the same as the percentage of cells expressing PU.1 and an empty vector (supplementary material Fig. S6B), and Tcf7-shRNA did not increase diversion (not shown). Co-expression of Gfi1 or HEBalt with PU.1 also did not block induction of Mac1 (supplementary material Fig. S6C,D). In fact, Gfi1 actually exacerbated the diversion response in the Scid.adh.2C2 cells, and in fetal thymocytes, when co-expressed with PU.1 (supplementary material Fig. S6C and data not shown). Thus, although incomplete, the protective effect of Myb against diversion was specific, implicating Myb and Notch signaling as separate control points for resistance to diversion.

The gene that was most affected by Myb overexpression, one that Myb rendered most resistant to PU.1, was Gata3. Indeed, Myb-transduced cells expressed higher levels of Gata3 RNA than controls with or without PU.1, raising the issue of whether Gata3 could help to resist diversion. Gata3 is essential and rate limiting for T-lineage development and is specifically downregulated in Mac1+ cells (Fig. 3C). It was lower priority to test for control of pro-T-cell lineage fidelity only because the magnitudes of PU.1 and Notch effects on Gata3 RNA were small. To test for Gata3 effects more sensitively at the single-cell level, we performed intracellular staining of the Gata3 protein in Scid.adh.2C2 cells with and without overexpressed PU.1 (Fig. 6E). In fact, PU.1 overexpression markedly downregulated Gata3 protein levels in one subset of the transduced cells, even though it slightly upregulated Gata3 protein, relative to controls, in another subset. This split had the same all-or-none quality as the diversion response itself. Gata3 downregulation was seen at a much greater frequency in cells expressing high levels of PU.1 (Fig. 6E), in which Gata3 levels were 5- to 10-fold reduced. Those cells that downregulated Gata3 protein were also the ones that upregulated Mac1 (Fig. 6F).

Myb may positively regulate Gata3 in later T-cell development (Maurice et al., 2007; Gimferrer et al., 2011). To test whether Myb could also maintain Gata3 despite PU.1 overexpression, we compared Gata3 protein levels in Scid.adh.2C2 cells co-expressing PU.1 and Myb with Gata3 in cells co-expressing PU.1 and an empty vector (Fig. 6G). Cells co-expressing PU.1 with an empty vector showed lowered Gata3 protein levels, but Gata3 was rescued to normal levels in cells co-expressing PU.1 and Myb (Fig. 6G). Guaranteed expression of Myb thus seems to protect Scid.adh.2C2 cells from the PU.1-driven mechanism that downregulates Gata3 protein.

To investigate whether Gata3 downregulation was simply a marker or actually caused differences in the ability of PU.1 to divert the cells, we used shRNA to reduce Gata3 expression in PU.1-transduced cells and measured the impact on diversion. Scid.adh.2C2 cells were first infected with a construct expressing a short hairpin RNA against Gata3 (Hernández-Hoyos and Alberola-Ila, 2005); then after 24 hours the cells were infected with a PU.1-expressing vector and cultured for 48 hours more (Fig. 7) before analysis. The Gata3-shRNA alone knocked down Gata3 protein to levels that were comparable with the lowest Gata3 protein levels in PU.1-expressing cells (Fig. 7A). Unexpectedly, the Scid.adh.2C2 cells co-expressing Gata3 shRNA together with PU.1 had even lower levels of Gata3 protein, some with 20-fold reduction compared with unperturbed cells. This Gata3 reduction made PU.1-expressing cells more susceptible to diversion. As shown in Fig. 7B, the fraction of cells remaining Mac1^- CD11c^- was halved, while increased percentages of cells acquired these myeloid markers.

Reduced Gata3 protein by itself had little effect on gene expression in the Scid.adh.2C2 cells, but the combination of PU.1 expression and Gata3 knockdown had a powerful effect on gene expression (Fig. 7E; supplementary material Table S2E). Gata3 knockdown did not generally cause further upregulation of genes induced by PU.1, and it did not exacerbate PU.1-mediated repression of several Notch targets (supplementary material Table S2E, group 3). However, we found that the multiple T-cell genes downregulated by PU.1 were further downregulated in cells with PU.1 and lowered Gata3 (supplementary material Table S2E, group 4). Notch1 and Notch3 themselves were affected. Loss of Gata3 thus sensitizes cells to the effects of PU.1, with a potency comparable to Notch inhibition.

These results imply that Gata3 downregulation can complement the inhibition of Notch responses by PU.1 and make cells susceptible to diversion. However, our earlier results indicate that direct manipulations of Notch signaling were also sufficient to regulate PU.1-driven diversion, despite little detectable effect of Notch inhibition on Gata3 RNA. In fact, dnMAML alone could slightly elevate Gata3 RNA (Fig. 4B). To revisit whether there is any convergence between these two regulatory mechanisms for protecting T-cell identity, we tested whether manipulations of Notch signaling in the context of PU.1 activity might have clearer effects on Gata3 protein.

By themselves, transduction with dnMAML or ICN1 had virtually no effect on Gata3 protein levels in Scid.adh.2C2 cells (supplementary material Fig. S7). However, when PU.1 transduction was combined with dnMAML or ICN1, the effect on Gata3 was dramatic (Fig. 7C). Cells co-expressing PU.1 and ICN1 uniformly expressed Gata3 at the highest level. By contrast, cells co-expressing PU.1 and dnMAML shifted almost completely to the low level of Gata3, normally seen only in cells with the highest expression of PU.1. Thus, Notch signaling affects not only Notch target gene expression but also the mechanism for Gata3 stabilization, with later impact on Gata3 targets.

Kinetically, the impact of Notch inhibition on Gata3 levels could precede appearance of the diverted phenotype (Fig. 7D). Mac1 expression is not evident on PU.1-transduced Scid.adh.2C2 cells until 48 hours (Fig. 7D, right panel). However, the combination of PU.1 and dnMAML began to downregulate Gata3 protein in the whole population of transduced Scid.adh.2C2 cells by 24 hours, falling lower by 48 hours. Thus, any decrease in Notch signaling undermined the resistance of Gata3 in the cells to inhibition by PU.1, precipitating the positive-feedback cascade that eventually silences genes dependent on Gata3 and Notch signaling alike.

T-lineage specification of blood-cell precursors is promoted by Notch interaction, with Delta expressed in the thymic micro-environment. However, throughout multiple cell cycles in this environment, the differentiating precursors continue to express transcription factors such as PU.1 that are associated with multipotentiality. Their access to other fates is revealed if removed from the thymus. How does the thymus predictably manage to impose a T-cell fate on virtually all these cells, despite their intrinsic delay of commitment? Our results reveal the architecture of a regulatory gene network switch circuit through which environmental Notch signaling interacts with PU.1 to determine T-cell, myeloid or progenitor-cell status (Fig. 8).

Two branches of this network are positively regulated by PU.1. One involves upregulation of myeloid genes such as Itgam and Csf1r, whereas another involves expression of stem-cell or progenitor-cell genes such as Bcl11a, Lyl1 and possibly also Bambi. Environmental Notch signaling blocks activation of Itgam and Csf1r, but not PU.1-dependent activation generally. Concomitantly, there are two network branches through which PU.1 can negatively regulate the T-cell differentiation program. Extinction of T-lineage regulatory gene expression is most tightly correlated with a switch to myeloid fate. One branch involves the ability of PU.1 to attenuate transcriptional responses to Notch signaling: PU.1 raises the threshold of Notch signaling needed for expression of Notch target genes. This occurs in part through inhibition of an E protein - Notch-positive feedback circuit. In parallel, however, we show that PU.1 also antagonizes expression of a second set of T-cell regulatory genes, including Myb and its activation target Gata3. These seem crucial for sustaining Gfi1, Zfpm1 (Fog1) and Tcf7 expression in the presence of PU.1. All these genes can also be protected against PU.1 by Notch signals, but are not otherwise Notch dependent, implying that the protective effect of Notch on this gene set is indirect, e.g. via maintenance of Gata3. Myeloid-lymphoid lineage choice is thus a bifurcation between opposing feed-forward network circuits, one dominated by PU.1, and the other by Notch signals that protect both Notch-E protein targets and Myb-Gata3 targets.

Our results suggest that the balance may be tipped from resistance to diversion by initial weakening of either protective mechanism in PU.1-expressing cells. Reduction either of Gata3 or of Notch signaling can sensitize the cells to diversion, and Notch signaling not only protects Gata3 but also protects its positive regulator, Myb, from inhibition by PU.1. However, it is notable that when PU.1 and Notch signals ‘balance’, T-cell regulatory gene expression can be maintained, along with expression of specific progenitor-associated PU.1 target genes. This is exactly the situation in early T-cell precursors before lineage commitment (Fig. 1A). Our results with Bcl11a, Lyl1 and possibly also Hhex and Bambi regulation, all of which are naturally expressed in early thymocytes, thus open the way for PU.1 to play a stage-specific positive role for early T cells.

Our results are drawn from both primary fetal thymocytes and a DN3-like clonal cell line, and the relationships are similar if not completely identical. Scid.adh.2C2 cells do not perfectly match the gene expression states of the primary cells, and as magnitudes of specific gene expression responses to PU.1 change with normal developmental progression, they also differ between the cell line and the primary cells. These probably reflect differences in basal Notch transduction machinery, E protein activity and Gata3 expression between these cell types (data not shown). Tcf7 is less protected by Notch signaling in the primary cells than in Scid.adh.2C2 cells, whereas genes such as Ptcra are more protected. However, these are not the PU.1 repression targets that appear to set the threshold against diversion. Instead, the key components of the network core architecture shown in Fig. 8 are consistent with results in both types of cells.

PU.1 opposition to Gata3 recalls the PU.1:GATA1 opposition that underlies erythroid/myeloid fate determination, which is based in part on protein-protein interaction (reviewed by Cantor and Orkin, 2002; Laiosa et al., 2006a). Here, Gata3 appears to be important for Zfpm1 maintenance against PU.1, like GATA1 in erythroid development. ChIP-seq analysis shows that Zfpm1, Gfi1, Myb and Tcf7 are all linked with Gata3 binding sites in early T cells (Zhang et al., 2012) (Bambi, Bcl11a, Itgam and Id2 are not), suggesting that high PU.1 may primarily inhibit these T-cell genes by blocking positive Gata3 inputs. However, Gata3-PU.1 antagonism itself is more conditional. Although PU.1 reduces Gata3 protein when Notch signaling is inhibited, PU.1 slightly upregulates Gata3 when Notch signals are active. PU.1 binds multiple sites around Gata3 in early T cells (Zhang et al., 2012), potentially contributing to both effects. PU.1 can be repressed by high-level Gata3 (Taghon et al., 2007), but the genomic sites through which GATA1 silences PU.1 expression (Chou et al., 2009) are not bound by Gata3 in early T cells (Zhang et al., 2012). Furthermore, although reduced Gata3 makes cells more diversion sensitive, increased Gata3 cannot bypass the need for Notch signaling to make cells diversion-resistant (data not shown). This suggests that PU.1-Gata3 relationships are probably asymmetric.

The relationship between PU.1 and Notch signaling provides a discrete, micro-environmental threshold setter for lymphoid precursor fate determination. In normal thymocytes though not in Scid.adh.2C2 cells, the signals actually received depend on environmental density of Notch ligands. Within the pro-T cells, signaling not only requires E proteins to maintain Notch1 expression but also a positive-feedback loop with E protein activity, as expression of both Id2 and Id3 E protein antagonists increases when Notch signaling is reduced. The molecular mechanism through which PU.1 inhibits Notch-dependent transcription still requires more investigation. However, our results show that the expression of PU.1 in the earliest T-cell precursors itself becomes a sensor that determines what level of Notch signal from the environment will suffice to promote entry and forward progression along the T-cell pathway.

We thank Avinash Bhandoola, Warren Pear and Gabriela Hernandez-Hoyos for constructs; the entire Rothenberg lab for help and valuable discussions; Robert Butler for technical expertise; Va Si for pilot Gata3 staining experiments; Rochelle Diamond, Diana Perez and Josh Verceles for cell sorting; and Scott Washburn for mouse care.