Combined heterozygous loss of Ebf1 and Pax5 allows for T-lineage conversion of B cell progenitors

To investigate the functional consequence of combined reductions of Ebf1 and Pax5 dose, we generated mice transheterozygous (TH) for mutations in the Ebf1 and Pax5 genes. These animals did not display any decrease in the total number of the pro–B cells as compared with that of the Wt or the single mutant mice (Fig. 1 A). In contrast, the Lin-B220⁺CD19⁺CD43^low/negIgM⁻ (pre–B cell) compartment was reduced in TH mice as compared with what was observed in any of the single mutant mice. The number of Lin⁻B220⁺CD19⁺CD43^low/negIgM⁺IgD⁻ cells was reduced to levels comparable to single Ebf1^+/− mice in the Pax5^+/−Ebf1^+/− animals (Fig. 1 A). The formation of early progenitor cells in the Pax5^+/−Ebf1^+/− mice suggested that the combined dose reduction in Ebf1 and Pax5 would not result in a complete disruption of the autoregulatory loop between Ebf1 and Pax5 and collapse of the genetic program. Consistent with this notion, Q-PCR analysis of primary pro–B cells from Wt and Pax5^+/−Ebf1^+/− mice (Fig. 1 B) suggested a twofold down-regulation of Ebf1 and Pax5 transcripts, respectively in pro–B cells from the TH mice, as would be expected from loss of one functional allele.

To investigate the functional potential of Pax5^+/−Ebf1^+/− B cell progenitors in vivo, we transplanted CD19⁺IgM⁻ cells from Wt and Pax5^+/−Ebf1^+/− CD45.2 mice into Rag1^−/− CD45.1 animals. Six to nine weeks after transplantation, we analyzed the cellular composition of CD45.2⁺ cells in the spleen of the transplanted mice. While over 93% of the cells in the mice transplanted with Wt progenitors expressed CD19, only one of the analyzed mice transplanted with Pax5^+/−Ebf1^+/− B cell progenitors presented with large numbers of CD19⁺ cells (Fig. 1 C). The majority of the mice transplanted with Pax5^+/−Ebf1^+/− progenitors presented a large fraction of Thy1.2⁺ cells, representing a most limited population in mice transplanted with Wt cells (Fig. 1 C). Donor cells expressing NK1.1 or Gr1/Mac1 were the most limited or undetectable (Fig. 1 C). Analysis of the CD19⁺ compartment revealed that the major part of the CD19⁺ cells generated from both Wt and Pax5^+/−Ebf1^+/− B cell progenitors represented IgM⁺ or IgM⁺IgD⁺ B cells (Fig. 1 D). However, in the three Pax5^+/−Ebf1^+/− progenitor transplanted mice displaying the largest fraction of CD19⁺ cells, a major part of the cells displayed the phenotype of pro–B cells with lack of IgM expression and high expression of CD43 (Fig. 1 D). These data reveal that Pax5^+/−Ebf1^+/− B cell progenitors are deficient in their ability to generate mature B-lineage cells, and the presence of large amounts of pro–B cells in the spleen of three of the transplanted mice indicate that these cells may be prone to malignant transformation. Whereas the majority of the mice transplanted with Wt cells presented a limited frequency of Thy1.2 high cells, this marker was highly expressed on the majority of the cells generated from Pax5^+/−Ebf1^+/− B cell progenitors. High Thy1.2 expression is normally restricted to T-lineage cells and to investigate the identity of these cells in more detail, we stained spleen cells with antibodies detecting CD4, CD8, and TCRβ. Virtually all the Thy1.2⁺ cells expressed TCRβ, and we could observe CD4 and CD8 single-positive as well as double-positive donor cells in the spleen (Fig. 1 E), strongly supporting the notion that the B cell progenitors from Pax5^+/−Ebf1^+/− mice had preferentially generated T-lineage cells after transplantation in vivo.

To investigate if we could detect evidence of immunoglobulin heavy chain VDJ recombination in the generated T-lineage cells, we sorted CD3⁺TCRβ⁺ cells from mice transplanted with TH cells, extracted DNA, and investigated the presence of VDJ recombination by PCR using one primer in J3 in combination with three degenerated primers directed toward the V-gene families J558, Q52, and 7183 (Schlissel et al., 1991). This indicated that the T cell generated from transplanted Pax5^+/−Ebf1^+/− cells carried detectable polyclonal immunoglobulin VDJ recombination events, supporting the notion that the cells were generated from B-lineage progenitors (Fig. 1 F). Analysis of the thymus of transplanted mice revealed that Pax5^+/−Ebf1^+/− B cell progenitors repopulated the thymus more efficiently than Wt cells (Fig. 1 G). Although the cells from Wt progenitors displayed a substantial amount of CD19⁺ cells in the thymus of the recipients, with the exception of one mouse with peripheral pro–B cell expansion, Pax5^+/−Ebf1^+/− B cell progenitors generated Thy1.2⁺ cells being either CD4 or CD8 single positive. A small fraction of double-positive cells could also be detected in the thymus of transplanted mice. In contrast to what was observed in the spleen, we could also detect a small fraction of NK1.1⁺ cells generated from both Wt and Pax5^+/−Ebf1^+/− B cell progenitors. Even though we did detect T-lineage cells in mice transplanted with Wt CD19⁺IgM⁻ cells, the frequency was significantly lower than what we observed in mice transplanted with Pax5^+/−Ebf1^+/− progenitors, revealing that combined reduction of Pax5 and Ebf1 dose results in abnormal lineage plasticity in vivo after transplantation to lymphocyte-deficient mice.

The finding that CD19⁺ progenitors from Pax5^+/−Ebf1^+/− mice generated T-lineage cells in vivo suggest that even though these cells express CD19, they are not stably committed to B-lineage cell fate. To investigate the lineage fidelity of B cell progenitors generated with reduced levels of Pax5 and Ebf1 in more detail, we sorted pro–B cells and seeded them on OP9-DL1 stroma cells, generating conditions permissive for T-lineage development of noncommitted progenitors (Schmitt and Zuñiga-Pflucker, 2002). Seeding 10 Wt pro–B cells per well, we detected growth in 32% of the wells. The overall cloning frequency of Pax5^+/− cells was 7%, a decrease as compared with Wt cells (P < 0.0001), whereas ∼24% of the wells seeded with Ebf1^+/− cells generated colonies, suggesting that the cloning frequency was reduced by loss of one allele of Pax5. Pax5^+/−Ebf1^+/− cells generated colonies in almost 50% of the wells. Upon analysis of the cellular content of the cultures, 100% of the colonies generated from Wt progenitors contained only CD19⁺ cells, whereas wells seeded with Pax5^+/− or Ebf1^+/− cells generated a low frequency of colonies with CD19⁻ cells expressing Thy1.2 (Fig. 2, A and B). Analysis of the content in wells seeded with pro–B cells from Pax5^+/−Ebf1^+/− mice revealed that the majority of these cultures were composed of a mixture of cells, including CD19⁻Thy1.2⁺ and CD3⁺ cells (Fig. 2, A and B), supporting an increased T cell potential in Pax5^+/−Ebf1^+/− as compared with Wt (P < 0.0001) or single heterozygous mice. To verify plasticity at the single-cell level, we seeded single Wt or Pax5^+/−Ebf1^+/− pro–B cells on OP9-DL1 stromal cells. Analysis of the cellular content of the single cell cultures revealed that although all the Wt cells generated CD19⁺ cells, a majority of the Pax5^+/−Ebf1^+/− cells generated mixed colonies (Fig. 2 B), revealing plasticity at the single-cell level. Inclusion of the γ-secretase inhibitor DAPT in the cell cultures inhibited the formation of CD19⁻Thy1.2⁺ cells, confirming a need of Notch signaling to initiate the phenotypic change (Fig. 2 B). We could also verify that the plasticity was critically dependent on the Ebf1 dose in the pro–B cells because retroviral transduction of Pax5^+/−Ebf1^+/− pro–B cells with an Ebf1 encoding retrovirus blocked the formation of CD19⁻Thy1.2⁺ cells (Fig. 2 E). Even though high expression of Thy1.2 and CD3 normally serves as reliable markers for cells that have entered the T-lineage pathway, we extended our analysis by performing a Q-PCR analysis of colonies containing either CD19⁺ or Thy1.2⁺ cells generated from Wt pro–B cells or from Pax5^+/−Ebf1^+/− pro–B cells, respectively. Although the expression of B-lineage genes such as CD19 and Pax5 was decreased as compared with Wt cells cultured under the same conditions (Fig. 2 C), analysis of the levels of T cell–associated transcripts including CD3e, Pre-Tα, Lck, and CD27 suggested that these genes were all expressed in the generated Thy1.2⁺ cells, verifying that the cells had initiated development toward T-lineage. Culturing Wt or Pax5^+/−Ebf1^+/− pro–B cells on OP9-DL1 cells for 21 d, followed by replacement of IL-7 with IL-2 for 10 additional days, resulted in cell death among the Wt cells, whereas the Pax5^+/−Ebf1^+/− cells generated offspring expressing TCRβ on the surface (Fig. 2 D).

Calculating the efficiency of the seeded pro–B cells to generate CD3⁺ offspring in vitro revealed that although none of the Wt cells generated CD3⁺ cells, 1 out of 1,819 of the Ebf1^+/− and 1 out of 2,223 Pax5^+/− cells displayed this ability. Among the seeded Pax5^+/−Ebf1^+/− cells, 1 out of 37 seeded cells was able to generate CD3⁺ progeny. Hence, even though the overall cloning frequency differed (1/25 for Wt, 1/39 for Ebf1^+/−, 1/166 for Pax^+/−, and 1/19 for Pax5^+/−Ebf1^+/− pro–B cells), the combined reduction of Ebf1 and Pax5 dose results in a synergistic increase in T cell potential in CD19⁺ B cell progenitors. This argues against the idea that the change in lineage would be the result of selection of a rare population of plastic cells enriched in the Pax5^+/−Ebf1^+/− pro–B cell population.

Lineage plasticity in Ebf1^−/− or Pax5^−/− B cell progenitors has been linked to rather drastic changes in gene expression patterns, including increased expression of T-lineage associated genes such as Notch1 (Souabni et al., 2002; Nechanitzky et al., 2013). To investigate how the combined dose reduction of Pax5 and Ebf1 would impact the transcriptional program in the pro–B cell compartment, we sorted pro–B cells from Wt, Ebf1^+/−, Pax5^+/−, and Pax5^+/−Ebf1^+/− animals and performed RNA sequencing analysis. Analysis of the data suggested that even though the transcriptome of Wt and Pax5^+/−Ebf1^+/− pro–B cells differed in expression at 204 genes (Fig. 3 A), the levels of classical B- or T-lineage genes, including Ebf1 and Pax5 target genes such as CD79α, CD79β, Notch1, or Gata3, were not significantly altered in mice carrying combined heterozygous deletion of Ebf1 and Pax5. K.E.G.G pathway analysis of the differentially expressed genes highlighted a significant difference (Benjamini-Hochberg corrected, P < 0.05) in the categories Cytokine–cytokine receptor interactions, Acute Myeloid Leukemia, Nod-like receptor signaling pathway, Leukocyte transendothelial migration, and NK cell–mediated cytotoxicity. Even though genes involved in cellular signaling were differentially expressed, we did not detect any obvious changes that could be linked to the cellular plasticity observed in Pax5^+/−Ebf1^+/− pro–B cells. Analysis of the gene expression data suggested that a major part of the genes differentially expressed in the Pax5^+/−Ebf1^+/− cells was dependent on the combined loss of Ebf1 and Pax5, as the RNA levels were not significantly altered in the single heterozygous cells. To investigate if the differentially expressed genes were direct targets for Ebf1, Pax5, or both, we investigated the binding of these transcription factors in the proximity of the differentially expressed genes using existing Rag2^−/− pro–B cell ChIP-sequencing data (Revilla-I-Domingo et al., 2012; Vilagos et al., 2012). This indicated that 85 (42%) of the differentially expressed genes (≥2-fold) contained either overlapping or nonoverlapping binding sites for both Ebf1 and Pax5 (Fig. 3, B and C), whereas an additional 32% contained binding sites for either Ebf1 or Pax5 (Fig. 3 B). Hence, the majority of the differentially regulated genes are direct targets for Ebf1, Pax5, or both. 29% of the up-regulated and 33% of the down-regulated genes contained overlapping peaks (Fig. 3 C), supporting the notion that Pax5 and Ebf1 share a substantial number of regulatory elements.

To investigate how a reduction in Ebf1 dose impacts DNA binding, we performed Ebf1-ChIP-seq analysis on Wt, Ebf1^+/−, Pax5^+/−, as well as Ebf1^+/−Pax5^+/− pro–B cells. This allowed for the identification of 8161 peaks in the Wt samples, whereas 7114 peaks were detected in the TH cells (Fig. 3 D). 4742 of the peaks were detected as bound by Ebf1 in both Wt and Ebf1^+/−Pax5^+/− pro–B cells, whereas 3419 and 2373 unique peaks were identified in the Wt and TH, respectively. Motif enrichment analysis suggested that the transcription factor dose had a limited impact on the selection of either Ebf1 or associated binding sites (Table S1). Sites conforming to the Ebf1 consensus site were detected in 64% (P = 10⁻⁴⁸⁷⁸) of the Wt peaks and in 73% (P = 10⁻⁵⁰¹³) of the peaks generated from TH cells. In addition to the Ebf1 consensus site, we detected enrichment in binding sites for Ets and Runx proteins as well as an E-box motif, independently of genotype (Table S1). To investigate how Ebf1 binding to regulatory regions was modulated by a reduction in transcription factor dose, we compared Ebf1 binding in Wt and TH cells on sites occupied by Ebf1 in Wt cells. This revealed that a total of 1,635 sites were differentially bound (P < 0.05), with approximately as many displaying reduced as increased binding in the TH cells. Increasing the stringency in the analysis, demanding a fourfold change in binding, identified 444 peaks. Investigating Ebf1 binding in a 5-kb window centered on peaks defined as Wt peaks with differentially binding revealed that a majority (341) of these 444 differentially bound peaks had a decrease in Ebf1 binding in TH cells as compared with Wt cells (Fig. 3 D). 104 of the genes displayed increased binding of Ebf1 in the TH as compared with the Wt cells. To link changes in gene expression to Ebf1 binding, we identified the genes closest to the 1,653 sites differentially bound by Ebf1 in Wt and TH cells. This resulted in the identification of 1,467 genes (Fig. 3 E), out of which 24 were among the 85 genes identified as differentially expressed direct targets for both Ebf1 and Pax5 in Fig. 3 B. Ebf1 binding was studied in a 5-kb window around differentially bound Ebf1 peaks between Wt and TH (P < 0.05) found within a 100-kb window of up- or down-regulated genes. The result suggests that genes down-regulated in the Ebf1^+/−Pax5^+/− pro–B cells displayed reduced Ebf1 binding on these sites, whereas up-regulated genes instead displayed an increased Ebf1 binding (Fig. 3 E), suggesting a correlation with changes in Ebf1 binding and gene expression. Hence, although the consensus-binding sites remain the same, combined heterozygote deletion of the Ebf1 and Pax5 genes impacts the binding patterns of Ebf1 to target genes in the pro–B cells.

Although the generation of T-lineage cells from Pax5 and Ebf1 deficient B or pro–B cells has been reported to be a result of dedifferentiation (Cobaleda et al., 2007; Nechanitzky et al., 2013), the generation of macrophages from B-lineage cells has been suggested to be a result of direct lineage conversion (Xie et al., 2004). To understand the basic mechanism for the generation of T-lineage cells from Pax5^+/−Ebf1^+/− pro–B cells, we analyzed the donor derived progenitor compartments in transplanted Rag1^−/− mice. However, we did not detect CD45.2⁺ common lymphoid progenitors (CLPs) or lymphoid-primed multipotent progenitors (LMPPs) in either BM or spleen. Hence, we were unable to detect signs of dedifferentiation of pro–B cells into classical hematopoietic progenitor compartments, arguing against the formation of T-lineage cells involving genuine dedifferentiation into conventional hematopoietic progenitors. To investigate the process in more detail, we sorted Wt and Pax5^+/−Ebf1^+/− pro–B cells, seeded them on OP9-DL1 cells, and followed the formation of T-lineage cells over time (Fig. 4, A and B). After 4 d of culture, the major part of the cells expressed CD19 but no, or low, levels of Thy1.2. After 8 d of incubation, we noted an increase in Thy1.2 expression generating CD19⁺Thy1.2⁺ cells, and after 10 d of incubation, a subpopulation of the cells lost expression of CD19. This subpopulation increased with time and, after 14 d of incubation, this was the dominant population in the cultures. At 16 d after incubation, we observed an increasing population of CD3⁺ cells. Initiation of normal T cell development has been shown to involve Gata3 and TCF7 (Weber et al., 2011; Banerjee et al., 2013); however, not until 10 d after the initiation of the experiments, at a time when we started to observe CD19⁻ cells (Fig. 4, A and B), did we detect increased expression of Gata3 and Tcf7 in the Pax5^+/−Ebf1^+/− cells by Q-PCR (Fig. 4 C). This indicates that the expression of these transcription factors is not significantly preceding the development of CD19⁻ cells in this process suggesting that the T cells are generated via a Thy1.2⁺CD19⁺ double positive state, indicative of lineage conversion rather than dedifferentiation. To establish links between the functional lineage conversion process and the defined populations, we sorted CD19⁺Thy1^High, as well as CD19⁻Thy1^High cells from a 10-d OP9-DL1 culture of Pax5^+/−Ebf1^+/− pro–B cells. The cells were then reseeded on either OP9 or OP9-DL1 cells and analyzed for the formation of B- or T-lineage cells after an additional 10 d in culture (Fig. 4 D). This revealed that CD19⁺Thy1^High cells were able to generate CD19⁺ cells when reseeded on OP9 cells, whereas, if seeded on OP9-DL1 cells, the absolute majority of the CD19⁺Thy1^High cells lost the expression of CD19. CD19⁻Thy1^High cells were unable to generate CD19⁺ cells even when cultured in the absence of Delta ligand; however, only OP9-DL1 cells supported the formation of CD3-expressing cells (Fig. 4 D). These data support the idea that the conversion process involves the formation of a bivalent state capable of adopting either B-or T-lineage cell fate.

The change of lineage fate in Pax5^−/− or Ebf1^−/− mice has been attributed to increased expression of Notch1, presumably making the cells more sensitive to Notch ligand in the environment (Souabni et al., 2002; Nechanitzky et al., 2013). However, in contrast to what has been reported from these models of differentiation, we did not detect any significant increase in Notch1 in Pax5^+/−Ebf1^+/− pro–B cells. Hence, another mechanism of action should be operating in this model system. To investigate the direct cellular response to Notch ligand, we incubated Wt and Pax5^+/−Ebf1^+/− pro–B cells on OP9-DL1 cells for 24 h and analyzed the formation of intracellular Notch (ICN) in the two cell types (Fig. 5 A). This suggested that both the Wt and the Pax5^+/−Ebf1^+/− pro–B cells responded with the formation of ICN; however, Q-PCR analysis suggested that the Notch target genes Deltex1 and Notch1 were not induced in the Wt cells as observed in the Pax5^+/−Ebf1^+/− pro–B cells (Fig. 5 B). To investigate the response to Notch signals in more detail, we analyzed the induction of genes in response to Notch signal by RNA sequencing analysis in Wt and Pax5^+/−Ebf1^+/− pro–B cells 24 h after seeding onto OP9-DL1 cells (Fig. 5 C). This revealed that whereas both Wt and Pax5^+/−Ebf1^+/− pro–B cells responded to the Notch signal by up-regulation of known Notch target genes, such as Fosl1, Fabp4, Dll1, and Heyl, only the Pax5^+/−Ebf1^+/− cells up-regulated the expression of Dtx1 (Fig. 5 B). To verify the RNA seq data, we performed Q-PCR analysis examining the expression of Fabp4 (Fig. 5 D). This verified the RNA seq data, suggesting that both Wt and TH cells responded by induction of Fabp4 transcription (Fig. 5 D), whereas Deltex1 was selectively activated in the Pax5^+/−Ebf1^+/− cells (Fig. 5 B). This supports the idea that both Wt and Pax5^+/−Ebf1^+/− cells respond to Notch signals but that the response differs in regard to the activation of target genes. The induction of several genes, including Fabp4, Fosl1, and Dll1 were as strong or stronger than what we observed for Pax5^+/−Ebf1^+/− cells, arguing against substantial difference in signal strength as an explanation for the difference in Notch response. Our analysis, suggested that Wt cells in total up-regulated 180 genes more than twofold whereas Pax5^+/−Ebf1^+/− cells up-regulated 51 genes (Fig. 5 C). Extracting information about genes more than twofold down-regulated by Notch signaling in the Wt and Pax5^+/−Ebf1^+/− cells suggested that although 6,423 genes were down-regulated by Notch signaling in the TH cells, only 245 genes were down-regulated in the Wt cells. K.E.G.G pathway analysis suggested that 46 of the genes down-regulated in the Pax5^+/−Ebf1^+/− cells belonged to the category B cell receptor-signaling pathway (P = 7.0 × 10⁻⁷). Analysis of the data revealed that the down-regulated genes included CD19, CD79α, CD79β, as well as Ebf1 and Pax5 themselves, whereas none of these genes were affected in the Wt cells. This idea was verified by Q-PCR analysis revealing significant down-regulation of Pax5 after Notch activation selectively in the TH cells (Fig. 5 D). Hence, Pax5^+/−Ebf1^+/− cells respond to Notch signaling by a general down-regulation of the transcriptional program defining the early B-lineage cells in a manner not observed in Wt cells. To investigate if the modulated genes are direct targets of Ebf1 and/or Pax5, we analyzed this in relation to the Rag2^−/− pro–B GEO ChIP seq datasets (Revilla-I-Domingo et al., 2012; Vilagos et al., 2012), which revealed that a majority (4,351 out of 6,474) of these differentially regulated genes were directly bound by either one or both of these transcription factors (Fig. 5, E and F). 18% of the genes up-regulated and 17% of those down-regulated displayed overlapping peaks (Fig. 5 F). Performing KEGG pathway analysis of up-regulated genes with combined Ebf1/Pax5 binding revealed that the only significantly ordered group of genes was in the Notch signaling pathway. Performing the same type of analysis on down-regulated genes revealed the B cell receptor signaling pathway as the most significantly affected pathway. To investigate if the differential response to Notch signaling was reflected in differential Ebf1 binding before Notch activation, we investigated the binding of Ebf1 (defined by the total filtered Wt list) in a 100-kb window, to genes with a stronger response (twofold) to Notch stimuli in TH that were either up- or down-regulated in the TH cells as compared with nonstimulated cells. This revealed that whereas Ebf1 binding to genes up-regulated by Notch signaling was not decreased, the binding to genes down-regulated displayed reduced Ebf1 binding, even before Notch activation (Fig. 5 G).

The data presented in this report support the idea that Ebf1 and Pax5 act in a coordinated, dose-dependent manner to preserve B-lineage cell fate. This is well in line with the findings that Ebf1 and Pax5 act in an autoregulatory loop and that complete loss of function of either of these proteins results in increased plasticity of B cell progenitors. Just as for normal T cell development (Radtke et al., 1999; Wilson et al., 2001), Notch signaling had an essential role in the lineage conversion process in vitro because the phenotypic changes were only observed on OP9-DL1 cells and prevented by the inclusion of DAPT (Fig. 2 B). It is reasonable to presume that activation of Notch signaling would be a requirement in vivo where the cells could be exposed to Notch ligands in the thymus (Felli et al., 1999; Schmitt et al., 2004) but also in other locations such as the spleen (Caton et al., 2007). Exposure to Notch ligands in peripheral organs might also explain the presence of CD4CD8 double positive cells in the spleens of Rag mice transplanted with Ebf1^+/−Pax5^+/− pro–B cells (Fig. 1 E). Even though we do detect some nonconventional populations of both B and T cells in the periphery of the transplanted mice, the majority of the TCRβ-expressing cells are either CD4 or CD8 single-positive, resembling major T cell populations in normal mice. In contrast, the majority of the TCRβ⁺ cells generated in vitro (Fig. 2 D) did not express either CD4 or CD8. Even though CD3⁺TCR⁺CD4⁻CD8⁻ immunoregulatory T cells has been identified in the periphery of normal mice and humans (Hillhouse and Lesage, 2013), the generation of both CD4 and CD8 single-positive T cells in vivo argues against this indicating that the Ebf1^+/−Pax5^+/− pro–B cells would be primed toward development into any specific T cell subtype.

It has been shown that the generation of T-lineage cells from Pax5^−/− (Cobaleda et al., 2007) or Ebf1^−/− (Nechanitzky et al., 2013) B cell progenitors involves a component of dedifferentiation. This is somewhat in contrast to our data suggesting that the generation of T-lineage cells from Ebf1^+/−Pax5^+/− CD19⁺ B cell progenitors instead involves the formation of a bivalent state displaying features of both B and T-lineage cells and with ability to adopt either of the two cell fates. This does in many regards resemble the lineage conversion process reported to generate macrophages from B cells in response to expression of Cebpα (Xie et al., 2004; Di Tullio et al., 2011). However, it should be noted that the dedifferentiation results reported in Pax5- and Ebf1-deficient mice involves the generation of rather atypical progenitor cells in some aspects resembling early pro–B cells (Cobaleda et al., 2007; Nechanitzky et al., 2013) and that these cells may in some regards correspond to the CD19⁺ pro–B cells we detect several weeks after transplantation of Pax5^+/−Ebf1^+/− cells (Fig. 1 D). We could, however, not detect the development of myeloid cells in any of our mice, in contrast to what has been reported for Pax5^−/− B cells (Cobaleda et al., 2007). A potential discrepancy could be a result of that as opposed to what has been reported for Pax5^−/− or Ebf1^−/− cells (Cobaleda et al., 2007; Nechanitzky et al., 2013) the progenitor compartments in Pax5^+/−Ebf1^+/− mice express a full B-lineage program and possess normal survival capability, eliminating the need of the transgenic expression of Bcl2 used in previous studies (Cobaleda et al., 2007; Nechanitzky et al., 2013). Even though differences in survival and potential competition between different populations in the cell cultures make it difficult to exactly estimate the frequency of Pax5^+/−Ebf1^+/− CD19⁺ progenitor cells that can undergo the conversion into T-linage cells, our data argue against the idea that this would be the result of selection of a small population of plastic cells. The in vivo experiments generated T cells with polyclonal VDJ rearrangements and 80% of the single cells able to generate colonies generated T-lineage cells in vitro. The calculated T cell cloning frequency of 1/37 was comparable to that of CD19⁺ cells in the same experiment, hence, even though we do not want to conclude that all cells possess the ability to change lineage, a significant portion of the CD19⁺ progenitors clearly display a plasticity toward T-lineage development.

Whereas the complete loss of function of either Ebf1 or Pax5 has been shown to result in increased expression of genes associated with other cell fates, including Notch1 (Souabni et al., 2002; Nechanitzky et al., 2013), the combined heterozygote loss of Ebf1 and Pax5 does not appear to cause any major disruptions in the genetic program of the pro–B cells (Fig. 3 A). Even though we did detect significant differential expression of a set of genes, several of which appear as direct targets for Ebf1 and Pax5, we were unable to find significant changes in genes that, to our knowledge, directly repress functional Notch signaling. Even though the complexity of this signaling pathway calls for caution in the interpretation of negative data, we believe that our short term gene expression analysis provide support for the idea that both Wt and Ebf1^+/−Pax5^+/− cells respond to Delta1 (Dll1) stimulation. Wt and Ebf1^+/−Pax5^+/− cells responded by increased expression of a set of genes suggested to be Notch targets, including Fabp4 (Harjes et al., 2014) and Heyl (Maier and Gessler, 2000), and also of the Notch ligand Dll1. In contrast, only the Ebf1^+/−Pax5^+/− cells responded by induction of the Notch target Deltex1 (Deftos et al., 2000). This was rather unexpected considering that Deltex1 is a powerful inhibitor of Notch signaling, driving progenitor cells into a B cell fate (Izon et al., 2002) and capable of preventing the induction of T cell fate in Pax5^−/− cells (Zandi et al., 2012). However, because this gene has regions of overlapping binding of Ebf1 and Pax5, it is likely directly under the control of these transcription factors in normal cells. Binding of Ebf1 and or Pax5 could be detected in 27 of the 50 genes up-regulated by the Notch signal in the Ebf1^+/−Pax5^+/− cells; however, because 14 of these 27 also responded in the Wt cells, binding of Ebf1 or Pax5 is not fully correlated to selective activation in the mutant cells. Even though there were differences in the spectra of activated genes, the most dramatic discrepancy between Wt and Ebf1^+/−Pax5^+/− cells was the down-regulation of a large number of genes, including a large part of the B cell–specific program, in the TH cells in response to the Notch signal. This included down-regulation of Pax5 and Ebf1 as well as of a large set of target genes, indicating that the Notch signal directly targets the Ebf1–Pax5 network. Notch1 has been shown to target Ebf1 posttranscriptionally (Smith et al., 2005), possibly adding to any direct effects on Ebf1 or Pax5 transcription. This, together with the finding that Gata3 and Tcf7 are significantly up-regulated first after 10–12 d of OP9-DL1 stimulation (Fig. 4 C), indicates that the initial response of the CD19⁺CD43⁺IgM⁻ TH cells to the Notch signal is to repress the B cell program, whereas the T cell identity genes in general requires more time and are up-regulated at a later phase of the conversion process.

In conclusion, we believe that our data support the idea that Ebf1 and Pax5 collaborate to preserve lineage identity and homeostasis in early B cell development. Even though this is well in line with previous work in this area, the fact that we are able to look at lineage conversion in cells with a well preserved transcriptional program allows us to obtain an increased understanding of a lineage specification process. This insight strongly suggests that Ebf1 and Pax5 counteract an activated Notch signal by repressing the functional response initiated by the ICN protein, thereby preserving the lineage identity of B cell progenitors.

Pax5^+/− (Urbánek et al., 1994) or Ebf1^+/− (Lin and Grosschedl, 1995) were all on C57BL/6 (CD45.2) background (Backcrossed for >10 generations) whereas Rag1^−/− animals were on C57BL/6 (CD45.1). Animal procedures were performed with consent from the local ethics committee at Linköping University (Linköping, Sweden).

Adoptive transfers were performed by tail vein injection. Sublethally irradiated (4.5 Gy) Rag1^−/− CD45.1 animals were transplanted with ∼2 million Lin⁻B220⁺CD19⁺IgM⁻ Pax5^+/−Ebf1^+/− or Wt cells as controls.

For analysis and cell sorting of B lineage cells, CD16/CD32 (FC)-blocked (93; eBioscience) cells were stained with antibodies against lineage markers CD11b/Mac1 (M1/70), Gr1 (RB6-8C5), TER119 (Ter119), CD3 (17A2; BD), CD11c (N418), and NK1.1 (PK136), followed by additional staining with CD19 (ID3), CD45R/B220 (RA3-6B2), CD43 (S7), IgM (RMM-1), and IgD (11–26; eBioscience). Analysis and cell sorting was performed on a FACSAria (BD) using propidium iodide (PI; Invitrogen) as a viability marker. Samples in Fig. 1 A were analyzed with a FACSCantoII (BD) and dead cells were excluded with 7-aminoactinomycin D (7-AAD; Sigma-Aldrich). Lymph node cells were FC-blocked and stained with CD11b/Mac1 (M1/70), Gr1 (RB6-8C5), CD3 (17A2), CD19 (ID3), CD45R/B220 (RA3-6B2), CD43 (S7), IgM (RMM-1), and propidium iodide and analyzed and sorted on FACSAria. Samples from transplanted mice were analyzed on a FACSAria with antibodies for CD45.1 (A20), CD45.2 (104), NK1.1 (PK136), CD3 (17A2), CD19 (ID3), CD45R/B220 (RA3-6B2), CD43 (S7), and IgM (RMM-1). All antibodies were purchased from BioLegend unless stated otherwise. Gates were set according to fluorescence minus one (FMO) controls. The cellular composition in the transplanted Rag1^−/− mice was analyzed using frozen single-cell suspensions from thymus, BM, and spleen.

Total RNA was isolated from four Wt CD19⁺CD43^highIgM⁻ and four Pax5^+/−Ebf1^+/− pro–B samples using RNAeasy MicroKit (QIAGEN) according to the manufacturer’s recommendations. RNA was sent to UCLA Clinical Microarray Core for library preparation and subjected to 50 cycles of HiSeq 2000 SBS sequencing generating 20–30 million reads/sample. RNA sequencing of OP9-DL1 stimulated cells was performed using 50 cycles of Next Seq 500 sequencing generating 20–30 million reads from four independent experiments. Data analysis was performed with Arraystar (DNASTAR). For analysis of RNA-seq experiments the reads were aligned to mouse reference genome (mm9, NCBI 37) and RPKM normalized. Statistical analysis in Arraystar was performed with a Student’s t test with correction for Multiple Testing (Benjamini Hochberg). P < 0.05 was considered to be statistically significant.

For evaluation of B and T cell potential, cells were deposited (using FACSAria) directly into 96-well plates containing preplated (2,000 cells/well) stroma cells. T cell cultures (on OP9-DL1 stroma layers) were supplemented with 10 ng/ml KIT ligand, 10 ng/ml Fms-like tyrosine kinase 3 ligand (FLT3L), and 10 ng/ml IL-7. All cytokines were acquired from PeproTech. Cultures were substituted with fresh cytokines after 7 d. OptiMEM supplemented with 10% heat-inactivated fetal calf serum, 25 mM Hepes, 50 µg/ml Gentamicin, and 50 µM β-mercaptoethanol was used for maintaining the OP9/OP9-DL1 stroma cell-lines (Schmitt and Zuñiga-Pflucker, 2002), as well as for the cocultures. Cocultures were evaluated at day 14 by FACS staining with CD19 (ID3), CD90.2/Thy1.2 (53–2.1), CD3 (17A2) and 7-AAD using a FACSCantoII (BD).

Q-PCR analysis of sorted cells was performed as previously described (Mansson et al., 2008). Assays-on-Demand probes (Applied Biosystems) used were: Hprt; Mm00446968_m1, Pax5; Mm00515420_m1, Cd79α/Mb-1;Mm00432423_m1, CD79β/B29;Mm00434143_m1, CD19 Mm00515420_m1, CD27 Mm01185210_m1, PreTa Mm00478361_m1, Lck Mm00802897_m1, CD3e Mm00599683_m1, TCF7 Mm493445_m1, Gata3 Mm00484683_m1, Notch1 Mm00435245_m1, Hes1 Mm00468601_m1, Deltex1 Mm00492294_m1. Assay on demands, Lambda5 (Igll1) (Mansson et al., 2010) and Ebf1, Forward primer, 5′-TCATGTTTGGGATCCAGGAAAG-3′; Reverse primer, 5′-GTTGGATTTCCGCAGGTTAGA-3′.

Live cells were sorted and DNA was extracted and subjected to PCR based VDJ analysis as in (Schlissel et al., 1991). The Schlissel et al. (1991)–generated PCR products were transferred to a nylon membrane and hybridized to a P³²-labeled oligonucleotide acting as an internal J3 probe.

Cultivated CD43⁺IgM⁻ cells were spun down and fixed in 1 mg/ml DSG (Thermo Fisher Scientific) in PBS for 30 min at RT, followed by addition of up to 1% formaldehyde and an additional 10 min fixation at RT. The cross-link reaction was quenched with 0.125 M glycine, washed in PBS, and snap frozen or used immediately for ChIP.

Nuclei for ChIP were isolated by 10 min incubation in Nuclei Isolation buffer (50 mM Tris, pH 8.0, 60 mM KCl, and 0.5% NP-40) + protease inhibitor cocktail (PIC; Roche) on ice. Pelleted nuclei were dissolved in Lysis buffer (0.5% SDS, 10 mM EDTA, 0.5 mM EGTA, 50 mM Tris-HCl, pH 8.0) + PIC and sonicated on a Bioruptor (Diagenode) 18 cycles, max power for 30 s followed by 30 s of rest. Sonication was followed by pelleting of debris, and then the supernatant was transferred to new tube and chromatin was diluted 5X in Dilution Buffer (1% Triton, 2 mM EDTA, 150 mM NaCl, 20 mM Tris-HCl, pH 8.0 + PIC). 5 µg anti-Ebf1 (ABE1294), rabbit, poly, 1 mg/ml (lot Q2399134) was hybridized to 20 µl Dynabeads M-280 sheep anti–rabbit IgG (Life Technologies) in 100 µl PBS + PIC for 4 h, washed twice with PBS, resuspended in 20 µl PBS + PIC and added to the diluted chromatin. ChIP was performed over night at 4°C. and subsequently washed (1 time with 500 µl Low Salt Immune Complex Wash Buffer, 1 time with 200 µl High Salt Immune Complex Wash Buffer, 1 time with 200 µl LiCl Immune Complex Wash Buffer, and 2 times with 200 µl TE buffer) and eluted for 4 h at 65°C (20 mM Tris-HCl, pH 7.5, 5 mM EDTA, 50 mM NaCl, 1% SDS, 100 µg RNase A, and 50 µg proteinase K), treated, and cleaned up using Zymo ChIP DNA Clean & Concentrator before ChIP-seq libary preparation. 50-bp single-read sequencing was performed on an Illumina HiSeq 2000 at 23–39 million reads per sample.

Reads from high-throughput sequencing were aligned to a mouse reference genome (mm9, NCBI 37) using Bowtie (Langmead et al., 2009) with best match parameters (bowtie -m 1–sam–best–strata -v 2). Further analyses were made using the HOMER package (Heinz et al., 2010). Replicate Ebf1 ChIP-seq runs on each genotype (Wt, TH, Ebf1^+/−, and Pax5^+/−) and corresponding inputs were pooled into one dataset and analyzed as one combined sample per genotype.

Peaks were identified using findPeaks.pl with the –style factor parameter and normalized to sequencing inputs. Motif enrichment analysis was performed with the findMotifsGenome.pl command of the HOMER package using a 200-bp window.

Because cells in the different cell populations could be from male or female mice, for a more detailed comparison of Wt and TH samples, new peak-lists were constructed by filtering out peaks on Chr. X, Chr. Y, Chr. M, and Random Chr from original peak-files. Overlapping peaks between Wt and TH samples were identified using mergePeaks.pl (default parameters). Sequencing data are deposited at the Gene Expression Omnibus under accession no. GSE69227.

To find differentially bound peaks between Wt and TH, the modified Wt peakfile was annotated with the tag counts from all experiments using annotatePeaks.pl with –noadj. Next, the differentially bound peaks (P < 0.05 and P < 0.05 + fourfold F.C) were called with the getDiffExpression.pl with the –peak option. Tag density plots and heatmaps were created with annotatePeaks.pl (–hist or –hist & -ghist, respectively), normalizing data to 10 million mapped reads per experiment, and visualized using Excel and Java Treeview (Version 1.1.6r4; Saldanha, 2004).

To analyze peaks around a certain gene, a customized Perl script was used to identify peaks flanking each defined gene in a 100-kb window.

Venn diagrams were created using Venn Diagram (http://www.bioinformatics.lu/venn.php) with lists for closest gene from the HOMER-annotated peak file in comparison with gene lists from RNA-seq experiments.

In addition, Rag2^−/− pro–B Ebf1 and Pax5 ChIP-seq data (Revilla-I-Domingo et al., 2012; Vilagos et al., 2012) were obtained from the Gene Expression Omnibus (under accession nos. GSM932921, GSM932922, GSM932923, GSM932924, GSM876622, GSM876623, GSM876635, GSM876636, GSM876637, GSM876638, GSM876639, and GSM876640) and pooled into one dataset per ChIP and further analyzed using the HOMER package (Heinz et al., 2010) as described above. Annotated peak-lists from these datasets were used to overlay closest genes to peaks with lists from RNA-sequencing datasets. Overlapping Ebf1 and Pax5 peaks were identified with mergePeaks.pl (default parameters).

The intracellular Notch 1 level was determined by Western blot analysis. Pro–B cells were cultured on a monolayer for OP9DL1 stroma cells for 24 h and the stimulation was terminated by placing the cells on ice. The pro–B cells were removed carefully without disturbing the OP9DL1 stroma cells, centrifuged, and lysed in RIPA buffer supplemented with Protease inhibitor (Roche; 11697498001) and PMSF (1 µM; Sigma-Aldrich; P7626). Protein concentration was determined by Bradford reagent (Sigma-Aldrich; B6916), 50 µg of protein from each sample was boiled for 5 min in Laemmli buffer, and the proteins were separated on SDS-PAGE (456–8081; Bio-Rad Laboratories). The proteins were electrophoretically transferred to PVDF membranes (Immobulin FL; Millipore). Membranes were blocked with 5% BSA and probed with antibodies against intracellular Notch1 (Val1744; Cell Signaling Technology; 2421; 1:250 dilution), and equal loading was determined by probing blots with antibodies against GAPDH (ab9483; Abcam). Proteins were detected by IR dyes (Odyssey) and the membrane were scanned using Odyssey CLx Infrared Imaging System.

Retrovirus transduction was performed as in (Zandi et al., 2012) using either a MIGR1-GFP control virus or a virus where the open reading frame of Ebf1 was cloned upsteam of the IRES and GFP reporter gene. The transduced cells were sorted for GFP expression and incubated on OP9DL1 stroma cells before FACS analysis.

Table S1 shows an analysis of Ebf1 ChIP-seq data in Wt, TH (Ebf1^+/−Pax5^+/−), Ebf1^+/−, and Pax5^+/− pro–B cells.

We wish to thank our colleges for help with transgenic mice and cell lines. We also wish to thank Liselotte Lenner and Linda Bergström for advice and assistance.

This work was supported by grants from the Swedish Cancer Society, the Swedish Research Council, including a Center grant to Hematolinne, Knut and Alice Wallenbers Stiftelse, The Swedish Childhood cancer foundation, and Linköping University.

The authors declare no competing financial interests.

Author contributions: J. Ungerbäck, J. Åhsberg, R. Somasundaram, T. Strid, and M. Sigvardsson conducted, designed, and interpreted the analysis of the mouse model. All authors participated in the writing of the manuscript.