Lineage specification and plasticity in CD19⁻ early B cell precursors

To identify hematopoietic cells transiting from MLPs to the CD19⁺ pro–B stage, we adopted the approach pioneered by Muller-Sieburg et al. (24) and Spangrude et al. (25) using a mixture of lineage-specific staining reagents to deplete erythroid, myeloid, and T lineage cells from bone marrow. To simultaneously analyze the potential B lineage precursors, we added reagents specific for proteins that define stages of B cell development after CD19 expression, including B220, CD43, CD24, CD93, and IgM (13). We also included reagents recognizing CD117/cKIT and CD127/IL-7Rα, which have been used to identify the MLP and CLP fractions (10). Fig. 1 shows analysis with these reagents, eliminating cell doublets/aggregates by forward light scatter height/area gating; focusing on intermediate-size cells by forward/side light scatter gating; eliminating dead cells, highly autofluorescent cells, and T cells by CD3/PI gating; and finally eliminating cells expressing other non–B “lineage marker” proteins by gating with reagents specific for monocyte/macrophage/granulocyte/dendritic cells (CD11b, GR1, and Ly6c) and erythroid cells (Ter119).

Further gating based on a display of CD19 versus CD24/HSA allowed elimination of CD19⁺ (pro–B and later) B lineage cells, focusing on CD19⁻ cells, most of which expressed low levels of CD24. Based on previous findings that CD93/AA4.1 is expressed from hematopoietic stem cells through the immature B cell stage (26–28), we selected CD93^high cells, most of which expressed intermediate levels of CD43, similar to that found on CD93^hiCD43^medCD19⁺ pro–B cells (5). This cell fraction contained B220⁻ and B220⁺ subsets. Analyzing both subsets for expression of CD117 versus CD127 identified the following three cell populations: (a) B220⁻ cells, some with high levels of CD117 and lacking CD127, others with intermediate levels of CD117 and bearing CD127; and (b) B220⁺ cell, most bearing CD127 and many with intermediate levels of CD117. Based on the similarity of the B220⁻ fractions to very early multilineage precursors and common lymphoid progenitors described previously, we provisionally referred to these cells as MLPs and CLPs (10). We refer to the B220⁺ fraction as pre-pro–B “Fr. A” based on their expression of B220 and low-level expression of CD24. We excluded the infrequent and variable numbers of B220⁺ cells having undetectable levels of CD117 because such cells were not found in the B220⁻ (CLP) fraction and showed only lower level expression of a RAG-2-GFP reporter compared with CLP, Fr. A, and pro–B stage cells (not depicted).

We then used a “back-gating” analysis, examining the distribution of cell surface proteins used in delineating these subsets to assess whether we might have excluded significant portions of cells belonging to the CD117^hiCD127⁻ and CD117^medCD127⁺ cell populations, considered to identify the MLP and CLP stages, respectively (9, 10). Although all CD117^medCD127⁺ were included in the CD93^hiCD43^med gate, this analysis revealed that only a portion of the CD117^hiCD127⁻ cells were CD93^hiCD43^med, with many more exhibiting a CD93^medCD43^hi phenotype. Most cells in the CD93^medCD43^hi population belong to the LIN⁻ Sca1⁺cKit⁺ (LSK) fraction of very early hematopoietic cell precursors, including stem cells (Fig. S1). Thus, it seems likely that the earliest stages of hematopoietic cell differentiation, such as LSK, have lower levels of CD93 (and higher levels of CD43) than the MLP, CLP, and Fr. A stage cells that are the focus of this work. As Fig. S2 shows, most LIN⁻CD19⁻ cells expressing very high levels of a RAG-2 GFP reporter (see next section) fall within the CD117^medCD127⁺ gate and these cells are predominantly CD43^med and CD24^low (i.e., are CLP and Fr. A as we define them). We conclude that using the gates in Fig. 1 encompasses essentially all early B lineage precursors.

In a second gating approach with the same stained sample, we focused on cell surface proteins whose expression identifies Fr. A through Fr. C′, focusing on B220⁺CD43^med cells, subdividing them on the basis of CD24 and CD19. Among these cells, most of those lacking CD19 have a distinctively low to undetectable level of CD24 (Fig. 1, right side). Furthermore, in contrast to the relatively unimodal high-level CD93 expression found with CD19⁺ stage cells, there is considerable heterogeneity in the CD24^low fraction. Again, based on previous work showing that CD93⁻ cells with this phenotype do not belong to the B lineage (5), we excluded CD93⁻ cells and identified a set of CD93⁺CD117^medCD127⁺ cells, corresponding to the pre-pro–B population (Fr. A) identified in the first gating analysis described above. We also distinguish two subsets of CD43⁺CD19⁺ stage cells, with intermediate and high levels of CD24, respectively. Both of these fractions, corresponding to stages termed pro–B (Fr. B and Fr. C) and early pre–B (Fr. C′; reference 13), express homogenous high levels of CD93 and are CD127⁺, but have variable (in B and C) to undetectable (in C′) levels of CD117. Table I summarizes the surface markers used in identifying the CD19⁻ fractions and CD19⁺ pro–B fractions examined in this work.

Next, we tested the capacity of cells in these cell fractions to engraft B, T, NK, and myeloid cells using a standard Ly5-congenic competition assay (Fig. 2). Thus, we injected Ly5.2 sorted cells mixed with a constant amount of Ly5.1 unfractionated bone marrow i.v. into lethally irradiated Ly5.1 recipient mice and examined the Ly5.1/Ly5.2 chimerism in different hematopoietic cell lineages 3 wk after transfer. This approach shows how a hematopoietic progenitor's fate is read out in the context of the whole organism environment.

Fig. 2 A shows the percent of each specific cell type contributed by the purified precursor population in various tissues, demonstrating robust multilineage reconstitution by the CD117^hiCD127⁻ (MLP) fraction, where we obtained large numbers of Ly5.2⁺ myeloid cells and macrophages in bone marrow, T lineage cells in thymus and spleen, and B cells in spleen. We also detect excellent engraftment of DX5⁺ NK cells. T cell reconstituting capacity at 3 wk after transfer was most clearly revealed by analysis of thymus engraftment, the primary site of T cell development. Therefore, Fig. 2 B presents the absolute numbers of myeloid, T, and B cells generated in bone marrow, thymus, and spleen, setting the values obtained with the MLP fraction to 1.0 and expressing the others relative to it. The greater numbers engrafted by MLP stage cells likely represent their capacity to proliferate significantly as precursors, before differentiating into the various hematopoietic cell types analyzed here. Although fewer cells were produced with transfers of CD117^medCD127⁺B220⁻ (CLP) fraction cells, we nonetheless observed a clear lymphoid-restricted repopulation of T, B, and NK cells (and few myeloid cells). Transfer of CD117^medCD127⁺B220⁺ (Fr. A) cells engrafted far fewer T cells (evident from analysis of thymus; Fig. 2 B) and few NK cells, and still generated a significant population of B cells, suggesting a clear delineation of Fr. A pre-pro–B cells. As expected, the CD19⁺ Fr. B subset produced the clearest B cell–restricted repopulation. In summary, cell transfer data support the designation of these stages as MLP, CLP, and Fr. A.

Next, we asked whether all CLP stage cells are lymphoid committed. Although the cell transfer analysis appeared to indicate such restriction, this assay suffers from at least two deficits: (a) lineage plasticity may be masked by failure of cells to migrate into inducing microenvironments for all potential alternate cell lineages; and (b) the seeding efficiency may vary among different cell subsets, making it difficult to estimate the frequency and homogeneity of subsets under analysis. Therefore, we assessed lineage plasticity of cells in these early stages by a series of clonal (or near-clonal) in vitro assays. First, we investigated B/myeloid potential using a bipotential assay (29), depositing individual cells (1 cell/well) into a 96-well plate containing a preestablished S17 stromal cell monolayer and medium supplemented with SCF, Flt3 ligand, and IL-7 (Fig. 3 A). This combination of cytokines and stromal cells supports the growth and differentiation of most early hematopoietic cells (30) and, under these conditions, both myeloid and B lymphoid cells can develop clonally with high efficiency. Most clones developing from MLPs were myeloid (CD11b⁺/CD19⁻) with occasional B lineage colonies. CLP stage cells generated more B lineage colonies, but we also observed a significant number of myeloid colonies. In contrast, plates sorted with the B220⁺ Fr. A and the CD19⁺ pro–B Fr. B and C contained very few myeloid colonies. Thus, although CLPs generated numerous myeloid colonies, few were seen in plates containing sorted Fr. A cells.

One possible explanation for our detection of myeloid potential from CLPs is heterogeneity within the cell fraction we identified, with only a portion being “true” CLPs. To address this, we examined the expression of a RAG-2 fluorescent reporter BAC transgene (31) in these early B lineage fractions. Expression of recombinase activating genes has been considered one of the hallmarks of lymphoid specification (17). As Fig. 4 A shows, we found that all RAG⁺ cells are CD93⁺. Furthermore, using the high-level CD93 gate we use (see Fig. 1), essentially all MLP cells express low but detectable levels of GFP and some express higher levels (Fig. 4 B). Cells in the CLP fraction show a 10-fold higher level than MLP, and those in Fr. A express a yet higher level. Although most cells in the CLP fraction show very high levels of RAG-2–GFP, there are some cells with a lower level, similar to MLP, so we fractionated cells based on levels of RAG-2–GFP to determine whether enriching for the RAG-2/ GFP highest-expressing fraction would eliminate the myeloid colonies for MLP or CLP cultures. This was not the case, as we found many myeloid colonies with MLP fraction cells expressing the highest levels of RAG-2/GFP and also obtained significant numbers of myeloid colonies with RAG-2/ GFP-gated CLPs (Fig. 3 B). Thus, the myeloid capacity in CLP stage cells detected by the bipotential culture assay does not reflect a heterogeneity in the cell population that can be fractionated based on RAG expression. Rather, appearance of surface B220 expression on CD117^medCD127⁺ cells, recognized as Fr. A, represents a clearer indicator of the loss of myeloid potential as B cell development progresses.

Having observed that CLP has myeloid potential, the question arises whether a more clearly B/T-restricted “CLP” stage could be identified. To examine T lineage potential, we undertook fetal thymic organ culture (FTOC) using the high oxygen submersion modification developed by Dou et al. (32) that facilitates clonal generation of T cells. We compared the capacity of the CLP, Fr. A, and pro–B stage cells (3 cells/well) to produce T lineage cells in alymphoid lobes isolated from RAG-2/common γ chain double-deficient embryos that lack endogenous lymphoid/NK lineage development (Fig. 5 A). It was striking that equal numbers of wells containing early T lineage cells (∼60%) arose from both CLP and Fr. A. And, as would be expected, none were generated using CD19⁺ pro–B cells. These early T lineage cells, generated under conditions of high levels of IL-7, typically showed a very immature phenotype, expressing CD90 and CD25, but lacking CD3, CD4, or CD8. They were not stained by reagents specific for B lineage (CD19), myeloid lineage (CD11b and GR1), or NK lineage (NK1.1 and DX5) cells.

Although our FTOC data indicated a significant capacity for T cell production from Fr. A, comparable to the CLP fraction, limitations on the number of thymic lobes required a 3 cell/well assay to generate significant numbers of engrafted lobes. Therefore, we undertook a single cell assay using the recently described Delta-like 1 (DL1)-OP9 system (33). Engagement of Notch-1 by DL1 delivers a critical signal specifying the T cell fate in developing hematopoietic cells (34, 35), and Schmitt et al. (33) found that DL1-transduced OP9 stromal cells could induce T cell development from early precursors in vitro. The OP9 cell line also supports efficient development of B lineage cells under these culture conditions and thus a T/B bipotential assay is possible using this approach. We found that the assay works well at the single cell/well level and compared the capacity of CLP and Fr. A to produce T and B lineage cells. As shown in Fig. 5 B, similar to the FTOC assay, we observe efficient T lineage generation from both CLP and Fr. A, but with a bias toward B lineage development apparent in Fr. A (Fig. 5 B). Interestingly, in our analysis, T cell generation occurs at the expense of B cell generation, with similar numbers of total clones in both control and DL1 cultures. Thus, under conditions favorable for generating B cells, Fr. A can generate T cells if a Notch-1 signal is provided.

While our RAG-2 reporter analysis (Fig. 4 B) indicated that RAG transcription is already activated in nearly 100% of MLP stage cells, it was unclear when D_HJ_H rearrangement initiated because chromatin configuration at the heavy chain locus will play a key role in determining heavy chain locus accessibility. Therefore, we then focused on this issue. To determine the frequency of cells with Ig heavy chain locus recombination we used a single cell DNA PCR assay (7, 36, 37). This approach uses two rounds of amplification with nested primers, one set that amplifies a germline band and another set that detects many of the possible D_HJ_H rearrangements, allowing us to assess the extent of heavy chain rearrangement at the single cell level. Importantly, this assay was very efficient, recovering signals from 80% of the cells analyzed (Table II). The earliest fraction, MLP, showed germline bands with little detectable D_HJ_H rearrangement (<2%). Yet, we detected D_HJ_H rearrangement in 48% of signal-positive CLP stage cells, and rearrangement increased to 88% of Fr. A cells (Table II, see values in parenthesis). Thus, even cells with extensive DJ rearrangement retain the capacity to develop into T cells. In comparison, CD19⁺ pro–B stage cells had extensive D_HJ_H bands with essentially no germline signal, indicating that most cells had D_HJ_H rearrangements on both chromosomes, consistent with a previous analysis using a germline-loss bulk PCR assay (13).

To more completely understand the progressive restriction of lineage potential in each cell fraction at these developmental stages, we next performed quantitative analysis of gene expression using a TaqMan PCR assay for a set of genes previously used by us and others to characterize early stages in B cell development, in contrast with that of myeloid and T lineage cells (Fig. 6). A very clear finding was the presence of mRNA for genes involved in Ig recombination earlier than conventionally expected. TdT, RAG-1, and RAG-2 were all found at significant levels in CLP stage cells, well before surrogate light chain message (Fig. 6 A). The activation of these key components of the recombinase system explains the observation of extensive D-J rearrangement in CLP stage cells shown above. TdT, previously suggested as a marker for B lineage– (15, 38) or lymphoid-restricted precursors (16), is expressed at significant levels even before the CLP stage. MLP stage cells show ∼10–15% of the peak levels detected in CLP and Fr. A (Fig. 6, A and B). As we noted earlier (5), Ig-β mRNA (B29) is expressed early in the B lineage pathway, before detectable Ig-α mRNA (MB1), and we confirm this in our quantitative analysis.

Several transcription factors act to regulate the B lineage gene program (2) and several key players are shown in Fig. 6 C. Interestingly, mRNA for the early acting and critical E2A proteins E12 and E47, diagnostic of B lineage potential (39), are well above background in all fractions tested (Fig. 6 C), suggesting that the MLP fraction identified by high levels of CD93 and characterized by significant RAG-2–GFP expression is already being induced along the B lineage pathway. In contrast, two other key B lineage transcription factors, EBF (40, 41) and Pax-5 (42–45), showed later induction in the progression from CLP (B220⁻) to Fr. A (B220⁺). PU.1, a transcription factor regulating B/myeloid potential by activating growth factor receptors at different levels of expression (46), decreases from MLP to CLP, and then decreases further from Fr. A to Fr. B. This is consistent with a decrease in myeloid potential by loss of myeloid growth factor receptors and induction of lymphoid potential by expression of the IL-7R during the course of this progression.

We also surveyed expression of genes representative of alternative lineage fates, including T lineage, GATA3 (47) and Notch-1 (33, 34), and myeloid lineage, Csf1r (48) and C/EBPα (49). These genes showed very significant decreases in the progression from CLP to Fr. A to the CD19⁺ pro–B stage (Fig. 6 D). In fact, the expression of these genes was reciprocal to EBF and Pax-5, consistent with a progressive restriction to the B lineage fate as cells pass through these three stages. Consistent with the myeloid potential in CLPs, there is little change in Csf1r mRNA levels from MLP to CLP, whereas the level of GATA3 rises sharply in CLPs, possibly indicating the activation of a T lineage program that is then extinguished as cells progress to express B220 and then CD19 in the absence of Notch-1 signaling. Significantly, Notch-1 expression is sharply up-regulated at the CLP stage, declining thereafter, providing a mechanism for T lineage specification via Notch-1 signaling in CLP and Fr. A stage cells.

Finally, we analyzed RNA prepared from these four early B lineage stages from mouse bone marrow to identify sets of genes that were coordinately regulated as cells progressed down the B cell development pathway. We generated ratios of signal from amplified RNA to a common reference RNA for two samples per fraction. Genes showing a statistically significant difference in at least one stage were identified by ANOVA, and the resulting set (∼1,000 genes) was analyzed by KMeans clustering. The results obtained were visualized by a “heat map” display (Fig. 7) where individual gene levels are coded green for low, black for intermediate, and red for high. Using this approach we can identify a set of genes that show low-level expression in MLP and pro–B stages, but higher expression in CLP and Fr. A (cluster A); another that is low in MLP and CLP, but increasing in Fr. A and Fr. B and C; and yet others where high-level expression found early diminishes in later fractions. Importantly, we could identify genes present in each cluster that were consistent with the patterns of expression shown in Fig. 6. Thus, cluster A includes the IL-7Rα gene, consistent with the highest staining for this protein on CLP and Fr. A. Cluster B contains well-known B lineage–associated transcription factors (Pou2af1/OCAB, EBF-1, PBX-1, LEF1, IRF4, and SPI-B) and a large number of lymphoid/B lineage–associated genes, including RAG-1, Blnk, CD19, CD79a, Cd79b, and VpreB. Cluster C is a group of genes sharply up-regulated at the pro–B stage and includes ABL-1. Cluster D includes cKit and one of the myeloid colony-stimulating factor receptor genes. Cluster E contains Csf1, the myeloid growth factor receptor gene that we analyzed by quantitative PCR and found was present in CLP and down-regulated in Fr. A (the pattern shown here). Cluster F includes Notch1 and GATA 3, consistent with the T lineage potential of Fr. A that is lost in Fr. B and C. In summary, the patterns of expression that we observe are consistent with our quantitative PCR determination for every gene examined that is contained in this set of differentially expressed genes, and this analysis identifies a large number of additional genes.

Here we identify and characterize very early stages in B cell development in mouse bone marrow, focusing on those before CD19 expression. Fig. 8 summarizes the features of the cell stages we have defined. The earliest stage, MLP, generates all lineages assayed in vivo, including B, T, NK, and myeloid. The next stage, identified by cell surface markers as CLP, gives a lymphoid-restricted repopulation in vivo but has the potential to generate myeloid cells in culture, while half of the cells already contain D_HJ_H rearrangements. The third stage, Fr. A, identified by expression of B220, generates predominantly B cells in vivo but has the potential to produce T cells in culture, while showing an even higher frequency of D_HJ_H rearrangement than CLP. Gene expression profiles are consistent with B lineage activation at the MLP stage, resulting in expression of a large number of B lineage–specific genes at the CLP and Fr. A stages.

T lineage potential in Fr. A was unexpected. Previously we showed that a fraction of B220⁺CD19⁻ cells expressing CD43 and CD93 (“Fr. A2”) were B lineage precursors that failed to generate T cells in i.v. and intrathymic assays (7). Using the present approach, applying greater purification criteria than previously, we still find predominantly B lineage repopulation by a “refined Fr. A” using the standard Ly-5 marked competitive repopulation assay. Although these B220⁺ CD19⁻ Fr. A cells constitute only 0.05–0.1% of total bone marrow cells, they constitute an important decision point in B cell development. However, while behaving as B lineage precursors in some assays, their lineage plasticity is revealed in the HOS-FTOC and DL1-OP9 assays. One explanation for this behavior is their expression of Notch-1, which can be activated in the thymic microenvironment or by DL1 interaction in culture, redirecting their lineage fate. The contrasting i.v. repopulation results likely indicate that Fr. A stage cells only inefficiently home to the thymus or else rapidly progress to the irreversibly B lineage–committed CD19⁺ stage compared with earlier stages like CLP or MLP.

Myeloid cell generation from the CLP fraction was also surprising. Our analysis shows that although these cells have become considerably more lymphoid specified than MLP, expressing higher levels of TdT and RAG message (and functional RAG protein, as indicated by DHJH rearrangement), they nevertheless retain significant myeloid capacity as revealed in the B/myeloid bipotential assay. As with Fr. A, the i.v. repopulation assay shows a more restricted lineage potential, either because these cells predominantly home to microenvironmental niches that disfavor myeloid development or else rapidly progress to Fr. A in vivo. We can only speculate that this in vitro myeloid potential was overlooked previously due to differences in stromal culture conditions that favored lymphoid progression or else to differences in sensitivity of detection of CD45R/B220 that might have included Fr. A in the CLP population, increasing its apparent lymphoid restriction. Therefore, the subdivision of B220⁻ and B220⁺ fractions of CD117^medCD127⁺ cells is important because it reveals clear differences in lineage restriction read out in both in vivo and in vitro. That is, in contrast with CLP, most Fr. A cells did not respond to myeloid-inducing signals in stromal cell culture. Thus, Fr. A pre-pro–B cells behave as a strict “common lymphoid progenitor” in terms of their lineage potential.

The finding of B220⁺ CLP is reminiscent of a previous report from Martin et al. (50) describing a lymphoid-restricted “CLP2” cell type. However, in their studies, the B220⁺ CLP-like cells were reported to lack CD117, clearly not the case with Fr. A. Furthermore, the CLP2 cells were described functionally as efficiently homing to the thymus, which we do not observe; Fr. A cells injected i.v. generate far fewer thymocytes compared with classical (i.e., B220⁻) CLP stage cells (Fig. 2 B). The capacity of CLP2 stage cells to home to the thymus led Martin et al. to propose that these cells are a founder population for thymic T cell development. In contrast, we would suggest that Fr. A stage cells are an intermediate between CLP and CD19⁺ pro–B stage cells, in a B lineage–specified, but not yet committed, state. Recent work from Allman et al. (51) has identified an early thymic progenitor that appears distinct from CLP2. Clearly, this issue will require further investigation to clarify the relationships among B220⁺CD19⁻ subsets from bone marrow and thymus in terms of lineage potentials.

A very recent report by Balciunaite et al. (21) described the presence of a B220⁺CD117⁺CD19⁻ hematopoietic progenitor with B, T, and myeloid potential in vitro and lymphoid potential in vivo. Although these authors' limiting dilution analysis suggests a precursor with some similarity to ours, their population appears more similar to the B220⁻ CLP stage we report here; the expression of B220 in our experience greatly reduces myeloid potential (by 10-fold) found in CLPs. Furthermore, the decrease in csfR1, a gene encoding the receptor for colony stimulating factor 1 (a key myeloid growth factor), in the progression from CLP to Fr. A provides a mechanistic explanation for the difference we observe. In fact, we found that B220 expression was a better marker for the loss of myeloid potential than induction of a RAG-2–GFP reporter, sounding a cautionary note on the use of RAG reporters for identifying “early lymphoid progenitors” (17).

Both CLP and Fr. A stage cells generated T lineage cells in vitro, revealing T lineage potential, but this does not necessarily mean that either are normal intermediates in a developmental pathway from hematopoietic stem cells to T cells. Although the existence of T cell lines or thymocytes with Ig D_HJ_H rearrangement have been reported (52), there are no T lineage precursors among triple negative (CD3⁻4⁻8⁻) thymocytes with a surface phenotype corresponding to bone marrow CLP (51), and we detect no cells corresponding to Fr. A in the thymus (not depicted). Our results seem most consistent with a type of developmental model where B lineage “specification” precedes B lineage commitment in bone marrow (53–55). In this model, B lineage genes are induced and non–B lineage genes are repressed in progressive stages of hematopoietic development, mediated by a hierarchy of transcription factors, resulting in a B lineage–specified stage, coincident with D_HJ_H rearrangement and the high-level expression of a set of early B lineage genes, such as Igα/β, λ5/VpreB, and RAG-1/2. However, absolute irreversible lineage commitment occurs at a later stage, coincident with high-level expression of functional Pax-5 (42, 43).

In fact, it appears that B lineage specification initiates even before the CLP stage, as MLP cells express some lymphoid/B lineage genes, including TdT, RAG-2, and Ig-β. We also note that most MLP stage cells show activation of a RAG reporter transgene, similar to the previously described CD117^hi early lymphoid progenitors fraction that lacks CD127 (17). Nevertheless, we found a robust myeloid lineage engraftment with MLP, suggesting that RAG-2 gene transcriptional activation, along with a significant component of the B lineage program, initiates in a cell fraction that maintains considerable myeloid potential as revealed using the cell engraftment competition assay.

Finally, our microarray analysis illustrates the progressive nature of B cell development. We identify clusters of genes with expression shared between MLP and CLP, between CLP and Fr. A, and between Fr. A and CD19⁺ pro–B stage cells. Continuing examination of the members of these clusters will help to elucidate more fully the gene program resulting in progressive restriction to B lineage development, along with the key microenvironmental interactions that foster this process.

6–12-wk-old BALB/c (ICR) female mice, bred in our animal facility, were used in most experiments. For the i.v. competition cell transfer assay, B6.Ly5.2 female mice were obtained from the National Cancer Institute. RAG-2–GFP BAC transgenic mice were obtained from M. Nussenzweig (Rockefeller University, New York, NY) and bred in our animal facility. RAG-2/common γ chain double-deficient (RAG2°γC°) a-lymphoid animals, originally described by Colucci et al. (56), were obtained from D. Wiest at our institute, and timed matings for generating fetal thymic lobes were performed in our animal facility. C57BL/6 mice were obtained from our animal facility production colony. All experiments with mice were conducted under an approved animal protocol.

The selected population was sorted from B6.Ly5.2 bone marrow and the yield derived from two bone marrow equivalents was transferred i.v. per recipient, typically 5–10 × 10³ for CD19⁻ fractions and 5–10 × 10⁴ for CD19⁺ cells, together with 10⁵ unfractionated Ly5.1 (recipient type) bone marrow. 2-mo-old C57/B6 female recipient mice were lethally irradiated (9 Gy) 1 d before transfer and provided neomycin polymixin B antibiotic in water. Animals were analyzed 3 wk after transfer, using antibodies specific for Ly5.1 and Ly5.2 in addition to reagents specific for T, B, NK, and myeloid cells (CD3, CD4, and CD8 for T cells; B220, CD19, and IgM for B cells; NK1.1 and DX5 for NK cells; and CD11b and GR-1 for myeloid cells/granulocytes).

S17 stromal cells were grown in a 37°C humidified, 10% CO₂ gassed incubator in 5% FBS/RPMI 1640 medium (supplemented with 1× glutamine, 10 mM Hepes, 5 × 10⁻⁵ M 2-ME, and 0.5 mg/ml gentamicin). Cells were passaged into 96-well plates in the same medium and allowed to reach confluence 3–5 d before use. Immediately before cloning, medium was replaced with fresh medium containing cytokines (10 ng/ml SCF, 10 ng/ml Flt3L, and 100 U/ml IL-7; R&D Systems). Individual cells of the selected population were sorted using the single cell deposition unit of a FACSVantage/DiVa flow cytometer and then returned to the incubator. 7–10 d later wells were examined for colonies using an inverted microscope. Positive wells were harvested by pipetting, and then cells were stained with antibody to CD19 and CD11b and analyzed by flow cytometry.

The modified FTOC culture under high oxygen tension was performed as described previously (32, 57), except that fetal thymic lobes from RAG2°γ_c° embryos were used. Three cells of selected population were sorted per well using the automated cell deposition unit. Cells were analyzed after 14 d by flow cytometry, staining for CD19, CD90/Thy-1, CD25/IL-2Rα, DX5, CD11b, and CD11c. Early T lineage cells were recognized as CD90⁺CD25⁺ cells, lacking other markers tested. Alternately, individual cells selected by sorting were deposited into microplate wells containing preestablished stromal cells, either DL1-transduced OP9 or GFP-OP9 (33). Cultures were performed essentially as described previously (33) and analyzed for generation of T or B cells using the staining procedure described above at 7–10 d.

Sorting was performed using a BD Biosciences FACSVantageSE/DiVa, equipped with three laser excitation lines (407, 488, and 600 nm) for 12-color detection. Early B lineage precursors were isolated using the following staining combination: FL-Ter119, FL–anti-Ly6c, PE-anti–IL-7Rα (SB/199), PI detected in TR-PE channel, Cy5PE–anti-CD3 (500A-A2), Cy55PE-CD93 (AA4.1), Cy7PE-CD43 (S7), Alexa 594–anti-CD24/HSA (30F1), APC-CD117/cKit (2B8), Cy55APC–anti-CD19 (1D3), Cy7APC–anti-CD11b/Mac-1(M1/70), Cy7APC-GR1, CasBlue–anti-IgM (331.12), and Bi–anti-CD45R/B220 (RA3-6B2). Biotin reagent was revealed by second-step incubation with QDot605-streptavidin. Analysis was performed using either this flow cytometer or a BD Biosciences LSR-II with three lasers (407, 488, and 630 nm) equipped for 10-color detection. All reagents were made in our laboratory, except for QDot605-streptavidin, which was purchased from Quantum Dot Corporation, and DX5 from BD Biosciences. Cells from RAG-2–GFP reporter mice were analyzed by detecting GFP in the FL channel, using the staining combination described above, except that Ter119 and anti-Ly6c antibody were labeled with Cy5PE.

Analysis was performed using a modification of procedures described previously (7, 36, 37). Cells were sorted directly into 96-well plates (Applied Biosystems) containing 20 μL/well lysis buffer (1× PCR buffer [Applied Biosystems] with 2.5 mM MgCl₂, 9.2 μg/ml tRNA [Sigma-Aldrich], and 100 μg/ml gelatin). Plates were stored at −80°C and then just before PCR, plates were thawed, treated with 0.5 mg/ml proteinase K for 1 h at 55°C, and then heated for 10 min at 95°C. After digestion, a two-round nested PCR was used to detect a germline DNA segment (lost upon heavy chain rearrangement) and potential D_HJ_H rearrangements. The PCR program was 95°C for 1 min, 63°C for 1 min, and 72°C for 1.5 min for 30 cycles, with a 10-min end extension at 72°C. For round 1: GL5-1, CCCGGACAGAGCAGGCAGGTGG; D_H5-1, ACAAGCTTCAAAGCACAATGCCTGGCT; and D_H3-1, AGGCTCTGAGATCCCTAGACAG. For round 2, the following two separate reactions were performed: for germline GL5-2, GAGTTGACTGAGAGGACAG, and GL3-2, CGAAGTACCAGTAGCAC; and for D_HJ_H rearrangements D_H5-2, ACGTCGACTTTT(G/C)TCAAGGGATCTACTACTGT, and D_H3-2, GGGTCTAGACTCTCAGCCGGCTCCCTCAGGG. In the first round, PCR amplification was performed with the contents of each well in a total volume of 50 μl containing 0.2 μM dNTPs, 1 μl BD Advantage cDNA polymerase mix, 1× BD Advantage PCR buffer, 0.5 mg/ml BSA, and 0.4 μM of each primer. In the second round, 1 μl of each first round product was added to a 50-μl reaction volume using the same conditions described above, except that second round primers (GL-5-2/GL-3-2 or D_H-5-2/D_H-3-2) were used at 2 μM. Products were visualized on 1.5% agarose gels stained with ethidium bromide. The validity of these PCR products was verified by sequence analysis.

Total RNA was prepared by sorting cells into Solution D lysis/denaturing solution, followed by acid-phenol extraction and isopropanol precipitation, as described previously (5). cDNA was synthesized by adding 1 μl oligo-(dT)_12–18 primer (0.5 μg/μL; Invitrogen) to 20 μl total RNA, heating at 70°C for 10 min, cooling on ice for 2 min, adding 8 μl 5× first-strand buffer (Invitrogen), 4 μl 0.1 M DTT (Invitrogen), 4 μl dNTPs (each dNTP at 10 mM; Promega), 1 μl random hexamer primers (20 U/ml; GE Healthcare), 2 μl RNAsin (40 U/ml; Promega), and 2 μl Superscript II (200 U/ml; Invitrogen), and then incubating at 42°C for 2 h. Gene expression was quantitated by real-time PCR. Analyses were performed in triplicate in 25-μl volumes using an ABI7500 thermal cycler. For each tube, 12.5 μl ABI TaqMan 2× Mastermix (polymerase and dNTPs), 1.25 μl probe mix (ABI), 9.25 μl DEPC-H₂0, and 2 μl template (typically diluted 1:3 from cDNA synthesis volume) were added. ABI software was used to quantify/calculate Ct values and determine relative gene expression levels, standardizing using β-actin values. All quantitative PCR ABI assay IDs and sequences for custom-designed sets are available upon request.

Cells were sorted directly into RNA lysis buffer (6 M guanidine thiocyanate, 0.67% Na N-lauroylsarcosine, 33 mM sodium citrate, and 133 mM 2-mercaptoethanol), and total RNA was extracted using the acid phenol method as described previously (5). Integrity and quantity of total RNA samples were analyzed using a 2100 Bioanalyzer (Agilent Technologies). 40 ng total RNA was used for RNA amplification. RNA amplification was performed using the Ovation Aminoallyl RNA Amplification and Labeling System (NuGEN Technologies, Inc.) in accordance with the manufacturer's protocols. 2 μg amplified cDNA was used for dye-coupling with Alexa Fluor 555 and Alexa Fluor 647 (Invitrogen). Quantification of the fluorescent-labeled probes was performed using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies). 50 pmol of fluorescent-labeled cDNA (∼1.5 μg amplified cDNA) of each probe (experimental sample and reference) was loaded per slide. For use as a reference, total RNA from six cell lines (R1, RS2, J558, EL4, DN3, and P388d) was prepared using TriReagent (Molecular Research Center, Inc.) and 30 μg (5 μg from each cell line) was reverse transcribed at 42°C for 2 h in a total volume of 20 μl that included 1 μl Superscript II reverse transcriptase (200 U/μl; Invitrogen), 1 μl rRNAsin RNAase inhibitor (40 U/μl; Promega), and 2 μl of a mixture of dNTPs (5 mM each of dGTP, dATP, and dCTP; 2 mM of dTTP; and 3 mM of aminoallyl dUTP). RNA was hydrolyzed by the addition of 10 μl 1 M NaOH and incubation at 70°C for 10 min. 10 μl of 1 M HCl was then added to neutralize the sample, and cDNA was precipitated overnight by the addition of 4 μl of 3 M sodium acetate, pH 4.5, 1 μl glycogen (20 μg/μl), and 100 μl ethanol. cDNA was pelleted by centrifugation, washed once with 70% ethanol, and dissolved in 9 μl coupling buffer, followed by the same dye-coupling procedure described above for amplified cDNA.

Two samples were analyzed from each fraction sorted from separate pools of mouse bone marrow cells. RNA prepared from each sample was amplified, used for probe generation, and hybridized with a common reference RNA. Amplified RNA was labeled with both Alexa Flour 555 and Alexa Flour 647 and hybridized with the complementary labeled reference RNA (dye-flip replicates). Whole mouse genome 44K oligo microarray kits (Agilent Technologies) were used for hybridization. The hybridization and SSPE washing and drying procedures were all performed according to the manufacturer's recommendations. The slides were then scanned using an Agilent BA DNA Microarray Scanner. Data from scans were normalized using Agilent feature extraction software and then subject to statistical analysis using GeneSight software (BioDiscovery). Results were determined as ratios of experimental sample to reference, and dye-flip replicates were combined. The resulting data, two ratios for each gene from all four samples, were analyzed by ANOVA, selecting genes differentially expressed reproducibly in at least one stage with a p-value cutoff of <0.005. This set of ∼1,000 genes was then analyzed by KMeans clustering using a distance measure based on the Pearson correlation and assuming six clusters.

Fig. S1 is a flow cytometry analysis showing the relationship between CD43 and CD93 expression with B cell development, comparing very early precursors (LSK) with CLP and Fr. A. Fig. S2 is a flow cytometry analysis showing the expression of a RAG-2–GFP reporter BAC transgene in early stages of B cell development. The microarray data used to generate the cluster analysis shown in Fig. 7 is also available. The microarray data is available from ArrayExpress, European Bioinformatics Institute (http:www.ebi.ac.uk/arrayexpress), as accession no. E-MEXP-559.

We thank J. Oesterling for technical assistance with flow cytometry. We appreciate the advice and help of Y.-S. Li with the microarray analysis. The RAG-2–GFP BAC transgenic reporter mice were a kind gift of M. Nussenzweig (Rockefeller University). We thank J.C. Zúñiga-Pflücker (University of Toronto) for OP9 stromal cells expressing DL1. We acknowledge the help of Y. Katsura and H. Kawamoto (Kyoto University) for sharing a detailed protocol for the HOS-FTOC assay. We thank P. Kincade (Oklahoma Medical Research Foundation) for providing hybridoma-secreting antibodies to CD117 and CD127. We appreciate helpful comments and suggestions on the manuscript from K. Hayakawa, K. Campbell, D. Kappes, and D. Wiest, all at our institute.

This work was supported by grants from the National Institutes of Health (NIH) to R.R. Hardy (AI26782 and AI40946). L.L. Rumfelt and B.M. Rowley were supported by NIH training grant AI07492.

The authors have no conflicting financial interests.