Stem cell–like reprogramming is required for leukemia-initiating activity in B-ALL

Cell quiescence and self-renewal activity are distinctive characteristics of normal stem cells that are lost throughout the differentiation process. Nevertheless, it is established that molecular reprogramming occurring in cancer cells frequently leads to tumor dedifferentiation and the acquisition of stemness features (Bradner et al., 2017). This notion has been exploited to predict the clinical outcome of patients with solid cancers and hematological malignancies (Malta et al., 2018; Ng et al., 2016; Yan et al., 2020) and supports the idea that cellular plasticity, reprogramming, and cancer resistance are tightly intertwined. Among hematological malignancies, B cell acute lymphoblastic leukemia (B-ALL) is defined as the most frequent pediatric cancer. Although current chemotherapy is efficient at reducing the tumor load by targeting proliferating and metabolically active leukemic cells, the disease relapse points to the presence of resistant cells that escape treatment (Inaba and Mullighan, 2020). Thus, the biological properties of the cell-of-origin of leukemia, including sustained self-renewal activity, cell quiescence, and drug resistance, can significantly affect leukemia treatment and should be considered in the search for new targeted therapies (Fregona et al., 2021).

The evolution and the dynamic of subclones in B-ALL cells have been fully explored in several works through the comparison of the genetic landscape of paired diagnostic and relapse samples (Dobson et al., 2020; Mullighan et al., 2008; van Delft et al., 2011; Waanders et al., 2020). These studies predicted the existence of an ancestral clone of preleukemic stem cells (pre-LSCs) harboring a restricted number of genetic alterations, such as a founding chromosomal translocation, which is not yet transformed but at the apex of the clonal hierarchy. The notion of a multistep process of B-ALL development has also been gained from studies exploring leukemia initiation and progression in monochorionic twins (Greaves, 2018; Hong et al., 2008). These approaches demonstrated that a primary chromosomal translocation by itself does not have the capacity to induce the disease but establishes a preleukemic subclonal compartment in blood cells many years before the leukemia onset. Recurrent chromosomal translocations have been described in B-ALL and lead to the expression of chimeric fusion proteins. These oncoproteins often harbor domains of transcription factors (TFs) (e.g., ETV6::RUNX1, TCF3::PBX1, PAX5::ETV6, and PAX5::ELN) (Coyaud et al., 2010; Familiades et al., 2009; Mullighan et al., 2007). For instance, the TF PAX5, known to play an important role in the transcriptional regulatory networks (TNRs) of early B cells, the definitive B cell commitment, and the plasticity of committed B cells (Cobaleda et al., 2007a, 2007b), was also shown to promote leukemogenesis once translocated in mouse models (Jamrog et al., 2018; Jurado et al., 2022; Smeenk et al., 2017). Thus, it is widely thought that recurrent chromosomal translocations can act as a first oncogenic event in the early steps of B-ALL initiation, though the precise mechanisms at play are yet ill-known.

The development of transgenic mouse models with activated primary oncogenes not only helped to identify the oncogenic collaborators required for leukemia development but also allowed to identify pre-LSCs in the early steps of the disease. Various studies using in vivo models demonstrated that, in both lymphoid and myeloid malignancies, a primary genetic alteration can confer stem cell–like properties to normally committed progenitors and convert them into self-renewing pre-LSCs (Cozzio et al., 2003; Gerby et al., 2014; Huntly et al., 2004; Krivtsov et al., 2006; McCormack et al., 2010; Wojiski et al., 2009). Other studies using T-acute lymphoblastic leukemia (T-ALL) mouse models showed that pre-LSCs are resistant to chemotherapeutic agents because of their distinctive slow-division rate (Gerby et al., 2016; Tremblay et al., 2018). These findings support the view that self-renewal and cell quiescence are early and obligatory events in leukemia initiation, two specific features of the cell-of-origin, and differ from the propagating activity of fully transformed leukemic blasts. However, how a primary oncogene reprograms committed B cell progenitors into pre-LSCs with altered self-renewal and survival properties remains to be fully explored (Fregona et al., 2021).

To this end, we took advantage of a genetically engineered mouse model expressing the human PAX5::ELN fusion oncoprotein (Jamrog et al., 2018), recurrently identified in B-ALL patients (Bousquet et al., 2007). PAX5::ELN-expressing mice efficiently develop B-ALL, reproduce accurately the key features of leukemia development including secondary mutations acquired during the multistep process of human B-ALL, and provide an unrestrictive and reproducible access to a pre-LSC population (Jamrog et al., 2018). Here, by combining genetic and functional assays, together with gene expression and chromatin accessibility profiling, we deciphered the biological mechanisms by which PAX5::ELN reprograms normal B cell progenitors and leads to the emergence of quiescent pre-LSCs. Our data demonstrate that B-lineage identity is perturbed in quiescent pre-LSCs, associated with the reactivation of an immature molecular program. Moreover, our TNR predicted a list of TFs to control pre-LSC reprogramming. Among them, Early growth response 1 (Egr1) represents a prime candidate TF to regulate pre-LSC quiescence. Finally, our analyses suggest that the pre-LSC signature mimics the resistance/quiescence of human B-ALL cells. In particular, we show that EGR1 expression is associated with poor clinical outcomes and is activated in quiescent/resistant leukemic blasts from patients.

Adult B cell development is initiated in the bone marrow (BM) by the entry of hematopoietic progenitors into the B cell lineage transcription program and the sequential rearrangement of immunoglobulin heavy (IgH) and light (IgL) chain loci through V(D)J recombination (Hardy et al., 2000). Phenotypic characterization of the different B cell subsets in mice using a combination of cell surface markers was established (Fig. S1 A) (Aurrand-Lions and Mancini, 2018). We previously developed a transgenic mouse model in which the human cDNA encoding the PAX5::ELN oncoprotein was inserted within the IgH locus to ensure an early and B cell–restricted expression (Jamrog et al., 2018). To determine whether PAX5::ELN oncoprotein perturbs the normal B cell development at the preleukemic stage, we performed a multiparametric staining by FACS covering exhaustively the steps of B cell differentiation (Fig. S1 A and Table S1). Unsupervised clustering allowed the visualization of the natural phenotypic progression of B cells throughout differentiation and showed that all the normal B cell subpopulations are present in preleukemic PAX5::ELN (PE^tg) mice as compared with wild type (wt) littermates (Fig. 1, A and B; and Fig. S1, B and C). Indeed, we distinguished in both lines the presence of pre-pro-B (B220⁺CD19⁻CD2⁻), pro-B (B220⁺CD19⁺Kit⁺), pre-BI (B220⁺CD19⁺BP1⁺), early pre-BII (B220⁺CD19⁺CD2⁺IL7r⁺), late pre-BII (B220⁺CD19⁺CD2⁺IL7r⁻), immature B (B220⁺CD19⁺Igκλ⁺IgM⁺), and circulating/mature-B (B220⁺CD19⁺Igκλ⁺IgM⁺CD23⁺) cell populations, corresponding to the different stages of differentiation (Fig. 1, A–C; and Fig. S1, B and C). However, our clustering approach revealed an aberrant population of B cell progenitors (aberrant B) in the PE^tg BMs (Fig. 1, A and B), featuring low expression of B220 (B220^low) and inappropriate expression of Kit (CD117), IL7r (CD127), IgM, and λ5 (CD179b) (Fig. 1 C), which is directly associated with the transgene expression (Fig. S1 D). Indeed, ELN expression is restricted to aberrant B cells (CD19⁺B220^low) and is negative in all the normal B cell subpopulations (CD19⁺B220⁺) as well as in hematopoietic progenitors (CD19⁻B220⁻Kit^+/low) from PE^tg mice (Fig. S1 D). Interestingly, we noticed through the exploration of BP1, CD2, CD25 (IL2rα), and Igκλ markers, a phenotypic progression within the aberrant B population that is reminiscent of normal B cell differentiation (Fig. 1 A and Fig. S1 E). Moreover, in addition to the strong expression of Kit, we also observed a concomitant expression of the stem cell surface marker Sca-1 (Morcos et al., 2017) in about 30% of the aberrant B cells from the preleukemic PE^tg BM (Fig. S1 F). Together, these results indicate that PAX5::ELN oncoprotein induces a partial blockade of differentiation associated with the emergence of an aberrant B cell population in a steady-state condition.

To explore the functional impact of PAX5::ELN oncoprotein on preleukemic B cells, we performed serial transplantations of wt and PE^tg preleukemic total B cells (Fig. 1 D). Short-term reconstitution potential was assessed by analyzing the BM engraftment 2 wk after transplantations in primary and secondary recipient mice. In contrast to normal B cells that are devoid of self-renewal potential, preleukemic B cells efficiently engrafted over primary and secondary recipients (Fig. 1 E). The phenotypic characterization of engrafted cells confirmed that this short-term self-renewal activity is associated with the selection and expansion of the aberrant B cell population (Fig. 1 F). Indeed, this phenotype was strongly stable over primary and secondary recipients (Fig. S2 A), as exemplified by the high level of Kit expression (Fig. 1 F). Moreover, tertiary transplantations of preleukemic B cells eventually led to long-term B-ALL development in about 5 mo (Fig. 1 G), which is the delay required to acquire the full pattern of transforming secondary mutations (Jamrog et al., 2018). Strikingly, fully transformed leukemic blasts from tertiary recipients exhibited a distinct phenotype (Fig. 1 F and Fig. S2 A) and were able to efficiently propagate B-ALL a few days only after quaternary transplantations, even with low doses of injected cells (Fig. S2 B). Finally, Kit⁺ and Kit⁻ B cells from PE^tg BMs were purified to compare their short-term engraftment capacity after transplantation in primary mice (Fig. S2 C). We controlled that purified Kit⁺ cells from wt mice were devoid of engraftment potential, and that the aberrant self-renewal potential of PE^tg B cells was well enriched in the Kit⁺ compartment (Fig. 1 H). Moreover, we confirmed that only Kit⁺ preleukemic B cells sustained long-term B-ALL development after secondary transplantation (Fig. S2 C and Fig. 1 I). Together, our results suggest that PAX5::ELN oncoprotein converts Kit⁺ B cell progenitors into pre-LSCs that have acquired de novo self-renewing activity in the first steps of the disease.

Cell-cycle restriction is a critical feature of normal hematopoietic stem cells (HSCs) to preserve their self-renewal function (Wilson et al., 2008). We therefore hypothesized that cell quiescence could be a biological mark of pre-LSCs in the PE^tg model. To test this hypothesis, we analyzed Ki67 expression in the different B cell subsets of wt and PE^tg preleukemic mice in steady-state condition (Fig. 2 A and Table S1). As expected, Ki67 expression followed the physiological proliferation rate of normal B cell differentiation (Tomura et al., 2013) in wt and PE^tg mice (Fig. 2, A and B; and Fig. S2 D). Furthermore, non-cycling Ki67⁻ cells were detected within the aberrant PE^tg B cell population (Fig. 2, A and B; and Fig. S2 D). Since cell-cycle restriction is also an important characteristic of treatment-resistant cells in B-ALL (Ebinger et al., 2016; Turati et al., 2021), we assessed in vitro dose responses of doxorubicin (DOXO), methotrexate (MTX), and vincristine (VCR) on the aberrant PE^tg B cell population and compared the sensitivity of negative and positive Ki67 cells (Fig. 2 C and Fig. S2 E). We observed that Ki67⁻ cells were still detected after 48 h of coculture and were strongly selected upon the treatment (Fig. S2 E). Moreover, the analysis of the absolute numbers of each fraction after the treatment showed that Ki67⁻ cells were less sensitive to the three chemotherapeutic agents than Ki67⁺ cells (Fig. 2 C).

To further address the question of whether PAX5::ELN perturbs the cell division kinetic of preleukemic B cells, wt and PE^tg Kit⁺ B cells were labeled with the cell trace violet (CTV) and cultured for 3 days (Fig. 2 D). This in vitro cell division assay indicated that PAX5::ELN reduced the global proliferation rate of preleukemic cells and revealed the persistence of an undivided (D0) population (Fig. 2, E and F) that maintained the expression of Kit (Fig. S2 F). Next, we purified the slow-cycling (D0–2) PE^tg B cells population, which represents <1% of total generated cells and are undetectable in CTV-labeled wt counterparts cultured in the same experimental conditions (Fig. 2, E and F). High-cycling (D8–9) cells from both wt and PE^tg conditions were also purified as controls (Fig. 2 E). RNA sequencing (RNA-seq) demonstrated the presence of PAX5::ELN fusion transcript in both D0–2 and D8–9 preleukemic cells (Fig. S2 G) and their polyclonality of V_H(D_H)J_H and V_LJ_L rearrangements (Fig. 2 G), confirming their preleukemic state. These cells were then treated in vitro separately by DOXO, MTX, or VCR (Fig. 2, D and H) with a sublethal concentration that we previously determined by dose-response experiments on total wt and PE^tg Kit⁺ B cells (Fig. S2 H). This assay revealed a significant resistance of slow-cycling PE^tg D0–2 cells in contrast to wt and PE^tg D8–9 cells (Fig. 2 H). To address at the functional level the role of cell-cycle restriction in pre-LSC activity, equal numbers of D0–2 and D8–9 preleukemic cells were transplanted (Fig. 2 D). Both D0–2 and D8–9 preleukemic PE^tg B cells were able to engraft the BM of recipient mice 3 wk after transplantation (Fig. 2 I). However, only donor-derived D0–2 PE^tg cells maintained high-level expression of Kit in vivo (Fig. S2 I), sustained the reconstitution 16 wk after transplantation (Fig. 2 I), and the long-term B-ALL development (Fig. 2 J). Together, our results indicate that cell-cycle restriction of preleukemic cells is associated with chemoresistance and leukemia initiation.

PAX5::ELN perturbs B cell differentiation (Fig. 1) and the cell cycle of preleukemic cells (Fig. 2). Thus, we asked whether the main B cell differentiation molecular programs are affected by PAX5::ELN in slow- and high-cycling preleukemic cells. Early B cell development is controlled by the IL7 receptor and the pre-B cell receptor (pre-BCR), two interacting signaling systems that tightly coordinate alternative phases of cell survival and differentiation through the JAK-STAT and pre-BCR/BLNK pathways, respectively (Clark et al., 2014; Reth and Nielsen, 2014). Each pathway has antagonistic and balanced functions to coordinate the proliferation and differentiation switch at the pre-B stage (Fig. 3 A). Interestingly, IL7r is strongly expressed in the aberrant PE^tg B cell population, together with the accumulation of the surrogate light chain λ5 at their surface (Fig. 1 C). Thus, we analyzed IL7r and pre-BCR/BLNK pathways by using a phosphoflow cytometry approach after ligand and chemical ex vivo stimulations, respectively (Fig. S3 A and Fig. 3 B). Ligand stimulation of IL7r efficiently induced the phosphorylation of its two downstream effectors, STAT5 and FOXO1, in the aberrant B (B220^low) PE^tg population, in a similar way to normal (B220^high) pro-B cells (Fig. S3 A and Fig. 3 B, left panels). Accordingly, the presence of IL-7 protected aberrant PE^tg B cells from apoptosis (Fig. 3 C) and efficiently promoted their proliferation (Fig. 3 D) in vitro. In addition, this process was abrogated in the presence of tofacitinib, a well-known JAK-STAT inhibitor (Fig. 3, C and D), showing that IL-7r signaling is critical to maintaining the survival of preleukemic cells. We also assessed the (pre-)BCR signaling response by mimicking its downstream activation ex vivo with hydrogen peroxide (H₂O₂) (Fig. S3 A and Fig. 3 B, right panels), a well-known inducer of BLNK/PLCγ2 phosphorylation cascade (Patterson et al., 2015; Reth, 2002). As expected, H₂O₂ induced the phosphorylation of BLNK and PLCγ2 along the normal B cell differentiation pathway. In contrast, the aberrant PE^tg B cell population was unresponsive to BLNK/PLCγ2 stimulation (Fig. S3 A and Fig. 3 B, right panels).

Interestingly, ligand stimulation of IL7r efficiently induced the phosphorylation of STAT5 in both Ki67⁺ and Ki67⁻ cells within the aberrant PE^tg B cell population, which is abrogated in the presence of tofacitinib (Fig. 3 E, right panels). In contrast, IL7r activation is concentrated in cycling Ki67⁺ cells from normal B cells (Fig. 3 E, left panels), corresponding to cycling pro-B cells (Fig. 2 B and Fig. S2 D). Moreover, gene set enrichment analysis (GSEA) from the RNA-seq of purified PE^tg D0–2, PE^tg D8–9, and wt D8–9 cells (Fig. 2 D and Table S2) confirmed that the IL7r/JAK-STAT axis was significantly upregulated by PAX5::ELN oncoprotein in slow- and high-cycling preleukemic cells (Fig. S3 B). This observation was reinforced by the comparison of our PAX5::ELN-modified genes with the Stat5-binding genes arising from the ChIP sequencing of murine Stat5b-induced B-ALL cells (Katerndahl et al., 2017), showing a substantial enrichment of Stat5 targets in PAX5::ELN upregulated genes in slow- (D0–2) and high- (D8–9) cycling preleukemic cells (Fig. 3 F and Table S2). Accordingly, the downstream targets of IL7r/JAK-STAT axis, including the prosurvival genes Cdkn2a, Il2rα, Tnfrsf13b, and Socs3, were significantly activated in both D0–2 and D8–9 cells (Fig. 3 G, left panels). In contrast, several components of the pre-BCR/BLNK signaling including CD79b, CD79a, Lyn, Syk, Blnk, Ikzf3, and Irf8 were downregulated in both D0–2 and D8–9 cells (Fig. 3 G, right panels). Among them, the pre-B cell differentiation genes Blnk, Ikzf3, and Syk also belong to Stat5 targets (Fig. 3 F), reinforcing the notion that IL7r/JAK-STAT axis negatively regulates the B cell differentiation (Fig. 3 A).

Collectively, combined with our phosphoflow cytometry results, our RNA-seq data suggest that the activation of IL7r/JAK-STAT pathway favors the maintenance and survival of preleukemic cells, in slow- and high-cycling fractions.

While the above results demonstrate that the leukemia-initiating potential is enriched in slow-cycling population, the imbalance between IL7r/JAK-STAT and pre-BCR/BLNK molecular circuits is a shared feature of all PAX5::ELN preleukemic cells. Thus, to further identify and characterize quiescent pre-LSCs, we generated PE^tg::H2B-GFP^tg mice, derived from doxycycline-inducible H2B-GFP^tg mice, an in vivo model allowing the study of cell quiescence in normal HSCs (Foudi et al., 2009; Wilson et al., 2008). After a doxycycline pulse, we compared the cell division kinetics in vivo between wt and preleukemic PE^tg B cells based on the loss of GFP expression during a time course of doxycycline chase (Fig. 4 A). At the end of the doxycycline pulse, almost all B cells in both wt and preleukemic BM expressed the H2B-GFP marker (Fig. 4 B). However, the decrease of GFP after the withdrawal (chase period) of doxycycline revealed a substantial delay of the global GFP loss in preleukemic PE^tg B cells, with the persistence of GFP^highPE^tg cells after 1-wk chase of doxycycline (Fig. 4 B). Based on the loss of GFP after 2- and 3-wk chase of doxycycline, we defined GFP^high cells as quiescent rather than dormant cells, such as reported for normal stem cells (Foudi et al., 2009; Wilson et al., 2008). Therefore, we adapted the multiparametric staining to identify the quiescent GFP^high B cell subset. Quiescent cells were exclusively found in the aberrant B cell population of PE^tg BM (Fig. S3 C and Fig. 4 C). Additionally, we observed that GFP loss was proportional to the acquisition of the Ki67 marker, almost all GFP^high cells did not express Ki67 (Fig. 4 D), and only this non-cycling population sustained engraftment capacity after transplantation (Fig. 4 E) and the long-term B-ALL development (Fig. S3 D).

To refine the molecular characterization of quiescent pre-LSCs in vivo, GFP^high and GFP^low cells were purified from aberrant B cells of PE^tgH2B-GFP^tg mice after 1-wk chase of doxycycline, and an integrative molecular approach was performed comparing both gene expression profiles by RNA-seq and chromatin accessibility by assay for transposase-accessible chromatin sequencing (ATAC-seq) (Fig. 4 F) (Buenrostro et al., 2013). Transcriptome analysis identified 836 upregulated and 789 downregulated genes with an expression difference of >1.5-fold and an adjusted P value of <0.05 in quiescent GFP^high pre-LSCs (Fig. 4 F and Table S3). Combined with an expected downregulation of cell cycle–related pathways, the analysis revealed that cytokine signaling pathways were upregulated in quiescent GFP^high pre-LSCs (Fig. S3 E). Indeed, these modifications, involving the downregulation of E2f/Myc targets and the upregulation of TNFα signaling (Fig. S3 E), are crucial for HSC and cancer stem cell survival (Cabezas-Wallscheid et al., 2017; Vazquez-Santillan et al., 2015; Yamashita and Passegué, 2019; Zhang et al., 2022). Differentially accessible chromatin regions were revealed by ATAC-seq analysis in preleukemic GFP^high (7,439 peaks) and GFP^low (5,360 peaks) cells (Fig. 4 F and Table S4). Their associated genes were enriched for molecular pathways similar to those identified at the mRNA level (Fig. S3, E and F). These observations highlight the direct relationship between the expression of genes and their surrounding accessibilities, allowing the identification of specific regulators.

To this end, we combined footprinting analysis using hint-ATAC and “diffTF” computational tools (Berest et al., 2019) to identify and classify TF motif accessibilities that were differentially regulated in quiescent GFP^high pre-LSCs (Fig. 4 G). We observed that the motif accessibility of critical TFs involved in the B cell identity, such as Ebf and Pax, or in the cell cycle, such as E2f, was significantly downregulated in quiescent GFP^high pre-LSCs (Fig. 4 G and Fig. S4 A). While the accessibility footprint (Tn5 insertion site) of Ebf and Pax was decreased in quiescent GFP^high pre-LSCs (Fig. S4 A), we observed that the expression of Ebf1 and total Pax5 was tightly increased at the transcriptional level (significantly for Ebf1 but not for total Pax5, Fig. S4 B). Of note, the expressions of Pax5 WT and PAX5-ELN were quietly similar between H2B-GFP^high and H2B-GFP^low cells (Fig. S4 B). This observation indicates that the binding activity of these two TF is affected in H2B-GFP^high pre-LSCs, but not their expression. Interestingly, we observed that Pax5-repressed genes, but not Pax5-activated genes (Revilla-I-Domingo et al., 2012), were drastically enriched in GFP^high pre-LSCs (Fig. 4 H and Fig. S4 C). Together, these results strongly suggest that Pax5 has lost its repressive activity in quiescent pre-LSCs, allowing the upregulation of a non-B cell molecular program. Indeed, we observed the upregulation of the chromatin accessibility to Spi and Runx motifs in quiescent GFP^high pre-LSCs (Fig. 4 G and Fig. S4 A). This was associated with increased expression of Spi1 (PU.1) and Runx1 (Fig. S4 B), two critical TFs regulating the quiescence and the myeloid differentiation process of normal HSCs (Chavez et al., 2021; Imperato et al., 2015; Staber et al., 2014). This observation was also associated with a downstream activation of a myeloid program in quiescent GFP^high pre-LSCs at the transcriptional level. Indeed, the analysis mainly revealed the upregulation of genes involved in TNFα signaling and in myeloid differentiation in GFP^high pre-LSCs (Fig. 4 I). Finally, we observed a strong enrichment of both murine (Fig. 4 J and Table S3) (Venezia et al., 2004) and human (Fig. S4 D and Table S3) (Garcia-Prat et al., 2021) dormant/quiescent HSC signature in the GFP^high pre-LSCs, suggesting redundant molecular mechanisms between pre-LSCs and HSCs. Collectively, our data indicate that quiescent pre-LSCs partially lose B-lineage molecular identity and concomitantly reactivate an inappropriate and immature transcriptional program.

The above findings raise the question as to what extent stem cell–like molecular features are reprogrammed in quiescent pre-LSCs. Thus, we integrated our results within ATAC-seq and RNA-seq data from the Immunological Genome Project (ImmGen) consortium (https://www.immgen.org/). First, we extracted both chromatin accessibility and gene expression profiles from normal HSPCs (LT-HSCs, MPP1, ST-HSCs, MPP4), MLPs, common lymphoid progenitor (CLP), Fr.A (i.e., Pre-pro-B), and B cells (Fr.B/Pro-B to Fr.E/immature-B) (Fig. 5 A). Then, differentially accessible peaks and expressed genes from preleukemic GFP^high and GFP^low cells were clustered by a k-means approach according to their signal in the different steps of normal differentiation (Fig. 5, A–C and Table S5). Strikingly, ATAC-seq data revealed that open chromatin regions from HSPCs, including LT-HSCs and other multipotent immature progenitors (clusters C1 to C4), were enriched in quiescent GFP^high pre-LSCs, contrasting in this regard with GFP^low cycling cells that were rather enriched in accessible regions from committed B cells (clusters C7, C8) (Fig. 5 B). Furthermore, RNA-seq data identified four clusters of genes in preleukemic GFP^high and three clusters of genes in GFP^low cells. This analysis notably revealed a unique “HSPC^like cluster” in quiescent GFP^high cells, composed of genes mostly upregulated in LT-HSCs, MPP1, ST-HSCs, and MPP4 fractions (Fig. 5 C). An “MLP^like cluster” was also defined in quiescent GFP^high cells, composed of genes upregulated in CLP and pre-pro-B precursors, two populations that retain multilymphoid potential (Fig. 5 C). This MLP signature, which is strongly activated in quiescent GFP^high cells, was also observed in human B-ALL chemoresistant cells (Turati et al., 2021).

These observations revealed at the transcriptional level the plasticity and reprogramming of quiescent pre-LSCs that can partially reactivate a non-B cell molecular program.

According to the global upregulation of the cell-cycle program in GFP^low cells (Fig. 4 I and Fig. S3 E), we identified a cluster of genes named “cell-cycle^like cluster,” whose expression pattern followed the normal proliferation rate of each subset in the differentiation (Fig. 5 C). Conversely, a “quiescence^like cluster” corresponding to the mirror image of the “cell-cycle^like cluster” was detected within GFP^high cells (Fig. 5 C).

To predict the “core” TFs defining pre-LSC identity, and in particular pre-LSC quiescence, we built a TRN from ATAC-seq (Fig. 5 B) and RNA-seq (Fig. 5 C) data using LASSO-StARS algorithm (Miraldi et al., 2019; Pokrovskii et al., 2019). This algorithm integrates (i) motif analyses, (ii) differential accessibility motifs, and (iii) gene expression to construct a set of TF–gene interactions. This analysis predicted TFs that positively (activating) or negatively (repressive) regulate their target genes. Thus, we contrasted five specific TRNs (#1 to #5) to one shared TRN involved in the pre-LSC signature (Fig. 5 D). In particular, a quiescence module (TRN #1) was identified in GFP^high pre-LSCs composed of four TFs (Egr1, Mnt, Klf6, and Atf3) (Fig. 5 D). Among them, Early growth response 1 (Egr1) regulates the higher number of target genes (121 activated and 242 repressed genes). Moreover, Egr1 was the top predicted TF to regulate positively the quiescence and negatively the cell cycle (TRN#1, Fig. 5 D). Accordingly, we observed a strong downregulation of the quiescence^like cluster and upregulation of the cell-cycle^like cluster in human HSPCs in which EGR1 has been silenced by RNA interference (Stoddart et al., 2022) (Fig. S4 E). Consistent with the critical role of Egr1 in controlling normal HSC dormancy and functions (Cabezas-Wallscheid et al., 2017; Min et al., 2008; Scheicher et al., 2015), our result suggests that it may be also relevant in pre-LSCs.

To address this question, we used a loss-of-function approach by CRISPR/Cas9 strategy (Fig. S4 F). We generated PE^tgCas9^tg mice that constitutively express the Cas9 endonuclease (Tzelepis et al., 2016) and designed two single-guide RNAs (sgRNAs) (sgEgr1 T5 and sgEgr1 T7) that target the genomic region encoding the exon 1 of Egr1 gene (Table S1). Aberrant B cells from preleukemic PE^tgCas9^tg mice were purified and transduced with lentiviral vectors encoding sgEgr1 T5 or sgEgr1 T7 or a non-targeted control sgRNA (sgCTL). Transduced PE^tgCas9^tg GFP⁺ cells were then purified (Fig. S4 G) and genome editing efficiency was determined by analyzing the rate of insertions/deletions (InDels) induced in the targeted region. Although sgEgr1 T5 did not induce InDels in the Egr1 locus, sgEgr1 T7 allowed for almost a complete (93%) genome editing of the Egr1 locus (Fig. S4 H). Subsequent in vitro cell division assay indicated that Egr1 genomic editing by sgEgr1 T7 slightly but significantly increases the global proliferation rate of aberrant B cells (Fig. S4, I and J). Interestingly, this observation was associated with a reduction of the proportion of slow-cycling (D0–2) cells (Fig. 5 E), with a decrease of Kit expression and an increase of IgM expression at the surface of aberrant B cells (Fig. 5 F). Thus, our results show that Egr1 genomic editing reverses, at least partially, the quiescence and the aberrant expression of Kit of preleukemic cells.

Finally, we built the molecular network of genes involved in TRN#1 from Fig. 5 D (Fig. S4 K). Strikingly, several members of the network such as the TFs Atf3 (Liu et al., 2020) and Klf6 (Adelman et al., 2019), the GTPase Gimap5 (Chen et al., 2011), and the transcriptional coactivator Taz (Althoff et al., 2020) have been also documented to control HSC functions.

Collectively, the data arising from our integrative molecular approach predict a list of TFs and their target genes to control molecular reprogramming in pre-LSCs and highlights Egr1 as a strong candidate to regulate pre-LSC quiescence.

Finally, we addressed the relevance of the expression modifications of murine pre-LSCs in human-resistant and quiescent leukemic cells. Interestingly, the downregulation of E2f/Myc targets (Fig. S3 E) and cell cycle/oxidative phosphorylation program (Fig. 4 I), combined with the upregulation of TNFα signaling (Fig. S3 E and Fig. 4 I) and HSPC/MLP signatures (Fig. 5 C) that we identified in GFP^high pre-LSCs from the PAX5::ELN model, have been also observed in treatment-resistant (minimal residual disease [MRD]) cells and in CFSE label-retaining cells (LRC) from B-ALL patient-derived xenografts (PDXs) in two independent studies (Ebinger et al., 2016; Turati et al., 2021). These molecular similarities (Fig. 6, A and B; and Fig. S5 A) suggest that PAX5::ELN pre-LSC signature mimics in some aspect the resistance/quiescence of human B-ALL cells.

Thus, we explored EGR1 expression in resistant and/or quiescent human B-ALL cells. We transplanted leukemic blasts from two “de novo” ETV6-RUNX1 B-ALL patients (HB#010 and HB#007) into immunodeficient NOD.Cg-Prkdc^scidIl2rg^tmWjl (referred to as NSG) mice and monitored for engraftment before treatment with a chemotherapeutic cocktail (dexamethasone [DEXA] + VCR) or vehicle control (Fig. 6, C and D; and Fig. S5, B and C). Strikingly, B-ALL progression was prevented in chemotherapy-treated mice characterized by a significant reduction of the leukemic burden in the BM and the spleen (Fig. 6, E and F; and Fig. S5, D and E). This drastic reduction of human B-ALL cells in vivo was associated with an enrichment of quiescent Ki67⁻ leukemic blasts (Fig. 6 G and Fig. S5 F). This resistance was also associated with a significant upregulation of EGR1 transcript in purified residual leukemic blasts (Fig. 6 H and Fig. S5 G). Interestingly, we observed that residual leukemic blasts developed leukemia in secondary recipients with a significant delay as compared with untreated cells (Fig. S5 H). This result reinforces the notion that quiescent/resistant cells from PDX models exhibit a long-term leukemia-initiating activity and mimics, in some aspects, relapse-initiating cells from B-ALL patients.

Based on our previous observations, we hypothesized that EGR1 expression was already upregulated in quiescent cells within the tumor burden of B-ALL. Thus, leukemic blasts from two other ETV6::RUNX1 B-ALL patients (HB#002 and HB#008) were labeled with CTV and cocultured on MS5 stromal cells (Fig. 6 I and Fig. S5 I). Strikingly, our data showed the feasibility of identifying undividing/slow-dividing cells from “de novo” B-ALL samples using this in vitro cell division approach (Fig. 6 J and Fig. S5 J). Of note, we observed that these cells were enriched in the CD45^neg/low phenotype (Fig. 6 J and Fig. S5 J), which has been associated with drug-tolerant leukemic clones leading to B-ALL relapse in patients (Dobson et al., 2020). Therefore, we confirmed that EGR1 expression was higher in purified undivided (CTV^high) B-ALL cells from the two patients as compared with cycling (CTV^low) leukemic cells (Fig. 6 K and Fig. S5 K). Together, our observations suggest that EGR1 overexpression is a “de novo” molecular characteristic of quiescent/resistant cells within the tumor bulk.

Finally, we took advantage of the RNA-seq and clinical data of a large cohort of B-ALL patients (Gu et al., 2019) to explore EGR1 expression and its clinical outcome. The analysis of the event-free survival (EFS) curves showed that the clinical outcome was drastically affected for all B-ALL patients with high levels of EGR1 (Fig. S5 L). This is partially explained by the observation that EGR1 expression was higher in high-risk genetic subgroups (i.e., Ph-like and KMT2A-rearranged subgroups) than in B-ALL subgroups associated with a better prognosis (i.e., ETV6::RUNX1, TCF3::PBX1, hyperdiploid, and PAX5alt) (Inaba and Pui, 2021) (Fig. S5 M). Furthermore, the poor clinical outcome was also strikingly associated with EGR1 expression within the different B-ALL oncogenic subgroups (Fig. 6 L). Collectively, our observations strongly suggest that cell quiescence and EGR1 activation are common and oncogene-independent features of treatment-resistant B-ALL cells.

In the present study, we found that a single primary oncogenic event of B-ALL is sufficient to confer stem cell–like features to a subset of committed B cell progenitors, converting them to pre-LSCs. Combined with an unsupervised clustering representation of FACS data, our multiparametric immunophenotypic approach allowed for a precise visualization of the normal progression of B cells throughout differentiation. In addition, this strategy revealed concomitantly the presence of an aberrant B cell population in preleukemic PE^tg mice, characterized by the abnormal co-occurrence of phenotypic markers. To date, this exhaustive coverage of the B cell differentiation has not yet been described in leukemic mouse models, mainly limited by the number of simultaneously assayed markers. However, the identification of aberrant antigen coexpression is commonly used in clinics to define the MRD in B-ALL (Cherian and Soma, 2021; Hassanein et al., 2009) based on the principle that leukemic cells express phenotypic features that can be used to distinguish them from normal hematopoietic cells. Moreover, the development of single-cell mass cytometry (CyTOF) approach, quantifying simultaneously a large number of surface markers and intracellular phosphorylated proteins, allowed for the identification of several features involved in normal and pathological B cell development in humans with unprecedented resolution (Bendall et al., 2014; Good et al., 2018). Thus, our approach reinforces the notion that exploring abnormal coexpression using a large number of phenotypic markers is a valuable approach to predict the aberrant differentiation state occurring in cells.

Combined with RNA-seq, the integration of intracellular phosphorylated proteins established that the ectopic activation of IL7r/JAK-STAT pathway, which is normally shut down by the pre-BCR/BLNK axis, favors the differentiation blockade of preleukemic cells. Thus, the imbalance between the IL7r/JAK-STAT pathway and pre-BCR/BLNK axis appears as a critical feature of B-ALL initiation. Interestingly, this molecular imbalance, including unresponsive pre-BCR signaling, has been shown to predict patient relapse in B-ALL (Good et al., 2018). Moreover, secondary alterations in human B-ALL frequently mimic cytokine-receptor signaling through the constitutive phosphorylation of STAT5 (IL7R, CRLF2, and JAK mutations) or mimic the pre-BCR downstream pathway through the constitutive activation of the RAS/MAPK signaling (NRAS, KRAS, and PTPN11 mutations) (Jamrog et al., 2018; Mullighan et al., 2007). It has been recently demonstrated that mutations in these two pathways are mutually exclusive in human B-ALL (Chan et al., 2020). In addition, the concurrent reactivation of one pathway affects the activity of the other, reciprocally, leading to the subversion of B-ALL transformation (Chan et al., 2020). B-ALL development in the PAX5::ELN model is also associated with the acquisition of secondary mutations in genes of either JAK/STAT (Jak3 mutations) or RAS/MAPK (Kras and Ptpn11 mutations) signaling pathways (Jamrog et al., 2018). These observations, therefore, suggest that secondary mutations either reinforce the aberrant activation of the JAK/STAT pathway already primed in preleukemic cells or allow for a bypass of pre-BCR signaling deficiency to trigger B-ALL transformation.

Various studies suggest that the restricted cell cycle is an important mechanism of therapeutic resistance in both human and murine leukemia models (Ebinger et al., 2016; Gerby et al., 2016; Prost et al., 2015; Tremblay et al., 2018; Trumpp et al., 2010; Turati et al., 2021). In parallel with an in vitro cell division assay, we used the doxycycline-inducible H2B-GFP^tg mouse model to demonstrate in vivo the existence of a slow-cycling and drug-resistant population of pre-LSCs in the PAX5::ELN-induced B-ALL model. In particular, our work corroborates that self-renewal and clonal evolution could be restricted to a rare and slow-cycling population of preleukemic cells, as previously observed in the Lmo2-induced T-ALL model (Tremblay et al., 2018). This not only reinforces the importance of cell-cycle restriction in pre-LSC activities in two distinct lineages but also makes the H2B-GFP system a powerful in vivo tool to develop strategies targeting the quiescence and the resistance of relapse-inducing clones in different models.

Additionally, the present study brings new insights regarding the molecular mechanisms of the cell-of-origin of B-ALL demonstrating that quiescent pre-LSCs are characterized by a partial loss of B cell lineage identity and activate an HSPC-like molecular program. This finding is in agreement with pioneer works in acute myeloid leukemia and T-ALL demonstrating that a primary oncogene can activate an aberrant stem cell–like program in committed myeloid progenitors (Krivtsov et al., 2006) or thymocytes (McCormack et al., 2010). While the presence of PAX5::ELN induces the emergence of quiescent cells in committed B cells, its expression level in quiescent and cycling cells is quite similar. Notwithstanding the fact that functional and molecular stem cell–like features were focused in quiescent cells, this observation raises at least two questions: (i) whether cell-cycle restriction is a common aberrant characteristic of pre-LSCs or is a critical property to prime leukemia initiation, and (ii) whether stem cell properties of pre-LSCs such as self-renewal, drug resistance, and leukemia-initiating activity are reversible processes, as demonstrated on human B-ALL blasts (Ebinger et al., 2016).

The question of whether a primary genetic alteration induces sufficient molecular and functional plasticity to cause lineage subversion remains poorly explored. A previous investigation on the cell-of-origin in mixed phenotype acute leukemia supports the notion that a founding alteration, rather than the secondary events, induces lineage plasticity in preleukemic clones (Alexander et al., 2018). Another study using lineage-specific oncogene activation also shows that ETV6::RUNX1 fusion protein can induce both B-ALL and T-ALL, and that leukemia transformation is dependent on the nature of the secondary mutations (Rodríguez-Hernández et al., 2021). While ETV6::RUNX1 is associated with B-ALL in humans, this work suggests that preleukemic clones can exhibit a T cell potential when expressed in the appropriate lineage. Our results revealed that the TF activity of PAX5 and EBF1, considered master factors to maintain B cell identity by suppressing alternative lineage choices (Busslinger, 2004; Nutt et al., 1999; Ramamoorthy et al., 2020), is downregulated in quiescent pre-LSCs. The downregulation of PAX5 and EBF1 activity is associated with the upregulation of PU-1 and RUNX1 activity, and with the activation of a non-B cell molecular program. Integrating chromatin accessibility and gene expression data recently brought major insights and facilitated the identification of putative regulatory regions and their candidate binding TFs, thus allowing a more precise definition of the regulatory networks in immune cells (Miraldi et al., 2019; Pokrovskii et al., 2019). Here, we applied this computational method to build cluster-specific TRNs that define the molecular identity and plasticity of quiescent pre-LSCs in our model. In particular, this approach allowed us to identify Egr1 as a strong TF candidate regulating positively pre-LSC quiescence. In addition, our loss-of-function approach demonstrated that Egr1 inhibition reverses, at least partially, the cell quiescence of preleukemic PAX5-ELN cells. This observation is consistent with its critical role in controlling normal HSC dormancy, localization, and functions that have been described using both Egr1^−/− mouse model (Min et al., 2008) and RNA interference in human CD34⁺ cord blood–derived HSPCs (Stoddart et al., 2022). Finally, molecular similarities with resistant clones in B-ALL patients after chemotherapy (Ebinger et al., 2016; Turati et al., 2021), including the upregulation of EGR1, strongly make our PAX5::ELN quiescent pre-LSCs a relevant drug-resistant cellular platform to develop new targeted therapies.

PAX5::ELN (PE^tg) transgenic mouse model was developed by our team as previously described (Jamrog et al., 2018). Briefly, cDNA encoding the human PAX5::ELN fusion protein was inserted at the IgH locus under the control of a VH promoter (PVH) and the endogenous Eμ enhancer to trigger PAX5::ELN expression in the early phase of B cell development. The previously developed inducible TetOP-H2B-GFP (H2B-GFP^tg) mouse model (Foudi et al., 2009) was purchased from Jackson Laboratory. Rosa26-Cas9 (Cas9^tg) mice (Tzelepis et al., 2016) were generously provided by the Wellcome Sanger Institute of Cambridge (UK). Homozygous PE^tg, H2B-GFP^tg, and Cas9^tg mice were maintained by crossbreeding, their genotypes were verified by PCR, and all the experiments were performed in heterozygous mice. C57BL6 (Ly6.2, CD45.2) mice were purchased from Charles River Laboratories and Pep3b (Ly6.1, CD45.1) B6.SJL congenic mice were initially obtained from The Jackson Laboratory. All animals were housed in pathogen-free conditions (Anexplo US006 CREFRE) in accordance with the European Directive 2010/63/EU and the French Institutional Guidelines for animal handling. Mice were handled according to protocols approved by the Regional Ethical Committee (agreement #A31555010). Mice were backcrossed into the C57BL6 background for >10 generations.

Single-cell suspensions were prepared from BM of wt and PE^tg mice of the indicated ages in Iscove’s modified Dulbecco’s medium (IMDM; Gibco) supplemented with 2% fetal bovine serum (FBS) (Stemcell Technologies). Immunostainings were performed using antibodies for flow cytometry obtained from Pharmingen (BD Biosciences), Invitrogen, and Miltenyi are listed in Table S1. FACS analysis was performed on a Fortessa cytometer (BD Biosciences) using FlowJo (BD Biosciences) software. For surface staining, cells were incubated with the antibodies for 20 min in IMDM 2% FBS at 4°C and washed twice with PBS1X before analysis. For cell-cycle analysis, cells were fixed and permeabilized (CytofixCytoperm Plus; BD Bioscience) for 30 min before staining with the anti-Ki67 antibody, in accordance with the manufacturer’s instructions. For cell sorting, BM cells from wt or preleukemic PE^tg mice were flushed, and B cell fraction were enriched by B220 magnetics microbead sorting (Miltenyi Biotech). B cell fraction was subsequently incubated with anti-CD19, -CD23, and -Kit antibodies listed in Table S1. Cell sorting of Kit⁺ and Kit⁻ (CD19⁺CD23⁻) B cells was performed on a MoFlo Astrios sorter (Beckman Coulter).

UMAP analysis was performed using FlowJo software. At least four samples per condition were pregated on single cells of the population of interest. Then, informatic cleaning and normalization were performed using FlowAI package (V2.1) (Monaco et al., 2016) and CytoNorm package (V1.0) (Van Gassen et al., 2020) for each sample and the same number of cells were then concatenated. Arc sin transformation was performed manually for each marker to discriminate the different B cell populations. The dimension reduction algorithm UMAP (umap-learn Python package v2.4.0) (Becht et al., 2018) was run using Euclidian distance with 15 nearest neighbors and 0.5 distance parameters. Thus, the UMAP analysis was performed using a multiparametric staining by FACS integrating the 12 markers shown in Fig. S1 A covering all the steps of the B cell differentiation. UMAP representations of the cell density and the clustering analysis of the different B cell subpopulations are based on the gating strategy shown in Fig. S1, B and C. In the UMAP analyses, each subset is represented by one color. Black arrow indicates the physiological phenotypic progression of B cells in the differentiation, and red dotted line delimits the aberrant preleukemic B cell population induced by PAX5::ELN.

B cell populations were cocultured on MS5 stromal cells in a B cell medium composed of IMDM 10% FBS, 0.05 mM β-mercaptoethanol (Sigma-Aldrich), 2 mM L-glutamine (Invitrogen), penicillin (100 U/ml), and streptomycin (100 U/ml), and supplemented with 5 ng/ml of murine Il7, 10 ng/ml of murine Flt3L, and 10 ng/ml of murine stem cell factor (SCF) (Peprotech).

Donor and recipient mice were on C57BL6 (Ly6.2, CD45.2) and Pep3b (Ly6.1, CD45.1) backgrounds, respectively, allowing for the discrimination of host- and donor-derived cells on the basis of CD45 alleles. For the serial transplantation assay, total B cells (B220⁺) from the BM of wt or preleukemic PE^tg mice (30 days old; CD45.2⁺) were purified and transplanted intravenously in equal numbers (4.5 × 10⁶ cells per mouse) into primary (I) recipient mice (6–8 wk-old; CD45.1⁺). Then, one fifth of the total BM cells from primary mice were transplanted in equal numbers (∼20 × 10⁶ cells per mouse) into secondary (II) recipients. To measure the short-term reconstitution potential of preleukemic B cells, the chimerism was analyzed by flow cytometry (FACS) 2 wk after each transplantation and was illustrated by the percentage of donor-derived cells (% CD45.2⁺) found in the BM of recipient mice. Finally, one fifth of the total BM cells from secondary mice were transplanted in equal numbers (∼20 × 10⁶ cells per mouse) into tertiary (III) recipients to assess the long-term B-ALL development, represented by Kaplan–Meier survival curves. All the recipient mice (CD45.1⁺) were pretreated, 24 h before transplantation, with 30 mg/kg of Busulfan (Busilvex; Pierre Fabre) (related to Fig. 1, D–G).

Kit⁺ and Kit⁻ (CD19⁺CD23⁻) B cells from the BM of preleukemic PE^tg mice (CD45.2) were purified and intravenously transplanted in equal numbers (5 × 10⁴ per mouse) in primary (I) recipient CD45.1 mice. Kit⁺ (CD19⁺CD23⁻) B cells from the BM of wt mice were purified and transplanted (5 × 10⁴ per mouse) as control. BM cells from the primary transplantation with preleukemic PE^tg Kit⁺ et Kit⁻ cells were transplanted in secondary (II) recipients (related to Fig. S2 C and Fig. 1, H and I).

Purified D0–2 and D8–9 cells (CD45.2⁺) from CTV-labeled preleukemic PE^tg cells were transplanted in equal numbers (2 × 10³ per mouse) in recipient CD45.1 mice (related to Fig. 2, D, I, and J).

Purified GFP^high and GFP^low cells from preleukemic PE^tgH2B-GFP^tg mice (CD45.2⁺) were transplanted in equal numbers (1.5 × 10⁴ per mouse) in recipient CD45.1 mice. Engraftment was monitored by FACS and was illustrated by the percentage of donor-derived cells (% CD45.2⁺) found in recipient mice after BM punctions. For all transplantation experiments, the number of positive mice and the median of engraftment are indicated. Statistical significance was calculated using an unpaired t test. For Kaplan–Meyer survival curves, the median of time required to develop the B-ALL is indicated on the curve (related to Fig. 4 E and Fig. S3 D).

Cell division of Kit⁺ (CD19⁺CD23⁻) wt and PE^tg B cells were analyzed by using CTV dilution. Briefly, total B cells (10⁶ cells/ml) from wt and preleukemic PE^tg mice (30 days old) were labeled with 5 μM CTV (C34557; Thermo Fisher Scientific) in PBS1X during 20 min at 37°C. CTV-labeled B cells were then washed twice in PBS1X 1% FBS according to the manufacturer’s recommendations and plated on MS5 stromal cells for 24 h to stabilize the staining. Then, Kit⁺ (CD19⁺CD23⁻) CTV-labeled B cells were purified by cell sorting and cocultured for 3 days on an MS5 cell line in a supplemented B cell medium. Since dividing cells equally distribute the dye to daughter cells, the number of cell divisions can be followed by a decrease in the fluorescence intensity. Cell division was analyzed using Winlist (Verity Software House), and proliferation index (PI) was calculated as described (Roederer, 2011). Slow- (D0–2) and high- (D8–9) cycling PE^tg cells were purified by cell sorting. High- (D8–9) cycling wt cells were also purified as a control to perform the subsequent functional and molecular experiments as indicated in the experimental procedure detailed in Fig. 2 D.

Wt or homozygous PE^tg mice were bred with homozygous H2B-GFP^tg mice to obtain heterozygous H2B-GFP^tg and PE^tgH2B-GFP^tg mice, respectively. To induce H2B-GFP transgene expression, 20-day-old mice were intraperitoneally treated with 40 mg/kg of doxycyxline (DOX, D9891; Sigma-Aldrich) and 2 mg/ml of DOX, supplemented with sucrose (S9378; Sigma-Aldrich) at 10 mg/ml, was added to drinking water for 5 wk (DOX pulse). GFP expression in all B cell subpopulations was then examined following withdrawal of doxycycline (DOX chase) for 0, 1, 2 and 3 wk. Quantification of GFP was determined for each time point by ranking 10⁴ wt or PE^tg cells from the lowest to the highest value of their GFP fluorescence intensity and the area under the curve (AUC) was then calculated. White and black dotted lines indicate the standard error of the mean (SEM) of wt and PE^tg AUC, respectively. The proportion of persisting GFP-retained (GFP^high) cells is indicated for each time point (related to Fig. 4, A–C). Purification of GFP^high, GFP^med, and GFP^low B cells was performed after 1 wk of DOX chase using a MoFlo Astrios cell sorter (Beckman Coulter).

Phosphoflow cytometry experiments are related to Fig. 3, B and E; and Fig. S3 A to explore IL7r and pre-BCR signaling pathways in preleukemic B cells. As shown in Fig. 3 A, the IL7r signal induces Stat5 phosphorylation that activates the transcription of Cdkn2a, IL2rα, Tnfrsf13b, and Socs3 to promote cell survival and proliferation. In parallel, IL7r activates the PI3K signaling pathway that phosphorylates and represses Foxo1, the inducer of Rag1 and Rag2 gene transcription, therefore inhibiting Igκ gene recombination. In contrast, pre-BCR signal induces Syk/Blnk/Plcγ2 phosphorylation cascade that converges to the transcription of Ikzf3 and Irf8 that inhibit pre-B cell proliferation and initiate Igκ gene recombination, respectively. Together, each receptor has antagonistic and balanced functions to coordinate the proliferation and differentiation switch in pre-B cells.

1.5 × 10⁶ cells from each wt and preleukemic PE^tg BM were mixed, and surface markers were stained in RPMI (Gibco) supplemented with 2% FBS for 10 min at 37°C and washed with RPMI (centrifugation, 5 min; 1,200 rpm).

IL7r-ligand stimulation was assessed by phosphoflow cytometry after the ex vivo activation of preleukemic PE^tg BM cells with IL-7 (+IL-7). Unstimulated (−IL-7) cells were used as control. Normal B cells and aberrant B PE^tg cells were distinguished according to the level of B220 expression, and pStat5 and pFoxo1 expression were detected by FACS. IL7r-ligand stimulation was performed by incubating the cells with RPMI 2% FBS supplemented or not with 30 ng/ml IL-7 (Peprotech) for 15 min at 37°C for the detection of pStat5 and for 16 h at 37°C for the detection of pFoxo1. For the staining of pStat5, cells were then fixed with prewarmed Lyse/Fix buffer 1× for 12 min at 37°C, washed twice (centrifugation, 7 min; 500 rpm) in RPMI 2% FBS, and permeabilized with cold Perm Buffer III for 30 min at 4°C, according to the manufacturer’s instructions (BD Phosphoflow). Cells were washed twice in PermWash buffer and the intracellular immunostainings with the anti-pStat5 antibody combined with the anti-Ki67 antibody were performed for 30 min at room temperature, followed by two washes before FACS analysis. For the staining of pFoxo1, cells were fixed and permeabilized with cold CytofixCytoperm Plus buffer (BD Bioscience) for 24 h at 4°C and washed twice (centrifugation, 7 min; 500 rpm) in PermWash buffer (BD Bioscience). The intracellular immunostainings with the anti-pFoxo1 antibody combined with the anti-Ki67 antibody were performed in PermWash buffer for 1 h at room temperature, followed by two washes (centrifugation, 7 min; 500 rpm) before FACS analysis.

(Pre-)BCR stimulation was assessed by phosphoflow cytometry after the ex vivo activation of preleukemic PE^tg BM cells with H₂O₂ (H₂O₂⁺). Unstimulated (H₂O₂⁻) cells were used as control. Normal B cells and aberrant B PE^tg cells were distinguished according to the level of B220 expression, and pBlnk and pPlcγ2 expression was detected by FACS. (Pre-)BCR stimulation was performed by incubating the cells with RPMI 2% FBS supplemented or not with 1 mM H₂O₂ for 5 min at 37°C. Cells were then fixed and permeabilized with cold CytofixCytoperm Plus buffer (BD Bioscience) for 45 min at 4°C and washed twice (centrifugation, 7 min; 500 rpm) in PermWash buffer (BD Bioscience). The intracellular immunostainings with the anti-pBlnk or the anti-pPlcγ2 antibodies combined with the anti-Ki67 antibody were performed in PermWash buffer for 30 min at room temperature, followed by two washes (centrifugation, 7 min; 500 rpm) before FACS analysis. Immunostaining combinations used for the phospflow cytometry are listed in Table S1.

sgRNAs targeting the exon 1 of Egr1 (Table S1) were designed using CCTop-CRISPR/Cas9 target online predictor software (Center for Organismal Studies Heidelberg) and subcloned in pLKO5-sgRNA-EFS-GFP, in which the sgRNA and GFP coding sequences are controlled by the U6 and the EF1α promoters, respectively (Heckl et al., 2014). Concentrated VSV/G pseudotyped lentiviral vectors pLKO5-sgRNA-EFS-GFP encoding sgRNA Egr1 (sgEgr1 T5 and T7) or sgCTL were produced by the vectorology facility Vect’UB (TBMCore—CNRS UAR3427, INSERM US005, Universtiy of Bordeaux, Bordeaux, France). Titrations of produced lentiviral vectors were performed on the HEK293T cell line. Homozygous PE^tg mice were bred with homozygous Cas9^tg mice to obtain heterozygous PE^tgCas9^tg mice. Aberrant B cells from the BM of preleukemic PE^tgCas9^tg mice were purified by cell sorting, plated in suspension culture in B cell medium and transduced with lentiviral vectors expressing sgCTL or sgEgr1 T5 and sgEgr1 T7 at a multiplicity of infection of 30 in the presence of 4 μg/ml of polybrene (Sigma-Aldrich) for 24 h. Transduced cells were then washed in PBS1X and cocultured on MS5 stromal cells in a fresh supplemented B cell medium for 3 days. Transduced GFP⁺ cells were then purified by cell sorting to perform in vitro cell division assay and cell differentiation assay in coculture on MS5 stromal cells in a B cell medium supplemented with a low concentration of Il-7 (0.5 ng/ml) and with Flt3L (10 ng/ml) and SCF (10 ng/ml) (related to Fig. 5, E and F; and Fig. S4, F–J).

The targeted region (exon 1) of Egr1 was performed by NGS, and the genomic editing efficiency was analyzed by the proportion (%) of insertions and deletions (InDel) induced by the sgRNA Egr1 T5 and T7. Genomic DNA was extracted and purified using QIAamp DNA mini kit (Qiagen). Sequencing was performed using a MiSeq sequencer (Illumina) using a Miseq Reagent kit V2 (paired-end sequencing 2 × 150 cycles). Genome editing efficiency of the targeted Egr1 region (exon 1) was analyzed using the online CRISPResso software (https://crispresso.pinellolab.partners.org/submission; Canver et al., 2018). Targets and primers are listed in Table S1.

Human B-ALL samples were obtained from the Toulouse University Hospital (Toulouse, France) after signed written informed consent for research use in accordance with the Declaration of Helsinki and stored at the Hémopathies malignes de l’Inserm en Midi-Pyrénées (HIMIP) collection (BB-0033-00060). According to the French law, HIMIP biobank collection has been declared to the Ministry of Higher Education and Research (DC 2008-307, collection 1) and obtained a transfer agreement for research applications (AC 2008-129) after approbation by our institutional review board and ethics committee (Comité de Protection des Personnes Sud-Oust et Outremer II). Biological annotations of the samples have been declared to the Comité National Informatique et Libertés (i.e., Data Processing and Liberties National Committee). For in vitro cell division assay using CTV dilution, human B-ALL blasts (HB#002 and HB#008) were stained as previously described, and CTV-labeled cells were grown in cocultured on MS5 cell line in supplemented B cell medium and analyzed at the indicated time points in the Fig. 6 J and Fig. S5 J.

NSG mice were produced at the Génotoul-Anexplo platform in Toulouse, France, using breeders obtained from Charles River Laboratories. Mice were housed in sterile conditions using high-efficiency particulate arrestance-filtered microisolators and fed with irradiated food and sterile water. Mice (6–9 wk old) were sublethally treated with busulfan 30 mg/kg 24 h before injection of leukemic cells. Leukemic blasts from B-ALL patients were thawed at room temperature, washed twice in PBS, and transplanted by intravenous injection (2 × 10⁵ cells per mouse). At the indicated time points in Fig. 6 D and Fig. S5 C, leukemia engraftment was monitored by FACS with hCD45 (557748; BD Bioscience) and hCD19 (562294; BD Bioscience) staining after BM punctions. For combination therapy, a chemotherapeutic cocktail composed of DEXA (10 mg/kg) and VCR (0.5 mg/kg) was administered weekly for 3 wk (HB#010; Fig. 6 C) and for 1 wk (HB#007; Fig. S5 B) via intraperitoneal injection. A daily administration of 10 mg/kg DEXA was supplemented during the period of treatment. Mice were euthanized and analyzed after the treatment.

RNA was isolated using the Trizol method, and cDNA was synthesized using SuperScript VILO cDNA Synthesis Kit (Invitrogen) according to the manufacturer’s instructions. Quantitative SYBR Green PCR was performed on a LightCycler480 II System (Roche) to quantify human EGR1 cDNAs expression in Fig. 6, H and K; and Fig. S5, G and K using LightCycler480 SYBR Green I Master (Roche Diagnostics GmbH) according to the manufacturer’s instructions. All PCR were carried out as follows in a 20 μl volume: 5 min at 95°C, followed by 45 cycles of 10 s at 95°C, 10 s at annealing temperature of 60°C and 10 s at 72°C. Quantification was performed using the ΔCt method with normalization to the human ABL1 gene expression levels. Data were analyzed using the LC480 software (Roche Diagnostics). The forward (Fw) and reverse (Rev) primers used were as follows: for EGR1 (Fw: 5′-CAG​CCC​TAC​GAG​CAC​CTG​AC-3′ and Rev: 5′-GTG​GTT​TGG​CTG​GGG​TAA​CT-3′); for ABL1 (Fw: 5′-TGG​AGA​TAA​CAC​TCT​AAG​CAT​AAC​TAA​AGG​T-3′ and Rev: 5′-GAT​GTA​GTT​GCT​TGG​GAC​CCA-3′).

The number of biological replicates is indicated in the relevant figure legends. Error bars for pooled replicates represent standard deviation (SD). Statistical differences were determined using a two-tailed unpaired Student’s t test for comparison of quantitative variables, assuming normality and equal distribution of variance between the different groups analyzed. All statistical analyses were performed using GraphPad Prism software, version 7 (GraphPad). A P value of <0.05 was considered statistically significant (*P < 0.05, **P < 0.01, ***P < 0.001). Survival in mouse experiments was represented with Kaplan–Meier curves (GraphPad Prism).

RNA from purified PE^tg D0–2 (n = 2) and D8–9 (n = 2) cells and from wt D8–9 (n = 2) cells was isolated with the RNeasy Plus Mini Kit (Qiagen). RNA-seq libraries were prepared using TruSeq Stranded mRNA Low Sample kit (Illumina) according to the manufacturer’s protocol, starting with 300 ng mRNA. Cluster generation and sequencing were carried out by using the Illumina NEXTSEQ 550 system and NEXTSEQ 500/550 HIGH OUTPUT KIT v2.5 (150 cycles) with a read length of 2 × 75 nucleotides according to the manufacturer’s guidelines. The expression of mRNA transcripts rearranged for the IgH and IgL loci was analyzed using Mixcr software and represented by Circos diagrams using VDJ tool software (related to Fig. 2 G).

GFP^high and GFP^low cells from the aberrant B population of preleukemic PE^tgH2B-GFP^tg mice after 1 wk of DOX chase were purified by cell sorting (Melody, BD) directly in RLT buffer, and the RNA was isolated with the RNeasy plus micro kit (Qiagen). Smart-seqv4 libraries were prepared as previously described (Picelli et al., 2014) using the Takara SMART-Seq v4 full-length transcriptome analysis kit according to the manufacture’s guidelines. Paired-end sequencing was performed on an Illumina NextSeq 500 using 2 × 75 bp reads. Low-quality reads were trimmed, and adaptor sequence was removed using fastp (v). Short reads were then mapped to mm10 genome using RNA-STAR (v2.7.8a) with doublepass parameter. Low-quality mapping and duplicated reads were removed and the remaining reads was count using featurecount (v) with default paired-end parameter.

A specific region in the exon 7 of Pax5 (Chr4:44,609,844–44,609,786), which is enriched in SNPs, was for the quantification of reads from murine Pax5 (Pax5 WT) and from human PAX5 (PAX5-ELN) (related to Fig. S4 B).

GSEA of top 500 differentially expressed genes (bulk RNA-seq EGAD00001006133) of residual human B-ALL cells after chemotherapies published in Turati et al. (2021) (corresponding to acutely treated versus untreated PDX from patient 2) was performed in preleukemic PE^tgH2B-GFP^tg (Fig. S5 A).

For the integration of the ImmGen database, the same preprocessing pipeline was used for all the ImmGen fastq (GSE109125) except that the reads of the PE^tgH2B-GFP^tg samples were trimmed to 25 bp to match the ImmGen reads length and avoid mappability bias. The ComBat function in the sva R package was used to correct the batch effect between our and ImmGen samples. Regularized log-transformed (rlog) values were calculated by DESeq2, used to calculate k-means, and perform our integrative approach.

Sorted PE^tgH2B-GFP^tg populations were prepared according to Buenrostro et al. (2013) using ATAC-Seq kit (#53150; Active-motif). The size distribution and concentration of the libraries were assessed on Tapestation with a DNA High Sensitivity kit (Agilent Technologies). Paired-end 37.5 bp sequences were generated from samples on an NextSeq 500 (Illumina), generating an average of 125 millions of reads per sample.

First, FastQC was used to assess the sequence quality. Foreign sequences removal and trimming are realized with Sickle (qual threshold 20 and length threshold 20). Sequences were mapped to the murine genome with Bowtie2 (2.4.2) (Langmead and Salzberg, 2012) with -X 2,000 (maximal fragment length), very sensitive, and against mm10. Next, we performed various cleaning steps according to Berest et al. (2019): in brief, removing mitochondrial reads, filtering reads with mapping quality <20, removing duplicate reads, and adjusting read start sites as described previously (Buenrostro et al., 2013) (+4 −5). Lastly, a GC bias diagnosis and correction using deepTools was run for each sample. The output of this preprocessing pipeline was used to peak calling using MACS2 with -SE -200 -100 -lambdafix parameter and removal of blacklisted regions.

The complete diffTF pipeline (Berest et al., 2019) was run using TF binding sites generated by PWMscan analysis (cutoff P val: 0.00001; background base composition: 0.29;0.21;0.21;0.29) of each JASPAR2020 PWM motif to obtain the position of 644 nonredundant and specific motifs. The analytical approach was preferred due to the small size of the sample, and paired design was used for DeSeq2 parameter. For the plotting of individual activity for each TF, we used Hint-atac pipeline (Li et al., 2019) with a standard parameter on the combination of the bam for each condition to scan the JASPAR2020 database.

Fastq ATAC-seq data of interest from the ImmGen database was extracted from GSE100738 preprocessed as described previously, and the totality of the reads from all samples was normalized together using local loess normalization as described in Reske et al. (2020) with csaw package. Finally, to represent pile-up traces of the integrated data, Becorrect (Gontarz et al., 2020) was used on each bedgraph file with their corresponding normalized counts.

Peaks were associated with putative TF binding events and target genes to generate a prior network: PεR^{|genes| × |TFs|} of TF–gene interactions. We used vertebrate motifs from JASPAR2020 to scan the differential PE^tgH2B-GFP^tg peaks, running construct_atac_prior.R script (P val cutoff = 10⁻⁵; window_feature = 25 kb; background set as A = 0.29, C = 0.21, G = 0.21, T = 0.29). We built gene expression models according to Miraldi et al. (2019) and included in our study a target gene expression matrix containing the 1,625 core PE^tgH2B-GFP^tg genes for a total of 26 samples: LT-HSC (n = 2), MPP1 (n = 2), ST-HSC (n = 2), MPP4 (n = 2), CLP (n = 2), FrA (n = 2), FrBC (n = 2), FrE (n = 2), H2B^high (n = 5), and H2B^low (n = 5). We considered 275 potential TF regulators corresponding to most variable TFs across the samples. The parameters used were moderate prior reinforcement (λbias = 0.05), 50 subsamples of size 0.90 × |samples|, and an instability cutoff of 0.05 to solve the TF–gene interactions. The PE^tgH2B-GFP^tg TRN size was set to an average of 15 TFs per gene and resulted in a total of 24,275 TF–gene interactions.

Fig. S1 describes the gating strategy of the different B cell subpopulations from the BM of wt and PE^tg preleukemic mice, and the expression of intracytoplasmic ELN and of Sca-1 surface marker are explored. Fig. S2 displays the immunophenotype of PE^tg B cells after serial transplantations; the survival curve of quaternary recipient mice after transplantation of PE^tg B-ALL cells; the cell-cycle status (Ki67) of wt and PE^tg preleukemic B cells in steady state and in response to chemotherapy; the expression of Kit in CTV-labeled wt and preleukemic PE^tg cells; the expression of PAX5-ELN in PE^tg D0–2, PE^tg D8–9, and wt D8–9 cells; the dose-response of chemotherapy on wt and preleukemic PE^tg Kit⁺ cells; and the immunophenotype of engrafted cells in recipient mice transplanted with D0–2 and D8–9 preleukemic PE^tg cells. Fig. S3 shows FACS analysis of pStat5/pFoxo1 and of pBlnk/pPlcγ2 on PE^tg B cells after IL-7 and H₂O₂ stimulation respectively; signaling pathways that are up- and down-regulated in PE^tg D0–2 and PE^tg D8–9 cells; the kinetic of GFP expression in the different B cell subsets from wt H2B-GFP^tg; and preleukemic PE^tgH2B-GFP^tg mice after DOX chase, the survival curve of recipient mice transplanted with purified GFP^high and GFP^low cells from PE^tgH2B-GFP^tg mice and the heatmaps of signaling pathways up- and downregulated in GFP^high from RNA-seq and ATAC-seq data. Fig. S4 shows the accessibility footprints (ATAC-seq) and the expression levels (RNA-seq) of Ebf1, Spi1, Runx1, and Pax5 in GFP^high and GFP^low cells from PE^tgH2B-GFP^tg mice; GSEA highlighting the quiescent state of GFP^high cells; the transduction efficiency of preleukemic PE^tgCas9^tg B cells with sgCTL, sgEgr1 T5, and sgEgr1 T7 lentiviral vectors combined to the genomic editing efficiency of the Egr1 locus; the impact of Egr1 editing on the cell division and proliferation of preleukemic cells; and the molecular network of genes involving Egr1 from TRN#1. Fig. S5 displays the molecular similarities between GFP^high cells and residual human B-ALL cells after chemotherapies and shows EGR1 activation in quiescent and therapy-resistant human B-ALL. Table S1 contains the list of FACS antibodies and the sequence of oligonucleotide primers used in the study. Table S2 contains the RNA-seq data of PE^tg D0–2, PE^tg D8–9, and Wt D8–9 cells. Tables S3 and S4 contain the RNA-seq data and the ATAC-seq data of GFP^high and GFP^low cells, respectively. Table S5 presents the clustering data of modified genes and peaks from GFP^high and GFP^low cells according to their signal in the different steps of the normal differentiation.

The RNA-seq and ATAC-seq data reported in this study (Tables S2, S3, and S4) are available at the at the Gene Expression Omnibus repository under the accession number GSE210926.

We acknowledge Pr. Tariq Enver and Pr. Javier Herrero for having shared their molecular data published in Turati et al. (2021). We acknowledge Manon Farcé from the cytometry and cell sorting facility of the Cancer Research Center of Toulouse (Institut National de la Santé et de la Recherche Médicale [INSERM] U1037) and the Anexplo/Genotoul platforms (UMS006) for technical assistance. We thank the vectorology facility Vect’UB for providing lentiviral particles and for technical support, TBMCore (Centre National de la Recherche Scientifique [CNRS] UAR3427, INSERM US005, University of Bordeaux, Bordeaux, France).

This study was supported by institutional grants from INSERM, from CNRS, the Institut National du Cancer (INCa-2020-096), the Agence Nationale de la Recherche (ANR-18-CE13-0002-01), and the Fondation ARC pour la Recherche sur le Cancer (PJA-20181207977). The team is supported by the Ligue Contre le Cancer and the associations “Laurette Fugain,” “111 des arts,” “Cassandra,” and “Constance la petite guerrière astronaute.”

Author contributions: V. Fregona designed the study, performed and analyzed the experiments, and wrote the manuscript. M. Bayet, M. Bouttier, L. Largeaud, C. Hamelle, L.A. Jamrog, N. Prade, S. Lagarde, S. Hebrard, I. Luquet, V. Mansat-De Mas, and M. Nolla performed experiments. M. Pasquet, C. Didier, A.A. Khamlichi, C. Broccardo, É. Delabesse, and S.J.C. Mancini reviewed the data and the manuscript. B. Gerby performed and analyzed experiments, conceived and supervised the project, and wrote the manuscript. All authors contributed to the final draft.