Phf6 Loss Enhances HSC Self-Renewal Driving Tumor Initiation and Leukemia Stem Cell Activity in T-ALL Free

Somatic loss-of-function mutations in the plant homeodomain factor 6 (PHF6) are highly prevalent in T-cell acute lymphoblastic leukemia (T-ALL; ref. 1) and are also present in acute myeloid leukemia (AML; refs. 2, 3). In addition, PHF6 is the causative gene mutated in Börjeson–Forssman–Lehman syndrome (BFLS; MIM#301900), a rare X-linked Mendelian disease characterized by mental retardation, obesity, hypogonadism, gynecomastia, digit abnormalities, large ears, and coarse and characteristic facial features (4).

Isolation of protein complexes has shown the interaction of PHF6 with the nucleosome remodeling deacetylase (NuRD) complex, a major chromatin regulator controlling nucleosome positioning and transcription with important roles in development, genome integrity, and cell-cycle progression (5, 6). In addition, PHF6 localizes to the nucleolus and interacts with the PAF1 transcription elongation complex (7) implicated in the control of RNA polymerase I activity and ribosomal DNA (rDNA) transcription (8), and with UBF (7, 9), a transcriptional activator in the RNA Pol I preinitiation complex, supporting a role for PHF6 in the control of ribosome biogenesis.

PHF6 mutations seem restricted to hematologic tumors, are most frequently found in tumors from male patients (1, 2), and are typically nonsense and frameshift truncating alleles resulting in complete loss of protein expression (1–3, 10, 11). In all, genetic loss of PHF6 as a result of deletions or mutations is present in about 20% of T-ALLs, in 20% to 25% of mixed phenotype acute leukemias (MPAL) with early T-cell precursor (ETP) and T/myeloid characteristics, and in 3% of AML cases (1–3, 10, 11). Interestingly, the development of pediatric T-ALL in a male patient with BFLS harboring a germline PHF6 nonsense mutation (12) and the presence of PHF6 mutations in preleukemic clonal hematopoiesis (13, 14) support a role for this tumor suppressor in leukemia initiation and hematopoietic stem cell (HSC) self-renewal, respectively.

To evaluate the potential role of PHF6 loss as a leukemia-initiating event, we analyzed the timing of somatically acquired mutations in T-ALL using Integrated Sequential Network (ISN; ref. 15) analysis of clonal evolution and mutation dynamics using whole-exome sequencing data from diagnostic and relapse leukemias. This analysis revealed that somatic mutations in PHF6 occur as early lesions in the natural history of T-ALL (P = 0.03; Fig. 1A), prompting us to evaluate a mechanistic link between the loss of Phf6, HSC function, and transformation leading to leukemia initiation. Toward this goal, we generated conditional Phf6 knockout mice (Supplementary Fig. S1A–S1C) and crossed them with a Vav-iCre line to inactivate Phf6 in fetal HSCs. Analysis of 8-week-old Phf6^−/YVav-iCre animals revealed an expansion of total immature hematopoietic LSK progenitors (Fig. 1B–D) resulting from increased numbers of multipotent MPP2 and MPP3 populations compared with controls (Phf6^+/YVav-iCre; Fig. 1B–G) and no differences in more mature myeloid (MEP, CMP, and GMP) and uncommitted lymphoid (CLP) bone marrow progenitors (Supplementary Fig. S1D–S1G). Phf6 knockout mice showed no significant differences in bone marrow B-cell precursors (Supplementary Fig. S2A and S2B), and analysis of thymic populations revealed only a modest but significant reduction of double-negative DN2 and DN3 thymic progenitors (Supplementary Fig. S2C–S2G).

The observed expansion of LSK cells in hematopoietic-specific Phf6 knockout mice further supports a potential role for PHF6 in the control of HSC activity. To test this possibility, we evaluated HSC function in mixed bone marrow chimera serial transplantation assays. Primary transplants of Phf6 knockout (Phf6^−/YVav-iCre) cells showed improved early hematopoietic reconstitution 5 and 10 weeks after transplant, yet equivalent long-term total lymphoid and myeloid regeneration capacity 15 weeks after transplant compared with Phf6^+/YVav-iCre wild-type controls (Fig. 1H; Supplementary Fig. S3A and S3B). In secondary transplants, Phf6 knockout (Phf6^−/YVav-iCre)cells showed increased total hematopoietic reconstitution and pronounced improvement in both lymphoid and myeloid-lineage repopulation compared with controls (Fig. 1H; Supplementary Fig. S3A). This phenotype was further accentuated in tertiary transplants, in which Phf6 knockout cells markedly outcompeted their Phf6 wild-type counterparts (Fig. 1H). In these experiments, Phf6 wild-type serially transplanted HSCs showed a myeloid bias and decreased ability to generate B cells (Supplementary Fig. S3A and S3C), features characteristically associated with HSC aging and functional exhaustion (16, 17). In contrast, HSCs from hematopoietic-specific Phf6 knockout mice effectively contributed to both lymphoid and myeloid reconstitution in tertiary transplants, demonstrating preservation of lymphoid-lineage differentiation potential (Supplementary Fig. S3A and S3C).

Most somatically acquired mutations in T-ALL presumably occur postnatally in the adult hematopoietic system. To specifically evaluate the effects of Phf6 deletion in adult HSC function, we analyzed the hematopoietic system of recipient mice engrafted with bone marrow cells from control wild-type (tamoxifen-treated Phf6^f/Y Rosa26^+/+; referred to simply as Phf6^f/Y Rosa26^+/+) and adult Phf6 knockout mice (tamoxifen-treated Phf6^−/Y Rosa26^+/Cre-ERT2; hereafter referred to as Phf6^–/Y Rosa26^+/Cre-ERT2). In this setting, loss of Phf6 in adult HSCs resulted in expansion of immature hematopoietic LSK progenitors, resulting from increased numbers of MPP2 and MPP3 multipotent progenitors, as well as lymphoid-committed MPP4 and CLP populations (Fig. 2A and B; Supplementary Fig. S4A and B). Furthermore, we also observed a modest increase in CMP-committed myeloid bone marrow progenitors, with no changes in MEP and GMP populations (Supplementary Fig. S4C and S4D). We observed no differences in cell-cycle distribution between wild-type and Phf6 knockout–derived HSCs (LSK CD48⁻) and MPP cells (LSK CD48⁺; Supplementary Fig. S4E and S4F). Animals transplanted with Phf6^−/Y Rosa26^+/Cre-ERT2 bone marrow showed largely preserved B-cell development, with only a modest decrease in pre-B and immature B-precursor cell populations (Supplementary Fig. S5A and S5B). In addition, and as noted in Phf6^−/Y Vav-iCre knockout mice, analysis of the thymus in mice reconstituted with Phf6^−/Y Rosa26^+/Cre-ERT2 cells revealed a modest but significant reduction in DN2 and DN3 thymic progenitor cells (Supplementary Fig. S5C–S5G). Consistently, mixed bone marrow chimera serial transplantation assays showed that Phf6^−/Y Rosa26^+/Cre-ERT2 adult HSCs display improved competitive multilineage reconstitution, increased serial repopulation capacity (Fig. 2C), and successively enhanced retention of lymphoid potential in secondary and tertiary transplants compared with Phf6^f/Y Rosa26^+/+ controls (Supplementary Fig. S6A–S6C). Moreover, as a result of exhaustion of HSC function in these experiments, we noted a progressive decrease in the number of LSK cells with each consecutive transplantation of Phf6^f/Y Rosa26^+/+ wild-type bone marrow (Fig. 2D), which was accompanied by an increase in the relative frequency of long term HSCs (LT-HSC) as a result of decreased hematopoietic output (Fig. 2E–H).

In comparison, secondary and tertiary recipients of Phf6^−/Y Rosa26^+/Cre-ERT2 knockout cells had a less severe and delayed decline in LSK numbers (Fig. 2D), showed preservation of hematopoietic function, improved MPP output, and lacked the characteristic increase in the relative frequency of LT-HSCs observed in mice serially transplanted with wild-type cells (Fig. 2E–H; Supplementary Fig. S6D and S6E). In addition, in-depth analysis of bone marrow MPP progenitor populations revealed preserved numbers of Phf6^−/Y knockout-derived lymphoid-restricted MPP4 cells in tertiary relative to primary recipients of adult-deleted Phf6 knockout bone marrow chimeras, a finding consistent with their distinctive long-term retention of lymphoid-lineage differentiation potential (Fig. 2G and H; Supplementary Fig. S6).

Next, to further and directly test the role of Phf6 in HSC self-renewal, we compared the reconstitution capacity of purified LT-HSCs from Phf6^−/Y Rosa26^+/Cre-ERT2 and wild-type Phf6^f/Y Rosa26^+/+ control mice after serial transplantation into isogenic recipients. In this experiment, primary recipient mice engrafted with Phf6-null LT-HSCs showed accelerated early-stage hematopoietic reconstitution at 5 weeks after transplantation. Increased long-term donor-derived reconstitution derived directly from Phf6-null LT-HSCs was maintained at 15 weeks after transplantation (Fig. 3A). Of note, equal numbers of primary transplanted Phf6-null and wild-type LT-HSCs gave rise to equivalent numbers of LT-HSCs in the bone marrow of recipient mice 16 weeks after transplant (Fig. 3B). Strikingly, Phf6-null LT-HSCs gave rise to increased numbers of Phf6-null–derived downstream progenitor cells compared with Phf6 wild-type LT-HSCs; increased numbers of Phf6-null donor-derived cells included ST-HSCs, MPP2-4 progenitor cells, and myeloid-committed MEP, CMP, and GMP progenitors (Fig. 3A and B). This phenotype was further augmented in recipients of secondary transplants performed with sorted LT-HSCs isolated 16 weeks after primary transplantation (Fig. 3C). Of note, and as observed in competitive bone marrow chimeras, Phf6 adult-deleted knockout LT-HSCs showed improved retention of B lymphoid–cell reconstitution potential compared with wild-type controls, which showed preferential myeloid-lineage differentiation (Fig. 3D). In all, these experiments indicate that loss of Phf6 results in increased LT-HSC self-renewal and hematopoietic reconstitution capacity with maintenance of long-term lymphoid-lineage commitment.

To explore the mechanisms mediating the effects of Phf6 in LT-HSCs, we performed genome-wide chromatin accessibility profiling by Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq). This analysis identified 140,485 transposase-accessible peaks in mouse LT-HSCs, of which 2,707 were significantly differentially accessible (adjusted P value: P_adj < 0.01; fold change > 4) in Phf6 knockout versus wild-type LT-HSCs. Of these significantly differentially accessible peaks, 2,334 (86%) became more accessible in Phf6 knockout LT-HSCs and 373 (14%) became less accessible, indicating a global increase in chromatin accessibility in the Phf6 knockout setting (Fig. 3E). Notably, although only 10% of the total ATAC-seq peaks identified in LT-HSCs fell within proximal promoter regions, 32% of the significantly differentially accessible peaks in Phf6 knockout LT-HSCs were located at the proximal promoter, indicating a preferential chromatin opening around transcription start sites (TSS; Fig. 3F and G). Functional enrichment analysis of promoter/TSS peaks with increased accessibility in Phf6 knockout LT-HSCs revealed significant enrichment of GO Term categories potentially related to HSC homeostasis, including cell cycle (GO:0007049, P = 9.67 × 10⁻¹²); translation (GO:0006412, P = 4.75 × 10⁻¹²) and ribosome biogenesis (GO:0042254, P = 2.49 × 10⁻⁷); MYC targets (MSIGDB:M15774, P = 5.3 × 10⁻⁹; MSIGDB:M17753, P = 1.807 × 10⁻¹⁰), and cellular response to stress (GO:0033554, P = 2.28 × 10⁻⁷). Consistently, we observed increased promoter accessibility at MYC target gene promoters in Phf6 knockout LT-HSCs (Fig. 3G). In addition, gene-expression profiling by RNA sequencing (RNA-seq) of sorted Phf6 wild-type and knockout LT-HSCs revealed modest changes in gene expression with 306 differentially expressed genes (P < 0.01; log₂ fold change > 0.3), of which 150 were upregulated and 156 were downregulated in Phf6 knockout LT-HSCs, respectively (Fig. 3H). Notably, this transcriptional program was significantly enriched in genes controlled by JAK1 in HSCs (ref. 18; Fig. 3I), supporting a potential convergent transcriptional mechanism between genetic inactivation of Phf6 and JAK–STAT signaling in HSC self-renewal.

Under homeostatic conditions, LT-HSCs are predominantly quiescent, a property characteristically required to maintain their self-renewal capacity and functionally protective under stress (19). However, following hematopoietic challenges such as genotoxic insult by radiation or chemotherapy, HSCs transiently proliferate to ensure efficient replenishment of blood and immune cellularity (19). Similarly, inflammatory antiviral responses and interferon signaling also induce the activation and proliferation of LT-HSCs, and their return to quiescence is essential to avoid accumulation of DNA damage and apoptosis (20). Loss of homeostatic equilibrium in the hematopoietic system in the context of chemotherapy or immune challenge has been implicated in the expansion of HSCs harboring clonal hematopoiesis-associated mutations and in leukemia development (21). In this context, we explored the capacity of Phf6-null HSCs to withstand stress-induced activation by examining the kinetics of recovery after myeloablative treatment with 5-fluorouracil (5-FU) compared with wild-type controls (22). In these experiments, mice reconstituted with Phf6^−/Y Rosa26^+/Cre-ERT2 knockout cells showed accelerated and improved expansion of LSK cells and total myeloid progenitors (MyP; Lin⁻ CD117⁺ Sca1⁻) on days 9 and 12 after treatment compared with wild-type Phf6^f/Y Rosa26^+/+ controls (Fig. 4A and B; Supplementary Fig. S7). In addition, recipients of Phf6 knockout bone marrow showed improved recovery of total bone marrow cellularity, yet lower levels of activation-induced expansion in the HSC compartment (LSK CD48⁻), following genotoxic ablation of hematopoietic progenitors with cyclophosphamide and rescue with G-CSF cytokine stimulation (ref. 23; Fig. 4C–E). To more directly explore a potential cell-intrinsic role of Phf6 loss in HSC quiescence, we tested the effects of interferon stimulation on HSC activation by polyinosinic:polycytidylic acid [poly (I:C)] treatment of mice reconstituted with control Phf6 wild-type (Phf6^f/Y Rosa26^+/+) and Phf6 knockout (Phf6^−/Y Rosa26^+/Cre-ERT2) bone marrow 8 weeks after transplant, a time frame in which the hematopoietic system has returned to homeostatic equilibrium. Consistently, wild-type and Phf6 knockout–derived HSCs (LSK CD48⁻) and MPP cells (LSK CD48⁺) showed no differences in cell-cycle distribution in basal conditions (Supplementary Fig. S4E and S4F). In contrast, and most notably, 24 hours after poly (I:C) administration, Phf6 knockout HSCs (LSK CD48⁻) showed a more limited proliferative response and improved retention of the quiescent G0 cell-cycle state compared with wild-type controls (Fig. 4F and G). Conversely, unrestrained cell-cycle reentry was observed in Phf6 knockout multipotent progenitor cells (LSK CD48⁺; Fig. 4F and G). Taken together, these results support that the loss of Phf6 limits the proliferative response of LT-HSCs in reaction to genotoxic stress and facilitates hematopoietic reconstitution after injury.

To formally and directly test the role of Phf6 loss as a leukemia-initiating event in T-ALL transformation, we infected hematopoietic progenitor cells from Phf6 knockout (Phf6^−/Y Vav-iCre) and wild-type (Phf6^+/YVav-iCre) controls with bicistronic GFP retroviruses driving expression of a constitutively active T-ALL–mutant form of the NOTCH1 receptor (NOTCH1-L1601P-ΔPEST; refs. 24, 25). Of note, activating mutations in NOTCH1 are highly prevalent in T-ALL (24) and frequently co-occur with mutational loss of PHF6 (1). In this model, forced expression of mutant NOTCH1 typically results in a wave of extrathymic T-cell lymphopoiesis 3 weeks after transplantation, which manifests as a transient appearance of GFP⁺ preleukemic CD4⁺ CD8⁺ double-positive (DP) T cells in peripheral blood, and is followed by full T-ALL leukemia transformation 10 to 15 weeks later (25). Mice transplanted with Phf6 knockout NOTCH1-L1601P-ΔPEST–expressing progenitors showed a consistent wave of GFP⁺ circulating cells in peripheral blood 3 weeks after transplant (Fig. 5A). Nonetheless, analysis of the thymic compartment in this setting revealed increased numbers of GFP⁺ cells in mice transplanted with Phf6 knockout NOTCH1-L1601P-ΔPEST–transduced bone marrow progenitor cells (Fig. 5B). Moreover, analysis of circulating tumor cells 5 and 11 weeks after transplant revealed accelerated leukemia development in mice transplanted with Phf6 knockout NOTCH1-L1601P-ΔPEST–expressing progenitors (Fig. 5C), which ultimately translated to increased leukemia penetrance and markedly accelerated leukemia-associated mortality in this group (Phf6 wild-type NOTCH1-L1601P-ΔPEST median survival = 23.6 weeks; Phf6 knockout NOTCH1-L1601P-ΔPEST median survival = 13.3 weeks; log-rank test P = 0.042; Fig. 5D). Interestingly, flow cytometry analysis of hematopoietic tissues at endpoint showed increased GFP⁺ leukemia lymphoblast infiltration in the thymus of animals bearing Phf6 knockout leukemia, despite nearly equivalent tumor burden in the bone marrow and spleen (Fig. 5E and F). Moreover, analysis of leukemia lymphoblasts recovered from leukemia-infiltrated tissues showed expression of T-cell surface markers with increased prevalence of a CD4⁺ CD8⁺ double-positive immunophenotype in Phf6 knockout tumors compared with Phf6 wild-type controls (Fig. 5G). Western blot analysis confirmed complete loss of PHF6 expression in Phf6 knockout tumor cells (Fig. 5H).

To better explore the specific role of Phf6 in the pathogenesis of T-ALL without the confounding effect of genetic and immunophenotypic heterogeneity of this model, we developed isogenic tumors with loss of Phf6. Toward this goal, we generated NOTCH1-L1601P-ΔPEST–induced leukemias as described above using hematopoietic progenitors from tamoxifen-inducible conditional Phf6 knockout (Phf6^f/YRosa26^+/CreERT2) mice. The resulting Phf6 conditional knockout leukemias were transplanted into secondary recipients and treated with vehicle only or tamoxifen to generate matched Phf6 wild-type and knockout tumors, respectively (Supplementary Fig. S8A–S8C). When transplanted into isogenic recipients, Phf6 knockout leukemia cells showed modestly accelerated tumor progression compared with Phf6 wild-type isogenic controls as documented by increased GFP⁺ cell counts in peripheral blood (Fig. 6A) and accelerated mortality (log-rank test P = 0.013; Fig. 6B). However, and most notably, limiting dilution transplantation assays demonstrated a marked increase in leukemia-initiating cell activity in Phf6 knockout tumors compared with wild-type isogenic controls, supporting a role for loss of Phf6 in promoting leukemia blast self-renewal (Fig. 6C and D; Supplementary Fig. S8D).

Analysis of transcriptional programs associated with Phf6 loss in matched isogenic Phf6 wild-type and Phf6-null leukemia samples revealed a distinct gene-expression signature dominated by upregulated transcripts (608 genes upregulated and 67 genes downregulated; P < 0.001, fold change > 2;Fig. 6E). Genes upregulated in Phf6-null leukemia lymphoblasts included the early hematopoietic marker Cd34, as well as genes associated with B-cell (Cd19, Pax5) and myeloid (Kit, Cd14, Cd38, Flt3, Csf3r, Fcgr1) differentiation, even though Phf6 wild-type and Phf6 knockout isogenic tumors showed no apparent differences in their immunophenotype (Fig. 6F). Notably, suppression of Phf6 in mouse B-cell lymphoma cells has been described to increase lineage plasticity, in this case resulting in decreased expression of B cell–lineage genes and upregulation of T-cell markers (26), and PHF6 mutations are prevalent in ETP ALLs and myeloid/T MPALs, which are characterized by mixed expression programs and lineage plasticity (1–3, 10, 11). In addition, genes upregulated upon Phf6 knockout in our NOTCH1 leukemia model include bona fide T-ALL oncogenes such as Lmo2, Mef2c, Zeb2, and Hex (ref. 27; Fig. 6F), which are characteristically expressed at higher levels in early immature ETP ALL leukemias (28, 29). In addition, and in agreement with an increase in leukemia-initiating cell activity, gene set enrichment analysis showed preferential activation of a gene-expression program associated with human leukemia stem cell activity (30) in Phf6 knockout NOTCH1-induced T-ALL cells compared with isogenic wild-type controls (Fig. 6G). Finally, and most notably, analysis of this leukemia-initiating cell signature in human T-ALLs validated and extended these results, showing a marked enrichment of leukemia stem cell genes (31) in human PHF6-mutant leukemias compared with matched PHF6 wild-type controls (Fig. 6H).

Chemoresistant leukemia-initiating cells with expanded self-renewal potential are major drivers of leukemia relapse and are associated with poor prognosis (32). Self-renewal potential and intrinsic resistance to therapeutic agents are characteristic properties of HSCs, and transcriptional profiling of human leukemia-initiating cell populations has established a link between the transcriptional programs operating in HSCs and those present in leukemia stem cells (33). These observations are in agreement with data that support HSCs representing the cell of origin of many human leukemias and further support an overlap between the circuitries governing HSC self-renewal and those operating in leukemia-initiating cell populations.

Here, we identify PHF6 as a new leukemia-initiating tumor suppressor gene implicated in the control of long-term self-renewal in HSCs and as a driver of leukemia-initiating cell activity in T-ALL. Mutations that increase the self-renewal capacity of hematopoietic progenitors are postulated to play important roles in leukemia initiation and as potential drivers of leukemia-initiating cell activity (21, 34). In this context, DNMT3A and TET2, two epigenetic factors governing HSC self-renewal, have emerged as prominent tumor suppressors frequently mutated in clonal hematopoiesis and as early events in the pathogenesis of AML and myelodysplastic syndromes (21, 34). As in the case of DNMT3A and TET2, PHF6 mutations can be found in clonal hematopoiesis associated with aging (14) and following recovery from aplastic anemia (13), and Phf6 inactivation increases the long-term self-renewal capacity of HSCs. However, although DNMT3A and TET2 both regulate DNA methylation, PHF6 is a nucleolar protein implicated in transcriptional control, ribosome biogenesis, and the regulation of nucleosome positioning via interaction with the NuRD nucleosome remodeling complex (35). Notably, analysis of Phf6 wild-type and knockout mice showed increased chromatin accessibility in Phf6-null LT-HSCs preferentially involving MYC targets and genes implicated in cell cycle, translation, and response to stress, in support of a chromatin remodeling function of PHF6 in HSC homeostasis. In addition, gene-expression profiling showed inverse effects of Phf6 inactivation and Jak1 knockout in LT-HSCs, potentially indicating a convergent mechanism of action between PHF6 inactivation and JAK–STAT signaling in promoting HSC self-renewal.

Clonal evolution mapping analysis supports that PHF6 mutations may occur early in T-ALL transformation, and we demonstrate experimentally that Phf6 loss can enhance the oncogenic activity of constitutively active mutant NOTCH1. Increased penetrance of NOTCH1-induced leukemia from Phf6-null cells may result from a priming effect resulting from increased chromatin accessibility in MYC target genes and in genes involved in cell growth and protein biosynthesis, programs that are tightly linked to T-ALL transformation and directly controlled by NOTCH1 (25, 36, 37). In addition, and most notably, secondary deletion of Phf6 in NOTCH1-induced leukemia cells increased leukemia-initiating cell activity, a phenotype linked to the effects of PHF6 inactivation in human leukemia by the enrichment in both mouse Phf6-null tumors and in PHF6-mutant human T-ALLs of a human leukemia-initiating cell transcriptional program. On the basis of these observations, we propose that loss of PHF6 favors leukemia initiation by enhancing the self-renewal capacity of HSCs and lowering the threshold for oncogenic transformation in hematopoietic progenitors, and postulate a role for PHF6 mutations as early drivers of leukemia-initiating cell activity in T-ALL.

Although the specific molecular functions of the PHF6 protein remain to be fully established (35), the association of PHF6 loss with increased chromatin accessibility warrants further evaluation of the role of PHF6 regulation of NuRD activity in HSCs and leukemia. Further studies will also characterize the interaction of Phf6 inactivation with leukemia oncogenes other than NOTCH1 and evaluate the role and mechanisms of Phf6 mutations in myeloid and mixed-lineage leukemias. Finally, analysis of Phf6 loss in the hematopoietic system of female mice will help clarify the mechanisms responsible for the observed male predominance of PHF6 mutations in human leukemia.

DNA from leukemic ALL blasts at diagnosis and relapse and matched remission samples were provided by the Hemato-Oncology Laboratory at University of Padua, Italy, the Children's Oncology Group, and the Eastern Cooperative Oncology Group (ECOG). Samples were collected under the supervision of local Institutional Review Boards for participating institutions and analyzed under the supervision of the Columbia University Medical Center Institutional Review Board (protocol number: IRB-AAAB3250). All samples were obtained with written informed consent at study entry.

All animals were housed in specific pathogen-free facilities at the Irving Cancer Research Center at the Columbia University Medical Center in accordance with NIH guidelines for the Care and Use of Laboratory Animals. All animal procedures were approved by the Columbia University Institutional Animal Care and Use Committee. To generate Phf6 conditional knockout mice, we used homologous recombination in C57BL/6 embryonic stem cells to introduce a LoxP site upstream of exon 4 in introns 3 to 4, and the LoxP/FRT-flanked Neo selection cassette downstream of exon 5 in introns 5 to 6. We generated chimeras in C57BL/6 albino blastocysts using three independent targeted embryonic stem cell clones identified by PCR analysis and verified by southern blot. We verified germline transmission in the offspring of highly chimeric male mice crossed with C57BL/6 females. To remove the neomycin selection cassette, we crossed mice harboring the targeting construct with an Flp germline deleter line [Gt(ROSA)26Sor^tm1(FLP1)Dym/RainJ; The Jackson Laboratory] and crossed the resulting mice with C57BL/6 to breed out the Flp allele. To generate conditional Phf6 knockout mice, we bred animals harboring the conditional targeted allele with Vav-iCre^+/− animals (B6.Cg-Commd10^{Tg(Vav1-icre)A2Kio}/J; The Jackson Laboratory; ref. 38) expressing the Cre recombinase under control of the pan-hematopoietic Vav1 oncogene promoter. To generate conditional inducible Phf6 knockout mice, we bred animals with Rosa26^+/Cre-ERT2 mice [Gt(ROSA)26Sor^{tm1(cre/ERT2)Thl}], which express a tamoxifen-inducible form of the Cre recombinase from the ubiquitous Rosa26 locus (39). Rosa26^+/Cre-ERT2 mice were a kind gift from T. Ludwig. Primers used for genotyping are provided in the Supplementary Methods section (Supplementary Table S1).

Phf6^−/YVav-iCre and Phf6^+/YVav-iCre male mice were used in experiments at 7 to 9 weeks of age. Phf6 inactivation in age-matched Phf6^f/YRosa26^+/Cre-ERT2 and Phf6^f/YRosa26^+/+ control mice was induced at 5 to 6 weeks of age by intraperitoneal administration of 1 mg/day tamoxifen for 3 consecutive days. Mice were used in experiments a minimum of 8 weeks after tamoxifen treatment.

Male C57BL/6J mice (6 weeks old) were purchased from The JacksonLaboratory (cat. #000664) and used as recipients in all mouse leukemia model experiments.

Male CD45.1 mice (B6.SJL-Ptprc^a/BoyAiTac) were purchased from Taconic Farms (cat. #4007) and were used as recipients in noncompetitive transplantation experiments and as both bone marrow donors and congenic recipients in competitive transplantation experiments.

In noncompetitive bone marrow transplantation experiments, 0.5 million lineage-negative bone marrow cells isolated from Phf6^−/Y Rosa26^+/Cre-ERT2 and Phf6^+/YRosa26^+/Cre-ERT2 bone marrow donor mice (CD45.2; 6–8 weeks old) 2 weeks after tamoxifen treatment were transplanted into the retro-orbital sinus of lethally irradiated (990 Rad) congenic CD45.1 recipients. All experimental procedures were performed 8 to 10 weeks after bone marrow transplantation. Competitive transplants were performed using 0.5 million lineage-negative Phf6^−/YRosa26^+/Cre-ERT2 and Phf6^+/YRosa26^+/Cre-ERT2–derived bone marrow cells 2 weeks after tamoxifen treatment (as above) combined with 0.5 million lineage-negative bone marrow cells derived from age-matched CD45.1 congenic mice. Lethally irradiated (990 Rad) 7- to 8-week-old congenic CD45.1 male mice served as primary mixed bone marrow recipients. Serial transplantation was performed after 24 weeks by transplanting 3 million total bone marrow cells from primary mixed bone marrow chimeras into secondary recipients, and from secondary mixed bone marrow chimeras into tertiary recipients. Donor frequency and lineage commitment were assessed every 4 weeks by flow-cytometric analysis of peripheral blood.

To generate NOTCH1-induced T-ALL tumors in mice, we performed retroviral transduction of lineage negative–enriched bone marrow cells (Lineage Cell Depletion kit, Miltenyi Biotec; cat. #130-090-858) with activated forms of the NOTCH1 oncogene (NOTCH1-L1601P-ΔPEST) tagged with GFP and transplanted them into the lethally irradiated (990 Rad) isogenic recipients (40). T-ALL tumor development was assessed by monitoring CD4⁺ CD8⁺ GFP⁺ cells in peripheral blood by flow cytometry.

To generate isogenic Phf6 wild-type and Phf6 knockout tumors, we generated NOTCH1-L1601P-ΔPEST–induced leukemia using Phf6^f/YRosa26^+/CreERT2 donor cells as before (40). In all subsequent in vivo studies, 1 million Phf6^f/YRosa26^+/Cre-ERT2 T-ALL tumor cells were transplanted into primary sublethally irradiated (450 Rad) congenic C57BL/6 recipients by retro-orbital injection. After 48 hours, mice received either 3 mg tamoxifen or an equivalent volume of corn oil by intraperitoneal injection. Resulting isogenic Phf6 wild-type (Phf6^f/Y Rosa26^+/Cre-ERT2) and Phf6 knockout (Phf6^−/YRosa26^+/Cre-ERT2) tumor cells recovered from primary recipient mice were transplanted via retro-orbital injection into isogenic secondary recipients for analysis of tumor progression (10,000 tumor cells/recipient; n = 10–15 animals per group) and leukemia-initiating cell activity (10,000–1 tumor cells/recipient; n = 6 animals per group).

Tamoxifen (cat. #T5648) and 5-FU (cat. #F6627) were purchased from Sigma-Aldrich. Tamoxifen was dissolved in corn oil at 55°C to a stock concentration of 30 mg/mL and stored at −20°C. In all leukemia model experiments, a single 3-mg dose of tamoxifen in 100 μL was administered by intraperitoneal injection. Vehicle-treated animals received 100 μL of corn oil. 5-FU was dissolved in PBS at 65°C to a final concentration of 15 mg/mL and administered by intraperitoneal injection at a dose of 150 mg/kg. A 1 mg/mL working solution of polyinosine-polycytidylic acid high-molecular-weight [poly (I:C); Invivogen; cat. #11C21-MM] was prepared in physiologic saline solution (NaCL 0.9%) and was administered at a dose of 5 μg/g body weight by intraperitoneal injection.

Western blot analysis was performed using standard protocols. Rabbit antibody against PHF6 (Sigma-Aldrich; cat. #HPA001023) was used at a 1:1,000 dilution, and rabbit anti-GAPDH antibody (Cell Signaling, cat. #2118) was used at a 1:10,000 dilution.

Single-cell suspensions of total bone marrow cells were prepared by crushing the long bones and passing cells through a 40-μm filter. For experiments involving HSC sorting, bone marrow cells were additionally isolated from the illia and spinal vertebrae. Single-cell suspensions of soft tissues were obtained by pressing the tissue through a 40-μm filter. All cell suspensions were prepared in PBS supplemented with 2% FCS, 1 mmol/L Hepes, and 0.5 μmol/L EDTA. Erythrocytes were eliminated using ACK lysing buffer. Absolute quantification of cell number was performed using CountBright absolute Counting Beads (Invitrogen; cat. #MP36950). B-cell progenitor populations were defined as follows: pre–pro-B (B220⁺ IgD⁻ CD19⁻), pro-B (B220⁺ IgD⁻ CD19⁻ IgM⁻ CD25⁻), pre-B (B220⁺ IgD⁻ CD19⁻ IgM⁻ CD25⁺), immature B (B220⁺ IgD⁻ CD19⁺ IgM⁺ CD25⁻), and total B cells (B220⁺ IgD⁻). A full list of fluorochrome-conjugated anti-mouse antibodies is included in the Supplementary Methods section (Supplementary Table S2).

Flow cytometry analysis was performed on a FACSCantoII, LSRII, and FortessaLSRII (BD Biosciences) and analyzed using FlowJo Treestar software (FlowJo LLC). Fluorescence-activated cell sorting was performed on a BD Influx cell sorter.

After staining of surface antigens, bone marrow cells were fixed using the Cytofix Fixation/Permeabilization Kit (BD Biosciences; cat. #554714) according to the manufacturer's instructions. Cells were stained with anti-Ki67 antibody (1:30 dilution) overnight on ice. One hour prior to analysis, cells were incubated with 1.62 μmol/L Hoechst 33342 (Invitrogen; cat. #H3570) at room temperature. Acquisition was performed using a BD LSRII (BD Biosciences) flow cytometry analyzer.

Mice were injected intraperitoneally with 200 mg/kg of cyclophosphamide (Cy; Calbiochem; cat. #239785) and then on succeeding days with 5 μg/day recombinant human G-SCF (Amgen; cat. #55513-924-10) administered by subcutaneous injection. The day of cyclophosphamide treatment was considered day −1 and the first day of G-CSF treatment was counted as day 0 (e.g., mice sacrificed on day 4 of the mobilization protocol were sacrificed on the day after the fourth G-CSF injection, designated as D4).

LT-HSCs (defined here as Lin⁻ Sca1⁺ CD117⁺ CD135⁻ CD150⁺ CD48⁻) were sorted from lineage-depleted bone marrow cells isolated from wild-type (Phf6^f/YRosa26^+/+) and Phf6 knockout (Phf6^−/YRosa26^+/Cre-ERT2) mice 8 weeks after tamoxifen treatment. Cell sorting was performed using a BD Influx cell sorter (BD Biosciences).

ATAC sequencing was performed as previously described (41) in 20,000 LT-HSCs sorted from three individual Phf6 knockout (Phf6^−/Y Rosa26^+/Cre-ERT2) and Phf6 wild-type (Phf6^f/Y Rosa26^+/+) mice 8 weeks after tamoxifen treatment. Nuclei were isolated and tagmented using Nextera Tn5 Transposase with single indexed, library-specific sequencing adapters. ATAC-seq libraries were quantified based on the average fragment size according to Bioanalyzer results, pooled at equimolar concentrations, and sequenced as a pool to a total read-depth target of approximately 7 × 10⁷, paired-end, 75-nucleotide-long reads per sample on an Illumina NextSeq 500. BCL files were demultiplexed, and FASTQ files were generated on the BaseSpace platform (Illumina). Reads were trimmed of contaminating adapter sequences using cutadapt and aligned to the GRCM38-build mouse genome using Bowtie2 in very sensitive mode. Peaks of transposase-accessible chromatin were called using MACS2 v2.1.1 (42) in SPMR mode on each sample independently and were subsequently centered by symmetrically extending peak summits in the 5′ and 3′ direction by 200 nucleotides. Overlapping peaks were merged into single regions, and the set of all nonoverlapping peaks was used for further analysis. We used the featureCounts method from the Rsubread v1.32.1 package to quantify the accessibility of each peak in all samples. Statistical analysis of differentially accessible regions was performed using total library size normalization and a negative binomial model in R using the DESeq2 1.22.1 package (43). We annotated ATAC-seq accessible peaks according to their nearest genomic feature using the annotatePeaks script from the HOMER package v4.10. Gene ontology analysis was also performed for all regions with significantly increased accessibility using HOMER. Enrichment of gene signatures was calculated using the hypergeometric distribution. Genome-wide distribution of chromatin accessibility was plotted using the Circlize v0.4.4 package in R. Global proximal promoter accessibility was plotted with deeptools, 2 Kb 5′ and 3′ to (and centered at) the TSS of all protein-coding transcripts annotated using the RPGC normalized signal for all protein-coding genes in the GRCm38.p6 comprehensive gene annotation from Gencode.

For RNA-seq analysis of isogenic Phf6 wild-type and knockout mouse T-ALL cells, RNA was extracted from 10⁶ cells (n = 3 per group) using the RNeasy Mini Kit (Qiagen; cat. #74106). cDNA synthesis and subsequent library synthesis were performed using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs; cat. #E7530). For RNA-seq analysis of Phf6 wild-type and knockout LT-HSCs, cells were sorted directly into buffer RLT Plus, and total RNA was isolated using the RNeasy Plus Micro Kit (Qiagen; cat. #74034). First-strand cDNA synthesis and full-length ds cDNA amplification were performed using the SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing (Clontech Laboratories; cat. #634888). Sequencing-ready libraries were generated using the Low Input Library Prep Kit v2 (Clontech Laboratories; cat. #634899). The Agencourt AMPure XP kit was used to purify cDNA (Beckman Coulter; cat. #A63880). Libraries were quantified using the Qubit dsDNA HS Assay Kit (Invitrogen; cat. #Q32854) and by quantitative PCR using KAPA Library Quantification Kit (Kapa Biosystems; cat. #KR0405). Libraries were pooled in equimolar concentrations and sequenced on an Illumina NextSeq 500 to a total target depth of 10 million 150-bp paired-end reads per sample. All protocols were performed following the manufacturer's instructions.

Raw reads were demultiplexed, and FASTQ files were generated using bcl2fastq v2.20.0.422 (Illumina). Reads were aligned to a hybrid mouse (GRCm38.p6)-ERCC genome using STAR v2.3.5.3a, and gene-count matrices were computed using featureCounts v1.6.0 (Rsubread). The removal of unwanted variance (RUV) method (44) was used to normalize based on the synthetic spike-in controls (Ambion; cat. #4456740) in RUVseq v1.14.0. Differential expression analysis was performed using the negative binomial distribution to estimate the variance-mean dependence and to test for statistical significance in DESeq2 v1.20.0 (43).

Gene set enrichment analysis (45) was performed using regularized log-transformed normalized count data in the javaGSEA program (Broad Institute). We used the Jak1 mouse knockout hematopoietic stem cell signature gene set from ref. 18 and the human 128-gene leukemia-initiating cell signature gene set from ref. 31. For analysis of mouse tumors, human gene identifiers were converted to mouse in R using the biomaRt bioconductor package. Analysis of human T-ALL tested the enrichment of the leukemia-initiating signature in PHF6 wild-type (n = 5) versus PHF6-mutant (n = 5) leukemias (46). Analysis was restricted to tumors in the TLX1 and TLX3 group to avoid the confounding effect of the association of TLX1/TLX3 translocation and overexpression with PHF6 mutations (1). Enrichment scores were computed, and 10,000 gene permutations were used to estimate P values.

RNA-seq and ATAC-seq data sets generated during this study, cited RNA-seq data sets, and microarray data sets are available in the Gene Expression Omnibus under accession numbers GSE117165, GSE85745, and GSE42328, respectively.

We illustrated the sequential order of somatic mutations in T-ALL using the ISN, which pools evolutionary paths across all patients. We selected recurrently mutated genes that were previously defined as drivers of pediatric T-ALL (27) and relapse-specific genes. Synonymous single-nucleotide variants were not included in the analysis. For each patient, we generated a sequential network that defined early events as mutations observed in both primary tumor and relapsed tumor. Each node represents a gene, and each arrow points from a gene with an early event to a gene with a late event. The ISN then pooled sequential networks across all patients. To test whether a gene within the ISN was significantly early or late, we used the binomial test based on in-degree and out-degree of each node. Specifically, the PHF6 P = 0.03 corresponds to the binomial cumulative distribution function for x = 0 arrows going into PHF6, n = 5 total trials, and probability P = 0.5 that any arrow will point toward PHF6 [binocdf(x = 0, N = 4, P = 0.5)].

Statistical analysis was performed using GraphPad Prism software v5.0 (GraphPad Software). We assumed normality and equal distribution of variance between groups, and we considered results with Student t test P < 0.05 as statistically significant. Kaplan–Meier curves were used to represent survival, where significance was calculated with the log-rank test. We analyzed serial limited dilution leukemia-initiating cell data using the ELDA online resource software (47).

J.M. Gastier-Foster reports receiving commercial research grants from Bristol-Myers Squibb and Incyte Corporation. M.L. Loh is a consultant/advisory board member for Medisix Therapeutics. R. Rabadan is a consultant/advisory board member for CNIO. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.A. Wendorff, A.A. Ferrando

Development of methodology: A.A. Wendorff, A. Ambesi-Impiombato, R. Rabadan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.A. Wendorff, A. Ambesi-Impiombato, M.S. Tallman, E. Paietta, M. Paganin, G. Basso, J.M. Gastier-Foster, M.L. Loh, P. Van Vlierberghe

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.A. Wendorff, S.A. Quinn, M. Rashkovan, C.J. Madubata, A. Ambesi-Impiombato, M.R. Litzow, R. Rabadan

Writing, review, and/or revision of the manuscript: A.A. Wendorff, S.A. Quinn, M.R. Litzow, M.S. Tallman, J.M. Gastier-Foster, M.L. Loh, A.A. Ferrando

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.A. Wendorff, S.A. Quinn, A. Ambesi-Impiombato, E. Paietta

Study supervision: R. Rabadan, A.A. Ferrando

This work was supported by the NIH grants R35 CA210065 (A.A. Ferrando), R01 CA155743 (A.A. Ferrando), CA180827 (E. Paietta), CA196172 (E. Paietta), U54 CA193313 (R. Rabadan), R01 CA185486 (R. Rabadan), U54 CA209997 (R. Rabadan), CA180820 (ECOG-ACRIN), CA189859 (ECOG-ACRIN), CA14958 (ECOG-ACRIN), CA180791 (ECOG-ACRIN), CA17145 (ECOG-ACRIN), U10 CA180827 (ECOG-ACRIN), U10 CA98543 (J.M. Gastier-Foster and M.L. Loh); the Human Specimen Banking Grant U24 CA114766 (J.M. Gastier-Foster), and P30 CA013696 (in support of the Herbert Irving Comprehensive Cancer Center Genomics, Flow Cytometry and Transgenic Mouse Shared Resources). A.A. Wendorff was supported by a Rally Foundation fellowship. P. Van Vlierberghe was supported by the Fund for Scientific Research (FWO) Flanders (postdoctoral fellowship and Odysseus type 2 grant). M. Rashkovan was supported by a Damon-Runyon Sohn Pediatric Cancer fellowship. We are grateful to T. Ludwig (The Ohio State University Comprehensive Cancer Center) for the Rosa26^+/Cre-ERT2 mouse. We thank E. Passegué for insightful suggestions in the experimental design and interpretation of our results.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.