PHF6 loss increases leukemia initiating cell frequency in Hoxa9-driven AML

To determine the prognostic significance of PHF6 mutations in human AML, we used publicly available mutational and survival data from the BEAT AML dataset [26]. Of 805 AML patients, 22 (2.7%) had PHF6 mutations. PHF6 mutations were associated with reduced overall survival in adverse risk patients (Fig. 1A) PHF6 gene mutational classes (frameshift, nonsense, missense) showed similar poor survival curves relative to each other (Fig. S1A–C).

Fig. 1: Phf6 loss increases leukemia initiating cell frequency in Hoxa9-driven AML.

A Kaplan-Meier survival curve for PHF6 mutated and unmutated adverse risk adult AML patients (ELN classification) from the BEAT AML dataset. B Experimental design for Hoxa9 retroviral transduction of Ctrl and cKO marrow, followed by colony forming unit assay (CFU), AML induction in mice, and serial transplantation of transformed leukemic cells. C Bar graph showing number of colony forming units (CFUs) obtained from 8 rounds of serial methylcellulose replating of 500 cells/plate of Ctrl+Hoxa9 and cKO+Hoxa9 transformed mouse bone marrow. (n = 6–12 biological replicates). D Left Bar graph showing average number of cells per colony (colony size) obtained after 7 rounds of serial methylcellulose replating of 500 cells/plate of Ctrl+Hoxa9 and cKO+Hoxa9 transformed mouse bone marrow. (n = 6–12 biological replicates) Right Representative photographs of colonies at the 3rd plating. Scale bar represents 500 μm. E Kaplan–Meier survival curves of Ctrl+Hoxa9 and cKO+Hoxa9 primary transplant recipients receiving 400 K, 100 K, or 30 K GFP+ cells. (n = 7–21 mice per cohort). F Kaplan–Meier survival curve of Ctrl+Hoxa9 and cKO+Hoxa9 secondary transplant recipients, receiving 100 K or 30 K GFP+ cells harvested from bone marrow of primary recipients 8 weeks after transplantation. (n = 7–10 mice per cohort). G Kaplan–Meier survival curve of Ctrl+Hoxa9 and cKO+Hoxa9 tertiary transplant recipients, receiving 100 K GFP+ cells harvested from bone marrow of secondary recipients 8 weeks after transplantation. (n = 7–11 mice per cohort). H, I Limiting dilution analysis for LIC calculation for (H) freshly Hoxa9-transduced cells, (I) 8 weeks primary transplanted leukemic cells. All bar graphs show mean ± SEM and statistical significance was calculated using the Student t-test. For all survival curves, statistical significance was calculated using the Log-rank (Mantel-Cox) test. *p < 0.05, **p < 0.01, ***p < 0.001; ****p < 0.0001, ns = not significant.

To assess the role of PHF6 loss in mouse AML, we generated conditional hematopoietic Phf6 knockout Vav-CreCre/+Phf6fl/y (cKO) mice (Fig. S1D) and compared them to their Vav-CreCre/+Phf6+/y (Ctrl) littermates. Published studies of hematopoietic Phf6 loss have reported no evidence of leukemic transformation [17,18,19, 27]. We also found no blood count abnormalities in cKO mice up to 9 months (Fig. S1E–K), further indicating that Phf6 loss alone is likely insufficient to initiate leukemia.

We next induced AML using the Hoxa9 retroviral transduction model. We picked this model due to its ability to produce AML with a relatively longer latency (lethality in ~3–6 months) [24], allowing us to test a potential role for Phf6 loss in accelerating AML kinetics. We transduced whole bone marrow from 5-FU-treated Ctrl and cKO mice with MSCV Hoxa9-IRES-GFP retrovirus, and investigated the effect of Phf6 loss on the ability of Hoxa9-transformed cells to form colonies in methylcellulose (Fig. 1B). We observed that Ctrl+Hoxa9 cells were nearly exhausted after 4 platings, whereas cKO+Hoxa9 cells demonstrated persistent colony-forming ability up to 8 platings (Fig. 1C), with larger colonies (Fig. 1D). Thus, Phf6 loss gives a replating advantage to Hoxa9-transformed marrow and delays its in vitro exhaustion. Conversely, MLL-AF9-IRES-GFP transformed marrow showed no evidence of exhaustion with replating, and no evidence of further acceleration with Phf6 loss (Fig. S2A, B). Given that 70% of human AMLs show high HOXA9 levels [25], we deemed the Hoxa9 overexpression mouse model as being broadly relevant to human leukemia biology, as opposed to MLL-AF9, which does not co-occur with PHF6 mutation in humans. We therefore proceeded with the Hoxa9-driven model for our studies.

To test the role of Phf6 loss in the development of AML in vivo, we transplanted Hoxa9-transduced marrow into lethally irradiated syngeneic recipients (Fig. 1B). We confirmed that Ctrl+Hoxa9 marrow produced lethality in recipient mice in ~3–5 months after transplantation (Fig. 1E) due to AML characterized by >20% blasts in the marrow (Fig. S2C), peripheral leukocytosis (Fig. S2D), splenomegaly with effacement of splenic architecture (Fig. S2E), and infiltration of leukemic cells in the liver (Fig. S2F). Survival was similar between Ctrl+Hoxa9 and cKO+Hoxa9 groups in primary recipients transplanted with multiple doses (400 K, 100 K, or 30 K cells) (Fig. 1E), with similar degrees of leukemic infiltrate (Fig. S3A–F) and splenomegaly at morbidity (Fig. S3G). However, secondary and tertiary transplantation of marrow showed progressively accelerated lethality in cKO+Hoxa9 compared to Ctrl+Hoxa9 (Fig. 1F, G, S3H). Thus, Phf6 loss accelerates Hoxa9-driven mouse AML on serial transplantation.

We next sought to determine the effect of Phf6 loss on the frequency of leukemia initiating cells (LIC), the sub-population of transformed marrow capable of initiating leukemia. We performed limiting dilution transplantation assays (LD) on freshly transduced marrow (GFP+ cells sorted 2 days after retroviral transduction) as well as on marrow from recipients (GFP+ bone marrow cells sorted from recipients 8 weeks after primary transplantation). We picked the 8-week time point based on the initiation of lethality in this model at ~12 weeks (Fig. 1E). LD of freshly transduced marrow showed that, at baseline, cKO+Hoxa9 cells had a 2-fold higher frequency of cells capable of leukemic transformation (Fig. 1H). LD on marrow extracted 8 weeks post-transplantation showed an approximately 5-fold greater frequency of leukemia initiating cells (LICs) in cKO+Hoxa9 marrow (Fig. 1I). Thus, Phf6 loss increases LIC frequency in Hoxa9-driven AML, with the increase occurring during in vivo AML evolution.

Phf6 loss increases leukemic disease burden

To characterize the effect of Phf6 loss further, we analyzed peripheral blood, splenic architecture, and bone marrow leukemic cell burden of primary recipients at 8 weeks after transplantation. Mice transplanted with cKO+Hoxa9 cells showed a higher frequency of GFP+ cells in peripheral blood at 8 weeks than mice receiving Ctrl+Hoxa9 cells (Fig. 2A). The cKO+Hoxa9 group also had greater leukocytosis (Fig. 2B) and more severe thrombocytopenia (Fig. 2C). Mice in both groups displayed comparable levels of anemia (Fig. S4A, B). The cKO+Hoxa9 group had increased spleen size and weight compared to the Ctrl+Hoxa9 group (Fig. 2D, E), and histopathological analysis showed greater effacement of splenic architecture (Fig. 2F). Splenic infiltration was quantified using a previously described leukemia infiltration score [28], and was found to be greater in cKO+Hoxa9 mice compared to Ctrl+Hoxa9 (Fig. 2G). Giemsa-stained cytospin preparations showed higher blast percentages in cKO+Hoxa9 at the 8-week timepoint (Fig. 2H, I), and flow cytometry showed higher absolute and percent GFP+ cells (Fig. 2J, S4C). All GFP+ cells were myeloid for both groups (Fig. 2K, S4D). Thus, while mice from both groups succumbed at similar times after primary transplant (Fig. 1F), analyses at matched time points before the onset of mortality revealed greater disease burden in cKO+Hoxa9 mice compared to Ctrl+Hoxa9.

Fig. 2: Phf6 loss increases leukemic disease burden.

A–C Bar graphs showing peripheral blood analysis at 8 weeks after transplantation of Ctrl+Hoxa9 and cKO+Hoxa9 cells. A Percentage of GFP+ cells in peripheral blood. B, C Counts of (B) WBCs and (C) platelets in peripheral blood. Normal range for WBCs: 2000–10,000/µl. Normal range for platelets: 900–1600 × 103/µl [41]. D Representative photograph of spleens at 8 weeks after transplantation. Ruler depicts length in centimeters. E Bar graph showing spleen weight (expressed as percentage of total body weight) of primary recipients at 8 weeks after transplantation of Ctrl+Hoxa9 and cKO+Hoxa9 cells. F Representative image of H&E staining of spleen from Ctrl+Hoxa9 and cKO+Hoxa9 at 8 weeks after transplantation. Scale bar is 100 µm at 10X. G Spleen weight (y-axis) and leukemia score (x-axis) from Ctrl+Hoxa9 and cKO+Hoxa9 primary recipients at 8 weeks. X-axis represents a previously published leukemia infiltration score [28] calculated based on splenic architecture. Intact white and red pulp was scored as 0, extramedullary hematopoiesis evident by aberrant cells in disturbed white pulp was scored as 1, while infiltration with leukemic blasts with high mitotic activity was scored as 2. H Representative image of Wright–Giemsa staining of cytospin of Ctrl+Hoxa9 and cKO+Hoxa9 bone marrow cells at 8 weeks after transplantation. Scale bar is 100 µM at 400X. Red arrows indicate representative blast cells. I Bar graph of percentage blast cells of total nucleated cells in bone marrow at 8 weeks after transplant. J, K Bar graphs showing (J) absolute number of GFP+ cells in marrow and (K) percentage of CD11b+ myeloid cells among all GFP+ cells at 8 weeks after transplantation. All bar graphs show mean ± SEM and statistical significance was calculated using the Student t-test. For all survival curves, statistical significance was calculated using the Log-rank (Mantel-Cox) test. *p < 0.05, **p < 0.01, ***p < 0.001; ****p < 0.0001, ns = not significant.

Phf6 loss increases the frequency of self-renewing, transplantable LICs

To characterize the immunophenotype of AML subpopulations (including LICs), we further analyzed the marrow of Ctrl+Hoxa9 recipients at 8 weeks after transplantation. The GFP+ cells did not express B or T cell markers (Fig. S5A). Immature AML cells are known to have high c-Kit expression [29], and leukemic stem cells (LSCs) in the MLL-AF9 retroviral mouse model aberrantly express mature myeloid lineage antigens such as Ly6C and CD11b [30]. To identify the corresponding subpopulation containing LICs in Hoxa9-only-driven AML, and to characterize the differentiation hierarchy of this model, we settled on a strategy using c-Kit and Ly6C expression to divide GFP+ marrow cells into three populations: (i) cKit+ Ly6C-, (ii) c-Kit+ Ly6C+, and (iii) c-Kit- Ly6C+ (Fig. 3A, S5B). The population at the top of the hierarchy was the cKit+ Ly6C- population, an immature population with expression of cKit, CD34, and dim CD11b, with no expression of Ly6C, Ly6G, or Sca-1, and mixed expression of CD16/32 (Fig. S5C). This population was capable of giving rise to more differentiated Ly6C+ cells within 2 days of culture (Fig. S5D), could produce colonies on methylcellulose plating (Fig. 3B), and could engraft into recipient mice (Fig. 3C). Based on this subpopulation’s ability to engraft, but cognizant that not all cells within it are LICs, we termed it the ‘LIC enriched’ (LIC-e) population (Fig. 3D). The second population was the c-Kit+ Ly6C+ population, also expressing CD11b, CD34, and CD16/32, but not Sca-1 (Fig. S5C). On culture, this population could only give rise to Ly6C+ cells, but not to any Ly6C- cells (Fig. S5D), indicating that it is irreversibly committed to differentiation. This population could produce a small number of colonies in methylcellulose (Fig. 3B), but could not engraft mice (Fig. 3C). We termed it the ‘committed’ leukemic cell population (Fig. 3D). The third population was the c-Kit- Ly6C+ population, which expressed Ly6C, CD11b, CD34, and CD16/32, but not Sca-1, and had mixed expression of Ly6G (Fig. S5C). It could not produce any colonies (Fig. 3B) nor engraft (Fig. 3C). In vitro, this population only gave rise to Ly6C+ cells (Fig. S5D). We termed it the “differentiated” leukemic cell population (Fig. 3D). We thus established the hierarchical organization of Hoxa9-driven AML, and used it to evaluate the effects of Phf6 loss.

Fig. 3: Phf6 loss increases the frequency of self-renewing, transplantable LICs.

A Representative flow cytometry plot of bone marrow GFP+ cells at 8 weeks after transplantation with Ctrl+Hoxa9 cells, compartmentalized into three subpopulations (i) LIC-e, (ii) Committed, and (iii) Differentiated leukemic cells using c-Kit and Ly6C expression. Note: The same flow cytometry plot has been shown in Fig. S5 with detailed immunophenotypic markers. B Bar graph showing number of colony forming units (CFUs) obtained on methylcellulose plating of 500 cells of sorted subpopulations of Ctrl+Hoxa9 transplanted marrow. (n = 6 biological replicates). C Bar graph showing frequencies of GFP+ cells in peripheral blood of secondary recipient mice at 4 weeks after transplantation of sorted subpopulations from Ctrl+Hoxa9 primary recipient marrow. (n = 4–5 biological replicates). D Schematic of hierarchical organization of leukemic cells (LIC-e, committed, and differentiated leukemic cells) within AML produced by retroviral Hoxa9 transduction. E Representative flow cytometry plots depicting subpopulations of Ctrl+Hoxa9 and cKO+Hoxa9 leukemia marrow at 8 weeks after primary transplant. F, G Stacked bar graphs showing (F) frequencies, and (G) absolute number per femur of LIC-e, committed, and differentiated leukemic populations from Ctrl+Hoxa9 and cKO+Hoxa9 marrow at 8 weeks after transplantation. Insets show frequencies and absolute numbers of the LIC-e subpopulation. (n = 10–11). H Bar graph showing number of CFUs obtained on methylcellulose plating of 500 cells of sorted LIC-e, committed, and differentiated leukemic populations from Ctrl+Hoxa9 and cKO+Hoxa9 primary recipient bone marrow at 8 weeks after transplantation. (n = 6 biological replicates). I Bar graph showing frequencies of GFP+ cells in the peripheral blood of secondary recipient mice at 4 weeks after transplantation with sorted LIC-e, committed, and differentiated leukemic cell subpopulations from Ctrl+Hoxa9 and cKO+Hoxa9 primary recipient bone marrow at 8 weeks after transplantation. (n = 5 biological replicates). Note: The Ctrl+Hoxa9 samples in (H) and (I) are the same as those depicted in (B) and (C)—the current figures are comparing Ctrl and cKO. J Kaplan–Meier curve of secondary transplant recipients receiving 50 K sorted LIC-e cells from Ctrl+Hoxa9 and cKO+Hoxa9 primary mouse bone marrow at 8 weeks after transplantation. (n = 4–5 biological replicates). K Left Representative flow cytometry plots for cell cycle analysis of LIC-e cells, with BrdU (marking cells in S phase) and 7-AAD (marking DNA) 2 h after BrdU injection into live mice at 8 weeks after transplantation. Right Stacked bar graph indicates frequencies of in vivo LIC-e cells in G0/G1, S, and G2/M phases. (n = 13 biological replicates). L Left Representative flow cytometry plots of apoptotic (Annexin V+, 7AAD−), and necrotic (7AAD+) LIC-e cells. Right Bar graph shows frequencies of apoptotic and necrotic LIC-e cells from Ctrl+Hoxa9 and cKO+Hoxa9 primary recipient bone marrow at 8 weeks after transplantation. All bar graphs show mean ± SEM and statistical significance was calculated using the Student t-test. For all survival curves, statistical significance was calculated using the Log-rank (Mantel-Cox) test. *p < 0.05, **p < 0.01, ***p < 0.001; ****p < 0.0001, ns = non significant.

The LIC-e population, though comprising a small minority of GFP+ cells, was expanded in 8-week cKO+Hoxa9 marrow compared to Ctrl+Hoxa9, while the relative fractions of committed and differentiated leukemic cells were similar (Fig. 3E, F). The difference in the LIC-e population was even more pronounced when absolute cell numbers were considered, showing a 5-fold increase in cKO+Hoxa9 (Fig. 3G). To determine functional differences between Ctrl+Hoxa9 and cKO+Hoxa9 subpopulations, we sorted equal numbers of cells from each subpopulation at 8 weeks after transplant and performed methylcellulose culture and secondary transplantation into irradiated recipients. Committed and differentiated leukemic cells of either group formed few to no colonies, while LIC-e cells showed comparable colony-forming ability (Fig. 3H). cKO+Hoxa9 LIC-e cells showed greater engraftment in secondary recipients at 4 weeks (Fig. 3I) and led to more rapid lethality than Ctrl+Hoxa9 LIC-e cells (Fig. 3J). We did not observe any difference in cell cycle distribution or apoptosis of LIC-e cells (Fig. 3K, L). Thus, Phf6 loss led to an expanded and more transplantable LIC-enriched AML subpopulation whose enhanced leukemic potential was not explained by differences in cell cycle or apoptosis.

Phf6 loss promotes a stemness gene network

We determined the transcriptional consequences of Phf6 loss on Hoxa9-transformed marrow by performing RNA-Seq on LIC-e and committed leukemic cells from marrow of transplanted recipients at 8 weeks. Committed leukemic cells showed no change in gene expression with Phf6 loss (Fig. S6A, B, Table S2), while the LIC-e population showed 91 downregulated and 65 upregulated genes in cKO+Hoxa9 compared to Ctrl+Hoxa9 (Fig. 4A, B, Table S2). Genes downregulated in cKO+Hoxa9 LIC-e cells showed Gene Ontology (GO) enrichment for myeloid differentiation terms (Fig. 4C). Gene set enrichment analysis (GSEA) [31] showed that the cKO+Hoxa9 LIC-e transcriptome showed positive enrichment for genesets related to high LSC potential [32] and leukemic GMPs (L-GMPs) [33] and negative enrichment for genesets related to myeloid differentiation [34], and mature neutrophils and monocytes [35] (Fig. 4D, S6C).

Fig. 4: Phf6 loss promotes a stemness gene network.

A Volcano plot showing differentially expressed genes in LIC-e cells from cKO+Hoxa9 compared to Ctrl+Hoxa9 bone marrow at 8 weeks after transplantation. (n = 3–4 biological replicates). B Heatmap of differential expression between Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells. Insets show selected downregulated (left) and upregulated (right) genes in cKO+Hoxa9 LIC-e compared with Ctrl+Hoxa9 LIC-e. C Top Gene Ontology terms enriched in genes downregulated in cKO+Hoxa9 LIC-e compared with Ctrl+Hoxa9 LIC-e. D Gene set enrichment analysis (GSEA) plots of the cKO+Hoxa9 LIC-e transcriptome compared to Ctrl+Hoxa9. Plots show positive enrichment of gene sets related to high LSC frequency (left) and leukemic GMPs (middle), and negative enrichment of a gene set related to myeloid development (right). Normalized Enrichment scores (NES) and FDR q values are shown. E Volcano plot showing differentially expressed genes in PHF6KO THP-1 cells compared to wildtype THP1 cells. (n = 4–5 clonal lines). F Top Gene Ontology terms enriched in genes downregulated in PHF6KO THP-1 cells compared to wildtype THP1 cells. G Gene set enrichment analysis (GSEA) plots of the PHF6KO THP-1 transcriptome compared to wildtype clones. Plots show positive enrichment of gene sets related to high HSC and Progenitors (left), and negative enrichment of a gene set related to granulocytes and monocytes (right). H Flow chart depicting choice of BEAT AML dataset samples picked for RNA-Seq analyses, with exclusion of cases classified as ELN favorable and those with low HOXA9 mRNA levels (which show rare/no PHF6 mutations). I Gene set enrichment analysis (GSEA) plots of the transcriptome of PHF6 mutated AML patients with high HOXA9 expression compared to PHF6 wildtype AML patients with high HOXA9 expression. Plots show positive enrichment of gene sets related to high HSC and Progenitors (left), and negative enrichment of a gene set related to granulocytes and monocytes (right).

We next sought to determine whether PHF6 loss produces similar transcriptional changes in human AML. We used CRISPR to generate PHF6 knockout (PHF6KO) clones of the THP-1 human AML cell line (Fig. S6D). RNA-Seq showed that genes downregulated in PHF6KO compared to wildtype clones were also enriched for myeloid differentiation GO terms (Fig. 4E, F, Table S3). PHF6KO clones also showed positive GSEA enrichment of HSC and progenitor signatures, and negative enrichment of granulocyte and monocyte signatures [36] (Fig. 4G). We further analyzed publicly available RNA-Seq data from the BEAT AML dataset [26]. A majority of samples had high HOXA9 mRNA levels, and PHF6 mutated cases exclusively fell within the HOXA9 high (HOXA9High) group (Fig. 4H). On comparing the transcriptomes of PHF6 mutant (PHF6Mut) to PHF6 wildtype (PHF6WT) AMLs within the HOXA9High group, we found similar positive GSEA enrichment of HSC and progenitor signatures, and negative enrichment of granulocyte and monocyte signatures [36] (Fig. 4I). Thus, Phf6 loss or mutation in both mouse and human AML skews their transcriptomes to a more stem-like and less differentiated state.

Phf6 loss prevents exhaustion of LIC-e cells by maintaining their self-renewal potential

To determine the kinetics of the effects of Phf6 loss on the behavior of LIC-e cells, we cultured Hoxa9-transduced mouse bone marrow in cytokine-supplemented media. The growth rate of bulk culture was similar for Ctrl+Hoxa9 and cKO+Hoxa9 marrow (Fig. S7A). When sorted LIC-e cells were cultured, most cells lost LIC-e identity within days (Fig. 5A). However, though both groups produced similar fractions of committed (c-Kit+ Ly6C+) and differentiated cells (c-Kit- Ly6C+), the Ctrl+Hoxa9 culture almost completely depleted its LIC-e population (<1%), while the cKO+Hoxa9 culture maintained this population, plateauing at 5–6% of the total culture after 5 days (Fig. 5A). Thus, Phf6 loss prevents exhaustion of the LIC-e population without impairing the rate of proliferation or differentiation of the bulk culture, recapitulating the in vivo LIC-e accumulation phenotype shown earlier (Fig. 3E–G).

Fig. 5: Phf6 loss prevents exhaustion of LIC-e cells by maintaining their self-renewal potential.

A Bar graph showing frequencies of subpopulations resulting from in vitro culture of LIC-e cells sorted 4 days after Hoxa9 transduction of Ctrl and cKO bone marrow. Inset bar graph depicts only LIC-e frequencies in the same culture. (n = 13 biological replicates). B Experimental design for study of in vitro cell cycle analysis (top) and self-renewal (bottom) of LIC-e cells using EdU chase and pulse-chase assay respectively. C Left, Bar graph showing frequencies of G0/G1, S, and G2/M phases in Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells in culture 2 h after addition of EdU. (n = 4–5 biological replicates) Right, Representative flow cytometry plots of same, with EdU marking cells in S phase and 7-AAD staining DNA. D Left, Representative flow cytometry plots showing kinetics of uptake of EdU by Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells over 10 h of culture, performed for calculation of cell cycle length. Time to whole population (>85%) EdU uptake represents G1 phase length, and time to EdU intensity maxima represents S phase length. Right, Bar graph represents the length of cell cycle phases of LIC-e cells in culture. (n = 4–5 biological replicates) E Stacked bar graph showing the percentage of self-renewing and differentiating progeny produced by Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells in culture. (n = 4–5 biological replicates). F Kaplan-Meier curve of primary transplant recipients receiving 10 K sorted Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells. (n = 5–6 biological replicates) All bar graphs show mean ± SEM and statistical significance was calculated using the Student t-test. For all survival curves, statistical significance was calculated using the Log-rank (Mantel-Cox) test. *p < 0.05, **p < 0.01, ***p < 0.001; ****p < 0.0001, ns = non significant.

To reconfirm that LIC-e accumulation wasn’t due to subtle cycling differences, we performed a 10-hour cell cycle analysis in culture by adding EdU to sorted Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells and harvesting them at serial time points (Fig. 5B). We observed that the distribution of Ctrl+Hoxa9 or cKO+Hoxa9 LIC-e cells in G1 and S phases showed no difference at the start of culture, and showed no difference in the kinetics or magnitude of EdU incorporation (Fig. 5C). Based on previously published rationale [37], we defined the time required for 85% of each sample to become EdU+ as the “G1 phase length”, and the time required for each sample to reach a maximal EdU fluorescent intensity as the ‘S phase length’. We did not observe any difference in G1 and S phase lengths between Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells (Fig. 5D). We therefore hypothesized that cKO+Hoxa9 LIC-e cells, while capable of producing differentiated progeny, may have a greater tendency to produce progeny with persistent LIC-e identity. To test this hypothesis, we performed an EdU pulse-chase experiment by incubating sorted LIC-e cells with EdU for 2 hours (pulse), followed by washing off the EdU and further culturing cells for 8 more hours (chase, for the length of S phase) (Fig. 5B) to allow all cells that were in S phase during the initial EdU pulse to complete mitosis, so that all EdU+ cells at the end of the 8 hour EdU-free chase would be daughter cells/progeny of the original EdU-uptaking cells. We determined the percentage of self-renewing progeny by calculating the percentage of total progeny (total EdU+ cells) that had LIC-e markers (Ly6C−EdU+), while the rest (Ly6C+ EdU+) were differentiating progeny. We observed that while 40.6% of the progeny of Ctrl+Hoxa9 LIC-e cells were also LIC-e cells, this fraction was increased to 60.0% in cKO+Hoxa9 (Fig. 5E). To confirm that in vitro LIC-e cells are functionally equivalent to their in vivo counterparts, we performed primary transplantation with 10 K LIC-e cells sorted after 2 days of culture. Recipients of cKO+Hoxa9 LIC-e cells succumbed faster than Ctrl+Hoxa9 LIC-e (Fig. 5F). Collectively, Phf6 loss prevents the exhaustion of LIC-e cells by increasing the fraction of their progeny that retain persistent LIC-e identity.

Effects of Phf6 loss on chromatin accessibility in LIC-e cells

To profile the effects of Phf6 loss on the accessibility landscape of LIC-e cells, we performed ATAC-Seq on sorted LIC-e cells from freshly transduced cKO+Hoxa9 and Ctrl+Hoxa9 marrow. We observed that cKO+Hoxa9 LIC-e cells showed a global reduction in chromatin accessibility compared to Ctrl+Hoxa9, with only a few regions showing increased accessibility (Fig. 6A, B). Regions that lost accessibility in cKO+Hoxa9 LIC-e cells showed enrichment for AP-1, HOX, SPI, and GATA family motifs, among others (Fig. 6C). Public ChIP-Seq tracks from leukemic or myeloid cells showed co-occupancy of these factors at ATAC peaks with reduced accessibility (Fig. 6D, Table S4). Promoters of multiple genes like Runx1, Selplg, and Aim2, which are downregulated in cKO+Hoxa9 LIC-e cells (Fig. 4B), showed occupancy by these factors and showed reduced chromatin accessibility in cKO+Hoxa9 LIC-e cells (Fig. S7B). Conversely, the small number of regions that gained accessibility in cKO+Hoxa9 LIC-e cells showed enrichment for NF-kB and IRF family motifs (Fig. 6E). Public ChIP-Seq tracks showed co-occupancy of NF-kB (RELA, RELB) and IRF8, IRF4 at regions of increased accessibility (Fig. 6F, Table S4). Overall, Phf6 loss, likely via a combination of direct and indirect effects, led to altered accessibility at sites bound by key hematopoietic transcription factors.

Fig. 6: Effects of Phf6 loss on chromatin accessibility in LIC-e cells.

A Volcano plot showing differentially accessible regions in LIC-e cells from cKO+Hoxa9 compared Ctrl+Hoxa9. (n = 3 biological replicates). B Representative signal profile (left) and metagene plots (right) showing genome-wide intensity of ATAC signal in Ctrl+Hoxa9 and cKO+Hoxa9 LIC-e cells. C HOMER analysis for regions of decreased chromatin accessibility in cKO+Hoxa9 LIC-e cells, showing enrichment of AP-1, HOX, GATA, SPI1, and MAF motifs. D Representative metagene plots at regions of decreased chromatin accessibility in cKO+Hoxa9 LIC-e, show ChIP-Seq signal for select proteins whose motifs are seen to be enriched through HOMER in (C). E HOMER analysis for regions of increased chromatin accessibility in cKO+Hoxa9 LIC-e cells showing enrichment of NF-kB and IRF motifs. F Representative metagene plots at regions of increased chromatin accessibility in cKO+Hoxa9 LIC-e, show ChIP-Seq signal for select proteins whose motifs are seen to be enriched through HOMER in (E). Publicly available ChIP-Seq datasets in leukemia or myeloid cells were used in metagene heatmaps (Table S4). All plots were centered around ATAC-Seq peaks. SeqPlots was used to draw all metagene plots.