Leukemia mutated proteins PHF6 and PHIP form a chromatin complex that represses acute myeloid leukemia stemness

Abstract

Myeloid leukemias are heterogeneous cancers with diverse mutations, sometimes in genes with unclear roles and unknown functional partners. PHF6 and PHIP are two poorly understood chromatin-binding proteins recurrently mutated in acute myeloid leukemia (AML). PHF6 mutations are associated with poorer outcomes, whereas PHIP was recently identified as the most common selective mutation in Black patients with AML. Here, we show that Phf6 knockout converts Flt3-ITD-driven mouse chronic myelomonocytic leukemia (CMML) into AML with reduced survival. Using cell line models, we show that PHF6 is a transcriptional repressor that suppresses a limited stemness gene network and that PHF6 missense mutations, classified by current clinical algorithms as variants of unknown significance, produce unstable or nonfunctional protein. We present multiple lines of evidence converging on a critical mechanistic connection between PHF6 and PHIP. We show that PHIP loss phenocopies PHF6 loss and that PHF6 requires PHIP to occupy chromatin and exert its downstream transcriptional program. Our work unifies PHF6 and PHIP, two disparate leukemia mutated proteins, into a common functional complex that suppresses AML stemness.

Keywords: chromatin; leukemia; myeloid; stemness.

Figure 1.

Phf6 loss transforms Flt3-ITD-driven CMML into AML. (A) Bar graph showing the percentages of additional genes mutated in AML patients with PHF6 mutations. Data were obtained from cBioPortal. (B) Lollipop plot of somatic PHF6 mutations in adults with myeloid (top) and lymphoid (bottom) hematological malignancies. Frameshift and nonsense mutations are shown at the left, and missense mutations are shown at the right. The plot was generated using COSMIC data visualized on the ProteinPaint portal. The ePHD1 and ePHD2 domains of PHF6 protein are indicated. (C) Kaplan–Meier survival curves of V, VP, VI, and VIP mice (n = 19–30 mice per cohort). Genotypes are described in the text. (D) Representative images of H&E staining of bone marrow from VI and VIP moribund mice and age-matched V and VP control mice. Scale bar, 100 μm at 500× magnification. (E) Spleen weight (left, n = 5–8) and leukemia score plotted against spleen weight (right, n = 3–4) from VI and VIP moribund mice and age-matched V and VP control mice. The X-axis in the right panel represents a previously described leukemia infiltration score (Spring et al. 2022) calculated based on changes in splenic architecture. (0) Intact white and red pulp, (1) extramedullary hematopoiesis with aberrant cells in disturbed white pulp, (2) leukemic blasts with high mitotic activity. (F) Representative images of H&E staining of spleens from VI and VIP moribund mice and age-matched V and VP control mice. Scale bar, 100 μm at 100× maginfication. (G–I) Bar graphs showing the percentage of LSK (Lineage⁻, cKit⁺, and Sca¹⁺) cells (G), LKSca⁻ (Lineage⁻, cKit⁺, and Sca1⁻) cells (H), and Ly6G⁺ granulocytes (I) in the bone marrow of VI and VIP moribund mice and age-matched V and VP control mice (n = 5–8 mice per cohort). (J) Table summarizing key phenotypic differences between VI and VIP mice. All bar graphs show mean ± standard error of mean (SEM). (ns) Not significant(P ≥ 0.05), (*) P = 0.01–0.05, (**) P = 0.001–0.01, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 2.

PHF6 suppresses stemness genes and promotes differentiation. (A) Immunoblot for PHF6 in WT and PHF6^KO THP-1 clones. GAPDH is shown as a loading control. (B) Heat map showing 853 differentially expressed genes in PHF6^KO compared with WT. (C) Gene set enrichment analysis (GSEA) plot showing positive enrichment of an HSC gene set in PHF6^KO compared with WT. (D) Bar graph showing normalized median fluorescence intensity (MFI) of myeloid surface markers in PHF6^KO compared with WT (n = 3). (E) Heat map showing the time course of effect of PHF6 rescue on genes differentially expressed in PHF6^KO compared with WT. Pearson correlation shows similarity between expression profiles of WT and PHF6 rescue clones at 48 h after doxycycline treatment. (F) GSEA plot showing positive enrichment of the myeloid cell gene set after 48 h of PHF6 rescue compared with the baseline KO state. (G) Bar graph showing normalized MFI of myeloid markers after 48 h of PHF6 rescue (n = 3). (H) Immunoblot for PHF6 in WT and Phf6^KO clones of the mouse ER-HoxB8 cell line. GAPDH is shown as a loading control. (I) Heat map showing 412 differentially expressed genes in Phf6^KO ER-HoxB8 clones compared with WT. (J) GSEA plots showing negative enrichment of the granulocyte gene set (left) and positive enrichment of the self-renewal gene set (right) in Phf6^KO ER-HoxB8 clones compared with WT. (K) Bar graphs showing normalized MFI (left) and the percentage of positive cells (right) for surface Ly6C expression in WT and Phf6^KO ER-HoxB8 clones at different time points after estradiol (E2) withdrawal (n = 4). All bar graphs show mean ± standard error of mean (SEM). (ns) not significant(P ≥ 0.05), (*) P = 0.01–0.05, (**) P = 0.001–0.01, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 3.

PHF6 binds gene promoters and represses transcription. (A) Heat maps (left) and meta-gene profiles (right) of three replicates of PHF6 ChIP-seq signal at open high-confidence PHF6 peaks, along with ATAC-seq and H3K27ac ChIP-seq. IgG ChIP-seq in WT and PHF6 ChIP-seq in PHF6^KO are shown as negative controls. (B) Pie chart showing categorization of PHF6 peaks based on overlap with ENCODE-defined cis-regulatory elements (CREs). (C) Heat maps showing PHF6 ChIP-seq along with selected active and repressive histone modifications (from us and publicly available data sets) (Supplemental Table S2) along bodies of genes with PHF6-bound promoters. (D) Scatter plot showing motifs and motif families enriched at PHF6-bound promoters. (E) Heat maps showing PHF6 co-occupancy with ETS family TFs, MEF2A, CEBPB, and MYB at PHF6-bound promoters. (F) Box plots showing differential expression of genes with or without PHF6 binding at promoters in PHF6^KO compared with WT. Box plots show median (lines), interquartile range (boxes), and minimum to maximum data range (whiskers). (G) Box plots showing differential expression following time-course PHF6 rescue of genes with or without PHF6 binding at promoters. Box plots show median (lines), interquartile range (boxes), and minimum to maximum data range (whiskers). (*) P = 0.01–0.05, (***) P = 0.001–0.0001, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 4.

R274Q is a functionally null point mutation. (A) Immunoblot (top) and bar graph (bottom) showing quantification of PHF6 protein in WT and R274Q clones in THP-1 cells. GAPDH is shown as a loading control (n = 5). (B) Bar graph showing RT-qPCR quantification of PHF6 mRNA levels in WT and R274Q (n = 3). (C) Representative immunofluorescence images showing localization of PHF6 protein in WT and R274Q clones. DNA stain DAPI marks the nucleoplasm, and nucleolin is a nucleolar marker. The stacked bar graph shows normalized distribution of PHF6 protein between the nucleolus and nucleoplasm in WT and R274Q clones (n = 40–60 cells). (D) Principal component analysis (PCA) plot of RNA-seq replicates of WT, PHF6^KO, and R274Q clones. (E) Heat maps showing the effect of R274Q mutation on the expression of genes differentially expressed in PHF6^KO compared with WT. Pearson correlation shows similarity between expression profiles of R274Q and PHF6^KO clones. (F) GSEA plot showing positive enrichment of the HSC gene set in R274Q compared with WT. (G) Bar graph showing normalized MFI of myeloid surface markers in R274Q compared with WT, with PHF6^KO shown for comparison (n = 3). (H) Heat maps (left) and meta-gene profiles (right) of replicates of PHF6 and R274Q ChIP-seq signal in doxycycline-inducible clones optimized to express identical levels of WT and mutant protein. PHF6 tracks are the same as those shown in Figure 3A. All bar graphs show mean ± standard error of mean (SEM). (ns) Not significant(P ≥ 0.05), (*) P = 0.01–0.05, (**) P = 0.001–0.01, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 5.

PHF6 missense mutations cause loss of function through compromised protein abundance and chromatin occupancy. (A) Lollipop plot depicting nine PHF6 missense somatic mutations selected for functional dissection. C242Y, D262V, R274Q, G287V, C297Y, and I314T are within the ePHD2 domain, whereas C20G, P153S, and E340K fall outside. (NoLS) Nucleolar localization signal. (B) Table summarizing functional characterization of PHF6 missense mutants (details shown in Supplemental Figs. S5, S6). Patient numbers were obtained from COSMIC through the ProteinPaint portal. Clinical classification of mutations was performed by the University of Pennsylvania Center for Personalized Diagnostics. Pathogenicity prediction was performed on ePHD2 mutants using four concordant meta-predictors: REVEL, MetaLR, MetaSVM, and Condel (Supplemental Fig. S5B). (ɸ) Mutants unable to be analyzed due to the unavailability of structure for non-ePHD2 domains, (↔) no change in protein or mRNA levels. ChIP signal is the average ChIP-qPCR signal at five PHF6 peaks. (†) Mutants skipped for ChIP-qPCR due to low protein level. Values marked in red are considered pathogenic or functionally detrimental for the analysis in question. (C, left) Immunoblots showing PHF6 protein level in one representative clone for each missense mutation compared with WT and PHF6^KO. GAPDH is shown as a loading control. (Right) Bar graph quantifying PHF6 protein in multiple replicate clones for each mutant (Supplemental Fig. 5SC), normalized to GAPDH. R274Q quantification shown here is the same as that shown in Figure 4A and is included here for completeness (n = 4–9 clones for each mutant). (D) Bar graph showing RT-qPCR quantification of PHF6 mRNA levels in mutant clones compared with WT. R274Q quantification shown here is the same as that shown in Figure 4B and is included here for completeness (n = 3). (E) Bar graph showing PHF6 ChIP-qPCR signal at a representative PHF6 peak in mutants compared with WT (n = 3). (†) Mutants skipped due to low protein levels (<70% of WT clones). All bar graphs show mean ± standard error of mean (SEM). (ns) Not significant (P ≥ 0.05), (*) P = 0.01–0.05, (**) P = 0.001–0.01, (***) P = 0.001–0.0001, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 6.

PHF6 shows functional similarity to PHIP, a newly described AML mutated protein. (A) Table showing the top five correlated gene dependencies for PHF6 in the Broad Institute DepMap project. (B) Scatter plot showing the correlation of CRISPR screen (Chronos) gene scores for PHF6 and PHIP in 1150 cell lines screened in DepMap. (C) Table showing the frequencies of PHF6 and PHIP mutations in databases of patients with myeloid neoplasms. Data were obtained from cBioPortal. (D) Table summarizing features of rare neurodevelopmental syndromes caused by germline mutations of PHF6 and PHIP. Features marked in red are common to both syndromes. (E) Immunoblots of PHF6 and PHIP in PHIP^KO and DKO clones. GAPDH and H3 are shown as loading controls. (F) Stacked bar graph showing normalized distribution of PHF6 protein between the nucleolus and nucleoplasm in WT and PHIP^KO clones (n = 40–60 cells). (G) PCA plot of RNA-seq replicates of WT, PHF6^KO, PHIP^KO, and DKO clones. (H) Heat map showing the effects in PHIP^KO and DKO on expression of genes differentially expressed in PHF6^KO compared with WT. Pearson correlation shows similarities between expression profiles of single- and double-knockout clones. (I) Bar graph showing normalized MFI of myeloid surface markers in PHIP^KO and DKO clones compared with WT clones (n = 3). (J) Bar graphs showing normalized MFI (left) and the percentage of positive cells (right) for surface Ly6C expression in WT, Phf6^KO, and Phip^KO clones at different time points after estradiol (E2) withdrawal (n = 3–4). All bar graphs show mean ± standard error of mean (SEM). (ns) Not significant(P ≥ 0.05), (*) P = 0.01–0.05, (**) P = 0.001–0.01, (***) P = 0.001–0.0001, (****) P < 0.0001; one-way ANOVA with Sidak’s multiple comparison testing.

Figure 7.

PHF6 cannot occupy chromatin without its functional partner, PHIP. (A) Heat maps showing PHIP ChIP-seq signal at PHF6 peaks. PHF6 tracks are the same as those shown in Figures 3A and 4H. (B) Venn diagram showing genomic overlap of PHF6 and PHIP peaks. (C) Immunoblots showing pull-down of PHF6 (wild type) and R274Q with PHIP-ChIP. IgG-ChIP and PHIP-ChIP in PHIP^KO and PHF6^KO clones are shown as negative controls. H3 is shown as a positive control for chromatin pull-down. (D) Heat maps (left) and meta-gene profiles (right) of replicates of PHF6 ChIP-seq signal in WT and PHIP^KO clones. PHF6 tracks are the same as those shown in Figure 3, A, H, and J. (E) Model of PHIP-dependent PHF6’s role in hematopoietic and leukemic stemness: PHF6 and PHIP form a complex on promoters bound by ETS factors and repress their transcription, thereby repressing a limited stemness gene network. In a subset of acute or chronic myeloid malignancies, loss of PHF6 chromatin occupancy—through loss of PHF6 itself, through missense mutations in PHF6 that impair its protein stability or chromatin occupancy, or through loss of PHIP—eliminates this repression and increases stemness.