HoxA9 transforms murine myeloid cells by a feedback loop driving expression of key oncogenes and cell cycle control genes

Abstract

Ectopic expression of the oncogenic transcription factor HoxA9 is a major cause of acute myeloid leukemia (AML). Here, we demonstrate that HoxA9 is a specific substrate of granule proteases. Protease knockout allowed the comprehensive determination of genome-wide HoxA9 binding sites by chromatin immunoprecipitation sequencing in primary murine cells and a human AML cell line. The kinetics of enhancer activity and transcription rates in response to alterations of an inducible HoxA9 were determined. This permitted identification of HoxA9-controlled enhancers and promoters, allocation to their respective transcription units, and discrimination against HoxA9-bound, but unresponsive, elements. HoxA9 triggered an elaborate positive-feedback loop that drove expression of the complete Hox-A locus. In addition, it controlled key oncogenic transcription factors Myc and Myb and directly induced the cell cycle regulators Cdk6 and CyclinD1, as well as telomerase, drawing the essential blueprint for perturbation of proliferation by leukemogenic HoxA9 expression.

Graphical abstract

Figure 1.

HoxA9 is degraded by neutrophil granule proteases. (A) HoxA9 is unstable in myeloblasts but not in epithelial cells. Triton-based lysates of primary murine bone marrow cells transformed by HA-tagged HoxA9 or extracts from 293T cells transfected with the same construct were incubated on ice with samples taken at the indicated time points, inactivated in hot SDS, and analyzed by western blotting. (B) HOXA9 is unstable in human AML cell lines. Extracts from THP1 and Molm13 cells derived from patients with MLL translocations and, therefore, expressing high levels of HOXA9, were extracted as above. Lysates were immediately denatured in SDS or incubated on ice for 10 minutes before western blot analysis for endogenous HOXA9. (C) High concentrations of the weak elastase inhibitor chymostatin inhibit HoxA9 degradation in myeloblast extracts. Triton-based extracts were produced from primary cells transformed by HA-tagged HoxA9 and supplemented with the indicated concentrations of chymostatin before western blotting. Controls were directly boiled in SDS before analysis. (D) Elastase degrades HoxA9. Extracts of 293T cells expressing HA-HoxA9 were supplemented with SDS or with different concentrations of purified elastase or were mixed 1:1 with myeloblast extract, as indicated. Lysates were incubated for 10 minutes on ice before western blotting. (E) HoxA9 is a substrate for elastase, proteinase 3, and cathepsin G. HoxA9-transformed primary HSPCs from Elane^−/− mice were further deleted for Prtn3 and Ctsg by Crispr-based knockout. Lysates of individual cell lines were tested for HoxA9 stability, as above. (F) HoxA9 is stable in EPC triple-knockout myeloblasts. HoxA9-transformed cells from EPC or WT animals were lysed, and the stability of HoxA9 was tested as before.

Figure 2.

Efficient and comprehensive ChIP-seq for HoxA9 is possible in myeloid cells from EPC mice and in a human cell line low in granule proteases. (A) Integrated Genome Viewer (IGV)–generated example of ChIP-seq profiles across the established HoxA9 target Myb. Results from murine cells (upper panel). Four replicate samples and an IgG control are shown. For comparison, HoxA9 precipitation data from T cells are included. Replicates I to III were prepared from EPC HSPCs transformed by HA-tagged HoxA9; for replicate IV, triple Flag–tagged HoxA9 was used. Chromatin was precipitated with anti-HA or anti-Flag antibodies. The color-coded ribbon represents a heat map of conservation between mammal species, with the highest conservation depicted in red. ChIP-seq results from human AML cell line MV4;11 that is naturally low in granule proteases (lower panel). Endogenous HOXA9 was precipitated by anti-HOXA9 antibodies. (B) Distribution of HoxA9 peaks across genomic elements. (C) De novo motif search across the 1000 top scoring and highest confidence HoxA9/HOXA9 peaks. Shown are all recurrent motifs with a significant homology to a consensus binding site for a TF (family) that occurred in >15% of all identified peaks. P values and frequency (%) are given. (D) Motif distribution across HoxA9/HOXA9 peaks, as in panel C. The localization of consensus sites with respect to the identified HoxA9/HOXA9 peak center is plotted. (E) The genomic sequence of HoxA9 peaks is conserved across higher mammals. The metagene plot depicts sequence-conservation scores of all HoxA9 peaks compared with 60 species of Euarchontoglires (a clade of mammals including rodents, lagomorphs, tree shrews, colugos, and primates; http://hgdownload.cse.ucsc.edu/goldenPath/mm10/phastCons60way/).

Figure 3.

HoxA9-regulated genes can be identified by nascent RNA-seq. (A) Schematic description of sample generation. HSPCs of WT mice transduced with a tamoxifen (TAM)-inducible HoxA9-ER fusion were used for these experiments. Cells were supplemented for 1 hour with 4-thiouridine at the indicated time points, and labeled RNA was isolated and sequenced to determine actual transcription rates in the presence of functional HoxA9 (0 hours), as well as after cessation of HoxA9 activity. (B) Venn diagram. Activated and repressed genes were defined at 24, 48, and 72 hours by a cumulated log2 fold change in transcription rate >1.5 (transcript up after inactivation of HoxA9 = repressed by HoxA9) or a cumulated log2 fold change in transcription rate less than −1.5 (transcript down in the absence of HoxA9 = activated by HoxA9). The overlap with 165 direct target genes of MLL-ENL, as identified previously, is indicated. (C) The 65-gene HoxA9/MLL-ENL core signature. Heat map illustrates transcript changes 72 hours after inactivation of the respective oncogene. Red hues correspond to active genes, and blue colors designate repressed genes. (D) Gene set enrichment analysis of the HoxA9-on vs HoxA9-off (72 hours) gene-expression signatures. Normalized enrichment scores (NES) and false-discovery rates (FDS) are given.

Figure 4.

ChIP-seq for H3K4me and H3K27ac assigns potential enhancers to transcripts under control of HoxA9. (A) Metagene plot and occupation heat map of H3K4me and H3K27ac around all 24 470 identified HoxA9 peaks at 0 hours (Hox active) and 72 hours (Hox inactive). Heat maps are drawn for 0 hours only. H3K4me and H3K27ac profiles were determined in WT cells transformed by HA-tagged HoxA9. (B) HoxA9 coincides with H3K4me and H3K27ac at promoters and enhancers. The example covers the Myc gene and its known enhancer ∼1.5 Mbp downstream. (C) Transcript response to HoxA9 correlates predominantly with H3K27ac in enhancers and promoters. Enhancer modifications change in response to HoxA9 activity (top panel). The situation at the Myc enhancer is shown at 0 hours (Hox on) and 72 hours (Hox off). HoxA9 ChIP, nascent RNA-seq, H3K27ac, and H3K4me IGV tracks are depicted. HoxA9 regulation allows association of enhancers with their cognate genes (middle panel). The example depicts the situation for Hmga2, in which a putative upstream enhancer is occupied and regulated by HoxA9. IGV tracks shown are as above. H3K27ac kinetics faithfully predicts HoxA9 activity (bottom panel). The intergenic region between Lcn2, a transcript repressed by HoxA9, and Ptges2, a gene bound, but not regulated, by HoxA9 is displayed. Please note that Ptges2 transcript abundances are too low to be visible in the RNA tracks at the scale necessary to visualize Lcn2 upregulation after inactivation of HoxA9. Potential promoter/enhancer sites are boxed. (D) Most HoxA9 sites are not involved in gene regulation. The Venn diagram depicts the numbers of acetylated regions changing modification density by ≥twofold 72 hours after inactivation of HoxA9 and the respective overlap with HoxA9 peaks.

Figure 5.

HoxA9 controls itself and the Hox-A locus. (A) Overview of the Hox-A locus in mouse and human genomes with conserved Hox/HOXA9 binding sites and associated nascent RNA and H3K27ac profiles. Potential HoxA9-dependent regulatory regions are boxed and labeled as follows: a1_put_enh = HoxA1 putative enhancer, a4_ac_reg = HoxA4 acetylated region, a6_put_enh = HoxA6 putative enhancer, a9_prom = HoxA9 promoter. For a detailed explanation, see the text and supplemental Figure 5. (B) Hox-A transcription is under control of HoxA9. Kinetics of nascent RNA Hox production after cessation of HoxA9 activity (left panel). RPKM values are shown. Differential qPCR for endogenous and viral-derived HoxA9 RNA (right panel). Expression was normalized to β-actin as a housekeeping gene, and it is plotted as relative value with 0 hours set to 1 unit. (C) Principle of epigenome editing by dCas9-KRAB. Targeting of a catalytically inactive Cas9-KRAB fusion to a specific locus by an sgRNA creates local heterochromatin through recruitment of a repressor complex. (D) Epigenome editing of HoxA1/A6 putative enhancers affects expression of anterior Hox-A genes. Expression levels of all Hox-A genes were recorded by qPCR in cells expressing sgRNAs targeting the HoxA1/A6 putative enhancers and a dCas-KRAB fusion or dCas-KRAB alone, as a control, as indicated. Relative expression normalized to β-actin is shown, with control expression set to 1 unit. (E) Pbx3 limits Meis1 protein availability. Nascent RNA production of Meis1 and Pbx3 RNAs in response to HoxA9 (upper panel). Western blot analysis of HoxA9-transformed cells virally coexpressing Pbx3 or vector only as control (lower left panel). qPCR quantification of Meis1 RNA in the same cells (lower right panel). HP1, heterochromatin protein 1; KAP1, KRAB-associated protein 1; KRAB, Krüppel-associated box; SETDB1, H3K9 methyltransferase.

Figure 6.

HoxA9 directly regulates cell cycle control genes. (A) Genomic environment, HoxA9 binding, RNA production, and H3K27ac status of cyclin-dependent kinase 6 (Cdk6), cyclin D1 (Ccnd1), and telomerase RNA component (Terc) loci. The line graph shows transcription rates of these genes in response to HoxA9 activity. (B) Core pathways of transformation by HoxA9.