Pathways, Processes, and Candidate Drugs Associated with a Hoxa Cluster-Dependency Model of Leukemia

High expression of the HOXA cluster correlates with poor clinical outcome in acute myeloid leukemias, particularly those harboring rearrangements of the mixed-lineage-leukemia gene (MLLr). Whilst decreased HOXA expression acts as a readout for candidate experimental therapies, the necessity of the HOXA cluster for leukemia maintenance has not been fully explored. Primary leukemias were generated in hematopoietic stem/progenitor cells from Cre responsive transgenic mice for conditional deletion of the Hoxa locus. Hoxa deletion resulted in reduced proliferation and colony formation in which surviving leukemic cells retained at least one copy of the Hoxa cluster, indicating dependency. Comparative transcriptome analysis of Hoxa wild type and deleted leukemic cells identified a unique gene signature associated with key pathways including transcriptional mis-regulation in cancer, the Fanconi anemia pathway and cell cycle progression. Further bioinformatics analysis of the gene signature identified a number of candidate FDA-approved drugs for potential repurposing in high HOXA expressing cancers including MLLr leukemias. Together these findings support dependency for an MLLr leukemia on Hoxa expression and identified candidate drugs for further therapeutic evaluation.


Introduction
The mixed lineage leukemia gene (MLL) encodes a 431 kD multifunctional protein recently re-designated as lysine-specific methyltransferase 2A (KMT2A). Gene rearrangements involving MLL, termed MLLr, are among the most potent oncogenic drivers of leukemia detected in over 70% of acute lymphoblastic (ALL), up to 50% of acute myeloid leukemia (AML) cases in infants [1][2][3][4] and approximately 10% of adult leukemia including refractory therapy-related cases [5][6][7][8]. Currently, there are no specific therapies approved for MLLr and identifying target and pathway dependencies is critical to improving clinical outcome for these patients.
The Class I homeobox (Hox) genes are well established as downstream targets of MLL, conserved from the ancestral Drosophila melanogaster orthologs HOM-C complex and Trithorax respectively [9][10][11][12][13]. In conjunction with polycomb repressor complex (PRC) proteins, MLL has an essential role in embryogenesis and definitive hematopoiesis through maintenance of Hox gene expression and instigation of progenitor cell proliferation and differentiation [14][15][16]. Complete knockout of Mll is embryonic lethal in mice due in part to loss of Hoxa7, Hoxa9, and Hoxc9, and heterozygous Mll wt/exhibit hematopoietic abnormalities, retarded growth and deformities of the axial skeleton due to dysregulated Hox expression [13,17]. Mll was also shown to be required for normal hematopoietic stem and progenitor cell (HSPC) activity in conditional knock-out models [18,19]. In human, high expression of HOXA cluster genes, is a hallmark of high-risk, refractory AML, particularly MLLr leukemias [16,[20][21][22][23][24] and direct binding of MLL-fusion proteins to HoxA promoter regions results in increased expression of these genes in leukemic models [25].
Whilst several studies indicate a need for Hoxa expression, primarily Hoxa9, in the establishment of MLLr leukemia their absolute requirement for disease progression and maintenance is less clear [26][27][28].
To address this, a MLL-AF9 AML transplantation model (MA9) was generated in a previously reported conditional Hoxa cluster (Hoxa flox/flox ) background [29]. Homozygous deletion of the Hoxa cluster was not tolerated by MLL-AF9. However, significant reduction in expression of Hoxa7, Hoxa9, Hoxa10, and Hoxa11 was demonstrated and a Hoxa del gene signature including upregulation of Mpo, Clk-1, Clk-4, and Ccl6 and downregulation of Angpt1, Emerin, Ddx6, and IL31ra was obtained. Gene set enrichment and associated bioinformatics analysis of the Hoxa del signature identified enriched processes including transcriptional repressor activity, myeloid, monocytic and leukemia differentiation, and canonical pathways including transcriptional mis-regulation in cancer and the Fanconi anemia pathway. Further bioinformatics analysis using connectivity mapping identified candidate drugs for potential repurposing in HOXA expressing cancer models including MLLr and associated leukemias.

Establishment and Validation of Conditional Hoxa Leukemia Models
Fresh HSPCs enriched from bone marrow of Hoxa flox/flox (AFF), MxCre + /Hoxa flox/flox (MAFF), or control CD45.1 mice were transduced with MA9 to generate leukemias. Extended serial re-plating in methylcellulose selected for transformed HSPCs with high proliferation potential. Condensed granulocyte-macrophage colonies (CFU-GM) were produced from MA9 transduction and serial re-plating resulted in increased condensation of the colony ( Figure 1A). Single colonies obtained from P3 cultures were used to generate cell lines. Transformed cells (P3-P5) were subsequently transplanted into sub-lethally irradiated recipient mice to generate leukemias. Recipient mice developed primary leukemias within reported time frames ( Figure 1B) and immunophenotypes (Table S1). Secondary leukemias, generated from direct transplantation of primary leukemias, were more aggressive with all mice succumbing to death within 50 days accompanied by tissue infiltration and splenomegaly ( Figure 1C,D). Comparative gene expression analysis demonstrated increased Hoxa expression in all MA9 leukemias generated, compared to normal bone marrow (NBM), independent of the genetic background ( Figure 2).  The mean values from triplicate experiments are plotted. Significance as calculated by 1 way ANOVA compared to control bone marrow is denoted as * p ≤ 0.05; ** p ≤ 0.01, *** p ≤ 0.001.

Reduced Leukemia Colony Formation Following IFNα-Induced Hoxa Deletion
Direct incubation with interferon-alpha (IFNα) activated Cre in the Mx-1 background only and resulted in visible reduction in MAFF-MA9 colony formation compared to control CD45.1-MA9 cells or MAFF-derived NBM ( Figure 3A upper panel). Direct colony counts demonstrated significant reduction in the number of colonies observed in MAFF-MA9 cells compared to PBS control and no measurable IFNα toxicity at the concentrations used (1U and 2.5U) in MAFF-derived NBM cells ( Figure 3A lower panel).

Reduced Leukemia Colony Formation Following IFNα-Induced Hoxa Deletion
Direct incubation with interferon-alpha (IFNα) activated Cre in the Mx-1 background only and resulted in visible reduction in MAFF-MA9 colony formation compared to control CD45.1-MA9 cells or MAFF-derived NBM ( Figure 3A   The mean values from triplicate experiments are plotted. Significance as calculated by 1 way ANOVA compared to control bone marrow is denoted as * p ≤ 0.05; ** p ≤ 0.01, *** p ≤ 0.001. PCR analysis of gDNA obtained from MAFF-MA9 cells showed levels of the Hoxa del amplicon equivalent with the Hoxa wt amplicon, presumably due to background Cre activation that was not markedly altered by IFNα treatment in bulk cells ( Figure 3B, left panel). To explore the possibility PCR analysis of gDNA obtained from MAFF-MA9 cells showed levels of the Hoxa del amplicon equivalent with the Hoxa wt amplicon, presumably due to background Cre activation that was not markedly altered by IFNα treatment in bulk cells ( Figure 3B, left panel). To explore the possibility that Hoxa del/del cells were out-competed by their Hoxa del/wt counterparts in bulk cultures, 100 individual MAFF-MA9 colonies from each treatment were picked and assayed by PCR. All 100 colonies demonstrated retention of at least one copy of the Hoxa cluster that had been unaffected by IFNα-treatment (sample gel Figure 3B, right panel). Isolation of the Hoxa del amplicon followed by Sanger sequencing confirmed deletion of the loci spanning~100 kb ( Figure 3C and Figure S1).

Extension in Survival Following In Vivo Deletion of Hoxa Cluster
To examine whether propagation of the leukemia was dependent on maintained Hoxa expression engrafted mice that received IFNα-or PBS-treated MAFF-MA9 leukemic cells were further treated in vivo with Poly I:C ( Figure 4A). A significant increased survival following Poly I:C treatment was observed compared to PBS controls ( Figure 4B). All control mice succumbed to disease by 21 days, whereas Poly I:C treated mice survived up to day 26, indicative of reduction in number or function of leukemia repopulating cells (LRCs) due to in vivo Hoxa deletion. A significant increase in survival was also observed in cells that had been pre-treated, ex vivo, with IFNα compared to PBS ( Figure 4B). Additional treatment of these cells with Poly I:C did not further increase survival suggesting that LRCs that escape Hoxa deletion are refractory to further Cre-treatment. In addition, reduced leukemia burden was observed in an independent (MAFF-MA9-Luc+) cohort of transplanted NOD-scid IL2rγnull (NSG) mice using bioimaging ( Figure 4C). Recipient mice that succumbed to leukemia from either treatment arm retained the Hoxa wt allele ( Figure S2). that Hoxa del/del cells were out-competed by their Hoxa del/wt counterparts in bulk cultures, 100 individual MAFF-MA9 colonies from each treatment were picked and assayed by PCR. All 100 colonies demonstrated retention of at least one copy of the Hoxa cluster that had been unaffected by IFNαtreatment (sample gel Figure 3B, right panel). Isolation of the Hoxa del amplicon followed by Sanger sequencing confirmed deletion of the loci spanning ~100 kb ( Figure 3C and Figure S1).

Extension in Survival Following In Vivo Deletion of Hoxa Cluster
To examine whether propagation of the leukemia was dependent on maintained Hoxa expression engrafted mice that received IFNα-or PBS-treated MAFF-MA9 leukemic cells were further treated in vivo with Poly I:C ( Figure 4A). A significant increased survival following Poly I:C treatment was observed compared to PBS controls ( Figure 4B). All control mice succumbed to disease by 21 days, whereas Poly I:C treated mice survived up to day 26, indicative of reduction in number or function of leukemia repopulating cells (LRCs) due to in vivo Hoxa deletion. A significant increase in survival was also observed in cells that had been pre-treated, ex vivo, with IFNα compared to PBS ( Figure 4B). Additional treatment of these cells with Poly I:C did not further increase survival suggesting that LRCs that escape Hoxa deletion are refractory to further Cre-treatment. In addition, reduced leukemia burden was observed in an independent (MAFF-MA9-Luc+) cohort of transplanted NOD-scid IL2rγnull (NSG) mice using bioimaging ( Figure 4C). Recipient mice that succumbed to leukemia from either treatment arm retained the Hoxa wt allele ( Figure S2).

Hoxa del Signature and Associated Pathways
Due to the potential loss of Hoxa del cells over time, Illumina BeadArray-based gene expression profiling was performed on freshly sorted Cre-GFP and GFP-control treated AFF-MA9 cells. Methylcellulose-based colony formation and PCR demonstrated efficient deletion of the Hoxa locus ( Figure S4). Overall output was characterized by plotting significance in the form of negative log10 (Adjusted p-value) versus log2 expression (fold change) for each gene ( Figure 5). Genes with -log10 (Adjusted p-value) > 2 and log2 (fold change) > 0.5 were considered differentially expressed. Transcripts induced or repressed at log2 fold change ≥ 0.5; p ≤ 0.05 were sorted in ascending order by adjusted P-value. One hundred and fifty three probes, representing 135 differentially expressed genes (Table S2) were identified for further bioinformatics evaluation. Due to the potential loss of Hoxa del cells over time, Illumina BeadArray-based gene expression profiling was performed on freshly sorted Cre-GFP and GFP-control treated AFF-MA9 cells. Methylcellulose-based colony formation and PCR demonstrated efficient deletion of the Hoxa locus ( Figure S4). Overall output was characterized by plotting significance in the form of negative log10 (Adjusted p-value) versus log2 expression (fold change) for each gene ( Figure 5). Genes with -log10 (Adjusted p-value) > 2 and log2 (fold change) > 0.5 were considered differentially expressed. Transcripts induced or repressed at log2 fold change ≥ 0.5; p ≤ 0.05 were sorted in ascending order by adjusted P-value. One hundred and fifty three probes, representing 135 differentially expressed genes (Table S2) were identified for further bioinformatics evaluation.

Gene Set Enrichment Analysis
To identify candidate pathways, ontologies and drug interactions, differentially expressed genes were further analyzed by Enrichr [30,31] against curated data sets. In total, 151 probes representing 135 transcripts (123 increased expression, 12 decreased expression) met the criteria for analysis (log2 fold change ≥ 0.5; p ≤ 0.05). A summary of the analysis indicates a role for the Hoxa cluster in key biological processes and pathways ( Figure 6). Cross reference of the Hoxa del signature to the NCBI drug signatures database for gene set analysis (DSigDB) identified several candidate drugs, previously shown to have reported effects in other cancers, for repurposing in HOXA-associated leukemia.

Gene Set Enrichment Analysis
To identify candidate pathways, ontologies and drug interactions, differentially expressed genes were further analyzed by Enrichr [30,31] against curated data sets. In total, 151 probes representing 135 transcripts (123 increased expression, 12 decreased expression) met the criteria for analysis (log2 fold change ≥ 0.5; p ≤ 0.05). A summary of the analysis indicates a role for the Hoxa cluster in key biological processes and pathways ( Figure 6). Cross reference of the Hoxa del signature to the NCBI drug signatures database for gene set analysis (DSigDB) identified several candidate drugs, previously shown to have reported effects in other cancers, for repurposing in HOXA-associated leukemia.  [30,31]. Combined scores (based on the log Fisher exact test p-value and z-score for deviation from expected rank) are plotted against the key pathway, process or drug interactions.

Discussion
The clinical impact of targeted therapy against oncogenic dependency is exemplified by successful treatment of chronic myeloid leukemia (CML) with tyrosine kinase inhibitors (TKIs). In a large cohort study (n = 2662) life expectancy of CML patients that respond to TKIs was reported to approach the life expectancy of the general population [32]. Similar approaches to develop effective targeted therapies for other cancers are ongoing for which identification of oncogenic dependency pathways and processes is a prerequisite.
The association of deregulated HOX gene expression with cancer is well established (reviewed by Grier et al. [33]) particularly in aggressive forms such as acute leukemias (reviewed by Alharbi et al. [34]) in which the Hoxa cluster is predominant. Loss-and gain-of-function studies have identified roles for individual Hoxa genes with normal hematopoiesis including regulation of lineage commitment [35,36], development [37,38], self-renewal, and stem cell expansion [39][40][41]. However, the need for continuous HOXA expression in leukemia initiation and progression remains unclear [26,28,42,43]. This is due in part to functional redundancy within the HOX network and the potential oncogenic roles of non-coding elements within the HOXA locus including HOTTIP [44] and HOXA-AS2 [45].
Herein, we demonstrate the criticality of the Hoxa cluster in the maintenance of established and aggressive MA9 leukemia in a murine model. Conditional Cre-recombinase based deletion of the Hoxa locus resulted in loss of leukemic phenotype in vitro and extension of survival in vivo. This phenotype was associated with differential expression of the Hoxa cluster, primarily Hoxa7-Hoxa11. Resistant leukemias and colonies retained one allele of the Hoxa locus, instead of upregulating paralogs, reflective of an 'escapee' rather than a 'functional redundancy' phenotype. Transplantation  [30,31]. Combined scores (based on the log Fisher exact test p-value and z-score for deviation from expected rank) are plotted against the key pathway, process or drug interactions.

Discussion
The clinical impact of targeted therapy against oncogenic dependency is exemplified by successful treatment of chronic myeloid leukemia (CML) with tyrosine kinase inhibitors (TKIs). In a large cohort study (n = 2662) life expectancy of CML patients that respond to TKIs was reported to approach the life expectancy of the general population [32]. Similar approaches to develop effective targeted therapies for other cancers are ongoing for which identification of oncogenic dependency pathways and processes is a prerequisite.
The association of deregulated HOX gene expression with cancer is well established (reviewed by Grier et al. [33]) particularly in aggressive forms such as acute leukemias (reviewed by Alharbi et al. [34]) in which the Hoxa cluster is predominant. Loss-and gain-of-function studies have identified roles for individual Hoxa genes with normal hematopoiesis including regulation of lineage commitment [35,36], development [37,38], self-renewal, and stem cell expansion [39][40][41]. However, the need for continuous HOXA expression in leukemia initiation and progression remains unclear [26,28,42,43]. This is due in part to functional redundancy within the HOX network and the potential oncogenic roles of non-coding elements within the HOXA locus including HOTTIP [44] and HOXA-AS2 [45].
Herein, we demonstrate the criticality of the Hoxa cluster in the maintenance of established and aggressive MA9 leukemia in a murine model. Conditional Cre-recombinase based deletion of the Hoxa locus resulted in loss of leukemic phenotype in vitro and extension of survival in vivo. This phenotype was associated with differential expression of the Hoxa cluster, primarily Hoxa7-Hoxa11. Resistant leukemias and colonies retained one allele of the Hoxa locus, instead of upregulating paralogs, reflective of an 'escapee' rather than a 'functional redundancy' phenotype. Transplantation of Hoxa del cells resulted in extension in time to leukemia. In contrast to the normal hematopoiesis setting, which tolerates biallelic deletion of the Hoxa locus, albeit with reduced HSPC proliferation and self-renewal [46] biallelic deletion was not tolerated in MA9 cells and all subsequent single-cell generated colonies retained at least one allele of the Hoxa locus. Together, this demonstrates unique dependency on a subset of the Hoxa cluster (Hoxa7, Hoxa9, Hoxa10, and Hoxa11) genes for maintenance of MA9 leukemia.
In association with the well-defined role for HOX in transcription regulation, hematopoietic stem cell self-renewal and maturation, several zinc finger, histone, cell cycle (e.g., Clk-1, Clk-4) and blood cell marker (e.g., Mpo) genes were over expressed following Hoxa cluster deletion. In combination with Hoxa7, a9, and a10, reduced expression of endothelial/adhesion molecules (Heg1 and Angptl1) may be associated with altered Hoxa del cell-niche interactions.
Bioinformatics analysis of the Hoxa del gene signature identified significant enrichment of biological processes associated with blood cell differentiation, nucleosome association and chromatin assembly along with molecular functions including transcription repressor activity. The key pathway associated with the Hoxa del signature was transcriptional misregulation in cancer. Several drugs with anti-cancer properties reported for solid tumors including, Mefloquine in glioblastoma [47], Digitoxigenin in renal carcinoma [48] and Emetine in bladder cancer [49] were associated with the Hoxa del signature. Furthermore, some of these drugs e.g., Emetine and Cephaline were recently shown to be highly active against primary chronic lymphocytic leukemia cells by repressing HIF-1α and disturbing intracellular redox homeostasis [50].
The most represented (n = 3) compound in the top 10 DSigDB drugs associated with the Hoxa del signature, Anisomycin, was recently identified as an expression mimic of mutant RUNX-1 (mt-RUNX-1) knockdown with specific potency against cells from AML patients with germline or somatic mt-RUNX-1 [51]. This is of particular interest as wild type RUNX-1 is essential for the maintenance of MLL-AF9 leukemia [52]. It is intriguing to postulate that upregulation of the HOXA cluster, central to progression in the most refractory AML subtypes and aggressive forms of cancer, is potentially druggable by such loss of expression mimics. Further analysis of the efficacy of such drugs across cell lines and patient samples in conjunction with toxicity profiles against normal cells is warranted to identify and de-risk candidates for potential clinical application.

Deletion of the Hoxa Cluster
In vitro deletion of the Hoxa locus was by IFNα (R&D Systems, Abingdon, UK) treatment at the indicated dose and time. In vivo deletion in MAFF-MA9 leukemic mice was achieved by intraperitoneal (IP) injection of Polyinosinic:polycytidylic acid (Poly I:C; GE Healthcare Life Sciences, Buckinghamshire, UK) as previously described [46]. Briefly, mice were given 10 µg/g Poly I:C or vehicle (PBS) up to a maximum of 250 µg/mouse. Ex vivo deletion was by direct exposure to MSCV-Cre-GFP or MSCV-GFP control retroviral supernatants followed by cell sorting and gene expression profiling. Mice transplanted with traceable (pSLIEW) leukemias were injected IP with 150 mg D-luciferin/kg (Gold Biotechnology, St. Louis, MO, USA) and imaged using the Xenogen IVIS 200 (PerkinElmer, Buckinghamshire, UK). Nested PCR was used to identify individual colony bands.

Gene Expression Analysis and Bioinformatics
Total RNA was isolated using Trizol ® (Life Technologies), and its quality was assessed using Agilent RNA 6000 Nano chip and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) according to the manufacturer's instructions. RNA integrity (RIN) values of > 9 were obtained for all samples prior to application to the Mouse Ref-8 expression v.2 BeadArray (Illumina, San Diego, CA, USA), according to the manufacturer's instructions and as previously reported [55]. Briefly, 550 ng of each RNA was reverse transcribed then in vitro transcribed to generate biotinylated cRNAs. Aliquots of labelled cRNAs (750 ng) were then hybridized to the BeadArray for 16-18 h at 58 • C. Signal detection was achieved using Amersham fluorolink streptavidin-Cy3 (GE Healthcare Bio-Sciences, Little Chalfont, UK) and scanned images obtained using the Illumina BeadArray confocal scanner. Bead level data of biological triplicates were transformed to log2 scale, normalized (quantile method) using beadarray R package [56] and annotated (Gene Symbols, Names, Entrez IDs, and Probe Quality Grades) using illuminaMousev2.db R package. Probes with poor quality grades (and Pgk-1 positive control for Cre-treatment) were removed. Differential expression was assessed by linear regression followed by parametric empirical Bayes analysis using limma R package [57] and output obtained by plotting significance versus log2 expression (fold change). Genes with expression ratios above log2 (0.5)-fold between treatments and control (p < 0.05 and p < 0.01) were identified. False discovery rates (FDR) were controlled using the Benjamini-Hochberg algorithm. Hierarchical clustering with complete linkage and Euclidean distance was performed. The Hoxa del signature obtained was submitted as a fuzzy list to the Enrichr platform [30,31] to identify association with 156-curated libraries. Outputs were tabulated to show the Combined Score as a function of the log Fisher exact test p-value and z-score for deviation from expected rank plotted against the key pathway, process or drug interactions.

Conclusions
In conclusion, our findings directly demonstrate, for the first time, absolute dependency on the Hoxa locus for the maintenance of MA9 leukemia and identifies molecular processes and key pathways for further analysis and candidate drugs for pre-clinical validation and redeployment to this highly refractory disease. Several of the candidates identified here, by association with a Hoxa del phenotype, have demonstrated efficacy in a variety of cancer subtypes. Extension of these findings to other MLLr leukemias or cancers associated with high HOXA expression warrants further investigation. Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6694/11/12/2036/s1, Figure S1: Direct sequence validation of the Hoxa del amplicon, Figure S2: Retention of the Hoxa cluster in treated MAFF-MA9 leukemic mice; Figure S3: Genotyping of MxCre + /Hoxa flox/flox (MAFF) mice, Table S1: Representative Immunophenotype analysis of MLL-AF9 primary leukemias (PL) and cell lines (CL), Table S2: Hoxa del signature (ranked), Table S3: List of oligomers used to define the genotype of experimental mice, bulk and individual colonies (nested).