CUX1 is a haploinsufficient tumor suppressor gene on chromosome 7 frequently inactivated in acute myeloid leukemia

Loss of chromosome 7 and del(7q) [−7/del(7q)] was first recognized as a frequent event in acute myeloid leukemia (AML) nearly 40 years ago.¹  −7/del(7q) occurs in 8% of de novo AML²  and 50% of therapy-related myeloid neoplasms (t-MNs).³  −7/del(7q) is also found in myelodysplastic syndromes, AMLs arising from myeloproliferative neoplasms, the blast phase of chronic myelogenous leukemia, Ph⁺ acute lymphoblastic leukemia, and AMLs associated with inherited syndromes.^4-10  −7/del(7q) is an adverse-risk prognostic indicator in myeloid disorders, and the long-term outcome for patients is typically poor. The median overall survival for patients with de novo AML or t-MNs with −7/del(7q) is ∼ 6 months.^2,3 

Loss of 1 or more tumor suppressor gene(s) (TSGs) is thought to contribute to leukemic growth in myeloid malignancies with −7/del(7q). Several groups have mapped a commonly deleted segment (CDS) of chromosome band 7q22 using polymorphic markers, conventional cytogenetic analysis, and FISH analysis.^11-13  In one study of 81 patients with malignant myeloid disorders characterized by chromosome 7 abnormalities, the CDS was mapped to a 2.52-Mb region of 7q22 by FISH using YAC clones.¹¹  However, deletion of a 2-Mb syntenic region in mice did not result in overt myeloid disease.¹⁴  Other studies have mapped rearrangements involving 7q22 and identified similar¹⁵  or slightly more centromeric intervals (Figure 1).^12,13,16 

In this study, we used single nucleotide polymorphism (SNP) arrays to map the 7q22 CDS with resolution of ∼ 1 kb. Overlaying transcriptome-sequencing with copy number aberrations, we identified the gene encoding the transcription factor, CUX1, to be within the CDS and expressed at haploinsufficient levels. Haploinsufficent levels of the CUX1 ortholog, cut (ct), in Drosophila melanogaster hemocytes led to hemocyte overproliferation and melanotic tumor formation in vivo. Similarly, decreased expression of CUX1 led to an engraftment advantage for human hematopoietic progenitors transplanted into immunodeficient mice. Thus, from invertebrates to humans, CUX1/cut is a conserved, haploinsufficient hematopoietic TSG.

This research was approved by the University of Chicago Institutional Review Board. Leukemia samples were obtained from the University of Chicago Hematopathology and Cancer Cytogenetics Laboratories (supplemental Table 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article) with informed consent per the Declaration of Helsinki.

DNA was analyzed on Infinium HumanOmni2.5-Quad Version 1.0 DNA Analysis BeadChips (Illumina). Log R ratio and B allele frequencies generated using GenomeStudio (Illumina) were used to identify copy number aberrations by GenoCN.¹⁷  GEO accession number is GSE42482.

Paired-end libraries were prepared following the standard protocol recommended by Illumina. In brief, purified mRNA (MicroPoly(A) Purist kit, Ambion) was fragmented for 7 minutes at 85°C, and first-strand cDNA generated (Superscript II Reverse Transcriptase, Invitrogen) with random hexamers (Invitrogen), followed by second-strand synthesis with RNaseH and DNA Polymerase I (New England Biolabs). cDNA was repaired and polished with T4 DNA polymerase, Klenow, and T4 PolyNucleotide Kinase (New England Biolabs), followed by adenosine addition with Klenow Fragment 3′ → 5′ exo⁻ (New England Biolabs). Paired-end adaptors (Illumina) were ligated with DNA T4 ligase, and libraries were amplified with p5 and p7 primers (Illumina) and Platinum Pfx taq polymerase (Invitrogen) for 18 PCR cycles; 450-bp fragments were gel-extracted using QIAquick Gel Extraction kit (QIAGEN).

Paired-end reads of lengths 36-100 bp were generated on the Illumina Genome Analyzer II. Reads were individually trimmed from the 3′ end, such that the trimmed 3′ bases had an average Phred-scaled quality score < 15. Reads were aligned to the human reference genome hg18 using TopHat (Version 1.3.1)¹⁸  and output in the BAM format.¹⁹  Further alignment manipulation was performed with SAMTools (Version 0.1.18),¹⁹  Picard (Version 1.52, http://picard.sourceforge.net), and custom perl scripts. Read alignments with mapping qualities < 10 were removed. Alignments were refined with the Genome Analysis Toolkit (Version 1.0.5) RealignerTargetCreator and IndelRealigner tools.²⁰  Cufflinks (Version 1.3.0) was used to estimate transcript abundance using RefSeq gene models and output as Fragments Per Kilobase per Million mapped reads (FPKM) with upper-quartile normalization (supplemental Table 2).²¹  Statistical tests were performed on log₂-transformed data after addition of 1 pseudocount. RNA-sequencing data are in the Short Read Archive SRA061655.

cDNA was generated using the High Capacity RNA-to-cDNA Master Mix (Applied Biosystems). CUX1 was amplified with Power SYBR Green PCR Master Mix (Invitrogen) on the ABI Step One Plus (Applied Biosystems). cut was amplified from cDNA from hemocytes pooled from 6 larvae per group, using EXPRESS qPCR Supermix (Invitrogen) and detected with a dual-labeled probe (IDT). Primer sequences are provided (supplemental Table 3). Samples were run in triplicate, and expression levels relative to a housekeeping control gene were determined using the comparative C_T method according to the manufacturer's software.

deFuse²²  (Version 0.5.0) was used to identify chimeric transcripts requiring > 7 split reads, > 7 spanning reads, splitr_span P value > 0.2, split position P value > 0.2, breakpoint homology < 11, and breakpoints in coding regions. To identify reads mapping to the fusion breakpoint junction, exon-exon junctions between the 2 genes were constructed and all reads were mapped to this junction library with BWA (Version 0.5.9).²³  Reads with < 3 mismatches and > 14 bp anchor region around the junction were identified.

K562, Kasumi-1, KG-1, NB4, Mono7, and U-937 human AML cell lines, and HeLa adenocarcinoma cells were maintained in RPMI 1640-GlutaMAX (Invitrogen) supplemented with 10% fetal bovine serum (Invitrogen), 100 U/mL penicillin, 100 μg/mL streptomycin, and 0.25 μg/mL amphotericin B (Invitrogen). Cells were grown at 37°C in 5% CO₂.

A total of 2 million cells were lysed in RIPA with sodium orthovanadate, protease inhibitor cocktail, and phenylmethylsulfonyl fluoride (Santa Cruz Biotechnology). A total of 40 μg of supernatant, quantified by Micro BCA kit (Thermo Scientific), was electrophoresed on a 4%-15% Mini-PROTEAN TGX gradient gel (Bio-Rad) and transferred to nitrocellulose. Blots were blocked with Odyssey Blocking buffer (LI-COR) and probed with anti-CUX1 at 1:200 (SC-6327, Santa Cruz Biotechnology), followed by donkey anti–goat AlexaFluor-680 (Invitrogen). Mouse anti–β-actin (SC-4778, Santa Cruz Biotechnology) followed by goat anti–mouse IRDye 800CW (LI-COR) was used for normalization. Fluorescence was quantified by Odyssey imaging (LI-COR).

Fly stocks were obtained from the Bloomington Drosophila Stock Center. To express the RNAi in larval hemocytes, hml^Δ-Gal4²⁴  or cg-Gal4²⁵  was used as the driver line to cross with the UAS-RNAi and UAS-2xEGFP lines. The following stocks were used: ct^{RNAi A}, ct^{RNAi B}, and luciferase RNAi (nonspecific control). RNAi-negative control larvae were from the crosses of hml^Δ-Gal4 or cg-Gal4,UAS-2xEGFP and w¹¹¹⁸. All the crosses were set up at 25°C on standard cornmeal media.

Circulating hemocytes from wandering third instar larvae were collected by opening the larval cuticles in 20 μL PBS. A total of 1 μL of hemocytes was fixed in 4% paraformaldehyde (Electron Microscopy Sciences), stained with 10 ng/mL Hoechst 33342 (Invitrogen), and counted using EBImage²⁶  under a DM5000 B microscope (Leica).

Sanger sequencing of CUX1 exons was performed on the 35 samples studied by SNP array plus an additional 109 leukemia samples consisting of 57 de novo AMLs, 47 t-MNs, and 5 AML cell lines (K562, Kasumi-1, KG-1, NB4, and U-937). Whole genome amplified DNA (REPLI-g, QIAGEN) was amplified using Taq 5X Master Mix (New England Biolabs) with primers for CUX1.²⁷  PCR product was treated with 3.3 units Exonuclease I and 0.3 units Shrimp Alkaline Phosphatase (Affymetrix) for 30 minutes at 37°C followed by 15 minutes at 80°C. Sequencing traces were analyzed using MutationSurveyor Version 4.00 software (SoftGenetics).

Lentiviral GFP-IRES-shRNA pGIPZ vectors purchased from Thermo Scientific are as follows: CUX1 shRNA A (clone V2LHS_151077), CUX1 shRNA B (clone V2LHS_151078), and nonspecific shRNA control. Lentivirus was produced by transient transfection of pGIPZ, pCMV-dR8.74, and pMD2 VSVG in HEK293T cells and ultracentrifugation of supernatant. Human cord blood was collected at the University of Chicago. Hematopoietic stem cells were purified from Ficoll gradient-separated white blood cells by Lineage Cell Depletion Kit (Miltenyi Biotec) and cultured for 24 hours in growth media: Stemline II media (Sigma-Aldrich) supplemented with 50 ng/mL each of IL-3, IL-6, thrombopoietin, SCF, and Fms-like tyrosine kinase 3 ligand (R&D Systems). Transduction was performed by spinning cells on Retronectin-coated plates (Takara) with viral stock in Stemline II media at 2500 rpm for 90 minutes at 30°C on day 1 and repeated on day 2. After the spin, supernatant was removed and replaced with growth media. On day 3, cells were assayed for GFP expression by flow and 1 × 10⁵ cells were injected via facial vein in 1- to 2-day-old NOD.Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ (NOD scid gamma, NSG, The Jackson Laboratory) pups after irradiation with 115 cGy. Retro-orbital blood was collected and stained with the following antibodies (eBioscience): anti–mouse CD45-allophycocyanin-Cy7, anti–human CD45- peridinin chlorophyll protein, anti–human CD13-PE, anti–human CD33-allophycocyanin, and anti–human CD3-Pe-Cy7. Flow cytometry was performed on an LSR II (BD Biosciences) and analyzed by FlowJo (TreeStar). Sorting was performed on a FACSAria II (BD Biosciences).

K562 cells were transfected with pGIPZ-shRNA by Nucleofector (Lonza) per the manufacturer's instructions and stably selected with 3 μg/mL puromycin (Invitrogen).

CUX1-associated genes were identified by regressing the expression level of each gene against the expression level of CUX1 using the function lm in the statistical package R. The significance of regression coefficients was quantified by t test. False discovery rates (FDR) were estimated by the Storey q-value method.²⁸  Unless otherwise noted, ± indicates SD, and P values were calculated by t test.

We performed SNP-array analysis of 35 de novo and t-MN leukemia samples and defined a 2.17-Mb CDS on chromosome band 7q22 (Figure 1). Cytogenetic analysis revealed clonal −7/del(7q) in 17 of these. None of the samples showed biallelic deletions of chromosome 7. As seen in previous reports,^11-13  most samples had large deletions of chromosome 7 and were uninformative for defining the CDS of 7q22. One sample, T12, was found to have an interstitial deletion encompassing 7q22 not visible by cytogenetic analysis (Figure 1). Samples T12 and T20 (with a del(7)(q22q36)) had similar proximal breakpoints and defined the proximal boundary, and sample T12 defined the distal boundary of the 2.17-Mb CDS (Figure 1). Although this CDS is defined by a small number of patients, this region overlaps with prior studies.^12,13,16  The CDS within the current study is only 500 kb centromeric to the CDS identified in Le Beau et al¹¹  (Figure 1). The relatively lower resolution of the 500- to 1.5-Mb YAC probes used in this earlier study may have contributed to imprecision in defining the borders of the deleted segment.

To identify genes that are differentially expressed in samples with −7/del(7q), RNA-sequencing was performed on 23 of the leukemia samples analyzed by SNP array. Eleven of the sequenced samples have loss of 7q22. Although sample A72 was reported to have−7[15%]/r(7)(p2?2q2?2)[75%] by cytogenetic analysis, we categorized it as having diploid 7q22 based on the SNP array (Figure 1). The percentage of blasts was similar in the 2 groups with 41.9% ± 30.1% and 49.7% ± 20.8% blasts in −7/del(7q) and other leukemias, respectively. Alignment statistics are provided in supplemental Table 4.

We performed an unbiased, genome-wide comparison of transcript levels in samples with −7/del(7q) versus other samples to define regions of haploinsufficient gene expression. Across all 20 139 expressed transcripts, 4 of the 5 most significantly, differentially expressed transcripts with > 1 fold change are located on chromosome band 7q22.1 and include CUX1 (Figure 2A). The RNA-sequencing dataset thereby provides an independent validation of the CDS defined by SNP-array (Figure 1). The 2.17-Mb CDS contains 49 RefSeq genes, 3 of which exhibited significantly decreased expression after Bonferroni correction (Figure 2B). CUX1 is the most significantly differentially expressed transcript within the CDS.

CUX1 transcript levels in samples with a single CUX1 allele are present at half the level (46.3% ± 20.2%) of samples with 2 copies of the gene (Figure 3A) consistent with a recent report.¹⁶  Transcript expression was confirmed by quantitative RT-PCR. Leukemia samples with loss of 7q22 express 45.8% ± 4.7% of the level of CUX1 expressed by other samples (Figure 3B). Loss of 1 copy of CUX1 is also associated with haploinsufficient protein levels in human AML cell lines (Figure 3C).

As an orthogonal dataset to identify the putative TSG on chromosome 7, RNA-sequencing data were mined for chimeric fusion reads involving genes within the CDS. Nine high-confidence fusion events were identified: 4 validated by Sanger sequencing, 2 were false positives, and there was insufficient patient material to confirm the remaining (Table 1; supplemental Table 5). MLL-MLLT1 and USP42-RUNX1 fusions have been previously described in AML.²⁹  Three expressed fusion events were identified in sample T20, 1 of which contains the first exon of CUX1 (7q22) upstream of exons 2-4 of CLND7 (chromosome band 17p13.1), in the same orientation (Figure 4). The resulting transcript is predicted to contain the codons for the first 21 amino acids of CUX1, followed by 8 amino acids in the wrong reading frame from CLDN7, followed by a premature stop codon, thus truncating 98.6% of the CUX1 transcript. The second allele of CUX1 in T20 probably remains intact based on RNA-sequencing (data not shown) and RT-PCR (Figure 4B). The level of CUX1 in T20 is 47.6% of the level in samples without loss of 7q22 as estimated by FPKM.

deFuse²²  did not identify CUX1 in fusion events in any of the other RNA-sequenced samples. Sanger sequencing of CUX1 exons in 144 AML/t-MNs samples, 39 of which have −7/del(7q), did not reveal any somatic mutations (data not shown). Similarly, there are no mutations in CUX1 reported in 200 AML samples by the Cancer Genome Atlas (tcga-data.nci.nih.gov, Level 2 DNA sequencing somatic mutations, Version 2.2.0). There is 1 reported case of CUX1 mutation acquired during leukemic transformation of a myeloproliferative neoplasm.³⁰  Thus, CUX1 function is disrupted primarily by chromosome loss or deletion (Figure 1) and infrequently by a translocation event (Figure 4) or mutation.³⁰ 

CUX1 is a compelling candidate TSG: it is a homeodomain-containing transcription factor that regulates cell cycle progression and apoptosis and has been implicated in tumorigenesis of solid tumors.³¹  Consistent with a TSG, we uncovered a CUX1-associated cell cycle program within the RNA-sequencing data. Of the 662 genes significantly associated with CUX1 expression (FDR < 0.20, P < .0083), there is a substantial enrichment for the mitotic cell cycle pathway (DAVID, GO:0000278, Benjamini P = 2.57 × 10⁻⁵).³² 

CUX1-associated genes replicate in an independent microarray dataset of 38 subtypes of normal human hematopoietic cells.³³  There is a significant overlap (P = .0063, permutation test) between the top 1000 CUX1-associated genes (FDR < 1.58 × 10⁻¹⁰) in normal hematopoietic cells and CUX1-associated genes in leukemia cells (2300 genes at P < .05). 73.8% of the 126 overlapping genes have a concordant direction (Pearson χ² test P = 6.24 × 10⁻⁷). CUX1-associated genes in normal cells are also enriched for cell cycle (GO:0007049, Benjamini P = .017). Thus, the CUX1-associated cell growth transcriptional profile is present in both leukemia and normal cells.

This transcriptional program is probably a result of direct regulation by CUX1, as many of these genes are bound by CUX1. We overlaid the CUX1-correlated leukemia gene set with previously reported CUX1 genome-wide promoter occupancy data from 8 human cancer cell lines, 5 of which are hematopoietic.³⁴  There is a significant enrichment for direct targets of CUX1 (defined as binding with a P < .0005 in any cell type) within the leukemia CUX1-associated genes (P = .0016, permutation test). The 125 overlapping genes are also enriched for cell cycle genes (GO:0022403, Benjamini P = .05, supplemental Table 6). Nine of the 10 cell cycle genes in the overlapping gene set are inversely correlated with CUX1 expression, further implicating CUX1 as a negative regulator of hematopoietic proliferation.

The hematopoietic system in Drosophila is composed of hemocytes with functions similar to human myeloid cells.³⁵  The development of hemocytes involves signaling molecules and transcription factors that are highly conserved in Drosophila and humans, such as GATA2, RUNX1, ZFPM1 (FOG-1), and NOTCH- and WNT-family orthologs.³⁵  Several genetic lesions in human leukemia result in hemocyte hyperproliferation in Drosophila, including gain-of-function mutations of RAS- and JAK-family genes, and expression of the human RUNX1-RUNX1T1 fusion protein.³⁵  Most human TSGs have Drosophila orthologs with evolutionarily conserved networks that regulate proliferation and differentiation; indeed, much of our understanding of TSGs comes from genetic screens in Drosophila. Ct is highly conserved between humans and Drosophila,³⁶  and ectopic expression of human CUX1 can rescue ct-deficient phenotypes in Drosophila.³⁷ 

To test the hypothesis that ct is a TSG in vivo, ct, which is expressed in hemocyte progenitors,³⁸  was targeted by RNAi specifically in developing Drosophila hemocytes. Expression of ct^{RNAi A} in developing hemocytes led to the development of melanotic pseudotumors³⁵  in both third instar larvae (Figure 5A) and pupae (data not shown). In a small number of cases, tumors were seen as early as the second instar (data not shown). A total of 23.8% (10 of 42) of larvae developed tumors, whereas none (0 of 30) of the control larvae did. To confirm that tumor formation was not the result of an off-target effect of the ct shRNA, we tested a second RNAi construct. Expression of ct^{RNAi B} in hemocytes also led to tumor formation in 16.1% (10 of 62) of larvae.

Melanotic tumors are a phenotype of hemocyte overgrowth³⁵ ; however, we confirmed that there are increased numbers of circulating hemocytes. ct^{RNAi A} and ct^{RNAi B} knockdown larvae have significantly increased numbers of hemocytes compared with nonspecific shRNA-expressing larvae (Figure 5B). In addition, we tested a second hemocyte-specific GAL4-driver.³⁹  Nine of 16 cg-GAL4/UAS-ct^{RNAi A} larvae developed tumors and had significantly increased numbers of circulating hemocytes (33 737 ± 5719 SE in cg-GAL4/UAS-ct^{RNAi A}, n = 7, vs 18 826 ± 1984 cg-GAL4 wild-type, n = 6, P = .042, data not shown). Residual expression of ct transcripts in hemocytes was 38.3% ± 3.8% (ct^{RNAi A}, Figure 5C) and 28.9% ± 6.8% (ct^{RNAi B}) of control, indicating that hemocyte overgrowth was a result of haploinsufficient levels of ct.

To determine whether CUX1 is a conserved TSG in humans, we targeted CUX1 for knockdown in human hematopoietic progenitors. CUX1 is normally expressed highly in multipotent hematopoietic stem cells (Figure 6A).³³ CUX1 expression decreases in common myeloid progenitors and varies significantly in terminally differentiated myeloid cells (Figure 6A).

We used 2 different CUX1-targeting shRNAs, shRNA A and shRNA B. K562 cells expressing shRNA A or shRNA B retain 24.1% ± 3.7% or 55.3% ± 10.9% residual CUX1 protein, respectively (Figure 6B). These shRNAs were transduced in a GFP-expressing lentiviral vector into lineage-depleted human cord blood progenitors and transplanted into NOD scid γ (NSG) mice. CUX1 shRNA-transduced progenitors retained 47.0%-66.9% residual CUX1 transcripts at the time of transplantation (Figure 6C). Transduction efficiency was similar in all groups (Figure 6D). Total engraftment of all human cells was not significantly different in CUX1-targeted recipient mice (Figure 6E). However, there was a significant expansion of the GFP⁺ population in the CUX1-knockdown xenografts (Figure 6E). Compared with nonspecific shRNA-transduced cells, there is a 40% increase in the percentage of CUX1-shRNA–transduced cells. All lineages appeared to be equally expanded in the peripheral blood (Figure 6F), suggesting that CUX1 TSG activity is not restricted to myeloid cells.

Loss of part or all of chromosome 7 is a frequent event in myeloid malignancies and is associated with a poor prognosis, yet it is unknown how these deletions contribute to leukemogenesis. We used high-density SNP arrays and transcriptome sequencing to identify a transcription factor gene, CUX1, as a haploinsufficient TSG within a CDS of chromosome 7. We also characterized a t-MNs sample with an inactivated CUX1 gene resulting from a translocation (Figure 4). CUX1 is normally highly expressed in multipotent hematopoietic progenitors and exhibits dynamic dosage changes during the course of differentiation (Figure 6A). Deletion of a single allele of CUX1 is associated with haploinsufficient transcript and protein levels (Figure 3). To test the TSG activity of CUX1, we first used Drosophila to demonstrate that haploinsufficient levels of the highly conserved Drosophila homolog, ct, leads to hemocyte overproliferation and melanotic tumor formation in developing larvae (Figure 5). In a second in vivo model, partial knockdown of CUX1 in human hematopoietic progenitors leads to increased hematopoietic engraftment on transplantation into NSG mice (Figure 6). These findings indicate that CUX1/cut is a conserved, haploinsufficient TSG that is essential for the regulation of normal hematopoietic cell growth. A proliferative gene signature was identified within CUX1-associated genes, a finding that was replicated in an independent dataset of normal hematopoietic cells (data not shown). A significant number of these genes are direct targets of CUX1 binding.³⁴  This suggests that CUX1 may act as a TSG in myeloid cells by regulating cell cycle components at the transcriptional level.

Deletions of chromosome 7 are typically large with heterogeneity in the breakpoints in myeloid diseases, making it difficult to map the CDS. It is probable that multiple TSGs on chromosome 7 cooperate in leukemogenesis. There may be at least 3 different CDS on chromosome 7, including an interval at 7q36 containing the EZH2 gene.^4,16  A precedent for this is seen in 5q deletions, which are also heterogeneous.⁴ 

We have not excluded the possibility that other differentially expressed genes within the CDS of 7q22.1 may also play a role in disease pathogenesis. However, CUX1, which was implicated as early as 1996,⁴⁰  was identified in a narrower CDS in 2 recent studies of AML arising from myeloproliferative neoplasms.^7,41  Four cryptic deletions were identified: 3 samples had deletions that spanned just the CUX1 and SH2B2 genes,⁴¹  and 1 sample had a deletion containing only CUX1.⁷  The CDS defined in the current study is larger, most probably because of smaller sample size.

We note that biallelic inactivation of CUX1 appears to be infrequent (Figure 1).^7,41  The remaining CUX1 allele is unlikely to be silenced by epigenetic modifications, as samples with −7/del(7q) express CUX1 at 46.3% ± 20.2% of the level of samples with both copies of CUX1 (Figure 3), consistent with a gene dosage effect and expression from the remaining allele. However, rare instances of homozygous deletions of the CUX1 locus and a CUX1 homeodomain mutation in a region of acquired uniparental disomy have been reported in AML arising from myeloproliferative neoplasms.^7,30  Mutations in CUX1 were not demonstrated by Sanger sequencing of CUX1 exons in 144 AML/t-MNs samples, 39 with −7/del(7q) (data not shown). A previous study also failed to find CUX1 mutations in pediatric myeloid neoplasias with −7.⁴² CUX1 is therefore most often inactivated by chromosome loss or deletion but can infrequently be inactivated by mutation³⁰  or translocation (Figure 4). As biallelic inactivation of CUX1 occurs only rarely in myeloid diseases, CUX1 appears to be a haploinsufficient TSG.

Studies in murine models support the hypothesis that CUX1 is a myeloid TSG in mammals. Cux1^ΔHD mice, which have a Cux1 truncation, have increased myeloid cell numbers and myeloid colony-forming units.⁴³  Transgenic expression of the p75 isoform of Cux1, containing 1 Cut repeat and the homeodomain, leads to a myeloproliferative disease in 33% of mice.⁴⁴  It is not clear whether the myeloid expansion is the result of increased Cux1 activity or a dominant negative effect of the transgene.

CUX1/cut was first described in Drosophila, wherein ct is essential for lineage specification in multiple tissues.³¹  Consistent with a developmental role in several cell types, altered Cux1 levels led to abnormal growth of multiple cell types in murine models.³¹  In humans, CUX1 mutations have been reported in 6.7% of lung squamous cell carcinoma⁴⁵  and 2.8% of colorectal carcinoma⁴⁶ ; however, the functional significance of these mutations remains to be tested. The TSG activity of CUX1 may have several mechanisms, including regulation of differentiation, cell cycle progression, genomic stability, and apoptosis, and the mechanism may be cell type specific.³¹  In one report, ct deletion in a Drosophila cancer model promoted invasive tumorigenesis because of both a block in differentiation and increased proliferation.⁴⁷ 

In conclusion, this is the first biologic confirmation of a haploinsufficient myeloid TSG on chromosome band 7q22. The concentrations of hematopoietic transcription factors play a critical role in hematopoietic cell fate, regulating lineage decisions, proliferative signals, and leading to leukemia when concentrations are above or below crucial thresholds. Examples of the detrimental consequences of abnormal gene product dosage can be found in many master regulators of hematopoiesis, including SPI1, RUNX1, and CBFB.⁴⁸  Intriguingly, these genes are typically not completely inactivated when mutated in leukemia but instead retain some level of activity.⁴⁸ CUX1 levels during normal human hematopoiesis appear to change significantly, depending on the developmental stage and lineage (Figure 6A). A gradient of Ct has been shown to be important in neural development in Drosophila: high, medium, and low levels of Ct, measured at the single-cell level, lead to distinct classes of neural cells.⁴⁹  Thus, changes in CUX1 levels may have important phenotypes during hematopoiesis and deleterious consequences when altered.

The authors thank Kevin Shannon for critical reading of the manuscript, Ronald Hause for assistance with statistical analysis, and the TRiP at Harvard Medical School for transgenic RNAi fly stocks. Next-generation sequencing was performed at the University of Chicago High-throughput Genome Analysis Core. Illumina SNP array services were performed at the Northwestern University Genomics Core. Sanger sequencing was performed at the University of Chicago Comprehensive Cancer Center Genomics Core.

This work was supported by a Leukemia & Lymphoma Society Fellow award (M.E.M), the Cancer Research Foundation, National Institutes of Health (CA40046; M.M.L.), and the Chicago Cancer Genomes Project.

National Institutes of Health

Contribution: M.E.M. designed research, performed experiments, analyzed and interpreted data, and wrote the manuscript; C.D.B. designed research, analyzed, and interpreted RNA-sequencing data, and edited the manuscript; X.W. performed Drosophila crosses and hemocyte counting; E.T.B. assisted in analyzing RNA-sequencing data; S.K. assisted in generating sequencing libraries; C.B. identified fusion genes; S.Y. assisted in analyzing SNP data and fusion validation; B.P.S., J.K., and J.M.C. designed and performed xenograft experiments; T.S. assisted in designing research; J.A. performed morphologic analysis and collected biospecimens; M.M.L. performed cytogenetic analysis of leukemia samples, designed research, collected biospecimens, interpreted data, and edited the manuscript; R.L.G. provided computational infrastructure and bioinformatics support; and K.P.W. designed research, interpreted data, and edited the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Megan E. McNerney, Institute for Genomics and Systems Biology, University of Chicago, 900 East 57th St, KCBD 10100A, Chicago, IL 60637; e-mail: megan.mcnerney@uchospitals.edu; and Kevin P. White, Institute for Genomics and Systems Biology, University of Chicago, 900 East 57th St, KCBD 10100A, Chicago, IL 60637; e-mail: kpwhite@uchicago.edu.