Background

Several proto-oncogenes have been demonstrated to be deregulated in human cancer. In particular, the development of the hematologic malignancies such as leukemia, is associated with aberrant expression or function of proto-oncogenes such as c-myc, evi-1, and c-abl. Many translocations with cytogenetic abnormalities that characterize leukemias involve rearrangement of transcription factors, including AML-ETO and Nup98-hox. Some of these leukemia-associated fusion proteins predict prognosis, e.g. t(8,21), t(15,17), and inv(16) are associated with a good prognosis in acute myeloid leukemia (AML) [1]. Approximately 50% of adult patients have been noted to have specific cytogenetic abnormalities. The overall survival of patients with AML is less than 50%. Since half of the patients diagnosed with AML have normal cytogenetic profiles, it is critical to understand the molecular pathways leading to leukemogenesis.

We identified that the cyclic AMP Response Element Binding Protein (CREB) was overexpressed in the majority of bone marrow samples from patients with acute leukemia [2, 3]. CREB is a leucine zipper transcription factor that is a member of the ATF/CREB family of proteins [4–6]. This transcription factor regulates proliferation, differentiation, and survival in a number of cell types, including neuronal and hematopoietic cells [4, 5]. CREB has been shown to be critical in memory and hippocampal development in mice [7, 8]. We previously described that CREB is phosphorylated at serine 133 downstream of signaling by the hematopoietic growth factor, Granulocyte Macrophage-Colony Stimulating Factor (GM-CSF) in myeloid cells [9–11]. We further demonstrated that CREB phosphorylation results from the activation of the Mitogen Activated Protein Kinase (MAPK) and pp90 Ribosomal S6 Kinase (pp90RSK) pathways in response to GM-CSF stimulation [9].

To understand the role of CREB in normal and neoplastichematopoiesis we investigated the expression of CREB in primary cells from patients with acute lymphoblastic (ALL) and myeloid leukemia and found that CREB was overexpressed in the majority of leukemia cells from patients with ALL and AML at the protein and mRNA levels [2, 3, 12]. Furthermore, overexpression of CREB was associated with a worse prognosis. We created CREB transgenic mice that overexpressed CREB in myeloid cells. These mice developed enlarged spleens, high monocyte count, and preleukemia (myeloproliferative disease) after one year. Bone marrow progenitor cells from CREB transgenic mice had increased proliferative capacity and were hypersensitive to growth factors compared to normal hematopoietic stems cells (HSCs). Overexpression of CREB in myeloid leukemia cell lines resulted in increased proliferation, survival, and numbers of cells in S phase [12]. Known target genes of CREB include the cyclins A1 and D [4, 5, 12, 13]. Both of these genes were upregulated in CREB overexpressing cells from mice and human cell lines [4, 5]. Thus, CREB is a critical regulator of leukemic proliferation and survival, at least in part, through its downstream target genes.

CREB target genes have been published on the website developed by Marc Montminy http://natural.salk.edu/CREB/ based on ChIP chip data [14]. Additional CREB target genes were described by Impey et al. [15]. In their studies, serial analysis of chromatin occupancy (SACO) was performed by combining chromatin immunoprecipitation (ChIP) with a modification of Serial Analysis of Gene Expression (SAGE). Using a SACO library derived from rat PC12 cells, approximately 41,000 genomic signature tags (GSTs) were identified that mapped to unique genomic loci. CREB binding was confirmed for all loci supported by multiple GSTs. Of the 6302 loci identified by multiple GSTs, 40% were within 2 kb of the transcriptional start of an annotated gene, 49% were within 1 kb of a CpG island, and 72% were within 1 kb of a putative cAMP-response element (CRE). A large fraction of the SACO loci delineated bidirectional promoters and novel antisense transcripts [15]. These studies suggest that CREB binds many promoters, but only a fraction of the associated genes are activated in any specific lineage. We therefore set out to measure the functional targets of CREB in a hematopoietic model system.

Since CREB is overexpressed in bone marrow cells from patients with acute leukemia compared to normal HSCs, this provides a potential target for leukemia therapy. To this end, we stably transduced myeloid leukemia cells with CREB shRNAlentivirus[16]. CREB knockdown by 80% resulted in decreased proliferation and differentiation of both normal myeloid cells and leukemia cells in vitro and in vivo [16]. However, downregulation of CREB did not affect short-term or long-term engraftment of normal HSCs in bone marrow transplantation assays [16]. To understand the pathways downstream of CREB, we investigated genes that were differentially regulated in CREB shRNA transduced cells. In this paper, we report expression profiling of genes that were differentially regulated in CREB knockdown K562 myeloid leukemia cells and could be potential targets for development of new therapies for acute leukemia.

Methods

Cell lines

The following human leukemia cell lines were transduced with shRNAs: K562 (Iscoves + 10% FCS) and TF-1 (RPMI + 10%FCS + rhGM-CSF. Cells were cultured at 37°C, 5% CO2 and split every 3 to 4 days. Primary AML bone marrow samples were processed as previously described [12]. All human samples were obtained with approval from the Institutional Review Board and consents were signed, according to the Helsinki protocol.

shRNA sequence design and constructs

The CREB specific shRNA sequences were selected and validated based on accepted parameters established by Tuschl et al. [17–19]http://www.rockefeller.edu/labheads/tuschl/sirna.html; CREB shRNA-1, CREB shRNA-2, CREB shRNA-3. Controls included empty vector, luciferaseshRNA, and scrambled shRNA. shRNA sequences are: CREB shRNA-1(5'GCAAATGACAGTTCAAGCCC3'), shRNA-2 (5'GTACAGCTGGCTAACAATGG3'), shRNA-3 (5'GAGAGAGGTCCGTCTAATG3'), LuciferaseshRNA (5'GCCATTCTATCCTCTAGAGGA3'), Scramble shRNA (5'GGACGAACCTGCTGAGATAT3'). Short-hairpin sequences were synthesized as oligonucleotides and annealed according to standard protocol. Annealed shRNAs were then subcloned into pSICO-R shRNA vectors from the Jacks laboratory at MIT [20]. The second generation SIN vector HIV-CSCG was used to produce human shRNA vectors [21].

Microarray analysis

Total RNA (10 μg) was extracted from K562 cells transduced with vector alone or CREB shRNA was submitted to the UCLA DNA Microarray Facility. RNA samples were labeled and hybridized by standard protocol to Affymetrix Gene Chip Human Genome U133+ Array Set HG-U133A array. Gene expression values were calculated using the MAS5 software. The expression values are quantile normalized across all arrays. We obtained the expression profiles for a control set and CREB downregulated K562 cells. A t-test is performed between the two groups to identify significantly differentially regulated genes. The analysis was performed using Matlab (Mathworks, Inc.). We find a significant number of differentially expressed genes, which are either direct or indirect targets of CREB.

To further characterize the data we have aligned CREB binding data from chromatin immunoprecipitation studies with our expression data. The chromatin immunoprecipitation data was obtained from the website http://natural.salk.edu/CREB/[14]. To identify genes that are most significantly bound by CREB and differentially expressed in our knockdown experiment we first filtered genes by their fold change (greater than 1.5 or less than 0.7). Finally, we ranked genes according to the product of the binding and expression P value (jerry_bind_data.xls) (see Additional file 1).

We characterize these genes using three types of analyses: Ingenuity Pathway Analysis (IPA), Gene Ontology term enrichment analysis and tissue distribution. For the former analysis, we used the Ingenuity Pathways Analysis tool on the lists of significant downregulated genes. We then identified functions that were overrepresented among these genes. For the second, we used the DAVID website http://david.abcc.ncifcrf.gov/home.jsp to identify Gene Ontology terms that were enriched in the list.

Finally, we compute the tissue distribution of the 200 genes we identified as functional CREB targets. The tissue specific expression profiles of each gene are obtained from HG_U133A/GNF1H and GNF1M Tissue Atlas Datasets.[22]. We first compute the logarithm of the ratio of the expression intensity of each gene in each tissue, divided by its average intensity across all tissues. We then perform hierarchical clustering of both the genes and the tissues.

Quantitative Real-time PCR

K562 transduced with CREBshRNA(5 × 106) were lysed in Trizol and stored at -80°C prior to RNA extraction. RNA extraction was performed according to a standard protocol supplied by the manufacturer (Invitrogen) and pellets were resuspended in RNAse free water. The cDNA was transcribed with a Superscript RT III based-protocol. DNAse treatment was not performed due to the selection of intron-spanning primers. Quantitative real-time PCR was performed with the SyberGreen reagent (Bio-Rad) in triplicates and analyzed by the standard curve method standardized to the housekeeping gene beta actin[23, 24].

Results and discussion

Since CREB has pleiotropic effects on cell function and potentially activates several genes in hematopoietic and leukemia cells, we performed microarray analysis with total RNA isolated from K562 chronic myeloid leukemia cells transduced with CREB or control shRNA. The comparison of transcriptional profiles in wild type and CREB shRNA transduced K562 cells revealed a large number of differentially expressed genes (see Additional file 2). Among these genes, some are direct targets of CREB, while others are indirect targets. To infer which of these genes was potentially directly regulated by CREB, we combined the expression data with the ChIP-chip data of CREB bound promoters as demonstrated by Marc Montminy[14]. As was previously observed CREB binding sites are highly conserved across different tissues. However, these sites are activated by cAMP in a tissues specific manner. Therefore by combining these two datasets we attempted to uncover the functional CREB sites in hematopoietic tissues.

Our hypothesis for discovering functional CREB sites in hematopoietic cells is that if a gene is found to be differentially expressed in the CREB shRNA K562 transduced cells, and bound by CREB it is likely to be a direct target. To identify these genes we developed a metric that accounts for both the significance of the expression change and binding data for each gene (described in detail in Methods).

Since CREB has been described as both a transcriptional activator (when phosphorylated) and a repressor, we were interested in genes that were both up and downregulated in CREB shRNA transduced cells. The resulting rank ordered list allows us to sort genes by their likelihood of being functional CREB targets in K562 cells. It is difficult to determine, however, where to draw a threshold between the true and false targets. We have decided to restrict our analysis to the top several hundred targets that had both significant changes in expression and binding, as we deemed these to be highly enriched for true versus false targets. However, we do not claim that these are the only functional CREB targets in K562 cells, as the exact number of true targets is difficult to determine. The top down and upregulated genes revealed by this analysis are listed in Tables 1 and 2, and the full list is found in the supplementary materials.

Table 1 Potential CREB target genes.
Full size table
Table 2 Potential CREB target genes.
Full size table

Genes within the downregulated list were BECLIN 1, UBE2B. Both these genes have a cAMP responsive element binding site(s) in their promoters. These genes were selected for further validation because they are known to be involved in autophagy/apoptosis (BECLIN 1), cell cycle/DNA repair (UBE2B) [25–28]. Quantitative real time-polymerase chain reaction (qRT-PCR) with mRNA from AML cell lines (K562 and TF-1) and primary leukemic blasts from a patient with M4-AML was performed. UBE2B expression was significantly reduced in CREB shRNA transduced TF-1 and K562 myeloid leukemia cells compared to controls (Figure 1, p < 0.05). BECLIN and UBE2B were downregulated in primary AML cells transduced with CREB shRNA (Figure 1, p < 0.05).

Figure 1
figure 1

Expression of potential target genes downstream of CREB in myeloid leukemia cells. Primers specific for the UBE2B, BECLIN1, and CREB genes were generated and utilized for quantitative real-time PCR by SyberGreen method (Bio-Rad Inc.) Relative gene expression normalized to the housekeeping gene actin is shown for the following transduced cells: (A) K562 myeloid leukemia cells, (B) TF-1 myeloid leukemia cells, and (C) Human AML-M4 blasts.

Full size image

Having confirmed the validity of our microarray results in these two test cases we set out to characterize the function of the complete list of CREB target genes using two annotation schemes. The first utilizes the annotation contained in the Ingenuity Pathway Analysis software (IPA). This analysis showed that there is a significant enrichment for cell cycle (P < 1e-3) and cancer (P < 1e-3) genes. The full list of genes associated with cancer is shown in Table 3. Many of these genes regulate cell cycle, signaling, DNA repair, or metabolism, which are consistent with previously published results [5, 15]. Furthermore, the role of CREB in the pathogenesis of leukemias has also been described in the literature [2, 3, 12, 29].

Table 3 The subset of CREB target genes associated with cancer according to Ingenuity Pathways Analysis.
Full size table

IPA also allows us to study CREB target genes in the context of protein-protein interactions networks. A network for downregulated genes interacting with CREB is shown in Figure 2, with a subset of the downregulated targets shown in grey, while other genes not in the target list that interact with these, shown in white. Here we see that there is prior literature supporting our analysis that CREB1 regulates PTGS2 (COX2), NR4A3 and TOM1, as depicted by the blue lines. Interestingly, COX2 is an important drug target, and suggests that commonly used COX2 inhibitors may provide a target for acute leukemia.

Figure 2
figure 2

A network depicting interactions between direct CREB targets (shown in grey) and proteins that these interact with (shown in white). PTGS2, NR4A3 and TOM1 are direct CREB targets whose regulation by CREB was previously described in the literature (clue lines). PTGS2 (COX2) emerges as a central player in this network, and is thus implicated as a potential regulator of leukemias.

Full size image

The second analysis that we performed used the terms from Gene Ontology to identify common characteristics among the top K562 CREB targets. Here we find the striking and unexpected result that ten percent of the downregulated targets code for histone genes (P < 1e-10, Table 4). We also performed an analysis of the top upregulated genes but did not find any significant GO terms. Although there is some prior literature indicating that CREB or CREB-related pathways may play a role in regulating histone modifications primarily through the histone acetylase CREB Binding Protein (CBP)[5, 30, 31], the fact that CREB directly regulates the transcription of histone genes in these cells is unexpected.

Table 4 Gene Ontology terms that are enriched among the top CREB targets.
Full size table

To further validate the hypothesis that CREB is an activator of these 20 histone genes, we utilized previously published analyses of the gene promoters to identify consensus CREB binding sequences. The results shown in Table 1 demonstrate that nearly all the histone genes contain CREB half sites along with a TATA box in the vicinity of these. Thus three lines of evidence support the assignment of these 20 histone genes as CREB targets in K562 cells: expression, binding and sequence based.

We examined the distribution of expression of these 20 histone genes across human tissues. The expression data were obtained from the GNF body atlas. We were able to extract expression profiles for 81 histone genes contained in the human genome. Fifteen of these overlapped with the 20 histone CREB targets. We show the expression of all 81 histone genes in Figure 3, where the identity of the 15 CREB target genes is shown in the last row. We see that the 15 genes are clustered into two groups containing more than one gene, with a third group consisting of a single histone HIST1H1C. One of the groups contains histones that are broadly expressed across human tissues, and particularly in all hematopoietic tissues. The second group is instead expressed in a very narrow range of tissues including K562 cells, bone marrow, prostate and thymus.

Figure 3
figure 3

The tissue specific expression of histone genes. Each row of the figure represents a tissue from the GNF Body Atlas (see methods). We show only the top 30 tissues with highest variance of expression of histone genes. Each column represents a histone gene. We use hierarchical clustering to order the rows and columns according to their similarity. Red indicates that the gene is over expressed relative to its mean expression levels across all tissues, and green that it is under expressed. The histone genes that we identify as direct targets of CREB are shown in red in the last row of the figure. We see that many of these are only expressed in a small subset of rapidly dividing tissues along with K562 cells.

Full size image

We examined the expression of three histones that are putative targets of CREB by real time PCR with mRNA from K562, TF-1, and primary cells from patients with AML. The three histones selected were based on our microarray analyses. Our results demonstrated a statistically significant decrease in histonesHIST1H2Bj, HIST1H3B, and HIST2H2AA in K562 and TF-1 cells (Figure 4). Interestingly, in primary cells from a patient with AML, only HIST1H3B and HIST2H2AA, but not HIST1H2BJ expression was decreased with CREB knockdown. These results suggest that histones are differentially expressed in AML and that specific histones are potential targets of CREB. This analysis supports the hypothesis that CREB regulates a subset of histone genes that are normally expressed in a small set of rapidly dividing tissues. These genes are presumably aberrantly activated in K562 and other leukemia cells, and could potentially contribute to the malignant phenotype.

Figure 4
figure 4

Expression of target histone genes is decreased in CREB knockdown myeloid leukemia cells. Primers specific for HIST1H2BJ, HIST1H3B, and HIST2H2AA were generated and utilized for quantitative real-time PCR by the SYBR Green method (Applied Biosystems). Relative gene expression normalized to the housekeeping gene actin is shown for the following transduced cells: (A) K562 myeloid leukemia cells, (B) TF-1 myeloid leukemia cells, and (C) primary AML cells.

Full size image

Conclusion

We have identified a high confidence list of CREB target genes in K562 myeloid leukemia cells. Several important CREB target genes that function in DNA repair, signaling, oncogenesis, and autophagy were identified. These genes provide potential mechanisms by which CREB contributes to the pathogenesis of acute leukemia. Expression of the genes beclin-1 and ube2b was found to be decreased in myeloid leukemia cell lines and primary AML cells in which CREB was downregulated. In addition, we speculate that CREB may have more global effects on transcription, primarily through the regulation of histone genes thereby altering the regulation of DNA replication during the cell cycle.