Gene array analysis reveals a common Runx transcriptional program controlling cell adhesion and survival

Abstract

The Runx genes play divergent roles in development and cancer, where they can act either as oncogenes or tumour suppressors. We compared the effects of ectopic Runx expression in established fibroblasts, where all three genes produce an indistinguishable phenotype entailing epithelioid morphology and increased cell survival under stress conditions. Gene array analysis revealed a strongly overlapping transcriptional signature, with no examples of opposing regulation of the same target gene. A common set of 50 highly regulated genes was identified after further filtering on regulation by inducible RUNX1-ER. This set revealed a strong bias towards genes with annotated roles in cancer and development, and a preponderance of targets encoding extracellular or surface proteins, reflecting the marked effects of Runx on cell adhesion. Furthermore, in silico prediction of resistance to glucocorticoid growth inhibition was confirmed in fibroblasts and lymphoid cells expressing ectopic Runx. The effects of fibroblast expression of common RUNX1 fusion oncoproteins (RUNX1-ETO, TEL-RUNX1, CBFB-MYH11) were also tested. While two direct Runx activation target genes were repressed (Ncam1, Rgc32), the fusion proteins appeared to disrupt regulation of down-regulated targets (Cebpd, Id2, Rgs2) rather than impose constitutive repression. These results elucidate the oncogenic potential of the Runx family and reveal novel targets for therapeutic inhibition.

Introduction

The mammalian genes play unique and non-redundant roles in normal development, as indicated by disruptions of Runx 1, 2 or 3, which give rise to severe deficits in definitive haemopoiesis, bone formation and neurogenesis respectively (Wang et al., 1996; Otto et al., 1997; Ducy et al., 1999; Levanon et al., 2002; Inoue et al., 2002). The gene family has also been widely implicated in cancer where they can play roles both as tumour suppressors and oncogenes according to context (Ito, 2004; Speck and Gilliland, 2002; Blyth et al., 2005). The oncogenic potential of the Runx genes has been demonstrated by their occurrence as targets for retroviral activation in lymphomas of CD2-MYC mice (Stewart et al., 1997; Stewart et al., 2002; Wotton et al., 2002) and confirmed in CD2-Runx2 transgenic mice which are prone to lymphoma development (Vaillant et al., 1999; Blyth et al., 2001). The human RUNX1 can also play a dominant oncogenic role in myeloid leukaemias and childhood B-cell leukaemias through the formation of fusion proteins such as RUNX1-ETO or TEL-RUNX1 by common chromosomal translocation events (Speck and Gilliland, 2002). The tumour suppressive potential of this family been demonstrated most clearly for RUNX1, where haploinsufficiency leads to familial platelet disorder, a rare syndrome that entails predisposition to myeloid leukaemia (Song et al., 1999), and somatic loss of function mutations are observed affecting one or both alleles of RUNX1 in acute myeloid leukaemias (Osato et al., 1999). Evidence for a tumour suppressor role for RUNX3 comes mainly from the observation that this gene is frequently down-regulated and hyper-methylated in various epithelial cancers (Ito, 2004).

The context-specific factors that determine which facet of Runx behaviour is observed are beginning to be understood. In primary murine fibroblasts Runx1 induces senescence-like growth arrest, but this response is lost in cells lacking Arf or p53 function (Linggi et al., 2002; Wotton et al., 2004), which instead acquire tumorigenic properties (Wotton et al., 2004). Ectopic expression of Runx2 in the thymus induces growth arrest in immature T-cells, but this effect is neutralised by co-expression of Myc, and the potent combination of Runx2 and Myc drives rapid tumour expansion (Vaillant et al., 2002; Blyth et al., 2001). While anti-apoptotic functions have been implicated (Blyth et al., 2006), the precise mechanisms by which the Runx genes manifest their oncogenic effects remain largely unknown. Another important question is the extent of functional redundancy between the Runx family members and whether the tissue-specific manifestations of their tumour suppressive potential are due merely to their distinctive patterns of transcription or reflect their unique functional attributes. The three Runx genes encode the α-chains of the core binding factor complex that bind directly to their consensus target sequences, with high-affinity interaction conferred by a common non DNA-binding co-factor, Cbfb (Huang X. et al., 1999). Binding of the core binding factor complex to target promoters can result in recruitment of either activating or repressing complexes (Lutterbach et al., 2000), and the classic example of this functional duality is provided by the opposing effects of Drosophila Runt on the slp1 gene in adjacent embryonal segments (Swantek and Gergen, 2004). The factors regulating this transcriptional switch are not fully understood, but are likely to involve the availability of cofactors as well as post-translational modifications of the Runx proteins themselves (Bae and Lee, 2006). A large number of Runx target genes have already been identified and these include essential components of lineage-specific differentiation programmes which operate in haemopoietic, osteoblastic and neurogenic precursor cells as well as regulators of basic cellular processes and cell cycle control (Otto et al., 2003). It is clear that the Runx transcriptome varies according to cell type, and that understanding Runx regulation in cancer requires the development of cell models that display a relevant phenotype. In this study, we have explored the effects of ectopic expression in immortalised fibroblasts. Our results show that Runx1, 2 and 3 induce a similar phenotype and direct a strongly overlapping transcriptional programme with a common set of target genes that provide novel insights into the pleomorphic effects of Runx expression on cell behaviour and growth regulation.

Results

The Runx genes induce epithelioid transformation and enhanced survival in established murine fibroblasts

We have shown previously that Runx1 can induce morphological transformation and promote tumorigenicity in p53 null primary murine fibroblasts (MEF) while wild-type cells undergo premature senescence in the presence of ectopic Runx expression (Wotton et al., 2004; Kilbey et al., 2007). For the present study we chose to use 3T3 fibroblasts that display a more uniform phenotype compared to MEFs but share permissiveness for Runx expression by virtue of their lack of Ink4a (p16/p19) expression. Into these cells we introduced the full-length P1 isoforms of Runx1, 2 and 3. The transduced cells displayed an epithelioid transformation phenomenon similar to that observed in MEFs (Figure 1a) (Wotton et al., 2004) with a phenotypic shift resembling mesenchymal to epithelial transition (MET) (Chaffer et al., 2007). Other key features of this phenotype included a profound alteration in the distribution of N-cadherin in favour of the plasmamembrane (Figure 1b) and markedly increased expression of integrin β5 (Figure 1c).

Figure 1.

Phenotypic effects of Runx expression in NIH 3T3 cells. a. epithelioid morphology (phase constrast light microscopy): b. increased cell-cell adherence (confocal microscopy, immunofluorescence labelling using N-cadherin anitibody): c. increased cell-matrix adherence (confocal microscopy, immunuofluorescence labelling with integrin β 5 antibody): d. increased survival at confluence (cumulative dead cell count, Trypan Blue exclusion assay of dead cells in suspension at each media change at 3 or 4 day intervals): e. Runx 1, 2 and 3 expression in 3T3 cells detected by western blotting using a pan-runx antibody. As this antibody reacts most efficiently with Runx3 and least efficiently with Runx2, a separate panel is shown where Runx2 is detected with a specific Runx 2 antibody. Cells over-expressing Runx1 (6i) and Runx2 (47i) are included as controls (Stewart et al., 1997; Wotton et al., 2002).

While the proliferation rate of Runx expressing cells was not found to be increased, a potentially important survival advantage was noted, particularly under conditions of stress such as medium exhaustion. This phenomenon is illustrated in Figure 1d where the death rates of control and Runx expressing cell cultures held at confluence, with periodic medium change, are compared. While results for Runx1 are illustrated, essentially identical observations were made for the other two genes. Ectopic Runx expression was stable and readily detected in these cells by western blot analysis (Figure 1e).

The Runx genes regulate a strongly overlapping transcriptional programme

To analyse the Runx phenotype at the level of global transcriptional changes, we harvested RNA from newly confluent cultures which displayed the most marked phenotypic alteration and carried out a gene expression microarray analysis, comparing cells expressing vector only with those transfected with each of the Runx genes. The primary array data have been deposited in the Gene Ontology Omnibus (accession number pending). Large numbers of genes were found to be changed, with a more or less equal number up- or down-regulated (Figure 2a). However, a striking overlap was observed between the three genes. This is illustrated clearly in the bar diagrams in Figure 2b, where the 150 genes falling within the 1% FDR cut-off are grouped according to their regulation by one or more of the Runx genes. Extending the window to 10% FDR increased representation to encompass 81% of the genes showing regulation by more than one family member. Moreover, analysis of genes beyond this cut off shows that, with few exceptions, these are similarly albeit more weakly regulated by other family members (not shown). Strikingly, we found no examples of genes that were significantly regulated by one family member (5% FDR) and regulated in the opposite direction by another family member. These results suggest that Runx transcriptional function is highly conserved, although clearly some divergence in regulatory potency has developed over evolutionary distances.

Figure 2.

Results of gene microarray analysis. a. The number of genes showing significant up- or down-regulation by Runx1, 2 and 3 at 1% and 5% false discovery rate cut-off (Rank Products method (Breitling et al., 2004)). b. Overlap of the Runx transcriptome is shown by the large number of genes significantly regulated by more than one family member. The chart includes 150 high confidence genes regulated by any Runx gene at 1% FDR. Relaxing stringency to 5% and 10% FDR results in inclusion of most of these target genes in a common transcriptome: c. Venn diagram highlighted the filtration strategy for further analysis. Of 69 regulated by all three constitutively expressed Runx genes at 5% FDR, 50 genes were selected on the basis of regulation by an inducible RUNX1-ER™ construct. The gel depicts the molecular weight shift of RUNX1-ER induced in the presence of 4-OHT.

To gain further insight into modes of Runx regulation, a functionally inducible Runx protein (RUNX1-ER) was introduced into the same fibroblast cells, and transcriptional profiles were compared after 24 hours of 4-OHT induction. Complete conversion of this fusion protein to its transcriptionally active form was confirmed by its rapid and stable post-translational modification in the presence of inducer (Figure 2c). Most of the target gene set displayed significant regulation under these conditions and we used this criterion as a further filter to define a subset of 50, Runx modulated genes that are candidates for direct regulation (Figure 2c).

Runx genes regulate many targets with established roles in cancer and development

Extensive validation was carried out to confirm the robustness of the array data. Figure 3a compares the results of the fold-change measurements for a subset of genes, on replicate chips with analysis by quantitative real-time PCR. As can be seen, there was a very strong correspondence across a large and representative subset of the target genes.

Figure 3.

a. Array data were validated for selected target genes by quantitative real-time PCR (SYBR green). Adjacent bars show fold-regulation as measured by Affymetrix array and qt-RT-PCR as indicated. b. Bioinformatic analysis of the selected 50 Runx target genes (see also Table 1). The most common Gene Ontology terms and associated P values (http://david.abccncifcrf.gov/) are listed at the top of the figure. Top biological functions and associated P values were based on Ingenuity Pathways Analysis (www.ingenuity.com). The range of P values refer to sub-categories within each heading. The top two categories are listed for diseases and disorders, molecular and cellular functions, and physiological system development and function. The group numbers correspond to functional annotations listed in Table 1.

The 50 gene Runx target set was analysed further with the aid of bioinformatics programmes (Figure 3b). The most highly over-represented Gene Ontology terms were identified (http://david.abcc.ncifcrf.gov/), showing a remarkable preponderance of genes encoding cell surface or extracellular ligands (36%). Regulated genes encoding cell surface proteins included Itgb5, indicating that transcriptional control is involved in its upregulation by Runx noted earlier (Figure 1c). Since smaller gene sets required fewer recorded hits to reach statistical significance, the occurrence of three enzymes involved in sphingolipid metabolism was also noteworthy (Sgpp1, Ugcg, St3gal5). Sorting according to over-represented biological functions (www.ingenuity.com) was also highly informative, revealing many with annotated roles in cancer (32%), and molecular and cellular processes including cellular development (26%), cellular growth and proliferation (40%). The top 50 genes are listed in Table 1 along with the fold-change estimates from the array analysis and annotations according to the bioinformatic groupings defined in Figure 3b.

Table 1. Common target genes regulated by all three Runx family members.

List of 50 common target genes selected on the basis of similar regulation by Runx1, 2 and 3 and inducible regulation by RUNX1-ER . The basis of filtration of the gene sets is shown in Figure 2. Upregulated and downregulated genes are listed separately in alphabetical order. The fold-change levels estimated from duplicate chips are indicated for each Runx gene.

Gene name	Probe ID	GeneID	Gene	Runx1	Runx2	Runx3	Functional annotation*
Up-regulated
Angiopoietin-like 4	1417130_s_at	NM_020581	Angptl4	2.14	1.85	2.28	1,4,6,9,11
Chemokine(C-C motif) ligand 2	1420380_at	AF065933	Ccl2	2.22	2	3.78	1,5,6,7,8,9,10,11
Chemokine(C-C motif) ligand 7	1421228_at	AF128193	Ccl7	2.92	2.34	3.29	1,5,6,8,9,10,11
Cytochrome P-450, family1, subfamily pptide1	1416612_at	B1251808	Cyp1b1	2.45	5.28	8.75	1,2,6
Elastin microfibril interfacer 1	1416414_at	NM_133918	Emilin1	3.17	4.4	9.14	1,5
IBR domain containing 3	1432478_a_at	AK015966	Ibrdc3	2.08	2.09	2.05
Integrin beta 5	1456195_x_at	BB543979	Itgb5	1.68	1.96	2.71	1,4,5,6,7,9,10,11
Lymphocyte antigen 6 containing complex	1453304_s_at	BM245572	Ly6e	2.8	3.06	2.02	1,9
Mannan-binding lectin serine protease 1	1425985_s_at	AB049755	Masp1	2.76	2.21	2.82	1
Neural cell adhesion molecule	1426865_a_at	BB698413	Ncam1	1.93	2.37	3.45	1,4,5,6,7,8,9,10
PCTAIRE-motif protein kinase 3	1449151_at	NM_008795	Pctk3	2.47	2.16	5.22
Retinol dehydrogenase 10 (all-trans)	1426968_a_at	BG073496	Rdh10	5.26	7.86	6.72
Response to complement gene 32	1418003_at	NM_025427	Rgc32	3.2	3.12	2.74	5
SERTA domain containing 4	1454877_at	BQ174721	Sertad4	3.05	2.46	4.3
Solute carrier family 40, member 1	1417061_at	AF226613	Slc40a1	2.78	1.99	6.18	2
ST3 β-galactoside alpha-2,3-sialyltransferase 5	1460241_a_at	BB829192	St3gal5	2.12	2.18	3.11	3
Tissue inhibitor of metalloproteinase 4	1423405_at	BI788452	Timp4	2.42	2.18	50.69	1,6,9,11
UDP-glucose ceramide transferase	1421269_at	AA591863	Ugcg	3.17	5.66	6.82	1,3,4,6,7
WW domain E3 ubiquitin protein ligase 2	1448145_at	AK004087	Wwp2	2.09	3.12	3.54
2610528A11Rik	1435639_at	BF580962	N/A	4.81	2.05	12.11
6330416G13Rik	1426315_a_at	AV326978	N/A	1.8	3.17	4.52
Down-regulated
Acid phosphatase-like 2	1456735_x_at	BB458645	Acpl2	−2.25	−2.25	−3.04
Adducin3 (γ)	1423297_at	BM239842	Add3	−1.68	−1.9	−2.68	5
Activated leukocyte cell adhesion molecule	1437467_at	AV315205	Alcam	−1.76	−2.16	−2.95	4,5,6
Cadherin-like 26	1456937_at	BB748621	Cdh26	−1.93	−3.52	−3.47	5
CCAAT/enhancer binding protein (C/EBP) δ	1423233_at	BB831146	Cebpd	−1.98	−3.32	−5.06	1,5,6,8,9,10
Erbb receptor feeback inhibitor 1 (MIG-6)	1416129_at	NM_133753	ErrFi1	−1.89	−2.24	−4.17	5,9
EGL nine homolog 3	1418649_at	BB284358	Egln3	−2.66	−2.07	−5.99	2,9
Enolase2, γ neuronal	1418829_a_at	NM_013509	Eno2	−1.87	−3.75	−3.89
Ferredoxin 1	1449108_at	D43690	Fdx1	−1.76	−2.26	−2.62	1
Fibronectin type III domain containing 3a	1426903_at	BC022140	Fndc3a	−1.69	−1.97	−2.64
Growth arrest specific 2	1450112_a_at	NM_00807	Gas2	−1.75	−2.17	−3.31	4,8,9,10
Inhibitor of DNA binding 2	1435176_a_at	BF19883	Id2	−3.21	−4.74	−6.55	4,5,6,7,8,9,10,11
Inhibitor of DNA binding 3	1416630_at	NM_008321	Id3	−2.33	−2.82	−2.64	4,5,6,8,9,10,11
Kit ligand	1415855_at	BB815530	Kitl	−1.63	−2.06	−2.69	1,5
Mesothelin	1460238_at	NM_018857	Msln	−1.82	−2.16	−4.92	1,9
Nuclear receptor subfamily 3, group C member 1*	1457635_s_at	BB096079	Nr3c1	−1.72	−1.97	−2.66	5,7,8,10
Odd Oz/ten-m homologue 4	1434530_at	BQ175876	Odz4	−2.35	−2.37	−2.66	4
Phosphodiesterase 8B	1437989_at	BB357157	Pde8b	−2.46	−4.41	−11.13
Procollagen lys, 2-oxoglutarate 5-dioxygenase 2	1416686_at	BC021352	Plod2	−1.92	−2.31	−2.28	1,2
Paired related homeobox 1	1425527_at	L06502	Prrx1	−2.03	−1.7	−3.73	4,8,9,10
Receptor (calcitonin) activity modifying protein 3	1420401_a_at	NM_019511	Ramp3	−1.82	−2.02	−3.06	1,10
Regulator of G-protein signalling 2	1447830_s_at	BB034265	Rgs2	−3.55	−4.4	−10.54	9,10,11
Ring finger protein 144	1438404_at	BB125272	Rnf144	−2.63	−1.84	−2.76
Sphingosine-1-phosphate phosphatase 1	1420821_at	NM_030750	Sgpp1	−1.75	−2.11	−2.96	3,6
Signal induced proliferation-associated 1 like 2	1434261_at	AV228782	Sipa1l2	−1.77	−1.94	−2.32	5
Sine oculis-related homeobox 1 homologue	1427277_at	BB137929	Six 1	−2.83	−1.89	−3.42	4,5
Transducer of ERBB2, 2	1448666_s_at	AV174616	Tob2	−1.64	−1.84	−2.36	9
Vasoactive intestinal polypeptide	1428664_at	AK018599	Vip	−3.77	−4.18	−4.01	1,5
Xanthine dehydrogenase	1451006_at	AV286265	Xdh	−1.86	−2.51	−3.46	2,4,6,7,8,9,10

Gene	Probe ID	GeneID (1)	Gene ID (2)	Gene name	RUNX1-ER uniind.FR	RUNX1-ER unind.FDR	RUNX1-ER ind.FR	RUNX1-ER ind. FDR	Runx1 FR	Runx1 FDR	Runx2 FR	Runx2 FDR	Runx3 FR	Runx3 FDR	Functional relevance
Esr1	1421244_at	NM_007956	Mm.9213	estrogen receptor 1 (alpha)	5.48	0
	1451382_at	BC025169	Mm.35083	1810008K03Rik	2.62	0
Mtus1	1436501_at	BB747681	Mm.149438	mitochondrial tumour suppressor 1	2.2	0.33
	1454824_s_at	BB699957	Mm.149438		1.75	0.76
	1436502_at	BB747681	Mm.149438		1.39	37.34
Gdap10	1420342_at	NM_010268	Mm.358597	Ganglioside-induced differentiation-associated-protein 10	2.38	0.25
	1451533_at	BC022687	Mm.186678	BC022687	1.97	0.4
Rx1t1	1437784_at	AW550878			2	0.5
Cbfa2t1h	1444615_x_at	AV327778	Mm.4909	CBFA2T1 identified gene homologue (human)	2.01	0.5
	1427640_a_at	X79989	Mm.4909		1.48	15.45
	1448785_at	BG072085	Mm.4909		1.4	37.34
Adm	1447839_x_at	AV378441	Mm.1408	Adrenomedullin	2.02	0.43
	1416077_at	NM_009627	Mm.1408		1.86	0.78
Lcp1	1415983_at	NM_008879	Mm.153911	lymphocyte cytosoloic protein 1	1.94	0.38
Ler3	14196347_a_at	NM_133662	Mm.25613	immediate early response 3	2	0.56
Arf6	1418824_at	BI24938	Mm.27308	ADP-ribosylation factor 6	2.19	0.67
	1422924_at	NM_009404		NM_009404	1.9	0.62

^*Functional annotation is based on the occurrence of each target gene in the sets defined in Figure 3b.

Few previously identified Runx targets were present in the gene set, emphasising the cell-type specific nature of the Runx transcriptome. For most targets listed in Table 1, the most intense regulation was noted for Runx3. This was probably due in most cases to the high Runx3 expression that was achieved in this experiment, as indicated by further biological replicates and real-time PCR measurements (not shown), but it is likely that there are some real quantitative differences as some genes showed markedly more potent regulation by Runx3 compared to constitutive Runx1, 2 expressing cells. A notable example is Timp4.

Validation of direct Runx targets

A number of known and novel Runx targets were selected for closer examination and determination of the regulatory mechanisms involved. A 24 hour induction period allows time for many indirect changes to transcript levels and gives only a preliminary indication of those genes under direct regulation. We therefore examined the kinetics of change in transcript levels on induction of RUNX1-ER with 4-OHT for a set of 6 genes (Figure 4a). This analysis showed that three of the selected genes were rapidly and strongly regulated (Cebpd, Rgc32, Rgs2) while two others showed a complex response where an initial dip in transcript levels preceded induction (Cyp1b1, Itgb5). The sixth gene, Nr3c1, showed an early but relatively modest level of regulation. As a further test for direct regulation we examined the requirement for protein synthesis by inducing RUNX1-ER in the presence of cycloheximide. While interpretation of these experiments was complicated by the non-specific effects of cycloheximide on stability of some transcripts, it was clear that induction of Rgc32 and repression of Rgs2 were insensitive to cycloheximide inhibition, while Nr3c1 down-regulation was completely abolished (not shown).

Figure 4.

4a. Kinetics of regulation of selected target genes by RUNX1-ER™. Transcript levels were measured by qt-RT-PCR analysis 1, 2 and 6 hours after induction of the construct with 4-OHT. 4b. Runx2 occupation of promoter regions of Rgs2,C/EBPδ, and Rgc32 in vivo as assessed by ChIP analysis. Specificity was demonstrated by lack of site recognition by a negative control polyclonal antisera, α-GST or when primary antibody was not included in the immunoprecipitation reaction. Primers flanking a previously defined Runx binding site in the osteocalcin gene promoter (Bglap2) were used as a positive control.

Further support for direct regulation of the most strongly regulated genes (Cebpd, Rgc32 and Rgs2) was obtained by scanning the promoter regions for consensus Runx target sites and confirming occupancy of these sites in vivo by chromatin immunoprecipitation (Figure 4b). It appears, therefore, that induction kinetics may be a useful predictor of the mode of Runx regulation.

Runx expression mediates resistance to glucocorticoid-mediated growth arrest and apoptosis

Despite the presence of many genes relevant to cancer, the highly regulated gene set revealed few candidate mediators of apoptosis control. A prominent exception was Nr3c1, the gene encoding glucocorticoid receptor, which was consistently down-regulated in the presence of each of the three Runx genes (Table 1) and quite rapidly modulated by RUNX1-ER, albeit by an apparently indirect mechanism. In silico prediction of glucocorticoid resistance was tested directly by exposing Runx expressing fibroblasts to dexamethasone (Figure 5a). In empty vector transduced cells, dexamethasone was found to be strongly growth inhibitory to control cells and induced death above background levels. Both effects were much less marked in cells expressing Runx 1.

Figure 5.

Resistance to dexamethasone growth arrest and apoptosis induced by ectopic Runx expression. a. 3T3 fibroblasts expressing Runx1 were compared to control vector-transduced cells. The graph shows live cell counts and % dead cells measured by Trypan Blue exclusion in the presence or absence of 10μM dexamethasone. Both live and dead counts are significantly different from control values (p = <0.001, 0.015 respectively); b. Lymphoma cells expressing ectopic Runx1 or transduced with the control vector were cultured at high density in the presence and absence of dexamethasone (10μM). Trypan Blue exclusion assay was used to measure live/dead cells over a 4 day time course. Significant resistance to dexamethasone-induced death was observed in Runx1 expressing cells at 2 days (p= 0.018), 3 days (p = <0.001), and 4 days (p = <0.001). The adjacent gel shows relative levels of expression in Runx1 transduced (+) and control (−) cells. The relative level of Nr3c1 transcripts in the same cells was assessed by SYBR green qt-RT-PCR. RQ denotes relative quantitation.

Runx expression is markedly anti-apoptotic in T-cell lymphoma in vivo (Blyth et al., 2006) so we extended the functional analysis to T-cells in vitro by expressing Runx1 in readily transduced T-cell lymphoma cell lines established from p53 null mice (Blyth et al., 1995). A similar level (2-fold) of down-regulation of Nr3c1 in the presence of ectopic Runx was found by qt-RT-PCR as had been observed in the fibroblasts. Although the lymphocytes over-expressing Runx1 showed no obvious survival advantage in the absence of stress, resistance to dexamethasone-induced apoptosis was seen in high density cultures. Results for one such line (pm97) are shown in Figure 5b.

RUNX1 oncoproteins disrupt Runx target gene regulation but do not behave as constitutive repressors

In human leukaemia, RUNX1 is a frequent target for chromosomal translocations that result in N or C-terminal fusions of truncated RUNX1 proteins to heterologous partners. Since the major RUNX1 fusion partners (ETO, TEL) harbour transcriptional repression domains that recruit co-repressors such as histone deacetylases to the target promoters, it has been postulated that these fusion events convert RUNX1 from a conditional transcriptional activator/repressor to a constitutive repressor of its target genes (Fenrick et al., 1999; Amann et al., 2001). A similar hypothesis has been advanced to explain the oncogenic potential of the CBFB/MYH11 fusion in which a truncated version of the common RUNX cofactor CBFB is conjoined with a smooth muscle myosin isoform (Durst et al., 2003). To test this model further, we expressed RUNX1-ETO, TEL-RUNX1 and CBFB-MYH11 in 3T3 cells (Figure 6a). These cells did not adopt the characteristic Runx morphology and displayed no significant difference in growth or survival characteristics (not shown).

Figure 6.

Regulation of common Runx target genes by fusion oncoproteins RUNX1-ETO, TEL-RUNX1 and CBFB/MYH11. 3T3 fibroblasts stably expressing each of the three Runx family members or one of the three fusion oncoproteins were analysed by qt-RT-PCR for expression of selected Runx upregulated genes (Ncam1, Rgc32) and down-regulated genes (Cebpd, Id2, Rgs2). Fold-change compared to control cells is indicated. The lower panel shows western blot analyses of the corresponding cell cultures confirming expression of the fusion oncoproteins (+) by comparison with vector transduced control cells (−). Endogenous ETO is indicated and expressed at high levels in control and RUNX1-ETO transduced cells.

Quantitative RT-PCR analysis of representative genes from the Runx target set is shown in Figure 6. Two of the target genes (Ncam1, Rgc32) behaved as predicted according to the constitutive expression model, with reduced expression in the presence of all three fusion proteins. However, none of the Runx-repressed target genes showed a similar response to the fusion oncoproteins, and showed instead a tendency towards de-repression compared to control cells. In particular Cepbd was markedly increased in the presence of the CBFB-MYH11 fusion. These results do not support the constitutive repressor model for the fusion oncoproteins.

Discussion

In this study we have shown that ectopic expression of the three Runx genes in the same cell background induces an indistinguishable phenotype and a strongly overlapping change in global transcription patterns. These observations confirm and greatly extend previous studies based on a small number of target genes which suggested a high degree of functional redundancy in the Runx family, despite their acquisition of unique developmental roles during vertebrate evolution (Javed et al., 2000). Quantitative differences may be significant, however, and could account for the failure of chimeric genes with C-terminal elements of Runx2 and Runx3 to completely rescue haemopoietic development in Runx1 deficient mice (Fukushima-Nakase et al., 2005).

The phenotype conferred by Runx expression in 3T3 fibroblasts is also of interest and potentially revealing with regard to the complex relationship of the Runx genes to cancer development. We have described the phenotype as epithelioid transformation, based on our previous observations on p53-/- MEFs which adopt a similar phenotype and display increased tumorigenicity under the influence of ectopic Runx expression (Wotton et al., 2004). This process resembles mesenchymal to epithelial transition, the mirror image of the process commonly associated with invasive potential in cancer. While this might appear to represent a tumour suppressive or anti-metastatic feature of Runx, it should be noted that MET may be important in establishing tumours at new niches after metastatic spread (Chaffer et al., 2007). Moreover, our previous studies in MEFs showed that Runx1 slows but does not prevent epithelial to mesenchymal transition in the presence of TGFβ (Wotton et al., 2004). The survival advantage we have described here may be more a decisive factor than cell morphology in the increased tumorigenicity of Runx expressing cells.

The transcriptional signature of ectopic Runx expression reflects both the epithelioid phenotype and the ambiguous roles of the Runx genes in cancer. Thus, the up-regulated set included genes over-expressed in cancer or associated with increased survival (Ccl2, Cyp1b1, Itgb5, Ncam1, Rgc32, Ugcg) while the down-regulated set included numerous genes with tumour suppressive or anti-proliferative potential (Cebpd, Egln3, Errfi1/Mig6, Gas2, Nr3c1, Rgs2, Sgpp1, Tob2). However, other genes displayed the contrary relationship with up-regulation of some ligands that have been reported to inhibit tumorigenic potential (Ccl7, Timp4) and down-regulation of putative oncogenic or pro-survival factors (Alcam, Kitl, Msln, Six1). Other genes from the set have context-specific roles analogous to the Runx family (Angptl4, Id2, Id3).

A surprising feature of the present study was the paucity of Runx regulated genes with an annotated role in control of apoptosis, suggesting that the Runx genes exploit a novel set of pathways to increase cell survival. As the glucocorticoid receptor gene Nr3c1 is down-regulated by all three Runx genes, albeit indirectly, we tested the sensitivity of Runx expressing cells to dexamethasone. We found significantly increased resistance in both fibroblast cells and in lymphoid cells. These results are of considerable interest in light of the profound anti-apoptotic effects of Runx2 expression in the lymphoid compartment in vivo, particularly in the context of Myc oncogene over-expression (Blyth et al., 2006). It seems unlikely that glucocorticoid receptor down-regulation alone is sufficient to explain the marked resistance we observed, and it is notable that the common Runx target set also includes enzymes involved in sphingolipid metabolism (Sgpp1, Ugcg, St3gal5) that have been shown to be involved in apoptosis regulation (Le Stunff et al., 2002; Bleicher and Cabot, 2002) and in cross-talk with glucocorticoid-induced inhibitory signals (Bianchini et al., 2006). It will be of interest to explore the wider role of Runx regulation in ceramide-sphingosine metabolism and tumour chemo-resistance. In this regard it is also interesting to note that the Runx genes emerged as preferred targets for retroviral activation in a murine model of imatinib resistance (Miething et al., 2007) and that the latter phenomenon has been linked in vitro to altered ceramide metabolism (Baran et al., 2007).

Finally, we tested the effects of ectopic expression of the three most important core binding factor fusion proteins which arise from chromosomal rearrangements in human leukaemia (RUNX1-ETO, TEL-RUNX1 and CBFB-MYH11). A model has been advanced, whereby the fusion proteins act as constitutive repressors of Runx gene targets, while the normal Runx complexes are able to activate or repress the same targets in a context-specific manner (Fenrick et al., 1999; Amann et al., 2001; Durst et al., 2003). These models are based largely on the use of exogenous reporter constructs based on model promoters or natural activation targets for Runx such as TCRβ. While we found that some activation targets of Runx behaved in the predicted manner (Ncam1, Rgc32), displaying strong activation by all three wild-type genes and repression by all three fusion oncoproteins, the Runx-repressed genes showed an unexpected pattern where the fusion proteins generally disrupted or reversed repression. This could not simply be ascribed to indirect regulation, as a number of unequivocal direct targets showed similar behaviour (Cebpd, Rgs2). We suggest that a more complex model is required to account for the behaviour of the fusion proteins that may be able, for example, to disrupt or destabilise Runx repressive complexes at target gene promoters. The operation of alternative Runx repressive mechanisms such as inhibition of transcriptional elongation should also be considered (Jiang et al., 2005) and it is notable that the Runx binding site we mapped in Rgs2 lies within the transcription unit. Whatever the mechanism involved, these results reinforce the view that the unique oncogenic properties of the fusion proteins entail more than merely loss of Runx function.

Materials and Methods

Cells and constructs

The fibroblast cell line was derived from the NIH 3T3 cell line and maintained as previously described (Mann et al., 1983). Lymphocyte cell lines were previously established from p53 null mice and grown as described (Blyth et al., 1995). Retroviral vectors were based on the pBabe plasmid (Morgenstern and Land, 1990), carrying the puromycin selectable marker. The Runx1 construct contains a 1.6kb cDNA fragment encoding the Runx1P1 isoform; the Runx2 is a 2.3kb fragment encoding the Runx2P1 isoform; the Runx 3 is a 1.3kb fragment encoding the Runx3P1 isoform . All were subcloned into the polylinker region of pBabe-PURO. The RUNX1-ER construct (pBabe-AML-ER) has been previously described (Lou et al., 2000) and encodes the RUNX1P1 isoform conjoined to a modified estrogen receptor (hereafter referred to as pBabe-RUNX1-ER) inducible in the presence of 4-hydroxytamoxifen (4-OHT, Sigma). 4-OHT dissolved in ethanol (0.1μM) was added at a final concentration of 200nM. Equivalent volumes of ethanol were added to control cultures. AML-1/ETO (RUNX1-ETO) and CBFB-MYH11 (Inv16) constructs were obtained as clones in the pBabe–PURO plasmid while the TEL-RUNX1 construct was kindly sub-cloned into pBabe-PURO by K Wolyniec. The vector control was the empty pBabe-PURO plasmid.

Transfections, transductions and culture conditions

Transfections were carried out using Superfect Transfection Reagent (Qiagen) according to the manufacturer's protocol for transient transfection of adherent cells. Fibroblasts were plated at 8×10⁵ in 10 cm dishes and incubated overnight before transfection. Transfected fibroblasts were then incubated overnight in normal medium after which selection was applied using puromycin (Sigma) at 2μg/ml for 4 days. Control fibroblasts died under these selection conditions. Puromycin selection was continued throughout. Medium changes were carried out every 3-4 days and post-confluent cultures were maintained in 25 cm² flasks without passage.

pBabe-Runx1P1-PURO and the control plasmid were introduced into lymphocytes by retroviral transduction using Phoenix packaging cells as described previously (Wotton et al., 2004). Viral supernatants were filtered through 0.45μm filters, supplemented with 4μg/ml polybrene (Sigma) and used to infect lymphocyte cultures (1×10⁶ cells). After an initial infection overnight, a second harvest of viral supernatant was added for a further 8 hours. Cells were allowed to recover overnight in fresh medium and selection was applied using 2μg/ml puromycin for 6 days.

Live/dead cell counts were carried out using a haemocytometer and trypan blue as a vital indicator. Graphs were generated with Sigma-Plot and significance values determined by students t-test. Error bars relate to standard deviations.

Western blotting and antibodies

Preparation of whole cell protein extracts was performed as described previously (Wotton et al., 2004). Samples equivalent to 30 μg total protein (Bio-Rad protein assay) were resolved on 8%, 10%, or 17% SDS-polyacrylamide gels and transferred to enhanced chemiluminescence (ECL; Amersham) nitrocellulose membranes. The antibodies used were ERα – sc542; β-actin – (I-19) sc1616 (Santa Cruz Biotechnology); Runx 3 (pan-Runx) rabbit polyclonal antibody (in house); Runx 2/ CBFa1 clone 8G5 (MWG # D130-3).

Immunofluorescent labelling and confocal microscopy

Immunofluorescence was performed as described previously (Terry et al., 2004). Runx expressing and control fibroblasts were plated at 2×10⁴/well and grown to 80% confluence on 13.3μg/ml poly-L-lysine-coated glass chamber slides (2 wells per slide). Primary antibodies used were 1:100 α-N-cadherin – 610921 BD Transduction Laboratories and 1:500 α-Integrin beta5-ab15459 ABCAM. An appropriate fluorescein (FITC)-conjugated antibody (Jackson Laboratories), diluted 1:100 in block buffer was used as secondary antibody.

Microarray

Runx expressing and control fibroblasts were grown to confluence in duplicate wells (7 days). Cells expressing the RUNX1-ER construct were treated for a further 24 hours with 4-OHT. Cells were harvested from duplicate wells into buffer RLT and RNA prepared using the RNeasy Mini kit (Qiagen) according to the manufacturers' protocol. cDNA preparation and the microarray assay and primary analysis was performed in the Sir Henry Wellcome Functional Genomics Facility at the University of Glasgow using standard Affymetrix protocols and GlaMA analysis approach (http:www.brc.dcs.gla.ac.uk/systems/glama/) implemented locally into the FunAlyse automated pipeline. Briefly, duplicate samples were hybridised to Affymetrix GeneChip® Mouse Genome 430.2.0 arrays representing over 34,000 genes. Raw data were then normalised by the Robust Multichip Average (RMA) method (Irizarry et al., 2003) and differentially expressed genes were identified by the Rank Products Algorithm (Breitling et al., 2004), which is particularly powerful for analysis of small numbers of biological replicates (Breitling and Herzyk, 2005; Jeffery et al., 2006).

Quantitative real time PCR

cDNA was prepared from RNA isolated as above or by RNA-Bee (ams biotechnology) as previously described (Kilbey et al., 2007). 5μl aliquots of cDNA were amplified in triplicate using primers for murine endogenous control Hprt as previously described (Kilbey et al., 2007) or primers for murine Cyp1b1, Itgb5, Rgc32, Timp4, Cebpd, ErrFi1, Nr3c1, Sgpp1, (Qiagen QuantiTect Primer Assays), Ncam1 (619F 5′acaaaggccgagatgtcattct 3′and 697R 5′atgcccctgatctgcaggta 3′), Ugcg (779F 5′tttgctcagtacattgctgaagatta 3′ and 861R 5′ acttgagtagacattgaaaacctccaa 3′), Id2 (370F 5′ccaccctgaacacggacatc 3′ and 456R 5′ agagtactttgctatcattcgacataagc 3′), Rgs2 (273F 5′ ggcagaagcatttgatgaactg 3′ and 375R 5′ caaccagaattcaatgttttcttcac 3′). Relative quantification was carried out and calibrated to vector control samples.

Chromatin Immunoprecipitation assay

Chromatin immunoprecipitation (ChIP) was performed as described for the Active Motif ChIP-IT kit with modifications. Approximately 5 × 10⁷ cells were treated with 1% (v/v) formaldehyde at room temperature for 10min. Cross-linking was terminated by adding 0.125M glycine for 5min. Cells were washed in ice cold PBS and collected by centrifugation at 4°C for 10min at 2500rpm. The pellet was resuspended in lysis buffer (ChIP-IT kit) and incubated on ice for 30min. Following Dounce homogenisation (15 strokes) the nuclei were collected by centrifugation at 4°C for 10min at 5000rpm, resuspended in 2ml shearing buffer (ChIP-IT kit) and sonicated on ice to a DNA size of 300-600bp. 50μl of the precleared chromatin was immunoprecipitated with 2μg of antibody and immuncomplexes collected by magnetic binding to Dyna beads protein G (Invitrogen 100.03D). Beads were washed 8 times in 800μl wash buffer (50mM HEPES pH 7.6., 0.5M LiCl., 1mM EDTA pH 8.0., 1% NP40., 7% sodium deoxycholate) and 3 times in 1X TE. Immunocomplexes were eluted twice in 50mM NaHCO₃ and 1% (v/v) SDS at 65°C for 15 mins, adjusted to 200mM NaCl and incubated overnight at 65°C to reverse cross-linking. Following treatment with RNaseA (100μg/ml) and Proteinase K (200μg/ml) each sample was purified through a DNA purification column (ChIP-IT kit). 1/10th tenth of the immunoprecipitated DNA and 1/100th of input DNA were analysed by PCR using oligonucleotides: Bglap2(f) - 5′– agcatccagtagcatttata-3′; Bglap2(r) - 5′–cttgtctctagggcgaccca-3′; Rgs2(f) - 5′-gtggctgacgcctccaggtc-3′; Rgs2(r) - 5′–ccatgggtacgcagtcgtgc-3′; Rgc32(upper)(f) - 5′– gcaaagggctgttggtcacc-3′; Rgc32(upper)(r) - 5′–ccaagggcctcagtagcctc-3′; Rgc32(lower)(f) - 5′–cagggtggtagagaaagcgg-3′; Rgc32(lower)(r) - 5′–cctgcactgcagtctgctcc-3′; C/ebpδ(upper)(f) - 5′–gccagagcacccaagatcc-3′; C/ebpδ(upper)(r) - 5′–ggcggtccagagtgagcg-3′; C/ebpδ(lower)(f) - 5′-cgggtccgcgaaccacgg-3′; C/ebpδ(lower)(r) - 5′ –cccgcactccttgccttccc- 3′. The hotstart PCR conditions were 5min at 95°C (1 cycle), 1min at 95°C, 1min at 55°C (Bglap2 – positive control) or 60°C (Rgs 2, C/ebpd, Rgc32 – test primers), 1min at 72°C (30 cycles), 10min at 72°C (1 cycle). The PCR products were resolved on a 1.5% TBE agarose gel.

The antibodies used were: α RUNX2 and α GST (Santa Cruz Biotechnology sc-10758., sc-459).

Dexamethasone treatment

Runx1 expressing fibroblasts or lymphocytes and control cells were plated at 1×10⁶/well/2ml and dexamethasone (10μM) (D2915, Sigma, UK) was added after 24 hours. Drug sensitivity was tested by Trypan Blue exclusion assay and live/dead counts.