To establish the impact of TAF6δ on global gene expression patterns, the expression of the endogenous TAF6δ splice variant was experimentally induced using splice-switching oligonucleotides (SSO) as previously documented 5. The HeLa cell line was chosen as a model system for transcriptome studies for three principle reasons. Firstly, these cells are readily transfectable and produce a robust apoptotic response to TAF6δ-inducing SSO 5. Secondly, the TAF6δ cDNA was cloned from a HeLa cell library 4 and therefore these cells provide a natural cellular context. Thirdly, HeLa cells express no detectable TAF6δ protein under standard culture conditions 5, thus these cells provide a stringently inducible model for SSO studies. To define the impact of TAF6δ expression on transcriptome dynamics we took advantage of an experimental approach that combines SSO treatment with high sensitivity microarray analysis 26. We note that to achieve the statistically significant overrepresentation of gene ontology pathways reported here it was necessary to employ optimized SSO sequences designed to more efficiently induce TAF6δ expression than those employed in a previous study 5. The improved SSO were transfected into HeLa cells and the induction of endogenous TAF6δ mRNA and protein was confirmed, as shown in Additional File 1. Total RNA was isolated 18 hours post-transfection and subjected to microarray analysis as previously detailed 26. Biological triplicates were performed with SSO T6-1 and, as a control to normalize for any non-specific SSO effects, a scrambled oligonucleotide (SSO ctrol). The statistical analysis and filtering of the raw microarray data was carried out as previously described 26 to identify significantly (P < 0.05) regulated mRNAs.

The induction of endogenous TAF6δ resulted in significant changes in the levels of 961 probes corresponding to 955 independent genes of 27,868 (Figure 1A). Remarkably, 90.5% of the mRNAs significantly changed by TAF6δ are upregulated and only 9.5% are downregulated (Figure 1A). These data are consistent with previous results obtained with a less efficient SSO in the HCT-116 cell line 5 and further demonstrate that TAF6δ acts primarily as a positive regulator of gene expression. The data also rule out the possibility that TAF6δ-induced cell death is a result of a global reduction in mRNA transcription. To validate the TAF6δ transcriptome signature we selected 14 genes (and the internal control beta-2-microglobulin gene) for quantitative RT-PCR confirmation. Gene expression changes measured by real-time RT-PCR for the 14 genes showed a strong correlation with changes measured by microarray analysis (Pearson correlation coefficient R² = 0.769, Figure 1B & 1C). To further confirm the specificity of the TAF6δ transcriptome signature we employed a distinct TAF6 targeting antisense oligonucleotide (SSO T6-3), whose binding to the TAF6 pre-mRNA is shifted five nucleotides downstream with respect to SSO T6-1. SSO T6-3 is slightly more efficient in inducing endogenous TAF6δ than SSO T6-1 (Additional File 1). Real-time RT-PCR shows that like SSO T6-1, SSO T6-3 also altered the expression of the same 14 TAF6δ target genes (Pearson correlation coefficient with microarray measurements R² = 0.836, Figure 1B & 1C). The above data confirm that the microarray results provide an accurate and reproducible measure of the TAF6δ-controlled transcriptome landscape.

To examine the specificity of the TAF6δ-induced transcriptome signature, we compared the microarray data with those obtained when the pro-apoptotic isoform of a distinct gene, Bcl-x, was induced by SSO under identical conditions 5. As shown in Additional File 2, the transcriptome signatures resulting from the induction of TAF6δ and Bcl-xS are highly distinct. The vast majority of TAF6δ-regulated transcripts (90.5%) were induced while Bcl-xS expression results in a majority of transcripts being repressed (58%). Only a minor fraction (3.4%) of the 870 genes upregulated by TAF6δ was also upregulated by Bcl-xS. Of the small number of transcripts repressed (46) by TAF6δ, 45 are also repressed by Bcl-xS, possibly reflecting a minor subset of genes that are repressed by both of these pro-apoptotic pathways. The portion of Bcl-xS repressed genes also repressed by TAF6δ was minor (15.8%). The fact that genes induced by TAF6δ share little overlap with those induced by Bcl-xS underscores the highly specific impact of the TAF6δ-inducing SSO on the transcriptome.

To shed light on the mechanisms underlying the pro-apoptotic capacity of TAF6δ, we performed gene ontology analysis to identify pathways that are statistically overrepresented within the microarray data (see details in Materials and Methods). Of 131 cellular pathways surveyed, those that are statistically (P < 0.05) overrepresented in the TAF6δ data set are the Notch, oxidative stress response, integrin, p53, apoptosis and p53 pathway feedback loops 2 pathways (Figure 2A). Genes in the angiogenesis pathway were also overrepresented (P = 0.0535, Figure 2A) in the TAF6δ-induced gene pool. The TAF6δ-regulated genes found in the overrepresented pathways are shown in Additional File 3. We next postulated that if the pathways activated by TAF6δ represent a physiologically coherent response, then the functional connections between these pathways would not be random. We therefore performed a statistical analysis to identify pathways that share two or more genes activated by TAF6δ at frequencies higher than those expected in a random sample. Statistically significant (P < 0.05) overrepresentation of shared genes between the integrin and angiogenesis pathways, as well as between the p53 and integrin pathways was observed (Figure 2B). To provide a visual framework that depicts the interconnections between the overrepresented pathways they were mapped onto cellular signaling networks using the Pajek algorithm 27. This global view reveals that the activated pathways are not randomly distributed and moreover that there is a network of interconnections between TAF6δ-regulated pathways, as seen by black points at which distinct colored lines intersect (Figure 2C). Taken together, the results above define a specific TAF6δ-driven transcriptome landscape that includes the induction of genes in the Notch, oxidative stress response, integrin, p53, apoptosis, p53 pathway feedback loops 2 and angiogenesis pathways.

To determine whether or not the changes in mRNA expression in response to TAF6δ expression result also in changes in protein levels we selected representative proteins from the TAF6δ transcriptome signature for verification by immunoblotting experiments. The microarray data showed expression of ARNT mRNA was repressed by TAF6δ, and immunoblotting showed the corresponding ARNT protein levels also decreased in response to TAF6δ-inducing SSO (Figure 3A). The levels of the transcription factor TBP were tested as a control for specificity and remained relatively constant in response to TAF6δ (Figure 3A). The levels of FOS, JUN, HES1, CDKN2B (p15INK4B), and PMAIP1 (NOXA) were assayed and TAF6δ expression resulted in increased protein levels that paralleled increased mRNA levels detected in the microarray experiments in each case (Figure 3A). Importantly, Bcl-x SSO treatment did not result in comparable changes of protein levels, showing the specificity of their response to TAF6δ expression (Figure 3A). These data demonstrate that for the proteins tested the impact of TAF6δ on gene expression programs occurs at both the mRNA and protein levels, including the induction of the known pro-apoptotic protein PMAIP1 (NOXA) 28.

Changes in mRNA levels detected by microarray analysis can in principle result from a number of effects including alterations in mRNA stability. To obtain evidence that TAF6δ can regulate gene expression in a promoter-dependent and promoter-specific fashion, we tested the ability of endogenous TAF6δ to increase target gene expression in luciferase reporter gene assays. We selected four promoters for analysis. The HES1, DUSP1 and ADM promoters were selected since the endogenous HES1, DUSP1 and ADM mRNA levels are induced in response to TAF6δ expression (Figure 1A & 1B). In the case of HES1, the levels of endogenous HES1 protein were also shown to be induced in response to TAF6δ expression (Figure 3A). The 3 selected genes act in several of the pathways activated by TAF6δ including the Notch (HES1) 29, angiogenesis (ADM) 30, oxidative stress and p53 pathways (DUSP1) 313233. The Bax promoter was also included because it is a p53-responsive and pro-apoptotic gene 34, yet is not induced by TAF6δ. The induction of TAF6δ in HeLa cells resulted in increased HES1, DUSP1, and ADM promoter-driven gene expression (Figure 3B). In contrast, the Bax promoter was not induced and even measurably repressed (Figure 3B). These results demonstrate that endogenous TAF6δ can act directly or indirectly to stimulate transcription in a promoter-dependent manner.

The only currently known functional distinction between pro-apoptotic TAF6δ and TAF6α is that TAF6δ cannot interact with TAF9 or TAF9b. We therefore analyzed the extent to which the transcription footprints resulting from TAF6δ induction resembles those reported following the depletion of TAF9 and/or TAF9b by treatment with siRNAs in HeLa cells 16. After mapping of the previous microarray data to enable comparison with our current microarray platform (see Materials and Methods), the datasets were co-filtered to compare transcriptome changes. Of the 961 TAF6δ-dependent mRNAs selected for comparative analysis, 803 mRNAs could be mapped between the datasets (Figure 4B). A global view of the magnitude of the changes for these 803 genes showed that changes in response to TAF6δ induction were more pronounced than those resulting from TAF9/TAF9b depletion as depicted by heat maps (Figure 4A). 204 mRNAs that are statistically significantly regulated by TAF9 and/or TAF9b were found within the TAF6δ-regulated transcripts (Figure 4B). Of the 204 genes, 50 showed regulation by both TAF9 and TAF9b. 90 showed regulation by TAF9 alone and 64 showed regulation by TAF9b alone (Figure 4B). To probe for communalities between TAF6δ, TAF9 and TAF9b-controlled transcriptomes, pathway analysis was performed on the genes regulated by TAF6δ and TAF9 as well as genes regulated by TAF6δ, TAF9, and TAF9b. The only pathway that was statistically significantly overrepresented in these subsets was the "p53 feedback loop 2" ontology (Figure 4C). Pathway analysis was also performed on the TAF9b-dependent gene set alone. Interestingly, the only pathway that was overrepresented was angiogenesis, a pathway also overrepresented in the TAF6δ transcriptome signature (Figure 4D). Finally, we compared the effects of TAF6δ, TAF9, and TAF9b on single genes from the subset of 50 genes that respond to changes in the expression all of TAF6δ, TAF9, and TAF9b. A majority of genes (~80%) had comparable changes in response to TAF6δ induction and TAF9 or TAF9b depletion of which 10 examples are shown in Figure 4D. In contrast, approximately 20% of the TAF6δ- TAF9- and TAF9b-dependent genes were positively regulated by TAF6δ induction whereas TAF9 or TAF9b depletion resulted in their suppression (Additional File 4). Taken together the analysis shows that TAF6δ induction has both overlapping and unique impacts on the transcriptome, when compared to TAF9 and/or TAF9b depletion.

TAF6α is known to interact with the p53 tumor suppressor protein (see Introduction), but whether the pro-apoptotic TAF6δ isoform could retain the capacity to interact with p53 is unknown. Our previous work demonstrated that TAF6δ induces apoptosis independently of p53 5. Interestingly, the current transcriptome data show that TAF6δ induces the expression of several p53 target genes (Figure 2A and Additional File 3). We therefore investigated the capacity of TAF6δ to interact with p53. Recombinant Histidine tagged TAF6δ was produced in bacteria and purified by nickel affinity chromatography. GST tagged p53 was produced and immobilized on glutathione-sepharose beads. Purified recombinant TAF6δ and TAF6α were assayed for their capacities to interact with immobilized GST-p53. As previously reported 17, TAF6α bound efficiently to GST-p53 (Figure 5A, lane 3, upper row). Terminal deoxynucleotidyl transferase (TdT) served as a control for specificity and did not bind to GST-p53 (Figure 5A, lane 3, lower row). Purified His-TAF6δ was efficiently retained by GST-p53 (Figure 5A, lane 3, middle row), but not by GST alone (Figure 5A, lane 2). Resistance to high ionic strength buffers can provide a measure of the strength of protein-protein interactions. We therefore challenged the p53-TAF6 interactions with 600 mM KCl washes and found that the interactions were stable in high salt conditions (Figure 5B). These data show a direct and selective interaction between TAF6δ and p53 in vitro.

To determine whether the TAF6δ-p53 interaction can occur in living cells we performed co-immunoprecipitation assays. As endogenous TAF6δ is highly labile and expressed at very low levels 5, TAF6δ was expressed by transfection of an expression vector into HCT-116 p53 -/- cells. Exogenous p53 was provided by co-transfection of an expression vector. Immunoprecipitation of TAF6δ resulted in co-immunoprecipitation of p53 (Figure 5C, lane 3), in contrast to the negative control immunoprecipitation of tubulin (Figure 5C, lane 2). In addition, the reciprocal experiment, immunoprecipitation of p53 resulted in recovery of TAF6δ (Figure 5C, lane 4). Together, these data show that the interaction between TAF6δ and p53 can occur in the cellular context.

We next sought to determine whether the interaction between TAF6δ and p53 has functional consequences. We took advantage of the fact that the DUSP1 promoter is activated by TAF6δ (Figure 3B) and is also activated by p53 33. A reporter construct expressing firefly luciferase under the control of the DUSP1 promoter was co-transfected with p53 expression vectors, as well as vectors expressing TAF6 variants. TAF6δ transfection resulted in enhanced DUSP1 expression when co-transfected with p53 (Figure 5D). A truncated version of TAF6 that lacks pro-apoptotic activity 4 failed to show significant co-activation (Figure 5D). The protein levels resulting from transfected plasmids were determined by immunoblotting experiments and are shown in Additional File 5. The data exclude the possibility that higher levels of TAF6δ protein (compared to the truncated negative control TAF6) contribute to the activation levels observed. These data show for the first time that TAF6δ can interact functionally with p53 to co-activate DUSP1 gene transcription.

Having established that TAF6δ can interact functionally with p53, we next sought to define the role of this interaction with endogenous TAF6δ and p53. Moreover we sought to define the potential crosstalk of TAF6δ with p53 upon endogenous genes and at the transcriptome-wide level. We therefore revisited microarray data derived from either wild-type HCT-116 or their p53 negative counterpart HCT-116 p53 -/- in which endogenous TAF6δ expression was induced by SSO (http://www.ncbi.nlm.nih.gov/geo/ under accession number GSE10795) to test for transcriptional crosstalk between TAF6δ and p53. We re-filtered the data specifically to determine the influence of TAF6δ versus TAF6α on p53-dependent genes. The effects of TAF6 isoforms are illustrated in Figure 6A, and reveal three classes of genes. Importantly, 20% of p53-regulated genes (e.g. CAMK2B & THEDC1) change only in the presence of TAF6δ (Figure 6B). 53% of the p53-regulated genes change specifically in the presence of TAF6α, for example TNFRSF10C and CHIA (Figure 6C). 25% of p53-regulated genes, such as FAS and ANGPT2, change expression in the presence of both TAF6δ and TAF6α (Figure 6D). The data show that the expression of TAF6δ versus TAF6α can dictate the outcome of p53-mediated transcriptional signals.

We also filtered the data to determine the reciprocal influence of p53 status upon previously identified TAF6δ-dependent mRNAs 5. 51% of the TAF6δ-regulated mRNAs changed significantly only in the absence of p53 (Figure 6E), for example TNFRSF6B (Figure 6F). 9% of the TAF6δ-regulated mRNAs changed independently of p53 status, including ACRC (Figure 6F). 40% of TAF6δ-regulated mRNAs changed significantly only in cells expressing p53, such as TP53I3 (Figure 6F). In general the influence of p53 on TAF6δ-dependent transcription was relatively subtle in magnitude, with at least one exception where the gene LOC342293 displayed opposing regulation in presence or absence of p53 (Figure 6F). Together, the above data establish reciprocal transcriptional crosstalk between the TAF6δ and p53 proteins.

HeLa cells were grown in DMEM containing 2.5% CS and 2.5% FCS. HCT-116 cells were grown in McCoy's media supplemented with 10% FCS.

2'-O-methyl-oligoribonucleoside phosphorothioate antisense 20-mers were from Sigma-Proligo. "SSO ctrol", "SSO T6-1" 5 and "SSO Bcl-x" 43 have been described. "SSO T6-3" 5'-CUGUGCGAUCUCUUUGAUGC-3' targets the 3' part of the alternative exon 2 of TAF6. SSOs were transfected at a final concentration of 200 nM with lipofectamine 2000 (Invitrogen) as a delivery agent (1.6 μl/ml) according to the manufacturer's recommendations. Plasmids were transfected using 1 μl DMRIE-C (Invitrogen) as a delivery agent in a 24 well plate according to the manufacturer's recommendations. All transfections were performed in OptiMEM medium (Invitrogen).

Plasmids expressing firefly luciferase under control of the HES1 29, DUSP1 33, Bax 44, and ADM 45 promoters have been described. Plasmids expressing TAF6α, TAF6δ and TAF6ΔAB 4 and p53 or its mutated form p53R175H 46 have been described. To construct vectors for bacterial production of His-tagged TAF6α and TAF6δ proteins, full length cDNAs were excised from pXJ42-TAF_II80α and pXJ42-TAF_II80δ(ΔA) 4 respectively, using NotI and XhoI sites. The NotI site was filled in by treatment with Klenow enzyme. The generated fragment was inserted into the SalI and Klenow filled HindIII sites of pQE31 vector (Qiagen), generating pQE31-TAF6α and pQE31-TAF6δ plasmids. pGST-p53Arg 47 and pGEX4-T-3 (GE Healthcare) were used for expression of GST-tagged p53 and GST proteins respectively.

Monoclonal antibodies directed against TAF6δ (37TA-1 & 37TA-2) 4, and TBP (3G3) 48 have been described. The pan-TAF6 monoclonal antibody was purchased from BD Transduction Laboratories. Antibodies against ARNT (sc-17811), JUN (sc-1694), FOS (sc-52), CDKN2B (sc-613) and His probe antibody (sc-803) were purchased from Santa Cruz Biotechnology. Antibodies against HES1 (AB5702), PMAIP1 (Ab13654), alpha-Tubulin (clone B-4-1-2), and p53 (clone PAb1801) were purchased from Millipore, Abcam, Sigma, and Calbiochem respectively.

RT-PCR conditions and primers for amplification of both TAF6α and TAF6δ have been described 5.

Immunolabelling of TAF6δ in fixed cells was performed as described 5.

Transcriptome analysis was performed as we previously detailed 26, using the NeONORM normalization method with k = 0.20 49. The published microarray data for TAF9 and TAF9b depletion by siRNA were generated on the Génopole Genomics Platform Strasbourg using custom technology. In order to be able to compare those data directly to data generated from commercial platforms, the unique probe-set identifiers were mapped to non-redundant NCBI and Ensemble gene IDs. Similarly, the AB1700 data generated for this study or from previous studies on an Applied Biosystems Microarray platform were mapped according to the published procedure 50 to the same set of gene IDs. After these mapping procedures >87.9% of unique probe-set or probe IDs could be directly compared which corresponds to > 93.2% comparable genes. For comparative pathway inference analyses the TAF9 and TAF9b data were mapped to, and treated as if AB1700 data to avoid any potential bias stemming from the use of different ontology annotation databases.

Gene Ontology (GO) and KEGG annotations were analyzed using the Panther Protein Classification System http://www.pantherdb.org to identify functional annotations that were significantly enriched in the different gene sets when compared to the whole set of genes present on the ABI microarray. Note that a given gene can be assigned to different pathways; in order to reduce multiple probing biases a gene is weighted by the inverse of the number of pathways it can be assigned to, leading to non-natural numbers for the gene counts. P-values are determined using a binominal distribution and a null hypothesis of a random set of genes with identical size. Pathway interconnectivity analysis was performed for the significantly overrepresented pathways based on genes that are annotated to be part of any combination of two of the selected pathways and that were significantly regulated in the subtraction profile analysis. Those numbers were then compared to the entire set of shared genes, and P-values were calculated as above.

Microarray data for the gene sets analyzed herein are provided as Additional Files 3; 6, 7, 8, 9, 10. The transcriptome-wide microarray data for all of the experiments described here were deposited in the M. ACE database http://mace.ihes.fr under accession numbers:

TAF9/TAF9b: 2833146766

TAF6δ signature: 2937831950; Bcl-x: 2156101006; p53: 2370552334

Real time PCR was performed as described 5. RNA was prepared with using an RNeasy mini Kit (Qiagen). 1 μg of total RNA was reverse transcribed using AMV-RT (Roche). Real-time PCR was performed in a final 25 μl reaction on 10 ng of cDNA with 12.5 μl 2× TaqMan^® Universal Master Mix (ABI) and 1.25 μl of the following 20× TaqMan^® probes: B2M as the internal control (Hs99999907_m1), ACRC (Hs00369516_m1), ADM (Hs00181605_m1), ATF3 (Hs00231069_m1), DDIT3 (Hs00358796_g1), DUSP1 (Hs00610256_g1), EFNA5 (Hs00157342_m1), HES1 (Hs00172878_m1), HOM-TES-103 (Hs00209961_m1), IFRD1 (Hs00155477_m1), IL6 (Hs00174131_m1), NR4A2 (Hs00428691_m1), PFKFβ4 (Hs00190096_m1), PMAIP1 (Hs00560402_m1), or TRIB3 (Hs00221754_m1).

Cells were washed with PBS and lysed with Passive lysis buffer (Promega). The luciferase activity was measured on a Lumistar luminometer (BMG Labtech), after injection of 2× Luciferin reagent; 270 μM CoenzymeA, 470 μM D-Luciferin, 530 μM ATP (all from Sigma-Aldrich) 40 mM Tris-Phosphate pH 7.8, 2.14 mM MgCl₂, 5.4 mM MgSO₄, 0.2 mM EDTA, 33.3 mM DTT.

Escherichia coli M15 cells were transformed with plasmids expressing His-TAF6α and His-TAF6δ fusion proteins and grown to log-phase before induction of protein expression with 1 mM IPTG (isopropyl-β-d-thiogalactopyranoside) for 18 h at 16.5°C. The cell pellets were resuspended in Ni buffer composed of 1 M NaCl, 30 mM Tris pH8.0 and supplemented with 25% glycerol and 1× Complete protease inhibitor cocktail (PIC) (Roche) and 0.5 mM PMSF (phenylmethyl-sulfonyl fluoride) and sonicated twice for 5 minutes on ice. After clarification, the supernatant was loaded on a His-trap column (GE Healthcare) equilibrated in Ni buffer containing 10 mM imidazole. The column was sequentially washed with Ni buffer containing 60 mM and 100 mM of imidazole and the proteins were finally eluted with 250 mM imidazole. Purified proteins were dialysed against buffer D (20 mM HEPES ph7.9, 100 mM KCl, 20% glycerol, 0.2 mM EDTA).

GST and GST-p53 fusion proteins were produced in Escherichia coli BL-21 cells by IPTG induction (1 mM) of a log-phase culture for 3 h. The cell pellets were resuspended in PBS containing 0.5% Triton X-100 and sonicated twice for 2 minutes. After clarification, the supernatant was incubated with glutathione-sepharose (GE Healthcare) for 1 h 30 at 4°C. The beads were washed three times in PBS and once in buffer D supplemented with 5 mM MgCl₂, 0.1% NP40 and EDTA up to 1 mM (buffer D+).

GST "pull-down" assays were performed essentially as previously described 17. GST- and His-tagged proteins were pretreated with 0.5 U DNase (Promega) for 10 minutes at 37°C before the interaction assay. Equal amounts of GST or GST-p53 linked to sepharose beads were then incubated with His-TAF6α, His-TAF6δ or His-TdT for 1 hour at room temperature in buffer D supplemented with 0.5 μg RNase A (USB) (buffer D+). After four washes with buffer D+ containing 100 mM or 600 mM KCl, bound proteins were analyzed by SDS-PAGE and immunoblotting.

Cells were lysed in RIPA buffer (50 mM Tris pH8, 1% NP40, 0.25% Na deoxycholate, 150 mM KCl, 1 mM EDTA) supplemented with 1× PIC and 0.5 mM PMSF. The lysate was diluted 1/10 with IP100 buffer (25 mM Tris pH8, 5 mM MgCl2, 10% glycerol, 100 mM KCl, 0.1% NP40, 0.3 mM DTT, PIC, PMSF) and precleared with proteinG-sepharose beads for 2 hours at 4°C. The precleared lysate was then incubated overnight at 4°C with anti-p53, anti-TAF6δ or anti-tubulin antibodies immobilized on protein G-sepharose beads. After extensive washes with IP100 buffer, complexes were analyzed by SDS-PAGE and immunoblotting.