Global expression changes were analyzed in actively proliferating T98G human glioblastoma cells, which are widely used as a model for studies of growth factor regulation of human cells 263233. Similar to previous studies in rat fibroblasts 31, these cells maintained a low level of PI 3-kinase signaling, which was 7 fold higher than in quiescent cells but approximately 5-fold lower than that observed following mitogenic stimulation of quiescent cells with either platelet-derived growth factor (PDGF) or 20% serum (Fig. 1A). PI 3-kinase was rapidly inhibited in proliferating cells by addition of LY294002 34, which inhibited phosphorylation of Akt within 15 minutes (Fig. 1B). Inhibition of PI 3-kinase led to the induction of apoptosis, which initiated as early as 3 hours after addition of LY294002 as indicated by DNA fragmentation (Fig. 1C). The fraction of cells undergoing apoptosis gradually increased over several hours (Fig. 1D), typically reaching levels of 25–50% after 24 hours of PI 3-kinase inhibition.

Microarray analyses were performed in three independent experiments to characterize the changes in gene expression resulting from PI 3-kinase inhibition. The number of genes with significantly altered expression (Log₂ > 0.9; p < 0.01) after 2, 4 and 8 hours of treatment with LY294002 are shown in Fig. 2. Inhibition of PI 3-kinase for 2 hours resulted in the up-regulation of 20 genes and down-regulation of 26 genes. After 4 hours of PI 3-kinase inhibition, the number of up-regulated genes slightly increased to a total of 23, whereas the number of down-regulated genes nearly doubled, to a total of 44. The increase in the number of down-regulated genes was most likely due to the slower rates of mRNA degradation for some genes, thus requiring a longer period for significant decreases in mRNA levels to occur. After 8 hours of PI 3-kinase inhibition, the numbers of both up-regulated and down-regulated genes increased substantially to 118 and 121, respectively, possibly resulting from secondary changes in gene expression.

The genes that were up- or down-regulated following 2 and 4 hours of PI 3-kinase inhibition are summarized in Table 1 (see Additional file 1 for expression changes and for genes with altered expression at 8 hours). Most of the up-regulated genes had increased levels of mRNA after both 2 and 4 hours of inhibition, however 4 genes were significantly up-regulated at 2 hours but returned to baseline at 4 hours, perhaps reflecting transient gene induction. Most down-regulated genes displayed reduced mRNA levels at both 2 and 4 hours, generally decreasing at the longer time of PI 3-kinase inhibition. The microarray data for 7 up-regulated and 11 down-regulated genes were validated by real-time RT-PCR. For all genes tested, the expression changes assessed by real-time RT-PCR after 4 hours of PI 3-kinase inhibition confirmed the microarray results (Additional file 1).

Since LY294002 can also inhibit PI 3-kinase-related protein kinases, such as mTOR, we also tested the effect of wortmannin, a more selective PI 3-kinase inhibitor that does not inhibit mTOR at the concentration used (50 nM) 35. 15 of the 18 genes were similarly up or down-regulated by both wortmannin and LY294002 (Additional file 2), indicating that PI 3-kinase was responsible for the changes in expression of these genes. It is possible that the 3 genes whose expression was affected by LY294002 but not wortmannin may be regulated by other targets of LY294002. However 50 nM wortmannin was less effective than LY294002 in inhibiting PI 3-kinase, yielding approximately 90% rather than complete inhibition of Akt phosphorylation (Additional file 2), so it is also possible that the 10% of PI 3-kinase signaling remaining after wortmannin treatment was sufficient to maintain normal expression of some genes. In either case, these results support the involvement of PI 3-kinase signaling in the regulation of at least 80–90% of the genes showing altered expression in response to LY294002.

To determine whether gene expression was similarly regulated in a different cell line, the response of representative genes to inhibition of PI 3-kinase was tested in U937 human promyelocytic leukemia cells (Table 2). The changes in expression of 7 out of 12 genes in response to PI 3-kinase inhibition were similar in both U937 and T98G cells, and 3 other genes were either up- or down-regulated, although to different extents, in both cell lines.

The genes that showed changes in expression after 2 and 4 hours of PI 3-kinase inhibition were largely overlapping and therefore were combined for subsequent analyses. This yielded a gene set in which altered expression was detected around the onset of apoptosis. These genes were compared to a group of 34 immediate-early genes that were induced by PI 3-kinase signaling following PDGF stimulation of quiescent T98G cells 26. There were only 4 genes in common between the two groups (CTGF, PLAU, CCL2 and F3), indicating that >90% of the genes regulated by continual PI 3-kinase signaling were distinct from immediate-early genes induced by PI 3-kinase in response to acute growth factor stimulation.

The genes with altered expression in response to inhibition of PI 3-kinase in T98G cells for 2 and 4 hours included well-known regulators of apoptosis and cell cycle progression. The ability of PI 3-kinase inhibition to induce apoptosis was apparent in the up-regulation of genes that can promote cell death, including DDIT3/CHOP 36, GADD45B 37, and PHLDA1 38, and the down-regulation of genes that promote cell survival, including BIRC3/cIAP2 39 and TNFRSF11B/osteoprotegerin 40. Additional expression changes included the up-regulation of genes that inhibit cell cycle progression, such as CCNG2 41, BTG1 42 and GADD45B 43, as well as the down-regulation of CCND1. Genes encoding several growth factors and cytokines that promote cell proliferation and survival (including FGF2, NTF3, EDN1, CCL2 and BDNF) were also down-regulated following inhibition of PI 3-kinase.

The Gene Ontology (GO) database was used to further characterize the functions of the genes regulated by PI 3-kinase. The frequencies of the GO terms assigned to the PI 3-kinase-regulated set of genes were compared to the frequencies of all annotated genes on the microarray. Biological Process GO terms that were significantly enriched (p < 0.001) and identified in at least 15% of the GO-annotated genes that were either up- or down-regulated following 2 and 4 hours of inhibition of PI 3-kinase are indicated in Table 3. The larger set of genes with altered expression after 8 hours of PI 3-kinase inhibition did not display enrichment of GO terms at this level of statistical significance. The up-regulated genes were enriched in GO categories related to development, differentiation, and response to stress, as well as in GO terms related to transcriptional regulation. These GO categories, particularly "response to stress", include such genes as DDIT3/CHOP 36 and GADD45B 37, and are consistent with the increased expression of genes involved in apoptosis and cell cycle arrest following inhibition of PI 3-kinase. The down-regulated genes were enriched in GO terms related to cell signaling and programmed cell death, consistent with the role of PI 3-kinase signaling in preventing apoptosis. These included genes encoding growth factors and cytokines as well as central negative regulators of apoptosis, such as BIRC3/cIAP2 39 and the TRAIL decoy receptor, TNFRSF11B/osteoprotogerin 40.

Co-expressed genes are often regulated by common transcription factors, therefore we analyzed the upstream regions of different sets of PI 3-kinase-regulated genes to identify over-represented transcription factor binding sites 263044. Predictions of functional transcription factor binding sites can be improved by incorporating phylogenetic data to identify sites that are evolutionarily conserved 45, so we focused on sites that were conserved in orthologous genomic regions of human and mouse. We independently analyzed the sets of genes that were either up- or down-regulated following PI 3-kinase inhibition, and also separately analyzed the genes that showed altered expression after 2 and 4 hours of PI 3-kinase inhibition and those that showed altered expression after 8 hours of PI 3-kinase inhibition. Regions 3 kb upstream of the transcription start sites in the human and corresponding mouse sequences were analyzed with the Match program using the minSUM threshold, and both the 546 vertebrate matrices in TRANSFAC Professional v8.4 and the 588 matrices in TRANSFAC Professional v11.1. Over-representation of transcription factor binding sites was assessed by comparing the frequency of predicted sites in each set of genes to the frequencies in a background set of genes that were expressed in proliferating T98G cells but not affected by PI 3-kinase inhibition (see Additional file 3 for complete results).

No significantly over-represented binding sites were identified in the sets of genes that were either up- or down-regulated following 8 hours of PI 3-kinase inhibition, perhaps because these large sets of genes are regulated by a diverse group of transcription factors or by other mechanisms, such as mRNA degradation. However, over-represented binding site matrices were identified in the sets of genes that were either up- or down-regulated after 2 and 4 hours of PI 3-kinase inhibition.

When the results from TRANSFAC v8.4 and v11.1 were combined, there were 9 matrices that were significantly over-represented in the up-regulated genes (Table 4). These included matrices for the forkhead family of transcription factors, as well as for Myc/Max and NF-Y. The most significant matrices represented forkhead family proteins, which together predicted binding sites in 82% of the up-regulated genes. The down-regulated genes were enriched in 4 matrices, 2 for NFκB and 2 for HMGI(Y). When the results from TRANSFAC v8.4 and v11.1 were combined, 27% of the genes had at least one NFκB binding site. The clear enrichment of forkhead and NFκB binding sites in the up- and down-regulated genes, respectively, as well as the fact that members of both of these families of transcription factors are known targets of Akt, led us to further investigate the roles of forkhead and NFκB in the transcriptional response of genes regulated by inhibition of PI 3-kinase.

The FOXO transcription factors are inhibited by Akt phosphorylation of Thr-24, Ser-256, and Ser-319 on FOXO1 and the equivalent sites on FOXO3 and FOXO4 16. Phosphorylation of these residues results in the binding of 14-3-3 proteins, which inhibits nuclear shuttling and transcriptional activity of the FOXOs. We investigated the expression of FOXO transcription factors in T98G cells and their response to PI 3-kinase inhibition by subcellular fractionation and immunoblotting (Fig. 3). FOXO1 and FOXO3a were expressed in T98G cells and were principally localized to the cytoplasm in cells maintained in serum (FOXO4 was not detectable, data not shown). Addition of LY294002 resulted in the nuclear accumulation of both FOXO1 and FOXO3a within 30 minutes, consistent with their involvement in the transcriptional response of T98G cells to PI 3-kinase inhibition.

Chromatin immunoprecipitation (ChIP) assays were therefore used to test the predicted FOXO binding sites in genes that were induced upon PI 3-kinase inhibition. Exponentially growing T98G cells were transiently transfected with a plasmid expressing a Flag-tagged FOXO3a-AAA mutant 46 and immunoprecipitations of sheared chromatin were performed with Flag-antibody. The Flag-FOXO3a-AAA is a constitutively active mutant in which the three Akt phosphorylation sites have been mutated to alanines, resulting in its constitutive nuclear localization. ChIP assays were performed on three independent transfected cell cultures to test predicted FOXO binding sites upstream of 21 up-regulated genes (two open reading frames with predicted FOXO binding sites were not tested, C1orf63 and C1orf183, due to their hypothetical protein designation). Flag-FOXO3a-AAA specifically bound to upstream regions of 10 of these 21 genes, compared to cells transfected with empty vector or to the negative control gene β-globin (Fig. 4). These verified FOXO binding sites are indicated in Table 5 (see Additional file 4 for all sites tested). These 10 genes included 4 with previously established FOXO upstream binding sites [CCNG2 4647, BTG1 48, BCL-6 4950 and ATROGIN-1 51 as well as 6 genes not previously identified as having binding sites for FOXO (ZBTB1, KIF1B, KLF6, KLHL24, DDIT3/CHOP and TXNIP).

In contrast to the forkhead family, Akt indirectly activates NFκB transcription factors, at least in part by phosphorylation of IκB kinase (IKK) 1920. There are 5 different NFκB family members (p65, c-Rel, RelB, p50 and p52), which all bind to a similar consensus sequence and can be activated by several pathways that converge on IKK 52. Because the majority of NFκB binding sites are occupied after stimulation with cytokines 53, we initially performed ChIP assays following treatment of T98G cells with tumor necrosis factor α (TNFα) to determine if p65 was able to bind to the sites predicted in the genes that were down-regulated upon PI 3-kinase inhibition. In addition to the sites predicted by the matrices indicated in Table 4, we tested conserved sites predicted by 7 additional NFκB matrices to ensure that most functional binding sites were identified. When all matrices were considered, 25 NFκB sites were predicted in a total of 20 of the down-regulated genes (See Additional file 4 for all sites).

TNFα activates NFκB through the canonical pathway, in which IKKβ phosphorylates IκB, which predominantly serves to activate the p50–p65 NFκB heterodimer. Cells treated with TNFα for 0.25, 0.5 or 1 hour were therefore analyzed by ChIP using anti-p65 antibody. Of the 25 predicted sites, 22 sites in 18 genes exhibited increased p65 binding in response to TNFα, which was significantly greater than binding to the β-globin negative control (Fig. 5). The binding sites (Table 6) included several that were previously known [BIRC3/cIAP2 54, CCL2 55, CCND1 56, EDN1 site I 57, F3 site II 58, GABPB2 53, IL7R site II 59, MAT2A site II 60, PLAU 61 and PTX3 62, in addition to novel binding sites in genes known to be targeted by NFκB [BDNF 63, EDN1 site II, F3 site I, IL7R site I and MAT2A site I], as well as binding sites in novel NFκB-regulated genes (AMIGO2, FST, G0S2, IGFBP5, RGS20, STC2 and TNFRSF11B/osteoprotegerin). Note that some of these verified NFκB binding sites were predicted with either Transfac v8.4 or v11.1, but not both.