Etoposide is known to induce apoptosis through DNA damage induced p53 activation 17. HepG2 cells incubated in the presence of 50 μM etoposide during 16 hours did indeed undergo apoptosis as shown by an increase in caspase activity, in active caspase 3 abundance, in PARP cleavage and in DNA fragmentation (Fig. 1). Hypoxia alone did not induce apoptosis since no increase in any of these parameters was observed after 16 hours incubation. On the other hand, hypoxia inhibited the etoposide-induced apoptosis: a marked decrease in caspase activity and DNA fragmentation was observed in addition to a nearly complete inhibition of PARP cleavage and caspase 3 activation (Fig. 1). Cell death was also investigated after a longer period of incubation in order to investigate whether the effect of hypoxia was sustained. Cells were incubated in the presence of etoposide under normoxia or hypoxia for 40 hours and LDH release was assessed. An increase in cell mortality was observed in the presence of etoposide which was significantly inhibited by hypoxia (Fig. 1E). Clonogenic assays also revealed that the hypoxia-induced protection from apoptosis was translated into actual survival of the cells (Fig. 1F).

The apoptosis induced by etoposide is controlled by numerous pro- and anti-apoptotic genes. We hypothesized that hypoxia could induce gene expression alterations that would then inhibit the etoposide-induced apoptosis. We thus compared differences and similarities in gene expression modifications in both conditions i.e. etoposide under normoxic and hypoxic conditions, using a low-density DNA microarray that comprises 123 capture probes allowing gene expression analysis for a set of key genes related to apoptosis (DualChip^® human apoptosis, Eppendorf). HepG2 cells were incubated during 16 hours in the presence or in the absence of etoposide at 50 μM under normoxia or hypoxia. Three independent experiments were performed for each condition and each sample was hybridized to three submicroarrays represented by the probes spotted in triplicate.

Gene expression data are presented in Table 1. The first list identified 35 genes found to be significantly up-regulated for at least one experimental condition. Some of the genes upregulated by etoposide, are known p53 target genes (GADD45, BAX, MDM2) while some of the genes upregulated by hypoxia alone are known HIF-1 target genes (BNIP3, ALDOA). The increased expression of the genes induced by hypoxia did not seem to be affected by the presence of etoposide. These results are coherent with the fact that etoposide did not influence HIF-1 activity under hypoxia (see below). The upregulation of gene expression induced by etoposide was (BAX, MDM2, BCL2A1) or was not (TNFRSF6) affected by hypoxia, indicating that hypoxia may influence some but not all the etoposide-activated pathways. There are also genes upregulated both by etoposide and by hypoxia alone (MCL1, BIRC3, IGFBP4) and the combined effect of both conditions may be additive for some but not all of them (CLK1, MCL1). There are also 23 genes that were not really affected by etoposide but were down-regulated by hypoxia (second list). The presence of etoposide did not influence the hypoxia-induced down-regulation of these genes. Finally, the third list shows 7 genes that were upregulated by etoposide, down-regulated by hypoxia (except AKT2 and E2F5) and displayed a different behavior when both etoposide and hypoxia were applied. BAK1 is interesting since it is a pro-apoptotic gene belonging to the Bcl-2 family which is up-regulated when apoptosis is induced (in the presence of etoposide) and down-regulated under hypoxia, when apoptosis was decreased.

The fact that our data are in good agreement with previous studies reporting modifications in HIF-1 target gene expression during hypoxia as well as in p53 target genes in the presence of etoposide already validates the DNA microarray used in this study. In order to further validate our data, we also performed SYBR Green quantitative real time PCR assays for some selected genes. Values obtained for a set of 5 genes that were up-regulated or down-regulated in response to at least one condition were confirmed (Table 2). For these genes, we found a very good correlation between relative transcript abundance data obtained by DNA microarray and by real time PCR.

The transcription of some of the etoposide-induced genes is regulated by p53 but there are also changes in gene expression observed in p53-negative cells 18. Moreover, other transcription factors than HIF-1 are activated under hypoxic conditions 19. We also observed changes in the expression of genes for which it is not yet known which transcription factor is involved. We thus investigated the activity of 10 transcription factors using several approaches to follow their nuclear abundance, their DNA binding activity and their transcriptional activity using reporter systems.

Immunofluorescence staining and confocal microscopy studies were performed to determine the abundance and the subcellular localization of the most relevant subunits of these transcription factors. The results show that HIF-1α was only present under hypoxic conditions. It was located exclusively in the nuclei and etoposide did not interfere with HIF-1α stabilization under hypoxia [additional file 1]. The expression of p53 was similarly investigated: p53 is present in the cytosol of unstimulated cells, etoposide markedly increased its abundance and induced its translocation into the nucleus. On the other hand, hypoxia decreased its abundance and in these conditions, etoposide also increased its abundance and induced its translocation into the nucleus, but to a lesser extent than in normoxia [additional file 1]. Upon DNA damage, p53 is activated by phosphorylation, serine 15 being the most important residue being modified 20. Immunofluorescence staining for the serine 15 phosphorylated form of p53 showed that this form of p53 was not present in unstimulated cells, was slightly induced in the nucleus of hypoxic cells and highly increased in the nucleus of cells incubated in the presence of etoposide under both normoxia and hypoxia [additional file 1]. These results indicate that the level of phosphorylation of p53 on serine 15 did not follow the abundance of the protein: indeed this level is enhanced under hypoxia while the protein abundance was decreased. A third transcription factor involved in transcriptional response to hypoxia 21 as well as to DNA damage 22 is AP-1. One of the most frequent subunit is c-jun and c-jun needs to be phosphorylated in order to be active. Hypoxia and to a larger extent etoposide, both increased c-jun protein abundance. There was no additive effect of both conditions and the protein was located in the nucleus [additional file 1]. On the other hand, etoposide induced c-jun phosphorylation on serine 63 while hypoxia had no effect and did not influence the effect of etoposide. As for p53, there was no correlation between c-jun protein abundance and its level of phosphorylation. Hypoxia also increased the abundance of MEF-2 while it decreased phospho-CREB and c-myc abundance and had no effect on phospho-elk1, phospho-ATF2, STAT-1. It had also different effects on the abundance of the different subunits constituting the NF-kB factor [additional file 1] (Table 3).

Transcription factor activity is regulated not only at the level of their expression but also by post-translational modifications as well as by protein-protein interactions. In a first approach to study the activity of the transcription factors of interest, we measured their DNA binding activity in vitro. To investigate the DNA binding activity of the transcription factors that are differentially regulated in these conditions, a multiplex assay has been used. This microarray comprises double-stranded DNA molecules fixed in triplicate on the support. Two types of DNA molecules are spotted: (i) "wild-type" DNA carrying the specific recognition sequence for the respective transcription factor and (ii) "mutated" DNA carrying point mutations that will not be recognized by the factor. Probes allowing the assay of the DNA binding activity for 8 transcription factors are present on the array (TFChip MAPK kit, Eppendorf). HepG2 cells were incubated 16 hours in the presence or in the absence of etoposide at 50 μM under normoxia or hypoxia. Nuclear extracts were recovered and hybridized on the microarray. Detection is performed by a cocktail of nine primary antibodies, one for each activated transcription factor and one recognizing the positive detection control, and labeled secondary antibodies. Three independent experiments were performed for each condition.

An example of such an experiment is presented in Fig. 2A. p53 was the most abundant of the 8 transcription factors. The three spots corresponding to the binding of p53 to its wild-type DNA probe are clearly visible on each array, as well as the difference in fluorescence intensity according to the experimental conditions. Figure 2B shows the quantitative data. There was no or very low DNA binding for c-myc, STAT-1 and NFATc1. DNA binding activity of elk-1 and MEF-2 decreased under hypoxic conditions while etoposide had no effect. Etoposide seemed to decrease ATF-2 DNA binding activity. Interesting data were obtained for AP-1 and p53: hypoxia increased AP-1 DNA binding activity, which was further enhanced in the presence of etoposide while etoposide had no effect under normoxic conditions. On the other hand, etoposide markedly increased p53 DNA binding activity while hypoxia inhibited it. An intermediate value was obtained in the presence of etoposide under hypoxic conditions.

In order to confirm these data, individual DNA binding assays (TransAM assay) were performed for some of these factors and for others which were not detectable with the array [additional file 2]. Similar profiles of DNA binding activity than the ones obtained with the array were observed for c-myc, elk-1 as well as for p53 and AP-1, using an antibody directed against c-jun. CREB DNA binding activity was not changed by any treatment. NF-kB DNA binding activity was diminished by hypoxia but enhanced by etoposide while HIF-1 activity was greatly enhanced by hypoxia with no influence of etoposide.

Finally, actual transcriptional activity was measured using a specific reporter system for 6 transcription factors (Fig. 3): transcriptional activity of p53, AP-1, HIF-1 and c-myc was parallel to their DNA binding activity assayed by the array and/or the TransAM assay. CREB and ATF-2 recognize the same DNA sequence and the results obtained with a reporter system containing this sequence shows an increase in luciferase expression under hypoxic conditions. Since there was no change in phospho-CREB and phospho-ATF-2 DNA binding activity nor in their nuclear abundance, this increase was probably due to the activation of another member of this family yet to be identified. Discrepancies were also observed for NF-kB since its transcriptional activity did not fit the DNA binding activity of p65. Here also another subunit may be involved. All these results indicate that hypoxia and etoposide have profound and different effects, alone and in combination, on the activity of numerous transcription factors. Table 3 summarizes all these data.

Gene expression microarrays generate a large sum of data. In addition, we also assessed the activity of 10 transcription factors in the four different experimental conditions studied here. In order to investigate which transcription factor would be responsible for the changes in expression observed for the different genes, cluster analysis was performed. The aim was to search for similar variation patterns. The expression values of up- and down-regulated genes as well as the activity profiles of the transcription factors of interest, measured with the DNA binding array and using a reporter system, were subjected to K-means clustering generating 10 clusters. These 10 clusters are illustrated in Fig. 4A and 4B. DNA binding activity and transcriptional activity were put in the same cluster for p53 but not for AP-1 and c-myc despite very similar profiles. This is due to the transformation of the actual values in log generating negative values if the original one is lower than the corresponding normoxic control. ATF-2 DNA binding activity was not either clustered with ATF-2/CREB transcriptional activity, probably because, as mentioned earlier, another subunit of this family was involved to increase luciferase activity under hypoxia. An interesting result was obtained for cluster 7: the expression profile of five genes were clustered with the activity of HIF-1. Three of these genes, MCL1, BNIP3 and JUN, are known to be HIF-1-target genes (respectively 231324), thus validating this bioinformatics approach. Similarly, cluster 8 correlated the activity of p53 with the expression of several of its target genes: CDKN1A, GPX1 and PCNA, as well as its own encoding gene (TP53). It has to be mentioned that the expression profile of other p53 target genes were clustered in another cluster (cluster 9): BAX, MDM2 and GADD45A. Cluster 8 contains genes whose expression was significantly diminished under hypoxia while the expression of those of cluster 9 was either unchanged or increased in these conditions. NF-kB activity was also clustering in the latter, together with IAP encoding genes (BIRC2 for cIPA-1 and BIRC3 for cIAP-2), whose expression is known to be regulated by this transcription factor 2526.

The ultimate goal of this study is to understand how hypoxia could decrease etoposide-induced apoptosis. In order to have a first insight in these hypoxia-induced mechanisms, we performed a second cluster analysis incorporating "phenotype" values from figure 1 (caspase activity and DNA fragmentation) expressed in log values of the ratio normalized to the corresponding normoxic control (Fig. 4C). Fig. 4D shows a summary of the 10 clusters. Clusters 8 and 9 contain p53 activity data, expression profile of most p53 target genes as well as data for caspase 3 activity. On the other hand, data for DNA fragmentation clustered with the expression profile of GADD45A, MDM2 and TNFRSF6 (Fas), three p53 target genes. There was no correlation of these phenotypic profiles with AP-1 or HIF-1 activity.

HIF-1 and p53 interact in many ways to finally dictate cell fate. We investigated whether HIF-1 was involved in the hypoxia-induced protection against the etoposide-induced apoptosis. For that, HIF-1α expression was silenced through RNA interference. Fig. 5A and 5B show that 50 nM HIF-1α siRNA inhibited HIF-1α protein expression and decreased mRNA level by more than 95 %. This was translated into an inhibition of HIF-1 target gene overexpression under hypoxic conditions as shown for EPOmRNA level (Fig. 5B). HIF-1α siRNA completely prevented the hypoxia-induced protection against the etoposide-induced apoptosis as measured by caspase 3 activity (Fig. 5C) or PARP cleavage (Fig. 5D): the protection was completely lost when HIF-1α expression was inhibited. The non-targeting siRNA did not influence apoptosis. It has to be noted that HIF-1α siRNA also increased cell death under hypoxia alone (without etoposide), which is understandable knowing that HIF-1α is the main regulator of cell adaptation to hypoxia. All together, these results indicate that HIF-1α is necessary for the hypoxia-induced resistance to apoptosis.

To gain further insight into modulation of apoptosis pathways by HIF-1α, we next assessed the effect of knockdown of HIF-1α on mRNA levels of pro- and anti-apoptotic proteins, using the low density DNA microarray used hereabove. HIF-1α siRNA completely prevented the upregulation of BNIP3 mRNA level induced by etoposide and hypoxia. This observation validated the effect of the siRNA since BNIP3 is known to be a HIF-1 target gene. Interestingly, we observed that HIF-1α siRNA prevented the hypoxia-induced decrease in BAK1 mRNA level without affecting BAX expression (Fig. 6A). It has to be noted that HIF-1α siRNA had no influence on TP53 mRNA level. All these results were confirmed by real-time RT-PCR assays (Table 4).