The aim of this study was to adopt a functional genomics approach to screen for human genes that induce apoptosis when over-expressed. In order to do this, our strategy required the construction of a microarray of a large number of human genes in mammalian expression vectors. In a previous study, GFP-tagged genes in Gateway expression vectors were used to examine the sub-cellular localisation of proteins over-expressed by reverse transfection . In common with other studies, we found that gene-tagging can disrupt the normal sub-cellular localisation and therefore presumably the function of the protein. In addition, sub-cloning of the gene inserts has the potential to introduce errors in the ORF during vector construction and inserting large numbers of genes into Gateway constructs is both costly and time consuming. Therefore, for the present study we decided to adopt a new approach to the construction of a high-content microarray of human genes by direct use of the readily available full-length MGC clones [9, 10]. This strategy was possible as many of the MGC clones were already in the pCMV-SPORT6 vector, which contains a CMV promoter to drive expression of the ORF in mammalian cells. Plasmid preparations were attempted for 2,976 MGC clones but we found that many did not yield enough/any product, despite repeated attempts. However, more than 2 μg plasmid DNA was recovered from 1,959 clones. In order to maximise the likelihood of observing the effects of over-expression on the cells covering arrays without compromising the array content, each of the 1,959 purified MGC clones was printed in quadruplicate onto a glass slide to form the array. After inclusion of control features, the array possessed 9,888 features in total. As such, this represents the largest over-expression cell-based reverse transfection microarray published to date.
Following growth of HEK293T cells over the array, the TUNEL assay was used to detect genes which had induced cell death when over-expressed. The assay was repeated on four separate arrays. When proteins only positive in two or more of the four assays were taken into account, 79 of the 1,959 genes (4%) appeared to be potentially inducing cell death. For verification, these 79 genes were then transfected in 6-well plates. The results from the 6-well plate assay indicated that out of the 79 positives from the array, 10 (12.7%) were true positives (Table 1). This would indicate that the arrays gave a fairly high false positive rate. Calculation of the false negative rate presents more of a challenge. However, in an attempt to evaluate this we determined the number of genes on the array that have been linked with apoptosis by Gene Ontology (GO) categorisation. Eleven genes on the array were listed under the apoptosis category, but only one of these genes (STK3) was observed in the final list of 10 positive genes. Assuming that all of these genes are capable of inducing apoptosis when over-expressed, the false negative rate of the array based assay was estimated as 91%. Whether all 11 genes are truly capable of inducing apoptosis if over-expressed and whether they would do so in all cell types is uncertain. The high false positive and estimated false negative rates of this assay are attributable to a number of factors. In order to maintain the positional address of the positive signals relative to the marker gene (GFP) transfections during manual inspection of the arrays, it was necessary to view the array at a relatively low magnification (×10 objective). At this resolution the relatively weak signal produced by TUNEL assay may have been overlooked or background signal attributed incorrectly. Employing automated high resolution scanning and image analysis tools would not only make scoring of cell-based arrays a lot easier, but would also improve the accuracy of scoring. The necessity for spot recognition software and the storage and analysis of the images present a considerable challenge. These issues are currently being explored and a microscope-based screening platform is being developed with automated sample preparation, image acquisition and data analysis [16, 17]. Improved tools for automated image analysis of cell-based arrays are also being developed elsewhere .
Of the 10 proteins found by this study to induce apoptosis when over-expressed a number have been linked previously with apoptosis, others have not. STK3 is known to be involved in Fas-mediated apoptosis and is cleaved/activated by CASP3 . In addition, it is possible to hypothesise mechanisms of action of four of the other proteins identified. XBP1 binds to the X-box of the HLA-DR-alpha promoter (MHC human class II gene)  and responds to accumulation of unfolded proteins in the ER . It is possible that this change may trigger apoptosis, although this has not been demonstrated previously. There is other evidence that XBP1 may be linked to apoptosis. In an expression profiling study of murine mammary epithelial cells expressing conditionally active STAT3 (which provides an essential death signal for mammary epithelial cells following weaning), XBP1 was highly up-regulated following STAT3 activation . CSTB maintains appropriate equilibrium between free cysteine proteases and their complexes . Cathepsins can be inhibited by CSTB and therefore this could prevent degradation of peptides and proteins, which in turn could possibly act as a pro-apoptotic stimulus. ACO1 represses ferritin and increases TFR translation , and its over-expression is therefore likely to cause a build up of free-iron within the cell. Previous studies have shown that increased levels of intracellular free-iron can induce apoptosis . EXOC7 is a component of the exocyst complex involved in the docking of exocystic vesicles with fusion sites on the plasma membrane  and potentially excessive removal of internal cell contents may cause apoptosis to occur. MLLT11 and CCBP2 are both found in tumours, but no other information is available to allow speculation on a possible mode of action for their induction of apoptosis. The same is true of the relatively uncharacterised genes C22ORF23, MGC5439 and LOC134285.
From even the limited number of 'pro-apoptotic' genes identified by this study, it would appear that they could be further classified into two categories; those that are 'true' apoptotic modulators, the proteins they encode being directly involved with the cell's apoptotic/survival machinery; and those that act indirectly, their over-expression leading to cytotoxic changes within the cell which then triggers a pro-apoptotic response. It is inevitable that screens such as this will expose genes in both categories and further work is necessary to determine their exact mode of action. Whilst 'true' apoptotic modulators might be considered more interesting in the pursuit of improved understanding of the apoptotic cascade, genes that give rise to apoptosis by indirect mechanisms may still have interest as novel therapeutic agents, for example in cancer gene therapy.
In an attempt to ascertain the timing and strength of apoptosis, a time course transfection study using HEK293T cells was undertaken on all 10 genes, a positive control STS and a mock transfected negative control at 12, 24, 36, 48 and 60 hours following transfection. Staurosporine (STS) was used as the positive control as it is an apoptotic effector classically linked to caspase activation . As the TUNEL assay has been shown to potentially detect necrotic death in addition to apoptotic death , a cleaved CASP3 assay was performed on the cultures in addition to the TUNEL assay. STS and all 10 genes identified by the reverse transfection screen led to cultures in 6-well plates exhibiting between 40–70% TUNEL and cleaved CASP3 positive cells after 60 hours compared with only 3–4% in the mock transfection cultures.
Apoptosis has been studied extensively and core pathways and events are generally well established. The regulation of apoptotic cell death is a complex interplay between proteins that promote cell survival and those that promote cell death. It is widely thought that the processes that control the balance between the life and death of a cell are regulated exclusively at the post-transcriptional level. As a result few observations have been made of the transcriptome during this process, although those that have  suggested that this would be a useful approach to further characterise the action of these genes.
STK3, ACO1 and XBP1 were selected over the other seven genes for expression profiling studies (Figure 2) as the proteins they encode had either been shown in previous studies to induce or be connected to apoptosis (STK3 and XBP1) or a clear apoptotic hypothesis could be postulated (ACO1). For the expression profiling study, samples were taken at 12, 24 and 48 hours following transfection to observe the early, mid and late transcriptional events associated with apoptosis. ACO1, STK3 and XBP1 were all found to be constitutively expressed within the cells but at different levels. It was interesting to note that despite differences in constitutive expression, all genes appeared to reach a similar level of up-regulation following transfection and that the changes in expression observed for these transcripts were the largest of all the transcripts represented on the expression profiling array.
It was envisaged that expression profiling experiments might reveal a transcriptional response in these cultures that would provide clues as to the mechanism by which the over-expression of these genes induces apoptosis. Overall, analysis of the microarray expression data indicated that the changes observed at all time points and across all conditions were relatively subtle. There was no strong tendency for the data to cluster according to treatment or time-point (Additional file 2 Figure Two). The lists of differentially expressed transcripts of genes prepared by comparing each time point with its respective mock transfection control, showed many genes to be significantly changing their expression, but on the whole these changes were relatively small i.e. the majority of changes were less than 2-fold in magnitude. For each transfection experiment, many more transcripts were down-regulated in expression than up-regulated during apoptosis progression. A similar observation was also reported in a previous study of the transcriptional events associated with apoptosis following removal of cell survival factors from cultures of HUVEC cells . However, with STS treatment the number of gene transcripts that were significantly down-regulated and up-regulated in expression overall was similar. What this indicates with respect to the fundamental mode of action for STS in apoptosis induction as opposed to the gene over-expression is uncertain. The number of transcripts identified as significantly changing with each treatment and at each time point varied considerably. In order to compare the condition or time-specific changes in transcript expression between the genes, a gene tree was plotted of the 997 transcripts that had changed in expression more than 1.4-fold and that were present in 3 or more of the 12 pair-wise treatment comparisons (Figure 5). There was a surprising degree of uniformity in behaviour of these genes across all conditions and time points. Genes that were up- or down-regulated in one comparison tended to show the same behaviour across all comparisons, although this change did not always reach the level of statistical significance. Indeed, we were unable to find any convincing condition-specific changes in transcriptome activity that might give clues as to the mechanism by which functionally distinct genes induce apoptosis when over-expressed. Whether this is because the events leading to the initiation of a pro-apoptotic response occurred post-transcriptionally or that the changes are too subtle to be recognised is unclear. Rather, these findings strongly indicate that the majority of the changes we observed were associated with a universal pattern of gene regulation during apoptosis, regardless of the initiating trigger.
In order to further explore the apoptotic signatures in this data, the University of Michigan list of apoptosis regulators  was used to identify other potentially interesting apoptosis-associated genes in the list of differentials. 130 genes were shared between this list and the list of genes found here to be differentially expressed. Many gene family members and known interactors that have been previously associated with either the pro-apoptotic or cell survival machinery were present within this new list of 130 apoptosis-associated genes. (The full list of these genes and an in depth discussion of their potential role in apoptosis induction or cell survival is available in the Additional file 4 Tables Two and Three). A literature search was performed on the 130 genes in the list to ascertain their function. Some apoptosis associated genes e.g. NR4A1, EGR1 SLIT2, CASP9, ADM, SMAD7, JUN and TIMP1 significantly changed in their expression in every transfection/treatment compared to the negative control; others were only observed to change under certain conditions. In order to provide a simplified view of this data, these genes were mapped onto a modified version of the KEGG apoptosis pathway if they were present in at least three of the four experimental conditions. Their action in either increasing or decreasing the likelihood of apoptosis and their directional change in expression is indicated (Figure 6, also see Additional file 5 Figures Three to Seven for apoptosis pathways and further discussion for each of the over-expressed gene and STS treatments). Overall, this approach supported the hypothesis that the transfected genes and STS were ultimately acting through similar pathways to induce cell death. In each case, numerous genes were observed to change which were associated with the MAPK8(JNK)/CASP3 pathway and in addition there were clear indications of suppression of the cell survival pathways. Whilst the KEGG pathway was helpful in visualising the apoptotic pathways, many of the genes on the University of Michigan list of apoptosis regulators  had to be added onto the pathway. In addition, there are most likely other genes in the lists of differentials that will be influencing the progression of apoptosis, but have not yet been recognised as being involved with the regulation of cell death.
Overall, the expression profiling studies have provided valuable insights into the transcription changes associated with apoptosis, as the transcriptional changes associated with programmed cell death have not been studied extensively. This current study supports the notion that there are discrete changes in the mRNA abundance of certain genes during apoptosis . As many of these transcripts encode proteins that are known regulators of cell survival and death, it would seem likely that transcriptional regulation of these mediators contributes to a cell's final decision to undergo a programmed cell death.