A Global Approach Combining Proteome Analysis and Phenotypic Screening with RNA Interference Yields Novel Apoptosis Regulators*S

Global approaches like proteome or transcriptome analyses have been performed extensively to identify candidate genes or proteins involved in biological and pathological processes. Here we describe the identification of proteins implicated in the regulation of apoptosis using proteome analysis and the functional validation of targets by RNA interference. A high-throughput platform for the validation of synthetic small interfering RNAs (siRNAs) by quantitative real-time PCR was established. Genes of the identified factors were silenced by automated siRNA transfection, and their role in apoptotic signaling was investigated. Using this strategy, nine new modulators of apoptosis were identified. A subsequent detailed study demonstrated that hepatoma-derived growth factor (HDGF) is required for TNFα-induced release of pro-apoptotic factors from mitochondria. The strategy described here may be used for hypothesis-free, global gene function analysis.

Global approaches like proteome or transcriptome analyses have been performed extensively to identify candidate genes or proteins involved in biological and pathological processes. Here we describe the identification of proteins implicated in the regulation of apoptosis using proteome analysis and the functional validation of targets by RNA interference. A high-throughput platform for the validation of synthetic small interfering RNAs (siRNAs) by quantitative real-time PCR was established. Genes of the identified factors were silenced by automated siRNA transfection, and their role in apoptotic signaling was investigated. Using this strategy, nine new modulators of apoptosis were identified. A subsequent detailed study demonstrated that hepatoma-derived growth factor (HDGF) is required for TNF␣-induced release of pro-apoptotic factors from mitochondria. The strategy described here may be used for hypothesis-free, global gene function analysis.

Molecular & Cellular Proteomics 4:44 -55, 2005.
Identification of new factors involved in physiological or pathological processes is of great interest for basic and applied research. In general, two basically different approaches for screens can be chosen. Hypothesis-driven approaches are based on existing knowledge of the process, and screens are designed for a set of factors fulfilling certain criteria like known activities or interactions (e.g. kinase substrate screens, two-hybrid screen). Hypothesis-free approaches aim to identify all factors involved in a process irrespective of their activity (e.g. proteome or transcriptome analysis). In theory, hypothesis-driven approaches will lead to a well-defined limited set of candidates for one defined aspect of a whole process. Factors identified by hypothesis-free approaches may be involved in all steps of the process but afterward their function remains to be defined.
Apoptosis, a physiological form of cell death, is implicated in embryonic development, cell homeostasis, and immune defense and plays a role in infectious diseases. Defects in apoptotic signaling have been connected to cancer development and degenerative diseases (1). The signals activating or preventing apoptosis regularly depend on posttranslational modification of apoptosis regulators; cleavage by caspases (2), phosphorylation (3), and/or translocation between different subcellular compartments (4).
A technology well suited for the analysis of complex protein compositions, modifications, and translocations is proteome analysis based on two-dimensional gel electrophoresis (2-DE) 1 (5). Proteome analysis has been used extensively to study apoptosis in different systems, and more than 100 proteins have been identified by this approach (for review see Ref. 6). However, only a few of these proteins were functionally analyzed for their role in apoptosis.
Silencing of gene expression by RNA interference (RNAi) has proven to be a robust and straightforward technique for gene function analysis. Synthetic small interfering RNAs (siR-NAs) have been widely used to transiently "knockdown" gene expression in cultured cells for loss of function (LoF) analyses. siRNAs consist of short 21-to 23-nt long double-stranded RNAs that, upon delivery into cells, directly activate the RNAi process (7). However, not all siRNAs designed with the currently available algorithms induce efficient silencing of gene expression. For the analysis of defined apoptotic signaling pathways, the efficiency of a siRNA in interfering with the expression of individual genes must be determined in order to draw any significant conclusions. Pools of different siRNAs directed against the same gene were also used in order to increase the probability of efficient inhibition of gene expression. Recent reports, however, show that siRNAs may reduce the expression of unrelated genes (8). Another potential difficulty in using RNAi for gene function analysis is the induction of the interferon (IFN) response (9). Therefore, working with single validated siRNAs minimizes the possibility of inducing unwanted off-target effects.
Here, we apply a general strategy for the identification, validation, and functional analysis of proteins involved in apoptosis regulation. Proteins modified during apoptosis were first identified by proteome analysis. An automated platform was used for the validation of siRNAs by quantitative real-time PCR (q-PCR). Functional siRNAs were subsequently used in LoF screens to analyze the role of modified proteins in modulating apoptosis.
Separation of the Compartments-Approximately 1 ϫ 10 8 Jurkat T cells were centrifuged for 10 min at 1,300 U/min at room temperature in a Megafuge 1.0R (Heraeus, Hanau, Germany). The supernatant was discarded and the pellet was washed twice with 10 ml PBS (Life Technologies, Inc.) and once with MB buffer (400 mM sucrose, 50 mM Tris, 1 mM EGTA, 5 mM 2-mercaptoethanol, 10 mM potassium hydrogenphosphate, pH 7.6, and 0.2% BSA) and centrifuged as above. The pellet was suspended in 4 ml of MB buffer and incubated on ice for 20 min. Subsequently, the cells were homogenized and centrifuged at 3,500 U/min for 1 min at 4°C (Rotor SS-34; RC5B; Sorvall, DuPont, Bad Hamburg, Germany). The supernatant contained the mitochondria/cytosol/membranes and the pellet contained the nuclei and unbroken cells. The mitochondrial fraction was pelleted by centrifugation at 8,600 U/min for 10 min at 4°C (Rotor SS-34; RC5B; Sorvall). The supernatant contained the cytosol and membranes. The pellet was suspended in 0.4 ml of MSM buffer (10 mM potassium hydrogenphosphate, pH 7.2, 0.3 mM mannitol, and 0.1% BSA) and purified by sucrose gradient centrifugation in 10 ml of SA buffer (1.6 M sucrose, 10 mM potassium hydrogenphosphate, pH 7.5, and 0.1% BSA) and 10 of ml SB buffer (1.2 M sucrose, 10 mM potassium hydrogenphosphate, pH 7.5, and 0.1% BSA) at 20,000 U/min, 1 h, 4°C (Rotor SW-28; L8-70 M Ultracentrifuge; Beckman, Fullerton, CA). The interphase, which contained the mitochondria, was collected, suspended in 4 volumes of MSM buffer, and cenrifuged again at 15,500 U/min for 10 min at 4°C (Rotor SS-34; RC5B; Sorvall). The pellet was suspended in MSM buffer without BSA and could be stored at Ϫ70°C. The supernatant with the cytosol and membrane was centrifuged at 100,000 U/min, 20 min, 4°C (Rotor TLA120.2; Optima TLX Ultracentrifuge; Beckman). The pellet contained the membranes. The pellet with the nucleus was suspended in 5 ml of PBS and centrifuged for 2 min at 3,500 U/min at 4°C (Rotor SS-34; RC5B; Sorvall). The pellet was suspended in 1 ml of NB buffer (10 mM HEPES, pH 7.4, 10 mM KCl, 2 mM MgCl 2 , 1 mM DTT, and 1 mM Pefabloc) and incubated for 1 h on ice, subsequently homogenized and applied to 10 ml of 30% sucrose in NB buffer. After the centrifugation with the Megafuge 1.0R (Heraeus) at 2,000 U/min for 10 min at 4°C, the pellet was washed twice with 6 ml of NB buffer, centrifuged as above, suspended in 1 ml of NB buffer, and centrifuged again at 10,000 U/min for 10 min at 4°C (Rotor SS-34; RC5B; Sorvall). The pellet could be stored at Ϫ70°C. Two-dimensional Gel Electrophoresis-The proteins were separated by large 2-DE gels (gel size 30 cm ϫ 23 cm) (10). Briefly, IEF rod gels were used for the first dimension with a diameter of 0.9 mm for analytical gels and 2.5 mm for preparative gels. SDS-PAGE gels with 15% w/v acrylamide and 0.2% bisacrylamide were used for the second dimension. Preparative gels were stained with Coomassie Brilliant Blue R-250 or G-250 (Serva, Heidelberg, Germany). Analytical gels were stained with silver nitrate.
Tryptic Digestion-The Coomassie blue-stained single gel spots from Jurkat T cells were excised with a scalpel for in-gel digestion with 0.1 g of trypsin (Promega, Madison, WI) in 20 l of 50 mM ammonium bicarbonate, pH 7.8. The samples were dissolved in 1 l 0.5% aqueous TFA/ACN (2:1) for the analysis by MS.
Database Searching-The proteins were identified by using the peptide mass fingerprinting analysis software MS-Fit (prospector.ucsf.edu/ucsfhtml3.2/msfit.htm) or Mascot (www.matrixscience-.com/cgi/search_form.pl?FORMVERϭ2&SEARCHϭPMF). The NCBI and Swiss-Prot databases were used for the searches by considering at maximum one missed cleavage site, pyro-Glu formation at Nterminal Gln, oxidation of methionine, acetylation of the N terminus, and modification of cysteines by acrylamide. The molecular masses and isoelectric points were calculated by employing the software Compute pI/Mw (www.expasy.ch/tool/pi_tool.html).
Validation of RNAi by q-PCR, Western Blot, and 2-DE-For validation of siRNAs by q-PCR, 10,000 -20,000 cells/well were seeded in a 96-well plate 1 day prior to transfection. Trypsinized cells were diluted in medium to final a concentration of 10,000 -20,000 cells per 100 l. These were plated by a BioRobot® 8000 system (Qiagen, Valencia, CA) into 96-well cell culture plates. Equal distribution of cells within each and between different wells was verified by microscopy. Transfection was performed with 0.1-0.25 g siRNA (final concentration 80 -200 nM) directed against target genes or Luciferase as control and 2 l of Transmessenger per well. Each siRNA was transfected in triplicates using the BioRobot® 8000 system (Qiagen) as follows: For each triplicate, 9 l of 0.1 g/l (7 M) siRNA solution were mixed with 44 l of EC-R buffer and incubated for 5 min. Then 7 l of Transmessenger reagent were mixed with 28 l of EC-R buffer, added to the siRNA and incubated for 15 min. After addition of 263 l of medium, 100 l of the transfection mixture were put on cells formerly seeded in 96-well plates and of which medium was aspirated directly before addition of transfection mixture. Knockdown measurements were performed three times independently. After 48 h, RNA was isolated using the RNeasy® 96 BioRobot® 8000 system (Qiagen). The relative amount of target mRNA was determined by q-PCR using Quantitect TM SYBR® Green RT-PCR Kit following manufacturer's instructions (Qiagen), and primers surrounding the siRNA binding site were employed. The relative expression levels of target mRNA were normalized against control transfected cells. GAPDH was used as an internal standard.
For validation of siRNAs by Western blot, 50,000 cells/well were seeded in 12-well plates 24 h prior to transfection. siRNAs against target genes and firefly luciferase as negative control were transfected using the Transmessenger transfection kit (Qiagen) at a final concentration of 80 -160 nM. For 2-DE analysis, HeLa cells were transfected with siRNAs in six-well plates according to manufacturer's instructions.
Apoptosis Assays-HeLa cells were transfected as described for validation of RNAi. One day posttransfection, cells were trypsinized and 15,000 cells/well were seeded in a 96-well plate. After an additional 48 h, apoptosis was induced using 50 ng/ml TNF-␣ and 10 g/ml cycloheximide for 5 h. ApoOne assays (Promega) were performed according to manufacturer's instructions. For TUNEL assays (Promega), cells were fixed with 3% paraformaldehyde for 30 min at room temperature, washed twice with PBS, and permeabilized with 0.2% Triton X-100 in PBS for 10 min. After washing with PBS, cells were covered with 25-35 l of equilibration buffer at room temperature for 10 min. Then cells were labeled with fluorescein-12-dUTP for 60 min. The reaction was stopped by addition of 2-fold saline-sodium citrate for 15 min, washed with PBS, and mounted on glass slides with Moviole. Experiments were performed three times independently.
Immunofluorescence Microscopy-Cells were fixed with 3% paraformaldehyde for 30 min at room temperature, washed with PBS, and permeabilized with 0.05% Triton X-100 in PBS. After blocking with 0.5% BSA, 1% goat serum, and 0.05% Triton X-100 in PBS, cells were incubated with a mouse monoclonal antibody for Smac/Diablo (Pharmingen), Cytochrome c (Pharmingen) or rabbit polyclonal antibody for HtrA2/Omi (Alexis, Grü nberg, Germany) diluted to 1:100 in 0.5% BSA and 1% goat serum in PBS for 60 min. Then cells were washed and re-probed with Cy-2-labeled anti-rabbit or anti-mouse antibody diluted to 1:100 in the same buffer for 60 min. Finally, chromatin was stained using Hoechst dye (33258) and mounted on to glass slides with Moviole.

Identification of Apoptosis-modified Proteins-In order to
identify new proteins involved in the regulation of apoptosis, a detailed study including the identification, validation, and functional analysis was performed (Fig. 1). Proteome analysis was chosen for the identification of new apoptosis factors. The involvement of the identified candidates in the regulation of apoptosis was then validated by LoF screens using RNAi. Only factors generating a phenotype where further characterized (Fig. 1).
Jurkat T cells were treated with anti-Fas (CD95/Apo-1) antibody for 6 h to induce apoptosis. The total cell lysate, purified cytosol, mitochondria, nuclei, and membranes of apoptotic and control cells were analyzed by 2-DE. A representative comparison of the 2-DE pattern of the nonapoptotic and apoptotic nuclear fraction is shown in Fig. 2. The purity of the cytosol, mitochondria, and nuclei was confirmed by their compartment-specific 2-DE pattern using previously identified marker proteins with known subcellular localization (11). The membrane fraction revealed a different 2-DE pattern compared with the other compartments; however, the identified marker proteins indicated that membrane fractions were not pure.
Peptide mass fingerprinting was used to identify 84 different proteins out of 184 spots (see Table I and supplemental Table S1). Most of these proteins were not previously known to be involved in apoptosis. Information about degradation, changes in charge, and translocation between subcellular compartments was obtained for many of the identified proteins (Table I and supplemental Table S1).
Protein degradation, most likely due to cleavage by caspases, was predicted to occur if a protein identified in the control gels was found with a reduced molecular mass in the gels of apoptotic cells. This was the case for FUSE binding protein 1, GAP SH3-domain binding protein 1, hnRNP K, KH-type splicing regulatory protein, lamin B1, lamin B2, NSassociated protein 1, proteasome subunit p50, Rho GDI 2, and splicing factor p54 nrb . Degradation was also predicted if a protein was identified only in apoptotic samples with a reduced molecular mass of more than 15% compared with the theoretical value. The group of candidate caspase substrates identified by this criteria included cGMP-dependent protein kinase I ␣ isozyme, GMP synthase, hnRNP R, KIAA1470, myosin heavy chain nonmuscle, nucleolin, and T-complex protein 1 ␤ subunit.
60S ribosomal protein P0 and the hnRNPs A/B, A3, C1/C2, and D0 were identified due to a change in their charge, as indicated by a shift of the isoelectric point without a change in molecular mass (Table I and supplemental Table S1).
Protein translocation during apoptosis was predicted by the identification of a particular protein in different compartments of control and apoptotic cells. Only cytosol, mitochondria, and nuclei were analyzed because of the ambiguous composition of the membrane fraction. Translocation of proteins between cellular compartments due to apoptosis could be found to occur for the proteins 3-hydroxyacyl-CoA dehydrogenase type II, 60S ribosomal protein P0, lamin B1, lamin B2, hnRNP A/B, A1, A3, D0, K, and the splicing factor p54 nrb (Table I and supplemental Table S1).
Proteins found only in control cells were either identified in the compartments they were supposed to localize according to database entries or in the membrane fraction. However, an exception was the nuclear protein RAD23 homolog B, which was found in the cytosolic fraction of the control cells.
The translocation of proteins identified only in apoptotic cells can be predicted as well (Table I and Table S1). In cells induced to undergo apoptosis, the elongation factor Tu, isocitrate dehydrogenase 2, and peroxiredoxin 3 seemed to translocate from the mitochondria to the nucleus, hnRNP F from the nucleus to the mitochondria, nucleolin from the nucleus to the cytosol, and platelet-activating factor acetylhydrolase IB ␥ subunit from the cytosol to the nucleus.
Setup of an Automated Protocol for siRNA Validation-In order to unravel whether the identified factors were directly involved in apoptosis regulation, LoF analysis by RNAi was performed. siRNAs specific for the genes of the identified factors were designed, and the efficiency of gene silencing in HeLa cells was tested by quantification of the respective mRNA by q-PCR (Fig. 3A). In the case of ATP synthase ␤ chain, RNAi-induced reduction of specific mRNAs by more than 99% (Fig. 3E) correlated with the decrease of protein expression by 96% (Fig. 3, B and F). The combined use of proteomics and RNAi can also be applied to test for the functionality of siRNAs by monitoring the reduction of the corresponding protein spot on 2-DE gels. The RNAi-induced inhibition of ATP synthase ␤ chain expression led to the clear reduction of the corresponding protein spot on silver-stained gels ( Fig. 3, C, D, and G). Therefore, monitoring RNAi-induced reduction of protein expression with 2-DE may allow analysis of siRNAs functionality independent of antibodies or other specific detection tools.
The efficiency of interference with gene expression varied significantly between different siRNAs. Therefore, the efficiency of each siRNA had to be tested. siRNAs were transfected in triplicate into HeLa cells seeded in 96-well plates, mRNA was prepared, and the amount of mRNA for the individual genes was quantified by q-PCR. In order to handle the large number of individual experiments, a protocol for the automated cell seeding, siRNA transfection, preparation of total RNA, and q-PCR sample preparation was established on a BioRobot® 8000 system (Qiagen). The automated siRNA validation platform theoretically allows 2,400 transfections (ϳ800 triplicates) per day and 200 complete siRNAs validations per week. Using this procedure, 45 of the 61 siRNAs tested were shown to reduce the respective mRNA by more than 75% (Fig. 3A, Table I).
Loss of Function Analysis of Apoptosis-modified Proteins-The validated siRNAs were transfected into HeLa cells and   TNF-␣/cycloheximide (TNF/CHX) was added 72 h posttransfection to test the effect of the knockdown of the identified factors. siRNAs directed against luciferase (siLuc, no cellular target present) and lamin A/C (cellular target present) were included on each plate as a reference. RNAi-induced inhibition of caspase-8 expression was expected to block TNF/ CHX-induced apoptosis (12) and, therefore, served as a positive control for apoptosis inhibition. Samples were analyzed for their caspase activity indicative for the induction of the apoptotic program. Caspase activity increased 3-to 4.5-fold in both negative controls (siLuc, lamin A/C), while cells lacking caspase-8 did not respond to TNF/CHX-treatment (1.2-fold) (Fig. 4A). RNAi-induced gene silencing of proteins found to be modified during apoptosis resulted in a range of different responses. No difference in response to TNF/CHX treatment, compared with the negative control, was observed for expression knockdown of 42 proteins. Inhibition of apoptosis was defined as follows: less than 2-fold caspase activation of the TNF/CHX-treated compared with untreated samples or less than 50% of the caspase activity measured in the TNF/CHXtreated siLuc-transfected control samples. According to these criteria, silencing of four candidate genes prevented apoptosis induction (Fig. 4, A-D). Sensitization to apoptosis was defined as more than a 9-fold caspase activation of the TNF/ CHX-treated compared with untreated samples or more than 300% of the caspase activity measured in the TNF/CHXtreated control samples. By these criteria, silencing the genes of five proteins resulted in a sensitization to TNF/CHX-induced apoptosis (Fig. 4, A-D). The effects measured for the silencing of the ATP synthase ␤ chain gene were probably due to a toxic effect generated by ATP depletion in these cells rather than a specific role in apoptotic signaling (Fig. 4B). In summary, down-regulation of nine genes previously not known to play a role in apoptosis generated a clear phenotype (Table II).
To confirm these results with a different apoptosis assay, the expression of the proteins identified in the caspase assay as new apoptosis regulators was inhibited by RNAi and DNA fragmentation was tested by terminal deoxynucleotidyl transferase-mediated dUTP nick-endlabeling (TUNEL) (Figs. 4, E-G, and 5). The results confirmed the caspase activity data obtained before because silencing the genes of hepatomaderived growth factor (HDGF), eukaryotic initiation factor 3 subunit 4 (eIF3-p42), proteasome subunit ␣ type 3 (PSMA3), and hnRNP C resulted in a complete inhibition of apoptosis, whereas silencing of nucleophosmin, SMARCE1, Cctb, hnRNP D, and HUMHFP sensitized cells to TNF/CHX-induced apoptosis (Figs. 4E and 5).
In contrast to long double-stranded RNAs, siRNAs were not previously believed to induce IFN responses (7). However, recent observations suggested that siRNAs may still elicit an IFN response (9). To rule out whether the observed phenotypes may depend on the induction of IFN-regulated genes, we tested the up-regulation of the marker genes protein kinase R or 2Ј-5Ј-oligoadenylate synthetase (8) in siRNA-transfected cells by q-PCR. None of the tested marker genes was up-regulated in the transfected samples, thus excluding an IFN type of response (data not shown).
In order to exclude that the unspecific knockdown of other genes (8) was responsible for the observed phenotypes, a second functional siRNA was tested. Suppression of the HDGF gene using two different siRNAs always led to a com-  Table I. B, HeLa cells were transfected with the indicated siRNAs. Three days posttransfection, cells were tested for the expression of ATP synthase ␤ chain (F-1 ␤) and ␣-tubulin by Western blot analysis as described in "Experimental Procedures." C and D, HeLa cells were transfected with (C) Luciferase (siLuc) and (D) ATP synthase ␤ chaindirected siRNAs (F-1 ␤). Three days posttransfection, cells were lysed and proteins were separated by 2-DE. The arrows point to the spots of ATP synthase ␤ chain. E-G, diagrams show the quantification of mRNA, the Western blot data shown in B, and the spots separated by 2-DE (C and D). Protein bands or spots were quantified using the TopSpot software. ATP synthase ␤ chain expression levels determined by Western blot were normalized against ␣-tubulin expression. Expression levels of ATP synthase ␤ chain transfected with siLuc was set as 100. Numbers above the columns indicate relative expression levels.
FIG. 4. Screen for caspase activity and DNA fragmentation. A and B, HeLa cells were transfected with siRNAs against the indicated genes in 96-well plates. Three days posttransfection, apoptosis was induced using 50 ng/ml TNF-␣ and 10 g/ml cycloheximide for 5 h. ApoOne assays were performed to measure the caspase activity. Gray columns represent relative caspase activity in cells without and black columns with TNF/CHX treatment. Caspase activities were normalized to Luciferase siRNA-transfected cells without TNF-␣ incubation. C and D, columns represent relative caspase activities for each siRNA with versus without TNF-␣. The arrows indicate the factors used for further analysis. E and G, indicated siRNAs were transfected into HeLa cells. After 3 days, apoptosis was induced as in A and B. Fragmented DNA was stained using TUNEL assay. Bars represent mean values of at least three independent experiments. F, shown are the mean values of the relative reduction of mRNA induced by the indicated siRNAs. Error bars represent Ϯ S.D. of the mean.

TABLE II List of proteins identified in the RNAi loss of function screen for apoptosis inhibition or sensitisation
Shown are the proteins abbreviations, the respective gene accession number used to design the siRNA, and the phenotype observed in knockdown cells upon TNF/CHX treatment.  (Figs. 4, F and G, and 5). Likewise, down-regulation of nucleophosmin and hnRNP D expression with two different siRNAs resulted in sensitization, suggesting a specific effect of nucleophosmin and hnRNP D on regulating TNF/CHX-induced apoptosis (Fig. 4, F and G). In the case of PSMA3, the degree by which siRNAs inhibited TNF/CHX-induced apoptosis correlated with their efficiency to inhibit PSMA3 RNA. In contrast, transfection of SMARCE1[-1] and SMARCE1[-2] increased the sensitivity of HeLa cells to TNF/CHX-induced apoptosis (Fig. 4, F and G), although only SMARCE1[-2], but not SMARCE1[-1], induced efficient mRNA degradation (Fig. 4, F and G). These data strongly argue for the use of validated siRNAs for gene function analysis, which allow the direct correlation of gene expression silencing and gene function analysis.

Modulation of Apoptosis by Silencing HDGF Expression-
Although loss of all nine candidates clearly affected apoptosis, we selected to focus on HDGF. Silencing the HDGF gene induced changes in morphology, even in the absence of apoptotic stimuli, a special feature of HDGF not seen with any other gene knockdown tested (Fig. 5). The morphology of siHDGF[-1]-and siHDGF[-3]-transfected cells was unchanged upon treatment with TNF/CHX (Fig. 5). Thus, suppression of HDGF expression causes a complete inhibition of biochemical (Fig. 4) and morphological features of apoptosis. However, because cells transfected with siHDGF[-1] and siHDGF[-3] rounded up, it was important to test whether inhibition of apoptosis was due to an overall toxic effect of the HDGF siRNAs. Because single transfections of siRNAs cause only a temporary knockdown of gene expression for up to 7 days, one would expect that the phenotype induced by siHDGF transfection is reversible if the transfected cells were not damaged. After 7 days, cells initially transfected with siH-DGF[-1] or siHDGF [-3] fully restored the morphology of control-transfected cells, suggesting that interfering with HDGF expression alone was not toxic for the cells (not shown).
During TNF-␣-induced apoptosis, mitochondrial factors are released to promote cell death. While cytochrome c is required for the activation of caspase-9 in the apoptosome (13), Smac/Diablo antagonizes the activity of inhibitor of apoptosis proteins (IAPs). The lack of effector caspase activity and the complete prevention of DNA fragmentation in HDGF knockdown cells raised the question if the block occurs upstream of mitochondrial membrane permeabilization. Neither Smac/Diablo nor HtrA2/Omi (not shown), another inhibitor of IAPs, were released into the cytosol in cells lacking HDGF in response to TNF-␣ induction (Fig. 6B). Likewise, cytochrome c was also retained in the mitochondria of these cells (Fig. 6A). These data indicate that the inhibition of apoptosis detected in HDGF knockdown cells FIG. 5. Morphology and TUNEL assay of cells with RNAi-induced suppression of candidate genes. HeLa cells were transfected with the indicated siRNAs, and apoptosis was induced 3 days posttransfection as described in Fig. 4, A and B. Fragmented DNA was stained using the TUNEL assay. Representative sections are shown in phase contrast and fluorescence images for fragmented DNA.
was probably due to the lack of mitochondrial membrane permeabilization.

DISCUSSION
One of the major challenges in the postgenomic era is the functional analysis of all human genes. One way to approach this goal is the performance of LoF screens, which have been used with overwhelming success for gene function analysis in simple model organisms (14,15). The high number of genes and the lack of straightforward genetic systems for LoF analyses hampered similar approaches in mammalian cells. RNA interference is very likely "the technology" that will allow global LoF approaches in mammalian cells in the near future (16). Here we describe a technology for the selection of a subset of proteins modified during apoptosis and their defined LoF analysis by RNAi.
Signals regulating apoptosis mainly depend on posttranslational modifications of apoptosis regulators. Using highresolution 2-DE, we have identified 84 proteins modified during apoptosis. As expected, the identified proteins belong to very different facets of cell life: factors regulating the cytoskeleton, transcription, splicing, chromatin structure, metabolic, and anabolic pathways were found. We previously showed that one of the caspase substrates identified in a different screen, c-Abl kinase, requires caspase-mediated cleavage for DNA damage-induced pro-apoptotic signaling (17). This ex-ample showed that information on the type of modification might be very important to unravel molecular mechanisms. Other modifications covered protein translocations, which could be followed from one compartment to the other. For instance, one of the proteins identified in the screen, peroxiredoxin 3, an oxidoreductase involved in the detoxification of peroxide (18), translocated from the mitochondria to the nucleus during apoptosis (supplemental Table S1). An intriguing parallel is given by apoptosis-inducing factor, an oxidoreductase, which translocates from the mitochondria to the nucleus and is involved in caspase-independent apoptotic signaling (19). The many translocation events identified in this screen are of potential interest, because many published examples underline the importance of protein translocation to various subcellular compartments for apoptotic signaling (4).
The subsequent screen for factors involved in TNF/CHXinduced apoptosis by suppressing gene expression unveiled nine new potential modulators of apoptosis. None of the identified proteins belong to the well-known apoptosis effectors like the caspases or inhibitors like the IAPs or Bcl-2 proteins (20,21). This demonstrates that hypothesis-free approaches are useful in finding new factors involved in wellstudied signaling pathways.
SiRNAs designed for silencing the SMARCE1 and nucleophosmin genes strongly sensitized HeLa cells to TNF/CHXinduced apoptosis. Loss of nucleophosmin expression has recently been shown to block pre-ribosomal RNA processing and induces cell death (22). In the case of SMARCE1, a component of the evolutionary conserved Swi/SNF complex involved in chromatin-remodeling (23), the mechanism of sensitization to apoptosis is not known. An interesting observation was, however, that both siRNAs, SMARCE1[-1], and SMARCE1[-2] sensitized HeLa cells similarly to TNF/CHXinduced apoptosis, although the efficiency to reduce SMARCE1 mRNA was 25 and 85%, respectively. A possible explanation might be that SMARCE1 functions in the Swi/SNF complex in a stoichiometric manner, where even slight disturbances of the complex composition results in a nonfunctional complex. An alternative explanation might be that both siR-NAs suppress SMARCE1 protein expression to a similar degree, SMARCE1[-1] via a micro-RNA-induced mechanism (24), which would result in the inhibition of SMARCE1 translation and SMARCE1[-2] by siRNA-induced RNAi leading to mRNA degradation.
Proteins identified so far in the current screen required for caspase activation and/or DNA degradation were PSMA3, eIF3-p42, hnRNP C, and HDGF. PSMA3 belongs to the 20S subunit of the proteasome, a large proteolytic complex (25). Proteins destined for degradation by the proteasome are usually ubiquitinated, recognized by the 19S subunit, and then degraded by the 20S subunit of the proteasome (25). Numerous proteins like p53, caspases, and Bcl-2 family members involved in the regulation of apoptosis are degraded by the FIG. 6. Modulation of apoptosis by HDGF expression knockdown. HeLa cells were transfected with the indicated siRNAs and apoptosis was induced as described in Fig. 4, A and B. Nuclei were stained with Hoechst, and the release of (A) cytochrome c (CytC) or (B) Smac/Diablo was monitored by immunofluorescence staining under a fluorescence microscope at 400ϫ magnification. proteasome (26). Ring-finger-containing members of the IAP family can function themselves as ubiquitin ligases to ubiquitinate their target proteins (27). Interestingly, although proteasomal degradation plays such an important role for apopototic signaling, only expression knockdown of the PSMA3 gene, but not of the genes for the proteasomal subunits p40 and p50, blocked apoptotic signaling. Therefore, the different subunits may fulfill specific functions during apoptosis. An intriguing result was the inhibition of apoptosis by RNAiinduced suppression of the eIF3-p42 gene. In general, protein synthesis is rapidly down-regulated in apoptotic cells, consistent with the cleavage and inactivation of a number of translational initiation factors by caspases like eIF4G, eIF4B, eIF2a, and the p35 subunit of eIF3 (28). How a translational initiation factor may contribute to proapoptotic signaling remains to be shown.
HDGF is a secreted growth factor initially purified from the HuH-7 cell line (29). It contains nuclear targeting sequences, and its growth-promoting activity has been shown to depend on the ability to enter the nucleus (30,31). HDGF expression might be of clinical relevance because tumor cells selected for radiation resistance expressed only low levels of HDGF (32). We have demonstrated that silencing the HDGF gene prevents TNF/CHX-induced apoptosis.
Upon binding TNF-␣, the TNF receptor recruits TNFR1associated protein (TRADD), which is then modified to recruit Fas-associated with death domain (FADD) and caspase-8 (12). However, TRADD also associates with TNF-␣ receptorassociated factor 2 (TRAF-2), which recruits the caspase inhibitors FLIP L and cIAP-1 (12). Caspase-8 is not activated unless the inhibitory complex is released from the TNF receptor (12). Deng and colleagues (33) recently showed that induction of apoptosis via the TNF receptor requires the release of the IAP antagonist Smac/Diablo from the mitochondria, which then disrupts the TRAF2-cIAP1 complex and permits activation of caspase-8 (33). RNAi-induced silencing of the HDGF gene prevented the release of Smac/Diablo upon TNF-␣ treatment, suggesting that lack of HDGF interferes with the release of proapoptotic factors from the mitochondria. Whether caspase-8 activation is also blocked in these cells and how the loss of HDGF interferes with the release of mitochondrial factors remains to be shown. We have shown that candidate screens by proteomics combined with phenotypic screens by RNAi are powerful approaches to identify new factors involved in apoptosis. The use of RNAi for the analysis of gene function allows the rapid screening for factors involved in discrete parts of complex signaling cascades, but one has to be cautious with the interpretation of the result. The careful validation of the RNAi reagents must be correlated to the phenotypic observation. In this respect, gene silencing by RNAi should be monitored on the protein levels, which in most cases is the relevant readout for the interpretation of observed phenotypes. The connection of proteomics and RNAi is, in this respect, the most promising combi-nation to meet the requirements of future gene function analysis.