Systematic Characterization of Nuclear Proteome during Apoptosis

Identification and characterization of the nuclear proteome is important for detailed understanding of multiple signaling events in eukaryotic cells. Toward this goal, we extensively characterized the nuclear proteome of human T leukemia cells by sequential extraction of nuclear proteins with different physicochemical properties using three buffer conditions. This large scale proteomic study also tested the feasibility and technical challenges associated with stable isotope labeling by amino acids in cell culture (SILAC) to uncover quantitative changes during apoptosis. Analyzing proteins from three nuclear fractions extracted from naive and apoptotic cells generated 780,530 MS/MS spectra that were used for database searching using the SEQUEST algorithm. This analysis resulted in the identification and quantification of 1,174 putative nuclear proteins. A number of known nuclear proteins involved in apoptosis as well as novel proteins not known to be part of the nuclear apoptotic machinery were identified and quantified. Consistent with SILAC-based quantifications, immunofluorescence staining of nucleus, mitochondria, and some associated proteins from both organelles revealed a dynamic recruitment of mitochondria into nuclear invaginations during apoptosis.

Identification and characterization of the nuclear proteome is important for detailed understanding of multiple signaling events in eukaryotic cells. Toward this goal, we extensively characterized the nuclear proteome of human T leukemia cells by sequential extraction of nuclear proteins with different physicochemical properties using three buffer conditions. This large scale proteomic study also tested the feasibility and technical challenges associated with stable isotope labeling by amino acids in cell culture (SILAC) to uncover quantitative changes during apoptosis. Analyzing proteins from three nuclear fractions extracted from naïve and apoptotic cells generated 780,530 MS/MS spectra that were used for database searching using the SEQUEST algorithm. This analysis resulted in the identification and quantification of 1,174 putative nuclear proteins. A number of known nuclear proteins involved in apoptosis as well as novel proteins not known to be part of the nuclear apoptotic machinery were identified and quantified. Consistent with SILACbased quantifications, immunofluorescence staining of nucleus, mitochondria, and some associated proteins from both organelles revealed a dynamic recruitment of mitochondria into nuclear invaginations during apoptosis.

Molecular & Cellular Proteomics 5:1131-1145, 2006.
Comprehensive understanding of the biology of the nucleus will require complete identification of the proteome of this important organelle. The nucleus of the cell is where a number of key enzymatic processes occur, and most of these processes are critical for cellular homeostasis: DNA synthesis, DNA replication and repair, transcription, higher order chromatin organization, and gene and chromosomal silencing (1)(2)(3)(4)(5)(6)(7)(8). A comprehensive list of proteins that reside in the nucleus, as well as the proteins that shuttle between multiple subcellular compartments and nucleus, has not been completed to date (9 -11). In addition, different classes of proteins are se-questered in the nucleus based on their affinity to DNA, nuclear matrix, nuclear membrane, nucleolus, specialized nuclear bodies such as speckles and PIKA (polymorphic interphase karyosomal association), or Cajal bodies (coiled bodies) (12)(13)(14)(15). Thus, large scale identification of nuclear proteins from human cells and characterization of their associated physicochemical properties will likely provide insights into the biology of the nucleus.
We have been interested in identifying proteins that control apoptosis in the nucleus. Apoptosis or programmed cell death is a process essential for the development and maintenance of cellular homeostasis of higher eukaryotes (16). One of the hallmarks of apoptosis is rapid chromatin condensation and DNA fragmentation (17). The exact mechanisms that control rapid DNA fragmentation and chromatin condensation in the nucleus are not fully understood, although some of the proteins that are crucial for these events have been identified (18 -25). For example, the caspase-activated DNase (CAD/DFF40) (18,19), lamin protease (Caspase 6) (20,21), Acinus (22), poly(ADP-ribose) polymerase (PARP) 1 (23), programmed cell death protein 8/apoptosis-inducing factor (PCD8/AIF) (24), and endonuclease G (25) have been implicated in DNA fragmentation and chromatin condensation. However, the exact mechanism of how rapid DNA fragmentation and chromatin condensation is regulated during apoptosis is not clear.
Here we applied a quantitative method, termed SILAC (stable isotope labeling by amino acids in cell culture) to compare nuclear proteins during the apoptotic signaling event (26,27). We wished to make significant improvements over the twodimensional gel electrophoresis methodology, which limits identification and quantification of low abundance and membrane proteins (28 -30). SILAC methodology has recently been found to be efficient for large scale protein identification and quantification (31)(32)(33)(34). In this study, we used SILAC and GeLC-MS/MS and rigorously tested the technical limitations associated with this methodology. We identified and quantified 1,174 proteins from nuclear extracts of human T leukemia cells among which are a significant number of mitochondria proteins. To further investigate the biological significance of these findings, we carefully examined some of the identified proteins by immunofluorescence staining and found dynamic nuclear invaginations that closely associate with mitochondria during apoptosis.

EXPERIMENTAL PROCEDURES
Cell Culture, SILAC, and Induction of Apoptosis-Human Jurkat T cells (clone A3, American Type Culture Collection, Manassas, VA) were cultured in SILAC medium as described previously with some modifications (26). The SILAC medium, RPMI 1640 medium deficient in leucine and lysine, was from JRH Biosciences (Lenexa, KA). This medium was supplemented with 10% dialyzed fetal bovine serum and light or heavy amino acids. Light L-leucine ( 12 C 6 ) and L-lysine ( 12 C 6 14 N 2 ) for the treated cells were purchased from Sigma, and heavy L-leucine ( 13 C 6 ) and L-lysine ( 13 C 6 15 N 2 ) were purchased from Cambridge Isotope Laboratories (Andover, MA). The final concentrations of leucine and lysine were 52 and 72.5 mg/liter, respectively. Both light and heavy labeled cell populations were grown for at least five passages in the SILAC media (ϳ10 population doublings) in a humidified incubator with 5% CO 2 at 37°C. For quantitative proteomic analysis during apoptosis, confluent cultures (ϳ8 ϫ 10 5 cells/ ml) of Jurkat cells with or without the heavy isotope labels were harvested, anti-human Fas IgM antibody (250 ng/ml) was introduced into the light isotope-labeled cells for 3.5 h (clone CH-11, Upstate Biotechnology, Lake Placid, NY), and nuclei were isolated for protein identification and quantification.
Isolation of Nuclei by Subcellular Fractionation-Jurkat T cells were collected by centrifugation at 400 ϫ g for 10 min and washed three times with ice-cold PBS at 4°C (137 mM NaCl, 2.7 mM KCl, 1.5 mM KH 2 PO 4 , and 8.0 mM Na 2 HPO 4 ). Cells were then incubated in 5 volumes of hypotonic buffer (Buffer A; 20 mM HEPES-KOH (pH 7.5), 10 mM KCl, 1 mM EDTA, 1 mM DTT, and one protease inhibitor tablet/50 ml (Roche Diagnostics)) for 2 min. Cells were then mixed with an equal volume of 0.5 M sucrose in Buffer A. After 10 min of incubation on ice, the cells were Dounce-homogenized until ϳ90% of the cells became trypan blue-positive. The nuclear pellets were then isolated by centrifugation at 600 ϫ g for 10 min at 4°C. The nuclear pellets were rinsed with 5 ml of homogenization buffer (Buffer B; 20 mM HEPES-KOH (pH 7.5), 10 mM KCl, 1 mM EDTA, 1 mM DTT, 0.25 M sucrose, and one protease inhibitor tablet/50 ml). We followed the methods described in the study by Enari et al. (35) except for the following modifications. 1) The nuclear pellets were gently resuspended in 2.5 ml of Buffer B, mixed with an equal volume of 2.3 M sucrose buffer (Buffer C; 20 mM HEPES-KOH (pH 7.5), 10 mM KCl, 1 mM EDTA, 1 mM DTT, 2.3 M sucrose, and one protease inhibitor tablet/50 ml), and then layered over 5 ml of Buffer C. The tubes were then centrifuged at an average of ϳ60,000 ϫ g (22,000 rpm) for 90 min in swinging bucket rotor SW41 of a Beckmann ultracentrifuge. 2) The pellets containing the purified nuclei were resuspended in 1 ml of Buffer A and centrifuged at 12,000 ϫ g for 10 min. Purified nuclear pellet was then sequentially extracted with three extraction buffers. First, soluble nuclear proteins were extracted with high salt (H. Salt) buffer (50 mM Tris-HCl (pH 8.3), 5 mM EDTA, 500 mM NaCl, and one protease inhibitor tablet/50 ml) with rotation for 30 min at 4°C. Second, weak insoluble nuclear proteins were extracted by modified radioimmunoprecipitation assay (mRIPA) buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM EDTA, 1% Nonidet P-40, 0.25% deoxycholate, and one protease inhibitor tablet/50 ml). Third, strong insoluble nuclear proteins were extracted by SDS buffer (50 mM Tris-HCl (pH 8.3), 5 mM EDTA, 0.5% SDS, and one protease inhibitor tablet/50 ml) and boiled at 100°C for 5 min. After each extraction, the pellets were rinsed twice with the same buffer. The protein concentration of each fraction was measured by the BCA protein assay kit (Pierce).
In-gel Digestion with Trypsin-The nuclear proteins from naïve and apoptotic cells of three fractions (50 g each) were mixed and separated by SDS-PAGE using a 10% NuPAGE gel (Invitrogen). The gel was lightly stained with Coomassie Brilliant Blue G-250 (0.04% Coomassie Brilliant Blue G-250, 3.5% (w/v) perchloric acid) for 5 min and destained overnight with Milli-Q filtered water. Each of the gel slices from three different fractions was in-gel digested with trypsin, and the tryptic peptides were extracted as described previously (36). Briefly the gel pieces (ϳ1-mm cubes) were dehydrated with 100% CH 3 CN, dried in a vacuum concentrator, and digested in a 12.5 ng/l trypsin solution in 50 mM NH 4 HCO 3 buffer overnight at 37°C. The peptides were extracted with 5% formic acid, 50% CH 3 CN three times and further analyzed by a -LC-MS/MS procedure.
-LC Mass Spectrometry Analysis and Protein Identification-Tryptic peptides from each of the gel slices were analyzed using an LTQ linear ion trap mass spectrometer (Thermo Finnigan, San Jose, CA) equipped with a commercial nanospray source (Thermo Finnigan). Samples were loaded into an in-house C 18 microcolumn (100m inner diameter, 360-m outer diameter, 10-cm length, 5-m bead size, 100-m pore size, Column Engineering Inc., Ontario, Canada) by a microautosampler (Famos, Dionex, Sunnyvale, CA) and separated by an Agilent 1100 high performance binary pump. Peptides were separated at a flow rate of ϳ200 nl/min by flow splitting. The solvent gradient of HPLC was linear from 100% solvent A (5% acetonitrile, 0.4% acetic acid, and 0.005% heptafluorobutyric acid) to 80% solvent B (100% acetonitrile, 0.4% acetic acid, and 0.005% heptafluorobutyric acid) for 108 min. The eluent was introduced directly into an LTQ mass spectrometer via electrospray ionization. Each full MS scan was followed by one MS/MS scan of the most intense ion with data-dependent selection using the dynamic exclusion option (Top 1 method). Thus, after the mass spectrometry, 22, 27, and 20 separate ".dat" files were generated from three different fractions, each .dat file representing the data from one gel slice. Next SEQUEST searches were performed on the 32 central processing unit Linux cluster sequentially where each .dat file was searched against the human protein database (37). The SEQUEST output files generated were in html format. The probabilities of peptides and proteins were computed using PeptideProphet and ProteinProphet software tools (38,39). All of the html files were clustered together from each of the nuclear fractions into a large protein page using the INTERACT software tool (30). All the peptides and their associated scores as well as the gel slice information are provided in Supplemental Table 2.
To analyze our dataset for protein redundancy, we utilized the "ProteinProphet" software tool that groups shared sequences within the database and provides a statistical probability for peptide (pcomp) and protein (Pcomp) assignments (Supplemental Table 6). The protein redundancy issue caused by multiple entries in the database for mRNA/cDNA/partial coding sequences or biologically conserved domains is resolved by grouping multiple protein entries and biological isoforms into a single entry (Supplemental Tables 2 and 6). The use of PeptideProphet and ProteinProphet tools essentially served as BLAST searches because each of the identified peptides is used for comparison against all the sequence entries in the human protein database.
Automated quantification was achieved using the XPRESS software tool (30). The human protein database (126,167 entries) from the Na-tional Center for Biotechnology Information (NCBI) was used for the SEQUEST searches. Uninterpreted MS/MS spectra were searched requiring tryptic cleavage sites and allowing one missed cleavage site. The search parameters also included peptide mass tolerance of 1.0 with differential modification of ϩ 6, ϩ 8, and ϩ 16 for heavy leucine, heavy lysine, and oxidized methionine residues, respectively. Data were further filtered with commonly accepted stringent criteria: cross-correlation (Xcorr) of 1.9, 2.2, and 3.7 for 1ϩ, 2ϩ, and 3ϩ charge state peptides, respectively, and delta correlation (⌬Cn) score greater than or equal to 0.1 (40). To remove false positive identifications, we excluded single peptide identification including different charge states and different modification. In addition, most of the peptide identifications were further filtered using the size of the peptide (must be greater than six amino acids), manual inspection of five or more consecutive b and y ions, molecular weight filtering using the excised gel regions, and parent ion information (Supplemental Table 2). Moreover for the estimation of false positive rates, we searched against a concatenated forward and reversed human protein database as outlined previously by Gygi's group (41). The protein list we are reporting here generated a Յ0.7% false positive rate of protein identification.
Extracting Quantitative Chromatogram from Mass Spectra-Using the scan number of the identified peptides from the MS/MS file, the XPRESS software isolates the [ 13 C 6 ]Leu and [ 13 C 6 , 15 N 2 ]Lys heavy isotope peptide elution profiles, determines the area of each peptide peak, and calculates the abundance ratio based on these areas in an automated fashion. The parent ion mass over charge (m/z) ratios with 6 amu (one leucine residue), 8 amu (one lysine residue), or any other combination of leucines and/or lysines on peptides were extracted in an automated fashion, and the areas were quantified. For each protein quantification data at least one ratio was confirmed by manual validation and correction. Standard deviations are given in Supplemental Table 3. The procedure for manual validation and correction is outlined below. First, we examined the quantification window available in XPRESS, which shows the area under the curves of the heavy and light peptides and their ratios. Second, we manually adjusted the scan range to include the entire peptide elution chromatogram (XPRESS has a scan range window for adjustment). Finally we "updated" the quantification by changing the values to the scan rangeadjusted values. The source codes for XPRESS modifications that allow quantification of up to three amino acid residues are now freely available for download through the Sashimi Sourceforge site (sashimi.sourceforge.net/software_pq.html #XPRESS).
Bioinformatic Analysis of Mass Spectrometry Datasets-The IN-TERACT differential (IADIFF) tool was used to compare identified proteins within multiple INTERACT files (42). This allowed for determination of the overlap among the three nuclear fractions. For functional and subcellular categorization, we utilized a software tool termed PROTEOME-3D. This tool is a previously described data exploration and knowledge discovery software tool developed in our laboratory (43). Briefly PROTEOME-3D utilizes the identified protein list as an input and creates a queryable annotated database of identified proteins from published literature. It also provides graphical tools for displaying proteome landscapes, proteome comparison among experiments, and subfractions. Furthermore this tool provides access to locally stored protein annotations through a query-building tool for systematic data analysis. For functional classification the user may choose to categorize proteins based on Gene Ontology (GO) terms, keywords, definitions, comments, or any combination of those fields by adding the desired fields to the query. Programs such as PROTEOME-3D and Linux scripts used in this study will be made available upon request.
Theoretical pI, Molecular Weight, and Hydrophobicity-Sequences of the proteins identified from high salt, mRIPA, and SDS extracts were processed using custom Linux shell scripts and Cϩϩ programs to calculate theoretical pI, molecular weight, and average hydrophobicity. The following pK a values were used to calculate the theoretical pI of the identified proteins: N terminus, 8.0; Lys, 10.0; Arg, 12.0; His, 6.5; C terminus, 3.1; Asp, 4.4; Glu, 4.4; Cys, 8.5; and Tyr, 10.0. A normalized consensus amino acid hydrophobicity scale was used to calculate the average hydrophobicity of the identified proteins (44). To calculate the pI, molecular weight, and hydrophobicity values of the whole human proteome, we used the International Protein Index (IPI) database (Version HUMAN v3.05 FASTA) from the European Bioinformatics Institute that contained ϳ49,000 protein entries.
Prediction of Nuclear Localization-Nuclear localization prediction was carried out using three tools: PredictNLS (45), NucPred (46), and PsortII (47). Source codes and/or executable programs were obtained from the authors of the above bioinformatics tools and run on a Linux computer. Custom Linux shell scripts were written to automatically analyze the sequences of 600 unannotated/uncharacterized proteins that we identified in the nuclear extracts and to format the outputs from the nuclear localization prediction tools. Criteria used for nuclear localization prediction are described in Supplemental Table 5.
Antibodies and Western Blotting-For Western blotting, mouse monoclonal anti-cytochrome c, anti-Ran, anti-HSP-90, anti-PARP, anti-PCD8/AIF, and anti-Vimentin (BD Biosciences); anti-CDC2 (Santa Cruz Biotechnology, Santa Cruz, CA); and anti-lactate dehydrogenase (LDH), anti-␤-Actin, and anti-␣-Tubulin (Sigma) were used. Anti-Caspase 8 antibody was purchased from Cell Signaling Technology Inc. (Beverly, MA); anti-Acinus antibody, anti-heterochromatin protein 1 homolog ␣ (HP1␣; or CBX5_HUMAN protein), and antihuman Fas IgM antibody were purchased form Upstate Biotechnology (Lake Placid, NY); and anti-Ku 70 was purchased from Neo-Markers (Freemont, CA). The purified fractions were separated by SDS-PAGE and transferred to nitrocellulose membranes by electroblotting. Nonspecific binding sites were blocked by incubation in PBS containing 5% nonfat dry milk and 0.05% Tween 20. The membranes were then incubated for 2 h at room temperature or overnight at 4°C with the indicated antibodies. Membranes were washed with phosphate-buffered saline containing 0.05% Tween 20 (PBST) and incubated with horseradish peroxidase-conjugated secondary antibody (Bio-Rad). After washing the membrane with PBST, signals were visualized with an enhanced chemiluminescence system (PerkinElmer Life Sciences). Densitometric quantification was performed using ImageQuant software (Version 1.2; Amersham Biosciences) to analyze the results of Western blotting analysis.
Immunostaining and Confocal Microscopy-The mitochondria and nuclei of live cells were stained with MitoTracker Red (Invitrogen) and Syto-13 (Invitrogen), respectively. Cells were washed with PBS, fixed with 4% paraformaldehyde for 15 min, and permeabilized with PBS containing 0.2% Triton X-100 (Sigma) or fixed and permeabilized with ice-cold 70% ethanol. Cells were then incubated overnight with the indicated primary antibody in 0.1% normal goat serum in PBS, washed with 0.05% Tween 20 in PBS, and visualized with Alexa 488 goat anti-mouse or anti-rabbit IgG, Alexa 546 goat anti-mouse or anti-rabbit IgG, and TOPRO-3 for nuclear staining (Invitrogen). The images were collected by a Zeiss inverted laser scanning confocal microscope, LSM-510 or LSM-510 META (Zeiss, Thornwood, NY), with 63 ϫ 1.25 or 63 ϫ 1.4 oil immersion objective lens. An excitation wavelength of 488 nm with a band pass 500 -530-nm emission filter was used to detect Alexa 488 IgG and Syto-13, a wavelength of 543 nm with a long pass 565-615-nm emission filter was used to detect Alexa 546 IgG and MitoTracker Red, and a wavelength of 647 nm with a band pass 650 -670-nm emission filter was used to detect TO-PRO-3. Images were analyzed with the LSM-510 Image Browser software, and 3D image reconstruction was processed with IMARIS software (Bitplane AG, Zurich, Switzerland).

Stable Isotope Labeling of Jurkat Cells and Subcellular
Fractionation-The SILAC method was introduced in 2002 for unbiased quantitative proteome analysis in cell culture using deuterated leucine (26). However, using a single heavy iso-tope-labeled amino acid such as a leucine residue alone does not support quantification of tryptic peptides that lack leucine residues. To improve labeling coverage, we used leucine and lysine heavy isotope amino acids in cell culture media (Fig.  1A). The control and apoptotic cells were labeled with heavy FIG. 1. A schematic procedure for SILAC and differential extraction of nuclear proteins during apoptosis. A, the control cells and apoptotic cells were grown in either the heavy leucine ( 13 C 6 ) and lysine ( 13 C 6 15 N 2 )-or light leucine ( 12 C 6 ) and lysine ( 12 C 6 14 N 2 )-containing medium. Nuclei were fractionated by sucrose gradient centrifugation, and proteins were differentially extracted as indicated. Differentially extracted proteins were then mixed 1:1 and separated by 1D SDS-PAGE, and protein identification was performed as described under "Experimental Procedures." B, assessment of purity of nuclear fractionation. The purity of isolated nuclei was shown using trypan blue staining and light microscopy. Magnification, ϫ20. C, SDS-PAGE analysis of differentially extracted nuclear proteins as visualized by Coomassie Blue staining. Note the clear differences in the distribution of proteins from differentially extracted fractions. D, validation of the purity of nuclear proteins by Western blotting. Total cell lysate (total), cytosolic fraction (cyto), and differentially extracted nuclear fractions (H. Salt, mRIPA, and SDS) were used for the detection of three nuclear proteins (PARP, PCD8/AIF, and HP1␣), two abundant cytosolic proteins (LDH and Caspase 8), and a protein distributed both in the nuclear and cytosolic fraction (Ku 70). Note the lack of detectable contamination of PARP in the cytosol and LDH in the nuclear fractions.
isotopes ([ 13 C 6 ]Leu and [ 13 C 6 , 15 N 2 ]Lys) and light isotopes ([ 12 C 6 ]Leu and [ 12 C 6 , 14 N 2 ]Lys), respectively, for at least five passages. The nuclei were fractionated by sucrose gradient as described under "Experimental Procedures." As shown in Fig. 1B, light microscopic characterization of purified nuclei revealed that homogenous intact nuclei were obtained by this method, and this method also effectively removed the unlysed Jurkat cells. To reduce the complexity of the nuclear fractions, we sequentially extracted the nuclear proteins by their solubility: first using H. Salt followed by mRIPA and then the final extraction by harsh SDS buffer (Fig. 1, A and C). As shown in Fig. 1C, clear differences in protein bands were seen among the three differential extractions when the SDS-PAGE gel was visualized by Coomassie staining. We measured the protein concentrations of differentially extracted nuclear fractions and found that the high salt buffer extracted ϳ43%, the mRIPA buffer extracted ϳ44%, and the SDS buffer extracted the remaining ϳ13% of the total nuclear proteins.
We next tested the efficiency of apoptotic signal transduction and the quality of our fractionation by Western blotting using antibodies specific to nuclear fraction, cytosolic fraction, and a common protein that is shared between the nuclear and cytosolic fractions. We chose to test a protein called PARP for the nuclear fraction because PARP is a nuclear protein that is cleaved during apoptosis. In addition, we tested the presence of PCD8/AIF and HP1␣ because these proteins are known to be localized in the nucleus. For the marker of the cytosolic fraction, we utilized antibodies against LDH and Caspase 8. We also used anti-Ku 70 antibody because it is known that Ku 70 is present both in the cytosol and in the nucleus. As shown in Fig. 1D, PARP protein was found mainly in the high salt fraction, PCD8/AIF was found in the mRIPA fraction, and HP1␣ was found in the SDS fraction. In contrast, the LDH and Caspase 8 were restricted mainly to the cytosolic fraction. As anticipated, Ku 70 was found in both the nuclear and the cytosolic fractions. Interestingly differential fractionation was found to be remarkably specific as a large majority of PARP, PCD8/AIF, and HP1␣ proteins were preferentially extracted by high salt, mRIPA, and SDS buffers, respectively. In addition, we found that the PARP protein was cleaved when anti-Fas IgM antibody was used to induce apoptosis in these cells. These results suggest that the apoptotic signal was successfully transmitted and that the nuclear protein preparation and differential extraction procedure that we used is efficient in isolating classes of proteins with different solubility.
To validate stable isotope labeling and this quantification methodology, we mixed light and heavy labeled samples from untreated cells in five increasing ratios: 1:1, 2:1, 4:1, 8:1, and 16:1 (Fig. 2, A and B). Multiple peptides from HSP-90␣ were used for the quantification, and the number of peptides for each of the mixing ratios is indicated (Fig. 2A). Plotting the expected ratios versus observed quantification generated from the extracted ion chromatograms for the HSP-90␣ re-vealed a strong linear correlation between observed and expected results with R 2 ϭ 0.9928 (Fig. 2A). These results further validate the methodologies for nuclear fractionation and SILAC analyses. Representative extracted ion chromatograms of HSP-90␣ are shown in Fig. 2B.
Identification and Quantification of Apoptotic Nuclear Proteins-We next performed the SILAC analysis of differentially extracted nuclear proteins from control and apoptotic cells. We mixed 50 g of heavy amino acid-labeled proteins with the same quantity of light amino acid-labeled proteins, separated the combined mixtures on a 4 -12% NuPAGE gradient gel, excised the gel bands, performed in-gel trypsin digestion, extracted the tryptic peptides, and analyzed the peptides by -capillary liquid chromatography-tandem mass spectrometry. We analyzed 22 gel bands from the high salt fraction, 27 from the mRIPA fraction, and 20 from the SDS fraction using a Finnigan LTQ ion trap mass spectrometer (Fig. 2C) (see "Experimental Procedures").
A total of 780,530 MS/MS spectra were generated from the three nuclear fractions. The spectra were subjected to database analysis by the SEQUEST algorithm using the NCBI human proteome database, which contained 126,167 sequence entries (37). The SEQUEST-matched peptides were then filtered with a set of stringent scoring criteria commonly accepted in the literature: Xcorr of 1.9, 2.2, and 3.7 or higher for 1ϩ, 2ϩ, and 3ϩ charge state peptides, respectively, and ⌬Cn score greater than or equal to 0.1. After filtering with these criteria, we excluded single peptide and non-applicable quantitative proteins, those with peptides containing a mixture of both heavy and light amino acids, resulting in high confident identification of 1,174 unique proteins from three nuclear fractions with ϳ0.4 Ϯ 0.3% false positive rates (Table  I and Supplemental Fig. 1B). The complete list of identified nuclear proteins, including entry names, fractions where they were identified, protein description, molecular weight, pI information, and peptide count numbers within each fraction are shown in Supplemental Table 1. The detailed gel slice information, scan numbers, charge state, experimentally determined peptide molecular weight and deviation from the predicted peptide mass, SEQUEST scores, peptide probability, duplicated number of proteins of each peptide sequence, and peptide sequence information of 16,548 peptides are shown in Supplemental Table 2. To address the issue of redundancy in the list of identified proteins, we utilized the ProteinProphet software tool, which groups the redundant proteins in a single entry based on all of the identified peptides (Supplemental Table 6). However, peptide quantifications resulting from peptides shared among multiple proteins (splice isoforms, protein families, etc.) are problematic and cannot be easily resolved.
We examined the efficiency of protein identification and quantification using SILAC, 1D gel separation, and the -capillary LC-MS/MS procedure. Because we identified PARP by Western analysis and demonstrated that PARP protein was cleaved during apoptosis, we wanted to know the efficiency of peptide coverage as well as the abundance changes of PARP. PARP has been shown to be cleaved from p116 to p85 and p25 by a number of cysteine proteases or caspases during apoptosis, and the proapoptotic activity of PARP has

FIG. 2. Validation of SILAC-based quantification.
A, the light and heavy nuclear proteins were mixed as indicated (1:1, 2:1, 4:1, 8:1, and 16:1) and separated by SDS-PAGE, protein bands at 90 kDa were excised, and multiple peptides from the identified HSP-90␣ were used to validate the quantification. The observed (measured ratio) versus expected (mixing ratio) values are shown. The averages of observed ratios for each of the mixing ratios are listed. B, representative extracted ion chromatograms of HSP-90␣ for each of the mixing and measured ratios are shown. C, the SDS-PAGE gel of H. Salt, mRIPA, and SDS fractions were excised as indicated and subjected to in-gel trypsin digestion followed by -capillary LC-MS/MS procedure. been shown to deplete ATP in cells (23). Therefore, we examined the protein coverage and quantification of PARP and found that 34% of the peptides from PARP were identified (Fig. 3A). An example of a representative MS/MS spectrum of a tryptic peptide from PARP is shown in Fig. 3B. In brief, all of the identified peptides were with stringent SEQUEST scores, and 34% peptide coverage from PARP protein even though we were analyzing a complex nuclear protein mixture indicated to us that our methodology is efficient for protein identification. Quantification of PARP protein from the abundance ratios of heavy (control PARP peptide) to light (PARP peptide from 3.5 h of apoptosis) revealed an ϳ8-fold reduction during apoptosis (Fig. 3C). Thus, these 1D GeLC-MS/MS results confirm the results of the Western analysis (Fig. 1D).
We next examined the robustness of proteome coverage using the SILAC technology. Toward this goal, we examined labeling efficiency with double labeling using [ 13 C 6 ]leucine and [ 13 C 6 , 15 N 2 ]lysine. The summary of this evaluation is shown in Supplemental Fig. 1A. Briefly we found that double labeling coverage comprises about 89% of the peptides and 94% of the multiple-peptide-hit proteins. The quantification also was improved over deuterium-labeled L-leucine (Leu-d 3 ) because [ 13 C 6 ]leucine-and [ 13 C 6 , 15 N 2 ]lysine-labeled peptides co-elute with the native peptides.
We next classified common and unique proteins from a total of 1,174 nuclear proteins that were distributed in three fractions using a software tool termed IADIFF (INTERACT differential) (Fig. 4A). 125 proteins were common in all three fractions; the number of unique proteins identified specifically in high salt, mRIPA, and SDS fractions was 271, 469, and 75, respectively (Fig. 4A). In addition, some proteins were distributed in two fractions. For example, the numbers of common proteins between high salt and mRIPA, high salt and SDS, and mRIPA and SDS were 174, 25, and 35, respectively.
We then compared the overall quantification of all the identified nuclear proteins and found that the trend showed an approximately Gaussian distribution but with higher than expected numbers of up-regulated nuclear proteins during apo- ptosis (Fig. 4B). We considered over 2-fold changes as a significant regulation based on our experimental observations that 1) the percentage of variation between mixed and measured ratios ( Fig. 2A) is Յ25%, 2) the percentage of variation between SILAC ratios and Western blotting (see Fig. 6A) is Յ28.4%, and 3) the average of computed relative standard deviations of abundance ratios of all quantifiable proteins is 20% (Supplemental Fig. 1C). The maximum fold change possible by chance using these criteria is ϳ1.8. Therefore we consider Ն2-fold change as significant.
Interestingly 96.3% of the quantified proteins were below 2-fold ratio changes. Of the rest, 15 proteins (0.9%) were highly down-regulated, and 50 proteins (2.9%) were highly up-regulated. The protein list and ratios of 1,174 quantified proteins from all three nuclear fractions are shown in Supplemental Table 3.
Physicochemical Characteristics of Differentially Extracted Nuclear Proteome-We compared the physicochemical characteristics of nuclear proteins that were extracted by high salt, mRIPA, and SDS buffers. Our intent was to classify nuclear proteins based on their experimentally observed solubility and compare with the theoretical predictions of physicochemical characteristics based on their amino acid composition. As shown in Fig. 4C, we found that the overall trend in protein hydrophobicity was quite similar between the three fractions and the IPI human database. However, SDS buffer preferentially extracted the hydrophilic proteins, whereas mRIPA buffer more efficiently extracted hydrophobic proteins. The frequency distribution of molecular weight showed no discernable trend among the three fractions (Fig. 4D). Approximately 40% of the identified nuclear proteins were distributed between 10 and 40 kDa molecular mass ranges. We next compared the predicted pI values of the proteins identified from the three fractions (Fig. 4E). Proteins with lower pI were extracted in each of the three fractions with comparable efficiency. However, during sequential extraction, both mRIPA and high salt buffers failed to efficiently extract proteins with alkaline pI that were efficiently extracted by SDS in the last stage. The overall physical property of the nuclear proteome compared with the human proteome revealed subtle differences in hydrophobicity and pI (Fig. 4, C-E). For example, the high salt fraction contained higher percentages of hydrophilic proteins, and the SDS fraction showed higher percentages of basic proteins when compared with the whole human proteome.
Functional and Subcellular Characterization of Identified Nuclear Proteins Using PROTEOME-3D Software Tool-To gain functional insights into the nuclear proteome, we utilized a previously described software tool that allows automated data retrieval and in depth analysis of identified proteins. This tool, termed PROTEOME-3D, allows the user to categorize the identified proteins into distinct functional or compartmental groups (43). Organizing the identified 1,174 proteins into 15 functional groups reveals differences in extractability of pro-teins belonging to different functional classes within each nuclear fraction (Fig. 5A and Supplemental Table 1). For example, among all the proteins identified from this experiment using three different buffers, 21% of the functionally classified proteins are part of the transcriptional component of the cells (Fig. 5A, bottom panel, category G). Even though a similar 16 -21% of the extracted proteins make up the transcriptional components from each of the three fractions, high salt, mRIPA, and SDS buffers extracted 18, 15, and six unique proteins belonging to this component, respectively. In contrast to the proteins that are involved in the transcriptional machinery, the ribosomal proteins (category M) and helicase proteins (category J) were more efficiently extracted by the SDS buffer, indicating that these proteins are either membrane-bound or subcompartmentalized in the nucleus (Fig.  5A, third panel). Although these proteins are hydrophilic, the large and insoluble protein complex in the subnuclear compartment was not extracted by high salt and mRIPA buffers efficiently. A plausible explanation for this finding is that these large, insoluble protein complexes (e.g. nucleolus) are not efficiently solubilized in the high salt or mRIPA buffers due to their physiochemical properties of the subnuclear compartments. In contrast, hydrophobic and insoluble vesicle proteins (category O) were sufficiently extracted by high salt and mRIPA buffers (Fig. 5A). Thus, apparent differential solubility of nuclear proteins can be characterized by sequential extraction of proteins using buffer conditions that progressively extract proteins based on their physicochemical properties in the nucleus.
We next examined the localization of previously characterized proteins that were extracted by the three buffer conditions. As shown in Fig. 5B, a large majority of proteins extracted by high salt (64%) and SDS (79%) buffers were found to be known nuclear and nucleolar proteins. In contrast, only 42% of the proteins that were extracted by mRIPA buffer were known to be nuclear or nucleolar proteins, and a significant number of mitochondrial proteins (24%), endoplasmic reticulum proteins (12%), and Golgi proteins (3%) were also extracted. In all three fractions, significant percentages of cytosolic proteins, which range from 19 to 30%, were also present (Fig. 5B). These results likely indicate the classes of proteins that are distributed in the nucleus of human T leukemia cells and implicate close association of mitochondria and nucleus in these cells.
Validation of Protein Identification and Quantification-We next attempted to rigorously validate the identification and SILAC quantification results. Toward this end, we used antibodies against seven proteins that we identified, and we quantified and performed Western analyses using high salt, mRIPA, and SDS buffer-extracted nuclear fractions. Our goal was to compare the identification and quantification results from the SILAC experiment with the results from the Western analyses, essentially comparing seven proteins in all three nuclear fractions.
We first compared three proteins that are involved in apoptosis signaling. Quantification of PCD8/AIF, cytochrome c, and Acinus was consistent between the tandem mass spec-trometry and Western analyses (Fig. 6A) as the fractions that were identified in the mass spectrometry showed detectable levels of proteins by Western analysis. Quantification results FIG. 5. Plotting functional categorization and subcellular localization of identified nuclear proteins using the PROTEOME-3D software tool. A, functional characterization of identified nuclear proteins based on the published literature. The functionally known proteins identified from three fractions (268 in high salt, 240 in mRIPA, and 137 in SDS fraction) and a combined total of 356 proteins were plotted into 15 known functional groups. Categories are as follows: A, cell cycle: the series of events involving the growth, replication, and division of a eukaryotic cell; B, kinase: any of various enzymes that catalyze the transfer of a phosphate group from a donor, such as ADP or ATP, to an acceptor; C, phosphatase: any of numerous enzymes that catalyze the hydrolysis of esters of phosphoric acid and are important in the absorption and metabolism of carbohydrates, nucleotides, and phospholipids and in the calcification of bone; D, chromosome: a threadlike linear strand of DNA and associated proteins in the nucleus of eukaryotic cells that carries the genes and functions in the transmission of hereditary information; E, DNA repair: the processes that minimize cell killing, mutations, replication errors, persistence of DNA damage, and genomic instability; F, replication: the process by which genetic material, a single celled organism, or a virus reproduces or makes a copy of itself; G, transcription: the process by which messenger RNA is synthesized from a DNA template resulting in the transfer of genetic information from the DNA molecule to the messenger RNA; H, translation: the process by which messenger RNA directs the amino acid sequence of a growing polypeptide during protein synthesis; I, polymerase: any of various enzymes, such as DNA polymerase, RNA polymerase, or reverse transcriptase, that catalyze the formation of polynucleotides of DNA or RNA using an existing strand of DNA or RNA as a template; J, helicase: any of various enzymes that catalyze the unwinding and separation of double-stranded DNA or RNA during its replication; K, splicing: the removal of introns and the joining of exons from mRNA precursors; L, chaperone: cytoplasmic proteins of both prokaryotes and eukaryotes that bind to nascent unfolded polypeptides and ensure correct folding or transport; M, ribosomal: a minute round cytoplasmic particle composed of RNA and protein that is the site of protein synthesis as directed by mRNA; N, structure: the arrangement or formation of the tissues, organs, or other parts of an organism; O, vesicle: a membranous and usually fluid-filled pouch (as a cyst, vacuole, or cell) in a plant or animal. B, subcellular locations of putative nuclear proteins. The number of localization-annotated proteins of high salt, mRIPA, and SDS fractions are 314 (ϳ53%), 458 (ϳ57%), and 164 (ϳ63%), respectively. Among the annotated proteins, known nuclear and nucleolar localized proteins of high salt, mRIPA, and SDS fractions are 64, 42, and 79%, respectively.
were comparable with variation of up to 28.4% between SILAC and Western blot-based quantification (Fig. 6A). Similar results with variability of less than 25% between two measurements were found when we tested for four additional proteins: Ran, CDC2, ␤-Actin, and ␣-Tubulin (Fig. 6A). Immunofluorescence experiments using mouse anti-CBX5/HP1␣, anti-HSP-90, anti-proliferating cell nuclear antigen, and anti-Lamin A/C antibodies and rabbit anti-Acinus, anti-cytochrome c, and anti-BAK (Bcl-2 homologous antagonist/killer) antibodies also confirmed the identification and quantification in the nucleus during apoptosis (Supplemental Fig. 5). These results validate mass spectrometry-based identification and SILACbased quantification.
Highly Up-or Down-regulated Nuclear Proteins during Apoptosis-To gain insights into apoptotic signaling in the nucleus, we examined proteins that were highly up-or downregulated during the apoptotic signal transduction. Among the 1,174 identified proteins, we found 59 proteins that were regulated over 2-fold during the apoptotic signaling event in the nucleus (Supplemental Table 4). Among these proteins are a number of important nuclear proteins such as histone H4, DNA replication protein RFC1, and NUP43 (NU43_HUMAN), a FIG. 6. Confirmation of SILAC-based quantification and close topographic association of mitochondria and nucleus. A, Western blotting of nuclear fractions using antibodies against a number of proteins that were identified and quantified from three different nuclear fractions. Comparison of Western blot-based semiquantification values and SILAC-based quantification values and the percentage of variability are shown. Note that for most cases, the variability is less than ϳ28%. The percentage of variation between Western blotting and SILAC is calculated by subtracting SILAC value from Western value, dividing the remainder by Western quantification value, and then multiplying the result by 100. B, immunofluorescence microscopy of fibrillarin (green), mitochondria (MitoTracker, red), and nucleus (Syto-13, blue) shows a close topological association of the mitochondria and nuclear invaginations (arrow) during apoptosis. The nucleolar marker fibrillarin is fragmented and/or aggregated during apoptosis. 3D reconstructions of blue (Syto-13) and red (mitochondria) channels for the control and apoptotic cells are shown in the right panel. Cross-section of the 3D reconstruction reveals significant numbers of mitochondria in the nuclear invagination. West., Western; Cyto, cytosolic fraction. bidirectional transport protein that participates in the transport of macromolecules between the cytoplasm and nucleus. Interestingly among many protein families identified, only specific proteins were found to be up-or down-regulated during apoptosis (Supplemental Table 4). These results suggest that chromatin condensation and DNA fragmentation during apoptosis may be regulated by many classes of proteins with diverse functions.
Close Topographic Association of Mitochondria and Nucleus during Apoptosis-Subcellular location analysis revealed that 24% of the identified proteins are known mitochondrial proteins, and 12% are known ER proteins (Fig. 5B). In addition, selecting proteins that are from control cells (heavy isotope label) versus apoptotic cells (light isotope label) revealed that 58 additional mitochondrial proteins were found in apoptotic nuclear fractions. These results suggest functional association between the nuclei and mitochondria and changes in association during apoptosis. Thus we performed immunofluorescence experiments for nuclei and mitochondria and identified proteins that are known to be localized in these organelles (Fig. 6B).
Careful analysis of nuclear morphology in control and apoptotic cells by fluorescence confocal microscopy revealed that nuclei in apoptotic cells, but less so in control cells, possess channels/invaginations that contain large numbers of mitochondria (Fig. 6B, white arrow). We also found that the invaginations of nuclei during apoptosis were deeper and wider than the interphase nuclei of control HeLa cells (Supplemental Fig. 3). Furthermore these nuclear channels showed close interaction and association with mitochondria during apoptosis ( Fig. 6B and Supplemental Fig. 3). Threedimensional reconstructions using optical Z-sections followed by vertical plane visualization revealed large numbers of mitochondria in apoptotic nuclear invaginations (Fig. 6B). To investigate how the invaginations occur during apoptosis, we observed live cells with differential interference contrast and fluorescence images of nuclei, mitochondria, and cell morphology. We found that an increase in nuclear invaginations occurred at the time of a generalized cell contraction during early apoptosis (Supplemental Fig. 4, A and B). We subsequently examined the distribution of Vimentin during apoptosis because this protein has been involved in cell contraction. As shown in Supplemental Fig. 4C, we found that Vimentin is present in nuclear invaginations (arrows), suggesting a functional association of contractile proteins with dynamic changes in nuclear channels. Such invaginations will increase the total contact area of mitochondria with the nucleus and may have a function in enhancing direct translocation of apoptotic mitochondrial proteins into the nucleus.
To further validate these results, we tested the localization of a known mitochondrial protein, BAK, and its changes in localization during apoptosis. From our SILAC analysis, BAK was identified by a single peptide and quantified as 2.5-fold up-regulated in the mRIPA fraction based on quantification from both light and heavy isotopes (Fig. 7, A and B). To validate this quantification in light of the close interaction of nucleus and mitochondria during apoptosis, we performed Western blotting and found a similar result (Fig. 7C). We next examined the localization of BAK in the mitochondria using MitoTracker and rabbit anti-BAK antibody (Fig. 7D) and found that BAK protein is also seen in the nuclear invaginations (Supplemental Fig. 5G), and it strongly co-localized with nuclear DNA and speckle marker SC-35 in later stages of apoptosis (Fig. 7E). These results further confirm our SILAC quantification and validate the close interaction and association of nucleus with mitochondria during apoptosis.

DISCUSSION
Characterization of Nuclear Proteome-The introduction of stable isotope labels in protein quantification studies allows detection of dynamically changing proteomes in multiple biological systems. The first introduction of ICAT technology for large scale proteomic profiling by Aebersold's group (48) in 1999 revolutionized the field. Multiple investigators now use the technology to identify biologically important proteins from complex mixtures. Introduction of the SILAC methodology by the use of leucine-d 3 in cell culture system in 2002 by Mann's group (26) is based on the same principle as the ICAT method. SILAC has more recently been modified to increase the coverage of proteins by using the amino acids lysine and arginine (49). Although arginine labeling has been successfully used by at least two groups, incorporation of arginine into proline residues complicates the quantification (32,34). To characterize nuclear proteome, we used a [ 13 C 6 ]leucine and [ 13 C 6 , 15 N 2 ]lysine combination to allow increased coverage while circumventing the complications associated with arginine.
In this study, we attempted to characterize nuclear proteins based on their extractability under three different buffer conditions. Our goal was to develop a methodology where most of the extractable nuclear proteins can be isolated, sequentially identified, and quantified by the use of stable isotope labeling and tandem mass spectrometry. We used three different buffer conditions: 1) high salt buffer for soluble nuclear proteins, 2) mRIPA buffer for hydrophobic proteins and nuclear membrane-associated proteins, and 3) SDS buffer for nuclear proteins that are likely to be intimately associated with the nuclear structural components. From our analyses, using a set of stringent filtering criteria on SEQUEST scores and removing the single hit peptides, we identified 1,174 multiple peptide-containing proteins.
Among all of the identified proteins, 574 proteins were assigned to nuclear, nucleolar, cytoplasm, mitochondria, ER, and Golgi subcellular location; however, the remaining 600 proteins were not annotated for subcellular localization. Hence we utilized three software tools to predict the nuclear localization of these unannotated proteins: PredictNLS, Nu-cPred, and PsortII (45)(46)(47). We found that 218 additional proteins from this category were predicted to reside in the nucleus by at least one of the prediction tools (Supplemental Fig. 2A and Supplemental Table 5). However, based on the literature or prediction tools, we were not able to confirm the experimental observation of an additional 382 proteins in the nucleus.
A recent study that described the nucleolar proteome reported 700 proteins from HeLa cells (50). We thus compared our protein list with the reported nucleolar proteins and found 244 (21%) common nucleolar proteins (Supplemental Fig. 2B). Detailed comparison by the three differentially extracted protein sets revealed that the high salt fraction contained 225 (38% of the high salt proteins), the mRIPA fraction contained 131 (16%), and the SDS fraction contained 164 (63%) of the reported nucleolar proteins. These results indicate that the complexity of nucleolar proteins, similar to nuclear proteins, can be reduced by differential extraction using the three buffer conditions described.
Non-nuclear Proteins-A large number of proteins detected in the three nuclear preparations were assigned by literature search or by prediction tools to other subcellular compartments (Fig. 5B). A similar result was also obtained from MS analysis of the human HeLa cell nucleolar proteome (50). Although we cannot rule out contamination during our purification of nuclei, the presence of classes of proteins from other subcellular compartments in the nuclear preparations may be physiologically relevant. For example, the known physical associations of outer nuclear membrane with ER membrane and mitochondria with ER and shuttling cytoplasmic signaling proteins between the cytosol and the nucleus may result in the co-purification of ER, mitochondrial, and cytosolic proteins with the nuclei (51). Also in smooth muscle cell, it is known that intermediate filaments form linkages between the nuclear envelope and mitochondria (52). Additionally proteins such as the Ran-binding protein 2 (RBP2_HUMAN), heterogeneous nuclear ribonucleoprotein K (ROK_HUMAN), and GTP-binding nuclear shuttling protein RAN are known to be localized in both the cytosolic and nuclear compartments (53)(54)(55). In addition, PCD8/AIF (PCD8_HUMAN) is known to reside in the mitochondria and nucleus (24). Similarly nuclear envelope pore membrane protein POM 121 (P121_HUMAN) is also known to reside in the endoplasmic reticulum during metaphase (56). These examples suggest that a significant number of non-nuclear proteins may be found in the nucleus.
Proteomic and Topographic Changes in Nucleus during Apoptosis-Apoptotic DNA condensation and fragmentation have been used to detect apoptotic cells in vitro and in vivo. However, the proteins that control this process are not well understood. Although similar but reversible DNA condensation is seen during the cell cycle progression, it is not known whether the chromatin condensing factors that control cell cycle-specific DNA condensation also participate in apoptotic DNA condensation. From our quantitative proteomic experiment, we were able to identify and quantify a number of candidate proteins that are known to control DNA replication and chromatin remodeling/condensation in the cell cycle. For example, among the DNA replication silencing factors that we identified are MCM2, MCM3, MCM4, MCM6, and MCM7. However, we found that the MCM family proteins were not significantly regulated during apoptosis. The final conclusion on the role of MCM proteins in apoptosis control will require not only their identification and quantification but the relative activity changes of these proteins during apoptosis.
We have provided a large scale analysis of nuclear proteins during apoptosis and a number of new candidate proteins that may participate in apoptotic DNA condensation and fragmentation. Furthermore immunofluorescence studies of identified nuclear and mitochondrial proteins revealed the close physical association of mitochondria with nuclear channels during apoptosis. These findings provide new insights into the mechanism of proapoptotic mitochondrial protein translocation into the nucleus. It is known that during apoptosis, a number of mitochondrial proteins are translocated into the nucleus, but the exact mechanism of this translocation is unclear. For example, it is not known how cytochrome c and other proapoptotic proteins such as AIF and endonuclease G are translocated from the mitochondria into the nucleus (24,25). Our results suggest that intimate association and recruitment of mitochondria into the nuclear invaginations may be a mechanism that allows efficient transport of mitochondria proteins into the nucleus. We found that during apoptosis nuclear invaginations contained substantially more mitochondria as quantified by MitoTracker Red fluorescence (Supplemental Fig. 3). The close proximity of apoptotic nucleus and mitochondria suggests that the contact sites between two organelles may facilitate protein transport and subsequent breakdown of both organelles during apoptosis. Consistent with this hypothesis, in addition to a number of well known mitochondrial proteins such as cytochrome c, endonuclease G, and AIF that cause nuclear breakdown, nuclear protein p53 has been shown to participate in the mitochondrial permeability transition during apoptosis (57,58). Similarly we found that one of the known mitochondrial proapoptotic proteins, BAK, was up-regulated in the mRIPA fraction (Fig. 7, A-C). The co-localization of BAK with mitochondria in control cells and the strong co-localization of BAK and nuclear speckle marker SC-35 in apoptotic cells, shown in Fig. 7, D and E, also support the notion that import of proapoptotic mitochondrial proteins plays a role in the process of apoptotic nuclear breakdown. Future experiments are necessary to elucidate the functions of highly regulated mitochondrial nuclear proteins in apoptotic chromatin condensation and DNA fragmentation.