Dynamics and structure of the Bax-Bak complex responsible for releasing mitochondrial proteins during apoptosis Free

Programmed cell death or apoptosis is a controlled cell-suicide process that has great importance in maintaining the normal physiological function of a biological organism (Vaux and Korsmeyer, 1999). The malfunction of apoptosis is also closely related to many diseases (MacFarlane and Williams, 2004; Nijhawan et al., 2000). In the past few years, a large number of studies have been conducted with the aim of understanding the process of apoptosis at a cell and molecular level. It was found that mitochondria often play a central role in the apoptotic process (Desagher and Martinou, 2000; Hengartner, 2000; Newmeyer et al., 1994; Wang, 2001). Many apoptotic stimuli can trigger mitochondria to release intermembrane proteins [including cytochrome c, Smac/DIABLO, the serine protease Omi (official symbol HTRA2), AIF (also known as AIFM1) and EndoG (or NUCG)] that activate the downstream apoptotic pathways (Degterev et al., 2003; Du et al., 2000; Hegde et al., 2002; Li et al., 2001; Liu et al., 1996; Susin et al., 1999; Verhagen et al., 2000). For example, the release of cytochrome c (Cyt c) and Smac from mitochondria is known to play a critical role in triggering the activation of caspase-9 (CASP9) and caspase-3 (CASP3) (Du et al., 2000; Lim et al., 2002; Liu et al., 1996). When Cyt c is released into the cytosol, it can bind with Apaf-1 (APAF) to activate the caspase cascade (Cai et al., 1998; Liu et al., 1996; Zou et al., 1999). By contrast, the release of Smac can facilitate activation of caspases indirectly by neutralizing a set of caspase inhibitors, known as apoptosis inhibitor IAP (Du et al., 2000; Verhagen et al., 2000). Induction of apoptosis also leads to the loss of mitochondrial membrane potential (ΔΨ_m) (Zhou et al., 2005), which results in reducing the production of ATP.

Thus, there is a great interest in understanding the process that triggers the release of the mitochondrial intermembrane proteins. So far, two major mechanisms have been proposed: (1) nonspecific rupture of the mitochondrial outer membrane (MOM) because of a swelling of the mitochondrial matrix caused by the opening of PT pores (Desagher and Martinou, 2000; Marzo et al., 1998; Vander Heiden et al., 1999); (2) formation of specific channels or pores at the MOM by a family of Bax-like proteins (Antonsson et al., 1997; Basanez et al., 1999; Desagher and Martinou, 2000; Shimizu et al., 1999). Previously, we used the single live cell analysis technique to investigate the matrix-swelling model (i.e. hypothesis 1) (Gao et al., 2001). We found that the release of Cyt c occurred before mitochondrial swelling during UV-induced apoptosis in HeLa cells (Gao et al., 2001). Our results did not support the matrix-swelling model. This suggested that the release of mitochondrial intermembrane proteins is more likely to be due to a direct permeabilization of the MOM caused by the pro-apoptotic B-cell lymphoma-2 (Bcl-2, official symbol BCL2) family proteins (mainly Bax and Bak).

Bcl-2-associated X protein (Bax) and Bcl-2-antagonist killer (Bak) are the most important pro-apoptotic Bcl-2 family proteins (Antonsson, 2001; Mikhailov et al., 2003; Nechushtan et al., 2001). It has been shown that genes encoding both Bax and Bak must be knocked out in order to prevent apoptosis (Wei et al., 2001). Translocation of Bax from the cytosol to mitochondria was required for triggering apoptosis in many cell types (Antonsson et al., 2001; Deng et al., 2002; Finucane et al., 1999; Hockenbery et al., 1993). It appears that the Bax protein undergoes several changes during apoptosis, including translocation to mitochondria (Wolter et al., 1997), change in conformation (Nechushtan et al., 1999), dimerization or oligomerization (Annis et al., 2005; Antonsson et al., 2001; Gross et al., 1998) and membrane integration (Nakamura et al., 2000; Tan et al., 2006). Bak has a similar function, because it is required for the induction of apoptosis in response to a variety of death signals (Lindsten et al., 2000; Wei et al., 2001). However, unlike Bax, Bak normally resides in mitochondria; it can form oligomers or clusters at the MOM during apoptosis without the translocation processes (Mikhailov et al., 2003; Nechushtan et al., 2001).

At present, it is not clear how Bax or Bak can permeabilize the MOM. Based on the ability of Bax to form channels in synthetic lipid membranes (Antonsson et al., 1997; Saito et al., 2000) and other findings, several models have been proposed: (1) formation of channels comprised of Bax or Bak at the MOM (Antonsson et al., 1997; Saito et al., 2000); (2) formation of a novel pore by Bax in conjunction with the voltage-dependent anion channel (VDAC) (Shimizu et al., 1999); (3) formation of large lipid-protein complexes at the MOM comprised mainly of Bax-like proteins (Basanez et al., 1999; Desagher and Martinou, 2000; Kluck et al., 1999). In order to provide useful information to evaluate these proposed models, we used live cell imaging techniques to examine the dynamics changes of Bax and Bak protein distribution within a cell undergoing UV-induced apoptosis. Our objective was to answer the following questions: (1) what is the molecular process that enables Bax-Bak to permeabilize the MOM? (2) what is the structure of the `pore' formed by Bax-Bak at the MOM that is responsible for the release of mitochondrial intermembrane proteins?

It is known that the release of cytochrome c (Cyt c) from the mitochondria can trigger the activation of caspase-3. This was confirmed in our model system. Using UV as an apoptotic inducer, we observed that the release of Cyt c into the cytosol was clearly correlated with caspase-3 activation and the cleavage of PARP, a well known substrate of caspase-3 (Fig. 1A). However, in addition, results of our western blot analysis indicated that the release of Cyt c was also correlated with the timing of Bax translocation from the cytosol to mitochondria (Fig. 1A,B). Both events occurred around 2 hours after UV irradiation. The time resolution based on the western blot analysis, however, was not high enough to determine the temporal order between these two events. Furthermore, the apoptotic process could not be synchronized within a population of cells. Thus, in order to accurately determine the time courses of Bax translocation and Cyt c release, one needs to conduct single live cell analysis.

To conduct such a study, we labelled the Bax molecule with GFP by gene fusion and used live cell imaging to measure the dynamic redistribution of Bax during UV-induced apoptosis. To ensure that the dynamic redistribution of GFP-Bax faithfully reflects that of the endogenous Bax within the apoptotic cell, we introduced the GFP-Bax fusion gene into HeLa cells by electroporation and compared the translocation pattern of GFP-Bax with endogenous Bax using western blot analysis. Our results showed that the translocation pattern of GFP-Bax from the cytosol to mitochondria was indeed similar to that of the endogenous Bax (Fig. 1C,D).

Results of our imaging measurements of the dynamic redistribution of GFP-Bax in HeLa cells undergoing UV-induced apoptosis are shown in Fig. 2. Here, we also used Mitotracker Red CMXRos, a red fluorescent dye that specifically accumulates in the mitochondria (Minamikawa et al., 1999), to monitor the morphology of mitochondria in the same cells. At 1-2 hours following the UV treatment, we observed that the distribution of GFP-Bax changed into different patterns (Fig. 2A-D). These Bax distributions could be classified into four different stages: stage I, GFP-Bax was diffusely distributed in the cytosol (Fig. 2A); stage II, GFP-Bax showed a filamentous pattern that was co-localized with mitochondria (Fig. 2B); stage III, GFP-Bax formed small complexes in the MOM (Fig. 2C); stage IV, GFP-Bax formed large clusters, that were mostly (but not exclusively) localized to mitochondria.

The distribution patterns of stage I and stage IV appeared to be more stable and they were similar to those observed in earlier reports (Nechushtan et al., 2001; Wolter et al., 1997). The distribution patterns of stage II and stage III were transient in nature and have not been investigated in detail before. When we magnified the images in stage II (in the region marked by the box shown in Fig. 2B), we found that a significant portion of GFP-Bax proteins were distributed uniformly as a hollow structure in mitochondria (Fig. 2E). At this time, some GFP-Bax proteins remained in the cytosol (Fig. 2B). The observed hollow structure in mitochondria suggest that the GFP-Bax proteins were most likely distributed at the MOM, because the mitochondrial inner membrane (MIM) generally folds into the matrix, in which no GFP-Bax was detected in our experiment. When we magnified the images in stage III (the region boxed in Fig. 2C), we found that Bax molecules started to form complexes at the MOM, while a portion of Bax was still distributed uniformly in the membrane (Fig. 2F). When we magnified the images in stage IV (in the region marked by the box shown in Fig. 2D), we observed that all GFP-Bax proteins had aggregated into large clusters, mostly localized to the external periphery of the mitochondria, which, by now, had undergone fission and become swollen (Fig. 2D,G). There was no more Bax distributed in the cytosol. (Note that the diameter of mitochondria within one cell can vary from 0.5 to 1 μm. To show the distribution of GFP-Bax at the MOM more clearly, we selected to display mitochondria with larger diameters in Fig. 2.)

Before the activation of Bax, Bax was not totally excluded from the mitochondria; a small amount of Bax appeared to be loosely associated with mitochondria. The concentration of these mitochondria-associated Bax proteins, however, was relatively low and did not appear to be at a higher concentration than that in the cytosol (Fig. 2A). We could detect an increase of Bax concentration at the MOM (i.e. appearing in a filamentous distribution) only after Bax was activated during apoptosis (Fig. 2B).

The dynamic redistribution of Bax shown in Fig. 2 appeared to be common for most apoptotic pathways and was not limited to UV-induced apoptosis. For example, when HeLa cells were treated with other apoptotic inducers, such as co-treatment with TNFα and CHX (cycloheximide), we also observed four stages of GFP-Bax distribution (supplementary material Fig. S1A-D). Furthermore, the dynamic distribution pattern of Bax was independent of the cell type. We found that GFP-Bax also underwent a four-stage distribution in UV-treated PtK2 cells (supplementary material Fig. S1M-P) or U2OS cells (supplementary material Fig. S1Q-T).

Based on results of these imaging measurements, we noted that the distribution of Bax must undergo the following dynamic changes during the apoptotic process. (1) Initially, Bax was distributed diffusely in the cytosol. (2) In response to an apoptotic stress, GFP-Bax was translocated to mitochondria and distributed uniformly at the MOM. (3) Then, Bax started to form small complexes by concentrating the Bax molecules within the MOM into isolated spots. (4) Finally, the Bax small complexes expanded by recruiting additional Bax molecules from both the MOM and the cytosol to form large clusters.

To verify the temporal relationship among these four stages of Bax distribution (shown in Fig. 2) and to correlate these with the change of the mitochondrial function, we used a confocal microscope operated under the two-channel mode to conduct a time-dependent measurement on both the distribution of GFP-Bax and Mitotracker within the same live cell. A sample record of this time series measurement is shown in Fig. 3A-H, where two neighbouring cells undergoing UV-induced apoptosis were observed to initiate Bax translocation at slightly different time. Here, the two cells were numbered Cell 1 for the upper cell and Cell 2 for the lower cell. Initially, Bax was diffusely distributed in the cytosol. The distribution pattern of Mitotracker was the same as that of mitochondria in the normal cell, that is it appeared as a filamentous structure (see Cell 1 and Cell 2 at 85 minutes after UV treatment) (Fig. 3A). About 1 hour later, some GFP-Bax molecules started to translocate to mitochondria to show a partially filamentous structure (see Cell 1 at 159 minutes or Cell 2 at 170 minutes). In the minutes after this, Bax began to form small complexes in the mitochondria (see Cell 1 at 160 minutes and Cell 2 at 187 minutes). At that time, the morphology of mitochondria remained filamentous. However, the fluorescence intensity of Mitotracker started to decrease, indicating a decrease of the mitochondrial membrane potential (ΔΨ_m). Later, more and more GFP-Bax molecules were translocated into the Bax complexes to form large clusters (see Cell 1 at 185 minutes and Cell 2 at 193 minutes). Soon after that, the cells started to shrink, indicating that the cytoplasm had been dismantled by the fully activated caspases. The sequence of events shown in Fig. 3 is representative of results observed in 58 live HeLa cells in our study.

In order to determine the temporal relationship between Bax translocation and the release of Cyt c and Smac, we decided to use the change of ΔΨ_m as a reference event. Here, we used Mitotracker Red CMXRos to monitor the change of ΔΨ_m in live HeLa cells. The accumulation of Mitotracker Red CMXRos in the mitochondria is dependent upon membrane potential. Thus, the reduction of the fluorescent signal of Mitotracker Red CMXRos is a known indicator of the loss of ΔΨ_m (Macho et al., 1996; Pendergrass et al., 2004; Poot et al., 1996). Here, we conducted a quantitative analysis of the time-dependent measurement shown in Fig. 3A-H. To quantify the distribution of GFP-Bax and the change of ΔΨ_m, we calculated the s.d. of pixel intensities of GFP-Bax and the average intensity of Mitotracker, respectively, within the same cell using the Metamorph software. The calculated GFP-Bax translocation index (i.e. the s.d. of GFP-Bax distribution) and the normalized fluorescence intensity of Mitotracker were plotted over time in Fig. 3I for Cell 2. From these results, we found that the formation of Bax small complexes was temporally associated with the decrease of ΔΨ_m (Fig. 3I). Using a similar imaging measurement, we further showed that a decrease of ΔΨ_m was always observed immediately following the Cyt c-GFP releasing process (Fig. 3J). In fact, we have shown in a previous study that the release of Cyt c and Smac both coincide with the mitochondrial membrane potential depolarization during UV-induced apoptosis (Zhou et al., 2005). Based on these results and those shown in Fig. 3, we concluded that the three apoptotic events, including the formation of Bax small complexes at the MOM, the release of Cyt c and Smac from mitochondria and the reduction of ΔΨ_m, all occurred simultaneously (i.e. within minutes) during UV-induced apoptosis (see supplementary material Movies 1-3).

To further verify the temporal relationship between Bax aggregation and the release of Cyt c, HeLa cells transfected with GFP-Bax fusion gene were fixed at different time points following UV irradiation. The distribution of Cyt c was visualized by immunostaining cells with anti-Cyt-c antibody. A fluorescent microscope was used to record both the distribution patterns of GFP-Bax and endogenous Cyt c in the same cells. The results are shown in Fig. 4A-H and could be classified into the following categories: (1) GFP-Bax was diffusely distributed in the cytosol, whereas Cyt c was concentrated in the mitochondria (Fig. 4A,E); (2) GFP-Bax showed partially filamentous distribution in the mitochondria, while Cyt c was still concentrated in the mitochondria (Fig. 4B,F); (3) GFP-Bax was distributed in a filamentous form, but some GFP-Bax started to form very small complexes in the mitochondria. The majority of Cyt c was still in the mitochondria (see the right-hand cell in Fig. 4C,G); (4) GFP-Bax formed small complexes in the mitochondria, whereas Cyt c was distributed entirely in the cytosol (Fig. 4D,H); (5) GFP-Bax formed large clusters and Cyt c was distributed solely in the cytosol (data not shown). The observations of the Category 3 and 4 cells were most interesting; they support the fact that cells started to release Cyt c only after GFP-Bax formed small complexes at the MOM.

We have conducted a statistical analysis of the distribution of various categories in 2032 cells expressing GFP-Bax. The percentage of Category 3 cells was very small: only about 1.97%. This fact indicated that the time window between the formation of Bax small complexes and the release of Cyt c must be very short. This finding, together with the results of our live cell measurements, suggested that Cyt c release occurred almost immediately following the formation of Bax small complexes at the MOM during UV-induced apoptosis.

In the preceding experiments, we used mainly GFP-labelled Bax to study the dynamic redistribution of the Bax protein. Using an immunostaining method, we have verified that the endogenous Bax also underwent the same four stages of distribution during apoptosis (Fig. 4I-L). Furthermore, we showed that the release of Cyt c occurred only after the endogenous Bax formed small complexes in MOM (Fig. 4K,O). Thus, the dynamic redistribution observed in GFP-Bax appeared to be truly representative of the behavior of the endogenous Bax.

Smac/DIABLO, a 25 kDa mitochondrial protein, is released as a tetramer (100 kDa) from mitochondria into the cytosol during apoptosis to neutralize a set of caspase inhibitors, known as IAP (Du et al., 2000; Verhagen et al., 2000). In this study, we measured directly the dynamic redistribution of Bax and Smac in a live cell during apoptosis by cotransfecting CFP-Bax and Smac-YFP into HeLa cells (see supplementary material Movie 4). Fig. 5A-E showed a sample record of this time series measurement. Up to 147 minutes following UV treatment, Bax had not yet formed small complexes and Smac showed a very clear filamentous pattern in the mitochondria (Fig. 5A,B). Four minutes later (t=151 minutes), Bax began forming small complexes in the mitochondria, and at the same time, Smac was released from mitochondria into the cytosol (Fig. 5C,D). The release of Smac-YFP was reflected in an increase of fluorescence intensity of Smac-YFP in the cytosol as well as a decrease of Smac-YFP in the mitochondria (Fig. 5C,D). Note that unlike the release of Cyt c-GFP, which was an `all-or-nothing' event, the release of Smac-YFP from mitochondria was not 100% (Zhou et al., 2005). We also conducted a quantitative analysis of the time-dependent Bax translocation and Smac release by comparing the s.d. of pixel intensities of both CFP-Bax and Smac-YFP within the same cell (Fig. 5F). The results clearly showed that, during apoptosis, the release of Smac occurred at the same time as the formation of Bax small complex within the HeLa cell, just like the release of Cyt c.

Previously, it was reported that Bak can play a similar role to Bax in the mitochondria-dependent apoptotic pathway (Lindsten et al., 2000). However, unlike Bax, Bak normally resides in the mitochondria and thus does not require a translocation process. Although it was known that Bak can form oligomers or clusters in the mitochondria under certain apoptotic stimuli (Mikhailov et al., 2003; Nechushtan et al., 2001; Sundararajan et al., 2001), it is not clear whether Bak undergoes a dynamic reorganization similar to that of Bax observed in stage II to stage IV. Thus, we measured the dynamic distribution of YFP-Bak in HeLa cells undergoing UV-induced apoptosis. Results of this imaging study are shown in Fig. 6. Following the UV treatment, we observed three different patterns of Bak distribution: (1) YFP-Bak localized to MOM showing a filamentous pattern (Fig. 6A-D); (2) YFP-Bak forming small complexes at the MOM (Fig. 6E-H); (3) YFP-Bak forming large clusters in mitochondria (Fig. 6I-L). These three patterns of YFP-Bak distribution were very similar to that of GFP-Bax at stage II, stage III and stage IV.

We also investigated the time relationship between Bak distribution and the decrease of mitochondrial membrane potential. We expressed YFP-Bak in HeLa cells and stained them with Mitotracker. We found that, within some UV-induced apoptotic cells, Bak could start to form small complexes in a few mitochondria first. Under this situation, only these mitochondria showed a reduced membrane potential, as indicated by their decreasing fluorescent intensity of Mitotracker (supplementary material Fig. S2). This result indicates that formation of small complex by Bak is associated with the permeabilization of MOM.

Furthermore, to determine whether Bax and Bak form small complexes at precisely the same time during the apoptotic process, we used live cell imaging to directly measure the dynamic process of Bax and Bak reorganization at the MOM in apoptotic cells. HeLa cells were cotransfected with Cherry-Bax and YFP-Bak genes. After UV irradiation, we recorded the redistribution of Bax and Bak at different time points. A sample record of this time series measurement is shown in Fig. 7. In this particular cell, neither Bax nor Bak had formed aggregates at the MOM at t=112 minutes 54 seconds following UV treatment. Bax was distributed uniformly in the cytosol and MOM, whereas Bak was clearly distributed at the MOM only (Fig. 7B). One minute later (at t=114 minutes 1 second), both Bax and Bak were observed to start forming small complexes at the MOM (Fig. 7C). Half a minute later, Bax and Bak had formed small complexes in most of the mitochondria (Fig. 7D). These results suggest that Bax and Bak are activated together to form small complexes during apoptosis.

To further determine whether Bax and Bak can form complexes with each other, we cotransfected CFP-Bax and YFP-Bak into HeLa cells and examined their colocalization during apoptosis. In the merged images of CFP-Bax (labeled here in red) and YFP-Bak (green) (Fig. 8A-C), we found that the majority of Bax and Bak complexes colocalized. A few Bax or Bak complexes, however, were found to form independently (see arrows 1 and 2 in Fig. 8C). This result clearly indicated that both Bax and Bak can form complexes at the MOM either with each other or by themselves. This observation may explain why Bax and Bak have similar and redundant functions.

Of course, colocalization of Bax and Bak complexes at the MOM is not sufficient to prove that Bax and Bak can form hetero-oligomers in vivo. In order to test whether Bax and Bak can interact tightly within Bax-Bak complexes at the MOM, we used a fluorescence resonance energy transfer (FRET) method to assay their interaction. It is known that when two fluorescent molecules bind with each other, exciting one molecule (the donor) can transfer the energy to its partner (the acceptor). As a result, fluorescent light is emitted from the acceptor instead of the donor (Heim and Tsien, 1996). This phenomenon is called FRET. FRET is considered to be the most powerful technique for monitoring protein interactions in living cells with precise spatial (angstrom to nanometer) and temporal (nanosecond) resolution (Piehler, 2005; Sekar and Periasamy, 2003). A successful FRET system yields information about the location and exact timing of the interaction that is not available from conventional detection systems such as coimmunoprecipitation, crosslinking and yeast two-hybrid analysis (Truong and Ikura, 2001). We conducted a FRET measurement between CFP-Bax and YFP-Bak within an apoptotic HeLa cell by exciting the cell using a CFP filter and recording the image using the YFP filter. In the absence of FRET, very little light would be detected. But if FRET does occur, a large fluorescent signal would be observed.

Results of our FRET measurements are shown in Fig. 8D-G. We found a strong FRET effect between CFP-Bax and YFP-Bak in their coaggregation at the MOM. As shown in Fig. 8D, there were three types of aggregates in the apoptotic cell: (1) aggregates formed by CFP-Bax alone, (2) aggregates formed by YFP-Bak alone, and (3) aggregates formed by CFP-Bax and YFP-Bak (labelled with arrows 1, 2 and 3, respectively in Fig. 8D). Here, the aggregates formed by Bax alone or Bak alone were used as negative controls. To analyze the FRET data, we first drew lines to pass through the three types of aggregates (arrows 1, 2 and 3) in the merged image (Fig. 8D). Then, we copied these lines in the corresponding areas of the FRET images (taken with CFP excitation and YFP emission) and measured the FRET signal intensity along these lines (Fig. 8D). The results were plotted in Fig. 8E-G. It is clear that in the aggregates composed of Bax or Bak alone (Fig. 8E,F), no FRET signal was found (as expected). But in the co-aggregate formed by both Bax and Bak, a strong FRET signal was detected (Fig. 8G).

To further verify our FRET results, we used a second method to detect the FRET effect between CFP-Bax and YFP-Bak. After Bax-Bak complexes were formed at the MOM during UV-induced apoptosis, we measured both the CFP and YFP images under CFP excitation. Then, we photobleached the YFP-Bak molecules in a small region of the cell (marked by the box in Fig. 8H) using YFP laser excitation light (514 nm). Following this photobleaching step, we measured the CFP and YFP images again under the CFP excitation. We found that within the photobleaching region, the CFP-Bax signal was significantly enhanced, whereas the YFP-Bak signal was greatly reduced (Fig. 8H). This result indicates that there was a strong FRET effect between the donor (CFP-Bax) and the acceptor (YFP-Bak). Results of a more quantitative analysis of this experiment using a line-scanning are shown in Fig. 8I,J. Thus, results of our FRET studies strongly suggest that Bax and Bak do interact with each other to form hetero-oligomers at the MOM during apoptosis.

From the single live cell image measurements, we concluded that the formation of Bax-Bak small complexes is responsible for the release of Cyt c and Smac. But it is still unclear what the nature of these small complexes is and how many molecules are contained in these small complexes. It has been suggested that Bax may form dimers, tetramers, or even 30-mers, which could be detected biochemically by gel filtration or chemical crosslinking (Antonsson et al., 2001; Tan et al., 1999). The size of the Bax small complexes we observed appeared to be significantly larger. We used an imaging method to directly determine the sizes of GFP-Bax small complexes in vivo by calibrating them against man-made fluorescent beads of different sizes of 0.2 μm, 0.5 μm, 1 μm to 2 μm (FluoSpheres Fluorescent Microspheres kit) (Fig. 9A). First, using a Zeiss Axiovert 200M microscope, we measured the pixel number of different sizes of man-made fluorescent beads (Fig. 9B). Then we constructed a calibration curve between the pixel number and diameter of beads (Fig. 9C). Based on the standard curve, we determined the size of GFP-Bax small complexes at the MOM by measuring the pixel number of GFP-Bax small complexes (Fig. 9D). From measurements in six HeLa cells, we calculated that the size of Bax complexes was 0.25±0.04 μm (n=316) (Fig. 9E). We would like to point out that, since the optical resolution is about 0.2 μm, the measured size of the Bax complexes was very close to the limit of the resolution of a light microscope. The distribution of GFP-Bax complexes smaller than 0.2 μm could not be determined accurately.

To measure the number of Bax molecules within a single aggregate, we employed a quantitative fluorescence imaging technique. First, using a sonication method, we generated purified recombinant GFP (0.5 μg/μl) droplets of different sizes in an oil bath. Then, using a fluorescent microscope, we imaged these GFP droplets under identical conditions as our live cell measurements (Fig. 9F). By knowing the volume of the droplets (based on the measurement of the diameter of GFP droplets) and its GFP concentration, we could calculate the number of GFP molecules in a single GFP droplet. Thus, we could generate a calibration curve linking the fluorescence intensity of GFP droplets with the number of GFP molecules (Fig. 9G). Using this calibration curve, we were able to determine the number of GFP-Bax molecules within a Bax complex based on its integrated fluorescence intensity (Fig. 9H). Our quantitative measurements showed that, when Bax started to form small complexes to permeabilize the MOM, the average number of GFP-Bax molecules in an aggregate was 147±34 (n=150).

This value is very likely to be an underestimate. First, the Bax complex should also contain endogenous Bax, which gave no fluorescent signal. Second, we have shown previously that Bax and Bak interacted to form small complexes in the mitochondria during apoptosis; an unknown number of Bak molecules must also be contained in the Bax-Bak small complexes. Thus, the small complex formed by Bax-Bak during apoptosis, which was responsible for the release of Cyt c and Smac, is likely to contain a couple of hundred Bax-Bak molecules.

Using the same methods, we also determined the size of the Bax clusters at the later apoptotic stage after mitochondria swelled (stage IV shown in Fig. 2D). The average size of these Bax clusters was 0.62±0.11 μm. The number of GFP-Bax molecules in a typical Bax cluster was calculated to be 1021±356 (n=120). These values were consistent with results reported by an earlier study, in which the number of Bax molecules in the apoptotic clusters was estimated to be 1×10³ to 2×10⁴ (Nechushtan et al., 2001).

In this study, we used single live cell analysis to directly measure the dynamic changes of Bax distribution during apoptosis (Figs 2, 3). We found that in response to the stress signal, Bax first translocated to mitochondria and distributed uniformly at the MOM. It appeared as a filamentous structure (Fig. 2B,E). Within a few minutes, Bax started to diffuse laterally to aggregate into one or two isolated spots at the MOM (Fig. 2F and Fig. 3). As soon as Bax complexes were formed, Cyt c and Smac started to be released from mitochondria (Figs 3, 4, 5). About 10 minutes later, the Bax complexes further expanded into large clusters (Fig. 2G and Fig. 3).

Previously, several groups have tried to relate the distribution patterns of Bax with Cyt c release (Capano and Crompton, 2002; Valentijn et al., 2003; Wolter et al., 1997). Their conclusions differed widely. Some suggested that the release of Cyt c occurred at the filamentous stage of Bax distribution (Capano and Crompton, 2002), whereas others suggested that Cyt c was released after Bax formed clusters (Valentijn et al., 2003). In this study, we showed that the release of Cyt c and Smac was initiated when Bax formed small complexes at the MOM, but not at the Bax filamentous stage, or in the later cluster formation stage (Figs 3, 4, 5).

Our previous study showed that the release of Cyt c and Smac from mitochondria is through the same pathway (Zhou et al., 2005). Recently, it was further demonstrated by Munoz-Pinedo et al. that many different mitochondrial intermembrane proteins, including Cyt c, Smac, Omi and AK2, are released together during apoptosis in a manner that is co-ordinately initiated (Munoz-Pinedo et al., 2006). These findings, together with our dynamic measurement of Bax and Bak distribution, support a model in which the permeabilization of the MOM is due to the rapid formation of pores by the aggregation of Bax and Bak; these pores allow the release of many soluble proteins. We estimate that this pore formation time is less than 5 minutes (Fig. 3).

What then is the molecular structure of the `pore' formed by Bax (or Bak) to permeabilize the MOM? At present, several models have been proposed, including: (1) several Bax proteins may insert into the MOM, resulting in forming a Bax-oligomer channel (Saito et al., 2000); (2) Bax may bind with VDACs to form a hybrid channel (Shimizu et al., 1999); (3) A large number of Bax molecules may aggregate at the MOM and destabilize it by forming a lipid-protein complex (Basanez et al., 1999; Desagher and Martinou, 2000; Epand et al., 2003; Garcia-Saez et al., 2007; Kuwana et al., 2002; Lucken-Ardjomande and Martinou, 2005; Terrones et al., 2004). In this study, we experimentally determined the number of molecules contained in the Bax complexes. We found that the small complex consisted of more than 100 Bax molecules (Fig. 9). Hence, the size of the pore formed by Bax is far larger than a typical channel, such as those reconstituted into pure liposomes (Saito et al., 2000). Thus, our data do not support model 1 or model 2 listed above. Our conclusion is consistent with the findings of Kuwana et al. who reported that Bax could produce mega openings at the MOM, which could release large molecules of up to 2000 kDa (Kuwana et al., 2002). It is difficult to conceive that a `channel' can have such a large `pore'. Based on results of this study, we think that the `pore' structure responsible for MOM permeabilization during apoptosis is more like a leaky membrane protein complex, which is formed by a large number of Bax-Bak molecules inserting into a local region of the MOM. Such an insertion disrupts the bilayer structure of the membrane and allows many different mitochondrial intermembrane proteins to be released. Thus, our data are more in line with model 3 above.

Although we do not think Bax or Bak forms channels by itself or with VDACs, it is not yet clear whether VDACs might play a role in Bax aggregation. It is conceivable that Bax may use other membrane proteins as a nucleation site to begin its aggregation at the MOM. In this case, the interaction between Bax and these membrane proteins could be functionally important. These membrane proteins could include VDACs or TOM (translocase outer membrane) (Bellot et al., 2007; Shimizu et al., 1999).

In conclusion, the results of this study suggest that Bax undergoes several dynamic redistribution stages during UV-induced apoptosis as shown in Fig. 10. In stage I, before its activation, Bax is distributed uniformly in the cytosol. In stage II, Bax is activated to translocate from the cytosol to the MOM. This activation is believed to be caused by a conformational change in the Bax molecule that exposes its transmembrane domain (Nechushtan et al., 1999). At this time, Bax is evenly distributed at the MOM and does not appear to interact with Bak. In stage III, Bax starts to diffuse laterally to aggregate into one or two isolated spots at the MOM (Fig. 2F). Bax can form hetero-oligomers with Bak during this process. The formation of Bax-Bak small complexes appears to destabilize the bilayer structure of MOM in the local region and thus permeabilize the MOM. As a result, a variety of mitochondrial intermembrane proteins, including Cyt c and Smac, begin to be released from mitochondria into the cytosol (Figs 3, 4, 5). In stage IV, the Bax-Bak complexes can further expand to become large clusters (Fig. 2G and Fig. 3), but the dysfunction of mitochondria is already complete before reaching this stage.

Mitotracker Red CMXRos and FluoSpheres kit were from Molecular Probes (Eugene, OR); Rabbit anti-Bax polyclonal antibodies (2772 and 2774) and Rabbit anti-caspase-3 polyclonal antibody were from Cell Signaling Tech; Rabbit anti-cytochrome c (H-104) polyclonal antibody and mouse anti-PARP (F-2) monoclonal antibody were from Santa Cruz; mouse anti-cytochrome-c monoclonal antibody (native) was from PharMingen for immunostaining; goat anti-rabbit IgG antibody (FITC) and goat anti-mouse IgG antibody (Rhodamine) were from Calbiochem; goat anti-mouse IgG (HRP) and goat anti-rabbit IgG (HRP) were from Bio-Rad. Other chemicals were from Sigma (St Louis, MO).

HeLa cells were grown on glass coverslips at 37°C in a humidified atmosphere containing 5% CO₂ in modified Eagle's Medium (MEM) supplemented with 10% fetal bovine serum (FBS) plus 100 U/ml penicillin and 100 μg/ml streptomycin. Apoptosis was induced by exposing cells to UV irradiation (300 μW/cm²) for 3 minutes.

Human BAX gene was isolated from pRSET-Bax (Narita et al., 1998) with primers 5′-CCCAAGCTTCGATGGACGGGTCCGGGGAG-3′ (sense) and 5′-GGAATTCAGCCCATCTTCTTCCAGATG-3′ (anti-sense), then cloned between the HindIII and EcoRI sites in a pEGFP-C1 (Clontech), pECFP-C1 (Clontech) or pCherry-C1 vector [modified from pRSET-Cherry (Shaner et al., 2004)]. Smac-YFP was described before (Zhou et al., 2005). pEYFP-C1-Bak was kindly supplied by Richard J. Youle (Nechushtan et al., 2001). HeLa cells were transfected with plasmid of fusion gene using an electroporation method (Chang, 1997) and were allowed to express the fusion gene for 24 hours before the experiment.

Cells grown in 60 mm Petri dishes were collected at the indicated time points. The cytosolic and mitochondrial fractions were separated using a digitonin-based subcellular fractionation technique (Adrain et al., 2001). For western blotting, proteins (50 μg per lane) were analysed by SDS-PAGE and followed by electrophoretic transfer onto the nitrocellulose membrane. Then, the membrane was probed (or reprobed) with specific antibodies using standard procedures.

Cells grown on coverslip were fixed with 4% paraformadehyde plus 0.1% glutaraldehyde for 15 minutes at room temperature, then permeabilized in 0.2% Triton X-100. Next, cells were immunostained using the standard method (Li et al., 1999). Rabbit anti-Bax (2774) (1:50 dilution, 4°C overnight) and goat anti-rabbit IgG (FITC) (1:100 dilution) were used to stain endogenous Bax. Mouse anti-native cytochrome c (1:200 dilution, 4°C overnight) and goat anti-mouse IgG (Rhodamine) (1:200 dilution) were used to stain endogenous Cyt c.

To stain mitochondria and measure the change of ΔΨ_m, HeLa cells were incubated with MEM containing 0.5 μM Mitotracker Red for 5 minutes followed by washing with MEM once. After apoptotic treatment, cells grown on round coverslips were mounted onto a chamber containing an observation medium (HEPES-buffered Dulbecco's MEM containing 4 mM glutathione, 1 mM L-ascorbic acid, 0.5 mM DTT, pH 7.4). The cells were examined under a microscope at 37°C with a temperature controller.

Three imaging systems were used for live cell observation: (1) An inverted fluorescence microscope (Axiovert 200M, Zeiss) to image Mitotracker and proteins labeled with Cherry, GFP, CFP or YFP. The images were recorded using a cooled CCD camera (Diagnostic Instruments) under a 100×/1.4 NA oil objective. (2) A confocal microscope (Bio-Rad MRC-600) equipped with a krypton/argon laser to image GFP (Ex: 488 nm) and Mitotracker (Ex: 568 nm). (3) A confocal microscope (Leica TCS SP2) for the photobleaching/FRET experiment. FRET experiments were done either using the Zeiss imaging system (Ex: 440 nm; Em: 535 nm) or the Leica confocal microscope (Ex: 458 nm; Em: 520-580 nm). YFP-Bak was photobleached with 514 nm excitation radiation. We used the MetaMorph v6.0 software (Universal Image, West Chester, PA) to analyze digital images.

We thank Lingli Zhou and Cheung-Ming Chow for technical assistance. We also want to thank Y. Tsujimoto for kindly supplying us the pRSET-Bax gene, Xiaodong Wang for the Smac-flag gene, Roger Y. Tsien for the pRSET-Cherry gene and Richard J. Youle for the YFP-Bak gene. This work was supported by the Research Grants Council of Hong Kong (HKUST6466/05M, 660207 and N_HKUST616/05).