Research article

Bax and Bcl-xL exert their regulation on different sites of the ceramide channel

Meenu N. Perera, Shang H. Lin, Yuri K. Peterson, Alicja Bielawska, Zdzislaw M. Szulc, Robert Bittman, Marco Colombini


The present study demonstrates the important structural features of ceramide required for proper regulation, binding and identification by both pro-apoptotic and anti-apoptotic Bcl-2 family proteins. The C-4=C-5 trans-double bond has little influence on the ability of Bax and Bcl-xL to identify and bind to these channels. The stereochemistry of the headgroup and access to the amide group of ceramide is indispensible for Bax binding, indicating that Bax may interact with the polar portion of the ceramide channel facing the bulk phase. In contrast, Bcl-xL binding to ceramide channels is tolerant of stereochemical changes in the headgroup. The present study also revealed that Bcl-xL has an optimal interaction with long-chain ceramides that are elevated early in apoptosis, whereas short-chain ceramides are not well regulated. Inhibitors specific for the hydrophobic groove of Bcl-xL, including 2-methoxyantimycin A3, ABT-737 and ABT-263 provide insights into the region of Bcl-xL involved in binding to ceramide channels. Molecular docking simulations of the lowest-energy binding poses of ceramides and Bcl-xL inhibitors to Bcl-xL were consistent with the results of our functional studies and propose potential binding modes.

  • ABT-263
  • ABT-737
  • apoptosis
  • Bcl-2 family protein
  • ceramide analogue
  • ceramide channel
  • 2-methoxyantimicin A3
  • mitochondrion


The intrinsic pathway of apoptosis, a type of programmed cell death, is regulated at the mitochondrial level by the Bcl-2 family of proteins. Permeabilization of the MOM (mitochondrial outer membrane) leads to the release of key IMS (intermembrane space) proteins, such as cytochrome c, which irreversibly commit the cell to the execution phase of apoptosis. The Bcl-2 proteins have pro-survival members, such as Bcl-xL, Bcl-2 and Mcl-1, that maintain the integrity of the MOM in normal cells and pro-apoptotic members, such as Bax, Bak and Bid, that initiate the apoptotic process and facilitate IMS protein release when the cell is damaged or an apoptogenic signal is delivered. Whether the anti-apoptotic or pro-apoptotic Bcl-2 proteins prevail will determine whether IMS proteins are released by a pore through the MOM and thus the fate of the cell. There are multiple pathways proposed for the release of IMS proteins from mitochondria [14], and even Bax itself is able to form channels [5,6]. Whatever the nature of the pathway, it must be tightly regulated to prevent unwanted cell death. One such proposed pathway for IMS protein release is formed by the bioactive sphingolipid ceramide.

The involvement of ceramide in apoptosis is widely recognized, and intracellular ceramide levels increase in response to many different apoptotic stimuli; specifically, it is the elevation of ceramide in mitochondria that is pro-apoptotic [710]. Elevation of ceramide in mitochondria can occur in a number of ways [11,12], including transfer of ceramides from mitochondria-associated membranes [13]. In fact, whereas Bax/Bak-knockout cells are known to be highly resistant to apoptosis [14,15], it was shown recently that this resistance is accompanied by a deficit in ceramide generation, as Bak regulates ceramide metabolism by activating a critical ceramide synthase to elevate mitochondrial ceramide levels [16]. Moreover, whereas metabolism of ceramide to other forms, such as glucosylceramide, can render cells less susceptible to cytotoxic agents [1719], blocking ceramide metabolism can sensitize multidrug-resistant cells to death [2022]. Thus the steady-state level of ceramide is an important step in the apoptotic programme in a variety of cell types [18,23], and cytotoxic stimuli [8,9,20] and ceramide channel formation can be demonstrated at these physiological levels of ceramide in mammalian and yeast mitochondria [1,24,25]. Ceramide channels have also been visualized in liposomes using transmission electron microscopy [26]. These structures are stable and have been shown to be large enough for the release of all known IMS proteins from mitochondria during apoptosis [25].

The current model of the ceramide channel highlights the structural features that are important for channel formation [27] (see Figure 1 for structures of ceramide and analogues). The amide nitrogen and carbonyl group allow for the formation of ceramide columns. Adjacent ceramide columns are held together by the C-1 and C-3 hydroxy groups, forming the lumen of the channel. The length of the hydrocarbon chains of ceramide, although physiologically important [28,29], is relatively unimportant for channel formation [30]. The sphingoid base is typically 18 carbons in length with a C-4=C-5 trans-double bond. The ceramide species in mammalian cells have fatty-acyl chains that range from 12 to 24 carbons in length. It is important to note that ceramide channels are highly stable rigid structures that are held together by a vast network of hydrogen bonds [27,31] and not simply lipidic pores of a transient non-specific nature.

Figure 1 Structures of ceramides and analogues used

Coded boxes highlight the key areas of the ceramide molecule that are important for ceramide channel formation (top). The codes indicate the location on the ceramide molecule where changes were introduced (S for stereochemistry, H for hydroxy group, T for trans-double bond, A for amide linkage and L for hydrophobic chain length) and the analogues are named throughout the paper using these short codes followed by a number as indicated.

It was found that ceramide channels are regulated by the Bcl-2 family of proteins. The pro-survival Bcl-2 members, Bcl-xL and Ced-9 (the Caenorhabditis elegans homologue of Bcl-2) were shown to antagonize channel formation and disassemble ceramide channels [24]. Conversely, a pro-apoptotic Bcl-2 member, Bax, can enhance ceramide channels directly [31]. The regulation of ceramide channels was demonstrated by the addition of purified Bcl-2 family proteins to isolated mammalian mitochondria (which have other Bcl-2 family proteins naturally present in those membranes), yeast mitochondria (which do not have potentially interfering Bcl-2 family proteins) and planar phospholipid bilayers (which are devoid of any proteins or specialized lipids). These results highlight the ability of Bcl-2 family proteins to identify and regulate ceramide channels. These interactions are highly specific and do not require other proteins or specialized lipids, demonstrating for the first time that a channel formed of lipids can be regulated by a protein. Remarkably, this regulation was in accordance with the known physiological function of these Bcl-2 proteins.

In the present study, we set out to understand further the interaction of ceramide channels and Bcl-2 family proteins. We provide insights into the specific sites on the ceramide channel where Bax and Bcl-xL exert their regulatory effects. Moreover, to investigate the binding site on Bcl-xL for ceramide, we tested specific inhibitors of Bcl-xL that bind to a known location on the protein. Molecular docking studies were conducted with ceramides and Bcl-xL inhibitors to probe for the ceramide-binding site and reveal the lowest-energy binding modes of these molecules to Bcl-xL. We offer mechanistic insights as to how the specific interactions that we discovered could lead to the regulation of ceramide channels and discuss the implications of these results.



C2-Cer (N-acetyl-D-erythro-sphingosine), C4-Cer (N-butyryl-D-erythro-sphingosine), C8-Cer (N-octanoyl-D-erythro-sphingosine), C16-Cer (N-palmitoyl-D-erythro-sphingosine), C18:1-Cer (N-oleoyl-D-erythro-sphingosine), C20-Cer (N-arachidoyl-D-erythro-sphingosine) and C24-Cer (N-lignoceroyl-D-erythro-sphingosine) were obtained from Avanti Polar Lipids. Antimycin A, DNP (2,4-dinitrophenol), and fatty-acid-depleted BSA were purchased from Sigma. The analogues of D-erythro-C16-ceramide and the analogues T4 and L5 (Figure 1) were synthesized as described previously [3236]. Horse heart cytochrome c was purchased from Acros, 2-MeAA3 (2-methoxyantimycin A3) was purchased from Enzo Life Sciences, and ABT-737 and ABT-263 were purchased from ChemieTek.

Preparation of rat liver mitochondria

Rat liver mitochondria were isolated from male Sprague–Dawley rats by differential centrifugation of tissue homogenates as described previously [37] with modifications described in [25]. BSA medium was used to wash the liver during the excision and initial centrifugation steps and was removed by centrifugation at 8700 g (twice) in BSA-free medium. The final mitochondrial pellet was resuspended in ~3–5 ml of ice-cold sucrose-free isotonic FH buffer (280 mM mannitol, 0.1 mM EGTA and 2 mM Hepes, pH 7.4). The mitochondrial intactness was determined from the rate of cytochrome c oxidation compared with the rate measured after mild hypotonic shock [38] as described below. The animal use protocols were approved by the Institutional Animal Care and Use Committee. The animals were killed by a procedure consistent with the Panel on Euthanasia of the AVMA (American Veterinary Medical Association). The animal facility used to house the animals is accredited by AAALAC (Association for Assessment and Accreditation of Laboratory Animal Care).

Purification of recombinant proteins

Recombinant human Bax was purified as described previously [39] with modifications described in [31]. One change made was to reduce the Tris concentration from 20 mM to 10 mM in both dialysis steps; this change had no effect of the activity of the protein. The protein was shell-frozen in small aliquots in thin-walled glass tubes using ethanol/solid CO2 and stored at −84°C. A fresh monomeric Bax sample was thawed before each experiment. Standardizing the concentration of Bax to 5–10 μg/ml with FH buffer (10 mM Tris/HCl, pH 8.0, solution could also be used) before activation improved reproducibility between different Bax preparations. This monomeric Bax was activated using 10% β-octyl glucoside to a final concentration of 1% and incubated for 30 min on ice; this activated Bax was used in the experiments throughout the study. Full-length recombinant Bcl-xL was isolated as described previously [40] and the following changes are recommended. Only 5 ml of overnight culture was inoculated into 1 litre of LB (Luria–Bertani) medium and this was allowed to incubate with shaking until the D600 reached 0.600. The culture was then induced with 0.01 mM IPTG (isopropyl β-D-thiogalactopyranoside) for 2 h at 37°C. After the cells were harvested by centrifugation, the pellet was resuspended in 40 ml of PBS and incubated on ice for 20 min with 140 units of lysozyme and 35 μM PMSF (final concentration). The cells were then subjected to two passes through a French press and centrifuged to remove cell fragments. The lysate was then incubated with glutathione–agarose beads for 2 h at 4°C and packed into a column. The beads were washed with 10 column volumes of ice-cold PBS containing 35 μM PMSF and another 10 column volumes of 20 mM Tris/HCl (pH 8.0). Then, 10 ml of the same buffer supplemented with 5 units of biotinylated thrombin was added to the column and allowed to incubate overnight at 4°C. The protein was eluted, and 80 μl of streptavidin beads were added to remove the remaining thrombin. After a 30 min incubation with rotation, the beads were removed by centrifugation and the sample was filter-sterilized. Glycerol (10%) was added to the protein and 100 μl aliquots were rapidly shell-frozen in ethanol/solid CO2. The published method included the addition of Triton X-100 to the cells along with the PMSF and lysozyme before French pressing of the cells. This greatly increased the yield (1.4 mg/ml with Triton X-100 compared with 165 μg/ml without Triton X-100). Both methods resulted in proteins with identical function in our assays.

Cytochrome c oxidation assay

The ability of cytochrome c to translocate through the MOM and become oxidized by cytochrome c oxidase (found on the mitochondrial inner membrane) is the rate-limiting step, thus this assay is a measurement of the permeability of the MOM. Whereas measurements from mitochondrial release assays indicate that mitochondria were permeabilized at some time point, perhaps transiently, one of the advantages of using this dynamic cytochrome c oxidation assay is that real-time permeability of the MOM can be measured. Thus this assay allows us to garner more information about the nature and stability of the channels in question. This assay was performed as described previously [41] with the following modifications to optimize reproducibility as mitochondria were found to be more stable at higher concentrations. In a typical experiment, unless noted otherwise in the Figure legends, mitochondria were diluted in FH buffer at 4°C to a concentration of 0.5 mg/ml, in small batches just before the assay. Then, 50 μl aliquots were dispersed in 650 μl of room temperature (21–24°C) reaction buffer [160 mM mannitol, 0.1 mM EDTA, 2 mM Hepes, pH 7.25, 60 mM KCl, 5 mM DNP and 1.3 μM antimycin A (antimycin A was tested in our system with Bcl-xL and ceramide and no effect was seen at the highest concentration tested which was 2.7 μM)]. The final mitochondrial protein concentration was 36 μg/ml. The mitochondria were allowed to acclimate at room temperature for 5 min in a microfuge tube. Then ceramide or a ceramide analogue (dissolved in propan-2-ol at 1 mg/ml) was delivered to the mitochondria, while the suspension was vortex-mixed for 30 s to achieve effective dispersal of the sphingolipid. After dispersal, the mixture was incubated for 10 min at room temperature followed by addition of cytochrome c (12 μl; final concentration 27 μM) and immediate measurement of the absorbance at 550 nm for a period of 2 min. The initial rate of decline of absorbance of reduced cytochrome c was used as a measure of the permeability of the MOM to cytochrome c; ϵ550 (red-ox)=18.5 mM−1·cm−1. The amount of vehicle added was less than 4% of the sample volume, and all vehicle controls were treated in an identical way. Rates were corrected for the rate of oxidation observed with vehicle alone and this was very close to the untreated rate arising from a small number of damaged mitochondria. The maximal rate of cytochrome c oxidation was assessed by subjecting the mitochondria to mild hypotonic shock as described previously [38] with the following modifications. Briefly, mitochondria were diluted in double-distilled water to 1 mg/ml and incubated on ice for 10 min before adding 2×FH buffer (FH buffer with double the amount of additives) to restore the isotonic conditions. Then, 50 μl of this 0.5 mg/ml mitochondrial stock was added to the reaction buffer and tested for cytochrome c oxidation as described above. In these experiments, the intactness of the mitochondrial preparations was greater than 85%. The rate of cytochrome c oxidation was converted into the percentage permeabilization of mitochondria by taking this rate as a percentage of the maximal rate of oxidation obtained by subjecting the mitochondria to mild hypotonic shock. The permeabilization induced by ceramide alone is given for each experiment in the Figure legends. It is important to note that even a relatively low percentage of permeabilization of mitochondria, as seen with our Bax experiments, is physiologically relevant [1,42,43], and even low levels of cytochrome c release are sufficient to induce apoptosis in whole cells [44,45].

It should be pointed out that, unlike liposomes, mitochondria are complex organelles whose function decays with time. Thus we are limited in what experiments can be performed within the functional time window. Also, it is necessary to monitor the state of the mitochondria. Before experiments are performed, the mitochondria are tested for the ability of C16-Cer to permeabilize the MOM and for the ability of the Bcl-2 family proteins to alter this permeabilization. These tests are repeated as time passes to ensure that any results obtained in the experimental trials are meaningful. The disadvantages of using a ‘living’ organelle are more than compensated for by performing experiments in the natural membrane environment. Clearly, the greatest level of understanding is reached by performing experiments at various levels of complexity, from the intact cell to the pure components in a defined system.

Testing Bcl-2 proteins on channels formed by ceramide and its analogues

The sensitivity of isolated mitochondria to permeabilization by added ceramide or analogues varied from one preparation to another. Therefore experiments with analogues were always performed in parallel with experiments with C16-Cer. Furthermore, the amounts of C16-Cer and analogue added were adjusted (dilutions were made to keep the volume added to the sample constant at 10 μl) so as to induce approximately the same extent of MOMP (MOM permeabilization). This was done because some analogues of ceramide may insert more or less readily into mitochondria or have a different propensity for channel formation. By finding conditions that result in the same degree of MOMP, we achieve the same level of channel formation. The permeabilization was measured as the rate of cytochrome c oxidation. Bax, Bcl-xL, 2-MeAA3, ABT-737 and ABT-263 were added along with the mitochondria for 5 min of incubation at room temperature before sphingolipid dispersal. Stock solutions of 2-MeAA3 (1.9 mM in propan-2-ol), ABT-737 (0.7 mM in DMSO) and ABT-263 (0.7 mM in DMSO) were diluted to the appropriate concentration with the same solvent so that the volume added to reach the concentration indicated in the Figures remained constant (10, 5 and 5 μl respectively). Results are shown as a percentage change comparing the effect of adding a Bcl-2 protein in combination with ceramide (or an analogue) compared with the addition of ceramide (or the analogue) alone.

Molecular docking simulations of potential Bcl-xL inhibitors

Modelling, simulations and visualizations were performed using MOE (Molecular Operating Environment) version 2011.10 (Chemical Computing Group). Simulations were performed on a Dell E8500 with an Intel Core 2 Duo @ 3.16 GHz using the Windows XP operating system. All other computational procedures were performed using a Dell XPS M1530 with an Intel Core2 Duo processor T8300 @2.40 GHz with 2 GB RAM using the Windows Vista operating system. The structural file used as input for analysis and docking simulations was PDB code 1PQ1. Before analysis and simulations, the main-chain protein was protonated at pH 7.5 with a salt concentration of 0.2 M and the structures were energy-minimized using the Amber99 forcefield and Born solvation model. Simulations focused on the entire protein surface and Bcl-xL was held rigid while the ligand was flexed. Initial placement calculated 500 poses per molecule using triangle matching placement with London dG scoring, the top 250 poses were then refined using forcefield placement and Affinity dG scoring. The resulting interaction after refinement is called E Score 2. This is a measure of the strength of the interaction, with stronger interactions yielding a more negative number. Ligands were prepared in MOE, protonated at pH 7.5, and chirality formalized within the ligand database. The top poses generated from each ligand were then minimized in place as a system with Bcl-xL to allow the protein to flex as well as the ligand. System minimization also used the Amber99 forcefield and Born solvation model.

Statistical analysis

All results shown are means±S.E.M. for three or four trials, and each experiment was repeated on three separate mitochondrial preparations. Student's t test was performed to obtain P values and identify a statistically significant change from the additive value (for Bax experiments) or individual C16-Cer or analogue addition (for Bcl-xL experiments). *P<0.05, **P<0.01, ***P<0.001.


Previous results [24,31] have indicated that the recognition between the ceramide channel and Bcl-2 proteins is highly specific. These findings prompted the present study in which we used a dynamic cytochrome c accessibility assay to assess the influence of Bax and Bcl-xL on the ability of ceramide and its analogues to permeabilize the MOM. To determine the specific site(s) of interaction of these Bcl-2 proteins on ceramide channels, we prepared ceramide analogues with changes in different functional groups thought to be important for channel formation. The analogues that retained channel-forming ability are illustrated in Figure 1. Experiments with ceramide and an analogue were performed in parallel. Before each experiment, the amounts of ceramide or ceramide analogue were adjusted so as to induce approximately the same extent of MOMP. In this way, we compensated for any differences in the propensity for channel formation and tested the Bcl-2 family proteins on the same degree of channel presence. As a result, the ability of the Bcl-2 family proteins to alter the MOMP induced by ceramide could be compared with that for the analogue.

When different doses of ceramide or ceramide analogues were applied to achieve a comparable degree of permeabilization, the amount of Bcl-2 family protein added was kept the same. This is because the amount of protein added greatly exceeded the number of channels present (for additional details, see the Supplementary Online Data at Thus the binding is not of very high affinity or stoichiometry. Previous studies showed that the channels are responsive to the effective concentration of added protein in a manner consistent with a dynamic equilibrium. Thus despite adding different amounts of the sphingolipid to achieve the same degree of permeabilization, it makes sense to use the same effective dose of Bcl-2 family protein to detect differences in effect on channel-mediated permeability.

The results described were obtained at room temperature, whereas apoptosis, in mammalian cells, typically takes place at 37°C. Thus differences in rates of reaction, fluidity, etc. resulting from a difference in temperature could easily affect the reported values, but are unlikely to change the fundamental conclusions about the specificity of the sites of interaction with ceramide associated with Bax and Bcl-xL. Indeed, key experiments reproduced at 37°C yield similar results as seen at 23°C (see Figures 4A and 5C, and Supplementary Figure S1 at, indicating that the physical properties of this interaction remain at physiological temperatures.

The importance of the trans-double bond

The following changes were introduced to the C-4=C-5 trans-double bond of the sphingoid backbone of ceramide: C16-phytoceramide (T4), which has a hydroxy group at the C-4 position instead of the double bond; C16-4,5-tb-ceramide (T5), which has a triple bond in place of the double bond; and C16-(4,5-allene)-ceramide (T6), which is a 4,5-cumulated diene. Previous results showed that the addition of either detergent-activated Bax or tBid (truncated Bid)-activated Bax along with C16-Cer to isolated rat liver mitochondria resulted in a much greater permeabilization of the MOM than either entity alone, indicating that Bax enhances the formation of ceramide channels [31]. We looked for a similar result on ceramide analogues. An example of the result of a typical paired experiment with C16-Cer and an analogue (T5) is shown in Figure 2(A). Detergent-activated Bax (‘Bax’) induces a small permeabilization, whereas C16-Cer (‘C16’) produces a more significant permeabilization. However, when both are added together (‘C16+Bax’), the resulting permeabilization is much greater than the value obtained by adding the values of the ‘Bax’ and ‘C16’ bars (‘Predicted C16+Bax’). The permeabilization induced by T5 was enhanced by the addition of Bax to a similar extent, indicating that the modification had not influenced the ability of Bax to act on T5 channels. Similar paired experiments were performed with the other analogues. Figure 2(B) shows that Bax enhanced the permeabilization induced by T4 to the same extent as T5, but was 40% less effective on T6.

Figure 2 Testing the ability of Bax and Bcl-xL to regulate channels formed in isolated rat liver mitochondria by ceramide analogues with changes in the C-4=C-5 trans-double bond or extent of unsaturation

The cytochrome c oxidation assay was used to determine the degree of MOMP induced by the addition of ceramide or an analogue and the enhancement or inhibition of this permeability in the presence of Bax or Bcl-xL respectively. (A) A typical Bax/C16-Cer/analogue experiment: the amount of C16-Cer (‘C16’, 10 μg) and T5 (5 μg) added was adjusted so that they induced approximately the same extent of MOMP when added alone. The addition of Bax (3.5 nM) along with C16-Cer (‘C16+Bax’) and T5 (‘T5+Bax’) resulted in enhancement of the MOM permeability induced by ceramide and the analogue. These combinations tested are compared with the permeability predicted by adding the effect of the respective individual treatments (‘Predicted’ columns, i.e. ‘Bax’+‘C16’=‘Predicted C16+Bax’). (B) Bax (3.5nM) can enhance channels formed by T4 (5 μg), T5 (5 μg) and T6 (5 μg) and is shown as a percentage increase in MOMP. (C) A typical Bcl-xL/C16-Cer/analogue experiment is shown here and is conducted similarly to that in (A). Bcl-xL (0.4 μM) inhibits C16-Cer (10 μg) channels and T5 (6 μg) channels to a similar extent. As Bcl-xL does not induce a detectible MOMP (see Figure 7A), the combination treatments of ‘C16+Bcl-xL’ and ‘T5+Bcl-xL’ were compared with the individual treatments of C16-Cer and T5 respectively and subject to the same statistical analysis as stated below. (D) Bcl-xL (0.4 μM) inhibited C16-Cer (10 μg), T4 (7.5 μg), T5 (6 μg) and T6 (5 μg) channels to the same extent and is shown as a percentage decrease in MOMP.

In the Bcl-xL experiments (Figure 2C), as in the Bax experiments, the doses of ceramide and analogue used were adjusted so that the extent of MOMP induced was approximately the same. When Bcl-xL is present (‘C16+Bcl-xL’ and ‘T5+Bcl-xL’), the permeabilization was greatly reduced in both cases and by the same amount. The percentage decrease in permeabilization is shown in Figure 2(D). Bcl-xL (0.4 μM) inhibited ceramide channel formation in isolated mitochondria by nearly 80% (Figure 2D). The same dose of Bcl-xL was similarly effective on T4, T5 and T6.

Analogues in which changes to the double bond were made by introducing a hydroxy group in its place (T4), changing it to a triple bond (T5) or adding another double bond next to it (T6) were well tolerated by both Bax and Bcl-xL. Thus the location of the C-4=C-5 double bond and the degree of unsaturation at this position have little influence on the ability of Bax or Bcl-xL to regulate these lipid channels.

Importance of the C-1 and C-3 polar groups

The C-1 and C-3 hydroxy groups of the ceramide are thought to hold adjacent columns of ceramide together to form the lumen of the ceramide channel. 3-O-Methyl-C16-ceramide (H2) has the C-3 hydroxy group blocked, yet this analogue can still form channels (Figure 3A), suggesting that the C-1 hydroxy group is sufficient to stabilize the channel. Both Bax and Bcl-xL can exert their respective effects on H2, indicating that the C-3 hydroxy group is not a site at which either protein interacts (Figures 3A and 3B).

Figure 3 Tests of the ability of Bax and Bcl-xL to regulate ceramide analogues with changes to polar headgroups or stereochemistry

(A and B) Blocking the C-3 hydroxy group does not reduce the channel's sensitivity to 3.5 nM Bax (5 μg of C16-Cer, 2 μg of H2, yielding ~10% MOMP) or 0.4 μM Bcl-xL (7.5 μg of C16-Cer, 5 of μg H2, yielding ~12% MOMP). (C) Bax (2 nM) is sensitive to changes in the stereochemistry found in analogue S1. Mitochondria were treated with 10 μg each of S1 and C16-Cer (~20% MOMP). (D) Channels formed by analogue S2 (5 μg) interact with Bax (3.5 nM) similarly to those formed by C16-Cer (5 μg) (~8% MOMP). (E) Bcl-xL (0.4 μM) inhibits C16-Cer (3.5 μg) channels more strongly than those formed by S1 and S2 (10 and 3.5 μg respectively, yielding ~10% MOMP).

The effect of changing the configuration at C-2 or C-3 of ceramide was explored by using two stereoisomers: the enantiomer (2R,3S,4E)-C16-ceramide (S1) and the diastereomer (2S,3S,4E)-C16-ceramide (S2). Bax was unable to enhance channels formed by S1 (Figure 3C). S1 is the mirror image of the natural ceramide and Bax is unable to recognize channels formed by S1. However, Bax was able to enhance channels formed by S2 to a similar degree as C16-Cer (Figure 3D), where only the orientation of the C-3 hydroxy group is altered. Bcl-xL was able to inhibit channels formed by S1 and S2 (Figure 3E), although to a lesser degree than those formed by C16-Cer. These results indicate that the ability of Bax to interact with ceramide channels does not require access to the C-3 hydroxy group (H2), but is highly sensitive to the configuration at C-2 of ceramide.

The importance of the amide group of ceramide

In the current model of the ceramide channel, the amide group is thought to form a hydrogen bond with the carbonyl group of another ceramide molecule [27]. These intermolecular interactions allow for the assembly of ceramide columns. Analogue A3, N-methyl-C16-ceramide, features a methyl group instead of the amide group. MOMP induced by A3 was not enhanced by the addition of 3.5 nM Bax, whereas parallel experiments with C16-Cer resulted in a 40% enhancement (Figure 4A and Supplementary Figure S1A). Thus Bax is unable to influence the structure of channels formed by A3, possibly due to a requirement for binding to the amide group. Bcl-xL also has a much reduced interaction with A3, but the effect is not completely diminished as with Bax (Figure 4B). (2S,3R,4E)-C16-urea-ceramide (A4) is a potent channel former as half as much is needed to permeabilize the MOM to the level achieved by a given amount of C16-Cer (Figure 4C). This could be attributed to the enhanced hydrogen-bonding ability afforded by the urea group and thus the potential to further stabilize ceramide columns. This change is opposite to that of A3 in terms of hydrogen-bonding ability of the molecule and Bax addition resulted in a greater enhancement of A4 channels than C16-Cer channels. Perhaps the increased hydrogen-bonding ability also serves to increase the strength of the interaction with Bax. Bcl-xL inhibits the A4 channels to the same extent as the C16-Cer channels (Figure 4D), indicating that Bcl-xL is not sensitive to the modification.

Figure 4 Bax is more sensitive than Bcl-xL to changes in the amide group of ceramide

(A) Blocking the amide nitrogen with a methyl group prevents Bax (3.5 nM) from enhancing channels made by A3 (10 μg of C16-Cer and A3 were added to induce approximately 10% MOMP). (B) Bcl-xL (0.1 μM) is approximately 50% less effective on A3 (20 μg) than is C16-Cer (6 μg); however, Bcl-xL can still cause an approximately 35% decrease in the permeability induced by A3 channels (~11% MOMP). (C and D) Analogue A4 contains a urea group, which increases the hydrogen-bonding ability of the molecule. Whereas Bax is more effective on A4 than C16-Cer, Bcl-xL is similarly effective on both. For these experiments, 3.5 nM Bax, 0.1 μM Bcl-xL and 10 μg of C16-Cer and 5 μg of A4 were used to induce approximately 22% MOMP.

The importance of tail length and apolar bulk

The importance of the apolar region of ceramide for Bcl-2 protein interaction was tested with a minimal analogue, N-palmitoylserinol (L5). It only has one aliphatic chain. Bax was able to interact with L5 (Figure 5A) almost as well as C16-Cer. Surprisingly, Bcl-xL was not able to inhibit these channels at all (Figure 5B). Likewise, the permeability induced by a very-short-chain ceramide analogue, C2-Cer, was also not inhibited by Bcl-xL (Figure 5C and Supplementary Figure S1B). To test whether this unexpected result was due to kinetic delays, the added doses of the ceramides was adjusted to achieve a similar MOMP by C2-Cer and C16-Cer alone. Incubations of 15 and 30 min with Bcl-xL resulted in the same degrees of inhibition (Figure 5D). No effect of Bcl-xL on C2-Cer was seen at either time point, whereas the same substantial level of inhibition of C16-Cer permeabilization occurred at each time point. Thus the inability of Bcl-xL to inhibit C2-Cer channels is not dependent on incubation time.

Figure 5 Effects of analogues with altered hydrophobic regions on the regulation by Bcl-xL and Bax

(A) Bax (2 nM) enhances MOMP formed by L5 (10 μg), which has a truncated sphingoid base (10 μg of C16-Cer was used, ~30% MOMP). (B) Bcl-xL (0.4 μM) does not inhibit channels formed by L5 (3 μg); 7.5 μg of C16-Cer was added (~12% MOMP). (C) Varying the length of the N-acyl chain of ceramide reveals an optimal acyl chain length for Bcl-xL (0.4 μM) inhibition of ceramide channels. of C2-Cer–C18-Cer and C18:1-Cer were each used at 10 μg; C20-Cer and C24-Cer were each used at 7.7 μg added in a volume of 13 μl. Before addition of Bcl-xL, the amount of ceramide or ceramide analogue used induced approximately 19% MOMP. (D) Analysis of the time-dependence of Bcl-xL interaction. The addition of 6 μg of C2-Cer and 12 μg of C16-Cer in a volume of 12 μl was performed to induce a similar level of mitochondrial permeabilization (~40%). The addition of 0.1 μM Bcl-xL results in 60% inhibition of C16-Cer channels, but does not cause any change in the permeability of C2-Cer channels during the time of a typical experiment, 15 min. Increasing the incubation time by 15 min does not affect the inability of Bcl-xL to recognize C2-Cer channels.

Ceramide analogues with tail lengths varying between C2 and C24 were tested and the results show an optimal length for the acyl chain of C16, C18 and C20. Bcl-xL is known to have a hydrophobic groove and perhaps the optimal hydrocarbon bulk results in optimal binding. The introduction of an unsaturated bond in the acyl chain at C-9–C-10 of C18-Cer (C18:1-Cer) resulted in a large decrease in sensitivity to Bcl-xL inhibition. The kink in the fatty acyl chain may interfere with binding to Bcl-xL, indicating a very specific interaction.

These results have important implications. Various long-chain and very-long-chain ceramides are found in the MOM, and the specific N-acyl chains of the naturally occurring ceramides in membranes are thought to play an important physiological role [28,29,46]. In fact, ceramide synthases show specificity for different acyl-CoA substrates [4751], and some preferentially generate long-chain ceramides such as C16-Cer or C24-Cer [48,49]. In the early stages of apoptosis, cells are reported to generate primarily C16-Cer along with some C18-Cer and C20-Cer [16,52]. C24-Cer generation occurs at later stages [52]. This fits well with our observation that Bcl-xL interacts optimally with C16-, C18- and C20-Cer, as those species are elevated when the fate of the cell is still being decided. During this initiation stage of apoptosis, before the mitochondria have become permeabilized, various apoptotic signals result in the elevation of ceramides. Anti-apoptotic Bcl-2 family proteins, such as Bcl-xL and Bcl-2, could prevent these higher ceramide levels from forming channels resulting in MOMP. Bax activation and heterodimerization with Bcl-xL would relieve this inhibition and, at the same time, any excess Bax could favour ceramide channel formation and/or growth. Clearly, these interactions result in many opportunities for controlling this key decision-making step.

Many studies have been carried out with cell-permeant C2- and C6-Cers to trigger apoptosis in cancer cells. It is also evident that many enzymes that metabolize ceramide do not recognize C2-Cers [53,54], thus allowing these molecules to act over a longer time scale. Interestingly, the ability of short- and long-chain ceramides to form channels is about the same, and mixtures of long- and short-chain ceramides are actually more potent [55]. Thus it is possible that these short-chain ceramides may be more effective than long-chain ceramides therapeutically in promoting apoptosis because Bcl-xL is unable to recognize or regulate these species, while Bax retains the ability to promote channel formation by these ceramides.

Inhibitors provide insights into the region of Bcl-xL that binds to the ceramide channel

Cells overexpressing pro-survival proteins Bcl-xL and Bcl-2 are highly resistant to cell death, and many types of cancer indeed display this phenotype. Three known small-molecule inhibitors of Bcl-xL and Bcl-2 are 2-MeAA3, ABT-737 and ABT-263. These inhibitors bind to the hydrophobic groove on Bcl-2 and Bcl-xL [24,5658], effectively neutralizing the effects of pro-survival protein overexpression and sensitizing cells to death. Since Bcl-xL interacts with the ceramide channel through the hydrophobic region of ceramide, the inhibitors should interfere with the interaction if binding involves the hydrophobic groove.

Results from the 2-MeAA3 experiments are shown in Figure 6(A). Whereas the highest concentration of 2-MeAA3 tested (2.7 μM) had no effect on ceramide-induced MOMP, Bcl-xL (0.7 μM) inhibited the ceramide-induced MOMP by approximately 50%. 2-MeAA3 produced a dose-dependent recovery of ceramide-induced MOMP, indicating that 2-MeAA3 might bind to the same site on Bcl-xL as the ceramide channel. Unlike 2-MeAA3, ABT-737 displayed a strong inhibition of ceramide permeabilization at 2.5 and 5 μM, whereas lower concentrations had either no effect or a slightly stimulatory effect on ceramide permeabilization (Figure 6B). In the presence of 0.7 μM Bcl-xL, ABT-737 produced a dose-dependent recovery of ceramide permeabilization with 50% recovery at approximately 1 μM and complete recovery at 2.5 μM. A strong inhibition of ceramide channels by ABT-737 is evident at 5 μM even in the presence of Bcl-xL. ABT-263, the orally available analogue of ABT-737, was also able to cause a dose-dependent recovery of ceramide permeabilization in the presence of Bcl-xL, but did not inhibit ceramide permeabilization at any of the concentrations tested when Bcl-xL was not present (Figure 6C). Thus the inhibitory effects of Bcl-xL on ceramide channels can be reversed by the addition of Bcl-xL-targeted inhibitors that bind to the hydrophobic groove, suggesting competition and that the hydrophobic groove is likely to be the site at which Bcl-xL binds to the ceramide channel.

Figure 6 The hydrophobic groove of Bcl-xL is involved in binding to ceramide channels

Inhibitors known to bind to the hydrophobic groove of Bcl-xL were assayed for ability to prevent Bcl-xL binding to ceramide channels in isolated rat liver mitochondria. Increasing concentrations of 2-MeAA3 (A), ABT-737 (B) or ABT-263 (C) were tested with C16-Cer alone (10 μg) and the combination of C16-Cer and Bcl-xL (0.7μM).

Molecular docking of ceramides and inhibitors to Bcl-xL were conducted to determine the lowest-energy site and mode of binding on Bcl-xL. The region of lowest-energy binding was first determined probing the whole surface of Bcl-xL and the hydrophobic groove was the best, or near best, scored region for all ligands tested. The top ten poses for each of the ligands are shown in Figure 7 and the average energy of these were calculated and are shown as a horizontal bar. The more negative the number, the stronger the interaction. ABT-737 and ABT-263 had the lowest-energy score, and therefore best binding, followed by C16-Cer. Especially noteworthy is the weaker average binding of C2-Cer compared with C16-Cer, consistent with the experimental results. To explore further the potential interaction between Bcl-xL and C16-Cer, the lowest-energy pose was subjected to global system energy minimization to allow both the protein and the ligand to flex. These results were considered a plausible prediction of the interaction between ceramide and Bcl-xL (Figure 8B).

Figure 7 E Score 2 of ten best poses of molecular docking simulations

Docking simulations of Bcl-xL to C16-Cer (C16), C2-Cer (C2), ABT-737, ABT-263, 2-MeAA3 or palmitoylserine (P-Serine) yielded structures of various degrees of interaction energy. The lower the E Score, the better the interaction. The average of the ten best interacting structures (poses) for each ligand is shown as a horizontal line.

Figure 8 Molecular structures of C16-Cer bound to Bcl-xL from docking simulations and possible interaction with a ceramide channel

(A) How Bcl-xL might bind to a ceramide molecule at the end of a ceramide column that is part of a ceramide channel. One of the top poses of C16-Cer bound to Bcl-xL leaves one chain free. This was lined up with a segment of the proposed structure of a ceramide channel, showing how Bcl-xL might bind to the end of one ceramide column, disrupting the column's interaction with the adjacent phospholipid bilayer. (B) C16-Cer nestled in the hydrophobic grove of Bcl-xL. This is one of the top ten poses after energy minimization.

One of the top ten poses shows ceramide binding to Bcl-xL with only one chain (Figure 8A). When this is placed on top of a segment of the model of a ceramide channel, this provides mechanistic support for a hypothetical mechanism by which Bcl-xL destabilizes a ceramide channel.

Molecular dynamics simulations [26,27] indicate that the ceramide channel forms an hourglass shape in the membrane, allowing the channel to properly interface with the phospholipid bilayer and thus avoiding exposure of apolar chains to the aqueous environment (Supplementary Figures S2 and S3 at This results in a curvature in the ceramide columns forming the channel that is positive in the direction normal to the plane of the membrane [27]. This curvature reduces the distortion of the phospholipids in close proximity to the ceramide channel. Thus interfacing the ceramide molecules that would tend to have their long axis parallel with the membrane plane with the phospholipids whose long axis is normal to that plane requires a compromise distortion of both to minimize the overall energy level. This low-energy compromise might be interfered with by the interaction with Bcl-xL. Bcl-xL could bind at this interface, displacing the local phospholipids and allowing the ceramide column to relax by reducing the positive curvature. Evidence from the present study indicates that the hydrophobic groove of Bcl-xL is interacting with the hydrophobic chains of the ceramide molecules in a channel, chains that are buried within the membrane apolar environment. This is consistent with previous results by Siskind et al. [24], who demonstrated that a Bcl-xL deletion mutant lacking the transmembrane domain is unable to disassemble ceramide channels, suggesting that the membrane-active form of Bcl-xL is required for ceramide channel regulation. Thus the binding of Bcl-xL to the end of one column and resulting relaxation of that column (Supplementary Figure S4 at would result in a mismatch with the curvature of adjacent ceramide columns and thus the generation of mechanical stress. The highly hydrogen-bonded structure of the channel would result in the propagation of the mechanical stress to the rest of the channel (Supplementary Figure S5 at This high-energy structure would then lead to the destabilization of the channel, shifting the dynamic equilibrium towards channel disassembly. This mechanism is consistent with previous results [24] that indicate the formation of a 1:1 complex between Bcl-xL and the channel, requiring that the effects of this binding propagate through the channel structure in an allosteric manner.

As for Bax, the results indicate that the chirality of the molecule's polar region and specifically the accessibility of the amide group are key for Bax binding. This indicates that Bax may also be acting at the interface of the channel and the phospholipid membrane where there is access to the carboxamide group. However, unlike Bcl-xL, Bax binding might favour the positive curvature of the ceramide columns, stabilizing the channel at the channel–membrane interface. This stabilization can result in the enlargement of ceramide channels to an optimal size (Supplementary Figure S6 at Again, these results are consistent with previous findings [31], as the binding of Bax influences the ceramide channel to enlarge to a size that fits optimally with the binding site of Bax. Previous results also indicated that multiple Bax structures bind to a single ceramide channel, presumably all acting to stabilize the ceramide channel in a particular state.

Concluding remarks

Beyond the details, it is worth noting that highly organized lipid assemblies, ceramide channels in this case, take a form dictated by the properties of their monomeric subunits and the environment in which they exist. However, the forms of these superstructures can be modified by interactions with regulating molecules, in this case Bcl-2 family proteins. Thus cells can control these superstructures at various levels: by controlling the supply of specific monomers at the membrane or cell compartment of interest, by changing the lipid environment of the membrane, and by producing proteins (or other regulating factors) that alter the form or stability of the final structure. Perhaps, in thinking about lipids, one must go beyond thinking about lipid phases, fluidity and phase transitions. Lipids can form superstructures with unique properties and the ability to be regulated much like the structures formed by proteins.


Meenu Perera designed the study, purified the proteins used, performed all mitochondrial experiments, analysed data and wrote the paper. Shang Lin aided in repeating a portion of the mitochondrial experiments. Yuri Peterson designed and ran molecular docking simulations and analysed output data. Alicja Bielawska, Zdzislaw Szulc and Robert Bittman provided ceramide analogues. Marco Colombini provided intellectual guidance and wrote the paper. All authors contributed to editing the paper before submission.


This work was supported by the National Science Foundation [grant number MCB-1023008 (to M.C.)], National Institutes of Health [grant number HL-083187 (to R.B.)], National Institutes of Health/National Cancer Institute [grant number P01 CA097132-01A1 (to A.B.)] and National Institutes of Health/National Center for Research Resources South Carolina COBRE (Centers of Biomedical Research Excellence) [grant number P20 RR017677 (to Z.M.S.)].


We extend our gratitude to Richard Youle and Antonella Antignani for providing the plasmid we used to express full-length Bax and for assistance with the purification procedure. We also extend our heartiest thanks to Marie Hardwick for the full-length Bcl-xL plasmid and Heather Lamb for recommending the modifications to the purification procedure. We thank Richard Trager for technical assistance with the molecular simulations. Finally, we thank Vidyaramanan Ganesan for isolating rat liver mitochondria.

Abbreviations: C2-Cer, N-acetyl-D-erythro-sphingosine; C16-Cer, N-palmitoyl-D-erythro-sphingosine; C18:1-Cer, N-oleoyl-D-erythro-sphingosine; C20-Cer, N-arachidoyl-D-erythro-sphingosine; C24-Cer, N-lignoceroyl-D-erythro-sphingosine; DNP, 2,4-dinitrophenol; IMS, intermembrane space; 2-MeAA3, 2-methoxyantimycin A3; MOE, Molecular Operating Environment; MOM, mitochondrial outer membrane; MOMP, MOM permeabilization


View Abstract