Ceramides bind VDAC2 to trigger mitochondrial apoptosis

Abstract

Ceramides draw wide attention as tumor suppressor lipids that act directly on mitochondria to trigger apoptotic cell death. However, molecular details of the underlying mechanism are largely unknown. Using a photoactivatable ceramide probe, we here identify the voltage-dependent anion channels VDAC1 and VDAC2 as mitochondrial ceramide binding proteins. Coarse-grain molecular dynamics simulations reveal that both channels harbor a ceramide binding site on one side of the barrel wall. This site includes a membrane-buried glutamate that mediates direct contact with the ceramide head group. Substitution or chemical modification of this residue abolishes photolabeling of both channels with the ceramide probe. Unlike VDAC1 removal, loss of VDAC2 or replacing its membrane-facing glutamate with glutamine renders human colon cancer cells largely resistant to ceramide-induced apoptosis. Collectively, our data support a role of VDAC2 as direct effector of ceramide-mediated cell death, providing a molecular framework for how ceramides exert their anti-neoplastic activity.

Fig. 1

A chemical screen for mitochondrial ceramide-binding proteins yields VDAC1 and -2. a Structure of the photoactive and clickable C₁₅-ceramide analog, pacCer. b Mitochondria isolated from HeLa cells were incubated with liposomes containing increasing amounts of pacCer, UV irradiated, and then click reacted with AF647-N₃. Samples were processed for SDS-PAGE, subjected to in-gel fluorescence (IGF, red), and stained with Coomassie blue (CB, blue). p33•Cer denotes a prominently photolabeled protein band of ~33 kDa. c Strategy for the identification of p33•Cer. d p33•Cer was purified from pacCer-labeled and TAMRA/biotin click-reacted mitochondria using NeutrAvidin-beads, imaged by IGF, excised from the gel, digested by trypsin, and then identified by LC-MS/MS. Data from two independent experiments revealed that p33•Cer corresponds to VDAC1 and VDAC2. e Specificity of anti-VDAC antibodies was validated by immunoblotting of mitochondria isolated from HeLa cells treated with non-silencing (siNS) or VDAC-targeting siRNAs (siVDAC1, siVDAC2). Note that the anti-VDAC3 antibody cross-reacts with VDAC2. f Mitochondria isolated from siVDAC1/2-treated HeLa cells were photolabeled with pacCer, click-reacted with AF647-N₃, and subjected to IGF analysis followed by CB staining. g Fractions obtained during affinity purification of p33•Cer were subjected to SDS-PAGE, transferred on nitrocellulose, analyzed by on-blot-fluorescence (OBF) and probed with antibodies against VDAC1, TOM40, and p60-Mito. h Fractions obtained during affinity purification of p33•Cer were processed as in g and probed with anti-VDAC2 and anti-VDAC3 antibodies. T total mitochondria extract, FT flow-through, W wash, E eluate

Fig. 2

MD simulations uncover a putative ceramide-binding site on VDAC1 and -2. a Ceramide head group contact occupancy of mouse VDAC1, VDAC2, and VDAC3 in an OMM model containing 5 mol% ceramide, with 1.0 corresponding to a ceramide contact during the entire simulation time. A threshold of 15% occupancy, based on the occupancies of non-binding site residues, is indicated by a dotted blue line. VDAC1 and VDAC2 have a clear ceramide-binding site, comprising residues 58–62, 71–75, and 81–85; this site is lacking in VDAC3. b Sequence alignment revealing the position of a bilayer-facing Glu residue in VDAC1 (E73) and VDAC2 (E84), which is replaced by Gln in VDAC3 (Q73). c Stills from an MD simulation, showing the approach and binding of a ceramide molecule to VDAC1 in close proximity of the bilayer-facing Glu residue in its deprotonated state. Protein surface colors mark polar (green), apolar (white), cationic (blue), or anionic (red) residues. d Space-filling and wireframe models of VDAC1, VDAC1^E73Q, VDAC2, and VDAC2^E84Q with deprotonated E73/E84. Indicated are the volumes for which there is ceramide occupancy greater than 10% (orange) or cholesterol occupancy greater than 20% (yellow). e Distribution of the durations of ceramide contacts with VDAC1, VDAC1^E73Q, VDAC2, and VDAC2^E84Q at the preferred binding site as in d. The y-axis indicates the fraction of the total system time spent in binding events of the duration indicated by x. Summing all points’ y-values yields the fraction of total simulation time when ceramide was bound

Fig. 3

Ceramide binding by VDACs relies on the protonation state of the bilayer-facing Glu. a Space-filling and wireframe models of VDAC1 and VDAC2 simulated with the bilayer-facing Glu residue in a protonated or deprotonated state. Indicated are the volumes for which there is ceramide occupancy greater than 10% (orange) or cholesterol occupancy greater than 20% (yellow). b Distribution of the durations of ceramide contacts with VDAC1 and VDAC2 at the preferred binding site as in a. The y-axis indicates the fraction of the total system time spent in binding events of the duration indicated by x. Summing all points’ y-values yields the fraction of total simulation time when ceramide was bound. c Human VDAC1 and VDAC1^E73Q were produced in E. coli, purified, reconstituted in liposomes, and then photolabeled at the indicated pH with pacCer added from an ethanolic stock. Samples were click-reacted with AF647-N₃, subjected to SDS-PAGE, and analyzed by IGF and CB staining. d Quantitative analysis of relative pacCer photolabeling efficiencies of reconstituted VDAC1 treated as in c. Data are means ± s.d.; n = 4; **p < 0.01 by two-tailed paired t-test. Source data

Fig. 4

The bilayer-facing Glu is a critical determinant of pacCer photolabeling of VDACs. a Human VDAC1, VDAC1^E73Q, VDAC2, and VDAC2^E84Q were produced in E. coli, purified, reconstituted in liposomes, and incubated for 30 min at 37 °C with liposomes containing 1 mol% of pacCer or photoactive and clickable analogs of diacylglycerol (pacDAG), phosphatidylcholine (pacPC), phosphatidylethanolamine (pacPE), or cholesterol (pacChol). Samples were UV irradiated and then click-reacted with AF647-N₃, subjected to SDS-PAGE, and analyzed by IGF and CB staining. b Quantitative analysis of relative labeling efficiencies of reconstituted VDAC1, VDAC1^E73Q, VDAC2, and VDAC2^E84Q with pacLipids as indicated in a. Data are means ± s.d.; n ≥ 3. Source data

Fig. 5

Competitive inhibition of pacCer photolabeling of VDACs by C₁₆-ceramide. a VDAC1, VDAC1^E73Q, VDAC2, and VDAC1^E84Q proteoliposomes were photolabeled with the indicated amount of pacCer or pacChol added from ethanolic stocks. Next, samples were click-reacted with AF647-N₃, subjected to SDS-PAGE, and analyzed by IGF and CB staining. b VDAC1, VDAC1^E73Q, VDAC2, and VDAC1^E84Q proteoliposomes were photolabeled with 0.2 nmol pacCer or pacChol in the presence of the indicated amounts of natural C₁₆-ceramide (Cer) added from ethanolic stocks and then processed as in a. Relative labeling efficiencies were quantified and expressed as % of control (0.2 nmol pacCer or pacChol in the absence of Cer). Data are means ± s.d.; n = 3; *p < 0.05, **p < 0.01, and ***p < 0.001 by two-tailed paired t-test. Source data

Fig. 6

VDAC2 removal disrupts ceramide-induced apoptosis. a Schematic outline of ceramide transfer protein CERT, mitoCERT, and mitoCERTΔSTART. MitoCERT was created by swapping the Golgi-targeting pleckstrin homology domain of CERT against the OMM anchor of AKAP1. Removal of the ceramide transfer or START domain yielded mitoCERTΔSTART. All three proteins bind the ER-resident protein VAP-A via their FFAT motif (F). Cer ceramide, PI(4)P phosphatidylinositol-4-phosphate, TGN trans-Golgi network. b Ceramides (Cer) are synthesized through N-acylation of long chain bases (LCB) by ceramide synthases (CerS) on the cytosolic surface of the ER and require CERT-mediated transfer to the Golgi for metabolic conversion into sphingomyelin (SM) by a Golgi-resident SM synthase (SMS). Expression of mitoCERT causes a diversion of this biosynthetic ceramide flow to mitochondria, triggering Bax-dependent apoptosis. c Wild type (WT), VDAC1-KO (ΔVDAC1), VDAC2-KO (ΔVDAC1), and VDAC1/2 double KO (ΔVDAC1/2) human colon cancer HCT116 cells were transfected with empty vector (EV), Flag-tagged mitoCERT, or Flag-tagged mitoCERTΔSTART. At 24 h post transfection, cells were processed for immunoblotting with antibodies against PARP1, the Flag-epitope, VDAC1, VDAC2, and β-actin. The percentage of PARP1 cleavage was quantified. Data are means ± s.d.; n = 3; *p < 0.05 and **p < 0.01 by two-tailed paired t-test. Source data

Fig. 7

Ceramide-induced apoptosis critically relies on Glu84 in VDAC2. a WT and ΔVDAC1/2 HCT116 cells stably transduced with HA-tagged VDAC1, VDAC1^E73Q, VDAC2, or VDAC2^E84Q were transfected with empty vector (EV), Flag-tagged mitoCERT, or Flag-tagged mitoCERTΔSTART. At 24 h post transfection, cells were processed for immunoblotting with antibodies against PARP1, cleaved caspase-3 (Casp3), the Flag-epitope, the HA-epitope, and β-actin. FL full-length, CL cleaved. b Quantitative analysis of PARP1 cleavage in cells treated as in a. Data are means ± s.d.; n = 4; *p < 0.05, **p < 0.01 and ***p < 0.001 by two-tailed paired t-test. c Quantitative analysis of cleaved Casp3 in cells treated as in a. Data are means ± s.e.; n = 2. Source data