Neuron-specific Bcl-2 homology 3 domain-only splice variant of Bak is anti-apoptotic in neurons, but pro-apoptotic in non-neuronal cells.

We have identified and characterized N-Bak, a neuron-specific isoform of the pro-apoptotic Bcl-2 family member Bak. N-Bak is generated by neuron-specific splicing of a novel 20-base pair exon, which changes the previously described Bak, containing Bcl-2 homology (BH) domains BH1, BH2, and BH3, into a shorter BH3-only protein. As demonstrated by reverse transcription-polymerase chain reaction and RNase protection assay, N-Bak transcripts are expressed only in central and peripheral neurons, but not in other cells, whereas the previously described Bak is expressed ubiquitously, but not in neurons. Neonatal sympathetic neurons microinjected with N-Bak resisted apoptotic death caused by nerve growth factor (NGF) removal, whereas microinjected Bak accelerated NGF deprivation-induced death. Overexpressed Bak killed sympathetic neurons in the presence of NGF, whereas N-Bak did not. N-Bak was, however, still death-promoting when overexpressed in non-neuronal cells. Thus, N-Bak is an anti-apoptotic BH3-only protein, but only in the appropriate cellular environment. This is the first example of a neuron-specific Bcl-2 family member.

During development, two opposite processes, proliferation and naturally occurring cell death (apoptosis), regulate cell number in almost all tissues and organs (1). In the developing nervous system, for example, 30 -80% of the initially produced neurons die, mostly due to deficiency of neurotrophic factors that neutralize the death program in neurons (2,3). The cells of an organism retain a potential to die apoptotically during their entire lifetime (4). Apoptotic pathways are therefore delicately balanced and controlled positively and negatively at several levels (5,6). The life and death decisions of a cell are critically controlled by the proteins of the Bcl-2 family that are either anti-apoptotic or pro-apoptotic. When overexpressed, antiapoptotic members protect cells against death stimuli, and their lack in vivo promotes developmental death or sensitivity to death stimuli. Conversely, pro-apoptotic Bcl-2 family members kill the cells even in the presence of life-promoting stimuli, and their deficiency reduces apoptotic death (7). All anti-apoptotic Bcl-2 family members (Bcl-2, Bcl-x L , Bcl-w, Mcl-1, A1, and Boo/DIVA) contain four conserved BH 1 domains (BH1, BH2, BH3, and BH4), whereas pro-apoptotic members contain either three BH domains (BH1, BH2, and BH3) (Bax, Bak, and Mtd/Bok) or only the BH3 domain (Bad, Bid, Bik, Blk, Hrk, Bim, Rad9, and Noxa) (7)(8)(9)(10). Interestingly, the recently discovered Bcl-G contains only BH3 and BH2 domains (11).
How the Bcl-2 family proteins work is still poorly understood. According to the current understanding, pro-apoptotic members containing several BH domains (Bax and Bak), when activated by a death stimulus, generate pores to the mitochondrial outer membrane that lead to the release of cytochrome c and other mitochondrial molecules to the cytoplasm, where they trigger activation of caspases at the apoptosome (5, 6). Once activated above certain threshold levels, caspases irreversibly execute cell death (6,12). In addition, heterodimerization between certain pro-and anti-apoptotic Bcl-2 family proteins nullifies the activity of the partners, and the fate of the cell is determined by the member that is in excess (7). BH3-only proteins are believed to regulate or modulate the activity of multi-BH domain members. Thus, Bid activates Bax (13,14), whereas Bad inactivates Bcl-x L (15), with the net result being cell death in both cases. Alternatively spliced transcripts have been described for many Bcl-2 family members, with the activity of the protein isoforms remaining unchanged in most cases (16). However, BH3-only splice variants encoding proteins with pro-apoptotic activity have been described for the anti-apoptotic Bcl-x (17) and 19) and also for Bcl-G (11). Here we describe a BH3-only splice variant of the pro-apoptotic Bak (20 -24) and show that it is expressed exclusively in neurons and encodes a protein isoform that is anti-apoptotic in neurons, but promotes death in non-neuronal cells.

Cloning and Analysis of Bak Splice Variants-Full-length N-Bak and
Bak cDNAs were generated by RT-PCR from P1 mouse brain RNA, inserted into the pCR3.1 expression vector (Invitrogen, Groningen, The Netherlands), and verified by sequencing. To check whether the expression plasmids produce the respective proteins in cells, COS-7 cells in semiconfluent 10-cm dishes were transfected with Bak-or N-Bak-encoding expression plasmids or with the empty pCR3.1 vector (5 g/dish) using Fugene transfection reagent (Roche Molecular Biochemicals). To protect cells against death caused by overexpression of pro-apoptotic proteins, plasmid encoding human Bcl-x L (5 g/dish) was cotransfected. Also, plasmid pGreenLantern-1, encoding green fluorescent protein (GFP) (0.5 g/dish; Life Technologies, Inc.), was included to reveal transfection efficiency. The next day, cells were lysed in phosphatebuffered saline containing 10 mM KCl, 2 mM EDTA, 1% SDS, and protease inhibitors (Roche Molecular Biochemicals) and analyzed by Western blotting with anti-Bak antibodies (65606E; Pharmingen, San Diego, CA) (see Fig. 3A). The peptide sequence used to generate these antibodies is present in both Bak and N-Bak (25). Some of the transfected cells were fixed with 4% paraformaldehyde in phosphate-buffered saline, permeabilized with 0.5% Triton X-100, and stained with the same anti-Bak antibodies. Both the Bak-and N-Bak-transfected cultures, but not the vector-transfected or untransfected cultures, contained strongly Bak-immunoreactive cells (data not shown). To demonstrate endogenous N-Bak protein, newborn mouse brain, cultured SCG or hippocampal neurons, or Neuro2A neuroblastoma cells were lysed in buffer containing either 2% Triton X-100 and 1% SDS or 1% SDS and 8 M urea. Either Bak proteins were immunoprecipitated from the lysates with the anti-Bak antibodies (Pharmingen 65606E), or the crude lysate was used; and the filter was probed with the same antibody. In all samples, only Bak (but not N-Bak) protein was visible (data not shown), whereas both N-Bak and Bak proteins were detected in control COS-7 cells transiently overexpressing N-Bak and Bcl-x L (similar to the results shown in Fig. 3A). The same result was obtained with anti-Bak antibodies from Oncogene Research Products (AM04; Darmstadt, Germany) or from Santa Cruz Biotechnology (H-211; Santa Cruz, CA) (data not shown).
To show that the non-neuronal cells overexpressing N-Bak die apoptotically, HeLa cells were transiently transfected with expression vector for GFP-N-Bak or with the empty pEGFP-C1 vector. 2 h after transfection, GFP-N-Bak-transfected cells exhibited weak fluorescence, became round, and began to detach from the substrate, whereas GFPexpressing cells remained healthy and flat. The cells were fixed with 4% paraformaldehyde in phosphate-buffered saline and stained with 1 M 4,6-diamidino-2-phenylindole. Images were acquired on an Olympus AX 70 Provis microscope and analyzed by Adobe Photoshop software. For the DNA ladder assay, HeLa cells were transiently transfected with expression plasmids encoding N-Bak or GFP. After 3 h, DNA was prepared and analyzed for the presence of internucleosomal degradation fragments according to a published protocol (26).
RT-PCR and RNase Protection-Total RNAs from different rat and mouse tissues or from cultured purified neurons or non-neuronal cells were isolated with Trizol reagent (Life Technologies, Inc.). RNAs from human tissues were obtained from CLONTECH. Rat cortical and hippocampal neurons were cultured as described (27). First strand cDNAs were synthesized using oligo(dT) 15 or dN 6 random primers (Roche Molecular Biochemicals) with enhanced avian reverse transcriptase (Sigma) or Superscript II (Life Technologies, Inc.). 2-l cDNA aliquots were processed by PCR with the High Fidelity PCR system (Roche Molecular Biochemicals), and Bak-specific primers (forward, 5Ј-CCACCATGGCA-TCTGGACAAGGACCAG; and reverse, 5Ј-TCATGATCTGAAGAATCT-GTGTACC) were used to amplify full-length N-Bak and Bak cDNAs. For nested PCR, a different pair of N-Bak-specific primers was used after amplifications with the primers shown above: forward, 5Ј-TTGC-CCAGGACACAGAGGAGGT; and reverse, 5Ј-GAATTGGCCCAACA-GAACCACACC. Both primers were localized to different exons to distinguish amplification products from cDNA and genomic DNA. In addition, with either pair of primers, fragments of the size of Bak or N-Bak cDNAs were not amplified from mouse genomic DNA (data not shown). PCR was performed for 35 cycles with the following program: 95°C for 45 s, 60°for 45 s, and 72°C for 45 s. PCR products (10 -20 l) were separated on 1.5% agarose gel to reveal the 20-bp difference in the Bak fragments. Fragments of interest were excised and cloned into the pCR2.1 vector or the pCR3.1 expression vector. Nucleotide sequences were verified by sequencing. To prepare neurons completely devoid of non-neuronal cells, 1000 -2000 neurons dissociated from neonatal mouse SCG were seeded onto a restricted area on the culture dish, and all non-neuronal cells were manually killed by a micromanipulatordriven needle. Absence of non-neuronal cells was then checked for several days, and the few non-neuronal cells that remained in the culture were killed. This approach was not feasible for hippocampal cultures. Therefore, neurons from 5-day cultures of E16 mouse hippocampi were individually picked up with glass pipettes with the help of a micromanipulator under visual control with an inverted microscope essentially as described (28). About 30 neurons were individually picked up with separate glass pipettes with an aperture slightly bigger than the cell diameter and collected into Trizol reagent. RNA isolation and RT-PCR were performed as described above. No fragments were amplified from the culture medium collected in the same way as the neurons (data not shown). The RNase protection assay (RPA) was performed as described (29). Full-length N-Bak cDNA was used to generate 32 P-labeled antisense RNA probe. The assay resulted in one fragment of 572 bp for N-Bak and two fragments of 322 and 230 bp for Bak. Only the 572-and 230-bp fragments are shown in Fig. 2. For a positive control, the mouse ␤-actin riboprobe (Ambion Inc., Austin, TX) was used.
Transfection of Cultured Primary Cells-Nuclei of the SCG neurons, cultured for 5-6 days with 30 ng/ml NGF, were pressure-injected under direct visual control with expression plasmids encoding Bak or N-Bak or with the empty pCR3.1 vector, all 50 ng/l. All injection solutions contained also 10 ng/l pGreenLantern-1. A Model MMO-220 micromanipulator (Narishige International Ltd., London) and a Model 5246 Transjector (Eppendorf Scientific, Westbury, NY) were used for injection. Neurons were grown further with NGF or in NGF-free medium with function-blocking anti-NGF antibodies (Roche Molecular Biochemicals). Initial neurons surviving the procedure were counted 3-4 h later. 50 -100 neurons were successfully injected with each plasmid combination in every experiment. To later follow all injected neurons individually, the positions of the injected neurons were mapped according to the grid scratched in the bottom of the dish. Healthy fluorescent neurons with phase-bright cytoplasm and an intact neuritic tree, identified according to the fluorescence and the map, were counted daily and expressed as a percentage of the uninjected cells. Few neurons that lost fluorescence during experiment were subtracted from the initial neurons. Uninjected neurons were counted from one dish. Non-neuronal cells from dissociated P1 mouse SCG were cultured in Dulbecco's modified Eagle's medium containing 10% fetal calf serum without neurotrophic factors until all neurons were dead and injected as described above for neurons. Initial cells were counted 2 h (40 -50 cells for every plasmid) and living fluorescent cells 18 -20 h after injection. During that period, injected cells neither moved nor divided significantly, so they were recognizable according to the fluorescence and the map. Experiments were repeated three times on independent cultures. In every repeat, all treatment groups presented in Fig. 4 (A-C) were studied collectively in the same culture. Statistical significance of the data was analyzed by one-way analysis of variance, followed by Tuckey's post hoc test at a significance level of ␣ ϭ 0.05.
For intracellular localization studies, GFP was fused to the N terminus of full-length N-Bak using the pEGFP-C1 vector (CLONTECH). N-Bak cDNA without a putative transmembrane domain (lacking amino acids 125-150) was generated by PCR and cloned into the pEGFP-C1 vector. Functional properties of this deletion mutant together with other mutated forms of N-Bak are currently being studied in our laboratory. Neurons microinjected with these constructs (50 ng/l) were grown with or without NGF for 48 h and fixed with 4% paraformaldehyde in phosphate-buffered saline. In some experiments, neurons were treated with 1 M Mitotracker Red CMXRos (Molecular Probes, Inc.) for 15 min and washed extensively before fixation. Although a classical punctate pattern of Mitotracker staining was obtained with the non-neuronal cells in culture, the neurons showed more broad and diffuse staining where individual mitochondria were difficult to discern. Images were acquired on an OZ confocal laser scanning microscope (Noran Instruments Inc., Middleton, WI) and processed with Adobe Photoshop software.

N-Bak, a Novel Isoform of Bak, Is a BH3-only Protein-When
verifying the nucleotide sequences of RT-PCR fragments of Bak from mouse brain, we noticed a 20-bp sequence, GCCAGCAG-CAACATGCACAG (Fig. 1A), not present in the published mouse Bak sequence (23). We cloned and sequenced full-length Bak cDNA from mouse brain. The 20-bp insert was found in several sequenced cDNA clones. Also some minor differences from the published Bak sequence were detected in all clones: there is an additional guanine between positions 147 and 148 as well as two additional cytidines between positions 153 and 154 of the published mouse Bak sequence. We do not know whether these minor differences result from sequencing errors of Ulrich et al. (23) or whether they represent polymorphisms or mutations in the Bak gene of the 3T3 cells used by Ulrich et al. to clone it. In the predicted Bak protein deduced from the corrected nucleotide sequence, residues 50 and 51 are both alanines, but not arginine and proline, as published by Ulrich et al. (23), followed by an additional alanine absent in the published sequence (Fig. 1D).
The 20-bp sequence is inserted at position 344 (corrected nucleotide numbering) of Bak cDNA (Fig. 1A), which corresponds exactly with the junction of exons 4 and 5 of the mouse Bak gene (23), suggesting that the 20-bp insert is a hitherto undescribed exon. We determined a partial sequence of the intron between exons 4 and 5 of the Bak gene. The 20-bp sequence was found in the intronic sequence flanked by intronexon junction sequences corresponding to the GT-AG rule (Fig.  1C). In addition, the 3Ј-splice site of the 20-bp exon deviated from the respective consensus sequence (Pyr) 12 CAG (where Pyr is pyrimidine), with several purines interrupting the polypyrimidine tract (Fig. 1C). The majority of the alternatively spliced exons have weak 3Ј-splice sites with higher purine content compared with the 3Ј-sites of constitutively spliced exons (31,32). We designated this exon as exon N (for neuron-specific exon; see below) and the transcript using this exon as N-Bak. The corrected genomic organization of the mouse Bak gene and the two transcripts generated from it by alternative use of exon N are schematically presented in Fig. 1.
Use of exon N would cause a translational frameshift, resulting in a changed amino acid sequence and a truncated protein due to a premature stop codon (Fig. 1D). The predicted protein translated from N-Bak contains 150 amino acids, with a calculated molecular mass of 16.4 kDa and a calculated pI of 4.48. The previously described Bak isoform has 208 residues, with a calculated molecular mass of 23.3 kDa and a calculated pI of 6.08. The novel C-terminal amino acid stretch RPAATCTAYL-RVASAGAAWWLSWALATVWPCTSTSVV of N-Bak (Fig. 1D) has no homology to any of the known proteins. As predicted by the Sosui and TopPred2 programs, 24 C-terminal residues of this novel sequence (VASAGAAWWLSWALATVWPCTSTS) may form a transmembrane ␣-helix. No other known structural motifs were found in this novel sequence by the PSORT II program. Use of exon N would lead to the change of the BH1 domain of the Bak protein into a different amino acid sequence. The BH2 as well as transmembrane domains would not be translated, whereas the BH3 domain would remain unchanged (Fig. 1D). Thus, use of exon N would convert the three BH domain-containing Bak protein into a BH3-only protein with a novel putative transmembrane domain. We designated the short protein isoform (encoded by N-Bak) as N-Bak and the previously known longer protein isoform as Bak.
A similar 20-bp sequence (GCCAGCAGCAACACCCACAG) was found in four independent Bak cDNA clones from human brain with two nucleotide differences from mouse exon N. This human sequence is inserted at the position identical to that in mouse Bak cDNA, leading to the same changes in amino acid sequence as in mouse N-Bak: a truncated protein of 150 amino acids, a novel C-terminal amino acid sequence (RPAATPTA-CLRVASIGAVWWLFWASATVWPYTSTSMA) with a predicted transmembrane domain, and a lack of BH1 and BH2 domains.
N-Bak Transcripts Are Expressed in Neurons, but Are Absent in Non-neuronal Cells-Expression of Bak splice variants in different tissues was studied by RT-PCR and RPA. RT-PCR analysis showed that Bak was expressed in all studied rat tissues, whereas N-Bak was expressed exclusively in the nervous tissue (Fig. 2A). Identical results were obtained when RNAs from human tissues were analyzed by RT-PCR (data not shown). RPA also revealed expression of Bak transcripts in all adult mouse tissues analyzed, whereas N-Bak was detected only in brain. The levels of both Bak and N-Bak transcripts were rather similar in different brain regions (Fig. 2B). Furthermore, Bak and N-Bak transcripts were differently regulated during mouse brain development. Low levels of N-Bak were present at E13, and the levels increased in the late embryogenesis and early postnatal days, with the peak being around birth (Fig. 2C). In contrast to this, the levels of Bak were high in E13 brain and decreased gradually during development (Fig. 2C).
Both transcripts of Bak were detected in cultured cells of rat cerebral cortex and hippocampus ( Fig. 2A) that contained neurons as well as non-neuronal cells. In contrast, the non-neuronal cells from rat cerebral cortical culture expressed only Bak, but not N-Bak. Similarly, the rat sciatic nerve, known to contain mostly Schwann cells, expressed Bak, but not N-Bak ( Fig.  2A). To clarify the cell-type specificity of expression of the two Bak transcripts, we manually separated cultured neonatal mouse SCG or E16 mouse hippocampal neurons from all nonneuronal cells (see "Experimental Procedures"). Neuron-free cultures of non-neuronal cells from both sources were also prepared. We then analyzed expression of Bak splice variants in purified cell populations by RT-PCR. Both SCG and hippocampal neurons expressed only N-Bak, but not Bak. Conversely, only Bak (but not N-Bak) was expressed in non-neuronal cells (Fig. 2D). The experiments with purified SCG neurons were repeated three times with identical results. Absence of Bak in neurons was further verified with totally nonneuronal cell-free SCG neurons by nested PCR. Bak message was still not detected (data not shown). Although we did not analyze other neuronal populations, it is probable that the Bak transcript in brain tissues (Fig. 2C) may be of glial origin. Thus, expression of N-Bak is strictly neural tissue-specific and, at least in SCG and hippocampus, strictly neuron-specific, whereas Bak is expressed almost ubiquitously (20, 23), but is absent (or below the detection limit) in the neurons. Exon N, as well as neuron-specific expression of N-Bak, is conserved in the mouse, rat, and human species. However, as our attempts to demonstrate endogenous N-Bak protein failed (see "Experimental Procedures"), we do not have evidence that endogenous N-Bak mRNA is translated into protein in neurons.

N-Bak Is a Survival-promoting Protein in Primary Neurons, but a Death-promoting Protein in Non-neuronal Cells-To
study the functional activity of N-Bak in its natural cellular environment, we microinjected cultured neonatal mouse sympathetic SCG neurons with the expression plasmid encoding N-Bak or Bak and maintained the neurons further with or without NGF. Cultured neonatal SCG neurons are known to die apoptotically when deprived of NGF (33). The respective proteins were produced from the corresponding expression plasmids in transiently transfected COS-7 cells as shown by Western blotting (Fig. 3A). An ϳ28-kDa band was greatly enhanced in Bak-transfected cells, comigrating with the endogenous Bak in COS-7 cells, whereas a smaller band of ϳ22 kDa was revealed only in N-Bak-transfected cells (Fig. 3A). For some unknown reason, both proteins migrated somewhat slower than predicted by their amino acid sequences. N-Bak was also produced by the injected expression plasmid in the neurons, as shown by staining the injected neurons with anti-Bak antibodies (Fig. 3B). As shown previously (34), overexpressed Bak rapidly killed the neurons even in the presence of NGF (Fig. 4A). In contrast, 93.7 Ϯ 2.4% (mean Ϯ S.E.) of the NGF-maintained N-Bak-expressing neurons remained healthy, which does not differ significantly from vector-injected or uninjected neurons (Fig. 4A). The same percentage of neurons survived when the concentration of injected N-Bak-encoding plasmid was raised from 50 to 100 ng/l (data not shown). When deprived of NGF, 19.4 Ϯ 6.7% of the uninjected neurons and 17.8 Ϯ 8.6% of the vector-injected neurons remained alive 72-75 h after injection (Fig. 4B). By that time, overexpressed Bak had killed almost all neurons (Fig. 4B), as reported previously (21). In contrast, 78.0 Ϯ 2.2% of the N-Bak-expressing neurons resisted NGF removal, which is statistically significantly different from the vector-injected (p Ͻ 0.001) and uninjected (p Ͻ 0.001) neurons (Fig. 4B). A similar percentage of neurons were maintained alive when the concentration of the N-Bak-encoding plasmid was raised to 100 ng/l (data not shown). Thus, in neurons, alternative splicing converts proapoptotic Bak into anti-apoptotic N-Bak. To our knowledge, this is the first demonstration of the anti-apoptotic activity of a BH3-only protein.
The survival-promoting activity of N-Bak was unexpected since we had noted earlier that its overexpression triggers death of transfected COS-7 cells (see "Experimental Procedures"). To address this controversy, we studied the effects of overexpressed Bak isoforms in non-neuronal cells more closely. To this end, we microinjected expression plasmids encoding  (71). D, RT-PCR analysis of Bak and N-Bak expression in cultured neonatal mouse SCG and hippocampal neurons free of non-neuronal cells, non-neuronal SCG and hippocampal cells free of neurons, or undissociated hippocampal tissue (Total). Note that the RT-PCR analyses were not quantitative and show only the presence or absence of the transcripts analyzed. ctx, cortex; ctx g, cortical non-neuronal cells; hip, hippocampus; sp c, spinal cord; sc n, sciatic nerve of adult rat; ad, adult; hrt, heart; kid, kidney; liv, liver; mus, skeletal muscle; spl, spleen; tes, testis; thyr, thyroid; thym, thymus; olf, olfactory bulb; str, striatum; tha, thalamus; col, colliculi superior and inferior; v midbr, ventral midbrain; cbl, cerebellum; med, medulla; total, whole P60 brain; tRNA, yeast tRNA.
Bak, N-Bak, or empty vector into primary non-neuronal cells of dissociated P1 mouse SCG. These cultures, containing mostly glial cells, but also epithelial and fibroblastic cells, expressed mRNA for Bak, but not for N-Bak (Fig. 2D). 24 h after injection, 90.7 Ϯ 1.2% (mean Ϯ S.E.) of the vector-injected control cells had survived, showing that these cells tolerate well the procedure and ectopic DNA. As expected, only 10.6 Ϯ 1.3% of the Bak-injected cells remained alive. N-Bak was an equally potent death inducer in these cells, as only 12.4 Ϯ 0.5% of the injected cells were living 24 h after injection (Fig. 4C). When COS-7 cells or non-neuronal cells from E18 mouse trigeminal ganglia were transfected with plasmid encoding Bak, N-Bak, or empty vector together with a Bcl-x L -encoding plasmid and stained the next day with anti-Bak antibodies, strongly immunopositive cells were found only in Bak-and N-Bak-transfected cultures, but not in vector-transfected cultures (data not shown). However, no Bak-immunopositive cells were found when Bcl-x L was omitted (data not shown), suggesting that Bak and N-Bak killed the cells in the absence of Bcl-x L . Death of non-neuronal cells overexpressing N-Bak was rapid, as already 1-2 h after transfection with expression plasmid encoding GFP-N-Bak, the weakly fluorescent cells become round, detached from the culture substrate, and died soon thereafter. Nuclei of the GFP-N-Bak-transfected HeLa cells exhibited apoptotic features, being fragmented and containing condensed chromatin, as shown by staining with 4,6-diamidino-2-phenylindole, whereas the nuclei of the cells overexpressing GFP were intact and contained diffuse chromatin (Fig. 3C). In addition, a fragmented DNA ladder typical for apoptotic cells was induced in HeLa cells by forced expression of N-Bak, but not GFP (data not shown). Thus, in non-neuronal cells, ectopically expressed N-Bak is a potent apoptosis-promoting protein.
Overxpressed N-Bak Is Associated with Intracellular Membranes in SCG Neurons-Several pro-apoptotic proteins, e.g. Bax, Bad, and Bid, have been shown to be inactive cytosolic proteins in healthy cells, but are translocated to mitochondria during apoptosis (7). To study the intracellular localization of N-Bak in neurons, we microinjected expression vector encoding N-Bak with N-terminally fused GFP into SCG neurons. The N-terminal GFP moiety did not affect the survival-promoting properties of N-Bak in microinjected NGF-maintained or NGFdeprived neurons (data not shown). When examined 48 h after injection, GFP-N-Bak had a clustered distribution both in NGF-maintained and NGF-deprived neurons (Fig. 5), suggesting that GFP-N-Bak is localized to intracellular membranes. Indeed, a similar distribution pattern was observed for GFP-Bcl-2 (Fig. 5), which is known to be associated with intracellular membranes (35). Deletion of the putative transmembrane domain abolished the clustered localization of GFP-N-Bak. Instead, a diffuse cytosolic pattern was observed that was similar to the localization of GFP (Fig. 5). Thus, the predicted transmembrane domain of N-Bak is used in SCG neurons to anchor N-Bak to intracellular membranes. Green fluorescent clusters overlapped with Mitotracker Red CMXRos in GFP-N-Bak-and GFP-Bcl-2-expressing neurons. However, Mitotracker was not specifically concentrated in the clusters of GFP-N-Bak and GFP-Bcl-2 (data not shown). More studies are necessary to ascertain the identity of neuronal membranes associated with clustered localization of GFP-N-Bak. DISCUSSION We describe here N-Bak, a splice variant of Bak, that is generated by the use of a novel exon. Insertion of this exon causes a translational frameshift in Bak mRNA, thereby changing the C-terminal amino acid sequence of the protein product. In N-Bak, only some structural elements of Bak, including the BH3 domain, are preserved, whereas the BH1, BH2, and transmembrane domains, as well as the predicted pore-forming hydrophobic ␣5and ␣6-helices (36), are missing due to a translational frameshift and premature stop codon. Splice variants encoding different protein isoforms have been described for several other Bcl-2 family members. In most cases, as for Bcl-x␤ (37) (46). The functional activity of these splice variants remains mainly unchanged. The pro-apoptotic activity of three splice variants of Bim is increased as the protein is shortened (47). There are, however, a few cases where alternative splicing generates shorter BH3-only variants of multi-BH domain Bcl-2 family proteins, whose activity is converted opposite to that of the longer isoform. So far, such splice variants have been described only for the anti-apoptotic proteins Bcl-x (17) and Mcl-1 (18,19). Thus, the multi-BH domain Bcl-x L and Mcl-1 L isoforms protect against apoptosis, whereas their short BH3-only variants, Bcl-x S (which, however, possesses also a BH4 domain) and Mcl-1 S , promote death. The recently described Bcl-G has also two splice variants: Bcl-G L , which contains the BH3 and BH2 domains and which has only minor effect on apoptosis, and Bcl-G S , a BH3-only protein with strong death-promoting activity (11). In contrast, N-Bak, a BH3-only variant of the pro-apoptotic protein Bak described here, is anti-apoptotic in cell types in which it is endogenously expressed. Thus, the change of activity to the opposite direction may be a general phenomenon when a multi-BH domain Bcl-2 family protein is converted into a BH3-only protein by alternative splicing.
The explanation, as their domain arrangement is similar to that of BH3-only proteins. Based on the determined three-dimensional structures of Bid (36,48) and Bcl-x L with or without a bound BH3 domain of Bak (49 -51), a ligand-receptor model has been proposed for the heterodimerization of pro-and anti-apoptotic Bcl-2 family proteins (8,36). The BH3-only proteins (and also activated Bax and Bak) expose their BH3 domains at the surface (so-called ligand state), whereas in the anti-apoptotic family members, the BH1, BH2, and BH3 domains form a hydrophobic surface groove (a receptor state). Pro-apoptotic members insert their BH3 domains into the surface pocket of the antiapoptotic partners, thereby inactivating them and favoring death. In Bcl-x S and Mcl-1 S , hydrophobic grooves cannot form due to loss of the BH1 and BH2 domains. Instead, Bcl-x S is predicted to take a ligand conformation with an exposed BH3 domain (36). Although we do not know the spatial structure of N-Bak, it probably has a conformation similar to that of Bcl-x S and should be intrinsically pro-apoptotic. Indeed, N-Bak was strongly pro-apoptotic in non-neuronal cells. This is, however, an artificial situation, as N-Bak is not endogenously expressed in these cells.
In neurons, in which N-Bak is endogenously expressed, it has no intrinsic apoptosis-promoting activity. Apparently, something in the neuronal environment blocks the killing activity of N-Bak. In neurons that would otherwise die due to the absence of NGF signaling, overexpressed N-Bak actively neutralized the death program. A survival-promoting effect of N-Bak in neurons was surprising, as to our knowledge, there are no other examples where a BH3-only protein protects the cells against apoptosis. Actually, the BH3 domain has intrinsic apoptosis-promoting properties, being called a minimal death domain. Indeed, oligopeptides encompassing little more than the BH3 domain of Bak are sufficient to release cytochrome c from isolated mitochondria, to activate caspases, and to promote cell death (52)(53)(54)(55). The same was shown for the BH3 domains of Bax, Bid, Bad, and Bik (15,52,54,56,57). Moreover, the BH3 domains of the anti-apoptotic Bcl-2 and Bcl-x L also facilitate apoptosis (58). How then can overexpressed N-Bak with an intact BH3 domain protect neurons from apoptotic death? Binding to and neutralizing the long Bak isoform, as shown for Bcl-x L /Bcl-x S and Mcl-1 L /Mcl-1 S (18,19,58), are not probable in SCG neurons, as they do not express Bak (although very low levels of Bak transcripts may still remain undetected by our assay). Heterodimerization with other Bcl-2 family members, e.g. neutralizing Bax or releasing Bcl-x L from some blocking constraint, cannot be excluded. However, we have not found any difference in the expression of other Bcl-2 family members in neurons and non-neuronal cells of cultured mouse SCG. Instead, almost all the known Bcl-2 family members were found in both cell types. 2 Therefore, we do not believe that simple heterodimerization with known Bcl-2 family members can explain the opposite behavior of N-Bak in neurons and non-neuronal cells. We propose that N-Bak is intrinsically apoptosis-promoting, as predicted by its domain structure, but that in neurons, it protects against apoptosis indirectly via some other neuron-specific molecules. The molecular interactions of N-Bak in neurons are currently being studied in our laboratory. There are also other examples where some Bcl-2 family members do not behave according to the ligand-receptor model described above. In NGF-deprived chick sensory neurons, overexpressed Bax protected the neurons, although it killed NGF-maintained neurons (59). Similarly, overexpressed Bak inhibited apoptosis in a serum-deprived Epstein-Barr virus-transformed lymphoblastoid cell line (22), whereas Bax protected central neurons against Sindbis virus-induced apoptotic death in vivo (60). Also, Bax-protected cells against tumor necrosis factor-induced apoptosis, although it promoted cell death in other circumstances (41). Currently, there are no good explanations for these results. Our results also show that the presence of four BH domains is not always necessary for a Bcl-2 family protein to be anti-apoptotic. Indeed, the activity of N-Bak in neurons is indistinguishable from that of Bcl-w, Bcl-2, or Bcl-x L microinjected into sympathetic or sensory neurons deprived of neurotrophic factors (43,(61)(62)(63)(64)(65).
We are aware that the study of a protein overexpressed at high levels does not necessarily reflect the functional properties of the endogenous protein. Moreover, our failure to demonstrate endogenous N-Bak protein suggests that the levels of the protein may be very low, in contrast to highly overexpressed N-Bak in our microinjection experiments. Therefore, we cannot claim that endogenous N-Bak is also anti-apoptotic in neurons.
N-Bak is expressed strictly in the nervous tissue and, at least in SCG and hippocampus, strictly in neurons and not in nonneuronal cells. This is, to our knowledge, the first neuronspecific member of the Bcl-2 family. Our results call for caution when interpreting functional studies of Bcl-2 family proteins in an inappropriate cellular environment. Indeed, overexpressed N-Bak had an opposite effect in different cell types. Several other Bcl-2 family members regulate life and death preferentially in some (but not all) tissues. Bid is the main mediator of Fas-induced death in hepatocytes, but not in thymocytes (66), whereas the primary cells affected by the absence of Bim are T and B lymphocytes (67). Bax-␥, a pro-apoptotic splice variant of Bax, is expressed mostly in immune organs and plays a critical role in T cell survival after T cell receptor activation (37). Expression of Bcl-x S is more restricted than that of Bcl-x L (68). Overexpressed Bcl-x S promotes apoptosis in 3T3 cells (69), whereas in 293 kidney epithelial cells or yeast cells, it only neutralizes the survival-promoting effect of Bcl-x L , but is not apoptotic by itself (58). Interestingly, whereas Bcl-G L is expressed in many tissues, its BH3-only isoform is exclusively testis-specific (11).
The function of N-Bak in the nervous system remains elusive, but a specific role is anticipated by its strict neuronspecific expression. Characterization of the nervous system of Bak-deficient mice has not yet been reported (70). Transcripts of N-Bak are expressed at the highest levels in perinatal and newborn mouse brain, at the time of active neurogenesis and programmed cell death. Presumably, N-Bak is part of the molecular network controlling the delicate processes that regulate neuronal number in the developing nervous system, but its explicit role remains to be studied.