A Developmentally Regulated, Neuron-specific Splice Variant of the Variable Subunit B (cid:1) Targets Protein Phosphatase 2A to Mitochondria and Modulates Apoptosis*

Heterotrimeric protein phosphatase 2A (PP2A) is a major Ser/Thr phosphatase composed of catalytic, structural, and regulatory subunits. Here, we characterize B (cid:1) 2, a novel splice variant of the neuronal B (cid:1) regulatory subunit with a unique N-terminal tail. B (cid:1) 2 is expressed predominantly in forebrain areas, and PP2A holoenzymes containing B (cid:1) 2 are about 10-fold less abundant than those containing the B (cid:1) 1 (previously B (cid:1) ) isoform. B (cid:1) 2 mRNA is dramatically induced postnatally and in response to neuronal differentiation of a hippocampal progenitor cell line. The divergent N terminus of B (cid:1) 2 does not affect phosphatase activity but encodes a subcellular targeting signal. B (cid:1) 2, but not B (cid:1) 1 or an N-ter-minal truncation mutant, colocalizes with mitochondria in neuronal PC12 cells. Moreover, the B (cid:1) 2 N-terminal tail is sufficient to target green fluorescent protein to this organelle. Inducible or transient expression of B (cid:1) 2, but neither B (cid:1) 1, B (cid:2) , nor a B (cid:1)

cells; rather it forms complexes with a variety of other proteins. The PP2A core dimer is composed of the C subunit and the scaffolding A (or PR65) subunit, and several other complexes containing one, but not the other subunit have also been described (5)(6)(7).
The predominant form of PP2A, however, is the trimeric holoenzyme consisting of the core dimer complexed to a third variable regulatory subunit. In mammals, regulatory subunits are encoded by four gene families denoted B (or PR55), BЈ (PR61, B56), BЉ (PR48, PR59, PR72/130), and Bٞ (striatin, SG2NA). Proposed functions of these subunits include regulation of catalytic activity, substrate specificity, and subcellular localization of PP2A. The PP2A B subunit family has four members (B␣Ϫ␦). The five BЈ subunit genes (BЈ␣Ϫ⑀) encode phosphoproteins with diverse functions including regulation of wnt/␤-catenin signaling (8,9). The BЉ family consists of four polypeptides that arise from three genes (PR72/130, PR48, PR59). BЉ subunits are nuclear proteins that bind calcium and have been implicated in the regulation of the G 1 /S cell cycle transition (10 -12). Recent RNA interference studies in Drosophila cells have demonstrated that B family subunits regulate mitogen-activated protein kinase signaling, whereas BЈ family subunits protect cells from apoptosis (13,14).
Even though they were the first PP2A regulatory subunits to be identified, few functions of the mammalian B family have been uncovered to date. Structurally, B family subunits resemble ␤ subunits of heteromeric G proteins in that they contain seven WD repeat motifs predicted to fold into a ␤-propeller (15). B␣, the most abundant B family member, is expressed in a variety of cell types and mediates dephosphorylation of the cytoskeletal proteins tau and vimentin by PP2A (16 -18). The recently identified B␦ subunit is most similar to B␣ and is also expressed in multiple tissues (19). B␤ and B␥ genes, on the other hand, give rise to neuron-specific members of the B family of PP2A subunits with distinct temporal and spatial expression patterns in brain (20). Forced expression of B␥, but not other PP2A regulatory subunits, promotes neuronal differentiation of PC12 cells, an effect that appears to be mediated by activation of the mitogen-activated protein kinase cascade at the level or upstream of the Ser/Thr kinase B-Raf (21).
An important role of B␤ in neuronal survival was suggested by the discovery that the neurodegenerative disorder spinocerebellar ataxia type 12 is caused by a trinucleotide repeat expansion in the promoter region of the human B␤ gene (PPP2R2B) (22). Thus, dysregulated B␤ gene expression may be detrimental to neurons, ultimately leading to the massive cerebral and cerebellar atrophy seen in spinocerebellar ataxia type 12 patients. In this report, we characterize a novel splice product of the B␤ gene which is induced upon neuronal differentiation. The unique N-terminal extension of B␤2 is shown to target the protein to mitochondria, where B␤2 accelerates neuronal cell death after survival factor deprivation.

EXPERIMENTAL PROCEDURES
Isolation of the B␤2 cDNA-Duplicate filters containing 1 ϫ 10 6 plaque-forming units from a rat brain cDNA library in the Zap II vector (Stratagene) were screened with a random primed, [␣-32 P]dCTPlabeled probe corresponding to the full-length mouse B␤1 cDNA (a gift from Dr. Nat Heintz, Rockefeller University). After four rounds of screening, a partial cDNA containing the 5Ј-UTR and the N-terminal third of the B␤2 coding sequence was isolated. The full-length coding sequence for B␤2 was obtained by reverse transcription-PCR (RT-PCR) from total rat brain RNA with primers complementary to the unique 5Ј-coding sequence and common 3Ј-UTR (5Ј-coding sequence/forward primer: 5Ј-AAA TGC TTC TCT CGT TAC CT-3Ј; 3Ј-UTR/reverse primer: 5Ј-GGT TTG ACT AGT ATT CAG TAT GTG-3Ј). The B␤2 cDNA sequence was submitted to GenBank and is available under accession number AY251277.
Generation of FLAG-and Green Fluorescent Protein (GFP)-tagged B␤ Constructs and Site-directed Mutagenesis-Primers complementary to the N terminus of B␤1 and B␤2 and to the beginning of the common region (TEAD) fitted with a HindIII cloning site in conjunction with two nested reverse primers including sequence complementary to the B␤ C-terminal end, the FLAG epitope tag, and a SalI cloning site were used to PCR amplify B␤1, B␤2, and B␤⌬N, respectively. PCR fragments were ligated into pcDNA5/TO or pEGFP-N1 to generate fusion proteins with C-terminal FLAG and FLAG-GFP sequences, respectively. B␤1 1-32 -GFP and B␤2 1-35 -GFP were constructed by excising C-terminal sequences from the full-length B␤1/2-pEGFP-N1 plasmids by EcoRI/ XmaI digestion, filling in the overhangs with Klenow polymerase, and religating the plasmids. The B␤2 RR168EE mutant was constructed by full plasmid synthesis using Pfu Ultra polymerase according to instructions for the QuikChange mutagenesis kit (Stratagene). All constructs were fully sequenced at the University of Iowa DNA Facility.
Antibodies-A peptide derived from the N terminus of B␤2 (CFS-RYLPYIFRPPNT) was coupled to keyhole limpet hemocyanin via the sulfhydryl group of the N-terminal cysteine, and polyclonal antibodies were generated in rabbits and affinity purified by standard techniques (23). B␤1 and pan-B subunit antibodies have been described previously (20). Monoclonal antibodies to the PP2A A subunit were a kind gift from Gernot Walter (University of California San Diego), and PP2A C subunit antibodies were purchased from Transduction Laboratories. The adenine nucleotide translocase antibody was provided by Harmut Wohlrab (Boston Biomedical Research Institute).
Ribonuclease Protection Analyses-The B␤2 cDNA library clone was subcloned into pBluescript KSϩ and in vitro transcribed using T7 polymerase. Total RNA was isolated from selected rat organs and brain regions using TriZol reagent according to the manufacturer's instruction (Molecular Research Center). Ribonuclease protection was carried out as described previously (20,24).
Competitive RT-PCR-Total RNA (0.5-1.0 g) was reverse transcribed and PCR amplified in the same 25-l reaction with reagents from the Access RT-PCR kit (Promega, Madison, WI) and the following primers (0.5 M each): common reverse, 5Ј-GAC ATC AAG CCA GCC AAC ATG GAG G-3Ј; B␤1 forward, 5Ј-TGC CCC CCT CTC CTG TGA GAC-3Ј; B␤2 forward, 5Ј-ACC ATC CTC TCT TCC AGC TGC C-3Ј. Aliquots of PCRs were separated on 1% agarose gels and ethidium bromide-stained bands were quantified by image analysis using NIH Image software. The ratio of the 749-bp B␤1 and 619-bp B␤2 PCR products was found to be independent of the number of PCR cycles; 35 cycles were routinely used.
Cell Culture-COS-M6 and PC6-3 cells were cultured and transfected as described previously (15,21). The adult hippocampal progenitor cell line HC2S2 was generously provided by Fred Gage (Salk Institute) and cultured in the presence of 20 ng/ml fibroblast growth factor-2 on laminin-and polyornithine-coated plates according to published protocols (25,26).
Immunoprecipitation and Phosphatase Activity Assays-COS-M6 cells were transfected in six-well plates using LipofectAMINE 2000 (BD Biosciences), and C-terminally FLAG-tagged B␤ subunits were immunoprecipitated with FLAG antibody-agarose conjugates (Sigma) as described (15), except that the immunoprecipitation buffer lacked protein phosphatase inhibitors. Aliquots of immunoprecipitates were solubilized in SDS sample buffer for immunoblot analyses. For PP2A activity assays, immunoprecipitates were stored at Ϫ20°C in 50% glycerol, 10% ethylene glycol, 20 mM Tris, pH 7.5, 5 mM dithiothreitol, 2 mM EDTA, and 0.1% Triton X-100. 33 P-Labeled substrates (see below) were diluted to 0.2-0.5 mg/ml (2,000 -10,000 cpm/l) in 2 mg/ml bovine serum albumin, 50 mM Tris, pH 7.5, 2 mM EDTA, 2 mM EGTA, 2 mM dithiothreitol, 1 mM benzamidine, 1 mg/ml leupeptin. Phosphatase reactions were started by the addition of 5 l of PP2A immunoprecipitates to 20 l of diluted substrate, incubated for 30 min at 30°C with intermittent agitation on an Eppendorf shaking incubator, and terminated by the addition of trichloroacetic acid to a final concentration of 20%. After centrifugation at 22,000 ϫ g, acid-soluble 33 P was quantified by liquid scintillation counting. Less than 20% substrate dephosphorylation occurred under our assay conditions, and activities were inhibited completely by 2.5 nM okadaic acid, a specific inhibitor of PP2A at this concentration.
Confocal Imaging of GFP Fusion Proteins-PC6-3 cells were seeded on collagen-coated, chambered no. 1 cover glasses (20-mm 2 chamber, Nalge Nunc) and transfected with 1 g of GFP fusion protein plasmids using LipofectAMINE 2000. 24 -48 h post-transfection, cells were imaged live on a Zeiss LSM 510 laser scanning confocal microscope at the Central Microscopy Facility of the University of Iowa. In some experiments, MitoTracker Red CMXRos (Molecular Probes) was added to 100 nM to stain mitochondria.
Subcellular Fractionation-PC6-3 cells were plated at 2 * 10 6 cells/ dish in 60-mm dishes and transfected 1 day later with 5 g/dish GFP fusion protein plasmids using LipofectAMINE 2000. Three days later, cells were dislodged by scraping in 0.5 ml of mitochondria isolation buffer (0.25 M sucrose, 20 mM HEPES, pH 7.4, 1 mM EGTA, 0.5 mM EDTA, 1 mM dithiothreitol, 1 mM benzamidine, 5 g/ml leupeptin, 0.5 mM phenylmethylsulfonyl fluoride. After addition of 2 mM MgCl 2 , cells were disrupted by nitrogen cavitation (20 min, 1,000 p.s.i). Unbroken cells and nuclei were removed by two successive centrifugations at 800 ϫ g for 5 min. The postnuclear supernatant was supplemented with 5 mM EDTA and further fractionated by centrifugation at 22,000 ϫ g for 15 min. The supernatant was designated the soluble protein fraction; the pellet was washed once in mitochondria isolation buffer and designated the crude membrane fraction containing mitochondria.
Generation of Tetracycline-inducible PC6-3 Cell Lines-Tetracyclineinducible (T-Rex system, BD Biosciences) PC6-3 cell lines stably expressing FLAG epitope-tagged B family PP2A regulatory subunits were generated as described previously (21). Between 40 and 60 blasticidineand hygromycin-resistant clones were expanded and tested for inducible expression by immunoblotting for the FLAG epitope tag. In positive clones, maximum protein expression was achieved after 24 h treatment with 1 g/ml doxycycline.
Cell Death Assays-Tetracycline-inducible PC6-3 cell lines were seeded at 10,000 cells/well in collagen-coated 96-well plates and grown for 72 h in regular growth medium (10% horse serum, 5% fetal bovine serum in RPMI 1640) in the presence of vehicle (0.1% ethanol) or 1 g/ml doxycycline. After two washes, serum-free RPMI 1640 Ϯ doxycycline was added, and cell density was assayed by MTS tetrazolium reduction to formazan according to the manufacturer's instructions (CellTiter 96 ® AQ ueous nonradioactive cell proliferation assay, Promega). Formazan production was quantified after 3 h by absorbance measurement at 490 nm using a 96-well plate reader. The MTS assay was repeated after 24 h in serum-free medium, and cell survival was expressed as the ratio of the two measurements. Previous apoptosis studies with PC6-3 cells have documented excellent correlation between cell counts and metabolic activity as assayed by tetrazolium salt reduction (27).
For nuclear morphology assays, native PC6-3 cells or tetracyclineinducible cells were seeded at 200,000 cells/well in 20-mm 2 chambered cover glasses. Native PC6-3 cells were transiently transfected with 1 g/chamber GFP fusion protein plasmids using LipofectAMINE 2000 and cultured for 48 h, whereas inducible cells were treated with vehicle or doxycycline for 72 h prior to serum deprivation. After 24 h under serum-free conditions, cultures were fixed in 3.7% paraformaldehyde in phosphate-buffered saline, incubated with the blue fluorescent nuclear stain Hoechst 33342 at 1 g/ml for 5 min and mounted on slides. Random microscopic fields (6 -12 fields/culture, 50 -200 cells/field) were captured on an epifluorescence microscope, and images were coded and analyzed blind to the experimental condition. Cells with condensed, irregular, or fragmented nuclei were scored as apoptotic.

RESULTS
Identification of a Novel PP2A Regulatory Subunit-A rat brain cDNA library was screened with a PP2A/B␤ cDNA probe to identify novel isoforms of this brain-specific PP2A regulatory subunit. A clone was isolated with an insert of 431 bp, of which the first 223 bases are novel, and the last 208 bp are identical to the coding sequence for amino acid residues 23-91 of rat B␤ (28). Conceptual translation of this partial cDNA predicts an isoform of B␤ with a novel 5Ј-UTR and N-terminal extension of 24 amino acids. The full-length cDNA for B␤2 was isolated by RT-PCR from rat brain RNA using primers flanking the coding sequence (GenBank accession number AY251277).
Data base searches with the unique rat B␤2 sequence identified several human and mouse ESTs with high degrees of sequence conservation at the nucleotide level and 100% amino acid identity in the coding region. The murine B␤2 ortholog was recently described and named B␤.2 (29). EST data base searches also identified a B␤2 ortholog from rainbow trout (accession number CA376753) which has three conservative substitutions in the N-terminal tail. No B␤2-related sequences were found in other EST or genome data bases, suggesting that B␤2 has evolved in the vertebrate subphylum.
The chromosomal organization of exons encoding human B␤1 and B␤2 was determined by computer-aided alignment of the B␤ cDNAs with human and murine genome data bases and is shown in Fig. 1A. The gene structure of the human and murine B␤ genes is highly conserved as has been noted previously (29). The alternate N termini of B␤1 and B␤2 are encoded by exons separated by ϳ150 kb. Because the transcription start site for the human B␤1 mRNA is less than 600 nucleotides upstream of the initiation codon (30), the B␤2 transcript ap-pears to be generated by use of an alternate promoter upstream of exon 1.2 and splicing of exon 1.2 to the first common exon (Fig. 1A). Of note, human EST BC031790 predicts an alternate B␤ mRNA in which exon 1.2 is fused out of frame to exon 1.1 and the rest of the coding sequence. The existence of this apparently incompletely spliced EST supports the notion that B␤2 is generated by cis splicing of a huge pre-mRNA (ϳ500 kb) spanning the B␤ locus. Based on prior structure modeling and site-directed mutagenesis of the related B␥ subunit (15), the variant B␤2 N terminus is predicted to extend from a ␤-propeller core structure encoded by common exons 2-9 ( Fig. 1B).
Characterization of B␤2 Expression-To demonstrate that B␤2 is expressed at the protein level, we generated polyclonal antibodies by immunizing rabbits with a peptide from the unique N-terminal tail of B␤2. The resulting antibody reacted specifically with heterologously expressed B␤2 and displayed no cross-reactivity with B␤1 or other PP2A regulatory subunits (Fig. 2, and not shown). Although B␤1 could be detected in total brain lysates (Ref. 20 and Fig. 2), antibody detection of a protein with the size predicted for B␤2 (52,000) necessitated enrichment of PP2A holoenzymes by microcystin-Sepharose affinity purification (20). COS cell lysates expressing FLAG epitope-tagged B␤ splice variants were used as standards and immunoblotted with B␤ isoform-specific and FLAG-directed antibodies to compare detection strengths of B␤1 and B␤2 antibodies. Thus normalizing for antibody affinities and titers, the relative abundance of B␤1-and B␤2-containing PP2A holoenzymes in rat brain was estimated to be ϳ10:1.
The low abundance of B␤2 precluded an analysis of its spatial and temporal expression pattern at the protein level. Therefore, we performed ribonuclease protection assays with probes corresponding to the divergent domains to map the expression of B␤ isoforms in rat brain regions. B␤1 and B␤2 transcripts were detected at comparable levels in all forebrain We reported previously that B family regulatory subunits exhibit distinct developmental expression profiles in brain (20). The neuronal B␥ isoform is induced during postnatal brain development, whereas B␣ and B␤1 show constant expression and a slight postnatal decline in expression, respectively. Ribonuclease protection assays with a B␤2-specific probe revealed that this isoform has an expression pattern similar to B␥, with near undetectable expression at birth rising to adult levels by postnatal day 14 (Fig. 4A). Thus, alternative promoter use and splicing of the B␤ gene appear to be regulated developmentally.
We developed a competitive RT-PCR protocol to assay changes in relative abundance of B␤1 and B␤2 transcripts rapidly. In this assay, reverse transcription of mRNA is carried out using a reverse primer that anneals to the common region of B␤1 and B␤2 followed by PCR amplification with the common primer and two competing forward primers complementary to B␤1-and B␤2-specific sequences (Fig. 4B). This technique was tested by analyzing changes in B␤ gene expression during rat brain maturation (Fig. 4C). Although this assay clearly showed postnatal up-regulation of the B␤2 transcript, the 70% drop in B␤1 message levels detected by ribonuclease protection could not be demonstrated by RT-PCR for reasons that are unclear at present. Because B␤1 protein levels show an ϳ2-fold decline during postnatal development (20), the ribonuclease protection data are likely a better indicator of the absolute abundance of the B␤1 mRNA.
Although all neuronal cell lines tested expressed B␤1, none had detectable levels of B␤2 (data not shown). To examine alternative splicing of the B␤ locus in a cell line that more closely resembles primary forebrain neurons, we turned to HC2S2 cells, a neuronal progenitor cell line derived from rat hippocampus (25). HC2S2 cells are conditionally immortalized by a tet-racycline-repressible v-myc oncogene and differentiate into phenotypic neurons upon addition of tetracycline or doxycycline to the medium (Fig. 4D). Competitive RT-PCR with B␤1 and B␤2 primers was performed on HC2S2 cultures treated for up to 4 days with doxycycline. B␤2 mRNA was already detectable in dividing HC2S2 cells, and levels increased further relative to B␤1 as cells differentiated into neurons (Fig. 4E). These data strongly indicate a neuronal locus of B␤2 expression. The time course of B␤ isoform expression in differentiating HC2S2 cells closely parallels that seen in postnatal maturation of the brain (compare Fig. 4, C and E). With the caveats inherent to a comparison between neuronal differentiation in vitro and brain development in the intact organism, these data suggest that the HC2S2 cell line may be an appropriate model system for B␤ gene regulation studies.
The B␤2 N Terminus Does Not Affect Holoenzyme Formation or Catalytic Activity-Previous structure-function studies indicated that the variable N terminus of B␥ is dispensable for binding to the A and C subunits (15). A role of the B␥ N terminus in directing the PP2A holoenzyme to specific substrates in the mitogen-activated protein kinase pathway was suggested by analyses of chimeras between this neuronal specific regulatory protein and the widely expressed B␣ subunit (21). To investigate whether the differentially spliced N termini of B␤1 and B␤2 play a role in formation or catalytic activity of the PP2A holoenzyme, the two isoforms were FLAG epitope tagged at the C terminus and transiently expressed in COS-M6  3. B␤2 transcripts are brain-specific and expressed widely in rat brain regions. Ribonuclease protection analyses were carried out with probes that correspond to the unique 5Ј-UTR and N-terminal coding regions of B␤1 and B␤2. A, total RNA from the indicated organs of adult rats (Br, brain; Lu, lung; Li, liver; Sp, spleen; Ki, kidney; Ov, ovary; Pl, placenta; Te, testis; He, heart) was analyzed for B␤2 and cyclophilin (cyc, internal control) transcript levels. B, total RNA from the indicated brain regions of adult rats (St, striatum; OB, olfactory bulb; Mi, midbrain; Hi, hindbrain; Di, diencephalons; Co, cortex; BS, brain stem; Ce, cerebellum) was subjected to ribonuclease protection analysis. The graph shows B␤1 and B␤2 densities normalized to the cyclophilin internal control and to the average of each series (means Ϯ S.E. of three sets of RNA preparations). The B␤1 expression data are replotted from Ref. 20 to facilitate comparison. A representative autoradiogram of B␤2 expression is shown at the bottom. cells. A deletion mutant lacking the divergent N terminus, B␤⌬N, was also constructed and analyzed in parallel. The ectopically expressed proteins were immuno-isolated with anti-FLAG resin and analyzed for association with endogenous A and C subunits by immunoblotting. B␤1, B␤2, and B␤⌬N could be expressed to similar levels and associated with equivalent amounts of A and C subunits (Fig. 5A). Aliquots of the immunoprecipitates were then assayed for dephosphorylation of two model substrates, myelin basic protein and casein phosphorylated in vitro by protein kinase A (Fig. 5B). Myelin basic protein was a better substrate than the more acidic casein in these assays. Importantly, the three PP2A heterotrimers had equivalent activities toward these substrates. Therefore, we conclude that the divergent N termini of B␤ isoforms are not involved in formation of the PP2A heterotrimer and do not influence substrate recognition, at least in these in vitro assays.
The B␤2 N Terminus Encodes a Mitochondrial Localization Signal-B␣, B␤1, and B␥ regulatory subunits are found in different subcellular fractions from brain and localize differentially to neuronal somata and processes (20). We explored a possible function of the B␤1 and B␤2 N termini in directing the protein to different places in the cell by transiently transfecting the PC6-3 subline of neuronal PC12 cells with expression plasmids encoding B␤1, B␤2, and B␤⌬N tagged at the C terminus with GFP. The three GFP fusion proteins were not degraded appreciably and were expressed at similar levels in PC6-3 cells. Furthermore, addition of the GFP moiety did not compromise coimmunoprecipitation of B␤ with endogenous A and C subunits (data not shown). Live confocal microscopy revealed that in every transfected cell, B␤1-GFP was diffusely localized throughout the cytoplasm and clearly excluded from the nucleus. B␤2-GFP, on the other hand, showed a punctate localization in addition to diffuse cytoplasmic fluorescence (Fig. 6A). The degree of punctate versus diffuse B␤2-GFP fluorescence varied between cells and transfections (10 -50% cells with discernible punctate). The most straightforward interpretation of these results is that the B␤2 N terminus is responsible for localizing the protein to punctate structures in cells. Alternatively, the B␤1 N terminus may encode a cytoplasmic targeting FIG. 5. In vitro characterization of PP2A holoenzymes containing B␤ isoforms. A, FLAG epitope-tagged B␤1, B␤2, or a truncation mutant lacking the divergent N terminus (B␤⌬N) was transiently expressed in COS-M6 cells; transfections with empty vector served as controls. FLAG immunoprecipitates (IP) were probed for transfected B subunits and endogenous A and C subunits. B, PP2A holoenzymes containing the indicated FLAG-tagged B␤ subunits were immuno-isolated as in A and assayed for activity toward exogenous, 33 P-labeled substrates (protein kinase A-phosphorylated myelin basic protein, MBP; protein kinase A-phosphorylated casein). Shown are the means Ϯ S.D. of duplicate determinations.

FIG. 4. B␤ gene expression is regulated developmentally.
A, ribonuclease protection analyses of total RNA isolated from whole rat brains of the indicated ages. Cyclophilin-normalized B␤1 and B␤2 intensities are plotted relative to the maximum value of each series (means Ϯ S.E. of analyses of three sets of RNA preparations). The B␤1 data are replotted from Ref. 20. B, diagram of the competitive RT-PCR procedure. cDNAs reversetranscribed with a reverse primer annealing to the common region of B␤1/2 are amplified by two competing, isoform-specific forward primers yielding PCR products distinguishable by size. C, total RNA was isolated from rat brains of the indicated ages and subjected to competitive RT-PCR to assess relative expression levels of B␤1 and B␤2. A gray scale inverted image of an ethidium bromide-stained agarose gel is shown. D, morphology of the hippocampal progenitor cell line HC2S2 in the dividing state (left) and 4 days after addition of doxycycline (Dox) to induce neuronal differentiation (right). E, competitive RT-PCR analysis of relative expression levels of B␤1 and B␤2 mRNA in HC2S2 cells treated for the indicated days with doxycycline.
signal that overrides a "default" address in the common region of B␤ for punctate localization. The latter interpretation can be discounted because the subcellular distribution the N-terminal deletion mutant (B␤⌬N-GFP) was indistinguishable from B␤1-GFP (Fig. 6A).
The puncta labeled by B␤2-GFP were identified as mitochon-dria in double labeling experiments with the red rosamine derivative dye MitoTracker, which accumulates in actively respiring mitochondria (Fig. 6B). B␤1-GFP, in contrast, appeared deplete in areas with high densities of mitochondria.
To investigate whether the B␤2 N terminus is sufficient for targeting to mitochondria, the first 35 amino acids of B␤2, FIG. 6. Mitochondrial targeting of B␤2. A, B␤1, B␤2, or the N-terminal deletion mutant B␤⌬N was fused at the C terminus to GFP and transiently expressed in PC6-3 cells for fluorescence imaging of live cells. B, confocal images of live PC6-3 cells expressing B␤1and B␤2-GFP fusion proteins (green) stained with MitoTracker dye to label mitochondria (red). Colocalization of signals is apparent as yellow color in the red-green channel-merged images. C, the first 32 and 35 residues including the divergent N termini and 11 residues of the common domain of B␤1 and B␤2, respectively, were fused to the N terminus of GFP (B␤1 1-32 , B␤2 1-35 , see diagram) and analyzed for colocalization with mitochondria as in B. Scale bars ϭ 10 m.

B␤2 Targets PP2A to Mitochondria 24981
including the unique 24 residues and 11 residues shared with 〉␤1, were fused to the N terminus of GFP (B␤2 1-35 -GFP). The corresponding N-terminal fusion of B␤1 (B␤1 1-32 -GFP) served as a control. The N terminus of B␤2, but not B␤1, was capable of targeting GFP to mitochondria in PC6-3 cells (Fig. 6C). In contrast to the full-length protein, B␤2 1-35 -GFP showed a strikingly discrete mitochondrial localization in virtually every transfected cell, with little if any diffuse fluorescence. It is conceivable that the common C-terminal region of the B␤ splice variants associates with cytoplasmic proteins/structures, which gives rise to the mixed diffuse/mitochondrial localization of full-length B␤2-GFP. Mitochondrial localization of the B␤2 N terminus was also demonstrated by subcellular fractionation. Transient expression of B␤2 1-35 -GFP gave rise to two GFP immunoreactive bands with mobilities of 31,000 and 33,000; the predicted molecular weight is 31,727 (Fig. 7). This heterogeneity may be a consequence of proteolysis, internal translation initiation, or post-translational modification. The lower mobility, presumably full-length or post-translationally processed B␤2 N-terminal fusion protein cofractionated with mitochondria, whereas the smaller protein and the B␤1 N terminus (B␤1 1-32 -GFP) were mostly soluble.
B␤2 Promotes Apoptosis-In addition to performing critical functions in biosynthesis and energy metabolism, mitochondria are central to apoptotic signal transduction (31). To explore a possible function of mitochondria-targeted B␤2 in neuronal apoptosis, we generated a panel of stable, clonal PC6-3 cell lines that express B␤1, B␤2, or B␥ under control of a tetracycline-inducible cytomegalovirus promoter (21). The PC6-3 subline of PC12 cells was established by Pittman and coworkers (27) as a neuronal apoptosis model that more closely resembles sympathetic neurons than the parental PC12 cell line. Undifferentiated PC6-3 cells express primarily B␣, whereas nerve growth factor-differentiated cells additionally express B␤1 and B␥ (21). Endogenous B␤2 expression is undetectable by RT-PCR under either condition (data not shown).
Growth of the stable PC6-3 cell lines in doxycycline-containing medium led to the induction of comparable levels of B␤1, B␤2, and B␥, as detected with an antibody to the FLAG epitope tag (Fig. 8A). Approximately 2-fold overexpression was achieved over the endogenous B␣ subunit, as visualized with a pan-B subunit antibody. Levels of A and C subunits, as well as levels of members of the BЈ regulatory subunit family, were unaltered following B subunit induction (Fig. 8A and data not  shown). Growth rates were unaffected by doxycycline treatment in two independently isolated B␤2-expressing cell lines (data not shown). We also assayed cell viability in serum-containing medium and found that B␤2 induction is not toxic to cells (Fig. 8B).
In two different clonal cell lines, inducible B␤2 expression decreased survival by 30 -40% assayed 24 h after serum withdrawal (Fig. 8C). Accelerated cell death was specific for this mitochondria-localized B␤ splice variant because B␤1 or B␥ induction had little to no effect on cell survival. Nuclear condensation and fragmentation are hallmarks of late stage apoptosis. We examined nuclear morphology after staining with a DNA dye to demonstrate that B␤2 decreases survival by promoting apoptosis. Inducible expression of B␤2, but not B␤1, almost doubled the number of cells with apoptotic nuclei after 24 h in serum-free medium (Fig. 8D).
B␤2 Requires Incorporation into the PP2A Heterotrimer to Promote Apoptosis-It is conceivable that binding of the B␤2 N terminus to mitochondria has a nonspecific toxic effect on cells.
To address this issue, we carried out apoptosis experiments in which various B family regulatory subunit constructs tagged at the C terminus with GFP were transiently transfected into PC6-3 cells. 24 h after serum removal, GFP-positive cells with apoptotic nuclei were counted. In agreement with the data from inducible cell lines, we found that transient expression of B␤2, but not B␤1 or B␥, increased the number of apoptotic cells compared with transfection with GFP alone (30% versus 5%, Fig. 9A).
Arg-165 and Arg-166 of B␥ form critical salt bridges with Glu-100 and Glu-101 of the A␣ subunit (15). We replaced the corresponding pair of arginines in B␤2 with glutamates to generate a mutant (RR168EE) that cannot associate with the AC dimer (Fig. 9B). B␤2 RR168EE was able to bind efficiently to an A␣ subunit carrying the opposite charge reversal mutation (EE100RR), demonstrating that the mutant B␤2 protein folds normally (data not shown). As expected, we also failed to detect binding of PP2A A and C subunits to the mitochondriatargeting N terminus of 〉␤2 (B␤2 1-35 , Fig. 9B). Neither the monomeric B␤2 point mutant nor the B␤2 N terminus fused to GFP was able to promote apoptosis after growth factor deprivation in transient transfection assays (Fig. 9, C and D) even though fluorescence levels were equivalent to (B␤2 RR168EE, Fig. 9C) or much greater (B␤2 1-35 ) than full-length, wild-type B␤2. These data strongly support a model in which B␤2 modulates neuronal survival by targeting an active PP2A heterotrimer to dephosphorylate mitochondrial substrates.

DISCUSSION
Protein kinases recognize their substrates via primary sequence motifs surrounding phosphorylatable residues. In addition, the spatial constraint imposed by tethering protein kinases to organelles and other subcellular structures further enhances the fidelity of intracellular signaling via these enzymes (32). In contrast, consensus sequences appear to contribute little to substrate recognition by protein phosphatases (33), and mechanisms for specific subcellular targeting of these enzymes remain relatively unexplored. This report challenges the view that phosphatases lack specificity by documenting the first example of a mitochondria-localized protein phosphatase subunit.
We show that the gene for the B␤ regulatory subunit of PP2A gives rise to an alternative splice variant, termed B␤2. B␤2 mRNA is highly expressed in forebrain structures and is induced during postnatal brain development and during differentiation of an adult hippocampal progenitor cell line. The unique N-terminal extension of B␤2 is shown to be necessary and sufficient for targeting to mitochondria. A functional consequence of mitochondrial localization appears to be modula- Complexity of B␤ Gene Expression-The B␤ gene (PPP2R2B) is unique among genes encoding B family PP2A regulatory subunits in that it gives rise to multiple variants. In a recent report, Schmidt et al. (29) reported the cloning of cDNAs for two novel murine B␤ isoforms: B␤.2 is the murine ortholog of B␤2, the subject of the present report. B␤.1, which has a distinct N-terminal tail, may be a murine-specific B␤ splice form because no orthologs are present in other EST data bases, and RT-PCR failed to detect the presence of this isoform in rat brain. 2 The number of potential B␤ isoforms appears to be even greater because RT-PCR from human brain samples combined with EST and genome data base searches identified several other B␤ gene transcripts that have unique 5Ј-UTR and Nterminal sequences. 3 Judging by the number of entries for B␤2 in human and mouse EST data bases, B␤2 appears to be the second most common B␤ isoform, although it is much less abundant than B␤1 at the protein level (Fig. 2). It seems therefore likely that the remaining, uncharacterized B␤ isoforms are expressed at extremely low levels or in only small subsets of neurons.
B␤ has recently attracted the attention of the research community because of its involvement in the neurodegenerative disorder spinocerebellar ataxia type 12. A CAG trinucleotide repeat expansion immediately upstream of the transcription initiation site of B␤1 was found to be responsible for this disorder (22), which is a relatively common type of spinocerebellar ataxia in India (34). It will be important to address the effect of this repeat expansion on mRNA and protein levels of not only B␤1, but also other B␤ isoforms, especially in light of our finding that B␤2 overexpression promotes neuronal apoptosis.
Structural Implications-Members of the B family of PP2A regulatory subunits are greater than 80% identical at the amino acid level, with greatest sequence divergence at the N terminus. These proteins are predicted to adopt a toroidal, ␤-propeller structure that makes multiple contacts with the AC dimer (15). Pairs of conserved arginines that bind directly to a adjacent glutamates in the A subunit, as well as other amino acids critical for holoenzyme association map to WD repeats 3 and 4 in the mid portion of the B subunit molecule. Based on these data, we arrived at the model topology shown in Fig. 1B. The divergent N terminus is located opposite the AC dimer interface, where it is free to engage in macromolecular interactions that determine the subcellular localization of the PP2A holoenzyme. Consistent with this model and our previous mutagenesis data with B␥ (15), we find that the N-terminal tail can be deleted from B␤ without disrupting the holoenzyme (Fig. 5A). Furthermore, neither the presence nor the identity of the Nterminal tail has any effect on phosphatase activities toward two model substrates (Fig. 5B), arguing that the divergent residues are not involved in direct binding to substrates. Instead, the N terminus of B␤2 was found to encode a subcellular targeting module that can direct GFP to mitochondria as shown by microscopy and biochemistry (Figs. 6 and 7). By analogy, we propose that the differential localization of other B family subunits (20) is also a function of their N-terminal sequences.
How the B␤2 N terminus interacts with mitochondria is unknown at present. Most nuclear encoded proteins destined for the mitochondrial matrix contain N-terminal sequences that are cleaved by signal peptidases (35). If proteolysis of the B␤2 N terminus occurs at all, it is restricted to the cytosol (Fig.  7). In addition, mitochondrial protein import involves unraveling of tertiary structure (36), which would be incompatible with holoenzyme association of B␤2. Because of these considerations, we hypothesize that the B␤2 N terminus binds to a protein or lipid constituent of the outer mitochondrial membrane. B␤2 may be targeted to mitochondria in a manner similar to hexokinase I, whose N terminus binds to the outer mitochondrial membrane protein porin (37). Alternatively, the B␤2 N terminus may interact with mitochondrial lipids such as cardiolipin (38), possibly subsequent to acylation of specific residues. For instance, mitochondria association of the small GTPase Rab32 is thought to depend on fatty acid modification of two cysteine residues near its C terminus (39). Inspection of the B␤2 N terminus does not reveal any sequence similarities to other outer mitochondrial membrane targeting sequences (37,40,41), but it is noteworthy that two cysteine residues at positions 3 and 23 are conserved in mammalian and fish B␤2 orthologs.
PP2A in Apoptosis-The pheochromocytoma PC12 cell line and its PC6-3 subline are established model systems for neuronal differentiation and apoptosis studies (27,42). We have generated stable, clonal PC6-3 lines that express neuronal PP2A regulatory subunits under control of a tetracycline-inducible promoter to investigate their involvement in apoptosis signal transduction pathways. Inducible expression levels of neuronal regulatory subunits are similar to the endogenous B␣ subunit and are not accompanied by any changes in A, C, or other regulatory subunit levels (Fig. 8A). The lack of any compensatory changes in combination with the known instability of monomeric B family regulatory subunits (15) suggests that the induced regulatory subunits complex to a pool of free PP2A dimer in the cell (43). Using this system, we show that inducible expression of the mitochondria-targeted B␤2 subunit potentiates neuronal death after growth factor withdrawal without affecting cell viability in the presence of serum (Fig. 8). It is important to point out that B␤1, B␤2, and B␥ induction levels in PC6-3 cell lines are likely considerably higher than in native neurons, where B␣ is the most abundant B family regulatory subunit (20). Demonstrating convincingly that endogenous B␤2 is proapoptotic will necessitate knocking down its expression in neurons by gene targeting or RNA interference techniques.
The decision between cell survival and apoptotic cell death depends on relative expressions levels and phosphorylation states of pro-and antiapoptotic members of the BCL-2 family of proteins (31). We hypothesize that targeting of PP2A to mitochondria via the divergent N terminus of B␤2 tips the balance toward dephosphorylation, facilitating the activation of proapoptotic proteins or the inactivation of antiapoptotic proteins when cells are challenged by removal of survival factors.
Several lines of evidence support the idea that the balance of kinase and phosphatase activities at the mitochondrial membrane is pivotal for survival signaling. In a set of experiments complementary to the present study, Affaitati et al. (44) showed FIG. 9. B␤2 proapoptotic activity requires incorporation into the PP2A holoenzyme. A, C-terminal GFP fusion proteins of the indicated B family subunits or GFP alone were transiently expressed in PC6-3 cells. 24 h after serum removal, apoptotic nuclei of GFP-positive cells were quantified as in Fig. 8D. B, B␤2 wild-type (w.t.), RR168EE mutant (RE), or the first 35 residues (1-35) of B␤2 tagged at the C terminus with the FLAG epitope and GFP were expressed in COS-M6 cells, immunoprecipitated with FLAG antibodies, and immunoblotted for association with endogenous PP2A A and C subunits. The positions of molecular mass markers (in kDa) are indicated on the left. Arrowheads point to full-length B␤2 (closed) and B␤2  (open) GFP fusion proteins. C, representative microscopic fields (top, GFP fluorescence; bottom, Hoechst 33342 nuclear stain) of PC6-3 cells transiently expressing wildtype or RR168EE mutant B␤2 after 24 h without serum. The arrows point to two apoptotic cells expressing wild-type B␤2-GFP. D, the GFP fusion proteins characterized in B or GFP alone were transiently expressed in PC6-3 cells and scored 24 h after serum withdrawal for apoptosis-promoting activity as in Fig.  8D. Significant differences from GFP control: **, p Ͻ 0.0001. that inducible overexpression of the mitochondria-targeted A kinase anchoring protein (AKAP) 121 promotes survival of PC12 cells. The prosurvival effect of AKAP121 was suggested to involve enhanced phosphorylation by protein kinase A and cytosolic sequestration of BAD, a proapoptotic BCL-2 family protein. Significantly, a PP2A-like activity was implicated in BAD dephosphorylation and apoptosis of lymphoid cells after interleukin-3 removal (45). In another set of studies with a leukemia-derived cell line, toxic concentrations of the lipid second messenger ceramide were shown to activate PP2A, promote its translocation to the mitochondrial membrane, and cause dephosphorylation of the prosurvival protein BCL-2 at Ser-70 (46,47). A member of the BЈ family of PP2A regulatory subunits highly expressed in non-neuronal tissues, BЈ␣, was implicated in targeting PP2A to BCL-2 in the latter studies, suggesting that induction of apoptosis may involve multiple PP2A holoenzymes. Identifying the mitochondrial substrates of the B␤2-containing PP2A holoenzyme is a goal of ongoing experiments and should provide further insights into the function of B␤2 in neurons and possibly the etiology of spinocerebellar ataxia type 12.
The observation that B␤2 is induced during postnatal brain development (Fig. 4) is intriguing in the context of the increased vulnerability of the aging brain to a variety of stressors and insults (48). Inhibition of B␤2 expression or subcellular targeting may therefore provide an attractive avenue for the treatment of brain injuries and neurodegenerative disorders.