Arginine/Serine-rich Protein Interaction Domain-dependent Modulation of a Tau Exon 10 Splicing Enhancer

Tau exon 10 splicing is altered by autosomal dominant mutations that cause frontotemporal dementia with parkinsonism chromosome 17-type and by unknown mechanisms in other related neurodegenerative disorders. Identifying cis- and trans-regulators of tau exon 10 splicing is therefore crucial for understanding disease mechanisms. We previously identified several splicing enhancers and silencers within exon 10 and intron 10. Here, we show that splicing factors SF2/ASF, Tra2β, and a 50-kDa nuclear protein bind in vitro to the polypurine enhancer at the 5′ end of exon 10. Disease splicing mutations N279K and Δ280K disrupt the enhancer and alter associations with these factors. N279K targets robustly bind Tra2β compared with the normal enhancer, which may explain why N279K enhances exon 10 splicing in vivo. In contrast, factor associations with Δ280K targets are nearly undetectable, explaining why Δ280K almost abolishes exon 10 splicing in vivo. Small interfering RNA-mediated suppression of endogenous SF2/ASF and Tra2β significantly reduces exon 10 splicing. Exogenous SF2/ASF dramatically enhances normal exon 10 splicing and efficiently rescues the Δ280K splicing defect. Domain deletion analyses show that the C-terminal RS domains of SF2/ASF and Tra2β are required for normal exon 10 splicing in vivo. In contrast to Tra2β, the SF2/ASF RS domain remains essential in the presence of a strengthened enhancer or when either weak splice site is strengthened. The data suggest that SF2/ASF has both essential and regulatory roles, whereas Tra2β has a supporting role in exon 10 splicing.

In the central nervous system, the microtubule-associated protein tau is highly expressed in neuronal axons and at lower levels in glial cells. Tau functions in microtubule assembly, influences microtubule stability and is important for neurogenesis, axonal maintenance, and axonal transport (1,2). In the human tau gene (MAPT), exons 9 -12 each encode a 31-32amino acid imperfect repeat that comprises the microtubule-binding domain of tau. In adult human brain, exons 2, 3, and 10 are alternatively spliced to produce six different tau isoforms (3)(4)(5). Exon 10 (E10) 2 inclusion generates isoforms with four microtubule repeats called four-repeat (4R) tau. When E10 is skipped, the result is 3R tau isoforms. The 4R/3R ratio in normal adult brain is ϳ1. In fetal brain, only the shortest 3R tau isoform is expressed because of constitutive exclusion of exons 2, 3, and 10 (6, 7).
Pre-mRNA splicing regulation is complex requiring multiple interactions between small ribonucleoprotein particles and non-small ribonucleoprotein particle splicing factors with conserved and nonconserved splicing elements in the unprocessed RNA transcript (8). Conserved elements are the 3Ј and 5Ј splice site signals that demarcate the exonintron boundary. Nonconserved sequences that promote splicing are called exon splicing enhancers (ESE) and intron splicing enhancers (ISE) (9). Serine/arginine-rich (SR) proteins usually bind to enhancer elements and are essential at multiple steps in the splicing pathway (10). Most alternatively spliced exons contain a weak splice site, whose recognition requires SR factor associations with ESEs and ISEs (11). Alternative splicing is also influenced by splicing inhibitory sequences called exon splicing silencers and intron splicing silencers. An additional function of some enhancers is to counteract the inhibitory effects of neighboring silencer elements (12)(13)(14)(15)(16)(17).
For MAPT, cis-regulation of E10 alternative splicing has been extensively studied, where multiple enhancers and silencers regulate use of the weak E10 3Ј and 5Ј splice sites (14,18). E10 sequences include three nonredundant, weak ESEs in the first half of the exon that include an SC35-like ESE, a polypurine enhancer (PPE), and an A/C-rich enhancer separated by a central exon splicing silencer from additional ESE sequences at the 3Ј end of E10. Intronic sequences immediately downstream of the E10 5Ј splice site include an intron splicing silencer and an adjacent intron splicing modulator, which counteracts the intron splicing silencer.
Silent and intronic mutations do not change the protein sequence of tau but affect E10 splicing and ultimately tau function. Thus, subtle changes in the ratio of normal tau isoforms cause severe neurodegeneration.
Identifying proteins that associate with cis-elements is necessary to elucidate how E10 splicing is regulated. Here we focus on identifying protein interactions with the E10 PPE, a 9-nucleotide purine-rich ESE sequence between E10 positions ϩ16 and ϩ24 (see Fig. 1A). The PPE sequence contains two copies of an AAG motif and one copy of a GAR (where R is a purine) motif that occur frequently in natural, purine-rich ESEs. These motifs are often high affinity binding sites for SR and SRrelated proteins. FTDP-17 mutations N279K and ⌬280K alter the normal PPE sequence by adding or removing an AAG copy, respectively (see Fig. 1A). Consequently these mutations have opposite effects on E10 splicing, where N279K increases and ⌬280K decreases E10 inclusion. Our earlier work predicted that the normal PPE function requires SR factor interactions, which are altered by disease mutations ⌬280K and N279K (26). Indeed, the SR-like splicing factor Tra2␤ is essential for E10 inclusion in splicing assays and associates in vitro with both normal and mutant PPE templates (31). Here we show specific in vitro associations of a trio of factors with the normal and strengthened (mutation N279K) PPE but not with a disrupted PPE (mutation ⌬280K) and identify two of these proteins as the SR domain-containing factors SF2/ASF and Tra2␤. We corroborate their specific effects on E10 splicing by gene knock-down assays and present in vivo mechanisms for their varied effects on E10 splicing. Overexpression of SF2/ASF rescues the splicing defect in mutation ⌬280K. Domain deletion assays in non-neuronal and neuronal cells show that the C-terminal protein-interacting domains of both SF2/ASF and Tra2␤ are required for E10 splicing. Our data reveal PPE-dependent and PPE-independent roles for SF2/ASF for use of the weak 3Ј and 5Ј splice sites, respectively. Thus, SF2/ASF has complex essential and regulatory roles on E10 splicing, whereas the role of Tra2␤ appears secondary to SF2/ASF.

EXPERIMENTAL PROCEDURES
Plasmid Construction and DNA Mutagenesis-Splicing vector hN has been described (18) and contains the entire 93-bp MAPT E10 sequence with 33 and 51 bp of flanking introns (see Fig. 3A). Minigene construct E10AH (see Fig. 2) contains entire MAPT exons 9, 10, and 11 as well as 324 and 559 bp of intron 9 and I10 sequences, respectively, inserted into expression vector pRcRSV (Invitrogen). Intron 9 in E10AH contains 224 bp of sequence immediately 5Ј of E9 and 100 bp of sequence immediately 3Ј of E10. I10 in E10AH contains 100 bp of sequence immediately 5Ј of E10 and 459 bp of sequence immediately 5Ј of E11. FTDP-17 mutations ⌬280K and N279K were introduced into hN (18) and E10AH by PCR mutagenesis.
Cell Culture, Transfections, and RNA Isolation-Maintenance and transfection of PC12 cells is described elsewhere (14). HeLa cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Invitrogen). All of the transfections were performed in triplicate with Lipfectamine (Invitrogen) in six-well plates. HeLa cells were transfected with 2 g of plasmid DNA and 6 l of Lipofectamine in 1 ml of OptiMEM (Invitrogen) for 5 h at 37°C (5% CO 2 ), after which 1 ml of Dulbecco's modified Eagle's medium supplemented with 20% fetal bovine serum was added. For coexpression experiments in PC12 cells, 2.5 g of total plasmid containing 0.5 g of an E10 splicing vector and 2 g (4-fold excess) of either pSKII control vector (Stratagene) or individual splicing factor expression constructs were used with 6 l of Lipofectamine. Total RNA was isolated with TRIzol reagent (Invitrogen) as previously described (26).
siRNA Synthesis and Transfection-Commercially synthesized (Dharmacon Inc.) siCONTROL nontargeting siRNA number 1 and gene-specific siRNA pools were used for endogenous suppression of SF2/ASF and Tra2␤. siRNA transfections were optimized in 24-well plates containing cells at 70% confluency using Oligofectamine (Invitrogen) with 25, 50, 75, and 100 nmol of each siRNA in the pool (see Fig.  2A). Maximal inhibition was observed with 100 nmol of each siRNA. Protein was harvested 48 -72 h post-transfection, and expression of targeted factors was tested by immunoblot analyses (described below). To test the effects of siRNA treatment on E10 splicing vectors, the cells were seeded into 12-well plates, and siRNA transfections were scaled up. After 24 h, the cells were rinsed in phosphate-buffered saline and fed 450 l of antibiotic-free Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. One g of normal or mutant splicing vector E10AH was mixed with 3 l of Lipofectamine in 50 l of OptiMEM and added to each well. Total RNA was isolated 48 h later for RT-PCR analyses with TRIzol reagent (Invitrogen).
Quantitation of Tau E10 Splicing by RT-PCR-Tau E10 splicing from transiently transfected splicing vectors was analyzed by RT-PCR assays as described previously (26). E10ϩ and E10Ϫ transcripts were amplified by RT-PCR using vector-specific primer pairs SD6/SA2 for hN and pREP/BGHPA2 for E10AH as previously reported (18,33). Amplified products from hN are 261 bp (E10Ϫ) and 354 bp (E10ϩ) and from E10AH are 447 bp (E10Ϫ) and 540 bp (E10ϩ). The products were resolved on 5-7.5% acrylamide gels and quantitated using a Molecular Imager system (Bio-Rad). For each mutant construct, the values presented are the averages of at least three different transfection experiments with the normal E10 construct transfected in parallel. Statistical comparisons were made using a two-tailed Student's t test. Criteria for significance were calculated using a Bonferroni correction for multiple comparisons.
In Vitro RNA Synthesis-A primer-based strategy was used to in vitro synthesize short, radiolabeled RNAs containing normal or mutant PPE sequences. Forward and reverse primers were designed such that on annealing they form a double-stranded T7 promoter region upstream of the RNA target to be transcribed. A universal forward primer, T7F3 (5Ј-CGCTCTAGATAATACGACTCACTATAGGGAGTAT-3Ј) containing the core T7 promoter sequence (italics) was used. Nine additional nucleotides upstream of the T7 promoter sequence maximize transcription efficiency, and five additional nucleotides downstream of the promoter serve as a linker that overlaps with the RNA target. The reverse primer contains the reverse complement of the core T7 promoter (underlined) with tau sequences (italics) that include the normal or mutant PPE element (bold type). The reverse primers used were: T7PPE (5Ј-GCCTCCTGGATCCAGCTTCTTATTATACTCCCTAT-AGTGAGTCGTATTAGCG-3Ј), T7279K (5Ј-GCCTCCTGGATCC-AGCTTCTTCTTATACTCCCTATAGTGAGTCGTATTAGCG-3Ј), and T7⌬280K (5Ј-GCCTCCTGGATCCAGCTTATTATACTCCCT-ATAGTGAGTCGTATTAGCG-3Ј). To anneal universal and reverse primers, 100 ng of each in 0.1 M NaCl was heated to 94°C for 3 min and cooled at room temperature for 10 min. For in vitro transcription, 1 l of the annealed primer reaction was added to 19 l of a transcription mixture containing 1ϫ T7 polymerase buffer (Invitrogen), 20 mM dithiothreitol, 20 units of RNasin (Promega), 5 M [ 32 P]␣ATP or [ 32 P]␣GTP (800 Ci/mmol), 0.5 mM each of the remaining three NTPs and 50 units of T7 RNA polymerase. The reactions were incubated at 31°C for 2 h, heat-denatured at 75°C for 2 min, quenched on ice for 2 min, and treated with 10 units of DNase I at 37°C for 20 min to remove primers. 7 l of formamide loading dye was added to samples, heated at 95°C for 5 min, and resolved on 12% urea-PAGE gels. Full-length transcripts were visualized by autoradiography, excised, and eluted in 400 l of elution mix (0.5 M ammonium acetate, 2.5 mM EDTA, 0.5% SDS) for 1 h at room temperature. The eluate was ethanol-precipitated and resuspended in 100 l of RNase-free H 2 O, and 1 l was used for quantitation in a scintillation counter.
HeLa Nuclear Extract Preparation, UV Cross-linking, and Affinity Purification-HeLa nuclear extracts were prepared as described (34). The cell pellets were obtained from HeLa-S3 spinner cultures grown at the National Cell Culture Center, a research resource facility established by the National Center of Resources for Research, National Institutes of Health.
For UV cross-linking, radiolabeled (30 -100 ϫ 10 5 cpm) gel-purified, full-length RNA targets were incubated with HeLa nuclear extract in 10-l reactions under splicing-permissive conditions (0.6 mM ATP, 0.1 M creatinine phosphate, 1.5 mM MgCl 2 , 50 -75 g of HeLa extract, 4 g each of tRNA and heparin, and 2 l of 10ϫ binding buffer containing 20 mM HEPES (pH 7.9), 1 M potassium glutamate, 0.2 mM EDTA, 0.5 mM dithiothreitol, and 20% glycerol). The reactions were incubated at 30°C for 30 min, exposed to UV light (0.2 J for 12 min) at a distance of 4 cm on ice to form covalent RNA-protein complexes before treating with RNase A to remove unbound RNA sequences. 11 l of 2ϫ SDS protein loading buffer was added to the samples before boiling for 5 min. The bound proteins were resolved on 10 -15% SDS-PAGE gels and visualized by autoradiography.
For RNA affinity purification, 5Ј-Ome-modified normal PPE and mutant (⌬280K,279K) RNA target sequences were commercially synthesized (Dharmacon Inc.) and are identical in sequence to in vitro transcribed templates used in UV cross-linking assays. RNA-agarose affinity columns were prepared according to methods described before (35) and incubated with 200 l of HeLa nuclear extract under splicing conditions at 30°C for 20 min to allow binding of splicing factors. Protein-RNA-agarose complexes were collected without exposure to UV light and washed extensively in Dignam Buffer D before eluting bound factors in 2ϫ SDS gel loading buffer. The samples were boiled for 5 min, resolved on 10 -12% SDS-PAGE gels, and electroblotted for immunodetection assays.
Immunoblotting-Monoclonal antibody mAb104 (1:100) was used to detect the SR family of splicing factors that range in size from 20 to 85 kDa (36). SF2/ASF was detected by a monoclonal antibody (1:5000) (37). Tra2␤ was detected using a polyclonal antibody Rb3505 (1:1500) raised against a synthetic peptide containing the N-terminal 15 amino acids (MSDSGEQNYGERESR) (Invitrogen Antibody Services). Anti-␤-tubulin monoclonal antibody E7 was obtained from the NICHD Developmental Studies Hybridoma Bank maintained by The University of Iowa (Department of Biological Sciences, Iowa City, IA) and was used at dilution 1:3000 as an internal protein control. Rabbit secondary antibodies (Jackson ImmunoResearch) used at 1:2500 dilution were antimouse IgM for mAb104 and anti-mouse IgG for both SF2/ASF and ␤-tubulin. Chemiluminescent immunodetection was achieved with horseradish peroxidase conjugated to protein A (Sigma) at a dilution of 1:3000 and the ECL plus detection system (Amersham Biosciences).

RESULTS
The following experiments were designed to identify the trans-acting proteins that recognize the normal PPE, to determine whether this protein-RNA interaction is disrupted by mutations that eliminate E10 inclusion, and to determine whether PPE mutations that enhance E10 inclusion work by strengthening normal protein-RNA interactions or cause different proteins to interact with the extended PPE.
Altered Association of Splicing Factors with Normal and Mutant PPE Sequences-To identify proteins associations, radiolabeled RNA targets containing the normal and mutant PPE sequences (Fig. 1A) were incubated with HeLa nuclear extracts under splicing conditions and treated with UV light to crosslink RNA-protein complexes. Because purinerich enhancers such as the PPE bind SR proteins, we first tested four different HeLa nuclear extract preparations for SR protein content by immunoblot analysis. The pan-SR monoclonal antibody mAb104 recognizes phosphorylated epitopes within the RS domain of prominent SR protein family members SRp75, SRp55, SRp40, and SRp30 (36) that were consistently observed in all four nuclear extract preparations (Fig.  1B). UV cross-linking experiments reveal three proteins with apparent molecular masses of 52, 45, and 36 kDa that individually cross-link to normal PPE and 279K RNA targets (Fig. 1C). The 45-kDa factor interacts more strongly with 279K than normal PPE templates. In contrast, almost no SR protein cross-links to the ⌬280K target, although overexposure of the immunoblot showed very weak signals from the 52-and 36-kDa factors but no signal from the 45-kDa factor (not shown). Similar results were obtained with different HeLa extract preparations. Thus, mutation ⌬280K may inhibit E10 splicing by abolishing productive SR protein interactions. The near absence of detectable proteins with ⌬280K targets verifies the binding specificity and rules out nonspecific interactions with flanking linker sequences. Candidate proteins for the 45-kDa band are SR factor SRp40 and the SR-like factor Tra2␤. Based on molecular size, the candidates for the 36-kDa protein are SF2/ ASF (SRp30a), SC35 (SRp30b), and SRp30c.
To identify these proteins, RNA affinity columns were constructed by coupling the normal and mutant RNA sequences to agarose. Columns were incubated with HeLa nuclear extract under splicing conditions as above but without exposure to UV light. Bound proteins were eluted and analyzed by immunoblotting with Tra2␤-and SF2/ASF-specific antibodies. We probed for Tra2␤ because it preferentially binds AAG repeats (38), present in two and three copies, respectively, in normal and 279K templates (Fig. 1A). Consistent with this observation and similar to the binding profiles in UV cross-linking assays, we identify the 45and 36-kDa proteins as Tra2␤ and SF2/ASF, respectively. Tra2␤ eluted robustly from the 279K column compared with the normal PPE column (Fig. 1D) and SF2/ASF eluted at similar levels from both columns. Interestingly, only SF2/ASF was eluted at very low levels from ⌬280K. The in vitro affinity of SF2/ASF and Tra2␤ for normal PPE and 279K mutant templates correlates well with their match to their respective binding consensus sequences (Fig. 1A). Missense mutation N279K strengthens the SF2/ASF binding consensus and may allow stronger SF2/ASF binding, although the observed amounts of SF2/ASF bound to 279K and normal PPE targets are similar. ESE sequences are usually degenerate to accommodate their presence within diverse coding sequences. This degeneracy allows more than one SR protein to recognize the same ESE (39,40). Our results show that the normal PPE sequence is an overlapping low affinity site for both SF2/ASF and Tra2␤. Other examples have been described where SF2/ASF associates with an ESE that is also recognized by Tra2␤ (41). The identity of the 50-kDa factor detected in the UV cross-linking experiments is unknown and is being pursued. In addition to mutation N279K (AAT 3 AAG), we previously showed that a different transversion (AAT 3 AAA) at the same nucleotide position strengthened the PPE by increasing the number of consecutive purines to enhance E10 splicing (18). Because these substitutions enhanced E10 splicing to different extents, the purine context is an important determinant of PPE strength. Although the T to A change increases the number of consecutive purines, unlike in N279K, it does not add an extra AAG copy and is less efficient in E10 enhancement, possibly because of a decreased affinity for Tra2␤.
Effects of SF2/ASF and Tra2␤ Depletion on E10 Splicing-To establish their in vivo requirement on E10 splicing, SF2/ASF and Tra2␤ were depleted individually or together in cell culture using double-stranded siRNAs (42). HeLa cells treated with increasing concentrations of SF2/ ASF-specific siRNAs show a significant reduction in endogenous SF2/ ASF levels compared with ␤-tubulin as a control ( Fig. 2A, left panel). Similar assays with increasing concentrations of Tra2␤-specific siRNAs show almost total reduction in endogenous Tra2␤ levels ( Fig. 2A, middle panel). The gene-specific siRNAs show appropriate target specificity. si-SF2/ASF does not affect Tra2␤ levels, and si-TRa2␤ does not affect SF2/ASF levels ( Fig. 2A, right panel, lanes 3 and 4). The lack of an effect between mock and siCONTROL-treated cells rules out nonspecific effects that may result from siRNA exposure ( Fig. 2A, right panel,  lanes 1 and 2). Cells simultaneously exposed to si-SF2/ASF and si-Tra2␤ show a depletion of both factors ( Fig. 2A, right panel, last lane). However, the extent of depletion is lower compared with the individual siRNA treatments, because the gene-specific siRNA concentration in the simultaneous treatment is half that of the individual treatment. The effects of SF2/ASF and Tra2␤ depletion in non-neuronal HeLa and neuronal PC12 cells were tested on the transiently transfected E10 minigene splicing vector E10AH, which contains MAPT exons 9 -11 with shortened intron 9 and 10 sequences (Fig. 2B). E10AH splicing is dramatically reduced by similar amounts from 80% to 53% and 49% in SF2/ASF-and Tra2␤-depleted HeLa cells, respectively (Fig. 2C). Similar reductions of E10 levels in SF2/ASF-and Tra2␤-depleted PC12 cells (from 76% to 62% and 54%, respectively) confirm that SF2/ASF and Tra2␤ are required for E10 splicing in vivo. E10AH splicing remains unchanged in HeLa or PC12 cells treated with the control siRNA compared with mock treated cells (Fig. 2D, left panel). Because si-SF2/ASF does not completely deplete endogenous SF2/ASF levels, the extent of E10 splicing inhibition in si-SF2/ASF-treated cells is very likely an underestimate and preliminarily suggests that SF2/ASF is more prominently required for E10 splicing. When HeLa and PC12 cells are simultaneously treated with both si-SF2/ASF and si-Tra2␤, E10AH splicing is also reduced by similar levels to 57%, which is not significantly different from the single treatments. These results suggest that SF2/ASF and Tra2␤, although essential, may not act synergistically on normal E10 splicing and may substitute for each other.

Altered SR Factor Interactions and Frontotemporal Dementia
Coexpressing hN with 4-fold excess expression vector SF2/ASF increases E10 inclusion in PC12 cells, whereas overexpressing Tra2␤ does not affect E10 splicing (Fig. 3C). Similar assays do not significantly increase E10 splicing from minigene vector E10AH (Fig. 3C). One explanation for why SF2/ASF specifically stimulates E10 inclusion in hN but not in E10AH is that splicing in E10AH may be too efficient to be influenced by excess splicing factors. Therefore, the inefficiently spliced minigene vector E10AH/⌬280 was also tested in SF2/ASF coexpression experiments. The result was that SF2/ASF stimulates E10 inclusion from 10 to 32%, similar to the 2.5-fold enhancement seen with hN (from 26 to 70%). Coexpression of Tra2␤ shows no change in E10 splicing in all three vectors. The effect of hnRNP G overexpression was examined because this protein has been shown to interact with Tra2␤ (32) and is a candidate for the 50-kDa protein that cross-links to PPE templates. However, hnRNP G does not influence E10 inclusion. This is in contrast to a recent report showing that exogenous hnRNP G reduced E10 splicing from a minigene vector in COS-7 cells (43). Because Tra2␤ interacts with SF2/ASF in vitro and in yeast two-hybrid assays (44,45), both factors were tested together with hN, E10AH, or E10AH/⌬280K. E10 inclusion was enhanced but was no different from that seen with SF2/ ASF alone. Thus, results from depletion and overexpression assays do not reveal a functional synergy between these two factors in enhancing E10 splicing as seen with other exons (41). Because SF2/ASF associations with ⌬280K targets are almost undetectable in vitro, the enhancement of E10AH/⌬280K splicing in the presence of exogenous SF2/ASF suggests that in vivo, either abnormally high concentrations of SF2/ASF can bind to the ⌬280K-shortened PPE or that SF2/ASF has a PPE-independent function, where it binds to other sites present in E10 or interacts with factors already bound to the pre-mRNA. Because the enhancing effect of SF2/ASF overexpression is observed in both minigene and heterologous constructs that contain different flanking exons, the effect appears specific for E10 sequences. Expression of intact and domain deletion constructs (described in the next section) were confirmed by immunoblotting (Fig. 3D). Monoclonal antibody ␣SF2/ASF recognizes an epitope within the N-terminal 90 residues that is retained in the faster migrating SF2/ASF-⌬RS (Fig. 3D, left panel). The polyclonal antibody Rb3505 recognizes an epitope in the N-terminal 15 residues of Tra2␤.
Similar assays performed with the inefficiently spliced E10AH/ ⌬280K construct reveal that SF2/ASF overexpression suppresses the splicing defect in E10AH/⌬280K and enhances splicing from 28 to 46% (Fig. 4B, bottom gel). Deletion of the SF2/ASF RRM domain (SF2-⌬RRM) has no effect (28% E10ϩ), whereas deletion of the RS domain (SF2-⌬RS) almost abolishes E10 inclusion (from 28 to 9%). Surprisingly, SF2-⌬RS also dramatically reduces E10 splicing in the efficiently spliced E10AH/279K construct (from 88 to 59%), suggesting that the RS domain of SF2/ASF plays an essential role in normal and N279K-enhanced E10 splicing (Fig. 4B, lanes 9 and 10). Interestingly, the only effect seen in Tra2␤ expression constructs is a reduction in E10 splicing (from 78 to 60%) when the C-terminal RS2 domain is deleted. Overexpression of intact or domain deletion Tra2␤ constructs have no significant effect on E10AH/⌬280K splicing, suggesting that E10AH/⌬280K is not responsive to normal or mutant Tra2␤ overexpression, presumably because Tra2␤ does not  FEBRUARY 3, 2006 • VOLUME 281 • NUMBER 5 associate with ⌬280K targets that are missing one of the two AAG repeats found in the normal sequence.

Altered SR Factor Interactions and Frontotemporal Dementia
Coexpression assays in PC12 cells were also carried out to determine whether intact or domain deletion SF2/ASF and Tra2␤ constructs have a similar effect on E10 splicing in splicing vectors hN/279K and hN/⌬280K. In hN/279K, the nearly constitutive inclusion of E10 is drastically reduced (from 91 to 32%) when cotransfected with SF2-⌬RS (Fig.  4C). In hN/⌬280K, excess SF2/ASF shows a 2.5-fold increase in E10 inclusion (from 20 to 48%). Thus, mutant E10 sequences in both minigene and heterologous settings show a similar requirement for SF2/ ASF. Tra2␤ expression constructs had no significant effect on hN/279K or hN/⌬280K templates.
RS Domain-dependent Interactions with the E10 Splice Sites-We previously reported that E10 has suboptimal 3Ј and 5Ј splice sites (14,18). A weak polypyrimidine tract contributes to the suboptimal 3Ј splice site, because replacing inhibitory purines with pyrimidines at positions Ϫ3, Ϫ4, and Ϫ7 upstream of E10 (Ϫ3T-7T; Fig. 5A) strengthened the polypyrimidine tract and allowed constitutive E10 splicing. However, when the PPE was disrupted by introducing mutation ⌬280K into construct Ϫ3T-7T, constitutive splicing was either abolished or severely reduced in COS-7 and PC12 cells, respectively. In contrast, the nearly constitutive E10 splicing achieved by strengthening the weak 5Ј splice site using FTDP-17 mutation E10ϩ3 (Fig. 5A) was only marginally reduced (from 97 to 87%) when the PPE was similarly disrupted. Thus, the normal PPE functions bidirectionally but is primarily required to promote use of the 3Ј splice site, allowing it to efficiently establish communication with its weak 5Ј splice site partner.
Because protein interactions with ⌬280K templates are nearly abolished in vitro, we asked whether altered enhancer dependence of the weak E10 splice sites is mediated through protein interactions with SF2/ASF or Tra2␤. RS domain deletion constructs SF2/ASF-⌬RS and Tra2␤-⌬RS2 were coexpressed individually with heterologous E10 splicing constructs hN/Ϫ3T-7T or hN/E10ϩ3 that contain the normal PPE. As in our previous reports (14,18), E10 inclusion in the above splicing constructs is almost constitutive (Fig. 5B). Coexpression of SF2/ ASF-⌬RS with construct hN/E10ϩ3 severely diminishes E10 splicing (from 97 to 36%). Thus, the constitutive effect of the strong 5Ј splice site, although shown previously to be relatively PPE-independent, strongly depends on the SF2/ASF RS domain. Coexpression of SF2/ASF-⌬RS with construct hN/Ϫ3T-7T partially reduces E10 splicing (from 98 to 80%). Thus, strengthening the 3Ј splice site appears to decrease its dependence on the SF2/ASF RS domain. Because splicing of hN/Ϫ3T-7T was previously shown to require an intact PPE, interactions of endogenous SF2/ASF or of the other members in the PPEbinding complex may be strengthened in the presence of a strong 3Ј splice site that overrides the effect of SF2/ASF-⌬RS. Similar assays with Tra2␤-⌬RS2 had no effect on either splicing vector. These results further support both PPE-dependent and PPE-independent roles for SF2/ ASF on E10 splicing regulation. Because both E10 splice sites are normally weak, the requirement for SF2/ASF is expected to be stronger.

DISCUSSION
FTDP-17 highlights the association between splicing mutations that affect multiple cis-sequences and the pronounced variability in phenotype that is characteristic of inherited disorders. Productive interactions that positively define E10 are mediated in part by the PPE element we previously identified. Here we identify factors that associate with the PPE and extend our understanding of how the normal PPE functions and how it is disrupted by FTDP-17 mutations ⌬280K and N279K.
The strength of an ESE is determined by its affinity for available SR factors and the activity of their RS domains (45). FTDP-17 mutations ⌬280K and N279K highlight the physiological relevance of the AAG repeat composition, characteristic of prototypical purine-rich ESEs, on PPE strength and function. Our in vitro binding data provide an explanation for the antagonistic effects of these two mutations on E10 splicing in vivo. Furthermore, our siRNA-mediated depletion assays validate the specific in vivo requirement of both SF2/ASF and Tra2␤ for normal E10 splicing. These results also give the first indication that normal E10 splicing shows a stronger requirement for SF2/ASF than Tra2␤, because incomplete depletion of SF2/ASF strongly reduced E10 transcripts to levels achieved only by complete depletion of Tra2␤. The second indication is that coexpression of SF2/ASF rather than Tra2␤ in neuronal PC12 cells dramatically enhanced E10 splicing in heterologous constructs hN and hN/⌬280K (2.5-fold each) as well as in minigene construct E10AH/⌬280K (3-fold). hN, hN/⌬280K, and E10AH/⌬280K splice inefficiently and appear more responsive to excess SF2/ASF in contrast to vectors E10AH and E10AH/279K, which splice almost constitutively. It is interesting that SF2/ASF coexpression in vivo substan-  (13). B, the bar graph shows averages of RT-PCR assays from HeLa cells co-transfected in triplicate with splicing vectors hN/E10ϩ3 or hN/Ϫ3T-7T and 4-fold excess SF2-⌬RS or Tra2␤-⌬RS2 expression constructs. A corrected significance criteria of p Ͻ 0.025 was used. Significance levels for comparisons between samples are indicated above bars by previously described symbols and include "," p Ͻ1 ϫ 10 Ϫ6 . C, autoradiograph shows representative RT-PCR samples from co-transfected HeLa cells.
tially rescued the splicing defect in hN/⌬280K and E10AH/⌬280K despite a disrupted PPE, whose in vitro association with SF2/ASF is barely detectable. One explanation is that enhancers increase the local concentration of specific SR factors (51), also achieved by their exogenous overexpression. It is also possible that SF2/ASF associates with other sites within E10 in addition to the PPE. The potential for multiple SF2/ASF high affinity binding sites upstream of the E10 5Ј splice site is predicted (not shown) by the ESEfinder program (48). Although the function of these predicted sites requires testing, in vivo analyses in PC12 cells reveal an RNA-dependent function for SF2/ASF on normal E10 splicing. Deletion of the entire SF2/ASF RRM domain marginally decreased E10AH splicing (from 77 to 67%) but had no effect on splicing of E10AH/⌬280K and E10AH/279K. Thus, the RNA binding function of SF2/ASF is required to maintain normal E10 levels in E10AH and, when in excess, its ability to promote E10 inclusion in E10AH and hN. Clues to the PPE-independent functions of SF2/ASF lie in its ability to rescue the splicing defect in constructs E10AH/⌬280K and hN/⌬280K. The third indication of the important role for SF2/ASF in E10 splicing is apparent when its RS domain is deleted. SF2-⌬RS coexpression not only reduced E10 splicing by similar levels in constructs containing a normal (E10AH) or strengthened (E10AH/279K, hN/279K) PPE but also drastically diminished splicing (by 68%) in the splicing-deficient construct E10AH/ ⌬280K. Thus, SF2/ASF-mediated protein interactions are a critical function for regulated and enhanced tau E10 splicing.
In contrast to SF2/ASF, only normal E10 showed a slight but nonsignificant increase in splicing levels when coexpressed with Tra2␤ in vivo (E10AH; Fig. 3C). One possibility is that endogenous levels of Tra2␤ are functionally adequate in maintaining E10 splicing as suggested by depletion assays. Coexpressing SF2/ASF with Tra2␤ did not show a synergistic increase in E10 splicing (Fig. 3C), although both factors have been shown to interact in other studies and depletion of either factor alone or together reduced normal E10 splicing by similar levels in vivo (Fig. 2C). Another possibility is that Tra2␤ mediates its effect by interacting with a limiting factor, which remains to be identified. Also, deletion of the Tra2␤ RRM domain did not affect E10 splicing, suggesting either that endogenous Tra2␤ association with the PPE competes efficiently with Tra2␤-⌬RRM to maintain normal E10 levels or that Tra2␤-⌬RRM may have an RNA-independent effect and is localized by factors already bound to the PPE. Deletion of only the C-terminal RS2 domain of Tra2␤ specifically decreased normal E10AH splicing, indicating a requirement for protein interactions mediated specifically by this domain (Fig. 4B). The C-terminal RS deletion mutants of SF2/ASF and Tra2␤ may behave in a dominant-negative fashion, where they may occupy the PPE but prevent productive protein interactions. Nevertheless, the strength and type of protein interactions mediated by the RS2 domain of Tra2␤ are distinct from the RS domain of SF2/ASF because both show different functional specificities in regulating E10 templates with a normal or mutant PPE. Use of a weak splice site is regulated both by the availability of SR factors and the strength of RNA-protein interactions as determined by the sequence of the recognition site. Our results show that the specific requirements for SF2/ASF and Tra2␤ protein interactions are also distinguished when either the E10 5Ј or 3Ј splice site is strengthened. SF2/ASF shows PPE-dependent and PPE-independent functions for the 3Ј and 5Ј splice sites, respectively. SF2/ASF protein interactions are critical for use of the weak 3Ј splice site despite the presence of a strong 5Ј splice site. Strengthening the 3Ј splice site almost completely compensates for the lack of SF2/ASF protein interactions with the weak 5Ј splice site (Fig. 5). Recently, the RS domain of SF2/ASF tethered to an ESE was shown to directly bind the branchpoint sequence region in the upstream intron and promote the assembly of splicing complexes (49).
An earlier report showed that the RS domain of SF2/ASF is required to recruit the U2 small ribonucleoprotein particle-associated heterodimer U2AF 65 /U2AF 35 to introns that contain a weak 3Ј splice site (50). Because the requirement for the SF2/ASF RS domain is dependent on the 3Ј splice site strength, an essential function of SF2/ASF bound to the PPE, which is in close proximity to the 3Ј splice site, would be to stabilize the association of splicing machinery components with the weak polypyrimidine tract upstream of E10. On the other hand Tra2␤ protein interactions are required only for normal E10 splicing but may be dispensable when either splice site is strengthened. Thus, SF2/ASF and Tra2␤ have distinct roles in E10 splicing where SF2/ASF appears to play a primary essential role, whereas Tra2␤ plays a regulatory role.
The difference in normal and enhanced tau E10 inclusion may simply reflect a difference in the strength of associations at the enhancer. Because only one factor tethered to an ESE interacts with splicing machinery components (51), the preferential association of SF2/ASF versus Tra2␤ may present different protein-protein interaction specificities and strengths specified by their RS domains. Thus, E10 splicing may be vulnerable to subtle alterations in the levels and activities of SF2/ASF and Tra2␤. Our observations highlight the requirement of SF2/ASF for use of the weak E10 3Ј and 5Ј splice sites as observed in other mRNAs (46,52,53). Further work is required to distinguish between the enhancer-dependent functions of SF2/ASF versus Tra2␤ and their interaction targets within the splicing complex.
Others have attempted to identify E10 trans-acting factors using templates that encompass most or all of E10 including the PPE (31,47). N279K targets bound endogenous Tra2␤ in UV cross-linking assays and exogenously added Tra2␤ in affinity purification assays (31). In contrast to our results, Tra2␤ was also affinity-purified from targets with mutation ⌬280K or with the entire nine-nucleotide PPE deleted. Another report showed increased immunoprecipitation of exogenous Tra2␤, SF2/ASF, and SRp30c with N279K targets, although significant associations were also detected with ⌬280K (47). Because both studies used almost the entire E10 sequence as a template and because binding in the absence of UV cross-linking was observed with ⌬280K and PPE deletion constructs, these experiments cannot distinguish between direct binding to the PPE, indirect binding through factors already bound to the PPE, and the presence of additional SF2/ASF and Tra2␤ E10-binding sites outside of the PPE. While this manuscript was in preparation, Wang et al. (56) recently reported that Tra2␤ overexpression enhances E10 splicing from N279K templates by antagonizing SRp55 and SRp30c associations with a regulatory sequence located immediately upstream of the PPE.
The cumulative interactions between multiple cis-elements and trans-factors result in balanced E10 splicing, which maintains a normal 4R/3R ratio of 1. The requirement for multiple ESEs in E10 and the intron splicing modulator in I10 reflects the need to overcome the presence of: weak splice sites, silencer sequences in E10 and I10, and large introns that flank E10. Our results provide a biochemical explanation for the opposite effects of mutations ⌬280K and N279K on E10 splicing. The altered associations and distinct regulatory functions of SF2/ASF and Tra2␤ proteins may contribute to the different clinical phenotypes associated with these splicing mutations. The altered activity (phosphorylation pattern) (54,55) or expression levels of SF2/ASF and Tra2␤ in different brain regions provide a pathogenic mechanism for abnormal 4R/3R tau isoform ratios observed not only in FTDP-17 but also in related disorders progressive supranuclear palsy, corticobasal degeneration, and Pick disease. SF2/ASF and Tra2␤ are thus modifiers of disease. Because tau is involved in over 22 neurodegenerative disorders including Alzheimer disease, revealing normal and aberrant mecha-nisms in tau expression will be valuable not only in understanding its pathogenic role in the central nervous system but also in the potential for designing RNA-or protein-based therapeutics.