Tau Exons 2 and 10, Which Are Misregulated in Neurodegenerative Diseases, Are Partly Regulated by Silencers Which Bind a SRp30c·SRp55 Complex That Either Recruits or Antagonizes htra2β1*

Tau is a microtubule-associated protein whose transcript undergoes complex regulated splicing in the mammalian nervous system. Exon 2 modulates the tau N-terminal domain, which interacts with the axonal membrane. Exon 10 codes for a microtubule binding domain, increasing the affinity of tau for microtubules. Both exons are excluded from fetal brain, but their default behavior is inclusion, suggesting that silencers are involved in their regulation. Exon 2 is significantly reduced in myotonic dystrophy type 1, whose symptoms include dementia. Mutations that affect exon 10 splicing cause frontotemporal dementia (FTDP). In this study, we investigated three regulators of exon 2 and 10 splicing: serine/arginine-rich (SR) proteins SRp55, SRp30c, and htra2β1. The first two inhibit both exons; htra2β1 inhibits exon 2 but activates exon 10. By deletion analysis, we identified splicing silencers located at the 5′ end of each exon. Furthermore, we demonstrated that SRp30c and SRp55 bind to both silencers and to each other. In exon 2, htra2β1 binds to the inhibitory heterodimer through its RS1 domain but not to exon 2, whereas in exon 10 the heterodimer may sterically interfere with htra2β1 binding to a purine-rich enhancer (defined by FTDP mutation E10-Δ5 = Δ280K) directly downstream of the silencer. Increased exon 10 inclusion in FTDP mutant ENH (N279K) may arise from abolishing SRp30c binding. Also, htra2β3, a naturally occurring variant of htra2β1, no longer inhibits exon 2 splicing but can partially rescue splicing of exon 10 in FTDP mutation E10-Δ5. This work provides interesting insights into the splicing regulation of the tau gene.

Alternative splicing is a versatile and widespread mechanism for generating multiple mRNAs from a single transcript (1,2). Splicing choices are spatially and temporally regulated, and the ensuing mRNAs produce functionally diverse proteins, contributing significantly to proteomic complexity (2,3).
Splicing is carried out by the spliceosome, a large and dy-namic complex of proteins and small RNAs (4,5). A major question in splicing, and an obvious point of regulation, is how the spliceosome recognizes authentic splicing sites. The rules governing splice site selection are not fully understood; combinatorial control and "weighing" of splice element strength are used to enable precise recognition of the short and degenerate splice sites (6). Despite the high fidelity of exon recognition in vivo, it is currently impossible to accurately predict alternative exons (7). Exonic and intronic enhancers and silencers are involved in splicing regulation (8,9). Their mutation can result in human disease by causing aberrant splicing (10,11).
On the trans side of regulation, mammalian splicing regulators mostly belong to two superfamilies, the serine/argininerich (SR) 1 proteins and the heterogeneous ribonuclear proteins (hnRNPs), neither of which is exclusively involved in alternative splicing (12,13). The former are also components of the spliceosome, whereas the latter are also involved in pre-mRNA transport, RNA stability, and translational regulation. Several mammalian splicing factors are enhanced in or restricted to neurons, among them htra2␤3, a splicing variant of htra2␤1 (14). Nevertheless, it appears that the exquisite calibration of mammalian alternative splicing is primarily achieved by SR and hnRNP proteins, which show distinct tissue and developmental ratios, despite their ubiquitous distribution (15,16).
Tau is a microtubule-associated protein enriched in axons of mature and growing neurons (17), although it is also found in other cell compartments and types (18). Hyperphosphorylated, microtubule-dissociated tau protein is the major component of neurofibrillary tangles, a hallmark of several neurodegenerative diseases (18,19).
The tau gene produces three transcripts that undergo complex alternative splicing: 6 of the 16 tau exons are regulated cassettes (18). The N terminus of the tau protein interacts with the plasma membrane (20). The structure and function of the tau N terminus are modulated by cassette exons 2 and 3. The C terminus of the tau protein contains four imperfect repeats (encoded by exons 9 -12) that act as microtubule binding domains (21).
Exons 2 and 10 are adult-specific in rodents and humans but with a crucial difference relevant to neurodegeneration: in adult rodents, exon 2 remains regulated, but exon 10 becomes constitutive (22)(23)(24). In contrast, in adult humans exon 10, like exon 2, remains regulated in the central nervous system (23,25). The difference most likely arises from the details of the cis sequences flanking exon 10 in various organisms (26,27), which in turn affect the regulation exerted by trans factors.
Altered splicing factor ratios suppress exon 2 splicing in myotonic dystrophy type 1, a multisystemic disorder whose symptoms include dementia (28). Altered splicing regulation of tau exon 10 can cause inherited frontotemporal dementia (FTDP) with Parkinsonism (17), almost certainly by disturbing the normal tau isoform ratio (18,19). Thus, the splicing regulation of these two exons is directly relevant to neurodegeneration.
Both exons show a splicing default pattern of inclusion (25,29), are affected by intronic and exonic sequences (25-27, 29 -35), and are regulated by several splicing factors, most acting as inhibitors (25, 27, 36 -38). In this report we show that these two tau exons are partially regulated by exonic silencers located at their respective 5Ј ends, which bind splicing factors SRp30c and SRp55. A third factor, htra2␤1, acts in opposite fashion on the two exons: it inhibits exon 2 by binding to the SRp30c⅐SRp55 heterodimer through its RS1 domain but activates exon 10 by binding to an exonic enhancer located directly downstream of the silencer that recognizes the SRp30c⅐SRp55 heterodimer. In the case of exon 10, htra2␤1 and the SRp30c⅐SRp55 complex may sterically interfere with each other.
Protein Expression in Cells and Bacteria and GST Pull-down Assays-For eukaryotic expression, cDNAs were expressed from the following promoters: cytomegalovirus (CMV) for htra2␤1, SRp30c, SRp55, and the variants of htra2␤1 and SRp30c (14,40,41); and SV40 for the variants of SRp55 (42). Western blots show that the factor variants express stable proteins in equivalent amounts (data not shown).
For protein interactions in vitro, htra2␤1 and SRp55 were cloned into pGEX-4T1 and pGEX-4T3, respectively (Amersham Biosciences). SRp30c (a generous gift from Dr. Chris Lorson) was in vector pRSET-c (Invitrogen). The recombinant proteins were produced and purified from Escherichia coli strain BL21 (Stratagene) according to the vendor's instructions. For protein and RNA pull-downs, htra2␤1 and SRp55 were cloned into pFLAG-CMV-6c, and SRp30c was cloned into pFLAG-CMV-6b (Sigma). Protein expression was verified by using anti-FLAG M2 monoclonal antibody (Sigma) on Western blots of crude lysates from cells transfected with each construct.
1 l of a GST-factor fusion or GST alone was incubated with 1 l of SRp30c-His or with 1 l of htra2␤1-FLAG in 0.5 ml of loading buffer (1% Triton X-100, 10% glycerol, 0.25 M NaCl, 1 mM EDTA, and 50 mM Tris, pH 7.5) for 1 h at 4°C and then incubated with 10 l (bed volume) of glutathione-agarose beads (Sigma) for 1 h at 4°C. The beads were washed three times with 500 l of wash buffer (0.05% Triton X-100, 10% glycerol, 50 mM NaCl, and 50 mM Hepes, pH 7.5). The retained proteins were run on 12% SDS-PAGE and immunoblotted with mouse monoclonal antibodies as primary antibodies (anti-His (Invitrogen) at 1:3000 or anti-FLAG M2 (Sigma) at 1:2500) and goat anti-mouse IgG (Zymed Laboratories Inc.) at 1:10,000 as secondary antibody.
Cell Culture, Transfections, and RNA Preparation-Monkey kidney (COS) and human epithelial (HeLa) cells were maintained in Dulbecco's modified Eagle's medium plus 10% fetal calf serum. Plasmid DNA was prepared by Qiagen Tip-50s and introduced into cells by lipofection (LT1; Panvera). Total RNA was isolated by the TRIzol method (Invitrogen). Reverse Transcription and PCRs-We performed PCR analysis of RNA as previously described (30) with primer pair SPL-LS/SPL-LN or SVP2/SPL-LN (Table I). The isoform ratio was calculated by scanning the bands from three independent transfections using the One-Dscan program and the Scanalytics IPLab software.
For UV-cross-linking, exon 2 riboprobes and GST fusions of htra2␤1, SRp30c, and SRp55 were incubated at room temperature for 20 min in binding buffer (10 mM Hepes, pH 7.9, 1 mM EDTA, 1 mM dithiothreitol, 5% glycerol, 50 mM KCl, and 1 g of tRNA). Reaction mixtures were pre-chilled on ice and irradiated with UV light (254 nm) for 10 min. The samples were digested with RNase A at 37°C for 1 h and run on 12% SDS-PAGE. The gels were fixed, dried, and subjected to autoradiography.
For RNA pull-downs, equal counts of the riboprobes were added to lysates from HeLa cells transfected with htra2␤1, SRp30c, and SRp55 FLAG fusions. The vectors were used as negative controls. We immunoprecipitated the proteins using anti-FLAG monoclonal antibody conjugated with agarose (Sigma) according to the Novagen instructions and counted the radioactivity of the precipitated pellet in a beta scintillation counter.

Scanning of Exon 2 by Deletions Uncovers an Exonic
Silencer-Because the default behavior of exon 2 and 10 is inclusion (25,29), the most likely explanation for their exclusion in fetal brain (18) is that in that tissue an inhibitory protein recognizes a silencer sequence on each exon. In our previous work we found one weak silencer within exon 2 by point mutagenesis (30). However, this element is context-specific and is not influenced by factors that strongly inhibit exon 2 splicing (30), suggesting that the point mutation may strengthen an enhancer rather than disrupt a silencer. In this study, we decided to employ a strategy used successfully for tau exon 10 (37, 38) by creating nine deletions that scan exon 2 (Fig. 1A). Fig. 1B shows the splicing behavior of the nine deletions. Two (E2-⌬7 and E2-⌬8) leave exon 2 splicing essentially unchanged. Four (E2-⌬3, E2-⌬5, E2-⌬6, and E2-⌬9) show a significant decrease of exon 2 inclusion, and one (E2-⌬4) shows a moderate decrease of exon 2 inclusion, suggesting that these deletions excise exonic enhancers; E2-⌬4 and E2-⌬5 partly overlap an enhancer previously defined by a point mutation (E2-6, shown in Fig. 1A) (30). Finally, two (E2-⌬1 and E2-⌬2) increase exon 2 inclusion, indicating that the first ϳ20 residues of exon 2 act as a silencer.
Combinations of Deletion Constructs and Regulators Pinpoint Possible Binding Sites for SRp30c, SRp55, and htra2␤1 within Exons 2 and 10 -Our previous work showed that SRp30c and SRp55 strongly inhibit exon 2 splicing (30), whereas htra2␤1 strongly inhibits exon 2 (30) but moderately activates exon 10 (27). This situation gives an ideal handle for comparing how the three factors distinctly regulate two similarly behaved exons. We first did a preliminary correlation by co-transfection experiments: we paired our exon 2 and 10 mu-tants with each member of the trio.
Similarly, both SRp30c and SRp55 lose their ability to inhibit exon 10 splicing in mutant E10-⌬2/3/4 (Fig. 2, C and D), implying that they exert their effect through this region. Neither of the two other exon 10 silencer deletions relieves inhibi-FIG. 1. Exon 2 contains an extended silencer at its 5 end, whereas exon 10 has several silencers throughout its length. A, scanning deletions within exon 2. Deleted nucleotides are in bold, and each deletion is underlined. Shown above each deletion is its action, if any: (W)E ϭ (weak) enhancer, (W)S ϭ (weak) silencer, N ϭ none. E2-6 defines a purine-rich enhancer from previous work (30). B, RT-PCR of the mutated exon 2 constructs in COS cells. The identities of the spliced species are indicated. Primer pair: SPL-LS/SPL-LN. The gray bar across the graph shows the level of the wild-type construct. The percentage of exon inclusion was calculated by scanning the bands from three independent transfections and measuring their areas using the OneDscan analysis program. C, silencers and an enhancer within exon 10. The silencer defined by E10-⌬2/3/4 and the purine-rich enhancer defined by FTDP mutation E10-⌬5 (⌬280K) are indicated. Also shown are the M280 double mutation, two additional deletions (E10-⌬8/9 and E10-⌬15) that define silencers, FTDP point mutations that overlap the silencers (N279K, L284L, and N296N), and the M3 FTDP mutation that is the background of M280. tion by SRp55 or SRp30c (Fig. 2C shows the SRp30c results; the SRp55 data are not shown). Interestingly, E10-⌬2/3/4 is directly upstream of a purine-rich enhancer defined by FTDP mutation ⌬280K (E10-⌬5), which binds htra2␤1 in its role as an activator of exon 10 splicing (27,37).
To discover which regulatory action is dominant on exon 10, we did dosage co-transfections of SRp55, SRp30c, and htra2␤1 with wild-type exon 10 as well as mutant ENH (Fig. 3), which changes a nucleotide right at the boundary between the silencer and the enhancer (Fig. 1C). The results suggest that the two inhibitors "trump" the activator, a finding consistent with the fact that htra2␤1 activates exon 10 splicing rather weakly (Ref. 24 and this work). In the wild type, 0.5 g of each inhibitor dominate 1 g of htra2␤1 (Fig. 3A). SRp55 still inhibits mutant ENH, although more of it is required to achieve full inhibition (Fig. 3B, lanes 2-4), but SRp30c no longer inhibits ENH (Fig. 3B, lane 6). This result strongly implies that 1) the altered nucleotide in ENH comprises an integral part of the SRp30c binding site and 2) the two inhibitors act independently.
Domains of Splicing Factors Required for Regulating the Splicing of Exons 2 and 10 -To determine which domains of the three splicing factors are required for exon 2 and 10 regulation, we next co-transfected wild-type exon 2 and 10 with variants of htra2␤1, SRp55, and SRp30c. Figs. 4A, 5A, and 6A diagram the variants we used for the assays; the functional domains of these three factors are designated according to the conventions of the splicing field: RRM for RNA recognition motif domain(s) and RS for the serine/arginine-rich domain(s).
The N-terminal RS domain of htra2␤1 is required for inhibition of exon 2 splicing, but the C-terminal one is dispensable (Fig. 4B, lanes 4 and 5). The RRM domain of htra2␤1 cannot inhibit exon 2 by itself (Fig. 4B, mutant RRM, lane 3). These results imply that htra2␤1 does not bind directly to exon 2.
Unlike htra2␤1, the SRp55 RS domain is not required for suppression of exon 2 splicing; in fact, SRp55 mutant RS, which only contains the RS domain, activates exon 2 splicing (Fig. 5B,  lane 8). Its RRM2 domain is sufficient to suppress exon 2 splicing as strongly as the wild type, whether it is alone or with RRM1 or RS (Fig. 5B, lanes 3, 6, and 7). Finally, SRp30c no longer suppresses splicing of exon 2 when it is truncated down to either its RNA or protein binding segment (Fig. 6B).
These results strongly suggest that SRp55 modulates exon 2 splicing by direct binding to the mRNA, whereas htra2␤1 influences exon 2 splicing through a protein interaction mediated by its N-terminal RS domain. This result is consistent with the fact that the exon 2 silencer contains a consensus site for SRp55 but not for htra2␤1. SRp30c requires its entire length to regulate exon 2, implying that it binds directly to exon 2 but must also interact with SRp55, htra2␤1, or both to discharge its function.
The RRM domain of htra2␤1 suffices to activate exon 10 splicing and is a slightly better activator than the full-length htra2␤1 (Fig. 4C, mutant RRM, lane 3). In the presence of the RRM domain, the N-terminal RS domain cannot activate exon 10 splicing (Fig. 4C, mutant ⌬RS2, lane 4), whereas the Cterminal RS domain activates exon 10 splicing as strongly as full-length htra2␤1 (Fig. 4C, mutant ⌬RS1, lane 5). Mutant ENH responds to the htra2␤1 variants weakly because of its high inclusion of exon 10 (Fig. 4D), whereas mutant M280 responds more strongly than wild type (ϳ3x rather than ϳ1.5x; Fig. 4E); nevertheless, the relative order of activation strength for both mutants remains the same as that for wild type. Finally, neither full-length htra2␤1 nor the RRM domain can activate mutant E10-⌬5 and ⌬RS2 slightly inhibits it, but ⌬RS1 still activates it strongly (Fig. 4F). These results imply that 1) htra2␤1 still binds to the mutated but still purine-rich M280 site and 2) its RS2 domain is crucial for the activation.
The two RRM domains of SRp55 combined and either RRM domain, alone or in combination with the RS domain, inhibit splicing of exon 10 (Fig. 5C, lanes 3-7). RRM2, alone or with the RS domain, is as good an inhibitor as the full-length SRp55 (Fig. 5C, lanes 6 and 7), whereas RRM1 inhibits only partially (Fig. 5C, lanes 4 and 5). The RS domain alone activates splicing of exon 10 (Fig. 5C, lane 8). The behavior of SRp55 suggests that RRM2 is the primary binding partner of exon 10. SRp30c loses its ability to suppress splicing of exon 10 when it is truncated down to either its RNA or protein binding segment (Fig. 6C). Like SRp55, the RS domain alone of SRp30c activates splicing of exon 10 (Fig. 6C, lane 3). These results strongly imply that SRp55 and SRp30c, like htra2␤1 (32), modulate exon 10 splicing by direct binding to the mRNA.
RNA and Protein Pull-downs Elucidate How Splicing Factors htra2␤1, SRp55, and SRp30c Modulate Exon 2 and Exon 10 Splicing-To further dissect the mechanism of action of the three inhibitory factors on exons 2 and 10, we investigated whether htra2␤1, SRp55, or SRp30c associates with the exon 2 and 10 RNAs in cell lysates or in vitro and whether they interact with each other in GST pull-down assays.
Consistent with the results of the factor variant co-transfections with exon 2 (previous section, Figs. 4BϪ6B), htra2␤1 does not associate with the wild-type exon 2 riboprobe (Fig. 7A,  lanes 10 -12), whereas both SRp55 and SRp30c strongly associate with it, the latter showing stronger binding than the former (Fig. 7A, lanes 3 and 7). Both deletions of the silencer region reduce the ability of SRp30c to bind by 50% (Fig. 7A,  lanes 8 and 9), correlating very well with the co-transfection results ( Fig. 2A, lanes 6 and 9). Deletion E2-⌬2 also greatly decreases binding to SRp55 (Fig. 7A, lane 5), as expected from the co-transfection results (Fig. 2B, lane 6). However, deletions E2-⌬1 and E2-⌬4 show enhanced binding to SRp55 (Fig. 7A,  lanes 4 and 6), even though SRp55 no longer inhibits exon 2 splicing in the presence of these mutants (Fig. 2B, lanes 4 and  8). The Western blots of the immunoprecipitated proteins (Fig.  7B) show that htra2␤1 is equivalently expressed and is immunoprecipitated as efficiently as the other two factors, further confirming the absence of htra2␤1 binding in exon 2. The findings support the model that SRp30c binds directly to exon 2 through a site that overlaps deletions E2-⌬1 and E2-⌬2; SRp55 binds to the region defined by E2-⌬2; finally, htra2␤1 exerts its influence through protein-protein interactions mediated by one of its RS domains.
These conclusions are strengthened by the UV-cross-linking experiments (Fig. 7, C and D). Fig. 7C shows that htra2␤1 does not bind to any of the riboprobes (wild-type exon 2 or deletions E2-⌬1 and E2-⌬2), whereas SRp30c binds strongly to wild-type exon 2, but not to the two deletions. These results are completely consistent with the behavior of the two factors in RNA pull-downs (Fig. 7A).
The SRp55 profile is slightly complicated by the fact that during cross-linking the protein degrades into two major species (Fig. 7D). Taking this into account, SRp55 shows a complex but explicable cross-linking profile. The behavior of the top band, which is full-length SRp55, is congruent with the RNA pull-down results (Fig. 7A): it binds to wild-type exon 2 and binds even more strongly to deletion E2-⌬1, but does not bind to deletions E2-⌬2 and E2-⌬4. The fact that the full-size SRp55 band does not bind to E2-⌬4 when purified, whereas SRp55 binds strongly to E2-⌬4 in cell lysates, almost certainly means that the effect of SRp55 on E2-⌬4 is mediated by yet another factor.
The bottom band in Fig. 7D is truncated SRp55. The truncation, combined with the lack of phosphorylation in bacteria, may allow the positively charged RS domain of SRp55 to interact nonspecifically with the negatively charged riboprobe. Still, it binds more strongly to E2-⌬1 than to the wild type and binds most weakly to deletion E2-⌬2, which lacks the SRp55 consensus binding site.
To confirm our suspicion that the three factors can act as a complex, we did co-immunoprecipitations of the factors in pairs. Fig. 8, A and B, shows that htra2␤1 interacts with SRp55 and that SRp30c interacts with both SRp55 and htra2␤1.

One of the Regulatory Elements of Tau Exon 2 Is an Exonic
Silencer Modulated by a Complex Consisting of htra2␤1, SRp55, and SRp30c-In our present foray we discovered that exon 2 contains a silencer at its 5Ј end (defined by deletions E2-⌬1 and E2-⌬2; Fig. 1B). Our assays (Figs. 2 and 4 -8) further indicate that this element exerts its effect by binding a complex comprising htra2␤1, SRp55, and SRp30c, which binds to exon 2 via the two latter proteins.
Exonic silencers are postulated to work by recruiting inhibitory factors or by blocking spliceosome assembly (45). Our results suggest that SRp30c and SRp55 directly bind to the E2-⌬1/⌬2 and E2-⌬2 elements, respectively, because at least one of their RRM domains is required for exon 2 exclusion. The activation of exon 2 splicing by the RS mutant of SRp55 has been seen with RS mutants of other SR proteins (40) and suggests that overexpression of this domain sequesters interacting inhibitors (in the case of exon 2, htra2␤1).
Besides tau exons 2 and 10, htra2␤1, SRp55, and SRp30c can either activate or inhibit splicing of other alternatively spliced exons, including tau exon 3 (46 -50). Htra2␤1 activates splicing by binding to purine-rich or -enriched motifs (37,43,44). However, in exon 2, it does not exert its effect through direct binding, although exon 2 has a purine-rich enhancer defined by mutants E2-⌬6 and E2-⌬5 (Fig. 1, A and B). Instead, htra2␤1 no longer inhibits exon 2 in mutants E2-⌬1 and E2-⌬2 ( Fig.  2A), and its N-terminal RS domain is required for this activity (Fig. 4B). The naturally occurring variant htra2␤3 lacks this RS domain and is predominantly expressed in brain after day E18 in the rat (14). This suggests that the relative inclusion of exon 2 during development may partly depend on htra2␤ isoform ratios: the increase of the 2 ϩ tau isoform in adulthood correlates with the increase of htra2␤3, which cannot inhibit exon 2 splicing as htra2␤1 can.
Therefore, it appears that the three factors inhibit exon 2 splicing by forming a complex that recognizes the E2-⌬1/⌬2 region through the RRM domains of SRp30c and the RRM2 domain of SRp55. Htra2␤1 exerts its influence by binding to both SR proteins via its N-terminal RS domain (RS1). SRp30c is the core element of the inhibitory complex acting on exon 2, and it binds to E2-⌬1 and part of E2-⌬2. E2-⌬2 is an SRp55 consensus site (9) and would be expected to bind SRp55 very strongly, yet SRp55 is the weakest inhibitor of exon 2 among the three factors in cellular assays ( Fig. 2A, lanes 2 and 3; Fig.  2B, lane 2). Binding of SRp30c partly masks the perfect binding site of SRp55 in E2-⌬2, thereby decreasing the influence of SRp55 on exon 2. In deletion E2-⌬1, SRp30c can no longer bind and therefore allows strong binding of SRp55.
The regulatory complex may inhibit exon 2 splicing by sterically interfering with the U2 small nuclear riboprotein that would form directly upstream of the silencer, at the 3Ј splice site of exon 2. However, the strong SRp55 binding in deletion E2-⌬4 (Fig. 7A) with the concurrent loss of ability of SRp55 to regulate exon 2 splicing in that deletion (Fig. 2B) suggests that this region regulates exon 2 splicing by binding yet another factor that interacts with SRp55. A final player is the purinerich enhancer defined by mutants E2-⌬5 and E2-6 ( Fig. 1A) that may sterically interfere with the action of E2-⌬4 by binding an activator.
In our previous work (30), we tentatively proposed that htra2␤1, SRp30c, and SRp55 may act as a complex that inhibits exon 2 splicing, possibly by binding to a then-unidentified silencer. In the present study, we identify the silencer and confirm that it is regulated by an htra2␤1⅐SRp55⅐SRp30c complex. In Fig. 9A, we show an updated model of exon 2 splicing regulation that incorporates this new knowledge.
The region directly upstream of E10-⌬2/3/4 defines a consensus SRp55 binding site (TGCAGA) according to the ESEfinder (9). Yet SRp55 binds to E10-⌬2/3/4 and not to its consensus site, which jibes with the fact that the latter region acts as an enhancer (32). 2 SRp30c also appears to bind directly to the E10-⌬2/3/4 region, which consists exclusively of A and T residues. This composition resembles the 5Ј half of the hnRNPA1 CE9 intronic element, to which SRp30c binds as a splicing inhibitor (46). The weak residual binding of SRp55 to E10-⌬3/4 (Fig. 7E, lane 5) indicates that the region defined by E10-⌬2 is part of its binding site. The abolition of SRp30c splicing inhibition in mutant ENH (N279K) (Fig. 3B) confirms that it binds downstream from SRp55. This mutation overlaps the beginning of E10-⌬5, which defines the start of a purine-rich enhancer that binds htra2␤1.
Previous work (27,37) led to the conclusion that htra2␤1 activates splicing of exon 10 by binding to the purine-rich enhancer defined by region E10-⌬5/6. Our work suggests that the increase of exon 10 inclusion in mutant ENH may arise not from strengthening the binding of htra2␤1 (Fig. 4D) but rather from abolishing the binding of SRp30c (Fig. 3B, lane 6). The behavior of SRp55 and SRp30c indicates that the two inhibit splicing of tau exon 10 by binding to a region (E10-⌬2/3/4) directly adjacent to the htra2␤1 site (E10-⌬5/6), thereby probably sterically interfering with the activator. The two inhibitors can inhibit exon 10 individually, as shown by the behavior of mutant ENH (Fig. 3B).
SRp30c and SRp55 apparently inhibit exon 10 splicing by binding to the E10-⌬2/3/4 region through their RRM domains. SRp30c binds downstream of SRp55, and the RRM2 of SRp55 is the primary binding component for that factor. As is the case with exon 2, splicing of exon 10 is also activated by the RS mutants of SRp55 and SRp30c, strengthening the theory that overexpression of this domain may relieve inhibition by sequestering partnering factor(s).
Htra2␤1 activates exon 10 splicing through direct binding to a region adjacent to the SRp55/SRp30c binding site, but it can be outcompeted easily by the two inhibitors. The RRM domain alone of htra2␤1 is a better activator than the full-length factor (Fig. 4C, lane 3), implying that SRp30c may interfere with htra2␤1 by interacting with the RS1 domain of htra2␤1. Finally, the RS2 domain of htra2␤1 is crucial for activation of 2 Y. Wang, L. Gao, and A. Andreadis, unpublished results. exon 10 because its combination with the RRM gives the strongest activation regardless of the strength of the RNA binding site (Fig. 4, CϪF, lane 5). Interestingly, htra2␤1 mutant ⌬RS1 can still activate exon 10 deletion ⌬5, suggesting that htra2␤1 can function even if its binding to exon 10 is greatly weakened. These results, combined with the fact that htra2␤1 activates exon 10 relatively weakly, suggest the possible involvement of a co-activator. Alternatively, each of the two RS domains may alter the conformation of htra2␤1. Fig. 9B shows a model of exon 10 splicing regulation that incorporates these results.
Comparisons of the Splicing Regulatory Mechanisms That Operate on Tau Exons 2 and 10 -Tau exons 2 and 10 show intriguing similarities in their expression patterns and behavior. Both are adult-specific, and their relative levels of expression are congruent across different nervous tissues (18). Also, the default splicing behavior of both exons is inclusion (25,29), yet most SR and hnRNP proteins inhibit their splicing (25,27,30). This collective behavior is unusual, although it is consistent with the theory that exons 2 and 10 must be primarily regulated through inhibition.
The cumulative results from our experiments let us formulate speculative models of splicing regulation of tau exons 2 and 10 ( Fig. 9, A and B). Here, too, interesting similarities and contrasts emerge. Both exons have a silencer at or near their 5Ј end, although that of exon 2 is pyrimidine-rich, whereas that of exon 10 is AT-rich (Fig. 1, A and C). In both cases, SRp55 and SRp30c act as an inhibitory heterodimer that binds to the silencer at the 5Ј end of each exon. Both domains of SRp30c and the RRM2 domain of SRp55 are required for suppression of either exon, and their RS domains alone moderately activate splicing of both exons, almost certainly through a sequestering effect. However, in exon 2, SRp30c is apparently situated upstream of SRp55 on the silencer, but their relative locations are switched in exon 10. Finally, although htra2␤1 acts in opposite ways on the two exons, it shows an interesting parallel nevertheless: htra2␤1 mutant ⌬RS1 can be considered an "activator" for both exons because it no longer inhibits exon 2 but is the strongest activator of exon 10 among the htra2 mutants. Interestingly, mutant ⌬RS1 corresponds to naturally occurring variant htra2␤3, which is predominantly expressed in brain and increases during development (14). Thus, the shift toward more htra2␤3 favors inclusion of both tau exons, but by different mechanisms. A and E, 32 P-labeled riboprobes containing exon 2 or 10 were incubated with extracts from HeLa cells transfected with plasmids of FLAG-tagged factors, and they were immunoprecipitated by anti-FLAG monoclonal antibody. Amounts of each SR protein bound to RNA of wildtype and mutant tau exon 2 (A) or exon 10 (E). Values were calculated relative to the nonspecific binding of the mock vector transfection (means Ϯ S.D. of three analyses). B and F, Western blots of immunoprecipitated proteins (IPP) from cell lysates that were transfected with each FLAG-tagged factor by anti-FLAG antibody in the presence of riboprobes for (B) exon 2 and (F) exon 10. Transfection efficiency and expression levels of each tagged SR protein are equivalent. A and B, SRp30c and SRp55, but not htra2␤1, bind directly to tau exon 2 pre-mRNA in HeLa cells, and their binding is altered in deletion mutants E2-⌬1, E2-⌬2, and E2-⌬4. C and D, SRp55 and SRp30c, but not htra2␤1, bind directly to the exon 2 pre-mRNA in UV-cross-linking assays. GST versions of the three proteins were UV-cross-linked to wild-type and deleted (E2-⌬1, E2-⌬2, and E2-⌬4) exon 2 riboprobes. Cross-linking of (C) SRp30c and htra2␤1 and (D) SRp55. The top band in D is full-length SRp55. The bottom band in D is a truncated degradation product, discussed under "Results." E and F, htra2␤1, SRp55, and SRp30c bind directly to tau exon 10 pre-mRNA in HeLa cells, and the binding of the latter two is abolished in mutant E10-⌬3/4.

Effects of Modulation of the Tau Projection Domain on the
Structure and Function of the Tau Protein-The tau exon 2/3 splicing events are developmentally regulated (18,(22)(23)(24). The resulting adult variants (2 ϩ 3 Ϫ , 2 ϩ 3 ϩ ) undoubtedly influence the conformation of the tau molecule, given the predominance of acidic residues encoded by the two exons. These changes alter tau interactions with other ligands and hence may regulate tau influence on axonal stability and neuronal plasticity.
The specific functions and ligands of the protein domains encoded by tau exons 2 and 3 are unknown. They are not involved in microtubule binding or spacing, functions that are localized to the C terminus and hinge region of tau and specifically influenced by the presence of exon 10 (18,53,54). The N terminus may be important for microtubule bundling because N-terminal deletion mutants display reduced capacity for microtubule bundling (55). The N terminus of the tau protein (also called the projection domain) interacts with the plasma membrane (20) and is phosphorylated by the non-receptor tyrosine kinase fyn (56).

Regulation of Tau Exon Splicing and Frontotemporal
Dementia-The accumulation of abnormal tau filaments into tangles is a hallmark of many neurodegenerative diseases, including Alzheimer disease. Formation of neurofibrillary tangles is an early event in the dementia cascade, and the number of neurofibrillary tangles correlates with disease severity. In several neurodegenerative diseases, now collectively termed tauopathies, tau pathology is solely and directly responsible for neuronal death and development of the clinical dementia manifestations (19).
The tauopathy pedigrees analyzed thus far predominantly show mutations in tau exon 10, although several pedigrees carry mutations in tau exons 1,9,11,12, and 13 that influence either microtubule binding or protein conformation (19). The exon 10 mutations fall in two categories, those that influence microtubule binding and those that alter the ratio of exon 10 isoforms. However, one atypical tauopathy (myotonic dystrophy type 1) is characterized by tau hyperphosphorylation, formation of intraneuronal aggregates, and significant reduction of tau isoforms containing exon 2 (28). Myotonic dystrophy type 1 arises from a CTG repeat extension at the 3Ј-untranslated region of the DM protein kinase gene (57). More recent studies indicate that exon 10 is also affected in myotonic dystrophy type 1, although its relative suppression is much less than that of exon 2 (58). Almost certainly, the repeat extension acts as a sink for the CELF family of splicing factors (52) that affect tau exons 2 and 10 (27,30). Thus, tau isoform ratios can cause neurodegeneration either directly (by cis mutations in tau exons) or indirectly (by variations in levels of trans factors that regulate tau exons).  8. SRp55, SRp30c, and htra2␤1 interact with each other. In vitro interaction assays were done using recombinant htra2␤1-GST, SRp55-GST, and SRp30c-His fusion proteins produced in E. coli and FLAG-tagged htra2␤1 produced in HeLa cells. A, htra2␤1 interacts with SRp55 (lane 5). htra2␤1-FLAG was expressed in HeLa cells (lane 2). FLAG and htra2␤1-FLAG proteins were incubated with GST and SRp55-GST proteins. GST alone/htra2␤1 and FLAG alone/SRp55 were used as the negative controls (lanes 3 and 4). B, SRp30c interacts with SRp55 and htra2␤1 (lanes 6 and 7). Fusion proteins were expressed in E. coli (lanes 1-4). SRp30-His was incubated with GST alone and GST fusion proteins. GST alone was used as the negative control (lane 5).
FIG. 9. Speculative model of splicing regulation for the silencers located at the 5 end of tau exons 2 and 10. The cellular functions of the tau domains are shown below the diagram. For the exons, shaded regions represent enhancers, and plaid regions represent silencers (not drawn to scale). For the factors, circles represent RRM domains, and squares represent RS domains. A, exon 2. SRp30c directly binds to regions E2-⌬1 and E2-⌬2. SRp55 directly binds to region E2-⌬2 and indirectly influences regions E2-⌬1 and E2-⌬4 through SRp30c and an unknown activator, respectively. SRp30c, SRp55, and htra2␤1 bind to each other through their RS domains (in the case of htra2␤1, the RS1 domain). B, exon 10. SRp30c and SRp55 bind to region E10-⌬2/3/4, whereas htra2␤1 binds to region E10-⌬5/6. The two inhibitors interact with (or sterically interfere with) the RS1 domain of htra2␤1; its RS2 domain may interact with a putative co-activator that may bind to either E10-⌬6/7 or E10-⌬10/11/12. Continued work on the basic molecular biology of the tau molecule may give us the tools to comprehend and combat not only FTDP but also other types of dementia. These diseases vary widely both in clinical phenotype and brain pathology, but they share tangles as an invariable defining characteristic (19).