Determinants of 4-Repeat Tau Expression

Mutations in the tau gene are pathogenic causing autosomal dominant frontotemporal dementia with Parkinsonism-chromosome 17 type (FTDP-17). Some mutations intau exon 10 (E10) and immediately adjacent sequences cause disease by altering E10 splicing. To determine the mechanism of normal E10 splicing regulation and how FTDP-17 mutations alter splicing, mutational analysis of E10 was performed. The results show that E10 contains a complex array of both enhancer and inhibitorcis-acting elements that modulate usage of a weak 5′ splice site. The 5′ end of E10 contains a previously unrecognized multipartite exon splicing enhancer (ESE) composed of an SC35-like binding sequence, a purine-rich sequence, and an AC-rich element. Downstream of this ESE is a purine-rich exon splicing inhibitor. Intronic sequences immediately downstream of E10 also are inhibitory. The results support an alternative model in which I10 inhibitory sequences appear to function as a linear sequence. The cis-elements described are not redundant, and all appear required for normal E10 splicing. Results with double mutations demonstrate that the ESE and the intronic inhibitory element collaborate to regulate splicing. Thus splicing oftau E10 is regulated by a complex set ofcis-acting elements that span nearly the entire exon and also include intronic sequences.

Mutations in the tau gene are pathogenic causing autosomal dominant frontotemporal dementia with Parkinsonism-chromosome 17 type (FTDP-17). Some mutations in tau exon 10 (E10) and immediately adjacent sequences cause disease by altering E10 splicing. To determine the mechanism of normal E10 splicing regulation and how FTDP-17 mutations alter splicing, mutational analysis of E10 was performed. The results show that E10 contains a complex array of both enhancer and inhibitor cis-acting elements that modulate usage of a weak 5 splice site. The 5 end of E10 contains a previously unrecognized multipartite exon splicing enhancer (ESE) composed of an SC35-like binding sequence, a purine-rich sequence, and an AC-rich element. Downstream of this ESE is a purine-rich exon splicing inhibitor. Intronic sequences immediately downstream of E10 also are inhibitory. The results support an alternative model in which I10 inhibitory sequences appear to function as a linear sequence. The cis-elements described are not redundant, and all appear required for normal E10 splicing. Results with double mutations demonstrate that the ESE and the intronic inhibitory element collaborate to regulate splicing. Thus splicing of tau E10 is regulated by a complex set of cis-acting elements that span nearly the entire exon and also include intronic sequences.
Tau is a microtubule-associated protein that in vitro promotes microtubule assembly and stability. In vivo, tau is important in neuronal morphogenesis, axon polarity, and axonal transport (1,2). In the central nervous system, tau is expressed primarily in neurons, although it is also present in glial cells at lower levels. During central nervous system development, the gene encoding tau (tau) is highly regulated by alternative exon splicing. In the human fetal brain, a single tau isoform is produced consisting of exons 1, 4, 5, 7, 9, and 11-13 (3). In the adult human central nervous system, 6 tau isoforms are made from the same exons plus alternative splicing of exons 2, 3, and 10 (4). tau exon 10 (E10) 1 encodes a microtubule-binding motif that is repeated 3 times in tau isoforms lacking E10 sequences (3 repeat or 3R tau) and 4 times when E10 is present (4R tau).
Mutations in tau cause frontotemporal dementia with Parkinsonism-chromosome 17 type (FTDP-17) (5-7), a group of autosomal dominantly inherited neurodegenerative disorders with varied clinical and neuropathologic phenotypes. A common consequence of most if not all tau mutations is abnormal filamentous aggregates of tau in central nervous system neurons and in some cases in glial cells (8). For some mutations, the aggregated tau is present as paired helical filaments in neurofibrillary tangles. These tangles are composed of all 6 tau isoforms and are structurally and biochemically indistinguishable from those found in Alzheimer's disease (AD) (9,10). For other FTDP-17 mutations, tau is present as paired straight filaments (11) or paired helical filaments with different dimensions than those of AD filaments (12,13). Also the isoform ratio can be different from that found in AD.
tau missense mutations fall into 2 categories as follows: mutations that alter protein function, and mutations that affect alternative splicing of E10. Mutations G272V (E9), P301L, P301S, ⌬280K (E10), V337M (E12), and R406W (E13) reduce the affinity of tau binding to microtubules (14) and/or reduce the ability of tau to promote microtubule assembly when compared with normal tau (11, 14 -17). Thus, these mutations alter tau protein function. Since E9, E12, and E13 are constitutively included in tau transcripts, mutations in these exons affect all tau isoforms, whereas E10 mutations P301L and P301S only affect 4R tau function. E10 mutations N279K and S305N do not alter the ability of tau to interact with microtubules but do affect the regulation of E10 splicing (15,18). In splicing assays, the N279K and S305N mutations result in almost constitutive inclusion of E10. Similarly, the L284L E10 silent mutation and intron 10 (I10) mutations immediately adjacent to the 3Ј end of E10 also increase E10 inclusion (6,15). Another FTDP-17 mutation, an in-frame 3-base deletion in E10 (⌬280K), is unique in that it alters tau protein function and also completely abolishes E10 inclusion in tau transcripts (15,17). The effect of ⌬280K on splicing is presumed to be the disease-causative mechanism since, with E10 being completely excluded, no ⌬280K protein would be produced. The mutations that affect E10 splicing cause FTDP-17 by altering the 4R/3R tau ratio, and in some cases, the elevated 4R tau produced has the normal tau amino acid sequence (e.g. L284L and I10 mutations; Fig. 1) (14,19). As expected, when E10 usage is increased, the aggregated tau formed has excess 4R tau isoforms (7,14,19). Different mutations that increase E10 usage result in different clinical and neuropathologic phenotypes, presumably because these mutations affect different cis-acting regulatory elements controlling E10 splicing. Thus, understanding the mechanisms that regulate E10 usage is important to understanding FTDP-17.
Our previous work examining the effects of FTDP-17 mutations on tau E10 splicing indicated that multiple cis-acting sequences regulate E10 alternative splicing (15). An additional intronic inhibitory element immediately adjacent to the end of E10 in I10 was revealed by other FTDP-17 mutations (6, 7). Here we examine exon and intron sequences that control tau E10 inclusion. We performed mutation and deletion studies to identify splicing regulatory sequences that control E10 splicing and to explore the functional interactions between elements in orchestrating E10 usage. Our results reveal that there are at least 4 cis-acting regulatory sequences within E10 that are non-redundant and that function cooperatively to regulate E10 inclusion. These E10 cis-elements interact with a novel I10 inhibitory element that potentially acts by forming a complex secondary structure, although evidence that the inhibitory element appears to function as a linear sequence is presented here. Correct regulation of tau E10 is the result of contributions from each of these elements.

EXPERIMENTAL PROCEDURES
Plasmid Construction and DNA Mutagenesis-Vector pSPL3 (Life Technologies, Inc.) contains an HIV genomic fragment with truncated tat exons 2 and 3 inserted into rabbit ␤-globin coding sequences (20). The resulting hybrid exons in pSPL3 are globin E1E2-tat exon 2 and tat exon 3-globin E3 separated by more than 2.5 kilobase pairs of tat intron sequence. pSPL3 contains a multiple cloning sequence (MCS) around 300 nucleotides downstream of the tat exon 2 5Ј splice site. The SV40 promoter and polyadenylation signal allow for enhanced expression in COS-7 cells. pSPL3 was further modified here by deleting a 500-base pair (bp) tat intronic fragment between the EcoRV (nucleotide 1050) and StuI (nucleotide 1550) sites (GenBank TM accession number U19867) to generate pSPES. This deletion removes a previously identified cryptic exon (20). Note that the portion of HIV tat exon 3 used in pSPL3 does not contain the splicing inhibitory sequence TTAG (21) that is also present in tau E10. A 177-bp genomic fragment containing human tau E10 with 33 and 51 bp of flanking intron sequences was amplified from P1 artificial chromosome clone 4139 by PCR using forward and reverse primers I9X2 (5Ј-CCACTCGAGCGTGTCACT-CATCCTTTTTC-3Ј) and I10B2 (5Ј-CGGGATCCTAATAATTCAAGCCA-CAG-3Ј) that contain an XhoI and BamHI site, respectively. The XhoI/ BamHI-digested PCR product was inserted into the pSPES MCS (XhoI/ BamHI) to generate hN (Fig. 1). Similarly, a 174-bp mouse genomic fragment containing E10 with 30 and 51 bp of flanking I9 and I10, respectively, was amplified from mouse genomic DNA by PCR and inserted into the XhoI/BamHI sites of pSPES to generate mWT. The mouse and human 3Ј ends of E10 (the last 13 nucleotides beginning at a SmaI site) along with the first 51 nucleotides of I10 (ending at a BamHI site) were exchanged by excising this section as a SmaI-BamHI fragment. All nucleotide changes and deletions in tau sequences were performed by standard PCR using a mutagenized primer with primers I9X2 or I10B2. PCR products were digested with XhoI/BamHI and ligated into pSPES. The DNA sequence of the normal human and mouse inserts and all sequence alterations were confirmed by dye terminator cycle sequencing with TaqFS DNA polymerase (Perkin-Elmer) using an ABI377 DNA sequencer.
When the cryptic exon present in pSPL3 was removed to create pSPES, splicing experiments revealed an additional cryptic splice site between the cloning site and tat exon 3. This new site is not used when no test exon is placed in pSPES. When the normal tau E10 or mutant E10 with a weak 5Ј splice site is inserted, small amounts of E10 join with this cryptic site yielding a ϳ60-bp product, although most of E10 is joined to tat exon 3, yielding a 354-bp product. In calculating the amount of E10 included in the final transcript, the amount of E10 joined to the cryptic site was included in the total.
Cell Culture, Transient Transfection, and RNA Isolation-Cell culture and transfection reagents were from Life Technologies, Inc. COS-7 cells were maintained in Dulbecco's modified Eagle's medium supple-mented with 10% fetal calf serum. Transient transfections were performed in triplicate using 1 g of plasmid DNA with 6 l of Lipo-fectAMINE in a total of 700 l of Opti-MEM per 35-mm well. Cells were exposed to the lipid-DNA complex for 5 h at 37°C in a 5% CO 2 incubator and allowed to recover with 700 l of Dulbecco's modified Eagle's medium containing 20% fetal calf serum. Total cellular RNA was isolated 48 h later with TRIzol (Life Technologies, Inc.). RNA samples were DNase I-treated (Amersham Pharmacia Biotech) prior to reverse transcription.
Quantitation of Tau E10 Splicing by Reverse Transcription-PCR (RT-PCR)-Splicing was assayed essentially as described previously (15). Total RNA (2-2.5 g) was reverse-transcribed with random hexamers using the GeneAmp RNA PCR kit (Perkin-Elmer). PSPESspliced products were amplified by PCR using forward and reverse primers SD6 and SA2, respectively (Life Technologies, Inc.), that are specific for rabbit ␤-globin sequences in the vector. 1 ng of 32 P-labeled SA2 was included in PCRs that were performed for 18 cycles to obtain linear amplification. Products were resolved on 5% acrylamide gels and quantitated using a PhosphorImager. For the normal tau E10 construct hN, the percent E10 inclusion is the mean value from eight independent transfection experiments performed on different days. For each mutant construct, values presented are the average of at least three different transfection experiments. Also, for each transfection experiment, hN was transfected in parallel, and the experiment was considered valid if the value for hN was Ϯ2 S.D. of 45% (36.6 -51.5%). Statistical comparisons were made using a two-tailed Student's t test. Criteria for significance were calculated using a Bonferroni correction for multiple comparisons by dividing the initial p value of 0.05 by the number of comparisons made.
RNA Structure Analyses-RNA secondary structures and free energies were predicted using the GCG version of the MFOLD program (22). The free energies of the most stable structures computed for each sequence were compared.

RESULTS
Previous work on the effects of FTDP-17 mutations on tau E10 splicing showed that mutations at nucleotides 15-18 within a purine-rich region in E10 and a mutation at nucleotide 30 altered splicing of this exon (15). To identify the regulatory elements affected by these mutations and to reveal other potential regulatory sequences, we generated in-frame 3 nucleotide deletions from the beginning of E10 exon to nucleotide 33 (E⌬1-E⌬11). Nine nucleotide deletions were generated from nucleotide 34 to the end of the exon (E⌬12-E⌬17) (Fig. 1). Additional substitutions were used to identify critical nucleotides and to determine if effects seen with deletions were due to removal of critical sequences or caused by changes in distances between other cis-acting elements. Splicing was assayed by inserting the normal or modified E10 with 33 nucleotides of I9 and 51 nucleotides of I10 ( Fig. 1) between HIV tat exons 2 and 3. When the normal E10 is assayed in COS-7 cells, 45% of the transcripts contain E10 (Fig. 1). The cis-acting regulatory elements identified are described individually below.
Purine-rich ESE Element-Deletions E⌬5 to E⌬7 severely reduce E10 inclusion ( Fig. 1) indicating the presence of an ESE sequence. This ESE is designated the polypurine enhancer (PPE) because of the high purine content of this sequence (AAGAAGCTG) and is the cis-acting element affected by FTDP-17 mutations N279K and ⌬280K. E⌬5 and E⌬6 are structurally equivalent and the same as FTDP-17 mutation ⌬280K. This deletion results in no detectable E10ϩ transcript. The downstream flanking deletion E⌬7 reduces the splicing ratio, whereas E⌬8 does not, thus defining the 3Ј limits of the PPE. The 5Ј end of the PPE is more difficult to define. Deletions E⌬3 and E⌬4 result in higher than normal E10 inclusion, suggesting that these sequences may inhibit function of the PPE.
The sequence requirements for PPE function were explored with single nucleotide substitutions alone or in combination with the E⌬5 deletion (Fig. 2). Mutation 15T3 G (FTDP-17 mutation N279K) increases E10 incorporation to 79%. The 15T3 G change possibly acts by increasing the number of GAR sequences (R is a purine) from 1 to 2 when compared with the normal sequence or by increasing the number of AAG repeats from 2 to 3. Both GAR and AAG repeats are known to enhance splicing in other systems (23,24). Compatible with both explanations is that E⌬5, which eliminates E10 inclusion, deletes the lone GAR repeat and 1 of the 2 normal AAG repeats. When an A substitution is used rather than a G (15T3 A), E10 inclusion is increased but to a lesser extent than with the 15T3 G change (61 compared with 79%, respectively). Likewise, in double mutants when the normal function of the PPE is severely compromised by E⌬5, the G substitution gives much higher E10 incorporation than the A substitution (71 versus 16%, respectively). In the context of the deletion, the G substitution, like the normal sequence, maintains 1 GAR repeat and 2 AAG repeats, whereas the A substitution yields only 1 AAG. Thus, although both mutations increase the number of consecutive purines in this region by 3 nucleotides, clearly the types of purine and thus the specific sequence of the element are critical.
AC-rich ESE Element-Deletions E⌬9 and E⌬10 were de-FIG. 1. Deletion analysis of tau E10. A, nucleotide sequence of tau E10. Exon nucleotides are in capital letters and intron nucleotides in lowercase letters. FTDP-17 mutations are indicated by solid arrows, except for ⌬280K which is shown with a triangle spanning the deleted nucleotides. E10 sequence differences between human and mouse are shown above the sequence (open arrows), and mouse nucleotides are in italics (differences in intronic sequences are not shown). For FTDP-17 mutations and substitutions used in this study, the location of E10 nucleotide changes is indicated by the nucleotide number within the exon (the first nucleotide at the 5Ј end of E10 is 1). The normal and the substituted nucleotide are shown in superscript. Thus the FTDP-17 mutation in E10 at nucleotide 15 where the normal base is a T and the mutant base is a G is indicated as 15T3 G. Amino acid change locations are listed with the normal amino acid shown in superscript before the codon number and the mutant amino acid shown after the codon number. Codon numbering is based on the longest common tau isoform in brain containing E2, E3, and E10. The locations of I10 mutations are shown after a ϩ symbol as the nucleotide number in I10 with the first nucleotide of the intron being 1 (e.g. E10ϩ16c3t). The normal and mutant nucleotides are shown in superscript. Deletions used for E10 analysis are indicated with bars directly below the sequence and are labeled E⌬1-E⌬17. The nucleotides present in a potential stem-loop are shown with a bar above the sequence with the loop region as a dotted line. The locations of potential cis-acting regulatory regions are shown as hatched boxes below the sequence. B, autoradiograph of E10 splicing analyzed by RT-PCR. The E10 minus (E10Ϫ) fragment is 261 bp and the E10ϩ fragment is 354 bp. A longer exposure of the last 4 lanes (E⌬14 -E⌬17) is presented below the original exposure. C, quantitation of E10 splicing. Radioactivity in each band was quantitated using a PhosphorImager. Each bar represents the mean of at least three separate transfection experiments and 100% is the sum of E10Ϫ and E10ϩ. Error bars are standard deviations. A corrected significance criteria of p Ͻ 0.003 was used, and p values for comparison of each construct to normal human E10 are indicated with the following symbols: ‡, p Ͻ 1 ϫ 10 Ϫ3 ; , p Ͻ 1 ϫ 10 Ϫ4 ; †, p Ͻ 1 ϫ 10 Ϫ5 ; *, p Ͻ 1 ϫ 10 Ϫ6 ; §, p Ͻ 1 ϫ 10 Ϫ7 ; and ϩ, p Ͻ 1 ϫ 10 Ϫ8 . signed to disrupt the region of E10 affected by FTDP-17 mutation L284L (30T3 C). This mutation alters the sequence TTAG, which acts as an ESS sequence in HIV-1 tat exon 3 (21). These deletions were expected to increase splicing due to deletion of the putative TTAG silencer. However, unlike mutation 30T3 C that increases E10 inclusion (87%, Fig. 2), these deletions either gave near normal splicing (E⌬9, 51%) or substantially reduced E10 inclusion (E⌬10, 17%; Fig. 1). E⌬11 and E⌬12, like E⌬10, also reduce the amount of E10ϩ produced (6 and 33%, respectively). Thus, in the tau E10 context, unlike in HIV-I tat exon 3, TTAG does not function as an ESS. Rather, the sequence in the E⌬9-E⌬12 region contains an ESE that may be an AC-rich enhancer (ACE) (25), particularly when the 30T3 C mutation is present. This region of E10 resembles ACE elements observed in other systems (Fig. 2) (26). In agreement with this hypothesis, transitions within the TTAG sequence that increase the AC content (29T3 C and 32G3 A), like 30T3 C, increase E10 inclusion with the T3 C substitutions yielding the largest increase. However, 31A3 G, which decreases the AC content, also increases E10 inclusion. Thus the sequence requirements for this ESE are more complex than simply a high AC content.
Additional cis-Acting Sequences in E10 -Deletions E⌬14 and E⌬15 markedly increase E10 inclusion (69 and 90%, respectively, Fig. 1), suggesting removal of a previously unknown inhibitory element at the 3Ј end of the exon. Flanking deletions resulted in either slightly elevated (E⌬13) or slightly reduced (E⌬16 and E⌬17) E10 incorporation, indicating that the putative inhibitory element is primarily confined to the E⌬14/E⌬15 segment. A series of point mutations (14a, 15a, and 15b, Fig. 3) were introduced into the same region replacing the clusters of As with G, C, and T nucleotides. Each different substitution as well as combinations of substitutions increased E10 incorporation (Fig. 3). These point mutations show that the E⌬14/E⌬15 region contains a purine-rich ESS and that the effects seen with these mutations are not simply due to changes in E10 length or altering the proximity of other cis-acting elements. Purine-rich ESSs downstream of an ESE are also present in the human fibronectin EDA exon (27) and in HIV tat exons 2 and 3 (28,29).
Another potential regulatory region is at the 5Ј end of E10. E⌬1 dramatically reduces E10 inclusion to 12% (Fig. 1). This deletion alters the sequence TGCAGAT that resembles the degenerate binding consensus TGCNGYY for the SR protein SC35 (30). Deletion of the AT-rich sequences in E⌬3 and E⌬4 results in increased E10 inclusion suggesting that either an inhibitory sequence has been removed or that a reduction in the distance between the SC35-like and the purine-rich ESE sequences is important for both to function.
Role of the 5Ј Splice Site in E10 Inclusion-The E10 5Ј splice site is predicted to be weak because the sequence (GTgtgagt) differs from the optimal consensus sequence (AGgt(a/g)agt) for U1/U6 snRNP binding (31). The role of the 5Ј splice site and immediately adjacent exon sequences was investigated using single nucleotide substitutions in the 5Ј splice site and by replacing tau E10 sequences with the 3Ј end of human ␤-globin E2. The ␤-globin E2 sequences used were the last 10 nucleotides of the exon and the first 6 nucleotides of the intron (the entire 5Ј splice site). This constitutively spliced middle exon was chosen because its 5Ј splice site can easily be converted to that of tau E10 by altering the Ϫ1 position. Predictably, the hybrid construct spliced to completion as it contains a strong 5Ј splice site GGgtgagt (Fig. 4). However, when a single nucleotide substitution changes the Ϫ1 from the ␤-globin to the tau nucleotide (Glo Ϫ1), E10 inclusion returns to levels seen with normal tau (39%). Likewise, changing the last nucleotide of tau E10 to a G (93T3 G) as in globin E2 strengthened the 5Ј splice site, and constitutive E10 splicing was observed. The same results are seen when the penultimate E10 nucleotide is changed to strengthen the splice site sequence (FTDP-17 mutation 92G3 A, Fig. 5). Thus a strong 5Ј splice site sequence overrides other inhibitory elements in E10 and the intron splicing silencer (ISS, see below) in I10.
The 5Ј Splice Site-ISS Interaction-Intronic FTDP-17 mutations (E10ϩ3g3a, E10ϩ12c3t, E10ϩ13a3g, E10ϩ14c3t, and E10ϩ16c3t; Fig. 1) immediately adjacent to the 5Ј splice site dramatically increase E10 inclusion, demonstrating that this region is an ISS element. One proposed mechanism for how this ISS functions is that the normal sequence results in a stem-loop spanning the E10/I10 junction; this structure sequesters the 5Ј splice site, inhibiting binding to U1 snRNP (32) and subsequent splicing (33). In this model, the FTDP-17 intronic mutations and the E10 mutation 92G3 A enhance E10 inclusion by disrupting base pairing in this stem-loop. Three different stem-loops have been proposed, each containing dif- FIG. 3. Substitution analysis within the E10 ESS sequences. A, E10 ESS sequence from nucleotides 55-72 with the E⌬14 and E⌬15 deletions and additional substitutions. Filled arrows above the exon sequence show contiguous A residues that were substituted in constructs 14a, 15a, and 15b. Mutant 15ab contains substitutions from both 15a and 15b, and mutant 14aϩ15ab contains substitutions from both 14a and 15ab. For the single nucleotide difference in this region between human and mouse, the mouse nucleotide is shown with an open arrow. B, autoradiograph of RT-PCR showing E10 inclusion for substitution mutants. Assays were performed as described in Fig. 1. C, quantitation of E10 inclusion. A corrected significance criteria of p Ͻ 0.007 was used. Significance levels for each mutant were compared with normal human E10 are indicated with symbols used in Fig. 1. A corrected significance criteria of p Ͻ 0.007 was used. Significance levels are for comparison of each construct to the normal human construct (hN). For WTGlo and 93 T3 G no E10-transcript was observed. ferent lengths of E10 sequences (Fig. 6). The longer versions extend into E10 raising the possibility that the inhibitory element includes both exon and intron sequences. However, E⌬17 reduces E10 inclusion to 26% (Fig. 1), a result that is inconsistent with these exon sequences being part of an inhibitory element. Also, substitutions Ϫ4⌬Ϫ9 and Ϫ5⌬Ϫ9 (Fig. 4) introduced into the E10 sequences immediately adjacent to the 5Ј splice either did not alter E10 inclusion (Ϫ4⌬Ϫ9) or only slightly increased E10 inclusion to 58%. The substituted nucleotides were mostly transitions, chosen so as not to base pair with their corresponding partners in I10. These substitutions and the E⌬17 deletion would be expected to disrupt the lower portions of the 25-and 30-nucleotide stem-loop structures (Fig.  6A) without disrupting the 18-nucleotide structure at the top. These results show that either secondary structure is not important for the inhibitory activity of the ISS or that only the shortest stem-loop is important for regulation.
Interactions between cis-Acting Elements-The interactions between different regulatory sequences were examined by comparing E10 splicing in constructs with mutations in single elements to constructs with mutations in two different elements. The E⌬5 mutation was used to abolish function of the PPE. FTDP-17 mutation 92G3 A or E10ϩ3g3a was used to strengthen the 5Ј splice site. Double mutants 92G3 A/E⌬5 and E10ϩ3g3a/E⌬5 yielded high levels of E10 inclusion (80 and 87%, respectively) demonstrating that in the presence of a strong 5Ј splice site, the PPE is not required for efficient E10 splicing (Fig. 5). When the FTDP-17 ISS mutations E10ϩ12c3t, E10ϩ13a3g, E10ϩ14c3t, and E10ϩ16c3t were combined with E⌬5, E10 inclusion was reduced compared with the results when only the ISS mutations were present. Thus the ISS and the PPE must collaborate to regulate E10 inclusion. Removal of the PPE with the E⌬5 mutation had varied effects when combined with the different ISS mutations. The E⌬5 mutation with E10ϩ12c3t or E10ϩ14c3t only modestly reduced E10 inclusion compared with results for only the single E10ϩ12c3t or E10ϩ14c3t mutations; E10 inclusion was still greater (59 and 65%, respectively) than seen with the normal E10 sequence. In contrast, for double mutants containing E10ϩ13a3g and E10ϩ16c3t, E10 inclusion was dramatically reduced (28 and 26%, respectively) to levels below those seen with the normal sequence. These results imply that not all nucleotides within the ISS are equivalent. The PPE also interacts with the ACE element. The effect of the FTDP-17 mutation 30T3 C that strengthens the AC-rich ESE and enhances E10 inclusion is modulated when the PPE is abolished in the double mutant 30T3 C/E⌬5 (Fig. 2).
Comparison of Human and Mouse E10 Splicing-Mice and humans regulate E10 splicing differently. In fetal brain in both organisms, E10 is not incorporated into tau transcripts. In adult mouse brain, all tau transcripts have E10, and only 4R tau protein is produced. In contrast, in adult human brain, both E10ϩ and E10Ϫ tau transcripts and 3R and 4R tau protein are found in roughly equal proportions (14). E10 is highly conserved between these two species with only 3 nucleotides being different (Fig. 1A), and the 5Ј and 3Ј splice sites are identical. However, the adjacent intronic regions are more divergent (Fig.  7). To assess the role of these divergent sequences on E10 regulation, a wild-type mouse splicing template (mWT) was generated that is analogous to the human sequences used here and is comprised of mouse E10 (E M ) flanked by 30 and 51 nucleotides from mouse I9 (I9 M ) and I10 (I10 M ), respectively. In COS-7 cells, the efficiency of E10 inclusion (34%) was somewhat less than the human equivalent.
To evaluate the effects of mouse and human intronic sequence differences on splicing, constructs were generated with different combinations of human and mouse sequences (Fig. 7). These combinations permitted evaluation of the role of I10 and E10 sequence differences separately. Mouse I10 sequences (I10 M ) are substantially more inhibitory than I10 H sequences; when human I9 and E10 (I9 H E H ) are coupled to I10 M , splicing is reduced (compare hN to I9 H E H I10 M ; 45 and 5%, respectively). Conversely, when mouse I9 and E10 sequences are coupled to I10 H , E10 incorporation is increased (compare mWT to I9 M E M I10 H ; 34 and 61%, respectively). The same strong inhibition by I10 M is also seen when mouse and human divergent FIG. 5. Interactions between exon and intron splicing elements. The effects of 5Ј splice site (92G3 A) and intronic (E10ϩ12, E10ϩ13, E10ϩ14, and E10ϩ16) FTDP-17 mutations on E10 splicing were analyzed in the absence of a functional ESE sequence. ESE function was disrupted using the E⌬5 mutation. A, autoradiograph showing results for E10 splicing assays for normal and mutant constructs. B, quantitation of E10 inclusion. A corrected significance criteria of p Ͻ 0.002 was used. Assays were performed as in Fig. 1. Significance levels for comparison of each mutant to normal human E10 are indicated by symbols above bars as described in Fig. 1 Exon sequence differences between human and mouse tau also affect splicing. When in hN, E10 nucleotide 57 is changed from A found in humans to G found in mouse (I9 H E GAC I10 H ), and E10 inclusion is increased from 45 to 89%. This result is consistent with an ESS element in the E⌬14/E⌬15 region of E10 (Fig. 3). When mouse E10 nucleotides at positions 84 and 87 were inserted into hN, splicing was reduced to 14% (compare hN to I9 H E ATA I10 H ; Fig. 7). Human nucleotides at positions 84 and 87 were also replaced individually to determine which is responsible for the decrease in splicing. At position 87, when the human C nucleotide was replaced with the mouse A, little change in splicing was observed (compare hN to I9 H E AAA I10 H , 45 versus 50%, respectively, Fig. 7). In contrast, the single nucleotide replacement of the human A at position 84 with either the mouse T or another pyrimidine (C) resulted in a dramatic reduction in splicing to 11 and 5%, respectively (compare hN to I9 H E ATA I10 H and I9 H E ACA I10 H ; Fig. 7). Thus, nucleotide 84 is critical, and the reduction in splicing observed when this nucleotide is changed is consistent with results seen with E⌬16 and E⌬17 (Fig. 1) and suggests that a stimulatory element is present immediately upstream of the end of E10. However, presently we cannot exclude the possibility that deletions E⌬16 and E⌬17 affect splicing by bringing the E10 ESS closer to the 5Ј splice site.
When human nucleotides were inserted into mWT at positions 84 and 87, E10 inclusion was reduced (compare mWT to I9 M E GAC I10 M , 34 versus 20%, respectively). However, the inhibitory effects of the substituted nucleotides were overridden when the downstream intron was human (compare I9 M E GAC I10 M to I9 M E GAC I10 H ; 20 and 93%, respectively). In contrast, no change in splicing occurred on replacing the mouse G at nucleotide 57 for the human A at this position (compare mWT to I9 M E ATA I10 M ; 34 and 35%, respectively). Presumably the strong inhibitory effect of mouse I10 overrides other elements (e.g. the ESS) in mouse E10. The mouse and human I9 sequences used here also differentially affect E10 splicing. The mouse I9 sequence is more permissive for E10 inclusion when compared with the human I9 sequence (compare I9 M E ATA I10 M to I9 H E ATA I10 M ; 35 and 3%, respectively). DISCUSSION The ratio of tau E10ϩ to E10Ϫ transcripts is regulated by a complex combination of intronic and exonic sequences. Nearly the entire exon is involved in controlling the inclusion of E10 in tau transcripts. The in vivo importance of this complex regulation is revealed by FTDP-17 mutations that disrupt critical cis-acting elements within or closely flanking E10. In the adult under normal conditions, the E10ϩ/E10Ϫ ratio is close to 1 (14). Autosomal dominant FTDP-17 splicing mutations that only affect 1 copy of the tau gene, elevate this ratio to 2:1 (e.g. N279K mutation) (14,19) or 3:1 (I10 mutations) (6, 7) or decrease this ratio to 1:3 (⌬280K mutation) (15,17). The consequence of these relatively subtle changes is onset of severe neurodegenerative disease in mid-life. The work described here demonstrates that the regulation of E10 splicing is complex involving both ESE and ESS elements within the exon. In addition, intronic sequences immediately adjacent to the 3Ј end of E10 inhibit splicing. Other more distant intronic sequences not studied here also affect E10 splicing.
The deletion and substitution analysis described above demonstrates that tau E10 splicing is regulated by a multipartite ESE composed of at least three functional motifs as follows: a potential 5Ј SC35-like binding element, a central purine-rich enhancer (PPE), and a 3Ј ACE-like element. In addition, results from deletions E⌬3 and E⌬4 show that the sequence between the SC35-like element and the PPE element is critical for splicing regulation and is a potential inhibitory element. Deletions in each ESE motif reduce E10 inclusion indicating that all three ESE motifs are required for the level of E10 inclusion observed for normal tau. Thus, these individual ESE components are not functionally redundant. Whereas only the PPE element appears to be absolutely required for E10 splicing as shown by results from the E⌬5/E⌬6 deletion, correct regulation of E10 splicing requires that all elements be present for exon definition. Double mutation experiments show that when the PPE is silenced by the E⌬5 deletion, either other enhancer sequences must be strengthened or inhibitory sequences weakened before E10 can be recognized. Although in the present work, the function of ESE was defined in COS-7 cells, this ESE clearly functions in vivo because the N279K mutation in this ESE results in an increase of 4R tau (19). This increase in the 4R/3R ratio at the protein level is consistent with this mutation causing increased E10 inclusion and resulting FTDP-17. Also, two other mutations in the ESE (⌬280K and L284L) cause FTDP-17.
The PPE element has a core requirement for GAR or possibly AAG repeats. The mechanism by which the FTDP-17 mutation N279K (15T3 G) causes increased E10 inclusion is somewhat difficult to define because this mutation is at the border between the PPE and a potential inhibitory element defined by deletions E⌬3 and E⌬4. This mutation appears to act by extending the purine tract of the PRE as either an A or G at position 15 increases E10 inclusion. The G at position 15 is clearly more effective than an A as seen when either are assayed alone or in combination with the E⌬5 deletion (Fig. 2). The normal T at position 15 may negatively modulate the activity of PPE. Others (23) have shown that a T within a purine-rich enhancer sequence can be inhibitory.
The splicing enhancing effects of the FTDP-17 mutation L284L were originally hypothesized to be due to disruption of an ISS with the core sequence of TTAG (15). This prediction was based on previous work (21) showing that this core sequence inhibits splicing of HIV tat exon 3. However, deletions E⌬9 and E⌬10 that span the TTAG sequence, rather than increasing E10 inclusion, do not alter or actually reduce E10 inclusion, respectively (Fig. 1). Deletions E⌬10 and E⌬11 re-duce splicing, demonstrating that this region contains an enhancer element. Mutations introduced to alter each nucleotide of TTAG all increase E10 inclusion (Fig. 2). These results suggest that the TTAG inhibitory sequence is directly adjacent to an enhancer region (ACE, Fig. 1). The FTDP-17 mutation L284L (E30T3 C) that increases splicing may act by disrupting the inhibitory TTAG sequence and by extending the ACE element to include an additional CA repeat. Thus, both FTDP-17 mutations L284L and N279K are at critical nucleotides between closely adjacent elements that regulate E10 splicing. An alternative explanation is that the TTAG sequence, within the context of tau E10, does not strongly influence splicing, as shown by the fact that E⌬9 that deletes the first two nucleotides of this sequence only slightly increases splicing (Fig. 1).
The multipartite ESE and the inhibitory sequences within E10 act in the context of a weak 5Ј splice site. When the 5Ј  . 1A) are indicated as a superscript above E10 with E10 AAC being the human sequence and E10 GTA being the mouse sequence. SD is standard deviation. A corrected significance criteria of p Ͻ 0.003 was used. Significance levels are given for comparisons for each construct to results for both hN and mWT constructs. splice is strengthened either by mutations that bring the sequence closer to the U1/U6 snRNP hybridization consensus sequence (e.g. FTDP-17 mutations 92G3 A) or when the splice site is replaced by a ␤-globin splice site known to give constitutive splicing (e.g. WTGlo), E10 inclusion is nearly complete. When a strong splice site is present, destruction of the ESE function by E⌬5 only slightly decreases E10 splicing (Fig. 5), showing that a strong splice site is dominant over other regulatory elements.
The downstream I10 sequence immediately adjacent to the 5Ј splice site is clearly inhibitory. One hypothesis on how this sequence acts is that a stem-loop forms that spans the splice site, blocks hybridization of U1/U6 snRNP, and inhibits initiation of splicing. A second hypothesis is that a stem-loop is the binding substrate for a trans-acting factor. One ambiguity of these hypotheses is that the stem-loop structure and stability is dependent on how many nucleotides from E10/I10 are included ( Fig. 6 and Table I). Even more extensive structures are predicted when additional flanking nucleotides are included (not shown). Although at least some of these stem-loops do form in vitro (32,33), it is not clear if any of these structures form in vivo where multiple splicing regulatory proteins (e.g. SR proteins) coat the pre-mRNA, potentially altering or abolishing predicted secondary structure. An alternative hypothesis is that the I10 linear sequence is the critical functional inhibitory unit that acts as a single-stranded sequence and binds a transacting splicing factor(s).
Evidence for the stem-loop hypothesis is that FTDP-17 mutations predicted to disrupt the stem-loop increase E10 inclusion. Thus, mutations within the splice site (92G3 A and E10ϩ3) and those 3Ј to the splice site (E10ϩ12, E10ϩ13, E10ϩ14, and E10ϩ16) destabilize the stem-loop and increase splicing from the normal 45 to 88 -97%. Interpretation of the results for 92G3 A and E10ϩ3 is complicated because these mutations may act by directly strengthening the 5Ј splice site interaction with U1/U6 snRNP rather than by disrupting an inhibitory element or structure. Another line of evidence supporting the stem-loop hypothesis is that substitutions in nucleotides within the loop do not alter splicing (33). Also, lengthening the stem portion of the structure by introducing complementary nucleotides at I10 positions 17 and 18 (GT3 TG) reduces splicing. The stem-loop hypothesis can also be tested by generating double mutants that restore the secondary structure disrupted by mutations in the stem. Double mutants 92G3 A/E10ϩ16 and E10ϩ3/E10ϩ12 do not restore normal splicing as expected, even though these changes stabilize the stem-loop, as shown both empirically in melting exper-iments (32) and as predicted by free energy calculations (22). This result appears to argue against the stem-loop hypothesis although another interpretation is that E10ϩ3 and 92G3 A result in a strong splice site that over-rides the silencing effects of the stem-loop. As shown here, a strong 5Ј splice site is dominant over other cis-acting elements that control E10 splicing. E10 mutations that disrupt the longer possible stem-loop structures (Ϫ4⌬Ϫ9, Ϫ5⌬9, E⌬17, and the ␤-globin sequences used in Glo-1) do not significantly alter splicing. Thus the sequences immediately 5Ј to the end of the exon do not participate in regulation, and the longer stem-loop structures are probably not involved in E10 regulation.
Evidence against the short 18-nucleotide stem-loop being a regulatory structure comes from splicing experiments with FTDP-17 mutations E10ϩ12c3t, E10ϩ13a3g, E10ϩ14c3t, and E10ϩ16c3t in combination with the E⌬5 mutation. Singly, these mutations increase splicing so that E10 is nearly constitutively included, which correlates with the expected destabilization of the stem-loop. However, when E⌬5 is present, splicing is dramatically variable, ranging from 65% for E⌬5/ E10ϩ14c3t to 26% for E⌬5/E10ϩ16c3t, even though these I10 mutants would produce stem-loops with similar free energies (Table I). Thus, for these double mutants, splicing does not correlate with stem-loop stability for any of the structures described in Fig. 6 and Table I. Other less-direct evidence against the stem-loop hypothesis is that when mouse sequences are used either alone or in hybrid constructs with human sequences, splicing and predicted stem-loop free energies also do not correlate with E10 splicing levels ( Fig. 6 and Table I). The conclusion from our work is consistent with an alternative possibility for E10 splicing regulation, where the linear sequence starting at I10 position 12 and extending at least to nucleotide 18 and perhaps farther is an ISS.
E10 splicing is regulated differently in mice and humans. Whereas no E10ϩ is produced in fetal brain of either species, in adult mouse brain only E10ϩ mRNA is generated, whereas in adult human brain approximately equal amounts of E10ϩ and E10Ϫ transcripts are made. The factors controlling the switching from fetal to adult pattern of regulation is unknown, as is the reason for the difference in regulation between mouse and human in adult brain. Hutton and co-workers (33) suggested that the difference in regulation between these two species is due to the fact that a stem-loop forms at the E10/I10 junction in human but not mouse pre-mRNA. The lack of a stem-loop in mouse would result in constitutive E10 inclusion in mouse brain. As shown in Fig. 6, the mouse sequence can form an E10/I10 junction stem-loop that is different from the human structure and is less stable (⌬G ϭ Ϫ4.3 and Ϫ6.5 kcal, respectively, for the shortest structures). The lower stability of the mouse stem-loop would be consistent with more E10 inclusion in the adult mouse. However, when hN and mWT are compared in COS-7 cells, the mouse construct actually results in slightly lower levels of E10 inclusion (Fig. 7). Also, mouse I10 ISS sequences are more inhibitory than the human ISS sequence. This observation is not consistent with the E10/I10 junction being the mechanistic difference between mouse and human splicing regulation. However, the possibility remains that this ISS may act differently in adult neurons, compared with COS-7 cells. A more likely explanation is that the constructs used here lack a regulatory element(s) that controls the switch between fetal and adult patterns and the element(s) that controls E10ϩ/ E10Ϫ ratios in the adult brain. This interpretation is supported by the fact that when the human and mouse E10 constructs used here are assayed in rat fetal neuron cultures, approximately equal amounts of E10ϩ and E10Ϫ transcripts are made. In contrast, the endogenous rat tau gene almost exclusively produces E10Ϫ transcripts in the same cultured cells, 2 suggesting that additional regulatory elements have been deleted in the abbreviated hN and mWT constructs. The work described here shows that there are at least five different regulatory elements within tau E10 and an inhibitory element in the intronic sequences immediately downstream of this exon. Other work 2 indicates that other intronic sequences also control E10 splicing. FTDP-17 mutations alter function of three of these elements (the PPE, ACE, and the I10 ISS), demonstrating that these regulatory sequences function in vivo by a mechanism similar to that observed in the model system used here. To our knowledge N279K and L284L are the first examples of disease mutations that actually enhance splicing by strengthening an existing ESE. Complete definition of each enhancer and silencer element in their normal and mutant contexts may require testing in heterologous splicing constructs. Additional work is needed to determine what splicing factors bind specifically to these regulatory elements and whether these factors are active in neurons and glial cells in the adult brain.