5′ Splice Site Mutations in tau Associated with the Inherited Dementia FTDP-17 Affect a Stem-Loop Structure That Regulates Alternative Splicing of Exon 10*

Missense and splice site mutations in the microtubule-associated protein tau gene were recently found associated with fronto-temporal dementia and parkinsonism linked to chromosome 17 (Poorkaj et al. (1998) Ann. Neurol. 43, 815–825; Hutton et al. (1998)Nature 393, 702–705; Spillantini et al. (1998)Proc. Natl. Acad. Sci. U. S. A. 95, 7737–7741). The mutations in the 5′ splice site of exon 10 were shown to increase the ratio of tau mRNAs containing exon 10 and thus the proportion of Tau protein isoforms with 4 microtubule binding repeat domains, although how this increase leads to neurodegeneration is presently unclear. The mechanism by which these mutations increasetau exon 10 splicing was not determined, although the mutations were predicted to disrupt a potential stem-loop structure that was likely involved in the regulation of exon 10 alternative splicing. Here we describe in vitro splicing assays and RNA structural analysis that demonstrate that the mutations do indeed act through disruption of the stem-loop structure and that the stability of this secondary structure feature at least partially determines the ratio of tau exon 10+/− transcripts. In addition, we provide evidence that the stability of the stem-loop structure underlies the alternative splicing of this exon in other species.

The microtubule-associated protein Tau plays an important role in the polymerization and stabilization of neuronal microtubules (for review, see Ref. 4). Tau is thus crucial to both maintenance of the neuronal cytoskeleton and axonal transport (4). Abnormal intraneuronal inclusions, termed neurofibrillary tangles, composed of Tau are a feature of the pathology in several neurodegenerative conditions (5) including Alzheimer's disease, Pick's disease, fronto-temporal dementia (6), progressive supra-nuclear palsy, and the Lytico and Bodig diseases of Guam.
The tau gene is localized to chromosome 17q21 (7) and consists of 15 exons (8) of which 11 encode the six major Tau protein isoforms in human brain. The six different Tau isoforms are generated by alternative splicing of exons 2, 3, and 10 (9). Exons 9 -12 encode four microtubule-binding domains that are imperfect repeats of 31 or 32 residues (10). Alternative splicing of exon 10 gives rise to Tau isoforms with 3 (exon 10Ϫ) or 4 (exon 10ϩ) microtubule binding domains (11). The recent identification of four missense (1,2) and four splice site mutations (2,3) in the tau gene, associated with fronto-temporal dementia and parkinsonism linked to chromosome 17 (FTDP-17) 1 (6), has demonstrated that Tau dysfunction can lead to neuronal cell death and is not simply a secondary consequence of neurodegenerative disease.
FTDP-17 (6) is inherited as an autosomal dominant condition characterized clinically by behavioral, cognitive, and motor disturbance (many cases of this disease continue to be described clinically as "Pick's disease"). The age of onset is highly variable but is usually 45-65 years. At autopsy, patients with FTDP-17 display pronounced fronto-temporal atrophy with neuronal cell loss, gray and white matter gliosis, and superficial cortical spongiform changes. In addition, virtually all FTDP-17 cases have abnormal intraneuronal Tau inclusions, with glial Tau inclusions present in some families (6,12). The morphology and isoform composition of the Tau filaments that compose the inclusions also varies in the different families (12). The identification of different mutations in the tau gene has largely explained the variability in Tau pathology observed in FTDP-17 (2,3,13). Families with missense (2, 14) (P301L, N279K) or splice site mutations (2, 3) that affect exon 10, and thus 4 repeat Tau isoforms, have intraneuronal and glial Tau inclusions consisting predominantly of four repeat Tau isoforms. The Tau filaments in these cases have a longer periodicity than the paired helical filaments that comprise the neurofibrillary tangles observed in AD (12,13). In contrast, families with missense mutations (G272V, V337M, and R406W) outside of exon 10 (1, 2), that affect all Tau isoforms, have neuronal inclusions (glial inclusions are absent) that are composed of all six Tau isoforms and are made up of filaments identical to the paired helical filaments observed in AD (12,13,15).
All but one of the reported (1,2,14) missense mutations (G272V, N279K, P301L, and V337M) associated with FTDP-17 occur within the microtubule binding domains of Tau and four missense mutations (G272V, P301L, V337M, and R406W) have been demonstrated to disrupt the interaction between Tau and the microtubules in vitro (16,17). In contrast, the splice site mutations in FTDP-17 affect alternative splicing of exon 10 such that an increased proportion of Tau exon 10ϩ transcripts are generated; this leads to an increase in Tau isoforms with four, as opposed to three, binding repeats (2,3). The mechanism by which this increase leads to neurodegeneration and FTDP-17 is presently unclear; however, it demonstrates that the ratio of four repeat to three repeat isoforms is crucial to the correct functioning of Tau (2,3). This is consistent with the observation that alternative splicing of exon 10 is developmentally regulated with three repeat Tau alone, in the absence of four repeat Tau, present in fetal brain tissue in multiple species (11,18). The predominance of three repeat Tau during neuronal development has clear significance for the likely function of these isoforms in adult brain perhaps implying that human neurons require three repeat Tau to maintain plasticity (4). In addition, the ratio of four to three repeat Tau isoforms varies markedly in different species with a slight predominance of three repeat Tau present in adult human brain (11), while in adult mouse neurons only four repeat Tau is observed (18). The ratio of Tau isoforms also varies in different neuronal populations with the granule cells of the human dentate gyrus being reported to contain only three repeat Tau isoforms (11).
We previously employed RT-PCR analysis of FTDP-17 brains to demonstrate that the 5Ј splice site mutations are associated in vivo with a 2-6-fold increase in the ratio of exon 10ϩ to exon 10Ϫ tau mRNA (2). In addition, we utilized an exon trapping protocol (2,19) as a splicing assay to demonstrate in vitro that the 5Ј splice site mutations were also capable of increasing the incorporation of tau exon 10 into artificial transcripts. However while both RT-PCR analysis of FTDP-17 brains and splicing assays demonstrated that the 5Ј splice site mutations act by increasing the incorporation of exon 10 into tau mRNAs, neither method indicated the mechanism by which the mutations affected splicing (2). Examination of the intronic sequence downstream of exon 10 revealed that each of the mutations was predicted to disrupt a potential stem-loop structure that was likely involved in the regulation of exon 10 alternative splicing by competing with the U1 snRNP for binding to the 5Ј splice site (2). Stem loop structures have previously been implicated in regulating the selection of alternative 5Ј splice sites (20) and distant branch points (21) and also in the tissue-specific splicing of the chicken ␤-tropomyosin exons 6A and 6B (22,23).
In this study we again utilize in vitro exon trapping (splicing) assays (2,19) to test the hypothesis that the stability of the potential stem-loop structure in the 5Ј splice site of tau exon 10 at least partially determines the ratio of tau exon 10ϩ/Ϫ transcripts and that the splice site mutations act by disrupting this structure. We also examined the sequence of the exon 10 5Ј splice site in the tau gene from bovine, rabbit, rat, mouse, and two other primates to determine if this pre-mRNA structure might regulate alternative splicing in other mammals. Finally, we utilize RNA secondary structure analysis to demonstrate that a stem-loop is present in the normal 5Ј splice site of tau exon 10 and that this structure is disrupted by the FTDP-17 splice site mutations.

EXPERIMENTAL PROCEDURES
Generation of Exon Trapping Constructs-Mutant and wild-type versions of tau exon 10 were amplified from the DNA of the patients with the FTDP-17 ϩ13, ϩ14, and ϩ16 splice site mutations (residues numbered from the exon 10 5Ј splice site) and from an unaffected individual. PCR products contained exon 10 and flanking intron sequence at either end ( Fig. 1). PCR products were cloned into the splicing vector pSPL3b using XhoI and PstI sites incorporated into the amplification products. Mutant and wild-type constructs were identified by sequence analysis. Site-directed mutagenesis was performed on these constructs using the Transformer site-directed mutagenesis kit (CLONTECH). Mutagenic primers for introducing the ϩ17(G/T)/ϩ18(T/G) "extended stem" mutations were generated for the wild-type construct and for each of the three FTDP-17 splice site mutant (ϩ13, ϩ14, and ϩ16) constructs. Also generated were mutagenic primers for the mutations for the complementary rescue analysis (see Figs. 2-5 for list of mutations). A selection primer was used in each mutagenesis reaction to remove a single HpaI site from the pSPL3b vector. For each reaction vector DNA was denatured at 100°C, mismatch and selection primers were annealed and T4 DNA polymerase, and T4 DNA ligase were used to synthesize the second strand. Each mutant was transformed into BMH 71-18 mutS Escherichia coli and the plasmid DNA isolated using the Wizard miniprep kit (Promega). The DNA was treated with HpaI as a primary selection process and then re-transformed into DH10B. Colonies were screened by PCR and restriction enzyme digestion, and the identity of each mutant was confirmed by sequencing.
Wild-type and FTDP-17 mutant (ϩ13, ϩ14, and ϩ16) exon trapping constructs containing additional flanking intronic sequence (1 kb) on either side of tau exon 10 were also generated in pSPL3b to ensure that the short intronic sequences used in the original constructs ( Fig. 1) were not artifactually affecting splicing assay results. These constructs were generated by performing PCR on control genomic DNA with primers (5Ј-CACCCTCGAGGGAAGACGTTCTCACTGATCTG-3Ј and 5Ј-GTG-GCGGATCCATGGCTCCTTGCAACTTC-3Ј) designed to more distant FIG. 1. Exon trapping assay constructs. A, design of the basic exon trapping construct is shown with tau exon 10 and flanking intronic sequences cloned into the XhoI and PstI sites in the pSPL3b splicing vector. The cloned tau sequence is flanked by artificial introns (311 and 2039 bp) from the human immunodeficiency virus tat gene. RT-PCR primers SD2 and SA4 were designed to the human immunodeficiency virus tat exons that are incorporated into the artificial mRNAs. Mutagenesis was performed on the basic constructs as required for each experiment. Extended exon 10 constructs (not shown) containing ϳ1 kilobase pair of flanking 5Ј and 3Ј intronic sequence were also generated in pSPL3b to ensure that similar exon trapping results were obtained when additional intronic sequence was included. B, schematic splicing diagrams displaying the RT-PCR product sizes produced by mRNAs containing tau exon 10 (246 bp) and mRNAs with tau exon 10 skipped (153 bp).
5Ј and 3Ј intronic sequences (GenBank TM ). Amplification was performed with the high fidelity PCR system (Boehringer Mannheim). The 2082-bp PCR products were cloned into pSPL3b using XhoI and BamHI sites incorporated into the amplification products and sequenced. Mutagenesis was then performed, as described above, on the wild-type extended splicing construct to introduce the FTDP-17 exon 10 5Ј splice site mutations (ϩ13, ϩ14, and ϩ16). These constructs were analyzed by exon trapping in identical fashion to the original wild type and mutant pSPL3b constructs (see below).
Exon Trapping-For splicing assays, the "exon trapping system" from Life Technologies, Inc. was used. Briefly, COS-7 cells were transfected in triplicate with 1 g of each construct using LipofectACE (Life Technologies, Inc.). Cells were collected 24 h post-transfection and RNA prepared using Trizol reagent (Life Technologies, Inc.). First-strand synthesis was performed using reagents supplied with the system, using the conditions described in the manufacturer's instructions. PCR was performed using the primers SD2 and SA4 (the sequence for these primers was taken from the exon trapping system) and Taq Gold (Perkin-Elmer). To enable quantification of PCR products on an automated sequencer, the SD2 primer was labeled with a TET (Applied Biosystems) fluorescent dye tag. An initial denaturation stage of 5 min at 94°C was followed by six stage 1 cycles (94°C for 1 min, 60°C for 1 min, 72°C for 5 min), 24 stage 2 cycles (94°C for 1 min, 60°C for 1 min, 72°C for 4 min), and a final extension phase of 72°C for 10 min. Performing PCR in a single stage for 30 cycles with shorter extension times (45 s at 72°C) did not significantly alter the results. To verify that the RT-PCR was quantitative, the total number of amplification cycles was varied (24 -39 cycles, data not shown). PCR products were visualized on 2% agarose gels; exon 10ϩ products were 246 bp, while exon 10Ϫ products were 153 bp. Quantification of products and estimation of the molar ratio of exon 10ϩ and exon 10Ϫ mRNA was performed using an Applied Biosystems 377 automated sequencer running Genescan software. Triplicate independent transfections were used to determine the mean exon 10ϩ/Ϫ ratio and S.D. for each experiment. The identity of RT-PCR products (exon 10ϩ/Ϫ) was confirmed by sequencing.
Sequencing of the 5Ј Splice Site in tau Exon 10 from Different Species-Analysis was performed on genomic DNA from rhesus, marmoset, bovine, rabbit, rat, and mouse. PCR was performed between tau exon 10 to exon 11 using the Expand Long Template PCR system (Roche Molecular Biochemicals) with primers (10F, 5Ј-GATCTTAGCAACGTC-CAGTCCAAGTG-3Ј and 11R, 5Ј-GTTGCCTAATGAGCCACACTTG-GAGGTC-3Ј) generated from conserved exonic sequence. For the PCR, an initial denaturation stage of 92.5°C for 2 min was followed by 10 stage 1 cycles (92.5°C for 10 s, 54°C for 30 s, 68°C for 12 min), 20 stage 2 cycles (92.5°C for 10 s, 54°C for 30 s, 68°C for 12 min increasing by 20 s each cycle), and a final extension phase of 10 min at 72°C. Sequencing of the PCR product was performed after agarose gel purification using the Big Dye terminator cycle sequencing kit (Perkin-Elmer). Sequencing reactions were analyzed on an Applied Biosystems 377 automated sequencer running Factura software (Perkin-Elmer). Sequences of the tau exon 10 5Ј splice site were aligned by the Sequence Navigator Package (Perkin-Elmer).
RT-PCR Analysis of Exon 10 Alternative Splicing in Different Species-RNA was isolated from human frontal lobe and from rabbit, rat, and mouse whole brain using the Trizol reagent and protocol (Life Technologies, Inc.). Reverse transcription was performed using the Superscript pre-amplification system (Life Technologies, Inc.) on 1-4 g of brain RNA with an oligo(dT) primer. PCR was performed using primers designed to exonic sequence that was conserved in each species from exon 9 (forward, 5Ј-CTGAAGCACCAGCCAGGAGG-3Ј) and exon 13 (reverse, 5Ј-TGGTCTGTCTTGGCTTTGGC-3Ј). Amplification involved 30 cycles of 94°C for 30 s, 65-50°C for 30 s, and 72°C for 45 s with a final 72°C extension phase for 10 min. We demonstrated previously that these conditions were quantitative by performing this amplification using a range of PCR cycles (18 -37) (2). PCR products were visualized on 2% agarose gels with exon 10ϩ RT-PCR products being 367 bp, while exon 10Ϫ products were 274 bp.
Human Genomic tau Transgenic Mouse Generation and Analysis-Transgenic mice expressing the entire normal human tau gene under the control of the human tau promoter were generated through microinjection of a tau containing PAC (P1-derived artificial chromosome) (24i13, ϳ200 kb) into mouse embryos. RNA was isolated from hemibrains from transgenic and littermate control animals, from two different lines, using the Trizol reagent and protocol (Life Technologies, Inc.). For analysis of exon 10 alternative splicing in mouse endogenous and human transgene-derived tau mRNA, RT-PCR was performed between exon 9 and exon 11 with mouse-specific (9F, 5Ј-CACCAAAATCCG-GAGAACGA-3Ј and 11R, 5Ј-CTTTGCTCAGGTCCACCGGC-3Ј) and hu-man specific primers (9F, 5Ј-CTCCAAAATCAGGGGATCGC-3Ј and 11R, 5Ј-CCTTGCTCAGGTCAACTGGT-3Ј). PCR was performed (30 cycles of 94°C for 30 s, 62°C for 30 s and 72°C for 45 s with a final 72°C extension phase for 10 min). Mouse-specific and human-specific PCRs were analyzed on agarose gels; products corresponding to exon 10ϩ tau mRNA gave a band at 390 bp, while products corresponding to exon 10Ϫ mRNA gave a band at 297 bp). A complete description of the generation and neuropathological analysis of these human tau transgenic mice will be published separately. 2 RNA Structural Mapping and Gel Migration Analysis-Wild-type and mutant (ϩ8, ϩ13, ϩ14, ϩ16, and 10-bp stem) constructs containing tau exon 10 and flanking intronic sequence ( Fig. 1) were cloned into pBluescript(KSϪ) using XhoI and PstI. The plasmid DNA was linearized with BamHI to generate a template for in vitro transcription with T3 RNA polymerase. In vitro transcription was performed using the Ribomax system (Promega) according to manufacturer's instructions. RNase digests were performed using 4 g of RNA and 0.08 unit of V1 double strand-specific RNase (Amersham Pharmacia Biotech) or 0.5 unit of T1 single strand-specific RNase (Ambion) for 15 min at 25°C in a 25-l reaction, using the manufacturers' recommended buffers. Control samples for each RNA were treated identically, to the RNase digests, but enzyme was omitted. Samples were then immediately purified by phenol/chloroform extraction and resuspended in 15 l of diethyl pyrocarbonate-treated water. RNA recovery was assessed UV absorption at 260 nM. Primer extension was performed on 1 pmol of digested RNA using 50,000 cpm of 32 P-labeled oligonucleotide (5Ј-CCGGGCTGCAGACACCACTTCC-3Ј) and Superscript II reverse transcriptase (Life Technologies, Inc.) in a 20-l reaction according to manufacturer's instructions. The RNA template was digested with RNase H (Life Technologies, Inc.), and the resultant primer extension products were isopropyl alcohol precipitated, washed in 70% ethanol, and resuspended in 5:1 formamide/blue dye loading mix. Sequencing reactions of the wild-type construct were performed using 33 P-labeled terminator sequencing system (Amersham Pharmacia Biotech) and the same, unlabeled, oligonucleotide as was used for the primer extension reactions. Primer extension samples were run on a 6% polyacrylamide sequencing gel alongside the sequencing reactions and the results visualized by autoradiography. Approximate equal gel loading was verified by comparison of bands generated from full-length cDNA products from control reactions.
To further demonstrate the effect of the FTDP-17 splice site mutants on RNA secondary structure, the variable migration of in vitro transcribed wild-type and mutant (ϩ13, ϩ14, ϩ16, and 10-bp extended stem) RNA transcripts was analyzed on both nondenaturing 4% Metaphor agarose (FMC) gels and on denaturing 4% agarose/formaldehyde gels. In vitro transcription was performed as described above. 1 g of each RNA was loaded onto each gel. Nondenaturing 4% Metaphor agarose gels were run at 100 V in TAE buffer (Tris/acetic acid/EDTA). Denaturing 4% agarose/formaldehyde gels were run at 100 V in 1 ϫ MOPS buffer, pH 7.0 (20 mM MOPS, 5 mM sodium acetate, 1 mM EDTA), samples were preheated to 55°C for 10 min in 50% formamide, 6.5% formaldehyde 1 ϫ MOPS loading buffer prior to loading.

Mutations That Disrupt the Predicted Stem-Loop
Increase the Incorporation tau Exon 10 -To demonstrate that mutations that disrupt the predicted stem-loop in the 5Ј splice site increase splicing of tau exon 10 in vitro a series of mutations were introduced into exon trapping constructs. Mutations were made at positions Ϫ2, Ϫ1, ϩ3, ϩ11, ϩ13, ϩ14, and ϩ16 ( Fig. 2) relative to the 5Ј splice site that are predicted to lie within the stem, while mutations at ϩ8 and ϩ9 affected loop residues that were not expected to affect the stability of the stem-loop. Mutations at ϩ3 (G to A), ϩ13 (A to G), ϩ14 (C to T), and ϩ16 (C to T) were those that had been shown previously to increase splicing of exon 10 in vivo and to be pathogenic in FTDP-17 (2,3). Exon trapping transfections were performed in triplicate on each of the mutant constructs as well as on the wild-type construct.
The results of these experiments (Fig. 2) demonstrated that each of the FTDP-17-associated splice mutants (ϩ3, ϩ13, ϩ14, and ϩ16) significantly increased the incorporation of tau exon 10 into the artificial transcripts generated by the exon trapping assay, compared with the wild-type constructs consistent with previous reports of the effects of these mutants in this in vitro system (2). In addition artificial mutants at Ϫ2 (G to A), Ϫ1 (T to C), and ϩ11 (T to C), which are predicted to disrupt the stem-loop, also increased splicing of exon 10 relative to the wild-type construct. Mutations at Ϫ2 and ϩ3 are also predicted to increase the stability of U1 snRNP binding to the 5Ј splice site (Fig. 5), which has also been shown to increase incorporation of an alternatively spliced exon (24). The result of this double effect on the splicing mechanism is that these mutations (ϩ2 and ϩ3) cause the greatest increase in exon 10 incorporation of the mutations tested (Fig. 2, ratios Ͼ10). The ϩ3 (G to A) FTDP-17 splice site mutation (3) is thus predicted to increase U1 snRNP binding as well as disrupt the stem-loop structure, suggesting that both factors are likely to play a role in the mechanism of this pathogenic mutation.
In contrast, mutations at ϩ8 and ϩ9 had little impact on exon 10 splicing in this assay relative to the wild-type construct (Fig. 2). This was expected since these mutants occur within the loop region and therefore are not predicted to affect the stability of the stem-loop structure. The small increase in splicing observed with the ϩ8 mutation probably reflects the small increase in U1 snRNP binding stability to the 5Ј splice site that is predicted to be generated by this change. Taken together the results from this series of experiments are entirely consistent with the hypothesis that the predicted stem-loop structure at the 5Ј splice site plays a major role in regulating the alternative splicing of tau exon 10 in vitro. However the precise ratio of exon 10ϩ to 10Ϫ RNA is likely to be affected by other factors such as the stability of U1 snRNP binding to the 5Ј splice site. Indeed it is significant that mutations that were predicted to reduce U1 snRNP base pairing at positions ϩ3 (G to C), ϩ4 (A to G), and ϩ5 (G to C) resulted in the total loss of exon 10 splicing (Fig. 5). These results, while uninterpretable in terms of demonstrating the role of the stem-loop, do indicate the minimal U1 snRNP binding stability required for this splice site to remain functional in vitro.
Increasing the Stability of the Predicted Stem-Loop Reduces the Incorporation tau Exon 10 -To further test the hypothesis that the stability of the predicted stem-loop at least partially determines the ratio of exon 10ϩ to exon 10Ϫ tau mRNAs, we generated wild-type and mutant (ϩ13, ϩ14, and ϩ16) constructs in which residues ϩ17/ϩ18 downstream of the exon 10 5Ј splice site were converted from GT to TG by site-directed mutagenesis (Fig. 3). This had the effect of increasing the length of the "stem" from 6 to 10 bp (although the exact structure of the stem-loop has yet to be determined), since residues ϩ19 and ϩ20 also pair with Ϫ5 and Ϫ6 (Fig. 3), significantly increasing the stability of the potential stem-loop structure in the 5Ј splice site sequence. Exon trapping was performed in triplicate on each of the 4 constructs (wild-type and ϩ13, ϩ14, and ϩ16 splice site mutants) with the extended (10 bp) stemloop sequence as well as on wild-type and mutant constructs with the normal (6 bp) stem-loop (Fig. 3).
As before, each of the FTDP-17-associated splice mutants significantly increased the incorporation of tau exon 10 into the artificial transcripts generated by the exon trapping assay, compared with the wild-type constructs. This increase was present regardless of whether comparisons were between wildtype and mutant constructs with the normal 6 bp or with the extended 10 bp stem-loop structure (Fig. 3). The mutation with the greatest effect on splicing in both the 6-and 10-bp constructs was the ϩ14 mutation (observed in the disinhibition dementia parkinsonism amyotrophy complex family (25)), which also causes the greatest loss in stem-loop stability (2) of the three mutations tested.
Increasing the length of the stem from 6 to 10 bp, and thereby the stability of the stem-loop structures, consistently resulted in a dramatic reduction in the ratio of exon 10ϩ to exon 10Ϫ transcripts, resulting from increased skipping of exon 10. This effect was observed in both the wild-type construct and in each of the three FTDP-17 mutant constructs (Fig. 3). Thus increasing the stability of the stem-loop in this manner reduced the incorporation of exon 10 consistent with the stem-loop structure regulating alternative splicing of this exon. The observed reduction in the incorporation of exon 10 with the extended stem (10 bp) constructs effectively rescues the effect of the FTDP-17 mutations by restabilizing the disrupted stemloop. However as mentioned earlier the FTDP-17 mutants even with the 10 bp stem continued to have a higher exon 10ϩ/Ϫ mRNA ratio than observed in the equivalent 10-bp stem wildtype construct (Fig. 3). Again this result is consistent with the hypothesis that the overall stability of the stem-loop deter-  (2, 3)). B, the results of exon trapping analysis shows that mutations which are predicted to disrupt the stem-loop structure (Ϫ2, Ϫ1, ϩ3, ϩ11, ϩ13, ϩ14, ϩ16) cause increased exon 10ϩ (246-bp product) to exon 10Ϫ (153 bp) RNA ratios. In contrast, mutations in the loop region (ϩ8 and ϩ9) are not predicted to affect the stability of the stem-loop and do not alter exon 10 splicing. Molar ratios and S.D. values (shown beneath gel lanes) were calculated from three independent transfections. Qualitatively identical results were obtained with wild-type and mutant (ϩ13, ϩ14, ϩ16) tau exon 10 constructs containing ϳ1 kb of 5Ј and 3Ј flanking intronic sequence (not shown). mines the proportion of transcripts into which exon 10 is incorporated.

Rescue Analysis of Stem-Loop Mutations by Alteration of Complementary Bases in the 5Ј Splice Site of Exon 10 -Analy-
sis of the stem-loop structure was also performed by mutating a residue on one side of the stem and then attempting to rescue the observed increase in splicing by altering the corresponding residue on the opposite side of the stem such that base pairing is restored (Fig. 4). However this analysis is complicated by the short length of the stem-loop (6 bp) as all of the residues on one side of the stem occur within the U1 snRNP binding site. As a result several of the residues could not be mutated without causing the complete loss of exon 10 splicing (ϩ1, ϩ2, ϩ3 (G to C), and ϩ4, Fig. 5). In addition, other mutations are predicted to cause an increase in U1 snRNP binding, which in turn leads to an increase in splicing independent of stem-loop stability (Ϫ2 (G to A), ϩ3 (A to G), Fig. 5).
However, rescue analysis was performed using constructs with artificial mutations at positions Ϫ1 (U to C) and ϩ16 (C to G) with rescue constructs where the opposing residues were altered, ϩ15 (A to G), Ϫ2 (G to C), such that the stem-loop in each case is restabilized (Fig. 4). The Ϫ1 and ϩ16 mutations are predicted to reduce the stability of the stem-loop while the Ϫ1/ϩ15 and ϩ16/Ϫ2 rescue constructs generate G-C base pairs that restore stem-loop stability. The Ϫ1/ϩ15 stem-loop is more stable than in the wild-type construct, since an A-T pair is replaced with a G-C pair; this results in a group of three consecutive G-C pairs.
As predicted the artificial mutations at Ϫ1 and ϩ16 resulted in an increase in the proportion of exon 10ϩ mRNA (ratios 3.16 and 1.55, respectively), consistent with destabilization of the stem-loop (Fig. 4). The Ϫ1/ϩ15 rescue construct reduced exon 10 splicing to levels significantly lower (ratio 0.44) than those observed with wild-type constructs (ratio 0.82). This reflects restabilization of the stem-loop by the opposing ϩ15 mutation, which gives a stem-loop that is more stable than with the wild-type sequence (Fig. 4). The exceptionally low exon 10ϩ/Ϫ ratio observed with the Ϫ1/ϩ15 rescue (it was the lowest ratio observed where splicing was not abolished) probably reflected the generation of three consecutive G-C base pairs, over the splice site, which would be expected to significantly increase stem-loop stability. The ϩ16/Ϫ2 rescue construct also reduced splicing significantly (ratio 1.32) relative to the ϩ16 mutant construct (ratio 1.55) consistent with restabilization of the stem-loop. However the rescue construct still gave a higher exon 10ϩ/Ϫ ratio than wild-type (0.82). The reason for this difference is unclear although it may reflect a subtle effect on either stem-loop stability, U1 snRNP binding or another part of the splicing process produced by replacing a purine residue with a pyrimidine (G to C) at position Ϫ2.
Exon trapping analysis with the Ϫ1, ϩ16 mutants, and Ϫ1/ ϩ15, ϩ16/Ϫ2 rescues (Fig. 4) yielded results that are consistent and opposing bases (ϩ15, Ϫ2) that are altered to generate rescue constructs. Note the generation of three consecutive G-C pairs in the Ϫ1/ϩ15 rescue construct stem-loop. B, exon trapping analysis (exon 10ϩ ϭ 246 bp, exon 10Ϫ ϭ 153 bp). Artificial Ϫ1 and ϩ16 mutations increase exon 10 splicing compared with wild-type (WT), and this effect is rescued when the stem-loop is restabilized by the opposing ϩ15 and Ϫ2 mutations, respectively (Ϫ1/ϩ15, ϩ16/Ϫ2). The Ϫ1/ϩ15 rescue construct gives the lowest observed exon 10ϩ/Ϫ ratio (0.44), which presumably reflects the generation of a more stable stem-loop (compared with wild-type in the Ϫ1/ϩ15 construct that contains three consecutive G-C pairs over the 5Ј splice site. Molar ratios and S.D. values (shown beneath gel lanes) were calculated from three independent transfections.
FIG. 3. Tau exon 10 alternative splicing in vitro is regulated by the stability of the stem-loop. A, normal (6 bp) and extended (10 bp) stem-loops are shown with the location of FTDP-17 splice site mutations (ϩ13, ϩ14, and ϩ16). The 10-bp stem-loop was generated by mutagenesis of residues ϩ17/ϩ18 (shaded box). B, the results of exon trapping analysis with the 6-and 10-bp stem-loop constructs demonstrated that increasing the stability of the stem-loop resulted in a reduction in the ratio of exon 10ϩ (246-bp product) to exon 10Ϫ (153-bp product) mRNAs in wild-type (WT) and mutants (ϩ13, ϩ14, and ϩ16). Note that the mutants (ϩ13, ϩ14, and ϩ16) consistently increased the proportion of exon 10ϩ mRNA compared with wild-type (WT) regardless of whether comparisons were made among 6-or 10-bp stem-loop constructs. The ϩ14 mutant gives the largest increase in exon 10ϩ/Ϫ ratio consistent with this mutation producing the greatest reduction in stem-loop stability (2). Molar ratios and S.D. values (shown beneath gel lanes) were calculated from three independent transfections.
with data from the extended (10 bp) stem-loop constructs (Fig.  3). Together these studies demonstrate that the stability of the predicted stem-loop in the 5Ј splice site has a major influence on the alternative splicing of tau exon 10 in vitro.

Alteration of Complementary Bases Does Not Rescue the Affects of FTDP-17 Splice Mutants at Positions ϩ16 and ϩ3-
Rescue analysis was also attempted with constructs designed to restabilize FTDP-17 splice site mutants at ϩ16 (C to T) (2) and ϩ3 (G to A) (3). Similar analysis could not be performed for FTDP-17 splice mutations at ϩ13 (2) and ϩ14 (2), since altering the residue on the opposite side of the stem-loop, to each mutant, would result in the loss of the ϩ1/ϩ2 GT sequence that is minimally required in a 5Ј splice site. This in turn would have resulted in an uninterpretable loss of exon 10 splicing.
Rescue constructs for the ϩ16 (C to T) and ϩ3 (G to A) FTDP-17 mutants were generated by altering the equivalent residue (Ϫ2 and ϩ12, respectively) on the opposite side of the stem-loop (Ϫ2, G to A; ϩ12, C to U) to restore the base pairing at this position (Fig. 5). However in both rescue constructs the predicted stem-loop structure is significantly less stable than in the wild-type splice site, since a G-C pair is replaced with A-U with the result that in both rescues consecutive A-U base pairs are created in the stem-loop. The anticipated result was that the two rescue constructs would show reduced incorporation of exon 10 relative to the ϩ16 and ϩ3 mutants alone. However, neither rescue construct significantly reduced the splicing of exon 10, compared with the ϩ3 and ϩ16 mutant constructs, and indeed the Ϫ2/ϩ16 construct displayed a small (1.8-fold), but significant, increase in the proportion of exon 10ϩ mRNA (Fig. 5). The explanation for these results is likely to be a combination of two factors: first both of the complementary "rescue" constructs have stem-loops that are significantly less stable than in the wild-type construct, with a G-C pair replaced with A-T; second the ϩ3 FTDP-17 mutant and the Ϫ2/ϩ16 rescue (to the FTDP-17 ϩ16 mutant) are both predicted to increase the base pairing of the 5Ј splice site with the U1 snRNP (Fig. 5). Increased binding of the U1 snRNP to the 5Ј splice site would be expected to increase splicing of exon 10 (24), independent of stem-loop stability. In both cases, the overall result is that the restoration of base pairing in the rescue constructs (ϩ3/ϩ12 and Ϫ2/ϩ16) is insufficient to significantly reduce exon 10 splicing compared with the FTDP-17 mutants (ϩ3 and ϩ16).

Analysis of the tau Exon 10 5Ј Splice Site in Other Species Confirms the Role of Pre-mRNA Secondary Structure in the Regulation of Alternative Splicing of This Exon-
The results from the exon trapping analysis demonstrate that the effect of the FTDP-17-associated 5Ј splice site mutations on the alternative splicing of tau exon 10 is mediated through destabilization of the predicted stem-loop structure. However exon trapping is an artificial system that does not fully replicate the alternative splicing of tau exon 10 in vivo. Therefore to determine if the predicted stem-loop structure was also likely to play a role in the regulation of tau exon 10 alternative splicing in vivo, we examined the sequence of the 5Ј splice site in a range of other mammalian species (human, rhesus, marmoset, bovine, rabbit, rat, and mouse).
The sequence of the 5Ј splice site in primates (human, rhesus, and marmoset) and bovine had an identical predicted stem-loop structure with a 6-bp stem and a 6-base loop region (Fig. 6). In contrast in each of the rodent species (rabbit, rat, and mouse) some part of the sequence that makes up this structure was absent (Fig. 6), resulting in a predicted stem-loop of reduced stability. In the rabbit the ϩ11 residue was not conserved (C in rabbit, U in human), resulting in a predicted 5-bp stem and 7-base loop, which mimics the artificial ϩ11 (T to C) mutation employed in exon trapping studies. In the rat residue ϩ13 is not conserved (G in rat and A in human), which mimics the ϩ13 (A to G) FTDP-17 mutation, and results in a 6-bp stem with an internal mismatch (G-U). In the mouse both the ϩ13 and the ϩ16 residues are not conserved. Thus the order of predicted stem-loop stability in the different mammalian species analyzed is primates/bovine Ͼ rabbit Ͼ rat Ͼ mouse (Fig. 6). In order to relate the stability of the stem-loop in the 5Ј splice site to the alternative splicing of exon 10, we performed RT-PCR analysis on RNA isolated from the brains of the different species (human frontal lobe, rabbit, rat, and mouse). The results of this analysis demonstrated clearly (Fig.  6) that there was an inverse relationship between the predicted stability of the stem-loop in the sequence of the splice site in each species and the ratio of tau exon 10ϩ/Ϫ mRNAs, thus the order of different species in exon 10ϩ/Ϫ ratio was: mouse Ͼ rat Ͼ rabbit Ͼ human frontal lobe (bovine and other primates not tested). The exon 10ϩ/Ϫ ratios observed in the human, rabbit, and rat brain RNA by RT-PCR are also highly similar to the ratios obtained through in vitro exon trapping studies performed with equivalent tau exon 10 constructs: wild-type, ϩ11 (T to C) and ϩ13 (A to G), respectively (Fig. 2). The results are consistent with the predicted stem-loop structure in the 5Ј  (G to A), ϩ16) and opposing bases (ϩ12, Ϫ2) that are altered to generate rescue constructs. B, predicted U1 snRNP binding to the tau exon 10 5Ј splice site and the effect of Ϫ2, ϩ3 (G to A), ϩ3 (G to C), ϩ4, ϩ5, ϩ8 mutations. Note that mutations at ϩ3 (G to C), ϩ4 (A to G), and ϩ5 (G to C), which are predicted to reduce U1 snRNP base pairing to the 5Ј splice site, resulted in the complete loss of exon 10 splicing (data not shown) C, exon trapping analysis (exon 10ϩ ϭ 246 bp, exon 10Ϫ ϭ 153 bp). The FTDP-17 mutations ϩ16 and ϩ3 (G to A) increase exon 10 splicing compared with wild-type (WT); however, splicing is not reduced by alteration of the corresponding residue Ϫ2 and ϩ12, respectively, that restores base pairing at each site. This likely reflects incomplete restabilization of the stem-loop in the rescue constructs (ϩ16/Ϫ2 and ϩ3/ ϩ12) as a G-C pair is replaced with A-T and increased U1 snRNP base pairing in the ϩ3 and ϩ16/Ϫ2 constructs. Molar ratios and standard deviations (shown below gel lanes) were calculated from three independent transfections. splice site of exon 10 regulating the alternative splicing of this exon in multiple species and that the stability of this structure at least partially determines the proportion of exon 10ϩ mRNA. It should be noted, however, that other factors in addition to the stem-loop must also be involved in the regulation of exon 10 alternative splicing, since only three repeat Tau (exon 10Ϫ) is observed in the mammalian fetal brain (18). One possible explanation of this phenomenon is that an inhibitory splicing factor is present in fetal brain that is absent in the majority of adult neurons; previous studies have also suggested that thyroid hormone expression regulates the generation of four repeat Tau during brain development (26).
In the mouse, tau exon 10Ϫ mRNA was not detected by RT-PCR analysis of whole brain RNA, in agreement with previous reports of the Tau isoform composition in adult mouse brain (18). The large difference in Tau isoform composition between adult human and mouse brains (and other rodents) is likely to reflect, if not underlie, some fundamental, undetermined differences in the functions and characteristics of neurons from these species.

Transgenic Mice Expressing the Human tau Gene Demonstrate Alternative Splicing of Exon 10 Consistent with the Involvement of Cis-acting Elements in Regulation of This Splice
Event-Alternative splicing of exon 10 was investigated in the brains of transgenic mice expressing a human PAC (P1-derived artificial chromosome) (ϳ200 kb) transgene containing the entire tau gene. RT-PCR analysis with human and mouse tauspecific primers was performed on RNA isolated from the brains of adult transgenic and littermate control animals. This analysis (Fig. 6C) demonstrated that splicing of tau exon 10 in the endogenous mouse gene was unaffected in the transgenic animals (only exon 10ϩ mouse tau RNA was observed). In contrast, with human-specific primers products corresponding to both exon 10ϩ and exon 10Ϫ mRNA were detected in the brains of transgenic mice expressing the human tau gene (Fig.  6C). This demonstrated that human-like alternative splicing of exon 10 was occurring in pre-RNA generated from the human transgene. Control animals gave no RT-PCR product with the human-specific primers. Western blot analysis of Tau protein in the transgenic mouse brains also demonstrated the presence of human Tau isoforms, generated by alternative splicing of pre-mRNA, that are absent in endogenous mouse Tau protein (not shown).
The fact that the human tau transgene generates pre-mRNA that undergoes alternative splicing of exon 10 similar to that seen in the human brain (although the exon 10ϩ/Ϫ ratio is somewhat altered in these mice) clearly suggests that cis-acting sequence elements specific to the human gene regulate alternative splicing of exon 10 (Fig. 6C). This is obviously consistent with the hypothesis that the stem-loop sequence in the 5Ј splice site of human exon 10, which is absent in the mouse gene, is at least partially involved in the regulation of alternative splicing of exon 10.

RNA Structural Mapping Confirms the Presence of a Stem-Loop Structure in the 5Ј Splice Site of tau Exon 10 That Is
Disrupted by the FTDP-17 Splice Site Mutations-RNA secondary structure prediction analysis with the program RNAFOLD (28) suggested two possible structures of similar stability for the region around the 5Ј splice site of exon 10 (Fig. 7). In both structures the minimal 6-bp stem-loop is maintained, however, beyond this region it is possible for either a short second stem to form (structure 1) or a lateral stem-loop (structure 2). The structures are of similar stability, because they are both formed with an identical 9-base sequence (ACACGUCCC) that is re-FIG. 6. Analysis of tau exon 10 alternative splicing in other mammals and transgenic mice. A, aligned 5Ј splice site sequences. Boxes denote the location of the stem-loop with nonconserved residues, in the rodents, excluded. B, RT-PCR analysis of tau exon 10 alternative splicing in human frontal lobe and rabbit, rat, and mouse whole brain RNA. PCR was between exon 9 and exon 11 (exon 10ϩ products are 367 bp and exon 10Ϫ products are 274 bp), and the identity of PCR products was confirmed by sequencing. The results demonstrate an inverse relationship between stem-loop stability and tau exon 10ϩ/Ϫ mRNA ratio. In Humans (most stable stem-loop) the lowest tau exon 10ϩ/Ϫ RNA ratio is observed while in mouse (least stable stem-loop) exon 10Ϫ mRNA could not be detected. C, RT-PCR analysis of tau exon 10 alternative splicing in transgenic mice expressing the entire human tau gene (TG) and in littermate controls (Non-TG). PCR was performed on RNA isolated from adult mouse brains between exon 9 and exon 11. Primers were designed to be human-or mouse-specific. Mouse tau-specific primers gave only exon 10ϩ products (390 bp), demonstrating that splicing of the endogenous mouse gene is unaffected by the transgene. In contrast, human tau pre-mRNA in the transgenic mice was shown to undergo exon 10 alternative splicing similar to that observed in Human brain. This result is consistent with cis-elements in the human gene (such as the stem-loop) regulating alternative splicing of this exon. peated at virtually equidistant positions from the 5Ј splice site (Ϫ14 to Ϫ22 and ϩ13 to ϩ21). The significance of this repeated sequence is unclear however its position on either side of the splice site would suggest that it may be involved in the regulation of alternative splicing of exon 10 in some manner possibly as a binding site for a splicing factor. The FTDP-17 mutations ϩ13, ϩ14, and ϩ16 will alter this repeat sequence, and thus it is possible that this might be an additional mechanism by which these mutations affect the alternative splicing of exon 10 beyond the disruption of the predicted stem-loop structure that is the subject of this study.
Initial characterization of the effects of the FTDP-17 splice site mutants on the secondary structure of tau pre-mRNA was performed by examining the migration of in vitro transcribed RNAs containing exon 10 and flanking intronic sequences (247 bases) on denaturing and nondenaturing agarose gels (Fig. 8A). RNAs containing wild-type exon 10, the ϩ8, ϩ13, ϩ14, ϩ16 splice site mutants and the extended (10 bp) stem-loop (Fig. 3) were analyzed. All five in vitro transcribed RNAs ran as single bands with identical migration on denaturing (formaldehyde/ agarose) gels consistent with each RNA being the same size (247 bases). In contrast, migration on nondenaturing agarose gels differed significantly between different RNAs (Fig. 8A). The wild-type and ϩ8 mutant RNAs gave essentially identical migration patterns consistent with the ϩ8 mutation occurring in the predicted loop region and therefore not significantly affecting the stability of the stem-loop. The wild-type and ϩ8 RNAs migrated as a doublet band corresponding to one major product and to one slower minor product, suggesting that these RNAs exist in at least two confirmations. This is consistent with the prediction of the RNAFOLD analysis that there are two possible secondary structures for this region with similar stability (Fig. 7). In contrast, the FTDP-17 splice mutant (ϩ13, ϩ14, and ϩ16) RNAs, which are predicted to disrupt the stemloop, were observed as a single major band with faster migration than the wild-type RNA. This is consistent with these mutations altering RNA secondary structure presumably resulting in a more flexible molecule. Weaker minor products in the FTDP-17 mutant RNAs (ϩ13, ϩ14, and ϩ16) are also visible that appear to co-migrate with the wild-type RNA major product, suggesting that a proportion of these mutant RNA molecules maintain the wild-type secondary structure (Fig.  8A). Interestingly these minor products are strongest with the FTDP-17 ϩ16 mutation, which also produces the smallest increase in exon 10ϩ to 10Ϫ RNA ratio in the exon trapping studies, compared with the ϩ13 and ϩ14 mutations (Figs. 2  and 3). The extended (10 bp) stem RNA migrated as a single band however this product displayed marginally slower migration compared with the major wild-type product consistent with the presence of a more stable stem-loop creating a less flexible molecule. The data from this study are completely consistent with the presence of a dynamic stem-loop structure in the exon 10 RNAs that is disrupted by the FTDP-17 splice site mutants but which is stabilized by the lengthening of the In contrast, different migration rates were observed for the different tau exon 10 RNAs in nondenaturing gels consistent with predicted differences in secondary structure. B, RNA secondary structure mapping (20). RNAs in vitro transcribed from wild-type (WT), FTDP-17 ϩ14 mutant (ϩ14), and the extended 10-bp stem-loop (EXT) constructs were analyzed. Digestion sites for V1 (double strand-specific) and T1 (single strand-specific) RNases were mapped by primer extension. Control samples (C) were treated identically to RNase digestion but the enzyme was omitted. The location and sequence of the predicted stem regions are indicated both for the 6-bp and artificial 10-bp stem-loops by vertical bars.
FIG. 7. RNA secondary structure predictions for the tau exon 10 5 splice site. Proposed structures for the tau exon 10 pre-mRNA 5Ј splice site. The minimal 6-bp stem-loop in addition to alternative structures 1 and 2 predicted using RNAFOLD (28) are shown. Structure 1 (⌬G ϭ Ϫ11.9 Kcal/mol) is similar to that proposed in Ref. 3 with additional base pairing in the lower stem region; structure 2 (⌬G ϭ Ϫ12.8 Kcal/mol) is similar in stability to structure 1 but with an alternative lateral stem-loop. A repeated ACACGUCCC sequence equidistant from the splice site (Ϫ14 to Ϫ22 and ϩ13 to ϩ21, indicated by boxes) allows for the formation of either structure 1 or 2 with similar stability. predicted 6-bp stem, to 10 bp.
To further investigate the likely pre-RNA secondary structure around tau exon 10, we performed mapping analysis with RNase enzymes that recognize double-stranded (V1 RNase) and single stranded (T1 RNase) RNA (20). The sites of cleavage were mapped by primer extension with an oligonucleotide complementary to a sequence downstream of the region to be studied (see "Experimental Procedures"). In vitro transcribed RNAs containing exon 10 (wild-type, ϩ14 mutant, 10-bp stem-loop mutant) were subject to this analysis (Fig. 8B).
The extended 10-bp stem RNA gave the strongest signals for V1 (double-stranded) digestion in regions predicted to be double-stranded in the extended 10-bp stem-loop (Fig. 8B). The V1 digestion was strongest at residues (Ϫ3 to ϩ3) in the 5Ј side of the stem. Wild-type RNA also gave significant V1 signals in the predicted double strand region, on both the 5Ј and 3Ј sides of the stem-loop, although the shorter stem was reflected in the absence of bands at ϩ17 and ϩ18. In contrast, markedly weaker V1 signals were observed from the ϩ14 mutant RNA in the predicted stem region. This is consistent with the proposed disruption of the wild-type secondary structure in this region by the ϩ14 mutation. Interestingly, however, additional V1 products that are specific to the ϩ14 mutant RNA are observed in the predicted loop region (ϩ9, ϩ10), suggesting that this mutation may cause the formation of an alternative, and presumably less stable, secondary structure feature in this region. Digestion with T1 enzyme gave little obvious difference between each of the RNAs (strong bands in the 10-bp stem construct at ϩ19 to ϩ21 are also present in the control and reflect a cDNA polymerization stop). The lack of digestion in the terminal loop region with the T1 RNase may reflect the base specificity of this enzyme, cutting after a guanine residue, and the fact that the G residue at ϩ5 in the loop is predicted to base pair to the uracil residue at ϩ10 (Fig. 7). A strong stop in cDNA polymerization is observed in the control lane for the extended stem (10 bp) RNA just before the predicted stem-loop structure. This stop is specific for this RNA and is consistent with the introduction of a more stable structure at this position. While loading of RNA samples was equalized, and enzyme conditions were designed to generate one cleavage per molecule, some artifactual strengthening of V1 signals in the 10-bp stem RNA 3Ј of the stem-loop region is likely due to secondary cutting caused by the strength of digestion at the Ϫ3 to ϩ3 positions. However this artifact is still a reflection of the stabilized stemloop that is present in this RNA.
The results from the secondary structure mapping analysis are not sufficiently clear to enable a prediction of the precise structure of the stem-loop. Indeed results from RNAFOLD predictions and gel migration analyses would suggest that there may be multiple (at least two) confirmations formed in this region. However the mapping results from different RNAs, wild-type and mutant (ϩ14 and 10-bp stem), are clearly consistent with the presence of a stem-loop, which is disrupted by the FTDP-17 ϩ14 mutation. This conclusion is further supported by the gel migration studies of different tau exon 10 wild-type and mutant RNAs (described above).
Concluding Remarks-Our results demonstrate that alternative splicing of tau exon 10 (in humans and other mammals) is at least partially regulated by a stem-loop structure that forms in the pre-mRNA at the 5Ј splice site. In addition, the ratio of exon 10ϩ/Ϫ mRNAs is largely determined by the stability of this structure. Stem-loop formation at the 5Ј splice site of exon 10 is likely to compete with the binding of specific factors required for the early stages of spliceosome assembly, most likely the U1 snRNP, after transcription has occurred. Inserted stem-loops were shown previously in vivo to block U1 snRNP binding to a 5Ј splice site in the yeast RP51A gene (27). Formation of the stem-loop at the 5Ј splice site of tau exon 10 will therefore block exon 10 definition and result in skipping of exon 10 in tau mRNA. Further experiments are needed to confirm the role of the U1 snRNP in this mechanism; however, sequence changes in the 5Ј splice site, which are predicted to alter the stability of U1 snRNP binding, result in altered exon 10 splicing in vitro independent of the of stem-loop stability. The exact structure of the stem-loop region has yet to be determined with two different confirmations predicted by RNAFOLD (28) analysis (Fig. 7). It is thus clear that the stem-loop structure plays a major role in the regulation of tau exon 10 alternative splicing however additional levels of splicing control may also be mediated through the 5Ј splice site. In particular, a 9-base region that is repeated at equidistant positions from the splice site may also be involved in splicing regulation possibly by providing a binding site for a splicing factor.
The 5Ј splice site mutations that are associated with FTDP-17 (2, 3) act by disrupting this stem-loop, which leads to increased incorporation of exon 10 in tau mRNA. This in turn leads to an increase in the proportion of Tau isoforms with four microtubule-binding domains, although how this leads to neurodegeneration is yet to be determined. The identification of splice site mutations in tau associated with FTDP-17 has demonstrated the significance of Tau isoform composition to neuronal function. In addition, these mutations in tau are the first shown to cause human disease through disruption of pre-mRNA secondary structure that regulates alternative splicing.