Regulated tissue-specific alternative splicing of enhanced green fluorescent protein transgenes conferred by alpha-tropomyosin regulatory elements in transgenic mice.

The mutually exclusive exons 2 and 3 of alpha-tropomyosin (alphaTM) have been used as a model system for strictly regulated alternative splicing. Exon 2 inclusion is only observed at high levels in smooth muscle (SM) tissues, whereas striated muscle and non-muscle cells use predominantly exon 3. Experiments in cell culture have shown that exon 2 selection results from repression of exon 3 and that this repression is mediated by regulatory elements flanking exon 3. We have now tested the cell culture-derived model in transgenic mice. We show that by harnessing the intronic splicing regulatory elements, expression of an enhanced green fluorescent protein transgene with a constitutively active promoter can be restricted to SM cells. Splicing of both endogenous alphaTM and a series of transgenes carrying regulatory element mutations was analyzed by reverse transcriptasePCR. These studies indicated that although SM-rich tissues are equipped to regulate splicing of high levels of endogenous or transgene alphaTM RNA, other non-SM tissues such as spleen, which express lower amounts of alphaTM, also splice significant proportions of exon 2, and this splicing pattern can be recapitulated by transgenes expressed at low levels. We confirm the importance in vivo of the negatively acting regulatory elements for regulated skipping of exon 3. Moreover, we provide evidence that some of the regulatory factors responsible for exon 3 skipping appear to be titratable, with loss of regulated splicing sometimes being associated with high transgene expression levels.

Alternative splicing is a major form of gene regulation in multicellular eukaryotes. It allows the repertoire of expressed proteins to far exceed the number of available genes by the generation from single genes of multiple mRNAs encoding functionally distinct proteins (1)(2)(3). Bioinformatic analyses coupling expressed sequence tag and genomic sequence data bases (4) and, more recently, splice-sensitive microarrays (5) have indicated that most human genes are alternatively spliced. In addition to allowing generation of qualitatively distinct proteins, a large number of human alternative splicing events lead to quantitative regulation of gene expression by generating RNAs that are substrates for degradation by non-sense-mediated decay (NMD) 1 (6). Alternative splicing is therefore of interest as a basic mechanism that allows regulated gene expression. It is likewise of interest as a technical tool that could be used to provide regulated restriction of transgene expression, perhaps in combination with regulated promoters.
Most studies of regulated mammalian alternative splicing have used tissue culture lines to analyze splicing of transfected minigene constructs, in combination with in vitro analysis in cell extracts (7). Although this approach has allowed much progress to be made in the characterization of regulatory RNA sequences and the trans-acting factors with which they interact, it is an approach with some limitations. For example, many alternative splicing events are regulated in tissue types for which there are no reliable cultured cell lines. Smooth muscle is a good example of such a tissue. Smooth muscle cells are important in various locations in the body, and smooth muscle dysfunction underlies a number of pathologies (8). Vascular smooth muscle is responsible for controlling blood flow and systemic blood pressure. Alterations in the state of differentiation of vascular smooth muscle cells (VSMCs) from a contractile to a proliferative phenotype are associated with the development of atherosclerotic plaques and restenosis (9). The transition between the normal contractile phenotype and the proliferative VSMC phenotype is accompanied by various alterations in gene expression at the transcriptional and splicing levels (8). Despite the biomedical importance of SMCs, there are no cultured SMC lines that reliably maintain the differentiated phenotype.
We have previously used the ␣-tropomyosin (TM) gene as a model system for regulated splicing (10 -17) (Fig. 1). Exons 2 and 3 are spliced in a mutually exclusive manner, with predominant inclusion of exon 2 only in smooth muscle tissues, whereas exon 3 is used almost exclusively in striated muscles (cardiac and skeletal) and most other tissues where ␣TM expression could be observed (18,19), including proliferating VSMCs (20). Likewise, transfection of minigene constructs into almost all cell types resulted in predominant or exclusive inclusion of exon 3 (10). The exception was in PAC-1 cells, a VSMC line derived from embryonic rat pulmonary artery (11,21). The combination of restriction of exon 2 selection to smooth muscle tissues and PAC-1 cells for transfected constructs led to the concept that exon 3 represents the "default" splicing pathway, with exon 2 selection resulting only from specific regulation in SM cells.
PAC-1 cells typically show ϳ20 -50% inclusion of exon 2 in both their endogenous ␣TM gene and in transiently expressed minigenes, and have therefore facilitated characterization of the regulatory mechanisms responsible for switching selection between ␣TM exons 2 and 3 (11)(12)(13)(14)(15). The regulatory switch involves inhibition of exon 3 and does not require the presence of exon 2. Regulation is mediated by four sequence elements (11). The polypyrimidine tract of exon 3 (P3), one of the essential 3Ј splice site elements, contains three optimal binding sites for polypyrimidine tract-binding protein (PTB), a splicing repressor (22) (Fig. 1B). Midway between the P3 pyrimidine tract and exon 3, which are separated by an unusually long distance (100 nt), there is an upstream regulatory element (URE) consisting of a short cluster of UGC (or CUG) repeats (11,13). Downstream of the exon there is an essential downstream regulatory element (DRE). The DRE is composed of both a short stretch of UGC/CUG motifs, known as DUGC and a pyrimidine tract (DY). Whereas both the P3 and DY pyrimidine tracts are known to bind the repressor protein PTB (12,14,16,22), the factors that bind to the URE and DUGC elements are, as yet, uncharacterized. These regulatory elements are capable of conferring PAC-1 cell-regulated splicing to heterologous exons (15), suggesting that they may be sufficient to restrict gene expression in transgenic animals.
Whereas PAC-1 cells have proven very useful in characterizing these regulatory elements, their partial and variable state of differentiation calls into question the extent to which these findings can be generalized, especially in other smooth muscle tissues in vivo. We therefore decided to analyze control of ␣TM alternative splicing in transgenic mice, using reporter constructs based on enhanced green fluorescence protein (EGFP), in which EGFP expression was dependent upon the regulated exclusion of the ␣TM exon. Our data confirm the importance of the previously characterized regulatory elements, but also challenge the simple concept of a regulated "smooth muscle" splicing pattern and a default splicing pattern elsewhere. Finally, our data demonstrate that the intronic elements flanking ␣TM exon 3 can be harnessed to drive highly tissue-restricted gene expression in transgenic animals, and so could be a useful technical tool, especially in conjunction with tissue-restricted promoters with overlapping specificity.

MATERIALS AND METHODS
Plasmids and Cloning-The vector pCAGGS was a generous gift of Jun-ichi Miazaki (University of Osaka). To ensure appropriate initiation of translation, the Kozak sequence in the EGFP gene of pCAGGS was converted to a consensus sequence (GCCACCATGG) by PCR. The introns were inserted between bases 5 and 6 of the EGFP open reading frame in a two-step process. First, a short sequence containing a consensus 5Ј splice site and 3Ј splice site separated by unique AflII and SmaI sites was inserted. pGTM2 and pGTM3 were generated by amplification of the ␣-TM exon 2 or 3 containing cassettes from pNeoTM2 and pNeoTM3, respectively (15). In the case of GTM2, the XhoI site at the 5Ј end of exon 2 was infilled, resulting in a 4-nt frameshifting insertion. In GTM3 a G to A mutation at nt 106 of exon 3 introduced a TGA stop codon in the EGFP open reading frame. The P3⌬123, ⌬URE, ⌬DRE, and 3URE mutations were cassette-cloned into the pGTM3 background from the mutant TM constructs reported previously (12)(13)(14).
Generation of Transgenic Mice-Transgenic mice were generated by pronuclear injection of SalI-linearized plasmid DNA into CBA ϫ C57Bl/6 fertilized eggs. Viable offspring were genotyped by PCR of tail samples for the presence of the EGFP transgene.
Isolation of Tissues and RNA-Tissues were harvested from 8-to 10-week-old transgenic mice killed by CO 2 asphyxiation. To ensure consistent parts of each tissue were used, the same dissection and tissue preparation protocol was used for all animals. All tissues were placed into RNA-Later solution immediately after dissection except for the heart and aorta which were removed as a single unit and placed into phosphate-buffered saline. Excess fatty tissue around the junction of the heart and aorta was removed before the aorta was cut away from the heart. The aorta was stripped of fat before immersion in RNA-Later solution. The atria were removed, and the ventricles were placed into RNA-Later. All tissue was stored at Ϫ20°C prior to processing. RNA was isolated using the RN-Easy kit according to the manufacturer's instructions. The RNA was quantified by fluorescence using the ribogreen kit (Molecular Probes).
RT-PCR-250 ng of RNA was reverse-transcribed in a 20-l reaction using Superscript II (Invitrogen) according to the manufacturer's instructions. After heating to 70°C to inactivate the enzyme, the reactions were diluted to 100 l. For PCR detection of GTM2 splicing (Fig.  2B), 0.5 l of diluted RT reaction was used in 25-l PCRs with a 1 mM MgCl 2 -containing buffer. An 80°C hot start was followed by 30 cycles of 94°C for 30 s, 58°C for 30 s, and 72°C for 30 s using the primers 5Ј CGTM (5Ј GGCAAAGAATTCGCCACCA 3Ј) and 3Ј CGTM (5Ј GGGTG-TCGCCCTCGAACTT 3Ј). Exon inclusion gave a band of 506 bp, and exon skipping gave a band of 376 bp. For endogenous ␣-tropomyosin splicing ( Fig. 2A), the PCRs used 3 l of RT template in 25-l reactions. An 80°C hot start was followed by 40 cycles of 94°C for 30 s, 63°C for 30 s, and 72°C for 30 s using the primers mTM1 (5Ј AGAACGCCTT-GGATCGA 3Ј) and mTM4 (5Ј AACAGACGCATCCAGCT 3Ј). PCR products were then phenol/chloroform-extracted, ethanol-precipitated, and then digested with XhoI, which cuts specifically in exon 2. Exon 3 products were 204 bp, whereas exon 2 products were 150 and 54 bp.
Quantitative Real Time PCR-Q-RT-PCR was carried out using SYBR green detection for the endogenous ␣-tropomyosin transcripts and Taqman detection for the spliced EGFP transcripts.
For the endogenous ␣-tropomyosin transcripts, 5 l of RT product was added to a mixture containing 12.5 l of SYBR Green Master Mix (Applied Biosystems), 2.5 l of primer E (10 pmol/l) or F (10 pmol/l), and 2.5 l of primer G (10 pmol/l) in a final volume of 25 l. Primer E detected transcripts including ␣-⌻⌴ exon 3 and had the sequence CA-AATACTCCGAGGCTCTCAAAG. Primer F detected transcripts ␣-TM exon 2 and had the sequence GCTGGAGGAGCTGCACAAG. Primer G had the sequence GCGATCCAACTCCTCCTCAA. Samples were incubated at 50°C for 2 min and heated to 95°C for 10 min. PCR was performed using 40 cycles of 95°C for 15 s and 60°C for 1 min in an ABI 7000. Ribosomal RNA was quantified in parallel reactions using 12.5 l of Universal Master Mix (Applied Biosystems), 0.125 l of rRNA forward primer, 0.125 l of rRNA reverse primer, 0.125 l of VIC-labeled rRNA Taqman probe, and 5 l of cDNA in a total volume of 25 l.
Each PCR was shown to produce a single correct PCR product with no primer dimer formation. To test the ability of each PCR to detect the correct transcript in the presence of the alternative form of ␣-tropomyosin cDNA, mixtures of cDNAs consisting of ␣-TM exons 1-2-4 and ␣-TM exons 1-3-4 were prepared from known concentrations of each cDNA. The total concentration of DNA in these mixtures ranged from 100 to 0.1 fg, and the percentage of ␣-TM exons 1-2-4 cDNA was 100, 75, 50, 25, and 0% with the remainder being ␣-TM exons 1-3-4 cDNA. This analysis showed that neither the PCR for ␣TM2 nor the PCR for ␣TM3 detected the other cDNA. Quantitation of these data showed that the amplifications produced linear standard curves for each of the cDNAs.
For the EGFP transcripts 5 l of RT product was added to a mixture containing 12.5 l of Universal Master Mix (Applied Biosystems), 0.125 l of rRNA forward primer, 0.125 l of rRNA reverse primer, 0.125 l of VIC-labeled rRNA Taqman probe, 2.5 l of primer A (10 pmol/l) or B (10pmol/l), 2.5 l of primer C (10 pmol/l), and 1 l of probe D in a final volume of 25 l. The rRNA primers are pre-designed Taqman primers (Applied Biosystems). Primer A detected transcripts including the transgene TM exon 3 and had the sequence ACGGAGAAAAAGG-CCACAGAT. Primer B detected transcripts excluding the transgene TM exon 3 and had the sequence CGCCACCATGGTGAGCA. Primer C had the sequence GCTGAACTTGTGGCCGTTTAC, and the FAM-labeled probe had the sequence TGGTGCCCATCCTGGTCGAGCT. Samples were incubated at 50°C for 2 min and heated to 95°C for 10 min. PCR was performed using 40 cycles of 95°C for 15 s and 60°C for 1 min in an ABI 7000.
To validate this PCR, plasmids containing the EGFP gene spliced to include or skip the central exon were prepared, diluted to known concentrations (10 ng/l to 160 fg/l), and mixed. Q-RT-PCR of the mixtures showed that the PCRs produced linear standard curves over this range. Because of the similarity in the sequences and the requirement to use identical upstream primers, both PCRs detect the other form of the transcript. However, in both cases the sensitivity of detection of the alternative form was ϳ1000-fold than for the intended target. As a result, the presence of either form below 0.1% of the total product could not be determined.
Quantitation of Real Time PCR Data-Data from real time PCR experiments were analyzed by determining the cycle threshold (C t ). C t corresponds to the cycle number at which the fluorescence reaches a threshold above the base-line value. Each reaction was performed in triplicate, and the C t value for each test reaction was normalized to the rRNA value for the same sample (for determination using FAM-labeled probes) or to the rRNA value from parallel experiments (when SYBR green detection was used).
Fluorescence Microscopy-Tissues from mice were removed and embedded in Cryo-M-Bed. Ten-micron sections were cut, thaw-mounted onto superfrost plus slides, and mounted in glycerol. Sections were analyzed on a Nikon Eclipse E600 microscope. For confocal microscopy the sections were fixed in 4% paraformaldehyde for 10 min. The nuclei were stained with 4,6-diamidino-2-phenylindole. Images were captured using a Leica single photon confocal microscope with Leica TLS NT software.

Regulation of Transgene Expression by Alternative
Splicing-The intronic region upstream of ␣TM exon 3, including the branch point sequence, P3 pyrimidine tract, and URE, together with the downstream intron including the DRE are sufficient to cause enhanced skipping of ␣TM exon 3 or a heterologous exon in PAC-1 cells (15). We decided to determine whether the characterized intronic elements are able to confer appropriate tissue-specific regulation upon splicing of a heterologous exon in transgenic mice. Our reporter constructs were all based upon EGFP under the control of the ubiquitously expressed CAGGS promoter (23). In a control construct the EGFP was present as a single exon (pCAGE), whereas in the other constructs the EGFP was present as a mini-gene consisting of three exons (Fig. 1C). The first exon contained the translational initiator ATG, whereas the third exon contained the remainder of the EGFP open reading frame. The central exon was flanked by the intron regulatory elements that usually surround ␣TM exon 3. Inclusion of this exon would disrupt the EGFP open reading frame, whereas regulated exon skipping to the 3Ј end of the downstream regulatory element (DRE). The exon 3 polypyrimidine tract (P3) contains three optimal PTB-binding sites P3-1 to P3-3. In mutant P3⌬123 these elements were altered by the transition mutations below the main sequence. The upstream regulatory element (URE) was deleted in mutant ⌬URE, indicated by the dashes below, accompanied by an adjacent C to U mutation. The downstream regulatory element (DRE), consisting of a UGC rich region (DUGC) and a pyrimidine tract (DY) containing an optimal PTB-binding site (underlined), was deleted in mutant ⌬DRE, as indicated by the dashes below the sequence. C, EGFP-based splicing reporters were driven by a cytomegalovirus (CMV) enhancer and ␤-actin promoter. The EGFP open reading frame initiates in the 5Ј exon and continues in the third exon. A central exon, which was ␣TM exon 3 in all transgenes except GTM2, was flanked by the regulatory intronic regions from either side of ␣TM exon 3 and was inserted between the EGFP exons. In GTM2 the exon sequences were from ␣TM exon 2. Inclusion of the central exon led to disruption of the EGFP reading frame, whereas exon skipping, the expected smooth muscle splicing pattern, would lead to functional EGFP. would lead to an mRNA with an intact open reading frame for EGFP.
Most of our constructs used TM exon 3 as the central exon, but as an initial test we used ␣TM exon 2 (construct GTM2, Fig.  1C). This presents a particularly stringent test, because ␣TM exon 2 contains exon splicing enhancer elements stronger than those in exon 3 (24) and is usually selected in smooth muscle as a consequence of regulated repression of exon 3. Although the exon 2 exon splicing enhancers do not show cell-specific activity in cultured cells, it remains possible that they could mediate SM-specific activity in vivo, which would antagonize the negative regulation imposed by the intronic elements. In order to disrupt the EGFP open reading frame upon exon inclusion, a 4-bp insert in ␣TM exon 2 was created by infill of an XhoI site. This resulted in a frameshift such that the downstream GFP exon was out of frame.
The two initial constructs (pCAGE and pGTM2) were used to generate lines of transgenic mice by pronuclear injection. Consistent with previous reports pCAGE, transgenic mice expressed EGFP in all the tissues examined (23). These mice could be genotyped by examination under fluorescent light, and all tissues showed strong green fluorescence (data not shown). Conversely, pGTM2 transgenic mice showed no green fluorescence in the skin. Furthermore, no fluorescence was detected in the majority of tissues either as whole organs or in tissue sections. However, EGFP fluorescence was detected in the smooth muscle cells of the gut, bladder, uterus, and vasculature but not within other cell types in these tissues ( Fig. 2A). This pattern of fluorescence was observed in three independent transgenic lines. These observations are consistent with restriction of EGFP protein expression to smooth muscle cells by regulated alternative splicing, despite the widespread transcriptional activity of the CAGGS promoter.
In order to verify that the GTM2 transgene was transcribed widely and that restricted EGFP expression was because of regulated splicing, RT-PCR was carried out using RNA harvested from various tissues (Fig. 2B). RNA from the transgene could be detected in all the tissue samples, although the levels of expression appeared to be low in the aorta and liver samples. In smooth muscle tissues (e.g. aorta, bladder, gut, and uterus) products corresponding to both inclusion and skipping of the transgene central exon were observed (Fig. 2B). In the kidney and heart no exon skipping was observed, whereas the lung sample showed predominant inclusion of the exon, with a small proportion of skipping. Surprisingly, in the liver both forms of the mRNA were detected even though EGFP was not detectable, presumably due to the low levels of expression. An additional strong band was observed in the kidney, heart, and lung samples about 100 bp above the exon included band. Sequencing of this product showed that it contained an additional 107-nt insert between the TM exon and the downstream GFP exon. This insert corresponds to a pseudo-exon, flanked by a consensus 5Ј splice site and a nonconsensus GAG 3Ј splice site just downstream of the DRE. The pseudo-exon has been shown to be activated by a number of simple mutations in transfected constructs. 2 To compare splicing of the endogenous gene with the transgene, the samples were analyzed by PCR for ␣⌻⌴ using primers in exon 1 and exon 4. The amplicons generated from both ␣TM mRNA isoforms containing exons 2 or 3 are identical in size but can be distinguished by digestion with XhoI, which cleaves within exon 2 but not 3. This analysis showed that the exon 2-containing isoform was detectable in all the smooth muscle tissues and the liver but was not detected in the heart, kidney, or lung indicating that the transgene-splicing pattern mimicked the splicing pattern of the endogenous gene (Fig. 2C). The only exception was that there was no evidence of the pseudo-exon containing band with the endogenous gene.
These initial data from the GTM2 transgenic mice clearly indicated that the regulatory elements flanking ␣TM exon 3 are sufficient to confer tissue-specific repression of splicing upon a heterologous exon and that this can be harnessed to allow regulated gene expression. We therefore decided to analyze in more detail the regulation of ␣TM splicing in vivo and the role of the regulatory elements previously defined in transient PAC-1 cell transfections.
Quantitative Analysis of Endogenous ␣TM Splicing-Our initial RT-PCR analysis had indicated an unexpected degree of exon 2 selection in liver (Fig. 2C), which is not enriched in smooth muscle cells, calling into question the cell type specificity of ␣TM splicing. Before further analyzing splicing of the EGFP transgenes, we therefore decided to quantitatively ana-2 N. Grellscheid and C. W. J. Smith, unpublished observations.

FIG. 2. Smooth muscle restricted EGFP expression in GTM2
transgenic mice. The GTM2 transgene has exon 2 of ␣TM as the central exon (Fig 1C). In the endogenous ␣TM gene exon 2 is normally included in SM as a result of exon 3 repression. In the GTM2 transgene, exon 2 should be repressed in SM. A, EGFP fluorescence in tissues of GTM2 mice. Top panels, epifluorescence microscopy of tissue sections from GTM2 mice. 10-m sections of unfixed tissue prepared as described under "Materials and Methods" were photographed on a Nikon Eclipse E600 microscope using a ϫ10 objective. Lower panels, confocal microscopy of tissue sections from GTM2 mice. 10-m sections of tissue were prepared for confocal microscopy as described under "Materials and Methods." EGFP fluorescence is shown in green, and 4,6-diamidino-2-phenylindole staining of nuclei is shown in blue. CT, connective tissue; L, laminae; M, mucosa; My, myocardium; S, smooth muscle tissue; TE, transitional epithelium. B, RT-PCR analysis of GTM2 expression in mouse tissues. The exon-skipped product was mainly observed in SM tissues (gut, bladder, uterus, and aorta) as well as liver. Other tissues showed mainly the exon-included product. Kidney, heart, and lung also showed an additional band corresponding to inclusion of both the central ␣TM2 exon and a 107-nt pseudo exon that lies just downstream of the DRE. C, RT-PCR analysis of endogenous ␣-TM in the same RNA samples as B. After PCR, the products were digested with XhoI, which cuts specifically within exon 2. As for the GTM2 transgene the regulated splicing pattern was observed in gut, bladder, uterus, aorta, and liver. lyze splicing of exons 2 and 3 of ␣TM across a range of tissues. This had previously been surveyed qualitatively by nuclease protection and was restricted primarily to those tissues with high levels of expression (18,19).
The levels of ␣TM mRNAs containing either exon 2 or exon 3 were determined by quantitative real time PCR and compared with the levels of rRNA in the same mouse tissue RNA samples. The proportion of ␣TM mRNA containing exon 2 (the "regulated" or SM pathway) was plotted against ␣TM mRNA expression level relative to rRNA (Fig. 3). Tissues with the highest levels of expression (heart, skeletal muscle, bladder, and aorta) showed the most extreme variations in splicing pattern. The striated muscle tissues, heart and skeletal muscle, which had the highest level of ␣TM expression, showed less than 0.2% exon 2 selection. In contrast, aorta and bladder, with the next highest levels of ␣TM expression, had more than 95% exon 2 selection. The next highest amount of ␣TM mRNA was present in the small intestine and lung, two tissues with a significant smooth muscle content. The amount of the exon 2-containing ␣TM in these samples was 80% for the small intestine and 29% in the lung. The brain had the lowest level of exon 2 inclusion (7%) apart from heart and skeletal muscle. Surprisingly, the other three tissues, none of which have a major smooth muscle component, showed much higher proportions of exon 2 inclusion: kidney 31%, spleen 55%, and liver 75%. Notably, these tissues had the lowest levels of ␣TM expression, with levels in liver more than 1000-fold lower than in heart and skeletal muscles. These expression data indicate that the concept of a regulated splicing pattern in smooth muscle cells, which splice exclusively or predominantly exon 2, and a default inclusion of exon 3 in all other cell types needs to be re-examined (see "Discussion").
Variable Regulation of ␣TM Exon 3 Splicing in Different Mouse Lines-Having established the splicing patterns of endogenous ␣TM, we proceeded to generate a series of transgenic mouse lines based upon the pGTM3 construct (Fig. 1C), in which the central exon was ␣TM exon 3 flanked by its own regulatory introns. A single point mutation (G3 A) at position 110 of exon 3 generated a stop codon in the EGFP reading frame. Three independent transgenic mouse lines were initially generated using the GTM3 transgene. The pattern of splicing in a range of tissues was determined by reverse transcription followed by Q-RT-PCR for exon-containing and skipped RNA isoforms. Examination of the products by agarose gel electrophoresis showed a single band for both reactions in most tissues (data not shown). The exon inclusion PCR also detected small amounts of product containing exon 3 spliced to the pseudo-exon in heart, skeletal muscle, and brain. Because this product contains exon 3, it was scored as exon inclusion for the purposes of quantitation. All three transgenic GTM3 lines (denoted r, s, and x) showed clear regulation of splicing ( Fig. 4 and Table I). This was most evident in the negligible levels of exon skipping in heart, skeletal muscle, and brain (Ͻ1.2%) and very high levels of exon 3 skipping in bladder (Ͼ78%). Splicing patterns in the other tissues were much more variable. Aorta gave high levels of exon skipping (Ͼ48%), but with greater variability than bladder, whereas levels of exon skipping in the small intestine varied between 5 and 86%. The splicing patterns in liver, kidney, spleen, and lung were intermediate between the striated and smooth muscle tissues but with a high degree of variability. When the degree of transgene exon skipping was plotted against the percentage of endogenous ␣-TM exon 2 splicing (the equivalent splicing pattern to transgene exon skipping) in the same tissues, none of the mouse lines produced a straight line, which would be the expected pattern if transgene and endogenous ␣-TM splicing were regulated identically. Rather, all three mouse lines produced a curved plot, with lower regulation of transgene splicing compared with endogenous ␣-TM particularly evident in liver, spleen, and kidney. Only line "s" showed a reasonable correlation between endogenous and construct splicing (Fig. 4). The other two lines (r and x) showed a greater variation between endogenous ␣TM and GTM3 splicing. Whereas the high degree of regulation between bladder and striated muscles (cardiac and skeletal) was maintained in all three lines, there were lower levels of transgene exon skipping in small intestine, liver, spleen, and to a lesser extent kidney and lung in lines "r" and "x" as compared with s. Notably, line s, which showed the best correspondence between transgene and endogenous ␣TM splicing, had the lowest levels of transgene expression in all 10 tissues examined (data not shown). This suggests that whereas small intestine, liver, spleen, kidney, and lung are able to splice the endogenous ␣TM gene to give significant levels of exon 2 incorporation, they do not provide a "robust" regulatory environment that is able to support similar levels of regulated splicing of overexpressed transgenes (see below).
To analyze further the variability of transgene splicing, five  Table I). The data are plotted as % transgene exon skipping against % ␣TM exon 2 inclusion. Perfect co-regulation of transgene and endogenous ␣TM splicing would therefore result in a straight line with a gradient of 1 that crosses the origin. Diamonds, line r; squares, line s; triangles, line x. Bl, bladder; A, aorta; SI, small intestine; Li, liver; Sp, spleen; K, kidney; Br, brain; Lu, lung; H, heart; Sk, skeletal muscle. Line s showed the best correlation between endogenous and transgene splicing (gradient ϭ 1.00, correlation ϭ 0.92). more independent pGTM3 transgenic mice were generated, and splicing patterns were determined in RNA isolated from the aorta, lung, small intestine, and spleen. Transgene exon skipping was highly variable in small intestine, kidney, and spleen but less variable in aorta (Fig. 5, A-D, and Table I). A weak trend of decreased transgene exon skipping associated with increased transgene expression was observed in some, but not all, tissues examined. For each tissue the mouse line with the highest expression levels gave one of the lowest levels of exon skipping (line r in aorta and small intestine, z in lung, and y in spleen, see Fig. 5).
The variability in tissue specificity of GTM3 transgene splicing between individual mouse lines posed a potential problem for our subsequent goal of analyzing the effects of mutations in the splicing regulatory elements. However, the extreme splicing patterns of exon skipping in bladder and inclusion in skeletal muscle were robustly maintained in all the GTM3 mice lines, irrespective of expression levels (Table I and Fig. 4). We therefore adopted a regulation index based upon the arithmetic difference in exon skipping between bladder and skeletal muscle (Table III). The alternative "fold difference" approach is potentially more meaningful in terms of the underlying mechanism and tissue specificity of regulation, but it is vulnerable to large fluctuations arising from variations in the very small values of transgene exon skipping determined in skeletal muscle. Conversely, these variations have negligible impact upon the arithmetic index; the arithmetic regulation index for GTM3 mice was 79.2 Ϯ 2.2, whereas the fold regulation index was 239 Ϯ 90 (Table III). Use of the arithmetic regulation index subsequently allowed a meaningful analysis of the effects of mutations in splicing regulatory elements despite the line-toline variability of transgene splicing in other tissues.
Effect of Mutations to the Known Splicing Regulatory Sequences-To analyze the role of the regulatory elements in the introns flanking exon 3, transgenic mice were generated containing a series of derivatives of pGTM3 with mutations in the defined regulatory elements (Fig. 1C). The P3⌬123 mutant contains four pyrimidine transitions in the three optimal PTBbinding sites within the polypyrimidine tract of exon 3, ⌬URE contains a 15-nt deletion of the URE, and ⌬DRE has a 90-nt deletion of the DRE (Fig. 1B). These three mutations all cause a decrease of ␣TM exon 3 skipping in PAC-1 cells (12,14). In contrast, the 3URE mutant contains two additional copies of the URE and in various cell lines increases the levels of exon 3 skipping (13).
Consistent with the previously reported effects in transfected cells, all four mutations had obvious effects in vivo. The P3⌬123, ⌬URE, and ⌬DRE mutations caused a progressively severe reduction of transgene exon skipping in all cell types, reflected in a marked reduction in the regulation index (Table   III). Deletion of the DRE (three lines) had a greater effect than deletion of the URE (three lines) in all the transgenic mice studied. In all tissues deletion of the DRE reduced exon skipping to less than 1% (Table II). This strong effect is consistent with the fact that the DRE consists of two separable regulatory elements, the UGC region and the DY pyrimidine tract. The ⌬URE and P3⌬123 mutations had a quantitatively similar effect as ⌬DRE in a number of non-smooth muscle tissues, reducing exon skipping to below 1%. In tissues with a significant smooth muscle content (bladder, aorta, and small intestine), there was significant residual transgene exon skipping with ⌬URE (as high as 9% in bladder and 22% in aorta of line "j"), whereas P3⌬123 mutants retained up to 28% exon skipping in the bladder and 32% in the aorta. The increasingly severe impact of the mutations (P3⌬123 Ͻ ⌬URE Ͻ ⌬DRE) was particularly evident in the arithmetic regulation index (Table  III). When regulation was assessed by the fold difference in exon skipping, the P3⌬123 and ⌬URE mutations had a less obvious effect, whereas the more severe ⌬DRE mutation impaired regulation using either index. These data clearly demonstrate that the regulatory elements identified by transfection of cultured cells play a crucial role in regulation of ␣TM splicing in vivo.
Multimerization of the URE-Multimerization of the URE enhances TM exon 3 skipping in PAC-1 and various other cell lines (13), with a concomitant decrease in regulation because of the elevated skipping in the non-SM cells. We therefore generated transgenic mice expressing a spliced transgene containing three tandem copies of the URE (3URE). Widespread EGFP fluorescence was observed in all lines of the 3URE mice (data not shown), suggesting reduced restriction of transgene expression via regulated splicing. Analysis of transgene RNA showed  the 3URE mutation to have a more complex effect than the other mutations, with a higher degree of line-to-line variability within some tissues but not others (Table II). All 3URE lines produced more of the exon-skipped mRNA than the GTM3 lines in brain, heart, and kidney and to a modest extent in skeletal muscle, i.e. the tissues that usually exhibit the lowest level of exon skipping. In many of the other tissues the two transgenes showed similar splicing patterns, but in bladder and aorta the 3URE lines showed higher variability and overall lower levels of exon skipping compared with the wild type GTM3 transgene. This variability showed a clear correlation with transgene expression levels (Fig. 6). In all of the tissues examined increased transgene expression was accompanied by reduced exon skipping, even though different mouse lines showed the highest levels of expression in different tissues. In a number of tissues exon skipping decreased in a linear manner over a particular range of expression levels (e.g. bladder, skeletal muscle, spleen, small intestine, and aorta, Fig. 6). The URE multimerization had no obvious impact upon the regulation indices (Table III), mainly because of the limited effect upon the very low levels of exon skipping in skeletal muscle (Table II).
Titratable Regulators?-The preceding observation of a trend for decreased transgene exon skipping with higher expression levels suggested that regulatory factors responsible for skipping of ␣TM exon 3 are titratable by the 3URE, and to a lesser extent the GTM3, pre-mRNAs. In this case, one would predict that splicing of the endogenous ␣TM RNA might be affected by transgene expression in some tissues. To test this prediction we analyzed endogenous ␣TM splicing in spleen and aorta, the two tissues that showed the clearest relationship between 3URE transgene expression levels and splicing (Fig.  6). Exon 2 inclusion in the spleen varied from 20 to 40% of the total ␣TM (Fig. 7, A and B), and there was no direct correlation with the level of transgene expression (Fig. 7B). Furthermore, the amount of endogenous ␣TM splicing using exon 2 did not correlate with the amount of the transgene exon skipping (Fig.  7A). This suggests that any regulatory factors that are titrated by high levels of 3URE RNA in spleen are not responsible for regulation of the endogenous ␣TM splicing.
By contrast, in the aorta there was a relationship between the amount of regulated splicing of the endogenous ␣TM exon 2 and exon skipping in the 3URE transgene (Fig. 7C). The mouse lines with the highest level of 3URE expression also had the lowest levels of regulated splicing of both the transgene and the endogenous ␣TM (Fig. 7D). Moreover, when ␣TM exon 2 selection was plotted against transgene expression levels, a clear decrease in exon 2 selection was observed to accompany increased expression of the 3URE transgene in aorta (Fig. 7D). In contrast, none of the GTM3 lines showed any variation in ␣TM exon 2 splicing (Fig. 7, C and D). These observations suggest that over the expression range analyzed in the aorta, the 3URE, but not the GTM3, transcript is able to titrate factors that are involved in regulation of the endogenous ␣TM splicing. DISCUSSION Our data from transgenic mice have validated earlier findings by using the partially differentiated PAC-1 cells and have a Arithmetic regulation index ϭ % exon 3 skipping in bladder minus exon 3 skipping in skeletal muscle.
b Fold regulation index ϭ % exon 3 skipping in bladder divided by exon 3 skipping in skeletal muscle.
FIG. 6. Relationship of 3URE transgene splicing and expression levels. The percentage of 3URE transgene exon skipping is plotted against transgene expression level relative to rRNA for lung (A), spleen (B), small intestine (C), and aorta (D) from five lines of 3URE mice. In contrast to the "wild type" GTM3 mice (Fig. 5), there was a clear relationship between 3URE expression levels and splicing, particularly in aorta and spleen. Higher levels of expression were accompanied by lower percentages of transgene exon skipping. As a reference, the horizontal dashed line indicates the splicing pattern of endogenous ␣TM (for which % exon 3 skipping is equivalent to exon 2 inclusion). also provided important new insights not available by using cultured cells. The first important observation was that the splicing of endogenous ␣TM does not follow a simple smooth muscle versus default model. This was evident both from analysis of the endogenous ␣TM gene (Fig. 3) and from the GTM3 lines that most closely recapitulated native regulation (Fig. 4, line s). Whereas the smooth muscle rich tissues such as bladder and aorta showed nearly complete exon 2 inclusion and the striated muscles showed full inclusion of exon 3, a number of other tissues with little smooth muscle content, such as liver and spleen, showed significant percentages of exon 2 inclusion. Notably, these tissues had very low levels of ␣TM expression, which had not been detectable using nuclease protection or Northern blot (18). Subsequent analyses of GTM3 transgene splicing indicated that bladder and aorta provided a robust regulatory environment that was able to support significant levels of transgene exon skipping in all mouse lines, even with high levels of transgene expression. The degree of regulation was not as great as with the endogenous gene (79% compared with 99%, Table III), possibly resulting from the lack of a competing mutually exclusive exon in the transgene (Fig. 1), or from the truncations in the flanking introns. In contrast to bladder and aorta, the other tissues were more sensitive to the levels of transgene expression and could only support exon skipping at lower levels of expression. Expression of exon 2 containing ␣TM isoforms has been detected previously in murine ES cells at stages of embryonic development preceding the formation of SM tissues (25). However, that analysis did not look at the contemporaneous expression of exon 3-containing isoforms, so it is not clear whether the detected exon 2-containing ␣TM was a major or minor isoform. Similar challenges to simple default versus regulated models for splicing mechanisms have also been reported in other systems (26,27). Although the concept of smooth muscle and default splicing patterns does not accurately describe the ratio of exon 2-and 3-containing ␣TM isoforms across various tissues, it neverthe-less appears to reflect a property unique to smooth muscle of having the capacity to confer regulated splicing even to high levels of both endogenous and ectopically expressed RNA. Likewise, the striated muscle tissues (skeletal and cardiac muscle) are able to robustly splice high levels of endogenous or transgene RNA to include exclusively ␣TM exon 3. Exon 3 inclusion does not require specific regulatory elements (Tables II and III) (11)(12)(13), so it appears most likely that striated muscles contain very low levels of the regulators responsible for the SM-splicing pathway. This could result from very low absolute levels. Alternatively, the low activity of regulatory factors in striated muscles could result from their titration by the very high levels of ␣TM expression in these tissues.
Various lines of evidence have indicated that use of heterologous promoters can sometimes alter the regulation of downstream alternative splicing events (reviewed in Ref. 28). This could present a serious impediment to the approach that we adopted, in which a widely active promoter drives expression of an alternatively spliced transgene. Therefore, it was gratifying that, despite the considerable line-to-line variability, in some mouse lines splicing of the GTM3 transgene was regulated in parallel with endogenous ␣TM splicing (Fig. 4). The variability in GTM3 splicing behavior between individual mouse lines probably arises from more than one cause. First, variations in absolute levels of expression in individual tissues such as spleen appeared to give rise to titration effects, such that higher levels of transgene expression were accompanied by lower proportions of exon skipping (Figs. 5 and 7). Further variability may have arisen in tissues such as small intestine that have a mixture of SM and other cell types. Although the pCAGGS promoter is active in most cell types, it nevertheless shows variable activity between tissues. Furthermore, we observed that the hierarchy of activity in different tissues varied between individual mouse lines. Therefore, within tissues such as small intestine higher levels of transgene expression in the non-SM cells would lead to an apparent unregulated splicing pattern, even though the fluorescence microscopy showed that EGFP expression was restricted to the SM cells of the gut (Fig.  2). The combination of these two factors may explain the observation that small intestine showed the greatest variation in splicing pattern of the GTM3 transgene (Table II, 5-86% exon  skipping). Such line-to-line variability that we observed could be circumvented by targeted insertion of the transgene by homologous recombination into the Hprt locus (29). However, integration at the Hprt locus facilitates high level constitutive transcription, so it is possible that mouse lines would not express the transgene at sufficiently low levels to recapitulate appropriate tissue-specific regulation.
The effects of transgene expression levels upon regulated splicing of both transgene and endogenous ␣TM indicate that regulatory factors distinct from the general splicing machinery can be titrated by the transgene RNA. The strongest titration effects were observed with the 3URE transgene, which at high levels led to lower exon 2 selection for endogenous ␣TM in the aorta (Fig.  7). At lower expression levels in non-SM tissues, the 3URE construct generally led to higher levels of exon skipping compared with GTM3, consistent with the concept that URE multimers can help to recruit limiting amounts of regulatory factors (13). In contrast, in bladder and aorta, where these factors are presumably not limiting, 3URE behaved similarly to GTM3, at least at low levels of expression. The URE consists of a short region of UGC or CUG repeats, and thus far the regulatory factors that bind to it have not been identified. Obvious candidates include members of the CELF family of RNA-binding proteins, which bind to CUG or UG repeats (30,31). However, overexpression of known family members (CUG-BP, ETR3, and CELF4) antago- FIG. 7. Alteration of endogenous ␣TM splicing in aorta of 3URE but not GTM3 transgenic mice. The relationship between 3URE transgene expression levels and endogenous ␣TM splicing was investigated in spleen (A and B) and aorta (C and D) samples. A and C, transgene exon skipping was plotted against ␣TM exon 2 inclusion. B and D, transgene expression level (relative to rRNA) was plotted against ␣TM exon 2 inclusion. Although there was no relationship between transgene splicing/expression and endogenous ␣TM splicing in spleen (A and B), in aorta the ␣TM exon 2 inclusion correlated with 3URE transgene expression (D) and splicing (C). For comparison, the data from aorta for GTM3 transgenic lines is also shown in C and D.
nized regulation of ␣TM splicing, suggesting that the tested proteins were not authentic regulators of this system (13). In the aorta, expression of 3URE appeared to titrate the factors responsible for regulating splicing of both the transgene and endogenous ␣TM. In contrast, the situation in the spleen was more complex. Although the 3URE transgene showed a clear decrease in exon skipping as its level of expression increased, there was no apparent effect upon splicing of the endogenous ␣TM. This suggests that in this case there are distinct pools of regulatory factors, one of which was used by the 3URE transgene and another by the endogenous ␣TM.
Knock-out of ␣TM in transgenic mice has shown that homozygous null mutants are embryonic lethal but that loss of a single allele produces no phenotype, probably due to compensating mechanisms that maintain the quantity of the ␣TM protein (32,33). There have been no reports of isoform-specific knockouts, although this was the intention of at least one previous investigation (33). An interesting consequence of the titratability of some of the splicing regulators is that in the highest expressing 3URE mouse line the level of exon 2 inclusion in the aorta was reduced from the usual ϳ90% to less than 10%. This suggests that depletion of the exon 2 isoform does not cause a lethal phenotype, although it is possible that more detailed investigation of such animals might reveal a specific phenotype in the vasculature or other smooth muscle tissues. It is also possible that the reduction in splicing of the TM2 isoform resulted in up-regulation of ␣TM expression to compensate for the splicing dependent loss of ␣TM. Consistent with this suggestion, the mice with the highest levels of 3URE expression and only 4% TM exon 2 inclusion had higher overall levels of TM transcripts in the aorta, so that the reduction in absolute levels of the TM exon 2 isoform was only ϳ4-fold.
In addition to exon inclusion and skipping, a third splicing pathway was observed with both the GTM2 and GTM3 constructs. This involved both normal exon inclusion as well as an additional 107 nt from the downstream intron. We have observed splicing of this 107-nt "pseudo-exon" in transfected mutant constructs in HeLa and PAC-1 cells, 2 resulting from simple mutations that activate its splice sites. Inclusion of the pseudo-exon results in the introduction of in-frame premature termination codons in RNA from both construct and endogenous ␣TM. Because these premature termination codons are more than 55-nt upstream of the splice junction between the pseudo-exon and the GFP exon (or ␣TM exon 4), they are predicted to provoke NMD (34,35). This means that in our quantification of transgene splicing, we will have underestimated the level of exon inclusion, because products containing the pseudo-exon are likely to be at levels that under-represent the actual frequency of the splicing event. Nevertheless, because pseudo-exon inclusion mainly occurred in tissues such as heart and skeletal muscle, in which exon inclusion is predominant, NMD of pseudo-exon-containing products is unlikely to cause a large distortion of the final quantitated data. Despite the clear inclusion of the pseudo-exon in the construct RNA, its inclusion in endogenous ␣TM RNA is much more difficult to detect. There are a number of possible explanations for the apparent contrast between the behavior of the pseudo-exon in the transgene and endogenous ␣TM. First, the smaller size of the intron downstream from the pseudo-exon in the transgene (174 bp) compared with the endogenous ␣TM gene (ϳ12 kbp) may favor pseudo-exon inclusion. Second, in the transgene there is a single intron downstream of the pseudo-exon stop codon, whereas in the endogenous gene there are 7-8. Although a single intron provoked NMD as efficiently as 2-4 downstream introns in the TPI gene (36), the additional exon junction complexes in ␣TM might lead to much more efficient NMD of the endogenous product compared with the transgene.
Whereas investigations of alternative splicing in genetically tractable organisms such as Drosophila melanogaster and Caenorhabditis elegans have routinely used transgenic animals, studies of mammalian alternative splicing have typically focused on cell culture in order to analyze regulatory elements. However, transgenic mice have been used in some analyses of alternative splicing mechanisms. The earliest case involved widespread expression of a calcitonin/CGRP transgene, showing that the majority of tissues were able to produce calcitonin as a default product, even though it is usually only produced in the thyroid. In contrast, transgene processing to CGRP was limited to neurons (37). More recently, an artificial transgene with an arrangement of competing 3Ј end processing and splice sites, analogous to the immunoglobulin (Ig) heavy chain gene, was found to be co-regulated with the Ig gene in transgenic mice, arguing against the need for sequence-specific regulators of Ig processing (27). Moreover, because the transgene was more widely expressed than the Ig gene, it could be seen that neither of the splicing patterns in resting or lipopolysaccharide-stimulated B cells represented a default or regulated state. Analysis of transgene tissue-specific splicing has been carried out for exon 9 of the F1␥ gene (38). However, only two mutant minigenes were tested in transgenic animals so that a direct comparison with a wild type construct could not be made. Alternative splicing of Smn exon 7 has also been analyzed in transgenic mice, although no tissue specificity of splicing was observed (39). Our investigation is one of the most detailed quantitative analyses to date of a mammalian model alternative splicing system in vivo. It has served to validate findings from the partially differentiated PAC1 cell line, highlighted some potential complications with transgenic models, and also given a number of important insights that would not have been available using previous methods.