Regulatory Sequences of the Mouse Villin Gene That Efficiently Drive Transgenic Expression in Immature and Differentiated Epithelial Cells of Small and Large Intestines*

Villin is an early marker of epithelial cells from the digestive and urogenital tracts. Indeed villin is expressed in the stem cells and the proliferative cells of the intestinal crypts. To investigate the underlying molecular mechanisms and particularly those responsible for the restricted tissue specificity, a large genomic region of the mouse villin gene has been analyzed. A 9-kilobase (kb) regulatory region of the mouse villin gene (harboring 3.5 kb upstream the transcription start site and 5.5 kb of the first intron) was able to promote transcription of the LacZ reporter gene in the small and large intestines of transgenic mice, in a transmissible manner, and thus efficiently directed subsequent β-galactosidase expression in epithelial cells along the entire crypt-villus axis. In the kidney, the transgene was also expressed in the epithelial cells of the proximal tubules but is likely sensitive to the site of integration. A construct lacking the first intron restricted β-galactosidase expression to the small intestine. Thus, the 9-kb genomic region contains the necessary cis-acting elements to recapitulate the tissue-specific expression pattern of the endogenous villin gene. Hence, these regulatory sequences can be used to target heterologous genes in immature and differentiated epithelial cells of the small and/or large intestinal mucosa.

Transgenic mice are routinely used to study the molecular and cellular basis of normal and pathological states in intestinal mucosa (1)(2)(3)(4)(5). The major limitation regarding the targeting of exogenous transgenes in this tissue is that the epithelium of the mouse intestinal mucosa is renewed every 2-5 days (6 -8). The epithelial cells arise from multipotent stem cells functionally anchored at the base (more precisely in the lower third) of the proliferative compartment of the epithelium, the crypts of Lieberkü hn. These crypts display a monoclonal organization because they are each derived from a single progenitor cell (9). Descendants of stem cells multiply in the middle portion of each crypt (10) and gradually differentiate into four principal cell types. In the small intestine, absorptive enterocytes (constituting Ͼ80% of the epithelial cells), mucus-producing goblet cells, and enteroendocrine cells migrate upward from the crypts to the apex of surrounding villi (whose colonic counterparts are hexagonal-shaped cuffs) (11), where they become apoptotic and are exfoliated into the gut lumen (12). In contrast, antimicrobial peptides secreting Paneth cells migrate to the bottom of the crypts, where they reside for about 20 days (13).
Given the remarkable protective effect of this epithelium, it is not surprising that most previous studies aiming to induce neoplastic transformation in intestinal mucosa of transgenic mice have failed (14,15). In these reports, the use of promoter sequences that direct oncogenes in nonproliferating enterocytes located in the upper third of crypts produce only minor phenotypic abnormalities without tumorigenic consequences in the gut epithelium, suggesting that the residence time of these villus-associated cells may not be sufficient for the oncogenes to exert their effects. Furthermore this suggests that transgenic mouse models of neoplasia may require an efficient targeting of oncogenes in crypt stem cells or their immediate descendants.
Villin is a cytoskeletal protein that is mainly produced in epithelial cells that develop a brush border responsible for absorption as in the digestive apparatus (epithelial cells of the large and small intestines) and in the urogenital tract (epithelial cells of the kidney proximal tubules). Because it is expressed in the proliferative stem cells of the intestinal crypts (16,17), it is believed to be an early marker for committed intestinal cells. The multiple levels of regulation control villin gene activity during mouse embryogenesis (18 -20) and account for the strict pattern of tissue-specific expression observed in adults. Moreover, the expression of the villin gene in intestinal epithelial cells is conspicuously maintained in their corresponding carcinomas (21)(22)(23)(24).
The specific expression pattern of villin suggests that it is an appropriate candidate for the characterization of regulatory sequences that could allow targeting of heterologous genes into a selected population of cells in the mouse digestive tract. With this goal in mind, the human villin gene has been isolated and characterized (25). A 2-kb 1 5Ј-flanking region has been found to contain sufficient regulatory elements to promote tissue-specific expression of a reporter gene in intestinal and renal cell lines (26). In transgenic mice, this regulatory region is able to drive the expression of the human Ha-ras oncogene in the tissues in which the endogenous gene is actively transcribed. However, low levels of expression were observed that did not trigger malignant tissue appearance into the gut of these animals. 2 These observations led us to further analyze an extended genomic region of the mouse villin gene with the goal of map-ping additional elements localized 5Ј and/or 3Ј and involved in promoting high levels of transgenic expression in the intestinal mucosa. Here we report the analysis of tissue-specific expression of the mouse villin gene using: (i) DNase I-hypersensitive sites assays, (ii) transient transfection assays, and (iii) transgenic mice.

EXPERIMENTAL PROCEDURES
Cell Culture and ex Vivo Transient Transfection-Human colon carcinoma enterocyte-like CaCo2 cells, pig kidney proximal tubules-derived LLCPK1 cells, and distal tubules-derived Madin-Darby canine kidney cells were cultured as described (26). Cells cultures were cotransfected using 15 l of Lipofectin reagent (Life Technologies, Inc.) with 5 g each of ␤-galactosidase reporter plasmid construct and the control plasmid pRSVLuc, which contains the luciferase gene under the control of the Rous sarcoma virus promoter. Cells were harvested 48 h later, and cell extracts were assayed by chemiluminescent detection of both ␤-galactosidase (Galacto-Light, Tropix, Inc.) and luciferase (Luciferase Assay Kit, Tropix, Inc.) activities using a luminometer (Berthold). ␤-Galactosidase activity (light units) was corrected for variations in transfection efficiencies as determined by luciferase activity. All transfections were repeated at least three times. Results are expressed as fold induction over that of the vector without promoter, pBasic.
Primer Extension Analysis-Total RNA was isolated from mouse intestine with RNA NOW reagent (Biogentex). For primer extension assay, 2 ng of 32 P-labeled oligonucleotide probe (5Ј-GAGTGGTGATGT-TGAGAGAGCCT-3Ј) complementary to nucleotides ϩ81 to ϩ103 of the murine villin cDNA (GenBank TM accession number M98454) was hybridized with 30 g of total RNA at 60°C (0.25 M KCl, 10 mM Tris-HCl, pH 7.5, 1 mM EDTA) for 90 min. Transcription with 5 units/l of Moloney murine leukemia virus reverse transcriptase (Life Technologies, Inc.) was carried out at 37°C for 90 min in 300 l of a solution containing 75 mM KCl, 3 mM MgCl 2 , 50 mM Tris-HCl (pH 8.3), 10 mM dithiothreitol, 0.75 mM deoxynucleoside triphosphates, 75 g/ml actinomycin D, and 0.3 units/l RNasin. The primer extension products were separated by electrophoresis in denaturing 8% polyacrylamide gels. The full-length extension product (105 nucleotides) was obtained by comparison with the length of the comigrating sequencing reaction products. A primer extension control experiment was performed on the same total RNA preparation using a 32 P-labeled oligonucleotide probe (5Ј-CATAGTTCTCGTTCCGGT-3Ј) complementary to nucleotides ϩ63 to ϩ80 of the mouse intestinal fatty acid-binding protein cDNA and generating an 81-nucleotide extension product (27).
DNase I-hypersensitive Site Analysis-Tissues from 30 mice were used per assay of intestine, kidney, liver, and spleen. Nuclei preparation and DNase I digestion were performed as described (28) with minor modifications. Nuclei were digested without or with 20 to 160 units of DNase I (DPRF Worthington) for 10 min at 0°C. 10 g of purified DNA was digested overnight with restriction enzyme (BamHI or BglII). The DNase I-hypersensitive sites were analyzed by Southern blotting using the BglII-PstI probe (0.5 kb) (as indicated in Fig. 2).
Plasmid Construction-All constructs described were subcloned into the pBluescript II KS vector (Stratagene) with fragments isolated from a DASHII phage containing a 16.3-kb region (9 kb upstream and 7.3 kb downstream from the translation initiation codon) of the mouse villin gene (kindly furnished by G. Tremp, Rhône-Poulenc Rorer) (29). The pD1 construct (as described in the legend to Fig. 3B) was prepared by ligating a BamHI fragment of 5.1 kb (1.8 kb upstream from the ATG translation initiation codon of the mouse villin gene, subcloned 5Ј to the nuclear localization signal-␤-galactosidase gene-SV40 polyadenylation site, using a polymerase chain reaction (PCR) strategy) at the BamHI site in a plasmid containing the 3.7-kb region of the mouse villin gene (immediately 5Ј to the 1.8-kb region described above). The pA1 and pA2 (containing an internal 1-kb deletion) constructs have resulted from several steps based on the BstEII sites present in the 3.7-kb region described above and in a plasmid containing the 3.5-kb region of the mouse villin gene (immediately 5Ј to the 3.7-kb region). The pC1 and pC2 constructs were derived from the pA1 and pA2 plasmids cut with ApaI and re-ligated, respectively. To generate the pB1 construct, a BglII fragment (480 bp) from the 3.5-kb region described above was excised and cloned into the KpnI site of the pC1 plasmid. The pA3, pB3, and pC3 constructs correspond to the pA1, pB1, and pC1 deleted from the intron 1 (see Fig. 3B). The sequence between the transcription initiation start site and the translation initiation codon, excluding the intron 1, was deduced from that of the murine villin cDNA and was introduced into the BglII-NcoI sites of the pC1 construct by using a dimerized oligodimer made of a coding-strand oligonucleotide (5Ј-GAT-CTCCCAGGTGGTGGCTGCCTCTTCCAGACAGGCTCGTCCAC-3Ј) and a noncoding-strand oligonucleotide (5Ј-CATGGTGGACGAGCCTG-TCTGGAAGAGGCAGCCACCACCTGGGA-3Ј), resulting in the pB3 construct. The pA3 and the pC3 constructs were derived from the pB3 plasmid by ligating an ApaI fragment (3.1 kb) and a BglII fragment (480 bp), both from the 3.5-kb region described above, at the ApaI site in the pB3 plasmid, respectively. Subcloning steps were confirmed by DNA sequencing.
Transgenic Mice Generation-The transgenes digested with XhoI-NotI were injected into the pronuclei of the fertilized eggs of the B6/D2 mice in collaboration with the Service d'Experimentation Animale et de Transgénèse, CNRS (Villejuif, France). Mice carrying transgenes were first identified by PCR of genomic DNA to confirm the presence of the ␤-galactosidase gene and then analyzed by Southern blotting to determine the copy number of the integrated transgene. Each founder animal harbored one copy of the transgene per genome. Small intestine, colon, kidney, stomach, liver, heart, lung, thymus, brain, spleen, and muscle were dissected from transgenic mice and either prepared for total RNA extraction or embedded to perform cryosections.
Reverse Transcription-PCR Analysis-Total RNA was isolated with SV Total RNA Isolation System (Promega). 20 ng of pd(N) 6 random primer (Pharmacia) were hybridized with 2 g of total RNA at 70°C for 10 min in distilled water. Reverse transcription with 200 units of Moloney murine leukemia virus reverse transcriptase (SuperScript II, Life Technologies, Inc.) was carried out at 37°C for 90 min in a 20-l solution of 1ϫ First Strand Buffer (Life Technologies, Inc), 10 mM dithiothreitol, 0.5 mM deoxynucleoside triphosphate, and 0.4 units/l RNasin. 2 l of the resulting cDNAs were amplified by PCR reaction in 50 l for 40 cycles. Each cycle consisted of 60 s at 94°C, 60 s at 51°C (for transgene and villin) and 57°C (for TFIID), and 30 s at 72°C. For the transgene primers, 5Ј-CAACTTCCTAAGATCTCC-3Ј coding strand and 5Ј-ATTCAGGCTGCGCAACTGTT-3Ј noncoding strand were used, generating a 250-bp product. For villin amplification 5Ј-CAACTTC-CTAAGATCTCC-3Ј coding strand primer and 5Ј-GCAACAGTCGCTG-GACATCACAGG-3Ј noncoding strand primer were used, generating a 473-bp product; for TFIID amplification 5Ј-CCACGGACAACTGCGTT-GAT-3Ј coding strand primer and 5Ј-GGCTCATAGCTACTGAACTG-3Ј noncoding strand primer were used, generating a 220-bp product. Onefifth of the PCR product was run on an ethidium bromide-containing agarose gel.

Determination of the Transcription Start
Site-To determine the transcriptional start site of the mouse villin gene, total RNA was isolated from intestine and analyzed by primer extension assay using an oligonucleotide complementary to the mouse villin cDNA downstream of the ATG initiation codon. The efficiency of the reaction was confirmed by primer extension of the mouse intestinal fatty acid-binding protein gene (fabpi) from the same RNA preparation (27). Analysis of the fabpi extension product on a sequencing gel by comparison with a sequence ladder (Fig. 1A) revealed a strong signal band of a size of 81 bp as expected. The extension product of villin was 105 bp, indicating that the transcriptional start site (an adenosine residue subsequently designed as nucleotide ϩ1) was 57 nucleotides upstream of the translation initiation codon of the murine villin cDNA (Fig. 1B). Comparison of the genomic sequence encompassing 9 kb upstream from the ATG initiation codon with the cDNA sequence, position of splice site consensus sequences in the 9-kb genomic sequence (Fig. 1B), and determination of the transcription start site reveal that the mouse villin gene has one transcription start site that is separated from the ATG initiation codon by a 5.5-kb intronic region (Fig. 1C).
DNase I-hypersensitive Sites in the Mouse Villin Gene-To characterize the key regulatory regions involved in the specific control of villin expression, we have mapped the DNase I-hypersensitive sites (31) in the mouse villin gene (along a region extending 9 kb upstream and 4.4 kb downstream from the translation initiation codon, as represented in Fig. 2A). The chromatin form of the mouse villin gene in different tissues (intestine, kidney, liver, and spleen) was submitted to limited DNase I digestion and subsequently digested with the appropriate restriction enzymes. Using BglII digestion and a 0.5-kb probe homologous 5Ј of the 7.5-kb BglII fragment (Fig. 2B), two sets of DNase I incubation-related fragments were detected, migrating at 5.5 and 2.7 kb, and corresponding to hypersensitive sites designated as HS I (located at approximately ϩ5.5 kb downstream from the transcription start (ϩ1) site, just upstream from the ATG initiation codon) and HS II (located at approximately ϩ3 kb downstream from the (ϩ1) site), respectively. HS I was observed in nuclei isolated from intestine, kidney, and liver, whereas HS II was only present in intestinal tissue. No specific hypersensitive sites were detected in nuclei isolated from spleen. The presence and location of these hypersensitive bands were confirmed by hybridization with the 0.8-kb probe ( Fig. 2A) homologous to the 3Ј end of the 7.5-kb BglII fragment (data not shown). Using BamHI digestion and the 0.5-kb probe (Fig. 2C), five sets of DNase I-treated nucleirelated fragments were detected, migrating at 3.4, 4.3, and 4.7 kb and approximately 10 and 15 kb, corresponding to the hypersensitive sites HS II, HS III (located at Ϫ0.5 kb upstream from the (ϩ1) site), HS IV (located at Ϫ1 kb upstream from the (ϩ1) site), HS V (located at Ϫ10 kb upstream from the (ϩ1) site), and HS VI (located at Ϫ15 kb upstream from the (ϩ1) site), respectively. HS III was observed in nuclei isolated from both intestine and kidney, whereas HS IV was only present in intestinal tissue as HS II. The hypersensitive sites HS V and HS VI were only present in liver tissue (in which villin is weakly expressed) and were located far upstream from the transcription start site in regions (i) that have not been subcloned and (ii) that could belong to an adjacent gene; for these reasons, these hepatic-specific hypersensitive sites were not analyzed further. As for BglII digestion, no specific hypersensitive sites were detected in nuclei isolated from spleen. Using other independant restriction digestions (EcoRI and HindIII) and the 0.5, 0.8, and 1.25 kb probes (as represented in Fig. 2A), similar results were obtained (data not shown).
In conclusion, four major distinct DNase I-hypersensitive sites (HS I to HS IV) were shown to be present in the region extending from Ϫ1 kb to ϩ5.5 kb in respect to the transcription start site (Fig. 3A) of the mouse villin gene. These sites were detected in intestine (HS I to HS IV), kidney (HS I and HS III), and liver (HS I), tissues in which villin is expressed, but they were not found in spleen, a tissue that does not produce villin. These findings correlate with the tissue-specific control of villin gene expression and suggest that the putative critical regulatory elements lie within these regions. HS II and HS IV were only detected in intestine and are probably associated with tissue-specific transcription factor-binding sites involved in the positive control of villin gene intestinal expression.
Analysis of Promoter Activity by Transient Expression-To test the effects of the segments containing the DNase I-hypersensitive sites (Fig. 3A) on transcriptional activity and to define more precisely the element(s) controlling villin gene expression in the intestine, segments were subcloned upstream of a promoterless LacZ plasmid (Fig. 3B). The construct pA1 contained all the subcloned regions downstream from the ATG initiation codon, encompassing the four DNase I-hypersensitive sites (HS I to HS IV) described above and the 5.5-kb intronic sequence, intron 1. Plasmids pA2 and pA3 were identical to pA1 except for the presence of intestine-specific hypersensitive site HS II and intron 1, respectively. Plasmid pB1 and plasmid pC1 were similar to plasmid pA1 but lacked the regions extending from Ϫ480 bp to Ϫ3.5 kb and Ϫ100 bp to Ϫ3.5 kb according to the transcription start site, respectively. Plasmid pC2 was identical to pA2 but lacked the region extending from Ϫ100 bp to Ϫ3.5 kb. Plasmids pB3 and pC3 were identical to pB1 and pC1 except for the presence of intron 1, respectively. The plasmid pD1 was identical to pA1 except for the presence of the transcription start site and the region extending upstream from this site. The plasmid pBasic, which does not contain a promoter or enhancer, and a pControl plasmid, which possesses the SV40 promoter, were also tested in each experiment.
Transient transfections with these recombinant plasmids were performed in CaCo2 and LLCPK1 cell lines, which express villin, and in Madin-Darby canine kidney epithelial cells, in which no villin expression is detected. Transcription from the villin promoter was measured by assaying ␤-galactosidase activity in extracts made from the transfected cells, and the results were expressed as fold induction over that of the promoterless vector, pBasic (Fig. 3C). High levels of ␤-galactosidase activity in the pControl transfected cell lines (CaCo2 cells, 50-fold over that of pBasic; LLCPK1 cells, 98-fold) demonstrated the presence of efficient general transcription/translation mechanisms in these cells (data not shown). Very low levels of ␤-galactosidase activity in pD1 both transfected cells compared with pBasic transfected cells showed that the transcription start site was necessary for an efficient specific transcription of the reporter gene and that nonspecific transcription was not initiated elsewhere in the villin regulatory sequences. The construct pA1 expressed the ␤-galactosidase gene at the highest level in CaCo2 cells (8- decreased ␤-galactosidase expression in CaCo2 cells (2-fold over pBasic) to about 25% of that of pA1, demonstrating that a major element that confers intestinal activity was confined within this fragment. Similar results were obtained when the region upstream from the transcription start site (encompassing HS III and HS IV) was almost wholly deleted with or without HS II (pC1 and pC2, respectively). The deletion of the intronic region alone (pA3) or in combination with deleted sequences upstream from the transcription start site (pB3 and pC3 extend only from Ϫ480 and Ϫ100 bp, respectively) affected to a lesser extent ␤-galactosidase expression in the same intestinal cells (5.5-fold over pBasic), with a decrease to only about 65% of that of pA1, demonstrating that the regulatory elements that lay within 100 bp were sufficient to promote transcription in cultured cells. However, the level of ␤-galactosidase activity increased strongly when the plasmids pA3, pB3, and pC3 were transfected in LLCPK1 cells (10-, 44-, and 45-fold over pBasic, respectively), showing that the absence of the first intron, in combination with the lack of intestine-specific HS IV, was able to promote transcription in a kidney cell line. This would suggest that negative elements that confer repression in kidney transcription are confined in these elements.
To test specificity, the villin promoter-related constructs were transfected in Madin-Darby canine kidney cells, which do not express villin. After transfection, these cells showed only base-line levels of ␤-galactosidase activity when compared with pBasic-related activity (data not shown), demonstrating that the villin regulatory sequences were unable to promote efficient transcription in nonexpressing villin cells and that consequently the expression of the reporter gene in CaCo2 and LLCPK1 cells is specifically dependent upon these regulatory sequences. Taken together, these results from transient transfection of cultured cells demonstrate that (i) the mouse villin genomic sequence, extending from Ϫ3.5 to ϩ5.5 kb, specifically directs an efficient level expression of the ␤-galactosidase reporter gene in intestine-derived cells, (ii) this level is dramatically reduced when the intronic intestine-specific hypersensitive site HS II or the region upstream from the (ϩ1) site is deleted, (iii) lack of the entire first intron seems to partially restore the intestine-related ability to promote transcription, and (iv) lack of the entire first intron in combination with intestine-specific hypersensitive site HS IV is correlated with a strong increase of ability in promoting transcription in kidneyderived cells.
Analysis of Transgene Expression in Mice-Because the Ϫ3.5 to ϩ5.5-kb region of the mouse villin contained the enterocytelike-specific promoter/enhancer activity in transient transfection assays, we examined the ability of this region to drive intestine-specific expression of the ␤-galactosidase reporter gene in transgenic mice. Five founder animals that contained the pA1 construct as a transgene were obtained. The founder mice were analyzed for mRNA reporter gene expression in several adult tissues by reverse transcription-PCR analysis. From the same cDNA samples, products encoding ␤-galactosidase, villin, and TFIID were analyzed. The PCR assays enabled only the detection of spliced transcribed mRNA, excluding that from genomic DNA itself, by means of an exon connection strategy by combination of a 5Ј PCR primer from within the mouse villin promoter sequence just upstream of the splice donor site and the 3Ј primers from within the ␤-galactosidase gene or the villin gene. For each founder, no reporter gene expression was detected in the tissues in which villin mRNAs were not detected using the PCR assay (Fig. 4). For all founder mice, the reporter gene transcription was detected along the cephalocaudal axis of the gut (duodenum, jejunum, ileum, proximal, and distal colon) following the intestine-specific expression of the villin gene (Fig. 4). In the kidney, the transgene was only transcribed in one founder of five animals obtained (Fig.  4). TFIID mRNA was present in all samples from tissues in which the reporter gene expression could not be detected (Fig.  4), confirming the quality of RNA from these tissues.
To examine the precise cellular distribution of transgene expression within the tissues, cryostat sections of small intestine, colon, and kidney were prepared and subsequently stained for ␤-galactosidase enzyme activity. Similar results were obtained by immunofluorescence analysis of ␤-galactosidase expression (data not shown). For four of five transgenic mice, a heterogeneous pattern of expression in small intestine and colon was observed. This heterogeneity might be due to mosaicism because we examined founder animals. The expression was confined to the nucleus of the epithelial cells, as expected because the ␤-galactosidase gene contains a nuclear localization sequence signal (Fig. 5). The staining was detected by a stronger signal in the villi migrating cells when compared with the crypt cells, of both small intestine (Fig. 5A) and colon (Fig. 5C) epithelium, thus confirming that the Ϫ3.5 to ϩ5.5-kb region of the mouse villin gene is able to recapitulate precisely the cellular pattern of expression, along the crypt-villus differentiation axis, of the endogeneous villin gene (17). A continuous labeling of all cells of the crypt (Fig. 5, B and D) was observed, suggesting the expression of the transgene in the stem cells (10). It is noteworthy that the intensity of the ␤-galactosidase staining was similar to that of intestinal sections from chimeric animals which possess a ␤-galactosidase gene integrated at the villin locus by homologous recombination procedure (32), indicating that the Ϫ3.5 to ϩ5.5-kb region of the mouse villin gene was able to promote intestinal transcription as efficiently as the mouse villin gene itself. In the kidney of the founder mouse in which the transgene was detected by reverse transcription-PCR, the staining was only observed in the epithelial cells of the proximal tubules where the villin gene is expressed (data not shown). The founder animals were able to transmit the transgene to their offspring with a similar pattern of ␤-galactosidase expression (data not shown). In our attempt to direct an efficient expression of the reporter gene in the intestinal epithelium with shorter regulatory sequences, plasmids pA3, pB3, and pC3 were used to generate transgenic mice, because these constructs display efficient levels of ␤-galactosidase activity in intestine-derived CaCo2 cells. The presence of the transgene assessed by ␤-galactosidase staining procedure was observed in three of the four independent lines of pA3 transgenic mice generated. These three lines expressed the reporter gene only in the small intestine (in both the immature and differentiated epithelial cells along the crypt-villus axis), and all three lines failed to express the transgene in the other tissues tested, particularly noteworthy is the lack of expression in the colon and the kidney (data not shown).
These results demonstrate that (i) the 3.5-kb regulatory region upstream from the transcription start site of the mouse villin gene is necessary and sufficient to sustain expression strictly in small intestine of transgenic mice, (ii) the first intron of the mouse villin gene is required for colon and kidney expression in transgenic mice. Concerning the pB3 and pC3 transgenic mice, no transgene expression was observed in all tissues examined, including small intestine, colon, and kidney. Thus, the key cis-acting elements of the villin gene required for intestinal and/or kidney-related expression of transgene(s) in transgenic mice are not located only within the region encompassing Ϫ480 bp upstream from the transcription start site, as observed in the cultured epithelial cells. DISCUSSION In this report, we demonstrate that cis-acting sequences located within a 9-kb region (-3.5 to ϩ5.5 kb from the start site of transcription) of the mouse villin gene are sufficient to direct both correct tissue-specific and high expression level of the ␤-galactosidase reporter gene in transgenic mice, when compared with the endogenous gene (19). Reporter gene expression is detected in the whole intestinal tube and appropriately restricted to epithelial cells along the crypt-villus axis of both small intestine and colon. In addition, these regulatory elements can maintain a gradient of ␤-galactosidase gene expression from the crypts of Lieberkü hn to the tips of villi that precisely reproduce the gradient exhibited by the murine villin gene (17). Similarities between transgene and endogenous gene expression were also noticed as judged by a comparison with the staining intensity of ␤-galactosidase activity in intestinal sections from our transgenic mice and mice in which the reporter gene has been inserted at the natural villin locus by homologous recombination (32).
In the kidney, for only one animal of five analyzed, mouse reporter gene expression was restricted to epithelial cells of the proximal tubules recapitulating the villin expression pattern in this tissue. This suggests that transcriptional mechanisms specifying gene expression to intestine and kidney tissues are in the Ϫ3.5 to ϩ5.5-kb region of the mouse villin gene and that those related to kidney may be sensitive to positional effects. Indeed it is known that the transgene expression is dependent on the site of chromosomal integration and can be influenced by regulatory regions in the vicinity, presumably acting on chromatin conformation (33). The construct that entirely lacks the first intron of 5.5 kb but that harbors 3.5 kb 5Ј to the start site of transcription of the mouse villin gene, placed in front of the ␤-galactosidase gene, restricts the in vivo expression of the reporter gene only into the epithelial cells along the crypt-villus axis of the small intestine. The extinction of the reporter gene expression in the kidney might be due to strong positional effects, as reported above, whereas the extinction related to the colon might be due to the absence of regulatory elements of the intron 1, such as the intestine-specific DNase I-hypersensitive site HS II.
Constructs harboring only the first 480 and 100 bp 5Ј to the start site of transcription, in combination with the lack of the first intron, placed in front of the ␤-galactosidase gene, both failed to drive intestine-specific and kidney-specific expression of ␤-galactosidase, suggesting that the intestine-specific DNase I-hypersensitive site HS IV localized just upstream from the 480 bp might play an important role in promoting reporter gene expression into the epithelial cells of the small intestine. Thus, distinct and separable regulatory elements in the mouse villin gene may direct transgene expression along the cephalocaudal axis of the gut: the regulatory elements required for transgene expression in the small intestine might be localized in the 3.5-kb region (i.e. the HS IV site) upstream from the transcription start site, whereas those necessary for the colonic expression might be localized in the first intron (i.e. the HS II site).
The inability of shorter regulatory sequences of the mouse villin gene to direct correct expression of the reporter gene in the whole intestine of transgenic mice might also be explained by spatial rearrangement of chromatine structure due to the lack of the entire first intron. In fact, the results described here are reminiscent of those of the adenosine deaminase gene (34) and the aldolase B gene (35), in which elements located in the first intron are required for transgene expression in vivo, because they may contain cis-acting tissue-specific enhancer elements and/or elements involved in promoting decondensation of the chromatin structure, allowing the accessibility for transcription factors and RNA polymerase. Further investigations will be required to elucidate the precise role of the first intronic region of the mouse villin gene.
To explain the discrepancy seen in the ability of the mouse villin gene regulatory elements to promote transcription of the reporter gene in cell cultures versus transgenic animals, we may argue that the regulation of gene expression in the intestinal epithelium occurs as cells differentiate and migrate along the crypt-villus axis. This process depends on the contacts that these cells maintain with other neighboring cells on one hand and with the extracellular matrix on the other hand (36). Thus, an ex vivo system as the intestine-derived CaCo2 cell line used in our study, is limited by its weak ability to recapitulate the temporal and spatial complexities of this epithelium and emphasizes the importance of using in vivo models to define a function for specific regulatory sequences (37,38).
Previous studies carried out in transgenic mice to map transcriptional regulatory elements responsible for intestinal expression have been performed using cis-acting sequences of genes expressed in villus associated-enterocytes of small intes- tine (4, 5, 38 -40). In some of these cases, precocious activation in the crypts in combination with extended expression in the colon occurs in an inappropriate manner. Thus, to our knowledge, the 9-kb regulatory region of the mouse villin gene represents the only characterized cis-acting sequences reported today that allow the expression of a heterologous gene in small intestine and colon epithelial cells of transgenic mice reproducing with great fidelity the tissue-specific and cell-specific pattern of expression when compared with that of the endogenous gene itself. In addition, the mice lines that drive a transgenic expression exclusively restricted to the intestinal mucosa could already be studied after selection of those that will not display expression into the kidney because of the positional effects.
The ability to target genes of interest in transgenic mice following the villin-restricted pattern of expression and particularly in the crypt stem cells should lead to the development of targeted genes in animal models. Experimental mouse models reproducing several steps of human colorectal carcinogenesis (a possible genetic pathway has been proposed by Fearon and Vogelstein (41)) could for instance be obtained by efficiently targeting the associated oncogenes or mutated tumor suppressor genes to colonocytes using the villin regulatory region. Another use could lie in the establishment of new cell lines derived from the digestive tract by targeting a thermosensitive SV40 T antigen to the crypt-resident progenitors of intestinal cells, as used in other systems (42)(43)(44).