E-cadherin intron 2 contains cis-regulatory elements essential for gene expression

E-cadherin-mediated cell-cell adhesion plays an important role in cell sorting, migration and tissue remodeling during several morphogenetic events in embryogenesis and organogenesis. E-cadherin protein has been well-studied during development and in adult tissues. Its altered expression often correlates with the generation of new cell types and tissues(Butz and Larue, 1995; Hatta et al., 1987; Huber et al., 1996a; Takeichi, 1988). Already during mouse preimplantation development, E-cadherin is expressed and essential for blastocyst formation (Larue et al., 1994; Riethmacher et al., 1995; Vestweber and Kemler, 1984), but subsequently cells of the trophectoderm and parietal endoderm gradually lose E-cadherin expression(Butz and Larue, 1995; Nose and Takeichi, 1986). During gastrulation, formation of mesoderm is achieved only if E-cadherin is properly downregulated in delaminating epiblast cells at the primitive streak(Butz and Larue, 1995; Carver et al., 2001; Huber et al., 1996a). Likewise, E-cadherin expression in the ectoderm is turned off at neurulation but remains high at the ectoderm-neurectoderm borders, where it is actively involved in neural tube closure (Detrick et al., 1990; Fujimori et al.,1990; Takeichi,1988). During skin development, the formation of hair follicles involves mesenchymal-epithelial interactions to establish follicle buds(Hardy, 1992; Hogan, 1999). In this process E-cadherin becomes downregulated and is replaced by P-cadherin(Hirai et al., 1989; Jamora et al., 2003). Conversely, E-cadherin transcription is re-initiated in cells undergoing mesenchymal-epithelial transitions during kidney organogenesis and in specific areas of the developing brain, as well as in differentiated neurons(Fannon and Colman, 1996; Shimamura et al., 1992; Shimamura and Takeichi, 1992; Vestweber et al., 1985). During these events, the expression of E-cadherin is often switched between`on' and `off' to determine the status of daughter cells.

Downregulation of E-cadherin is also a frequent event in tumorigenesis(Berx et al., 1998; Thiery, 2002), when the epithelial cell phenotype is lost during tumor progression. In many cases, the loss of E-cadherin, either by mutation within the coding sequence or by transcriptional downregulation, is a necessary step that promotes invasiveness(Berx et al., 1998; Perl et al., 1998; Thiery, 2002).

Although much information has been gathered about E-cadherin protein during development, organogenesis and tumor formation, little is known about the trancriptional regulation of E-cadherin, particularly how expression is activated and maintained in a developmentally and cell-type-specific manner. Several transcriptional repressors, all binding to the E-cadherin promoter region, have been identified that are able to downregulate the E-cadherin gene in specific contexts. The zinc-finger proteins Snail, Slug, δEF1/ZEB-1 and Sip-1/ZEB-2, and the basic helix-loop-helix transcription factors Twist and E12/E47 inhibit E-cadherin expression(Batlle et al., 2000; Cano et al., 2000; Carver et al., 2001; Comijn et al., 2001; Conacci-Sorrell et al., 2003; Grooteclaes and Frisch, 2000; Peinado et al., 2004; Perez-Moreno et al., 2001; Yang et al., 2004). These regulatory factors bind to a common DNA sequence known as the E-box motif,present three times in the E-cadherin promoter. In addition, mediators of Wnt signaling, namely β-catenin and Lef-1, downregulate E-cadherin in hair follicle bud formation (Jamora et al.,2003). Lef-1 binds to a single Lef/Tcf motif upstream of the E-boxes (Huber et al., 1996b). Besides these precisely defined cis-regulatory elements at the promoter, an enhancer element in intron 1 has been identified(Behrens et al., 1991; Bussemakers et al., 1994; Hennig et al., 1995; Hennig et al., 1996; Ringwald et al., 1991; Sorkin et al., 1993). Recently, we provided evidence that the above mentioned elements are insufficient to give E-cadherin-specific expression in transgenic mice(Stemmler et al., 2003). In addition, we identified sequences in the first third of intron 2 (15 kb), that conferred some cell-type-specific gene activation(Stemmler et al., 2003). Although promising, the use of large fragments of the E-cadherin gene (between-6 and +16 kb from the transcription start) still did not recapitulate the complete endogenous expression pattern, indicating that important regulatory elements were missing in this analysis. However, this work pointed to the possibility that important regulatory sequences may be located in intron 2 of the E-cadherin gene.

Here, we have investigated the function of intron 2 sequences in proper E-cadherin gene regulation by deleting the entire intron 2 of E-cadherin by gene targeting in ES cells. We show that these sequences are essential for gene activation in early embryonic development. During late embryogenesis,intron 2 strongly enhances transcription. Additionally, we show that intron 2 is required for maintenance of E-cadherin expression after initial transcriptional activation.

The different targeting vectors were generated by using standard techniques(Sambrook et al., 1989). For targeting vector 1 (TV1), a genomic fragment of the mouse E-cadherin gene from-0.1 kb to +11 kb relative to the transcriptional start site was combined with a promoter fragment from -1.5 kb to -0.1 kb together with a HSV-tkcassette. A betageo cassette was inserted into the ATG codon(Stemmler et al., 2003) and a loxP site was inserted at the ClaI site at +1.2 kb, at the 5′ end of intron 2 (Fig. 1). The second targeting vector (TV2) was generated based on a genomic mouse E-cadherin fragment (BS11) containing exon 3(Ringwald et al., 1991). A BstEII site 300 bp 3′ of exon 3 was used to insert a PGK-hyg^r cassette flanked by FRT sites and a single loxP site at the 3′ end of intron 2. The homologous recombination at the start codon was achieved by electroporation of 30 μg SwaI-linearized TV1 DNA into 10⁷ E14.1 ES cells(Hooper et al., 1987; Kuhn et al., 1991), which were then selected with G418 (Sigma, 250 μg/ml) and Ganciclovir (Cymeven, 2μM). A twofold enrichment of G418-resistant clones was observed upon additional selection with Ganciclovir. Resistant clones were analyzed by Southern blotting and PCR for correct homologous recombination events at the 5′ and 3′ end of the locus. One correctly recombined clone was expanded and used for a second electroporation with 30 μg XhoI linearized TV2 DNA and selection of transfectants with hygromycin (Calbiochem,200 μg/ml). An analysis similar to the first gene targeting was then carried out with resistant colonies after TV2 electroporation. Double-targeted clones were analyzed using pulse-field gel electrophoresis (PFGE) for separation of large fragments (Carle et al., 1986; Chu et al.,1986; Schwartz and Cantor,1984) and subsequent Southern blotting to identify clones with both homologous recombination events on the same chromosome. Two independent clones were injected into C57BL/6 blastocysts, and embryos were transferred into pseudopregnant NMRI females. Chimeric males, identified by their coat color, were mated to C57BL/6 females to generate an Ecad-In2floxFRT mouse strain. Crossing of Ecad-In2floxFRT mice with ACT-Flpe mice(Dymecki, 1996) led to a deletion of the hyg^r cassette (Ecad-In2flox) and, with expression of CMV-Cre (Schwenk et al., 1995), to the removal of intron 2 (Ecad-In2floxdel). Embryos were obtained from crosses of different strains to NMRI females or from crosses of CK14-Cre (Hafner et al., 2004) or CK19-Cre(Harada et al., 1999) males to Ecad-In2flox females. Detailed information about targeting vector sequences,PCR primers and Southern blot probes is available upon request.

Embryonic stages were determined by assuming that the appearance of a vaginal plug corresponds to embryonic day 0.5. Either whole-mount embryos,isolated organs, teratomas or ES cells were fixed in PBS/1% formaldehyde/0.2%glutaraldehyde/2 mM MgCl₂/5 mM EGTA/0.02% NP-40 for 5-90 minutes,washed three times with PBS/0.02% NP-40 and incubated overnight in PBS/2 mM MgCl₂/5 mM K₃Fe(CN)₆/5 mM K₄Fe(CN)₆/0.01% sodium desoxycholate/0.02% NP-40/1 mg/ml X-gal (Whiting et al., 1991). After post-fixation with 4% PFA, some specimens were embedded in paraffin,sectioned at 7 μm, and counterstained with Eosin or Hematoxylin/Eosin(Wilkinson and Green,1990).

ES cells grown on embryonic fibroblasts were trypsinized and resuspended in PBS. Of these, 10⁷ cells in a volume of 100 μl were injected peritoneally into 129/Sv mice. After 3 weeks, teratomas were isolated and stained with X-gal for β-galactosidase activity.

RNA was isolated from embryonic halves of E7.5 embryos with an RNeasy Kit(Qiagen) and from yolk sacs with RNA-Bee reagent (ams biotechnology). RNA of one or two embryos or 2 μg total RNA was used to synthesize cDNA with oligo(dT)-primer and a Superscript II Kit (Invitrogen). Amplification of betageo RNA was carried out with the primer pair 5′-TTACTGCCGCCTGTTTTGAC-3′ and 5′-TAGCCGAATAGCCTCTCCAC-3′, and that of Gapd with the primer pair 5′-ACCACAGTCCATGCCATCACT-3′ and 5′-GTCCACCACCCTGTTGCTGTA-3′ [in both cases using FastStart DNA Master^PLUS (Roche) in the LightCycler Instrument (Roche) according to the manufacturer's instructions]. Transcripts were normalized to Gapd expression. Values in arbitrary units are the mean of three separate experiments comparing Ecad-In2flox and Ecad-In2floxdel samples.

We performed an in silico comparative genomics approach of large sequence parts, including the E-cadherin locus for mouse, rat, human, chimp and dog(see Fig. S1 in the supplementary material). No significant evolutionary conservation was detected further upstream of the previously analyzed region(-6 kb of the transcription start site)(Stemmler et al., 2003). But interestingly, several blocks of sequence conservation over all five species were identified throughout the large intron 2. This suggested that additional,not yet functionally analyzed, sequences in intron 2 are required for proper E-cadherin gene function.

A scheme for the deletion of the entire intron 2 of the E-cadherin gene (45 kb genomic sequence) is depicted in Fig. 1A. Two independent homologous recombination events were used to insert loxP sites 5′ and 3′ of intron 2. Additionally, we inserted a betageo reporter gene at the start codon of E-cadherin to monitor the transcriptional activity of the targeted locus (TV1, Fig. 1A). More than 80% of ES-cell clones were homologously recombined(Fig. 1B) after electroporation of TV1. A 6.2 kb wild-type fragment and a 9 kb fragment of the mutated allele were detected with probe a in Southern blot analysis after BamHI digestion (Fig. 1B). One recombined ES-cell clone was taken for the second gene targeting. The 3′ loxP site was inserted by homologous recombination at exon 3 with targeting vector 2 (TV2, Fig. 1A, right side). Southern blot analysis showed homologous recombination at the 3′ end of the locus with a frequency of 10% (Fig. 1C). A BamHI digest probed with probe f revealed a 12 kb wild-type fragment and a 7 kb fragment of the mutated allele due to the insertion of a BamHI site at the loxP site. To identify recombination events which had occurred on the same allele, pulse-field gel electrophoresis separation and Southern blot analysis were performed. Hybridization with probes e and c(Fig. 1A,D) revealed a fragment that migrates at the predicted size corresponding to recombination in cis(clones 2, 6, 8-11, arrowhead, Fig. 1D). By contrast, in addition to the wild-type fragment of∼400 kb (arrow in Fig. 1D),a fragment of ∼300 kb with probe e(Fig. 1D, left) and of 100 kb with probe c (Fig. 1D,right) appeared in cases where the homologous recombination event occurred in trans (clones 3-5, 7, open arrow, Fig. 1D). Three ES-cell clones with both homologous recombination events in cis were used to generate transgenic mice. Neither a potential fused mRNA between betageo and E-cadherin sequences as a result of the knock-in nor a hypomorphic fusion protein was detected in heterozygous mice(data not shown). Because of the betageo insertion at the ATG codon of E-cadherin, the targeted allele should result in a null phenotype. Consistent with the null having an early lethal phenotype(Larue et al., 1994; Riethmacher et al., 1995),interbreeding of mice heterozygous for the targeted allele failed to generate any viable homozygous knock-in offspring (data not shown).

First insights into the regulatory function of sequences in intron 2 were obtained with the targeted ES cells (Ecad-In2flox), which, after transient transfection with a Cre expression vector(Gu et al., 1993), removed intron 2 (Ecad-In2floxdel), as demonstrated by PCR and Southern blot(Fig. 2A,B). X-Gal staining of Ecad-In2flox ES cells revealed β-galactosidase (β-gal) activity,albeit in a heterogeneous pattern (Fig. 2C). By contrast, no β-gal staining was detectable in Ecad-In2floxdel ES cells (Fig. 2D). Teratomas were produced in isogenic mice from Ecad-In2flox and Ecad-In2floxdel ES cells and in both cases these tumors contained the well-known typical variety of different tissues and cell types. Reporter gene activity was observed throughout teratomas derived from Ecad-In2flox cells(Fig. 2E) and was particularly strong in cysts and polarized epithelia(Fig. 2G). However, in teratomas derived from Ecad-In2floxdel cells, only partial and weakerβ-gal expression was observed (Fig. 2F), and this did not coincide with the locations of cysts(Fig. 2H). Importantly,epithelia of Ecad-In2floxdel teratomas did not stain for β-gal(Fig. 2H). These results provide strong evidence that intron 2 is necessary for the expression of E-cadherin in ES cells and in teratoma-derived differentiated epithelia. To study the differences in gene activity that are due to the function of intron 2, we compared the abundance of betageo transcripts in Ecad-In2flox versus Ecad-In2floxdel ES cells using a semi-quantitative PCR approach. Transcripts for betageo were detected in Ecad-In2flox samples, and these were much less abundant in Ecad-In2floxdel samples(Fig. 2I, upper panel). This result was verified by quantitative PCR, which showed a 95% reduction in gene activity after deletion of intron 2 (Fig. 2I, lower panel), thus confirming the pivotal role for intron 2 in activating E-cadherin gene expression.

Next, we analyzed the contribution of the intron 2 sequences to E-cadherin gene expression during development by crossing of Ecad-In2flox mice to a Cre-deleter strain. Reporter gene activity of Ecad-In2flox and Ecad-In2floxdel mice was monitored by X-gal staining on embryos of different stages. Embryos from Ecad-ATG (see Fig. S2 in the supplementary material) and Ecad-In2flox mice exhibited comparable profiles and both reporter lines reflected the endogenous E-cadherin expression pattern. Particularly at E6.5,β-gal expression was found in the extra-embryonic ectoderm in higher amounts compared with the embryonic part(Fig. 3A). β-Gal staining was increased in the embryonic part at E7.5(Fig. 3B), downregulated in the mesoderm at gastrulation, and maintained in ectoderm and endoderm(Fig. 3F), all in accordance with the known endogenous E-cadherin expression. Intense β-gal expression was observed in the definitive gut endoderm between E8.5 and E10.5, with increasing expression in the surface ectoderm(Fig. 3C-E,G). From E8.5 onwards, expression in the yolk sac was detected, and this increased until E10.5 (Fig. 3E). Importantly,β-gal expression was not found in embryos carrying the Ecad-In2floxdel locus prior to E9.5-10.0 (Fig. 3H-K). In particular, cells or tissues positive for β-gal expression from the Ecad-In2flox allele were all negative when intron 2 was absent, e.g. the extra-embryonic ectoderm at E6.5 (compare Fig. 3A and H), ectoderm and endoderm at E7.5 (compare Fig. 3B,F with 3I,M, respectively), or definitive gut endoderm at E8.5 to E10.5 (compare Fig. 3C,D,E,G with J,K,L,N,respectively). Generally, no β-gal expression of the Ecad-In2floxdel locus was seen in most high-level E-cadherin expression domains such as the lens. Exceptions to this rule are weak activities at the apical ectodermal ridge (AER) of the forelimb buds at E10.5(Fig. 3L) and between the first and second branchial arches around E11.0 (data not shown). Interestingly, in extra-embryonic cells of the yolk sac, β-gal activity was found at comparable levels in Ecad-In2flox and Ecad-In2floxdel embryos at E10.5. Differences in gene activity between the two alleles were examined by semi-quantitative RT-PCR from embryonic cups of E7.5 embryos and revealed reduced mRNA levels after deletion of intron 2(Fig. 3O, upper panel). Additional analysis by real-time PCR showed a 85% reduction in transcript abundancy in Ecad-In2floxdel embryos at E7.5(Fig. 3O, lower panel). These data demonstrate that during early embryogenesis the cis-regulatory elements in intron 2 are absolutely required for gene expression from the E-cadherin locus and that the promoter alone is insufficient to drive expression.

The results of the expression analysis in early embryogenesis up to E10.5 support a pivotal role for intron 2 in establishing high-level gene activity of the E-cadherin locus. Next, we analyzed whether these cis-regulatory elements have a similar important function later in development and in organogenesis. Endogenous background β-galactosidase enzymatic activity was detectable at low levels from E14.5 onwards, but this was clearly distinguishable from reporter gene-specific expression (data not shown). At E11.5, β-gal was present at high levels in the surface ectoderm of Ecad-In2flox embryos (Fig. 4A). This expression was almost completely absent in corresponding Ecad-In2floxdel embryos (Fig. 4F), with only weak β-gal expression in the facial region, mandibulary and maxillary components of branchial arches, and AER. These differences were maintained at E12.5, when additional expression domains appeared in the follicles of vibrissae in Ecad-In2flox embryos (Fig. 4B) which were very weak in Ecad-In2floxdel embryos(Fig. 4G). During skin development between E14.5 to E16.5, the differences in β-gal activity between Ecad-In2flox and Ecad-In2floxdel became less pronounced, but were still obvious after only a brief incubation (45 minutes) in X-gal solution(compare Fig. 4C with 4H). Similarly, epithelia of the inner organs of E16.5 Ecad-In2floxdel embryos showed residual β-gal expression that was much weaker than that in Ecad-In2flox embryos (compare Fig. 4D,E with 4I,J, respectively).

When E11.5 Ecad-In2flox and Ecad-In2floxdel embryos were sectioned, high levels of β-gal expression were detected in the lens and ectoderm surrounding the eye of Ecad-In2flox embryos, whereas expression in this area was absent in Ecad-In2floxdel embryos (Fig. 4K,R). Likewise and in contrast to Ecad-In2flox embryos(Fig. 4L-Q), β-gal expression was not found in the surface ectoderm of the back(Fig. 4S), nasal cavity(Fig. 4T), stomach(Fig. 4U), gut(Fig. 4W) or metanephros(Fig. 4X) of Ecad-In2floxdel embryos. Only faint β-gal expression was observed in the pancreas primordium of Ecad-In2floxdel embryos (Fig. 4V). Collectively, these results suggest that the E-cadherin locus can be activated in later stages of development in a tissue-specific manner,even without the cis-regulatory elements of intron 2, but to a very reduced extent. During this later phase of development, sequences of intron 2 strongly enhance transcription of the E-cadherin reporter gene.

The results described above revealed that the presence of intron 2 had a more global enhancing effect on activation of E-cadherin transcription,particularly in later stages of development. During this analysis it became apparent that the β-gal expression in the yolk sac was independent of intron 2 sequences. Whereas yolk sacs of wild-type embryos do not show endogenous β-galactosidase expression at E10.5(Fig. 5A) and only faint staining was observed at E12.5 (Fig. 5C), the yolk sacs of Ecad-In2flox and Ecad-In2floxdel embryos showed high-level reporter gene-derived β-gal expression(Fig. 5B,D). Remarkably,β-gal expression was equally high in the yolk sacs of both genotypes,although a clear difference was observed between the respective embryos(Fig. 5B,D). Semi-quantitative and real-time PCR corroborated the X-gal staining data showing intron 2-independent expression of β-gal in yolk sacs at E10.5 and E16.5(Fig. 5E).

In contrast to the situation in yolk sac tissue, two different E-cadherin expression domains were identified where reporter gene expression was never detected in Ecad-In2floxdel even in late embryonic stages. Whereas intense blue X-gal staining was observed in the lenses of Ecad-In2flox embryos at E10.5 (Fig. 6A), E12.5(Fig. 6C) and E14.5(Fig. 6E), expression was absent in the lenses of Ecad-In2floxdel embryos(Fig. 6B,D,F). Similarly, in contrast to Ecad-In2flox (Fig. 6G,I, arrow), no β-gal expression was observed in salivary glands of Ecad-In2floxdel embryos at E16.5(Fig. 6H,M, arrow), whereas expression in skin, thyroid glands (Fig. 6H,N, open arrowhead) and meninges(Fig. 6H,P, arrowhead) is still detected because of weak activity of the reporter gene during later embryonic development. These results show different requirements of intron 2 sequences for E-cadherin-specific β-gal expression in different organs.

To investigate whether intron 2 is required also for maintenance of transcription after the initial activation of the locus, we deleted these DNA sequences conditionally during later development, after epithelia had already formed and E-cadherin expression had been initiated. The deletion of intron 2 from established epithelia was performed with two different transgenic Cre-recombinase-expressing mouse strains: CK14-Cre and CK19-Cre (Hafner et al.,2004; Harada et al.,1999). In CK14-Cre mice, Cre-expression is controlled by the cytokeratin 14 promoter, which drives expression in the developing skin (Hafner et al.,2004; Wang et al.,1997). In CK19-Cre mice, Cre-expression is driven by the cytokeratin 19 locus (knock-in) in the trophectoderm, and, from E8.0 onwards, in the notochord, definitive gut endoderm and endoderm-derived epithelia. At later stages, CK19-Cre is also expressed in the surface ectoderm (Harada et al., 1999; Tamai et al., 2000).

Using CK14-Cre to recombine the Ecad-In2flox locus, no difference in β-gal expression between the Ecad-In2flox and Ecad-In2flox/CK14-Cre was detected before E12.5 (data not shown). At E12.5, a slight reduction in β-gal expression was observed in the surface ectoderm of Ecad-In2flox embryos carrying the CK14-Cre allele(Fig. 7A, right, +/Δ)when compared with CK14-Cre negative embryos(Fig. 7A, left, +/flox). This difference became more evident at E13.5 and E14.5(Fig. 7B,C). Interestingly,β-gal expression persisted in the lens and the gut loops of Ecad-In2flox/CK14-Cre embryos (compare left/right Fig. 7B), because the CK14-Cre is not expressed in these tissues(Hafner et al., 2004; Wang et al., 1997). At E16.5,intense β-gal expression was visible in the skin of control embryos(Fig. 7D, left), but only faint staining was observed in the skin of Ecad-In2flox/CK14-Cre embryos(Fig. 7D, right; compare with Fig. 4D,I).

We obtained similar results when using CK19-Cre to ablate intron 2; at E9.5 and earlier, no difference in reporter gene activity was observed(Fig. 7E), but at E10.5, a significant reduction in β-gal expression level was observable in Ecad-In2flox/CK19-Cre embryos. Staining of the endoderm and also in the ectoderm was reduced compared with control embryos(Fig. 7F). This reduction was more obvious when examining sections of the stained embryos; β-gal expression in the gut tube was significantly reduced after CK19-Creablation of intron 2 (Fig. 7G,H). In the pharynx region, β-gal expression was mosaic following deletion of intron 2 sequences, presumably owing to incomplete deletion (Fig. 7I,J). These results indicate that intron 2 is required for maintaining E-cadherin gene activity in the gut epithelium and the skin, in addition to its role in initiation of transcription.

To monitor gene activity of the E-cadherin locus, we used the enzyme activity derived from the E-cadherin-betageo knock-in allele. In order to validate this approach, it was important to show that E-cadherin expression and β-gal activity coincide in a spatiotemporal manner. Both the Ecad-ATG (see Fig. S2 in the supplementary material) and the Ecad-In2flox knock-in alleles faithfully recapitulated all E-cadherin expression domains,and we did not observe any ectopic expression of the reporter gene. β-Gal activity was present as soon as zygotic E-cadherin expression is detected in four-cell stage embryos and was downregulated during gastrulation when mesodermal cells are formed. Thus, all changes in E-cadherin transcriptional activity are correctly reflected by β-gal activity.

The position of cis-regulatory elements on genomic DNA sequences can be indicated by the presence of DNase-I-hypersensitive sites (DHSs). DHSs arise from nucleosome-free chromatin that is highly accessible to DNaseI and result from bound transcription factors. The occurrence of DHSs and the presence of cis-regulatory elements correlate in other genes(Harju et al., 2002; Kintscher et al., 2004; Lefevre et al., 2001; Murakami et al., 2004). At the E-cadherin locus, only one DHS is found upstream of the transcription start site at position -0.1 kb. The absence of additional DHSs further upstream of the E-cadherin promoter region and the accumulated occurrence of DHSs in the 5′ part of intron 2 in E-cadherin-expressing cells hint at cis-regulatory elements in intron 2(Stemmler et al., 2003). A high degree of sequence conservation in mouse, rat, human, chimp and dog around these DNA elements in the part of intron 2 that has been analyzed for DHSs (-15 to +18 kb) further supports this notion (see Fig. S1 in the supplementary material). Because additional areas of significant sequence conservation in intron 2 outside of the DHS-mapped region were found, this suggested the existence of other regulatory elements spread over the entire intron 2.

In our previous transgenic reporter gene approach, we had demonstrated that a -6 to +0.1 kb promoter fragment is insufficient to drive E-cadherin expression. The first 15 kb of intron 2 sequences were beneficial towards properly regulating an E-cadherin transgene(Stemmler et al., 2003). This work also demonstrated that sequences required for E-cadherin-specific expression in the endoderm are found between +0.1 and +11 kb of the E-cadherin gene, a general enhancer between +11 and +16 kb, and a brain-specific enhancer between -6 and -1.5 kb. Nevertheless, it became evident that not all regulatory sequences have been covered by this analysis.

Nonetheless, encouraged by these findings and the fact that the entire intron 2 contained conserved sequences (see Fig. S1 in the supplementary material), we analyzed the function of these sequences in vivo by ablating the entire intron 2 using gene targeting. We were able to show that, if these sequences are deleted, E-cadherin expression is completely lost during early embryogenesis. Only during later embryonic development can the locus be activated without intron 2, but with significantly reduced expression levels. In addition, our analyses revealed even more complex regulatory functions of intron 2. In general, in expression domains that are affected by the absence of intron 2, these sequences are required for both activation of the locus and maintenance of expression. We found that, in the lens and the salivary glands,expression is absolutely controlled by cis-regulatory elements of intron 2,whereas, in the yolk sac, expression can be activated regardless of the presence of these sequences.

Based on our previous findings in transgenic mice(Stemmler et al., 2003) and the data presented here, we suggest the following model of regulating E-cadherin gene activity (Fig. 8). Whereas E-boxes at the promoter contribute to downregulating the locus (small red boxes, Fig. 8), E-cadherin gene activation is initiated and maintained due to intron 2 sequences. Importantly, in Ecad-In2floxdel embryos the endoderm-specific expression of the E-cadherin reporter gene was lost,probably owing to the lack of sequences between +1.2 and +11 kb (endoderm, Fig. 8). Entire loss ofβ-gal expression in Ecad-In2floxdel embryos can be partially ascribed to the general enhancer between +11 and +16 kb (enh., Fig. 8). However,ectoderm-specific expression is not at all detectable until E11.5 in Ecad-In2floxdel embryos nor was it consistently observed in the transgenic analysis (Stemmler et al.,2003). This indicates that additional, so far undescribed cis-regulatory elements in intron 2 are present between +18 and +47 kb to drive expression in the ectoderm (indicated by `tse' in Fig. 8). E-cadherin-specific reporter gene expression in the brain due to the function of cis-regulatory elements between -6 and -1.5 kb (brain, Fig. 8) needs to be restricted to the E-cadherin expression domain by an as yet unknown brain-specific silencer (sil., Fig. 8),because we observed additional ectopic β-gal activity in the brain of transgenic embryos (Stemmler et al.,2003). Because this was not the case in Ecad-In2floxdel embryos,we conclude that the postulated brain-specific silencer must be located outside of intron 2.

A common mechanism of gene regulation of classical cadherins involving cis-regulatory elements in proximity to the transcription start site is suggested by the following observations. The genomic structure of classical cadherins exhibits a high degree of conservation between different species, as well as between different members of the cadherin family. They share a similar large second intron, and the promoter regions, e.g. those of E- and P-cadherin, have conserved DNA motifs(Faraldo and Cano, 1993). Additionally, it has been shown that the 5′-proximal promoters of other cadherins are insufficient to faithfully reflect endogenous expression. However, intron 2 is involved in proper gene function, for example of the chicken N-cadherin or the LCAM genes (Li et al., 1997; Sorkin et al.,1993). In addition to the local and promoter-proximal cis-regulatory elements provided by intron 2, correct expression of each member in the cadherin cluster may be achieved by the function of a higher order control element over a greater distance. This is supported by the lack of significant sequence similarities in different species between the 3′end of the P-cadherin gene and -6 kb of the E-cadherin gene. Because of this,the presence of additional and so far unconsidered cis-regulatory elements 5′ of the promoter is unlikely. A locus or general control region (LCR, Fig. 8) might exist at the cadherin cluster for proper expression of each member of the cluster, similar to the regulation of the Hoxd cluster or of Mrf4 and Myf5 (Fomin et al.,2004; Spitz et al.,2003). For the correct transcriptional control of the E-cadherin locus, the gene is then linked to this element via the proximal elements of intron 2 by factors that interact with the complex formed at the LCR. A similar mechanism can be postulated for classical cadherins outside of this cluster.

We observed that, despite the lack of intron 2, the E-cadherin locus was activated in many cell types of epithelial origin during late embryogenesis after E10.5. This suggests that the E-cadherin locus can be activated by two independent mechanisms. One mechanism acts during early embryogenesis and requires intron 2 for the onset of expression, and the second one functions at later stages. This second mechanism initiates E-cadherin expression independently of intron 2, although for high-level expression the support of the intron 2 enhancer elements is still required. The onset of the second wave of expression becomes apparent around E12.5 in the surface ectoderm(coinciding with the differentiation of the surface ectoderm and ongoing skin development) and in the gut endoderm. Presumably, this second, alternative activation mechanism is regulated by a common subset of transcription factors active in the specialized epithelia and might be achieved at the promoter or the intron 1 enhancer (Fig. 8).

Even more complexity of E-cadherin gene regulation emerged from the analysis of expression in the yolk sac, lens and salivary glands. The initiation of high-level reporter gene expression in the yolk sac is achieved independently of cis-regulatory elements of intron 2 and could reflect a gene-regulation mechanism specific to extra-embryonic tissues. By contrast,E-cadherin expression in the lens and the salivary glands is absolutely dependent on intron 2. Surprisingly, E-cadherin expression differs in tissues that originate from similar germ-layers. The lens develops from the lens placode, which is derived from surface ectoderm from E9.5 onwards. Whereas E-cadherin reporter gene expression is initiated by the second wave of expression in Ecad-In2floxdel embryos in the surface ectoderm of later stage embryos, no gene activation was found in the lens. Similarly, in epithelia of salivary glands of Ecad-In2floxdel embryos β-gal was never expressed,although they share the same germ-layer origin with epithelia of other inner organs. The postulated factors that are able to initiate E-cadherin transcription in later embryogenesis without intron 2 do not seem to be present in epithelia of salivary glands or in the lens. To explain the intron 2-dependent and independent E-cadherin expression, we propose that different tissue-specific enhancers probably exist that mediate E-cadherin expression in the yolk sac or in the lens and the salivary glands. This difference probably coincides with the different functions of specialized epithelia.

The data presented here reveal and emphasize the pivotal role of intron 2 in E-cadherin gene regulation during embryonic development. The importance of intron 2 sequences in gene regulation may also have an impact on tumorigenesis. The invasive property of cancer cells is often linked to loss of E-cadherin expression, in several cases owing to transcriptional downregulation (Berx et al.,1998). Accordingly, dysregulated expression of E-cadherin may be linked to mutations in intron 2 in cancer cells in which no mutation in the promoter or the coding sequence and no activation of a transcriptional repressor could be described. In some tumor cell lines, CpG-hypermethylation of the E-cadherin gene was discovered, but no mutation was found that might be responsible for this epigenic inactivation of the locus(Berx et al., 1998; Yoshiura et al., 1995). The mutations that are responsible for E-cadherin downregulation and subsequent CpG-hypermethylation may be located in intron 2. The identification of intron 2 mutations would underline the role of intron 2 in gene regulation in tumorigenesis. To be able to assess the impact of such mutations, a more precise description of the location and architecture of regulatory elements in intron 2 is required. Further gene targeting or transgenic mouse studies will concentrate on locating single tissue-specific cis-regulatory elements. An integrated in silico search for transcription factor binding sites can be used to determine which transcription factors bind to the putative regulatory sequences of intron 2. Together, these approaches will lead to better understanding of the complex interplay of multiple regulatory regions dispersed throughout large parts of the E-cadherin locus.

We thank T. Krieg and C. Niessen for CK14-Cre mice, and M. Taketo for CK19-Cre mice. We are grateful to M. Leitges for providing E14.1 ES cells and sharing expert knowledge in gene targeting, and to B. Kanzler, E. Huber, L. Morawiec, N. Klemm and Y. Joos for support in the generation of transgenic mice. We thank K. Bruser and K. Hansen for excellent technical assistance, V. Taylor for helpful discussions, and R. Cassada for critically reading the manuscript. This work was supported by the GIF Research grant No. I-747-147.2/2002.