Regulation of Hoxc-8 during mouse embryonic development: identification and characterization of critical elements involved in early neural tube expression

The determination of positional identities along the anteroposterior axis represents a complex and as yet unresolved problem in pattern formation. The positional cues for the generation of spatial organization are highly conserved among metazoans (Krumlauf, 1994; Lawrence and Morata, 1994; Ruddle et al., 1994). In insects, the Homeotic-Complex (HOM-C) encodes a family of regionally expressed genes that are largely responsible for the specification of segmental identity along the anteroposterior axis (Lawrence and Morata, 1994). The mammalian genome contains four separate clusters of genes, termed Hox genes, which share a high degree of sequence, organizational and functional similarity to the Drosophila HOM-C genes (Krumlauf, 1994). Furthermore, as in Drosophila, the mammalian Hox genes exhibit a collinear relationship between their order of arrangement on the chromosome and their anterior limits of expression along the embryonic axis: the more 3′ the gene is within a cluster, the more anterior is its limit of expression (Graham et al., 1989; Duboule and Dollé, 1989). Since proper spatial patterns of Hox gene expression are necessary for normal development, understanding how Hox genes are regulated in precise spatial patterns along the anteroposterior axis will help explain an important step in morphogenesis.

Hox gene expression is regulated both spatially and temporally primarily at the level of transcription. Factors that govern the spatial expression of Hox genes may form a hierarchy that reflects developmental decisions made prior to Hox gene activation. A critical first step in dissecting this molecular hierarchy is to examine transcriptional regulation of Hox genes. This will require isolation of cis-acting DNA elements that mediate spatial regulation, characterization of critical nucleotides that affect the regulatory activity of these sequences and identification of trans-acting factors whose activity represents a previous level of hierarchy controlling pattern formation.

Employing reporter gene analysis in transgenic mice, several investigators have identified cis-regulatory regions of mouse Hox genes (see Krumlauf, 1994). Many of these studies have demonstrated that relatively large genomic regions are capable of regulating different aspects of Hox gene expression but specific elements have been identified in only a few cases (Sham et al., 1993; Marshall et al., 1994; Studer et al., 1994). Many of these analyses suggest that interactions among various cis-regulatory regions are required for Hox gene expression. However, the critical nature of such combinatorial interactions has not been directly demonstrated.

We and others have previously studied the region-specific expression and function of mouse Hoxc-8 (Awgulewitsch et al., 1986; Utset et al., 1987; Breir et al., 1988; Gaunt, 1988; Le Mouellic et al., 1988, 1992; Awgulewitsch and Jacobs, 1990; Pollock et al., 1992, 1995). The expression of Hoxc-8, like that of many other Hox genes, can be divided into an early or ‘establishment phase’ and a late or ‘maintenance phase’ (Deschamps and Wijgerde, 1993; Gaunt and Strachan, 1994). At 8.5 days pc (post coitum), Hoxc-8 expression extends from the base of the allantois to the segmental plate mesoderm and to a more anterior region in the neurectoderm. The anterior boundary of Hoxc-8 at 9.5 days pc is located in the neural tube at the level of the 9th somite and in the somitic mesoderm at the level of the 13th somite. Later in development, posterior expression of Hoxc-8 decreases while intense expression is maintained within the brachial region of the neural tube.

In previous studies of the regulation of the Hoxc-8 gene, we described a 5.1 kb cis-regulatory region that mediates expression in posterior neurectoderm and mesoderm reconstituting the early but not the late phase of Hoxc-8 expression (Bieberich et al., 1990). In the studies reported here, we have performed extensive deletion and mutational analyses of this regulatory region to delineate elements involved in the establishment of region-specific patterns of gene expression. We describe the components of a posterior early neural tube (ENT) enhancer and provide evidence that four distinct elements located within a 135 bp region interact to determine the expression pattern in the neural tube. Comparison of nucleotide sequences critical for enhancer activity reveal potential sites for interactions with families of known transcription factors including caudal, forkhead and homeodomain proteins.

Hsp68-lacZ gene was obtained from Dr J. Rossant. The BamHI fragment of phspPTlacZpA was subcloned into a modified pGem (Promega) vector termed pSafyre (L. Bogarad unpublished data) at the BamHI site (pHSF5). pHSF5 was used as a basic subcloning vector for all constructs. All restriction enzymes were from New England Biolabs. A 2.1 kb EcoRI-HindIII fragment was isolated as a HindIII fragment (HindIII site in the polylinker) from a plasmid which contained a 5.1 kb EcoRI fragment of the Hoxc-8 upstream region (see Fig. 1 for restriction map). This fragment was cloned at the HindIII site in the polylinker to generate two constructs: construct 1, which contained the nucleotide sequence in the same orientation as that of the genomic Hoxc-8 sequence with respect to its transcription site, and construct 2, which contained the nucleotide sequence in the opposite orientation. The 1.4 kb EcoRI-DraI fragment was isolated as a HindIII- DraI fragment from construct 1 and cloned at the HindIII site of pHSF5 by blunt end ligation (construct 3). Construct 1 was digested with BspEI and SmaI (in the polylinker), end filled and self-ligated to generate construct 5. Construct 6 was generated by isolating a 0.7 kb DraI-HindIII fragment from construct 1 and subcloned at the HindIII site of pHSF5 by blunt end ligation. Construct 3 was digested with BspEI and SalI, end filled and self-ligated to generate construct 7. Digestion of construct 7 with XhoI and PstI (enzymes in the polylinker) released the 335 bp fragment. This fragment was further digested with RsaI and the resulting fragments XhoI-RsaI (203 bp) and RsaI-PstI (132 bp) were cloned directionally into pHSF5, which was rendered blunt at HindIII and staggered at XhoI or PstI to generate constructs 8 and 9, respectively. Construct 10 was made by isolating a 132 bp ApaI- PstI (in the polylinker) fragment from construct 9, further digesting it with HaeIII, and subcloning the resulting HaeIII-PstI fragment directionally into pHSF5, which had been digested with HindIII, rendered blunt and digested with PstI. Similarly, the AluI-PstI fragment from the above fragment was cloned to obtain construct 11.

Remaining constructs were generated by the polymerase chain reaction (PCR). PCR was carried out in a 50 μl reaction containing 100 ng each of template DNA and two primers in a buffer containing 10 mM Tris, pH 8.4; 50 mM KCl, 1.5 mM MgCl₂, 200 μ? each of the four dNTPs and 1 unit of Taq polymerase (Boehringer-Mannheim). The reaction mix was overlaid with a drop of mineral oil and placed in a thermocycler (Hybaid, Omnigene). The conditions for amplification were 94°C, 3 minutes, 1 cycle; 94°C, 1 minute; 55°C, 1 minute; 72°C, 1 minute 15 seconds; 20 cycles; 72°C 5 minutes, 1 cycle. At the end of the reaction, 10 μl of the reaction product was analyzed by agarose gel electrophoresis to confirm the amplification of correct-sized product. The fragments were diluted 3- to 5-fold and subcloned into PCR™ vector (Invitrogen) according to the manufacturer’s instructions.

The nucleotide sequence of the 335 bp BspEI-DraI fragment is shown in Fig. 4 and the position of the nucleotides included in the constructs is shown in Fig. 3 (constructs 12-15). The alterations in the 335 bp reporter construct were introduced by an overlapping PCR strategy, using synthetic oligonucleotide primers containing appropriate changes in the nucleotide sequence. The changes introduced are indicated in Fig. 4 (constructs 16-24).

All amplified products were sequenced by the dideoxy sequencing method (Sanger et al., 1977) using sequencing reagents (Pharmacia), 1 μCi of (³⁵S) dATP (Amersham), 3 U Sequenase (USB Biochemicals) and 10 ng T7 or SP6 primers (Promega) in each reaction. These plasmids were then digested with XhoI and PstI, and the desired fragments were isolated and subcloned at similar sites in pHSF5. Plasmid DNAs were routinely isolated by an alkaline lysis method followed by ultracentrifugation on a CsCl gradient (Sambrook et al., 1989). Prior to injection, DNA fragments were excised from vector sequences by digesting with XhoI (or ApaI or SapI) and NotI in the polylinkers and isolated by ultracentrifugation on a sucrose density gradient (Sambrook et al., 1989). DNA fragments were dialyzed extensively against 10 mM Tris, pH 7.5 and 0.25 mM EDTA, and their concentration was adjusted to 1-3 μg/ml.

Transgenic mice were generated by injecting DNA into pronuclei of fertilized oocytes of inbred FVB mice (Gordon et al., 1980). Eggs that survived injection were transplanted into the oviduct of pseudopregnant CD-1 or B6/CBAF1 fosters. The founder embryos were analyzed for the expression of the transgene on 9.5 or 10.5 days pc. Alternatively, some of the embryos were allowed to go to term and transgene expression was analyzed in the F₁ embryos obtained by timed mating. For staging embryos, the day that the plug was observed was considered to be 0.5 days pc.

Southern blot analysis was carried out as described (Sambrook et al., 1989). Genomic DNA was isolated from placenta or tails of young mice, 4-5 weeks old as described (Hogan et al., 1986; Laird et al., 1991). 10 μg of genomic DNA was digested with PstI, electrophoresed on an agarose gel and transferred to nitrocellulose (Schleicher & Schuell, Inc.) or Hybond N filters (Amersham). The filters were hybridized radioactively with random-labeled 3.7 kb HindIII- BamHI fragment from pCH110 (Pharmacia). Alternatively, the probes were labeled nonradioactively with ‘Genius Nonradioactive Nucleic Acid Labeling and Detection Kit’ (Boehringer-Mannheim) and the hybridized products were visualized by following the manufacturer’s instructions.

The embryos were dissected free from the maternal tissues into cold PBS, fixed for 30 minutes in 0.25% glutaraldehyde, washed in PBS and stained for β-galactosidase activity as described previously (Bieberich et al., 1990).

Hoxc-8 proteins were detected by using the whole-mount staining method (Lumsden and Keynes, 1989). The production and characterization of anti-Hoxc-8 monoclonal antibodies will be described elsewhere (H. B. and F. H. R., unpublished data).

Total protein extracts were prepared from 9.5 dpc embryos obtained from staged matings as follows. The embryos were dissected free of maternal tissue, separated from the yolk sac, washed in cold PBS and frozen by liquid Nitrogen as quickly as possible. Frozen embryos collected from 5-7 pregnant mice were thawed in 5 ml of Buffer A (50 mM Tris, pH 7.6; 5 mM MgCl₂, 25 mM KCl, 0.2 mM EDTA, 250 mM sucrose and protease inhibitor cocktail containing 1 mM phenylmethylsulfonylfluoride; PMSF from Sigma, 100 ng/ml aprotinin, 500 ng/ml leupeptin and 1 μg/ml pepstatin from Boehringer-Mannheim). The embryos were homogenized on ice in a tight-fitting Dounce homogenizer with 20 strokes. The extract was then brought to 0.5 M NaCl and 0.5% Triton X-100 and incubated on ice for 30 minutes with occasional mixing. At the end of incubation, the extract was repeatedly passed through a 26-gauge needle to reduce the viscosity of the solution and then centrifuged at 105,000 g for 60 minutes at 4°C. The supernatant was collected and dialyzed against Buffer A for 8-12 hours with repeated changes of the buffer. After dialysis, the extract was centrifuged at 15,000 g for 20 minutes at 4°C. The supernatant was aliquoted (100 μl) and frozen in liquid Nitrogen. The amount of protein in the extract was quantitated by Lowry’s method with minor modifications (Lowry et al., 1951).

To determine protein binding to the 5′ region of the 135 bp, the following complementary oligonucleotides were synthesized: 5′AGCTTTTATGGCCCTGTTTGTCTCCCTGCTCTA3′ and 5′AGCTAGAGCAGGGAGACAAACAGGGCCATAAAA3′. To perform scanning mutations, a series of oligonucleotides, each containing three nucleotide substitutions (A to C; G to T and vice versa) were synthesized beginning from position 5. These oligonucleotides were purified by electrophoresis on a 7 M urea-15% polyacrylamide gel and isolated as described above. The complementary singlestranded oligonucleotides were then mixed in equal proportions and annealed in buffer containing 10 mM Tris, pH 7.4 and 100 mM NaCl. The reaction mixture was heated at 65°C for 15 minutes and allowed to cool gradually to room temperature. The resulting double-stranded oligonucleotides were purified on a 15% polyacrylamide gel. The oligonucleotides were then end labeled with high specific activity ³²P-dCTP and cold dATP, dGTP and dTTP with Klenow fragment. The labeled oligonucleotides were separated from free nucleotides on a spin column of Sephadex G-75.

Electrophoretic mobility shift assay was carried out by mixing different amounts of protein (10-40 μg) with approximately 1 ng labeled DNA (20,000 cts/minute) and 1 μg poly(dI).poly(dC) in a 50 μl reaction mix containing buffer 50 mM Tris, pH 7.6; 5 mM MgCl₂, 25 mM KCl, 0.2 mM EDTA, 250 mM sucrose and 6 μg/ml Bovine serum albumin (Ausubel et al., 1987). The reaction was incubated for 20 minutes on ice and then electrophoresed on a 5% polyacrylamide gel. The gels were dried and exposed to Kodak X-ray film for 12-18 hours.

Previous studies identified a 5.1 kb DNA fragment from the Hoxc-8 locus that was capable of directing a region-specific pattern of gene expression in transgenic mice (Bieberich et al., 1990). This fragment included the Hoxc-8 promoter (Awgule-witsch et al., 1990) and was shown to direct a pattern of reporter gene expression that was consistent with the early phase of Hoxc-8 expression (Bieberich et al., 1990). In this study, deletion of the distal 2.1 kb EcoRI-HindIII fragment (Fig. 1) completely abolished activity, suggesting that elements within this fragment are required for Hoxc-8 transcriptional activity (data not shown). This 2.1 kb DNA fragment was linked to a heterologous promoter, mouse heat-shock promoter, hsp68, that by itself does not confer region-specific expression on a β-galactosidase reporter gene (Kothary et al.,1989; Rossant et al., 1991). In either orientation, the 2.1 kb enhancer conferred on the hsp68 promoter an expression pattern that was very similar, though not identical, to the early phase of Hoxc-8 expression (Fig. 2A,B). Like the endogenous Hoxc-8 gene and the previously described 5.1 kb transgene, the 2.1 kb enhancer became transcriptionally active around 8.0 days pc, established appropriate anterior boundaries of expression in the neural tube and paraxial mesoderm in 8.5 days pc embryos and directed expression in most cells of the tail bud region at this stage. However, the expression pattern of the 2.1 kb enhancer-containing construct differed from the 5.1 kb transgene in that expression was stronger in the somites than in the lateral plate mesoderm and was down regulated at later stages of development (i.e., after 10.5 days pc; data not shown). Thus the 2.1 kb fragment, which is located 3 kb from the transcriptional start sites of the Hoxc-8, can function as an enhancer directing posterior early neural tube and mesoderm expression in the context of a heterologous promoter.

Fragments derived from the 2.1 kb enhancer were tested for activity in transgenic mice as a prelude to a detailed mutational analysis. Two partially overlapping fragments, a 1.4 kb EcoRI-DraI and a 1.0 kb BspEI-HindIII fragment (constructs 3 and 4 respectively; Fig. 1) were capable of driving reporter gene expression in the neural tube and mesoderm (Fig. 2C,D), whereas fragments lacking the region of overlap (constructs 5 and 6) were inactive. The two active constructs direct similar anterior bound-aries of expression in the neural tube at somite levels 12 and 11 in 9.5 days pc. embryos, respectively. Practically all cells in the tail bud region including those in the allantois showed intense staining for β-galactosidase activity. However, the level of β-galactosidase activity in the somites differed between the two constructs. Construct 3 showed strong expression of the transgene in somites in the most posterior region but weaker expression more anteriorly. In contrast, somitic expression was uniformly strong in embryos containing construct 4. These differences were verified at the cellular level in serial sections (data not shown). A comparison between the expression pattern of construct 4 and the Hoxc-8 protein distribution, shown in Fig. 2E and F, demonstrates the similar anterior boundaries of expression in the neural tube and mesoderm in 8.5 days pc embryos.

The above results suggest that the region of overlap between construct 3 and construct 4, a 335 bp BspEI-DraI fragment, contains elements that support the expression of the transgene in the posterior neural tube. The reporter gene construct containing only this 335 bp region overlap (construct 7) was shown to drive the expression in the posterior neural tube and mesoderm surrounding the caudal regions of the neural tube (Fig. 2G,H). Compared to larger constructs (constructs 3 and 4), the enhancer activity of construct 7 was less pronounced as reflected by the longer time required to stain the embryo for β-galactosidase activity, as well as by a slight temporal delay in the onset of expression (data not shown). In addition, the anterior boundary of expression with construct 7 was 1-to 2-somite levels posterior to that observed with the larger constructs. Moreover, construct 7 showed no activity in somites when assayed by either whole-mount staining or by histological analysis of tissue sections (Fig. 2H). However, some mesoderm in the tail bud was weakly positive for β-galactosidase activity.

It can be concluded that the 335 bp fragment contains essential regulatory elements for early expression in the posterior neural tube. Although sequences surrounding this fragment may contribute to enhanced neural tube expression, especially at the anterior border of Hoxc-8 expression, no independent cis-acting elements were detected outside of this region by reporter gene analysis.

Deletion analyses were performed to delineate regulatory regions within the 335 bp enhancer (BspEI-DraI fragment) that direct early neural tube expression. Assays for β-galactosidase activity of these deletion constructs were performed on founder generation embryos at 9.5 days pc because of the high level of β-galac-tosidase activity detected at this stage. Construct 8, containing the 5′ 200 bp of the BspEI-DraI fragment, did not contain specific enhancer activity (Fig. 3) and the inclusion of increasing amounts of 3′ sequence (BspEI-237, 272, 284 and 314 in constructs 12-15 respectively) did not reconstitute the neural tube expression pattern completely, suggesting the presence of additional element(s) at the 3′ terminus of the 335 bp fragment. In contrast, construct 9, containing the 3′ 135 bp of the BspEI-DraI fragment, showed β-galactosidase activity in the posterior neural tube and mesoderm in the tail bud region (Fig. 2I). Although construct 9 clearly showed region-specific expression, the expression was somewhat posterior to that driven by the 335 bp fragment.

Further deletion within the 135 bp region from the 5′ end resulted in loss of enhancer activity in the neural tube. Construct 10, containing a 98 bp HaeIII-DraI fragment, showed expression in only a few cells at the caudal neuropore (Fig. 2J). Hence a 32 bp region between the RsaI and a HaeIII sites (see Fig. 4 for nucleotide positions) contains a regulatory element essential for early enhancer activity. These results, taken together, indicate the presence of regulatory elements located at the 5′ and 3′ termini of the 135 bp fragment that are essential for neural tube expression.

To gain insight into critical nucleotides involved in the posterior neural tube enhancer activity, we next identified elements that serve as specific binding sites for proteins in the mouse embryo extracts. DNA fragments from termini of the 135 bp fragment (a 52 bp RsaI-EcoNI fragment and a 30 bp MaeI-DraI fragment; see Fig 4) bound proteins present in 9.5 dpc mouse embryo extracts in DNA-mobility shift assays (data not shown). Based on comparisons of nucleotide sequences to the binding sites for known transcription factors, each terminus was identified to contain two distinct sites, designated as A and B at the 5′ end, and C and D at the 3′ end (Fig. 4). The nucleotide sequences of sites A and D represent potential binding sites for caudal-related proteins (Dearolf et al., 1989; Margalit et al., 1993; Suh et al., 1994), site B represents a potential binding site for HNF-3/forkhead-related proteins ?Liu et al., 1991; Jackson et al., 1993; Overdier et al., 1994) and site C is a typical homeo-domain protein-binding sequence (Triesman et al., 1992).

To identify the nucleotides involved in the binding activity at the 5′ terminus, a synthetic oligonucleotide containing sequences 227-259 was tested for binding to proteins using a 9.5 dpc mouse embryonic extract in a DNA-mobility shift assay (Fig. 5 lanes 1-3; I). We further synthesized a series of oligonucleotides, each containing three nucleotide substitutions and tested them for protein binding. Mutations in the 5′TTTTATGGC3′ sequence resulted in reduced protein-binding activity (Fig. 5 lanes 1-3; II-IV), whereas mutations in the adjacent sequences (Fig. 5 lanes 1-3; V) did not affect protein binding. This sequence is identical to the protein-binding sites for the caudal-related proteins from different organisms (Dearolf et al., 1989; Margalit et al., 1993; Suh et al., 1994). Preliminary results indicate that hamster Cdx-3 protein can bind to both sites A and D of the ENT enhancer (A. Carr and C. S. S., unpublished observations).

To determine if these elements were critical for enhancer activity in vivo, three nucleotide substitutions were introduced in sites A and D individually (TTTTATGG ²³ to TTTGCGGG) in the 335 bp ₂₄ reporter construct. Nucleotide substitutions in site A (construct 16; Fig. 4) dramat-ically reduced enhancer activity. Weak expression of the transgene was noticeable in a few embryos in cells at the caudal neuropore (Fig. 6A) in a pattern similar to that observed with construct 10 (Fig. 2J) in which element A is deleted. These results indicate that site A is essential for enhancer activity. In contrast, mutating site D (construct 19) did not affect the neural tube expression of the transgene but did result in weaker expression in the mesoderm (Fig. 6D). This demonstrates that sites A and D, although identical in their core nucleotide sequences, are functionally distinct.

Individual alterations at the sequences adjacent to sites A and D (designated as B and C; Fig. 4) did not affect the neural tube expression of the enhancer. Deletion of site B (18 bp HaeIII-EcoNI sequence) in the 335 bp enhancer (construct 17) or mutations in site C (TTAATTG to TTCCTTG) in the 335 bp enhancer (construct 18) showed no observable changes in the neural tube expression of the reporter genes in transgenic embryos (Fig. 6B,C). However, mutations introduced simulta-neously at sites B, C and D did result in pronounced changes in the neural tube expression patterns as discussed below.

Pair-wise alterations at sites B, C and D were introduced in various combinations. The construct containing a deletion at site B and substitutions at site C (construct 20) led to a loss of neural tube expression except in the caudal limit (Fig. 5E), as did construct 21, which contained substitutions at sites B and C (data not shown).

When nucleotide substitutions were introduced at both site B and site D (construct 22) or at both site C and site D (construct 24), reduced levels of neural tube expression were seen (construct 22, Fig. 6F; construct 24, not shown) and the anterior extent of the neural tube expression was at least 6 somite levels more posterior than that observed in embryos carrying an unaltered 335 bp construct. The level of neural tube expression obtained with these constructs was much higher than that obtained with con-structs that contained alterations at sites B and C.

In summary, interactions at four distinct sites are critical for the expression of the transgene in the neural tube (Fig. 7). Site A is essential but not sufficient for enhancer activity. Interac-tions at sites B, C and D are required for the determination of the anterior extent of transgene expression driven by the ENT enhancer. Further, the nucleotide sequences of these sites implicate members of caudal, forkhead and homeodomain-protein families as transcription factors involved in the regu-lation of the Hoxc-8 ENT enhancer.

We have characterized an early-acting enhancer from the 5′ flanking region of the mouse Hoxc-8 gene. This enhancer, termed the early neural tube (ENT) enhancer, directs expression in the posterior neural tube. The key elements that mediate ENT tran-scriptional regulation have been mapped to a 135 bp fragment, although other elements located elsewhere may also contribute towards neural tube expression (C. S. S., unpublished observa-tions). Within this fragment, we have identified four distinct but interdependent elements designated AB and CD that occur in pairs at its termini (Fig. 6). The ENT enhancer requires combinatorial interactions between these elements and presumably specific tran-scription factors for appropriate transcriptional regulation. Element A is essential but not sufficient to direct ENT enhancer activity. At least two other elements in conjunction with A are required to reconstitute enhancer activity (i.e., ABC, ABD and ACD). Elements A and D are identical with respect to sequence but differ functionally: mutations at site D alone do not eliminate expression. In contrast, elements B and C represent dissimilar nucleotide sequences, yet mutations at these sites individually or in conjunction with mutations at site D lead to very similar reporter gene expression patterns, suggesting that these elements are functionally equivalent. These results, taken together, implicate site A as having primary importance in activating expression, whereas the other sites play a secondary role as mod-ulators. Moreover, element B in conjunction with C and D appears to influence the anterior limit of expression of the reporter in the neural tube. Thus, ENT enhancer activity depends on combinato-rial interactions among its components. Multiple cis-regulatory regions have been implicated in the regulation of Hox gene expression in other analyses performed to date. However, these studies were confined to regions of DNA significantly larger than the ENT enhancer (see Krumlauf, 1994, for references).

Characterization of critical nucleotides required for ENT enhancer activity presents us with an opportunity to identify trans-acting factors acting upstream of Hoxc-8. Such factors have been detected in mouse embryo extracts (C. S. S., unpublished observations) but their exact identity remains to be deter-mined. A comparison of the nucleotide sequence of the ENT enhancer with the binding sites for known transcription factors reveals that three of the four elements are potential sites for inter-actions with homeodomain proteins and that the fourth element is a potential binding site for forkhead-related proteins. Two classes of homeodomain proteins, caudal and Hox proteins, may interact with the nucleotide sequences of the ENT enhancer.

A comparison with the zebra stripe enhancer of the Drosophila ftz gene suggests that caudal-related proteins can interact with elements A and D. Three murine caudal-related genes have been identified to date: Cdx-1, Cdx-2 (homologous to hamster Cdx-3) and Cdx-4 (Duprey et al., 1988; Hu et al, 1993; James et al., 1994; Gamer and Wright, 1993; German et al., 1992; Suh et al., 1994). The sequences to which caudal proteins bind are similar to the ‘TTTATG’ sequence found in sites A and D of the ENT enhancer (Dearolf et al., 1989; German et al., 1992; Margalit et al., 1993; Suh et al., 1994). Significantly, these binding sites are grouped in pairs or associated with binding sites for other transcription factors (Dearolf et al., 1989; German et al., 1992; Suh et al., 1994). Mouse Cdx proteins are expressed during gastrulation, prior to Hoxc-8 gene expression and their spatial and temporal distribution makes them candidates for the regulation of the ENT enhancer (Meyer and Gruss, 1993; Gamer and Wright, 1993; F. Beck, personal communication).

The second class of proteins that may interact with the ENT enhancer includes members of the Hox gene family. Element C (TTAATTGT) contains a common TAAT core shared by most Hox-binding sites (Triesman et al., 1992). Since Hox genes are expressed in overlapping expression patterns, many Hox proteins may interact with this element of the ENT enhancer at different developmental stages. In Drosophila, several auto- and cross-regulatory interactions among homeodomain proteins have been demonstrated by biochemical and genetic studies. In mammals, binding of Hox proteins to several promoters in vitro and their transcriptional regulation in cell transfection assays have been demonstrated (for example, see Goomer et al., 1994; Guazzi et al., 1994). Circumstantial evidence suggests that the H0XD4 gene is autoregulated similarly to its Drosophila ortholog, deformed (Pöpperl and Featherstone, 1992). However, examples of in vivo regulation of enhancers by Hox proteins are few (Pöpperl et al., 1995). Although Hox proteins are expected to demonstrate exquisite in vivo specificity, it is difficult to demonstrate such specificity in vitro, as most Hox proteins bind to related sequences with very similar affinities. It has been suggested that the specificity of Hox gene interactions is provided in part by protein-protein interactions as in the case of Drosophila extradenticle and ultrabithorax proteins, which both regulate the activity of the downstream decapentaplegic gene (Chan et al., 1994; van Dijk and Murre, 1994). The ENT enhancer provides an excellent opportunity to investigate protein-protein interactions involving mammalian Hox proteins. The third type of nucleotide sequence, defined by element B, includes a binding sequence for HNF3/forkhead-related proteins (Liu et al., 1991; Jackson et al., 1993; Overdier et al., 1994). Many of these proteins are expressed early during gas-trulation (Ang et al., 1993; Sasaki and Hogan, 1993; Pierrou et al., 1994; Kaestner et al., 1994).

Thus, a number of developmentally important proteins appear to have the potential to regulate the activity of the ENT enhancer. The elucidation of these interactions will aid in the identification of networks of transcription factors necessary for Hox gene regulation. To date, two types of upstream factors have been implicated in the regulation of Hox genes. These are retinoids and the zinc finger protein, Krox 20, in the regulation of Hoxb-1 and Hoxb-2 genes, respectively (Sham et al., 1993; Marshall et al., 1994; Studer et al., 1994). On this basis, it has been argued that the factors acting upstream of mammalian Hox genes are distinct from those acting in Drosophila, reflect-ing differences in their modes of early development. However, the nature of the ENT enhancer described here suggests that some of the regulators of mammalian Hox genes may still be evolutionarily derived from Drosophila counterparts such as caudal and other homeodomain proteins.

This work was supported by a grant (GM 09966-33) from the National Institutes of Health. We wish to acknowledge the following for critical discussions on experiments and/or writing of the manu-script: Drs Michael Salbaum, Suzanne Bradshaw, Kevin Bentley, Wendy Bailey, Manuel Utset, Alexander Awgulewitsch and Trevor Williams. We also thank Dr Aruna Pawashe for advice on overlap-ping PCR and Zhiling Jiang for oligonucleotide synthesis.

The nucleotide sequence reported in the paper has previously been described (Awgulewitsch et al., 1990; GeneBank accession no. M35603).