Pax2 and homeodomain proteins cooperatively regulate a 435 bp enhancer of the mouse Pax5 gene at the midbrain-hindbrain boundary

The midbrain and cerebellum develop from an organizing center which is formed at the midbrain-hindbrain boundary (mhb; also known as isthmus) of the vertebrate embryo (Wassef and Joyner, 1997). Genetic studies in mouse and zebrafish revealed that the development and/or activity of this isthmic organizer critically depends on the mhb-specific expression of secreted proteins (Wnt1, Fgf8) and transcription factors (Otx2, Gbx2, En1, En2, Pax2, Pax5). Elucidation of the genetic hierarchy of these regulators is, however, hampered by the fact that targeted inactivation of many of these genes results in a similar phenotype, i.e. in loss of the mhb organizer (reviewed by Wassef and Joyner, 1997). Hence, all these factors appear to participate in a feed-back regulatory network controlling the function of the mhb organizer.

Pax2, Pax5 and Pax8 constitute a distinct class of Pax transcription factors which are all expressed in temporally and spatially overlapping patterns at the mhb of the vertebrate embryo (Nornes et al., 1990; Plachov et al., 1990; Adams et al., 1992; Pfeffer et al., 1998). Pax2 expression is initiated already during gastrulation in the prospective mhb region of the mouse embryo (Rowitch and McMahon, 1995). Pax5 transcription is next activated at the 3-to 5-somite stage in the mhb region (Urbánek et al., 1994; Rowitch and McMahon, 1995) followed by Pax8 expression at approx. 9 somites (Pfeffer et al., 1998). In agreement with this temporal onset of gene expression, a severe, mild or no phenotype was observed in the midbrain and cerebellum of mice lacking Pax2 (Favor et al., 1996), Pax5 (Urbánek et al., 1994) or Pax8 (Mansouri et al., 1998), respectively. The effect of the Pax2 mutation is, however, strongly influenced by the genetic background of the mouse strain analyzed, ranging from deletion of the posterior midbrain and cerebellum (Favor et al., 1996) or exencephaly (Torres et al., 1996) to almost normal development of these brain structures (Schwarz et al., 1997). In contrast, deletion of the midbrain and cerebellum was consistently observed in Pax2, Pax5 double-mutant embryos, thus demonstrating dosage-dependent cooperation of these transcription factors in the development of the mhb region (Urbánek et al., 1997; Schwarz et al., 1997).

The zebrafish genome contains four members of the Pax2/5/8 family due to an additional duplication of the Pax2 gene. All of these zebrafish genes are expressed at the mhb in similar patterns to their mouse orthologues (Pfeffer et al., 1998). Surprisingly however, the no isthmus (noi) mutation of the Pax2.1 gene results in consistent deletion of the entire mhb region (Brand et al., 1996), indicating that other members of the Pax2/5/8 family are unable to compensate for the loss of Pax2.1 function in the fish embryo. In agreement with this finding, the expression of Pax5 and Pax8 is never initiated at the mhb of noi mutant embryos (Pfeffer et al., 1998). Hence, the Pax2.1 protein is involved, either directly or indirectly, in the regulation of Pax5 and Pax8, thus revealing a genetic hierarchy of the Pax2/5/8 genes in mhb development.

Here we report the characterization of a 435 bp minimal enhancer of the mouse Pax5 gene which directs lacZ reporter gene expression in the correct temporal and spatial pattern at the mhb of transgenic mouse embryos. This minimal enhancer contains functional binding sites for homeodomain proteins and members of the Pax2/5/8 family. Double mutation of both binding sites abolished enhancer activity, whereas single mutation of either site affected transgene expression most severely in the dorsal mhb region. Similarly, expression of the endogenous Pax5 gene was initiated only near the midline in Pax2 mutant embryos, but failed to be expressed in the lateral neural plate which, upon neural tube closure, becomes the dorsal mhb region. Hence, the 435 bp enhancer of Pax5 is a direct target of Pax2 and requires Pax2 function for correct activation at the mhb of the mouse embryo, indicating that the brain-specific regulation of Pax5 has been conserved throughout vertebrate evolution.

pTrap: the 3.6 kb SmaI fragment of pBGZ40, containing the minimal β-globin promoter, lacZ gene and SV40 poly(A) signal (Yee and Rigby, 1993), was cloned into the SmaI site of pPolyIII-I to generate pTrap. This vector has unique SalI, MscI, HindIII, XbaI, KpnI and SphI sites upstream of the rabbit β-globin TATA-box. 25iz: a 19 kb NaeI and 3.6 kb NaeI-XmaI fragment of mPax5 cosmid 14 were inserted into pSP65 together with a 3.9 kb XmaI-XhoI fragment of pGNA-Pax5 containing exon 2 fused in frame to nuclear localization, lacZ and SV40 polyadenylation sequences (Urbánek et al., 1994). In1z: a 5.9 kb fragment was PCR-amplified from cosmid 14 with primers 5′-AAGATCGATGTAGGAACACCGCGAAACGGGATTC-AT-3′ and 5′-AAGCGGCCGCGAGCGCAGGGCCCTCCGTCTCA-3′ followed by insertion into pTrap. In2z: a 5 kb fragment was PCR-amplified from cosmid 14 with primers 5′-ATAGCGGCCGCA-TAGTAGGCGGCGAGCTCGCTTC-3′ and 5′-GAGACTAGTGAT-TCACTCCTCCATGTCCTGAAACAG-3′ and cloned into pBGZ40. 14z-1.6z: these constructs were generated by inserting the relevant fragments (Fig. 1) into pBGZ40. 1.2z-0.39z: the following fragments were cloned into pTrap. 1.2z contains a 1.2 kb fragment amplified from plasmid 3z with primers 5′-CCACTCGAGTTTCTTTCTCCA-GATGCA-3′ (a) and 5′-CCAGTCGACTTTGGTATCTCATTTCTC-AT-3′ (b). 0.6z: construct 1.2z was cut with SphI and religated. 0.5z contains the BsaI-BanII fragment of plasmid 1.2z. 0.43z contains a PCR fragment generated with 5′-TCTCGTCGACAAAAACCAGC-TCCAAGC-3′ and oligonucleotide 1.2z-b. 0.39z contains the KpnI-PacI fragment of plasmid 1.2z which was resected with T4 DNA polymerase prior to cloning.

Transgene 0.6z was mutated by using the QuikChange kit (Stratagene) and 33-42 nucleotide long primers containing the mutations (Figs 5A,6A) in their center. The mutated SalI-SphI fragments were recloned into pTrap and verified by DNA sequencing. Construct 0.6z-Δwas generated by removing the ends of the NsiI site by T4 DNA polymerase treatment.

C57BL/6×CBA F₁ mice were used for the generation of transgenic mice which were identified by PCR analysis with the lacZ primers 5′-ATACTGTCGTCGTCCCCTCAAACTG-3′ and 5′-TTCAACCACC-GCACGATAGAGATTC-3′. Pax5^+/lacZ and Pax5^+/Δ mice were maintained on the C57BL/6×129/Sv background (Urbánek et al., 1994; M. Horcher and M. B., unpublished data). Pax2−/− mice were generated by in-frame lacZ insertion into Pax2 exon 2 (M. B., unpublished data).

Embryos were stained at 30°C for 1-15 hours in 1 mg/ml X-gal (5-bromo-4-chloro-3-indolyl-β-D-galactoside), postfixed in 4% formaldehyde at 20˚C for 1 hour, dehydrated in methanol and then isopropanol for 30 minutes each and cleared in tetrahydronaphthalene prior to photography. Whole-mount in situ hybridization with single-stranded, digoxigenin-UTP-labeled lacZ or Pax5 probes (Adams et al., 1992) was performed as described by Pfeffer et al. (1998).

Fertilized chicken eggs were incubated for 2-3 days at 39°C in a humidified incubator. All subsequent steps were carried out on ice. Mhb tissue was dissected by cutting within the posterior third of the midbrain vesicle and near the rhombomere 1/2 constriction. Forebrain tissue was obtained by cutting anterior to the midbrain vesicle. The dissected tissue of 100-200 embryos was resuspended in 2.3 volumes of a buffer containing 0.25 M sucrose, 10 mM Hepes pH 7.6, 25 mM KCl, 0.15 mM spermine, 0.5 mM spermidine, 1 mM EDTA, 10% glycerol, 1 mM DTT, 0.1 mM Pefabloc, 5 μg/ml aprotinin, pepstatinA, leupeptin, 2 μg/ml antipain, chymostatin and 2 mM benzamidine hydrochloride, in an Eppendorf tube. This mixture was homogenized by hand, using a pellet pestle (Sigma; Z35994-7). NUN buffer of a 5× solution (5 M urea, 1.5 M NaCl, 5% NP40, 125 mM Hepes pH 7.6, 5 mM DTT) was added to a final 1× concentration. This mixture was vortexed for 5 seconds, incubated on ice for 15 minutes and microfuged at full speed for 10 minutes. Glycerol was added to 10%, and the extracts were flash-frozen in liquid nitrogen. For DNase I protection analyses, the extracts were dialysed for 1.5 h in Tube-O-DIALYZER™ tubes (Novus Molecular Inc.) against 2× 500 ml of 25 mM Hepes pH 7.6, 40 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 10% glycerol, 1 mM DTT, 0.1 mM Pefabloc.

Probe A was subcloned as a blunt-ended SfuI-SalI fragment from plasmid 0.6z into the SmaI site of pBluescript. The EcoRV-BamHI insert of this plasmid was 3′ end-labeled, and 1 μl of this probe (approx. 30,000 cts/minute, 0.1 pmol) was incubated with protein extracts followed by DNase I digestion and electrophoretic analysis as described by Barberis et al. (1987).

The mouse Pax2b, Pax3, Pax5, Pax6, Oct1, Otx2, Gbx2, En2 and Xenopus HoxD1 cDNAs were cloned into the expression vector pKW2T, and proteins were synthesized by in vitro transcription-translation. These proteins (0.02-2 μl) or tissue extracts (approx. 0.1 μl) were used for EMSA with end-labeled probes as described by Barberis et al. (1987). The oligonucleotides shown in Figs 5A and 6A were used as competitor DNA at a 100-fold molar excess. The following oligonucleotides were furthermore used: HD, 5′-AGCTCCAAATTTAATTGAAGAGTG-3′; Sp1, 5′-AATTCGATC-GGGGCGGGGCGAGCG-3′; Oct (H2A octamer) 5′-GTCTTTT-GTGCAGCTTATGCAAATGAGGGTAGG-3′; Pit (Pit1-binding site of rat growth hormone gene) 5′-GTCTTTGTGCAGGCCATGAATA-AATGATAGGTG-3′.

The 5′ sequences of the mouse and human Pax5 genes were submitted to GenBank (AF222993, AF222994).

The mouse Pax5 gene is transcribed from two independent promoters which are separated by 6.5 kb of intervening sequences and give rise to two distinct mRNAs by splicing of alternative 5′ exons (1A or 1B) to the common coding sequences of exon 2-10 (Busslinger et al., 1996) (Fig. 1A). The distal promoter serving exon 1A is predominantly active during B-lymphopoiesis, whereas the proximal promoter directs Pax5 expression to the embryonic mhb region in addition to the B-lymphoid lineage (Busslinger et al., 1996). Due to the presence of two separate promoters in the Pax5 locus, we initiated our search for cis-acting regulatory elements by fusing 25 kb of the Pax5 5′ region in frame in exon 2 to a lacZ reporter gene (Fig. 1). Mice carrying this transgene (25iz) were generated by pronuclear DNA injection and then analysed for lacZ expression by X-gal staining of transgenic embryos (Fig. 2E-H). Whole-mount in situ hybridization of transgenic embryos was used to compare Pax5 and lacZ expression, which validated the lacZ reporter assay for monitoring Pax5 expression (Fig. 2O,P). Hence, heterozygous Pax5^+/lacZ embryos, which contained an identical in-frame lacZ insertion in the targeted Pax5 locus (Urbánek et al., 1994), were stained in parallel to reveal the endogenous Pax5 expression pattern (Fig. 2A-D). These analyses unequivocally demonstrated that the transgene 25iz is correctly initiated in the mhb region of 8.5-day embryos and then continues to faithfully reflect the endogenous Pax5 expression pattern at the mhb until day 11.5 (Fig. 2A-H). Hence, 25 kb from the Pax5 5′ region contain all necessary information to direct early and late expression of Pax5 at the mhb.

Next, the 25 kb region was systematically resected to determine the number and location of mhb-specific regulatory element(s). For this purpose, we used a transgenic vector consisting of a minimal β-globin promoter linked to a lacZ gene which shows little ectopic expression on its own (Yee and Rigby, 1993). The different DNA fragments from the Pax5 5′?region were inserted upstream of the β-globin TATA-box of this vector, and their activity was directly tested in injected founder (G₀) embryos at day 11.5. The transgene 14z containing the 5′ flanking region of Pax5 was correctly expressed at the mhb (Fig. 1B), whereas intron 1 and 2 sequences were inactive in this assay (Fig. 1A). By analyzing a series of 11 overlapping fragments, the mhb enhancer could be mapped within the 5′ flanking sequences to a minimal region of 435 base pairs (construct 0.43z) which was located?5.6 kb upstream of exon 1A (Fig. 1A). Importantly, this minimal enhancer gave rise to the same temporal and spatial expression pattern at the mhb (Fig. 2I-N) as the endogenous Pax5 locus (Fig. 2A-D).

Surprisingly, all transgenes containing inserts shorter than 14z were more vigorously expressed at the mhb and additionally showed consistent expression in two ectopic locations, i.e. in the basal plate (extending from rhombomere 1 to the diencephalon) and in a transverse stripe of the forebrain (at the prosomere 2/3 boundary known as zona limitans intrathalamica; Figs 1D,E, 2I-N). These transgenes contained neither the proximal nor distal promoter of Pax5 with the exception of construct 8z (Fig. 1A). The DNA insert of this transgene was analyzed in both orientations relative to the TATA-box of the lacZ gene, indicating that the activity of the mhb enhancer is indeed orientation-independent. Interestingly, the transgene 8z only gave rise to ectopic expression in the basal plate if the exon 1A promoter of Pax5 was present in opposite direction to the lacZ gene and was thus unable to drive β-galactosidase expression (Fig. 1A,C). It is therefore likely that transcription factors binding to the endogenous Pax5 promoters restrict the activity of the enhancer to the mhb region of the embryo.

The sequence and relative position of the mhb enhancer have been highly conserved in the human and mouse Pax5 loci (Fig. 3), further indicating an important role of this regulatory element in the control of Pax5 expression. The minimal enhancer sequences, defined by transgene 0.43z, were 89% identical between the two species. As this high degree of sequence conservation did not allow us to pinpoint potential regulatory elements, we used a biochemical approach to identify transcription factor-binding sites in the enhancer. For this purpose we prepared protein extracts from dissected mhb tissue of early embryos. In the interest of procuring sufficient brain material, we switched to chick embryos, particularly since the temporal and spatial expression pattern of Pax5 is similar, if not identical, at the mhb in this distantly related vertebrate (Funahashi et al., 1999). Chick embryos at day 3 (HH stage 20), which corresponds approximately to embryonic day 11 of the mouse (Hamburger and Hamilton, 1992), were used for micro-dissection of the mhb region and protein extract preparation. This extract was analyzed by DNase I protection assay for factors binding to three overlapping DNA fragments which spanned the entire enhancer of transgene 0.6z. One of these DNA fragments (probe A; nucleotides 451-666) gave rise to a strong DNase I footprint which mapped to a potential Pax-binding site (Fig. 4A). This finding was confirmed by electrophoretic mobility shift assay (EMSA), using in vitro translated Pax proteins (Fig. 4B). Pax2 and Pax5 bound with high affinity to probe A in contrast to Pax3 and Pax6, which represent two other Pax subfamilies expressed in the developing CNS (Fig. 4B). Hence, members of the Pax2/5/8 family preferentially interact with an enhancer sequence which has been entirely conserved in mammalian Pax5 genes (Fig. 3).

EMSA analysis of the 3-day mhb extract with probe A revealed at least five different proteins which bound to the proximal 216 bp sequence of the Pax5 enhancer (Fig. 5B, lane 4). The formation of these five protein-DNA complexes was prevented in the presence of a 100-fold molar excess of a short oligonucleotide (E1; 465-497) encompassing the Pax-binding site (lane 5). Moreover, oligonucleotide E1, when used as a probe, detected a similar cluster of six protein-DNA complexes in the 3-day mhb extract (lane 16). Hence, a relatively short sequence of the enhancer interacts with multiple proteins and is thus referred to as enhancer element 1.

Sequence inspection of element 1 revealed a good match to the consensus binding site of Pax2/5/8 proteins and identified potential recognition sequences for POU transcription factors (Fig. 5A). Indeed, several members of the POU protein family are expressed in the developing CNS of the mouse embryo and are known to recognize DNA via the bipartite POU domain which consists of two structurally independent DNA-binding units referred to as POU-specific (POU_S) and POU homeo (POU_H) domains (Herr and Cleary, 1995). The POU_S domain interacts with the sequence ATGC and the POU_H domain with the motif (T/A)AAT (Herr and Cleary, 1995). These two recognition sequences are present in both orientations on enhancer element 1 (Fig. 5A) and may be bound by POU proteins in any possible combination, as the spacing, orientation and positioning of the POU_S and POU_H domains on DNA is highly flexible (Herr and Cleary, 1995). As the POU-and Pax-binding sites overlap (Fig. 5A), we introduced specific mutations into element 1 to identify the different proteins involved in complex formation. The effect of these mutations on protein binding was investigated by using the mutant oligonucleotides either as competitor DNA (Fig. 5B) or end-labeled DNA probes (Fig. 5C) for EMSA analysis with the 3-day mhb extract.

The complexes P1, P2 and P3 were identified by several criteria as Pax2a, Pax2b and Pax5, respectively. First, an excess of the unrelated high-affinity Pax-binding site of the CD19 gene (Czerny and Busslinger, 1995) prevented the generation of all three complexes (Fig. 5, lane 9). Second, a neutralizing antibody directed against the Pax5 paired domain (Adams et al., 1992) interfered with complex formation (lanes 2, 17). Third, deletion (Δ) or substitution (S1) of four base pairs within the Pax recognition sequence abolished binding of all three proteins (lanes 6, 7, 18, 19). Fourth, in vitro translated Pax2b and Pax5 co-migrated with complexes P2 and P3, respectively (lanes 23, 24). Complex P1 is most likely formed by the isoform Pax2a, whose size exceeds that of Pax2b by 23 amino acids (Dressler et al., 1990). Lastly, a mhb extract prepared from earlier, 2-day chick embryos (HH stage 12) only gave rise to complexes P1 and P2 (lane 10) in agreement with the fact that expression of Pax2 precedes that of Pax5 during mhb development (Rowitch and McMahon, 1995; Pfeffer et al., 1998).

Likewise, the three complexes referred to as X1, X2 and X3 were generated by POU proteins present in the 3-day mhb extract. First, these complexes were specifically competed away by unrelated octamer and Pit-1 recognition sequences (Fig. 5, lanes 13, 14). Second, an octamer probe generated the same three complexes as enhancer element 1 (lane 29). Third, the in vitro translated POU protein Oct1 also interacted with element 1 (lane 25,30). Fourth, mutation of the POU_H-binding sites (S2) abrogated the formation of all three complexes (lanes 7, 20). Finally, mutations (Δ, S1, S3) affecting the recognition motifs of the POU_S domain also interfered with protein binding (lanes 6, 15, 18-20). The effect of substitution S1 differed, however, between the two types of experiments (lanes 7, 18), suggesting that this mutation only reduced, but did not abolish the binding of POU proteins. A recent analysis of comparable mhb extracts prepared from mouse embryos resulted in the identification of three POU proteins with similar electrophoretic mobilities as our complexes X1, X2 and X3, which may therefore be generated by the proteins Brn1, Brn2 and Brn4, respectively (Mihailescu et al., 1999).

In summary, we conclude that members of the Pax2/5/8 and POU protein families bind to overlapping sequences in element 1.These proteins differ, however, in their sequence requirements as the mutations S2 and S3 specifically interfered with binding of the POU proteins (Fig. 5, lanes 8, 15, 27, 28).

To assess the in vivo significance of enhancer element 1, we next introduced the same mutations into the Pax-and POU-binding sites of transgene 0.6z. Fig. 7K summarizes the generation and expression analyses of the different transgenic embryos. The mutations Δ and S1, both of which interfere with binding of Pax and POU proteins, resulted in a similar reduction of transgene expression at the mhb (Fig. 7A-D). Both mutations interfered with β-galactosidase expression on the dorsal side of the neural tube, while the effect on ventral expression was variable (Fig. 7K). One embryo expressed the transgene only in the basal plate (Fig. 7A), while others showed reduced β-galactosidase expression within the mhb (Fig. 7B-D). In contrast, all embryos carrying the S2 or S3 mutation reproducibly revealed intense β-galactosidase staining at the mhb (Fig. 7E,F,K), as seen with the wild-type 0.6z transgene (Fig. 2L). Hence, the mhb enhancer of Pax5 is fully functional in the absence of POU protein binding which is selectively abrogated by the A to G mutation S2 or the 4 bp substitution S3 (Fig. 5). These experiments therefore identified the Pax-binding site as the critical regulatory element in the proximal enhancer region.

The transgene 0.43z contains the smallest enhancer region that was still able to yield mhb expresssion in our transgenic assay (Fig. 2M,N). Further deletion from the 5′ end led to inactivation of the enhancer, indicating that the 45 base pairs missing in transgene 0.39z must contain essential regulatory sequences (Figs 1A, 3). Sequence inspection revealed two overlapping homeodomain-binding sites (TAATTA; Gehring et al., 1994), an Otx-binding site (TAATCC; Gan et al., 1995) and a recognition sequence for Sp1-like zinc finger proteins (Briggs et al., 1986) (Fig. 6A). Indeed, probe B (nucleotides 170-303 in Fig. 3), which encompasses this module of recognition sequences (referred to as element 2), was shown by EMSA analysis to interact with multiple proteins present in the 3-day mhb extract (Fig. 6, lane 1). Moreover, all these proteins bound to sequences within element 2, as an excess of the corresponding oligonucleotide E2 prevented complex formation (lane 2). Conversely, the same oligonucleotide used as a probe also detected the various different proteins (lane 10). The two slowly migrating complexes are generated by Sp1-like zinc finger proteins as they could not be competed with an excess of the oligonucleotide M4 containing a mutation in the Sp1-binding site (Fig. 6, lane 6). Conversely, the formation of both complexes was specifically inhibited by competition with a consensus Sp1-binding site (lane 7). The second cluster of faster migrating complexes was caused by homeodomain proteins which could be titrated away by an excess of an unrelated homeodomain recognition sequence (lane 8) and failed to bind to oligonucleotide M1 containing a mutation in the homeodomain-binding site of element 2 (lanes 3,12). This finding was confirmed by in vitro translated Gbx2 and HoxD1, two representative homeodomain proteins, which bound the wild-type (E2), but not the mutant (M1) element 2 sequence (lane 13-16). However, in vitro translated En proteins were unable to bind to element 2, nor could an En-specific antiserum (Davis et al., 1991) affect the protein binding pattern in the mhb extract (lane 11, unpublished data). Mutation of the Otx recognition sequence (M3) did also not influence protein binding in the mhb extract (lane 5), although in vitro synthesized Otx2 had the potential to interact with element 2 (lanes 17, 18). It is therefore unlikely that Otx2 binds in vivo to element 2 consistent with the fact that the Otx recognition sequence has not been conserved in the human PAX5 enhancer (Fig. 3). In conclusion, these EMSA analyses identified a distal enhancer region which contains binding sites for homeodomain and zinc finger proteins.

The importance of the homeodomain-and zinc finger-binding sites was next assessed by analyzing the expression of 0.6z transgenes carrying the M1 and M4 mutations, respectively (Fig. 7K). These experiments revealed that the activity of the mhb enhancer is critically dependent on the homeodomain-binding sequence TAATTA, as its mutation (M1) strongly reduced transgene expression at the mbh (Fig. 7G,H). Although the loss of β-galactosidase staining varied among different transgenic embryos, it was in all cases most prominent in the dorsal region of the mhb, similar to the effects observed with Pax-binding site mutations in element 1. In contrast, a transgene carrying the M4 mutation in the zinc finger recognition sequence was normally expressed at the mhb in all embryos analyzed (Fig. 7I,K). Moreover, the zinc finger recognition sequence did not cooperate with the adjacent homeodomain-binding site, as the variability of transgene expression at the mhb was similar in the presence of the double M1/M4 or single M1 mutation (Fig. 7K). We conclude therefore that the homeodomain-binding site of element 2 is a second regulatory component of the mhb enhancer.

The identification of two functional elements which, upon mutation, similarly affected the activity of the mhb enhancer suggested that the Pax-and homeodomain-binding sites may functionally compensate for each other’s loss. To test this hypothesis, we have generated a 0.6z transgene carrying both homeodomain (M1)-and Pax (Δ)-binding site mutations. Although this double mutant transgene was expressed at ectopic locations in 7 different embryos, it was only once found at the mhb (Fig. 7J,K). This embryo showed β-galactosidase staining only in the dorsal region of the mhb where the expression of transgenes with corresponding single mutations was consistently reduced (Fig. 7). Hence, the staining pattern of this embryo did not correlate with the introduced mutations and was also considered to be ectopic due to an intergration site effect. In summary, these data demonstrate that the homeodomain-and Pax-binding sites cooperatively control the activity of the mhb enhancer.

Pax2/5/8-binding site in the mhb enhancer of Pax5 raised the question of which protein of the Pax2/5/8 family is involved in the regulation of this enhancer. All three members of this Pax subfamily are known to be expressed at the mhb in a. defined temporal sequence with Pax2 being activated first followed by Pax5 and later by Pax8 (see Pfeffer et al., 1998). In addition, we have recently demonstrated that the Pax2 and Pax5 proteins can substitute for each other and are thus equivalent with regard to their function in mhb development (M. B., unpublished data). The above question can therefore be rephrased to ask whether the initation of Pax5 expression depends on cross-regulation by Pax2 and/or whether Pax5 expression is later maintained by autoregulation.

To investigate an autoregulatory role of Pax5, we compared the β-galactosidase staining of heterozygous Pax5^+/lacZ and homozygous Pax5^lacZ/lacZ embryos which contained a lacZ gene insertion in the endogenous, targeted Pax5 allele (Urbánek et al., 1994). Embryos of the two genotypes failed, however, to reveal any difference in staining pattern at days 10.5 and 12.5 (Fig. 8A; unpublished data). The absence of functional Pax5 protein in homozygous mutant embryos thus indicates that autoregulation does not significantly contribute to Pax5 expression at the mhb. We next examined a possible dependency of the minimal mhb enhancer on Pax5 function. For this purpose, we crossed the transgene 0.6z into a mouse strain containing a mutant Pax5 allele which was inactivated by deletion of exon 2 (Δ) rather than by lacZ insertion (M. Horcher and M. B., unpublished data). β-Galactosidase staining at days 10.5 and 12.5 also revealed no difference between heterozygous and homozygous mutant embryos (Fig. 8B), indicating that the transgene, like the endogenous Pax5 locus, is not under autoregulatory control by Pax5.

The initiation of endogenous Pax5 expression was, however, shown by whole-mount in situ hybridization to be significantly delayed in homozygous Pax2^−/−embryos compared to heterozygous control embryos (Fig. 8C). In homozygous embryos, Pax5 expression could be first detected at the 7-to 8-somite stage, whereas Pax5 transcription is already initiated at 3-5 somites in heterozygous and wild-type embryos (Urbánek et al., 1994; Rowitch and McMahon, 1995). Furthermore, the expression of Pax5 commences in homozygous embryos near the midline which, on neural tube folding, becomes the ventral region of the mhb (Fig. 8C). At the same stage, Pax5 expression could not be detected in the lateral (later dorsal) aspect of the neural plate (Fig. 8C). However, the lacZ gene inserted in the targeted Pax2 locus was expressed throughout the entire mhb region in Pax2^−/−embryos, indicating that the lateral mhb tissue has not been lost at this early stage (Fig. 8E). Moreover, Pax5 expression increased up to the 12-somite stage (Fig. 8D), but could then not be followed further due to onset of tissue deletion in Pax2^−/−embryos raised on the C3H/He strain background (M. B., unpublished results). The delayed and medially restricted Pax5 expression in Pax2^−/−embryos unequivocally demonstrates cross-regulation of Pax5 by Pax2 and correlates well with the severe effect of the Pax-binding site mutation on enhancer activity in the dorsal mhb region. Taken together, these data indicated that the Pax recognition sequence in the minimal Pax5 enhancer mediates cross-regulation by Pax2 rather than autoregulation by Pax5.

The transcription factor Pax5 (BSAP) plays an essential role in B-lymphopoiesis and in the development of the mhb region which gives rise to the midbrain and cerebellum of the vertebrate embryo (Urbánek et al., 1994). Here we have demonstrated that a 25 kb fragment from the 5′ region of the Pax5 locus contains a 435 bp enhancer which directs gene expression in the correct spatial and temporal pattern at the mhb of transgenic embryos. This 5′ region was, however, unable to activate transgene expression in B-lymphocytes, indicating that the B-cell-specific control region(s) must reside at a more distant location from the Pax5 promoters. The mhb enhancer of the human and mouse Pax5 genes has been conserved in sequence and position at −5.6 kb upstream of the distal promoter. Surprisingly however, neither sequence comparison nor transgenic analysis allowed us to identify a mhb enhancer in the 5′ flanking region of the chick and pufferfish Pax5 genes (unpublished data), although Pax5 expression is very similar at the mhb in these vertebrates compared to the mouse embryo (Pfeffer et al., 1998; Funahashi et al., 1999). It appears therefore that the function but not the location of the mhb enhancer has been conserved during vertebrate evolution.

Several arguments suggest that the identified enhancer is the only mhb-specific control region of the mammalian Pax5 locus. First, we have identified only a single mhb enhancer within the 25 kb 5′ region analyzed. Second, this enhancer accurately reflects the early and late phases of Pax5 expression at the mhb in both wild-type and Pax5 mutant embryos. Third, the minimal 435 bp enhancer is sufficient to confer the correct mhb-specific activity to a heterologous β-globin promoter. When analyzed outside of the Pax5 context, the minimal enhancer can give rise to ectopic expression in the forebrain and basal plate of the neural tube, suggesting that communication between the enhancer and the endogenous Pax5 promoters is important for restricting its activity to the mhb region of the embryo.

The minimal mhb enhancer of Pax5 is composed of at least two elements which contain binding sites for homeodomain, zinc finger, POU and Pax proteins. Mutation of individual binding sites revealed that only the homeodomain and Pax recognition sequences are critically involved in the control of the mhb enhancer. Inactivation of either binding site affected the activity of the enhancer in a similar dorsoventral gradient, with the strongest reduction observed in the dorsal mhb region. Simultaneous mutation of both sites resulted in complete inactivation of the enhancer, demonstrating that the homeodomain-and Pax-binding sites cooperatively regulate the activity of the mhb enhancer.

The three members of the Pax2/5/8 family arose by gene duplications at the onset of vertebrate evolution and have since largely been conserved with regard to their expression patterns (Pfeffer et al., 1998) and DNA-binding specificity (Czerny et al., 1997). The temporal sequence of gene activation at the mhb has also been maintained with Pax2 expression being initated before Pax5 and Pax8 expression (Rowitch and McMahon, 1995; Pfeffer et al., 1998). Our recent expression analysis of Pax2/5/8 genes in zebrafish embryos revealed that the mhb-specific expression of Pax5 is entirely dependent on Pax2, as it is never initiated in noi mutant embryos lacking the Pax2.1 protein (Pfeffer et al., 1998). In contrast, the absence of Pax2 did not seem to affect expression of the mouse Pax5 gene at the mhb of Pax2^−/−embryos (Torres et al., 1996). However, the brain phenotype of the Pax2 mutation proved to be strongly dependent on the background of the mouse strain analyzed, ranging from almost normal development of the midbrain and cerebellum on the C57BL/6 background (Schwarz et al., 1997) to complete deletion of these brain structures in the C3H/He strain (Favor et al., 1996; M.Bouchard, unpublished data). By analyzing the Pax2 mutation on the non-permissive C3H/He background, we have now demonstrated that the expression of Pax5 is delayed at the mhb of Pax2^−/−embryos and fails to be initiated in the lateral neural plate which, upon neural tube closure, becomes the dorsal region of the mhb. Similarly, mutation of the Pax-binding site in the minimal Pax5 enhancer affected transgene expression most severely in the dorsal mhb region, suggesting that Pax2 exerts its control by directly interacting with the mhb enhancer. Protein-binding assays indeed demonstrated that the Pax2 and Pax5 proteins of the Pax transcription factor family bind with highest affinity to the Pax recognition sequence of the mhb enhancer. Autoregulatory control of the enhancer by Pax5 is, however, unlikely as lacZ genes under the control of the endogenous Pax5 locus or the minimal enhancer were normally expressed at the mhb of Pax5 mutant embryos. In summary, our data therefore indicate that the 435 bp enhancer of the mouse Pax5 gene is a direct target of Pax2 and requires Pax2 function for its initial activation at the mhb. Hence, the Pax5 gene has been evolutionarily conserved also with regard to its regulation at the mhb. The fish and mouse Pax5 genes differ only in the degree of their dependency on Pax2 function, as the loss of Pax2 prevents activation of the fish Pax5 gene at the mhb (Pfeffer et al., 1998), while it is partly compensated for by homeodomain proteins binding to the mouse Pax5 enhancer (this study).

Pax2 is already expressed during late gastrulation almost a day before the onset of Pax5 expression in the mhb region (Rowitch and McMahon, 1995), indicating that Pax2 can activate the Pax5 enhancer only in cooperation with other regulators. Homeodomain proteins binding to the functional TAATTA sequence of the mhb enhancer are likely candidates for these cooperating transcription factors. Different homeodomain proteins bind, however, promiscuously in vitro to the same TAAT motif which hampers biochemical identification of the particular regulator responsible for a specific in vivo phenotype (Gehring et al., 1994). Normal development of the mhb region depends on the homeodomain proteins Otx2, Gbx2, En1 and En2 (Joyner et al., 1991; Wurst et al., 1994; Matsuo et al., 1995; Wasserman et al., 1997). Of these proteins, only Gbx2 was able to interact with the functional TAAT motif of the Pax5 enhancer, while Otx2, En1 and En2 failed to bind even in the presence of Pbx1, which is known to enhance the binding of En proteins by cooperative dimerization on DNA (Peltenburg and Murre, 1996). The expression domains of Gbx2 and Otx2 overlap only partially with that of Pax2 in the mhb region (Hidalgo-Sánchez et al., 1999), thus rendering these two proteins less likely candidates for regulating the mhb enhancer. In contrast, the En1 gene is activated entirely within the Pax2 expression domain at the 1-somite stage just before the onset of Pax5 expression at 3-5 somites (Rowitch and McMahon, 1995). A direct role of En1 in the activation of the Pax5 enhancer is, however, excluded both by the inability of En1 to interact with the enhancer and by the fact that En proteins are constitutive repressors of gene transcription (Jiménez et al., 1997). It is therefore likely that a hitherto unknown homeodomain protein with an En1-like expression pattern cooperates with Pax2 in the regulation of the mhb enhancer of Pax5.

Patterning of the early neural tube critically depends on members of different Pax subfamilies. In particular, Pax6 plays an essential role in the development of the forebrain and spinal cord (Stoykova et al., 1996; Ericson et al., 1997), whereas Pax2 and Pax5 are together required for normal morphogenesis of the midbrain and cerebellum (Urbánek et al., 1994; Schwarz et al., 1997, 1999). Negative cross-regulation between Pax2/5 and Pax6 has been implicated in the establishment of mutually exclusive expression patterns of these Pax genes in the developing CNS (Schwarz et al., 1999). At the molecular level, this hypothesis has been verified in the developing eye where Pax6 and Pax2 are expressed in adjacent domains of the neural retina and optic stalk, respectively. In the absence of Pax2, the expression domain of Pax6 extends into the optic stalk concomitant with a similar expansion of the pigmented neural retina (Torres et al., 1996). Interestingly, the retina-specific enhancer of the Pax6 gene contains Pax2-binding sites which mediate transcriptional repression of Pax6 in the Pax2 expression domain of the optic stalk (Kammandel et al., 1999; P. Gruss, personal communication). Negative cross-regulation between members of different Pax subfamilies is therefore used to specify adjacent regions in the developing brain.

In contrast, genes of the same Pax subfamily arose by more recent gene duplications and then assumed non-redundant functions in the same developmental pathway (reviewed by Mansouri et al., 1996). A hierarchical relationship among these genes was frequently established by intercalation of one of the duplicated genes below the other in the same genetic cascade. Here we have presented genetic and biochemical evidence that Pax2 directly activates the Pax5 gene during development of the mhb region. A similar situation has recently been described for the duplicated Drosophila Pax6 genes toy and ey, both of which play key roles in eye development of the fly (Czerny et al., 1999). Toy was shown to function upstream of ey by binding to and directly regulating the eye-specific enhancer of the ey gene (Czerny et al., 1999). Hence, positive cross-regulation between members of the same Pax subfamily has independently evolved in the insect and vertebrate lineages to diversify the function of Pax proteins within a given developmental pathway.

We thank T. Jenuwein for initial help in generating transgenic mice, M. Horcher for providing Pax5+/Δ mice, R. Krumlauf for transgenic vectors, G. Martin, S.-L. Ang, W. Wurst and W. Herr for cDNA clones, A. Joyner for anti-En antibodies, U. Schibler for advice on nuclear extract preparation, G. Reim for analysis of the chick and pufferfish Pax5 genes and G. Schaffner for DNA sequencing. This work was supported by the IMP and the Austrian Science Foundation (grant P11025-MED).