Developmental expression of the mouse Evx-2 gene: relationship with the evolution of the HOM/Hox complex

The mouse genome contains 38 Hox genes which are clustered in four genomic loci (the HOXA,B,C and D complexes). These complexes have similar features of structure and organization as they arose by large-scale duplication of an ancestral complex (reviewed in McGinnis and Krumlauf, 1992; Dolle and Duboule, 1993). During development, Hox genes are activated during gastrulation, while the embyo establishes its anteroposterior axis. Hox transcripts are found in partially overlapping domains along the embryonic axis, with genes located towards the 5’ extremity of one complex being expressed in progressively more posterior (caudal) areas of the embryo (Gaunt et al., 1988). All these genes are transcribed from the same DNA strand so that each complex can be assigned with a general 5’ (posterior) to 3’ (anterior) orienta tion. The importance of this gene family to the establishment and realization of the vertebrate body plan has been recently demonstrated by gene inactivation experiments (LeMouellic et al., 1992; Ramirez-Solis et al., 1993; Dolle et al., 1993; Condie and Capecchi, 1993). The results of such inactivations suggest that the function of Hox genes could be mediated through the control of either the local growth or patterning (or both) of structures of neurectodermal and/or mesodermal origin.

In their 5’ regions, the HOXA, C and D complexes contain 4-5 Hox genes that are related to the Drosophila Abdominal-B homeotic gene (lzpisua-Belmonte et al., 1991). The tandem duplication of these Hox sub-groups is thought to have occurred before large scale HOX cluster duplication and may, perhaps, have been linked to evolution of appendicular struc tures, since such genes are coordinately expressed in the devel oping limbs and genital tubercle (Dolle et al., 1989, 1991a; Yokouchi et al., 1991) in which they exert an important function (Dolle et al., 1993; Small and Potter, 1993).

The coordinate expression of Hox genes may result from a successive temporal activation of these genes, in a 3’ to 5’ sequence, the temporal colinearity (see Dolle et al., 1989; Izpisua-Belmonte et al., 1991; Duboule, 1992). As a conse quence, genes located in 5’ positions display transcript domains that are always more distal (or posterior and distal at earlier stages) than their 3’ neighbour genes (the spatial colinearity). In the genital tubercle, quantitative differences are observed rather than true spatial colinear expression domains (Dolle et al., 1991a). The fact that some of these genes are functional in all axial skeleton structures has been confirmed by the exper imental knock-out of Hoxd-13, which alters both the growth and patterning of skeletal elements arising from the limb buds and genital tubercle (Dolle et al., 1993).

Evx-1 and Evx-2 are the two genes identified to date in mammals, which contain a homeobox homologous to that of the even-skipped (eve) gene from Drosophila (Bastian and Gross, 1990). A genomic walk along the upstream (5’) part of the human HOXD locus revealed that the human EVX.2 gene is located in the close vicinity of HOXDJ 3 (D’Esposito et al., 1991) and a similar linkage was reported in mice (Bastian et al., 1992). The human EVX.2 homeobox is found about 13 kilobases (kb) away from that of HOXD13. Similarly, the EVXJ gene is linked to the HOXA complex. In this case, however, its distance from the complex is larger, as it lies approximately 45 kb upstream of the HOXAJ 3 gene (Faiella et al., 1992). Interestingly, both EVXJ and EVX.2 are transcribed from the DNA strands that are opposite to those from which all HOX sequences are transcribed (Faiella et al., 1992; D’Esposito et al., 1991).

The developmental expression of the murine Evx-1 gene has been studied in detail; Evx-1 starts to be expressed at approx. 6.5 days post-coitum (dpc) in part of the embryo epiblast, and then throughout gastrulation (6.5-8.5 dpc) in a graded fashion along the primitive streak and the involuting mesodermal cells, but not anterior to the primitive streak (Dush and Martin, 1992). Later on, Evx-1 is still expressed in the tail bud, which corresponds to the primitive streak in its histogenic potential to generate various tail structures. Starting at approx. 10.5 dpc, Evx-1 is also expressed in discrete neural areas along the spinal cord and hindbrain (Bastian and Gross, 1990). Finally, Evx-1 transcripts are transiently detected in the limb buds, where they first appear in the posterior distal mesenchyme, and are subse quently maintained only in very distal (subectodermal) cells (Niswander and Martin, 1993). Thus, none of the three com ponents of the Evx-1 pattern during development (early, neural and limb expression) resembles the coordinate regulation of the neigh-bouring Hox genes, which are7.5 dpc embryo sections (Fig. 2A). In contrast, a graded disactivated at later stages of gastrulation and in progressively more posterior areas (Yokouchi et al., 1991; Haack and Gross, 1993).

We have analysed further the linkage of Evx-2 to the mouse HOXD complex and show that it lies in close contact with the Hoxd-13 gene (approx. 8 kb). Such a distance is in the range of the usual spacing between Hox genes themselves and we investigated whether Evx-2 gene expression could, in part, be controlled by regulatory mechanisms acting on the neig bouring Hoxd genes, or instead, would be regulated independently from the complex, as seems to be the case for Evx-1. We report here that Evx-2 is regulated as expected for a gene belonging to the HOXD complex, in both the limb buds and the genitalia. Interestingly, Evx-2 expression follows temporal colinearity (i.e. shortly after the Hoxd-13 gene), and thus does not share the early expression phase of Evx-1 during gas trulation. However, an important aspect of Evx-2 expression (in the CNS) makes it different from other Hoxd genes but similar to Evx-1. The relationships between the respective genomic locations of Evx genes and the potential control of their expression by regula tory mechanisms acting upon the HOX complexes are discussed from an evolutionary perspective.

The cloning and sequence analysis of DNA fragments located upstream the Hoxd-13 gene on genomic cosB (Dolle et al., 1991b) confirmed the linkage between the Evx-2 sequences and the HOXD complex. Sequences were identified that corre spond to the previously reported sequence of an Evx-2 cDNA (Dush and Martin, 1992) and encode the N terminus of the Evx-2 product. We localized the Evx-2 ATG initiation codon to 8 kb from the Hoxd-13 initiation codon, in an inverted orienta tion (Fig. 1). The spacing between Evx-2 and Hoxd-13 coding regions was thus found to be in the range of that observed between all five neighbouring Abdominal-B related Hox genes (from 5 to 10 kb; Fig. 1). This genomic organization is distinct from that of the EVXJ locus, which was mapped, in human, at about 45 kb from HOXA13 (Faiella et al., 1992; Fig. 1).

Dush and Martin (1992) reported the activation of Evx-1 shortly before the onset of gastrolation; here we compared Evx-1 and Evx-2 expression at corresponding stages by in situ hybridization. No detectable Evx-2 labeling was observed in tribution of Evx-1 transcripts was een in the primiLive sLreak, the embryonic mesoderm and the epiblast (Fig. 2A; and Dush and Martin, 1992). Evx-2 expression was al o not detected in 8.5 dpc embryos, a stage al which the Evx-I signal was present in caudal areas (Fig. 2B). At this stage. a restricted Hoxd-/3 signal wa. seen in the most posterior embryonic mesoderm and hindgut (Fig. 2B. arrow). This probably represent. the time of Hoxd-I3 activation. slightly earlier than previously reported (Dolle et al., 1991b). At later stages of development, no Evx- 2 labeling wasdetected in the tail bud (Fig. 2C). a region equiv alent to the primitive streak (Tam. 1984), even though. pecific signals were observed in other areas of the same embryos (see below). In contrast. both Evx-I and Hoxd-I3 Lranscripts were detected in the tail bud, although with different distributions (Fig. 2C). Evx-2 expression was not merely delayed in the primitive streak and tail bud, with respect to Evx-1; instead, Evx-2 was not activated in these structures.

The earliest signs of Evx-2 transcription were seen in 9.5 dpc embryos, in the cloacal epithelium and surrounding mesoderm (Fig. 3A), from which the genital bud will arise. Hoxd-13 was also expressed in these structures, but extending over a larger area (Fig. 3A). Evx-1 was also expressed in this future genital region (Fig. 3A). Evx-2 transcripts were not visible in 9.5 or 10.0 dpc limb buds while Hoxd-13 labeling was clearly detected in posterior mesenchyme (Fig. 3B). The first detection of Evx-2 expression in the forelimb bud was at 10.5 dpc, when it was restricted to a subset of Hoxd-13-expressing cells in distal and posterior mesenchyme, within an area that also expressed Evx-1 (Fig. 3C). The co-expression of Evx-1 and Evx-2 in both the genital area and limb buds was transient and the two expression patterns diverged by 11.5 dpc (not shown). In 12.5 dpc fetuses, Evx-1 was expressed in the very distal extremities of the limb mesenchyme, while Evx-2 displayed a Hoxd-type of expression pattern, that is, across a wider distal region (Fig. 3D; see below).

Hoxd genes are expressed in the limb buds along spatial domains which are colinear with the ordering of the genes along the chromosome (Dolle et al., 1989). In both 11.5 dpc fore- and hindlimbs, the distribution of Evx-2 transcripts was consistent with this gene being subject to a similar mechanism of regulation. Successive sections of hindlimb buds showed that Hoxd-13 transcripts were distributed in slightly more extended domains than those of Evx-2, both more proximally, along the posterior margin of the bud (Fig. 4A), and more ante riorly, when progressing towards the tip of the bud (Fig. 4B,C). These differences were still visible in 12.5 dpc limbs (Fig. 4D). Both Evx-2 and Hoxd-13 transcripts were restricted to the foot plates; however, the Evx-2 domain was not as proximal as the Hoxd-13 domain along the posterior margin and the anterior margin (Fig. 4D). The weak expression of Evx-2 in the limbs of 13.5 dpc fetuses indicated that its transcript domain extended into all five digit primordia, up to the base of the most anterior digit I (Fig. 4E), a domain equivalent to that in which Hoxd-13 is expressed and functional (Dolle et al., 1993). However, while transcripts of 5’-located Hoxd genes are known to accumulate at high levels in the limb extremities, with Hoxd-13 displaying the strongest expression until at least 17.5 dpc (Dolle et al., 1991b; unpublished data), the Evx-2 signals remained much weaker at 12.5 and 13.5 dpc (see Fig. 4D-E) and were hardly detectable at 14.5 dpc (not shown).

Hoxd genes are expressed in the genital tubercle where 5’-gene transcripts are more abundant, and maintained for longer times, than 3’-gene transcripts (Dolle et al., 1991a). Serial sections of 11.5 dpc embryos showed that Evx-2 also had a specific expression domain in the genital bud and urogenital mesenchyme. This domain is similar to that of Hoxd-13, both being centred in the genital bud, and around the urogenital sinus and rectum (Fig. 5A). However, Evx-2 transcripts were not detected in more lateral regions of the genital bud (Fig. 5B, see also Fig. 4B,C), suggesting either that Evx-2 transcripts may be restricted to a subset of Hoxd-13-expressing cells in the genital bud and urogenital mesenchyme or that the tran script level is too low to be detected.

Evx-2 was expressed in discrete cell layers within the embryonic central nervous system (CNS). This was first detected at 10.5 dpc and was clearly apparent until 12.5 dpc. The labeling was observed in a thin and continuous column of cells along the entire spinal cord in near mid-sagittal sections (Fig. 6A) as well as in more scattered cells of the ventral spinal cord, visible in more lateral sections (Fig. 6B). As this labeling was reminiscent of that reported for Evx-1 (Bastian and Gruss, 1990), we compared the expression of both genes at various levels of the CNS on both transverse and coronal sections. Double-labeling experiments were performed in which the two probes were hybridized, either separately, or together on a third set of sections in order to assess possible co-expression. On transverse sections of 12.5 dpc embryo spinal cord, Evx-2 labeling was indistinguishable from, although weaker than, that of Evx-1 (Fig. 6C). In the cervical spinal cord, Evx-2 signal was found in two symmetrical columns in the medial part of the ventricular zone (Fig. 6C, arrowheads; also in Fig. 6A), as well as in cells located between the ventral horns and the ven tricular zone (Fig. 6C, open arrows; also in Fig. 6B) and in the most dorsal layers of the spinal cord (Fig. 6C, filled arrows). Only the former two areas were labeled in more caudal regions of the spinal cord (thoracic and lumbar; data not shown). Sections hybridized to both probes together showed no detectable extension of the signal, which indicated a probable co-expression of the two genes in the same cells (Fig. 6C; panel ‘1 +2’). A similar co-expression was found more rostrally, at the level of the developing hindbrain (Fig. 6D). In the CNS, Evx-1 and Evx-2 appeared to be activated at a similar devel opmental stage, in 10.5 dpc embryo (data not shown).

In contrast to the similarities in the spinal cord, a striking difference was seen in the extent of Evx-2 and Evx-1 transcripts domains towards more rostral regions of the developing brain. While Evx-1 transcripts were expressed up to the rhomben cephalic isthmus area (the metencephalon-mesencephalon boundary, see Bastian and Gruss, 1990; Figs 6D, 7), Evx-2 transcripts extended into the superficial layer of the entire midbrain (Fig. 7A,B). The point of divergence between Evx-1 and Evx-2 transcript distributions in the marginal layer was seen on coronal sections of 11.5 dpc brain (Fig. 7A), where it appeared that the Evx-2 signal was found more rostrally than Evx-1 in the same layer. Therefore, the expression of Evx-2 in the superficial layer of the midbrain (Fig. 7B) corresponds to a rostral extension of domains where Evx-2 and Evx-1 are coexpressed.

Members of the vertebrate HOX gene complexes are expressed, during development, according to a set of rules that control the coordination of their functions. An interesting aspect of these rules is that they rely on the genomic organiz ation of this gene family, that is, they rely on the respective position of each gene within its complex. Thus, a particular position will coincide with both the time of activation (temporal colinearity. Dolle et al., 1989; lzpisua-Belmonte et al., 1991) and the craniocaudal position of the expression domain (spatial colinearity, Gaunt et al., 1988). Though the mechanism underlying these rules have yet to be discovered. it has been hypothesized that spatial and temporal colinearities were major contributory factors to the con ervation of this clustered organization during evolution (Duboule, 1992. 1994).

The two vertebrate genes Evx-1 and Evx-2 are related to the Drosophila even-skipped gene and are closely linked to the 5’ extremities of the HOXA (Evx-1) and HOXD (Evx-2) complexes (D”Esposito et al., 1991: Faiella et al., 1992; Bastian et al., 1992). Interestingly, while the distances between paralogous Hox genes have been relatively well conserved among the various HOX complexes (an average of approx. 10 kb), the distances between the Evx genes and the HOXA and D complexes have not been equally conserved: EVX I is separated from HOXA13 by 45 kb. whereas EVX2 was found in the close vicinity of HOXD13 (D’Esposito et al., 1991; Faiella et al., 1992). We have further analysed the mouse Evx- 2 locus and found that this gene lies 8 kb from the Hoxd-13 gene. in an inverted orientation. Thus, while the distance between EVX I and HOXA13 is well above those usually found between vertebrate Hox genes. the spacing between Evx-2 and Hoxd-13 i. in the range of the distances separating Hoxd genes themselves.

The expression profile of Evx-1 during development (Bastian and Gruss. 1990, Dush and Martin, 1992, this work) suggests that its physical linkage to the HOXA complex is not associated with a ‘Hox type’ of regulation. Itsearly expression. at the onset of ga trulation. does not follow temporal colin earity (it is expressed well before Hoxa-13) and, subsequently. no restriction to posterior regions isobserved in the spinal cord expression (as is the case for Hoxa-13). Even though Evx-1 u anscripts are found in both limb buds and genital tubercle, their distribution in time is distinct from that of Hox transcripts in these structures. As the Evx-2 gene was found to be physi cally closer to the HOXD complex, we analysed whether its expression could be subject to regulatory influences of Lhe nearby HOXD complex. Our results show that the expression of Evx-2 corresponds. in many respects, Lo thal of a genuine Hoxd gene. However, an important a pect of regulation conserved between Evx-1 and Evx-2 relates Lo expression in the central nervous system.

Evx-2 expres ion seems to comply with temporal colinear ity of the HOXD complex (Dolle et al., 1989; Izpisua-Belmonte et al., 1991), as there was no area in the embryo where Evx-2 transcripts could be detected prior to those of Hoxd-13. Evx-2 transcripts were detected earliest in 9.5 dpc embryos, in a region of the genital anlage corresponding to a subset of the Hoxd-/3 domain. This suggests that Evx-2 is activated slightly after Hoxd-13 in the cell population that will eventually give rise to the genitalia. This delay was even clearer in limb buds, where Evx-2 expression was detectable only by day 10.5 in posterior forelimb mesenchyme, a region that expressed Hoxd-13 at 9.5-10.0 dpc. Possibly as a result of this temporal progre ion, the Evx-2 transcript domain in developing limbs appears to be more restricted than that of Hoxd-13. This is in good agreement with the spatial colinear ity observed amongst Hoxd genes in limbs (Dolle et al., 1989).

Temporal colinearity may be respected in the activation of Evx-2 in the central nervous system as well, even though this is a structure where the Evx-2 expression pattern is clearly distinct from that of all the neighbouring Hoxd genes. Therefore, there is a correlation between the close association of Evx-2 to the HOXD complex and the fact that its develop mental expression follows the rules of temporal and spatial col inearities acting upon the adjacent complex. fn this respect, and in contrast to Evx-1, Evx-2 could be considered as an additional Hoxd gene. Such an assimilation is based on the location and regulation of Evx-2 and does not necessarily imply similar functions of the gene products. Furthermore, the following reservations are noteworthy; first, the Evx-2 signal was always much weaker than that of Hox genes, making it possible that the distributions of transcripts, in both space and time, were underestimated due to technical problems with detection of gene activity. However, results obtained with other mouse Evx- 2 probes confirm that the gene is expressed at a low level, as is also suggested by comparison of Evx-2 versus Hoxd-13 expre. sion in zebrafish (P. Sordino and D. D., unpublished data). Second, the expression of Evx-2 in the CNS clearly con tradicts spatial colinearity as Hoxd-13 is not expressed in the CNS anterior to the sacral region. In addition, it should be noted that Evx and Hoxd genes are not expressed in the same cellular subsets in the spinal cord.

Previous analyses of vertebrate and invertebrate genes from the even-skipped family revealed two features of expression in common. First, there are similarities in the early expression pattern, which is initiated prior to gastrulation and subse quently restricted to posterior parts of the embryo. This is exemplified by the expression, during gastrulation, of mouse Evx-1 (Dush and Martin, 1992) or zebrafish eve-I (Joly et al., 1993), and by grasshopper and beetle eve genes until caudal extension (Patel et al., 1992, 1994). In vertebrates, this early expression is maintained in the tail bud. Second, these genes have specific expression patterns in the CNS, for example, eve in Drosophila (Doe et al., 1988). Evx-2 is the first example of an eve-related gene that does not show either the early posterior expression, or its persistence in the tail bud. As such, this gene is rather atypical amongst the eve gene family. A posterior expression is however resumed later in development, at a position similar to Hoxd gene expression in the genital anlage and following the activation of Hoxd-13. In limbs and genitalia, Evx-1 and Evx-2 expression were similar at first, but soon diverged, essentially through a decrease of Evx-1 expression. Such differences in the dynamics of the expression domains suggest that the systems of expression maintainance may have differentially evolved between the two genes.

In contrast, Evx-2 expression in the CNS was almost identical to that of Evx-1, both in timing and distribution. Both genes were activated around day 10 of development and Evx-2 transcripts showed the same cell-type specificity, mostly in post-mitotic cells (early differentiating neurons). These highly restricted expression domains were located at presumptive areas for interneurons (Bastian and Gruss, 1990). It is therefore probable that the function of both Evx-2 and Evx-1 in the CNS, is to identify particular types of neural cells; a role with parallels to the Drosophila eve gene (Doe et al., 1988). This aspect of Evx-2 regulation, clearly shared by Evx-1, is likely confered by some CNS-specific regulatory sequences which, were conserved after large-scale duplication of an original HOM/Evx complex.

Although both genes are probably co-expressed along the spinal cord and hindbrain, the cranial boundaries of the Evx-1 and Evx-2 expression domains in the CNS are different. While the rostral limit of Evx-1 expression is found at the rhomben cephalic isthmus (the hindbrain-midbrain transition), Evx-2 transcripts extend more rostrally into the midbrain, up to the midbrain-forebrain junction. In the midbrain, the unique expression of Evx-2 is observed in cell layers similar to those expressing both Evx genes at more caudal levels of the CNS. It is thus conceivable that this additional Evx-2 rostral domain reflects an anterior extension of the original Evx function. This functional extension may have followed the duplication of an original HOX/Evx complex, in prechordates or early in the ver tebrate lineage (see below), in parallel to the acquisition of a complex anterior nervous system. It is noteworthy that both Evx-1 and Evx-2 rostral limits of expression coincide with two important morphological sub-divisions of the embryonic brain, between the mesencephalic-metencephalic and metencephalic diencepalic vesicles, respectively.

These results, together with previously published data, suggest the following scheme for the history of the functional rela tionships between eve/HOM related genes. The conservation of an original linkage between an ancestral eve gene and an ancestral HOM complex (presumably formed originally by horizontal gene duplication) may have been favoured by a pri mordial function of eve in the early specification of the ‘posterior’ in development. This assumption is supported by the presence of an eve-like gene linked to a potential HOM complex in a Cnidarian (Miller and Miles, 1993). Before the separation between protostomes and deuterostomes, this ancestral gene was secondarily recruited for neuronal specifi cation, probably in association with the evolution of different cell types in an ancient CNS. At this point, temporal colinear ity must have been imposed on the HOM complex (if one assumes that this mechanism is used in some present day arthropods or annelids; Duboule, 1992). However, because of its early function, it is probable that the ancestral eve gene was not subject to this additional regulation. It nevertheless remained at the vicinity of the complex, perhaps because of the regulatory control determining posterior specificity. In many arthropods, a similar configuration could still exist nowadays, where an eve-like gene would be next to the HOM complex (on the ‘posterior’ side), playing an important function during early posterior development (Patel et al., 1992). In long germ band insects, a novel function (in addition to neural and posterior functions) was developed for eve, as reflected by the pair rule expression pattern observed in Diptera eve (see Patel, this volume). In fact, a reminiscence of the ancestral posterior pattern may be seen in flies, with the expression of eve in the primordial proctodeum at the end of germ band extension (Goto et al., 1989).

In early deuterostomes, a primitive Evx gene was thus linked to an ancestral HOX complex and probably had functions during both gastrulation and neuronal specification. The present configuration of the vertebrate HOXA/Evx-1 complex may reflect this ancestral situation, with the Evx gene relatively independent from HOX regulation. In cephalochordates, a situation quite similar to the ancestral one may be observed, since these close relatives of the vertebrates still retain a single HOX complex (see Holland et al., this volume). In the lineage leading to vertebrates, either in parallel with, or soon after, the large scale complex duplications, a rearrangement might have taken place in the upstream part of the HOXD complex resulting in a reduction in the distance separating Hoxd-13 from Evx-2. (The alternative, that the Evx-1 gene could have been secondarily removed from the immediate proximity of Hoxa-13 would imply that an ancestral Evx gene, before dupli cation, had no early posterior function. We consider this unlikely, as discussed above.) Since the large scale HOX complex duplications probably occurred in parrallel to the design of a complex central nervous system, it is possible that at this time one of the duplicated Evx genes (Evx-2) was also recruited to extend its original function up to a more anterior region.

The close vicinity of the Evx-2 gene to the HOXD complex correlates with the two essential differences between Evx-2 and other members of the eve gene family. On the one hand, the aquisition by Evx-2 of expression features that are characteris tic of Hox genes; on the other hand, the suppression of a major trait of the eve family, the early phase of expression. We suggest that these two novel components of Evx-2 expression pattern arose by this gene coming under the influence of the neighbouring HOXD complex. As a consequence, while the Evx-2 gene became expressed in limbs and genitalia through the control of some of the Hoxd-13 regulatory (enhancer) elements, its early phase of expression was suppressed by temporal colinearity. Indeed this latter mechanism would not allow the Evx-2 gene to be transcribed earlier than Hoxd-13, that is not before 8.5-9 days of gestation. The molecular mech anisms involved in temporal colinearity are not known. One potential explanation is the sequential opening (accessibility) of the HOX complexes, from 3’ to 5’, to the transcription machinery during gastrulation (Dolle et al., 1989; Duboule, 1992). In this context, the inactivation, in the course of evolution, of the early phase of Evx-2 expression by its translo cation next to the 5’ extremity of the HOXD complex (i.e. near an early ‘inactive configuration’) may be mechanistically similar to position effect variegation in Drosophila (Reuter and Spierer, 1992). If this were the case, however, one might not expect Evx-2 to be expressed anterior to Hoxd-13, even in different cell types. Analysis of the location and structure of the regulatory element(s) conferring CNS specificity on Evx-2 may, therefore, give insight into the mechanism involved not only in Evx-2 regulation, but also in temporal colinearity of HOX complexes.

We thank Professor P. Chambon and Drs J.-C. Izpisua-Belmonte and T. Schimmang for their support and contributions to various phases of this project and J. Zakany and for critical reading of the manuscript. We thank M. Philippe for technical assistance as well as D. Decimo for suggestions in the in situ hybridization procedure. We also thank Drs M. Dush and G. Martin for the Evx-1 probe and B. Boulay, S. Metz, V. Stengel and C. Werle for help in preparing the manuscript and illustrations. This work was supported by funds from the Institut National de la Sante et de la Recherche Medicale, the Centre National de la Recherche Scientifique, the Centre Hospitalier Universitaire Regional, the Association pour la Recherche sur le Cancer, the Fondation pour la Recherche Medicale, the Human Science Frontier Program, the EMBL, the Swiss National Science Foundation, the University of Geneva and the Foundation Claraz.