Function of posterior HoxD genes in the morphogenesis of the anal sphincter

The key role of Hox genes in the organization of the vertebrate body plan is well documented. Genetics experiments involving either gain-of-function or loss-of-function alleles have shown that they are required to properly build structures along the major body axes (reviewed in Krumlauf, 1994). Their developmental expression patterns have revealed that genes located at the 3′ extremities of the complexes are expressed anteriorly while genes located at 5′ positions are expressed in progressively more restricted posterior areas (Gaunt et al., 1988; Dollé et al., 1989; see Krumlauf, 1994). This type of colinearity, analogous to that observed in Drosophila (Lewis, 1978), is visible not only within presumptive skeletal structures such as vertebrae or appendicular bones, but also in internal organs or systems such as in the genitourinary organs (Dollé et al., 1991a,b) or the digestive tract (Yokouchi et al., 1995; Roberts et al., 1995). As a consequence of colinearity, genes located at the 5′ extremity of the Hox complexes, and related to the Drosophila gene AbdB, are expected to exert their functions at the distal ends of the various body axes. This was first verified by the inactivation of the Hoxd-13 gene, which led to alterations in the sacrum, the digits and the penian bone (Dollé et al., 1993).

Amongst the various sites of expression of Hox genes, the alimentary canal is of particular interest as it represents one of the phylogenetically oldest structures wherein such genes have been speculated to have a function. Unlike skeletal elements, which are present only in a minority of metazoa and emerged rather recently in evolution, most animals have a ‘digestive’ tube. It may thus represent one of the oldest signs of an anterior-posterior asymmetry and, thereby, be associated with the original requirement for a rostrocaudal specification. The regionalization of the gut, a necessity for the achievement of a variety of important functions, may thus rely on the Hox gene complexes (Roberts et al., 1995; Yokouchi et al., 1995).

We previously reported the expression of the two most posterior Hoxd genes, Hoxd-13 and Hoxd-12, in the terminal part of the hindgut (Dollé et al., 1991a,b). More recently, expression analyses revealed that 5′-located genes belonging to either the HOXD or HOXA chicken complexes are expressed in a colinear fashion and that the transition between expression domains correlates with anatomical divisions (Roberts et al., 1995; Yokouchi et al., 1995). Therefore, it seems plausible that colinear expression of Hox genes in the hindgut may be informative regarding changes in morphologies, much like what is known to happen in the vertebral column. Such a function for vertebrate Hox genes in the gut is particularly intriguing as some Drosophila homeotic genes too, are required for proper development of midgut mesoderm, mutant flies showing morphological alterations in muscle constrictions (e.g. Bienz, 1993). The potential importance of AbdB-related genes in terminal gut morphogenesis is also supported by the expression of egl-5, the C. elegans gene orthologous to AbdB, in rectal blast cells, a group of cells lining the anus (Chisholm, 1991), as well as by the presence of both Hoxd-13 and Hoxa-13 transcripts in the developing cloaca of zebrafish (van der Hoeven et al., 1996a; P. Sordino, T. K. and D. D., unpublished).

Specialization of the gut epithelia is thought to occur through an induction from the visceral mesoderm (e.g. Haffen et al., 1987), while earlier inductive signals emanating from the endoderm may instruct mesodermal tissue to undergo gut differentiation (Kédinger et al., 1990). The molecules involved in these early interactions are unknown, although expression studies suggest that the signaling molecule Sonic hedgehog (Shh) may be part of the cascade, upstream of Hox genes (Roberts et al., 1995). So far, however, a genetic demonstration of the function of these genes during gut morphogenesis was lacking. In order to document this point, we have analysed the effect of the inactivation of two Hoxd genes on the morphology of the anorectal region. Both the Hoxd-13 and Hoxd12 genes were shown to be expressed, during mouse development, in the terminal part of the hindgut mesoderm, in the presumptive musculosa, which will generate, after birth, the circular and longitudinal layers of muscles. Expression of both genes was strong and overlapping, up to the most distal part, i.e. the level of the internal sphincter that demarcates the anorectal transition (Dollé et al., 1991). In addition, Hoxd-13, but not Hoxd-12, was heavily expressed in rectal epithelium, which is derived from definitive endoderm, while undetectable in squamous anal epithelium, which derives from the ectoderm.

Both Hoxd-13 (Dollé et al., 1993) and Hoxd-12 (this work) have been inactivated and we report here that, in homozygous mutants of either genes, the morphology of the internal sphincter was deeply disorganized. In particular, the longitudinal muscle layer was severely reduced or partially absent. We show that this defect becomes apparent soon after birth, after the two muscle layers differentiate. In the case of Hoxd13, a high proportion of aged homozygous males developed an anal prolapsus, likely due to an incapacity of the internal sphincter. These results demonstrate the requirement of Hox genes for proper gut morphogenesis. The phylogenetic impli- cations of this observation are discussed in the context of a potential origin of colinearity in evolution.

For the construction of the replacement targeting vector, a KpnI-SalI 5.5 kb fragment of mouse genomic DNA containing the Hoxd-12 gene was subcloned from a cosmid clone (cosB, Izpisúa-Belmonte et al., 1991). The bacterial lacZ gene, coding for the β-galactosidase (β-gal) enzyme, followed by a SV40 poly(A) sequence (lacZ cassette), was inserted in frame in the Hoxd-12 protein coding sequence at the StyI site in the first exon and a loxP-PGKneo-loxP cassette was introduced next to the lacZ cassette, in the same transcriptional orientation. 10 μg of the construct, digested with both KpnI and SalI, was electroporated into 1×10⁷ ES cells (D3, Doetschman et al., 1985). One internal and two external probes were used to verify the targeting; the internal probe was a 0.3 kb ClaI-EcoRV fragment within lacZ (Fig. 1; lac probe), while the external probes were either a 0.2 kb KpnI fragment (Fig. 1; kk200 probe) or a 0.6 kb SalI-Eco47III fragment (Fig. 1; se₄₇600 probe). Two out of 108 G418 resistant colonies were correctly targeted, amplified and injected into C57BL/6 blastocysts. As both clones gave equivalent results, as judged by β-gal staining of progenies from chimeric males, only one of them was selected to excise the loxP-PGK-neo-loxP cassette. This was done by electroporating 10 μg of the pMC-CRE plasmid producing the Cre recombinase from bacteriophage P1 (Gu et al., 1993). Clones were selected for the deletion of the PGK-neo cassette by using the se₄₇600 DNA fragment as a probe (Fig. 1) and one such clone was injected into blastocysts and transferred into the germ line.

Whole-mount enzyme histochemical detection of β-galactosidase activity in embryos was carried out as described previously (Zákány et al., 1988). Embryos were isolated at appropriate stages, fixed in 4% paraformaldehyde and stained with X-gal cocktail. Stained animals were embedded in paraffin and sectioned. As β-gal staining was weak on 10 μm sections, the picture in Fig. 2E was taken directly from the paraffin block, at a mid-sagittal level. For histological analyses, animals and organs were fixed with Bouin’s fluid and embedded in paraffin following standard procedures. 7 μm thick sections were stained with eosin and hematoxylin or van Guison staining solution (for goblet cells). The Hoxd-13 probe as well as the protocol used for in situ hybridizations were as described previously (Dollé et al., 1991).

For skeletal preparations, animals at 1, 6, 13 and 70 days after birth were killed. Kidneys or tails were taken for isolation of DNA and animals were genotyped by Southern hybridization. Whole-mount skeletal preparations of newborn, juvenile and adult mice were carried out according to established procedures (Inouye, 1976; Dollé et al., 1993).

In order to follow the expression of Hoxd-12, we inserted the lacZ reporter sequences in frame with the Hoxd-12 gene. To preserve all Hoxd regulatory elements, no deletion of genomic sequences was generated by the targeting construct (Fig. 1AC). The PGKneo selectable gene was subsequently removed in order to avoid possible interference with the surrounding Hox transcription units. The targeted allele was identified as a 6.5 kb or 7.5 kb XhoI fragment with probe kk200 or probe se₄₇600, respectively, instead of the 8.5 kb band expected for both probes in the wild-type configuration (Fig. 1D). A lacZ probe was used to verify that a single insertion event had occurred. Excision of the PGKneo cassette (Fig. 1D) was produced by subsequent electroporation of a plasmid expressing the Cre recombinase of bacteriophage P1. The PGKneo deleted locus was identified as a 5.5 kb XhoI fragment with probe se₄₇600 (Fig. 1D,E). Two neo-positive ES cell clones as well as one neo-negative clone (derived from one of the former) were injected into C57BL/6 blastocysts. In all cases, germ-line chimeras were obtained and heterozygous animals from F₁ progenies were crossed to generate homozygous Hoxd-12deficient mice. Such animals were viable and fertile.

β-galactosidase activity was first assayed in undifferentiated ES cell clones containing the Hoxd-12/lacZ fusion gene. No βgal staining could be recovered, either in the absence or in the presence of all-trans retinoic acid (4 days of treatment), in contrast to results obtained in the human NT2/D1 teratocarcinoma cell line where this gene was reported to be transcribed in absence of RA (Simeone et al., 1991). Similar results were obtained with a Hoxd-13/lacZ fusion gene (T. K. and D. D, unpublished) and may be explained by expression levels undetectable by this technology.

We used the β-gal activity of the targeted Hoxd-12 allele to visualize the expression of Hoxd-12. At both 9 and 15 days post coitum (p.c.), heterozygous and homozygous animals derived from all three clones were stained. All mouse lines showed identical staining patterns. This indicated that the function of the Hoxd-12 promoter was not modified by the presence of the PGK promoter. Secondly, as no qualitative staining difference was observed between heterozygous and homozygous animals, we concluded that autoregulation of Hoxd-12 does not have an essential role in the expression of the gene. This did not exclude, however, the possibility of cross-regulation by other paralogous or non-paralogous Hox genes, such as Hoxc-12 or group 11 and/or 13 genes.

The earliest detection of β-gal activity was at mid day 9 p.c. Embryos of this age showed a restricted expression in the area of the proctodeum, the region from which the anus subsequently forms (Fig. 2A, arrow). The staining was visible along each side of the proctodeal membrane, extending towards the future base of the hindlimb buds. At 10.5 days p.c., the staining became visible in the posterior region of the forelimb buds (Fig. 2D) as well as in more dorsal parts of the tail (Fig. 2B,D). In 11.5 days p.c. embryos, expression in limbs evolved with the appearance of a second domain covering the presumptive digit area (Fig. 2C,D). This distal domain was similar to that observed for Hoxd-13 (Dollé et al., 1993) and Hoxd-11 (Sordino et al., 1995) and likely originated at the posterodistal margin of the developing bud (Duboule, 1994). Interestingly, however, this Hoxd-12 digit domain was slightly smaller than that of Hoxd-13 and did not involve the future thumb, as previously described for birds (Fig. 2; Yokouchi et al., 1991). At this stage, weak expression was detected in the zeugopodium region (Fig. 2D). In older animals, the staining became more restricted. While the distal tips of the developing digits and the zeugopodium were negative in day 15 p.c. fetuses, strong expression was observed at the levels of both the first phalangeal (P1) and the distal part of the metacarpal chondrogenic condensations (Fig. 2D). Staining did not extend into the most anterior handplate, in contrast to both Hoxd-11 and Hoxd-13 (Dollé et al., 1993).

Hoxd-12 was also expressed in the genital eminence (Fig. 2C). After a rather homogenous expression at day 12 p.c., 14and 15-day-old fetuses showed strong staining in the future corpus cavernosum of the penis (Fig. 2E). The outer layer of the bladder, especially at the junction with the penian urethra, was positive as well as the presumptive sites of origin of accessory glands such as the prostate, at the level of the urogenital sinus. The mesodermal layer of the rectum, from the anorectal junction up to the first few turns of the hindgut expressed Hoxd-12 (Fig. 2E). In contrast, Hoxd-13 was transcribed both in the rectal epithelium and mesodermal derivatives (Fig. 2F,G), until adulthood (not shown). While both Hoxd-12 and Hoxd-13 were expressed in the internal and external muscle layers, their anterior boundaries of expression in hindgut were not precisely established. Neither Hoxd-12 nor Hoxd-13 were expressed in the anal epithelium (Fig. 2G; data not shown).

Mice homozygous for the Hoxd-12 mutation appeared normal in all respects. No particular phenotype could distinguish them from wild-type littermates and a detailed macroscopic observation and anatomical dissections did not reveal any obvious alteration of structures or internal organs which expressed this gene during development. Upon microscopic analysis, however, a consistent histological defect was uncovered in the rectum of both male and female homozygous animals.

The terminal part of the large intestine, the rectum, contains two well-defined subserosal smooth muscles layers, the inner, circular muscle layer (cm) and the outer, longitudinal muscle layer (lm; Fig. 3A). While the outer muscle layer is rather thin anteriorly, in the colon, it becomes almost as thick as the inner muscle layer in the terminal part of the rectum (Fig. 3A,B). In its terminal part, at the level of the anorectal transition, the inner layer extends slightly more distal than the outer layer and thickens to form a circular ring at the basis of which the outer layer comes to attach. This thick muscular structure, the internal anal sphincter, delimits the end of the rectum. This autonomous sphincter is surrounded by an external sphincter, composed of striated muscle masses, which do not express either Hoxd-12 or Hoxd-13. Smooth muscle layers of the rectum are particularly important as they receive enteric innervation responsible for the constipation reflex. As shown in Fig. 2G, both smooth muscle layers, including the internal sphincter, expressed the Hoxd-12 and Hoxd-13 genes, from mid-gestation stages up to adulthood.

Histological analysis of homozygous Hoxd-12 mutant mice showed a marked disorganization of rectal muscle layers in the region of the sphincter (Fig. 3C-E). Most frequently, the outer muscle layer was abnormally thin, or had a degenerated aspect (Fig. 3C). Often, it was interrupted at many places, so that isolated blocks of muscles cells were found scattered in the most distal region of the rectum (Fig. 3E). In addition, the muscle layers of Hoxd-12-deficient mice did not properly stop at the anorectal transition. Instead of forming the internal sphincter, muscle cells extended to a more distal position, such that the circular layer often overlapped with the position of the anal epithelium (e.g. Fig. 3C, left), in contrast to the very regular organisation of this terminal structure in wild-type mice. In each observed specimen, the sphincter was ill-formed and this alteration allowed us to identify mutants animals on histological sections independently from genotyping.

Hoxd-13-deficient animals showed a similar, though stronger, rectum defect (Fig. 4). As for Hoxd-12-deficient animals, the longitudinal muscle layer was abnormally thin (Fig. 4A) and, sometimes, locally absent (Fig. 4E, right side). The usual thickening of this muscle layer in the terminal part of the rectum (Fig. 4C) was never clearly detected in these mutant mice (Fig. 4D-F). This severe defect was observed from the very beginning of gut muscle differentiation, as newborn animals already exhibited some histological alterations (data not shown). 14 days after birth, Hoxd-13 mutant animals showed the same defect as in adults (Fig. 4E), indicating that the cause was likely a morphogenetic problem rather than a late degenerescence. At this age, while the two smooth muscle layers were already well apparent in wild-type littermates, mutant animals had disrupted muscle blocks (Fig. 4E) which did not reach the distal end of the circular layer, thus preventing formation of a sphincter. In some cases, the longitudinal layer was spectacularly absent (Fig. 4E). In other Hoxd-13deficient animals, the terminal part of the smooth muscle layer turned towards the external sides leading to a kind of ‘internally prolapsed’ muscle (Fig. 4A,D).

Upon dissection, the anus of mutant animals appeared abnormally opened and relaxed, with a funnel-like shape, probably due to insufficient muscular tonus in the sphincter. This was also visible on histological sections, where the muscle masses of the circular layer adopted an open, evaginated position (Fig. 4D,F,G). In such animals, the internal anal sphincter was virtually absent and this situation eventually led to an anal prolapse (Fig. 4B). More than 30 percent of the males after the age of 18 months developed this severe pathology (Fig. 4B,G). Interestingly, this was rarely observed in females, even though histology revealed identical alterations of the sphincter.

Anal prolapses can result from a variety of causes and may occur spontaneously in most animal colonies, for example due to the presence of intestinal parasites. While we sporadically observed such cases, the observed frequency of prolapse amongst the Hoxd-13 mutant mice, as well as the quasi absence of such manifestations in mice from other genotypes, makes an epigenetic etiology very unlikely. In addition, histological sections of the prolapsed material revealed that the external (longitudinal) smooth muscle layer was absent (Fig. 4G,H). In contrast, histological analyses of an anal prolapse that occurred in an unrelated mouse showed the presence of normal muscle layers (not shown). In the case of Hoxd-13, heterozygous animals already showed a tendency for disorganization of the sphincter and, in a single case, a prolapse was seen. As the Hoxd-12 mutation was produced more recently, we cannot exclude that anal prolapses will appear in older animals. Using different types of colorations (for enterocytes and goblet cells), we did not detect any alterations in the rectal epithelium of Hoxd-13 mutant mice, even though this gene was strongly expressed there (Fig. 4F). In summary, the Hoxd-13 phenotype was comparable, though stronger, to that of Hoxd-12.

Hoxd-12 mutant animals were analysed for their limb skeletal patterns. Surprisingly, no obvious defect was scored. However, upon direct measurement of the length of the skeletal elements, homozygous animals exhibited a minor reduction of the second phalange (P2) of digit V (Fig. 5A). As such a reduction of P2 was reminiscent of the Hoxd-13 phenotype, and since the two genes are co-expressed in digits, we produced Hoxd-12/Hoxd13 trans-heterozygous animals. Trans-heterozygous intercross generated progeny that had segregated the various alleles so that phenotypes could be compared between littermates. Animals homozygous for the Hoxd-13 mutation showed a strong digit phenotype, as previously reported (Dollé et al., 1993). Hoxd-13 heterozygous mice had no significant defects except that about half of them had a sign of postminimus. In Hoxd-12/Hoxd-13 trans-heterozygote animals, reduction of digits V and II was readily observed in forelimbs (Fig. 5), together with an occurrence of postminimus in 80% of the animals (not shown). The reduction in digit size was most prominent in the second phalanges, as expected for a Hoxd-13derived phenotype, confirming a recent study from Davis and Capecchi (1996). This result indicated that Hoxd-12 can have a genuine function during limb development. However, this function is largely masked in the presence of a full complement of Hoxd-13 product.

We have examined the function of the Hoxd-12 gene by targeted mutagenesis. Introduction of the lacZ gene in the Hoxd-12 reading frame allowed us to reexamine its expression pattern during development. While only a minor phenotype was detected in the skeleton, a fully penetrant malformation appeared in the anal sphincter. A related defect was uncovered in mice mutant for the 5′ neighbouring gene and most posterior one of the HoxD complex, Hoxd-13.

In several instances, Hox genes have been reported to be expressed during gut development (e.g. Dollé et al., 1991). More recently, comparative analyses have suggested that they may contribute to the establishment of morphological transitions in the intestine through the colinear distribution of their products (Yokouchi et al., 1995; Roberts et al., 1995), similarly to segmented mesodermal derivatives such as vertebrae. In such a view, the most posterior Hox genes of the HoxD complex were expected to be involved in the functional organization of the most distal hindgut structure, i.e. the anal sphincter. We show here that both Hoxd-12 and Hoxd-13 are required for proper morphogenesis of this terminal structure.

The rectal musculosa is characterized by a marked thickening of the external layer, while the internal layer remains more or less of the same width. At the distal end, however, a conspicuous thickening of the internal layer forms the internal anal sphincter, at the base of which it comes into contact with the external layer. In both Hoxd-12 and Hoxd-13 mutant mice, the thickening of the external layer failed to occur. As a consequence, the muscle layers became disorganized and no efficient sphincter formed. This absence of thickening led to a structural aspect of the rectal muscle layers which resembled more anterior parts of the large intestine, the sigmoid for example, where no thick external layer is present. In this view, it seems that both Hoxd-12 and Hoxd-13 are required to build up the terminal smooth muscle structure, on the top of a ‘hindgut-like’ generic muscle distribution. The function of both genes may be to increase the thickness of smooth muscles, leading to the formation of a sphincter. Consequently, the absence of this ‘terminal’ signal results in the distal extension of the thin muscle layers. No alteration was observed in more anterior parts of the intestine.

In aged Hoxd-13 mutant males, the disorganization of the internal anal sphincter led to an evagination of both the anus and the terminal part of the rectum. While the cause(s) of such anal prolapses are difficult to establish, it likely resulted from an overall weakness in the muscular tonus in the anorectal region. This weakness could result from both the disorganization of the muscle mass and the consequent problem in the innervation by the enteric system controlling the constipation reflex. In the prolapsed material, the rectal epithelium appeared hyperplastic and somewhat disorganized. However, this was probably a secondary effect as both the number and histological aspect of epithelial cells in the rectum, in similarly aged but not yet prolapsed mutants, was apparently normal. While females showed the same alterations of the internal sphincter, they did not develop anal prolapses, indicating that, for some as yet unknown reasons, males are more susceptible than females to this alteration.

β-gal staining of embryos and fetuses gave a precise account of the Hoxd-12 expression pattern during limb development. The dynamics of the transcript domain was very similar to that observed for both Hoxd-13 (Dollé et al., 1993) and Hoxd-11 (Sordino et al., 1995). An initial posterior (post-axial) domain extended distally and anteriorly in subsequent developmental stages, to finally cover most of the distal autopodium. The similarity between the three Hox digit domains (d-11 to d-13), likely due to the use of one and the same control mechanism (van der Hoeven et al., 1996b), suggested that the three genes may collaborate in digit morphogenesis. However, in contrast to Hoxd-13, but similarly to Hoxd-11, Hoxd-12 mutant animals showed only minor alterations in the skeleton of digits. This may have different causes. Firstly, Hoxd-12 may not exert a function in digits and may be restricted to more proximal regions (Davis et al., 1995; Davis and Capecchi, 1996). Alternatively, Hoxd-12 may function in digits but in a way somehow redundant with Hoxd-13 so that the latter gene could ‘rescue’ or compensate for the absence of Hoxd-12 function. Finally, Hoxd-12 may have a function in digits only in absence of Hoxd-13 product, consistent with a strict interpretation of posterior prevalence (Duboule and Morata, 1994). In such a view, Hoxd-12 would be dispensable in digits of normal animals, but may exhibit functional capacities whenever the level of the Hoxd-13 (or Hoxa-13, or both) gene product is reduced.

Trans-heterozygous Hoxd-12/Hoxd-13 animals did show some alterations in the digits, thus ruling out the first possibility. However, this result did not allow us to discriminate between the second and third alternatives. Therefore, in contrast to the original proposal of Davis et al. (1995), Hoxd12 does have an apparent function in digits which became visible when one dose of Hoxd-13 was removed. This confirmed that AbdB-related genes functionally cooperate during the morphogenesis of digits (Sordino et al., 1995; Davis and Capecchi, 1996). These results also suggested that Hox gene loss-of-function experiments could lead to the functional ‘activation’ of other Hox genes in some areas where their functions would otherwise be suppressed. We believe that this point is important to consider, as apparent loss-of-function phenotypes may be partly contributed by artificial gain-offunction of other Hox genes. For example, it is possible that part of the described Hoxd-13 phenotype (Dollé et al., 1993) may result from a functional (not transcriptional) gain-offunction of the otherwise suppressed Hoxd-12 function. The prevalence (or stronger functional impact) of Hoxd-13 over Hoxd-12 in limbs is also evident in the gut, where the Hoxd13^−/− phenotype was clearly more severe than that of Hoxd-12. Inactivations of several such genes at once will be informative in this respect.

Based on comparison of expression patterns, it was proposed that vertebrate Hox genes exert their functions within a morphological window that extends from the anterior to the posterior parts of the digestive tract, i.e. from the level of the foregut pocket, in the branchial region, to the proctodeal area (van der Hoeven et al., 1996a). One argument in support of this hypothesis is the systematic expression, in various metazoans, of the most posterior Hox genes (AbdB-related) in the region of the future anus or in the terminal part of the hindgut (with the exception of shrimps, where transcripts of the AbdB cognate were not reported in the analia; Averof and Akam, 1995). This suggested that the terminal Hox genes are involved in the morphogenesis of the distal part of the digestive system. The results presented here support this proposal, as they show that both Hoxd-12 and Hoxd-13 are required for the proper making of the anal sphincter. The question remains as to how ancestral this ‘gut’ function is? It is conceivable that one of the most ancient signs of a longitudinal axial structure in a metazoan body plan was linked to the design of a genuine alimentary tract, i.e. a region of transit demarcated, at both extremities, by specialized structures. It is therefore possible that colinearity in Hox gene expression co-evolved with the emergence of such a structure. In this view, the digestive tract is considered as the original site of deployment of Hox genes, this genetic system being recruited subsequently in neural and mesodermal derivatives.

Although such evolutionary arguments are, by definition, difficult to substantiate, two pieces of indirect evidence deserve to be discussed in this context: firstly, the search for terminal Hox genes in ‘primitive’ animals and secondly, the expression patterns of posterior Hox genes, as viewed from an phylogenetic perspective. In recent years, Hox genes have been extensively looked for in essentially all kind of animals, from the diblastic hydra (e.g. Schummer et al., 1992) to planarians (Balavoine and Telford, 1995), nematodes (Kenyon and Wang, 1991), crustaceans (Averof and Akam, 1995) and a variety of other animals. In all cases, genes have been isolated belonging to the major (ancestral) groups of paralogy. As far as AbdBrelated genes are concerned, however, the situation is less clear since this class of gene has not yet been found, either in hydra, or in planarians. One cannot exclude that this is due to important divergences in DNA sequences (such genes may be soon isolated), but it is intriguing that neither cnidarians, nor planarians have a bipolarity in their ‘alimentary tracts’ which, in both cases, works through a unique syphoon or pharynx. It thus appears conceivable that an AbdB ancestral gene appeared together with the evolution of a ‘posterior’ opening in the digestive system.

The second, more complex, argument relies on the mechanistic bases of temporal colinearity. Hox genes are activated following a temporal sequence that reflects their positions along the complex (Duboule, 1994). Experimental Hox gene transpositions at an ectopic position within the HoxD complex have revealed that early active genes are repressed until late stages, when placed at a more posterior position (van der Hoeven et al., 1996b). This latter set of experiments supports a view in which posterior (late) genes are ‘closed for business’ early on, for instance through a mechanism relying on a high order regulatory control such as a progressive transition in the structure of chromatin. As a consequence, vertebrate Hox genes may have been refractory to an evolutionary recruitment for additional functions earlier on during development, as this would have implied a premature opening of the complex, a situation probably detrimental to the animal. Therefore, the earliest sign of expression of a given Hox gene (in animals where temporal colinearity applies) may indicate its most ancestral functionality. In the case of Hoxd-12, transcripts are first visible in the proctodeal region. We do not wish to give a strict Haeckelian interpretation to this reasoning, i.e. to imply that the order of appearance of transcript domains during development would reflect the phylogeny of the functions, since, for instance, some functions may have been subsequently lost. Nevertheless, the initial site of expression during ontogeny may well represent the ancestral function, for it is unlikely that the function for which the system of colinear activation would have originally evolved (whatever it is) be lost in subsequent evolutionary steps.

We would like to thank Drs R. Kemler, A. Smith, P. Kastner and K. Rajewski for kindly providing us with the D3 ES cells, the LIF expression vector, a loxP-PGKneo-loxP cassette and the Cre recombinase expression vector, respectively. We thank M. Friedli, N. Fraudeau and V. Fraulob for their help with cell culture, histology and in situ hybridization. J. Z. is on leave from the Institute of Genetics (BRC) of the Hungarian Academy of Sciences, Szeged. T. K. was supported by a grant from the Uehara Memorial Foundation (Japan). This work was supported by funds from the Swiss National Research Fund, the Canton de Genève, the Claraz Foundation, the Latsis Foundation and the HFSPO.