Homeotic genes and the control of segment diversity

The bodies of insects, and of most other arthropods, are visibly subdivided into distinct regions, or tagmata (Snodgrass, 1935). In insects, the organization of segments within tagmata is very constant. There are three gnathal or mouthpart segments (often incorporated into the head), three thoracic segments and an abdomen with a maximum of 11 segments (Fig. 1).

The tagmata reflect functional subdivisions of the body, but in varying degree they also reflect domains of the body where each of several segments follows a similar developmental pathway. In Drosophila, for example, the thoracic segments will each establish imaginal disc primordia – groups of cells destined to proliferate during the larval period – whereas abdominal segments establish histoblasts that will be quiescent until puparium formation (Bodenstein, 1950).

In Drosophila, and presumably in all insects, the differences between segments are controlled by homeotic genes (Lewis, 1963, 1978; see Mahaffey & Kaufman, 1987). We might therefore expect that these major morphological divisions of the body would be reflected in the domains of action of homeotic genes. Little trace of this is evident, however, in the descriptions of the domains of homeotic gene activity defined by genetic analysis (Lewis, 1978; Sanchez-Herrero et al. 1985). In the bithorax complex, for example, the gene Ultrabithorax (Ubx) is assigned a role controlling both thorax and abdomen, whereas the very similar abdominal segments A4 and A5 are assigned to the control of different homeotic genes (Fig. 1).

These genetic subdivisions of the fly are clearly correct – the Abdominal-B gene (Abd-B) does play a visible role in the development of A5 but not A4. However, we believe that they are misleading in our attempts to understand the control of development. Analysis of the deployment of homeotic genes during development (for review, see Akam, 1987) suggests that a single gene may play qualitatively different roles in different segments. Typically, each homeotic gene is activated ‘metamerically’ in certain segments (or more precisely, parasegments (Martinez-Arias & Lawrence, 1985)). By this, we mean that it is activated at the blastoderm stage, in all or virtually all cells of the parasegment primordium. In these parasegments the activity of the gene is probably required to establish the basic characteristics of the segment. In other segments, the same homeotic gene may be deployed adventitiously, only in subsets of cells and at later stages of development. In these segments, the gene must play a secondary role in defining developmental pathways.

We argue below that the early ‘metameric’ domains of gene expression do correlate with the major morphological subdivisions of the fly, suggesting a relatively direct relationship between the expression of particular homeotic genes and the decisions taken by individual cells in early development. This relationship allows us to suggest a scenario for the evolution of homeotic genes in relation to the evolving morphological organization of the arthropod body plan in the insect myriapod lineage. To approach this question, we first review how the molecular analysis of homeotic gene expression has altered our understanding of homeotic gene function.

Current models relating the expression of homeotic genes to the control of segment development are based on the pioneering work of Lewis (1963, 1978, 1981), and relate primarily to the role of genes in the bithorax complex in the control of development of the posterior thorax and abdomen.

Lewis suggested that the bithorax complex contained a series of genes, each of which would be turned on at a different position along the anteroposterior axis of the fly, in partially overlapping domains (Lewis, 1963, 1978). These partially overlapping domains of gene expression would give each segment a different combination of active homeotic genes, and this combination would define segment identity.

This model has been integrated with the idea of ‘selector’ genes, or genes serving to provide a genetic address for a polyclonal group of cells (Garcia-Bellido, 1975). In this form (Struhl, 1982; Lawrence & Morata, 1983) the model focuses on the assumption that within a domain – be it segment (Lewis, 1978), parasegment (Martinez-Arias & Lawrence, 1985) or compartment (Garcia-Bellido, 1975) – every cell takes the same decision about which homeotic gene(s) to turn on at an early stage of development, and remembers that decision through all subsequent cell divisions.

For example, all cells of T2 (parasegment 4) were assumed to express Antennapedia (Antp) alone, whereas all cells of T1 (parasegment 3) would express both Antp and Sex combs reduced (Scr), and all cells of T3 (parasegment 5) would express Antp and Ubx (Fig. 2). By extension, each segment of the embryo would express a unique combination of homeotic genes that constituted a specific codeword, defining segment identity.

Three aspects of the available molecular data suggest that this model needs to be revised. First, there are not enough genes, and probably not enough different gene products, to give a unique combination of products in every segment. All functions of the bithorax complex depend on the activity of just three homeotic genes – Ubx, abdominal-A (abd-A) and Abd-B (Sanchez-Herrero et al. 1985; for review see Duncan, 1987). Together these control the different identities of at least nine parasegments.

A more serious revision is required by the observation that, at the level of protein expression, the cells of a segment do not all express the same combination of homeotic genes. In the blastoderm and very early gastrula, parasegments do appear as domains of uniform homeotic gene expression (Akam & Martinez-Arias, 1985; Martinez-Arias, 1986), but once a pattern of different cell types has been generated within each segment, most segments become mosaics of cells expressing different homeotic proteins (White & Wilcox, 1985; Carroll et al. 1988) and, indeed, by late stages of embryogenesis, most individual cells express only a single homeotic protein at high level, as the expression of one protein represses the transcription of other homeotic genes in a fixed hierarchy (Hafen et al. 1984; Struhl & White, 1985).

Finally, certain cells switch from expressing one homeotic gene in very early development to expressing another, or none at all, in later stages of development, thus confounding the strict lineage model of homeotic gene expression. One clear example of this is seen in the ectoderm of parasegment 3, which expresses only Antp protein in the very early extended germ band, but then expresses Scr at later stages (Martinez-Arias et al. 1987; Carroll et al. 1988).

Many of these data can be reconciled with Lewis’ model, if for the idea of ‘turning on a gene’ we substitute the phrase ‘make accessible an enhancer’ or region of DNA containing many independent enhancers. This is essentially the ‘open-for-business’ model developed by Welcome Bender and his colleagues (Peifer et al. 1988).

This model is based on the observation that, although there appear to be only three protein-coding ‘homeotic genes’ in the bithorax complex, each of these genes is controlled by multiple, independent regulatory regions (Bender et al. 1983; Karch et al.1985). Thus the Ubx gene controlled by the bithorax (bx) regulatory region in parasegment (PS) 5, and the bithoraxoid (bxd) regulatory region in PS6 (Fig. 3). Similarly the abd-A and Abd-B genes are controlled by a series of ‘infra-abdominal’ (iab) regulatory regions – approximately one per segment (or parasegment; see Duncan, 1987).

The essence of the open-for-business model is that each cell decides, early in development, which regulatory elements or regulatory domains are ‘open for business’ according to position in the embryo. Each regulatory domain contains elements that interact independently with trans-acting factors to control the temporal and cell-specific activity of each homeotic gene. So for example, the ‘bx’ regulators are ‘open for business’ in PS5, and are the only elements to control Ubx expression in this parasegment. The ‘bxd’ regulators are ‘open for business’ in PS6 and more posteriorly, and act, perhaps with the PS5 regulators, to control Ubx expression in more posterior regions (Peifer et al. 1987).

While this model conforms to the formal structure of Lewis’ model, its implications for the control of development are very different from those of the combinatorial selector gene model. For example, in PS6, Ubx protein is expressed in all or virtually all cells from shortly after the time of gastrulation (White & Wilcox, 1985). Thus, Ubx can play a role in early development throughout this parasegment. However, in PS5 the regulators of Ubx are ‘off’ at very early stages of development. Ubx expression first appears in a limited number of cells around the forming tracheal pits. Later, it appears in the nervous system, though at the extended germ band stage the precursors for the neural cells of PS5 show no trace of Ubx transcription (Akam & Martinez-Arias, 1985; White & Wilcox, 1985).

The deployment of Abd-B illustrates the difference between these two models particularly clearly. Abd-B gives rise to at least two distinct sets of transcripts, probably from multiple promoters (Sanchez-Herrero & Crosby, 1988; DeLorenzi et al. 1988). One set are present only in the extreme posterior (PS 14–15) under the control of the iab-8,9 regulators, whereas the other transcripts are expressed more anteriorly, under the control of different regulators, some of which (iab 5,6,7) lie 3′ to the Abd-B gene (Fig. 3).

In early development, in the blastoderm and early extended germ band, Abd-B transcripts are expressed only in the extreme posterior of the presumptive abdomen, from parasegments 13–14. As development proceeds, detectable levels of transcripts appear in PS12, then 11, then 10 (Sanchez-Herrero & Crosby, 1988). Although the iab 5,6,7 regulatory elements may perhaps be ‘open for business’ from blastoderm stages onwards in PS10 to 12, the Abd-B gene can play no role in controlling the development of these segments until a later stage of development. Thus the developmental role of Abd-B in PS 10–12, like that of Ubx in PS5, must be more limited than the role of these same genes in their ‘metameric’ domains – PS6 for Ubx; PS13–14 for Abd-B.

If we consider, not the entire domain of action of each homeotic gene, but rather these metameric domains, then we see a closer relationship between homeotic gene expression and segment development. This is illustrated in Fig. 4, which summarizes homeotic gene expression in the ectoderm when the germ band first forms. To make this correlation clear we must look in more detail at the morphology of the Diptera.

The tagmata of the insect body cannot simply be equated with groups of segments sharing common developmental origins. The gnathal segments, for example, probably did not evolve by the diversification of an archetypal gnathal segment. Comparative morphology suggests rather that they evolved successively, presumably by the independent evolution of mechanisms to modify the most anterior remaining trunk segments (Anderson, 1973).

There are better grounds for believing that the ancestors of modern insects possessed multiple, near identical, thoracic and abdominal segments, as such regions are still present in the trunk of myriapods, and in the abdomen of many insects. The diversity of segments within these tagmata may therefore be regarded as deriving secondarily through the diversification of generalized thoracic and abdominal progenitors.

Using the term tagma in this developmental and evolutionary sense, we would limit the abdomen to include only the progenital abdominal segments, A1–A7 (Snodgrass, 1935; Matsuda, 1975). Segments of the posterior abdomen (PS13–15, A8–A11) or tail (Jurgens, 1987) differ from the anterior abdominal segments, and from one another, in many respects, and give rise to the specialized reproductive and cereal appendages. Although functionally part of the abdomen, there are few grounds for regarding these tail segments as derived by modification of a segment resembling those in the preabdomen. Differentiated genital and terminal segments exist in the myriapods, so the equivalent structures of insects are more likely to derive from independent modification of an unspecialized trunk segment.

Within the abdomen, the first segment (Al) is generally unique, and clearly distinguishable even in early embryonic stages from A2–A7 (Anderson, 1972). These remaining segments of the preabdomen are strikingly similar during early embryogenesis. Their similarity is retained even by the adults of a typical lower dipteran (e.g. Trichocera, Fig. 5), but in Drosophila and other higher Diptera, segments A5–A7 of the adult have become specialized (McAlpine et al. 1981). None the less, in the Drosophila embryo, segments A2–A7 remain remarkably similar. Even in such complex patterns as the somatic musculature (Hooper, 1986) and the peripheral nervous system (Ghysen et al. 1986; Dambly-Chaudiere & Ghysen, 1986), these segments are indistinguishable.

These morphological domains are clearly correlated with the metameric domains of homeotic gene expression established by early germ band stages. For example, Ubx and/or abd-A are expressed throughout PS6–12, which is the parasegmental equivalent of the preabdomen. The early domain of Abd-B expression is limited exclusively to the postabdomen (Fig. 4).

Ubx alone is expressed in PS6 (Al), but both Ubx and abd-A are expressed in the more posterior parts of the preabdomen. Where their expression overlaps, at least some of their functions are redundant. For example, the activity of either gene will result in normal development of the tracheal system (Lewis, 1981) and of certain characteristic structures of the nervous system (Ghysen & Lewis, 1986) in the abdominal segments. These two genes are the most similar of the Drosophila homeotic genes in the sequence of their homeoboxes (Fig. 6). Most remarkably, a mutation that fuses the amino terminus of the abd-A protein-coding region with the homeobox and carboxy terminus of Ubx allows expression of a functional product that rescues many aspects of both the Ubx and abd-A mutant phenotypes (Rowe & Akam, 1988). All of these observations suggest that both of these genes are able to lay down the ground plan for the preabdominal segments – albeit each is ‘fine tuned’ to make different structures.

The equivalent early role of Abd-B is probably to define the genital and postgenital regions of the abdomen, but development of this region depends also on the activity of another homeobox-containing gene, caudal (Macdonald & Struhl, 1986; Mlodzik & Gehring, 1987). The early expression domains of these two genes may overlap, though in later development they appear to be expressed in different structures (Sanchez-Herrero & Crosby, 1988), with caudal required for the development of the most posterior segmental structures (anal pads, A10 or All).

In the more anterior regions, Deformed (Dfd) is expressed metamerically in and anterior to parasegment 1, while Scr is expressed initially throughout parasegment 2 (Martinez-Arias et al. 1987; Chadwick & McGinnis, 1987; Riley et al. 1987). These genes together specify the development of the gnathal segments. Antp is expressed metamerically in parasegment 4, and slightly later throughout the ectoderm of parasegments 3 and 5, thus extending throughout the thoracic parasegments (Martinez-Arias, 1986 and personal communication). However, the conceptual ‘metameric’ expression of Antp is fleeting because, in PS5, limited expression of Ubx appears shortly after germ band extension and, in PS3, the expression of Antp is ‘overwritten’ by Scr in the epidermis at about the same time (Carroll et al. 1988).

The correlation presented above ignores the inevitable overlaps that arise in aligning tagma, defined in segmental terms, with the early parasegmental domains of homeotic gene expression. Effectively, it assigns the identity of each parasegment to the tagma containing its A compartment. There is some justification for this. The primordia for the P compartments, defined by engrailed expression, include only about a quarter of the cells in each parasegment (DiNardo et al. 1985), and include few of the specific pattern elements on the trunk (though more on the appendages). The mesoderm of each segment derives entirely from the same parasegment as its A compartment (Lawrence, 1985). Moreover, in cases where a shift occurs from parasegmental to segmental expression of a homeotic gene (as for Dfd and Scr in the gnathal buds (Martinez-Arias et al. 1987; Jack et al. 1988)) it is the P compartment that switches to match expression of the anteriorly adjacent A compartment.

The relationship presented here suggests that the expression of homeotic genes in the very early germ band may directly control developmental decisions that establish what is best-called ‘tagmatic identity’ for each parasegment; not by a complicated decoding process, but as a relatively direct consequence of the homeotic gene first expressed. Once this decision has been taken, changing patterns of homeotic gene expression modify the development of segments within tagmata.

We can relate these suggestions to the evolution of segment diversity by making two assumptions. First, we follow Lewis (1978) in assuming that the evolution of the homeotic gene family parallels, and in some degree directs, the evolution of segment diversity in the myriapod – insect lineage. Secondly, we suggest, in what is essentially a molecular version of Von Baer’s rule (see Gould, 1977), that the patterns of gene deployment seen in early development – or more properly in the earliest phylotypic stage (Sander, 1983) – are likely to represent phylogenically ancient roles for the gene. Conversely, patterns of deployment that first appear in later development are more likely to represent recent uses of the gene.

Although the homeobox was first characterized in homeotic and segmentation genes, it is becoming clear that proteins containing this structural motif are involved in a very wide range of developmental processes, in both segmented and nonsegmented organisms (e.g. Saint et al. 1988; Way & Chalfie, 1988).

On the basis of gene structure, this diverse family falls into a number of more or less well-defined subfamilies. One of these, the Antennapedia-like subfamily, includes most of the homeotic, segment-selector genes. Antp, Ubx,abd-A and Scr share particularly similar homeobox sequences (Fig. 6), and also share short conserved peptide motifs in other regions of the protein.

Genes sharing all of these structural features are also found in vertebrates (Krumlauf et al. 1987), indicating that this Anip-like subfamily arose before the divergence of the major animal phyla. At least within the homeobox, the sequence of the modern Antp gene appears to be very close to the ancestral sequence for this subfamily (Gehring, 1986).

The sequence of the Dfd homeobox is somewhat more diverged from the Antp-like concensus, although Dfd shares with the Antp-like genes at least one of the conserved peptides outside of the homeobox. However, homeotic genes that are clearly related to Dfd rather than Antp are found in vertebrates, indicating that the distinction between Dfd-like and Amp-like genes predates the vertebrate-arthropod divergence (Regulski et al. 1987).

The same is probably true for the Abd-B gene of the bithorax complex, although no very close homologues of Abd-B have yet been identified in vertebrates. The sequence of the Abd-B gene places it outside the close Antp family; its homeobox sequence is no more closely related to Antp than are the Drosophila genes caudal and labial, and, of these, at least labial is represented in vertebrates by distinct and well-conserved homeobox homologue (Mlodzik et al. 1988).

In Drosophila, Abd-B, caudal and labial are all expressed in defined spatial domains either at the ends of, or beyond the limits of, the metameric region (Hoey et al. 1986; Akam, 1987). In vertebrates, the homeobox genes are also expressed in different regions along the A-P axis of the animal, suggesting that their role in regionalization may be a very ancient one (Holland & Hogan, 1988).

These data suggest that the earliest ancestor we can reasonably imagine for the myriapod–insect lineage already possesed distinct homeobox genes related to labial, Deformed, Antp and probably also Abd-B and/or caudal (Fig. 7; see also Martinez-Arias, 1987). We suggest that this animal utilized the Antp gene to control various aspects of the development of a region that has become the trunk in arthropods, and that this region of the body was already bounded by domains expressing homeobox genes related to Deformed and caudal. We would guess, however, that this common ancestor did not have specific homologues for the different genes of the Amp-like family that now specify the diversity of parasegments 2–12 in Drosophila.

In the line that led to the myriapods, a mechanism evolved that allowed the most anterior trunk segments to develop specialized mouthparts, quite different from the typical trunk segments. The deployment of homeotic genes in Drosophila suggests that this process utilized the Dfd gene, and may have been associated with the origin of an Scr-like gene, but it is not possible from what we know at present to guess when, or from what ancestral gene(s), Scr arose. At some stage after this, distinct thorax/abdomen tag-mosis evolved in the insect lineage. Our view of the role of homeotic genes leads us to suggest that a functional equivalent of Ubx or abd-A arose at this point.

With this array of homeotic genes, and with appropriate mechanisms to control segment number, our ancient ancestor could have been transformed into a primitive insect, with three similar thoracic segments and an array of similar preabdominal segments. To make a typical modern insect requires a sophisticated mechanism to make each segment different. Two rather early innovations may have been the use of Scr to make T1 different from the other thoracic segments, and also the origin of distinct Ubx and abdominal-A genes to make Al different from the rest of the preabdomen. The evolution of the higher Diptera, leading to Drosophila, may have been achieved without any additional homeotic genes, but by the evolution of more complex regulatory mechanisms to modulate the expression of existing genes.

The scenario present above points to mechanisms whereby the current elaboration of segment specialization may have arisen. We are suggesting that one of the more rapid ways a homeotic gene may mediate evolutionary change is by acquiring new regulatory signals that work on existing products, independent of, and in addition to, existing regulation. We may be seeing the results of such evolutionary changes in the patterns of deployment of Ubx in PS5, and Abd-B in PS10–12. It is perhaps worth noting that in these two cases the cis-acting regulatory elements for the ‘non-metameric’ expression of these genes are clearly distinct from the DNA regions required for the metameric domain of expression, and in both cases lie distant from (and 3′ to) the promoter (Fig. 3).

Clearly, the evolution of novel regulatory mechanisms can proceed with less constraint if gene duplication occurs, allowing closely related proteins that have extensively overlapping functions to come under totally independent regulation. To some extent the Ubx and abd-A genes of Drosophila may be viewed as being still in this state. However, once genes have duplicated, functional divergence can create products with different regulatory effects. Such processes presumably occurred in the origin of the abdominal-like genes from an Antp-like ancestor, in association with the origin of thorax-abdomen tagmosis.

The scheme presented above makes predictions about which organisms should share genes with specific Scr-like or Ubx-like properties, and what features of the expression of these genes we should expect to see conserved, which different. For instance, it ‘predicts’ that the bx functions of Ubx and the iab 5–7 functions of Abd-B may not be conserved among all insects, whereas the early embryonic PS6–12 domains of Ubx, and PS 13–14 domains of Abd-B should be.

To test these predictions, we are cloning the homeotic genes of the Antp-like family from the locust Schistocerca gregaria, a representative of one of the older and embryologically more primitive groups of the Neoptera (I. Dawson, G. Tear, A. Martinez-Arias and M. Akam, in preparation).

The locust is a short-germ insect. That is to say, at the onset of gastrulation only the anterior part of the body plan is defined; posterior segments are generated by a growth process during embryogenesis. This difference raises questions concerning the establishment of segment pattern, intimately related to the problems of segment specification. These are discussed in the article by Tear, Bate and Martinez-Arias in this volume.

We have sequenced the homeobox regions of three homeotic genes from the locust. On the basis of sequence, two are clearly the locust homologues of specific Drosophila genes – Scr and abd-A. The homeoboxes are identical or almost identical (Fig. 8) and there is enough homology in the regions flanking the homeobox to identify the genes unambiguously, independent of the homeobox sequence itself.

We have only just started to analyse the pattern of expression of these genes. The most striking result is how similar are the patterns of expression between Drosophila and Schistocerca, at least in relatively late germ band stages. In both species, Scr is expressed in part of the mandibular segment, in the labium and in the anterior part of T1. The abd-A gene is expressed in the epidermis of the abdominal segments from posterior Al backwards (I. Dawson, G. Tear, A. Martinez-Arias and M. Akam, in preparation).

The third homeobox is much less easy to identify. Its sequence places it firmly among the Antp-like family, but it differs in nine amino acids from Antp and more from all other known Drosophila homeoboxes. At present, we know only that this gene is expressed in the nervous system, without obvious segment specificity; in this respect it most resembles the neural expression of fushi-tarazu, the only segmentation gene that shares an Antp-like homeobox. If a perfect homologue of this third homeobox existed in Drosophila, the sequence would almost certainly have been detected by hybridization with Antp probes. It is therefore likely that this gene represents either a function that is no longer present in the Drosophila genome, or a homeobox subject to less stringent evolutionary constraint than the abd-A and Scr genes.

It remains to be seen whether such changes in gene structure, and changes in the pattern of deployment of genes during development, can be correlated with the evolution of new developmental strategies.

We acknowledge many discussions with Alphonso Martinez-Arias that have contributed to the development of the ideas presented above. We thank Michael Bate for teaching us the embryology of locusts, and Adrian Friday for calculating the network illustrated in Fig. 6. The evolutionary speculation presented above was prompted by an invitation to speak on the topic of ‘Evolution and Development’ at the Markey Symposium held in honour of E. B. Lewis at Caltech in April, 1988. Experimental work from our Laboratory is supported by the Medical Research Council.