The molecular genetics of head development in Drosophila melanogaster

‘It would be too bad if the question of head segmentation ever should be finally settled; it has been for so long such fertile ground for theorizing that arthropodists would miss it as a field for mental exercise’. Snodgrass, 1960 (cited by Rempel, 1975)

The development of the Drosophila embryo has provided an excellent model system for the analysis of embryonic pattern formation. In little more than a decade, many of the elements of the molecular genetic cascade underlying early development have been identified. However, these studies have focused primarily on the development of the central, overtly segmented region of the fruitfly embryo. As the influential insect morphologist R. E. Snodgrass pointed out, understanding of the formation of the complex embryonic head region has advanced quite slowly. Recently, however, substantial progress has been made in the identification and analysis of the genes determining head development. These studies are generating increasing hope that both the structure of the head and how it is formed will soon be significantly clearer. What is perhaps most interesting is that there are already a number of hints that the rules governing head formation may differ from the paradigm established for the central region of the embryo.

In this review, we will focus on the process of embryonic head development in Drosophila melanogaster. First, we will describe the evolution of the head, its formation, and some of the difficulties involved in analyzing its structure. We will then discuss what has been learned about the genes responsible for embryonic head development. Finally, the implications of this molecular genetic data for models of head development will be explored.

The dipteran Drosophila embryo is one of the most highly evolved of the arthropod embryos. Arthropods, annelids, and other members of the articulate group (see Fig. 1) all show marked similarities in body plan. Most obviously, both arthropods and annelids are metameric in structure, consisting of a series of segmental units. An early definition of a metameric unit was proposed by Snodgrass (1935) as simply a body division of the embryo. This definition, however, was rather crude and was later supplanted by a series of more precise criteria. Rempel (1975), for example, summarized the attributes of a metamere as: (1) a pair of mesodermal somites (which give rise to the muscles) (2) a pair of appendages (3) a pair of apodemes (inner projections of the ectoderm which form muscle attachment sites) and (4) a neuromere (which produces a ganglion and its associated lateral nerves). Although these characteristics are easily recognizable in the metameres of most annelids, in many arthropod segments they are often less evident. In particular regions of arachnids and crustaceans, for example, metamerism is quite difficult to detect. This loss of segmental attributes (which as we shall see is an important issue in Drosophila head development) has occurred in several ways. In a number of cases, particular segments have been lost or fused together during evolution. In various species, segmental appendages are greatly reduced and apodemes and neuromeres difficult to discern.

In addition to a metameric body plan, both arthropods and annelids have other similarities in general body structure. Both groups have dorsal hearts and related nervous systems, consisting of an anterior, dorsally located brain and a ventral nerve cord formed of a series of ganglia. Because of these structural similarities, present day articulates are thought to have evolved from a primitive annelid-like organism (see Fig. 1 and, for example, Anderson, 1973). This hypothesized ancestor consisted of an array of metameric units, which showed little evidence of functional or structural specialization. These metameres, as in present-day annelids, were likely to have been generated sequentially from a central growth zone. The ends of this animal were primitive non-segmented structures. The anterior end was probably most similar to the anterior ‘prostomial’ region of modem annelids, housing a simple brain (archicerebrum). This ‘head’ area was involved in sensory and feeding functions. The posterior ‘tail’ region most likely consisted of a nonsegmented area surrounding the anal opening.

The evolutionary process of head formation involved the progressive incorporation of structures into the head region (see Fig. 2). In animals even more primitive than the annelids (e.g. the Platyhelminthes), there is already an increasing concentration of sensory and feeding structures (eyes, ganglia and tentacles) in the anterior region. A critical step in this process of ‘cephalization’ occurred when several of the anterior-most trunk segments became incorporated into the beginnings of a clearly recognizable head region. This process is already evident in certain annelid embryos (e.g. the polychaete Nereis) in which the head includes regions called the prostomium (anterior to the mouth opening) and the peristomium (surrounding the mouth). Although the precise origin of the peristomium is unclear, it is generally thought to include one or more former trunk segments. One piece of evidence for this is that the peristomium contains pairs of cirri, which are likely the remnants of the parapodia, the appendages present on annelid trunk metameres. Cephalization continued as evolution proceeded and is evident in present-day myriapods (centipedes and millipedes). The insect head, consisting of an asegmental terminal region and cephalized trunk segments, first appeared in primitive form during the Devonian period (Smart and Hughes, 1972). However, because of the scarcity of fossil material from this period, its structural form has been deduced largely through the comparative analysis of more recent species.

As cephalization occurred, several of the anterior-most trunk segments shifted forward, so that they came to lie in front of the mouth opening. These segments became increasingly specialized and capable of performing more ‘head specific’ roles. For example, the specialization and anterior position of the antennal segment allowed it to acquire an important sensory function. The preoral domain of the head is often referred to as the procephalon. It is in this region that segmental identities are most difficult to establish. Behind the mouth (postoral), is the section of the head known as the gnathocephalon. In this region, cephalized trunk segments have become adapted for feeding functions. As evolution proceeded, the remaining trunk segments became subdivided into thoracic and abdominal units with clearly distinct functional roles. Further regional specialization throughout the embryo led to the complex Drosophila body plan.

In the Drosophila embryo, part of this process of head specialization has been the evolution of a series of complex morphological movements which occur during hours 9 –12 of development. During this time window, a parallel series of complex events (which will not be discussed here) also occurs in the posterior anal region (Turner and Mahowald, 1977). Much of our understanding of the process of head formation described below stems from the detailed scanning electron microscopy (SEM) studies of Turner and Mahowald (1979). Initially, the ‘germ band’ of the embryo is fully extended and segmental furrows are clearly visible (stage 12; Campos-Ortega and Hartenstein, 1985). In the head region, a series of six lobes can be identified. These include the procephalic, clypeolabral and hypopharyngeal lobes as well as three gnathal lobes (mandibular, maxillary and labial). At this time, the germ band begins to retract and a series of rotations and fusions of the head lobes begins that culminates with the process of ‘head involution’. During head involution, the gnathal lobes move closer to the stomodeal opening at the anterior end of the embryo and eventually become largely internalized (see Fig. 3). As a result, the labial lobe becomes the floor of the mouth, the maxillary and mandibular lobes fuse and form the lateral sides of the mouth, and the internalized clypeolabrum forms the roof of the mouth. The completed head structures then retract beneath the first thoracic segment. This highly specialized process of head involution is a relatively recent evolutionary development. It results in the acephalic appearance of the Drosophila larva, a consequence of the fact that most of the head structures are brought to positions inside the anterior end of the embryo.

Following head involution, a layer of epidermal cells contained within the involuted head region secretes a variety of cuticular and sensory structures (see Fig. 4). These structures include various easily recognizable ‘skeletal’ elements (which occur mostly in pairs and constitute the cephalopharyngeal skeleton), as well as the mouth hooks, labrum, cirri and several segmentspecific sensory organs. In the head of the first instar larva, these structures can be relatively easily identified in cuticular preparations. This has two important implications First, it permits fate mapping studies to be performed in which regions of the early embryo are disrupted, and the subsequent effects on head development monitored. Second, the effects of mutations that perturb head development on the formation of head structures can be determined. Both these approaches have been critical to understanding the structure of the head domain.

What Rempel (1975) referred to as the ‘endless dispute’ about insect head development involves the determination of the number and nature of the segments included in the head. This problem has been approached using several different methodologies. Traditional analyses have involved the comparative morphological study of the heads of a wide variety of insects. Such investigations have produced numerous hypotheses to describe head segmentation (summarized in Rempel, 1975). There have been two critical points of disagreement in these models. The first issue is whether there is a domain of the embryonic head that is in fact unsegmented in structure. Although most current hypotheses propose that there is an asegmental region (referred to as the acron), there have been observers who maintain that the insect head is entirely metameric in composition. The second controversy, which is clearly related to the first, has involved the number of segments present in the head. This number has been estimated to be as low as three and as high as seven (e.g. Roonwal, 1938). There has been no disagreement about the gnathocephalon, which is composed of three easily visible segments. These gnathal subdivisions, the mandibular, maxillary and labial segments, clearly meet the requirements for metameric identity discussed in the previous section. In the procephalon, however, segmental identities are more difficult to establish. More recent morphological studies (Scholl, 1969; Rempel and Church, 1971) have proposed that the anterior region is composed of the asegmental acron and three distinct segments. From anterior to posterior, these include the labral, antennal and intercalary (premandibular) segments. However, establishing the metameric nature of the labral and intercalary segments in particular has been difficult. In addition, some observers have hypothesized that the posterior region of the acron is in fact an additional pre-antennal segment (e.g. Roonwal, 1938).

In Drosophila, the highly evolved form of the procephalon has made segmental identification particularly difficult by morphological methods. As a result, various other approaches have been utilized. By histological analysis of staged embryos, Poulson (1950) was able to assign various larval organs to specific positions on the blastoderm fate map. More recently, Campos-Ortega and colleagues (Hartenstein and Campos-Ortega, 1985; Technau and Campos-Ortega, 1985; Hartenstein et al. 1985) have refined this fate map by following individual cells injected with horseradish peroxidase (HRP). This technique has been particularly useful in determining the pattern of mitoses in various embryonic primordia, including the cephalic region. Specifically, HRP injection has allowed better definition of the boundaries of the gnathal segments and of the size of the blastoderm anlagen that give rise to them. In particular, it was shown that as one proceeds in the anterior direction, each gnathal segment is derived from an increasingly small region of the blastoderm fate map. In addition, a population of about 80 cells was defined as constituting the procephalic neurogenic region. This region shows a pattern of mitotic behavior independent of that of the more well-characterized ventral neurogenic domain. Although the procephalon was demonstrated to consist of several distinct groups of cells in this study, specific segmental identities were not proposed.

Another approach to the structural analysis of the head has been genetic. This method has relied on the analysis of marked clones of cells to deduce the regions of the blastoderm that give rise to adult, rather than embryonic head structures. The adult head of Drosophila is formed from three pairs of imaginal discs which develop during larval stages. Most head structures, including the head capsule, eyes, antennae and maxillary palp are derived from the eye-antennal discs (Bryant, 1978). The proboscis and the clypeolabrum are formed from the labial and cibarial discs respectively. By analyzing clones of cells in wild-type and mutant flies, Morata and Lawrence (1978, 1979) showed that the eye-antennal disc, like other imaginal discs, is divided into anterior and posterior compartments. This division was shown to require the activity of the engrailed gene. However, the relative positions of the anterior and posterior compartments of this disc were shown to be reversed, with the posterior compartment giving rise to the extreme anterior region of the adult head. Struhl (1981) extended this analysis using gynandromorph fate mapping. He was able to show that the eye-antennal disc, which is presumably derived from a single blastoderm (antennal) segment, becomes spatially reversed as a result of a rotation that occurs after the blastoderm stage. This gynandromorph analysis also yielded two important results relevant to head segmentation. The first is that the clypeolabrum of the adult head is derived from a region anterior to the antennal segment, supporting the existence of a labral segment. The second finding is that, assuming segments initially arise from equally sized regions of the blastoderm, there is sufficient ‘room’ on the fate map for three segments between the antennal and labial primordia. Therefore, this genetic analysis supports the existence of the intercalary segment, and is consistent with a six segment model of the insect head.

The final technique that has been used to study head structure is analyzing the effects of destroying cells at the blastoderm stage of development. This has been done mechanically (Bownes and Sang, 1974b; Underwood et al. 1980), by microcautery (Bownes and Sang, 1974a), and by UV-laser irradiation (Lohs-Schardin et al. 1979; Jurgens et al. 1986). We will discuss the work of Jurgens and colleagues, since it has produced the highest level of resolution. In these experiments, Drosophila embryos were irradiated at either the cellular blastoderm stage or at the end of the extended germ band stage (approximately 9h of development). In each case, irradiation was performed in small designated regions of the presumptive head region (the anterior 40% of the embryo). By scoring the cuticle defects that occurred in the heads of the larvae that developed from irradiated embryos, the regions of the blastoderm and the extended germ band embryo that give rise to particular,head structures could be determined. It should,be noted however, that in this-study, only epidermal derivatives of the head and not internal tissues were considered. The detailed blastoderm fate map derived from this work is shown in Fig. 4. The information in this fate map was combined with the results of irradiating regions within the morphologically recognizable head lobes in the extended germ band embryo. It was concluded that the Drosophila larval head is derived from the unsegmented acron and six segmental anlagen present at the cellular blastoderm stage. Not all of these proposed segments are composed of symmetric, circumferential strips of embryonic cells. The most anterior segments in particular consist of small localized regions of the blastoderm (see Fig. 4).

In general, the current evidence can be summarized as follows. The Drosophila head, like the ‘generalized’ insect head, is bipartite in structure. The first region is the asegmental acron, which gives rise to the brain (including the optic lobes) and several of the elements of the cephalopharygeal skeleton. The remaining paired cuticular and sensory structures shown in Fig. 4 are derived from the second, segmented region of the head. This region comprises six segments which are, from anterior to posterior, the labral, antennal, intercalary, mandibular, maxillary and labial segments. There are two areas about which some controversy still exists. The first involves the acron. As some investigators have postulated, there may be regions of the acron that are in fact the remnants of ancestral segments. The second area about which there is still some dispute is the existence of the six head segments. There remains some doubt about whether the labral region is truly a metamere and whether the intercalary segment still exists in the highly evolved Drosophila head. As we will see, molecular genetic studies have provided new information relevant to these issues of head structure.

In the Drosophila embryo, pattern formation is specified during the first few hours of development (for review, see Akam, 1987; Ingham, 1988). The ‘coordinates’ of the embryo are established by four classes of maternal gene products. These gene products are encoded by the terminal, anterior, posterior and dorsal/ventral groups of genes. In the central trunk region, maternal information from the anterior and posterior groups is translated into the metameric pattern of the embryo through the activation of a cascade of zygotic genes. The gap, pair-rule, and segment polarity genes subdivide the embryo into parasegments, the initial, transient units of segmentation. The identities of these parasegments are specified by the homeotic genes of the Antennapedia complex (ANT-C) and Bithorax complex (BX-C). Parasegmental divisions are ultimately replaced by segments, which persist throughout embryogenesis.

As noted above, the head is composed of two parts, the asegmental acron and a segmental domain. These two regions are primarily specified by the genes of the terminal and anterior classes, respectively. As will be described below, however, the labral segment is in fact included in the region of the head specified by the terminal group of genes. In this section, we will review what is currently known about the genes in these two classes that are required for head development. For convenience, we will group these genes using the categories that have been established for the trunk region of the embryo. However, as will be discussed, the functions of these genes are not necessarily analogous in the head region.

The termini of the embryo are established by a signal transduction pathway which translates a localized signal to generate the formation of various structural derivatives. The maternal genes involved in this pathway include torso, torsolike, trunk, fs(l)Nasrat, fs(l)pole-hole and l(l)polehole (reviewed in Manseau and Schupbach, 1989). Mutations in any of these genes result in the loss of terminal structures, which in the anterior head region include the acron and the derivatives of the labral segment. In each case, posterior terminal elements are also deleted. The torso (tor) gene encodes a putative tyrosine kinase receptor which is uniformly distributed throughout the embryo (Sprenger et al. 1989; Casanova and Struhl, 1989). Upon being activated by a presumably localized ligand, the tor protein transmits its signal via a downstream serine-threonine kinase encoded by the l(l)polehole gene, which is the Drosophila homologue of the raf oncogene (Nishida et al. 1988; Ambrosio et al. 1989a,b). This mode of activation appears to be similar to the mammalian signal transduction pathway involving the PDGF receptor and the c-raf gene.

Recently, several genes have been identified that are potential zygotic elements of the torso signaling pathway. Each of these genes is involved in the specification of both ends of the embryo, but we will focus primarily on their anterior functions.

Loss of function mutations at the tailless (tll) locus result in the deletion of a subset of the structures that are derived from the anterior terminal domain. Specifically, tll mutant embryos are missing most of the brain (including the supraesophageal ganglion and optic lobe, but probably not the most anterior region of the brain) and parts of the cephalopharyngeal skeleton (the dorsal bridge and dorsal arms). These derivatives originate from the posterior part of the anterior terminal region. huckebein (hkb) appears to be required for the establishment of the extreme poles of the embryo, hkb mutations in fact affect the anterior midgut, which is a structure outside the anterior limits of the ior-requiring region. Although tll and hkb are together probably sufficient for the development of the posterior terminal domain, they are probably not the only zygotic genes required for establishing the anterior terminus. This can be deduced from the fact that the labrum, which is absent in maternal mutants of the tor class, is present in both tll and hkb embryos.

The tll gene has been isolated by Pignoni and colleagues (1990) and shown to encode a protein that is quite similar to the members of the family of steroid hormone receptors. This similarity extends over both the putative DNA-binding zinc finger region and the ligand binding domain. Although it is initially transcribed in mirror-image terminal ‘caps’, anterior tll RNA expression quickly becomes localized to a smaller region which probably roughly coincides with its domain of function. This domain includes cells that appear to be procephalic neuroblasts which give rise to the developing brain.

The homology to steroid hormone receptors suggests a role for the tll gene product as a transcriptional regulatory molecule. This is consistent with the fact that tll expression is required for transcription of the caudal, hunchback and fork head genes in particular terminal regions (Mlodzik and Gehring, 1987; Schroder et al. 1988; Weigel et al. 1990). In addition, in its anterior domain, tll represses the expression of hb, fushi tarazu (ftz) and Deformed, and may thereby function as an ‘anti-segmentation’ gene (Reinitz and Levine, 1990). The tll gene thus appears to be a critical element (downstream of torso and l(l)polehole) in the pathway that induces the formation of the unsegmented ends of the embryo. The hkb gene has also recently been isolated and shown to encode a zinc finger-containing protein (G. Bronner and H. Jackie, personal communication), again suggesting a function in transcriptional regulation.

Recently, two loci have been identified which appear to represent a novel class of homeotic genes. These genes, fork head (fkh;, Jurgens and Weigel, 1988) and spalt (sab, Jurgens, 1988), function outside the trunk region, where parasegmental identities are established by the homeotic selector genes of the ANT-C and BX-C. In addition, the activities of both/Ui and sal appear to be genetically independent of those of the homeotic selector genes. Although the sal gene actually affects the segmented domain of the head, we will include it here because of its functional similarities to fkh.

fkh and sal mutations affect both the head and the tail regions of the embryo. In fkh mutant embryos, unsegmented terminal regions (both anterior and posterior) become transformed into segmental derivatives. At the anterior end, the structures affected include preoral head derivatives (the dorsal bridge and the labrum) and the foregut (including the esophagus and proventriculus, which are derived from the stomodeum).fkz mutations result in the transformation of these asegmental structures into segmented, postoral head elements. Analogous transformations also occur at the posterior end of the embryo. In both cases, loss of fkh activity appears to produce transformations directed towards the center of the embryo. Although the posterior.fkz domain lies within the region affected by the tor group, the anterior domain, which extends to the tip .of the embryo, includes anterior parts of the gut beyond the tor region. Therefore, the anterior domain must require additional maternal inputs, perhaps from the bed system.

Mutations in sal also produce centrally directed homeotic transformations near each end of the embryo. Compared to fkh, the regions affected are more posterior (in the head) and more anterior (in the tail). Specifically, posterior head structures (derived mostly from the labial segment) are changed into anterior thoracic structures while anterior tail segments become posterior abdominal segments.

The functions of fkh and sal are distinct from those of the homeotic selector genes. This independence is supported by the genetic relationship between these mutations and mutations in the Poly comb (Pc) group. Genes in the Pc class have been shown to regulate negatively the genes of the ANT-C and BX-C (Struhl and Akam, 1985; Wedeen et al. 1986). In Pc group mutant embryos, these homeotic genes are ectopically expressed, causing a range of phenotypic defects. However, the domain of sal action is not similarly expanded in such embryos, demonstrating that the region affected by sal mutations is distinct from the trunk (Jurgens, 1988). An equivalent genetic analysis showed that the fkh domain is not expanded in Pc embryos (Jurgens and Weigel, 1988). The functional difference between sal and fkh and the genes of the ANT-C and BX-C is also supported by molecular analysis. The fkh gene encodes a nuclear protein which has no homeodomain but may have a novel DNA-binding motif (Weigel et al. 1989; Weigel and Jackie, 1990). sal is predicted to encode a small protein with interesting repetitive stretches but also lacking a homeodomain (Frei et al. 1988). The expression patterns of the fkh protein and sal RNA during embryogenesis include the regions that give rise to the structures affected by each mutation. Further experiments will be required to determine the modes of action of each predicted gene product.

The central trunk domain of the embryo (including the thoracic and abdominal segments) derives from the region of the blastoderm extending from 20 – 60 % EL (parasegments 3 – 13). As mentioned earlier, a hierarchy of genes has been shown to be responsible for the progressively more refined subdivision of the embryo. In the anterior region of the trunk, the critical maternal morphogen has been shown to be bicoid (Frohnhofer and Nusslein-Volhard, 1986). The bicoid mRNA is initially localized by the maternal products of the swallow and exuperantia genes (Berleth et al. 1988; Stephenson et al. 1988). bicoid protein subsequently forms a concentration gradient (Driever and Nusslein-Volhard, 1988) that declines in the posterior direction. However, in embryos lacking maternal bed product, not only the thoracic segments, but the entire head region is deleted and replaced by a duplication of asegmental tail structures. This indicates that bed is also the key maternal requirement for the establishment of all the head segments. The zygotic components of this process will be discussed in the following sections.

The only zygotic gene that has been demonstrated to be a direct transcriptional target of bed is hunchback (hb; Driever and Nusslein-Volhard, 1989a). In embryos homozygous for strong hb mutations, two embryonic domains are deleted (Lehmann and Nusslein-Volhard, 1987). These are the labial and thoracic segments (anteriorly) and the 8th and part of the 7th abdominal segments (posteriorly). The deletion of the labial anlagen, the most posterior of the head segments, is indicated by the absence of the labial sense organ and the H-piece, a component of the cephalopharyngeal skeleton (see Fig. 4). All the more anterior head segments appear to be present in mutant embryos. The hb gene has been shown to encode a putative zinc finger-containing transcription factor which is initially expressed in the anterior half of the blastoderm (Tautz et al. 1987).

A second mutation with a gap-like phenotype that affects head segments is giant (gt). As in the case of hb, loss-of-function mutations in this gene affect more than one region of the embryo. In the head, structures derived from the labral and labial segments are deleted (Mohler et al. 1989; Petschek et al. 1990). Posteriorly, abdominal segments 5 – 7 are affected although not all the cuticular tissue derived from this region is lost. Perhaps as a consequence of the anterior deletions, head involution fails to occur properly, resulting in parts of the head skeleton being extruded from the embryo. The gt gene has recently been isolated and shown to encode a protein containing a ‘leucine zipper’ motif (V. Pirrotta, personal communication), suggesting that it functions (perhaps in conjunction with other gene products) as a transcriptional regulator of downstream segmentation genes. The expression pattern of gt RNA during early embryogenesis is interesting but somewhat difficult to explain fully. At the cellular blastoderm stage, gt is expressed in four stripes, each about 5 – 6 cells in width. Stripes 1, 3 and 4 are expressed in regions that give rise to labral, labial and abdominal structures, respectively. However, in the case of stripe 2 (at approximately 80% EL), no corresponding defects can be identified. In addition, stripe 3 covers a significantly wider region than the labial segment primordium. The function of this seemingly superfluous transcription may be clarified when antibodies to the gt protein become available. Some of the interactions between gt and other segmentation genes have also recently been identified. These include cross-regulatory effects among gt and other gap genes as well as the determination of particular homeotic gene boundaries of expression by gt (Reinitz and Levine, 1990).

As mentioned above, bcd mutant embryos lack all head and thoracic structures. Since hb is only required for development of the labial head segment, there must be other bed-regulated genes responsible for more anterior head segments. Accordingly, Driever et al. (1989) have proposed the existence of an additional gap gene (‘gene X’) or genes which would be activated at higher bed concentrations than is hb. Loss-of-function mutations at such a locus should result in the deletion of a block of adjacent head segments.

One of the problems involved in attempting to identify candidate head gap genes is the difficulty in scoring head structures in mutant embryos. Mutations that disrupt head involution can result in gross abnormalities in the head cuticle which do not necessarily represent true structural deletions (see for example, the subsequent discussion of the labial gene). There are, however, head structures whose presence or absence is relatively easy to assess (e.g. segment specific sensory organs). In addition, as will be discussed shortly, there are genes whose expression patterns may function as segmental markers in anterior head regions.

Three mutations have recently been characterized which affect anterior head segments. All three loci were originally identified in a large scale screen for zygotically acting mutations affecting segmental patterning (Wieschaus et al. 1984; Jurgens et al. 1984). The first, empty spiracles (ems), has been phenotypically and molecularly characterized by Dalton et al. (1989). Named because it is required for the development of the tracheal system in abdominal segment 8, ems mutations also result in the deletion of specific anterior head structures. The interpretation of the ems head phenotype, however, is not entirely straightforward. Dalton and colleagues argue that ems is a homeotic selector gene controlling the identities of the antennal and mandibular segments. However, as they point out, no obvious homeotic transformations can be seen to occur. Cohen and Jurgens (1990) have proposed a different segmental interpretation of the ems head phenotype, which will be discussed in section III. The ems gene has been isolated and shown to contain a canonical homeobox. At the blastoderm stage, the protein is expressed in an anterior circumferential stripe. This stripe is under the regulation of the maternal bed product. Embryos with varying dosage of bed form the ems stripe at different anterior-posterior positions. Later, the ems protein becomes localized to specific head regions of the extended germ band embryo.

The second candidate for an anterior gap gene is orthodenticle (otd). otd mutations result in the deletion of an overlapping set of head structures more anterior than those affected by ems (Finkelstein and Perrimon, 1990). For example, both mutations cause the deletion of the antennal sense organ, while only otd deletes the dorsomedial papilla (DMP) and only ems deletes the dorsolateral papilla (DLP; the DLP and DMP are peripheral parts of the maxillary sense organ that are not derived from the maxillary segment). As in the case of ems, no homeotic transformations can be found in otd mutant embryos. The otd gene has also been isolated and shown to be expressed in a circumferential anterior stripe at the cellular blastoderm stage. This otd expression, like that of ems, is under .bed control. However, the anterior limit of otd expression (at approximately 90 % EL) is determined by the maternal tor product which represses otd expression in the anterior 10% of the blastoderm, otd also encodes a predicted homeodomain protein (Finkelstein et al. 1990). At residue 9 of the ‘recognition helix’, otd is the only homeodomain protein with the same amino acid as bed, which suggests a similar binding specificity for the two proteins (Hanes and Brent, 1989; Treisman et al. 1989). otd has indeed been shown to bind to consensus bed-binding sites in the hb promoter (M. Simpson and C. Desplan, unpublished observations). The regulatory significance of this binding is not yet clear.

A third mutation that causes deletions in this anterior segmented region is buttonhead (btd). As in the cases of ems and otd, head involution fails in btd mutant embryos. The structures deleted by btd form a third overlapping set that extends more posteriorly than those of otd or ems (Cohen and Jurgens, 1990). Again, no evidence for homeotic transformations can be found. The btd gene has not yet been characterized, so the molecular mechanism of its effects remains unclear.

We will discuss this class only briefly, because little is understood about the role of pair-rule genes in terminal and segmental head development. A number of pairrule genes are expressed in stripes in the gnathal region and are required for the correct establishment of the gnathal segment boundaries. Amorphic /iz mutations, for example, result in the absence of the boundary between the maxillary and labial segments (Wakimoto et al. 1984). hairy (h), paired (prd) and runt (run) have been shown to be expressed in more anterior regions at the blastoderm stage (Ingham et al. 1985; Kilchherr et al. 1986; Gergen and Butler, 1988). In each case, anteriormost expression is confined to a dorsal region of the embryo, hairy, for example, is expressed in a dorsal region from 85 – 95 % EL as well as in a circumferential stripe at about 70 – 75 % EL. Null mutations at the h locus affect labral derivatives and delete the mandibular-maxillary boundary. Strong mutations in both the prd and run genes also cause segmental deletions and fusions (Nusslein-Volhard and Wieschaus, 1980) but the role of these genes in the pregnathal regions is not yet known.

In the trunk region, segment polarity genes act to define and maintain cell fates within segmental units (Martinez-Arias et al. 1988). Mutations in any of these genes cause deletions (and often accompanying duplications) of specific intrasegmental pattern elements. Several of the genes in this class are first expressed in narrow stripes one cell in width, which define positions within each parasegment (and ultimately segment). For example, the expression of the engrailed (en) and wingless (wg) genes respectively demarcate the anterior and posterior compartments of each parasegment (DiNardo et al. 1985; Baker, 1988; van den Heuvel et al. 1989).

In addition to their expression in the trunk, en and wg are expressed in stripes in the three gnathal segments and in discrete patterns in more anterior head regions. This anterior expression occurs in the postulated positions of the intercalary and antennal segmental primordia as well as in preantennal regions. The two segment polarity genes in the gooseberry locus (BSH4 and BSH9) also appear to be expressed in all six head segments (Baumgartner et al. 1987). As will be discussed, these expression patterns are extremely useful in analyzing these anterior regions. Because the effects of specific mutations on anterior head segments are difficult to assess, segment polarity gene expression is providing an important molecular marker for these segments.

Parasegmental identities in the trunk region are specified by the homeotic genes of the ANT-C (thoracic region) and BX-C (abdominal region). In addition to several other genes, the ANT-C contains five homeotic genes: labial (lab), proboscipedia (pb), Deformed (Dfd), Sex combs reduced (Scr) and Antennapedia (Antp). Three of these genes (lab, Dfd and Scr) are required for the embryonic head formation and will be discussed below. It should first be noted, however, that these three genes were originally labelled ‘homeotic’ because of the segmental transformations of adult head structures caused by particular alleles. Only in the case of Scr, however, is it clear that similar transformations can be seen in the embryonic head region.

The lab mutation was originally named because two of the affected structures (the H piece and salivary glands) are thought to be derivatives of the labial segment. In mutant embryos, head involution fails, and several other larval head structures (derived from the gnathal segments) are also disrupted (Merrill et al. 1989). As with other mutations discussed previously, no obvious homeotic transformations can be seen, lab embryos develop normally until the onset of head involution, when the required fusions and movements of the head lobes do not occur correctly.

The molecular analysis of lab has been quite important in the interpretation of the mutant phenotype. The labial protein contains a homeodomain and is the most anteriorly expressed of the ANTP-C genes (Diederich et al. 1989). It is not expressed in any of the gnathal lobes (including the labial lobe) from which the phenotype was thought to derive. Instead, it is expressed in more anterior regions, including the lateral margins of the procephalic lobe. Diederich and colleagues propose that labial expression occurs in regions necessary for correct head involution. A critical region in this process may be the intercalary segment, for which labial could be a marker. This argument is partly based on a comparison of the position of lab protein expression in the Drosophila embryo with the position of the intercalary segment in more primitive insect heads. The expression pattern of the lab protein suggests that some of the head defects are a secondary consequence of the failure of head involution.

Dfd was originally identified as a dominant mutation that affects the formation of the adult head (Lindsley and Grell, 1968). Subsequently, recessive loss of function alleles were recovered that were shown to affect embryonic head development. However, the interpretations of the head phenotype of embryos lacking Dfd function differ. Merrill et al. (1987) found that mutant embryos have disrupted head involution and that larval head structures of maxillary, and to a lesser extent, mandibular and antennal origin are deleted or perturbed. A second group (Regulski et al. 1987) saw similar deletions, but also reported the duplication of an anterior portion of the larval head skeleton. This homeotic transformation was only seen in a single allelic combination though, and may not reflect the true loss of function phenotype. However, ectopic expression of the Dfd gene driven by a heat shock promoter does induce homeotic transformations of many head and thoracic segments towards a maxillary identity (Kuziora and McGinnis, 1988).

The predicted Dfd gene product also contains a homeodomain (Regulski et al. 1987). The protein is expressed in a circumferential stripe at the blastoderm stage and later becomes concentrated in the region of the mandibular and maxillary segments. Elegant analyses by Jack and colleagues (1988, 1990) have shown that the initial stripe of Dfd expression requires input from maternal, gap and pair-rule genes for its correct establishment.

The Scr gene is necessary for the formation of the labial segment, as well as for the prothoracic segments of the embryo (Wakimoto and Kaufman, 1981). In mutant embryos, a partial transformation of the labial segment to a maxillary identity is suggested by the duplication of part of the maxillary sense organ in the labial segmental region (Sato et al. 1985). The Scr gene has been shown to contain a homeobox (LeMotte et al. 1989) and to be expressed in the regions of the embryo affected by mutations. In the head region, the protein is first expressed at gastrulation and becomes increasingly concentrated in the labial lobes (Riley et al. 1987; LeMotte et al. 1989). As in the case of Dfd, Scr expression is under combinatorial control and has been shown to be altered by mutations in different classes of segmentation genes (Riley et al. 1987).

How does pattern formation in the Drosophila head compare with the development of the central region of the embryo? In principle, two basic types of models of head development are possible. The first model is that head formation follows the same hierarchy of genetic interactions that govern the trunk. In this model, the roles played by the various classes of regulatory genes (coordinate, gap, pair-rule, segment polarity, homeotic) would be equivalent in the head and trunk. Only the structural difficulties in identifying head segments make these roles difficult to assess. The second model of head development is that, in order to promote a higher degree of specialization, a different, or modified, system of genetic control has evolved in the head region. Such a model might be expected to be more combinatorial in nature in order to permit a higher degree of flexibility. In addition, it might include new classes of gene products as well as not utilizing all the classes functioning in the trunk.

The current molecular and genetic evidence supports the second type of model of head formation. To begin with, both morphological and genetic studies demonstrate that the head is composed of two distinct domains. The anterior terminal domain (which includes the acron and labral segment) is specified primarily by a genetic system (the torso class) which is quite different from that used in the trunk. This system is initiated by what appears to be a phosphorylation cascade that ultimately activates zygotic transcription factors. Within the terminal class, there are novel genetic activities (i.e. the fork head gene) which promote terminal development over a segmental ground state. Such functions, as discussed by Jurgens and Weigel (1988), are quite ancient, since they would be required even in the annelid-like ancestor of the articulates (see section I). In addition, the terminal head domain requires input from the bicoid system to specify its anterior character. In the absence of maternal bicoid activity, the anterior region is replaced by posterior terminal structures.

The second domain of the head, the segmented region, is specified by the anterior (bicoid) class of genes. However, currently available evidence suggests that head segmentation is governed by different, or modified, rules compared with segmentation in the trunk region. Cohen and Jurgens (1990), for example, have proposed a novel model for the establishment of head segmentation. Through a comparative study of otd, ems and btd mutant embryos, they have postulated that these three genes form an overlapping set of head gap genes. This analysis was undertaken in an attempt to understand the regulation of the Distal-less gene, which is also expressed in a segmental pattern throughout the embryo (Cohen, 1990). In addition to analyzing cuticular structures, they demonstrated that each of these mutations deletes three adjacent segments, as defined by the en/wg expression pattern (Fig. 5). The deletions, which include two postulated preantennal ‘segments’, are each out of register by one segment, such that btd, ems and otd act in increasingly anterior domains. In addition, the deletions appear (based on the en/wg pattern) to be in segmental, rather than parasegmental register. These investigators proposed that, in the head, these genes may be required not only to establish contiguous blocs of segments, but also to specify segmental identity. If this is true, otd, ems and btd would be functioning simultaneously as gap and homeotic selector genes. In such a combinatorial model, the identity of each head segment could be specified by the combination of (three) ‘gaplike’ genes expressed within its boundaries. It will be important to determine if there are additional overlapping gap genes that are required in the posterior head segments. It is also possible, however, that specification of the gnathal segments is more similar to that of the trunk region.

It also appears that other classes of genes that function in the trunk are used differently in the head region. The analysis by Mahaffey and colleagues (1989) is interesting in this regard. By examining the protein distribution patterns of three of the genes from the ANTP-C, they found that these genes are expressed in non-overlapping domains in the head. This differs sharply from the overlapping expression of homeotic selector genes in the trunk. In addition, the various pair-rule genes are either not expressed in anterior head regions, or else appear to show less specificity of expression. It may be, if the overlapping gap gene model discussed earlier is correct, that pair-rule gene expression is not required for specifying segments in this region.

Finally, the ‘region specific’ activity of the spalt gene appears to have evolved during the process of head specialization, sal, which promotes head segmental development above a trunk ground state, would be required for the cephalization of anterior trunk segments (the first evolutionary step in Fig. 2). This function represents the real beginning of the development of a more complex head. Future studies will determine if there are other members of this class of genes required to specify the head (and tail) of the embryo as distinct from the central trunk domain.

The molecular genetic data is also relevant to the controversial area of metameric identity in the head. The expression patterns of the segment polarity genes are proving useful in defining head segments. It is reasonable to assume that these genes, which mark intrasegmental compartments in the trunk, are also expressed in segmental patterns in the head. The expression patterns of, for example, the engrailed and wingless genes, support the existence of all six head segments described in section I. In addition, these patterns suggest that there is cryptic segmentation in preantennal regions. These preantennal ‘segments’ are no longer true metameres, but their possible existence is significant in interpreting the phenotypes of mutations that affect head formation.

In order to establish just how different development is in the head and trunk, it will be critical to determine the precise hierarchy of gene regulation. For example, are the putative head gap genes btd, ems and otd directly regulated by bed? Is pair-rule gene activity required in the head to establish segment polarity gene expression? How does the molecular cascade in the terminal domains differ in the head and tail? As in the trunk region, the availability of this kind of information should clarify our models of head formation. It should also be emphasized that there are other mutations that affect head development that we have not discussed (e.g. Distal-less), as well as additional loci perhaps not yet identified.

An area not discussed in this review is the development of the adult head in Drosophila. This process (which occurs during the larval stages from imaginal disc primordia) is essentially independent of embryonic head formation, but should nevertheless cast increasing light on the events discussed here. For example, many of the genes mentioned here play important roles in the development of the adult head. In several cases, the absence of expression of these genes during adult head development results in clear homeotic transformations not seen during embryonic stages. It will be important to determine whether such differences represent true variations in function, or simply reflect current inadequacies in our understanding.

It is likely that understanding Drosophila head formation will have important implications for higher organisms. Recently, for example, an increasing amount of evidence indicates that the vertebrate hindbrain develops as a series of segments or rhombo-meres (for review, see Lumsden, 1990). Furthermore, in situ hybridization experiments have demonstrated that various homeobox and zinc-finger genes appear to be expressed in patterns that respect rhombomere borders. For example, there are vertebrate homologues of the genes of the Bithorax and Antennapedia complexes which are arranged in clusters on the chromosome resembling those in Drosophila. The anterior limits of expression of these ‘Hox’ genes, as in the fruitfly, appear to correlate with their positions within these clusters (Graham et al. 1989; Wilkinson et al. 1989). It is not yet clear whether the vertebrate forebrain and midbrain are also segmentally organized. In this regard, it will be important to determine whether there are vertebrate homologues of the fly genes discussed here which are expressed in more anterior head regions. Our increasing understanding of the rules governing Drosophila head development is certain to contribute to models of our own morphogenesis.

We would like to thank the members of the Perrimon laboratory for critical reading of the manuscript. We also thank Stephen Cohen for helpful discussions and H. Jackie and V. Pirrotta for communicating results prior to publication. We apologize to any authors whose work we have overlooked or neglected to cite. We are particularly indebted to Gerd Jurgens and colleagues for many of the observations upon which this review is based. Finally, we are grateful to the two reviewers of this manuscript, who significantly aided its improvement.

This work was supported by the Howard Hughes Medical Institute and by an NIH postdoctoral fellowship to R.F.