Drosophila segmentation: after the first three hours

Not long ago, there were streams of illuminating discoveries about early Drosophila development and these have built a coherent picture of the first three hours (Nüsslein-Volhard, 1991). Now these streams have slowed: attempts to find clarifying principles have been held up at a particular phase in development, the period when segment polarity genes begin to take over the task of patterning the segment. At that time, the pair rule genes (Nüsslein-Volhard and Wieschaus, 1980) have already laid down the segmented body plan. It was suggested that the prime function of pair rule genes is “to locate boundaries that delimit fields or gradients of positional information” (Lawrence, 1987). The purposes of this speculative essay are to assess this hypothesis with respect to new evidence about the segment polarity genes, to consider the functions of these genes and to question some modern trends. (There is a swarm of reviews in this field; please see them for primary references: e.g. Martinez-Arias, 1989; DiNardo and Heemskerk, 1990; Ingham, 1991; Hooper and Scott, 1992; Ingham and Martinez-Arias, 1992; Peifer and Bejsovec, 1992; Woods and Bryant, 1992).

There is now good evidence that the parasegmental borders are the first and most fundamental signs of segmentation. They are delineated by the anterior boundaries of expression of the pair rule genes fushi tarazu and evenskipped. They coincide with the parasegmental grooves. There was evidence that the parasegmental border coincided with a lineage restriction, a compartment border (GarciaBellido et al., 1973) and now this has been demonstrated by observing the division pattern and gene expression of cells directly (Vincent and O’Farrell, 1992). This means that, by the end of the blastoderm stage, cells are unequivocally allocated to specific parasegments. The parasegmental boundaries become the limits of expression and requirement for homeobox genes such as Ultrabithorax, while the mechanisms that position and fix these limits are becoming clearer (see Müller and Bienz, 1991).

Although the pair rule genes locate the parasegment boundaries, they cannot maintain them beyond the first three hours, for many of the pair rule genes switch off and their products disappear around that time. Yet, in our view, these boundaries have to be maintained; for example to impede adjoining populations of dividing cells from mixing, to delimit selector gene expression and to contain gradients of positional information. We propose that the segment polarity genes are largely responsible for these boundary properties. We mean that their products would be instrumental in cell adhesion and cell affinities (lineage restriction?) and act as components of intercellular junctions (limits of fields?). The gradients of positional information set up in relation to these boundaries would subsequently determine cell fate and cell polarity.

There are two main types of evidence consistent with our view. First, the pattern of expression of segment polarity genes, if limited to parts of the parasegment, is either to abut the boundary (cubitus interruptus, engrailed, hedgehog, patched, wingless) or form a narrow stripe overlapping the border (gooseberry). Second, all the mutant phenotypes can be interpreted as due to border failures or to extra borders. To understand this, we need to look back to earlier work.

Although most of the experiments on insect segmental gradients were done 20–30 years ago, they have not been superseded, nor have the phenomena found any other explanation. They are concerned with exactly the same type of patterns seen in the Drosophila embryo after the first three hours; these are the array of cell types within the segment, their arrangement along the anteroposterior axis and their polarity as indicated in the orientation of anisotropic structures in the cuticle. The models generated from these studies can be applied directly to the Drosophila patterns even though they were not designed for Drosophila. The system that we envisage is a gradient, probably though not necessarily, a concentration gradient of a morphogen. In the model, the level of the gradient gives the cells information of their position in the segment, determines the patterns of gene expression and thus the types of differentiation. The direction of slope of the gradient gives the cells their polarity. The gradient is reiterated in each segmental unit. Thus, if one segment has a broader band of denticles than another, this is not because the gradient differs in each segment. Rather, common fields of positional information are interpreted uniquely in each segment, the interpretation depending on the subset of homeotic genes that are active there. Any changes in the gradient, whether brought about by mutations or experiments, would therefore apply equally to all the segments and produce consistent changes in the pattern of each one (reviewed in Lawrence, 1992).

Figs 1 and 2 show an example of a gradient model, in which parasegment borders form barriers between adjacent gradients. It suggests a simple way to interpret all segment polarity phenotypes. These phenotypes, either produced by mutation or by universal expression of the genes, can be arranged into two classes. In the first, the borders ‘leak’, resulting in a flatter gradient landscape. As shown in Fig. 1, this failure can be massive (as in null alleles of wingless and hedgehog, or when very little armadillo protein is present) or slight (as in gooseberry⁻, or when there is some wingless, hedgehog or armadillo proteins). Most segment polarity mutants belong to this class. In the second class, which includes naked⁻ and patched⁻, ectopic borders appear and these are associated with altered patterns and polarity (Figs 1, 2). For example, in the naked⁻ mutant embryo, the cuticle loses the denticle bands and ectopic grooves mark the new boundaries. Using the gradient model, all these changes in the cuticle, in each segment, find a common cause in an altered landscape, which itself derives from the persistence of the original border and the arrival of a new one (see Figs 1, 2). The gradient model provides one explanation for the many patterns in the different segments including such curious features as the elimination of all patches of denticles except the beard in the first thoracic segment and not only why the beard persists but also why it has a reflected pattern of denticles (compare Figs 2 and 3).

This suggests that all segment polarity genes are needed for normal parasegment borders. If this is so, what are their wild-type functions? Some gene products might be involved in making a special kind of ‘sealing’ junction at the border (Sampedro et al., 1993). This special junction might depend on intracellular localisation or stabilisation of the armadillo protein (a component of adherens junctions, Peifer et al., 1993), a localisation or stabilisation that requires wingless (Riggleman et al., 1990). After generalised expression of wingless, the altered localisation of armadillo protein can be seen in every cell (Noordermeer et al., 1992), but borders only appear at their normal location and at an ectopic site in the middle of the parasegment. One possible explanation is that normal and ectopic borders might form only where cells with different adhesion properties or ‘affinities’ meet (Morata and Lawrence, 1975). Other segment polarity gene products might themselves be instrumental in cell affinities. If the segment polarity genes are mainly responsible for borders, which genes are responsible for the segmental gradient itself? We mean gene products such as those that make the morphogen, are receptors for it or allow it to be transferred or transduced from cell to cell. We conjecture that the activities of these genes need not be localised (since the placing of boundaries alone might be sufficient to determine the positions, extent and shape of the gradients). Unlocalised gene products are usually deposited in the egg by the mother; possibly, therefore, maternal effect genes may include the genes responsible for making positional information – a search for maternal effect mutations that give segmentation phenotypes has already proved fruitful (Perrimon et al., 1989).

We imagine that, even though the patterns of expression of the segment polarity genes are initially determined by the pair rule genes, they later become dependent on the gradient that they themselves have helped to generate. Once the gradient is established, it is the height, the scalar, of the gradient that locates subsequent gene expression. In this way, the gradient landscapes model not only the abnormal cuticle patterns but also the changed patterns of gene expression found in the mutants (Figs 1, 2). In the gradient model, in strong mutants such as wingless⁻, the borders fail, the gradient cannot form properly and a flattish landscape results at a level normally found near the middle of the segment (Fig. 1). In the absence of engrailed, the landscape is also flattened. These flattened landscapes mean that neither engrailed nor wingless can be expressed any longer −because expression of each of these genes is maintained only between certain characteristic levels of the gradient and those levels are no longer there. In other words, winglessand engrailed-expressing cells are postulated to be mutually interdependent because they each perform essential tasks in the maintenance of the borders of the segment gradient. We suggest this feedback between borders and gradient values is also the basis of the autoregulation that has been observed for both wingless and engrailed (Bejsovec and MartinezArias, 1991; Heemskerk et al., 1991). In the naked⁻ mutant, the alterations in the engrailed and wingless stripes are also modelled by the new landscape (Fig. 1).

These ideas differ radically from the various assumptions and models favoured in nearly all of the many recent publications on segment polarity genes (see reviews listed above); models that generally assume that these genes directly determine cell states within the segment: For example, in a row of cells, the products of two segment polarity genes might be present, a combination that would specify cell fate in that row (e.g. Martinez-Arias et al., 1988; Dougan and DiNardo, 1992). Most of the current models (reviews listed above) have been built up step by step, and modified bit by bit, to accommodate new experimental results (e.g. compare the model in Martinez-Arias et al., 1988 with those of Ingham, 1991 and Dougan and DiNardo, 1992). This type of model building (‘carpentry’, Crick, 1988) tends to produce intricate explanations that, naturally, fit the data, but can contain circular arguments and hide simple clarifying principles (if they exist!). In the following list, we compare the two ways of interpreting these genes with respect to mutant phenotypes, wild-type functions and future experiments.

• (1) The segment polarity genes have been seen as subdividing the segments into qualitatively different parts, seen as defining ‘domains’ (as the gap genes do) or ‘cell states’ (as the homeotic genes do). For example, the individual denticle rows could be considered as discontinuous outcomes of discrete cell states, row 1 being qualitatively different from row 2, row 3 from 4 and so on.

  In the gradient model, pattern and polarity depend on a variable that changes continuously across the segment.

• (2) The segment polarity mutant phenotypes have been interpreted as if specific elements of the pattern are dependent on single genes and cut out, piece by piece, in the mutants. Thus in the naked mutant, the denticle belt is ‘deleted’; in some wingless mutations, cells secreting naked cuticle are ‘transformed’ into cells making denticles.

  In the gradient model, the phenotypes are interpreted as reorganisations of the whole pattern, which derive from effects on the boundaries. Polarised denticles and naked cuticle are produced as a direct readout of positional information.

• (3) Most current papers tend to avoid trying to explain pattern in the cuticle, and concentrate rather on the genetic interactions between segment polarity genes. They apply the digital logic of epistasis and arrange the genes in hierarchies and pathways. The logic of epistasis can be safely applied to linear pathways where a series of genes act one after the other; however, it may be precarious to apply the same logic to segment polarity genes. There are three main difficulties: First, some segment polarity genes may encode structural proteins; if the gene products join together to form structures, it is not easy, and may not be useful, to arrange them in a hierarchy. Second, the segment polarity genes act in a sheet of cells in which the neighbours change as the cells move and divide; pathways of cell interactions may easily become inscrutable. Third, these genes are expressed over a long period while both their patterns of expression and their functions, evolve (e.g. Heemskerk et al., 1991.) One unwarranted outcome of the current papers is the general perception that the main function of the segment polarity genes is to regulate each other, each one becoming an element in a network of ‘cross-regulatory interactions’.

  In the gradient model, the genes are involved in the function of the borders that delimit the gradients; the emphasis is not so much on how they ‘interact’ with each other, but on what their protein products might do, to set up the borders, to give the cell junctions particular properties, or otherwise to help establish the gradients. The ‘interactions’ are seen as indirect consequences of how the genes affect the borders and the gradients.

• (4) One key observation is the interdependence of wingless and engrailed; remove either gene and expression of the other is no longer maintained (DiNardo et al., 1988; Martinez-Arias et al., 1988). This interdependence has contributed to the hypothesis that the wingless (and other Wnt genes) encode signalling molecules that carry messages from secreting to responding cells. The wingless protein is supposed to cross the parasegment border to keep the engrailed gene turned on in any cells that are near enough to receive the ‘wingless signal’ (DiNardo et al., 1988). This signal hypothesis has good evidence in its favour; for example, the wingless protein is secreted and antigen can be seen in nearby cells (van den Heuvel et al., 1989; González et al., 1991); in the gut, wingless, when expressed in the visceral mesoderm, influences gene expression in the endoderm (Immerglück et al., 1990). Also, when engrailedexpressing cells are isolated from embryos, they lose engrailed expression, but this does not happen if they are mixed with Drosophila tissue culture cells that produce wingless protein (Cumberledge and Krasnow, 1993). There are other views of the function of wingless: in the leg disc, small groups of cells that express wingless protein ectopically can reorganise the pattern of wild-type cells nearby, suggesting that the wingless protein itself could be a morphogen (Struhl and Basler, 1993). Although all these results are suggestive, they do not prove that the wingless protein itself embodies a message. In the epidermis of the embryo, it has been shown that uniformly distributed wingless protein can emulate the wild-type function of the gene, undermining theories that demand that the protein be unevenly distributed – as it must be if it is to act as a morphogen gradient or as a locally instructive signal (Sampedro et al., 1993).

  Because continued expression of wingless depends on engrailed function, there is postulated to be another signal passing from the engrailed-expressing cells to their wingless-expressing neighbours. It is thought that this latter signal might be the hedgehog protein and that this might interact with patched (Ingham et al., 1991).

  The gradient model provides an alternative explanation for the mutual dependence of engrailed and wingless (Figs 1, 2).

• (5) Digital models have led to many experiments in which segment polarity genes are eliminated in twos or even threes, in which gene products are removed or added at different stages of development and the effects on the other genes studied. Again these experiments emphasise the epistatic relations between the segment polarity genes.

The gradient model helps in the interpretation of some of these double mutants. For example, we have classified the mutants into those in which ectopic borders affect the pattern (e.g. patched) and those that impair border function (e.g. hedgehog). This perspective might suggest that a mutant of the latter class could correct the phenotype of the former class; indeed the patched; hedgehog double mutant has a less severe phenotype than either mutant alone (Ingham et al., 1991).

The gradient model leads to different kinds of experiments. Here are some examples: When dyes are injected into cells, a restriction in dye-spread shows that intercellular permeability can change at compartment borders (Warner and Lawrence, 1982; Blennerhasset and Caveney, 1984). This suggests the presence of special junctions there, or an agent that blocks intercellular channels between cells in different compartments. As far as we know, experiments on dyespread and electrical coupling in the Drosophila embryo have not been reported. More attempts could be made to identify the morphogen. As it may well depend on genes which are transcribed in the mother, during oogenesis, more of these could be looked for. In general terms, the advantage of the gradient explanation is that the model is internally consistent and aims to explain coherently all aspects of the pattern including cell polarity; the disadvantage is that the molecular basis of the gradient is still unknown. However, gradient molecules are no longer mythical; in the last few years several morphogens have been identified in the Drosophila embryo, both in the syncytial stage before cells are formed (Driever and Nüsslein-Volhard, 1988; Struhl et al., 1992) and after cellularisation (Ferguson and Anderson, 1992).

We thank Gary Struhl and our colleagues at the LMB and other scientific institutions for their advice. J. S. is supported by a Fleming Fellowship of the Ministerio de Educacion y Ciencia.