Neurogenetics and Behaviour in Insects

Behavioural neurogenetics is an extension of developmental genetics. Its aim is to elucidate the role of genes in brain development, and in the emergence of speciesspecific behaviour. For decades, the genetics of behaviour has been governed by the concepts and tools of population genetics. The distinction between the two approaches is not clear to everyone and discussions about genes and behaviour tend to confuse the developmental and population genetical point of view. We shall therefore begin this review with a brief outline of the main concepts of these approaches.

Population genetics is concerned with the variability of characters in animal populations and its genetic basics. Variability can result from two factors: variations between individual genomes and between particular environments. If the total variance of character is assumed to be simply the sum of the genetically and environmentally caused variances, its heritability is defined as that fraction of the total which is caused by genetic variance. It can be measured, for instance, in isogenic populations under normal environmental variability (Fuller & Thompson, 1960). This has been tried for human characters (including intelligence) in studies of twins (e.g. Jensen, 1970).

Heritability describes an important evolutionary parameter. The higher the heritability, the faster selection works. However, because of the way heritability is defined, ‘non-heritable’ characters might as well be under tight genetic control. For example, the variability of limb number in any animal population is due mainly to environmental factors. The highly invariable genetic control causes a heritability near zero. Furthermore, it follows from the definition that the heritability of characters in a homozygous inbred strain (or in an isogenic clone) is close to zero. For these reasons, the term heritability for the genetically caused variance fraction is not an appropriate choice. It will certainly continue to cause misunderstandings.

Tolman (1924) was the first to show that behavioural variability in animal populations can be due to variation at the genetic level. By selecting strains of fast and slow learners from a population of rats, he also demonstrated that the development of behavioural traits (even of learning performance) is actually influenced by the genome, an inference which was not self-evident at that time.

The heritability of a character tends to be low when it is under a strong stabilizing selection pressure, for example mating behaviour in Drosophila (Fulker, 1966) and the reactivity of Drosophila to mechanical stimulation (Hay, 1972). Phototactic and geotactic behaviour of Drosophila have been most extensively explored by the techniques of population genetics (see Grossfield, 1978). The heritability is generally below 20% (Michutta, Krause & Kohler, 1981). This is, however, high enough to allow selection for positive or negative taxis in 20–100 generations. As expected, the heritability of the behaviour decreases as selection proceeds (Dobzhansky, Spassky & Sved, 1969), since under most conditions, selection is a reduction of genetic variability

Selection experiments have certainly provided important insights into evolutionary mechanisms. However, for the developmental geneticist, analysing a particular behaviour, the draw-backs of selective breeding as a starting point are obvious. The results strongly depend upon the genetic variability which happened to occur in the initial population subjected to selection. Thus, the final genetic analysis of selected strains cannot be interpreted as showing that the behaviour is mainly under the control of one or another chromosome. This fact is sometimes neglected. Another major disadvantage of the selection method is that it requires mass screening which excludes the investigation of more sophisticated behaviour.

In contrast to population genetics, developmental genetics elucidates the role of the genome in the individual ontogenetic process. In the strict sense, genes do not determine development; outside factors can always interfere with, or stop the process. However, the existence of genetic guidance is obvious. In any given species the outcome of development, beginning with a fertilized egg, is fairly predictable for any morphological traits as well as for species-specific action patterns. Ethologists had this in mind when they coined the term ‘innate behaviour’ (Lorenz & Tinbergen, 1938). They observed that the emergence of some behaviour is a normal and predictable outcome of the developmental process, fairly invariant to environmental fluctuations, including social isolation.

At the beginning, ethologists contrasted ‘learned’ and ‘inborn’ behaviour, but from our present point of view, this makes little sense (as already recognized by Lehrman, 1953). Genes acquire meaning only in a narrowly defined biological context. Thus it is difficult to exclude from ‘innate behaviour’ all that is learned. Whenever the contents of what is learned are predictable (due to invariant environmental factors), learning may simply be regarded as one form of epigenetic mechanism. For example, auditory feedback plays an important role in the development of normal speciesspecific song patterns in isolated canaries (Marler & Maser, 1977). In Drosophila, parameters of visual flight control are influenced by experience (Heisenberg & Wolf, 1984). Development of so-called ‘fixed action patterns’ may therefore involve learning.

We conclude that the development of species-specific features of brain and behaviour may be called ‘genetically controlled’ (or ‘innate’), whenever the outcome is a reliable event. This convention does not refer to the actual developmental mechanisms involved.

The introductory remarks show that observation of behaviour alone is not sufficient to provide a thorough understanding of its genetic basis. It is also essential to investigate the genetic control of brain development. Due to the bilateral symmetry of insect brains, comparison of the two halves may yield information about the extent of genetic control of brain structure. Genetically, both halves are identical, and environmental factors are also likely to be very similar. Deviations from symmetry are therefore an upper estimate of the effect of developmental noise. At the level of the light microscope, these differences are strikingly small. However, symmetry is strongly influenced by the state of the genome (see below).

Intra- and interclonal comparison of isogenic animals is another way of estimating the relative importance of genetic and non-genetic factors on brain development.Macagno, Lopresti & Levinthal (1973) investigated the visual system of isogenic Daphnia by serial ultrathin sections and found that the overall structural pattern was highly invariant; however, the positions of cell bodies and major branching points of axons were variable within a range of a few micrometers. Similar results were obtained by Ward, Thomson, White & Brenner (1975) for Caenorhabditis.

In insects, Goodman (1976, 1977, 1979) reported the occurrence of duplications and deletions of large and small identified ocellar interneurones in the locust. Typically, extra cells were indistinguishable, morphologically, from their siblings. Comparison of different clones of isogenic grasshoppers indicated a high specificity of the genetic control of cell number. The occurrence of duplications or deletions in one cluster of cells in a given clone was not correlated with their occurrence in other cell clusters.

The expression of a new phenotype with an increased tendency towards duplicate tions or deletions, was never as stable as the wild-type phenotype in other clones. A random bilateral asymmetry was observed that indicates a substantial contribution of noise to the developmental process. In wild-type animals this is obviously smaller, presumably due to a better concordance of the whole genotype. A similar phenomenon of unstable new phenotypes is observed in some structural brain mutants of Drosophila (Heisenberg & Bohl, 1979; Heisenberg, 1980).

The existence of genetically controlled duplications and deletions of neurones is of vital importance for an understanding of brain evolution. One main theme of this review is the parallel organization of brain functions. Duplications of existing neuronal pathways may provide the substrate for later functional differentiation.

Not only number, but also neuronal shape is under tight genetic control. One example is the HS neurones of the dipteran lobula plate (Fig. 1). Hausen (1982) reported that although the dendritic branching pattern may vary considerably between HS neurones (of the left and of the right lobula plates within individual Calliphora erythrocephala), the dendritic fields of homologous horizontal neurones are nearly identical, even when neurones of different animals are compared. In addition, the density of dendritic branching seems to be constant. Fig. 1 demonstrates that these parameters are also similar for HS neurones of different species. Furthermore, in Drosophila, the similarity of major branching patterns between homologous HS neurones of different animals is conspicuous.

Goodman (1979), using the large ocellar interneurones in different isogenic clones of grasshoppers, found that most clones did not show any morphological abnormality. One clone, however, produced abnormal morphologies in a particular pair of neurones in 88% of all individuals.

Goodman & Heitler (1977) described the effects of genetic variability on the physiology of identified neurones. The spike threshold of the fast extensor tibiae motor neurone (FETi) decreases with increasing temperature (Heitler, Goodman & Rowell, 1977), a factor that correlates with the increase in jumping frequency. Among 30 parthenogenic clones of locust, two showed altered behaviour. In one of these, a low probability of jumping and lack of temperature dependence was accompanied by an abnormal increase with temperature of the spike threshold of the FETi motor neurone. Intraclonal variability was low.

To summarize, we can conclude that number, morphology and physiology of neurones are under genetic control.

In the preceding section, variability in the brain which is not due to genetic factors was classified as ‘developmental noise’. In this section, some factors contributing to this component of variability are briefly discussed.

Regulatory cell death is a well-known phenomenon, occurring, for example, in the developing optic lobes of insects (Nordlander & Edwards, 1968; Fischbach & Technau, 1984). It has been assumed that there is an overproduction of neurones which competion for a limited number of functional contacts. Neurones that lose the corna petition eventually die. This process (as opposed to programmed cell death) can be regarded as a probabilistic process. This implies that the final outcome (i.e. which cells survive and which die) is not determined by genetic information in the zygote and is not essential for proper functioning of the optic lobes.

Path finding in fibre growth is thought of as a similar probabilistic mechanism. The outgrowth of filopodia from the growth cone is a trial-and-error process (e.g. Johnston & Wessels, 1980), so some minor deviations in axonal projections between isogenic animals can be expected.

Sprouting of neurones in response to experimental (Schneider, 1973; Cotman & Lynch, 1976) or congenital synaptic deprivation (Fischbach, 1983a,b) reveals epigenetic regulation of the number of functional contacts. The resulting irregular shapes of neurones suggest that the establishment and elimination of new contacts is a selective trial-and-error process.

Technau (1984) has shown that the number of axons in the peduncle of the mushroom bodies of Drosophila melanogaster changes during adult life. Furthermore, he found that axon number is relatively low after sensory deprivation of adult flies (see Fig. 2). In honey bees, morphological changes of Kenyon cell spines occur during the first orientation flight (Coss & Brandon, 1983). In the coleopteran Aleochara curtida (Bieber & Fuldner, 1979) considerable growth of the mushroom bodies occurs during the life-time of the imago. Whether some of these structural changes are related to long-term memory is an open question.

Genetic control of learning ability will be discussed below.

Genetic analysis of brain and behaviour requires the investigation of individual genes. In the last 20 years, a large number of single gene mutations have been isolated in Drosophila melanogaster that affect nervous tissue and behaviour. These have recently been comprehensively reviewed by Hall (1982), so that our task here will be to demonstrate, using selected examples, the scope and perspective of single gene analysis.

We begin with a simple motor pattern, the jumping response. This is driven by a pair of giant fibres (GF) which project from the brain through the cervical connective and terminate ipsilaterally in the ventral part of the mesothoracic neuromere (Koto et al. 1981). They form an electrical synapse with a motor neurone (TTMm) of the tergotrochanteral jump muscle (TTM; King & Wyman, 1980). Thomas & Wyman (1982) have isolated several X-linked non-jumping mutants with altered physiology and morphology of neurones. One of those, bendless (ben), lacks the electrical synapse between the GF and the TTMm as seen with anatomical and physiological techniques. In wild-type flies, intracellular stimulation of the GF leads to a TTM response after only 0·8 ms, but in the ben mutant the latency is 2·2 ms. The response latencies of other muscles driven by the GF are not different in mutant and wild-type flies.

The specificity of this mutational effect is remarkable. However, it certainly should not reanimate the ‘one gene -one synapse’ hypothesis. Thomas & Wyman (1982) suggest that the mutation may affect some molecule necessary for recognition between the GF and TTMm. Genetic mosaic analysis should tell whether or not the wild-type ben gene product is required in the GF or TTMm for proper synapse formation Obviously, the search for genes specifying parameters for the establishment of connectivity is an exciting prospect.

The visual system has been a favourite subject for single gene analysis over the last two decades, and as a result a host of mutations affecting it at different levels has accumulated (66 genes are listed by Hall, 1982, and many additional ones have been isolated since). Recent reviews (Hall & Greenspan, 1979; Heisenberg, 1979; Pak, 1979; Hall, 1982; Heisenberg & Wolf, 1984) have emphasized the value of mutants as tools in vision research. Our main objective here is to extract relevant information about the way the genome instructs the organization of visual functions.

The main anatomical constituents of the ‘visual system’ are shown in Fig. 3 : compound eye, lamina, medulla, lobula, lobula plate and optic foci in the central brain. Not depicted are the ocelli which interact with the processing of visual information mediated by the compound eyes (Fischbach & Reichert, 1978; Miller, Hansen & Stark, 1981). A more detailed account of the cellular structure of the visual system is available (see Fischbach, 1983a for Drosophila and Strausfeld, 1976 for Musca.

Total blindness can be caused by disrupting the visual system at different levels. Some mutations interfere with the formation of the compound eyes (e.g. eyeless, sine oculis ; Power, 1943 ; Fischbach, 1983b ; Fischbach &Technau, 1984) ; others prevent the establishment of connections between the compound eyes and the central brain (disconnected-, K. F. Fischbach, unpublished; Fig. 11). Blindness also results when mutations disrupt the function or formation of the photoreceptor cells. The normal product of the no-receptor potential A (norpA) gene is essential for photoreceptor function (Pak & Grabowsky, 1978; Pak, 1979). Some alleles cause a failure to transduce the conformational change of xanthopsin (Vogt, 1983) into a receptor potential.

Mutations in the X-chromosomal gene, retina degeneration A (rdgA), induce blindness by degeneration of photoreceptors during the first week of adult life. Degeneration seems to be due to a phototransduction defect caused by a block in the synthesis of phosphatidic acid (Hotta, 1984). Alleles may differ in the degree to which they affect receptors Rl-6, R7/8 and the ocelli (Harris & Stark, 1977; Homyk, Pye & Pak, 1981; Johnson, Frayer & Stark, 1982).

Mutations causing total blindness are often highly specific insofar as other sensory modalities are not affected.

What follows is a description of some mutants suffering from partial blindness. The point is made that, in most cases ‘partial blindness’, does not involve attenuation of all functions, but rather elimination or attenuation of specific functions.

The ommatidium of a dipteran compound eye contains eight receptor cells which are arranged in a typical pattern. Six receptor cells (Rl–6) are situated peripherally, forming unfused rhabdomeres of large diameter which extend from the crystalline cone to the basal membrane. The central rhabdomere (which is smaller in diameter is formed distally by receptor R7 and, proximally, by receptor R8. The three receptor types (Rl–6, R7, R8) differ in spectral sensitivity (Harris, Stark & Walker, 1976). In the lamina, axons from receptors with identical optical axes converge on so-called ‘cartridges’ (Braitenberg, 1967; Kirschfeld, 1973). The axons of retinula cells Rl–6 terminate in the lamina, while R7 and R8 project into the neuropile of the distal medulla (see Fig. 3).

Several gene mutations specifically affect the different types of receptor cells. The sevenless^LY3(sev) mutation causes non-formation of the R7 retinula cells (Harris et al. 1976). It is cell autonomous, i.e. genetically wild-type R7 precursor cells develop normally in otherwiseser; flies (Campos-Ortega, 1980). The absence of the R7 rhabdomere impairs visual input to retinula cell R8 since the distal end of its rhabdomere lies beyond the focal plane of the corneal lens. The absence of retinula cell R7 should cause a decrease in sensitivity and acuity in visual functions normally mediated by the central rhabdomeres, but Heisenberg & Buchner (1977) have shown that in several responses to visual motion, the sev mutant performs as well as the wild-type. Thus retinula cells R7 and R8 do not seem to play any role in movement detection. This observation is corroborated by the behaviour of mutants with defective receptors Rl–6.

Outer rhabdomeres absent (ora) is probably a structural gene for Rl-6 opsin (Schinz, Lo, Larrivee & Pak, 1982). The mutation acts autonomously in receptors Rl–6 (Stark, Srygley & Greenberg, 1981) and prevents formation of normal rhabdomeresinreceptors 1–6 (Harris et al. 1976). The small distal rudiments are depleted of membrane particles (Schinz et al. 1982). Accordingly the optomotor yaw responses of ora are severely reduced (Heisenberg & Buchner, 1977).

Mutations in the receptor degeneration B (rdgB) gene cause degeneration of Rl–6 receptor cells in response to illumination. The normal gene product is required for phototransduction (Harris & Stark, 1977). Retinula cells R7 and R8 are not affected (Stark, Chen, Johnson & Frayer, 1983). Some rdgB flies show no optomotor yaw response, others do (Heisenberg & Buchner, 1977). This behavioural heterogeneity is probably due to variable expression of the mutant defect in individual flies (it should be noted that these behavioural tests are more sensitive than electrophysiological tests for judging the extent of degeneration).

In summary, the observations on receptor mutants strongly support the hypothesis that, in wild-type flies, responses to motion are mainly mediated by receptors Rl–6. The mutant analysis also suggests that the central receptor types R7 and R8 function in phototaxis and in colour discrimination (Harris et al. 1976; Hu & Stark, 1977; Heisenberg & Buchner, 1977; Fischbach, 1979; Miller et al. 1981).

This example of a structural separation of different functions at the level of the retina in Drosophila is not an isolated case. Optomotor responses of Phormia regina (Kaiser, 1968) and bees (Kaiser & Liske, 1974) are not elicited by moving patterns without intensity contrast. In bees, the spectral sensitivity of the optomotor yaw response corresponds to that of the green receptors, and even in man, colour blindness of certain visual subsystems has been demonstrated (Wolfe, 1983).

Heisenberg (1972) and Heisenberg & Götz (1975) categorized several visual mutants according to their selective defects in the optomotor yaw response. This ihavioural element has been systematically studied over the last three decades (e.g. Reichardt, 1970; Götz, 1968; Heisenberg & Wolf, 1984). The mutants were impaired either in their turning response to movement of narrow stripes in bright light, or to movement of broad stripes in dim light. This observation fitted the notions of a ‘low-sensitivity, high-acuity’ (HAS) and a ‘high-sensitivity, low-acuity’ (HSS) system (Eckert, 1971). Later, Heisenberg & Buchner (1977) showed that the optomotor response in dim and bright light is mediated by receptors Rl–6 alone (see above) ; the HAS and HSS systems therefore represent two adaptational states of the Rl-6 pathway. Support for this conclusion comes from the observation that in both types of mutants, defects of the lamina potential of the ERG can be observed. Furthermore, in Musca domestica, Pick & Buchner (1979) discovered that the distance between the two sampling points of elementary movement detectors increases in dim light, and so at different light intensities different sets of neurones seem to be involved. This may explain the specific defects in the HAS and HSS mutant types.

Five examples of structural mutants of the optic lobes will be briefly described below. In these mutants, different, and partially overlapping, sets of visual neurones are defective.

Vacuolar medullc^KS74(Vam), at present studied by P. Coombe (in preparation), is an X-chromosomal dominant mutation causing degeneration of laminar and medullar cell types. This process begins at eclosion and continues throughout adult life to produce a densely packed array of vacuoles in the distal medulla. In hemizygous males and homozygous females, degeneration becomes apparent during the first few minutes after eclosion, but in heterozygous females, degenerating cell bodies are not visible until at least 1 day after eclosion. With progressing degeneration, the lamina potential in the electroretinogram disappears, as does the optomotor response. However, degeneration obviously does not affect all functional pathways, because certain behaviour patterns persist. The orientation of freely walking Vam-files towards a black vertical stripe (width = 20°) is almost normal. The mutants ora and rdgB fail in this test. Thus, the orientation response in Vam may still be mediated by retinula cells Rl–6 via a set of non-degenerating lamina interneurones. Electron microscopy of the mutant’s lamina should reveal their identity.

While in Vam, the optomotor pathway is disrupted, in the mutant small optic lobes^KS58(sol; see Fig. 6) it operates normally (Fischbach & Heisenberg, 1981), although the number of neurones in the medulla and in the lobula complex is reduced to about 50 % by tissue autonomous degeneration of ganglion cells during the first half of pupal development (Fischbach & Technau, 1984). In sol, visual acuity is not affected and colour discrimination persists (Fischbach, 1981a). For the optomotor yaw response, the light intensity threshold and the upper and lower threshold for contrast frequency are about normal. Thus, the neuronal network responsible for the optomotor yaw response must be intact in sol flies (Fischbach & Heisenberg, 1981). Strong evidence for this conclusion also comes from the relatively normal pattern of radioactive labelling with 2-deoxyglucose (Buchner & Buchner, 1983) in the medulla during responses to movement ( Nicod, 1983 ). The sol mutant may, therefore, be of help in identifying the neurones of the optomotor pathway. However, the number of neuronal types in the medulla is very high ; inMusca it is probably of the order of 120 (Campos-Ortega & Strausfeld, 1972), and in Drosophila wild-type 64 types have already been described (Fischbach, 1983a). An understanding of this complex structure is facilitated by its organization in columns and layers (see Fig. 3). Evidence for the existence of multiple parallel pathways comes from the distribution of the branching pattern of Golgi-stained neurones; for example, one pathway seems to involve two layers of LI arborizations in the distal and the most proximal medulla layer. These layers share a large number of interneurones ( Fischbach,1983a) and the most proximal medulla layer is the one which is most intimately connected to the lobula plate (e.g. via T4 neurones; Fischbach, 1983a). In sol flies, the ‘LI-subsystem’ is probably retained. The drastic reduction of cell number per medullar column is mainly explained by the absence of neurones normally participating in the formation of other layers ; for example, layer 3 of the medulla, which is closely connected via Tm-neurones to deep layers of the lobula, is most reduced in adult sol flies (Fischbach, 1983a). sol flies are impaired in the evaluation of patterns and show a low specificity in the releasing mechanism for landing (Fischbach, 1981a). In addition, they are deficient in an object-background discrimination task (Heisenberg & Wolf, 1984) and in several types of visual learning (see below).

The complexity of the optic lobes can be further diminished with the minibrain mutation (mnb;Fig. 5). This reduces brain volume, including the optic neuropiles (an exception is the lamina), by about 40–50% with a drastic reduction of cell number. As in sol, the number of columns in the optic lobes is normal (Heidenreich, 1982) and mnb flies still show optomotor yaw and landing responses.

Mutations in different genes often act additively. This may be expected if they act specifically on different subsystems. Examples of such additive effects are given by the double mutants sev ora and rdgB sev (Harris et al. 1976) at the receptor level, and at the level of the optic lobes by sol so (Fischbach & Lyly-Hünerberg, 1983) and possibly by mnb sol (Fig. 7). The latter double mutant has normal sized eyes and normal sized laminae. Golgi studies have shown that every columnar neurone of the lamina is retained, including C2, C3 and T1, but the volume of the medulla and lobula complex neuropiles is drastically reduced, although still well structured. The lobula plate, especially, is very thin and seems to consist of little more than giant fibres. The extensive sprouting of these giant neurones into the medulla of the double mutant is not seen in mnb and sol to such an extent. In mnb sol, the reduction in the number of input neurones may have reached a threshold which triggers this additional growth. At this level of resolution, the effects of sol and mnb are synergistic.

Optomotor yaw responses of the double mutant are reduced, but not abolished. Possibly, new functional contacts of the lobula plate giant neurones partially compensate for the loss of normal input neurones.

Other mutants with reduced optic lobes are being crossed with the double mutant. We hope to end up with an even smaller optic lobe containing a minimal number of cell types. However, sprouting will distort their normal shape.

The HS neurones of the lobula plate (see Fig. 1 ; Pierantoni, 1973) are part of the optomotor pathway. This hypothesis relies on electrophysiological studies with Calliphoridae (review: Hausen, 1981) and on the finding that the behavioural Drosophila mutant optomotor-blind^H31(omb) lacks the giant neurones of the lobula plate (compare Figs 8, 9; Heisenberg, Wonneberger & Wolf, 1978; Blondeau & Heisenberg, 1982). It has also been supported by laser ablation experiments with Musca (Geiger & Nässel, 1981, 1982) and surgery in Calliphora (Hausen & Wehrhahn, 1983). However, a closer inspection of the responses of the omb mutant revealed that only the roll response is reduced to zero. Yaw responses in flight range from 20–60%, depending on the stimulus parameters (Heisenberg & Wolf, 1984). This result is puzzling, since in most mutant flies no anatomical trace of the three HS neurones can be found. Is the loss of these neurones compensated by some developmental mechanism, or is the contribution of a second functional pathway to the turning response revealed by the omb defect? The latter seems to be the case for recent advances in the fine analysis of flight control in Drosophila (Bülthoff, 1980; Heisenberg & Wolf, 1984) have shown that in wild-type flies two separate functional subsystems contribute to the yaw response. One, the ‘classical’ optomotor yaw response system, is a large field course control system concerned with flight stabilization against ‘involuntary’ rotation. It enables a fly to fly straight. It responds equally well to front-to-back and back-to-front motion. The other system responds only to front-to-back movement. In free flight, this movement results from the fly’s forward motion in the vicinity of an object. It is called an ‘object response’ system.

The two systems can be separated in the wild-type under appropriate stimulus conditions. A single vertical black stripe rotating around the fly specifically elicits strong object responses, whereas a homogeneously textured rotating background stimulates the large field course control system only (Heisenberg & Wolf, 1984). In the mutant omb, large field course control is blocked, but the object response is close to normal (Heisenberg & Wolf, 1984). Therefore, the remaining yaw responses are not due to compensatory mechanisms, but to a different functional pathway. The object response does not require the HS neurones, it even uses other elementary movement detectors than the large field course control system (Bausenwein, 1984).

The above conclusions are supported by results from a second mutant with a defective lobula plate which is called lobula plate-less^N684(lop;Fig. 10). In this mutant, most columnar neurones of the lobula plate are missing due to their selective degeneration in the first half of pupal life. As a result, only a small rudiment of the lobula plate is present in the adult, but it still contains TmY-cell terminals and a few T4 and T5 neurones (Fischbach, 1983a). The shortage of presynaptic columnar neurones in the lobula plate has a dramatic effect on what presumably are the VS cells and some other neurones of large diameter. They project in a thick bundle through the second optic chiasma into the upper frontal medulla (Fischbach, 1983a; see Fig. 10).

In lop there is no response to roll and pitch (Paschma, 1982). The abnormal growth of the VS neurones, which are thought to mediate the roll response in the wild-type, probably does not compensate for the lack of a normal input. However, lop flies still show significant optomotor yaw responses (Paschma, 1982). The object response does not account for all of it. Large field course control is reduced, but still operates. This may be explained by the presence of the small lobula plate rudiment which still contains the main HS dendrites (Paschma, 1982).

Single gene analysis of chemosensory behaviour has been hampered by the limited knowledge about the general organization of olfaction and taste in insects. The naive concept of such behaviour being a press-button mechanism (sugar-reception -proboscis extension), ignores the problems in evaluating quality and quantity in mixtures of chemicals. Thus, progress from the genetic point of view has largely been confined to the receptor level.

The problem of how to obtain taste mutants has been elegantly solved. Differential feeding on two food sources is conveniently monitored with food-dyes which after ingestion can be seen in living flies (Falk & Atidia, 1975; Tanimura & Shimada, 1981). Again we will not give a detailed description of the various mutants and their phenotypes, but will instead mention two examples which highlight the manner in which differential gene expression specifies neural cell types.

It has been learned from mutant analysis in Drosophila that every sugar receptor cell in the labellar setae contains at least three different sugar receptor molecules. Two mutants have been found in which the sugar cells have reduced sensitivity to pyranose sugars (e.g. sucrose, glucose) while responding normally to fructose, trehalose and other sugars (Isono & Kikuchi, 1974; Siddiqi & Rodrigues, 1980). A putative structural gene for the second receptor, trehalose, has recently been isolated by Tanimura (1984). The third receptor, the one for fructose, can be specifically eliminated by treatment with papain or trypsin. Obviously, Drosophila would be able to distinguish between these sugars, if the genes for the three receptors were expressed in different cells. Such a distinction, however, seems to be of no particular advantage for the fly.

The mutant gust B originally was thought to have a reduced salt sensitivity (Rodrigues & Siddiqi, 1978), but recently gust B flies have been shown to be attracted by salt (instead of being repelled as is wild-type).Ina search for an explanation, Arora & Rodrigues (1983) found that in the setae of this mutant both the salt and the sugar receptor cells show sensitivity to salt. This observation suggests that in gust B the gene responsible for salt sensitivity is expressed in the wrong cell type. This emphasizes the importance of differential gene expression for behaviour.

Olfactory neurogenetics of Drosophila has recently been summarized by Siddiqi (1984). Mutants with specific anosmias have been reported, but so far none of them have been shown to be affected in the structural gene of a receptor protein. The closest may be the mutant olf C. In flies of this strain, there is a reduced response to acetate esters, but normal responses to alcohols or aldehydes. However, it remains to be determined whether this reflects the presence of several genetically independent acetate receptors in the wild-type, as has been proposed by Siddiqi (1984) on the basis of single unit recording in the antenna, or whether the olf C allele is a hypomorph. A broad screen for specific anosmias combined with more refined genetic and physiological tests (e.g. Borst, 1984; Siddiqi, 1984) could determine the number of receptor cell types, the number of different receptor molecules, and their distribution in the sensory cells. This information would be valuable for investigating olfactory behaviour.

The most interesting result emerging from the mutant analysis of biological oscillations is that the mechanisms of long-(e.g. circadian rhythms) and short-term oscillations (e.g. courtship song) share certain components. By now five genes are known which influence the period length of both.

Konopka & Benzer (1971) isolated three alleles of theperiod (per) gene which either shorten (per^s), lengthen (per¹) or abolish (per°) the oscillation period. Homozygous per⁰ animals have a reduced synthesis of octopamine due to a decreased concentration of tyrosine decarboxylase (Livingstone, 1981). The concentration of this enzyme in per⁰, however, is not zero. This and the fact that heterozygous per°/+ females have wild-type enzyme levels is taken as evidence against tyrosine decarboxylase being the gene product of per⁺. One hypothesis is that the per gene may interfere with the development of octopamine synthesizing neurones. per° flies show scattering of certain neurosecretory cells in ectopic positions (Konopka & Wells, 1980). An involvement of neurosecretion is also suggested by transplantation experiments (Handler & Konopka, 1979). Implantation of per^s brains into the abdomen of per⁰ adults sometimes leads to the establishment of a periodicity characteristic of the donor.

Kyriacou & Hall (1980) discovered that the period of the short-term oscillations in the courtship song of male flies is affected by the per locus in much the same way as the circadian rhythm. Preliminary results from mosaic studies indicate that expression of the per⁺ allele in the brain is required for the normal circadian rhythm, while expression in the thoracic ganglion is required for the normal courtship song to occur (Hall, 1984).

Genetic coupling of the period of the song and circadian rhythm is also revealed by mutations in the phase-angle-2 (psi-2), phase-angle-3 (psi-3) and gat genes (Jackson, 1983; Kyriacou & Jackson as cited in Hall, 1984) and in the CLK. gene (Konopka, 1984). However, at least one gene (Andante;Konopka, 1984) is known to affect the period length of the circadian rhythm, but not the song cycle (Zehring & Hall as cited in Hall, 1984). This important finding shows that the overlap in the set of molecular components between the circadian rhythm and the song cycle is not complete.

It should be noted that parameters of biological oscillations can be changed without serious pleiotropic effects in other functions of the organism, e.g. deletions of the per gene are not lethal (Young & Judd, 1978; Smith & Konopka, 1981).

Single gene mutations affecting learning in Drosophila melanogaster are now numerous and may be categorized as ‘biochemical’ or ‘structural’.

Associative and non-associative learning (the latter comprising habituation and sensitization) probably use common biochemical mechanisms (Hawkins & Kandel, 1984). The work of Kandel and co-workers with the mollusc Aplysia has shown that habituation is correlated with a decrease of transmitter release of the habituating synapse. Such a synapse can be sensitized by a heterosynaptic pathway which may use serotonin as a transmitter. Serotonin activates the serotonin receptor which in turn stimulates adenylate cyclase. The subsequent increase in the level of cAMP activates a protein kinase which closes a K⁺ channel by phosphorylation. The decrease in the number of K⁺ channels results in a broadening of action potentials which in turn allows more Ca²⁺ to enter. The resulting high Ca²⁺ concentration causes more transmitter to be released per action potential.

According to Hawkins & Kandel (1984), classical conditioning is an extension of sensitization (conditioned sensitization). This requires that the adenylate cyclase is not only activated via the serotonin receptor, but also by Ca^z+. If this is the case, simultaneous stimulation of the heterosynaptic sensitizing pathway and the primary pathway would result in a synergistic effect yielding high amounts of cAMP. This then causes long-lasting changes in membrane properties.

The isolation of learning and memory mutants in Drosophila (Dudai et al. 1976; Quinn, Sziber & Booker, 1979; Aceves-Pina & Quinn, 1979; Tempel & Quinn, 1980) and the finding that several of them show defects of enzymes playing a role in the basic ‘Aplsia-model’ of learning (Byers, Davis & Kiger, 1981 ; Shotwell & Konopka, 1982; Livingstone, Sziber & Quinn, 1982; Uzzan & Dudai, 1982) has strengthened the hypothesis that basic learning mechanisms may be the same throughout the animal kingdom (Quinn, 1984). Furthermore, mutants in Drosophila may be used to demonstrate coupling between non-associative learning and associative learning ‘mechanisms, and to answer the question whether different molecular mechanisms may coexist for a given form of learning.

Duerr & Quinn (1982) tested habituation and sensitization of the proboscis extension reflex to tarsal stimulation in the mutants dunce, turnip, rutabaga and amnesiac, dunce is the structural gene for the phosphodiesterase II which normally degrades cAMP (Kauvar, 1982) ; turnip blocks the serotonin receptor (Smith as cited in Quinn, 1984), whereas rutabaga shows decreased adenylate cyclase activity (Livingstone et al. 1982). In amnesiac flies the level of adenylate cyclase activity in membrane fractions is higher than normal (Uzzan & Dudai, 1982). The results of Duerr & Quinn (1982) suggest that these mutants, which were selected in associative olfactory learning paradigms, also tend to be defective in habituation and sensitization of the proboscis extension reflex. The mutants dunce, turnip and rutabaga habituate more slowly than wild-type, while sensitization wanes much more rapidly in dunce and rutabaga flies. In amnesiac flies there is an increased threshold for elicitation of the proboscis extension reflex.

Kyriacou &Hall (1984) report that dunce and rutabaga mutants are deficient in what is probably an acoustic sensitization: receptivity of wild-type females is enhanced for some minutes by artificial courtship songs. This does not occur in mutant females. Sensitization wanes much faster than in wild-type. The genetic connection between associative and non-associative learning and the possible applicability of the ‘Aplysia-model’ in Drosophila is exciting, but is the basic molecular mechanism of learning really as general as suggested by these experiments?

A comparison of the mutants in olfactory and visual learning behaviour seems to indicate a qualitative difference. While learning of dunce mutants in the original olfactory learning paradigm (Dudai et al. 1976) is close to zero, the dunce’ and dunce² mutants have been found to learn normally in a visual learning paradigm (Dudai & Bicker, 1978). Folkers (1982) reports a decreased, but still highly significant learning of dunce ¹ in the same visual paradigm. She also tested amnesiac, turnip and rutabaga, and found them all able to learn, although not as well as the wild type Canton-S. In addition, amnesiac, originally characterized as a ‘memory mutant’ (Quinn et al. 1979), remembers in the visual test as well as the wild-type (Folkers, 1982).

The ‘biochemical’ learning and memory mutants have also been tested in a visual paradigm for habituation and sensitization. The landing response of stationary flying Drosophila to visual stimuli habituates readily and can be sensitized, e.g. by actual landing (Fischbach, 1981b). The landing response to unilateral front-to-back motion is also sensitized for some seconds by contralateral stimuli (‘contralateral sensitization’; Fischbach, 1981b). The most apparent effect of the amnesiac, dunce and rutabaga mutations (turnip flies did not fly) on the landing response is a dramatic, stimulus specific decrease in excitability (Fig. 12), not a change in its plasticity. Habituation is about normal in the mutants and contralateral sensitization in rutabaga and dunce flies is neither impaired in its amplitude nor in its time course, although the basic level of responsiveness to front-to-back motion is low in these strains. amnesiac flies could not be tested due to their low overall responsiveness (Fischbach, 1983a).

These results suggest that visual associative and non-associative conditioning may use special mechanisms. ‘Contralateral sensitization’, for instance, may not change the ‘central excitatory state’ (Dethier, 1976). It rather may be a property of the detector system for relative retinal expansion signalling the ‘time to collision’ with an object (Wagner, 1982). The low excitability of the mutants in only certain visual pathways might indicate biochemical differentiation of separate channels of visual information processing.

Irrespective of whether one or several molecular mechanisms will be found to underly learning and short-term memory, it is apparent from the work on molluscs and flies that learning and memory are properties of specific behaviour patterns. This point is clearly made by structural brain mutants. Mutants with different structural impairments in the brain have specific learning defects in different behavioural tasks.

Mutants with reduced mushroom bodies are all deficient in olfactory discrimination learning (M. Heisenberg, in preparation). One of them, the mutant mushroom bodies lleranged^KS65(mbd) has normal colour discrimination learning and normal visuo-fnotor coordination learning (Heisenberg & Wolf, 1984). The structural learning mutants are not insensitive to the conditioned (odours) and unconditioned stimuli (sugar, electric shock). The mutants, mbd and mushroom body miniature^N337(mbm) for instance, are able to perceive and distinguish the odours in the same situations in which they fail to learn. Thus, odour perception and evaluation can be genetically separated from learning of the same odours. Mushroom bodies are not necessary for the first task but are required for the second. These functions, therefore, also seem to be structurally separated. A conspicuous anatomical feature of the olfactory system is that the output fibres of the antennal lobes, which project through the antenno-glomerular tract, bifurcate in the dorsal brain sending terminals into the calyx of the mushroom bodies and into the lateral protocerebrum (Heisenberg, 1980; A. Borst & K. F. Fischbach, in preparation). The first projection area may be involved in learning, the latter in immediate evaluation and behaviour.

Mutants with structural defects in the visual system perform well in the olfactory learning paradigm. For instance, the sol mutant (see above) learns almost as well as the wild-type in the olfactory tasks, but is severely impaired in visuo-motor coordination learning (Götz, 1983 ; Heisenberg & Wolf, 1984) and in habituation and sensitization of the landing response. While habituation is slowed down, sensitization wanes unusually quickly (Fischbach, 1983a). This is comparable to the effects of the dunce and rutabaga mutations on habituation and sensitization of the proboscis extension reflex (see above).

The mutants indicate that behavioural plasticity needs to be understood not only at the biochemical level but also at the structural level. Several questions remain to be answered; for example, what is the relationship of the missing neurones to learning performance? What is the extent of the structural separation of ‘learning’ from primary information processing of sensory information?

The two genders of a species normally show genetically determined differences in behaviour and often also in brain structure. Striking examples are known in a variety of insects, but only in Drosophila is the underlying genetic control beginning to emerge. Aside from courtship, copulation and egg-laying, behavioural differences between males and females in Drosophila are small. A large variety of laboratory tests for visual, olfactory and learning performance, some of which have been mentioned above, give very similar results for males and females. Accordingly, the vast majority of known genes of neurological interest are expressed in both genders. Thus, concerning only a subset of genes and a few items of the behavioural repertoire, sexual differentiation may be of particular interest for studying the relation between genes and behaviour.

Is sex-specific behaviour mediated by specified neural pathways or is it mainly due to (e.g. neurohumoral) modulation of brain function? Extending the view beyond Drosophila one finds ample evidence for the first proposal. Sexual dimorphisms in sensory organs, brain structure and musculature are very common. A thoroughly investigated example is the visual system of some flies. Male Musca and Calliphora catch their mates in flight. They exhibit a special chasing behaviour in which they follow small dark objects from below (Wehrhahn, Poggio & Bülthoff, 1982). This male-specific behaviour has a structural correlate. Female and most of the male ommatidia contain the typical pattern of large peripheral and small central rhab-domeres as explained above for Drosophila. In the dorso-frontal region of the male eye, the ommatidia are different: R7 rhabdomeres are of similar diameter and have the same spectral sensitivity as Rl–6 (Hardie, Franceschini, Ribi & Kirschfeld, 1981 ; Franceschini, Hardie, Ribi & Kirschfeld, 1981).

Apart from this specialization in the eye, there is also a male-specific differentiation in the visual neuropile. The R7 axons of the male-specific region terminate in the lamina (Franceschini et al. 1981 ; Hardie, 1983) rather than in the medulla. Furthermore, in the lobula, certain giant neurones covering the projection area of the dorso-frontal eye region are found in males and not in females (Strausfeld, 1980; Hausen & Strausfeld, 1980). Thus chasing behaviour of males probably is mediated by a special circuitry in the optic lobes. Apparently, colour vision in the dorso-frontal part of the visual field is sacrificed for optimal contrast sensitivity so that the distance at which a female can be detected is increased. Whether this network is a modification of a homologous network in the female, or whether it uses additional neurones is not known.

In Bibionidae, the sexual dimorphism in the visual system is carried to an extreme. Males have large dorsal compound eyes which are not present in females. These eyes do not mediate visual course control (as the ventral eyes do) but seem to be specialized for the detection of small dark objects in the sky (Zeil, 1983a,b).

Another example of sexually dimorphic circuitry in the central nervous system is the antennal lobes of male moths. The antennae of many male moths have a large number of pheromone receptors which send their axons to one large glomerulus (macroglomerular complex, MGC; e.g. Rospars, 1983). Several male-specific neurone types innervating the MGC have been identified (Boeckh & Boeckh, 1979; Matsumoto & Hildebrand, 1981). As in the dipteran eye, these antennal and neural specializations of the male serve to detect the female at the largest possible distance (for review see Bell & Tobin, 1982). A male antenna can grow in an otherwise female organism, if a male imaginal disk is transplanted into a female larva. The transplant induces growth of an MGC and of identified male-specific interneurones innervating it (Schneidermann, Matsumoto & Hildebrand, 1982). What is even more surprising is that in adult moths, the transplant gives rise to the male-specific up-wind flight response elicited by female pheromones (J. G. Hildebrand, personal communication).

In the central brain, sexual dimorphisms are expressed in the size of particular structural subunits. For instance, the mushroom bodies of worker bees are considerably larger than those of the drones. The behavioural correlate of this particular difference is not obvious but may involve learning (Menzel, Erber & Masuhr, 1974). The sexual dimorphism of mushroom bodies in Drosophila is not as obvious as in bees, but it can be uncovered by certain mutations (see below).

It has long been known that sex in Drosophila ultimately depends upon the X: autosome (X: A) ratio. In recent years a set of at least five regulatory genes which are controlled by the X: A ratio have been described. Like a complicated switch, these genes in turn control the whole battery of genes responsible for the expression of male or female traits (Baker & Ridge, 1980). For example, three female-specific genes that code for yoke proteins are regulated at the transcriptional level (Baker, 1984). With respect to behaviour and brain structure, only few, if any, of these effector genes have been studied (Hall, 1981) and their mode of regulation is not known.

J. M. Belote & B. S. Baker (personal communication) observed that the whole male courtship sequence in Drosophila can be induced by a temperature shift in adult flies carrying a temperature-sensitive allele of one of the regulatory genes (transformer-2^tsl ; Belote & Baker, 1982). Chromosomally female (XX) tra-2^tsl flies grown at low temperature ( 16 °C) emerge from the pupa as male-like intersexes which, however, do not display male courtship. After 1 week at high temperature (29 °C), 30 % of the flies begin to behave like males. It is possible that in these flies the ventral ganglion was structurally reorganized. Alternatively, a pre-existing circuitry may merely have been switched on by the lack of functional tra-2 gene product.

The identification of genes responsible for the expression of sexual dimorphisms in the central nervous system (CNS) would be most valuable. Unfortunately, in Drosophila the only known structural difference in the CNS between the two genders is the size of the mushroom bodies, which in the female contains about 8% more Kenyon cell fibres than in the male (see Fig. 2). However, in the mutant mbm (see above) most of the female mushroom body degenerates in late second-or early third-instar larvae. Thus, in mbm, the male flies have mushroom bodies which appear to be normal whereas in females this structure is missing or unusually small (M. Heisenberg, in preparation). This finding suggests that the degree of sexual dimorphism of mushroom bodies in Drosophila is higher than apparent from their gross structural appearance and that the mbm gene is essential for their female expression. It is not clear whether mbm is expressed in one gender only or in both, nor is it known in which cells and at which time of development the gene is expressed. In principle such questions can now be answered with the help of molecular genetics.

An intriguing problem in neurogenetics is the question of how presumably homologous, but recognizably different behavioural subroutines and neuronal networks of different species are expressed in their hybrid offspring. Rarely will the interspecific differences under study be due to differences in only one or a few genes (one such case seems to be the difference in courtship songs of Drosophila melanogas-ter and D. simulans, which map solely to the X-chromosome, Kyriacou & Hall, as cited in Hall, 1984). But bastardization experiments may illuminate aspects of the genotype-phenotype relationship not assessed by the single gene approach. An interesting example is the acoustic communication system of grasshoppers (von Hel-aigrsen, 1975a,b) : The courtship songs of two species, Chorthippus biguttulus and Ch. mollis, are distinguished by number, length and sequence of pattern elements. The songs and their recognition by females normally prevent bastardization, but mating under laboratory condition can produce vital hybrids. Analysis of the song pattern of individual hybrids showed that some pattern elements were intermediate between the parental ones, whilst others were more-or-less independently superimposed. The authors proposed that to ‘some extent independent pattern-generating neuronal networks may be formed during ontogeny, corresponding to the speciesspecific information of the two parental genomes’ (von Helversen, 1975a). A similar result was obtained with the song-specific innate releasing mechanism of females. Hybrid females normally answer only weakly to the song of their hybrid brothers, preferring both or one parental song (Fig. 13). Therefore, their innate releasing mechanism is not intermediate. In some hybrid females both parental releasing mechanisms seem to exist in parallel.

Hybrid female individuals differ in their preference for the parental songs and also in their own song pattern. Interestingly, there is no correlation between their own song pattern and their song preference. Therefore, a functional coupling between song pattern and innate releasing mechanism does not exist in grasshoppers (von Helversen, 1975b).

However, Hoy (1974) and Hoy, Hahn & Paul (1977) report that in hybrids of Teleo-gryllus commodus and T. oceanicus, the Fl females respond best to the hybrid song of Fl males. They therefore postulate a genetic coupling between sender and receiver.

It would be fascinating to complement the results of such bastardization experiments with neurostructural and physiological data.

It is generally agreed that genetics provides useful tools for the analysis of neurobiological and developmental problems. However, the genetic guidance of bra(a development has not been as generally accepted. For example, G. Stent (in Gerhard et al. 1982) states: ‘… the genetic approach to development resembles the quantum mechanical approach to genetics that had some vogue in the 1930s and 1940s’. It seems necessary to point to the simple fact that diversification of phenotypes during evolution is achieved by modifications at the genomic level. This implies that the state of the genome controls the invariances of the developmental process. One aim of developmental biology is to understand how the genetic information is used at different stages of epigenesis.

The present paper has reviewed studies about genetic control of brain structure and function in insects. Two general principles emerging will now be discussed.

The organization of brains does not only reflect functional aspects. Severe constraints are imposed by evolutionary change. For example, often modification of only certain functions is of selective advantage. Therefore, functional subsystems are – at least to some degree – under independent genetic control. This – besides function – limits the number of possible solutions for the design of a brain. It favours a parallel organization.

The notion of a genetic and structural separation of functional subsystems seems to be trivial as far as different sensory modalities are considered. Therefore, we may concentrate our discussion on the extent of the parallel organization of the visual system of insects. ‘Parallel organization’ in our context does not refer to the multiplicity of receptor inputs and the retinotopic organization of the optic ganglia. ‘Parallel organization of functions’ means peripheral separation of information processing channels, and is reflected in the multiplicity of neurones inside single visual columns and is related to the formation of layers in the optic ganglia (Fischbach, 1983a; Buchner & Buchner, 1984). Structural separation of visual functions begins at the receptor level. Trivial examples are those where different regions of the visual field serve different functions, e.g. the dorsal eye region of many insects is specialized for the perception of u.v. light (reviewed by Wehner, 1981), or chasing behaviour of male Musca relies on its specialized dorso-frontal eye region (Franceschini et al. 1981). But separation of functions can also take place at the receptor level when the functions subserve the same part of the visual field. We have seen, for example, that movement detection in bees and flies uses only one receptor type, while colour vision uses all.

Mutants of Drosophila have also shown that different visual functions relying on movement detection are separated at the neuronal level (e.g. landing response, large field course control and object response). In this context it is of interest to mention that the large field course control system is in principle able to mediate a response towards objects (Poggio, Reichart & Hausen, 1981), and vice versa, the object response system is not blind to large field movements (Heisenberg & Wolf, 1984). The formation of apparently redundant parallel subsystems probably secures optimal genetic modulation of the respective functions. The visual system is not a single ‘parallel computer’ with different context dependent states. It consists of several, structurally distinct information processing subsystems which are retinotopically organized.

This emerging picture of the organization of the visual system is partly derived from studies of visual mutants. In many cases, these have indicated for the first time a separation of certain functional pathways. Retrospectively, we notice that it is the parallel organization of functions which enables specific mutations.

The insect nervous system and its organization into functional pathways requires an abundance of neuronal cell types characterized by their connectivity and physiology. The differentiation of neurones involves gene regulation. Different sets of genes are expressed in different neurones. This is obvious for sensory neurones (e.g. the trehalose receptor molecule in the sugar receptor or the opsin in retinula cells) and for neurone-specific transmitter metabolism (not discussed in this review, but see for instance Hall, 1982). Impairment of certain neurone types can be due to mutations in the structural gene specifying a certain (neuronal) function and in regulatory genes which specify the ‘tissue address’ for a structural gene. A possible example for the latter case is the gust B mutation which causes salt sensitivity to be expressed in sugar receptors (see above).

So far, in none of the structural brain mutants discussed has the primary gene product of the wild-type allele and its role in brain development been identified. Many of them lead to degeneration of ganglion cells, a process which in the case of the sol mutant has been shown to be tissue-(most probably cell-) autonomous (Fischbach & Technau, 1984). Here, the final differentiation of neuronal types seems to be affected. Other mutants (e.g. omb) are presumably cell lineage mutants similar to those isolated in Caenorhabditis (Sulston & Horvitz, 1981). They may help to identify the genetic basis of the progressive determination of cell types during development.

This review has emphasized that brain development and neuronal functions are under tight genetic control. It may thus be justifiable to speak of genetic control of behaviour. However, we propose not to use this term, as it is easily misunderstood. In any behavioural situation, organisms are not puppets on genetic strings. Genes do not control actual behaviour. Once a brain has developed, it generates behaviour. Genetic functions are still required, but now control is rather the other way around: expression of certain genes may well depend on what is perceived and learned or done by the organism.

We are grateful to Drs A. Borst, E. Buchner, P. Coombe and H. Mariath for discussions and critical reading of the manuscript. C. R. Götz and A. Dittrich prepared the figures and contributed much to the working atmosphere. We also thank H. Karwath, G. Kruschel, C. R. Götz, M. Weltner, D. Richter and G. Schaflein for participating in our screens for structural brain mutants. The reconstructions of the HS cells of Musca and Calliphora are a kind gift of Dr K. Hausen. The work of the authors was supported by the DFG Grants Fi 336/1-3 and He 986/5-3.