Origin and specification of type II sensory neurons in Drosophila

The peripheral nervous system (PNS) of Drosophila embryos has provided many insights into the genetic mechanisms of how neural precursor cells are determined and assume a particular developmental pathway of differentiation (for reviews see Campuzano and Modellel, 1992; Ghysen and Dambly-Chaudiere, 1993; Jan and Jan, 1993). The model that has emerged suggests that a given area of the early ectoderm where sensory organs will develop becomes competent for producing neural precursors due to the action of ‘proneural’ genes, such as atonal and genes of the achaete-scute-Complex (AS-C) (e.g. Cabrera et al., 1987; Dambly-Chaudiere and Ghysen, 1987; Romani et al., 1989; Ruiz-Gomez and Ghysen, 1993; Jarman et al., 1993). Some of these genes are expressed in a limited number of ectodermal cells endowing each of these cells with the potential to become a neural precursor. Another set of genes, the ‘neurogenic’ genes, such as Notch and Delta, are then required to limit the number of cells that will become neural pre-cursors to one per cluster (Hartenstein and Campos-Ortega, 1986; Goriely et al., 1991; Bodmer et al., 1993; for review on neurogenic genes see Artavanis-Tsakonas and Simpson, 1991; Campos-Ortega, 1993). Once a neural precursor has been singled out, selector-type genes, such as cut and poxneuro (poxn), are required to initiate the correct developmental program of a particular type of sensory organ (Bodmer et al., 1987; Blochlinger et al., 1991; Dambly-Chaudiere et al., 1992). Other genes are then responsible for specifying cellular identi-ties of sublineages of neural precursor cells. The gene numb, for example, is required for the correct specification of second order precursor cells (Uemura et al., 1989; Rhyu et al., 1994; for other genes affecting PNS development, see Salzberg et al., 1994).

Type I sensory neurons in Drosophila innervate the sensory organs to which they are related by lineage (Bodmer et al., 1989; Hartenstein and Posakony, 1989). Each of these sensory organs is thought to derive from a single ectodermal precursor (SOP) which gives rise to one or several monodendritic neurons and several support cells. Type I sensory organs have been classified into two major classes: mechano- or chemosensory organs that have external sensory structures in the cuticle such as bristles, campaniform and basiconical sensilla (es organs) and chordotonal organs that are internally located stretch receptors (ch organs). In addition, the larval PNS also contains numerous type II neurons that have multiple dendrites (md neurons, Ghysen et al., 1986; Bodmer and Jan, 1987). In contrast to es and ch neurons, md neurons (with one exception) do not seem to be associated with support cells. Md neurons are thought to function as stretch or touch receptors. Three different subclasses of md neurons have been distinguished based on their morphology (Bodmer and Jan, 1987): md-da neurons are the most abundant subclass and have extensive subepidermal dendritic arborisations, md-bd neurons have bipolar dendrites and md-td neurons extend their dendrites along tracheal branches. The origin and lineage relations of these cells with other PNS cells has not been established. Moreover, the genetic basis for md neuron differentiation is poorly understood.

Mutations in some genes involved in sensory organ (es and/or ch) development have also been shown to affect the formation of md neurons. It is not clear, however, if md neurons are affected independently of sensory organs or as a consequence of alterations in sensory organ development. ch and es organs are absent in atonal and AS-C mutants, respectively. Many md neurons are also missing in these mutants (Dambly-Chaudiere and Ghysen, 1987; Jarman et al., 1993). It is possible that md neurons have their own atonal- or AS-C-dependent precursor cells or, alternatively, md neurons may derive from es or ch organ lineages. In the latter case, a failure to recruit es and ch SOPs will automatically result in the loss of md neurons.

In numb mutants, a normal number of es or ch SOPs are formed, but the second order precursor cells give rise mainly to support cells instead of neurons and their glial-like sibling cells. Many md neurons are also missing in numb mutants (Uemura et al., 1989; Rhyu et al., 1994). The neural selector gene, cut, is required for specifying es organ identity: in cut mutant embryos, the number of sensory organs and md neurons is unchanged but es precursor cells develop as ch organs (Bodmer et al., 1987). In addition to the es organs, about two thirds of the md neurons also express cut (Blochlinger et al., 1990), many of them are in close physical association with es organs. Interestingly, in deficiencies of AS-C all Cut-positive md neurons are absent, whereas, in atonal mutants, a complementary subset of (Cut-negative) md neurons is missing (Dambly-Chaudiere and Ghysen, 1987; Jarman et al., 1993; R. B. unpublished). In order to better understand the role of these (and other) genes in the specification of md neurons, it is necessary to know their origins and lineages.

Observation of cell division patterns of sensory organs in other insects had suggested that the cells within an es or ch sensory organ derive from a common precursor that divides near the organ’s final location (e.g., Wigglesworth, 1953; Lawrence, 1966; Jagers-Rohr, 1968; for review see Bate, 1978). Due to the relatively small size of Drosophila cells, it has not been possible to confirm these lineages by simple visualization of SOP divisions. BrdU labelling of Drosophila embryos provided further insights concerning the division pattern of the putative SOPs (Bodmer et al., 1989). Exposure to BrdU at progressively later times resulted in fewer and fewer labelled cells within the postulated sensory organ lineages, since many precursor cells had already completed their last S-phase (i.e. last possibility for BrdU incorporation) at the time of exposure. A likely pattern of SOP division was inferred from the patterns of BrdU-labelled cells (Bodmer et al., 1989). Md neurons were shown to be generated around the same time as type I SOPs, and it was speculated that they derive from type II ectodermal precursors that divide close to their final position. The main drawback of BrdU incorporation studies with respect to SOP lineages stems from the inability to mark individual SOPs. Since all dividing PNS cells get labeled, the interpretation of sensory organ lineages heavily relied on the underlying assumption that the cells within a sensory organ lineage derive from a common precursor.

In order to reassess the lineages of type I sensory organs and determine the lineages of type II (md) neurons in the PNS of Drosophila, we have used the flp/FRT site-specific recombination system of yeast (Golic and Lindquist, 1989; Struhl and Basler, 1993). Using this method, we show that each type I sensory organ derives indeed from a single ectodermal precursor cell. In addition, we have determined that many of the md neurons are part of the type I sensory organ lineages. Our results indicate that one subset of the md neurons is related to es organs, a second subset is related to ch organs and a third set does not appear to be related to either es or ch organs. Consistent with these proposed lineage relationships, we find that, in numb mutants, the sensory organ related md neurons are usually absent or transformed into support cells. We also show that, in cut mutant embryos, the identity of AS-C-dependent md neurons (many of which are related to es organs) has changed to what seems to be characteristic of a subset of atonal-dependent md neurons. This suggests that AS-C and cut are not only required for es organ formation, but also for specifying a subset of md neurons.

The fly stocks used to generate lacZ-positive clones were: hsp 70-flp (flipase construct inserted on the X chromosome, Struhl and Basler, 1993) and Act-Draf-nuclacZ (construct inserted on the third chromo-some; Struhl and Basler, 1993). The Act-Draf-lacZ stock has a constitutive actin promoter separated from a lacZ reporter gene by a segment of DNA that contains a transcriptional stop codon and is flanked by two FRT repeats. The presence of Draf in this construct is incidental. The transformant flies were provided by G. Struhl.

In order to determine whether md neurons are affected in numb mutants, numbⁿ⁷/CyO males (Uemura et al., 1989) were crossed to females from the E7-3-36 P-element enhancer trap line (second chromosome insertion), which marks all md neurons (Bier et al., 1989). Since the numb mutation and the E7-2-36 P-element insertion are both on the second chromosome, recombinants from this cross were selected (numb, E7-2-36/CyO).

Transformations of md neurons in cut mutants were assessed using the enhancer trap line, E7-3-49 (third chromosome insertion), which marks a subset of md neurons (Bier et al., 1989). This line was crossed into a cut mutant background of the genotype: yw,ct^db7/FM7c. ct^db7 is a small deletion in the cut gene of ∼1 kb that is homozygous lethal and considered a null mutation (Blochlinger et al., 1988).

lacZ-expressing clones were generated using the yeast flp/FRT site-specific recombination method according to Struhl and Basler (1993). Homozygous hsp 70-flp virgin females were crossed to Act-Draf-nuclacZ males. F₁ embryos, hemizygous for hsp70-flp and heterozy-gous for Act-Draf-nuclacZ, were collected on grape plates with a dab of yeast and submitted to various levels of heat shock (28 °, 30 °, 32 ° or 34 °C) for 30–40 minutes, at one of the following developmental stages: 2–4 hours, 4–6 hours or 6–8 hours of development after egg laying. Heat shocks were administered either in a water bath or in an incubator in which water had been prewarmed to the desired temperature. The embryos were then allowed to develop at 18 °C until they reached 14-16 hours of development. They were then fixed and stained with antibodies. In order to avoid scoring the same clones twice, the coordinates of each PNS clone were taken. Embryos submitted to a mild heat-shock treatment (30 °C) had less than 20 clones and on average less than one clone per embryo that included PNS cells embryos heat-shocked at 32 °C had on average two or three PNS-containing clones. These two treatments were used in the vast majority of the collected data. An excess of 5000 embryos was scored. To test the efficacy of the flipase-mediated recombination, embryos were heat shocked at 37 °C for 30-40 minutes or given two consecutive 20–30 minute heat shocks at 37 °C with an interval of 30 minutes at room temperature. With this regime, embryos expressed lacZ in virtually all cells. Another control consisted of raising the embryos at 18 °C until they reached 14–16 hours of development without heat shock. These embryos had very few lacZ-expressing clones (data not shown). Embryos bearing only the Act-Draf-nuclacZ construct were also tested for lacZ expression (after heat shock for 30 minutes at 32 °C). These embryos did not express the reporter gene.

hsp70-flp;Act-Draf-nuclacZ embryos or mutant embryos (for cut and numb) marked with enhancer trap lines E7-2-36 or E7-3–49 were double labeled with a rabbit antibody against β-galactosidase (β-gal, Cappel, 1:2000) to monitor the lacZ-positive clones or md neurons, and one of the following antibodies: monoclonal antibody 22C10 (1:100; all PNS neurons marked), monoclonal antibody 21A6 (1:10, labels scolopales and dendritic caps; Zipursky et al., 1984), anti-Prospero antibody (1:4, stains all thecogen and scolopale cells; Spana and Doe, personal communication) or the antibody RK2 (1:500, stains glial cells and ligament cells; Campbell et al., 1994). Embryos were then incubated with the appropriate HRP-coupled secondary antibody (Biorad). The lacZ construct that we have used confers nuclear localization of this β-gal reporter gene product. Since the intensity of the anti-β-gal staining tended to mask that of 22C10, the two antibodies were used sequentially. The staining procedure is essentially the same as described in Bodmer and Jan (1987). In some cases, the DAB (Sigma) color reaction for one antibody was carried out in the presence of 8% nickel chloride, which results in a black product that can then be distinguished from the brown DAB product. numb homozygous mutants were recognized using the following criteria: absence of most peripheral neurons (determined with 22C10 anti-bodies) or absence and disorganization of md neurons (determined with anti-β-gal staining). cut mutant embryos were recognized by the transformation of dendritic caps into scolopales (determined with 21A6 antibodies, see Bodmer et al., 1987).

The PNS of Drosophila embryos is composed of numerous es and ch organs as well as md neurons that are arranged in a segmental, highly stereotyped fashion, allowing reliable identification of all cellular components (Fig. 1A,B; Ghysen et al., 1986; Bodmer and Jan, 1987; Hartenstein, 1988). We have studied cell lineages in the PNS by inducing random clones in the early embryo using a site-specific recombination method from yeast, the flp/FRT system (Golic and Lindquist, 1989; Struhl and Basler, 1993). Small clones of lacZ-positive cells were generated randomly in embryos, before or during the division of PNS precursors (Fig. 1C, see also Materials and methods). These clones were induced by activation of a site-specific recombinase, flipase (flp), which fuses a constitutive promoter (of actin) to the coding region of a lacZ reporter gene (constructs and transgenic flies of this system are described in Struhl and Basler, 1993). Prior to flipase expression, these sequences are separated by two flipase recombination target sites (FRTs) in between which a stop codon is located. The lacZ gene product can therefore only be detected in the cells in which the intervening FRT/stop/FRT sequence has been recombined away. A heat-shock promoter-flp construct was used to control the timing and the level of flp expression. A flp-mediated recombination event can occur at any time after flp induction, thus marking all or a subset of the cells that belong to a given lineage.

We reasoned that comparison of larger with smaller clones should enable us to determine the sequence of cell divisions of a given SOP. Since the SOPs of the embryonic PNS start dividing between 5 and 7 hours of development, we heat-shocked embryos at blastoderm (2-4 hours of development) to maximize the labelling frequency of SOPs, or after gastrulation (4-6 or 6-8 hours of development) expecting to label a higher proportion of SOP sub-lineages (Fig. 2). Embryos were aged to 14-16 hours of development (stage 16, Campos-Ortega and Hartenstein, 1985) and stained for lacZ expression. All cells in clones that included PNS cells had approximately the same level of staining of lacZ expression, which indicates that the flipase was not expressed preferentially in any given cell type (a representative embryo is shown in Fig. 1C). To help identify PNS clones, the embryos in some experiments were double labelled with the monoclonal antibodies 22C10 (a neuronal membrane marker) or 21A6 (which stains the dendritic caps of es organs and the scolopales of chordotonal organs; Zipursky et al., 1984; see Bodmer et al., 1987).

First we examined clones containing only es or ch organ cells, because the outcome of these studies could be compared with the findings of previous lineage studies of the embryonic PNS (Bodmer et al., 1989). The cells of es organs tend to be tightly grouped together. We chose to concentrate on the lineages of two dorsal es organs (desC/D) and one lateral es organs (lesC) since individual cells in these organs are spaced further apart (see Fig. 1A for exact location within a segment). The neuron was easily recognized with the use of the 22C10 marker and the neuron-associated glial cell, the thecogen, was localized on the basis of its vicinity to the neuron along its dendrite. The support cells, the tormogen and trichogen, were distinguished based on their position adjacent to the 21A6-positive dendritic cap, which is located at the tip of the dendrite at the base of the future cuticular structure associated with the sensory organ (Hartenstein, 1988).

Three types of clones were observed in desC/D and lesC (Fig. 3A-D; Table 1). 42 clones were composed of all es organ cells: neuron, thecogen, tormogen and trichogen (Fig. 3A,B). This indicates that the cells within es organs in Drosophila embryos are indeed clonally related. Many of these clones, however, also included a few labelled ectodermal cells (asterisks in Fig. 3) indicating that the recombination event took place in an ectodermal cell that underwent one or more rounds of cell divisions before the SOP was formed. 13 clones were composed of neuron/thecogen (Fig. 3C) and 10 clones were composed of tormogen/trichogen (Fig. 3D; Table 1). These findings are consistent with the previously reported BrdU data (Bodmer et al., 1989) indicating that es organs are generated through two sets of cell divisions: neuron/thecogen derive from one SOP daughter cell, tormogen/trichogen from the other (Fig. 5A). Individually labeled es organ cells were also observed, consistent with a late recombination event (see Fig. 2).

An abdominal segment is composed of three single ch organs (v′ch1, vchB and vchA) and a cluster of five ch organs (lch5) (Fig. 1A). Unlike most es organs, the cells of ch organs are neatly arranged in a row. Whereas the ligament cells of the ch organs (scolopidia) in the lch5 cluster are clearly visible, this is not always the case for the ligament cells in v′ch1, vchB and vchA. We therefore restricted our analysis of ch lineages to the lch5 cluster.

The majority of the clones that we observed in the lch5 cluster (82 cases) were composed of all four cells within one scolopidium (ligament, neuron, scolopale and cap cell). These clones frequently included a few ectodermal cells, as was also observed for es organ clones (Figs 1C, 3E,F; Table 1). This argues in favor of a clonal relationship between the cells of individual ch organs. 34 clones consisted of the neuron and scolopale cell (Fig. 3H; Table 1) indicating that they are sibling cells. 25 cases clones were composed of three cells that consistently included neuron, scolopale and ligament cell (Fig. 3G; Table 1), which is indicative of a serial mode of ch SOP division: the cap cell is produced first then the ligament cell and finally neuron and scolopale cell (Fig. 5D). In this model, the ligament/neuron/scolopale derive from a common second order precursor. BrdU-labelling studies also suggested a serial division pattern for ch SOPs; however, the ligament cell was thought to be produced first. This sequence of cell divisions implies that cap/neuron/scolopale are generated from a common second order precursor (Bodmer et al., 1989). These two contradictory observations can be reconciled in a model where the first ch SOP division produces two second order pre-cursors, one that gives rise to the ligament/neuron/scolopale cell lineage, and the other that generates the cap and another (ectodermal) cell. This other, cap/ectodermal cell precursor, then replicates somewhat later than the first division of the secondary precursor for the ligament, neuron and scolopale cell (see Fig. 5D). Consistent with this model is the observation that an ectodermal cell, dorsal to the cap cells, was often co-labelled in clones that expressed lacZ in all cells of a scolopidium (ec-labelled cell in Fig. 3E). In 40% of the cases, where only the cap cell was labelled to the exclusion of other ch cells, this dorsal ectodermal cell was also labelled (Table 1). Indeed, two ectodermal cells have been identified in the position of these cap-related ectodermal cells as attachment cells for lch5 (Matthews et al., 1990; see also Fig. 1A). In many cases, however, it seems the cap-related ectodermal cells fail to differentiate in this ectodermal position (perhaps because they degenerate), thus escaping detection at the stage of analysis. In addition, we have observed one or two labelled ectodermal cells in embryos that have incorporated BrdU at the time of PNS neurogenesis (R. B. and R. B., unpublished). These cells were in the same position as the ch attachment cells and only labelled when cap cells were also labelled. These observations further support the proposed model of ch lineages (Fig. 5D).

A considerable number of clones (29) observed in the lch5 cluster consisted of two or more fully labelled scolopidia (individual ch organs) (data not shown). Since the frequency of clones observed in the PNS is relatively low, most if not all of these larger clones are probably not due to independent recombination events. This raises the possibility that two or more scolopidia within the lch5 cluster derive from a common precursor. Alternatively, the labelled scolopidia may have originated from a recombination event in an early ectodermal cell (at blastoderm stage), which, after further cell divisions, gave rise to independent ch SOPs (within the lch5 proneural region). Clones of less than four cells per ch organ were never observed when more than one scolopidium of lch5 was labelled. Therefore, it is unlikely that swapping of equivalent cells amongst neighboring scolopidia occurs with appreciable frequency.

Md neurons constitute the third class of sensory neurons in the Drosophila PNS. The majority of these neurons have extensive dendritic arbors below the epidermis. Md neurons are easily identifiable by location and shape, using the 22C10 marker. Some md neurons are in close association with es or ch organs (see Fig. 1A). We explored the possibility that these md neurons could be related by lineage to their neighboring type I sensory organs. A few PNS cells have been shown to migrate away from their point of origin. For example, lch5 originates in the dorsal region and ends up more laterally (Bodmer et al., 1989; Salzberg et al., 1994), and v′ch1 originates in a ventral cluster and migrates dorsally (Ghysen and O’Kane, 1989). Therefore, we scored all subectodermal cells that were labelled close to and at a distance from a labelled md neuron. Our results suggest that md neurons derive from three types of lineages (summarized in Fig. 5). One type of lineage probably gives rise to md neurons exclusively (termed ‘solo-md’ lineage), whiles the others produce md neurons and es organs (‘md/es’ lineage) or md neurons and ch organs (‘md/ch’ lineage).

Several md neurons of the da subclass are located adjacent to es organs (Fig. 1A). In practically every case (129 out of 131 cases), when all cells of one of these es organs were labelled, the neighboring md neuron was also labelled (Table 1): lesA/ldaA (27 cases, Fig. 4A), lesB/ldaB (21 cases, Fig. 4B), v′esB/v′ada (33 cases, Fig. 4C), v′es2/v′pda (68 cases, Fig. 4D). This strongly suggests that a subset of md neurons are descendants of es SOPs. Since lesA, lesB and v′esB are monoinnervated es organs and v′es2 is a polyinnervated es organs, we will refer to these lineages as md/mono-es and md/poly-es, respectively.

In order to determine the pattern of cell division of the cells belonging to the md/mono-es lineages, we scored clones in which only part of the md/es cells were labelled. For v′esA/v′ada, 2 cases consisted of the es neuron, thecogen and md neuron; and for ldaA/lesA, 4 cases consisted of the es neuron and md neuron (Table 1). This suggests that the md neuron derives from a second order precursor, which also generates the es neuron and thecogen, probably by an additional division of the neuronal precursor. Due to the small number of such clones, the order in which the md neuron, es neuron and thecogen cell are produced could not be determined unequivocally.

The situation is similar for poly-es neurons (Table 1): in 12 cases, we observed that the md neuron (v′pda) was co-labelled with the thecogen cell and the two es neurons of v′es2. This suggests that, in md/poly-es lineages, the md neuron is also generated from a second order precursor cell derived from the first division of the v′es2/v′pda precursors. In 78 clones, the md neuron and both es neurons co-labelled exclusively (Fig. 4E). Co-labelling of the es neurons and the thecogen to the exclusion of the md neuron was almost never observed (Table 1). This means that, in the case of md/v′es2, the md and es neurons are generated by additional divisions of the neuronal precursor. The two v′es2 neurons were stained alone in 17 cases (Fig. 4F) and the md neuron was labeled alone in 27 cases (Table 1), suggesting that the es neurons are siblings. In 5 cases, we observed that only the tormogen and trichogen were co-labelled suggesting that they are also siblings (Table 1). We conclude that the first SOP division gives rise to the tormogen/trichogen precursor and the precursor for the thecogen cell and the md/es neurons. The latter secondary precursor then generates the thecogen cell and the precursor for the md and es neurons. These lineages are consistent with the BrdU-labelling studies, which show that the tormogen/trichogen precursor replicates first, followed by the precursor of the thecogen and the neurons. In the case of md/v′es2, the precursor of the two es neurons replicates last. It is likely that the division pattern for an md/poly-es SOP is essentially the same as the division pattern that generates md/mono-es, except that the precursor for the neurons of v′es2 undergoes another division to generate two es neurons. The proposed lineages for md/mono-es and md/poly-es are summarized in Fig. 5B,C.

Two types of clones that contained co-labeled md neurons and ch organs were observed: one type of clone was composed of the ventral v′td2 (an md neuron belonging to the td subclass) and vchA; the other type of clone included the dorsal v′td2 neuron and vchB (see Fig. 1A). The v′td2 neurons are located at some distance from vchA or vchB in the v′ cluster. Table 1 (bottom half) shows the number and types of chordotonal organ-associated md clones that we have observed. Since the ligament cells of vchA and vchB were not easily visible in most clones (too close to the ch neurons), we excluded them from the analysis. Clones composed of an md neuron and all cells of a vch organ were observed in 82 cases (Fig. 4H,I; Table 1). These data suggest that a subset of md neurons derives from ch organ lineages. The other abdominal ch organs, lch5 and v′ch, do not seem to have lineage relations to md neurons. In some embryos, we observed vpda, an md neuron belonging to the da subclass, co-labelled with v′td/vchA clones or with clones including both v′td/vchA and v′td/vchB organs (Fig. 4G,H). As discussed below, this probably means that an early recombination event generated several neighboring PNS precursors, similar to the situation with lch5.

In order to determine more precisely the relation between vpda and v′td2 to vch organs, we scored smaller clones. 29 md/vchB clones were composed of the dorsal v′td2 and a partially labelled ch organ (i.e. the neuron and scolopale were labelled to the exclusion of the cap cell). In most of these cases, the ch neuron alone was co-labelled with the dorsal v′td2 (Fig. 4J). In contrast, only 3 cases showed the vchB neuron and scolopale cell not co-labelled with the dorsal v′td2. These observations suggest that the md/vchB lineage is likely to be similar to other ch organ lineages, the only difference being that the ch neuron undergoes another cell division to generate an md neuron (Fig. 5E). The md/vchA lineage is most likely identical to the md/vchB lineage with respect to the td neuron (Table 1), but it is unclear if the other vchA-associated md neuron, vpda, is indeed part of the md/vchA lineage. Only 13 out of 25 cases show co-labelling of vpda with the other md/vchA cells. This means that vpda is either generated first in the md/vchA lineage or that it arises from an independent ectodermal precursor in close proximity to the md/vchA precursor (Fig. 5E). Consistent with the latter interpretation is the observation that vpda is occasionally co-labelled with a vchB clone (5 cases) and often by itself (40 cases) (Fig. 4M). We classify vpda tentatively as a solo-md neuron (see below).

A number of md neurons were not usually or consistently colabelled with other sensory organ cells. These md neurons are the dorsal bd neuron (dbd), the da neurons in the ventral vmd5 cluster and possibly a few da neurons in the dorsal cluster (see Fig. 1A). 48 clones were observed that labelled the dbd neuron and/or its neighboring glial cell. 20 of these clones label both the dbd neuron and the glial cell, excluding other PNS cells (Fig. 4K), which suggests that they are sibling cells and that they derive from their own ectodermal precursor (Fig. 5F, top). This is further supported by BrdU-labelling studies, which showed that these cells divide around the same time and are thus likely to be siblings.

The clones that include vmd5 neurons were quite heterogeneous, varying between 1 and 4 labelled vmd5 neurons (Fig. 4L), often in conjunction with ves organs. In 19 cases, vmd5 neurons were labelled to the exclusion of other PNS cells and, in 19 other cases, vmd5 neurons were co-labelled with vesA, vesB or vesC. There did not seem to be a consistent pattern in which es organs were most frequently co-labelled with vmd5 neurons. Although we can not rule out a complex lineage relationship of ves organs and vmd5 neurons, it seems most likely that the vmd5 neurons are generated from md-specific precursors, in the vicinity of ves SOPs (Fig. 5F). Some or all of the dorsal md neurons are probably generated by a similar mechanism although clones including these cells have not been quantified (since they are difficult to identify individually).

It has been suggested that in numb mutants the first SOP division generates two identical second order precursors, which results in the overproduction of sensory organ support cells (tormogen/trichogen) at the expense of neuron/thecogen cells (Uemura et al., 1989; Rhyu et al., 1994). Our lineage analysis suggests that md neurons related to es or ch organs are descendants of the neuronal second order precursors (Fig. 5B,C,E). In numb mutants, we would thus expect that these SO-related md neurons are absent. To test this hypothesis, we have crossed an enhancer trap line, which expresses the lacZ reporter gene exclusively in all md neurons (E7-2-36; Bier et al., 1989), into a numb mutant background. Embryos doubly labelled for all PNS neurons and for lacZ expression show that the sensory organ-associated md neurons are usually missing in numb mutants (Fig. 6A,B). Although this observation is consistent with a lineage relationship between a subset of md neurons and sensory organs, one can not rule out that the md neurons are affected independently by numb as well. This is supported by the observation that the solo-md neurons, vpda and some vmd5 (and several dmds) were often missing in numb (Fig. 6B).

The transformation of the neuron and thecogen cells into tormogen/trichogen-like cells is frequently incomplete (see Uemura et al., 1989). We reasoned that, if the es-related md neuron (md/es) and the es neuron derive from the same second order precursor lineage, both es and md/es neuron should either be present (incomplete transformation) or absent (complete transformation). Embryos, doubly labelled for all PNS neurons and for lacZ reporter gene expression in md neurons, show that this prediction was correct: both the es and md/es neurons either did or did not express neural markers simultaneously (Fig. 6B). Similar observations were made in numb mutant embryos that were doubly labelled for md neurons and thecogen cells: every time an md/es neuron was present the neighboring thecogen cell was also labelled (Fig. 6C,D). Interestingly, two or three cells positive for the thecogen marker were frequently present in the location of md neurons/mdrelated es organs (indicated by brackets in Fig. 6D). This suggests that the first SOP division appears to be normal in some cases, but the secondary precursor for the thecogen and neurons now generates only thecogens (Fig. 6G, top panel). These results also indicate that numb not only acts in generating asymmetry during the first SOP division but also during both secondary SOP divisions (see also Rhyu et al., 1994). Since our md marker is only weakly expressed in md-td neurons associated with ch organs, we have not examined their fate in numb mutants. Taken together, our analysis of numb mutants strongly supports the conclusions from the lineage studies.

To determine if numb also affects other md lineages, we examined the fate of solo-md neurons in these mutants. We observed only 2–3 neurons in vmd5 or the dorsal cluster. Due to the lack of other markers, we do not know if a hypothetical solo-md precursor divides less or if the fate of its progeny is altered. The md-bd neuron of the dorsal cluster is also absent in numb mutants (Fig. 6E,F). Our lineage studies suggest that dbd is generated from a dbd-specific ectodermal precursor, which divides once to give rise to a glial cell and dbd (Fig. 6E). In numb mutants, instead of one dbd neuron and one glial cell, two glial cells are formed at the expense of the dbd neuron (Fig. 6F,G, lower panel). Therefore, numb not only affects the es and ch lineages but also other components of the PNS.

It had previously been shown that in numb mutants an excess of cap cells are produced but the fate of the ligament cell was not clear. Since lch5 neurons are rarely formed in numb mutants (Uemura et al., 1989), we expected that the second order precursor is often transformed into a cap-like precursor. Consistent with our proposed ch lineages, we find that in lch5 not only neurons and scolopales are missing but also the ligament cells (Fig. 6E,F).

The da subclass of md neurons appears to be a morphologically homogeneous population of cells (Bodmer and Jan, 1987). The observation that these cells are generated by different types of lineages (this study) and that the axonal projections to the CNS are not identical for all da-md neurons (Merritt and Whitington, 1995), raises the possibility that there are distinct subpopulations of da neurons. To address this question and identify genes that may play a role in the specification of md neurons, we sought for cell markers that differentially label subsets of da neurons. One such marker is the cut gene product. cut is first expressed in es SOPs and later in their progeny including the es-related md-da neurons (see Figs 5B,C, 7C; Blochlinger et al., 1990). Since cut functions as a developmental switch to specify the correct identity of es organs (Bodmer et al., 1987; Blochlinger et al., 1991), we wondered if cut may also be involved in the specification of da neurons. In cut mutants, the morphology and position of md neurons is not affected (data not shown). To test for more subtle changes, we used an enhancer trap line, E7-3-49 (Bier et al., 1989), which marks a subset of md neurons that is non-overlapping with the Cut-positive da neurons: vpda, dbd and 3-4 dorsal da neurons (Fig. 7A,D). At least one of the cells marked with E7-3-49 (vpda) is atonal-dependent (Jarman et al., 1993).

The lacZ expression of the E7-3-49 enhancer trap line is dramatically altered in cut mutant embryos. In addition to the cells in which it is normally expressed, β-gal staining is now observed in the da neurons that are normally Cut positive: all dorsal cluster da neurons, ldaA&B, v′ada, v′pda and 3-4 neurons of vmd5 (Fig. 7A-E). We conclude from these observations that there are at least two subclasses of da neurons, Cutnegative and Cut-positive da neurons, and that cut is likely to be involved in specifying the identity not only of es organs but also of a subpopulation of da neurons (Fig. 7F).

The analysis of patterns of cell division in a number of insects and BrdU-labelling studies in Drosophila embryos and wing imaginal discs have provided evidence that the cells within individual es and ch organs are derived from single precursor cells (reviewed in Bate 1978; Bodmer et al., 1989; Hartenstein and Posakony, 1989). The pattern of BrdU incorporation in embryos was suggestive of a cell division pattern that was different for es organs and for ch organs (Bodmer et al., 1989). Using the yeast flipase method to generate small clones in the embryo (Golic and Lindquist, 1989; Struhl and Basler, 1993), we have reassessed the lineages of type I sensory organs (es and ch) and examined the lineages of type II sensory neurons (md neurons) and their relationship to type I sensory organs, by scoring an excess of 5000 embryos.

We scored clones that included cells of the dorsal most es organs (desC/D) and the lateral ch organs (lch5). We observed that a majority of clones were composed of all cells belonging to an es or ch organ, which indicates that es and ch organs derive from individual SOPs. The cellular compositions of clones that included only a fraction (but more than one) of labelled cells within an individual es or ch organ allowed us to infer the most likely lineage relationships. For es organs, we observed co-labelling of either the neuron/thecogen or the tormogen/trichogen, confirming the previously proposed division pattern for es SOPs (Fig. 5A; Bodmer et al., 1989). For ch organs, we observed the following combinations of labelled cells: ligament/scolopale/neuron, cap/attachment cell or scolopale/neuron. This suggests a division pattern for ch organs that is a modification of what had been previously proposed, but which is also consistent with the BrdU studies (illustrated in Fig. 5D). In this model, the two second order precursors replicate at different times: the ligament/ scolopale/neuron precursor does so before the cap and attachment cell precursor. In agreement with the proposed ch lineage pattern is the finding of several clones in which the cap and attachment cell are co-labelled (Table 1), indicating that the recombination event took place in the immediate precursor of these two cells.

A relatively large number of multiple ch organs was labelled in lch5 (the lateral chordotonal cluster), suggesting that ‘super SOPs’ could give rise to more than one ch organ. It has previously been suggested that there are at least two ch organ precursors for lch5 that emerge during early stage 10 (Ghysen and O’Kane, 1989). Since BrdU-labelling studies have shown that the precursors that give rise to lch5, divide in a graded fashion, always proceeding from anterior to posterior in each segment (Bodmer et al., 1989), we expected that clonally related ch organs would be adjacent to one another. However, the number of adjacent scolopidia that were co-labelled was far lower than the number of non-adjacent clones (R. B. and R. B., unpublished). Moreover, the clones observed were composed of random combinations of scolopidia (identified by their position along the anterior-posterior axis). This makes it unlikely that multiple scolopidia in lch5 are generated by a fixed lineage pattern but rather argues in favor of the existence of multiple precursors (possibly five) for the lch5 cluster. We speculate that these precursors emerge independently, in close vicinity to one another, in the posterior lateral region of each segment (Ghysen and O’Kane, 1989). One cannot rule out, however, that clonally related precursor cells rearrange themselves randomly after they have been generated.

Three types of clones were observed that included md neurons: a subset of md neurons that were almost always co-labelled with closely juxtaposed es organs (md/es), another subset of md neurons that co-labelled with ch organs (md/ch) and a third group of md neurons that did not co-label with other sensory organs in a consistent and reproducible pattern (solo mds). We conclude from this clonal analysis that es- and ch-associated md neurons are related to these sensory organs by lineage and that they share a common SOP (lineages are summarized in Fig. 5). Clones that labeled all cells of the es organs, v′es2, v′esB, lesA and lesB, always included the associated md-da neuron, whereas the chordotonal organs vchA and vchB co-labelled with one of the v′td2 md neurons. We inferred the patterns of cell division from smaller clones where only a fraction of a sensory organ was labelled (see Fig. 4; Table 1): SO-related md neurons derive from secondary SOPs that also give rise to the SO neuron(s).

The vpda md neuron often co-labelled with vchA and vmd5 neurons often co-labelled with one of the ves organs. These md neurons may also have lineage relations to sensory organs, possibly generated by an early ectodermal division (see Fig. 5E). As we argued for co-labelling of multiple scolopidia of lch5, it is likely that the high frequency of co-labelled vpda/vchA or vmd5/ves simply reflects the close proximity of independently emerging SOPs, and is not due to a necessary lineage relation. This possibility is supported by the finding that in mutations of the rhomboid gene, vchA is usually absent without the concomitant deletion of vpda (Bier et al., 1990). Therefore, we classify vpda and vmd5 as solo-md neurons. The dorsal bd neuron and its sibling glial cell were never consistently co-labelled with SO cells and was therefore also classified as a solo md neuron.

The putative es-related md neurons also happen to be among the md neurons that are missing in AS-C mutants (Dambly-Chaudiere and Ghysen, 1987). This suggests that genes of AS-C are required for the formation of precursors common to es organs and a subset of md neurons. Similarly, the atonal gene seems to be required for the precursors common to vchA, vchB, the v′td2-md neurons and vpda (Jarman et al., 1993).

In mutants of the numb gene, the second order precursor of es organs which gives rise to the neuron and thecogen cell is usually transformed into its sibling, the tormogen/trichogen second order precursor (Uemura et al., 1989). Our proposed model of md/mono-es lineages predicts that the fate of the md neurons related to es organs may also be altered in numb mutants since these md neurons appear to have the same second order precursor as the neuron and thecogen cell. The finding that most md neurons are absent in numb mutants supports this finding, but it could be argued that md neurons and sensory organs are affected independently. Close examination of partially transformed md/es organs (see also Uemura et al., 1989) indicate that, whenever a thecogen cell and an es neuron is formed, the associated md neuron is formed as well (Fig. 6B). This strongly supports that es-related md neurons derive from the same second order precursor as neuron and thecogen cell.

numb not only affects the lineage of es and ch organs in the PNS but also that of solo-md neurons, since most of them are absent (Uemura et al., 1989). Although the fate of most solo-md neurons in numb mutants could not be determined due to lack of specific markers, the dorsal bd neuron (dbd), is a notable exception. In numb mutants, dbd is transformed into its associated glial cell, consistent with them being siblings and requiring numb for distinguishing between a neuronal and glial cell fate.

In poxn mutants, the poly-es organs are transformed into mono-es organs (Dambly-Chaudiere et al., 1992). At least for v′es2, the formation of the associated md neuron is not affected in these mutants (C. Dambly-Chaudiere and A. Ghysen, unpublished data). This means that, although the identity of the whole v′es2 has changed in poxn mutants, the only lineage defect concerns the immediate precursor of the es neurons.

Md neurons have previously been classified into three broad categories: da neurons characterized by large dendritic arrays, td neurons whose dendrites extend along trachea and bd neurons which have bipolar dendrites (Bodmer and Jan, 1987). Our clonal analysis of the PNS suggests that md neurons can be further classified into different subtypes defined by lineage. Two markers, Cut and E7-3-49 label non-overlapping subsets of md neurons. In addition to es organs, cut is expressed in the es-related da neurons and some of the potential solo da neurons (e.g. in vmd5). E7-3-49 expresses the lacZ reporter in a set of da neurons that is essentially complementary to the cut-expressing da neurons (Fig. 7C,D). In cut mutant embryos, the E7-3-49 driven lacZ expression in md neurons is expanded to encompass virtually all da-md neurons including all es-related da neurons (see Fig. 7B,E). In contrast, when the Cut protein is overexpressed via a heat-shock promoter (Blochlinger et al., 1991) during the time of PNS neurogenesis in embryos that are wild-type for cut, E7-3-49 driven lacZ expression is greatly reduced or absent in those md neurons that normally express lacZ (i.e. vpda and the dorsal cluster da neurons)(R. B. and R. B., unpublished observations). Therefore, cut seems to be a necessary and sufficient component for the specification of the correct identity of es-related da neurons, similar to its role in es organ development (Bodmer et al., 1987; Blochlinger et al., 1991). Since the cut gene product contains a homeodomain (Blochlinger et al., 1988), cut could act directly on downstream genes as a transcriptional activator and/or as a repressor. A ver-tebrate Cut-related protein, CDP/Clox/Cux1 (Neufeld et al., 1992; Andres et al., 1992; Valarche et al., 1993), has been shown to function as a transcriptional repressor in co-transfection experiments (Andres et al., 1992; Valarche et al., 1993; Dufort and Nepveu, 1994; see also Skalnik et al., 1991). A possible repressor function by Drosophila Cut is supported by our finding that E7-3-49 driven lacZ expression is apparently suppressed in md neurons that normally or ectopically express Cut. Thus, cut function seems not only required for the specification of es organs but also for a subset of md neurons. Other selector-type genes, which may function in parallel or downstream of cut, are likely to be required to further specify the fate of different cell types, including es-related md neurons and es neurons.

Morphologically, md-da neurons look quite similar (Bodmer et al., 1987). However, the fact that md-da neurons derive from distinct lineages, raises the possibility that these cells form functionally distinct subpopulations, and the axonal projections of different types of md neurons in the CNS may reflect these differences. Merritt et al. (1993) have previously shown that the central projections of es neurons are distinct from those of ch neurons and that cut is required for the correct projections of es organs. Md neurons also have distinct central projections (Merritt and Whitington, 1995). (1) Most md-da neurons form one class and project into a discrete longitudinal CNS fascicle. They are dependent on AS-C and virtually all express cut. (2) In contrast, the atonal-dependent v′td2 and vpda md neurons have a projection pattern that is different from the majority of the md-da neurons. (3) The dbd-md neuron and one of the dorsal md-da neurons do not depend on a known proneural gene and their projections seem to be distinct from all others. This suggests that the central projections of md neurons reflects their requirement for proneural genes (Merritt and Whitington, 1995).

How do the lineage relations of md neurons correlate with a particular projection pattern? The es-related da neurons and most solo-da neurons seem to have an undistinguishable projection pattern, but the ch-related v′td2 neurons and the dbd-md neuron have distinct projections. Although lineage relationships may not be an absolutely reliable indicator of md neuronal identity, md neurons of similar lineages also have similar projections in the CNS (e.g. all es-related md neurons project into the same fascicle). Thus, it seems that proneural genes, selector genes (cut) and lineage relationships, as well as other unknown factors, have influential roles in determining the identity of md neurons.

We are much indebted to Gary Struhl for generously providing us with the flipase and FRT-lacZ transformant flies. We thank Greg Gibson, Alain Ghysen and David Merrit for critical reading of the manuscript. We also thank Corey Goodman for the 22C10 antibody, Andrew Tomlinson for the RK2 antibody, Chris Doe for the anti-prospero antibody and Seymour Benzer for the 21A6 antibody. We also thank the Dr Yuh-Nung Jan and the Bloomington Stock Center for sending fly stocks. This work was supported by a grant from NIH to R. Bodmer.