The Drosophila miti-mere gene, a member of the POU family, is required for the specification of the RP2/sibling lineage during neurogenesis

The development of the central nervous system (CNS) in the Drosophila embryo begins with the specification of the neural precursor cells in the ventral neurogenic region. Approximately 25 cells per hemisegment delaminate from the ectoderm to form a subepidermal population of cells known as the neuro-blast (NB) stem cells (Jimenez and Camposortega, 1990; Camposortega and Jan, 1991; reviewed by Cabrera, 1992; Doe, 1992). Once formed these stem cells undergo a series of asymmetric divisions producing neuroblasts and ganglion mother cells (GMCs). The GMCs are committed to a differentiation pathway and typically divide once to yield two cells which then differentiate into one or two neurons (Thomas et al., 1984; Doe et al., 1988). When the cytodifferentiation of all the neurons is complete, the embryonic CNS is believed to consist of about 250 distinct and highly specialized neurons per hemisegment.

The initial choice between the neural and epidermal fate depends upon the combined action of the proneural (e.g., achaete-scute and daughterless) and the neurogenic genes (e.g., Notch and Delta) in the ventral ectoderm (see Cabrera, 1992). The proneural genes appear to establish the neurogenic potential of the ventral ectoderm cells, while the neurogenic genes permit only a subset of these cells to differentiate into the CNS stem cells, the neuroblasts (NB). Depending upon their position in the CNS, each neuroblast in a hemisegment is thought to assume a specific identity under the control of a group of ‘neuroblast identity’ genes (see Cabrera, 1992; Doe, 1992; Chu-LaGraf and Doe, 1993). Once the identities of the individual NBs are established, these cells function much like the stem cells in other systems; they undergo a series of asymmetric divisions producing self-renewing NBs and daughter GMCs with defined fates. The specification of the GMC fate appears to be governed by the identity of the parental NB stem cell and by the order of birth of the GMC from the NB (i.e., first division, second division, etc.; Huff et al., 1989; Doe and Goodman, 1985). Like the NBs, the GMCs also undergo asymmetric division; however, this division is not self-renewing and instead the GMC produces two different daughter cells that terminally differentiate into neurons. Although studies on the grasshopper suggest that the divisions leading from NB to GMC to neuron are invariable and cell autonomous, little is understood about the underlying mechanisms or the genes involved in this process. Thus, little is known about the mechanisms that govern the identity of the two daughter cells that are produced from the asymmetric division of the neuroblast — one the self-renewing NB stem cell and the other the GMC that is committed to a differentiation pathway. Similarly, little is known about the mechanisms governing the identity of the two daughter cells that are produced from the asymmetric division of the GMC.

To begin dissecting the mechanisms that govern the elaboration of stem cell lineages like those found in the Drosophila CNS, it is important to identify genes that are involved in these specification processes. One group of genes is the POU genes (reviewed recently by Scholer, 1991; Rosenfeld, 1991; Ruvkun and Finney, 1991; Wegner et al., 1993) which are thought to function as transcriptional regulators and contain two highly conserved domains, a POU-specific domain and a POU homeodomain. Perhaps the best documented case for the role of these genes in the specification of a cell lineage is the control of the elaboration of neuronal lineages by the Unc-86 gene in Caenorhabditis elegans and loss-of-function mutations result in the reiteration of the parental NB (Finney and Ruvkun, 1990). In mammals POU proteins have been implicated in maintaining pluripotential cell identity, and the loss of the pluripotential state has been correlated with down-regulation of POU genes. Oct4, for example, is expressed in totipotent and pluripotent stem cells of the early mouse embryo, and is turned off when these cells begin to differentiate (Rosner et al., 1990). SCIP is expressed in the progenitors of oligodendrocytes, and is down regulated when these cells are induced to differentiate (Collarini et al., 1992).

In a differential screen (see Materials and Methods), we isolated a Drosophila POU gene, miti-mere (miti). This gene, as well as a second POU gene, pdml, which is located about 50 kb distal to miti, have been cloned independently by several other laboratories (pdm2 and pdm1, Billin et al., 1991; Lloyd and Sakonju, 1991; dPOU28 and dPOU19, Dick et al., 1991; Affolter et al., 1993). Both POU genes are expressed in a complex pattern during embryonic development suggesting that they may function in several distinct developmental programs. Initially, at blastoderm formation, the expression pattern of the two POU genes resembles that of the gap genes; during gastrulation the pattern is similar to that observed for many pair-rule and segment polarity genes, while later in embryonic development the two genes are expressed predominantly in the CNS and PNS. In the CNS, miti is transcribed in a subset of NBs and GMCs, but does not appear to be active in the postmitotic neurons. Recent work by Yang et al. (1993) has suggested that miti may function in neurogenesis in the specification of cell lineages. They found that ectopic expression of a full-length miti cDNA (pdm2/dPOU28) results in the duplication of the RP2 motoneuron. However, this is a gain-of-function phenotype and it is not entirely clear how it relates to the normal functions of miti in the CNS (see for example Parkhurst and Ish-Horowicz, 1991). Interpretation of this gain-of-function phenotype is further complicated by the fact that there are no known loss-of-function mutations in the miti POU gene.

Traditional genetic analysis of miti is potentially complicated by the presence of a closely linked second POU gene, pdm1. These two POU genes not only share a high degree of sequence similarity but also quite similar expression patterns and may perform redundant functions. For this reason, we decided to take a different approach for studying the role of miti during development. We generated a dominant negative ‘synthetic antimorphic mutation’ by expressing a truncated version of the wild-type miti gene product from either a con-stitutive hsp83 promoter or an inducible hsp70 promoter. Using these synthetic antimorph constructs, we demonstrate that miti is required early in embryogenesis in segmentation and later in the neurogenesis. In the CNS, it functions to establish the identity of the first ganglion mother cell, GMC-1, produced from the asymmetric division of NB4-2. This GMC gives rise to the RP2 motoneuron and its sibling cell. We have confirmed the proposed function of miti in the GMC-l→RP2/sib lineage by the analysis of deletions and a duplication that include the gene. In addition, we show that miti function in segmentation is distinct and separable from its role in the elaboration of the RP2/sib lineage. Finally, we present evidence that the down-regulation of miti is essential for the asymmetric division of GMC-l into RP2 and its sibling cell.

Deficiencies prd1.7 (cytology: 33B2-3; 34A1-2) and Proxless, Prl (cytology: 32F1-3; 33F1-2), were obtained from the Indiana stock center (see Lindsley and Zimm, 1992). The distal breakpoints of prd1.7 and Prl were previously mapped to DNA (Frei et al., 1988), while the proximal breakpoint of Prl was mapped by us and Cockerill et al. (1993). The second chromosome duplication, Dp (2;2)GYL (cytology: 33B1-2; 35C1-3 inserted into 50A1-4; see Lindsley and Zimm, 1992), was obtained from John Roote, Cambridge University, UK. The Df (2L) Prl; Dp (2;2) GYL chromosome was constructed by recombin-ing the Prl chromosome with the GYL chromosome and balancing it with the CyO, β–gal balancer chromosome. The lines used in this study were S83-2/CyO; β–gal ; S83-9/TM3, β–gal and S83-11/TM3, β–gal (the transgene in these lines is under the control of hsp83 promoter); and S70-1, S70-8 (the transgene is under the control of hsp70 promoter). Flies were raised in standard corn meal fly food for maintenance and genetic work.

Genomic library was previously constructed by Schedl (unpublished), 0-4 hour embryonic cDNA library was a gift from Dr Markus Noll, the hsp 83 and hsp 70 promoter CaSpeR vectors were from Jamila Horabin. Northern and Southern analysis, library screening and other standard molecular biology procedures were done according to the protocols described in Sambrook et al. (1989).

The miti-mere gene was originally isolated in a differential screen. For this screen, a lambda phage cDNA library was prepared from pole cell RNA by PCR amplification (using random primers) of cDNAs. This library was screened using the PCR amplified cDNAs, cDNAs prepared from tissue culture cells and labeled genomic DNA. A small number of phage clones, which hybridized with the pole cell cDNAs but not the other two probes, were selected for further study. Since the inserts were small (∼100-200 bp) we sequenced several cDNAs. One of these cDNAs showed significant sequence similarity to the Oct 1 and Oct 2 POU domains (Rosenfeld, 1991). We subsequently isolated and sequenced a near full-length cDNA from an 0-4 hour embryonic cDNA library obtained from Dr Markus Noll. This analysis revealed that the sequence of our cDNA is similar to that of the pdm2 gene (Lloyd and Sakonju, 1991; Billin et al., 1991; Dick et al., 1991; Affolter et al., 1993). In this paper, we have re-named pdm2 as miti-mere based on the dose-sensitivity of the ganglion mother cell, GMC-1 for miti (miti is Sanskrit for ‘measure’ and mere is French for ‘mother’.)

A 1.2 kb PstI fragment from our cDNA (extending from the PstI site at bp 975 in the sequence of Lloyd and Sakonju, 199l, through the end of cDNA into the vector) was inserted downstream of the hsp 83 or hsp70 promoters in the hsp83 or hsp70-CaSpeR transformation vector of J. Horabin. In the cDNA there is an internal ATG at bp 1024 (aa 278). If this ATG is used as a start codon, the resulting protein would lack 277 aa from the N terminus, and would be 221 aa in length (see Lloyd and Sakonju, 1991). Only 8 amino acids upstream of the POU-specific domain will be included in the truncated miti protein. For P-element-mediated transformation, we used the white1 allele.

The embryos were staged according to Wieschaus and Nusslein-Volhard, (1986) at room temperature (∼22°C). We also used the development of the other GMCs (in particular the GMCs for CQ neurons and aCC/pCC neurons) in relation to the development of the GMC-1→RP2/sib cells. This second criterion cannot be applied very strictly since the development of these cells appears slightly asynchronous in different hemisegments.

The whole-mount in situ hybridizations were performed according to Tautz and Pfeifle (1989). Immunostaining with antibodies were performed as described by Patel et al. (1989b). A polyclonal antibody directed against eve was a gift from Manfred Frasch and a monoclonal for eve was a gift from Nipam Patel; antibody against 22C10 (Fujita et al., 1982) was from Yash Hiromi. Antibodies against pdm1 and miti (pdm2) were a gift from Steve Poole. For antibody staining, embryos were fixed 20 minutes for eve, pdm1 and miti/pdm2 and 40 minutes for 22C10. eve antibody was used at 1:2000 dilution and 22C10 was at 1:4 dilution, pdm1 and miti was at 1: 25 dilution. For double staining with eve and miti antibody or eve and pdm1 antibody, a monoclonal against eve was used. Following the double staining, the detection was done by confocal microscopy. The images presented in Fig. 4 were the summed images of a series of sections taken across the CNS. Embryos were mounted in 65% glycerol before photography. The CNS from older embryos were dissected before mounting. Finally, embryonic cuticles were prepared essentially as described by Ashburner (1989).

For the heat-shock experiments, we used the S70-1 line, which carries a Δmiti cDNA fragment fused to a hsp70 promoter. Embryo collec-tions were at 22°C for 15 minutes. Early cleavage stage embryos were hand picked and aged for different durations as shown in Fig. 8. The heat shock was performed for 15 minutes at 37°C by immersing the embryos in prewarmed halocarbon oil. Following heat shock, the embryos were aged at 22°C for different lengths of time in order to monitor the effects of ectopic Δmiti expression on the development of the cuticle or the NB4-2→GMC-1→RP2/sib lineage. We then processed the embryos to examine either the cuticle or the RP2/sib lineage. For cuticle analysis, embryos were aged for 36 hours before proceeding with the cuticle preparation.

Domain swapping experiments have demonstrated that many transcriptional regulators have a bipartite structure with distinct DNA-binding and ‘activator’ domains that are often capable of functioning independently (Ma and Ptashne, 1987a,b, Ptashne, 1988; Hahn, 1993). This bipartite structure is also found in POU proteins such as miti (Rosenfeld, 1991; Ruvkun and Finney, 199l; Wegner et al., 1993). As shown in Fig. 1B, the DNA-binding domains of the miti protein, the POU-specific domain and the POU homeodomain, are located in the C-terminal half of the protein, while the N-terminal half has features commonly found in the ‘activator’ domains of Oct-1 and Oct-2, including an acidic blob of about 50 amino acids as well as a glutamine-rich region (Sturm et al., 1988; Clerc et al., 1988). Functional studies on POU proteins have shown that the C-terminal POU-specific and POU home-odomains are responsible for DNA binding (Sturm and Herr, 1988; Ingraham et al., 1990; Verrijzer et al., 1992), while the N-terminal domain is responsible for cell-type-specific trans-activation (Ingraham et al., 1990; Monuki et al., 1993). We reasoned that expression of a truncated miti protein, consisting of the C-terminal POU-specific and POU homeodomains but lacking the N-terminal ‘activator’ domain, might behave as a dominant negative, and interfere with the activity of the wild-type miti gene (cf., Han and Sternberg, 199l; Coffman et al., 1993). The truncated protein should, in principle, be able to compete with the full-length miti protein for binding to the cis-acting elements associated with downstream target genes. However, since it lacks the N-terminal ‘activator’ domain, it would be unable to control downstream gene activity appropriately. Hence, if sufficient quantities of the truncated protein could be expressed in cells that normally require miti activity, it should mimic the phenotypic effects of loss-of-function mutations in the miti gene by interfering with the functioning of the wild-type protein. This approach offers an additional advantage in the event that miti and the nearby POU gene, pdml (see Fig.1A), perform redundant functions. While this redundancy would make a traditional genetic approach difficult, the synthetic antimorph would be expected to interfere with the activity of both POU genes.

To generate this dominant negative gene, we inserted sequences from the 3′ half of a miti cDNA downstream of the ‘constitutive’ hsp83 promoter (see Fig. 1C). In this construct, there is an in frame AUG codon located just beyond the restriction site in the cDNA that was used for cloning. If this AUG codon is used for translation initiation in vivo, the protein expressed by the hsp83Δmiti transgene would be expected to begin at amino acid residue 278 of the miti protein which is just 8 amino acids upstream of the POU-specific domain. It would extend from this AUG through the POU-specific domain and the POU homeodomain to the normal C terminus of the miti protein. From studies on other POU proteins, this Δmiti protein would be expected to have DNA-binding activity, but be unable to function in transcriptional regulation (Sturm and Herr, 1988; Ingraham et al., 1990; Monuki et al., 1993). Transgenic lines carrying the hsp83Δmiti transgene were then isolated by P-element-mediated transformation.

We obtained 8 independent lines carrying the hsp83Δmiti transgene. The pattern of expression of Δmiti RNA from one of these lines, S83-2, and of RNA from the endogenous miti gene during early embryogenesis and neurogenesis is shown in Fig. 2. Although there is some maternally deposited Δmiti RNA, it appears to turnover rather rapidly, and is largely gone by blastoderm formation (data not shown). The endogenous miti gene is initially expressed at the cellular blastoderm stage (early stage 5) in a pattern similar to that of the gap genes; there is a strong band spanning the Al/A2 abdominal region, and somewhat weaker bands in the A5/A6 abdominal region, and in the head region (see Fig. 2A; Lloyd and Sakonju, 1991; Cockerill et al., 1993). In embryos carrying the Δmiti transgene, we observe the nearly ubiquitous expression of miti RNA from the transgene superimposed upon this endogenous pattern (Fig. 2B). As the embryo gastrulates, there is a dramatic change in the expression pattern of the endogenous miti gene and, like many pair-rule and segment polarity genes, 14 stripes are observed. The ubiquitous expression of the Δmiti transgene persists somewhat weakly through this period (data not shown). Later in embryogenesis, the stripe pattern of the endogenous miti gene disappears and is replaced by expression in the ventral neurogenic region. Initially, most of the cells in the medial neurogenic region express miti; however, as the CNS develops, expression becomes restricted to only a subset of the neuroblasts and GMCs. miti expression in neuroectoderm of a wild-type embryo at about stage 8 is shown in Fig. 2C. A more ubiquitous expression pattern is typically observed for the Δmiti transgene (see Fig. 2D).

Analysis of several of the hsp83 transgenic lines indicated that the Δmiti transgene has deleterious effects. These effects were first evident when we attempted to increase the copy number of the hsp83Δmiti transgene. As indicated for the S83-2 transgenic line in Table 1, two copies of the transgene were found to cause a substantial reduction in viability. A similar result was obtained with other independently isolated lines, as well as for trans-heterozygotes between two different lines (thus ruling out the possibility that lethality is simply due to inactivation of an essential gene by the inserted transgene).

Two hypotheses could be advanced to explain the lethality observed when we attempted to increase the dose of the hsp83Δmiti transgene. In the first, the truncated miti protein has acquired a ‘new’, apparently toxic activity and hence is behaving like a neomorphic mutation. In the second, the truncated miti protein is toxic because it interferes with the normal functioning of the wild-type gene product and hence behaves like a dominant negative or antimorphic mutation. It is possible to distinguish between these two hypotheses by examining the effects of gene dose (Muller, 1932). If the truncated miti protein expressed by the hsp83Δmiti transgene has acquired a new activity, then the phenotypic effects of this neomorphic mutation will not be altered by changing the dose of the wild-type gene. In contrast, if the truncated miti protein behaves as an antimorph, then reducing the dose of the wild-type gene should enhance the phenotypic effects of the transgene.

To examine the effects of gene dose, the hsp83Δmiti transgene was introduced into animals carrying the deficiency prd 1.7, which deletes the wild-type miti gene (see Fig. 1). For this experiment, we used the S83-2 line, which has an hsp83Δmiti transgene inserted on the second chromosome. As can be seen in Table 1, the hsp83Δmiti transgene causes a significant reduction in the viability of animals heterozygous for the prd 1. 7 deficiency. When the prd 1. 7 deficiency is crossed to animals carrying a 2nd chromosome lacking the transgene (+/Cyo or white1 in Table 1), close to the expected percentage of viable transheterozygotes (prd 1.7/+) is obtained. In contrast, when the prd 1.7 deficiency is crossed to S83-2 animals, we recover only between 12 and 15% transheterozygotes (prd 1.7/hsp83Δmiti) instead of the expected 33%. Thus, as is characteristic of antimorphic mutations, the lethal effects of hsp83Δmiti appear to depend on the relative ratio of the transgene to the wild-type gene.

While these results indicate that the hsp83Δmiti transgene behaves like an antimorphic mutation in miti, there is a potential complication with this interpretation. As can be seen in Fig. lA, the prd 1.7 deficiency removes not only the proximal miti POU gene but also the more distal POU gene pdml. Consequently, the lethality observed for the prd/Δmiti transheterozygote could be due to dominant negative interactions not only with the miti protein but also with the product of the pdm1 gene (see Introduction). Indeed, the two POU proteins share extensive amino acid sequence homology in the POU domain and exhibit similar (though not precisely identical) expression patterns during early embryogenesis and neurogenesis. At this moment, it is not possible to resolve this issue entirely with the existing deficiencies. The only other known deficiency that affects either POU gene is Prl. The Prl chromosome has an intact miti gene, while the pdm1 gene is disrupted by the Prl deficiency breakpoint (Cockerill et al., 1993; our unpublished results). As can be seen from comparing the frequency of transheterozygotes when Prl is crossed to a wild-type second chromosome or to the S83-2 second chromosome, the hsp83Δmiti transgene has no detectable effect on the viability of Prl animals (Table 1). Thus, the removal of the pdm1 POU gene and all of the other genes located centromere-distal to pdm1 that are deleted in both prd1.7 and Prl is not in itself sufficient to cause detectable lethality (see Fig. 1A). This would map the region of lethal interaction between the centromere proximal break-points of Prl and prd1.7, making miti a good candidate for the anti-morphic interactions with the Δmiti transgene. While these findings would suggest that the hsp83Δmiti transgene interferes with the miti gene, it should be pointed out that we can not exclude the possibility that the transgene must simultaneously interfere with both miti and pdm1 in order to generate the lethal effects.

If the Δmiti transgene behaves as an antimorphic mutation, then the developmental perturbations observed in transgenic animals should closely correlate with the expression pattern of the endogenous miti gene. To explore this possibility, we examined the cuticles of embryos carrying the hsp83Δmiti transgene for evidence of patterning defects. As described above, miti is first transcribed in the early embryo in a gap-like pattern with strongest expression in Al/A2 and weaker expression in A5/A6 and in the head region. Consistent with a dominant negative interaction between the Δmiti transgene and the endogenous gene, the most common cuticle defect in trans-genic animals is the deletion of the abdominal segment A2, the segment in which miti is expressed at the highest level in blastoderm embryos (see Table 2). This can be seen by comparing the cuticle of a wild-type embryo with that of an embryo carrying two copies of Δmiti (see Fig. 3A,B). Infrequently, we also observe embryos showing defects in A6 (see Table 2; Fig. 3C) and in the head region (see Fig. 3C).

To test whether the cuticular defects induced by the Δmiti transgene are due to dominant negative interactions between the transgene and the endogenous miti gene, we examined the effects of gene dose. The results of this analysis, presented in Table 2, are consistent with the antimorphic character of the hsp83Δmiti transgene. In a wild-type background (miti+/miti+) a single copy of the hsp83Δmiti transgene results in a very low frequency (∼1%) of embryos showing cuticle defects (see Table 2). This frequency increases significantly when there are two copies of the transgene. Similarly, reducing the dose of the wild-type miti gene (miti+/prd l.7) in the presence of a single copy of Δmiti also results in a substantial increase in the frequency of cuticle defects. Moreover, the cuticle phenotype, like the lethal interaction, requires the deletion of the miti gene; in Prl /Δmiti transheterozygotes the frequency of cuticle defects is equivalent to that observed when Δmiti is crossed to wild type.

While miti is initially transcribed in many cells in the medial neurogenic region, expression becomes restricted to only a subset of the neuroblasts, GMCs and neurons as the CNS develops. It was of interest to determine whether the hsp83Δmiti transgene affects the elaboration of neuronal lineages which normally express the endogenous miti gene. One of the lineages in which miti is expressed, the lineage giving rise to RP2/sib (see also Yang et al. 1993), has been especially well characterized (Thomas et al., 1984; Patel et al., 1989a; Doe, 1992; Doe et al., 1991; Chu-LaGraf and Doe, 1993). The stem cell of this lineage is NB4-2. Its first asymmetric division gives rise to GMC-1, which in turn divides to produce the RP2 and sib cells (Patel et al., 1989a; Doe, 1992; Doe et al., 1991). While the fate of the sib cell is unknown, the RP2 cell differentiates into a motor neuron innervating muscle no. 2 (see Vactor et al., 1993). In order to correlate the possible effects of the Δmiti transgene on the elaboration of this lineage, we first examined in detail the pattern of expression of the miti protein. The development of the RP2/sib lineage can be followed with several different markers. One such marker is even-skipped (eve), which is useful for tracing the lineage from shortly after the formation of GMC-1 through its division into RP2 and sib. Fig. 4 shows a confocal analysis of the CNS, double stained with eve and miti antibodies, at two different time points, approximately 7 and 7.5 hours of development (at 22°C). GMC-1 can be identified in each hemisegment by its position relative to other eve-positive GMCs, i.e., those for the aCC/pCC and CQ neurons. As can be seen in Fig. 4A, the aCC/pCC and CQ GMCs (large arrowhead) stain heavily with eve antibody at ∼7 hours of development while only very low levels of eve protein are observed in GMC-1 (see arrow). Fig. 4B shows the same field as Fig. 4A stained with miti antibody, while Fig. 4C shows the two staining patterns superimposed. In contrast to the eve staining pattern, very high levels of miti protein are detected in GMC-1 (while no miti protein appears to be present in the aCC/pCC or CQ cells). A large difference in the level of eve and miti protein in GMC-1 is typically observed at this stage of development and may indicate that miti is turned on earlier than eve. miti does not, however, appear to be expressed in the progenitor of GMC-1, NB4-2 (data not shown). These results suggest that miti is initially activated in the RP2/sib lineage at or soon after GMC-1 is formed.

Although the level of miti protein in early GMC-1 cells is higher than eve, in older GMC-1 cells the level of miti protein appears to drop while the level of eve protein increases substantially. Fig. 4D-F show eve and miti protein in the CNS of a ∼8 hour embryo, just around the time when the GMC-1 cells begin to divide. In the hemisegment on the left, the GMC-1 has already divided to form RP2 (long arrow) and sib (short arrow). Typically, just after the GMC-1 division, the RP2 cell has high levels of eve protein (and appears to be the larger cell) while the sib cell has low levels of eve (and appears to be the smaller cell) (see Fig. 4D). A similar pattern is observed for miti protein; there are high levels in RP2 and lower levels in the sib (see Fig. 4E,F). eve protein then persists in the RP2 cell during its subsequent differentiation into a neuron, while it gradually disappears from the sib. In contrast, miti protein dis-appears from both cells soon after they are formed (data not shown). Also shown in panels D-F is a GMC-1 which appears to be in the process of dividing (see the hemisegment on the right). eve protein in this cell is not uniformly distributed; instead, there is a higher concentration of protein on one side (bottom arrow) and a lower concentration on the other side (top arrow). Miti protein also appears to be unevenly distributed in this cell and it follows a pattern similar to that observed for eve.

If the Δmiti protein interferes with the normal functioning of the wild-type miti protein in the RP2/sib lineage, the transgene should affect the elaboration of this lineage. Consistent with this expectation, two types of phenotypic defects in the RP2 lineage are observed. First, in about 72% of the S83-2 homozygotes (18% of the embryos from the cross), we observe hemisegments that appear to be missing the GMC→RP2/sib lineage (see Table 2). In these affected embryos, four hemiseg-ments on average (see Table 2) lacked the eve-positive cells from this lineage (either GMC or RP2/sib, depending on the stage). Shown in panel A of Fig. 5 is a wild-type ∼7 hr embryo stained with eve antibody. The GMC-1 cells (illustrated by thick arrow) in each hemisegment of this embryo are located immediately anterior to the GMCs for aCC/pCC (large arrowhead) and CQ (small arrowhead) neurons, but are laterally displaced from the midline. In the S83-2 homozygote shown in Fig. 5C, the laterally displaced eve-positive GMC-1s are absent (see arrow) while the aCC/pCC and the CQ GMCs (see arrowheads) appear normal. Shortly before the GMC-1 cell divides, it migrates inward toward the ventral midline. Fig. 5C shows the CNS of a ∼8 hr wild-type embryo in which the GMC-1 cells have migrated towards the ventral midline, and in some cases have already divided. As can be seen in the 8 hour S83-2 homozygote in Fig. 5D, eve-positive GMC-1 cells and/or their daughters the RP2 and sib cells are missing (expected position of the eve-positive GMC-1, or RP2/sib is indicated by arrow). Consistent with the apparent absence of GMC-1 cells in 7-8 hour embryos, eve-positive cells at the position expected for the differentiated RP2 neuron are not observed in hemisegments from 13 hour S83-2 homozygotes (compare the wild type in Fig. 5F with the S83-2 homozygote in Fig. 5G).

In addition to the apparent deletion of the GMC-1→RP2/sib lineage, a second phenotype is observed (in 1-3 hemisegments) in about 5 % of the 13 hour S83-2 homozygous embryos. Although there is an eve-positive cell at the longitudinal position expected for RP2 (upward arrow), this cell is displaced laterally from the midline compared to a normal RP2 cell (see for example, the two RP2 cells in the hemisegment above). In fact, this lateral position is close to that of a GMC-1 cell, suggesting that this cell might be a GMC-1 cell that has failed to migrate and divide into RP2 and sib.

To confirm that the phenotypic effects of the Δmiti transgene on the GMC-1→RP2/sib lineage are due to dominant negative interactions between the transgene and the endogenous miti gene, we next examined the effect of the wild-type miti gene dose. Table 2 shows the percentage of embryos showing defects in the RP2 lineage as judged by staining with eve antibody. Whereas a single copy of the Δmiti transgene had little effect on this lineage in the presence of two wild-type copies of the miti gene, a high percentage of the embryos showed defects when there was only a single copy of the wild-type gene (see S83-2 × prd 1.7 in Table 2 and Fig. 5E,I). The percentage of embryos showing defects was nearly as high as that observed when there were two copies of both the transgene and the wild-type gene. In contrast, the Δmiti transgene had no apparent effect on the RP2 lineage in trans to the Prl deficiency, which removes the nearby POU gene pdm1, but not the miti POU gene.

To provide further evidence for the apparent deletion of the GMC-1→RP2/sib lineage, we used another marker, the 22C10 antibody (Fujita et al., 1982), to examine the RP2 neurons in the CNS of 13 hour or older S83-2 homozygotes. The 22C10 antibody recognizes a membrane glycoprotein and stains mature RP2 neurons as well as a subset of other neurons and their axonal growth cones in wild-type embryos (Fujita et al., 1982). As shown in the wild-type CNS in Fig. 6A, this antibody can be used to identify the RP1-RP4 cell bodies in the anterior commissure and their axonal trajectories out of the RP-cluster into the intersegmental or ipsilateral nerve (ISN) (Thomas et al., 1984; see Camposortega and Hartenstein, 1985). The axonal trajectories from RP1, RP3 and RP4 (Chiba et al., 1993) join with the aCC neuron (which is located in the posterior commissure) and extend posteriorly into the ISN (thin arrow in Fig. 6A). Since the axonal trajectory from the RP2 cell body (long arrow in Fig. 6A) extends anteriorly (arrowhead) not posteriorly, each ISN receives its RP2 axon from the adjacent (posterior) neuromere. Located immediately posterior to the ISN is the segmental nerve (SN: thick arrow in Fig. 6A). While this nerve appears to consist of more than one axon, the main cell body of the SN lies lateral to the RP cluster. It can be easily distinguished from RP2 by its more lateral position, its axonal trajectory and its fasciculation with the ventral sensory neurons of the peripheral nervous system (see Ghysen et al., 1986) (see Fig. 6A).

Analysis of the 22C10 staining pattern in S83-2 homozygotes confirms that the Δmiti transgene interferes with the formation of the RP2 neuron. As shown in the S83-2 CNS in Fig. 6B, the cell bodies of the RP1-3 cluster and the aCC neuron appear normal and their fascicles (thin arrows) extend properly into the ISN; however, no RP2 cell body (long arrow) or axonal projection (arrowhead) can be detected in the neuromere. We found that the RP2 neuron, as judged by the 22C10 staining pattern, is absent in over 40 % of the hemisegments (n=168) of 13 hour S83-2 homozygotes. This frequency is somewhat higher than that predicted from the experiment with eve antibody where only 28 % of the hemisegments (from S83-2 homozygotes) were missing an eve-positive cell. In contrast, this number is close to the combined frequency of hemisegments missing eve-positive cells altogether (28 %) or containing eve-positive cells that are laterally displaced from normal RP2 position (7 %). This finding would suggest that the eve-positive cells that fail to migrate to the normal position of RP2 never differentiate into RP2 neurons. One possibility is that cells are blocked at some stage in GMC-1 development, and never progress to the point where they divide to produce RP2 and sib.

To obtain further evidence that the miti gene plays a role in the specification of the GMC-1→RP2/sib lineage, we examined the eve staining pattern in the CNS of embryos homozygous for either the prd 1.7 or the Prl deficiency. Both deficiencies remove the segmentation gene paired, and as a result embryos homozygous for either deletion fail to form segments T1,T3, A2, A4, A6 and A8 while the development of the remaining segments is almost normal. Thus, it should be possible to examine the RP2 lineage in these deficiency embryos in the segments unaffected by the lack of paired activity. The results of this analysis are presented in Fig. 7. In prd 1.7 (which is deleted for both miti and pdm1 genes) eve-positive RP2/sib GMCs are not observed in the ∼7.5 hour deficiency embryos (Fig. 7C). In contrast, in Prl (which deletes pdm1 but not miti) the RP2 lineage (in segments not requiring paired) appears normal, and eve-positive cells are observed at the location expected for the RP2/sib GMC in ∼7.5 hour embryos (Fig. 7B). This result provides further support for the conclusion that miti (or miti together with pdm1) is required for the specification of the RP2/sib lineage. It would also argue that the hsp83 Δmiti transgene mimics a hypomorphic loss-of-function mutation in miti.

Analysis of the phenotypic defects associated with hypomorphic or null mutations can be extremely valuable in elucidating the biological roles of a specific gene during development. However, distinguishing between the primary and secondary effects of a mutation can sometimes prove difficult, particularly in cases where a gene may function in several different developmental processes. Similar problems are encountered in interpreting the phenotypic defects induced by the hsp83Δmiti transgene — for example, are the defects in neurogenesis a direct consequence of interfering with miti function in the CNS, or are they an indirect consequence of the earlier dis-ruptions in segmentation? Indeed, as discussed in the previous section, the RP2/sib lineage is completely missing in those segments that are affected by the deletion of the segmentation gene paired in prd1.7 and Prl (see Fig. 7A,B). Typically, conditional mutations, such as temperature-sensitive alleles, can be used to distinguish between primary and secondary effects particularly in those cases where the developmental processes in question, such as segmentation and neurogenesis, are temporally distinct (see for example, Doe et al., 1988; Duffy et al., 199l; Chu-LaGraf and Doe, 1993). By shifting the conditional mutants from permissive to non-permissive conditions (or vice versa), gene function at different points in development can often be separated and examined independently. We reasoned that it might be possible to generate a ‘conditional’ miti allele by fusing the truncated Δmiti gene to an ‘inducible’ hsp70 promoter (see Fig. 1C). Under permissive conditions, ∼22°C, where the hsp70 promoter is off, the hsp70Δmiti gene should have no deleterious effects on transgenic animals. However, we should be able to observe the dominant negative effects of the Δmiti protein by elevating the temperature and inducing the transgene. The newly synthesized Δmiti protein would be expected to interfere with the endogenous protein, blocking its function at that particular developmental stage. (The tightness of this non-permissive window will, however, depend on at least three factors: the level of Δmiti protein produced by induction of the hsp70 transgene, the subsequent perdurance of the Δmiti protein, and the sensitivity of the development step to interference by the Δmiti protein.).

The results obtained with the transgenic lines carrying the ‘inducible’ hsp70Δmiti transgene indicate that it behaves as a ‘conditional’ dominant negative mutation. In contrast to the hsp83Δmiti transgene, even two copies of the hsp70Δmiti transgene had little or no effect on development or viability when the animals were grown at ∼22°C. However, significant lethality and developmental abnormalities could be induced by growth at elevated temperatures (29°C) or by a brief heat shock at 37°C. In one of the lines, the hsp70 transgene was inserted on the X chromosome and we used an attached X chromosome to generate a stock in which only males inherit the transgene. While this stock shows normal male viability at 22°C, at 29°C less than 15% of the adult flies were males. Similarly, in a line which is homozygous for a transgene insert on the second chromosome (S70-1) the viability was reduced to <2% at 29°C. Moreover, the phenotypic defects observed at 29°C were the same as those generated by the hsp83 transgene, though the defects were typically less frequent and not as severe. Cuticle preparations from (29°C) S70-1 homozygotes indicated that about 16% of the embryos showed segmentation defects and, in most cases, were missing either the A2 or the A2 and A6 denticle belts. In staining with eve antibody, 10% of the embryos showed defects in the RP2/sib lineage and these defects were usually restricted to only two or three hemisegments.

To determine when the Δmiti transgene interferes with specific developmental processes — the formation of the A2/A6 segment, and the elaboration of the NB4-2→GMC-1→RP2/sib lineage — we shifted S70-1 homozygous embryos from the permissive temperature 22°C to the non-permissive temperature 37°C for ∼15 minutes at different times and then returned the embryos to 22°C to allow development to proceed. As indicated in Fig. 8, we find that a brief heat pulse (15 minutes) in the first 2 hours of development (early cleavage to syncitial blastoderm) induced with high efficiency very specific segmentation defects in transgenic embryos — virtually all embryos showed pattern deletions in A2 or A2 and A6 (Fig. 9D). By contrast, heat shock did not induce these very specific segmentation defects in similarly staged parental white1 embryos. (Though segmentation defects could be found in the parental embryos, the segments affected varied from embryo to embryo and the penetrance was low.) Similarly no cuticular defects were evident when the transgenic animals were heat shocked at times later in development (cf, the embryo heat shocked at 6 hours shown in Fig. 9E). In this context it may be of interest that transcripts from the endogenous miti gene are first detected around 2.5 hours of development (at 22°C). Hence, it would appear that the truncated miti protein must already be present in order to interfere with the functioning of the endogenous miti protein in the segmentation process.

Significantly, induction of defects in the RP2/sib lineage follows a quite different time course from that of the segmentation defects. While heat shock of precellular blastoderm embryos induces A2/A6 segmentation defects at a high frequency, it does not affect the elaboration of the RP2/sib lineage. This is illustrated by the eve stained embryo shown in Fig. 9A. Outside of the A2 region, the organization of the CNS appears normal, and RP2 cells are present in all hemisegments (illustrated by arrows). In the A2 region the organization of eve-positive cells in the CNS is abnormal, probably the con-sequence of the segmentation defect (cf Fig. 9D); however, RP2 cells still appear to be present.

As demonstrated by the results presented in Fig. 8, defects in the RP2/sib lineage can only be induced by heat pulses between 6 and 8 hours of development. By far the most sensitive period is between 6 and 7 hours (mid-to late-stage 9), while the frequency drops dramatically thereafter. Moreover, the lineage is not responsive to heat shock at earlier or later times. The GMC-1→RP2/sib lineage defects can be seen first as either missing GMC-1 cells (see the ∼7 hour embryos in Fig. 9C) or GMC-1 cells in which eve expression appears to be abnormally low (see Fig. 9F). Consistent with the missing GMC-1 cells, later in development we observe hemisegments with missing RP2 and sib cells (see the ∼12 hour embryo in Fig. 9B which lacks RP2 cells in several hemisegments).

In a previous study, Yang et al. (1993) found that ectopic expression of a full-length miti (pdm2/dPOU28) protein under the control of an hsp70 promoter results in the duplication of both RP2 and its sibling cell. This duplication is only observed when the hsp70miti transgene is induced during the GMC state, and does not occur when the transgene is induced after the GMC has divided to form RP2 and its sib. To explain these findings, Yang et al. (1993) suggested that the ectopically expressed miti protein causes the GMC-1 to divide ‘symmetrically’ producing two GMC-1 progeny instead of RP2 and its sib. The duplicated GMC-1s then divide to produce two RP2s and two sibs. The duplication of the GMC-1 observed in their experiments could, in principle, occur because induction of the hsp70 miti transgene mimics a neomorphic mutation — the miti protein has acquired a new function because it is expressed in an inappropriate cell and/or at inappropriate time. Alternatively, the induction of the hsp70 miti transgene may mimic a hypermorphic mutation — too much miti protein accumulates in the GMC-1. Our experiments which indicate that miti is required for the specification of GMC-1 would seem to argue in favor of the second inter-pretation, namely that the duplication of GMC-1 occurs because there is too much miti protein. It should be possible to further distinguish between these two interpretations — neomorphic versus hypermorphic — by determining whether increasing the dose of the wild-type miti gene has similar effects on the GMC-1 lineage as the hsp70miti transgene.

For this purpose, we took advantage of a second chromosome which carries a duplication for chromosomal bands 33B1-2 to 35C1-3 at 50A1-4, Dp (2;2) GYL (see Lindsley and Zimm, 1992). Since this chromosome has two copies of both the miti gene and pdm1, we recombined the duplication chromosome with the Prl deficiency chromosome (see Fig. 1A). The resulting recombinant chromosome now carries 2 copies of miti, but only a single copy of pdm1. We then examined the eve expression pattern in the CNS of embryos homozygous for this duplication/deficiency chromosome. A number of findings are of interest. First, in the CNS of ∼7 hour embryos, we observed hemisegments (∼13 %, n=112 hemisegments) with two eve-positive cells, instead of one, in the location of GMC-1 (Fig. 10A). In most cases, one of the two cells had a lower level of eve expression (as might be expected for a presump-tive sibling cell). This finding would indicate that additional copies of the miti gene may cause the precocious division of GMC-1. Second, in the CNS of ∼8 hour embryos, some of the hemisegments had three, instead of two, eve-positive cells at the position of RP2/sib, and this pattern persists until eve staining is lost from either one or two of the cells (presumably sib). In no cases, however, do we observe four eve-positive cells as was found by Yang et al. (1993) when the hsp70miti transgene was induced. As illustrated in Fig. 10, some of these three cell clusters contain one strongly staining and two weakly A staining eve-positive cells (Fig. 10B) while other clusters contain two strongly staining and one somewhat weakly staining eve-positive cells (Fig. 10C). The stronger eve-positive cells are expected to correspond to the RP2, while the weaker should correspond to the sib. These results suggest that additional copies of the miti gene can result in the duplication of either RP2 or the sibling. This suggestion is confirmed by the presence of two eve-positive cells in the location of RP2 in ∼8% of the hemisegments (n=84) in the CNS of 13 hour embryos (Fig. 10D) and also by the presence of two 22C10-positive RP2 neurons with their trajectories extending into the ISN (Fig. 10E). It may be significant that clusters of three eve-positive cells are found (in ∼8-8.5 hour embryos) at more than twice the frequency of duplicated RP2 neurons (i.e. two eve- or 22C10-positive cells in 13 hour embryos). This may indicate that duplication of the sibling cell is preferred over the duplication of RP2.

These results suggest that increasing the dose of the wild-type miti gene alters the GMC-1 lineage, ultimately leading to the duplication of either the sib or the RP2 neuron. This would argue that the hsp70 miti transgene of Yang et al. (1993) is mimicking a hypermorphic mutation, not a neomorphic mutation. There is, however, one critical difference between our results and those of Yang et al. (1993). They observe four eve-positive cells, while we observe only three. This differ-ence would suggest that the elaboration of the GMC-1→RP2/sib lineage may be sensitive to the level of the miti protein. At very high levels of the protein, the initial division of the GMC-1 would be symmetric as suggested by Yang et al. (1993) leading first to the duplication of the GMC-1, and subsequently to the duplication of both RP2 and sib (see Fig. 11). When the miti protein is present at only twice the normal level, the initial division of the GMC-1 would be asymmetric, but abnormal in that one of the daughters apparently retains the GMC-1 identity while the other daughter becomes either RP2 or sib (see Fig. 11). The cell retaining the GMC-1 identity would then undergo a normal asymmetric division to produce RP2 and sib. The alternative possibility that the duplication occurred at either the sibling level or at the RP2 neuronal level (GMC-1→RP2, sib and then RP2→RP2 or GMC-1→RP2, sib and then sib→sib) is unlikely since only GMC-1 was found to be sensitive to miti protein levels in the experiments of Yang et al. (1993) and neither RP2 nor sib responded to high levels of miti induced from hsp70miti transgene once they were formed. Our results would also argue against the hypoth-esis that GMC-2 (of NB4-2) is transformed into GMC-1 (a possible interpretation of the data from Yang et al., 1993), since we observed the duplication of either the sib or the RP2 neuron but not both.

Ectopic expression experiments have provided a valuable tool for analyzing the function of regulatory molecules in Drosophila and other developmental systems. In most of these experiments, a cDNA encoding the regulatory molecule of interest is placed under the control of a heterologous promoter such as the heat-inducible hsp70 promoter (Blochlinger et al., 199l; Parkhurst and Ish-Horowicz, 1991; Gonzales-Reyes and Morata, 1990; Cabson and Gehring, 1988; Kuziora and McGinnis, 1988). The heterologous promoter then drives the inappropriate expression of the regulatory molecule with respect to time, cell type or level and the developmental con-sequences if any are determined. In order to observe pheno-typic effects, misexpression of the regulatory molecule must typically mimic either a hypermorphic or a neomorphic mutation. While the gain-of-function phenotypes produced by such constructs can provide useful insights into the normal function of the regulatory molecule, they can also be difficult to interpret and even misleading.

Since there are no known mutations in the miti gene or in its relative, pdm1, we decided to modify the strategy normally used in ectopic expression experiments so that we could mimic not a gain-of-function, but rather a loss-of-function mutation. For this purpose, we used either the constitutive hsp83 promoter or the inducible hsp70 promoter to express a truncated version of the miti protein that contains the C-terminal DNA-binding domain, but lacks the N-terminal ‘activator’ domain. This truncated protein should bind to the same regulatory targets as the full-length miti protein, but be unable to regulate properly these targets (Sturm and Herr, 1988; Ingraham et al., 1990; Verrijzer et al., 1992; Monuki et al., 1993). As a consequence, it should behave as a dominant negative — when expressed in cells that normally require miti function, it should interfere with the activity of the wild-type gene product, and mimic a loss-of-function mutation. This approach offers a potential advantage over traditional genetic analysis in the event that miti and its closely linked relative, pdm1, are functionally redundant. If this were the case, a dominant negative should interfere with the functioning of both miti and pdm1.

The genetic and phenotypic properties of the hsp83- and hsp70Δmiti transgenes are generally consistent with those expected for a dominant negative or antimorphic mutation. The hsp83 transgene is most useful in assessing the genetic prop-erties of the Δmiti protein. As anticipated the hsp83Δmiti construct has deleterious effects on transgenic animals. Moreover, like a classical antimorphic mutation (Muller, 1932), the effects of the hsp83 transgene are dependent on the relative dose of the transgene and the wild-type gene. While animals carrying only one copy of the transgene are viable, two copies are not tolerated. Similarly, when only one copy of the wild-type gene is present, even a single hsp83Δmiti transgene is sufficient to cause lethality. Our analysis of deficiencies that remove miti and/or pdm1, suggest that the Δmiti transgene must, at the minimum, interfere with the functioning of the wild-type miti gene. At the maximum, the dominant negative effects may require that Δmiti interfere with both the miti and pdm1 genes. Consistent with this later possibility, the POU domains of the miti and pdm1 proteins are closely related. In addition, they exhibit quite similar expression patterns (see dis-cussion below). However, resolving this question will require the isolation of mutations in each gene.

The conclusions reached from this genetic analysis of the hsp83 Δmiti transgene are further supported by a second line of evidence. As expected for a dominant negative, the pheno-typic disruptions induced by the Δmiti transgene closely correlate with the expression pattern of the endogenous miti (and pdm1) gene. At cellular blastoderm miti is expressed in two broad bands, at high levels in A1/A2 and at lower levels in A5/6. Consistent with a dominant negative interaction between the transgene and the endogenous miti gene during this early period of development, the most common segmen-tation defect is the deletion of the A2 denticle belt (or less fre-quently both A2 and A6). Moreover, the segmentation defects induced by the transgene, like the lethality, depend upon gene dose. Later in development, the miti gene is expressed in the CNS in a subset of the neuroblast lineages. We have examined the effects of the hsp83 Δmiti transgene on the elaboration of one of the miti-(and pdm1-) expressing lineages, NB4-2→GMC-1→RP2/sib; we find that the hsp83 transgene perturbs the development of this lineage in a dose-dependent manner. By contrast, the development of other neuroblast lineages (aCC/pCC, CQ, RP1, RP3 and RP4), which do not express the miti gene, does not appear to be affected by the hsp83 transgene.

Taken together, these findings indicate that the hsp83 Δmiti transgene behaves like a classical antimorph, mimicking the phenotypic effects of a loss-of-function mutation. Moreover, our results would suggest that miti (or miti together with pdm1)is required in segmentation for the formation of A2 and A6, and in the CNS for the elaboration of RP2/sib lineage. Given the variable penetrance of these segmentation and CNS phenotypes, it seems likely that the hsp83 Δmiti transgene mimics a hypomorphic, not a null mutation. Consequently, functions, in addition to these, can certainly be expected. (Indeed, in the CNS other miti-expressing lineages are also affected by the hsp83 Δmiti transgene; unpublished data.)

While the hsp83 transgene is useful for genetic analysis, drawing firm conclusions about the functions of the miti gene from the phenotypic effects of the transgene at different stages of development is not necessarily straightforward. For example, it could be argued that defects in CNS development are a secondary consequence of earlier disruptions in the seg-mentation process (cf, the effect of the paird mutation on RP2/sib lineage, Fig. 7A,B). Difficulties in sorting out primary from secondary effects is a problem that is not unique to the hsp83 Δmiti transgene, but is typically encountered with non-conditional loss-of-function mutations. To better distinguish between primary and secondary effects, we generated a second Δmiti transgene using the heat-inducible hsp70 promoter. This transgene mimics the effects of a conditional loss-of-function mutation. When grown at the permissive temperature, even two copies of the hsp70 Δmiti transgene have no effect on development; however, a brief heat shock can induce the same type of developmental perturbations as are observed with the hsp83Δmiti transgene. Moreover, by heat shocking transgenic embryos at different times of development, it is possible to separate miti function in segmentation and CNS development. Segmentation defects are induced only in the first 2 hours of embryogenesis (early cleavage to syncitial stage), while defects in the elaboration of the RP2/sib lineage are induced much later in embryogenesis between 6 and 7 hours of devel-opment (mid to late stage 9). In both cases, this is just prior to the time when expression of the endogenous miti (and pdm1) gene can first be detected.

Our work together with the studies of Yang et al. (1993) suggests that the miti gene plays a key role in the elaboration of the NB4-2→GMC-1→RP2/sib lineage. We found that the hsp83– and hsp 70 Δmiti transgenes interfere with the initial specification and/or subsequent development of GMC-1. In some cases, we could not detect eve-positive GMC-1 cells. In other cases, they are only very weakly stained with eve antibody. Since the endogenous miti gene appears to be first expressed after GMC-1 is formed, it is unlikely to be required for the division of NB4-2. This would suggest that NB4-2 probably divides to form a ‘GMC-1’-like cell, but the identity of this cell is not correctly specified and as a consequence it fails to properly express the eve marker. Whether this incorrectly specified ‘GMC-1’-like cell subsequently dies, or assume some other fate (eg., that of its ‘sister’ GMC-2) because of the failure in miti function remains to be deter-mined. We also find cases in which eve-positive GMC-1 cells are formed but they do not migrate to their proper position close to the ventral midline. It would appear that these cells assume at least some aspects of GMC-1 identity (eg., apparently normal eve expression) but are then defective in fully executing the GMC-1 developmental program. That the specification of GMC-1 represents a normal function of miti is supported by our analysis of the eve staining pattern in a deficiency that removes the miti gene (and the related pdm1 POU gene). In this deficiency, eve-positive GMC-1 cells are absent. In contrast, eve-positive GMC-1 cells can be found in a defi-ciency which retains miti but removes pdm1.

High levels of miti are detected in GMC-1 very soon after it is formed, and we suppose that this newly synthesized protein plays a critical role in establishing GMC-1 identity. In this view, the Δmiti transgenes would interfere with the deter-mination process by reducing the level of functional miti protein below a critical threshold. That threshold levels of miti are a key factor in establishing cell identity in this lineage is supported by our analysis of the effects of increasing the dose of the wild-type miti gene. Normally GMC-1 divides into two eve-positive cells, one of which is the sib, while the other dif-ferentiates into the RP2 neuron. When 4 copies of miti are present, three eve-positive cells are sometimes observed. Since it is possible to demonstrate a duplication of the RP2 neuron, we presume that these three cells represent either two RP2 cells and one sib, or the converse, one RP2 cell and two sibs. When even high levels of the full-length miti protein are expressed from an hsp70 transgene, four, not three, eve-positive cells are observed (Yang et al., 1993).

Taken together these results would suggest a model in which miti function at different points in the RP2/sib lineage depends upon threshold protein concentrations (Fig. 11). In this model a critical threshold level of miti protein would initially be required to establish GMC-1 identity when this cell is first formed from the asymmetric division of NB4-2. As the GMC-1 cell prepares to divide the concentration of miti drops sub-stantially. This drop in the level of protein would then allow the two daughter cells to assume fates distinct from GMC-1, and they would be able to follow appropriate differentiation pathways (cf, Bhat et al., 1988). In the presence of elevated amounts of miti protein in animals carrying four copies of the gene, the division of GMC-1 would still be asymmetric. However, one cell would reiterate the GMC-1 state, while the other would become either RP2 or sib. In this scenario, the con-centration of miti protein in the cell which reiterates the GMC-1 state is above the critical threshold value, while it is below this threshold in the cell that becomes committed to the differ-entiation pathway. Finally, at very high levels of protein, as would be produced by the hsp70miti transgene of Yang et al. (1993) the initial division of GMC-1 would be symmetric, giving two GMC-1 cells, which ultimately give rise to two RP2s and two sibs. The selection of different cell fates depending upon the concentration of a key ‘regulatory’ molecule as proposed here for the RP2/sib lineage is not unique, but in fact seems to be a fairly common theme in development. For example, the formation of anterior structures in the Drosophila embryo depends upon a concentration gradient of the bicoid protein, while the specification of cell fate along the dorsal-ventral axis depends upon the nuclear concentration of the dorsal protein (Driever and Nusslein-Volhard, 1988; Steward, 1989; Roth et al., 1989).

We thank Drs Yash Hiromi, Akira Chiba, Henrik Gyurkovics, Janos Gausz and Nipam Patel for their suggestions and comments on this work, and Dr Steve Poole and Andrew Billin for pdm1 and miti/pdm2 antibodies. We also acknowledge Stella Han for technical assistance, Jamila Horabin for hsp83 and hsp 70 promoter CaSpeR vectors, and Kathy Mathews at the Drosophila Stock Center, Indiana University for various fly stocks, Joe Goodhouse for help in confocal microscopy and Gorden Grey for fly food. Help from Prema Bhat for handling various Drosophila stocks is much appreciated. This work has been supported by a grant from NIH.

Recently we have isolated lethal mutations in miti and pdm1 genes. The preliminary results obtained with miti mutants are consistent with the conclusion of the paper that miti is involved in the specification of the GMC-1. As with Δmiti, in miti mutant eve positive GMC-1 cells are absent. Further, the pen-etrance of this defect can be increased by removing one copy of the pdm1 gene suggesting that during the specification of this lineage the two genes interact with each other and that there may be some partial redundancy between the two in this lineage.