Headless flies produced by mutations in the paralogous Pax6 genes eyeless and twin of eyeless

Pax genes, discovered by a test of the gene network hypothesis (Bopp et al., 1986; Frigerio et al., 1986), play crucial roles in the development of the central nervous system (CNS) and brain, as well as of the peripheral nervous system and sensory organs (Noll, 1993; Dahl et al., 1997). They encode transcription factors that bind to specific DNA sequences by virtue of a conserved paired-domain and have been divided into four different classes according to the four primordial Pax genes present at the time of protostome-deuterostome separation (Noll, 1993). Two classes include paired-domain proteins with a second DNA-binding domain, a prd-type homeodomain. Pax genes belonging to the same class frequently overlap in their spatiotemporal expression patterns because the product of one gene activates transcription of the other (Noll, 1993). In contrast, paired-domain proteins of separate classes avoid such overlaps, presumably because they activate interfering developmental pathways (Jiao et al., 2001). In Drosophila, for example, the product of the pair-rule gene paired activates the segment-polarity gene gooseberry in the epidermis (Li et al., 1993), whose product in turn activates gooseberry-neuro in the neuroectoderm (Gutjahr et al., 1993). All three genes belong to the same class as the mammalian Pax3/Pax7 genes. Similarly, the four Drosophila genes eyeless (ey), twin of eyeless (toy), eye gone (eyg), and twin of eyg (toe) that belong to the vertebrate Pax6 class are expressed in the same developmental pathway controlling eye development (Quiring et al., 1994; Jun et al., 1998; Czerny et al., 1999; Jang and Sun, 2001). Also in this case, it has been proposed that the Toy protein is required for the activation of the ey gene, both of which were thought to be required for proper development of the eye disc, yet not of the antennal disc (Czerny et al., 1999). In the absence of toy mutants, however, this proposal was based entirely on experiments testing for ectopic activation of ey upon ectopic expression of toy.

In order to clarify the roles of the two Pax genes ey and toy in the development of normal rather than ectopic eyes, we examined the effect on eye development of the first toy mutations and of a much stronger ey allele than previously characterized. We concentrated particularly on the earliest period of ey and toy expression in the eye-antennal primordia in embryos (Czerny et al., 1999). This period has been shown to be particularly sensitive to developmental pathway interference, which generates headless flies (Jiao et al., 2001), a phenotype much stronger than that observed for ey mutants (Quiring et al., 1994; Halder et al., 1998). As this indicated much wider roles for toy and ey in head development, we set out to find strong mutants of both toy and ey in order to examine if these displayed a headless rather than the milder eyeless phenotype.

We found that strong mutants of either toy or ey indeed lack all structures derived from eye-antennal discs and thus exhibit a spectacular headless phenotype. In addition, activation of ey in the primordia of eye-antennal discs does not strictly depend on the presence of the product of the toy gene, but becomes temperature-sensitive in its absence. In the absence of Toy protein, moderate levels of Ey protein are sufficient to rescue the headless phenotype, while high Ey levels are necessary to rescue the eyeless phenotype as well. In contrast, very low levels of Ey apparently suffice to rescue the headless phenotype in the presence of wild-type Toy protein. These findings suggest a delicate balance of Ey protein levels regulating eye-antennal disc development and a partial redundancy of Ey and Toy functions in these discs, in which Toy acts not only through Ey, but also in a pathway parallel to Ey. Finally, inhibition of apoptosis by the baculovirus P35 protein is able, in the absence of functional Ey, to rescue the headless, but not the eyeless phenotype. Hence, Ey is not responsible for the development of head structures derived from the antennal disc, but is primarily required to inhibit cell death and to promote eye development.

The preparation of genomic libraries, isolation of poly(A)⁺ RNAs or cDNAs, 5′- and 3′-RACE of mRNAs, northern blot analysis, PCR, DNA sequencing, and in situ hybridization to salivary gland chromosomes were carried out according to standard procedures. Mutant stocks not obtained from the Umea and Bloomington Stock Centers have been described previously (Fu et al., 1998; Jiao et al., 2001) or are characterized here.

To assay the lethal phase of homozygous toy^hdl mutants, newly hatched first instar larvae were picked from a toy^hdl/spa^Cat stock and the fractions that reached the pupal and adult stages determined. These fractions were the same at all temperatures tested (18, 23, 25, and 28°C). All viable adults were heterozygotes because all spa^Cat homozygotes die during embryogenesis (Hochman, 1976). From a total of 3250 larvae, 2728 developed to pupae and 1853 to viable adults. Assuming that of the 15% of heterozygotes that fail to reach the adult stage half each die as larvae or pupae, we find that the lethality of toy^hdl mutants is 34±2% during the larval and 66±3% during the pupal stage.

In situ hybridization with antisense RNA probes to staged embryos was carried out according to standard procedures. To determine what fraction of toy^hdl mutants showed a reduced level of ey transcripts in the eye-antennal primordia, their embryonic stage was determined and the level of ey transcripts compared to that of late wild-type embryos of the same stage. Among embryos derived from toy^hdl/l(4)2C2 parents, the level was reduced in 32 out of 127 late stage 16 embryos at 25°C, which is as expected if it was reduced in all homozygous toy^hdl embryos; and in 10 out of 124 such embryos at 18°C, i.e., in a third of all toy^hdl embryos (Fig. 5A,C,E). For embryos derived from Df(4)spa⁶⁶/l(4)2C2 parents, it was first verified that 25% of all stage 16 embryos were homozygous for the deficiency, by the absence of toy transcripts after in situ hybridization with a toy cDNA probe, before the same analysis for ey transcript levels was carried out. Again the level was reduced in a quarter (24/100) of late stage 16 embryos at 25°C, and in about half (10/71) of the expected number of homozygous Df(4)spa⁶⁶ embryos of the same stage at 18°C (Fig. 5G,H). Similar results were obtained for embryos from Df(4)spa³⁰/l(4)2C2 parents: ey transcripts in the eye-antennal primordia were clearly reduced in 17 out of 62 embryos at 25°C, while they were reduced in only 13 out of 100 embryos at 18°C.

Both Drosophila homologs of the vertebrate Pax6 gene, ey and its paralog toy, are located on the small fourth chromosome (Czerny et al., 1999). No mutants of toy have been described, while only hypomorphic ey alleles have been identified, the strongest of which give rise to eyeless adult flies (Quiring et al., 1994; Halder et al., 1998). This is surprising since the fourth chromosome has been subjected to an extensive systematic screen for lethal mutations in an attempt to establish its total number of such loci (Hochman et al., 1964; Hochman, 1971; Hochman, 1976). However, a change in nomenclature of the mutant alleles, by which new alleles (Hochman, 1971; Hochman, 1976) were given names of alleles found in a previous screen (Hochman et al., 1964), caused considerable confusion, and many of these mutants were lost. In May 1994, in the course of examining the phenotype of some of these mutant stocks more closely, one of us (Å. R.-L.) discovered that homozygous l(4)8 flies (Hochman et al., 1964), which die as pharate adults inside their pupal case, have a spectacular phenotype. When raised at 28°C, 65% of these pharates were headless, developing no structures derived from the eye-antennal discs (Fig. 1A,B), while less extreme phenotypes had partially developed heads like ‘half-heads’ (Fig. 1C) and ‘cleft-heads’ (Fig. 1D). The weakest phenotype was a ‘small-eye’ (Fig. 1E), similar to heterozygous ey^D flies, or it was indistinguishable from wild type (Fig. 1F). According to its most extreme phenotype, the l(4)8 mutant allele was called headless (hdl).

The relative proportions of strong, intermediate and weak phenotypes observed among homozygous hdl flies depend on the temperature at which they are raised. Thus, at 18°C about 3% of these flies survived to apparently wild-type fertile adults, while 65% of the pharates had normal heads and only about 3% were headless. The remaining pharates developed partial heads and consisted of about equal portions of half-heads (Fig. 1C), cleft-heads (Fig. 1D) and small-eyed flies (Fig. 1E). In addition, about 1% each of the flies died as pharates in which only one eye and nearly no other head structures developed (one-eyed pharates) or as headless pharates with an eye inside their thorax (eye-in-thorax pharates). In contrast, at 28°C no hdl flies survived to viable adults, 65% of the pharate adults were headless (Fig. 1A,B), and only about 10% exhibited an apparently wild-type head phenotype. The remaining 25% of the pharates consisted of about equal portions of eye-in-thorax, one-eyed, and half-head flies, while only about 1% each of the pharates were small-eyed or cleft-heads.

This strong temperature sensitivity of the hdl allele permits us to test whether its temperature-sensitive functions are restricted to a critical period or required throughout development. To this end, temperature-shift experiments were carried out during which embryos developing at 19.5°C were transferred to 28°C (shift-up experiments; Fig. 2A) or embryos developing at 28°C were shifted to 19.5°C (shift-down experiments; Fig. 2B) after various time intervals. In both temperature-shift experiments, the temperature-sensitive period was limited to a short interval from stage 12 to 16 (Campos-Ortega and Hartenstein, 1997), when toy and ey transcripts begin to appear in the anlagen of the eye-antennal discs (Jürgens and Hartenstein, 1993; Czerny et al., 1999). Temperature shifts after this phenocritical period showed the same proportions of headless phenotypes as if no shift had occurred.

In addition, about one third of the hdl homozygotes die as larvae (compare Materials and Methods), in agreement with the observed lethal phase of this mutant (Hochman, 1976). In contrast to the headless phenotype, larval lethality is independent of temperature between 18°C and 28°C. Thus, the hdl allele affects not only functions crucial for the proper development of the eye-antennal disc, but also additional vital functions required for survival to the pupal stage.

Headless flies have been observed previously when ectopic expression of transcription factors in the eye-antennal primordia interfered with the developmental program initiated by the toy and/or ey genes (Jiao et al., 2001), both of which are located on the fourth chromosome (Czerny et al., 1999). Because hdl also maps to the fourth chromosome, we suspected that it is either the first mutant allele of toy or an allele of ey that is stronger than any of the previously characterized ey alleles. To test this hypothesis, we mapped hdl by complementation analysis, using a number of deficiencies that had been isolated in screens for D-Pax2 mutants (Fu et al., 1998), a deficiency uncovering the ey locus (a generous gift from K. Basler) (Brunner, 1997), three ey alleles, and a few additional mutant alleles thought to belong to the same complementation group as ey or hdl on the basis of Hochman’s complementation analysis (Hochman, 1971). The resulting genetic map shows that the hdl mutation is located distal to ey and proximal to D-Pax2 (Fig. 3A). Therefore, hdl is not an ey allele, but its genetic location is consistent with the possibility that it is allelic to toy. To test this possibility, we isolated genomic clones of toy from wild-type and hdl λ phage libraries. Comparison of their restriction maps indeed suggested that sequences downstream of exon 5 of toy are deleted in hdl mutants (Fig. 3B). Isolation of the deficiency by PCR and subsequent DNA sequence analysis corroborated this conclusion and demonstrated that hdl is a 5,863 bp deletion of the 3′ moiety of the toy transcript (Fig. 3B). It follows that hdl is the first mutant allele of toy and hence was named toy^hdl. Our finding is further consistent with a recent report, mentioning a headless phenotype of toy mutants as unpublished results (Kammermeier et al., 2001).

A developmental profile of toy transcripts obtained by northern blot analysis shows a strong and a weak mRNA band at about 3.8 kb and 2.7 kb, respectively, that are present throughout development as well as in adult flies (Fig. 4A). In addition, another weak band of 2.4 kb becomes visible in third instar larvae and continues to be expressed in subsequent stages. Since no maternal toy transcripts are detected, all bands are of zygotic origin. Their sizes are consistent with the mRNA lengths derived from toy cDNAs, extended by 5′- and 3′-RACE (Fig. 3B). Thus, the major toy mRNA includes a rather long 1.6 kb untranslated trailer sequence, while the minor bands reflect much shorter trailers (Fig. 3B). Analysis of toy mRNAs produced in homozygous or heterozygous toy^hdl flies reveals relatively strong bands at 4.7 kb and 2.8 kb, and weaker bands at 3.6 kb and 1.7 kb (Fig. 4B). The lengths of these new mRNAs indicate that the stronger bands correspond to the major 3.8 kb, the weaker ones to the minor 2.7 kb wild-type mRNA, if one assumes that intron 5 was either not removed or removed by splicing to a cryptic 3′ acceptor site close to the deficiency breakpoint in exon 9 (Fig. 3B). No change in ey mRNAs is observed in toy^hdl mutants, which confirms that the toy^hdl chromosome is wild-type at the ey locus (Fig. 4).

The observation that the levels of wild-type and toy^hdl mRNAs are about the same in adults (Fig. 4B) suggests that the deficiency does not remove a major enhancer, active in adults, and that the stabilities of the toy^hdl mRNAs are not significantly different from those of their wild-type counterparts. This result has been extended to embryos by in situ hybridization of a probe complementary to the 5′ portion of the toy mRNA that is not deleted by the toy^hdl deficiency. No difference in the patterns of toy transcripts was observed between homozygous toy^hdl and wild-type embryos (not shown). In contrast, when a probe covering the deleted exons 6-8 was used, a quarter of the embryos derived from heterozygous parents displayed no signal. It follows that none of the toy enhancers active in embryos or adults is affected by the toy^hdl deficiency. Moreover, we expect that the mutant produces a Toy^hdl protein that includes the N-terminal paired-domain but is truncated in the homeodomain (Fig. 3C).

It has been proposed that the Toy protein activates ey in the eye developmental pathway. In the absence of toy mutants, this proposal was based (i) on the observation that ectopic eye formation induced by ectopic expression of toy in leg discs was dependent on a functional ey gene, and (ii) on in vitro binding studies of Toy to sites in a minimal ey enhancer whose activation of a reporter gene in the eye-antennal primordia was reduced upon mutation of the Toy binding sites (Czerny et al., 1999). Although these results indicate that the Toy protein activates the ey gene in the eye primordia, they do not prove it for two reasons. First, studies on ectopic eye formation do not strictly provide information about the normal pathway of eye formation, particularly since ectopic eye formation is possible in the absence of toy expression (Czerny et al., 1999). Second, mutation of the Toy binding sites in the minimal ey enhancer did not eliminate reporter gene expression in the entire eye-antennal primordium, but strong expression was still observed in its posterior portion, presumably where the eye primordium is located, perhaps because the minimal enhancer did not include all cis-regulatory elements required for proper expression of the ey gene during eye development (Czerny et al., 1999). Moreover, as evident from the toy^hdl phenotype (Fig. 1A,B), Toy is required not only for eye formation, but for the development of the entire eye-antennal disc and exhibits a phenocritical period at the time when the eye-antennal anlage forms and begins to express ey (Fig. 2).

The identification of deficiencies uncovering the toy gene and of hdl as a toy mutation (Fig. 3A,B), however, allows us to perform a more critical test of whether toy acts upstream of ey during normal eye and head development: the analysis of the expression of the ey gene in the eye-antennal primordia of toy mutant embryos at the end of the phenocritical period (Fig. 5C-H). Transcription of ey was strongly reduced, though not completely eliminated, in the eye-antennal primordia of most toy^hdl as compared to wild-type embryos when they developed at 25°C (compare Fig. 5C with 5A). If development occurred at 18°C, however, ey transcript levels were not reduced as much, but still clearly reduced in a third of the toy^hdl embryos (Fig. 5E) and indistinguishable from wild type in the remaining toy^hdl embryos. These findings prove that toy acts upstream of ey in the eye-antennal anlage at the time of the initial ey activation. In addition, they provide a simple explanation for the temperature sensitivity of the toy^hdl allele since the fractions of strong (Fig. 1A,B) and less severe headless phenotypes (Fig. 1C-E) correspond well to the fractions of embryos exhibiting reduced levels of ey transcripts in the eye-antennal primordia at the high (Fig. 5C) and low temperature (Fig. 5E). We conclude that ey transcription in the eye-antennal primordia of toy^hdl embryos is temperature-dependent and that the most severe headless phenotypes arise only if transcriptional activation of the ey gene in the eye-antennal primordia does not exceed a threshold level that is clearly detectable (Fig. 5C).

In contrast to ey transcription in the eye-antennal primordia, transcription of ey in the CNS appears to remain unaffected in toy^hdl embryos (Fig. 5B,D). This confirms an earlier conclusion based on the observation that ey transcripts precede those of toy and display a pattern different from toy transcripts in the ventral nerve cord (Czerny et al., 1999).

The temperature dependence of ey transcription in the eye-antennal primordia of toy mutants does not prove that the Toy^hdl protein is the cause of the temperature sensitivity of the headless phenotype. It is possible that even in the absence of Toy protein activation of ey transcription remains temperature dependent. To test this possibility, we examined ey expression in embryos homozygous for Df(4)spa³⁰ or Df(4)spa⁶⁶, both of which fail to express toy transcripts (Materials and Methods). Surprisingly, in all homozygous mutant embryos the level of ey transcripts was similarly reduced as in toy^hdl embryos, but not eliminated, and still depended on temperature to the same degree (Fig. 5G,H). It follows that it is not the truncated Toy^hdl protein that is temperature dependent, but rather the activation of ey transcription in the absence of a functional Toy protein. This conclusion is consistent with our finding that similar fractions of homozygous toy^hdl pharates and of toy^hdl pharates transheterozygous for one of the two deficiencies show a headless phenotype when tested at the same temperature between 18°C and 28°C. We conclude that toy^hdl behaves as a null allele with respect to the activation of ey at the phenocritical stage as well as with regard to the severity of the resulting headless phenotypes.

Since the severity of the headless phenotype correlates inversely with the level of ey transcripts in the eye-antennal primordia, we would expect that strong ey mutants also produce headless pharates if toy acts mainly through ey in eye-antennal development. None of the two defined ey mutants, ey² and ey^R, exhibits, however, a headless phenotype (Bridges, 1935a; Quiring et al., 1994; Halder et al., 1998), possibly because both are viable hypomorphs, whereas ey null alleles, all of which have been lost, are pupal lethals (Hochman, 1976). An additional putative ey mutation, ey^D, X-ray-induced by Muller in 1927 and mapped to the fourth chromosome (Patterson and Muller, 1930), has not been analyzed at the molecular level. Although its allelism to ey has been questioned (Bridges, 1935a), it is pupal lethal (Patterson and Muller, 1930; Bridges, 1935b) and has been characterized to be associated with the translocation of about a dozen chromosomal bands that are inserted as reversed repeat into the region 102D of the fourth chromosome (Bridges, 1935b) (Fig. 6C-E), known to include the ey locus (Quiring et al., 1994). Moreover, complementation analysis indicated that ey^D might be allelic to ey (Patterson and Muller, 1930) (Fig. 3A).

To test if ey^D is a lethal mutation of the ey gene, we isolated ey DNA clones from a genomic library of ey^D mutants and compared their EcoRI restriction maps to that of the wild-type ey locus (Quiring et al., 1994; Hauck et al., 1999). This analysis revealed that the large insertion on the fourth chromosome of ey^D mutants indeed interrupts the ey gene (Fig. 6A). Sequence analysis of mutant clones covering the breakpoints of the insertion showed that the ey gene is interrupted in exon 5 and that the adjacent 320 bp of exon 5 are replaced by a large insertion originating from the second chromosome (Fig. 6A,C). This insertion consists of a large reversed repeat (Bridges, 1935b), whose point of inversion is at 23D1,2 (Fig. 6C). Further analysis (Fig. 6D,E) showed that the ey gene is transcribed towards the telomere (Fig. 6A).

In situ hybridization demonstrates that in about half of the homozygous ey^D embryos ey transcripts of the first five exons are as abundant in the eye-antennal primordia as in wild-type embryos, while in the other half they are reduced to about half the wild-type level, both at 18°C and 25°C (not shown). In contrast, no transcripts of the exons downstream of the ey^D insertion are detectable in homozygous ey^D embryos (not shown). These findings imply that the insertion has no strong effect on the embryonic activities of the ey enhancers and that the stability of the truncated ey^D transcripts is comparable to that of wild-type ey mRNA, independent of temperature within the examined range. It is therefore reasonable to assume that ey^D transcripts are translated to produce a truncated Ey protein that includes the N-terminal paired-domain and a Ser/Thr-rich domain, possibly an activation domain, but lacks all C-terminal domains, including the homeodomain (Fig. 6B).

In about a fifth of the homozygous or heterozygous ey^D embryos, the first five exons of ey were ectopically expressed in a metameric pattern of 11 lateral, later ventrolateral, posterior segmental stripes after germband retraction (Fig. 5I,J). Hence, a nearby enhancer of the inserted second chromosome sequences seems to activate the ey^D gene at low penetrance.

The homozygous ey^D condition is lethal (Patterson and Muller, 1930) with a lethal phase during the pupal stage (Bridges, 1935b), in agreement with the lethal phase of other strong ey alleles (Hochman, 1976) that, however, were lost. To investigate if homozygous ey^D mutants display a headless phenotype as expected, we dissected ey^D pharate adults from their pupal case and examined them by scanning electron microscopy. We found that indeed about half of these pharates showed a strong headless phenotype (Fig. 7B,C) missing all or most structures derived from wild-type eye-antennal discs (Fig. 7A), in agreement with an earlier report (Arking et al., 1975). The remaining pharates developed a considerable portion of the head and antennae, yet no eyes, and exhibited phenotypes most of which were stronger than the eyeless phenotype shown in Fig. 7D.

Both ey² and ey^R, shown to consist of transposon insertions into the eye-specific enhancer of ey (Quiring et al., 1994), on average enhance the heterozygous ey^D phenotype to produce smaller eyes (cf. Fig. 7E,F with 7G), but never headless flies. This finding strongly suggests that these two ey alleles are not null alleles, but retain some functions in developing eye-antennal discs even though they seem not to be expressed in these discs (Quiring et al., 1994) (our unpublished observation). The dominant phenotype of ey^D/+ adults whose eye size is moderately reduced (Fig. 7G) is probably the result of a dominant-negative effect of the truncated Ey^D protein that inhibits the wild-type Ey protein by competing for its targets during larval eye development, while haplo-insufficiency of ey is excluded by the wild-type phenotype of Df(4)BA/+ adults that are hemizygous for ey (Fig. 3A). Competition of Ey^D with targets of Toy appears improbable because the similarly truncated Toy^hdl protein does not exhibit a dominant-negative effect.

In combination with nearby deficiencies, some of which delete the toy gene but none the ey gene (Fig. 3A), the phenotype of heterozygous ey^D flies is affected in various ways. Thus, the size of the eye in ey^D/Df(4)spa³⁰ flies varies over a wider range than, but on average is similar to, that of ey^D/+ flies (Fig. 7H), while Df(4)spa⁶⁶ reduces the average eye size of heterozygous ey^D flies to about 10% of wild type (not shown). In contrast, Df(4)spa⁴⁷, which does not delete toy (Fig. 3A), seems to suppress the ey^D/+ phenotype significantly (not shown). These effects on the eyeless phenotype may result in part or entirely from different genetic backgrounds. For example, continued selection for small eye sizes among y w; ey^D/Dp(1;Y;4)y⁺, spa^pol offspring obtained from repeated crossings inter se, yields increasing numbers of flies that have no eyes (Fig. 7I). Such variable effects of the genetic background on the ey^D/+ phenotype may explain why ey^D could not be identified unambiguously as an allele of ey by mere genetics.

Since about half of the ey^D pharates lacked most derivatives of the eye-antennal discs, we investigated whether inhibition of apoptosis by expression of the baculovirus P35 protein (Hay et al., 1994), under the indirect control of the eye-antennal specific enhancer of the ey gene, had any effect on eye-antennal disc development. Astonishingly, more than half of such ey^D flies are rescued by ey-Gal4>UAS-P35 to viable adults that are eyeless, but have developed most other head structures derived from the eye-antennal discs, usually including all three ocelli (Fig. 7J). The heads of all these flies are rescued since their phenotypes are much weaker than the weakest ey^D phenotype (Fig. 7D). These results imply that the headless phenotype of ey^D mutants is the result of considerable cell death in the eye-antennal disc, a process that is inhibited by the wild-type Ey protein.

We have shown that an old mutant, l(4)8, induced by X-ray in 1957 by Gloor and Green (Hochman et al., 1964), displays a headless phenotype. This toy^hdl allele turned out to be the first mutant allele of the Pax gene toy, previously isolated as a paralog of the ey gene and proposed to activate ey to initiate eye development (Czerny et al., 1999). The dramatic phenotype of toy^hdl mutants, however, implies that toy is not only necessary for proper development of the eye, but for that of all structures derived from the eye-antennal disc. Moreover, we have shown that the same is true for the ey gene by demonstrating that ey^D is an allele of ey and much stronger than any ey alleles previously characterized (Quiring et al., 1994). The headless phenotype of ey^D pharates results from the induction of apoptosis in the eye-antennal disc, as evident from its rescue by the expression of the baculovirus P35 protein in eye-antennal primordia and discs. Our findings are consistent with an old conclusion that “at no time in development (except at the very end) is there a group of identified and committed cells which will give rise to the eye and to nothing else” (Martinez-Arias and Lawrence, 1985).

Temperature shift experiments with toy^hdl mutants show that the headless phenotype critically depends on the absence of Toy protein activity during stages 12-16 of embryogenesis (Fig. 2). In toy^hdl mutants, up to 80% of the pharate adults are headless at 28°C, whereas this phenotype is nearly completely suppressed at 18°C. Shifting the temperature down to the permissive temperature at the end of stage 16 demonstrates that after this stage Toy^hdl is unable to provide any function that would be able to rescue the headless phenotype, while shifting the temperature up to the non-permissive temperature at this time shows that Toy^hdl can provide all the functions necessary, if any, to rescue the headless phenotype at all temperatures after stage 16. It follows that the temperature-sensitive function of Toy^hdl with regard to the headless phenotype is restricted to the phenocritical period during stages 12 to 16.

Homozygous ey^D pharates exhibit the same headless phenotype as toy^hdl mutants. Since we have demonstrated that ey^D is a strong allele of the ey gene, we conclude that the truncated Ey^D protein, if translated from the ey^D mRNA, is unable to provide the functions necessary for eye-antennal disc development. It was therefore important to know if ey transcription depended on Toy during the phenocritical period, as had been shown previously for ectopic eye formation, but could not be tested for normal eye development because of the lack of toy mutants (Czerny et al., 1999). We found that ey transcription indeed depends on toy activity. Surprisingly, however, even in the absence of Toy protein ey transcription, (i) remains temperature-dependent during the phenocritical period, and (ii) is not completely eliminated. Hence, the observed temperature-sensitivity of the headless phenotype of toy^hdl mutants is not a property of the truncated Toy^hdl protein, but rather of the transcriptional activation of the ey gene in the absence of a functional Toy protein.

We propose that during the phenocritical period, in addition to Toy, other transcription factors bind to the eye-antennal enhancer of the ey gene to activate its transcription in the eye-antennal primordia. In the absence of Toy, these factors are able to activate ey transcription sufficiently at low but not at high temperatures to ensure normal eye-antennal development. The simplest explanation for this temperature-sensitive activation of ey is that formation of the transcription factor complex bound to the eye-antennal enhancer or its activation of the basal transcription machinery becomes temperature-dependent in the absence of Toy protein. Thus, the main function of Toy is to stabilize this transcriptional activator complex on the eye-antennal enhancer of ey. It is interesting to note that three Toy binding sites have been found in the eye-antennal enhancer of ey, two of which are immediate neighbors (Czerny et al., 1999). Truncated Toy^hdl probably binds with similar affinity to these sites as wild-type Toy protein (Punzo et al., 2001). However, its lack of C-terminal activation domains may fail to stabilize its own binding and that of cofactors as a result of which the basal transcription machinery is not efficiently recruited and activated. The fact that a third of the toy^hdl mutants die during larval stages and that toy^hdl mutants that show normal eye-antennal development still die as pharate adults shows that Toy is more strictly required for the activation of other enhancers necessary for the development of viable adults.

Interestingly, ey transcription is not abolished, but reduced levels of ey transcripts remain detectable in eye-antennal primordia of toy^hdl embryos, even at temperatures at which most of them develop to headless pharates. This finding implies that there is a delicate balance of ey activation for inducing eye-antennal development. If transcript levels do not exceed a relatively high threshold value, the program for eye-antennal disc development cannot proceed. This delicate balance is particularly evident when it is temperature-dependent in the absence of a functional Toy protein, for example, in toy^hdl mutants. In many instances, unequal ey transcript levels have been observed in the left and right eye-antennal primordia of toy^hdl embryos. Accordingly, ey transcripts may surpass the threshold in only one of the two eye-antennal primordia and thus give rise to pharates with only one half of the head developing normally (Fig. 1C). Even within the eye-antennal primordium, ey transcript levels may vary in toy^hdl mutants (Fig. 5E) and thus explain phenotypes like the cleft-head (Fig. 1D).

In toy^hdl mutants, reduced yet detectable (moderate) levels of ey transcripts in eye-antennal primordia are unable to rescue the headless phenotype. This is evident from the reciprocal correlation between ey transcript levels in eye-antennal primordia during the phenocritical period and the fraction of headless pharates at different temperatures of development. In contrast, ey² mutants, in which no ey transcripts have been detected in eye-antennal primordia, never display a headless phenotype and mostly have eyes of only slightly reduced size (Quiring et al., 1994) (our unpublished observation), whereas ey^D mutants show a high penetrance of headless pharates. The following considerations illustrate that this apparent contradiction is resolved by the assumptions that (i) Toy and Ey share partial functional redundancy in eye-antennal disc development, and (ii) ey² expresses very low levels of wild-type Ey protein whose mRNA escapes detection (Fig. 8). In the absence of functional Ey protein, as in ey^D mutants, normal levels of functional Toy protein rescue the headless phenotype at low efficiency. In ey² mutants, however, very low levels of functional Ey protein are sufficient to rescue the headless phenotype completely in the presence of normal Toy levels and are even able to promote nearly normal eye development (our unpublished results). In contrast, in the absence of functional Toy protein, as in toy^hdl mutants, much higher levels of ey transcripts are necessary to rescue the headless phenotype, which is achieved more efficiently at lower temperatures. Consequently, the headless phenotype is observed only in the complete absence of functional Ey protein or in the absence of functional Toy if Ey does not exceed a moderate level.

This partial redundancy of Toy and Ey functions implies that Toy does not act exclusively through the activation of ey, but also in a pathway parallel to ey, to promote eye-antennal disc development (Fig. 8). Such an ey-independent pathway downstream of toy may include the optix gene, a paralog of the sine oculis gene (Seimiya and Gehring, 2000) which is in turn a target of ey (Halder et al., 1998; Niimi et al., 1999), and the eyg gene (Jun et al., 1998; Gehring and Ikeo, 1999).

The most crucial function of ey and most sensitive to the level of Ey during eye-antennal development is the inhibition of cell death. This became evident from a successful attempt to rescue the ey^D headless phenotype by the expression in eye-antennal discs of the baculovirus P35 protein, an inhibitor of apoptosis (Fig. 7J). Astonishingly, P35 is able to rescue more than half of the ey^D embryos to viable adults with normal head structures, with the exception of the eyes that still do not develop.

The fact that inhibition of apoptosis in ey^D eye-antennal discs is unable to rescue eye development argues that additional functions of ey during eye-antennal development are required and restricted to the eye disc, in agreement with its expression pattern in eye-antennal discs (Quiring et al., 1994) and a recent analysis (Kumar and Moses, 2001). It has been shown that interference with the program of eye-antennal development by ectopic expression of transcription factors also generates headless pharates if interference is restricted to exactly the same phenocritical period observed here (Jiao et al., 2001). However, in this case interference could not be antagonized by P35 expression, but only by overexpression of Ey, CycE or Myc. It was concluded that one of the earliest functions of ey must be the activation of the cell cycle (Jiao et al., 2001). Combining this conclusion with the results reported here, we propose that in the presence of developmental pathway interference it is impossible to obtain rescue by the inhibition of apoptosis in the absence of cell cycle activation, whereas inhibition of apoptosis appears to suffice in ey^D mutants, possibly because the truncated Ey^D and/or the Toy protein are able to activate the cell cycle.

We are grateful to K. Basler and the Bloomington and Umea Stock Centers for fly stocks, U. Jauch (Institut für Pflanzenbiologie, University of Zürich) and P. Hörstedt for scanning electron microscopy, K. Kristiansson for technical assistance, and F. Ochsenbein for art work. We thank H. Noll for criticism and comments on the manuscript. This work has been supported by the Lawski Foundation (to J. K.), the Swiss National Science Foundation grants 31-40874.94 and 31-56817.99 (to M. N.), and by the Kanton Zürich.