Transcriptional regulation of string (cdc25): a link between developmental programming and the cell cycle

From the past 15 years of genetic and molecular studies of Drosophila, a molecular description of the initial stages of embryonic pattern formation has emerged (see Lawrence, 1992; Bate and Martinez-Arias, 1993, for reviews). A cascade of genetic interactions directs the expression of numerous transcription factors in localized patterns in the blastoderm stage embryo and these transcription factors direct subsequent morphogenesis in the regions where they are expressed. But how do localized transcription factors direct the events of morphogenesis? Neither the molecules nor the mechanisms involved are known. We have addressed this question by examining the mechanisms directing one important aspect of morphogenesis, namely embryonic cell proliferation.

In Drosophila, mechanisms of cell cycle control change as development progresses. The first 13 embryonic cell cycles are rapid and synchronous, and are driven by a maternally installed mechanism that does not require zygotic gene expression (Foe et al., 1993; Edgar et al., 1994). In cycle 14, decay of a maternal message that encodes the Cdc2 activator, String (Cdc25), leads to a transient arrest of the cell cycle in G₂ (Edgar and O’Farrell, 1989; Edgar et al., 1994). The next three cell cycles are controlled at the G₂/M transition by the availability of zygotically produced String (Edgar and O’Farrell, 1990). In cycle 17 many of the embryonic cells arrest in an extended G₁ (or G₀) phase (Edgar and O’Farrell, 1990) due to the down-regulation of a G₁ cyclin, cyclin E, which is required for DNA synthesis (Knoblich et al., 1994). Here, we focus on the three G₂-regulated cell cycles that occur between removal of maternal string function in cycle 14 and the introduction of this new rate-limiting step in cycle 17. These cell cycles progress according to an intricate spatiotemporal pattern that is essentially invariant and appears to be under genetic control (Fig. 1; Hartenstein and Campos-Ortega, 1985; Foe, 1989; Foe and Odell, 1989; Arora and Nusslein-Volhard, 1992). Three observations indicate that the pattern of these cycles is directed by modulated expression of string. First, loss-of-function string mutants are blocked in interphase 14 (Edgar and O’Farrell, 1989); second, ectopic string expression is sufficient to drive ectopic cell divisions during cycles 14, 15 and 16 (Edgar and O’Farrell, 1990); and third, string is normally expressed in a dynamic spatiotemporal pattern that is identical to the mitotic pattern, but precedes it (Edgar and O’Farrell, 1989; Fig. 2).

This experimental evidence is further supported by biochemical characterization of string and its influence on cell cycle regulators. string encodes a conserved Cdc25 type tyrosine phosphatase that triggers mitosis by dephosphorylating and activating the mitotic kinase, Cdc2 (Edgar and O’Farrell, 1989; Kumagai and Dunphy, 1991; Gautier et al., 1991; Edgar et al., 1994). Active Cdc2 phosphorylates many substrates in the cell, directly catalyzing mitotic events such as chromatin condensation, nuclear envelope breakdown and spindle formation (Nigg, 1993). Other factors required for cell proliferation, including Cdc2 itself, its cyclin cofactors and factors required for DNA replication, are expressed in excess in the embryo during cycles 14-16 and are not rate-limiting for cell cycle progression (Lehner and O’Farrell, 1989; 1990a,b; Stern et al., 1993; Richardson et al., 1993; Knoblich et al., 1994). Expression of cytoplasmic string RNA parallels the expression of nascent nuclear transcripts (O’Farrell et al., 1989) and string protein expression patterns likewise track the mRNA expression patterns very closely (Edgar et al., 1994; and unpublished). In addition, both string protein and mRNA are extremely unstable (T_G<15 minutes), and thus all available evidence indicates that the modulation of string activity is regulated largely at the level of transcription. Here, we show that cis-acting sequences that regulate string transcription serve as the interface between the pattern-formation gene network and cell cycle control. These sequences integrate numerous position-specific signals to generate the complex spatiotemporal program of string transcription and this in turn executes the pattern of embryonic cell divisions.

In situ hybridization was done with digoxigenin-labeled DNA probes (Boehringer Mannheim) essentially as described by Tautz and Pfeifle (1989) and in Ashburner (1989). Our major modification was that the template used for the string DNA probe was size-reduced to less than 200 bp to improve the signal and reduce background staining. Size reduction of a gel-purified 2.3 kb fragment of the string cDNA was accomplished by digestion with a cocktail of six restriction enzymes, and digoxigenin labeling by the random primer reaction was done at 22°C for 16 hours. string protein expression patterns were visualized using an affinity-purified rabbit polyclonal antibody (AP4; Edgar et al., 1994), preabsorbed goat anti-rabbit-biotin secondary antibodies (Jackson Labs) and streptavidin-HRP (Chemicon) according to standard protocols (Ashburner, 1989). In vivo labeling with BrdU was used to assay cell cycle patterns according to Bodmer et al. (1989) and Edgar and O’Farrell (1990).

Pattern-formation mutants were obtained in large part from the Mid-America Drosophila Stock Center. For most lines, mutant embryos were identified by characteristic defects in morphology and string expression. In some cases, lines carrying a lacZ marked balancer chromosome were stained for both string and lacZ RNA, and mutant embryos were identified by the absence of lacZ expression. string, cyclinA and cyclinA, cyclinB double-mutant embryos (stg^7B/stg^7B, cyclin A^neo114/cyclin A^neo114 and cyclin B^Df(59AB)/cyclin B^Df(59AB); cyclin A^neo114/cyclin A^neo114, respectively) were identified by their lack of mitotic figures and abnormally low cell densities for morphological stage. cyclin E-deficient embryos were generated by the cross: Df(cycE)GW1/CyO wg-lacZ × Df(cycE)GW3/CyO wg-lacZ. Trans-heterozygosity for these two deficiencies deletes only the cyclin E gene (Knoblich et al., 1994; R. Duronio, personal communication). Transheterozygotes were identified by the absence of wg-lacZ expression from the balancer. In addition, we analysed string expression in cycE^AR95/cycE^AR95 mutant embryos derived from l(2)Dd^AR95pr cn wx^wxtbw/CyO ftz-lacZ parents (see Knoblich et al., 1994). These homozygous embryos were identified by the absence of ftz-lacZ expression from the balancer, with similar results.

A 10.5 kb string transgene was isolated from a phage lambda clone (#G2A), cloned into the Carnegie 20 transformation vector and three independent lines with integrations on chromosome 2 were generated by standard P-element-mediated germline transformation (Spradling, 1986). Expression patterns and rescue of this fragment were assessed in embryos homozygous for a transcription-null allele at the stg locus (genotype P[(ry⁺)stg10.5]/+; stg^3A1/stg^3A1). The 31.3 kb transgene was isolated from a genomic cosmid library (NotBamNot-CoSpeR provided by J. W. Tamkun) and two independent transformant lines, one on chromosome 2 and one on chromosome 3, were isolated by P-mediated transformation. Expression patterns and rescue were assessed in the genotypes P[(w⁺)stg31.3]/+; stg^3A1/stg^3A1 and P[(w⁺)stg31.3] stg^3A1/P[(w⁺)stg31.3] stg^3A1. The 15.3 kb transgene was isolated as a SalI fragment from the 31.3 kb clone, inserted into CaSpeR and three independent lines on the X chromosome were generated. Expression patterns and rescue were assessed in the genotypes P[(w⁺)stg15.3]/P[(w⁺)stg15.3/Y; stg^AR2/stg^AR2.

We generated 225 w⁻ revertants by excision of P[w⁺ AA53] (Fig. 4). This insertion is homozygous viable and has only minor effects on string expression patterns, but is 90% lethal over null alleles of string (stg^3A1 and stg^7B). Eight stg⁻ mutants (stg^AR1-8) were then identified among these w⁻ revertants by their failure to complement the null allele stg^3A1. Lesions at the string locus in these mutants were analyzed by genomic Southern blotting using probes covering the P element and much of the string genomic region (−28.6 kb to +4.5 kb). Probes were generally the EcoRI fragments shown in Fig. 4 and genomic DNA was likewise restricted with EcoRI. To unambiguously map the deletions in stg^AR2 and stg^AR5, genomic DNA was extracted from about 0.5 ml of homozygous mutant embryos. These embryos were generated by collecting eggs from the cross stg^AR/+ × stg^AR/+, allowing the eggs to age for 36 hours and then collecting the unhatched embryos. Heterozygous and wild-type larvae (from hatched eggs) were effectively removed from the egg collection plates by placing yeast paste, into which larvae crawl, at the edge of the plate and replacing it three times. This procedure allowed the analysis of purely mutant DNA, greatly simplifying mapping of the deletions. Southern blots were done using digoxigenin-labeled DNA probes, Hybond-N membranes (Amersham) and Lumiphos chemiluminescent detection, according to the manufacturer (Boehringer Mannheim).

The embryonic expression pattern of string mRNA, assayed by in situ hybridization, is shown in Fig. 2. Although the pattern is too intricate to be described exhaustively here, several aspects are noteworthy. After the degradation of maternal string transcripts during early interphase 14 (Fig. 2A,B), expression occurs in a sequence of brief pulses that are timed differently in different regions of the embryo. The first wave of pulses precedes and matches the cycle 14 mitotic domains (MDs) mapped by Foe (1989; compare Figs 1 and 2C-G, M-Q). The order of string mRNA appearance generally corresponds to the order of mitoses, but shows deviations in some tissues. For instance, expression in MD10 (the mesoderm) precedes expression in MD2 (part of the head) by at least 5 minutes, even though mitosis in MD10 follows mitosis in MD 2 by 10 minutes (compare Fig. 2M and N). However, accumulation of transcripts proceeds more slowly in MD10 than in MD2, suggesting that, in addition to the time of initiation of string transcription, string RNA level is an important parameter in timing mitosis. The three postblastoderm cell cycles (14, 15 and 16) are regulated in three distinct, though related, spatial patterns. Consistent with its role in governing these cycles, string RNA accumulates prior to each division in patterns that anticipate the division patterns (Fig. 2).

During cycles 14 and 15 string transcription in most MDs ceases immediately following mitosis and does not resume until just prior to the next mitosis, 40-100 minutes later. In contrast, string mRNA is not rapidly down-regulated following mitosis 16. Instead, high levels of RNA persist in virtually all external tissues for about 2 hours after mitosis (stages 11-13; Fig. 2K,U,V). Expression is then extinguished in the epidermis (which has ceased proliferation) and is maintained only in proliferating cells of the peripheral and central nervous systems (Fig. 2L,W,X). string RNA expression in proliferating neuroblasts of the CNS would appear to be periodic, since not all of the neuroblasts express string at any given time. As neuroblasts cease proliferation they too extinguish string RNA expression (not shown).

The cycle 14 mitotic domains coincide precisely with features of the blastoderm fate map (Foe, 1989) and embryos with mutations in the pattern-formation genes that generate this fate map have characteristic defects in mitotic patterns (Foe and Odell, 1989; Rushlow and Arora, 1990; Arora and Nusslein-Volhard, 1992). Since mitotic patterns are executed by regulated transcription of string and since many of the pattern-formation genes encode position-specific transcription factors, we have proposed that string transcription is controlled, directly or indirectly, by pattern-formation gene products (Edgar and O’Farrell, 1989; O’Farrell et al., 1989).

To investigate this possible connection, we visualized string mRNA and protein patterns in a representative collection of pattern-formation mutants. Almost without exception, these mutations alter string expression in the regions where they affect cell fate (see Fig. 3 and Table 1 for summary data). In some mutants (such as twi, sna and btd), string expression is cleanly deleted in a specific domain that corresponds, spatially and temporally, to the normal expression of the mutant gene. This is consistent with direct regulation. In other mutants (such as bcd, hb and Kr), string expression patterns are not deleted, but are globally distorted. This suggests indirect, combinatorial, or concentration-dependent regulation. The earliest onset of abnormal expression in the mutant embryos is also informative. For instance, the pair-rule periodicity of string expression in MD11 (Fig. 2N; lateral epidermis) is not significantly affected in pair-rule mutants, but is altered in gap mutants (Fig. 3). Pair-rule mutants first show defects in string expression in late cycle 14, in the segmentally reiterated patterns of MDs 16, 17 and 21. Likewise, segment-polarity genes, homeotic genes and genes involved in neural patterning do not have significant effects until late cycle 14, in the intricately patterned ventral neurogenic region (MDs 16, 17, 21, 25, N and M). Such observations indicate that string is a sophisticated pattern integrator that responds to genes at all levels of the pattern-formation network.

To assess the size of the string gene, we isolated large fragments of the locus, introduced these into flies via P-element-mediated germline transformation, and tested their ability to rescue mutations of string. Fragments of approximate sizes 10.5 kb, 15.3 kb and 31.3 kb that included the complete 3.4 kb string transcription unit were tested (Fig. 4). Each of these fragments provided string function, as assessed by restoration of some cell division to string null mutants (stg^7B, stg^AR2 and stg^3A1). Nevertheless, none of these transgenes restored cell division to all mitotic domains, nor viability to string null mutants, and thus none appears to include all the essential regulatory sequences.

The patterns of string RNA expressed by these string transgenes were assessed following crosses to introduce each transgene into embryos homozygous for transcription-null alleles of stg (stg^3A1 or stg^AR2). The largest transgene (stg-31.3 kb) is expressed in the majority of the cycle 14 mitotic domains as well as many mitotic domains of cycles 15 and 16 (Figs 5, 6). At all stages, expression is correctly timed and positioned. Preceding mitosis 14, stg-31.3 kb is expressed in mitotic domains of the head (MDs 1, 2, 3, 8, 15, 20, 23, 24), the mesoderm (MD 10) and the ventral neurogenic region (MDs 16, 17, 21, N). Expression is clearly missing in other domains: the head (parts of MDs 5, 9), the lateral epidermis (MDs 6, 7, 11, 19), the tail (MDs 4, 12), the mesectoderm (MD 14) and about half of the ventral neuroectoderm (MDs 25, M). During cycle 15, stg-31.3 kb expression is confined to analogous domains (Figs 5, 6). In addition, a subset of lateral epidermal cells show expression in cycle 15 (Fig. 5C), as do many delaminated ventral neuroblasts. In cycle 16, stg-31.3 kb expression occurs in the tracheal placodes, many neuroblasts in the head and the ventral nerve cord, and in a few external cells of the head (Fig. 5D-F). Expression in the epidermis, which is strong and persistent from the endogenous gene after mitosis 16 (Fig. 2K,V), is not observed from stg-31.3 kb (Fig. 5F). During later embryogenesis, the 31.3 kb transgene continues to be expressed in many of the cells that normally continue dividing, namely neuroblasts in the peripheral nervous system, the ventral nerve cord and the brain (Fig. 5G). Analysis of String protein, mitotic figures and BrdU incorporation patterns in the transgenic, stg⁻ mutants confirmed that protein expression, mitoses and S phases occur in all regions exhibiting string mRNA expression (not shown).

Removal of 16 kb of 3′ sequences from the 31.3 kb transgene, producing the stg-15.3 kb transgene (Fig. 4), did not alter the expression patterns described above in any detectable way. While it remains possible that these 3′ sequences contain redundant transcriptional elements, or that there are regulatory sequences lying more than 16 kb 3′ of the gene, we suggest that few, if any, string regulatory sequences lie 3′ to +4.5kb (region E, Fig. 4).

In contrast, removal of 4.8 kb of 5′ sequences, producing the stg-10.5 kb transgene, greatly restricted expression. stg-10.5 kb expression is limited to cycles 14 and 15 and occurs only in the ventral neurogenic region (MDs 16, 17, 21, N; Fig. 6A). We detected no strong expression in any of the other MDs that are driven by stg-31.3 kb or stg-15.3 kb. Thus, we infer that the position-specific elements (PSEs) responsible for expression in these other MDs reside between −5 kb and −9.8 kb (region B, Fig. 4). We could not assess expression in the mesoderm (MD 10) due to interference from a mesodermal enhancer of a rosy gene that marked the stg-10.5 kb transgene (but see below).

P[w⁺ AA53] is a 12 kb transposon inserted 2.1 kb upstream of the string transcription start site (Fig. 4). It is homozygous viable and retains substantially normal spatial patterns of string expression, though it is semilethal in trans to string null alleles. Since P[w⁺ AA53] lies between many of the string PSEs and the basal string promoter, correct spacing of these PSEs relative to the promotor is evidently not crucial for their function.

As an alternative method for dissecting the string regulatory region, we generated deletion mutations in vivo by excision of the P[w⁺ AA53] transposon. One of these (stg^AR2) is deleted for the entire transposon and downstream sequences, including the string transcribed region. Not surprisingly, this mutant produces no string transcripts. A more pertinent mutation, stg^AR5, deletes all of P[w⁺ AA53] and sequences extending upstream to at least −28.6 kb. This deletion eliminates string expression in almost all tissues: cycle 14 expression is maintained only in the mesoderm (MD 10) and invaginated regions of the head (MDs 8, 15), which lie next to the mesoderm and form the anterior midgut (Fig. 7B,C). During later stages, we noted weak, uniform expression throughout the embryo and scattered high level expression in a few randomly located cells. In vivo labeling with BrdU confirmed that the mesoderm underwent two postblastoderm cell cycles in stg^AR5/stg^AR5 mutant embryos (Fig. 7D). In addition, scattered internal cells, perhaps mesodermal, incorporated BrdU at later stages, as did a few cells in the lateral epidermis. These latter cell cycles appeared to be driven by very low levels of string expression and may not reflect normal patterning. Since the 31.3 and 15.3 kb transgenes also exhibited mesodermal expression, we conclude that a mesoderm PSE is located between −2.1 kb and +4.5 kb (Fig. 4, region D).

In summary, our experiments define four string PSEs that drive transcription in distinct sets of cells. These regulatory regions are denoted A-D in Fig. 4 and their inferred functions are listed in Table 2. Region D contains a mesoderm PSE, region C contains an early acting ventral neuroectoderm PSE, region B contains a PSE that acts in a number of cell types in the head, the nervous system and the trachea, and region E appears to be beyond the boundary of the string locus. By elimination, we assign all unidentified PSEs to region A, upstream of tested sequences. Nevertheless, it is possible that we have missed PSEs within the tested region due to weak expression or disruption by breakpoints. We also note that, since we assessed expression in embryos that lacked cell proliferation in certain tissues, the late-stage expression patterns we saw may have had some abnormalities. However, the PSE locations we infer from these studies correlate well with the locations being mapped by an independent method, using string-lacZ reporter gene fusions (B. A. E. and D. A. L. unpublished).

In Drosophila, progress of the developmental program continues quite normally even after the cell cycle is blocked (see Hartenstein and Posakony, 1990; Gould et al., 1990). Since string transcription is evidently driven by developmental regulators, we expected that progression of the string transcription program would be cell cycle independent. Nevertheless, the coincidence of string transcriptional shut off with passage through mitoses 14 and 15 suggests that the cell cycle does have a role in orchestrating string expression patterns (see above). To test the influence of cell cycle progression on string transcription, we studied embryos arrested by different cell cycle mutations. Maternal supplies of the various cell cycle regulators decay with different kinetics and thus mutations in these regulators arrest the cell cycle at different developmen-tal stages. string mutants arrest in G₂ of cycle 14 (Edgar and O’Farrell, 1989), cyclin A, cyclin B double-mutants arrest in G₂ of cycle 15 (Knoblich and Lehner, 1993), cyclin A single mutants arrest in G₂ of cycle 16 (Lehner and O’Farrell, 1989) and cyclin E mutants arrest in G₁ of cycle 17 (Knoblich et al., 1994).

In embryos arrested in G₂ of cycle 14 by stg^7B, an EMS-induced allele that produces an inactive protein but has no deficit in RNA expression, string transcription commences in a normal sequence of mitotic domain patterns, but is not extinguished normally. Instead, high levels of string mRNA (both cytoplasmic and nuclear) persist continuously in most ectodermal cells during the interval that normally encompasses mitoses 14, 15 and 16 (Fig. 8A,B). This persistent expression could be extinguished by delivering active String from an inducible HSP70-string fusion gene and triggering mitosis (not shown, see Edgar and O’Farrell, 1990). Thus, string activity, or its consequence (Cdc2 activation and mitosis), contributes to the shut-off of string transcription in the ectoderm at the close of cycle 14.

While mitosis extinguishes string transcription in many of the cycle 14 expression domains, it is not uniformly essential. Many cells in the heads of arrested stg^7B embryos extinguish transcription normally, at the time corresponding to interphase of cycle 15 (Fig. 8A; arrow), and additional cells shut-off expression at the time of cycle 16 (Fig. 8B; arrows). Moreover, the inactivation of transcription during the interval corresponding to interphase 17 is normal in stg^7B embryos: inactivation starts in the dorsal epidermis, spreads to cells of the presumed PNS and finally encompasses the ventral nerve cord (Fig. 8C,D). At very late stages (after stage 16), stg^7B mutant embryos maintain expression in only a few cells of the rudimentary brain, just as do wild-type embryos (compare Fig. 2X to Fig. 8D). Thus the developmental programing of later transcriptional activations and inactivations continues even though the timing of the initial shut-off is disturbed.

Mitosis plays a minor role in extinguishing string transcription after cycle 14. In cycle 15-arrested, cyclin A, cyclin B mutants, we noted prolonged expression of cycle 15 string RNA patterns in some parts of the lateral epidermis and the ventral neurogenic region during stages 9-11. However, most cells shut off their cycle 15 expression patterns with near normal timing (Fig. 8E; arrows), and essentially all cells activate and inactivate cycle 16 expression patterns correctly (Fig. 8F,G). In embryos arrested in G₂ of cycle 16 by cyclin A mutations, the shut off of string expression after arrest is also essentially normal, as is continued expression in the brain and CNS (not shown). Likewise, embryos arrested in G₁ of cycle 17 by mutations in the cyclin E gene (Knoblich et al., 1994) show continued string expression after the cell cycle arrest. This is limited to neuroblasts of the peripheral nervous system, brain and ventral nerve cord, where proliferation continues after cycle 17 in wild-type embryos. We did note, however, that fewer neuroblasts express string in these late-stage cyclin E mutant embryos than in wild-type embryos (Fig. 8H). We conclude that, while mitosis or an associated S phase event may contribute to the abruptness of the shut-off of string expression, this effect is rather slight except in some of the cycle 14 mitotic domains. Importantly, there seems to be little influence of the cell cycle on the periodic activation of string transcription.

Since the isolation of a large collection of pattern mutants in the early 1980s (Jurgens et al., 1984; Nusslein-Volhard et al., 1984; Wieschaus et al., 1984), studies in Drosophila have revealed the mechanisms that specify positional information during early embryogenesis (see Lawrence, 1992; Bate and Martinez-Arias, 1993, for reviews). Despite this, our understanding of how positional information is translated into cell behaviors, and thus into morphogenesis, remains rudimentary. Here we address how embryonic positional information is used to orchestrate one fundamental aspect of morphogenesis, namely patterns of cell proliferation. During much of Drosophila embryogenesis cell proliferation is regulated by the expression of string, which encodes a conserved Cdc25-type phosphatase that activates Cdc2 (Edgar and O’Farrell, 1989; Millar and Russell 1992; Edgar et al., 1994). string expression is controlled at the transcriptional level and we show here that it is patterned by positional information supplied by a large set of genes that determine many aspects of cell fate (see also Foe and Odell, 1989; Rushlow and Arora, 1990; Arora and Nusslein-Volhard, 1992). Extensive (>15.3 kb) regulatory sequences of string integrate this positional information to generate the complex patterns of string transcription that execute the mitotic program.

The regulatory DNA of string contains separable position-specific elements (PSEs) that drive transcription in specified regions of the embryo at specified times (Fig. 4; Table 2). We have defined four such PSEs, but these are large, and in some cases multifunctional, and we expect that each one is a conglomeration of smaller, more specific PSEs. Like the control regions of several pattern-formation genes that have been studied in detail, string’s control region appears to be a patchwork of elements that can function independently and which, when summed, generate the overall expression pattern (see Small and Levine, 1991; Pankratz and Jäckle, 1993; for reviews).

Perturbations of string expression in mutant backgrounds indicate that string’s expression program depends on known patterning genes, many of which encode position-specific regulators of transcription (Fig. 3; Table 1). For example, twist and snail mutants fail to express string specifically in the mesoderm (MD10) and buttonhead mutants fail to express string specifically in a single stripe in the head (MD2). Since twist, snail and buttonhead encode transcription factors that are expressed specifically in MD10 (twi and sna) and MD2 (btd), they may be direct effectors of string transcription in these domains (see Thisse et al., 1987, 1988; Boulay et al., 1987; Wimmer et al., 1993). Our mapping indicates that the MD10 and MD2 PSEs reside in regions B and D of Fig. 4, and thus we might expect to find binding sites for these factors in these regions.

Although these examples suggest simple regulation, complex regulatory relationships are indicated as well. For example, string expression in MD11 (dorsal ectoderm) shows a transient pair-rule periodicity (Fig. 2N), but this pattern is not significantly affected by mutations in pair-rule genes. Rather, MD11 expression is affected by mutations in an earlier-acting set of genes, the gap genes (Fig. 3; Table 1). Thus string expression in MD11 is probably regulated by combinations of gap-gene products, in a concentration-dependent manner, in the same fashion that pair-rule genes are regulated (Small and Levine, 1991; Small et al., 1992; Pankratz and Jäckle, 1993). Expression of string in MD14 (the mesectoderm) provides another example of independent regulation of related patterns. Although string expression in MD14 is precisely coincident with expression of single-minded, a transcription factor involved in mesectoderm specification (Fig. 2E; Nambu et al., 1991), string expression is unaffected by single-minded mutations. We presume that string and single-minded are regulated independently, in parallel, by similar mechanisms. The PSE driving string expression in MD14 most likely responds, like single-minded, to combinations of broadly distributed dorsoventral pattern gene products to produce its highly restricted expression pattern (see Kasai et al., 1992).

Patterning genes that encode cell signaling molecules presumably influence string transcription by altering activity or expression of a transcription factor. For example, dpp, which encodes an TGF-β type signaling molecule, is known to regulate the spatial patterns of twist and zen, two transcription factors shown to alter string expression. Similarly, neurogenic genes such as Notch are likely to exert their influence on string by alterations in the activity of the helix-loop-helix proteins encoded by the achaete/scute complex.

Given the complexities of integrating positional information, it seems likely that different PSEs that share the same spatiotemporal response have a common evolutionary origin. Perhaps the string PSEs arose as regulators of other genes and were spliced into string piecemeal during evolution. In this regard, it is interesting to consider that string homologues appear to regulate G₂/M transitions in eukaryotic cells ranging from yeast to humans and where it has been studied, this regulation occurs at the transcriptional level (Moreno et al., 1990; Sadhu et al., 1990; Kakizuka et al., 1992). Thus, like its catalytic function, string’s mode of responding to positional information through complex transcriptional control may be evolutionarily conserved.

Studies of the cell cycle in cleavage-stage frog and marine invertebrate embryos led to the suggestion that cell cycle timing is governed by an autonomous cell cycle oscillator. While this may be an accurate characterization of the relentless progress of some early embryonic divisions, ultimately, embryogenesis requires the coordination of cell proliferation with other aspects of morphogenesis (see Edgar and O’Farrell, 1990). Although this coordination might be achieved through modulation of the rate of an autonomous oscillator, the Drosophila embryo appears to follow a different alternative. Its cell cycle oscillator is interrupted by loss of one oscillator component, String, after mitosis 13 (Edgar et al., 1994). Subsequently, developmental regulators determine the timing of string transcription and thus control progression of the cycle until removal of a second oscillator component, cyclin E, leads to a G₁ arrest in cycle 17, or later (Knoblich et al., 1994).

The constellation of stage-and position-specific transcription factors that regulate string control multiple aspects of cell fate and are not specialized excusively for control of the cell cycle. For this reason, we were not surprised to find that string transcription continues to be periodically activated and inactivated according to normal spatial patterns even after the embryonic cell cycle is arrested. The independence of string’s transcription program from cell cycle progression is most clearly documented by our finding that arrest in cycles 15, 16 or 17, (achieved by the various cyclin mutations) causes little perturbation in the dynamics of string transcription following arrest (Fig. 8). Thus, like a number of DNA synthesis genes expressed at the G₁ to S transition, string is not actually a ‘cell cycle-regulated’ gene in vivo (see Knoblich et al., 1994; Duronio and O’Farrell, 1994).

Although developmental cues are evidently the primary regulators of string expression, we did find that cell cycle progression plays a role in the shut-off of transcription in some cell types at some stages. This influence is most significant in cycle 14. If embryos are arrested in cycle 14 (by a string mutation), a major subset of ectodermal cells fails to deactivate string transcription on schedule (Fig. 8). However, in other regions of the embryo and, at other stages, the timely inactivation of string transcription does not require passage through mitosis. This complexity might be attributed to variation in the stability of the different factors that presumably activate string transcription in different cells at different stages. The transcription complexes that activate string in the ectoderm in cycle 14 may require mitotic phosphorylation or DNA replication to be disrupted and inactivated on schedule. Perhaps because of this, the reprograming of string transcription in the cycle 14 ectoderm appears to require mitosis. In contrast, later reprogramming events may involve distinct transcriptional complexes that are less stable and do not require passage through the cell cycle for their inactivation. Such complexes might be inactivated by newly expressed repressors, or simply by the cessation of expression of some of their components.

Finally, we would like to offer a perspective somewhat different than that emphasized in many investigations of the molecular basis of developmental patterning. We suggest that the distinctions between different body structures such as wings and legs are largely organizational, and are not likely to be understood in terms of the induction of distinct tissuespecific gene products. Studies of string may provide a paradigm for another type of developmental control, in which the upstream regulatory regions of a variety of ‘housekeeping’ genes control the spatial and temporal programming of cellular events (such as cell division, cell adhesion or cytoskeletal changes) that have profound, specific effects on tissue organization and structure (see Costa et al., 1994).

We would like to thank Rick Terle, Briony Patterson and Robert Saint for freely sharing their maps of the string locus and the string genomic clone DRL13. Louisa Wu and Charles Zucker provided the P[w⁺ AA53] mutant, John Tamkun provided his genomic cosmid library, Christian Lehner provided unpublished data and the cyclin B and cycE^AR95 mutants and Bob Duronio provided the cyclin E deficiencies. Doris McLean, Luke Alphey and David Glover supplied a useful DNA sequence of the string promotor. We also thank Phil Ingham, David Ish-Horowicz and Paul Nurse for hosting B. A. E. in their laboratories, where some of this work was done. B. A. E thanks the Lucille P. Markey Charitable Trust for support. The effort in Pat O’Farrell’s laboratory was supported by NIH RO1 GM37193.