The Drosophila embryo is well-known for its simultaneous mode of segmentation, in which a segmental pattern is laid down at the end of the blastoderm stage, prior to significant morphogenetic movements. Fourteen prospective parasegment boundaries appear at this stage, marked by segmental stripes of segment-polarity gene expression (DiNardo et al., 1985; Baumgartner et al., 1987; Baker, 1988; Lee et al., 1992; Grossniklaus et al., 1992).

Sandwiched in between parasegment boundary 14 (PSB14; see Box 1) and the broad posterior domain of wg (thought to correspond to prospective hindgut; Baker, 1988) are about four cell rows of ectoderm that remain unpatterned by segment-polarity genes at the end of the blastoderm stage (Figure 1A, stage 6). This ‘tail’ region (see Box 1) goes on to form the most terminal structures of the larva (Turner and Mahowald, 1979; Jürgens, 1987), including a 15th parasegment boundary (Kuhn et al., 1995; Kuhn et al., 2000), various sensory organs (Sato and Denell, 1986; Jürgens, 1987; Kuhn et al., 1992), and the anal pads (external organs involved in ion transport; Jarial, 1987).

The segmental nature of the tail is unclear. The tissue just posterior to PSB15 is abdominal segment 10 (A10; Figure 1C). Some authors consider the region to contain a cryptic 11th abdominal segment as well (Jürgens, 1987; Baumgartner et al., 1987), but most do not (see Discussion: ‘The segmental character of the Drosophila tail’) and, to the best of our knowledge, a 16th parasegment boundary has not been described. To investigate this issue, we used multiplexed HCR ISH to re-examine the expression of the parasegment boundary markers wingless (wg; Baker, 1987; Rijsewijk et al., 1987), engrailed (en; Kornberg et al., 1985; Fjose et al., 1985), sloppy-paired (slp; Grossniklaus et al., 1992), and even-skipped (eve; Macdonald et al., 1986) during germband extension and extended germband stages (Figure 1; Figure 1—figure supplement 1).

Given that Drosophila shows distinct segmentation dynamics in the trunk and the tail, we examined the expression of the timer genes (cad, D, and opa) in these regions during blastoderm stages and early germband extension (for an earlier survey using an inferior in situ hybridisation method, see Clark and Peel, 2018). To account for the movement of nuclei/cells during blastoderm (Keränen et al., 2006) and gastrulation stages, we co-stained the timer genes with wg and used the posterior wg domain as a fiducial marker. (The posterior wg domain appears to be stable relative to nuclei, as nuclear transcription foci are not offset anteriorly or posteriorly relative to cytoplasmic transcripts.) To aid with fine-scale staging of embryos, we have divided stage 5, which lasts ∼40 min at 25° C, into five timeclasses based on gene expression and morphology (see Appendix 1).

The relative spatiotemporal expression dynamics of the timer genes are suggestive of cross-regulation. To investigate this possibility, we examined timer gene expression in opa^-, D^-, and cad^- mutants (Figure 4; Figure 4—figure supplement 1) and discovered a variety of cross-regulatory effects. As cad is expressed maternally as well as zygotically, we examined cad maternal mutants (cad^m-z+) and cad zygotic mutants (cad^m+z-) in addition to cad null mutants (cad^m-z-) in order to disentangle maternal and zygotic effects (Figure 4—figure supplement 3). We also examined timer gene expression in wg^- mutants, but did not observe any aberrant expression in these embryos during our stages of interest (Figure 4—figure supplement 4).

We next wanted to understand why timer gene expression differs between the trunk, tail, and prospective gut regions; i.e., how the timer gene network is spatially regulated. We therefore examined how timer gene expression relates to the expression domains of the zygotic terminal system genes tll (Jurgens et al., 1984; Strecker et al., 1986; Pignoni et al., 1990) and huckebein (hkb; Weigel et al., 1990; Brönner and Jäckle, 1991), the obvious candidates for providing this spatial information.

To determine whether Hkb and Tll spatially regulate the timer genes, we investigated timer gene expression in hkb^- mutants, tll^- mutants, and torso (tor^-) mutants (Figure 6). Tor (Klingler et al., 1988; Sprenger et al., 1989; Casanova and Struhl, 1989) is a maternally provided receptor necessary for transducing the extracellular signal-regulated kinase (ERK) signal that specifies the poles of the embryo (reviewed in Duffy and Perrimon, 1994; Li, 2005; Goyal et al., 2018), and therefore tor^- mutants express neither hkb nor tll (Brönner and Jäckle, 1991; Pignoni et al., 1992).

Having found that Tll is necessary for patterning both the tail region and the posterior wg domain (prospective hindgut), we next asked how these regions are distinguished from each other. Forkhead (Fkh) is a zygotic transcription factor that is expressed in the posterior of the embryo from stage 4.4 (nuclear cycle 13) downstream of Tor (Weigel et al., 1989; Weigel et al., 1990) and is required for the specification of hindgut identity (Jürgens and Weigel, 1988; Weigel et al., 1989; Kuhn et al., 1995; Hoch and Pankratz, 1996).

From looking at how gene expression is affected in various mutant genotypes, we have inferred a network of regulatory interactions between the timer genes and the posterior terminal genes (Figure 8A; Appendix 3—table 1). Most (11/18) of these proposed interactions originate from this study, although we also find support for previously proposed interactions related to the patterning of tll, hkb, fkh, and wg (Figure 8B). (For a recent quantitative model of posterior gut specification using a network similar to Figure 8B, see Keenan et al., 2022.)

We now formalise the regulatory network in Figure 8A as a logical model, and see whether it reproduces the patterning dynamics that we observed in the embryo. For the purposes of this study, we are aiming for a minimal, qualitative explanation of timer gene patterning, commensurable with the essentially qualitative developmental genetic paradigm we have been working within. We are interested in the relative ordering of gene expression domains in time and space, abstracted away from specific domain sizes, expression levels or expression kinetics. To the extent that the model is able to recapitulate the essential features of both wild-type and mutant genotypes, our confidence in the network will be increased.

The modelling framework we have chosen is very simple (for a full description, see Appendix 4). Briefly, Hkb and Tll are assumed to be extrinsic inputs to the system (we ignore the cross-regulation of tll by Hkb), and we model how Fkh, Wg, Cad, D, and Opa are expressed in response. Each of these seven factors is modelled as a logical variable, some of which (Hkb, Tll, D, Opa) may take one of three levels of expression (off/weak/strong), while the others (Fkh, Wg, Cad) may take only two (off/on). The AP axis is modelled as four discrete regions, 1–4 (corresponding to trunk, tail, hindgut, and posterior midgut, respectively), which differ in their hard-coded Hkb and Tll inputs over time. (Note that we do not include any dorsoventral input to the system, nor attempt to model the D neuroectodermal domain.) Each simulation consists of four time points, t0–t3 (corresponding to nuclear cycle 13, early stage 5, mid stage 5, and stage 6, respectively). At t0, Cad is on in all regions, and the other output factors are off. Expression at subsequent timepoints is computed from expression at t(n − 1), according to factor-specific logical rules (which remain the same for all timepoints). Mutants are simulated by keeping the relevant factor(s) turned off for all timepoints.

We simulated the patterning model for the wild-type condition (Figure 8C) and eight mutant genotypes examined in this study (fkh^-, cad^m-z-, D^-, opa^-, tor^-, hkb^-, tll^-, and tll^- opa^-; Figure 8D–K). A genotype-by-genotype explanation of the simulated expression dynamics is provided in Appendix 4, along with a table cross-referencing the simulated expression data with the corresponding observations from real embryos (Appendix 4—table 1). Allowing for the simple, qualitative nature of the model, the simulations were remarkably accurate at recapitulating the patterning dynamics of each genotype.

This work provides evidence for a set of cross-regulatory interactions between cad, D, and opa that helps generate dynamic, sequential expression. In particular, we find that Cad activates D (i.e., promotes the expression of the next gene in the sequence), while D represses cad and Opa represses cad and D (i.e., both inhibit the previous gene(s) in the sequence). opa is not cross-regulated, however, making it an ‘input-only’ component of the three gene network (at least in the blastoderm context).

Timer gene expression is also shaped by extrinsic spatiotemporal regulation. In this work, we show how the timer gene network interacts with the posterior terminal system: most notably, Tll differentially represses cad, D and opa in the tail region, indirectly allowing cad expression to be maintained. The localised inputs from the posterior terminal system are overlaid on global temporal regulation provided by the nuclear:cytoplasmic ratio (which is particularly important for regulating the onset of opa transcription; Lu et al., 2009) as well as the levels of maternal factors such as Tramtrack (Harrison and Travers, 1990; Brown et al., 1991; Read et al., 1992), Zelda (Liang et al., 2008; Harrison et al., 2011; Nien et al., 2011; McDaniel et al., 2019), Stat92e (Yan et al., 1996; Hou et al., 1996; Tsurumi et al., 2011), and GAGA Factor/Trithorax-like (Farkas et al., 1994; Bhat et al., 1996; Moshe and Kaplan, 2017; Gaskill et al., 2021). Ironically, precisely because these maternal factors are so crucial to development, their patterning roles remain less well understood than those of the zygotic patterning genes, which are less pleiotropic and therefore easier to study.

Recent work in the Drosophila blastoderm has demonstrated the extensive effects of timer genes on developmental gene expression. Opa has been shown to act as a pioneer factor, reshaping gene expression genome-wide by opening chromatin at hundreds of target enhancers (Soluri et al., 2020; Koromila et al., 2020). Cad and D are also known to regulate expression across the genome (Li et al., 2008; MacArthur et al., 2009; Aleksic et al., 2013). Here, we have found that early Cad expression appears to be necessary for the correct timing of later developmental events because neuroectodermal gene expression turns on precociously in cad^m-z- embryos. The vertebrate Cad ortholog Cdx4 has also been shown to temporally regulate neural differentiation, in the developing spinal cord (Joshi et al., 2019), a tissue in which D and Opa orthologs play key developmental roles (reviewed in Graham et al., 2003; Merzdorf, 2007; Houtmeyers et al., 2013; Stevanovic et al., 2021). More generally, comparative evidence suggests that Cad/Cdx plays a deeply conserved role in the formation of the posterior body and the patterning of the posterior gut (Copf et al., 2004; Wu and Lengyel, 1998; van Rooijen et al., 2012; Zhong et al., 2020). In this context, Drosophila cad^m-z- mutants offer a rare opportunity to study the genome-wide effects of a total loss of Cad/Cdx function without also catastrophically perturbing early developmental events.

In this work, we have investigated blastoderm gene expression downstream of the posterior terminal system, revisiting a patterning network that was most intensely studied in the late 1980s and early 1990s (Strecker et al., 1986; Mahoney and Lengyel, 1987; Mlodzik and Gehring, 1987b; Strecker et al., 1988; Jürgens and Weigel, 1988; Weigel et al., 1990; Casanova, 1990; Brönner and Jäckle, 1991; Wu and Lengyel, 1998). The modern availability of marked balancers and multiplexed imaging techniques has allowed us to clarify the topology and spatiotemporal dynamics of the network, and incorporate genes (D and opa) that had not been cloned at the time most of the original work was completed. All told, we have identified 11 new regulatory interactions involved in Drosophila AP patterning, put forward the first formalised model (to our knowledge) for the patterning of the tail, and provided a solid foundation for future quantitative analyses of this system.

Although simple, our model provides new insights into how the tail and hindgut regions are specified in the early embryo. Both regions, along with segment A8, have long been known to depend on Tll expression (Strecker et al., 1986; Diaz et al., 1996). tll alleles can be arranged into a coherent phenotypic series in which the most posterior structures within the Tll-dependent region are the most sensitive to tll perturbation and the most anterior structures are the least (Strecker et al., 1986; Diaz et al., 1996), suggesting that this part of the blastoderm fate map is patterned by a gradient of Tll activity (Casanova, 1990). However, it has not been clear at the network level how graded Tll activity would be transduced into a specific series of boundaries and domains.

We found that tll expression was strong and persistent within the hindgut region, but weaker and transient in the tail region, with the anterior border of the expression domain retracting posteriorly across nuclei over time. We additionally found that Tll effectively patterned both the anterior and posterior boundaries of the tail region by differentially repressing D and opa relative to cad. Crucially, D and opa were repressed even where Tll expression was transient and weak, but cad was not repressed (and fkh was not activated) unless Tll expression was stronger, helping explain the transition from tail fate to hindgut fate as Tll levels increase. Furthermore, the retraction of the Tll domain over time explains the posterior shifting dynamics we found for the timer genes within the tail region, which contrasts with the anterior shifting dynamics previously described for the pair-rule and gap genes (Jaeger et al., 2004; Keränen et al., 2006; Surkova et al., 2008; Lim et al., 2018).

We also discovered, to our surprise, that there is a concerted posterior fate map shift in hkb^- embryos, apparently mediated by a reduction in the size of the tll domain. (A subtle anterior fate map shift additionally occurs in D^- embryos, which might also be mediated by Tll.) Although further research is necessary to determine the mechanism by which Hkb cross-regulates tll, the phenotype implies that the size of the tll domain is not an unmediated response to terminal signalling. (Indeed, there are hints in the existing literature that tll and hkb may be zygotically cross-regulated by other AP patterning genes as well; see Casanova et al., 1994; Greenwood and Struhl, 1997; de las Heras and Casanova, 2006.) These findings may complicate the interpretation of recent studies that have characterised the input:output relationships between terminal signalling and tll and hkb expression using optogenetics (Johnson and Toettcher, 2019; Johnson et al., 2020; Keenan et al., 2020).

The ancestral insect body plan has 11 true abdominal segments plus the periproct/telson, but this number has been reduced in many extant insect lineages (Snodgrass, 1935; Demerec, 1950; Matsuda, 1976; Chapman et al., 2013). In Drosophila, the most common view has been that the embryo makes 10 abdominal segments (i.e., 15 parasegment boundaries), with the anal pads located in PS15/A10 (Turner and Mahowald, 1979; DiNardo et al., 1985; Sato and Denell, 1986; Perkins and Perrimon, 1991; Kuhn et al., 1992; Schmidt-Ott et al., 1994). In particular, territories corresponding to A8, A9, and A10 are visible at the morphological level during embryogenesis (Turner and Mahowald, 1979), and surveys of en, wg, hh, and slp staining have found evidence for (at most) 15 parasegment boundaries (DiNardo et al., 1985; Baker, 1987; Baker, 1988; Kuhn et al., 1992; Grossniklaus et al., 1992; Mohler and Vani, 1992; Tabata et al., 1992; Lee et al., 1992; Tashiro et al., 1993; Kuhn et al., 1995). However, fate mapping experiments (Jürgens, 1987) and surveys of gooseberry expression (Baumgartner et al., 1987; Gutjahr et al., 1993) have suggested that the embryo makes 16 parasegment boundaries, with the anal pads located in PS16/A11. There is also some evidence for A11 from patterns of gene expression in adult genital discs (Freeland and Kuhn, 1996).

Given the small size of the tail region within the embryo, the fact that it is covered by amnioserosa during key stages of patterning, and the fact that it later undergoes complicated morphogenetic rearrangements and fusions that obscure its metameric nature, it is perhaps unsurprising that the number of Drosophila segments has not been unambiguously resolved. In this study, we present evidence for a vestigial 16th parasegment boundary in the embryo by identifying additional domains of slp and wg expression and reinterpreting previously described domains of eve and en. These observations suggest that the anal pads are located in PS16. (Whether the tissue between PSB16 and the anus should be classified as a true 11th abdominal segment or a non-segmental periproct/telson is beyond the scope of this article.) However, PSB16 appears extremely dorsoventrally restricted and may have little functional significance in the organism. As the number of abdominal segments varies across insects (Matsuda, 1976), the mechanistic basis of this evolutionary reduction would be interesting to study within a comparative developmental framework.

Our findings suggest that the Drosophila embryo sequentially patterns two parasegment boundaries after gastrulation, and that in both cases the new boundary is patterned by abutting stripes of slp and eve. In PS15 and PS16, the relative arrangement of slp, eve, wg, and en expressing cells is the same conserved pattern that is found at parasegment boundaries in the Drosophila trunk and throughout the arthropod phylum (reviewed in Clark et al., 2019). However, tail segmentation differs from trunk segmentation in that resolved, stable eve stripes emerge de novo and with single-segmental periodicity, rather than from a dynamic and double-segmental phase of pair-rule gene expression.

Intriguingly, a remarkably similar switch from double-segment to single-segment periodicity occurs towards the end of segmentation in the centipede Strigamia maritima, where stable, resolved eve stripes start appearing de novo in the anterior segmentation zone instead of emerging from posterior oscillatory expression (Brena and Akam, 2013). A possible switch from double-segmental to single-segmental patterning has also been reported for terminal segments in the beetle Tribolium (Janssen, 2014). These observations hint that terminal and trunk segments may be homonomous at the level of segment-polarity gene expression but derived from distinct ontogenetic programs. More work is needed to determine how such a developmental switch—if present—is controlled, as well as its relationship to the more general problem of terminating axial development.

We end this study by assessing the relevance of our findings from Drosophila to the development of other insect species. Which aspects of the Drosophila network are likely to be conserved in other insect species that have been used to study segmentation, such as Tribolium, Nasonia vitripennis, and Oncopeltus fasciatus? And how might the Drosophila network differ from that of its sequentially segmenting ancestors?

The cross-regulatory interactions that we found between the timer genes might be quite widely conserved in insect segmentation. Activation of D by Cad, repression of cad by Opa, and repression of D by Opa are all consistent with a segment addition zone that is subdivided into a posterior region that expresses Cad and D and an anterior region that expresses Opa, as seen, for example, in Tribolium (Clark and Peel, 2018). However, repression of cad by D would need to be reconciled with the sustained expression of both cad and D in the posterior segment addition zone. Intriguingly, some of the timer gene cross-regulatory interactions may even be important for regulating expression dynamics in completely different developmental contexts, given that Opa has recently been found to repress D during the temporal patterning of Drosophila intermediate neural progenitors (Abdusselamoglu et al., 2019).

The different components of the Drosophila terminal system seem to have acquired their posterior patterning roles at different times: posterior tll expression is found across diverse holometabolan species (Schroder et al., 2000; Lynch et al., 2006; Wilson and Dearden, 2009; García-Solache et al., 2010; Lemke et al., 2010; Klomp et al., 2015) although not in hemipterans (Weisbrod et al., 2013; Bickel et al., 2013), whereas hkb and tor appear to have been recruited to terminal patterning roles more recently (García-Solache et al., 2010; Kittelmann et al., 2013; Duncan et al., 2013). In Tribolium, tll is expressed downstream of tor (as in Drosophila), and tor RNAi embryos fail to express cad and wg in the posterior of the embryo, resulting in AP truncation (Schoppmeier and Schröder, 2005). In Nasonia, tll RNAi results in a reduction of posterior cad, as well as in gap gene misregulation that disrupts much of abdominal segmentation (Lynch et al., 2006). It will be instructive to test whether these losses of cad expression in Tribolium and Nasonia are mediated by ectopic expression of Opa, as we found for tll^- and tor^- mutants in Drosophila. If so, it would suggest that the initial spatial regulation of the timer gene network by Tll in the posterior blastoderm might be conserved across holometabolan embryos, despite their varying modes of development.

So, how does timer gene regulation differ between sequentially segmenting embryos (which establish a persistent segment addition zone) and simultaneously segmenting embryos like Drosophila? One key difference is likely to be the role of a posterior Wnt signalling centre: there is evidence from many different sequentially segmenting species that Wnt signalling is important for activating cad expression and maintaining the segment addition zone (reviewed in Clark et al., 2019), whereas we found that timer gene expression was unaffected in Drosophila wg^- mutants, at least during our stages of interest. In addition, it seems probable that timer gene cross-regulation of opa is important in sequentially segmenting species, with this having been lost from the Drosophila lineage during the evolution of simultaneous patterning.

If we modify the Drosophila timer gene network to incorporate these additional features (Appendix 4), we can see how appropriate segment addition zone dynamics might naturally emerge (Figure 8—figure supplement 1). It therefore seems plausible that the cross-regulatory interactions between the Drosophila timer genes may represent an evolutionary vestige of a ‘dynamical module’ that was originally involved in axial elongation (Clark and Peel, 2018; Clark, 2021). Functional experiments in sequentially segmenting species will be necessary to test this hypothesis.

Stock maintenance and embryo fixation (20 min with 4% formaldehyde in PBS) was performed as described in Sullivan et al., 2000. ‘Wild-type’ flies were Oregon-R. The mutant alleles used were wg^l-8 (Bloomington #5351), cad³ (gift from H. Skaer), cad² (Bloomington #7091), D^r72 (gift from S. Russell), opa⁸ (Bloomington #5340), tor^XR1 (gift from T. Johnson), hkb^A321R1 (Bloomington #2059), Df(3R)Exel6217 (Bloomington #7695, a deficiency covering the tll locus), and fkh⁶ (gift from K. Roeper). Mutant lines obtained from the Bloomington Drosophila Stock Centre were verified by cuticle preparations as described in Sullivan et al., 2000. The tll^- opa^- double mutant was generated by the Cambridge Fly Facility by recombining Df(3R)Exel6217 and opa⁸. Mutants were balanced over marked balancer chromosomes expressing lacZ during early embryogenesis: CyO, hb-lacZ (Bloomington #6650) for the second chromosome and TM6C, twi-lacZ Sb¹Tb¹ (Bloomington #7251) or TM3, hb-lacZ Sb¹ (gift from S. Russell) for the third.

cad^- germline clones were generated using the heatshock induced FLP/FRT system as described in Selva and Stronach, 2007. Briefly, eight vials of 30 cad² FRT40A/CyO virgin females (Bloomington #7091) were each crossed with 10 hsFLP w; ovoD1 FRT40A/CyO males (constructed by crossing Bloomington #6 hsFLP w; Adv/CyO females with Bloomington #2121 ovoD1 FRT40A/CyO, but note that #2121 is no longer listed in Bloomington). Adults were flipped to new vials every 2 days, resulting in a total of ∼100 vials. When crawling L3 larvae were visible, vials were heatshocked at 37°C in a waterbath for 1 hr, allowed to recover at 25°C for 24 hr, then heatshocked again at 37°C for 1 hr. Approximately 600 non-CyO virgin females (some presumably with cad²/cad² ovaries) were collected from the heatshocked vials and crossed with ∼300 cad³/CyO, hb-lacZ males. Resulting embryos without lacZ expression lacked both maternal and zygotic cad (cad^m-z-), while embryos with lacZ expression were paternal rescues (cad^m-z+). Zygotic cad mutants (cad^m+z-) were offspring from cad³/CyO, hb-lacZ parents that lacked lacZ expression; note that this genotype is also heterozygous for maternal cad.

Clone FI01113 containing opa coding sequence was obtained from the Drosophila Genomics Resource Center. Gateway attB primers were designed to express 386 amino acids from the N-terminus of Opa (amino acids 3–389), spanning the zinc finger region in the centre of the protein. The forward primer included a Shine-Dalgarno sequence; the reverse primer was designed to be in-frame with the C-terminal fusion of the Gateway expression vector pET-DEST42 (Thermo Fisher Scientific). A two-stage PCR procedure was used to obtain a final amplicon carrying the attB-sequences at each end of the N-terminal opa sequence.

Primers for the first amplification were

paDM-F: AAAAAGCAGGCTTCGAAGGAGATAGAACCATGAACGCCTTCATTGAGC

paA-R: AGAAAGCTGGGTTGTCGTAGCCGTGGGATG

Overlapping primers for the second amplification to complete the attB regions were

attB1adap-F: GGGGACAAGTTTGTACAAAAAAGCAGGCT

attB2adap-R: GGGGACCACTTTGTACAAGAAAGCTGGGT

The attB-opa amplicon was obtained by PCR with Phusion proofreading polymerase (Thermo Fisher Scientific) using primers opaDM-F and opaA-R. This first amplicon was diluted 1000-fold, then Phusion PCR was repeated with primers attB1adap-F and attB2adap-R. This attB-opa amplicon was recombined into Gateway donor vector pDONR (Thermo Fisher Scientific) using the BP Clonase II kit (Thermo Fisher Scientific). Plasmid DNA from a sequence-verified clone was then recombined into pET-DEST42 using the LR Clonase II kit (Thermo Fisher Scientific).

For expression of the fusion protein, plasmid DNA was transformed into One Shot BL21 Star (DE3) chemically competent Escherichia coli (Thermo Fisher Scientific). Opa protein was expressed in two ways, firstly by IPTG induction of exponentially growing cells (0.75 mM IPTG for 2.75 hr), secondly by overnight culture in TB Overnight Express (Novagen). The Opa fusion protein in pET-DEST42 had a C-terminal 6-His tag. Protein was purified from bacterial pellets, each from 100 ml of cells induced in IPTG or TB Overnight Express. Purification was carried out using Ni-NTA Agarose (QIAGEN), under 8 M urea denaturing conditions according to the manufacturer’s protocol. Purified protein was dialysed against water, then concentrated using an Amicon Ultra-Ultracel 5 kDa centrifugal filter (Millipore). Antibodies were raised in two guinea pigs by Eurogentec. Aliquots are available from EC on request.

Prior to staining, fixed embryos stored in methanol were put through a rehydration series of 5 min each at 75, 50, and 25% methanol in PBS + 0.1% Tween-20, then washed three times with PBS + 0.1% Tween-20.

HCR in situ hybridisation was performed using probes and hairpins produced by Molecular Instruments, following the protocol for whole-mount fruit fly embryos included in Choi et al., 2016, adapted for v3.0 probes as described in Choi et al., 2018, with the following changes. Treatment of fixed embryos with ethanol, xylene, and proteinase K was omitted. The percentage of dextran sulphate in the probe hybridisation and amplification buffers was reduced from 10% w/v to 5% w/v, to reduce viscosity and allow the embryos to settle more easily in the tube. A 20 min postfix step (4% formaldehyde in 5× SSC + 0.1% Tween-20) was added at the end of the protocol to stabilise the signal.

For antibody staining following HCR, embryos were incubated for 30 min in blocking solution (5% Normal Goat Serum [Vector Laboratories] in 5× SSC + 0.1% Triton X-100), at room temperature with rocking. Embryos were then incubated overnight in preabsorbed primary antibody diluted in blocking solution, at 4°C with rocking. Embryos were washed four times for 15 min in 5× SSC + 0.1% Triton X-100, at room temperature with rocking, then incubated for 30 min in blocking solution, at room temperature with rocking. Embryos were then incubated for 2 hr with fluorescently labelled secondary antibody diluted in blocking solution at room temperature with rocking. Embryos were washed four times for 15 min then one time for 30 min with 5× SSC + 0.1% Triton X-100 at room temperature with rocking. Antibody staining without prior HCR was performed as above with the exception that PBS was used instead of 5× SSC. Primary antibodies were guinea pig anti-Opa (this work) at 1:5000, rabbit anti-Dichaete (Soriano and Russell, 1998) at 1:10, rabbit anti-Tll (Kosman et al., 1998) at 1:100, and rat anti-Hkb (Ashyraliyev et al., 2009) at 1:100. Secondary antibodies were goat anti-guinea pig Alexa Fluor 647 (Invitrogen A-21450), goat anti-rabbit Alexa Fluor 488 (Invitrogen A-11034), goat anti-rabbit Alexa Fluor 555 (Invitrogen A-21429), and goat anti-rat Alexa Fluor 488 (Invitrogen A-11006), diluted 1:1 with 100% glycerol for storage and used at 1:500 (1:1000 overall).

Following HCR and/or antibody staining, embryos were incubated for 30 min with 1 ng/μL DAPI (Thermo Fisher Scientific) in 5× SSC + 0.1% Tween-20, at room temperature with rocking, then washed three times for 30 min in 5× SSC + 0.1% Tween-20, at room temperature with rocking. Prior to mounting, embryos were stored in 1.5 mL tubes in SlowFade Gold Antifade Mountant (Thermo Fisher Scientific).

Embryos were mounted in SlowFade Gold Antifade Mountant (Thermo Fisher) on glass microscope slides (Thermo Scientific) with #1.5 coverslips (Corning). #1.5 coverslips were used as bridges to prevent embryos from being squashed. Clear nail varnish was used to seal the edges of the slide.

Microscopy was performed on an Olympus FV3000 confocal microscope at the Department of Zoology Imaging Facility (University of Cambridge). Acquired images were 12-bit, with a 1024 × 768 scan format and a 2 μs/pixel dwell time. Whole embryo images were acquired using an Olympus UPlanSApo 30 ×1.05 NA silicon immersion oil objective, a physical pixel size of 0.47 μm × 0.47 μm, and a z-stack step size of 1.5 μm. The close-ups in Figure 1 and Figure 3 were acquired using an Olympus UPlanSApo 60 × 1.3 NA silicon immersion oil objective, a physical pixel size of 0.21 μm × 0.21 μm, and a z-stack step size of 0.8 μm. Each z-stack was specified so as to span from just above the top surface of the focal embryo through to the middle of its yolk.

In each experiment, embryos had been stained for up to four transcripts and/or proteins of interest plus nuclei, generally using Alexa Fluor 488, Alexa Fluor 546, Alexa Fluor 594, Alexa Fluor 647, and DAPI. (For mutant experiments, a lacZ probe or a probe to a gene covered by a deficiency was additionally labelled with one of these same fluorophores, so that homozygous mutant embryos could be easily identified.) All imaging channels were acquired sequentially to minimise cross-talk. The laser lines and collection windows were: 405 laser and 443–472 nm window for DAPI; 488 laser and 500–536 nm window for Alexa Fluor 488; 561 laser and 566–584 nm window for Alexa Fluor 546 or Alexa Fluor 555; 594 laser and 610–631 nm window for Alexa Fluor 594; 640 laser and 663–713 nm window for Alexa Fluor 647. Alexa Fluor 514 (514 laser and 519–540 nm window) was used in place of Alexa Fluor 488 for a round of HCR experiments carried out when the 488 laser was awaiting repair. When necessary, a transmitted light channel was also collected to allow for embryo staging based on the progress of cellularisation.

Embryo staging was based on Bownes stages (Bownes, 1975; Campos-Ortega and Hartenstein, 1997), with subdivision of particular stages into substages where necessary (details in Appendix 1). Fiji (Schindelin et al., 2012) was used for routine inspection of imaging data and certain image adjustments (details in Appendix 2). Image processing and analysis scripts were written in Python 3 (https://www.python.org) using the libraries NumPy (Harris et al., 2020), SciPy (Virtanen et al., 2020), scikit-image (van der Walt et al., 2014), and matplotlib (Hunter, 2007); see Appendix 2 for details. Figures were assembled in Affinity Designer (Serif Europe). Embryo outlines were drawn manually in Affinity. Image look-up tables (LUTs) were either chosen from the ‘ChrisLUTs’ LUT package for ImageJ (Christophe Leterrier and Scott Harden; https://github.com/cleterrier/ChrisLUTs; ‘NeuroCyto LUTs’ update site in Fiji) or generated for custom colours using a macro provided by Nicolás De Francesco (https://github.com/ndefrancesco).

Models were implemented in Python using NumPy (Harris et al., 2020), and outputs were plotted using matplotlib (Hunter, 2007). See Appendix 4 for details.