A key role for foxQ2 in anterior head and central brain patterning in insects

Anterior patterning in bilaterian animals is based on a set of highly conserved transcription factors, such as orthodenticle/otx, empty spiracles/emx, eyeless/Pax6 and other genes, which have comparable expression and function from flies to mice (Hirth et al., 1995; Leuzinger et al., 1998; Quiring et al., 1994; Simeone et al., 1992). Likewise, canonical Wnt signaling needs to be repressed in order to allow anterior pattern formation in most animals examined, with Drosophila being a notable exception (Fu et al., 2012; Martin and Kimelman, 2009; Petersen and Reddien, 2009). Recently, additional genes have been studied that are expressed anterior to the orthodenticle/otx region at the anterior pole of embryos of all major clades of Bilateria and their sister group, the Cnidaria. These data suggested that a distinct but highly conserved anterior gene regulatory network (aGRN) governs anteriormost patterning (Lowe et al., 2003; Marlow et al., 2014; Posnien et al., 2011b; Sinigaglia et al., 2013; Steinmetz et al., 2010; Yaguchi et al., 2008). This region gives rise to the apical organ in sea urchins, annelids and hemichordates, and other structures. Most of the respective orthologs were shown to be expressed in the vertebrate anterior neural plate as well as in the anterior insect head (Posnien et al., 2011b), although it remains disputed which, if any, tissue is homologous to the apical organ in these species (Hunnekuhl and Akam, 2014; Marlow et al., 2014; Santagata et al., 2012; Sinigaglia et al., 2013; Telford et al., 2008). Detailed interactions of this aGRN have been determined only in a few model systems (Range and Wei, 2016; Sinigaglia et al., 2013; Yaguchi et al., 2008).

The aGRN of sea urchins as representative of deuterostomes is best studied in Strongylocentrotus purpuratus and Hemicentrotus pulcherrimus. Here, six3 (sine oculis homeobox homolog 3/optix) is the most upstream regulator, which is initially co-expressed with foxQ2. Both genes are restricted to the anterior pole by repression by posterior Wnt signaling (Range and Wei, 2016; Wei et al., 2009; Yaguchi et al., 2008). six3 in turn is able to repress Wnt signaling (Wei et al., 2009) and to activate a large number of genes including rx, nk2.1 and foxQ2 (Wei et al., 2009). Subsequently, foxQ2 represses six3 but activates nk2.1 expression at the anteriormost tip. In this tissue freed of six3 expression, foxQ2 is responsible for establishing a signaling center involved in the differentiation of the apical organ (Range and Wei, 2016). In addition, foxQ2 expression is activated by nodal signaling (Yaguchi et al., 2016). six3 knockdown leads to a strong morphological phenotype, including change of embryonic epidermal shape and loss of neural cells (Wei et al., 2009). In foxQ2 knockdown, by contrast, an epidermal phenotype, other than animal plate thickening, was not observed (Yaguchi et al., 2008) but foxQ2 appears to be essential for the specification of neural cell types (Yaguchi et al., 2008, 2012, 2016).

Nematostella vectensis (Cnidaria) represents the sister group to bilaterian animals (Sinigaglia et al., 2013). Here, both Nv-six3 and Nv-foxQ2 are initially activated by Wnt/β-catenin signaling, whereas at later stages there seems to be an antagonism between Wnt/β-catenin signaling and Nv-six3 (Leclère et al., 2016). Nv-six3 activates Nv-foxQ2 and several other genes of the aGRN (Marlow et al., 2013; Sinigaglia et al., 2013). Like in the sea urchin, knockdown of Nv-six3 leads to strong morphological defects including loss of the apical organ, whereas Nv-foxQ2 knockdown does not affect the morphology of the embryo but the apical organ is reduced (Sinigaglia et al., 2013). In contrast to sea urchin, Nv-foxQ2 does not appear to regulate Nv-six3. The repression of six3 and foxQ2 by Wnt signaling has been shown for a hemichordate (deuterostome) (Darras et al., 2011; Fritzenwanker et al., 2014) and for foxQ2 in a hydrozoan (cnidarian) (Momose et al., 2008). Neither sea urchins nor cnidarians possess a highly centralized nervous system, such that a function of foxQ2 in brain development could not be tested.

Within protostomes, the expression of aGRN genes has been studied extensively in postembryonic stages of the annelid Platynereis dumerilii (Lophotrochozoa). However, the only functional interaction tested was the repression of Pd-six3 and Pd-foxQ2 by Wnt signaling, which was activated by pharmacological treatment (Marlow et al., 2014). Within arthropods (Ecdysozoa), the red flour beetle Tribolium castaneum has become the main model system for studying anterior patterning (Posnien et al., 2010), because head development is more representative for insects than the involuted head of Drosophila. Indeed, the anterior morphogen bicoid is present only in some dipterans, while the canonical anterior repression of Wnt signaling is observed in Tribolium only (Brown et al., 2001; Stauber et al., 1999; Fu et al., 2012). Neither terminal Torso signaling nor the terminal gap gene huckebein has an influence on head formation in Tribolium (Schoppmeier and Schröder, 2005; Kittelmann et al., 2013). Apart from a region with similarity to vertebrate neural plate patterning there is also a largely non-neural anterior median region (AMR) patterned by other genes (Kittelmann et al., 2013; Posnien et al., 2011b). Interestingly, Tc-six3 is a central regulator of anterior head and brain patterning, which represses Tc-wg expression, among other genes, but does not regulate Tc-rx or Tc-nk2.1 (Posnien et al., 2011b).

In an ongoing genome-wide RNAi screen in Tribolium (iBeetle screen) (Dönitz et al., 2015; Schmitt-Engel et al., 2015) a head phenotype similar to that of Tc-six3 RNAi was induced by the dsRNA fragment iB_03837. The targeted gene was the Tribolium ortholog of foxQ2 (Tc-foxQ2), which encodes a Forkhead transcription factor. All members of this family share the Forkhead DNA-binding domain and they are involved in development and disease (Benayoun et al., 2011). Although highly conserved among animals, this gene was lost from placental mammals (Mazet et al., 2003; Yu et al., 2008). Within arthropods, anterior expression was described for the Drosophila ortholog fd102C (CG11152) (Lee and Frasch, 2004) and Strigamia maritima (myriapod) foxQ2 (Hunnekuhl and Akam, 2014). However, the function of this gene has not been studied in any protostome so far.

We studied the expression and function of Tc-foxQ2 by RNAi and heat shock-mediated misexpression and found that it has a much more central role in the insect aGRN than in sea urchin and cnidarians. Surprisingly, Tc-foxQ2 and Tc-six3 form a regulatory module with mutual activation, contrasting with the clear upstream role of six3 in other species. Another difference is that Tc-foxQ2 knockdown led to a strong epidermal phenotype. Further, we found a novel role of Tc-foxQ2 in CNS patterning. Specifically, it was required for the development of the mushroom bodies (MBs) and the central complex (CX), both of which are higher-order processing centers of the insect brain (Heisenberg, 2003; Pfeiffer and Homberg, 2014). Finally, we present the most comprehensive aGRN available for protostomes.

In the iBeetle screen, injection of the dsRNA fragment iB_03837 led to first instar larval cuticles with reduced or absent labrum (cyan area in Fig. 1) with high penetrance (Dönitz et al., 2015; Schmitt-Engel et al., 2015) (Fig. 1A). The targeted gene was TC004761 (Tcas_OGS 3.0), which was revealed as the sole Tribolium ortholog of foxQ2 (Tc-foxQ2) (Fig. S1).

Quantitative analyses of parental RNAi experiments with two non-overlapping dsRNA fragments (Tc-foxQ2^RNAi_a and Tc-foxQ2^RNAi_b; 1.5 µg/µl) (Fig. 1A,B, Tables S1-S4) in two different genetic backgrounds (Fig. S2, Tables S3-S6) revealed the same morphological phenotype, arguing against off-target effects or strong influence of the genetic background (Kitzmann et al., 2013). The proportions of eggs without cuticles or with cuticle remnants (strong defects) were within the range observed in wild type (WT) (van der Zee et al., 2005). Weak Tc-foxQ2 RNAi phenotypes (Fig. 1C,D) showed a labrum that was reduced in size and loss of one or both labral setae (Fig. 1E,F, yellow dots). Intermediate phenotypes (Fig. 1C,D) were marked by a reduced labrum and loss of one or both anterior vertex triplet setae (Fig. 1E,G, red dots), indicating additional deletions of the head capsule. In strong phenotypes (Fig. 1C,D), the labrum was substantially reduced or deleted along with several setae marking the anterior head and/or the labrum (Fig. 1E,H). No other specific L1 cuticle phenotypes were detected. We tested higher dsRNA concentrations (2 µg/µl and 3.1 µg/µl; data not shown) as well as double RNAi using both dsRNA fragments together (1.5 µg/µl each; data not shown). None of these variations resulted in a stronger cuticle phenotype. Taken together, Tc-foxQ2 is required for epidermal patterning of anterior head structures. Interestingly, the RNAi phenotype was similar to that of Tc-six3, although somewhat weaker and less penetrant (Posnien et al., 2011b).

The labrum is an appendage-like structure (Posnien et al., 2009a) and its outgrowth requires cell proliferation regulated by Tc-serrate (Tc-ser) (Siemanowski et al., 2015). We examined when the size of the labrum decreased after Tc-foxQ2 RNAi and whether cell proliferation or cell death was involved. We found that the labral buds in Tc-foxQ2^RNAi embryos were decreased and fused from fully elongated germ band stages onwards (Fig. 2Aa′-d′). We found no regulation of Tc-ser by Tc-foxQ2 (see below).

Next, we quantified apoptosis in embryos 6-26 h after egg laying (AEL) (Table S7) using an antibody against cleaved death caspase-1 (Dcp-1) (Florentin and Arama, 2012). We quantified apoptotic cells in the labral region (region 1 in Fig. 2B) and a control region (region 3). Fully elongated germ bands showed a 6-fold increase in the number of apoptotic cells in the labral region compared with the control region after Tc-foxQ2 RNAi (P=4.1×10⁻⁴, n=15; Fig. 2C) coinciding with the stage of morphological reduction. Hence, Tc-foxQ2 prevents apoptosis in the growing labrum.

The phenotypic similarity to Tc-six3^RNAi embryos (Posnien et al., 2011b) and the expression of Tc-foxQ2 in neuroectodermal tissue (see below) prompted us to check for brain phenotypes. We performed parental RNAi in the background of the brainy line, which marks glia by ECFP (black signal in Fig. 3A-C) (Koniszewski et al., 2016). In the weakest phenotypes, the medial lobes of the MBs were reduced and appeared to be medially fused (Fig. 3, compare open arrowheads in A with B). Further, the central body (CB; part of the CX) was shortened (Fig. 3B) and the brain hemispheres were slightly fused (Fig. 3, compare solid arrowheads in B with A). In stronger phenotypes, the CB was clearly reduced in size and the MBs were not detectable. Further, the brain hemispheres appeared fused at the midline (Fig. 3C).

We performed RNAi in the background of the transgenic MB-green line, which marks MBs by EGFP (Koniszewski et al., 2016) (the magenta-framed black signal in Fig. 3D-H). We observed a similar range of MB body phenotypes. In addition to fused MBs, we found misarranged MBs that had lost their medial contact (Fig. 3F,G, arrowheads), interdigitated MBs (Fig. 3G), and highly reduced or even absent MBs (Fig. 3H). Importantly, the strength of epidermal and neural phenotypes correlated. Larvae with weak neural defects showed a decreased labrum, while strong neural phenotypes correlated with lack of the entire labrum. Taken together, we found Tc-foxQ2 to be required for brain formation, with the MBs, CB and the midline being strongly affected. Again, these defects are similar to those reported for Tc-six3 loss of function (Posnien et al., 2011b).

The expression of Tc-foxQ2 started in two domains at the anterior terminus of the germ rudiment (Fig. 4A,B). During elongation, the two domains approach each other as a consequence of morphological movements (Fig. 4C-E). At late elongating germ band stages, the domains split into several subdomains in the AMR and the labrum anlagen (arrowhead in Fig. 4F,H) and domains lateral to the stomodeum (arrow in Fig. 4G,I,J); in addition, there are domains in the neuroectoderm (e.g. arrow in Fig. 4F,L; see approximate fate map in Fig. S3). Very weak staining in the ocular region was detected using the tyramide signal amplification (TSA) system (Fig. 4K). These data are comparable to the anterior expression of foxQ2 orthologs in other animals and are consistent with the phenotype in Tribolium.

We mapped the expression of Tc-foxQ2 relative to other genes of the aGRN by double in situ hybridization in WT embryos. Expression overlaps are outlined (dotted lines) in Fig. 5 and Fig. S4. First, we tested for genes that interact with foxQ2 in other species or are required for labrum formation (Coulcher and Telford, 2012; Economou and Telford, 2009; Kittelmann et al., 2013; Posnien et al., 2009a,b). We found no overlap with Tc-wingless/wnt1 (Tc-wg) expression until retraction, when the emerging stomodeal Tc-wg domain overlapped with Tc-foxQ2 expression (Fig. 5A). By contrast, we found complete overlap with Tc-six3 expression at early embryonic stages (Fig. 5Ba), which developed into a mutually exclusive expression at intermediate elongating germ bands (Fig. 5Bc). Thereafter, these genes remained mutually exclusive, except for a small anterior median neuroectodermal region (lateral area marked in Fig. 5Bd-g) and the labrum anlagen (median area marked in Fig. 5Bf,g). These data are in agreement with the interactions described for sea urchin, where six3 initially activates foxQ2 but at later stages six3 becomes repressed by foxQ2 (see Introduction). The later coexistence of mutually exclusive and co-expression domains along with the many different expression domains of Tc-foxQ2 indicate a complex and region-specific regulation. Early co-expression developing into partially overlapping expression patterns was observed for both Tc-cap‘n'collar (Tc-cnc) and Tc-scarecrow/nk2.1 (Tc-scro) (Fig. 5C,D). For Tc-crocodile (Tc-croc), a small overlap was observed, which remained throughout development (Fig. 5E).

Next we studied genes that, based on their expression and function, are thought to be downstream components of the aGRN. Tc-retinal homeobox (Tc-rx) was expressed in a largely non-overlapping pattern, apart from small domains in the labrum and neuroectoderm at late stages (Fig. S4A). Expression of Tc-chx and Tc-ser initially largely overlaps with that of Tc-foxQ2, but they later resolve to mainly non-overlapping patterns (Fig. S4B,E). Tc-forkhead/foxA (Tc-fkh) expression was essentially non-overlapping with Tc-foxQ2 (Fig. S4C). Tc-six4 marks a region with molecular similarity to the vertebrate placodes (Posnien et al., 2011a). No co-expression was observed until late stages, when a small domain expresses both genes (Fig. S4D). In summary, Tc-foxQ2 expression appears to have a central and dynamic role in the aGRN.

We first tested the interactions of foxQ2 known from other species. Tc-foxQ2 was virtually absent in Tc-six3 RNAi embryos (Fig. 6 Ca-c, compare with Aa-c) indicating a conserved role of Tc-six3 in Tc-foxQ2 activation. Only at later stages was some residual Tc-foxQ2 expression seen in the stomodeal region (not shown). Unexpectedly, Tc-six3 expression was strongly reduced in Tc-foxQ2^RNAi germ rudiments (compare Fig. 6Ba with Da), which contrasts with findings in cnidarians and sea urchins. At later stages, medial Tc-six3 expression emerged (Fig. 6Db) and developed a similar shape and intensity as in WT (Fig. 6Dc). However, the lateral neuroectodermal domains remained strongly reduced or even absent (Fig. 6Db,c, arrowheads). Efficient knockdown of Tc-foxQ2 by RNAi was shown by in situ hybridization and RT-qPCR (Figs S5 and S6). The reduction of Tc-six3 in Tc-foxQ2^RNAi animals was confirmed by RT-qPCR experiments (–25.6±0.05%; Fig. S6).

At early embryonic stages, Tc-wg expression was not altered in Tc-foxQ2^RNAi (not shown), whereas later the labral Tc-wg expression was lost (Fig. S7). Likewise, inhibition of canonical Wnt signaling by Tc-arr RNAi (Bolognesi et al., 2009) did not alter Tc-foxQ2 expression at early stages (Fig. 6Ea,b). Later, the neuroectodermal Tc-foxQ2 expression domains were expanded (Fig. 6Ec, open arrowhead), whereas the anterior labral Tc-foxQ2 expression domain was lost (Fig. 6Ec, solid arrowhead). The Tc-six3 expression domains were expanded in Tc-arr^RNAi germ rudiments covering the anterior region (Fig. 6Fa). Later, the neuroectodermal Tc-six3 expression domains were strongly expanded (Fig. 6Fb,c, arrowheads), whereas the median domains were unchanged. Overactivation of canonical Wnt signaling via Tc-axin RNAi led to a strong reduction of Tc-foxQ2 (Fig. 6Ga,b) and Tc-six3 (Fig. 6Ha,b) expression. In summary, as in other model systems, we found activation of Tc-foxQ2 by Tc-six3 and repression by Wnt signaling. In contrast to other clades, Tc-foxQ2 was required for early Tc-six3 expression.

The AMR of the insect head harbors the labrum and the stomodeum. Tc-cnc and Tc-croc are upstream factors required for anterior and posterior AMR patterning, respectively (Economou and Telford, 2009; Hunnekuhl and Akam, 2014; Kittelmann et al., 2013). The AMR expression domain of Tc-cnc in the labrum was strongly reduced after knockdown of Tc-foxQ2 (Fig. 7Ba-d, compare with Aa-c). Likewise, Tc-croc expression was affected, but the reduction was restricted to the anterior boundary of expression (arrowheads in Fig. 7Da-d, compare with Ca-c); posterior expression around the stomodeum was largely unchanged. Conversely, in Tc-croc and Tc-cnc RNAi we observed no alteration of Tc-foxQ2 expression at early stages (not shown), indicating an upstream role of Tc-foxQ2. However, at later stages, expression of Tc-foxQ2 was reduced in the labrum in both treatments (Fig. S8). Next, we tested the medial AMR markers Tc-scro and Tc-fkh. Tc-scro was reduced anteriorly and laterally in Tc-foxQ2^RNAi embryos in early elongating germ bands (Fig. 7Fa, arrowhead, compare with Ea), but its posterior aspects were only moderately altered. In contrast to WT embryos, in Tc-foxQ2^RNAi the stomodeal/labral expression of Tc-scro remained connected to the lateral expression in neuroectoderm (Fig. 7Fb-d, arrows, compare with Eb-d). Conversely, Tc-foxQ2 was not altered in early Tc-scro^RNAi embryos, whereas in later embryos changes were observed (Fig. S8). The expression of the stomodeum marker Tc-fkh was not considerably altered in Tc-foxQ2^RNAi embryos (Fig. S9), whereas later aspects of Tc-foxQ2 expression were altered upon Tc-fkh RNAi, probably by indirect effects (Fig. S9).

The Notch pathway ligand Tc-ser and the ubiquitin ligase Tc-mind bomb1 (Tc-mib1) are required for Notch signaling, and knockdown of both leads to loss of the labrum (Siemanowski et al., 2015). However, we detected no difference in Tc-ser expression in Tc-foxQ2^RNAi (not shown). Conversely, in early and intermediate elongating Tc-mib1^RNAi embryos only the lateral aspects of Tc-foxQ2 expression appeared mildly decreased (Fig. S10), arguing against a strong interaction with the Notch pathway. At later stages, by contrast, some lateral and labral Tc-foxQ2 expression domains were clearly reduced in Tc-mib1^RNAi embryos. Taken together, these results demonstrated an upstream role of Tc-foxQ2 in early anterior AMR patterning and indicated that the later interactions of the aGRN differ from the early interactions.

The anterior neuroectoderm is marked by the expression of several highly conserved transcription factors and harbors the anlagen of the insect head placode (Posnien et al., 2011a), the pars intercerebralis and the pars lateralis (Posnien et al., 2011b) (Fig. S3). Furthermore, it corresponds to the region where in grasshoppers several neuroblasts arise, which are required for CX development (Boyan and Reichert, 2011). Tc-chx expression was completely lost in early elongating Tc-foxQ2^RNAi germ bands (Fig. 8Ba, arrowhead, compare with Aa) and highly reduced at later stages (arrows in Fig. 8Bb-d, compare with Ab-d). The Tc-six4 domain was much reduced in early Tc-foxQ2^RNAi embryos, showing only small spots of expression at the anterior rim (Fig. 8Da, arrow, compare with Ca). Later, the lateral expression developed normally, whereas a median aspect of its expression was lost (Fig. 8Db-d, arrows, compare with Cb-d). Tc-rx expression at early elongating germ band stages was absent after Tc-foxQ2^RNAi (Fig. 8Fa, compare with Ea). This was unexpected because Tc-rx is largely expressed outside the Tc-foxQ2 expression domain (Fig. S4Aa-c), arguing against a direct effect. Indeed, our misexpression studies indicate a repressive role (see below). At later stages, the lateral aspects of Tc-rx expression recovered but the labral expression domain remained reduced or lost, in line with labral co-expression (Fig. 8Fb,c, arrowheads, compare with Eb,c).

Next, we scored Tc-foxQ2 expression in Tc-chx^RNAi, Tc-six4^RNAi and Tc-rx^RNAi embryos. In neither treatment were early aspects of Tc-foxQ2 expression affected, whereas at later stages we found expression differences within the neurogenic region (Fig. S11). These results confirm the upstream role of Tc-foxQ2 in early anterior patterning and they confirm that the interactions of the aGRN at later stages differ from those at early stages. We found no change in cell death in the neurogenic region of Tc-foxQ2^RNAi embryos until retraction, when cell death was significantly increased 1.5-fold (P=0.023) (Fig. S12). Hence, the observed changes in expression domains are likely to be due to regulatory interactions and not loss of tissue.

Heat shock-mediated misexpression has been established in Tribolium (Schinko et al., 2012). We generated eight independent transgenic lines and selected one for our experiments that showed the most evenly distributed misexpression upon heat shock (Fig. S13). Heat shock-induced misexpression led to a reproducible pleiotropic cuticle phenotype. The bristle pattern of the anterior head showed diverse signs of mild disruption (Fig. S14). Ectopic Tc-foxQ2 expression led to a reduced number of segments in the legs, the abdomen and the terminus. For our experiments, we used the earliest possible time point of misexpression (9-13 h AEL), which led to a higher incidence of anterior defects than at 14-20 h AEL and 20-25 h AEL (not shown). At 14-18 h AEL (5 h after a 10 min heat shock) the embryos were fixed and marker gene expression was scored. WT embryos undergoing the same procedure were used as negative control.

Heat shock-induced expression starts at late blastoderm stages (Schinko et al., 2012). Therefore, we were not able to test for the early interactions of the aGRN. Hence, comparably mild alterations in expression were found for Tc-wg, Tc-six3 and Tc-croc after ectopic Tc-foxQ2 expression (Fig. S15).

The strongest effects were on genes with late-onset expression, i.e. Tc-rx, Tc-six4, Tc-scro and Tc-cnc (Fig. 9). Tc-rx expression was reduced to a punctate pattern (Fig. 9B,C, compare with A). This repressive function of Tc-foxQ2 on Tc-rx is in line with the non-overlapping, adjacent expression of these two genes (Fig. S4Aa-c). Therefore, we assume that the reduction of Tc-rx found in Tc-foxQ2 RNAi is due to secondary effects, although this remains speculative (see above). Ectopic Tc-foxQ2 expression caused a premature onset and an expansion of Tc-six4 expression (Fig. 9D,F) and an additional ectopic domain was found in the posterior head (Fig. 9E,F, arrowhead). Tc-scro expression emerged precociously and was expanded (Fig. 9H,I and K,L; compare with G and J, respectively). Later, the domain was altered in shape and had a punctate appearance. In the case of Tc-cnc, a posterior expansion in expression was observed (Fig. 9N,O, arrowhead, compare with M).

With this work we present the first functional analysis of foxQ2 in protostomes and, together with previous work, we present the most comprehensive aGRN in that clade (Fig. 10). We found that Tc-foxQ2 is required at the top of the gene regulatory network to pattern the anteriormost part of the beetle embryo.

Interestingly, we identified a positive regulatory loop between Tc-six3 and Tc-foxQ2 at early embryonic stages (germ rudiment), forming a novel regulatory module. In support of this, several genetic interactions with downstream genes are similar (Posnien et al., 2011b) (see Fig. 10). The six3/foxQ2 regulatory module regulates a large number of genes that are required for anterior AMR patterning (e.g. anterior Tc-cnc and Tc-croc) and neuroectoderm patterning (e.g. Tc-chx). Hence, it lies at the top of the aGRN governing insect anterior head and brain development and both components elicit similar phenotypes when knocked down. Such a self-regulatory module at the top of a gene network is not without precedence. For instance, eyes absent, eyeless, dachshund and sine oculis form a positive regulatory loop at the top of the Drosophila eye gene regulatory network and, in consequence, all four genes are required for eye development and have similar mutant phenotypes (i.e. complete loss of the eyes) (Wagner, 2007). With our data, we are not able to distinguish between alternative modes of interaction: Tc-foxQ2 and Tc-six3 could cooperate in the regulation of the same targets, they could individually regulate different subsets, or one gene could be the regulator of all target genes while the other would just be required for the initiation of expression.

Within the regulatory module, Tc-six3 appears to be the more upstream of the two based on a number of criteria. First, Tc-six3 expression starts a bit earlier than that of Tc-foxQ2 (at the differentiated blastoderm stage, as compared with the germ rudiment for Tc-foxQ2), and it has a larger expression domain, within which early Tc-foxQ2 is completely nested. Further, the loss of Tc-foxQ2 in Tc-six3 RNAi is more complete, even at later stages, than the loss of Tc-six3 in Tc-foxQ2 in RNAi (Fig. 6). Finally, the Tc-six3 cuticle phenotype is more penetrant and comprises a slightly larger region of the dorsal head (compare with Posnien et al., 2011b).

Interestingly, the mutual activation of the six3/foxQ2 module does not persist, as the initially overlapping expression patterns diverge to become largely distinct at later embryonic stages (Fig. 5Bc-g). Just small domains in the neuroectoderm and the labrum continue to co-express both genes (encircled in Fig. 5Bd-g). Given their almost mutually exclusive expression, they might even switch to mutual repression at these later stages. In line with this scenario, the heat shock-induced misexpression of Tc-foxQ2 led to a reduction in lateral aspects of Tc-six3 expression (Fig. S15). Hence, it could be that later repression of six3 by foxQ2 is conserved between sea urchin and beetles.

Functional studies of foxQ2 orthologs were restricted to a sea urchin, as a model for deuterostomes, and the sea anemone, a cnidarian, representing the sister group to the bilaterian animals. This functional study in protostomes allows the first conclusions to be drawn on the evolution of foxQ2 function. Compared with sea urchin and sea anemone, Tc-foxQ2 plays a much more important role in the aGRN of our protostome model. First, it is clearly required for epidermal development, as evidenced by the loss of the entire labrum in knockdown animals. This is in contrast to sea urchin and sea anemone, where no epidermal phenotype was described apart from a thickened animal plate (Sinigaglia et al., 2013; Yaguchi et al., 2008). Second, Tc-foxQ2 is required for the development of two brain parts involved in higher-order processing, namely the CX and the MBs. In other models, FoxQ2 affects specification of certain neural cell types but does not lead to tissue loss. (Sinigaglia et al., 2013; Yaguchi et al., 2008, 2010). Finally, Tc-foxQ2 is required for Tc-six3 expression in Tribolium, which is not the case in the other models (Range and Wei, 2016; Sinigaglia et al., 2013; Yaguchi et al., 2008). Many of these novel functions may be explained by Tc-foxQ2 gaining control over Tc-six3 expression in our protostome model system.

Based on previous expression data, it has been suggested that foxQ2 orthologs played a role in the anterior development of all animals and that this involved interaction with six3 orthologs (Fig. S16) (Fritzenwanker et al., 2014; Hunnekuhl and Akam, 2014; Marlow et al., 2014; Martín-Durán et al., 2015; Santagata et al., 2012; Sinigaglia et al., 2013; Tu et al., 2006; Wei et al., 2009). In line with a conserved function, at early stages foxQ2 shows co-expression with six3 in deuterostomes and cnidarians and a nested expression within protostomes. In all cases, foxQ2 arises within the six3 domain (Fig. S16, left column). At later stages, by contrast, expression is more diverse. Some species retain nested or co-expression with six3 (Sinigaglia et al., 2013; Marlow et al., 2014; Martín-Durán et al., 2015), whereas other species develop six3-negative/foxQ2-positive domains (Hunnekuhl and Akam, 2014; Martín-Durán et al., 2015; Fritzenwanker et al., 2014; Wei et al., 2009; Yaguchi et al., 2008), similar to what we found in the beetle. In other species, the anteriormost region is cleared of expression of both genes (Fig. S16, right columns). This variation is not clearly linked to certain clades, indicating that the regulatory interactions at later stages might have evolved independently. In addition to the novel functions described above, there are conserved functional aspects as well. Initial activation of foxQ2 by six3 is found in all three model species. Repression of six3 by foxQ2 was found in Strongylocentrotus and Tribolium (at later stages) but not Nematostella (Range and Wei, 2016; Sinigaglia et al., 2013).

Together, these data indicate that at the base of metazoans foxQ2 and six3 were involved in early anterior patterning, with six3 being upstream of foxQ2. In bilaterians, the expression of both genes might have become restricted to the anterior by repression by posterior Wnt signaling (Leclère et al., 2016; Darras et al., 2011; Fritzenwanker et al., 2014; Marlow et al., 2014; Range and Wei, 2016; Sinigaglia et al., 2013; Wei et al., 2009; Yaguchi et al., 2008). After the split from Cnidaria, the Urbilateria foxQ2 evolved a repressive function on six3, leading to more complex expression and increased diversity of the molecular code specifying cells at the anterior pole (Range and Wei, 2016). This diversification might have been required for the evolution of more diverse neural cell types. In protostomes, foxQ2 additionally evolved control over early six3 expression. Indeed, that six3 is slightly more upstream of foxQ2 in the regulatory module of Tribolium might be a remnant of its ancestrally more important role. A curiosity is the loss of such a highly conserved gene in the genome of placental mammals, whereas other vertebrates still have the gene (Mazet et al., 2003; Yu et al., 2008). Unfortunately, foxQ2 function in vertebrates remains unstudied and our evolutionary hypothesis needs to be tested in other models representing deuterostomes, Lophotrochozoa and Ecdysozoa.

Animals (Tribolium castaneum) were reared under standard conditions at 32°C (Brown et al., 2009). The San Bernadino (SB) wild-type strain was used for all experiments, except for initial reproduction of the phenotype, where the black (Sokoloff, 1974) and the Pig-19/pBA19 (Lorenzen et al., 2003) strains were used as in the iBeetle screen. The Tc-vermillion^white (v_w) strain (Lorenzen et al., 2002) was used for transgenesis and heat shock experiments. Transgenic lines MB-green line (G11410) and brainy marking parts of the brain were described previously (Koniszewski et al., 2016).

Tc-foxQ2 full coding sequence (1633 bp; GenBank accession number XM_008202469) was obtained from the Tribolium genome browser (http://bioinf.uni-greifswald.de/gb2/gbrowse/tcas5/) and the sequence was confirmed by cloning the full coding sequence from cDNA. Phylogenetic analysis was performed using MEGA v.5 (Tamura et al., 2011). The multiple sequence alignment was conducted with the ClustalW algorithm using the parameters preset in MEGA v.5. Positions containing gaps were eliminated from the dataset. The phylogenetic tree was constructed using the neighbor-joining method with the Dayhoff matrix-based substitution model (Schwartz and Dayhoff, 1979). Bootstrap tests (Felsenstein, 1985) were conducted using 1000 replicates to test the robustness of the phylogenetic tree. Tc-Lin31 was the second best hit in the NCBI BLASTp (Altschul et al., 1997) search (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and used as outgroup.

To test for RNAi efficiency, we examined Tc-foxQ2 mRNA in Tc-foxQ2 RNAi embryos (6-26 h AEL) by in situ hybridization (ISH). As expected, no signal was detected using regular detection settings (not shown) but increased exposure time revealed residual Tc-foxQ2 expression at advanced embryonic stages (Fig. S5, note the increased background in B,D,F,H). These domains reflected only part of the WT expression pattern, indicating autoregulatory interactions restricted to some domains. Our subsequent analyses focused on early patterning, where the RNAi knockdown was shown to be very efficient (Fig. S5A,B). Furthermore, RT-qPCR (supplementary Materials and Methods) experiments confirmed efficient knockdown of the Tc-foxQ2 mRNA level (−91.3%) using the Tc-foxQ2^RNAi_a dsRNA fragment (Fig. S6). Embryonic RNAi phenotypes were found with a penetrance of over 90%, except for Tc-foxQ2, Tc-rx and Tc-six4, which showed a penetrance of over 40%.

The templates for the non-overlapping dsRNA fragments used in this study were generated by PCR from a plasmid template (for primer and dsRNA template sequences see Table S8). The dsRNAs were synthesized using the MEGAscript T7 Transcription Kit (Invitrogen) or purchased from Eupheria BioTech. The transcribed dsRNA was extracted via isopropanol precipitation (Tc-foxQ2^RNAi_a) or phenol/chloroform extraction (Tc-foxQ2^RNAi_b) and dissolved in injection buffer (1.4 mM NaCl, 0.07 mM Na₂HPO₄, 0.03 mM KH₂PO₄, 4 mM KCl, pH 6.8). The injected dsRNA concentrations for parental RNAi with Tc-foxQ2^RNAi_a and Tc-foxQ2^RNAi_b were 1.0 µg/µl, 1.5 µg/µl and 3.1 µg/µl. Unless stated otherwise, 1.5 µg/µl dsRNA was used. Pupal injections were performed as previously described (Bucher et al., 2002; Posnien et al., 2009b). The dsRNA was injected using FemtoJet Express (Eppendorf). Cuticles of the L1 larval offspring were prepared as described (Wohlfrom et al., 2006).

Standard immunostaining was performed using the cleaved Drosophila Dcp-1 (Asp216) rabbit antibody (Cell Signaling Technology, #9578) at 1:100 dilution. Anti-rabbit secondary antibody coupled with Alexa Fluor 488 (Thermo Fisher Scientific, A-11070) was used for detection at 1:1000 dilution. ISH (alkaline phosphatase/NBT/BCIP) and double ISH (alkaline phosphatase/NBT/BCIP, and horseradish peroxidase-mediated TSA reaction; the Dylight550 conjugate was synthesized by Georg Oberhofer) were performed as described previously (Oberhofer et al., 2014; Schinko et al., 2009; Siemanowski et al., 2015).

The regions of interest were set based on head morphology. Cell counting was performed using the Fiji cell counter plug-in (Schindelin et al., 2012). The number of apoptotic cells was positively correlated with age. Hence, to circumvent systematic errors due to staging, the apoptotic cell number in the posterior procephalon was used to normalize the data. This region was chosen because it was outside the foxQ2 expression domain and unaffected by our RNAi experiments. The correction value was calculated by dividing the mean number of apoptotic cells of RNAi embryos by the mean number of apoptotic cells in WT embryos in the control region. For the normalization, the data were divided by the correction value. Raw counts are shown in Table S7.

The normalized data were tested with R (v.2.14.2; http://www.R-project.org/) for the homogeneity of the variances via the box plot, and for normal distribution via the Shapiro-Wilk test. To test for significance, three statistical tests were conducted: Welch's t-test, two-sample t-test, and the Wilcoxon rank-sum test. All three tests showed the same levels of significance. P-values are based on the Wilcoxon rank-sum test results.

The foxQ2 heat shock construct was based on the constructs developed by Schinko et al. (2012) and germline transformation was performed as described previously (Berghammer et al., 1999; Schinko et al., 2012) using the injection buffer used for RNAi experiments and the mammalian codon-optimized hyperactive transposase (Yusa et al., 2011) flanked by the Tc-hsp68 sequences (gift from Stefan Dippel, Göttingen University). All animals for heat shock experiments were kept at 32°C. Heat shock was performed as described previously (Schinko et al., 2012) for 10 min at 48°C, with egg collections for cuticle preparations at 0-24, 9-13, 14-20 and 20-25 h AEL, and egg collections for ISHs at 9-13 h AEL (fixed 5 h after heat shock). For each gene, one batch of embryos was used. Phenotypes shown in figures were exhibited by over 50% of the embryos in the batch.

Cuticle preparations and L1 larval brains were imaged as described (Posnien et al., 2011b; Wohlfrom et al., 2006) using an LSM 510 or Axioplan 2 microscope (Zeiss) and processed using Amira (v.5.3.2; FEI) using ‘voltex’ projections. Stacks were visualized as average or maximum projections using Fiji (v.1.49i). All images were level-adjusted and assembled in Photoshop CS (Adobe) and labeled using Illustrator CS5 (Adobe).

We thank Claudia Hinners for excellent technical help and Stefan Dippel for the hyperactive helper construct. The phenotype was initially identified in the iBeetle project (DFG Research Unit FOR1234).