Non-canonical dorsoventral patterning in the moth midge Clogmia albipunctata

Bone morphogenetic proteins (BMPs) are of central importance for dorsal–ventral (DV) axis specification. They are core components of a signalling cascade that includes the BMP ligand decapentaplegic (DPP) and its antagonist short gastrulation (SOG) in Drosophila melanogaster. These components are very ancient, with orthologs involved in DV patterning in both protostomes and deuterostomes. Despite such strong conservation, recent comparative work in insects has revealed interesting differences in the way the patterning function of the DV system is achieved in different species. In this paper, we characterise the expression patterns of the principal components of the BMP DV patterning system, as well as its signalling outputs and downstream targets, in the non-cyclorrhaphan moth midge Clogmia albipunctata (Diptera: Psychodidae). We previously reported ventral expression patterns of dpp in the pole regions of C. albipunctata blastoderm embryos. Strikingly, we also find ventral sog and posteriorly restricted tkv expression, as well as expanded polar activity of pMad. We use our results from gene knock-down by embryonic RNA interference to propose a mechanism of polar morphogen shuttling in C. albipunctata. We compare these results to available data from other species and discuss scenarios for the evolution of DV signalling in the holometabolan insects. A comparison of gene expression patterns across hemipteran and holometabolan insects reveals that expression of upstream signalling factors in the DV system is very variable, while signalling output is highly conserved. This has two major implications: first, as long as ligand shuttling and other upstream regulatory mechanisms lead to an appropriately localised activation of BMP signalling at the dorsal midline, it is of less importance exactly where the upstream components of the DV system are expressed. This, in turn, explains why the early-acting components of the DV patterning system in insects exhibit extensive amounts of developmental systems drift constrained by highly conserved downstream signalling output.


Background
Signalling molecules of the TGF-β family are widely conserved across the animal kingdom. They are key factors in developmental processes such as dorsal-ventral (DV) axis specification, appendage formation, patterning of the central nervous system, and cell proliferation (reviewed in [1]). Bone morphogenetic proteins (BMPs) are of particular interest for DV axis specification, because they The resulting free ligand dimers attach to their receptors [encoded by punt (put), saxophone (sax), and thick veins (tkv)], which form transmembrane complexes [9,12] (Fig. 1c). This binding event triggers an intracellular signalling process: once the ligand-receptor complex has been established, SAX and TKV phosphorylate Mothers against dpp (MAD). Phosphorylated MAD (pMAD) (associated with the signal transducer Medea) translocates to the nucleus (Fig. 1d) to activate or repress a number of target genes (such as brinker, Dorsocross, tailup, twist, snail, pannier, among others), whose localised regulation induces different tissue fates along the DV axis. In the DV patterning system, initially triggered by the Toll signalling pathway [13,14], dpp is a key factor for the determination of dorsal tissues and acts as a morphogen for the specification of the dorsal ectoderm and the amnioserosa [6,15]; see [9,10] for a review.
In D. melanogaster, the heteromeric shuttling complex (DPP-SOG-SCW-TSG) transports the dpp protein ligand dorsally from the lateral regions of the broad dpp expression domain (Fig. 1). SOG is the component responsible for the diffusive transport of the complex from the ventral-lateral region, where sog is expressed, towards the dorsal region, where the concentration of SOG is low [9,11,16]. Once close to the dorsal midline, the complex is cleaved by TLD (Fig. 1c). This process of complex formation, diffusion, and subsequent local cleavage creates a sink for the shuttling complex, and DPP in particular, around the dorsal midline. High concentrations of SOG laterally ensure capture of DPP for transport. Two additional factors help to ensure that DPP signalling only occurs in a restricted dorsal range. First, pnt and sax are ubiquitously expressed, but tkv is restricted to dorsal regions, and all three receptors are needed for signalling to occur. Second, low concentrations of SOG dorsally make rebinding to DPP unlikely, which therefore remains free to attach predominantly to its receptors in this region (Fig. 1). A similar shuttling mechanism is proposed to be in place in vertebrates, although the DV polarity of the system is reversed in these organisms [17].
The function of the genes involved in this signalling cascade appears to be conserved in insects, and some components of the system are very ancient, controlling DV patterning in both protostomes and deuterostomes [18,19]. Despite such strong conservation, recent work on holometabolan insects, including the mosquito Anopheles gambiae [20], the honey bee Apis mellifera [4], the jewel wasp Nasonia vitripennis [5,21,22], and the red flour beetle Tribolium castaneum [3,23,24], together with hemipteran milkweed bug Oncopeltus fasciatus has revealed some interesting differences in the way the patterning function of the system is achieved (see Fig. 1e for a phylogeny of the main species compared in this study). For example, one major difference between D. melanogaster and these insects is the absence of scw, a gene which originated recently within the lineage of the cyclorrhaphan flies via tandem duplication of gbb [25]. Other differences include the expression of key genes like sog and dpp, and the implications these altered patterns have for the signalling pathway. Hence, the DV patterning system in insects provides an interesting case study for how evolutionary changes in the distribution and timing of gene expression affect an ancient patterning system based on morphogenetic gradients.
In this paper, we characterise the expression patterns of the principal components of the DV patterning system and its downstream targets in Clogmia albipunctata (Diptera: Psychodidae). C. albipunctata is a member of the Psychodomorpha, a basally branching lineage of dipterans [26,27]. It is an emerging model for evolutionary and developmental studies (see [25,26,[28][29][30][31][32]). We have previously reported that a single copy dpp is expressed in the ventral pole regions of the blastoderm embryo [25]. This species lacks scw, while a single copy of gbb is expressed at the blastoderm stage in a large central domain, excluding the poles of the embryo [25]. In addition, we assay gene expression in embryos depleted of dpp, gbb, and sog transcripts via RNA interference (RNAi). We compare the expression patterns observed in C. albipunctata to those in other insects (A. mellifera, N. vitripennis, T. castaneum, A. gambiae, and O. fasciatus). Our comparative analysis and RNAi investigation suggest specific mechanisms for the establishment of the DV signalling morphogen gradients and indicate evolution of DV patterning by extensive developmental system drift [33,34].

Gene identification, cloning, and synthesis of RNA constructs
We searched the early embryonic transcriptome of C. albipunctata ([26]; http://diptex.crg.es) using BLAST with sequences from D. melanogaster retrieved from GenBank. Gene identity was confirmed via reciprocal BLAST against the D. melanogaster genome. PCR primers were designed from transcriptome sequences ( Table 1). Amplified sequences for C. albipunctata have been deposited in GenBank (see Table 1 for accession numbers). Fragments were cloned into the PCRII-TOPO vector (Invitrogen) and used to make DIG or FITClabelled riboprobes for whole-mount in situ hybridisation, as well as double-stranded RNA constructs for RNAi. Constructs for RNAi against C. albipunctata dpp and gbb were synthesised from clones of sequences KC810051 and KC810052, respectively, and for sog from MF457413. We utilised primers containing a T7 promoter at their 5′ end to amplify a DNA template. This template was then purified and used to simultaneously synthesise sense and antisense RNA strands using T7 RNA polymerase. The residual DNA was then digested using RNAse-free DNAse before being extracted three times using phenol/chloroform/isoamylalcohol followed by precipitation in ethanol. The resulting pellet was washed in 70% ethanol before being resuspended in injection buffer (100 mM NaPO4, pH 6.8; 5 mM KCl). RNA was then annealed to form double-stranded (ds) RNA in a thermocycler by cycling at 94 °C for 40 s, then reducing the temperature by − 0.1 °C per cycle for 750 cycles. dsRNA was checked on an agarose gel. We obtained dsRNAs of the following length: dpp, 1550 bp; gbb, 1640 bp; and sog, 1460 bp. Constructs were quantified and, if necessary, diluted and reannealed to a concentration of 4-6 µM before being stored at − 20 °C. Before each use dsRNA was defrosted and centrifuged for 15 min at 12,000g in a desktop centrifuge to remove and debris to prevent clogging of the microinjection needle.

Embryo collection and fixation
Wild-type and RNAi-treated embryos of C. albipunctata were collected at blastoderm and post-gastrulation stages as described previously in [28]. For whole-mount in situ hybridisation, embryos were heat-fixed using a protocol adapted from [35]. In brief, embryos were dechorionated at the desired developmental stage by submerging in 25% bleach for 45 s. They were then heatfixed for 20 s in boiling fixing solution (0.7% NaCl; 0.05% Tween-20); heat fixation was stopped by adding water. These embryos then underwent a second fixation with formaldehyde (5%) and PBS/methanol. Devitellinisation was achieved in 50% heptane-methanol by vigorous shaking for 20 s, and embryos were preserved in methanol. For RNAi treatment, embryos were dechorionated by hand and fixed as described above. For immunostaining, embryos were fixed using a solution of PEMS/methanol, formaldehyde, and heptane for 25 min at room temperature [36], followed by devitellinisation and storage as described above.

Whole-mount in situ hybridisation and immunohistochemistry
Whole-mount in situ hybridisation was performed as described previously in [26]; and references therein. In brief, after rehydration for 5 min in 50% MetOH-PBT followed by PBT washes, embryos were permeabilised by proteinase K treatment (8 mg/ml PBT) for 7 min at room temperature (RT), followed by 25-min refixation (5% formaldehyde in PBT). After washes and prehybridisation for 90 min at 56 °C, hybridisation was carried out overnight at 56 °C with a labelled probe at a concentration of 0.5-1 ng/μl. After washes and blocking in 5% heat-inactivated goat serum, antibody antidigoxigenin or fluorescein conjugated with alkaline phosphatase (Sigma) was used at 1:2000 for 2 h at RT followed by washing overnight. Staining was achieved using NBT/BCIP or fast red, and embryos were counterstained with DAPI and mounted in 70% glycerol. Enzymatic immunodetection of pMAD was done using a cross-reactive monoclonal anti-Smad3 antibody (phosphor S423 + S425) [EP823Y] (Abcam, UK, cat number ab52903) at a 1:10 dilution following the protocol described previously in [30]. In brief, after blocking in PBTB, embryos were incubated with the antibody at a concentration 1:10 overnight at 4 °C. Embryos were incubated in secondary antibodies (antirabbit conjugated with alkaline phosphatase) for 2 h at RT. Staining was achieved using NBT/BCIP, counterstaining with DAPI, and embryos were mounted in 70% glycerol.

RNA interference (RNAi)
RNAi treatment was carried out based on protocols established in other dipteran species [37][38][39]. In brief, activated embryos were allowed to develop for 2 h before submerging them in 25% bleach for 10 s to weaken the chorions. Embryos were then rinsed in water under a tap for 1 min before being aligned on a microscope slide against a glass capillary (Hilgenberg 1421602 65 mm × OD 0.25 mm) and covered with a 3:1 mixture of 10:27 halocarbon oil. Embryos were injected while maintaining a constant flow of liquid to avoid blocking of the needle. Aluminosilicate (rather than borosilicate) capillaries were pulled in Sutter P-97 Flaming/Brown Micropipette Puller to produce needles better able to penetrate with the hard extraembryonic membranes of C. albipunctata without breaking. After injection embryos were allowed to develop for 7 h before being sprayed into a mesh disc with water from a squeeze bottle. The mesh and embryos were then transferred into a 50-ml tube to be fixed and collected as previously described [35]. We injected buffer only as a negative control, along with in situ hybridisation staining for depleted transcripts. sog dsRNA was injected four independent times, with a survival to blastoderm stage of 22%, obtaining approximately 300 embryos, dpp dsRNA was injected three times, with survival to blastoderm of 15%, leading to approximately 150 embryos, and gbb dsRNA was injected two times, with survival to blastoderm stage of 18%, obtaining around 120 embryos. For cuticle preparations, embryos were allowed to develop until first larval instar stages and were then manually extracted from the eggshell under oil using tungsten needles. Additional oil was dissolved in heptane, and the soft tissue removed in an acetic acid/ glycerol solution for one day. Cuticles were then transferred to a slide in Hoyers medium and lactic acid (2:1) and covered with a cover slip; residual soft tissue was removed by placing at 65 °C overnight. Only dsRNAinjected cuticles presented abnormal phenotypes.

Cryosections of stained embryos
Embryos were embedded in optimal cutting temperature (OCT) media and preserved at − 80 °C. Using a CM3050 S cryostat (Leica, Germany), 20-μm transversal sections were obtained.

Expression patterns of the DV patterning system components and its downstream targets in C. albipunctata
We examined the expression patterns of the major components of the BMP signalling cascade comprising the DV patterning system as well as its core signalling output and key downstream targets in blastoderm and postgastrulation embryos of the moth midge C. albipunctata.
Where necessary, we utilised double in situ hybridisation to establish the relative position of expression patterns with regard to each other.

Identification of the major components of the DV patterning system
We previously isolated single copies of dpp and gbb from an early embryonic transcriptome [26]. We used the same genomic resource to identify other members of the DV patterning system in C. albipunctata (Table 1) and assigned orthology by reciprocal BLAST. Only Doc had more than one copy in the transcriptome. Two paralogues were found that appear to have arisen from an independent duplication within the lineage leading to C. albipunctata because they branch together in a phylogenetic tree, (http://phylomedb.org/?q=search_ tree&phyd=174&seqid=CAL_comp7148_c0_seq1). We used the longer sequence to design primers. No duplicates of any other DV genes analysed here were present in the transcriptome. Cloned sequences were deposited in GenBank with the accession numbers as listed in Table 1.

Expression of the major components of the DV patterning system
In previous work, we detected dpp transcripts in two . We also cloned the other two D. melanogaster DPP receptors, punt and saxophone. Unfortunately, we could not detect any clear expression patterns of these genes by in situ hybridisation in C. albipunctata embryos. Finally, the protease tld is expressed dorsally in early blastoderm embryos (Fig. 2g,

Spatial distribution of BMP signalling outputs
We detected BMP signalling activity by immunostaining against pMAD in blastoderm-stage embryos of C. albipuncata using a cross-reactive antibody (see Materials and Methods). We found pMAD to be localised in a contiguous narrow domain along the dorsal midline ( Fig. 3a-d, blue). This domain expands laterally at both poles of the embryo (Fig. 3b, c), an effect not seen in D. melanogaster (Fig. 3f, g), if a subtle posterior expansion has been described before [40]. The expanded dorsal-polar domains of pMAD are similar in size and antero-posterior extent to the more ventral expression domains of dpp in these regions (Fig. 3d). Therefore, both expression patterns seem to mirror each other in the polar regions ( Fig. 3b-d).

Downstream targets of BMP signalling
We characterise the expression patterns of downstream targets of D. melanogaster BMP signalling in C. albipunctata. In D. melanogaster, low levels of BMP signalling coincide with the ventral activation of twist (twi) and snail (sna) defining the mesodermal region of the embryo [41,42]. Similarly, in C. albipunctata, twi and sna are expressed in domains along the ventral midline of the blastoderm embryo (Fig. 4a, b). In D. melanogaster, brinker (brk) is a transcriptional repressor of DPP target genes negatively regulated by DPP signalling activity [43]. In C. albipunctata, we find that brk is expressed in two ventral-lateral domains (Fig. 4c). In D. melanogaster, pannier (pnr) and Dorsocross (Doc) are activated by high levels of DPP signalling activity along the dorsal midline [2,44,45]. They are involved dorsal closure [46], and in the specification of the amnioserosa [45], respectively. In C. albipunctata, we detect Doc expression in a dorsal domain (excluding the serosa) and a head stripe during the blastoderm stage . Stains as indicated in panels: a dpp (blue)/zen (red); b dpp (red)/sog (blue); c sog (blue)/zen (red); d sog (blue); e tkv (blue)/zen (red); f tkv (blue); g tld (blue)/sog (red); h tld (blue)/zen (red); g′, h′ DAPI nuclear counterstaining of embryos in g, h. Counterstains were used for staging as described in [30]: f/f′ is at cleavage stage 11 (C11), and g/g′ is at cleavage stage 14 (C14) (Fig. 4d). C. albipunctata pnr is not unambiguously detectable until after gastrulation; it then shows expression at the dorsal ectoderm ridge (Fig. 4e, arrow) and in the head lobes ( Fig. 4e, arrowheads). In addition, we looked at the laterally expressed markers of the neurogenic ectoderm, the columnar genes vnd, ind, and msh [47][48][49]. In C. albipunctata, these genes are only detectable after gastrulation, in expression patterns restricted to tissues of the nervous system. We see no obvious DV polarity in the expression of vnd, with its broad anterior head domain (Fig. 4f). ind has a ventral expression pattern in two spotted stripes along the anteroposterior axis (Fig. 4g). msh is expressed in a similar pattern, dorsal to ind (Fig. 4h). All in all, the expression patterns of these target genes are very similar to those found in D. melanogaster and other insects.

Functional characterisation of the DV patterning system in C. albipunctata
To functionally characterise the DV patterning system in C. albipunctata, we knocked down expression of dpp, gbb, and sog using RNA interference (RNAi).
Attempts to obtain cuticles from dpp-depleted embryos failed due to late embryonic lethality. Larval cuticles from sog-depleted embryos show severe defects in the abdominal region including one or a combination of the following: failure of dorsal closure (5/16) (Fig. 5a), reduction of the dorsal abdominal region adjacent to the thorax (6/16) (Fig. 5b), compressed abdominal segments (5/16) (Fig. 5c), and constricted or absent terminal region (12/16) (Fig. 5 a-c, arrowheads). Given the nearly abutting expression pattern of dpp and sog, we wondered whether there was a transcriptional regulatory interaction between the two genes. Such transcriptional effects have been described before in O. fasciatus [50], where sog is repressed by BMP signalling, as it is in spiders, vertebrates, and sea anemones [51][52][53][54]. An effect of sog on dpp transcription has been reported previously in Drosophila, where sog blocks dpp autoactivation in the neuroectoderm [54], and RNAi knock-down of sog in T. castaneum leads to downregulation of dorsal expression of dpp during segment formation [3]. Although dpp-depleted embryos did not show any effects on sog  (Fig. 5d).
To identify the effect of dpp knock-down on downstream target genes we stained embryos treated with dpp RNAi for markers of neurogenic ectoderm (brk) and mesoderm (twi). brk shows a strong expansion of gene expression dorsally (18/22; Fig. 5f, arrow), indicating an expansion of neurogenic ectodermal tissue to the dorsal midline. In contrast, brk expression remains low along the ventral midline (Fig. 5f, arrowhead). Consistent with this, twi expression in ventral mesoderm is also not affected (n = 18; Fig. 5i). To rule out a redundant role for the other BMP ligand present in C. albipunctata, we depleted embryos of gbb transcripts using RNAi, but could not detect any discernible effect on brk expression compared to the wild type (n = 15; Fig. 5g).

Expression patterns of BMP DV signalling genes in insects
Our expression analysis for DV patterning genes in C. albipunctata reveals an unexpected amount of upstream variation despite highly conserved target gene output (Fig. 6). Perhaps most surprising is the previously reported restriction of dpp expression to two ventral polar domains [25] (Fig. 2a, b), which stands in stark contrast to the dorsal expression domain observed in D. melanogaster [15]. This represents an extreme case of the very widespread expression variability for BMP ligands across insect species. In A. gambiae, dpp expression is much broader than in D. melanogaster, expanding into the ventral-lateral region of the embryo [20]. In A. mellifera, dpp is expressed in an anterior and a posterior domain that show no obvious DV polarity [55]. In N. vitripennis, dpp shows very faint ubiquitous expression before gastrulation [22]. Finally, in early embryos of T. castaneum, dpp expression is uniform along the DV axis, with a small posterior cap expression, only becoming restricted dorsally after gastrulation [3,56]. Outside of the holometabola, in blastoderm embryos of the hemipteran O. fasciatus dpp is not detected [50], but is later expressed at the posterior pole, along the dorsal edge of the site of germband invagination [57]. These expression differences may reflect underlying differences in dpp regulation. RNAi knock-down of sog in C. albipunctata leads to expansion of dpp expression into the dorsal posterior pole region and ectopic expression in the middle of the embryo (Fig. 5d). Moreover, it leads to abnormalities in the abdominal and posterior patterning as well as dorsal closure (Fig. 5a-c). A transcriptional effect of sog on dpp has also been described in D. melanogaster [54] and in T. castaneum [3]. The expansion of dpp expression along the AP axis could suggest an implication of the BMP signalling pathway in the AP patterning, as seen in A. mellifera [55] and O. fasciatus [50]. Such an AP role is further supported by the compression of the abdominal segments in the cuticle phenotypes, although an analysis of the mechanism by which this could occur is beyond the scope of the present study.
A similar amount of variation can be observed for sog expression. In C. albipunctata, we detect a single ventral expression domain (Fig. 2b-d, g) similar to that seen in T. castaneum and A. gambiae [3,20]. In O. fasciatus, early blastoderm embryos have ubiquitous expression of sog, and only at mid-blastoderm does its expression become ventral, similar to the expression described here [50]. These expression patterns differ from the two mediolateral domains observed in D. melanogaster [58], which therefore may be evolutionarily derived. Interestingly, sog appears to have lost its ancestral role in DV patterning in hymenoptera: in A. mellifera it is only expressed after gastrulation [4], while in N. vitripennis it appears to be completely absent [21].
Other components of the DV patterning system also show variation in expression patterns. In C. albipunctata, we observe expression of the ligand tkv in a dorsal midline domain, covering the posterior-most ~ 25% of the embryo (Fig. 2e, f ). In D. melanogaster, tkv is largely, but not exclusively, restricted to the dorsal half of the embryo along the entire antero-posterior axis [59]. In A. gambiae, it is initially expressed throughout the dorsal ectoderm, later to be excluded from the presumptive serosa [20]. In N. vitripennis, it is faintly expressed ubiquitously [22]. In A. mellifera, it is restricted to the posterior third of the embryo, with no DV polarity [55]. This indicates that the localisation of tkv expression cannot explain the dorsal localisation of pMAD, as suggested for D. melanogaster [60,61].
Finally, the protease tld shows a very dynamic expression pattern in C. albipunctata. At the early blastoderm stage, it shows a dorsal expression domain (Fig. 2g), whereas later this expression has completely shifted to the ventral side of the embryo (Fig. 2h). This surprising switch is not seen in any other insect described so far, but expression variations for tld in other species are also very striking. Whereas tld expression is strictly dorsal in D. melanogaster [62], in N. vitripennis it is expressed in a small anterior dorsal domain and does not have any function in DV patterning [5]. In A. gambiae, tld expression is limited to lateral regions and is excluded from the dorsal ectoderm [20]. In T. castaneum, tld expression occurs in the presumptive germ rudiment and shows the highest levels of expression in a broad anterior domain [24]. In O. fasciatus, tld is expressed uniformly across the embryo circumference [50].
Taken together, the available evidence reveals a surprisingly large amount of expression variation among upstream signalling factors in the DV patterning system of insects. It is striking that the localisation and extent of expression along the DV axis are not particularly conserved for any of these factors.

Localised pMAD activity in insect embryos
Compared to the expression patterns of upstream signalling factors, the localisation of pMAD activity is much more conserved across holometabolan insect species. In C. albipunctata, pMAD is localised in a narrow stripe along the dorsal midline (Fig. 3a-d), similar to pMAD distributions in D. melanogaster (Fig. 3e-g; also see Dorfman and Shilo [63]) and N. vitripennis [5]. In A. gambiae the dorsal domain of pMAD activity is broader [20]. In T. castaneum, pMAD covers the dorsal 50% of the serosa in the anterior, narrowing posteriorly to cover a region of about 20% of the DV axis along the dorsal midline of the germ rudiment [3]. In addition to these similarities, C. albipunctata shows a peculiar deviation from the canonical dorsal holometabolan pattern of pMAD. Its DV domain of localisation expands at both anterior and posterior poles of the embryo, to dorsally mirror the ventral polar domains of dpp expression (Fig. 3b-d). This polar broadening of the pMAD domain is not seen in any other insect studied so far, although much more subtle polar expansions, predominantly around the posterior pole, are also seen in Drosophila and Megaselia [40,64]. The complementary polar patterns of dpp and pMAD suggest a correlation between the two. We discuss this observation in the context of a potential BMP ligand shuttling mechanism in C. albipunctata below.
In D. melanogaster, DPP signalling represses expression of brk, while brk in turn is a transcriptional repressor of other DPP target genes [43]. The dorsal morphogen gradient activates brk expression in the ventral neurogenic ectoderm, which restricts dpp expression to the dorsal half of the embryo [67]. This results in opposite activity gradients of pMAD and brk [2]. In C. albipunctata, we observe a conserved brk expression pattern in two ventral-lateral domains (Fig. 4c), very similar to D. melanogaster, A. gambiae, and N. vitripennis [20,21,43]. In contrast, brk is not expressed in the early embryo of T. castaneum and sog alone is responsible for restricting dpp expression to the dorsal side of the embryo [3]. It is not clear whether brk was recruited independently into DV patterning in hymenopterans and dipterans, or whether it was lost in the coleopteran lineage, although the evidence slightly favours the former scenario (see below).
In D. melanogaster, pnr and Doc are activated by DPP signalling along the dorsal midline of the blastoderm and are involved in dorsal closure and the specification of the extraembryonic amnioserosa, respectively [2,45,46]. In C. albipunctata, Doc is expressed in a dorsal domain excluding the serosa (Fig. 4d), plus a head stripe, similar to D. melanogaster, A. gambiae, and late blastoderm embryos of N. vitripennis [20,21]. In A. gambiae, Doc expression is restricted to the amnion, but repressed in the serosa [20]. In T. castaneum, Doc plays a role in extraembryonic tissue morphogenesis but not specification; it is expressed early in a dorsal anterior domain, then though the entire serosa, but most strongly in its dorsal region [3,68]. This indicates some variability in the role of Doc for determining extraembryonic tissues, which reflects the rapid evolution of these tissues among holometabolan insects [69].
In D. melanogaster, N. vitripennis, and T. castaneum, pnr is expressed in a broad dorsal domain during the blastoderm stage [3,21,70]. In C. albipunctata, it is not detectable until after gastrulation, exhibiting a pattern in the dorsal epidermis and head lobes of the embryo (Fig. 4e). This pattern is similar to that found in A. mellifera, where pnr shows post-gastrulation expression at the ventral edges of the amnion, resembling post-gastrulation expression in D. melanogaster and N. vitripennis as well [21]. This indicates that the onset of pnr expression seems to vary between species, while its post-gastrulation expression pattern is strongly conserved.
Columnar genes vnd, ind, and msh are lateral markers of the neurogenic ectoderm: vnd is required for the specification of ventral column neuroectoderm, ind for the specification of intermediate column neuroectoderm, and msh labels all of the dorsal column neuroectoderm [47]. Interaction between DPP and SOG is necessary to restrict the expression of these genes to their respective expression sites [48]. In N. vitripennis, like in D. melanogaster, these genes are expressed in ventral-lateral stripes [21,[47][48][49]. In C. albipunctata, ind and msh show similar expression patterns as in other species (Fig. 4g, h). In contrast, vnd can only be clearly detected in the head region, with potential additional expression (although faint and diffused) at the posterior pole and the ventral side of the embryo (Fig. 4f ). Unlike in N. vitripennis, D. melanogaster, and A. gambiae, we can only detect expression of all three columnar genes in C. albipunctata after gastrulation.
Despite some interesting differences in target gene expression between species, it is evident that downstream DV genes are much more conserved than the upstream signalling factors. In general, we observe a trend towards increasing conservation of expression and localisation patterns as we move downstream in the DV patterning cascade. This parallels the situation in the antero-posterior patterning system, where the most downstream tier of the segment-polarity gene network shows the highest degree of conservation [71][72][73].

BMP shuttling in C. albipunctata and other insect species
The complex nature of the post-translational shuttling mechanism for BMP ligands poses a challenge for the mechanistic interpretation of our evidence. Comparisons to vertebrate DV patterning suggest that this shuttling mechanism is extremely conserved [10,17,74]. What do our data reveal about shuttling of BMP ligands in C. albipunctata? The discrepancy between ventral dpp expression and dorsal pMAD localisation strongly suggests that some sort of ligand transport must be involved in DV pattern formation in this species.
One important feature that both D. melanogaster and C. albipuncatata share is the complementarity of their dpp and sog expression patterns. In D. melanogaster, the border between the two domains occurs in the ventral-lateral region of the embryo along the entire length of the antero-posterior axis [58]. In C. albipunctata, it is restricted to an anterior and a posterior interface, where sog expression abuts, with a slight overlap, the ventrally localised polar dpp domains (Fig. 2b). Shuttling is likely to occur at these interfaces as there will be opposing protein gradients of DPP and SOG present at these sites. This suggests that a significant amount of DPP would be shuttled around the anterior and posterior poles of the embryo in C. albipunctata, involving transport along the antero-posterior as well as the DV axis, rather than straightforward ventral-to-dorsal transport throughout the embryo as observed in D. melanogaster. If our interpretation of the evidence is correct, this can explain the intensified and expanded domains of pMAD in the anterior and posterior polar regions of the C. albipunctata blastoderm embryo (Fig. 3b, c).
There are additional differences between the two species. One concerns the posterior-only expression of tkv in C. albipunctata (Fig. 2e, f ). The discrepancy between tkv expression and pMAD localisation may indicate that other BMP receptors, such as PUT, may be required for signal transduction in the anterior of the embryo. Even more difficult to explain is the switch of tld expression from dorsal to ventral during the late blastoderm stage in C. albipunctata (Fig. 2g, h). Its function (if any) remains mysterious, but it either suggests that BMP ligand shuttling must be very dynamic in this species, or that all relevant protein cleavage by TLD and subsequent DPP signalling activity must happen before the dorsal-toventral transition in tld expression. Further experimental work including immunostaining against the relevant proteins and/or enzymatic activity assays will be required to gain further insight.
A final difference between species is the absence of scw in the genome of the non-cyclorrhaphan fly C. albipunctata [25]. Comparative analyses suggest that scw arose in the cyclorrhaphan lineage from a duplication of the ancestral gbb homolog [25,75]. The SCW ligand is essential for DPP transport in D. melanogaster [7]. Is it possible that gbb is fulfilling its role in C. albipunctata? gbb is required for DPP signalling in N. vitripennis [5]. Moreover, it is expressed dorsally in the blastoderm embryo of the scuttle fly Megaselia abdita [76], and in a ubiquitous pattern excluding the poles in C. albipunctata [25], similar to that seen in O. fasciatus [50]. Despite this, our evidence indicates that gbb is not needed for DV patterning in C. albipunctata. Expression of the target gene brk shows no detectable defects in gbb knock-down embryos (Fig. 5g), indicating that dorsally localised DPP signalling is occurring correctly in the absence of GBB protein. Taken together, our evidence suggests that DPP is the only BMP ligand contributing to DV patterning in C. albipunctata.
Our evidence indicates that BMP shuttling is likely to occur in C. albipunctata, although the exact set of factors involved and the spatio-temporal dynamics differ compared to D. melanogaster. This further suggests that BMP ligand shuttling is a conserved phenomenon in dipteran insects. The situation is more complicated in other holometabolan taxa. In T. castaneum, dorsal localisation of DPP signalling activity depends on SOG and TLD as in flies [3] but does not involve tsg [24]. Still, BMP ligand shuttling is probably happening in this species. In contrast, sog is not expressed in embryos of N. vitripennis [21] and is only expressed at late embryonic stages in A. mellifera [4]. Yet, BMP signalling is still responsible for the patterning of the DV axis in N. vitripennis [5]. It has been proposed that maternal localisation of BMP receptors, combined with zygotic regulatory feedback, could take the role of dorsal shuttling in hymenoptera [22]. It is not entirely clear whether this condition is ancestral or derived, although the fact that BMP ligand shuttling has been proposed to occur in vertebrates [17] would favour the latter alternative.
Fundamental differences between DV patterning in dipterans and hymenopterans are further supported by the following evidence: RNAi knock-down of dpp in C. albipunctata leads to an expansion of brk expression to the dorsal midline of the embryo (Fig. 5f ). This is similar to brk expression in dpp mutants of D. melanogaster [43], but very different to dpp knock-down in N. vitripennis, where brk is restricted to an antero-dorsal expression domain by an otherwise ubiquitous expansion of twi [5]. In A. mellifera, embryos treated with dpp RNAi also show dorsal expansion of twi [55], although to a lesser degree than in N. vitripennis. In contrast, C. albipunctata embryos treated with dpp RNAi show wild-type twi expression (Fig. 5i), similar to dpp mutants in D. melanogaster [5] (supporting the difference in the determination of the mesodermal fate by Toll or BMP signalling in dipterans versus hymenopterans). This indicates that dpp downregulation in dipterans induces dorsal expansion of the neurogenic ectoderm, while in hymenopterans it leads to an expansion of mesodermal markers. Such fundamental differences in the role of brk between dipterans and hymenopterans favour a scenario where brk was independently recruited into DV patterning in each lineage [5].

Conclusions
In this paper, we have analysed the expression of DV patterning factors in the moth midge C. albipunctata. A comparison of these expression patterns to those in other insects reveals that expression of upstream signalling factors in the DV system is very variable, while signalling output is highly conserved. This has two major implications.
The first of these two implications is the following: variable expression patterns of upstream factors, in particular the highly unusual ventral expression of dpp [25] and the dorsal-to-ventral switch of tld in C. albipunctata (Fig. 2g,  h), complicate the simple traditional picture in which protostomes show dorsal expression of BMP ligands and ventral expression of their inhibitors (e.g. sog), while vertebrates show the exact opposite pattern [19]. Our results and other work on BMP DV patterning in insects [3,5,[20][21][22][23][24][25]55] reveal that BMP ligands, as well as other factors involved in their transport, can be expressed at many different positions along the DV axis in different insect species (Fig. 6). This does not argue against the general possibility of a dorsal inversion of the DV axis [18,19,77], but indicates that such an inversion would involve more complex spatio-temporal changes in regulation and gene expression than a simple DV switch involving BMP ligands and their inhibitors. As long as ligand shuttling leads to an appropriate localised activation of BMP signalling at either pole of the DV axis, it does not seem to matter all that much where exactly the component factors of the DV system are expressed.
As in the first case, the second implication is also connected to the fact that conservation of gene expression increases for downstream factors in the DV system. This suggests that the DV patterning system evolves by developmental system drift (DSD). DSD is a mode of evolution where a regulatory network gets rewired while maintaining a constant patterning output [33,34]. It explains that homologous characters can be highly conserved despite substantial divergence in the underlying gene regulatory and signalling mechanisms. DSD has been proposed for many evolving developmental processes such as vulval specification in nematode worms [78], the establishment of left-right asymmetry in vertebrates [79], and the dynamics of antero-posterior patterning in insects [72,73,80]. Our results suggest that the early-acting components of the DV patterning system in insects are another strong example of DSD in a generally highly conserved signalling cascade. While most of the examples listed above focus on changes in transcriptional regulation, our work suggests that post-translational regulation-as in transport of BMP ligands-makes important contributions to DSD and will need to be more strongly considered in future studies of the phenomenon.