DamID transcriptional profiling identifies the Snail/Scratch transcription factor Kahuli as an Alk target in the Drosophila visceral mesoderm

Receptor tyrosine kinase (RTK) signaling enables transduction of extracellular signals into the cell and is essential in a wide range of developmental processes. Ligand-dependent RTK activation is conserved among metazoans, leading to engagement of signal transduction adaptor proteins, serine/threonine kinases and transcription factors (TFs) that regulate gene expression and promote a wide range of intracellular responses. In Drosophila melanogaster, the Anaplastic Lymphoma Kinase (Alk) RTK and its ligand Jelly Belly (Jeb), are involved in the development of the visceral mesoderm (VM), where they drive a signaling pathway required for specification of muscle founder cells (FCs) (Englund et al., 2003; Lee et al., 2003; Stute et al., 2004). Ligand-stimulated activation of Alk acts through the guanosine triphosphatase Ras, the scaffold protein connector enhancer of kinase suppressor of Ras (Cnk) and Aveugle/Hyphen (Ave/Hyp) to drive the mitogen-activated protein kinase/extracellular signal-regulated kinase (MAPK/ERK) pathway (Englund et al., 2003; Lee et al., 2003; Stute et al., 2004; Wolfstetter et al., 2017).

During Drosophila development, the mesoderm is partitioned along the dorso-ventral axis due to inductive inputs from the ectoderm, such as Decapentaplegic (Dpp), which induces high levels of Tinman (Tin) and subsequently Bagpipe (Bap), leading to VM specification (Frasch, 1995). Early VM consists of naïve Alk-expressing myoblasts that are specified to become either FCs or fusion-competent myoblasts (FCMs). FC specification requires activation of Alk signaling by Jeb secreted from the adjacent somatic mesoderm (Englund et al., 2003; Lee et al., 2003; Stute et al., 2004). After specification, FCs fuse with FCMs, forming binucleate myotubes (Campos-Ortega and Hartenstein, 1997; Klapper et al., 2002; Lee et al., 2006; Martin et al., 2001; Poulson, 1950). Fusion is required for the formation of circular visceral muscles, upon which the longitudinal muscle precursors migrate, forming a web of interconnected muscles surrounding the midgut endoderm (Georgias et al., 1997; Klapper et al., 2002; Kusch and Reuter, 1999; Martin et al., 2001; Rudolf et al., 2014). Although Alk is expressed throughout the VM, only the ventral-most row of cells are exposed to Jeb (Englund et al., 2003; Lee et al., 2003; Lorén et al., 2001; Stute et al., 2004). Alk signaling regulates transcription of FC-specific genes, including Hand, optomotor-blind-related-gene-1 (org-1) and kin of irre (kirre) (Englund et al., 2003; Lee et al., 2003; Stute et al., 2004; Varshney and Palmer, 2006). Animals devoid of FCs do not undergo visceral myoblast fusion and the gut musculature fails to develop (Englund et al., 2003; Lee et al., 2003; Lorén et al., 2001; Stute et al., 2004). The influence of Alk signaling on later events in visceral myogenesis is unclear; however, Alk activity is required for visceral decapentaplegic (dpp) expression in the VM (Shirinian et al., 2007).

Although Alk has been widely studied during Drosophila embryogenesis, assaying transcriptional effects specifically in the VM is challenging using traditional transcriptomics, and few Alk transcriptional targets have been described. We used Targeted DamID (TaDa) to determine genome-wide Alk-regulated transcriptional events in the embryonic VM. TaDa exploits the activity of bacterial DNA adenine methyltransferase (Dam) fused to any protein of interest to determine cell type-specific DNA-binding profiles, and has previously been used with RNA polymerases, TFs and histone modifiers, among others (Aughey and Southall, 2016). TaDa can further be refined to address DNA-binding profiles in specific tissues and time points using the well-established GAL4/UAS expression system (Brand and Perrimon, 1993; Southall et al., 2013). This tissue-specific approach revealed known and previously unidentified Alk VM target genes. Among these, we validated the Snail/Scratch family TF Kahuli (Kah) as a previously unreported VM target of Alk. Loss of Alk signaling resulted in reduced Kah mRNA expression in FCs, while activation of Alk increased Kah expression. To characterize Kah function, we generated Kah loss-of-function mutants, which fail to form the first midgut constriction at later embryonic stages. We show that this defect in Kah mutants is similar to that previously described for pnt mutants, suggesting that Kah and Pnt may function together to regulate this process. Publicly available ChIP datasets for Kah and Pnt revealed a number of common targets, reinforcing the hypothesis that Kah and Pnt work together in midgut morphogenesis. Thus, our Alk activity-dependent DamID approach successfully identified a number of Alk-regulated transcriptional targets in the embryonic VM, including the Kah TF that is required for correct midgut constriction.

To characterize Alk-regulated transcriptional activity in vivo, we employed TaDa in the embryonic VM in which we genetically manipulated Alk signaling output. We used transgenic Drosophila expressing Dam methylase fused to RNA-Pol II (here Dam-PolII) (Southall et al., 2013), driven either generally in the mesoderm (twi2xPE-GAL4) or specifically in the VM (bap-GAL4), resulting in methylation at GATC sites in targeted tissue (Fig. 1A; Movies 1-4). To manipulate Alk signaling we used combinatorial expression of either UAS-Jeb, which leads to ectopic activation of Alk, or UAS-Alk.EC.MYC, encoding a MYC-tagged extracellular domain (ECD) of Alk that inhibits Alk signaling in a dominant-negative manner (further referred to as UAS-Alk.DN) (Bazigou et al., 2007) (Fig. 1A,C-E). Animals expressing Dam-PolII alone in a wild-type background were employed to control for basal Dam-PolII signal (Fig. 1A,C). Stage 10-13 embryos were collected, representing a developmental window during which Alk is activated in future visceral FCs, and experimental sampling was performed in triplicate. Methylated DNA was isolated and digested with the methylation-specific DpnI restriction endonuclease, followed by next-generation sequencing to identify Alk transcriptional targets in the VM (Fig. 1B). Our analysis of this dataset was based on previous pipelines developed for DamID (Maksimov et al., 2016; Tosti et al., 2018). The number of quality reads obtained were comparable between samples and replicates (>15M reads/sample, Fig. S1A). After alignment to the Drosophila genome, sequencing depth was >60%, with exception of one sample (Fig. S1B). A high degree of reproducibility was observed between biological replicates overexpressing Alk.DN and Dam-PolII. In contrast, samples expressing Jeb displayed substantial variation (Fig. S1C). Therefore, we employed twi2xPE-GAL4- and bap-Gal4-driven UAS-jeb NGS TaDa datasets for qualitative analyses only.

To assess whether our DamID approach recapitulates transcriptionally active regions of the genome, we performed a meta-analysis of Dam-PolII occupancy, as indicated by GATC associated reads (see Materials and Methods for details), relative to the distance to the closest transcription start site (TSS). When comparing all GATC motifs (Non Dam-PolII) and random regions in the genome to Dam-PolII methylated GATC sites (Dam-PolII), we observed a tendency for methylated GATC sites to accumulate close to TSSs (Fig. S2A). In addition, we compared our DamID results with previously published RNA-seq data from isolated mesoderm cells (NCBI BioProject, PRJEB11879). In agreement with our previous observations, the Dam-PolII-binding profile along all annotated genes is consistent with a mesodermal RNA expression profile (Fig. S2B,C), demonstrating that the Dam-PolII binding in our analyses reflects PolII in vivo occupancy.

To detect differential gene expression between Dam-PolII and Alk.DN samples, we clustered neighboring GATC-associated reads, a maximum of 350 bp apart (median GATC fragment distance for the Drosophila genome) into peaks (Tosti et al., 2018). Most peaks were associated with a single GATC (Fig. 2A). We then calculated the mean fold change ratios for all GATCs falling into each peak across annotated transcripts (Alk.DN/Dam-Pol II), and a false discovery rate (FDR) was assigned to each peak. Each gene along the genome was assigned an overlapping peak with the minimum FDR value, and its logFC and FDR were used for differential expression analysis visualization on a volcano plot (Fig. 2B).

For statistical analyses, a FDR <0.01 was considered significant. In total, we identified significant change in Dam-PolII occupancy on 1739 genes in the twi.2xPE-GAL4 samples (Twi) and 2107 genes in the bap-GAL4 samples (Bap), with an overlap of 1052 genes between samples (Fig. 2C). Identified genes included known targets of Alk signaling, such as Hand, org-1, kirre and dpp (Fig. 2B, Table S1) (Lee et al., 2003; Lorén et al., 2003; Shirinian et al., 2007; Stute et al., 2004; Varshney and Palmer, 2006), demonstrating that the TaDa approach was successfully able to identify Alk targets in the VM.

Alk signaling controls FC specification in the VM; therefore, we expected transcriptional activation of factors involved in this process. With TaDa, we observed peak enrichment for TFs in both Bap and Twi datasets, including Hand and org-1 (Fig. 2B,D). We also observed genes involved in VM cell fusion, such as kirre (upregulated), and sticks and stones (sns) and Verprolin 1 (Vrp1) (downregulated) (Fig. 2B). Moreover, we identified factors involved in signaling pathways known to be active during development of the mesoderm, musculature and nervous system (Fig. 2E).

Qualitative analysis of a selection of peak-associated genes with low P-values showed differential Dam-PolII occupancy between Jeb and Alk.DN overexpression samples. At individual gene levels, Dam-PolII occupancy revealed similar binding profile dynamics for Jeb (positive peaks) and Alk.DN samples (negative peaks) (Fig. 3A-F). Known Alk targets, such as Hand and org-1, exhibit highly specific expression in the VM (Fig. 3A,B). Further validation of a selection of highly significant differentially expressed genes by in situ showed them to be also actively expressed in the VM. These included CG11658, the transmembrane protein failed axon connections (fax), the Kahuli (Kah) TF and the SUMO family protein Sumo (previously referred to as smt3) (Fig. 3C-F). Taken together, our bioinformatics analysis and experimental validation supports TaDa as an effective approach in identification of novel transcriptional regulation events that are potentially downstream of Alk in the VM.

To validate identified candidates, we employed single-cell RNA-sequencing (scRNA-seq) profiling on cells isolated from stage 10-13 embryos, using live/dead cell markers to isolate living cells through flow cytometric cell sorting (Fig. 4A). After quality control filtering in Seurat R and Scanpy toolkits, 1055 cells from wild-type embryos were further analyzed (Satija et al., 2015; Wolf et al., 2018). Unsupervised clustering of cells based on gene expression profiles identified 13 cell clusters with distinct transcriptional profiles that could be assigned to distinct cell lineages [epidermis, somatic mesoderm, early trunk mesoderm, amnioserosa (AS)+endoderm, neuroblasts, neurons, hemocytes, endoderm, visceral mesoderm, fat body, ectoderm, trachea and glia] (Fig. 4B,C, Fig. S3). When visualized in a two-dimensional uniform manifold approximation and projection (UMAP) plot, clusters distributed into four main groups, one of which comprised clusters of mesodermal origin (Fig. 4B,C, Fig. S3). Within this group, the cluster representing VM was identified by plotting combinatorial gene expression of known factors involved in VM development, such as biniou (bin), bagpipe (bap), org-1, Hand, Fasciclin 3 (Fas3) and Alk.

We next analyzed expression levels of TaDa-identified candidates within the VM cluster. However, this was not possible for our whole embryo dataset due to (1) the overall low number of VM cells and (2) the low proportion of VM FCs that precluded a rigorous interrogation of TaDa candidate expression in relation to Alk activity. To achieve this, GFP-expressing cells were purified by flow cytometric cell sorting from HandC-GFP;twi2xPE-Gal4>UAS-jeb embryos with an enlarged visceral FC population. After quality filtering, we identified 888 cells that distributed as five clusters based on gene expression profiles (Fig. 4E-H). Four of these clusters exhibited VM marker gene expression. The remaining cluster represented HandC-GFP-positive cells of the cardiac mesoderm (CM), as indicated by combinatorial expression of CM-specific genes (Fig. 4G,H). In earlier work, we observed a number of proliferating cells in the VM shortly after FC specification (Popichenko et al., 2013). Further analysis allowed identification of a cluster of cells that strongly expressed both FC- and cell-cycle/cell-proliferation markers, which we termed VM proliferating cells (Fig. 4E,H, Fig. S4). Heatmap analysis revealed an enrichment of TaDa-identified, Alk-regulated genes in the VM that was most prominent in the VM clusters, when compared with the CM cluster (Fig. 4G). Moreover, in agreement with our in situ analyses (Fig. 3C-F), we confirmed that CG11658, fax, Kah and Sumo were strongly expressed in the VM of our whole embryo scRNA-seq dataset (Fig. S5). Taken together, our scRNA-seq analysis supports our TaDa-based identification of novel Alk signaling targets in the VM.

Our TaDa analysis identified 151 TFs that could be potentially regulated by Alk signaling activity (Fig. 2C,D) (Table S1). We chose to further investigate one of these: the Snail family TF Kahuli (Kah; Fig. 3E). The Snail TF family in Drosophila comprises Snail (Sna), Worniu (Wor) and Escargot (Esg), while the Scratch A and B families comprise Scratch (Scrt), CG12605 and Kah, which share a common domain structure with a zinc-finger C2H2-type DNA-binding domain (Fig. 5A,A′). Kah is the only Scratch A family member and lacks the Scratch domain found in Scratch B family members (Kerner et al., 2009). Kah mRNA is expressed in a dynamic pattern, initially detected in early embryogenesis (Fig. S6). Expression is seen at stage 10 in the developing VM, and later is enriched in FCs, as revealed by both mRNA in situ and our VM-enriched scRNA-seq dataset (Fig. 5B,C). Kah expression was also observed in the somatic mesoderm (SM) (Fig. 5C), and in the brain and ventral nerve cord (VNC) (Fig. S6). Our in situ data identifying Kah expression in both the VM and SM was confirmed in our scRNA-seq datasets (Fig. S5).

The robust Kah mRNA signal observed in visceral FCs was consistent with our hypothesis that Kah could be a target of Alk activity in the VM. To test this, we assessed Kah expression in the VM upon twist(2xPE)-Gal4-driven Jeb overexpression, as well as in jeb mutants. Ectopic Alk activation led to strong, uniform Kah expression throughout the entire VM, whereas loss of Alk signaling in jeb mutants reduced, although did not eliminate, Kah expression (Fig. 5D, Fig. S6). As expected, Kah expression was still robustly expressed in the embryonic SM, as this tissue was unaffected by loss of Alk signaling (Fig. 5D, Fig. S6).

To visualize Kah protein during embryo development, we employed BAC clone CH322-97G04-derived strain from the modERN (model organism Encyclopedia of Regulatory Networks), which carries an extra copy of the Kah locus encoding a C-terminally GFP::FLAG-tagged variant of Kah, under control of endogenous regulatory elements (Kudron et al., 2018). This strain was generated by targeted genomic integration of the Kah.GFP recombining BAC into an intronic region of the Msp300 gene, and does not compromise fly viability (Fig. S7A). Kah-GFP was detected throughout the VM and SM, in agreement with Kah mRNA expression (Fig. S8). However, in contrast to our mRNA analyses, Kah-GFP was observed in both VM FCs and FCMs (Fig. S8). Given the large size of GFP and its tendency to form homodimers at high concentrations (Yang et al., 1996), we were concerned this might impact function and stability of the Kah-fusion protein, prompting us to generate a Kah allele with a C-terminal 3xOLLAS tag using CRISPR/Cas9-induced homology-directed repair (HDR) (Fig. 5E-G′, Fig. 6A, Fig. S9; referred to as Kah^Cterm.OLLAS). Viable Kah^Cterm.OLLAS animals were obtained that displayed nuclear OLLAS-tag staining in the visceral and somatic muscle (Fig. 5E-G′). Kah^Cterm.OLLAS was enriched in nuclei of VM FCs at stage 11/12, in keeping with our Kah mRNA observations (Fig. 5F,F′,H,H′,J,J′). To investigate whether Alk signaling was crucial for Kah expression, we examined Kah^Cterm.OLLAS in an Alk¹⁰ (Lorén et al., 2003) mutant background (Fig. 5H-K′). Kah^Cterm.OLLAS was weakly expressed and uniformly distributed in all VM cells of Alk¹⁰ mutants at stage 11/12 (Fig. 5I,I′,K,K′), suggesting that Alk signaling is not crucial for the initial expression of Kah, but regulates or maintains Kah expression after visceral FCs are specified.

Given the expression of Kah in the VM, we next addressed a potential role for Kah in this tissue during embryonic development. We initially characterized a Kah^f06749 PiggyBac insertion that was lethal in combination with Df(3L)Exel6085, which deletes the entire Kah locus [Fig. 6A, Figs S7, S10A; 0% of transheterozygous Kah^f06749/Df(3L)Exel6085 survived to L2 larvae (n=200)]. To explore Kah function further, we generated two additional alleles using CRISPR/Cas9: (1) Kah^ΔATG, deleting most of exon 2, including the predicted ATG start codon; and (2) Kah^ΔZnF, carrying an in-frame deletion that removes the region encoding the Kah zinc-finger domains (Fig. 6A). Surprisingly, and in contrast to Kah^f06749, Kah^ΔATG and Kah^ΔZnF were viable over Df(3L)Exel6085 (Fig. S10A). To investigate whether Kah mutants exhibit defects in visceral cell fate specification or VM morphology, we visualized Alk, Fas3 and Org-1 at stage 11/12 in Kah^f06749/Df(3L)Exel6085, Kah^ΔATG and Kah^ΔZnF embryos. For all Kah alleles, we noted that Alk expression and localization were similar to controls (Fig. 6B). In addition, loss of Kah did not affect early VM cell identity, as indicated by FC-specific expression of Org-1 (Fig. 6B).

To identify a function for Kah in the embryonic VM, we next performed RNA-seq on Kah mutants to identify putative targets of this previously uncharacterized TF. RNA-seq was performed on both Kah^ΔATG and Kah^ΔZnF mutants at embryonic stages 11-16. We noted 1664/2640 upregulated genes and 1191/2827 downregulated genes in Kah^ΔATG and Kah^ΔZnF mutants, respectively [threshold values: log2FC≤−0.59 (FC≤−1.5), Padj≤0.05] (Fig. 6C,D, Table S2). Further comparison identified 2524 overlapping differentially regulated genes [log2FC≥0.59 and ≤−0.59 (FC≥1.5 and ≤−1.5), Padj≤0.05] in both Kah mutants (1464 present at increased/1034 at decreased expression levels), the expression of which was correlated significantly (Fig. 6E,F). Many genes identified have yet to be investigated and represent interesting candidates for future functional characterization. As Kah encodes a TF, we performed over-representation analysis on these datasets using WebGestaltR (Liao et al., 2019). This led to the identification of BMP (Dpp), Toll, Notch and Hedgehog (Hh) as enriched pathways commonly upregulated in both Kah^ΔATG and Kah^ΔZnF embryos (Fig. 6G, Fig. S11).

As Dpp signaling has been reported to be required for late midgut development (Hursh et al., 1993; Masucci and Hoffmann, 1993), we examined Kah mutant midguts [Kah^f06749/Df(3L)Exel6085, Df(3L)Exel6085, Kah^ΔATG and Kah^ΔZnF] at later stages using antibody staining and live imaging (Fig. 7A,B, Fig. S10B,C). By stage 16, the midgut of wild-type embryos (Fig. 7A,B,D,E) has acquired three constrictions that subdivide it into four chambers (Campos-Ortega and Hartenstein, 1997; Poulson, 1950; Reuter and Scott, 1990; Schröter et al., 2006). Live imaging of Kah^ΔATG and Kah^ΔZnF mutants identified abnormalities in the first midgut constriction of both mutants (Fig. 7A; Movies 5-7). Using HandC-GFP and Fas3 to visualize the midgut, we found the first midgut constriction was frequently not formed or was incomplete (Fig. 7B,E, quantified in 7C; Kah^ΔATG 80% penetrance, n=89; Kah^ΔZnF 70% penetrance, n=109; Movies 5-7). This phenotype was also observed in Kah^f06749/Df(3L)Exel6085 and in Df(3L)BSC362/Df(3L)Exel6085 embryos in which Kah is entirely deleted (Figs S7B and S10B). As Wingless (Wg) and Dpp signaling events are important for midgut constriction, and our RNA-Seq analysis identified differential gene expression in Dpp signaling components in Kah mutants, we investigated Wg, Dpp and pMAD expression in Kah mutants. Wg protein, Dpp mRNA and robust pMAD expression was observed (Fig. 7B, Fig. S10C), suggesting that the defective midgut constriction observed in the absence of Kah is not due to loss of Wg or Dpp signaling. Although no decrease in pMAD was observed, we cannot exclude increased signaling through this pathway, which would reflect the increased expression of pathway components in our Kah mutant RNA-seq datasets and that may result in disruption of the midgut constriction process.

We next turned towards TFs with reported VM constriction phenotypes, such as Org-1 and Pointed (Pnt), as well as Hand and H2O, which exhibit FC-specific expression but have no obvious phenotypes in the embryonic VM (Barad et al., 1991; Bilder et al., 1998; Lo et al., 2007; Schaub and Frasch, 2013). Interestingly, a physical interaction of the ETS TF Pnt with Kah has been reported by Y2H (Thurmond et al., 2019), http://flybi.hms.harvard.edu/results.php, prompting us to further investigate similarities between pnt and Kah mutants. Analysis of the amorphic pnt^Δ88 allele confirmed the previously described midgut constriction phenotype (Fig. 7D) (Bilder et al., 1998). Using the HandC-GFP reporter, we revealed that Kah mutant embryos have an increased number of visceral nuclei in the midgut (Fig. 7E; quantified in 7F, n=30, P<0.001). These HandC-GFP-positive nuclei were also highly disorganized when compared with controls where visceral muscle nuclei are aligned in four rows (Fig. 7E). We next performed epistasis experiments between our Kah loss-of-function alleles and pnt^Δ88, identifying incomplete anterior midgut constriction formation in ∼12% of late stage transheterozygous embryos (n=154). Live imaging of Kah^ΔATG, pnt^Δ88 as well as Kah^ΔATG,pnt^Δ88 double mutant embryos revealed that the pnt^Δ88 mutant constriction phenotype increased in severity in Kah^ΔATG, pnt^Δ88 double mutants, which completely failed to initiate anterior constriction (Movies 8, 9, representative images in Fig. 7G; quantified in 7H). Taken together, these findings suggest that Kah and Pnt may function together to accomplish the anterior midgut constriction. Finally, we asked whether Alk signaling could impact on the midgut constriction process. To do this, we expressed dominant-negative Alk (bap-Gal4>UAS-Alk.DN) in the VM, observing defects in midgut constriction at low penetrance (34%, n=190) (Fig. 7I). Interestingly, we also observed a range of late gut phenotypes upon pan-mesodermal overexpression of Jeb (2xPE-Gal4>UAS-jeb), ranging from absent or incomplete midgut constriction with occasional bulging of the visceral muscle layer (56%, n=37) to severe defects that affected overall midgut formation (44%, n=37) (Fig. 7I). Thus, Alk signaling appears to be important for later events in midgut development, particularly midgut constriction formation.

To better understand Kah function, we employed ChIP-seq data from Kah-eGFP embryos that has been deposited publicly by the modENCODE project (accession number ENCSR161YRO) (Roy et al., 2010). This Kah-eGFP ChIP dataset contains a predominance of promoter regions with a peak in the vicinity of transcription start sites (TSS) (Fig. 8A, Fig. S12A,B). A Basic Motif search in regions 50 bp and 200 bp around the peaks led to the generation of a de novo motif for Kah with highest scoring similarity to the related Snail TF (Fig. 8B), in agreement with a recent study (Reddington et al., 2020). Among those genes containing a Kah motif in the vicinity of the TSS, we noted a number expressed in the visceral mesoderm, such as Antennapedia (Antp), mind bomb 2 (mib2) and Netrin-B (NetB) (Table S3). Comparison of differentially expressed genes identified in Kah mutants by RNA-seq with the Kah ChIP-seq dataset showed that 31% (339/1094) of genes in the Kah ChIP-seq dataset were differentially regulated in Kah mutants (Fig. 8C, Table S3). Further interrogation of our whole-embryo scRNA-seq dataset (Fig. 4) showed that these genes were more highly expressed across the visceral mesoderm cluster (Fig. 8D).

ChIP-seq data from transgenic Pnt-eGFP embryos has also been deposited publicly by the modENCODE project (Fig. 8C, Fig. S12C) (accession number ENCSR997UIM) (Roy et al., 2010), allowing us to compare Pnt and Kah binding locations throughout the genome, together with genes differentially expressed in Kah mutants. This analysis revealed that 46% (510/1094) of Kah-ChIP targets are potentially occupied by both Kah and Pnt (Table S3). Furthermore, 30% (157/510) of these common ChIP targets overlapped with genes differentially expressed in Kah mutants (Table S3). These overlapping genes include Antp, which is known to play an important role in setting up the first midgut constriction (Bilder et al., 1998; Roy et al., 1997), as well as Kah itself. We also found nej, put, Mad and Ras85D, which were upregulated in both RNA-seq datasets of Kah^ΔATG and Kah^ΔZnF mutants (Fig. 6G) as common Kah- and Pnt-ChIP targets. Taken together, our ChIP analysis together with our Kah mutant RNA-seq datasets identified a set of genes that are potentially regulated by Kah and Pnt downstream of Alk signaling during midgut constriction, some of which likely play important roles in this process (Fig. 8G).

Specification of FCs in the VM depends on Alk signaling in response to Jeb secretion from the somatic mesoderm (Englund et al., 2003; Lee et al., 2003; Stute et al., 2004). Signaling via Alk activates the Ras/MAPK pathway, translocating the FCM fate-promoting TF Lameduck (Lmd) from the nucleus to the cytoplasm (Popichenko et al., 2013). A similar mechanism has been suggested for a still unknown FC-fate repressor triggering the FC-specific transcriptional program in the VM (Popichenko et al., 2013; Zhou et al., 2019). This transcriptional program remains relatively unexplored with only a few identified targets reported, such as Hand, org-1, kirre, dpp and Alk itself (Englund et al., 2003; Lee et al., 2003; Mendoza-García et al., 2017; Shirinian et al., 2007; Varshney and Palmer, 2006). Although ChIP has been the predominant approach for mapping protein-chromatin interactions, it requires significant amounts of starting material and specific antibodies (Wu et al., 2016). RNA-seq has also been intensely employed for transcriptomic analyses and, although straightforward for cell culture studies, isolation of the VM would be required for its use in identifying Alk transcriptional targets in Drosophila. Therefore, in our efforts to identify novel Alk transcriptional targets in the VM, we employed TaDa, which allows genome-wide RNA PolII occupancy to be investigated in the specific tissue of choice (Southall et al., 2013). TaDa requires less starting material and provides cell-type specificity, although resolution can be less accurate compared with RNA-seq and ChIP-seq due to its dependency on frequency of GATC sites in the genome (Marshall et al., 2016).

Our experimental design was based on either activating or inhibiting Alk signaling throughout the VM, followed by TaDa analysis. Comparison of our TaDa dataset with previously published RNA-seq data (NCBI BioProject, PRJEB11879) from cells isolated from the mesoderm suggest our data recapitulated endogenous binding of RNA PolII. Our dataset also agreed with current understanding of Alk signaling and induction of cell fate specification in the trunk VM (Lee et al., 2003; Lorén et al., 2003; Stute et al., 2004), including observed differential expression of previously identified Alk transcriptional targets, such as Hand, org-1, kirre and dpp. However, these FC-enriched targets were not the most significantly expressed genes within our TaDa dataset, perhaps reflecting lower levels or smaller temporal windows of active transcription that may additionally be complicated by differentially stable mRNA or protein products. Taken together, a combination of different analyses supported our approach as replicating transcriptional events in the VM and led us to validate of TaDa-identified genes in the VM as targets of Alk-driven signaling events.

A number of differentially expressed genes were validated by in situ hybridization during embryogenesis. mRNA was indeed visualized in the visceral musculature, and VM expression was confirmed in our HandC-GFP scRNA-seq dataset for fax, Sumo and CG11658. Expression of fax was observed in the VM and CNS, as previously reported (Hill et al., 1995). Interestingly, Fax has been identified in a screen for diet-regulated proteins in the Drosophila ovary. Insulin signaling in response to diet promotes activation of the Ribosomal protein S6 Kinase (S6K), which drives fax expression (Hsu and Drummond-Barbosa, 2017; Su et al., 2018). Notably, Alk modulates insulin signaling in the brain during nutrient restriction (Cheng et al., 2011; Okamoto and Nishimura, 2015), making Fax interesting for further study. Another interesting candidate is Sumo, which encodes the Drosophila SUMO-1 homologue (Abed et al., 2018). Sumo is not exclusively expressed in the VM, but is rather expressed maternally and ubiquitously throughout the embryo (Shigenobu et al., 2006). Functional studies have identified a role for Sumo in the post-translational modification of several TFs, as well as in modulation of signaling in the fly (Huang et al., 2011; Sánchez et al., 2010). Overall, our TaDa analysis identified numerous uncharacterized genes, and further investigation will be crucial to decipher their role in the developing visceral muscles.

One interesting uncharacterized target was Kah, which encodes a Snail family TF. Kah overexpression in the thorax has been reported to block development of thoracic bristles, revealing a potential to drive changes in cell identity (Singari et al., 2014). We were able to validate Kah as an Alk target locus in the embryonic VM, with differences in Kah expression when Alk signaling was either blocked or activated. However, we also noted Alk-independent Kah transcription in the early VM in addition to the Alk-modulated transcription that is reminiscent of the VM expression reported for org-1 (Schaub and Frasch, 2013). Currently, the TFs downstream of Alk that regulate Kah transcription are unknown, although this will be interesting to study in the future.

Alk signaling in the VM drives FC specification via Ras/MAPK pathway activation, leading to the transcription of FC-specific genes such as Hand, org-1, kirre, dpp and also Kah (Englund et al., 2003; Lee et al., 2003; Shirinian et al., 2007; Stute et al., 2004; Varshney and Palmer, 2006). Loss of Hand, org-1 and dpp does not alter FC specification in the VM, suggesting a complex temporal regulation that assures FC specification and eventually formation of visceral muscles (Hursh et al., 1993; Masucci and Hoffmann, 1993; Schaub and Frasch, 2013; Varshney and Palmer, 2006). Our characterization of Kah mutant alleles suggest that, similar to Hand and org-1, Kah is dispensable for VM FC specification, although formally Kah could be responsible for FC-specific transcriptional changes of yet unidentified targets.

Kah mutants exhibit defects in midgut constriction formation. A number of players are implicated in this event, such as Wg, Dpp, Ultrabithorax (Ubx), Pnt, Extra macrochaetae (Emc) and Org-1 (Bilder et al., 1998; Ellis et al., 1990; Müller et al., 1989; Panganiban et al., 1990; Reuter et al., 1990; Schaub and Frasch, 2013). Interestingly, Alk signaling activity is important for VM Dpp expression and maintenance of Org-1 in FCs (Popichenko et al., 2013; Shirinian et al., 2007). Although it is difficult to define a role for Alk in VM events post FC specification, we show that Alk may indeed play a role in later midgut development. We also noted that inappropriate activation of Alk signaling via Jeb expression results in a range of later gut defects. Thus, perturbation of Alk signaling appears to disrupt later events in midgut constriction. However, whether this is a consequence of earlier, Alk-dependent, visceral FC specification or refers to an independent role of Alk signaling in the VM is unclear. We were particularly interested in Pnt, as this ETS domain TF has been reported by the FlyBi-project (http://flybi.hms.harvard.edu/) to bind Kah in high-throughput Y2H (Thurmond et al., 2019), and pnt mutant embryos also exhibit a midgut constriction phenotype (Bilder et al., 1998). Like Kah, Pnt is not required for VM FC specification (Zhou et al., 2019). The observation of low penetrance midgut constriction defects in transheterozygous Kah^ΔATG/pnt^Δ88 embryos suggest that Kah and Pnt may function together in this process, which is further supported by live imaging, through which we observed stronger midgut constriction phenotypes in Kah, pnt^Δ88 double mutants than in pnt^Δ88 alone. Employing publicly available Kah-ChIP datasets (Roy et al., 2010), we could define a Kah-binding motif similar to that of Snail. Analysis of publicly available Pnt-ChIP datasets (Roy et al., 2010) identified numerous targets of both Kah and Pnt, including Kah itself, nej, put, Mad and Ras85D. Interestingly, although several components of the Dpp signaling appear to be misregulated in Kah mutants, Wg and dpp expression appear normal and robust pMAD activity is observed in both Kah^ΔATG and Kah^ΔZnF mutants. Earlier work reported that Wg and dpp expression are also normal in the VM of pnt mutant embryos (Bilder et al., 1998). It is likely that as yet unidentified players function downstream of Kah in this process, and we have not yet been able to identify a key component downstream of Kah that could explain the underlying molecular mechanisms. It is also important to note that, although in this work we have focused on a role in the VM, Kah is also expressed in the embryonic SM and CNS. Our analysis of Kah mutant RNA-seq together with Kah-ChIP and Pnt-ChIP datasets identifies candidates to be focused on in future studies.

The TaDa approach successfully allowed us to identify transcriptional targets of Alk signaling in the developing mesoderm, including the transcriptional regulator Kahuli described here. Many of these targets are currently uncharacterized and future studies should allow their function(s) in the VM to be elucidated. Our in-depth study of Kah highlights a role for this TF in later visceral musculature development, where it appears to work in concert with other factors, including Pnt, to regulate midgut constriction. Combined ChIP and RNA-seq analyses highlights a group of interesting and largely uncharacterized genes, which should shed further light on the midgut constriction process.

LacZ or GFP balancer chromosomes were used to distinguish progeny of crosses. The following stock lines were obtained from the Bloomington Drosophila Stock Center (BDSC): TM3 Sb Ubx-lacZ (#9120), Df(3L)BSC362 (Kah deficiency, #24386), Df(3L)Exel8065 (Kah deficiency, #7564), Kah^f06749 (PiggyBac insertion mapped to the second intron of Kah, #19006), Kah-GFP.FPTB (#64829), twi.2xPE-Gal4 (#2517) and Df(2R)BSC199 (jeb deficiency, #9626). Additional stocks used in this study were: UAS-LT3-NDam-Pol II (Southall et al., 2013), rP298-lacZ (Nose et al., 1998), which is an enhancer trap in the kirre locus (Ruiz-Gómez et al., 2000), HandC-GFP (Sellin et al., 2006), bap3-Gal4 (Zaffran et al., 2001), UAS-jeb.V (Varshney and Palmer, 2006), UAS-Alk.EC.MYC (Bazigou et al., 2007) (Alk extracellular domain that functions as dominant negative, here referred to as UAS-Alk.DN), jebweli (Stute et al., 2004) and Alk¹⁰ (Lorén et al., 2003).

Twi.2xPE-Gal4 and bap3-Gal4 lines were used to test Dam-Pol II toxicity and further RNA-Pol II profiling. Embryos were collected over a 4 h period and aged at 25°C to stages 10-13, followed by dechorionation in 2% sodium hypochlorite solution for 2 min and subsequent washing steps in PBS. A total of 50 µl embryos per sample was used as starting material. Genomic DNA was extracted (QIAGEN DNeasy Blood and Tissue kit, #69504) and methylated DNA was processed and amplified as previously described (Choksi et al., 2006; Sun et al., 2003), with the following modifications. Upon verification of non-sheared gDNA, the DpnI digestion was set up in 50 µl. After overnight DpnI digestion, the DNA was purified (QIAGEN MinElute PCR purification kit, #28004) into 30 µl of deionized water, from which 15 µl were used for the adaptors-ligation step. Amplified DNA from experimental and Dam-only embryos was again purified (QIAGEN MinElute PCR purification kit) into 20 µl of deionized water and 200 ng aliquots were run in a 1% agarose gel to verify amplification of different fragments (visualized as a smear from 500 bp to 2-3 kb). Purified PCR products were used for PCR-free library preparation, followed by pair-end sequencing on Illumina HiSeq X Ten platform (BGI Tech Solutions, Hong Kong). Three biological replicates were performed for transcriptional profiling of the visceral mesoderm on each of the experimental genetic backgrounds.

The Drosophila genome (FASTA) and genes (GTF) version r6.21 were downloaded from FlyBase (Gramates et al., 2017; Thurmond et al., 2019) and all GATC regions extracted in BED format using fuzznuc (Rice et al., 2000). The paired FASTQ files from 18 samples (background Dam, Jeb and DN with three replicates each for twi.2xPE-Gal4 and bap3-Gal4) were aligned to the Drosophila genome using bowtie2 (--very-sensitive-local) (Langmead and Salzberg, 2012). Sambamba (merge) (Tarasov et al., 2015) was used to combine replicates and the log-fold changes between DN/Jeb and Dam, obtained using bamCompare (--centerReads --normalizeTo1x 142573017 --smoothLength 5 -bs 1) from deepTools (Ramírez et al., 2014). Counts of reads mapped on edge to GATC fragments were generated using a script (GATC_mapper.pl) from DamID-Seq pipeline (Maksimov et al., 2016). RNA-seq data from a public dataset (PRJEB11879) was used to quantify the expression of genes (only 6-8 h mesoderm samples used). The GATC level counts were converted to gene level counts using intersectBed from Bedtools (Quinlan, 2014) and compared against the gene expression (only background Dam samples) at TPM level. GATC sites were merged into peaks following the methods prescribed in a previous study (Tosti et al., 2018). In brief, logFC for individual GATCs were generated using Limma (Jeb versus Dam and DN versus Dam) (P<1e-5) and the GATC sites were merged into peaks based on median GATC fragment distance in the Drosophila genome using mergeWindows and combineTests function from the csaw package (Lun and Smyth, 2016). The peaks were assigned to overlapping genes and filtered for FDR at 0.01. The final results were taken only for the DN versus Dam comparison. The Jeb samples had low rate of alignment, hence Jeb versus Dam is only used as a visual confirmation of the DN versus Dam peaks at specific locations. Enrichment of TFs in the peaks generated was performed by using Fisher test against list of published Drosophila TFs (Kudron et al., 2018). Enrichment for GO and KEGG terms for the genes assigned to significant peaks was performed using WebGestalt (Liao et al., 2019). All statistical analysis was performed in the R programming environment.

Embryos were collected on apple juice agar plates and aged to stages 10-13 (as confirmed by microscopic visualization of a small fraction). Embryos were dechorionated in 2% sodium hypochlorite solution for 2 min and washed in ice-cold PBS. Subsequently, embryos were incubated in dissociating solution (1 mg/ml trypsin, 0.5 collagenase I, 2% BSA) for 1 h and vortexed every 10 min, after which the reaction was stopped by addition of 10 volumes of ice-cold PBS. Dissociated cell solution was sieved through 70 µm and 40 µm cell strainers to remove cell clumps. Dead cells or debris from the dissociated samples were removed using the EasySep Dead Cell Removal (Anexin V) Kit (STEMCELL, 17899) according to the manufacturer's guidelines. The remaining cells were respectively labelled with aqua-fluorescent reactive dye (dying cells) and calcein violet AM (living cells) using the LIVE/DEAD Violet Viability/Vitality Kit (Molecular Probes, L34958) under manufacturer's guidelines. Finally, each sample was washed twice in PBS/2% fetal bovine serum and resuspended in 500 µl PBS/2% fetal bovine serum. Living cells were enriched using a FACSAria III cell sorter (BD biosciences) based on the LIVE/DEAD staining and, when appropriate, GFP expression driven by the HandC-GFP construct (Sellin et al., 2006). The cells were sorted using an 85 µm nozzle into Eppendorf tubes that had been pre-coated with PBS/2% BSA.

Approximately 2500 sorted cells were directly loaded in sheath fluid onto one lane of a Chromium 10X chip (10X Genomics) and libraries prepared using the normal workflow for Single Cell 3´ v3 libraries (10X Genomics). Libraries were sequenced on the NextSeq 500 platform (Illumina), and the raw format base call file (BCL) sequences were demultiplexed using cellranger mkfastq version 3.1. After read QC, mapping was performed with the Drosophila melanogaster genome using STAR aligner. For analysis, unique molecular identified (UMI) count matrix were imported into the Seurat R toolkit version 3.1. For quality filtering, cells with fewer than 1000 genes and more than 5000 expressed genes were excluded. In addition, cells that expressed more than 25% mitochondrial genes were removed. Subsequent count normalization, scaling, feature selection, clustering (PCA) and dimensionality reduction (UMAP, Uniform Manifold Approximation and Projection) were performed using Seurat (Stuart et al., 2019) and Scanpy (Wolf et al., 2018). After preprocessing, 1055 and 888 cells with a total of 11,029 and 10,602 RNA features remained, for the whole embryo and HandC-GFP scRNA-seq datasets, respectively.

To identify relationships and correlations between clusters, hierarchical clustering and correlation matrices were calculated using Pearson correlation (Scanpy). Based on canonical markers and previously known markers, the cellular heterogeneity of the whole embryo with 13 cell types [epidermis, somatic mesoderm, early trunk mesoderm, amnioserosa (AS) and endoderm, neuroblasts, neurons, hemocytes, endoderm, visceral mesoderm, fat body, ectoderm, trachea and glia] and the HandC-GFP scRNA-seq with five clusters (VM founder cells, early visceral muscle, cardiac mesoderm, VM proliferating cells and late visceral muscle) were determined. Clusters were identified using the Louvain algorithm (Blondel et al., 2008) and visualized by UMAP projection. Feature plots, violin plots, matrix plots, heatmaps and dot plots were plotted for visualizing the marker genes expression and percentage of cell distribution. Gene Ontology (GO) term enrichment was performed on genes commonly upregulated in both Kah^ΔATG and Kah^ΔZnF mutants using WebGestalt (PANTHER functional database) (Liao et al., 2019). Markers employed for determination of VM proliferating cells were defined by gene ontology analysis using g:Profiler (Biological Process) (Raudvere et al., 2019).

Deletion (Kah^ΔATG and Kah^ΔZnF) and endogenous tagged (Kah^Cterm.OLLAS) Kah alleles were generated using the CRISPR/Cas9 system. CRISPR target sites were identified and evaluated using flyCRISPR Optimal Target Finder tool (Gratz et al., 2015). Single guide RNA (sgRNA) target sequences (sequences available in Table S4) were cloned into pU6-BbsI-chiRNA vector (Addgene plasmid #45946;Gratz et al., 2013) and injected into vasa-Cas9 (BDSC, #51323) embryos (BestGene). For Kah^Cterm.OLLAS a donor construct was added to the injection mix. Injected animals were crossed to third chromosome balancer flies and resulting progeny PCR screened for positive deletion/insertion events. Positive candidates were confirmed further by Sanger sequencing (Eurofins Genomics).

Endogenously tagged Kah^Cterm.OLLAS was generated using CRISPR/Cas9-induced homology-directed repair (HDR) at the Kah C-terminal. Three CRISPR sgRNA sequences (sequences available in Table S4) were used. One sgRNA was designed to target upstream to the Kah stop codon and the other two to target directly the Kah stop codon. In addition, a DNA donor cassette was synthesized (Integrated DNA Technologies) before Gibson assembly cloning into the pBluescript II KS[-] HDR plasmid. This donor cassette codes for the remaining part of the Kah C-terminal followed by six glycines (linker), three copies of the OLLAS-tag and a TAG stop codon; flanked by two homology arms (495 bp upstream and 500 bp downstream of the respective Cas9 cutting sites, with codon optimized target sequences).

Publicly available dm6 aligned Kah ChIP-seq data input libraries (ENCSR161YRO and ENCSR664RUV) and the raw format FASTQ sequences of Pnt ChIP-seq (ENCSR997UIM and ENCSR249WKC) were retrieved from modEncodeID. For Pnt ChIP-seq data, the base quality of each sequenced read was assessed using the FASTQC program. The reads were aligned to the Drosophila melanogaster (BDGP6) reference genome using Bowtie2. Owing to the ambiguity of reads that align to multiple locations across the genome, only reads that uniquely mapped were considered for subsequent analysis. Post-alignment processes were performed with samtools and BEDtools, and Homer suite v4.1 program ‘findpeaks’ (Heinz et al., 2010) with default TF finding parameters (-style factor) used for peak calling. Resulting peaks from each replicate were annotated using ChIPseeker v1.2 (Yu et al., 2015) and merged to be used as input for genome-wide motif enrichment scanning using the ‘findMotifsGenome.pl’ script from the Homer suite. Regions of 50 bp and 200 bp around the peak center were analyzed for motif enrichment.

Embryos were collected at 25°C (5-16 h after egg laying) and dechorionated in 2% sodium hypochlorite solution for 4 min. After subsequent washing with embryo wash (0.8% NaCl and 0.05% Triton X-100) and H₂O, embryos were stored at −80°C. One embryo collection for three biological replicates per genotype was obtained. RNA-extraction was carried out according to the manufacturer's protocol (Promega ReliaPrepTM RNA Tissue Miniprep System, REF-Z6111). Total RNA was measured using NanoDrop OneC (Thermo Scientific) for its concentration and RNA integrity was checked using gel electrophoresis. Then 9-15 µg of total RNA/biological replicate was shipped to Novogene for sequencing. Prior to making the library, samples were reassessed for quality with the Agilent 2100 Bioanalyzer system. Sequencing was performed on an Illumina platform and paired-end reads were produced. Over 40 million reads/genotype were generated and mapped to the genome at a rate of over 96%. Drosophila melanogaster (ensemble bdgp6_gca_000001215_4 genome assembly) was used. HISAT2 algorithm for alignment and DESeq2 R package (Anders and Huber, 2010) for differential gene expression was used. Subsequent analyses were performed with GraphPad Prism 9. Fold change ≥1.5 and ≤−1.5 (log2FC≥0.59 and ≤−0.59) for up- and downregulated genes, respectively, and P_adj≤0.05 were used for statistical significance.

Embryos were fixated and stained as described previously (Müller, 2008). Primary antibodies used were: guinea pig anti-Alk (1:1000; Lorén et al., 2003), guinea pig anti-Jeb (1:1000; Englund et al., 2003), rabbit anti-Alk (1:750; Lorén et al., 2003), chicken anti-β-galactosidase (1:200; Abcam ab9361), mouse anti-Fasciclin3 (1:50; DSHB 7G10), rabbit anti-GFP (1:500; Abcam ab290), chicken anti-GFP (1:300; Abcam ab13970), mouse anti-Wg (1:50; DSHB 4D4), rat anti-OLLAS (1:200, pre-absorbed on w1118 embryos; Abnova), rabbit anti-Org-1 (1:1000; Mendoza-García et al., 2017), sheep anti-digoxygenin-AP fab fragment (1:4000, Roche), rabbit anti-phospho-Smad1/5 (41D10) (1:500; Cell Signaling Technologies 9516). Alexa Fluor-conjugated secondary antibodies were from Jackson Immuno Research. Embryos were dehydrated in an ascending ethanol series before clearing and mounting in methylsalicylate. Images were acquired with a Zeiss LSM800 confocal microscope or Axiocam 503 camera, processed and analyzed employing Zeiss ZEN2 (Blue Edition) imaging software. HandC-GFP positive VM nuclei quantification (Fig. 7F) was performed in triplicate for each genotype (n=10 per replicate, in total n=30 per genotype) with ImageJ software (Schneider et al., 2012). Raw images were converted into binary format and nuclei were quantified using 3D nuclei counter package (Bolte and Cordelieres, 2006). To identify statistical differences between Kah mutants and controls (w¹¹¹⁸), t-test analysis was performed.

Overnight collection of embryos was performed at 25°C followed by dechorionation in 2% sodium hypochlorite solution for 4 min. After subsequent washing with embryo wash solution (0.8% NaCl and 0.05% Triton X-100) and H₂O, stage 14-early 16 embryos were sorted manually with a stereo/epi-fluorescence microscope. Afterwards embryos were transferred into 96-well plates and live imaged with a Zeiss Cell Discoverer 7 at 1 min intervals with time series setting. Further processing was performed with FIJI ImageJ. For quantification of the midgut phenotype of pnt^Δ88 and Kah^ΔATGpnt^Δ88 double mutant embryos, manual scoring of live imaged embryos (stage 15 onward) was performed. Scoring was based on two criteria: (1) 1st midgut constriction started but not completed; and (2) no apparent constriction. Results are shown as a percentage of embryos scored.

For in situ hybridization, fragments of the respective CDS were PCR amplified from genomic DNA, cloned into the dual promoter pCRII-TA vector (ThermoFisher, K207040) and used as a template to generate DIG-labeled in situ probes with SP6/T7 polymerases (Roche, 10999644001). Whole-mount in situ hybridization was carried out according to Lécuyer et al. (2008), with modifications adapted from Pfeifer et al. (2012). Samples were mounted using in situ mounting media (Electron Microscopy Sciences). Images were acquired with a Zeiss Axio Imager.Z2 microscope, processed and analyzed using Zeiss ZEN2 (Blue Edition) imaging software.

We thank A. Brand, M. Takeichi, M. Frasch, A. Paululat, A. Nose, M. Ruiz-Gomez and J. Bateman for sharing fly stocks and reagents, and B. Paul and C. Schaub for valuable help and suggestions. We acknowledge Bloomington Drosophila Stock Center (NIH P40OD018537) for fly stocks used in this study. The Fasciclin 3, Wg and Antp antibodies developed by C. Goodman, S. M. Cohen and D. Brower, respectively, were obtained from the Developmental Studies Hybridoma Bank, created by the NICHD of the NIH and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199465.