All epithelial components of the inner ear, including sensory hair cells and innervating afferent neurons, arise by patterning and differentiation of epithelial progenitors residing in a simple sphere, the otocyst. Here, we identify the transcriptional repressors TBX2 and TBX3 as novel regulators of these processes in the mouse. Ablation of Tbx2 from the otocyst led to cochlear hypoplasia, whereas loss of Tbx3 was associated with vestibular malformations. The loss of function of both genes (Tbx2/3cDKO) prevented inner ear morphogenesis at midgestation, resulting in indiscernible cochlear and vestibular structures at birth. Morphogenetic impairment occurred concomitantly with increased apoptosis in ventral and lateral regions of Tbx2/3cDKO otocysts around E10.5. Expression analyses revealed partly disturbed regionalisation, and a posterior-ventral expansion of the neurogenic domain in Tbx2/3cDKO otocysts at this stage. We provide evidence that repression of FGF signalling by TBX2 is important to restrict neurogenesis to the anterior-ventral otocyst and implicate another T-box factor, TBX1, as a crucial mediator in this regulatory network.
The mammalian inner ear is a highly asymmetric sensory organ that is characterised by an elaborated system of fluid-filled epithelial ducts and chambers. Ventro-medially in this membraneous labyrinth lies the spirally coiled cochlear duct, which is essential for the perception of sound. Dorso-laterally resides the vestibular organ, in which anterior, posterior and lateral semicircular canals and two membraneous sacs, the saccule and the utricle, mediate the perception of angular and linear acceleration, as well as of gravity. In both the cochlea and the vestibular organ, sensory stimuli are converted into electrical impulses by clusters of mechanosensitive hair cells and are transmitted to the brain by sensory neurons of the cochlear and the vestibular ganglia.
Hair cells, their supporting cells, non-sensory epithelial cells as well as sensory neurons of the inner ear all derive from a common precursor tissue, the otic placode, which in mice is established in the ectoderm lateral to the hindbrain around embryonic day (E) 8.5 (reviewed by Wu and Sandell, 2016). This epithelial patch invaginates and pinches off the surrounding ectoderm to form at E9.5 an epithelial sphere, the otocyst. Within the next 8 days, localised changes of cell shapes, cell migration, proliferation and apoptosis drive morphogenesis of the semicircular canals and the endolymphatic duct from the dorsal and dorso-medial aspect of the otocyst, respectively; of utricle and saccule from the intermediate part; and of the cochlear duct from the posterior-ventral region (Alsina and Whitfield, 2017; Martin and Swanson, 1993; Morsli et al., 1998; Nishikori et al., 1999).
Around E9.0, epithelial cells in the ventral otocyst initiate a neurosensory differentiation programme. From this neurogenic domain, which is rapidly restricted to the anterior subregion of the otocyst until E10.5, neuroblasts delaminate and coalesce to form the statoacoustic ganglion (SAG), which later subdivides and innervates the auditory and vestibular inner ear components (Carney and Silver, 1983; Ma et al., 1998; Raft et al., 2004). Remaining cells of the proneurosensory domain, as well as cells of the neighbouring prosensory domain, segregate to different regions of the inner ear where they eventually differentiate into mechanosensitive hair cells and supporting cells of the vestibular and cochlear sensory patches (reviewed by Raft and Groves, 2015; Wu and Sandell, 2016).
Spatially confined morphogenetic and neurosensory differentiation programmes rely on prior patterning of the otocyst along three perpendicular axes. WNTs and BMPs from the dorsal hindbrain, and SHH from the basal plate of the hindbrain and the underlying notochord provide asymmetric cues to generate opposing signalling gradients along the dorso-ventral and medio-lateral axes of the otocyst (Hatch et al., 2007; Ohta et al., 2016; Pirvola et al., 2000; Riccomagno et al., 2002, 2005). Combinatorial signalling activities confer regionalised expression of transcription factors that regulate distinct morphogenetic programmes in subregions of the otocyst (reviewed by Ohta and Schoenwolf, 2018). Polarisation along the anterior-posterior axis is triggered by retinoic acid (RA) from paraxial mesoderm at the posterior edge of the forming otocyst (Bok et al., 2011). In the posterior otocyst, high levels of RA induce the expression of the T-box transcription factor TBX1, which suppresses neurogenesis (Bok et al., 2011; Raft et al., 2004). Commitment to the neuronal lineage within the anterior-ventral otocyst is regulated by the basic helix-loop-helix (bHLH) transcription factor NEUROG1, which acts upstream of NEUROD1 and DLL1 (Ma et al., 2000, 1998). DLL1 limits the number of neuronal precursors by NOTCH-dependent lateral inhibition (Brooker et al., 2006; Daudet et al., 2007), and NEUROD1 controls the subsequent delamination and differentiation of neuroblasts (Kim, 2013; Kim et al., 2001; Liu et al., 2000).
TBX2 and TBX3 are two closely related members of a T-box transcription factor subfamily that is distinct from that of TBX1 (Agulnik et al., 1996). Both proteins are expressed at many sites of the developing vertebrate embryo, where they regulate diverse cellular processes in early organogenesis as transcriptional repressors (reviewed by Papaioannou, 2014). Based on previous reports that these genes are also expressed in the early mouse otocyst (Bollag et al., 1994; Mesbah et al., 2012), we used a conditional gene-targeting approach to analyse their function in murine inner ear development. Here, we show that Tbx2 and Tbx3 are individually and combinatorially required for inner ear morphogenesis. We trace the developmental onset of these changes to the early otocyst, and characterise the cellular and molecular alterations at this stage. We provide evidence for a role of TBX2 in the anterior-ventral restriction of otic neurogenesis by repressing FGF signalling and maintaining Tbx1 expression in the posterior-ventral otocyst.
Tbx2 and Tbx3 show spatially restricted but partially overlapping patterns of expression in early murine inner ear development
Previous work reported expression of Tbx2 in the E8.5 otic placode (Barrionuevo et al., 2008), and of Tbx2 and Tbx3 in the E9.5 otocyst (Bollag et al., 1994; Mesbah et al., 2012). However, a spatially and temporally resolved profile of (co-)expression of the two genes in the early development of the murine inner ear was not reported. RNA in situ hybridisation analysis confirmed that Tbx2 expression commences in the otic placode at E8.5 (Fig. 1A). At E9.5, Tbx2 was expressed in the epithelium of the entire otocyst with exception of the dorsolateral quadrant (Fig. 1B). From E10.5 to E12.0, expression of Tbx2 was found ventrally in the developing cochlear duct, dorso-medially in the endolymphatic duct and in the central region, the vestibule, but not in the dorso-lateral region, which gives rise to the vertical canal plate, the primordium of the posterior and anterior semicircular canals (Morsli et al., 1998). Otic expression of Tbx3 started at E9.5 in the lateral otocyst (Fig. 1A,B). From E10.5 onwards, expression was confined to the lateral aspects of the developing vestibular system and ventro-medially to the epithelium of the cochlear duct. Three-dimensional (3D) reconstructions from serial sections confirmed unique expression of Tbx2 in the developing endolymphatic duct and in the posterior aspect of the cochlear duct, and of Tbx3 in the dorsal vestibular system whereas expression of the two genes largely overlapped in the central-lateral part of the developing vestibular system and in the anterior cochlear duct (Fig. 1C). Expression of TBX2 and TBX3 protein mirrored the patterns of the mRNA (Fig. 1D). Together, this analysis argues for both unique and combined functions of Tbx2 and Tbx3 in the development of specific subregions of the otocyst.
Individual and combined inactivation of Tbx2 and Tbx3 in the otic epithelium disrupts inner ear morphogenesis
To test the individual and combined function of Tbx2 and Tbx3 in inner ear development, we used a tissue-specific gene inactivation approach with a Pax2-cre transgenic line that mediates recombination in the otic epithelium from E8.5 onwards (Trowe et al., 2011) (Fig. S1A), and floxed alleles of Tbx2 (Tbx2flox) (Wakker et al., 2010) and Tbx3 (Tbx3flox) (Frank et al., 2013). Complete loss of TBX2 and TBX3 protein in the otic epithelium of Pax2-cre/+;Tbx2flox/flox;Tbx3flox/flox (Tbx2/3cDKO) embryos at E9.5 and E10.5 confirmed the validity of this approach (Fig. S1B).
Mice with individual and combined inactivation of Tbx2 and Tbx3 presented in the expected Mendelian ratio at E18.5 (Table S1) allowing for the inspection of inner ear morphology by serial histological sections and subsequent 3D reconstructions of the cavities of all mutant combinations at this stage (Fig. 2A, Fig. S2A). Embryos with an individual deletion of Tbx2 (Pax2-cre/+;Tbx2flox/flox, Tbx2cKO) exhibited a severely shortened cochlear duct, a rudimentary endolymphatic duct, malformation of the vestibular system with dilation of the vestibule and crus commune, loss of the saccular cavity and variably truncated semicircular canals. In Pax2-cre/+;Tbx3flox/flox (Tbx3cKO) embryos, the cochlear and endolymphatic ducts were established normally, but the vestibular system was variably hypo- or dysplastic. In Tbx2/3cDKO embryos, hypodysplasia of the cochlear and endolymphatic duct and of the vestibular system was strongly enhanced compared with single mutants. The membranous labyrinth was severely reduced in size and collapsed. This shows that Tbx2 is uniquely required for morphogenesis of the cochlea and the endolymphatic duct, whereas Tbx2 and Tbx3 redundantly regulate the formation of the vestibular system.
Morphological changes of the inner ear in Tbx2- and/or Tbx3-deficient embryos occur shortly after otocyst formation
To define both the onset and progression of these morphological defects, we analysed inner ear development from E9.5, when the otocyst is formed, to E15.5, when all epithelial ducts of the cochlea and vestibular system are established (Fig. 2B, Fig. S2B). In Tbx2cKO embryos, the first phenotypic changes were detected at E10.5 as a thickening of the otic epithelium in the ventro-lateral region of the otocyst (Fig. S2B, brackets). In 3D reconstructions, the protrusion of the endolymphatic duct at the dorso-medial side was much shorter than in controls. At subsequent stages, the endolymphatic duct remained short and did not separate from the vestibule. The cochlear duct underwent minimal elongation and no coiling. Formation of the posterior semicircular canal was delayed and separation of the vestibule and saccular cavity failed. Inner ears of Tbx3cKO embryos appeared normal until E11.5, but exhibited delayed semicircular canal morphogenesis at E12.5, and lacked the posterior semicircular canal at E15.5. In Tbx2/3cDKO inner ears, phenotypic changes started at E10.5, mirroring the findings in Tbx2cKO embryos. Unlike in Tbx2cKO embryos, the endolymphatic and the cochlear duct were absent at E11.5 and subsequent stages. The size and shape of the otocyst remained largely unchanged at E12.5. At E15.5, the cavity was collapsed. These findings point to onset of morphological changes in Tbx2cKO and Tbx2/3cDKO inner ears at E10.5, defining a requirement of Tbx2 and Tbx3 shortly after the otocyst has formed.
Tbx2/3cDKO embryos exhibit ectopic apoptosis in the dorso-lateral and ventral regions of the otocyst
We next investigated whether these morphological changes are preceded and/or accompanied by alterations in cell survival and/or proliferation in the otic epithelium. A bromodeoxyuridine (BrdU) incorporation assay revealed normal cell proliferation in four quadrants of Tbx2/3cDKO otocysts at E10.5 (Fig. 3A,B, Table S2). In contrast, terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) and anti-CASPASE-3 staining detected altered patterns of apoptosis (Fig. 3C, Fig. S3). A small cluster of apoptotic cells at the ventral-most tip of the otocyst was present in control and Tbx3cKO embryos at E10.5 but not in Tbx2cKO and Tbx2/3cDKO embryos (arrowheads in Fig. 3C). In the latter genotypes, ectopic programmed cell death occurred in the ventro-lateral and dorso-lateral region at this stage (arrows in Fig. 3C). At E11.5, Tbx2cKO and Tbx2/3cDKO embryos exhibited markedly increased apoptosis in the ventral-most area of the otocyst, and in cells and ganglion-like structures located in the adjacent periotic mesenchyme (arrows in Fig. 3C, Fig. S3).
Dorso-ventral and medio-lateral regionalisation of Tbx2/3cDKO otocysts is partly disturbed
To identify in an unbiased fashion molecular changes that correlate or even account for the observed cellular and morphological defects in Tbx2/3cDKO inner ears, we performed microarray-based transcriptional profiling using RNA from four independent pools of dissected otocysts of E10.5 Tbx2/3cDKO embryos and their Cre-negative littermates as controls. Using Qlucore Omics Explorer software, we found a total of 2860 differentially expressed genes with q<0.05 (Table S3); additional thresholds for signal intensity (INT≥100) and fold change (FC≥2) resulted in 234 robustly upregulated and 169 robustly downregulated genes in Tbx2/3cDKO otocysts (Fig. 4A, Tables S4 and S5).
Functional annotation using the DAVID software tool (david.ncifcrf.gov) revealed strongest enrichment of the gene ontology (GO) term ‘inner ear morphogenesis’ in the group of 169 genes with reduced expression (Fig. 4B, Table S6). Among the top 20 downregulated transcripts was the transcription factor gene Otx1 (−5.4), which is required for lateral semicircular canal formation and cochlear morphogenesis (Morsli et al., 1999), and Smpx (−11.4), loss of which leads to hearing impairment in humans (Huebner et al., 2011) (Fig. 4C). Manual inspection of the list of downregulated genes uncovered additional regulators of inner ear morphogenesis, including Otx2 (−4.9) (Miyazaki et al., 2006; Morsli et al., 1999), Hmx2/3 (−4.1/−3.1) (Wang et al., 2001, 2004), Gata2/3 (−3.0/−2.2) (Haugas et al., 2010; Lilleväli et al., 2006), Gbx2 (−2.9) (Lin et al., 2005; Miyazaki et al., 2006), Tbx1 (−2.7) (Raft et al., 2004; Vitelli et al., 2003), Dlx5 (−2.3) (Robledo and Lufkin, 2006), Fgf10 (−1.7) (Lilleväli et al., 2006; Pauley et al., 2003) and Lmo4 (−1.6) (Deng et al., 2010) (Fig. 4D).
RNA in situ hybridisation analysis uncovered that regionalised expression of genes from the top 20 list as well as of further selected candidates was differentially affected in E10.5 Tbx2/3cDKO otocysts (Fig. 4E). Expression of Smpx and Sstr1 in the dorso-medial region was lost, and that of Gbx2, Pax8 and Wnt2b was strongly reduced. Aldh1a2 and Hmx3 were no longer expressed in the dorso-lateral region, and Gata3 and Lmo4 were downregulated and their expression domain was smaller. The ventral expression limit of the lateral Dlx5 domain was shifted dorsally. Otx2 expression was completely abolished from the ventro-lateral region, and the expression domain of Otx1 in the lateral otic epithelium was reduced. Cldn22 was no longer transcribed in the ventral otocyst. Gata3 expression in the ventral tip of the otocyst was reduced, and expression of Gdf6, Fgf10 and Tnni1 was not detected in this region. Other transcription factor genes implicated in dorso-ventral patterning of the otocyst (Dach1, Eya1, Pax2, Six1) (Ozaki et al., 2004; Torres et al., 1996; Xu et al., 1999; Zheng et al., 2003), were unchanged in the microarray and exhibited minor (Pax2 was ventrally expanded) or no changes (Dach1, Eya1, Six1) in their expression pattern at E10.5 (Fig. 4F). Together, these findings indicate that the dorso-ventral and medio-lateral axes are established but that regionalised morphogenetic programmes along these axes are partly compromised upon combined loss of Tbx2 and Tbx3 in the otic epithelium.
Genes associated with neuronal differentiation are upregulated in Tbx2/3cDKO otocysts
As Tbx2 and Tbx3 encode transcriptional repressors (Brummelkamp et al., 2002; Carreira et al., 1998; Jacobs et al., 2000; Lingbeek et al., 2002), it is likely that the set of genes with increased expression in Tbx2/3cDKO otocysts (Table S5) better reflects their primary regulatory function. Functional annotation analysis of upregulated transcripts showed an enrichment of genes associated with the terms ‘neurogenesis’ and ‘regionalisation’ (Fig. 5A, Table S7). Many of these ‘neuronal’ genes are among the top 40 upregulated genes, including En2 (+17.1), Sox3 (+5.2), Fgf8 (+4.6) and Nefm (+4.3) (Fig. 5B). Other highly upregulated neuronal genes were Pou4f1 (+3.1), Insm1 (+2.3), Nhlh2 (+2.2) (Fig. 5C). Additional genes with known function or expression in otic neurogenesis were found among significantly upregulated genes that had a FC<2, including the neurogenic marker genes Dll1 (+1.5) and Neurod1 (+1.7), Heyl (+1.8), Nhlh1 (+1.6) and Tubb3 (+1.7) (Jalali et al., 2011; Ma et al., 1998) (Fig. 5C, Table S3).
Using RNA in situ hybridisation analysis, we validated expression of the top 40 upregulated genes as well as of selected candidates for which we were able to obtain probes (Fig. 5D). Ten of the genes showed ectopic expression in Tbx2/3cDKO otocysts: Cd83, En2, Lamc2, Papss2 and Sox3 in the ventral region; Ctxn3 and Kcns3 in the ventro-lateral region; and Insm1, Prkaa2 and Sowaha in single cells in the ventral region as well as in the adjacent mesenchyme. Expression of Nhlh2 in neuroblasts was unaffected. Dorso-lateral expression of Ntn1 and medial expression of Heyl was expanded ventrally and ventro-laterally, respectively, in mutant otocysts. Celf3, Dll3, Kcnmb2, Pou4f1 and Tmeff2 were expressed in the SAG in both control and Tbx2/3cDKO embryos. Many of these genes (Celf3, Dll3, Heyl, Insm1, Nhlh2, Pou4f1, Sox3) were also transcribed in the neural tube, confirming their possible involvement in neurogenesis. Collectively, these results indicate that neurogenesis occurs ectopically in the ventral region of Tbx2/3cDKO otocysts.
The neurogenic domain is posteriorly expanded in Tbx2-deficient otocysts
In wild-type embryos, neurogenesis commences in the entire ventral aspect of the otocyst around E9.5 but becomes rapidly restricted to an anterior subregion at E10.5 (Raft et al., 2004). To judge whether this regionalisation process is compromised in Tbx2/3cDKO embryos, we analysed the expression of genes involved in this programme on transverse sections along the anterior-posterior axis of the otocyst at E9.5 and E10.5. At E9.5, expression of the proneurosensory markers Sox2 and Lnfg, and of the neurogenic marker Neurog1 in the ventral region along the anterior-posterior axis was unaffected as was the restriction of Tbx1 to the posterior-lateral otocyst (Fig. S4).
At E10.5, Neurog1 and its targets, Dll1 and Neurod1 (Ma et al., 1998) were found in the anterior ventro-lateral otocyst as well as in the periphery of the adjacent SAG in control embryos; Nefm, a gene transcribed in maturing neurons (Romand et al., 1990), was expressed in the centre of the SAG (Fig. 6A). In Tbx2/3cDKO otocysts, expression of Neurog1, Dll1 and Neurod1 was expanded posterior-ventrally into the non-neurogenic region of the otocyst, and was also observed in cells underneath the epithelium. Ectopic ganglion-like structures were found in the adjacent periotic mesenchyme (Fig. 6A, arrows), of which some were Nefm positive. 3D reconstructions confirmed posterior-ventral expansion of resident and delaminating neuroblasts (Neurod1) (Fig. 6B). However, only a subset of these ectopic neuroblasts managed to differentiate further (Nefm) as the slightly lateralised SAG was not expanded posteriorly (Fig. 6C). Importantly, posterior expansion of the neurogenic domain occurred in Tbx2cKO but not in the Tbx3cKO otocysts (Fig. S5), indicating that TBX2 accounts for the anterior restriction of the otic neurogenic programme after E9.5.
Tbx1, the major inhibitor of otic neurogenesis (Raft et al., 2004), was expressed in the entire otocyst except for the anterior-ventral neurogenic domain of control and Tbx3cKO embryos at E10.5. In Tbx2cKO embryos, Tbx1 expression was absent from the ventral aspect of the otocyst along the complete anterior-posterior axis. In Tbx2/3cDKO mutants, expression of Tbx1 was additionally abolished from the anterior-dorsal otocyst (Fig. S6A, Fig. 6D). In Tbx1-deficient otocysts, expression of Tbx2 expanded into a small anterior-dorsal region but was otherwise unaffected; Tbx3 was lost in the posterior-medial region (Fig. S6B). Hence, ventral expression of Tbx1 depends on Tbx2 whereas Tbx2/Tbx3 expression gets (minor) input from Tbx1 function.
Ectopic FGF signalling contributes to the posterior expansion of the neurogenic domain in Tbx2/3cDKO otocysts at E10.5
Our transcriptional profiling revealed increased expression of Fgf8 (+4.6) and of members of its synexpression group: FGF8 effectors Etv1 (+1.5), Etv4 (+1.6), Etv5 (+1.5), and counter-regulators Dusp6 (+1.8) and Il17rd (+1.4) in Tbx2/3cDKO otocysts (Fig. 5B, Table S3). Because FGF8 promotes commitment of otic epithelial cells to the neuronal lineage by inducing Sox3 (increased 5-fold here) and downstream Neurod1 expression in the chick (Abelló et al., 2010), we wished to explore changes of the spatial pattern of Fgf8 and Sox3 (Fig. 7A). In control embryos, Ffg8 was restricted to the periphery of the SAG in the anterior region; Sox3 was not expressed. In Tbx2/3cDKO embryos, we detected ectopic expression of both genes in the ventral otocyst along the anterior-posterior axis. Importantly, this ectopic expression correlated with strongly enhanced FGF signalling, as indicated by ectopic posterior expression of Etv4 and Etv5, bona fide target genes of this pathway (Raible and Brand, 2001). Notably, expression of another FGF ligand gene, Fgf3, which is confined to the anterior-lateral (neural) region of the otocyst (McKay et al., 1996), occurred ectopically in a dorso-medial position at the posterior region of the Tbx2/3cDKO otocyst at this stage (Fig. 7A). Ectopic expression of Fgf8, Etv4 and Sox3 was also found in Tbx2cKO but not in Tbx3cKO otocysts at E10.5, implicating specifically TBX2 in the repression of Fgf8 and FGF signalling (Fig. S7).
In Tbx1KO otocysts, Fgf8 and Sox3 were not ectopically expressed; the anterior-ventro-lateral expression of Fgf3 was expanded posteriorly; and Etv4 and Etv5 expression occurred ectopically in the posterior-medial region (Fig. S8). These results suggest that loss of Tbx1 in Tbx2cKO and Tbx2/3cDKO otocysts does not account for ectopic posterior-ventral activation of Fgf8/Sox3 expression and FGF signalling.
Because signalling pathways are frequently interconnected in tissue patterning, we also analysed changes of the spatial activity of other pathways important for otocyst patterning by RNA in situ hybridisation of known targets in E10.5 Tbx2/3cDKO otocysts (Fig. S9). Id3, target of BMP signalling (Hollnagel et al., 1999), was lost from the posterior-ventral and downregulated in the dorsal region. Bmp4, a ligand of this pathway, was ectopically expressed in the dorsolateral otocyst, but its expression was lost at ventro-medial sides in the posterior region. Axin2, a target of WNT signalling (Jho et al., 2002), was weakly expanded from the dorso-medial into the ventro-medial region whereas ventral expression of Ptch1, a target of SHH signalling (Ingham and McMahon, 2001), was slightly reduced in this region in Tbx2/3cDKO otocysts. Rarb, a target of RA signalling (Mendelsohn et al., 1991), was found neither in control nor in Tbx2/3cDKO otocysts.
At E9.5, expression of Etv4 and Etv5 indicated FGF signalling in the entire medial region of control otocysts. In Tbx2/3cDKO otocysts, expression of both genes was enhanced and laterally expanded, correlating with ectopic Fgf3 expression in this region. Fgf8 expression was not detected at this stage (Fig. S10A). BMP, WNT, SHH and RA signalling were marginally affected (Fig. S10B). We conclude that ectopic neurogenesis in the posterior-ventral region of E10.5 Tbx2-deficient otocysts coincides with ectopic Fgf8 expression and FGF signalling at this stage.
We used pharmacological inhibition to determine whether increased FGF signalling contributes to ectopic neurogenesis in Tbx2/3cDKO mutants (Fig. 7B,C). Halves of E9.5 embryo heads were cultured in the presence of the highly selective FGFR inhibitor AZD4547 (1 µM in DMSO) (Gavine et al., 2012; Gudernova et al., 2016), or with DMSO vehicle as a control. After 24 h of treatment, expression of Etv4 was abolished in wild-type and Tbx2/3cDKO otocysts, validating the assay in this tissue (Fig. 7B). Upon AZD4547 treatment, expression of Neurog1 was restricted to the anterior-ventral domain in both control and Tbx2/3cDKO otocysts, suggesting that TBX2 restricts neurogenesis to the anterior-ventral otocyst after E9.5 by suppression of posterior-ventral FGF signalling (Fig. 7C).
Tbx2 and Tbx3 exert unique and redundant functions in early inner ear development
The T-box transcription factors TBX2 and TBX3 regulate a diverse set of developmental programmes. Here, we have extended their functional analysis in mouse development and found that they are individually and combinatorially required for the formation of the cochlea and the vestibular organ. Our morphological analysis revealed strong hypoplasia of the cochlea and the endolymphatic duct combined with moderate vestibular defects in Tbx2cKO embryos, hypodysplasia of the semicircular canals in Tbx3cKO embryos and indiscernible cochlear and vestibular structures in Tbx2/3cDKO embryos shortly before birth. These morphological defects were preceded by markedly increased apoptosis in the ventral otocyst of Tbx2cKO and Tbx2/Tbx3cDKO, and in the dorsolateral domain of Tbx2/3cDKO, as well as by ectopic posterior-ventral neurogenesis in Tbx2-deficient otocysts. These individual and combined defects reflect the unique and overlapping sites of expression of the two genes in the early otocyst, i.e. of Tbx2 in the dorsal-medial and ventral region, of Tbx3 in the dorsal region, and of Tbx2 and Tbx3 in the central and ventro-medial region, from which the epithelial ducts and chambers of the membraneous labyrinth arise in a region-specific manner. Similar to other developmental contexts (Aydoğdu et al., 2018; Lüdtke et al., 2016; Mesbah et al., 2012; Singh et al., 2012; Zirzow et al., 2009), Tbx2 and Tbx3 therefore have individual as well as redundant functions in early inner ear development depending on their co-expression patterns.
In humans, heterozygous mutations of TBX2 have been associated with vertebral anomalies, and variable endocrine and T-cell dysfunctions (Liu et al., 2018), and heterozygous mutations of TBX3 have been linked to ulnar-mammary syndrome (Bamshad et al., 1997). In neither case were deficits in hearing and balance reported. Our findings, together with recent reports that the heterozygous loss of several genes, including TBX2, in a microdeletion at 17q23.1q23.2 is associated with sensorineural hearing loss in humans (Ballif et al., 2010; Nimmakayalu et al., 2011) suggest that it might be beneficial to screen individuals with known mutations in TBX2 or TBX3 for subtle defects of inner ear functions.
Tbx2 and Tbx3 exert crucial patterning functions in early inner ear development
Our morphological analysis of midgestation stages revealed that inner ear defects in Tbx2 and Tbx3 single and double mutant embryos manifested around E10.5, indicating that Tbx2/Tbx3 are not involved in formation of the otic placode and the otocyst but are required for the subsequent regionalised outgrowth of the otic epithelium. Although Tbx2 and Tbx3 have been implicated in the maintenance of cell proliferation by direct repression of cell cycle inhibitors in vitro and in vivo (Brummelkamp et al., 2002; Jacobs et al., 2000; Lingbeek et al., 2002; Lüdtke et al., 2013), neither our BrdU assay nor our transcriptional profiling provided support for reduced proliferation as a cause of morphogenetic impairment in Tbx2/3cDKO otocysts. However, we detected a massive increase in apoptotic cells in the dorso-lateral and ventro-lateral aspects of the Tbx2/3cDKO otocyst at E10.5, and in the ventral otocyst at E11.5, arguing that increased apoptosis depletes the epithelial progenitor pool required for localised outgrowth of the epithelial ducts and chambers similar to other mouse mutants with morphogenetic defects in the inner ear (Merlo et al., 2002; Vitelli et al., 2003; Xu et al., 1999; Zheng et al., 2003).
Involvement of Tbx2 and Tbx3 in the negative regulation of apoptosis has been demonstrated in several contexts (Du et al., 2017; Singhvi et al., 2008). However, we think that apoptosis, and hence morphogenetic impairment in the mutant otocyst, is secondary to a disturbed establishment of regional identities. Apoptosis at E10.5 was found in regions that exhibited specific loss or reduction of regional markers, including Aldh1a2, Gata3, Hmx3 and Lmo4 in the dorso-lateral, and Otx1 and Otx2 in the ventro-lateral otocyst, that have been implicated in vestibular morphogenesis (Deng et al., 2010; Karis et al., 2001; Morsli et al., 1999; Wang et al., 1998). Apoptosis in the ventral region of Tbx2cKO and Tbx2/3cDKO otocysts at E11.5 was preceded by gain of a neurogenic programme and by loss or reduction of genes that are essential for cochlear outgrowth (Gata3, Gdf6, Fgf10) (Bademci et al., 2020; Karis et al., 2001; Urness et al., 2015), and possibly for survival of delaminating neuroblasts (Fgf10) (Pirvola et al., 2000; Vázquez-Echeverría et al., 2008).
Previous work implicated WNT signals from the dorsal hindbrain and SHH signals from the ventral hindbrain and notochord in the establishment of dorsal or ventral fates, respectively, in the otocyst. These signals induce regionalised expression of transcription factor genes such as Dlx5/6, Hmx2/3 and Gbx2 (dorsal) and Pax2 and Otx2 (ventral) that regulate distinct morphogenetic subprogrammes along the dorso-ventral axis of the otocyst (Hatch et al., 2007; Ohta et al., 2016; Pirvola et al., 2000; Riccomagno et al., 2002, 2005). Our expression analysis characterised weak yet discernible expansion of WNT signalling into ventro-medial regions, a downregulation of ventral SHH signalling, and a loss of Otx1/Otx2 expression in the ventro-lateral otocyst of Tbx2/Tbx3cDKO embryos at E10.5, suggesting a (partial) reduction of ventral and a possible expansion of dorsal fates. However, many ventral genes, including Six1, Eya1, Pax2, Gata3, were still expressed and dorsal genes, including Dlx5, Gbx2, Aldh1a2, Smpx, were not ventrally expanded, but reduced in their dorsal expression indicating that dorso-ventral polarity was normally established and that changes in the activity of WNT and SHH signalling contribute only in a minor fashion to the morphogenetic defects of Tbx2/3cDKO otocysts.
In contrast, we think that altered BMP and FGF signalling may impact on axial patterning and regionalisation in Tbx2/3cDKO otocysts. Loss- and gain-of-function experiments in chicken and mouse demonstrated that dorsal BMP signalling is both required and sufficient to mediate a dorsal vestibular fate, possibly via induction of Dlx5 and Hmx3, and by repression of SHH-dependent ventral genes (Chang et al., 2008; Gerlach et al., 2000; Ohta and Schoenwolf, 2018; Ohta et al., 2016). Given the dorsal downregulation of Id3 expression in Tbx2/3cDKO otocysts, and hence of BMP signalling, it is likely that reduced expression of some dorsal genes, including Hmx3, can be attributed to changes in this pathway. Loss of Fgf3 from the hindbrain leads to a failure of endolymphatic duct and common crus formation, accompanied by epithelial dilatation and reduced cochlear coiling (Hatch et al., 2007). FGF10 derived from the developing cristae affects vestibular morphogenesis (Pauley et al., 2003). Conceivably, ectopic expression of Fgf8 in the ventral and of Fgf3 in the posterior-dorso-medial region of Tbx2/3cDKO otocysts may constitute novel signalling centres that directly interfere with regionalisation along the otocyst axes, and/or affect the activity of other signalling pathways, particularly BMP signalling, required for otocyst patterning as shown in many other developmental contexts (reviewed by Schliermann and Nickel, 2018; Teven et al., 2014). We conclude that TBX2 and TBX3 are essential regulators of signalling activities that confer regional identities important for localised outgrowth in the otocyst.
TBX2 restricts otic neurogenesis via repression of posterior FGF signalling and maintenance of Tbx1
Neurogenesis is initiated in the entire ventral half of the otocyst and becomes restricted to a small anterior-ventral region until E10.5 (Raft et al., 2004). Our analysis showed that neurogenesis was correctly initiated in the entire ventral region of Tbx2/3cDKO otocysts at E9.5 but failed to restrict to the anterior subregion at E10.5. Such a phenotype can be conceptualised by maintenance or establishment of a neural inducer in the posterior subregion and/or lack of a posterior repressor of this programme.
Previous studies in the mouse but also in other vertebrates provided ample evidence that FGF signals provide an initial cue for neural specification both in the otic but also in other sensory placodes (reviewed by Lassiter et al., 2014; Maier et al., 2014). Pharmacological inhibition of FGF signalling in cultured mouse otocysts abolished formation of Neurog1+ neuroblasts (Brown and Epstein, 2011), and loss of Fgf3, which is expressed in the neurosensory domain, disturbed otic ganglion formation (Mansour et al., 1993). In chicken, Fgf10, which is also expressed in the neurosensory domain, specifies neuronal fate (Alsina et al., 2004), and treatment with FGF2 enhances neuronal differentiation via an unknown mechanism (Adamska et al., 2001). In zebrafish, fgf3 and fgf8 are implicated in the specification of otic neuroblasts (Vemaraju et al., 2012).
In Tbx2-deficient otocysts, we detected ectopic FGF signalling in the posterior-ventral region at E9.5 and E10.5. At E9.5, ectopic FGF signalling correlated with posteriorly expanded Fgf3 expression, whereas at E10.5 we detected ectopic Fgf8 expression, which coincided with the posterior expansion of the neurogenic domain and ectopic induction of the neurosensory competence factor Sox3. Additionally, pharmacological inhibition of FGF signalling after E9.5 abrogated ectopic expression of Neurog1 in the posterior-ventral otocyst, suggesting that Tbx2 is required to repress FGF-induced neuronal competence in the posterior-ventral otocyst. Lack of reduction of Neurog1 expression at the anterior otocyst may reflect that neurogenesis on this side depends on a different regulatory programme, or that the neurogenic lineage has already been determined and no longer needs FGF signalling at this stage. The discrepancy compared with an earlier report (Brown and Epstein, 2011) may be due to usage of a different inhibitor under different experimental conditions.
In contrast to chicken, in which FGF8-mediated signalling induces otic neurogenic fate via induction of the proneural gene Sox3 (Abelló et al., 2010) or zebrafish, in which sox3 promotes neural competence in the otic epithelium (Gou et al., 2018), neither Fgf8 nor Sox3 have been associated with otic neurogenesis in the mouse. Fgf8 is expressed in maturing neuroblasts after delamination but is absent from the otic epithelium, as is Sox3 (Vitelli et al., 2003; this study). Hence, ectopic neural competence in Tbx2/3cDKO otocysts may be due to de-repression of an Fgf8-Sox3/Sox2-Neurog1 regulatory module that has been lost in mammalian evolution.
Concomitant with the upregulation of FGF signalling, we detected reduced BMP signalling in the posterior-ventral region of Tbx2/3cDKO otocysts, indicating a signalling antagonism similar to other biological settings (reviewed by Schliermann and Nickel, 2018). BMP signalling has been implicated in the inhibition of neurogenesis in many developmental contexts (Bond et al., 2012; Imayoshi and Kageyama, 2014; Shou et al., 1999). However, pharmacological inhibition of BMP signalling and activation of the pathway by overexpression of a constitutively active ALK3 at the otic placode stage in chick embryos caused a downregulation and anterior expansion of the posterior marker Lmx1b, respectively, but had no effect on the establishment of the neurogenic fate (Abelló et al., 2010). We, therefore, posit that loss of ventral BMP signalling in Tbx2/3cDKO otocysts does not directly account for the expansion of neurogenesis.
Previous work demonstrated that Tbx1 is both essential and sufficient to restrict neural fate to the anterior-ventral otocyst (Raft et al., 2004; Xu et al., 2007). Loss of Tbx1 expression in the posterior-ventral region of Tbx2-deficient otocysts at E10.5, is therefore likely to contribute in a major fashion to the observed phenotypic changes. Our analyses revealed ventral apoptosis (Fig. S11) and expanded posterior FGF signalling in Tbx1KO otocysts, similar to Tbx2/3cDKO mutants. However, ectopic posterior expression of Fgf8 and Sox3 and strong ventral induction of FGF signalling did not occur in Tbx1KO otocysts, indicating a functional diversity between TBX1 and TBX2.
In different contexts, FGF and RA have been shown to act antagonistically (Diez del Corral and Storey, 2004; Marklund et al., 2004), suggesting that FGF inhibits RA-mediated Tbx1 induction. However, normal expression of Tbx1 at E9.5 argues against such a scenario. Given the similar expression patterns of Tbx1 and Id3 in controls as well as loss of Tbx1 expression and BMP signalling in the posterior-ventral otocyst of E10.5 Tbx2/3cDKO embryos, it is tempting to speculate that induction of Fgf8 expression leads to reduced BMP signalling which, in turn, causes loss of Tbx1. Irrespective of the precise mechanism, our findings add to the overarching theme that TBX1, TBX2, TBX3 and most other T-box transcription factors act in cross-regulatory networks in a variety of developmental settings (Goering et al., 2003; Greulich et al., 2011; Mesbah et al., 2012; Sheeba and Logan, 2017).
In summary, we suggest that in Tbx2-deficient otocysts ectopic FGF signalling accounts for posterior expansion of the neurogenic region in a dual manner. First, FGF signalling activates the neurogenic regulatory network via Sox3 and Neurog1. Second, FGF signalling leads to repression of the neural repressor Tbx1, possibly via reduced BMP signalling (Fig. 8).
TBX2 may directly repress Fgf8 and other neuronal genes
Given the upregulation of Fgf8 and Fgf3, and a possible role of FGF signalling in inducing the loss of Tbx1 and the gain of neurogenesis in Tbx2/3cDKO otocysts, it is tempting to speculate that TBX2 directly represses Fgf8 and/or other FGF ligand genes. In order to identify TBX2 target genes, we performed a ChIP-seq experiment from more than 1000 E10.5 otocysts. Possibly owing to the paucity of tissue or antibody limitations, we failed to enrich for TBX2-bound fragments. However, interrogation of the web-platform ChIP-Atlas (chip-atlas.org; Oki et al., 2018) in which ChIP-Seq data sets for mouse and human tissues are deposited, identified TBX2/TBX3 bound genomic regions associated with Fgf8 (Tables S8-S10). Moreover, the web-tool oPOSSUM (opossum.cisreg.ca; Kwon et al., 2012) revealed evolutionarily conserved binding sites for TBX proteins in Fgf8 (Tables S8 and S11). These in silico data, together with the finding that TBX1 binds to and transactivates an ultraconserved cis-regulatory element downstream of Fgf8 (Castellanos et al., 2014), supports the notion that Fgf8 might present a direct target of TBX2 repressive activity in the otocyst.
Interestingly, additional neuronal genes with increased expression in Tbx2/3cDKO otocysts are associated with TBX2/TBX3 ChIP peaks in other tissues (Pou4f1, Nefm, Nefl, Foxn4, Dll1, Insm1, Nhlh2, Neurog1, Neurod1), and harbour TBX-binding sites (Pou4f1, Dll1, Neurod1) (Tables S8-S11), suggesting that TBX2 suppresses the neurogenic gene regulatory network at multiple levels beyond the repression of Fgf8 and the (indirect) maintenance of Tbx1. Supporting that notion, recent work showed that TBX3 represses NEUROD1 target genes via DNA binding in the absence of NEUROD1, thereby suppressing the neuronal lineage, but is displaced from cis-regulatory regions once NEUROD1 binds (Pataskar et al., 2016).
We conclude that TBX2 and TBX3 are novel regulators of otocyst patterning and that TBX2 exerts its function in the posterior-ventral region at least partly by repression of a neurogenic gene regulatory network.
MATERIALS AND METHODS
Mice and genotyping
Mice carrying a null allele of Tbx1 (Tbx1tm1Pa) (Jerome and Papaioannou, 2001), mice with a conditional floxed allele of Tbx2 (Tbx2tm2.1Vmc, synonym Tbx2flox) (Wakker et al., 2010) or Tbx3 (Tbx3tm1Pa, synonym Tbx3flox) (Frank et al., 2013), the double fluorescent Cre reporter line Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J (synonym R26mTmG) (Muzumdar et al., 2007) and the transgenic cre driver line Tg(Pax2-cre)1AKis (synonym Pax2-cre) (Trowe et al., 2011) were maintained on an NMRI outbred background. Embryos for Tbx2 and Tbx3 gene expression analyses were derived from matings of NMRI wild-type mice. Pax2-cre/+;Tbx2flox/flox;Tbx3flox/flox;R26mTmG/+ (Tbx2/3cDKO) mice were obtained from matings of Pax2-cre/+;Tbx2flox/+;Tbx3flox/+ males and Tbx2flox/flox;Tbx3flox/flox;R26mTmG/mTmG females. For generation of Pax2-cre/+;Tbx2flox/flox (Tbx2cKO) embryos, Pax2-cre/+;Tbx2flox/+ males were mated to Tbx2flox/flox females. Pax2-cre/+;Tbx3flox/flox (Tbx3cKO) mice were obtained from matings of Pax2-cre/+;Tbx3flox/+ males and Tbx3flox/flox females. For conditional mutant mice, cre-negative littermates served as controls. Tbx1-deficient embryos were generated from matings of Tbx1/+ animals. The presence of a vaginal plug on the morning after mating defined midnight as time of fertilisation; noon was correspondingly defined as E0.5. Pregnant females were sacrificed by cervical dislocation. Embryos were dissected in PBS, fixed in 4% paraformaldehyde (PFA)/PBS overnight and stored in 100% methanol at −20°C until use. Somite numbers were used for embryo staging. Genotyping was carried out by PCR on genomic DNA prepared from embryonic tissues or ear clips.
All animal work conducted for this study was performed according to European and German legislation. The breeding and handling of mice lines was approved by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit (permit number AZ33.12-42502-04-13/1356).
Histological and immunofluorescence analyses
For histological analysis, 5-μm-thick paraffin sections were stained with Haematoxylin and Eosin or with Alcian Blue and Eosin. 3D reconstruction of stained serial sections was performed with the Amira software (version 5.3.3, ThermoFisher Scientific).
For detection of antigens on 5-µm-thick paraffin sections, labelling with primary antibodies was performed at 4°C overnight after antigen retrieval (15 min at 100°C; H-3300, Vector Laboratories), blocking of endogenous peroxidases with 3% H2O2/PBS for 15 min and incubation in blocking buffer (TNB) provided by the Tyramide Signal Amplification (TSA) kit (NEL702001KT, Perkin Elmer) for 45 min. The following primary antibodies were used: rabbit-anti-TBX2 (1:4000, 07-318, Merck Millipore), polyclonal goat-anti-TBX3 (1:500, sc-31656, Santa Cruz Biotechnology), polyclonal rabbit-anti-cleaved CASPASE-3 (1:400, 9661, Cell Signaling Technology), monoclonal mouse-anti-GFP (1:250, 11814460001, Roche). Primary antibodies were visualised with either biotinylated Fab fragment goat-anti-rabbit IgG (1:200, 111-067-003, Dianova), biotinylated donkey-anti-goat IgG (1:200, 705-065-147, Dianova) and the TSA system (NEL702001KT, Perkin Elmer) or Alexa 488-conjugated donkey-anti-mouse IgG (1:200, A21202, Invitrogen). Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI, 6335.1, Carl Roth) according to the manufacturer's instruction. At least three specimens of each genotype were used for each experiment.
Proliferation and apoptosis assays
Cell proliferation rates were investigated by the detection of incorporated BrdU on 5-µm-thick transverse paraffin sections according to published protocols (Bussen et al., 2004). For each specimen (n=6 per genotype), eight adjacent sections were assessed. The BrdU labelling index was defined as the number of BrdU-positive nuclei relative to the total number of nuclei as detected by DAPI counterstaining in arbitrarily defined regions. Statistical analysis was performed using the two-tailed Student's t-test. Data were expressed as mean±s.d. Differences were considered significant when the P-value was below 0.05. Apoptosis was assessed by TUNEL assay using the ApopTag Plus In Situ Apoptosis Fluorescein Detection Kit (S7111, Merck Millipore) and by anti-cleaved CASPASE-3 staining on at least three specimens per genotype.
RNA in situ hybridisation analysis
Whole-mount RNA in situ hybridisation was performed following a standard procedure with digoxigenin-labelled antisense riboprobes (Wilkinson and Nieto, 1993). Stained specimens were transferred to 80% glycerol prior to documentation. RNA in situ hybridisation on 10-μm-thick paraffin sections (5-µm-thick paraffin sections for reconstructions) was performed as previously described (Moorman et al., 2001). For each marker at least three independent specimens were analysed. Primers for PCR amplification of new DNA templates for in vitro transcription of RNA probes are listed in Table S12.
For explant cultures of otocysts, the part anterior of the 4th branchial arch of E9.0-E9.5 (23-28 somites) embryos was dissected and cut in half along the neural tube in L-15 Leibovitz medium (F1315, Biochrom). The tissue pieces were placed on 0.4 µm polyester membrane Transwell supports (3450, Corning) with the medial side down, and incubated at 37°C and 5% CO2 in organ culture medium (DMEM/F12) (21331-020, ThermoFisher Scientific) supplemented with 10% fetal calf serum (S0115, Biochrom), 100 units/ml penicillin/100 µg/ml streptomycin (15140-122, ThermoFisher Scientific), 1 mM sodium pyruvate (11360-039, ThermoFisher Scientific), 1× MEM NEAA (11140-035, ThermoFisher Scientific) and 1× GlutaMAX (35050-038, ThermoFisher Scientific) at the air-liquid interface for 24 h. For inhibition of FGF signalling, AZD4547 (HY-13330, MedChemExpress) was added to the medium at a final concentration of 1 µM. At the end of the culture period, the tissue was fixed in 4% PFA overnight at 4°C, and then processed for whole-mount RNA in situ hybridisation. Otocysts were punctured to avoid probe trapping.
Microarray experiments and data analysis
Total RNA was extracted from pools of otocysts (n=14-18) mechanically dissected from E10.5 control and Tbx2/3cDKO embryos using the peqGOLD RNAPure reagent (732-3312, Peqlab) and sent to the Research Core Unit Transcriptomics of Hannover Medical School where it was hybridised to Agilent Whole Mouse Genome Oligo v2 (4×44 K) Microarrays (G4846A, Agilent Technologies) in a dual-colour mode. Data were averaged from four independent biological samples per genotype.
Log2-converted expression data were imported into Qlucore Omics Explorer (version 3.6, Lund, Sweden) and normalised (mean=0, Var=1). For the identification of deregulated genes, probe sets with the highest values for a gene were used in a two-group comparison (t-test). Genes with a FDR-adjusted P-value (q-value)<0.05 were considered deregulated, and further subjected to filtering using a signal intensity threshold (≥100 either for control or mutant pool) and a fold change threshold (FC≥2, FC≤−2). Functional enrichment analysis was performed using DAVID 6.8 software (Huang et al., 2009a,b), and terms were selected based on Gene Functional Annotation Clustering on GO-BP-FAT annotations, ‘medium’ stringency and final group membership of ≥5.
For the prediction of direct target genes of TBX2/TBX3, we used the ‘Target Genes’ tool of the public ChIP-seq database ChIP-Atlas (http://Chip-atlas.org) (Oki et al., 2018). MACS2 scores for all genes were retrieved for TBX2 (in human, hg38) and TBX3 (in mouse, mm10) (setting ‘Distance from TSS’: ±10 kb), and joined to genes upregulated in our microarray based on the gene symbol. Evolutionarily conserved T-box transcription factor-binding sites were identified using the oPOSSUM3 software (http://opossum.cisreg.ca) (Kwon et al., 2012) using the JASPAR Core profile for Brachyury (T) and 10 kb up/downstream sequence.
Sections were photographed using a Leica DM5000 microscope with a Leica DFC300FX digital camera. Whole-mount specimens were photographed on a Leica M420 with Fujix digital camera HC-300Z. All images were processed in Adobe Photoshop CS4.
Results are expressed as mean±s.d. Normal distribution of data was tested by Shapiro–Wilk test (P>0.05) and Q-Q-plots. For comparison of two groups, an unpaired two-tailed Student's t-test was performed. P<0.05 was considered statistically significant. Statistical analyses were performed using Microsoft Excel (version 16.16.24) and JASP (version 0.13.1).
We thank Robert Kelly (Marseille) for the Tbx1 mice and the Research Core Unit Transcriptomics of Hannover Medical School for microarray analysis.
Conceptualization: A.K., M.-O.T.; Formal analysis: M.K., A.K., M.-O.T.; Investigation: M.K., I.W., C.R., T.H.L., M.-O.T.; Resources: V.M.C., A.M.; Data curation: M.K., M.-O.T.; Writing - original draft: M.K., A.K., M.-O.T.; Writing - review & editing: M.K., I.W., C.R., T.H.L., V.M.C., A.M., A.K., M.-O.T.; Visualization: M.K.; Supervision: A.K., M.-O.T.; Project administration: M.-O.T.; Funding acquisition: M.-O.T.
This work was supported by the German Research Council (Deutsche Forschungsgemeinschaft) [DFG TR1325/1-1 to M.-O.T.]
Microarray data have been deposited in Gene Expression Omnibus under accession number GSE154305.
Peer review history
The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.195651
The authors declare no competing or financial interests.