Facial whisker pattern is not sufficient to instruct a whisker-related topographic map in the mouse somatosensory brainstem

Relay of somatosensory stimuli from the body surface to higher brain centres is highly organised, allowing the sensing of positional origin of an input. Facial somatosensory inputs are serially relayed through the trigeminal circuit to the brainstem, thalamus and neocortex. The trigeminal circuit is somatotopically organised, such that topographic maps of connectivity matching the distribution and density of sensory receptors of facial dermatomes are generated at all levels of the pathway (Erzurumlu and Killackey, 1983; Erzurumlu et al., 2010; Ma, 1991, 1993; Ma and Woolsey, 1984; Schlaggar and O'Leary, 1993; Van Der Loos, 1976; Woolsey and Van der Loos, 1970).

Distinct facial dermatomes are innervated by the peripheral axonal processes of trigeminal ganglion (TG) primary sensory neurons, whose central axons form the trigeminal nerve (nV) and project to innervate second order neurons in the brainstem trigeminal column, composed of the rostral principal (PrV) and the caudal spinal (SpV) sensory nuclei. TG neurons bridge the facial sensory periphery and the brainstem where facial maps are first formed. During prenatal development, somatotopic segregation of TG cell bodies contributes to the segregation of the trigeminal nerve into its three main divisions – the mandibular, maxillary and ophtalmic branches, which peripherally innervate the corresponding facial dermatomes (Arvidsson and Rice, 1991; Erzurumlu and Jhaveri, 1992; Erzurumlu and Killackey, 1983; Erzurumlu et al., 2010; Hodge et al., 2007).

In mouse, the largest portion of the facial somatosensory map is devoted to the representation of mystacial whiskers which are organised into five rows of four to seven follicles at invariant positions on the snout. Whisker inputs are somatotopically mapped at each level of the pathway as spatially ordered neuronal modules, called barrelettes (brainstem), barreloids (thalamus), and barrels (cortex) (Ma and Woolsey, 1984; Van Der Loos, 1976; Woolsey and Van der Loos, 1970), reproducing facial whisker distribution.

The central axons of the mouse trigeminal nerve divisions start sending radially oriented collaterals at about embryonic day (E) 14.5 to innervate the PrV and SpV brainstem nuclei, and at about E16.5 begin to arborise, forming dense terminals (Erzurumlu et al., 2006; Ozdinler and Erzurumlu, 2002). In the developing PrV, mandibular axon collaterals selectively target the dorsal portion (dPrV) whereas whisker-related afferent collaterals preferentially target the ventral portion (vPrV) with a dorsoventral row-specific organisation (Erzurumlu and Killackey, 1983; Erzurumlu et al., 2010; Hodge et al., 2007; Oury et al., 2006; Xiang et al., 2010; this study). Thus, the spatial segregation of collateral targeting by distinct trigeminal divisions in PrV provides an early template to build topographic equivalence between the face and the brainstem.

To what extent peripheral signals and/or patterns are sufficient to impose a central somatotopic pattern is still debated. One approach to understanding a potential instructive role of the periphery in imposing a central somatotopic pattern has been to manipulate the number and/or spatial organisation of whiskers within the whisker pad. Such peripheral changes were reflected on the somatotopy of the barrel map (Ohsaki et al., 2002; Van der Loos et al., 1984). Moreover, retrograde signalling from the developing face was shown to be involved in establishing spatial patterns of gene expression in the TG and transferring somatotopic information to the brainstem (da Silva et al., 2011; Hodge et al., 2007). Such experiments seemed to indicate an instructive role of signalling from facial receptors to establish ordered connectivity at the peripheral and central level for somatotopic map generation.

By contrast, the early steps of cortical development and arealisation rely on intrinsic patterning mechanisms that are independent of sub-cortical input relaying information from the periphery (Cohen-Tannoudji et al., 1994; Grove and Fukuchi-Shimogori, 2003; López-Bendito and Molnár, 2003; O'Leary et al., 2007; Sur and Rubenstein, 2005). Furthermore, recent studies have begun to uncover some of the intrinsic molecular mechanisms underlying the establishment of somatotopic organisation at brainstem level (Erzurumlu et al., 2010; Oury et al., 2006). Thus, although facial pattern provides a template for map formation, it is still unclear whether it is sufficient to impose a central somatotopic pattern. Moreover, the relative importance of facial pattern to organise TG and central topography of connectivity at pre-natal stages, before facial maps are apparent at sub-cortical and cortical levels, is still poorly understood.

By using multicolour tracing methods, we first determined that a coarse row- and single whisker-specific somatotopic map is already built into somatosensory brainstem nuclei from the outset of the targeting process at pre-natal stages, with a relative degree of precision and peripheral positional discrimination. To investigate the importance of peripheral organisation in imposing a topographic pattern we took advantage of mutant mice exhibiting an ectopic whisker pattern on the lower jaw. In Edn1 null mutant mice, the lower jaw is morphologically transformed into an upper jaw-like structure, including a partial duplication of the whisker pad with a second set of ordered whisker rows located in an ectopic, mandibular position (Fig. S1A) (Clouthier et al., 1998; Kurihara et al., 1994; Ozeki et al., 2004).

The Edn1^−/− mutation is lethal at birth, thus preventing analysis at postnatal stages. Nonetheless, we found that, in Edn1^−/− mutants, the ectopic whisker pad is innervated by mandibular, though not maxillary, TG neurons whose central axon collaterals initially targeted the dPrV. However, unlike maxillary neurons, ectopic whisker-innervating mandibular neurons and pre-target axons did not segregate into a row-specific pattern nor target the dPrV with a topographic pattern. Following periphery-driven molecular re-patterning to a maxillary-like identity, mandibular neurons redirected their central projections from dorsal to ventral PrV. Thus, a spatially ordered lower jaw ectopic whisker pattern is not sufficient to impose pre-target organisation of the mandibular tract nor a whisker-related matching pattern in dPrV, but it is still able to induce maxillary-like molecular features resulting in vPrV final targeting.

Distinct rows in the whisker pad are innervated by neuronal populations with segregated latero-medial distribution within the TG maxillary portion (da Silva et al., 2011; Erzurumlu and Jhaveri, 1992; Erzurumlu and Killackey, 1983; Hodge et al., 2007), whereas barrelette topography is not prefigured by a pre-ordered position of neurons within row-specific TG sub-populations (da Silva et al., 2011). We therefore investigated the establishment of whisker-specific afferent patterns in the developing brainstem and carried out row- and whisker-specific neuronal tracings with multicolour lipophylic NeuroVue fluorescent dyes, which allow controlled diffusion by progressive dye release from a filter membrane (Jensen-Smith et al., 2007).

By applying tiny pieces of NeuroVue-coated filters we simultaneously traced selected subsets of nerve endings innervating individual rows at E14.0. Single row-innervating primary neuron cell bodies segregated along the lateromedial axis of the maxillary TG portion (Fig. S2E). Furthermore, each set of single row-innervating nerve fascicles remained segregated in their central projections with a row-specific topography along the dorsoventral axis of the trigeminal tract maxillary division (Fig. S2A-E). These results confirmed a high degree of order in the trigeminal peripheral system during prenatal development (Erzurumlu and Killackey, 1983) and supported the validity of our tracing procedure.

To investigate the spatial arrangement of whisker-specific central collateral targeting, we simultaneously labelled distinct anteroposterior whisker follicle positions in different rows of E14.5 and E17.5 whisker pads (Fig. S2F-I). The insertion of NeuroVue filters at single follicles allowed targeting of the follicle terminals and just their close neighbours. To improve tracing precision, we selectively labelled whiskers at distinct positions where they are bigger and less densely spaced (position 1-5) (Fig. S2J). Along the dorsoventral axes of the vPrV and SpVi nuclei, collaterals maintained the row-specific somatotopy observed in the trigeminal tract (Fig. S2F,H and not shown) (Xiang et al., 2010). Notably, collaterals from afferents innervating anterior whiskers projected deeper medially into the vPrV nucleus than collaterals of posterior whisker positions (colour bars, Fig. S2F,H). The early somatotopic pattern of whisker-specific collateral targeting was maintained and further refined by E17.5 (Fig. S2G,I). At this stage, collaterals showed dense arborisation, prefiguring maturation of the final whisker map and barrelette formation as observed by cytochrome oxidase staining at postnatal (P) stages (P6, Fig. S2K).

In summary, these results further extend previous work. They show that the topographic mapping of row-specific and single whisker-specific afferent targeting in brainstem nuclei is established at the onset of collateralisation with a relative degree of precision and peripheral positional discrimination, and is refined throughout late prenatal and early postnatal stages.

We first asked whether the ectopic whisker pad of Edn1 null mutants is innervated and by which TG division. We carried out triple retrograde labelling at E14.5 and applied NeuroVue at the ectopic whisker pad on the mutant lower jaw and at two dorsoventral positions in the maxillary whisker pad (Fig. 1C,D). The neurons innervating the ectopic whiskers, traced in a retrograde manner, were located in the mandibular portion of TG and were segregated from those in the maxillary division wiring the normal whisker pad (Fig. 1E,F). Thus, the ectopic whisker pad on the lower jaw of Edn1^−/− mutants is not able to attract and redirect the peripheral projections of maxillary primary sensory neurons.

Furthermore, the central axon of neurons innervating the duplicated whisker pad did not intermingle with the axons of the neurons innervating the maxillary whisker pad and maintained the same dorsally segregated position in the trigeminal tract of Edn1^−/− fetuses as that of mandibular axons in wild type (Fig. 1G,H).

Bmp4 and TGF-β retrograde signalling from the whisker pad and/or follicles has been suggested to be involved in organising topographic patterns of TG neuron central connectivity (da Silva et al., 2011; Hodge et al., 2007). In particular, Bmp4-dependent signalling regulates early positional differences of Tbx3 and Hmx1 transcription factor expression in distinct divisions of the TG (Hodge et al., 2007). Bmp4 expression is readily detected in the duplicate arrays of whisker primordia on the Edn1^−/− mutant lower jaw (Ozeki et al., 2004). We therefore asked whether maxillary-like spatial patterns of transcription factor expression were ectopically induced in the mandibular division of Edn1^−/− mutants upon innervation of the duplicated whisker pad.

In wild-type E11.5 and E13.5 TG, Hmx1 and OC1 (Onecut1 – Mouse Genome Informatics) are selectively expressed in the mandibular division, whereas Tbx3 is highly expressed in ophthalmic and maxillary neurons with only sparse expression in mandibular neurons (Fig. 2B-D) (Hodge et al., 2007). In Edn1^−/− mutants, expression of the general marker Drg11 (Prrxl1 – Mouse Genome Informatics) was normally maintained throughout the TG (Fig. 2A,E,L,N). By contrast, Tbx3 was induced in a larger number of mandibular neurons than in wild type, whereas Hmx1⁺ and OC1⁺ cells were concomitantly reduced (Fig. 2F-H). This was further confirmed with Hmx1/Tbx3 double fluorescent in situ hybridisation (FISH) and quantified (Fig. 2I-K). This effect was not a result of increased cell death in the mandibular TG component, as assessed by activated caspase-3 immunostaining (Fig. S3A,B). Moreover, no Tbx3 expression differences were observed at E10.25 between wild-type and Edn1^−/− TG neurons (Fig. S3G,L), indicating that the molecular changes observed in E11.5 Edn1^−/− mandibular neurons are induced as their axons grow into the duplicated whisker pad.

To further support a periphery-induced re-patterning of mandibular neurons in Edn1^−/− mutants, we next searched for additional TG maxillary-specific expression markers in the Allen Developing Mouse Brain Atlas (http://developingmouse.brain-map.org). At E13.5, cadherin 13 (Cdh13) is selectively expressed in the maxillary TG division, whereas it is excluded from the mandibular division (Fig. 2M). By combining retrograde dextran tracing of row-specific afferents and Cdh13 in situ hybridisation at E13.5, we further assessed that Cdh13 is preferentially expressed in TG neurons innervating row C-E (Fig. S3C-F). As these rows are duplicated in a mirror pattern in Edn1^−/− mutants (Ozeki et al., 2004) (Fig. S1), we assessed Cdh13 expression in E13.5 mutant TG. Notably, Cdh13 was ectopically induced in the mandibular division of Edn1^−/− mutants (Fig. 2O).

In summary, these findings demonstrate that positional molecular differences are induced in Edn1^−/− mutant TG, leading a subset of mandibular neurons to acquire molecular features normally expressed by maxillary primary sensory neurons innervating mystacial whiskers.

We next asked whether the observed molecular changes were sufficient to induce corresponding maxillary-like changes in the central topography of TG mandibular neurons in Edn1^−/− fetuses. We assessed whether the tract innervating the duplicated arrays of whiskers in Edn1^−/− mutants also became somatotopically organised, in the same manner as maxillary tract organisation (Fig. S1). In wild-type fetuses (n=4/stage), NeuroVue labelling of two distinct positions at E14.5 and E16.5 on the lower jaw (Fig. 3A, Fig. S4A) revealed intermingled non-segregated primary neuron cell bodies in the mandibular TG division (Fig. 3C, Fig. S4B). This was further depicted in the lack of spatial segregation of mandibular afferent central axons (Fig. 3E, Fig. S4C-D), contrasting with the strict somatotopic organisation of the maxillary tract (n=4/stage) (Fig. 1C, Fig. S2E).

Similarly, in E14.5 and E16.5 Edn1^−/− mutant fetuses (n=4/stage), despite the ectopic whisker arrays on the lower jaw, NeuroVue labelling of two distinct ectopic row positions (Fig. 3B, Fig. S4E) also resulted in a non-segregated distribution of retrogradely labelled cell bodies in the TG (Fig. 3D, Fig. S4F) and a lack of somatotopic organisation of central axons (Fig. 3F, Fig. S4G,H), similar to wild-type mandibular branch organisation. Thus, the maxillary-like molecular changes induced in neurons innervating the duplicated whisker rows in Edn1^−/− mutants (Fig. 2) are not sufficient to impose a maxillary-like topography to their central afferents.

We then investigated the central pattern of collateral targeting in E14.5 Edn1^−/− mutants. In wild-type fetuses, mandibular TG axons project radially oriented collaterals into rhombomere (r)2-derived dPrV, whereas r3-derived vPrV selectively receives collateral input from whisker-related afferents (Oury et al., 2006). These distinct populations of targeting collaterals remain spatially segregated within the PrV and never cross each other (Fig. 1H, Fig. 3G, Fig. 4C,D,I,J) (Oury et al., 2006). In Edn1^−/− mutants, axon afferents from the duplicated whisker arrays did not project collaterals into the vPrV area which normally hosts the whisker map but instead targeted the dPrV area which normally generates the lower jaw map (Fig. 1J). Moreover, labelling of distinct peripheral positions on the lower jaw did not result in spatially segregated collaterals along either the row-specific dorsoventral or whisker-specific lateromedial axes in the dPrV (Fig. 3H), in contrast to the whisker-related collateral topographic mapping observed in the vPrV (Fig. 1I, Fig. S2F-I).

Remarkably, unlike at E14.5, at E16.5 ectopic whisker (Md*)-innervating collaterals turned from dPrV (and dorsal SpV), and navigated into vPrV (and ventral SpV) (arrows, Fig. 4G,H, Fig. S4G,H). Upon labelling of two distinct ectopic row positions no clear sorting of collaterals navigating into vPrV or ventral SpV was observed (arrows, Fig. S4G-H). The fraction of collaterals attracted ventrally was variable (roughly 5-40%) (see Fig. 4G-H, Fig. S4G-H). This variability likely depended on the precision and/or amount of labelling of the ectopic whisker follicles versus the surrounding lower jaw tissue in each experiment and/or on the extent of mandibular neuron repatterning among distinct Edn1^−/− mutant individuals. Nonetheless, such a mis-targeting behaviour of Md* collaterals was observed in all mutants (n=6/6) and never in wild type (n=6/6) (quantified in Fig. 4I,J).

Thus, the Md* primary neurons innervating the duplicated whisker pad in Edn1^−/− mutants initially project central axon collaterals into dPrV, similar to normal lower jaw input. Despite the observed maxillary-like molecular changes as early as E11.5 (Fig. 2), these Md* incoming afferents are unable to generate ordered ectopic whisker-related patterns of collateral targeting in dPrV, unlike the maxillary whisker afferents in vPrV. However, the molecular changes in Md* TG neurons might enable ectopic whisker afferents to respond to late vPrV-specific targeting cues prior to barrelette formation (Fig. 4K,L).

Brainstem whisker maps develop through interplay between peripheral signals, intrinsic pre-patterning of TG neurons with their axonal processes, and brainstem target neurons, establishing a prenatal coarse topographic connectivity map. Further refinement by whisker-related activity and/or retrograde molecular signalling from periphery contributes to achieve final one-to-one topography at postnatal stages. Recent studies addressed the importance of facial signals to refine central projections (da Silva et al., 2011; Hodge et al., 2007) whereas others highlighted the importance of brainstem patterning to receive appropriate and spatially restricted peripheral projections and allow topographic representation (Erzurumlu et al., 2010; Oury et al., 2006). However, the relative importance of facial signals versus TG pre-target axon sorting versus central pre-patterning in the building of topographic equivalence between the face and the brainstem is still unclear. In this study, we asked whether an ordered array of whisker receptors on the face is sufficient to impose a matching topographic connectivity into the brainstem, in order to unveil the relative contributions of specific TG neuron pre-sorting events and/or central pre-patterning.

A number of conclusions can be drawn from our study. Firstly, we provide evidence that, in addition to a row-specific somatotopy of maxillary pre-targeting axons, a single-whisker map of afferent collateral targeting is set out in vPrV from the outset of the collateralisation process with a significant degree of precision and peripheral positional discrimination, which is progressively refined through late prenatal and postnatal stages (Fig. S2). Thus, postnatal activity-dependent refinement processes might only act locally to stabilise pre- and post-synaptic elements, but they are not the main contributors to building an ordered whisker-specific topography within the brainstem sensory nuclei. This is supported by the analysis of NMDA receptor knockout mice, in which barrelette neuron dendrite remodelling and patterning is impaired, but the general topographic organisation of afferents is maintained in mutant PrV (Lee et al., 2005).

Secondly, we show that the central axons of TG mandibular neurons innervating distinct lower jaw surface positions display poor spatial segregation within the mandibular tract, as do their collateral patterns in dorsal PrV. Our analysis of Edn1^−/− mutants further demonstrate that ectopic whisker arrays are selectively targeted by the mandibular branch, according to their position on the lower jaw, and that this innervation in turn induces significant maxillary-like molecular changes in mandibular TG neurons, supporting whisker-specific retrograde signalling (da Silva et al., 2011; Hodge et al., 2007). However, such molecular changes are not sufficient to instruct a maxillary-like, row-specific, pre-target axon sorting of the mandibular tract, nor to instruct the establishment of row- and/or whisker-related patterns of collateral targeting in the dorsal ‘mandibular’ area of PrV. Together, these findings highlight a fundamental intrinsic pre-ordering difference between TG maxillary and mandibular primary axons which will be important to address in future studies. Pre-ordering of somatosensory thalamocortical axons is essential for the transfer of precise topographic equivalence between thalamic and cortical whisker maps (Lokmane et al., 2013). Our current results suggest that maxillary neuron pre-target axon pre-ordering (Erzurumlu and Jhaveri, 1992; Erzurumlu and Killackey, 1983; reviewed in Erzurumlu et al., 2010) might be intrinsically organised, independently of facial or brainstem influences, and might be an important spatial requisite to match intrinsic positional information in the vPrV (see summary diagram in Fig. 4K,L).

Lastly, even though ectopic whisker-innervating mandibular axons initially target the dPrV, following periphery-driven repatterning their collaterals turn and navigate from dorsal to ventral PrV, converging into the whisker-related area already occupied by the processes of maxillary neurons innervating the normal whisker pad on the snout (Fig. 4). These latter findings unveil a temporal sequence of events underlying the transfer of peripherally induced whisker-specific information to the brainstem and suggest the importance of a vPrV-specific intrinsic patterning program to match whisker-specific input. Further studies will be required to address the molecular determinants of such a program. In this respect, it is however noteworthy that Hoxa2 displays a differential expression pattern between dPrV and vPrV, and has an important role in prenatal PrV patterning (Oury et al., 2006). Hoxa2 is expressed in all PrV mitotic progenitors but differentially maintained in the post-mitotic neurons of vPrV, though not dPrV (Oury et al., 2006). Moreover, Hoxa2 is not expressed in TG neurons or facial mesenchyme (Oury et al., 2006). The early dorsoventral differential expression of Hoxa2 regulates the specific targeting of mandibular versus maxillary trigeminal axon collaterals between the dorsal and ventral components of the developing PrV (Erzurumlu et al., 2010; Oury et al., 2006). In particular, conditional Hoxa2 inactivation selectively impaired whisker-related afferent targeting in the vPrV (Oury et al., 2006), indicating that Hoxa2 might regulate the expression of whisker afferent collateralisation factor(s) in vPrV. Brainstem maturation is a crucial regulator of the onset of arborisation of trigeminal ganglion afferents in brainstem nuclei (Erzurumlu et al., 2010; Ozdinler and Erzurumlu, 2002). Prenatal row- and whisker-specific afferent topographic collateralisation could be under the control of collateral-target interactions mediated by differentially expressed surface-bound and/or secreted guidance molecules (Erzurumlu et al., 2010). In this regard, Hoxa2 has been shown to positively regulate Epha4 and Epha7 expression in vPrV (Oury et al., 2006). Moreover, our finding that Cdh13 displays row-related expression in TG neurons suggests that matching adhesive cues between central afferent axons and brainstem target neurons might also contribute to topographic collateralisation into vPrV. Lastly, dPrV is devoid of Hoxa2 expression at the stage of collateral formation (Oury et al., 2006). Thus, the lack of topographic patterns of collateral targeting observed in Edn1 mutant dPrV, which receives selective input from the duplicated whisker arrays (Fig. 4), might at least partly account for the lack of a Hoxa2-dependent patterning program that normally instructs prenatal afferent targeting topography and establishment of whisker-related patterns in vPrV.

Edn1^tm1Utj (referred to as Edn1^−/−) homozygous mutant mice were described (Kurihara et al., 1994). Each in situ hybridisation and tracing experiment at any given stage was carried out on at least n=3 Edn1^−/− mutant and control embryos and fetuses; some tracing experiments were carried out on n=4 or n=6 mutant and control fetuses, respectively, as indicated in the main text. All animal experiments were approved by local veterinary authorities and conducted in accordance with the Guide for Care and Use of Laboratory Animals.

Embryos were dissected and fixed in 4% paraformaldehyde/PBS (PBS–PFA4%) overnight at 4°C. Tissues were cryoprotected in 20% sucrose and embedded in gelatine with 7.5%/10% sucrose/PBS. Cryostat sections (20-25 μm) in coronal and sagittal orientations were frozen at −80°C until processing. For chromogenic in situ hybridisation, the OC1, Tbx3 and Hmx1 (Hodge et al., 2007) probes were a gift from F. Wang. The Cdh13 probe was generated by cloning nucleotides 1139-2145 from the coding sequence into a pCRII-TOPO vector using a TOPO TA cloning kit (Life Technologies). Fluorescent in situ hybridisations were performed using the RNAscope Multiplex Fluorescent Kit according to the manufacturer's protocol [Advanced Cell Diagnostics (ACD), ref: 320850]. In brief, sections were air dried for 30 min at room temperature (RT), and a hydrophobic barrier was drawn around sections with an Immedge hydrophobic barrier pen. Sections were incubated for 20 min with the protease solution Pretreat 4 (RNAscope Pretreatment Kit) at RT in a humid chamber, washed twice with PBS and then incubated with the mixture of Tbx3 and Hmx1 probes for 2 h at 40°C in an oven. cDNA probe sets were designed and generated by ACD. Targeted sequences were: Mm-Tbx3-C1, nucleotides 460-1597 of accession number NM_011535.3 and Mm-Hmx1-C2, nucleotide 109-1483 of accession number NM_010445.2. A probe against the gene encoding POL2RA, a protein expressed in mammalian cells, was used as positive control, and a probe against Escherichia coli dapB (not expressed in mammalian cells) was used as a negative control (data not shown). Sections were then incubated with preamplifier and amplifier probes by applying AMP1 (40°C for 30 min), AMP2 (40°C for 15 min), and AMP3 (40°C for 30 min), followed by incubation with AMP4 AltA (40°C, 15 min), containing fluorescently labelled probes to detect Tbx3 RNA in green (Alexa Fluor 488, ACD), and Hmx1 RNA in orange (Alexa Fluor 550, ACD). A final incubation of slides with DAPI (ACD) for 30 s was performed before mounting them with Prolong Gold antifade reagent (Molecular Probes, P36934). Imaging of fluorescent signals was performed using an Axio Imager Z2 upright microscope coupled to a LSM700 Zeiss laser scanning confocal at 40×. Maximum intensity projections and stitching of double fluorescent in situs were performed using ZEN Software (Zeiss).

Cryosections were immunolabelled with a rabbit primary antibody against active Caspase-3 (1:500; Promega, G748A), and goat anti-rabbit Alexa Fluor 568-conjugated antibody (1:200; Molecular Probes, A-11011), counterstained with DAPI and mounted with Prolong Gold antifade reagent.

For cytochrome oxidase (CO) histochemical staining (Fig. S2K), 40 μm cryostat sections were cut in the coronal plane to visualise barrelettes at brainstem levels. The CO staining was performed according to the procedure described in Wong-Riley (1979).

NV Red (FS-1002), NV Maroon (FS-1001) and NV Jade (FS-1006) NeuroVue dye-coated filters were from Molecular Targeting Technologies (West Chester, PA, USA). Embryos were fixed in PBS–PFA4% overnight at 4°C. Small pieces (<1 mm²) of NeuroVue filter dyes were cut and placed into the specimen at specific locations. Embryos and fetuses were incubated at 37°C for 2 weeks (E10.5, E11.5), 3 weeks (E12.5), 4 weeks (E14.0, E14.5) or 6 weeks (E16.5, E17.5) in PBS–PFA4%. Diffusion of staining around the injection sites was monitored using a MacroFluo Z6 APO (Leica Microsystems). Whole-mount brains were dissected, keeping trigeminal ganglion attached to the brain, and imaged. Flat-mount preparations of E10.5 and E11.5 embryos were obtained by dissecting the neural tube, mounting in PBS between a slide and a cover slip and imaging. Older embryos were embedded into 4% agarose. Vibratome sections of 50-100 μm were mounted in Aqua-Poly/Mount (Polysciences) and analysed under MacroFluo Z16 APO and confocal microscope (LSM700, Zeiss).

Biotin-conjugated lysine-fixable dextran (Invitrogen, D-7135) was employed for retrograde labelling of specific whisker rows in E13.5 mouse fetuses. Dextran crystals were prepared according to Stirling et al. (1995). E13.5 fetuses were dissected and labelled with dextran crystals in given whisker follicle, cultured for 6 h as described previously (Oury et al., 2006; Stirling et al., 1995), then fixed in PBS–PFA4% and prepared for cryosection as described above. Sagittal sections of 25 μm were collected and processed for chromogenic in situ hybridisation against Cdh13. Biotin was revealed afterward using fluorescent-conjugated streptavidin (1:200; Molecular Probes, S11223), mounted and analysed under fluorescence microscope (Olympus).

Brightness, contrast and gamma (not for in situ hybridisation) were adjusted in Adobe Photoshop 2015 for better visualisation, and figures assembled in Adobe Illustrator 2015. Quantification of double fluorescent in situ in Fig. 2 was realised using Slidebook 6 (Intelligent Imaging Innovations). Briefly, after background subtraction of the image, an automatic threshold was used to determine the percentage of positive surface area (i.e. pixels above threshold) belonging to the hand-drawn mandibular area. Average and t-test statistics were calculated using Excel (Microsoft).

We thank R. Erzurumlu and G. Levi for discussions at an early stage of the project. We also wish to thank F. Wang for the kind gift of probes and S. Ducret for excellent technical assistance.