A Simple Method for 3D Analysis of Immunolabeled Axonal Tracts in a Transparent Nervous System

Clearing techniques have been developed to trans-parentize mouse brains, thereby preserving 3D structure, but their complexity has limited their use. Here, we show that immunolabeling of axonal tracts followed by optical clearing with solvents (3DISCO) and light-sheet microscopy reveals brain connectivity in mouse embryos and postnatal brains. We show that the Robo3 receptor is selec-tively expressed by medial habenula axons forming the fasciculus retroﬂexus (FR) and analyzed the development of this commissural tract in mutants of the Slit/Robo and DCC/Netrin pathways. Netrin-1 and DCC are required to attract FR axons to the


In Brief
Clearing techniques have recently been developed to look at mouse brains, but they are complex and expensive. Belle et al. now describe a simple procedure that combines immunolabeling, solventbased clearing, and light-sheet fluorescence microscopy. This technique allows large-scale screening of axon guidance defects and other developmental disorders in mutant mice.

INTRODUCTION
3D imaging of solvent-cleared organ (3DISCO) is a simple solvent-based clearing method used for transparentizing the brain of adult transgenic mice expressing fluorescent proteins such as GFP (Ertü rk et al., 2012a) and Alexa-conjugated axonal tracers (Ertü rk et al., 2012a(Ertü rk et al., , 2012b. In combination with light-sheet fluorescence microscopy (LSM), 3DISCO allows one to quickly generate 3D images of axonal tracts (Dodt et al., 2007;Ertü rk et al., 2012aErtü rk et al., , 2012b. However, the fluorescence of GFP and related proteins rapidly vanishes after clearing. A handful of clearing methods using solvent-free reagents have since been described with which the fluorescence is better preserved (Chung et al., 2013;Hama et al., 2011;Ke et al., 2013;Susaki et al., 2014;Yang et al., 2014), but they are more complex technically. Clearing takes several days or even weeks and requires large volumes of expensive reagents, such as polymers, that impregnate the samples. Moreover, the number of fluorescent mouse lines that have been validated with these methods is still limited (in most cases, only Thy1-YFP transgenic lines were tested), and in any case, using them to study axonal projections in mutant mice would require time-consuming intercrosses. Therefore, we reasoned that there was still a need for simple, rapid, and inexpensive methods to clear and image brain samples.
We first thought that performing whole-mount immunostaining prior to clearing could be a good strategy, as this would alleviate the use of fluorescent reporters. Immunostaining is compatible with clarity (Chung et al., 2013), CUBIC (Susaki et al., 2014), and PACT (Yang et al., 2014) but is performed after clearing and therefore lengthens these procedures. We show here that 3DISCO clearing can be performed after whole-mount immunostaining on embryo or postnatal brain and preserves the activity of fluorescent dyes for several months. We also show that Robo3, a receptor of the roundabout family , is a unique marker of medial habenula (mHb) axons, which extend through the fasciculus retroflexus (FR) to the interpeduncular nucleus (IPN; Beretta et al., 2012). The FR exists in all vertebrate embryos (Figdor and Stern, 1993) and primarily targets the IPN, an unpaired structure extending across the ventral midline at the midbrain/hindbrain boundary (Beretta et al., 2012). FR axons pathfinding in the thalamus is controlled by Sema3F/Neuropilin-2 repulsion (Chen et al., 2000;Sahay et al., 2003). In vertebrates, the behavior of FR axons at the ventral midline is rather unique, as they cross it multiple times (Ramon y Cajal, 1911;Bianco et al., 2008;Iwahori et al., 1993). Surprisingly, the role of guidance cues such as Slits and Netrin-1 in midline crossing of FR axons had not been studied. (E) Ventral view of the brain of an E13 embryo labeled with anti-Foxp2 antibodies. The entire expression pattern of FoxP2+ neurons can be observed in a single brain. Foxp2 is highly expressed in the inferior olivary nucleus (ION), Purkinje cells in the cerebellum (Cer), neocortex (Cx), and anterior olfactory nucleus (AON).
(legend continued on next page)

Immunolabeling of Mouse Embryos Is Compatible with 3DISCO Clearing
First, we tested the technique on embryonic day 12 (E12)-E14 mouse embryos immunostained with antibodies against choline acetyl transferase (ChAT) to label motor projections and transient axonal glycoprotein 1 (TAG-1/Contactin-2), which is expressed by many axons (Yamamoto et al., 1986), including sensory ganglia axons ( Figures 1A-1D). The precise 3D pattern of motor and sensory projections could be visualized, including oculomotor nerves and limb innervation (Movies S1, S2, and S3). Postacquisition treatments using Imaris allowed us to obtain high-resolution images of regions of interest. The samples, or part of it, such as a limb ( Figure 1D; Movie S4), could be optically sliced in all orientations, thereby allowing one to generate classic sagittal, horizontal, and coronal sections with a single embryo. Next, we used anti-FoxP2 (forkhead box P2) antibody, which recognizes a transcription factor expressed by neurons in multiple brain areas (Ferland et al., 2003;Fujita and Sugihara, 2012). The comprehensive distribution of FoxP2 immunoreactive nuclei was revealed ( Figure 1E). Notably, automatic 3D counting of the number of FoxP2+ neurons could also be done (data not shown).
Transgenic mice expressing fluorescent proteins are increasingly used to study axonal circuits and gene expression patterns.
To determine if the quenching of fluorescent proteins after 3DISCO could be overcome by antibody staining, we performed anti-GFP staining. Tau-lox-Stop-lox-mGFP-IRES-nls-lacZ mice (Tau mGFP ), which express a membrane-tethered GFP in axons following Cre-mediated recombination (Hippenmeyer et al., 2005), were crossed with Wnt1:Cre mice. In this line, Cre recombinase is targeted to some hindbrain and spinal cord commissural neurons and sensory ganglia (Danielian et al., 1997). In E11.5 Wnt1:Cre;Tau GFP embryos, GFP-immunoreactive axons could be followed throughout the CNS and peripheral nervous system (Figures 1F and 1G; Figure S1A; Movies S5 and S6). Likewise, anti-dsRed immunostaining was performed on E14 Ptf1a: Cre ERTM ;Rosa26 dTomato embryos that had received tamoxifen at E12.5. In this line, Cre is activated upon tamoxifen injection in subsets of neurons in the CNS (Kopinke et al., 2012), and this induces the expression of a red fluorescent protein from the Tomato transgene inserted into the ubiquitous Rosa locus (Madisen et al., 2010). After anti-dsRed immunostaining and clearing, the distribution of Tomato-positive neurons, derived from Ptf1a+ progenitors, could be observed, including Purkinje cells in the cerebellum ( Figure 1H). This showed that 3DISCO, in combination with immunostaining, can bypass the problem of native fluorescence instability. Last, we performed immunostaining against Robo3, a transmembrane receptor of the round-about family, transiently expressed by commissural axons in the mouse hindbrain and spinal cord (Marillat et al., 2002;Sabatier et al., 2004). In the E11 spinal cord, the whole array of commissural axons crossing the floor plate was observed in the spinal cord and hindbrain ( Figure 1I; Figure S1B; Movies S7 and S8).

Robo3 Is a Selective Marker of the Fasciculus Retroflexus
We next performed whole-mount anti-TAG1 immunostaining on E16 brain and could observe that the 3D framework of TAG-1+ axons such as the lateral olfactory tract, anterior commissure, and commissural axons in the hindbrain (Figure 2A). By contrast, at E16, Robo3 was still detected in the hindbrain, such as in latemigrating pontine neurons, but it was absent from the forebrain ( Figure 2B; Movie S9). Surprisingly, only the FR was immunoreactive for Robo3 in the diencephalon ( Figures 2B-2E). Robo3 was only expressed in the mHb, as previously described (Quina et al., 2009;Schmidt et al., 2014; data not shown). FR axons cross multiple times and zigzag at the ventral midline (Ramon y Cajal, 1911;Iwahori et al., 1993 ;Bianco et al., 2008). This feature was seen on 3D images and following optical slicing ( Figure 2E; Movie S10). The presence of Robo3 in E16 FR axons, which reach the ventral midline around E13-E14 (Funato et al., 2000), was unexpected, as this receptor is usually downregulated in postcrossing axons (Marillat et al., 2002;Sabatier et al., 2004). Tissues shrink after clearing with 3DISCO (Ertü rk and Bradke, 2013), but the FR length ( Figure S1C) was similar between E16 brains (862.9 mm ± 10.5 SEM; n = 20 E16 brains; p = 0.93; one-sample t test). It was also equivalent between the right (867.1 mm ± 14.8 SEM) and left (858.7 mm ± 15.3 SEM; p = 0.34; paired t test) side. This demonstrates that the shrinkage is homothetic and that comparing relative dimensions between cases is possible. Although LSM is the most rapid and efficient method to image cleared samples (see also Tomer et al., 2014), confocal microcopy could be used ( Figure S1D). Lastly, the same sample could be imaged several times without significant quenching of the fluorescence, which was only slightly diminished 6 months after clearing ( Figure S1E). We next attempted to perform whole-mount Robo3 immunostaining and 3DISCO clearing on postnatal brain (postnatal day 0 [P0] and P5). This revealed that FR axons still expressed high levels of Robo3 and that no other tracts were labeled in the CNS ( Figures  2F-2H). However, Robo3 was not expressed in the adult mHb (data not shown). The FR could also be stained with antibodies against the netrin-1 receptor deleted in colorectal cancer (DCC) and TAG-1 (Figures 2J and 2K; Figure S1F), as previously shown (Wolfer et al., 1994;Yamamoto et al., 1986;Funato et al., 2000;Schmidt et al., 2014). Therefore, we next studied FR structure in a collection of knockout mice for axon guidance molecules involved in midline crossing.

3D Analysis of the Development of the FR in Midline Guidance Mutants
Previous studies showed that Netrin-1 attracts FR axons and that the FR is disorganized in Netrin-1 and DCC mutants (Funato et al., 2000;Schmidt et al., 2014). However, the phenotypic analysis of Netrin-1/DCC knockout embryos was done on cryostat sections that only provide an incomplete visualization of the guidance defects. Robo3 immunostaining and 3DISCO clearing revealed that the FR was severely perturbed in E16 DCC and Netrin-1 knockout embryos (n = 8 and n = 6, respectively). Interestingly, this method allowed us to compare FR axonal defects between cases and score embryos as done in invertebrate species and zebrafish. In each case, the position and orientation of all the Robo3+ fascicles was determined.
In all DCC knockout embryos (n = 8/8), a large axonal bundle projected rostrally from the mHb instead of caudally and a second one grew along the normal pathway to the IPN level (Figures 3D-3I; Movie S11). These axons passed the IPN and extended along the midline. Floor plate crossing was severely reduced in all cases, although a few axons still crossed in six of eight cases ( Figure 3F; Figures S2D and S2E). In five of the eight embryos, a small tract projected dorsally from the mHb and aberrant crossing of the dorsal midline at the level of the mHb was observed. Overall, the embryos could be grouped in two main categories (containing five and three embryos, respectively) based on their combination of axonal defects.
Next we studied Netrin-1 knockout embryos ( Figures 3J-3O). The spectrum of Robo3+ FR axon guidance defects was distinct from DCC knockouts. Compared to a single tract in wild-type (n = 5/5) and Netrin-1 +/À (n = 7/7) embryos, the mHb projections were strongly defasciculated in Netrin-1 À/À embryos (n = 6/6), but cases were heterogeneous. In five out of six Netrin-1 À/À embryos, a fascicle failed to grow toward the midline and projected caudally from the mHb, and in two out of six cases, a small tract extended rostrally from the mHb (Figures 3J-3O; Movie S12). Like in DCC knockouts, dorsal midline crossing was observed (four out of six cases; Figures S2B and S2C). All other axons extended along the normal FR pathway, but one or two large fascicles left the main tract dorsally before the IPN. The remaining axons reached the IPN and formed a commissure (six out of six embryos) from which a few axons extend pass the IPN parallel to the ventral midline (two out of six). In a single embryo ( Figures 3M-3O), a fascicle of axons extended ventrorostrally instead of caudally when approaching the IPN. These data show that although a pattern of FR axon guidance defects could established, their combination was almost unique to each embryo. This confirms that Netrin-1 and DCC play a major role in FR axon guidance but indicates that their role is not limited to floor plate crossing.
Habenula neurons and FR axons express Robo1 and Robo2 receptors and Slit ligands (Marillat et al., 2002;Schmidt et al., 2014), but the consequence of Slit or Robo loss of function on FR development had not been studied. Therefore, we applied our clearing procedure to the analysis of FR organization in Slit and Robo mutants. In Robo À/À ;Robo À/À double knockouts (n = 6), FR axons formed a single tract and reached the IPN (Figures 4A-4C; Movie S13), where they defasciculated into smaller bundles. Some axons crossed the midline, but many remained on the ipsilateral side without extending further caudally (n = 6/ 6). In only one embryo, a few axons crossed the dorsal midline at the level of the mHb (data not shown). Next, we used anti-Tag1 immunostaining to study the organization of FR projections in Robo3 knockouts (n = 3). FR axons extended to the IPN level and crossed the floor plate ( Figures 4D-4F; Movie S14), but they next turned back toward the midline and coalesced at the midline. Together, the FR wiring defects in Robo mutants were unexpected and distinct from what has been described for other commissural tracts (see Discussion). We next studied various combinations of Slit knockouts, including a conditional Slit2 lox allele. To delete Slit2 from the floor plate, Slit2 lox mice were crossed to the Shh:Cre line, in which Cre is highly expressed throughout the floor plate. In Slit1 À/À knockouts (n = 1) and single and compound heterozygous controls (n = 3), Robo3+ FR axons projected as in wild-type embryos (Figures 4G-4I and data not shown). However, severe midline crossing defects were observed in Shh:Cre;Slit1 À/À ;Slit2 lox/lox and Shh:Cre;Slit1 À/À ; Slit2 lox/lox ;Slit3 À/À embryos (n = 2 for each genotype). In both mutants, FR axons projected to the IPN, but their growth at the midline was perturbed. In Shh:Cre;Slit1 À/À ;Slit2 lox/lox embryos, FR axons defasciculated at the IPN level but then followed the floor plate in both directions ( Figures 4J-4K; Movie S15) toward the diencephalon or hindbrain. In Shh:Cre;Slit1 À/À ;Slit2 lox/lox ; Slit3 À/À embryos, each FR divided into two branches that crossed the midline and joined axons from the contralateral FR to form two commissures. In addition, a bundle of axons escaped the anterior commissure to grow along the ventral midline ( Figure 4O; Movie S16). Together, these results show that 3DISCO clearing after whole-mount immunostaining reveals unexpected axonal defect in midline mutants.

DISCUSSION
Whole-Mount Immunostaining and 3DISCO Clearing: An Optimal Recipe for Studying Brain Connectivity? Embryos from most invertebrate species, such as Drosophila or C. elegans, or some vertebrates, including zebrafish or Xenopus, are optically transparent. Therefore, one can visualize and reconstruct entirely axonal tracts and their arborization using specific transgenes or whole-mount immunostaining. This has facilitated the phenotypic characterization of axon guidance mutants and large-scale genetic screens (Baier et al., 1996;Kolodkin et al., 1993;Seeger et al., 1993;Zallen et al., 1998). By contrast, mammalian embryos and postnatal brains are opaque and current imaging techniques mostly restrict the 3D analysis of axonal connections to young embryos or brain samples at immature stages. Older brains need to be cut, which is time consuming and only provides fragmentary information on brain connectivity. Serial electron microscopy was used to reconstruct neuronal networks in small pieces of tissue (Helmstaedter et al., 2013), but the technique is still in its infancy and inaccessible to most laboratories. Using it for genetic screens or to assess interspecimen variability would also be extremely challenging.
A few forward genetic screens for axon guidance mutants have been performed in mice using N-ethyl-N-nitrosourea mutagenesis (Lewcock et al., 2007;Merte et al., 2010) but restricted to E11.5-E12.5 embryos. We show here that our imaging method is  (1) and another one caudally (2). Other FR axons (3) reach the IPN, but only a few cross the midline (arrowhead in F). Note that many axons cross the dorsal midline at the level of the habenula (arrowhead in E). (G-I) FR defects in another DCC À/À embryo (#2). The rostral tract is also present (1). Other FR axons reach the IPN (3) but fail to cross the midline (arrowhead in I). Axons do not cross the dorsal midline (arrowhead in H).
(legend continued on next page) fast, inexpensive, and scalable enough for processing a large number of embryos. Moreover, it is compatible with multiple labeling and therefore several types of axonal tracts can be stained in a single embryo. The procedure is technically simple and does not require any specific device (perfusion is not even needed). Another main advantage of this immunostaining/clearing procedure over existing ones is its low cost (about 10 Euros [V] per sample, including 3V of antibodies and only 3.5V of clearing solutions), which can be reduced if several embryos are processed simultaneously. Last, brains and embryos can be stored for weeks before staining and after clearing. As with every immunohistochemical procedure, the staining protocol will have to be optimized for each antibody and specific needs. Our work also confirms that LSM is a very powerful imaging technique for 3D analysis of brain connectivity.

Unique Features of Midline Guidance in the Habenular System
The habenular complex is an important relay between the limbic forebrain and caudal brain nuclei, in particular monoaminergic ones (Herkenham and Nauta, 1979;Hikosaka, 2010). The medial habenula project mostly to the IPN (Kuhar et al., 1975) and plays a role in nicotine intake (Fowler et al., 2011) and anxiety (Yamaguchi et al., 2013), among other emotional behaviors. The habenula has fascinated neuroatomists for being an asymmetric brain structure in most vertebrate species, excepting mammals. This lateralization (in size, neurotransmitter content, and connectivity) is most obvious in anamniotes, including fish (Amo et al., 2010). In zebrafish, the axons from both sides cross the midline multiple times in a different pattern depending on their lateral origin in the dorsal habenula (Bianco et al., 2008). This unusual, and seemingly unique, midline recrossing behavior was also described in mammals (Ramon y Cajal, 1911;Iwahori et al., 1993). This raised the question of the underlying axon guidance mechanisms, as in other commissural systems, midline recrossing is prevented (Ché dotal, 2011). We show here that Slit/Robo signaling plays a major role in the control of midline recrossing. Although FR pathfinding to the midline is not affected in Robo1/2, Robo3 knockouts or in mice lacking all Slit expression at the floor plate, their final arborization is severely perturbed. Midline recrossing is strongly reduced in Robo1/2 double knockouts. This is somehow counterintuitive, as one would have expected axons to recross or coalesce at the midline, as described in other commissural systems (Farmer et al., 2008;Jaworski et al., 2010). Likewise, the FR phenotype of Robo3 knockout is unexpected, as axons cross the midline and even fail to leave it. Moreover, Robo3 is still expressed by FR axons several days after they crossed the midline, unlike in other commissural systems (Marillat et al., 2004;Sabatier et al., 2004). The analysis of Slit triple knockouts also reveals that Slits control the defasciculation of FR axons at the midline and prevent them from growing caudally and rostrally, thereby confining them to the IPN level.
By contrast, the analysis of Netrin-1 and DCC knockouts demonstrates that they are essential for guiding FR axons to the midline and promoting crossing. However, they also favor FR axon fasciculation and prevent axons from crossing the dorsal midline or from extending rostrally, as previously described (Schmidt et al., 2014). The distinct fascicles that form in these mutants might reflect the molecular diversity of the adult medial habenula (Yamaguchi et al., 2013). This heterogeneity of guidance defects is unlikely to rely on a differential expression of unc5 receptors in mHb neurons, as they only express unc5a, and homogeneously (van den Heuvel et al., 2013).
Interestingly, the spectrum of axon defects is specific for each mutant line, and a common signature of pathfinding errors can be established. However, a clear interindividual variability exists in each knockout line. For Netrin-1 knockouts, this could be related to a hypomorphic allele (Serafini et al., 1996), but this should not be the case for the other mutants. This suggests that FR axon rewiring in these mutant is partially stochastic.
In conclusion, this method will facilitate the development of large-scale forward genetic screens and of 3D atlases of immunolabeled tissues. It will be a valuable tool to implement ongoing initiatives aimed at establishing the brain connectome (see, for instance, http://connectivity.brain-map.org/ and http://www.gensat.org/index.html).
Embryos were collected at E12, E14, E16, and E18 and transferred to ice-cold PBS 1X (Invitrogen). From E16, the nervous system was dissected and a small hole was made in the telencephalon to facilitate the diffusion of the antibodies. At postnatal ages (P0 to P8), brains were dissected and fixed by immersion in 4% paraformaldehyde (PFA; Merck) for 3 hr at room temperature (RT) or overnight at 4 C. Samples were kept at 4 C in PBS 1X until use.
All animal procedures were carried out in accordance to institutional guidelines (UPMC and INSERM).
(J-L) FR defects in a Netrin-1 À/À mutant (#1). Some axons project rostrally over the medial habenula (1) without crossing the midline, a bundle grow caudally from the habenula (2), and other axons extend along the normal FR pathway (3). Two bundles leave the main tract to project dorsally before reaching the IPN (4 and 5), whereas other axons reach the IPN level. They form a large commissure (6), but a few axons extend more caudally parallel to the floor plate, and only a few cross it (arrowhead in K and L). (M-O) Distinct FR guidance defects in another Netrin-1 À/À mutant (#2). The dorsal defasciculation at habenula level is also seen (1), but most axons grow along the normal FR pathway (3). Upon reaching the floor plate, a large bundle is deflected dorsally (4) and a smaller ventral fascicle also forms in the ventral diencephalon (7). Other axons reach the IPN level and cross the midline to form a small commissure (6)  Whole-Mount Immunostaining The procedure was similar for single and multiple labeling. Samples were first incubated at RT on a rotating shaker in a solution (PBSGT) of PBS 1X containing 0.2% gelatin (Prolabo), 0.5% Triton X-100 (Sigma-Aldrich) and 0.01% thimerosal (Sigma-Aldrich) for 3 hr (E12), 24 hr (E14-E18 and P0), or 48 hr (P5 and P8). Samples were next transferred to PBSGT containing the primary antibodies (Table S1) and placed at 37 C, with rotation at 100 rpm, for 3 days (E12), 1 week (E14, E16, and E18), 10 days (P0), or 14 days (P5 and P8). This was followed by six washes of 30 min in PBSGT at RT. Next, samples were incubated in secondary antibodies (Table S2) diluted in PBSGT overnight (E12-E18) or for 2 days (P0-P8) at 37 C. After six washes of 30 min in PBSGT at RT, samples were stored at 4 C in PBS until clearing.

Tissue Clearing
We used the 3DISCO clearing procedure (Ertü rk et al., 2012b) and slightly adapted it to our samples. All incubation steps were performed at RT in a fume hood, on a tube rotator (SB3, Stuart) at 14 rpm, using a 15 ml centrifuge tube (TPP, Dutscher) covered with aluminum foil to avoid contact with light. Samples were first dehydrated in a graded series (50%, 80%, and 100%) of tetrahydroflurane (THF; anhydrous, containing 250 ppm butylated hydroxytoluene inhibitor, Sigma-Aldrich) diluted in H 2 O, during 1hr (E12-P0) or 90 min (P5) for each step. This was followed by a delipidation step of 20 min (E12 to P0) or 40 min (P5-P8) in dichloromethane (DCM; Sigma-Aldrich). Samples were transferred to 100% DCM until they have sunk. Finally, samples were cleared overnight in dibenzylether (DBE; Sigma-Aldrich). Samples should be stored in brown glass vial (Rotilabo, Roth) filled with DBE, in the dark and at RT. THF, DCM, and DBE are toxic (gloves must be worn at all steps) and flammable. Waste should be treated and eliminated accordingly. DBE must be stored in glass containers.

Imaging
Ultramicroscopy 3D imaging was primarily performed with an ultramicroscope (LaVision BioTec) using ImspectorPro software (LaVision BioTec). The light sheet was generated by a laser (wavelength 488 or 561 nm, Coherent Sapphire Laser, LaVision BioTec) and two cylindrical lenses. A binocular stereomicroscope (MXV10, Olympus) with a 23 objective (MVPLAPO, Olympus) was used at different magnifications (1.63, 43, 53, and 6.33). Samples were placed in an imaging reservoir made of 100% quartz (LaVision BioTec) filled with DBE and illuminated from the side by the laser light. Different dipping caps were used to image large samples with a high working distance or small samples with a low working distance. Images were acquired with a PCO Edge SCMOS CCD camera (2,560 3 2,160 pixel size, LaVision BioTec). The step size between each image was fixed at 1 mm.
The acquisition time with the ultramicroscope depends on the number of lasers (one side or two sides) used to generate and focus the light sheet. It took only 10 min to image (1,488 sections) the habenula and FR with a single light source. For the entire brain, the two light sources were used and the total acquisition time (1,856 sections) was $1 hr. Each resulting imaris (.ims) file was $15 Gb in size at a 1 mm z resolution (16-bit images).

Confocal Microscopy
For imaging with an upright confocal microscope (Olympus FV1000), samples were placed on a glass slide, in a homemade PDMS cuvette (DBE resistant) with a rim of 5 mm (Sylgard 184/Silicone elastomer, Dow Corning). The cuvette was filled with DBE and covered with a glass coverslip. Images were obtained with a 103 objective (Olympus UPlanSApo 103/0.40 numerical aperture objective Royal Microscopical Society, infinity corrected 0.31 mm working distance). Each individual image (.oif file) was 1.08 Gb from confocal for 1 mm z resolution. The acquisition time was of about 4 hr for the FR and therefore much longer than with an ultramicroscope. 3D Imaging and Image Processing Images, 3D volume, and movies were generated using Imaris x64 software (version 7.6.1, Bitplane). Stack images were first converted to imaris file (.ims) using ImarisFileConverter. File size was next reduced to 8 bits. 3D reconstruction of the sample was performed using ''volume rendering'' (Imaris). The sample could be optically sliced in any angle using the ''orthoslicer'' or ''obliqueslicer'' tools. Air bubbles and crystals that might form at the surface of the samples could be eliminated using the ''surface'' tool by creating a mask around the each volume. 3D pictures and movies were generated using the ''snapshot'' and ''animation'' tools. Finally, images were cropped and, if required, their brightness was adjusted evenly using Photoshop CS4 (Adobe). Movies legends were generated using iMovie 10.0.2.
Alternatively, movies and 3D analysis could be done with free software such as Vaa3D (http://www.vaa3d.org) or Fiji 3D project plugin (Schindelin et al., 2012). However, the quality of the 3D images was not as good as with Imaris (data not shown).
The distance separating the base of the habenula from the IPN was obtained using the ''measurement'' tool (Imaris), and statistical analysis was performed with Prism 6 (GraphPad).

SUPPLEMENTAL INFORMATION
Supplemental Information includes two figures, two tables, and 16 movies and can be found with this article online at http://dx.doi.org/10.1016/j.celrep.2014. 10.037.

ACKNOWLEDGMENTS
This work was supported by grants from the Fondation pour la Recherche Mé dicale (Programme é quipe FRM; reference DEQ20120323700) and the Agence Nationale de la Recherche (ANR-08-MNP-030) to A.C. It was also performed in the frame of the LABEX LIFESENSES (reference ANR-10-LABX-65) supported by French state funds managed by the ANR within the Investissements d'Avenir programme under reference ANR-11-IDEX-0004-02. F.H. was the recipient of fellowships from the Wings for Life Foundation, the Institute for Research on Paraplegia (IRP), and the DFG. The ultramicroscope was purchased with a grant from the Fé dé ration pour la Recherche sur le Cerveau (FRC, Programme espoir en tête, Rotary). We also thank Dr. Domna Karagogeos for providing one of the anti-TAG-1 antibodies and Dr. Chris Wright for the Ptf1a:Cre ERTM line.