Peculiar protrusions along tanycyte processes face diverse neural and nonneural cell types in the hypothalamic parenchyma

Abstract Tanycytes are highly specialized ependymal cells that line the bottom and the lateral walls of the third ventricle. In contact with the cerebrospinal fluid through their cell bodies, they send processes into the arcuate nucleus, the ventromedial nucleus, and the dorsomedial nucleus of the hypothalamus. In the present work, we combined transgenic and immunohistochemical approaches to investigate the neuroanatomical associations between tanycytes and neural cells present in the hypothalamic parenchyma, in particular in the arcuate nucleus. The specific expression of tdTomato in tanycytes first allowed the observation of peculiar subcellular protrusions along tanycyte processes and at their endfeet such as spines, swelling, en passant boutons, boutons, or claws. Interestingly, these protrusions contact different neural cells in the brain parenchyma including blood vessels and neurons, and in particular NPY and POMC neurons in the arcuate nucleus. Using both fluorescent and electron microscopy, we finally observed that these tanycyte protrusions contain ribosomes, mitochondria, diverse vesicles, and transporters, suggesting dense tanycyte/neuron and tanycyte/blood vessel communications. Altogether, our results lay the neuroanatomical basis for tanycyte/neural cell interactions, which will be useful to further understand cell‐to‐cell communications involved in the regulation of neuroendocrine functions.

This heterogeneity among tanycyte populations allows them to participate in the regulation of numerous neuroendocrine functions-such as the control of energy balance, reproduction and seasonal adaptationsthrough their diverse cellular properties including blood/brain traffic controllers, metabolic sensors, modulators of cell function, and neural stem/ progenitor cells (Langlet, 2014(Langlet, , 2019Prevot et al., 2018). However, their interactions with different neural populations and their integration within different neuronal networks to ensure these regulations are still poorly described. The aim of this study is to define the neuroanatomical basis for tanycyte/neural cell interactions. Focusing on tanycytes lining the ARH, the VMH, and the DMH, we observed along their processes peculiar protrusions, of which we extensively characterized the morphometry, the partners and the composition. Altogether, our results allowed us to speculate about specific tanycyte-to-neural cell communications.

| Animals, and tdTomato expression in tanycytes
Two-to-four-month-old male Rosa26-floxed stop tdTomato mice (n = 14, initially obtained from Charles River), two-month-old male Rosa26-floxed stop tdTomato:NPY-GFP mice (n = 4) and two-monthold male Rosa26-floxed stop tdTomato:POMC-GFP mice (n = 3) were used in this study. Animals were housed in groups (from 2 to 5 mice per cage), and maintained in a temperature-controlled room (at 22-23 C) on a 12 hr light/dark cycle with ad libitum access to chow diet (Diet 3,436; Provimi Kliba AG, Kaiseraugst, Switzerland). All animal procedures were performed at the University of Lausanne, and were reviewed and approved by the Veterinary Office of Canton de Vaud. To induce tdTomato expression in tanycytes, TAT-CRE fusion protein (Excellgen, EG-1001) was stereotactically infused into the lateral ventricle (2 μl over 2 min at 2 mg/ml; at the coordinates from the bregma of anteroposterior = −0.3 mm; mediolateral = −1 mm and dorsoventral = −2.3 mm from the cortex surface) of ketamine/xylazine-anesthetized mice (100 mg/kg and 20 mg/kg, respectively) 72 hr before experiments. This injection through the lateral ventricle avoids local inflammation around the 3V and sparsely label 3V tanycytes, optimizing their morphological analysis.

| Tissue preparation
For immunostaining, mice were anesthetized with isoflurane, and perfused transcardially with 0.9% saline followed by an ice-cold solution of 4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4. Brains were quickly removed, postfixed in the same fixative for 2 hr at 4 C, and immersed in 20% sucrose in 0.1 M phosphate buffered saline (PBS) at 4 C overnight. Brains were finally embedded in ice-cold OCT medium (optimal cutting temperature embedding medium, Tissue Tek, Sakura) and frozen on dry ice or in liquid nitrogen-cooled isopentane.
To visualize blood vessels using fluorescent dextran, mice were given i.v. injections of 70 kDa fluorescein isothiocyanate-dextran (25 mg/ml, Cat Nb 90,718, lot # BCBV4422, Sigma, France) in sterile 0.9% saline (100 μl) into the tail vein and killed by decapitation 5 min later. Brains were then immersed in an ice-cold solution of 4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4, for 24 hr at 4 C, followed by 20% sucrose in 0.1 M PBS at 4 C overnight. Brains were finally embedded in ice-cold OCT medium and frozen on dry ice or in liquid nitrogen-cooled isopentane.
For electron microscopy, mice were anesthetized with isoflurane, and perfused transcardially with 0.9% saline followed by an ice-cold solution of 2% paraformaldehyde/2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4. Brains were quickly removed, postfixed in the same fixative overnight at 4 C. Two hundred micrometer-thick hypothalamic slices were then cut using vibratome. TdTomato fluorescence in tanycytes was then acquired using ZEISS Axio Imager.M2 microscope equipped with ApoTome.2 in order to give coordinates to each protrusion in the slice. Afterwards, the samples were incubated in 2% (wt/vol) osmium tetroxide and 1.5% (wt/vol) K4[Fe(CN)6] in 0.1 M PB buffer for 1 hr, following by one-hour incubation in 1% (wt/vol) tannic acid in 0.1 M PB buffer. Subsequently, brain slices were incubated in 1% (wt/vol) uranyl acetate for 1 hr and dehydrated at the end of standard gradual dehydration cycles in ethanol. Samples were flat embedded in Epon-Araldite mix (Kolotuev, 2014;Kolotuev, Schwab, & Labouesse, 2009). All procedures were performed at room temperature.

| Immunohistochemistry
Brains were cut using a cryostat into 20-μm-thick coronal, horizontal, or sagittal sections and processed for immunohistochemistry as described previously (Langlet, Mullier, et al., 2013). For most of the antibodies, slide-mounted sections were (a) blocked for 30 min using a solution containing 4% normal goat serum and 0.3% Triton X-100; (b) incubated overnight at 4 C with primary antibodies (Table 1) followed by 2 hr at room temperature with a cocktail of secondary

| Antibody characterization
All primary and secondary antibodies used are listed in Tables 1 and 2 respectively. These antibodies are in the Antibody Registry.
The chicken polyclonal antibody to VIM (Vimentin) (Millipore Cat# AB5733, RRID:AB_11212377) produced a pattern of staining associated to tanycytes, ependymal cells and endothelial cells, similar to that described elsewhere in the literature Langlet, Mullier, et al., 2013Parkash et al., 2015). This expression profile replicates the pattern of mRNA expression determined by in situ hybridization in the adult mouse (Allen brain atlas).
The rabbit polyclonal antibody to RFP (Red fluorescent protein) (Rockland Cat# 600-401-379, RRID:AB_2209751) was prepared against RFP fusion protein corresponding to the full length amino acid sequence (234aa) derived from the mushroom polyp coral Discosoma.
The antibody labeled cells expressing tdTomato.
The mouse monoclonal antibody to NeuN (Neuron-specific nuclear protein) (Millipore Cat# MAB377, RRID:AB_2298772) produced a pattern of staining associated to neuronal cells, similar to that described elsewhere in the literature (Z. Liu & Martin, 2006). According to the man- The rabbit polyclonal antibody to GFAP (Glial Fibrillary Acidic Protein) (Agilent Cat# Z0334, RRID:AB_10013382) produced a pattern of staining associated to astrocytes, and some tanycytes lining the dorsal ARH, similar to that described elsewhere in the literature (Langlet, Mullier, et al., 2013). Specificity of this antibody in mouse brain was also confirmed by immunohistochemistry in GFAP knockout mice (Hanbury, Ling, Wuu, & Kordower, 2003).
The rabbit polyclonal antibody to OLIG2 (Oligodendrocyte Tran- and produced a pattern of staining similar to that described elsewhere in the literature (Fekete et al., 2015). According to the manufacturer, this antibody recognizes the expected band at 60 kDa by western blot of synaptic vesicle fraction of rat brain.
The rabbit polyclonal antibody to VGLUT2 (Solute Carrier Family 17 Member 6) (Synaptic Systems Cat# 135403, RRID:AB_887883) targets the vesicular glutamate transporter 2 (VGLUT2) and produced a pattern of staining similar to that described elsewhere in the literature (Toyoshima et al., 2009). According to the manufacturer, this antibody recognizes the expected band at 62 kDa by western blot of synaptic vesicle fraction of rat brain.
The rabbit polyclonal antibody to NKB (Neurokinin B) (Ciofi P. laboratory, IS-39, RRID: AB_2819032) was prepared against NKB precursor peptide-2 coupled to human serum albumin by glutaraldehyde. This antibody was characterized previously (Ciofi, Leroy, & Tramu, 2006): its labeling is not abolished by preabsorption with other tachykinins (neurokinin A (NKA) and substance P (SP)). Its expression profile replicates the pattern of mRNA expression determined by in situ hybridization in the adult mouse (Allen brain atlas).
The rabbit polyclonal antibody to TH (Tyrosine Hydroxylase)

| Microscopic imaging
Sections were analyzed using an ZEISS Axio Imager.M2 microscope, equipped with ApoTome.2 and a Camera Axiocam 702 mono (Zeiss, Germany Quintuple-ApoTome frames were collected in a stepwise fashion over a defined z-focus range corresponding to all visible fluorescence within the section: basically, multiple-plane frames were collected at a step of 0.3 μm while using x63 objective (between 35 and 45 frames per image) and 1 μm while using ×20 objective (between 4 and 10 frames per image). Weak deconvolution was finally applied on images following the acquisition. All images were then saved in .cvi, processed to get orthogonal and maximal intensity projections, and finally export in .tiff. Multiple-plane frames were collected at a step of 0.25 μm while using ×63 objective and 0.5 μm while using ×20 objective, over a defined zfocus range corresponding to all visible fluorescence within the section. All images were then saved in .lsm for Imaris ® analysis.

| Electron microscopy
The particularity of tanycyte endfeet analysis using electron microscopy is that they are rather large to be observed by a classical transmission electron microscopy (TEM) method for the volume acquisitions. Array tomography approach (Smith, 2018) consequently allowed us both to cover the large surface and make an efficient screening for the desired ROI. Moreover, as it is not a destructive technique such as Focused Ion Beam (FIB), it also permitted to go back to the area of interest and concentrate on desired details. To do so, polymerized flat blocks were trimmed using 90 diamond trim tool, and the arrays of 80 nm sections were obtained using 35 ATC diamond knife (Diatome, Biel, Switzerland) mounted on Leica UC6 microtome (Leica, Vienna). Sections were directly transferred to 2 × 4 cm pieces of silicon wafers using a modified array tomography procedure (Burel et al., 2018). During the sectioning phase, reliable landmarks were used to improve our chances to find tanycyte endfeet by applying a semi-correlative approach: our regions of interest (ROI) were defined before cutting by superimposing the images of the fluorescent acquisition from the vibratome sections with the images of the embedded samples (Burel et al., 2018;Kolotuev, 2014;Kolotuev et al., 2009). Wafers were analyzed using FEI Helios Nanolab 650 scanning electron microscope (Thermo Fischer, Eindhoven). The imaging settings were as follows: MD detector, accelerating voltage 2 kV, current 0.8 nA, dwell time 4-6 μs. Images were collected manually or using the AT module of MAPs program (Thermo Fischer, Eindhoven). Single images were aligned and reconstructed with the IMOD software package (Kremer, Mastronarde, & McIntosh, 1996).
For electron microscopy data interpretation, previous reports in the literature were used to recognize the different neural cell types based on their ultrastructural characteristics (Luse, 1956).

| Morphometric analysis
To quantitatively analyze tanycyte morphology and the subcellular protrusions observed along their process, three male mice were used.
Three-dimensional reconstructions of the image volumes were then prepared using Imaris ® visualization software to perform morphometric analysis. For ependyma/tanycyte ratio, the length of the ventricle occupied by tanycytes was reported to the total length of the ventricle: the length of the ventricle occupied by tanycytes (in μm) represents the distance from the bottom of the ventricle up to the last tanycyte measured using tdTomato fluorescence, whereas the total length of the ventricle (in μm) represents the distance from the bottom to the top of the ventricle measured using DAPI counterstaining. Three ratios per anteroposterior zone were used for quantification. For nucleus area occupied by tanycyte processes, the area of nucleus containing tanycyte processes was reported to the total area of the nucleus: the area of nucleus containing tanycyte processes (in μm 2 ) represents the area within the nucleus of interest measured by delineating tdTomato fluorescence, whereas the total area of the nucleus (in μm 2 ) represents the area of the nucleus of interest measured using DAPI counterstaining.
Two ratios per anteroposterior zone and per nucleus were used for quantification. For cell body analysis, the maximal length of four cell bodies per anteroposterior zone and per nucleus were measured on an anteroposterior, ventrodorsal, and mediolateral direction. For process thickness, the maximal diameter of four processes per anteroposterior zone and per nucleus was measured at the proximal, medial, and distal portion of the process. For protrusion analysis, each protrusion was first defined as Surface using Imaris ® software, and their surface area, volume and sphericity were then quantified using Imaris ® algorithm.

| Tanycyte partner analysis
To quantitatively analyze the proportion of tanycyte interactions with different neural cells, we first counted the number of tanycyte protrusions (e.g., swelling and boutons) in the region of interest, and then the number of these protrusions in contact with a neural partner. The cell identity of these partners was assessed by immunohistochemistry for known markers (i.e., HuC/HuD for neuronal cells, CD31 for vessels, GFAP for astrocytes, NG2 and MBP for oligodendrocytes, and TMEM19 for microglia), while NPY and POMC were visualized using transgenic reporter mice. The regions of interest (i.e., ARH, VMH, and DMH) were identified based on DAPI staining. The analysis was performed in three mice per staining, in 2 sections per anteroposterior zone.

| Statistical analysis
All values are expressed as means ± SEM. Data were analyzed for statistical significance with Graph Prism 5 software (Version 11.0), using one-way analysis of variance (ANOVA) followed by a Tukey's post hoc test or twoway ANOVA followed by a Bonferroni's post hoc test when appropriate.
p-values of less than .05 were considered to be statistically significant.

| Location and direction of tanycyte processes within the hypothalamic parenchyma
To examine the morphology of tanycytes lining the lateral wall of the 3V (commonly called dorsal β1 and α-tanycytes), we filled their cytoplasm with the red fluorescent tdTomato protein using a transgenic approach. To do so, TAT-CRE fusion protein was stereotactically infused into the lateral ventricle of tdTomato loxP/+ cre reporter mice to induce tdTomato expression in ependymoglial cells, including tanycytes, as described previously (Langlet, Levin, et al., 2013;Parkash et al., 2015) and confirmed by vimentin immunostaining (VIM, Figure 1a-c). As TAT-CRE mainly incorporates into cells located close to the site of injection (Langlet, Levin, et al., 2013), this injection in the lateral ventricle allowed us to sparsely label 3V tanycytes (Figure 1a-c), facilitating their morphological analysis. Using this approach, we confirm that 3V hypothalamic tanycytes are present from Bregma corresponding to Zone 1 (from bregma −1.3 to −1.6 mm), Zone 2 (from bregma −1.6 to −1.8 mm), Zone 3 (from bregma −1.8 to −2.1 mm) and Zone 4 (from bregma −2.1 to −2.5 mm) ( Figure S1). From a neuroanatomical point of view, these subdivisions were defined based on the shape of the ventricle, and the presence/absence of hypothalamic nuclei along the 3 V ( Figure S1). For some analyses, ARH was also divided into ventromedial ARH (vmARH) versus dorsomedial ARH (dmARH). This anteroposterior and ventrodorsal analysis first shows that tanycytes are mainly located at the bottom of the 3V and that the tanycyte/ependymal cell ratio progressively grows from the rostral to the caudal region of the brain reaching up to 60% of the ventricular wall occupied by tanycytes in Zones 3 and 4 ( Figure 1d). Secondly, we show that the entire ARH typi- their processes into the brain parenchyma following a dorsolateral trajectory followed by a lateral trajectory in Zones 3-4 (Figure 2e-g).

| Presence of subcellular protrusions along tanycyte processes
TdTomato fluorescent protein fills the entire cytoplasm of the cell: our approach consequently allowed us to study cell morphology in detail (Figure 3a-c). Tanycytes lining the lateral wall of the third ventricle share a similar shape composed of a somatic region, a long process and an endfoot (Figures 3a, 4a). The process may be additionally  (Figure 3j). In contrast, vmARH tanycyte processes are quite smooth, and end at the pial surface forming club-shaped endfeet laterally or fork-shaped endfeet more medially (Figures 3k and S1).

| Tanycyte protrusions in close proximity to different neural cells
We next analyzed which type of neural cells these protrusions-in particular sleeves, boutons, swellings and spines-are in contact with by using fluorescent dye and immunohistochemistry ( Figure 5, Table 3).
As previously described, the most frequently identified tanycyte partners are blood vessels visualized using i.v. injected fluorescent dextran (Figure 5a,b) or anti-CD31 antibodies (Figure 5e) (up to 71% association; Table 3). These associations occur through two different tanycyte protrusions: sleeve-like shapes formed by numerous tanycyte endfeet around the blood vessel, and boutons (arrows and empty arrows respectively, Figure 5a,b). In some cases, in particular in the vmARH, tanycyte processes were also observed surrounding a blood vessel before continuing their way into the brain parenchyma (data not shown).
Besides capillaries and neurons, associations with other neural cell types were also detected (Table 3). In particular, tanycytes contact astrocytes visualized by GFAP immunostaining: some tanycyte boutons end on the astrocyte cell body (Figure 5f), whereas astrocyte processes also appear to contact numerous neighboring tanycyte pro-  Table 3) but are also wrapped by their processes (Figure 5i,j, and l, Table 3).
Finally, tanycyte associations with glutamatergic and GABAergic terminals were analyzed using VGLUT2 and VGAT immunostainings, respectively ( Figure 6). Along tanycyte processes, swellings and boutons wrap GABAergic (Figure 6a,c) and, to a lesser extent, glutamatergic terminals in the ARH (Figure 6d-f). In contrast, at the proximal neck portion along the VMH and DMH, GABAergic and glutamatergic terminals end on tanycyte spines (Figure 6g-i and j-l, respectively).

| Tanycyte terminal boutons contact diverse arcuate neuronal subpopulations
As the entire ARH contains tanycyte processes, we next wanted to determine which arcuate neuronal populations are in contact with tanycyte protrusions. Both immunostaining and genetic mouse models reveal the presence of numerous contacts with NPY and POMC neurons (Figure 7a,b, respectively) (up to 24 and 10% association, respectively; Table 3), through tanycyte swellings as well as tanycyte boutons. A few contacts were also observed with TH-positive neurons (Figure 7c), and KNDy neurons (Figure 7d,e).

| Composition of tanycyte terminal boutons
To understand the putative function of these tanycyte protrusions, we next examined their composition using immunohistochemistry (Figures 8-11).
Labelings for vimentin, GFAP, and actin were first performed to identify the cytoskeleton proteins present in tanycyte boutons T A B L E 3 Proportion of tanycyte protrusions in contact with a neural partner along the anteroposterior axis (see Figure S1)  As α-tanycytes are considered as modulators of neuronal activity (Coppola et al., 2007;Lanfray et al., 2013), we also evaluated the pres- Finally, as tanycytes are involved in transport activity (Balland et al., 2014;Collden et al., 2015), we sought for vesicular system markers (i.e., caveolin, clathrin, CD9). Clathrin is located in some tanycyte boutons (Figure 11a-c), whereas caveolin is only present in blood-brain barrier (BBB) vessels (data not shown) but not in tanycyte endfeet (Figure 11d-f). As previously described (Horiguchi et al., 2019), exosomal marker CD9 is expressed by tanycytes: interestingly, CD9 is located in tanycyte endfeet (Figure 11g-i).

| Ultrastructural characterization of tanycyte terminal boutons
To support our immunohistochemical observations, we finally examined tanycyte terminal boutons using electron microscopy ( Figures 12-14, S2, and Videos S2,S3). Our regions of interest were defined before cutting by applying a semi-correlative approach (Kolotuev, 2014): we combined the images of tdTomato fluorescent F I G U R E 6 Tanycyte protrusions are in close proximity to synaptic terminals. (a-l) High-magnification z-stack images (×63) showing the distribution of tdTomato (red, a, d, g, and j), VGAT (green, b and h), VGLUT2 (green in e and k) immunoreactivity, and merge (c, f, i, and l) with DAPI counterstaining (blue, a-l). Images were acquired in coronal sections of the arcuate nucleus, in Zone 2. Pictures (a-l) are the maximal intensity projections of z-stack acquisition. Inset on the top in a-l panels shows the orthogonal view on the horizontal line; inset on the right in a-l panels shows the orthogonal view on the vertical line. acquired prior sample preparation with the images of the embedded samples. After sectioning, we found tanycyte endfeet throughout the brain parenchyma using morphological landmarks (i.e., ventricle and blood vessels), as well as ultrastructural characteristics of neural cells previously reported in the literature (Luse, 1956; 2019) ( Figure S2). Once tanycyte endfeet were recognized, the acquisition was done for the complete series of sections spanning a significant volume to finally build 3D reconstructions of tanycyte process and endfoot (Burel et al., 2018) (Videos S2 and S3).
As reported in our morphometric analysis, tanycyte processes are 1 μm thick, display some dilated portions and form 5 μm wide bouton-like endfeet (Figure 12a, Videos S2 and S3). These tanycyte boutons contact the basal lamina surrounding BBB microvessels, where they share the surface with other tanycyte endfeet (Figure 12), glial endfeet as well as pericytes (Figures 12-13). Interestingly, electron densities corresponding to junctional complexes are located at the site of contacts between tanycytes, glial cells, and pericytes (arrows in Figures 12 and 13). and GFAP (Figure 8a-f). In contrast, the endfeet mainly appear to be filled with a filamentous material (Figures 12 and 13). Among the cytoplasm material, elongated mitochondria run along the process and are present at the endfeet ( Figure 12). The endfeet also contain a variable number of ribosomes, and cisternae of endoplasmic reticulum (Figures 12 and 13). Interestingly, endoplasmic reticulum is mainly smooth, and closely contact the plasma membrane (Video S2). Finally, the main feature of tanycyte endfeet relies on the diversity of their vesicular system (Figures 12 and 13). Indeed, phagosome, doublemembrane vesicles, single membrane vesicles, multivesicular bodies, as well as dense-core vesicles are present in tanycyte boutons ( Figures 12 and 13). In particular, we report here multivesicular bodies at the interface with brain parenchyma, in particular close to mitochondria ( Figure 13b, Video S3).
Our electron microscopy 3D reconstructions lastly reveal peculiar associations with neurons. First, tanycyte endfeet directly contact neuronal cell body, close to their primary cilia, and separate them from blood vessels (Figure 12b, Video S2), as astrocyte endfeet do at the BBB. Moreover, as previously reported for tanycytes (Rodríguez et al., 2019) and astrocytes, neurons and tanycytes establish special

| DISCUSSION
In the present study, we used genetic approaches to express the fluorescent tdTomato protein in tanycytes in order to systematically examine their morphology. In contrast to vimentin immunostaining, this approach allowed us to reveal the presence of peculiar protrusions along tanycytes processes, contacting diverse neural cells throughout the hypothalamic parenchyma.
Like in other mammals, these protrusions are less abundant than in amphibians or reptiles: tanycyte processes are quite straight and their protrusions are mainly located at the proximal (i.e., spines) and the distal portions (i.e., swellings, boutons). Interestingly, differences exist between mice and rats (Bleier, 1971;Joy & Sathyanesan, 1981;Millhouse, 1971), although these two species are closely related to one another. In particular, spines in the neck region are more numerous in the rat and present up to the median eminence, whereas they are restricted to the DMH and VMH in mice.
The present study also shows that, besides interspecies differences, tanycyte morphology differs within the same animal, revealing an inadequacy in the current tanycyte classification (i.e., β1, β2, α1, and α2). Along the ventrodorsal axis, at least four morphologically distinct tanycyte populations would line the lateral wall of the third ventricle, facing respectively in the vmARH, the dmARH, the VMH and the DMH. Indeed, tanycytes lining the DMH and the VMH-currently called α1-tanycytes-present differences regarding the morphology of their cell body as well as their protrusions (i.e., boutons and swellings).
Moreover, morphological differences between tanycytes facing the dmARH and vmARH are striking, especially regarding their protrusions (i.e., spines) and their endfeet (i.e., endfeet within the parenchyma or at the pial surface). However, a clear delimitation between these different subgroups remains uneasy to make: indeed, some protrusions, such as spines present at the neck portion along the VMH and DMH, progressively disappear in the ARH, describing a ventrodorsal "gradient" rather than a feature of clear-cut tanycyte subgroups. Additionally, our study describes morphological differences from tanycyte to tanycyte along the anteroposterior axis, especially concerning their diameter and the direction of their processes. Therefore, as discussed previously (Langlet, 2019), improving tanycyte classification taking into account both the anteroposterior and ventrodorsal axis is crucial and constitutes our next challenge to further understand tanycyte biology.
In contrast to Golgi's method that stains many different neural cell types at random (Bleier, 1971;Card & Rafols, 1978;Fasolo & Franzoni, 1974;Joy & Sathyanesan, 1981;Millhouse, 1971), our approach inducing tdTomato expression in the ependymal layer allows us to visualize with certainty the interactions between tanycytes and other neural cells present in the hypothalamic parenchyma. In this study, we first show that, while tanycytes contact each other mainly through their cell bodies, spine-to-spine and endfeet-to-endfeet contacts were also detected between distinct tanycytes at the proximal and distal region of their processes, respectively. The formation of these close contacts confirm the importance of tanycyte-to-tanycyte communications, presumably for the synchronization of their functions. These communications could occur through gap junction protein Connexin-43 given that its selective depletion in astrocytes and tanycytes disrupts tanycyte-coupled network (Recabal et al., 2018).
Secondly, this study settles that tanycytes are in contact with multiple other neural cells. Their main heterotypic partners confirmed in our F I G U R E 1 2 Fine structure of tanycyte endfeet. (a) Inverse contrast scanning electron microscopy micrograph showing the ultrastructure of the distal process and endfoot of three α-tanycytes (tan, red) contacting a blood vessel (Vs) in the compact DMH in Zone 3 in adult male mice. Distal processes are filled with microtubules (mt) in their center, and display dilated segments containing elongated mitochondria (m). Around the blood vessel, tanycytes share the surface with glial processes (GP, violet) -probably belonging to astrocytes or tanycytes-and pericytes (Peri). Tanycytes also insulate neurons (Neu) from blood vessel. (b,c) Area of contact with the blood-brain barrier vessel and the pericyte. At the endfoot, mitochondria (m), ribosomes (r) and cisternae of the smooth endoplasmic reticulum and rough endoplasmic reticulum (ER) are present, but no microtubules. Double-membrane vesicles (dmv) surrounding a synapse (syn), single-membrane vesicles (smv), as well as dense-core vesicles (dcv) are observed. Around the blood vessel (Vs), tanycytes are in contact with the basal lamina (bm), and establish interactions through junctional complexes (arrowheads) with pericytes and other glial processes. Tanycytes are also observed in contact with neuronal cell body close to primary cilia (ci). TJ, tight junctions. study are blood vessels. This tanycyte/blood vessel relationship was largely described in the literature (Rodríguez et al., 2019): linking the blood and the cerebrospinal fluid, tanycytes are therefore described as gateways to the brain (Langlet, 2014). Here, we also observe that many tanycyte endfeet share the space around the blood vessels with pericytes and astrocytes: multiple way exchanges between these different cell types, in particular for the formation and maintenance of the hypothalamic blood-brain barrier, remain to be explored. Contacts with other glial cells such as oligodendrocytes are also present, confirming previous reports (Recabal et al., 2018). Additionally, we describe here associations with different neuronal populations, suggesting that tanycytes integrate into multiple neuronal networks, resulting in the regulation of diverse physiological functions. In particular, tanycytes contact NPY and POMC neuronal populations as well as KNDy neurons: such associations likely play a role in the regulation of energy balance and reproduction, respectively. So far, tanycyte/ neuron interactions were a matter of controversy in the literature.
Close anatomical contacts with NPY neurons were already reported (Coppola et al., 2007), but those with other neuronal populations were never described before. Besides neuronal cell bodies, our study also reveals tanycyte associations with GABAergic terminals and glutamatergic terminals. The first type of associations consists of synaptoid contacts on tanycyte processes as previously reported in rats for β and α tanycytes (Rodríguez et al., 2019;Rodríguez et al., 2005). The functional significance of these synaptoid contacts remains to be elucidated, but suggests that tanycyte/neuron communications are likely bi-directional. By this way, neurons could control tanycyte functions, as it was previously hypothesized for the regulation of volume transmission (Alpár, Benevento, Romanov, Hökfelt, & Harkany, 2019).
Here, we additionally provide the description of tanycyte protrusions surrounding synapses suggesting a role for tanycytes in the modulation of synaptic transmission. This hypothesis is strengthened by the presence of GLT1 glutamate transporter in tanycyte boutons necessary to control extracellular glutamate in hypothalamic nuclei, and to ensure brain homeostasis and synaptic transmission. Further experiments are consequently needed to analyze tanycyte proximity with synapses as well as their plasticity, and to evaluate if tanycytes, mimicking astrocyte function, could play a role in functional tripartite synapses. Finally, we also show that tanycyte endfeet encapsulate synapses, suggesting a role for tanycytes in synapse elimination. While phagocytosis is primarily attributed to microglia, it also was described in other glial cells such as astrocytes or oligodendrocytes in order to maintain homeostasis in the brain (Lee & Chung, 2019). By this way, tanycytes could contribute to synaptic plasticity.
Although we focused our study on the ARH, these different tanycyte partners were also observed in the VMH and DMH, suggesting that tanycytes are involved in different neural network regulating diverse physiological functions. Moreover, tanycytes-like cells being present in other brain areas (Langlet, Mullier, et al., 2013), similar contacts could also be observed in these regions: some differences and/or specificity between brain areas cannot be nevertheless excluded. Finally, it is worth to note that these different partners could impact tanycyte morphology per se. Indeed, cells sense the presence of interaction partners and, specifically respond by changing the expression of many target genes resulting in a specific cell morphology adapted to their functions and communications. Therefore, the existence of different neuronal populations with different anteroposterior and ventrodorsal distributions within the same nucleus may influence tanycyte morphology and functions on both axes. Based on that, we can consequently predict that a higher number of different tanycyte subgroups exist.
The numerous protrusions present along tanycyte processes and contacting neurons and/or blood vessels strongly suggest a functional

| CONCLUSION
Based on this neuroanatomical study, we propose that tanycytes serve as a communication system between the cerebrospinal fluid, brain capillaries and neural cells within the hypothalamus. The different protrusions found along tanycyte processes would allow them to integrate information coming from the brain through the CSF and the periphery through blood vessel, and to redistribute it to neural cells throughout the hypothalamus.