Characterisation of endogenous players in fibroblast growth factor‐regulated functions of hypothalamic tanycytes and energy‐balance nuclei

Abstract The mammalian hypothalamus regulates key homeostatic and neuroendocrine functions ranging from circadian rhythm and energy balance to growth and reproductive cycles via the hypothalamic‐pituitary and hypothalamic‐thyroid axes. In addition to its neurones, tanycytes are taking centre stage in the short‐ and long‐term augmentation and integration of diverse hypothalamic functions, although the genetic regulators and mediators of their involvement are poorly understood. Exogenous interventions have implicated fibroblast growth factor (FGF) signalling, although the focal point of the action of FGF and any role for putative endogenous players also remains elusive. We carried out a comprehensive high‐resolution screen of FGF signalling pathway mediators and modifiers using a combination of in situ hybridisation, immunolabelling and transgenic reporter mice, aiming to map their spatial distribution in the adult hypothalamus. Our findings suggest that β‐tanycytes are the likely focal point of exogenous and endogenous action of FGF in the third ventricular wall, utilising FGF receptor (FGFR)1 and FGFR2 IIIc isoforms, but not FGFR3. Key IIIc‐activating endogenous ligands include FGF1, 2, 9 and 18, which are expressed by a subset of ependymal and parenchymal cells. In the parenchymal compartment, FGFR1‐3 show divergent patterns, with FGFR1 being predominant in neuronal nuclei and expression of FGFR3 being associated with glial cell function. Intracrine FGFs are also present, suggestive of multiple modes of FGF function. Our findings provide a testable framework for understanding the complex role of FGFs with respect to regulating the metabolic endocrine and neurogenic functions of hypothalamus in vivo.


| INTRODUC TI ON
Growing evidence shows that, in addition to hypothalamic neurones, tanycytes act as sensors, mediators and effectors for critical processes that underpin the homeostatic and neuroendocrine functions of the hypothalamus; for example, sensing of metabolites and micronutrients and trafficking of peripheral and central signals such as leptin and gonadotrophin-releasing hormone. 1,2 Moreover, a subset of tanycytes acts as bona fide neural stem/progenitor cells in the postnatal and adult hypothalamus, capable of generating new energy balance-regulating neurones and glia. [3][4][5] Tanycytes are residual radial glial-like cells that occupy the floor and ventrolateral walls of the third ventricle (3V). They have been subdivided into four main subtypes: β2, β1 and α2, α1, according to a combination of projection trajectory, morphology, barrier properties, cilia arrangement and positioning within or outside marker domains within the 3V wall. 1 Commonly, their apical surface is exposed to the cerebrospinal fluid, whereas their basal processes either contact the portal capillaries of the central (β2-tanycytes) and lateral (β1-tanycytes) median eminence, or terminate within the arcuate or ventromedial and dorsomedial nuclei C γ. 6,7 The remaining FGFs (FGF11-14) function intracellularly in a receptor-independent manner. 8 FGFR signalling can modulate diverse aspects of cell behaviour in a cell-type specific manner. This is achieved partly via differential levels of signalling, established by intracellular negative feedback loop regulators such as Sproutys (Spry) and Map-kinase phosphatases (Mkp). 9 Tissue specificity of the action of FGF is governed by target cell expression of the socalled IIIb or IIIc alternatively-spliced receptor isoforms that engage a mutually-exclusive set of FGFs. 10 Moreover, co-factors such as sulphated proteoglycans and β-Klotho molecules not only facilitate focal FGF/FGFR signalling, but also determine the range of extracellular FGF diffusion. For example, peripherally generated FGF19 and FGF21 evade entrapment by heparan sulphate proteoglycans to act as circulating hormones in the metabolic control of energy homeostasis, with potential clinical applications in the management of type 2 diabetes. 11,12 Exogenous FGF2 is a potent mitogen for hypothalamic ependymal and neural cells in vitro and in vivo. 5,13 Levels of FGF1 show a dramatic postprandial rise in the cerebrospinal fluid, 14,15 whereas experimental elevation or attenuation of canonical FGF signalling, via peripheral or ic.v. application of FGF ligands, soluble FGFR fragments or neutralising antibodies, can induce hypo-and hyperphagia, respectively. [16][17][18][19] Exogenous FGF1 also accelerates glucose clearance and induces sustained remission of diabetes symptoms in a diabetic mouse model. 20 A full understanding of how exogenous FGFs function and/or whether endogenous FGFs have similar or unique roles requires the identification and spatial distribution of the key players: ligands, receptors, receptor co-factors and signalling modifiers. To address this need, we carried out a detailed survey of FGF signalling pathway components in the mediobasal hypothalamus, using reverse transcriptasepolymerase chain reaction (RT-PCR) screens, in conjunction with in situ hybridisation (ISH), immunolabelling and analysis of transgenic reporter mice. Our investigations reveal intricate and restricted domains of FGF and FGFR receptor expression amongst tanycytes, in addition to both distinct and overlapping patterns of receptor and ligand expression within the neighbouring energy balance regulating nuclei.

| Animals
Both male and female mice ranging in age from 30 to 80 days of age were used. All mice were bred and maintained on a mixed C57BL6/129Ola background, under a 12:12 hour light/dark cycle, in accordance with local and national regulations and licenses governing experimental work with animals. Tissues from Fgf9-lacZ; Fgf18-CreER and Etv4-GFP transgenic reporter mice were kindly provided by Professors David Ornitz and Saverio Bellusci, as reported previously, 21,22 or recently generated strains. To detect Fgf18-expressing cells in the brain, mice carrying Fgf18-CreERT2::Rosa-tdTomato-dsred alleles were injected i.p. with tamoxifen (300 mg kg -1 body weight) at postnatal day (P)50, P51 and P52, before tissue harvest at P53.

| Tissue isolation and preparation
Mice were killed by CO 2 asphyxiation and brains were then dissected out and fixed for 4-16 hours overnight in 4% paraformaldehyde (PFA) (pH 7.0) at 4°C. For fresh tissue and mRNA isolation, mice were killed by cervical dislocation and microdissected hypothalami were flash frozen in cryotubes on dry ice and stored at −80°C. To generate cryostat sections, the brains were washed with diethyl pyrocarbonate (DEPC)-treated PBS (DEPC-PBS) and cryo-protected in 30% w/v sucrose/DEPC-PBS solution for 48 hours at 4°C. Thereafter, they were placed in cryomoulds filled with optimal cutting temperature (OCT) compound, allowed to set on dry ice, and stored at −80°C.
Just before use, brains were rehydrated stepwise back to PBS and embedded in 3% w/v agar overnight. Coronal sections, 60 μm thick, containing median eminence (bregma −1.46 to −2.18) were then generated using a vibrating microtome (Leica Microsystems, Wetzlar, Germany) and stored in PBS until use.

| RNA isolation
Microdissected hypothalami were homogenised by trituration in 1 mL of Trizol (Invitrogen, Carlsbad, CA, USA) before addition of 200 μL of chloroform. Following centrifugation (15 minutes at 12 000 g), RNA was precipitated from the aqueous layer by addition of 500 μL of isopropanol. Samples were briefly vortexed, allowed to stand for 10 minutes at room temperature and then centrifuged (15 minutes at 12 000 g). The resulting supernatant was carefully removed before adding 1 mL of 75% ethanol diluted in molecular grade water (Sigma, St Louis, MO, USA) and further centrifugation (12 minutes at 5200 g). The RNA pellet was then briefly air dried, re-suspended in molecular grade water and incubated at 55°C for 10 minutes. RNA concentration and quality was determined using a spectrophotometer (Nanodrop ND-1000; Thermo Fisher Scientific) and samples were stored at −80°C until use.

| Primer design and RT-PCR
Gene specific primer pairs (see Supporting information, Table S1) for use in RT-PCR reactions were designed using NCBI primer-BLAST ( https ://www.ncbi.nlm.nih.gov/tools/ primer-blast ), aiming for regions of least sequence conservation between related family members, 23 and amplicons that are not only small, but are also comparable in size across the related family members (see Supporting information, Table S1). To rapidly generate gene-specific templates for in vitro transcription of antisense cRNA probes (see below), some reverse primers were additionally tagged at their 5′ end with a T7 polymerase-encoding promoter sequence (TAATACGACTCACTATAGGG).

| Riboprobe synthesis and In situ hybridisation (ISH)
Digoxigenin-labelled antisense RNA probes were generated using T7 RNA polymerase enzyme (Thermo Fisher Scientific) and either T7-tagged RT-PCR amplicons (above) or linearised plasmids encoding gene-specific cDNAs (see Supporting information, Table S2), as templates. Riboprobes were purified using Chroma Spin columns (Clontech, Mountain View, CA, USA) and stored at −20°C until use.
ISH reactions were performed after optimisation of previously reported protocols, 24 as detailed below. Briefly, cryosections were allowed to equilibrate at room temperature for 1 hour before a 5 minutes fixation in 4% PFA, followed by three 5 minutes washes Hybridisation was carried out at between 68 and 72°C in a hybridisation oven for 16-18 hours, using 500 ng mL -1 riboprobes diluted in hybridisation buffer. Post hybridisation washes were: 5 minutes with 5 × SSC at room temperature, followed by three 30 minutes washes with 0.2 × SSC at hybridisation temperature and a wash with 0.2 × SSC at room temperature. Sections were then washed with Tris-HCl/saline buffer (100 mmol L -1 Tris-HCl, pH 7.5, 150 mmol L -1 NaCl) at room temperature and blocked in Tris-HCl/saline buffer with 10% heat-inactivated normal goat serum (hiNGS) before an overnight incubation at 4°C with alkaline phosphatase-conjugated anti-digoxigenin conjugated antibodies (see Supporting information, Table S2) diluted in Tris-HCl/saline buffer, containing 3% hiNGS.
Negative controls included the omission of antisense and/or use of sense cRNA probes (Fgf9, Fgf20, Klb, Sef, Spry1, Spry2). Positive controls included probes for Neuropeptide Y (Npy), an anorexigenic neurotransmitter expressed by some arcuate neurones, which invariably yielded a very strong and specific signal.

| Immunohistochemistry, immmunofluorescence labelling and X-gal staining
Non-specific binding sites were blocked by a 2-hour incubation in a solution of 20% hiNGS/1% Triton X-100 (TX)/PBS, before application of primary antibodies (see Supporting information, Table S2) overnight at 4°C in 0.2% hiNGS/0.1% TX/PBS. Excess primary antibodies were removed by five 1-hour washes at room temperature with a 0.2% hiNGS/0.1% TX solution and this was followed by overnight incubation with the relevant secondary fluorophore conjugated antibodies (see Supporting information, Table S2) diluted in 0.2% hiNGS/0.5% NP-40/in PBS at 4°C. The next day, sections were washed six times (30 minutes per wash) before counterstaining with nuclear DNA marker, Hoechst, and subsequent mounting in Vectashield (Vector Laboratories, Burlingame, CA, USA).
For combined ISH and immunohistochemistry, after completion of the ISH protocol, sections were incubated for 1 hour in blocking solution (10% hiNGS/0.3% TX/PBS) followed by overnight incubation at 4°C with mouse anti-glial fibrillary acidic protein (GFAP), rabbit anti-GFAP or mouse anti-S100β antibodies (see Supporting information, Table S2) diluted in 0.2% hiNGS/0.1% TX/PBS. After three 5-minute PBS washes, the 3,3'-diaminobenzidine (DAB) staining reaction was performed using a DAB peroxidase substrate kit (Vector Laboratories) in accordance with the manufacturer's instructions. After five 1-minute washes with ddH 2 O, sections were preserved by coverslipping with 50% glycerol/PBS solution.
To detect the product of the lacZ reporter gene, β-galactosidase, brains of Fgf10-lacZ reporter mice were preserved in Mirsky's fixative (National Diagnostics, Atlanta, GA, USA) overnight at 4ºC, washed three times with PBT (PBS with 0.1% Tween-20) before being incubated with the X-gal substrate solution (2 mmol L -1 MgCl 2 ,

| Image capture and analysis
All images were captured using a Axioplan 2 microscope and axiovision, version 4.8 (Carl Zeiss, Oberkochen, Germany). High magnification of phophorylated extracellular signal-regulated kinase (pERK) immunostaining was imaged and analysed by 3D reconstruction of 1μm optical sections (z stacks), captured using an Apotome attachment.

| Global analysis: detection of distinct FGFs and signalling modulators and predominance of FGFR IIIc isoforms
To determine which components of the FGF signalling system need to be mapped spatially in the hypothalamus, we first undertook a gross RT-PCR screen of mRNA expression derived from 30-45day-old (P30-P45) animals. Industrially pre-aliquoted substrates (illustra Ready-To-Go beads; GE Healthcare Life Sciences) were used to amplify comparable amplicon sizes. We opted for an RNA-based and gene-reporter analysis (here and below) because antibodies to FGFs often detect multiple related family members and, in the case of FGFRs, they fail to discriminate between isoforms in immunolabelling studies. Animals older than P30 were used because the full spectrum of tanycytes subtypes, as well as the connectivity of hypothalamic neurones, is fully established by this age. 25 FGFR1-3 exist as alternatively-spliced isoforms termed "IIIb"  Table S1) detected products of the expected size for Fgf1, 2,5,7,8,9,10,11,12,13,14,16,17,18,20  Underlined numbers and letters denote a lack of detection of the relevant genes. Faint bands for Fgfr2 IIIb isoform and Fgf3 are nonspecific products, whereas all other RT-PCR products match the expected amplicon size, as predicted by use of primers listed in the Supporting information (Table S1). Multiple Fgf5 and Fgf8 products possibly represent isoforms of these genes Fgfr- KIβ Spry4   1IIIb  2IIIb  3IIIb  1IIIc  2IIIc  4  3IIIc closely-related isoforms. 26 Co-receptor β-Klotho, which is required for the action of FGF19 and FGF21, as well as FGFR signalling modulators Spry1, Spry2, Spry3 and Mkp3 but not Sef or Spry4, were also detectable, with Spry2 and Spry3 showing stronger expression ( Figure 1C). Pea3 (also known as Etv4), a transcriptional target of FGF signalling, was also present.
This broad screen reveals that various components of the FGF signalling pathway are in place to mediate exogenous and to potentially augment endogenous FGF signalling in the hypothalamus.

| Predominance of FGFR1 and FGFR2, in the βtanycyte domain
The gross RT-PCR screen (above) was performed using bulk tissue that contains multiple cell compartments: the median eminence  Figure S1), the ISH analysis was combined with mouse anti-GFAP immunolabelling to investigate the selective expression of candidate genes within α-vs β-tanycyte subtypes.
We found that Fgfr1 and Fgfr2 are expressed predominantly in the GFAP-ve β-tanycyte domain (Figure 2A,A′,B). Interestingly, the strongest Fgfr1 signal was observed in the floor of the 3V, which harbours the β2 subtype ( Figure 2A′), whereas Fgfr2 was equally strong in both β2-and β1-tanycyte domains, as well as the transition zone between β1-and α-tanycytes ( Figure 2B). We combined Fgfr3 in situ with immunolabelling for glial/progenitor cell marker, S100β, only to find prevalent co-expression of these markers, confirming the glial identity of most Fgfr3-expressing cells

| Expression of FGF signalling modulators, mediators and targets overlaps with FGFRs
A key feature of FGF signalling is the transcriptional activation of its own downstream negative regulators to fine tune intracellular levels of signalling. 9 Exploiting this phenomenon, we investigated the domains of Spry family and Mkp3 expression to determine exactly where endogenous FGFR signalling may be occurring in the mediobasal hypothalamus. ISH with Spry1 riboprobes (Figure 3A,A′) yielded a strong signal in β-tanycytes and cells of the ARC, although expression in VMN, DMN and LHA was also evident. Co-labelling with GFAP antibodies revealed a weak expression in the α-tanycyte domain ( Figure 3A′). By contrast, use of Spry2 riboprobes ( Figure 3B) showed a more restricted pattern, composed of a weak signal in β-tanycytes, and scattered expression in the ARC, LHA and subependymal ME.
Spry3 ISH reactions produced a weak background signal. Mkp3 signal was restricted to scant cells in the LHA/tuberal nucleus ( Figure 3C).

| FGF ligands are expressed in tanycytes in two distinct patterns
The presence of FGFRs (IIIc isoforms) and FGF signalling modulators in the hypothalamus is suggestive of a necessity for cognate ligand(s). To delineate the cohort of putative ligands involved, we performed ISH using Fgf-specific cRNA probes and used transgenic Interestingly, ventricular expression of Fgf9, as judged by X-gal staining of Fgf9-lacZ brain, was also confined to rostral sections and absent from caudal bregma co-ordinate −1.82 onwards. Parenchymal staining for these FGFs (Figure 4A,B; see also Supporting information, Figure S2

| Restriction of intracellular FGFs to the parenchymal cell compartment
ISH with cRNA probes for intracellular FGFs (11 to 14) revealed a predominant expression in the hypothalamic parenchyma and subependymal cells in the ME but a notable absence from tanycytes themselves.

| D ISCUSS I ON
Extrinsic modulation of FGF signalling in rodents can elicit dramatic changes in metabolism, body weight and plasma glucose clearance, TA B L E 1 Distribution of key FGF signalling players within different cellular compartments of the mediobasal hypothalamus Note: Tapered colouring denotes a gradient of intensity of staining (darker ends, higher levels). A single star (*), denotes scattered cells expressing the gene and double stars (**) denote higher levels of expression or tighter clusters of gene-expressing cells. with the hypothalamus being envisaged as an anchor for these effects.
The therapeutic use of FGFs requires a better understanding of the foci and mechanisms by which FGFs operate, and whether endogenous hypothalamic FGF signalling normally has a role to play in these processes.
On face value, our results, summarised in Table 1 and Figure 5 Figure S3) in a similar pattern to that reported previously. 37 The presence of FGF pathway genes that we have characterised at celllular resolution is supported by a recent drop-seq gene profiling of cells in the median eminence and the ARC. 34 However, it is possible that, with the use of more sensitive methods, such as isotopic ISH or quantitative PCR on cell sorted subpopulations, new and additional members may be identified.

| Potential modes and significance of FGF/FGFR function in tanycytes and parenchymal nuclei
Because canonical FGF signalling requires the activation of FGFRs, specifically their IIIc isoforms in the hypothalamus ( Figure 1A), FGFRs become the focus of debate when considering FGF function. Activation of FGFRs can yield a multitude of effects, typically involving cell proliferation, differentiation, migration or cytoskeletal remodelling. 7 We showed that FGFR1 and FGFR2 are restricted to β-tanycytes, where they could regulate cell proliferation and differentation of these newly-identified stem/progenitor cells. 38 Indeed, some studies have envisaged changes in postnatal hypothalamic neurogliogenesis as a contributory mechanism to body weight change and insulin-independent glucose lowering effects of exogenous FGFs. 11,39 Equally, these effects may involve short-or long-term alterations in barrier properties, nutrient sensing, and cargoing or trafficking of metabolic signals and hormones by β-tanycytes. 2,38,40 These cells are certainly enriched in the endocytotic pathway molecules, such as caveolins, 41 which have been shown to control FGF2/ FGFR1 signalling in other cell types and settings. 42 Within the 3V epithelium, β-tanycytes also uniquely possess primary cilia, 33 as well as receptors for VEGFR3, 43 and other key regulators of energy homeostasis such as ciliary neurotrophic factor 44 and leptin. 38 This suggests that β-tanycytes may also be responsive to Hedgehog signalling and/ or act as a hub of cell signalling, or that their biology relies on synergy between multiple signalling pathways. Therefore, fine dissection of other signalling pathways in tanycytes is warranted.
β2-tanycytes are radial glial-like cells whose processes span the width of the median eminence. Therefore, it is possible that FGFR1 and FGFR2 operate not only on their apical surfaces, exposed to ventricular-derived FGFs, but also on their basal surfaces, in close proximity to the capillary plexus of the outer ME, where capillary fenestrations would readily allow exposure to circulating FGFs.
Hints for the heterogeneity and paracine response of β-tanycytes to endogenous FGFs comes from caudal expression of pERK in β1-tanycytes, complemented by rostral expression of Fgf9 in α-tanycytes ( Figure 4B). The rostrocaudal complimentarity of ligands and receptor is intriguing but highly reminiscent of developmental settings, where mesenchymal and epithelial cells express mutually exclusive cohorts of FGF ligands and receptors to ensure paracrine and unidirectional FGF signalling. 9 In the hypothalamus, FGFs that are secreted into the 3V space may well be distributed by beating cilia of 3V ependymal cells 45 to activate distant receptors. This would be akin to critical asymmetrical distribution of FGF8 and Nodal by beating cilia during establishment of left-right asymmetry in early vertebrate embryos. 46 Similarly, strong expression of pERK in the ARC ( Figure 3E) and Etv4-GFP and Mkp3 in the LHA is, complemented by strong expression of Fgf9 in the neighbouring VMN ( Figure 4B).
ERK activation could also reflect receptor activation by FGF19 and

| Lessons from human mutations and transgenic mouse models of the FGF signalling pathway
Is the prediction of FGFR1-3 IIIc and FGF1, 2, 9 and 18 as key endogenous canonical FGF signalling partners in mouse hypothalamus, borne out by naturally-occuring human mutations or experimental transgenic alleles? With minor exceptions (e.g. murine Fgf15 vs human Fgf19), both the human and mouse FGF signalling apparatus is highly conserved in expression domain and function. Moreover, tanycytes have also been identified in the human hypothalamus 54 and human and rodents broadly share common mechanisms for the regulation of energy balance. A spectrum of dominant acting mutations in human FGFR1 and FGFR2 causes rare congenital skeletal syndromes, such as Apert and Pfeiffer. 6 However, no association between these syndromes and predisposition or resistance to diabetes and obesity has been reported. By contrast, a significant number of achondroplasia patients that carry activating FGFR3 mutations also manifest atypical obesity. 55 Difficulties that prevent drawing firm conclusions from human patients include: the germline nature of these mutations, which would additionally impact peripheral organs or indeed multiple brain regions; their rareness, which may preclude statistically valid associations with metabolic/neuroendocrine defects or diet; and their varied molecular mode of function, with some mutant receptors operating in a FGF ligand-independent manner.
With respect to engineered mouse models, germline deletion of FGFR1 and FGFR2, or FGF9 or FGF18 56 causes prenatal or early postnatal lethality, whereas the loss of FGFR3 induces abnormal bone growth. FGF1-and FGF2-deficient mice show no overt abnormalities, even as compound mutants. After challenge with a high-fat diet, however, these mice show hyperglycaemia and insulin resistance, likely as a result of the critical regulation of adipose tissue by FGF1. 57 Mice carrying a gain-of-Fgfr1 function mutation appear normal, 58 although those with a gain-of Fgfr2 mutation are hypoglycaemic and show growth retardation and early death. 59 Fortunately, a plethora of conditional loss and gain of function alleles for these and other FGFs and FGFRs have been developed to circumvent embryonic/neonatal lethality. Their use in cell-type specific and stage-dependent gene targeting, as exemplified by recent work 28,60 will prove valuable for unravelling the exact role(s) of endogenous FGF signalling, alone or in combination with other signalling pathways, in tanycyte biology and the regulation of hypothalamic neuronal functions.

ACK N OWLED G EM ENTS
We are grateful to Stuart Nayar for insightful comments and critique.
BK was supported by a BBSRC Doctoral Training Programme (DTP) Studentship allocated to MKH. This work was supported in part by a BBSRC research grant (BB/L003406/1) to MKH. SB was supported by EXC2026 (project id 390649896). The authors apologise to colleagues whose work could not be cited because of space limitations.

DATA AVA I L A B I L I T Y
The data that support the findings of the present study are available from the corresponding author upon reasonable request.