Novel hypophysiotropic AgRP2 neurons and pineal cells revealed by BAC transgenesis in zebrafish

The neuropeptide agouti-related protein (AgRP) is expressed in the arcuate nucleus of the mammalian hypothalamus and plays a key role in regulating food consumption and energy homeostasis. Fish express two agrp genes in the brain: agrp1, considered functionally homologous with the mammalian AgRP, and agrp2. The role of agrp2 and its relationship to agrp1 are not fully understood. Utilizing BAC transgenesis, we generated transgenic zebrafish in which agrp1- and agrp2-expressing cells can be visualized and manipulated. By characterizing these transgenic lines, we showed that agrp1-expressing neurons are located in the ventral periventricular hypothalamus (the equivalent of the mammalian arcuate nucleus), projecting throughout the hypothalamus and towards the preoptic area. The agrp2 gene was expressed in the pineal gland in a previously uncharacterized subgroup of cells. Additionally, agrp2 was expressed in a small group of neurons in the preoptic area that project directly towards the pituitary and form an interface with the pituitary vasculature, suggesting that preoptic AgRP2 neurons are hypophysiotropic. We showed that direct synaptic connection can exist between AgRP1 and AgRP2 neurons in the hypothalamus, suggesting communication and coordination between AgRP1 and AgRP2 neurons and, therefore, probably also between the processes they regulate.

In the zebrafish, agrp1 is expressed exclusively in the hypothalamus, and, like mammalian Agrp, was shown to be involved in feeding behavior 12,27 . Zebrafish agrp2 mRNA is expressed in the pineal gland [28][29][30] and was suggested, on the basis of morpholino-mediated knock-down experiments, to regulate background pigment adaptation for camouflage 30 . Pineal gland expression of agrp2 was reported in sea bass (Dicentrarchus labrax) 20 , and we have also detected its expression in the pineal gland of Nile tilapia (Oreochromis niloticus) and rainbow trout (Oncorhynchus mykiss; unpublished data). This suggests a conserved pineal gland expression of agrp2 among teleosts. However, the targets and role of AgRP2 and its relationship to AgRP1 are not fully understood.
To facilitate elucidation of the functions of AgRP1 and AgRP2 in zebrafish, we generated transgenic fish in which agrp1-and agrp2-expressing cells are fluorescently labelled. Here we describe the anatomical organization of these cells and provide tools for their manipulation. Focusing mainly on AgRP2, we report previously uncharacterized agrp2-expressing pineal cells, novel hypophysiotropic AgRP2 neurons, and a novel AgRP1-AgRP2 neuronal interaction. Together, these observations and use of the novel transgenic fish lines may eventually lead to identification of hitherto undiscovered neuroendocrine functions for AgRP neuropeptides.

Generation of transgenic zebrafish lines for visualization of AgRP1 and AgRP2 neurons.
To visualize AgRP1 and AgRP2 neurons we employed the bacterial artificial chromosome (BAC) transgenesis approach to express fluorescent markers within these neurons. To increase the utility of the generated transgenic lines, we employed the Gal4-VP16 transactivation system, generating BACs in which Gal4-VP16 is under the control of agrp1 and agrp2 regulatory regions. BAC clones containing AgRP1-or AgRP2-coding sequences, as well as 45− 60 kb downstream and upstream flanking regions, were modified by recombineering according to an established protocol 31 and used for transgenesis ( Supplementary Fig. S1). These BACs are likely to contain all of the cis-regulatory elements needed to replicate the endogenous spatial and temporal expression of agrp1 and agrp2. The generated transgenic lines were registered in the Zebrafish Model Organism Database (ZFIN) as TgBAC(agrp:Gal4-VP16) tlv04 and TgBAC(agrp2:Gal4-VP16) tlv05 . To fluorescently label the Gal4-VP16-expressing cells, we crossed these lines with Tg(UAS:nfsB-mCherry) c264 fish 32 . For simplicity, the resulting double-transgenic fish, TgBAC(agrp:Gal4-VP16) tlv04 ;Tg(UAS:nfsB-mCherry) c264 and TgBAC(agrp2:Gal4-VP16) tlv05 ;Tg(UAS:nfsB-mCherry) c264 , will be referred to as agrp1:mCherry and agrp2:mCherry, respectively. In addition, we generated a BAC transgenic line expressing enhanced green fluorescent protein (EGFP) under agrp2 regulatory regions, registered in ZFIN as TgBAC(agrp2:EGFP) tlv06 . For simplicity, these transgenic fish will be referred to as agrp2:EGFP.
Whole mount in-situ hybridization (ISH) of agrp1 and agrp2 mRNAs in non-transgenic wild-type larvae was used to determine whether the transgene expression patterns accurately report the endogenous expression of agrp1 and agrp2. Endogenous agrp1 mRNA expression was found to be restricted to the ventral periventricular hypothalamus (Fig. 1a,b), as previously described 12 . This expression pattern was similar to the distribution of fluorescently labelled cells in agrp1:mCherry larvae (Fig. 1c). Whole-mount ISH showed that agrp2 mRNA is strongly expressed in the pineal gland, as previously described [28][29][30] . In addition, a domain with weaker agrp2 expression was revealed in the preoptic area (Fig. 1d,e). The spatial expression of agrp2 mRNA was matched by expression of the transgene in agrp2:mCherry and agrp2:EGFP larvae (Fig. 1f). Both lines showed strong fluorescence in the pineal gland, with a weaker fluorescent signal in the preoptic area.
The expression pattern of fluorescent markers in those transgenic lines also correlated with the temporal expression of the endogenous genes. agrp1 mRNA was detected as early as 2 days post-fertilization (dpf) as previously described 12 , and mCherry expression in the agrp1:mCherry line was detected by 3 dpf. agrp2 mRNA was detected in the pineal at 30 hours post-fertilization and in the preoptic area at 2− 3 dpf. mCherry expression in the transgenic line was detected at the same developmental stages, while EGFP expression in the agrp2:EGFP line was weaker and appeared later (data not shown).
Together, the above findings indicated that the generated transgenic lines reliably report the endogenous expression of agrp1 and agrp2, and can therefore be used to examine AgRP1 and AgRP2 neuronal systems. For the remainder of this study we focused mainly on the two populations of AgRP2 cells in the pineal and preoptic areas of the forebrain, and also on potential connectivity between AgRP1 and AgRP2 neurons. agrp2 is expressed in a subset of uncharacterized pineal cells. The pineal gland of teleost fish contains at least three main types of cells: photoreceptor cells, projection neurons, and interstitial cells [33][34][35][36] . We used transgenic and immunohistochemical approaches to determine the identity of the AgRP2 cells. Pineal photoreceptor cells express opsins and melatonin-synthesizing enzymes such as arylalkylamine N-acetyltransferase2 (AANAT2), and are fluorescently labelled in Tg(aanat2:EGFP) y9 fish 37 . Foxd3 is an early marker of pre-migratory neural crest cells and of certain non-crest-derived cell types including pineal projecting neurons; the latter are fluorescently labelled in Tg(foxd3:EGFP) zf104 fish [38][39][40] . We crossed agrp2:mCherry with either Tg(aanat2:EGFP) y9 or Tg(foxd3:EGFP) zf104 fish. Examination of the resulting double-labelled progeny revealed that the expression of mCherry in AgRP2 cells does not co-localize with EGFP-expressing cells in either line, indicating that they are not likely to be pineal photoreceptors or foxd3-expressing pineal neurons (Fig. 2a,b, Supplementary Movie S1). To further examine whether AgRP2 cells are a sub-population of pineal neurons lacking foxd3 expression, we performed double immunohistochemistry for mCherry and the neuronal marker HuC in agrp2:mCherry larvae. This revealed AgRP2 cells to be HuC negative (Fig. 2c, Supplementary Movie S1). These experiments strongly suggested that AgRP2 cells are neither photoreceptors nor HuC-positive neurons. We then examined the possibility that pineal AgRP2 cells might be interstitial cells. Glial fibrillary acidic protein (GFAP) is a marker for pineal interstitial cells 33,41 . Double immunohistochemistry for GFAP and GFP in agrp2:EGFP larvae revealed different patterns of expression. This strongly suggested that AgRP2 cells do not express GFAP and are therefore not pineal interstitial cells (Fig. 2d, Supplementary Movie S1). Altogether, these results indicated that AgRP2 cells may represent a novel, uncharacterized pineal cell population. AgRP2 preoptic neurons project to the adenohypophysis. We next investigated the other main agrp2 expression domain, a small bilateral group of cells in the preoptic area (4− 5 cells in each hemisphere, Fig. 3a). Confocal imaging of agrp2:mCherry revealed that these cells project towards the pituitary (Fig. 3b, Supplementary Movie S2). Examination of transgene expression in a cross between agrp2:mCherry and Tg(oxt:EGFP) wz01 fish 42 showed that AgRP2-positive cells and oxytocin neurons are adjacent (Fig. 3a) and project towards the pituitary (Fig. 3d). The presence of axon-like processes and the sharing of axonal tracts (Fig. 3d) suggest that these cells are neurons. We therefore examined these AgRP2 axonal projections. A cross between agrp2:mCherry and Tg(pomc:EGFP) zf44 fish (which expresses EGFP in pituitary POMC cells only 43 ) revealed that AgRP2 axons terminate near the anterior group of pituitary POMC cells (Fig. 3c), suggesting that the preoptic AgRP2 neurons project towards the adenohypophysis. This anatomical location was confirmed by comparing the terminations of AgRP2 and oxytocin axons in the pituitary: a clear segregation was observed between these To verify that the AgRP2 adenohypophysial terminations are indeed axonal terminals, we crossed agrp2:m-Cherry fish with Tg(UAS:SYP-EGFP) biu5 fish. This UAS-Gal4 driven line expresses a synaptophysin-EGFP fusion protein that aggregates in pre-synaptic vesicles 44,45 . In the progeny of this cross, AgRP2 cell bodies and projections are labelled with mCherry, while pre-synaptic varicosities are labelled with EGFP. Examination of the larvae resulting from this cross revealed a field of EGFP-positive pre-synaptic boutons in the AgRP2 pituitary terminals (Fig. 3e). Neurovascular connections reminiscent of the tetrapod hypophysial portal system were recently described in the zebrafish adenohypophysis 46 . To examine the possibility of an AgRP2 neurovascular connection in the hypophysis, we crossed agrp2:mCherry with Tg(kdrl:EGFP) S843 fish (which expresses EGFP in vascular endothelial cells 47 ). Examination of transgene expression in this cross showed that AgRP2 terminals are juxtaposed with vessels that have been designated as the hypophysial artery 42 (Fig. 3f,g). AgRP2 projections towards the pituitary vasculature could be detected as early as 3 dpf; the complexity of the pituitary vasculature and the magnitude of AgRP2 innervation subsequently increased (described in Supplementary Fig. S2).
These results indicate that hypothalamic AgRP2 neurons are hypophysiotropic and probably terminate directly onto the hypophysial vasculature. Whether AgRP2 acts on adenohypophysial cells to induce hormonal secretion or is directly released into the circulation and has a neuroendocrine role in the periphery remains to be determined.
AgRP1 projections and AgRP1-AgRP2 interactions. The zebrafish AgRP1 neuronal system has been previously described in a detailed immunohistochemical study 27 . In the present study we closely examined the generated agrp1:mCherry fish to determine whether it accurately represents the previously described distribution. These investigations revealed that AgRP1 neurons are located in the ventral periventricular hypothalamus (10− 12 neurons in each hemisphere) and project axons towards the rostral, intermediate and dorsal hypothalamus, the preoptic area, the anterior commissure, the post-optic commissure, and the ventral tegmental commissure (Fig. 4a,b, Supplementary Movie S3), as previously described 27 . In contrast to the previous interpretation of immunohistochemical data 27,48 , however, the mCherry-expressing AgRP1 neurons in our study did not appear to project towards the pituitary. This was further validated in the double-transgenic larvae agrp1:m-Cherry, Tg(pomc:EGFP) zf44 , which exhibited no AgRP1 projections or terminals at the pituitary territory (Fig. 4c,  Supplementary Movie S4). AgRP2 projections towards the pituitary may have been previously mistaken for AgRP1 projections because the antibody used in those studies was shown to bind both AgRP1 and AgRP2 27,48 , and because agrp2 expression was considered at that time to be restricted to the pineal.
To examine the possibility of a relationship between AgRP1 and AgRP2 neurons in the hypothalamus, we crossed agrp1:mCherry with agrp2:EGFP fish. Imaging of the resulting larvae revealed the presence of what appeared to be 'en passant' synapses of AgRP2 preoptic projections on AgRP1 somata. Thus, AgRP2 axons were in close proximity to AgRP1 cell bodies and appeared to innervate them while projecting to the pituitary (Fig. 5a,b). Reciprocal connectivity also appeared to occur, as the AgRP1 axonal projections towards the preoptic area innervated AgRP2 somata (Fig. 5c). There was also close apposition between hypothalamic AgRP1 and AgRP2 axons as they traversed the same tracts between the preoptic area and the ventral periventricular hypothalamus. This anatomical evidence suggested close association and possible physiological inter-regulation between AgRP1 and AgRP2 neurons (as shown schematically in Fig. 5d,e).

Discussion
The ancestors of teleost fish, after diverging from the ancestors of tetrapods, underwent a whole-genome duplication event. Many of the resulting paralogous genes 22,23 , called ohnologues, quickly acquired mutations leading to one of a few possible outcomes: degeneration of one of the paralogous genes while the other retained its original function; partition of the pre-existing function (subfunctionalization); or emergence of new functions (neofunctionalization) 49,50 . Four genes belonging to the agouti family have been identified in fish, agrp1, agrp2, agouti-signaling protein 1 (asip1) and asip2. Their phylogenetic relationship is still unclear; specifically, whether agrp1 and agrp2 are duplicates or agrp2 is a paralog of asip1, is a matter of debate 24,26,51 . The anatomical distribution of the AgRP1 and AgRP2 neurons and of their projections in the zebrafish brain suggests that both subfunctionalization and neofunctionalization may have occurred: in the preoptic area and the hypothalamus each of the two agrp genes may have retained partial roles of the ancestral agrp, while in the pineal gland agrp2 has acquired a new function.
The pineal gland of non-mammalian vertebrates is photoreceptive, contains an intrinsic circadian oscillator, and influences daily rhythms and seasonal changes 33,52,53 . The role of the pineal gland in transducing photoperiodic information is particularly apparent in many teleost fish species, where it is located just underneath a thin, translucent area of the skull overlaid by less pigmented skin known as the pineal window, which facilitates the entry of light 52 . Three main types of cells have been characterized in the fish pineal gland: photoreceptor cells, projection neurons, and interstitial cells [33][34][35][36] . Data mining of pineal gland transcriptomes in zebrafish 28,29,54 reveals that agrp2 mRNA is among the most highly expressed transcripts in this tissue (data not shown). Remarkably, the pineal AgRP2 cells described here do not express known markers for pineal photoreceptor cells (AANAT2), pineal neurons (FoxD3 and HuC), or glia (GFAP), and therefore may represent a previously uncharacterized type of pineal cell. Expression of melanocortin receptors in the pineal gland ( Supplementary Fig. S3) suggests that the AgRP2 peptide exerts intracellular communication within the pineal gland. The current identification of previously uncharacterized AgRP2 pineal cells possibly reflects neofunctionalization of the agrp2 gene. The functional properties of these cells clearly warrant further thorough investigation.
Compared with the pineal AgRP2 cells, the preoptic AgRP2 neuronal system appears to have more characteristics in common with the ancestral AgRP neuronal system. The functional and anatomical properties of the AgRP neuronal network have been extensively studied in mammals [55][56][57][58] . AgRP terminals were shown to be widely spread throughout several hypothalamic structures as well as the amygdala, other forebrain regions and brainstem 55,56 . AgRP terminals were also found in the internal layer of the median eminence in rodents 56,59 , the infundibulum in rhesus monkeys 60 , and the infundibulum and median eminence in ducks 61 . Those findings suggest that AgRP may also act as a hypophysiotropic factor and thus exert neuroendocrine functions, a potential property of AgRP that has been largely overlooked. In the present study we show that zebrafish AgRP1 neurons project axons to multiple locations including rostral, intermediate and dorsal areas of the hypothalamus and the preoptic area, but do not project to the pituitary. AgRP2 preoptic neurons, on the other hand, project directly towards the pituitary, where they form a neurovascular interface with the hypophysial artery. This finding suggests that AgRP2 neurons are hypophysiotropic and may affect pituitary hormonal secretion. Expression of agrp2 mRNA was previously detected by PCR in the brains of sea bass 20 , Atlantic salmon (Salmo salar) 18 and pufferfish (Takifugu rubripes) 17 , suggesting that the preoptic AgRP2 system may not be specific to zebrafish. Thus, the present anatomical findings, together with previous studies in zebrafish, suggest this is an example of subfunctionalization, in which agrp1 has retained its function as a central regulator of food consumption 27,62 , while agrp2 has taken over the putative neuroendocrine role of agrp in the pituitary.
A possible neuroendocrine role for AgRP2 in zebrafish is in the regulation of background adaptation. Transient knock-down of AgRP2, using morpholino-modified oligonucleotides, prevented pigment aggregation when larvae were placed on a white background 30 . Since agrp2 was thought to be expressed exclusively in the pineal, it was suggested that AgRP2 is expressed in pineal neurons that regulate pigmentation through MC1R signaling and activation of hypothalamic melanin-concentrating hormone neurons 30 . We now propose an alternative mechanism of action for AgRP2. The current findings of preoptic hypophysiotropic AgRP2 neurons, together with the absence of pineal AgRP2 projections to the hypothalamus, suggest that the effect of AgRP2 on background adaptation might be exerted directly via the synapses of AgRP2 neurons in the hypophysial vasculature. Possible mechanisms for the effect of hypophysiotropic AgRP2 on pigmentation are: a) release of AgRP2 into the general circulation, thus directly affecting MC1R signaling throughout the skin and allowing for rapid background adaption; and b) modulation of pituitary melanotrope cells by AgRP2. Pituitary α -MSH affects background adaption in teleost 3 , and zebrafish melanosomes disperse rapidly in response to α -MSH 63 . However, the lack of melanocortin receptor expression in the pituitary 64 (and Supplementary Fig. S3) would suggest that such hypophysiotropic activity, if exists, is mediated by a different type of receptor. In both cases, knock-down of AgRP2 30 would be expected to increase MC1R activation by α -MSH, leading to the dispersion of melanosomes and decreased camouflage capability.
Intriguingly, AgRP2 axonal projections coursing towards the pituitary form what appear to be en passant synapses with AgRP1 cell bodies, suggesting that AgRP2 neurons can directly modulate AgRP1 neuronal activity. These connections appear to be reciprocal, with AgRP1 projections towards the preoptic area overlapping with AgRP2 projections and cell bodies. This suggests a possible mutual modulation, in which AgRP2 neurons modulate feeding via AgRP1 neurons and AgRP1 neurons modulate background adaptation via AgRP2 neurons. Whilst it is certainly highly speculative to suggest, it is possible that this mutual modulation reflects the relationship between food foraging and predation avoidance. Thus, background conditions that increase the likelihood of predation might simultaneously stimulate the induction of camouflage and suppression of appetite.
In this study we generated genetically modified zebrafish lines, thereby enabling anatomical and functional investigations of the AgRP systems. We propose that whereas zebrafish hypothalamic AgRP1 neurons function as central regulators of feeding, preoptic AgRP2 neurons have a neuroendocrine function, possibly related to pigmentation and camouflage. We showed that the two types of AgRP neurons may communicate and modulate one another. Pineal agrp2 expression appears to be a unique feature of teleosts, but its importance for pineal function remains unclear. The Gal4 transgenic lines generated here will facilitate future investigations of the AgRP1 and AgRP2 systems through the use of a variety of UAS transgenic lines. Approaches such as neuronal ablation and silencing, neuronal activation, and the monitoring of neuronal activity in these systems should now be feasible. Overall, these transgenic fish provide exciting new models for studying the neuronal mechanisms regulating food consumption and feeding behavior as well as the neuroendocrine roles of hypothalamic AgRP2 and the novel functions of pineal AgRP2.

Materials and Methods
Ethics statement. All procedures were approved by the Tel Aviv University Animal Care Committee and conducted in accordance with the requirements of the Council for Experimentation on Animal Subjects, Ministry of Health, Israel.
Fish maintenance. Adult zebrafish were raised in a recirculation water system under 12 h light:12 h dark cycles at 28 °C and fed twice a day. Embryos were generated by natural mating, placed in 10 cm Petri dishes with egg water containing methylene blue (0.3 ppm), and raised in a light-controlled incubator at 28 °C (light intensity, 12 W/m 2 ). To prevent pigmentation for ISH and immunostaining analysis, the fish water was supplemented with 0.2 mM phenylthiourea from 1-dpf onward.
BAC recombineering and transgenesis. BAC clones CH73-27D24 and CH73-352F23, containing AgRP1 and AgRP2 coding sequences, respectively, were obtained from the BACPAC Resources Center. The BAC plasmids were introduced by electroporation into Escherichia coli SW102 strain (kindly provided by Donald Court), which contains heat-inducible recombinase functions 65 . Tol2 and reporter gene cassettes (GFP or Gal4-VP16, kindly provided by Maximiliano Suster) were recombineered into the BAC clones according to established protocol 31,66 (Supplementary Fig. S1). Reporter gene cassettes were amplified by PCR together with 50 bp homology arms of agrp1 or agrp2 5′ untranslated region and first exon, with consequent recombination of the reporter gene after the AgRP1-or AgRP2-translation initiation site (Supplementary Table S1). Tol2 was amplified and recombineered as described 31 . Recombineered BAC clones were purified using HiPure Plasmid Midiprep Kit (Invitrogen), according to the manufacturer's instructions, and co-injected with Tol2 transposase RNA into single-cell embryos as described 31 . Embryos were raised and their progeny screened for positive germline transmission.
Whole-mount in-situ hybridization. agrp2 mRNA sequence (reference sequence NM_001271291.1, bp 41-544) and agrp1 mRNA sequence (reference sequence NM_001328012.1, bp 23-584) were cloned into pGEM-T Easy (Promega), linearized, and used as template for in-vitro synthesis of digoxigenin (DIG)-labelled anti-sense riboprobes (DIG RNA labelling kit, Roche). Wild-type 6-dpf larvae were fixed in 4% paraformaldehyde and stored in 100% methanol. Whole-mount ISH analysis was carried out according to an established protocol 67 . For image analysis, the ISH-stained larvae were placed in 70% glycerol and photographed with a dissecting microscope (SZX12, Olympus) equipped with a digital camera (DP70, Olympus). The ISH was repeated three times, each time with 15− 30 larvae for each gene.
Immunostaining. Transgenic 5-dpf larvae were fixed in 4% paraformaldehyde and stored in 100% methanol. The immunostaining protocol was performed as previously described 68  Confocal microscopy imaging. Transgenic 3− 12-dpf larvae were anesthetized and placed in low melting agarose for dorsal imaging. For lateral and ventral imaging the larvae were fixed in 4% paraformaldehyde, washed in PBS, and their eyes and jaws were dissected. At least five larvae of each genotype combination were analyzed. All images were obtained using a Leica TCS SP8 confocal laser scanning microscope equipped with Leica LAS AF image acquisition software.